projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: Encode i_generation in the meta data lvb
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / ocfs2 / dlmglue.c
CommitLineData
ccd979bd
MF		 1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmglue.c
5 *
6 * Code which implements an OCFS2 specific interface to our DLM.
7 *
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/types.h>
27#include <linux/slab.h>
28#include <linux/highmem.h>
29#include <linux/mm.h>
30#include <linux/smp_lock.h>
31#include <linux/crc32.h>
32#include <linux/kthread.h>
33#include <linux/pagemap.h>
34#include <linux/debugfs.h>
35#include <linux/seq_file.h>
36
37#include <cluster/heartbeat.h>
38#include <cluster/nodemanager.h>
39#include <cluster/tcp.h>
40
41#include <dlm/dlmapi.h>
42
43#define MLOG_MASK_PREFIX ML_DLM_GLUE
44#include <cluster/masklog.h>
45
46#include "ocfs2.h"
47
48#include "alloc.h"
d680efe9 49#include "dcache.h"
ccd979bd
MF		 50#include "dlmglue.h"
51#include "extent_map.h"
52#include "heartbeat.h"
53#include "inode.h"
54#include "journal.h"
55#include "slot_map.h"
56#include "super.h"
57#include "uptodate.h"
58#include "vote.h"
59
60#include "buffer_head_io.h"
61
62struct ocfs2_mask_waiter {
63 struct list_head mw_item;
64 int mw_status;
65 struct completion mw_complete;
66 unsigned long mw_mask;
67 unsigned long mw_goal;
68};
69
70static void ocfs2_inode_ast_func(void *opaque);
71static void ocfs2_inode_bast_func(void *opaque,
72 int level);
d680efe9
MF		 73static void ocfs2_dentry_ast_func(void *opaque);
74static void ocfs2_dentry_bast_func(void *opaque,
75 int level);
ccd979bd
MF		 76static void ocfs2_super_ast_func(void *opaque);
77static void ocfs2_super_bast_func(void *opaque,
78 int level);
79static void ocfs2_rename_ast_func(void *opaque);
80static void ocfs2_rename_bast_func(void *opaque,
81 int level);
82
d680efe9
MF		 83/*
84 * Return value from ocfs2_convert_worker_t functions.
85 *
86 * These control the precise actions of ocfs2_generic_unblock_lock()
87 * and ocfs2_process_blocked_lock()
88 *
89 */
90enum ocfs2_unblock_action {
91 UNBLOCK_CONTINUE = 0, /* Continue downconvert */
92 UNBLOCK_CONTINUE_POST = 1, /* Continue downconvert, fire
93 * ->post_unlock callback */
94 UNBLOCK_STOP_POST = 2, /* Do not downconvert, fire
95 * ->post_unlock() callback. */
96};
97
98struct ocfs2_unblock_ctl {
99 int requeue;
100 enum ocfs2_unblock_action unblock_action;
101};
102
ccd979bd
MF		 103/* so far, all locks have gotten along with the same unlock ast */
104static void ocfs2_unlock_ast_func(void *opaque,
105 enum dlm_status status);
ccd979bd 106static int ocfs2_unblock_meta(struct ocfs2_lock_res *lockres,
d680efe9 107 struct ocfs2_unblock_ctl *ctl);
ccd979bd 108static int ocfs2_unblock_data(struct ocfs2_lock_res *lockres,
d680efe9 109 struct ocfs2_unblock_ctl *ctl);
ccd979bd 110static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res *lockres,
d680efe9
MF		 111 struct ocfs2_unblock_ctl *ctl);
112static int ocfs2_unblock_dentry_lock(struct ocfs2_lock_res *lockres,
113 struct ocfs2_unblock_ctl *ctl);
ccd979bd 114static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res *lockres,
d680efe9
MF		 115 struct ocfs2_unblock_ctl *ctl);
116
117static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
118 struct ocfs2_lock_res *lockres);
ccd979bd
MF		 119
120struct ocfs2_lock_res_ops {
121 void (*ast)(void *);
122 void (*bast)(void *, int);
123 void (*unlock_ast)(void *, enum dlm_status);
d680efe9
MF		 124 int (*unblock)(struct ocfs2_lock_res *, struct ocfs2_unblock_ctl *);
125 void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
ccd979bd
MF		 126};
127
d680efe9
MF		 128typedef int (ocfs2_convert_worker_t)(struct ocfs2_lock_res *, int);
129static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb,
130 struct ocfs2_lock_res *lockres,
131 struct ocfs2_unblock_ctl *ctl,
132 ocfs2_convert_worker_t *worker);
133
ccd979bd
MF		 134static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
135 .ast = ocfs2_inode_ast_func,
136 .bast = ocfs2_inode_bast_func,
137 .unlock_ast = ocfs2_unlock_ast_func,
138 .unblock = ocfs2_unblock_inode_lock,
139};
140
141static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = {
142 .ast = ocfs2_inode_ast_func,
143 .bast = ocfs2_inode_bast_func,
144 .unlock_ast = ocfs2_unlock_ast_func,
145 .unblock = ocfs2_unblock_meta,
146};
147
ccd979bd
MF		 148static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = {
149 .ast = ocfs2_inode_ast_func,
150 .bast = ocfs2_inode_bast_func,
151 .unlock_ast = ocfs2_unlock_ast_func,
152 .unblock = ocfs2_unblock_data,
153};
154
155static struct ocfs2_lock_res_ops ocfs2_super_lops = {
156 .ast = ocfs2_super_ast_func,
157 .bast = ocfs2_super_bast_func,
158 .unlock_ast = ocfs2_unlock_ast_func,
159 .unblock = ocfs2_unblock_osb_lock,
160};
161
162static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
163 .ast = ocfs2_rename_ast_func,
164 .bast = ocfs2_rename_bast_func,
165 .unlock_ast = ocfs2_unlock_ast_func,
166 .unblock = ocfs2_unblock_osb_lock,
167};
168
d680efe9
MF		 169static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
170 .ast = ocfs2_dentry_ast_func,
171 .bast = ocfs2_dentry_bast_func,
172 .unlock_ast = ocfs2_unlock_ast_func,
173 .unblock = ocfs2_unblock_dentry_lock,
174 .post_unlock = ocfs2_dentry_post_unlock,
175};
176
ccd979bd
MF		 177static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
178{
179 return lockres->l_type == OCFS2_LOCK_TYPE_META ||
180 lockres->l_type == OCFS2_LOCK_TYPE_DATA ||
181 lockres->l_type == OCFS2_LOCK_TYPE_RW;
182}
183
184static inline int ocfs2_is_super_lock(struct ocfs2_lock_res *lockres)
185{
186 return lockres->l_type == OCFS2_LOCK_TYPE_SUPER;
187}
188
189static inline int ocfs2_is_rename_lock(struct ocfs2_lock_res *lockres)
190{
191 return lockres->l_type == OCFS2_LOCK_TYPE_RENAME;
192}
193
194static inline struct ocfs2_super *ocfs2_lock_res_super(struct ocfs2_lock_res *lockres)
195{
196 BUG_ON(!ocfs2_is_super_lock(lockres)
197 && !ocfs2_is_rename_lock(lockres));
198
199 return (struct ocfs2_super *) lockres->l_priv;
200}
201
202static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
203{
204 BUG_ON(!ocfs2_is_inode_lock(lockres));
205
206 return (struct inode *) lockres->l_priv;
207}
208
d680efe9
MF		 209static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
210{
211 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
212
213 return (struct ocfs2_dentry_lock *)lockres->l_priv;
214}
215
ccd979bd
MF		 216static int ocfs2_lock_create(struct ocfs2_super *osb,
217 struct ocfs2_lock_res *lockres,
218 int level,
219 int dlm_flags);
220static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
221 int wanted);
222static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
223 struct ocfs2_lock_res *lockres,
224 int level);
225static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
226static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
227static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
228static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
229static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
230 struct ocfs2_lock_res *lockres);
231static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
232 int convert);
233#define ocfs2_log_dlm_error(_func, _stat, _lockres) do { \
234 mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \
235 "resource %s: %s\n", dlm_errname(_stat), _func, \
236 _lockres->l_name, dlm_errmsg(_stat)); \
237} while (0)
238static void ocfs2_vote_on_unlock(struct ocfs2_super *osb,
239 struct ocfs2_lock_res *lockres);
240static int ocfs2_meta_lock_update(struct inode *inode,
241 struct buffer_head **bh);
242static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
243static inline int ocfs2_highest_compat_lock_level(int level);
244static inline int ocfs2_can_downconvert_meta_lock(struct inode *inode,
245 struct ocfs2_lock_res *lockres,
246 int new_level);
247
ccd979bd
MF		 248static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
249 u64 blkno,
250 u32 generation,
251 char *name)
252{
253 int len;
254
255 mlog_entry_void();
256
257 BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
258
b0697053
MF		 259 len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
260 ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
261 (long long)blkno, generation);
ccd979bd
MF		 262
263 BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
264
265 mlog(0, "built lock resource with name: %s\n", name);
266
267 mlog_exit_void();
268}
269
34af946a 270static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
ccd979bd
MF		 271
272static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
273 struct ocfs2_dlm_debug *dlm_debug)
274{
275 mlog(0, "Add tracking for lockres %s\n", res->l_name);
276
277 spin_lock(&ocfs2_dlm_tracking_lock);
278 list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
279 spin_unlock(&ocfs2_dlm_tracking_lock);
280}
281
282static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
283{
284 spin_lock(&ocfs2_dlm_tracking_lock);
285 if (!list_empty(&res->l_debug_list))
286 list_del_init(&res->l_debug_list);
287 spin_unlock(&ocfs2_dlm_tracking_lock);
288}
289
290static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
291 struct ocfs2_lock_res *res,
292 enum ocfs2_lock_type type,
ccd979bd
MF		 293 struct ocfs2_lock_res_ops *ops,
294 void *priv)
295{
ccd979bd
MF		 296 res->l_type = type;
297 res->l_ops = ops;
298 res->l_priv = priv;
299
300 res->l_level = LKM_IVMODE;
301 res->l_requested = LKM_IVMODE;
302 res->l_blocking = LKM_IVMODE;
303 res->l_action = OCFS2_AST_INVALID;
304 res->l_unlock_action = OCFS2_UNLOCK_INVALID;
305
306 res->l_flags = OCFS2_LOCK_INITIALIZED;
307
308 ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
309}
310
311void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
312{
313 /* This also clears out the lock status block */
314 memset(res, 0, sizeof(struct ocfs2_lock_res));
315 spin_lock_init(&res->l_lock);
316 init_waitqueue_head(&res->l_event);
317 INIT_LIST_HEAD(&res->l_blocked_list);
318 INIT_LIST_HEAD(&res->l_mask_waiters);
319}
320
321void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
322 enum ocfs2_lock_type type,
323 struct inode *inode)
324{
325 struct ocfs2_lock_res_ops *ops;
326
327 switch(type) {
328 case OCFS2_LOCK_TYPE_RW:
329 ops = &ocfs2_inode_rw_lops;
330 break;
331 case OCFS2_LOCK_TYPE_META:
332 ops = &ocfs2_inode_meta_lops;
333 break;
334 case OCFS2_LOCK_TYPE_DATA:
335 ops = &ocfs2_inode_data_lops;
336 break;
337 default:
338 mlog_bug_on_msg(1, "type: %d\n", type);
339 ops = NULL; /* thanks, gcc */
340 break;
341 };
342
d680efe9
MF		 343 ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
344 inode->i_generation, res->l_name);
345 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
346}
347
348static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
349{
350 __be64 inode_blkno_be;
351
352 memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
353 sizeof(__be64));
354
355 return be64_to_cpu(inode_blkno_be);
356}
357
358void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
359 u64 parent, struct inode *inode)
360{
361 int len;
362 u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
363 __be64 inode_blkno_be = cpu_to_be64(inode_blkno);
364 struct ocfs2_lock_res *lockres = &dl->dl_lockres;
365
366 ocfs2_lock_res_init_once(lockres);
367
368 /*
369 * Unfortunately, the standard lock naming scheme won't work
370 * here because we have two 16 byte values to use. Instead,
371 * we'll stuff the inode number as a binary value. We still
372 * want error prints to show something without garbling the
373 * display, so drop a null byte in there before the inode
374 * number. A future version of OCFS2 will likely use all
375 * binary lock names. The stringified names have been a
376 * tremendous aid in debugging, but now that the debugfs
377 * interface exists, we can mangle things there if need be.
378 *
379 * NOTE: We also drop the standard "pad" value (the total lock
380 * name size stays the same though - the last part is all
381 * zeros due to the memset in ocfs2_lock_res_init_once()
382 */
383 len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
384 "%c%016llx",
385 ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
386 (long long)parent);
387
388 BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
389
390 memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
391 sizeof(__be64));
392
393 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
394 OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
395 dl);
ccd979bd
MF		 396}
397
398static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
399 struct ocfs2_super *osb)
400{
401 /* Superblock lockres doesn't come from a slab so we call init
402 * once on it manually. */
403 ocfs2_lock_res_init_once(res);
d680efe9
MF		 404 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
405 0, res->l_name);
ccd979bd 406 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
ccd979bd
MF		 407 &ocfs2_super_lops, osb);
408}
409
410static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
411 struct ocfs2_super *osb)
412{
413 /* Rename lockres doesn't come from a slab so we call init
414 * once on it manually. */
415 ocfs2_lock_res_init_once(res);
d680efe9
MF		 416 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
417 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
ccd979bd
MF		 418 &ocfs2_rename_lops, osb);
419}
420
421void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
422{
423 mlog_entry_void();
424
425 if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
426 return;
427
428 ocfs2_remove_lockres_tracking(res);
429
430 mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
431 "Lockres %s is on the blocked list\n",
432 res->l_name);
433 mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
434 "Lockres %s has mask waiters pending\n",
435 res->l_name);
436 mlog_bug_on_msg(spin_is_locked(&res->l_lock),
437 "Lockres %s is locked\n",
438 res->l_name);
439 mlog_bug_on_msg(res->l_ro_holders,
440 "Lockres %s has %u ro holders\n",
441 res->l_name, res->l_ro_holders);
442 mlog_bug_on_msg(res->l_ex_holders,
443 "Lockres %s has %u ex holders\n",
444 res->l_name, res->l_ex_holders);
445
446 /* Need to clear out the lock status block for the dlm */
447 memset(&res->l_lksb, 0, sizeof(res->l_lksb));
448
449 res->l_flags = 0UL;
450 mlog_exit_void();
451}
452
453static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
454 int level)
455{
456 mlog_entry_void();
457
458 BUG_ON(!lockres);
459
460 switch(level) {
461 case LKM_EXMODE:
462 lockres->l_ex_holders++;
463 break;
464 case LKM_PRMODE:
465 lockres->l_ro_holders++;
466 break;
467 default:
468 BUG();
469 }
470
471 mlog_exit_void();
472}
473
474static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
475 int level)
476{
477 mlog_entry_void();
478
479 BUG_ON(!lockres);
480
481 switch(level) {
482 case LKM_EXMODE:
483 BUG_ON(!lockres->l_ex_holders);
484 lockres->l_ex_holders--;
485 break;
486 case LKM_PRMODE:
487 BUG_ON(!lockres->l_ro_holders);
488 lockres->l_ro_holders--;
489 break;
490 default:
491 BUG();
492 }
493 mlog_exit_void();
494}
495
496/* WARNING: This function lives in a world where the only three lock
497 * levels are EX, PR, and NL. It *will* have to be adjusted when more
498 * lock types are added. */
499static inline int ocfs2_highest_compat_lock_level(int level)
500{
501 int new_level = LKM_EXMODE;
502
503 if (level == LKM_EXMODE)
504 new_level = LKM_NLMODE;
505 else if (level == LKM_PRMODE)
506 new_level = LKM_PRMODE;
507 return new_level;
508}
509
510static void lockres_set_flags(struct ocfs2_lock_res *lockres,
511 unsigned long newflags)
512{
513 struct list_head *pos, *tmp;
514 struct ocfs2_mask_waiter *mw;
515
516 assert_spin_locked(&lockres->l_lock);
517
518 lockres->l_flags = newflags;
519
520 list_for_each_safe(pos, tmp, &lockres->l_mask_waiters) {
521 mw = list_entry(pos, struct ocfs2_mask_waiter, mw_item);
522 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
523 continue;
524
525 list_del_init(&mw->mw_item);
526 mw->mw_status = 0;
527 complete(&mw->mw_complete);
528 }
529}
530static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
531{
532 lockres_set_flags(lockres, lockres->l_flags | or);
533}
534static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
535 unsigned long clear)
536{
537 lockres_set_flags(lockres, lockres->l_flags & ~clear);
538}
539
540static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
541{
542 mlog_entry_void();
543
544 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
545 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
546 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
547 BUG_ON(lockres->l_blocking <= LKM_NLMODE);
548
549 lockres->l_level = lockres->l_requested;
550 if (lockres->l_level <=
551 ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
552 lockres->l_blocking = LKM_NLMODE;
553 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
554 }
555 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
556
557 mlog_exit_void();
558}
559
560static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
561{
562 mlog_entry_void();
563
564 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
565 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
566
567 /* Convert from RO to EX doesn't really need anything as our
568 * information is already up to data. Convert from NL to
569 * *anything* however should mark ourselves as needing an
570 * update */
571 if (lockres->l_level == LKM_NLMODE)
572 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
573
574 lockres->l_level = lockres->l_requested;
575 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
576
577 mlog_exit_void();
578}
579
580static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
581{
582 mlog_entry_void();
583
584 BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY));
585 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
586
587 if (lockres->l_requested > LKM_NLMODE &&
588 !(lockres->l_flags & OCFS2_LOCK_LOCAL))
589 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
590
591 lockres->l_level = lockres->l_requested;
592 lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
593 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
594
595 mlog_exit_void();
596}
597
598static void ocfs2_inode_ast_func(void *opaque)
599{
600 struct ocfs2_lock_res *lockres = opaque;
601 struct inode *inode;
602 struct dlm_lockstatus *lksb;
603 unsigned long flags;
604
605 mlog_entry_void();
606
607 inode = ocfs2_lock_res_inode(lockres);
608
b0697053
MF		 609 mlog(0, "AST fired for inode %llu, l_action = %u, type = %s\n",
610 (unsigned long long)OCFS2_I(inode)->ip_blkno, lockres->l_action,
ccd979bd
MF		 611 ocfs2_lock_type_string(lockres->l_type));
612
613 BUG_ON(!ocfs2_is_inode_lock(lockres));
614
615 spin_lock_irqsave(&lockres->l_lock, flags);
616
617 lksb = &(lockres->l_lksb);
618 if (lksb->status != DLM_NORMAL) {
619 mlog(ML_ERROR, "ocfs2_inode_ast_func: lksb status value of %u "
b0697053
MF		 620 "on inode %llu\n", lksb->status,
621 (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 622 spin_unlock_irqrestore(&lockres->l_lock, flags);
623 mlog_exit_void();
624 return;
625 }
626
627 switch(lockres->l_action) {
628 case OCFS2_AST_ATTACH:
629 ocfs2_generic_handle_attach_action(lockres);
630 lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
631 break;
632 case OCFS2_AST_CONVERT:
633 ocfs2_generic_handle_convert_action(lockres);
634 break;
635 case OCFS2_AST_DOWNCONVERT:
636 ocfs2_generic_handle_downconvert_action(lockres);
637 break;
638 default:
639 mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
640 "lockres flags = 0x%lx, unlock action: %u\n",
641 lockres->l_name, lockres->l_action, lockres->l_flags,
642 lockres->l_unlock_action);
643
644 BUG();
645 }
646
647 /* data and rw locking ignores refresh flag for now. */
648 if (lockres->l_type != OCFS2_LOCK_TYPE_META)
649 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
650
651 /* set it to something invalid so if we get called again we
652 * can catch it. */
653 lockres->l_action = OCFS2_AST_INVALID;
654 spin_unlock_irqrestore(&lockres->l_lock, flags);
655 wake_up(&lockres->l_event);
656
657 mlog_exit_void();
658}
659
660static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
661 int level)
662{
663 int needs_downconvert = 0;
664 mlog_entry_void();
665
666 assert_spin_locked(&lockres->l_lock);
667
668 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
669
670 if (level > lockres->l_blocking) {
671 /* only schedule a downconvert if we haven't already scheduled
672 * one that goes low enough to satisfy the level we're
673 * blocking. this also catches the case where we get
674 * duplicate BASTs */
675 if (ocfs2_highest_compat_lock_level(level) <
676 ocfs2_highest_compat_lock_level(lockres->l_blocking))
677 needs_downconvert = 1;
678
679 lockres->l_blocking = level;
680 }
681
682 mlog_exit(needs_downconvert);
683 return needs_downconvert;
684}
685
686static void ocfs2_generic_bast_func(struct ocfs2_super *osb,
687 struct ocfs2_lock_res *lockres,
688 int level)
689{
690 int needs_downconvert;
691 unsigned long flags;
692
693 mlog_entry_void();
694
695 BUG_ON(level <= LKM_NLMODE);
696
697 spin_lock_irqsave(&lockres->l_lock, flags);
698 needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
699 if (needs_downconvert)
700 ocfs2_schedule_blocked_lock(osb, lockres);
701 spin_unlock_irqrestore(&lockres->l_lock, flags);
702
d680efe9
MF		 703 wake_up(&lockres->l_event);
704
ccd979bd
MF		 705 ocfs2_kick_vote_thread(osb);
706
ccd979bd
MF		 707 mlog_exit_void();
708}
709
710static void ocfs2_inode_bast_func(void *opaque, int level)
711{
712 struct ocfs2_lock_res *lockres = opaque;
713 struct inode *inode;
714 struct ocfs2_super *osb;
715
716 mlog_entry_void();
717
718 BUG_ON(!ocfs2_is_inode_lock(lockres));
719
720 inode = ocfs2_lock_res_inode(lockres);
721 osb = OCFS2_SB(inode->i_sb);
722
b0697053
MF		 723 mlog(0, "BAST fired for inode %llu, blocking %d, level %d type %s\n",
724 (unsigned long long)OCFS2_I(inode)->ip_blkno, level,
725 lockres->l_level, ocfs2_lock_type_string(lockres->l_type));
ccd979bd
MF		 726
727 ocfs2_generic_bast_func(osb, lockres, level);
728
729 mlog_exit_void();
730}
731
732static void ocfs2_generic_ast_func(struct ocfs2_lock_res *lockres,
733 int ignore_refresh)
734{
735 struct dlm_lockstatus *lksb = &lockres->l_lksb;
736 unsigned long flags;
737
738 spin_lock_irqsave(&lockres->l_lock, flags);
739
740 if (lksb->status != DLM_NORMAL) {
741 mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n",
742 lockres->l_name, lksb->status);
743 spin_unlock_irqrestore(&lockres->l_lock, flags);
744 return;
745 }
746
747 switch(lockres->l_action) {
748 case OCFS2_AST_ATTACH:
749 ocfs2_generic_handle_attach_action(lockres);
750 break;
751 case OCFS2_AST_CONVERT:
752 ocfs2_generic_handle_convert_action(lockres);
753 break;
754 case OCFS2_AST_DOWNCONVERT:
755 ocfs2_generic_handle_downconvert_action(lockres);
756 break;
757 default:
758 BUG();
759 }
760
761 if (ignore_refresh)
762 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
763
764 /* set it to something invalid so if we get called again we
765 * can catch it. */
766 lockres->l_action = OCFS2_AST_INVALID;
ccd979bd
MF		 767
768 wake_up(&lockres->l_event);
d680efe9 769 spin_unlock_irqrestore(&lockres->l_lock, flags);
ccd979bd
MF		 770}
771
772static void ocfs2_super_ast_func(void *opaque)
773{
774 struct ocfs2_lock_res *lockres = opaque;
775
776 mlog_entry_void();
777 mlog(0, "Superblock AST fired\n");
778
779 BUG_ON(!ocfs2_is_super_lock(lockres));
780 ocfs2_generic_ast_func(lockres, 0);
781
782 mlog_exit_void();
783}
784
785static void ocfs2_super_bast_func(void *opaque,
786 int level)
787{
788 struct ocfs2_lock_res *lockres = opaque;
789 struct ocfs2_super *osb;
790
791 mlog_entry_void();
792 mlog(0, "Superblock BAST fired\n");
793
794 BUG_ON(!ocfs2_is_super_lock(lockres));
795 osb = ocfs2_lock_res_super(lockres);
796 ocfs2_generic_bast_func(osb, lockres, level);
797
798 mlog_exit_void();
799}
800
801static void ocfs2_rename_ast_func(void *opaque)
802{
803 struct ocfs2_lock_res *lockres = opaque;
804
805 mlog_entry_void();
806
807 mlog(0, "Rename AST fired\n");
808
809 BUG_ON(!ocfs2_is_rename_lock(lockres));
810
811 ocfs2_generic_ast_func(lockres, 1);
812
813 mlog_exit_void();
814}
815
816static void ocfs2_rename_bast_func(void *opaque,
817 int level)
818{
819 struct ocfs2_lock_res *lockres = opaque;
820 struct ocfs2_super *osb;
821
822 mlog_entry_void();
823
824 mlog(0, "Rename BAST fired\n");
825
826 BUG_ON(!ocfs2_is_rename_lock(lockres));
827
828 osb = ocfs2_lock_res_super(lockres);
829 ocfs2_generic_bast_func(osb, lockres, level);
830
831 mlog_exit_void();
832}
833
d680efe9
MF		 834static void ocfs2_dentry_ast_func(void *opaque)
835{
836 struct ocfs2_lock_res *lockres = opaque;
837
838 BUG_ON(!lockres);
839
840 ocfs2_generic_ast_func(lockres, 1);
841}
842
843static void ocfs2_dentry_bast_func(void *opaque, int level)
844{
845 struct ocfs2_lock_res *lockres = opaque;
846 struct ocfs2_dentry_lock *dl = lockres->l_priv;
847 struct ocfs2_super *osb = OCFS2_SB(dl->dl_inode->i_sb);
848
849 mlog(0, "Dentry bast: level: %d, name: %s\n", level,
850 lockres->l_name);
851
852 ocfs2_generic_bast_func(osb, lockres, level);
853}
854
ccd979bd
MF		 855static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
856 int convert)
857{
858 unsigned long flags;
859
860 mlog_entry_void();
861 spin_lock_irqsave(&lockres->l_lock, flags);
862 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
863 if (convert)
864 lockres->l_action = OCFS2_AST_INVALID;
865 else
866 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
867 spin_unlock_irqrestore(&lockres->l_lock, flags);
868
869 wake_up(&lockres->l_event);
870 mlog_exit_void();
871}
872
873/* Note: If we detect another process working on the lock (i.e.,
874 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
875 * to do the right thing in that case.
876 */
877static int ocfs2_lock_create(struct ocfs2_super *osb,
878 struct ocfs2_lock_res *lockres,
879 int level,
880 int dlm_flags)
881{
882 int ret = 0;
883 enum dlm_status status;
884 unsigned long flags;
885
886 mlog_entry_void();
887
888 mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level,
889 dlm_flags);
890
891 spin_lock_irqsave(&lockres->l_lock, flags);
892 if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
893 (lockres->l_flags & OCFS2_LOCK_BUSY)) {
894 spin_unlock_irqrestore(&lockres->l_lock, flags);
895 goto bail;
896 }
897
898 lockres->l_action = OCFS2_AST_ATTACH;
899 lockres->l_requested = level;
900 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
901 spin_unlock_irqrestore(&lockres->l_lock, flags);
902
903 status = dlmlock(osb->dlm,
904 level,
905 &lockres->l_lksb,
906 dlm_flags,
907 lockres->l_name,
f0681062 908 OCFS2_LOCK_ID_MAX_LEN - 1,
ccd979bd
MF		 909 lockres->l_ops->ast,
910 lockres,
911 lockres->l_ops->bast);
912 if (status != DLM_NORMAL) {
913 ocfs2_log_dlm_error("dlmlock", status, lockres);
914 ret = -EINVAL;
915 ocfs2_recover_from_dlm_error(lockres, 1);
916 }
917
918 mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name);
919
920bail:
921 mlog_exit(ret);
922 return ret;
923}
924
925static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
926 int flag)
927{
928 unsigned long flags;
929 int ret;
930
931 spin_lock_irqsave(&lockres->l_lock, flags);
932 ret = lockres->l_flags & flag;
933 spin_unlock_irqrestore(&lockres->l_lock, flags);
934
935 return ret;
936}
937
938static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
939
940{
941 wait_event(lockres->l_event,
942 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
943}
944
945static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
946
947{
948 wait_event(lockres->l_event,
949 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
950}
951
952/* predict what lock level we'll be dropping down to on behalf
953 * of another node, and return true if the currently wanted
954 * level will be compatible with it. */
955static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
956 int wanted)
957{
958 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
959
960 return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
961}
962
963static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
964{
965 INIT_LIST_HEAD(&mw->mw_item);
966 init_completion(&mw->mw_complete);
967}
968
969static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
970{
971 wait_for_completion(&mw->mw_complete);
972 /* Re-arm the completion in case we want to wait on it again */
973 INIT_COMPLETION(mw->mw_complete);
974 return mw->mw_status;
975}
976
977static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
978 struct ocfs2_mask_waiter *mw,
979 unsigned long mask,
980 unsigned long goal)
981{
982 BUG_ON(!list_empty(&mw->mw_item));
983
984 assert_spin_locked(&lockres->l_lock);
985
986 list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
987 mw->mw_mask = mask;
988 mw->mw_goal = goal;
989}
990
991/* returns 0 if the mw that was removed was already satisfied, -EBUSY
992 * if the mask still hadn't reached its goal */
993static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
994 struct ocfs2_mask_waiter *mw)
995{
996 unsigned long flags;
997 int ret = 0;
998
999 spin_lock_irqsave(&lockres->l_lock, flags);
1000 if (!list_empty(&mw->mw_item)) {
1001 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1002 ret = -EBUSY;
1003
1004 list_del_init(&mw->mw_item);
1005 init_completion(&mw->mw_complete);
1006 }
1007 spin_unlock_irqrestore(&lockres->l_lock, flags);
1008
1009 return ret;
1010
1011}
1012
1013static int ocfs2_cluster_lock(struct ocfs2_super *osb,
1014 struct ocfs2_lock_res *lockres,
1015 int level,
1016 int lkm_flags,
1017 int arg_flags)
1018{
1019 struct ocfs2_mask_waiter mw;
1020 enum dlm_status status;
1021 int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1022 int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1023 unsigned long flags;
1024
1025 mlog_entry_void();
1026
1027 ocfs2_init_mask_waiter(&mw);
1028
1029again:
1030 wait = 0;
1031
1032 if (catch_signals && signal_pending(current)) {
1033 ret = -ERESTARTSYS;
1034 goto out;
1035 }
1036
1037 spin_lock_irqsave(&lockres->l_lock, flags);
1038
1039 mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1040 "Cluster lock called on freeing lockres %s! flags "
1041 "0x%lx\n", lockres->l_name, lockres->l_flags);
1042
1043 /* We only compare against the currently granted level
1044 * here. If the lock is blocked waiting on a downconvert,
1045 * we'll get caught below. */
1046 if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1047 level > lockres->l_level) {
1048 /* is someone sitting in dlm_lock? If so, wait on
1049 * them. */
1050 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1051 wait = 1;
1052 goto unlock;
1053 }
1054
1055 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1056 /* lock has not been created yet. */
1057 spin_unlock_irqrestore(&lockres->l_lock, flags);
1058
1059 ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0);
1060 if (ret < 0) {
1061 mlog_errno(ret);
1062 goto out;
1063 }
1064 goto again;
1065 }
1066
1067 if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1068 !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1069 /* is the lock is currently blocked on behalf of
1070 * another node */
1071 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1072 wait = 1;
1073 goto unlock;
1074 }
1075
1076 if (level > lockres->l_level) {
1077 if (lockres->l_action != OCFS2_AST_INVALID)
1078 mlog(ML_ERROR, "lockres %s has action %u pending\n",
1079 lockres->l_name, lockres->l_action);
1080
1081 lockres->l_action = OCFS2_AST_CONVERT;
1082 lockres->l_requested = level;
1083 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1084 spin_unlock_irqrestore(&lockres->l_lock, flags);
1085
1086 BUG_ON(level == LKM_IVMODE);
1087 BUG_ON(level == LKM_NLMODE);
1088
1089 mlog(0, "lock %s, convert from %d to level = %d\n",
1090 lockres->l_name, lockres->l_level, level);
1091
1092 /* call dlm_lock to upgrade lock now */
1093 status = dlmlock(osb->dlm,
1094 level,
1095 &lockres->l_lksb,
1096 lkm_flags|LKM_CONVERT|LKM_VALBLK,
1097 lockres->l_name,
f0681062 1098 OCFS2_LOCK_ID_MAX_LEN - 1,
ccd979bd
MF		 1099 lockres->l_ops->ast,
1100 lockres,
1101 lockres->l_ops->bast);
1102 if (status != DLM_NORMAL) {
1103 if ((lkm_flags & LKM_NOQUEUE) &&
1104 (status == DLM_NOTQUEUED))
1105 ret = -EAGAIN;
1106 else {
1107 ocfs2_log_dlm_error("dlmlock", status,
1108 lockres);
1109 ret = -EINVAL;
1110 }
1111 ocfs2_recover_from_dlm_error(lockres, 1);
1112 goto out;
1113 }
1114
1115 mlog(0, "lock %s, successfull return from dlmlock\n",
1116 lockres->l_name);
1117
1118 /* At this point we've gone inside the dlm and need to
1119 * complete our work regardless. */
1120 catch_signals = 0;
1121
1122 /* wait for busy to clear and carry on */
1123 goto again;
1124 }
1125
1126 /* Ok, if we get here then we're good to go. */
1127 ocfs2_inc_holders(lockres, level);
1128
1129 ret = 0;
1130unlock:
1131 spin_unlock_irqrestore(&lockres->l_lock, flags);
1132out:
1133 /*
1134 * This is helping work around a lock inversion between the page lock
1135 * and dlm locks. One path holds the page lock while calling aops
1136 * which block acquiring dlm locks. The voting thread holds dlm
1137 * locks while acquiring page locks while down converting data locks.
1138 * This block is helping an aop path notice the inversion and back
1139 * off to unlock its page lock before trying the dlm lock again.
1140 */
1141 if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1142 mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1143 wait = 0;
1144 if (lockres_remove_mask_waiter(lockres, &mw))
1145 ret = -EAGAIN;
1146 else
1147 goto again;
1148 }
1149 if (wait) {
1150 ret = ocfs2_wait_for_mask(&mw);
1151 if (ret == 0)
1152 goto again;
1153 mlog_errno(ret);
1154 }
1155
1156 mlog_exit(ret);
1157 return ret;
1158}
1159
1160static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
1161 struct ocfs2_lock_res *lockres,
1162 int level)
1163{
1164 unsigned long flags;
1165
1166 mlog_entry_void();
1167 spin_lock_irqsave(&lockres->l_lock, flags);
1168 ocfs2_dec_holders(lockres, level);
1169 ocfs2_vote_on_unlock(osb, lockres);
1170 spin_unlock_irqrestore(&lockres->l_lock, flags);
1171 mlog_exit_void();
1172}
1173
d680efe9
MF		 1174int ocfs2_create_new_lock(struct ocfs2_super *osb,
1175 struct ocfs2_lock_res *lockres,
1176 int ex)
ccd979bd 1177{
d680efe9 1178 int level = ex ? LKM_EXMODE : LKM_PRMODE;
ccd979bd
MF		 1179 unsigned long flags;
1180
1181 spin_lock_irqsave(&lockres->l_lock, flags);
1182 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1183 lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1184 spin_unlock_irqrestore(&lockres->l_lock, flags);
1185
d680efe9 1186 return ocfs2_lock_create(osb, lockres, level, LKM_LOCAL);
ccd979bd
MF		 1187}
1188
1189/* Grants us an EX lock on the data and metadata resources, skipping
1190 * the normal cluster directory lookup. Use this ONLY on newly created
1191 * inodes which other nodes can't possibly see, and which haven't been
1192 * hashed in the inode hash yet. This can give us a good performance
1193 * increase as it'll skip the network broadcast normally associated
1194 * with creating a new lock resource. */
1195int ocfs2_create_new_inode_locks(struct inode *inode)
1196{
1197 int ret;
d680efe9 1198 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1199
1200 BUG_ON(!inode);
1201 BUG_ON(!ocfs2_inode_is_new(inode));
1202
1203 mlog_entry_void();
1204
b0697053 1205 mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 1206
1207 /* NOTE: That we don't increment any of the holder counts, nor
1208 * do we add anything to a journal handle. Since this is
1209 * supposed to be a new inode which the cluster doesn't know
1210 * about yet, there is no need to. As far as the LVB handling
1211 * is concerned, this is basically like acquiring an EX lock
1212 * on a resource which has an invalid one -- we'll set it
1213 * valid when we release the EX. */
1214
d680efe9 1215 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1);
ccd979bd
MF		 1216 if (ret) {
1217 mlog_errno(ret);
1218 goto bail;
1219 }
1220
d680efe9 1221 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_meta_lockres, 1);
ccd979bd
MF		 1222 if (ret) {
1223 mlog_errno(ret);
1224 goto bail;
1225 }
1226
d680efe9 1227 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_data_lockres, 1);
ccd979bd
MF		 1228 if (ret) {
1229 mlog_errno(ret);
1230 goto bail;
1231 }
1232
1233bail:
1234 mlog_exit(ret);
1235 return ret;
1236}
1237
1238int ocfs2_rw_lock(struct inode *inode, int write)
1239{
1240 int status, level;
1241 struct ocfs2_lock_res *lockres;
1242
1243 BUG_ON(!inode);
1244
1245 mlog_entry_void();
1246
b0697053
MF		 1247 mlog(0, "inode %llu take %s RW lock\n",
1248 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1249 write ? "EXMODE" : "PRMODE");
1250
1251 lockres = &OCFS2_I(inode)->ip_rw_lockres;
1252
1253 level = write ? LKM_EXMODE : LKM_PRMODE;
1254
1255 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1256 0);
1257 if (status < 0)
1258 mlog_errno(status);
1259
1260 mlog_exit(status);
1261 return status;
1262}
1263
1264void ocfs2_rw_unlock(struct inode *inode, int write)
1265{
1266 int level = write ? LKM_EXMODE : LKM_PRMODE;
1267 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1268
1269 mlog_entry_void();
1270
b0697053
MF		 1271 mlog(0, "inode %llu drop %s RW lock\n",
1272 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1273 write ? "EXMODE" : "PRMODE");
1274
1275 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1276
1277 mlog_exit_void();
1278}
1279
1280int ocfs2_data_lock_full(struct inode *inode,
1281 int write,
1282 int arg_flags)
1283{
1284 int status = 0, level;
1285 struct ocfs2_lock_res *lockres;
1286
1287 BUG_ON(!inode);
1288
1289 mlog_entry_void();
1290
b0697053
MF		 1291 mlog(0, "inode %llu take %s DATA lock\n",
1292 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1293 write ? "EXMODE" : "PRMODE");
1294
1295 /* We'll allow faking a readonly data lock for
1296 * rodevices. */
1297 if (ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) {
1298 if (write) {
1299 status = -EROFS;
1300 mlog_errno(status);
1301 }
1302 goto out;
1303 }
1304
1305 lockres = &OCFS2_I(inode)->ip_data_lockres;
1306
1307 level = write ? LKM_EXMODE : LKM_PRMODE;
1308
1309 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level,
1310 0, arg_flags);
1311 if (status < 0 && status != -EAGAIN)
1312 mlog_errno(status);
1313
1314out:
1315 mlog_exit(status);
1316 return status;
1317}
1318
1319/* see ocfs2_meta_lock_with_page() */
1320int ocfs2_data_lock_with_page(struct inode *inode,
1321 int write,
1322 struct page *page)
1323{
1324 int ret;
1325
1326 ret = ocfs2_data_lock_full(inode, write, OCFS2_LOCK_NONBLOCK);
1327 if (ret == -EAGAIN) {
1328 unlock_page(page);
1329 if (ocfs2_data_lock(inode, write) == 0)
1330 ocfs2_data_unlock(inode, write);
1331 ret = AOP_TRUNCATED_PAGE;
1332 }
1333
1334 return ret;
1335}
1336
1337static void ocfs2_vote_on_unlock(struct ocfs2_super *osb,
1338 struct ocfs2_lock_res *lockres)
1339{
1340 int kick = 0;
1341
1342 mlog_entry_void();
1343
1344 /* If we know that another node is waiting on our lock, kick
1345 * the vote thread * pre-emptively when we reach a release
1346 * condition. */
1347 if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1348 switch(lockres->l_blocking) {
1349 case LKM_EXMODE:
1350 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1351 kick = 1;
1352 break;
1353 case LKM_PRMODE:
1354 if (!lockres->l_ex_holders)
1355 kick = 1;
1356 break;
1357 default:
1358 BUG();
1359 }
1360 }
1361
1362 if (kick)
1363 ocfs2_kick_vote_thread(osb);
1364
1365 mlog_exit_void();
1366}
1367
1368void ocfs2_data_unlock(struct inode *inode,
1369 int write)
1370{
1371 int level = write ? LKM_EXMODE : LKM_PRMODE;
1372 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_data_lockres;
1373
1374 mlog_entry_void();
1375
b0697053
MF		 1376 mlog(0, "inode %llu drop %s DATA lock\n",
1377 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1378 write ? "EXMODE" : "PRMODE");
1379
1380 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)))
1381 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1382
1383 mlog_exit_void();
1384}
1385
1386#define OCFS2_SEC_BITS 34
1387#define OCFS2_SEC_SHIFT (64 - 34)
1388#define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1)
1389
1390/* LVB only has room for 64 bits of time here so we pack it for
1391 * now. */
1392static u64 ocfs2_pack_timespec(struct timespec *spec)
1393{
1394 u64 res;
1395 u64 sec = spec->tv_sec;
1396 u32 nsec = spec->tv_nsec;
1397
1398 res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
1399
1400 return res;
1401}
1402
1403/* Call this with the lockres locked. I am reasonably sure we don't
1404 * need ip_lock in this function as anyone who would be changing those
1405 * values is supposed to be blocked in ocfs2_meta_lock right now. */
1406static void __ocfs2_stuff_meta_lvb(struct inode *inode)
1407{
1408 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1409 struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
1410 struct ocfs2_meta_lvb *lvb;
1411
1412 mlog_entry_void();
1413
1414 lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1415
4d3b83f7 1416 lvb->lvb_version = OCFS2_LVB_VERSION;
ccd979bd
MF		 1417 lvb->lvb_isize = cpu_to_be64(i_size_read(inode));
1418 lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
1419 lvb->lvb_iuid = cpu_to_be32(inode->i_uid);
1420 lvb->lvb_igid = cpu_to_be32(inode->i_gid);
1421 lvb->lvb_imode = cpu_to_be16(inode->i_mode);
1422 lvb->lvb_inlink = cpu_to_be16(inode->i_nlink);
1423 lvb->lvb_iatime_packed =
1424 cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
1425 lvb->lvb_ictime_packed =
1426 cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
1427 lvb->lvb_imtime_packed =
1428 cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
ca4d147e 1429 lvb->lvb_iattr = cpu_to_be32(oi->ip_attr);
f9e2d82e 1430 lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
ccd979bd
MF		 1431
1432 mlog_meta_lvb(0, lockres);
1433
1434 mlog_exit_void();
1435}
1436
1437static void ocfs2_unpack_timespec(struct timespec *spec,
1438 u64 packed_time)
1439{
1440 spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
1441 spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
1442}
1443
1444static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
1445{
1446 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1447 struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
1448 struct ocfs2_meta_lvb *lvb;
1449
1450 mlog_entry_void();
1451
1452 mlog_meta_lvb(0, lockres);
1453
1454 lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1455
1456 /* We're safe here without the lockres lock... */
1457 spin_lock(&oi->ip_lock);
1458 oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
1459 i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
1460
ca4d147e
HP		 1461 oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
1462 ocfs2_set_inode_flags(inode);
1463
ccd979bd
MF		 1464 /* fast-symlinks are a special case */
1465 if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
1466 inode->i_blocks = 0;
1467 else
1468 inode->i_blocks =
1469 ocfs2_align_bytes_to_sectors(i_size_read(inode));
1470
1471 inode->i_uid = be32_to_cpu(lvb->lvb_iuid);
1472 inode->i_gid = be32_to_cpu(lvb->lvb_igid);
1473 inode->i_mode = be16_to_cpu(lvb->lvb_imode);
1474 inode->i_nlink = be16_to_cpu(lvb->lvb_inlink);
1475 ocfs2_unpack_timespec(&inode->i_atime,
1476 be64_to_cpu(lvb->lvb_iatime_packed));
1477 ocfs2_unpack_timespec(&inode->i_mtime,
1478 be64_to_cpu(lvb->lvb_imtime_packed));
1479 ocfs2_unpack_timespec(&inode->i_ctime,
1480 be64_to_cpu(lvb->lvb_ictime_packed));
1481 spin_unlock(&oi->ip_lock);
1482
1483 mlog_exit_void();
1484}
1485
f9e2d82e
MF		 1486static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
1487 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 1488{
1489 struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1490
f9e2d82e
MF		 1491 if (lvb->lvb_version == OCFS2_LVB_VERSION
1492 && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
ccd979bd
MF		 1493 return 1;
1494 return 0;
1495}
1496
1497/* Determine whether a lock resource needs to be refreshed, and
1498 * arbitrate who gets to refresh it.
1499 *
1500 * 0 means no refresh needed.
1501 *
1502 * > 0 means you need to refresh this and you MUST call
1503 * ocfs2_complete_lock_res_refresh afterwards. */
1504static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
1505{
1506 unsigned long flags;
1507 int status = 0;
1508
1509 mlog_entry_void();
1510
1511refresh_check:
1512 spin_lock_irqsave(&lockres->l_lock, flags);
1513 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
1514 spin_unlock_irqrestore(&lockres->l_lock, flags);
1515 goto bail;
1516 }
1517
1518 if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
1519 spin_unlock_irqrestore(&lockres->l_lock, flags);
1520
1521 ocfs2_wait_on_refreshing_lock(lockres);
1522 goto refresh_check;
1523 }
1524
1525 /* Ok, I'll be the one to refresh this lock. */
1526 lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
1527 spin_unlock_irqrestore(&lockres->l_lock, flags);
1528
1529 status = 1;
1530bail:
1531 mlog_exit(status);
1532 return status;
1533}
1534
1535/* If status is non zero, I'll mark it as not being in refresh
1536 * anymroe, but i won't clear the needs refresh flag. */
1537static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
1538 int status)
1539{
1540 unsigned long flags;
1541 mlog_entry_void();
1542
1543 spin_lock_irqsave(&lockres->l_lock, flags);
1544 lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
1545 if (!status)
1546 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
1547 spin_unlock_irqrestore(&lockres->l_lock, flags);
1548
1549 wake_up(&lockres->l_event);
1550
1551 mlog_exit_void();
1552}
1553
1554/* may or may not return a bh if it went to disk. */
1555static int ocfs2_meta_lock_update(struct inode *inode,
1556 struct buffer_head **bh)
1557{
1558 int status = 0;
1559 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1560 struct ocfs2_lock_res *lockres;
1561 struct ocfs2_dinode *fe;
1562
1563 mlog_entry_void();
1564
1565 spin_lock(&oi->ip_lock);
1566 if (oi->ip_flags & OCFS2_INODE_DELETED) {
b0697053 1567 mlog(0, "Orphaned inode %llu was deleted while we "
ccd979bd 1568 "were waiting on a lock. ip_flags = 0x%x\n",
b0697053 1569 (unsigned long long)oi->ip_blkno, oi->ip_flags);
ccd979bd
MF		 1570 spin_unlock(&oi->ip_lock);
1571 status = -ENOENT;
1572 goto bail;
1573 }
1574 spin_unlock(&oi->ip_lock);
1575
1576 lockres = &oi->ip_meta_lockres;
1577
1578 if (!ocfs2_should_refresh_lock_res(lockres))
1579 goto bail;
1580
1581 /* This will discard any caching information we might have had
1582 * for the inode metadata. */
1583 ocfs2_metadata_cache_purge(inode);
1584
1585 /* will do nothing for inode types that don't use the extent
1586 * map (directories, bitmap files, etc) */
1587 ocfs2_extent_map_trunc(inode, 0);
1588
f9e2d82e 1589 if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
b0697053
MF		 1590 mlog(0, "Trusting LVB on inode %llu\n",
1591 (unsigned long long)oi->ip_blkno);
ccd979bd
MF		 1592 ocfs2_refresh_inode_from_lvb(inode);
1593 } else {
1594 /* Boo, we have to go to disk. */
1595 /* read bh, cast, ocfs2_refresh_inode */
1596 status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno,
1597 bh, OCFS2_BH_CACHED, inode);
1598 if (status < 0) {
1599 mlog_errno(status);
1600 goto bail_refresh;
1601 }
1602 fe = (struct ocfs2_dinode *) (*bh)->b_data;
1603
1604 /* This is a good chance to make sure we're not
1605 * locking an invalid object.
1606 *
1607 * We bug on a stale inode here because we checked
1608 * above whether it was wiped from disk. The wiping
1609 * node provides a guarantee that we receive that
1610 * message and can mark the inode before dropping any
1611 * locks associated with it. */
1612 if (!OCFS2_IS_VALID_DINODE(fe)) {
1613 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
1614 status = -EIO;
1615 goto bail_refresh;
1616 }
1617 mlog_bug_on_msg(inode->i_generation !=
1618 le32_to_cpu(fe->i_generation),
b0697053 1619 "Invalid dinode %llu disk generation: %u "
ccd979bd 1620 "inode->i_generation: %u\n",
b0697053
MF		 1621 (unsigned long long)oi->ip_blkno,
1622 le32_to_cpu(fe->i_generation),
ccd979bd
MF		 1623 inode->i_generation);
1624 mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
1625 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
b0697053
MF		 1626 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
1627 (unsigned long long)oi->ip_blkno,
1628 (unsigned long long)le64_to_cpu(fe->i_dtime),
ccd979bd
MF		 1629 le32_to_cpu(fe->i_flags));
1630
1631 ocfs2_refresh_inode(inode, fe);
1632 }
1633
1634 status = 0;
1635bail_refresh:
1636 ocfs2_complete_lock_res_refresh(lockres, status);
1637bail:
1638 mlog_exit(status);
1639 return status;
1640}
1641
1642static int ocfs2_assign_bh(struct inode *inode,
1643 struct buffer_head **ret_bh,
1644 struct buffer_head *passed_bh)
1645{
1646 int status;
1647
1648 if (passed_bh) {
1649 /* Ok, the update went to disk for us, use the
1650 * returned bh. */
1651 *ret_bh = passed_bh;
1652 get_bh(*ret_bh);
1653
1654 return 0;
1655 }
1656
1657 status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1658 OCFS2_I(inode)->ip_blkno,
1659 ret_bh,
1660 OCFS2_BH_CACHED,
1661 inode);
1662 if (status < 0)
1663 mlog_errno(status);
1664
1665 return status;
1666}
1667
1668/*
1669 * returns < 0 error if the callback will never be called, otherwise
1670 * the result of the lock will be communicated via the callback.
1671 */
1672int ocfs2_meta_lock_full(struct inode *inode,
1673 struct ocfs2_journal_handle *handle,
1674 struct buffer_head **ret_bh,
1675 int ex,
1676 int arg_flags)
1677{
1678 int status, level, dlm_flags, acquired;
1679 struct ocfs2_lock_res *lockres;
1680 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1681 struct buffer_head *local_bh = NULL;
1682
1683 BUG_ON(!inode);
1684
1685 mlog_entry_void();
1686
b0697053
MF		 1687 mlog(0, "inode %llu, take %s META lock\n",
1688 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1689 ex ? "EXMODE" : "PRMODE");
1690
1691 status = 0;
1692 acquired = 0;
1693 /* We'll allow faking a readonly metadata lock for
1694 * rodevices. */
1695 if (ocfs2_is_hard_readonly(osb)) {
1696 if (ex)
1697 status = -EROFS;
1698 goto bail;
1699 }
1700
1701 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1702 wait_event(osb->recovery_event,
1703 ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1704
1705 acquired = 0;
1706 lockres = &OCFS2_I(inode)->ip_meta_lockres;
1707 level = ex ? LKM_EXMODE : LKM_PRMODE;
1708 dlm_flags = 0;
1709 if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
1710 dlm_flags |= LKM_NOQUEUE;
1711
1712 status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
1713 if (status < 0) {
1714 if (status != -EAGAIN && status != -EIOCBRETRY)
1715 mlog_errno(status);
1716 goto bail;
1717 }
1718
1719 /* Notify the error cleanup path to drop the cluster lock. */
1720 acquired = 1;
1721
1722 /* We wait twice because a node may have died while we were in
1723 * the lower dlm layers. The second time though, we've
1724 * committed to owning this lock so we don't allow signals to
1725 * abort the operation. */
1726 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1727 wait_event(osb->recovery_event,
1728 ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1729
1730 /* This is fun. The caller may want a bh back, or it may
1731 * not. ocfs2_meta_lock_update definitely wants one in, but
1732 * may or may not read one, depending on what's in the
1733 * LVB. The result of all of this is that we've *only* gone to
1734 * disk if we have to, so the complexity is worthwhile. */
1735 status = ocfs2_meta_lock_update(inode, &local_bh);
1736 if (status < 0) {
1737 if (status != -ENOENT)
1738 mlog_errno(status);
1739 goto bail;
1740 }
1741
1742 if (ret_bh) {
1743 status = ocfs2_assign_bh(inode, ret_bh, local_bh);
1744 if (status < 0) {
1745 mlog_errno(status);
1746 goto bail;
1747 }
1748 }
1749
1750 if (handle) {
1751 status = ocfs2_handle_add_lock(handle, inode);
1752 if (status < 0)
1753 mlog_errno(status);
1754 }
1755
1756bail:
1757 if (status < 0) {
1758 if (ret_bh && (*ret_bh)) {
1759 brelse(*ret_bh);
1760 *ret_bh = NULL;
1761 }
1762 if (acquired)
1763 ocfs2_meta_unlock(inode, ex);
1764 }
1765
1766 if (local_bh)
1767 brelse(local_bh);
1768
1769 mlog_exit(status);
1770 return status;
1771}
1772
1773/*
1774 * This is working around a lock inversion between tasks acquiring DLM locks
1775 * while holding a page lock and the vote thread which blocks dlm lock acquiry
1776 * while acquiring page locks.
1777 *
1778 * ** These _with_page variantes are only intended to be called from aop
1779 * methods that hold page locks and return a very specific *positive* error
1780 * code that aop methods pass up to the VFS -- test for errors with != 0. **
1781 *
1782 * The DLM is called such that it returns -EAGAIN if it would have blocked
1783 * waiting for the vote thread. In that case we unlock our page so the vote
1784 * thread can make progress. Once we've done this we have to return
1785 * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up
1786 * into the VFS who will then immediately retry the aop call.
1787 *
1788 * We do a blocking lock and immediate unlock before returning, though, so that
1789 * the lock has a great chance of being cached on this node by the time the VFS
1790 * calls back to retry the aop. This has a potential to livelock as nodes
1791 * ping locks back and forth, but that's a risk we're willing to take to avoid
1792 * the lock inversion simply.
1793 */
1794int ocfs2_meta_lock_with_page(struct inode *inode,
1795 struct ocfs2_journal_handle *handle,
1796 struct buffer_head **ret_bh,
1797 int ex,
1798 struct page *page)
1799{
1800 int ret;
1801
1802 ret = ocfs2_meta_lock_full(inode, handle, ret_bh, ex,
1803 OCFS2_LOCK_NONBLOCK);
1804 if (ret == -EAGAIN) {
1805 unlock_page(page);
1806 if (ocfs2_meta_lock(inode, handle, ret_bh, ex) == 0)
1807 ocfs2_meta_unlock(inode, ex);
1808 ret = AOP_TRUNCATED_PAGE;
1809 }
1810
1811 return ret;
1812}
1813
1814void ocfs2_meta_unlock(struct inode *inode,
1815 int ex)
1816{
1817 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1818 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres;
1819
1820 mlog_entry_void();
1821
b0697053
MF		 1822 mlog(0, "inode %llu drop %s META lock\n",
1823 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1824 ex ? "EXMODE" : "PRMODE");
1825
1826 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)))
1827 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1828
1829 mlog_exit_void();
1830}
1831
1832int ocfs2_super_lock(struct ocfs2_super *osb,
1833 int ex)
1834{
1835 int status;
1836 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1837 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1838 struct buffer_head *bh;
1839 struct ocfs2_slot_info *si = osb->slot_info;
1840
1841 mlog_entry_void();
1842
1843 if (ocfs2_is_hard_readonly(osb))
1844 return -EROFS;
1845
1846 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1847 if (status < 0) {
1848 mlog_errno(status);
1849 goto bail;
1850 }
1851
1852 /* The super block lock path is really in the best position to
1853 * know when resources covered by the lock need to be
1854 * refreshed, so we do it here. Of course, making sense of
1855 * everything is up to the caller :) */
1856 status = ocfs2_should_refresh_lock_res(lockres);
1857 if (status < 0) {
1858 mlog_errno(status);
1859 goto bail;
1860 }
1861 if (status) {
1862 bh = si->si_bh;
1863 status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0,
1864 si->si_inode);
1865 if (status == 0)
1866 ocfs2_update_slot_info(si);
1867
1868 ocfs2_complete_lock_res_refresh(lockres, status);
1869
1870 if (status < 0)
1871 mlog_errno(status);
1872 }
1873bail:
1874 mlog_exit(status);
1875 return status;
1876}
1877
1878void ocfs2_super_unlock(struct ocfs2_super *osb,
1879 int ex)
1880{
1881 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1882 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1883
1884 ocfs2_cluster_unlock(osb, lockres, level);
1885}
1886
1887int ocfs2_rename_lock(struct ocfs2_super *osb)
1888{
1889 int status;
1890 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1891
1892 if (ocfs2_is_hard_readonly(osb))
1893 return -EROFS;
1894
1895 status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0);
1896 if (status < 0)
1897 mlog_errno(status);
1898
1899 return status;
1900}
1901
1902void ocfs2_rename_unlock(struct ocfs2_super *osb)
1903{
1904 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1905
1906 ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE);
1907}
1908
d680efe9
MF		 1909int ocfs2_dentry_lock(struct dentry *dentry, int ex)
1910{
1911 int ret;
1912 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1913 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
1914 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
1915
1916 BUG_ON(!dl);
1917
1918 if (ocfs2_is_hard_readonly(osb))
1919 return -EROFS;
1920
1921 ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
1922 if (ret < 0)
1923 mlog_errno(ret);
1924
1925 return ret;
1926}
1927
1928void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
1929{
1930 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1931 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
1932 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
1933
1934 ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
1935}
1936
ccd979bd
MF		 1937/* Reference counting of the dlm debug structure. We want this because
1938 * open references on the debug inodes can live on after a mount, so
1939 * we can't rely on the ocfs2_super to always exist. */
1940static void ocfs2_dlm_debug_free(struct kref *kref)
1941{
1942 struct ocfs2_dlm_debug *dlm_debug;
1943
1944 dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
1945
1946 kfree(dlm_debug);
1947}
1948
1949void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
1950{
1951 if (dlm_debug)
1952 kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
1953}
1954
1955static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
1956{
1957 kref_get(&debug->d_refcnt);
1958}
1959
1960struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
1961{
1962 struct ocfs2_dlm_debug *dlm_debug;
1963
1964 dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
1965 if (!dlm_debug) {
1966 mlog_errno(-ENOMEM);
1967 goto out;
1968 }
1969
1970 kref_init(&dlm_debug->d_refcnt);
1971 INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
1972 dlm_debug->d_locking_state = NULL;
1973out:
1974 return dlm_debug;
1975}
1976
1977/* Access to this is arbitrated for us via seq_file->sem. */
1978struct ocfs2_dlm_seq_priv {
1979 struct ocfs2_dlm_debug *p_dlm_debug;
1980 struct ocfs2_lock_res p_iter_res;
1981 struct ocfs2_lock_res p_tmp_res;
1982};
1983
1984static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
1985 struct ocfs2_dlm_seq_priv *priv)
1986{
1987 struct ocfs2_lock_res *iter, *ret = NULL;
1988 struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
1989
1990 assert_spin_locked(&ocfs2_dlm_tracking_lock);
1991
1992 list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
1993 /* discover the head of the list */
1994 if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
1995 mlog(0, "End of list found, %p\n", ret);
1996 break;
1997 }
1998
1999 /* We track our "dummy" iteration lockres' by a NULL
2000 * l_ops field. */
2001 if (iter->l_ops != NULL) {
2002 ret = iter;
2003 break;
2004 }
2005 }
2006
2007 return ret;
2008}
2009
2010static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2011{
2012 struct ocfs2_dlm_seq_priv *priv = m->private;
2013 struct ocfs2_lock_res *iter;
2014
2015 spin_lock(&ocfs2_dlm_tracking_lock);
2016 iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2017 if (iter) {
2018 /* Since lockres' have the lifetime of their container
2019 * (which can be inodes, ocfs2_supers, etc) we want to
2020 * copy this out to a temporary lockres while still
2021 * under the spinlock. Obviously after this we can't
2022 * trust any pointers on the copy returned, but that's
2023 * ok as the information we want isn't typically held
2024 * in them. */
2025 priv->p_tmp_res = *iter;
2026 iter = &priv->p_tmp_res;
2027 }
2028 spin_unlock(&ocfs2_dlm_tracking_lock);
2029
2030 return iter;
2031}
2032
2033static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2034{
2035}
2036
2037static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2038{
2039 struct ocfs2_dlm_seq_priv *priv = m->private;
2040 struct ocfs2_lock_res *iter = v;
2041 struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2042
2043 spin_lock(&ocfs2_dlm_tracking_lock);
2044 iter = ocfs2_dlm_next_res(iter, priv);
2045 list_del_init(&dummy->l_debug_list);
2046 if (iter) {
2047 list_add(&dummy->l_debug_list, &iter->l_debug_list);
2048 priv->p_tmp_res = *iter;
2049 iter = &priv->p_tmp_res;
2050 }
2051 spin_unlock(&ocfs2_dlm_tracking_lock);
2052
2053 return iter;
2054}
2055
2056/* So that debugfs.ocfs2 can determine which format is being used */
2057#define OCFS2_DLM_DEBUG_STR_VERSION 1
2058static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2059{
2060 int i;
2061 char *lvb;
2062 struct ocfs2_lock_res *lockres = v;
2063
2064 if (!lockres)
2065 return -EINVAL;
2066
d680efe9
MF		 2067 seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2068
2069 if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2070 seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2071 lockres->l_name,
2072 (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2073 else
2074 seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2075
2076 seq_printf(m, "%d\t"
ccd979bd
MF		 2077 "0x%lx\t"
2078 "0x%x\t"
2079 "0x%x\t"
2080 "%u\t"
2081 "%u\t"
2082 "%d\t"
2083 "%d\t",
ccd979bd
MF		 2084 lockres->l_level,
2085 lockres->l_flags,
2086 lockres->l_action,
2087 lockres->l_unlock_action,
2088 lockres->l_ro_holders,
2089 lockres->l_ex_holders,
2090 lockres->l_requested,
2091 lockres->l_blocking);
2092
2093 /* Dump the raw LVB */
2094 lvb = lockres->l_lksb.lvb;
2095 for(i = 0; i < DLM_LVB_LEN; i++)
2096 seq_printf(m, "0x%x\t", lvb[i]);
2097
2098 /* End the line */
2099 seq_printf(m, "\n");
2100 return 0;
2101}
2102
2103static struct seq_operations ocfs2_dlm_seq_ops = {
2104 .start = ocfs2_dlm_seq_start,
2105 .stop = ocfs2_dlm_seq_stop,
2106 .next = ocfs2_dlm_seq_next,
2107 .show = ocfs2_dlm_seq_show,
2108};
2109
2110static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2111{
2112 struct seq_file *seq = (struct seq_file *) file->private_data;
2113 struct ocfs2_dlm_seq_priv *priv = seq->private;
2114 struct ocfs2_lock_res *res = &priv->p_iter_res;
2115
2116 ocfs2_remove_lockres_tracking(res);
2117 ocfs2_put_dlm_debug(priv->p_dlm_debug);
2118 return seq_release_private(inode, file);
2119}
2120
2121static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2122{
2123 int ret;
2124 struct ocfs2_dlm_seq_priv *priv;
2125 struct seq_file *seq;
2126 struct ocfs2_super *osb;
2127
2128 priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2129 if (!priv) {
2130 ret = -ENOMEM;
2131 mlog_errno(ret);
2132 goto out;
2133 }
2134 osb = (struct ocfs2_super *) inode->u.generic_ip;
2135 ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2136 priv->p_dlm_debug = osb->osb_dlm_debug;
2137 INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2138
2139 ret = seq_open(file, &ocfs2_dlm_seq_ops);
2140 if (ret) {
2141 kfree(priv);
2142 mlog_errno(ret);
2143 goto out;
2144 }
2145
2146 seq = (struct seq_file *) file->private_data;
2147 seq->private = priv;
2148
2149 ocfs2_add_lockres_tracking(&priv->p_iter_res,
2150 priv->p_dlm_debug);
2151
2152out:
2153 return ret;
2154}
2155
4b6f5d20 2156static const struct file_operations ocfs2_dlm_debug_fops = {
ccd979bd
MF		 2157 .open = ocfs2_dlm_debug_open,
2158 .release = ocfs2_dlm_debug_release,
2159 .read = seq_read,
2160 .llseek = seq_lseek,
2161};
2162
2163static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2164{
2165 int ret = 0;
2166 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2167
2168 dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2169 S_IFREG|S_IRUSR,
2170 osb->osb_debug_root,
2171 osb,
2172 &ocfs2_dlm_debug_fops);
2173 if (!dlm_debug->d_locking_state) {
2174 ret = -EINVAL;
2175 mlog(ML_ERROR,
2176 "Unable to create locking state debugfs file.\n");
2177 goto out;
2178 }
2179
2180 ocfs2_get_dlm_debug(dlm_debug);
2181out:
2182 return ret;
2183}
2184
2185static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2186{
2187 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2188
2189 if (dlm_debug) {
2190 debugfs_remove(dlm_debug->d_locking_state);
2191 ocfs2_put_dlm_debug(dlm_debug);
2192 }
2193}
2194
2195int ocfs2_dlm_init(struct ocfs2_super *osb)
2196{
2197 int status;
2198 u32 dlm_key;
2199 struct dlm_ctxt *dlm;
2200
2201 mlog_entry_void();
2202
2203 status = ocfs2_dlm_init_debug(osb);
2204 if (status < 0) {
2205 mlog_errno(status);
2206 goto bail;
2207 }
2208
2209 /* launch vote thread */
78427043 2210 osb->vote_task = kthread_run(ocfs2_vote_thread, osb, "ocfs2vote");
ccd979bd
MF		 2211 if (IS_ERR(osb->vote_task)) {
2212 status = PTR_ERR(osb->vote_task);
2213 osb->vote_task = NULL;
2214 mlog_errno(status);
2215 goto bail;
2216 }
2217
2218 /* used by the dlm code to make message headers unique, each
2219 * node in this domain must agree on this. */
2220 dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str));
2221
2222 /* for now, uuid == domain */
2223 dlm = dlm_register_domain(osb->uuid_str, dlm_key);
2224 if (IS_ERR(dlm)) {
2225 status = PTR_ERR(dlm);
2226 mlog_errno(status);
2227 goto bail;
2228 }
2229
2230 ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
2231 ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
2232
2233 dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb);
2234
2235 osb->dlm = dlm;
2236
2237 status = 0;
2238bail:
2239 if (status < 0) {
2240 ocfs2_dlm_shutdown_debug(osb);
2241 if (osb->vote_task)
2242 kthread_stop(osb->vote_task);
2243 }
2244
2245 mlog_exit(status);
2246 return status;
2247}
2248
2249void ocfs2_dlm_shutdown(struct ocfs2_super *osb)
2250{
2251 mlog_entry_void();
2252
2253 dlm_unregister_eviction_cb(&osb->osb_eviction_cb);
2254
2255 ocfs2_drop_osb_locks(osb);
2256
2257 if (osb->vote_task) {
2258 kthread_stop(osb->vote_task);
2259 osb->vote_task = NULL;
2260 }
2261
2262 ocfs2_lock_res_free(&osb->osb_super_lockres);
2263 ocfs2_lock_res_free(&osb->osb_rename_lockres);
2264
2265 dlm_unregister_domain(osb->dlm);
2266 osb->dlm = NULL;
2267
2268 ocfs2_dlm_shutdown_debug(osb);
2269
2270 mlog_exit_void();
2271}
2272
2273static void ocfs2_unlock_ast_func(void *opaque, enum dlm_status status)
2274{
2275 struct ocfs2_lock_res *lockres = opaque;
2276 unsigned long flags;
2277
2278 mlog_entry_void();
2279
2280 mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
2281 lockres->l_unlock_action);
2282
2283 spin_lock_irqsave(&lockres->l_lock, flags);
2284 /* We tried to cancel a convert request, but it was already
2285 * granted. All we want to do here is clear our unlock
2286 * state. The wake_up call done at the bottom is redundant
2287 * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't
2288 * hurt anything anyway */
2289 if (status == DLM_CANCELGRANT &&
2290 lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2291 mlog(0, "Got cancelgrant for %s\n", lockres->l_name);
2292
2293 /* We don't clear the busy flag in this case as it
2294 * should have been cleared by the ast which the dlm
2295 * has called. */
2296 goto complete_unlock;
2297 }
2298
2299 if (status != DLM_NORMAL) {
2300 mlog(ML_ERROR, "Dlm passes status %d for lock %s, "
2301 "unlock_action %d\n", status, lockres->l_name,
2302 lockres->l_unlock_action);
2303 spin_unlock_irqrestore(&lockres->l_lock, flags);
2304 return;
2305 }
2306
2307 switch(lockres->l_unlock_action) {
2308 case OCFS2_UNLOCK_CANCEL_CONVERT:
2309 mlog(0, "Cancel convert success for %s\n", lockres->l_name);
2310 lockres->l_action = OCFS2_AST_INVALID;
2311 break;
2312 case OCFS2_UNLOCK_DROP_LOCK:
2313 lockres->l_level = LKM_IVMODE;
2314 break;
2315 default:
2316 BUG();
2317 }
2318
2319 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
2320complete_unlock:
2321 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
2322 spin_unlock_irqrestore(&lockres->l_lock, flags);
2323
2324 wake_up(&lockres->l_event);
2325
2326 mlog_exit_void();
2327}
2328
2329typedef void (ocfs2_pre_drop_cb_t)(struct ocfs2_lock_res *, void *);
2330
2331struct drop_lock_cb {
2332 ocfs2_pre_drop_cb_t *drop_func;
2333 void *drop_data;
2334};
2335
2336static int ocfs2_drop_lock(struct ocfs2_super *osb,
2337 struct ocfs2_lock_res *lockres,
2338 struct drop_lock_cb *dcb)
2339{
2340 enum dlm_status status;
2341 unsigned long flags;
2342
2343 /* We didn't get anywhere near actually using this lockres. */
2344 if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
2345 goto out;
2346
2347 spin_lock_irqsave(&lockres->l_lock, flags);
2348
2349 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
2350 "lockres %s, flags 0x%lx\n",
2351 lockres->l_name, lockres->l_flags);
2352
2353 while (lockres->l_flags & OCFS2_LOCK_BUSY) {
2354 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2355 "%u, unlock_action = %u\n",
2356 lockres->l_name, lockres->l_flags, lockres->l_action,
2357 lockres->l_unlock_action);
2358
2359 spin_unlock_irqrestore(&lockres->l_lock, flags);
2360
2361 /* XXX: Today we just wait on any busy
2362 * locks... Perhaps we need to cancel converts in the
2363 * future? */
2364 ocfs2_wait_on_busy_lock(lockres);
2365
2366 spin_lock_irqsave(&lockres->l_lock, flags);
2367 }
2368
2369 if (dcb)
2370 dcb->drop_func(lockres, dcb->drop_data);
2371
2372 if (lockres->l_flags & OCFS2_LOCK_BUSY)
2373 mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
2374 lockres->l_name);
2375 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2376 mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
2377
2378 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
2379 spin_unlock_irqrestore(&lockres->l_lock, flags);
2380 goto out;
2381 }
2382
2383 lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
2384
2385 /* make sure we never get here while waiting for an ast to
2386 * fire. */
2387 BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
2388
2389 /* is this necessary? */
2390 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2391 lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
2392 spin_unlock_irqrestore(&lockres->l_lock, flags);
2393
2394 mlog(0, "lock %s\n", lockres->l_name);
2395
2396 status = dlmunlock(osb->dlm, &lockres->l_lksb, LKM_VALBLK,
2397 lockres->l_ops->unlock_ast, lockres);
2398 if (status != DLM_NORMAL) {
2399 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2400 mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
2401 dlm_print_one_lock(lockres->l_lksb.lockid);
2402 BUG();
2403 }
2404 mlog(0, "lock %s, successfull return from dlmunlock\n",
2405 lockres->l_name);
2406
2407 ocfs2_wait_on_busy_lock(lockres);
2408out:
2409 mlog_exit(0);
2410 return 0;
2411}
2412
2413/* Mark the lockres as being dropped. It will no longer be
2414 * queued if blocking, but we still may have to wait on it
2415 * being dequeued from the vote thread before we can consider
2416 * it safe to drop.
2417 *
2418 * You can *not* attempt to call cluster_lock on this lockres anymore. */
2419void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
2420{
2421 int status;
2422 struct ocfs2_mask_waiter mw;
2423 unsigned long flags;
2424
2425 ocfs2_init_mask_waiter(&mw);
2426
2427 spin_lock_irqsave(&lockres->l_lock, flags);
2428 lockres->l_flags |= OCFS2_LOCK_FREEING;
2429 while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
2430 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
2431 spin_unlock_irqrestore(&lockres->l_lock, flags);
2432
2433 mlog(0, "Waiting on lockres %s\n", lockres->l_name);
2434
2435 status = ocfs2_wait_for_mask(&mw);
2436 if (status)
2437 mlog_errno(status);
2438
2439 spin_lock_irqsave(&lockres->l_lock, flags);
2440 }
2441 spin_unlock_irqrestore(&lockres->l_lock, flags);
2442}
2443
d680efe9
MF		 2444void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
2445 struct ocfs2_lock_res *lockres)
ccd979bd 2446{
d680efe9 2447 int ret;
ccd979bd 2448
d680efe9
MF		 2449 ocfs2_mark_lockres_freeing(lockres);
2450 ret = ocfs2_drop_lock(osb, lockres, NULL);
2451 if (ret)
2452 mlog_errno(ret);
2453}
ccd979bd 2454
d680efe9
MF		 2455static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
2456{
2457 ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
2458 ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
ccd979bd
MF		 2459}
2460
2461static void ocfs2_meta_pre_drop(struct ocfs2_lock_res *lockres, void *data)
2462{
2463 struct inode *inode = data;
2464
2465 /* the metadata lock requires a bit more work as we have an
2466 * LVB to worry about. */
2467 if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
2468 lockres->l_level == LKM_EXMODE &&
2469 !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2470 __ocfs2_stuff_meta_lvb(inode);
2471}
2472
2473int ocfs2_drop_inode_locks(struct inode *inode)
2474{
2475 int status, err;
2476 struct drop_lock_cb meta_dcb = { ocfs2_meta_pre_drop, inode, };
2477
2478 mlog_entry_void();
2479
2480 /* No need to call ocfs2_mark_lockres_freeing here -
2481 * ocfs2_clear_inode has done it for us. */
2482
2483 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2484 &OCFS2_I(inode)->ip_data_lockres,
2485 NULL);
2486 if (err < 0)
2487 mlog_errno(err);
2488
2489 status = err;
2490
2491 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2492 &OCFS2_I(inode)->ip_meta_lockres,
2493 &meta_dcb);
2494 if (err < 0)
2495 mlog_errno(err);
2496 if (err < 0 && !status)
2497 status = err;
2498
2499 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2500 &OCFS2_I(inode)->ip_rw_lockres,
2501 NULL);
2502 if (err < 0)
2503 mlog_errno(err);
2504 if (err < 0 && !status)
2505 status = err;
2506
2507 mlog_exit(status);
2508 return status;
2509}
2510
2511static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
2512 int new_level)
2513{
2514 assert_spin_locked(&lockres->l_lock);
2515
2516 BUG_ON(lockres->l_blocking <= LKM_NLMODE);
2517
2518 if (lockres->l_level <= new_level) {
2519 mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n",
2520 lockres->l_level, new_level);
2521 BUG();
2522 }
2523
2524 mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
2525 lockres->l_name, new_level, lockres->l_blocking);
2526
2527 lockres->l_action = OCFS2_AST_DOWNCONVERT;
2528 lockres->l_requested = new_level;
2529 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2530}
2531
2532static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
2533 struct ocfs2_lock_res *lockres,
2534 int new_level,
2535 int lvb)
2536{
2537 int ret, dlm_flags = LKM_CONVERT;
2538 enum dlm_status status;
2539
2540 mlog_entry_void();
2541
2542 if (lvb)
2543 dlm_flags |= LKM_VALBLK;
2544
2545 status = dlmlock(osb->dlm,
2546 new_level,
2547 &lockres->l_lksb,
2548 dlm_flags,
2549 lockres->l_name,
f0681062 2550 OCFS2_LOCK_ID_MAX_LEN - 1,
ccd979bd
MF		 2551 lockres->l_ops->ast,
2552 lockres,
2553 lockres->l_ops->bast);
2554 if (status != DLM_NORMAL) {
2555 ocfs2_log_dlm_error("dlmlock", status, lockres);
2556 ret = -EINVAL;
2557 ocfs2_recover_from_dlm_error(lockres, 1);
2558 goto bail;
2559 }
2560
2561 ret = 0;
2562bail:
2563 mlog_exit(ret);
2564 return ret;
2565}
2566
2567/* returns 1 when the caller should unlock and call dlmunlock */
2568static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
2569 struct ocfs2_lock_res *lockres)
2570{
2571 assert_spin_locked(&lockres->l_lock);
2572
2573 mlog_entry_void();
2574 mlog(0, "lock %s\n", lockres->l_name);
2575
2576 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2577 /* If we're already trying to cancel a lock conversion
2578 * then just drop the spinlock and allow the caller to
2579 * requeue this lock. */
2580
2581 mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
2582 return 0;
2583 }
2584
2585 /* were we in a convert when we got the bast fire? */
2586 BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
2587 lockres->l_action != OCFS2_AST_DOWNCONVERT);
2588 /* set things up for the unlockast to know to just
2589 * clear out the ast_action and unset busy, etc. */
2590 lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
2591
2592 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
2593 "lock %s, invalid flags: 0x%lx\n",
2594 lockres->l_name, lockres->l_flags);
2595
2596 return 1;
2597}
2598
2599static int ocfs2_cancel_convert(struct ocfs2_super *osb,
2600 struct ocfs2_lock_res *lockres)
2601{
2602 int ret;
2603 enum dlm_status status;
2604
2605 mlog_entry_void();
2606 mlog(0, "lock %s\n", lockres->l_name);
2607
2608 ret = 0;
2609 status = dlmunlock(osb->dlm,
2610 &lockres->l_lksb,
2611 LKM_CANCEL,
2612 lockres->l_ops->unlock_ast,
2613 lockres);
2614 if (status != DLM_NORMAL) {
2615 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2616 ret = -EINVAL;
2617 ocfs2_recover_from_dlm_error(lockres, 0);
2618 }
2619
2620 mlog(0, "lock %s return from dlmunlock\n", lockres->l_name);
2621
2622 mlog_exit(ret);
2623 return ret;
2624}
2625
2626static inline int ocfs2_can_downconvert_meta_lock(struct inode *inode,
2627 struct ocfs2_lock_res *lockres,
2628 int new_level)
2629{
2630 int ret;
2631
2632 mlog_entry_void();
2633
2634 BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE);
2635
2636 if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2637 ret = 0;
2638 mlog(0, "lockres %s currently being refreshed -- backing "
2639 "off!\n", lockres->l_name);
2640 } else if (new_level == LKM_PRMODE)
2641 ret = !lockres->l_ex_holders &&
2642 ocfs2_inode_fully_checkpointed(inode);
2643 else /* Must be NLMODE we're converting to. */
2644 ret = !lockres->l_ro_holders && !lockres->l_ex_holders &&
2645 ocfs2_inode_fully_checkpointed(inode);
2646
2647 mlog_exit(ret);
2648 return ret;
2649}
2650
2651static int ocfs2_do_unblock_meta(struct inode *inode,
2652 int *requeue)
2653{
2654 int new_level;
2655 int set_lvb = 0;
2656 int ret = 0;
2657 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres;
2658 unsigned long flags;
2659
2660 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2661
2662 mlog_entry_void();
2663
2664 spin_lock_irqsave(&lockres->l_lock, flags);
2665
2666 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
2667
2668 mlog(0, "l_level=%d, l_blocking=%d\n", lockres->l_level,
2669 lockres->l_blocking);
2670
2671 BUG_ON(lockres->l_level != LKM_EXMODE &&
2672 lockres->l_level != LKM_PRMODE);
2673
2674 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
2675 *requeue = 1;
2676 ret = ocfs2_prepare_cancel_convert(osb, lockres);
2677 spin_unlock_irqrestore(&lockres->l_lock, flags);
2678 if (ret) {
2679 ret = ocfs2_cancel_convert(osb, lockres);
2680 if (ret < 0)
2681 mlog_errno(ret);
2682 }
2683 goto leave;
2684 }
2685
2686 new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
2687
2688 mlog(0, "l_level=%d, l_blocking=%d, new_level=%d\n",
2689 lockres->l_level, lockres->l_blocking, new_level);
2690
2691 if (ocfs2_can_downconvert_meta_lock(inode, lockres, new_level)) {
2692 if (lockres->l_level == LKM_EXMODE)
2693 set_lvb = 1;
2694
2695 /* If the lock hasn't been refreshed yet (rare), then
2696 * our memory inode values are old and we skip
2697 * stuffing the lvb. There's no need to actually clear
2698 * out the lvb here as it's value is still valid. */
2699 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2700 if (set_lvb)
2701 __ocfs2_stuff_meta_lvb(inode);
2702 } else
2703 mlog(0, "lockres %s: downconverting stale lock!\n",
2704 lockres->l_name);
2705
2706 mlog(0, "calling ocfs2_downconvert_lock with l_level=%d, "
2707 "l_blocking=%d, new_level=%d\n",
2708 lockres->l_level, lockres->l_blocking, new_level);
2709
2710 ocfs2_prepare_downconvert(lockres, new_level);
2711 spin_unlock_irqrestore(&lockres->l_lock, flags);
2712 ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb);
2713 goto leave;
2714 }
2715 if (!ocfs2_inode_fully_checkpointed(inode))
2716 ocfs2_start_checkpoint(osb);
2717
2718 *requeue = 1;
2719 spin_unlock_irqrestore(&lockres->l_lock, flags);
2720 ret = 0;
2721leave:
2722 mlog_exit(ret);
2723 return ret;
2724}
2725
2726static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb,
2727 struct ocfs2_lock_res *lockres,
d680efe9 2728 struct ocfs2_unblock_ctl *ctl,
ccd979bd
MF		 2729 ocfs2_convert_worker_t *worker)
2730{
2731 unsigned long flags;
2732 int blocking;
2733 int new_level;
2734 int ret = 0;
2735
2736 mlog_entry_void();
2737
2738 spin_lock_irqsave(&lockres->l_lock, flags);
2739
2740 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
2741
2742recheck:
2743 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
d680efe9 2744 ctl->requeue = 1;
ccd979bd
MF		 2745 ret = ocfs2_prepare_cancel_convert(osb, lockres);
2746 spin_unlock_irqrestore(&lockres->l_lock, flags);
2747 if (ret) {
2748 ret = ocfs2_cancel_convert(osb, lockres);
2749 if (ret < 0)
2750 mlog_errno(ret);
2751 }
2752 goto leave;
2753 }
2754
2755 /* if we're blocking an exclusive and we have *any* holders,
2756 * then requeue. */
2757 if ((lockres->l_blocking == LKM_EXMODE)
2758 && (lockres->l_ex_holders || lockres->l_ro_holders)) {
2759 spin_unlock_irqrestore(&lockres->l_lock, flags);
d680efe9 2760 ctl->requeue = 1;
ccd979bd
MF		 2761 ret = 0;
2762 goto leave;
2763 }
2764
2765 /* If it's a PR we're blocking, then only
2766 * requeue if we've got any EX holders */
2767 if (lockres->l_blocking == LKM_PRMODE &&
2768 lockres->l_ex_holders) {
2769 spin_unlock_irqrestore(&lockres->l_lock, flags);
d680efe9 2770 ctl->requeue = 1;
ccd979bd
MF		 2771 ret = 0;
2772 goto leave;
2773 }
2774
2775 /* If we get here, then we know that there are no more
2776 * incompatible holders (and anyone asking for an incompatible
2777 * lock is blocked). We can now downconvert the lock */
2778 if (!worker)
2779 goto downconvert;
2780
2781 /* Some lockres types want to do a bit of work before
2782 * downconverting a lock. Allow that here. The worker function
2783 * may sleep, so we save off a copy of what we're blocking as
2784 * it may change while we're not holding the spin lock. */
2785 blocking = lockres->l_blocking;
2786 spin_unlock_irqrestore(&lockres->l_lock, flags);
2787
d680efe9
MF		 2788 ctl->unblock_action = worker(lockres, blocking);
2789
2790 if (ctl->unblock_action == UNBLOCK_STOP_POST)
2791 goto leave;
ccd979bd
MF		 2792
2793 spin_lock_irqsave(&lockres->l_lock, flags);
2794 if (blocking != lockres->l_blocking) {
2795 /* If this changed underneath us, then we can't drop
2796 * it just yet. */
2797 goto recheck;
2798 }
2799
2800downconvert:
d680efe9 2801 ctl->requeue = 0;
ccd979bd
MF		 2802 new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
2803
2804 ocfs2_prepare_downconvert(lockres, new_level);
2805 spin_unlock_irqrestore(&lockres->l_lock, flags);
2806 ret = ocfs2_downconvert_lock(osb, lockres, new_level, 0);
2807leave:
2808 mlog_exit(ret);
2809 return ret;
2810}
2811
d680efe9
MF		 2812static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
2813 int blocking)
ccd979bd
MF		 2814{
2815 struct inode *inode;
2816 struct address_space *mapping;
2817
ccd979bd
MF		 2818 inode = ocfs2_lock_res_inode(lockres);
2819 mapping = inode->i_mapping;
2820
2821 if (filemap_fdatawrite(mapping)) {
b0697053
MF		 2822 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
2823 (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 2824 }
2825 sync_mapping_buffers(mapping);
2826 if (blocking == LKM_EXMODE) {
2827 truncate_inode_pages(mapping, 0);
2828 unmap_mapping_range(mapping, 0, 0, 0);
2829 } else {
2830 /* We only need to wait on the I/O if we're not also
2831 * truncating pages because truncate_inode_pages waits
2832 * for us above. We don't truncate pages if we're
2833 * blocking anything < EXMODE because we want to keep
2834 * them around in that case. */
2835 filemap_fdatawait(mapping);
2836 }
2837
d680efe9 2838 return UNBLOCK_CONTINUE;
ccd979bd
MF		 2839}
2840
2841int ocfs2_unblock_data(struct ocfs2_lock_res *lockres,
d680efe9 2842 struct ocfs2_unblock_ctl *ctl)
ccd979bd
MF		 2843{
2844 int status;
2845 struct inode *inode;
2846 struct ocfs2_super *osb;
2847
2848 mlog_entry_void();
2849
2850 inode = ocfs2_lock_res_inode(lockres);
2851 osb = OCFS2_SB(inode->i_sb);
2852
b0697053
MF		 2853 mlog(0, "unblock inode %llu\n",
2854 (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd 2855
d680efe9 2856 status = ocfs2_generic_unblock_lock(osb, lockres, ctl,
ccd979bd
MF		 2857 ocfs2_data_convert_worker);
2858 if (status < 0)
2859 mlog_errno(status);
2860
b0697053 2861 mlog(0, "inode %llu, requeue = %d\n",
d680efe9 2862 (unsigned long long)OCFS2_I(inode)->ip_blkno, ctl->requeue);
ccd979bd
MF		 2863
2864 mlog_exit(status);
2865 return status;
2866}
2867
2868static int ocfs2_unblock_inode_lock(struct ocfs2_lock_res *lockres,
d680efe9 2869 struct ocfs2_unblock_ctl *ctl)
ccd979bd
MF		 2870{
2871 int status;
2872 struct inode *inode;
2873
2874 mlog_entry_void();
2875
2876 mlog(0, "Unblock lockres %s\n", lockres->l_name);
2877
2878 inode = ocfs2_lock_res_inode(lockres);
2879
2880 status = ocfs2_generic_unblock_lock(OCFS2_SB(inode->i_sb),
d680efe9 2881 lockres, ctl, NULL);
ccd979bd
MF		 2882 if (status < 0)
2883 mlog_errno(status);
2884
2885 mlog_exit(status);
2886 return status;
2887}
2888
d680efe9
MF		 2889static int ocfs2_unblock_meta(struct ocfs2_lock_res *lockres,
2890 struct ocfs2_unblock_ctl *ctl)
ccd979bd
MF		 2891{
2892 int status;
2893 struct inode *inode;
2894
2895 mlog_entry_void();
2896
2897 inode = ocfs2_lock_res_inode(lockres);
2898
b0697053
MF		 2899 mlog(0, "unblock inode %llu\n",
2900 (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd 2901
d680efe9 2902 status = ocfs2_do_unblock_meta(inode, &ctl->requeue);
ccd979bd
MF		 2903 if (status < 0)
2904 mlog_errno(status);
2905
b0697053 2906 mlog(0, "inode %llu, requeue = %d\n",
d680efe9 2907 (unsigned long long)OCFS2_I(inode)->ip_blkno, ctl->requeue);
ccd979bd
MF		 2908
2909 mlog_exit(status);
2910 return status;
2911}
2912
d680efe9
MF		 2913/*
2914 * Does the final reference drop on our dentry lock. Right now this
2915 * happens in the vote thread, but we could choose to simplify the
2916 * dlmglue API and push these off to the ocfs2_wq in the future.
2917 */
2918static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
2919 struct ocfs2_lock_res *lockres)
2920{
2921 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
2922 ocfs2_dentry_lock_put(osb, dl);
2923}
2924
2925/*
2926 * d_delete() matching dentries before the lock downconvert.
2927 *
2928 * At this point, any process waiting to destroy the
2929 * dentry_lock due to last ref count is stopped by the
2930 * OCFS2_LOCK_QUEUED flag.
2931 *
2932 * We have two potential problems
2933 *
2934 * 1) If we do the last reference drop on our dentry_lock (via dput)
2935 * we'll wind up in ocfs2_release_dentry_lock(), waiting on
2936 * the downconvert to finish. Instead we take an elevated
2937 * reference and push the drop until after we've completed our
2938 * unblock processing.
2939 *
2940 * 2) There might be another process with a final reference,
2941 * waiting on us to finish processing. If this is the case, we
2942 * detect it and exit out - there's no more dentries anyway.
2943 */
2944static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
2945 int blocking)
2946{
2947 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
2948 struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
2949 struct dentry *dentry;
2950 unsigned long flags;
2951 int extra_ref = 0;
2952
2953 /*
2954 * This node is blocking another node from getting a read
2955 * lock. This happens when we've renamed within a
2956 * directory. We've forced the other nodes to d_delete(), but
2957 * we never actually dropped our lock because it's still
2958 * valid. The downconvert code will retain a PR for this node,
2959 * so there's no further work to do.
2960 */
2961 if (blocking == LKM_PRMODE)
2962 return UNBLOCK_CONTINUE;
2963
2964 /*
2965 * Mark this inode as potentially orphaned. The code in
2966 * ocfs2_delete_inode() will figure out whether it actually
2967 * needs to be freed or not.
2968 */
2969 spin_lock(&oi->ip_lock);
2970 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2971 spin_unlock(&oi->ip_lock);
2972
2973 /*
2974 * Yuck. We need to make sure however that the check of
2975 * OCFS2_LOCK_FREEING and the extra reference are atomic with
2976 * respect to a reference decrement or the setting of that
2977 * flag.
2978 */
2979 spin_lock_irqsave(&lockres->l_lock, flags);
2980 spin_lock(&dentry_attach_lock);
2981 if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
2982 && dl->dl_count) {
2983 dl->dl_count++;
2984 extra_ref = 1;
2985 }
2986 spin_unlock(&dentry_attach_lock);
2987 spin_unlock_irqrestore(&lockres->l_lock, flags);
2988
2989 mlog(0, "extra_ref = %d\n", extra_ref);
2990
2991 /*
2992 * We have a process waiting on us in ocfs2_dentry_iput(),
2993 * which means we can't have any more outstanding
2994 * aliases. There's no need to do any more work.
2995 */
2996 if (!extra_ref)
2997 return UNBLOCK_CONTINUE;
2998
2999 spin_lock(&dentry_attach_lock);
3000 while (1) {
3001 dentry = ocfs2_find_local_alias(dl->dl_inode,
3002 dl->dl_parent_blkno, 1);
3003 if (!dentry)
3004 break;
3005 spin_unlock(&dentry_attach_lock);
3006
3007 mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
3008 dentry->d_name.name);
3009
3010 /*
3011 * The following dcache calls may do an
3012 * iput(). Normally we don't want that from the
3013 * downconverting thread, but in this case it's ok
3014 * because the requesting node already has an
3015 * exclusive lock on the inode, so it can't be queued
3016 * for a downconvert.
3017 */
3018 d_delete(dentry);
3019 dput(dentry);
3020
3021 spin_lock(&dentry_attach_lock);
3022 }
3023 spin_unlock(&dentry_attach_lock);
3024
3025 /*
3026 * If we are the last holder of this dentry lock, there is no
3027 * reason to downconvert so skip straight to the unlock.
3028 */
3029 if (dl->dl_count == 1)
3030 return UNBLOCK_STOP_POST;
3031
3032 return UNBLOCK_CONTINUE_POST;
3033}
3034
3035static int ocfs2_unblock_dentry_lock(struct ocfs2_lock_res *lockres,
3036 struct ocfs2_unblock_ctl *ctl)
3037{
3038 int ret;
3039 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3040 struct ocfs2_super *osb = OCFS2_SB(dl->dl_inode->i_sb);
3041
3042 mlog(0, "unblock dentry lock: %llu\n",
3043 (unsigned long long)OCFS2_I(dl->dl_inode)->ip_blkno);
3044
3045 ret = ocfs2_generic_unblock_lock(osb,
3046 lockres,
3047 ctl,
3048 ocfs2_dentry_convert_worker);
3049 if (ret < 0)
3050 mlog_errno(ret);
3051
3052 mlog(0, "requeue = %d, post = %d\n", ctl->requeue, ctl->unblock_action);
3053
3054 return ret;
3055}
3056
ccd979bd
MF		 3057/* Generic unblock function for any lockres whose private data is an
3058 * ocfs2_super pointer. */
3059static int ocfs2_unblock_osb_lock(struct ocfs2_lock_res *lockres,
d680efe9 3060 struct ocfs2_unblock_ctl *ctl)
ccd979bd
MF		 3061{
3062 int status;
3063 struct ocfs2_super *osb;
3064
3065 mlog_entry_void();
3066
3067 mlog(0, "Unblock lockres %s\n", lockres->l_name);
3068
3069 osb = ocfs2_lock_res_super(lockres);
3070
3071 status = ocfs2_generic_unblock_lock(osb,
3072 lockres,
d680efe9 3073 ctl,
ccd979bd
MF		 3074 NULL);
3075 if (status < 0)
3076 mlog_errno(status);
3077
3078 mlog_exit(status);
3079 return status;
3080}
3081
3082void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3083 struct ocfs2_lock_res *lockres)
3084{
3085 int status;
d680efe9 3086 struct ocfs2_unblock_ctl ctl = {0, 0,};
ccd979bd
MF		 3087 unsigned long flags;
3088
3089 /* Our reference to the lockres in this function can be
3090 * considered valid until we remove the OCFS2_LOCK_QUEUED
3091 * flag. */
3092
3093 mlog_entry_void();
3094
3095 BUG_ON(!lockres);
3096 BUG_ON(!lockres->l_ops);
3097 BUG_ON(!lockres->l_ops->unblock);
3098
3099 mlog(0, "lockres %s blocked.\n", lockres->l_name);
3100
3101 /* Detect whether a lock has been marked as going away while
3102 * the vote thread was processing other things. A lock can
3103 * still be marked with OCFS2_LOCK_FREEING after this check,
3104 * but short circuiting here will still save us some
3105 * performance. */
3106 spin_lock_irqsave(&lockres->l_lock, flags);
3107 if (lockres->l_flags & OCFS2_LOCK_FREEING)
3108 goto unqueue;
3109 spin_unlock_irqrestore(&lockres->l_lock, flags);
3110
d680efe9 3111 status = lockres->l_ops->unblock(lockres, &ctl);
ccd979bd
MF		 3112 if (status < 0)
3113 mlog_errno(status);
3114
3115 spin_lock_irqsave(&lockres->l_lock, flags);
3116unqueue:
d680efe9 3117 if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
ccd979bd
MF		 3118 lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3119 } else
3120 ocfs2_schedule_blocked_lock(osb, lockres);
3121
3122 mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
d680efe9 3123 ctl.requeue ? "yes" : "no");
ccd979bd
MF		 3124 spin_unlock_irqrestore(&lockres->l_lock, flags);
3125
d680efe9
MF		 3126 if (ctl.unblock_action != UNBLOCK_CONTINUE
3127 && lockres->l_ops->post_unlock)
3128 lockres->l_ops->post_unlock(osb, lockres);
3129
ccd979bd
MF		 3130 mlog_exit_void();
3131}
3132
3133static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3134 struct ocfs2_lock_res *lockres)
3135{
3136 mlog_entry_void();
3137
3138 assert_spin_locked(&lockres->l_lock);
3139
3140 if (lockres->l_flags & OCFS2_LOCK_FREEING) {
3141 /* Do not schedule a lock for downconvert when it's on
3142 * the way to destruction - any nodes wanting access
3143 * to the resource will get it soon. */
3144 mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
3145 lockres->l_name, lockres->l_flags);
3146 return;
3147 }
3148
3149 lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
3150
3151 spin_lock(&osb->vote_task_lock);
3152 if (list_empty(&lockres->l_blocked_list)) {
3153 list_add_tail(&lockres->l_blocked_list,
3154 &osb->blocked_lock_list);
3155 osb->blocked_lock_count++;
3156 }
3157 spin_unlock(&osb->vote_task_lock);
3158
3159 mlog_exit_void();
3160}
3161
3162/* This aids in debugging situations where a bad LVB might be involved. */
3163void ocfs2_dump_meta_lvb_info(u64 level,
3164 const char *function,
3165 unsigned int line,
3166 struct ocfs2_lock_res *lockres)
3167{
3168 struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
3169
3170 mlog(level, "LVB information for %s (called from %s:%u):\n",
3171 lockres->l_name, function, line);
f9e2d82e
MF		 3172 mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
3173 lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
3174 be32_to_cpu(lvb->lvb_igeneration));
b0697053
MF		 3175 mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
3176 (unsigned long long)be64_to_cpu(lvb->lvb_isize),
3177 be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
3178 be16_to_cpu(lvb->lvb_imode));
3179 mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
ca4d147e 3180 "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
b0697053
MF		 3181 (long long)be64_to_cpu(lvb->lvb_iatime_packed),
3182 (long long)be64_to_cpu(lvb->lvb_ictime_packed),
ca4d147e
HP		 3183 (long long)be64_to_cpu(lvb->lvb_imtime_packed),
3184 be32_to_cpu(lvb->lvb_iattr));
ccd979bd 3185}
kernel source for telekom puls (alcatel/mediatek)