xref: /openbmc/linux/fs/gfs2/glock.c (revision b1e058da50f7938e9c9e963e978b0730bba4ad32)
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
4  *
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.
8  */
9 
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>
32 
33 #include "gfs2.h"
34 #include "incore.h"
35 #include "glock.h"
36 #include "glops.h"
37 #include "inode.h"
38 #include "lm.h"
39 #include "lops.h"
40 #include "meta_io.h"
41 #include "quota.h"
42 #include "super.h"
43 #include "util.h"
44 
45 struct gfs2_gl_hash_bucket {
46         struct hlist_head hb_list;
47 };
48 
49 struct glock_iter {
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             */
55 };
56 
57 typedef void (*glock_examiner) (struct gfs2_glock * gl);
58 
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);
64 
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;
70 
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)
74 
75 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
76 static struct dentry *gfs2_root;
77 
78 /*
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.
83  *
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.
87  */
88 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
89 	defined(CONFIG_PROVE_LOCKING)
90 
91 #ifdef CONFIG_LOCKDEP
92 # define GL_HASH_LOCK_SZ        256
93 #else
94 # if NR_CPUS >= 32
95 #  define GL_HASH_LOCK_SZ       4096
96 # elif NR_CPUS >= 16
97 #  define GL_HASH_LOCK_SZ       2048
98 # elif NR_CPUS >= 8
99 #  define GL_HASH_LOCK_SZ       1024
100 # elif NR_CPUS >= 4
101 #  define GL_HASH_LOCK_SZ       512
102 # else
103 #  define GL_HASH_LOCK_SZ       256
104 # endif
105 #endif
106 
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
111 #endif
112 
113 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
114 
115 static inline rwlock_t *gl_lock_addr(unsigned int x)
116 {
117 	return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
118 }
119 #else /* not SMP, so no spinlocks required */
120 static inline rwlock_t *gl_lock_addr(unsigned int x)
121 {
122 	return NULL;
123 }
124 #endif
125 
126 /**
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
131  *
132  * Returns: 1 if the locks are compatible, 0 otherwise
133  */
134 
135 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
136 				   int flags)
137 {
138 	if (actual == requested)
139 		return 1;
140 
141 	if (flags & GL_EXACT)
142 		return 0;
143 
144 	if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
145 		return 1;
146 
147 	if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
148 		return 1;
149 
150 	return 0;
151 }
152 
153 /**
154  * gl_hash() - Turn glock number into hash bucket number
155  * @lock: The glock number
156  *
157  * Returns: The number of the corresponding hash bucket
158  */
159 
160 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
161 			    const struct lm_lockname *name)
162 {
163 	unsigned int h;
164 
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;
169 
170 	return h;
171 }
172 
173 /**
174  * glock_free() - Perform a few checks and then release struct gfs2_glock
175  * @gl: The glock to release
176  *
177  * Also calls lock module to release its internal structure for this glock.
178  *
179  */
180 
181 static void glock_free(struct gfs2_glock *gl)
182 {
183 	struct gfs2_sbd *sdp = gl->gl_sbd;
184 	struct inode *aspace = gl->gl_aspace;
185 
186 	gfs2_lm_put_lock(sdp, gl->gl_lock);
187 
188 	if (aspace)
189 		gfs2_aspace_put(aspace);
190 
191 	kmem_cache_free(gfs2_glock_cachep, gl);
192 }
193 
194 /**
195  * gfs2_glock_hold() - increment reference count on glock
196  * @gl: The glock to hold
197  *
198  */
199 
200 void gfs2_glock_hold(struct gfs2_glock *gl)
201 {
202 	atomic_inc(&gl->gl_ref);
203 }
204 
205 /**
206  * gfs2_glock_put() - Decrement reference count on glock
207  * @gl: The glock to put
208  *
209  */
210 
211 int gfs2_glock_put(struct gfs2_glock *gl)
212 {
213 	int rv = 0;
214 	struct gfs2_sbd *sdp = gl->gl_sbd;
215 
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 		gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
221 		gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
222 		gfs2_assert(sdp, list_empty(&gl->gl_holders));
223 		gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
224 		gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
225 		glock_free(gl);
226 		rv = 1;
227 		goto out;
228 	}
229 	write_unlock(gl_lock_addr(gl->gl_hash));
230 out:
231 	return rv;
232 }
233 
234 /**
235  * search_bucket() - Find struct gfs2_glock by lock number
236  * @bucket: the bucket to search
237  * @name: The lock name
238  *
239  * Returns: NULL, or the struct gfs2_glock with the requested number
240  */
241 
242 static struct gfs2_glock *search_bucket(unsigned int hash,
243 					const struct gfs2_sbd *sdp,
244 					const struct lm_lockname *name)
245 {
246 	struct gfs2_glock *gl;
247 	struct hlist_node *h;
248 
249 	hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
250 		if (!lm_name_equal(&gl->gl_name, name))
251 			continue;
252 		if (gl->gl_sbd != sdp)
253 			continue;
254 
255 		atomic_inc(&gl->gl_ref);
256 
257 		return gl;
258 	}
259 
260 	return NULL;
261 }
262 
263 /**
264  * gfs2_glock_find() - Find glock by lock number
265  * @sdp: The GFS2 superblock
266  * @name: The lock name
267  *
268  * Returns: NULL, or the struct gfs2_glock with the requested number
269  */
270 
271 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
272 					  const struct lm_lockname *name)
273 {
274 	unsigned int hash = gl_hash(sdp, name);
275 	struct gfs2_glock *gl;
276 
277 	read_lock(gl_lock_addr(hash));
278 	gl = search_bucket(hash, sdp, name);
279 	read_unlock(gl_lock_addr(hash));
280 
281 	return gl;
282 }
283 
284 static void glock_work_func(struct work_struct *work)
285 {
286 	struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
287 
288 	spin_lock(&gl->gl_spin);
289 	if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags))
290 		set_bit(GLF_DEMOTE, &gl->gl_flags);
291 	run_queue(gl);
292 	spin_unlock(&gl->gl_spin);
293 	gfs2_glock_put(gl);
294 }
295 
296 /**
297  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
298  * @sdp: The GFS2 superblock
299  * @number: the lock number
300  * @glops: The glock_operations to use
301  * @create: If 0, don't create the glock if it doesn't exist
302  * @glp: the glock is returned here
303  *
304  * This does not lock a glock, just finds/creates structures for one.
305  *
306  * Returns: errno
307  */
308 
309 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
310 		   const struct gfs2_glock_operations *glops, int create,
311 		   struct gfs2_glock **glp)
312 {
313 	struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
314 	struct gfs2_glock *gl, *tmp;
315 	unsigned int hash = gl_hash(sdp, &name);
316 	int error;
317 
318 	read_lock(gl_lock_addr(hash));
319 	gl = search_bucket(hash, sdp, &name);
320 	read_unlock(gl_lock_addr(hash));
321 
322 	if (gl || !create) {
323 		*glp = gl;
324 		return 0;
325 	}
326 
327 	gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
328 	if (!gl)
329 		return -ENOMEM;
330 
331 	gl->gl_flags = 0;
332 	gl->gl_name = name;
333 	atomic_set(&gl->gl_ref, 1);
334 	gl->gl_state = LM_ST_UNLOCKED;
335 	gl->gl_demote_state = LM_ST_EXCLUSIVE;
336 	gl->gl_hash = hash;
337 	gl->gl_owner_pid = 0;
338 	gl->gl_ip = 0;
339 	gl->gl_ops = glops;
340 	gl->gl_req_gh = NULL;
341 	gl->gl_req_bh = NULL;
342 	gl->gl_vn = 0;
343 	gl->gl_stamp = jiffies;
344 	gl->gl_tchange = jiffies;
345 	gl->gl_object = NULL;
346 	gl->gl_sbd = sdp;
347 	gl->gl_aspace = NULL;
348 	INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
349 
350 	/* If this glock protects actual on-disk data or metadata blocks,
351 	   create a VFS inode to manage the pages/buffers holding them. */
352 	if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
353 		gl->gl_aspace = gfs2_aspace_get(sdp);
354 		if (!gl->gl_aspace) {
355 			error = -ENOMEM;
356 			goto fail;
357 		}
358 	}
359 
360 	error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
361 	if (error)
362 		goto fail_aspace;
363 
364 	write_lock(gl_lock_addr(hash));
365 	tmp = search_bucket(hash, sdp, &name);
366 	if (tmp) {
367 		write_unlock(gl_lock_addr(hash));
368 		glock_free(gl);
369 		gl = tmp;
370 	} else {
371 		hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
372 		write_unlock(gl_lock_addr(hash));
373 	}
374 
375 	*glp = gl;
376 
377 	return 0;
378 
379 fail_aspace:
380 	if (gl->gl_aspace)
381 		gfs2_aspace_put(gl->gl_aspace);
382 fail:
383 	kmem_cache_free(gfs2_glock_cachep, gl);
384 	return error;
385 }
386 
387 /**
388  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
389  * @gl: the glock
390  * @state: the state we're requesting
391  * @flags: the modifier flags
392  * @gh: the holder structure
393  *
394  */
395 
396 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
397 		      struct gfs2_holder *gh)
398 {
399 	INIT_LIST_HEAD(&gh->gh_list);
400 	gh->gh_gl = gl;
401 	gh->gh_ip = (unsigned long)__builtin_return_address(0);
402 	gh->gh_owner_pid = get_pid(task_pid(current));
403 	gh->gh_state = state;
404 	gh->gh_flags = flags;
405 	gh->gh_error = 0;
406 	gh->gh_iflags = 0;
407 	gfs2_glock_hold(gl);
408 }
409 
410 /**
411  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
412  * @state: the state we're requesting
413  * @flags: the modifier flags
414  * @gh: the holder structure
415  *
416  * Don't mess with the glock.
417  *
418  */
419 
420 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
421 {
422 	gh->gh_state = state;
423 	gh->gh_flags = flags;
424 	gh->gh_iflags = 0;
425 	gh->gh_ip = (unsigned long)__builtin_return_address(0);
426 }
427 
428 /**
429  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
430  * @gh: the holder structure
431  *
432  */
433 
434 void gfs2_holder_uninit(struct gfs2_holder *gh)
435 {
436 	put_pid(gh->gh_owner_pid);
437 	gfs2_glock_put(gh->gh_gl);
438 	gh->gh_gl = NULL;
439 	gh->gh_ip = 0;
440 }
441 
442 static void gfs2_holder_wake(struct gfs2_holder *gh)
443 {
444 	clear_bit(HIF_WAIT, &gh->gh_iflags);
445 	smp_mb__after_clear_bit();
446 	wake_up_bit(&gh->gh_iflags, HIF_WAIT);
447 }
448 
449 static int just_schedule(void *word)
450 {
451         schedule();
452         return 0;
453 }
454 
455 static void wait_on_holder(struct gfs2_holder *gh)
456 {
457 	might_sleep();
458 	wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
459 }
460 
461 static void gfs2_demote_wake(struct gfs2_glock *gl)
462 {
463 	gl->gl_demote_state = LM_ST_EXCLUSIVE;
464         clear_bit(GLF_DEMOTE, &gl->gl_flags);
465         smp_mb__after_clear_bit();
466         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
467 }
468 
469 static void wait_on_demote(struct gfs2_glock *gl)
470 {
471 	might_sleep();
472 	wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
473 }
474 
475 /**
476  * rq_mutex - process a mutex request in the queue
477  * @gh: the glock holder
478  *
479  * Returns: 1 if the queue is blocked
480  */
481 
482 static int rq_mutex(struct gfs2_holder *gh)
483 {
484 	struct gfs2_glock *gl = gh->gh_gl;
485 
486 	list_del_init(&gh->gh_list);
487 	/*  gh->gh_error never examined.  */
488 	set_bit(GLF_LOCK, &gl->gl_flags);
489 	clear_bit(HIF_WAIT, &gh->gh_iflags);
490 	smp_mb();
491 	wake_up_bit(&gh->gh_iflags, HIF_WAIT);
492 
493 	return 1;
494 }
495 
496 /**
497  * rq_promote - process a promote request in the queue
498  * @gh: the glock holder
499  *
500  * Acquire a new inter-node lock, or change a lock state to more restrictive.
501  *
502  * Returns: 1 if the queue is blocked
503  */
504 
505 static int rq_promote(struct gfs2_holder *gh)
506 {
507 	struct gfs2_glock *gl = gh->gh_gl;
508 
509 	if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
510 		if (list_empty(&gl->gl_holders)) {
511 			gl->gl_req_gh = gh;
512 			set_bit(GLF_LOCK, &gl->gl_flags);
513 			spin_unlock(&gl->gl_spin);
514 			gfs2_glock_xmote_th(gh->gh_gl, gh);
515 			spin_lock(&gl->gl_spin);
516 		}
517 		return 1;
518 	}
519 
520 	if (list_empty(&gl->gl_holders)) {
521 		set_bit(HIF_FIRST, &gh->gh_iflags);
522 		set_bit(GLF_LOCK, &gl->gl_flags);
523 	} else {
524 		struct gfs2_holder *next_gh;
525 		if (gh->gh_state == LM_ST_EXCLUSIVE)
526 			return 1;
527 		next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
528 				     gh_list);
529 		if (next_gh->gh_state == LM_ST_EXCLUSIVE)
530 			 return 1;
531 	}
532 
533 	list_move_tail(&gh->gh_list, &gl->gl_holders);
534 	gh->gh_error = 0;
535 	set_bit(HIF_HOLDER, &gh->gh_iflags);
536 
537 	gfs2_holder_wake(gh);
538 
539 	return 0;
540 }
541 
542 /**
543  * rq_demote - process a demote request in the queue
544  * @gh: the glock holder
545  *
546  * Returns: 1 if the queue is blocked
547  */
548 
549 static int rq_demote(struct gfs2_glock *gl)
550 {
551 	if (!list_empty(&gl->gl_holders))
552 		return 1;
553 
554 	if (gl->gl_state == gl->gl_demote_state ||
555 	    gl->gl_state == LM_ST_UNLOCKED) {
556 		gfs2_demote_wake(gl);
557 		return 0;
558 	}
559 
560 	set_bit(GLF_LOCK, &gl->gl_flags);
561 	set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
562 
563 	if (gl->gl_demote_state == LM_ST_UNLOCKED ||
564 	    gl->gl_state != LM_ST_EXCLUSIVE) {
565 		spin_unlock(&gl->gl_spin);
566 		gfs2_glock_drop_th(gl);
567 	} else {
568 		spin_unlock(&gl->gl_spin);
569 		gfs2_glock_xmote_th(gl, NULL);
570 	}
571 
572 	spin_lock(&gl->gl_spin);
573 	clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
574 
575 	return 0;
576 }
577 
578 /**
579  * run_queue - process holder structures on a glock
580  * @gl: the glock
581  *
582  */
583 static void run_queue(struct gfs2_glock *gl)
584 {
585 	struct gfs2_holder *gh;
586 	int blocked = 1;
587 
588 	for (;;) {
589 		if (test_bit(GLF_LOCK, &gl->gl_flags))
590 			break;
591 
592 		if (!list_empty(&gl->gl_waiters1)) {
593 			gh = list_entry(gl->gl_waiters1.next,
594 					struct gfs2_holder, gh_list);
595 			blocked = rq_mutex(gh);
596 		} else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
597 			blocked = rq_demote(gl);
598 			if (gl->gl_waiters2 && !blocked) {
599 				set_bit(GLF_DEMOTE, &gl->gl_flags);
600 				gl->gl_demote_state = LM_ST_UNLOCKED;
601 			}
602 			gl->gl_waiters2 = 0;
603 		} else if (!list_empty(&gl->gl_waiters3)) {
604 			gh = list_entry(gl->gl_waiters3.next,
605 					struct gfs2_holder, gh_list);
606 			blocked = rq_promote(gh);
607 		} else
608 			break;
609 
610 		if (blocked)
611 			break;
612 	}
613 }
614 
615 /**
616  * gfs2_glmutex_lock - acquire a local lock on a glock
617  * @gl: the glock
618  *
619  * Gives caller exclusive access to manipulate a glock structure.
620  */
621 
622 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
623 {
624 	spin_lock(&gl->gl_spin);
625 	if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
626 		struct gfs2_holder gh;
627 
628 		gfs2_holder_init(gl, 0, 0, &gh);
629 		set_bit(HIF_WAIT, &gh.gh_iflags);
630 		list_add_tail(&gh.gh_list, &gl->gl_waiters1);
631 		spin_unlock(&gl->gl_spin);
632 		wait_on_holder(&gh);
633 		gfs2_holder_uninit(&gh);
634 	} else {
635 		gl->gl_owner_pid = current->pid;
636 		gl->gl_ip = (unsigned long)__builtin_return_address(0);
637 		spin_unlock(&gl->gl_spin);
638 	}
639 }
640 
641 /**
642  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
643  * @gl: the glock
644  *
645  * Returns: 1 if the glock is acquired
646  */
647 
648 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
649 {
650 	int acquired = 1;
651 
652 	spin_lock(&gl->gl_spin);
653 	if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
654 		acquired = 0;
655 	} else {
656 		gl->gl_owner_pid = current->pid;
657 		gl->gl_ip = (unsigned long)__builtin_return_address(0);
658 	}
659 	spin_unlock(&gl->gl_spin);
660 
661 	return acquired;
662 }
663 
664 /**
665  * gfs2_glmutex_unlock - release a local lock on a glock
666  * @gl: the glock
667  *
668  */
669 
670 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
671 {
672 	spin_lock(&gl->gl_spin);
673 	clear_bit(GLF_LOCK, &gl->gl_flags);
674 	gl->gl_owner_pid = 0;
675 	gl->gl_ip = 0;
676 	run_queue(gl);
677 	spin_unlock(&gl->gl_spin);
678 }
679 
680 /**
681  * handle_callback - process a demote request
682  * @gl: the glock
683  * @state: the state the caller wants us to change to
684  *
685  * There are only two requests that we are going to see in actual
686  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
687  */
688 
689 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
690 			    int remote, unsigned long delay)
691 {
692 	int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
693 
694 	spin_lock(&gl->gl_spin);
695 	set_bit(bit, &gl->gl_flags);
696 	if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
697 		gl->gl_demote_state = state;
698 		gl->gl_demote_time = jiffies;
699 		if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
700 		    gl->gl_object) {
701 			gfs2_glock_schedule_for_reclaim(gl);
702 			spin_unlock(&gl->gl_spin);
703 			return;
704 		}
705 	} else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
706 			gl->gl_demote_state != state) {
707 		if (test_bit(GLF_DEMOTE_IN_PROGRESS,  &gl->gl_flags))
708 			gl->gl_waiters2 = 1;
709 		else
710 			gl->gl_demote_state = LM_ST_UNLOCKED;
711 	}
712 	spin_unlock(&gl->gl_spin);
713 }
714 
715 /**
716  * state_change - record that the glock is now in a different state
717  * @gl: the glock
718  * @new_state the new state
719  *
720  */
721 
722 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
723 {
724 	int held1, held2;
725 
726 	held1 = (gl->gl_state != LM_ST_UNLOCKED);
727 	held2 = (new_state != LM_ST_UNLOCKED);
728 
729 	if (held1 != held2) {
730 		if (held2)
731 			gfs2_glock_hold(gl);
732 		else
733 			gfs2_glock_put(gl);
734 	}
735 
736 	gl->gl_state = new_state;
737 	gl->gl_tchange = jiffies;
738 }
739 
740 /**
741  * xmote_bh - Called after the lock module is done acquiring a lock
742  * @gl: The glock in question
743  * @ret: the int returned from the lock module
744  *
745  */
746 
747 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
748 {
749 	struct gfs2_sbd *sdp = gl->gl_sbd;
750 	const struct gfs2_glock_operations *glops = gl->gl_ops;
751 	struct gfs2_holder *gh = gl->gl_req_gh;
752 	int prev_state = gl->gl_state;
753 	int op_done = 1;
754 
755 	gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
756 	gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
757 	gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
758 
759 	state_change(gl, ret & LM_OUT_ST_MASK);
760 
761 	if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
762 		if (glops->go_inval)
763 			glops->go_inval(gl, DIO_METADATA);
764 	} else if (gl->gl_state == LM_ST_DEFERRED) {
765 		/* We might not want to do this here.
766 		   Look at moving to the inode glops. */
767 		if (glops->go_inval)
768 			glops->go_inval(gl, 0);
769 	}
770 
771 	/*  Deal with each possible exit condition  */
772 
773 	if (!gh) {
774 		gl->gl_stamp = jiffies;
775 		if (ret & LM_OUT_CANCELED) {
776 			op_done = 0;
777 		} else {
778 			spin_lock(&gl->gl_spin);
779 			if (gl->gl_state != gl->gl_demote_state) {
780 				gl->gl_req_bh = NULL;
781 				spin_unlock(&gl->gl_spin);
782 				gfs2_glock_drop_th(gl);
783 				gfs2_glock_put(gl);
784 				return;
785 			}
786 			gfs2_demote_wake(gl);
787 			spin_unlock(&gl->gl_spin);
788 		}
789 	} else {
790 		spin_lock(&gl->gl_spin);
791 		list_del_init(&gh->gh_list);
792 		gh->gh_error = -EIO;
793 		if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
794 			goto out;
795 		gh->gh_error = GLR_CANCELED;
796 		if (ret & LM_OUT_CANCELED)
797 			goto out;
798 		if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
799 			list_add_tail(&gh->gh_list, &gl->gl_holders);
800 			gh->gh_error = 0;
801 			set_bit(HIF_HOLDER, &gh->gh_iflags);
802 			set_bit(HIF_FIRST, &gh->gh_iflags);
803 			op_done = 0;
804 			goto out;
805 		}
806 		gh->gh_error = GLR_TRYFAILED;
807 		if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
808 			goto out;
809 		gh->gh_error = -EINVAL;
810 		if (gfs2_assert_withdraw(sdp, 0) == -1)
811 			fs_err(sdp, "ret = 0x%.8X\n", ret);
812 out:
813 		spin_unlock(&gl->gl_spin);
814 	}
815 
816 	if (glops->go_xmote_bh)
817 		glops->go_xmote_bh(gl);
818 
819 	if (op_done) {
820 		spin_lock(&gl->gl_spin);
821 		gl->gl_req_gh = NULL;
822 		gl->gl_req_bh = NULL;
823 		clear_bit(GLF_LOCK, &gl->gl_flags);
824 		spin_unlock(&gl->gl_spin);
825 	}
826 
827 	gfs2_glock_put(gl);
828 
829 	if (gh)
830 		gfs2_holder_wake(gh);
831 }
832 
833 /**
834  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
835  * @gl: The glock in question
836  * @state: the requested state
837  * @flags: modifier flags to the lock call
838  *
839  */
840 
841 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
842 {
843 	struct gfs2_sbd *sdp = gl->gl_sbd;
844 	int flags = gh ? gh->gh_flags : 0;
845 	unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
846 	const struct gfs2_glock_operations *glops = gl->gl_ops;
847 	int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
848 				 LM_FLAG_NOEXP | LM_FLAG_ANY |
849 				 LM_FLAG_PRIORITY);
850 	unsigned int lck_ret;
851 
852 	if (glops->go_xmote_th)
853 		glops->go_xmote_th(gl);
854 
855 	gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
856 	gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
857 	gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
858 	gfs2_assert_warn(sdp, state != gl->gl_state);
859 
860 	gfs2_glock_hold(gl);
861 	gl->gl_req_bh = xmote_bh;
862 
863 	lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
864 
865 	if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
866 		return;
867 
868 	if (lck_ret & LM_OUT_ASYNC)
869 		gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
870 	else
871 		xmote_bh(gl, lck_ret);
872 }
873 
874 /**
875  * drop_bh - Called after a lock module unlock completes
876  * @gl: the glock
877  * @ret: the return status
878  *
879  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
880  * Doesn't drop the reference on the glock the top half took out
881  *
882  */
883 
884 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
885 {
886 	struct gfs2_sbd *sdp = gl->gl_sbd;
887 	const struct gfs2_glock_operations *glops = gl->gl_ops;
888 	struct gfs2_holder *gh = gl->gl_req_gh;
889 
890 	gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
891 	gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
892 	gfs2_assert_warn(sdp, !ret);
893 
894 	state_change(gl, LM_ST_UNLOCKED);
895 
896 	if (glops->go_inval)
897 		glops->go_inval(gl, DIO_METADATA);
898 
899 	if (gh) {
900 		spin_lock(&gl->gl_spin);
901 		list_del_init(&gh->gh_list);
902 		gh->gh_error = 0;
903 		spin_unlock(&gl->gl_spin);
904 	}
905 
906 	spin_lock(&gl->gl_spin);
907 	gfs2_demote_wake(gl);
908 	gl->gl_req_gh = NULL;
909 	gl->gl_req_bh = NULL;
910 	clear_bit(GLF_LOCK, &gl->gl_flags);
911 	spin_unlock(&gl->gl_spin);
912 
913 	gfs2_glock_put(gl);
914 
915 	if (gh)
916 		gfs2_holder_wake(gh);
917 }
918 
919 /**
920  * gfs2_glock_drop_th - call into the lock module to unlock a lock
921  * @gl: the glock
922  *
923  */
924 
925 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
926 {
927 	struct gfs2_sbd *sdp = gl->gl_sbd;
928 	const struct gfs2_glock_operations *glops = gl->gl_ops;
929 	unsigned int ret;
930 
931 	if (glops->go_xmote_th)
932 		glops->go_xmote_th(gl);
933 
934 	gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
935 	gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
936 	gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
937 
938 	gfs2_glock_hold(gl);
939 	gl->gl_req_bh = drop_bh;
940 
941 	ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
942 
943 	if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
944 		return;
945 
946 	if (!ret)
947 		drop_bh(gl, ret);
948 	else
949 		gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
950 }
951 
952 /**
953  * do_cancels - cancel requests for locks stuck waiting on an expire flag
954  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
955  *
956  * Don't cancel GL_NOCANCEL requests.
957  */
958 
959 static void do_cancels(struct gfs2_holder *gh)
960 {
961 	struct gfs2_glock *gl = gh->gh_gl;
962 
963 	spin_lock(&gl->gl_spin);
964 
965 	while (gl->gl_req_gh != gh &&
966 	       !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
967 	       !list_empty(&gh->gh_list)) {
968 		if (gl->gl_req_bh && !(gl->gl_req_gh &&
969 				     (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
970 			spin_unlock(&gl->gl_spin);
971 			gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
972 			msleep(100);
973 			spin_lock(&gl->gl_spin);
974 		} else {
975 			spin_unlock(&gl->gl_spin);
976 			msleep(100);
977 			spin_lock(&gl->gl_spin);
978 		}
979 	}
980 
981 	spin_unlock(&gl->gl_spin);
982 }
983 
984 /**
985  * glock_wait_internal - wait on a glock acquisition
986  * @gh: the glock holder
987  *
988  * Returns: 0 on success
989  */
990 
991 static int glock_wait_internal(struct gfs2_holder *gh)
992 {
993 	struct gfs2_glock *gl = gh->gh_gl;
994 	struct gfs2_sbd *sdp = gl->gl_sbd;
995 	const struct gfs2_glock_operations *glops = gl->gl_ops;
996 
997 	if (test_bit(HIF_ABORTED, &gh->gh_iflags))
998 		return -EIO;
999 
1000 	if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1001 		spin_lock(&gl->gl_spin);
1002 		if (gl->gl_req_gh != gh &&
1003 		    !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1004 		    !list_empty(&gh->gh_list)) {
1005 			list_del_init(&gh->gh_list);
1006 			gh->gh_error = GLR_TRYFAILED;
1007 			run_queue(gl);
1008 			spin_unlock(&gl->gl_spin);
1009 			return gh->gh_error;
1010 		}
1011 		spin_unlock(&gl->gl_spin);
1012 	}
1013 
1014 	if (gh->gh_flags & LM_FLAG_PRIORITY)
1015 		do_cancels(gh);
1016 
1017 	wait_on_holder(gh);
1018 	if (gh->gh_error)
1019 		return gh->gh_error;
1020 
1021 	gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1022 	gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1023 						   gh->gh_flags));
1024 
1025 	if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1026 		gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1027 
1028 		if (glops->go_lock) {
1029 			gh->gh_error = glops->go_lock(gh);
1030 			if (gh->gh_error) {
1031 				spin_lock(&gl->gl_spin);
1032 				list_del_init(&gh->gh_list);
1033 				spin_unlock(&gl->gl_spin);
1034 			}
1035 		}
1036 
1037 		spin_lock(&gl->gl_spin);
1038 		gl->gl_req_gh = NULL;
1039 		gl->gl_req_bh = NULL;
1040 		clear_bit(GLF_LOCK, &gl->gl_flags);
1041 		run_queue(gl);
1042 		spin_unlock(&gl->gl_spin);
1043 	}
1044 
1045 	return gh->gh_error;
1046 }
1047 
1048 static inline struct gfs2_holder *
1049 find_holder_by_owner(struct list_head *head, struct pid *pid)
1050 {
1051 	struct gfs2_holder *gh;
1052 
1053 	list_for_each_entry(gh, head, gh_list) {
1054 		if (gh->gh_owner_pid == pid)
1055 			return gh;
1056 	}
1057 
1058 	return NULL;
1059 }
1060 
1061 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1062 {
1063 	va_list args;
1064 
1065 	va_start(args, fmt);
1066 	if (gi) {
1067 		vsprintf(gi->string, fmt, args);
1068 		seq_printf(gi->seq, gi->string);
1069 	}
1070 	else
1071 		vprintk(fmt, args);
1072 	va_end(args);
1073 }
1074 
1075 /**
1076  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1077  * @gh: the holder structure to add
1078  *
1079  */
1080 
1081 static void add_to_queue(struct gfs2_holder *gh)
1082 {
1083 	struct gfs2_glock *gl = gh->gh_gl;
1084 	struct gfs2_holder *existing;
1085 
1086 	BUG_ON(gh->gh_owner_pid == NULL);
1087 	if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1088 		BUG();
1089 
1090 	if (!(gh->gh_flags & GL_FLOCK)) {
1091 		existing = find_holder_by_owner(&gl->gl_holders,
1092 						gh->gh_owner_pid);
1093 		if (existing) {
1094 			print_symbol(KERN_WARNING "original: %s\n",
1095 				     existing->gh_ip);
1096 			printk(KERN_INFO "pid : %d\n",
1097 					pid_nr(existing->gh_owner_pid));
1098 			printk(KERN_INFO "lock type : %d lock state : %d\n",
1099 			       existing->gh_gl->gl_name.ln_type,
1100 			       existing->gh_gl->gl_state);
1101 			print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1102 			printk(KERN_INFO "pid : %d\n",
1103 					pid_nr(gh->gh_owner_pid));
1104 			printk(KERN_INFO "lock type : %d lock state : %d\n",
1105 			       gl->gl_name.ln_type, gl->gl_state);
1106 			BUG();
1107 		}
1108 
1109 		existing = find_holder_by_owner(&gl->gl_waiters3,
1110 						gh->gh_owner_pid);
1111 		if (existing) {
1112 			print_symbol(KERN_WARNING "original: %s\n",
1113 				     existing->gh_ip);
1114 			print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1115 			BUG();
1116 		}
1117 	}
1118 
1119 	if (gh->gh_flags & LM_FLAG_PRIORITY)
1120 		list_add(&gh->gh_list, &gl->gl_waiters3);
1121 	else
1122 		list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1123 }
1124 
1125 /**
1126  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1127  * @gh: the holder structure
1128  *
1129  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1130  *
1131  * Returns: 0, GLR_TRYFAILED, or errno on failure
1132  */
1133 
1134 int gfs2_glock_nq(struct gfs2_holder *gh)
1135 {
1136 	struct gfs2_glock *gl = gh->gh_gl;
1137 	struct gfs2_sbd *sdp = gl->gl_sbd;
1138 	int error = 0;
1139 
1140 restart:
1141 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1142 		set_bit(HIF_ABORTED, &gh->gh_iflags);
1143 		return -EIO;
1144 	}
1145 
1146 	spin_lock(&gl->gl_spin);
1147 	add_to_queue(gh);
1148 	run_queue(gl);
1149 	spin_unlock(&gl->gl_spin);
1150 
1151 	if (!(gh->gh_flags & GL_ASYNC)) {
1152 		error = glock_wait_internal(gh);
1153 		if (error == GLR_CANCELED) {
1154 			msleep(100);
1155 			goto restart;
1156 		}
1157 	}
1158 
1159 	return error;
1160 }
1161 
1162 /**
1163  * gfs2_glock_poll - poll to see if an async request has been completed
1164  * @gh: the holder
1165  *
1166  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1167  */
1168 
1169 int gfs2_glock_poll(struct gfs2_holder *gh)
1170 {
1171 	struct gfs2_glock *gl = gh->gh_gl;
1172 	int ready = 0;
1173 
1174 	spin_lock(&gl->gl_spin);
1175 
1176 	if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1177 		ready = 1;
1178 	else if (list_empty(&gh->gh_list)) {
1179 		if (gh->gh_error == GLR_CANCELED) {
1180 			spin_unlock(&gl->gl_spin);
1181 			msleep(100);
1182 			if (gfs2_glock_nq(gh))
1183 				return 1;
1184 			return 0;
1185 		} else
1186 			ready = 1;
1187 	}
1188 
1189 	spin_unlock(&gl->gl_spin);
1190 
1191 	return ready;
1192 }
1193 
1194 /**
1195  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1196  * @gh: the holder structure
1197  *
1198  * Returns: 0, GLR_TRYFAILED, or errno on failure
1199  */
1200 
1201 int gfs2_glock_wait(struct gfs2_holder *gh)
1202 {
1203 	int error;
1204 
1205 	error = glock_wait_internal(gh);
1206 	if (error == GLR_CANCELED) {
1207 		msleep(100);
1208 		gh->gh_flags &= ~GL_ASYNC;
1209 		error = gfs2_glock_nq(gh);
1210 	}
1211 
1212 	return error;
1213 }
1214 
1215 /**
1216  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1217  * @gh: the glock holder
1218  *
1219  */
1220 
1221 void gfs2_glock_dq(struct gfs2_holder *gh)
1222 {
1223 	struct gfs2_glock *gl = gh->gh_gl;
1224 	const struct gfs2_glock_operations *glops = gl->gl_ops;
1225 	unsigned delay = 0;
1226 
1227 	if (gh->gh_flags & GL_NOCACHE)
1228 		handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1229 
1230 	gfs2_glmutex_lock(gl);
1231 
1232 	spin_lock(&gl->gl_spin);
1233 	list_del_init(&gh->gh_list);
1234 
1235 	if (list_empty(&gl->gl_holders)) {
1236 		if (glops->go_unlock) {
1237 			spin_unlock(&gl->gl_spin);
1238 			glops->go_unlock(gh);
1239 			spin_lock(&gl->gl_spin);
1240 		}
1241 		gl->gl_stamp = jiffies;
1242 	}
1243 
1244 	clear_bit(GLF_LOCK, &gl->gl_flags);
1245 	spin_unlock(&gl->gl_spin);
1246 
1247 	gfs2_glock_hold(gl);
1248 	if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1249 	    !test_bit(GLF_DEMOTE, &gl->gl_flags))
1250 		delay = gl->gl_ops->go_min_hold_time;
1251 	if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1252 		gfs2_glock_put(gl);
1253 }
1254 
1255 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1256 {
1257 	struct gfs2_glock *gl = gh->gh_gl;
1258 	gfs2_glock_dq(gh);
1259 	wait_on_demote(gl);
1260 }
1261 
1262 /**
1263  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1264  * @gh: the holder structure
1265  *
1266  */
1267 
1268 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1269 {
1270 	gfs2_glock_dq(gh);
1271 	gfs2_holder_uninit(gh);
1272 }
1273 
1274 /**
1275  * gfs2_glock_nq_num - acquire a glock based on lock number
1276  * @sdp: the filesystem
1277  * @number: the lock number
1278  * @glops: the glock operations for the type of glock
1279  * @state: the state to acquire the glock in
1280  * @flags: modifier flags for the aquisition
1281  * @gh: the struct gfs2_holder
1282  *
1283  * Returns: errno
1284  */
1285 
1286 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1287 		      const struct gfs2_glock_operations *glops,
1288 		      unsigned int state, int flags, struct gfs2_holder *gh)
1289 {
1290 	struct gfs2_glock *gl;
1291 	int error;
1292 
1293 	error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1294 	if (!error) {
1295 		error = gfs2_glock_nq_init(gl, state, flags, gh);
1296 		gfs2_glock_put(gl);
1297 	}
1298 
1299 	return error;
1300 }
1301 
1302 /**
1303  * glock_compare - Compare two struct gfs2_glock structures for sorting
1304  * @arg_a: the first structure
1305  * @arg_b: the second structure
1306  *
1307  */
1308 
1309 static int glock_compare(const void *arg_a, const void *arg_b)
1310 {
1311 	const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1312 	const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1313 	const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1314 	const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1315 
1316 	if (a->ln_number > b->ln_number)
1317 		return 1;
1318 	if (a->ln_number < b->ln_number)
1319 		return -1;
1320 	BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1321 	return 0;
1322 }
1323 
1324 /**
1325  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1326  * @num_gh: the number of structures
1327  * @ghs: an array of struct gfs2_holder structures
1328  *
1329  * Returns: 0 on success (all glocks acquired),
1330  *          errno on failure (no glocks acquired)
1331  */
1332 
1333 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1334 		     struct gfs2_holder **p)
1335 {
1336 	unsigned int x;
1337 	int error = 0;
1338 
1339 	for (x = 0; x < num_gh; x++)
1340 		p[x] = &ghs[x];
1341 
1342 	sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1343 
1344 	for (x = 0; x < num_gh; x++) {
1345 		p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1346 
1347 		error = gfs2_glock_nq(p[x]);
1348 		if (error) {
1349 			while (x--)
1350 				gfs2_glock_dq(p[x]);
1351 			break;
1352 		}
1353 	}
1354 
1355 	return error;
1356 }
1357 
1358 /**
1359  * gfs2_glock_nq_m - acquire multiple glocks
1360  * @num_gh: the number of structures
1361  * @ghs: an array of struct gfs2_holder structures
1362  *
1363  *
1364  * Returns: 0 on success (all glocks acquired),
1365  *          errno on failure (no glocks acquired)
1366  */
1367 
1368 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1369 {
1370 	struct gfs2_holder *tmp[4];
1371 	struct gfs2_holder **pph = tmp;
1372 	int error = 0;
1373 
1374 	switch(num_gh) {
1375 	case 0:
1376 		return 0;
1377 	case 1:
1378 		ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1379 		return gfs2_glock_nq(ghs);
1380 	default:
1381 		if (num_gh <= 4)
1382 			break;
1383 		pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1384 		if (!pph)
1385 			return -ENOMEM;
1386 	}
1387 
1388 	error = nq_m_sync(num_gh, ghs, pph);
1389 
1390 	if (pph != tmp)
1391 		kfree(pph);
1392 
1393 	return error;
1394 }
1395 
1396 /**
1397  * gfs2_glock_dq_m - release multiple glocks
1398  * @num_gh: the number of structures
1399  * @ghs: an array of struct gfs2_holder structures
1400  *
1401  */
1402 
1403 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1404 {
1405 	unsigned int x;
1406 
1407 	for (x = 0; x < num_gh; x++)
1408 		gfs2_glock_dq(&ghs[x]);
1409 }
1410 
1411 /**
1412  * gfs2_glock_dq_uninit_m - release multiple glocks
1413  * @num_gh: the number of structures
1414  * @ghs: an array of struct gfs2_holder structures
1415  *
1416  */
1417 
1418 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1419 {
1420 	unsigned int x;
1421 
1422 	for (x = 0; x < num_gh; x++)
1423 		gfs2_glock_dq_uninit(&ghs[x]);
1424 }
1425 
1426 /**
1427  * gfs2_lvb_hold - attach a LVB from a glock
1428  * @gl: The glock in question
1429  *
1430  */
1431 
1432 int gfs2_lvb_hold(struct gfs2_glock *gl)
1433 {
1434 	int error;
1435 
1436 	gfs2_glmutex_lock(gl);
1437 
1438 	if (!atomic_read(&gl->gl_lvb_count)) {
1439 		error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1440 		if (error) {
1441 			gfs2_glmutex_unlock(gl);
1442 			return error;
1443 		}
1444 		gfs2_glock_hold(gl);
1445 	}
1446 	atomic_inc(&gl->gl_lvb_count);
1447 
1448 	gfs2_glmutex_unlock(gl);
1449 
1450 	return 0;
1451 }
1452 
1453 /**
1454  * gfs2_lvb_unhold - detach a LVB from a glock
1455  * @gl: The glock in question
1456  *
1457  */
1458 
1459 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1460 {
1461 	gfs2_glock_hold(gl);
1462 	gfs2_glmutex_lock(gl);
1463 
1464 	gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1465 	if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1466 		gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1467 		gl->gl_lvb = NULL;
1468 		gfs2_glock_put(gl);
1469 	}
1470 
1471 	gfs2_glmutex_unlock(gl);
1472 	gfs2_glock_put(gl);
1473 }
1474 
1475 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1476 			unsigned int state)
1477 {
1478 	struct gfs2_glock *gl;
1479 	unsigned long delay = 0;
1480 	unsigned long holdtime;
1481 	unsigned long now = jiffies;
1482 
1483 	gl = gfs2_glock_find(sdp, name);
1484 	if (!gl)
1485 		return;
1486 
1487 	holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1488 	if (time_before(now, holdtime))
1489 		delay = holdtime - now;
1490 
1491 	handle_callback(gl, state, 1, delay);
1492 	if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1493 		gfs2_glock_put(gl);
1494 }
1495 
1496 /**
1497  * gfs2_glock_cb - Callback used by locking module
1498  * @sdp: Pointer to the superblock
1499  * @type: Type of callback
1500  * @data: Type dependent data pointer
1501  *
1502  * Called by the locking module when it wants to tell us something.
1503  * Either we need to drop a lock, one of our ASYNC requests completed, or
1504  * a journal from another client needs to be recovered.
1505  */
1506 
1507 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1508 {
1509 	struct gfs2_sbd *sdp = cb_data;
1510 
1511 	switch (type) {
1512 	case LM_CB_NEED_E:
1513 		blocking_cb(sdp, data, LM_ST_UNLOCKED);
1514 		return;
1515 
1516 	case LM_CB_NEED_D:
1517 		blocking_cb(sdp, data, LM_ST_DEFERRED);
1518 		return;
1519 
1520 	case LM_CB_NEED_S:
1521 		blocking_cb(sdp, data, LM_ST_SHARED);
1522 		return;
1523 
1524 	case LM_CB_ASYNC: {
1525 		struct lm_async_cb *async = data;
1526 		struct gfs2_glock *gl;
1527 
1528 		down_read(&gfs2_umount_flush_sem);
1529 		gl = gfs2_glock_find(sdp, &async->lc_name);
1530 		if (gfs2_assert_warn(sdp, gl))
1531 			return;
1532 		if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1533 			gl->gl_req_bh(gl, async->lc_ret);
1534 		if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1535 			gfs2_glock_put(gl);
1536 		up_read(&gfs2_umount_flush_sem);
1537 		return;
1538 	}
1539 
1540 	case LM_CB_NEED_RECOVERY:
1541 		gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1542 		if (sdp->sd_recoverd_process)
1543 			wake_up_process(sdp->sd_recoverd_process);
1544 		return;
1545 
1546 	case LM_CB_DROPLOCKS:
1547 		gfs2_gl_hash_clear(sdp, NO_WAIT);
1548 		gfs2_quota_scan(sdp);
1549 		return;
1550 
1551 	default:
1552 		gfs2_assert_warn(sdp, 0);
1553 		return;
1554 	}
1555 }
1556 
1557 /**
1558  * demote_ok - Check to see if it's ok to unlock a glock
1559  * @gl: the glock
1560  *
1561  * Returns: 1 if it's ok
1562  */
1563 
1564 static int demote_ok(struct gfs2_glock *gl)
1565 {
1566 	const struct gfs2_glock_operations *glops = gl->gl_ops;
1567 	int demote = 1;
1568 
1569 	if (test_bit(GLF_STICKY, &gl->gl_flags))
1570 		demote = 0;
1571 	else if (glops->go_demote_ok)
1572 		demote = glops->go_demote_ok(gl);
1573 
1574 	return demote;
1575 }
1576 
1577 /**
1578  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1579  * @gl: the glock
1580  *
1581  */
1582 
1583 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1584 {
1585 	struct gfs2_sbd *sdp = gl->gl_sbd;
1586 
1587 	spin_lock(&sdp->sd_reclaim_lock);
1588 	if (list_empty(&gl->gl_reclaim)) {
1589 		gfs2_glock_hold(gl);
1590 		list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1591 		atomic_inc(&sdp->sd_reclaim_count);
1592 	}
1593 	spin_unlock(&sdp->sd_reclaim_lock);
1594 
1595 	wake_up(&sdp->sd_reclaim_wq);
1596 }
1597 
1598 /**
1599  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1600  * @sdp: the filesystem
1601  *
1602  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1603  * different glock and we notice that there are a lot of glocks in the
1604  * reclaim list.
1605  *
1606  */
1607 
1608 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1609 {
1610 	struct gfs2_glock *gl;
1611 
1612 	spin_lock(&sdp->sd_reclaim_lock);
1613 	if (list_empty(&sdp->sd_reclaim_list)) {
1614 		spin_unlock(&sdp->sd_reclaim_lock);
1615 		return;
1616 	}
1617 	gl = list_entry(sdp->sd_reclaim_list.next,
1618 			struct gfs2_glock, gl_reclaim);
1619 	list_del_init(&gl->gl_reclaim);
1620 	spin_unlock(&sdp->sd_reclaim_lock);
1621 
1622 	atomic_dec(&sdp->sd_reclaim_count);
1623 	atomic_inc(&sdp->sd_reclaimed);
1624 
1625 	if (gfs2_glmutex_trylock(gl)) {
1626 		if (list_empty(&gl->gl_holders) &&
1627 		    gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1628 			handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1629 		gfs2_glmutex_unlock(gl);
1630 	}
1631 
1632 	gfs2_glock_put(gl);
1633 }
1634 
1635 /**
1636  * examine_bucket - Call a function for glock in a hash bucket
1637  * @examiner: the function
1638  * @sdp: the filesystem
1639  * @bucket: the bucket
1640  *
1641  * Returns: 1 if the bucket has entries
1642  */
1643 
1644 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1645 			  unsigned int hash)
1646 {
1647 	struct gfs2_glock *gl, *prev = NULL;
1648 	int has_entries = 0;
1649 	struct hlist_head *head = &gl_hash_table[hash].hb_list;
1650 
1651 	read_lock(gl_lock_addr(hash));
1652 	/* Can't use hlist_for_each_entry - don't want prefetch here */
1653 	if (hlist_empty(head))
1654 		goto out;
1655 	gl = list_entry(head->first, struct gfs2_glock, gl_list);
1656 	while(1) {
1657 		if (!sdp || gl->gl_sbd == sdp) {
1658 			gfs2_glock_hold(gl);
1659 			read_unlock(gl_lock_addr(hash));
1660 			if (prev)
1661 				gfs2_glock_put(prev);
1662 			prev = gl;
1663 			examiner(gl);
1664 			has_entries = 1;
1665 			read_lock(gl_lock_addr(hash));
1666 		}
1667 		if (gl->gl_list.next == NULL)
1668 			break;
1669 		gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1670 	}
1671 out:
1672 	read_unlock(gl_lock_addr(hash));
1673 	if (prev)
1674 		gfs2_glock_put(prev);
1675 	cond_resched();
1676 	return has_entries;
1677 }
1678 
1679 /**
1680  * scan_glock - look at a glock and see if we can reclaim it
1681  * @gl: the glock to look at
1682  *
1683  */
1684 
1685 static void scan_glock(struct gfs2_glock *gl)
1686 {
1687 	if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1688 		return;
1689 
1690 	if (gfs2_glmutex_trylock(gl)) {
1691 		if (list_empty(&gl->gl_holders) &&
1692 		    gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1693 			goto out_schedule;
1694 		gfs2_glmutex_unlock(gl);
1695 	}
1696 	return;
1697 
1698 out_schedule:
1699 	gfs2_glmutex_unlock(gl);
1700 	gfs2_glock_schedule_for_reclaim(gl);
1701 }
1702 
1703 /**
1704  * clear_glock - look at a glock and see if we can free it from glock cache
1705  * @gl: the glock to look at
1706  *
1707  */
1708 
1709 static void clear_glock(struct gfs2_glock *gl)
1710 {
1711 	struct gfs2_sbd *sdp = gl->gl_sbd;
1712 	int released;
1713 
1714 	spin_lock(&sdp->sd_reclaim_lock);
1715 	if (!list_empty(&gl->gl_reclaim)) {
1716 		list_del_init(&gl->gl_reclaim);
1717 		atomic_dec(&sdp->sd_reclaim_count);
1718 		spin_unlock(&sdp->sd_reclaim_lock);
1719 		released = gfs2_glock_put(gl);
1720 		gfs2_assert(sdp, !released);
1721 	} else {
1722 		spin_unlock(&sdp->sd_reclaim_lock);
1723 	}
1724 
1725 	if (gfs2_glmutex_trylock(gl)) {
1726 		if (list_empty(&gl->gl_holders) &&
1727 		    gl->gl_state != LM_ST_UNLOCKED)
1728 			handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1729 		gfs2_glmutex_unlock(gl);
1730 	}
1731 }
1732 
1733 /**
1734  * gfs2_gl_hash_clear - Empty out the glock hash table
1735  * @sdp: the filesystem
1736  * @wait: wait until it's all gone
1737  *
1738  * Called when unmounting the filesystem, or when inter-node lock manager
1739  * requests DROPLOCKS because it is running out of capacity.
1740  */
1741 
1742 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1743 {
1744 	unsigned long t;
1745 	unsigned int x;
1746 	int cont;
1747 
1748 	t = jiffies;
1749 
1750 	for (;;) {
1751 		cont = 0;
1752 		for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1753 			if (examine_bucket(clear_glock, sdp, x))
1754 				cont = 1;
1755 		}
1756 
1757 		if (!wait || !cont)
1758 			break;
1759 
1760 		if (time_after_eq(jiffies,
1761 				  t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1762 			fs_warn(sdp, "Unmount seems to be stalled. "
1763 				     "Dumping lock state...\n");
1764 			gfs2_dump_lockstate(sdp);
1765 			t = jiffies;
1766 		}
1767 
1768 		down_write(&gfs2_umount_flush_sem);
1769 		invalidate_inodes(sdp->sd_vfs);
1770 		up_write(&gfs2_umount_flush_sem);
1771 		msleep(10);
1772 	}
1773 }
1774 
1775 /*
1776  *  Diagnostic routines to help debug distributed deadlock
1777  */
1778 
1779 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1780                               unsigned long address)
1781 {
1782 	char buffer[KSYM_SYMBOL_LEN];
1783 
1784 	sprint_symbol(buffer, address);
1785 	print_dbg(gi, fmt, buffer);
1786 }
1787 
1788 /**
1789  * dump_holder - print information about a glock holder
1790  * @str: a string naming the type of holder
1791  * @gh: the glock holder
1792  *
1793  * Returns: 0 on success, -ENOBUFS when we run out of space
1794  */
1795 
1796 static int dump_holder(struct glock_iter *gi, char *str,
1797 		       struct gfs2_holder *gh)
1798 {
1799 	unsigned int x;
1800 	struct task_struct *gh_owner;
1801 
1802 	print_dbg(gi, "  %s\n", str);
1803 	if (gh->gh_owner_pid) {
1804 		print_dbg(gi, "    owner = %ld ",
1805 				(long)pid_nr(gh->gh_owner_pid));
1806 		gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1807 		if (gh_owner)
1808 			print_dbg(gi, "(%s)\n", gh_owner->comm);
1809 		else
1810 			print_dbg(gi, "(ended)\n");
1811 	} else
1812 		print_dbg(gi, "    owner = -1\n");
1813 	print_dbg(gi, "    gh_state = %u\n", gh->gh_state);
1814 	print_dbg(gi, "    gh_flags =");
1815 	for (x = 0; x < 32; x++)
1816 		if (gh->gh_flags & (1 << x))
1817 			print_dbg(gi, " %u", x);
1818 	print_dbg(gi, " \n");
1819 	print_dbg(gi, "    error = %d\n", gh->gh_error);
1820 	print_dbg(gi, "    gh_iflags =");
1821 	for (x = 0; x < 32; x++)
1822 		if (test_bit(x, &gh->gh_iflags))
1823 			print_dbg(gi, " %u", x);
1824 	print_dbg(gi, " \n");
1825         gfs2_print_symbol(gi, "    initialized at: %s\n", gh->gh_ip);
1826 
1827 	return 0;
1828 }
1829 
1830 /**
1831  * dump_inode - print information about an inode
1832  * @ip: the inode
1833  *
1834  * Returns: 0 on success, -ENOBUFS when we run out of space
1835  */
1836 
1837 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1838 {
1839 	unsigned int x;
1840 
1841 	print_dbg(gi, "  Inode:\n");
1842 	print_dbg(gi, "    num = %llu/%llu\n",
1843 		  (unsigned long long)ip->i_no_formal_ino,
1844 		  (unsigned long long)ip->i_no_addr);
1845 	print_dbg(gi, "    type = %u\n", IF2DT(ip->i_inode.i_mode));
1846 	print_dbg(gi, "    i_flags =");
1847 	for (x = 0; x < 32; x++)
1848 		if (test_bit(x, &ip->i_flags))
1849 			print_dbg(gi, " %u", x);
1850 	print_dbg(gi, " \n");
1851 	return 0;
1852 }
1853 
1854 /**
1855  * dump_glock - print information about a glock
1856  * @gl: the glock
1857  * @count: where we are in the buffer
1858  *
1859  * Returns: 0 on success, -ENOBUFS when we run out of space
1860  */
1861 
1862 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1863 {
1864 	struct gfs2_holder *gh;
1865 	unsigned int x;
1866 	int error = -ENOBUFS;
1867 	struct task_struct *gl_owner;
1868 
1869 	spin_lock(&gl->gl_spin);
1870 
1871 	print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
1872 		   (unsigned long long)gl->gl_name.ln_number);
1873 	print_dbg(gi, "  gl_flags =");
1874 	for (x = 0; x < 32; x++) {
1875 		if (test_bit(x, &gl->gl_flags))
1876 			print_dbg(gi, " %u", x);
1877 	}
1878 	if (!test_bit(GLF_LOCK, &gl->gl_flags))
1879 		print_dbg(gi, " (unlocked)");
1880 	print_dbg(gi, " \n");
1881 	print_dbg(gi, "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
1882 	print_dbg(gi, "  gl_state = %u\n", gl->gl_state);
1883 	if (gl->gl_owner_pid) {
1884 		gl_owner = find_task_by_pid(gl->gl_owner_pid);
1885 		if (gl_owner)
1886 			print_dbg(gi, "  gl_owner = pid %d (%s)\n",
1887 				  gl->gl_owner_pid, gl_owner->comm);
1888 		else
1889 			print_dbg(gi, "  gl_owner = %d (ended)\n",
1890 				  gl->gl_owner_pid);
1891 	} else
1892 		print_dbg(gi, "  gl_owner = -1\n");
1893 	print_dbg(gi, "  gl_ip = %lu\n", gl->gl_ip);
1894 	print_dbg(gi, "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1895 	print_dbg(gi, "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1896 	print_dbg(gi, "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1897 	print_dbg(gi, "  object = %s\n", (gl->gl_object) ? "yes" : "no");
1898 	print_dbg(gi, "  reclaim = %s\n",
1899 		   (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1900 	if (gl->gl_aspace)
1901 		print_dbg(gi, "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1902 			   gl->gl_aspace->i_mapping->nrpages);
1903 	else
1904 		print_dbg(gi, "  aspace = no\n");
1905 	print_dbg(gi, "  ail = %d\n", atomic_read(&gl->gl_ail_count));
1906 	if (gl->gl_req_gh) {
1907 		error = dump_holder(gi, "Request", gl->gl_req_gh);
1908 		if (error)
1909 			goto out;
1910 	}
1911 	list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1912 		error = dump_holder(gi, "Holder", gh);
1913 		if (error)
1914 			goto out;
1915 	}
1916 	list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1917 		error = dump_holder(gi, "Waiter1", gh);
1918 		if (error)
1919 			goto out;
1920 	}
1921 	list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1922 		error = dump_holder(gi, "Waiter3", gh);
1923 		if (error)
1924 			goto out;
1925 	}
1926 	if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1927 		print_dbg(gi, "  Demotion req to state %u (%llu uS ago)\n",
1928 			  gl->gl_demote_state, (unsigned long long)
1929 			  (jiffies - gl->gl_demote_time)*(1000000/HZ));
1930 	}
1931 	if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1932 		if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1933 			list_empty(&gl->gl_holders)) {
1934 			error = dump_inode(gi, gl->gl_object);
1935 			if (error)
1936 				goto out;
1937 		} else {
1938 			error = -ENOBUFS;
1939 			print_dbg(gi, "  Inode: busy\n");
1940 		}
1941 	}
1942 
1943 	error = 0;
1944 
1945 out:
1946 	spin_unlock(&gl->gl_spin);
1947 	return error;
1948 }
1949 
1950 /**
1951  * gfs2_dump_lockstate - print out the current lockstate
1952  * @sdp: the filesystem
1953  * @ub: the buffer to copy the information into
1954  *
1955  * If @ub is NULL, dump the lockstate to the console.
1956  *
1957  */
1958 
1959 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1960 {
1961 	struct gfs2_glock *gl;
1962 	struct hlist_node *h;
1963 	unsigned int x;
1964 	int error = 0;
1965 
1966 	for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1967 
1968 		read_lock(gl_lock_addr(x));
1969 
1970 		hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1971 			if (gl->gl_sbd != sdp)
1972 				continue;
1973 
1974 			error = dump_glock(NULL, gl);
1975 			if (error)
1976 				break;
1977 		}
1978 
1979 		read_unlock(gl_lock_addr(x));
1980 
1981 		if (error)
1982 			break;
1983 	}
1984 
1985 
1986 	return error;
1987 }
1988 
1989 /**
1990  * gfs2_scand - Look for cached glocks and inodes to toss from memory
1991  * @sdp: Pointer to GFS2 superblock
1992  *
1993  * One of these daemons runs, finding candidates to add to sd_reclaim_list.
1994  * See gfs2_glockd()
1995  */
1996 
1997 static int gfs2_scand(void *data)
1998 {
1999 	unsigned x;
2000 	unsigned delay;
2001 
2002 	while (!kthread_should_stop()) {
2003 		for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2004 			examine_bucket(scan_glock, NULL, x);
2005 		if (freezing(current))
2006 			refrigerator();
2007 		delay = scand_secs;
2008 		if (delay < 1)
2009 			delay = 1;
2010 		schedule_timeout_interruptible(delay * HZ);
2011 	}
2012 
2013 	return 0;
2014 }
2015 
2016 
2017 
2018 int __init gfs2_glock_init(void)
2019 {
2020 	unsigned i;
2021 	for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2022 		INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2023 	}
2024 #ifdef GL_HASH_LOCK_SZ
2025 	for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2026 		rwlock_init(&gl_hash_locks[i]);
2027 	}
2028 #endif
2029 
2030 	scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
2031 	if (IS_ERR(scand_process))
2032 		return PTR_ERR(scand_process);
2033 
2034 	glock_workqueue = create_workqueue("glock_workqueue");
2035 	if (IS_ERR(glock_workqueue)) {
2036 		kthread_stop(scand_process);
2037 		return PTR_ERR(glock_workqueue);
2038 	}
2039 
2040 	return 0;
2041 }
2042 
2043 void gfs2_glock_exit(void)
2044 {
2045 	destroy_workqueue(glock_workqueue);
2046 	kthread_stop(scand_process);
2047 }
2048 
2049 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
2050 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
2051 
2052 static int gfs2_glock_iter_next(struct glock_iter *gi)
2053 {
2054 	struct gfs2_glock *gl;
2055 
2056 restart:
2057 	read_lock(gl_lock_addr(gi->hash));
2058 	gl = gi->gl;
2059 	if (gl) {
2060 		gi->gl = hlist_entry(gl->gl_list.next,
2061 				     struct gfs2_glock, gl_list);
2062 		if (gi->gl)
2063 			gfs2_glock_hold(gi->gl);
2064 	}
2065 	read_unlock(gl_lock_addr(gi->hash));
2066 	if (gl)
2067 		gfs2_glock_put(gl);
2068 	if (gl && gi->gl == NULL)
2069 		gi->hash++;
2070 	while(gi->gl == NULL) {
2071 		if (gi->hash >= GFS2_GL_HASH_SIZE)
2072 			return 1;
2073 		read_lock(gl_lock_addr(gi->hash));
2074 		gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
2075 				     struct gfs2_glock, gl_list);
2076 		if (gi->gl)
2077 			gfs2_glock_hold(gi->gl);
2078 		read_unlock(gl_lock_addr(gi->hash));
2079 		gi->hash++;
2080 	}
2081 
2082 	if (gi->sdp != gi->gl->gl_sbd)
2083 		goto restart;
2084 
2085 	return 0;
2086 }
2087 
2088 static void gfs2_glock_iter_free(struct glock_iter *gi)
2089 {
2090 	if (gi->gl)
2091 		gfs2_glock_put(gi->gl);
2092 	kfree(gi);
2093 }
2094 
2095 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2096 {
2097 	struct glock_iter *gi;
2098 
2099 	gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2100 	if (!gi)
2101 		return NULL;
2102 
2103 	gi->sdp = sdp;
2104 	gi->hash = 0;
2105 	gi->seq = NULL;
2106 	gi->gl = NULL;
2107 	memset(gi->string, 0, sizeof(gi->string));
2108 
2109 	if (gfs2_glock_iter_next(gi)) {
2110 		gfs2_glock_iter_free(gi);
2111 		return NULL;
2112 	}
2113 
2114 	return gi;
2115 }
2116 
2117 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2118 {
2119 	struct glock_iter *gi;
2120 	loff_t n = *pos;
2121 
2122 	gi = gfs2_glock_iter_init(file->private);
2123 	if (!gi)
2124 		return NULL;
2125 
2126 	while(n--) {
2127 		if (gfs2_glock_iter_next(gi)) {
2128 			gfs2_glock_iter_free(gi);
2129 			return NULL;
2130 		}
2131 	}
2132 
2133 	return gi;
2134 }
2135 
2136 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2137 				 loff_t *pos)
2138 {
2139 	struct glock_iter *gi = iter_ptr;
2140 
2141 	(*pos)++;
2142 
2143 	if (gfs2_glock_iter_next(gi)) {
2144 		gfs2_glock_iter_free(gi);
2145 		return NULL;
2146 	}
2147 
2148 	return gi;
2149 }
2150 
2151 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2152 {
2153 	struct glock_iter *gi = iter_ptr;
2154 	if (gi)
2155 		gfs2_glock_iter_free(gi);
2156 }
2157 
2158 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2159 {
2160 	struct glock_iter *gi = iter_ptr;
2161 
2162 	gi->seq = file;
2163 	dump_glock(gi, gi->gl);
2164 
2165 	return 0;
2166 }
2167 
2168 static const struct seq_operations gfs2_glock_seq_ops = {
2169 	.start = gfs2_glock_seq_start,
2170 	.next  = gfs2_glock_seq_next,
2171 	.stop  = gfs2_glock_seq_stop,
2172 	.show  = gfs2_glock_seq_show,
2173 };
2174 
2175 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2176 {
2177 	struct seq_file *seq;
2178 	int ret;
2179 
2180 	ret = seq_open(file, &gfs2_glock_seq_ops);
2181 	if (ret)
2182 		return ret;
2183 
2184 	seq = file->private_data;
2185 	seq->private = inode->i_private;
2186 
2187 	return 0;
2188 }
2189 
2190 static const struct file_operations gfs2_debug_fops = {
2191 	.owner   = THIS_MODULE,
2192 	.open    = gfs2_debugfs_open,
2193 	.read    = seq_read,
2194 	.llseek  = seq_lseek,
2195 	.release = seq_release
2196 };
2197 
2198 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2199 {
2200 	sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2201 	if (!sdp->debugfs_dir)
2202 		return -ENOMEM;
2203 	sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2204 							 S_IFREG | S_IRUGO,
2205 							 sdp->debugfs_dir, sdp,
2206 							 &gfs2_debug_fops);
2207 	if (!sdp->debugfs_dentry_glocks)
2208 		return -ENOMEM;
2209 
2210 	return 0;
2211 }
2212 
2213 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2214 {
2215 	if (sdp && sdp->debugfs_dir) {
2216 		if (sdp->debugfs_dentry_glocks) {
2217 			debugfs_remove(sdp->debugfs_dentry_glocks);
2218 			sdp->debugfs_dentry_glocks = NULL;
2219 		}
2220 		debugfs_remove(sdp->debugfs_dir);
2221 		sdp->debugfs_dir = NULL;
2222 	}
2223 }
2224 
2225 int gfs2_register_debugfs(void)
2226 {
2227 	gfs2_root = debugfs_create_dir("gfs2", NULL);
2228 	return gfs2_root ? 0 : -ENOMEM;
2229 }
2230 
2231 void gfs2_unregister_debugfs(void)
2232 {
2233 	debugfs_remove(gfs2_root);
2234 	gfs2_root = NULL;
2235 }
2236