[GFS2] [-mm patch] fs/gfs2/: possible cleanups
[linux-2.6.git] / fs / gfs2 / glock.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2005 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 v.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/kref.h>
19 #include <linux/kallsyms.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <asm/semaphore.h>
22 #include <asm/uaccess.h>
23
24 #include "gfs2.h"
25 #include "lm_interface.h"
26 #include "incore.h"
27 #include "glock.h"
28 #include "glops.h"
29 #include "inode.h"
30 #include "lm.h"
31 #include "lops.h"
32 #include "meta_io.h"
33 #include "quota.h"
34 #include "super.h"
35 #include "util.h"
36
37 /*  Must be kept in sync with the beginning of struct gfs2_glock  */
38 struct glock_plug {
39         struct list_head gl_list;
40         unsigned long gl_flags;
41 };
42
43 struct greedy {
44         struct gfs2_holder gr_gh;
45         struct work_struct gr_work;
46 };
47
48 typedef void (*glock_examiner) (struct gfs2_glock * gl);
49
50 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
51
52 /**
53  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
54  * @actual: the current state of the lock
55  * @requested: the lock state that was requested by the caller
56  * @flags: the modifier flags passed in by the caller
57  *
58  * Returns: 1 if the locks are compatible, 0 otherwise
59  */
60
61 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
62                                    int flags)
63 {
64         if (actual == requested)
65                 return 1;
66
67         if (flags & GL_EXACT)
68                 return 0;
69
70         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
71                 return 1;
72
73         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
74                 return 1;
75
76         return 0;
77 }
78
79 /**
80  * gl_hash() - Turn glock number into hash bucket number
81  * @lock: The glock number
82  *
83  * Returns: The number of the corresponding hash bucket
84  */
85
86 static unsigned int gl_hash(struct lm_lockname *name)
87 {
88         unsigned int h;
89
90         h = jhash(&name->ln_number, sizeof(uint64_t), 0);
91         h = jhash(&name->ln_type, sizeof(unsigned int), h);
92         h &= GFS2_GL_HASH_MASK;
93
94         return h;
95 }
96
97 /**
98  * glock_free() - Perform a few checks and then release struct gfs2_glock
99  * @gl: The glock to release
100  *
101  * Also calls lock module to release its internal structure for this glock.
102  *
103  */
104
105 static void glock_free(struct gfs2_glock *gl)
106 {
107         struct gfs2_sbd *sdp = gl->gl_sbd;
108         struct inode *aspace = gl->gl_aspace;
109
110         gfs2_lm_put_lock(sdp, gl->gl_lock);
111
112         if (aspace)
113                 gfs2_aspace_put(aspace);
114
115         kmem_cache_free(gfs2_glock_cachep, gl);
116 }
117
118 /**
119  * gfs2_glock_hold() - increment reference count on glock
120  * @gl: The glock to hold
121  *
122  */
123
124 void gfs2_glock_hold(struct gfs2_glock *gl)
125 {
126         kref_get(&gl->gl_ref);
127 }
128
129 /* All work is done after the return from kref_put() so we
130    can release the write_lock before the free. */
131
132 static void kill_glock(struct kref *kref)
133 {
134         struct gfs2_glock *gl = container_of(kref, struct gfs2_glock, gl_ref);
135         struct gfs2_sbd *sdp = gl->gl_sbd;
136
137         gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
138         gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
139         gfs2_assert(sdp, list_empty(&gl->gl_holders));
140         gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
141         gfs2_assert(sdp, list_empty(&gl->gl_waiters2));
142         gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
143 }
144
145 /**
146  * gfs2_glock_put() - Decrement reference count on glock
147  * @gl: The glock to put
148  *
149  */
150
151 int gfs2_glock_put(struct gfs2_glock *gl)
152 {
153         struct gfs2_sbd *sdp = gl->gl_sbd;
154         struct gfs2_gl_hash_bucket *bucket = gl->gl_bucket;
155         int rv = 0;
156
157         mutex_lock(&sdp->sd_invalidate_inodes_mutex);
158
159         write_lock(&bucket->hb_lock);
160         if (kref_put(&gl->gl_ref, kill_glock)) {
161                 list_del_init(&gl->gl_list);
162                 write_unlock(&bucket->hb_lock);
163                 BUG_ON(spin_is_locked(&gl->gl_spin));
164                 glock_free(gl);
165                 rv = 1;
166                 goto out;
167         }
168         write_unlock(&bucket->hb_lock);
169  out:
170         mutex_unlock(&sdp->sd_invalidate_inodes_mutex);
171         return rv;
172 }
173
174 /**
175  * queue_empty - check to see if a glock's queue is empty
176  * @gl: the glock
177  * @head: the head of the queue to check
178  *
179  * This function protects the list in the event that a process already
180  * has a holder on the list and is adding a second holder for itself.
181  * The glmutex lock is what generally prevents processes from working
182  * on the same glock at once, but the special case of adding a second
183  * holder for yourself ("recursive" locking) doesn't involve locking
184  * glmutex, making the spin lock necessary.
185  *
186  * Returns: 1 if the queue is empty
187  */
188
189 static inline int queue_empty(struct gfs2_glock *gl, struct list_head *head)
190 {
191         int empty;
192         spin_lock(&gl->gl_spin);
193         empty = list_empty(head);
194         spin_unlock(&gl->gl_spin);
195         return empty;
196 }
197
198 /**
199  * search_bucket() - Find struct gfs2_glock by lock number
200  * @bucket: the bucket to search
201  * @name: The lock name
202  *
203  * Returns: NULL, or the struct gfs2_glock with the requested number
204  */
205
206 static struct gfs2_glock *search_bucket(struct gfs2_gl_hash_bucket *bucket,
207                                         struct lm_lockname *name)
208 {
209         struct gfs2_glock *gl;
210
211         list_for_each_entry(gl, &bucket->hb_list, gl_list) {
212                 if (test_bit(GLF_PLUG, &gl->gl_flags))
213                         continue;
214                 if (!lm_name_equal(&gl->gl_name, name))
215                         continue;
216
217                 kref_get(&gl->gl_ref);
218
219                 return gl;
220         }
221
222         return NULL;
223 }
224
225 /**
226  * gfs2_glock_find() - Find glock by lock number
227  * @sdp: The GFS2 superblock
228  * @name: The lock name
229  *
230  * Returns: NULL, or the struct gfs2_glock with the requested number
231  */
232
233 static struct gfs2_glock *gfs2_glock_find(struct gfs2_sbd *sdp,
234                                           struct lm_lockname *name)
235 {
236         struct gfs2_gl_hash_bucket *bucket = &sdp->sd_gl_hash[gl_hash(name)];
237         struct gfs2_glock *gl;
238
239         read_lock(&bucket->hb_lock);
240         gl = search_bucket(bucket, name);
241         read_unlock(&bucket->hb_lock);
242
243         return gl;
244 }
245
246 /**
247  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
248  * @sdp: The GFS2 superblock
249  * @number: the lock number
250  * @glops: The glock_operations to use
251  * @create: If 0, don't create the glock if it doesn't exist
252  * @glp: the glock is returned here
253  *
254  * This does not lock a glock, just finds/creates structures for one.
255  *
256  * Returns: errno
257  */
258
259 int gfs2_glock_get(struct gfs2_sbd *sdp, uint64_t number,
260                    struct gfs2_glock_operations *glops, int create,
261                    struct gfs2_glock **glp)
262 {
263         struct lm_lockname name;
264         struct gfs2_glock *gl, *tmp;
265         struct gfs2_gl_hash_bucket *bucket;
266         int error;
267
268         name.ln_number = number;
269         name.ln_type = glops->go_type;
270         bucket = &sdp->sd_gl_hash[gl_hash(&name)];
271
272         read_lock(&bucket->hb_lock);
273         gl = search_bucket(bucket, &name);
274         read_unlock(&bucket->hb_lock);
275
276         if (gl || !create) {
277                 *glp = gl;
278                 return 0;
279         }
280
281         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
282         if (!gl)
283                 return -ENOMEM;
284
285         memset(gl, 0, sizeof(struct gfs2_glock));
286
287         INIT_LIST_HEAD(&gl->gl_list);
288         gl->gl_name = name;
289         kref_init(&gl->gl_ref);
290
291         spin_lock_init(&gl->gl_spin);
292
293         gl->gl_state = LM_ST_UNLOCKED;
294         INIT_LIST_HEAD(&gl->gl_holders);
295         INIT_LIST_HEAD(&gl->gl_waiters1);
296         INIT_LIST_HEAD(&gl->gl_waiters2);
297         INIT_LIST_HEAD(&gl->gl_waiters3);
298
299         gl->gl_ops = glops;
300
301         gl->gl_bucket = bucket;
302         INIT_LIST_HEAD(&gl->gl_reclaim);
303
304         gl->gl_sbd = sdp;
305
306         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
307         INIT_LIST_HEAD(&gl->gl_ail_list);
308
309         /* If this glock protects actual on-disk data or metadata blocks,
310            create a VFS inode to manage the pages/buffers holding them. */
311         if (glops == &gfs2_inode_glops ||
312             glops == &gfs2_rgrp_glops ||
313             glops == &gfs2_meta_glops) {
314                 gl->gl_aspace = gfs2_aspace_get(sdp);
315                 if (!gl->gl_aspace) {
316                         error = -ENOMEM;
317                         goto fail;
318                 }
319         }
320
321         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
322         if (error)
323                 goto fail_aspace;
324
325         write_lock(&bucket->hb_lock);
326         tmp = search_bucket(bucket, &name);
327         if (tmp) {
328                 write_unlock(&bucket->hb_lock);
329                 glock_free(gl);
330                 gl = tmp;
331         } else {
332                 list_add_tail(&gl->gl_list, &bucket->hb_list);
333                 write_unlock(&bucket->hb_lock);
334         }
335
336         *glp = gl;
337
338         return 0;
339
340  fail_aspace:
341         if (gl->gl_aspace)
342                 gfs2_aspace_put(gl->gl_aspace);
343
344  fail:
345         kmem_cache_free(gfs2_glock_cachep, gl); 
346
347         return error;
348 }
349
350 /**
351  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
352  * @gl: the glock
353  * @state: the state we're requesting
354  * @flags: the modifier flags
355  * @gh: the holder structure
356  *
357  */
358
359 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
360                       struct gfs2_holder *gh)
361 {
362         INIT_LIST_HEAD(&gh->gh_list);
363         gh->gh_gl = gl;
364         gh->gh_ip = (unsigned long)__builtin_return_address(0);
365         gh->gh_owner = current;
366         gh->gh_state = state;
367         gh->gh_flags = flags;
368         gh->gh_error = 0;
369         gh->gh_iflags = 0;
370         init_completion(&gh->gh_wait);
371
372         if (gh->gh_state == LM_ST_EXCLUSIVE)
373                 gh->gh_flags |= GL_LOCAL_EXCL;
374
375         gfs2_glock_hold(gl);
376 }
377
378 /**
379  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
380  * @state: the state we're requesting
381  * @flags: the modifier flags
382  * @gh: the holder structure
383  *
384  * Don't mess with the glock.
385  *
386  */
387
388 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
389 {
390         gh->gh_state = state;
391         gh->gh_flags = flags;
392         if (gh->gh_state == LM_ST_EXCLUSIVE)
393                 gh->gh_flags |= GL_LOCAL_EXCL;
394
395         gh->gh_iflags &= 1 << HIF_ALLOCED;
396         gh->gh_ip = (unsigned long)__builtin_return_address(0);
397 }
398
399 /**
400  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
401  * @gh: the holder structure
402  *
403  */
404
405 void gfs2_holder_uninit(struct gfs2_holder *gh)
406 {
407         gfs2_glock_put(gh->gh_gl);
408         gh->gh_gl = NULL;
409         gh->gh_ip = 0;
410 }
411
412 /**
413  * gfs2_holder_get - get a struct gfs2_holder structure
414  * @gl: the glock
415  * @state: the state we're requesting
416  * @flags: the modifier flags
417  * @gfp_flags: __GFP_NOFAIL
418  *
419  * Figure out how big an impact this function has.  Either:
420  * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
421  * 2) Leave it like it is
422  *
423  * Returns: the holder structure, NULL on ENOMEM
424  */
425
426 static struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl,
427                                            unsigned int state,
428                                            int flags, gfp_t gfp_flags)
429 {
430         struct gfs2_holder *gh;
431
432         gh = kmalloc(sizeof(struct gfs2_holder), gfp_flags);
433         if (!gh)
434                 return NULL;
435
436         gfs2_holder_init(gl, state, flags, gh);
437         set_bit(HIF_ALLOCED, &gh->gh_iflags);
438         gh->gh_ip = (unsigned long)__builtin_return_address(0);
439         return gh;
440 }
441
442 /**
443  * gfs2_holder_put - get rid of a struct gfs2_holder structure
444  * @gh: the holder structure
445  *
446  */
447
448 static void gfs2_holder_put(struct gfs2_holder *gh)
449 {
450         gfs2_holder_uninit(gh);
451         kfree(gh);
452 }
453
454 /**
455  * rq_mutex - process a mutex request in the queue
456  * @gh: the glock holder
457  *
458  * Returns: 1 if the queue is blocked
459  */
460
461 static int rq_mutex(struct gfs2_holder *gh)
462 {
463         struct gfs2_glock *gl = gh->gh_gl;
464
465         list_del_init(&gh->gh_list);
466         /*  gh->gh_error never examined.  */
467         set_bit(GLF_LOCK, &gl->gl_flags);
468         complete(&gh->gh_wait);
469
470         return 1;
471 }
472
473 /**
474  * rq_promote - process a promote request in the queue
475  * @gh: the glock holder
476  *
477  * Acquire a new inter-node lock, or change a lock state to more restrictive.
478  *
479  * Returns: 1 if the queue is blocked
480  */
481
482 static int rq_promote(struct gfs2_holder *gh)
483 {
484         struct gfs2_glock *gl = gh->gh_gl;
485         struct gfs2_sbd *sdp = gl->gl_sbd;
486         struct gfs2_glock_operations *glops = gl->gl_ops;
487
488         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
489                 if (list_empty(&gl->gl_holders)) {
490                         gl->gl_req_gh = gh;
491                         set_bit(GLF_LOCK, &gl->gl_flags);
492                         spin_unlock(&gl->gl_spin);
493
494                         if (atomic_read(&sdp->sd_reclaim_count) >
495                             gfs2_tune_get(sdp, gt_reclaim_limit) &&
496                             !(gh->gh_flags & LM_FLAG_PRIORITY)) {
497                                 gfs2_reclaim_glock(sdp);
498                                 gfs2_reclaim_glock(sdp);
499                         }
500
501                         glops->go_xmote_th(gl, gh->gh_state,
502                                            gh->gh_flags);
503
504                         spin_lock(&gl->gl_spin);
505                 }
506                 return 1;
507         }
508
509         if (list_empty(&gl->gl_holders)) {
510                 set_bit(HIF_FIRST, &gh->gh_iflags);
511                 set_bit(GLF_LOCK, &gl->gl_flags);
512         } else {
513                 struct gfs2_holder *next_gh;
514                 if (gh->gh_flags & GL_LOCAL_EXCL)
515                         return 1;
516                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
517                                      gh_list);
518                 if (next_gh->gh_flags & GL_LOCAL_EXCL)
519                          return 1;
520         }
521
522         list_move_tail(&gh->gh_list, &gl->gl_holders);
523         gh->gh_error = 0;
524         set_bit(HIF_HOLDER, &gh->gh_iflags);
525
526         complete(&gh->gh_wait);
527
528         return 0;
529 }
530
531 /**
532  * rq_demote - process a demote request in the queue
533  * @gh: the glock holder
534  *
535  * Returns: 1 if the queue is blocked
536  */
537
538 static int rq_demote(struct gfs2_holder *gh)
539 {
540         struct gfs2_glock *gl = gh->gh_gl;
541         struct gfs2_glock_operations *glops = gl->gl_ops;
542
543         if (!list_empty(&gl->gl_holders))
544                 return 1;
545
546         if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
547                 list_del_init(&gh->gh_list);
548                 gh->gh_error = 0;
549                 spin_unlock(&gl->gl_spin);
550                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
551                         gfs2_holder_put(gh);
552                 else
553                         complete(&gh->gh_wait);
554                 spin_lock(&gl->gl_spin);
555         } else {
556                 gl->gl_req_gh = gh;
557                 set_bit(GLF_LOCK, &gl->gl_flags);
558                 spin_unlock(&gl->gl_spin);
559
560                 if (gh->gh_state == LM_ST_UNLOCKED ||
561                     gl->gl_state != LM_ST_EXCLUSIVE)
562                         glops->go_drop_th(gl);
563                 else
564                         glops->go_xmote_th(gl, gh->gh_state, gh->gh_flags);
565
566                 spin_lock(&gl->gl_spin);
567         }
568
569         return 0;
570 }
571
572 /**
573  * rq_greedy - process a queued request to drop greedy status
574  * @gh: the glock holder
575  *
576  * Returns: 1 if the queue is blocked
577  */
578
579 static int rq_greedy(struct gfs2_holder *gh)
580 {
581         struct gfs2_glock *gl = gh->gh_gl;
582
583         list_del_init(&gh->gh_list);
584         /*  gh->gh_error never examined.  */
585         clear_bit(GLF_GREEDY, &gl->gl_flags);
586         spin_unlock(&gl->gl_spin);
587
588         gfs2_holder_uninit(gh);
589         kfree(container_of(gh, struct greedy, gr_gh));
590
591         spin_lock(&gl->gl_spin);                
592
593         return 0;
594 }
595
596 /**
597  * run_queue - process holder structures on a glock
598  * @gl: the glock
599  *
600  */
601 static void run_queue(struct gfs2_glock *gl)
602 {
603         struct gfs2_holder *gh;
604         int blocked = 1;
605
606         for (;;) {
607                 if (test_bit(GLF_LOCK, &gl->gl_flags))
608                         break;
609
610                 if (!list_empty(&gl->gl_waiters1)) {
611                         gh = list_entry(gl->gl_waiters1.next,
612                                         struct gfs2_holder, gh_list);
613
614                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
615                                 blocked = rq_mutex(gh);
616                         else
617                                 gfs2_assert_warn(gl->gl_sbd, 0);
618
619                 } else if (!list_empty(&gl->gl_waiters2) &&
620                            !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
621                         gh = list_entry(gl->gl_waiters2.next,
622                                         struct gfs2_holder, gh_list);
623
624                         if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
625                                 blocked = rq_demote(gh);
626                         else if (test_bit(HIF_GREEDY, &gh->gh_iflags))
627                                 blocked = rq_greedy(gh);
628                         else
629                                 gfs2_assert_warn(gl->gl_sbd, 0);
630
631                 } else if (!list_empty(&gl->gl_waiters3)) {
632                         gh = list_entry(gl->gl_waiters3.next,
633                                         struct gfs2_holder, gh_list);
634
635                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
636                                 blocked = rq_promote(gh);
637                         else
638                                 gfs2_assert_warn(gl->gl_sbd, 0);
639
640                 } else
641                         break;
642
643                 if (blocked)
644                         break;
645         }
646 }
647
648 /**
649  * gfs2_glmutex_lock - acquire a local lock on a glock
650  * @gl: the glock
651  *
652  * Gives caller exclusive access to manipulate a glock structure.
653  */
654
655 void gfs2_glmutex_lock(struct gfs2_glock *gl)
656 {
657         struct gfs2_holder gh;
658
659         gfs2_holder_init(gl, 0, 0, &gh);
660         set_bit(HIF_MUTEX, &gh.gh_iflags);
661
662         spin_lock(&gl->gl_spin);
663         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
664                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
665         else
666                 complete(&gh.gh_wait);
667         spin_unlock(&gl->gl_spin);
668
669         wait_for_completion(&gh.gh_wait);
670         gfs2_holder_uninit(&gh);
671 }
672
673 /**
674  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
675  * @gl: the glock
676  *
677  * Returns: 1 if the glock is acquired
678  */
679
680 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
681 {
682         int acquired = 1;
683
684         spin_lock(&gl->gl_spin);
685         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
686                 acquired = 0;
687         spin_unlock(&gl->gl_spin);
688
689         return acquired;
690 }
691
692 /**
693  * gfs2_glmutex_unlock - release a local lock on a glock
694  * @gl: the glock
695  *
696  */
697
698 void gfs2_glmutex_unlock(struct gfs2_glock *gl)
699 {
700         spin_lock(&gl->gl_spin);
701         clear_bit(GLF_LOCK, &gl->gl_flags);
702         run_queue(gl);
703         BUG_ON(!spin_is_locked(&gl->gl_spin));
704         spin_unlock(&gl->gl_spin);
705 }
706
707 /**
708  * handle_callback - add a demote request to a lock's queue
709  * @gl: the glock
710  * @state: the state the caller wants us to change to
711  *
712  */
713
714 static void handle_callback(struct gfs2_glock *gl, unsigned int state)
715 {
716         struct gfs2_holder *gh, *new_gh = NULL;
717
718  restart:
719         spin_lock(&gl->gl_spin);
720
721         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
722                 if (test_bit(HIF_DEMOTE, &gh->gh_iflags) &&
723                     gl->gl_req_gh != gh) {
724                         if (gh->gh_state != state)
725                                 gh->gh_state = LM_ST_UNLOCKED;
726                         goto out;
727                 }
728         }
729
730         if (new_gh) {
731                 list_add_tail(&new_gh->gh_list, &gl->gl_waiters2);
732                 new_gh = NULL;
733         } else {
734                 spin_unlock(&gl->gl_spin);
735
736                 new_gh = gfs2_holder_get(gl, state, LM_FLAG_TRY,
737                                          GFP_KERNEL | __GFP_NOFAIL),
738                 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
739                 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
740
741                 goto restart;
742         }
743
744  out:
745         spin_unlock(&gl->gl_spin);
746
747         if (new_gh)
748                 gfs2_holder_put(new_gh);
749 }
750
751 /**
752  * state_change - record that the glock is now in a different state
753  * @gl: the glock
754  * @new_state the new state
755  *
756  */
757
758 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
759 {
760         int held1, held2;
761
762         held1 = (gl->gl_state != LM_ST_UNLOCKED);
763         held2 = (new_state != LM_ST_UNLOCKED);
764
765         if (held1 != held2) {
766                 if (held2)
767                         gfs2_glock_hold(gl);
768                 else
769                         gfs2_glock_put(gl);
770         }
771
772         gl->gl_state = new_state;
773 }
774
775 /**
776  * xmote_bh - Called after the lock module is done acquiring a lock
777  * @gl: The glock in question
778  * @ret: the int returned from the lock module
779  *
780  */
781
782 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
783 {
784         struct gfs2_sbd *sdp = gl->gl_sbd;
785         struct gfs2_glock_operations *glops = gl->gl_ops;
786         struct gfs2_holder *gh = gl->gl_req_gh;
787         int prev_state = gl->gl_state;
788         int op_done = 1;
789
790         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
791         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
792         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
793
794         state_change(gl, ret & LM_OUT_ST_MASK);
795
796         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
797                 if (glops->go_inval)
798                         glops->go_inval(gl, DIO_METADATA | DIO_DATA);
799         } else if (gl->gl_state == LM_ST_DEFERRED) {
800                 /* We might not want to do this here.
801                    Look at moving to the inode glops. */
802                 if (glops->go_inval)
803                         glops->go_inval(gl, DIO_DATA);
804         }
805
806         /*  Deal with each possible exit condition  */
807
808         if (!gh)
809                 gl->gl_stamp = jiffies;
810
811         else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
812                 spin_lock(&gl->gl_spin);
813                 list_del_init(&gh->gh_list);
814                 gh->gh_error = -EIO;
815                 spin_unlock(&gl->gl_spin);
816
817         } else if (test_bit(HIF_DEMOTE, &gh->gh_iflags)) {
818                 spin_lock(&gl->gl_spin);
819                 list_del_init(&gh->gh_list);
820                 if (gl->gl_state == gh->gh_state ||
821                     gl->gl_state == LM_ST_UNLOCKED)
822                         gh->gh_error = 0;
823                 else {
824                         if (gfs2_assert_warn(sdp, gh->gh_flags &
825                                         (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) == -1)
826                                 fs_warn(sdp, "ret = 0x%.8X\n", ret);
827                         gh->gh_error = GLR_TRYFAILED;
828                 }
829                 spin_unlock(&gl->gl_spin);
830
831                 if (ret & LM_OUT_CANCELED)
832                         handle_callback(gl, LM_ST_UNLOCKED); /* Lame */
833
834         } else if (ret & LM_OUT_CANCELED) {
835                 spin_lock(&gl->gl_spin);
836                 list_del_init(&gh->gh_list);
837                 gh->gh_error = GLR_CANCELED;
838                 spin_unlock(&gl->gl_spin);
839
840         } else if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
841                 spin_lock(&gl->gl_spin);
842                 list_move_tail(&gh->gh_list, &gl->gl_holders);
843                 gh->gh_error = 0;
844                 set_bit(HIF_HOLDER, &gh->gh_iflags);
845                 spin_unlock(&gl->gl_spin);
846
847                 set_bit(HIF_FIRST, &gh->gh_iflags);
848
849                 op_done = 0;
850
851         } else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
852                 spin_lock(&gl->gl_spin);
853                 list_del_init(&gh->gh_list);
854                 gh->gh_error = GLR_TRYFAILED;
855                 spin_unlock(&gl->gl_spin);
856
857         } else {
858                 if (gfs2_assert_withdraw(sdp, 0) == -1)
859                         fs_err(sdp, "ret = 0x%.8X\n", ret);
860         }
861
862         if (glops->go_xmote_bh)
863                 glops->go_xmote_bh(gl);
864
865         if (op_done) {
866                 spin_lock(&gl->gl_spin);
867                 gl->gl_req_gh = NULL;
868                 gl->gl_req_bh = NULL;
869                 clear_bit(GLF_LOCK, &gl->gl_flags);
870                 run_queue(gl);
871                 spin_unlock(&gl->gl_spin);
872         }
873
874         gfs2_glock_put(gl);
875
876         if (gh) {
877                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
878                         gfs2_holder_put(gh);
879                 else
880                         complete(&gh->gh_wait);
881         }
882 }
883
884 /**
885  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
886  * @gl: The glock in question
887  * @state: the requested state
888  * @flags: modifier flags to the lock call
889  *
890  */
891
892 void gfs2_glock_xmote_th(struct gfs2_glock *gl, unsigned int state, int flags)
893 {
894         struct gfs2_sbd *sdp = gl->gl_sbd;
895         struct gfs2_glock_operations *glops = gl->gl_ops;
896         int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
897                                  LM_FLAG_NOEXP | LM_FLAG_ANY |
898                                  LM_FLAG_PRIORITY);
899         unsigned int lck_ret;
900
901         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
902         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
903         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
904         gfs2_assert_warn(sdp, state != gl->gl_state);
905
906         if (gl->gl_state == LM_ST_EXCLUSIVE) {
907                 if (glops->go_sync)
908                         glops->go_sync(gl,
909                                        DIO_METADATA | DIO_DATA | DIO_RELEASE);
910         }
911
912         gfs2_glock_hold(gl);
913         gl->gl_req_bh = xmote_bh;
914
915         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state,
916                                lck_flags);
917
918         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
919                 return;
920
921         if (lck_ret & LM_OUT_ASYNC)
922                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
923         else
924                 xmote_bh(gl, lck_ret);
925 }
926
927 /**
928  * drop_bh - Called after a lock module unlock completes
929  * @gl: the glock
930  * @ret: the return status
931  *
932  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
933  * Doesn't drop the reference on the glock the top half took out
934  *
935  */
936
937 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
938 {
939         struct gfs2_sbd *sdp = gl->gl_sbd;
940         struct gfs2_glock_operations *glops = gl->gl_ops;
941         struct gfs2_holder *gh = gl->gl_req_gh;
942
943         clear_bit(GLF_PREFETCH, &gl->gl_flags);
944
945         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
946         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
947         gfs2_assert_warn(sdp, !ret);
948
949         state_change(gl, LM_ST_UNLOCKED);
950
951         if (glops->go_inval)
952                 glops->go_inval(gl, DIO_METADATA | DIO_DATA);
953
954         if (gh) {
955                 spin_lock(&gl->gl_spin);
956                 list_del_init(&gh->gh_list);
957                 gh->gh_error = 0;
958                 spin_unlock(&gl->gl_spin);
959         }
960
961         if (glops->go_drop_bh)
962                 glops->go_drop_bh(gl);
963
964         spin_lock(&gl->gl_spin);
965         gl->gl_req_gh = NULL;
966         gl->gl_req_bh = NULL;
967         clear_bit(GLF_LOCK, &gl->gl_flags);
968         run_queue(gl);
969         spin_unlock(&gl->gl_spin);
970
971         gfs2_glock_put(gl);
972
973         if (gh) {
974                 if (test_bit(HIF_DEALLOC, &gh->gh_iflags))
975                         gfs2_holder_put(gh);
976                 else
977                         complete(&gh->gh_wait);
978         }
979 }
980
981 /**
982  * gfs2_glock_drop_th - call into the lock module to unlock a lock
983  * @gl: the glock
984  *
985  */
986
987 void gfs2_glock_drop_th(struct gfs2_glock *gl)
988 {
989         struct gfs2_sbd *sdp = gl->gl_sbd;
990         struct gfs2_glock_operations *glops = gl->gl_ops;
991         unsigned int ret;
992
993         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
994         gfs2_assert_warn(sdp, queue_empty(gl, &gl->gl_holders));
995         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
996
997         if (gl->gl_state == LM_ST_EXCLUSIVE) {
998                 if (glops->go_sync)
999                         glops->go_sync(gl,
1000                                        DIO_METADATA | DIO_DATA | DIO_RELEASE);
1001         }
1002
1003         gfs2_glock_hold(gl);
1004         gl->gl_req_bh = drop_bh;
1005
1006         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
1007
1008         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
1009                 return;
1010
1011         if (!ret)
1012                 drop_bh(gl, ret);
1013         else
1014                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1015 }
1016
1017 /**
1018  * do_cancels - cancel requests for locks stuck waiting on an expire flag
1019  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1020  *
1021  * Don't cancel GL_NOCANCEL requests.
1022  */
1023
1024 static void do_cancels(struct gfs2_holder *gh)
1025 {
1026         struct gfs2_glock *gl = gh->gh_gl;
1027
1028         spin_lock(&gl->gl_spin);
1029
1030         while (gl->gl_req_gh != gh &&
1031                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1032                !list_empty(&gh->gh_list)) {
1033                 if (gl->gl_req_bh &&
1034                     !(gl->gl_req_gh &&
1035                       (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1036                         spin_unlock(&gl->gl_spin);
1037                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1038                         msleep(100);
1039                         spin_lock(&gl->gl_spin);
1040                 } else {
1041                         spin_unlock(&gl->gl_spin);
1042                         msleep(100);
1043                         spin_lock(&gl->gl_spin);
1044                 }
1045         }
1046
1047         spin_unlock(&gl->gl_spin);
1048 }
1049
1050 /**
1051  * glock_wait_internal - wait on a glock acquisition
1052  * @gh: the glock holder
1053  *
1054  * Returns: 0 on success
1055  */
1056
1057 static int glock_wait_internal(struct gfs2_holder *gh)
1058 {
1059         struct gfs2_glock *gl = gh->gh_gl;
1060         struct gfs2_sbd *sdp = gl->gl_sbd;
1061         struct gfs2_glock_operations *glops = gl->gl_ops;
1062
1063         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1064                 return -EIO;
1065
1066         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1067                 spin_lock(&gl->gl_spin);
1068                 if (gl->gl_req_gh != gh &&
1069                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1070                     !list_empty(&gh->gh_list)) {
1071                         list_del_init(&gh->gh_list);
1072                         gh->gh_error = GLR_TRYFAILED;
1073                         run_queue(gl);
1074                         spin_unlock(&gl->gl_spin);
1075                         return gh->gh_error;
1076                 }
1077                 spin_unlock(&gl->gl_spin);
1078         }
1079
1080         if (gh->gh_flags & LM_FLAG_PRIORITY)
1081                 do_cancels(gh);
1082
1083         wait_for_completion(&gh->gh_wait);
1084
1085         if (gh->gh_error)
1086                 return gh->gh_error;
1087
1088         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1089         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state,
1090                                                    gh->gh_state,
1091                                                    gh->gh_flags));
1092
1093         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1094                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1095
1096                 if (glops->go_lock) {
1097                         gh->gh_error = glops->go_lock(gh);
1098                         if (gh->gh_error) {
1099                                 spin_lock(&gl->gl_spin);
1100                                 list_del_init(&gh->gh_list);
1101                                 spin_unlock(&gl->gl_spin);
1102                         }
1103                 }
1104
1105                 spin_lock(&gl->gl_spin);
1106                 gl->gl_req_gh = NULL;
1107                 gl->gl_req_bh = NULL;
1108                 clear_bit(GLF_LOCK, &gl->gl_flags);
1109                 run_queue(gl);
1110                 spin_unlock(&gl->gl_spin);
1111         }
1112
1113         return gh->gh_error;
1114 }
1115
1116 static inline struct gfs2_holder *
1117 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1118 {
1119         struct gfs2_holder *gh;
1120
1121         list_for_each_entry(gh, head, gh_list) {
1122                 if (gh->gh_owner == owner)
1123                         return gh;
1124         }
1125
1126         return NULL;
1127 }
1128
1129 /**
1130  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1131  * @gh: the holder structure to add
1132  *
1133  */
1134
1135 static void add_to_queue(struct gfs2_holder *gh)
1136 {
1137         struct gfs2_glock *gl = gh->gh_gl;
1138         struct gfs2_holder *existing;
1139
1140         BUG_ON(!gh->gh_owner);
1141
1142         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1143         if (existing) {
1144                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1145                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1146                 BUG();
1147         }
1148
1149         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1150         if (existing) {
1151                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1152                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1153                 BUG();
1154         }
1155
1156         if (gh->gh_flags & LM_FLAG_PRIORITY)
1157                 list_add(&gh->gh_list, &gl->gl_waiters3);
1158         else
1159                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);  
1160 }
1161
1162 /**
1163  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1164  * @gh: the holder structure
1165  *
1166  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1167  *
1168  * Returns: 0, GLR_TRYFAILED, or errno on failure
1169  */
1170
1171 int gfs2_glock_nq(struct gfs2_holder *gh)
1172 {
1173         struct gfs2_glock *gl = gh->gh_gl;
1174         struct gfs2_sbd *sdp = gl->gl_sbd;
1175         int error = 0;
1176
1177  restart:
1178         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1179                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1180                 return -EIO;
1181         }
1182
1183         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1184
1185         spin_lock(&gl->gl_spin);
1186         add_to_queue(gh);
1187         run_queue(gl);
1188         spin_unlock(&gl->gl_spin);
1189
1190         if (!(gh->gh_flags & GL_ASYNC)) {
1191                 error = glock_wait_internal(gh);
1192                 if (error == GLR_CANCELED) {
1193                         msleep(100);
1194                         goto restart;
1195                 }
1196         }
1197
1198         clear_bit(GLF_PREFETCH, &gl->gl_flags);
1199
1200         return error;
1201 }
1202
1203 /**
1204  * gfs2_glock_poll - poll to see if an async request has been completed
1205  * @gh: the holder
1206  *
1207  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1208  */
1209
1210 int gfs2_glock_poll(struct gfs2_holder *gh)
1211 {
1212         struct gfs2_glock *gl = gh->gh_gl;
1213         int ready = 0;
1214
1215         spin_lock(&gl->gl_spin);
1216
1217         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1218                 ready = 1;
1219         else if (list_empty(&gh->gh_list)) {
1220                 if (gh->gh_error == GLR_CANCELED) {
1221                         spin_unlock(&gl->gl_spin);
1222                         msleep(100);
1223                         if (gfs2_glock_nq(gh))
1224                                 return 1;
1225                         return 0;
1226                 } else
1227                         ready = 1;
1228         }
1229
1230         spin_unlock(&gl->gl_spin);
1231
1232         return ready;
1233 }
1234
1235 /**
1236  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1237  * @gh: the holder structure
1238  *
1239  * Returns: 0, GLR_TRYFAILED, or errno on failure
1240  */
1241
1242 int gfs2_glock_wait(struct gfs2_holder *gh)
1243 {
1244         int error;
1245
1246         error = glock_wait_internal(gh);
1247         if (error == GLR_CANCELED) {
1248                 msleep(100);
1249                 gh->gh_flags &= ~GL_ASYNC;
1250                 error = gfs2_glock_nq(gh);
1251         }
1252
1253         return error;
1254 }
1255
1256 /**
1257  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1258  * @gh: the glock holder
1259  *
1260  */
1261
1262 void gfs2_glock_dq(struct gfs2_holder *gh)
1263 {
1264         struct gfs2_glock *gl = gh->gh_gl;
1265         struct gfs2_glock_operations *glops = gl->gl_ops;
1266
1267         if (gh->gh_flags & GL_SYNC)
1268                 set_bit(GLF_SYNC, &gl->gl_flags);
1269
1270         if (gh->gh_flags & GL_NOCACHE)
1271                 handle_callback(gl, LM_ST_UNLOCKED);
1272
1273         gfs2_glmutex_lock(gl);
1274
1275         spin_lock(&gl->gl_spin);
1276         list_del_init(&gh->gh_list);
1277
1278         if (list_empty(&gl->gl_holders)) {
1279                 spin_unlock(&gl->gl_spin);
1280
1281                 if (glops->go_unlock)
1282                         glops->go_unlock(gh);
1283
1284                 if (test_bit(GLF_SYNC, &gl->gl_flags)) {
1285                         if (glops->go_sync)
1286                                 glops->go_sync(gl, DIO_METADATA | DIO_DATA);
1287                 }
1288
1289                 gl->gl_stamp = jiffies;
1290
1291                 spin_lock(&gl->gl_spin);
1292         }
1293
1294         clear_bit(GLF_LOCK, &gl->gl_flags);
1295         run_queue(gl);
1296         spin_unlock(&gl->gl_spin);
1297 }
1298
1299 /**
1300  * gfs2_glock_prefetch - Try to prefetch a glock
1301  * @gl: the glock
1302  * @state: the state to prefetch in
1303  * @flags: flags passed to go_xmote_th()
1304  *
1305  */
1306
1307 static void gfs2_glock_prefetch(struct gfs2_glock *gl, unsigned int state,
1308                                 int flags)
1309 {
1310         struct gfs2_glock_operations *glops = gl->gl_ops;
1311
1312         spin_lock(&gl->gl_spin);
1313
1314         if (test_bit(GLF_LOCK, &gl->gl_flags) ||
1315             !list_empty(&gl->gl_holders) ||
1316             !list_empty(&gl->gl_waiters1) ||
1317             !list_empty(&gl->gl_waiters2) ||
1318             !list_empty(&gl->gl_waiters3) ||
1319             relaxed_state_ok(gl->gl_state, state, flags)) {
1320                 spin_unlock(&gl->gl_spin);
1321                 return;
1322         }
1323
1324         set_bit(GLF_PREFETCH, &gl->gl_flags);
1325         set_bit(GLF_LOCK, &gl->gl_flags);
1326         spin_unlock(&gl->gl_spin);
1327
1328         glops->go_xmote_th(gl, state, flags);
1329 }
1330
1331 /**
1332  * gfs2_glock_force_drop - Force a glock to be uncached
1333  * @gl: the glock
1334  *
1335  */
1336 #if 0
1337 void gfs2_glock_force_drop(struct gfs2_glock *gl)
1338 {
1339         struct gfs2_holder gh;
1340
1341         gfs2_holder_init(gl, LM_ST_UNLOCKED, 0, &gh);
1342         set_bit(HIF_DEMOTE, &gh.gh_iflags);
1343
1344         spin_lock(&gl->gl_spin);
1345         list_add_tail(&gh.gh_list, &gl->gl_waiters2);
1346         run_queue(gl);
1347         spin_unlock(&gl->gl_spin);
1348
1349         wait_for_completion(&gh.gh_wait);
1350         gfs2_holder_uninit(&gh);
1351 }
1352 #endif  /*  0  */
1353
1354 static void greedy_work(void *data)
1355 {
1356         struct greedy *gr = data;
1357         struct gfs2_holder *gh = &gr->gr_gh;
1358         struct gfs2_glock *gl = gh->gh_gl;
1359         struct gfs2_glock_operations *glops = gl->gl_ops;
1360
1361         clear_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1362
1363         if (glops->go_greedy)
1364                 glops->go_greedy(gl);
1365
1366         spin_lock(&gl->gl_spin);
1367
1368         if (list_empty(&gl->gl_waiters2)) {
1369                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1370                 spin_unlock(&gl->gl_spin);
1371                 gfs2_holder_uninit(gh);
1372                 kfree(gr);
1373         } else {
1374                 gfs2_glock_hold(gl);
1375                 list_add_tail(&gh->gh_list, &gl->gl_waiters2);
1376                 run_queue(gl);
1377                 spin_unlock(&gl->gl_spin);
1378                 gfs2_glock_put(gl);
1379         }
1380 }
1381
1382 /**
1383  * gfs2_glock_be_greedy -
1384  * @gl:
1385  * @time:
1386  *
1387  * Returns: 0 if go_greedy will be called, 1 otherwise
1388  */
1389
1390 int gfs2_glock_be_greedy(struct gfs2_glock *gl, unsigned int time)
1391 {
1392         struct greedy *gr;
1393         struct gfs2_holder *gh;
1394
1395         if (!time ||
1396             gl->gl_sbd->sd_args.ar_localcaching ||
1397             test_and_set_bit(GLF_GREEDY, &gl->gl_flags))
1398                 return 1;
1399
1400         gr = kmalloc(sizeof(struct greedy), GFP_KERNEL);
1401         if (!gr) {
1402                 clear_bit(GLF_GREEDY, &gl->gl_flags);
1403                 return 1;
1404         }
1405         gh = &gr->gr_gh;
1406
1407         gfs2_holder_init(gl, 0, 0, gh);
1408         set_bit(HIF_GREEDY, &gh->gh_iflags);
1409         INIT_WORK(&gr->gr_work, greedy_work, gr);
1410
1411         set_bit(GLF_SKIP_WAITERS2, &gl->gl_flags);
1412         schedule_delayed_work(&gr->gr_work, time);
1413
1414         return 0;
1415 }
1416
1417 /**
1418  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1419  * @gh: the holder structure
1420  *
1421  */
1422
1423 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1424 {
1425         gfs2_glock_dq(gh);
1426         gfs2_holder_uninit(gh);
1427 }
1428
1429 /**
1430  * gfs2_glock_nq_num - acquire a glock based on lock number
1431  * @sdp: the filesystem
1432  * @number: the lock number
1433  * @glops: the glock operations for the type of glock
1434  * @state: the state to acquire the glock in
1435  * @flags: modifier flags for the aquisition
1436  * @gh: the struct gfs2_holder
1437  *
1438  * Returns: errno
1439  */
1440
1441 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, uint64_t number,
1442                       struct gfs2_glock_operations *glops, unsigned int state,
1443                       int flags, struct gfs2_holder *gh)
1444 {
1445         struct gfs2_glock *gl;
1446         int error;
1447
1448         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1449         if (!error) {
1450                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1451                 gfs2_glock_put(gl);
1452         }
1453
1454         return error;
1455 }
1456
1457 /**
1458  * glock_compare - Compare two struct gfs2_glock structures for sorting
1459  * @arg_a: the first structure
1460  * @arg_b: the second structure
1461  *
1462  */
1463
1464 static int glock_compare(const void *arg_a, const void *arg_b)
1465 {
1466         struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1467         struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1468         struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1469         struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1470         int ret = 0;
1471
1472         if (a->ln_number > b->ln_number)
1473                 ret = 1;
1474         else if (a->ln_number < b->ln_number)
1475                 ret = -1;
1476         else {
1477                 if (gh_a->gh_state == LM_ST_SHARED &&
1478                     gh_b->gh_state == LM_ST_EXCLUSIVE)
1479                         ret = 1;
1480                 else if (!(gh_a->gh_flags & GL_LOCAL_EXCL) &&
1481                          (gh_b->gh_flags & GL_LOCAL_EXCL))
1482                         ret = 1;
1483         }
1484
1485         return ret;
1486 }
1487
1488 /**
1489  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1490  * @num_gh: the number of structures
1491  * @ghs: an array of struct gfs2_holder structures
1492  *
1493  * Returns: 0 on success (all glocks acquired),
1494  *          errno on failure (no glocks acquired)
1495  */
1496
1497 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1498                      struct gfs2_holder **p)
1499 {
1500         unsigned int x;
1501         int error = 0;
1502
1503         for (x = 0; x < num_gh; x++)
1504                 p[x] = &ghs[x];
1505
1506         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1507
1508         for (x = 0; x < num_gh; x++) {
1509                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1510
1511                 error = gfs2_glock_nq(p[x]);
1512                 if (error) {
1513                         while (x--)
1514                                 gfs2_glock_dq(p[x]);
1515                         break;
1516                 }
1517         }
1518
1519         return error;
1520 }
1521
1522 /**
1523  * gfs2_glock_nq_m - acquire multiple glocks
1524  * @num_gh: the number of structures
1525  * @ghs: an array of struct gfs2_holder structures
1526  *
1527  * Figure out how big an impact this function has.  Either:
1528  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1529  * 2) Forget async stuff and just call nq_m_sync()
1530  * 3) Leave it like it is
1531  *
1532  * Returns: 0 on success (all glocks acquired),
1533  *          errno on failure (no glocks acquired)
1534  */
1535
1536 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1537 {
1538         int *e;
1539         unsigned int x;
1540         int borked = 0, serious = 0;
1541         int error = 0;
1542
1543         if (!num_gh)
1544                 return 0;
1545
1546         if (num_gh == 1) {
1547                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1548                 return gfs2_glock_nq(ghs);
1549         }
1550
1551         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1552         if (!e)
1553                 return -ENOMEM;
1554
1555         for (x = 0; x < num_gh; x++) {
1556                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1557                 error = gfs2_glock_nq(&ghs[x]);
1558                 if (error) {
1559                         borked = 1;
1560                         serious = error;
1561                         num_gh = x;
1562                         break;
1563                 }
1564         }
1565
1566         for (x = 0; x < num_gh; x++) {
1567                 error = e[x] = glock_wait_internal(&ghs[x]);
1568                 if (error) {
1569                         borked = 1;
1570                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1571                                 serious = error;
1572                 }
1573         }
1574
1575         if (!borked) {
1576                 kfree(e);
1577                 return 0;
1578         }
1579
1580         for (x = 0; x < num_gh; x++)
1581                 if (!e[x])
1582                         gfs2_glock_dq(&ghs[x]);
1583
1584         if (serious)
1585                 error = serious;
1586         else {
1587                 for (x = 0; x < num_gh; x++)
1588                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1589                                           &ghs[x]);
1590                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1591         }
1592
1593         kfree(e);
1594
1595         return error;
1596 }
1597
1598 /**
1599  * gfs2_glock_dq_m - release multiple glocks
1600  * @num_gh: the number of structures
1601  * @ghs: an array of struct gfs2_holder structures
1602  *
1603  */
1604
1605 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1606 {
1607         unsigned int x;
1608
1609         for (x = 0; x < num_gh; x++)
1610                 gfs2_glock_dq(&ghs[x]);
1611 }
1612
1613 /**
1614  * gfs2_glock_dq_uninit_m - release multiple glocks
1615  * @num_gh: the number of structures
1616  * @ghs: an array of struct gfs2_holder structures
1617  *
1618  */
1619
1620 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1621 {
1622         unsigned int x;
1623
1624         for (x = 0; x < num_gh; x++)
1625                 gfs2_glock_dq_uninit(&ghs[x]);
1626 }
1627
1628 /**
1629  * gfs2_glock_prefetch_num - prefetch a glock based on lock number
1630  * @sdp: the filesystem
1631  * @number: the lock number
1632  * @glops: the glock operations for the type of glock
1633  * @state: the state to acquire the glock in
1634  * @flags: modifier flags for the aquisition
1635  *
1636  * Returns: errno
1637  */
1638
1639 void gfs2_glock_prefetch_num(struct gfs2_sbd *sdp, uint64_t number,
1640                              struct gfs2_glock_operations *glops,
1641                              unsigned int state, int flags)
1642 {
1643         struct gfs2_glock *gl;
1644         int error;
1645
1646         if (atomic_read(&sdp->sd_reclaim_count) <
1647             gfs2_tune_get(sdp, gt_reclaim_limit)) {
1648                 error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1649                 if (!error) {
1650                         gfs2_glock_prefetch(gl, state, flags);
1651                         gfs2_glock_put(gl);
1652                 }
1653         }
1654 }
1655
1656 /**
1657  * gfs2_lvb_hold - attach a LVB from a glock
1658  * @gl: The glock in question
1659  *
1660  */
1661
1662 int gfs2_lvb_hold(struct gfs2_glock *gl)
1663 {
1664         int error;
1665
1666         gfs2_glmutex_lock(gl);
1667
1668         if (!atomic_read(&gl->gl_lvb_count)) {
1669                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1670                 if (error) {
1671                         gfs2_glmutex_unlock(gl);
1672                         return error;
1673                 }
1674                 gfs2_glock_hold(gl);
1675         }
1676         atomic_inc(&gl->gl_lvb_count);
1677
1678         gfs2_glmutex_unlock(gl);
1679
1680         return 0;
1681 }
1682
1683 /**
1684  * gfs2_lvb_unhold - detach a LVB from a glock
1685  * @gl: The glock in question
1686  *
1687  */
1688
1689 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1690 {
1691         gfs2_glock_hold(gl);
1692         gfs2_glmutex_lock(gl);
1693
1694         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1695         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1696                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1697                 gl->gl_lvb = NULL;
1698                 gfs2_glock_put(gl);
1699         }
1700
1701         gfs2_glmutex_unlock(gl);
1702         gfs2_glock_put(gl);
1703 }
1704
1705 #if 0
1706 void gfs2_lvb_sync(struct gfs2_glock *gl)
1707 {
1708         gfs2_glmutex_lock(gl);
1709
1710         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count));
1711         if (!gfs2_assert_warn(gl->gl_sbd, gfs2_glock_is_held_excl(gl)))
1712                 gfs2_lm_sync_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1713
1714         gfs2_glmutex_unlock(gl);
1715 }
1716 #endif  /*  0  */
1717
1718 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1719                         unsigned int state)
1720 {
1721         struct gfs2_glock *gl;
1722
1723         gl = gfs2_glock_find(sdp, name);
1724         if (!gl)
1725                 return;
1726
1727         if (gl->gl_ops->go_callback)
1728                 gl->gl_ops->go_callback(gl, state);
1729         handle_callback(gl, state);
1730
1731         spin_lock(&gl->gl_spin);
1732         run_queue(gl);
1733         spin_unlock(&gl->gl_spin);
1734
1735         gfs2_glock_put(gl);
1736 }
1737
1738 /**
1739  * gfs2_glock_cb - Callback used by locking module
1740  * @fsdata: Pointer to the superblock
1741  * @type: Type of callback
1742  * @data: Type dependent data pointer
1743  *
1744  * Called by the locking module when it wants to tell us something.
1745  * Either we need to drop a lock, one of our ASYNC requests completed, or
1746  * a journal from another client needs to be recovered.
1747  */
1748
1749 void gfs2_glock_cb(lm_fsdata_t *fsdata, unsigned int type, void *data)
1750 {
1751         struct gfs2_sbd *sdp = (struct gfs2_sbd *)fsdata;
1752
1753         switch (type) {
1754         case LM_CB_NEED_E:
1755                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1756                 return;
1757
1758         case LM_CB_NEED_D:
1759                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1760                 return;
1761
1762         case LM_CB_NEED_S:
1763                 blocking_cb(sdp, data, LM_ST_SHARED);
1764                 return;
1765
1766         case LM_CB_ASYNC: {
1767                 struct lm_async_cb *async = data;
1768                 struct gfs2_glock *gl;
1769
1770                 gl = gfs2_glock_find(sdp, &async->lc_name);
1771                 if (gfs2_assert_warn(sdp, gl))
1772                         return;
1773                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1774                         gl->gl_req_bh(gl, async->lc_ret);
1775                 gfs2_glock_put(gl);
1776                 return;
1777         }
1778
1779         case LM_CB_NEED_RECOVERY:
1780                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1781                 if (sdp->sd_recoverd_process)
1782                         wake_up_process(sdp->sd_recoverd_process);
1783                 return;
1784
1785         case LM_CB_DROPLOCKS:
1786                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1787                 gfs2_quota_scan(sdp);
1788                 return;
1789
1790         default:
1791                 gfs2_assert_warn(sdp, 0);
1792                 return;
1793         }
1794 }
1795
1796 /**
1797  * gfs2_try_toss_inode - try to remove a particular inode struct from cache
1798  * sdp: the filesystem
1799  * inum: the inode number
1800  *
1801  */
1802
1803 void gfs2_try_toss_inode(struct gfs2_sbd *sdp, struct gfs2_inum *inum)
1804 {
1805         struct gfs2_glock *gl;
1806         struct gfs2_inode *ip;
1807         int error;
1808
1809         error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops,
1810                                NO_CREATE, &gl);
1811         if (error || !gl)
1812                 return;
1813
1814         if (!gfs2_glmutex_trylock(gl))
1815                 goto out;
1816
1817         ip = gl->gl_object;
1818         if (!ip)
1819                 goto out_unlock;
1820
1821         if (atomic_read(&ip->i_count))
1822                 goto out_unlock;
1823
1824         gfs2_inode_destroy(ip, 1);
1825
1826  out_unlock:
1827         gfs2_glmutex_unlock(gl);
1828
1829  out:
1830         gfs2_glock_put(gl);
1831 }
1832
1833 /**
1834  * gfs2_iopen_go_callback - Try to kick the inode/vnode associated with an
1835  *                          iopen glock from memory
1836  * @io_gl: the iopen glock
1837  * @state: the state into which the glock should be put
1838  *
1839  */
1840
1841 void gfs2_iopen_go_callback(struct gfs2_glock *io_gl, unsigned int state)
1842 {
1843         struct gfs2_glock *i_gl;
1844
1845         if (state != LM_ST_UNLOCKED)
1846                 return;
1847
1848         spin_lock(&io_gl->gl_spin);
1849         i_gl = io_gl->gl_object;
1850         if (i_gl) {
1851                 gfs2_glock_hold(i_gl);
1852                 spin_unlock(&io_gl->gl_spin);
1853         } else {
1854                 spin_unlock(&io_gl->gl_spin);
1855                 return;
1856         }
1857
1858         if (gfs2_glmutex_trylock(i_gl)) {
1859                 struct gfs2_inode *ip = i_gl->gl_object;
1860                 if (ip) {
1861                         gfs2_try_toss_vnode(ip);
1862                         gfs2_glmutex_unlock(i_gl);
1863                         gfs2_glock_schedule_for_reclaim(i_gl);
1864                         goto out;
1865                 }
1866                 gfs2_glmutex_unlock(i_gl);
1867         }
1868
1869  out:
1870         gfs2_glock_put(i_gl);
1871 }
1872
1873 /**
1874  * demote_ok - Check to see if it's ok to unlock a glock
1875  * @gl: the glock
1876  *
1877  * Returns: 1 if it's ok
1878  */
1879
1880 static int demote_ok(struct gfs2_glock *gl)
1881 {
1882         struct gfs2_sbd *sdp = gl->gl_sbd;
1883         struct gfs2_glock_operations *glops = gl->gl_ops;
1884         int demote = 1;
1885
1886         if (test_bit(GLF_STICKY, &gl->gl_flags))
1887                 demote = 0;
1888         else if (test_bit(GLF_PREFETCH, &gl->gl_flags))
1889                 demote = time_after_eq(jiffies,
1890                                     gl->gl_stamp +
1891                                     gfs2_tune_get(sdp, gt_prefetch_secs) * HZ);
1892         else if (glops->go_demote_ok)
1893                 demote = glops->go_demote_ok(gl);
1894
1895         return demote;
1896 }
1897
1898 /**
1899  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1900  * @gl: the glock
1901  *
1902  */
1903
1904 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1905 {
1906         struct gfs2_sbd *sdp = gl->gl_sbd;
1907
1908         spin_lock(&sdp->sd_reclaim_lock);
1909         if (list_empty(&gl->gl_reclaim)) {
1910                 gfs2_glock_hold(gl);
1911                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1912                 atomic_inc(&sdp->sd_reclaim_count);
1913         }
1914         spin_unlock(&sdp->sd_reclaim_lock);
1915
1916         wake_up(&sdp->sd_reclaim_wq);
1917 }
1918
1919 /**
1920  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1921  * @sdp: the filesystem
1922  *
1923  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1924  * different glock and we notice that there are a lot of glocks in the
1925  * reclaim list.
1926  *
1927  */
1928
1929 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1930 {
1931         struct gfs2_glock *gl;
1932
1933         spin_lock(&sdp->sd_reclaim_lock);
1934         if (list_empty(&sdp->sd_reclaim_list)) {
1935                 spin_unlock(&sdp->sd_reclaim_lock);
1936                 return;
1937         }
1938         gl = list_entry(sdp->sd_reclaim_list.next,
1939                         struct gfs2_glock, gl_reclaim);
1940         list_del_init(&gl->gl_reclaim);
1941         spin_unlock(&sdp->sd_reclaim_lock);
1942
1943         atomic_dec(&sdp->sd_reclaim_count);
1944         atomic_inc(&sdp->sd_reclaimed);
1945
1946         if (gfs2_glmutex_trylock(gl)) {
1947                 if (gl->gl_ops == &gfs2_inode_glops) {
1948                         struct gfs2_inode *ip = gl->gl_object;
1949                         if (ip && !atomic_read(&ip->i_count))
1950                                 gfs2_inode_destroy(ip, 1);
1951                 }
1952                 if (queue_empty(gl, &gl->gl_holders) &&
1953                     gl->gl_state != LM_ST_UNLOCKED &&
1954                     demote_ok(gl))
1955                         handle_callback(gl, LM_ST_UNLOCKED);
1956                 gfs2_glmutex_unlock(gl);
1957         }
1958
1959         gfs2_glock_put(gl);
1960 }
1961
1962 /**
1963  * examine_bucket - Call a function for glock in a hash bucket
1964  * @examiner: the function
1965  * @sdp: the filesystem
1966  * @bucket: the bucket
1967  *
1968  * Returns: 1 if the bucket has entries
1969  */
1970
1971 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1972                           struct gfs2_gl_hash_bucket *bucket)
1973 {
1974         struct glock_plug plug;
1975         struct list_head *tmp;
1976         struct gfs2_glock *gl;
1977         int entries;
1978
1979         /* Add "plug" to end of bucket list, work back up list from there */
1980         memset(&plug.gl_flags, 0, sizeof(unsigned long));
1981         set_bit(GLF_PLUG, &plug.gl_flags);
1982
1983         write_lock(&bucket->hb_lock);
1984         list_add(&plug.gl_list, &bucket->hb_list);
1985         write_unlock(&bucket->hb_lock);
1986
1987         for (;;) {
1988                 write_lock(&bucket->hb_lock);
1989
1990                 for (;;) {
1991                         tmp = plug.gl_list.next;
1992
1993                         if (tmp == &bucket->hb_list) {
1994                                 list_del(&plug.gl_list);
1995                                 entries = !list_empty(&bucket->hb_list);
1996                                 write_unlock(&bucket->hb_lock);
1997                                 return entries;
1998                         }
1999                         gl = list_entry(tmp, struct gfs2_glock, gl_list);
2000
2001                         /* Move plug up list */
2002                         list_move(&plug.gl_list, &gl->gl_list);
2003
2004                         if (test_bit(GLF_PLUG, &gl->gl_flags))
2005                                 continue;
2006
2007                         /* examiner() must glock_put() */
2008                         gfs2_glock_hold(gl);
2009
2010                         break;
2011                 }
2012
2013                 write_unlock(&bucket->hb_lock);
2014
2015                 examiner(gl);
2016         }
2017 }
2018
2019 /**
2020  * scan_glock - look at a glock and see if we can reclaim it
2021  * @gl: the glock to look at
2022  *
2023  */
2024
2025 static void scan_glock(struct gfs2_glock *gl)
2026 {
2027         if (gfs2_glmutex_trylock(gl)) {
2028                 if (gl->gl_ops == &gfs2_inode_glops) {
2029                         struct gfs2_inode *ip = gl->gl_object;
2030                         if (ip && !atomic_read(&ip->i_count))
2031                                 goto out_schedule;
2032                 }
2033                 if (queue_empty(gl, &gl->gl_holders) &&
2034                     gl->gl_state != LM_ST_UNLOCKED &&
2035                     demote_ok(gl))
2036                         goto out_schedule;
2037
2038                 gfs2_glmutex_unlock(gl);
2039         }
2040
2041         gfs2_glock_put(gl);
2042
2043         return;
2044
2045  out_schedule:
2046         gfs2_glmutex_unlock(gl);
2047         gfs2_glock_schedule_for_reclaim(gl);
2048         gfs2_glock_put(gl);
2049 }
2050
2051 /**
2052  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
2053  * @sdp: the filesystem
2054  *
2055  */
2056
2057 void gfs2_scand_internal(struct gfs2_sbd *sdp)
2058 {
2059         unsigned int x;
2060
2061         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2062                 examine_bucket(scan_glock, sdp, &sdp->sd_gl_hash[x]);
2063                 cond_resched();
2064         }
2065 }
2066
2067 /**
2068  * clear_glock - look at a glock and see if we can free it from glock cache
2069  * @gl: the glock to look at
2070  *
2071  */
2072
2073 static void clear_glock(struct gfs2_glock *gl)
2074 {
2075         struct gfs2_sbd *sdp = gl->gl_sbd;
2076         int released;
2077
2078         spin_lock(&sdp->sd_reclaim_lock);
2079         if (!list_empty(&gl->gl_reclaim)) {
2080                 list_del_init(&gl->gl_reclaim);
2081                 atomic_dec(&sdp->sd_reclaim_count);
2082                 spin_unlock(&sdp->sd_reclaim_lock);
2083                 released = gfs2_glock_put(gl);
2084                 gfs2_assert(sdp, !released);
2085         } else {
2086                 spin_unlock(&sdp->sd_reclaim_lock);
2087         }
2088
2089         if (gfs2_glmutex_trylock(gl)) {
2090                 if (gl->gl_ops == &gfs2_inode_glops) {
2091                         struct gfs2_inode *ip = gl->gl_object;
2092                         if (ip && !atomic_read(&ip->i_count))
2093                                 gfs2_inode_destroy(ip, 1);
2094                 }
2095                 if (queue_empty(gl, &gl->gl_holders) &&
2096                     gl->gl_state != LM_ST_UNLOCKED)
2097                         handle_callback(gl, LM_ST_UNLOCKED);
2098
2099                 gfs2_glmutex_unlock(gl);
2100         }
2101
2102         gfs2_glock_put(gl);
2103 }
2104
2105 /**
2106  * gfs2_gl_hash_clear - Empty out the glock hash table
2107  * @sdp: the filesystem
2108  * @wait: wait until it's all gone
2109  *
2110  * Called when unmounting the filesystem, or when inter-node lock manager
2111  * requests DROPLOCKS because it is running out of capacity.
2112  */
2113
2114 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
2115 {
2116         unsigned long t;
2117         unsigned int x;
2118         int cont;
2119
2120         t = jiffies;
2121
2122         for (;;) {
2123                 cont = 0;
2124
2125                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
2126                         if (examine_bucket(clear_glock, sdp,
2127                                            &sdp->sd_gl_hash[x]))
2128                                 cont = 1;
2129
2130                 if (!wait || !cont)
2131                         break;
2132
2133                 if (time_after_eq(jiffies,
2134                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
2135                         fs_warn(sdp, "Unmount seems to be stalled. "
2136                                      "Dumping lock state...\n");
2137                         gfs2_dump_lockstate(sdp);
2138                         t = jiffies;
2139                 }
2140
2141                 /* invalidate_inodes() requires that the sb inodes list
2142                    not change, but an async completion callback for an
2143                    unlock can occur which does glock_put() which
2144                    can call iput() which will change the sb inodes list.
2145                    invalidate_inodes_mutex prevents glock_put()'s during
2146                    an invalidate_inodes() */
2147
2148                 mutex_lock(&sdp->sd_invalidate_inodes_mutex);
2149                 invalidate_inodes(sdp->sd_vfs);
2150                 mutex_unlock(&sdp->sd_invalidate_inodes_mutex);
2151                 yield();
2152         }
2153 }
2154
2155 /*
2156  *  Diagnostic routines to help debug distributed deadlock
2157  */
2158
2159 /**
2160  * dump_holder - print information about a glock holder
2161  * @str: a string naming the type of holder
2162  * @gh: the glock holder
2163  *
2164  * Returns: 0 on success, -ENOBUFS when we run out of space
2165  */
2166
2167 static int dump_holder(char *str, struct gfs2_holder *gh)
2168 {
2169         unsigned int x;
2170         int error = -ENOBUFS;
2171
2172         printk(KERN_INFO "  %s\n", str);
2173         printk(KERN_INFO "    owner = %ld\n",
2174                    (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
2175         printk(KERN_INFO "    gh_state = %u\n", gh->gh_state);
2176         printk(KERN_INFO "    gh_flags =");
2177         for (x = 0; x < 32; x++)
2178                 if (gh->gh_flags & (1 << x))
2179                         printk(" %u", x);
2180         printk(" \n");
2181         printk(KERN_INFO "    error = %d\n", gh->gh_error);
2182         printk(KERN_INFO "    gh_iflags =");
2183         for (x = 0; x < 32; x++)
2184                 if (test_bit(x, &gh->gh_iflags))
2185                         printk(" %u", x);
2186         printk(" \n");
2187         print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
2188
2189         error = 0;
2190
2191         return error;
2192 }
2193
2194 /**
2195  * dump_inode - print information about an inode
2196  * @ip: the inode
2197  *
2198  * Returns: 0 on success, -ENOBUFS when we run out of space
2199  */
2200
2201 static int dump_inode(struct gfs2_inode *ip)
2202 {
2203         unsigned int x;
2204         int error = -ENOBUFS;
2205
2206         printk(KERN_INFO "  Inode:\n");
2207         printk(KERN_INFO "    num = %llu %llu\n",
2208                     ip->i_num.no_formal_ino, ip->i_num.no_addr);
2209         printk(KERN_INFO "    type = %u\n", IF2DT(ip->i_di.di_mode));
2210         printk(KERN_INFO "    i_count = %d\n", atomic_read(&ip->i_count));
2211         printk(KERN_INFO "    i_flags =");
2212         for (x = 0; x < 32; x++)
2213                 if (test_bit(x, &ip->i_flags))
2214                         printk(" %u", x);
2215         printk(" \n");
2216         printk(KERN_INFO "    vnode = %s\n", (ip->i_vnode) ? "yes" : "no");
2217
2218         error = 0;
2219
2220         return error;
2221 }
2222
2223 /**
2224  * dump_glock - print information about a glock
2225  * @gl: the glock
2226  * @count: where we are in the buffer
2227  *
2228  * Returns: 0 on success, -ENOBUFS when we run out of space
2229  */
2230
2231 static int dump_glock(struct gfs2_glock *gl)
2232 {
2233         struct gfs2_holder *gh;
2234         unsigned int x;
2235         int error = -ENOBUFS;
2236
2237         spin_lock(&gl->gl_spin);
2238
2239         printk(KERN_INFO "Glock (%u, %llu)\n",
2240                     gl->gl_name.ln_type,
2241                     gl->gl_name.ln_number);
2242         printk(KERN_INFO "  gl_flags =");
2243         for (x = 0; x < 32; x++)
2244                 if (test_bit(x, &gl->gl_flags))
2245                         printk(" %u", x);
2246         printk(" \n");
2247         printk(KERN_INFO "  gl_ref = %d\n", atomic_read(&gl->gl_ref.refcount));
2248         printk(KERN_INFO "  gl_state = %u\n", gl->gl_state);
2249         printk(KERN_INFO "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
2250         printk(KERN_INFO "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
2251         printk(KERN_INFO "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
2252         printk(KERN_INFO "  object = %s\n", (gl->gl_object) ? "yes" : "no");
2253         printk(KERN_INFO "  le = %s\n",
2254                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
2255         printk(KERN_INFO "  reclaim = %s\n",
2256                     (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
2257         if (gl->gl_aspace)
2258                 printk(KERN_INFO "  aspace = %lu\n",
2259                             gl->gl_aspace->i_mapping->nrpages);
2260         else
2261                 printk(KERN_INFO "  aspace = no\n");
2262         printk(KERN_INFO "  ail = %d\n", atomic_read(&gl->gl_ail_count));
2263         if (gl->gl_req_gh) {
2264                 error = dump_holder("Request", gl->gl_req_gh);
2265                 if (error)
2266                         goto out;
2267         }
2268         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
2269                 error = dump_holder("Holder", gh);
2270                 if (error)
2271                         goto out;
2272         }
2273         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
2274                 error = dump_holder("Waiter1", gh);
2275                 if (error)
2276                         goto out;
2277         }
2278         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
2279                 error = dump_holder("Waiter2", gh);
2280                 if (error)
2281                         goto out;
2282         }
2283         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
2284                 error = dump_holder("Waiter3", gh);
2285                 if (error)
2286                         goto out;
2287         }
2288         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
2289                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
2290                     list_empty(&gl->gl_holders)) {
2291                         error = dump_inode(gl->gl_object);
2292                         if (error)
2293                                 goto out;
2294                 } else {
2295                         error = -ENOBUFS;
2296                         printk(KERN_INFO "  Inode: busy\n");
2297                 }
2298         }
2299
2300         error = 0;
2301
2302  out:
2303         spin_unlock(&gl->gl_spin);
2304
2305         return error;
2306 }
2307
2308 /**
2309  * gfs2_dump_lockstate - print out the current lockstate
2310  * @sdp: the filesystem
2311  * @ub: the buffer to copy the information into
2312  *
2313  * If @ub is NULL, dump the lockstate to the console.
2314  *
2315  */
2316
2317 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
2318 {
2319         struct gfs2_gl_hash_bucket *bucket;
2320         struct gfs2_glock *gl;
2321         unsigned int x;
2322         int error = 0;
2323
2324         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2325                 bucket = &sdp->sd_gl_hash[x];
2326
2327                 read_lock(&bucket->hb_lock);
2328
2329                 list_for_each_entry(gl, &bucket->hb_list, gl_list) {
2330                         if (test_bit(GLF_PLUG, &gl->gl_flags))
2331                                 continue;
2332
2333                         error = dump_glock(gl);
2334                         if (error)
2335                                 break;
2336                 }
2337
2338                 read_unlock(&bucket->hb_lock);
2339
2340                 if (error)
2341                         break;
2342         }
2343
2344
2345         return error;
2346 }
2347