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