[GFS2] Allow bmap to allocate extents
[linux-2.6.git] / fs / gfs2 / glock.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/delay.h>
16 #include <linux/sort.h>
17 #include <linux/jhash.h>
18 #include <linux/kallsyms.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/list.h>
21 #include <linux/lm_interface.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <linux/rwsem.h>
25 #include <asm/uaccess.h>
26 #include <linux/seq_file.h>
27 #include <linux/debugfs.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
30 #include <linux/workqueue.h>
31 #include <linux/jiffies.h>
32
33 #include "gfs2.h"
34 #include "incore.h"
35 #include "glock.h"
36 #include "glops.h"
37 #include "inode.h"
38 #include "lops.h"
39 #include "meta_io.h"
40 #include "quota.h"
41 #include "super.h"
42 #include "util.h"
43
44 struct gfs2_gl_hash_bucket {
45         struct hlist_head hb_list;
46 };
47
48 struct glock_iter {
49         int hash;                     /* hash bucket index         */
50         struct gfs2_sbd *sdp;         /* incore superblock         */
51         struct gfs2_glock *gl;        /* current glock struct      */
52         struct seq_file *seq;         /* sequence file for debugfs */
53         char string[512];             /* scratch space             */
54 };
55
56 typedef void (*glock_examiner) (struct gfs2_glock * gl);
57
58 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
59 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl);
60 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
61 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
62 static void run_queue(struct gfs2_glock *gl);
63
64 static DECLARE_RWSEM(gfs2_umount_flush_sem);
65 static struct dentry *gfs2_root;
66 static struct task_struct *scand_process;
67 static unsigned int scand_secs = 5;
68 static struct workqueue_struct *glock_workqueue;
69
70 #define GFS2_GL_HASH_SHIFT      15
71 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
72 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
73
74 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
75 static struct dentry *gfs2_root;
76
77 /*
78  * Despite what you might think, the numbers below are not arbitrary :-)
79  * They are taken from the ipv4 routing hash code, which is well tested
80  * and thus should be nearly optimal. Later on we might tweek the numbers
81  * but for now this should be fine.
82  *
83  * The reason for putting the locks in a separate array from the list heads
84  * is that we can have fewer locks than list heads and save memory. We use
85  * the same hash function for both, but with a different hash mask.
86  */
87 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
88         defined(CONFIG_PROVE_LOCKING)
89
90 #ifdef CONFIG_LOCKDEP
91 # define GL_HASH_LOCK_SZ        256
92 #else
93 # if NR_CPUS >= 32
94 #  define GL_HASH_LOCK_SZ       4096
95 # elif NR_CPUS >= 16
96 #  define GL_HASH_LOCK_SZ       2048
97 # elif NR_CPUS >= 8
98 #  define GL_HASH_LOCK_SZ       1024
99 # elif NR_CPUS >= 4
100 #  define GL_HASH_LOCK_SZ       512
101 # else
102 #  define GL_HASH_LOCK_SZ       256
103 # endif
104 #endif
105
106 /* We never want more locks than chains */
107 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
108 # undef GL_HASH_LOCK_SZ
109 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
110 #endif
111
112 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
113
114 static inline rwlock_t *gl_lock_addr(unsigned int x)
115 {
116         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
117 }
118 #else /* not SMP, so no spinlocks required */
119 static inline rwlock_t *gl_lock_addr(unsigned int x)
120 {
121         return NULL;
122 }
123 #endif
124
125 /**
126  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
127  * @actual: the current state of the lock
128  * @requested: the lock state that was requested by the caller
129  * @flags: the modifier flags passed in by the caller
130  *
131  * Returns: 1 if the locks are compatible, 0 otherwise
132  */
133
134 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
135                                    int flags)
136 {
137         if (actual == requested)
138                 return 1;
139
140         if (flags & GL_EXACT)
141                 return 0;
142
143         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
144                 return 1;
145
146         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
147                 return 1;
148
149         return 0;
150 }
151
152 /**
153  * gl_hash() - Turn glock number into hash bucket number
154  * @lock: The glock number
155  *
156  * Returns: The number of the corresponding hash bucket
157  */
158
159 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
160                             const struct lm_lockname *name)
161 {
162         unsigned int h;
163
164         h = jhash(&name->ln_number, sizeof(u64), 0);
165         h = jhash(&name->ln_type, sizeof(unsigned int), h);
166         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
167         h &= GFS2_GL_HASH_MASK;
168
169         return h;
170 }
171
172 /**
173  * glock_free() - Perform a few checks and then release struct gfs2_glock
174  * @gl: The glock to release
175  *
176  * Also calls lock module to release its internal structure for this glock.
177  *
178  */
179
180 static void glock_free(struct gfs2_glock *gl)
181 {
182         struct gfs2_sbd *sdp = gl->gl_sbd;
183         struct inode *aspace = gl->gl_aspace;
184
185         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
186                 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl->gl_lock);
187
188         if (aspace)
189                 gfs2_aspace_put(aspace);
190
191         kmem_cache_free(gfs2_glock_cachep, gl);
192 }
193
194 /**
195  * gfs2_glock_hold() - increment reference count on glock
196  * @gl: The glock to hold
197  *
198  */
199
200 static void gfs2_glock_hold(struct gfs2_glock *gl)
201 {
202         atomic_inc(&gl->gl_ref);
203 }
204
205 /**
206  * gfs2_glock_put() - Decrement reference count on glock
207  * @gl: The glock to put
208  *
209  */
210
211 int gfs2_glock_put(struct gfs2_glock *gl)
212 {
213         int rv = 0;
214         struct gfs2_sbd *sdp = gl->gl_sbd;
215
216         write_lock(gl_lock_addr(gl->gl_hash));
217         if (atomic_dec_and_test(&gl->gl_ref)) {
218                 hlist_del(&gl->gl_list);
219                 write_unlock(gl_lock_addr(gl->gl_hash));
220                 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
221                 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
222                 gfs2_assert(sdp, list_empty(&gl->gl_holders));
223                 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
224                 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
225                 glock_free(gl);
226                 rv = 1;
227                 goto out;
228         }
229         write_unlock(gl_lock_addr(gl->gl_hash));
230 out:
231         return rv;
232 }
233
234 /**
235  * search_bucket() - Find struct gfs2_glock by lock number
236  * @bucket: the bucket to search
237  * @name: The lock name
238  *
239  * Returns: NULL, or the struct gfs2_glock with the requested number
240  */
241
242 static struct gfs2_glock *search_bucket(unsigned int hash,
243                                         const struct gfs2_sbd *sdp,
244                                         const struct lm_lockname *name)
245 {
246         struct gfs2_glock *gl;
247         struct hlist_node *h;
248
249         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
250                 if (!lm_name_equal(&gl->gl_name, name))
251                         continue;
252                 if (gl->gl_sbd != sdp)
253                         continue;
254
255                 atomic_inc(&gl->gl_ref);
256
257                 return gl;
258         }
259
260         return NULL;
261 }
262
263 /**
264  * gfs2_glock_find() - Find glock by lock number
265  * @sdp: The GFS2 superblock
266  * @name: The lock name
267  *
268  * Returns: NULL, or the struct gfs2_glock with the requested number
269  */
270
271 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
272                                           const struct lm_lockname *name)
273 {
274         unsigned int hash = gl_hash(sdp, name);
275         struct gfs2_glock *gl;
276
277         read_lock(gl_lock_addr(hash));
278         gl = search_bucket(hash, sdp, name);
279         read_unlock(gl_lock_addr(hash));
280
281         return gl;
282 }
283
284 static void glock_work_func(struct work_struct *work)
285 {
286         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
287
288         spin_lock(&gl->gl_spin);
289         if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags))
290                 set_bit(GLF_DEMOTE, &gl->gl_flags);
291         run_queue(gl);
292         spin_unlock(&gl->gl_spin);
293         gfs2_glock_put(gl);
294 }
295
296 static int gfs2_lm_get_lock(struct gfs2_sbd *sdp, struct lm_lockname *name,
297                      void **lockp)
298 {
299         int error = -EIO;
300         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
301                 error = sdp->sd_lockstruct.ls_ops->lm_get_lock(
302                                 sdp->sd_lockstruct.ls_lockspace, name, lockp);
303         return error;
304 }
305
306 /**
307  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
308  * @sdp: The GFS2 superblock
309  * @number: the lock number
310  * @glops: The glock_operations to use
311  * @create: If 0, don't create the glock if it doesn't exist
312  * @glp: the glock is returned here
313  *
314  * This does not lock a glock, just finds/creates structures for one.
315  *
316  * Returns: errno
317  */
318
319 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
320                    const struct gfs2_glock_operations *glops, int create,
321                    struct gfs2_glock **glp)
322 {
323         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
324         struct gfs2_glock *gl, *tmp;
325         unsigned int hash = gl_hash(sdp, &name);
326         int error;
327
328         read_lock(gl_lock_addr(hash));
329         gl = search_bucket(hash, sdp, &name);
330         read_unlock(gl_lock_addr(hash));
331
332         if (gl || !create) {
333                 *glp = gl;
334                 return 0;
335         }
336
337         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
338         if (!gl)
339                 return -ENOMEM;
340
341         gl->gl_flags = 0;
342         gl->gl_name = name;
343         atomic_set(&gl->gl_ref, 1);
344         gl->gl_state = LM_ST_UNLOCKED;
345         gl->gl_demote_state = LM_ST_EXCLUSIVE;
346         gl->gl_hash = hash;
347         gl->gl_owner_pid = NULL;
348         gl->gl_ip = 0;
349         gl->gl_ops = glops;
350         gl->gl_req_gh = NULL;
351         gl->gl_stamp = jiffies;
352         gl->gl_tchange = jiffies;
353         gl->gl_object = NULL;
354         gl->gl_sbd = sdp;
355         gl->gl_aspace = NULL;
356         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
357
358         /* If this glock protects actual on-disk data or metadata blocks,
359            create a VFS inode to manage the pages/buffers holding them. */
360         if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
361                 gl->gl_aspace = gfs2_aspace_get(sdp);
362                 if (!gl->gl_aspace) {
363                         error = -ENOMEM;
364                         goto fail;
365                 }
366         }
367
368         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
369         if (error)
370                 goto fail_aspace;
371
372         write_lock(gl_lock_addr(hash));
373         tmp = search_bucket(hash, sdp, &name);
374         if (tmp) {
375                 write_unlock(gl_lock_addr(hash));
376                 glock_free(gl);
377                 gl = tmp;
378         } else {
379                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
380                 write_unlock(gl_lock_addr(hash));
381         }
382
383         *glp = gl;
384
385         return 0;
386
387 fail_aspace:
388         if (gl->gl_aspace)
389                 gfs2_aspace_put(gl->gl_aspace);
390 fail:
391         kmem_cache_free(gfs2_glock_cachep, gl);
392         return error;
393 }
394
395 /**
396  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
397  * @gl: the glock
398  * @state: the state we're requesting
399  * @flags: the modifier flags
400  * @gh: the holder structure
401  *
402  */
403
404 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
405                       struct gfs2_holder *gh)
406 {
407         INIT_LIST_HEAD(&gh->gh_list);
408         gh->gh_gl = gl;
409         gh->gh_ip = (unsigned long)__builtin_return_address(0);
410         gh->gh_owner_pid = get_pid(task_pid(current));
411         gh->gh_state = state;
412         gh->gh_flags = flags;
413         gh->gh_error = 0;
414         gh->gh_iflags = 0;
415         gfs2_glock_hold(gl);
416 }
417
418 /**
419  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
420  * @state: the state we're requesting
421  * @flags: the modifier flags
422  * @gh: the holder structure
423  *
424  * Don't mess with the glock.
425  *
426  */
427
428 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
429 {
430         gh->gh_state = state;
431         gh->gh_flags = flags;
432         gh->gh_iflags = 0;
433         gh->gh_ip = (unsigned long)__builtin_return_address(0);
434 }
435
436 /**
437  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
438  * @gh: the holder structure
439  *
440  */
441
442 void gfs2_holder_uninit(struct gfs2_holder *gh)
443 {
444         put_pid(gh->gh_owner_pid);
445         gfs2_glock_put(gh->gh_gl);
446         gh->gh_gl = NULL;
447         gh->gh_ip = 0;
448 }
449
450 static void gfs2_holder_wake(struct gfs2_holder *gh)
451 {
452         clear_bit(HIF_WAIT, &gh->gh_iflags);
453         smp_mb__after_clear_bit();
454         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
455 }
456
457 static int just_schedule(void *word)
458 {
459         schedule();
460         return 0;
461 }
462
463 static void wait_on_holder(struct gfs2_holder *gh)
464 {
465         might_sleep();
466         wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
467 }
468
469 static void gfs2_demote_wake(struct gfs2_glock *gl)
470 {
471         gl->gl_demote_state = LM_ST_EXCLUSIVE;
472         clear_bit(GLF_DEMOTE, &gl->gl_flags);
473         smp_mb__after_clear_bit();
474         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
475 }
476
477 static void wait_on_demote(struct gfs2_glock *gl)
478 {
479         might_sleep();
480         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
481 }
482
483 /**
484  * rq_mutex - process a mutex request in the queue
485  * @gh: the glock holder
486  *
487  * Returns: 1 if the queue is blocked
488  */
489
490 static int rq_mutex(struct gfs2_holder *gh)
491 {
492         struct gfs2_glock *gl = gh->gh_gl;
493
494         list_del_init(&gh->gh_list);
495         /*  gh->gh_error never examined.  */
496         set_bit(GLF_LOCK, &gl->gl_flags);
497         clear_bit(HIF_WAIT, &gh->gh_iflags);
498         smp_mb();
499         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
500
501         return 1;
502 }
503
504 /**
505  * rq_promote - process a promote request in the queue
506  * @gh: the glock holder
507  *
508  * Acquire a new inter-node lock, or change a lock state to more restrictive.
509  *
510  * Returns: 1 if the queue is blocked
511  */
512
513 static int rq_promote(struct gfs2_holder *gh)
514 {
515         struct gfs2_glock *gl = gh->gh_gl;
516
517         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
518                 if (list_empty(&gl->gl_holders)) {
519                         gl->gl_req_gh = gh;
520                         set_bit(GLF_LOCK, &gl->gl_flags);
521                         spin_unlock(&gl->gl_spin);
522                         gfs2_glock_xmote_th(gh->gh_gl, gh);
523                         spin_lock(&gl->gl_spin);
524                 }
525                 return 1;
526         }
527
528         if (list_empty(&gl->gl_holders)) {
529                 set_bit(HIF_FIRST, &gh->gh_iflags);
530                 set_bit(GLF_LOCK, &gl->gl_flags);
531         } else {
532                 struct gfs2_holder *next_gh;
533                 if (gh->gh_state == LM_ST_EXCLUSIVE)
534                         return 1;
535                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
536                                      gh_list);
537                 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
538                          return 1;
539         }
540
541         list_move_tail(&gh->gh_list, &gl->gl_holders);
542         gh->gh_error = 0;
543         set_bit(HIF_HOLDER, &gh->gh_iflags);
544
545         gfs2_holder_wake(gh);
546
547         return 0;
548 }
549
550 /**
551  * rq_demote - process a demote request in the queue
552  * @gh: the glock holder
553  *
554  * Returns: 1 if the queue is blocked
555  */
556
557 static int rq_demote(struct gfs2_glock *gl)
558 {
559         if (!list_empty(&gl->gl_holders))
560                 return 1;
561
562         if (gl->gl_state == gl->gl_demote_state ||
563             gl->gl_state == LM_ST_UNLOCKED) {
564                 gfs2_demote_wake(gl);
565                 return 0;
566         }
567
568         set_bit(GLF_LOCK, &gl->gl_flags);
569         set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
570
571         if (gl->gl_demote_state == LM_ST_UNLOCKED ||
572             gl->gl_state != LM_ST_EXCLUSIVE) {
573                 spin_unlock(&gl->gl_spin);
574                 gfs2_glock_drop_th(gl);
575         } else {
576                 spin_unlock(&gl->gl_spin);
577                 gfs2_glock_xmote_th(gl, NULL);
578         }
579
580         spin_lock(&gl->gl_spin);
581         clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
582
583         return 0;
584 }
585
586 /**
587  * run_queue - process holder structures on a glock
588  * @gl: the glock
589  *
590  */
591 static void run_queue(struct gfs2_glock *gl)
592 {
593         struct gfs2_holder *gh;
594         int blocked = 1;
595
596         for (;;) {
597                 if (test_bit(GLF_LOCK, &gl->gl_flags))
598                         break;
599
600                 if (!list_empty(&gl->gl_waiters1)) {
601                         gh = list_entry(gl->gl_waiters1.next,
602                                         struct gfs2_holder, gh_list);
603                         blocked = rq_mutex(gh);
604                 } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
605                         blocked = rq_demote(gl);
606                         if (test_bit(GLF_WAITERS2, &gl->gl_flags) &&
607                                      !blocked) {
608                                 set_bit(GLF_DEMOTE, &gl->gl_flags);
609                                 gl->gl_demote_state = LM_ST_UNLOCKED;
610                         }
611                         clear_bit(GLF_WAITERS2, &gl->gl_flags);
612                 } else if (!list_empty(&gl->gl_waiters3)) {
613                         gh = list_entry(gl->gl_waiters3.next,
614                                         struct gfs2_holder, gh_list);
615                         blocked = rq_promote(gh);
616                 } else
617                         break;
618
619                 if (blocked)
620                         break;
621         }
622 }
623
624 /**
625  * gfs2_glmutex_lock - acquire a local lock on a glock
626  * @gl: the glock
627  *
628  * Gives caller exclusive access to manipulate a glock structure.
629  */
630
631 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
632 {
633         spin_lock(&gl->gl_spin);
634         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
635                 struct gfs2_holder gh;
636
637                 gfs2_holder_init(gl, 0, 0, &gh);
638                 set_bit(HIF_WAIT, &gh.gh_iflags);
639                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
640                 spin_unlock(&gl->gl_spin);
641                 wait_on_holder(&gh);
642                 gfs2_holder_uninit(&gh);
643         } else {
644                 gl->gl_owner_pid = get_pid(task_pid(current));
645                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
646                 spin_unlock(&gl->gl_spin);
647         }
648 }
649
650 /**
651  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
652  * @gl: the glock
653  *
654  * Returns: 1 if the glock is acquired
655  */
656
657 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
658 {
659         int acquired = 1;
660
661         spin_lock(&gl->gl_spin);
662         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
663                 acquired = 0;
664         } else {
665                 gl->gl_owner_pid = get_pid(task_pid(current));
666                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
667         }
668         spin_unlock(&gl->gl_spin);
669
670         return acquired;
671 }
672
673 /**
674  * gfs2_glmutex_unlock - release a local lock on a glock
675  * @gl: the glock
676  *
677  */
678
679 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
680 {
681         struct pid *pid;
682
683         spin_lock(&gl->gl_spin);
684         clear_bit(GLF_LOCK, &gl->gl_flags);
685         pid = gl->gl_owner_pid;
686         gl->gl_owner_pid = NULL;
687         gl->gl_ip = 0;
688         run_queue(gl);
689         spin_unlock(&gl->gl_spin);
690
691         put_pid(pid);
692 }
693
694 /**
695  * handle_callback - process a demote request
696  * @gl: the glock
697  * @state: the state the caller wants us to change to
698  *
699  * There are only two requests that we are going to see in actual
700  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
701  */
702
703 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
704                             int remote, unsigned long delay)
705 {
706         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
707
708         spin_lock(&gl->gl_spin);
709         set_bit(bit, &gl->gl_flags);
710         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
711                 gl->gl_demote_state = state;
712                 gl->gl_demote_time = jiffies;
713                 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
714                     gl->gl_object) {
715                         gfs2_glock_schedule_for_reclaim(gl);
716                         spin_unlock(&gl->gl_spin);
717                         return;
718                 }
719         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
720                         gl->gl_demote_state != state) {
721                 if (test_bit(GLF_DEMOTE_IN_PROGRESS,  &gl->gl_flags)) 
722                         set_bit(GLF_WAITERS2, &gl->gl_flags);
723                 else 
724                         gl->gl_demote_state = LM_ST_UNLOCKED;
725         }
726         spin_unlock(&gl->gl_spin);
727 }
728
729 /**
730  * state_change - record that the glock is now in a different state
731  * @gl: the glock
732  * @new_state the new state
733  *
734  */
735
736 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
737 {
738         int held1, held2;
739
740         held1 = (gl->gl_state != LM_ST_UNLOCKED);
741         held2 = (new_state != LM_ST_UNLOCKED);
742
743         if (held1 != held2) {
744                 if (held2)
745                         gfs2_glock_hold(gl);
746                 else
747                         gfs2_glock_put(gl);
748         }
749
750         gl->gl_state = new_state;
751         gl->gl_tchange = jiffies;
752 }
753
754 /**
755  * drop_bh - Called after a lock module unlock completes
756  * @gl: the glock
757  * @ret: the return status
758  *
759  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
760  * Doesn't drop the reference on the glock the top half took out
761  *
762  */
763
764 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
765 {
766         struct gfs2_sbd *sdp = gl->gl_sbd;
767         const struct gfs2_glock_operations *glops = gl->gl_ops;
768
769         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
770         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
771         gfs2_assert_warn(sdp, !ret);
772
773         state_change(gl, LM_ST_UNLOCKED);
774
775         if (glops->go_inval)
776                 glops->go_inval(gl, DIO_METADATA);
777
778         spin_lock(&gl->gl_spin);
779         gfs2_demote_wake(gl);
780         clear_bit(GLF_LOCK, &gl->gl_flags);
781         spin_unlock(&gl->gl_spin);
782         gfs2_glock_put(gl);
783 }
784
785 /**
786  * xmote_bh - Called after the lock module is done acquiring a lock
787  * @gl: The glock in question
788  * @ret: the int returned from the lock module
789  *
790  */
791
792 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
793 {
794         struct gfs2_sbd *sdp = gl->gl_sbd;
795         const struct gfs2_glock_operations *glops = gl->gl_ops;
796         struct gfs2_holder *gh = gl->gl_req_gh;
797         int prev_state = gl->gl_state;
798         int op_done = 1;
799
800         if (!gh && (ret & LM_OUT_ST_MASK) == LM_ST_UNLOCKED) {
801                 drop_bh(gl, ret);
802                 return;
803         }
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                 if (ret & LM_OUT_CANCELED) {
826                         op_done = 0;
827                 } else {
828                         spin_lock(&gl->gl_spin);
829                         if (gl->gl_state != gl->gl_demote_state) {
830                                 spin_unlock(&gl->gl_spin);
831                                 gfs2_glock_drop_th(gl);
832                                 gfs2_glock_put(gl);
833                                 return;
834                         }
835                         gfs2_demote_wake(gl);
836                         spin_unlock(&gl->gl_spin);
837                 }
838         } else {
839                 spin_lock(&gl->gl_spin);
840                 list_del_init(&gh->gh_list);
841                 gh->gh_error = -EIO;
842                 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 
843                         goto out;
844                 gh->gh_error = GLR_CANCELED;
845                 if (ret & LM_OUT_CANCELED) 
846                         goto out;
847                 if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
848                         list_add_tail(&gh->gh_list, &gl->gl_holders);
849                         gh->gh_error = 0;
850                         set_bit(HIF_HOLDER, &gh->gh_iflags);
851                         set_bit(HIF_FIRST, &gh->gh_iflags);
852                         op_done = 0;
853                         goto out;
854                 }
855                 gh->gh_error = GLR_TRYFAILED;
856                 if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
857                         goto out;
858                 gh->gh_error = -EINVAL;
859                 if (gfs2_assert_withdraw(sdp, 0) == -1)
860                         fs_err(sdp, "ret = 0x%.8X\n", ret);
861 out:
862                 spin_unlock(&gl->gl_spin);
863         }
864
865         if (glops->go_xmote_bh)
866                 glops->go_xmote_bh(gl);
867
868         if (op_done) {
869                 spin_lock(&gl->gl_spin);
870                 gl->gl_req_gh = NULL;
871                 clear_bit(GLF_LOCK, &gl->gl_flags);
872                 spin_unlock(&gl->gl_spin);
873         }
874
875         gfs2_glock_put(gl);
876
877         if (gh)
878                 gfs2_holder_wake(gh);
879 }
880
881 static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
882                                  unsigned int cur_state, unsigned int req_state,
883                                  unsigned int flags)
884 {
885         int ret = 0;
886         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
887                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock, cur_state,
888                                                          req_state, flags);
889         return ret;
890 }
891
892 /**
893  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
894  * @gl: The glock in question
895  * @state: the requested state
896  * @flags: modifier flags to the lock call
897  *
898  */
899
900 static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
901 {
902         struct gfs2_sbd *sdp = gl->gl_sbd;
903         int flags = gh ? gh->gh_flags : 0;
904         unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
905         const struct gfs2_glock_operations *glops = gl->gl_ops;
906         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
907                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
908                                  LM_FLAG_PRIORITY);
909         unsigned int lck_ret;
910
911         if (glops->go_xmote_th)
912                 glops->go_xmote_th(gl);
913
914         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
915         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
916         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
917         gfs2_assert_warn(sdp, state != gl->gl_state);
918
919         gfs2_glock_hold(gl);
920
921         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
922
923         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
924                 return;
925
926         if (lck_ret & LM_OUT_ASYNC)
927                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
928         else
929                 xmote_bh(gl, lck_ret);
930 }
931
932 static unsigned int gfs2_lm_unlock(struct gfs2_sbd *sdp, void *lock,
933                                    unsigned int cur_state)
934 {
935         int ret = 0;
936         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
937                 ret =  sdp->sd_lockstruct.ls_ops->lm_unlock(lock, cur_state);
938         return ret;
939 }
940
941 /**
942  * gfs2_glock_drop_th - call into the lock module to unlock a lock
943  * @gl: the glock
944  *
945  */
946
947 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
948 {
949         struct gfs2_sbd *sdp = gl->gl_sbd;
950         const struct gfs2_glock_operations *glops = gl->gl_ops;
951         unsigned int ret;
952
953         if (glops->go_xmote_th)
954                 glops->go_xmote_th(gl);
955
956         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
957         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
958         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
959
960         gfs2_glock_hold(gl);
961
962         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
963
964         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
965                 return;
966
967         if (!ret)
968                 drop_bh(gl, ret);
969         else
970                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
971 }
972
973 /**
974  * do_cancels - cancel requests for locks stuck waiting on an expire flag
975  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
976  *
977  * Don't cancel GL_NOCANCEL requests.
978  */
979
980 static void do_cancels(struct gfs2_holder *gh)
981 {
982         struct gfs2_glock *gl = gh->gh_gl;
983         struct gfs2_sbd *sdp = gl->gl_sbd;
984
985         spin_lock(&gl->gl_spin);
986
987         while (gl->gl_req_gh != gh &&
988                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
989                !list_empty(&gh->gh_list)) {
990                 if (!(gl->gl_req_gh && (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
991                         spin_unlock(&gl->gl_spin);
992                         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
993                                 sdp->sd_lockstruct.ls_ops->lm_cancel(gl->gl_lock);
994                         msleep(100);
995                         spin_lock(&gl->gl_spin);
996                 } else {
997                         spin_unlock(&gl->gl_spin);
998                         msleep(100);
999                         spin_lock(&gl->gl_spin);
1000                 }
1001         }
1002
1003         spin_unlock(&gl->gl_spin);
1004 }
1005
1006 /**
1007  * glock_wait_internal - wait on a glock acquisition
1008  * @gh: the glock holder
1009  *
1010  * Returns: 0 on success
1011  */
1012
1013 static int glock_wait_internal(struct gfs2_holder *gh)
1014 {
1015         struct gfs2_glock *gl = gh->gh_gl;
1016         struct gfs2_sbd *sdp = gl->gl_sbd;
1017         const struct gfs2_glock_operations *glops = gl->gl_ops;
1018
1019         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1020                 return -EIO;
1021
1022         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1023                 spin_lock(&gl->gl_spin);
1024                 if (gl->gl_req_gh != gh &&
1025                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1026                     !list_empty(&gh->gh_list)) {
1027                         list_del_init(&gh->gh_list);
1028                         gh->gh_error = GLR_TRYFAILED;
1029                         run_queue(gl);
1030                         spin_unlock(&gl->gl_spin);
1031                         return gh->gh_error;
1032                 }
1033                 spin_unlock(&gl->gl_spin);
1034         }
1035
1036         if (gh->gh_flags & LM_FLAG_PRIORITY)
1037                 do_cancels(gh);
1038
1039         wait_on_holder(gh);
1040         if (gh->gh_error)
1041                 return gh->gh_error;
1042
1043         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1044         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1045                                                    gh->gh_flags));
1046
1047         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1048                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1049
1050                 if (glops->go_lock) {
1051                         gh->gh_error = glops->go_lock(gh);
1052                         if (gh->gh_error) {
1053                                 spin_lock(&gl->gl_spin);
1054                                 list_del_init(&gh->gh_list);
1055                                 spin_unlock(&gl->gl_spin);
1056                         }
1057                 }
1058
1059                 spin_lock(&gl->gl_spin);
1060                 gl->gl_req_gh = NULL;
1061                 clear_bit(GLF_LOCK, &gl->gl_flags);
1062                 run_queue(gl);
1063                 spin_unlock(&gl->gl_spin);
1064         }
1065
1066         return gh->gh_error;
1067 }
1068
1069 static inline struct gfs2_holder *
1070 find_holder_by_owner(struct list_head *head, struct pid *pid)
1071 {
1072         struct gfs2_holder *gh;
1073
1074         list_for_each_entry(gh, head, gh_list) {
1075                 if (gh->gh_owner_pid == pid)
1076                         return gh;
1077         }
1078
1079         return NULL;
1080 }
1081
1082 static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
1083 {
1084         va_list args;
1085
1086         va_start(args, fmt);
1087         if (gi) {
1088                 vsprintf(gi->string, fmt, args);
1089                 seq_printf(gi->seq, gi->string);
1090         }
1091         else
1092                 vprintk(fmt, args);
1093         va_end(args);
1094 }
1095
1096 /**
1097  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1098  * @gh: the holder structure to add
1099  *
1100  */
1101
1102 static void add_to_queue(struct gfs2_holder *gh)
1103 {
1104         struct gfs2_glock *gl = gh->gh_gl;
1105         struct gfs2_holder *existing;
1106
1107         BUG_ON(gh->gh_owner_pid == NULL);
1108         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1109                 BUG();
1110
1111         if (!(gh->gh_flags & GL_FLOCK)) {
1112                 existing = find_holder_by_owner(&gl->gl_holders, 
1113                                                 gh->gh_owner_pid);
1114                 if (existing) {
1115                         print_symbol(KERN_WARNING "original: %s\n", 
1116                                      existing->gh_ip);
1117                         printk(KERN_INFO "pid : %d\n",
1118                                         pid_nr(existing->gh_owner_pid));
1119                         printk(KERN_INFO "lock type : %d lock state : %d\n",
1120                                existing->gh_gl->gl_name.ln_type, 
1121                                existing->gh_gl->gl_state);
1122                         print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1123                         printk(KERN_INFO "pid : %d\n",
1124                                         pid_nr(gh->gh_owner_pid));
1125                         printk(KERN_INFO "lock type : %d lock state : %d\n",
1126                                gl->gl_name.ln_type, gl->gl_state);
1127                         BUG();
1128                 }
1129                 
1130                 existing = find_holder_by_owner(&gl->gl_waiters3, 
1131                                                 gh->gh_owner_pid);
1132                 if (existing) {
1133                         print_symbol(KERN_WARNING "original: %s\n", 
1134                                      existing->gh_ip);
1135                         print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1136                         BUG();
1137                 }
1138         }
1139
1140         if (gh->gh_flags & LM_FLAG_PRIORITY)
1141                 list_add(&gh->gh_list, &gl->gl_waiters3);
1142         else
1143                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1144 }
1145
1146 /**
1147  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1148  * @gh: the holder structure
1149  *
1150  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1151  *
1152  * Returns: 0, GLR_TRYFAILED, or errno on failure
1153  */
1154
1155 int gfs2_glock_nq(struct gfs2_holder *gh)
1156 {
1157         struct gfs2_glock *gl = gh->gh_gl;
1158         struct gfs2_sbd *sdp = gl->gl_sbd;
1159         int error = 0;
1160
1161 restart:
1162         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1163                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1164                 return -EIO;
1165         }
1166
1167         spin_lock(&gl->gl_spin);
1168         add_to_queue(gh);
1169         run_queue(gl);
1170         spin_unlock(&gl->gl_spin);
1171
1172         if (!(gh->gh_flags & GL_ASYNC)) {
1173                 error = glock_wait_internal(gh);
1174                 if (error == GLR_CANCELED) {
1175                         msleep(100);
1176                         goto restart;
1177                 }
1178         }
1179
1180         return error;
1181 }
1182
1183 /**
1184  * gfs2_glock_poll - poll to see if an async request has been completed
1185  * @gh: the holder
1186  *
1187  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1188  */
1189
1190 int gfs2_glock_poll(struct gfs2_holder *gh)
1191 {
1192         struct gfs2_glock *gl = gh->gh_gl;
1193         int ready = 0;
1194
1195         spin_lock(&gl->gl_spin);
1196
1197         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1198                 ready = 1;
1199         else if (list_empty(&gh->gh_list)) {
1200                 if (gh->gh_error == GLR_CANCELED) {
1201                         spin_unlock(&gl->gl_spin);
1202                         msleep(100);
1203                         if (gfs2_glock_nq(gh))
1204                                 return 1;
1205                         return 0;
1206                 } else
1207                         ready = 1;
1208         }
1209
1210         spin_unlock(&gl->gl_spin);
1211
1212         return ready;
1213 }
1214
1215 /**
1216  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1217  * @gh: the holder structure
1218  *
1219  * Returns: 0, GLR_TRYFAILED, or errno on failure
1220  */
1221
1222 int gfs2_glock_wait(struct gfs2_holder *gh)
1223 {
1224         int error;
1225
1226         error = glock_wait_internal(gh);
1227         if (error == GLR_CANCELED) {
1228                 msleep(100);
1229                 gh->gh_flags &= ~GL_ASYNC;
1230                 error = gfs2_glock_nq(gh);
1231         }
1232
1233         return error;
1234 }
1235
1236 /**
1237  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1238  * @gh: the glock holder
1239  *
1240  */
1241
1242 void gfs2_glock_dq(struct gfs2_holder *gh)
1243 {
1244         struct gfs2_glock *gl = gh->gh_gl;
1245         const struct gfs2_glock_operations *glops = gl->gl_ops;
1246         unsigned delay = 0;
1247
1248         if (gh->gh_flags & GL_NOCACHE)
1249                 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1250
1251         gfs2_glmutex_lock(gl);
1252
1253         spin_lock(&gl->gl_spin);
1254         list_del_init(&gh->gh_list);
1255
1256         if (list_empty(&gl->gl_holders)) {
1257                 if (glops->go_unlock) {
1258                         spin_unlock(&gl->gl_spin);
1259                         glops->go_unlock(gh);
1260                         spin_lock(&gl->gl_spin);
1261                 }
1262                 gl->gl_stamp = jiffies;
1263         }
1264
1265         clear_bit(GLF_LOCK, &gl->gl_flags);
1266         spin_unlock(&gl->gl_spin);
1267
1268         gfs2_glock_hold(gl);
1269         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1270             !test_bit(GLF_DEMOTE, &gl->gl_flags))
1271                 delay = gl->gl_ops->go_min_hold_time;
1272         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1273                 gfs2_glock_put(gl);
1274 }
1275
1276 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1277 {
1278         struct gfs2_glock *gl = gh->gh_gl;
1279         gfs2_glock_dq(gh);
1280         wait_on_demote(gl);
1281 }
1282
1283 /**
1284  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1285  * @gh: the holder structure
1286  *
1287  */
1288
1289 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1290 {
1291         gfs2_glock_dq(gh);
1292         gfs2_holder_uninit(gh);
1293 }
1294
1295 /**
1296  * gfs2_glock_nq_num - acquire a glock based on lock number
1297  * @sdp: the filesystem
1298  * @number: the lock number
1299  * @glops: the glock operations for the type of glock
1300  * @state: the state to acquire the glock in
1301  * @flags: modifier flags for the aquisition
1302  * @gh: the struct gfs2_holder
1303  *
1304  * Returns: errno
1305  */
1306
1307 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1308                       const struct gfs2_glock_operations *glops,
1309                       unsigned int state, int flags, struct gfs2_holder *gh)
1310 {
1311         struct gfs2_glock *gl;
1312         int error;
1313
1314         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1315         if (!error) {
1316                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1317                 gfs2_glock_put(gl);
1318         }
1319
1320         return error;
1321 }
1322
1323 /**
1324  * glock_compare - Compare two struct gfs2_glock structures for sorting
1325  * @arg_a: the first structure
1326  * @arg_b: the second structure
1327  *
1328  */
1329
1330 static int glock_compare(const void *arg_a, const void *arg_b)
1331 {
1332         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1333         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1334         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1335         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1336
1337         if (a->ln_number > b->ln_number)
1338                 return 1;
1339         if (a->ln_number < b->ln_number)
1340                 return -1;
1341         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1342         return 0;
1343 }
1344
1345 /**
1346  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1347  * @num_gh: the number of structures
1348  * @ghs: an array of struct gfs2_holder structures
1349  *
1350  * Returns: 0 on success (all glocks acquired),
1351  *          errno on failure (no glocks acquired)
1352  */
1353
1354 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1355                      struct gfs2_holder **p)
1356 {
1357         unsigned int x;
1358         int error = 0;
1359
1360         for (x = 0; x < num_gh; x++)
1361                 p[x] = &ghs[x];
1362
1363         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1364
1365         for (x = 0; x < num_gh; x++) {
1366                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1367
1368                 error = gfs2_glock_nq(p[x]);
1369                 if (error) {
1370                         while (x--)
1371                                 gfs2_glock_dq(p[x]);
1372                         break;
1373                 }
1374         }
1375
1376         return error;
1377 }
1378
1379 /**
1380  * gfs2_glock_nq_m - acquire multiple glocks
1381  * @num_gh: the number of structures
1382  * @ghs: an array of struct gfs2_holder structures
1383  *
1384  *
1385  * Returns: 0 on success (all glocks acquired),
1386  *          errno on failure (no glocks acquired)
1387  */
1388
1389 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1390 {
1391         struct gfs2_holder *tmp[4];
1392         struct gfs2_holder **pph = tmp;
1393         int error = 0;
1394
1395         switch(num_gh) {
1396         case 0:
1397                 return 0;
1398         case 1:
1399                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1400                 return gfs2_glock_nq(ghs);
1401         default:
1402                 if (num_gh <= 4)
1403                         break;
1404                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1405                 if (!pph)
1406                         return -ENOMEM;
1407         }
1408
1409         error = nq_m_sync(num_gh, ghs, pph);
1410
1411         if (pph != tmp)
1412                 kfree(pph);
1413
1414         return error;
1415 }
1416
1417 /**
1418  * gfs2_glock_dq_m - release multiple glocks
1419  * @num_gh: the number of structures
1420  * @ghs: an array of struct gfs2_holder structures
1421  *
1422  */
1423
1424 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1425 {
1426         unsigned int x;
1427
1428         for (x = 0; x < num_gh; x++)
1429                 gfs2_glock_dq(&ghs[x]);
1430 }
1431
1432 /**
1433  * gfs2_glock_dq_uninit_m - release multiple glocks
1434  * @num_gh: the number of structures
1435  * @ghs: an array of struct gfs2_holder structures
1436  *
1437  */
1438
1439 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1440 {
1441         unsigned int x;
1442
1443         for (x = 0; x < num_gh; x++)
1444                 gfs2_glock_dq_uninit(&ghs[x]);
1445 }
1446
1447 static int gfs2_lm_hold_lvb(struct gfs2_sbd *sdp, void *lock, char **lvbp)
1448 {
1449         int error = -EIO;
1450         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1451                 error = sdp->sd_lockstruct.ls_ops->lm_hold_lvb(lock, lvbp);
1452         return error;
1453 }
1454
1455 /**
1456  * gfs2_lvb_hold - attach a LVB from a glock
1457  * @gl: The glock in question
1458  *
1459  */
1460
1461 int gfs2_lvb_hold(struct gfs2_glock *gl)
1462 {
1463         int error;
1464
1465         gfs2_glmutex_lock(gl);
1466
1467         if (!atomic_read(&gl->gl_lvb_count)) {
1468                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1469                 if (error) {
1470                         gfs2_glmutex_unlock(gl);
1471                         return error;
1472                 }
1473                 gfs2_glock_hold(gl);
1474         }
1475         atomic_inc(&gl->gl_lvb_count);
1476
1477         gfs2_glmutex_unlock(gl);
1478
1479         return 0;
1480 }
1481
1482 /**
1483  * gfs2_lvb_unhold - detach a LVB from a glock
1484  * @gl: The glock in question
1485  *
1486  */
1487
1488 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1489 {
1490         struct gfs2_sbd *sdp = gl->gl_sbd;
1491
1492         gfs2_glock_hold(gl);
1493         gfs2_glmutex_lock(gl);
1494
1495         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1496         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1497                 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1498                         sdp->sd_lockstruct.ls_ops->lm_unhold_lvb(gl->gl_lock, gl->gl_lvb);
1499                 gl->gl_lvb = NULL;
1500                 gfs2_glock_put(gl);
1501         }
1502
1503         gfs2_glmutex_unlock(gl);
1504         gfs2_glock_put(gl);
1505 }
1506
1507 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1508                         unsigned int state)
1509 {
1510         struct gfs2_glock *gl;
1511         unsigned long delay = 0;
1512         unsigned long holdtime;
1513         unsigned long now = jiffies;
1514
1515         gl = gfs2_glock_find(sdp, name);
1516         if (!gl)
1517                 return;
1518
1519         holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1520         if (time_before(now, holdtime))
1521                 delay = holdtime - now;
1522
1523         handle_callback(gl, state, 1, delay);
1524         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1525                 gfs2_glock_put(gl);
1526 }
1527
1528 /**
1529  * gfs2_glock_cb - Callback used by locking module
1530  * @sdp: Pointer to the superblock
1531  * @type: Type of callback
1532  * @data: Type dependent data pointer
1533  *
1534  * Called by the locking module when it wants to tell us something.
1535  * Either we need to drop a lock, one of our ASYNC requests completed, or
1536  * a journal from another client needs to be recovered.
1537  */
1538
1539 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1540 {
1541         struct gfs2_sbd *sdp = cb_data;
1542
1543         switch (type) {
1544         case LM_CB_NEED_E:
1545                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1546                 return;
1547
1548         case LM_CB_NEED_D:
1549                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1550                 return;
1551
1552         case LM_CB_NEED_S:
1553                 blocking_cb(sdp, data, LM_ST_SHARED);
1554                 return;
1555
1556         case LM_CB_ASYNC: {
1557                 struct lm_async_cb *async = data;
1558                 struct gfs2_glock *gl;
1559
1560                 down_read(&gfs2_umount_flush_sem);
1561                 gl = gfs2_glock_find(sdp, &async->lc_name);
1562                 if (gfs2_assert_warn(sdp, gl))
1563                         return;
1564                 xmote_bh(gl, async->lc_ret);
1565                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1566                         gfs2_glock_put(gl);
1567                 up_read(&gfs2_umount_flush_sem);
1568                 return;
1569         }
1570
1571         case LM_CB_NEED_RECOVERY:
1572                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1573                 if (sdp->sd_recoverd_process)
1574                         wake_up_process(sdp->sd_recoverd_process);
1575                 return;
1576
1577         case LM_CB_DROPLOCKS:
1578                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1579                 gfs2_quota_scan(sdp);
1580                 return;
1581
1582         default:
1583                 gfs2_assert_warn(sdp, 0);
1584                 return;
1585         }
1586 }
1587
1588 /**
1589  * demote_ok - Check to see if it's ok to unlock a glock
1590  * @gl: the glock
1591  *
1592  * Returns: 1 if it's ok
1593  */
1594
1595 static int demote_ok(struct gfs2_glock *gl)
1596 {
1597         const struct gfs2_glock_operations *glops = gl->gl_ops;
1598         int demote = 1;
1599
1600         if (test_bit(GLF_STICKY, &gl->gl_flags))
1601                 demote = 0;
1602         else if (glops->go_demote_ok)
1603                 demote = glops->go_demote_ok(gl);
1604
1605         return demote;
1606 }
1607
1608 /**
1609  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1610  * @gl: the glock
1611  *
1612  */
1613
1614 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1615 {
1616         struct gfs2_sbd *sdp = gl->gl_sbd;
1617
1618         spin_lock(&sdp->sd_reclaim_lock);
1619         if (list_empty(&gl->gl_reclaim)) {
1620                 gfs2_glock_hold(gl);
1621                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1622                 atomic_inc(&sdp->sd_reclaim_count);
1623                 spin_unlock(&sdp->sd_reclaim_lock);
1624                 wake_up(&sdp->sd_reclaim_wq);
1625         } else
1626                 spin_unlock(&sdp->sd_reclaim_lock);
1627 }
1628
1629 /**
1630  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1631  * @sdp: the filesystem
1632  *
1633  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1634  * different glock and we notice that there are a lot of glocks in the
1635  * reclaim list.
1636  *
1637  */
1638
1639 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1640 {
1641         struct gfs2_glock *gl;
1642
1643         spin_lock(&sdp->sd_reclaim_lock);
1644         if (list_empty(&sdp->sd_reclaim_list)) {
1645                 spin_unlock(&sdp->sd_reclaim_lock);
1646                 return;
1647         }
1648         gl = list_entry(sdp->sd_reclaim_list.next,
1649                         struct gfs2_glock, gl_reclaim);
1650         list_del_init(&gl->gl_reclaim);
1651         spin_unlock(&sdp->sd_reclaim_lock);
1652
1653         atomic_dec(&sdp->sd_reclaim_count);
1654         atomic_inc(&sdp->sd_reclaimed);
1655
1656         if (gfs2_glmutex_trylock(gl)) {
1657                 if (list_empty(&gl->gl_holders) &&
1658                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1659                         handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1660                 gfs2_glmutex_unlock(gl);
1661         }
1662
1663         gfs2_glock_put(gl);
1664 }
1665
1666 /**
1667  * examine_bucket - Call a function for glock in a hash bucket
1668  * @examiner: the function
1669  * @sdp: the filesystem
1670  * @bucket: the bucket
1671  *
1672  * Returns: 1 if the bucket has entries
1673  */
1674
1675 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1676                           unsigned int hash)
1677 {
1678         struct gfs2_glock *gl, *prev = NULL;
1679         int has_entries = 0;
1680         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1681
1682         read_lock(gl_lock_addr(hash));
1683         /* Can't use hlist_for_each_entry - don't want prefetch here */
1684         if (hlist_empty(head))
1685                 goto out;
1686         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1687         while(1) {
1688                 if (!sdp || gl->gl_sbd == sdp) {
1689                         gfs2_glock_hold(gl);
1690                         read_unlock(gl_lock_addr(hash));
1691                         if (prev)
1692                                 gfs2_glock_put(prev);
1693                         prev = gl;
1694                         examiner(gl);
1695                         has_entries = 1;
1696                         read_lock(gl_lock_addr(hash));
1697                 }
1698                 if (gl->gl_list.next == NULL)
1699                         break;
1700                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1701         }
1702 out:
1703         read_unlock(gl_lock_addr(hash));
1704         if (prev)
1705                 gfs2_glock_put(prev);
1706         cond_resched();
1707         return has_entries;
1708 }
1709
1710 /**
1711  * scan_glock - look at a glock and see if we can reclaim it
1712  * @gl: the glock to look at
1713  *
1714  */
1715
1716 static void scan_glock(struct gfs2_glock *gl)
1717 {
1718         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1719                 return;
1720
1721         if (gfs2_glmutex_trylock(gl)) {
1722                 if (list_empty(&gl->gl_holders) &&
1723                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1724                         goto out_schedule;
1725                 gfs2_glmutex_unlock(gl);
1726         }
1727         return;
1728
1729 out_schedule:
1730         gfs2_glmutex_unlock(gl);
1731         gfs2_glock_schedule_for_reclaim(gl);
1732 }
1733
1734 /**
1735  * clear_glock - look at a glock and see if we can free it from glock cache
1736  * @gl: the glock to look at
1737  *
1738  */
1739
1740 static void clear_glock(struct gfs2_glock *gl)
1741 {
1742         struct gfs2_sbd *sdp = gl->gl_sbd;
1743         int released;
1744
1745         spin_lock(&sdp->sd_reclaim_lock);
1746         if (!list_empty(&gl->gl_reclaim)) {
1747                 list_del_init(&gl->gl_reclaim);
1748                 atomic_dec(&sdp->sd_reclaim_count);
1749                 spin_unlock(&sdp->sd_reclaim_lock);
1750                 released = gfs2_glock_put(gl);
1751                 gfs2_assert(sdp, !released);
1752         } else {
1753                 spin_unlock(&sdp->sd_reclaim_lock);
1754         }
1755
1756         if (gfs2_glmutex_trylock(gl)) {
1757                 if (list_empty(&gl->gl_holders) &&
1758                     gl->gl_state != LM_ST_UNLOCKED)
1759                         handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1760                 gfs2_glmutex_unlock(gl);
1761         }
1762 }
1763
1764 /**
1765  * gfs2_gl_hash_clear - Empty out the glock hash table
1766  * @sdp: the filesystem
1767  * @wait: wait until it's all gone
1768  *
1769  * Called when unmounting the filesystem, or when inter-node lock manager
1770  * requests DROPLOCKS because it is running out of capacity.
1771  */
1772
1773 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1774 {
1775         unsigned long t;
1776         unsigned int x;
1777         int cont;
1778
1779         t = jiffies;
1780
1781         for (;;) {
1782                 cont = 0;
1783                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1784                         if (examine_bucket(clear_glock, sdp, x))
1785                                 cont = 1;
1786                 }
1787
1788                 if (!wait || !cont)
1789                         break;
1790
1791                 if (time_after_eq(jiffies,
1792                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1793                         fs_warn(sdp, "Unmount seems to be stalled. "
1794                                      "Dumping lock state...\n");
1795                         gfs2_dump_lockstate(sdp);
1796                         t = jiffies;
1797                 }
1798
1799                 down_write(&gfs2_umount_flush_sem);
1800                 invalidate_inodes(sdp->sd_vfs);
1801                 up_write(&gfs2_umount_flush_sem);
1802                 msleep(10);
1803         }
1804 }
1805
1806 /*
1807  *  Diagnostic routines to help debug distributed deadlock
1808  */
1809
1810 static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
1811                               unsigned long address)
1812 {
1813         char buffer[KSYM_SYMBOL_LEN];
1814
1815         sprint_symbol(buffer, address);
1816         print_dbg(gi, fmt, buffer);
1817 }
1818
1819 /**
1820  * dump_holder - print information about a glock holder
1821  * @str: a string naming the type of holder
1822  * @gh: the glock holder
1823  *
1824  * Returns: 0 on success, -ENOBUFS when we run out of space
1825  */
1826
1827 static int dump_holder(struct glock_iter *gi, char *str,
1828                        struct gfs2_holder *gh)
1829 {
1830         unsigned int x;
1831         struct task_struct *gh_owner;
1832
1833         print_dbg(gi, "  %s\n", str);
1834         if (gh->gh_owner_pid) {
1835                 print_dbg(gi, "    owner = %ld ",
1836                                 (long)pid_nr(gh->gh_owner_pid));
1837                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1838                 if (gh_owner)
1839                         print_dbg(gi, "(%s)\n", gh_owner->comm);
1840                 else
1841                         print_dbg(gi, "(ended)\n");
1842         } else
1843                 print_dbg(gi, "    owner = -1\n");
1844         print_dbg(gi, "    gh_state = %u\n", gh->gh_state);
1845         print_dbg(gi, "    gh_flags =");
1846         for (x = 0; x < 32; x++)
1847                 if (gh->gh_flags & (1 << x))
1848                         print_dbg(gi, " %u", x);
1849         print_dbg(gi, " \n");
1850         print_dbg(gi, "    error = %d\n", gh->gh_error);
1851         print_dbg(gi, "    gh_iflags =");
1852         for (x = 0; x < 32; x++)
1853                 if (test_bit(x, &gh->gh_iflags))
1854                         print_dbg(gi, " %u", x);
1855         print_dbg(gi, " \n");
1856         gfs2_print_symbol(gi, "    initialized at: %s\n", gh->gh_ip);
1857
1858         return 0;
1859 }
1860
1861 /**
1862  * dump_inode - print information about an inode
1863  * @ip: the inode
1864  *
1865  * Returns: 0 on success, -ENOBUFS when we run out of space
1866  */
1867
1868 static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
1869 {
1870         unsigned int x;
1871
1872         print_dbg(gi, "  Inode:\n");
1873         print_dbg(gi, "    num = %llu/%llu\n",
1874                   (unsigned long long)ip->i_no_formal_ino,
1875                   (unsigned long long)ip->i_no_addr);
1876         print_dbg(gi, "    type = %u\n", IF2DT(ip->i_inode.i_mode));
1877         print_dbg(gi, "    i_flags =");
1878         for (x = 0; x < 32; x++)
1879                 if (test_bit(x, &ip->i_flags))
1880                         print_dbg(gi, " %u", x);
1881         print_dbg(gi, " \n");
1882         return 0;
1883 }
1884
1885 /**
1886  * dump_glock - print information about a glock
1887  * @gl: the glock
1888  * @count: where we are in the buffer
1889  *
1890  * Returns: 0 on success, -ENOBUFS when we run out of space
1891  */
1892
1893 static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
1894 {
1895         struct gfs2_holder *gh;
1896         unsigned int x;
1897         int error = -ENOBUFS;
1898         struct task_struct *gl_owner;
1899
1900         spin_lock(&gl->gl_spin);
1901
1902         print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
1903                    (unsigned long long)gl->gl_name.ln_number);
1904         print_dbg(gi, "  gl_flags =");
1905         for (x = 0; x < 32; x++) {
1906                 if (test_bit(x, &gl->gl_flags))
1907                         print_dbg(gi, " %u", x);
1908         }
1909         if (!test_bit(GLF_LOCK, &gl->gl_flags))
1910                 print_dbg(gi, " (unlocked)");
1911         print_dbg(gi, " \n");
1912         print_dbg(gi, "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
1913         print_dbg(gi, "  gl_state = %u\n", gl->gl_state);
1914         if (gl->gl_owner_pid) {
1915                 gl_owner = pid_task(gl->gl_owner_pid, PIDTYPE_PID);
1916                 if (gl_owner)
1917                         print_dbg(gi, "  gl_owner = pid %d (%s)\n",
1918                                   pid_nr(gl->gl_owner_pid), gl_owner->comm);
1919                 else
1920                         print_dbg(gi, "  gl_owner = %d (ended)\n",
1921                                   pid_nr(gl->gl_owner_pid));
1922         } else
1923                 print_dbg(gi, "  gl_owner = -1\n");
1924         print_dbg(gi, "  gl_ip = %lu\n", gl->gl_ip);
1925         print_dbg(gi, "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1926         print_dbg(gi, "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1927         print_dbg(gi, "  object = %s\n", (gl->gl_object) ? "yes" : "no");
1928         print_dbg(gi, "  reclaim = %s\n",
1929                    (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1930         if (gl->gl_aspace)
1931                 print_dbg(gi, "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1932                            gl->gl_aspace->i_mapping->nrpages);
1933         else
1934                 print_dbg(gi, "  aspace = no\n");
1935         print_dbg(gi, "  ail = %d\n", atomic_read(&gl->gl_ail_count));
1936         if (gl->gl_req_gh) {
1937                 error = dump_holder(gi, "Request", gl->gl_req_gh);
1938                 if (error)
1939                         goto out;
1940         }
1941         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1942                 error = dump_holder(gi, "Holder", gh);
1943                 if (error)
1944                         goto out;
1945         }
1946         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1947                 error = dump_holder(gi, "Waiter1", gh);
1948                 if (error)
1949                         goto out;
1950         }
1951         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1952                 error = dump_holder(gi, "Waiter3", gh);
1953                 if (error)
1954                         goto out;
1955         }
1956         if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
1957                 print_dbg(gi, "  Demotion req to state %u (%llu uS ago)\n",
1958                           gl->gl_demote_state, (unsigned long long)
1959                           (jiffies - gl->gl_demote_time)*(1000000/HZ));
1960         }
1961         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1962                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1963                         list_empty(&gl->gl_holders)) {
1964                         error = dump_inode(gi, gl->gl_object);
1965                         if (error)
1966                                 goto out;
1967                 } else {
1968                         error = -ENOBUFS;
1969                         print_dbg(gi, "  Inode: busy\n");
1970                 }
1971         }
1972
1973         error = 0;
1974
1975 out:
1976         spin_unlock(&gl->gl_spin);
1977         return error;
1978 }
1979
1980 /**
1981  * gfs2_dump_lockstate - print out the current lockstate
1982  * @sdp: the filesystem
1983  * @ub: the buffer to copy the information into
1984  *
1985  * If @ub is NULL, dump the lockstate to the console.
1986  *
1987  */
1988
1989 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1990 {
1991         struct gfs2_glock *gl;
1992         struct hlist_node *h;
1993         unsigned int x;
1994         int error = 0;
1995
1996         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1997
1998                 read_lock(gl_lock_addr(x));
1999
2000                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
2001                         if (gl->gl_sbd != sdp)
2002                                 continue;
2003
2004                         error = dump_glock(NULL, gl);
2005                         if (error)
2006                                 break;
2007                 }
2008
2009                 read_unlock(gl_lock_addr(x));
2010
2011                 if (error)
2012                         break;
2013         }
2014
2015
2016         return error;
2017 }
2018
2019 /**
2020  * gfs2_scand - Look for cached glocks and inodes to toss from memory
2021  * @sdp: Pointer to GFS2 superblock
2022  *
2023  * One of these daemons runs, finding candidates to add to sd_reclaim_list.
2024  * See gfs2_glockd()
2025  */
2026
2027 static int gfs2_scand(void *data)
2028 {
2029         unsigned x;
2030         unsigned delay;
2031
2032         while (!kthread_should_stop()) {
2033                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2034                         examine_bucket(scan_glock, NULL, x);
2035                 if (freezing(current))
2036                         refrigerator();
2037                 delay = scand_secs;
2038                 if (delay < 1)
2039                         delay = 1;
2040                 schedule_timeout_interruptible(delay * HZ);
2041         }
2042
2043         return 0;
2044 }
2045
2046
2047
2048 int __init gfs2_glock_init(void)
2049 {
2050         unsigned i;
2051         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2052                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2053         }
2054 #ifdef GL_HASH_LOCK_SZ
2055         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2056                 rwlock_init(&gl_hash_locks[i]);
2057         }
2058 #endif
2059
2060         scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
2061         if (IS_ERR(scand_process))
2062                 return PTR_ERR(scand_process);
2063
2064         glock_workqueue = create_workqueue("glock_workqueue");
2065         if (IS_ERR(glock_workqueue)) {
2066                 kthread_stop(scand_process);
2067                 return PTR_ERR(glock_workqueue);
2068         }
2069
2070         return 0;
2071 }
2072
2073 void gfs2_glock_exit(void)
2074 {
2075         destroy_workqueue(glock_workqueue);
2076         kthread_stop(scand_process);
2077 }
2078
2079 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
2080 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
2081
2082 static int gfs2_glock_iter_next(struct glock_iter *gi)
2083 {
2084         struct gfs2_glock *gl;
2085
2086 restart:
2087         read_lock(gl_lock_addr(gi->hash));
2088         gl = gi->gl;
2089         if (gl) {
2090                 gi->gl = hlist_entry(gl->gl_list.next,
2091                                      struct gfs2_glock, gl_list);
2092                 if (gi->gl)
2093                         gfs2_glock_hold(gi->gl);
2094         }
2095         read_unlock(gl_lock_addr(gi->hash));
2096         if (gl)
2097                 gfs2_glock_put(gl);
2098         if (gl && gi->gl == NULL)
2099                 gi->hash++;
2100         while(gi->gl == NULL) {
2101                 if (gi->hash >= GFS2_GL_HASH_SIZE)
2102                         return 1;
2103                 read_lock(gl_lock_addr(gi->hash));
2104                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
2105                                      struct gfs2_glock, gl_list);
2106                 if (gi->gl)
2107                         gfs2_glock_hold(gi->gl);
2108                 read_unlock(gl_lock_addr(gi->hash));
2109                 gi->hash++;
2110         }
2111
2112         if (gi->sdp != gi->gl->gl_sbd)
2113                 goto restart;
2114
2115         return 0;
2116 }
2117
2118 static void gfs2_glock_iter_free(struct glock_iter *gi)
2119 {
2120         if (gi->gl)
2121                 gfs2_glock_put(gi->gl);
2122         kfree(gi);
2123 }
2124
2125 static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
2126 {
2127         struct glock_iter *gi;
2128
2129         gi = kmalloc(sizeof (*gi), GFP_KERNEL);
2130         if (!gi)
2131                 return NULL;
2132
2133         gi->sdp = sdp;
2134         gi->hash = 0;
2135         gi->seq = NULL;
2136         gi->gl = NULL;
2137         memset(gi->string, 0, sizeof(gi->string));
2138
2139         if (gfs2_glock_iter_next(gi)) {
2140                 gfs2_glock_iter_free(gi);
2141                 return NULL;
2142         }
2143
2144         return gi;
2145 }
2146
2147 static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
2148 {
2149         struct glock_iter *gi;
2150         loff_t n = *pos;
2151
2152         gi = gfs2_glock_iter_init(file->private);
2153         if (!gi)
2154                 return NULL;
2155
2156         while(n--) {
2157                 if (gfs2_glock_iter_next(gi)) {
2158                         gfs2_glock_iter_free(gi);
2159                         return NULL;
2160                 }
2161         }
2162
2163         return gi;
2164 }
2165
2166 static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
2167                                  loff_t *pos)
2168 {
2169         struct glock_iter *gi = iter_ptr;
2170
2171         (*pos)++;
2172
2173         if (gfs2_glock_iter_next(gi)) {
2174                 gfs2_glock_iter_free(gi);
2175                 return NULL;
2176         }
2177
2178         return gi;
2179 }
2180
2181 static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
2182 {
2183         struct glock_iter *gi = iter_ptr;
2184         if (gi)
2185                 gfs2_glock_iter_free(gi);
2186 }
2187
2188 static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
2189 {
2190         struct glock_iter *gi = iter_ptr;
2191
2192         gi->seq = file;
2193         dump_glock(gi, gi->gl);
2194
2195         return 0;
2196 }
2197
2198 static const struct seq_operations gfs2_glock_seq_ops = {
2199         .start = gfs2_glock_seq_start,
2200         .next  = gfs2_glock_seq_next,
2201         .stop  = gfs2_glock_seq_stop,
2202         .show  = gfs2_glock_seq_show,
2203 };
2204
2205 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
2206 {
2207         struct seq_file *seq;
2208         int ret;
2209
2210         ret = seq_open(file, &gfs2_glock_seq_ops);
2211         if (ret)
2212                 return ret;
2213
2214         seq = file->private_data;
2215         seq->private = inode->i_private;
2216
2217         return 0;
2218 }
2219
2220 static const struct file_operations gfs2_debug_fops = {
2221         .owner   = THIS_MODULE,
2222         .open    = gfs2_debugfs_open,
2223         .read    = seq_read,
2224         .llseek  = seq_lseek,
2225         .release = seq_release
2226 };
2227
2228 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2229 {
2230         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2231         if (!sdp->debugfs_dir)
2232                 return -ENOMEM;
2233         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
2234                                                          S_IFREG | S_IRUGO,
2235                                                          sdp->debugfs_dir, sdp,
2236                                                          &gfs2_debug_fops);
2237         if (!sdp->debugfs_dentry_glocks)
2238                 return -ENOMEM;
2239
2240         return 0;
2241 }
2242
2243 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2244 {
2245         if (sdp && sdp->debugfs_dir) {
2246                 if (sdp->debugfs_dentry_glocks) {
2247                         debugfs_remove(sdp->debugfs_dentry_glocks);
2248                         sdp->debugfs_dentry_glocks = NULL;
2249                 }
2250                 debugfs_remove(sdp->debugfs_dir);
2251                 sdp->debugfs_dir = NULL;
2252         }
2253 }
2254
2255 int gfs2_register_debugfs(void)
2256 {
2257         gfs2_root = debugfs_create_dir("gfs2", NULL);
2258         return gfs2_root ? 0 : -ENOMEM;
2259 }
2260
2261 void gfs2_unregister_debugfs(void)
2262 {
2263         debugfs_remove(gfs2_root);
2264         gfs2_root = NULL;
2265 }