projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: Remember rw lock level during direct io
[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"
7f1a37e3 52#include "file.h"
ccd979bd
MF		 53#include "heartbeat.h"
54#include "inode.h"
55#include "journal.h"
56#include "slot_map.h"
57#include "super.h"
58#include "uptodate.h"
59#include "vote.h"
60
61#include "buffer_head_io.h"
62
63struct ocfs2_mask_waiter {
64 struct list_head mw_item;
65 int mw_status;
66 struct completion mw_complete;
67 unsigned long mw_mask;
68 unsigned long mw_goal;
69};
70
54a7e755
MF		 71static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
72static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
ccd979bd 73
d680efe9 74/*
cc567d89 75 * Return value from ->downconvert_worker functions.
d680efe9 76 *
b5e500e2 77 * These control the precise actions of ocfs2_unblock_lock()
d680efe9
MF		 78 * and ocfs2_process_blocked_lock()
79 *
80 */
81enum ocfs2_unblock_action {
82 UNBLOCK_CONTINUE = 0, /* Continue downconvert */
83 UNBLOCK_CONTINUE_POST = 1, /* Continue downconvert, fire
84 * ->post_unlock callback */
85 UNBLOCK_STOP_POST = 2, /* Do not downconvert, fire
86 * ->post_unlock() callback. */
87};
88
89struct ocfs2_unblock_ctl {
90 int requeue;
91 enum ocfs2_unblock_action unblock_action;
92};
93
810d5aeb
MF		 94static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
95 int new_level);
96static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
97
cc567d89
MF		 98static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
99 int blocking);
100
cc567d89
MF		 101static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
102 int blocking);
d680efe9
MF		 103
104static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
105 struct ocfs2_lock_res *lockres);
ccd979bd 106
f625c979
MF		 107/*
108 * OCFS2 Lock Resource Operations
109 *
110 * These fine tune the behavior of the generic dlmglue locking infrastructure.
0d5dc6c2
MF		 111 *
112 * The most basic of lock types can point ->l_priv to their respective
113 * struct ocfs2_super and allow the default actions to manage things.
114 *
115 * Right now, each lock type also needs to implement an init function,
116 * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
117 * should be called when the lock is no longer needed (i.e., object
118 * destruction time).
f625c979 119 */
ccd979bd 120struct ocfs2_lock_res_ops {
54a7e755
MF		 121 /*
122 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
123 * this callback if ->l_priv is not an ocfs2_super pointer
124 */
125 struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
b5e500e2 126
0d5dc6c2
MF		 127 /*
128 * Optionally called in the downconvert (or "vote") thread
129 * after a successful downconvert. The lockres will not be
130 * referenced after this callback is called, so it is safe to
131 * free memory, etc.
132 *
133 * The exact semantics of when this is called are controlled
134 * by ->downconvert_worker()
135 */
d680efe9 136 void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
f625c979 137
16d5b956
MF		 138 /*
139 * Allow a lock type to add checks to determine whether it is
140 * safe to downconvert a lock. Return 0 to re-queue the
141 * downconvert at a later time, nonzero to continue.
142 *
143 * For most locks, the default checks that there are no
144 * incompatible holders are sufficient.
145 *
146 * Called with the lockres spinlock held.
147 */
148 int (*check_downconvert)(struct ocfs2_lock_res *, int);
149
5ef0d4ea
MF		 150 /*
151 * Allows a lock type to populate the lock value block. This
152 * is called on downconvert, and when we drop a lock.
153 *
154 * Locks that want to use this should set LOCK_TYPE_USES_LVB
155 * in the flags field.
156 *
157 * Called with the lockres spinlock held.
158 */
159 void (*set_lvb)(struct ocfs2_lock_res *);
160
cc567d89
MF		 161 /*
162 * Called from the downconvert thread when it is determined
163 * that a lock will be downconverted. This is called without
164 * any locks held so the function can do work that might
165 * schedule (syncing out data, etc).
166 *
167 * This should return any one of the ocfs2_unblock_action
168 * values, depending on what it wants the thread to do.
169 */
170 int (*downconvert_worker)(struct ocfs2_lock_res *, int);
171
f625c979
MF		 172 /*
173 * LOCK_TYPE_* flags which describe the specific requirements
174 * of a lock type. Descriptions of each individual flag follow.
175 */
176 int flags;
ccd979bd
MF		 177};
178
f625c979
MF		 179/*
180 * Some locks want to "refresh" potentially stale data when a
181 * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
182 * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
183 * individual lockres l_flags member from the ast function. It is
184 * expected that the locking wrapper will clear the
185 * OCFS2_LOCK_NEEDS_REFRESH flag when done.
186 */
187#define LOCK_TYPE_REQUIRES_REFRESH 0x1
188
b80fc012 189/*
5ef0d4ea
MF		 190 * Indicate that a lock type makes use of the lock value block. The
191 * ->set_lvb lock type callback must be defined.
b80fc012
MF		 192 */
193#define LOCK_TYPE_USES_LVB 0x2
194
ccd979bd 195static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
54a7e755 196 .get_osb = ocfs2_get_inode_osb,
f625c979 197 .flags = 0,
ccd979bd
MF		 198};
199
200static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = {
54a7e755 201 .get_osb = ocfs2_get_inode_osb,
810d5aeb
MF		 202 .check_downconvert = ocfs2_check_meta_downconvert,
203 .set_lvb = ocfs2_set_meta_lvb,
b80fc012 204 .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
ccd979bd
MF		 205};
206
ccd979bd 207static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = {
54a7e755 208 .get_osb = ocfs2_get_inode_osb,
cc567d89 209 .downconvert_worker = ocfs2_data_convert_worker,
f625c979 210 .flags = 0,
ccd979bd
MF		 211};
212
213static struct ocfs2_lock_res_ops ocfs2_super_lops = {
f625c979 214 .flags = LOCK_TYPE_REQUIRES_REFRESH,
ccd979bd
MF		 215};
216
217static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
f625c979 218 .flags = 0,
ccd979bd
MF		 219};
220
d680efe9 221static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
54a7e755 222 .get_osb = ocfs2_get_dentry_osb,
d680efe9 223 .post_unlock = ocfs2_dentry_post_unlock,
cc567d89 224 .downconvert_worker = ocfs2_dentry_convert_worker,
f625c979 225 .flags = 0,
d680efe9
MF		 226};
227
50008630
TY		 228static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
229 .get_osb = ocfs2_get_inode_osb,
230 .flags = 0,
231};
232
ccd979bd
MF		 233static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
234{
235 return lockres->l_type == OCFS2_LOCK_TYPE_META ||
236 lockres->l_type == OCFS2_LOCK_TYPE_DATA ||
50008630
TY		 237 lockres->l_type == OCFS2_LOCK_TYPE_RW ||
238 lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
ccd979bd
MF		 239}
240
ccd979bd
MF		 241static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
242{
243 BUG_ON(!ocfs2_is_inode_lock(lockres));
244
245 return (struct inode *) lockres->l_priv;
246}
247
d680efe9
MF		 248static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
249{
250 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
251
252 return (struct ocfs2_dentry_lock *)lockres->l_priv;
253}
254
54a7e755
MF		 255static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
256{
257 if (lockres->l_ops->get_osb)
258 return lockres->l_ops->get_osb(lockres);
259
260 return (struct ocfs2_super *)lockres->l_priv;
261}
262
ccd979bd
MF		 263static int ocfs2_lock_create(struct ocfs2_super *osb,
264 struct ocfs2_lock_res *lockres,
265 int level,
266 int dlm_flags);
267static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
268 int wanted);
269static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
270 struct ocfs2_lock_res *lockres,
271 int level);
272static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
273static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
274static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
275static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
276static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
277 struct ocfs2_lock_res *lockres);
278static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
279 int convert);
280#define ocfs2_log_dlm_error(_func, _stat, _lockres) do { \
281 mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \
282 "resource %s: %s\n", dlm_errname(_stat), _func, \
283 _lockres->l_name, dlm_errmsg(_stat)); \
284} while (0)
285static void ocfs2_vote_on_unlock(struct ocfs2_super *osb,
286 struct ocfs2_lock_res *lockres);
287static int ocfs2_meta_lock_update(struct inode *inode,
288 struct buffer_head **bh);
289static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
290static inline int ocfs2_highest_compat_lock_level(int level);
ccd979bd 291
ccd979bd
MF		 292static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
293 u64 blkno,
294 u32 generation,
295 char *name)
296{
297 int len;
298
299 mlog_entry_void();
300
301 BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
302
b0697053
MF		 303 len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
304 ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
305 (long long)blkno, generation);
ccd979bd
MF		 306
307 BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
308
309 mlog(0, "built lock resource with name: %s\n", name);
310
311 mlog_exit_void();
312}
313
34af946a 314static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
ccd979bd
MF		 315
316static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
317 struct ocfs2_dlm_debug *dlm_debug)
318{
319 mlog(0, "Add tracking for lockres %s\n", res->l_name);
320
321 spin_lock(&ocfs2_dlm_tracking_lock);
322 list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
323 spin_unlock(&ocfs2_dlm_tracking_lock);
324}
325
326static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
327{
328 spin_lock(&ocfs2_dlm_tracking_lock);
329 if (!list_empty(&res->l_debug_list))
330 list_del_init(&res->l_debug_list);
331 spin_unlock(&ocfs2_dlm_tracking_lock);
332}
333
334static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
335 struct ocfs2_lock_res *res,
336 enum ocfs2_lock_type type,
ccd979bd
MF		 337 struct ocfs2_lock_res_ops *ops,
338 void *priv)
339{
ccd979bd
MF		 340 res->l_type = type;
341 res->l_ops = ops;
342 res->l_priv = priv;
343
344 res->l_level = LKM_IVMODE;
345 res->l_requested = LKM_IVMODE;
346 res->l_blocking = LKM_IVMODE;
347 res->l_action = OCFS2_AST_INVALID;
348 res->l_unlock_action = OCFS2_UNLOCK_INVALID;
349
350 res->l_flags = OCFS2_LOCK_INITIALIZED;
351
352 ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
353}
354
355void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
356{
357 /* This also clears out the lock status block */
358 memset(res, 0, sizeof(struct ocfs2_lock_res));
359 spin_lock_init(&res->l_lock);
360 init_waitqueue_head(&res->l_event);
361 INIT_LIST_HEAD(&res->l_blocked_list);
362 INIT_LIST_HEAD(&res->l_mask_waiters);
363}
364
365void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
366 enum ocfs2_lock_type type,
24c19ef4 367 unsigned int generation,
ccd979bd
MF		 368 struct inode *inode)
369{
370 struct ocfs2_lock_res_ops *ops;
371
372 switch(type) {
373 case OCFS2_LOCK_TYPE_RW:
374 ops = &ocfs2_inode_rw_lops;
375 break;
376 case OCFS2_LOCK_TYPE_META:
377 ops = &ocfs2_inode_meta_lops;
378 break;
379 case OCFS2_LOCK_TYPE_DATA:
380 ops = &ocfs2_inode_data_lops;
381 break;
50008630
TY		 382 case OCFS2_LOCK_TYPE_OPEN:
383 ops = &ocfs2_inode_open_lops;
384 break;
ccd979bd
MF		 385 default:
386 mlog_bug_on_msg(1, "type: %d\n", type);
387 ops = NULL; /* thanks, gcc */
388 break;
389 };
390
d680efe9 391 ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
24c19ef4 392 generation, res->l_name);
d680efe9
MF		 393 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
394}
395
54a7e755
MF		 396static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
397{
398 struct inode *inode = ocfs2_lock_res_inode(lockres);
399
400 return OCFS2_SB(inode->i_sb);
401}
402
d680efe9
MF		 403static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
404{
405 __be64 inode_blkno_be;
406
407 memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
408 sizeof(__be64));
409
410 return be64_to_cpu(inode_blkno_be);
411}
412
54a7e755
MF		 413static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
414{
415 struct ocfs2_dentry_lock *dl = lockres->l_priv;
416
417 return OCFS2_SB(dl->dl_inode->i_sb);
418}
419
d680efe9
MF		 420void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
421 u64 parent, struct inode *inode)
422{
423 int len;
424 u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
425 __be64 inode_blkno_be = cpu_to_be64(inode_blkno);
426 struct ocfs2_lock_res *lockres = &dl->dl_lockres;
427
428 ocfs2_lock_res_init_once(lockres);
429
430 /*
431 * Unfortunately, the standard lock naming scheme won't work
432 * here because we have two 16 byte values to use. Instead,
433 * we'll stuff the inode number as a binary value. We still
434 * want error prints to show something without garbling the
435 * display, so drop a null byte in there before the inode
436 * number. A future version of OCFS2 will likely use all
437 * binary lock names. The stringified names have been a
438 * tremendous aid in debugging, but now that the debugfs
439 * interface exists, we can mangle things there if need be.
440 *
441 * NOTE: We also drop the standard "pad" value (the total lock
442 * name size stays the same though - the last part is all
443 * zeros due to the memset in ocfs2_lock_res_init_once()
444 */
445 len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
446 "%c%016llx",
447 ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
448 (long long)parent);
449
450 BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
451
452 memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
453 sizeof(__be64));
454
455 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
456 OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
457 dl);
ccd979bd
MF		 458}
459
460static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
461 struct ocfs2_super *osb)
462{
463 /* Superblock lockres doesn't come from a slab so we call init
464 * once on it manually. */
465 ocfs2_lock_res_init_once(res);
d680efe9
MF		 466 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
467 0, res->l_name);
ccd979bd 468 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
ccd979bd
MF		 469 &ocfs2_super_lops, osb);
470}
471
472static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
473 struct ocfs2_super *osb)
474{
475 /* Rename lockres doesn't come from a slab so we call init
476 * once on it manually. */
477 ocfs2_lock_res_init_once(res);
d680efe9
MF		 478 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
479 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
ccd979bd
MF		 480 &ocfs2_rename_lops, osb);
481}
482
483void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
484{
485 mlog_entry_void();
486
487 if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
488 return;
489
490 ocfs2_remove_lockres_tracking(res);
491
492 mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
493 "Lockres %s is on the blocked list\n",
494 res->l_name);
495 mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
496 "Lockres %s has mask waiters pending\n",
497 res->l_name);
498 mlog_bug_on_msg(spin_is_locked(&res->l_lock),
499 "Lockres %s is locked\n",
500 res->l_name);
501 mlog_bug_on_msg(res->l_ro_holders,
502 "Lockres %s has %u ro holders\n",
503 res->l_name, res->l_ro_holders);
504 mlog_bug_on_msg(res->l_ex_holders,
505 "Lockres %s has %u ex holders\n",
506 res->l_name, res->l_ex_holders);
507
508 /* Need to clear out the lock status block for the dlm */
509 memset(&res->l_lksb, 0, sizeof(res->l_lksb));
510
511 res->l_flags = 0UL;
512 mlog_exit_void();
513}
514
515static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
516 int level)
517{
518 mlog_entry_void();
519
520 BUG_ON(!lockres);
521
522 switch(level) {
523 case LKM_EXMODE:
524 lockres->l_ex_holders++;
525 break;
526 case LKM_PRMODE:
527 lockres->l_ro_holders++;
528 break;
529 default:
530 BUG();
531 }
532
533 mlog_exit_void();
534}
535
536static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
537 int level)
538{
539 mlog_entry_void();
540
541 BUG_ON(!lockres);
542
543 switch(level) {
544 case LKM_EXMODE:
545 BUG_ON(!lockres->l_ex_holders);
546 lockres->l_ex_holders--;
547 break;
548 case LKM_PRMODE:
549 BUG_ON(!lockres->l_ro_holders);
550 lockres->l_ro_holders--;
551 break;
552 default:
553 BUG();
554 }
555 mlog_exit_void();
556}
557
558/* WARNING: This function lives in a world where the only three lock
559 * levels are EX, PR, and NL. It *will* have to be adjusted when more
560 * lock types are added. */
561static inline int ocfs2_highest_compat_lock_level(int level)
562{
563 int new_level = LKM_EXMODE;
564
565 if (level == LKM_EXMODE)
566 new_level = LKM_NLMODE;
567 else if (level == LKM_PRMODE)
568 new_level = LKM_PRMODE;
569 return new_level;
570}
571
572static void lockres_set_flags(struct ocfs2_lock_res *lockres,
573 unsigned long newflags)
574{
575 struct list_head *pos, *tmp;
576 struct ocfs2_mask_waiter *mw;
577
578 assert_spin_locked(&lockres->l_lock);
579
580 lockres->l_flags = newflags;
581
582 list_for_each_safe(pos, tmp, &lockres->l_mask_waiters) {
583 mw = list_entry(pos, struct ocfs2_mask_waiter, mw_item);
584 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
585 continue;
586
587 list_del_init(&mw->mw_item);
588 mw->mw_status = 0;
589 complete(&mw->mw_complete);
590 }
591}
592static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
593{
594 lockres_set_flags(lockres, lockres->l_flags | or);
595}
596static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
597 unsigned long clear)
598{
599 lockres_set_flags(lockres, lockres->l_flags & ~clear);
600}
601
602static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
603{
604 mlog_entry_void();
605
606 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
607 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
608 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
609 BUG_ON(lockres->l_blocking <= LKM_NLMODE);
610
611 lockres->l_level = lockres->l_requested;
612 if (lockres->l_level <=
613 ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
614 lockres->l_blocking = LKM_NLMODE;
615 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
616 }
617 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
618
619 mlog_exit_void();
620}
621
622static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
623{
624 mlog_entry_void();
625
626 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
627 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
628
629 /* Convert from RO to EX doesn't really need anything as our
630 * information is already up to data. Convert from NL to
631 * *anything* however should mark ourselves as needing an
632 * update */
f625c979
MF		 633 if (lockres->l_level == LKM_NLMODE &&
634 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
ccd979bd
MF		 635 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
636
637 lockres->l_level = lockres->l_requested;
638 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
639
640 mlog_exit_void();
641}
642
643static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
644{
645 mlog_entry_void();
646
647 BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY));
648 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
649
650 if (lockres->l_requested > LKM_NLMODE &&
f625c979
MF		 651 !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
652 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
ccd979bd
MF		 653 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
654
655 lockres->l_level = lockres->l_requested;
656 lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
657 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
658
659 mlog_exit_void();
660}
661
ccd979bd
MF		 662static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
663 int level)
664{
665 int needs_downconvert = 0;
666 mlog_entry_void();
667
668 assert_spin_locked(&lockres->l_lock);
669
670 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
671
672 if (level > lockres->l_blocking) {
673 /* only schedule a downconvert if we haven't already scheduled
674 * one that goes low enough to satisfy the level we're
675 * blocking. this also catches the case where we get
676 * duplicate BASTs */
677 if (ocfs2_highest_compat_lock_level(level) <
678 ocfs2_highest_compat_lock_level(lockres->l_blocking))
679 needs_downconvert = 1;
680
681 lockres->l_blocking = level;
682 }
683
684 mlog_exit(needs_downconvert);
685 return needs_downconvert;
686}
687
aa2623ad 688static void ocfs2_blocking_ast(void *opaque, int level)
ccd979bd 689{
aa2623ad
MF		 690 struct ocfs2_lock_res *lockres = opaque;
691 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
ccd979bd
MF		 692 int needs_downconvert;
693 unsigned long flags;
694
ccd979bd
MF		 695 BUG_ON(level <= LKM_NLMODE);
696
aa2623ad
MF		 697 mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n",
698 lockres->l_name, level, lockres->l_level,
699 ocfs2_lock_type_string(lockres->l_type));
700
ccd979bd
MF		 701 spin_lock_irqsave(&lockres->l_lock, flags);
702 needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
703 if (needs_downconvert)
704 ocfs2_schedule_blocked_lock(osb, lockres);
705 spin_unlock_irqrestore(&lockres->l_lock, flags);
706
d680efe9
MF		 707 wake_up(&lockres->l_event);
708
ccd979bd 709 ocfs2_kick_vote_thread(osb);
ccd979bd
MF		 710}
711
e92d57df 712static void ocfs2_locking_ast(void *opaque)
ccd979bd 713{
e92d57df 714 struct ocfs2_lock_res *lockres = opaque;
ccd979bd
MF		 715 struct dlm_lockstatus *lksb = &lockres->l_lksb;
716 unsigned long flags;
717
718 spin_lock_irqsave(&lockres->l_lock, flags);
719
720 if (lksb->status != DLM_NORMAL) {
721 mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n",
722 lockres->l_name, lksb->status);
723 spin_unlock_irqrestore(&lockres->l_lock, flags);
724 return;
725 }
726
727 switch(lockres->l_action) {
728 case OCFS2_AST_ATTACH:
729 ocfs2_generic_handle_attach_action(lockres);
e92d57df 730 lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
ccd979bd
MF		 731 break;
732 case OCFS2_AST_CONVERT:
733 ocfs2_generic_handle_convert_action(lockres);
734 break;
735 case OCFS2_AST_DOWNCONVERT:
736 ocfs2_generic_handle_downconvert_action(lockres);
737 break;
738 default:
e92d57df
MF		 739 mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
740 "lockres flags = 0x%lx, unlock action: %u\n",
741 lockres->l_name, lockres->l_action, lockres->l_flags,
742 lockres->l_unlock_action);
ccd979bd
MF		 743 BUG();
744 }
745
ccd979bd
MF		 746 /* set it to something invalid so if we get called again we
747 * can catch it. */
748 lockres->l_action = OCFS2_AST_INVALID;
ccd979bd
MF		 749
750 wake_up(&lockres->l_event);
d680efe9 751 spin_unlock_irqrestore(&lockres->l_lock, flags);
ccd979bd
MF		 752}
753
ccd979bd
MF		 754static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
755 int convert)
756{
757 unsigned long flags;
758
759 mlog_entry_void();
760 spin_lock_irqsave(&lockres->l_lock, flags);
761 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
762 if (convert)
763 lockres->l_action = OCFS2_AST_INVALID;
764 else
765 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
766 spin_unlock_irqrestore(&lockres->l_lock, flags);
767
768 wake_up(&lockres->l_event);
769 mlog_exit_void();
770}
771
772/* Note: If we detect another process working on the lock (i.e.,
773 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
774 * to do the right thing in that case.
775 */
776static int ocfs2_lock_create(struct ocfs2_super *osb,
777 struct ocfs2_lock_res *lockres,
778 int level,
779 int dlm_flags)
780{
781 int ret = 0;
c271c5c2 782 enum dlm_status status = DLM_NORMAL;
ccd979bd
MF		 783 unsigned long flags;
784
785 mlog_entry_void();
786
787 mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level,
788 dlm_flags);
789
790 spin_lock_irqsave(&lockres->l_lock, flags);
791 if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
792 (lockres->l_flags & OCFS2_LOCK_BUSY)) {
793 spin_unlock_irqrestore(&lockres->l_lock, flags);
794 goto bail;
795 }
796
797 lockres->l_action = OCFS2_AST_ATTACH;
798 lockres->l_requested = level;
799 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
800 spin_unlock_irqrestore(&lockres->l_lock, flags);
801
802 status = dlmlock(osb->dlm,
803 level,
804 &lockres->l_lksb,
805 dlm_flags,
806 lockres->l_name,
f0681062 807 OCFS2_LOCK_ID_MAX_LEN - 1,
e92d57df 808 ocfs2_locking_ast,
ccd979bd 809 lockres,
aa2623ad 810 ocfs2_blocking_ast);
ccd979bd
MF		 811 if (status != DLM_NORMAL) {
812 ocfs2_log_dlm_error("dlmlock", status, lockres);
813 ret = -EINVAL;
814 ocfs2_recover_from_dlm_error(lockres, 1);
815 }
816
817 mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name);
818
819bail:
820 mlog_exit(ret);
821 return ret;
822}
823
824static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
825 int flag)
826{
827 unsigned long flags;
828 int ret;
829
830 spin_lock_irqsave(&lockres->l_lock, flags);
831 ret = lockres->l_flags & flag;
832 spin_unlock_irqrestore(&lockres->l_lock, flags);
833
834 return ret;
835}
836
837static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
838
839{
840 wait_event(lockres->l_event,
841 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
842}
843
844static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
845
846{
847 wait_event(lockres->l_event,
848 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
849}
850
851/* predict what lock level we'll be dropping down to on behalf
852 * of another node, and return true if the currently wanted
853 * level will be compatible with it. */
854static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
855 int wanted)
856{
857 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
858
859 return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
860}
861
862static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
863{
864 INIT_LIST_HEAD(&mw->mw_item);
865 init_completion(&mw->mw_complete);
866}
867
868static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
869{
870 wait_for_completion(&mw->mw_complete);
871 /* Re-arm the completion in case we want to wait on it again */
872 INIT_COMPLETION(mw->mw_complete);
873 return mw->mw_status;
874}
875
876static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
877 struct ocfs2_mask_waiter *mw,
878 unsigned long mask,
879 unsigned long goal)
880{
881 BUG_ON(!list_empty(&mw->mw_item));
882
883 assert_spin_locked(&lockres->l_lock);
884
885 list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
886 mw->mw_mask = mask;
887 mw->mw_goal = goal;
888}
889
890/* returns 0 if the mw that was removed was already satisfied, -EBUSY
891 * if the mask still hadn't reached its goal */
892static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
893 struct ocfs2_mask_waiter *mw)
894{
895 unsigned long flags;
896 int ret = 0;
897
898 spin_lock_irqsave(&lockres->l_lock, flags);
899 if (!list_empty(&mw->mw_item)) {
900 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
901 ret = -EBUSY;
902
903 list_del_init(&mw->mw_item);
904 init_completion(&mw->mw_complete);
905 }
906 spin_unlock_irqrestore(&lockres->l_lock, flags);
907
908 return ret;
909
910}
911
912static int ocfs2_cluster_lock(struct ocfs2_super *osb,
913 struct ocfs2_lock_res *lockres,
914 int level,
915 int lkm_flags,
916 int arg_flags)
917{
918 struct ocfs2_mask_waiter mw;
919 enum dlm_status status;
920 int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
921 int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
922 unsigned long flags;
923
924 mlog_entry_void();
925
926 ocfs2_init_mask_waiter(&mw);
927
b80fc012
MF		 928 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
929 lkm_flags |= LKM_VALBLK;
930
ccd979bd
MF		 931again:
932 wait = 0;
933
934 if (catch_signals && signal_pending(current)) {
935 ret = -ERESTARTSYS;
936 goto out;
937 }
938
939 spin_lock_irqsave(&lockres->l_lock, flags);
940
941 mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
942 "Cluster lock called on freeing lockres %s! flags "
943 "0x%lx\n", lockres->l_name, lockres->l_flags);
944
945 /* We only compare against the currently granted level
946 * here. If the lock is blocked waiting on a downconvert,
947 * we'll get caught below. */
948 if (lockres->l_flags & OCFS2_LOCK_BUSY &&
949 level > lockres->l_level) {
950 /* is someone sitting in dlm_lock? If so, wait on
951 * them. */
952 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
953 wait = 1;
954 goto unlock;
955 }
956
957 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
958 /* lock has not been created yet. */
959 spin_unlock_irqrestore(&lockres->l_lock, flags);
960
961 ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0);
962 if (ret < 0) {
963 mlog_errno(ret);
964 goto out;
965 }
966 goto again;
967 }
968
969 if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
970 !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
971 /* is the lock is currently blocked on behalf of
972 * another node */
973 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
974 wait = 1;
975 goto unlock;
976 }
977
978 if (level > lockres->l_level) {
979 if (lockres->l_action != OCFS2_AST_INVALID)
980 mlog(ML_ERROR, "lockres %s has action %u pending\n",
981 lockres->l_name, lockres->l_action);
982
983 lockres->l_action = OCFS2_AST_CONVERT;
984 lockres->l_requested = level;
985 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
986 spin_unlock_irqrestore(&lockres->l_lock, flags);
987
988 BUG_ON(level == LKM_IVMODE);
989 BUG_ON(level == LKM_NLMODE);
990
991 mlog(0, "lock %s, convert from %d to level = %d\n",
992 lockres->l_name, lockres->l_level, level);
993
994 /* call dlm_lock to upgrade lock now */
995 status = dlmlock(osb->dlm,
996 level,
997 &lockres->l_lksb,
b80fc012 998 lkm_flags|LKM_CONVERT,
ccd979bd 999 lockres->l_name,
f0681062 1000 OCFS2_LOCK_ID_MAX_LEN - 1,
e92d57df 1001 ocfs2_locking_ast,
ccd979bd 1002 lockres,
aa2623ad 1003 ocfs2_blocking_ast);
ccd979bd
MF		 1004 if (status != DLM_NORMAL) {
1005 if ((lkm_flags & LKM_NOQUEUE) &&
1006 (status == DLM_NOTQUEUED))
1007 ret = -EAGAIN;
1008 else {
1009 ocfs2_log_dlm_error("dlmlock", status,
1010 lockres);
1011 ret = -EINVAL;
1012 }
1013 ocfs2_recover_from_dlm_error(lockres, 1);
1014 goto out;
1015 }
1016
1017 mlog(0, "lock %s, successfull return from dlmlock\n",
1018 lockres->l_name);
1019
1020 /* At this point we've gone inside the dlm and need to
1021 * complete our work regardless. */
1022 catch_signals = 0;
1023
1024 /* wait for busy to clear and carry on */
1025 goto again;
1026 }
1027
1028 /* Ok, if we get here then we're good to go. */
1029 ocfs2_inc_holders(lockres, level);
1030
1031 ret = 0;
1032unlock:
1033 spin_unlock_irqrestore(&lockres->l_lock, flags);
1034out:
1035 /*
1036 * This is helping work around a lock inversion between the page lock
1037 * and dlm locks. One path holds the page lock while calling aops
1038 * which block acquiring dlm locks. The voting thread holds dlm
1039 * locks while acquiring page locks while down converting data locks.
1040 * This block is helping an aop path notice the inversion and back
1041 * off to unlock its page lock before trying the dlm lock again.
1042 */
1043 if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1044 mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1045 wait = 0;
1046 if (lockres_remove_mask_waiter(lockres, &mw))
1047 ret = -EAGAIN;
1048 else
1049 goto again;
1050 }
1051 if (wait) {
1052 ret = ocfs2_wait_for_mask(&mw);
1053 if (ret == 0)
1054 goto again;
1055 mlog_errno(ret);
1056 }
1057
1058 mlog_exit(ret);
1059 return ret;
1060}
1061
1062static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
1063 struct ocfs2_lock_res *lockres,
1064 int level)
1065{
1066 unsigned long flags;
1067
1068 mlog_entry_void();
1069 spin_lock_irqsave(&lockres->l_lock, flags);
1070 ocfs2_dec_holders(lockres, level);
1071 ocfs2_vote_on_unlock(osb, lockres);
1072 spin_unlock_irqrestore(&lockres->l_lock, flags);
1073 mlog_exit_void();
1074}
1075
da66116e
AB		 1076static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1077 struct ocfs2_lock_res *lockres,
1078 int ex,
1079 int local)
ccd979bd 1080{
d680efe9 1081 int level = ex ? LKM_EXMODE : LKM_PRMODE;
ccd979bd 1082 unsigned long flags;
24c19ef4 1083 int lkm_flags = local ? LKM_LOCAL : 0;
ccd979bd
MF		 1084
1085 spin_lock_irqsave(&lockres->l_lock, flags);
1086 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1087 lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1088 spin_unlock_irqrestore(&lockres->l_lock, flags);
1089
24c19ef4 1090 return ocfs2_lock_create(osb, lockres, level, lkm_flags);
ccd979bd
MF		 1091}
1092
1093/* Grants us an EX lock on the data and metadata resources, skipping
1094 * the normal cluster directory lookup. Use this ONLY on newly created
1095 * inodes which other nodes can't possibly see, and which haven't been
1096 * hashed in the inode hash yet. This can give us a good performance
1097 * increase as it'll skip the network broadcast normally associated
1098 * with creating a new lock resource. */
1099int ocfs2_create_new_inode_locks(struct inode *inode)
1100{
1101 int ret;
d680efe9 1102 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1103
1104 BUG_ON(!inode);
1105 BUG_ON(!ocfs2_inode_is_new(inode));
1106
1107 mlog_entry_void();
1108
b0697053 1109 mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 1110
1111 /* NOTE: That we don't increment any of the holder counts, nor
1112 * do we add anything to a journal handle. Since this is
1113 * supposed to be a new inode which the cluster doesn't know
1114 * about yet, there is no need to. As far as the LVB handling
1115 * is concerned, this is basically like acquiring an EX lock
1116 * on a resource which has an invalid one -- we'll set it
1117 * valid when we release the EX. */
1118
24c19ef4 1119 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
ccd979bd
MF		 1120 if (ret) {
1121 mlog_errno(ret);
1122 goto bail;
1123 }
1124
24c19ef4
MF		 1125 /*
1126 * We don't want to use LKM_LOCAL on a meta data lock as they
1127 * don't use a generation in their lock names.
1128 */
1129 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_meta_lockres, 1, 0);
ccd979bd
MF		 1130 if (ret) {
1131 mlog_errno(ret);
1132 goto bail;
1133 }
1134
24c19ef4 1135 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_data_lockres, 1, 1);
ccd979bd
MF		 1136 if (ret) {
1137 mlog_errno(ret);
1138 goto bail;
1139 }
1140
50008630
TY		 1141 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1142 if (ret) {
1143 mlog_errno(ret);
1144 goto bail;
1145 }
1146
ccd979bd
MF		 1147bail:
1148 mlog_exit(ret);
1149 return ret;
1150}
1151
1152int ocfs2_rw_lock(struct inode *inode, int write)
1153{
1154 int status, level;
1155 struct ocfs2_lock_res *lockres;
c271c5c2 1156 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1157
1158 BUG_ON(!inode);
1159
1160 mlog_entry_void();
1161
b0697053
MF		 1162 mlog(0, "inode %llu take %s RW lock\n",
1163 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1164 write ? "EXMODE" : "PRMODE");
1165
c271c5c2
SM		 1166 if (ocfs2_mount_local(osb))
1167 return 0;
1168
ccd979bd
MF		 1169 lockres = &OCFS2_I(inode)->ip_rw_lockres;
1170
1171 level = write ? LKM_EXMODE : LKM_PRMODE;
1172
1173 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1174 0);
1175 if (status < 0)
1176 mlog_errno(status);
1177
1178 mlog_exit(status);
1179 return status;
1180}
1181
1182void ocfs2_rw_unlock(struct inode *inode, int write)
1183{
1184 int level = write ? LKM_EXMODE : LKM_PRMODE;
1185 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
c271c5c2 1186 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1187
1188 mlog_entry_void();
1189
b0697053
MF		 1190 mlog(0, "inode %llu drop %s RW lock\n",
1191 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1192 write ? "EXMODE" : "PRMODE");
1193
c271c5c2
SM		 1194 if (!ocfs2_mount_local(osb))
1195 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
ccd979bd
MF		 1196
1197 mlog_exit_void();
1198}
1199
50008630
TY		 1200/*
1201 * ocfs2_open_lock always get PR mode lock.
1202 */
1203int ocfs2_open_lock(struct inode *inode)
1204{
1205 int status = 0;
1206 struct ocfs2_lock_res *lockres;
1207 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1208
1209 BUG_ON(!inode);
1210
1211 mlog_entry_void();
1212
1213 mlog(0, "inode %llu take PRMODE open lock\n",
1214 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1215
1216 if (ocfs2_mount_local(osb))
1217 goto out;
1218
1219 lockres = &OCFS2_I(inode)->ip_open_lockres;
1220
1221 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1222 LKM_PRMODE, 0, 0);
1223 if (status < 0)
1224 mlog_errno(status);
1225
1226out:
1227 mlog_exit(status);
1228 return status;
1229}
1230
1231int ocfs2_try_open_lock(struct inode *inode, int write)
1232{
1233 int status = 0, level;
1234 struct ocfs2_lock_res *lockres;
1235 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1236
1237 BUG_ON(!inode);
1238
1239 mlog_entry_void();
1240
1241 mlog(0, "inode %llu try to take %s open lock\n",
1242 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1243 write ? "EXMODE" : "PRMODE");
1244
1245 if (ocfs2_mount_local(osb))
1246 goto out;
1247
1248 lockres = &OCFS2_I(inode)->ip_open_lockres;
1249
1250 level = write ? LKM_EXMODE : LKM_PRMODE;
1251
1252 /*
1253 * The file system may already holding a PRMODE/EXMODE open lock.
1254 * Since we pass LKM_NOQUEUE, the request won't block waiting on
1255 * other nodes and the -EAGAIN will indicate to the caller that
1256 * this inode is still in use.
1257 */
1258 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
1259 level, LKM_NOQUEUE, 0);
1260
1261out:
1262 mlog_exit(status);
1263 return status;
1264}
1265
1266/*
1267 * ocfs2_open_unlock unlock PR and EX mode open locks.
1268 */
1269void ocfs2_open_unlock(struct inode *inode)
1270{
1271 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1272 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1273
1274 mlog_entry_void();
1275
1276 mlog(0, "inode %llu drop open lock\n",
1277 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1278
1279 if (ocfs2_mount_local(osb))
1280 goto out;
1281
1282 if(lockres->l_ro_holders)
1283 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1284 LKM_PRMODE);
1285 if(lockres->l_ex_holders)
1286 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
1287 LKM_EXMODE);
1288
1289out:
1290 mlog_exit_void();
1291}
1292
ccd979bd
MF		 1293int ocfs2_data_lock_full(struct inode *inode,
1294 int write,
1295 int arg_flags)
1296{
1297 int status = 0, level;
1298 struct ocfs2_lock_res *lockres;
c271c5c2 1299 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1300
1301 BUG_ON(!inode);
1302
1303 mlog_entry_void();
1304
b0697053
MF		 1305 mlog(0, "inode %llu take %s DATA lock\n",
1306 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1307 write ? "EXMODE" : "PRMODE");
1308
1309 /* We'll allow faking a readonly data lock for
1310 * rodevices. */
1311 if (ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) {
1312 if (write) {
1313 status = -EROFS;
1314 mlog_errno(status);
1315 }
1316 goto out;
1317 }
1318
c271c5c2
SM		 1319 if (ocfs2_mount_local(osb))
1320 goto out;
1321
ccd979bd
MF		 1322 lockres = &OCFS2_I(inode)->ip_data_lockres;
1323
1324 level = write ? LKM_EXMODE : LKM_PRMODE;
1325
1326 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level,
1327 0, arg_flags);
1328 if (status < 0 && status != -EAGAIN)
1329 mlog_errno(status);
1330
1331out:
1332 mlog_exit(status);
1333 return status;
1334}
1335
1336/* see ocfs2_meta_lock_with_page() */
1337int ocfs2_data_lock_with_page(struct inode *inode,
1338 int write,
1339 struct page *page)
1340{
1341 int ret;
1342
1343 ret = ocfs2_data_lock_full(inode, write, OCFS2_LOCK_NONBLOCK);
1344 if (ret == -EAGAIN) {
1345 unlock_page(page);
1346 if (ocfs2_data_lock(inode, write) == 0)
1347 ocfs2_data_unlock(inode, write);
1348 ret = AOP_TRUNCATED_PAGE;
1349 }
1350
1351 return ret;
1352}
1353
1354static void ocfs2_vote_on_unlock(struct ocfs2_super *osb,
1355 struct ocfs2_lock_res *lockres)
1356{
1357 int kick = 0;
1358
1359 mlog_entry_void();
1360
1361 /* If we know that another node is waiting on our lock, kick
1362 * the vote thread * pre-emptively when we reach a release
1363 * condition. */
1364 if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1365 switch(lockres->l_blocking) {
1366 case LKM_EXMODE:
1367 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1368 kick = 1;
1369 break;
1370 case LKM_PRMODE:
1371 if (!lockres->l_ex_holders)
1372 kick = 1;
1373 break;
1374 default:
1375 BUG();
1376 }
1377 }
1378
1379 if (kick)
1380 ocfs2_kick_vote_thread(osb);
1381
1382 mlog_exit_void();
1383}
1384
1385void ocfs2_data_unlock(struct inode *inode,
1386 int write)
1387{
1388 int level = write ? LKM_EXMODE : LKM_PRMODE;
1389 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_data_lockres;
c271c5c2 1390 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1391
1392 mlog_entry_void();
1393
b0697053
MF		 1394 mlog(0, "inode %llu drop %s DATA lock\n",
1395 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1396 write ? "EXMODE" : "PRMODE");
1397
c271c5c2
SM		 1398 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
1399 !ocfs2_mount_local(osb))
ccd979bd
MF		 1400 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1401
1402 mlog_exit_void();
1403}
1404
1405#define OCFS2_SEC_BITS 34
1406#define OCFS2_SEC_SHIFT (64 - 34)
1407#define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1)
1408
1409/* LVB only has room for 64 bits of time here so we pack it for
1410 * now. */
1411static u64 ocfs2_pack_timespec(struct timespec *spec)
1412{
1413 u64 res;
1414 u64 sec = spec->tv_sec;
1415 u32 nsec = spec->tv_nsec;
1416
1417 res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
1418
1419 return res;
1420}
1421
1422/* Call this with the lockres locked. I am reasonably sure we don't
1423 * need ip_lock in this function as anyone who would be changing those
1424 * values is supposed to be blocked in ocfs2_meta_lock right now. */
1425static void __ocfs2_stuff_meta_lvb(struct inode *inode)
1426{
1427 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1428 struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
1429 struct ocfs2_meta_lvb *lvb;
1430
1431 mlog_entry_void();
1432
1433 lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1434
24c19ef4
MF		 1435 /*
1436 * Invalidate the LVB of a deleted inode - this way other
1437 * nodes are forced to go to disk and discover the new inode
1438 * status.
1439 */
1440 if (oi->ip_flags & OCFS2_INODE_DELETED) {
1441 lvb->lvb_version = 0;
1442 goto out;
1443 }
1444
4d3b83f7 1445 lvb->lvb_version = OCFS2_LVB_VERSION;
ccd979bd
MF		 1446 lvb->lvb_isize = cpu_to_be64(i_size_read(inode));
1447 lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
1448 lvb->lvb_iuid = cpu_to_be32(inode->i_uid);
1449 lvb->lvb_igid = cpu_to_be32(inode->i_gid);
1450 lvb->lvb_imode = cpu_to_be16(inode->i_mode);
1451 lvb->lvb_inlink = cpu_to_be16(inode->i_nlink);
1452 lvb->lvb_iatime_packed =
1453 cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
1454 lvb->lvb_ictime_packed =
1455 cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
1456 lvb->lvb_imtime_packed =
1457 cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
ca4d147e 1458 lvb->lvb_iattr = cpu_to_be32(oi->ip_attr);
f9e2d82e 1459 lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
ccd979bd 1460
24c19ef4 1461out:
ccd979bd
MF		 1462 mlog_meta_lvb(0, lockres);
1463
1464 mlog_exit_void();
1465}
1466
1467static void ocfs2_unpack_timespec(struct timespec *spec,
1468 u64 packed_time)
1469{
1470 spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
1471 spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
1472}
1473
1474static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
1475{
1476 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1477 struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
1478 struct ocfs2_meta_lvb *lvb;
1479
1480 mlog_entry_void();
1481
1482 mlog_meta_lvb(0, lockres);
1483
1484 lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1485
1486 /* We're safe here without the lockres lock... */
1487 spin_lock(&oi->ip_lock);
1488 oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
1489 i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
1490
ca4d147e
HP		 1491 oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
1492 ocfs2_set_inode_flags(inode);
1493
ccd979bd
MF		 1494 /* fast-symlinks are a special case */
1495 if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
1496 inode->i_blocks = 0;
1497 else
8110b073 1498 inode->i_blocks = ocfs2_inode_sector_count(inode);
ccd979bd
MF		 1499
1500 inode->i_uid = be32_to_cpu(lvb->lvb_iuid);
1501 inode->i_gid = be32_to_cpu(lvb->lvb_igid);
1502 inode->i_mode = be16_to_cpu(lvb->lvb_imode);
1503 inode->i_nlink = be16_to_cpu(lvb->lvb_inlink);
1504 ocfs2_unpack_timespec(&inode->i_atime,
1505 be64_to_cpu(lvb->lvb_iatime_packed));
1506 ocfs2_unpack_timespec(&inode->i_mtime,
1507 be64_to_cpu(lvb->lvb_imtime_packed));
1508 ocfs2_unpack_timespec(&inode->i_ctime,
1509 be64_to_cpu(lvb->lvb_ictime_packed));
1510 spin_unlock(&oi->ip_lock);
1511
1512 mlog_exit_void();
1513}
1514
f9e2d82e
MF		 1515static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
1516 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 1517{
1518 struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1519
f9e2d82e
MF		 1520 if (lvb->lvb_version == OCFS2_LVB_VERSION
1521 && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
ccd979bd
MF		 1522 return 1;
1523 return 0;
1524}
1525
1526/* Determine whether a lock resource needs to be refreshed, and
1527 * arbitrate who gets to refresh it.
1528 *
1529 * 0 means no refresh needed.
1530 *
1531 * > 0 means you need to refresh this and you MUST call
1532 * ocfs2_complete_lock_res_refresh afterwards. */
1533static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
1534{
1535 unsigned long flags;
1536 int status = 0;
1537
1538 mlog_entry_void();
1539
1540refresh_check:
1541 spin_lock_irqsave(&lockres->l_lock, flags);
1542 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
1543 spin_unlock_irqrestore(&lockres->l_lock, flags);
1544 goto bail;
1545 }
1546
1547 if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
1548 spin_unlock_irqrestore(&lockres->l_lock, flags);
1549
1550 ocfs2_wait_on_refreshing_lock(lockres);
1551 goto refresh_check;
1552 }
1553
1554 /* Ok, I'll be the one to refresh this lock. */
1555 lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
1556 spin_unlock_irqrestore(&lockres->l_lock, flags);
1557
1558 status = 1;
1559bail:
1560 mlog_exit(status);
1561 return status;
1562}
1563
1564/* If status is non zero, I'll mark it as not being in refresh
1565 * anymroe, but i won't clear the needs refresh flag. */
1566static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
1567 int status)
1568{
1569 unsigned long flags;
1570 mlog_entry_void();
1571
1572 spin_lock_irqsave(&lockres->l_lock, flags);
1573 lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
1574 if (!status)
1575 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
1576 spin_unlock_irqrestore(&lockres->l_lock, flags);
1577
1578 wake_up(&lockres->l_event);
1579
1580 mlog_exit_void();
1581}
1582
1583/* may or may not return a bh if it went to disk. */
1584static int ocfs2_meta_lock_update(struct inode *inode,
1585 struct buffer_head **bh)
1586{
1587 int status = 0;
1588 struct ocfs2_inode_info *oi = OCFS2_I(inode);
be9e986b 1589 struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
ccd979bd 1590 struct ocfs2_dinode *fe;
c271c5c2 1591 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1592
1593 mlog_entry_void();
1594
be9e986b
MF		 1595 if (ocfs2_mount_local(osb))
1596 goto bail;
1597
ccd979bd
MF		 1598 spin_lock(&oi->ip_lock);
1599 if (oi->ip_flags & OCFS2_INODE_DELETED) {
b0697053 1600 mlog(0, "Orphaned inode %llu was deleted while we "
ccd979bd 1601 "were waiting on a lock. ip_flags = 0x%x\n",
b0697053 1602 (unsigned long long)oi->ip_blkno, oi->ip_flags);
ccd979bd
MF		 1603 spin_unlock(&oi->ip_lock);
1604 status = -ENOENT;
1605 goto bail;
1606 }
1607 spin_unlock(&oi->ip_lock);
1608
be9e986b
MF		 1609 if (!ocfs2_should_refresh_lock_res(lockres))
1610 goto bail;
ccd979bd
MF		 1611
1612 /* This will discard any caching information we might have had
1613 * for the inode metadata. */
1614 ocfs2_metadata_cache_purge(inode);
1615
be9e986b 1616 if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
b0697053
MF		 1617 mlog(0, "Trusting LVB on inode %llu\n",
1618 (unsigned long long)oi->ip_blkno);
ccd979bd
MF		 1619 ocfs2_refresh_inode_from_lvb(inode);
1620 } else {
1621 /* Boo, we have to go to disk. */
1622 /* read bh, cast, ocfs2_refresh_inode */
1623 status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno,
1624 bh, OCFS2_BH_CACHED, inode);
1625 if (status < 0) {
1626 mlog_errno(status);
1627 goto bail_refresh;
1628 }
1629 fe = (struct ocfs2_dinode *) (*bh)->b_data;
1630
1631 /* This is a good chance to make sure we're not
1632 * locking an invalid object.
1633 *
1634 * We bug on a stale inode here because we checked
1635 * above whether it was wiped from disk. The wiping
1636 * node provides a guarantee that we receive that
1637 * message and can mark the inode before dropping any
1638 * locks associated with it. */
1639 if (!OCFS2_IS_VALID_DINODE(fe)) {
1640 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
1641 status = -EIO;
1642 goto bail_refresh;
1643 }
1644 mlog_bug_on_msg(inode->i_generation !=
1645 le32_to_cpu(fe->i_generation),
b0697053 1646 "Invalid dinode %llu disk generation: %u "
ccd979bd 1647 "inode->i_generation: %u\n",
b0697053
MF		 1648 (unsigned long long)oi->ip_blkno,
1649 le32_to_cpu(fe->i_generation),
ccd979bd
MF		 1650 inode->i_generation);
1651 mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
1652 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
b0697053
MF		 1653 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
1654 (unsigned long long)oi->ip_blkno,
1655 (unsigned long long)le64_to_cpu(fe->i_dtime),
ccd979bd
MF		 1656 le32_to_cpu(fe->i_flags));
1657
1658 ocfs2_refresh_inode(inode, fe);
1659 }
1660
1661 status = 0;
1662bail_refresh:
be9e986b 1663 ocfs2_complete_lock_res_refresh(lockres, status);
ccd979bd
MF		 1664bail:
1665 mlog_exit(status);
1666 return status;
1667}
1668
1669static int ocfs2_assign_bh(struct inode *inode,
1670 struct buffer_head **ret_bh,
1671 struct buffer_head *passed_bh)
1672{
1673 int status;
1674
1675 if (passed_bh) {
1676 /* Ok, the update went to disk for us, use the
1677 * returned bh. */
1678 *ret_bh = passed_bh;
1679 get_bh(*ret_bh);
1680
1681 return 0;
1682 }
1683
1684 status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1685 OCFS2_I(inode)->ip_blkno,
1686 ret_bh,
1687 OCFS2_BH_CACHED,
1688 inode);
1689 if (status < 0)
1690 mlog_errno(status);
1691
1692 return status;
1693}
1694
1695/*
1696 * returns < 0 error if the callback will never be called, otherwise
1697 * the result of the lock will be communicated via the callback.
1698 */
1699int ocfs2_meta_lock_full(struct inode *inode,
ccd979bd
MF		 1700 struct buffer_head **ret_bh,
1701 int ex,
1702 int arg_flags)
1703{
1704 int status, level, dlm_flags, acquired;
c271c5c2 1705 struct ocfs2_lock_res *lockres = NULL;
ccd979bd
MF		 1706 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1707 struct buffer_head *local_bh = NULL;
1708
1709 BUG_ON(!inode);
1710
1711 mlog_entry_void();
1712
b0697053
MF		 1713 mlog(0, "inode %llu, take %s META lock\n",
1714 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1715 ex ? "EXMODE" : "PRMODE");
1716
1717 status = 0;
1718 acquired = 0;
1719 /* We'll allow faking a readonly metadata lock for
1720 * rodevices. */
1721 if (ocfs2_is_hard_readonly(osb)) {
1722 if (ex)
1723 status = -EROFS;
1724 goto bail;
1725 }
1726
c271c5c2
SM		 1727 if (ocfs2_mount_local(osb))
1728 goto local;
1729
ccd979bd
MF		 1730 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1731 wait_event(osb->recovery_event,
1732 ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1733
ccd979bd
MF		 1734 lockres = &OCFS2_I(inode)->ip_meta_lockres;
1735 level = ex ? LKM_EXMODE : LKM_PRMODE;
1736 dlm_flags = 0;
1737 if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
1738 dlm_flags |= LKM_NOQUEUE;
1739
1740 status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
1741 if (status < 0) {
1742 if (status != -EAGAIN && status != -EIOCBRETRY)
1743 mlog_errno(status);
1744 goto bail;
1745 }
1746
1747 /* Notify the error cleanup path to drop the cluster lock. */
1748 acquired = 1;
1749
1750 /* We wait twice because a node may have died while we were in
1751 * the lower dlm layers. The second time though, we've
1752 * committed to owning this lock so we don't allow signals to
1753 * abort the operation. */
1754 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1755 wait_event(osb->recovery_event,
1756 ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1757
c271c5c2 1758local:
24c19ef4
MF		 1759 /*
1760 * We only see this flag if we're being called from
1761 * ocfs2_read_locked_inode(). It means we're locking an inode
1762 * which hasn't been populated yet, so clear the refresh flag
1763 * and let the caller handle it.
1764 */
1765 if (inode->i_state & I_NEW) {
1766 status = 0;
c271c5c2
SM		 1767 if (lockres)
1768 ocfs2_complete_lock_res_refresh(lockres, 0);
24c19ef4
MF		 1769 goto bail;
1770 }
1771
ccd979bd
MF		 1772 /* This is fun. The caller may want a bh back, or it may
1773 * not. ocfs2_meta_lock_update definitely wants one in, but
1774 * may or may not read one, depending on what's in the
1775 * LVB. The result of all of this is that we've *only* gone to
1776 * disk if we have to, so the complexity is worthwhile. */
1777 status = ocfs2_meta_lock_update(inode, &local_bh);
1778 if (status < 0) {
1779 if (status != -ENOENT)
1780 mlog_errno(status);
1781 goto bail;
1782 }
1783
1784 if (ret_bh) {
1785 status = ocfs2_assign_bh(inode, ret_bh, local_bh);
1786 if (status < 0) {
1787 mlog_errno(status);
1788 goto bail;
1789 }
1790 }
1791
ccd979bd
MF		 1792bail:
1793 if (status < 0) {
1794 if (ret_bh && (*ret_bh)) {
1795 brelse(*ret_bh);
1796 *ret_bh = NULL;
1797 }
1798 if (acquired)
1799 ocfs2_meta_unlock(inode, ex);
1800 }
1801
1802 if (local_bh)
1803 brelse(local_bh);
1804
1805 mlog_exit(status);
1806 return status;
1807}
1808
1809/*
1810 * This is working around a lock inversion between tasks acquiring DLM locks
1811 * while holding a page lock and the vote thread which blocks dlm lock acquiry
1812 * while acquiring page locks.
1813 *
1814 * ** These _with_page variantes are only intended to be called from aop
1815 * methods that hold page locks and return a very specific *positive* error
1816 * code that aop methods pass up to the VFS -- test for errors with != 0. **
1817 *
1818 * The DLM is called such that it returns -EAGAIN if it would have blocked
1819 * waiting for the vote thread. In that case we unlock our page so the vote
1820 * thread can make progress. Once we've done this we have to return
1821 * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up
1822 * into the VFS who will then immediately retry the aop call.
1823 *
1824 * We do a blocking lock and immediate unlock before returning, though, so that
1825 * the lock has a great chance of being cached on this node by the time the VFS
1826 * calls back to retry the aop. This has a potential to livelock as nodes
1827 * ping locks back and forth, but that's a risk we're willing to take to avoid
1828 * the lock inversion simply.
1829 */
1830int ocfs2_meta_lock_with_page(struct inode *inode,
ccd979bd
MF		 1831 struct buffer_head **ret_bh,
1832 int ex,
1833 struct page *page)
1834{
1835 int ret;
1836
4bcec184 1837 ret = ocfs2_meta_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
ccd979bd
MF		 1838 if (ret == -EAGAIN) {
1839 unlock_page(page);
4bcec184 1840 if (ocfs2_meta_lock(inode, ret_bh, ex) == 0)
ccd979bd
MF		 1841 ocfs2_meta_unlock(inode, ex);
1842 ret = AOP_TRUNCATED_PAGE;
1843 }
1844
1845 return ret;
1846}
1847
7f1a37e3
TY		 1848int ocfs2_meta_lock_atime(struct inode *inode,
1849 struct vfsmount *vfsmnt,
1850 int *level)
1851{
1852 int ret;
1853
1854 mlog_entry_void();
1855 ret = ocfs2_meta_lock(inode, NULL, 0);
1856 if (ret < 0) {
1857 mlog_errno(ret);
1858 return ret;
1859 }
1860
1861 /*
1862 * If we should update atime, we will get EX lock,
1863 * otherwise we just get PR lock.
1864 */
1865 if (ocfs2_should_update_atime(inode, vfsmnt)) {
1866 struct buffer_head *bh = NULL;
1867
1868 ocfs2_meta_unlock(inode, 0);
1869 ret = ocfs2_meta_lock(inode, &bh, 1);
1870 if (ret < 0) {
1871 mlog_errno(ret);
1872 return ret;
1873 }
1874 *level = 1;
1875 if (ocfs2_should_update_atime(inode, vfsmnt))
1876 ocfs2_update_inode_atime(inode, bh);
1877 if (bh)
1878 brelse(bh);
1879 } else
1880 *level = 0;
1881
1882 mlog_exit(ret);
1883 return ret;
1884}
1885
ccd979bd
MF		 1886void ocfs2_meta_unlock(struct inode *inode,
1887 int ex)
1888{
1889 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1890 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres;
c271c5c2 1891 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1892
1893 mlog_entry_void();
1894
b0697053
MF		 1895 mlog(0, "inode %llu drop %s META lock\n",
1896 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1897 ex ? "EXMODE" : "PRMODE");
1898
c271c5c2
SM		 1899 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
1900 !ocfs2_mount_local(osb))
ccd979bd
MF		 1901 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1902
1903 mlog_exit_void();
1904}
1905
1906int ocfs2_super_lock(struct ocfs2_super *osb,
1907 int ex)
1908{
c271c5c2 1909 int status = 0;
ccd979bd
MF		 1910 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1911 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1912 struct buffer_head *bh;
1913 struct ocfs2_slot_info *si = osb->slot_info;
1914
1915 mlog_entry_void();
1916
1917 if (ocfs2_is_hard_readonly(osb))
1918 return -EROFS;
1919
c271c5c2
SM		 1920 if (ocfs2_mount_local(osb))
1921 goto bail;
1922
ccd979bd
MF		 1923 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1924 if (status < 0) {
1925 mlog_errno(status);
1926 goto bail;
1927 }
1928
1929 /* The super block lock path is really in the best position to
1930 * know when resources covered by the lock need to be
1931 * refreshed, so we do it here. Of course, making sense of
1932 * everything is up to the caller :) */
1933 status = ocfs2_should_refresh_lock_res(lockres);
1934 if (status < 0) {
1935 mlog_errno(status);
1936 goto bail;
1937 }
1938 if (status) {
1939 bh = si->si_bh;
1940 status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0,
1941 si->si_inode);
1942 if (status == 0)
1943 ocfs2_update_slot_info(si);
1944
1945 ocfs2_complete_lock_res_refresh(lockres, status);
1946
1947 if (status < 0)
1948 mlog_errno(status);
1949 }
1950bail:
1951 mlog_exit(status);
1952 return status;
1953}
1954
1955void ocfs2_super_unlock(struct ocfs2_super *osb,
1956 int ex)
1957{
1958 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1959 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1960
c271c5c2
SM		 1961 if (!ocfs2_mount_local(osb))
1962 ocfs2_cluster_unlock(osb, lockres, level);
ccd979bd
MF		 1963}
1964
1965int ocfs2_rename_lock(struct ocfs2_super *osb)
1966{
1967 int status;
1968 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1969
1970 if (ocfs2_is_hard_readonly(osb))
1971 return -EROFS;
1972
c271c5c2
SM		 1973 if (ocfs2_mount_local(osb))
1974 return 0;
1975
ccd979bd
MF		 1976 status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0);
1977 if (status < 0)
1978 mlog_errno(status);
1979
1980 return status;
1981}
1982
1983void ocfs2_rename_unlock(struct ocfs2_super *osb)
1984{
1985 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1986
c271c5c2
SM		 1987 if (!ocfs2_mount_local(osb))
1988 ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE);
ccd979bd
MF		 1989}
1990
d680efe9
MF		 1991int ocfs2_dentry_lock(struct dentry *dentry, int ex)
1992{
1993 int ret;
1994 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1995 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
1996 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
1997
1998 BUG_ON(!dl);
1999
2000 if (ocfs2_is_hard_readonly(osb))
2001 return -EROFS;
2002
c271c5c2
SM		 2003 if (ocfs2_mount_local(osb))
2004 return 0;
2005
d680efe9
MF		 2006 ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2007 if (ret < 0)
2008 mlog_errno(ret);
2009
2010 return ret;
2011}
2012
2013void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2014{
2015 int level = ex ? LKM_EXMODE : LKM_PRMODE;
2016 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2017 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2018
c271c5c2
SM		 2019 if (!ocfs2_mount_local(osb))
2020 ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
d680efe9
MF		 2021}
2022
ccd979bd
MF		 2023/* Reference counting of the dlm debug structure. We want this because
2024 * open references on the debug inodes can live on after a mount, so
2025 * we can't rely on the ocfs2_super to always exist. */
2026static void ocfs2_dlm_debug_free(struct kref *kref)
2027{
2028 struct ocfs2_dlm_debug *dlm_debug;
2029
2030 dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2031
2032 kfree(dlm_debug);
2033}
2034
2035void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2036{
2037 if (dlm_debug)
2038 kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2039}
2040
2041static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2042{
2043 kref_get(&debug->d_refcnt);
2044}
2045
2046struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2047{
2048 struct ocfs2_dlm_debug *dlm_debug;
2049
2050 dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2051 if (!dlm_debug) {
2052 mlog_errno(-ENOMEM);
2053 goto out;
2054 }
2055
2056 kref_init(&dlm_debug->d_refcnt);
2057 INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2058 dlm_debug->d_locking_state = NULL;
2059out:
2060 return dlm_debug;
2061}
2062
2063/* Access to this is arbitrated for us via seq_file->sem. */
2064struct ocfs2_dlm_seq_priv {
2065 struct ocfs2_dlm_debug *p_dlm_debug;
2066 struct ocfs2_lock_res p_iter_res;
2067 struct ocfs2_lock_res p_tmp_res;
2068};
2069
2070static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2071 struct ocfs2_dlm_seq_priv *priv)
2072{
2073 struct ocfs2_lock_res *iter, *ret = NULL;
2074 struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2075
2076 assert_spin_locked(&ocfs2_dlm_tracking_lock);
2077
2078 list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2079 /* discover the head of the list */
2080 if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2081 mlog(0, "End of list found, %p\n", ret);
2082 break;
2083 }
2084
2085 /* We track our "dummy" iteration lockres' by a NULL
2086 * l_ops field. */
2087 if (iter->l_ops != NULL) {
2088 ret = iter;
2089 break;
2090 }
2091 }
2092
2093 return ret;
2094}
2095
2096static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2097{
2098 struct ocfs2_dlm_seq_priv *priv = m->private;
2099 struct ocfs2_lock_res *iter;
2100
2101 spin_lock(&ocfs2_dlm_tracking_lock);
2102 iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2103 if (iter) {
2104 /* Since lockres' have the lifetime of their container
2105 * (which can be inodes, ocfs2_supers, etc) we want to
2106 * copy this out to a temporary lockres while still
2107 * under the spinlock. Obviously after this we can't
2108 * trust any pointers on the copy returned, but that's
2109 * ok as the information we want isn't typically held
2110 * in them. */
2111 priv->p_tmp_res = *iter;
2112 iter = &priv->p_tmp_res;
2113 }
2114 spin_unlock(&ocfs2_dlm_tracking_lock);
2115
2116 return iter;
2117}
2118
2119static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2120{
2121}
2122
2123static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2124{
2125 struct ocfs2_dlm_seq_priv *priv = m->private;
2126 struct ocfs2_lock_res *iter = v;
2127 struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2128
2129 spin_lock(&ocfs2_dlm_tracking_lock);
2130 iter = ocfs2_dlm_next_res(iter, priv);
2131 list_del_init(&dummy->l_debug_list);
2132 if (iter) {
2133 list_add(&dummy->l_debug_list, &iter->l_debug_list);
2134 priv->p_tmp_res = *iter;
2135 iter = &priv->p_tmp_res;
2136 }
2137 spin_unlock(&ocfs2_dlm_tracking_lock);
2138
2139 return iter;
2140}
2141
2142/* So that debugfs.ocfs2 can determine which format is being used */
2143#define OCFS2_DLM_DEBUG_STR_VERSION 1
2144static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2145{
2146 int i;
2147 char *lvb;
2148 struct ocfs2_lock_res *lockres = v;
2149
2150 if (!lockres)
2151 return -EINVAL;
2152
d680efe9
MF		 2153 seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2154
2155 if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2156 seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2157 lockres->l_name,
2158 (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2159 else
2160 seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2161
2162 seq_printf(m, "%d\t"
ccd979bd
MF		 2163 "0x%lx\t"
2164 "0x%x\t"
2165 "0x%x\t"
2166 "%u\t"
2167 "%u\t"
2168 "%d\t"
2169 "%d\t",
ccd979bd
MF		 2170 lockres->l_level,
2171 lockres->l_flags,
2172 lockres->l_action,
2173 lockres->l_unlock_action,
2174 lockres->l_ro_holders,
2175 lockres->l_ex_holders,
2176 lockres->l_requested,
2177 lockres->l_blocking);
2178
2179 /* Dump the raw LVB */
2180 lvb = lockres->l_lksb.lvb;
2181 for(i = 0; i < DLM_LVB_LEN; i++)
2182 seq_printf(m, "0x%x\t", lvb[i]);
2183
2184 /* End the line */
2185 seq_printf(m, "\n");
2186 return 0;
2187}
2188
2189static struct seq_operations ocfs2_dlm_seq_ops = {
2190 .start = ocfs2_dlm_seq_start,
2191 .stop = ocfs2_dlm_seq_stop,
2192 .next = ocfs2_dlm_seq_next,
2193 .show = ocfs2_dlm_seq_show,
2194};
2195
2196static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2197{
2198 struct seq_file *seq = (struct seq_file *) file->private_data;
2199 struct ocfs2_dlm_seq_priv *priv = seq->private;
2200 struct ocfs2_lock_res *res = &priv->p_iter_res;
2201
2202 ocfs2_remove_lockres_tracking(res);
2203 ocfs2_put_dlm_debug(priv->p_dlm_debug);
2204 return seq_release_private(inode, file);
2205}
2206
2207static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2208{
2209 int ret;
2210 struct ocfs2_dlm_seq_priv *priv;
2211 struct seq_file *seq;
2212 struct ocfs2_super *osb;
2213
2214 priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2215 if (!priv) {
2216 ret = -ENOMEM;
2217 mlog_errno(ret);
2218 goto out;
2219 }
8e18e294 2220 osb = inode->i_private;
ccd979bd
MF		 2221 ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2222 priv->p_dlm_debug = osb->osb_dlm_debug;
2223 INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2224
2225 ret = seq_open(file, &ocfs2_dlm_seq_ops);
2226 if (ret) {
2227 kfree(priv);
2228 mlog_errno(ret);
2229 goto out;
2230 }
2231
2232 seq = (struct seq_file *) file->private_data;
2233 seq->private = priv;
2234
2235 ocfs2_add_lockres_tracking(&priv->p_iter_res,
2236 priv->p_dlm_debug);
2237
2238out:
2239 return ret;
2240}
2241
4b6f5d20 2242static const struct file_operations ocfs2_dlm_debug_fops = {
ccd979bd
MF		 2243 .open = ocfs2_dlm_debug_open,
2244 .release = ocfs2_dlm_debug_release,
2245 .read = seq_read,
2246 .llseek = seq_lseek,
2247};
2248
2249static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2250{
2251 int ret = 0;
2252 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2253
2254 dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2255 S_IFREG|S_IRUSR,
2256 osb->osb_debug_root,
2257 osb,
2258 &ocfs2_dlm_debug_fops);
2259 if (!dlm_debug->d_locking_state) {
2260 ret = -EINVAL;
2261 mlog(ML_ERROR,
2262 "Unable to create locking state debugfs file.\n");
2263 goto out;
2264 }
2265
2266 ocfs2_get_dlm_debug(dlm_debug);
2267out:
2268 return ret;
2269}
2270
2271static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2272{
2273 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2274
2275 if (dlm_debug) {
2276 debugfs_remove(dlm_debug->d_locking_state);
2277 ocfs2_put_dlm_debug(dlm_debug);
2278 }
2279}
2280
2281int ocfs2_dlm_init(struct ocfs2_super *osb)
2282{
c271c5c2 2283 int status = 0;
ccd979bd 2284 u32 dlm_key;
c271c5c2 2285 struct dlm_ctxt *dlm = NULL;
ccd979bd
MF		 2286
2287 mlog_entry_void();
2288
c271c5c2
SM		 2289 if (ocfs2_mount_local(osb))
2290 goto local;
2291
ccd979bd
MF		 2292 status = ocfs2_dlm_init_debug(osb);
2293 if (status < 0) {
2294 mlog_errno(status);
2295 goto bail;
2296 }
2297
2298 /* launch vote thread */
78427043 2299 osb->vote_task = kthread_run(ocfs2_vote_thread, osb, "ocfs2vote");
ccd979bd
MF		 2300 if (IS_ERR(osb->vote_task)) {
2301 status = PTR_ERR(osb->vote_task);
2302 osb->vote_task = NULL;
2303 mlog_errno(status);
2304 goto bail;
2305 }
2306
2307 /* used by the dlm code to make message headers unique, each
2308 * node in this domain must agree on this. */
2309 dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str));
2310
2311 /* for now, uuid == domain */
2312 dlm = dlm_register_domain(osb->uuid_str, dlm_key);
2313 if (IS_ERR(dlm)) {
2314 status = PTR_ERR(dlm);
2315 mlog_errno(status);
2316 goto bail;
2317 }
2318
c271c5c2
SM		 2319 dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb);
2320
2321local:
ccd979bd
MF		 2322 ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
2323 ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
2324
ccd979bd
MF		 2325 osb->dlm = dlm;
2326
2327 status = 0;
2328bail:
2329 if (status < 0) {
2330 ocfs2_dlm_shutdown_debug(osb);
2331 if (osb->vote_task)
2332 kthread_stop(osb->vote_task);
2333 }
2334
2335 mlog_exit(status);
2336 return status;
2337}
2338
2339void ocfs2_dlm_shutdown(struct ocfs2_super *osb)
2340{
2341 mlog_entry_void();
2342
2343 dlm_unregister_eviction_cb(&osb->osb_eviction_cb);
2344
2345 ocfs2_drop_osb_locks(osb);
2346
2347 if (osb->vote_task) {
2348 kthread_stop(osb->vote_task);
2349 osb->vote_task = NULL;
2350 }
2351
2352 ocfs2_lock_res_free(&osb->osb_super_lockres);
2353 ocfs2_lock_res_free(&osb->osb_rename_lockres);
2354
2355 dlm_unregister_domain(osb->dlm);
2356 osb->dlm = NULL;
2357
2358 ocfs2_dlm_shutdown_debug(osb);
2359
2360 mlog_exit_void();
2361}
2362
2a45f2d1 2363static void ocfs2_unlock_ast(void *opaque, enum dlm_status status)
ccd979bd
MF		 2364{
2365 struct ocfs2_lock_res *lockres = opaque;
2366 unsigned long flags;
2367
2368 mlog_entry_void();
2369
2370 mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
2371 lockres->l_unlock_action);
2372
2373 spin_lock_irqsave(&lockres->l_lock, flags);
2374 /* We tried to cancel a convert request, but it was already
2375 * granted. All we want to do here is clear our unlock
2376 * state. The wake_up call done at the bottom is redundant
2377 * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't
2378 * hurt anything anyway */
2379 if (status == DLM_CANCELGRANT &&
2380 lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2381 mlog(0, "Got cancelgrant for %s\n", lockres->l_name);
2382
2383 /* We don't clear the busy flag in this case as it
2384 * should have been cleared by the ast which the dlm
2385 * has called. */
2386 goto complete_unlock;
2387 }
2388
2389 if (status != DLM_NORMAL) {
2390 mlog(ML_ERROR, "Dlm passes status %d for lock %s, "
2391 "unlock_action %d\n", status, lockres->l_name,
2392 lockres->l_unlock_action);
2393 spin_unlock_irqrestore(&lockres->l_lock, flags);
2394 return;
2395 }
2396
2397 switch(lockres->l_unlock_action) {
2398 case OCFS2_UNLOCK_CANCEL_CONVERT:
2399 mlog(0, "Cancel convert success for %s\n", lockres->l_name);
2400 lockres->l_action = OCFS2_AST_INVALID;
2401 break;
2402 case OCFS2_UNLOCK_DROP_LOCK:
2403 lockres->l_level = LKM_IVMODE;
2404 break;
2405 default:
2406 BUG();
2407 }
2408
2409 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
2410complete_unlock:
2411 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
2412 spin_unlock_irqrestore(&lockres->l_lock, flags);
2413
2414 wake_up(&lockres->l_event);
2415
2416 mlog_exit_void();
2417}
2418
ccd979bd 2419static int ocfs2_drop_lock(struct ocfs2_super *osb,
0d5dc6c2 2420 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 2421{
2422 enum dlm_status status;
2423 unsigned long flags;
b80fc012 2424 int lkm_flags = 0;
ccd979bd
MF		 2425
2426 /* We didn't get anywhere near actually using this lockres. */
2427 if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
2428 goto out;
2429
b80fc012
MF		 2430 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
2431 lkm_flags |= LKM_VALBLK;
2432
ccd979bd
MF		 2433 spin_lock_irqsave(&lockres->l_lock, flags);
2434
2435 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
2436 "lockres %s, flags 0x%lx\n",
2437 lockres->l_name, lockres->l_flags);
2438
2439 while (lockres->l_flags & OCFS2_LOCK_BUSY) {
2440 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2441 "%u, unlock_action = %u\n",
2442 lockres->l_name, lockres->l_flags, lockres->l_action,
2443 lockres->l_unlock_action);
2444
2445 spin_unlock_irqrestore(&lockres->l_lock, flags);
2446
2447 /* XXX: Today we just wait on any busy
2448 * locks... Perhaps we need to cancel converts in the
2449 * future? */
2450 ocfs2_wait_on_busy_lock(lockres);
2451
2452 spin_lock_irqsave(&lockres->l_lock, flags);
2453 }
2454
0d5dc6c2
MF		 2455 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2456 if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
2457 lockres->l_level == LKM_EXMODE &&
2458 !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2459 lockres->l_ops->set_lvb(lockres);
2460 }
ccd979bd
MF		 2461
2462 if (lockres->l_flags & OCFS2_LOCK_BUSY)
2463 mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
2464 lockres->l_name);
2465 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2466 mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
2467
2468 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
2469 spin_unlock_irqrestore(&lockres->l_lock, flags);
2470 goto out;
2471 }
2472
2473 lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
2474
2475 /* make sure we never get here while waiting for an ast to
2476 * fire. */
2477 BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
2478
2479 /* is this necessary? */
2480 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2481 lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
2482 spin_unlock_irqrestore(&lockres->l_lock, flags);
2483
2484 mlog(0, "lock %s\n", lockres->l_name);
2485
b80fc012 2486 status = dlmunlock(osb->dlm, &lockres->l_lksb, lkm_flags,
2a45f2d1 2487 ocfs2_unlock_ast, lockres);
ccd979bd
MF		 2488 if (status != DLM_NORMAL) {
2489 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2490 mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
2491 dlm_print_one_lock(lockres->l_lksb.lockid);
2492 BUG();
2493 }
2494 mlog(0, "lock %s, successfull return from dlmunlock\n",
2495 lockres->l_name);
2496
2497 ocfs2_wait_on_busy_lock(lockres);
2498out:
2499 mlog_exit(0);
2500 return 0;
2501}
2502
2503/* Mark the lockres as being dropped. It will no longer be
2504 * queued if blocking, but we still may have to wait on it
2505 * being dequeued from the vote thread before we can consider
2506 * it safe to drop.
2507 *
2508 * You can *not* attempt to call cluster_lock on this lockres anymore. */
2509void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
2510{
2511 int status;
2512 struct ocfs2_mask_waiter mw;
2513 unsigned long flags;
2514
2515 ocfs2_init_mask_waiter(&mw);
2516
2517 spin_lock_irqsave(&lockres->l_lock, flags);
2518 lockres->l_flags |= OCFS2_LOCK_FREEING;
2519 while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
2520 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
2521 spin_unlock_irqrestore(&lockres->l_lock, flags);
2522
2523 mlog(0, "Waiting on lockres %s\n", lockres->l_name);
2524
2525 status = ocfs2_wait_for_mask(&mw);
2526 if (status)
2527 mlog_errno(status);
2528
2529 spin_lock_irqsave(&lockres->l_lock, flags);
2530 }
2531 spin_unlock_irqrestore(&lockres->l_lock, flags);
2532}
2533
d680efe9
MF		 2534void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
2535 struct ocfs2_lock_res *lockres)
ccd979bd 2536{
d680efe9 2537 int ret;
ccd979bd 2538
d680efe9 2539 ocfs2_mark_lockres_freeing(lockres);
0d5dc6c2 2540 ret = ocfs2_drop_lock(osb, lockres);
d680efe9
MF		 2541 if (ret)
2542 mlog_errno(ret);
2543}
ccd979bd 2544
d680efe9
MF		 2545static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
2546{
2547 ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
2548 ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
ccd979bd
MF		 2549}
2550
ccd979bd
MF		 2551int ocfs2_drop_inode_locks(struct inode *inode)
2552{
2553 int status, err;
ccd979bd
MF		 2554
2555 mlog_entry_void();
2556
2557 /* No need to call ocfs2_mark_lockres_freeing here -
2558 * ocfs2_clear_inode has done it for us. */
2559
2560 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
50008630 2561 &OCFS2_I(inode)->ip_open_lockres);
ccd979bd
MF		 2562 if (err < 0)
2563 mlog_errno(err);
2564
2565 status = err;
2566
50008630
TY		 2567 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2568 &OCFS2_I(inode)->ip_data_lockres);
2569 if (err < 0)
2570 mlog_errno(err);
2571 if (err < 0 && !status)
2572 status = err;
2573
ccd979bd 2574 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
0d5dc6c2 2575 &OCFS2_I(inode)->ip_meta_lockres);
ccd979bd
MF		 2576 if (err < 0)
2577 mlog_errno(err);
2578 if (err < 0 && !status)
2579 status = err;
2580
2581 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
0d5dc6c2 2582 &OCFS2_I(inode)->ip_rw_lockres);
ccd979bd
MF		 2583 if (err < 0)
2584 mlog_errno(err);
2585 if (err < 0 && !status)
2586 status = err;
2587
2588 mlog_exit(status);
2589 return status;
2590}
2591
2592static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
2593 int new_level)
2594{
2595 assert_spin_locked(&lockres->l_lock);
2596
2597 BUG_ON(lockres->l_blocking <= LKM_NLMODE);
2598
2599 if (lockres->l_level <= new_level) {
2600 mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n",
2601 lockres->l_level, new_level);
2602 BUG();
2603 }
2604
2605 mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
2606 lockres->l_name, new_level, lockres->l_blocking);
2607
2608 lockres->l_action = OCFS2_AST_DOWNCONVERT;
2609 lockres->l_requested = new_level;
2610 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2611}
2612
2613static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
2614 struct ocfs2_lock_res *lockres,
2615 int new_level,
2616 int lvb)
2617{
2618 int ret, dlm_flags = LKM_CONVERT;
2619 enum dlm_status status;
2620
2621 mlog_entry_void();
2622
2623 if (lvb)
2624 dlm_flags |= LKM_VALBLK;
2625
2626 status = dlmlock(osb->dlm,
2627 new_level,
2628 &lockres->l_lksb,
2629 dlm_flags,
2630 lockres->l_name,
f0681062 2631 OCFS2_LOCK_ID_MAX_LEN - 1,
e92d57df 2632 ocfs2_locking_ast,
ccd979bd 2633 lockres,
aa2623ad 2634 ocfs2_blocking_ast);
ccd979bd
MF		 2635 if (status != DLM_NORMAL) {
2636 ocfs2_log_dlm_error("dlmlock", status, lockres);
2637 ret = -EINVAL;
2638 ocfs2_recover_from_dlm_error(lockres, 1);
2639 goto bail;
2640 }
2641
2642 ret = 0;
2643bail:
2644 mlog_exit(ret);
2645 return ret;
2646}
2647
2648/* returns 1 when the caller should unlock and call dlmunlock */
2649static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
2650 struct ocfs2_lock_res *lockres)
2651{
2652 assert_spin_locked(&lockres->l_lock);
2653
2654 mlog_entry_void();
2655 mlog(0, "lock %s\n", lockres->l_name);
2656
2657 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2658 /* If we're already trying to cancel a lock conversion
2659 * then just drop the spinlock and allow the caller to
2660 * requeue this lock. */
2661
2662 mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
2663 return 0;
2664 }
2665
2666 /* were we in a convert when we got the bast fire? */
2667 BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
2668 lockres->l_action != OCFS2_AST_DOWNCONVERT);
2669 /* set things up for the unlockast to know to just
2670 * clear out the ast_action and unset busy, etc. */
2671 lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
2672
2673 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
2674 "lock %s, invalid flags: 0x%lx\n",
2675 lockres->l_name, lockres->l_flags);
2676
2677 return 1;
2678}
2679
2680static int ocfs2_cancel_convert(struct ocfs2_super *osb,
2681 struct ocfs2_lock_res *lockres)
2682{
2683 int ret;
2684 enum dlm_status status;
2685
2686 mlog_entry_void();
2687 mlog(0, "lock %s\n", lockres->l_name);
2688
2689 ret = 0;
2690 status = dlmunlock(osb->dlm,
2691 &lockres->l_lksb,
2692 LKM_CANCEL,
2a45f2d1 2693 ocfs2_unlock_ast,
ccd979bd
MF		 2694 lockres);
2695 if (status != DLM_NORMAL) {
2696 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2697 ret = -EINVAL;
2698 ocfs2_recover_from_dlm_error(lockres, 0);
2699 }
2700
2701 mlog(0, "lock %s return from dlmunlock\n", lockres->l_name);
2702
ccd979bd
MF		 2703 mlog_exit(ret);
2704 return ret;
2705}
2706
b5e500e2
MF		 2707static int ocfs2_unblock_lock(struct ocfs2_super *osb,
2708 struct ocfs2_lock_res *lockres,
2709 struct ocfs2_unblock_ctl *ctl)
ccd979bd
MF		 2710{
2711 unsigned long flags;
2712 int blocking;
2713 int new_level;
2714 int ret = 0;
5ef0d4ea 2715 int set_lvb = 0;
ccd979bd
MF		 2716
2717 mlog_entry_void();
2718
2719 spin_lock_irqsave(&lockres->l_lock, flags);
2720
2721 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
2722
2723recheck:
2724 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
d680efe9 2725 ctl->requeue = 1;
ccd979bd
MF		 2726 ret = ocfs2_prepare_cancel_convert(osb, lockres);
2727 spin_unlock_irqrestore(&lockres->l_lock, flags);
2728 if (ret) {
2729 ret = ocfs2_cancel_convert(osb, lockres);
2730 if (ret < 0)
2731 mlog_errno(ret);
2732 }
2733 goto leave;
2734 }
2735
2736 /* if we're blocking an exclusive and we have *any* holders,
2737 * then requeue. */
2738 if ((lockres->l_blocking == LKM_EXMODE)
f7fbfdd1
MF		 2739 && (lockres->l_ex_holders || lockres->l_ro_holders))
2740 goto leave_requeue;
ccd979bd
MF		 2741
2742 /* If it's a PR we're blocking, then only
2743 * requeue if we've got any EX holders */
2744 if (lockres->l_blocking == LKM_PRMODE &&
f7fbfdd1
MF		 2745 lockres->l_ex_holders)
2746 goto leave_requeue;
2747
2748 /*
2749 * Can we get a lock in this state if the holder counts are
2750 * zero? The meta data unblock code used to check this.
2751 */
2752 if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
2753 && (lockres->l_flags & OCFS2_LOCK_REFRESHING))
2754 goto leave_requeue;
ccd979bd 2755
16d5b956
MF		 2756 new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
2757
2758 if (lockres->l_ops->check_downconvert
2759 && !lockres->l_ops->check_downconvert(lockres, new_level))
2760 goto leave_requeue;
2761
ccd979bd
MF		 2762 /* If we get here, then we know that there are no more
2763 * incompatible holders (and anyone asking for an incompatible
2764 * lock is blocked). We can now downconvert the lock */
cc567d89 2765 if (!lockres->l_ops->downconvert_worker)
ccd979bd
MF		 2766 goto downconvert;
2767
2768 /* Some lockres types want to do a bit of work before
2769 * downconverting a lock. Allow that here. The worker function
2770 * may sleep, so we save off a copy of what we're blocking as
2771 * it may change while we're not holding the spin lock. */
2772 blocking = lockres->l_blocking;
2773 spin_unlock_irqrestore(&lockres->l_lock, flags);
2774
cc567d89 2775 ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
d680efe9
MF		 2776
2777 if (ctl->unblock_action == UNBLOCK_STOP_POST)
2778 goto leave;
ccd979bd
MF		 2779
2780 spin_lock_irqsave(&lockres->l_lock, flags);
2781 if (blocking != lockres->l_blocking) {
2782 /* If this changed underneath us, then we can't drop
2783 * it just yet. */
2784 goto recheck;
2785 }
2786
2787downconvert:
d680efe9 2788 ctl->requeue = 0;
ccd979bd 2789
5ef0d4ea
MF		 2790 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2791 if (lockres->l_level == LKM_EXMODE)
2792 set_lvb = 1;
2793
2794 /*
2795 * We only set the lvb if the lock has been fully
2796 * refreshed - otherwise we risk setting stale
2797 * data. Otherwise, there's no need to actually clear
2798 * out the lvb here as it's value is still valid.
2799 */
2800 if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2801 lockres->l_ops->set_lvb(lockres);
2802 }
2803
ccd979bd
MF		 2804 ocfs2_prepare_downconvert(lockres, new_level);
2805 spin_unlock_irqrestore(&lockres->l_lock, flags);
5ef0d4ea 2806 ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb);
ccd979bd
MF		 2807leave:
2808 mlog_exit(ret);
2809 return ret;
f7fbfdd1
MF		 2810
2811leave_requeue:
2812 spin_unlock_irqrestore(&lockres->l_lock, flags);
2813 ctl->requeue = 1;
2814
2815 mlog_exit(0);
2816 return 0;
ccd979bd
MF		 2817}
2818
d680efe9
MF		 2819static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
2820 int blocking)
ccd979bd
MF		 2821{
2822 struct inode *inode;
2823 struct address_space *mapping;
2824
ccd979bd
MF		 2825 inode = ocfs2_lock_res_inode(lockres);
2826 mapping = inode->i_mapping;
2827
7f4a2a97
MF		 2828 /*
2829 * We need this before the filemap_fdatawrite() so that it can
2830 * transfer the dirty bit from the PTE to the
2831 * page. Unfortunately this means that even for EX->PR
2832 * downconverts, we'll lose our mappings and have to build
2833 * them up again.
2834 */
2835 unmap_mapping_range(mapping, 0, 0, 0);
2836
ccd979bd 2837 if (filemap_fdatawrite(mapping)) {
b0697053
MF		 2838 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
2839 (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 2840 }
2841 sync_mapping_buffers(mapping);
2842 if (blocking == LKM_EXMODE) {
2843 truncate_inode_pages(mapping, 0);
ccd979bd
MF		 2844 } else {
2845 /* We only need to wait on the I/O if we're not also
2846 * truncating pages because truncate_inode_pages waits
2847 * for us above. We don't truncate pages if we're
2848 * blocking anything < EXMODE because we want to keep
2849 * them around in that case. */
2850 filemap_fdatawait(mapping);
2851 }
2852
d680efe9 2853 return UNBLOCK_CONTINUE;
ccd979bd
MF		 2854}
2855
810d5aeb
MF		 2856static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
2857 int new_level)
2858{
2859 struct inode *inode = ocfs2_lock_res_inode(lockres);
2860 int checkpointed = ocfs2_inode_fully_checkpointed(inode);
2861
2862 BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE);
2863 BUG_ON(lockres->l_level != LKM_EXMODE && !checkpointed);
2864
2865 if (checkpointed)
2866 return 1;
2867
2868 ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb));
2869 return 0;
2870}
2871
2872static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
2873{
2874 struct inode *inode = ocfs2_lock_res_inode(lockres);
2875
2876 __ocfs2_stuff_meta_lvb(inode);
2877}
2878
d680efe9
MF		 2879/*
2880 * Does the final reference drop on our dentry lock. Right now this
2881 * happens in the vote thread, but we could choose to simplify the
2882 * dlmglue API and push these off to the ocfs2_wq in the future.
2883 */
2884static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
2885 struct ocfs2_lock_res *lockres)
2886{
2887 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
2888 ocfs2_dentry_lock_put(osb, dl);
2889}
2890
2891/*
2892 * d_delete() matching dentries before the lock downconvert.
2893 *
2894 * At this point, any process waiting to destroy the
2895 * dentry_lock due to last ref count is stopped by the
2896 * OCFS2_LOCK_QUEUED flag.
2897 *
2898 * We have two potential problems
2899 *
2900 * 1) If we do the last reference drop on our dentry_lock (via dput)
2901 * we'll wind up in ocfs2_release_dentry_lock(), waiting on
2902 * the downconvert to finish. Instead we take an elevated
2903 * reference and push the drop until after we've completed our
2904 * unblock processing.
2905 *
2906 * 2) There might be another process with a final reference,
2907 * waiting on us to finish processing. If this is the case, we
2908 * detect it and exit out - there's no more dentries anyway.
2909 */
2910static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
2911 int blocking)
2912{
2913 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
2914 struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
2915 struct dentry *dentry;
2916 unsigned long flags;
2917 int extra_ref = 0;
2918
2919 /*
2920 * This node is blocking another node from getting a read
2921 * lock. This happens when we've renamed within a
2922 * directory. We've forced the other nodes to d_delete(), but
2923 * we never actually dropped our lock because it's still
2924 * valid. The downconvert code will retain a PR for this node,
2925 * so there's no further work to do.
2926 */
2927 if (blocking == LKM_PRMODE)
2928 return UNBLOCK_CONTINUE;
2929
2930 /*
2931 * Mark this inode as potentially orphaned. The code in
2932 * ocfs2_delete_inode() will figure out whether it actually
2933 * needs to be freed or not.
2934 */
2935 spin_lock(&oi->ip_lock);
2936 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2937 spin_unlock(&oi->ip_lock);
2938
2939 /*
2940 * Yuck. We need to make sure however that the check of
2941 * OCFS2_LOCK_FREEING and the extra reference are atomic with
2942 * respect to a reference decrement or the setting of that
2943 * flag.
2944 */
2945 spin_lock_irqsave(&lockres->l_lock, flags);
2946 spin_lock(&dentry_attach_lock);
2947 if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
2948 && dl->dl_count) {
2949 dl->dl_count++;
2950 extra_ref = 1;
2951 }
2952 spin_unlock(&dentry_attach_lock);
2953 spin_unlock_irqrestore(&lockres->l_lock, flags);
2954
2955 mlog(0, "extra_ref = %d\n", extra_ref);
2956
2957 /*
2958 * We have a process waiting on us in ocfs2_dentry_iput(),
2959 * which means we can't have any more outstanding
2960 * aliases. There's no need to do any more work.
2961 */
2962 if (!extra_ref)
2963 return UNBLOCK_CONTINUE;
2964
2965 spin_lock(&dentry_attach_lock);
2966 while (1) {
2967 dentry = ocfs2_find_local_alias(dl->dl_inode,
2968 dl->dl_parent_blkno, 1);
2969 if (!dentry)
2970 break;
2971 spin_unlock(&dentry_attach_lock);
2972
2973 mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
2974 dentry->d_name.name);
2975
2976 /*
2977 * The following dcache calls may do an
2978 * iput(). Normally we don't want that from the
2979 * downconverting thread, but in this case it's ok
2980 * because the requesting node already has an
2981 * exclusive lock on the inode, so it can't be queued
2982 * for a downconvert.
2983 */
2984 d_delete(dentry);
2985 dput(dentry);
2986
2987 spin_lock(&dentry_attach_lock);
2988 }
2989 spin_unlock(&dentry_attach_lock);
2990
2991 /*
2992 * If we are the last holder of this dentry lock, there is no
2993 * reason to downconvert so skip straight to the unlock.
2994 */
2995 if (dl->dl_count == 1)
2996 return UNBLOCK_STOP_POST;
2997
2998 return UNBLOCK_CONTINUE_POST;
2999}
3000
ccd979bd
MF		 3001void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3002 struct ocfs2_lock_res *lockres)
3003{
3004 int status;
d680efe9 3005 struct ocfs2_unblock_ctl ctl = {0, 0,};
ccd979bd
MF		 3006 unsigned long flags;
3007
3008 /* Our reference to the lockres in this function can be
3009 * considered valid until we remove the OCFS2_LOCK_QUEUED
3010 * flag. */
3011
3012 mlog_entry_void();
3013
3014 BUG_ON(!lockres);
3015 BUG_ON(!lockres->l_ops);
ccd979bd
MF		 3016
3017 mlog(0, "lockres %s blocked.\n", lockres->l_name);
3018
3019 /* Detect whether a lock has been marked as going away while
3020 * the vote thread was processing other things. A lock can
3021 * still be marked with OCFS2_LOCK_FREEING after this check,
3022 * but short circuiting here will still save us some
3023 * performance. */
3024 spin_lock_irqsave(&lockres->l_lock, flags);
3025 if (lockres->l_flags & OCFS2_LOCK_FREEING)
3026 goto unqueue;
3027 spin_unlock_irqrestore(&lockres->l_lock, flags);
3028
b5e500e2 3029 status = ocfs2_unblock_lock(osb, lockres, &ctl);
ccd979bd
MF		 3030 if (status < 0)
3031 mlog_errno(status);
3032
3033 spin_lock_irqsave(&lockres->l_lock, flags);
3034unqueue:
d680efe9 3035 if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
ccd979bd
MF		 3036 lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3037 } else
3038 ocfs2_schedule_blocked_lock(osb, lockres);
3039
3040 mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
d680efe9 3041 ctl.requeue ? "yes" : "no");
ccd979bd
MF		 3042 spin_unlock_irqrestore(&lockres->l_lock, flags);
3043
d680efe9
MF		 3044 if (ctl.unblock_action != UNBLOCK_CONTINUE
3045 && lockres->l_ops->post_unlock)
3046 lockres->l_ops->post_unlock(osb, lockres);
3047
ccd979bd
MF		 3048 mlog_exit_void();
3049}
3050
3051static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3052 struct ocfs2_lock_res *lockres)
3053{
3054 mlog_entry_void();
3055
3056 assert_spin_locked(&lockres->l_lock);
3057
3058 if (lockres->l_flags & OCFS2_LOCK_FREEING) {
3059 /* Do not schedule a lock for downconvert when it's on
3060 * the way to destruction - any nodes wanting access
3061 * to the resource will get it soon. */
3062 mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
3063 lockres->l_name, lockres->l_flags);
3064 return;
3065 }
3066
3067 lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
3068
3069 spin_lock(&osb->vote_task_lock);
3070 if (list_empty(&lockres->l_blocked_list)) {
3071 list_add_tail(&lockres->l_blocked_list,
3072 &osb->blocked_lock_list);
3073 osb->blocked_lock_count++;
3074 }
3075 spin_unlock(&osb->vote_task_lock);
3076
3077 mlog_exit_void();
3078}
3079
3080/* This aids in debugging situations where a bad LVB might be involved. */
3081void ocfs2_dump_meta_lvb_info(u64 level,
3082 const char *function,
3083 unsigned int line,
3084 struct ocfs2_lock_res *lockres)
3085{
3086 struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
3087
3088 mlog(level, "LVB information for %s (called from %s:%u):\n",
3089 lockres->l_name, function, line);
f9e2d82e
MF		 3090 mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
3091 lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
3092 be32_to_cpu(lvb->lvb_igeneration));
b0697053
MF		 3093 mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
3094 (unsigned long long)be64_to_cpu(lvb->lvb_isize),
3095 be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
3096 be16_to_cpu(lvb->lvb_imode));
3097 mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
ca4d147e 3098 "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
b0697053
MF		 3099 (long long)be64_to_cpu(lvb->lvb_iatime_packed),
3100 (long long)be64_to_cpu(lvb->lvb_ictime_packed),
ca4d147e
HP		 3101 (long long)be64_to_cpu(lvb->lvb_imtime_packed),
3102 be32_to_cpu(lvb->lvb_iattr));
ccd979bd 3103}
kernel source for telekom puls (alcatel/mediatek)