1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/buffer_head.h>
13 #include <linux/delay.h>
14 #include <linux/sort.h>
15 #include <linux/hash.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <linux/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/percpu.h>
33 #include <linux/list_sort.h>
34 #include <linux/lockref.h>
35 #include <linux/rhashtable.h>
48 #define CREATE_TRACE_POINTS
49 #include "trace_gfs2.h"
51 struct gfs2_glock_iter {
52 struct gfs2_sbd *sdp; /* incore superblock */
53 struct rhashtable_iter hti; /* rhashtable iterator */
54 struct gfs2_glock *gl; /* current glock struct */
55 loff_t last_pos; /* last position */
58 typedef void (*glock_examiner) (struct gfs2_glock * gl);
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
61 static void __gfs2_glock_dq(struct gfs2_holder *gh);
63 static struct dentry *gfs2_root;
64 static struct workqueue_struct *glock_workqueue;
65 struct workqueue_struct *gfs2_delete_workqueue;
66 static LIST_HEAD(lru_list);
67 static atomic_t lru_count = ATOMIC_INIT(0);
68 static DEFINE_SPINLOCK(lru_lock);
70 #define GFS2_GL_HASH_SHIFT 15
71 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
73 static const struct rhashtable_params ht_parms = {
74 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
75 .key_len = offsetofend(struct lm_lockname, ln_type),
76 .key_offset = offsetof(struct gfs2_glock, gl_name),
77 .head_offset = offsetof(struct gfs2_glock, gl_node),
80 static struct rhashtable gl_hash_table;
82 #define GLOCK_WAIT_TABLE_BITS 12
83 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
84 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
86 struct wait_glock_queue {
87 struct lm_lockname *name;
88 wait_queue_entry_t wait;
91 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
94 struct wait_glock_queue *wait_glock =
95 container_of(wait, struct wait_glock_queue, wait);
96 struct lm_lockname *wait_name = wait_glock->name;
97 struct lm_lockname *wake_name = key;
99 if (wake_name->ln_sbd != wait_name->ln_sbd ||
100 wake_name->ln_number != wait_name->ln_number ||
101 wake_name->ln_type != wait_name->ln_type)
103 return autoremove_wake_function(wait, mode, sync, key);
106 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
108 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
110 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
114 * wake_up_glock - Wake up waiters on a glock
117 static void wake_up_glock(struct gfs2_glock *gl)
119 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
121 if (waitqueue_active(wq))
122 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
125 static void gfs2_glock_dealloc(struct rcu_head *rcu)
127 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
129 kfree(gl->gl_lksb.sb_lvbptr);
130 if (gl->gl_ops->go_flags & GLOF_ASPACE)
131 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
133 kmem_cache_free(gfs2_glock_cachep, gl);
137 * glock_blocked_by_withdraw - determine if we can still use a glock
140 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
141 * when we're withdrawn. For example, to maintain metadata integrity, we should
142 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks, like
143 * iopen or the transaction glocks may be safely used because none of their
144 * metadata goes through the journal. So in general, we should disallow all
145 * glocks that are journaled, and allow all the others. One exception is:
146 * we need to allow our active journal to be promoted and demoted so others
147 * may recover it and we can reacquire it when they're done.
149 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
151 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
153 if (likely(!gfs2_withdrawn(sdp)))
155 if (gl->gl_ops->go_flags & GLOF_NONDISK)
157 if (!sdp->sd_jdesc ||
158 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
163 void gfs2_glock_free(struct gfs2_glock *gl)
165 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
167 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
168 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
171 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
172 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
173 wake_up(&sdp->sd_glock_wait);
177 * gfs2_glock_hold() - increment reference count on glock
178 * @gl: The glock to hold
182 void gfs2_glock_hold(struct gfs2_glock *gl)
184 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
185 lockref_get(&gl->gl_lockref);
189 * demote_ok - Check to see if it's ok to unlock a glock
192 * Returns: 1 if it's ok
195 static int demote_ok(const struct gfs2_glock *gl)
197 const struct gfs2_glock_operations *glops = gl->gl_ops;
199 if (gl->gl_state == LM_ST_UNLOCKED)
202 * Note that demote_ok is used for the lru process of disposing of
203 * glocks. For this purpose, we don't care if the glock's holders
204 * have the HIF_MAY_DEMOTE flag set or not. If someone is using
205 * them, don't demote.
207 if (!list_empty(&gl->gl_holders))
209 if (glops->go_demote_ok)
210 return glops->go_demote_ok(gl);
215 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
217 if (!(gl->gl_ops->go_flags & GLOF_LRU))
220 spin_lock(&lru_lock);
222 list_move_tail(&gl->gl_lru, &lru_list);
224 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
225 set_bit(GLF_LRU, &gl->gl_flags);
226 atomic_inc(&lru_count);
229 spin_unlock(&lru_lock);
232 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
234 if (!(gl->gl_ops->go_flags & GLOF_LRU))
237 spin_lock(&lru_lock);
238 if (test_bit(GLF_LRU, &gl->gl_flags)) {
239 list_del_init(&gl->gl_lru);
240 atomic_dec(&lru_count);
241 clear_bit(GLF_LRU, &gl->gl_flags);
243 spin_unlock(&lru_lock);
247 * Enqueue the glock on the work queue. Passes one glock reference on to the
250 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
251 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
253 * We are holding the lockref spinlock, and the work was still
254 * queued above. The queued work (glock_work_func) takes that
255 * spinlock before dropping its glock reference(s), so it
256 * cannot have dropped them in the meantime.
258 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
259 gl->gl_lockref.count--;
263 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
264 spin_lock(&gl->gl_lockref.lock);
265 __gfs2_glock_queue_work(gl, delay);
266 spin_unlock(&gl->gl_lockref.lock);
269 static void __gfs2_glock_put(struct gfs2_glock *gl)
271 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
272 struct address_space *mapping = gfs2_glock2aspace(gl);
274 lockref_mark_dead(&gl->gl_lockref);
276 gfs2_glock_remove_from_lru(gl);
277 spin_unlock(&gl->gl_lockref.lock);
278 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
280 truncate_inode_pages_final(mapping);
281 if (!gfs2_withdrawn(sdp))
282 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
284 trace_gfs2_glock_put(gl);
285 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
289 * Cause the glock to be put in work queue context.
291 void gfs2_glock_queue_put(struct gfs2_glock *gl)
293 gfs2_glock_queue_work(gl, 0);
297 * gfs2_glock_put() - Decrement reference count on glock
298 * @gl: The glock to put
302 void gfs2_glock_put(struct gfs2_glock *gl)
304 /* last put could call sleepable dlm api */
307 if (lockref_put_or_lock(&gl->gl_lockref))
310 __gfs2_glock_put(gl);
314 * may_grant - check if it's ok to grant a new lock
316 * @current_gh: One of the current holders of @gl
317 * @gh: The lock request which we wish to grant
319 * With our current compatibility rules, if a glock has one or more active
320 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
321 * @current_gh; they are all the same as far as compatibility with the new @gh
324 * Returns true if it's ok to grant the lock.
327 static inline bool may_grant(struct gfs2_glock *gl,
328 struct gfs2_holder *current_gh,
329 struct gfs2_holder *gh)
332 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
334 switch(current_gh->gh_state) {
335 case LM_ST_EXCLUSIVE:
337 * Here we make a special exception to grant holders
338 * who agree to share the EX lock with other holders
339 * who also have the bit set. If the original holder
340 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
341 * holders with the bit set.
343 return gh->gh_state == LM_ST_EXCLUSIVE &&
344 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
345 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
349 return gh->gh_state == current_gh->gh_state;
356 if (gl->gl_state == gh->gh_state)
358 if (gh->gh_flags & GL_EXACT)
360 if (gl->gl_state == LM_ST_EXCLUSIVE) {
361 return gh->gh_state == LM_ST_SHARED ||
362 gh->gh_state == LM_ST_DEFERRED;
364 if (gh->gh_flags & LM_FLAG_ANY)
365 return gl->gl_state != LM_ST_UNLOCKED;
369 static void gfs2_holder_wake(struct gfs2_holder *gh)
371 clear_bit(HIF_WAIT, &gh->gh_iflags);
372 smp_mb__after_atomic();
373 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
374 if (gh->gh_flags & GL_ASYNC) {
375 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
377 wake_up(&sdp->sd_async_glock_wait);
382 * do_error - Something unexpected has happened during a lock request
384 * @ret: The status from the DLM
387 static void do_error(struct gfs2_glock *gl, const int ret)
389 struct gfs2_holder *gh, *tmp;
391 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
392 if (!test_bit(HIF_WAIT, &gh->gh_iflags))
394 if (ret & LM_OUT_ERROR)
396 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
397 gh->gh_error = GLR_TRYFAILED;
400 list_del_init(&gh->gh_list);
401 trace_gfs2_glock_queue(gh, 0);
402 gfs2_holder_wake(gh);
407 * demote_incompat_holders - demote incompatible demoteable holders
408 * @gl: the glock we want to promote
409 * @new_gh: the new holder to be promoted
411 static void demote_incompat_holders(struct gfs2_glock *gl,
412 struct gfs2_holder *new_gh)
414 struct gfs2_holder *gh;
417 * Demote incompatible holders before we make ourselves eligible.
418 * (This holder may or may not allow auto-demoting, but we don't want
419 * to demote the new holder before it's even granted.)
421 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
423 * Since holders are at the front of the list, we stop when we
424 * find the first non-holder.
426 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
428 if (test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags) &&
429 !may_grant(gl, new_gh, gh)) {
431 * We should not recurse into do_promote because
432 * __gfs2_glock_dq only calls handle_callback,
433 * gfs2_glock_add_to_lru and __gfs2_glock_queue_work.
441 * find_first_holder - find the first "holder" gh
445 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
447 struct gfs2_holder *gh;
449 if (!list_empty(&gl->gl_holders)) {
450 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
452 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
459 * find_first_strong_holder - find the first non-demoteable holder
462 * Find the first holder that doesn't have the HIF_MAY_DEMOTE flag set.
464 static inline struct gfs2_holder *
465 find_first_strong_holder(struct gfs2_glock *gl)
467 struct gfs2_holder *gh;
469 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
470 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
472 if (!test_bit(HIF_MAY_DEMOTE, &gh->gh_iflags))
479 * gfs2_instantiate - Call the glops instantiate function
482 * Returns: 0 if instantiate was successful, 2 if type specific operation is
483 * underway, or error.
485 int gfs2_instantiate(struct gfs2_holder *gh)
487 struct gfs2_glock *gl = gh->gh_gl;
488 const struct gfs2_glock_operations *glops = gl->gl_ops;
492 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
496 * Since we unlock the lockref lock, we set a flag to indicate
497 * instantiate is in progress.
499 if (test_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
500 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
501 TASK_UNINTERRUPTIBLE);
503 * Here we just waited for a different instantiate to finish.
504 * But that may not have been successful, as when a process
505 * locks an inode glock _before_ it has an actual inode to
506 * instantiate into. So we check again. This process might
507 * have an inode to instantiate, so might be successful.
512 set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
514 ret = glops->go_instantiate(gh);
516 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
517 clear_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
518 smp_mb__after_atomic();
519 wake_up_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG);
524 * do_promote - promote as many requests as possible on the current queue
527 * Returns: 1 if there is a blocked holder at the head of the list, or 2
528 * if a type specific operation is underway.
531 static int do_promote(struct gfs2_glock *gl)
532 __releases(&gl->gl_lockref.lock)
533 __acquires(&gl->gl_lockref.lock)
535 struct gfs2_holder *gh, *tmp, *first_gh;
536 bool incompat_holders_demoted = false;
541 first_gh = find_first_strong_holder(gl);
542 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
543 lock_released = false;
544 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
546 if (!may_grant(gl, first_gh, gh)) {
548 * If we get here, it means we may not grant this holder for
549 * some reason. If this holder is the head of the list, it
550 * means we have a blocked holder at the head, so return 1.
552 if (gh->gh_list.prev == &gl->gl_holders)
557 if (!incompat_holders_demoted) {
558 demote_incompat_holders(gl, first_gh);
559 incompat_holders_demoted = true;
562 if (test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags) &&
563 !(gh->gh_flags & GL_SKIP) && gl->gl_ops->go_instantiate) {
564 lock_released = true;
565 spin_unlock(&gl->gl_lockref.lock);
566 ret = gfs2_instantiate(gh);
567 spin_lock(&gl->gl_lockref.lock);
572 list_del_init(&gh->gh_list);
573 trace_gfs2_glock_queue(gh, 0);
574 gfs2_holder_wake(gh);
578 set_bit(HIF_HOLDER, &gh->gh_iflags);
579 trace_gfs2_promote(gh);
580 gfs2_holder_wake(gh);
582 * If we released the gl_lockref.lock the holders list may have
583 * changed. For that reason, we start again at the start of
593 * find_first_waiter - find the first gh that's waiting for the glock
597 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
599 struct gfs2_holder *gh;
601 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
602 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
609 * state_change - record that the glock is now in a different state
611 * @new_state: the new state
614 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
618 held1 = (gl->gl_state != LM_ST_UNLOCKED);
619 held2 = (new_state != LM_ST_UNLOCKED);
621 if (held1 != held2) {
622 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
624 gl->gl_lockref.count++;
626 gl->gl_lockref.count--;
628 if (new_state != gl->gl_target)
629 /* shorten our minimum hold time */
630 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
632 gl->gl_state = new_state;
633 gl->gl_tchange = jiffies;
636 static void gfs2_set_demote(struct gfs2_glock *gl)
638 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
640 set_bit(GLF_DEMOTE, &gl->gl_flags);
642 wake_up(&sdp->sd_async_glock_wait);
645 static void gfs2_demote_wake(struct gfs2_glock *gl)
647 gl->gl_demote_state = LM_ST_EXCLUSIVE;
648 clear_bit(GLF_DEMOTE, &gl->gl_flags);
649 smp_mb__after_atomic();
650 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
654 * finish_xmote - The DLM has replied to one of our lock requests
656 * @ret: The status from the DLM
660 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
662 const struct gfs2_glock_operations *glops = gl->gl_ops;
663 struct gfs2_holder *gh;
664 unsigned state = ret & LM_OUT_ST_MASK;
667 spin_lock(&gl->gl_lockref.lock);
668 trace_gfs2_glock_state_change(gl, state);
669 state_change(gl, state);
670 gh = find_first_waiter(gl);
672 /* Demote to UN request arrived during demote to SH or DF */
673 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
674 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
675 gl->gl_target = LM_ST_UNLOCKED;
677 /* Check for state != intended state */
678 if (unlikely(state != gl->gl_target)) {
679 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
680 /* move to back of queue and try next entry */
681 if (ret & LM_OUT_CANCELED) {
682 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
683 list_move_tail(&gh->gh_list, &gl->gl_holders);
684 gh = find_first_waiter(gl);
685 gl->gl_target = gh->gh_state;
688 /* Some error or failed "try lock" - report it */
689 if ((ret & LM_OUT_ERROR) ||
690 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
691 gl->gl_target = gl->gl_state;
697 /* Unlocked due to conversion deadlock, try again */
700 do_xmote(gl, gh, gl->gl_target);
702 /* Conversion fails, unlock and try again */
705 do_xmote(gl, gh, LM_ST_UNLOCKED);
707 default: /* Everything else */
708 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
709 gl->gl_target, state);
712 spin_unlock(&gl->gl_lockref.lock);
716 /* Fast path - we got what we asked for */
717 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
718 gfs2_demote_wake(gl);
719 if (state != LM_ST_UNLOCKED) {
720 if (glops->go_xmote_bh) {
721 spin_unlock(&gl->gl_lockref.lock);
722 rv = glops->go_xmote_bh(gl);
723 spin_lock(&gl->gl_lockref.lock);
734 clear_bit(GLF_LOCK, &gl->gl_flags);
736 spin_unlock(&gl->gl_lockref.lock);
739 static bool is_system_glock(struct gfs2_glock *gl)
741 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
742 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
744 if (gl == m_ip->i_gl)
750 * do_xmote - Calls the DLM to change the state of a lock
751 * @gl: The lock state
752 * @gh: The holder (only for promotes)
753 * @target: The target lock state
757 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
758 __releases(&gl->gl_lockref.lock)
759 __acquires(&gl->gl_lockref.lock)
761 const struct gfs2_glock_operations *glops = gl->gl_ops;
762 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
763 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
766 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
767 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
769 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
771 GLOCK_BUG_ON(gl, gl->gl_state == target);
772 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
773 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
776 * If another process is already doing the invalidate, let that
777 * finish first. The glock state machine will get back to this
778 * holder again later.
780 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
783 do_error(gl, 0); /* Fail queued try locks */
786 set_bit(GLF_BLOCKING, &gl->gl_flags);
787 if ((gl->gl_req == LM_ST_UNLOCKED) ||
788 (gl->gl_state == LM_ST_EXCLUSIVE) ||
789 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
790 clear_bit(GLF_BLOCKING, &gl->gl_flags);
791 spin_unlock(&gl->gl_lockref.lock);
792 if (glops->go_sync) {
793 ret = glops->go_sync(gl);
794 /* If we had a problem syncing (due to io errors or whatever,
795 * we should not invalidate the metadata or tell dlm to
796 * release the glock to other nodes.
799 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
800 fs_err(sdp, "Error %d syncing glock \n", ret);
801 gfs2_dump_glock(NULL, gl, true);
806 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
808 * The call to go_sync should have cleared out the ail list.
809 * If there are still items, we have a problem. We ought to
810 * withdraw, but we can't because the withdraw code also uses
811 * glocks. Warn about the error, dump the glock, then fall
812 * through and wait for logd to do the withdraw for us.
814 if ((atomic_read(&gl->gl_ail_count) != 0) &&
815 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
816 gfs2_glock_assert_warn(gl,
817 !atomic_read(&gl->gl_ail_count));
818 gfs2_dump_glock(NULL, gl, true);
820 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
821 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
827 * Check for an error encountered since we called go_sync and go_inval.
828 * If so, we can't withdraw from the glock code because the withdraw
829 * code itself uses glocks (see function signal_our_withdraw) to
830 * change the mount to read-only. Most importantly, we must not call
831 * dlm to unlock the glock until the journal is in a known good state
832 * (after journal replay) otherwise other nodes may use the object
833 * (rgrp or dinode) and then later, journal replay will corrupt the
834 * file system. The best we can do here is wait for the logd daemon
835 * to see sd_log_error and withdraw, and in the meantime, requeue the
838 * We make a special exception for some system glocks, such as the
839 * system statfs inode glock, which needs to be granted before the
840 * gfs2_quotad daemon can exit, and that exit needs to finish before
841 * we can unmount the withdrawn file system.
843 * However, if we're just unlocking the lock (say, for unmount, when
844 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
845 * then it's okay to tell dlm to unlock it.
847 if (unlikely(sdp->sd_log_error && !gfs2_withdrawn(sdp)))
848 gfs2_withdraw_delayed(sdp);
849 if (glock_blocked_by_withdraw(gl) &&
850 (target != LM_ST_UNLOCKED ||
851 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
852 if (!is_system_glock(gl)) {
853 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
856 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
860 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
862 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
863 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
864 target == LM_ST_UNLOCKED &&
865 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
866 finish_xmote(gl, target);
867 gfs2_glock_queue_work(gl, 0);
869 fs_err(sdp, "lm_lock ret %d\n", ret);
870 GLOCK_BUG_ON(gl, !gfs2_withdrawn(sdp));
872 } else { /* lock_nolock */
873 finish_xmote(gl, target);
874 gfs2_glock_queue_work(gl, 0);
877 spin_lock(&gl->gl_lockref.lock);
881 * run_queue - do all outstanding tasks related to a glock
882 * @gl: The glock in question
883 * @nonblock: True if we must not block in run_queue
887 static void run_queue(struct gfs2_glock *gl, const int nonblock)
888 __releases(&gl->gl_lockref.lock)
889 __acquires(&gl->gl_lockref.lock)
891 struct gfs2_holder *gh = NULL;
894 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
897 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
899 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
900 gl->gl_demote_state != gl->gl_state) {
901 if (find_first_holder(gl))
905 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
906 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
907 gl->gl_target = gl->gl_demote_state;
909 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
910 gfs2_demote_wake(gl);
911 ret = do_promote(gl);
916 gh = find_first_waiter(gl);
917 gl->gl_target = gh->gh_state;
918 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
919 do_error(gl, 0); /* Fail queued try locks */
921 do_xmote(gl, gh, gl->gl_target);
926 clear_bit(GLF_LOCK, &gl->gl_flags);
927 smp_mb__after_atomic();
928 gl->gl_lockref.count++;
929 __gfs2_glock_queue_work(gl, 0);
933 clear_bit(GLF_LOCK, &gl->gl_flags);
934 smp_mb__after_atomic();
938 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
940 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
942 if (ri->ri_magic == 0)
943 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
944 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
945 ri->ri_generation_deleted = cpu_to_be64(generation);
948 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
950 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
952 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
954 return generation <= be64_to_cpu(ri->ri_generation_deleted);
957 static void gfs2_glock_poke(struct gfs2_glock *gl)
959 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
960 struct gfs2_holder gh;
963 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
964 error = gfs2_glock_nq(&gh);
967 gfs2_holder_uninit(&gh);
970 static bool gfs2_try_evict(struct gfs2_glock *gl)
972 struct gfs2_inode *ip;
973 bool evicted = false;
976 * If there is contention on the iopen glock and we have an inode, try
977 * to grab and release the inode so that it can be evicted. This will
978 * allow the remote node to go ahead and delete the inode without us
979 * having to do it, which will avoid rgrp glock thrashing.
981 * The remote node is likely still holding the corresponding inode
982 * glock, so it will run before we get to verify that the delete has
985 spin_lock(&gl->gl_lockref.lock);
987 if (ip && !igrab(&ip->i_inode))
989 spin_unlock(&gl->gl_lockref.lock);
991 struct gfs2_glock *inode_gl = NULL;
993 gl->gl_no_formal_ino = ip->i_no_formal_ino;
994 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
995 d_prune_aliases(&ip->i_inode);
998 /* If the inode was evicted, gl->gl_object will now be NULL. */
999 spin_lock(&gl->gl_lockref.lock);
1002 inode_gl = ip->i_gl;
1003 lockref_get(&inode_gl->gl_lockref);
1004 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
1006 spin_unlock(&gl->gl_lockref.lock);
1008 gfs2_glock_poke(inode_gl);
1009 gfs2_glock_put(inode_gl);
1016 static void delete_work_func(struct work_struct *work)
1018 struct delayed_work *dwork = to_delayed_work(work);
1019 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1020 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1021 struct inode *inode;
1022 u64 no_addr = gl->gl_name.ln_number;
1024 spin_lock(&gl->gl_lockref.lock);
1025 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
1026 spin_unlock(&gl->gl_lockref.lock);
1028 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1030 * If we can evict the inode, give the remote node trying to
1031 * delete the inode some time before verifying that the delete
1032 * has happened. Otherwise, if we cause contention on the inode glock
1033 * immediately, the remote node will think that we still have
1034 * the inode in use, and so it will give up waiting.
1036 * If we can't evict the inode, signal to the remote node that
1037 * the inode is still in use. We'll later try to delete the
1038 * inode locally in gfs2_evict_inode.
1040 * FIXME: We only need to verify that the remote node has
1041 * deleted the inode because nodes before this remote delete
1042 * rework won't cooperate. At a later time, when we no longer
1043 * care about compatibility with such nodes, we can skip this
1046 if (gfs2_try_evict(gl)) {
1047 if (gfs2_queue_delete_work(gl, 5 * HZ))
1053 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1054 GFS2_BLKST_UNLINKED);
1055 if (!IS_ERR_OR_NULL(inode)) {
1056 d_prune_aliases(inode);
1063 static void glock_work_func(struct work_struct *work)
1065 unsigned long delay = 0;
1066 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1067 unsigned int drop_refs = 1;
1069 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1070 finish_xmote(gl, gl->gl_reply);
1073 spin_lock(&gl->gl_lockref.lock);
1074 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1075 gl->gl_state != LM_ST_UNLOCKED &&
1076 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1077 unsigned long holdtime, now = jiffies;
1079 holdtime = gl->gl_tchange + gl->gl_hold_time;
1080 if (time_before(now, holdtime))
1081 delay = holdtime - now;
1084 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1085 gfs2_set_demote(gl);
1090 /* Keep one glock reference for the work we requeue. */
1092 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1094 __gfs2_glock_queue_work(gl, delay);
1098 * Drop the remaining glock references manually here. (Mind that
1099 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1102 gl->gl_lockref.count -= drop_refs;
1103 if (!gl->gl_lockref.count) {
1104 __gfs2_glock_put(gl);
1107 spin_unlock(&gl->gl_lockref.lock);
1110 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1111 struct gfs2_glock *new)
1113 struct wait_glock_queue wait;
1114 wait_queue_head_t *wq = glock_waitqueue(name);
1115 struct gfs2_glock *gl;
1118 init_wait(&wait.wait);
1119 wait.wait.func = glock_wake_function;
1122 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1125 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1126 &new->gl_node, ht_parms);
1130 gl = rhashtable_lookup_fast(&gl_hash_table,
1133 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1140 finish_wait(wq, &wait.wait);
1145 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1146 * @sdp: The GFS2 superblock
1147 * @number: the lock number
1148 * @glops: The glock_operations to use
1149 * @create: If 0, don't create the glock if it doesn't exist
1150 * @glp: the glock is returned here
1152 * This does not lock a glock, just finds/creates structures for one.
1157 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1158 const struct gfs2_glock_operations *glops, int create,
1159 struct gfs2_glock **glp)
1161 struct super_block *s = sdp->sd_vfs;
1162 struct lm_lockname name = { .ln_number = number,
1163 .ln_type = glops->go_type,
1165 struct gfs2_glock *gl, *tmp;
1166 struct address_space *mapping;
1167 struct kmem_cache *cachep;
1170 gl = find_insert_glock(&name, NULL);
1178 if (glops->go_flags & GLOF_ASPACE)
1179 cachep = gfs2_glock_aspace_cachep;
1181 cachep = gfs2_glock_cachep;
1182 gl = kmem_cache_alloc(cachep, GFP_NOFS);
1186 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1188 if (glops->go_flags & GLOF_LVB) {
1189 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1190 if (!gl->gl_lksb.sb_lvbptr) {
1191 kmem_cache_free(cachep, gl);
1196 atomic_inc(&sdp->sd_glock_disposal);
1197 gl->gl_node.next = NULL;
1198 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1200 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1201 gl->gl_lockref.count = 1;
1202 gl->gl_state = LM_ST_UNLOCKED;
1203 gl->gl_target = LM_ST_UNLOCKED;
1204 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1208 /* We use the global stats to estimate the initial per-glock stats */
1209 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1211 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1212 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1213 gl->gl_tchange = jiffies;
1214 gl->gl_object = NULL;
1215 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1216 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1217 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1218 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1220 mapping = gfs2_glock2aspace(gl);
1222 mapping->a_ops = &gfs2_meta_aops;
1223 mapping->host = s->s_bdev->bd_inode;
1225 mapping_set_gfp_mask(mapping, GFP_NOFS);
1226 mapping->private_data = NULL;
1227 mapping->writeback_index = 0;
1230 tmp = find_insert_glock(&name, gl);
1242 kfree(gl->gl_lksb.sb_lvbptr);
1243 kmem_cache_free(cachep, gl);
1244 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1245 wake_up(&sdp->sd_glock_wait);
1252 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
1254 * @state: the state we're requesting
1255 * @flags: the modifier flags
1256 * @gh: the holder structure
1260 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1261 struct gfs2_holder *gh, unsigned long ip)
1263 INIT_LIST_HEAD(&gh->gh_list);
1266 gh->gh_owner_pid = get_pid(task_pid(current));
1267 gh->gh_state = state;
1268 gh->gh_flags = flags;
1271 gfs2_glock_hold(gl);
1275 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1276 * @state: the state we're requesting
1277 * @flags: the modifier flags
1278 * @gh: the holder structure
1280 * Don't mess with the glock.
1284 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1286 gh->gh_state = state;
1287 gh->gh_flags = flags;
1289 gh->gh_ip = _RET_IP_;
1290 put_pid(gh->gh_owner_pid);
1291 gh->gh_owner_pid = get_pid(task_pid(current));
1295 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1296 * @gh: the holder structure
1300 void gfs2_holder_uninit(struct gfs2_holder *gh)
1302 put_pid(gh->gh_owner_pid);
1303 gfs2_glock_put(gh->gh_gl);
1304 gfs2_holder_mark_uninitialized(gh);
1308 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1309 unsigned long start_time)
1311 /* Have we waited longer that a second? */
1312 if (time_after(jiffies, start_time + HZ)) {
1313 /* Lengthen the minimum hold time. */
1314 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1320 * gfs2_glock_wait - wait on a glock acquisition
1321 * @gh: the glock holder
1323 * Returns: 0 on success
1326 int gfs2_glock_wait(struct gfs2_holder *gh)
1328 unsigned long start_time = jiffies;
1331 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1332 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1333 return gh->gh_error;
1336 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1340 for (i = 0; i < num_gh; i++)
1341 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1347 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1348 * @num_gh: the number of holders in the array
1349 * @ghs: the glock holder array
1351 * Returns: 0 on success, meaning all glocks have been granted and are held.
1352 * -ESTALE if the request timed out, meaning all glocks were released,
1353 * and the caller should retry the operation.
1356 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1358 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1359 int i, ret = 0, timeout = 0;
1360 unsigned long start_time = jiffies;
1365 * Total up the (minimum hold time * 2) of all glocks and use that to
1366 * determine the max amount of time we should wait.
1368 for (i = 0; i < num_gh; i++)
1369 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1372 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1373 !glocks_pending(num_gh, ghs), timeout))
1374 ret = -ESTALE; /* request timed out. */
1377 * If dlm granted all our requests, we need to adjust the glock
1378 * minimum hold time values according to how long we waited.
1380 * If our request timed out, we need to repeatedly release any held
1381 * glocks we acquired thus far to allow dlm to acquire the remaining
1382 * glocks without deadlocking. We cannot currently cancel outstanding
1383 * glock acquisitions.
1385 * The HIF_WAIT bit tells us which requests still need a response from
1388 * If dlm sent us any errors, we return the first error we find.
1390 keep_waiting = false;
1391 for (i = 0; i < num_gh; i++) {
1392 /* Skip holders we have already dequeued below. */
1393 if (!gfs2_holder_queued(&ghs[i]))
1395 /* Skip holders with a pending DLM response. */
1396 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags)) {
1397 keep_waiting = true;
1401 if (test_bit(HIF_HOLDER, &ghs[i].gh_iflags)) {
1403 gfs2_glock_dq(&ghs[i]);
1405 gfs2_glock_update_hold_time(ghs[i].gh_gl,
1409 ret = ghs[i].gh_error;
1416 * At this point, we've either acquired all locks or released them all.
1422 * handle_callback - process a demote request
1424 * @state: the state the caller wants us to change to
1425 * @delay: zero to demote immediately; otherwise pending demote
1426 * @remote: true if this came from a different cluster node
1428 * There are only two requests that we are going to see in actual
1429 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1432 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1433 unsigned long delay, bool remote)
1436 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1438 gfs2_set_demote(gl);
1439 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1440 gl->gl_demote_state = state;
1441 gl->gl_demote_time = jiffies;
1442 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1443 gl->gl_demote_state != state) {
1444 gl->gl_demote_state = LM_ST_UNLOCKED;
1446 if (gl->gl_ops->go_callback)
1447 gl->gl_ops->go_callback(gl, remote);
1448 trace_gfs2_demote_rq(gl, remote);
1451 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1453 struct va_format vaf;
1456 va_start(args, fmt);
1459 seq_vprintf(seq, fmt, args);
1464 pr_err("%pV", &vaf);
1471 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1472 * @gh: the holder structure to add
1474 * Eventually we should move the recursive locking trap to a
1475 * debugging option or something like that. This is the fast
1476 * path and needs to have the minimum number of distractions.
1480 static inline void add_to_queue(struct gfs2_holder *gh)
1481 __releases(&gl->gl_lockref.lock)
1482 __acquires(&gl->gl_lockref.lock)
1484 struct gfs2_glock *gl = gh->gh_gl;
1485 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1486 struct list_head *insert_pt = NULL;
1487 struct gfs2_holder *gh2;
1490 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1491 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1492 GLOCK_BUG_ON(gl, true);
1494 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1495 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1496 struct gfs2_holder *first_gh;
1498 first_gh = find_first_strong_holder(gl);
1499 try_futile = !may_grant(gl, first_gh, gh);
1501 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1505 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1506 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
1507 (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK) &&
1508 !test_bit(HIF_MAY_DEMOTE, &gh2->gh_iflags)))
1509 goto trap_recursive;
1511 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1513 gh->gh_error = GLR_TRYFAILED;
1514 gfs2_holder_wake(gh);
1517 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1519 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1520 insert_pt = &gh2->gh_list;
1522 trace_gfs2_glock_queue(gh, 1);
1523 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1524 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1525 if (likely(insert_pt == NULL)) {
1526 list_add_tail(&gh->gh_list, &gl->gl_holders);
1527 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1531 list_add_tail(&gh->gh_list, insert_pt);
1533 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
1534 if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1535 spin_unlock(&gl->gl_lockref.lock);
1536 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1537 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1538 spin_lock(&gl->gl_lockref.lock);
1543 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1544 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1545 fs_err(sdp, "lock type: %d req lock state : %d\n",
1546 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1547 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1548 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1549 fs_err(sdp, "lock type: %d req lock state : %d\n",
1550 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1551 gfs2_dump_glock(NULL, gl, true);
1556 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1557 * @gh: the holder structure
1559 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1561 * Returns: 0, GLR_TRYFAILED, or errno on failure
1564 int gfs2_glock_nq(struct gfs2_holder *gh)
1566 struct gfs2_glock *gl = gh->gh_gl;
1569 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1572 if (test_bit(GLF_LRU, &gl->gl_flags))
1573 gfs2_glock_remove_from_lru(gl);
1575 spin_lock(&gl->gl_lockref.lock);
1577 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1578 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1579 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1580 gl->gl_lockref.count++;
1581 __gfs2_glock_queue_work(gl, 0);
1584 spin_unlock(&gl->gl_lockref.lock);
1586 if (!(gh->gh_flags & GL_ASYNC))
1587 error = gfs2_glock_wait(gh);
1593 * gfs2_glock_poll - poll to see if an async request has been completed
1596 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1599 int gfs2_glock_poll(struct gfs2_holder *gh)
1601 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1604 static inline bool needs_demote(struct gfs2_glock *gl)
1606 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1607 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1610 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1612 struct gfs2_glock *gl = gh->gh_gl;
1613 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1618 * This while loop is similar to function demote_incompat_holders:
1619 * If the glock is due to be demoted (which may be from another node
1620 * or even if this holder is GL_NOCACHE), the weak holders are
1621 * demoted as well, allowing the glock to be demoted.
1625 * If we're in the process of file system withdraw, we cannot
1626 * just dequeue any glocks until our journal is recovered, lest
1627 * we introduce file system corruption. We need two exceptions
1628 * to this rule: We need to allow unlocking of nondisk glocks
1629 * and the glock for our own journal that needs recovery.
1631 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1632 glock_blocked_by_withdraw(gl) &&
1633 gh->gh_gl != sdp->sd_jinode_gl) {
1634 sdp->sd_glock_dqs_held++;
1635 spin_unlock(&gl->gl_lockref.lock);
1637 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1638 TASK_UNINTERRUPTIBLE);
1639 spin_lock(&gl->gl_lockref.lock);
1643 * This holder should not be cached, so mark it for demote.
1644 * Note: this should be done before the check for needs_demote
1647 if (gh->gh_flags & GL_NOCACHE)
1648 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1650 list_del_init(&gh->gh_list);
1651 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1652 trace_gfs2_glock_queue(gh, 0);
1655 * If there hasn't been a demote request we are done.
1656 * (Let the remaining holders, if any, keep holding it.)
1658 if (!needs_demote(gl)) {
1659 if (list_empty(&gl->gl_holders))
1664 * If we have another strong holder (we cannot auto-demote)
1665 * we are done. It keeps holding it until it is done.
1667 if (find_first_strong_holder(gl))
1671 * If we have a weak holder at the head of the list, it
1672 * (and all others like it) must be auto-demoted. If there
1673 * are no more weak holders, we exit the while loop.
1675 gh = find_first_holder(gl);
1678 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1679 gfs2_glock_add_to_lru(gl);
1681 if (unlikely(!fast_path)) {
1682 gl->gl_lockref.count++;
1683 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1684 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1685 gl->gl_name.ln_type == LM_TYPE_INODE)
1686 delay = gl->gl_hold_time;
1687 __gfs2_glock_queue_work(gl, delay);
1692 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1693 * @gh: the glock holder
1696 void gfs2_glock_dq(struct gfs2_holder *gh)
1698 struct gfs2_glock *gl = gh->gh_gl;
1700 spin_lock(&gl->gl_lockref.lock);
1701 __gfs2_glock_dq(gh);
1702 spin_unlock(&gl->gl_lockref.lock);
1705 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1707 struct gfs2_glock *gl = gh->gh_gl;
1710 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1714 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1715 * @gh: the holder structure
1719 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1722 gfs2_holder_uninit(gh);
1726 * gfs2_glock_nq_num - acquire a glock based on lock number
1727 * @sdp: the filesystem
1728 * @number: the lock number
1729 * @glops: the glock operations for the type of glock
1730 * @state: the state to acquire the glock in
1731 * @flags: modifier flags for the acquisition
1732 * @gh: the struct gfs2_holder
1737 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1738 const struct gfs2_glock_operations *glops,
1739 unsigned int state, u16 flags, struct gfs2_holder *gh)
1741 struct gfs2_glock *gl;
1744 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1746 error = gfs2_glock_nq_init(gl, state, flags, gh);
1754 * glock_compare - Compare two struct gfs2_glock structures for sorting
1755 * @arg_a: the first structure
1756 * @arg_b: the second structure
1760 static int glock_compare(const void *arg_a, const void *arg_b)
1762 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1763 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1764 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1765 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1767 if (a->ln_number > b->ln_number)
1769 if (a->ln_number < b->ln_number)
1771 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1776 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1777 * @num_gh: the number of structures
1778 * @ghs: an array of struct gfs2_holder structures
1779 * @p: placeholder for the holder structure to pass back
1781 * Returns: 0 on success (all glocks acquired),
1782 * errno on failure (no glocks acquired)
1785 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1786 struct gfs2_holder **p)
1791 for (x = 0; x < num_gh; x++)
1794 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1796 for (x = 0; x < num_gh; x++) {
1797 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1799 error = gfs2_glock_nq(p[x]);
1802 gfs2_glock_dq(p[x]);
1811 * gfs2_glock_nq_m - acquire multiple glocks
1812 * @num_gh: the number of structures
1813 * @ghs: an array of struct gfs2_holder structures
1816 * Returns: 0 on success (all glocks acquired),
1817 * errno on failure (no glocks acquired)
1820 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1822 struct gfs2_holder *tmp[4];
1823 struct gfs2_holder **pph = tmp;
1830 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1831 return gfs2_glock_nq(ghs);
1835 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1841 error = nq_m_sync(num_gh, ghs, pph);
1850 * gfs2_glock_dq_m - release multiple glocks
1851 * @num_gh: the number of structures
1852 * @ghs: an array of struct gfs2_holder structures
1856 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1859 gfs2_glock_dq(&ghs[num_gh]);
1862 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1864 struct gfs2_holder mock_gh = { .gh_gl = gl, .gh_state = state, };
1865 unsigned long delay = 0;
1866 unsigned long holdtime;
1867 unsigned long now = jiffies;
1869 gfs2_glock_hold(gl);
1870 spin_lock(&gl->gl_lockref.lock);
1871 holdtime = gl->gl_tchange + gl->gl_hold_time;
1872 if (!list_empty(&gl->gl_holders) &&
1873 gl->gl_name.ln_type == LM_TYPE_INODE) {
1874 if (time_before(now, holdtime))
1875 delay = holdtime - now;
1876 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1877 delay = gl->gl_hold_time;
1880 * Note 1: We cannot call demote_incompat_holders from handle_callback
1881 * or gfs2_set_demote due to recursion problems like: gfs2_glock_dq ->
1882 * handle_callback -> demote_incompat_holders -> gfs2_glock_dq
1883 * Plus, we only want to demote the holders if the request comes from
1884 * a remote cluster node because local holder conflicts are resolved
1887 * Note 2: if a remote node wants this glock in EX mode, lock_dlm will
1888 * request that we set our state to UNLOCKED. Here we mock up a holder
1889 * to make it look like someone wants the lock EX locally. Any SH
1890 * and DF requests should be able to share the lock without demoting.
1892 * Note 3: We only want to demote the demoteable holders when there
1893 * are no more strong holders. The demoteable holders might as well
1894 * keep the glock until the last strong holder is done with it.
1896 if (!find_first_strong_holder(gl)) {
1897 if (state == LM_ST_UNLOCKED)
1898 mock_gh.gh_state = LM_ST_EXCLUSIVE;
1899 demote_incompat_holders(gl, &mock_gh);
1901 handle_callback(gl, state, delay, true);
1902 __gfs2_glock_queue_work(gl, delay);
1903 spin_unlock(&gl->gl_lockref.lock);
1907 * gfs2_should_freeze - Figure out if glock should be frozen
1908 * @gl: The glock in question
1910 * Glocks are not frozen if (a) the result of the dlm operation is
1911 * an error, (b) the locking operation was an unlock operation or
1912 * (c) if there is a "noexp" flagged request anywhere in the queue
1914 * Returns: 1 if freezing should occur, 0 otherwise
1917 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1919 const struct gfs2_holder *gh;
1921 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1923 if (gl->gl_target == LM_ST_UNLOCKED)
1926 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1927 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1929 if (LM_FLAG_NOEXP & gh->gh_flags)
1937 * gfs2_glock_complete - Callback used by locking
1938 * @gl: Pointer to the glock
1939 * @ret: The return value from the dlm
1941 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1942 * to use a bitfield shared with other glock state fields.
1945 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1947 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1949 spin_lock(&gl->gl_lockref.lock);
1952 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1953 if (gfs2_should_freeze(gl)) {
1954 set_bit(GLF_FROZEN, &gl->gl_flags);
1955 spin_unlock(&gl->gl_lockref.lock);
1960 gl->gl_lockref.count++;
1961 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1962 __gfs2_glock_queue_work(gl, 0);
1963 spin_unlock(&gl->gl_lockref.lock);
1966 static int glock_cmp(void *priv, const struct list_head *a,
1967 const struct list_head *b)
1969 struct gfs2_glock *gla, *glb;
1971 gla = list_entry(a, struct gfs2_glock, gl_lru);
1972 glb = list_entry(b, struct gfs2_glock, gl_lru);
1974 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1976 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1983 * gfs2_dispose_glock_lru - Demote a list of glocks
1984 * @list: The list to dispose of
1986 * Disposing of glocks may involve disk accesses, so that here we sort
1987 * the glocks by number (i.e. disk location of the inodes) so that if
1988 * there are any such accesses, they'll be sent in order (mostly).
1990 * Must be called under the lru_lock, but may drop and retake this
1991 * lock. While the lru_lock is dropped, entries may vanish from the
1992 * list, but no new entries will appear on the list (since it is
1996 static void gfs2_dispose_glock_lru(struct list_head *list)
1997 __releases(&lru_lock)
1998 __acquires(&lru_lock)
2000 struct gfs2_glock *gl;
2002 list_sort(NULL, list, glock_cmp);
2004 while(!list_empty(list)) {
2005 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2006 list_del_init(&gl->gl_lru);
2007 clear_bit(GLF_LRU, &gl->gl_flags);
2008 if (!spin_trylock(&gl->gl_lockref.lock)) {
2010 list_add(&gl->gl_lru, &lru_list);
2011 set_bit(GLF_LRU, &gl->gl_flags);
2012 atomic_inc(&lru_count);
2015 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2016 spin_unlock(&gl->gl_lockref.lock);
2017 goto add_back_to_lru;
2019 gl->gl_lockref.count++;
2021 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2022 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2023 __gfs2_glock_queue_work(gl, 0);
2024 spin_unlock(&gl->gl_lockref.lock);
2025 cond_resched_lock(&lru_lock);
2030 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2031 * @nr: The number of entries to scan
2033 * This function selects the entries on the LRU which are able to
2034 * be demoted, and then kicks off the process by calling
2035 * gfs2_dispose_glock_lru() above.
2038 static long gfs2_scan_glock_lru(int nr)
2040 struct gfs2_glock *gl;
2045 spin_lock(&lru_lock);
2046 while ((nr-- >= 0) && !list_empty(&lru_list)) {
2047 gl = list_first_entry(&lru_list, struct gfs2_glock, gl_lru);
2049 /* Test for being demotable */
2050 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2051 list_move(&gl->gl_lru, &dispose);
2052 atomic_dec(&lru_count);
2057 list_move(&gl->gl_lru, &skipped);
2059 list_splice(&skipped, &lru_list);
2060 if (!list_empty(&dispose))
2061 gfs2_dispose_glock_lru(&dispose);
2062 spin_unlock(&lru_lock);
2067 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2068 struct shrink_control *sc)
2070 if (!(sc->gfp_mask & __GFP_FS))
2072 return gfs2_scan_glock_lru(sc->nr_to_scan);
2075 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2076 struct shrink_control *sc)
2078 return vfs_pressure_ratio(atomic_read(&lru_count));
2081 static struct shrinker glock_shrinker = {
2082 .seeks = DEFAULT_SEEKS,
2083 .count_objects = gfs2_glock_shrink_count,
2084 .scan_objects = gfs2_glock_shrink_scan,
2088 * glock_hash_walk - Call a function for glock in a hash bucket
2089 * @examiner: the function
2090 * @sdp: the filesystem
2092 * Note that the function can be called multiple times on the same
2093 * object. So the user must ensure that the function can cope with
2097 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2099 struct gfs2_glock *gl;
2100 struct rhashtable_iter iter;
2102 rhashtable_walk_enter(&gl_hash_table, &iter);
2105 rhashtable_walk_start(&iter);
2107 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2108 if (gl->gl_name.ln_sbd == sdp)
2112 rhashtable_walk_stop(&iter);
2113 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2115 rhashtable_walk_exit(&iter);
2118 bool gfs2_queue_delete_work(struct gfs2_glock *gl, unsigned long delay)
2122 spin_lock(&gl->gl_lockref.lock);
2123 queued = queue_delayed_work(gfs2_delete_workqueue,
2124 &gl->gl_delete, delay);
2126 set_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2127 spin_unlock(&gl->gl_lockref.lock);
2131 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2133 if (cancel_delayed_work(&gl->gl_delete)) {
2134 clear_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2139 bool gfs2_delete_work_queued(const struct gfs2_glock *gl)
2141 return test_bit(GLF_PENDING_DELETE, &gl->gl_flags);
2144 static void flush_delete_work(struct gfs2_glock *gl)
2146 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2147 if (cancel_delayed_work(&gl->gl_delete)) {
2148 queue_delayed_work(gfs2_delete_workqueue,
2154 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2156 glock_hash_walk(flush_delete_work, sdp);
2157 flush_workqueue(gfs2_delete_workqueue);
2161 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2162 * @gl: The glock to thaw
2166 static void thaw_glock(struct gfs2_glock *gl)
2168 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2170 if (!lockref_get_not_dead(&gl->gl_lockref))
2172 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2173 gfs2_glock_queue_work(gl, 0);
2177 * clear_glock - look at a glock and see if we can free it from glock cache
2178 * @gl: the glock to look at
2182 static void clear_glock(struct gfs2_glock *gl)
2184 gfs2_glock_remove_from_lru(gl);
2186 spin_lock(&gl->gl_lockref.lock);
2187 if (!__lockref_is_dead(&gl->gl_lockref)) {
2188 gl->gl_lockref.count++;
2189 if (gl->gl_state != LM_ST_UNLOCKED)
2190 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2191 __gfs2_glock_queue_work(gl, 0);
2193 spin_unlock(&gl->gl_lockref.lock);
2197 * gfs2_glock_thaw - Thaw any frozen glocks
2198 * @sdp: The super block
2202 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2204 glock_hash_walk(thaw_glock, sdp);
2207 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2209 spin_lock(&gl->gl_lockref.lock);
2210 gfs2_dump_glock(seq, gl, fsid);
2211 spin_unlock(&gl->gl_lockref.lock);
2214 static void dump_glock_func(struct gfs2_glock *gl)
2216 dump_glock(NULL, gl, true);
2220 * gfs2_gl_hash_clear - Empty out the glock hash table
2221 * @sdp: the filesystem
2223 * Called when unmounting the filesystem.
2226 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2228 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2229 flush_workqueue(glock_workqueue);
2230 glock_hash_walk(clear_glock, sdp);
2231 flush_workqueue(glock_workqueue);
2232 wait_event_timeout(sdp->sd_glock_wait,
2233 atomic_read(&sdp->sd_glock_disposal) == 0,
2235 glock_hash_walk(dump_glock_func, sdp);
2238 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
2240 struct gfs2_glock *gl = ip->i_gl;
2243 ret = gfs2_truncatei_resume(ip);
2244 gfs2_glock_assert_withdraw(gl, ret == 0);
2246 spin_lock(&gl->gl_lockref.lock);
2247 clear_bit(GLF_LOCK, &gl->gl_flags);
2249 spin_unlock(&gl->gl_lockref.lock);
2252 static const char *state2str(unsigned state)
2255 case LM_ST_UNLOCKED:
2259 case LM_ST_DEFERRED:
2261 case LM_ST_EXCLUSIVE:
2267 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2270 if (flags & LM_FLAG_TRY)
2272 if (flags & LM_FLAG_TRY_1CB)
2274 if (flags & LM_FLAG_NOEXP)
2276 if (flags & LM_FLAG_ANY)
2278 if (flags & LM_FLAG_PRIORITY)
2280 if (flags & LM_FLAG_NODE_SCOPE)
2282 if (flags & GL_ASYNC)
2284 if (flags & GL_EXACT)
2286 if (flags & GL_NOCACHE)
2288 if (test_bit(HIF_HOLDER, &iflags))
2290 if (test_bit(HIF_WAIT, &iflags))
2292 if (test_bit(HIF_MAY_DEMOTE, &iflags))
2294 if (flags & GL_SKIP)
2301 * dump_holder - print information about a glock holder
2302 * @seq: the seq_file struct
2303 * @gh: the glock holder
2304 * @fs_id_buf: pointer to file system id (if requested)
2308 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2309 const char *fs_id_buf)
2311 struct task_struct *gh_owner = NULL;
2315 if (gh->gh_owner_pid)
2316 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2317 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2318 fs_id_buf, state2str(gh->gh_state),
2319 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2321 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
2322 gh_owner ? gh_owner->comm : "(ended)",
2327 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2329 const unsigned long *gflags = &gl->gl_flags;
2332 if (test_bit(GLF_LOCK, gflags))
2334 if (test_bit(GLF_DEMOTE, gflags))
2336 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2338 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2340 if (test_bit(GLF_DIRTY, gflags))
2342 if (test_bit(GLF_LFLUSH, gflags))
2344 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2346 if (test_bit(GLF_REPLY_PENDING, gflags))
2348 if (test_bit(GLF_INITIAL, gflags))
2350 if (test_bit(GLF_FROZEN, gflags))
2352 if (!list_empty(&gl->gl_holders))
2354 if (test_bit(GLF_LRU, gflags))
2358 if (test_bit(GLF_BLOCKING, gflags))
2360 if (test_bit(GLF_PENDING_DELETE, gflags))
2362 if (test_bit(GLF_FREEING, gflags))
2364 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2366 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2373 * gfs2_dump_glock - print information about a glock
2374 * @seq: The seq_file struct
2376 * @fsid: If true, also dump the file system id
2378 * The file format is as follows:
2379 * One line per object, capital letters are used to indicate objects
2380 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2381 * other objects are indented by a single space and follow the glock to
2382 * which they are related. Fields are indicated by lower case letters
2383 * followed by a colon and the field value, except for strings which are in
2384 * [] so that its possible to see if they are composed of spaces for
2385 * example. The field's are n = number (id of the object), f = flags,
2386 * t = type, s = state, r = refcount, e = error, p = pid.
2390 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2392 const struct gfs2_glock_operations *glops = gl->gl_ops;
2393 unsigned long long dtime;
2394 const struct gfs2_holder *gh;
2395 char gflags_buf[32];
2396 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2397 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2398 unsigned long nrpages = 0;
2400 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2401 struct address_space *mapping = gfs2_glock2aspace(gl);
2403 nrpages = mapping->nrpages;
2405 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2406 if (fsid && sdp) /* safety precaution */
2407 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2408 dtime = jiffies - gl->gl_demote_time;
2409 dtime *= 1000000/HZ; /* demote time in uSec */
2410 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2412 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2413 "v:%d r:%d m:%ld p:%lu\n",
2414 fs_id_buf, state2str(gl->gl_state),
2415 gl->gl_name.ln_type,
2416 (unsigned long long)gl->gl_name.ln_number,
2417 gflags2str(gflags_buf, gl),
2418 state2str(gl->gl_target),
2419 state2str(gl->gl_demote_state), dtime,
2420 atomic_read(&gl->gl_ail_count),
2421 atomic_read(&gl->gl_revokes),
2422 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2424 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2425 dump_holder(seq, gh, fs_id_buf);
2427 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2428 glops->go_dump(seq, gl, fs_id_buf);
2431 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2433 struct gfs2_glock *gl = iter_ptr;
2435 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2436 gl->gl_name.ln_type,
2437 (unsigned long long)gl->gl_name.ln_number,
2438 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2439 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2440 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2441 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2442 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2443 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2444 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2445 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2449 static const char *gfs2_gltype[] = {
2463 static const char *gfs2_stype[] = {
2464 [GFS2_LKS_SRTT] = "srtt",
2465 [GFS2_LKS_SRTTVAR] = "srttvar",
2466 [GFS2_LKS_SRTTB] = "srttb",
2467 [GFS2_LKS_SRTTVARB] = "srttvarb",
2468 [GFS2_LKS_SIRT] = "sirt",
2469 [GFS2_LKS_SIRTVAR] = "sirtvar",
2470 [GFS2_LKS_DCOUNT] = "dlm",
2471 [GFS2_LKS_QCOUNT] = "queue",
2474 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2476 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2478 struct gfs2_sbd *sdp = seq->private;
2479 loff_t pos = *(loff_t *)iter_ptr;
2480 unsigned index = pos >> 3;
2481 unsigned subindex = pos & 0x07;
2484 if (index == 0 && subindex != 0)
2487 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2488 (index == 0) ? "cpu": gfs2_stype[subindex]);
2490 for_each_possible_cpu(i) {
2491 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2494 seq_printf(seq, " %15u", i);
2496 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2497 lkstats[index - 1].stats[subindex]);
2499 seq_putc(seq, '\n');
2503 int __init gfs2_glock_init(void)
2507 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2511 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2512 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2513 if (!glock_workqueue) {
2514 rhashtable_destroy(&gl_hash_table);
2517 gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
2518 WQ_MEM_RECLAIM | WQ_FREEZABLE,
2520 if (!gfs2_delete_workqueue) {
2521 destroy_workqueue(glock_workqueue);
2522 rhashtable_destroy(&gl_hash_table);
2526 ret = register_shrinker(&glock_shrinker);
2528 destroy_workqueue(gfs2_delete_workqueue);
2529 destroy_workqueue(glock_workqueue);
2530 rhashtable_destroy(&gl_hash_table);
2534 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2535 init_waitqueue_head(glock_wait_table + i);
2540 void gfs2_glock_exit(void)
2542 unregister_shrinker(&glock_shrinker);
2543 rhashtable_destroy(&gl_hash_table);
2544 destroy_workqueue(glock_workqueue);
2545 destroy_workqueue(gfs2_delete_workqueue);
2548 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2550 struct gfs2_glock *gl = gi->gl;
2555 if (!lockref_put_not_zero(&gl->gl_lockref))
2556 gfs2_glock_queue_put(gl);
2559 gl = rhashtable_walk_next(&gi->hti);
2560 if (IS_ERR_OR_NULL(gl)) {
2561 if (gl == ERR_PTR(-EAGAIN)) {
2568 if (gl->gl_name.ln_sbd != gi->sdp)
2571 if (!lockref_get_not_dead(&gl->gl_lockref))
2575 if (__lockref_is_dead(&gl->gl_lockref))
2583 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2586 struct gfs2_glock_iter *gi = seq->private;
2590 * We can either stay where we are, skip to the next hash table
2591 * entry, or start from the beginning.
2593 if (*pos < gi->last_pos) {
2594 rhashtable_walk_exit(&gi->hti);
2595 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2598 n = *pos - gi->last_pos;
2601 rhashtable_walk_start(&gi->hti);
2603 gfs2_glock_iter_next(gi, n);
2604 gi->last_pos = *pos;
2608 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2611 struct gfs2_glock_iter *gi = seq->private;
2614 gi->last_pos = *pos;
2615 gfs2_glock_iter_next(gi, 1);
2619 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2622 struct gfs2_glock_iter *gi = seq->private;
2624 rhashtable_walk_stop(&gi->hti);
2627 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2629 dump_glock(seq, iter_ptr, false);
2633 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2636 if (*pos >= GFS2_NR_SBSTATS)
2641 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2645 if (*pos >= GFS2_NR_SBSTATS)
2650 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2655 static const struct seq_operations gfs2_glock_seq_ops = {
2656 .start = gfs2_glock_seq_start,
2657 .next = gfs2_glock_seq_next,
2658 .stop = gfs2_glock_seq_stop,
2659 .show = gfs2_glock_seq_show,
2662 static const struct seq_operations gfs2_glstats_seq_ops = {
2663 .start = gfs2_glock_seq_start,
2664 .next = gfs2_glock_seq_next,
2665 .stop = gfs2_glock_seq_stop,
2666 .show = gfs2_glstats_seq_show,
2669 static const struct seq_operations gfs2_sbstats_sops = {
2670 .start = gfs2_sbstats_seq_start,
2671 .next = gfs2_sbstats_seq_next,
2672 .stop = gfs2_sbstats_seq_stop,
2673 .show = gfs2_sbstats_seq_show,
2676 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2678 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2679 const struct seq_operations *ops)
2681 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2683 struct seq_file *seq = file->private_data;
2684 struct gfs2_glock_iter *gi = seq->private;
2686 gi->sdp = inode->i_private;
2687 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2689 seq->size = GFS2_SEQ_GOODSIZE;
2691 * Initially, we are "before" the first hash table entry; the
2692 * first call to rhashtable_walk_next gets us the first entry.
2696 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2701 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2703 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2706 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2708 struct seq_file *seq = file->private_data;
2709 struct gfs2_glock_iter *gi = seq->private;
2712 gfs2_glock_put(gi->gl);
2713 rhashtable_walk_exit(&gi->hti);
2714 return seq_release_private(inode, file);
2717 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2719 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2722 static const struct file_operations gfs2_glocks_fops = {
2723 .owner = THIS_MODULE,
2724 .open = gfs2_glocks_open,
2726 .llseek = seq_lseek,
2727 .release = gfs2_glocks_release,
2730 static const struct file_operations gfs2_glstats_fops = {
2731 .owner = THIS_MODULE,
2732 .open = gfs2_glstats_open,
2734 .llseek = seq_lseek,
2735 .release = gfs2_glocks_release,
2738 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2740 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2742 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2744 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2747 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2748 &gfs2_glstats_fops);
2750 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2751 &gfs2_sbstats_fops);
2754 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2756 debugfs_remove_recursive(sdp->debugfs_dir);
2757 sdp->debugfs_dir = NULL;
2760 void gfs2_register_debugfs(void)
2762 gfs2_root = debugfs_create_dir("gfs2", NULL);
2765 void gfs2_unregister_debugfs(void)
2767 debugfs_remove(gfs2_root);