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