Merge tag 'gfs2-4.11.addendum' of git://git.kernel.org/pub/scm/linux/kernel/git/gfs2...
[sfrench/cifs-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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/buffer_head.h>
16 #include <linux/delay.h>
17 #include <linux/sort.h>
18 #include <linux/jhash.h>
19 #include <linux/kallsyms.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/list.h>
22 #include <linux/wait.h>
23 #include <linux/module.h>
24 #include <linux/uaccess.h>
25 #include <linux/seq_file.h>
26 #include <linux/debugfs.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/workqueue.h>
30 #include <linux/jiffies.h>
31 #include <linux/rcupdate.h>
32 #include <linux/rculist_bl.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/percpu.h>
35 #include <linux/list_sort.h>
36 #include <linux/lockref.h>
37 #include <linux/rhashtable.h>
38
39 #include "gfs2.h"
40 #include "incore.h"
41 #include "glock.h"
42 #include "glops.h"
43 #include "inode.h"
44 #include "lops.h"
45 #include "meta_io.h"
46 #include "quota.h"
47 #include "super.h"
48 #include "util.h"
49 #include "bmap.h"
50 #define CREATE_TRACE_POINTS
51 #include "trace_gfs2.h"
52
53 struct gfs2_glock_iter {
54         struct gfs2_sbd *sdp;           /* incore superblock           */
55         struct rhashtable_iter hti;     /* rhashtable iterator         */
56         struct gfs2_glock *gl;          /* current glock struct        */
57         loff_t last_pos;                /* last position               */
58 };
59
60 typedef void (*glock_examiner) (struct gfs2_glock * gl);
61
62 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
63
64 static struct dentry *gfs2_root;
65 static struct workqueue_struct *glock_workqueue;
66 struct workqueue_struct *gfs2_delete_workqueue;
67 static LIST_HEAD(lru_list);
68 static atomic_t lru_count = ATOMIC_INIT(0);
69 static DEFINE_SPINLOCK(lru_lock);
70
71 #define GFS2_GL_HASH_SHIFT      15
72 #define GFS2_GL_HASH_SIZE       BIT(GFS2_GL_HASH_SHIFT)
73
74 static struct rhashtable_params ht_parms = {
75         .nelem_hint = GFS2_GL_HASH_SIZE * 3 / 4,
76         .key_len = sizeof(struct lm_lockname),
77         .key_offset = offsetof(struct gfs2_glock, gl_name),
78         .head_offset = offsetof(struct gfs2_glock, gl_node),
79 };
80
81 static struct rhashtable gl_hash_table;
82
83 void gfs2_glock_free(struct gfs2_glock *gl)
84 {
85         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
86
87         if (gl->gl_ops->go_flags & GLOF_ASPACE) {
88                 kmem_cache_free(gfs2_glock_aspace_cachep, gl);
89         } else {
90                 kfree(gl->gl_lksb.sb_lvbptr);
91                 kmem_cache_free(gfs2_glock_cachep, gl);
92         }
93         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
94                 wake_up(&sdp->sd_glock_wait);
95 }
96
97 /**
98  * gfs2_glock_hold() - increment reference count on glock
99  * @gl: The glock to hold
100  *
101  */
102
103 static void gfs2_glock_hold(struct gfs2_glock *gl)
104 {
105         GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
106         lockref_get(&gl->gl_lockref);
107 }
108
109 /**
110  * demote_ok - Check to see if it's ok to unlock a glock
111  * @gl: the glock
112  *
113  * Returns: 1 if it's ok
114  */
115
116 static int demote_ok(const struct gfs2_glock *gl)
117 {
118         const struct gfs2_glock_operations *glops = gl->gl_ops;
119
120         if (gl->gl_state == LM_ST_UNLOCKED)
121                 return 0;
122         if (!list_empty(&gl->gl_holders))
123                 return 0;
124         if (glops->go_demote_ok)
125                 return glops->go_demote_ok(gl);
126         return 1;
127 }
128
129
130 void gfs2_glock_add_to_lru(struct gfs2_glock *gl)
131 {
132         spin_lock(&lru_lock);
133
134         if (!list_empty(&gl->gl_lru))
135                 list_del_init(&gl->gl_lru);
136         else
137                 atomic_inc(&lru_count);
138
139         list_add_tail(&gl->gl_lru, &lru_list);
140         set_bit(GLF_LRU, &gl->gl_flags);
141         spin_unlock(&lru_lock);
142 }
143
144 static void gfs2_glock_remove_from_lru(struct gfs2_glock *gl)
145 {
146         spin_lock(&lru_lock);
147         if (!list_empty(&gl->gl_lru)) {
148                 list_del_init(&gl->gl_lru);
149                 atomic_dec(&lru_count);
150                 clear_bit(GLF_LRU, &gl->gl_flags);
151         }
152         spin_unlock(&lru_lock);
153 }
154
155 /**
156  * gfs2_glock_put() - Decrement reference count on glock
157  * @gl: The glock to put
158  *
159  */
160
161 void gfs2_glock_put(struct gfs2_glock *gl)
162 {
163         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
164         struct address_space *mapping = gfs2_glock2aspace(gl);
165
166         if (lockref_put_or_lock(&gl->gl_lockref))
167                 return;
168
169         lockref_mark_dead(&gl->gl_lockref);
170
171         gfs2_glock_remove_from_lru(gl);
172         spin_unlock(&gl->gl_lockref.lock);
173         rhashtable_remove_fast(&gl_hash_table, &gl->gl_node, ht_parms);
174         GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
175         GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
176         trace_gfs2_glock_put(gl);
177         sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
178 }
179
180 /**
181  * may_grant - check if its ok to grant a new lock
182  * @gl: The glock
183  * @gh: The lock request which we wish to grant
184  *
185  * Returns: true if its ok to grant the lock
186  */
187
188 static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
189 {
190         const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
191         if ((gh->gh_state == LM_ST_EXCLUSIVE ||
192              gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
193                 return 0;
194         if (gl->gl_state == gh->gh_state)
195                 return 1;
196         if (gh->gh_flags & GL_EXACT)
197                 return 0;
198         if (gl->gl_state == LM_ST_EXCLUSIVE) {
199                 if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
200                         return 1;
201                 if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
202                         return 1;
203         }
204         if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
205                 return 1;
206         return 0;
207 }
208
209 static void gfs2_holder_wake(struct gfs2_holder *gh)
210 {
211         clear_bit(HIF_WAIT, &gh->gh_iflags);
212         smp_mb__after_atomic();
213         wake_up_bit(&gh->gh_iflags, HIF_WAIT);
214 }
215
216 /**
217  * do_error - Something unexpected has happened during a lock request
218  *
219  */
220
221 static void do_error(struct gfs2_glock *gl, const int ret)
222 {
223         struct gfs2_holder *gh, *tmp;
224
225         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
226                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
227                         continue;
228                 if (ret & LM_OUT_ERROR)
229                         gh->gh_error = -EIO;
230                 else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
231                         gh->gh_error = GLR_TRYFAILED;
232                 else
233                         continue;
234                 list_del_init(&gh->gh_list);
235                 trace_gfs2_glock_queue(gh, 0);
236                 gfs2_holder_wake(gh);
237         }
238 }
239
240 /**
241  * do_promote - promote as many requests as possible on the current queue
242  * @gl: The glock
243  * 
244  * Returns: 1 if there is a blocked holder at the head of the list, or 2
245  *          if a type specific operation is underway.
246  */
247
248 static int do_promote(struct gfs2_glock *gl)
249 __releases(&gl->gl_lockref.lock)
250 __acquires(&gl->gl_lockref.lock)
251 {
252         const struct gfs2_glock_operations *glops = gl->gl_ops;
253         struct gfs2_holder *gh, *tmp;
254         int ret;
255
256 restart:
257         list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
258                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
259                         continue;
260                 if (may_grant(gl, gh)) {
261                         if (gh->gh_list.prev == &gl->gl_holders &&
262                             glops->go_lock) {
263                                 spin_unlock(&gl->gl_lockref.lock);
264                                 /* FIXME: eliminate this eventually */
265                                 ret = glops->go_lock(gh);
266                                 spin_lock(&gl->gl_lockref.lock);
267                                 if (ret) {
268                                         if (ret == 1)
269                                                 return 2;
270                                         gh->gh_error = ret;
271                                         list_del_init(&gh->gh_list);
272                                         trace_gfs2_glock_queue(gh, 0);
273                                         gfs2_holder_wake(gh);
274                                         goto restart;
275                                 }
276                                 set_bit(HIF_HOLDER, &gh->gh_iflags);
277                                 trace_gfs2_promote(gh, 1);
278                                 gfs2_holder_wake(gh);
279                                 goto restart;
280                         }
281                         set_bit(HIF_HOLDER, &gh->gh_iflags);
282                         trace_gfs2_promote(gh, 0);
283                         gfs2_holder_wake(gh);
284                         continue;
285                 }
286                 if (gh->gh_list.prev == &gl->gl_holders)
287                         return 1;
288                 do_error(gl, 0);
289                 break;
290         }
291         return 0;
292 }
293
294 /**
295  * find_first_waiter - find the first gh that's waiting for the glock
296  * @gl: the glock
297  */
298
299 static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
300 {
301         struct gfs2_holder *gh;
302
303         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
304                 if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
305                         return gh;
306         }
307         return NULL;
308 }
309
310 /**
311  * state_change - record that the glock is now in a different state
312  * @gl: the glock
313  * @new_state the new state
314  *
315  */
316
317 static void state_change(struct gfs2_glock *gl, unsigned int new_state)
318 {
319         int held1, held2;
320
321         held1 = (gl->gl_state != LM_ST_UNLOCKED);
322         held2 = (new_state != LM_ST_UNLOCKED);
323
324         if (held1 != held2) {
325                 GLOCK_BUG_ON(gl, __lockref_is_dead(&gl->gl_lockref));
326                 if (held2)
327                         gl->gl_lockref.count++;
328                 else
329                         gl->gl_lockref.count--;
330         }
331         if (held1 && held2 && list_empty(&gl->gl_holders))
332                 clear_bit(GLF_QUEUED, &gl->gl_flags);
333
334         if (new_state != gl->gl_target)
335                 /* shorten our minimum hold time */
336                 gl->gl_hold_time = max(gl->gl_hold_time - GL_GLOCK_HOLD_DECR,
337                                        GL_GLOCK_MIN_HOLD);
338         gl->gl_state = new_state;
339         gl->gl_tchange = jiffies;
340 }
341
342 static void gfs2_demote_wake(struct gfs2_glock *gl)
343 {
344         gl->gl_demote_state = LM_ST_EXCLUSIVE;
345         clear_bit(GLF_DEMOTE, &gl->gl_flags);
346         smp_mb__after_atomic();
347         wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
348 }
349
350 /**
351  * finish_xmote - The DLM has replied to one of our lock requests
352  * @gl: The glock
353  * @ret: The status from the DLM
354  *
355  */
356
357 static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
358 {
359         const struct gfs2_glock_operations *glops = gl->gl_ops;
360         struct gfs2_holder *gh;
361         unsigned state = ret & LM_OUT_ST_MASK;
362         int rv;
363
364         spin_lock(&gl->gl_lockref.lock);
365         trace_gfs2_glock_state_change(gl, state);
366         state_change(gl, state);
367         gh = find_first_waiter(gl);
368
369         /* Demote to UN request arrived during demote to SH or DF */
370         if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
371             state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
372                 gl->gl_target = LM_ST_UNLOCKED;
373
374         /* Check for state != intended state */
375         if (unlikely(state != gl->gl_target)) {
376                 if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
377                         /* move to back of queue and try next entry */
378                         if (ret & LM_OUT_CANCELED) {
379                                 if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
380                                         list_move_tail(&gh->gh_list, &gl->gl_holders);
381                                 gh = find_first_waiter(gl);
382                                 gl->gl_target = gh->gh_state;
383                                 goto retry;
384                         }
385                         /* Some error or failed "try lock" - report it */
386                         if ((ret & LM_OUT_ERROR) ||
387                             (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
388                                 gl->gl_target = gl->gl_state;
389                                 do_error(gl, ret);
390                                 goto out;
391                         }
392                 }
393                 switch(state) {
394                 /* Unlocked due to conversion deadlock, try again */
395                 case LM_ST_UNLOCKED:
396 retry:
397                         do_xmote(gl, gh, gl->gl_target);
398                         break;
399                 /* Conversion fails, unlock and try again */
400                 case LM_ST_SHARED:
401                 case LM_ST_DEFERRED:
402                         do_xmote(gl, gh, LM_ST_UNLOCKED);
403                         break;
404                 default: /* Everything else */
405                         pr_err("wanted %u got %u\n", gl->gl_target, state);
406                         GLOCK_BUG_ON(gl, 1);
407                 }
408                 spin_unlock(&gl->gl_lockref.lock);
409                 return;
410         }
411
412         /* Fast path - we got what we asked for */
413         if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
414                 gfs2_demote_wake(gl);
415         if (state != LM_ST_UNLOCKED) {
416                 if (glops->go_xmote_bh) {
417                         spin_unlock(&gl->gl_lockref.lock);
418                         rv = glops->go_xmote_bh(gl, gh);
419                         spin_lock(&gl->gl_lockref.lock);
420                         if (rv) {
421                                 do_error(gl, rv);
422                                 goto out;
423                         }
424                 }
425                 rv = do_promote(gl);
426                 if (rv == 2)
427                         goto out_locked;
428         }
429 out:
430         clear_bit(GLF_LOCK, &gl->gl_flags);
431 out_locked:
432         spin_unlock(&gl->gl_lockref.lock);
433 }
434
435 /**
436  * do_xmote - Calls the DLM to change the state of a lock
437  * @gl: The lock state
438  * @gh: The holder (only for promotes)
439  * @target: The target lock state
440  *
441  */
442
443 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
444 __releases(&gl->gl_lockref.lock)
445 __acquires(&gl->gl_lockref.lock)
446 {
447         const struct gfs2_glock_operations *glops = gl->gl_ops;
448         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
449         unsigned int lck_flags = (unsigned int)(gh ? gh->gh_flags : 0);
450         int ret;
451
452         lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
453                       LM_FLAG_PRIORITY);
454         GLOCK_BUG_ON(gl, gl->gl_state == target);
455         GLOCK_BUG_ON(gl, gl->gl_state == gl->gl_target);
456         if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
457             glops->go_inval) {
458                 set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
459                 do_error(gl, 0); /* Fail queued try locks */
460         }
461         gl->gl_req = target;
462         set_bit(GLF_BLOCKING, &gl->gl_flags);
463         if ((gl->gl_req == LM_ST_UNLOCKED) ||
464             (gl->gl_state == LM_ST_EXCLUSIVE) ||
465             (lck_flags & (LM_FLAG_TRY|LM_FLAG_TRY_1CB)))
466                 clear_bit(GLF_BLOCKING, &gl->gl_flags);
467         spin_unlock(&gl->gl_lockref.lock);
468         if (glops->go_sync)
469                 glops->go_sync(gl);
470         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
471                 glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
472         clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
473
474         gfs2_glock_hold(gl);
475         if (sdp->sd_lockstruct.ls_ops->lm_lock) {
476                 /* lock_dlm */
477                 ret = sdp->sd_lockstruct.ls_ops->lm_lock(gl, target, lck_flags);
478                 if (ret == -EINVAL && gl->gl_target == LM_ST_UNLOCKED &&
479                     target == LM_ST_UNLOCKED &&
480                     test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags)) {
481                         finish_xmote(gl, target);
482                         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
483                                 gfs2_glock_put(gl);
484                 }
485                 else if (ret) {
486                         pr_err("lm_lock ret %d\n", ret);
487                         GLOCK_BUG_ON(gl, 1);
488                 }
489         } else { /* lock_nolock */
490                 finish_xmote(gl, target);
491                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
492                         gfs2_glock_put(gl);
493         }
494
495         spin_lock(&gl->gl_lockref.lock);
496 }
497
498 /**
499  * find_first_holder - find the first "holder" gh
500  * @gl: the glock
501  */
502
503 static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
504 {
505         struct gfs2_holder *gh;
506
507         if (!list_empty(&gl->gl_holders)) {
508                 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
509                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
510                         return gh;
511         }
512         return NULL;
513 }
514
515 /**
516  * run_queue - do all outstanding tasks related to a glock
517  * @gl: The glock in question
518  * @nonblock: True if we must not block in run_queue
519  *
520  */
521
522 static void run_queue(struct gfs2_glock *gl, const int nonblock)
523 __releases(&gl->gl_lockref.lock)
524 __acquires(&gl->gl_lockref.lock)
525 {
526         struct gfs2_holder *gh = NULL;
527         int ret;
528
529         if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
530                 return;
531
532         GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
533
534         if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
535             gl->gl_demote_state != gl->gl_state) {
536                 if (find_first_holder(gl))
537                         goto out_unlock;
538                 if (nonblock)
539                         goto out_sched;
540                 set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
541                 GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
542                 gl->gl_target = gl->gl_demote_state;
543         } else {
544                 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
545                         gfs2_demote_wake(gl);
546                 ret = do_promote(gl);
547                 if (ret == 0)
548                         goto out_unlock;
549                 if (ret == 2)
550                         goto out;
551                 gh = find_first_waiter(gl);
552                 gl->gl_target = gh->gh_state;
553                 if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
554                         do_error(gl, 0); /* Fail queued try locks */
555         }
556         do_xmote(gl, gh, gl->gl_target);
557 out:
558         return;
559
560 out_sched:
561         clear_bit(GLF_LOCK, &gl->gl_flags);
562         smp_mb__after_atomic();
563         gl->gl_lockref.count++;
564         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
565                 gl->gl_lockref.count--;
566         return;
567
568 out_unlock:
569         clear_bit(GLF_LOCK, &gl->gl_flags);
570         smp_mb__after_atomic();
571         return;
572 }
573
574 static void delete_work_func(struct work_struct *work)
575 {
576         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
577         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
578         struct inode *inode;
579         u64 no_addr = gl->gl_name.ln_number;
580
581         /* If someone's using this glock to create a new dinode, the block must
582            have been freed by another node, then re-used, in which case our
583            iopen callback is too late after the fact. Ignore it. */
584         if (test_bit(GLF_INODE_CREATING, &gl->gl_flags))
585                 goto out;
586
587         inode = gfs2_lookup_by_inum(sdp, no_addr, NULL, GFS2_BLKST_UNLINKED);
588         if (inode && !IS_ERR(inode)) {
589                 d_prune_aliases(inode);
590                 iput(inode);
591         }
592 out:
593         gfs2_glock_put(gl);
594 }
595
596 static void glock_work_func(struct work_struct *work)
597 {
598         unsigned long delay = 0;
599         struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
600         int drop_ref = 0;
601
602         if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
603                 finish_xmote(gl, gl->gl_reply);
604                 drop_ref = 1;
605         }
606         spin_lock(&gl->gl_lockref.lock);
607         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
608             gl->gl_state != LM_ST_UNLOCKED &&
609             gl->gl_demote_state != LM_ST_EXCLUSIVE) {
610                 unsigned long holdtime, now = jiffies;
611
612                 holdtime = gl->gl_tchange + gl->gl_hold_time;
613                 if (time_before(now, holdtime))
614                         delay = holdtime - now;
615
616                 if (!delay) {
617                         clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags);
618                         set_bit(GLF_DEMOTE, &gl->gl_flags);
619                 }
620         }
621         run_queue(gl, 0);
622         spin_unlock(&gl->gl_lockref.lock);
623         if (!delay)
624                 gfs2_glock_put(gl);
625         else {
626                 if (gl->gl_name.ln_type != LM_TYPE_INODE)
627                         delay = 0;
628                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
629                         gfs2_glock_put(gl);
630         }
631         if (drop_ref)
632                 gfs2_glock_put(gl);
633 }
634
635 /**
636  * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
637  * @sdp: The GFS2 superblock
638  * @number: the lock number
639  * @glops: The glock_operations to use
640  * @create: If 0, don't create the glock if it doesn't exist
641  * @glp: the glock is returned here
642  *
643  * This does not lock a glock, just finds/creates structures for one.
644  *
645  * Returns: errno
646  */
647
648 int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
649                    const struct gfs2_glock_operations *glops, int create,
650                    struct gfs2_glock **glp)
651 {
652         struct super_block *s = sdp->sd_vfs;
653         struct lm_lockname name = { .ln_number = number,
654                                     .ln_type = glops->go_type,
655                                     .ln_sbd = sdp };
656         struct gfs2_glock *gl, *tmp = NULL;
657         struct address_space *mapping;
658         struct kmem_cache *cachep;
659         int ret, tries = 0;
660
661         rcu_read_lock();
662         gl = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
663         if (gl && !lockref_get_not_dead(&gl->gl_lockref))
664                 gl = NULL;
665         rcu_read_unlock();
666
667         *glp = gl;
668         if (gl)
669                 return 0;
670         if (!create)
671                 return -ENOENT;
672
673         if (glops->go_flags & GLOF_ASPACE)
674                 cachep = gfs2_glock_aspace_cachep;
675         else
676                 cachep = gfs2_glock_cachep;
677         gl = kmem_cache_alloc(cachep, GFP_NOFS);
678         if (!gl)
679                 return -ENOMEM;
680
681         memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
682
683         if (glops->go_flags & GLOF_LVB) {
684                 gl->gl_lksb.sb_lvbptr = kzalloc(GFS2_MIN_LVB_SIZE, GFP_NOFS);
685                 if (!gl->gl_lksb.sb_lvbptr) {
686                         kmem_cache_free(cachep, gl);
687                         return -ENOMEM;
688                 }
689         }
690
691         atomic_inc(&sdp->sd_glock_disposal);
692         gl->gl_node.next = NULL;
693         gl->gl_flags = 0;
694         gl->gl_name = name;
695         gl->gl_lockref.count = 1;
696         gl->gl_state = LM_ST_UNLOCKED;
697         gl->gl_target = LM_ST_UNLOCKED;
698         gl->gl_demote_state = LM_ST_EXCLUSIVE;
699         gl->gl_ops = glops;
700         gl->gl_dstamp = 0;
701         preempt_disable();
702         /* We use the global stats to estimate the initial per-glock stats */
703         gl->gl_stats = this_cpu_ptr(sdp->sd_lkstats)->lkstats[glops->go_type];
704         preempt_enable();
705         gl->gl_stats.stats[GFS2_LKS_DCOUNT] = 0;
706         gl->gl_stats.stats[GFS2_LKS_QCOUNT] = 0;
707         gl->gl_tchange = jiffies;
708         gl->gl_object = NULL;
709         gl->gl_hold_time = GL_GLOCK_DFT_HOLD;
710         INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
711         INIT_WORK(&gl->gl_delete, delete_work_func);
712
713         mapping = gfs2_glock2aspace(gl);
714         if (mapping) {
715                 mapping->a_ops = &gfs2_meta_aops;
716                 mapping->host = s->s_bdev->bd_inode;
717                 mapping->flags = 0;
718                 mapping_set_gfp_mask(mapping, GFP_NOFS);
719                 mapping->private_data = NULL;
720                 mapping->writeback_index = 0;
721         }
722
723 again:
724         ret = rhashtable_lookup_insert_fast(&gl_hash_table, &gl->gl_node,
725                                             ht_parms);
726         if (ret == 0) {
727                 *glp = gl;
728                 return 0;
729         }
730
731         if (ret == -EEXIST) {
732                 ret = 0;
733                 rcu_read_lock();
734                 tmp = rhashtable_lookup_fast(&gl_hash_table, &name, ht_parms);
735                 if (tmp == NULL || !lockref_get_not_dead(&tmp->gl_lockref)) {
736                         if (++tries < 100) {
737                                 rcu_read_unlock();
738                                 cond_resched();
739                                 goto again;
740                         }
741                         tmp = NULL;
742                         ret = -ENOMEM;
743                 }
744                 rcu_read_unlock();
745         } else {
746                 WARN_ON_ONCE(ret);
747         }
748         kfree(gl->gl_lksb.sb_lvbptr);
749         kmem_cache_free(cachep, gl);
750         atomic_dec(&sdp->sd_glock_disposal);
751         *glp = tmp;
752
753         return ret;
754 }
755
756 /**
757  * gfs2_holder_init - initialize a struct gfs2_holder in the default way
758  * @gl: the glock
759  * @state: the state we're requesting
760  * @flags: the modifier flags
761  * @gh: the holder structure
762  *
763  */
764
765 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, u16 flags,
766                       struct gfs2_holder *gh)
767 {
768         INIT_LIST_HEAD(&gh->gh_list);
769         gh->gh_gl = gl;
770         gh->gh_ip = _RET_IP_;
771         gh->gh_owner_pid = get_pid(task_pid(current));
772         gh->gh_state = state;
773         gh->gh_flags = flags;
774         gh->gh_error = 0;
775         gh->gh_iflags = 0;
776         gfs2_glock_hold(gl);
777 }
778
779 /**
780  * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
781  * @state: the state we're requesting
782  * @flags: the modifier flags
783  * @gh: the holder structure
784  *
785  * Don't mess with the glock.
786  *
787  */
788
789 void gfs2_holder_reinit(unsigned int state, u16 flags, struct gfs2_holder *gh)
790 {
791         gh->gh_state = state;
792         gh->gh_flags = flags;
793         gh->gh_iflags = 0;
794         gh->gh_ip = _RET_IP_;
795         put_pid(gh->gh_owner_pid);
796         gh->gh_owner_pid = get_pid(task_pid(current));
797 }
798
799 /**
800  * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
801  * @gh: the holder structure
802  *
803  */
804
805 void gfs2_holder_uninit(struct gfs2_holder *gh)
806 {
807         put_pid(gh->gh_owner_pid);
808         gfs2_glock_put(gh->gh_gl);
809         gfs2_holder_mark_uninitialized(gh);
810         gh->gh_ip = 0;
811 }
812
813 /**
814  * gfs2_glock_wait - wait on a glock acquisition
815  * @gh: the glock holder
816  *
817  * Returns: 0 on success
818  */
819
820 int gfs2_glock_wait(struct gfs2_holder *gh)
821 {
822         unsigned long time1 = jiffies;
823
824         might_sleep();
825         wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE);
826         if (time_after(jiffies, time1 + HZ)) /* have we waited > a second? */
827                 /* Lengthen the minimum hold time. */
828                 gh->gh_gl->gl_hold_time = min(gh->gh_gl->gl_hold_time +
829                                               GL_GLOCK_HOLD_INCR,
830                                               GL_GLOCK_MAX_HOLD);
831         return gh->gh_error;
832 }
833
834 /**
835  * handle_callback - process a demote request
836  * @gl: the glock
837  * @state: the state the caller wants us to change to
838  *
839  * There are only two requests that we are going to see in actual
840  * practise: LM_ST_SHARED and LM_ST_UNLOCKED
841  */
842
843 static void handle_callback(struct gfs2_glock *gl, unsigned int state,
844                             unsigned long delay, bool remote)
845 {
846         int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
847
848         set_bit(bit, &gl->gl_flags);
849         if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
850                 gl->gl_demote_state = state;
851                 gl->gl_demote_time = jiffies;
852         } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
853                         gl->gl_demote_state != state) {
854                 gl->gl_demote_state = LM_ST_UNLOCKED;
855         }
856         if (gl->gl_ops->go_callback)
857                 gl->gl_ops->go_callback(gl, remote);
858         trace_gfs2_demote_rq(gl, remote);
859 }
860
861 void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
862 {
863         struct va_format vaf;
864         va_list args;
865
866         va_start(args, fmt);
867
868         if (seq) {
869                 seq_vprintf(seq, fmt, args);
870         } else {
871                 vaf.fmt = fmt;
872                 vaf.va = &args;
873
874                 pr_err("%pV", &vaf);
875         }
876
877         va_end(args);
878 }
879
880 /**
881  * add_to_queue - Add a holder to the wait queue (but look for recursion)
882  * @gh: the holder structure to add
883  *
884  * Eventually we should move the recursive locking trap to a
885  * debugging option or something like that. This is the fast
886  * path and needs to have the minimum number of distractions.
887  * 
888  */
889
890 static inline void add_to_queue(struct gfs2_holder *gh)
891 __releases(&gl->gl_lockref.lock)
892 __acquires(&gl->gl_lockref.lock)
893 {
894         struct gfs2_glock *gl = gh->gh_gl;
895         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
896         struct list_head *insert_pt = NULL;
897         struct gfs2_holder *gh2;
898         int try_futile = 0;
899
900         BUG_ON(gh->gh_owner_pid == NULL);
901         if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
902                 BUG();
903
904         if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
905                 if (test_bit(GLF_LOCK, &gl->gl_flags))
906                         try_futile = !may_grant(gl, gh);
907                 if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
908                         goto fail;
909         }
910
911         list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
912                 if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
913                     (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
914                         goto trap_recursive;
915                 if (try_futile &&
916                     !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
917 fail:
918                         gh->gh_error = GLR_TRYFAILED;
919                         gfs2_holder_wake(gh);
920                         return;
921                 }
922                 if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
923                         continue;
924                 if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
925                         insert_pt = &gh2->gh_list;
926         }
927         set_bit(GLF_QUEUED, &gl->gl_flags);
928         trace_gfs2_glock_queue(gh, 1);
929         gfs2_glstats_inc(gl, GFS2_LKS_QCOUNT);
930         gfs2_sbstats_inc(gl, GFS2_LKS_QCOUNT);
931         if (likely(insert_pt == NULL)) {
932                 list_add_tail(&gh->gh_list, &gl->gl_holders);
933                 if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
934                         goto do_cancel;
935                 return;
936         }
937         list_add_tail(&gh->gh_list, insert_pt);
938 do_cancel:
939         gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
940         if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
941                 spin_unlock(&gl->gl_lockref.lock);
942                 if (sdp->sd_lockstruct.ls_ops->lm_cancel)
943                         sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
944                 spin_lock(&gl->gl_lockref.lock);
945         }
946         return;
947
948 trap_recursive:
949         pr_err("original: %pSR\n", (void *)gh2->gh_ip);
950         pr_err("pid: %d\n", pid_nr(gh2->gh_owner_pid));
951         pr_err("lock type: %d req lock state : %d\n",
952                gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
953         pr_err("new: %pSR\n", (void *)gh->gh_ip);
954         pr_err("pid: %d\n", pid_nr(gh->gh_owner_pid));
955         pr_err("lock type: %d req lock state : %d\n",
956                gh->gh_gl->gl_name.ln_type, gh->gh_state);
957         gfs2_dump_glock(NULL, gl);
958         BUG();
959 }
960
961 /**
962  * gfs2_glock_nq - enqueue a struct gfs2_holder onto a glock (acquire a glock)
963  * @gh: the holder structure
964  *
965  * if (gh->gh_flags & GL_ASYNC), this never returns an error
966  *
967  * Returns: 0, GLR_TRYFAILED, or errno on failure
968  */
969
970 int gfs2_glock_nq(struct gfs2_holder *gh)
971 {
972         struct gfs2_glock *gl = gh->gh_gl;
973         struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
974         int error = 0;
975
976         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
977                 return -EIO;
978
979         if (test_bit(GLF_LRU, &gl->gl_flags))
980                 gfs2_glock_remove_from_lru(gl);
981
982         spin_lock(&gl->gl_lockref.lock);
983         add_to_queue(gh);
984         if (unlikely((LM_FLAG_NOEXP & gh->gh_flags) &&
985                      test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))) {
986                 set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
987                 gl->gl_lockref.count++;
988                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
989                         gl->gl_lockref.count--;
990         }
991         run_queue(gl, 1);
992         spin_unlock(&gl->gl_lockref.lock);
993
994         if (!(gh->gh_flags & GL_ASYNC))
995                 error = gfs2_glock_wait(gh);
996
997         return error;
998 }
999
1000 /**
1001  * gfs2_glock_poll - poll to see if an async request has been completed
1002  * @gh: the holder
1003  *
1004  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
1005  */
1006
1007 int gfs2_glock_poll(struct gfs2_holder *gh)
1008 {
1009         return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
1010 }
1011
1012 /**
1013  * gfs2_glock_dq - dequeue a struct gfs2_holder from a glock (release a glock)
1014  * @gh: the glock holder
1015  *
1016  */
1017
1018 void gfs2_glock_dq(struct gfs2_holder *gh)
1019 {
1020         struct gfs2_glock *gl = gh->gh_gl;
1021         const struct gfs2_glock_operations *glops = gl->gl_ops;
1022         unsigned delay = 0;
1023         int fast_path = 0;
1024
1025         spin_lock(&gl->gl_lockref.lock);
1026         if (gh->gh_flags & GL_NOCACHE)
1027                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1028
1029         list_del_init(&gh->gh_list);
1030         clear_bit(HIF_HOLDER, &gh->gh_iflags);
1031         if (find_first_holder(gl) == NULL) {
1032                 if (glops->go_unlock) {
1033                         GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
1034                         spin_unlock(&gl->gl_lockref.lock);
1035                         glops->go_unlock(gh);
1036                         spin_lock(&gl->gl_lockref.lock);
1037                         clear_bit(GLF_LOCK, &gl->gl_flags);
1038                 }
1039                 if (list_empty(&gl->gl_holders) &&
1040                     !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1041                     !test_bit(GLF_DEMOTE, &gl->gl_flags))
1042                         fast_path = 1;
1043         }
1044         if (!test_bit(GLF_LFLUSH, &gl->gl_flags) && demote_ok(gl) &&
1045             (glops->go_flags & GLOF_LRU))
1046                 gfs2_glock_add_to_lru(gl);
1047
1048         trace_gfs2_glock_queue(gh, 0);
1049         spin_unlock(&gl->gl_lockref.lock);
1050         if (likely(fast_path))
1051                 return;
1052
1053         gfs2_glock_hold(gl);
1054         if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
1055             !test_bit(GLF_DEMOTE, &gl->gl_flags) &&
1056             gl->gl_name.ln_type == LM_TYPE_INODE)
1057                 delay = gl->gl_hold_time;
1058         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1059                 gfs2_glock_put(gl);
1060 }
1061
1062 void gfs2_glock_dq_wait(struct gfs2_holder *gh)
1063 {
1064         struct gfs2_glock *gl = gh->gh_gl;
1065         gfs2_glock_dq(gh);
1066         might_sleep();
1067         wait_on_bit(&gl->gl_flags, GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
1068 }
1069
1070 /**
1071  * gfs2_glock_dq_uninit - dequeue a holder from a glock and initialize it
1072  * @gh: the holder structure
1073  *
1074  */
1075
1076 void gfs2_glock_dq_uninit(struct gfs2_holder *gh)
1077 {
1078         gfs2_glock_dq(gh);
1079         gfs2_holder_uninit(gh);
1080 }
1081
1082 /**
1083  * gfs2_glock_nq_num - acquire a glock based on lock number
1084  * @sdp: the filesystem
1085  * @number: the lock number
1086  * @glops: the glock operations for the type of glock
1087  * @state: the state to acquire the glock in
1088  * @flags: modifier flags for the acquisition
1089  * @gh: the struct gfs2_holder
1090  *
1091  * Returns: errno
1092  */
1093
1094 int gfs2_glock_nq_num(struct gfs2_sbd *sdp, u64 number,
1095                       const struct gfs2_glock_operations *glops,
1096                       unsigned int state, u16 flags, struct gfs2_holder *gh)
1097 {
1098         struct gfs2_glock *gl;
1099         int error;
1100
1101         error = gfs2_glock_get(sdp, number, glops, CREATE, &gl);
1102         if (!error) {
1103                 error = gfs2_glock_nq_init(gl, state, flags, gh);
1104                 gfs2_glock_put(gl);
1105         }
1106
1107         return error;
1108 }
1109
1110 /**
1111  * glock_compare - Compare two struct gfs2_glock structures for sorting
1112  * @arg_a: the first structure
1113  * @arg_b: the second structure
1114  *
1115  */
1116
1117 static int glock_compare(const void *arg_a, const void *arg_b)
1118 {
1119         const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1120         const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1121         const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1122         const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1123
1124         if (a->ln_number > b->ln_number)
1125                 return 1;
1126         if (a->ln_number < b->ln_number)
1127                 return -1;
1128         BUG_ON(gh_a->gh_gl->gl_ops->go_type == gh_b->gh_gl->gl_ops->go_type);
1129         return 0;
1130 }
1131
1132 /**
1133  * nq_m_sync - synchonously acquire more than one glock in deadlock free order
1134  * @num_gh: the number of structures
1135  * @ghs: an array of struct gfs2_holder structures
1136  *
1137  * Returns: 0 on success (all glocks acquired),
1138  *          errno on failure (no glocks acquired)
1139  */
1140
1141 static int nq_m_sync(unsigned int num_gh, struct gfs2_holder *ghs,
1142                      struct gfs2_holder **p)
1143 {
1144         unsigned int x;
1145         int error = 0;
1146
1147         for (x = 0; x < num_gh; x++)
1148                 p[x] = &ghs[x];
1149
1150         sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare, NULL);
1151
1152         for (x = 0; x < num_gh; x++) {
1153                 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1154
1155                 error = gfs2_glock_nq(p[x]);
1156                 if (error) {
1157                         while (x--)
1158                                 gfs2_glock_dq(p[x]);
1159                         break;
1160                 }
1161         }
1162
1163         return error;
1164 }
1165
1166 /**
1167  * gfs2_glock_nq_m - acquire multiple glocks
1168  * @num_gh: the number of structures
1169  * @ghs: an array of struct gfs2_holder structures
1170  *
1171  *
1172  * Returns: 0 on success (all glocks acquired),
1173  *          errno on failure (no glocks acquired)
1174  */
1175
1176 int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1177 {
1178         struct gfs2_holder *tmp[4];
1179         struct gfs2_holder **pph = tmp;
1180         int error = 0;
1181
1182         switch(num_gh) {
1183         case 0:
1184                 return 0;
1185         case 1:
1186                 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1187                 return gfs2_glock_nq(ghs);
1188         default:
1189                 if (num_gh <= 4)
1190                         break;
1191                 pph = kmalloc(num_gh * sizeof(struct gfs2_holder *), GFP_NOFS);
1192                 if (!pph)
1193                         return -ENOMEM;
1194         }
1195
1196         error = nq_m_sync(num_gh, ghs, pph);
1197
1198         if (pph != tmp)
1199                 kfree(pph);
1200
1201         return error;
1202 }
1203
1204 /**
1205  * gfs2_glock_dq_m - release multiple glocks
1206  * @num_gh: the number of structures
1207  * @ghs: an array of struct gfs2_holder structures
1208  *
1209  */
1210
1211 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
1212 {
1213         while (num_gh--)
1214                 gfs2_glock_dq(&ghs[num_gh]);
1215 }
1216
1217 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
1218 {
1219         unsigned long delay = 0;
1220         unsigned long holdtime;
1221         unsigned long now = jiffies;
1222
1223         gfs2_glock_hold(gl);
1224         holdtime = gl->gl_tchange + gl->gl_hold_time;
1225         if (test_bit(GLF_QUEUED, &gl->gl_flags) &&
1226             gl->gl_name.ln_type == LM_TYPE_INODE) {
1227                 if (time_before(now, holdtime))
1228                         delay = holdtime - now;
1229                 if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
1230                         delay = gl->gl_hold_time;
1231         }
1232
1233         spin_lock(&gl->gl_lockref.lock);
1234         handle_callback(gl, state, delay, true);
1235         spin_unlock(&gl->gl_lockref.lock);
1236         if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
1237                 gfs2_glock_put(gl);
1238 }
1239
1240 /**
1241  * gfs2_should_freeze - Figure out if glock should be frozen
1242  * @gl: The glock in question
1243  *
1244  * Glocks are not frozen if (a) the result of the dlm operation is
1245  * an error, (b) the locking operation was an unlock operation or
1246  * (c) if there is a "noexp" flagged request anywhere in the queue
1247  *
1248  * Returns: 1 if freezing should occur, 0 otherwise
1249  */
1250
1251 static int gfs2_should_freeze(const struct gfs2_glock *gl)
1252 {
1253         const struct gfs2_holder *gh;
1254
1255         if (gl->gl_reply & ~LM_OUT_ST_MASK)
1256                 return 0;
1257         if (gl->gl_target == LM_ST_UNLOCKED)
1258                 return 0;
1259
1260         list_for_each_entry(gh, &gl->gl_holders, gh_list) {
1261                 if (test_bit(HIF_HOLDER, &gh->gh_iflags))
1262                         continue;
1263                 if (LM_FLAG_NOEXP & gh->gh_flags)
1264                         return 0;
1265         }
1266
1267         return 1;
1268 }
1269
1270 /**
1271  * gfs2_glock_complete - Callback used by locking
1272  * @gl: Pointer to the glock
1273  * @ret: The return value from the dlm
1274  *
1275  * The gl_reply field is under the gl_lockref.lock lock so that it is ok
1276  * to use a bitfield shared with other glock state fields.
1277  */
1278
1279 void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
1280 {
1281         struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
1282
1283         spin_lock(&gl->gl_lockref.lock);
1284         gl->gl_reply = ret;
1285
1286         if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags))) {
1287                 if (gfs2_should_freeze(gl)) {
1288                         set_bit(GLF_FROZEN, &gl->gl_flags);
1289                         spin_unlock(&gl->gl_lockref.lock);
1290                         return;
1291                 }
1292         }
1293
1294         gl->gl_lockref.count++;
1295         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1296         spin_unlock(&gl->gl_lockref.lock);
1297
1298         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1299                 gfs2_glock_put(gl);
1300 }
1301
1302 static int glock_cmp(void *priv, struct list_head *a, struct list_head *b)
1303 {
1304         struct gfs2_glock *gla, *glb;
1305
1306         gla = list_entry(a, struct gfs2_glock, gl_lru);
1307         glb = list_entry(b, struct gfs2_glock, gl_lru);
1308
1309         if (gla->gl_name.ln_number > glb->gl_name.ln_number)
1310                 return 1;
1311         if (gla->gl_name.ln_number < glb->gl_name.ln_number)
1312                 return -1;
1313
1314         return 0;
1315 }
1316
1317 /**
1318  * gfs2_dispose_glock_lru - Demote a list of glocks
1319  * @list: The list to dispose of
1320  *
1321  * Disposing of glocks may involve disk accesses, so that here we sort
1322  * the glocks by number (i.e. disk location of the inodes) so that if
1323  * there are any such accesses, they'll be sent in order (mostly).
1324  *
1325  * Must be called under the lru_lock, but may drop and retake this
1326  * lock. While the lru_lock is dropped, entries may vanish from the
1327  * list, but no new entries will appear on the list (since it is
1328  * private)
1329  */
1330
1331 static void gfs2_dispose_glock_lru(struct list_head *list)
1332 __releases(&lru_lock)
1333 __acquires(&lru_lock)
1334 {
1335         struct gfs2_glock *gl;
1336
1337         list_sort(NULL, list, glock_cmp);
1338
1339         while(!list_empty(list)) {
1340                 gl = list_entry(list->next, struct gfs2_glock, gl_lru);
1341                 list_del_init(&gl->gl_lru);
1342                 if (!spin_trylock(&gl->gl_lockref.lock)) {
1343 add_back_to_lru:
1344                         list_add(&gl->gl_lru, &lru_list);
1345                         atomic_inc(&lru_count);
1346                         continue;
1347                 }
1348                 if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
1349                         spin_unlock(&gl->gl_lockref.lock);
1350                         goto add_back_to_lru;
1351                 }
1352                 clear_bit(GLF_LRU, &gl->gl_flags);
1353                 gl->gl_lockref.count++;
1354                 if (demote_ok(gl))
1355                         handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1356                 WARN_ON(!test_and_clear_bit(GLF_LOCK, &gl->gl_flags));
1357                 if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1358                         gl->gl_lockref.count--;
1359                 spin_unlock(&gl->gl_lockref.lock);
1360                 cond_resched_lock(&lru_lock);
1361         }
1362 }
1363
1364 /**
1365  * gfs2_scan_glock_lru - Scan the LRU looking for locks to demote
1366  * @nr: The number of entries to scan
1367  *
1368  * This function selects the entries on the LRU which are able to
1369  * be demoted, and then kicks off the process by calling
1370  * gfs2_dispose_glock_lru() above.
1371  */
1372
1373 static long gfs2_scan_glock_lru(int nr)
1374 {
1375         struct gfs2_glock *gl;
1376         LIST_HEAD(skipped);
1377         LIST_HEAD(dispose);
1378         long freed = 0;
1379
1380         spin_lock(&lru_lock);
1381         while ((nr-- >= 0) && !list_empty(&lru_list)) {
1382                 gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
1383
1384                 /* Test for being demotable */
1385                 if (!test_bit(GLF_LOCK, &gl->gl_flags)) {
1386                         list_move(&gl->gl_lru, &dispose);
1387                         atomic_dec(&lru_count);
1388                         freed++;
1389                         continue;
1390                 }
1391
1392                 list_move(&gl->gl_lru, &skipped);
1393         }
1394         list_splice(&skipped, &lru_list);
1395         if (!list_empty(&dispose))
1396                 gfs2_dispose_glock_lru(&dispose);
1397         spin_unlock(&lru_lock);
1398
1399         return freed;
1400 }
1401
1402 static unsigned long gfs2_glock_shrink_scan(struct shrinker *shrink,
1403                                             struct shrink_control *sc)
1404 {
1405         if (!(sc->gfp_mask & __GFP_FS))
1406                 return SHRINK_STOP;
1407         return gfs2_scan_glock_lru(sc->nr_to_scan);
1408 }
1409
1410 static unsigned long gfs2_glock_shrink_count(struct shrinker *shrink,
1411                                              struct shrink_control *sc)
1412 {
1413         return vfs_pressure_ratio(atomic_read(&lru_count));
1414 }
1415
1416 static struct shrinker glock_shrinker = {
1417         .seeks = DEFAULT_SEEKS,
1418         .count_objects = gfs2_glock_shrink_count,
1419         .scan_objects = gfs2_glock_shrink_scan,
1420 };
1421
1422 /**
1423  * examine_bucket - Call a function for glock in a hash bucket
1424  * @examiner: the function
1425  * @sdp: the filesystem
1426  * @bucket: the bucket
1427  *
1428  * Note that the function can be called multiple times on the same
1429  * object.  So the user must ensure that the function can cope with
1430  * that.
1431  */
1432
1433 static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
1434 {
1435         struct gfs2_glock *gl;
1436         struct rhashtable_iter iter;
1437
1438         rhashtable_walk_enter(&gl_hash_table, &iter);
1439
1440         do {
1441                 gl = ERR_PTR(rhashtable_walk_start(&iter));
1442                 if (gl)
1443                         continue;
1444
1445                 while ((gl = rhashtable_walk_next(&iter)) && !IS_ERR(gl))
1446                         if ((gl->gl_name.ln_sbd == sdp) &&
1447                             lockref_get_not_dead(&gl->gl_lockref))
1448                                 examiner(gl);
1449
1450                 rhashtable_walk_stop(&iter);
1451         } while (cond_resched(), gl == ERR_PTR(-EAGAIN));
1452
1453         rhashtable_walk_exit(&iter);
1454 }
1455
1456 /**
1457  * thaw_glock - thaw out a glock which has an unprocessed reply waiting
1458  * @gl: The glock to thaw
1459  *
1460  */
1461
1462 static void thaw_glock(struct gfs2_glock *gl)
1463 {
1464         if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
1465                 goto out;
1466         set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
1467         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0) {
1468 out:
1469                 gfs2_glock_put(gl);
1470         }
1471 }
1472
1473 /**
1474  * clear_glock - look at a glock and see if we can free it from glock cache
1475  * @gl: the glock to look at
1476  *
1477  */
1478
1479 static void clear_glock(struct gfs2_glock *gl)
1480 {
1481         gfs2_glock_remove_from_lru(gl);
1482
1483         spin_lock(&gl->gl_lockref.lock);
1484         if (gl->gl_state != LM_ST_UNLOCKED)
1485                 handle_callback(gl, LM_ST_UNLOCKED, 0, false);
1486         spin_unlock(&gl->gl_lockref.lock);
1487         if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
1488                 gfs2_glock_put(gl);
1489 }
1490
1491 /**
1492  * gfs2_glock_thaw - Thaw any frozen glocks
1493  * @sdp: The super block
1494  *
1495  */
1496
1497 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
1498 {
1499         glock_hash_walk(thaw_glock, sdp);
1500 }
1501
1502 static void dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
1503 {
1504         spin_lock(&gl->gl_lockref.lock);
1505         gfs2_dump_glock(seq, gl);
1506         spin_unlock(&gl->gl_lockref.lock);
1507 }
1508
1509 static void dump_glock_func(struct gfs2_glock *gl)
1510 {
1511         dump_glock(NULL, gl);
1512 }
1513
1514 /**
1515  * gfs2_gl_hash_clear - Empty out the glock hash table
1516  * @sdp: the filesystem
1517  * @wait: wait until it's all gone
1518  *
1519  * Called when unmounting the filesystem.
1520  */
1521
1522 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
1523 {
1524         set_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags);
1525         flush_workqueue(glock_workqueue);
1526         glock_hash_walk(clear_glock, sdp);
1527         flush_workqueue(glock_workqueue);
1528         wait_event_timeout(sdp->sd_glock_wait,
1529                            atomic_read(&sdp->sd_glock_disposal) == 0,
1530                            HZ * 600);
1531         glock_hash_walk(dump_glock_func, sdp);
1532 }
1533
1534 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
1535 {
1536         struct gfs2_glock *gl = ip->i_gl;
1537         int ret;
1538
1539         ret = gfs2_truncatei_resume(ip);
1540         gfs2_assert_withdraw(gl->gl_name.ln_sbd, ret == 0);
1541
1542         spin_lock(&gl->gl_lockref.lock);
1543         clear_bit(GLF_LOCK, &gl->gl_flags);
1544         run_queue(gl, 1);
1545         spin_unlock(&gl->gl_lockref.lock);
1546 }
1547
1548 static const char *state2str(unsigned state)
1549 {
1550         switch(state) {
1551         case LM_ST_UNLOCKED:
1552                 return "UN";
1553         case LM_ST_SHARED:
1554                 return "SH";
1555         case LM_ST_DEFERRED:
1556                 return "DF";
1557         case LM_ST_EXCLUSIVE:
1558                 return "EX";
1559         }
1560         return "??";
1561 }
1562
1563 static const char *hflags2str(char *buf, u16 flags, unsigned long iflags)
1564 {
1565         char *p = buf;
1566         if (flags & LM_FLAG_TRY)
1567                 *p++ = 't';
1568         if (flags & LM_FLAG_TRY_1CB)
1569                 *p++ = 'T';
1570         if (flags & LM_FLAG_NOEXP)
1571                 *p++ = 'e';
1572         if (flags & LM_FLAG_ANY)
1573                 *p++ = 'A';
1574         if (flags & LM_FLAG_PRIORITY)
1575                 *p++ = 'p';
1576         if (flags & GL_ASYNC)
1577                 *p++ = 'a';
1578         if (flags & GL_EXACT)
1579                 *p++ = 'E';
1580         if (flags & GL_NOCACHE)
1581                 *p++ = 'c';
1582         if (test_bit(HIF_HOLDER, &iflags))
1583                 *p++ = 'H';
1584         if (test_bit(HIF_WAIT, &iflags))
1585                 *p++ = 'W';
1586         if (test_bit(HIF_FIRST, &iflags))
1587                 *p++ = 'F';
1588         *p = 0;
1589         return buf;
1590 }
1591
1592 /**
1593  * dump_holder - print information about a glock holder
1594  * @seq: the seq_file struct
1595  * @gh: the glock holder
1596  *
1597  */
1598
1599 static void dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
1600 {
1601         struct task_struct *gh_owner = NULL;
1602         char flags_buf[32];
1603
1604         rcu_read_lock();
1605         if (gh->gh_owner_pid)
1606                 gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
1607         gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %pS\n",
1608                        state2str(gh->gh_state),
1609                        hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
1610                        gh->gh_error,
1611                        gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
1612                        gh_owner ? gh_owner->comm : "(ended)",
1613                        (void *)gh->gh_ip);
1614         rcu_read_unlock();
1615 }
1616
1617 static const char *gflags2str(char *buf, const struct gfs2_glock *gl)
1618 {
1619         const unsigned long *gflags = &gl->gl_flags;
1620         char *p = buf;
1621
1622         if (test_bit(GLF_LOCK, gflags))
1623                 *p++ = 'l';
1624         if (test_bit(GLF_DEMOTE, gflags))
1625                 *p++ = 'D';
1626         if (test_bit(GLF_PENDING_DEMOTE, gflags))
1627                 *p++ = 'd';
1628         if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
1629                 *p++ = 'p';
1630         if (test_bit(GLF_DIRTY, gflags))
1631                 *p++ = 'y';
1632         if (test_bit(GLF_LFLUSH, gflags))
1633                 *p++ = 'f';
1634         if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
1635                 *p++ = 'i';
1636         if (test_bit(GLF_REPLY_PENDING, gflags))
1637                 *p++ = 'r';
1638         if (test_bit(GLF_INITIAL, gflags))
1639                 *p++ = 'I';
1640         if (test_bit(GLF_FROZEN, gflags))
1641                 *p++ = 'F';
1642         if (test_bit(GLF_QUEUED, gflags))
1643                 *p++ = 'q';
1644         if (test_bit(GLF_LRU, gflags))
1645                 *p++ = 'L';
1646         if (gl->gl_object)
1647                 *p++ = 'o';
1648         if (test_bit(GLF_BLOCKING, gflags))
1649                 *p++ = 'b';
1650         *p = 0;
1651         return buf;
1652 }
1653
1654 /**
1655  * gfs2_dump_glock - print information about a glock
1656  * @seq: The seq_file struct
1657  * @gl: the glock
1658  *
1659  * The file format is as follows:
1660  * One line per object, capital letters are used to indicate objects
1661  * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
1662  * other objects are indented by a single space and follow the glock to
1663  * which they are related. Fields are indicated by lower case letters
1664  * followed by a colon and the field value, except for strings which are in
1665  * [] so that its possible to see if they are composed of spaces for
1666  * example. The field's are n = number (id of the object), f = flags,
1667  * t = type, s = state, r = refcount, e = error, p = pid.
1668  *
1669  */
1670
1671 void gfs2_dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
1672 {
1673         const struct gfs2_glock_operations *glops = gl->gl_ops;
1674         unsigned long long dtime;
1675         const struct gfs2_holder *gh;
1676         char gflags_buf[32];
1677
1678         dtime = jiffies - gl->gl_demote_time;
1679         dtime *= 1000000/HZ; /* demote time in uSec */
1680         if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
1681                 dtime = 0;
1682         gfs2_print_dbg(seq, "G:  s:%s n:%u/%llx f:%s t:%s d:%s/%llu a:%d v:%d r:%d m:%ld\n",
1683                   state2str(gl->gl_state),
1684                   gl->gl_name.ln_type,
1685                   (unsigned long long)gl->gl_name.ln_number,
1686                   gflags2str(gflags_buf, gl),
1687                   state2str(gl->gl_target),
1688                   state2str(gl->gl_demote_state), dtime,
1689                   atomic_read(&gl->gl_ail_count),
1690                   atomic_read(&gl->gl_revokes),
1691                   (int)gl->gl_lockref.count, gl->gl_hold_time);
1692
1693         list_for_each_entry(gh, &gl->gl_holders, gh_list)
1694                 dump_holder(seq, gh);
1695
1696         if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
1697                 glops->go_dump(seq, gl);
1698 }
1699
1700 static int gfs2_glstats_seq_show(struct seq_file *seq, void *iter_ptr)
1701 {
1702         struct gfs2_glock *gl = iter_ptr;
1703
1704         seq_printf(seq, "G: n:%u/%llx rtt:%llu/%llu rttb:%llu/%llu irt:%llu/%llu dcnt: %llu qcnt: %llu\n",
1705                    gl->gl_name.ln_type,
1706                    (unsigned long long)gl->gl_name.ln_number,
1707                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTT],
1708                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVAR],
1709                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTB],
1710                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SRTTVARB],
1711                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRT],
1712                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_SIRTVAR],
1713                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_DCOUNT],
1714                    (unsigned long long)gl->gl_stats.stats[GFS2_LKS_QCOUNT]);
1715         return 0;
1716 }
1717
1718 static const char *gfs2_gltype[] = {
1719         "type",
1720         "reserved",
1721         "nondisk",
1722         "inode",
1723         "rgrp",
1724         "meta",
1725         "iopen",
1726         "flock",
1727         "plock",
1728         "quota",
1729         "journal",
1730 };
1731
1732 static const char *gfs2_stype[] = {
1733         [GFS2_LKS_SRTT]         = "srtt",
1734         [GFS2_LKS_SRTTVAR]      = "srttvar",
1735         [GFS2_LKS_SRTTB]        = "srttb",
1736         [GFS2_LKS_SRTTVARB]     = "srttvarb",
1737         [GFS2_LKS_SIRT]         = "sirt",
1738         [GFS2_LKS_SIRTVAR]      = "sirtvar",
1739         [GFS2_LKS_DCOUNT]       = "dlm",
1740         [GFS2_LKS_QCOUNT]       = "queue",
1741 };
1742
1743 #define GFS2_NR_SBSTATS (ARRAY_SIZE(gfs2_gltype) * ARRAY_SIZE(gfs2_stype))
1744
1745 static int gfs2_sbstats_seq_show(struct seq_file *seq, void *iter_ptr)
1746 {
1747         struct gfs2_sbd *sdp = seq->private;
1748         loff_t pos = *(loff_t *)iter_ptr;
1749         unsigned index = pos >> 3;
1750         unsigned subindex = pos & 0x07;
1751         int i;
1752
1753         if (index == 0 && subindex != 0)
1754                 return 0;
1755
1756         seq_printf(seq, "%-10s %8s:", gfs2_gltype[index],
1757                    (index == 0) ? "cpu": gfs2_stype[subindex]);
1758
1759         for_each_possible_cpu(i) {
1760                 const struct gfs2_pcpu_lkstats *lkstats = per_cpu_ptr(sdp->sd_lkstats, i);
1761
1762                 if (index == 0)
1763                         seq_printf(seq, " %15u", i);
1764                 else
1765                         seq_printf(seq, " %15llu", (unsigned long long)lkstats->
1766                                    lkstats[index - 1].stats[subindex]);
1767         }
1768         seq_putc(seq, '\n');
1769         return 0;
1770 }
1771
1772 int __init gfs2_glock_init(void)
1773 {
1774         int ret;
1775
1776         ret = rhashtable_init(&gl_hash_table, &ht_parms);
1777         if (ret < 0)
1778                 return ret;
1779
1780         glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
1781                                           WQ_HIGHPRI | WQ_FREEZABLE, 0);
1782         if (!glock_workqueue) {
1783                 rhashtable_destroy(&gl_hash_table);
1784                 return -ENOMEM;
1785         }
1786         gfs2_delete_workqueue = alloc_workqueue("delete_workqueue",
1787                                                 WQ_MEM_RECLAIM | WQ_FREEZABLE,
1788                                                 0);
1789         if (!gfs2_delete_workqueue) {
1790                 destroy_workqueue(glock_workqueue);
1791                 rhashtable_destroy(&gl_hash_table);
1792                 return -ENOMEM;
1793         }
1794
1795         ret = register_shrinker(&glock_shrinker);
1796         if (ret) {
1797                 destroy_workqueue(gfs2_delete_workqueue);
1798                 destroy_workqueue(glock_workqueue);
1799                 rhashtable_destroy(&gl_hash_table);
1800                 return ret;
1801         }
1802
1803         return 0;
1804 }
1805
1806 void gfs2_glock_exit(void)
1807 {
1808         unregister_shrinker(&glock_shrinker);
1809         rhashtable_destroy(&gl_hash_table);
1810         destroy_workqueue(glock_workqueue);
1811         destroy_workqueue(gfs2_delete_workqueue);
1812 }
1813
1814 static void gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
1815 {
1816         while ((gi->gl = rhashtable_walk_next(&gi->hti))) {
1817                 if (IS_ERR(gi->gl)) {
1818                         if (PTR_ERR(gi->gl) == -EAGAIN)
1819                                 continue;
1820                         gi->gl = NULL;
1821                         return;
1822                 }
1823                 /* Skip entries for other sb and dead entries */
1824                 if (gi->sdp == gi->gl->gl_name.ln_sbd &&
1825                     !__lockref_is_dead(&gi->gl->gl_lockref))
1826                         return;
1827         }
1828 }
1829
1830 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
1831 {
1832         struct gfs2_glock_iter *gi = seq->private;
1833         loff_t n = *pos;
1834         int ret;
1835
1836         if (gi->last_pos <= *pos)
1837                 n = (*pos - gi->last_pos);
1838
1839         ret = rhashtable_walk_start(&gi->hti);
1840         if (ret)
1841                 return NULL;
1842
1843         do {
1844                 gfs2_glock_iter_next(gi);
1845         } while (gi->gl && n--);
1846
1847         gi->last_pos = *pos;
1848         return gi->gl;
1849 }
1850
1851 static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
1852                                  loff_t *pos)
1853 {
1854         struct gfs2_glock_iter *gi = seq->private;
1855
1856         (*pos)++;
1857         gi->last_pos = *pos;
1858         gfs2_glock_iter_next(gi);
1859         return gi->gl;
1860 }
1861
1862 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
1863 {
1864         struct gfs2_glock_iter *gi = seq->private;
1865
1866         gi->gl = NULL;
1867         rhashtable_walk_stop(&gi->hti);
1868 }
1869
1870 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
1871 {
1872         dump_glock(seq, iter_ptr);
1873         return 0;
1874 }
1875
1876 static void *gfs2_sbstats_seq_start(struct seq_file *seq, loff_t *pos)
1877 {
1878         preempt_disable();
1879         if (*pos >= GFS2_NR_SBSTATS)
1880                 return NULL;
1881         return pos;
1882 }
1883
1884 static void *gfs2_sbstats_seq_next(struct seq_file *seq, void *iter_ptr,
1885                                    loff_t *pos)
1886 {
1887         (*pos)++;
1888         if (*pos >= GFS2_NR_SBSTATS)
1889                 return NULL;
1890         return pos;
1891 }
1892
1893 static void gfs2_sbstats_seq_stop(struct seq_file *seq, void *iter_ptr)
1894 {
1895         preempt_enable();
1896 }
1897
1898 static const struct seq_operations gfs2_glock_seq_ops = {
1899         .start = gfs2_glock_seq_start,
1900         .next  = gfs2_glock_seq_next,
1901         .stop  = gfs2_glock_seq_stop,
1902         .show  = gfs2_glock_seq_show,
1903 };
1904
1905 static const struct seq_operations gfs2_glstats_seq_ops = {
1906         .start = gfs2_glock_seq_start,
1907         .next  = gfs2_glock_seq_next,
1908         .stop  = gfs2_glock_seq_stop,
1909         .show  = gfs2_glstats_seq_show,
1910 };
1911
1912 static const struct seq_operations gfs2_sbstats_seq_ops = {
1913         .start = gfs2_sbstats_seq_start,
1914         .next  = gfs2_sbstats_seq_next,
1915         .stop  = gfs2_sbstats_seq_stop,
1916         .show  = gfs2_sbstats_seq_show,
1917 };
1918
1919 #define GFS2_SEQ_GOODSIZE min(PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER, 65536UL)
1920
1921 static int gfs2_glocks_open(struct inode *inode, struct file *file)
1922 {
1923         int ret = seq_open_private(file, &gfs2_glock_seq_ops,
1924                                    sizeof(struct gfs2_glock_iter));
1925         if (ret == 0) {
1926                 struct seq_file *seq = file->private_data;
1927                 struct gfs2_glock_iter *gi = seq->private;
1928
1929                 gi->sdp = inode->i_private;
1930                 gi->last_pos = 0;
1931                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1932                 if (seq->buf)
1933                         seq->size = GFS2_SEQ_GOODSIZE;
1934                 gi->gl = NULL;
1935                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti, GFP_KERNEL);
1936         }
1937         return ret;
1938 }
1939
1940 static int gfs2_glocks_release(struct inode *inode, struct file *file)
1941 {
1942         struct seq_file *seq = file->private_data;
1943         struct gfs2_glock_iter *gi = seq->private;
1944
1945         gi->gl = NULL;
1946         rhashtable_walk_exit(&gi->hti);
1947         return seq_release_private(inode, file);
1948 }
1949
1950 static int gfs2_glstats_open(struct inode *inode, struct file *file)
1951 {
1952         int ret = seq_open_private(file, &gfs2_glstats_seq_ops,
1953                                    sizeof(struct gfs2_glock_iter));
1954         if (ret == 0) {
1955                 struct seq_file *seq = file->private_data;
1956                 struct gfs2_glock_iter *gi = seq->private;
1957                 gi->sdp = inode->i_private;
1958                 gi->last_pos = 0;
1959                 seq->buf = kmalloc(GFS2_SEQ_GOODSIZE, GFP_KERNEL | __GFP_NOWARN);
1960                 if (seq->buf)
1961                         seq->size = GFS2_SEQ_GOODSIZE;
1962                 gi->gl = NULL;
1963                 ret = rhashtable_walk_init(&gl_hash_table, &gi->hti, GFP_KERNEL);
1964         }
1965         return ret;
1966 }
1967
1968 static int gfs2_sbstats_open(struct inode *inode, struct file *file)
1969 {
1970         int ret = seq_open(file, &gfs2_sbstats_seq_ops);
1971         if (ret == 0) {
1972                 struct seq_file *seq = file->private_data;
1973                 seq->private = inode->i_private;  /* sdp */
1974         }
1975         return ret;
1976 }
1977
1978 static const struct file_operations gfs2_glocks_fops = {
1979         .owner   = THIS_MODULE,
1980         .open    = gfs2_glocks_open,
1981         .read    = seq_read,
1982         .llseek  = seq_lseek,
1983         .release = gfs2_glocks_release,
1984 };
1985
1986 static const struct file_operations gfs2_glstats_fops = {
1987         .owner   = THIS_MODULE,
1988         .open    = gfs2_glstats_open,
1989         .read    = seq_read,
1990         .llseek  = seq_lseek,
1991         .release = gfs2_glocks_release,
1992 };
1993
1994 static const struct file_operations gfs2_sbstats_fops = {
1995         .owner   = THIS_MODULE,
1996         .open    = gfs2_sbstats_open,
1997         .read    = seq_read,
1998         .llseek  = seq_lseek,
1999         .release = seq_release,
2000 };
2001
2002 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)
2003 {
2004         struct dentry *dent;
2005
2006         dent = debugfs_create_dir(sdp->sd_table_name, gfs2_root);
2007         if (IS_ERR_OR_NULL(dent))
2008                 goto fail;
2009         sdp->debugfs_dir = dent;
2010
2011         dent = debugfs_create_file("glocks",
2012                                    S_IFREG | S_IRUGO,
2013                                    sdp->debugfs_dir, sdp,
2014                                    &gfs2_glocks_fops);
2015         if (IS_ERR_OR_NULL(dent))
2016                 goto fail;
2017         sdp->debugfs_dentry_glocks = dent;
2018
2019         dent = debugfs_create_file("glstats",
2020                                    S_IFREG | S_IRUGO,
2021                                    sdp->debugfs_dir, sdp,
2022                                    &gfs2_glstats_fops);
2023         if (IS_ERR_OR_NULL(dent))
2024                 goto fail;
2025         sdp->debugfs_dentry_glstats = dent;
2026
2027         dent = debugfs_create_file("sbstats",
2028                                    S_IFREG | S_IRUGO,
2029                                    sdp->debugfs_dir, sdp,
2030                                    &gfs2_sbstats_fops);
2031         if (IS_ERR_OR_NULL(dent))
2032                 goto fail;
2033         sdp->debugfs_dentry_sbstats = dent;
2034
2035         return 0;
2036 fail:
2037         gfs2_delete_debugfs_file(sdp);
2038         return dent ? PTR_ERR(dent) : -ENOMEM;
2039 }
2040
2041 void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp)
2042 {
2043         if (sdp->debugfs_dir) {
2044                 if (sdp->debugfs_dentry_glocks) {
2045                         debugfs_remove(sdp->debugfs_dentry_glocks);
2046                         sdp->debugfs_dentry_glocks = NULL;
2047                 }
2048                 if (sdp->debugfs_dentry_glstats) {
2049                         debugfs_remove(sdp->debugfs_dentry_glstats);
2050                         sdp->debugfs_dentry_glstats = NULL;
2051                 }
2052                 if (sdp->debugfs_dentry_sbstats) {
2053                         debugfs_remove(sdp->debugfs_dentry_sbstats);
2054                         sdp->debugfs_dentry_sbstats = NULL;
2055                 }
2056                 debugfs_remove(sdp->debugfs_dir);
2057                 sdp->debugfs_dir = NULL;
2058         }
2059 }
2060
2061 int gfs2_register_debugfs(void)
2062 {
2063         gfs2_root = debugfs_create_dir("gfs2", NULL);
2064         if (IS_ERR(gfs2_root))
2065                 return PTR_ERR(gfs2_root);
2066         return gfs2_root ? 0 : -ENOMEM;
2067 }
2068
2069 void gfs2_unregister_debugfs(void)
2070 {
2071         debugfs_remove(gfs2_root);
2072         gfs2_root = NULL;
2073 }