Merge branches 'core-fixes-for-linus' and 'irq-fixes-for-linus' of git://git.kernel...
[pandora-kernel.git] / fs / gfs2 / glock.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/buffer_head.h>
14 #include <linux/delay.h>
15 #include <linux/sort.h>
16 #include <linux/jhash.h>
17 #include <linux/kallsyms.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/list.h>
20 #include <linux/wait.h>
21 #include <linux/module.h>
22 #include <asm/uaccess.h>
23 #include <linux/seq_file.h>
24 #include <linux/debugfs.h>
25 #include <linux/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/workqueue.h>
28 #include <linux/jiffies.h>
29 #include <linux/rcupdate.h>
30 #include <linux/rculist_bl.h>
31 #include <linux/bit_spinlock.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 #include "bmap.h"
44 #define CREATE_TRACE_POINTS
45 #include "trace_gfs2.h"
46
47 struct gfs2_glock_iter {
48         int hash;                       /* hash bucket index         */
49         struct gfs2_sbd *sdp;           /* incore superblock         */
50         struct gfs2_glock *gl;          /* current glock struct      */
51         char string[512];               /* scratch space             */
52 };
53
54 typedef void (*glock_examiner) (struct gfs2_glock * gl);
55
56 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
57 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
58 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
59
60 static struct dentry *gfs2_root;
61 static struct workqueue_struct *glock_workqueue;
62 struct workqueue_struct *gfs2_delete_workqueue;
63 static LIST_HEAD(lru_list);
64 static atomic_t lru_count = ATOMIC_INIT(0);
65 static DEFINE_SPINLOCK(lru_lock);
66
67 #define GFS2_GL_HASH_SHIFT      15
68 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
69 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
70
71 static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
72 static struct dentry *gfs2_root;
73
74 /**
75  * gl_hash() - Turn glock number into hash bucket number
76  * @lock: The glock number
77  *
78  * Returns: The number of the corresponding hash bucket
79  */
80
81 static unsigned int gl_hash(const struct gfs2_sbd *sdp,
82                             const struct lm_lockname *name)
83 {
84         unsigned int h;
85
86         h = jhash(&name->ln_number, sizeof(u64), 0);
87         h = jhash(&name->ln_type, sizeof(unsigned int), h);
88         h = jhash(&sdp, sizeof(struct gfs2_sbd *), h);
89         h &= GFS2_GL_HASH_MASK;
90
91         return h;
92 }
93
94 static inline void spin_lock_bucket(unsigned int hash)
95 {
96         hlist_bl_lock(&gl_hash_table[hash]);
97 }
98
99 static inline void spin_unlock_bucket(unsigned int hash)
100 {
101         hlist_bl_unlock(&gl_hash_table[hash]);
102 }
103
104 static void gfs2_glock_dealloc(struct rcu_head *rcu)
105 {
106         struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
107
108         if (gl->gl_ops->go_flags & GLOF_ASPACE)
109                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
110         else
111                 kmem_cache_free(gfs2_glock_cachep, gl);
112 }
113
114 void gfs2_glock_free(struct gfs2_glock *gl)
115 {
116         struct gfs2_sbd *sdp = gl->gl_sbd;
117
118         call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
119         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
120                 wake_up(&sdp->sd_glock_wait);
121 }
122
123 /**
124  * gfs2_glock_hold() - increment reference count on glock
125  * @gl: The glock to hold
126  *
127  */
128
129 void gfs2_glock_hold(struct gfs2_glock *gl)
130 {
131         GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
132         atomic_inc(&gl->gl_ref);
133 }
134
135 /**
136  * demote_ok - Check to see if it's ok to unlock a glock
137  * @gl: the glock
138  *
139  * Returns: 1 if it's ok
140  */
141
142 static int demote_ok(const struct gfs2_glock *gl)
143 {
144         const struct gfs2_glock_operations *glops = gl->gl_ops;
145
146         if (gl->gl_state == LM_ST_UNLOCKED)
147                 return 0;
148         if (!list_empty(&gl->gl_holders))
149                 return 0;
150         if (glops->go_demote_ok)
151                 return glops->go_demote_ok(gl);
152         return 1;
153 }
154
155
156 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
157 {
158         spin_lock(&lru_lock);
159
160         if (!list_empty(&gl->gl_lru))
161                 list_del_init(&gl->gl_lru);
162         else
163                 atomic_inc(&lru_count);
164
165         list_add_tail(&gl->gl_lru, &lru_list);
166         set_bit(GLF_LRU, &gl->gl_flags);
167         spin_unlock(&lru_lock);
168 }
169
170 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
171 {
172         spin_lock(&lru_lock);
173         if (!list_empty(&gl->gl_lru)) {
174                 list_del_init(&gl->gl_lru);
175                 atomic_dec(&lru_count);
176                 clear_bit(GLF_LRU, &gl->gl_flags);
177         }
178         spin_unlock(&lru_lock);
179 }
180
181 /**
182  * __gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
183  * @gl: the glock
184  *
185  * If the glock is demotable, then we add it (or move it) to the end
186  * of the glock LRU list.
187  */
188
189 static void __gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
190 {
191         if (demote_ok(gl))
192                 gfs2_glock_add_to_lru(gl);
193 }
194
195 /**
196  * gfs2_glock_put_nolock() - Decrement reference count on glock
197  * @gl: The glock to put
198  *
199  * This function should only be used if the caller has its own reference
200  * to the glock, in addition to the one it is dropping.
201  */
202
203 void gfs2_glock_put_nolock(struct gfs2_glock *gl)
204 {
205         if (atomic_dec_and_test(&gl->gl_ref))
206                 GLOCK_BUG_ON(gl, 1);
207 }
208
209 /**
210  * gfs2_glock_put() - Decrement reference count on glock
211  * @gl: The glock to put
212  *
213  */
214
215 void gfs2_glock_put(struct gfs2_glock *gl)
216 {
217         struct gfs2_sbd *sdp = gl->gl_sbd;
218         struct address_space *mapping = gfs2_glock2aspace(gl);
219
220         if (atomic_dec_and_test(&gl->gl_ref)) {
221                 spin_lock_bucket(gl->gl_hash);
222                 hlist_bl_del_rcu(&gl->gl_list);
223                 spin_unlock_bucket(gl->gl_hash);
224                 gfs2_glock_remove_from_lru(gl);
225                 GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
226                 GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
227                 trace_gfs2_glock_put(gl);
228                 sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
229         }
230 }
231
232 /**
233  * search_bucket() - Find struct gfs2_glock by lock number
234  * @bucket: the bucket to search
235  * @name: The lock name
236  *
237  * Returns: NULL, or the struct gfs2_glock with the requested number
238  */
239
240 static struct gfs2_glock *search_bucket(unsigned int hash,
241                                         const struct gfs2_sbd *sdp,
242                                         const struct lm_lockname *name)
243 {
244         struct gfs2_glock *gl;
245         struct hlist_bl_node *h;
246
247         hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
248                 if (!lm_name_equal(&gl->gl_name, name))
249                         continue;
250                 if (gl->gl_sbd != sdp)
251                         continue;
252                 if (atomic_inc_not_zero(&gl->gl_ref))
253                         return gl;
254         }
255
256         return NULL;
257 }
258
259 /**
260  * may_grant - check if its ok to grant a new lock
261  * @gl: The glock
262  * @gh: The lock request which we wish to grant
263  *
264  * Returns: true if its ok to grant the lock
265  */
266
267 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
268 {
269         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
270         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
271              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
272                 return 0;
273         if (gl->gl_state == gh->gh_state)
274                 return 1;
275         if (gh->gh_flags & GL_EXACT)
276                 return 0;
277         if (gl->gl_state == LM_ST_EXCLUSIVE) {
278                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
279                         return 1;
280                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
281                         return 1;
282         }
283         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
284                 return 1;
285         return 0;
286 }
287
288 static void gfs2_holder_wake(struct gfs2_holder *gh)
289 {
290         clear_bit(HIF_WAIT, &gh->gh_iflags);
291         smp_mb__after_clear_bit();
292         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
293 }
294
295 /**
296  * do_error - Something unexpected has happened during a lock request
297  *
298  */
299
300 static inline void do_error(struct gfs2_glock *gl, const int ret)
301 {
302         struct gfs2_holder *gh, *tmp;
303
304         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
305                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
306                         continue;
307                 if (ret & LM_OUT_ERROR)
308                         gh->gh_error = -EIO;
309                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
310                         gh->gh_error = GLR_TRYFAILED;
311                 else
312                         continue;
313                 list_del_init(&gh->gh_list);
314                 trace_gfs2_glock_queue(gh, 0);
315                 gfs2_holder_wake(gh);
316         }
317 }
318
319 /**
320  * do_promote - promote as many requests as possible on the current queue
321  * @gl: The glock
322  * 
323  * Returns: 1 if there is a blocked holder at the head of the list, or 2
324  *          if a type specific operation is underway.
325  */
326
327 static int do_promote(struct gfs2_glock *gl)
328 __releases(&gl->gl_spin)
329 __acquires(&gl->gl_spin)
330 {
331         const struct gfs2_glock_operations *glops = gl->gl_ops;
332         struct gfs2_holder *gh, *tmp;
333         int ret;
334
335 restart:
336         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
337                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
338                         continue;
339                 if (may_grant(gl, gh)) {
340                         if (gh->gh_list.prev == &gl->gl_holders &&
341                             glops->go_lock) {
342                                 spin_unlock(&gl->gl_spin);
343                                 /* FIXME: eliminate this eventually */
344                                 ret = glops->go_lock(gh);
345                                 spin_lock(&gl->gl_spin);
346                                 if (ret) {
347                                         if (ret == 1)
348                                                 return 2;
349                                         gh->gh_error = ret;
350                                         list_del_init(&gh->gh_list);
351                                         trace_gfs2_glock_queue(gh, 0);
352                                         gfs2_holder_wake(gh);
353                                         goto restart;
354                                 }
355                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
356                                 trace_gfs2_promote(gh, 1);
357                                 gfs2_holder_wake(gh);
358                                 goto restart;
359                         }
360                         set_bit(HIF_HOLDER, &gh->gh_iflags);
361                         trace_gfs2_promote(gh, 0);
362                         gfs2_holder_wake(gh);
363                         continue;
364                 }
365                 if (gh->gh_list.prev == &gl->gl_holders)
366                         return 1;
367                 do_error(gl, 0);
368                 break;
369         }
370         return 0;
371 }
372
373 /**
374  * find_first_waiter - find the first gh that's waiting for the glock
375  * @gl: the glock
376  */
377
378 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
379 {
380         struct gfs2_holder *gh;
381
382         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
383                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
384                         return gh;
385         }
386         return NULL;
387 }
388
389 /**
390  * state_change - record that the glock is now in a different state
391  * @gl: the glock
392  * @new_state the new state
393  *
394  */
395
396 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
397 {
398         int held1, held2;
399
400         held1 = (gl->gl_state != LM_ST_UNLOCKED);
401         held2 = (new_state != LM_ST_UNLOCKED);
402
403         if (held1 != held2) {
404                 if (held2)
405                         gfs2_glock_hold(gl);
406                 else
407                         gfs2_glock_put_nolock(gl);
408         }
409         if (held1 && held2 && list_empty(&gl->gl_holders))
410                 clear_bit(GLF_QUEUED, &gl->gl_flags);
411
412         gl->gl_state = new_state;
413         gl->gl_tchange = jiffies;
414 }
415
416 static void gfs2_demote_wake(struct gfs2_glock *gl)
417 {
418         gl->gl_demote_state = LM_ST_EXCLUSIVE;
419         clear_bit(GLF_DEMOTE, &gl->gl_flags);
420         smp_mb__after_clear_bit();
421         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
422 }
423
424 /**
425  * finish_xmote - The DLM has replied to one of our lock requests
426  * @gl: The glock
427  * @ret: The status from the DLM
428  *
429  */
430
431 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
432 {
433         const struct gfs2_glock_operations *glops = gl->gl_ops;
434         struct gfs2_holder *gh;
435         unsigned state = ret & LM_OUT_ST_MASK;
436         int rv;
437
438         spin_lock(&gl->gl_spin);
439         trace_gfs2_glock_state_change(gl, state);
440         state_change(gl, state);
441         gh = find_first_waiter(gl);
442
443         /* Demote to UN request arrived during demote to SH or DF */
444         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
445             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
446                 gl->gl_target = LM_ST_UNLOCKED;
447
448         /* Check for state != intended state */
449         if (unlikely(state != gl->gl_target)) {
450                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
451                         /* move to back of queue and try next entry */
452                         if (ret & LM_OUT_CANCELED) {
453                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
454                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
455                                 gh = find_first_waiter(gl);
456                                 gl->gl_target = gh->gh_state;
457                                 goto retry;
458                         }
459                         /* Some error or failed "try lock" - report it */
460                         if ((ret & LM_OUT_ERROR) ||
461                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
462                                 gl->gl_target = gl->gl_state;
463                                 do_error(gl, ret);
464                                 goto out;
465                         }
466                 }
467                 switch(state) {
468                 /* Unlocked due to conversion deadlock, try again */
469                 case LM_ST_UNLOCKED:
470 retry:
471                         do_xmote(gl, gh, gl->gl_target);
472                         break;
473                 /* Conversion fails, unlock and try again */
474                 case LM_ST_SHARED:
475                 case LM_ST_DEFERRED:
476                         do_xmote(gl, gh, LM_ST_UNLOCKED);
477                         break;
478                 default: /* Everything else */
479                         printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
480                         GLOCK_BUG_ON(gl, 1);
481                 }
482                 spin_unlock(&gl->gl_spin);
483                 return;
484         }
485
486         /* Fast path - we got what we asked for */
487         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
488                 gfs2_demote_wake(gl);
489         if (state != LM_ST_UNLOCKED) {
490                 if (glops->go_xmote_bh) {
491                         spin_unlock(&gl->gl_spin);
492                         rv = glops->go_xmote_bh(gl, gh);
493                         spin_lock(&gl->gl_spin);
494                         if (rv) {
495                                 do_error(gl, rv);
496                                 goto out;
497                         }
498                 }
499                 rv = do_promote(gl);
500                 if (rv == 2)
501                         goto out_locked;
502         }
503 out:
504         clear_bit(GLF_LOCK, &gl->gl_flags);
505 out_locked:
506         spin_unlock(&gl->gl_spin);
507 }
508
509 /**
510  * do_xmote - Calls the DLM to change the state of a lock
511  * @gl: The lock state
512  * @gh: The holder (only for promotes)
513  * @target: The target lock state
514  *
515  */
516
517 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
518 __releases(&gl->gl_spin)
519 __acquires(&gl->gl_spin)
520 {
521         const struct gfs2_glock_operations *glops = gl->gl_ops;
522         struct gfs2_sbd *sdp = gl->gl_sbd;
523         unsigned int lck_flags = gh ? gh->gh_flags : 0;
524         int ret;
525
526         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
527                       LM_FLAG_PRIORITY);
528         GLOCK_BUG_ON(gl, gl->gl_state == target);
529         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
530         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
531             glops->go_inval) {
532                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
533                 do_error(gl, 0); /* Fail queued try locks */
534         }
535         gl->gl_req = target;
536         spin_unlock(&gl->gl_spin);
537         if (glops->go_xmote_th)
538                 glops->go_xmote_th(gl);
539         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
540                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
541         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
542
543         gfs2_glock_hold(gl);
544         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
545                 /* lock_dlm */
546                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
547                 GLOCK_BUG_ON(gl, ret);
548         } else { /* lock_nolock */
549                 finish_xmote(gl, target);
550                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
551                         gfs2_glock_put(gl);
552         }
553
554         spin_lock(&gl->gl_spin);
555 }
556
557 /**
558  * find_first_holder - find the first "holder" gh
559  * @gl: the glock
560  */
561
562 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
563 {
564         struct gfs2_holder *gh;
565
566         if (!list_empty(&gl->gl_holders)) {
567                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
568                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
569                         return gh;
570         }
571         return NULL;
572 }
573
574 /**
575  * run_queue - do all outstanding tasks related to a glock
576  * @gl: The glock in question
577  * @nonblock: True if we must not block in run_queue
578  *
579  */
580
581 static void run_queue(struct gfs2_glock *gl, const int nonblock)
582 __releases(&gl->gl_spin)
583 __acquires(&gl->gl_spin)
584 {
585         struct gfs2_holder *gh = NULL;
586         int ret;
587
588         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
589                 return;
590
591         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
592
593         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
594             gl->gl_demote_state != gl->gl_state) {
595                 if (find_first_holder(gl))
596                         goto out_unlock;
597                 if (nonblock)
598                         goto out_sched;
599                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
600                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
601                 gl->gl_target = gl->gl_demote_state;
602         } else {
603                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
604                         gfs2_demote_wake(gl);
605                 ret = do_promote(gl);
606                 if (ret == 0)
607                         goto out_unlock;
608                 if (ret == 2)
609                         goto out;
610                 gh = find_first_waiter(gl);
611                 gl->gl_target = gh->gh_state;
612                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
613                         do_error(gl, 0); /* Fail queued try locks */
614         }
615         do_xmote(gl, gh, gl->gl_target);
616 out:
617         return;
618
619 out_sched:
620         clear_bit(GLF_LOCK, &gl->gl_flags);
621         smp_mb__after_clear_bit();
622         gfs2_glock_hold(gl);
623         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
624                 gfs2_glock_put_nolock(gl);
625         return;
626
627 out_unlock:
628         clear_bit(GLF_LOCK, &gl->gl_flags);
629         smp_mb__after_clear_bit();
630         return;
631 }
632
633 static void delete_work_func(struct work_struct *work)
634 {
635         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
636         struct gfs2_sbd *sdp = gl->gl_sbd;
637         struct gfs2_inode *ip;
638         struct inode *inode;
639         u64 no_addr = gl->gl_name.ln_number;
640
641         ip = gl->gl_object;
642         /* Note: Unsafe to dereference ip as we don't hold right refs/locks */
643
644         if (ip)
645                 inode = gfs2_ilookup(sdp->sd_vfs, no_addr, 1);
646         else
647                 inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
648         if (inode && !IS_ERR(inode)) {
649                 d_prune_aliases(inode);
650                 iput(inode);
651         }
652         gfs2_glock_put(gl);
653 }
654
655 static void glock_work_func(struct work_struct *work)
656 {
657         unsigned long delay = 0;
658         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
659         int drop_ref = 0;
660
661         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
662                 finish_xmote(gl, gl->gl_reply);
663                 drop_ref = 1;
664         }
665         spin_lock(&gl->gl_spin);
666         if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
667             gl->gl_state != LM_ST_UNLOCKED &&
668             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
669                 unsigned long holdtime, now = jiffies;
670                 holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
671                 if (time_before(now, holdtime))
672                         delay = holdtime - now;
673                 set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
674         }
675         run_queue(gl, 0);
676         spin_unlock(&gl->gl_spin);
677         if (!delay ||
678             queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
679                 gfs2_glock_put(gl);
680         if (drop_ref)
681                 gfs2_glock_put(gl);
682 }
683
684 /**
685  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
686  * @sdp: The GFS2 superblock
687  * @number: the lock number
688  * @glops: The glock_operations to use
689  * @create: If 0, don't create the glock if it doesn't exist
690  * @glp: the glock is returned here
691  *
692  * This does not lock a glock, just finds/creates structures for one.
693  *
694  * Returns: errno
695  */
696
697 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
698                    const struct gfs2_glock_operations *glops, int create,
699                    struct gfs2_glock **glp)
700 {
701         struct super_block *s = sdp->sd_vfs;
702         struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
703         struct gfs2_glock *gl, *tmp;
704         unsigned int hash = gl_hash(sdp, &name);
705         struct address_space *mapping;
706         struct kmem_cache *cachep;
707
708         rcu_read_lock();
709         gl = search_bucket(hash, sdp, &name);
710         rcu_read_unlock();
711
712         *glp = gl;
713         if (gl)
714                 return 0;
715         if (!create)
716                 return -ENOENT;
717
718         if (glops->go_flags & GLOF_ASPACE)
719                 cachep = gfs2_glock_aspace_cachep;
720         else
721                 cachep = gfs2_glock_cachep;
722         gl = kmem_cache_alloc(cachep, GFP_KERNEL);
723         if (!gl)
724                 return -ENOMEM;
725
726         atomic_inc(&sdp->sd_glock_disposal);
727         gl->gl_flags = 0;
728         gl->gl_name = name;
729         atomic_set(&gl->gl_ref, 1);
730         gl->gl_state = LM_ST_UNLOCKED;
731         gl->gl_target = LM_ST_UNLOCKED;
732         gl->gl_demote_state = LM_ST_EXCLUSIVE;
733         gl->gl_hash = hash;
734         gl->gl_ops = glops;
735         snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
736         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
737         gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
738         gl->gl_tchange = jiffies;
739         gl->gl_object = NULL;
740         gl->gl_sbd = sdp;
741         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
742         INIT_WORK(&gl->gl_delete, delete_work_func);
743
744         mapping = gfs2_glock2aspace(gl);
745         if (mapping) {
746                 mapping->a_ops = &gfs2_meta_aops;
747                 mapping->host = s->s_bdev->bd_inode;
748                 mapping->flags = 0;
749                 mapping_set_gfp_mask(mapping, GFP_NOFS);
750                 mapping->assoc_mapping = NULL;
751                 mapping->backing_dev_info = s->s_bdi;
752                 mapping->writeback_index = 0;
753         }
754
755         spin_lock_bucket(hash);
756         tmp = search_bucket(hash, sdp, &name);
757         if (tmp) {
758                 spin_unlock_bucket(hash);
759                 kmem_cache_free(cachep, gl);
760                 atomic_dec(&sdp->sd_glock_disposal);
761                 gl = tmp;
762         } else {
763                 hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
764                 spin_unlock_bucket(hash);
765         }
766
767         *glp = gl;
768
769         return 0;
770 }
771
772 /**
773  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
774  * @gl: the glock
775  * @state: the state we're requesting
776  * @flags: the modifier flags
777  * @gh: the holder structure
778  *
779  */
780
781 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
782                       struct gfs2_holder *gh)
783 {
784         INIT_LIST_HEAD(&gh->gh_list);
785         gh->gh_gl = gl;
786         gh->gh_ip = (unsigned long)__builtin_return_address(0);
787         gh->gh_owner_pid = get_pid(task_pid(current));
788         gh->gh_state = state;
789         gh->gh_flags = flags;
790         gh->gh_error = 0;
791         gh->gh_iflags = 0;
792         gfs2_glock_hold(gl);
793 }
794
795 /**
796  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
797  * @state: the state we're requesting
798  * @flags: the modifier flags
799  * @gh: the holder structure
800  *
801  * Don't mess with the glock.
802  *
803  */
804
805 void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
806 {
807         gh->gh_state = state;
808         gh->gh_flags = flags;
809         gh->gh_iflags = 0;
810         gh->gh_ip = (unsigned long)__builtin_return_address(0);
811         if (gh->gh_owner_pid)
812                 put_pid(gh->gh_owner_pid);
813         gh->gh_owner_pid = get_pid(task_pid(current));
814 }
815
816 /**
817  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
818  * @gh: the holder structure
819  *
820  */
821
822 void gfs2_holder_uninit(struct gfs2_holder *gh)
823 {
824         put_pid(gh->gh_owner_pid);
825         gfs2_glock_put(gh->gh_gl);
826         gh->gh_gl = NULL;
827         gh->gh_ip = 0;
828 }
829
830 /**
831  * gfs2_glock_holder_wait
832  * @word: unused
833  *
834  * This function and gfs2_glock_demote_wait both show up in the WCHAN
835  * field. Thus I've separated these otherwise identical functions in
836  * order to be more informative to the user.
837  */
838
839 static int gfs2_glock_holder_wait(void *word)
840 {
841         schedule();
842         return 0;
843 }
844
845 static int gfs2_glock_demote_wait(void *word)
846 {
847         schedule();
848         return 0;
849 }
850
851 static void wait_on_holder(struct gfs2_holder *gh)
852 {
853         might_sleep();
854         wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
855 }
856
857 static void wait_on_demote(struct gfs2_glock *gl)
858 {
859         might_sleep();
860         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
861 }
862
863 /**
864  * handle_callback - process a demote request
865  * @gl: the glock
866  * @state: the state the caller wants us to change to
867  *
868  * There are only two requests that we are going to see in actual
869  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
870  */
871
872 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
873                             unsigned long delay)
874 {
875         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
876
877         set_bit(bit, &gl->gl_flags);
878         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
879                 gl->gl_demote_state = state;
880                 gl->gl_demote_time = jiffies;
881         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
882                         gl->gl_demote_state != state) {
883                 gl->gl_demote_state = LM_ST_UNLOCKED;
884         }
885         if (gl->gl_ops->go_callback)
886                 gl->gl_ops->go_callback(gl);
887         trace_gfs2_demote_rq(gl);
888 }
889
890 /**
891  * gfs2_glock_wait - wait on a glock acquisition
892  * @gh: the glock holder
893  *
894  * Returns: 0 on success
895  */
896
897 int gfs2_glock_wait(struct gfs2_holder *gh)
898 {
899         wait_on_holder(gh);
900         return gh->gh_error;
901 }
902
903 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
904 {
905         struct va_format vaf;
906         va_list args;
907
908         va_start(args, fmt);
909
910         if (seq) {
911                 struct gfs2_glock_iter *gi = seq->private;
912                 vsprintf(gi->string, fmt, args);
913                 seq_printf(seq, gi->string);
914         } else {
915                 vaf.fmt = fmt;
916                 vaf.va = &args;
917
918                 printk(KERN_ERR " %pV", &vaf);
919         }
920
921         va_end(args);
922 }
923
924 /**
925  * add_to_queue - Add a holder to the wait queue (but look for recursion)
926  * @gh: the holder structure to add
927  *
928  * Eventually we should move the recursive locking trap to a
929  * debugging option or something like that. This is the fast
930  * path and needs to have the minimum number of distractions.
931  * 
932  */
933
934 static inline void add_to_queue(struct gfs2_holder *gh)
935 __releases(&gl->gl_spin)
936 __acquires(&gl->gl_spin)
937 {
938         struct gfs2_glock *gl = gh->gh_gl;
939         struct gfs2_sbd *sdp = gl->gl_sbd;
940         struct list_head *insert_pt = NULL;
941         struct gfs2_holder *gh2;
942         int try_lock = 0;
943
944         BUG_ON(gh->gh_owner_pid == NULL);
945         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
946                 BUG();
947
948         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
949                 if (test_bit(GLF_LOCK, &gl->gl_flags))
950                         try_lock = 1;
951                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
952                         goto fail;
953         }
954
955         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
956                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
957                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
958                         goto trap_recursive;
959                 if (try_lock &&
960                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
961                     !may_grant(gl, gh)) {
962 fail:
963                         gh->gh_error = GLR_TRYFAILED;
964                         gfs2_holder_wake(gh);
965                         return;
966                 }
967                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
968                         continue;
969                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
970                         insert_pt = &gh2->gh_list;
971         }
972         set_bit(GLF_QUEUED, &gl->gl_flags);
973         trace_gfs2_glock_queue(gh, 1);
974         if (likely(insert_pt == NULL)) {
975                 list_add_tail(&gh->gh_list, &gl->gl_holders);
976                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
977                         goto do_cancel;
978                 return;
979         }
980         list_add_tail(&gh->gh_list, insert_pt);
981 do_cancel:
982         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
983         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
984                 spin_unlock(&gl->gl_spin);
985                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
986                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
987                 spin_lock(&gl->gl_spin);
988         }
989         return;
990
991 trap_recursive:
992         print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
993         printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
994         printk(KERN_ERR "lock type: %d req lock state : %d\n",
995                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
996         print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
997         printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
998         printk(KERN_ERR "lock type: %d req lock state : %d\n",
999                gh->gh_gl->gl_name.ln_type, gh->gh_state);
1000         __dump_glock(NULL, gl);
1001         BUG();
1002 }
1003
1004 /**
1005  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
1006  * @gh: the holder structure
1007  *
1008  * if (gh->gh_flags & GL_ASYNC), this never returns an error
1009  *
1010  * Returns: 0, GLR_TRYFAILED, or errno on failure
1011  */
1012
1013 int gfs2_glock_nq(struct gfs2_holder *gh)
1014 {
1015         struct gfs2_glock *gl = gh->gh_gl;
1016         struct gfs2_sbd *sdp = gl->gl_sbd;
1017         int error = 0;
1018
1019         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1020                 return -EIO;
1021
1022         if (test_bit(GLF_LRU, &gl->gl_flags))
1023                 gfs2_glock_remove_from_lru(gl);
1024
1025         spin_lock(&gl->gl_spin);
1026         add_to_queue(gh);
1027         if ((LM_FLAG_NOEXP & gh->gh_flags) &&
1028             test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1029                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1030         run_queue(gl, 1);
1031         spin_unlock(&gl->gl_spin);
1032
1033         if (!(gh->gh_flags & GL_ASYNC))
1034                 error = gfs2_glock_wait(gh);
1035
1036         return error;
1037 }
1038
1039 /**
1040  * gfs2_glock_poll - poll to see if an async request has been completed
1041  * @gh: the holder
1042  *
1043  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1044  */
1045
1046 int gfs2_glock_poll(struct gfs2_holder *gh)
1047 {
1048         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1049 }
1050
1051 /**
1052  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1053  * @gh: the glock holder
1054  *
1055  */
1056
1057 void gfs2_glock_dq(struct gfs2_holder *gh)
1058 {
1059         struct gfs2_glock *gl = gh->gh_gl;
1060         const struct gfs2_glock_operations *glops = gl->gl_ops;
1061         unsigned delay = 0;
1062         int fast_path = 0;
1063
1064         spin_lock(&gl->gl_spin);
1065         if (gh->gh_flags & GL_NOCACHE)
1066                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1067
1068         list_del_init(&gh->gh_list);
1069         if (find_first_holder(gl) == NULL) {
1070                 if (glops->go_unlock) {
1071                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1072                         spin_unlock(&gl->gl_spin);
1073                         glops->go_unlock(gh);
1074                         spin_lock(&gl->gl_spin);
1075                         clear_bit(GLF_LOCK, &gl->gl_flags);
1076                 }
1077                 if (list_empty(&gl->gl_holders) &&
1078                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1079                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1080                         fast_path = 1;
1081         }
1082         if (!test_bit(GLF_LFLUSH, &gl->gl_flags))
1083                 __gfs2_glock_schedule_for_reclaim(gl);
1084         trace_gfs2_glock_queue(gh, 0);
1085         spin_unlock(&gl->gl_spin);
1086         if (likely(fast_path))
1087                 return;
1088
1089         gfs2_glock_hold(gl);
1090         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1091             !test_bit(GLF_DEMOTE, &gl->gl_flags))
1092                 delay = gl->gl_ops->go_min_hold_time;
1093         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1094                 gfs2_glock_put(gl);
1095 }
1096
1097 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1098 {
1099         struct gfs2_glock *gl = gh->gh_gl;
1100         gfs2_glock_dq(gh);
1101         wait_on_demote(gl);
1102 }
1103
1104 /**
1105  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1106  * @gh: the holder structure
1107  *
1108  */
1109
1110 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1111 {
1112         gfs2_glock_dq(gh);
1113         gfs2_holder_uninit(gh);
1114 }
1115
1116 /**
1117  * gfs2_glock_nq_num - acquire a glock based on lock number
1118  * @sdp: the filesystem
1119  * @number: the lock number
1120  * @glops: the glock operations for the type of glock
1121  * @state: the state to acquire the glock in
1122  * @flags: modifier flags for the acquisition
1123  * @gh: the struct gfs2_holder
1124  *
1125  * Returns: errno
1126  */
1127
1128 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1129                       const struct gfs2_glock_operations *glops,
1130                       unsigned int state, int flags, struct gfs2_holder *gh)
1131 {
1132         struct gfs2_glock *gl;
1133         int error;
1134
1135         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1136         if (!error) {
1137                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1138                 gfs2_glock_put(gl);
1139         }
1140
1141         return error;
1142 }
1143
1144 /**
1145  * glock_compare - Compare two struct gfs2_glock structures for sorting
1146  * @arg_a: the first structure
1147  * @arg_b: the second structure
1148  *
1149  */
1150
1151 static int glock_compare(const void *arg_a, const void *arg_b)
1152 {
1153         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1154         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1155         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1156         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1157
1158         if (a->ln_number > b->ln_number)
1159                 return 1;
1160         if (a->ln_number < b->ln_number)
1161                 return -1;
1162         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1163         return 0;
1164 }
1165
1166 /**
1167  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1168  * @num_gh: the number of structures
1169  * @ghs: an array of struct gfs2_holder structures
1170  *
1171  * Returns: 0 on success (all glocks acquired),
1172  *          errno on failure (no glocks acquired)
1173  */
1174
1175 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1176                      struct gfs2_holder **p)
1177 {
1178         unsigned int x;
1179         int error = 0;
1180
1181         for (x = 0; x < num_gh; x++)
1182                 p[x] = &ghs[x];
1183
1184         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1185
1186         for (x = 0; x < num_gh; x++) {
1187                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1188
1189                 error = gfs2_glock_nq(p[x]);
1190                 if (error) {
1191                         while (x--)
1192                                 gfs2_glock_dq(p[x]);
1193                         break;
1194                 }
1195         }
1196
1197         return error;
1198 }
1199
1200 /**
1201  * gfs2_glock_nq_m - acquire multiple glocks
1202  * @num_gh: the number of structures
1203  * @ghs: an array of struct gfs2_holder structures
1204  *
1205  *
1206  * Returns: 0 on success (all glocks acquired),
1207  *          errno on failure (no glocks acquired)
1208  */
1209
1210 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1211 {
1212         struct gfs2_holder *tmp[4];
1213         struct gfs2_holder **pph = tmp;
1214         int error = 0;
1215
1216         switch(num_gh) {
1217         case 0:
1218                 return 0;
1219         case 1:
1220                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1221                 return gfs2_glock_nq(ghs);
1222         default:
1223                 if (num_gh <= 4)
1224                         break;
1225                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1226                 if (!pph)
1227                         return -ENOMEM;
1228         }
1229
1230         error = nq_m_sync(num_gh, ghs, pph);
1231
1232         if (pph != tmp)
1233                 kfree(pph);
1234
1235         return error;
1236 }
1237
1238 /**
1239  * gfs2_glock_dq_m - release multiple glocks
1240  * @num_gh: the number of structures
1241  * @ghs: an array of struct gfs2_holder structures
1242  *
1243  */
1244
1245 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1246 {
1247         while (num_gh--)
1248                 gfs2_glock_dq(&ghs[num_gh]);
1249 }
1250
1251 /**
1252  * gfs2_glock_dq_uninit_m - release multiple glocks
1253  * @num_gh: the number of structures
1254  * @ghs: an array of struct gfs2_holder structures
1255  *
1256  */
1257
1258 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
1259 {
1260         while (num_gh--)
1261                 gfs2_glock_dq_uninit(&ghs[num_gh]);
1262 }
1263
1264 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1265 {
1266         unsigned long delay = 0;
1267         unsigned long holdtime;
1268         unsigned long now = jiffies;
1269
1270         gfs2_glock_hold(gl);
1271         holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
1272         if (test_bit(GLF_QUEUED, &gl->gl_flags)) {
1273                 if (time_before(now, holdtime))
1274                         delay = holdtime - now;
1275                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1276                         delay = gl->gl_ops->go_min_hold_time;
1277         }
1278
1279         spin_lock(&gl->gl_spin);
1280         handle_callback(gl, state, delay);
1281         spin_unlock(&gl->gl_spin);
1282         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1283                 gfs2_glock_put(gl);
1284 }
1285
1286 /**
1287  * gfs2_should_freeze - Figure out if glock should be frozen
1288  * @gl: The glock in question
1289  *
1290  * Glocks are not frozen if (a) the result of the dlm operation is
1291  * an error, (b) the locking operation was an unlock operation or
1292  * (c) if there is a "noexp" flagged request anywhere in the queue
1293  *
1294  * Returns: 1 if freezing should occur, 0 otherwise
1295  */
1296
1297 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1298 {
1299         const struct gfs2_holder *gh;
1300
1301         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1302                 return 0;
1303         if (gl->gl_target == LM_ST_UNLOCKED)
1304                 return 0;
1305
1306         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1307                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1308                         continue;
1309                 if (LM_FLAG_NOEXP & gh->gh_flags)
1310                         return 0;
1311         }
1312
1313         return 1;
1314 }
1315
1316 /**
1317  * gfs2_glock_complete - Callback used by locking
1318  * @gl: Pointer to the glock
1319  * @ret: The return value from the dlm
1320  *
1321  * The gl_reply field is under the gl_spin lock so that it is ok
1322  * to use a bitfield shared with other glock state fields.
1323  */
1324
1325 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1326 {
1327         struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
1328
1329         spin_lock(&gl->gl_spin);
1330         gl->gl_reply = ret;
1331
1332         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
1333                 if (gfs2_should_freeze(gl)) {
1334                         set_bit(GLF_FROZEN, &gl->gl_flags);
1335                         spin_unlock(&gl->gl_spin);
1336                         return;
1337                 }
1338         }
1339
1340         spin_unlock(&gl->gl_spin);
1341         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1342         smp_wmb();
1343         gfs2_glock_hold(gl);
1344         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1345                 gfs2_glock_put(gl);
1346 }
1347
1348
1349 static int gfs2_shrink_glock_memory(struct shrinker *shrink,
1350                                     struct shrink_control *sc)
1351 {
1352         struct gfs2_glock *gl;
1353         int may_demote;
1354         int nr_skipped = 0;
1355         int nr = sc->nr_to_scan;
1356         gfp_t gfp_mask = sc->gfp_mask;
1357         LIST_HEAD(skipped);
1358
1359         if (nr == 0)
1360                 goto out;
1361
1362         if (!(gfp_mask & __GFP_FS))
1363                 return -1;
1364
1365         spin_lock(&lru_lock);
1366         while(nr && !list_empty(&lru_list)) {
1367                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1368                 list_del_init(&gl->gl_lru);
1369                 clear_bit(GLF_LRU, &gl->gl_flags);
1370                 atomic_dec(&lru_count);
1371
1372                 /* Test for being demotable */
1373                 if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1374                         gfs2_glock_hold(gl);
1375                         spin_unlock(&lru_lock);
1376                         spin_lock(&gl->gl_spin);
1377                         may_demote = demote_ok(gl);
1378                         if (may_demote) {
1379                                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1380                                 nr--;
1381                         }
1382                         clear_bit(GLF_LOCK, &gl->gl_flags);
1383                         smp_mb__after_clear_bit();
1384                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1385                                 gfs2_glock_put_nolock(gl);
1386                         spin_unlock(&gl->gl_spin);
1387                         spin_lock(&lru_lock);
1388                         continue;
1389                 }
1390                 nr_skipped++;
1391                 list_add(&gl->gl_lru, &skipped);
1392                 set_bit(GLF_LRU, &gl->gl_flags);
1393         }
1394         list_splice(&skipped, &lru_list);
1395         atomic_add(nr_skipped, &lru_count);
1396         spin_unlock(&lru_lock);
1397 out:
1398         return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
1399 }
1400
1401 static struct shrinker glock_shrinker = {
1402         .shrink = gfs2_shrink_glock_memory,
1403         .seeks = DEFAULT_SEEKS,
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  */
1413
1414 static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
1415                           unsigned int hash)
1416 {
1417         struct gfs2_glock *gl;
1418         struct hlist_bl_head *head = &gl_hash_table[hash];
1419         struct hlist_bl_node *pos;
1420
1421         rcu_read_lock();
1422         hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
1423                 if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
1424                         examiner(gl);
1425         }
1426         rcu_read_unlock();
1427         cond_resched();
1428 }
1429
1430 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1431 {
1432         unsigned x;
1433
1434         for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
1435                 examine_bucket(examiner, sdp, x);
1436 }
1437
1438
1439 /**
1440  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1441  * @gl: The glock to thaw
1442  *
1443  * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
1444  * so this has to result in the ref count being dropped by one.
1445  */
1446
1447 static void thaw_glock(struct gfs2_glock *gl)
1448 {
1449         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1450                 return;
1451         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1452         gfs2_glock_hold(gl);
1453         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1454                 gfs2_glock_put(gl);
1455 }
1456
1457 /**
1458  * clear_glock - look at a glock and see if we can free it from glock cache
1459  * @gl: the glock to look at
1460  *
1461  */
1462
1463 static void clear_glock(struct gfs2_glock *gl)
1464 {
1465         gfs2_glock_remove_from_lru(gl);
1466
1467         spin_lock(&gl->gl_spin);
1468         if (gl->gl_state != LM_ST_UNLOCKED)
1469                 handle_callback(gl, LM_ST_UNLOCKED, 0);
1470         spin_unlock(&gl->gl_spin);
1471         gfs2_glock_hold(gl);
1472         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1473                 gfs2_glock_put(gl);
1474 }
1475
1476 /**
1477  * gfs2_glock_thaw - Thaw any frozen glocks
1478  * @sdp: The super block
1479  *
1480  */
1481
1482 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1483 {
1484         glock_hash_walk(thaw_glock, sdp);
1485 }
1486
1487 static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1488 {
1489         int ret;
1490         spin_lock(&gl->gl_spin);
1491         ret = __dump_glock(seq, gl);
1492         spin_unlock(&gl->gl_spin);
1493         return ret;
1494 }
1495
1496 static void dump_glock_func(struct gfs2_glock *gl)
1497 {
1498         dump_glock(NULL, 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         glock_hash_walk(clear_glock, sdp);
1512         flush_workqueue(glock_workqueue);
1513         wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
1514         glock_hash_walk(dump_glock_func, sdp);
1515 }
1516
1517 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1518 {
1519         struct gfs2_glock *gl = ip->i_gl;
1520         int ret;
1521
1522         ret = gfs2_truncatei_resume(ip);
1523         gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
1524
1525         spin_lock(&gl->gl_spin);
1526         clear_bit(GLF_LOCK, &gl->gl_flags);
1527         run_queue(gl, 1);
1528         spin_unlock(&gl->gl_spin);
1529 }
1530
1531 static const char *state2str(unsigned state)
1532 {
1533         switch(state) {
1534         case LM_ST_UNLOCKED:
1535                 return "UN";
1536         case LM_ST_SHARED:
1537                 return "SH";
1538         case LM_ST_DEFERRED:
1539                 return "DF";
1540         case LM_ST_EXCLUSIVE:
1541                 return "EX";
1542         }
1543         return "??";
1544 }
1545
1546 static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
1547 {
1548         char *p = buf;
1549         if (flags & LM_FLAG_TRY)
1550                 *p++ = 't';
1551         if (flags & LM_FLAG_TRY_1CB)
1552                 *p++ = 'T';
1553         if (flags & LM_FLAG_NOEXP)
1554                 *p++ = 'e';
1555         if (flags & LM_FLAG_ANY)
1556                 *p++ = 'A';
1557         if (flags & LM_FLAG_PRIORITY)
1558                 *p++ = 'p';
1559         if (flags & GL_ASYNC)
1560                 *p++ = 'a';
1561         if (flags & GL_EXACT)
1562                 *p++ = 'E';
1563         if (flags & GL_NOCACHE)
1564                 *p++ = 'c';
1565         if (test_bit(HIF_HOLDER, &iflags))
1566                 *p++ = 'H';
1567         if (test_bit(HIF_WAIT, &iflags))
1568                 *p++ = 'W';
1569         if (test_bit(HIF_FIRST, &iflags))
1570                 *p++ = 'F';
1571         *p = 0;
1572         return buf;
1573 }
1574
1575 /**
1576  * dump_holder - print information about a glock holder
1577  * @seq: the seq_file struct
1578  * @gh: the glock holder
1579  *
1580  * Returns: 0 on success, -ENOBUFS when we run out of space
1581  */
1582
1583 static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1584 {
1585         struct task_struct *gh_owner = NULL;
1586         char flags_buf[32];
1587
1588         if (gh->gh_owner_pid)
1589                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1590         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1591                        state2str(gh->gh_state),
1592                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1593                        gh->gh_error,
1594                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1595                        gh_owner ? gh_owner->comm : "(ended)",
1596                        (void *)gh->gh_ip);
1597         return 0;
1598 }
1599
1600 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1601 {
1602         const unsigned long *gflags = &gl->gl_flags;
1603         char *p = buf;
1604
1605         if (test_bit(GLF_LOCK, gflags))
1606                 *p++ = 'l';
1607         if (test_bit(GLF_DEMOTE, gflags))
1608                 *p++ = 'D';
1609         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1610                 *p++ = 'd';
1611         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1612                 *p++ = 'p';
1613         if (test_bit(GLF_DIRTY, gflags))
1614                 *p++ = 'y';
1615         if (test_bit(GLF_LFLUSH, gflags))
1616                 *p++ = 'f';
1617         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1618                 *p++ = 'i';
1619         if (test_bit(GLF_REPLY_PENDING, gflags))
1620                 *p++ = 'r';
1621         if (test_bit(GLF_INITIAL, gflags))
1622                 *p++ = 'I';
1623         if (test_bit(GLF_FROZEN, gflags))
1624                 *p++ = 'F';
1625         if (test_bit(GLF_QUEUED, gflags))
1626                 *p++ = 'q';
1627         if (test_bit(GLF_LRU, gflags))
1628                 *p++ = 'L';
1629         if (gl->gl_object)
1630                 *p++ = 'o';
1631         *p = 0;
1632         return buf;
1633 }
1634
1635 /**
1636  * __dump_glock - print information about a glock
1637  * @seq: The seq_file struct
1638  * @gl: the glock
1639  *
1640  * The file format is as follows:
1641  * One line per object, capital letters are used to indicate objects
1642  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1643  * other objects are indented by a single space and follow the glock to
1644  * which they are related. Fields are indicated by lower case letters
1645  * followed by a colon and the field value, except for strings which are in
1646  * [] so that its possible to see if they are composed of spaces for
1647  * example. The field's are n = number (id of the object), f = flags,
1648  * t = type, s = state, r = refcount, e = error, p = pid.
1649  *
1650  * Returns: 0 on success, -ENOBUFS when we run out of space
1651  */
1652
1653 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1654 {
1655         const struct gfs2_glock_operations *glops = gl->gl_ops;
1656         unsigned long long dtime;
1657         const struct gfs2_holder *gh;
1658         char gflags_buf[32];
1659         int error = 0;
1660
1661         dtime = jiffies - gl->gl_demote_time;
1662         dtime *= 1000000/HZ; /* demote time in uSec */
1663         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1664                 dtime = 0;
1665         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d\n",
1666                   state2str(gl->gl_state),
1667                   gl->gl_name.ln_type,
1668                   (unsigned long long)gl->gl_name.ln_number,
1669                   gflags2str(gflags_buf, gl),
1670                   state2str(gl->gl_target),
1671                   state2str(gl->gl_demote_state), dtime,
1672                   atomic_read(&gl->gl_ail_count),
1673                   atomic_read(&gl->gl_revokes),
1674                   atomic_read(&gl->gl_ref));
1675
1676         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1677                 error = dump_holder(seq, gh);
1678                 if (error)
1679                         goto out;
1680         }
1681         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1682                 error = glops->go_dump(seq, gl);
1683 out:
1684         return error;
1685 }
1686
1687
1688
1689
1690 int __init gfs2_glock_init(void)
1691 {
1692         unsigned i;
1693         for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
1694                 INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
1695         }
1696
1697         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1698                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1699         if (IS_ERR(glock_workqueue))
1700                 return PTR_ERR(glock_workqueue);
1701         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1702                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1703                                                 0);
1704         if (IS_ERR(gfs2_delete_workqueue)) {
1705                 destroy_workqueue(glock_workqueue);
1706                 return PTR_ERR(gfs2_delete_workqueue);
1707         }
1708
1709         register_shrinker(&glock_shrinker);
1710
1711         return 0;
1712 }
1713
1714 void gfs2_glock_exit(void)
1715 {
1716         unregister_shrinker(&glock_shrinker);
1717         destroy_workqueue(glock_workqueue);
1718         destroy_workqueue(gfs2_delete_workqueue);
1719 }
1720
1721 static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
1722 {
1723         return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
1724                               struct gfs2_glock, gl_list);
1725 }
1726
1727 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
1728 {
1729         return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
1730                               struct gfs2_glock, gl_list);
1731 }
1732
1733 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1734 {
1735         struct gfs2_glock *gl;
1736
1737         do {
1738                 gl = gi->gl;
1739                 if (gl) {
1740                         gi->gl = glock_hash_next(gl);
1741                 } else {
1742                         gi->gl = glock_hash_chain(gi->hash);
1743                 }
1744                 while (gi->gl == NULL) {
1745                         gi->hash++;
1746                         if (gi->hash >= GFS2_GL_HASH_SIZE) {
1747                                 rcu_read_unlock();
1748                                 return 1;
1749                         }
1750                         gi->gl = glock_hash_chain(gi->hash);
1751                 }
1752         /* Skip entries for other sb and dead entries */
1753         } while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
1754
1755         return 0;
1756 }
1757
1758 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1759 {
1760         struct gfs2_glock_iter *gi = seq->private;
1761         loff_t n = *pos;
1762
1763         gi->hash = 0;
1764         rcu_read_lock();
1765
1766         do {
1767                 if (gfs2_glock_iter_next(gi))
1768                         return NULL;
1769         } while (n--);
1770
1771         return gi->gl;
1772 }
1773
1774 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1775                                  loff_t *pos)
1776 {
1777         struct gfs2_glock_iter *gi = seq->private;
1778
1779         (*pos)++;
1780
1781         if (gfs2_glock_iter_next(gi))
1782                 return NULL;
1783
1784         return gi->gl;
1785 }
1786
1787 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1788 {
1789         struct gfs2_glock_iter *gi = seq->private;
1790
1791         if (gi->gl)
1792                 rcu_read_unlock();
1793         gi->gl = NULL;
1794 }
1795
1796 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1797 {
1798         return dump_glock(seq, iter_ptr);
1799 }
1800
1801 static const struct seq_operations gfs2_glock_seq_ops = {
1802         .start = gfs2_glock_seq_start,
1803         .next  = gfs2_glock_seq_next,
1804         .stop  = gfs2_glock_seq_stop,
1805         .show  = gfs2_glock_seq_show,
1806 };
1807
1808 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
1809 {
1810         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1811                                    sizeof(struct gfs2_glock_iter));
1812         if (ret == 0) {
1813                 struct seq_file *seq = file->private_data;
1814                 struct gfs2_glock_iter *gi = seq->private;
1815                 gi->sdp = inode->i_private;
1816         }
1817         return ret;
1818 }
1819
1820 static const struct file_operations gfs2_debug_fops = {
1821         .owner   = THIS_MODULE,
1822         .open    = gfs2_debugfs_open,
1823         .read    = seq_read,
1824         .llseek  = seq_lseek,
1825         .release = seq_release_private,
1826 };
1827
1828 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
1829 {
1830         sdp->debugfs_dir = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
1831         if (!sdp->debugfs_dir)
1832                 return -ENOMEM;
1833         sdp->debugfs_dentry_glocks = debugfs_create_file("glocks",
1834                                                          S_IFREG | S_IRUGO,
1835                                                          sdp->debugfs_dir, sdp,
1836                                                          &gfs2_debug_fops);
1837         if (!sdp->debugfs_dentry_glocks)
1838                 return -ENOMEM;
1839
1840         return 0;
1841 }
1842
1843 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
1844 {
1845         if (sdp && sdp->debugfs_dir) {
1846                 if (sdp->debugfs_dentry_glocks) {
1847                         debugfs_remove(sdp->debugfs_dentry_glocks);
1848                         sdp->debugfs_dentry_glocks = NULL;
1849                 }
1850                 debugfs_remove(sdp->debugfs_dir);
1851                 sdp->debugfs_dir = NULL;
1852         }
1853 }
1854
1855 int gfs2_register_debugfs(void)
1856 {
1857         gfs2_root = debugfs_create_dir("gfs2", NULL);
1858         return gfs2_root ? 0 : -ENOMEM;
1859 }
1860
1861 void gfs2_unregister_debugfs(void)
1862 {
1863         debugfs_remove(gfs2_root);
1864         gfs2_root = NULL;
1865 }