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 static LIST_HEAD(lru_list);
71 static atomic_t lru_count = ATOMIC_INIT(0);
72 static DEFINE_SPINLOCK(lru_lock);
74 #define GFS2_GL_HASH_SHIFT 15
75 #define GFS2_GL_HASH_SIZE BIT(GFS2_GL_HASH_SHIFT)
77 static const struct rhashtable_params ht_parms = {
78 .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
79 .key_len = offsetofend(struct lm_lockname, ln_type),
80 .key_offset = offsetof(struct gfs2_glock, gl_name),
81 .head_offset = offsetof(struct gfs2_glock, gl_node),
84 static struct rhashtable gl_hash_table;
86 #define GLOCK_WAIT_TABLE_BITS 12
87 #define GLOCK_WAIT_TABLE_SIZE (1 << GLOCK_WAIT_TABLE_BITS)
88 static wait_queue_head_t glock_wait_table[GLOCK_WAIT_TABLE_SIZE] __cacheline_aligned;
90 struct wait_glock_queue {
91 struct lm_lockname *name;
92 wait_queue_entry_t wait;
95 static int glock_wake_function(wait_queue_entry_t *wait, unsigned int mode,
98 struct wait_glock_queue *wait_glock =
99 container_of(wait, struct wait_glock_queue, wait);
100 struct lm_lockname *wait_name = wait_glock->name;
101 struct lm_lockname *wake_name = key;
103 if (wake_name->ln_sbd != wait_name->ln_sbd ||
104 wake_name->ln_number != wait_name->ln_number ||
105 wake_name->ln_type != wait_name->ln_type)
107 return autoremove_wake_function(wait, mode, sync, key);
110 static wait_queue_head_t *glock_waitqueue(struct lm_lockname *name)
112 u32 hash = jhash2((u32 *)name, ht_parms.key_len / 4, 0);
114 return glock_wait_table + hash_32(hash, GLOCK_WAIT_TABLE_BITS);
118 * wake_up_glock - Wake up waiters on a glock
121 static void wake_up_glock(struct gfs2_glock *gl)
123 wait_queue_head_t *wq = glock_waitqueue(&gl->gl_name);
125 if (waitqueue_active(wq))
126 __wake_up(wq, TASK_NORMAL, 1, &gl->gl_name);
129 static void gfs2_glock_dealloc(struct rcu_head *rcu)
131 struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
133 kfree(gl->gl_lksb.sb_lvbptr);
134 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
135 struct gfs2_glock_aspace *gla =
136 container_of(gl, struct gfs2_glock_aspace, glock);
137 kmem_cache_free(gfs2_glock_aspace_cachep, gla);
139 kmem_cache_free(gfs2_glock_cachep, gl);
143 * glock_blocked_by_withdraw - determine if we can still use a glock
146 * We need to allow some glocks to be enqueued, dequeued, promoted, and demoted
147 * when we're withdrawn. For example, to maintain metadata integrity, we should
148 * disallow the use of inode and rgrp glocks when withdrawn. Other glocks like
149 * the iopen or freeze glock may be safely used because none of their
150 * metadata goes through the journal. So in general, we should disallow all
151 * glocks that are journaled, and allow all the others. One exception is:
152 * we need to allow our active journal to be promoted and demoted so others
153 * may recover it and we can reacquire it when they're done.
155 static bool glock_blocked_by_withdraw(struct gfs2_glock *gl)
157 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
159 if (!gfs2_withdrawing_or_withdrawn(sdp))
161 if (gl->gl_ops->go_flags & GLOF_NONDISK)
163 if (!sdp->sd_jdesc ||
164 gl->gl_name.ln_number == sdp->sd_jdesc->jd_no_addr)
169 void gfs2_glock_free(struct gfs2_glock *gl)
171 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
173 gfs2_glock_assert_withdraw(gl, atomic_read(&gl->gl_revokes) == 0);
174 rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
177 call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
178 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
179 wake_up(&sdp->sd_kill_wait);
183 * gfs2_glock_hold() - increment reference count on glock
184 * @gl: The glock to hold
188 struct gfs2_glock *gfs2_glock_hold(struct gfs2_glock *gl)
190 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
191 lockref_get(&gl->gl_lockref);
196 * demote_ok - Check to see if it's ok to unlock a glock
199 * Returns: 1 if it's ok
202 static int demote_ok(const struct gfs2_glock *gl)
204 const struct gfs2_glock_operations *glops = gl->gl_ops;
206 if (gl->gl_state == LM_ST_UNLOCKED)
208 if (!list_empty(&gl->gl_holders))
210 if (glops->go_demote_ok)
211 return glops->go_demote_ok(gl);
216 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
218 if (!(gl->gl_ops->go_flags & GLOF_LRU))
221 spin_lock(&lru_lock);
223 list_move_tail(&gl->gl_lru, &lru_list);
225 if (!test_bit(GLF_LRU, &gl->gl_flags)) {
226 set_bit(GLF_LRU, &gl->gl_flags);
227 atomic_inc(&lru_count);
230 spin_unlock(&lru_lock);
233 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
235 if (!(gl->gl_ops->go_flags & GLOF_LRU))
238 spin_lock(&lru_lock);
239 if (test_bit(GLF_LRU, &gl->gl_flags)) {
240 list_del_init(&gl->gl_lru);
241 atomic_dec(&lru_count);
242 clear_bit(GLF_LRU, &gl->gl_flags);
244 spin_unlock(&lru_lock);
248 * Enqueue the glock on the work queue. Passes one glock reference on to the
251 static void __gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
252 if (!queue_delayed_work(glock_workqueue, &gl->gl_work, delay)) {
254 * We are holding the lockref spinlock, and the work was still
255 * queued above. The queued work (glock_work_func) takes that
256 * spinlock before dropping its glock reference(s), so it
257 * cannot have dropped them in the meantime.
259 GLOCK_BUG_ON(gl, gl->gl_lockref.count < 2);
260 gl->gl_lockref.count--;
264 static void gfs2_glock_queue_work(struct gfs2_glock *gl, unsigned long delay) {
265 spin_lock(&gl->gl_lockref.lock);
266 __gfs2_glock_queue_work(gl, delay);
267 spin_unlock(&gl->gl_lockref.lock);
270 static void __gfs2_glock_put(struct gfs2_glock *gl)
272 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
273 struct address_space *mapping = gfs2_glock2aspace(gl);
275 lockref_mark_dead(&gl->gl_lockref);
276 spin_unlock(&gl->gl_lockref.lock);
277 gfs2_glock_remove_from_lru(gl);
278 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
280 truncate_inode_pages_final(mapping);
281 if (!gfs2_withdrawing_or_withdrawn(sdp))
282 GLOCK_BUG_ON(gl, !mapping_empty(mapping));
284 trace_gfs2_glock_put(gl);
285 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
289 * Cause the glock to be put in work queue context.
291 void gfs2_glock_queue_put(struct gfs2_glock *gl)
293 gfs2_glock_queue_work(gl, 0);
297 * gfs2_glock_put() - Decrement reference count on glock
298 * @gl: The glock to put
302 void gfs2_glock_put(struct gfs2_glock *gl)
304 if (lockref_put_or_lock(&gl->gl_lockref))
307 __gfs2_glock_put(gl);
311 * may_grant - check if it's ok to grant a new lock
313 * @current_gh: One of the current holders of @gl
314 * @gh: The lock request which we wish to grant
316 * With our current compatibility rules, if a glock has one or more active
317 * holders (HIF_HOLDER flag set), any of those holders can be passed in as
318 * @current_gh; they are all the same as far as compatibility with the new @gh
321 * Returns true if it's ok to grant the lock.
324 static inline bool may_grant(struct gfs2_glock *gl,
325 struct gfs2_holder *current_gh,
326 struct gfs2_holder *gh)
329 GLOCK_BUG_ON(gl, !test_bit(HIF_HOLDER, ¤t_gh->gh_iflags));
331 switch(current_gh->gh_state) {
332 case LM_ST_EXCLUSIVE:
334 * Here we make a special exception to grant holders
335 * who agree to share the EX lock with other holders
336 * who also have the bit set. If the original holder
337 * has the LM_FLAG_NODE_SCOPE bit set, we grant more
338 * holders with the bit set.
340 return gh->gh_state == LM_ST_EXCLUSIVE &&
341 (current_gh->gh_flags & LM_FLAG_NODE_SCOPE) &&
342 (gh->gh_flags & LM_FLAG_NODE_SCOPE);
346 return gh->gh_state == current_gh->gh_state;
353 if (gl->gl_state == gh->gh_state)
355 if (gh->gh_flags & GL_EXACT)
357 if (gl->gl_state == LM_ST_EXCLUSIVE) {
358 return gh->gh_state == LM_ST_SHARED ||
359 gh->gh_state == LM_ST_DEFERRED;
361 if (gh->gh_flags & LM_FLAG_ANY)
362 return gl->gl_state != LM_ST_UNLOCKED;
366 static void gfs2_holder_wake(struct gfs2_holder *gh)
368 clear_bit(HIF_WAIT, &gh->gh_iflags);
369 smp_mb__after_atomic();
370 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
371 if (gh->gh_flags & GL_ASYNC) {
372 struct gfs2_sbd *sdp = gh->gh_gl->gl_name.ln_sbd;
374 wake_up(&sdp->sd_async_glock_wait);
379 * do_error - Something unexpected has happened during a lock request
381 * @ret: The status from the DLM
384 static void do_error(struct gfs2_glock *gl, const int ret)
386 struct gfs2_holder *gh, *tmp;
388 list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
389 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
391 if (ret & LM_OUT_ERROR)
393 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
394 gh->gh_error = GLR_TRYFAILED;
397 list_del_init(&gh->gh_list);
398 trace_gfs2_glock_queue(gh, 0);
399 gfs2_holder_wake(gh);
404 * find_first_holder - find the first "holder" gh
408 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
410 struct gfs2_holder *gh;
412 if (!list_empty(&gl->gl_holders)) {
413 gh = list_first_entry(&gl->gl_holders, struct gfs2_holder,
415 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
422 * gfs2_instantiate - Call the glops instantiate function
423 * @gh: The glock holder
425 * Returns: 0 if instantiate was successful, or error.
427 int gfs2_instantiate(struct gfs2_holder *gh)
429 struct gfs2_glock *gl = gh->gh_gl;
430 const struct gfs2_glock_operations *glops = gl->gl_ops;
434 if (!test_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags))
438 * Since we unlock the lockref lock, we set a flag to indicate
439 * instantiate is in progress.
441 if (test_and_set_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags)) {
442 wait_on_bit(&gl->gl_flags, GLF_INSTANTIATE_IN_PROG,
443 TASK_UNINTERRUPTIBLE);
445 * Here we just waited for a different instantiate to finish.
446 * But that may not have been successful, as when a process
447 * locks an inode glock _before_ it has an actual inode to
448 * instantiate into. So we check again. This process might
449 * have an inode to instantiate, so might be successful.
454 ret = glops->go_instantiate(gl);
456 clear_bit(GLF_INSTANTIATE_NEEDED, &gl->gl_flags);
457 clear_and_wake_up_bit(GLF_INSTANTIATE_IN_PROG, &gl->gl_flags);
463 return glops->go_held(gh);
468 * do_promote - promote as many requests as possible on the current queue
471 * Returns true on success (i.e., progress was made or there are no waiters).
474 static bool do_promote(struct gfs2_glock *gl)
476 struct gfs2_holder *gh, *current_gh;
478 current_gh = find_first_holder(gl);
479 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
480 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
482 if (!may_grant(gl, current_gh, gh)) {
484 * If we get here, it means we may not grant this
485 * holder for some reason. If this holder is at the
486 * head of the list, it means we have a blocked holder
487 * at the head, so return false.
489 if (list_is_first(&gh->gh_list, &gl->gl_holders))
494 set_bit(HIF_HOLDER, &gh->gh_iflags);
495 trace_gfs2_promote(gh);
496 gfs2_holder_wake(gh);
504 * find_first_waiter - find the first gh that's waiting for the glock
508 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
510 struct gfs2_holder *gh;
512 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
513 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
520 * find_last_waiter - find the last gh that's waiting for the glock
523 * This also is a fast way of finding out if there are any waiters.
526 static inline struct gfs2_holder *find_last_waiter(const struct gfs2_glock *gl)
528 struct gfs2_holder *gh;
530 if (list_empty(&gl->gl_holders))
532 gh = list_last_entry(&gl->gl_holders, struct gfs2_holder, gh_list);
533 return test_bit(HIF_HOLDER, &gh->gh_iflags) ? NULL : gh;
537 * state_change - record that the glock is now in a different state
539 * @new_state: the new state
542 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
546 held1 = (gl->gl_state != LM_ST_UNLOCKED);
547 held2 = (new_state != LM_ST_UNLOCKED);
549 if (held1 != held2) {
550 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
552 gl->gl_lockref.count++;
554 gl->gl_lockref.count--;
556 if (new_state != gl->gl_target)
557 /* shorten our minimum hold time */
558 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
560 gl->gl_state = new_state;
561 gl->gl_tchange = jiffies;
564 static void gfs2_set_demote(struct gfs2_glock *gl)
566 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
568 set_bit(GLF_DEMOTE, &gl->gl_flags);
570 wake_up(&sdp->sd_async_glock_wait);
573 static void gfs2_demote_wake(struct gfs2_glock *gl)
575 gl->gl_demote_state = LM_ST_EXCLUSIVE;
576 clear_bit(GLF_DEMOTE, &gl->gl_flags);
577 smp_mb__after_atomic();
578 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
582 * finish_xmote - The DLM has replied to one of our lock requests
584 * @ret: The status from the DLM
588 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
590 const struct gfs2_glock_operations *glops = gl->gl_ops;
591 struct gfs2_holder *gh;
592 unsigned state = ret & LM_OUT_ST_MASK;
594 spin_lock(&gl->gl_lockref.lock);
595 trace_gfs2_glock_state_change(gl, state);
596 state_change(gl, state);
597 gh = find_first_waiter(gl);
599 /* Demote to UN request arrived during demote to SH or DF */
600 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
601 state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
602 gl->gl_target = LM_ST_UNLOCKED;
604 /* Check for state != intended state */
605 if (unlikely(state != gl->gl_target)) {
606 if (gh && (ret & LM_OUT_CANCELED))
607 gfs2_holder_wake(gh);
608 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
609 /* move to back of queue and try next entry */
610 if (ret & LM_OUT_CANCELED) {
611 list_move_tail(&gh->gh_list, &gl->gl_holders);
612 gh = find_first_waiter(gl);
613 gl->gl_target = gh->gh_state;
618 /* Some error or failed "try lock" - report it */
619 if ((ret & LM_OUT_ERROR) ||
620 (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
621 gl->gl_target = gl->gl_state;
627 /* Unlocked due to conversion deadlock, try again */
630 do_xmote(gl, gh, gl->gl_target);
632 /* Conversion fails, unlock and try again */
635 do_xmote(gl, gh, LM_ST_UNLOCKED);
637 default: /* Everything else */
638 fs_err(gl->gl_name.ln_sbd, "wanted %u got %u\n",
639 gl->gl_target, state);
642 spin_unlock(&gl->gl_lockref.lock);
646 /* Fast path - we got what we asked for */
647 if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
648 gfs2_demote_wake(gl);
649 if (state != LM_ST_UNLOCKED) {
650 if (glops->go_xmote_bh) {
653 spin_unlock(&gl->gl_lockref.lock);
654 rv = glops->go_xmote_bh(gl);
655 spin_lock(&gl->gl_lockref.lock);
664 clear_bit(GLF_LOCK, &gl->gl_flags);
665 spin_unlock(&gl->gl_lockref.lock);
668 static bool is_system_glock(struct gfs2_glock *gl)
670 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
671 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
673 if (gl == m_ip->i_gl)
679 * do_xmote - Calls the DLM to change the state of a lock
680 * @gl: The lock state
681 * @gh: The holder (only for promotes)
682 * @target: The target lock state
686 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh,
688 __releases(&gl->gl_lockref.lock)
689 __acquires(&gl->gl_lockref.lock)
691 const struct gfs2_glock_operations *glops = gl->gl_ops;
692 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
693 unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
696 if (target != LM_ST_UNLOCKED && glock_blocked_by_withdraw(gl) &&
697 gh && !(gh->gh_flags & LM_FLAG_NOEXP))
700 lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP);
701 GLOCK_BUG_ON(gl, gl->gl_state == target);
702 GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
703 if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
706 * If another process is already doing the invalidate, let that
707 * finish first. The glock state machine will get back to this
708 * holder again later.
710 if (test_and_set_bit(GLF_INVALIDATE_IN_PROGRESS,
713 do_error(gl, 0); /* Fail queued try locks */
716 set_bit(GLF_BLOCKING, &gl->gl_flags);
717 if ((gl->gl_req == LM_ST_UNLOCKED) ||
718 (gl->gl_state == LM_ST_EXCLUSIVE) ||
719 (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
720 clear_bit(GLF_BLOCKING, &gl->gl_flags);
721 spin_unlock(&gl->gl_lockref.lock);
722 if (glops->go_sync) {
723 ret = glops->go_sync(gl);
724 /* If we had a problem syncing (due to io errors or whatever,
725 * we should not invalidate the metadata or tell dlm to
726 * release the glock to other nodes.
729 if (cmpxchg(&sdp->sd_log_error, 0, ret)) {
730 fs_err(sdp, "Error %d syncing glock \n", ret);
731 gfs2_dump_glock(NULL, gl, true);
736 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags)) {
738 * The call to go_sync should have cleared out the ail list.
739 * If there are still items, we have a problem. We ought to
740 * withdraw, but we can't because the withdraw code also uses
741 * glocks. Warn about the error, dump the glock, then fall
742 * through and wait for logd to do the withdraw for us.
744 if ((atomic_read(&gl->gl_ail_count) != 0) &&
745 (!cmpxchg(&sdp->sd_log_error, 0, -EIO))) {
746 gfs2_glock_assert_warn(gl,
747 !atomic_read(&gl->gl_ail_count));
748 gfs2_dump_glock(NULL, gl, true);
750 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
751 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
757 * Check for an error encountered since we called go_sync and go_inval.
758 * If so, we can't withdraw from the glock code because the withdraw
759 * code itself uses glocks (see function signal_our_withdraw) to
760 * change the mount to read-only. Most importantly, we must not call
761 * dlm to unlock the glock until the journal is in a known good state
762 * (after journal replay) otherwise other nodes may use the object
763 * (rgrp or dinode) and then later, journal replay will corrupt the
764 * file system. The best we can do here is wait for the logd daemon
765 * to see sd_log_error and withdraw, and in the meantime, requeue the
768 * We make a special exception for some system glocks, such as the
769 * system statfs inode glock, which needs to be granted before the
770 * gfs2_quotad daemon can exit, and that exit needs to finish before
771 * we can unmount the withdrawn file system.
773 * However, if we're just unlocking the lock (say, for unmount, when
774 * gfs2_gl_hash_clear calls clear_glock) and recovery is complete
775 * then it's okay to tell dlm to unlock it.
777 if (unlikely(sdp->sd_log_error) && !gfs2_withdrawing_or_withdrawn(sdp))
778 gfs2_withdraw_delayed(sdp);
779 if (glock_blocked_by_withdraw(gl) &&
780 (target != LM_ST_UNLOCKED ||
781 test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags))) {
782 if (!is_system_glock(gl)) {
783 handle_callback(gl, LM_ST_UNLOCKED, 0, false); /* sets demote */
785 * Ordinarily, we would call dlm and its callback would call
786 * finish_xmote, which would call state_change() to the new state.
787 * Since we withdrew, we won't call dlm, so call state_change
788 * manually, but to the UNLOCKED state we desire.
790 state_change(gl, LM_ST_UNLOCKED);
792 * We skip telling dlm to do the locking, so we won't get a
793 * reply that would otherwise clear GLF_LOCK. So we clear it here.
795 clear_bit(GLF_LOCK, &gl->gl_flags);
796 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
797 gfs2_glock_queue_work(gl, GL_GLOCK_DFT_HOLD);
800 clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
804 if (sdp->sd_lockstruct.ls_ops->lm_lock) {
806 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
807 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
808 target == LM_ST_UNLOCKED &&
809 test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
810 finish_xmote(gl, target);
811 gfs2_glock_queue_work(gl, 0);
813 fs_err(sdp, "lm_lock ret %d\n", ret);
814 GLOCK_BUG_ON(gl, !gfs2_withdrawing_or_withdrawn(sdp));
816 } else { /* lock_nolock */
817 finish_xmote(gl, target);
818 gfs2_glock_queue_work(gl, 0);
821 spin_lock(&gl->gl_lockref.lock);
825 * run_queue - do all outstanding tasks related to a glock
826 * @gl: The glock in question
827 * @nonblock: True if we must not block in run_queue
831 static void run_queue(struct gfs2_glock *gl, const int nonblock)
832 __releases(&gl->gl_lockref.lock)
833 __acquires(&gl->gl_lockref.lock)
835 struct gfs2_holder *gh = NULL;
837 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
840 GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
842 if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
843 gl->gl_demote_state != gl->gl_state) {
844 if (find_first_holder(gl))
848 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
849 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
850 gl->gl_target = gl->gl_demote_state;
852 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
853 gfs2_demote_wake(gl);
856 gh = find_first_waiter(gl);
857 gl->gl_target = gh->gh_state;
858 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
859 do_error(gl, 0); /* Fail queued try locks */
861 do_xmote(gl, gh, gl->gl_target);
865 clear_bit(GLF_LOCK, &gl->gl_flags);
866 smp_mb__after_atomic();
867 gl->gl_lockref.count++;
868 __gfs2_glock_queue_work(gl, 0);
872 clear_bit(GLF_LOCK, &gl->gl_flags);
873 smp_mb__after_atomic();
878 * glock_set_object - set the gl_object field of a glock
880 * @object: the object
882 void glock_set_object(struct gfs2_glock *gl, void *object)
886 spin_lock(&gl->gl_lockref.lock);
887 prev_object = gl->gl_object;
888 gl->gl_object = object;
889 spin_unlock(&gl->gl_lockref.lock);
890 if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == NULL)) {
891 pr_warn("glock=%u/%llx\n",
893 (unsigned long long)gl->gl_name.ln_number);
894 gfs2_dump_glock(NULL, gl, true);
899 * glock_clear_object - clear the gl_object field of a glock
901 * @object: object the glock currently points at
903 void glock_clear_object(struct gfs2_glock *gl, void *object)
907 spin_lock(&gl->gl_lockref.lock);
908 prev_object = gl->gl_object;
909 gl->gl_object = NULL;
910 spin_unlock(&gl->gl_lockref.lock);
911 if (gfs2_assert_warn(gl->gl_name.ln_sbd, prev_object == object)) {
912 pr_warn("glock=%u/%llx\n",
914 (unsigned long long)gl->gl_name.ln_number);
915 gfs2_dump_glock(NULL, gl, true);
919 void gfs2_inode_remember_delete(struct gfs2_glock *gl, u64 generation)
921 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
923 if (ri->ri_magic == 0)
924 ri->ri_magic = cpu_to_be32(GFS2_MAGIC);
925 if (ri->ri_magic == cpu_to_be32(GFS2_MAGIC))
926 ri->ri_generation_deleted = cpu_to_be64(generation);
929 bool gfs2_inode_already_deleted(struct gfs2_glock *gl, u64 generation)
931 struct gfs2_inode_lvb *ri = (void *)gl->gl_lksb.sb_lvbptr;
933 if (ri->ri_magic != cpu_to_be32(GFS2_MAGIC))
935 return generation <= be64_to_cpu(ri->ri_generation_deleted);
938 static void gfs2_glock_poke(struct gfs2_glock *gl)
940 int flags = LM_FLAG_TRY_1CB | LM_FLAG_ANY | GL_SKIP;
941 struct gfs2_holder gh;
944 __gfs2_holder_init(gl, LM_ST_SHARED, flags, &gh, _RET_IP_);
945 error = gfs2_glock_nq(&gh);
948 gfs2_holder_uninit(&gh);
951 static bool gfs2_try_evict(struct gfs2_glock *gl)
953 struct gfs2_inode *ip;
954 bool evicted = false;
957 * If there is contention on the iopen glock and we have an inode, try
958 * to grab and release the inode so that it can be evicted. This will
959 * allow the remote node to go ahead and delete the inode without us
960 * having to do it, which will avoid rgrp glock thrashing.
962 * The remote node is likely still holding the corresponding inode
963 * glock, so it will run before we get to verify that the delete has
966 spin_lock(&gl->gl_lockref.lock);
968 if (ip && !igrab(&ip->i_inode))
970 spin_unlock(&gl->gl_lockref.lock);
972 gl->gl_no_formal_ino = ip->i_no_formal_ino;
973 set_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
974 d_prune_aliases(&ip->i_inode);
977 /* If the inode was evicted, gl->gl_object will now be NULL. */
978 spin_lock(&gl->gl_lockref.lock);
981 clear_bit(GIF_DEFERRED_DELETE, &ip->i_flags);
982 if (!igrab(&ip->i_inode))
985 spin_unlock(&gl->gl_lockref.lock);
987 gfs2_glock_poke(ip->i_gl);
995 bool gfs2_queue_try_to_evict(struct gfs2_glock *gl)
997 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
999 if (test_and_set_bit(GLF_TRY_TO_EVICT, &gl->gl_flags))
1001 return queue_delayed_work(sdp->sd_delete_wq,
1005 static bool gfs2_queue_verify_evict(struct gfs2_glock *gl)
1007 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1009 if (test_and_set_bit(GLF_VERIFY_EVICT, &gl->gl_flags))
1011 return queue_delayed_work(sdp->sd_delete_wq,
1012 &gl->gl_delete, 5 * HZ);
1015 static void delete_work_func(struct work_struct *work)
1017 struct delayed_work *dwork = to_delayed_work(work);
1018 struct gfs2_glock *gl = container_of(dwork, struct gfs2_glock, gl_delete);
1019 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1020 struct inode *inode;
1021 u64 no_addr = gl->gl_name.ln_number;
1023 if (test_and_clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags)) {
1025 * If we can evict the inode, give the remote node trying to
1026 * delete the inode some time before verifying that the delete
1027 * has happened. Otherwise, if we cause contention on the inode glock
1028 * immediately, the remote node will think that we still have
1029 * the inode in use, and so it will give up waiting.
1031 * If we can't evict the inode, signal to the remote node that
1032 * the inode is still in use. We'll later try to delete the
1033 * inode locally in gfs2_evict_inode.
1035 * FIXME: We only need to verify that the remote node has
1036 * deleted the inode because nodes before this remote delete
1037 * rework won't cooperate. At a later time, when we no longer
1038 * care about compatibility with such nodes, we can skip this
1041 if (gfs2_try_evict(gl)) {
1042 if (test_bit(SDF_KILL, &sdp->sd_flags))
1044 if (gfs2_queue_verify_evict(gl))
1050 if (test_and_clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags)) {
1051 inode = gfs2_lookup_by_inum(sdp, no_addr, gl->gl_no_formal_ino,
1052 GFS2_BLKST_UNLINKED);
1053 if (IS_ERR(inode)) {
1054 if (PTR_ERR(inode) == -EAGAIN &&
1055 !test_bit(SDF_KILL, &sdp->sd_flags) &&
1056 gfs2_queue_verify_evict(gl))
1059 d_prune_aliases(inode);
1068 static void glock_work_func(struct work_struct *work)
1070 unsigned long delay = 0;
1071 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
1072 unsigned int drop_refs = 1;
1074 if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
1075 finish_xmote(gl, gl->gl_reply);
1078 spin_lock(&gl->gl_lockref.lock);
1079 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1080 gl->gl_state != LM_ST_UNLOCKED &&
1081 gl->gl_demote_state != LM_ST_EXCLUSIVE) {
1082 unsigned long holdtime, now = jiffies;
1084 holdtime = gl->gl_tchange + gl->gl_hold_time;
1085 if (time_before(now, holdtime))
1086 delay = holdtime - now;
1089 clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1090 gfs2_set_demote(gl);
1095 /* Keep one glock reference for the work we requeue. */
1097 if (gl->gl_name.ln_type != LM_TYPE_INODE)
1099 __gfs2_glock_queue_work(gl, delay);
1103 * Drop the remaining glock references manually here. (Mind that
1104 * __gfs2_glock_queue_work depends on the lockref spinlock begin held
1107 gl->gl_lockref.count -= drop_refs;
1108 if (!gl->gl_lockref.count) {
1109 __gfs2_glock_put(gl);
1112 spin_unlock(&gl->gl_lockref.lock);
1115 static struct gfs2_glock *find_insert_glock(struct lm_lockname *name,
1116 struct gfs2_glock *new)
1118 struct wait_glock_queue wait;
1119 wait_queue_head_t *wq = glock_waitqueue(name);
1120 struct gfs2_glock *gl;
1123 init_wait(&wait.wait);
1124 wait.wait.func = glock_wake_function;
1127 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1130 gl = rhashtable_lookup_get_insert_fast(&gl_hash_table,
1131 &new->gl_node, ht_parms);
1135 gl = rhashtable_lookup_fast(&gl_hash_table,
1138 if (gl && !lockref_get_not_dead(&gl->gl_lockref)) {
1145 finish_wait(wq, &wait.wait);
1150 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
1151 * @sdp: The GFS2 superblock
1152 * @number: the lock number
1153 * @glops: The glock_operations to use
1154 * @create: If 0, don't create the glock if it doesn't exist
1155 * @glp: the glock is returned here
1157 * This does not lock a glock, just finds/creates structures for one.
1162 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
1163 const struct gfs2_glock_operations *glops, int create,
1164 struct gfs2_glock **glp)
1166 struct super_block *s = sdp->sd_vfs;
1167 struct lm_lockname name = { .ln_number = number,
1168 .ln_type = glops->go_type,
1170 struct gfs2_glock *gl, *tmp;
1171 struct address_space *mapping;
1174 gl = find_insert_glock(&name, NULL);
1182 if (glops->go_flags & GLOF_ASPACE) {
1183 struct gfs2_glock_aspace *gla =
1184 kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_NOFS);
1189 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_NOFS);
1193 memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
1196 if (glops->go_flags & GLOF_LVB) {
1197 gl->gl_lksb.sb_lvbptr = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1198 if (!gl->gl_lksb.sb_lvbptr) {
1199 gfs2_glock_dealloc(&gl->gl_rcu);
1204 atomic_inc(&sdp->sd_glock_disposal);
1205 gl->gl_node.next = NULL;
1206 gl->gl_flags = glops->go_instantiate ? BIT(GLF_INSTANTIATE_NEEDED) : 0;
1208 lockdep_set_subclass(&gl->gl_lockref.lock, glops->go_subclass);
1209 gl->gl_lockref.count = 1;
1210 gl->gl_state = LM_ST_UNLOCKED;
1211 gl->gl_target = LM_ST_UNLOCKED;
1212 gl->gl_demote_state = LM_ST_EXCLUSIVE;
1215 /* We use the global stats to estimate the initial per-glock stats */
1216 gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
1218 gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
1219 gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
1220 gl->gl_tchange = jiffies;
1221 gl->gl_object = NULL;
1222 gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
1223 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
1224 if (gl->gl_name.ln_type == LM_TYPE_IOPEN)
1225 INIT_DELAYED_WORK(&gl->gl_delete, delete_work_func);
1227 mapping = gfs2_glock2aspace(gl);
1229 mapping->a_ops = &gfs2_meta_aops;
1230 mapping->host = s->s_bdev->bd_inode;
1232 mapping_set_gfp_mask(mapping, GFP_NOFS);
1233 mapping->i_private_data = NULL;
1234 mapping->writeback_index = 0;
1237 tmp = find_insert_glock(&name, gl);
1249 gfs2_glock_dealloc(&gl->gl_rcu);
1250 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
1251 wake_up(&sdp->sd_kill_wait);
1258 * __gfs2_holder_init - initialize a struct gfs2_holder in the default way
1260 * @state: the state we're requesting
1261 * @flags: the modifier flags
1262 * @gh: the holder structure
1266 void __gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
1267 struct gfs2_holder *gh, unsigned long ip)
1269 INIT_LIST_HEAD(&gh->gh_list);
1270 gh->gh_gl = gfs2_glock_hold(gl);
1272 gh->gh_owner_pid = get_pid(task_pid(current));
1273 gh->gh_state = state;
1274 gh->gh_flags = flags;
1279 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
1280 * @state: the state we're requesting
1281 * @flags: the modifier flags
1282 * @gh: the holder structure
1284 * Don't mess with the glock.
1288 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
1290 gh->gh_state = state;
1291 gh->gh_flags = flags;
1293 gh->gh_ip = _RET_IP_;
1294 put_pid(gh->gh_owner_pid);
1295 gh->gh_owner_pid = get_pid(task_pid(current));
1299 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
1300 * @gh: the holder structure
1304 void gfs2_holder_uninit(struct gfs2_holder *gh)
1306 put_pid(gh->gh_owner_pid);
1307 gfs2_glock_put(gh->gh_gl);
1308 gfs2_holder_mark_uninitialized(gh);
1312 static void gfs2_glock_update_hold_time(struct gfs2_glock *gl,
1313 unsigned long start_time)
1315 /* Have we waited longer that a second? */
1316 if (time_after(jiffies, start_time + HZ)) {
1317 /* Lengthen the minimum hold time. */
1318 gl->gl_hold_time = min(gl->gl_hold_time + GL_GLOCK_HOLD_INCR,
1324 * gfs2_glock_holder_ready - holder is ready and its error code can be collected
1325 * @gh: the glock holder
1327 * Called when a glock holder no longer needs to be waited for because it is
1328 * now either held (HIF_HOLDER set; gh_error == 0), or acquiring the lock has
1329 * failed (gh_error != 0).
1332 int gfs2_glock_holder_ready(struct gfs2_holder *gh)
1334 if (gh->gh_error || (gh->gh_flags & GL_SKIP))
1335 return gh->gh_error;
1336 gh->gh_error = gfs2_instantiate(gh);
1339 return gh->gh_error;
1343 * gfs2_glock_wait - wait on a glock acquisition
1344 * @gh: the glock holder
1346 * Returns: 0 on success
1349 int gfs2_glock_wait(struct gfs2_holder *gh)
1351 unsigned long start_time = jiffies;
1354 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1355 gfs2_glock_update_hold_time(gh->gh_gl, start_time);
1356 return gfs2_glock_holder_ready(gh);
1359 static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs)
1363 for (i = 0; i < num_gh; i++)
1364 if (test_bit(HIF_WAIT, &ghs[i].gh_iflags))
1370 * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions
1371 * @num_gh: the number of holders in the array
1372 * @ghs: the glock holder array
1374 * Returns: 0 on success, meaning all glocks have been granted and are held.
1375 * -ESTALE if the request timed out, meaning all glocks were released,
1376 * and the caller should retry the operation.
1379 int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs)
1381 struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd;
1382 int i, ret = 0, timeout = 0;
1383 unsigned long start_time = jiffies;
1387 * Total up the (minimum hold time * 2) of all glocks and use that to
1388 * determine the max amount of time we should wait.
1390 for (i = 0; i < num_gh; i++)
1391 timeout += ghs[i].gh_gl->gl_hold_time << 1;
1393 if (!wait_event_timeout(sdp->sd_async_glock_wait,
1394 !glocks_pending(num_gh, ghs), timeout)) {
1395 ret = -ESTALE; /* request timed out. */
1399 for (i = 0; i < num_gh; i++) {
1400 struct gfs2_holder *gh = &ghs[i];
1403 if (test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1404 gfs2_glock_update_hold_time(gh->gh_gl,
1407 ret2 = gfs2_glock_holder_ready(gh);
1414 for (i = 0; i < num_gh; i++) {
1415 struct gfs2_holder *gh = &ghs[i];
1424 * handle_callback - process a demote request
1426 * @state: the state the caller wants us to change to
1427 * @delay: zero to demote immediately; otherwise pending demote
1428 * @remote: true if this came from a different cluster node
1430 * There are only two requests that we are going to see in actual
1431 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
1434 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
1435 unsigned long delay, bool remote)
1438 set_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
1440 gfs2_set_demote(gl);
1441 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
1442 gl->gl_demote_state = state;
1443 gl->gl_demote_time = jiffies;
1444 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
1445 gl->gl_demote_state != state) {
1446 gl->gl_demote_state = LM_ST_UNLOCKED;
1448 if (gl->gl_ops->go_callback)
1449 gl->gl_ops->go_callback(gl, remote);
1450 trace_gfs2_demote_rq(gl, remote);
1453 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
1455 struct va_format vaf;
1458 va_start(args, fmt);
1461 seq_vprintf(seq, fmt, args);
1466 pr_err("%pV", &vaf);
1472 static inline bool pid_is_meaningful(const struct gfs2_holder *gh)
1474 if (!(gh->gh_flags & GL_NOPID))
1476 if (gh->gh_state == LM_ST_UNLOCKED)
1482 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1483 * @gh: the holder structure to add
1485 * Eventually we should move the recursive locking trap to a
1486 * debugging option or something like that. This is the fast
1487 * path and needs to have the minimum number of distractions.
1491 static inline void add_to_queue(struct gfs2_holder *gh)
1492 __releases(&gl->gl_lockref.lock)
1493 __acquires(&gl->gl_lockref.lock)
1495 struct gfs2_glock *gl = gh->gh_gl;
1496 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1497 struct list_head *insert_pt = NULL;
1498 struct gfs2_holder *gh2;
1501 GLOCK_BUG_ON(gl, gh->gh_owner_pid == NULL);
1502 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1503 GLOCK_BUG_ON(gl, true);
1505 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1506 if (test_bit(GLF_LOCK, &gl->gl_flags)) {
1507 struct gfs2_holder *current_gh;
1509 current_gh = find_first_holder(gl);
1510 try_futile = !may_grant(gl, current_gh, gh);
1512 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
1516 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1517 if (likely(gh2->gh_owner_pid != gh->gh_owner_pid))
1519 if (gh->gh_gl->gl_ops->go_type == LM_TYPE_FLOCK)
1521 if (!pid_is_meaningful(gh2))
1523 goto trap_recursive;
1525 list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
1527 !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
1529 gh->gh_error = GLR_TRYFAILED;
1530 gfs2_holder_wake(gh);
1533 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1536 trace_gfs2_glock_queue(gh, 1);
1537 gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
1538 gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
1539 if (likely(insert_pt == NULL)) {
1540 list_add_tail(&gh->gh_list, &gl->gl_holders);
1543 list_add_tail(&gh->gh_list, insert_pt);
1544 spin_unlock(&gl->gl_lockref.lock);
1545 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1546 sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1547 spin_lock(&gl->gl_lockref.lock);
1551 fs_err(sdp, "original: %pSR\n", (void *)gh2->gh_ip);
1552 fs_err(sdp, "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1553 fs_err(sdp, "lock type: %d req lock state : %d\n",
1554 gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1555 fs_err(sdp, "new: %pSR\n", (void *)gh->gh_ip);
1556 fs_err(sdp, "pid: %d\n", pid_nr(gh->gh_owner_pid));
1557 fs_err(sdp, "lock type: %d req lock state : %d\n",
1558 gh->gh_gl->gl_name.ln_type, gh->gh_state);
1559 gfs2_dump_glock(NULL, gl, true);
1564 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1565 * @gh: the holder structure
1567 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1569 * Returns: 0, GLR_TRYFAILED, or errno on failure
1572 int gfs2_glock_nq(struct gfs2_holder *gh)
1574 struct gfs2_glock *gl = gh->gh_gl;
1577 if (glock_blocked_by_withdraw(gl) && !(gh->gh_flags & LM_FLAG_NOEXP))
1580 if (gh->gh_flags & GL_NOBLOCK) {
1581 struct gfs2_holder *current_gh;
1584 spin_lock(&gl->gl_lockref.lock);
1585 if (find_last_waiter(gl))
1587 current_gh = find_first_holder(gl);
1588 if (!may_grant(gl, current_gh, gh))
1590 set_bit(HIF_HOLDER, &gh->gh_iflags);
1591 list_add_tail(&gh->gh_list, &gl->gl_holders);
1592 trace_gfs2_promote(gh);
1595 spin_unlock(&gl->gl_lockref.lock);
1599 if (test_bit(GLF_LRU, &gl->gl_flags))
1600 gfs2_glock_remove_from_lru(gl);
1603 spin_lock(&gl->gl_lockref.lock);
1605 if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
1606 test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
1607 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1608 gl->gl_lockref.count++;
1609 __gfs2_glock_queue_work(gl, 0);
1612 spin_unlock(&gl->gl_lockref.lock);
1615 if (!(gh->gh_flags & GL_ASYNC))
1616 error = gfs2_glock_wait(gh);
1622 * gfs2_glock_poll - poll to see if an async request has been completed
1625 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1628 int gfs2_glock_poll(struct gfs2_holder *gh)
1630 return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1633 static inline bool needs_demote(struct gfs2_glock *gl)
1635 return (test_bit(GLF_DEMOTE, &gl->gl_flags) ||
1636 test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags));
1639 static void __gfs2_glock_dq(struct gfs2_holder *gh)
1641 struct gfs2_glock *gl = gh->gh_gl;
1646 * This holder should not be cached, so mark it for demote.
1647 * Note: this should be done before the check for needs_demote
1650 if (gh->gh_flags & GL_NOCACHE)
1651 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1653 list_del_init(&gh->gh_list);
1654 clear_bit(HIF_HOLDER, &gh->gh_iflags);
1655 trace_gfs2_glock_queue(gh, 0);
1658 * If there hasn't been a demote request we are done.
1659 * (Let the remaining holders, if any, keep holding it.)
1661 if (!needs_demote(gl)) {
1662 if (list_empty(&gl->gl_holders))
1666 if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl))
1667 gfs2_glock_add_to_lru(gl);
1669 if (unlikely(!fast_path)) {
1670 gl->gl_lockref.count++;
1671 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1672 !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1673 gl->gl_name.ln_type == LM_TYPE_INODE)
1674 delay = gl->gl_hold_time;
1675 __gfs2_glock_queue_work(gl, delay);
1680 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1681 * @gh: the glock holder
1684 void gfs2_glock_dq(struct gfs2_holder *gh)
1686 struct gfs2_glock *gl = gh->gh_gl;
1687 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
1689 spin_lock(&gl->gl_lockref.lock);
1690 if (!gfs2_holder_queued(gh)) {
1692 * May have already been dequeued because the locking request
1693 * was GL_ASYNC and it has failed in the meantime.
1698 if (list_is_first(&gh->gh_list, &gl->gl_holders) &&
1699 !test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1700 spin_unlock(&gl->gl_lockref.lock);
1701 gl->gl_name.ln_sbd->sd_lockstruct.ls_ops->lm_cancel(gl);
1702 wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
1703 spin_lock(&gl->gl_lockref.lock);
1707 * If we're in the process of file system withdraw, we cannot just
1708 * dequeue any glocks until our journal is recovered, lest we introduce
1709 * file system corruption. We need two exceptions to this rule: We need
1710 * to allow unlocking of nondisk glocks and the glock for our own
1711 * journal that needs recovery.
1713 if (test_bit(SDF_WITHDRAW_RECOVERY, &sdp->sd_flags) &&
1714 glock_blocked_by_withdraw(gl) &&
1715 gh->gh_gl != sdp->sd_jinode_gl) {
1716 sdp->sd_glock_dqs_held++;
1717 spin_unlock(&gl->gl_lockref.lock);
1719 wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
1720 TASK_UNINTERRUPTIBLE);
1721 spin_lock(&gl->gl_lockref.lock);
1724 __gfs2_glock_dq(gh);
1726 spin_unlock(&gl->gl_lockref.lock);
1729 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1731 struct gfs2_glock *gl = gh->gh_gl;
1734 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1738 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1739 * @gh: the holder structure
1743 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1746 gfs2_holder_uninit(gh);
1750 * gfs2_glock_nq_num - acquire a glock based on lock number
1751 * @sdp: the filesystem
1752 * @number: the lock number
1753 * @glops: the glock operations for the type of glock
1754 * @state: the state to acquire the glock in
1755 * @flags: modifier flags for the acquisition
1756 * @gh: the struct gfs2_holder
1761 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1762 const struct gfs2_glock_operations *glops,
1763 unsigned int state, u16 flags, struct gfs2_holder *gh)
1765 struct gfs2_glock *gl;
1768 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1770 error = gfs2_glock_nq_init(gl, state, flags, gh);
1778 * glock_compare - Compare two struct gfs2_glock structures for sorting
1779 * @arg_a: the first structure
1780 * @arg_b: the second structure
1784 static int glock_compare(const void *arg_a, const void *arg_b)
1786 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1787 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1788 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1789 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1791 if (a->ln_number > b->ln_number)
1793 if (a->ln_number < b->ln_number)
1795 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1800 * nq_m_sync - synchronously acquire more than one glock in deadlock free order
1801 * @num_gh: the number of structures
1802 * @ghs: an array of struct gfs2_holder structures
1803 * @p: placeholder for the holder structure to pass back
1805 * Returns: 0 on success (all glocks acquired),
1806 * errno on failure (no glocks acquired)
1809 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1810 struct gfs2_holder **p)
1815 for (x = 0; x < num_gh; x++)
1818 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1820 for (x = 0; x < num_gh; x++) {
1821 error = gfs2_glock_nq(p[x]);
1824 gfs2_glock_dq(p[x]);
1833 * gfs2_glock_nq_m - acquire multiple glocks
1834 * @num_gh: the number of structures
1835 * @ghs: an array of struct gfs2_holder structures
1837 * Returns: 0 on success (all glocks acquired),
1838 * errno on failure (no glocks acquired)
1841 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1843 struct gfs2_holder *tmp[4];
1844 struct gfs2_holder **pph = tmp;
1851 return gfs2_glock_nq(ghs);
1855 pph = kmalloc_array(num_gh, sizeof(struct gfs2_holder *),
1861 error = nq_m_sync(num_gh, ghs, pph);
1870 * gfs2_glock_dq_m - release multiple glocks
1871 * @num_gh: the number of structures
1872 * @ghs: an array of struct gfs2_holder structures
1876 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1879 gfs2_glock_dq(&ghs[num_gh]);
1882 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1884 unsigned long delay = 0;
1885 unsigned long holdtime;
1886 unsigned long now = jiffies;
1888 gfs2_glock_hold(gl);
1889 spin_lock(&gl->gl_lockref.lock);
1890 holdtime = gl->gl_tchange + gl->gl_hold_time;
1891 if (!list_empty(&gl->gl_holders) &&
1892 gl->gl_name.ln_type == LM_TYPE_INODE) {
1893 if (time_before(now, holdtime))
1894 delay = holdtime - now;
1895 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1896 delay = gl->gl_hold_time;
1898 handle_callback(gl, state, delay, true);
1899 __gfs2_glock_queue_work(gl, delay);
1900 spin_unlock(&gl->gl_lockref.lock);
1904 * gfs2_should_freeze - Figure out if glock should be frozen
1905 * @gl: The glock in question
1907 * Glocks are not frozen if (a) the result of the dlm operation is
1908 * an error, (b) the locking operation was an unlock operation or
1909 * (c) if there is a "noexp" flagged request anywhere in the queue
1911 * Returns: 1 if freezing should occur, 0 otherwise
1914 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1916 const struct gfs2_holder *gh;
1918 if (gl->gl_reply & ~LM_OUT_ST_MASK)
1920 if (gl->gl_target == LM_ST_UNLOCKED)
1923 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1924 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1926 if (LM_FLAG_NOEXP & gh->gh_flags)
1934 * gfs2_glock_complete - Callback used by locking
1935 * @gl: Pointer to the glock
1936 * @ret: The return value from the dlm
1938 * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1939 * to use a bitfield shared with other glock state fields.
1942 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1944 struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1946 spin_lock(&gl->gl_lockref.lock);
1949 if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1950 if (gfs2_should_freeze(gl)) {
1951 set_bit(GLF_FROZEN, &gl->gl_flags);
1952 spin_unlock(&gl->gl_lockref.lock);
1957 gl->gl_lockref.count++;
1958 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1959 __gfs2_glock_queue_work(gl, 0);
1960 spin_unlock(&gl->gl_lockref.lock);
1963 static int glock_cmp(void *priv, const struct list_head *a,
1964 const struct list_head *b)
1966 struct gfs2_glock *gla, *glb;
1968 gla = list_entry(a, struct gfs2_glock, gl_lru);
1969 glb = list_entry(b, struct gfs2_glock, gl_lru);
1971 if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1973 if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1980 * gfs2_dispose_glock_lru - Demote a list of glocks
1981 * @list: The list to dispose of
1983 * Disposing of glocks may involve disk accesses, so that here we sort
1984 * the glocks by number (i.e. disk location of the inodes) so that if
1985 * there are any such accesses, they'll be sent in order (mostly).
1987 * Must be called under the lru_lock, but may drop and retake this
1988 * lock. While the lru_lock is dropped, entries may vanish from the
1989 * list, but no new entries will appear on the list (since it is
1993 static void gfs2_dispose_glock_lru(struct list_head *list)
1994 __releases(&lru_lock)
1995 __acquires(&lru_lock)
1997 struct gfs2_glock *gl;
1999 list_sort(NULL, list, glock_cmp);
2001 while(!list_empty(list)) {
2002 gl = list_first_entry(list, struct gfs2_glock, gl_lru);
2003 list_del_init(&gl->gl_lru);
2004 clear_bit(GLF_LRU, &gl->gl_flags);
2005 if (!spin_trylock(&gl->gl_lockref.lock)) {
2007 list_add(&gl->gl_lru, &lru_list);
2008 set_bit(GLF_LRU, &gl->gl_flags);
2009 atomic_inc(&lru_count);
2012 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
2013 spin_unlock(&gl->gl_lockref.lock);
2014 goto add_back_to_lru;
2016 gl->gl_lockref.count++;
2018 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2019 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
2020 __gfs2_glock_queue_work(gl, 0);
2021 spin_unlock(&gl->gl_lockref.lock);
2022 cond_resched_lock(&lru_lock);
2027 * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
2028 * @nr: The number of entries to scan
2030 * This function selects the entries on the LRU which are able to
2031 * be demoted, and then kicks off the process by calling
2032 * gfs2_dispose_glock_lru() above.
2035 static long gfs2_scan_glock_lru(int nr)
2037 struct gfs2_glock *gl, *next;
2041 spin_lock(&lru_lock);
2042 list_for_each_entry_safe(gl, next, &lru_list, gl_lru) {
2045 /* Test for being demotable */
2046 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
2047 if (!spin_trylock(&gl->gl_lockref.lock))
2049 if (gl->gl_lockref.count <= 1 &&
2050 (gl->gl_state == LM_ST_UNLOCKED ||
2052 list_move(&gl->gl_lru, &dispose);
2053 atomic_dec(&lru_count);
2056 spin_unlock(&gl->gl_lockref.lock);
2059 if (!list_empty(&dispose))
2060 gfs2_dispose_glock_lru(&dispose);
2061 spin_unlock(&lru_lock);
2066 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
2067 struct shrink_control *sc)
2069 if (!(sc->gfp_mask & __GFP_FS))
2071 return gfs2_scan_glock_lru(sc->nr_to_scan);
2074 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
2075 struct shrink_control *sc)
2077 return vfs_pressure_ratio(atomic_read(&lru_count));
2080 static struct shrinker *glock_shrinker;
2083 * glock_hash_walk - Call a function for glock in a hash bucket
2084 * @examiner: the function
2085 * @sdp: the filesystem
2087 * Note that the function can be called multiple times on the same
2088 * object. So the user must ensure that the function can cope with
2092 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
2094 struct gfs2_glock *gl;
2095 struct rhashtable_iter iter;
2097 rhashtable_walk_enter(&gl_hash_table, &iter);
2100 rhashtable_walk_start(&iter);
2102 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl)) {
2103 if (gl->gl_name.ln_sbd == sdp)
2107 rhashtable_walk_stop(&iter);
2108 } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
2110 rhashtable_walk_exit(&iter);
2113 void gfs2_cancel_delete_work(struct gfs2_glock *gl)
2115 clear_bit(GLF_TRY_TO_EVICT, &gl->gl_flags);
2116 clear_bit(GLF_VERIFY_EVICT, &gl->gl_flags);
2117 if (cancel_delayed_work(&gl->gl_delete))
2121 static void flush_delete_work(struct gfs2_glock *gl)
2123 if (gl->gl_name.ln_type == LM_TYPE_IOPEN) {
2124 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2126 if (cancel_delayed_work(&gl->gl_delete)) {
2127 queue_delayed_work(sdp->sd_delete_wq,
2133 void gfs2_flush_delete_work(struct gfs2_sbd *sdp)
2135 glock_hash_walk(flush_delete_work, sdp);
2136 flush_workqueue(sdp->sd_delete_wq);
2140 * thaw_glock - thaw out a glock which has an unprocessed reply waiting
2141 * @gl: The glock to thaw
2145 static void thaw_glock(struct gfs2_glock *gl)
2147 if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
2149 if (!lockref_get_not_dead(&gl->gl_lockref))
2151 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
2152 gfs2_glock_queue_work(gl, 0);
2156 * clear_glock - look at a glock and see if we can free it from glock cache
2157 * @gl: the glock to look at
2161 static void clear_glock(struct gfs2_glock *gl)
2163 gfs2_glock_remove_from_lru(gl);
2165 spin_lock(&gl->gl_lockref.lock);
2166 if (!__lockref_is_dead(&gl->gl_lockref)) {
2167 gl->gl_lockref.count++;
2168 if (gl->gl_state != LM_ST_UNLOCKED)
2169 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
2170 __gfs2_glock_queue_work(gl, 0);
2172 spin_unlock(&gl->gl_lockref.lock);
2176 * gfs2_glock_thaw - Thaw any frozen glocks
2177 * @sdp: The super block
2181 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
2183 glock_hash_walk(thaw_glock, sdp);
2186 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2188 spin_lock(&gl->gl_lockref.lock);
2189 gfs2_dump_glock(seq, gl, fsid);
2190 spin_unlock(&gl->gl_lockref.lock);
2193 static void dump_glock_func(struct gfs2_glock *gl)
2195 dump_glock(NULL, gl, true);
2198 static void withdraw_dq(struct gfs2_glock *gl)
2200 spin_lock(&gl->gl_lockref.lock);
2201 if (!__lockref_is_dead(&gl->gl_lockref) &&
2202 glock_blocked_by_withdraw(gl))
2203 do_error(gl, LM_OUT_ERROR); /* remove pending waiters */
2204 spin_unlock(&gl->gl_lockref.lock);
2207 void gfs2_gl_dq_holders(struct gfs2_sbd *sdp)
2209 glock_hash_walk(withdraw_dq, sdp);
2213 * gfs2_gl_hash_clear - Empty out the glock hash table
2214 * @sdp: the filesystem
2216 * Called when unmounting the filesystem.
2219 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
2221 set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
2222 flush_workqueue(glock_workqueue);
2223 glock_hash_walk(clear_glock, sdp);
2224 flush_workqueue(glock_workqueue);
2225 wait_event_timeout(sdp->sd_kill_wait,
2226 atomic_read(&sdp->sd_glock_disposal) == 0,
2228 glock_hash_walk(dump_glock_func, sdp);
2231 static const char *state2str(unsigned state)
2234 case LM_ST_UNLOCKED:
2238 case LM_ST_DEFERRED:
2240 case LM_ST_EXCLUSIVE:
2246 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
2249 if (flags & LM_FLAG_TRY)
2251 if (flags & LM_FLAG_TRY_1CB)
2253 if (flags & LM_FLAG_NOEXP)
2255 if (flags & LM_FLAG_ANY)
2257 if (flags & LM_FLAG_NODE_SCOPE)
2259 if (flags & GL_ASYNC)
2261 if (flags & GL_EXACT)
2263 if (flags & GL_NOCACHE)
2265 if (test_bit(HIF_HOLDER, &iflags))
2267 if (test_bit(HIF_WAIT, &iflags))
2269 if (flags & GL_SKIP)
2276 * dump_holder - print information about a glock holder
2277 * @seq: the seq_file struct
2278 * @gh: the glock holder
2279 * @fs_id_buf: pointer to file system id (if requested)
2283 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh,
2284 const char *fs_id_buf)
2286 const char *comm = "(none)";
2287 pid_t owner_pid = 0;
2291 if (pid_is_meaningful(gh)) {
2292 struct task_struct *gh_owner;
2295 owner_pid = pid_nr(gh->gh_owner_pid);
2296 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
2298 comm = gh_owner->comm;
2300 gfs2_print_dbg(seq, "%s H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
2301 fs_id_buf, state2str(gh->gh_state),
2302 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
2303 gh->gh_error, (long)owner_pid, comm, (void *)gh->gh_ip);
2307 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
2309 const unsigned long *gflags = &gl->gl_flags;
2312 if (test_bit(GLF_LOCK, gflags))
2314 if (test_bit(GLF_DEMOTE, gflags))
2316 if (test_bit(GLF_PENDING_DEMOTE, gflags))
2318 if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
2320 if (test_bit(GLF_DIRTY, gflags))
2322 if (test_bit(GLF_LFLUSH, gflags))
2324 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
2326 if (test_bit(GLF_REPLY_PENDING, gflags))
2328 if (test_bit(GLF_INITIAL, gflags))
2330 if (test_bit(GLF_FROZEN, gflags))
2332 if (!list_empty(&gl->gl_holders))
2334 if (test_bit(GLF_LRU, gflags))
2338 if (test_bit(GLF_BLOCKING, gflags))
2340 if (test_bit(GLF_FREEING, gflags))
2342 if (test_bit(GLF_INSTANTIATE_NEEDED, gflags))
2344 if (test_bit(GLF_INSTANTIATE_IN_PROG, gflags))
2346 if (test_bit(GLF_TRY_TO_EVICT, gflags))
2348 if (test_bit(GLF_VERIFY_EVICT, gflags))
2355 * gfs2_dump_glock - print information about a glock
2356 * @seq: The seq_file struct
2358 * @fsid: If true, also dump the file system id
2360 * The file format is as follows:
2361 * One line per object, capital letters are used to indicate objects
2362 * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
2363 * other objects are indented by a single space and follow the glock to
2364 * which they are related. Fields are indicated by lower case letters
2365 * followed by a colon and the field value, except for strings which are in
2366 * [] so that its possible to see if they are composed of spaces for
2367 * example. The field's are n = number (id of the object), f = flags,
2368 * t = type, s = state, r = refcount, e = error, p = pid.
2372 void gfs2_dump_glock(struct seq_file *seq, struct gfs2_glock *gl, bool fsid)
2374 const struct gfs2_glock_operations *glops = gl->gl_ops;
2375 unsigned long long dtime;
2376 const struct gfs2_holder *gh;
2377 char gflags_buf[32];
2378 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
2379 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
2380 unsigned long nrpages = 0;
2382 if (gl->gl_ops->go_flags & GLOF_ASPACE) {
2383 struct address_space *mapping = gfs2_glock2aspace(gl);
2385 nrpages = mapping->nrpages;
2387 memset(fs_id_buf, 0, sizeof(fs_id_buf));
2388 if (fsid && sdp) /* safety precaution */
2389 sprintf(fs_id_buf, "fsid=%s: ", sdp->sd_fsname);
2390 dtime = jiffies - gl->gl_demote_time;
2391 dtime *= 1000000/HZ; /* demote time in uSec */
2392 if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
2394 gfs2_print_dbg(seq, "%sG: s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d "
2395 "v:%d r:%d m:%ld p:%lu\n",
2396 fs_id_buf, state2str(gl->gl_state),
2397 gl->gl_name.ln_type,
2398 (unsigned long long)gl->gl_name.ln_number,
2399 gflags2str(gflags_buf, gl),
2400 state2str(gl->gl_target),
2401 state2str(gl->gl_demote_state), dtime,
2402 atomic_read(&gl->gl_ail_count),
2403 atomic_read(&gl->gl_revokes),
2404 (int)gl->gl_lockref.count, gl->gl_hold_time, nrpages);
2406 list_for_each_entry(gh, &gl->gl_holders, gh_list)
2407 dump_holder(seq, gh, fs_id_buf);
2409 if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
2410 glops->go_dump(seq, gl, fs_id_buf);
2413 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
2415 struct gfs2_glock *gl = iter_ptr;
2417 seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
2418 gl->gl_name.ln_type,
2419 (unsigned long long)gl->gl_name.ln_number,
2420 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
2421 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
2422 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
2423 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
2424 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
2425 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
2426 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
2427 (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
2431 static const char *gfs2_gltype[] = {
2445 static const char *gfs2_stype[] = {
2446 [GFS2_LKS_SRTT] = "srtt",
2447 [GFS2_LKS_SRTTVAR] = "srttvar",
2448 [GFS2_LKS_SRTTB] = "srttb",
2449 [GFS2_LKS_SRTTVARB] = "srttvarb",
2450 [GFS2_LKS_SIRT] = "sirt",
2451 [GFS2_LKS_SIRTVAR] = "sirtvar",
2452 [GFS2_LKS_DCOUNT] = "dlm",
2453 [GFS2_LKS_QCOUNT] = "queue",
2456 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
2458 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
2460 struct gfs2_sbd *sdp = seq->private;
2461 loff_t pos = *(loff_t *)iter_ptr;
2462 unsigned index = pos >> 3;
2463 unsigned subindex = pos & 0x07;
2466 if (index == 0 && subindex != 0)
2469 seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
2470 (index == 0) ? "cpu": gfs2_stype[subindex]);
2472 for_each_possible_cpu(i) {
2473 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
2476 seq_printf(seq, " %15u", i);
2478 seq_printf(seq, " %15llu", (unsigned long long)lkstats->
2479 lkstats[index - 1].stats[subindex]);
2481 seq_putc(seq, '\n');
2485 int __init gfs2_glock_init(void)
2489 ret = rhashtable_init(&gl_hash_table, &ht_parms);
2493 glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
2494 WQ_HIGHPRI | WQ_FREEZABLE, 0);
2495 if (!glock_workqueue) {
2496 rhashtable_destroy(&gl_hash_table);
2500 glock_shrinker = shrinker_alloc(0, "gfs2-glock");
2501 if (!glock_shrinker) {
2502 destroy_workqueue(glock_workqueue);
2503 rhashtable_destroy(&gl_hash_table);
2507 glock_shrinker->count_objects = gfs2_glock_shrink_count;
2508 glock_shrinker->scan_objects = gfs2_glock_shrink_scan;
2510 shrinker_register(glock_shrinker);
2512 for (i = 0; i < GLOCK_WAIT_TABLE_SIZE; i++)
2513 init_waitqueue_head(glock_wait_table + i);
2518 void gfs2_glock_exit(void)
2520 shrinker_free(glock_shrinker);
2521 rhashtable_destroy(&gl_hash_table);
2522 destroy_workqueue(glock_workqueue);
2525 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi, loff_t n)
2527 struct gfs2_glock *gl = gi->gl;
2532 if (!lockref_put_not_zero(&gl->gl_lockref))
2533 gfs2_glock_queue_put(gl);
2536 gl = rhashtable_walk_next(&gi->hti);
2537 if (IS_ERR_OR_NULL(gl)) {
2538 if (gl == ERR_PTR(-EAGAIN)) {
2545 if (gl->gl_name.ln_sbd != gi->sdp)
2548 if (!lockref_get_not_dead(&gl->gl_lockref))
2552 if (__lockref_is_dead(&gl->gl_lockref))
2560 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
2563 struct gfs2_glock_iter *gi = seq->private;
2567 * We can either stay where we are, skip to the next hash table
2568 * entry, or start from the beginning.
2570 if (*pos < gi->last_pos) {
2571 rhashtable_walk_exit(&gi->hti);
2572 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2575 n = *pos - gi->last_pos;
2578 rhashtable_walk_start(&gi->hti);
2580 gfs2_glock_iter_next(gi, n);
2581 gi->last_pos = *pos;
2585 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
2588 struct gfs2_glock_iter *gi = seq->private;
2591 gi->last_pos = *pos;
2592 gfs2_glock_iter_next(gi, 1);
2596 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
2599 struct gfs2_glock_iter *gi = seq->private;
2601 rhashtable_walk_stop(&gi->hti);
2604 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
2606 dump_glock(seq, iter_ptr, false);
2610 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
2613 if (*pos >= GFS2_NR_SBSTATS)
2618 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
2622 if (*pos >= GFS2_NR_SBSTATS)
2627 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
2632 static const struct seq_operations gfs2_glock_seq_ops = {
2633 .start = gfs2_glock_seq_start,
2634 .next = gfs2_glock_seq_next,
2635 .stop = gfs2_glock_seq_stop,
2636 .show = gfs2_glock_seq_show,
2639 static const struct seq_operations gfs2_glstats_seq_ops = {
2640 .start = gfs2_glock_seq_start,
2641 .next = gfs2_glock_seq_next,
2642 .stop = gfs2_glock_seq_stop,
2643 .show = gfs2_glstats_seq_show,
2646 static const struct seq_operations gfs2_sbstats_sops = {
2647 .start = gfs2_sbstats_seq_start,
2648 .next = gfs2_sbstats_seq_next,
2649 .stop = gfs2_sbstats_seq_stop,
2650 .show = gfs2_sbstats_seq_show,
2653 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
2655 static int __gfs2_glocks_open(struct inode *inode, struct file *file,
2656 const struct seq_operations *ops)
2658 int ret = seq_open_private(file, ops, sizeof(struct gfs2_glock_iter));
2660 struct seq_file *seq = file->private_data;
2661 struct gfs2_glock_iter *gi = seq->private;
2663 gi->sdp = inode->i_private;
2664 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
2666 seq->size = GFS2_SEQ_GOODSIZE;
2668 * Initially, we are "before" the first hash table entry; the
2669 * first call to rhashtable_walk_next gets us the first entry.
2673 rhashtable_walk_enter(&gl_hash_table, &gi->hti);
2678 static int gfs2_glocks_open(struct inode *inode, struct file *file)
2680 return __gfs2_glocks_open(inode, file, &gfs2_glock_seq_ops);
2683 static int gfs2_glocks_release(struct inode *inode, struct file *file)
2685 struct seq_file *seq = file->private_data;
2686 struct gfs2_glock_iter *gi = seq->private;
2689 gfs2_glock_put(gi->gl);
2690 rhashtable_walk_exit(&gi->hti);
2691 return seq_release_private(inode, file);
2694 static int gfs2_glstats_open(struct inode *inode, struct file *file)
2696 return __gfs2_glocks_open(inode, file, &gfs2_glstats_seq_ops);
2699 static const struct file_operations gfs2_glocks_fops = {
2700 .owner = THIS_MODULE,
2701 .open = gfs2_glocks_open,
2703 .llseek = seq_lseek,
2704 .release = gfs2_glocks_release,
2707 static const struct file_operations gfs2_glstats_fops = {
2708 .owner = THIS_MODULE,
2709 .open = gfs2_glstats_open,
2711 .llseek = seq_lseek,
2712 .release = gfs2_glocks_release,
2715 struct gfs2_glockfd_iter {
2716 struct super_block *sb;
2718 struct task_struct *task;
2723 static struct task_struct *gfs2_glockfd_next_task(struct gfs2_glockfd_iter *i)
2725 struct pid_namespace *ns = task_active_pid_ns(current);
2729 put_task_struct(i->task);
2734 pid = find_ge_pid(i->tgid, ns);
2736 i->tgid = pid_nr_ns(pid, ns);
2737 i->task = pid_task(pid, PIDTYPE_TGID);
2742 get_task_struct(i->task);
2748 static struct file *gfs2_glockfd_next_file(struct gfs2_glockfd_iter *i)
2757 struct inode *inode;
2759 i->file = task_lookup_next_fdget_rcu(i->task, &i->fd);
2765 inode = file_inode(i->file);
2766 if (inode->i_sb == i->sb)
2777 static void *gfs2_glockfd_seq_start(struct seq_file *seq, loff_t *pos)
2779 struct gfs2_glockfd_iter *i = seq->private;
2783 while (gfs2_glockfd_next_task(i)) {
2784 if (gfs2_glockfd_next_file(i))
2791 static void *gfs2_glockfd_seq_next(struct seq_file *seq, void *iter_ptr,
2794 struct gfs2_glockfd_iter *i = seq->private;
2799 if (gfs2_glockfd_next_file(i))
2802 } while (gfs2_glockfd_next_task(i));
2806 static void gfs2_glockfd_seq_stop(struct seq_file *seq, void *iter_ptr)
2808 struct gfs2_glockfd_iter *i = seq->private;
2813 put_task_struct(i->task);
2816 static void gfs2_glockfd_seq_show_flock(struct seq_file *seq,
2817 struct gfs2_glockfd_iter *i)
2819 struct gfs2_file *fp = i->file->private_data;
2820 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
2821 struct lm_lockname gl_name = { .ln_type = LM_TYPE_RESERVED };
2823 if (!READ_ONCE(fl_gh->gh_gl))
2826 spin_lock(&i->file->f_lock);
2827 if (gfs2_holder_initialized(fl_gh))
2828 gl_name = fl_gh->gh_gl->gl_name;
2829 spin_unlock(&i->file->f_lock);
2831 if (gl_name.ln_type != LM_TYPE_RESERVED) {
2832 seq_printf(seq, "%d %u %u/%llx\n",
2833 i->tgid, i->fd, gl_name.ln_type,
2834 (unsigned long long)gl_name.ln_number);
2838 static int gfs2_glockfd_seq_show(struct seq_file *seq, void *iter_ptr)
2840 struct gfs2_glockfd_iter *i = seq->private;
2841 struct inode *inode = file_inode(i->file);
2842 struct gfs2_glock *gl;
2844 inode_lock_shared(inode);
2845 gl = GFS2_I(inode)->i_iopen_gh.gh_gl;
2847 seq_printf(seq, "%d %u %u/%llx\n",
2848 i->tgid, i->fd, gl->gl_name.ln_type,
2849 (unsigned long long)gl->gl_name.ln_number);
2851 gfs2_glockfd_seq_show_flock(seq, i);
2852 inode_unlock_shared(inode);
2856 static const struct seq_operations gfs2_glockfd_seq_ops = {
2857 .start = gfs2_glockfd_seq_start,
2858 .next = gfs2_glockfd_seq_next,
2859 .stop = gfs2_glockfd_seq_stop,
2860 .show = gfs2_glockfd_seq_show,
2863 static int gfs2_glockfd_open(struct inode *inode, struct file *file)
2865 struct gfs2_glockfd_iter *i;
2866 struct gfs2_sbd *sdp = inode->i_private;
2868 i = __seq_open_private(file, &gfs2_glockfd_seq_ops,
2869 sizeof(struct gfs2_glockfd_iter));
2872 i->sb = sdp->sd_vfs;
2876 static const struct file_operations gfs2_glockfd_fops = {
2877 .owner = THIS_MODULE,
2878 .open = gfs2_glockfd_open,
2880 .llseek = seq_lseek,
2881 .release = seq_release_private,
2884 DEFINE_SEQ_ATTRIBUTE(gfs2_sbstats);
2886 void gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2888 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2890 debugfs_create_file("glocks", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2893 debugfs_create_file("glockfd", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2894 &gfs2_glockfd_fops);
2896 debugfs_create_file("glstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2897 &gfs2_glstats_fops);
2899 debugfs_create_file("sbstats", S_IFREG | S_IRUGO, sdp->debugfs_dir, sdp,
2900 &gfs2_sbstats_fops);
2903 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2905 debugfs_remove_recursive(sdp->debugfs_dir);
2906 sdp->debugfs_dir = NULL;
2909 void gfs2_register_debugfs(void)
2911 gfs2_root = debugfs_create_dir("gfs2", NULL);
2914 void gfs2_unregister_debugfs(void)
2916 debugfs_remove(gfs2_root);