projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: Fix extent lookup to return true size of holes
[deliverable/linux.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
1498 inode->i_blocks =
1499 ocfs2_align_bytes_to_sectors(i_size_read(inode));
1500
1501 inode->i_uid = be32_to_cpu(lvb->lvb_iuid);
1502 inode->i_gid = be32_to_cpu(lvb->lvb_igid);
1503 inode->i_mode = be16_to_cpu(lvb->lvb_imode);
1504 inode->i_nlink = be16_to_cpu(lvb->lvb_inlink);
1505 ocfs2_unpack_timespec(&inode->i_atime,
1506 be64_to_cpu(lvb->lvb_iatime_packed));
1507 ocfs2_unpack_timespec(&inode->i_mtime,
1508 be64_to_cpu(lvb->lvb_imtime_packed));
1509 ocfs2_unpack_timespec(&inode->i_ctime,
1510 be64_to_cpu(lvb->lvb_ictime_packed));
1511 spin_unlock(&oi->ip_lock);
1512
1513 mlog_exit_void();
1514}
1515
f9e2d82e
MF		 1516static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
1517 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 1518{
1519 struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1520
f9e2d82e
MF		 1521 if (lvb->lvb_version == OCFS2_LVB_VERSION
1522 && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
ccd979bd
MF		 1523 return 1;
1524 return 0;
1525}
1526
1527/* Determine whether a lock resource needs to be refreshed, and
1528 * arbitrate who gets to refresh it.
1529 *
1530 * 0 means no refresh needed.
1531 *
1532 * > 0 means you need to refresh this and you MUST call
1533 * ocfs2_complete_lock_res_refresh afterwards. */
1534static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
1535{
1536 unsigned long flags;
1537 int status = 0;
1538
1539 mlog_entry_void();
1540
1541refresh_check:
1542 spin_lock_irqsave(&lockres->l_lock, flags);
1543 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
1544 spin_unlock_irqrestore(&lockres->l_lock, flags);
1545 goto bail;
1546 }
1547
1548 if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
1549 spin_unlock_irqrestore(&lockres->l_lock, flags);
1550
1551 ocfs2_wait_on_refreshing_lock(lockres);
1552 goto refresh_check;
1553 }
1554
1555 /* Ok, I'll be the one to refresh this lock. */
1556 lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
1557 spin_unlock_irqrestore(&lockres->l_lock, flags);
1558
1559 status = 1;
1560bail:
1561 mlog_exit(status);
1562 return status;
1563}
1564
1565/* If status is non zero, I'll mark it as not being in refresh
1566 * anymroe, but i won't clear the needs refresh flag. */
1567static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
1568 int status)
1569{
1570 unsigned long flags;
1571 mlog_entry_void();
1572
1573 spin_lock_irqsave(&lockres->l_lock, flags);
1574 lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
1575 if (!status)
1576 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
1577 spin_unlock_irqrestore(&lockres->l_lock, flags);
1578
1579 wake_up(&lockres->l_event);
1580
1581 mlog_exit_void();
1582}
1583
1584/* may or may not return a bh if it went to disk. */
1585static int ocfs2_meta_lock_update(struct inode *inode,
1586 struct buffer_head **bh)
1587{
1588 int status = 0;
1589 struct ocfs2_inode_info *oi = OCFS2_I(inode);
be9e986b 1590 struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
ccd979bd 1591 struct ocfs2_dinode *fe;
c271c5c2 1592 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1593
1594 mlog_entry_void();
1595
be9e986b
MF		 1596 if (ocfs2_mount_local(osb))
1597 goto bail;
1598
ccd979bd
MF		 1599 spin_lock(&oi->ip_lock);
1600 if (oi->ip_flags & OCFS2_INODE_DELETED) {
b0697053 1601 mlog(0, "Orphaned inode %llu was deleted while we "
ccd979bd 1602 "were waiting on a lock. ip_flags = 0x%x\n",
b0697053 1603 (unsigned long long)oi->ip_blkno, oi->ip_flags);
ccd979bd
MF		 1604 spin_unlock(&oi->ip_lock);
1605 status = -ENOENT;
1606 goto bail;
1607 }
1608 spin_unlock(&oi->ip_lock);
1609
be9e986b
MF		 1610 if (!ocfs2_should_refresh_lock_res(lockres))
1611 goto bail;
ccd979bd
MF		 1612
1613 /* This will discard any caching information we might have had
1614 * for the inode metadata. */
1615 ocfs2_metadata_cache_purge(inode);
1616
be9e986b 1617 if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
b0697053
MF		 1618 mlog(0, "Trusting LVB on inode %llu\n",
1619 (unsigned long long)oi->ip_blkno);
ccd979bd
MF		 1620 ocfs2_refresh_inode_from_lvb(inode);
1621 } else {
1622 /* Boo, we have to go to disk. */
1623 /* read bh, cast, ocfs2_refresh_inode */
1624 status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno,
1625 bh, OCFS2_BH_CACHED, inode);
1626 if (status < 0) {
1627 mlog_errno(status);
1628 goto bail_refresh;
1629 }
1630 fe = (struct ocfs2_dinode *) (*bh)->b_data;
1631
1632 /* This is a good chance to make sure we're not
1633 * locking an invalid object.
1634 *
1635 * We bug on a stale inode here because we checked
1636 * above whether it was wiped from disk. The wiping
1637 * node provides a guarantee that we receive that
1638 * message and can mark the inode before dropping any
1639 * locks associated with it. */
1640 if (!OCFS2_IS_VALID_DINODE(fe)) {
1641 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
1642 status = -EIO;
1643 goto bail_refresh;
1644 }
1645 mlog_bug_on_msg(inode->i_generation !=
1646 le32_to_cpu(fe->i_generation),
b0697053 1647 "Invalid dinode %llu disk generation: %u "
ccd979bd 1648 "inode->i_generation: %u\n",
b0697053
MF		 1649 (unsigned long long)oi->ip_blkno,
1650 le32_to_cpu(fe->i_generation),
ccd979bd
MF		 1651 inode->i_generation);
1652 mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
1653 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
b0697053
MF		 1654 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
1655 (unsigned long long)oi->ip_blkno,
1656 (unsigned long long)le64_to_cpu(fe->i_dtime),
ccd979bd
MF		 1657 le32_to_cpu(fe->i_flags));
1658
1659 ocfs2_refresh_inode(inode, fe);
1660 }
1661
1662 status = 0;
1663bail_refresh:
be9e986b 1664 ocfs2_complete_lock_res_refresh(lockres, status);
ccd979bd
MF		 1665bail:
1666 mlog_exit(status);
1667 return status;
1668}
1669
1670static int ocfs2_assign_bh(struct inode *inode,
1671 struct buffer_head **ret_bh,
1672 struct buffer_head *passed_bh)
1673{
1674 int status;
1675
1676 if (passed_bh) {
1677 /* Ok, the update went to disk for us, use the
1678 * returned bh. */
1679 *ret_bh = passed_bh;
1680 get_bh(*ret_bh);
1681
1682 return 0;
1683 }
1684
1685 status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1686 OCFS2_I(inode)->ip_blkno,
1687 ret_bh,
1688 OCFS2_BH_CACHED,
1689 inode);
1690 if (status < 0)
1691 mlog_errno(status);
1692
1693 return status;
1694}
1695
1696/*
1697 * returns < 0 error if the callback will never be called, otherwise
1698 * the result of the lock will be communicated via the callback.
1699 */
1700int ocfs2_meta_lock_full(struct inode *inode,
ccd979bd
MF		 1701 struct buffer_head **ret_bh,
1702 int ex,
1703 int arg_flags)
1704{
1705 int status, level, dlm_flags, acquired;
c271c5c2 1706 struct ocfs2_lock_res *lockres = NULL;
ccd979bd
MF		 1707 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1708 struct buffer_head *local_bh = NULL;
1709
1710 BUG_ON(!inode);
1711
1712 mlog_entry_void();
1713
b0697053
MF		 1714 mlog(0, "inode %llu, take %s META lock\n",
1715 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1716 ex ? "EXMODE" : "PRMODE");
1717
1718 status = 0;
1719 acquired = 0;
1720 /* We'll allow faking a readonly metadata lock for
1721 * rodevices. */
1722 if (ocfs2_is_hard_readonly(osb)) {
1723 if (ex)
1724 status = -EROFS;
1725 goto bail;
1726 }
1727
c271c5c2
SM		 1728 if (ocfs2_mount_local(osb))
1729 goto local;
1730
ccd979bd
MF		 1731 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1732 wait_event(osb->recovery_event,
1733 ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1734
ccd979bd
MF		 1735 lockres = &OCFS2_I(inode)->ip_meta_lockres;
1736 level = ex ? LKM_EXMODE : LKM_PRMODE;
1737 dlm_flags = 0;
1738 if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
1739 dlm_flags |= LKM_NOQUEUE;
1740
1741 status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
1742 if (status < 0) {
1743 if (status != -EAGAIN && status != -EIOCBRETRY)
1744 mlog_errno(status);
1745 goto bail;
1746 }
1747
1748 /* Notify the error cleanup path to drop the cluster lock. */
1749 acquired = 1;
1750
1751 /* We wait twice because a node may have died while we were in
1752 * the lower dlm layers. The second time though, we've
1753 * committed to owning this lock so we don't allow signals to
1754 * abort the operation. */
1755 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1756 wait_event(osb->recovery_event,
1757 ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1758
c271c5c2 1759local:
24c19ef4
MF		 1760 /*
1761 * We only see this flag if we're being called from
1762 * ocfs2_read_locked_inode(). It means we're locking an inode
1763 * which hasn't been populated yet, so clear the refresh flag
1764 * and let the caller handle it.
1765 */
1766 if (inode->i_state & I_NEW) {
1767 status = 0;
c271c5c2
SM		 1768 if (lockres)
1769 ocfs2_complete_lock_res_refresh(lockres, 0);
24c19ef4
MF		 1770 goto bail;
1771 }
1772
ccd979bd
MF		 1773 /* This is fun. The caller may want a bh back, or it may
1774 * not. ocfs2_meta_lock_update definitely wants one in, but
1775 * may or may not read one, depending on what's in the
1776 * LVB. The result of all of this is that we've *only* gone to
1777 * disk if we have to, so the complexity is worthwhile. */
1778 status = ocfs2_meta_lock_update(inode, &local_bh);
1779 if (status < 0) {
1780 if (status != -ENOENT)
1781 mlog_errno(status);
1782 goto bail;
1783 }
1784
1785 if (ret_bh) {
1786 status = ocfs2_assign_bh(inode, ret_bh, local_bh);
1787 if (status < 0) {
1788 mlog_errno(status);
1789 goto bail;
1790 }
1791 }
1792
ccd979bd
MF		 1793bail:
1794 if (status < 0) {
1795 if (ret_bh && (*ret_bh)) {
1796 brelse(*ret_bh);
1797 *ret_bh = NULL;
1798 }
1799 if (acquired)
1800 ocfs2_meta_unlock(inode, ex);
1801 }
1802
1803 if (local_bh)
1804 brelse(local_bh);
1805
1806 mlog_exit(status);
1807 return status;
1808}
1809
1810/*
1811 * This is working around a lock inversion between tasks acquiring DLM locks
1812 * while holding a page lock and the vote thread which blocks dlm lock acquiry
1813 * while acquiring page locks.
1814 *
1815 * ** These _with_page variantes are only intended to be called from aop
1816 * methods that hold page locks and return a very specific *positive* error
1817 * code that aop methods pass up to the VFS -- test for errors with != 0. **
1818 *
1819 * The DLM is called such that it returns -EAGAIN if it would have blocked
1820 * waiting for the vote thread. In that case we unlock our page so the vote
1821 * thread can make progress. Once we've done this we have to return
1822 * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up
1823 * into the VFS who will then immediately retry the aop call.
1824 *
1825 * We do a blocking lock and immediate unlock before returning, though, so that
1826 * the lock has a great chance of being cached on this node by the time the VFS
1827 * calls back to retry the aop. This has a potential to livelock as nodes
1828 * ping locks back and forth, but that's a risk we're willing to take to avoid
1829 * the lock inversion simply.
1830 */
1831int ocfs2_meta_lock_with_page(struct inode *inode,
ccd979bd
MF		 1832 struct buffer_head **ret_bh,
1833 int ex,
1834 struct page *page)
1835{
1836 int ret;
1837
4bcec184 1838 ret = ocfs2_meta_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
ccd979bd
MF		 1839 if (ret == -EAGAIN) {
1840 unlock_page(page);
4bcec184 1841 if (ocfs2_meta_lock(inode, ret_bh, ex) == 0)
ccd979bd
MF		 1842 ocfs2_meta_unlock(inode, ex);
1843 ret = AOP_TRUNCATED_PAGE;
1844 }
1845
1846 return ret;
1847}
1848
7f1a37e3
TY		 1849int ocfs2_meta_lock_atime(struct inode *inode,
1850 struct vfsmount *vfsmnt,
1851 int *level)
1852{
1853 int ret;
1854
1855 mlog_entry_void();
1856 ret = ocfs2_meta_lock(inode, NULL, 0);
1857 if (ret < 0) {
1858 mlog_errno(ret);
1859 return ret;
1860 }
1861
1862 /*
1863 * If we should update atime, we will get EX lock,
1864 * otherwise we just get PR lock.
1865 */
1866 if (ocfs2_should_update_atime(inode, vfsmnt)) {
1867 struct buffer_head *bh = NULL;
1868
1869 ocfs2_meta_unlock(inode, 0);
1870 ret = ocfs2_meta_lock(inode, &bh, 1);
1871 if (ret < 0) {
1872 mlog_errno(ret);
1873 return ret;
1874 }
1875 *level = 1;
1876 if (ocfs2_should_update_atime(inode, vfsmnt))
1877 ocfs2_update_inode_atime(inode, bh);
1878 if (bh)
1879 brelse(bh);
1880 } else
1881 *level = 0;
1882
1883 mlog_exit(ret);
1884 return ret;
1885}
1886
ccd979bd
MF		 1887void ocfs2_meta_unlock(struct inode *inode,
1888 int ex)
1889{
1890 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1891 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres;
c271c5c2 1892 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1893
1894 mlog_entry_void();
1895
b0697053
MF		 1896 mlog(0, "inode %llu drop %s META lock\n",
1897 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1898 ex ? "EXMODE" : "PRMODE");
1899
c271c5c2
SM		 1900 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
1901 !ocfs2_mount_local(osb))
ccd979bd
MF		 1902 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1903
1904 mlog_exit_void();
1905}
1906
1907int ocfs2_super_lock(struct ocfs2_super *osb,
1908 int ex)
1909{
c271c5c2 1910 int status = 0;
ccd979bd
MF		 1911 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1912 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1913 struct buffer_head *bh;
1914 struct ocfs2_slot_info *si = osb->slot_info;
1915
1916 mlog_entry_void();
1917
1918 if (ocfs2_is_hard_readonly(osb))
1919 return -EROFS;
1920
c271c5c2
SM		 1921 if (ocfs2_mount_local(osb))
1922 goto bail;
1923
ccd979bd
MF		 1924 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1925 if (status < 0) {
1926 mlog_errno(status);
1927 goto bail;
1928 }
1929
1930 /* The super block lock path is really in the best position to
1931 * know when resources covered by the lock need to be
1932 * refreshed, so we do it here. Of course, making sense of
1933 * everything is up to the caller :) */
1934 status = ocfs2_should_refresh_lock_res(lockres);
1935 if (status < 0) {
1936 mlog_errno(status);
1937 goto bail;
1938 }
1939 if (status) {
1940 bh = si->si_bh;
1941 status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0,
1942 si->si_inode);
1943 if (status == 0)
1944 ocfs2_update_slot_info(si);
1945
1946 ocfs2_complete_lock_res_refresh(lockres, status);
1947
1948 if (status < 0)
1949 mlog_errno(status);
1950 }
1951bail:
1952 mlog_exit(status);
1953 return status;
1954}
1955
1956void ocfs2_super_unlock(struct ocfs2_super *osb,
1957 int ex)
1958{
1959 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1960 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1961
c271c5c2
SM		 1962 if (!ocfs2_mount_local(osb))
1963 ocfs2_cluster_unlock(osb, lockres, level);
ccd979bd
MF		 1964}
1965
1966int ocfs2_rename_lock(struct ocfs2_super *osb)
1967{
1968 int status;
1969 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1970
1971 if (ocfs2_is_hard_readonly(osb))
1972 return -EROFS;
1973
c271c5c2
SM		 1974 if (ocfs2_mount_local(osb))
1975 return 0;
1976
ccd979bd
MF		 1977 status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0);
1978 if (status < 0)
1979 mlog_errno(status);
1980
1981 return status;
1982}
1983
1984void ocfs2_rename_unlock(struct ocfs2_super *osb)
1985{
1986 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1987
c271c5c2
SM		 1988 if (!ocfs2_mount_local(osb))
1989 ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE);
ccd979bd
MF		 1990}
1991
d680efe9
MF		 1992int ocfs2_dentry_lock(struct dentry *dentry, int ex)
1993{
1994 int ret;
1995 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1996 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
1997 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
1998
1999 BUG_ON(!dl);
2000
2001 if (ocfs2_is_hard_readonly(osb))
2002 return -EROFS;
2003
c271c5c2
SM		 2004 if (ocfs2_mount_local(osb))
2005 return 0;
2006
d680efe9
MF		 2007 ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2008 if (ret < 0)
2009 mlog_errno(ret);
2010
2011 return ret;
2012}
2013
2014void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2015{
2016 int level = ex ? LKM_EXMODE : LKM_PRMODE;
2017 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2018 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2019
c271c5c2
SM		 2020 if (!ocfs2_mount_local(osb))
2021 ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
d680efe9
MF		 2022}
2023
ccd979bd
MF		 2024/* Reference counting of the dlm debug structure. We want this because
2025 * open references on the debug inodes can live on after a mount, so
2026 * we can't rely on the ocfs2_super to always exist. */
2027static void ocfs2_dlm_debug_free(struct kref *kref)
2028{
2029 struct ocfs2_dlm_debug *dlm_debug;
2030
2031 dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2032
2033 kfree(dlm_debug);
2034}
2035
2036void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2037{
2038 if (dlm_debug)
2039 kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2040}
2041
2042static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2043{
2044 kref_get(&debug->d_refcnt);
2045}
2046
2047struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2048{
2049 struct ocfs2_dlm_debug *dlm_debug;
2050
2051 dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2052 if (!dlm_debug) {
2053 mlog_errno(-ENOMEM);
2054 goto out;
2055 }
2056
2057 kref_init(&dlm_debug->d_refcnt);
2058 INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2059 dlm_debug->d_locking_state = NULL;
2060out:
2061 return dlm_debug;
2062}
2063
2064/* Access to this is arbitrated for us via seq_file->sem. */
2065struct ocfs2_dlm_seq_priv {
2066 struct ocfs2_dlm_debug *p_dlm_debug;
2067 struct ocfs2_lock_res p_iter_res;
2068 struct ocfs2_lock_res p_tmp_res;
2069};
2070
2071static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2072 struct ocfs2_dlm_seq_priv *priv)
2073{
2074 struct ocfs2_lock_res *iter, *ret = NULL;
2075 struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2076
2077 assert_spin_locked(&ocfs2_dlm_tracking_lock);
2078
2079 list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2080 /* discover the head of the list */
2081 if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2082 mlog(0, "End of list found, %p\n", ret);
2083 break;
2084 }
2085
2086 /* We track our "dummy" iteration lockres' by a NULL
2087 * l_ops field. */
2088 if (iter->l_ops != NULL) {
2089 ret = iter;
2090 break;
2091 }
2092 }
2093
2094 return ret;
2095}
2096
2097static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2098{
2099 struct ocfs2_dlm_seq_priv *priv = m->private;
2100 struct ocfs2_lock_res *iter;
2101
2102 spin_lock(&ocfs2_dlm_tracking_lock);
2103 iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2104 if (iter) {
2105 /* Since lockres' have the lifetime of their container
2106 * (which can be inodes, ocfs2_supers, etc) we want to
2107 * copy this out to a temporary lockres while still
2108 * under the spinlock. Obviously after this we can't
2109 * trust any pointers on the copy returned, but that's
2110 * ok as the information we want isn't typically held
2111 * in them. */
2112 priv->p_tmp_res = *iter;
2113 iter = &priv->p_tmp_res;
2114 }
2115 spin_unlock(&ocfs2_dlm_tracking_lock);
2116
2117 return iter;
2118}
2119
2120static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2121{
2122}
2123
2124static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2125{
2126 struct ocfs2_dlm_seq_priv *priv = m->private;
2127 struct ocfs2_lock_res *iter = v;
2128 struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2129
2130 spin_lock(&ocfs2_dlm_tracking_lock);
2131 iter = ocfs2_dlm_next_res(iter, priv);
2132 list_del_init(&dummy->l_debug_list);
2133 if (iter) {
2134 list_add(&dummy->l_debug_list, &iter->l_debug_list);
2135 priv->p_tmp_res = *iter;
2136 iter = &priv->p_tmp_res;
2137 }
2138 spin_unlock(&ocfs2_dlm_tracking_lock);
2139
2140 return iter;
2141}
2142
2143/* So that debugfs.ocfs2 can determine which format is being used */
2144#define OCFS2_DLM_DEBUG_STR_VERSION 1
2145static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2146{
2147 int i;
2148 char *lvb;
2149 struct ocfs2_lock_res *lockres = v;
2150
2151 if (!lockres)
2152 return -EINVAL;
2153
d680efe9
MF		 2154 seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2155
2156 if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2157 seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2158 lockres->l_name,
2159 (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2160 else
2161 seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2162
2163 seq_printf(m, "%d\t"
ccd979bd
MF		 2164 "0x%lx\t"
2165 "0x%x\t"
2166 "0x%x\t"
2167 "%u\t"
2168 "%u\t"
2169 "%d\t"
2170 "%d\t",
ccd979bd
MF		 2171 lockres->l_level,
2172 lockres->l_flags,
2173 lockres->l_action,
2174 lockres->l_unlock_action,
2175 lockres->l_ro_holders,
2176 lockres->l_ex_holders,
2177 lockres->l_requested,
2178 lockres->l_blocking);
2179
2180 /* Dump the raw LVB */
2181 lvb = lockres->l_lksb.lvb;
2182 for(i = 0; i < DLM_LVB_LEN; i++)
2183 seq_printf(m, "0x%x\t", lvb[i]);
2184
2185 /* End the line */
2186 seq_printf(m, "\n");
2187 return 0;
2188}
2189
2190static struct seq_operations ocfs2_dlm_seq_ops = {
2191 .start = ocfs2_dlm_seq_start,
2192 .stop = ocfs2_dlm_seq_stop,
2193 .next = ocfs2_dlm_seq_next,
2194 .show = ocfs2_dlm_seq_show,
2195};
2196
2197static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2198{
2199 struct seq_file *seq = (struct seq_file *) file->private_data;
2200 struct ocfs2_dlm_seq_priv *priv = seq->private;
2201 struct ocfs2_lock_res *res = &priv->p_iter_res;
2202
2203 ocfs2_remove_lockres_tracking(res);
2204 ocfs2_put_dlm_debug(priv->p_dlm_debug);
2205 return seq_release_private(inode, file);
2206}
2207
2208static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2209{
2210 int ret;
2211 struct ocfs2_dlm_seq_priv *priv;
2212 struct seq_file *seq;
2213 struct ocfs2_super *osb;
2214
2215 priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2216 if (!priv) {
2217 ret = -ENOMEM;
2218 mlog_errno(ret);
2219 goto out;
2220 }
8e18e294 2221 osb = inode->i_private;
ccd979bd
MF		 2222 ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2223 priv->p_dlm_debug = osb->osb_dlm_debug;
2224 INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2225
2226 ret = seq_open(file, &ocfs2_dlm_seq_ops);
2227 if (ret) {
2228 kfree(priv);
2229 mlog_errno(ret);
2230 goto out;
2231 }
2232
2233 seq = (struct seq_file *) file->private_data;
2234 seq->private = priv;
2235
2236 ocfs2_add_lockres_tracking(&priv->p_iter_res,
2237 priv->p_dlm_debug);
2238
2239out:
2240 return ret;
2241}
2242
4b6f5d20 2243static const struct file_operations ocfs2_dlm_debug_fops = {
ccd979bd
MF		 2244 .open = ocfs2_dlm_debug_open,
2245 .release = ocfs2_dlm_debug_release,
2246 .read = seq_read,
2247 .llseek = seq_lseek,
2248};
2249
2250static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2251{
2252 int ret = 0;
2253 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2254
2255 dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2256 S_IFREG|S_IRUSR,
2257 osb->osb_debug_root,
2258 osb,
2259 &ocfs2_dlm_debug_fops);
2260 if (!dlm_debug->d_locking_state) {
2261 ret = -EINVAL;
2262 mlog(ML_ERROR,
2263 "Unable to create locking state debugfs file.\n");
2264 goto out;
2265 }
2266
2267 ocfs2_get_dlm_debug(dlm_debug);
2268out:
2269 return ret;
2270}
2271
2272static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2273{
2274 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2275
2276 if (dlm_debug) {
2277 debugfs_remove(dlm_debug->d_locking_state);
2278 ocfs2_put_dlm_debug(dlm_debug);
2279 }
2280}
2281
2282int ocfs2_dlm_init(struct ocfs2_super *osb)
2283{
c271c5c2 2284 int status = 0;
ccd979bd 2285 u32 dlm_key;
c271c5c2 2286 struct dlm_ctxt *dlm = NULL;
ccd979bd
MF		 2287
2288 mlog_entry_void();
2289
c271c5c2
SM		 2290 if (ocfs2_mount_local(osb))
2291 goto local;
2292
ccd979bd
MF		 2293 status = ocfs2_dlm_init_debug(osb);
2294 if (status < 0) {
2295 mlog_errno(status);
2296 goto bail;
2297 }
2298
2299 /* launch vote thread */
78427043 2300 osb->vote_task = kthread_run(ocfs2_vote_thread, osb, "ocfs2vote");
ccd979bd
MF		 2301 if (IS_ERR(osb->vote_task)) {
2302 status = PTR_ERR(osb->vote_task);
2303 osb->vote_task = NULL;
2304 mlog_errno(status);
2305 goto bail;
2306 }
2307
2308 /* used by the dlm code to make message headers unique, each
2309 * node in this domain must agree on this. */
2310 dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str));
2311
2312 /* for now, uuid == domain */
2313 dlm = dlm_register_domain(osb->uuid_str, dlm_key);
2314 if (IS_ERR(dlm)) {
2315 status = PTR_ERR(dlm);
2316 mlog_errno(status);
2317 goto bail;
2318 }
2319
c271c5c2
SM		 2320 dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb);
2321
2322local:
ccd979bd
MF		 2323 ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
2324 ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
2325
ccd979bd
MF		 2326 osb->dlm = dlm;
2327
2328 status = 0;
2329bail:
2330 if (status < 0) {
2331 ocfs2_dlm_shutdown_debug(osb);
2332 if (osb->vote_task)
2333 kthread_stop(osb->vote_task);
2334 }
2335
2336 mlog_exit(status);
2337 return status;
2338}
2339
2340void ocfs2_dlm_shutdown(struct ocfs2_super *osb)
2341{
2342 mlog_entry_void();
2343
2344 dlm_unregister_eviction_cb(&osb->osb_eviction_cb);
2345
2346 ocfs2_drop_osb_locks(osb);
2347
2348 if (osb->vote_task) {
2349 kthread_stop(osb->vote_task);
2350 osb->vote_task = NULL;
2351 }
2352
2353 ocfs2_lock_res_free(&osb->osb_super_lockres);
2354 ocfs2_lock_res_free(&osb->osb_rename_lockres);
2355
2356 dlm_unregister_domain(osb->dlm);
2357 osb->dlm = NULL;
2358
2359 ocfs2_dlm_shutdown_debug(osb);
2360
2361 mlog_exit_void();
2362}
2363
2a45f2d1 2364static void ocfs2_unlock_ast(void *opaque, enum dlm_status status)
ccd979bd
MF		 2365{
2366 struct ocfs2_lock_res *lockres = opaque;
2367 unsigned long flags;
2368
2369 mlog_entry_void();
2370
2371 mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
2372 lockres->l_unlock_action);
2373
2374 spin_lock_irqsave(&lockres->l_lock, flags);
2375 /* We tried to cancel a convert request, but it was already
2376 * granted. All we want to do here is clear our unlock
2377 * state. The wake_up call done at the bottom is redundant
2378 * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't
2379 * hurt anything anyway */
2380 if (status == DLM_CANCELGRANT &&
2381 lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2382 mlog(0, "Got cancelgrant for %s\n", lockres->l_name);
2383
2384 /* We don't clear the busy flag in this case as it
2385 * should have been cleared by the ast which the dlm
2386 * has called. */
2387 goto complete_unlock;
2388 }
2389
2390 if (status != DLM_NORMAL) {
2391 mlog(ML_ERROR, "Dlm passes status %d for lock %s, "
2392 "unlock_action %d\n", status, lockres->l_name,
2393 lockres->l_unlock_action);
2394 spin_unlock_irqrestore(&lockres->l_lock, flags);
2395 return;
2396 }
2397
2398 switch(lockres->l_unlock_action) {
2399 case OCFS2_UNLOCK_CANCEL_CONVERT:
2400 mlog(0, "Cancel convert success for %s\n", lockres->l_name);
2401 lockres->l_action = OCFS2_AST_INVALID;
2402 break;
2403 case OCFS2_UNLOCK_DROP_LOCK:
2404 lockres->l_level = LKM_IVMODE;
2405 break;
2406 default:
2407 BUG();
2408 }
2409
2410 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
2411complete_unlock:
2412 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
2413 spin_unlock_irqrestore(&lockres->l_lock, flags);
2414
2415 wake_up(&lockres->l_event);
2416
2417 mlog_exit_void();
2418}
2419
ccd979bd 2420static int ocfs2_drop_lock(struct ocfs2_super *osb,
0d5dc6c2 2421 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 2422{
2423 enum dlm_status status;
2424 unsigned long flags;
b80fc012 2425 int lkm_flags = 0;
ccd979bd
MF		 2426
2427 /* We didn't get anywhere near actually using this lockres. */
2428 if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
2429 goto out;
2430
b80fc012
MF		 2431 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
2432 lkm_flags |= LKM_VALBLK;
2433
ccd979bd
MF		 2434 spin_lock_irqsave(&lockres->l_lock, flags);
2435
2436 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
2437 "lockres %s, flags 0x%lx\n",
2438 lockres->l_name, lockres->l_flags);
2439
2440 while (lockres->l_flags & OCFS2_LOCK_BUSY) {
2441 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2442 "%u, unlock_action = %u\n",
2443 lockres->l_name, lockres->l_flags, lockres->l_action,
2444 lockres->l_unlock_action);
2445
2446 spin_unlock_irqrestore(&lockres->l_lock, flags);
2447
2448 /* XXX: Today we just wait on any busy
2449 * locks... Perhaps we need to cancel converts in the
2450 * future? */
2451 ocfs2_wait_on_busy_lock(lockres);
2452
2453 spin_lock_irqsave(&lockres->l_lock, flags);
2454 }
2455
0d5dc6c2
MF		 2456 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2457 if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
2458 lockres->l_level == LKM_EXMODE &&
2459 !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2460 lockres->l_ops->set_lvb(lockres);
2461 }
ccd979bd
MF		 2462
2463 if (lockres->l_flags & OCFS2_LOCK_BUSY)
2464 mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
2465 lockres->l_name);
2466 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2467 mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
2468
2469 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
2470 spin_unlock_irqrestore(&lockres->l_lock, flags);
2471 goto out;
2472 }
2473
2474 lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
2475
2476 /* make sure we never get here while waiting for an ast to
2477 * fire. */
2478 BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
2479
2480 /* is this necessary? */
2481 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2482 lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
2483 spin_unlock_irqrestore(&lockres->l_lock, flags);
2484
2485 mlog(0, "lock %s\n", lockres->l_name);
2486
b80fc012 2487 status = dlmunlock(osb->dlm, &lockres->l_lksb, lkm_flags,
2a45f2d1 2488 ocfs2_unlock_ast, lockres);
ccd979bd
MF		 2489 if (status != DLM_NORMAL) {
2490 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2491 mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
2492 dlm_print_one_lock(lockres->l_lksb.lockid);
2493 BUG();
2494 }
2495 mlog(0, "lock %s, successfull return from dlmunlock\n",
2496 lockres->l_name);
2497
2498 ocfs2_wait_on_busy_lock(lockres);
2499out:
2500 mlog_exit(0);
2501 return 0;
2502}
2503
2504/* Mark the lockres as being dropped. It will no longer be
2505 * queued if blocking, but we still may have to wait on it
2506 * being dequeued from the vote thread before we can consider
2507 * it safe to drop.
2508 *
2509 * You can *not* attempt to call cluster_lock on this lockres anymore. */
2510void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
2511{
2512 int status;
2513 struct ocfs2_mask_waiter mw;
2514 unsigned long flags;
2515
2516 ocfs2_init_mask_waiter(&mw);
2517
2518 spin_lock_irqsave(&lockres->l_lock, flags);
2519 lockres->l_flags |= OCFS2_LOCK_FREEING;
2520 while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
2521 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
2522 spin_unlock_irqrestore(&lockres->l_lock, flags);
2523
2524 mlog(0, "Waiting on lockres %s\n", lockres->l_name);
2525
2526 status = ocfs2_wait_for_mask(&mw);
2527 if (status)
2528 mlog_errno(status);
2529
2530 spin_lock_irqsave(&lockres->l_lock, flags);
2531 }
2532 spin_unlock_irqrestore(&lockres->l_lock, flags);
2533}
2534
d680efe9
MF		 2535void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
2536 struct ocfs2_lock_res *lockres)
ccd979bd 2537{
d680efe9 2538 int ret;
ccd979bd 2539
d680efe9 2540 ocfs2_mark_lockres_freeing(lockres);
0d5dc6c2 2541 ret = ocfs2_drop_lock(osb, lockres);
d680efe9
MF		 2542 if (ret)
2543 mlog_errno(ret);
2544}
ccd979bd 2545
d680efe9
MF		 2546static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
2547{
2548 ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
2549 ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
ccd979bd
MF		 2550}
2551
ccd979bd
MF		 2552int ocfs2_drop_inode_locks(struct inode *inode)
2553{
2554 int status, err;
ccd979bd
MF		 2555
2556 mlog_entry_void();
2557
2558 /* No need to call ocfs2_mark_lockres_freeing here -
2559 * ocfs2_clear_inode has done it for us. */
2560
2561 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
50008630 2562 &OCFS2_I(inode)->ip_open_lockres);
ccd979bd
MF		 2563 if (err < 0)
2564 mlog_errno(err);
2565
2566 status = err;
2567
50008630
TY		 2568 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
2569 &OCFS2_I(inode)->ip_data_lockres);
2570 if (err < 0)
2571 mlog_errno(err);
2572 if (err < 0 && !status)
2573 status = err;
2574
ccd979bd 2575 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
0d5dc6c2 2576 &OCFS2_I(inode)->ip_meta_lockres);
ccd979bd
MF		 2577 if (err < 0)
2578 mlog_errno(err);
2579 if (err < 0 && !status)
2580 status = err;
2581
2582 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
0d5dc6c2 2583 &OCFS2_I(inode)->ip_rw_lockres);
ccd979bd
MF		 2584 if (err < 0)
2585 mlog_errno(err);
2586 if (err < 0 && !status)
2587 status = err;
2588
2589 mlog_exit(status);
2590 return status;
2591}
2592
2593static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
2594 int new_level)
2595{
2596 assert_spin_locked(&lockres->l_lock);
2597
2598 BUG_ON(lockres->l_blocking <= LKM_NLMODE);
2599
2600 if (lockres->l_level <= new_level) {
2601 mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n",
2602 lockres->l_level, new_level);
2603 BUG();
2604 }
2605
2606 mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
2607 lockres->l_name, new_level, lockres->l_blocking);
2608
2609 lockres->l_action = OCFS2_AST_DOWNCONVERT;
2610 lockres->l_requested = new_level;
2611 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2612}
2613
2614static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
2615 struct ocfs2_lock_res *lockres,
2616 int new_level,
2617 int lvb)
2618{
2619 int ret, dlm_flags = LKM_CONVERT;
2620 enum dlm_status status;
2621
2622 mlog_entry_void();
2623
2624 if (lvb)
2625 dlm_flags |= LKM_VALBLK;
2626
2627 status = dlmlock(osb->dlm,
2628 new_level,
2629 &lockres->l_lksb,
2630 dlm_flags,
2631 lockres->l_name,
f0681062 2632 OCFS2_LOCK_ID_MAX_LEN - 1,
e92d57df 2633 ocfs2_locking_ast,
ccd979bd 2634 lockres,
aa2623ad 2635 ocfs2_blocking_ast);
ccd979bd
MF		 2636 if (status != DLM_NORMAL) {
2637 ocfs2_log_dlm_error("dlmlock", status, lockres);
2638 ret = -EINVAL;
2639 ocfs2_recover_from_dlm_error(lockres, 1);
2640 goto bail;
2641 }
2642
2643 ret = 0;
2644bail:
2645 mlog_exit(ret);
2646 return ret;
2647}
2648
2649/* returns 1 when the caller should unlock and call dlmunlock */
2650static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
2651 struct ocfs2_lock_res *lockres)
2652{
2653 assert_spin_locked(&lockres->l_lock);
2654
2655 mlog_entry_void();
2656 mlog(0, "lock %s\n", lockres->l_name);
2657
2658 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2659 /* If we're already trying to cancel a lock conversion
2660 * then just drop the spinlock and allow the caller to
2661 * requeue this lock. */
2662
2663 mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
2664 return 0;
2665 }
2666
2667 /* were we in a convert when we got the bast fire? */
2668 BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
2669 lockres->l_action != OCFS2_AST_DOWNCONVERT);
2670 /* set things up for the unlockast to know to just
2671 * clear out the ast_action and unset busy, etc. */
2672 lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
2673
2674 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
2675 "lock %s, invalid flags: 0x%lx\n",
2676 lockres->l_name, lockres->l_flags);
2677
2678 return 1;
2679}
2680
2681static int ocfs2_cancel_convert(struct ocfs2_super *osb,
2682 struct ocfs2_lock_res *lockres)
2683{
2684 int ret;
2685 enum dlm_status status;
2686
2687 mlog_entry_void();
2688 mlog(0, "lock %s\n", lockres->l_name);
2689
2690 ret = 0;
2691 status = dlmunlock(osb->dlm,
2692 &lockres->l_lksb,
2693 LKM_CANCEL,
2a45f2d1 2694 ocfs2_unlock_ast,
ccd979bd
MF		 2695 lockres);
2696 if (status != DLM_NORMAL) {
2697 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2698 ret = -EINVAL;
2699 ocfs2_recover_from_dlm_error(lockres, 0);
2700 }
2701
2702 mlog(0, "lock %s return from dlmunlock\n", lockres->l_name);
2703
ccd979bd
MF		 2704 mlog_exit(ret);
2705 return ret;
2706}
2707
b5e500e2
MF		 2708static int ocfs2_unblock_lock(struct ocfs2_super *osb,
2709 struct ocfs2_lock_res *lockres,
2710 struct ocfs2_unblock_ctl *ctl)
ccd979bd
MF		 2711{
2712 unsigned long flags;
2713 int blocking;
2714 int new_level;
2715 int ret = 0;
5ef0d4ea 2716 int set_lvb = 0;
ccd979bd
MF		 2717
2718 mlog_entry_void();
2719
2720 spin_lock_irqsave(&lockres->l_lock, flags);
2721
2722 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
2723
2724recheck:
2725 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
d680efe9 2726 ctl->requeue = 1;
ccd979bd
MF		 2727 ret = ocfs2_prepare_cancel_convert(osb, lockres);
2728 spin_unlock_irqrestore(&lockres->l_lock, flags);
2729 if (ret) {
2730 ret = ocfs2_cancel_convert(osb, lockres);
2731 if (ret < 0)
2732 mlog_errno(ret);
2733 }
2734 goto leave;
2735 }
2736
2737 /* if we're blocking an exclusive and we have *any* holders,
2738 * then requeue. */
2739 if ((lockres->l_blocking == LKM_EXMODE)
f7fbfdd1
MF		 2740 && (lockres->l_ex_holders || lockres->l_ro_holders))
2741 goto leave_requeue;
ccd979bd
MF		 2742
2743 /* If it's a PR we're blocking, then only
2744 * requeue if we've got any EX holders */
2745 if (lockres->l_blocking == LKM_PRMODE &&
f7fbfdd1
MF		 2746 lockres->l_ex_holders)
2747 goto leave_requeue;
2748
2749 /*
2750 * Can we get a lock in this state if the holder counts are
2751 * zero? The meta data unblock code used to check this.
2752 */
2753 if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
2754 && (lockres->l_flags & OCFS2_LOCK_REFRESHING))
2755 goto leave_requeue;
ccd979bd 2756
16d5b956
MF		 2757 new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
2758
2759 if (lockres->l_ops->check_downconvert
2760 && !lockres->l_ops->check_downconvert(lockres, new_level))
2761 goto leave_requeue;
2762
ccd979bd
MF		 2763 /* If we get here, then we know that there are no more
2764 * incompatible holders (and anyone asking for an incompatible
2765 * lock is blocked). We can now downconvert the lock */
cc567d89 2766 if (!lockres->l_ops->downconvert_worker)
ccd979bd
MF		 2767 goto downconvert;
2768
2769 /* Some lockres types want to do a bit of work before
2770 * downconverting a lock. Allow that here. The worker function
2771 * may sleep, so we save off a copy of what we're blocking as
2772 * it may change while we're not holding the spin lock. */
2773 blocking = lockres->l_blocking;
2774 spin_unlock_irqrestore(&lockres->l_lock, flags);
2775
cc567d89 2776 ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
d680efe9
MF		 2777
2778 if (ctl->unblock_action == UNBLOCK_STOP_POST)
2779 goto leave;
ccd979bd
MF		 2780
2781 spin_lock_irqsave(&lockres->l_lock, flags);
2782 if (blocking != lockres->l_blocking) {
2783 /* If this changed underneath us, then we can't drop
2784 * it just yet. */
2785 goto recheck;
2786 }
2787
2788downconvert:
d680efe9 2789 ctl->requeue = 0;
ccd979bd 2790
5ef0d4ea
MF		 2791 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2792 if (lockres->l_level == LKM_EXMODE)
2793 set_lvb = 1;
2794
2795 /*
2796 * We only set the lvb if the lock has been fully
2797 * refreshed - otherwise we risk setting stale
2798 * data. Otherwise, there's no need to actually clear
2799 * out the lvb here as it's value is still valid.
2800 */
2801 if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2802 lockres->l_ops->set_lvb(lockres);
2803 }
2804
ccd979bd
MF		 2805 ocfs2_prepare_downconvert(lockres, new_level);
2806 spin_unlock_irqrestore(&lockres->l_lock, flags);
5ef0d4ea 2807 ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb);
ccd979bd
MF		 2808leave:
2809 mlog_exit(ret);
2810 return ret;
f7fbfdd1
MF		 2811
2812leave_requeue:
2813 spin_unlock_irqrestore(&lockres->l_lock, flags);
2814 ctl->requeue = 1;
2815
2816 mlog_exit(0);
2817 return 0;
ccd979bd
MF		 2818}
2819
d680efe9
MF		 2820static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
2821 int blocking)
ccd979bd
MF		 2822{
2823 struct inode *inode;
2824 struct address_space *mapping;
2825
ccd979bd
MF		 2826 inode = ocfs2_lock_res_inode(lockres);
2827 mapping = inode->i_mapping;
2828
7f4a2a97
MF		 2829 /*
2830 * We need this before the filemap_fdatawrite() so that it can
2831 * transfer the dirty bit from the PTE to the
2832 * page. Unfortunately this means that even for EX->PR
2833 * downconverts, we'll lose our mappings and have to build
2834 * them up again.
2835 */
2836 unmap_mapping_range(mapping, 0, 0, 0);
2837
ccd979bd 2838 if (filemap_fdatawrite(mapping)) {
b0697053
MF		 2839 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
2840 (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 2841 }
2842 sync_mapping_buffers(mapping);
2843 if (blocking == LKM_EXMODE) {
2844 truncate_inode_pages(mapping, 0);
ccd979bd
MF		 2845 } else {
2846 /* We only need to wait on the I/O if we're not also
2847 * truncating pages because truncate_inode_pages waits
2848 * for us above. We don't truncate pages if we're
2849 * blocking anything < EXMODE because we want to keep
2850 * them around in that case. */
2851 filemap_fdatawait(mapping);
2852 }
2853
d680efe9 2854 return UNBLOCK_CONTINUE;
ccd979bd
MF		 2855}
2856
810d5aeb
MF		 2857static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
2858 int new_level)
2859{
2860 struct inode *inode = ocfs2_lock_res_inode(lockres);
2861 int checkpointed = ocfs2_inode_fully_checkpointed(inode);
2862
2863 BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE);
2864 BUG_ON(lockres->l_level != LKM_EXMODE && !checkpointed);
2865
2866 if (checkpointed)
2867 return 1;
2868
2869 ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb));
2870 return 0;
2871}
2872
2873static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
2874{
2875 struct inode *inode = ocfs2_lock_res_inode(lockres);
2876
2877 __ocfs2_stuff_meta_lvb(inode);
2878}
2879
d680efe9
MF		 2880/*
2881 * Does the final reference drop on our dentry lock. Right now this
2882 * happens in the vote thread, but we could choose to simplify the
2883 * dlmglue API and push these off to the ocfs2_wq in the future.
2884 */
2885static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
2886 struct ocfs2_lock_res *lockres)
2887{
2888 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
2889 ocfs2_dentry_lock_put(osb, dl);
2890}
2891
2892/*
2893 * d_delete() matching dentries before the lock downconvert.
2894 *
2895 * At this point, any process waiting to destroy the
2896 * dentry_lock due to last ref count is stopped by the
2897 * OCFS2_LOCK_QUEUED flag.
2898 *
2899 * We have two potential problems
2900 *
2901 * 1) If we do the last reference drop on our dentry_lock (via dput)
2902 * we'll wind up in ocfs2_release_dentry_lock(), waiting on
2903 * the downconvert to finish. Instead we take an elevated
2904 * reference and push the drop until after we've completed our
2905 * unblock processing.
2906 *
2907 * 2) There might be another process with a final reference,
2908 * waiting on us to finish processing. If this is the case, we
2909 * detect it and exit out - there's no more dentries anyway.
2910 */
2911static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
2912 int blocking)
2913{
2914 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
2915 struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
2916 struct dentry *dentry;
2917 unsigned long flags;
2918 int extra_ref = 0;
2919
2920 /*
2921 * This node is blocking another node from getting a read
2922 * lock. This happens when we've renamed within a
2923 * directory. We've forced the other nodes to d_delete(), but
2924 * we never actually dropped our lock because it's still
2925 * valid. The downconvert code will retain a PR for this node,
2926 * so there's no further work to do.
2927 */
2928 if (blocking == LKM_PRMODE)
2929 return UNBLOCK_CONTINUE;
2930
2931 /*
2932 * Mark this inode as potentially orphaned. The code in
2933 * ocfs2_delete_inode() will figure out whether it actually
2934 * needs to be freed or not.
2935 */
2936 spin_lock(&oi->ip_lock);
2937 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2938 spin_unlock(&oi->ip_lock);
2939
2940 /*
2941 * Yuck. We need to make sure however that the check of
2942 * OCFS2_LOCK_FREEING and the extra reference are atomic with
2943 * respect to a reference decrement or the setting of that
2944 * flag.
2945 */
2946 spin_lock_irqsave(&lockres->l_lock, flags);
2947 spin_lock(&dentry_attach_lock);
2948 if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
2949 && dl->dl_count) {
2950 dl->dl_count++;
2951 extra_ref = 1;
2952 }
2953 spin_unlock(&dentry_attach_lock);
2954 spin_unlock_irqrestore(&lockres->l_lock, flags);
2955
2956 mlog(0, "extra_ref = %d\n", extra_ref);
2957
2958 /*
2959 * We have a process waiting on us in ocfs2_dentry_iput(),
2960 * which means we can't have any more outstanding
2961 * aliases. There's no need to do any more work.
2962 */
2963 if (!extra_ref)
2964 return UNBLOCK_CONTINUE;
2965
2966 spin_lock(&dentry_attach_lock);
2967 while (1) {
2968 dentry = ocfs2_find_local_alias(dl->dl_inode,
2969 dl->dl_parent_blkno, 1);
2970 if (!dentry)
2971 break;
2972 spin_unlock(&dentry_attach_lock);
2973
2974 mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
2975 dentry->d_name.name);
2976
2977 /*
2978 * The following dcache calls may do an
2979 * iput(). Normally we don't want that from the
2980 * downconverting thread, but in this case it's ok
2981 * because the requesting node already has an
2982 * exclusive lock on the inode, so it can't be queued
2983 * for a downconvert.
2984 */
2985 d_delete(dentry);
2986 dput(dentry);
2987
2988 spin_lock(&dentry_attach_lock);
2989 }
2990 spin_unlock(&dentry_attach_lock);
2991
2992 /*
2993 * If we are the last holder of this dentry lock, there is no
2994 * reason to downconvert so skip straight to the unlock.
2995 */
2996 if (dl->dl_count == 1)
2997 return UNBLOCK_STOP_POST;
2998
2999 return UNBLOCK_CONTINUE_POST;
3000}
3001
ccd979bd
MF		 3002void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3003 struct ocfs2_lock_res *lockres)
3004{
3005 int status;
d680efe9 3006 struct ocfs2_unblock_ctl ctl = {0, 0,};
ccd979bd
MF		 3007 unsigned long flags;
3008
3009 /* Our reference to the lockres in this function can be
3010 * considered valid until we remove the OCFS2_LOCK_QUEUED
3011 * flag. */
3012
3013 mlog_entry_void();
3014
3015 BUG_ON(!lockres);
3016 BUG_ON(!lockres->l_ops);
ccd979bd
MF		 3017
3018 mlog(0, "lockres %s blocked.\n", lockres->l_name);
3019
3020 /* Detect whether a lock has been marked as going away while
3021 * the vote thread was processing other things. A lock can
3022 * still be marked with OCFS2_LOCK_FREEING after this check,
3023 * but short circuiting here will still save us some
3024 * performance. */
3025 spin_lock_irqsave(&lockres->l_lock, flags);
3026 if (lockres->l_flags & OCFS2_LOCK_FREEING)
3027 goto unqueue;
3028 spin_unlock_irqrestore(&lockres->l_lock, flags);
3029
b5e500e2 3030 status = ocfs2_unblock_lock(osb, lockres, &ctl);
ccd979bd
MF		 3031 if (status < 0)
3032 mlog_errno(status);
3033
3034 spin_lock_irqsave(&lockres->l_lock, flags);
3035unqueue:
d680efe9 3036 if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
ccd979bd
MF		 3037 lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3038 } else
3039 ocfs2_schedule_blocked_lock(osb, lockres);
3040
3041 mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
d680efe9 3042 ctl.requeue ? "yes" : "no");
ccd979bd
MF		 3043 spin_unlock_irqrestore(&lockres->l_lock, flags);
3044
d680efe9
MF		 3045 if (ctl.unblock_action != UNBLOCK_CONTINUE
3046 && lockres->l_ops->post_unlock)
3047 lockres->l_ops->post_unlock(osb, lockres);
3048
ccd979bd
MF		 3049 mlog_exit_void();
3050}
3051
3052static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3053 struct ocfs2_lock_res *lockres)
3054{
3055 mlog_entry_void();
3056
3057 assert_spin_locked(&lockres->l_lock);
3058
3059 if (lockres->l_flags & OCFS2_LOCK_FREEING) {
3060 /* Do not schedule a lock for downconvert when it's on
3061 * the way to destruction - any nodes wanting access
3062 * to the resource will get it soon. */
3063 mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
3064 lockres->l_name, lockres->l_flags);
3065 return;
3066 }
3067
3068 lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
3069
3070 spin_lock(&osb->vote_task_lock);
3071 if (list_empty(&lockres->l_blocked_list)) {
3072 list_add_tail(&lockres->l_blocked_list,
3073 &osb->blocked_lock_list);
3074 osb->blocked_lock_count++;
3075 }
3076 spin_unlock(&osb->vote_task_lock);
3077
3078 mlog_exit_void();
3079}
3080
3081/* This aids in debugging situations where a bad LVB might be involved. */
3082void ocfs2_dump_meta_lvb_info(u64 level,
3083 const char *function,
3084 unsigned int line,
3085 struct ocfs2_lock_res *lockres)
3086{
3087 struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
3088
3089 mlog(level, "LVB information for %s (called from %s:%u):\n",
3090 lockres->l_name, function, line);
f9e2d82e
MF		 3091 mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
3092 lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
3093 be32_to_cpu(lvb->lvb_igeneration));
b0697053
MF		 3094 mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
3095 (unsigned long long)be64_to_cpu(lvb->lvb_isize),
3096 be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
3097 be16_to_cpu(lvb->lvb_imode));
3098 mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
ca4d147e 3099 "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
b0697053
MF		 3100 (long long)be64_to_cpu(lvb->lvb_iatime_packed),
3101 (long long)be64_to_cpu(lvb->lvb_ictime_packed),
ca4d147e
HP		 3102 (long long)be64_to_cpu(lvb->lvb_imtime_packed),
3103 be32_to_cpu(lvb->lvb_iattr));
ccd979bd 3104}
Linux kernel - Deliverables
This page took 0.298277 seconds and 5 git commands to generate.