[GFS2] fix bz 231369, gfs2 will oops if you specify an invalid mount option
[safe/jmp/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
29 #include "gfs2.h"
30 #include "incore.h"
31 #include "glock.h"
32 #include "glops.h"
33 #include "inode.h"
34 #include "lm.h"
35 #include "lops.h"
36 #include "meta_io.h"
37 #include "quota.h"
38 #include "super.h"
39 #include "util.h"
40
41 struct gfs2_gl_hash_bucket {
42         struct hlist_head hb_list;
43 };
44
45 struct glock_iter {
46         int hash;                     /* hash bucket index         */
47         struct gfs2_sbd *sdp;         /* incore superblock         */
48         struct gfs2_glock *gl;        /* current glock struct      */
49         struct hlist_head *hb_list;   /* current hash bucket ptr   */
50         struct seq_file *seq;         /* sequence file for debugfs */
51         char string[512];             /* scratch space             */
52 };
53
54 typedef void (*glock_examiner) (struct gfs2_glock * gl);
55
56 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
57 static void gfs2_glock_xmote_th(struct gfs2_holder *gh);
58 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
59 static DECLARE_RWSEM(gfs2_umount_flush_sem);
60 static struct dentry *gfs2_root;
61
62 #define GFS2_GL_HASH_SHIFT      15
63 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
64 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
65
66 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
67
68 /*
69  * Despite what you might think, the numbers below are not arbitrary :-)
70  * They are taken from the ipv4 routing hash code, which is well tested
71  * and thus should be nearly optimal. Later on we might tweek the numbers
72  * but for now this should be fine.
73  *
74  * The reason for putting the locks in a separate array from the list heads
75  * is that we can have fewer locks than list heads and save memory. We use
76  * the same hash function for both, but with a different hash mask.
77  */
78 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
79         defined(CONFIG_PROVE_LOCKING)
80
81 #ifdef CONFIG_LOCKDEP
82 # define GL_HASH_LOCK_SZ        256
83 #else
84 # if NR_CPUS >= 32
85 #  define GL_HASH_LOCK_SZ       4096
86 # elif NR_CPUS >= 16
87 #  define GL_HASH_LOCK_SZ       2048
88 # elif NR_CPUS >= 8
89 #  define GL_HASH_LOCK_SZ       1024
90 # elif NR_CPUS >= 4
91 #  define GL_HASH_LOCK_SZ       512
92 # else
93 #  define GL_HASH_LOCK_SZ       256
94 # endif
95 #endif
96
97 /* We never want more locks than chains */
98 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
99 # undef GL_HASH_LOCK_SZ
100 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
101 #endif
102
103 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
104
105 static inline rwlock_t *gl_lock_addr(unsigned int x)
106 {
107         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
108 }
109 #else /* not SMP, so no spinlocks required */
110 static inline rwlock_t *gl_lock_addr(unsigned int x)
111 {
112         return NULL;
113 }
114 #endif
115
116 /**
117  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
118  * @actual: the current state of the lock
119  * @requested: the lock state that was requested by the caller
120  * @flags: the modifier flags passed in by the caller
121  *
122  * Returns: 1 if the locks are compatible, 0 otherwise
123  */
124
125 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
126                                    int flags)
127 {
128         if (actual == requested)
129                 return 1;
130
131         if (flags & GL_EXACT)
132                 return 0;
133
134         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
135                 return 1;
136
137         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
138                 return 1;
139
140         return 0;
141 }
142
143 /**
144  * gl_hash() - Turn glock number into hash bucket number
145  * @lock: The glock number
146  *
147  * Returns: The number of the corresponding hash bucket
148  */
149
150 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
151                             const struct lm_lockname *name)
152 {
153         unsigned int h;
154
155         h = jhash(&name->ln_number, sizeof(u64), 0);
156         h = jhash(&name->ln_type, sizeof(unsigned int), h);
157         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
158         h &= GFS2_GL_HASH_MASK;
159
160         return h;
161 }
162
163 /**
164  * glock_free() - Perform a few checks and then release struct gfs2_glock
165  * @gl: The glock to release
166  *
167  * Also calls lock module to release its internal structure for this glock.
168  *
169  */
170
171 static void glock_free(struct gfs2_glock *gl)
172 {
173         struct gfs2_sbd *sdp = gl->gl_sbd;
174         struct inode *aspace = gl->gl_aspace;
175
176         gfs2_lm_put_lock(sdp, gl->gl_lock);
177
178         if (aspace)
179                 gfs2_aspace_put(aspace);
180
181         kmem_cache_free(gfs2_glock_cachep, gl);
182 }
183
184 /**
185  * gfs2_glock_hold() - increment reference count on glock
186  * @gl: The glock to hold
187  *
188  */
189
190 void gfs2_glock_hold(struct gfs2_glock *gl)
191 {
192         atomic_inc(&gl->gl_ref);
193 }
194
195 /**
196  * gfs2_glock_put() - Decrement reference count on glock
197  * @gl: The glock to put
198  *
199  */
200
201 int gfs2_glock_put(struct gfs2_glock *gl)
202 {
203         int rv = 0;
204         struct gfs2_sbd *sdp = gl->gl_sbd;
205
206         write_lock(gl_lock_addr(gl->gl_hash));
207         if (atomic_dec_and_test(&gl->gl_ref)) {
208                 hlist_del(&gl->gl_list);
209                 write_unlock(gl_lock_addr(gl->gl_hash));
210                 BUG_ON(spin_is_locked(&gl->gl_spin));
211                 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
212                 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
213                 gfs2_assert(sdp, list_empty(&gl->gl_holders));
214                 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
215                 gfs2_assert(sdp, list_empty(&gl->gl_waiters2));
216                 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
217                 glock_free(gl);
218                 rv = 1;
219                 goto out;
220         }
221         write_unlock(gl_lock_addr(gl->gl_hash));
222 out:
223         return rv;
224 }
225
226 /**
227  * search_bucket() - Find struct gfs2_glock by lock number
228  * @bucket: the bucket to search
229  * @name: The lock name
230  *
231  * Returns: NULL, or the struct gfs2_glock with the requested number
232  */
233
234 static struct gfs2_glock *search_bucket(unsigned int hash,
235                                         const struct gfs2_sbd *sdp,
236                                         const struct lm_lockname *name)
237 {
238         struct gfs2_glock *gl;
239         struct hlist_node *h;
240
241         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
242                 if (!lm_name_equal(&gl->gl_name, name))
243                         continue;
244                 if (gl->gl_sbd != sdp)
245                         continue;
246
247                 atomic_inc(&gl->gl_ref);
248
249                 return gl;
250         }
251
252         return NULL;
253 }
254
255 /**
256  * gfs2_glock_find() - Find glock by lock number
257  * @sdp: The GFS2 superblock
258  * @name: The lock name
259  *
260  * Returns: NULL, or the struct gfs2_glock with the requested number
261  */
262
263 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
264                                           const struct lm_lockname *name)
265 {
266         unsigned int hash = gl_hash(sdp, name);
267         struct gfs2_glock *gl;
268
269         read_lock(gl_lock_addr(hash));
270         gl = search_bucket(hash, sdp, name);
271         read_unlock(gl_lock_addr(hash));
272
273         return gl;
274 }
275
276 /**
277  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
278  * @sdp: The GFS2 superblock
279  * @number: the lock number
280  * @glops: The glock_operations to use
281  * @create: If 0, don't create the glock if it doesn't exist
282  * @glp: the glock is returned here
283  *
284  * This does not lock a glock, just finds/creates structures for one.
285  *
286  * Returns: errno
287  */
288
289 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
290                    const struct gfs2_glock_operations *glops, int create,
291                    struct gfs2_glock **glp)
292 {
293         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
294         struct gfs2_glock *gl, *tmp;
295         unsigned int hash = gl_hash(sdp, &name);
296         int error;
297
298         read_lock(gl_lock_addr(hash));
299         gl = search_bucket(hash, sdp, &name);
300         read_unlock(gl_lock_addr(hash));
301
302         if (gl || !create) {
303                 *glp = gl;
304                 return 0;
305         }
306
307         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
308         if (!gl)
309                 return -ENOMEM;
310
311         gl->gl_flags = 0;
312         gl->gl_name = name;
313         atomic_set(&gl->gl_ref, 1);
314         gl->gl_state = LM_ST_UNLOCKED;
315         gl->gl_hash = hash;
316         gl->gl_owner = NULL;
317         gl->gl_ip = 0;
318         gl->gl_ops = glops;
319         gl->gl_req_gh = NULL;
320         gl->gl_req_bh = NULL;
321         gl->gl_vn = 0;
322         gl->gl_stamp = jiffies;
323         gl->gl_object = NULL;
324         gl->gl_sbd = sdp;
325         gl->gl_aspace = NULL;
326         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
327
328         /* If this glock protects actual on-disk data or metadata blocks,
329            create a VFS inode to manage the pages/buffers holding them. */
330         if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
331                 gl->gl_aspace = gfs2_aspace_get(sdp);
332                 if (!gl->gl_aspace) {
333                         error = -ENOMEM;
334                         goto fail;
335                 }
336         }
337
338         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
339         if (error)
340                 goto fail_aspace;
341
342         write_lock(gl_lock_addr(hash));
343         tmp = search_bucket(hash, sdp, &name);
344         if (tmp) {
345                 write_unlock(gl_lock_addr(hash));
346                 glock_free(gl);
347                 gl = tmp;
348         } else {
349                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
350                 write_unlock(gl_lock_addr(hash));
351         }
352
353         *glp = gl;
354
355         return 0;
356
357 fail_aspace:
358         if (gl->gl_aspace)
359                 gfs2_aspace_put(gl->gl_aspace);
360 fail:
361         kmem_cache_free(gfs2_glock_cachep, gl);
362         return error;
363 }
364
365 /**
366  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
367  * @gl: the glock
368  * @state: the state we're requesting
369  * @flags: the modifier flags
370  * @gh: the holder structure
371  *
372  */
373
374 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
375                       struct gfs2_holder *gh)
376 {
377         INIT_LIST_HEAD(&gh->gh_list);
378         gh->gh_gl = gl;
379         gh->gh_ip = (unsigned long)__builtin_return_address(0);
380         gh->gh_owner = current;
381         gh->gh_state = state;
382         gh->gh_flags = flags;
383         gh->gh_error = 0;
384         gh->gh_iflags = 0;
385         gfs2_glock_hold(gl);
386 }
387
388 /**
389  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
390  * @state: the state we're requesting
391  * @flags: the modifier flags
392  * @gh: the holder structure
393  *
394  * Don't mess with the glock.
395  *
396  */
397
398 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
399 {
400         gh->gh_state = state;
401         gh->gh_flags = flags;
402         gh->gh_iflags &= 1 << HIF_ALLOCED;
403         gh->gh_ip = (unsigned long)__builtin_return_address(0);
404 }
405
406 /**
407  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
408  * @gh: the holder structure
409  *
410  */
411
412 void gfs2_holder_uninit(struct gfs2_holder *gh)
413 {
414         gfs2_glock_put(gh->gh_gl);
415         gh->gh_gl = NULL;
416         gh->gh_ip = 0;
417 }
418
419 /**
420  * gfs2_holder_get - get a struct gfs2_holder structure
421  * @gl: the glock
422  * @state: the state we're requesting
423  * @flags: the modifier flags
424  * @gfp_flags:
425  *
426  * Figure out how big an impact this function has.  Either:
427  * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
428  * 2) Leave it like it is
429  *
430  * Returns: the holder structure, NULL on ENOMEM
431  */
432
433 static struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl,
434                                            unsigned int state,
435                                            int flags, gfp_t gfp_flags)
436 {
437         struct gfs2_holder *gh;
438
439         gh = kmalloc(sizeof(struct gfs2_holder), gfp_flags);
440         if (!gh)
441                 return NULL;
442
443         gfs2_holder_init(gl, state, flags, gh);
444         set_bit(HIF_ALLOCED, &gh->gh_iflags);
445         gh->gh_ip = (unsigned long)__builtin_return_address(0);
446         return gh;
447 }
448
449 /**
450  * gfs2_holder_put - get rid of a struct gfs2_holder structure
451  * @gh: the holder structure
452  *
453  */
454
455 static void gfs2_holder_put(struct gfs2_holder *gh)
456 {
457         gfs2_holder_uninit(gh);
458         kfree(gh);
459 }
460
461 static void gfs2_holder_dispose_or_wake(struct gfs2_holder *gh)
462 {
463         if (test_bit(HIF_DEALLOC, &gh->gh_iflags)) {
464                 gfs2_holder_put(gh);
465                 return;
466         }
467         clear_bit(HIF_WAIT, &gh->gh_iflags);
468         smp_mb();
469         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
470 }
471
472 static int holder_wait(void *word)
473 {
474         schedule();
475         return 0;
476 }
477
478 static void wait_on_holder(struct gfs2_holder *gh)
479 {
480         might_sleep();
481         wait_on_bit(&gh->gh_iflags, HIF_WAIT, holder_wait, TASK_UNINTERRUPTIBLE);
482 }
483
484 /**
485  * rq_mutex - process a mutex request in the queue
486  * @gh: the glock holder
487  *
488  * Returns: 1 if the queue is blocked
489  */
490
491 static int rq_mutex(struct gfs2_holder *gh)
492 {
493         struct gfs2_glock *gl = gh->gh_gl;
494
495         list_del_init(&gh->gh_list);
496         /*  gh->gh_error never examined.  */
497         set_bit(GLF_LOCK, &gl->gl_flags);
498         clear_bit(HIF_WAIT, &gh->gh_iflags);
499         smp_mb();
500         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
501
502         return 1;
503 }
504
505 /**
506  * rq_promote - process a promote request in the queue
507  * @gh: the glock holder
508  *
509  * Acquire a new inter-node lock, or change a lock state to more restrictive.
510  *
511  * Returns: 1 if the queue is blocked
512  */
513
514 static int rq_promote(struct gfs2_holder *gh)
515 {
516         struct gfs2_glock *gl = gh->gh_gl;
517         struct gfs2_sbd *sdp = gl->gl_sbd;
518
519         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
520                 if (list_empty(&gl->gl_holders)) {
521                         gl->gl_req_gh = gh;
522                         set_bit(GLF_LOCK, &gl->gl_flags);
523                         spin_unlock(&gl->gl_spin);
524
525                         if (atomic_read(&sdp->sd_reclaim_count) >
526                             gfs2_tune_get(sdp, gt_reclaim_limit) &&
527                             !(gh->gh_flags & LM_FLAG_PRIORITY)) {
528                                 gfs2_reclaim_glock(sdp);
529                                 gfs2_reclaim_glock(sdp);
530                         }
531
532                         gfs2_glock_xmote_th(gh);
533                         spin_lock(&gl->gl_spin);
534                 }
535                 return 1;
536         }
537
538         if (list_empty(&gl->gl_holders)) {
539                 set_bit(HIF_FIRST, &gh->gh_iflags);
540                 set_bit(GLF_LOCK, &gl->gl_flags);
541         } else {
542                 struct gfs2_holder *next_gh;
543                 if (gh->gh_state == LM_ST_EXCLUSIVE)
544                         return 1;
545                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
546                                      gh_list);
547                 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
548                          return 1;
549         }
550
551         list_move_tail(&gh->gh_list, &gl->gl_holders);
552         gh->gh_error = 0;
553         set_bit(HIF_HOLDER, &gh->gh_iflags);
554
555         gfs2_holder_dispose_or_wake(gh);
556
557         return 0;
558 }
559
560 /**
561  * rq_demote - process a demote request in the queue
562  * @gh: the glock holder
563  *
564  * Returns: 1 if the queue is blocked
565  */
566
567 static int rq_demote(struct gfs2_holder *gh)
568 {
569         struct gfs2_glock *gl = gh->gh_gl;
570
571         if (!list_empty(&gl->gl_holders))
572                 return 1;
573
574         if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
575                 list_del_init(&gh->gh_list);
576                 gh->gh_error = 0;
577                 spin_unlock(&gl->gl_spin);
578                 gfs2_holder_dispose_or_wake(gh);
579                 spin_lock(&gl->gl_spin);
580         } else {
581                 gl->gl_req_gh = gh;
582                 set_bit(GLF_LOCK, &gl->gl_flags);
583                 spin_unlock(&gl->gl_spin);
584
585                 if (gh->gh_state == LM_ST_UNLOCKED ||
586                     gl->gl_state != LM_ST_EXCLUSIVE)
587                         gfs2_glock_drop_th(gl);
588                 else
589                         gfs2_glock_xmote_th(gh);
590
591                 spin_lock(&gl->gl_spin);
592         }
593
594         return 0;
595 }
596
597 /**
598  * run_queue - process holder structures on a glock
599  * @gl: the glock
600  *
601  */
602 static void run_queue(struct gfs2_glock *gl)
603 {
604         struct gfs2_holder *gh;
605         int blocked = 1;
606
607         for (;;) {
608                 if (test_bit(GLF_LOCK, &gl->gl_flags))
609                         break;
610
611                 if (!list_empty(&gl->gl_waiters1)) {
612                         gh = list_entry(gl->gl_waiters1.next,
613                                         struct gfs2_holder, gh_list);
614
615                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
616                                 blocked = rq_mutex(gh);
617                         else
618                                 gfs2_assert_warn(gl->gl_sbd, 0);
619
620                 } else if (!list_empty(&gl->gl_waiters2) &&
621                            !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
622                         gh = list_entry(gl->gl_waiters2.next,
623                                         struct gfs2_holder, gh_list);
624
625                         if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
626                                 blocked = rq_demote(gh);
627                         else
628                                 gfs2_assert_warn(gl->gl_sbd, 0);
629
630                 } else if (!list_empty(&gl->gl_waiters3)) {
631                         gh = list_entry(gl->gl_waiters3.next,
632                                         struct gfs2_holder, gh_list);
633
634                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
635                                 blocked = rq_promote(gh);
636                         else
637                                 gfs2_assert_warn(gl->gl_sbd, 0);
638
639                 } else
640                         break;
641
642                 if (blocked)
643                         break;
644         }
645 }
646
647 /**
648  * gfs2_glmutex_lock - acquire a local lock on a glock
649  * @gl: the glock
650  *
651  * Gives caller exclusive access to manipulate a glock structure.
652  */
653
654 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
655 {
656         struct gfs2_holder gh;
657
658         gfs2_holder_init(gl, 0, 0, &gh);
659         set_bit(HIF_MUTEX, &gh.gh_iflags);
660         if (test_and_set_bit(HIF_WAIT, &gh.gh_iflags))
661                 BUG();
662
663         spin_lock(&gl->gl_spin);
664         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
665                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
666         } else {
667                 gl->gl_owner = current;
668                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
669                 clear_bit(HIF_WAIT, &gh.gh_iflags);
670                 smp_mb();
671                 wake_up_bit(&gh.gh_iflags, HIF_WAIT);
672         }
673         spin_unlock(&gl->gl_spin);
674
675         wait_on_holder(&gh);
676         gfs2_holder_uninit(&gh);
677 }
678
679 /**
680  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
681  * @gl: the glock
682  *
683  * Returns: 1 if the glock is acquired
684  */
685
686 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
687 {
688         int acquired = 1;
689
690         spin_lock(&gl->gl_spin);
691         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
692                 acquired = 0;
693         } else {
694                 gl->gl_owner = current;
695                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
696         }
697         spin_unlock(&gl->gl_spin);
698
699         return acquired;
700 }
701
702 /**
703  * gfs2_glmutex_unlock - release a local lock on a glock
704  * @gl: the glock
705  *
706  */
707
708 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
709 {
710         spin_lock(&gl->gl_spin);
711         clear_bit(GLF_LOCK, &gl->gl_flags);
712         gl->gl_owner = NULL;
713         gl->gl_ip = 0;
714         run_queue(gl);
715         BUG_ON(!spin_is_locked(&gl->gl_spin));
716         spin_unlock(&gl->gl_spin);
717 }
718
719 /**
720  * handle_callback - add a demote request to a lock's queue
721  * @gl: the glock
722  * @state: the state the caller wants us to change to
723  *
724  * Note: This may fail sliently if we are out of memory.
725  */
726
727 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
728 {
729         struct gfs2_holder *gh, *new_gh = NULL;
730
731 restart:
732         spin_lock(&gl->gl_spin);
733
734         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
735                 if (test_bit(HIF_DEMOTE, &gh->gh_iflags) &&
736                     gl->gl_req_gh != gh) {
737                         if (gh->gh_state != state)
738                                 gh->gh_state = LM_ST_UNLOCKED;
739                         goto out;
740                 }
741         }
742
743         if (new_gh) {
744                 list_add_tail(&new_gh->gh_list, &gl->gl_waiters2);
745                 new_gh = NULL;
746         } else {
747                 spin_unlock(&gl->gl_spin);
748
749                 new_gh = gfs2_holder_get(gl, state, LM_FLAG_TRY, GFP_NOFS);
750                 if (!new_gh)
751                         return;
752                 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
753                 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
754                 set_bit(HIF_WAIT, &new_gh->gh_iflags);
755
756                 goto restart;
757         }
758
759 out:
760         spin_unlock(&gl->gl_spin);
761
762         if (new_gh)
763                 gfs2_holder_put(new_gh);
764 }
765
766 /**
767  * state_change - record that the glock is now in a different state
768  * @gl: the glock
769  * @new_state the new state
770  *
771  */
772
773 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
774 {
775         int held1, held2;
776
777         held1 = (gl->gl_state != LM_ST_UNLOCKED);
778         held2 = (new_state != LM_ST_UNLOCKED);
779
780         if (held1 != held2) {
781                 if (held2)
782                         gfs2_glock_hold(gl);
783                 else
784                         gfs2_glock_put(gl);
785         }
786
787         gl->gl_state = new_state;
788 }
789
790 /**
791  * xmote_bh - Called after the lock module is done acquiring a lock
792  * @gl: The glock in question
793  * @ret: the int returned from the lock module
794  *
795  */
796
797 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
798 {
799         struct gfs2_sbd *sdp = gl->gl_sbd;
800         const struct gfs2_glock_operations *glops = gl->gl_ops;
801         struct gfs2_holder *gh = gl->gl_req_gh;
802         int prev_state = gl->gl_state;
803         int op_done = 1;
804
805         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
806         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
807         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
808
809         state_change(gl, ret & LM_OUT_ST_MASK);
810
811         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
812                 if (glops->go_inval)
813                         glops->go_inval(gl, DIO_METADATA);
814         } else if (gl->gl_state == LM_ST_DEFERRED) {
815                 /* We might not want to do this here.
816                    Look at moving to the inode glops. */
817                 if (glops->go_inval)
818                         glops->go_inval(gl, 0);
819         }
820
821         /*  Deal with each possible exit condition  */
822
823         if (!gh)
824                 gl->gl_stamp = jiffies;
825         else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
826                 spin_lock(&gl->gl_spin);
827                 list_del_init(&gh->gh_list);
828                 gh->gh_error = -EIO;
829                 spin_unlock(&gl->gl_spin);
830         } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
831                 spin_lock(&gl->gl_spin);
832                 list_del_init(&gh->gh_list);
833                 if (gl->gl_state == gh->gh_state ||
834                     gl->gl_state == LM_ST_UNLOCKED) {
835                         gh->gh_error = 0;
836                 } else {
837                         if (gfs2_assert_warn(sdp, gh->gh_flags &
838                                         (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
839                                 fs_warn(sdp, "ret = 0x%.8X\n", ret);
840                         gh->gh_error = GLR_TRYFAILED;
841                 }
842                 spin_unlock(&gl->gl_spin);
843
844                 if (ret & LM_OUT_CANCELED)
845                         handle_callback(gl, LM_ST_UNLOCKED);
846
847         } else if (ret & LM_OUT_CANCELED) {
848                 spin_lock(&gl->gl_spin);
849                 list_del_init(&gh->gh_list);
850                 gh->gh_error = GLR_CANCELED;
851                 spin_unlock(&gl->gl_spin);
852
853         } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
854                 spin_lock(&gl->gl_spin);
855                 list_move_tail(&gh->gh_list, &gl->gl_holders);
856                 gh->gh_error = 0;
857                 set_bit(HIF_HOLDER, &gh->gh_iflags);
858                 spin_unlock(&gl->gl_spin);
859
860                 set_bit(HIF_FIRST, &gh->gh_iflags);
861
862                 op_done = 0;
863
864         } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
865                 spin_lock(&gl->gl_spin);
866                 list_del_init(&gh->gh_list);
867                 gh->gh_error = GLR_TRYFAILED;
868                 spin_unlock(&gl->gl_spin);
869
870         } else {
871                 if (gfs2_assert_withdraw(sdp, 0) == -1)
872                         fs_err(sdp, "ret = 0x%.8X\n", ret);
873         }
874
875         if (glops->go_xmote_bh)
876                 glops->go_xmote_bh(gl);
877
878         if (op_done) {
879                 spin_lock(&gl->gl_spin);
880                 gl->gl_req_gh = NULL;
881                 gl->gl_req_bh = NULL;
882                 clear_bit(GLF_LOCK, &gl->gl_flags);
883                 run_queue(gl);
884                 spin_unlock(&gl->gl_spin);
885         }
886
887         gfs2_glock_put(gl);
888
889         if (gh)
890                 gfs2_holder_dispose_or_wake(gh);
891 }
892
893 /**
894  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
895  * @gl: The glock in question
896  * @state: the requested state
897  * @flags: modifier flags to the lock call
898  *
899  */
900
901 void gfs2_glock_xmote_th(struct gfs2_holder *gh)
902 {
903         struct gfs2_glock *gl = gh->gh_gl;
904         struct gfs2_sbd *sdp = gl->gl_sbd;
905         int flags = gh->gh_flags;
906         unsigned state = gh->gh_state;
907         const struct gfs2_glock_operations *glops = gl->gl_ops;
908         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
909                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
910                                  LM_FLAG_PRIORITY);
911         unsigned int lck_ret;
912
913         if (glops->go_xmote_th)
914                 glops->go_xmote_th(gl);
915
916         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
917         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
918         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
919         gfs2_assert_warn(sdp, state != gl->gl_state);
920
921         gfs2_glock_hold(gl);
922         gl->gl_req_bh = xmote_bh;
923
924         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
925
926         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
927                 return;
928
929         if (lck_ret & LM_OUT_ASYNC)
930                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
931         else
932                 xmote_bh(gl, lck_ret);
933 }
934
935 /**
936  * drop_bh - Called after a lock module unlock completes
937  * @gl: the glock
938  * @ret: the return status
939  *
940  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
941  * Doesn't drop the reference on the glock the top half took out
942  *
943  */
944
945 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
946 {
947         struct gfs2_sbd *sdp = gl->gl_sbd;
948         const struct gfs2_glock_operations *glops = gl->gl_ops;
949         struct gfs2_holder *gh = gl->gl_req_gh;
950
951         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
952         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
953         gfs2_assert_warn(sdp, !ret);
954
955         state_change(gl, LM_ST_UNLOCKED);
956
957         if (glops->go_inval)
958                 glops->go_inval(gl, DIO_METADATA);
959
960         if (gh) {
961                 spin_lock(&gl->gl_spin);
962                 list_del_init(&gh->gh_list);
963                 gh->gh_error = 0;
964                 spin_unlock(&gl->gl_spin);
965         }
966
967         spin_lock(&gl->gl_spin);
968         gl->gl_req_gh = NULL;
969         gl->gl_req_bh = NULL;
970         clear_bit(GLF_LOCK, &gl->gl_flags);
971         run_queue(gl);
972         spin_unlock(&gl->gl_spin);
973
974         gfs2_glock_put(gl);
975
976         if (gh)
977                 gfs2_holder_dispose_or_wake(gh);
978 }
979
980 /**
981  * gfs2_glock_drop_th - call into the lock module to unlock a lock
982  * @gl: the glock
983  *
984  */
985
986 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
987 {
988         struct gfs2_sbd *sdp = gl->gl_sbd;
989         const struct gfs2_glock_operations *glops = gl->gl_ops;
990         unsigned int ret;
991
992         if (glops->go_drop_th)
993                 glops->go_drop_th(gl);
994
995         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
996         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
997         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
998
999         gfs2_glock_hold(gl);
1000         gl->gl_req_bh = drop_bh;
1001
1002         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1003
1004         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1005                 return;
1006
1007         if (!ret)
1008                 drop_bh(gl, ret);
1009         else
1010                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1011 }
1012
1013 /**
1014  * do_cancels - cancel requests for locks stuck waiting on an expire flag
1015  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1016  *
1017  * Don't cancel GL_NOCANCEL requests.
1018  */
1019
1020 static void do_cancels(struct gfs2_holder *gh)
1021 {
1022         struct gfs2_glock *gl = gh->gh_gl;
1023
1024         spin_lock(&gl->gl_spin);
1025
1026         while (gl->gl_req_gh != gh &&
1027                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1028                !list_empty(&gh->gh_list)) {
1029                 if (gl->gl_req_bh && !(gl->gl_req_gh &&
1030                                      (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1031                         spin_unlock(&gl->gl_spin);
1032                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1033                         msleep(100);
1034                         spin_lock(&gl->gl_spin);
1035                 } else {
1036                         spin_unlock(&gl->gl_spin);
1037                         msleep(100);
1038                         spin_lock(&gl->gl_spin);
1039                 }
1040         }
1041
1042         spin_unlock(&gl->gl_spin);
1043 }
1044
1045 /**
1046  * glock_wait_internal - wait on a glock acquisition
1047  * @gh: the glock holder
1048  *
1049  * Returns: 0 on success
1050  */
1051
1052 static int glock_wait_internal(struct gfs2_holder *gh)
1053 {
1054         struct gfs2_glock *gl = gh->gh_gl;
1055         struct gfs2_sbd *sdp = gl->gl_sbd;
1056         const struct gfs2_glock_operations *glops = gl->gl_ops;
1057
1058         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1059                 return -EIO;
1060
1061         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1062                 spin_lock(&gl->gl_spin);
1063                 if (gl->gl_req_gh != gh &&
1064                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1065                     !list_empty(&gh->gh_list)) {
1066                         list_del_init(&gh->gh_list);
1067                         gh->gh_error = GLR_TRYFAILED;
1068                         run_queue(gl);
1069                         spin_unlock(&gl->gl_spin);
1070                         return gh->gh_error;
1071                 }
1072                 spin_unlock(&gl->gl_spin);
1073         }
1074
1075         if (gh->gh_flags & LM_FLAG_PRIORITY)
1076                 do_cancels(gh);
1077
1078         wait_on_holder(gh);
1079         if (gh->gh_error)
1080                 return gh->gh_error;
1081
1082         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1083         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1084                                                    gh->gh_flags));
1085
1086         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1087                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1088
1089                 if (glops->go_lock) {
1090                         gh->gh_error = glops->go_lock(gh);
1091                         if (gh->gh_error) {
1092                                 spin_lock(&gl->gl_spin);
1093                                 list_del_init(&gh->gh_list);
1094                                 spin_unlock(&gl->gl_spin);
1095                         }
1096                 }
1097
1098                 spin_lock(&gl->gl_spin);
1099                 gl->gl_req_gh = NULL;
1100                 gl->gl_req_bh = NULL;
1101                 clear_bit(GLF_LOCK, &gl->gl_flags);
1102                 run_queue(gl);
1103                 spin_unlock(&gl->gl_spin);
1104         }
1105
1106         return gh->gh_error;
1107 }
1108
1109 static inline struct gfs2_holder *
1110 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1111 {
1112         struct gfs2_holder *gh;
1113
1114         list_for_each_entry(gh, head, gh_list) {
1115                 if (gh->gh_owner == owner)
1116                         return gh;
1117         }
1118
1119         return NULL;
1120 }
1121
1122 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1123 {
1124         va_list args;
1125
1126         va_start(args, fmt);
1127         if (gi) {
1128                 vsprintf(gi->string, fmt, args);
1129                 seq_printf(gi->seq, gi->string);
1130         }
1131         else
1132                 vprintk(fmt, args);
1133         va_end(args);
1134 }
1135
1136 /**
1137  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1138  * @gh: the holder structure to add
1139  *
1140  */
1141
1142 static void add_to_queue(struct gfs2_holder *gh)
1143 {
1144         struct gfs2_glock *gl = gh->gh_gl;
1145         struct gfs2_holder *existing;
1146
1147         BUG_ON(!gh->gh_owner);
1148         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1149                 BUG();
1150
1151         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1152         if (existing) {
1153                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1154                 printk(KERN_INFO "pid : %d\n", existing->gh_owner->pid);
1155                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1156                                 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1157                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1158                 printk(KERN_INFO "pid : %d\n", gh->gh_owner->pid);
1159                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1160                                 gl->gl_name.ln_type, gl->gl_state);
1161                 BUG();
1162         }
1163
1164         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1165         if (existing) {
1166                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1167                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1168                 BUG();
1169         }
1170
1171         if (gh->gh_flags & LM_FLAG_PRIORITY)
1172                 list_add(&gh->gh_list, &gl->gl_waiters3);
1173         else
1174                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1175 }
1176
1177 /**
1178  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1179  * @gh: the holder structure
1180  *
1181  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1182  *
1183  * Returns: 0, GLR_TRYFAILED, or errno on failure
1184  */
1185
1186 int gfs2_glock_nq(struct gfs2_holder *gh)
1187 {
1188         struct gfs2_glock *gl = gh->gh_gl;
1189         struct gfs2_sbd *sdp = gl->gl_sbd;
1190         int error = 0;
1191
1192 restart:
1193         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1194                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1195                 return -EIO;
1196         }
1197
1198         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1199
1200         spin_lock(&gl->gl_spin);
1201         add_to_queue(gh);
1202         run_queue(gl);
1203         spin_unlock(&gl->gl_spin);
1204
1205         if (!(gh->gh_flags & GL_ASYNC)) {
1206                 error = glock_wait_internal(gh);
1207                 if (error == GLR_CANCELED) {
1208                         msleep(100);
1209                         goto restart;
1210                 }
1211         }
1212
1213         return error;
1214 }
1215
1216 /**
1217  * gfs2_glock_poll - poll to see if an async request has been completed
1218  * @gh: the holder
1219  *
1220  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1221  */
1222
1223 int gfs2_glock_poll(struct gfs2_holder *gh)
1224 {
1225         struct gfs2_glock *gl = gh->gh_gl;
1226         int ready = 0;
1227
1228         spin_lock(&gl->gl_spin);
1229
1230         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1231                 ready = 1;
1232         else if (list_empty(&gh->gh_list)) {
1233                 if (gh->gh_error == GLR_CANCELED) {
1234                         spin_unlock(&gl->gl_spin);
1235                         msleep(100);
1236                         if (gfs2_glock_nq(gh))
1237                                 return 1;
1238                         return 0;
1239                 } else
1240                         ready = 1;
1241         }
1242
1243         spin_unlock(&gl->gl_spin);
1244
1245         return ready;
1246 }
1247
1248 /**
1249  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1250  * @gh: the holder structure
1251  *
1252  * Returns: 0, GLR_TRYFAILED, or errno on failure
1253  */
1254
1255 int gfs2_glock_wait(struct gfs2_holder *gh)
1256 {
1257         int error;
1258
1259         error = glock_wait_internal(gh);
1260         if (error == GLR_CANCELED) {
1261                 msleep(100);
1262                 gh->gh_flags &= ~GL_ASYNC;
1263                 error = gfs2_glock_nq(gh);
1264         }
1265
1266         return error;
1267 }
1268
1269 /**
1270  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1271  * @gh: the glock holder
1272  *
1273  */
1274
1275 void gfs2_glock_dq(struct gfs2_holder *gh)
1276 {
1277         struct gfs2_glock *gl = gh->gh_gl;
1278         const struct gfs2_glock_operations *glops = gl->gl_ops;
1279
1280         if (gh->gh_flags & GL_NOCACHE)
1281                 handle_callback(gl, LM_ST_UNLOCKED);
1282
1283         gfs2_glmutex_lock(gl);
1284
1285         spin_lock(&gl->gl_spin);
1286         list_del_init(&gh->gh_list);
1287
1288         if (list_empty(&gl->gl_holders)) {
1289                 spin_unlock(&gl->gl_spin);
1290
1291                 if (glops->go_unlock)
1292                         glops->go_unlock(gh);
1293
1294                 gl->gl_stamp = jiffies;
1295
1296                 spin_lock(&gl->gl_spin);
1297         }
1298
1299         clear_bit(GLF_LOCK, &gl->gl_flags);
1300         run_queue(gl);
1301         spin_unlock(&gl->gl_spin);
1302 }
1303
1304 /**
1305  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1306  * @gh: the holder structure
1307  *
1308  */
1309
1310 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1311 {
1312         gfs2_glock_dq(gh);
1313         gfs2_holder_uninit(gh);
1314 }
1315
1316 /**
1317  * gfs2_glock_nq_num - acquire a glock based on lock number
1318  * @sdp: the filesystem
1319  * @number: the lock number
1320  * @glops: the glock operations for the type of glock
1321  * @state: the state to acquire the glock in
1322  * @flags: modifier flags for the aquisition
1323  * @gh: the struct gfs2_holder
1324  *
1325  * Returns: errno
1326  */
1327
1328 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1329                       const struct gfs2_glock_operations *glops,
1330                       unsigned int state, int flags, struct gfs2_holder *gh)
1331 {
1332         struct gfs2_glock *gl;
1333         int error;
1334
1335         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1336         if (!error) {
1337                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1338                 gfs2_glock_put(gl);
1339         }
1340
1341         return error;
1342 }
1343
1344 /**
1345  * glock_compare - Compare two struct gfs2_glock structures for sorting
1346  * @arg_a: the first structure
1347  * @arg_b: the second structure
1348  *
1349  */
1350
1351 static int glock_compare(const void *arg_a, const void *arg_b)
1352 {
1353         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1354         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1355         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1356         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1357
1358         if (a->ln_number > b->ln_number)
1359                 return 1;
1360         if (a->ln_number < b->ln_number)
1361                 return -1;
1362         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1363         return 0;
1364 }
1365
1366 /**
1367  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1368  * @num_gh: the number of structures
1369  * @ghs: an array of struct gfs2_holder structures
1370  *
1371  * Returns: 0 on success (all glocks acquired),
1372  *          errno on failure (no glocks acquired)
1373  */
1374
1375 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1376                      struct gfs2_holder **p)
1377 {
1378         unsigned int x;
1379         int error = 0;
1380
1381         for (x = 0; x < num_gh; x++)
1382                 p[x] = &ghs[x];
1383
1384         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1385
1386         for (x = 0; x < num_gh; x++) {
1387                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1388
1389                 error = gfs2_glock_nq(p[x]);
1390                 if (error) {
1391                         while (x--)
1392                                 gfs2_glock_dq(p[x]);
1393                         break;
1394                 }
1395         }
1396
1397         return error;
1398 }
1399
1400 /**
1401  * gfs2_glock_nq_m - acquire multiple glocks
1402  * @num_gh: the number of structures
1403  * @ghs: an array of struct gfs2_holder structures
1404  *
1405  * Figure out how big an impact this function has.  Either:
1406  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1407  * 2) Forget async stuff and just call nq_m_sync()
1408  * 3) Leave it like it is
1409  *
1410  * Returns: 0 on success (all glocks acquired),
1411  *          errno on failure (no glocks acquired)
1412  */
1413
1414 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1415 {
1416         int *e;
1417         unsigned int x;
1418         int borked = 0, serious = 0;
1419         int error = 0;
1420
1421         if (!num_gh)
1422                 return 0;
1423
1424         if (num_gh == 1) {
1425                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1426                 return gfs2_glock_nq(ghs);
1427         }
1428
1429         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1430         if (!e)
1431                 return -ENOMEM;
1432
1433         for (x = 0; x < num_gh; x++) {
1434                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1435                 error = gfs2_glock_nq(&ghs[x]);
1436                 if (error) {
1437                         borked = 1;
1438                         serious = error;
1439                         num_gh = x;
1440                         break;
1441                 }
1442         }
1443
1444         for (x = 0; x < num_gh; x++) {
1445                 error = e[x] = glock_wait_internal(&ghs[x]);
1446                 if (error) {
1447                         borked = 1;
1448                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1449                                 serious = error;
1450                 }
1451         }
1452
1453         if (!borked) {
1454                 kfree(e);
1455                 return 0;
1456         }
1457
1458         for (x = 0; x < num_gh; x++)
1459                 if (!e[x])
1460                         gfs2_glock_dq(&ghs[x]);
1461
1462         if (serious)
1463                 error = serious;
1464         else {
1465                 for (x = 0; x < num_gh; x++)
1466                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1467                                           &ghs[x]);
1468                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1469         }
1470
1471         kfree(e);
1472
1473         return error;
1474 }
1475
1476 /**
1477  * gfs2_glock_dq_m - release multiple glocks
1478  * @num_gh: the number of structures
1479  * @ghs: an array of struct gfs2_holder structures
1480  *
1481  */
1482
1483 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1484 {
1485         unsigned int x;
1486
1487         for (x = 0; x < num_gh; x++)
1488                 gfs2_glock_dq(&ghs[x]);
1489 }
1490
1491 /**
1492  * gfs2_glock_dq_uninit_m - release multiple glocks
1493  * @num_gh: the number of structures
1494  * @ghs: an array of struct gfs2_holder structures
1495  *
1496  */
1497
1498 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1499 {
1500         unsigned int x;
1501
1502         for (x = 0; x < num_gh; x++)
1503                 gfs2_glock_dq_uninit(&ghs[x]);
1504 }
1505
1506 /**
1507  * gfs2_lvb_hold - attach a LVB from a glock
1508  * @gl: The glock in question
1509  *
1510  */
1511
1512 int gfs2_lvb_hold(struct gfs2_glock *gl)
1513 {
1514         int error;
1515
1516         gfs2_glmutex_lock(gl);
1517
1518         if (!atomic_read(&gl->gl_lvb_count)) {
1519                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1520                 if (error) {
1521                         gfs2_glmutex_unlock(gl);
1522                         return error;
1523                 }
1524                 gfs2_glock_hold(gl);
1525         }
1526         atomic_inc(&gl->gl_lvb_count);
1527
1528         gfs2_glmutex_unlock(gl);
1529
1530         return 0;
1531 }
1532
1533 /**
1534  * gfs2_lvb_unhold - detach a LVB from a glock
1535  * @gl: The glock in question
1536  *
1537  */
1538
1539 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1540 {
1541         gfs2_glock_hold(gl);
1542         gfs2_glmutex_lock(gl);
1543
1544         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1545         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1546                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1547                 gl->gl_lvb = NULL;
1548                 gfs2_glock_put(gl);
1549         }
1550
1551         gfs2_glmutex_unlock(gl);
1552         gfs2_glock_put(gl);
1553 }
1554
1555 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1556                         unsigned int state)
1557 {
1558         struct gfs2_glock *gl;
1559
1560         gl = gfs2_glock_find(sdp, name);
1561         if (!gl)
1562                 return;
1563
1564         handle_callback(gl, state);
1565
1566         spin_lock(&gl->gl_spin);
1567         run_queue(gl);
1568         spin_unlock(&gl->gl_spin);
1569
1570         gfs2_glock_put(gl);
1571 }
1572
1573 /**
1574  * gfs2_glock_cb - Callback used by locking module
1575  * @sdp: Pointer to the superblock
1576  * @type: Type of callback
1577  * @data: Type dependent data pointer
1578  *
1579  * Called by the locking module when it wants to tell us something.
1580  * Either we need to drop a lock, one of our ASYNC requests completed, or
1581  * a journal from another client needs to be recovered.
1582  */
1583
1584 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1585 {
1586         struct gfs2_sbd *sdp = cb_data;
1587
1588         switch (type) {
1589         case LM_CB_NEED_E:
1590                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1591                 return;
1592
1593         case LM_CB_NEED_D:
1594                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1595                 return;
1596
1597         case LM_CB_NEED_S:
1598                 blocking_cb(sdp, data, LM_ST_SHARED);
1599                 return;
1600
1601         case LM_CB_ASYNC: {
1602                 struct lm_async_cb *async = data;
1603                 struct gfs2_glock *gl;
1604
1605                 down_read(&gfs2_umount_flush_sem);
1606                 gl = gfs2_glock_find(sdp, &async->lc_name);
1607                 if (gfs2_assert_warn(sdp, gl))
1608                         return;
1609                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1610                         gl->gl_req_bh(gl, async->lc_ret);
1611                 gfs2_glock_put(gl);
1612                 up_read(&gfs2_umount_flush_sem);
1613                 return;
1614         }
1615
1616         case LM_CB_NEED_RECOVERY:
1617                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1618                 if (sdp->sd_recoverd_process)
1619                         wake_up_process(sdp->sd_recoverd_process);
1620                 return;
1621
1622         case LM_CB_DROPLOCKS:
1623                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1624                 gfs2_quota_scan(sdp);
1625                 return;
1626
1627         default:
1628                 gfs2_assert_warn(sdp, 0);
1629                 return;
1630         }
1631 }
1632
1633 /**
1634  * demote_ok - Check to see if it's ok to unlock a glock
1635  * @gl: the glock
1636  *
1637  * Returns: 1 if it's ok
1638  */
1639
1640 static int demote_ok(struct gfs2_glock *gl)
1641 {
1642         const struct gfs2_glock_operations *glops = gl->gl_ops;
1643         int demote = 1;
1644
1645         if (test_bit(GLF_STICKY, &gl->gl_flags))
1646                 demote = 0;
1647         else if (glops->go_demote_ok)
1648                 demote = glops->go_demote_ok(gl);
1649
1650         return demote;
1651 }
1652
1653 /**
1654  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1655  * @gl: the glock
1656  *
1657  */
1658
1659 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1660 {
1661         struct gfs2_sbd *sdp = gl->gl_sbd;
1662
1663         spin_lock(&sdp->sd_reclaim_lock);
1664         if (list_empty(&gl->gl_reclaim)) {
1665                 gfs2_glock_hold(gl);
1666                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1667                 atomic_inc(&sdp->sd_reclaim_count);
1668         }
1669         spin_unlock(&sdp->sd_reclaim_lock);
1670
1671         wake_up(&sdp->sd_reclaim_wq);
1672 }
1673
1674 /**
1675  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1676  * @sdp: the filesystem
1677  *
1678  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1679  * different glock and we notice that there are a lot of glocks in the
1680  * reclaim list.
1681  *
1682  */
1683
1684 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1685 {
1686         struct gfs2_glock *gl;
1687
1688         spin_lock(&sdp->sd_reclaim_lock);
1689         if (list_empty(&sdp->sd_reclaim_list)) {
1690                 spin_unlock(&sdp->sd_reclaim_lock);
1691                 return;
1692         }
1693         gl = list_entry(sdp->sd_reclaim_list.next,
1694                         struct gfs2_glock, gl_reclaim);
1695         list_del_init(&gl->gl_reclaim);
1696         spin_unlock(&sdp->sd_reclaim_lock);
1697
1698         atomic_dec(&sdp->sd_reclaim_count);
1699         atomic_inc(&sdp->sd_reclaimed);
1700
1701         if (gfs2_glmutex_trylock(gl)) {
1702                 if (list_empty(&gl->gl_holders) &&
1703                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1704                         handle_callback(gl, LM_ST_UNLOCKED);
1705                 gfs2_glmutex_unlock(gl);
1706         }
1707
1708         gfs2_glock_put(gl);
1709 }
1710
1711 /**
1712  * examine_bucket - Call a function for glock in a hash bucket
1713  * @examiner: the function
1714  * @sdp: the filesystem
1715  * @bucket: the bucket
1716  *
1717  * Returns: 1 if the bucket has entries
1718  */
1719
1720 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1721                           unsigned int hash)
1722 {
1723         struct gfs2_glock *gl, *prev = NULL;
1724         int has_entries = 0;
1725         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1726
1727         read_lock(gl_lock_addr(hash));
1728         /* Can't use hlist_for_each_entry - don't want prefetch here */
1729         if (hlist_empty(head))
1730                 goto out;
1731         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1732         while(1) {
1733                 if (gl->gl_sbd == sdp) {
1734                         gfs2_glock_hold(gl);
1735                         read_unlock(gl_lock_addr(hash));
1736                         if (prev)
1737                                 gfs2_glock_put(prev);
1738                         prev = gl;
1739                         examiner(gl);
1740                         has_entries = 1;
1741                         read_lock(gl_lock_addr(hash));
1742                 }
1743                 if (gl->gl_list.next == NULL)
1744                         break;
1745                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1746         }
1747 out:
1748         read_unlock(gl_lock_addr(hash));
1749         if (prev)
1750                 gfs2_glock_put(prev);
1751         return has_entries;
1752 }
1753
1754 /**
1755  * scan_glock - look at a glock and see if we can reclaim it
1756  * @gl: the glock to look at
1757  *
1758  */
1759
1760 static void scan_glock(struct gfs2_glock *gl)
1761 {
1762         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1763                 return;
1764
1765         if (gfs2_glmutex_trylock(gl)) {
1766                 if (list_empty(&gl->gl_holders) &&
1767                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1768                         goto out_schedule;
1769                 gfs2_glmutex_unlock(gl);
1770         }
1771         return;
1772
1773 out_schedule:
1774         gfs2_glmutex_unlock(gl);
1775         gfs2_glock_schedule_for_reclaim(gl);
1776 }
1777
1778 /**
1779  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1780  * @sdp: the filesystem
1781  *
1782  */
1783
1784 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1785 {
1786         unsigned int x;
1787
1788         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1789                 examine_bucket(scan_glock, sdp, x);
1790 }
1791
1792 /**
1793  * clear_glock - look at a glock and see if we can free it from glock cache
1794  * @gl: the glock to look at
1795  *
1796  */
1797
1798 static void clear_glock(struct gfs2_glock *gl)
1799 {
1800         struct gfs2_sbd *sdp = gl->gl_sbd;
1801         int released;
1802
1803         spin_lock(&sdp->sd_reclaim_lock);
1804         if (!list_empty(&gl->gl_reclaim)) {
1805                 list_del_init(&gl->gl_reclaim);
1806                 atomic_dec(&sdp->sd_reclaim_count);
1807                 spin_unlock(&sdp->sd_reclaim_lock);
1808                 released = gfs2_glock_put(gl);
1809                 gfs2_assert(sdp, !released);
1810         } else {
1811                 spin_unlock(&sdp->sd_reclaim_lock);
1812         }
1813
1814         if (gfs2_glmutex_trylock(gl)) {
1815                 if (list_empty(&gl->gl_holders) &&
1816                     gl->gl_state != LM_ST_UNLOCKED)
1817                         handle_callback(gl, LM_ST_UNLOCKED);
1818                 gfs2_glmutex_unlock(gl);
1819         }
1820 }
1821
1822 /**
1823  * gfs2_gl_hash_clear - Empty out the glock hash table
1824  * @sdp: the filesystem
1825  * @wait: wait until it's all gone
1826  *
1827  * Called when unmounting the filesystem, or when inter-node lock manager
1828  * requests DROPLOCKS because it is running out of capacity.
1829  */
1830
1831 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1832 {
1833         unsigned long t;
1834         unsigned int x;
1835         int cont;
1836
1837         t = jiffies;
1838
1839         for (;;) {
1840                 cont = 0;
1841                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1842                         if (examine_bucket(clear_glock, sdp, x))
1843                                 cont = 1;
1844                 }
1845
1846                 if (!wait || !cont)
1847                         break;
1848
1849                 if (time_after_eq(jiffies,
1850                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1851                         fs_warn(sdp, "Unmount seems to be stalled. "
1852                                      "Dumping lock state...\n");
1853                         gfs2_dump_lockstate(sdp);
1854                         t = jiffies;
1855                 }
1856
1857                 down_write(&gfs2_umount_flush_sem);
1858                 invalidate_inodes(sdp->sd_vfs);
1859                 up_write(&gfs2_umount_flush_sem);
1860                 msleep(10);
1861         }
1862 }
1863
1864 /*
1865  *  Diagnostic routines to help debug distributed deadlock
1866  */
1867
1868 /**
1869  * dump_holder - print information about a glock holder
1870  * @str: a string naming the type of holder
1871  * @gh: the glock holder
1872  *
1873  * Returns: 0 on success, -ENOBUFS when we run out of space
1874  */
1875
1876 static int dump_holder(struct glock_iter *gi, char *str,
1877                        struct gfs2_holder *gh)
1878 {
1879         unsigned int x;
1880
1881         print_dbg(gi, "  %s\n", str);
1882         print_dbg(gi, "    owner = %ld\n",
1883                    (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
1884         print_dbg(gi, "    gh_state = %u\n", gh->gh_state);
1885         print_dbg(gi, "    gh_flags =");
1886         for (x = 0; x < 32; x++)
1887                 if (gh->gh_flags & (1 << x))
1888                         print_dbg(gi, " %u", x);
1889         print_dbg(gi, " \n");
1890         print_dbg(gi, "    error = %d\n", gh->gh_error);
1891         print_dbg(gi, "    gh_iflags =");
1892         for (x = 0; x < 32; x++)
1893                 if (test_bit(x, &gh->gh_iflags))
1894                         print_dbg(gi, " %u", x);
1895         print_dbg(gi, " \n");
1896         if (gi)
1897                 print_dbg(gi, "    initialized at: 0x%x\n", gh->gh_ip);
1898         else
1899                 print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
1900
1901         return 0;
1902 }
1903
1904 /**
1905  * dump_inode - print information about an inode
1906  * @ip: the inode
1907  *
1908  * Returns: 0 on success, -ENOBUFS when we run out of space
1909  */
1910
1911 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1912 {
1913         unsigned int x;
1914
1915         print_dbg(gi, "  Inode:\n");
1916         print_dbg(gi, "    num = %llu/%llu\n",
1917                     ip->i_num.no_formal_ino, ip->i_num.no_addr);
1918         print_dbg(gi, "    type = %u\n", IF2DT(ip->i_inode.i_mode));
1919         print_dbg(gi, "    i_flags =");
1920         for (x = 0; x < 32; x++)
1921                 if (test_bit(x, &ip->i_flags))
1922                         print_dbg(gi, " %u", x);
1923         print_dbg(gi, " \n");
1924         return 0;
1925 }
1926
1927 /**
1928  * dump_glock - print information about a glock
1929  * @gl: the glock
1930  * @count: where we are in the buffer
1931  *
1932  * Returns: 0 on success, -ENOBUFS when we run out of space
1933  */
1934
1935 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1936 {
1937         struct gfs2_holder *gh;
1938         unsigned int x;
1939         int error = -ENOBUFS;
1940
1941         spin_lock(&gl->gl_spin);
1942
1943         print_dbg(gi, "Glock 0x%p (%u, %llu)\n", gl, gl->gl_name.ln_type,
1944                    (unsigned long long)gl->gl_name.ln_number);
1945         print_dbg(gi, "  gl_flags =");
1946         for (x = 0; x < 32; x++) {
1947                 if (test_bit(x, &gl->gl_flags))
1948                         print_dbg(gi, " %u", x);
1949         }
1950         print_dbg(gi, " \n");
1951         print_dbg(gi, "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
1952         print_dbg(gi, "  gl_state = %u\n", gl->gl_state);
1953         print_dbg(gi, "  gl_owner = %s\n", gl->gl_owner->comm);
1954         print_dbg(gi, "  gl_ip = %lu\n", gl->gl_ip);
1955         print_dbg(gi, "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1956         print_dbg(gi, "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1957         print_dbg(gi, "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1958         print_dbg(gi, "  object = %s\n", (gl->gl_object) ? "yes" : "no");
1959         print_dbg(gi, "  le = %s\n",
1960                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
1961         print_dbg(gi, "  reclaim = %s\n",
1962                    (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1963         if (gl->gl_aspace)
1964                 print_dbg(gi, "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1965                            gl->gl_aspace->i_mapping->nrpages);
1966         else
1967                 print_dbg(gi, "  aspace = no\n");
1968         print_dbg(gi, "  ail = %d\n", atomic_read(&gl->gl_ail_count));
1969         if (gl->gl_req_gh) {
1970                 error = dump_holder(gi, "Request", gl->gl_req_gh);
1971                 if (error)
1972                         goto out;
1973         }
1974         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1975                 error = dump_holder(gi, "Holder", gh);
1976                 if (error)
1977                         goto out;
1978         }
1979         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1980                 error = dump_holder(gi, "Waiter1", gh);
1981                 if (error)
1982                         goto out;
1983         }
1984         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
1985                 error = dump_holder(gi, "Waiter2", gh);
1986                 if (error)
1987                         goto out;
1988         }
1989         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1990                 error = dump_holder(gi, "Waiter3", gh);
1991                 if (error)
1992                         goto out;
1993         }
1994         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1995                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1996                         list_empty(&gl->gl_holders)) {
1997                         error = dump_inode(gi, gl->gl_object);
1998                         if (error)
1999                                 goto out;
2000                 } else {
2001                         error = -ENOBUFS;
2002                         print_dbg(gi, "  Inode: busy\n");
2003                 }
2004         }
2005
2006         error = 0;
2007
2008 out:
2009         spin_unlock(&gl->gl_spin);
2010         return error;
2011 }
2012
2013 /**
2014  * gfs2_dump_lockstate - print out the current lockstate
2015  * @sdp: the filesystem
2016  * @ub: the buffer to copy the information into
2017  *
2018  * If @ub is NULL, dump the lockstate to the console.
2019  *
2020  */
2021
2022 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2023 {
2024         struct gfs2_glock *gl;
2025         struct hlist_node *h;
2026         unsigned int x;
2027         int error = 0;
2028
2029         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2030
2031                 read_lock(gl_lock_addr(x));
2032
2033                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
2034                         if (gl->gl_sbd != sdp)
2035                                 continue;
2036
2037                         error = dump_glock(NULL, gl);
2038                         if (error)
2039                                 break;
2040                 }
2041
2042                 read_unlock(gl_lock_addr(x));
2043
2044                 if (error)
2045                         break;
2046         }
2047
2048
2049         return error;
2050 }
2051
2052 int __init gfs2_glock_init(void)
2053 {
2054         unsigned i;
2055         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2056                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2057         }
2058 #ifdef GL_HASH_LOCK_SZ
2059         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2060                 rwlock_init(&gl_hash_locks[i]);
2061         }
2062 #endif
2063         return 0;
2064 }
2065
2066 static int gfs2_glock_iter_next(struct glock_iter *gi)
2067 {
2068         while (1) {
2069                 if (!gi->hb_list) {  /* If we don't have a hash bucket yet */
2070                         gi->hb_list = &gl_hash_table[gi->hash].hb_list;
2071                         if (hlist_empty(gi->hb_list)) {
2072                                 gi->hash++;
2073                                 gi->hb_list = NULL;
2074                                 if (gi->hash >= GFS2_GL_HASH_SIZE)
2075                                         return 1;
2076                                 else
2077                                         continue;
2078                         }
2079                         if (!hlist_empty(gi->hb_list)) {
2080                                 gi->gl = list_entry(gi->hb_list->first,
2081                                                     struct gfs2_glock,
2082                                                     gl_list);
2083                         }
2084                 } else {
2085                         if (gi->gl->gl_list.next == NULL) {
2086                                 gi->hash++;
2087                                 gi->hb_list = NULL;
2088                                 continue;
2089                         }
2090                         gi->gl = list_entry(gi->gl->gl_list.next,
2091                                             struct gfs2_glock, gl_list);
2092                 }
2093                 if (gi->gl)
2094                         break;
2095         }
2096         return 0;
2097 }
2098
2099 static void gfs2_glock_iter_free(struct glock_iter *gi)
2100 {
2101         kfree(gi);
2102 }
2103
2104 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2105 {
2106         struct glock_iter *gi;
2107
2108         gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2109         if (!gi)
2110                 return NULL;
2111
2112         gi->sdp = sdp;
2113         gi->hash = 0;
2114         gi->gl = NULL;
2115         gi->hb_list = NULL;
2116         gi->seq = NULL;
2117         memset(gi->string, 0, sizeof(gi->string));
2118
2119         if (gfs2_glock_iter_next(gi)) {
2120                 gfs2_glock_iter_free(gi);
2121                 return NULL;
2122         }
2123
2124         return gi;
2125 }
2126
2127 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2128 {
2129         struct glock_iter *gi;
2130         loff_t n = *pos;
2131
2132         gi = gfs2_glock_iter_init(file->private);
2133         if (!gi)
2134                 return NULL;
2135
2136         while (n--) {
2137                 if (gfs2_glock_iter_next(gi)) {
2138                         gfs2_glock_iter_free(gi);
2139                         return NULL;
2140                 }
2141         }
2142
2143         return gi;
2144 }
2145
2146 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2147                                  loff_t *pos)
2148 {
2149         struct glock_iter *gi = iter_ptr;
2150
2151         (*pos)++;
2152
2153         if (gfs2_glock_iter_next(gi)) {
2154                 gfs2_glock_iter_free(gi);
2155                 return NULL;
2156         }
2157
2158         return gi;
2159 }
2160
2161 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2162 {
2163         /* nothing for now */
2164 }
2165
2166 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2167 {
2168         struct glock_iter *gi = iter_ptr;
2169
2170         gi->seq = file;
2171         dump_glock(gi, gi->gl);
2172
2173         return 0;
2174 }
2175
2176 static struct seq_operations gfs2_glock_seq_ops = {
2177         .start = gfs2_glock_seq_start,
2178         .next  = gfs2_glock_seq_next,
2179         .stop  = gfs2_glock_seq_stop,
2180         .show  = gfs2_glock_seq_show,
2181 };
2182
2183 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2184 {
2185         struct seq_file *seq;
2186         int ret;
2187
2188         ret = seq_open(file, &gfs2_glock_seq_ops);
2189         if (ret)
2190                 return ret;
2191
2192         seq = file->private_data;
2193         seq->private = inode->i_private;
2194
2195         return 0;
2196 }
2197
2198 static const struct file_operations gfs2_debug_fops = {
2199         .owner   = THIS_MODULE,
2200         .open    = gfs2_debugfs_open,
2201         .read    = seq_read,
2202         .llseek  = seq_lseek,
2203         .release = seq_release
2204 };
2205
2206 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2207 {
2208         sdp->debugfs_dentry = debugfs_create_file(sdp->sd_table_name,
2209                                                   S_IFREG | S_IRUGO,
2210                                                   gfs2_root, sdp,
2211                                                   &gfs2_debug_fops);
2212         if (!sdp->debugfs_dentry)
2213                 return -ENOMEM;
2214
2215         return 0;
2216 }
2217
2218 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2219 {
2220         if (sdp && sdp->debugfs_dentry)
2221                 debugfs_remove(sdp->debugfs_dentry);
2222 }
2223
2224 int gfs2_register_debugfs(void)
2225 {
2226         gfs2_root = debugfs_create_dir("gfs2", NULL);
2227         return gfs2_root ? 0 : -ENOMEM;
2228 }
2229
2230 void gfs2_unregister_debugfs(void)
2231 {
2232         debugfs_remove(gfs2_root);
2233 }