[GFS2] Compile fix for glock.c
[safe/jmp/linux-2.6] / fs / gfs2 / glock.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/delay.h>
16 #include <linux/sort.h>
17 #include <linux/jhash.h>
18 #include <linux/kallsyms.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/list.h>
21 #include <linux/lm_interface.h>
22 #include <linux/wait.h>
23 #include <asm/uaccess.h>
24
25 #include "gfs2.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 struct gfs2_gl_hash_bucket {
38         struct hlist_head hb_list;
39 };
40
41 typedef void (*glock_examiner) (struct gfs2_glock * gl);
42
43 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
44 static int dump_glock(struct gfs2_glock *gl);
45 static int dump_inode(struct gfs2_inode *ip);
46 static void gfs2_glock_xmote_th(struct gfs2_holder *gh);
47 static void gfs2_glock_drop_th(struct gfs2_glock *gl);
48
49 #define GFS2_GL_HASH_SHIFT      15
50 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
51 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
52
53 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
54
55 /*
56  * Despite what you might think, the numbers below are not arbitrary :-)
57  * They are taken from the ipv4 routing hash code, which is well tested
58  * and thus should be nearly optimal. Later on we might tweek the numbers
59  * but for now this should be fine.
60  *
61  * The reason for putting the locks in a separate array from the list heads
62  * is that we can have fewer locks than list heads and save memory. We use
63  * the same hash function for both, but with a different hash mask.
64  */
65 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
66         defined(CONFIG_PROVE_LOCKING)
67
68 #ifdef CONFIG_LOCKDEP
69 # define GL_HASH_LOCK_SZ        256
70 #else
71 # if NR_CPUS >= 32
72 #  define GL_HASH_LOCK_SZ       4096
73 # elif NR_CPUS >= 16
74 #  define GL_HASH_LOCK_SZ       2048
75 # elif NR_CPUS >= 8
76 #  define GL_HASH_LOCK_SZ       1024
77 # elif NR_CPUS >= 4
78 #  define GL_HASH_LOCK_SZ       512
79 # else
80 #  define GL_HASH_LOCK_SZ       256
81 # endif
82 #endif
83
84 /* We never want more locks than chains */
85 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
86 # undef GL_HASH_LOCK_SZ
87 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
88 #endif
89
90 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
91
92 static inline rwlock_t *gl_lock_addr(unsigned int x)
93 {
94         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
95 }
96 #else /* not SMP, so no spinlocks required */
97 static inline rwlock_t *gl_lock_addr(unsigned int x)
98 {
99         return NULL;
100 }
101 #endif
102
103 /**
104  * relaxed_state_ok - is a requested lock compatible with the current lock mode?
105  * @actual: the current state of the lock
106  * @requested: the lock state that was requested by the caller
107  * @flags: the modifier flags passed in by the caller
108  *
109  * Returns: 1 if the locks are compatible, 0 otherwise
110  */
111
112 static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
113                                    int flags)
114 {
115         if (actual == requested)
116                 return 1;
117
118         if (flags & GL_EXACT)
119                 return 0;
120
121         if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
122                 return 1;
123
124         if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
125                 return 1;
126
127         return 0;
128 }
129
130 /**
131  * gl_hash() - Turn glock number into hash bucket number
132  * @lock: The glock number
133  *
134  * Returns: The number of the corresponding hash bucket
135  */
136
137 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
138                             const struct lm_lockname *name)
139 {
140         unsigned int h;
141
142         h = jhash(&name->ln_number, sizeof(u64), 0);
143         h = jhash(&name->ln_type, sizeof(unsigned int), h);
144         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
145         h &= GFS2_GL_HASH_MASK;
146
147         return h;
148 }
149
150 /**
151  * glock_free() - Perform a few checks and then release struct gfs2_glock
152  * @gl: The glock to release
153  *
154  * Also calls lock module to release its internal structure for this glock.
155  *
156  */
157
158 static void glock_free(struct gfs2_glock *gl)
159 {
160         struct gfs2_sbd *sdp = gl->gl_sbd;
161         struct inode *aspace = gl->gl_aspace;
162
163         gfs2_lm_put_lock(sdp, gl->gl_lock);
164
165         if (aspace)
166                 gfs2_aspace_put(aspace);
167
168         kmem_cache_free(gfs2_glock_cachep, gl);
169 }
170
171 /**
172  * gfs2_glock_hold() - increment reference count on glock
173  * @gl: The glock to hold
174  *
175  */
176
177 void gfs2_glock_hold(struct gfs2_glock *gl)
178 {
179         atomic_inc(&gl->gl_ref);
180 }
181
182 /**
183  * gfs2_glock_put() - Decrement reference count on glock
184  * @gl: The glock to put
185  *
186  */
187
188 int gfs2_glock_put(struct gfs2_glock *gl)
189 {
190         int rv = 0;
191         struct gfs2_sbd *sdp = gl->gl_sbd;
192
193         write_lock(gl_lock_addr(gl->gl_hash));
194         if (atomic_dec_and_test(&gl->gl_ref)) {
195                 hlist_del(&gl->gl_list);
196                 write_unlock(gl_lock_addr(gl->gl_hash));
197                 BUG_ON(spin_is_locked(&gl->gl_spin));
198                 gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
199                 gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
200                 gfs2_assert(sdp, list_empty(&gl->gl_holders));
201                 gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
202                 gfs2_assert(sdp, list_empty(&gl->gl_waiters2));
203                 gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
204                 glock_free(gl);
205                 rv = 1;
206                 goto out;
207         }
208         write_unlock(gl_lock_addr(gl->gl_hash));
209 out:
210         return rv;
211 }
212
213 /**
214  * search_bucket() - Find struct gfs2_glock by lock number
215  * @bucket: the bucket to search
216  * @name: The lock name
217  *
218  * Returns: NULL, or the struct gfs2_glock with the requested number
219  */
220
221 static struct gfs2_glock *search_bucket(unsigned int hash,
222                                         const struct gfs2_sbd *sdp,
223                                         const struct lm_lockname *name)
224 {
225         struct gfs2_glock *gl;
226         struct hlist_node *h;
227
228         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
229                 if (!lm_name_equal(&gl->gl_name, name))
230                         continue;
231                 if (gl->gl_sbd != sdp)
232                         continue;
233
234                 atomic_inc(&gl->gl_ref);
235
236                 return gl;
237         }
238
239         return NULL;
240 }
241
242 /**
243  * gfs2_glock_find() - Find glock by lock number
244  * @sdp: The GFS2 superblock
245  * @name: The lock name
246  *
247  * Returns: NULL, or the struct gfs2_glock with the requested number
248  */
249
250 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
251                                           const struct lm_lockname *name)
252 {
253         unsigned int hash = gl_hash(sdp, name);
254         struct gfs2_glock *gl;
255
256         read_lock(gl_lock_addr(hash));
257         gl = search_bucket(hash, sdp, name);
258         read_unlock(gl_lock_addr(hash));
259
260         return gl;
261 }
262
263 /**
264  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
265  * @sdp: The GFS2 superblock
266  * @number: the lock number
267  * @glops: The glock_operations to use
268  * @create: If 0, don't create the glock if it doesn't exist
269  * @glp: the glock is returned here
270  *
271  * This does not lock a glock, just finds/creates structures for one.
272  *
273  * Returns: errno
274  */
275
276 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
277                    const struct gfs2_glock_operations *glops, int create,
278                    struct gfs2_glock **glp)
279 {
280         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
281         struct gfs2_glock *gl, *tmp;
282         unsigned int hash = gl_hash(sdp, &name);
283         int error;
284
285         read_lock(gl_lock_addr(hash));
286         gl = search_bucket(hash, sdp, &name);
287         read_unlock(gl_lock_addr(hash));
288
289         if (gl || !create) {
290                 *glp = gl;
291                 return 0;
292         }
293
294         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
295         if (!gl)
296                 return -ENOMEM;
297
298         gl->gl_flags = 0;
299         gl->gl_name = name;
300         atomic_set(&gl->gl_ref, 1);
301         gl->gl_state = LM_ST_UNLOCKED;
302         gl->gl_hash = hash;
303         gl->gl_owner = NULL;
304         gl->gl_ip = 0;
305         gl->gl_ops = glops;
306         gl->gl_req_gh = NULL;
307         gl->gl_req_bh = NULL;
308         gl->gl_vn = 0;
309         gl->gl_stamp = jiffies;
310         gl->gl_object = NULL;
311         gl->gl_sbd = sdp;
312         gl->gl_aspace = NULL;
313         lops_init_le(&gl->gl_le, &gfs2_glock_lops);
314
315         /* If this glock protects actual on-disk data or metadata blocks,
316            create a VFS inode to manage the pages/buffers holding them. */
317         if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
318                 gl->gl_aspace = gfs2_aspace_get(sdp);
319                 if (!gl->gl_aspace) {
320                         error = -ENOMEM;
321                         goto fail;
322                 }
323         }
324
325         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
326         if (error)
327                 goto fail_aspace;
328
329         write_lock(gl_lock_addr(hash));
330         tmp = search_bucket(hash, sdp, &name);
331         if (tmp) {
332                 write_unlock(gl_lock_addr(hash));
333                 glock_free(gl);
334                 gl = tmp;
335         } else {
336                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
337                 write_unlock(gl_lock_addr(hash));
338         }
339
340         *glp = gl;
341
342         return 0;
343
344 fail_aspace:
345         if (gl->gl_aspace)
346                 gfs2_aspace_put(gl->gl_aspace);
347 fail:
348         kmem_cache_free(gfs2_glock_cachep, gl);
349         return error;
350 }
351
352 /**
353  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
354  * @gl: the glock
355  * @state: the state we're requesting
356  * @flags: the modifier flags
357  * @gh: the holder structure
358  *
359  */
360
361 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
362                       struct gfs2_holder *gh)
363 {
364         INIT_LIST_HEAD(&gh->gh_list);
365         gh->gh_gl = gl;
366         gh->gh_ip = (unsigned long)__builtin_return_address(0);
367         gh->gh_owner = current;
368         gh->gh_state = state;
369         gh->gh_flags = flags;
370         gh->gh_error = 0;
371         gh->gh_iflags = 0;
372         gfs2_glock_hold(gl);
373 }
374
375 /**
376  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
377  * @state: the state we're requesting
378  * @flags: the modifier flags
379  * @gh: the holder structure
380  *
381  * Don't mess with the glock.
382  *
383  */
384
385 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
386 {
387         gh->gh_state = state;
388         gh->gh_flags = flags;
389         gh->gh_iflags &= 1 << HIF_ALLOCED;
390         gh->gh_ip = (unsigned long)__builtin_return_address(0);
391 }
392
393 /**
394  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
395  * @gh: the holder structure
396  *
397  */
398
399 void gfs2_holder_uninit(struct gfs2_holder *gh)
400 {
401         gfs2_glock_put(gh->gh_gl);
402         gh->gh_gl = NULL;
403         gh->gh_ip = 0;
404 }
405
406 /**
407  * gfs2_holder_get - get a struct gfs2_holder structure
408  * @gl: the glock
409  * @state: the state we're requesting
410  * @flags: the modifier flags
411  * @gfp_flags:
412  *
413  * Figure out how big an impact this function has.  Either:
414  * 1) Replace it with a cache of structures hanging off the struct gfs2_sbd
415  * 2) Leave it like it is
416  *
417  * Returns: the holder structure, NULL on ENOMEM
418  */
419
420 static struct gfs2_holder *gfs2_holder_get(struct gfs2_glock *gl,
421                                            unsigned int state,
422                                            int flags, gfp_t gfp_flags)
423 {
424         struct gfs2_holder *gh;
425
426         gh = kmalloc(sizeof(struct gfs2_holder), gfp_flags);
427         if (!gh)
428                 return NULL;
429
430         gfs2_holder_init(gl, state, flags, gh);
431         set_bit(HIF_ALLOCED, &gh->gh_iflags);
432         gh->gh_ip = (unsigned long)__builtin_return_address(0);
433         return gh;
434 }
435
436 /**
437  * gfs2_holder_put - get rid of a struct gfs2_holder structure
438  * @gh: the holder structure
439  *
440  */
441
442 static void gfs2_holder_put(struct gfs2_holder *gh)
443 {
444         gfs2_holder_uninit(gh);
445         kfree(gh);
446 }
447
448 static void gfs2_holder_dispose_or_wake(struct gfs2_holder *gh)
449 {
450         if (test_bit(HIF_DEALLOC, &gh->gh_iflags)) {
451                 gfs2_holder_put(gh);
452                 return;
453         }
454         clear_bit(HIF_WAIT, &gh->gh_iflags);
455         smp_mb();
456         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
457 }
458
459 static int holder_wait(void *word)
460 {
461         schedule();
462         return 0;
463 }
464
465 static void wait_on_holder(struct gfs2_holder *gh)
466 {
467         might_sleep();
468         wait_on_bit(&gh->gh_iflags, HIF_WAIT, holder_wait, TASK_UNINTERRUPTIBLE);
469 }
470
471 /**
472  * rq_mutex - process a mutex request in the queue
473  * @gh: the glock holder
474  *
475  * Returns: 1 if the queue is blocked
476  */
477
478 static int rq_mutex(struct gfs2_holder *gh)
479 {
480         struct gfs2_glock *gl = gh->gh_gl;
481
482         list_del_init(&gh->gh_list);
483         /*  gh->gh_error never examined.  */
484         set_bit(GLF_LOCK, &gl->gl_flags);
485         clear_bit(HIF_WAIT, &gh->gh_flags);
486         smp_mb();
487         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
488
489         return 1;
490 }
491
492 /**
493  * rq_promote - process a promote request in the queue
494  * @gh: the glock holder
495  *
496  * Acquire a new inter-node lock, or change a lock state to more restrictive.
497  *
498  * Returns: 1 if the queue is blocked
499  */
500
501 static int rq_promote(struct gfs2_holder *gh)
502 {
503         struct gfs2_glock *gl = gh->gh_gl;
504         struct gfs2_sbd *sdp = gl->gl_sbd;
505
506         if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
507                 if (list_empty(&gl->gl_holders)) {
508                         gl->gl_req_gh = gh;
509                         set_bit(GLF_LOCK, &gl->gl_flags);
510                         spin_unlock(&gl->gl_spin);
511
512                         if (atomic_read(&sdp->sd_reclaim_count) >
513                             gfs2_tune_get(sdp, gt_reclaim_limit) &&
514                             !(gh->gh_flags & LM_FLAG_PRIORITY)) {
515                                 gfs2_reclaim_glock(sdp);
516                                 gfs2_reclaim_glock(sdp);
517                         }
518
519                         gfs2_glock_xmote_th(gh);
520                         spin_lock(&gl->gl_spin);
521                 }
522                 return 1;
523         }
524
525         if (list_empty(&gl->gl_holders)) {
526                 set_bit(HIF_FIRST, &gh->gh_iflags);
527                 set_bit(GLF_LOCK, &gl->gl_flags);
528         } else {
529                 struct gfs2_holder *next_gh;
530                 if (gh->gh_state == LM_ST_EXCLUSIVE)
531                         return 1;
532                 next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
533                                      gh_list);
534                 if (next_gh->gh_state == LM_ST_EXCLUSIVE)
535                          return 1;
536         }
537
538         list_move_tail(&gh->gh_list, &gl->gl_holders);
539         gh->gh_error = 0;
540         set_bit(HIF_HOLDER, &gh->gh_iflags);
541
542         gfs2_holder_dispose_or_wake(gh);
543
544         return 0;
545 }
546
547 /**
548  * rq_demote - process a demote request in the queue
549  * @gh: the glock holder
550  *
551  * Returns: 1 if the queue is blocked
552  */
553
554 static int rq_demote(struct gfs2_holder *gh)
555 {
556         struct gfs2_glock *gl = gh->gh_gl;
557
558         if (!list_empty(&gl->gl_holders))
559                 return 1;
560
561         if (gl->gl_state == gh->gh_state || gl->gl_state == LM_ST_UNLOCKED) {
562                 list_del_init(&gh->gh_list);
563                 gh->gh_error = 0;
564                 spin_unlock(&gl->gl_spin);
565                 gfs2_holder_dispose_or_wake(gh);
566                 spin_lock(&gl->gl_spin);
567         } else {
568                 gl->gl_req_gh = gh;
569                 set_bit(GLF_LOCK, &gl->gl_flags);
570                 spin_unlock(&gl->gl_spin);
571
572                 if (gh->gh_state == LM_ST_UNLOCKED ||
573                     gl->gl_state != LM_ST_EXCLUSIVE)
574                         gfs2_glock_drop_th(gl);
575                 else
576                         gfs2_glock_xmote_th(gh);
577
578                 spin_lock(&gl->gl_spin);
579         }
580
581         return 0;
582 }
583
584 /**
585  * run_queue - process holder structures on a glock
586  * @gl: the glock
587  *
588  */
589 static void run_queue(struct gfs2_glock *gl)
590 {
591         struct gfs2_holder *gh;
592         int blocked = 1;
593
594         for (;;) {
595                 if (test_bit(GLF_LOCK, &gl->gl_flags))
596                         break;
597
598                 if (!list_empty(&gl->gl_waiters1)) {
599                         gh = list_entry(gl->gl_waiters1.next,
600                                         struct gfs2_holder, gh_list);
601
602                         if (test_bit(HIF_MUTEX, &gh->gh_iflags))
603                                 blocked = rq_mutex(gh);
604                         else
605                                 gfs2_assert_warn(gl->gl_sbd, 0);
606
607                 } else if (!list_empty(&gl->gl_waiters2) &&
608                            !test_bit(GLF_SKIP_WAITERS2, &gl->gl_flags)) {
609                         gh = list_entry(gl->gl_waiters2.next,
610                                         struct gfs2_holder, gh_list);
611
612                         if (test_bit(HIF_DEMOTE, &gh->gh_iflags))
613                                 blocked = rq_demote(gh);
614                         else
615                                 gfs2_assert_warn(gl->gl_sbd, 0);
616
617                 } else if (!list_empty(&gl->gl_waiters3)) {
618                         gh = list_entry(gl->gl_waiters3.next,
619                                         struct gfs2_holder, gh_list);
620
621                         if (test_bit(HIF_PROMOTE, &gh->gh_iflags))
622                                 blocked = rq_promote(gh);
623                         else
624                                 gfs2_assert_warn(gl->gl_sbd, 0);
625
626                 } else
627                         break;
628
629                 if (blocked)
630                         break;
631         }
632 }
633
634 /**
635  * gfs2_glmutex_lock - acquire a local lock on a glock
636  * @gl: the glock
637  *
638  * Gives caller exclusive access to manipulate a glock structure.
639  */
640
641 static void gfs2_glmutex_lock(struct gfs2_glock *gl)
642 {
643         struct gfs2_holder gh;
644
645         gfs2_holder_init(gl, 0, 0, &gh);
646         set_bit(HIF_MUTEX, &gh.gh_iflags);
647         if (test_and_set_bit(HIF_WAIT, &gh.gh_iflags))
648                 BUG();
649
650         spin_lock(&gl->gl_spin);
651         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
652                 list_add_tail(&gh.gh_list, &gl->gl_waiters1);
653         } else {
654                 gl->gl_owner = current;
655                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
656                 clear_bit(HIF_WAIT, &gh.gh_iflags);
657                 smp_mb();
658                 wake_up_bit(&gh.gh_iflags, HIF_WAIT);
659         }
660         spin_unlock(&gl->gl_spin);
661
662         wait_on_holder(&gh);
663         gfs2_holder_uninit(&gh);
664 }
665
666 /**
667  * gfs2_glmutex_trylock - try to acquire a local lock on a glock
668  * @gl: the glock
669  *
670  * Returns: 1 if the glock is acquired
671  */
672
673 static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
674 {
675         int acquired = 1;
676
677         spin_lock(&gl->gl_spin);
678         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
679                 acquired = 0;
680         } else {
681                 gl->gl_owner = current;
682                 gl->gl_ip = (unsigned long)__builtin_return_address(0);
683         }
684         spin_unlock(&gl->gl_spin);
685
686         return acquired;
687 }
688
689 /**
690  * gfs2_glmutex_unlock - release a local lock on a glock
691  * @gl: the glock
692  *
693  */
694
695 static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
696 {
697         spin_lock(&gl->gl_spin);
698         clear_bit(GLF_LOCK, &gl->gl_flags);
699         gl->gl_owner = NULL;
700         gl->gl_ip = 0;
701         run_queue(gl);
702         BUG_ON(!spin_is_locked(&gl->gl_spin));
703         spin_unlock(&gl->gl_spin);
704 }
705
706 /**
707  * handle_callback - add a demote request to a lock's queue
708  * @gl: the glock
709  * @state: the state the caller wants us to change to
710  *
711  * Note: This may fail sliently if we are out of memory.
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, GFP_NOFS);
737                 if (!new_gh)
738                         return;
739                 set_bit(HIF_DEMOTE, &new_gh->gh_iflags);
740                 set_bit(HIF_DEALLOC, &new_gh->gh_iflags);
741                 set_bit(HIF_WAIT, &new_gh->gh_iflags);
742
743                 goto restart;
744         }
745
746 out:
747         spin_unlock(&gl->gl_spin);
748
749         if (new_gh)
750                 gfs2_holder_put(new_gh);
751 }
752
753 /**
754  * state_change - record that the glock is now in a different state
755  * @gl: the glock
756  * @new_state the new state
757  *
758  */
759
760 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
761 {
762         int held1, held2;
763
764         held1 = (gl->gl_state != LM_ST_UNLOCKED);
765         held2 = (new_state != LM_ST_UNLOCKED);
766
767         if (held1 != held2) {
768                 if (held2)
769                         gfs2_glock_hold(gl);
770                 else
771                         gfs2_glock_put(gl);
772         }
773
774         gl->gl_state = new_state;
775 }
776
777 /**
778  * xmote_bh - Called after the lock module is done acquiring a lock
779  * @gl: The glock in question
780  * @ret: the int returned from the lock module
781  *
782  */
783
784 static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
785 {
786         struct gfs2_sbd *sdp = gl->gl_sbd;
787         const struct gfs2_glock_operations *glops = gl->gl_ops;
788         struct gfs2_holder *gh = gl->gl_req_gh;
789         int prev_state = gl->gl_state;
790         int op_done = 1;
791
792         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
793         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
794         gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
795
796         state_change(gl, ret & LM_OUT_ST_MASK);
797
798         if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
799                 if (glops->go_inval)
800                         glops->go_inval(gl, DIO_METADATA);
801         } else if (gl->gl_state == LM_ST_DEFERRED) {
802                 /* We might not want to do this here.
803                    Look at moving to the inode glops. */
804                 if (glops->go_inval)
805                         glops->go_inval(gl, 0);
806         }
807
808         /*  Deal with each possible exit condition  */
809
810         if (!gh)
811                 gl->gl_stamp = jiffies;
812         else if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
813                 spin_lock(&gl->gl_spin);
814                 list_del_init(&gh->gh_list);
815                 gh->gh_error = -EIO;
816                 spin_unlock(&gl->gl_spin);
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);
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                 gfs2_holder_dispose_or_wake(gh);
878 }
879
880 /**
881  * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
882  * @gl: The glock in question
883  * @state: the requested state
884  * @flags: modifier flags to the lock call
885  *
886  */
887
888 void gfs2_glock_xmote_th(struct gfs2_holder *gh)
889 {
890         struct gfs2_glock *gl = gh->gh_gl;
891         struct gfs2_sbd *sdp = gl->gl_sbd;
892         int flags = gh->gh_flags;
893         unsigned state = gh->gh_state;
894         const 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         if (glops->go_xmote_th)
901                 glops->go_xmote_th(gl);
902
903         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
904         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
905         gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
906         gfs2_assert_warn(sdp, state != gl->gl_state);
907
908         gfs2_glock_hold(gl);
909         gl->gl_req_bh = xmote_bh;
910
911         lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
912
913         if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
914                 return;
915
916         if (lck_ret & LM_OUT_ASYNC)
917                 gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
918         else
919                 xmote_bh(gl, lck_ret);
920 }
921
922 /**
923  * drop_bh - Called after a lock module unlock completes
924  * @gl: the glock
925  * @ret: the return status
926  *
927  * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
928  * Doesn't drop the reference on the glock the top half took out
929  *
930  */
931
932 static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
933 {
934         struct gfs2_sbd *sdp = gl->gl_sbd;
935         const struct gfs2_glock_operations *glops = gl->gl_ops;
936         struct gfs2_holder *gh = gl->gl_req_gh;
937
938         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
939         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
940         gfs2_assert_warn(sdp, !ret);
941
942         state_change(gl, LM_ST_UNLOCKED);
943
944         if (glops->go_inval)
945                 glops->go_inval(gl, DIO_METADATA);
946
947         if (gh) {
948                 spin_lock(&gl->gl_spin);
949                 list_del_init(&gh->gh_list);
950                 gh->gh_error = 0;
951                 spin_unlock(&gl->gl_spin);
952         }
953
954         if (glops->go_drop_bh)
955                 glops->go_drop_bh(gl);
956
957         spin_lock(&gl->gl_spin);
958         gl->gl_req_gh = NULL;
959         gl->gl_req_bh = NULL;
960         clear_bit(GLF_LOCK, &gl->gl_flags);
961         run_queue(gl);
962         spin_unlock(&gl->gl_spin);
963
964         gfs2_glock_put(gl);
965
966         if (gh)
967                 gfs2_holder_dispose_or_wake(gh);
968 }
969
970 /**
971  * gfs2_glock_drop_th - call into the lock module to unlock a lock
972  * @gl: the glock
973  *
974  */
975
976 static void gfs2_glock_drop_th(struct gfs2_glock *gl)
977 {
978         struct gfs2_sbd *sdp = gl->gl_sbd;
979         const struct gfs2_glock_operations *glops = gl->gl_ops;
980         unsigned int ret;
981
982         if (glops->go_drop_th)
983                 glops->go_drop_th(gl);
984
985         gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
986         gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
987         gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
988
989         gfs2_glock_hold(gl);
990         gl->gl_req_bh = drop_bh;
991
992         ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
993
994         if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
995                 return;
996
997         if (!ret)
998                 drop_bh(gl, ret);
999         else
1000                 gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
1001 }
1002
1003 /**
1004  * do_cancels - cancel requests for locks stuck waiting on an expire flag
1005  * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
1006  *
1007  * Don't cancel GL_NOCANCEL requests.
1008  */
1009
1010 static void do_cancels(struct gfs2_holder *gh)
1011 {
1012         struct gfs2_glock *gl = gh->gh_gl;
1013
1014         spin_lock(&gl->gl_spin);
1015
1016         while (gl->gl_req_gh != gh &&
1017                !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1018                !list_empty(&gh->gh_list)) {
1019                 if (gl->gl_req_bh && !(gl->gl_req_gh &&
1020                                      (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
1021                         spin_unlock(&gl->gl_spin);
1022                         gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
1023                         msleep(100);
1024                         spin_lock(&gl->gl_spin);
1025                 } else {
1026                         spin_unlock(&gl->gl_spin);
1027                         msleep(100);
1028                         spin_lock(&gl->gl_spin);
1029                 }
1030         }
1031
1032         spin_unlock(&gl->gl_spin);
1033 }
1034
1035 /**
1036  * glock_wait_internal - wait on a glock acquisition
1037  * @gh: the glock holder
1038  *
1039  * Returns: 0 on success
1040  */
1041
1042 static int glock_wait_internal(struct gfs2_holder *gh)
1043 {
1044         struct gfs2_glock *gl = gh->gh_gl;
1045         struct gfs2_sbd *sdp = gl->gl_sbd;
1046         const struct gfs2_glock_operations *glops = gl->gl_ops;
1047
1048         if (test_bit(HIF_ABORTED, &gh->gh_iflags))
1049                 return -EIO;
1050
1051         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
1052                 spin_lock(&gl->gl_spin);
1053                 if (gl->gl_req_gh != gh &&
1054                     !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
1055                     !list_empty(&gh->gh_list)) {
1056                         list_del_init(&gh->gh_list);
1057                         gh->gh_error = GLR_TRYFAILED;
1058                         run_queue(gl);
1059                         spin_unlock(&gl->gl_spin);
1060                         return gh->gh_error;
1061                 }
1062                 spin_unlock(&gl->gl_spin);
1063         }
1064
1065         if (gh->gh_flags & LM_FLAG_PRIORITY)
1066                 do_cancels(gh);
1067
1068         wait_on_holder(gh);
1069         if (gh->gh_error)
1070                 return gh->gh_error;
1071
1072         gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
1073         gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
1074                                                    gh->gh_flags));
1075
1076         if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
1077                 gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
1078
1079                 if (glops->go_lock) {
1080                         gh->gh_error = glops->go_lock(gh);
1081                         if (gh->gh_error) {
1082                                 spin_lock(&gl->gl_spin);
1083                                 list_del_init(&gh->gh_list);
1084                                 spin_unlock(&gl->gl_spin);
1085                         }
1086                 }
1087
1088                 spin_lock(&gl->gl_spin);
1089                 gl->gl_req_gh = NULL;
1090                 gl->gl_req_bh = NULL;
1091                 clear_bit(GLF_LOCK, &gl->gl_flags);
1092                 run_queue(gl);
1093                 spin_unlock(&gl->gl_spin);
1094         }
1095
1096         return gh->gh_error;
1097 }
1098
1099 static inline struct gfs2_holder *
1100 find_holder_by_owner(struct list_head *head, struct task_struct *owner)
1101 {
1102         struct gfs2_holder *gh;
1103
1104         list_for_each_entry(gh, head, gh_list) {
1105                 if (gh->gh_owner == owner)
1106                         return gh;
1107         }
1108
1109         return NULL;
1110 }
1111
1112 /**
1113  * add_to_queue - Add a holder to the wait queue (but look for recursion)
1114  * @gh: the holder structure to add
1115  *
1116  */
1117
1118 static void add_to_queue(struct gfs2_holder *gh)
1119 {
1120         struct gfs2_glock *gl = gh->gh_gl;
1121         struct gfs2_holder *existing;
1122
1123         BUG_ON(!gh->gh_owner);
1124         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
1125                 BUG();
1126
1127         existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner);
1128         if (existing) {
1129                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1130                 printk(KERN_INFO "pid : %d\n", existing->gh_owner->pid);
1131                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1132                                 existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
1133                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1134                 printk(KERN_INFO "pid : %d\n", gh->gh_owner->pid);
1135                 printk(KERN_INFO "lock type : %d lock state : %d\n",
1136                                 gl->gl_name.ln_type, gl->gl_state);
1137                 BUG();
1138         }
1139
1140         existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner);
1141         if (existing) {
1142                 print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
1143                 print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
1144                 BUG();
1145         }
1146
1147         if (gh->gh_flags & LM_FLAG_PRIORITY)
1148                 list_add(&gh->gh_list, &gl->gl_waiters3);
1149         else
1150                 list_add_tail(&gh->gh_list, &gl->gl_waiters3);
1151 }
1152
1153 /**
1154  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1155  * @gh: the holder structure
1156  *
1157  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1158  *
1159  * Returns: 0, GLR_TRYFAILED, or errno on failure
1160  */
1161
1162 int gfs2_glock_nq(struct gfs2_holder *gh)
1163 {
1164         struct gfs2_glock *gl = gh->gh_gl;
1165         struct gfs2_sbd *sdp = gl->gl_sbd;
1166         int error = 0;
1167
1168 restart:
1169         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
1170                 set_bit(HIF_ABORTED, &gh->gh_iflags);
1171                 return -EIO;
1172         }
1173
1174         set_bit(HIF_PROMOTE, &gh->gh_iflags);
1175
1176         spin_lock(&gl->gl_spin);
1177         add_to_queue(gh);
1178         run_queue(gl);
1179         spin_unlock(&gl->gl_spin);
1180
1181         if (!(gh->gh_flags & GL_ASYNC)) {
1182                 error = glock_wait_internal(gh);
1183                 if (error == GLR_CANCELED) {
1184                         msleep(100);
1185                         goto restart;
1186                 }
1187         }
1188
1189         return error;
1190 }
1191
1192 /**
1193  * gfs2_glock_poll - poll to see if an async request has been completed
1194  * @gh: the holder
1195  *
1196  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1197  */
1198
1199 int gfs2_glock_poll(struct gfs2_holder *gh)
1200 {
1201         struct gfs2_glock *gl = gh->gh_gl;
1202         int ready = 0;
1203
1204         spin_lock(&gl->gl_spin);
1205
1206         if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1207                 ready = 1;
1208         else if (list_empty(&gh->gh_list)) {
1209                 if (gh->gh_error == GLR_CANCELED) {
1210                         spin_unlock(&gl->gl_spin);
1211                         msleep(100);
1212                         if (gfs2_glock_nq(gh))
1213                                 return 1;
1214                         return 0;
1215                 } else
1216                         ready = 1;
1217         }
1218
1219         spin_unlock(&gl->gl_spin);
1220
1221         return ready;
1222 }
1223
1224 /**
1225  * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
1226  * @gh: the holder structure
1227  *
1228  * Returns: 0, GLR_TRYFAILED, or errno on failure
1229  */
1230
1231 int gfs2_glock_wait(struct gfs2_holder *gh)
1232 {
1233         int error;
1234
1235         error = glock_wait_internal(gh);
1236         if (error == GLR_CANCELED) {
1237                 msleep(100);
1238                 gh->gh_flags &= ~GL_ASYNC;
1239                 error = gfs2_glock_nq(gh);
1240         }
1241
1242         return error;
1243 }
1244
1245 /**
1246  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1247  * @gh: the glock holder
1248  *
1249  */
1250
1251 void gfs2_glock_dq(struct gfs2_holder *gh)
1252 {
1253         struct gfs2_glock *gl = gh->gh_gl;
1254         const struct gfs2_glock_operations *glops = gl->gl_ops;
1255
1256         if (gh->gh_flags & GL_NOCACHE)
1257                 handle_callback(gl, LM_ST_UNLOCKED);
1258
1259         gfs2_glmutex_lock(gl);
1260
1261         spin_lock(&gl->gl_spin);
1262         list_del_init(&gh->gh_list);
1263
1264         if (list_empty(&gl->gl_holders)) {
1265                 spin_unlock(&gl->gl_spin);
1266
1267                 if (glops->go_unlock)
1268                         glops->go_unlock(gh);
1269
1270                 gl->gl_stamp = jiffies;
1271
1272                 spin_lock(&gl->gl_spin);
1273         }
1274
1275         clear_bit(GLF_LOCK, &gl->gl_flags);
1276         run_queue(gl);
1277         spin_unlock(&gl->gl_spin);
1278 }
1279
1280 /**
1281  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1282  * @gh: the holder structure
1283  *
1284  */
1285
1286 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1287 {
1288         gfs2_glock_dq(gh);
1289         gfs2_holder_uninit(gh);
1290 }
1291
1292 /**
1293  * gfs2_glock_nq_num - acquire a glock based on lock number
1294  * @sdp: the filesystem
1295  * @number: the lock number
1296  * @glops: the glock operations for the type of glock
1297  * @state: the state to acquire the glock in
1298  * @flags: modifier flags for the aquisition
1299  * @gh: the struct gfs2_holder
1300  *
1301  * Returns: errno
1302  */
1303
1304 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1305                       const struct gfs2_glock_operations *glops,
1306                       unsigned int state, int flags, struct gfs2_holder *gh)
1307 {
1308         struct gfs2_glock *gl;
1309         int error;
1310
1311         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1312         if (!error) {
1313                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1314                 gfs2_glock_put(gl);
1315         }
1316
1317         return error;
1318 }
1319
1320 /**
1321  * glock_compare - Compare two struct gfs2_glock structures for sorting
1322  * @arg_a: the first structure
1323  * @arg_b: the second structure
1324  *
1325  */
1326
1327 static int glock_compare(const void *arg_a, const void *arg_b)
1328 {
1329         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1330         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1331         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1332         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1333
1334         if (a->ln_number > b->ln_number)
1335                 return 1;
1336         if (a->ln_number < b->ln_number)
1337                 return -1;
1338         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1339         return 0;
1340 }
1341
1342 /**
1343  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1344  * @num_gh: the number of structures
1345  * @ghs: an array of struct gfs2_holder structures
1346  *
1347  * Returns: 0 on success (all glocks acquired),
1348  *          errno on failure (no glocks acquired)
1349  */
1350
1351 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1352                      struct gfs2_holder **p)
1353 {
1354         unsigned int x;
1355         int error = 0;
1356
1357         for (x = 0; x < num_gh; x++)
1358                 p[x] = &ghs[x];
1359
1360         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1361
1362         for (x = 0; x < num_gh; x++) {
1363                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1364
1365                 error = gfs2_glock_nq(p[x]);
1366                 if (error) {
1367                         while (x--)
1368                                 gfs2_glock_dq(p[x]);
1369                         break;
1370                 }
1371         }
1372
1373         return error;
1374 }
1375
1376 /**
1377  * gfs2_glock_nq_m - acquire multiple glocks
1378  * @num_gh: the number of structures
1379  * @ghs: an array of struct gfs2_holder structures
1380  *
1381  * Figure out how big an impact this function has.  Either:
1382  * 1) Replace this code with code that calls gfs2_glock_prefetch()
1383  * 2) Forget async stuff and just call nq_m_sync()
1384  * 3) Leave it like it is
1385  *
1386  * Returns: 0 on success (all glocks acquired),
1387  *          errno on failure (no glocks acquired)
1388  */
1389
1390 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1391 {
1392         int *e;
1393         unsigned int x;
1394         int borked = 0, serious = 0;
1395         int error = 0;
1396
1397         if (!num_gh)
1398                 return 0;
1399
1400         if (num_gh == 1) {
1401                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1402                 return gfs2_glock_nq(ghs);
1403         }
1404
1405         e = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1406         if (!e)
1407                 return -ENOMEM;
1408
1409         for (x = 0; x < num_gh; x++) {
1410                 ghs[x].gh_flags |= LM_FLAG_TRY | GL_ASYNC;
1411                 error = gfs2_glock_nq(&ghs[x]);
1412                 if (error) {
1413                         borked = 1;
1414                         serious = error;
1415                         num_gh = x;
1416                         break;
1417                 }
1418         }
1419
1420         for (x = 0; x < num_gh; x++) {
1421                 error = e[x] = glock_wait_internal(&ghs[x]);
1422                 if (error) {
1423                         borked = 1;
1424                         if (error != GLR_TRYFAILED && error != GLR_CANCELED)
1425                                 serious = error;
1426                 }
1427         }
1428
1429         if (!borked) {
1430                 kfree(e);
1431                 return 0;
1432         }
1433
1434         for (x = 0; x < num_gh; x++)
1435                 if (!e[x])
1436                         gfs2_glock_dq(&ghs[x]);
1437
1438         if (serious)
1439                 error = serious;
1440         else {
1441                 for (x = 0; x < num_gh; x++)
1442                         gfs2_holder_reinit(ghs[x].gh_state, ghs[x].gh_flags,
1443                                           &ghs[x]);
1444                 error = nq_m_sync(num_gh, ghs, (struct gfs2_holder **)e);
1445         }
1446
1447         kfree(e);
1448
1449         return error;
1450 }
1451
1452 /**
1453  * gfs2_glock_dq_m - release multiple glocks
1454  * @num_gh: the number of structures
1455  * @ghs: an array of struct gfs2_holder structures
1456  *
1457  */
1458
1459 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1460 {
1461         unsigned int x;
1462
1463         for (x = 0; x < num_gh; x++)
1464                 gfs2_glock_dq(&ghs[x]);
1465 }
1466
1467 /**
1468  * gfs2_glock_dq_uninit_m - release multiple glocks
1469  * @num_gh: the number of structures
1470  * @ghs: an array of struct gfs2_holder structures
1471  *
1472  */
1473
1474 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1475 {
1476         unsigned int x;
1477
1478         for (x = 0; x < num_gh; x++)
1479                 gfs2_glock_dq_uninit(&ghs[x]);
1480 }
1481
1482 /**
1483  * gfs2_lvb_hold - attach a LVB from a glock
1484  * @gl: The glock in question
1485  *
1486  */
1487
1488 int gfs2_lvb_hold(struct gfs2_glock *gl)
1489 {
1490         int error;
1491
1492         gfs2_glmutex_lock(gl);
1493
1494         if (!atomic_read(&gl->gl_lvb_count)) {
1495                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1496                 if (error) {
1497                         gfs2_glmutex_unlock(gl);
1498                         return error;
1499                 }
1500                 gfs2_glock_hold(gl);
1501         }
1502         atomic_inc(&gl->gl_lvb_count);
1503
1504         gfs2_glmutex_unlock(gl);
1505
1506         return 0;
1507 }
1508
1509 /**
1510  * gfs2_lvb_unhold - detach a LVB from a glock
1511  * @gl: The glock in question
1512  *
1513  */
1514
1515 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1516 {
1517         gfs2_glock_hold(gl);
1518         gfs2_glmutex_lock(gl);
1519
1520         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1521         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1522                 gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
1523                 gl->gl_lvb = NULL;
1524                 gfs2_glock_put(gl);
1525         }
1526
1527         gfs2_glmutex_unlock(gl);
1528         gfs2_glock_put(gl);
1529 }
1530
1531 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1532                         unsigned int state)
1533 {
1534         struct gfs2_glock *gl;
1535
1536         gl = gfs2_glock_find(sdp, name);
1537         if (!gl)
1538                 return;
1539
1540         handle_callback(gl, state);
1541
1542         spin_lock(&gl->gl_spin);
1543         run_queue(gl);
1544         spin_unlock(&gl->gl_spin);
1545
1546         gfs2_glock_put(gl);
1547 }
1548
1549 /**
1550  * gfs2_glock_cb - Callback used by locking module
1551  * @sdp: Pointer to the superblock
1552  * @type: Type of callback
1553  * @data: Type dependent data pointer
1554  *
1555  * Called by the locking module when it wants to tell us something.
1556  * Either we need to drop a lock, one of our ASYNC requests completed, or
1557  * a journal from another client needs to be recovered.
1558  */
1559
1560 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1561 {
1562         struct gfs2_sbd *sdp = cb_data;
1563
1564         switch (type) {
1565         case LM_CB_NEED_E:
1566                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1567                 return;
1568
1569         case LM_CB_NEED_D:
1570                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1571                 return;
1572
1573         case LM_CB_NEED_S:
1574                 blocking_cb(sdp, data, LM_ST_SHARED);
1575                 return;
1576
1577         case LM_CB_ASYNC: {
1578                 struct lm_async_cb *async = data;
1579                 struct gfs2_glock *gl;
1580
1581                 gl = gfs2_glock_find(sdp, &async->lc_name);
1582                 if (gfs2_assert_warn(sdp, gl))
1583                         return;
1584                 if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
1585                         gl->gl_req_bh(gl, async->lc_ret);
1586                 gfs2_glock_put(gl);
1587                 return;
1588         }
1589
1590         case LM_CB_NEED_RECOVERY:
1591                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1592                 if (sdp->sd_recoverd_process)
1593                         wake_up_process(sdp->sd_recoverd_process);
1594                 return;
1595
1596         case LM_CB_DROPLOCKS:
1597                 gfs2_gl_hash_clear(sdp, NO_WAIT);
1598                 gfs2_quota_scan(sdp);
1599                 return;
1600
1601         default:
1602                 gfs2_assert_warn(sdp, 0);
1603                 return;
1604         }
1605 }
1606
1607 /**
1608  * demote_ok - Check to see if it's ok to unlock a glock
1609  * @gl: the glock
1610  *
1611  * Returns: 1 if it's ok
1612  */
1613
1614 static int demote_ok(struct gfs2_glock *gl)
1615 {
1616         const struct gfs2_glock_operations *glops = gl->gl_ops;
1617         int demote = 1;
1618
1619         if (test_bit(GLF_STICKY, &gl->gl_flags))
1620                 demote = 0;
1621         else if (glops->go_demote_ok)
1622                 demote = glops->go_demote_ok(gl);
1623
1624         return demote;
1625 }
1626
1627 /**
1628  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1629  * @gl: the glock
1630  *
1631  */
1632
1633 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1634 {
1635         struct gfs2_sbd *sdp = gl->gl_sbd;
1636
1637         spin_lock(&sdp->sd_reclaim_lock);
1638         if (list_empty(&gl->gl_reclaim)) {
1639                 gfs2_glock_hold(gl);
1640                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1641                 atomic_inc(&sdp->sd_reclaim_count);
1642         }
1643         spin_unlock(&sdp->sd_reclaim_lock);
1644
1645         wake_up(&sdp->sd_reclaim_wq);
1646 }
1647
1648 /**
1649  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1650  * @sdp: the filesystem
1651  *
1652  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1653  * different glock and we notice that there are a lot of glocks in the
1654  * reclaim list.
1655  *
1656  */
1657
1658 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1659 {
1660         struct gfs2_glock *gl;
1661
1662         spin_lock(&sdp->sd_reclaim_lock);
1663         if (list_empty(&sdp->sd_reclaim_list)) {
1664                 spin_unlock(&sdp->sd_reclaim_lock);
1665                 return;
1666         }
1667         gl = list_entry(sdp->sd_reclaim_list.next,
1668                         struct gfs2_glock, gl_reclaim);
1669         list_del_init(&gl->gl_reclaim);
1670         spin_unlock(&sdp->sd_reclaim_lock);
1671
1672         atomic_dec(&sdp->sd_reclaim_count);
1673         atomic_inc(&sdp->sd_reclaimed);
1674
1675         if (gfs2_glmutex_trylock(gl)) {
1676                 if (list_empty(&gl->gl_holders) &&
1677                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1678                         handle_callback(gl, LM_ST_UNLOCKED);
1679                 gfs2_glmutex_unlock(gl);
1680         }
1681
1682         gfs2_glock_put(gl);
1683 }
1684
1685 /**
1686  * examine_bucket - Call a function for glock in a hash bucket
1687  * @examiner: the function
1688  * @sdp: the filesystem
1689  * @bucket: the bucket
1690  *
1691  * Returns: 1 if the bucket has entries
1692  */
1693
1694 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1695                           unsigned int hash)
1696 {
1697         struct gfs2_glock *gl, *prev = NULL;
1698         int has_entries = 0;
1699         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1700
1701         read_lock(gl_lock_addr(hash));
1702         /* Can't use hlist_for_each_entry - don't want prefetch here */
1703         if (hlist_empty(head))
1704                 goto out;
1705         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1706         while(1) {
1707                 if (gl->gl_sbd == sdp) {
1708                         gfs2_glock_hold(gl);
1709                         read_unlock(gl_lock_addr(hash));
1710                         if (prev)
1711                                 gfs2_glock_put(prev);
1712                         prev = gl;
1713                         examiner(gl);
1714                         has_entries = 1;
1715                         read_lock(gl_lock_addr(hash));
1716                 }
1717                 if (gl->gl_list.next == NULL)
1718                         break;
1719                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1720         }
1721 out:
1722         read_unlock(gl_lock_addr(hash));
1723         if (prev)
1724                 gfs2_glock_put(prev);
1725         return has_entries;
1726 }
1727
1728 /**
1729  * scan_glock - look at a glock and see if we can reclaim it
1730  * @gl: the glock to look at
1731  *
1732  */
1733
1734 static void scan_glock(struct gfs2_glock *gl)
1735 {
1736         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1737                 return;
1738
1739         if (gfs2_glmutex_trylock(gl)) {
1740                 if (list_empty(&gl->gl_holders) &&
1741                     gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1742                         goto out_schedule;
1743                 gfs2_glmutex_unlock(gl);
1744         }
1745         return;
1746
1747 out_schedule:
1748         gfs2_glmutex_unlock(gl);
1749         gfs2_glock_schedule_for_reclaim(gl);
1750 }
1751
1752 /**
1753  * gfs2_scand_internal - Look for glocks and inodes to toss from memory
1754  * @sdp: the filesystem
1755  *
1756  */
1757
1758 void gfs2_scand_internal(struct gfs2_sbd *sdp)
1759 {
1760         unsigned int x;
1761
1762         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1763                 examine_bucket(scan_glock, sdp, x);
1764 }
1765
1766 /**
1767  * clear_glock - look at a glock and see if we can free it from glock cache
1768  * @gl: the glock to look at
1769  *
1770  */
1771
1772 static void clear_glock(struct gfs2_glock *gl)
1773 {
1774         struct gfs2_sbd *sdp = gl->gl_sbd;
1775         int released;
1776
1777         spin_lock(&sdp->sd_reclaim_lock);
1778         if (!list_empty(&gl->gl_reclaim)) {
1779                 list_del_init(&gl->gl_reclaim);
1780                 atomic_dec(&sdp->sd_reclaim_count);
1781                 spin_unlock(&sdp->sd_reclaim_lock);
1782                 released = gfs2_glock_put(gl);
1783                 gfs2_assert(sdp, !released);
1784         } else {
1785                 spin_unlock(&sdp->sd_reclaim_lock);
1786         }
1787
1788         if (gfs2_glmutex_trylock(gl)) {
1789                 if (list_empty(&gl->gl_holders) &&
1790                     gl->gl_state != LM_ST_UNLOCKED)
1791                         handle_callback(gl, LM_ST_UNLOCKED);
1792                 gfs2_glmutex_unlock(gl);
1793         }
1794 }
1795
1796 /**
1797  * gfs2_gl_hash_clear - Empty out the glock hash table
1798  * @sdp: the filesystem
1799  * @wait: wait until it's all gone
1800  *
1801  * Called when unmounting the filesystem, or when inter-node lock manager
1802  * requests DROPLOCKS because it is running out of capacity.
1803  */
1804
1805 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
1806 {
1807         unsigned long t;
1808         unsigned int x;
1809         int cont;
1810
1811         t = jiffies;
1812
1813         for (;;) {
1814                 cont = 0;
1815                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1816                         if (examine_bucket(clear_glock, sdp, x))
1817                                 cont = 1;
1818                 }
1819
1820                 if (!wait || !cont)
1821                         break;
1822
1823                 if (time_after_eq(jiffies,
1824                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1825                         fs_warn(sdp, "Unmount seems to be stalled. "
1826                                      "Dumping lock state...\n");
1827                         gfs2_dump_lockstate(sdp);
1828                         t = jiffies;
1829                 }
1830
1831                 invalidate_inodes(sdp->sd_vfs);
1832                 msleep(10);
1833         }
1834 }
1835
1836 /*
1837  *  Diagnostic routines to help debug distributed deadlock
1838  */
1839
1840 /**
1841  * dump_holder - print information about a glock holder
1842  * @str: a string naming the type of holder
1843  * @gh: the glock holder
1844  *
1845  * Returns: 0 on success, -ENOBUFS when we run out of space
1846  */
1847
1848 static int dump_holder(char *str, struct gfs2_holder *gh)
1849 {
1850         unsigned int x;
1851         int error = -ENOBUFS;
1852
1853         printk(KERN_INFO "  %s\n", str);
1854         printk(KERN_INFO "    owner = %ld\n",
1855                    (gh->gh_owner) ? (long)gh->gh_owner->pid : -1);
1856         printk(KERN_INFO "    gh_state = %u\n", gh->gh_state);
1857         printk(KERN_INFO "    gh_flags =");
1858         for (x = 0; x < 32; x++)
1859                 if (gh->gh_flags & (1 << x))
1860                         printk(" %u", x);
1861         printk(" \n");
1862         printk(KERN_INFO "    error = %d\n", gh->gh_error);
1863         printk(KERN_INFO "    gh_iflags =");
1864         for (x = 0; x < 32; x++)
1865                 if (test_bit(x, &gh->gh_iflags))
1866                         printk(" %u", x);
1867         printk(" \n");
1868         print_symbol(KERN_INFO "    initialized at: %s\n", gh->gh_ip);
1869
1870         error = 0;
1871
1872         return error;
1873 }
1874
1875 /**
1876  * dump_inode - print information about an inode
1877  * @ip: the inode
1878  *
1879  * Returns: 0 on success, -ENOBUFS when we run out of space
1880  */
1881
1882 static int dump_inode(struct gfs2_inode *ip)
1883 {
1884         unsigned int x;
1885         int error = -ENOBUFS;
1886
1887         printk(KERN_INFO "  Inode:\n");
1888         printk(KERN_INFO "    num = %llu %llu\n",
1889                     (unsigned long long)ip->i_num.no_formal_ino,
1890                     (unsigned long long)ip->i_num.no_addr);
1891         printk(KERN_INFO "    type = %u\n", IF2DT(ip->i_inode.i_mode));
1892         printk(KERN_INFO "    i_flags =");
1893         for (x = 0; x < 32; x++)
1894                 if (test_bit(x, &ip->i_flags))
1895                         printk(" %u", x);
1896         printk(" \n");
1897
1898         error = 0;
1899
1900         return error;
1901 }
1902
1903 /**
1904  * dump_glock - print information about a glock
1905  * @gl: the glock
1906  * @count: where we are in the buffer
1907  *
1908  * Returns: 0 on success, -ENOBUFS when we run out of space
1909  */
1910
1911 static int dump_glock(struct gfs2_glock *gl)
1912 {
1913         struct gfs2_holder *gh;
1914         unsigned int x;
1915         int error = -ENOBUFS;
1916
1917         spin_lock(&gl->gl_spin);
1918
1919         printk(KERN_INFO "Glock 0x%p (%u, %llu)\n", gl, gl->gl_name.ln_type,
1920                (unsigned long long)gl->gl_name.ln_number);
1921         printk(KERN_INFO "  gl_flags =");
1922         for (x = 0; x < 32; x++) {
1923                 if (test_bit(x, &gl->gl_flags))
1924                         printk(" %u", x);
1925         }
1926         printk(" \n");
1927         printk(KERN_INFO "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
1928         printk(KERN_INFO "  gl_state = %u\n", gl->gl_state);
1929         printk(KERN_INFO "  gl_owner = %s\n", gl->gl_owner->comm);
1930         print_symbol(KERN_INFO "  gl_ip = %s\n", gl->gl_ip);
1931         printk(KERN_INFO "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
1932         printk(KERN_INFO "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
1933         printk(KERN_INFO "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
1934         printk(KERN_INFO "  object = %s\n", (gl->gl_object) ? "yes" : "no");
1935         printk(KERN_INFO "  le = %s\n",
1936                    (list_empty(&gl->gl_le.le_list)) ? "no" : "yes");
1937         printk(KERN_INFO "  reclaim = %s\n",
1938                     (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
1939         if (gl->gl_aspace)
1940                 printk(KERN_INFO "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
1941                        gl->gl_aspace->i_mapping->nrpages);
1942         else
1943                 printk(KERN_INFO "  aspace = no\n");
1944         printk(KERN_INFO "  ail = %d\n", atomic_read(&gl->gl_ail_count));
1945         if (gl->gl_req_gh) {
1946                 error = dump_holder("Request", gl->gl_req_gh);
1947                 if (error)
1948                         goto out;
1949         }
1950         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1951                 error = dump_holder("Holder", gh);
1952                 if (error)
1953                         goto out;
1954         }
1955         list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
1956                 error = dump_holder("Waiter1", gh);
1957                 if (error)
1958                         goto out;
1959         }
1960         list_for_each_entry(gh, &gl->gl_waiters2, gh_list) {
1961                 error = dump_holder("Waiter2", gh);
1962                 if (error)
1963                         goto out;
1964         }
1965         list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
1966                 error = dump_holder("Waiter3", gh);
1967                 if (error)
1968                         goto out;
1969         }
1970         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
1971                 if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
1972                     list_empty(&gl->gl_holders)) {
1973                         error = dump_inode(gl->gl_object);
1974                         if (error)
1975                                 goto out;
1976                 } else {
1977                         error = -ENOBUFS;
1978                         printk(KERN_INFO "  Inode: busy\n");
1979                 }
1980         }
1981
1982         error = 0;
1983
1984 out:
1985         spin_unlock(&gl->gl_spin);
1986         return error;
1987 }
1988
1989 /**
1990  * gfs2_dump_lockstate - print out the current lockstate
1991  * @sdp: the filesystem
1992  * @ub: the buffer to copy the information into
1993  *
1994  * If @ub is NULL, dump the lockstate to the console.
1995  *
1996  */
1997
1998 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1999 {
2000         struct gfs2_glock *gl;
2001         struct hlist_node *h;
2002         unsigned int x;
2003         int error = 0;
2004
2005         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
2006
2007                 read_lock(gl_lock_addr(x));
2008
2009                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
2010                         if (gl->gl_sbd != sdp)
2011                                 continue;
2012
2013                         error = dump_glock(gl);
2014                         if (error)
2015                                 break;
2016                 }
2017
2018                 read_unlock(gl_lock_addr(x));
2019
2020                 if (error)
2021                         break;
2022         }
2023
2024
2025         return error;
2026 }
2027
2028 int __init gfs2_glock_init(void)
2029 {
2030         unsigned i;
2031         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
2032                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
2033         }
2034 #ifdef GL_HASH_LOCK_SZ
2035         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
2036                 rwlock_init(&gl_hash_locks[i]);
2037         }
2038 #endif
2039         return 0;
2040 }
2041