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>
36 #include <linux/pid_namespace.h>
37 #include <linux/fdtable.h>
38 #include <linux/file.h>
51 #define CREATE_TRACE_POINTS
52 #include "trace_gfs2.h"
54 struct gfs2_glock_iter {
55 struct gfs2_sbd *sdp; /* incore superblock */
56 struct rhashtable_iter hti; /* rhashtable iterator */
57 struct gfs2_glock *gl; /* current glock struct */
58 loff_t last_pos; /* last position */
61 typedef void (*glock_examiner) (struct gfs2_glock * gl);
63 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
64 static void __gfs2_glock_dq(struct gfs2_holder *gh);
65 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
66 unsigned long delay, bool remote);
68 static struct dentry *gfs2_root;
69 static struct workqueue_struct *glock_workqueue;
70 struct workqueue_struct *gfs2_delete_workqueue;
71 static LIST_HEAD(lru_list);
72 static atomic_t lru_count = ATOMIC_INIT(0);
73 static DEFINE_SPINLOCK(lru_lock);
75 #define GFS2_GL_HASH_SHIFT 15
76 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
78 static const struct rhashtable_params ht_parms = {
79 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
80 .key_len = offsetofend(struct lm_lockname, ln_type),
81 .key_offset = offsetof(struct gfs2_glock, gl_name),
82 .head_offset = offsetof(struct gfs2_glock, gl_node),
85 static struct rhashtable gl_hash_table;
87 #define GLOCK_WAIT_TABLE_BITS 12
88 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
89 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
91 struct wait_glock_queue {
92 struct lm_lockname *name;
93 wait_queue_entry_t wait;
96 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
99 struct wait_glock_queue *wait_glock =
100 container_of(wait, struct wait_glock_queue, wait);
101 struct lm_lockname *wait_name = wait_glock->name;
102 struct lm_lockname *wake_name = key;
104 if (wake_name->ln_sbd != wait_name->ln_sbd ||
105 wake_name->ln_number != wait_name->ln_number ||
106 wake_name->ln_type != wait_name->ln_type)
108 return autoremove_wake_function(wait, mode, sync, key);
111 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
113 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
115 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
119 * wake_up_glock - Wake up waiters on a glock
122 static void wake_up_glock(struct gfs2_glock *gl)
124 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
126 if (waitqueue_active(wq))
127 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
130 static void gfs2_glock_dealloc(struct rcu_head *rcu)
132 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
134 kfree(gl->gl_lksb.sb_lvbptr);
135 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
136 struct gfs2_glock_aspace *gla =
137 container_of(gl, struct gfs2_glock_aspace, glock);
138 kmem_cache_free(gfs2_glock_aspace_cachep, gla);
140 kmem_cache_free(gfs2_glock_cachep, gl);
144 * glock_blocked_by_withdraw - determine if we can still use a glock
147 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
148 * when we're withdrawn. For example, to maintain metadata integrity, we should
149 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
150 * iopen or the transaction glocks may be safely used because none of their
151 * metadata goes through the journal. So in general, we should disallow all
152 * glocks that are journaled, and allow all the others. One exception is:
153 * we need to allow our active journal to be promoted and demoted so others
154 * may recover it and we can reacquire it when they're done.
156 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
158 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
160 if (likely(!gfs2_withdrawn(sdp)))
162 if (gl->gl_ops->go_flags & GLOF_NONDISK)
164 if (!sdp->sd_jdesc ||
165 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
170 void gfs2_glock_free(struct gfs2_glock *gl)
172 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
174 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
175 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
178 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
179 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
180 wake_up(&sdp->sd_glock_wait);
184 * gfs2_glock_hold() - increment reference count on glock
185 * @gl: The glock to hold
189 void gfs2_glock_hold(struct gfs2_glock *gl)
191 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
192 lockref_get(&gl->gl_lockref);
196 * demote_ok - Check to see if it's ok to unlock a glock
199 * Returns: 1 if it's ok
202 static int demote_ok(const struct gfs2_glock *gl)
204 const struct gfs2_glock_operations *glops = gl->gl_ops;
206 if (gl->gl_state == LM_ST_UNLOCKED)
209 * Note that demote_ok is used for the lru process of disposing of
210 * glocks. For this purpose, we don't care if the glock's holders
211 * have the HIF_MAY_DEMOTE flag set or not. If someone is using
212 * them, don't demote.
214 if (!list_empty(&gl->gl_holders))
216 if (glops->go_demote_ok)
217 return glops->go_demote_ok(gl);
222 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
224 if (!(gl->gl_ops->go_flags & GLOF_LRU))
227 spin_lock(&lru_lock);
229 list_move_tail(&gl->gl_lru, &lru_list);
231 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
232 set_bit(GLF_LRU, &gl->gl_flags);
233 atomic_inc(&lru_count);
236 spin_unlock(&lru_lock);
239 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
241 if (!(gl->gl_ops->go_flags & GLOF_LRU))
244 spin_lock(&lru_lock);
245 if (test_bit(GLF_LRU, &gl->gl_flags)) {
246 list_del_init(&gl->gl_lru);
247 atomic_dec(&lru_count);
248 clear_bit(GLF_LRU, &gl->gl_flags);
250 spin_unlock(&lru_lock);
254 * Enqueue the glock on the work queue. Passes one glock reference on to the
257 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
258 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
260 * We are holding the lockref spinlock, and the work was still
261 * queued above. The queued work (glock_work_func) takes that
262 * spinlock before dropping its glock reference(s), so it
263 * cannot have dropped them in the meantime.
265 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
266 gl->gl_lockref.count--;
270 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
271 spin_lock(&gl->gl_lockref.lock);
272 __gfs2_glock_queue_work(gl, delay);
273 spin_unlock(&gl->gl_lockref.lock);
276 static void __gfs2_glock_put(struct gfs2_glock *gl)
278 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
279 struct address_space *mapping = gfs2_glock2aspace(gl);
281 lockref_mark_dead(&gl->gl_lockref);
283 gfs2_glock_remove_from_lru(gl);
284 spin_unlock(&gl->gl_lockref.lock);
285 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
287 truncate_inode_pages_final(mapping);
288 if (!gfs2_withdrawn(sdp))
289 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
291 trace_gfs2_glock_put(gl);
292 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
296 * Cause the glock to be put in work queue context.
298 void gfs2_glock_queue_put(struct gfs2_glock *gl)
300 gfs2_glock_queue_work(gl, 0);
304 * gfs2_glock_put() - Decrement reference count on glock
305 * @gl: The glock to put
309 void gfs2_glock_put(struct gfs2_glock *gl)
311 if (lockref_put_or_lock(&gl->gl_lockref))
314 __gfs2_glock_put(gl);
318 * may_grant - check if it's ok to grant a new lock
320 * @current_gh: One of the current holders of @gl
321 * @gh: The lock request which we wish to grant
323 * With our current compatibility rules, if a glock has one or more active
324 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
325 * @current_gh; they are all the same as far as compatibility with the new @gh
328 * Returns true if it's ok to grant the lock.
331 static inline bool may_grant(struct gfs2_glock *gl,
332 struct gfs2_holder *current_gh,
333 struct gfs2_holder *gh)
336 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
338 switch(current_gh->gh_state) {
339 case LM_ST_EXCLUSIVE:
341 * Here we make a special exception to grant holders
342 * who agree to share the EX lock with other holders
343 * who also have the bit set. If the original holder
344 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
345 * holders with the bit set.
347 return gh->gh_state == LM_ST_EXCLUSIVE &&
348 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
349 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
353 return gh->gh_state == current_gh->gh_state;
360 if (gl->gl_state == gh->gh_state)
362 if (gh->gh_flags & GL_EXACT)
364 if (gl->gl_state == LM_ST_EXCLUSIVE) {
365 return gh->gh_state == LM_ST_SHARED ||
366 gh->gh_state == LM_ST_DEFERRED;
368 if (gh->gh_flags & LM_FLAG_ANY)
369 return gl->gl_state != LM_ST_UNLOCKED;
373 static void gfs2_holder_wake(struct gfs2_holder *gh)
375 clear_bit(HIF_WAIT, &gh->gh_iflags);
376 smp_mb__after_atomic();
377 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
378 if (gh->gh_flags & GL_ASYNC) {
379 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
381 wake_up(&sdp->sd_async_glock_wait);
386 * do_error - Something unexpected has happened during a lock request
388 * @ret: The status from the DLM
391 static void do_error(struct gfs2_glock *gl, const int ret)
393 struct gfs2_holder *gh, *tmp;
395 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
396 if (!test_bit(HIF_WAIT, &gh->gh_iflags))
398 if (ret & LM_OUT_ERROR)
400 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
401 gh->gh_error = GLR_TRYFAILED;
404 list_del_init(&gh->gh_list);
405 trace_gfs2_glock_queue(gh, 0);
406 gfs2_holder_wake(gh);
411 * demote_incompat_holders - demote incompatible demoteable holders
412 * @gl: the glock we want to promote
413 * @current_gh: the newly promoted holder
415 * We're passing the newly promoted holder in @current_gh, but actually, any of
416 * the strong holders would do.
418 static void demote_incompat_holders(struct gfs2_glock *gl,
419 struct gfs2_holder *current_gh)
421 struct gfs2_holder *gh, *tmp;
424 * Demote incompatible holders before we make ourselves eligible.
425 * (This holder may or may not allow auto-demoting, but we don't want
426 * to demote the new holder before it's even granted.)
428 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
430 * Since holders are at the front of the list, we stop when we
431 * find the first non-holder.
433 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
435 if (gh == current_gh)
437 if (test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags) &&
438 !may_grant(gl, current_gh, gh)) {
440 * We should not recurse into do_promote because
441 * __gfs2_glock_dq only calls handle_callback,
442 * gfs2_glock_add_to_lru and __gfs2_glock_queue_work.
450 * find_first_holder - find the first "holder" gh
454 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
456 struct gfs2_holder *gh;
458 if (!list_empty(&gl->gl_holders)) {
459 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
461 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
468 * find_first_strong_holder - find the first non-demoteable holder
471 * Find the first holder that doesn't have the HIF_MAY_DEMOTE flag set.
473 static inline struct gfs2_holder *
474 find_first_strong_holder(struct gfs2_glock *gl)
476 struct gfs2_holder *gh;
478 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
479 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
481 if (!test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags))
488 * gfs2_instantiate - Call the glops instantiate function
489 * @gh: The glock holder
491 * Returns: 0 if instantiate was successful, or error.
493 int gfs2_instantiate(struct gfs2_holder *gh)
495 struct gfs2_glock *gl = gh->gh_gl;
496 const struct gfs2_glock_operations *glops = gl->gl_ops;
500 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
504 * Since we unlock the lockref lock, we set a flag to indicate
505 * instantiate is in progress.
507 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
508 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
509 TASK_UNINTERRUPTIBLE);
511 * Here we just waited for a different instantiate to finish.
512 * But that may not have been successful, as when a process
513 * locks an inode glock _before_ it has an actual inode to
514 * instantiate into. So we check again. This process might
515 * have an inode to instantiate, so might be successful.
520 ret = glops->go_instantiate(gl);
522 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
523 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
529 return glops->go_held(gh);
534 * do_promote - promote as many requests as possible on the current queue
537 * Returns: 1 if there is a blocked holder at the head of the list
540 static int do_promote(struct gfs2_glock *gl)
542 struct gfs2_holder *gh, *current_gh;
543 bool incompat_holders_demoted = false;
545 current_gh = find_first_strong_holder(gl);
546 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
547 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
549 if (!may_grant(gl, current_gh, gh)) {
551 * If we get here, it means we may not grant this
552 * holder for some reason. If this holder is at the
553 * head of the list, it means we have a blocked holder
554 * at the head, so return 1.
556 if (list_is_first(&gh->gh_list, &gl->gl_holders))
561 set_bit(HIF_HOLDER, &gh->gh_iflags);
562 trace_gfs2_promote(gh);
563 gfs2_holder_wake(gh);
564 if (!incompat_holders_demoted) {
566 demote_incompat_holders(gl, current_gh);
567 incompat_holders_demoted = true;
574 * find_first_waiter - find the first gh that's waiting for the glock
578 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
580 struct gfs2_holder *gh;
582 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
583 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
590 * state_change - record that the glock is now in a different state
592 * @new_state: the new state
595 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
599 held1 = (gl->gl_state != LM_ST_UNLOCKED);
600 held2 = (new_state != LM_ST_UNLOCKED);
602 if (held1 != held2) {
603 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
605 gl->gl_lockref.count++;
607 gl->gl_lockref.count--;
609 if (new_state != gl->gl_target)
610 /* shorten our minimum hold time */
611 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
613 gl->gl_state = new_state;
614 gl->gl_tchange = jiffies;
617 static void gfs2_set_demote(struct gfs2_glock *gl)
619 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
621 set_bit(GLF_DEMOTE, &gl->gl_flags);
623 wake_up(&sdp->sd_async_glock_wait);
626 static void gfs2_demote_wake(struct gfs2_glock *gl)
628 gl->gl_demote_state = LM_ST_EXCLUSIVE;
629 clear_bit(GLF_DEMOTE, &gl->gl_flags);
630 smp_mb__after_atomic();
631 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
635 * finish_xmote - The DLM has replied to one of our lock requests
637 * @ret: The status from the DLM
641 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
643 const struct gfs2_glock_operations *glops = gl->gl_ops;
644 struct gfs2_holder *gh;
645 unsigned state = ret & LM_OUT_ST_MASK;
647 spin_lock(&gl->gl_lockref.lock);
648 trace_gfs2_glock_state_change(gl, state);
649 state_change(gl, state);
650 gh = find_first_waiter(gl);
652 /* Demote to UN request arrived during demote to SH or DF */
653 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
654 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
655 gl->gl_target = LM_ST_UNLOCKED;
657 /* Check for state != intended state */
658 if (unlikely(state != gl->gl_target)) {
659 if (gh && (ret & LM_OUT_CANCELED))
660 gfs2_holder_wake(gh);
661 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
662 /* move to back of queue and try next entry */
663 if (ret & LM_OUT_CANCELED) {
664 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
665 list_move_tail(&gh->gh_list, &gl->gl_holders);
666 gh = find_first_waiter(gl);
667 gl->gl_target = gh->gh_state;
670 /* Some error or failed "try lock" - report it */
671 if ((ret & LM_OUT_ERROR) ||
672 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
673 gl->gl_target = gl->gl_state;
679 /* Unlocked due to conversion deadlock, try again */
682 do_xmote(gl, gh, gl->gl_target);
684 /* Conversion fails, unlock and try again */
687 do_xmote(gl, gh, LM_ST_UNLOCKED);
689 default: /* Everything else */
690 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
691 gl->gl_target, state);
694 spin_unlock(&gl->gl_lockref.lock);
698 /* Fast path - we got what we asked for */
699 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
700 gfs2_demote_wake(gl);
701 if (state != LM_ST_UNLOCKED) {
702 if (glops->go_xmote_bh) {
705 spin_unlock(&gl->gl_lockref.lock);
706 rv = glops->go_xmote_bh(gl);
707 spin_lock(&gl->gl_lockref.lock);
716 clear_bit(GLF_LOCK, &gl->gl_flags);
717 spin_unlock(&gl->gl_lockref.lock);
720 static bool is_system_glock(struct gfs2_glock *gl)
722 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
723 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
725 if (gl == m_ip->i_gl)
731 * do_xmote - Calls the DLM to change the state of a lock
732 * @gl: The lock state
733 * @gh: The holder (only for promotes)
734 * @target: The target lock state
738 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh,
740 __releases(&gl->gl_lockref.lock)
741 __acquires(&gl->gl_lockref.lock)
743 const struct gfs2_glock_operations *glops = gl->gl_ops;
744 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
745 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
748 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
749 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
752 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
754 GLOCK_BUG_ON(gl, gl->gl_state == target);
755 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
756 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
759 * If another process is already doing the invalidate, let that
760 * finish first. The glock state machine will get back to this
761 * holder again later.
763 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
766 do_error(gl, 0); /* Fail queued try locks */
769 set_bit(GLF_BLOCKING, &gl->gl_flags);
770 if ((gl->gl_req == LM_ST_UNLOCKED) ||
771 (gl->gl_state == LM_ST_EXCLUSIVE) ||
772 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
773 clear_bit(GLF_BLOCKING, &gl->gl_flags);
774 spin_unlock(&gl->gl_lockref.lock);
775 if (glops->go_sync) {
776 ret = glops->go_sync(gl);
777 /* If we had a problem syncing (due to io errors or whatever,
778 * we should not invalidate the metadata or tell dlm to
779 * release the glock to other nodes.
782 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
783 fs_err(sdp, "Error %d syncing glock \n", ret);
784 gfs2_dump_glock(NULL, gl, true);
789 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
791 * The call to go_sync should have cleared out the ail list.
792 * If there are still items, we have a problem. We ought to
793 * withdraw, but we can't because the withdraw code also uses
794 * glocks. Warn about the error, dump the glock, then fall
795 * through and wait for logd to do the withdraw for us.
797 if ((atomic_read(&gl->gl_ail_count) != 0) &&
798 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
799 gfs2_glock_assert_warn(gl,
800 !atomic_read(&gl->gl_ail_count));
801 gfs2_dump_glock(NULL, gl, true);
803 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
804 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
810 * Check for an error encountered since we called go_sync and go_inval.
811 * If so, we can't withdraw from the glock code because the withdraw
812 * code itself uses glocks (see function signal_our_withdraw) to
813 * change the mount to read-only. Most importantly, we must not call
814 * dlm to unlock the glock until the journal is in a known good state
815 * (after journal replay) otherwise other nodes may use the object
816 * (rgrp or dinode) and then later, journal replay will corrupt the
817 * file system. The best we can do here is wait for the logd daemon
818 * to see sd_log_error and withdraw, and in the meantime, requeue the
821 * We make a special exception for some system glocks, such as the
822 * system statfs inode glock, which needs to be granted before the
823 * gfs2_quotad daemon can exit, and that exit needs to finish before
824 * we can unmount the withdrawn file system.
826 * However, if we're just unlocking the lock (say, for unmount, when
827 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
828 * then it's okay to tell dlm to unlock it.
830 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
831 gfs2_withdraw_delayed(sdp);
832 if (glock_blocked_by_withdraw(gl) &&
833 (target != LM_ST_UNLOCKED ||
834 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
835 if (!is_system_glock(gl)) {
836 handle_callback(gl, LM_ST_UNLOCKED, 0, false); /* sets demote */
838 * Ordinarily, we would call dlm and its callback would call
839 * finish_xmote, which would call state_change() to the new state.
840 * Since we withdrew, we won't call dlm, so call state_change
841 * manually, but to the UNLOCKED state we desire.
843 state_change(gl, LM_ST_UNLOCKED);
845 * We skip telling dlm to do the locking, so we won't get a
846 * reply that would otherwise clear GLF_LOCK. So we clear it here.
848 clear_bit(GLF_LOCK, &gl->gl_flags);
849 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
850 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
853 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
857 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
859 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
860 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
861 target == LM_ST_UNLOCKED &&
862 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
863 finish_xmote(gl, target);
864 gfs2_glock_queue_work(gl, 0);
866 fs_err(sdp, "lm_lock ret %d\n", ret);
867 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
869 } else { /* lock_nolock */
870 finish_xmote(gl, target);
871 gfs2_glock_queue_work(gl, 0);
874 spin_lock(&gl->gl_lockref.lock);
878 * run_queue - do all outstanding tasks related to a glock
879 * @gl: The glock in question
880 * @nonblock: True if we must not block in run_queue
884 static void run_queue(struct gfs2_glock *gl, const int nonblock)
885 __releases(&gl->gl_lockref.lock)
886 __acquires(&gl->gl_lockref.lock)
888 struct gfs2_holder *gh = NULL;
890 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
893 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
895 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
896 gl->gl_demote_state != gl->gl_state) {
897 if (find_first_holder(gl))
901 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
902 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
903 gl->gl_target = gl->gl_demote_state;
905 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
906 gfs2_demote_wake(gl);
907 if (do_promote(gl) == 0)
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);
918 clear_bit(GLF_LOCK, &gl->gl_flags);
919 smp_mb__after_atomic();
920 gl->gl_lockref.count++;
921 __gfs2_glock_queue_work(gl, 0);
925 clear_bit(GLF_LOCK, &gl->gl_flags);
926 smp_mb__after_atomic();
930 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
932 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
934 if (ri->ri_magic == 0)
935 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
936 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
937 ri->ri_generation_deleted = cpu_to_be64(generation);
940 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
942 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
944 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
946 return generation <= be64_to_cpu(ri->ri_generation_deleted);
949 static void gfs2_glock_poke(struct gfs2_glock *gl)
951 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
952 struct gfs2_holder gh;
955 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
956 error = gfs2_glock_nq(&gh);
959 gfs2_holder_uninit(&gh);
962 static bool gfs2_try_evict(struct gfs2_glock *gl)
964 struct gfs2_inode *ip;
965 bool evicted = false;
968 * If there is contention on the iopen glock and we have an inode, try
969 * to grab and release the inode so that it can be evicted. This will
970 * allow the remote node to go ahead and delete the inode without us
971 * having to do it, which will avoid rgrp glock thrashing.
973 * The remote node is likely still holding the corresponding inode
974 * glock, so it will run before we get to verify that the delete has
977 spin_lock(&gl->gl_lockref.lock);
979 if (ip && !igrab(&ip->i_inode))
981 spin_unlock(&gl->gl_lockref.lock);
983 struct gfs2_glock *inode_gl = NULL;
985 gl->gl_no_formal_ino = ip->i_no_formal_ino;
986 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
987 d_prune_aliases(&ip->i_inode);
990 /* If the inode was evicted, gl->gl_object will now be NULL. */
991 spin_lock(&gl->gl_lockref.lock);
995 lockref_get(&inode_gl->gl_lockref);
996 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
998 spin_unlock(&gl->gl_lockref.lock);
1000 gfs2_glock_poke(inode_gl);
1001 gfs2_glock_put(inode_gl);
1008 static void delete_work_func(struct work_struct *work)
1010 struct delayed_work *dwork = to_delayed_work(work);
1011 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1012 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1013 struct inode *inode;
1014 u64 no_addr = gl->gl_name.ln_number;
1016 spin_lock(&gl->gl_lockref.lock);
1017 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1018 spin_unlock(&gl->gl_lockref.lock);
1020 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1022 * If we can evict the inode, give the remote node trying to
1023 * delete the inode some time before verifying that the delete
1024 * has happened. Otherwise, if we cause contention on the inode glock
1025 * immediately, the remote node will think that we still have
1026 * the inode in use, and so it will give up waiting.
1028 * If we can't evict the inode, signal to the remote node that
1029 * the inode is still in use. We'll later try to delete the
1030 * inode locally in gfs2_evict_inode.
1032 * FIXME: We only need to verify that the remote node has
1033 * deleted the inode because nodes before this remote delete
1034 * rework won't cooperate. At a later time, when we no longer
1035 * care about compatibility with such nodes, we can skip this
1038 if (gfs2_try_evict(gl)) {
1039 if (gfs2_queue_delete_work(gl, 5 * HZ))
1045 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1046 GFS2_BLKST_UNLINKED);
1047 if (IS_ERR(inode)) {
1048 if (PTR_ERR(inode) == -EAGAIN &&
1049 (gfs2_queue_delete_work(gl, 5 * HZ)))
1052 d_prune_aliases(inode);
1059 static void glock_work_func(struct work_struct *work)
1061 unsigned long delay = 0;
1062 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1063 unsigned int drop_refs = 1;
1065 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1066 finish_xmote(gl, gl->gl_reply);
1069 spin_lock(&gl->gl_lockref.lock);
1070 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1071 gl->gl_state != LM_ST_UNLOCKED &&
1072 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1073 unsigned long holdtime, now = jiffies;
1075 holdtime = gl->gl_tchange + gl->gl_hold_time;
1076 if (time_before(now, holdtime))
1077 delay = holdtime - now;
1080 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1081 gfs2_set_demote(gl);
1086 /* Keep one glock reference for the work we requeue. */
1088 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1090 __gfs2_glock_queue_work(gl, delay);
1094 * Drop the remaining glock references manually here. (Mind that
1095 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1098 gl->gl_lockref.count -= drop_refs;
1099 if (!gl->gl_lockref.count) {
1100 __gfs2_glock_put(gl);
1103 spin_unlock(&gl->gl_lockref.lock);
1106 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1107 struct gfs2_glock *new)
1109 struct wait_glock_queue wait;
1110 wait_queue_head_t *wq = glock_waitqueue(name);
1111 struct gfs2_glock *gl;
1114 init_wait(&wait.wait);
1115 wait.wait.func = glock_wake_function;
1118 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1121 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1122 &new->gl_node, ht_parms);
1126 gl = rhashtable_lookup_fast(&gl_hash_table,
1129 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1136 finish_wait(wq, &wait.wait);
1141 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1142 * @sdp: The GFS2 superblock
1143 * @number: the lock number
1144 * @glops: The glock_operations to use
1145 * @create: If 0, don't create the glock if it doesn't exist
1146 * @glp: the glock is returned here
1148 * This does not lock a glock, just finds/creates structures for one.
1153 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1154 const struct gfs2_glock_operations *glops, int create,
1155 struct gfs2_glock **glp)
1157 struct super_block *s = sdp->sd_vfs;
1158 struct lm_lockname name = { .ln_number = number,
1159 .ln_type = glops->go_type,
1161 struct gfs2_glock *gl, *tmp;
1162 struct address_space *mapping;
1165 gl = find_insert_glock(&name, NULL);
1173 if (glops->go_flags & GLOF_ASPACE) {
1174 struct gfs2_glock_aspace *gla =
1175 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1180 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1184 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1187 if (glops->go_flags & GLOF_LVB) {
1188 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1189 if (!gl->gl_lksb.sb_lvbptr) {
1190 gfs2_glock_dealloc(&gl->gl_rcu);
1195 atomic_inc(&sdp->sd_glock_disposal);
1196 gl->gl_node.next = NULL;
1197 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1199 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1200 gl->gl_lockref.count = 1;
1201 gl->gl_state = LM_ST_UNLOCKED;
1202 gl->gl_target = LM_ST_UNLOCKED;
1203 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1206 /* We use the global stats to estimate the initial per-glock stats */
1207 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1209 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1210 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1211 gl->gl_tchange = jiffies;
1212 gl->gl_object = NULL;
1213 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1214 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1215 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1216 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1218 mapping = gfs2_glock2aspace(gl);
1220 mapping->a_ops = &gfs2_meta_aops;
1221 mapping->host = s->s_bdev->bd_inode;
1223 mapping_set_gfp_mask(mapping, GFP_NOFS);
1224 mapping->private_data = NULL;
1225 mapping->writeback_index = 0;
1228 tmp = find_insert_glock(&name, gl);
1240 gfs2_glock_dealloc(&gl->gl_rcu);
1241 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1242 wake_up(&sdp->sd_glock_wait);
1249 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1251 * @state: the state we're requesting
1252 * @flags: the modifier flags
1253 * @gh: the holder structure
1257 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1258 struct gfs2_holder *gh, unsigned long ip)
1260 INIT_LIST_HEAD(&gh->gh_list);
1263 gh->gh_owner_pid = get_pid(task_pid(current));
1264 gh->gh_state = state;
1265 gh->gh_flags = flags;
1267 gfs2_glock_hold(gl);
1271 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1272 * @state: the state we're requesting
1273 * @flags: the modifier flags
1274 * @gh: the holder structure
1276 * Don't mess with the glock.
1280 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1282 gh->gh_state = state;
1283 gh->gh_flags = flags;
1285 gh->gh_ip = _RET_IP_;
1286 put_pid(gh->gh_owner_pid);
1287 gh->gh_owner_pid = get_pid(task_pid(current));
1291 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1292 * @gh: the holder structure
1296 void gfs2_holder_uninit(struct gfs2_holder *gh)
1298 put_pid(gh->gh_owner_pid);
1299 gfs2_glock_put(gh->gh_gl);
1300 gfs2_holder_mark_uninitialized(gh);
1304 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1305 unsigned long start_time)
1307 /* Have we waited longer that a second? */
1308 if (time_after(jiffies, start_time + HZ)) {
1309 /* Lengthen the minimum hold time. */
1310 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1316 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1317 * @gh: the glock holder
1319 * Called when a glock holder no longer needs to be waited for because it is
1320 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1321 * failed (gh_error != 0).
1324 int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1326 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1327 return gh->gh_error;
1328 gh->gh_error = gfs2_instantiate(gh);
1331 return gh->gh_error;
1335 * gfs2_glock_wait - wait on a glock acquisition
1336 * @gh: the glock holder
1338 * Returns: 0 on success
1341 int gfs2_glock_wait(struct gfs2_holder *gh)
1343 unsigned long start_time = jiffies;
1346 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1347 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1348 return gfs2_glock_holder_ready(gh);
1351 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1355 for (i = 0; i < num_gh; i++)
1356 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1362 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1363 * @num_gh: the number of holders in the array
1364 * @ghs: the glock holder array
1366 * Returns: 0 on success, meaning all glocks have been granted and are held.
1367 * -ESTALE if the request timed out, meaning all glocks were released,
1368 * and the caller should retry the operation.
1371 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1373 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1374 int i, ret = 0, timeout = 0;
1375 unsigned long start_time = jiffies;
1379 * Total up the (minimum hold time * 2) of all glocks and use that to
1380 * determine the max amount of time we should wait.
1382 for (i = 0; i < num_gh; i++)
1383 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1385 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1386 !glocks_pending(num_gh, ghs), timeout)) {
1387 ret = -ESTALE; /* request timed out. */
1391 for (i = 0; i < num_gh; i++) {
1392 struct gfs2_holder *gh = &ghs[i];
1395 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1396 gfs2_glock_update_hold_time(gh->gh_gl,
1399 ret2 = gfs2_glock_holder_ready(gh);
1406 for (i = 0; i < num_gh; i++) {
1407 struct gfs2_holder *gh = &ghs[i];
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);
1464 static inline bool pid_is_meaningful(const struct gfs2_holder *gh)
1466 if (!(gh->gh_flags & GL_NOPID))
1468 if (gh->gh_state == LM_ST_UNLOCKED)
1474 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1475 * @gh: the holder structure to add
1477 * Eventually we should move the recursive locking trap to a
1478 * debugging option or something like that. This is the fast
1479 * path and needs to have the minimum number of distractions.
1483 static inline void add_to_queue(struct gfs2_holder *gh)
1484 __releases(&gl->gl_lockref.lock)
1485 __acquires(&gl->gl_lockref.lock)
1487 struct gfs2_glock *gl = gh->gh_gl;
1488 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1489 struct list_head *insert_pt = NULL;
1490 struct gfs2_holder *gh2;
1493 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1494 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1495 GLOCK_BUG_ON(gl, true);
1497 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1498 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1499 struct gfs2_holder *current_gh;
1501 current_gh = find_first_strong_holder(gl);
1502 try_futile = !may_grant(gl, current_gh, gh);
1504 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1508 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1509 if (likely(gh2->gh_owner_pid != gh->gh_owner_pid))
1511 if (gh->gh_gl->gl_ops->go_type == LM_TYPE_FLOCK)
1513 if (test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags))
1515 if (!pid_is_meaningful(gh2))
1517 goto trap_recursive;
1519 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1521 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1523 gh->gh_error = GLR_TRYFAILED;
1524 gfs2_holder_wake(gh);
1527 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1529 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1530 insert_pt = &gh2->gh_list;
1532 trace_gfs2_glock_queue(gh, 1);
1533 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1534 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1535 if (likely(insert_pt == NULL)) {
1536 list_add_tail(&gh->gh_list, &gl->gl_holders);
1537 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1541 list_add_tail(&gh->gh_list, insert_pt);
1543 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1544 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1545 spin_unlock(&gl->gl_lockref.lock);
1546 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1547 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1548 spin_lock(&gl->gl_lockref.lock);
1553 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1554 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1555 fs_err(sdp, "lock type: %d req lock state : %d\n",
1556 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1557 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1558 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1559 fs_err(sdp, "lock type: %d req lock state : %d\n",
1560 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1561 gfs2_dump_glock(NULL, gl, true);
1566 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1567 * @gh: the holder structure
1569 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1571 * Returns: 0, GLR_TRYFAILED, or errno on failure
1574 int gfs2_glock_nq(struct gfs2_holder *gh)
1576 struct gfs2_glock *gl = gh->gh_gl;
1579 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1582 if (test_bit(GLF_LRU, &gl->gl_flags))
1583 gfs2_glock_remove_from_lru(gl);
1586 spin_lock(&gl->gl_lockref.lock);
1588 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1589 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1590 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1591 gl->gl_lockref.count++;
1592 __gfs2_glock_queue_work(gl, 0);
1595 spin_unlock(&gl->gl_lockref.lock);
1597 if (!(gh->gh_flags & GL_ASYNC))
1598 error = gfs2_glock_wait(gh);
1604 * gfs2_glock_poll - poll to see if an async request has been completed
1607 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1610 int gfs2_glock_poll(struct gfs2_holder *gh)
1612 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1615 static inline bool needs_demote(struct gfs2_glock *gl)
1617 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1618 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1621 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1623 struct gfs2_glock *gl = gh->gh_gl;
1624 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1629 * This while loop is similar to function demote_incompat_holders:
1630 * If the glock is due to be demoted (which may be from another node
1631 * or even if this holder is GL_NOCACHE), the weak holders are
1632 * demoted as well, allowing the glock to be demoted.
1636 * If we're in the process of file system withdraw, we cannot
1637 * just dequeue any glocks until our journal is recovered, lest
1638 * we introduce file system corruption. We need two exceptions
1639 * to this rule: We need to allow unlocking of nondisk glocks
1640 * and the glock for our own journal that needs recovery.
1642 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1643 glock_blocked_by_withdraw(gl) &&
1644 gh->gh_gl != sdp->sd_jinode_gl) {
1645 sdp->sd_glock_dqs_held++;
1646 spin_unlock(&gl->gl_lockref.lock);
1648 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1649 TASK_UNINTERRUPTIBLE);
1650 spin_lock(&gl->gl_lockref.lock);
1654 * This holder should not be cached, so mark it for demote.
1655 * Note: this should be done before the check for needs_demote
1658 if (gh->gh_flags & GL_NOCACHE)
1659 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1661 list_del_init(&gh->gh_list);
1662 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1663 trace_gfs2_glock_queue(gh, 0);
1666 * If there hasn't been a demote request we are done.
1667 * (Let the remaining holders, if any, keep holding it.)
1669 if (!needs_demote(gl)) {
1670 if (list_empty(&gl->gl_holders))
1675 * If we have another strong holder (we cannot auto-demote)
1676 * we are done. It keeps holding it until it is done.
1678 if (find_first_strong_holder(gl))
1682 * If we have a weak holder at the head of the list, it
1683 * (and all others like it) must be auto-demoted. If there
1684 * are no more weak holders, we exit the while loop.
1686 gh = find_first_holder(gl);
1689 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1690 gfs2_glock_add_to_lru(gl);
1692 if (unlikely(!fast_path)) {
1693 gl->gl_lockref.count++;
1694 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1695 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1696 gl->gl_name.ln_type == LM_TYPE_INODE)
1697 delay = gl->gl_hold_time;
1698 __gfs2_glock_queue_work(gl, delay);
1703 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1704 * @gh: the glock holder
1707 void gfs2_glock_dq(struct gfs2_holder *gh)
1709 struct gfs2_glock *gl = gh->gh_gl;
1711 spin_lock(&gl->gl_lockref.lock);
1712 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1713 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1714 spin_unlock(&gl->gl_lockref.lock);
1715 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1716 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1717 spin_lock(&gl->gl_lockref.lock);
1720 __gfs2_glock_dq(gh);
1721 spin_unlock(&gl->gl_lockref.lock);
1724 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1726 struct gfs2_glock *gl = gh->gh_gl;
1729 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1733 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1734 * @gh: the holder structure
1738 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1741 gfs2_holder_uninit(gh);
1745 * gfs2_glock_nq_num - acquire a glock based on lock number
1746 * @sdp: the filesystem
1747 * @number: the lock number
1748 * @glops: the glock operations for the type of glock
1749 * @state: the state to acquire the glock in
1750 * @flags: modifier flags for the acquisition
1751 * @gh: the struct gfs2_holder
1756 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1757 const struct gfs2_glock_operations *glops,
1758 unsigned int state, u16 flags, struct gfs2_holder *gh)
1760 struct gfs2_glock *gl;
1763 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1765 error = gfs2_glock_nq_init(gl, state, flags, gh);
1773 * glock_compare - Compare two struct gfs2_glock structures for sorting
1774 * @arg_a: the first structure
1775 * @arg_b: the second structure
1779 static int glock_compare(const void *arg_a, const void *arg_b)
1781 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1782 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1783 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1784 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1786 if (a->ln_number > b->ln_number)
1788 if (a->ln_number < b->ln_number)
1790 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1795 * nq_m_sync - synchronously acquire more than one glock in deadlock free order
1796 * @num_gh: the number of structures
1797 * @ghs: an array of struct gfs2_holder structures
1798 * @p: placeholder for the holder structure to pass back
1800 * Returns: 0 on success (all glocks acquired),
1801 * errno on failure (no glocks acquired)
1804 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1805 struct gfs2_holder **p)
1810 for (x = 0; x < num_gh; x++)
1813 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1815 for (x = 0; x < num_gh; x++) {
1816 error = gfs2_glock_nq(p[x]);
1819 gfs2_glock_dq(p[x]);
1828 * gfs2_glock_nq_m - acquire multiple glocks
1829 * @num_gh: the number of structures
1830 * @ghs: an array of struct gfs2_holder structures
1832 * Returns: 0 on success (all glocks acquired),
1833 * errno on failure (no glocks acquired)
1836 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1838 struct gfs2_holder *tmp[4];
1839 struct gfs2_holder **pph = tmp;
1846 return gfs2_glock_nq(ghs);
1850 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1856 error = nq_m_sync(num_gh, ghs, pph);
1865 * gfs2_glock_dq_m - release multiple glocks
1866 * @num_gh: the number of structures
1867 * @ghs: an array of struct gfs2_holder structures
1871 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1874 gfs2_glock_dq(&ghs[num_gh]);
1877 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1879 unsigned long delay = 0;
1880 unsigned long holdtime;
1881 unsigned long now = jiffies;
1883 gfs2_glock_hold(gl);
1884 spin_lock(&gl->gl_lockref.lock);
1885 holdtime = gl->gl_tchange + gl->gl_hold_time;
1886 if (!list_empty(&gl->gl_holders) &&
1887 gl->gl_name.ln_type == LM_TYPE_INODE) {
1888 if (time_before(now, holdtime))
1889 delay = holdtime - now;
1890 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1891 delay = gl->gl_hold_time;
1894 * Note 1: We cannot call demote_incompat_holders from handle_callback
1895 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1896 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1897 * Plus, we only want to demote the holders if the request comes from
1898 * a remote cluster node because local holder conflicts are resolved
1901 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1902 * request that we set our state to UNLOCKED. Here we mock up a holder
1903 * to make it look like someone wants the lock EX locally. Any SH
1904 * and DF requests should be able to share the lock without demoting.
1906 * Note 3: We only want to demote the demoteable holders when there
1907 * are no more strong holders. The demoteable holders might as well
1908 * keep the glock until the last strong holder is done with it.
1910 if (!find_first_strong_holder(gl)) {
1911 struct gfs2_holder mock_gh = {
1913 .gh_state = (state == LM_ST_UNLOCKED) ?
1914 LM_ST_EXCLUSIVE : state,
1915 .gh_iflags = BIT(HIF_HOLDER)
1918 demote_incompat_holders(gl, &mock_gh);
1920 handle_callback(gl, state, delay, true);
1921 __gfs2_glock_queue_work(gl, delay);
1922 spin_unlock(&gl->gl_lockref.lock);
1926 * gfs2_should_freeze - Figure out if glock should be frozen
1927 * @gl: The glock in question
1929 * Glocks are not frozen if (a) the result of the dlm operation is
1930 * an error, (b) the locking operation was an unlock operation or
1931 * (c) if there is a "noexp" flagged request anywhere in the queue
1933 * Returns: 1 if freezing should occur, 0 otherwise
1936 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1938 const struct gfs2_holder *gh;
1940 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1942 if (gl->gl_target == LM_ST_UNLOCKED)
1945 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1946 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1948 if (LM_FLAG_NOEXP & gh->gh_flags)
1956 * gfs2_glock_complete - Callback used by locking
1957 * @gl: Pointer to the glock
1958 * @ret: The return value from the dlm
1960 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1961 * to use a bitfield shared with other glock state fields.
1964 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1966 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1968 spin_lock(&gl->gl_lockref.lock);
1971 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1972 if (gfs2_should_freeze(gl)) {
1973 set_bit(GLF_FROZEN, &gl->gl_flags);
1974 spin_unlock(&gl->gl_lockref.lock);
1979 gl->gl_lockref.count++;
1980 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1981 __gfs2_glock_queue_work(gl, 0);
1982 spin_unlock(&gl->gl_lockref.lock);
1985 static int glock_cmp(void *priv, const struct list_head *a,
1986 const struct list_head *b)
1988 struct gfs2_glock *gla, *glb;
1990 gla = list_entry(a, struct gfs2_glock, gl_lru);
1991 glb = list_entry(b, struct gfs2_glock, gl_lru);
1993 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1995 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
2002 * gfs2_dispose_glock_lru - Demote a list of glocks
2003 * @list: The list to dispose of
2005 * Disposing of glocks may involve disk accesses, so that here we sort
2006 * the glocks by number (i.e. disk location of the inodes) so that if
2007 * there are any such accesses, they'll be sent in order (mostly).
2009 * Must be called under the lru_lock, but may drop and retake this
2010 * lock. While the lru_lock is dropped, entries may vanish from the
2011 * list, but no new entries will appear on the list (since it is
2015 static void gfs2_dispose_glock_lru(struct list_head *list)
2016 __releases(&lru_lock)
2017 __acquires(&lru_lock)
2019 struct gfs2_glock *gl;
2021 list_sort(NULL, list, glock_cmp);
2023 while(!list_empty(list)) {
2024 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2025 list_del_init(&gl->gl_lru);
2026 clear_bit(GLF_LRU, &gl->gl_flags);
2027 if (!spin_trylock(&gl->gl_lockref.lock)) {
2029 list_add(&gl->gl_lru, &lru_list);
2030 set_bit(GLF_LRU, &gl->gl_flags);
2031 atomic_inc(&lru_count);
2034 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2035 spin_unlock(&gl->gl_lockref.lock);
2036 goto add_back_to_lru;
2038 gl->gl_lockref.count++;
2040 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2041 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2042 __gfs2_glock_queue_work(gl, 0);
2043 spin_unlock(&gl->gl_lockref.lock);
2044 cond_resched_lock(&lru_lock);
2049 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2050 * @nr: The number of entries to scan
2052 * This function selects the entries on the LRU which are able to
2053 * be demoted, and then kicks off the process by calling
2054 * gfs2_dispose_glock_lru() above.
2057 static long gfs2_scan_glock_lru(int nr)
2059 struct gfs2_glock *gl;
2064 spin_lock(&lru_lock);
2065 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2066 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2068 /* Test for being demotable */
2069 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2070 list_move(&gl->gl_lru, &dispose);
2071 atomic_dec(&lru_count);
2076 list_move(&gl->gl_lru, &skipped);
2078 list_splice(&skipped, &lru_list);
2079 if (!list_empty(&dispose))
2080 gfs2_dispose_glock_lru(&dispose);
2081 spin_unlock(&lru_lock);
2086 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2087 struct shrink_control *sc)
2089 if (!(sc->gfp_mask & __GFP_FS))
2091 return gfs2_scan_glock_lru(sc->nr_to_scan);
2094 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2095 struct shrink_control *sc)
2097 return vfs_pressure_ratio(atomic_read(&lru_count));
2100 static struct shrinker glock_shrinker = {
2101 .seeks = DEFAULT_SEEKS,
2102 .count_objects = gfs2_glock_shrink_count,
2103 .scan_objects = gfs2_glock_shrink_scan,
2107 * glock_hash_walk - Call a function for glock in a hash bucket
2108 * @examiner: the function
2109 * @sdp: the filesystem
2111 * Note that the function can be called multiple times on the same
2112 * object. So the user must ensure that the function can cope with
2116 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2118 struct gfs2_glock *gl;
2119 struct rhashtable_iter iter;
2121 rhashtable_walk_enter(&gl_hash_table, &iter);
2124 rhashtable_walk_start(&iter);
2126 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2127 if (gl->gl_name.ln_sbd == sdp)
2131 rhashtable_walk_stop(&iter);
2132 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2134 rhashtable_walk_exit(&iter);
2137 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2141 spin_lock(&gl->gl_lockref.lock);
2142 queued = queue_delayed_work(gfs2_delete_workqueue,
2143 &gl->gl_delete, delay);
2145 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2146 spin_unlock(&gl->gl_lockref.lock);
2150 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2152 if (cancel_delayed_work(&gl->gl_delete)) {
2153 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2158 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2160 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2163 static void flush_delete_work(struct gfs2_glock *gl)
2165 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2166 if (cancel_delayed_work(&gl->gl_delete)) {
2167 queue_delayed_work(gfs2_delete_workqueue,
2173 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2175 glock_hash_walk(flush_delete_work, sdp);
2176 flush_workqueue(gfs2_delete_workqueue);
2180 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2181 * @gl: The glock to thaw
2185 static void thaw_glock(struct gfs2_glock *gl)
2187 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2189 if (!lockref_get_not_dead(&gl->gl_lockref))
2191 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2192 gfs2_glock_queue_work(gl, 0);
2196 * clear_glock - look at a glock and see if we can free it from glock cache
2197 * @gl: the glock to look at
2201 static void clear_glock(struct gfs2_glock *gl)
2203 gfs2_glock_remove_from_lru(gl);
2205 spin_lock(&gl->gl_lockref.lock);
2206 if (!__lockref_is_dead(&gl->gl_lockref)) {
2207 gl->gl_lockref.count++;
2208 if (gl->gl_state != LM_ST_UNLOCKED)
2209 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2210 __gfs2_glock_queue_work(gl, 0);
2212 spin_unlock(&gl->gl_lockref.lock);
2216 * gfs2_glock_thaw - Thaw any frozen glocks
2217 * @sdp: The super block
2221 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2223 glock_hash_walk(thaw_glock, sdp);
2226 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2228 spin_lock(&gl->gl_lockref.lock);
2229 gfs2_dump_glock(seq, gl, fsid);
2230 spin_unlock(&gl->gl_lockref.lock);
2233 static void dump_glock_func(struct gfs2_glock *gl)
2235 dump_glock(NULL, gl, true);
2238 static void withdraw_dq(struct gfs2_glock *gl)
2240 spin_lock(&gl->gl_lockref.lock);
2241 if (!__lockref_is_dead(&gl->gl_lockref) &&
2242 glock_blocked_by_withdraw(gl))
2243 do_error(gl, LM_OUT_ERROR); /* remove pending waiters */
2244 spin_unlock(&gl->gl_lockref.lock);
2247 void gfs2_gl_dq_holders(struct gfs2_sbd *sdp)
2249 glock_hash_walk(withdraw_dq, sdp);
2253 * gfs2_gl_hash_clear - Empty out the glock hash table
2254 * @sdp: the filesystem
2256 * Called when unmounting the filesystem.
2259 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2261 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2262 flush_workqueue(glock_workqueue);
2263 glock_hash_walk(clear_glock, sdp);
2264 flush_workqueue(glock_workqueue);
2265 wait_event_timeout(sdp->sd_glock_wait,
2266 atomic_read(&sdp->sd_glock_disposal) == 0,
2268 glock_hash_walk(dump_glock_func, sdp);
2271 static const char *state2str(unsigned state)
2274 case LM_ST_UNLOCKED:
2278 case LM_ST_DEFERRED:
2280 case LM_ST_EXCLUSIVE:
2286 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2289 if (flags & LM_FLAG_TRY)
2291 if (flags & LM_FLAG_TRY_1CB)
2293 if (flags & LM_FLAG_NOEXP)
2295 if (flags & LM_FLAG_ANY)
2297 if (flags & LM_FLAG_PRIORITY)
2299 if (flags & LM_FLAG_NODE_SCOPE)
2301 if (flags & GL_ASYNC)
2303 if (flags & GL_EXACT)
2305 if (flags & GL_NOCACHE)
2307 if (test_bit(HIF_HOLDER, &iflags))
2309 if (test_bit(HIF_WAIT, &iflags))
2311 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2313 if (flags & GL_SKIP)
2320 * dump_holder - print information about a glock holder
2321 * @seq: the seq_file struct
2322 * @gh: the glock holder
2323 * @fs_id_buf: pointer to file system id (if requested)
2327 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2328 const char *fs_id_buf)
2330 const char *comm = "(none)";
2331 pid_t owner_pid = 0;
2335 if (pid_is_meaningful(gh)) {
2336 struct task_struct *gh_owner;
2339 owner_pid = pid_nr(gh->gh_owner_pid);
2340 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2342 comm = gh_owner->comm;
2344 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2345 fs_id_buf, state2str(gh->gh_state),
2346 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2347 gh->gh_error, (long)owner_pid, comm, (void *)gh->gh_ip);
2351 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2353 const unsigned long *gflags = &gl->gl_flags;
2356 if (test_bit(GLF_LOCK, gflags))
2358 if (test_bit(GLF_DEMOTE, gflags))
2360 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2362 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2364 if (test_bit(GLF_DIRTY, gflags))
2366 if (test_bit(GLF_LFLUSH, gflags))
2368 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2370 if (test_bit(GLF_REPLY_PENDING, gflags))
2372 if (test_bit(GLF_INITIAL, gflags))
2374 if (test_bit(GLF_FROZEN, gflags))
2376 if (!list_empty(&gl->gl_holders))
2378 if (test_bit(GLF_LRU, gflags))
2382 if (test_bit(GLF_BLOCKING, gflags))
2384 if (test_bit(GLF_PENDING_DELETE, gflags))
2386 if (test_bit(GLF_FREEING, gflags))
2388 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2390 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2397 * gfs2_dump_glock - print information about a glock
2398 * @seq: The seq_file struct
2400 * @fsid: If true, also dump the file system id
2402 * The file format is as follows:
2403 * One line per object, capital letters are used to indicate objects
2404 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2405 * other objects are indented by a single space and follow the glock to
2406 * which they are related. Fields are indicated by lower case letters
2407 * followed by a colon and the field value, except for strings which are in
2408 * [] so that its possible to see if they are composed of spaces for
2409 * example. The field's are n = number (id of the object), f = flags,
2410 * t = type, s = state, r = refcount, e = error, p = pid.
2414 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2416 const struct gfs2_glock_operations *glops = gl->gl_ops;
2417 unsigned long long dtime;
2418 const struct gfs2_holder *gh;
2419 char gflags_buf[32];
2420 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2421 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2422 unsigned long nrpages = 0;
2424 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2425 struct address_space *mapping = gfs2_glock2aspace(gl);
2427 nrpages = mapping->nrpages;
2429 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2430 if (fsid && sdp) /* safety precaution */
2431 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2432 dtime = jiffies - gl->gl_demote_time;
2433 dtime *= 1000000/HZ; /* demote time in uSec */
2434 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2436 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2437 "v:%d r:%d m:%ld p:%lu\n",
2438 fs_id_buf, state2str(gl->gl_state),
2439 gl->gl_name.ln_type,
2440 (unsigned long long)gl->gl_name.ln_number,
2441 gflags2str(gflags_buf, gl),
2442 state2str(gl->gl_target),
2443 state2str(gl->gl_demote_state), dtime,
2444 atomic_read(&gl->gl_ail_count),
2445 atomic_read(&gl->gl_revokes),
2446 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2448 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2449 dump_holder(seq, gh, fs_id_buf);
2451 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2452 glops->go_dump(seq, gl, fs_id_buf);
2455 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2457 struct gfs2_glock *gl = iter_ptr;
2459 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2460 gl->gl_name.ln_type,
2461 (unsigned long long)gl->gl_name.ln_number,
2462 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2463 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2464 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2465 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2466 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2467 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2468 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2469 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2473 static const char *gfs2_gltype[] = {
2487 static const char *gfs2_stype[] = {
2488 [GFS2_LKS_SRTT] = "srtt",
2489 [GFS2_LKS_SRTTVAR] = "srttvar",
2490 [GFS2_LKS_SRTTB] = "srttb",
2491 [GFS2_LKS_SRTTVARB] = "srttvarb",
2492 [GFS2_LKS_SIRT] = "sirt",
2493 [GFS2_LKS_SIRTVAR] = "sirtvar",
2494 [GFS2_LKS_DCOUNT] = "dlm",
2495 [GFS2_LKS_QCOUNT] = "queue",
2498 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2500 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2502 struct gfs2_sbd *sdp = seq->private;
2503 loff_t pos = *(loff_t *)iter_ptr;
2504 unsigned index = pos >> 3;
2505 unsigned subindex = pos & 0x07;
2508 if (index == 0 && subindex != 0)
2511 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2512 (index == 0) ? "cpu": gfs2_stype[subindex]);
2514 for_each_possible_cpu(i) {
2515 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2518 seq_printf(seq, " %15u", i);
2520 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2521 lkstats[index - 1].stats[subindex]);
2523 seq_putc(seq, '\n');
2527 int __init gfs2_glock_init(void)
2531 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2535 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2536 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2537 if (!glock_workqueue) {
2538 rhashtable_destroy(&gl_hash_table);
2541 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2542 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2544 if (!gfs2_delete_workqueue) {
2545 destroy_workqueue(glock_workqueue);
2546 rhashtable_destroy(&gl_hash_table);
2550 ret = register_shrinker(&glock_shrinker, "gfs2-glock");
2552 destroy_workqueue(gfs2_delete_workqueue);
2553 destroy_workqueue(glock_workqueue);
2554 rhashtable_destroy(&gl_hash_table);
2558 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2559 init_waitqueue_head(glock_wait_table + i);
2564 void gfs2_glock_exit(void)
2566 unregister_shrinker(&glock_shrinker);
2567 rhashtable_destroy(&gl_hash_table);
2568 destroy_workqueue(glock_workqueue);
2569 destroy_workqueue(gfs2_delete_workqueue);
2572 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2574 struct gfs2_glock *gl = gi->gl;
2579 if (!lockref_put_not_zero(&gl->gl_lockref))
2580 gfs2_glock_queue_put(gl);
2583 gl = rhashtable_walk_next(&gi->hti);
2584 if (IS_ERR_OR_NULL(gl)) {
2585 if (gl == ERR_PTR(-EAGAIN)) {
2592 if (gl->gl_name.ln_sbd != gi->sdp)
2595 if (!lockref_get_not_dead(&gl->gl_lockref))
2599 if (__lockref_is_dead(&gl->gl_lockref))
2607 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2610 struct gfs2_glock_iter *gi = seq->private;
2614 * We can either stay where we are, skip to the next hash table
2615 * entry, or start from the beginning.
2617 if (*pos < gi->last_pos) {
2618 rhashtable_walk_exit(&gi->hti);
2619 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2622 n = *pos - gi->last_pos;
2625 rhashtable_walk_start(&gi->hti);
2627 gfs2_glock_iter_next(gi, n);
2628 gi->last_pos = *pos;
2632 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2635 struct gfs2_glock_iter *gi = seq->private;
2638 gi->last_pos = *pos;
2639 gfs2_glock_iter_next(gi, 1);
2643 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2646 struct gfs2_glock_iter *gi = seq->private;
2648 rhashtable_walk_stop(&gi->hti);
2651 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2653 dump_glock(seq, iter_ptr, false);
2657 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2660 if (*pos >= GFS2_NR_SBSTATS)
2665 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2669 if (*pos >= GFS2_NR_SBSTATS)
2674 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2679 static const struct seq_operations gfs2_glock_seq_ops = {
2680 .start = gfs2_glock_seq_start,
2681 .next = gfs2_glock_seq_next,
2682 .stop = gfs2_glock_seq_stop,
2683 .show = gfs2_glock_seq_show,
2686 static const struct seq_operations gfs2_glstats_seq_ops = {
2687 .start = gfs2_glock_seq_start,
2688 .next = gfs2_glock_seq_next,
2689 .stop = gfs2_glock_seq_stop,
2690 .show = gfs2_glstats_seq_show,
2693 static const struct seq_operations gfs2_sbstats_sops = {
2694 .start = gfs2_sbstats_seq_start,
2695 .next = gfs2_sbstats_seq_next,
2696 .stop = gfs2_sbstats_seq_stop,
2697 .show = gfs2_sbstats_seq_show,
2700 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2702 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2703 const struct seq_operations *ops)
2705 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2707 struct seq_file *seq = file->private_data;
2708 struct gfs2_glock_iter *gi = seq->private;
2710 gi->sdp = inode->i_private;
2711 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2713 seq->size = GFS2_SEQ_GOODSIZE;
2715 * Initially, we are "before" the first hash table entry; the
2716 * first call to rhashtable_walk_next gets us the first entry.
2720 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2725 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2727 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2730 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2732 struct seq_file *seq = file->private_data;
2733 struct gfs2_glock_iter *gi = seq->private;
2736 gfs2_glock_put(gi->gl);
2737 rhashtable_walk_exit(&gi->hti);
2738 return seq_release_private(inode, file);
2741 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2743 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2746 static const struct file_operations gfs2_glocks_fops = {
2747 .owner = THIS_MODULE,
2748 .open = gfs2_glocks_open,
2750 .llseek = seq_lseek,
2751 .release = gfs2_glocks_release,
2754 static const struct file_operations gfs2_glstats_fops = {
2755 .owner = THIS_MODULE,
2756 .open = gfs2_glstats_open,
2758 .llseek = seq_lseek,
2759 .release = gfs2_glocks_release,
2762 struct gfs2_glockfd_iter {
2763 struct super_block *sb;
2765 struct task_struct *task;
2770 static struct task_struct *gfs2_glockfd_next_task(struct gfs2_glockfd_iter *i)
2772 struct pid_namespace *ns = task_active_pid_ns(current);
2776 put_task_struct(i->task);
2781 pid = find_ge_pid(i->tgid, ns);
2783 i->tgid = pid_nr_ns(pid, ns);
2784 i->task = pid_task(pid, PIDTYPE_TGID);
2789 get_task_struct(i->task);
2795 static struct file *gfs2_glockfd_next_file(struct gfs2_glockfd_iter *i)
2804 struct inode *inode;
2806 i->file = task_lookup_next_fd_rcu(i->task, &i->fd);
2811 inode = file_inode(i->file);
2812 if (inode->i_sb != i->sb)
2814 if (get_file_rcu(i->file))
2821 static void *gfs2_glockfd_seq_start(struct seq_file *seq, loff_t *pos)
2823 struct gfs2_glockfd_iter *i = seq->private;
2827 while (gfs2_glockfd_next_task(i)) {
2828 if (gfs2_glockfd_next_file(i))
2835 static void *gfs2_glockfd_seq_next(struct seq_file *seq, void *iter_ptr,
2838 struct gfs2_glockfd_iter *i = seq->private;
2843 if (gfs2_glockfd_next_file(i))
2846 } while (gfs2_glockfd_next_task(i));
2850 static void gfs2_glockfd_seq_stop(struct seq_file *seq, void *iter_ptr)
2852 struct gfs2_glockfd_iter *i = seq->private;
2857 put_task_struct(i->task);
2860 static void gfs2_glockfd_seq_show_flock(struct seq_file *seq,
2861 struct gfs2_glockfd_iter *i)
2863 struct gfs2_file *fp = i->file->private_data;
2864 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
2865 struct lm_lockname gl_name = { .ln_type = LM_TYPE_RESERVED };
2867 if (!READ_ONCE(fl_gh->gh_gl))
2870 spin_lock(&i->file->f_lock);
2871 if (gfs2_holder_initialized(fl_gh))
2872 gl_name = fl_gh->gh_gl->gl_name;
2873 spin_unlock(&i->file->f_lock);
2875 if (gl_name.ln_type != LM_TYPE_RESERVED) {
2876 seq_printf(seq, "%d %u %u/%llx\n",
2877 i->tgid, i->fd, gl_name.ln_type,
2878 (unsigned long long)gl_name.ln_number);
2882 static int gfs2_glockfd_seq_show(struct seq_file *seq, void *iter_ptr)
2884 struct gfs2_glockfd_iter *i = seq->private;
2885 struct inode *inode = file_inode(i->file);
2886 struct gfs2_glock *gl;
2888 inode_lock_shared(inode);
2889 gl = GFS2_I(inode)->i_iopen_gh.gh_gl;
2891 seq_printf(seq, "%d %u %u/%llx\n",
2892 i->tgid, i->fd, gl->gl_name.ln_type,
2893 (unsigned long long)gl->gl_name.ln_number);
2895 gfs2_glockfd_seq_show_flock(seq, i);
2896 inode_unlock_shared(inode);
2900 static const struct seq_operations gfs2_glockfd_seq_ops = {
2901 .start = gfs2_glockfd_seq_start,
2902 .next = gfs2_glockfd_seq_next,
2903 .stop = gfs2_glockfd_seq_stop,
2904 .show = gfs2_glockfd_seq_show,
2907 static int gfs2_glockfd_open(struct inode *inode, struct file *file)
2909 struct gfs2_glockfd_iter *i;
2910 struct gfs2_sbd *sdp = inode->i_private;
2912 i = __seq_open_private(file, &gfs2_glockfd_seq_ops,
2913 sizeof(struct gfs2_glockfd_iter));
2916 i->sb = sdp->sd_vfs;
2920 static const struct file_operations gfs2_glockfd_fops = {
2921 .owner = THIS_MODULE,
2922 .open = gfs2_glockfd_open,
2924 .llseek = seq_lseek,
2925 .release = seq_release_private,
2928 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2930 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2932 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2934 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2937 debugfs_create_file("glockfd", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2938 &gfs2_glockfd_fops);
2940 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2941 &gfs2_glstats_fops);
2943 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2944 &gfs2_sbstats_fops);
2947 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2949 debugfs_remove_recursive(sdp->debugfs_dir);
2950 sdp->debugfs_dir = NULL;
2953 void gfs2_register_debugfs(void)
2955 gfs2_root = debugfs_create_dir("gfs2", NULL);
2958 void gfs2_unregister_debugfs(void)
2960 debugfs_remove(gfs2_root);