parisc: Remove unnecessary macros from entry.S
[safe/jmp/linux-2.6] / fs / gfs2 / glock.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 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 }
859
860 /**
861  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
862  * @gh: the holder structure
863  *
864  */
865
866 void gfs2_holder_uninit(struct gfs2_holder *gh)
867 {
868         put_pid(gh->gh_owner_pid);
869         gfs2_glock_put(gh->gh_gl);
870         gh->gh_gl = NULL;
871         gh->gh_ip = 0;
872 }
873
874 /**
875  * gfs2_glock_holder_wait
876  * @word: unused
877  *
878  * This function and gfs2_glock_demote_wait both show up in the WCHAN
879  * field. Thus I've separated these otherwise identical functions in
880  * order to be more informative to the user.
881  */
882
883 static int gfs2_glock_holder_wait(void *word)
884 {
885         schedule();
886         return 0;
887 }
888
889 static int gfs2_glock_demote_wait(void *word)
890 {
891         schedule();
892         return 0;
893 }
894
895 static void wait_on_holder(struct gfs2_holder *gh)
896 {
897         might_sleep();
898         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
899 }
900
901 static void wait_on_demote(struct gfs2_glock *gl)
902 {
903         might_sleep();
904         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
905 }
906
907 /**
908  * handle_callback - process a demote request
909  * @gl: the glock
910  * @state: the state the caller wants us to change to
911  *
912  * There are only two requests that we are going to see in actual
913  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
914  */
915
916 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
917                             unsigned long delay)
918 {
919         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
920
921         set_bit(bit, &gl->gl_flags);
922         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
923                 gl->gl_demote_state = state;
924                 gl->gl_demote_time = jiffies;
925         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
926                         gl->gl_demote_state != state) {
927                 gl->gl_demote_state = LM_ST_UNLOCKED;
928         }
929         if (gl->gl_ops->go_callback)
930                 gl->gl_ops->go_callback(gl);
931         trace_gfs2_demote_rq(gl);
932 }
933
934 /**
935  * gfs2_glock_wait - wait on a glock acquisition
936  * @gh: the glock holder
937  *
938  * Returns: 0 on success
939  */
940
941 int gfs2_glock_wait(struct gfs2_holder *gh)
942 {
943         wait_on_holder(gh);
944         return gh->gh_error;
945 }
946
947 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
948 {
949         va_list args;
950
951         va_start(args, fmt);
952         if (seq) {
953                 struct gfs2_glock_iter *gi = seq->private;
954                 vsprintf(gi->string, fmt, args);
955                 seq_printf(seq, gi->string);
956         } else {
957                 printk(KERN_ERR " ");
958                 vprintk(fmt, args);
959         }
960         va_end(args);
961 }
962
963 /**
964  * add_to_queue - Add a holder to the wait queue (but look for recursion)
965  * @gh: the holder structure to add
966  *
967  * Eventually we should move the recursive locking trap to a
968  * debugging option or something like that. This is the fast
969  * path and needs to have the minimum number of distractions.
970  * 
971  */
972
973 static inline void add_to_queue(struct gfs2_holder *gh)
974 __releases(&gl->gl_spin)
975 __acquires(&gl->gl_spin)
976 {
977         struct gfs2_glock *gl = gh->gh_gl;
978         struct gfs2_sbd *sdp = gl->gl_sbd;
979         struct list_head *insert_pt = NULL;
980         struct gfs2_holder *gh2;
981         int try_lock = 0;
982
983         BUG_ON(gh->gh_owner_pid == NULL);
984         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
985                 BUG();
986
987         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
988                 if (test_bit(GLF_LOCK, &gl->gl_flags))
989                         try_lock = 1;
990                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
991                         goto fail;
992         }
993
994         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
995                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
996                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
997                         goto trap_recursive;
998                 if (try_lock &&
999                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
1000                     !may_grant(gl, gh)) {
1001 fail:
1002                         gh->gh_error = GLR_TRYFAILED;
1003                         gfs2_holder_wake(gh);
1004                         return;
1005                 }
1006                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
1007                         continue;
1008                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
1009                         insert_pt = &gh2->gh_list;
1010         }
1011         if (likely(insert_pt == NULL)) {
1012                 list_add_tail(&gh->gh_list, &gl->gl_holders);
1013                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
1014                         goto do_cancel;
1015                 return;
1016         }
1017         trace_gfs2_glock_queue(gh, 1);
1018         list_add_tail(&gh->gh_list, insert_pt);
1019 do_cancel:
1020         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
1021         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
1022                 spin_unlock(&gl->gl_spin);
1023                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
1024                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
1025                 spin_lock(&gl->gl_spin);
1026         }
1027         return;
1028
1029 trap_recursive:
1030         print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
1031         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
1032         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1033                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
1034         print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
1035         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
1036         printk(KERN_ERR "lock type: %d req lock state : %d\n",
1037                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1038         __dump_glock(NULL, gl);
1039         BUG();
1040 }
1041
1042 /**
1043  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1044  * @gh: the holder structure
1045  *
1046  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1047  *
1048  * Returns: 0, GLR_TRYFAILED, or errno on failure
1049  */
1050
1051 int gfs2_glock_nq(struct gfs2_holder *gh)
1052 {
1053         struct gfs2_glock *gl = gh->gh_gl;
1054         struct gfs2_sbd *sdp = gl->gl_sbd;
1055         int error = 0;
1056
1057         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1058                 return -EIO;
1059
1060         spin_lock(&gl->gl_spin);
1061         add_to_queue(gh);
1062         run_queue(gl, 1);
1063         spin_unlock(&gl->gl_spin);
1064
1065         if (!(gh->gh_flags & GL_ASYNC))
1066                 error = gfs2_glock_wait(gh);
1067
1068         return error;
1069 }
1070
1071 /**
1072  * gfs2_glock_poll - poll to see if an async request has been completed
1073  * @gh: the holder
1074  *
1075  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1076  */
1077
1078 int gfs2_glock_poll(struct gfs2_holder *gh)
1079 {
1080         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1081 }
1082
1083 /**
1084  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1085  * @gh: the glock holder
1086  *
1087  */
1088
1089 void gfs2_glock_dq(struct gfs2_holder *gh)
1090 {
1091         struct gfs2_glock *gl = gh->gh_gl;
1092         const struct gfs2_glock_operations *glops = gl->gl_ops;
1093         unsigned delay = 0;
1094         int fast_path = 0;
1095
1096         spin_lock(&gl->gl_spin);
1097         if (gh->gh_flags & GL_NOCACHE)
1098                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1099
1100         list_del_init(&gh->gh_list);
1101         if (find_first_holder(gl) == NULL) {
1102                 if (glops->go_unlock) {
1103                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1104                         spin_unlock(&gl->gl_spin);
1105                         glops->go_unlock(gh);
1106                         spin_lock(&gl->gl_spin);
1107                         clear_bit(GLF_LOCK, &gl->gl_flags);
1108                 }
1109                 if (list_empty(&gl->gl_holders) &&
1110                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1111                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1112                         fast_path = 1;
1113         }
1114         trace_gfs2_glock_queue(gh, 0);
1115         spin_unlock(&gl->gl_spin);
1116         if (likely(fast_path))
1117                 return;
1118
1119         gfs2_glock_hold(gl);
1120         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1121             !test_bit(GLF_DEMOTE, &gl->gl_flags))
1122                 delay = gl->gl_ops->go_min_hold_time;
1123         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1124                 gfs2_glock_put(gl);
1125 }
1126
1127 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1128 {
1129         struct gfs2_glock *gl = gh->gh_gl;
1130         gfs2_glock_dq(gh);
1131         wait_on_demote(gl);
1132 }
1133
1134 /**
1135  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1136  * @gh: the holder structure
1137  *
1138  */
1139
1140 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1141 {
1142         gfs2_glock_dq(gh);
1143         gfs2_holder_uninit(gh);
1144 }
1145
1146 /**
1147  * gfs2_glock_nq_num - acquire a glock based on lock number
1148  * @sdp: the filesystem
1149  * @number: the lock number
1150  * @glops: the glock operations for the type of glock
1151  * @state: the state to acquire the glock in
1152  * @flags: modifier flags for the aquisition
1153  * @gh: the struct gfs2_holder
1154  *
1155  * Returns: errno
1156  */
1157
1158 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1159                       const struct gfs2_glock_operations *glops,
1160                       unsigned int state, int flags, struct gfs2_holder *gh)
1161 {
1162         struct gfs2_glock *gl;
1163         int error;
1164
1165         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1166         if (!error) {
1167                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1168                 gfs2_glock_put(gl);
1169         }
1170
1171         return error;
1172 }
1173
1174 /**
1175  * glock_compare - Compare two struct gfs2_glock structures for sorting
1176  * @arg_a: the first structure
1177  * @arg_b: the second structure
1178  *
1179  */
1180
1181 static int glock_compare(const void *arg_a, const void *arg_b)
1182 {
1183         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1184         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1185         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1186         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1187
1188         if (a->ln_number > b->ln_number)
1189                 return 1;
1190         if (a->ln_number < b->ln_number)
1191                 return -1;
1192         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1193         return 0;
1194 }
1195
1196 /**
1197  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1198  * @num_gh: the number of structures
1199  * @ghs: an array of struct gfs2_holder structures
1200  *
1201  * Returns: 0 on success (all glocks acquired),
1202  *          errno on failure (no glocks acquired)
1203  */
1204
1205 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1206                      struct gfs2_holder **p)
1207 {
1208         unsigned int x;
1209         int error = 0;
1210
1211         for (x = 0; x < num_gh; x++)
1212                 p[x] = &ghs[x];
1213
1214         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1215
1216         for (x = 0; x < num_gh; x++) {
1217                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1218
1219                 error = gfs2_glock_nq(p[x]);
1220                 if (error) {
1221                         while (x--)
1222                                 gfs2_glock_dq(p[x]);
1223                         break;
1224                 }
1225         }
1226
1227         return error;
1228 }
1229
1230 /**
1231  * gfs2_glock_nq_m - acquire multiple glocks
1232  * @num_gh: the number of structures
1233  * @ghs: an array of struct gfs2_holder structures
1234  *
1235  *
1236  * Returns: 0 on success (all glocks acquired),
1237  *          errno on failure (no glocks acquired)
1238  */
1239
1240 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1241 {
1242         struct gfs2_holder *tmp[4];
1243         struct gfs2_holder **pph = tmp;
1244         int error = 0;
1245
1246         switch(num_gh) {
1247         case 0:
1248                 return 0;
1249         case 1:
1250                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1251                 return gfs2_glock_nq(ghs);
1252         default:
1253                 if (num_gh <= 4)
1254                         break;
1255                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1256                 if (!pph)
1257                         return -ENOMEM;
1258         }
1259
1260         error = nq_m_sync(num_gh, ghs, pph);
1261
1262         if (pph != tmp)
1263                 kfree(pph);
1264
1265         return error;
1266 }
1267
1268 /**
1269  * gfs2_glock_dq_m - release multiple glocks
1270  * @num_gh: the number of structures
1271  * @ghs: an array of struct gfs2_holder structures
1272  *
1273  */
1274
1275 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1276 {
1277         unsigned int x;
1278
1279         for (x = 0; x < num_gh; x++)
1280                 gfs2_glock_dq(&ghs[x]);
1281 }
1282
1283 /**
1284  * gfs2_glock_dq_uninit_m - release multiple glocks
1285  * @num_gh: the number of structures
1286  * @ghs: an array of struct gfs2_holder structures
1287  *
1288  */
1289
1290 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1291 {
1292         unsigned int x;
1293
1294         for (x = 0; x < num_gh; x++)
1295                 gfs2_glock_dq_uninit(&ghs[x]);
1296 }
1297
1298 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1299 {
1300         unsigned long delay = 0;
1301         unsigned long holdtime;
1302         unsigned long now = jiffies;
1303
1304         gfs2_glock_hold(gl);
1305         holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1306         if (time_before(now, holdtime))
1307                 delay = holdtime - now;
1308         if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1309                 delay = gl->gl_ops->go_min_hold_time;
1310
1311         spin_lock(&gl->gl_spin);
1312         handle_callback(gl, state, delay);
1313         spin_unlock(&gl->gl_spin);
1314         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1315                 gfs2_glock_put(gl);
1316 }
1317
1318 /**
1319  * gfs2_glock_complete - Callback used by locking
1320  * @gl: Pointer to the glock
1321  * @ret: The return value from the dlm
1322  *
1323  */
1324
1325 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1326 {
1327         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1328         gl->gl_reply = ret;
1329         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1330                 struct gfs2_holder *gh;
1331                 spin_lock(&gl->gl_spin);
1332                 gh = find_first_waiter(gl);
1333                 if ((!(gh && (gh->gh_flags & LM_FLAG_NOEXP)) &&
1334                      (gl->gl_target != LM_ST_UNLOCKED)) ||
1335                     ((ret & ~LM_OUT_ST_MASK) != 0))
1336                         set_bit(GLF_FROZEN, &gl->gl_flags);
1337                 spin_unlock(&gl->gl_spin);
1338                 if (test_bit(GLF_FROZEN, &gl->gl_flags))
1339                         return;
1340         }
1341         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1342         gfs2_glock_hold(gl);
1343         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1344                 gfs2_glock_put(gl);
1345 }
1346
1347
1348 static int gfs2_shrink_glock_memory(int nr, gfp_t gfp_mask)
1349 {
1350         struct gfs2_glock *gl;
1351         int may_demote;
1352         int nr_skipped = 0;
1353         LIST_HEAD(skipped);
1354
1355         if (nr == 0)
1356                 goto out;
1357
1358         if (!(gfp_mask & __GFP_FS))
1359                 return -1;
1360
1361         spin_lock(&lru_lock);
1362         while(nr && !list_empty(&lru_list)) {
1363                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1364                 list_del_init(&gl->gl_lru);
1365                 atomic_dec(&lru_count);
1366
1367                 /* Test for being demotable */
1368                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1369                         gfs2_glock_hold(gl);
1370                         spin_unlock(&lru_lock);
1371                         spin_lock(&gl->gl_spin);
1372                         may_demote = demote_ok(gl);
1373                         if (may_demote) {
1374                                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1375                                 nr--;
1376                         }
1377                         clear_bit(GLF_LOCK, &gl->gl_flags);
1378                         smp_mb__after_clear_bit();
1379                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1380                                 gfs2_glock_put_nolock(gl);
1381                         spin_unlock(&gl->gl_spin);
1382                         spin_lock(&lru_lock);
1383                         continue;
1384                 }
1385                 nr_skipped++;
1386                 list_add(&gl->gl_lru, &skipped);
1387         }
1388         list_splice(&skipped, &lru_list);
1389         atomic_add(nr_skipped, &lru_count);
1390         spin_unlock(&lru_lock);
1391 out:
1392         return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1393 }
1394
1395 static struct shrinker glock_shrinker = {
1396         .shrink = gfs2_shrink_glock_memory,
1397         .seeks = DEFAULT_SEEKS,
1398 };
1399
1400 /**
1401  * examine_bucket - Call a function for glock in a hash bucket
1402  * @examiner: the function
1403  * @sdp: the filesystem
1404  * @bucket: the bucket
1405  *
1406  * Returns: 1 if the bucket has entries
1407  */
1408
1409 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1410                           unsigned int hash)
1411 {
1412         struct gfs2_glock *gl, *prev = NULL;
1413         int has_entries = 0;
1414         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1415
1416         read_lock(gl_lock_addr(hash));
1417         /* Can't use hlist_for_each_entry - don't want prefetch here */
1418         if (hlist_empty(head))
1419                 goto out;
1420         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1421         while(1) {
1422                 if (!sdp || gl->gl_sbd == sdp) {
1423                         gfs2_glock_hold(gl);
1424                         read_unlock(gl_lock_addr(hash));
1425                         if (prev)
1426                                 gfs2_glock_put(prev);
1427                         prev = gl;
1428                         examiner(gl);
1429                         has_entries = 1;
1430                         read_lock(gl_lock_addr(hash));
1431                 }
1432                 if (gl->gl_list.next == NULL)
1433                         break;
1434                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1435         }
1436 out:
1437         read_unlock(gl_lock_addr(hash));
1438         if (prev)
1439                 gfs2_glock_put(prev);
1440         cond_resched();
1441         return has_entries;
1442 }
1443
1444
1445 /**
1446  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1447  * @gl: The glock to thaw
1448  *
1449  * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1450  * so this has to result in the ref count being dropped by one.
1451  */
1452
1453 static void thaw_glock(struct gfs2_glock *gl)
1454 {
1455         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1456                 return;
1457         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1458         gfs2_glock_hold(gl);
1459         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1460                 gfs2_glock_put(gl);
1461 }
1462
1463 /**
1464  * clear_glock - look at a glock and see if we can free it from glock cache
1465  * @gl: the glock to look at
1466  *
1467  */
1468
1469 static void clear_glock(struct gfs2_glock *gl)
1470 {
1471         spin_lock(&lru_lock);
1472         if (!list_empty(&gl->gl_lru)) {
1473                 list_del_init(&gl->gl_lru);
1474                 atomic_dec(&lru_count);
1475         }
1476         spin_unlock(&lru_lock);
1477
1478         spin_lock(&gl->gl_spin);
1479         if (find_first_holder(gl) == NULL && gl->gl_state != LM_ST_UNLOCKED)
1480                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1481         spin_unlock(&gl->gl_spin);
1482         gfs2_glock_hold(gl);
1483         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1484                 gfs2_glock_put(gl);
1485 }
1486
1487 /**
1488  * gfs2_glock_thaw - Thaw any frozen glocks
1489  * @sdp: The super block
1490  *
1491  */
1492
1493 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1494 {
1495         unsigned x;
1496
1497         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1498                 examine_bucket(thaw_glock, sdp, x);
1499 }
1500
1501 /**
1502  * gfs2_gl_hash_clear - Empty out the glock hash table
1503  * @sdp: the filesystem
1504  * @wait: wait until it's all gone
1505  *
1506  * Called when unmounting the filesystem.
1507  */
1508
1509 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1510 {
1511         unsigned int x;
1512
1513         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1514                 examine_bucket(clear_glock, sdp, x);
1515         flush_workqueue(glock_workqueue);
1516         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1517         gfs2_dump_lockstate(sdp);
1518 }
1519
1520 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1521 {
1522         struct gfs2_glock *gl = ip->i_gl;
1523         int ret;
1524
1525         ret = gfs2_truncatei_resume(ip);
1526         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1527
1528         spin_lock(&gl->gl_spin);
1529         clear_bit(GLF_LOCK, &gl->gl_flags);
1530         run_queue(gl, 1);
1531         spin_unlock(&gl->gl_spin);
1532 }
1533
1534 static const char *state2str(unsigned state)
1535 {
1536         switch(state) {
1537         case LM_ST_UNLOCKED:
1538                 return "UN";
1539         case LM_ST_SHARED:
1540                 return "SH";
1541         case LM_ST_DEFERRED:
1542                 return "DF";
1543         case LM_ST_EXCLUSIVE:
1544                 return "EX";
1545         }
1546         return "??";
1547 }
1548
1549 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1550 {
1551         char *p = buf;
1552         if (flags & LM_FLAG_TRY)
1553                 *p++ = 't';
1554         if (flags & LM_FLAG_TRY_1CB)
1555                 *p++ = 'T';
1556         if (flags & LM_FLAG_NOEXP)
1557                 *p++ = 'e';
1558         if (flags & LM_FLAG_ANY)
1559                 *p++ = 'A';
1560         if (flags & LM_FLAG_PRIORITY)
1561                 *p++ = 'p';
1562         if (flags & GL_ASYNC)
1563                 *p++ = 'a';
1564         if (flags & GL_EXACT)
1565                 *p++ = 'E';
1566         if (flags & GL_NOCACHE)
1567                 *p++ = 'c';
1568         if (test_bit(HIF_HOLDER, &iflags))
1569                 *p++ = 'H';
1570         if (test_bit(HIF_WAIT, &iflags))
1571                 *p++ = 'W';
1572         if (test_bit(HIF_FIRST, &iflags))
1573                 *p++ = 'F';
1574         *p = 0;
1575         return buf;
1576 }
1577
1578 /**
1579  * dump_holder - print information about a glock holder
1580  * @seq: the seq_file struct
1581  * @gh: the glock holder
1582  *
1583  * Returns: 0 on success, -ENOBUFS when we run out of space
1584  */
1585
1586 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1587 {
1588         struct task_struct *gh_owner = NULL;
1589         char buffer[KSYM_SYMBOL_LEN];
1590         char flags_buf[32];
1591
1592         sprint_symbol(buffer, gh->gh_ip);
1593         if (gh->gh_owner_pid)
1594                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1595         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
1596                   state2str(gh->gh_state),
1597                   hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1598                   gh->gh_error, 
1599                   gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1600                   gh_owner ? gh_owner->comm : "(ended)", buffer);
1601         return 0;
1602 }
1603
1604 static const char *gflags2str(char *buf, const unsigned long *gflags)
1605 {
1606         char *p = buf;
1607         if (test_bit(GLF_LOCK, gflags))
1608                 *p++ = 'l';
1609         if (test_bit(GLF_DEMOTE, gflags))
1610                 *p++ = 'D';
1611         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1612                 *p++ = 'd';
1613         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1614                 *p++ = 'p';
1615         if (test_bit(GLF_DIRTY, gflags))
1616                 *p++ = 'y';
1617         if (test_bit(GLF_LFLUSH, gflags))
1618                 *p++ = 'f';
1619         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1620                 *p++ = 'i';
1621         if (test_bit(GLF_REPLY_PENDING, gflags))
1622                 *p++ = 'r';
1623         if (test_bit(GLF_INITIAL, gflags))
1624                 *p++ = 'I';
1625         if (test_bit(GLF_FROZEN, gflags))
1626                 *p++ = 'F';
1627         *p = 0;
1628         return buf;
1629 }
1630
1631 /**
1632  * __dump_glock - print information about a glock
1633  * @seq: The seq_file struct
1634  * @gl: the glock
1635  *
1636  * The file format is as follows:
1637  * One line per object, capital letters are used to indicate objects
1638  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1639  * other objects are indented by a single space and follow the glock to
1640  * which they are related. Fields are indicated by lower case letters
1641  * followed by a colon and the field value, except for strings which are in
1642  * [] so that its possible to see if they are composed of spaces for
1643  * example. The field's are n = number (id of the object), f = flags,
1644  * t = type, s = state, r = refcount, e = error, p = pid.
1645  *
1646  * Returns: 0 on success, -ENOBUFS when we run out of space
1647  */
1648
1649 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1650 {
1651         const struct gfs2_glock_operations *glops = gl->gl_ops;
1652         unsigned long long dtime;
1653         const struct gfs2_holder *gh;
1654         char gflags_buf[32];
1655         int error = 0;
1656
1657         dtime = jiffies - gl->gl_demote_time;
1658         dtime *= 1000000/HZ; /* demote time in uSec */
1659         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1660                 dtime = 0;
1661         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d r:%d\n",
1662                   state2str(gl->gl_state),
1663                   gl->gl_name.ln_type,
1664                   (unsigned long long)gl->gl_name.ln_number,
1665                   gflags2str(gflags_buf, &gl->gl_flags),
1666                   state2str(gl->gl_target),
1667                   state2str(gl->gl_demote_state), dtime,
1668                   atomic_read(&gl->gl_ail_count),
1669                   atomic_read(&gl->gl_ref));
1670
1671         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1672                 error = dump_holder(seq, gh);
1673                 if (error)
1674                         goto out;
1675         }
1676         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1677                 error = glops->go_dump(seq, gl);
1678 out:
1679         return error;
1680 }
1681
1682 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1683 {
1684         int ret;
1685         spin_lock(&gl->gl_spin);
1686         ret = __dump_glock(seq, gl);
1687         spin_unlock(&gl->gl_spin);
1688         return ret;
1689 }
1690
1691 /**
1692  * gfs2_dump_lockstate - print out the current lockstate
1693  * @sdp: the filesystem
1694  * @ub: the buffer to copy the information into
1695  *
1696  * If @ub is NULL, dump the lockstate to the console.
1697  *
1698  */
1699
1700 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1701 {
1702         struct gfs2_glock *gl;
1703         struct hlist_node *h;
1704         unsigned int x;
1705         int error = 0;
1706
1707         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1708
1709                 read_lock(gl_lock_addr(x));
1710
1711                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1712                         if (gl->gl_sbd != sdp)
1713                                 continue;
1714
1715                         error = dump_glock(NULL, gl);
1716                         if (error)
1717                                 break;
1718                 }
1719
1720                 read_unlock(gl_lock_addr(x));
1721
1722                 if (error)
1723                         break;
1724         }
1725
1726
1727         return error;
1728 }
1729
1730
1731 int __init gfs2_glock_init(void)
1732 {
1733         unsigned i;
1734         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1735                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1736         }
1737 #ifdef GL_HASH_LOCK_SZ
1738         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1739                 rwlock_init(&gl_hash_locks[i]);
1740         }
1741 #endif
1742
1743         glock_workqueue = create_workqueue("glock_workqueue");
1744         if (IS_ERR(glock_workqueue))
1745                 return PTR_ERR(glock_workqueue);
1746         gfs2_delete_workqueue = create_workqueue("delete_workqueue");
1747         if (IS_ERR(gfs2_delete_workqueue)) {
1748                 destroy_workqueue(glock_workqueue);
1749                 return PTR_ERR(gfs2_delete_workqueue);
1750         }
1751
1752         register_shrinker(&glock_shrinker);
1753
1754         return 0;
1755 }
1756
1757 void gfs2_glock_exit(void)
1758 {
1759         unregister_shrinker(&glock_shrinker);
1760         destroy_workqueue(glock_workqueue);
1761         destroy_workqueue(gfs2_delete_workqueue);
1762 }
1763
1764 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1765 {
1766         struct gfs2_glock *gl;
1767
1768 restart:
1769         read_lock(gl_lock_addr(gi->hash));
1770         gl = gi->gl;
1771         if (gl) {
1772                 gi->gl = hlist_entry(gl->gl_list.next,
1773                                      struct gfs2_glock, gl_list);
1774         } else {
1775                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1776                                      struct gfs2_glock, gl_list);
1777         }
1778         if (gi->gl)
1779                 gfs2_glock_hold(gi->gl);
1780         read_unlock(gl_lock_addr(gi->hash));
1781         if (gl)
1782                 gfs2_glock_put(gl);
1783         while (gi->gl == NULL) {
1784                 gi->hash++;
1785                 if (gi->hash >= GFS2_GL_HASH_SIZE)
1786                         return 1;
1787                 read_lock(gl_lock_addr(gi->hash));
1788                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1789                                      struct gfs2_glock, gl_list);
1790                 if (gi->gl)
1791                         gfs2_glock_hold(gi->gl);
1792                 read_unlock(gl_lock_addr(gi->hash));
1793         }
1794
1795         if (gi->sdp != gi->gl->gl_sbd)
1796                 goto restart;
1797
1798         return 0;
1799 }
1800
1801 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1802 {
1803         if (gi->gl)
1804                 gfs2_glock_put(gi->gl);
1805         gi->gl = NULL;
1806 }
1807
1808 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1809 {
1810         struct gfs2_glock_iter *gi = seq->private;
1811         loff_t n = *pos;
1812
1813         gi->hash = 0;
1814
1815         do {
1816                 if (gfs2_glock_iter_next(gi)) {
1817                         gfs2_glock_iter_free(gi);
1818                         return NULL;
1819                 }
1820         } while (n--);
1821
1822         return gi->gl;
1823 }
1824
1825 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1826                                  loff_t *pos)
1827 {
1828         struct gfs2_glock_iter *gi = seq->private;
1829
1830         (*pos)++;
1831
1832         if (gfs2_glock_iter_next(gi)) {
1833                 gfs2_glock_iter_free(gi);
1834                 return NULL;
1835         }
1836
1837         return gi->gl;
1838 }
1839
1840 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1841 {
1842         struct gfs2_glock_iter *gi = seq->private;
1843         gfs2_glock_iter_free(gi);
1844 }
1845
1846 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1847 {
1848         return dump_glock(seq, iter_ptr);
1849 }
1850
1851 static const struct seq_operations gfs2_glock_seq_ops = {
1852         .start = gfs2_glock_seq_start,
1853         .next  = gfs2_glock_seq_next,
1854         .stop  = gfs2_glock_seq_stop,
1855         .show  = gfs2_glock_seq_show,
1856 };
1857
1858 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1859 {
1860         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1861                                    sizeof(struct gfs2_glock_iter));
1862         if (ret == 0) {
1863                 struct seq_file *seq = file->private_data;
1864                 struct gfs2_glock_iter *gi = seq->private;
1865                 gi->sdp = inode->i_private;
1866         }
1867         return ret;
1868 }
1869
1870 static const struct file_operations gfs2_debug_fops = {
1871         .owner   = THIS_MODULE,
1872         .open    = gfs2_debugfs_open,
1873         .read    = seq_read,
1874         .llseek  = seq_lseek,
1875         .release = seq_release_private,
1876 };
1877
1878 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1879 {
1880         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1881         if (!sdp->debugfs_dir)
1882                 return -ENOMEM;
1883         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1884                                                          S_IFREG | S_IRUGO,
1885                                                          sdp->debugfs_dir, sdp,
1886                                                          &gfs2_debug_fops);
1887         if (!sdp->debugfs_dentry_glocks)
1888                 return -ENOMEM;
1889
1890         return 0;
1891 }
1892
1893 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1894 {
1895         if (sdp && sdp->debugfs_dir) {
1896                 if (sdp->debugfs_dentry_glocks) {
1897                         debugfs_remove(sdp->debugfs_dentry_glocks);
1898                         sdp->debugfs_dentry_glocks = NULL;
1899                 }
1900                 debugfs_remove(sdp->debugfs_dir);
1901                 sdp->debugfs_dir = NULL;
1902         }
1903 }
1904
1905 int gfs2_register_debugfs(void)
1906 {
1907         gfs2_root = debugfs_create_dir("gfs2", NULL);
1908         return gfs2_root ? 0 : -ENOMEM;
1909 }
1910
1911 void gfs2_unregister_debugfs(void)
1912 {
1913         debugfs_remove(gfs2_root);
1914         gfs2_root = NULL;
1915 }