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 struct gfs2_glock *gfs2_glock_hold(struct gfs2_glock *gl)
191 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
192 lockref_get(&gl->gl_lockref);
197 * demote_ok - Check to see if it's ok to unlock a glock
200 * Returns: 1 if it's ok
203 static int demote_ok(const struct gfs2_glock *gl)
205 const struct gfs2_glock_operations *glops = gl->gl_ops;
207 if (gl->gl_state == LM_ST_UNLOCKED)
209 if (!list_empty(&gl->gl_holders))
211 if (glops->go_demote_ok)
212 return glops->go_demote_ok(gl);
217 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
219 if (!(gl->gl_ops->go_flags & GLOF_LRU))
222 spin_lock(&lru_lock);
224 list_move_tail(&gl->gl_lru, &lru_list);
226 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
227 set_bit(GLF_LRU, &gl->gl_flags);
228 atomic_inc(&lru_count);
231 spin_unlock(&lru_lock);
234 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
236 if (!(gl->gl_ops->go_flags & GLOF_LRU))
239 spin_lock(&lru_lock);
240 if (test_bit(GLF_LRU, &gl->gl_flags)) {
241 list_del_init(&gl->gl_lru);
242 atomic_dec(&lru_count);
243 clear_bit(GLF_LRU, &gl->gl_flags);
245 spin_unlock(&lru_lock);
249 * Enqueue the glock on the work queue. Passes one glock reference on to the
252 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
253 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
255 * We are holding the lockref spinlock, and the work was still
256 * queued above. The queued work (glock_work_func) takes that
257 * spinlock before dropping its glock reference(s), so it
258 * cannot have dropped them in the meantime.
260 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
261 gl->gl_lockref.count--;
265 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
266 spin_lock(&gl->gl_lockref.lock);
267 __gfs2_glock_queue_work(gl, delay);
268 spin_unlock(&gl->gl_lockref.lock);
271 static void __gfs2_glock_put(struct gfs2_glock *gl)
273 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
274 struct address_space *mapping = gfs2_glock2aspace(gl);
276 lockref_mark_dead(&gl->gl_lockref);
278 gfs2_glock_remove_from_lru(gl);
279 spin_unlock(&gl->gl_lockref.lock);
280 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
282 truncate_inode_pages_final(mapping);
283 if (!gfs2_withdrawn(sdp))
284 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
286 trace_gfs2_glock_put(gl);
287 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
291 * Cause the glock to be put in work queue context.
293 void gfs2_glock_queue_put(struct gfs2_glock *gl)
295 gfs2_glock_queue_work(gl, 0);
299 * gfs2_glock_put() - Decrement reference count on glock
300 * @gl: The glock to put
304 void gfs2_glock_put(struct gfs2_glock *gl)
306 if (lockref_put_or_lock(&gl->gl_lockref))
309 __gfs2_glock_put(gl);
313 * may_grant - check if it's ok to grant a new lock
315 * @current_gh: One of the current holders of @gl
316 * @gh: The lock request which we wish to grant
318 * With our current compatibility rules, if a glock has one or more active
319 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
320 * @current_gh; they are all the same as far as compatibility with the new @gh
323 * Returns true if it's ok to grant the lock.
326 static inline bool may_grant(struct gfs2_glock *gl,
327 struct gfs2_holder *current_gh,
328 struct gfs2_holder *gh)
331 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
333 switch(current_gh->gh_state) {
334 case LM_ST_EXCLUSIVE:
336 * Here we make a special exception to grant holders
337 * who agree to share the EX lock with other holders
338 * who also have the bit set. If the original holder
339 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
340 * holders with the bit set.
342 return gh->gh_state == LM_ST_EXCLUSIVE &&
343 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
344 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
348 return gh->gh_state == current_gh->gh_state;
355 if (gl->gl_state == gh->gh_state)
357 if (gh->gh_flags & GL_EXACT)
359 if (gl->gl_state == LM_ST_EXCLUSIVE) {
360 return gh->gh_state == LM_ST_SHARED ||
361 gh->gh_state == LM_ST_DEFERRED;
363 if (gh->gh_flags & LM_FLAG_ANY)
364 return gl->gl_state != LM_ST_UNLOCKED;
368 static void gfs2_holder_wake(struct gfs2_holder *gh)
370 clear_bit(HIF_WAIT, &gh->gh_iflags);
371 smp_mb__after_atomic();
372 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
373 if (gh->gh_flags & GL_ASYNC) {
374 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
376 wake_up(&sdp->sd_async_glock_wait);
381 * do_error - Something unexpected has happened during a lock request
383 * @ret: The status from the DLM
386 static void do_error(struct gfs2_glock *gl, const int ret)
388 struct gfs2_holder *gh, *tmp;
390 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
391 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
393 if (ret & LM_OUT_ERROR)
395 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
396 gh->gh_error = GLR_TRYFAILED;
399 list_del_init(&gh->gh_list);
400 trace_gfs2_glock_queue(gh, 0);
401 gfs2_holder_wake(gh);
406 * find_first_holder - find the first "holder" gh
410 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
412 struct gfs2_holder *gh;
414 if (!list_empty(&gl->gl_holders)) {
415 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
417 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
424 * gfs2_instantiate - Call the glops instantiate function
425 * @gh: The glock holder
427 * Returns: 0 if instantiate was successful, or error.
429 int gfs2_instantiate(struct gfs2_holder *gh)
431 struct gfs2_glock *gl = gh->gh_gl;
432 const struct gfs2_glock_operations *glops = gl->gl_ops;
436 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
440 * Since we unlock the lockref lock, we set a flag to indicate
441 * instantiate is in progress.
443 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
444 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
445 TASK_UNINTERRUPTIBLE);
447 * Here we just waited for a different instantiate to finish.
448 * But that may not have been successful, as when a process
449 * locks an inode glock _before_ it has an actual inode to
450 * instantiate into. So we check again. This process might
451 * have an inode to instantiate, so might be successful.
456 ret = glops->go_instantiate(gl);
458 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
459 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
465 return glops->go_held(gh);
470 * do_promote - promote as many requests as possible on the current queue
473 * Returns: 1 if there is a blocked holder at the head of the list
476 static int do_promote(struct gfs2_glock *gl)
478 struct gfs2_holder *gh, *current_gh;
480 current_gh = find_first_holder(gl);
481 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
482 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
484 if (!may_grant(gl, current_gh, gh)) {
486 * If we get here, it means we may not grant this
487 * holder for some reason. If this holder is at the
488 * head of the list, it means we have a blocked holder
489 * at the head, so return 1.
491 if (list_is_first(&gh->gh_list, &gl->gl_holders))
496 set_bit(HIF_HOLDER, &gh->gh_iflags);
497 trace_gfs2_promote(gh);
498 gfs2_holder_wake(gh);
506 * find_first_waiter - find the first gh that's waiting for the glock
510 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
512 struct gfs2_holder *gh;
514 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
515 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
522 * state_change - record that the glock is now in a different state
524 * @new_state: the new state
527 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
531 held1 = (gl->gl_state != LM_ST_UNLOCKED);
532 held2 = (new_state != LM_ST_UNLOCKED);
534 if (held1 != held2) {
535 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
537 gl->gl_lockref.count++;
539 gl->gl_lockref.count--;
541 if (new_state != gl->gl_target)
542 /* shorten our minimum hold time */
543 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
545 gl->gl_state = new_state;
546 gl->gl_tchange = jiffies;
549 static void gfs2_set_demote(struct gfs2_glock *gl)
551 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
553 set_bit(GLF_DEMOTE, &gl->gl_flags);
555 wake_up(&sdp->sd_async_glock_wait);
558 static void gfs2_demote_wake(struct gfs2_glock *gl)
560 gl->gl_demote_state = LM_ST_EXCLUSIVE;
561 clear_bit(GLF_DEMOTE, &gl->gl_flags);
562 smp_mb__after_atomic();
563 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
567 * finish_xmote - The DLM has replied to one of our lock requests
569 * @ret: The status from the DLM
573 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
575 const struct gfs2_glock_operations *glops = gl->gl_ops;
576 struct gfs2_holder *gh;
577 unsigned state = ret & LM_OUT_ST_MASK;
579 spin_lock(&gl->gl_lockref.lock);
580 trace_gfs2_glock_state_change(gl, state);
581 state_change(gl, state);
582 gh = find_first_waiter(gl);
584 /* Demote to UN request arrived during demote to SH or DF */
585 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
586 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
587 gl->gl_target = LM_ST_UNLOCKED;
589 /* Check for state != intended state */
590 if (unlikely(state != gl->gl_target)) {
591 if (gh && (ret & LM_OUT_CANCELED))
592 gfs2_holder_wake(gh);
593 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
594 /* move to back of queue and try next entry */
595 if (ret & LM_OUT_CANCELED) {
596 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
597 list_move_tail(&gh->gh_list, &gl->gl_holders);
598 gh = find_first_waiter(gl);
599 gl->gl_target = gh->gh_state;
602 /* Some error or failed "try lock" - report it */
603 if ((ret & LM_OUT_ERROR) ||
604 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
605 gl->gl_target = gl->gl_state;
611 /* Unlocked due to conversion deadlock, try again */
614 do_xmote(gl, gh, gl->gl_target);
616 /* Conversion fails, unlock and try again */
619 do_xmote(gl, gh, LM_ST_UNLOCKED);
621 default: /* Everything else */
622 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
623 gl->gl_target, state);
626 spin_unlock(&gl->gl_lockref.lock);
630 /* Fast path - we got what we asked for */
631 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
632 gfs2_demote_wake(gl);
633 if (state != LM_ST_UNLOCKED) {
634 if (glops->go_xmote_bh) {
637 spin_unlock(&gl->gl_lockref.lock);
638 rv = glops->go_xmote_bh(gl);
639 spin_lock(&gl->gl_lockref.lock);
648 clear_bit(GLF_LOCK, &gl->gl_flags);
649 spin_unlock(&gl->gl_lockref.lock);
652 static bool is_system_glock(struct gfs2_glock *gl)
654 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
655 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
657 if (gl == m_ip->i_gl)
663 * do_xmote - Calls the DLM to change the state of a lock
664 * @gl: The lock state
665 * @gh: The holder (only for promotes)
666 * @target: The target lock state
670 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh,
672 __releases(&gl->gl_lockref.lock)
673 __acquires(&gl->gl_lockref.lock)
675 const struct gfs2_glock_operations *glops = gl->gl_ops;
676 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
677 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
680 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
681 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
684 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
686 GLOCK_BUG_ON(gl, gl->gl_state == target);
687 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
688 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
691 * If another process is already doing the invalidate, let that
692 * finish first. The glock state machine will get back to this
693 * holder again later.
695 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
698 do_error(gl, 0); /* Fail queued try locks */
701 set_bit(GLF_BLOCKING, &gl->gl_flags);
702 if ((gl->gl_req == LM_ST_UNLOCKED) ||
703 (gl->gl_state == LM_ST_EXCLUSIVE) ||
704 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
705 clear_bit(GLF_BLOCKING, &gl->gl_flags);
706 spin_unlock(&gl->gl_lockref.lock);
707 if (glops->go_sync) {
708 ret = glops->go_sync(gl);
709 /* If we had a problem syncing (due to io errors or whatever,
710 * we should not invalidate the metadata or tell dlm to
711 * release the glock to other nodes.
714 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
715 fs_err(sdp, "Error %d syncing glock \n", ret);
716 gfs2_dump_glock(NULL, gl, true);
721 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
723 * The call to go_sync should have cleared out the ail list.
724 * If there are still items, we have a problem. We ought to
725 * withdraw, but we can't because the withdraw code also uses
726 * glocks. Warn about the error, dump the glock, then fall
727 * through and wait for logd to do the withdraw for us.
729 if ((atomic_read(&gl->gl_ail_count) != 0) &&
730 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
731 gfs2_glock_assert_warn(gl,
732 !atomic_read(&gl->gl_ail_count));
733 gfs2_dump_glock(NULL, gl, true);
735 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
736 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
742 * Check for an error encountered since we called go_sync and go_inval.
743 * If so, we can't withdraw from the glock code because the withdraw
744 * code itself uses glocks (see function signal_our_withdraw) to
745 * change the mount to read-only. Most importantly, we must not call
746 * dlm to unlock the glock until the journal is in a known good state
747 * (after journal replay) otherwise other nodes may use the object
748 * (rgrp or dinode) and then later, journal replay will corrupt the
749 * file system. The best we can do here is wait for the logd daemon
750 * to see sd_log_error and withdraw, and in the meantime, requeue the
753 * We make a special exception for some system glocks, such as the
754 * system statfs inode glock, which needs to be granted before the
755 * gfs2_quotad daemon can exit, and that exit needs to finish before
756 * we can unmount the withdrawn file system.
758 * However, if we're just unlocking the lock (say, for unmount, when
759 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
760 * then it's okay to tell dlm to unlock it.
762 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
763 gfs2_withdraw_delayed(sdp);
764 if (glock_blocked_by_withdraw(gl) &&
765 (target != LM_ST_UNLOCKED ||
766 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
767 if (!is_system_glock(gl)) {
768 handle_callback(gl, LM_ST_UNLOCKED, 0, false); /* sets demote */
770 * Ordinarily, we would call dlm and its callback would call
771 * finish_xmote, which would call state_change() to the new state.
772 * Since we withdrew, we won't call dlm, so call state_change
773 * manually, but to the UNLOCKED state we desire.
775 state_change(gl, LM_ST_UNLOCKED);
777 * We skip telling dlm to do the locking, so we won't get a
778 * reply that would otherwise clear GLF_LOCK. So we clear it here.
780 clear_bit(GLF_LOCK, &gl->gl_flags);
781 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
782 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
785 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
789 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
791 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
792 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
793 target == LM_ST_UNLOCKED &&
794 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
795 finish_xmote(gl, target);
796 gfs2_glock_queue_work(gl, 0);
798 fs_err(sdp, "lm_lock ret %d\n", ret);
799 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
801 } else { /* lock_nolock */
802 finish_xmote(gl, target);
803 gfs2_glock_queue_work(gl, 0);
806 spin_lock(&gl->gl_lockref.lock);
810 * run_queue - do all outstanding tasks related to a glock
811 * @gl: The glock in question
812 * @nonblock: True if we must not block in run_queue
816 static void run_queue(struct gfs2_glock *gl, const int nonblock)
817 __releases(&gl->gl_lockref.lock)
818 __acquires(&gl->gl_lockref.lock)
820 struct gfs2_holder *gh = NULL;
822 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
825 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
827 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
828 gl->gl_demote_state != gl->gl_state) {
829 if (find_first_holder(gl))
833 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
834 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
835 gl->gl_target = gl->gl_demote_state;
837 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
838 gfs2_demote_wake(gl);
839 if (do_promote(gl) == 0)
841 gh = find_first_waiter(gl);
842 gl->gl_target = gh->gh_state;
843 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
844 do_error(gl, 0); /* Fail queued try locks */
846 do_xmote(gl, gh, gl->gl_target);
850 clear_bit(GLF_LOCK, &gl->gl_flags);
851 smp_mb__after_atomic();
852 gl->gl_lockref.count++;
853 __gfs2_glock_queue_work(gl, 0);
857 clear_bit(GLF_LOCK, &gl->gl_flags);
858 smp_mb__after_atomic();
863 * glock_set_object - set the gl_object field of a glock
865 * @object: the object
867 void glock_set_object(struct gfs2_glock *gl, void *object)
871 spin_lock(&gl->gl_lockref.lock);
872 prev_object = gl->gl_object;
873 gl->gl_object = object;
874 spin_unlock(&gl->gl_lockref.lock);
875 if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == NULL)) {
876 pr_warn("glock=%u/%llx\n",
878 (unsigned long long)gl->gl_name.ln_number);
879 gfs2_dump_glock(NULL, gl, true);
884 * glock_clear_object - clear the gl_object field of a glock
887 void glock_clear_object(struct gfs2_glock *gl, void *object)
891 spin_lock(&gl->gl_lockref.lock);
892 prev_object = gl->gl_object;
893 gl->gl_object = NULL;
894 spin_unlock(&gl->gl_lockref.lock);
895 if (gfs2_assert_warn(gl->gl_name.ln_sbd,
896 prev_object == object || prev_object == NULL)) {
897 pr_warn("glock=%u/%llx\n",
899 (unsigned long long)gl->gl_name.ln_number);
900 gfs2_dump_glock(NULL, gl, true);
904 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
906 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
908 if (ri->ri_magic == 0)
909 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
910 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
911 ri->ri_generation_deleted = cpu_to_be64(generation);
914 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
916 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
918 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
920 return generation <= be64_to_cpu(ri->ri_generation_deleted);
923 static void gfs2_glock_poke(struct gfs2_glock *gl)
925 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
926 struct gfs2_holder gh;
929 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
930 error = gfs2_glock_nq(&gh);
933 gfs2_holder_uninit(&gh);
936 static bool gfs2_try_evict(struct gfs2_glock *gl)
938 struct gfs2_inode *ip;
939 bool evicted = false;
942 * If there is contention on the iopen glock and we have an inode, try
943 * to grab and release the inode so that it can be evicted. This will
944 * allow the remote node to go ahead and delete the inode without us
945 * having to do it, which will avoid rgrp glock thrashing.
947 * The remote node is likely still holding the corresponding inode
948 * glock, so it will run before we get to verify that the delete has
951 spin_lock(&gl->gl_lockref.lock);
953 if (ip && !igrab(&ip->i_inode))
955 spin_unlock(&gl->gl_lockref.lock);
957 gl->gl_no_formal_ino = ip->i_no_formal_ino;
958 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
959 d_prune_aliases(&ip->i_inode);
962 /* If the inode was evicted, gl->gl_object will now be NULL. */
963 spin_lock(&gl->gl_lockref.lock);
966 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
967 if (!igrab(&ip->i_inode))
970 spin_unlock(&gl->gl_lockref.lock);
972 gfs2_glock_poke(ip->i_gl);
980 static void delete_work_func(struct work_struct *work)
982 struct delayed_work *dwork = to_delayed_work(work);
983 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
984 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
986 u64 no_addr = gl->gl_name.ln_number;
988 spin_lock(&gl->gl_lockref.lock);
989 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
990 spin_unlock(&gl->gl_lockref.lock);
992 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
994 * If we can evict the inode, give the remote node trying to
995 * delete the inode some time before verifying that the delete
996 * has happened. Otherwise, if we cause contention on the inode glock
997 * immediately, the remote node will think that we still have
998 * the inode in use, and so it will give up waiting.
1000 * If we can't evict the inode, signal to the remote node that
1001 * the inode is still in use. We'll later try to delete the
1002 * inode locally in gfs2_evict_inode.
1004 * FIXME: We only need to verify that the remote node has
1005 * deleted the inode because nodes before this remote delete
1006 * rework won't cooperate. At a later time, when we no longer
1007 * care about compatibility with such nodes, we can skip this
1010 if (gfs2_try_evict(gl)) {
1011 if (gfs2_queue_delete_work(gl, 5 * HZ))
1017 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1018 GFS2_BLKST_UNLINKED);
1019 if (IS_ERR(inode)) {
1020 if (PTR_ERR(inode) == -EAGAIN &&
1021 (gfs2_queue_delete_work(gl, 5 * HZ)))
1024 d_prune_aliases(inode);
1031 static void glock_work_func(struct work_struct *work)
1033 unsigned long delay = 0;
1034 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1035 unsigned int drop_refs = 1;
1037 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1038 finish_xmote(gl, gl->gl_reply);
1041 spin_lock(&gl->gl_lockref.lock);
1042 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1043 gl->gl_state != LM_ST_UNLOCKED &&
1044 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1045 unsigned long holdtime, now = jiffies;
1047 holdtime = gl->gl_tchange + gl->gl_hold_time;
1048 if (time_before(now, holdtime))
1049 delay = holdtime - now;
1052 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1053 gfs2_set_demote(gl);
1058 /* Keep one glock reference for the work we requeue. */
1060 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1062 __gfs2_glock_queue_work(gl, delay);
1066 * Drop the remaining glock references manually here. (Mind that
1067 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1070 gl->gl_lockref.count -= drop_refs;
1071 if (!gl->gl_lockref.count) {
1072 __gfs2_glock_put(gl);
1075 spin_unlock(&gl->gl_lockref.lock);
1078 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1079 struct gfs2_glock *new)
1081 struct wait_glock_queue wait;
1082 wait_queue_head_t *wq = glock_waitqueue(name);
1083 struct gfs2_glock *gl;
1086 init_wait(&wait.wait);
1087 wait.wait.func = glock_wake_function;
1090 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1093 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1094 &new->gl_node, ht_parms);
1098 gl = rhashtable_lookup_fast(&gl_hash_table,
1101 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1108 finish_wait(wq, &wait.wait);
1113 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1114 * @sdp: The GFS2 superblock
1115 * @number: the lock number
1116 * @glops: The glock_operations to use
1117 * @create: If 0, don't create the glock if it doesn't exist
1118 * @glp: the glock is returned here
1120 * This does not lock a glock, just finds/creates structures for one.
1125 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1126 const struct gfs2_glock_operations *glops, int create,
1127 struct gfs2_glock **glp)
1129 struct super_block *s = sdp->sd_vfs;
1130 struct lm_lockname name = { .ln_number = number,
1131 .ln_type = glops->go_type,
1133 struct gfs2_glock *gl, *tmp;
1134 struct address_space *mapping;
1137 gl = find_insert_glock(&name, NULL);
1145 if (glops->go_flags & GLOF_ASPACE) {
1146 struct gfs2_glock_aspace *gla =
1147 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1152 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1156 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1159 if (glops->go_flags & GLOF_LVB) {
1160 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1161 if (!gl->gl_lksb.sb_lvbptr) {
1162 gfs2_glock_dealloc(&gl->gl_rcu);
1167 atomic_inc(&sdp->sd_glock_disposal);
1168 gl->gl_node.next = NULL;
1169 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1171 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1172 gl->gl_lockref.count = 1;
1173 gl->gl_state = LM_ST_UNLOCKED;
1174 gl->gl_target = LM_ST_UNLOCKED;
1175 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1178 /* We use the global stats to estimate the initial per-glock stats */
1179 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1181 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1182 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1183 gl->gl_tchange = jiffies;
1184 gl->gl_object = NULL;
1185 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1186 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1187 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1188 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1190 mapping = gfs2_glock2aspace(gl);
1192 mapping->a_ops = &gfs2_meta_aops;
1193 mapping->host = s->s_bdev->bd_inode;
1195 mapping_set_gfp_mask(mapping, GFP_NOFS);
1196 mapping->private_data = NULL;
1197 mapping->writeback_index = 0;
1200 tmp = find_insert_glock(&name, gl);
1212 gfs2_glock_dealloc(&gl->gl_rcu);
1213 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1214 wake_up(&sdp->sd_glock_wait);
1221 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1223 * @state: the state we're requesting
1224 * @flags: the modifier flags
1225 * @gh: the holder structure
1229 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1230 struct gfs2_holder *gh, unsigned long ip)
1232 INIT_LIST_HEAD(&gh->gh_list);
1233 gh->gh_gl = gfs2_glock_hold(gl);
1235 gh->gh_owner_pid = get_pid(task_pid(current));
1236 gh->gh_state = state;
1237 gh->gh_flags = flags;
1242 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1243 * @state: the state we're requesting
1244 * @flags: the modifier flags
1245 * @gh: the holder structure
1247 * Don't mess with the glock.
1251 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1253 gh->gh_state = state;
1254 gh->gh_flags = flags;
1256 gh->gh_ip = _RET_IP_;
1257 put_pid(gh->gh_owner_pid);
1258 gh->gh_owner_pid = get_pid(task_pid(current));
1262 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1263 * @gh: the holder structure
1267 void gfs2_holder_uninit(struct gfs2_holder *gh)
1269 put_pid(gh->gh_owner_pid);
1270 gfs2_glock_put(gh->gh_gl);
1271 gfs2_holder_mark_uninitialized(gh);
1275 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1276 unsigned long start_time)
1278 /* Have we waited longer that a second? */
1279 if (time_after(jiffies, start_time + HZ)) {
1280 /* Lengthen the minimum hold time. */
1281 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1287 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1288 * @gh: the glock holder
1290 * Called when a glock holder no longer needs to be waited for because it is
1291 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1292 * failed (gh_error != 0).
1295 int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1297 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1298 return gh->gh_error;
1299 gh->gh_error = gfs2_instantiate(gh);
1302 return gh->gh_error;
1306 * gfs2_glock_wait - wait on a glock acquisition
1307 * @gh: the glock holder
1309 * Returns: 0 on success
1312 int gfs2_glock_wait(struct gfs2_holder *gh)
1314 unsigned long start_time = jiffies;
1317 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1318 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1319 return gfs2_glock_holder_ready(gh);
1322 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1326 for (i = 0; i < num_gh; i++)
1327 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1333 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1334 * @num_gh: the number of holders in the array
1335 * @ghs: the glock holder array
1337 * Returns: 0 on success, meaning all glocks have been granted and are held.
1338 * -ESTALE if the request timed out, meaning all glocks were released,
1339 * and the caller should retry the operation.
1342 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1344 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1345 int i, ret = 0, timeout = 0;
1346 unsigned long start_time = jiffies;
1350 * Total up the (minimum hold time * 2) of all glocks and use that to
1351 * determine the max amount of time we should wait.
1353 for (i = 0; i < num_gh; i++)
1354 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1356 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1357 !glocks_pending(num_gh, ghs), timeout)) {
1358 ret = -ESTALE; /* request timed out. */
1362 for (i = 0; i < num_gh; i++) {
1363 struct gfs2_holder *gh = &ghs[i];
1366 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1367 gfs2_glock_update_hold_time(gh->gh_gl,
1370 ret2 = gfs2_glock_holder_ready(gh);
1377 for (i = 0; i < num_gh; i++) {
1378 struct gfs2_holder *gh = &ghs[i];
1387 * handle_callback - process a demote request
1389 * @state: the state the caller wants us to change to
1390 * @delay: zero to demote immediately; otherwise pending demote
1391 * @remote: true if this came from a different cluster node
1393 * There are only two requests that we are going to see in actual
1394 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1397 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1398 unsigned long delay, bool remote)
1401 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1403 gfs2_set_demote(gl);
1404 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1405 gl->gl_demote_state = state;
1406 gl->gl_demote_time = jiffies;
1407 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1408 gl->gl_demote_state != state) {
1409 gl->gl_demote_state = LM_ST_UNLOCKED;
1411 if (gl->gl_ops->go_callback)
1412 gl->gl_ops->go_callback(gl, remote);
1413 trace_gfs2_demote_rq(gl, remote);
1416 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1418 struct va_format vaf;
1421 va_start(args, fmt);
1424 seq_vprintf(seq, fmt, args);
1429 pr_err("%pV", &vaf);
1435 static inline bool pid_is_meaningful(const struct gfs2_holder *gh)
1437 if (!(gh->gh_flags & GL_NOPID))
1439 if (gh->gh_state == LM_ST_UNLOCKED)
1445 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1446 * @gh: the holder structure to add
1448 * Eventually we should move the recursive locking trap to a
1449 * debugging option or something like that. This is the fast
1450 * path and needs to have the minimum number of distractions.
1454 static inline void add_to_queue(struct gfs2_holder *gh)
1455 __releases(&gl->gl_lockref.lock)
1456 __acquires(&gl->gl_lockref.lock)
1458 struct gfs2_glock *gl = gh->gh_gl;
1459 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1460 struct list_head *insert_pt = NULL;
1461 struct gfs2_holder *gh2;
1464 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1465 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1466 GLOCK_BUG_ON(gl, true);
1468 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1469 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1470 struct gfs2_holder *current_gh;
1472 current_gh = find_first_holder(gl);
1473 try_futile = !may_grant(gl, current_gh, gh);
1475 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1479 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1480 if (likely(gh2->gh_owner_pid != gh->gh_owner_pid))
1482 if (gh->gh_gl->gl_ops->go_type == LM_TYPE_FLOCK)
1484 if (!pid_is_meaningful(gh2))
1486 goto trap_recursive;
1488 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1490 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1492 gh->gh_error = GLR_TRYFAILED;
1493 gfs2_holder_wake(gh);
1496 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1498 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1499 insert_pt = &gh2->gh_list;
1501 trace_gfs2_glock_queue(gh, 1);
1502 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1503 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1504 if (likely(insert_pt == NULL)) {
1505 list_add_tail(&gh->gh_list, &gl->gl_holders);
1506 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1510 list_add_tail(&gh->gh_list, insert_pt);
1512 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1513 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1514 spin_unlock(&gl->gl_lockref.lock);
1515 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1516 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1517 spin_lock(&gl->gl_lockref.lock);
1522 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1523 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1524 fs_err(sdp, "lock type: %d req lock state : %d\n",
1525 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1526 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1527 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1528 fs_err(sdp, "lock type: %d req lock state : %d\n",
1529 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1530 gfs2_dump_glock(NULL, gl, true);
1535 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1536 * @gh: the holder structure
1538 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1540 * Returns: 0, GLR_TRYFAILED, or errno on failure
1543 int gfs2_glock_nq(struct gfs2_holder *gh)
1545 struct gfs2_glock *gl = gh->gh_gl;
1548 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1551 if (test_bit(GLF_LRU, &gl->gl_flags))
1552 gfs2_glock_remove_from_lru(gl);
1555 spin_lock(&gl->gl_lockref.lock);
1557 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1558 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1559 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1560 gl->gl_lockref.count++;
1561 __gfs2_glock_queue_work(gl, 0);
1564 spin_unlock(&gl->gl_lockref.lock);
1566 if (!(gh->gh_flags & GL_ASYNC))
1567 error = gfs2_glock_wait(gh);
1573 * gfs2_glock_poll - poll to see if an async request has been completed
1576 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1579 int gfs2_glock_poll(struct gfs2_holder *gh)
1581 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1584 static inline bool needs_demote(struct gfs2_glock *gl)
1586 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1587 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1590 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1592 struct gfs2_glock *gl = gh->gh_gl;
1597 * This holder should not be cached, so mark it for demote.
1598 * Note: this should be done before the check for needs_demote
1601 if (gh->gh_flags & GL_NOCACHE)
1602 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1604 list_del_init(&gh->gh_list);
1605 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1606 trace_gfs2_glock_queue(gh, 0);
1609 * If there hasn't been a demote request we are done.
1610 * (Let the remaining holders, if any, keep holding it.)
1612 if (!needs_demote(gl)) {
1613 if (list_empty(&gl->gl_holders))
1617 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1618 gfs2_glock_add_to_lru(gl);
1620 if (unlikely(!fast_path)) {
1621 gl->gl_lockref.count++;
1622 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1623 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1624 gl->gl_name.ln_type == LM_TYPE_INODE)
1625 delay = gl->gl_hold_time;
1626 __gfs2_glock_queue_work(gl, delay);
1631 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1632 * @gh: the glock holder
1635 void gfs2_glock_dq(struct gfs2_holder *gh)
1637 struct gfs2_glock *gl = gh->gh_gl;
1638 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1640 spin_lock(&gl->gl_lockref.lock);
1641 if (!gfs2_holder_queued(gh)) {
1643 * May have already been dequeued because the locking request
1644 * was GL_ASYNC and it has failed in the meantime.
1649 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1650 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1651 spin_unlock(&gl->gl_lockref.lock);
1652 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1653 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1654 spin_lock(&gl->gl_lockref.lock);
1658 * If we're in the process of file system withdraw, we cannot just
1659 * dequeue any glocks until our journal is recovered, lest we introduce
1660 * file system corruption. We need two exceptions to this rule: We need
1661 * to allow unlocking of nondisk glocks and the glock for our own
1662 * journal that needs recovery.
1664 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1665 glock_blocked_by_withdraw(gl) &&
1666 gh->gh_gl != sdp->sd_jinode_gl) {
1667 sdp->sd_glock_dqs_held++;
1668 spin_unlock(&gl->gl_lockref.lock);
1670 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1671 TASK_UNINTERRUPTIBLE);
1672 spin_lock(&gl->gl_lockref.lock);
1675 __gfs2_glock_dq(gh);
1677 spin_unlock(&gl->gl_lockref.lock);
1680 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1682 struct gfs2_glock *gl = gh->gh_gl;
1685 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1689 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1690 * @gh: the holder structure
1694 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1697 gfs2_holder_uninit(gh);
1701 * gfs2_glock_nq_num - acquire a glock based on lock number
1702 * @sdp: the filesystem
1703 * @number: the lock number
1704 * @glops: the glock operations for the type of glock
1705 * @state: the state to acquire the glock in
1706 * @flags: modifier flags for the acquisition
1707 * @gh: the struct gfs2_holder
1712 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1713 const struct gfs2_glock_operations *glops,
1714 unsigned int state, u16 flags, struct gfs2_holder *gh)
1716 struct gfs2_glock *gl;
1719 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1721 error = gfs2_glock_nq_init(gl, state, flags, gh);
1729 * glock_compare - Compare two struct gfs2_glock structures for sorting
1730 * @arg_a: the first structure
1731 * @arg_b: the second structure
1735 static int glock_compare(const void *arg_a, const void *arg_b)
1737 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1738 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1739 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1740 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1742 if (a->ln_number > b->ln_number)
1744 if (a->ln_number < b->ln_number)
1746 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1751 * nq_m_sync - synchronously acquire more than one glock in deadlock free order
1752 * @num_gh: the number of structures
1753 * @ghs: an array of struct gfs2_holder structures
1754 * @p: placeholder for the holder structure to pass back
1756 * Returns: 0 on success (all glocks acquired),
1757 * errno on failure (no glocks acquired)
1760 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1761 struct gfs2_holder **p)
1766 for (x = 0; x < num_gh; x++)
1769 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1771 for (x = 0; x < num_gh; x++) {
1772 error = gfs2_glock_nq(p[x]);
1775 gfs2_glock_dq(p[x]);
1784 * gfs2_glock_nq_m - acquire multiple glocks
1785 * @num_gh: the number of structures
1786 * @ghs: an array of struct gfs2_holder structures
1788 * Returns: 0 on success (all glocks acquired),
1789 * errno on failure (no glocks acquired)
1792 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1794 struct gfs2_holder *tmp[4];
1795 struct gfs2_holder **pph = tmp;
1802 return gfs2_glock_nq(ghs);
1806 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1812 error = nq_m_sync(num_gh, ghs, pph);
1821 * gfs2_glock_dq_m - release multiple glocks
1822 * @num_gh: the number of structures
1823 * @ghs: an array of struct gfs2_holder structures
1827 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1830 gfs2_glock_dq(&ghs[num_gh]);
1833 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1835 unsigned long delay = 0;
1836 unsigned long holdtime;
1837 unsigned long now = jiffies;
1839 gfs2_glock_hold(gl);
1840 spin_lock(&gl->gl_lockref.lock);
1841 holdtime = gl->gl_tchange + gl->gl_hold_time;
1842 if (!list_empty(&gl->gl_holders) &&
1843 gl->gl_name.ln_type == LM_TYPE_INODE) {
1844 if (time_before(now, holdtime))
1845 delay = holdtime - now;
1846 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1847 delay = gl->gl_hold_time;
1849 handle_callback(gl, state, delay, true);
1850 __gfs2_glock_queue_work(gl, delay);
1851 spin_unlock(&gl->gl_lockref.lock);
1855 * gfs2_should_freeze - Figure out if glock should be frozen
1856 * @gl: The glock in question
1858 * Glocks are not frozen if (a) the result of the dlm operation is
1859 * an error, (b) the locking operation was an unlock operation or
1860 * (c) if there is a "noexp" flagged request anywhere in the queue
1862 * Returns: 1 if freezing should occur, 0 otherwise
1865 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1867 const struct gfs2_holder *gh;
1869 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1871 if (gl->gl_target == LM_ST_UNLOCKED)
1874 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1875 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1877 if (LM_FLAG_NOEXP & gh->gh_flags)
1885 * gfs2_glock_complete - Callback used by locking
1886 * @gl: Pointer to the glock
1887 * @ret: The return value from the dlm
1889 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1890 * to use a bitfield shared with other glock state fields.
1893 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1895 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1897 spin_lock(&gl->gl_lockref.lock);
1900 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1901 if (gfs2_should_freeze(gl)) {
1902 set_bit(GLF_FROZEN, &gl->gl_flags);
1903 spin_unlock(&gl->gl_lockref.lock);
1908 gl->gl_lockref.count++;
1909 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1910 __gfs2_glock_queue_work(gl, 0);
1911 spin_unlock(&gl->gl_lockref.lock);
1914 static int glock_cmp(void *priv, const struct list_head *a,
1915 const struct list_head *b)
1917 struct gfs2_glock *gla, *glb;
1919 gla = list_entry(a, struct gfs2_glock, gl_lru);
1920 glb = list_entry(b, struct gfs2_glock, gl_lru);
1922 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1924 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1931 * gfs2_dispose_glock_lru - Demote a list of glocks
1932 * @list: The list to dispose of
1934 * Disposing of glocks may involve disk accesses, so that here we sort
1935 * the glocks by number (i.e. disk location of the inodes) so that if
1936 * there are any such accesses, they'll be sent in order (mostly).
1938 * Must be called under the lru_lock, but may drop and retake this
1939 * lock. While the lru_lock is dropped, entries may vanish from the
1940 * list, but no new entries will appear on the list (since it is
1944 static void gfs2_dispose_glock_lru(struct list_head *list)
1945 __releases(&lru_lock)
1946 __acquires(&lru_lock)
1948 struct gfs2_glock *gl;
1950 list_sort(NULL, list, glock_cmp);
1952 while(!list_empty(list)) {
1953 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
1954 list_del_init(&gl->gl_lru);
1955 clear_bit(GLF_LRU, &gl->gl_flags);
1956 if (!spin_trylock(&gl->gl_lockref.lock)) {
1958 list_add(&gl->gl_lru, &lru_list);
1959 set_bit(GLF_LRU, &gl->gl_flags);
1960 atomic_inc(&lru_count);
1963 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1964 spin_unlock(&gl->gl_lockref.lock);
1965 goto add_back_to_lru;
1967 gl->gl_lockref.count++;
1969 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1970 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1971 __gfs2_glock_queue_work(gl, 0);
1972 spin_unlock(&gl->gl_lockref.lock);
1973 cond_resched_lock(&lru_lock);
1978 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1979 * @nr: The number of entries to scan
1981 * This function selects the entries on the LRU which are able to
1982 * be demoted, and then kicks off the process by calling
1983 * gfs2_dispose_glock_lru() above.
1986 static long gfs2_scan_glock_lru(int nr)
1988 struct gfs2_glock *gl;
1993 spin_lock(&lru_lock);
1994 while ((nr-- >= 0) && !list_empty(&lru_list)) {
1995 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
1997 /* Test for being demotable */
1998 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1999 list_move(&gl->gl_lru, &dispose);
2000 atomic_dec(&lru_count);
2005 list_move(&gl->gl_lru, &skipped);
2007 list_splice(&skipped, &lru_list);
2008 if (!list_empty(&dispose))
2009 gfs2_dispose_glock_lru(&dispose);
2010 spin_unlock(&lru_lock);
2015 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2016 struct shrink_control *sc)
2018 if (!(sc->gfp_mask & __GFP_FS))
2020 return gfs2_scan_glock_lru(sc->nr_to_scan);
2023 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2024 struct shrink_control *sc)
2026 return vfs_pressure_ratio(atomic_read(&lru_count));
2029 static struct shrinker glock_shrinker = {
2030 .seeks = DEFAULT_SEEKS,
2031 .count_objects = gfs2_glock_shrink_count,
2032 .scan_objects = gfs2_glock_shrink_scan,
2036 * glock_hash_walk - Call a function for glock in a hash bucket
2037 * @examiner: the function
2038 * @sdp: the filesystem
2040 * Note that the function can be called multiple times on the same
2041 * object. So the user must ensure that the function can cope with
2045 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2047 struct gfs2_glock *gl;
2048 struct rhashtable_iter iter;
2050 rhashtable_walk_enter(&gl_hash_table, &iter);
2053 rhashtable_walk_start(&iter);
2055 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2056 if (gl->gl_name.ln_sbd == sdp)
2060 rhashtable_walk_stop(&iter);
2061 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2063 rhashtable_walk_exit(&iter);
2066 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2070 spin_lock(&gl->gl_lockref.lock);
2071 queued = queue_delayed_work(gfs2_delete_workqueue,
2072 &gl->gl_delete, delay);
2074 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2075 spin_unlock(&gl->gl_lockref.lock);
2079 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2081 if (cancel_delayed_work(&gl->gl_delete)) {
2082 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2087 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2089 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2092 static void flush_delete_work(struct gfs2_glock *gl)
2094 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2095 if (cancel_delayed_work(&gl->gl_delete)) {
2096 queue_delayed_work(gfs2_delete_workqueue,
2102 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2104 glock_hash_walk(flush_delete_work, sdp);
2105 flush_workqueue(gfs2_delete_workqueue);
2109 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2110 * @gl: The glock to thaw
2114 static void thaw_glock(struct gfs2_glock *gl)
2116 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2118 if (!lockref_get_not_dead(&gl->gl_lockref))
2120 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2121 gfs2_glock_queue_work(gl, 0);
2125 * clear_glock - look at a glock and see if we can free it from glock cache
2126 * @gl: the glock to look at
2130 static void clear_glock(struct gfs2_glock *gl)
2132 gfs2_glock_remove_from_lru(gl);
2134 spin_lock(&gl->gl_lockref.lock);
2135 if (!__lockref_is_dead(&gl->gl_lockref)) {
2136 gl->gl_lockref.count++;
2137 if (gl->gl_state != LM_ST_UNLOCKED)
2138 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2139 __gfs2_glock_queue_work(gl, 0);
2141 spin_unlock(&gl->gl_lockref.lock);
2145 * gfs2_glock_thaw - Thaw any frozen glocks
2146 * @sdp: The super block
2150 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2152 glock_hash_walk(thaw_glock, sdp);
2155 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2157 spin_lock(&gl->gl_lockref.lock);
2158 gfs2_dump_glock(seq, gl, fsid);
2159 spin_unlock(&gl->gl_lockref.lock);
2162 static void dump_glock_func(struct gfs2_glock *gl)
2164 dump_glock(NULL, gl, true);
2167 static void withdraw_dq(struct gfs2_glock *gl)
2169 spin_lock(&gl->gl_lockref.lock);
2170 if (!__lockref_is_dead(&gl->gl_lockref) &&
2171 glock_blocked_by_withdraw(gl))
2172 do_error(gl, LM_OUT_ERROR); /* remove pending waiters */
2173 spin_unlock(&gl->gl_lockref.lock);
2176 void gfs2_gl_dq_holders(struct gfs2_sbd *sdp)
2178 glock_hash_walk(withdraw_dq, sdp);
2182 * gfs2_gl_hash_clear - Empty out the glock hash table
2183 * @sdp: the filesystem
2185 * Called when unmounting the filesystem.
2188 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2190 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2191 flush_workqueue(glock_workqueue);
2192 glock_hash_walk(clear_glock, sdp);
2193 flush_workqueue(glock_workqueue);
2194 wait_event_timeout(sdp->sd_glock_wait,
2195 atomic_read(&sdp->sd_glock_disposal) == 0,
2197 glock_hash_walk(dump_glock_func, sdp);
2200 static const char *state2str(unsigned state)
2203 case LM_ST_UNLOCKED:
2207 case LM_ST_DEFERRED:
2209 case LM_ST_EXCLUSIVE:
2215 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2218 if (flags & LM_FLAG_TRY)
2220 if (flags & LM_FLAG_TRY_1CB)
2222 if (flags & LM_FLAG_NOEXP)
2224 if (flags & LM_FLAG_ANY)
2226 if (flags & LM_FLAG_PRIORITY)
2228 if (flags & LM_FLAG_NODE_SCOPE)
2230 if (flags & GL_ASYNC)
2232 if (flags & GL_EXACT)
2234 if (flags & GL_NOCACHE)
2236 if (test_bit(HIF_HOLDER, &iflags))
2238 if (test_bit(HIF_WAIT, &iflags))
2240 if (flags & GL_SKIP)
2247 * dump_holder - print information about a glock holder
2248 * @seq: the seq_file struct
2249 * @gh: the glock holder
2250 * @fs_id_buf: pointer to file system id (if requested)
2254 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2255 const char *fs_id_buf)
2257 const char *comm = "(none)";
2258 pid_t owner_pid = 0;
2262 if (pid_is_meaningful(gh)) {
2263 struct task_struct *gh_owner;
2266 owner_pid = pid_nr(gh->gh_owner_pid);
2267 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2269 comm = gh_owner->comm;
2271 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2272 fs_id_buf, state2str(gh->gh_state),
2273 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2274 gh->gh_error, (long)owner_pid, comm, (void *)gh->gh_ip);
2278 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2280 const unsigned long *gflags = &gl->gl_flags;
2283 if (test_bit(GLF_LOCK, gflags))
2285 if (test_bit(GLF_DEMOTE, gflags))
2287 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2289 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2291 if (test_bit(GLF_DIRTY, gflags))
2293 if (test_bit(GLF_LFLUSH, gflags))
2295 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2297 if (test_bit(GLF_REPLY_PENDING, gflags))
2299 if (test_bit(GLF_INITIAL, gflags))
2301 if (test_bit(GLF_FROZEN, gflags))
2303 if (!list_empty(&gl->gl_holders))
2305 if (test_bit(GLF_LRU, gflags))
2309 if (test_bit(GLF_BLOCKING, gflags))
2311 if (test_bit(GLF_PENDING_DELETE, gflags))
2313 if (test_bit(GLF_FREEING, gflags))
2315 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2317 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2324 * gfs2_dump_glock - print information about a glock
2325 * @seq: The seq_file struct
2327 * @fsid: If true, also dump the file system id
2329 * The file format is as follows:
2330 * One line per object, capital letters are used to indicate objects
2331 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2332 * other objects are indented by a single space and follow the glock to
2333 * which they are related. Fields are indicated by lower case letters
2334 * followed by a colon and the field value, except for strings which are in
2335 * [] so that its possible to see if they are composed of spaces for
2336 * example. The field's are n = number (id of the object), f = flags,
2337 * t = type, s = state, r = refcount, e = error, p = pid.
2341 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2343 const struct gfs2_glock_operations *glops = gl->gl_ops;
2344 unsigned long long dtime;
2345 const struct gfs2_holder *gh;
2346 char gflags_buf[32];
2347 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2348 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2349 unsigned long nrpages = 0;
2351 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2352 struct address_space *mapping = gfs2_glock2aspace(gl);
2354 nrpages = mapping->nrpages;
2356 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2357 if (fsid && sdp) /* safety precaution */
2358 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2359 dtime = jiffies - gl->gl_demote_time;
2360 dtime *= 1000000/HZ; /* demote time in uSec */
2361 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2363 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2364 "v:%d r:%d m:%ld p:%lu\n",
2365 fs_id_buf, state2str(gl->gl_state),
2366 gl->gl_name.ln_type,
2367 (unsigned long long)gl->gl_name.ln_number,
2368 gflags2str(gflags_buf, gl),
2369 state2str(gl->gl_target),
2370 state2str(gl->gl_demote_state), dtime,
2371 atomic_read(&gl->gl_ail_count),
2372 atomic_read(&gl->gl_revokes),
2373 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2375 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2376 dump_holder(seq, gh, fs_id_buf);
2378 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2379 glops->go_dump(seq, gl, fs_id_buf);
2382 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2384 struct gfs2_glock *gl = iter_ptr;
2386 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2387 gl->gl_name.ln_type,
2388 (unsigned long long)gl->gl_name.ln_number,
2389 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2390 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2391 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2392 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2393 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2394 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2395 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2396 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2400 static const char *gfs2_gltype[] = {
2414 static const char *gfs2_stype[] = {
2415 [GFS2_LKS_SRTT] = "srtt",
2416 [GFS2_LKS_SRTTVAR] = "srttvar",
2417 [GFS2_LKS_SRTTB] = "srttb",
2418 [GFS2_LKS_SRTTVARB] = "srttvarb",
2419 [GFS2_LKS_SIRT] = "sirt",
2420 [GFS2_LKS_SIRTVAR] = "sirtvar",
2421 [GFS2_LKS_DCOUNT] = "dlm",
2422 [GFS2_LKS_QCOUNT] = "queue",
2425 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2427 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2429 struct gfs2_sbd *sdp = seq->private;
2430 loff_t pos = *(loff_t *)iter_ptr;
2431 unsigned index = pos >> 3;
2432 unsigned subindex = pos & 0x07;
2435 if (index == 0 && subindex != 0)
2438 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2439 (index == 0) ? "cpu": gfs2_stype[subindex]);
2441 for_each_possible_cpu(i) {
2442 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2445 seq_printf(seq, " %15u", i);
2447 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2448 lkstats[index - 1].stats[subindex]);
2450 seq_putc(seq, '\n');
2454 int __init gfs2_glock_init(void)
2458 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2462 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2463 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2464 if (!glock_workqueue) {
2465 rhashtable_destroy(&gl_hash_table);
2468 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2469 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2471 if (!gfs2_delete_workqueue) {
2472 destroy_workqueue(glock_workqueue);
2473 rhashtable_destroy(&gl_hash_table);
2477 ret = register_shrinker(&glock_shrinker, "gfs2-glock");
2479 destroy_workqueue(gfs2_delete_workqueue);
2480 destroy_workqueue(glock_workqueue);
2481 rhashtable_destroy(&gl_hash_table);
2485 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2486 init_waitqueue_head(glock_wait_table + i);
2491 void gfs2_glock_exit(void)
2493 unregister_shrinker(&glock_shrinker);
2494 rhashtable_destroy(&gl_hash_table);
2495 destroy_workqueue(glock_workqueue);
2496 destroy_workqueue(gfs2_delete_workqueue);
2499 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2501 struct gfs2_glock *gl = gi->gl;
2506 if (!lockref_put_not_zero(&gl->gl_lockref))
2507 gfs2_glock_queue_put(gl);
2510 gl = rhashtable_walk_next(&gi->hti);
2511 if (IS_ERR_OR_NULL(gl)) {
2512 if (gl == ERR_PTR(-EAGAIN)) {
2519 if (gl->gl_name.ln_sbd != gi->sdp)
2522 if (!lockref_get_not_dead(&gl->gl_lockref))
2526 if (__lockref_is_dead(&gl->gl_lockref))
2534 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2537 struct gfs2_glock_iter *gi = seq->private;
2541 * We can either stay where we are, skip to the next hash table
2542 * entry, or start from the beginning.
2544 if (*pos < gi->last_pos) {
2545 rhashtable_walk_exit(&gi->hti);
2546 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2549 n = *pos - gi->last_pos;
2552 rhashtable_walk_start(&gi->hti);
2554 gfs2_glock_iter_next(gi, n);
2555 gi->last_pos = *pos;
2559 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2562 struct gfs2_glock_iter *gi = seq->private;
2565 gi->last_pos = *pos;
2566 gfs2_glock_iter_next(gi, 1);
2570 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2573 struct gfs2_glock_iter *gi = seq->private;
2575 rhashtable_walk_stop(&gi->hti);
2578 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2580 dump_glock(seq, iter_ptr, false);
2584 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2587 if (*pos >= GFS2_NR_SBSTATS)
2592 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2596 if (*pos >= GFS2_NR_SBSTATS)
2601 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2606 static const struct seq_operations gfs2_glock_seq_ops = {
2607 .start = gfs2_glock_seq_start,
2608 .next = gfs2_glock_seq_next,
2609 .stop = gfs2_glock_seq_stop,
2610 .show = gfs2_glock_seq_show,
2613 static const struct seq_operations gfs2_glstats_seq_ops = {
2614 .start = gfs2_glock_seq_start,
2615 .next = gfs2_glock_seq_next,
2616 .stop = gfs2_glock_seq_stop,
2617 .show = gfs2_glstats_seq_show,
2620 static const struct seq_operations gfs2_sbstats_sops = {
2621 .start = gfs2_sbstats_seq_start,
2622 .next = gfs2_sbstats_seq_next,
2623 .stop = gfs2_sbstats_seq_stop,
2624 .show = gfs2_sbstats_seq_show,
2627 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2629 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2630 const struct seq_operations *ops)
2632 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2634 struct seq_file *seq = file->private_data;
2635 struct gfs2_glock_iter *gi = seq->private;
2637 gi->sdp = inode->i_private;
2638 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2640 seq->size = GFS2_SEQ_GOODSIZE;
2642 * Initially, we are "before" the first hash table entry; the
2643 * first call to rhashtable_walk_next gets us the first entry.
2647 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2652 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2654 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2657 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2659 struct seq_file *seq = file->private_data;
2660 struct gfs2_glock_iter *gi = seq->private;
2663 gfs2_glock_put(gi->gl);
2664 rhashtable_walk_exit(&gi->hti);
2665 return seq_release_private(inode, file);
2668 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2670 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2673 static const struct file_operations gfs2_glocks_fops = {
2674 .owner = THIS_MODULE,
2675 .open = gfs2_glocks_open,
2677 .llseek = seq_lseek,
2678 .release = gfs2_glocks_release,
2681 static const struct file_operations gfs2_glstats_fops = {
2682 .owner = THIS_MODULE,
2683 .open = gfs2_glstats_open,
2685 .llseek = seq_lseek,
2686 .release = gfs2_glocks_release,
2689 struct gfs2_glockfd_iter {
2690 struct super_block *sb;
2692 struct task_struct *task;
2697 static struct task_struct *gfs2_glockfd_next_task(struct gfs2_glockfd_iter *i)
2699 struct pid_namespace *ns = task_active_pid_ns(current);
2703 put_task_struct(i->task);
2708 pid = find_ge_pid(i->tgid, ns);
2710 i->tgid = pid_nr_ns(pid, ns);
2711 i->task = pid_task(pid, PIDTYPE_TGID);
2716 get_task_struct(i->task);
2722 static struct file *gfs2_glockfd_next_file(struct gfs2_glockfd_iter *i)
2731 struct inode *inode;
2733 i->file = task_lookup_next_fd_rcu(i->task, &i->fd);
2738 inode = file_inode(i->file);
2739 if (inode->i_sb != i->sb)
2741 if (get_file_rcu(i->file))
2748 static void *gfs2_glockfd_seq_start(struct seq_file *seq, loff_t *pos)
2750 struct gfs2_glockfd_iter *i = seq->private;
2754 while (gfs2_glockfd_next_task(i)) {
2755 if (gfs2_glockfd_next_file(i))
2762 static void *gfs2_glockfd_seq_next(struct seq_file *seq, void *iter_ptr,
2765 struct gfs2_glockfd_iter *i = seq->private;
2770 if (gfs2_glockfd_next_file(i))
2773 } while (gfs2_glockfd_next_task(i));
2777 static void gfs2_glockfd_seq_stop(struct seq_file *seq, void *iter_ptr)
2779 struct gfs2_glockfd_iter *i = seq->private;
2784 put_task_struct(i->task);
2787 static void gfs2_glockfd_seq_show_flock(struct seq_file *seq,
2788 struct gfs2_glockfd_iter *i)
2790 struct gfs2_file *fp = i->file->private_data;
2791 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
2792 struct lm_lockname gl_name = { .ln_type = LM_TYPE_RESERVED };
2794 if (!READ_ONCE(fl_gh->gh_gl))
2797 spin_lock(&i->file->f_lock);
2798 if (gfs2_holder_initialized(fl_gh))
2799 gl_name = fl_gh->gh_gl->gl_name;
2800 spin_unlock(&i->file->f_lock);
2802 if (gl_name.ln_type != LM_TYPE_RESERVED) {
2803 seq_printf(seq, "%d %u %u/%llx\n",
2804 i->tgid, i->fd, gl_name.ln_type,
2805 (unsigned long long)gl_name.ln_number);
2809 static int gfs2_glockfd_seq_show(struct seq_file *seq, void *iter_ptr)
2811 struct gfs2_glockfd_iter *i = seq->private;
2812 struct inode *inode = file_inode(i->file);
2813 struct gfs2_glock *gl;
2815 inode_lock_shared(inode);
2816 gl = GFS2_I(inode)->i_iopen_gh.gh_gl;
2818 seq_printf(seq, "%d %u %u/%llx\n",
2819 i->tgid, i->fd, gl->gl_name.ln_type,
2820 (unsigned long long)gl->gl_name.ln_number);
2822 gfs2_glockfd_seq_show_flock(seq, i);
2823 inode_unlock_shared(inode);
2827 static const struct seq_operations gfs2_glockfd_seq_ops = {
2828 .start = gfs2_glockfd_seq_start,
2829 .next = gfs2_glockfd_seq_next,
2830 .stop = gfs2_glockfd_seq_stop,
2831 .show = gfs2_glockfd_seq_show,
2834 static int gfs2_glockfd_open(struct inode *inode, struct file *file)
2836 struct gfs2_glockfd_iter *i;
2837 struct gfs2_sbd *sdp = inode->i_private;
2839 i = __seq_open_private(file, &gfs2_glockfd_seq_ops,
2840 sizeof(struct gfs2_glockfd_iter));
2843 i->sb = sdp->sd_vfs;
2847 static const struct file_operations gfs2_glockfd_fops = {
2848 .owner = THIS_MODULE,
2849 .open = gfs2_glockfd_open,
2851 .llseek = seq_lseek,
2852 .release = seq_release_private,
2855 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2857 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2859 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2861 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2864 debugfs_create_file("glockfd", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2865 &gfs2_glockfd_fops);
2867 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2868 &gfs2_glstats_fops);
2870 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2871 &gfs2_sbstats_fops);
2874 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2876 debugfs_remove_recursive(sdp->debugfs_dir);
2877 sdp->debugfs_dir = NULL;
2880 void gfs2_register_debugfs(void)
2882 gfs2_root = debugfs_create_dir("gfs2", NULL);
2885 void gfs2_unregister_debugfs(void)
2887 debugfs_remove(gfs2_root);