projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Linux 2.6.19
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / ocfs2 / dlmglue.c
CommitLineData
ccd979bd
MF		 1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmglue.c
5 *
6 * Code which implements an OCFS2 specific interface to our DLM.
7 *
8 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26#include <linux/types.h>
27#include <linux/slab.h>
28#include <linux/highmem.h>
29#include <linux/mm.h>
30#include <linux/smp_lock.h>
31#include <linux/crc32.h>
32#include <linux/kthread.h>
33#include <linux/pagemap.h>
34#include <linux/debugfs.h>
35#include <linux/seq_file.h>
36
37#include <cluster/heartbeat.h>
38#include <cluster/nodemanager.h>
39#include <cluster/tcp.h>
40
41#include <dlm/dlmapi.h>
42
43#define MLOG_MASK_PREFIX ML_DLM_GLUE
44#include <cluster/masklog.h>
45
46#include "ocfs2.h"
47
48#include "alloc.h"
d680efe9 49#include "dcache.h"
ccd979bd
MF		 50#include "dlmglue.h"
51#include "extent_map.h"
52#include "heartbeat.h"
53#include "inode.h"
54#include "journal.h"
55#include "slot_map.h"
56#include "super.h"
57#include "uptodate.h"
58#include "vote.h"
59
60#include "buffer_head_io.h"
61
62struct ocfs2_mask_waiter {
63 struct list_head mw_item;
64 int mw_status;
65 struct completion mw_complete;
66 unsigned long mw_mask;
67 unsigned long mw_goal;
68};
69
54a7e755
MF		 70static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
71static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
ccd979bd 72
d680efe9 73/*
cc567d89 74 * Return value from ->downconvert_worker functions.
d680efe9 75 *
b5e500e2 76 * These control the precise actions of ocfs2_unblock_lock()
d680efe9
MF		 77 * and ocfs2_process_blocked_lock()
78 *
79 */
80enum ocfs2_unblock_action {
81 UNBLOCK_CONTINUE = 0, /* Continue downconvert */
82 UNBLOCK_CONTINUE_POST = 1, /* Continue downconvert, fire
83 * ->post_unlock callback */
84 UNBLOCK_STOP_POST = 2, /* Do not downconvert, fire
85 * ->post_unlock() callback. */
86};
87
88struct ocfs2_unblock_ctl {
89 int requeue;
90 enum ocfs2_unblock_action unblock_action;
91};
92
810d5aeb
MF		 93static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
94 int new_level);
95static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
96
cc567d89
MF		 97static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
98 int blocking);
99
cc567d89
MF		 100static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
101 int blocking);
d680efe9
MF		 102
103static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
104 struct ocfs2_lock_res *lockres);
ccd979bd 105
f625c979
MF		 106/*
107 * OCFS2 Lock Resource Operations
108 *
109 * These fine tune the behavior of the generic dlmglue locking infrastructure.
0d5dc6c2
MF		 110 *
111 * The most basic of lock types can point ->l_priv to their respective
112 * struct ocfs2_super and allow the default actions to manage things.
113 *
114 * Right now, each lock type also needs to implement an init function,
115 * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
116 * should be called when the lock is no longer needed (i.e., object
117 * destruction time).
f625c979 118 */
ccd979bd 119struct ocfs2_lock_res_ops {
54a7e755
MF		 120 /*
121 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
122 * this callback if ->l_priv is not an ocfs2_super pointer
123 */
124 struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
b5e500e2 125
0d5dc6c2
MF		 126 /*
127 * Optionally called in the downconvert (or "vote") thread
128 * after a successful downconvert. The lockres will not be
129 * referenced after this callback is called, so it is safe to
130 * free memory, etc.
131 *
132 * The exact semantics of when this is called are controlled
133 * by ->downconvert_worker()
134 */
d680efe9 135 void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
f625c979 136
16d5b956
MF		 137 /*
138 * Allow a lock type to add checks to determine whether it is
139 * safe to downconvert a lock. Return 0 to re-queue the
140 * downconvert at a later time, nonzero to continue.
141 *
142 * For most locks, the default checks that there are no
143 * incompatible holders are sufficient.
144 *
145 * Called with the lockres spinlock held.
146 */
147 int (*check_downconvert)(struct ocfs2_lock_res *, int);
148
5ef0d4ea
MF		 149 /*
150 * Allows a lock type to populate the lock value block. This
151 * is called on downconvert, and when we drop a lock.
152 *
153 * Locks that want to use this should set LOCK_TYPE_USES_LVB
154 * in the flags field.
155 *
156 * Called with the lockres spinlock held.
157 */
158 void (*set_lvb)(struct ocfs2_lock_res *);
159
cc567d89
MF		 160 /*
161 * Called from the downconvert thread when it is determined
162 * that a lock will be downconverted. This is called without
163 * any locks held so the function can do work that might
164 * schedule (syncing out data, etc).
165 *
166 * This should return any one of the ocfs2_unblock_action
167 * values, depending on what it wants the thread to do.
168 */
169 int (*downconvert_worker)(struct ocfs2_lock_res *, int);
170
f625c979
MF		 171 /*
172 * LOCK_TYPE_* flags which describe the specific requirements
173 * of a lock type. Descriptions of each individual flag follow.
174 */
175 int flags;
ccd979bd
MF		 176};
177
f625c979
MF		 178/*
179 * Some locks want to "refresh" potentially stale data when a
180 * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
181 * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
182 * individual lockres l_flags member from the ast function. It is
183 * expected that the locking wrapper will clear the
184 * OCFS2_LOCK_NEEDS_REFRESH flag when done.
185 */
186#define LOCK_TYPE_REQUIRES_REFRESH 0x1
187
b80fc012 188/*
5ef0d4ea
MF		 189 * Indicate that a lock type makes use of the lock value block. The
190 * ->set_lvb lock type callback must be defined.
b80fc012
MF		 191 */
192#define LOCK_TYPE_USES_LVB 0x2
193
ccd979bd 194static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
54a7e755 195 .get_osb = ocfs2_get_inode_osb,
f625c979 196 .flags = 0,
ccd979bd
MF		 197};
198
199static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = {
54a7e755 200 .get_osb = ocfs2_get_inode_osb,
810d5aeb
MF		 201 .check_downconvert = ocfs2_check_meta_downconvert,
202 .set_lvb = ocfs2_set_meta_lvb,
b80fc012 203 .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
ccd979bd
MF		 204};
205
ccd979bd 206static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = {
54a7e755 207 .get_osb = ocfs2_get_inode_osb,
cc567d89 208 .downconvert_worker = ocfs2_data_convert_worker,
f625c979 209 .flags = 0,
ccd979bd
MF		 210};
211
212static struct ocfs2_lock_res_ops ocfs2_super_lops = {
f625c979 213 .flags = LOCK_TYPE_REQUIRES_REFRESH,
ccd979bd
MF		 214};
215
216static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
f625c979 217 .flags = 0,
ccd979bd
MF		 218};
219
d680efe9 220static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
54a7e755 221 .get_osb = ocfs2_get_dentry_osb,
d680efe9 222 .post_unlock = ocfs2_dentry_post_unlock,
cc567d89 223 .downconvert_worker = ocfs2_dentry_convert_worker,
f625c979 224 .flags = 0,
d680efe9
MF		 225};
226
ccd979bd
MF		 227static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
228{
229 return lockres->l_type == OCFS2_LOCK_TYPE_META ||
230 lockres->l_type == OCFS2_LOCK_TYPE_DATA ||
231 lockres->l_type == OCFS2_LOCK_TYPE_RW;
232}
233
ccd979bd
MF		 234static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
235{
236 BUG_ON(!ocfs2_is_inode_lock(lockres));
237
238 return (struct inode *) lockres->l_priv;
239}
240
d680efe9
MF		 241static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
242{
243 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
244
245 return (struct ocfs2_dentry_lock *)lockres->l_priv;
246}
247
54a7e755
MF		 248static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
249{
250 if (lockres->l_ops->get_osb)
251 return lockres->l_ops->get_osb(lockres);
252
253 return (struct ocfs2_super *)lockres->l_priv;
254}
255
ccd979bd
MF		 256static int ocfs2_lock_create(struct ocfs2_super *osb,
257 struct ocfs2_lock_res *lockres,
258 int level,
259 int dlm_flags);
260static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
261 int wanted);
262static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
263 struct ocfs2_lock_res *lockres,
264 int level);
265static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
266static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
267static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
268static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
269static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
270 struct ocfs2_lock_res *lockres);
271static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
272 int convert);
273#define ocfs2_log_dlm_error(_func, _stat, _lockres) do { \
274 mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \
275 "resource %s: %s\n", dlm_errname(_stat), _func, \
276 _lockres->l_name, dlm_errmsg(_stat)); \
277} while (0)
278static void ocfs2_vote_on_unlock(struct ocfs2_super *osb,
279 struct ocfs2_lock_res *lockres);
280static int ocfs2_meta_lock_update(struct inode *inode,
281 struct buffer_head **bh);
282static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
283static inline int ocfs2_highest_compat_lock_level(int level);
ccd979bd 284
ccd979bd
MF		 285static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
286 u64 blkno,
287 u32 generation,
288 char *name)
289{
290 int len;
291
292 mlog_entry_void();
293
294 BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
295
b0697053
MF		 296 len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
297 ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
298 (long long)blkno, generation);
ccd979bd
MF		 299
300 BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
301
302 mlog(0, "built lock resource with name: %s\n", name);
303
304 mlog_exit_void();
305}
306
34af946a 307static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
ccd979bd
MF		 308
309static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
310 struct ocfs2_dlm_debug *dlm_debug)
311{
312 mlog(0, "Add tracking for lockres %s\n", res->l_name);
313
314 spin_lock(&ocfs2_dlm_tracking_lock);
315 list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
316 spin_unlock(&ocfs2_dlm_tracking_lock);
317}
318
319static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
320{
321 spin_lock(&ocfs2_dlm_tracking_lock);
322 if (!list_empty(&res->l_debug_list))
323 list_del_init(&res->l_debug_list);
324 spin_unlock(&ocfs2_dlm_tracking_lock);
325}
326
327static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
328 struct ocfs2_lock_res *res,
329 enum ocfs2_lock_type type,
ccd979bd
MF		 330 struct ocfs2_lock_res_ops *ops,
331 void *priv)
332{
ccd979bd
MF		 333 res->l_type = type;
334 res->l_ops = ops;
335 res->l_priv = priv;
336
337 res->l_level = LKM_IVMODE;
338 res->l_requested = LKM_IVMODE;
339 res->l_blocking = LKM_IVMODE;
340 res->l_action = OCFS2_AST_INVALID;
341 res->l_unlock_action = OCFS2_UNLOCK_INVALID;
342
343 res->l_flags = OCFS2_LOCK_INITIALIZED;
344
345 ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
346}
347
348void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
349{
350 /* This also clears out the lock status block */
351 memset(res, 0, sizeof(struct ocfs2_lock_res));
352 spin_lock_init(&res->l_lock);
353 init_waitqueue_head(&res->l_event);
354 INIT_LIST_HEAD(&res->l_blocked_list);
355 INIT_LIST_HEAD(&res->l_mask_waiters);
356}
357
358void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
359 enum ocfs2_lock_type type,
24c19ef4 360 unsigned int generation,
ccd979bd
MF		 361 struct inode *inode)
362{
363 struct ocfs2_lock_res_ops *ops;
364
365 switch(type) {
366 case OCFS2_LOCK_TYPE_RW:
367 ops = &ocfs2_inode_rw_lops;
368 break;
369 case OCFS2_LOCK_TYPE_META:
370 ops = &ocfs2_inode_meta_lops;
371 break;
372 case OCFS2_LOCK_TYPE_DATA:
373 ops = &ocfs2_inode_data_lops;
374 break;
375 default:
376 mlog_bug_on_msg(1, "type: %d\n", type);
377 ops = NULL; /* thanks, gcc */
378 break;
379 };
380
d680efe9 381 ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
24c19ef4 382 generation, res->l_name);
d680efe9
MF		 383 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
384}
385
54a7e755
MF		 386static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
387{
388 struct inode *inode = ocfs2_lock_res_inode(lockres);
389
390 return OCFS2_SB(inode->i_sb);
391}
392
d680efe9
MF		 393static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
394{
395 __be64 inode_blkno_be;
396
397 memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
398 sizeof(__be64));
399
400 return be64_to_cpu(inode_blkno_be);
401}
402
54a7e755
MF		 403static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
404{
405 struct ocfs2_dentry_lock *dl = lockres->l_priv;
406
407 return OCFS2_SB(dl->dl_inode->i_sb);
408}
409
d680efe9
MF		 410void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
411 u64 parent, struct inode *inode)
412{
413 int len;
414 u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
415 __be64 inode_blkno_be = cpu_to_be64(inode_blkno);
416 struct ocfs2_lock_res *lockres = &dl->dl_lockres;
417
418 ocfs2_lock_res_init_once(lockres);
419
420 /*
421 * Unfortunately, the standard lock naming scheme won't work
422 * here because we have two 16 byte values to use. Instead,
423 * we'll stuff the inode number as a binary value. We still
424 * want error prints to show something without garbling the
425 * display, so drop a null byte in there before the inode
426 * number. A future version of OCFS2 will likely use all
427 * binary lock names. The stringified names have been a
428 * tremendous aid in debugging, but now that the debugfs
429 * interface exists, we can mangle things there if need be.
430 *
431 * NOTE: We also drop the standard "pad" value (the total lock
432 * name size stays the same though - the last part is all
433 * zeros due to the memset in ocfs2_lock_res_init_once()
434 */
435 len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
436 "%c%016llx",
437 ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
438 (long long)parent);
439
440 BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
441
442 memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
443 sizeof(__be64));
444
445 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
446 OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
447 dl);
ccd979bd
MF		 448}
449
450static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
451 struct ocfs2_super *osb)
452{
453 /* Superblock lockres doesn't come from a slab so we call init
454 * once on it manually. */
455 ocfs2_lock_res_init_once(res);
d680efe9
MF		 456 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
457 0, res->l_name);
ccd979bd 458 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
ccd979bd
MF		 459 &ocfs2_super_lops, osb);
460}
461
462static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
463 struct ocfs2_super *osb)
464{
465 /* Rename lockres doesn't come from a slab so we call init
466 * once on it manually. */
467 ocfs2_lock_res_init_once(res);
d680efe9
MF		 468 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
469 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
ccd979bd
MF		 470 &ocfs2_rename_lops, osb);
471}
472
473void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
474{
475 mlog_entry_void();
476
477 if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
478 return;
479
480 ocfs2_remove_lockres_tracking(res);
481
482 mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
483 "Lockres %s is on the blocked list\n",
484 res->l_name);
485 mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
486 "Lockres %s has mask waiters pending\n",
487 res->l_name);
488 mlog_bug_on_msg(spin_is_locked(&res->l_lock),
489 "Lockres %s is locked\n",
490 res->l_name);
491 mlog_bug_on_msg(res->l_ro_holders,
492 "Lockres %s has %u ro holders\n",
493 res->l_name, res->l_ro_holders);
494 mlog_bug_on_msg(res->l_ex_holders,
495 "Lockres %s has %u ex holders\n",
496 res->l_name, res->l_ex_holders);
497
498 /* Need to clear out the lock status block for the dlm */
499 memset(&res->l_lksb, 0, sizeof(res->l_lksb));
500
501 res->l_flags = 0UL;
502 mlog_exit_void();
503}
504
505static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
506 int level)
507{
508 mlog_entry_void();
509
510 BUG_ON(!lockres);
511
512 switch(level) {
513 case LKM_EXMODE:
514 lockres->l_ex_holders++;
515 break;
516 case LKM_PRMODE:
517 lockres->l_ro_holders++;
518 break;
519 default:
520 BUG();
521 }
522
523 mlog_exit_void();
524}
525
526static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
527 int level)
528{
529 mlog_entry_void();
530
531 BUG_ON(!lockres);
532
533 switch(level) {
534 case LKM_EXMODE:
535 BUG_ON(!lockres->l_ex_holders);
536 lockres->l_ex_holders--;
537 break;
538 case LKM_PRMODE:
539 BUG_ON(!lockres->l_ro_holders);
540 lockres->l_ro_holders--;
541 break;
542 default:
543 BUG();
544 }
545 mlog_exit_void();
546}
547
548/* WARNING: This function lives in a world where the only three lock
549 * levels are EX, PR, and NL. It *will* have to be adjusted when more
550 * lock types are added. */
551static inline int ocfs2_highest_compat_lock_level(int level)
552{
553 int new_level = LKM_EXMODE;
554
555 if (level == LKM_EXMODE)
556 new_level = LKM_NLMODE;
557 else if (level == LKM_PRMODE)
558 new_level = LKM_PRMODE;
559 return new_level;
560}
561
562static void lockres_set_flags(struct ocfs2_lock_res *lockres,
563 unsigned long newflags)
564{
565 struct list_head *pos, *tmp;
566 struct ocfs2_mask_waiter *mw;
567
568 assert_spin_locked(&lockres->l_lock);
569
570 lockres->l_flags = newflags;
571
572 list_for_each_safe(pos, tmp, &lockres->l_mask_waiters) {
573 mw = list_entry(pos, struct ocfs2_mask_waiter, mw_item);
574 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
575 continue;
576
577 list_del_init(&mw->mw_item);
578 mw->mw_status = 0;
579 complete(&mw->mw_complete);
580 }
581}
582static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
583{
584 lockres_set_flags(lockres, lockres->l_flags | or);
585}
586static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
587 unsigned long clear)
588{
589 lockres_set_flags(lockres, lockres->l_flags & ~clear);
590}
591
592static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
593{
594 mlog_entry_void();
595
596 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
597 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
598 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
599 BUG_ON(lockres->l_blocking <= LKM_NLMODE);
600
601 lockres->l_level = lockres->l_requested;
602 if (lockres->l_level <=
603 ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
604 lockres->l_blocking = LKM_NLMODE;
605 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
606 }
607 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
608
609 mlog_exit_void();
610}
611
612static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
613{
614 mlog_entry_void();
615
616 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
617 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
618
619 /* Convert from RO to EX doesn't really need anything as our
620 * information is already up to data. Convert from NL to
621 * *anything* however should mark ourselves as needing an
622 * update */
f625c979
MF		 623 if (lockres->l_level == LKM_NLMODE &&
624 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
ccd979bd
MF		 625 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
626
627 lockres->l_level = lockres->l_requested;
628 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
629
630 mlog_exit_void();
631}
632
633static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
634{
635 mlog_entry_void();
636
637 BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY));
638 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
639
640 if (lockres->l_requested > LKM_NLMODE &&
f625c979
MF		 641 !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
642 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
ccd979bd
MF		 643 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
644
645 lockres->l_level = lockres->l_requested;
646 lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
647 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
648
649 mlog_exit_void();
650}
651
ccd979bd
MF		 652static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
653 int level)
654{
655 int needs_downconvert = 0;
656 mlog_entry_void();
657
658 assert_spin_locked(&lockres->l_lock);
659
660 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
661
662 if (level > lockres->l_blocking) {
663 /* only schedule a downconvert if we haven't already scheduled
664 * one that goes low enough to satisfy the level we're
665 * blocking. this also catches the case where we get
666 * duplicate BASTs */
667 if (ocfs2_highest_compat_lock_level(level) <
668 ocfs2_highest_compat_lock_level(lockres->l_blocking))
669 needs_downconvert = 1;
670
671 lockres->l_blocking = level;
672 }
673
674 mlog_exit(needs_downconvert);
675 return needs_downconvert;
676}
677
aa2623ad 678static void ocfs2_blocking_ast(void *opaque, int level)
ccd979bd 679{
aa2623ad
MF		 680 struct ocfs2_lock_res *lockres = opaque;
681 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
ccd979bd
MF		 682 int needs_downconvert;
683 unsigned long flags;
684
ccd979bd
MF		 685 BUG_ON(level <= LKM_NLMODE);
686
aa2623ad
MF		 687 mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n",
688 lockres->l_name, level, lockres->l_level,
689 ocfs2_lock_type_string(lockres->l_type));
690
ccd979bd
MF		 691 spin_lock_irqsave(&lockres->l_lock, flags);
692 needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
693 if (needs_downconvert)
694 ocfs2_schedule_blocked_lock(osb, lockres);
695 spin_unlock_irqrestore(&lockres->l_lock, flags);
696
d680efe9
MF		 697 wake_up(&lockres->l_event);
698
ccd979bd 699 ocfs2_kick_vote_thread(osb);
ccd979bd
MF		 700}
701
e92d57df 702static void ocfs2_locking_ast(void *opaque)
ccd979bd 703{
e92d57df 704 struct ocfs2_lock_res *lockres = opaque;
ccd979bd
MF		 705 struct dlm_lockstatus *lksb = &lockres->l_lksb;
706 unsigned long flags;
707
708 spin_lock_irqsave(&lockres->l_lock, flags);
709
710 if (lksb->status != DLM_NORMAL) {
711 mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n",
712 lockres->l_name, lksb->status);
713 spin_unlock_irqrestore(&lockres->l_lock, flags);
714 return;
715 }
716
717 switch(lockres->l_action) {
718 case OCFS2_AST_ATTACH:
719 ocfs2_generic_handle_attach_action(lockres);
e92d57df 720 lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
ccd979bd
MF		 721 break;
722 case OCFS2_AST_CONVERT:
723 ocfs2_generic_handle_convert_action(lockres);
724 break;
725 case OCFS2_AST_DOWNCONVERT:
726 ocfs2_generic_handle_downconvert_action(lockres);
727 break;
728 default:
e92d57df
MF		 729 mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
730 "lockres flags = 0x%lx, unlock action: %u\n",
731 lockres->l_name, lockres->l_action, lockres->l_flags,
732 lockres->l_unlock_action);
ccd979bd
MF		 733 BUG();
734 }
735
ccd979bd
MF		 736 /* set it to something invalid so if we get called again we
737 * can catch it. */
738 lockres->l_action = OCFS2_AST_INVALID;
ccd979bd
MF		 739
740 wake_up(&lockres->l_event);
d680efe9 741 spin_unlock_irqrestore(&lockres->l_lock, flags);
ccd979bd
MF		 742}
743
ccd979bd
MF		 744static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
745 int convert)
746{
747 unsigned long flags;
748
749 mlog_entry_void();
750 spin_lock_irqsave(&lockres->l_lock, flags);
751 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
752 if (convert)
753 lockres->l_action = OCFS2_AST_INVALID;
754 else
755 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
756 spin_unlock_irqrestore(&lockres->l_lock, flags);
757
758 wake_up(&lockres->l_event);
759 mlog_exit_void();
760}
761
762/* Note: If we detect another process working on the lock (i.e.,
763 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
764 * to do the right thing in that case.
765 */
766static int ocfs2_lock_create(struct ocfs2_super *osb,
767 struct ocfs2_lock_res *lockres,
768 int level,
769 int dlm_flags)
770{
771 int ret = 0;
772 enum dlm_status status;
773 unsigned long flags;
774
775 mlog_entry_void();
776
777 mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level,
778 dlm_flags);
779
780 spin_lock_irqsave(&lockres->l_lock, flags);
781 if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
782 (lockres->l_flags & OCFS2_LOCK_BUSY)) {
783 spin_unlock_irqrestore(&lockres->l_lock, flags);
784 goto bail;
785 }
786
787 lockres->l_action = OCFS2_AST_ATTACH;
788 lockres->l_requested = level;
789 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
790 spin_unlock_irqrestore(&lockres->l_lock, flags);
791
792 status = dlmlock(osb->dlm,
793 level,
794 &lockres->l_lksb,
795 dlm_flags,
796 lockres->l_name,
f0681062 797 OCFS2_LOCK_ID_MAX_LEN - 1,
e92d57df 798 ocfs2_locking_ast,
ccd979bd 799 lockres,
aa2623ad 800 ocfs2_blocking_ast);
ccd979bd
MF		 801 if (status != DLM_NORMAL) {
802 ocfs2_log_dlm_error("dlmlock", status, lockres);
803 ret = -EINVAL;
804 ocfs2_recover_from_dlm_error(lockres, 1);
805 }
806
807 mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name);
808
809bail:
810 mlog_exit(ret);
811 return ret;
812}
813
814static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
815 int flag)
816{
817 unsigned long flags;
818 int ret;
819
820 spin_lock_irqsave(&lockres->l_lock, flags);
821 ret = lockres->l_flags & flag;
822 spin_unlock_irqrestore(&lockres->l_lock, flags);
823
824 return ret;
825}
826
827static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
828
829{
830 wait_event(lockres->l_event,
831 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
832}
833
834static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
835
836{
837 wait_event(lockres->l_event,
838 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
839}
840
841/* predict what lock level we'll be dropping down to on behalf
842 * of another node, and return true if the currently wanted
843 * level will be compatible with it. */
844static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
845 int wanted)
846{
847 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
848
849 return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
850}
851
852static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
853{
854 INIT_LIST_HEAD(&mw->mw_item);
855 init_completion(&mw->mw_complete);
856}
857
858static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
859{
860 wait_for_completion(&mw->mw_complete);
861 /* Re-arm the completion in case we want to wait on it again */
862 INIT_COMPLETION(mw->mw_complete);
863 return mw->mw_status;
864}
865
866static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
867 struct ocfs2_mask_waiter *mw,
868 unsigned long mask,
869 unsigned long goal)
870{
871 BUG_ON(!list_empty(&mw->mw_item));
872
873 assert_spin_locked(&lockres->l_lock);
874
875 list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
876 mw->mw_mask = mask;
877 mw->mw_goal = goal;
878}
879
880/* returns 0 if the mw that was removed was already satisfied, -EBUSY
881 * if the mask still hadn't reached its goal */
882static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
883 struct ocfs2_mask_waiter *mw)
884{
885 unsigned long flags;
886 int ret = 0;
887
888 spin_lock_irqsave(&lockres->l_lock, flags);
889 if (!list_empty(&mw->mw_item)) {
890 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
891 ret = -EBUSY;
892
893 list_del_init(&mw->mw_item);
894 init_completion(&mw->mw_complete);
895 }
896 spin_unlock_irqrestore(&lockres->l_lock, flags);
897
898 return ret;
899
900}
901
902static int ocfs2_cluster_lock(struct ocfs2_super *osb,
903 struct ocfs2_lock_res *lockres,
904 int level,
905 int lkm_flags,
906 int arg_flags)
907{
908 struct ocfs2_mask_waiter mw;
909 enum dlm_status status;
910 int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
911 int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
912 unsigned long flags;
913
914 mlog_entry_void();
915
916 ocfs2_init_mask_waiter(&mw);
917
b80fc012
MF		 918 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
919 lkm_flags |= LKM_VALBLK;
920
ccd979bd
MF		 921again:
922 wait = 0;
923
924 if (catch_signals && signal_pending(current)) {
925 ret = -ERESTARTSYS;
926 goto out;
927 }
928
929 spin_lock_irqsave(&lockres->l_lock, flags);
930
931 mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
932 "Cluster lock called on freeing lockres %s! flags "
933 "0x%lx\n", lockres->l_name, lockres->l_flags);
934
935 /* We only compare against the currently granted level
936 * here. If the lock is blocked waiting on a downconvert,
937 * we'll get caught below. */
938 if (lockres->l_flags & OCFS2_LOCK_BUSY &&
939 level > lockres->l_level) {
940 /* is someone sitting in dlm_lock? If so, wait on
941 * them. */
942 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
943 wait = 1;
944 goto unlock;
945 }
946
947 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
948 /* lock has not been created yet. */
949 spin_unlock_irqrestore(&lockres->l_lock, flags);
950
951 ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0);
952 if (ret < 0) {
953 mlog_errno(ret);
954 goto out;
955 }
956 goto again;
957 }
958
959 if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
960 !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
961 /* is the lock is currently blocked on behalf of
962 * another node */
963 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
964 wait = 1;
965 goto unlock;
966 }
967
968 if (level > lockres->l_level) {
969 if (lockres->l_action != OCFS2_AST_INVALID)
970 mlog(ML_ERROR, "lockres %s has action %u pending\n",
971 lockres->l_name, lockres->l_action);
972
973 lockres->l_action = OCFS2_AST_CONVERT;
974 lockres->l_requested = level;
975 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
976 spin_unlock_irqrestore(&lockres->l_lock, flags);
977
978 BUG_ON(level == LKM_IVMODE);
979 BUG_ON(level == LKM_NLMODE);
980
981 mlog(0, "lock %s, convert from %d to level = %d\n",
982 lockres->l_name, lockres->l_level, level);
983
984 /* call dlm_lock to upgrade lock now */
985 status = dlmlock(osb->dlm,
986 level,
987 &lockres->l_lksb,
b80fc012 988 lkm_flags|LKM_CONVERT,
ccd979bd 989 lockres->l_name,
f0681062 990 OCFS2_LOCK_ID_MAX_LEN - 1,
e92d57df 991 ocfs2_locking_ast,
ccd979bd 992 lockres,
aa2623ad 993 ocfs2_blocking_ast);
ccd979bd
MF		 994 if (status != DLM_NORMAL) {
995 if ((lkm_flags & LKM_NOQUEUE) &&
996 (status == DLM_NOTQUEUED))
997 ret = -EAGAIN;
998 else {
999 ocfs2_log_dlm_error("dlmlock", status,
1000 lockres);
1001 ret = -EINVAL;
1002 }
1003 ocfs2_recover_from_dlm_error(lockres, 1);
1004 goto out;
1005 }
1006
1007 mlog(0, "lock %s, successfull return from dlmlock\n",
1008 lockres->l_name);
1009
1010 /* At this point we've gone inside the dlm and need to
1011 * complete our work regardless. */
1012 catch_signals = 0;
1013
1014 /* wait for busy to clear and carry on */
1015 goto again;
1016 }
1017
1018 /* Ok, if we get here then we're good to go. */
1019 ocfs2_inc_holders(lockres, level);
1020
1021 ret = 0;
1022unlock:
1023 spin_unlock_irqrestore(&lockres->l_lock, flags);
1024out:
1025 /*
1026 * This is helping work around a lock inversion between the page lock
1027 * and dlm locks. One path holds the page lock while calling aops
1028 * which block acquiring dlm locks. The voting thread holds dlm
1029 * locks while acquiring page locks while down converting data locks.
1030 * This block is helping an aop path notice the inversion and back
1031 * off to unlock its page lock before trying the dlm lock again.
1032 */
1033 if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1034 mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1035 wait = 0;
1036 if (lockres_remove_mask_waiter(lockres, &mw))
1037 ret = -EAGAIN;
1038 else
1039 goto again;
1040 }
1041 if (wait) {
1042 ret = ocfs2_wait_for_mask(&mw);
1043 if (ret == 0)
1044 goto again;
1045 mlog_errno(ret);
1046 }
1047
1048 mlog_exit(ret);
1049 return ret;
1050}
1051
1052static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
1053 struct ocfs2_lock_res *lockres,
1054 int level)
1055{
1056 unsigned long flags;
1057
1058 mlog_entry_void();
1059 spin_lock_irqsave(&lockres->l_lock, flags);
1060 ocfs2_dec_holders(lockres, level);
1061 ocfs2_vote_on_unlock(osb, lockres);
1062 spin_unlock_irqrestore(&lockres->l_lock, flags);
1063 mlog_exit_void();
1064}
1065
d680efe9
MF		 1066int ocfs2_create_new_lock(struct ocfs2_super *osb,
1067 struct ocfs2_lock_res *lockres,
24c19ef4
MF		 1068 int ex,
1069 int local)
ccd979bd 1070{
d680efe9 1071 int level = ex ? LKM_EXMODE : LKM_PRMODE;
ccd979bd 1072 unsigned long flags;
24c19ef4 1073 int lkm_flags = local ? LKM_LOCAL : 0;
ccd979bd
MF		 1074
1075 spin_lock_irqsave(&lockres->l_lock, flags);
1076 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1077 lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1078 spin_unlock_irqrestore(&lockres->l_lock, flags);
1079
24c19ef4 1080 return ocfs2_lock_create(osb, lockres, level, lkm_flags);
ccd979bd
MF		 1081}
1082
1083/* Grants us an EX lock on the data and metadata resources, skipping
1084 * the normal cluster directory lookup. Use this ONLY on newly created
1085 * inodes which other nodes can't possibly see, and which haven't been
1086 * hashed in the inode hash yet. This can give us a good performance
1087 * increase as it'll skip the network broadcast normally associated
1088 * with creating a new lock resource. */
1089int ocfs2_create_new_inode_locks(struct inode *inode)
1090{
1091 int ret;
d680efe9 1092 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1093
1094 BUG_ON(!inode);
1095 BUG_ON(!ocfs2_inode_is_new(inode));
1096
1097 mlog_entry_void();
1098
b0697053 1099 mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 1100
1101 /* NOTE: That we don't increment any of the holder counts, nor
1102 * do we add anything to a journal handle. Since this is
1103 * supposed to be a new inode which the cluster doesn't know
1104 * about yet, there is no need to. As far as the LVB handling
1105 * is concerned, this is basically like acquiring an EX lock
1106 * on a resource which has an invalid one -- we'll set it
1107 * valid when we release the EX. */
1108
24c19ef4 1109 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
ccd979bd
MF		 1110 if (ret) {
1111 mlog_errno(ret);
1112 goto bail;
1113 }
1114
24c19ef4
MF		 1115 /*
1116 * We don't want to use LKM_LOCAL on a meta data lock as they
1117 * don't use a generation in their lock names.
1118 */
1119 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_meta_lockres, 1, 0);
ccd979bd
MF		 1120 if (ret) {
1121 mlog_errno(ret);
1122 goto bail;
1123 }
1124
24c19ef4 1125 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_data_lockres, 1, 1);
ccd979bd
MF		 1126 if (ret) {
1127 mlog_errno(ret);
1128 goto bail;
1129 }
1130
1131bail:
1132 mlog_exit(ret);
1133 return ret;
1134}
1135
1136int ocfs2_rw_lock(struct inode *inode, int write)
1137{
1138 int status, level;
1139 struct ocfs2_lock_res *lockres;
1140
1141 BUG_ON(!inode);
1142
1143 mlog_entry_void();
1144
b0697053
MF		 1145 mlog(0, "inode %llu take %s RW lock\n",
1146 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1147 write ? "EXMODE" : "PRMODE");
1148
1149 lockres = &OCFS2_I(inode)->ip_rw_lockres;
1150
1151 level = write ? LKM_EXMODE : LKM_PRMODE;
1152
1153 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1154 0);
1155 if (status < 0)
1156 mlog_errno(status);
1157
1158 mlog_exit(status);
1159 return status;
1160}
1161
1162void ocfs2_rw_unlock(struct inode *inode, int write)
1163{
1164 int level = write ? LKM_EXMODE : LKM_PRMODE;
1165 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
1166
1167 mlog_entry_void();
1168
b0697053
MF		 1169 mlog(0, "inode %llu drop %s RW lock\n",
1170 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1171 write ? "EXMODE" : "PRMODE");
1172
1173 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1174
1175 mlog_exit_void();
1176}
1177
1178int ocfs2_data_lock_full(struct inode *inode,
1179 int write,
1180 int arg_flags)
1181{
1182 int status = 0, level;
1183 struct ocfs2_lock_res *lockres;
1184
1185 BUG_ON(!inode);
1186
1187 mlog_entry_void();
1188
b0697053
MF		 1189 mlog(0, "inode %llu take %s DATA lock\n",
1190 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1191 write ? "EXMODE" : "PRMODE");
1192
1193 /* We'll allow faking a readonly data lock for
1194 * rodevices. */
1195 if (ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb))) {
1196 if (write) {
1197 status = -EROFS;
1198 mlog_errno(status);
1199 }
1200 goto out;
1201 }
1202
1203 lockres = &OCFS2_I(inode)->ip_data_lockres;
1204
1205 level = write ? LKM_EXMODE : LKM_PRMODE;
1206
1207 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level,
1208 0, arg_flags);
1209 if (status < 0 && status != -EAGAIN)
1210 mlog_errno(status);
1211
1212out:
1213 mlog_exit(status);
1214 return status;
1215}
1216
1217/* see ocfs2_meta_lock_with_page() */
1218int ocfs2_data_lock_with_page(struct inode *inode,
1219 int write,
1220 struct page *page)
1221{
1222 int ret;
1223
1224 ret = ocfs2_data_lock_full(inode, write, OCFS2_LOCK_NONBLOCK);
1225 if (ret == -EAGAIN) {
1226 unlock_page(page);
1227 if (ocfs2_data_lock(inode, write) == 0)
1228 ocfs2_data_unlock(inode, write);
1229 ret = AOP_TRUNCATED_PAGE;
1230 }
1231
1232 return ret;
1233}
1234
1235static void ocfs2_vote_on_unlock(struct ocfs2_super *osb,
1236 struct ocfs2_lock_res *lockres)
1237{
1238 int kick = 0;
1239
1240 mlog_entry_void();
1241
1242 /* If we know that another node is waiting on our lock, kick
1243 * the vote thread * pre-emptively when we reach a release
1244 * condition. */
1245 if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1246 switch(lockres->l_blocking) {
1247 case LKM_EXMODE:
1248 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1249 kick = 1;
1250 break;
1251 case LKM_PRMODE:
1252 if (!lockres->l_ex_holders)
1253 kick = 1;
1254 break;
1255 default:
1256 BUG();
1257 }
1258 }
1259
1260 if (kick)
1261 ocfs2_kick_vote_thread(osb);
1262
1263 mlog_exit_void();
1264}
1265
1266void ocfs2_data_unlock(struct inode *inode,
1267 int write)
1268{
1269 int level = write ? LKM_EXMODE : LKM_PRMODE;
1270 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_data_lockres;
1271
1272 mlog_entry_void();
1273
b0697053
MF		 1274 mlog(0, "inode %llu drop %s DATA lock\n",
1275 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1276 write ? "EXMODE" : "PRMODE");
1277
1278 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)))
1279 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1280
1281 mlog_exit_void();
1282}
1283
1284#define OCFS2_SEC_BITS 34
1285#define OCFS2_SEC_SHIFT (64 - 34)
1286#define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1)
1287
1288/* LVB only has room for 64 bits of time here so we pack it for
1289 * now. */
1290static u64 ocfs2_pack_timespec(struct timespec *spec)
1291{
1292 u64 res;
1293 u64 sec = spec->tv_sec;
1294 u32 nsec = spec->tv_nsec;
1295
1296 res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
1297
1298 return res;
1299}
1300
1301/* Call this with the lockres locked. I am reasonably sure we don't
1302 * need ip_lock in this function as anyone who would be changing those
1303 * values is supposed to be blocked in ocfs2_meta_lock right now. */
1304static void __ocfs2_stuff_meta_lvb(struct inode *inode)
1305{
1306 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1307 struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
1308 struct ocfs2_meta_lvb *lvb;
1309
1310 mlog_entry_void();
1311
1312 lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1313
24c19ef4
MF		 1314 /*
1315 * Invalidate the LVB of a deleted inode - this way other
1316 * nodes are forced to go to disk and discover the new inode
1317 * status.
1318 */
1319 if (oi->ip_flags & OCFS2_INODE_DELETED) {
1320 lvb->lvb_version = 0;
1321 goto out;
1322 }
1323
4d3b83f7 1324 lvb->lvb_version = OCFS2_LVB_VERSION;
ccd979bd
MF		 1325 lvb->lvb_isize = cpu_to_be64(i_size_read(inode));
1326 lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
1327 lvb->lvb_iuid = cpu_to_be32(inode->i_uid);
1328 lvb->lvb_igid = cpu_to_be32(inode->i_gid);
1329 lvb->lvb_imode = cpu_to_be16(inode->i_mode);
1330 lvb->lvb_inlink = cpu_to_be16(inode->i_nlink);
1331 lvb->lvb_iatime_packed =
1332 cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
1333 lvb->lvb_ictime_packed =
1334 cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
1335 lvb->lvb_imtime_packed =
1336 cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
ca4d147e 1337 lvb->lvb_iattr = cpu_to_be32(oi->ip_attr);
f9e2d82e 1338 lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
ccd979bd 1339
24c19ef4 1340out:
ccd979bd
MF		 1341 mlog_meta_lvb(0, lockres);
1342
1343 mlog_exit_void();
1344}
1345
1346static void ocfs2_unpack_timespec(struct timespec *spec,
1347 u64 packed_time)
1348{
1349 spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
1350 spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
1351}
1352
1353static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
1354{
1355 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1356 struct ocfs2_lock_res *lockres = &oi->ip_meta_lockres;
1357 struct ocfs2_meta_lvb *lvb;
1358
1359 mlog_entry_void();
1360
1361 mlog_meta_lvb(0, lockres);
1362
1363 lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1364
1365 /* We're safe here without the lockres lock... */
1366 spin_lock(&oi->ip_lock);
1367 oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
1368 i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
1369
ca4d147e
HP		 1370 oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
1371 ocfs2_set_inode_flags(inode);
1372
ccd979bd
MF		 1373 /* fast-symlinks are a special case */
1374 if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
1375 inode->i_blocks = 0;
1376 else
1377 inode->i_blocks =
1378 ocfs2_align_bytes_to_sectors(i_size_read(inode));
1379
1380 inode->i_uid = be32_to_cpu(lvb->lvb_iuid);
1381 inode->i_gid = be32_to_cpu(lvb->lvb_igid);
1382 inode->i_mode = be16_to_cpu(lvb->lvb_imode);
1383 inode->i_nlink = be16_to_cpu(lvb->lvb_inlink);
1384 ocfs2_unpack_timespec(&inode->i_atime,
1385 be64_to_cpu(lvb->lvb_iatime_packed));
1386 ocfs2_unpack_timespec(&inode->i_mtime,
1387 be64_to_cpu(lvb->lvb_imtime_packed));
1388 ocfs2_unpack_timespec(&inode->i_ctime,
1389 be64_to_cpu(lvb->lvb_ictime_packed));
1390 spin_unlock(&oi->ip_lock);
1391
1392 mlog_exit_void();
1393}
1394
f9e2d82e
MF		 1395static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
1396 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 1397{
1398 struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
1399
f9e2d82e
MF		 1400 if (lvb->lvb_version == OCFS2_LVB_VERSION
1401 && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
ccd979bd
MF		 1402 return 1;
1403 return 0;
1404}
1405
1406/* Determine whether a lock resource needs to be refreshed, and
1407 * arbitrate who gets to refresh it.
1408 *
1409 * 0 means no refresh needed.
1410 *
1411 * > 0 means you need to refresh this and you MUST call
1412 * ocfs2_complete_lock_res_refresh afterwards. */
1413static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
1414{
1415 unsigned long flags;
1416 int status = 0;
1417
1418 mlog_entry_void();
1419
1420refresh_check:
1421 spin_lock_irqsave(&lockres->l_lock, flags);
1422 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
1423 spin_unlock_irqrestore(&lockres->l_lock, flags);
1424 goto bail;
1425 }
1426
1427 if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
1428 spin_unlock_irqrestore(&lockres->l_lock, flags);
1429
1430 ocfs2_wait_on_refreshing_lock(lockres);
1431 goto refresh_check;
1432 }
1433
1434 /* Ok, I'll be the one to refresh this lock. */
1435 lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
1436 spin_unlock_irqrestore(&lockres->l_lock, flags);
1437
1438 status = 1;
1439bail:
1440 mlog_exit(status);
1441 return status;
1442}
1443
1444/* If status is non zero, I'll mark it as not being in refresh
1445 * anymroe, but i won't clear the needs refresh flag. */
1446static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
1447 int status)
1448{
1449 unsigned long flags;
1450 mlog_entry_void();
1451
1452 spin_lock_irqsave(&lockres->l_lock, flags);
1453 lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
1454 if (!status)
1455 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
1456 spin_unlock_irqrestore(&lockres->l_lock, flags);
1457
1458 wake_up(&lockres->l_event);
1459
1460 mlog_exit_void();
1461}
1462
1463/* may or may not return a bh if it went to disk. */
1464static int ocfs2_meta_lock_update(struct inode *inode,
1465 struct buffer_head **bh)
1466{
1467 int status = 0;
1468 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1469 struct ocfs2_lock_res *lockres;
1470 struct ocfs2_dinode *fe;
1471
1472 mlog_entry_void();
1473
1474 spin_lock(&oi->ip_lock);
1475 if (oi->ip_flags & OCFS2_INODE_DELETED) {
b0697053 1476 mlog(0, "Orphaned inode %llu was deleted while we "
ccd979bd 1477 "were waiting on a lock. ip_flags = 0x%x\n",
b0697053 1478 (unsigned long long)oi->ip_blkno, oi->ip_flags);
ccd979bd
MF		 1479 spin_unlock(&oi->ip_lock);
1480 status = -ENOENT;
1481 goto bail;
1482 }
1483 spin_unlock(&oi->ip_lock);
1484
1485 lockres = &oi->ip_meta_lockres;
1486
1487 if (!ocfs2_should_refresh_lock_res(lockres))
1488 goto bail;
1489
1490 /* This will discard any caching information we might have had
1491 * for the inode metadata. */
1492 ocfs2_metadata_cache_purge(inode);
1493
1494 /* will do nothing for inode types that don't use the extent
1495 * map (directories, bitmap files, etc) */
1496 ocfs2_extent_map_trunc(inode, 0);
1497
f9e2d82e 1498 if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
b0697053
MF		 1499 mlog(0, "Trusting LVB on inode %llu\n",
1500 (unsigned long long)oi->ip_blkno);
ccd979bd
MF		 1501 ocfs2_refresh_inode_from_lvb(inode);
1502 } else {
1503 /* Boo, we have to go to disk. */
1504 /* read bh, cast, ocfs2_refresh_inode */
1505 status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno,
1506 bh, OCFS2_BH_CACHED, inode);
1507 if (status < 0) {
1508 mlog_errno(status);
1509 goto bail_refresh;
1510 }
1511 fe = (struct ocfs2_dinode *) (*bh)->b_data;
1512
1513 /* This is a good chance to make sure we're not
1514 * locking an invalid object.
1515 *
1516 * We bug on a stale inode here because we checked
1517 * above whether it was wiped from disk. The wiping
1518 * node provides a guarantee that we receive that
1519 * message and can mark the inode before dropping any
1520 * locks associated with it. */
1521 if (!OCFS2_IS_VALID_DINODE(fe)) {
1522 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
1523 status = -EIO;
1524 goto bail_refresh;
1525 }
1526 mlog_bug_on_msg(inode->i_generation !=
1527 le32_to_cpu(fe->i_generation),
b0697053 1528 "Invalid dinode %llu disk generation: %u "
ccd979bd 1529 "inode->i_generation: %u\n",
b0697053
MF		 1530 (unsigned long long)oi->ip_blkno,
1531 le32_to_cpu(fe->i_generation),
ccd979bd
MF		 1532 inode->i_generation);
1533 mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
1534 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
b0697053
MF		 1535 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
1536 (unsigned long long)oi->ip_blkno,
1537 (unsigned long long)le64_to_cpu(fe->i_dtime),
ccd979bd
MF		 1538 le32_to_cpu(fe->i_flags));
1539
1540 ocfs2_refresh_inode(inode, fe);
1541 }
1542
1543 status = 0;
1544bail_refresh:
1545 ocfs2_complete_lock_res_refresh(lockres, status);
1546bail:
1547 mlog_exit(status);
1548 return status;
1549}
1550
1551static int ocfs2_assign_bh(struct inode *inode,
1552 struct buffer_head **ret_bh,
1553 struct buffer_head *passed_bh)
1554{
1555 int status;
1556
1557 if (passed_bh) {
1558 /* Ok, the update went to disk for us, use the
1559 * returned bh. */
1560 *ret_bh = passed_bh;
1561 get_bh(*ret_bh);
1562
1563 return 0;
1564 }
1565
1566 status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1567 OCFS2_I(inode)->ip_blkno,
1568 ret_bh,
1569 OCFS2_BH_CACHED,
1570 inode);
1571 if (status < 0)
1572 mlog_errno(status);
1573
1574 return status;
1575}
1576
1577/*
1578 * returns < 0 error if the callback will never be called, otherwise
1579 * the result of the lock will be communicated via the callback.
1580 */
1581int ocfs2_meta_lock_full(struct inode *inode,
1582 struct ocfs2_journal_handle *handle,
1583 struct buffer_head **ret_bh,
1584 int ex,
1585 int arg_flags)
1586{
1587 int status, level, dlm_flags, acquired;
1588 struct ocfs2_lock_res *lockres;
1589 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1590 struct buffer_head *local_bh = NULL;
1591
1592 BUG_ON(!inode);
1593
1594 mlog_entry_void();
1595
b0697053
MF		 1596 mlog(0, "inode %llu, take %s META lock\n",
1597 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1598 ex ? "EXMODE" : "PRMODE");
1599
1600 status = 0;
1601 acquired = 0;
1602 /* We'll allow faking a readonly metadata lock for
1603 * rodevices. */
1604 if (ocfs2_is_hard_readonly(osb)) {
1605 if (ex)
1606 status = -EROFS;
1607 goto bail;
1608 }
1609
1610 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1611 wait_event(osb->recovery_event,
1612 ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1613
1614 acquired = 0;
1615 lockres = &OCFS2_I(inode)->ip_meta_lockres;
1616 level = ex ? LKM_EXMODE : LKM_PRMODE;
1617 dlm_flags = 0;
1618 if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
1619 dlm_flags |= LKM_NOQUEUE;
1620
1621 status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
1622 if (status < 0) {
1623 if (status != -EAGAIN && status != -EIOCBRETRY)
1624 mlog_errno(status);
1625 goto bail;
1626 }
1627
1628 /* Notify the error cleanup path to drop the cluster lock. */
1629 acquired = 1;
1630
1631 /* We wait twice because a node may have died while we were in
1632 * the lower dlm layers. The second time though, we've
1633 * committed to owning this lock so we don't allow signals to
1634 * abort the operation. */
1635 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
1636 wait_event(osb->recovery_event,
1637 ocfs2_node_map_is_empty(osb, &osb->recovery_map));
1638
24c19ef4
MF		 1639 /*
1640 * We only see this flag if we're being called from
1641 * ocfs2_read_locked_inode(). It means we're locking an inode
1642 * which hasn't been populated yet, so clear the refresh flag
1643 * and let the caller handle it.
1644 */
1645 if (inode->i_state & I_NEW) {
1646 status = 0;
1647 ocfs2_complete_lock_res_refresh(lockres, 0);
1648 goto bail;
1649 }
1650
ccd979bd
MF		 1651 /* This is fun. The caller may want a bh back, or it may
1652 * not. ocfs2_meta_lock_update definitely wants one in, but
1653 * may or may not read one, depending on what's in the
1654 * LVB. The result of all of this is that we've *only* gone to
1655 * disk if we have to, so the complexity is worthwhile. */
1656 status = ocfs2_meta_lock_update(inode, &local_bh);
1657 if (status < 0) {
1658 if (status != -ENOENT)
1659 mlog_errno(status);
1660 goto bail;
1661 }
1662
1663 if (ret_bh) {
1664 status = ocfs2_assign_bh(inode, ret_bh, local_bh);
1665 if (status < 0) {
1666 mlog_errno(status);
1667 goto bail;
1668 }
1669 }
1670
1671 if (handle) {
1672 status = ocfs2_handle_add_lock(handle, inode);
1673 if (status < 0)
1674 mlog_errno(status);
1675 }
1676
1677bail:
1678 if (status < 0) {
1679 if (ret_bh && (*ret_bh)) {
1680 brelse(*ret_bh);
1681 *ret_bh = NULL;
1682 }
1683 if (acquired)
1684 ocfs2_meta_unlock(inode, ex);
1685 }
1686
1687 if (local_bh)
1688 brelse(local_bh);
1689
1690 mlog_exit(status);
1691 return status;
1692}
1693
1694/*
1695 * This is working around a lock inversion between tasks acquiring DLM locks
1696 * while holding a page lock and the vote thread which blocks dlm lock acquiry
1697 * while acquiring page locks.
1698 *
1699 * ** These _with_page variantes are only intended to be called from aop
1700 * methods that hold page locks and return a very specific *positive* error
1701 * code that aop methods pass up to the VFS -- test for errors with != 0. **
1702 *
1703 * The DLM is called such that it returns -EAGAIN if it would have blocked
1704 * waiting for the vote thread. In that case we unlock our page so the vote
1705 * thread can make progress. Once we've done this we have to return
1706 * AOP_TRUNCATED_PAGE so the aop method that called us can bubble that back up
1707 * into the VFS who will then immediately retry the aop call.
1708 *
1709 * We do a blocking lock and immediate unlock before returning, though, so that
1710 * the lock has a great chance of being cached on this node by the time the VFS
1711 * calls back to retry the aop. This has a potential to livelock as nodes
1712 * ping locks back and forth, but that's a risk we're willing to take to avoid
1713 * the lock inversion simply.
1714 */
1715int ocfs2_meta_lock_with_page(struct inode *inode,
1716 struct ocfs2_journal_handle *handle,
1717 struct buffer_head **ret_bh,
1718 int ex,
1719 struct page *page)
1720{
1721 int ret;
1722
1723 ret = ocfs2_meta_lock_full(inode, handle, ret_bh, ex,
1724 OCFS2_LOCK_NONBLOCK);
1725 if (ret == -EAGAIN) {
1726 unlock_page(page);
1727 if (ocfs2_meta_lock(inode, handle, ret_bh, ex) == 0)
1728 ocfs2_meta_unlock(inode, ex);
1729 ret = AOP_TRUNCATED_PAGE;
1730 }
1731
1732 return ret;
1733}
1734
1735void ocfs2_meta_unlock(struct inode *inode,
1736 int ex)
1737{
1738 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1739 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres;
1740
1741 mlog_entry_void();
1742
b0697053
MF		 1743 mlog(0, "inode %llu drop %s META lock\n",
1744 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1745 ex ? "EXMODE" : "PRMODE");
1746
1747 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)))
1748 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
1749
1750 mlog_exit_void();
1751}
1752
1753int ocfs2_super_lock(struct ocfs2_super *osb,
1754 int ex)
1755{
1756 int status;
1757 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1758 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1759 struct buffer_head *bh;
1760 struct ocfs2_slot_info *si = osb->slot_info;
1761
1762 mlog_entry_void();
1763
1764 if (ocfs2_is_hard_readonly(osb))
1765 return -EROFS;
1766
1767 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
1768 if (status < 0) {
1769 mlog_errno(status);
1770 goto bail;
1771 }
1772
1773 /* The super block lock path is really in the best position to
1774 * know when resources covered by the lock need to be
1775 * refreshed, so we do it here. Of course, making sense of
1776 * everything is up to the caller :) */
1777 status = ocfs2_should_refresh_lock_res(lockres);
1778 if (status < 0) {
1779 mlog_errno(status);
1780 goto bail;
1781 }
1782 if (status) {
1783 bh = si->si_bh;
1784 status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0,
1785 si->si_inode);
1786 if (status == 0)
1787 ocfs2_update_slot_info(si);
1788
1789 ocfs2_complete_lock_res_refresh(lockres, status);
1790
1791 if (status < 0)
1792 mlog_errno(status);
1793 }
1794bail:
1795 mlog_exit(status);
1796 return status;
1797}
1798
1799void ocfs2_super_unlock(struct ocfs2_super *osb,
1800 int ex)
1801{
1802 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1803 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
1804
1805 ocfs2_cluster_unlock(osb, lockres, level);
1806}
1807
1808int ocfs2_rename_lock(struct ocfs2_super *osb)
1809{
1810 int status;
1811 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1812
1813 if (ocfs2_is_hard_readonly(osb))
1814 return -EROFS;
1815
1816 status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0);
1817 if (status < 0)
1818 mlog_errno(status);
1819
1820 return status;
1821}
1822
1823void ocfs2_rename_unlock(struct ocfs2_super *osb)
1824{
1825 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
1826
1827 ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE);
1828}
1829
d680efe9
MF		 1830int ocfs2_dentry_lock(struct dentry *dentry, int ex)
1831{
1832 int ret;
1833 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1834 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
1835 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
1836
1837 BUG_ON(!dl);
1838
1839 if (ocfs2_is_hard_readonly(osb))
1840 return -EROFS;
1841
1842 ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
1843 if (ret < 0)
1844 mlog_errno(ret);
1845
1846 return ret;
1847}
1848
1849void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
1850{
1851 int level = ex ? LKM_EXMODE : LKM_PRMODE;
1852 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
1853 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
1854
1855 ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
1856}
1857
ccd979bd
MF		 1858/* Reference counting of the dlm debug structure. We want this because
1859 * open references on the debug inodes can live on after a mount, so
1860 * we can't rely on the ocfs2_super to always exist. */
1861static void ocfs2_dlm_debug_free(struct kref *kref)
1862{
1863 struct ocfs2_dlm_debug *dlm_debug;
1864
1865 dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
1866
1867 kfree(dlm_debug);
1868}
1869
1870void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
1871{
1872 if (dlm_debug)
1873 kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
1874}
1875
1876static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
1877{
1878 kref_get(&debug->d_refcnt);
1879}
1880
1881struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
1882{
1883 struct ocfs2_dlm_debug *dlm_debug;
1884
1885 dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
1886 if (!dlm_debug) {
1887 mlog_errno(-ENOMEM);
1888 goto out;
1889 }
1890
1891 kref_init(&dlm_debug->d_refcnt);
1892 INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
1893 dlm_debug->d_locking_state = NULL;
1894out:
1895 return dlm_debug;
1896}
1897
1898/* Access to this is arbitrated for us via seq_file->sem. */
1899struct ocfs2_dlm_seq_priv {
1900 struct ocfs2_dlm_debug *p_dlm_debug;
1901 struct ocfs2_lock_res p_iter_res;
1902 struct ocfs2_lock_res p_tmp_res;
1903};
1904
1905static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
1906 struct ocfs2_dlm_seq_priv *priv)
1907{
1908 struct ocfs2_lock_res *iter, *ret = NULL;
1909 struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
1910
1911 assert_spin_locked(&ocfs2_dlm_tracking_lock);
1912
1913 list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
1914 /* discover the head of the list */
1915 if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
1916 mlog(0, "End of list found, %p\n", ret);
1917 break;
1918 }
1919
1920 /* We track our "dummy" iteration lockres' by a NULL
1921 * l_ops field. */
1922 if (iter->l_ops != NULL) {
1923 ret = iter;
1924 break;
1925 }
1926 }
1927
1928 return ret;
1929}
1930
1931static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
1932{
1933 struct ocfs2_dlm_seq_priv *priv = m->private;
1934 struct ocfs2_lock_res *iter;
1935
1936 spin_lock(&ocfs2_dlm_tracking_lock);
1937 iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
1938 if (iter) {
1939 /* Since lockres' have the lifetime of their container
1940 * (which can be inodes, ocfs2_supers, etc) we want to
1941 * copy this out to a temporary lockres while still
1942 * under the spinlock. Obviously after this we can't
1943 * trust any pointers on the copy returned, but that's
1944 * ok as the information we want isn't typically held
1945 * in them. */
1946 priv->p_tmp_res = *iter;
1947 iter = &priv->p_tmp_res;
1948 }
1949 spin_unlock(&ocfs2_dlm_tracking_lock);
1950
1951 return iter;
1952}
1953
1954static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
1955{
1956}
1957
1958static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
1959{
1960 struct ocfs2_dlm_seq_priv *priv = m->private;
1961 struct ocfs2_lock_res *iter = v;
1962 struct ocfs2_lock_res *dummy = &priv->p_iter_res;
1963
1964 spin_lock(&ocfs2_dlm_tracking_lock);
1965 iter = ocfs2_dlm_next_res(iter, priv);
1966 list_del_init(&dummy->l_debug_list);
1967 if (iter) {
1968 list_add(&dummy->l_debug_list, &iter->l_debug_list);
1969 priv->p_tmp_res = *iter;
1970 iter = &priv->p_tmp_res;
1971 }
1972 spin_unlock(&ocfs2_dlm_tracking_lock);
1973
1974 return iter;
1975}
1976
1977/* So that debugfs.ocfs2 can determine which format is being used */
1978#define OCFS2_DLM_DEBUG_STR_VERSION 1
1979static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
1980{
1981 int i;
1982 char *lvb;
1983 struct ocfs2_lock_res *lockres = v;
1984
1985 if (!lockres)
1986 return -EINVAL;
1987
d680efe9
MF		 1988 seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
1989
1990 if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
1991 seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
1992 lockres->l_name,
1993 (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
1994 else
1995 seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
1996
1997 seq_printf(m, "%d\t"
ccd979bd
MF		 1998 "0x%lx\t"
1999 "0x%x\t"
2000 "0x%x\t"
2001 "%u\t"
2002 "%u\t"
2003 "%d\t"
2004 "%d\t",
ccd979bd
MF		 2005 lockres->l_level,
2006 lockres->l_flags,
2007 lockres->l_action,
2008 lockres->l_unlock_action,
2009 lockres->l_ro_holders,
2010 lockres->l_ex_holders,
2011 lockres->l_requested,
2012 lockres->l_blocking);
2013
2014 /* Dump the raw LVB */
2015 lvb = lockres->l_lksb.lvb;
2016 for(i = 0; i < DLM_LVB_LEN; i++)
2017 seq_printf(m, "0x%x\t", lvb[i]);
2018
2019 /* End the line */
2020 seq_printf(m, "\n");
2021 return 0;
2022}
2023
2024static struct seq_operations ocfs2_dlm_seq_ops = {
2025 .start = ocfs2_dlm_seq_start,
2026 .stop = ocfs2_dlm_seq_stop,
2027 .next = ocfs2_dlm_seq_next,
2028 .show = ocfs2_dlm_seq_show,
2029};
2030
2031static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2032{
2033 struct seq_file *seq = (struct seq_file *) file->private_data;
2034 struct ocfs2_dlm_seq_priv *priv = seq->private;
2035 struct ocfs2_lock_res *res = &priv->p_iter_res;
2036
2037 ocfs2_remove_lockres_tracking(res);
2038 ocfs2_put_dlm_debug(priv->p_dlm_debug);
2039 return seq_release_private(inode, file);
2040}
2041
2042static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2043{
2044 int ret;
2045 struct ocfs2_dlm_seq_priv *priv;
2046 struct seq_file *seq;
2047 struct ocfs2_super *osb;
2048
2049 priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2050 if (!priv) {
2051 ret = -ENOMEM;
2052 mlog_errno(ret);
2053 goto out;
2054 }
8e18e294 2055 osb = inode->i_private;
ccd979bd
MF		 2056 ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2057 priv->p_dlm_debug = osb->osb_dlm_debug;
2058 INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2059
2060 ret = seq_open(file, &ocfs2_dlm_seq_ops);
2061 if (ret) {
2062 kfree(priv);
2063 mlog_errno(ret);
2064 goto out;
2065 }
2066
2067 seq = (struct seq_file *) file->private_data;
2068 seq->private = priv;
2069
2070 ocfs2_add_lockres_tracking(&priv->p_iter_res,
2071 priv->p_dlm_debug);
2072
2073out:
2074 return ret;
2075}
2076
4b6f5d20 2077static const struct file_operations ocfs2_dlm_debug_fops = {
ccd979bd
MF		 2078 .open = ocfs2_dlm_debug_open,
2079 .release = ocfs2_dlm_debug_release,
2080 .read = seq_read,
2081 .llseek = seq_lseek,
2082};
2083
2084static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2085{
2086 int ret = 0;
2087 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2088
2089 dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2090 S_IFREG|S_IRUSR,
2091 osb->osb_debug_root,
2092 osb,
2093 &ocfs2_dlm_debug_fops);
2094 if (!dlm_debug->d_locking_state) {
2095 ret = -EINVAL;
2096 mlog(ML_ERROR,
2097 "Unable to create locking state debugfs file.\n");
2098 goto out;
2099 }
2100
2101 ocfs2_get_dlm_debug(dlm_debug);
2102out:
2103 return ret;
2104}
2105
2106static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2107{
2108 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2109
2110 if (dlm_debug) {
2111 debugfs_remove(dlm_debug->d_locking_state);
2112 ocfs2_put_dlm_debug(dlm_debug);
2113 }
2114}
2115
2116int ocfs2_dlm_init(struct ocfs2_super *osb)
2117{
2118 int status;
2119 u32 dlm_key;
2120 struct dlm_ctxt *dlm;
2121
2122 mlog_entry_void();
2123
2124 status = ocfs2_dlm_init_debug(osb);
2125 if (status < 0) {
2126 mlog_errno(status);
2127 goto bail;
2128 }
2129
2130 /* launch vote thread */
78427043 2131 osb->vote_task = kthread_run(ocfs2_vote_thread, osb, "ocfs2vote");
ccd979bd
MF		 2132 if (IS_ERR(osb->vote_task)) {
2133 status = PTR_ERR(osb->vote_task);
2134 osb->vote_task = NULL;
2135 mlog_errno(status);
2136 goto bail;
2137 }
2138
2139 /* used by the dlm code to make message headers unique, each
2140 * node in this domain must agree on this. */
2141 dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str));
2142
2143 /* for now, uuid == domain */
2144 dlm = dlm_register_domain(osb->uuid_str, dlm_key);
2145 if (IS_ERR(dlm)) {
2146 status = PTR_ERR(dlm);
2147 mlog_errno(status);
2148 goto bail;
2149 }
2150
2151 ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
2152 ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
2153
2154 dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb);
2155
2156 osb->dlm = dlm;
2157
2158 status = 0;
2159bail:
2160 if (status < 0) {
2161 ocfs2_dlm_shutdown_debug(osb);
2162 if (osb->vote_task)
2163 kthread_stop(osb->vote_task);
2164 }
2165
2166 mlog_exit(status);
2167 return status;
2168}
2169
2170void ocfs2_dlm_shutdown(struct ocfs2_super *osb)
2171{
2172 mlog_entry_void();
2173
2174 dlm_unregister_eviction_cb(&osb->osb_eviction_cb);
2175
2176 ocfs2_drop_osb_locks(osb);
2177
2178 if (osb->vote_task) {
2179 kthread_stop(osb->vote_task);
2180 osb->vote_task = NULL;
2181 }
2182
2183 ocfs2_lock_res_free(&osb->osb_super_lockres);
2184 ocfs2_lock_res_free(&osb->osb_rename_lockres);
2185
2186 dlm_unregister_domain(osb->dlm);
2187 osb->dlm = NULL;
2188
2189 ocfs2_dlm_shutdown_debug(osb);
2190
2191 mlog_exit_void();
2192}
2193
2a45f2d1 2194static void ocfs2_unlock_ast(void *opaque, enum dlm_status status)
ccd979bd
MF		 2195{
2196 struct ocfs2_lock_res *lockres = opaque;
2197 unsigned long flags;
2198
2199 mlog_entry_void();
2200
2201 mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
2202 lockres->l_unlock_action);
2203
2204 spin_lock_irqsave(&lockres->l_lock, flags);
2205 /* We tried to cancel a convert request, but it was already
2206 * granted. All we want to do here is clear our unlock
2207 * state. The wake_up call done at the bottom is redundant
2208 * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't
2209 * hurt anything anyway */
2210 if (status == DLM_CANCELGRANT &&
2211 lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2212 mlog(0, "Got cancelgrant for %s\n", lockres->l_name);
2213
2214 /* We don't clear the busy flag in this case as it
2215 * should have been cleared by the ast which the dlm
2216 * has called. */
2217 goto complete_unlock;
2218 }
2219
2220 if (status != DLM_NORMAL) {
2221 mlog(ML_ERROR, "Dlm passes status %d for lock %s, "
2222 "unlock_action %d\n", status, lockres->l_name,
2223 lockres->l_unlock_action);
2224 spin_unlock_irqrestore(&lockres->l_lock, flags);
2225 return;
2226 }
2227
2228 switch(lockres->l_unlock_action) {
2229 case OCFS2_UNLOCK_CANCEL_CONVERT:
2230 mlog(0, "Cancel convert success for %s\n", lockres->l_name);
2231 lockres->l_action = OCFS2_AST_INVALID;
2232 break;
2233 case OCFS2_UNLOCK_DROP_LOCK:
2234 lockres->l_level = LKM_IVMODE;
2235 break;
2236 default:
2237 BUG();
2238 }
2239
2240 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
2241complete_unlock:
2242 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
2243 spin_unlock_irqrestore(&lockres->l_lock, flags);
2244
2245 wake_up(&lockres->l_event);
2246
2247 mlog_exit_void();
2248}
2249
ccd979bd 2250static int ocfs2_drop_lock(struct ocfs2_super *osb,
0d5dc6c2 2251 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 2252{
2253 enum dlm_status status;
2254 unsigned long flags;
b80fc012 2255 int lkm_flags = 0;
ccd979bd
MF		 2256
2257 /* We didn't get anywhere near actually using this lockres. */
2258 if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
2259 goto out;
2260
b80fc012
MF		 2261 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
2262 lkm_flags |= LKM_VALBLK;
2263
ccd979bd
MF		 2264 spin_lock_irqsave(&lockres->l_lock, flags);
2265
2266 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
2267 "lockres %s, flags 0x%lx\n",
2268 lockres->l_name, lockres->l_flags);
2269
2270 while (lockres->l_flags & OCFS2_LOCK_BUSY) {
2271 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2272 "%u, unlock_action = %u\n",
2273 lockres->l_name, lockres->l_flags, lockres->l_action,
2274 lockres->l_unlock_action);
2275
2276 spin_unlock_irqrestore(&lockres->l_lock, flags);
2277
2278 /* XXX: Today we just wait on any busy
2279 * locks... Perhaps we need to cancel converts in the
2280 * future? */
2281 ocfs2_wait_on_busy_lock(lockres);
2282
2283 spin_lock_irqsave(&lockres->l_lock, flags);
2284 }
2285
0d5dc6c2
MF		 2286 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2287 if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
2288 lockres->l_level == LKM_EXMODE &&
2289 !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2290 lockres->l_ops->set_lvb(lockres);
2291 }
ccd979bd
MF		 2292
2293 if (lockres->l_flags & OCFS2_LOCK_BUSY)
2294 mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
2295 lockres->l_name);
2296 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2297 mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
2298
2299 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
2300 spin_unlock_irqrestore(&lockres->l_lock, flags);
2301 goto out;
2302 }
2303
2304 lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
2305
2306 /* make sure we never get here while waiting for an ast to
2307 * fire. */
2308 BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
2309
2310 /* is this necessary? */
2311 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2312 lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
2313 spin_unlock_irqrestore(&lockres->l_lock, flags);
2314
2315 mlog(0, "lock %s\n", lockres->l_name);
2316
b80fc012 2317 status = dlmunlock(osb->dlm, &lockres->l_lksb, lkm_flags,
2a45f2d1 2318 ocfs2_unlock_ast, lockres);
ccd979bd
MF		 2319 if (status != DLM_NORMAL) {
2320 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2321 mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
2322 dlm_print_one_lock(lockres->l_lksb.lockid);
2323 BUG();
2324 }
2325 mlog(0, "lock %s, successfull return from dlmunlock\n",
2326 lockres->l_name);
2327
2328 ocfs2_wait_on_busy_lock(lockres);
2329out:
2330 mlog_exit(0);
2331 return 0;
2332}
2333
2334/* Mark the lockres as being dropped. It will no longer be
2335 * queued if blocking, but we still may have to wait on it
2336 * being dequeued from the vote thread before we can consider
2337 * it safe to drop.
2338 *
2339 * You can *not* attempt to call cluster_lock on this lockres anymore. */
2340void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
2341{
2342 int status;
2343 struct ocfs2_mask_waiter mw;
2344 unsigned long flags;
2345
2346 ocfs2_init_mask_waiter(&mw);
2347
2348 spin_lock_irqsave(&lockres->l_lock, flags);
2349 lockres->l_flags |= OCFS2_LOCK_FREEING;
2350 while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
2351 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
2352 spin_unlock_irqrestore(&lockres->l_lock, flags);
2353
2354 mlog(0, "Waiting on lockres %s\n", lockres->l_name);
2355
2356 status = ocfs2_wait_for_mask(&mw);
2357 if (status)
2358 mlog_errno(status);
2359
2360 spin_lock_irqsave(&lockres->l_lock, flags);
2361 }
2362 spin_unlock_irqrestore(&lockres->l_lock, flags);
2363}
2364
d680efe9
MF		 2365void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
2366 struct ocfs2_lock_res *lockres)
ccd979bd 2367{
d680efe9 2368 int ret;
ccd979bd 2369
d680efe9 2370 ocfs2_mark_lockres_freeing(lockres);
0d5dc6c2 2371 ret = ocfs2_drop_lock(osb, lockres);
d680efe9
MF		 2372 if (ret)
2373 mlog_errno(ret);
2374}
ccd979bd 2375
d680efe9
MF		 2376static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
2377{
2378 ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
2379 ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
ccd979bd
MF		 2380}
2381
ccd979bd
MF		 2382int ocfs2_drop_inode_locks(struct inode *inode)
2383{
2384 int status, err;
ccd979bd
MF		 2385
2386 mlog_entry_void();
2387
2388 /* No need to call ocfs2_mark_lockres_freeing here -
2389 * ocfs2_clear_inode has done it for us. */
2390
2391 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
0d5dc6c2 2392 &OCFS2_I(inode)->ip_data_lockres);
ccd979bd
MF		 2393 if (err < 0)
2394 mlog_errno(err);
2395
2396 status = err;
2397
2398 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
0d5dc6c2 2399 &OCFS2_I(inode)->ip_meta_lockres);
ccd979bd
MF		 2400 if (err < 0)
2401 mlog_errno(err);
2402 if (err < 0 && !status)
2403 status = err;
2404
2405 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
0d5dc6c2 2406 &OCFS2_I(inode)->ip_rw_lockres);
ccd979bd
MF		 2407 if (err < 0)
2408 mlog_errno(err);
2409 if (err < 0 && !status)
2410 status = err;
2411
2412 mlog_exit(status);
2413 return status;
2414}
2415
2416static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
2417 int new_level)
2418{
2419 assert_spin_locked(&lockres->l_lock);
2420
2421 BUG_ON(lockres->l_blocking <= LKM_NLMODE);
2422
2423 if (lockres->l_level <= new_level) {
2424 mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n",
2425 lockres->l_level, new_level);
2426 BUG();
2427 }
2428
2429 mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
2430 lockres->l_name, new_level, lockres->l_blocking);
2431
2432 lockres->l_action = OCFS2_AST_DOWNCONVERT;
2433 lockres->l_requested = new_level;
2434 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2435}
2436
2437static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
2438 struct ocfs2_lock_res *lockres,
2439 int new_level,
2440 int lvb)
2441{
2442 int ret, dlm_flags = LKM_CONVERT;
2443 enum dlm_status status;
2444
2445 mlog_entry_void();
2446
2447 if (lvb)
2448 dlm_flags |= LKM_VALBLK;
2449
2450 status = dlmlock(osb->dlm,
2451 new_level,
2452 &lockres->l_lksb,
2453 dlm_flags,
2454 lockres->l_name,
f0681062 2455 OCFS2_LOCK_ID_MAX_LEN - 1,
e92d57df 2456 ocfs2_locking_ast,
ccd979bd 2457 lockres,
aa2623ad 2458 ocfs2_blocking_ast);
ccd979bd
MF		 2459 if (status != DLM_NORMAL) {
2460 ocfs2_log_dlm_error("dlmlock", status, lockres);
2461 ret = -EINVAL;
2462 ocfs2_recover_from_dlm_error(lockres, 1);
2463 goto bail;
2464 }
2465
2466 ret = 0;
2467bail:
2468 mlog_exit(ret);
2469 return ret;
2470}
2471
2472/* returns 1 when the caller should unlock and call dlmunlock */
2473static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
2474 struct ocfs2_lock_res *lockres)
2475{
2476 assert_spin_locked(&lockres->l_lock);
2477
2478 mlog_entry_void();
2479 mlog(0, "lock %s\n", lockres->l_name);
2480
2481 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
2482 /* If we're already trying to cancel a lock conversion
2483 * then just drop the spinlock and allow the caller to
2484 * requeue this lock. */
2485
2486 mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
2487 return 0;
2488 }
2489
2490 /* were we in a convert when we got the bast fire? */
2491 BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
2492 lockres->l_action != OCFS2_AST_DOWNCONVERT);
2493 /* set things up for the unlockast to know to just
2494 * clear out the ast_action and unset busy, etc. */
2495 lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
2496
2497 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
2498 "lock %s, invalid flags: 0x%lx\n",
2499 lockres->l_name, lockres->l_flags);
2500
2501 return 1;
2502}
2503
2504static int ocfs2_cancel_convert(struct ocfs2_super *osb,
2505 struct ocfs2_lock_res *lockres)
2506{
2507 int ret;
2508 enum dlm_status status;
2509
2510 mlog_entry_void();
2511 mlog(0, "lock %s\n", lockres->l_name);
2512
2513 ret = 0;
2514 status = dlmunlock(osb->dlm,
2515 &lockres->l_lksb,
2516 LKM_CANCEL,
2a45f2d1 2517 ocfs2_unlock_ast,
ccd979bd
MF		 2518 lockres);
2519 if (status != DLM_NORMAL) {
2520 ocfs2_log_dlm_error("dlmunlock", status, lockres);
2521 ret = -EINVAL;
2522 ocfs2_recover_from_dlm_error(lockres, 0);
2523 }
2524
2525 mlog(0, "lock %s return from dlmunlock\n", lockres->l_name);
2526
ccd979bd
MF		 2527 mlog_exit(ret);
2528 return ret;
2529}
2530
b5e500e2
MF		 2531static int ocfs2_unblock_lock(struct ocfs2_super *osb,
2532 struct ocfs2_lock_res *lockres,
2533 struct ocfs2_unblock_ctl *ctl)
ccd979bd
MF		 2534{
2535 unsigned long flags;
2536 int blocking;
2537 int new_level;
2538 int ret = 0;
5ef0d4ea 2539 int set_lvb = 0;
ccd979bd
MF		 2540
2541 mlog_entry_void();
2542
2543 spin_lock_irqsave(&lockres->l_lock, flags);
2544
2545 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
2546
2547recheck:
2548 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
d680efe9 2549 ctl->requeue = 1;
ccd979bd
MF		 2550 ret = ocfs2_prepare_cancel_convert(osb, lockres);
2551 spin_unlock_irqrestore(&lockres->l_lock, flags);
2552 if (ret) {
2553 ret = ocfs2_cancel_convert(osb, lockres);
2554 if (ret < 0)
2555 mlog_errno(ret);
2556 }
2557 goto leave;
2558 }
2559
2560 /* if we're blocking an exclusive and we have *any* holders,
2561 * then requeue. */
2562 if ((lockres->l_blocking == LKM_EXMODE)
f7fbfdd1
MF		 2563 && (lockres->l_ex_holders || lockres->l_ro_holders))
2564 goto leave_requeue;
ccd979bd
MF		 2565
2566 /* If it's a PR we're blocking, then only
2567 * requeue if we've got any EX holders */
2568 if (lockres->l_blocking == LKM_PRMODE &&
f7fbfdd1
MF		 2569 lockres->l_ex_holders)
2570 goto leave_requeue;
2571
2572 /*
2573 * Can we get a lock in this state if the holder counts are
2574 * zero? The meta data unblock code used to check this.
2575 */
2576 if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
2577 && (lockres->l_flags & OCFS2_LOCK_REFRESHING))
2578 goto leave_requeue;
ccd979bd 2579
16d5b956
MF		 2580 new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
2581
2582 if (lockres->l_ops->check_downconvert
2583 && !lockres->l_ops->check_downconvert(lockres, new_level))
2584 goto leave_requeue;
2585
ccd979bd
MF		 2586 /* If we get here, then we know that there are no more
2587 * incompatible holders (and anyone asking for an incompatible
2588 * lock is blocked). We can now downconvert the lock */
cc567d89 2589 if (!lockres->l_ops->downconvert_worker)
ccd979bd
MF		 2590 goto downconvert;
2591
2592 /* Some lockres types want to do a bit of work before
2593 * downconverting a lock. Allow that here. The worker function
2594 * may sleep, so we save off a copy of what we're blocking as
2595 * it may change while we're not holding the spin lock. */
2596 blocking = lockres->l_blocking;
2597 spin_unlock_irqrestore(&lockres->l_lock, flags);
2598
cc567d89 2599 ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
d680efe9
MF		 2600
2601 if (ctl->unblock_action == UNBLOCK_STOP_POST)
2602 goto leave;
ccd979bd
MF		 2603
2604 spin_lock_irqsave(&lockres->l_lock, flags);
2605 if (blocking != lockres->l_blocking) {
2606 /* If this changed underneath us, then we can't drop
2607 * it just yet. */
2608 goto recheck;
2609 }
2610
2611downconvert:
d680efe9 2612 ctl->requeue = 0;
ccd979bd 2613
5ef0d4ea
MF		 2614 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2615 if (lockres->l_level == LKM_EXMODE)
2616 set_lvb = 1;
2617
2618 /*
2619 * We only set the lvb if the lock has been fully
2620 * refreshed - otherwise we risk setting stale
2621 * data. Otherwise, there's no need to actually clear
2622 * out the lvb here as it's value is still valid.
2623 */
2624 if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2625 lockres->l_ops->set_lvb(lockres);
2626 }
2627
ccd979bd
MF		 2628 ocfs2_prepare_downconvert(lockres, new_level);
2629 spin_unlock_irqrestore(&lockres->l_lock, flags);
5ef0d4ea 2630 ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb);
ccd979bd
MF		 2631leave:
2632 mlog_exit(ret);
2633 return ret;
f7fbfdd1
MF		 2634
2635leave_requeue:
2636 spin_unlock_irqrestore(&lockres->l_lock, flags);
2637 ctl->requeue = 1;
2638
2639 mlog_exit(0);
2640 return 0;
ccd979bd
MF		 2641}
2642
d680efe9
MF		 2643static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
2644 int blocking)
ccd979bd
MF		 2645{
2646 struct inode *inode;
2647 struct address_space *mapping;
2648
ccd979bd
MF		 2649 inode = ocfs2_lock_res_inode(lockres);
2650 mapping = inode->i_mapping;
2651
2652 if (filemap_fdatawrite(mapping)) {
b0697053
MF		 2653 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
2654 (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 2655 }
2656 sync_mapping_buffers(mapping);
2657 if (blocking == LKM_EXMODE) {
2658 truncate_inode_pages(mapping, 0);
2659 unmap_mapping_range(mapping, 0, 0, 0);
2660 } else {
2661 /* We only need to wait on the I/O if we're not also
2662 * truncating pages because truncate_inode_pages waits
2663 * for us above. We don't truncate pages if we're
2664 * blocking anything < EXMODE because we want to keep
2665 * them around in that case. */
2666 filemap_fdatawait(mapping);
2667 }
2668
d680efe9 2669 return UNBLOCK_CONTINUE;
ccd979bd
MF		 2670}
2671
810d5aeb
MF		 2672static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
2673 int new_level)
2674{
2675 struct inode *inode = ocfs2_lock_res_inode(lockres);
2676 int checkpointed = ocfs2_inode_fully_checkpointed(inode);
2677
2678 BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE);
2679 BUG_ON(lockres->l_level != LKM_EXMODE && !checkpointed);
2680
2681 if (checkpointed)
2682 return 1;
2683
2684 ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb));
2685 return 0;
2686}
2687
2688static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
2689{
2690 struct inode *inode = ocfs2_lock_res_inode(lockres);
2691
2692 __ocfs2_stuff_meta_lvb(inode);
2693}
2694
d680efe9
MF		 2695/*
2696 * Does the final reference drop on our dentry lock. Right now this
2697 * happens in the vote thread, but we could choose to simplify the
2698 * dlmglue API and push these off to the ocfs2_wq in the future.
2699 */
2700static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
2701 struct ocfs2_lock_res *lockres)
2702{
2703 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
2704 ocfs2_dentry_lock_put(osb, dl);
2705}
2706
2707/*
2708 * d_delete() matching dentries before the lock downconvert.
2709 *
2710 * At this point, any process waiting to destroy the
2711 * dentry_lock due to last ref count is stopped by the
2712 * OCFS2_LOCK_QUEUED flag.
2713 *
2714 * We have two potential problems
2715 *
2716 * 1) If we do the last reference drop on our dentry_lock (via dput)
2717 * we'll wind up in ocfs2_release_dentry_lock(), waiting on
2718 * the downconvert to finish. Instead we take an elevated
2719 * reference and push the drop until after we've completed our
2720 * unblock processing.
2721 *
2722 * 2) There might be another process with a final reference,
2723 * waiting on us to finish processing. If this is the case, we
2724 * detect it and exit out - there's no more dentries anyway.
2725 */
2726static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
2727 int blocking)
2728{
2729 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
2730 struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
2731 struct dentry *dentry;
2732 unsigned long flags;
2733 int extra_ref = 0;
2734
2735 /*
2736 * This node is blocking another node from getting a read
2737 * lock. This happens when we've renamed within a
2738 * directory. We've forced the other nodes to d_delete(), but
2739 * we never actually dropped our lock because it's still
2740 * valid. The downconvert code will retain a PR for this node,
2741 * so there's no further work to do.
2742 */
2743 if (blocking == LKM_PRMODE)
2744 return UNBLOCK_CONTINUE;
2745
2746 /*
2747 * Mark this inode as potentially orphaned. The code in
2748 * ocfs2_delete_inode() will figure out whether it actually
2749 * needs to be freed or not.
2750 */
2751 spin_lock(&oi->ip_lock);
2752 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2753 spin_unlock(&oi->ip_lock);
2754
2755 /*
2756 * Yuck. We need to make sure however that the check of
2757 * OCFS2_LOCK_FREEING and the extra reference are atomic with
2758 * respect to a reference decrement or the setting of that
2759 * flag.
2760 */
2761 spin_lock_irqsave(&lockres->l_lock, flags);
2762 spin_lock(&dentry_attach_lock);
2763 if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
2764 && dl->dl_count) {
2765 dl->dl_count++;
2766 extra_ref = 1;
2767 }
2768 spin_unlock(&dentry_attach_lock);
2769 spin_unlock_irqrestore(&lockres->l_lock, flags);
2770
2771 mlog(0, "extra_ref = %d\n", extra_ref);
2772
2773 /*
2774 * We have a process waiting on us in ocfs2_dentry_iput(),
2775 * which means we can't have any more outstanding
2776 * aliases. There's no need to do any more work.
2777 */
2778 if (!extra_ref)
2779 return UNBLOCK_CONTINUE;
2780
2781 spin_lock(&dentry_attach_lock);
2782 while (1) {
2783 dentry = ocfs2_find_local_alias(dl->dl_inode,
2784 dl->dl_parent_blkno, 1);
2785 if (!dentry)
2786 break;
2787 spin_unlock(&dentry_attach_lock);
2788
2789 mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
2790 dentry->d_name.name);
2791
2792 /*
2793 * The following dcache calls may do an
2794 * iput(). Normally we don't want that from the
2795 * downconverting thread, but in this case it's ok
2796 * because the requesting node already has an
2797 * exclusive lock on the inode, so it can't be queued
2798 * for a downconvert.
2799 */
2800 d_delete(dentry);
2801 dput(dentry);
2802
2803 spin_lock(&dentry_attach_lock);
2804 }
2805 spin_unlock(&dentry_attach_lock);
2806
2807 /*
2808 * If we are the last holder of this dentry lock, there is no
2809 * reason to downconvert so skip straight to the unlock.
2810 */
2811 if (dl->dl_count == 1)
2812 return UNBLOCK_STOP_POST;
2813
2814 return UNBLOCK_CONTINUE_POST;
2815}
2816
ccd979bd
MF		 2817void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
2818 struct ocfs2_lock_res *lockres)
2819{
2820 int status;
d680efe9 2821 struct ocfs2_unblock_ctl ctl = {0, 0,};
ccd979bd
MF		 2822 unsigned long flags;
2823
2824 /* Our reference to the lockres in this function can be
2825 * considered valid until we remove the OCFS2_LOCK_QUEUED
2826 * flag. */
2827
2828 mlog_entry_void();
2829
2830 BUG_ON(!lockres);
2831 BUG_ON(!lockres->l_ops);
ccd979bd
MF		 2832
2833 mlog(0, "lockres %s blocked.\n", lockres->l_name);
2834
2835 /* Detect whether a lock has been marked as going away while
2836 * the vote thread was processing other things. A lock can
2837 * still be marked with OCFS2_LOCK_FREEING after this check,
2838 * but short circuiting here will still save us some
2839 * performance. */
2840 spin_lock_irqsave(&lockres->l_lock, flags);
2841 if (lockres->l_flags & OCFS2_LOCK_FREEING)
2842 goto unqueue;
2843 spin_unlock_irqrestore(&lockres->l_lock, flags);
2844
b5e500e2 2845 status = ocfs2_unblock_lock(osb, lockres, &ctl);
ccd979bd
MF		 2846 if (status < 0)
2847 mlog_errno(status);
2848
2849 spin_lock_irqsave(&lockres->l_lock, flags);
2850unqueue:
d680efe9 2851 if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
ccd979bd
MF		 2852 lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
2853 } else
2854 ocfs2_schedule_blocked_lock(osb, lockres);
2855
2856 mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
d680efe9 2857 ctl.requeue ? "yes" : "no");
ccd979bd
MF		 2858 spin_unlock_irqrestore(&lockres->l_lock, flags);
2859
d680efe9
MF		 2860 if (ctl.unblock_action != UNBLOCK_CONTINUE
2861 && lockres->l_ops->post_unlock)
2862 lockres->l_ops->post_unlock(osb, lockres);
2863
ccd979bd
MF		 2864 mlog_exit_void();
2865}
2866
2867static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
2868 struct ocfs2_lock_res *lockres)
2869{
2870 mlog_entry_void();
2871
2872 assert_spin_locked(&lockres->l_lock);
2873
2874 if (lockres->l_flags & OCFS2_LOCK_FREEING) {
2875 /* Do not schedule a lock for downconvert when it's on
2876 * the way to destruction - any nodes wanting access
2877 * to the resource will get it soon. */
2878 mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
2879 lockres->l_name, lockres->l_flags);
2880 return;
2881 }
2882
2883 lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
2884
2885 spin_lock(&osb->vote_task_lock);
2886 if (list_empty(&lockres->l_blocked_list)) {
2887 list_add_tail(&lockres->l_blocked_list,
2888 &osb->blocked_lock_list);
2889 osb->blocked_lock_count++;
2890 }
2891 spin_unlock(&osb->vote_task_lock);
2892
2893 mlog_exit_void();
2894}
2895
2896/* This aids in debugging situations where a bad LVB might be involved. */
2897void ocfs2_dump_meta_lvb_info(u64 level,
2898 const char *function,
2899 unsigned int line,
2900 struct ocfs2_lock_res *lockres)
2901{
2902 struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
2903
2904 mlog(level, "LVB information for %s (called from %s:%u):\n",
2905 lockres->l_name, function, line);
f9e2d82e
MF		 2906 mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
2907 lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
2908 be32_to_cpu(lvb->lvb_igeneration));
b0697053
MF		 2909 mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
2910 (unsigned long long)be64_to_cpu(lvb->lvb_isize),
2911 be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
2912 be16_to_cpu(lvb->lvb_imode));
2913 mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
ca4d147e 2914 "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
b0697053
MF		 2915 (long long)be64_to_cpu(lvb->lvb_iatime_packed),
2916 (long long)be64_to_cpu(lvb->lvb_ictime_packed),
ca4d147e
HP		 2917 (long long)be64_to_cpu(lvb->lvb_imtime_packed),
2918 be32_to_cpu(lvb->lvb_iattr));
ccd979bd 2919}
kernel source for telekom puls (alcatel/mediatek)