Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[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/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/delay.h>
16 #include <linux/sort.h>
17 #include <linux/jhash.h>
18 #include <linux/kallsyms.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/list.h>
21 #include <linux/lm_interface.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <linux/rwsem.h>
25 #include <asm/uaccess.h>
26 #include <linux/seq_file.h>
27 #include <linux/debugfs.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
30 #include <linux/workqueue.h>
31 #include <linux/jiffies.h>
32
33 #include "gfs2.h"
34 #include "incore.h"
35 #include "glock.h"
36 #include "glops.h"
37 #include "inode.h"
38 #include "lops.h"
39 #include "meta_io.h"
40 #include "quota.h"
41 #include "super.h"
42 #include "util.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 DECLARE_RWSEM(gfs2_umount_flush_sem);
63 static struct dentry *gfs2_root;
64 static struct task_struct *scand_process;
65 static unsigned int scand_secs = 5;
66 static struct workqueue_struct *glock_workqueue;
67
68 #define GFS2_GL_HASH_SHIFT      15
69 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
70 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
71
72 static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
73 static struct dentry *gfs2_root;
74
75 /*
76  * Despite what you might think, the numbers below are not arbitrary :-)
77  * They are taken from the ipv4 routing hash code, which is well tested
78  * and thus should be nearly optimal. Later on we might tweek the numbers
79  * but for now this should be fine.
80  *
81  * The reason for putting the locks in a separate array from the list heads
82  * is that we can have fewer locks than list heads and save memory. We use
83  * the same hash function for both, but with a different hash mask.
84  */
85 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
86         defined(CONFIG_PROVE_LOCKING)
87
88 #ifdef CONFIG_LOCKDEP
89 # define GL_HASH_LOCK_SZ        256
90 #else
91 # if NR_CPUS >= 32
92 #  define GL_HASH_LOCK_SZ       4096
93 # elif NR_CPUS >= 16
94 #  define GL_HASH_LOCK_SZ       2048
95 # elif NR_CPUS >= 8
96 #  define GL_HASH_LOCK_SZ       1024
97 # elif NR_CPUS >= 4
98 #  define GL_HASH_LOCK_SZ       512
99 # else
100 #  define GL_HASH_LOCK_SZ       256
101 # endif
102 #endif
103
104 /* We never want more locks than chains */
105 #if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
106 # undef GL_HASH_LOCK_SZ
107 # define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
108 #endif
109
110 static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
111
112 static inline rwlock_t *gl_lock_addr(unsigned int x)
113 {
114         return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
115 }
116 #else /* not SMP, so no spinlocks required */
117 static inline rwlock_t *gl_lock_addr(unsigned int x)
118 {
119         return NULL;
120 }
121 #endif
122
123 /**
124  * gl_hash() - Turn glock number into hash bucket number
125  * @lock: The glock number
126  *
127  * Returns: The number of the corresponding hash bucket
128  */
129
130 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
131                             const struct lm_lockname *name)
132 {
133         unsigned int h;
134
135         h = jhash(&name->ln_number, sizeof(u64), 0);
136         h = jhash(&name->ln_type, sizeof(unsigned int), h);
137         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
138         h &= GFS2_GL_HASH_MASK;
139
140         return h;
141 }
142
143 /**
144  * glock_free() - Perform a few checks and then release struct gfs2_glock
145  * @gl: The glock to release
146  *
147  * Also calls lock module to release its internal structure for this glock.
148  *
149  */
150
151 static void glock_free(struct gfs2_glock *gl)
152 {
153         struct gfs2_sbd *sdp = gl->gl_sbd;
154         struct inode *aspace = gl->gl_aspace;
155
156         if (sdp->sd_lockstruct.ls_ops->lm_put_lock)
157                 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl->gl_lock);
158
159         if (aspace)
160                 gfs2_aspace_put(aspace);
161
162         kmem_cache_free(gfs2_glock_cachep, gl);
163 }
164
165 /**
166  * gfs2_glock_hold() - increment reference count on glock
167  * @gl: The glock to hold
168  *
169  */
170
171 static void gfs2_glock_hold(struct gfs2_glock *gl)
172 {
173         atomic_inc(&gl->gl_ref);
174 }
175
176 /**
177  * gfs2_glock_put() - Decrement reference count on glock
178  * @gl: The glock to put
179  *
180  */
181
182 int gfs2_glock_put(struct gfs2_glock *gl)
183 {
184         int rv = 0;
185
186         write_lock(gl_lock_addr(gl->gl_hash));
187         if (atomic_dec_and_test(&gl->gl_ref)) {
188                 hlist_del(&gl->gl_list);
189                 write_unlock(gl_lock_addr(gl->gl_hash));
190                 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_UNLOCKED);
191                 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_reclaim));
192                 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
193                 glock_free(gl);
194                 rv = 1;
195                 goto out;
196         }
197         write_unlock(gl_lock_addr(gl->gl_hash));
198 out:
199         return rv;
200 }
201
202 /**
203  * search_bucket() - Find struct gfs2_glock by lock number
204  * @bucket: the bucket to search
205  * @name: The lock name
206  *
207  * Returns: NULL, or the struct gfs2_glock with the requested number
208  */
209
210 static struct gfs2_glock *search_bucket(unsigned int hash,
211                                         const struct gfs2_sbd *sdp,
212                                         const struct lm_lockname *name)
213 {
214         struct gfs2_glock *gl;
215         struct hlist_node *h;
216
217         hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
218                 if (!lm_name_equal(&gl->gl_name, name))
219                         continue;
220                 if (gl->gl_sbd != sdp)
221                         continue;
222
223                 atomic_inc(&gl->gl_ref);
224
225                 return gl;
226         }
227
228         return NULL;
229 }
230
231 /**
232  * gfs2_glock_find() - Find glock by lock number
233  * @sdp: The GFS2 superblock
234  * @name: The lock name
235  *
236  * Returns: NULL, or the struct gfs2_glock with the requested number
237  */
238
239 static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
240                                           const struct lm_lockname *name)
241 {
242         unsigned int hash = gl_hash(sdp, name);
243         struct gfs2_glock *gl;
244
245         read_lock(gl_lock_addr(hash));
246         gl = search_bucket(hash, sdp, name);
247         read_unlock(gl_lock_addr(hash));
248
249         return gl;
250 }
251
252 /**
253  * may_grant - check if its ok to grant a new lock
254  * @gl: The glock
255  * @gh: The lock request which we wish to grant
256  *
257  * Returns: true if its ok to grant the lock
258  */
259
260 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
261 {
262         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
263         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
264              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
265                 return 0;
266         if (gl->gl_state == gh->gh_state)
267                 return 1;
268         if (gh->gh_flags & GL_EXACT)
269                 return 0;
270         if (gl->gl_state == LM_ST_EXCLUSIVE) {
271                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
272                         return 1;
273                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
274                         return 1;
275         }
276         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
277                 return 1;
278         return 0;
279 }
280
281 static void gfs2_holder_wake(struct gfs2_holder *gh)
282 {
283         clear_bit(HIF_WAIT, &gh->gh_iflags);
284         smp_mb__after_clear_bit();
285         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
286 }
287
288 /**
289  * do_promote - promote as many requests as possible on the current queue
290  * @gl: The glock
291  * 
292  * Returns: true if there is a blocked holder at the head of the list
293  */
294
295 static int do_promote(struct gfs2_glock *gl)
296 {
297         const struct gfs2_glock_operations *glops = gl->gl_ops;
298         struct gfs2_holder *gh, *tmp;
299         int ret;
300
301 restart:
302         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
303                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
304                         continue;
305                 if (may_grant(gl, gh)) {
306                         if (gh->gh_list.prev == &gl->gl_holders &&
307                             glops->go_lock) {
308                                 spin_unlock(&gl->gl_spin);
309                                 /* FIXME: eliminate this eventually */
310                                 ret = glops->go_lock(gh);
311                                 spin_lock(&gl->gl_spin);
312                                 if (ret) {
313                                         gh->gh_error = ret;
314                                         list_del_init(&gh->gh_list);
315                                         gfs2_holder_wake(gh);
316                                         goto restart;
317                                 }
318                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
319                                 gfs2_holder_wake(gh);
320                                 goto restart;
321                         }
322                         set_bit(HIF_HOLDER, &gh->gh_iflags);
323                         gfs2_holder_wake(gh);
324                         continue;
325                 }
326                 if (gh->gh_list.prev == &gl->gl_holders)
327                         return 1;
328                 break;
329         }
330         return 0;
331 }
332
333 /**
334  * do_error - Something unexpected has happened during a lock request
335  *
336  */
337
338 static inline void do_error(struct gfs2_glock *gl, const int ret)
339 {
340         struct gfs2_holder *gh, *tmp;
341
342         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
343                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
344                         continue;
345                 if (ret & LM_OUT_ERROR)
346                         gh->gh_error = -EIO;
347                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
348                         gh->gh_error = GLR_TRYFAILED;
349                 else
350                         continue;
351                 list_del_init(&gh->gh_list);
352                 gfs2_holder_wake(gh);
353         }
354 }
355
356 /**
357  * find_first_waiter - find the first gh that's waiting for the glock
358  * @gl: the glock
359  */
360
361 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
362 {
363         struct gfs2_holder *gh;
364
365         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
366                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
367                         return gh;
368         }
369         return NULL;
370 }
371
372 /**
373  * state_change - record that the glock is now in a different state
374  * @gl: the glock
375  * @new_state the new state
376  *
377  */
378
379 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
380 {
381         int held1, held2;
382
383         held1 = (gl->gl_state != LM_ST_UNLOCKED);
384         held2 = (new_state != LM_ST_UNLOCKED);
385
386         if (held1 != held2) {
387                 if (held2)
388                         gfs2_glock_hold(gl);
389                 else
390                         gfs2_glock_put(gl);
391         }
392
393         gl->gl_state = new_state;
394         gl->gl_tchange = jiffies;
395 }
396
397 static void gfs2_demote_wake(struct gfs2_glock *gl)
398 {
399         gl->gl_demote_state = LM_ST_EXCLUSIVE;
400         clear_bit(GLF_DEMOTE, &gl->gl_flags);
401         smp_mb__after_clear_bit();
402         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
403 }
404
405 /**
406  * finish_xmote - The DLM has replied to one of our lock requests
407  * @gl: The glock
408  * @ret: The status from the DLM
409  *
410  */
411
412 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
413 {
414         const struct gfs2_glock_operations *glops = gl->gl_ops;
415         struct gfs2_holder *gh;
416         unsigned state = ret & LM_OUT_ST_MASK;
417
418         spin_lock(&gl->gl_spin);
419         state_change(gl, state);
420         gh = find_first_waiter(gl);
421
422         /* Demote to UN request arrived during demote to SH or DF */
423         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
424             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
425                 gl->gl_target = LM_ST_UNLOCKED;
426
427         /* Check for state != intended state */
428         if (unlikely(state != gl->gl_target)) {
429                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
430                         /* move to back of queue and try next entry */
431                         if (ret & LM_OUT_CANCELED) {
432                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
433                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
434                                 gh = find_first_waiter(gl);
435                                 gl->gl_target = gh->gh_state;
436                                 goto retry;
437                         }
438                         /* Some error or failed "try lock" - report it */
439                         if ((ret & LM_OUT_ERROR) ||
440                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
441                                 gl->gl_target = gl->gl_state;
442                                 do_error(gl, ret);
443                                 goto out;
444                         }
445                 }
446                 switch(state) {
447                 /* Unlocked due to conversion deadlock, try again */
448                 case LM_ST_UNLOCKED:
449 retry:
450                         do_xmote(gl, gh, gl->gl_target);
451                         break;
452                 /* Conversion fails, unlock and try again */
453                 case LM_ST_SHARED:
454                 case LM_ST_DEFERRED:
455                         do_xmote(gl, gh, LM_ST_UNLOCKED);
456                         break;
457                 default: /* Everything else */
458                         printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
459                         GLOCK_BUG_ON(gl, 1);
460                 }
461                 spin_unlock(&gl->gl_spin);
462                 gfs2_glock_put(gl);
463                 return;
464         }
465
466         /* Fast path - we got what we asked for */
467         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
468                 gfs2_demote_wake(gl);
469         if (state != LM_ST_UNLOCKED) {
470                 if (glops->go_xmote_bh) {
471                         int rv;
472                         spin_unlock(&gl->gl_spin);
473                         rv = glops->go_xmote_bh(gl, gh);
474                         if (rv == -EAGAIN)
475                                 return;
476                         spin_lock(&gl->gl_spin);
477                         if (rv) {
478                                 do_error(gl, rv);
479                                 goto out;
480                         }
481                 }
482                 do_promote(gl);
483         }
484 out:
485         clear_bit(GLF_LOCK, &gl->gl_flags);
486         spin_unlock(&gl->gl_spin);
487         gfs2_glock_put(gl);
488 }
489
490 static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
491                                  unsigned int cur_state, unsigned int req_state,
492                                  unsigned int flags)
493 {
494         int ret = LM_OUT_ERROR;
495
496         if (!sdp->sd_lockstruct.ls_ops->lm_lock)
497                 return req_state == LM_ST_UNLOCKED ? 0 : req_state;
498
499         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
500                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock, cur_state,
501                                                          req_state, flags);
502         return ret;
503 }
504
505 /**
506  * do_xmote - Calls the DLM to change the state of a lock
507  * @gl: The lock state
508  * @gh: The holder (only for promotes)
509  * @target: The target lock state
510  *
511  */
512
513 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
514 {
515         const struct gfs2_glock_operations *glops = gl->gl_ops;
516         struct gfs2_sbd *sdp = gl->gl_sbd;
517         unsigned int lck_flags = gh ? gh->gh_flags : 0;
518         int ret;
519
520         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
521                       LM_FLAG_PRIORITY);
522         BUG_ON(gl->gl_state == target);
523         BUG_ON(gl->gl_state == gl->gl_target);
524         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
525             glops->go_inval) {
526                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
527                 do_error(gl, 0); /* Fail queued try locks */
528         }
529         spin_unlock(&gl->gl_spin);
530         if (glops->go_xmote_th)
531                 glops->go_xmote_th(gl);
532         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
533                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
534         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
535
536         gfs2_glock_hold(gl);
537         if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
538             gl->gl_state == LM_ST_DEFERRED) &&
539             !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
540                 lck_flags |= LM_FLAG_TRY_1CB;
541         ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, target, lck_flags);
542
543         if (!(ret & LM_OUT_ASYNC)) {
544                 finish_xmote(gl, ret);
545                 gfs2_glock_hold(gl);
546                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
547                         gfs2_glock_put(gl);
548         } else {
549                 GLOCK_BUG_ON(gl, ret != LM_OUT_ASYNC);
550         }
551         spin_lock(&gl->gl_spin);
552 }
553
554 /**
555  * find_first_holder - find the first "holder" gh
556  * @gl: the glock
557  */
558
559 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
560 {
561         struct gfs2_holder *gh;
562
563         if (!list_empty(&gl->gl_holders)) {
564                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
565                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
566                         return gh;
567         }
568         return NULL;
569 }
570
571 /**
572  * run_queue - do all outstanding tasks related to a glock
573  * @gl: The glock in question
574  * @nonblock: True if we must not block in run_queue
575  *
576  */
577
578 static void run_queue(struct gfs2_glock *gl, const int nonblock)
579 {
580         struct gfs2_holder *gh = NULL;
581
582         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
583                 return;
584
585         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
586
587         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
588             gl->gl_demote_state != gl->gl_state) {
589                 if (find_first_holder(gl))
590                         goto out;
591                 if (nonblock)
592                         goto out_sched;
593                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
594                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
595                 gl->gl_target = gl->gl_demote_state;
596         } else {
597                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
598                         gfs2_demote_wake(gl);
599                 if (do_promote(gl) == 0)
600                         goto out;
601                 gh = find_first_waiter(gl);
602                 gl->gl_target = gh->gh_state;
603                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
604                         do_error(gl, 0); /* Fail queued try locks */
605         }
606         do_xmote(gl, gh, gl->gl_target);
607         return;
608
609 out_sched:
610         gfs2_glock_hold(gl);
611         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
612                 gfs2_glock_put(gl);
613 out:
614         clear_bit(GLF_LOCK, &gl->gl_flags);
615 }
616
617 static void glock_work_func(struct work_struct *work)
618 {
619         unsigned long delay = 0;
620         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
621
622         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags))
623                 finish_xmote(gl, gl->gl_reply);
624         spin_lock(&gl->gl_spin);
625         if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
626             gl->gl_state != LM_ST_UNLOCKED &&
627             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
628                 unsigned long holdtime, now = jiffies;
629                 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
630                 if (time_before(now, holdtime))
631                         delay = holdtime - now;
632                 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
633         }
634         run_queue(gl, 0);
635         spin_unlock(&gl->gl_spin);
636         if (!delay ||
637             queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
638                 gfs2_glock_put(gl);
639 }
640
641 static int gfs2_lm_get_lock(struct gfs2_sbd *sdp, struct lm_lockname *name,
642                      void **lockp)
643 {
644         int error = -EIO;
645         if (!sdp->sd_lockstruct.ls_ops->lm_get_lock)
646                 return 0;
647         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
648                 error = sdp->sd_lockstruct.ls_ops->lm_get_lock(
649                                 sdp->sd_lockstruct.ls_lockspace, name, lockp);
650         return error;
651 }
652
653 /**
654  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
655  * @sdp: The GFS2 superblock
656  * @number: the lock number
657  * @glops: The glock_operations to use
658  * @create: If 0, don't create the glock if it doesn't exist
659  * @glp: the glock is returned here
660  *
661  * This does not lock a glock, just finds/creates structures for one.
662  *
663  * Returns: errno
664  */
665
666 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
667                    const struct gfs2_glock_operations *glops, int create,
668                    struct gfs2_glock **glp)
669 {
670         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
671         struct gfs2_glock *gl, *tmp;
672         unsigned int hash = gl_hash(sdp, &name);
673         int error;
674
675         read_lock(gl_lock_addr(hash));
676         gl = search_bucket(hash, sdp, &name);
677         read_unlock(gl_lock_addr(hash));
678
679         if (gl || !create) {
680                 *glp = gl;
681                 return 0;
682         }
683
684         gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
685         if (!gl)
686                 return -ENOMEM;
687
688         gl->gl_flags = 0;
689         gl->gl_name = name;
690         atomic_set(&gl->gl_ref, 1);
691         gl->gl_state = LM_ST_UNLOCKED;
692         gl->gl_target = LM_ST_UNLOCKED;
693         gl->gl_demote_state = LM_ST_EXCLUSIVE;
694         gl->gl_hash = hash;
695         gl->gl_ops = glops;
696         gl->gl_stamp = jiffies;
697         gl->gl_tchange = jiffies;
698         gl->gl_object = NULL;
699         gl->gl_sbd = sdp;
700         gl->gl_aspace = NULL;
701         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
702
703         /* If this glock protects actual on-disk data or metadata blocks,
704            create a VFS inode to manage the pages/buffers holding them. */
705         if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
706                 gl->gl_aspace = gfs2_aspace_get(sdp);
707                 if (!gl->gl_aspace) {
708                         error = -ENOMEM;
709                         goto fail;
710                 }
711         }
712
713         error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
714         if (error)
715                 goto fail_aspace;
716
717         write_lock(gl_lock_addr(hash));
718         tmp = search_bucket(hash, sdp, &name);
719         if (tmp) {
720                 write_unlock(gl_lock_addr(hash));
721                 glock_free(gl);
722                 gl = tmp;
723         } else {
724                 hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
725                 write_unlock(gl_lock_addr(hash));
726         }
727
728         *glp = gl;
729
730         return 0;
731
732 fail_aspace:
733         if (gl->gl_aspace)
734                 gfs2_aspace_put(gl->gl_aspace);
735 fail:
736         kmem_cache_free(gfs2_glock_cachep, gl);
737         return error;
738 }
739
740 /**
741  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
742  * @gl: the glock
743  * @state: the state we're requesting
744  * @flags: the modifier flags
745  * @gh: the holder structure
746  *
747  */
748
749 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
750                       struct gfs2_holder *gh)
751 {
752         INIT_LIST_HEAD(&gh->gh_list);
753         gh->gh_gl = gl;
754         gh->gh_ip = (unsigned long)__builtin_return_address(0);
755         gh->gh_owner_pid = get_pid(task_pid(current));
756         gh->gh_state = state;
757         gh->gh_flags = flags;
758         gh->gh_error = 0;
759         gh->gh_iflags = 0;
760         gfs2_glock_hold(gl);
761 }
762
763 /**
764  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
765  * @state: the state we're requesting
766  * @flags: the modifier flags
767  * @gh: the holder structure
768  *
769  * Don't mess with the glock.
770  *
771  */
772
773 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
774 {
775         gh->gh_state = state;
776         gh->gh_flags = flags;
777         gh->gh_iflags = 0;
778         gh->gh_ip = (unsigned long)__builtin_return_address(0);
779 }
780
781 /**
782  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
783  * @gh: the holder structure
784  *
785  */
786
787 void gfs2_holder_uninit(struct gfs2_holder *gh)
788 {
789         put_pid(gh->gh_owner_pid);
790         gfs2_glock_put(gh->gh_gl);
791         gh->gh_gl = NULL;
792         gh->gh_ip = 0;
793 }
794
795 static int just_schedule(void *word)
796 {
797         schedule();
798         return 0;
799 }
800
801 static void wait_on_holder(struct gfs2_holder *gh)
802 {
803         might_sleep();
804         wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
805 }
806
807 static void wait_on_demote(struct gfs2_glock *gl)
808 {
809         might_sleep();
810         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
811 }
812
813 /**
814  * handle_callback - process a demote request
815  * @gl: the glock
816  * @state: the state the caller wants us to change to
817  *
818  * There are only two requests that we are going to see in actual
819  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
820  */
821
822 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
823                             int remote, unsigned long delay)
824 {
825         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
826
827         set_bit(bit, &gl->gl_flags);
828         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
829                 gl->gl_demote_state = state;
830                 gl->gl_demote_time = jiffies;
831                 if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
832                     gl->gl_object)
833                         gfs2_glock_schedule_for_reclaim(gl);
834         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
835                         gl->gl_demote_state != state) {
836                 gl->gl_demote_state = LM_ST_UNLOCKED;
837         }
838 }
839
840 /**
841  * gfs2_glock_wait - wait on a glock acquisition
842  * @gh: the glock holder
843  *
844  * Returns: 0 on success
845  */
846
847 int gfs2_glock_wait(struct gfs2_holder *gh)
848 {
849         wait_on_holder(gh);
850         return gh->gh_error;
851 }
852
853 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
854 {
855         va_list args;
856
857         va_start(args, fmt);
858         if (seq) {
859                 struct gfs2_glock_iter *gi = seq->private;
860                 vsprintf(gi->string, fmt, args);
861                 seq_printf(seq, gi->string);
862         } else {
863                 printk(KERN_ERR " ");
864                 vprintk(fmt, args);
865         }
866         va_end(args);
867 }
868
869 /**
870  * add_to_queue - Add a holder to the wait queue (but look for recursion)
871  * @gh: the holder structure to add
872  *
873  * Eventually we should move the recursive locking trap to a
874  * debugging option or something like that. This is the fast
875  * path and needs to have the minimum number of distractions.
876  * 
877  */
878
879 static inline void add_to_queue(struct gfs2_holder *gh)
880 {
881         struct gfs2_glock *gl = gh->gh_gl;
882         struct gfs2_sbd *sdp = gl->gl_sbd;
883         struct list_head *insert_pt = NULL;
884         struct gfs2_holder *gh2;
885         int try_lock = 0;
886
887         BUG_ON(gh->gh_owner_pid == NULL);
888         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
889                 BUG();
890
891         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
892                 if (test_bit(GLF_LOCK, &gl->gl_flags))
893                         try_lock = 1;
894                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
895                         goto fail;
896         }
897
898         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
899                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
900                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
901                         goto trap_recursive;
902                 if (try_lock &&
903                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
904                     !may_grant(gl, gh)) {
905 fail:
906                         gh->gh_error = GLR_TRYFAILED;
907                         gfs2_holder_wake(gh);
908                         return;
909                 }
910                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
911                         continue;
912                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
913                         insert_pt = &gh2->gh_list;
914         }
915         if (likely(insert_pt == NULL)) {
916                 list_add_tail(&gh->gh_list, &gl->gl_holders);
917                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
918                         goto do_cancel;
919                 return;
920         }
921         list_add_tail(&gh->gh_list, insert_pt);
922 do_cancel:
923         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
924         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
925                 spin_unlock(&gl->gl_spin);
926                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
927                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl->gl_lock);
928                 spin_lock(&gl->gl_spin);
929         }
930         return;
931
932 trap_recursive:
933         print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
934         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
935         printk(KERN_ERR "lock type: %d req lock state : %d\n",
936                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
937         print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
938         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
939         printk(KERN_ERR "lock type: %d req lock state : %d\n",
940                gh->gh_gl->gl_name.ln_type, gh->gh_state);
941         __dump_glock(NULL, gl);
942         BUG();
943 }
944
945 /**
946  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
947  * @gh: the holder structure
948  *
949  * if (gh->gh_flags & GL_ASYNC), this never returns an error
950  *
951  * Returns: 0, GLR_TRYFAILED, or errno on failure
952  */
953
954 int gfs2_glock_nq(struct gfs2_holder *gh)
955 {
956         struct gfs2_glock *gl = gh->gh_gl;
957         struct gfs2_sbd *sdp = gl->gl_sbd;
958         int error = 0;
959
960         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
961                 return -EIO;
962
963         spin_lock(&gl->gl_spin);
964         add_to_queue(gh);
965         run_queue(gl, 1);
966         spin_unlock(&gl->gl_spin);
967
968         if (!(gh->gh_flags & GL_ASYNC))
969                 error = gfs2_glock_wait(gh);
970
971         return error;
972 }
973
974 /**
975  * gfs2_glock_poll - poll to see if an async request has been completed
976  * @gh: the holder
977  *
978  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
979  */
980
981 int gfs2_glock_poll(struct gfs2_holder *gh)
982 {
983         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
984 }
985
986 /**
987  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
988  * @gh: the glock holder
989  *
990  */
991
992 void gfs2_glock_dq(struct gfs2_holder *gh)
993 {
994         struct gfs2_glock *gl = gh->gh_gl;
995         const struct gfs2_glock_operations *glops = gl->gl_ops;
996         unsigned delay = 0;
997         int fast_path = 0;
998
999         spin_lock(&gl->gl_spin);
1000         if (gh->gh_flags & GL_NOCACHE)
1001                 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1002
1003         list_del_init(&gh->gh_list);
1004         if (find_first_holder(gl) == NULL) {
1005                 if (glops->go_unlock) {
1006                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1007                         spin_unlock(&gl->gl_spin);
1008                         glops->go_unlock(gh);
1009                         spin_lock(&gl->gl_spin);
1010                         clear_bit(GLF_LOCK, &gl->gl_flags);
1011                 }
1012                 gl->gl_stamp = jiffies;
1013                 if (list_empty(&gl->gl_holders) &&
1014                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1015                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1016                         fast_path = 1;
1017         }
1018         spin_unlock(&gl->gl_spin);
1019         if (likely(fast_path))
1020                 return;
1021
1022         gfs2_glock_hold(gl);
1023         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1024             !test_bit(GLF_DEMOTE, &gl->gl_flags))
1025                 delay = gl->gl_ops->go_min_hold_time;
1026         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1027                 gfs2_glock_put(gl);
1028 }
1029
1030 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1031 {
1032         struct gfs2_glock *gl = gh->gh_gl;
1033         gfs2_glock_dq(gh);
1034         wait_on_demote(gl);
1035 }
1036
1037 /**
1038  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1039  * @gh: the holder structure
1040  *
1041  */
1042
1043 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1044 {
1045         gfs2_glock_dq(gh);
1046         gfs2_holder_uninit(gh);
1047 }
1048
1049 /**
1050  * gfs2_glock_nq_num - acquire a glock based on lock number
1051  * @sdp: the filesystem
1052  * @number: the lock number
1053  * @glops: the glock operations for the type of glock
1054  * @state: the state to acquire the glock in
1055  * @flags: modifier flags for the aquisition
1056  * @gh: the struct gfs2_holder
1057  *
1058  * Returns: errno
1059  */
1060
1061 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1062                       const struct gfs2_glock_operations *glops,
1063                       unsigned int state, int flags, struct gfs2_holder *gh)
1064 {
1065         struct gfs2_glock *gl;
1066         int error;
1067
1068         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1069         if (!error) {
1070                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1071                 gfs2_glock_put(gl);
1072         }
1073
1074         return error;
1075 }
1076
1077 /**
1078  * glock_compare - Compare two struct gfs2_glock structures for sorting
1079  * @arg_a: the first structure
1080  * @arg_b: the second structure
1081  *
1082  */
1083
1084 static int glock_compare(const void *arg_a, const void *arg_b)
1085 {
1086         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1087         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1088         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1089         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1090
1091         if (a->ln_number > b->ln_number)
1092                 return 1;
1093         if (a->ln_number < b->ln_number)
1094                 return -1;
1095         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1096         return 0;
1097 }
1098
1099 /**
1100  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1101  * @num_gh: the number of structures
1102  * @ghs: an array of struct gfs2_holder structures
1103  *
1104  * Returns: 0 on success (all glocks acquired),
1105  *          errno on failure (no glocks acquired)
1106  */
1107
1108 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1109                      struct gfs2_holder **p)
1110 {
1111         unsigned int x;
1112         int error = 0;
1113
1114         for (x = 0; x < num_gh; x++)
1115                 p[x] = &ghs[x];
1116
1117         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1118
1119         for (x = 0; x < num_gh; x++) {
1120                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1121
1122                 error = gfs2_glock_nq(p[x]);
1123                 if (error) {
1124                         while (x--)
1125                                 gfs2_glock_dq(p[x]);
1126                         break;
1127                 }
1128         }
1129
1130         return error;
1131 }
1132
1133 /**
1134  * gfs2_glock_nq_m - acquire multiple glocks
1135  * @num_gh: the number of structures
1136  * @ghs: an array of struct gfs2_holder structures
1137  *
1138  *
1139  * Returns: 0 on success (all glocks acquired),
1140  *          errno on failure (no glocks acquired)
1141  */
1142
1143 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1144 {
1145         struct gfs2_holder *tmp[4];
1146         struct gfs2_holder **pph = tmp;
1147         int error = 0;
1148
1149         switch(num_gh) {
1150         case 0:
1151                 return 0;
1152         case 1:
1153                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1154                 return gfs2_glock_nq(ghs);
1155         default:
1156                 if (num_gh <= 4)
1157                         break;
1158                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1159                 if (!pph)
1160                         return -ENOMEM;
1161         }
1162
1163         error = nq_m_sync(num_gh, ghs, pph);
1164
1165         if (pph != tmp)
1166                 kfree(pph);
1167
1168         return error;
1169 }
1170
1171 /**
1172  * gfs2_glock_dq_m - release multiple glocks
1173  * @num_gh: the number of structures
1174  * @ghs: an array of struct gfs2_holder structures
1175  *
1176  */
1177
1178 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1179 {
1180         unsigned int x;
1181
1182         for (x = 0; x < num_gh; x++)
1183                 gfs2_glock_dq(&ghs[x]);
1184 }
1185
1186 /**
1187  * gfs2_glock_dq_uninit_m - release multiple glocks
1188  * @num_gh: the number of structures
1189  * @ghs: an array of struct gfs2_holder structures
1190  *
1191  */
1192
1193 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1194 {
1195         unsigned int x;
1196
1197         for (x = 0; x < num_gh; x++)
1198                 gfs2_glock_dq_uninit(&ghs[x]);
1199 }
1200
1201 static int gfs2_lm_hold_lvb(struct gfs2_sbd *sdp, void *lock, char **lvbp)
1202 {
1203         int error = -EIO;
1204         if (!sdp->sd_lockstruct.ls_ops->lm_hold_lvb)
1205                 return 0;
1206         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1207                 error = sdp->sd_lockstruct.ls_ops->lm_hold_lvb(lock, lvbp);
1208         return error;
1209 }
1210
1211 /**
1212  * gfs2_lvb_hold - attach a LVB from a glock
1213  * @gl: The glock in question
1214  *
1215  */
1216
1217 int gfs2_lvb_hold(struct gfs2_glock *gl)
1218 {
1219         int error;
1220
1221         if (!atomic_read(&gl->gl_lvb_count)) {
1222                 error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
1223                 if (error) 
1224                         return error;
1225                 gfs2_glock_hold(gl);
1226         }
1227         atomic_inc(&gl->gl_lvb_count);
1228
1229         return 0;
1230 }
1231
1232 /**
1233  * gfs2_lvb_unhold - detach a LVB from a glock
1234  * @gl: The glock in question
1235  *
1236  */
1237
1238 void gfs2_lvb_unhold(struct gfs2_glock *gl)
1239 {
1240         struct gfs2_sbd *sdp = gl->gl_sbd;
1241
1242         gfs2_glock_hold(gl);
1243         gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
1244         if (atomic_dec_and_test(&gl->gl_lvb_count)) {
1245                 if (sdp->sd_lockstruct.ls_ops->lm_unhold_lvb)
1246                         sdp->sd_lockstruct.ls_ops->lm_unhold_lvb(gl->gl_lock, gl->gl_lvb);
1247                 gl->gl_lvb = NULL;
1248                 gfs2_glock_put(gl);
1249         }
1250         gfs2_glock_put(gl);
1251 }
1252
1253 static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
1254                         unsigned int state)
1255 {
1256         struct gfs2_glock *gl;
1257         unsigned long delay = 0;
1258         unsigned long holdtime;
1259         unsigned long now = jiffies;
1260
1261         gl = gfs2_glock_find(sdp, name);
1262         if (!gl)
1263                 return;
1264
1265         holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1266         if (time_before(now, holdtime))
1267                 delay = holdtime - now;
1268
1269         spin_lock(&gl->gl_spin);
1270         handle_callback(gl, state, 1, delay);
1271         spin_unlock(&gl->gl_spin);
1272         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1273                 gfs2_glock_put(gl);
1274 }
1275
1276 /**
1277  * gfs2_glock_cb - Callback used by locking module
1278  * @sdp: Pointer to the superblock
1279  * @type: Type of callback
1280  * @data: Type dependent data pointer
1281  *
1282  * Called by the locking module when it wants to tell us something.
1283  * Either we need to drop a lock, one of our ASYNC requests completed, or
1284  * a journal from another client needs to be recovered.
1285  */
1286
1287 void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
1288 {
1289         struct gfs2_sbd *sdp = cb_data;
1290
1291         switch (type) {
1292         case LM_CB_NEED_E:
1293                 blocking_cb(sdp, data, LM_ST_UNLOCKED);
1294                 return;
1295
1296         case LM_CB_NEED_D:
1297                 blocking_cb(sdp, data, LM_ST_DEFERRED);
1298                 return;
1299
1300         case LM_CB_NEED_S:
1301                 blocking_cb(sdp, data, LM_ST_SHARED);
1302                 return;
1303
1304         case LM_CB_ASYNC: {
1305                 struct lm_async_cb *async = data;
1306                 struct gfs2_glock *gl;
1307
1308                 down_read(&gfs2_umount_flush_sem);
1309                 gl = gfs2_glock_find(sdp, &async->lc_name);
1310                 if (gfs2_assert_warn(sdp, gl))
1311                         return;
1312                 gl->gl_reply = async->lc_ret;
1313                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1314                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1315                         gfs2_glock_put(gl);
1316                 up_read(&gfs2_umount_flush_sem);
1317                 return;
1318         }
1319
1320         case LM_CB_NEED_RECOVERY:
1321                 gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
1322                 if (sdp->sd_recoverd_process)
1323                         wake_up_process(sdp->sd_recoverd_process);
1324                 return;
1325
1326         default:
1327                 gfs2_assert_warn(sdp, 0);
1328                 return;
1329         }
1330 }
1331
1332 /**
1333  * demote_ok - Check to see if it's ok to unlock a glock
1334  * @gl: the glock
1335  *
1336  * Returns: 1 if it's ok
1337  */
1338
1339 static int demote_ok(struct gfs2_glock *gl)
1340 {
1341         const struct gfs2_glock_operations *glops = gl->gl_ops;
1342         int demote = 1;
1343
1344         if (test_bit(GLF_STICKY, &gl->gl_flags))
1345                 demote = 0;
1346         else if (glops->go_demote_ok)
1347                 demote = glops->go_demote_ok(gl);
1348
1349         return demote;
1350 }
1351
1352 /**
1353  * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
1354  * @gl: the glock
1355  *
1356  */
1357
1358 void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
1359 {
1360         struct gfs2_sbd *sdp = gl->gl_sbd;
1361
1362         spin_lock(&sdp->sd_reclaim_lock);
1363         if (list_empty(&gl->gl_reclaim)) {
1364                 gfs2_glock_hold(gl);
1365                 list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
1366                 atomic_inc(&sdp->sd_reclaim_count);
1367                 spin_unlock(&sdp->sd_reclaim_lock);
1368                 wake_up(&sdp->sd_reclaim_wq);
1369         } else
1370                 spin_unlock(&sdp->sd_reclaim_lock);
1371 }
1372
1373 /**
1374  * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
1375  * @sdp: the filesystem
1376  *
1377  * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
1378  * different glock and we notice that there are a lot of glocks in the
1379  * reclaim list.
1380  *
1381  */
1382
1383 void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
1384 {
1385         struct gfs2_glock *gl;
1386         int done_callback = 0;
1387
1388         spin_lock(&sdp->sd_reclaim_lock);
1389         if (list_empty(&sdp->sd_reclaim_list)) {
1390                 spin_unlock(&sdp->sd_reclaim_lock);
1391                 return;
1392         }
1393         gl = list_entry(sdp->sd_reclaim_list.next,
1394                         struct gfs2_glock, gl_reclaim);
1395         list_del_init(&gl->gl_reclaim);
1396         spin_unlock(&sdp->sd_reclaim_lock);
1397
1398         atomic_dec(&sdp->sd_reclaim_count);
1399         atomic_inc(&sdp->sd_reclaimed);
1400
1401         spin_lock(&gl->gl_spin);
1402         if (find_first_holder(gl) == NULL &&
1403             gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl)) {
1404                 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1405                 done_callback = 1;
1406         }
1407         spin_unlock(&gl->gl_spin);
1408         if (!done_callback ||
1409             queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1410                 gfs2_glock_put(gl);
1411 }
1412
1413 /**
1414  * examine_bucket - Call a function for glock in a hash bucket
1415  * @examiner: the function
1416  * @sdp: the filesystem
1417  * @bucket: the bucket
1418  *
1419  * Returns: 1 if the bucket has entries
1420  */
1421
1422 static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
1423                           unsigned int hash)
1424 {
1425         struct gfs2_glock *gl, *prev = NULL;
1426         int has_entries = 0;
1427         struct hlist_head *head = &gl_hash_table[hash].hb_list;
1428
1429         read_lock(gl_lock_addr(hash));
1430         /* Can't use hlist_for_each_entry - don't want prefetch here */
1431         if (hlist_empty(head))
1432                 goto out;
1433         gl = list_entry(head->first, struct gfs2_glock, gl_list);
1434         while(1) {
1435                 if (!sdp || gl->gl_sbd == sdp) {
1436                         gfs2_glock_hold(gl);
1437                         read_unlock(gl_lock_addr(hash));
1438                         if (prev)
1439                                 gfs2_glock_put(prev);
1440                         prev = gl;
1441                         examiner(gl);
1442                         has_entries = 1;
1443                         read_lock(gl_lock_addr(hash));
1444                 }
1445                 if (gl->gl_list.next == NULL)
1446                         break;
1447                 gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
1448         }
1449 out:
1450         read_unlock(gl_lock_addr(hash));
1451         if (prev)
1452                 gfs2_glock_put(prev);
1453         cond_resched();
1454         return has_entries;
1455 }
1456
1457 /**
1458  * scan_glock - look at a glock and see if we can reclaim it
1459  * @gl: the glock to look at
1460  *
1461  */
1462
1463 static void scan_glock(struct gfs2_glock *gl)
1464 {
1465         if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
1466                 return;
1467         if (test_bit(GLF_LOCK, &gl->gl_flags))
1468                 return;
1469
1470         spin_lock(&gl->gl_spin);
1471         if (find_first_holder(gl) == NULL &&
1472             gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
1473                 gfs2_glock_schedule_for_reclaim(gl);
1474         spin_unlock(&gl->gl_spin);
1475 }
1476
1477 /**
1478  * clear_glock - look at a glock and see if we can free it from glock cache
1479  * @gl: the glock to look at
1480  *
1481  */
1482
1483 static void clear_glock(struct gfs2_glock *gl)
1484 {
1485         struct gfs2_sbd *sdp = gl->gl_sbd;
1486         int released;
1487
1488         spin_lock(&sdp->sd_reclaim_lock);
1489         if (!list_empty(&gl->gl_reclaim)) {
1490                 list_del_init(&gl->gl_reclaim);
1491                 atomic_dec(&sdp->sd_reclaim_count);
1492                 spin_unlock(&sdp->sd_reclaim_lock);
1493                 released = gfs2_glock_put(gl);
1494                 gfs2_assert(sdp, !released);
1495         } else {
1496                 spin_unlock(&sdp->sd_reclaim_lock);
1497         }
1498
1499         spin_lock(&gl->gl_spin);
1500         if (find_first_holder(gl) == NULL && gl->gl_state != LM_ST_UNLOCKED)
1501                 handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
1502         spin_unlock(&gl->gl_spin);
1503         gfs2_glock_hold(gl);
1504         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1505                 gfs2_glock_put(gl);
1506 }
1507
1508 /**
1509  * gfs2_gl_hash_clear - Empty out the glock hash table
1510  * @sdp: the filesystem
1511  * @wait: wait until it's all gone
1512  *
1513  * Called when unmounting the filesystem.
1514  */
1515
1516 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1517 {
1518         unsigned long t;
1519         unsigned int x;
1520         int cont;
1521
1522         t = jiffies;
1523
1524         for (;;) {
1525                 cont = 0;
1526                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1527                         if (examine_bucket(clear_glock, sdp, x))
1528                                 cont = 1;
1529                 }
1530
1531                 if (!cont)
1532                         break;
1533
1534                 if (time_after_eq(jiffies,
1535                                   t + gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
1536                         fs_warn(sdp, "Unmount seems to be stalled. "
1537                                      "Dumping lock state...\n");
1538                         gfs2_dump_lockstate(sdp);
1539                         t = jiffies;
1540                 }
1541
1542                 down_write(&gfs2_umount_flush_sem);
1543                 invalidate_inodes(sdp->sd_vfs);
1544                 up_write(&gfs2_umount_flush_sem);
1545                 msleep(10);
1546         }
1547 }
1548
1549 static const char *state2str(unsigned state)
1550 {
1551         switch(state) {
1552         case LM_ST_UNLOCKED:
1553                 return "UN";
1554         case LM_ST_SHARED:
1555                 return "SH";
1556         case LM_ST_DEFERRED:
1557                 return "DF";
1558         case LM_ST_EXCLUSIVE:
1559                 return "EX";
1560         }
1561         return "??";
1562 }
1563
1564 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1565 {
1566         char *p = buf;
1567         if (flags & LM_FLAG_TRY)
1568                 *p++ = 't';
1569         if (flags & LM_FLAG_TRY_1CB)
1570                 *p++ = 'T';
1571         if (flags & LM_FLAG_NOEXP)
1572                 *p++ = 'e';
1573         if (flags & LM_FLAG_ANY)
1574                 *p++ = 'a';
1575         if (flags & LM_FLAG_PRIORITY)
1576                 *p++ = 'p';
1577         if (flags & GL_ASYNC)
1578                 *p++ = 'a';
1579         if (flags & GL_EXACT)
1580                 *p++ = 'E';
1581         if (flags & GL_ATIME)
1582                 *p++ = 'a';
1583         if (flags & GL_NOCACHE)
1584                 *p++ = 'c';
1585         if (test_bit(HIF_HOLDER, &iflags))
1586                 *p++ = 'H';
1587         if (test_bit(HIF_WAIT, &iflags))
1588                 *p++ = 'W';
1589         if (test_bit(HIF_FIRST, &iflags))
1590                 *p++ = 'F';
1591         *p = 0;
1592         return buf;
1593 }
1594
1595 /**
1596  * dump_holder - print information about a glock holder
1597  * @seq: the seq_file struct
1598  * @gh: the glock holder
1599  *
1600  * Returns: 0 on success, -ENOBUFS when we run out of space
1601  */
1602
1603 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1604 {
1605         struct task_struct *gh_owner = NULL;
1606         char buffer[KSYM_SYMBOL_LEN];
1607         char flags_buf[32];
1608
1609         sprint_symbol(buffer, gh->gh_ip);
1610         if (gh->gh_owner_pid)
1611                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1612         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
1613                   state2str(gh->gh_state),
1614                   hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1615                   gh->gh_error, 
1616                   gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1617                   gh_owner ? gh_owner->comm : "(ended)", buffer);
1618         return 0;
1619 }
1620
1621 static const char *gflags2str(char *buf, const unsigned long *gflags)
1622 {
1623         char *p = buf;
1624         if (test_bit(GLF_LOCK, gflags))
1625                 *p++ = 'l';
1626         if (test_bit(GLF_STICKY, gflags))
1627                 *p++ = 's';
1628         if (test_bit(GLF_DEMOTE, gflags))
1629                 *p++ = 'D';
1630         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1631                 *p++ = 'd';
1632         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1633                 *p++ = 'p';
1634         if (test_bit(GLF_DIRTY, gflags))
1635                 *p++ = 'y';
1636         if (test_bit(GLF_LFLUSH, gflags))
1637                 *p++ = 'f';
1638         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1639                 *p++ = 'i';
1640         if (test_bit(GLF_REPLY_PENDING, gflags))
1641                 *p++ = 'r';
1642         *p = 0;
1643         return buf;
1644 }
1645
1646 /**
1647  * __dump_glock - print information about a glock
1648  * @seq: The seq_file struct
1649  * @gl: the glock
1650  *
1651  * The file format is as follows:
1652  * One line per object, capital letters are used to indicate objects
1653  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1654  * other objects are indented by a single space and follow the glock to
1655  * which they are related. Fields are indicated by lower case letters
1656  * followed by a colon and the field value, except for strings which are in
1657  * [] so that its possible to see if they are composed of spaces for
1658  * example. The field's are n = number (id of the object), f = flags,
1659  * t = type, s = state, r = refcount, e = error, p = pid.
1660  *
1661  * Returns: 0 on success, -ENOBUFS when we run out of space
1662  */
1663
1664 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1665 {
1666         const struct gfs2_glock_operations *glops = gl->gl_ops;
1667         unsigned long long dtime;
1668         const struct gfs2_holder *gh;
1669         char gflags_buf[32];
1670         int error = 0;
1671
1672         dtime = jiffies - gl->gl_demote_time;
1673         dtime *= 1000000/HZ; /* demote time in uSec */
1674         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1675                 dtime = 0;
1676         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llu f:%s t:%s d:%s/%llu l:%d a:%d r:%d\n",
1677                   state2str(gl->gl_state),
1678                   gl->gl_name.ln_type,
1679                   (unsigned long long)gl->gl_name.ln_number,
1680                   gflags2str(gflags_buf, &gl->gl_flags),
1681                   state2str(gl->gl_target),
1682                   state2str(gl->gl_demote_state), dtime,
1683                   atomic_read(&gl->gl_lvb_count),
1684                   atomic_read(&gl->gl_ail_count),
1685                   atomic_read(&gl->gl_ref));
1686
1687         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1688                 error = dump_holder(seq, gh);
1689                 if (error)
1690                         goto out;
1691         }
1692         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1693                 error = glops->go_dump(seq, gl);
1694 out:
1695         return error;
1696 }
1697
1698 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1699 {
1700         int ret;
1701         spin_lock(&gl->gl_spin);
1702         ret = __dump_glock(seq, gl);
1703         spin_unlock(&gl->gl_spin);
1704         return ret;
1705 }
1706
1707 /**
1708  * gfs2_dump_lockstate - print out the current lockstate
1709  * @sdp: the filesystem
1710  * @ub: the buffer to copy the information into
1711  *
1712  * If @ub is NULL, dump the lockstate to the console.
1713  *
1714  */
1715
1716 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
1717 {
1718         struct gfs2_glock *gl;
1719         struct hlist_node *h;
1720         unsigned int x;
1721         int error = 0;
1722
1723         for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
1724
1725                 read_lock(gl_lock_addr(x));
1726
1727                 hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
1728                         if (gl->gl_sbd != sdp)
1729                                 continue;
1730
1731                         error = dump_glock(NULL, gl);
1732                         if (error)
1733                                 break;
1734                 }
1735
1736                 read_unlock(gl_lock_addr(x));
1737
1738                 if (error)
1739                         break;
1740         }
1741
1742
1743         return error;
1744 }
1745
1746 /**
1747  * gfs2_scand - Look for cached glocks and inodes to toss from memory
1748  * @sdp: Pointer to GFS2 superblock
1749  *
1750  * One of these daemons runs, finding candidates to add to sd_reclaim_list.
1751  * See gfs2_glockd()
1752  */
1753
1754 static int gfs2_scand(void *data)
1755 {
1756         unsigned x;
1757         unsigned delay;
1758
1759         while (!kthread_should_stop()) {
1760                 for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1761                         examine_bucket(scan_glock, NULL, x);
1762                 if (freezing(current))
1763                         refrigerator();
1764                 delay = scand_secs;
1765                 if (delay < 1)
1766                         delay = 1;
1767                 schedule_timeout_interruptible(delay * HZ);
1768         }
1769
1770         return 0;
1771 }
1772
1773
1774
1775 int __init gfs2_glock_init(void)
1776 {
1777         unsigned i;
1778         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1779                 INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
1780         }
1781 #ifdef GL_HASH_LOCK_SZ
1782         for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
1783                 rwlock_init(&gl_hash_locks[i]);
1784         }
1785 #endif
1786
1787         scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
1788         if (IS_ERR(scand_process))
1789                 return PTR_ERR(scand_process);
1790
1791         glock_workqueue = create_workqueue("glock_workqueue");
1792         if (IS_ERR(glock_workqueue)) {
1793                 kthread_stop(scand_process);
1794                 return PTR_ERR(glock_workqueue);
1795         }
1796
1797         return 0;
1798 }
1799
1800 void gfs2_glock_exit(void)
1801 {
1802         destroy_workqueue(glock_workqueue);
1803         kthread_stop(scand_process);
1804 }
1805
1806 module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
1807 MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
1808
1809 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1810 {
1811         struct gfs2_glock *gl;
1812
1813 restart:
1814         read_lock(gl_lock_addr(gi->hash));
1815         gl = gi->gl;
1816         if (gl) {
1817                 gi->gl = hlist_entry(gl->gl_list.next,
1818                                      struct gfs2_glock, gl_list);
1819                 if (gi->gl)
1820                         gfs2_glock_hold(gi->gl);
1821         }
1822         read_unlock(gl_lock_addr(gi->hash));
1823         if (gl)
1824                 gfs2_glock_put(gl);
1825         if (gl && gi->gl == NULL)
1826                 gi->hash++;
1827         while (gi->gl == NULL) {
1828                 if (gi->hash >= GFS2_GL_HASH_SIZE)
1829                         return 1;
1830                 read_lock(gl_lock_addr(gi->hash));
1831                 gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
1832                                      struct gfs2_glock, gl_list);
1833                 if (gi->gl)
1834                         gfs2_glock_hold(gi->gl);
1835                 read_unlock(gl_lock_addr(gi->hash));
1836                 gi->hash++;
1837         }
1838
1839         if (gi->sdp != gi->gl->gl_sbd)
1840                 goto restart;
1841
1842         return 0;
1843 }
1844
1845 static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
1846 {
1847         if (gi->gl)
1848                 gfs2_glock_put(gi->gl);
1849         gi->gl = NULL;
1850 }
1851
1852 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1853 {
1854         struct gfs2_glock_iter *gi = seq->private;
1855         loff_t n = *pos;
1856
1857         gi->hash = 0;
1858
1859         do {
1860                 if (gfs2_glock_iter_next(gi)) {
1861                         gfs2_glock_iter_free(gi);
1862                         return NULL;
1863                 }
1864         } while (n--);
1865
1866         return gi->gl;
1867 }
1868
1869 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1870                                  loff_t *pos)
1871 {
1872         struct gfs2_glock_iter *gi = seq->private;
1873
1874         (*pos)++;
1875
1876         if (gfs2_glock_iter_next(gi)) {
1877                 gfs2_glock_iter_free(gi);
1878                 return NULL;
1879         }
1880
1881         return gi->gl;
1882 }
1883
1884 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1885 {
1886         struct gfs2_glock_iter *gi = seq->private;
1887         gfs2_glock_iter_free(gi);
1888 }
1889
1890 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1891 {
1892         return dump_glock(seq, iter_ptr);
1893 }
1894
1895 static const struct seq_operations gfs2_glock_seq_ops = {
1896         .start = gfs2_glock_seq_start,
1897         .next  = gfs2_glock_seq_next,
1898         .stop  = gfs2_glock_seq_stop,
1899         .show  = gfs2_glock_seq_show,
1900 };
1901
1902 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1903 {
1904         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1905                                    sizeof(struct gfs2_glock_iter));
1906         if (ret == 0) {
1907                 struct seq_file *seq = file->private_data;
1908                 struct gfs2_glock_iter *gi = seq->private;
1909                 gi->sdp = inode->i_private;
1910         }
1911         return ret;
1912 }
1913
1914 static const struct file_operations gfs2_debug_fops = {
1915         .owner   = THIS_MODULE,
1916         .open    = gfs2_debugfs_open,
1917         .read    = seq_read,
1918         .llseek  = seq_lseek,
1919         .release = seq_release_private,
1920 };
1921
1922 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1923 {
1924         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1925         if (!sdp->debugfs_dir)
1926                 return -ENOMEM;
1927         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1928                                                          S_IFREG | S_IRUGO,
1929                                                          sdp->debugfs_dir, sdp,
1930                                                          &gfs2_debug_fops);
1931         if (!sdp->debugfs_dentry_glocks)
1932                 return -ENOMEM;
1933
1934         return 0;
1935 }
1936
1937 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1938 {
1939         if (sdp && sdp->debugfs_dir) {
1940                 if (sdp->debugfs_dentry_glocks) {
1941                         debugfs_remove(sdp->debugfs_dentry_glocks);
1942                         sdp->debugfs_dentry_glocks = NULL;
1943                 }
1944                 debugfs_remove(sdp->debugfs_dir);
1945                 sdp->debugfs_dir = NULL;
1946         }
1947 }
1948
1949 int gfs2_register_debugfs(void)
1950 {
1951         gfs2_root = debugfs_create_dir("gfs2", NULL);
1952         return gfs2_root ? 0 : -ENOMEM;
1953 }
1954
1955 void gfs2_unregister_debugfs(void)
1956 {
1957         debugfs_remove(gfs2_root);
1958         gfs2_root = NULL;
1959 }