[GFS2] panic after can't parse mount arguments
[linux-2.6] / fs / gfs2 / glock.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 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
31 #include "gfs2.h"
32 #include "incore.h"
33 #include "glock.h"
34 #include "glops.h"
35 #include "inode.h"
36 #include "lm.h"
37 #include "lops.h"
38 #include "meta_io.h"
39 #include "quota.h"
40 #include "super.h"
41 #include "util.h"
42
43 struct gfs2_gl_hash_bucket {
44         struct hlist_head hb_list;
45 };
46
47 struct glock_iter {
48         int hash;                     /* hash bucket index         */
49         struct gfs2_sbd *sdp;         /* incore superblock         */
50         struct gfs2_glock *gl;        /* current glock struct      */
51         struct seq_file *seq;         /* sequence file for debugfs */
52         char string[512];             /* scratch space             */
53 };
54
55 typedef void (*glock_examiner) (struct gfs2_glock * gl);
56
57 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
58 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl);
59 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
60 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
61 static DECLARE_RWSEM(gfs2_umount_flush_sem);
62 static struct dentry *gfs2_root;
63 static struct task_struct *scand_process;
64 static unsigned int scand_secs = 5;
65
66 #define GFS2_GL_HASH_SHIFT      15
67 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
68 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
69
70 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
71 static struct dentry *gfs2_root;
72
73 /*
74  * Despite what you might think, the numbers below are not arbitrary :-)
75  * They are taken from the ipv4 routing hash code, which is well tested
76  * and thus should be nearly optimal. Later on we might tweek the numbers
77  * but for now this should be fine.
78  *
79  * The reason for putting the locks in a separate array from the list heads
80  * is that we can have fewer locks than list heads and save memory. We use
81  * the same hash function for both, but with a different hash mask.
82  */
83 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
84         defined(CONFIG_PROVE_LOCKING)
85
86 #ifdef CONFIG_LOCKDEP
87 # define GL_HASH_LOCK_SZ        256
88 #else
89 # if NR_CPUS >= 32
90 #  define GL_HASH_LOCK_SZ       4096
91 # elif NR_CPUS >= 16
92 #  define GL_HASH_LOCK_SZ       2048
93 # elif NR_CPUS >= 8
94 #  define GL_HASH_LOCK_SZ       1024
95 # elif NR_CPUS >= 4
96 #  define GL_HASH_LOCK_SZ       512
97 # else
98 #  define GL_HASH_LOCK_SZ       256
99 # endif
100 #endif
101
102 /* We never want more locks than chains */
103 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
104 # undef GL_HASH_LOCK_SZ
105 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
106 #endif
107
108 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
109
110 static inline rwlock_t *gl_lock_addr(unsigned int x)
111 {
112         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
113 }
114 #else /* not SMP, so no spinlocks required */
115 static inline rwlock_t *gl_lock_addr(unsigned int x)
116 {
117         return NULL;
118 }
119 #endif
120
121 /**
122  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
123  * @actual: the current state of the lock
124  * @requested: the lock state that was requested by the caller
125  * @flags: the modifier flags passed in by the caller
126  *
127  * Returns: 1 if the locks are compatible, 0 otherwise
128  */
129
130 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
131                                    int flags)
132 {
133         if (actual == requested)
134                 return 1;
135
136         if (flags & GL_EXACT)
137                 return 0;
138
139         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
140                 return 1;
141
142         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
143                 return 1;
144
145         return 0;
146 }
147
148 /**
149  * gl_hash() - Turn glock number into hash bucket number
150  * @lock: The glock number
151  *
152  * Returns: The number of the corresponding hash bucket
153  */
154
155 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
156                             const struct lm_lockname *name)
157 {
158         unsigned int h;
159
160         h = jhash(&name->ln_number, sizeof(u64), 0);
161         h = jhash(&name->ln_type, sizeof(unsigned int), h);
162         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
163         h &= GFS2_GL_HASH_MASK;
164
165         return h;
166 }
167
168 /**
169  * glock_free() - Perform a few checks and then release struct gfs2_glock
170  * @gl: The glock to release
171  *
172  * Also calls lock module to release its internal structure for this glock.
173  *
174  */
175
176 static void glock_free(struct gfs2_glock *gl)
177 {
178         struct gfs2_sbd *sdp = gl->gl_sbd;
179         struct inode *aspace = gl->gl_aspace;
180
181         gfs2_lm_put_lock(sdp, gl->gl_lock);
182
183         if (aspace)
184                 gfs2_aspace_put(aspace);
185
186         kmem_cache_free(gfs2_glock_cachep, gl);
187 }
188
189 /**
190  * gfs2_glock_hold() - increment reference count on glock
191  * @gl: The glock to hold
192  *
193  */
194
195 void gfs2_glock_hold(struct gfs2_glock *gl)
196 {
197         atomic_inc(&gl->gl_ref);
198 }
199
200 /**
201  * gfs2_glock_put() - Decrement reference count on glock
202  * @gl: The glock to put
203  *
204  */
205
206 int gfs2_glock_put(struct gfs2_glock *gl)
207 {
208         int rv = 0;
209         struct gfs2_sbd *sdp = gl->gl_sbd;
210
211         write_lock(gl_lock_addr(gl->gl_hash));
212         if (atomic_dec_and_test(&gl->gl_ref)) {
213                 hlist_del(&gl->gl_list);
214                 write_unlock(gl_lock_addr(gl->gl_hash));
215                 BUG_ON(spin_is_locked(&gl->gl_spin));
216                 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
217                 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
218                 gfs2_assert(sdp, list_empty(&gl->gl_holders));
219                 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
220                 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
221                 glock_free(gl);
222                 rv = 1;
223                 goto out;
224         }
225         write_unlock(gl_lock_addr(gl->gl_hash));
226 out:
227         return rv;
228 }
229
230 /**
231  * search_bucket() - Find struct gfs2_glock by lock number
232  * @bucket: the bucket to search
233  * @name: The lock name
234  *
235  * Returns: NULL, or the struct gfs2_glock with the requested number
236  */
237
238 static struct gfs2_glock *search_bucket(unsigned int hash,
239                                         const struct gfs2_sbd *sdp,
240                                         const struct lm_lockname *name)
241 {
242         struct gfs2_glock *gl;
243         struct hlist_node *h;
244
245         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
246                 if (!lm_name_equal(&gl->gl_name, name))
247                         continue;
248                 if (gl->gl_sbd != sdp)
249                         continue;
250
251                 atomic_inc(&gl->gl_ref);
252
253                 return gl;
254         }
255
256         return NULL;
257 }
258
259 /**
260  * gfs2_glock_find() - Find glock by lock number
261  * @sdp: The GFS2 superblock
262  * @name: The lock name
263  *
264  * Returns: NULL, or the struct gfs2_glock with the requested number
265  */
266
267 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
268                                           const struct lm_lockname *name)
269 {
270         unsigned int hash = gl_hash(sdp, name);
271         struct gfs2_glock *gl;
272
273         read_lock(gl_lock_addr(hash));
274         gl = search_bucket(hash, sdp, name);
275         read_unlock(gl_lock_addr(hash));
276
277         return gl;
278 }
279
280 /**
281  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
282  * @sdp: The GFS2 superblock
283  * @number: the lock number
284  * @glops: The glock_operations to use
285  * @create: If 0, don't create the glock if it doesn't exist
286  * @glp: the glock is returned here
287  *
288  * This does not lock a glock, just finds/creates structures for one.
289  *
290  * Returns: errno
291  */
292
293 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
294                    const struct gfs2_glock_operations *glops, int create,
295                    struct gfs2_glock **glp)
296 {
297         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
298         struct gfs2_glock *gl, *tmp;
299         unsigned int hash = gl_hash(sdp, &name);
300         int error;
301
302         read_lock(gl_lock_addr(hash));
303         gl = search_bucket(hash, sdp, &name);
304         read_unlock(gl_lock_addr(hash));
305
306         if (gl || !create) {
307                 *glp = gl;
308                 return 0;
309         }
310
311         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
312         if (!gl)
313                 return -ENOMEM;
314
315         gl->gl_flags = 0;
316         gl->gl_name = name;
317         atomic_set(&gl->gl_ref, 1);
318         gl->gl_state = LM_ST_UNLOCKED;
319         gl->gl_hash = hash;
320         gl->gl_owner_pid = 0;
321         gl->gl_ip = 0;
322         gl->gl_ops = glops;
323         gl->gl_req_gh = NULL;
324         gl->gl_req_bh = NULL;
325         gl->gl_vn = 0;
326         gl->gl_stamp = jiffies;
327         gl->gl_object = NULL;
328         gl->gl_sbd = sdp;
329         gl->gl_aspace = NULL;
330         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
331
332         /* If this glock protects actual on-disk data or metadata blocks,
333            create a VFS inode to manage the pages/buffers holding them. */
334         if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
335                 gl->gl_aspace = gfs2_aspace_get(sdp);
336                 if (!gl->gl_aspace) {
337                         error = -ENOMEM;
338                         goto fail;
339                 }
340         }
341
342         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
343         if (error)
344                 goto fail_aspace;
345
346         write_lock(gl_lock_addr(hash));
347         tmp = search_bucket(hash, sdp, &name);
348         if (tmp) {
349                 write_unlock(gl_lock_addr(hash));
350                 glock_free(gl);
351                 gl = tmp;
352         } else {
353                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
354                 write_unlock(gl_lock_addr(hash));
355         }
356
357         *glp = gl;
358
359         return 0;
360
361 fail_aspace:
362         if (gl->gl_aspace)
363                 gfs2_aspace_put(gl->gl_aspace);
364 fail:
365         kmem_cache_free(gfs2_glock_cachep, gl);
366         return error;
367 }
368
369 /**
370  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
371  * @gl: the glock
372  * @state: the state we're requesting
373  * @flags: the modifier flags
374  * @gh: the holder structure
375  *
376  */
377
378 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
379                       struct gfs2_holder *gh)
380 {
381         INIT_LIST_HEAD(&gh->gh_list);
382         gh->gh_gl = gl;
383         gh->gh_ip = (unsigned long)__builtin_return_address(0);
384         gh->gh_owner_pid = current->pid;
385         gh->gh_state = state;
386         gh->gh_flags = flags;
387         gh->gh_error = 0;
388         gh->gh_iflags = 0;
389         gfs2_glock_hold(gl);
390 }
391
392 /**
393  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
394  * @state: the state we're requesting
395  * @flags: the modifier flags
396  * @gh: the holder structure
397  *
398  * Don't mess with the glock.
399  *
400  */
401
402 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
403 {
404         gh->gh_state = state;
405         gh->gh_flags = flags;
406         gh->gh_iflags = 0;
407         gh->gh_ip = (unsigned long)__builtin_return_address(0);
408 }
409
410 /**
411  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
412  * @gh: the holder structure
413  *
414  */
415
416 void gfs2_holder_uninit(struct gfs2_holder *gh)
417 {
418         gfs2_glock_put(gh->gh_gl);
419         gh->gh_gl = NULL;
420         gh->gh_ip = 0;
421 }
422
423 static void gfs2_holder_wake(struct gfs2_holder *gh)
424 {
425         clear_bit(HIF_WAIT, &gh->gh_iflags);
426         smp_mb__after_clear_bit();
427         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
428 }
429
430 static int just_schedule(void *word)
431 {
432         schedule();
433         return 0;
434 }
435
436 static void wait_on_holder(struct gfs2_holder *gh)
437 {
438         might_sleep();
439         wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
440 }
441
442 static void gfs2_demote_wake(struct gfs2_glock *gl)
443 {
444         clear_bit(GLF_DEMOTE, &gl->gl_flags);
445         smp_mb__after_clear_bit();
446         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
447 }
448
449 static void wait_on_demote(struct gfs2_glock *gl)
450 {
451         might_sleep();
452         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
453 }
454
455 /**
456  * rq_mutex - process a mutex request in the queue
457  * @gh: the glock holder
458  *
459  * Returns: 1 if the queue is blocked
460  */
461
462 static int rq_mutex(struct gfs2_holder *gh)
463 {
464         struct gfs2_glock *gl = gh->gh_gl;
465
466         list_del_init(&gh->gh_list);
467         /*  gh->gh_error never examined.  */
468         set_bit(GLF_LOCK, &gl->gl_flags);
469         clear_bit(HIF_WAIT, &gh->gh_iflags);
470         smp_mb();
471         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
472
473         return 1;
474 }
475
476 /**
477  * rq_promote - process a promote request in the queue
478  * @gh: the glock holder
479  *
480  * Acquire a new inter-node lock, or change a lock state to more restrictive.
481  *
482  * Returns: 1 if the queue is blocked
483  */
484
485 static int rq_promote(struct gfs2_holder *gh)
486 {
487         struct gfs2_glock *gl = gh->gh_gl;
488         struct gfs2_sbd *sdp = gl->gl_sbd;
489
490         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
491                 if (list_empty(&gl->gl_holders)) {
492                         gl->gl_req_gh = gh;
493                         set_bit(GLF_LOCK, &gl->gl_flags);
494                         spin_unlock(&gl->gl_spin);
495
496                         if (atomic_read(&sdp->sd_reclaim_count) >
497                             gfs2_tune_get(sdp, gt_reclaim_limit) &&
498                             !(gh->gh_flags & LM_FLAG_PRIORITY)) {
499                                 gfs2_reclaim_glock(sdp);
500                                 gfs2_reclaim_glock(sdp);
501                         }
502
503                         gfs2_glock_xmote_th(gh->gh_gl, gh);
504                         spin_lock(&gl->gl_spin);
505                 }
506                 return 1;
507         }
508
509         if (list_empty(&gl->gl_holders)) {
510                 set_bit(HIF_FIRST, &gh->gh_iflags);
511                 set_bit(GLF_LOCK, &gl->gl_flags);
512         } else {
513                 struct gfs2_holder *next_gh;
514                 if (gh->gh_state == LM_ST_EXCLUSIVE)
515                         return 1;
516                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
517                                      gh_list);
518                 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
519                          return 1;
520         }
521
522         list_move_tail(&gh->gh_list, &gl->gl_holders);
523         gh->gh_error = 0;
524         set_bit(HIF_HOLDER, &gh->gh_iflags);
525
526         gfs2_holder_wake(gh);
527
528         return 0;
529 }
530
531 /**
532  * rq_demote - process a demote request in the queue
533  * @gh: the glock holder
534  *
535  * Returns: 1 if the queue is blocked
536  */
537
538 static int rq_demote(struct gfs2_glock *gl)
539 {
540         if (!list_empty(&gl->gl_holders))
541                 return 1;
542
543         if (gl->gl_state == gl->gl_demote_state ||
544             gl->gl_state == LM_ST_UNLOCKED) {
545                 gfs2_demote_wake(gl);
546                 return 0;
547         }
548         set_bit(GLF_LOCK, &gl->gl_flags);
549         if (gl->gl_demote_state == LM_ST_UNLOCKED ||
550             gl->gl_state != LM_ST_EXCLUSIVE) {
551                 spin_unlock(&gl->gl_spin);
552                 gfs2_glock_drop_th(gl);
553         } else {
554                 spin_unlock(&gl->gl_spin);
555                 gfs2_glock_xmote_th(gl, NULL);
556         }
557         spin_lock(&gl->gl_spin);
558
559         return 0;
560 }
561
562 /**
563  * run_queue - process holder structures on a glock
564  * @gl: the glock
565  *
566  */
567 static void run_queue(struct gfs2_glock *gl)
568 {
569         struct gfs2_holder *gh;
570         int blocked = 1;
571
572         for (;;) {
573                 if (test_bit(GLF_LOCK, &gl->gl_flags))
574                         break;
575
576                 if (!list_empty(&gl->gl_waiters1)) {
577                         gh = list_entry(gl->gl_waiters1.next,
578                                         struct gfs2_holder, gh_list);
579
580                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
581                                 blocked = rq_mutex(gh);
582                         else
583                                 gfs2_assert_warn(gl->gl_sbd, 0);
584
585                 } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
586                         blocked = rq_demote(gl);
587                 } else if (!list_empty(&gl->gl_waiters3)) {
588                         gh = list_entry(gl->gl_waiters3.next,
589                                         struct gfs2_holder, gh_list);
590
591                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
592                                 blocked = rq_promote(gh);
593                         else
594                                 gfs2_assert_warn(gl->gl_sbd, 0);
595
596                 } else
597                         break;
598
599                 if (blocked)
600                         break;
601         }
602 }
603
604 /**
605  * gfs2_glmutex_lock - acquire a local lock on a glock
606  * @gl: the glock
607  *
608  * Gives caller exclusive access to manipulate a glock structure.
609  */
610
611 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
612 {
613         struct gfs2_holder gh;
614
615         gfs2_holder_init(gl, 0, 0, &gh);
616         set_bit(HIF_MUTEX, &gh.gh_iflags);
617         if (test_and_set_bit(HIF_WAIT, &gh.gh_iflags))
618                 BUG();
619
620         spin_lock(&gl->gl_spin);
621         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
622                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
623         } else {
624                 gl->gl_owner_pid = current->pid;
625                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
626                 clear_bit(HIF_WAIT, &gh.gh_iflags);
627                 smp_mb();
628                 wake_up_bit(&gh.gh_iflags, HIF_WAIT);
629         }
630         spin_unlock(&gl->gl_spin);
631
632         wait_on_holder(&gh);
633         gfs2_holder_uninit(&gh);
634 }
635
636 /**
637  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
638  * @gl: the glock
639  *
640  * Returns: 1 if the glock is acquired
641  */
642
643 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
644 {
645         int acquired = 1;
646
647         spin_lock(&gl->gl_spin);
648         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
649                 acquired = 0;
650         } else {
651                 gl->gl_owner_pid = current->pid;
652                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
653         }
654         spin_unlock(&gl->gl_spin);
655
656         return acquired;
657 }
658
659 /**
660  * gfs2_glmutex_unlock - release a local lock on a glock
661  * @gl: the glock
662  *
663  */
664
665 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
666 {
667         spin_lock(&gl->gl_spin);
668         clear_bit(GLF_LOCK, &gl->gl_flags);
669         gl->gl_owner_pid = 0;
670         gl->gl_ip = 0;
671         run_queue(gl);
672         BUG_ON(!spin_is_locked(&gl->gl_spin));
673         spin_unlock(&gl->gl_spin);
674 }
675
676 /**
677  * handle_callback - process a demote request
678  * @gl: the glock
679  * @state: the state the caller wants us to change to
680  *
681  * There are only two requests that we are going to see in actual
682  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
683  */
684
685 static void handle_callback(struct gfs2_glock *gl, unsigned int state, int remote)
686 {
687         spin_lock(&gl->gl_spin);
688         if (test_and_set_bit(GLF_DEMOTE, &gl->gl_flags) == 0) {
689                 gl->gl_demote_state = state;
690                 gl->gl_demote_time = jiffies;
691                 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
692                     gl->gl_object) {
693                         struct inode *inode = igrab(gl->gl_object);
694                         spin_unlock(&gl->gl_spin);
695                         if (inode) {
696                                 d_prune_aliases(inode);
697                                 iput(inode);
698                         }
699                         return;
700                 }
701         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
702                         gl->gl_demote_state != state) {
703                 gl->gl_demote_state = LM_ST_UNLOCKED;
704         }
705         spin_unlock(&gl->gl_spin);
706 }
707
708 /**
709  * state_change - record that the glock is now in a different state
710  * @gl: the glock
711  * @new_state the new state
712  *
713  */
714
715 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
716 {
717         int held1, held2;
718
719         held1 = (gl->gl_state != LM_ST_UNLOCKED);
720         held2 = (new_state != LM_ST_UNLOCKED);
721
722         if (held1 != held2) {
723                 if (held2)
724                         gfs2_glock_hold(gl);
725                 else
726                         gfs2_glock_put(gl);
727         }
728
729         gl->gl_state = new_state;
730 }
731
732 /**
733  * xmote_bh - Called after the lock module is done acquiring a lock
734  * @gl: The glock in question
735  * @ret: the int returned from the lock module
736  *
737  */
738
739 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
740 {
741         struct gfs2_sbd *sdp = gl->gl_sbd;
742         const struct gfs2_glock_operations *glops = gl->gl_ops;
743         struct gfs2_holder *gh = gl->gl_req_gh;
744         int prev_state = gl->gl_state;
745         int op_done = 1;
746
747         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
748         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
749         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
750
751         state_change(gl, ret & LM_OUT_ST_MASK);
752
753         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
754                 if (glops->go_inval)
755                         glops->go_inval(gl, DIO_METADATA);
756         } else if (gl->gl_state == LM_ST_DEFERRED) {
757                 /* We might not want to do this here.
758                    Look at moving to the inode glops. */
759                 if (glops->go_inval)
760                         glops->go_inval(gl, 0);
761         }
762
763         /*  Deal with each possible exit condition  */
764
765         if (!gh) {
766                 gl->gl_stamp = jiffies;
767                 if (ret & LM_OUT_CANCELED) {
768                         op_done = 0;
769                 } else {
770                         spin_lock(&gl->gl_spin);
771                         if (gl->gl_state != gl->gl_demote_state) {
772                                 gl->gl_req_bh = NULL;
773                                 spin_unlock(&gl->gl_spin);
774                                 gfs2_glock_drop_th(gl);
775                                 gfs2_glock_put(gl);
776                                 return;
777                         }
778                         gfs2_demote_wake(gl);
779                         spin_unlock(&gl->gl_spin);
780                 }
781         } else {
782                 spin_lock(&gl->gl_spin);
783                 list_del_init(&gh->gh_list);
784                 gh->gh_error = -EIO;
785                 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 
786                         goto out;
787                 gh->gh_error = GLR_CANCELED;
788                 if (ret & LM_OUT_CANCELED) 
789                         goto out;
790                 if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
791                         list_add_tail(&gh->gh_list, &gl->gl_holders);
792                         gh->gh_error = 0;
793                         set_bit(HIF_HOLDER, &gh->gh_iflags);
794                         set_bit(HIF_FIRST, &gh->gh_iflags);
795                         op_done = 0;
796                         goto out;
797                 }
798                 gh->gh_error = GLR_TRYFAILED;
799                 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
800                         goto out;
801                 gh->gh_error = -EINVAL;
802                 if (gfs2_assert_withdraw(sdp, 0) == -1)
803                         fs_err(sdp, "ret = 0x%.8X\n", ret);
804 out:
805                 spin_unlock(&gl->gl_spin);
806         }
807
808         if (glops->go_xmote_bh)
809                 glops->go_xmote_bh(gl);
810
811         if (op_done) {
812                 spin_lock(&gl->gl_spin);
813                 gl->gl_req_gh = NULL;
814                 gl->gl_req_bh = NULL;
815                 clear_bit(GLF_LOCK, &gl->gl_flags);
816                 run_queue(gl);
817                 spin_unlock(&gl->gl_spin);
818         }
819
820         gfs2_glock_put(gl);
821
822         if (gh)
823                 gfs2_holder_wake(gh);
824 }
825
826 /**
827  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
828  * @gl: The glock in question
829  * @state: the requested state
830  * @flags: modifier flags to the lock call
831  *
832  */
833
834 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
835 {
836         struct gfs2_sbd *sdp = gl->gl_sbd;
837         int flags = gh ? gh->gh_flags : 0;
838         unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
839         const struct gfs2_glock_operations *glops = gl->gl_ops;
840         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
841                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
842                                  LM_FLAG_PRIORITY);
843         unsigned int lck_ret;
844
845         if (glops->go_xmote_th)
846                 glops->go_xmote_th(gl);
847
848         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
849         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
850         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
851         gfs2_assert_warn(sdp, state != gl->gl_state);
852
853         gfs2_glock_hold(gl);
854         gl->gl_req_bh = xmote_bh;
855
856         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
857
858         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
859                 return;
860
861         if (lck_ret & LM_OUT_ASYNC)
862                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
863         else
864                 xmote_bh(gl, lck_ret);
865 }
866
867 /**
868  * drop_bh - Called after a lock module unlock completes
869  * @gl: the glock
870  * @ret: the return status
871  *
872  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
873  * Doesn't drop the reference on the glock the top half took out
874  *
875  */
876
877 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
878 {
879         struct gfs2_sbd *sdp = gl->gl_sbd;
880         const struct gfs2_glock_operations *glops = gl->gl_ops;
881         struct gfs2_holder *gh = gl->gl_req_gh;
882
883         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
884         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
885         gfs2_assert_warn(sdp, !ret);
886
887         state_change(gl, LM_ST_UNLOCKED);
888         gfs2_demote_wake(gl);
889
890         if (glops->go_inval)
891                 glops->go_inval(gl, DIO_METADATA);
892
893         if (gh) {
894                 spin_lock(&gl->gl_spin);
895                 list_del_init(&gh->gh_list);
896                 gh->gh_error = 0;
897                 spin_unlock(&gl->gl_spin);
898         }
899
900         spin_lock(&gl->gl_spin);
901         gl->gl_req_gh = NULL;
902         gl->gl_req_bh = NULL;
903         clear_bit(GLF_LOCK, &gl->gl_flags);
904         run_queue(gl);
905         spin_unlock(&gl->gl_spin);
906
907         gfs2_glock_put(gl);
908
909         if (gh)
910                 gfs2_holder_wake(gh);
911 }
912
913 /**
914  * gfs2_glock_drop_th - call into the lock module to unlock a lock
915  * @gl: the glock
916  *
917  */
918
919 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
920 {
921         struct gfs2_sbd *sdp = gl->gl_sbd;
922         const struct gfs2_glock_operations *glops = gl->gl_ops;
923         unsigned int ret;
924
925         if (glops->go_drop_th)
926                 glops->go_drop_th(gl);
927
928         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
929         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
930         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
931
932         gfs2_glock_hold(gl);
933         gl->gl_req_bh = drop_bh;
934
935         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
936
937         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
938                 return;
939
940         if (!ret)
941                 drop_bh(gl, ret);
942         else
943                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
944 }
945
946 /**
947  * do_cancels - cancel requests for locks stuck waiting on an expire flag
948  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
949  *
950  * Don't cancel GL_NOCANCEL requests.
951  */
952
953 static void do_cancels(struct gfs2_holder *gh)
954 {
955         struct gfs2_glock *gl = gh->gh_gl;
956
957         spin_lock(&gl->gl_spin);
958
959         while (gl->gl_req_gh != gh &&
960                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
961                !list_empty(&gh->gh_list)) {
962                 if (gl->gl_req_bh && !(gl->gl_req_gh &&
963                                      (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
964                         spin_unlock(&gl->gl_spin);
965                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
966                         msleep(100);
967                         spin_lock(&gl->gl_spin);
968                 } else {
969                         spin_unlock(&gl->gl_spin);
970                         msleep(100);
971                         spin_lock(&gl->gl_spin);
972                 }
973         }
974
975         spin_unlock(&gl->gl_spin);
976 }
977
978 /**
979  * glock_wait_internal - wait on a glock acquisition
980  * @gh: the glock holder
981  *
982  * Returns: 0 on success
983  */
984
985 static int glock_wait_internal(struct gfs2_holder *gh)
986 {
987         struct gfs2_glock *gl = gh->gh_gl;
988         struct gfs2_sbd *sdp = gl->gl_sbd;
989         const struct gfs2_glock_operations *glops = gl->gl_ops;
990
991         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
992                 return -EIO;
993
994         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
995                 spin_lock(&gl->gl_spin);
996                 if (gl->gl_req_gh != gh &&
997                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
998                     !list_empty(&gh->gh_list)) {
999                         list_del_init(&gh->gh_list);
1000                         gh->gh_error = GLR_TRYFAILED;
1001                         run_queue(gl);
1002                         spin_unlock(&gl->gl_spin);
1003                         return gh->gh_error;
1004                 }
1005                 spin_unlock(&gl->gl_spin);
1006         }
1007
1008         if (gh->gh_flags & LM_FLAG_PRIORITY)
1009                 do_cancels(gh);
1010
1011         wait_on_holder(gh);
1012         if (gh->gh_error)
1013                 return gh->gh_error;
1014
1015         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1016         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1017                                                    gh->gh_flags));
1018
1019         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1020                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1021
1022                 if (glops->go_lock) {
1023                         gh->gh_error = glops->go_lock(gh);
1024                         if (gh->gh_error) {
1025                                 spin_lock(&gl->gl_spin);
1026                                 list_del_init(&gh->gh_list);
1027                                 spin_unlock(&gl->gl_spin);
1028                         }
1029                 }
1030
1031                 spin_lock(&gl->gl_spin);
1032                 gl->gl_req_gh = NULL;
1033                 gl->gl_req_bh = NULL;
1034                 clear_bit(GLF_LOCK, &gl->gl_flags);
1035                 run_queue(gl);
1036                 spin_unlock(&gl->gl_spin);
1037         }
1038
1039         return gh->gh_error;
1040 }
1041
1042 static inline struct gfs2_holder *
1043 find_holder_by_owner(struct list_head *head, pid_t pid)
1044 {
1045         struct gfs2_holder *gh;
1046
1047         list_for_each_entry(gh, head, gh_list) {
1048                 if (gh->gh_owner_pid == pid)
1049                         return gh;
1050         }
1051
1052         return NULL;
1053 }
1054
1055 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1056 {
1057         va_list args;
1058
1059         va_start(args, fmt);
1060         if (gi) {
1061                 vsprintf(gi->string, fmt, args);
1062                 seq_printf(gi->seq, gi->string);
1063         }
1064         else
1065                 vprintk(fmt, args);
1066         va_end(args);
1067 }
1068
1069 /**
1070  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1071  * @gh: the holder structure to add
1072  *
1073  */
1074
1075 static void add_to_queue(struct gfs2_holder *gh)
1076 {
1077         struct gfs2_glock *gl = gh->gh_gl;
1078         struct gfs2_holder *existing;
1079
1080         BUG_ON(!gh->gh_owner_pid);
1081         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1082                 BUG();
1083
1084         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner_pid);
1085         if (existing) {
1086                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1087                 printk(KERN_INFO "pid : %d\n", existing->gh_owner_pid);
1088                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1089                                 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1090                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1091                 printk(KERN_INFO "pid : %d\n", gh->gh_owner_pid);
1092                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1093                                 gl->gl_name.ln_type, gl->gl_state);
1094                 BUG();
1095         }
1096
1097         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner_pid);
1098         if (existing) {
1099                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1100                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1101                 BUG();
1102         }
1103
1104         if (gh->gh_flags & LM_FLAG_PRIORITY)
1105                 list_add(&gh->gh_list, &gl->gl_waiters3);
1106         else
1107                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1108 }
1109
1110 /**
1111  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1112  * @gh: the holder structure
1113  *
1114  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1115  *
1116  * Returns: 0, GLR_TRYFAILED, or errno on failure
1117  */
1118
1119 int gfs2_glock_nq(struct gfs2_holder *gh)
1120 {
1121         struct gfs2_glock *gl = gh->gh_gl;
1122         struct gfs2_sbd *sdp = gl->gl_sbd;
1123         int error = 0;
1124
1125 restart:
1126         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1127                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1128                 return -EIO;
1129         }
1130
1131         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1132
1133         spin_lock(&gl->gl_spin);
1134         add_to_queue(gh);
1135         run_queue(gl);
1136         spin_unlock(&gl->gl_spin);
1137
1138         if (!(gh->gh_flags & GL_ASYNC)) {
1139                 error = glock_wait_internal(gh);
1140                 if (error == GLR_CANCELED) {
1141                         msleep(100);
1142                         goto restart;
1143                 }
1144         }
1145
1146         return error;
1147 }
1148
1149 /**
1150  * gfs2_glock_poll - poll to see if an async request has been completed
1151  * @gh: the holder
1152  *
1153  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1154  */
1155
1156 int gfs2_glock_poll(struct gfs2_holder *gh)
1157 {
1158         struct gfs2_glock *gl = gh->gh_gl;
1159         int ready = 0;
1160
1161         spin_lock(&gl->gl_spin);
1162
1163         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1164                 ready = 1;
1165         else if (list_empty(&gh->gh_list)) {
1166                 if (gh->gh_error == GLR_CANCELED) {
1167                         spin_unlock(&gl->gl_spin);
1168                         msleep(100);
1169                         if (gfs2_glock_nq(gh))
1170                                 return 1;
1171                         return 0;
1172                 } else
1173                         ready = 1;
1174         }
1175
1176         spin_unlock(&gl->gl_spin);
1177
1178         return ready;
1179 }
1180
1181 /**
1182  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1183  * @gh: the holder structure
1184  *
1185  * Returns: 0, GLR_TRYFAILED, or errno on failure
1186  */
1187
1188 int gfs2_glock_wait(struct gfs2_holder *gh)
1189 {
1190         int error;
1191
1192         error = glock_wait_internal(gh);
1193         if (error == GLR_CANCELED) {
1194                 msleep(100);
1195                 gh->gh_flags &= ~GL_ASYNC;
1196                 error = gfs2_glock_nq(gh);
1197         }
1198
1199         return error;
1200 }
1201
1202 /**
1203  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1204  * @gh: the glock holder
1205  *
1206  */
1207
1208 void gfs2_glock_dq(struct gfs2_holder *gh)
1209 {
1210         struct gfs2_glock *gl = gh->gh_gl;
1211         const struct gfs2_glock_operations *glops = gl->gl_ops;
1212
1213         if (gh->gh_flags & GL_NOCACHE)
1214                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1215
1216         gfs2_glmutex_lock(gl);
1217
1218         spin_lock(&gl->gl_spin);
1219         list_del_init(&gh->gh_list);
1220
1221         if (list_empty(&gl->gl_holders)) {
1222                 spin_unlock(&gl->gl_spin);
1223
1224                 if (glops->go_unlock)
1225                         glops->go_unlock(gh);
1226
1227                 spin_lock(&gl->gl_spin);
1228                 gl->gl_stamp = jiffies;
1229         }
1230
1231         clear_bit(GLF_LOCK, &gl->gl_flags);
1232         run_queue(gl);
1233         spin_unlock(&gl->gl_spin);
1234 }
1235
1236 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1237 {
1238         struct gfs2_glock *gl = gh->gh_gl;
1239         gfs2_glock_dq(gh);
1240         wait_on_demote(gl);
1241 }
1242
1243 /**
1244  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1245  * @gh: the holder structure
1246  *
1247  */
1248
1249 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1250 {
1251         gfs2_glock_dq(gh);
1252         gfs2_holder_uninit(gh);
1253 }
1254
1255 /**
1256  * gfs2_glock_nq_num - acquire a glock based on lock number
1257  * @sdp: the filesystem
1258  * @number: the lock number
1259  * @glops: the glock operations for the type of glock
1260  * @state: the state to acquire the glock in
1261  * @flags: modifier flags for the aquisition
1262  * @gh: the struct gfs2_holder
1263  *
1264  * Returns: errno
1265  */
1266
1267 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1268                       const struct gfs2_glock_operations *glops,
1269                       unsigned int state, int flags, struct gfs2_holder *gh)
1270 {
1271         struct gfs2_glock *gl;
1272         int error;
1273
1274         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1275         if (!error) {
1276                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1277                 gfs2_glock_put(gl);
1278         }
1279
1280         return error;
1281 }
1282
1283 /**
1284  * glock_compare - Compare two struct gfs2_glock structures for sorting
1285  * @arg_a: the first structure
1286  * @arg_b: the second structure
1287  *
1288  */
1289
1290 static int glock_compare(const void *arg_a, const void *arg_b)
1291 {
1292         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1293         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1294         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1295         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1296
1297         if (a->ln_number > b->ln_number)
1298                 return 1;
1299         if (a->ln_number < b->ln_number)
1300                 return -1;
1301         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1302         return 0;
1303 }
1304
1305 /**
1306  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1307  * @num_gh: the number of structures
1308  * @ghs: an array of struct gfs2_holder structures
1309  *
1310  * Returns: 0 on success (all glocks acquired),
1311  *          errno on failure (no glocks acquired)
1312  */
1313
1314 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1315                      struct gfs2_holder **p)
1316 {
1317         unsigned int x;
1318         int error = 0;
1319
1320         for (x = 0; x < num_gh; x++)
1321                 p[x] = &ghs[x];
1322
1323         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1324
1325         for (x = 0; x < num_gh; x++) {
1326                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1327
1328                 error = gfs2_glock_nq(p[x]);
1329                 if (error) {
1330                         while (x--)
1331                                 gfs2_glock_dq(p[x]);
1332                         break;
1333                 }
1334         }
1335
1336         return error;
1337 }
1338
1339 /**
1340  * gfs2_glock_nq_m - acquire multiple glocks
1341  * @num_gh: the number of structures
1342  * @ghs: an array of struct gfs2_holder structures
1343  *
1344  *
1345  * Returns: 0 on success (all glocks acquired),
1346  *          errno on failure (no glocks acquired)
1347  */
1348
1349 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1350 {
1351         struct gfs2_holder *tmp[4];
1352         struct gfs2_holder **pph = tmp;
1353         int error = 0;
1354
1355         switch(num_gh) {
1356         case 0:
1357                 return 0;
1358         case 1:
1359                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1360                 return gfs2_glock_nq(ghs);
1361         default:
1362                 if (num_gh <= 4)
1363                         break;
1364                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1365                 if (!pph)
1366                         return -ENOMEM;
1367         }
1368
1369         error = nq_m_sync(num_gh, ghs, pph);
1370
1371         if (pph != tmp)
1372                 kfree(pph);
1373
1374         return error;
1375 }
1376
1377 /**
1378  * gfs2_glock_dq_m - release multiple glocks
1379  * @num_gh: the number of structures
1380  * @ghs: an array of struct gfs2_holder structures
1381  *
1382  */
1383
1384 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1385 {
1386         unsigned int x;
1387
1388         for (x = 0; x < num_gh; x++)
1389                 gfs2_glock_dq(&ghs[x]);
1390 }
1391
1392 /**
1393  * gfs2_glock_dq_uninit_m - release multiple glocks
1394  * @num_gh: the number of structures
1395  * @ghs: an array of struct gfs2_holder structures
1396  *
1397  */
1398
1399 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1400 {
1401         unsigned int x;
1402
1403         for (x = 0; x < num_gh; x++)
1404                 gfs2_glock_dq_uninit(&ghs[x]);
1405 }
1406
1407 /**
1408  * gfs2_lvb_hold - attach a LVB from a glock
1409  * @gl: The glock in question
1410  *
1411  */
1412
1413 int gfs2_lvb_hold(struct gfs2_glock *gl)
1414 {
1415         int error;
1416
1417         gfs2_glmutex_lock(gl);
1418
1419         if (!atomic_read(&gl->gl_lvb_count)) {
1420                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1421                 if (error) {
1422                         gfs2_glmutex_unlock(gl);
1423                         return error;
1424                 }
1425                 gfs2_glock_hold(gl);
1426         }
1427         atomic_inc(&gl->gl_lvb_count);
1428
1429         gfs2_glmutex_unlock(gl);
1430
1431         return 0;
1432 }
1433
1434 /**
1435  * gfs2_lvb_unhold - detach a LVB from a glock
1436  * @gl: The glock in question
1437  *
1438  */
1439
1440 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1441 {
1442         gfs2_glock_hold(gl);
1443         gfs2_glmutex_lock(gl);
1444
1445         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1446         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1447                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1448                 gl->gl_lvb = NULL;
1449                 gfs2_glock_put(gl);
1450         }
1451
1452         gfs2_glmutex_unlock(gl);
1453         gfs2_glock_put(gl);
1454 }
1455
1456 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1457                         unsigned int state)
1458 {
1459         struct gfs2_glock *gl;
1460
1461         gl = gfs2_glock_find(sdp, name);
1462         if (!gl)
1463                 return;
1464
1465         handle_callback(gl, state, 1);
1466
1467         spin_lock(&gl->gl_spin);
1468         run_queue(gl);
1469         spin_unlock(&gl->gl_spin);
1470
1471         gfs2_glock_put(gl);
1472 }
1473
1474 /**
1475  * gfs2_glock_cb - Callback used by locking module
1476  * @sdp: Pointer to the superblock
1477  * @type: Type of callback
1478  * @data: Type dependent data pointer
1479  *
1480  * Called by the locking module when it wants to tell us something.
1481  * Either we need to drop a lock, one of our ASYNC requests completed, or
1482  * a journal from another client needs to be recovered.
1483  */
1484
1485 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1486 {
1487         struct gfs2_sbd *sdp = cb_data;
1488
1489         switch (type) {
1490         case LM_CB_NEED_E:
1491                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1492                 return;
1493
1494         case LM_CB_NEED_D:
1495                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1496                 return;
1497
1498         case LM_CB_NEED_S:
1499                 blocking_cb(sdp, data, LM_ST_SHARED);
1500                 return;
1501
1502         case LM_CB_ASYNC: {
1503                 struct lm_async_cb *async = data;
1504                 struct gfs2_glock *gl;
1505
1506                 down_read(&gfs2_umount_flush_sem);
1507                 gl = gfs2_glock_find(sdp, &async->lc_name);
1508                 if (gfs2_assert_warn(sdp, gl))
1509                         return;
1510                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1511                         gl->gl_req_bh(gl, async->lc_ret);
1512                 gfs2_glock_put(gl);
1513                 up_read(&gfs2_umount_flush_sem);
1514                 return;
1515         }
1516
1517         case LM_CB_NEED_RECOVERY:
1518                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1519                 if (sdp->sd_recoverd_process)
1520                         wake_up_process(sdp->sd_recoverd_process);
1521                 return;
1522
1523         case LM_CB_DROPLOCKS:
1524                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1525                 gfs2_quota_scan(sdp);
1526                 return;
1527
1528         default:
1529                 gfs2_assert_warn(sdp, 0);
1530                 return;
1531         }
1532 }
1533
1534 /**
1535  * demote_ok - Check to see if it's ok to unlock a glock
1536  * @gl: the glock
1537  *
1538  * Returns: 1 if it's ok
1539  */
1540
1541 static int demote_ok(struct gfs2_glock *gl)
1542 {
1543         const struct gfs2_glock_operations *glops = gl->gl_ops;
1544         int demote = 1;
1545
1546         if (test_bit(GLF_STICKY, &gl->gl_flags))
1547                 demote = 0;
1548         else if (glops->go_demote_ok)
1549                 demote = glops->go_demote_ok(gl);
1550
1551         return demote;
1552 }
1553
1554 /**
1555  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1556  * @gl: the glock
1557  *
1558  */
1559
1560 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1561 {
1562         struct gfs2_sbd *sdp = gl->gl_sbd;
1563
1564         spin_lock(&sdp->sd_reclaim_lock);
1565         if (list_empty(&gl->gl_reclaim)) {
1566                 gfs2_glock_hold(gl);
1567                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1568                 atomic_inc(&sdp->sd_reclaim_count);
1569         }
1570         spin_unlock(&sdp->sd_reclaim_lock);
1571
1572         wake_up(&sdp->sd_reclaim_wq);
1573 }
1574
1575 /**
1576  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1577  * @sdp: the filesystem
1578  *
1579  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1580  * different glock and we notice that there are a lot of glocks in the
1581  * reclaim list.
1582  *
1583  */
1584
1585 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1586 {
1587         struct gfs2_glock *gl;
1588
1589         spin_lock(&sdp->sd_reclaim_lock);
1590         if (list_empty(&sdp->sd_reclaim_list)) {
1591                 spin_unlock(&sdp->sd_reclaim_lock);
1592                 return;
1593         }
1594         gl = list_entry(sdp->sd_reclaim_list.next,
1595                         struct gfs2_glock, gl_reclaim);
1596         list_del_init(&gl->gl_reclaim);
1597         spin_unlock(&sdp->sd_reclaim_lock);
1598
1599         atomic_dec(&sdp->sd_reclaim_count);
1600         atomic_inc(&sdp->sd_reclaimed);
1601
1602         if (gfs2_glmutex_trylock(gl)) {
1603                 if (list_empty(&gl->gl_holders) &&
1604                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1605                         handle_callback(gl, LM_ST_UNLOCKED, 0);
1606                 gfs2_glmutex_unlock(gl);
1607         }
1608
1609         gfs2_glock_put(gl);
1610 }
1611
1612 /**
1613  * examine_bucket - Call a function for glock in a hash bucket
1614  * @examiner: the function
1615  * @sdp: the filesystem
1616  * @bucket: the bucket
1617  *
1618  * Returns: 1 if the bucket has entries
1619  */
1620
1621 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1622                           unsigned int hash)
1623 {
1624         struct gfs2_glock *gl, *prev = NULL;
1625         int has_entries = 0;
1626         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1627
1628         read_lock(gl_lock_addr(hash));
1629         /* Can't use hlist_for_each_entry - don't want prefetch here */
1630         if (hlist_empty(head))
1631                 goto out;
1632         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1633         while(1) {
1634                 if (!sdp || gl->gl_sbd == sdp) {
1635                         gfs2_glock_hold(gl);
1636                         read_unlock(gl_lock_addr(hash));
1637                         if (prev)
1638                                 gfs2_glock_put(prev);
1639                         prev = gl;
1640                         examiner(gl);
1641                         has_entries = 1;
1642                         read_lock(gl_lock_addr(hash));
1643                 }
1644                 if (gl->gl_list.next == NULL)
1645                         break;
1646                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1647         }
1648 out:
1649         read_unlock(gl_lock_addr(hash));
1650         if (prev)
1651                 gfs2_glock_put(prev);
1652         cond_resched();
1653         return has_entries;
1654 }
1655
1656 /**
1657  * scan_glock - look at a glock and see if we can reclaim it
1658  * @gl: the glock to look at
1659  *
1660  */
1661
1662 static void scan_glock(struct gfs2_glock *gl)
1663 {
1664         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1665                 return;
1666
1667         if (gfs2_glmutex_trylock(gl)) {
1668                 if (list_empty(&gl->gl_holders) &&
1669                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1670                         goto out_schedule;
1671                 gfs2_glmutex_unlock(gl);
1672         }
1673         return;
1674
1675 out_schedule:
1676         gfs2_glmutex_unlock(gl);
1677         gfs2_glock_schedule_for_reclaim(gl);
1678 }
1679
1680 /**
1681  * clear_glock - look at a glock and see if we can free it from glock cache
1682  * @gl: the glock to look at
1683  *
1684  */
1685
1686 static void clear_glock(struct gfs2_glock *gl)
1687 {
1688         struct gfs2_sbd *sdp = gl->gl_sbd;
1689         int released;
1690
1691         spin_lock(&sdp->sd_reclaim_lock);
1692         if (!list_empty(&gl->gl_reclaim)) {
1693                 list_del_init(&gl->gl_reclaim);
1694                 atomic_dec(&sdp->sd_reclaim_count);
1695                 spin_unlock(&sdp->sd_reclaim_lock);
1696                 released = gfs2_glock_put(gl);
1697                 gfs2_assert(sdp, !released);
1698         } else {
1699                 spin_unlock(&sdp->sd_reclaim_lock);
1700         }
1701
1702         if (gfs2_glmutex_trylock(gl)) {
1703                 if (list_empty(&gl->gl_holders) &&
1704                     gl->gl_state != LM_ST_UNLOCKED)
1705                         handle_callback(gl, LM_ST_UNLOCKED, 0);
1706                 gfs2_glmutex_unlock(gl);
1707         }
1708 }
1709
1710 /**
1711  * gfs2_gl_hash_clear - Empty out the glock hash table
1712  * @sdp: the filesystem
1713  * @wait: wait until it's all gone
1714  *
1715  * Called when unmounting the filesystem, or when inter-node lock manager
1716  * requests DROPLOCKS because it is running out of capacity.
1717  */
1718
1719 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1720 {
1721         unsigned long t;
1722         unsigned int x;
1723         int cont;
1724
1725         t = jiffies;
1726
1727         for (;;) {
1728                 cont = 0;
1729                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1730                         if (examine_bucket(clear_glock, sdp, x))
1731                                 cont = 1;
1732                 }
1733
1734                 if (!wait || !cont)
1735                         break;
1736
1737                 if (time_after_eq(jiffies,
1738                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1739                         fs_warn(sdp, "Unmount seems to be stalled. "
1740                                      "Dumping lock state...\n");
1741                         gfs2_dump_lockstate(sdp);
1742                         t = jiffies;
1743                 }
1744
1745                 down_write(&gfs2_umount_flush_sem);
1746                 invalidate_inodes(sdp->sd_vfs);
1747                 up_write(&gfs2_umount_flush_sem);
1748                 msleep(10);
1749         }
1750 }
1751
1752 /*
1753  *  Diagnostic routines to help debug distributed deadlock
1754  */
1755
1756 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1757                               unsigned long address)
1758 {
1759         char buffer[KSYM_SYMBOL_LEN];
1760
1761         sprint_symbol(buffer, address);
1762         print_dbg(gi, fmt, buffer);
1763 }
1764
1765 /**
1766  * dump_holder - print information about a glock holder
1767  * @str: a string naming the type of holder
1768  * @gh: the glock holder
1769  *
1770  * Returns: 0 on success, -ENOBUFS when we run out of space
1771  */
1772
1773 static int dump_holder(struct glock_iter *gi, char *str,
1774                        struct gfs2_holder *gh)
1775 {
1776         unsigned int x;
1777         struct task_struct *gh_owner;
1778
1779         print_dbg(gi, "  %s\n", str);
1780         if (gh->gh_owner_pid) {
1781                 print_dbg(gi, "    owner = %ld ", (long)gh->gh_owner_pid);
1782                 gh_owner = find_task_by_pid(gh->gh_owner_pid);
1783                 if (gh_owner)
1784                         print_dbg(gi, "(%s)\n", gh_owner->comm);
1785                 else
1786                         print_dbg(gi, "(ended)\n");
1787         } else
1788                 print_dbg(gi, "    owner = -1\n");
1789         print_dbg(gi, "    gh_state = %u\n", gh->gh_state);
1790         print_dbg(gi, "    gh_flags =");
1791         for (x = 0; x < 32; x++)
1792                 if (gh->gh_flags & (1 << x))
1793                         print_dbg(gi, " %u", x);
1794         print_dbg(gi, " \n");
1795         print_dbg(gi, "    error = %d\n", gh->gh_error);
1796         print_dbg(gi, "    gh_iflags =");
1797         for (x = 0; x < 32; x++)
1798                 if (test_bit(x, &gh->gh_iflags))
1799                         print_dbg(gi, " %u", x);
1800         print_dbg(gi, " \n");
1801         gfs2_print_symbol(gi, "    initialized at: %s\n", gh->gh_ip);
1802
1803         return 0;
1804 }
1805
1806 /**
1807  * dump_inode - print information about an inode
1808  * @ip: the inode
1809  *
1810  * Returns: 0 on success, -ENOBUFS when we run out of space
1811  */
1812
1813 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1814 {
1815         unsigned int x;
1816
1817         print_dbg(gi, "  Inode:\n");
1818         print_dbg(gi, "    num = %llu/%llu\n",
1819                   (unsigned long long)ip->i_no_formal_ino,
1820                   (unsigned long long)ip->i_no_addr);
1821         print_dbg(gi, "    type = %u\n", IF2DT(ip->i_inode.i_mode));
1822         print_dbg(gi, "    i_flags =");
1823         for (x = 0; x < 32; x++)
1824                 if (test_bit(x, &ip->i_flags))
1825                         print_dbg(gi, " %u", x);
1826         print_dbg(gi, " \n");
1827         return 0;
1828 }
1829
1830 /**
1831  * dump_glock - print information about a glock
1832  * @gl: the glock
1833  * @count: where we are in the buffer
1834  *
1835  * Returns: 0 on success, -ENOBUFS when we run out of space
1836  */
1837
1838 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1839 {
1840         struct gfs2_holder *gh;
1841         unsigned int x;
1842         int error = -ENOBUFS;
1843         struct task_struct *gl_owner;
1844
1845         spin_lock(&gl->gl_spin);
1846
1847         print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
1848                    (unsigned long long)gl->gl_name.ln_number);
1849         print_dbg(gi, "  gl_flags =");
1850         for (x = 0; x < 32; x++) {
1851                 if (test_bit(x, &gl->gl_flags))
1852                         print_dbg(gi, " %u", x);
1853         }
1854         if (!test_bit(GLF_LOCK, &gl->gl_flags))
1855                 print_dbg(gi, " (unlocked)");
1856         print_dbg(gi, " \n");
1857         print_dbg(gi, "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
1858         print_dbg(gi, "  gl_state = %u\n", gl->gl_state);
1859         if (gl->gl_owner_pid) {
1860                 gl_owner = find_task_by_pid(gl->gl_owner_pid);
1861                 if (gl_owner)
1862                         print_dbg(gi, "  gl_owner = pid %d (%s)\n",
1863                                   gl->gl_owner_pid, gl_owner->comm);
1864                 else
1865                         print_dbg(gi, "  gl_owner = %d (ended)\n",
1866                                   gl->gl_owner_pid);
1867         } else
1868                 print_dbg(gi, "  gl_owner = -1\n");
1869         print_dbg(gi, "  gl_ip = %lu\n", gl->gl_ip);
1870         print_dbg(gi, "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1871         print_dbg(gi, "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1872         print_dbg(gi, "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1873         print_dbg(gi, "  object = %s\n", (gl->gl_object) ? "yes" : "no");
1874         print_dbg(gi, "  le = %s\n",
1875                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
1876         print_dbg(gi, "  reclaim = %s\n",
1877                    (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1878         if (gl->gl_aspace)
1879                 print_dbg(gi, "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1880                            gl->gl_aspace->i_mapping->nrpages);
1881         else
1882                 print_dbg(gi, "  aspace = no\n");
1883         print_dbg(gi, "  ail = %d\n", atomic_read(&gl->gl_ail_count));
1884         if (gl->gl_req_gh) {
1885                 error = dump_holder(gi, "Request", gl->gl_req_gh);
1886                 if (error)
1887                         goto out;
1888         }
1889         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1890                 error = dump_holder(gi, "Holder", gh);
1891                 if (error)
1892                         goto out;
1893         }
1894         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1895                 error = dump_holder(gi, "Waiter1", gh);
1896                 if (error)
1897                         goto out;
1898         }
1899         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1900                 error = dump_holder(gi, "Waiter3", gh);
1901                 if (error)
1902                         goto out;
1903         }
1904         if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1905                 print_dbg(gi, "  Demotion req to state %u (%llu uS ago)\n",
1906                           gl->gl_demote_state, (unsigned long long)
1907                           (jiffies - gl->gl_demote_time)*(1000000/HZ));
1908         }
1909         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1910                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1911                         list_empty(&gl->gl_holders)) {
1912                         error = dump_inode(gi, gl->gl_object);
1913                         if (error)
1914                                 goto out;
1915                 } else {
1916                         error = -ENOBUFS;
1917                         print_dbg(gi, "  Inode: busy\n");
1918                 }
1919         }
1920
1921         error = 0;
1922
1923 out:
1924         spin_unlock(&gl->gl_spin);
1925         return error;
1926 }
1927
1928 /**
1929  * gfs2_dump_lockstate - print out the current lockstate
1930  * @sdp: the filesystem
1931  * @ub: the buffer to copy the information into
1932  *
1933  * If @ub is NULL, dump the lockstate to the console.
1934  *
1935  */
1936
1937 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1938 {
1939         struct gfs2_glock *gl;
1940         struct hlist_node *h;
1941         unsigned int x;
1942         int error = 0;
1943
1944         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1945
1946                 read_lock(gl_lock_addr(x));
1947
1948                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1949                         if (gl->gl_sbd != sdp)
1950                                 continue;
1951
1952                         error = dump_glock(NULL, gl);
1953                         if (error)
1954                                 break;
1955                 }
1956
1957                 read_unlock(gl_lock_addr(x));
1958
1959                 if (error)
1960                         break;
1961         }
1962
1963
1964         return error;
1965 }
1966
1967 /**
1968  * gfs2_scand - Look for cached glocks and inodes to toss from memory
1969  * @sdp: Pointer to GFS2 superblock
1970  *
1971  * One of these daemons runs, finding candidates to add to sd_reclaim_list.
1972  * See gfs2_glockd()
1973  */
1974
1975 static int gfs2_scand(void *data)
1976 {
1977         unsigned x;
1978         unsigned delay;
1979
1980         while (!kthread_should_stop()) {
1981                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1982                         examine_bucket(scan_glock, NULL, x);
1983                 if (freezing(current))
1984                         refrigerator();
1985                 delay = scand_secs;
1986                 if (delay < 1)
1987                         delay = 1;
1988                 schedule_timeout_interruptible(delay * HZ);
1989         }
1990
1991         return 0;
1992 }
1993
1994
1995
1996 int __init gfs2_glock_init(void)
1997 {
1998         unsigned i;
1999         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2000                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2001         }
2002 #ifdef GL_HASH_LOCK_SZ
2003         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2004                 rwlock_init(&gl_hash_locks[i]);
2005         }
2006 #endif
2007
2008         scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
2009         if (IS_ERR(scand_process))
2010                 return PTR_ERR(scand_process);
2011
2012         return 0;
2013 }
2014
2015 void gfs2_glock_exit(void)
2016 {
2017         kthread_stop(scand_process);
2018 }
2019
2020 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
2021 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
2022
2023 static int gfs2_glock_iter_next(struct glock_iter *gi)
2024 {
2025         struct gfs2_glock *gl;
2026
2027 restart:
2028         read_lock(gl_lock_addr(gi->hash));
2029         gl = gi->gl;
2030         if (gl) {
2031                 gi->gl = hlist_entry(gl->gl_list.next,
2032                                      struct gfs2_glock, gl_list);
2033                 if (gi->gl)
2034                         gfs2_glock_hold(gi->gl);
2035         }
2036         read_unlock(gl_lock_addr(gi->hash));
2037         if (gl)
2038                 gfs2_glock_put(gl);
2039         if (gl && gi->gl == NULL)
2040                 gi->hash++;
2041         while(gi->gl == NULL) {
2042                 if (gi->hash >= GFS2_GL_HASH_SIZE)
2043                         return 1;
2044                 read_lock(gl_lock_addr(gi->hash));
2045                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
2046                                      struct gfs2_glock, gl_list);
2047                 if (gi->gl)
2048                         gfs2_glock_hold(gi->gl);
2049                 read_unlock(gl_lock_addr(gi->hash));
2050                 gi->hash++;
2051         }
2052
2053         if (gi->sdp != gi->gl->gl_sbd)
2054                 goto restart;
2055
2056         return 0;
2057 }
2058
2059 static void gfs2_glock_iter_free(struct glock_iter *gi)
2060 {
2061         if (gi->gl)
2062                 gfs2_glock_put(gi->gl);
2063         kfree(gi);
2064 }
2065
2066 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2067 {
2068         struct glock_iter *gi;
2069
2070         gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2071         if (!gi)
2072                 return NULL;
2073
2074         gi->sdp = sdp;
2075         gi->hash = 0;
2076         gi->seq = NULL;
2077         gi->gl = NULL;
2078         memset(gi->string, 0, sizeof(gi->string));
2079
2080         if (gfs2_glock_iter_next(gi)) {
2081                 gfs2_glock_iter_free(gi);
2082                 return NULL;
2083         }
2084
2085         return gi;
2086 }
2087
2088 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2089 {
2090         struct glock_iter *gi;
2091         loff_t n = *pos;
2092
2093         gi = gfs2_glock_iter_init(file->private);
2094         if (!gi)
2095                 return NULL;
2096
2097         while(n--) {
2098                 if (gfs2_glock_iter_next(gi)) {
2099                         gfs2_glock_iter_free(gi);
2100                         return NULL;
2101                 }
2102         }
2103
2104         return gi;
2105 }
2106
2107 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2108                                  loff_t *pos)
2109 {
2110         struct glock_iter *gi = iter_ptr;
2111
2112         (*pos)++;
2113
2114         if (gfs2_glock_iter_next(gi)) {
2115                 gfs2_glock_iter_free(gi);
2116                 return NULL;
2117         }
2118
2119         return gi;
2120 }
2121
2122 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2123 {
2124         struct glock_iter *gi = iter_ptr;
2125         if (gi)
2126                 gfs2_glock_iter_free(gi);
2127 }
2128
2129 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2130 {
2131         struct glock_iter *gi = iter_ptr;
2132
2133         gi->seq = file;
2134         dump_glock(gi, gi->gl);
2135
2136         return 0;
2137 }
2138
2139 static const struct seq_operations gfs2_glock_seq_ops = {
2140         .start = gfs2_glock_seq_start,
2141         .next  = gfs2_glock_seq_next,
2142         .stop  = gfs2_glock_seq_stop,
2143         .show  = gfs2_glock_seq_show,
2144 };
2145
2146 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2147 {
2148         struct seq_file *seq;
2149         int ret;
2150
2151         ret = seq_open(file, &gfs2_glock_seq_ops);
2152         if (ret)
2153                 return ret;
2154
2155         seq = file->private_data;
2156         seq->private = inode->i_private;
2157
2158         return 0;
2159 }
2160
2161 static const struct file_operations gfs2_debug_fops = {
2162         .owner   = THIS_MODULE,
2163         .open    = gfs2_debugfs_open,
2164         .read    = seq_read,
2165         .llseek  = seq_lseek,
2166         .release = seq_release
2167 };
2168
2169 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2170 {
2171         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2172         if (!sdp->debugfs_dir)
2173                 return -ENOMEM;
2174         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2175                                                          S_IFREG | S_IRUGO,
2176                                                          sdp->debugfs_dir, sdp,
2177                                                          &gfs2_debug_fops);
2178         if (!sdp->debugfs_dentry_glocks)
2179                 return -ENOMEM;
2180
2181         return 0;
2182 }
2183
2184 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2185 {
2186         if (sdp && sdp->debugfs_dir) {
2187                 if (sdp->debugfs_dentry_glocks) {
2188                         debugfs_remove(sdp->debugfs_dentry_glocks);
2189                         sdp->debugfs_dentry_glocks = NULL;
2190                 }
2191                 debugfs_remove(sdp->debugfs_dir);
2192                 sdp->debugfs_dir = NULL;
2193         }
2194 }
2195
2196 int gfs2_register_debugfs(void)
2197 {
2198         gfs2_root = debugfs_create_dir("gfs2", NULL);
2199         return gfs2_root ? 0 : -ENOMEM;
2200 }
2201
2202 void gfs2_unregister_debugfs(void)
2203 {
2204         debugfs_remove(gfs2_root);
2205         gfs2_root = NULL;
2206 }