2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/delay.h>
16 #include <linux/sort.h>
17 #include <linux/jhash.h>
18 #include <linux/kallsyms.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/list.h>
21 #include <linux/lm_interface.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <linux/rwsem.h>
25 #include <asm/uaccess.h>
26 #include <linux/seq_file.h>
27 #include <linux/debugfs.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
30 #include <linux/workqueue.h>
31 #include <linux/jiffies.h>
45 struct gfs2_gl_hash_bucket {
46 struct hlist_head hb_list;
50 int hash; /* hash bucket index */
51 struct gfs2_sbd *sdp; /* incore superblock */
52 struct gfs2_glock *gl; /* current glock struct */
53 struct seq_file *seq; /* sequence file for debugfs */
54 char string[512]; /* scratch space */
57 typedef void (*glock_examiner) (struct gfs2_glock * gl);
59 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
60 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl);
61 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
62 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
63 static void run_queue(struct gfs2_glock *gl);
65 static DECLARE_RWSEM(gfs2_umount_flush_sem);
66 static struct dentry *gfs2_root;
67 static struct task_struct *scand_process;
68 static unsigned int scand_secs = 5;
69 static struct workqueue_struct *glock_workqueue;
71 #define GFS2_GL_HASH_SHIFT 15
72 #define GFS2_GL_HASH_SIZE (1 << GFS2_GL_HASH_SHIFT)
73 #define GFS2_GL_HASH_MASK (GFS2_GL_HASH_SIZE - 1)
75 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
76 static struct dentry *gfs2_root;
79 * Despite what you might think, the numbers below are not arbitrary :-)
80 * They are taken from the ipv4 routing hash code, which is well tested
81 * and thus should be nearly optimal. Later on we might tweek the numbers
82 * but for now this should be fine.
84 * The reason for putting the locks in a separate array from the list heads
85 * is that we can have fewer locks than list heads and save memory. We use
86 * the same hash function for both, but with a different hash mask.
88 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
89 defined(CONFIG_PROVE_LOCKING)
92 # define GL_HASH_LOCK_SZ 256
95 # define GL_HASH_LOCK_SZ 4096
97 # define GL_HASH_LOCK_SZ 2048
99 # define GL_HASH_LOCK_SZ 1024
101 # define GL_HASH_LOCK_SZ 512
103 # define GL_HASH_LOCK_SZ 256
107 /* We never want more locks than chains */
108 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
109 # undef GL_HASH_LOCK_SZ
110 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
113 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
115 static inline rwlock_t *gl_lock_addr(unsigned int x)
117 return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
119 #else /* not SMP, so no spinlocks required */
120 static inline rwlock_t *gl_lock_addr(unsigned int x)
127 * relaxed_state_ok - is a requested lock compatible with the current lock mode?
128 * @actual: the current state of the lock
129 * @requested: the lock state that was requested by the caller
130 * @flags: the modifier flags passed in by the caller
132 * Returns: 1 if the locks are compatible, 0 otherwise
135 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
138 if (actual == requested)
141 if (flags & GL_EXACT)
144 if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
147 if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
154 * gl_hash() - Turn glock number into hash bucket number
155 * @lock: The glock number
157 * Returns: The number of the corresponding hash bucket
160 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
161 const struct lm_lockname *name)
165 h = jhash(&name->ln_number, sizeof(u64), 0);
166 h = jhash(&name->ln_type, sizeof(unsigned int), h);
167 h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
168 h &= GFS2_GL_HASH_MASK;
174 * glock_free() - Perform a few checks and then release struct gfs2_glock
175 * @gl: The glock to release
177 * Also calls lock module to release its internal structure for this glock.
181 static void glock_free(struct gfs2_glock *gl)
183 struct gfs2_sbd *sdp = gl->gl_sbd;
184 struct inode *aspace = gl->gl_aspace;
186 gfs2_lm_put_lock(sdp, gl->gl_lock);
189 gfs2_aspace_put(aspace);
191 kmem_cache_free(gfs2_glock_cachep, gl);
195 * gfs2_glock_hold() - increment reference count on glock
196 * @gl: The glock to hold
200 void gfs2_glock_hold(struct gfs2_glock *gl)
202 atomic_inc(&gl->gl_ref);
206 * gfs2_glock_put() - Decrement reference count on glock
207 * @gl: The glock to put
211 int gfs2_glock_put(struct gfs2_glock *gl)
214 struct gfs2_sbd *sdp = gl->gl_sbd;
216 write_lock(gl_lock_addr(gl->gl_hash));
217 if (atomic_dec_and_test(&gl->gl_ref)) {
218 hlist_del(&gl->gl_list);
219 write_unlock(gl_lock_addr(gl->gl_hash));
220 BUG_ON(spin_is_locked(&gl->gl_spin));
221 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
222 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
223 gfs2_assert(sdp, list_empty(&gl->gl_holders));
224 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
225 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
230 write_unlock(gl_lock_addr(gl->gl_hash));
236 * search_bucket() - Find struct gfs2_glock by lock number
237 * @bucket: the bucket to search
238 * @name: The lock name
240 * Returns: NULL, or the struct gfs2_glock with the requested number
243 static struct gfs2_glock *search_bucket(unsigned int hash,
244 const struct gfs2_sbd *sdp,
245 const struct lm_lockname *name)
247 struct gfs2_glock *gl;
248 struct hlist_node *h;
250 hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
251 if (!lm_name_equal(&gl->gl_name, name))
253 if (gl->gl_sbd != sdp)
256 atomic_inc(&gl->gl_ref);
265 * gfs2_glock_find() - Find glock by lock number
266 * @sdp: The GFS2 superblock
267 * @name: The lock name
269 * Returns: NULL, or the struct gfs2_glock with the requested number
272 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
273 const struct lm_lockname *name)
275 unsigned int hash = gl_hash(sdp, name);
276 struct gfs2_glock *gl;
278 read_lock(gl_lock_addr(hash));
279 gl = search_bucket(hash, sdp, name);
280 read_unlock(gl_lock_addr(hash));
285 static void glock_work_func(struct work_struct *work)
287 struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
289 spin_lock(&gl->gl_spin);
290 if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags))
291 set_bit(GLF_DEMOTE, &gl->gl_flags);
293 spin_unlock(&gl->gl_spin);
298 * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
299 * @sdp: The GFS2 superblock
300 * @number: the lock number
301 * @glops: The glock_operations to use
302 * @create: If 0, don't create the glock if it doesn't exist
303 * @glp: the glock is returned here
305 * This does not lock a glock, just finds/creates structures for one.
310 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
311 const struct gfs2_glock_operations *glops, int create,
312 struct gfs2_glock **glp)
314 struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
315 struct gfs2_glock *gl, *tmp;
316 unsigned int hash = gl_hash(sdp, &name);
319 read_lock(gl_lock_addr(hash));
320 gl = search_bucket(hash, sdp, &name);
321 read_unlock(gl_lock_addr(hash));
328 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
334 atomic_set(&gl->gl_ref, 1);
335 gl->gl_state = LM_ST_UNLOCKED;
336 gl->gl_demote_state = LM_ST_EXCLUSIVE;
338 gl->gl_owner_pid = 0;
341 gl->gl_req_gh = NULL;
342 gl->gl_req_bh = NULL;
344 gl->gl_stamp = jiffies;
345 gl->gl_tchange = jiffies;
346 gl->gl_object = NULL;
348 gl->gl_aspace = NULL;
349 lops_init_le(&gl->gl_le, &gfs2_glock_lops);
350 INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
352 /* If this glock protects actual on-disk data or metadata blocks,
353 create a VFS inode to manage the pages/buffers holding them. */
354 if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
355 gl->gl_aspace = gfs2_aspace_get(sdp);
356 if (!gl->gl_aspace) {
362 error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
366 write_lock(gl_lock_addr(hash));
367 tmp = search_bucket(hash, sdp, &name);
369 write_unlock(gl_lock_addr(hash));
373 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
374 write_unlock(gl_lock_addr(hash));
383 gfs2_aspace_put(gl->gl_aspace);
385 kmem_cache_free(gfs2_glock_cachep, gl);
390 * gfs2_holder_init - initialize a struct gfs2_holder in the default way
392 * @state: the state we're requesting
393 * @flags: the modifier flags
394 * @gh: the holder structure
398 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
399 struct gfs2_holder *gh)
401 INIT_LIST_HEAD(&gh->gh_list);
403 gh->gh_ip = (unsigned long)__builtin_return_address(0);
404 gh->gh_owner_pid = current->pid;
405 gh->gh_state = state;
406 gh->gh_flags = flags;
413 * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
414 * @state: the state we're requesting
415 * @flags: the modifier flags
416 * @gh: the holder structure
418 * Don't mess with the glock.
422 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
424 gh->gh_state = state;
425 gh->gh_flags = flags;
427 gh->gh_ip = (unsigned long)__builtin_return_address(0);
431 * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
432 * @gh: the holder structure
436 void gfs2_holder_uninit(struct gfs2_holder *gh)
438 gfs2_glock_put(gh->gh_gl);
443 static void gfs2_holder_wake(struct gfs2_holder *gh)
445 clear_bit(HIF_WAIT, &gh->gh_iflags);
446 smp_mb__after_clear_bit();
447 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
450 static int just_schedule(void *word)
456 static void wait_on_holder(struct gfs2_holder *gh)
459 wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
462 static void gfs2_demote_wake(struct gfs2_glock *gl)
464 BUG_ON(!spin_is_locked(&gl->gl_spin));
465 gl->gl_demote_state = LM_ST_EXCLUSIVE;
466 clear_bit(GLF_DEMOTE, &gl->gl_flags);
467 smp_mb__after_clear_bit();
468 wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
471 static void wait_on_demote(struct gfs2_glock *gl)
474 wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
478 * rq_mutex - process a mutex request in the queue
479 * @gh: the glock holder
481 * Returns: 1 if the queue is blocked
484 static int rq_mutex(struct gfs2_holder *gh)
486 struct gfs2_glock *gl = gh->gh_gl;
488 list_del_init(&gh->gh_list);
489 /* gh->gh_error never examined. */
490 set_bit(GLF_LOCK, &gl->gl_flags);
491 clear_bit(HIF_WAIT, &gh->gh_iflags);
493 wake_up_bit(&gh->gh_iflags, HIF_WAIT);
499 * rq_promote - process a promote request in the queue
500 * @gh: the glock holder
502 * Acquire a new inter-node lock, or change a lock state to more restrictive.
504 * Returns: 1 if the queue is blocked
507 static int rq_promote(struct gfs2_holder *gh)
509 struct gfs2_glock *gl = gh->gh_gl;
511 if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
512 if (list_empty(&gl->gl_holders)) {
514 set_bit(GLF_LOCK, &gl->gl_flags);
515 spin_unlock(&gl->gl_spin);
516 gfs2_glock_xmote_th(gh->gh_gl, gh);
517 spin_lock(&gl->gl_spin);
522 if (list_empty(&gl->gl_holders)) {
523 set_bit(HIF_FIRST, &gh->gh_iflags);
524 set_bit(GLF_LOCK, &gl->gl_flags);
526 struct gfs2_holder *next_gh;
527 if (gh->gh_state == LM_ST_EXCLUSIVE)
529 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
531 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
535 list_move_tail(&gh->gh_list, &gl->gl_holders);
537 set_bit(HIF_HOLDER, &gh->gh_iflags);
539 gfs2_holder_wake(gh);
545 * rq_demote - process a demote request in the queue
546 * @gh: the glock holder
548 * Returns: 1 if the queue is blocked
551 static int rq_demote(struct gfs2_glock *gl)
553 if (!list_empty(&gl->gl_holders))
556 if (gl->gl_state == gl->gl_demote_state ||
557 gl->gl_state == LM_ST_UNLOCKED) {
558 gfs2_demote_wake(gl);
562 set_bit(GLF_LOCK, &gl->gl_flags);
563 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
565 if (gl->gl_demote_state == LM_ST_UNLOCKED ||
566 gl->gl_state != LM_ST_EXCLUSIVE) {
567 spin_unlock(&gl->gl_spin);
568 gfs2_glock_drop_th(gl);
570 spin_unlock(&gl->gl_spin);
571 gfs2_glock_xmote_th(gl, NULL);
574 spin_lock(&gl->gl_spin);
575 clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
581 * run_queue - process holder structures on a glock
585 static void run_queue(struct gfs2_glock *gl)
587 struct gfs2_holder *gh;
591 if (test_bit(GLF_LOCK, &gl->gl_flags))
594 if (!list_empty(&gl->gl_waiters1)) {
595 gh = list_entry(gl->gl_waiters1.next,
596 struct gfs2_holder, gh_list);
597 blocked = rq_mutex(gh);
598 } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
599 blocked = rq_demote(gl);
600 if (gl->gl_waiters2 && !blocked) {
601 set_bit(GLF_DEMOTE, &gl->gl_flags);
602 gl->gl_demote_state = LM_ST_UNLOCKED;
605 } else if (!list_empty(&gl->gl_waiters3)) {
606 gh = list_entry(gl->gl_waiters3.next,
607 struct gfs2_holder, gh_list);
608 blocked = rq_promote(gh);
618 * gfs2_glmutex_lock - acquire a local lock on a glock
621 * Gives caller exclusive access to manipulate a glock structure.
624 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
626 struct gfs2_holder gh;
628 gfs2_holder_init(gl, 0, 0, &gh);
629 if (test_and_set_bit(HIF_WAIT, &gh.gh_iflags))
632 spin_lock(&gl->gl_spin);
633 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
634 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
636 gl->gl_owner_pid = current->pid;
637 gl->gl_ip = (unsigned long)__builtin_return_address(0);
638 clear_bit(HIF_WAIT, &gh.gh_iflags);
640 wake_up_bit(&gh.gh_iflags, HIF_WAIT);
642 spin_unlock(&gl->gl_spin);
645 gfs2_holder_uninit(&gh);
649 * gfs2_glmutex_trylock - try to acquire a local lock on a glock
652 * Returns: 1 if the glock is acquired
655 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
659 spin_lock(&gl->gl_spin);
660 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
663 gl->gl_owner_pid = current->pid;
664 gl->gl_ip = (unsigned long)__builtin_return_address(0);
666 spin_unlock(&gl->gl_spin);
672 * gfs2_glmutex_unlock - release a local lock on a glock
677 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
679 spin_lock(&gl->gl_spin);
680 clear_bit(GLF_LOCK, &gl->gl_flags);
681 gl->gl_owner_pid = 0;
684 BUG_ON(!spin_is_locked(&gl->gl_spin));
685 spin_unlock(&gl->gl_spin);
689 * handle_callback - process a demote request
691 * @state: the state the caller wants us to change to
693 * There are only two requests that we are going to see in actual
694 * practise: LM_ST_SHARED and LM_ST_UNLOCKED
697 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
698 int remote, unsigned long delay)
700 int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
702 spin_lock(&gl->gl_spin);
703 set_bit(bit, &gl->gl_flags);
704 if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
705 gl->gl_demote_state = state;
706 gl->gl_demote_time = jiffies;
707 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
709 gfs2_glock_schedule_for_reclaim(gl);
710 spin_unlock(&gl->gl_spin);
713 } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
714 gl->gl_demote_state != state) {
715 if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
718 gl->gl_demote_state = LM_ST_UNLOCKED;
720 spin_unlock(&gl->gl_spin);
724 * state_change - record that the glock is now in a different state
726 * @new_state the new state
730 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
734 held1 = (gl->gl_state != LM_ST_UNLOCKED);
735 held2 = (new_state != LM_ST_UNLOCKED);
737 if (held1 != held2) {
744 gl->gl_state = new_state;
745 gl->gl_tchange = jiffies;
749 * xmote_bh - Called after the lock module is done acquiring a lock
750 * @gl: The glock in question
751 * @ret: the int returned from the lock module
755 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
757 struct gfs2_sbd *sdp = gl->gl_sbd;
758 const struct gfs2_glock_operations *glops = gl->gl_ops;
759 struct gfs2_holder *gh = gl->gl_req_gh;
760 int prev_state = gl->gl_state;
763 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
764 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
765 gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
767 state_change(gl, ret & LM_OUT_ST_MASK);
769 if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
771 glops->go_inval(gl, DIO_METADATA);
772 } else if (gl->gl_state == LM_ST_DEFERRED) {
773 /* We might not want to do this here.
774 Look at moving to the inode glops. */
776 glops->go_inval(gl, 0);
779 /* Deal with each possible exit condition */
782 gl->gl_stamp = jiffies;
783 if (ret & LM_OUT_CANCELED) {
786 spin_lock(&gl->gl_spin);
787 if (gl->gl_state != gl->gl_demote_state) {
788 gl->gl_req_bh = NULL;
789 spin_unlock(&gl->gl_spin);
790 gfs2_glock_drop_th(gl);
794 gfs2_demote_wake(gl);
795 spin_unlock(&gl->gl_spin);
798 spin_lock(&gl->gl_spin);
799 list_del_init(&gh->gh_list);
801 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
803 gh->gh_error = GLR_CANCELED;
804 if (ret & LM_OUT_CANCELED)
806 if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
807 list_add_tail(&gh->gh_list, &gl->gl_holders);
809 set_bit(HIF_HOLDER, &gh->gh_iflags);
810 set_bit(HIF_FIRST, &gh->gh_iflags);
814 gh->gh_error = GLR_TRYFAILED;
815 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
817 gh->gh_error = -EINVAL;
818 if (gfs2_assert_withdraw(sdp, 0) == -1)
819 fs_err(sdp, "ret = 0x%.8X\n", ret);
821 spin_unlock(&gl->gl_spin);
824 if (glops->go_xmote_bh)
825 glops->go_xmote_bh(gl);
828 spin_lock(&gl->gl_spin);
829 gl->gl_req_gh = NULL;
830 gl->gl_req_bh = NULL;
831 clear_bit(GLF_LOCK, &gl->gl_flags);
832 spin_unlock(&gl->gl_spin);
838 gfs2_holder_wake(gh);
842 * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
843 * @gl: The glock in question
844 * @state: the requested state
845 * @flags: modifier flags to the lock call
849 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
851 struct gfs2_sbd *sdp = gl->gl_sbd;
852 int flags = gh ? gh->gh_flags : 0;
853 unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
854 const struct gfs2_glock_operations *glops = gl->gl_ops;
855 int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
856 LM_FLAG_NOEXP | LM_FLAG_ANY |
858 unsigned int lck_ret;
860 if (glops->go_xmote_th)
861 glops->go_xmote_th(gl);
863 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
864 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
865 gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
866 gfs2_assert_warn(sdp, state != gl->gl_state);
869 gl->gl_req_bh = xmote_bh;
871 lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
873 if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
876 if (lck_ret & LM_OUT_ASYNC)
877 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
879 xmote_bh(gl, lck_ret);
883 * drop_bh - Called after a lock module unlock completes
885 * @ret: the return status
887 * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
888 * Doesn't drop the reference on the glock the top half took out
892 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
894 struct gfs2_sbd *sdp = gl->gl_sbd;
895 const struct gfs2_glock_operations *glops = gl->gl_ops;
896 struct gfs2_holder *gh = gl->gl_req_gh;
898 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
899 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
900 gfs2_assert_warn(sdp, !ret);
902 state_change(gl, LM_ST_UNLOCKED);
905 glops->go_inval(gl, DIO_METADATA);
908 spin_lock(&gl->gl_spin);
909 list_del_init(&gh->gh_list);
911 spin_unlock(&gl->gl_spin);
914 spin_lock(&gl->gl_spin);
915 gfs2_demote_wake(gl);
916 gl->gl_req_gh = NULL;
917 gl->gl_req_bh = NULL;
918 clear_bit(GLF_LOCK, &gl->gl_flags);
919 spin_unlock(&gl->gl_spin);
924 gfs2_holder_wake(gh);
928 * gfs2_glock_drop_th - call into the lock module to unlock a lock
933 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
935 struct gfs2_sbd *sdp = gl->gl_sbd;
936 const struct gfs2_glock_operations *glops = gl->gl_ops;
939 if (glops->go_xmote_th)
940 glops->go_xmote_th(gl);
942 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
943 gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
944 gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
947 gl->gl_req_bh = drop_bh;
949 ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
951 if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
957 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
961 * do_cancels - cancel requests for locks stuck waiting on an expire flag
962 * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
964 * Don't cancel GL_NOCANCEL requests.
967 static void do_cancels(struct gfs2_holder *gh)
969 struct gfs2_glock *gl = gh->gh_gl;
971 spin_lock(&gl->gl_spin);
973 while (gl->gl_req_gh != gh &&
974 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
975 !list_empty(&gh->gh_list)) {
976 if (gl->gl_req_bh && !(gl->gl_req_gh &&
977 (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
978 spin_unlock(&gl->gl_spin);
979 gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
981 spin_lock(&gl->gl_spin);
983 spin_unlock(&gl->gl_spin);
985 spin_lock(&gl->gl_spin);
989 spin_unlock(&gl->gl_spin);
993 * glock_wait_internal - wait on a glock acquisition
994 * @gh: the glock holder
996 * Returns: 0 on success
999 static int glock_wait_internal(struct gfs2_holder *gh)
1001 struct gfs2_glock *gl = gh->gh_gl;
1002 struct gfs2_sbd *sdp = gl->gl_sbd;
1003 const struct gfs2_glock_operations *glops = gl->gl_ops;
1005 if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1008 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1009 spin_lock(&gl->gl_spin);
1010 if (gl->gl_req_gh != gh &&
1011 !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1012 !list_empty(&gh->gh_list)) {
1013 list_del_init(&gh->gh_list);
1014 gh->gh_error = GLR_TRYFAILED;
1016 spin_unlock(&gl->gl_spin);
1017 return gh->gh_error;
1019 spin_unlock(&gl->gl_spin);
1022 if (gh->gh_flags & LM_FLAG_PRIORITY)
1027 return gh->gh_error;
1029 gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1030 gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1033 if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1034 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1036 if (glops->go_lock) {
1037 gh->gh_error = glops->go_lock(gh);
1039 spin_lock(&gl->gl_spin);
1040 list_del_init(&gh->gh_list);
1041 spin_unlock(&gl->gl_spin);
1045 spin_lock(&gl->gl_spin);
1046 gl->gl_req_gh = NULL;
1047 gl->gl_req_bh = NULL;
1048 clear_bit(GLF_LOCK, &gl->gl_flags);
1050 spin_unlock(&gl->gl_spin);
1053 return gh->gh_error;
1056 static inline struct gfs2_holder *
1057 find_holder_by_owner(struct list_head *head, pid_t pid)
1059 struct gfs2_holder *gh;
1061 list_for_each_entry(gh, head, gh_list) {
1062 if (gh->gh_owner_pid == pid)
1069 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1073 va_start(args, fmt);
1075 vsprintf(gi->string, fmt, args);
1076 seq_printf(gi->seq, gi->string);
1084 * add_to_queue - Add a holder to the wait queue (but look for recursion)
1085 * @gh: the holder structure to add
1089 static void add_to_queue(struct gfs2_holder *gh)
1091 struct gfs2_glock *gl = gh->gh_gl;
1092 struct gfs2_holder *existing;
1094 BUG_ON(!gh->gh_owner_pid);
1095 if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1098 if (!(gh->gh_flags & GL_FLOCK)) {
1099 existing = find_holder_by_owner(&gl->gl_holders,
1102 print_symbol(KERN_WARNING "original: %s\n",
1104 printk(KERN_INFO "pid : %d\n", existing->gh_owner_pid);
1105 printk(KERN_INFO "lock type : %d lock state : %d\n",
1106 existing->gh_gl->gl_name.ln_type,
1107 existing->gh_gl->gl_state);
1108 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1109 printk(KERN_INFO "pid : %d\n", gh->gh_owner_pid);
1110 printk(KERN_INFO "lock type : %d lock state : %d\n",
1111 gl->gl_name.ln_type, gl->gl_state);
1115 existing = find_holder_by_owner(&gl->gl_waiters3,
1118 print_symbol(KERN_WARNING "original: %s\n",
1120 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1125 if (gh->gh_flags & LM_FLAG_PRIORITY)
1126 list_add(&gh->gh_list, &gl->gl_waiters3);
1128 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1132 * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1133 * @gh: the holder structure
1135 * if (gh->gh_flags & GL_ASYNC), this never returns an error
1137 * Returns: 0, GLR_TRYFAILED, or errno on failure
1140 int gfs2_glock_nq(struct gfs2_holder *gh)
1142 struct gfs2_glock *gl = gh->gh_gl;
1143 struct gfs2_sbd *sdp = gl->gl_sbd;
1147 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1148 set_bit(HIF_ABORTED, &gh->gh_iflags);
1152 spin_lock(&gl->gl_spin);
1155 spin_unlock(&gl->gl_spin);
1157 if (!(gh->gh_flags & GL_ASYNC)) {
1158 error = glock_wait_internal(gh);
1159 if (error == GLR_CANCELED) {
1169 * gfs2_glock_poll - poll to see if an async request has been completed
1172 * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1175 int gfs2_glock_poll(struct gfs2_holder *gh)
1177 struct gfs2_glock *gl = gh->gh_gl;
1180 spin_lock(&gl->gl_spin);
1182 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1184 else if (list_empty(&gh->gh_list)) {
1185 if (gh->gh_error == GLR_CANCELED) {
1186 spin_unlock(&gl->gl_spin);
1188 if (gfs2_glock_nq(gh))
1195 spin_unlock(&gl->gl_spin);
1201 * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1202 * @gh: the holder structure
1204 * Returns: 0, GLR_TRYFAILED, or errno on failure
1207 int gfs2_glock_wait(struct gfs2_holder *gh)
1211 error = glock_wait_internal(gh);
1212 if (error == GLR_CANCELED) {
1214 gh->gh_flags &= ~GL_ASYNC;
1215 error = gfs2_glock_nq(gh);
1222 * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1223 * @gh: the glock holder
1227 void gfs2_glock_dq(struct gfs2_holder *gh)
1229 struct gfs2_glock *gl = gh->gh_gl;
1230 const struct gfs2_glock_operations *glops = gl->gl_ops;
1233 if (gh->gh_flags & GL_NOCACHE)
1234 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1236 gfs2_glmutex_lock(gl);
1238 spin_lock(&gl->gl_spin);
1239 list_del_init(&gh->gh_list);
1241 if (list_empty(&gl->gl_holders)) {
1242 if (glops->go_unlock) {
1243 spin_unlock(&gl->gl_spin);
1244 glops->go_unlock(gh);
1245 spin_lock(&gl->gl_spin);
1247 gl->gl_stamp = jiffies;
1250 clear_bit(GLF_LOCK, &gl->gl_flags);
1251 spin_unlock(&gl->gl_spin);
1253 gfs2_glock_hold(gl);
1254 if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1255 !test_bit(GLF_DEMOTE, &gl->gl_flags))
1256 delay = gl->gl_ops->go_min_hold_time;
1257 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1261 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1263 struct gfs2_glock *gl = gh->gh_gl;
1269 * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1270 * @gh: the holder structure
1274 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1277 gfs2_holder_uninit(gh);
1281 * gfs2_glock_nq_num - acquire a glock based on lock number
1282 * @sdp: the filesystem
1283 * @number: the lock number
1284 * @glops: the glock operations for the type of glock
1285 * @state: the state to acquire the glock in
1286 * @flags: modifier flags for the aquisition
1287 * @gh: the struct gfs2_holder
1292 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1293 const struct gfs2_glock_operations *glops,
1294 unsigned int state, int flags, struct gfs2_holder *gh)
1296 struct gfs2_glock *gl;
1299 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1301 error = gfs2_glock_nq_init(gl, state, flags, gh);
1309 * glock_compare - Compare two struct gfs2_glock structures for sorting
1310 * @arg_a: the first structure
1311 * @arg_b: the second structure
1315 static int glock_compare(const void *arg_a, const void *arg_b)
1317 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1318 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1319 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1320 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1322 if (a->ln_number > b->ln_number)
1324 if (a->ln_number < b->ln_number)
1326 BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1331 * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1332 * @num_gh: the number of structures
1333 * @ghs: an array of struct gfs2_holder structures
1335 * Returns: 0 on success (all glocks acquired),
1336 * errno on failure (no glocks acquired)
1339 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1340 struct gfs2_holder **p)
1345 for (x = 0; x < num_gh; x++)
1348 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1350 for (x = 0; x < num_gh; x++) {
1351 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1353 error = gfs2_glock_nq(p[x]);
1356 gfs2_glock_dq(p[x]);
1365 * gfs2_glock_nq_m - acquire multiple glocks
1366 * @num_gh: the number of structures
1367 * @ghs: an array of struct gfs2_holder structures
1370 * Returns: 0 on success (all glocks acquired),
1371 * errno on failure (no glocks acquired)
1374 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1376 struct gfs2_holder *tmp[4];
1377 struct gfs2_holder **pph = tmp;
1384 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1385 return gfs2_glock_nq(ghs);
1389 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1394 error = nq_m_sync(num_gh, ghs, pph);
1403 * gfs2_glock_dq_m - release multiple glocks
1404 * @num_gh: the number of structures
1405 * @ghs: an array of struct gfs2_holder structures
1409 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1413 for (x = 0; x < num_gh; x++)
1414 gfs2_glock_dq(&ghs[x]);
1418 * gfs2_glock_dq_uninit_m - release multiple glocks
1419 * @num_gh: the number of structures
1420 * @ghs: an array of struct gfs2_holder structures
1424 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1428 for (x = 0; x < num_gh; x++)
1429 gfs2_glock_dq_uninit(&ghs[x]);
1433 * gfs2_lvb_hold - attach a LVB from a glock
1434 * @gl: The glock in question
1438 int gfs2_lvb_hold(struct gfs2_glock *gl)
1442 gfs2_glmutex_lock(gl);
1444 if (!atomic_read(&gl->gl_lvb_count)) {
1445 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1447 gfs2_glmutex_unlock(gl);
1450 gfs2_glock_hold(gl);
1452 atomic_inc(&gl->gl_lvb_count);
1454 gfs2_glmutex_unlock(gl);
1460 * gfs2_lvb_unhold - detach a LVB from a glock
1461 * @gl: The glock in question
1465 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1467 gfs2_glock_hold(gl);
1468 gfs2_glmutex_lock(gl);
1470 gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1471 if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1472 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1477 gfs2_glmutex_unlock(gl);
1481 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1484 struct gfs2_glock *gl;
1485 unsigned long delay = 0;
1486 unsigned long holdtime;
1487 unsigned long now = jiffies;
1489 gl = gfs2_glock_find(sdp, name);
1493 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1494 if (time_before(now, holdtime))
1495 delay = holdtime - now;
1497 handle_callback(gl, state, 1, delay);
1498 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1503 * gfs2_glock_cb - Callback used by locking module
1504 * @sdp: Pointer to the superblock
1505 * @type: Type of callback
1506 * @data: Type dependent data pointer
1508 * Called by the locking module when it wants to tell us something.
1509 * Either we need to drop a lock, one of our ASYNC requests completed, or
1510 * a journal from another client needs to be recovered.
1513 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1515 struct gfs2_sbd *sdp = cb_data;
1519 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1523 blocking_cb(sdp, data, LM_ST_DEFERRED);
1527 blocking_cb(sdp, data, LM_ST_SHARED);
1531 struct lm_async_cb *async = data;
1532 struct gfs2_glock *gl;
1534 down_read(&gfs2_umount_flush_sem);
1535 gl = gfs2_glock_find(sdp, &async->lc_name);
1536 if (gfs2_assert_warn(sdp, gl))
1538 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1539 gl->gl_req_bh(gl, async->lc_ret);
1540 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1542 up_read(&gfs2_umount_flush_sem);
1546 case LM_CB_NEED_RECOVERY:
1547 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1548 if (sdp->sd_recoverd_process)
1549 wake_up_process(sdp->sd_recoverd_process);
1552 case LM_CB_DROPLOCKS:
1553 gfs2_gl_hash_clear(sdp, NO_WAIT);
1554 gfs2_quota_scan(sdp);
1558 gfs2_assert_warn(sdp, 0);
1564 * demote_ok - Check to see if it's ok to unlock a glock
1567 * Returns: 1 if it's ok
1570 static int demote_ok(struct gfs2_glock *gl)
1572 const struct gfs2_glock_operations *glops = gl->gl_ops;
1575 if (test_bit(GLF_STICKY, &gl->gl_flags))
1577 else if (glops->go_demote_ok)
1578 demote = glops->go_demote_ok(gl);
1584 * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1589 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1591 struct gfs2_sbd *sdp = gl->gl_sbd;
1593 spin_lock(&sdp->sd_reclaim_lock);
1594 if (list_empty(&gl->gl_reclaim)) {
1595 gfs2_glock_hold(gl);
1596 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1597 atomic_inc(&sdp->sd_reclaim_count);
1599 spin_unlock(&sdp->sd_reclaim_lock);
1601 wake_up(&sdp->sd_reclaim_wq);
1605 * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1606 * @sdp: the filesystem
1608 * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1609 * different glock and we notice that there are a lot of glocks in the
1614 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1616 struct gfs2_glock *gl;
1618 spin_lock(&sdp->sd_reclaim_lock);
1619 if (list_empty(&sdp->sd_reclaim_list)) {
1620 spin_unlock(&sdp->sd_reclaim_lock);
1623 gl = list_entry(sdp->sd_reclaim_list.next,
1624 struct gfs2_glock, gl_reclaim);
1625 list_del_init(&gl->gl_reclaim);
1626 spin_unlock(&sdp->sd_reclaim_lock);
1628 atomic_dec(&sdp->sd_reclaim_count);
1629 atomic_inc(&sdp->sd_reclaimed);
1631 if (gfs2_glmutex_trylock(gl)) {
1632 if (list_empty(&gl->gl_holders) &&
1633 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1634 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1635 gfs2_glmutex_unlock(gl);
1642 * examine_bucket - Call a function for glock in a hash bucket
1643 * @examiner: the function
1644 * @sdp: the filesystem
1645 * @bucket: the bucket
1647 * Returns: 1 if the bucket has entries
1650 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1653 struct gfs2_glock *gl, *prev = NULL;
1654 int has_entries = 0;
1655 struct hlist_head *head = &gl_hash_table[hash].hb_list;
1657 read_lock(gl_lock_addr(hash));
1658 /* Can't use hlist_for_each_entry - don't want prefetch here */
1659 if (hlist_empty(head))
1661 gl = list_entry(head->first, struct gfs2_glock, gl_list);
1663 if (!sdp || gl->gl_sbd == sdp) {
1664 gfs2_glock_hold(gl);
1665 read_unlock(gl_lock_addr(hash));
1667 gfs2_glock_put(prev);
1671 read_lock(gl_lock_addr(hash));
1673 if (gl->gl_list.next == NULL)
1675 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1678 read_unlock(gl_lock_addr(hash));
1680 gfs2_glock_put(prev);
1686 * scan_glock - look at a glock and see if we can reclaim it
1687 * @gl: the glock to look at
1691 static void scan_glock(struct gfs2_glock *gl)
1693 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1696 if (gfs2_glmutex_trylock(gl)) {
1697 if (list_empty(&gl->gl_holders) &&
1698 gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1700 gfs2_glmutex_unlock(gl);
1705 gfs2_glmutex_unlock(gl);
1706 gfs2_glock_schedule_for_reclaim(gl);
1710 * clear_glock - look at a glock and see if we can free it from glock cache
1711 * @gl: the glock to look at
1715 static void clear_glock(struct gfs2_glock *gl)
1717 struct gfs2_sbd *sdp = gl->gl_sbd;
1720 spin_lock(&sdp->sd_reclaim_lock);
1721 if (!list_empty(&gl->gl_reclaim)) {
1722 list_del_init(&gl->gl_reclaim);
1723 atomic_dec(&sdp->sd_reclaim_count);
1724 spin_unlock(&sdp->sd_reclaim_lock);
1725 released = gfs2_glock_put(gl);
1726 gfs2_assert(sdp, !released);
1728 spin_unlock(&sdp->sd_reclaim_lock);
1731 if (gfs2_glmutex_trylock(gl)) {
1732 if (list_empty(&gl->gl_holders) &&
1733 gl->gl_state != LM_ST_UNLOCKED)
1734 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1735 gfs2_glmutex_unlock(gl);
1740 * gfs2_gl_hash_clear - Empty out the glock hash table
1741 * @sdp: the filesystem
1742 * @wait: wait until it's all gone
1744 * Called when unmounting the filesystem, or when inter-node lock manager
1745 * requests DROPLOCKS because it is running out of capacity.
1748 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1758 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1759 if (examine_bucket(clear_glock, sdp, x))
1766 if (time_after_eq(jiffies,
1767 t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1768 fs_warn(sdp, "Unmount seems to be stalled. "
1769 "Dumping lock state...\n");
1770 gfs2_dump_lockstate(sdp);
1774 down_write(&gfs2_umount_flush_sem);
1775 invalidate_inodes(sdp->sd_vfs);
1776 up_write(&gfs2_umount_flush_sem);
1782 * Diagnostic routines to help debug distributed deadlock
1785 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1786 unsigned long address)
1788 char buffer[KSYM_SYMBOL_LEN];
1790 sprint_symbol(buffer, address);
1791 print_dbg(gi, fmt, buffer);
1795 * dump_holder - print information about a glock holder
1796 * @str: a string naming the type of holder
1797 * @gh: the glock holder
1799 * Returns: 0 on success, -ENOBUFS when we run out of space
1802 static int dump_holder(struct glock_iter *gi, char *str,
1803 struct gfs2_holder *gh)
1806 struct task_struct *gh_owner;
1808 print_dbg(gi, " %s\n", str);
1809 if (gh->gh_owner_pid) {
1810 print_dbg(gi, " owner = %ld ", (long)gh->gh_owner_pid);
1811 gh_owner = find_task_by_pid(gh->gh_owner_pid);
1813 print_dbg(gi, "(%s)\n", gh_owner->comm);
1815 print_dbg(gi, "(ended)\n");
1817 print_dbg(gi, " owner = -1\n");
1818 print_dbg(gi, " gh_state = %u\n", gh->gh_state);
1819 print_dbg(gi, " gh_flags =");
1820 for (x = 0; x < 32; x++)
1821 if (gh->gh_flags & (1 << x))
1822 print_dbg(gi, " %u", x);
1823 print_dbg(gi, " \n");
1824 print_dbg(gi, " error = %d\n", gh->gh_error);
1825 print_dbg(gi, " gh_iflags =");
1826 for (x = 0; x < 32; x++)
1827 if (test_bit(x, &gh->gh_iflags))
1828 print_dbg(gi, " %u", x);
1829 print_dbg(gi, " \n");
1830 gfs2_print_symbol(gi, " initialized at: %s\n", gh->gh_ip);
1836 * dump_inode - print information about an inode
1839 * Returns: 0 on success, -ENOBUFS when we run out of space
1842 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1846 print_dbg(gi, " Inode:\n");
1847 print_dbg(gi, " num = %llu/%llu\n",
1848 (unsigned long long)ip->i_no_formal_ino,
1849 (unsigned long long)ip->i_no_addr);
1850 print_dbg(gi, " type = %u\n", IF2DT(ip->i_inode.i_mode));
1851 print_dbg(gi, " i_flags =");
1852 for (x = 0; x < 32; x++)
1853 if (test_bit(x, &ip->i_flags))
1854 print_dbg(gi, " %u", x);
1855 print_dbg(gi, " \n");
1860 * dump_glock - print information about a glock
1862 * @count: where we are in the buffer
1864 * Returns: 0 on success, -ENOBUFS when we run out of space
1867 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1869 struct gfs2_holder *gh;
1871 int error = -ENOBUFS;
1872 struct task_struct *gl_owner;
1874 spin_lock(&gl->gl_spin);
1876 print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
1877 (unsigned long long)gl->gl_name.ln_number);
1878 print_dbg(gi, " gl_flags =");
1879 for (x = 0; x < 32; x++) {
1880 if (test_bit(x, &gl->gl_flags))
1881 print_dbg(gi, " %u", x);
1883 if (!test_bit(GLF_LOCK, &gl->gl_flags))
1884 print_dbg(gi, " (unlocked)");
1885 print_dbg(gi, " \n");
1886 print_dbg(gi, " gl_ref = %d\n", atomic_read(&gl->gl_ref));
1887 print_dbg(gi, " gl_state = %u\n", gl->gl_state);
1888 if (gl->gl_owner_pid) {
1889 gl_owner = find_task_by_pid(gl->gl_owner_pid);
1891 print_dbg(gi, " gl_owner = pid %d (%s)\n",
1892 gl->gl_owner_pid, gl_owner->comm);
1894 print_dbg(gi, " gl_owner = %d (ended)\n",
1897 print_dbg(gi, " gl_owner = -1\n");
1898 print_dbg(gi, " gl_ip = %lu\n", gl->gl_ip);
1899 print_dbg(gi, " req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1900 print_dbg(gi, " req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1901 print_dbg(gi, " lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1902 print_dbg(gi, " object = %s\n", (gl->gl_object) ? "yes" : "no");
1903 print_dbg(gi, " le = %s\n",
1904 (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
1905 print_dbg(gi, " reclaim = %s\n",
1906 (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1908 print_dbg(gi, " aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1909 gl->gl_aspace->i_mapping->nrpages);
1911 print_dbg(gi, " aspace = no\n");
1912 print_dbg(gi, " ail = %d\n", atomic_read(&gl->gl_ail_count));
1913 if (gl->gl_req_gh) {
1914 error = dump_holder(gi, "Request", gl->gl_req_gh);
1918 list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1919 error = dump_holder(gi, "Holder", gh);
1923 list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1924 error = dump_holder(gi, "Waiter1", gh);
1928 list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1929 error = dump_holder(gi, "Waiter3", gh);
1933 if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1934 print_dbg(gi, " Demotion req to state %u (%llu uS ago)\n",
1935 gl->gl_demote_state, (unsigned long long)
1936 (jiffies - gl->gl_demote_time)*(1000000/HZ));
1938 if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1939 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1940 list_empty(&gl->gl_holders)) {
1941 error = dump_inode(gi, gl->gl_object);
1946 print_dbg(gi, " Inode: busy\n");
1953 spin_unlock(&gl->gl_spin);
1958 * gfs2_dump_lockstate - print out the current lockstate
1959 * @sdp: the filesystem
1960 * @ub: the buffer to copy the information into
1962 * If @ub is NULL, dump the lockstate to the console.
1966 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1968 struct gfs2_glock *gl;
1969 struct hlist_node *h;
1973 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1975 read_lock(gl_lock_addr(x));
1977 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1978 if (gl->gl_sbd != sdp)
1981 error = dump_glock(NULL, gl);
1986 read_unlock(gl_lock_addr(x));
1997 * gfs2_scand - Look for cached glocks and inodes to toss from memory
1998 * @sdp: Pointer to GFS2 superblock
2000 * One of these daemons runs, finding candidates to add to sd_reclaim_list.
2004 static int gfs2_scand(void *data)
2009 while (!kthread_should_stop()) {
2010 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2011 examine_bucket(scan_glock, NULL, x);
2012 if (freezing(current))
2017 schedule_timeout_interruptible(delay * HZ);
2025 int __init gfs2_glock_init(void)
2028 for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2029 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2031 #ifdef GL_HASH_LOCK_SZ
2032 for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2033 rwlock_init(&gl_hash_locks[i]);
2037 scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
2038 if (IS_ERR(scand_process))
2039 return PTR_ERR(scand_process);
2041 glock_workqueue = create_workqueue("glock_workqueue");
2042 if (IS_ERR(glock_workqueue)) {
2043 kthread_stop(scand_process);
2044 return PTR_ERR(glock_workqueue);
2050 void gfs2_glock_exit(void)
2052 destroy_workqueue(glock_workqueue);
2053 kthread_stop(scand_process);
2056 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
2057 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
2059 static int gfs2_glock_iter_next(struct glock_iter *gi)
2061 struct gfs2_glock *gl;
2064 read_lock(gl_lock_addr(gi->hash));
2067 gi->gl = hlist_entry(gl->gl_list.next,
2068 struct gfs2_glock, gl_list);
2070 gfs2_glock_hold(gi->gl);
2072 read_unlock(gl_lock_addr(gi->hash));
2075 if (gl && gi->gl == NULL)
2077 while(gi->gl == NULL) {
2078 if (gi->hash >= GFS2_GL_HASH_SIZE)
2080 read_lock(gl_lock_addr(gi->hash));
2081 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
2082 struct gfs2_glock, gl_list);
2084 gfs2_glock_hold(gi->gl);
2085 read_unlock(gl_lock_addr(gi->hash));
2089 if (gi->sdp != gi->gl->gl_sbd)
2095 static void gfs2_glock_iter_free(struct glock_iter *gi)
2098 gfs2_glock_put(gi->gl);
2102 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2104 struct glock_iter *gi;
2106 gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2114 memset(gi->string, 0, sizeof(gi->string));
2116 if (gfs2_glock_iter_next(gi)) {
2117 gfs2_glock_iter_free(gi);
2124 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2126 struct glock_iter *gi;
2129 gi = gfs2_glock_iter_init(file->private);
2134 if (gfs2_glock_iter_next(gi)) {
2135 gfs2_glock_iter_free(gi);
2143 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2146 struct glock_iter *gi = iter_ptr;
2150 if (gfs2_glock_iter_next(gi)) {
2151 gfs2_glock_iter_free(gi);
2158 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2160 struct glock_iter *gi = iter_ptr;
2162 gfs2_glock_iter_free(gi);
2165 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2167 struct glock_iter *gi = iter_ptr;
2170 dump_glock(gi, gi->gl);
2175 static const struct seq_operations gfs2_glock_seq_ops = {
2176 .start = gfs2_glock_seq_start,
2177 .next = gfs2_glock_seq_next,
2178 .stop = gfs2_glock_seq_stop,
2179 .show = gfs2_glock_seq_show,
2182 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2184 struct seq_file *seq;
2187 ret = seq_open(file, &gfs2_glock_seq_ops);
2191 seq = file->private_data;
2192 seq->private = inode->i_private;
2197 static const struct file_operations gfs2_debug_fops = {
2198 .owner = THIS_MODULE,
2199 .open = gfs2_debugfs_open,
2201 .llseek = seq_lseek,
2202 .release = seq_release
2205 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2207 sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2208 if (!sdp->debugfs_dir)
2210 sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2212 sdp->debugfs_dir, sdp,
2214 if (!sdp->debugfs_dentry_glocks)
2220 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2222 if (sdp && sdp->debugfs_dir) {
2223 if (sdp->debugfs_dentry_glocks) {
2224 debugfs_remove(sdp->debugfs_dentry_glocks);
2225 sdp->debugfs_dentry_glocks = NULL;
2227 debugfs_remove(sdp->debugfs_dir);
2228 sdp->debugfs_dir = NULL;
2232 int gfs2_register_debugfs(void)
2234 gfs2_root = debugfs_create_dir("gfs2", NULL);
2235 return gfs2_root ? 0 : -ENOMEM;
2238 void gfs2_unregister_debugfs(void)
2240 debugfs_remove(gfs2_root);