Merge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw
[pandora-kernel.git] / fs / gfs2 / glock.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/buffer_head.h>
14 #include <linux/delay.h>
15 #include <linux/sort.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <asm/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29
30 #include "gfs2.h"
31 #include "incore.h"
32 #include "glock.h"
33 #include "glops.h"
34 #include "inode.h"
35 #include "lops.h"
36 #include "meta_io.h"
37 #include "quota.h"
38 #include "super.h"
39 #include "util.h"
40 #include "bmap.h"
41 #define CREATE_TRACE_POINTS
42 #include "trace_gfs2.h"
43
44 struct gfs2_gl_hash_bucket {
45         struct hlist_head hb_list;
46 };
47
48 struct gfs2_glock_iter {
49         int hash;                       /* hash bucket index         */
50         struct gfs2_sbd *sdp;           /* incore superblock         */
51         struct gfs2_glock *gl;          /* current glock struct      */
52         char string[512];               /* scratch space             */
53 };
54
55 typedef void (*glock_examiner) (struct gfs2_glock * gl);
56
57 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
58 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
59 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
60 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
61
62 static struct dentry *gfs2_root;
63 static struct workqueue_struct *glock_workqueue;
64 struct workqueue_struct *gfs2_delete_workqueue;
65 static LIST_HEAD(lru_list);
66 static atomic_t lru_count = ATOMIC_INIT(0);
67 static DEFINE_SPINLOCK(lru_lock);
68
69 #define GFS2_GL_HASH_SHIFT      15
70 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
71 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
72
73 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
74 static struct dentry *gfs2_root;
75
76 /*
77  * Despite what you might think, the numbers below are not arbitrary :-)
78  * They are taken from the ipv4 routing hash code, which is well tested
79  * and thus should be nearly optimal. Later on we might tweek the numbers
80  * but for now this should be fine.
81  *
82  * The reason for putting the locks in a separate array from the list heads
83  * is that we can have fewer locks than list heads and save memory. We use
84  * the same hash function for both, but with a different hash mask.
85  */
86 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
87         defined(CONFIG_PROVE_LOCKING)
88
89 #ifdef CONFIG_LOCKDEP
90 # define GL_HASH_LOCK_SZ        256
91 #else
92 # if NR_CPUS >= 32
93 #  define GL_HASH_LOCK_SZ       4096
94 # elif NR_CPUS >= 16
95 #  define GL_HASH_LOCK_SZ       2048
96 # elif NR_CPUS >= 8
97 #  define GL_HASH_LOCK_SZ       1024
98 # elif NR_CPUS >= 4
99 #  define GL_HASH_LOCK_SZ       512
100 # else
101 #  define GL_HASH_LOCK_SZ       256
102 # endif
103 #endif
104
105 /* We never want more locks than chains */
106 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
107 # undef GL_HASH_LOCK_SZ
108 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
109 #endif
110
111 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
112
113 static inline rwlock_t *gl_lock_addr(unsigned int x)
114 {
115         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
116 }
117 #else /* not SMP, so no spinlocks required */
118 static inline rwlock_t *gl_lock_addr(unsigned int x)
119 {
120         return NULL;
121 }
122 #endif
123
124 /**
125  * gl_hash() - Turn glock number into hash bucket number
126  * @lock: The glock number
127  *
128  * Returns: The number of the corresponding hash bucket
129  */
130
131 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
132                             const struct lm_lockname *name)
133 {
134         unsigned int h;
135
136         h = jhash(&name->ln_number, sizeof(u64), 0);
137         h = jhash(&name->ln_type, sizeof(unsigned int), h);
138         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
139         h &= GFS2_GL_HASH_MASK;
140
141         return h;
142 }
143
144 /**
145  * glock_free() - Perform a few checks and then release struct gfs2_glock
146  * @gl: The glock to release
147  *
148  * Also calls lock module to release its internal structure for this glock.
149  *
150  */
151
152 static void glock_free(struct gfs2_glock *gl)
153 {
154         struct gfs2_sbd *sdp = gl->gl_sbd;
155         struct address_space *mapping = gfs2_glock2aspace(gl);
156         struct kmem_cache *cachep = gfs2_glock_cachep;
157
158         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
159         trace_gfs2_glock_put(gl);
160         if (mapping)
161                 cachep = gfs2_glock_aspace_cachep;
162         sdp->sd_lockstruct.ls_ops->lm_put_lock(cachep, gl);
163 }
164
165 /**
166  * gfs2_glock_hold() - increment reference count on glock
167  * @gl: The glock to hold
168  *
169  */
170
171 void gfs2_glock_hold(struct gfs2_glock *gl)
172 {
173         GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
174         atomic_inc(&gl->gl_ref);
175 }
176
177 /**
178  * demote_ok - Check to see if it's ok to unlock a glock
179  * @gl: the glock
180  *
181  * Returns: 1 if it's ok
182  */
183
184 static int demote_ok(const struct gfs2_glock *gl)
185 {
186         const struct gfs2_glock_operations *glops = gl->gl_ops;
187
188         if (gl->gl_state == LM_ST_UNLOCKED)
189                 return 0;
190         if (!list_empty(&gl->gl_holders))
191                 return 0;
192         if (glops->go_demote_ok)
193                 return glops->go_demote_ok(gl);
194         return 1;
195 }
196
197 /**
198  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
199  * @gl: the glock
200  *
201  */
202
203 static void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
204 {
205         int may_reclaim;
206         may_reclaim = (demote_ok(gl) &&
207                        (atomic_read(&gl->gl_ref) == 1 ||
208                         (gl->gl_name.ln_type == LM_TYPE_INODE &&
209                          atomic_read(&gl->gl_ref) <= 2)));
210         spin_lock(&lru_lock);
211         if (list_empty(&gl->gl_lru) && may_reclaim) {
212                 list_add_tail(&gl->gl_lru, &lru_list);
213                 atomic_inc(&lru_count);
214         }
215         spin_unlock(&lru_lock);
216 }
217
218 /**
219  * gfs2_glock_put_nolock() - Decrement reference count on glock
220  * @gl: The glock to put
221  *
222  * This function should only be used if the caller has its own reference
223  * to the glock, in addition to the one it is dropping.
224  */
225
226 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
227 {
228         if (atomic_dec_and_test(&gl->gl_ref))
229                 GLOCK_BUG_ON(gl, 1);
230         gfs2_glock_schedule_for_reclaim(gl);
231 }
232
233 /**
234  * gfs2_glock_put() - Decrement reference count on glock
235  * @gl: The glock to put
236  *
237  */
238
239 int gfs2_glock_put(struct gfs2_glock *gl)
240 {
241         int rv = 0;
242
243         write_lock(gl_lock_addr(gl->gl_hash));
244         if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
245                 hlist_del(&gl->gl_list);
246                 if (!list_empty(&gl->gl_lru)) {
247                         list_del_init(&gl->gl_lru);
248                         atomic_dec(&lru_count);
249                 }
250                 spin_unlock(&lru_lock);
251                 write_unlock(gl_lock_addr(gl->gl_hash));
252                 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
253                 glock_free(gl);
254                 rv = 1;
255                 goto out;
256         }
257         spin_lock(&gl->gl_spin);
258         gfs2_glock_schedule_for_reclaim(gl);
259         spin_unlock(&gl->gl_spin);
260         write_unlock(gl_lock_addr(gl->gl_hash));
261 out:
262         return rv;
263 }
264
265 /**
266  * search_bucket() - Find struct gfs2_glock by lock number
267  * @bucket: the bucket to search
268  * @name: The lock name
269  *
270  * Returns: NULL, or the struct gfs2_glock with the requested number
271  */
272
273 static struct gfs2_glock *search_bucket(unsigned int hash,
274                                         const struct gfs2_sbd *sdp,
275                                         const struct lm_lockname *name)
276 {
277         struct gfs2_glock *gl;
278         struct hlist_node *h;
279
280         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
281                 if (!lm_name_equal(&gl->gl_name, name))
282                         continue;
283                 if (gl->gl_sbd != sdp)
284                         continue;
285
286                 atomic_inc(&gl->gl_ref);
287
288                 return gl;
289         }
290
291         return NULL;
292 }
293
294 /**
295  * may_grant - check if its ok to grant a new lock
296  * @gl: The glock
297  * @gh: The lock request which we wish to grant
298  *
299  * Returns: true if its ok to grant the lock
300  */
301
302 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
303 {
304         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
305         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
306              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
307                 return 0;
308         if (gl->gl_state == gh->gh_state)
309                 return 1;
310         if (gh->gh_flags & GL_EXACT)
311                 return 0;
312         if (gl->gl_state == LM_ST_EXCLUSIVE) {
313                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
314                         return 1;
315                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
316                         return 1;
317         }
318         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
319                 return 1;
320         return 0;
321 }
322
323 static void gfs2_holder_wake(struct gfs2_holder *gh)
324 {
325         clear_bit(HIF_WAIT, &gh->gh_iflags);
326         smp_mb__after_clear_bit();
327         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
328 }
329
330 /**
331  * do_promote - promote as many requests as possible on the current queue
332  * @gl: The glock
333  * 
334  * Returns: 1 if there is a blocked holder at the head of the list, or 2
335  *          if a type specific operation is underway.
336  */
337
338 static int do_promote(struct gfs2_glock *gl)
339 __releases(&gl->gl_spin)
340 __acquires(&gl->gl_spin)
341 {
342         const struct gfs2_glock_operations *glops = gl->gl_ops;
343         struct gfs2_holder *gh, *tmp;
344         int ret;
345
346 restart:
347         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
348                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
349                         continue;
350                 if (may_grant(gl, gh)) {
351                         if (gh->gh_list.prev == &gl->gl_holders &&
352                             glops->go_lock) {
353                                 spin_unlock(&gl->gl_spin);
354                                 /* FIXME: eliminate this eventually */
355                                 ret = glops->go_lock(gh);
356                                 spin_lock(&gl->gl_spin);
357                                 if (ret) {
358                                         if (ret == 1)
359                                                 return 2;
360                                         gh->gh_error = ret;
361                                         list_del_init(&gh->gh_list);
362                                         trace_gfs2_glock_queue(gh, 0);
363                                         gfs2_holder_wake(gh);
364                                         goto restart;
365                                 }
366                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
367                                 trace_gfs2_promote(gh, 1);
368                                 gfs2_holder_wake(gh);
369                                 goto restart;
370                         }
371                         set_bit(HIF_HOLDER, &gh->gh_iflags);
372                         trace_gfs2_promote(gh, 0);
373                         gfs2_holder_wake(gh);
374                         continue;
375                 }
376                 if (gh->gh_list.prev == &gl->gl_holders)
377                         return 1;
378                 break;
379         }
380         return 0;
381 }
382
383 /**
384  * do_error - Something unexpected has happened during a lock request
385  *
386  */
387
388 static inline void do_error(struct gfs2_glock *gl, const int ret)
389 {
390         struct gfs2_holder *gh, *tmp;
391
392         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
393                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
394                         continue;
395                 if (ret & LM_OUT_ERROR)
396                         gh->gh_error = -EIO;
397                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
398                         gh->gh_error = GLR_TRYFAILED;
399                 else
400                         continue;
401                 list_del_init(&gh->gh_list);
402                 trace_gfs2_glock_queue(gh, 0);
403                 gfs2_holder_wake(gh);
404         }
405 }
406
407 /**
408  * find_first_waiter - find the first gh that's waiting for the glock
409  * @gl: the glock
410  */
411
412 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
413 {
414         struct gfs2_holder *gh;
415
416         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
417                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
418                         return gh;
419         }
420         return NULL;
421 }
422
423 /**
424  * state_change - record that the glock is now in a different state
425  * @gl: the glock
426  * @new_state the new state
427  *
428  */
429
430 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
431 {
432         int held1, held2;
433
434         held1 = (gl->gl_state != LM_ST_UNLOCKED);
435         held2 = (new_state != LM_ST_UNLOCKED);
436
437         if (held1 != held2) {
438                 if (held2)
439                         gfs2_glock_hold(gl);
440                 else
441                         gfs2_glock_put_nolock(gl);
442         }
443
444         gl->gl_state = new_state;
445         gl->gl_tchange = jiffies;
446 }
447
448 static void gfs2_demote_wake(struct gfs2_glock *gl)
449 {
450         gl->gl_demote_state = LM_ST_EXCLUSIVE;
451         clear_bit(GLF_DEMOTE, &gl->gl_flags);
452         smp_mb__after_clear_bit();
453         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
454 }
455
456 /**
457  * finish_xmote - The DLM has replied to one of our lock requests
458  * @gl: The glock
459  * @ret: The status from the DLM
460  *
461  */
462
463 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
464 {
465         const struct gfs2_glock_operations *glops = gl->gl_ops;
466         struct gfs2_holder *gh;
467         unsigned state = ret & LM_OUT_ST_MASK;
468         int rv;
469
470         spin_lock(&gl->gl_spin);
471         trace_gfs2_glock_state_change(gl, state);
472         state_change(gl, state);
473         gh = find_first_waiter(gl);
474
475         /* Demote to UN request arrived during demote to SH or DF */
476         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
477             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
478                 gl->gl_target = LM_ST_UNLOCKED;
479
480         /* Check for state != intended state */
481         if (unlikely(state != gl->gl_target)) {
482                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
483                         /* move to back of queue and try next entry */
484                         if (ret & LM_OUT_CANCELED) {
485                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
486                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
487                                 gh = find_first_waiter(gl);
488                                 gl->gl_target = gh->gh_state;
489                                 goto retry;
490                         }
491                         /* Some error or failed "try lock" - report it */
492                         if ((ret & LM_OUT_ERROR) ||
493                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
494                                 gl->gl_target = gl->gl_state;
495                                 do_error(gl, ret);
496                                 goto out;
497                         }
498                 }
499                 switch(state) {
500                 /* Unlocked due to conversion deadlock, try again */
501                 case LM_ST_UNLOCKED:
502 retry:
503                         do_xmote(gl, gh, gl->gl_target);
504                         break;
505                 /* Conversion fails, unlock and try again */
506                 case LM_ST_SHARED:
507                 case LM_ST_DEFERRED:
508                         do_xmote(gl, gh, LM_ST_UNLOCKED);
509                         break;
510                 default: /* Everything else */
511                         printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
512                         GLOCK_BUG_ON(gl, 1);
513                 }
514                 spin_unlock(&gl->gl_spin);
515                 return;
516         }
517
518         /* Fast path - we got what we asked for */
519         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
520                 gfs2_demote_wake(gl);
521         if (state != LM_ST_UNLOCKED) {
522                 if (glops->go_xmote_bh) {
523                         spin_unlock(&gl->gl_spin);
524                         rv = glops->go_xmote_bh(gl, gh);
525                         spin_lock(&gl->gl_spin);
526                         if (rv) {
527                                 do_error(gl, rv);
528                                 goto out;
529                         }
530                 }
531                 rv = do_promote(gl);
532                 if (rv == 2)
533                         goto out_locked;
534         }
535 out:
536         clear_bit(GLF_LOCK, &gl->gl_flags);
537 out_locked:
538         spin_unlock(&gl->gl_spin);
539 }
540
541 static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
542                                  unsigned int req_state,
543                                  unsigned int flags)
544 {
545         int ret = LM_OUT_ERROR;
546
547         if (!sdp->sd_lockstruct.ls_ops->lm_lock)
548                 return req_state == LM_ST_UNLOCKED ? 0 : req_state;
549
550         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
551                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock,
552                                                          req_state, flags);
553         return ret;
554 }
555
556 /**
557  * do_xmote - Calls the DLM to change the state of a lock
558  * @gl: The lock state
559  * @gh: The holder (only for promotes)
560  * @target: The target lock state
561  *
562  */
563
564 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
565 __releases(&gl->gl_spin)
566 __acquires(&gl->gl_spin)
567 {
568         const struct gfs2_glock_operations *glops = gl->gl_ops;
569         struct gfs2_sbd *sdp = gl->gl_sbd;
570         unsigned int lck_flags = gh ? gh->gh_flags : 0;
571         int ret;
572
573         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
574                       LM_FLAG_PRIORITY);
575         BUG_ON(gl->gl_state == target);
576         BUG_ON(gl->gl_state == gl->gl_target);
577         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
578             glops->go_inval) {
579                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
580                 do_error(gl, 0); /* Fail queued try locks */
581         }
582         spin_unlock(&gl->gl_spin);
583         if (glops->go_xmote_th)
584                 glops->go_xmote_th(gl);
585         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
586                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
587         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
588
589         gfs2_glock_hold(gl);
590         if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
591             gl->gl_state == LM_ST_DEFERRED) &&
592             !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
593                 lck_flags |= LM_FLAG_TRY_1CB;
594         ret = gfs2_lm_lock(sdp, gl, target, lck_flags);
595
596         if (!(ret & LM_OUT_ASYNC)) {
597                 finish_xmote(gl, ret);
598                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
599                         gfs2_glock_put(gl);
600         } else {
601                 GLOCK_BUG_ON(gl, ret != LM_OUT_ASYNC);
602         }
603         spin_lock(&gl->gl_spin);
604 }
605
606 /**
607  * find_first_holder - find the first "holder" gh
608  * @gl: the glock
609  */
610
611 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
612 {
613         struct gfs2_holder *gh;
614
615         if (!list_empty(&gl->gl_holders)) {
616                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
617                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
618                         return gh;
619         }
620         return NULL;
621 }
622
623 /**
624  * run_queue - do all outstanding tasks related to a glock
625  * @gl: The glock in question
626  * @nonblock: True if we must not block in run_queue
627  *
628  */
629
630 static void run_queue(struct gfs2_glock *gl, const int nonblock)
631 __releases(&gl->gl_spin)
632 __acquires(&gl->gl_spin)
633 {
634         struct gfs2_holder *gh = NULL;
635         int ret;
636
637         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
638                 return;
639
640         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
641
642         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
643             gl->gl_demote_state != gl->gl_state) {
644                 if (find_first_holder(gl))
645                         goto out_unlock;
646                 if (nonblock)
647                         goto out_sched;
648                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
649                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
650                 gl->gl_target = gl->gl_demote_state;
651         } else {
652                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
653                         gfs2_demote_wake(gl);
654                 ret = do_promote(gl);
655                 if (ret == 0)
656                         goto out_unlock;
657                 if (ret == 2)
658                         goto out;
659                 gh = find_first_waiter(gl);
660                 gl->gl_target = gh->gh_state;
661                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
662                         do_error(gl, 0); /* Fail queued try locks */
663         }
664         do_xmote(gl, gh, gl->gl_target);
665 out:
666         return;
667
668 out_sched:
669         clear_bit(GLF_LOCK, &gl->gl_flags);
670         smp_mb__after_clear_bit();
671         gfs2_glock_hold(gl);
672         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
673                 gfs2_glock_put_nolock(gl);
674         return;
675
676 out_unlock:
677         clear_bit(GLF_LOCK, &gl->gl_flags);
678         smp_mb__after_clear_bit();
679         return;
680 }
681
682 static void delete_work_func(struct work_struct *work)
683 {
684         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
685         struct gfs2_sbd *sdp = gl->gl_sbd;
686         struct gfs2_inode *ip = NULL;
687         struct inode *inode;
688         u64 no_addr = 0;
689
690         spin_lock(&gl->gl_spin);
691         ip = (struct gfs2_inode *)gl->gl_object;
692         if (ip)
693                 no_addr = ip->i_no_addr;
694         spin_unlock(&gl->gl_spin);
695         if (ip) {
696                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr);
697                 if (inode) {
698                         d_prune_aliases(inode);
699                         iput(inode);
700                 }
701         }
702         gfs2_glock_put(gl);
703 }
704
705 static void glock_work_func(struct work_struct *work)
706 {
707         unsigned long delay = 0;
708         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
709         int drop_ref = 0;
710
711         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
712                 finish_xmote(gl, gl->gl_reply);
713                 drop_ref = 1;
714         }
715         spin_lock(&gl->gl_spin);
716         if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
717             gl->gl_state != LM_ST_UNLOCKED &&
718             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
719                 unsigned long holdtime, now = jiffies;
720                 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
721                 if (time_before(now, holdtime))
722                         delay = holdtime - now;
723                 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
724         }
725         run_queue(gl, 0);
726         spin_unlock(&gl->gl_spin);
727         if (!delay ||
728             queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
729                 gfs2_glock_put(gl);
730         if (drop_ref)
731                 gfs2_glock_put(gl);
732 }
733
734 /**
735  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
736  * @sdp: The GFS2 superblock
737  * @number: the lock number
738  * @glops: The glock_operations to use
739  * @create: If 0, don't create the glock if it doesn't exist
740  * @glp: the glock is returned here
741  *
742  * This does not lock a glock, just finds/creates structures for one.
743  *
744  * Returns: errno
745  */
746
747 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
748                    const struct gfs2_glock_operations *glops, int create,
749                    struct gfs2_glock **glp)
750 {
751         struct super_block *s = sdp->sd_vfs;
752         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
753         struct gfs2_glock *gl, *tmp;
754         unsigned int hash = gl_hash(sdp, &name);
755         struct address_space *mapping;
756
757         read_lock(gl_lock_addr(hash));
758         gl = search_bucket(hash, sdp, &name);
759         read_unlock(gl_lock_addr(hash));
760
761         *glp = gl;
762         if (gl)
763                 return 0;
764         if (!create)
765                 return -ENOENT;
766
767         if (glops->go_flags & GLOF_ASPACE)
768                 gl = kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_KERNEL);
769         else
770                 gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
771         if (!gl)
772                 return -ENOMEM;
773
774         atomic_inc(&sdp->sd_glock_disposal);
775         gl->gl_flags = 0;
776         gl->gl_name = name;
777         atomic_set(&gl->gl_ref, 1);
778         gl->gl_state = LM_ST_UNLOCKED;
779         gl->gl_target = LM_ST_UNLOCKED;
780         gl->gl_demote_state = LM_ST_EXCLUSIVE;
781         gl->gl_hash = hash;
782         gl->gl_ops = glops;
783         snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
784         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
785         gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
786         gl->gl_tchange = jiffies;
787         gl->gl_object = NULL;
788         gl->gl_sbd = sdp;
789         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
790         INIT_WORK(&gl->gl_delete, delete_work_func);
791
792         mapping = gfs2_glock2aspace(gl);
793         if (mapping) {
794                 mapping->a_ops = &gfs2_meta_aops;
795                 mapping->host = s->s_bdev->bd_inode;
796                 mapping->flags = 0;
797                 mapping_set_gfp_mask(mapping, GFP_NOFS);
798                 mapping->assoc_mapping = NULL;
799                 mapping->backing_dev_info = s->s_bdi;
800                 mapping->writeback_index = 0;
801         }
802
803         write_lock(gl_lock_addr(hash));
804         tmp = search_bucket(hash, sdp, &name);
805         if (tmp) {
806                 write_unlock(gl_lock_addr(hash));
807                 glock_free(gl);
808                 gl = tmp;
809         } else {
810                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
811                 write_unlock(gl_lock_addr(hash));
812         }
813
814         *glp = gl;
815
816         return 0;
817 }
818
819 /**
820  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
821  * @gl: the glock
822  * @state: the state we're requesting
823  * @flags: the modifier flags
824  * @gh: the holder structure
825  *
826  */
827
828 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
829                       struct gfs2_holder *gh)
830 {
831         INIT_LIST_HEAD(&gh->gh_list);
832         gh->gh_gl = gl;
833         gh->gh_ip = (unsigned long)__builtin_return_address(0);
834         gh->gh_owner_pid = get_pid(task_pid(current));
835         gh->gh_state = state;
836         gh->gh_flags = flags;
837         gh->gh_error = 0;
838         gh->gh_iflags = 0;
839         gfs2_glock_hold(gl);
840 }
841
842 /**
843  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
844  * @state: the state we're requesting
845  * @flags: the modifier flags
846  * @gh: the holder structure
847  *
848  * Don't mess with the glock.
849  *
850  */
851
852 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
853 {
854         gh->gh_state = state;
855         gh->gh_flags = flags;
856         gh->gh_iflags = 0;
857         gh->gh_ip = (unsigned long)__builtin_return_address(0);
858         if (gh->gh_owner_pid)
859                 put_pid(gh->gh_owner_pid);
860         gh->gh_owner_pid = get_pid(task_pid(current));
861 }
862
863 /**
864  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
865  * @gh: the holder structure
866  *
867  */
868
869 void gfs2_holder_uninit(struct gfs2_holder *gh)
870 {
871         put_pid(gh->gh_owner_pid);
872         gfs2_glock_put(gh->gh_gl);
873         gh->gh_gl = NULL;
874         gh->gh_ip = 0;
875 }
876
877 /**
878  * gfs2_glock_holder_wait
879  * @word: unused
880  *
881  * This function and gfs2_glock_demote_wait both show up in the WCHAN
882  * field. Thus I've separated these otherwise identical functions in
883  * order to be more informative to the user.
884  */
885
886 static int gfs2_glock_holder_wait(void *word)
887 {
888         schedule();
889         return 0;
890 }
891
892 static int gfs2_glock_demote_wait(void *word)
893 {
894         schedule();
895         return 0;
896 }
897
898 static void wait_on_holder(struct gfs2_holder *gh)
899 {
900         might_sleep();
901         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
902 }
903
904 static void wait_on_demote(struct gfs2_glock *gl)
905 {
906         might_sleep();
907         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
908 }
909
910 /**
911  * handle_callback - process a demote request
912  * @gl: the glock
913  * @state: the state the caller wants us to change to
914  *
915  * There are only two requests that we are going to see in actual
916  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
917  */
918
919 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
920                             unsigned long delay)
921 {
922         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
923
924         set_bit(bit, &gl->gl_flags);
925         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
926                 gl->gl_demote_state = state;
927                 gl->gl_demote_time = jiffies;
928         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
929                         gl->gl_demote_state != state) {
930                 gl->gl_demote_state = LM_ST_UNLOCKED;
931         }
932         if (gl->gl_ops->go_callback)
933                 gl->gl_ops->go_callback(gl);
934         trace_gfs2_demote_rq(gl);
935 }
936
937 /**
938  * gfs2_glock_wait - wait on a glock acquisition
939  * @gh: the glock holder
940  *
941  * Returns: 0 on success
942  */
943
944 int gfs2_glock_wait(struct gfs2_holder *gh)
945 {
946         wait_on_holder(gh);
947         return gh->gh_error;
948 }
949
950 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
951 {
952         va_list args;
953
954         va_start(args, fmt);
955         if (seq) {
956                 struct gfs2_glock_iter *gi = seq->private;
957                 vsprintf(gi->string, fmt, args);
958                 seq_printf(seq, gi->string);
959         } else {
960                 printk(KERN_ERR " ");
961                 vprintk(fmt, args);
962         }
963         va_end(args);
964 }
965
966 /**
967  * add_to_queue - Add a holder to the wait queue (but look for recursion)
968  * @gh: the holder structure to add
969  *
970  * Eventually we should move the recursive locking trap to a
971  * debugging option or something like that. This is the fast
972  * path and needs to have the minimum number of distractions.
973  * 
974  */
975
976 static inline void add_to_queue(struct gfs2_holder *gh)
977 __releases(&gl->gl_spin)
978 __acquires(&gl->gl_spin)
979 {
980         struct gfs2_glock *gl = gh->gh_gl;
981         struct gfs2_sbd *sdp = gl->gl_sbd;
982         struct list_head *insert_pt = NULL;
983         struct gfs2_holder *gh2;
984         int try_lock = 0;
985
986         BUG_ON(gh->gh_owner_pid == NULL);
987         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
988                 BUG();
989
990         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
991                 if (test_bit(GLF_LOCK, &gl->gl_flags))
992                         try_lock = 1;
993                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
994                         goto fail;
995         }
996
997         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
998                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
999                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
1000                         goto trap_recursive;
1001                 if (try_lock &&
1002                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
1003                     !may_grant(gl, gh)) {
1004 fail:
1005                         gh->gh_error = GLR_TRYFAILED;
1006                         gfs2_holder_wake(gh);
1007                         return;
1008                 }
1009                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1010                         continue;
1011                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1012                         insert_pt = &gh2->gh_list;
1013         }
1014         if (likely(insert_pt == NULL)) {
1015                 list_add_tail(&gh->gh_list, &gl->gl_holders);
1016                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1017                         goto do_cancel;
1018                 return;
1019         }
1020         trace_gfs2_glock_queue(gh, 1);
1021         list_add_tail(&gh->gh_list, insert_pt);
1022 do_cancel:
1023         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1024         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1025                 spin_unlock(&gl->gl_spin);
1026                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1027                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1028                 spin_lock(&gl->gl_spin);
1029         }
1030         return;
1031
1032 trap_recursive:
1033         print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1034         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1035         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1036                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1037         print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1038         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1039         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1040                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1041         __dump_glock(NULL, gl);
1042         BUG();
1043 }
1044
1045 /**
1046  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1047  * @gh: the holder structure
1048  *
1049  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1050  *
1051  * Returns: 0, GLR_TRYFAILED, or errno on failure
1052  */
1053
1054 int gfs2_glock_nq(struct gfs2_holder *gh)
1055 {
1056         struct gfs2_glock *gl = gh->gh_gl;
1057         struct gfs2_sbd *sdp = gl->gl_sbd;
1058         int error = 0;
1059
1060         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1061                 return -EIO;
1062
1063         spin_lock(&gl->gl_spin);
1064         add_to_queue(gh);
1065         run_queue(gl, 1);
1066         spin_unlock(&gl->gl_spin);
1067
1068         if (!(gh->gh_flags & GL_ASYNC))
1069                 error = gfs2_glock_wait(gh);
1070
1071         return error;
1072 }
1073
1074 /**
1075  * gfs2_glock_poll - poll to see if an async request has been completed
1076  * @gh: the holder
1077  *
1078  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1079  */
1080
1081 int gfs2_glock_poll(struct gfs2_holder *gh)
1082 {
1083         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1084 }
1085
1086 /**
1087  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1088  * @gh: the glock holder
1089  *
1090  */
1091
1092 void gfs2_glock_dq(struct gfs2_holder *gh)
1093 {
1094         struct gfs2_glock *gl = gh->gh_gl;
1095         const struct gfs2_glock_operations *glops = gl->gl_ops;
1096         unsigned delay = 0;
1097         int fast_path = 0;
1098
1099         spin_lock(&gl->gl_spin);
1100         if (gh->gh_flags & GL_NOCACHE)
1101                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1102
1103         list_del_init(&gh->gh_list);
1104         if (find_first_holder(gl) == NULL) {
1105                 if (glops->go_unlock) {
1106                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1107                         spin_unlock(&gl->gl_spin);
1108                         glops->go_unlock(gh);
1109                         spin_lock(&gl->gl_spin);
1110                         clear_bit(GLF_LOCK, &gl->gl_flags);
1111                 }
1112                 if (list_empty(&gl->gl_holders) &&
1113                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1114                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1115                         fast_path = 1;
1116         }
1117         trace_gfs2_glock_queue(gh, 0);
1118         spin_unlock(&gl->gl_spin);
1119         if (likely(fast_path))
1120                 return;
1121
1122         gfs2_glock_hold(gl);
1123         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1124             !test_bit(GLF_DEMOTE, &gl->gl_flags))
1125                 delay = gl->gl_ops->go_min_hold_time;
1126         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1127                 gfs2_glock_put(gl);
1128 }
1129
1130 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1131 {
1132         struct gfs2_glock *gl = gh->gh_gl;
1133         gfs2_glock_dq(gh);
1134         wait_on_demote(gl);
1135 }
1136
1137 /**
1138  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1139  * @gh: the holder structure
1140  *
1141  */
1142
1143 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1144 {
1145         gfs2_glock_dq(gh);
1146         gfs2_holder_uninit(gh);
1147 }
1148
1149 /**
1150  * gfs2_glock_nq_num - acquire a glock based on lock number
1151  * @sdp: the filesystem
1152  * @number: the lock number
1153  * @glops: the glock operations for the type of glock
1154  * @state: the state to acquire the glock in
1155  * @flags: modifier flags for the aquisition
1156  * @gh: the struct gfs2_holder
1157  *
1158  * Returns: errno
1159  */
1160
1161 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1162                       const struct gfs2_glock_operations *glops,
1163                       unsigned int state, int flags, struct gfs2_holder *gh)
1164 {
1165         struct gfs2_glock *gl;
1166         int error;
1167
1168         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1169         if (!error) {
1170                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1171                 gfs2_glock_put(gl);
1172         }
1173
1174         return error;
1175 }
1176
1177 /**
1178  * glock_compare - Compare two struct gfs2_glock structures for sorting
1179  * @arg_a: the first structure
1180  * @arg_b: the second structure
1181  *
1182  */
1183
1184 static int glock_compare(const void *arg_a, const void *arg_b)
1185 {
1186         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1187         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1188         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1189         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1190
1191         if (a->ln_number > b->ln_number)
1192                 return 1;
1193         if (a->ln_number < b->ln_number)
1194                 return -1;
1195         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1196         return 0;
1197 }
1198
1199 /**
1200  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1201  * @num_gh: the number of structures
1202  * @ghs: an array of struct gfs2_holder structures
1203  *
1204  * Returns: 0 on success (all glocks acquired),
1205  *          errno on failure (no glocks acquired)
1206  */
1207
1208 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1209                      struct gfs2_holder **p)
1210 {
1211         unsigned int x;
1212         int error = 0;
1213
1214         for (x = 0; x < num_gh; x++)
1215                 p[x] = &ghs[x];
1216
1217         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1218
1219         for (x = 0; x < num_gh; x++) {
1220                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1221
1222                 error = gfs2_glock_nq(p[x]);
1223                 if (error) {
1224                         while (x--)
1225                                 gfs2_glock_dq(p[x]);
1226                         break;
1227                 }
1228         }
1229
1230         return error;
1231 }
1232
1233 /**
1234  * gfs2_glock_nq_m - acquire multiple glocks
1235  * @num_gh: the number of structures
1236  * @ghs: an array of struct gfs2_holder structures
1237  *
1238  *
1239  * Returns: 0 on success (all glocks acquired),
1240  *          errno on failure (no glocks acquired)
1241  */
1242
1243 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1244 {
1245         struct gfs2_holder *tmp[4];
1246         struct gfs2_holder **pph = tmp;
1247         int error = 0;
1248
1249         switch(num_gh) {
1250         case 0:
1251                 return 0;
1252         case 1:
1253                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1254                 return gfs2_glock_nq(ghs);
1255         default:
1256                 if (num_gh <= 4)
1257                         break;
1258                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1259                 if (!pph)
1260                         return -ENOMEM;
1261         }
1262
1263         error = nq_m_sync(num_gh, ghs, pph);
1264
1265         if (pph != tmp)
1266                 kfree(pph);
1267
1268         return error;
1269 }
1270
1271 /**
1272  * gfs2_glock_dq_m - release multiple glocks
1273  * @num_gh: the number of structures
1274  * @ghs: an array of struct gfs2_holder structures
1275  *
1276  */
1277
1278 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1279 {
1280         unsigned int x;
1281
1282         for (x = 0; x < num_gh; x++)
1283                 gfs2_glock_dq(&ghs[x]);
1284 }
1285
1286 /**
1287  * gfs2_glock_dq_uninit_m - release multiple glocks
1288  * @num_gh: the number of structures
1289  * @ghs: an array of struct gfs2_holder structures
1290  *
1291  */
1292
1293 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1294 {
1295         unsigned int x;
1296
1297         for (x = 0; x < num_gh; x++)
1298                 gfs2_glock_dq_uninit(&ghs[x]);
1299 }
1300
1301 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1302 {
1303         unsigned long delay = 0;
1304         unsigned long holdtime;
1305         unsigned long now = jiffies;
1306
1307         gfs2_glock_hold(gl);
1308         holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1309         if (time_before(now, holdtime))
1310                 delay = holdtime - now;
1311         if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1312                 delay = gl->gl_ops->go_min_hold_time;
1313
1314         spin_lock(&gl->gl_spin);
1315         handle_callback(gl, state, delay);
1316         spin_unlock(&gl->gl_spin);
1317         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1318                 gfs2_glock_put(gl);
1319 }
1320
1321 /**
1322  * gfs2_glock_complete - Callback used by locking
1323  * @gl: Pointer to the glock
1324  * @ret: The return value from the dlm
1325  *
1326  */
1327
1328 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1329 {
1330         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1331         gl->gl_reply = ret;
1332         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1333                 struct gfs2_holder *gh;
1334                 spin_lock(&gl->gl_spin);
1335                 gh = find_first_waiter(gl);
1336                 if ((!(gh && (gh->gh_flags & LM_FLAG_NOEXP)) &&
1337                      (gl->gl_target != LM_ST_UNLOCKED)) ||
1338                     ((ret & ~LM_OUT_ST_MASK) != 0))
1339                         set_bit(GLF_FROZEN, &gl->gl_flags);
1340                 spin_unlock(&gl->gl_spin);
1341                 if (test_bit(GLF_FROZEN, &gl->gl_flags))
1342                         return;
1343         }
1344         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1345         gfs2_glock_hold(gl);
1346         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1347                 gfs2_glock_put(gl);
1348 }
1349
1350
1351 static int gfs2_shrink_glock_memory(int nr, gfp_t gfp_mask)
1352 {
1353         struct gfs2_glock *gl;
1354         int may_demote;
1355         int nr_skipped = 0;
1356         LIST_HEAD(skipped);
1357
1358         if (nr == 0)
1359                 goto out;
1360
1361         if (!(gfp_mask & __GFP_FS))
1362                 return -1;
1363
1364         spin_lock(&lru_lock);
1365         while(nr && !list_empty(&lru_list)) {
1366                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1367                 list_del_init(&gl->gl_lru);
1368                 atomic_dec(&lru_count);
1369
1370                 /* Test for being demotable */
1371                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1372                         gfs2_glock_hold(gl);
1373                         spin_unlock(&lru_lock);
1374                         spin_lock(&gl->gl_spin);
1375                         may_demote = demote_ok(gl);
1376                         if (may_demote) {
1377                                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1378                                 nr--;
1379                         }
1380                         clear_bit(GLF_LOCK, &gl->gl_flags);
1381                         smp_mb__after_clear_bit();
1382                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1383                                 gfs2_glock_put_nolock(gl);
1384                         spin_unlock(&gl->gl_spin);
1385                         spin_lock(&lru_lock);
1386                         continue;
1387                 }
1388                 nr_skipped++;
1389                 list_add(&gl->gl_lru, &skipped);
1390         }
1391         list_splice(&skipped, &lru_list);
1392         atomic_add(nr_skipped, &lru_count);
1393         spin_unlock(&lru_lock);
1394 out:
1395         return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1396 }
1397
1398 static struct shrinker glock_shrinker = {
1399         .shrink = gfs2_shrink_glock_memory,
1400         .seeks = DEFAULT_SEEKS,
1401 };
1402
1403 /**
1404  * examine_bucket - Call a function for glock in a hash bucket
1405  * @examiner: the function
1406  * @sdp: the filesystem
1407  * @bucket: the bucket
1408  *
1409  * Returns: 1 if the bucket has entries
1410  */
1411
1412 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1413                           unsigned int hash)
1414 {
1415         struct gfs2_glock *gl, *prev = NULL;
1416         int has_entries = 0;
1417         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1418
1419         read_lock(gl_lock_addr(hash));
1420         /* Can't use hlist_for_each_entry - don't want prefetch here */
1421         if (hlist_empty(head))
1422                 goto out;
1423         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1424         while(1) {
1425                 if (!sdp || gl->gl_sbd == sdp) {
1426                         gfs2_glock_hold(gl);
1427                         read_unlock(gl_lock_addr(hash));
1428                         if (prev)
1429                                 gfs2_glock_put(prev);
1430                         prev = gl;
1431                         examiner(gl);
1432                         has_entries = 1;
1433                         read_lock(gl_lock_addr(hash));
1434                 }
1435                 if (gl->gl_list.next == NULL)
1436                         break;
1437                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1438         }
1439 out:
1440         read_unlock(gl_lock_addr(hash));
1441         if (prev)
1442                 gfs2_glock_put(prev);
1443         cond_resched();
1444         return has_entries;
1445 }
1446
1447
1448 /**
1449  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1450  * @gl: The glock to thaw
1451  *
1452  * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1453  * so this has to result in the ref count being dropped by one.
1454  */
1455
1456 static void thaw_glock(struct gfs2_glock *gl)
1457 {
1458         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1459                 return;
1460         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1461         gfs2_glock_hold(gl);
1462         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1463                 gfs2_glock_put(gl);
1464 }
1465
1466 /**
1467  * clear_glock - look at a glock and see if we can free it from glock cache
1468  * @gl: the glock to look at
1469  *
1470  */
1471
1472 static void clear_glock(struct gfs2_glock *gl)
1473 {
1474         spin_lock(&lru_lock);
1475         if (!list_empty(&gl->gl_lru)) {
1476                 list_del_init(&gl->gl_lru);
1477                 atomic_dec(&lru_count);
1478         }
1479         spin_unlock(&lru_lock);
1480
1481         spin_lock(&gl->gl_spin);
1482         if (find_first_holder(gl) == NULL && gl->gl_state != LM_ST_UNLOCKED)
1483                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1484         spin_unlock(&gl->gl_spin);
1485         gfs2_glock_hold(gl);
1486         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1487                 gfs2_glock_put(gl);
1488 }
1489
1490 /**
1491  * gfs2_glock_thaw - Thaw any frozen glocks
1492  * @sdp: The super block
1493  *
1494  */
1495
1496 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1497 {
1498         unsigned x;
1499
1500         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1501                 examine_bucket(thaw_glock, sdp, x);
1502 }
1503
1504 /**
1505  * gfs2_gl_hash_clear - Empty out the glock hash table
1506  * @sdp: the filesystem
1507  * @wait: wait until it's all gone
1508  *
1509  * Called when unmounting the filesystem.
1510  */
1511
1512 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1513 {
1514         unsigned int x;
1515
1516         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1517                 examine_bucket(clear_glock, sdp, x);
1518         flush_workqueue(glock_workqueue);
1519         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1520         gfs2_dump_lockstate(sdp);
1521 }
1522
1523 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1524 {
1525         struct gfs2_glock *gl = ip->i_gl;
1526         int ret;
1527
1528         ret = gfs2_truncatei_resume(ip);
1529         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1530
1531         spin_lock(&gl->gl_spin);
1532         clear_bit(GLF_LOCK, &gl->gl_flags);
1533         run_queue(gl, 1);
1534         spin_unlock(&gl->gl_spin);
1535 }
1536
1537 static const char *state2str(unsigned state)
1538 {
1539         switch(state) {
1540         case LM_ST_UNLOCKED:
1541                 return "UN";
1542         case LM_ST_SHARED:
1543                 return "SH";
1544         case LM_ST_DEFERRED:
1545                 return "DF";
1546         case LM_ST_EXCLUSIVE:
1547                 return "EX";
1548         }
1549         return "??";
1550 }
1551
1552 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1553 {
1554         char *p = buf;
1555         if (flags & LM_FLAG_TRY)
1556                 *p++ = 't';
1557         if (flags & LM_FLAG_TRY_1CB)
1558                 *p++ = 'T';
1559         if (flags & LM_FLAG_NOEXP)
1560                 *p++ = 'e';
1561         if (flags & LM_FLAG_ANY)
1562                 *p++ = 'A';
1563         if (flags & LM_FLAG_PRIORITY)
1564                 *p++ = 'p';
1565         if (flags & GL_ASYNC)
1566                 *p++ = 'a';
1567         if (flags & GL_EXACT)
1568                 *p++ = 'E';
1569         if (flags & GL_NOCACHE)
1570                 *p++ = 'c';
1571         if (test_bit(HIF_HOLDER, &iflags))
1572                 *p++ = 'H';
1573         if (test_bit(HIF_WAIT, &iflags))
1574                 *p++ = 'W';
1575         if (test_bit(HIF_FIRST, &iflags))
1576                 *p++ = 'F';
1577         *p = 0;
1578         return buf;
1579 }
1580
1581 /**
1582  * dump_holder - print information about a glock holder
1583  * @seq: the seq_file struct
1584  * @gh: the glock holder
1585  *
1586  * Returns: 0 on success, -ENOBUFS when we run out of space
1587  */
1588
1589 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1590 {
1591         struct task_struct *gh_owner = NULL;
1592         char buffer[KSYM_SYMBOL_LEN];
1593         char flags_buf[32];
1594
1595         sprint_symbol(buffer, gh->gh_ip);
1596         if (gh->gh_owner_pid)
1597                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1598         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
1599                   state2str(gh->gh_state),
1600                   hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1601                   gh->gh_error, 
1602                   gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1603                   gh_owner ? gh_owner->comm : "(ended)", buffer);
1604         return 0;
1605 }
1606
1607 static const char *gflags2str(char *buf, const unsigned long *gflags)
1608 {
1609         char *p = buf;
1610         if (test_bit(GLF_LOCK, gflags))
1611                 *p++ = 'l';
1612         if (test_bit(GLF_DEMOTE, gflags))
1613                 *p++ = 'D';
1614         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1615                 *p++ = 'd';
1616         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1617                 *p++ = 'p';
1618         if (test_bit(GLF_DIRTY, gflags))
1619                 *p++ = 'y';
1620         if (test_bit(GLF_LFLUSH, gflags))
1621                 *p++ = 'f';
1622         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1623                 *p++ = 'i';
1624         if (test_bit(GLF_REPLY_PENDING, gflags))
1625                 *p++ = 'r';
1626         if (test_bit(GLF_INITIAL, gflags))
1627                 *p++ = 'I';
1628         if (test_bit(GLF_FROZEN, gflags))
1629                 *p++ = 'F';
1630         *p = 0;
1631         return buf;
1632 }
1633
1634 /**
1635  * __dump_glock - print information about a glock
1636  * @seq: The seq_file struct
1637  * @gl: the glock
1638  *
1639  * The file format is as follows:
1640  * One line per object, capital letters are used to indicate objects
1641  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1642  * other objects are indented by a single space and follow the glock to
1643  * which they are related. Fields are indicated by lower case letters
1644  * followed by a colon and the field value, except for strings which are in
1645  * [] so that its possible to see if they are composed of spaces for
1646  * example. The field's are n = number (id of the object), f = flags,
1647  * t = type, s = state, r = refcount, e = error, p = pid.
1648  *
1649  * Returns: 0 on success, -ENOBUFS when we run out of space
1650  */
1651
1652 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1653 {
1654         const struct gfs2_glock_operations *glops = gl->gl_ops;
1655         unsigned long long dtime;
1656         const struct gfs2_holder *gh;
1657         char gflags_buf[32];
1658         int error = 0;
1659
1660         dtime = jiffies - gl->gl_demote_time;
1661         dtime *= 1000000/HZ; /* demote time in uSec */
1662         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1663                 dtime = 0;
1664         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d r:%d\n",
1665                   state2str(gl->gl_state),
1666                   gl->gl_name.ln_type,
1667                   (unsigned long long)gl->gl_name.ln_number,
1668                   gflags2str(gflags_buf, &gl->gl_flags),
1669                   state2str(gl->gl_target),
1670                   state2str(gl->gl_demote_state), dtime,
1671                   atomic_read(&gl->gl_ail_count),
1672                   atomic_read(&gl->gl_ref));
1673
1674         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1675                 error = dump_holder(seq, gh);
1676                 if (error)
1677                         goto out;
1678         }
1679         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1680                 error = glops->go_dump(seq, gl);
1681 out:
1682         return error;
1683 }
1684
1685 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1686 {
1687         int ret;
1688         spin_lock(&gl->gl_spin);
1689         ret = __dump_glock(seq, gl);
1690         spin_unlock(&gl->gl_spin);
1691         return ret;
1692 }
1693
1694 /**
1695  * gfs2_dump_lockstate - print out the current lockstate
1696  * @sdp: the filesystem
1697  * @ub: the buffer to copy the information into
1698  *
1699  * If @ub is NULL, dump the lockstate to the console.
1700  *
1701  */
1702
1703 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1704 {
1705         struct gfs2_glock *gl;
1706         struct hlist_node *h;
1707         unsigned int x;
1708         int error = 0;
1709
1710         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1711
1712                 read_lock(gl_lock_addr(x));
1713
1714                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1715                         if (gl->gl_sbd != sdp)
1716                                 continue;
1717
1718                         error = dump_glock(NULL, gl);
1719                         if (error)
1720                                 break;
1721                 }
1722
1723                 read_unlock(gl_lock_addr(x));
1724
1725                 if (error)
1726                         break;
1727         }
1728
1729
1730         return error;
1731 }
1732
1733
1734 int __init gfs2_glock_init(void)
1735 {
1736         unsigned i;
1737         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1738                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1739         }
1740 #ifdef GL_HASH_LOCK_SZ
1741         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1742                 rwlock_init(&gl_hash_locks[i]);
1743         }
1744 #endif
1745
1746         glock_workqueue = create_workqueue("glock_workqueue");
1747         if (IS_ERR(glock_workqueue))
1748                 return PTR_ERR(glock_workqueue);
1749         gfs2_delete_workqueue = create_workqueue("delete_workqueue");
1750         if (IS_ERR(gfs2_delete_workqueue)) {
1751                 destroy_workqueue(glock_workqueue);
1752                 return PTR_ERR(gfs2_delete_workqueue);
1753         }
1754
1755         register_shrinker(&glock_shrinker);
1756
1757         return 0;
1758 }
1759
1760 void gfs2_glock_exit(void)
1761 {
1762         unregister_shrinker(&glock_shrinker);
1763         destroy_workqueue(glock_workqueue);
1764         destroy_workqueue(gfs2_delete_workqueue);
1765 }
1766
1767 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1768 {
1769         struct gfs2_glock *gl;
1770
1771 restart:
1772         read_lock(gl_lock_addr(gi->hash));
1773         gl = gi->gl;
1774         if (gl) {
1775                 gi->gl = hlist_entry(gl->gl_list.next,
1776                                      struct gfs2_glock, gl_list);
1777         } else {
1778                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1779                                      struct gfs2_glock, gl_list);
1780         }
1781         if (gi->gl)
1782                 gfs2_glock_hold(gi->gl);
1783         read_unlock(gl_lock_addr(gi->hash));
1784         if (gl)
1785                 gfs2_glock_put(gl);
1786         while (gi->gl == NULL) {
1787                 gi->hash++;
1788                 if (gi->hash >= GFS2_GL_HASH_SIZE)
1789                         return 1;
1790                 read_lock(gl_lock_addr(gi->hash));
1791                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1792                                      struct gfs2_glock, gl_list);
1793                 if (gi->gl)
1794                         gfs2_glock_hold(gi->gl);
1795                 read_unlock(gl_lock_addr(gi->hash));
1796         }
1797
1798         if (gi->sdp != gi->gl->gl_sbd)
1799                 goto restart;
1800
1801         return 0;
1802 }
1803
1804 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1805 {
1806         if (gi->gl)
1807                 gfs2_glock_put(gi->gl);
1808         gi->gl = NULL;
1809 }
1810
1811 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1812 {
1813         struct gfs2_glock_iter *gi = seq->private;
1814         loff_t n = *pos;
1815
1816         gi->hash = 0;
1817
1818         do {
1819                 if (gfs2_glock_iter_next(gi)) {
1820                         gfs2_glock_iter_free(gi);
1821                         return NULL;
1822                 }
1823         } while (n--);
1824
1825         return gi->gl;
1826 }
1827
1828 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1829                                  loff_t *pos)
1830 {
1831         struct gfs2_glock_iter *gi = seq->private;
1832
1833         (*pos)++;
1834
1835         if (gfs2_glock_iter_next(gi)) {
1836                 gfs2_glock_iter_free(gi);
1837                 return NULL;
1838         }
1839
1840         return gi->gl;
1841 }
1842
1843 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1844 {
1845         struct gfs2_glock_iter *gi = seq->private;
1846         gfs2_glock_iter_free(gi);
1847 }
1848
1849 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1850 {
1851         return dump_glock(seq, iter_ptr);
1852 }
1853
1854 static const struct seq_operations gfs2_glock_seq_ops = {
1855         .start = gfs2_glock_seq_start,
1856         .next  = gfs2_glock_seq_next,
1857         .stop  = gfs2_glock_seq_stop,
1858         .show  = gfs2_glock_seq_show,
1859 };
1860
1861 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1862 {
1863         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1864                                    sizeof(struct gfs2_glock_iter));
1865         if (ret == 0) {
1866                 struct seq_file *seq = file->private_data;
1867                 struct gfs2_glock_iter *gi = seq->private;
1868                 gi->sdp = inode->i_private;
1869         }
1870         return ret;
1871 }
1872
1873 static const struct file_operations gfs2_debug_fops = {
1874         .owner   = THIS_MODULE,
1875         .open    = gfs2_debugfs_open,
1876         .read    = seq_read,
1877         .llseek  = seq_lseek,
1878         .release = seq_release_private,
1879 };
1880
1881 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1882 {
1883         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1884         if (!sdp->debugfs_dir)
1885                 return -ENOMEM;
1886         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1887                                                          S_IFREG | S_IRUGO,
1888                                                          sdp->debugfs_dir, sdp,
1889                                                          &gfs2_debug_fops);
1890         if (!sdp->debugfs_dentry_glocks)
1891                 return -ENOMEM;
1892
1893         return 0;
1894 }
1895
1896 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1897 {
1898         if (sdp && sdp->debugfs_dir) {
1899                 if (sdp->debugfs_dentry_glocks) {
1900                         debugfs_remove(sdp->debugfs_dentry_glocks);
1901                         sdp->debugfs_dentry_glocks = NULL;
1902                 }
1903                 debugfs_remove(sdp->debugfs_dir);
1904                 sdp->debugfs_dir = NULL;
1905         }
1906 }
1907
1908 int gfs2_register_debugfs(void)
1909 {
1910         gfs2_root = debugfs_create_dir("gfs2", NULL);
1911         return gfs2_root ? 0 : -ENOMEM;
1912 }
1913
1914 void gfs2_unregister_debugfs(void)
1915 {
1916         debugfs_remove(gfs2_root);
1917         gfs2_root = NULL;
1918 }