projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2_dlmfs: Move to its own directory
[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>
ccd979bd
MF		 30#include <linux/kthread.h>
31#include <linux/pagemap.h>
32#include <linux/debugfs.h>
33#include <linux/seq_file.h>
8ddb7b00 34#include <linux/time.h>
9e33d69f 35#include <linux/quotaops.h>
ccd979bd 36
ccd979bd
MF		 37#define MLOG_MASK_PREFIX ML_DLM_GLUE
38#include <cluster/masklog.h>
39
40#include "ocfs2.h"
d24fbcda 41#include "ocfs2_lockingver.h"
ccd979bd
MF		 42
43#include "alloc.h"
d680efe9 44#include "dcache.h"
ccd979bd
MF		 45#include "dlmglue.h"
46#include "extent_map.h"
7f1a37e3 47#include "file.h"
ccd979bd
MF		 48#include "heartbeat.h"
49#include "inode.h"
50#include "journal.h"
24ef1815 51#include "stackglue.h"
ccd979bd
MF		 52#include "slot_map.h"
53#include "super.h"
54#include "uptodate.h"
9e33d69f 55#include "quota.h"
8dec98ed 56#include "refcounttree.h"
ccd979bd
MF		 57
58#include "buffer_head_io.h"
59
60struct ocfs2_mask_waiter {
61 struct list_head mw_item;
62 int mw_status;
63 struct completion mw_complete;
64 unsigned long mw_mask;
65 unsigned long mw_goal;
8ddb7b00
SM		 66#ifdef CONFIG_OCFS2_FS_STATS
67 unsigned long long mw_lock_start;
68#endif
ccd979bd
MF		 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);
cf8e06f1 73static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
9e33d69f 74static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
ccd979bd 75
d680efe9 76/*
cc567d89 77 * Return value from ->downconvert_worker functions.
d680efe9 78 *
b5e500e2 79 * These control the precise actions of ocfs2_unblock_lock()
d680efe9
MF		 80 * and ocfs2_process_blocked_lock()
81 *
82 */
83enum ocfs2_unblock_action {
84 UNBLOCK_CONTINUE = 0, /* Continue downconvert */
85 UNBLOCK_CONTINUE_POST = 1, /* Continue downconvert, fire
86 * ->post_unlock callback */
87 UNBLOCK_STOP_POST = 2, /* Do not downconvert, fire
88 * ->post_unlock() callback. */
89};
90
91struct ocfs2_unblock_ctl {
92 int requeue;
93 enum ocfs2_unblock_action unblock_action;
94};
95
cb25797d
JK		 96/* Lockdep class keys */
97struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES];
98
810d5aeb
MF		 99static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
100 int new_level);
101static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
102
cc567d89
MF		 103static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
104 int blocking);
105
cc567d89
MF		 106static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
107 int blocking);
d680efe9
MF		 108
109static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
110 struct ocfs2_lock_res *lockres);
ccd979bd 111
9e33d69f 112static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
6cb129f5 113
8dec98ed
TM		 114static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
115 int new_level);
116static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
117 int blocking);
118
6cb129f5
AB		 119#define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
120
121/* This aids in debugging situations where a bad LVB might be involved. */
122static void ocfs2_dump_meta_lvb_info(u64 level,
123 const char *function,
124 unsigned int line,
125 struct ocfs2_lock_res *lockres)
126{
a641dc2a 127 struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
6cb129f5
AB		 128
129 mlog(level, "LVB information for %s (called from %s:%u):\n",
130 lockres->l_name, function, line);
131 mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
132 lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
133 be32_to_cpu(lvb->lvb_igeneration));
134 mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
135 (unsigned long long)be64_to_cpu(lvb->lvb_isize),
136 be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
137 be16_to_cpu(lvb->lvb_imode));
138 mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
139 "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
140 (long long)be64_to_cpu(lvb->lvb_iatime_packed),
141 (long long)be64_to_cpu(lvb->lvb_ictime_packed),
142 (long long)be64_to_cpu(lvb->lvb_imtime_packed),
143 be32_to_cpu(lvb->lvb_iattr));
144}
145
146
f625c979
MF		 147/*
148 * OCFS2 Lock Resource Operations
149 *
150 * These fine tune the behavior of the generic dlmglue locking infrastructure.
0d5dc6c2
MF		 151 *
152 * The most basic of lock types can point ->l_priv to their respective
153 * struct ocfs2_super and allow the default actions to manage things.
154 *
155 * Right now, each lock type also needs to implement an init function,
156 * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
157 * should be called when the lock is no longer needed (i.e., object
158 * destruction time).
f625c979 159 */
ccd979bd 160struct ocfs2_lock_res_ops {
54a7e755
MF		 161 /*
162 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
163 * this callback if ->l_priv is not an ocfs2_super pointer
164 */
165 struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
b5e500e2 166
0d5dc6c2 167 /*
34d024f8
MF		 168 * Optionally called in the downconvert thread after a
169 * successful downconvert. The lockres will not be referenced
170 * after this callback is called, so it is safe to free
171 * memory, etc.
0d5dc6c2
MF		 172 *
173 * The exact semantics of when this is called are controlled
174 * by ->downconvert_worker()
175 */
d680efe9 176 void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
f625c979 177
16d5b956
MF		 178 /*
179 * Allow a lock type to add checks to determine whether it is
180 * safe to downconvert a lock. Return 0 to re-queue the
181 * downconvert at a later time, nonzero to continue.
182 *
183 * For most locks, the default checks that there are no
184 * incompatible holders are sufficient.
185 *
186 * Called with the lockres spinlock held.
187 */
188 int (*check_downconvert)(struct ocfs2_lock_res *, int);
189
5ef0d4ea
MF		 190 /*
191 * Allows a lock type to populate the lock value block. This
192 * is called on downconvert, and when we drop a lock.
193 *
194 * Locks that want to use this should set LOCK_TYPE_USES_LVB
195 * in the flags field.
196 *
197 * Called with the lockres spinlock held.
198 */
199 void (*set_lvb)(struct ocfs2_lock_res *);
200
cc567d89
MF		 201 /*
202 * Called from the downconvert thread when it is determined
203 * that a lock will be downconverted. This is called without
204 * any locks held so the function can do work that might
205 * schedule (syncing out data, etc).
206 *
207 * This should return any one of the ocfs2_unblock_action
208 * values, depending on what it wants the thread to do.
209 */
210 int (*downconvert_worker)(struct ocfs2_lock_res *, int);
211
f625c979
MF		 212 /*
213 * LOCK_TYPE_* flags which describe the specific requirements
214 * of a lock type. Descriptions of each individual flag follow.
215 */
216 int flags;
ccd979bd
MF		 217};
218
f625c979
MF		 219/*
220 * Some locks want to "refresh" potentially stale data when a
221 * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
222 * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
223 * individual lockres l_flags member from the ast function. It is
224 * expected that the locking wrapper will clear the
225 * OCFS2_LOCK_NEEDS_REFRESH flag when done.
226 */
227#define LOCK_TYPE_REQUIRES_REFRESH 0x1
228
b80fc012 229/*
5ef0d4ea
MF		 230 * Indicate that a lock type makes use of the lock value block. The
231 * ->set_lvb lock type callback must be defined.
b80fc012
MF		 232 */
233#define LOCK_TYPE_USES_LVB 0x2
234
ccd979bd 235static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
54a7e755 236 .get_osb = ocfs2_get_inode_osb,
f625c979 237 .flags = 0,
ccd979bd
MF		 238};
239
e63aecb6 240static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
54a7e755 241 .get_osb = ocfs2_get_inode_osb,
810d5aeb
MF		 242 .check_downconvert = ocfs2_check_meta_downconvert,
243 .set_lvb = ocfs2_set_meta_lvb,
f1f54068 244 .downconvert_worker = ocfs2_data_convert_worker,
b80fc012 245 .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
ccd979bd
MF		 246};
247
ccd979bd 248static struct ocfs2_lock_res_ops ocfs2_super_lops = {
f625c979 249 .flags = LOCK_TYPE_REQUIRES_REFRESH,
ccd979bd
MF		 250};
251
252static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
f625c979 253 .flags = 0,
ccd979bd
MF		 254};
255
6ca497a8 256static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
257 .flags = 0,
258};
259
83273932
SE		 260static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = {
261 .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
262};
263
d680efe9 264static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
54a7e755 265 .get_osb = ocfs2_get_dentry_osb,
d680efe9 266 .post_unlock = ocfs2_dentry_post_unlock,
cc567d89 267 .downconvert_worker = ocfs2_dentry_convert_worker,
f625c979 268 .flags = 0,
d680efe9
MF		 269};
270
50008630
TY		 271static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
272 .get_osb = ocfs2_get_inode_osb,
273 .flags = 0,
274};
275
cf8e06f1
MF		 276static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
277 .get_osb = ocfs2_get_file_osb,
278 .flags = 0,
279};
280
9e33d69f
JK		 281static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
282 .set_lvb = ocfs2_set_qinfo_lvb,
283 .get_osb = ocfs2_get_qinfo_osb,
284 .flags = LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
285};
286
8dec98ed
TM		 287static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = {
288 .check_downconvert = ocfs2_check_refcount_downconvert,
289 .downconvert_worker = ocfs2_refcount_convert_worker,
290 .flags = 0,
291};
292
ccd979bd
MF		 293static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
294{
295 return lockres->l_type == OCFS2_LOCK_TYPE_META ||
50008630
TY		 296 lockres->l_type == OCFS2_LOCK_TYPE_RW ||
297 lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
ccd979bd
MF		 298}
299
ccd979bd
MF		 300static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
301{
302 BUG_ON(!ocfs2_is_inode_lock(lockres));
303
304 return (struct inode *) lockres->l_priv;
305}
306
d680efe9
MF		 307static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
308{
309 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
310
311 return (struct ocfs2_dentry_lock *)lockres->l_priv;
312}
313
9e33d69f
JK		 314static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
315{
316 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
317
318 return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
319}
320
8dec98ed
TM		 321static inline struct ocfs2_refcount_tree *
322ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
323{
324 return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
325}
326
54a7e755
MF		 327static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
328{
329 if (lockres->l_ops->get_osb)
330 return lockres->l_ops->get_osb(lockres);
331
332 return (struct ocfs2_super *)lockres->l_priv;
333}
334
ccd979bd
MF		 335static int ocfs2_lock_create(struct ocfs2_super *osb,
336 struct ocfs2_lock_res *lockres,
337 int level,
bd3e7610 338 u32 dlm_flags);
ccd979bd
MF		 339static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
340 int wanted);
cb25797d
JK		 341static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
342 struct ocfs2_lock_res *lockres,
343 int level, unsigned long caller_ip);
344static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb,
345 struct ocfs2_lock_res *lockres,
346 int level)
347{
348 __ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_);
349}
350
ccd979bd
MF		 351static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
352static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
353static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
354static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
355static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
356 struct ocfs2_lock_res *lockres);
357static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
358 int convert);
c74ff8bb
SM		 359#define ocfs2_log_dlm_error(_func, _err, _lockres) do { \
360 if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY) \
361 mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n", \
362 _err, _func, _lockres->l_name); \
363 else \
364 mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n", \
365 _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name, \
366 (unsigned int)ocfs2_get_dentry_lock_ino(_lockres)); \
ccd979bd 367} while (0)
34d024f8
MF		 368static int ocfs2_downconvert_thread(void *arg);
369static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
370 struct ocfs2_lock_res *lockres);
e63aecb6 371static int ocfs2_inode_lock_update(struct inode *inode,
ccd979bd
MF		 372 struct buffer_head **bh);
373static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
374static inline int ocfs2_highest_compat_lock_level(int level);
de551246
JB		 375static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
376 int new_level);
cf8e06f1
MF		 377static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
378 struct ocfs2_lock_res *lockres,
379 int new_level,
de551246
JB		 380 int lvb,
381 unsigned int generation);
cf8e06f1
MF		 382static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
383 struct ocfs2_lock_res *lockres);
384static int ocfs2_cancel_convert(struct ocfs2_super *osb,
385 struct ocfs2_lock_res *lockres);
386
ccd979bd 387
ccd979bd
MF		 388static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
389 u64 blkno,
390 u32 generation,
391 char *name)
392{
393 int len;
394
395 mlog_entry_void();
396
397 BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
398
b0697053
MF		 399 len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
400 ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
401 (long long)blkno, generation);
ccd979bd
MF		 402
403 BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
404
405 mlog(0, "built lock resource with name: %s\n", name);
406
407 mlog_exit_void();
408}
409
34af946a 410static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
ccd979bd
MF		 411
412static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
413 struct ocfs2_dlm_debug *dlm_debug)
414{
415 mlog(0, "Add tracking for lockres %s\n", res->l_name);
416
417 spin_lock(&ocfs2_dlm_tracking_lock);
418 list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
419 spin_unlock(&ocfs2_dlm_tracking_lock);
420}
421
422static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
423{
424 spin_lock(&ocfs2_dlm_tracking_lock);
425 if (!list_empty(&res->l_debug_list))
426 list_del_init(&res->l_debug_list);
427 spin_unlock(&ocfs2_dlm_tracking_lock);
428}
429
8ddb7b00
SM		 430#ifdef CONFIG_OCFS2_FS_STATS
431static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
432{
433 res->l_lock_num_prmode = 0;
434 res->l_lock_num_prmode_failed = 0;
435 res->l_lock_total_prmode = 0;
436 res->l_lock_max_prmode = 0;
437 res->l_lock_num_exmode = 0;
438 res->l_lock_num_exmode_failed = 0;
439 res->l_lock_total_exmode = 0;
440 res->l_lock_max_exmode = 0;
441 res->l_lock_refresh = 0;
442}
443
444static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
445 struct ocfs2_mask_waiter *mw, int ret)
446{
447 unsigned long long *num, *sum;
448 unsigned int *max, *failed;
449 struct timespec ts = current_kernel_time();
450 unsigned long long time = timespec_to_ns(&ts) - mw->mw_lock_start;
451
452 if (level == LKM_PRMODE) {
453 num = &res->l_lock_num_prmode;
454 sum = &res->l_lock_total_prmode;
455 max = &res->l_lock_max_prmode;
456 failed = &res->l_lock_num_prmode_failed;
457 } else if (level == LKM_EXMODE) {
458 num = &res->l_lock_num_exmode;
459 sum = &res->l_lock_total_exmode;
460 max = &res->l_lock_max_exmode;
461 failed = &res->l_lock_num_exmode_failed;
462 } else
463 return;
464
465 (*num)++;
466 (*sum) += time;
467 if (time > *max)
468 *max = time;
469 if (ret)
470 (*failed)++;
471}
472
473static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
474{
475 lockres->l_lock_refresh++;
476}
477
478static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
479{
480 struct timespec ts = current_kernel_time();
481 mw->mw_lock_start = timespec_to_ns(&ts);
482}
483#else
484static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
485{
486}
487static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
488 int level, struct ocfs2_mask_waiter *mw, int ret)
489{
490}
491static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
492{
493}
494static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
495{
496}
497#endif
498
ccd979bd
MF		 499static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
500 struct ocfs2_lock_res *res,
501 enum ocfs2_lock_type type,
ccd979bd
MF		 502 struct ocfs2_lock_res_ops *ops,
503 void *priv)
504{
ccd979bd
MF		 505 res->l_type = type;
506 res->l_ops = ops;
507 res->l_priv = priv;
508
bd3e7610
JB		 509 res->l_level = DLM_LOCK_IV;
510 res->l_requested = DLM_LOCK_IV;
511 res->l_blocking = DLM_LOCK_IV;
ccd979bd
MF		 512 res->l_action = OCFS2_AST_INVALID;
513 res->l_unlock_action = OCFS2_UNLOCK_INVALID;
514
515 res->l_flags = OCFS2_LOCK_INITIALIZED;
516
517 ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
8ddb7b00
SM		 518
519 ocfs2_init_lock_stats(res);
cb25797d
JK		 520#ifdef CONFIG_DEBUG_LOCK_ALLOC
521 if (type != OCFS2_LOCK_TYPE_OPEN)
522 lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type],
523 &lockdep_keys[type], 0);
524 else
525 res->l_lockdep_map.key = NULL;
526#endif
ccd979bd
MF		 527}
528
529void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
530{
531 /* This also clears out the lock status block */
532 memset(res, 0, sizeof(struct ocfs2_lock_res));
533 spin_lock_init(&res->l_lock);
534 init_waitqueue_head(&res->l_event);
535 INIT_LIST_HEAD(&res->l_blocked_list);
536 INIT_LIST_HEAD(&res->l_mask_waiters);
537}
538
539void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
540 enum ocfs2_lock_type type,
24c19ef4 541 unsigned int generation,
ccd979bd
MF		 542 struct inode *inode)
543{
544 struct ocfs2_lock_res_ops *ops;
545
546 switch(type) {
547 case OCFS2_LOCK_TYPE_RW:
548 ops = &ocfs2_inode_rw_lops;
549 break;
550 case OCFS2_LOCK_TYPE_META:
e63aecb6 551 ops = &ocfs2_inode_inode_lops;
ccd979bd 552 break;
50008630
TY		 553 case OCFS2_LOCK_TYPE_OPEN:
554 ops = &ocfs2_inode_open_lops;
555 break;
ccd979bd
MF		 556 default:
557 mlog_bug_on_msg(1, "type: %d\n", type);
558 ops = NULL; /* thanks, gcc */
559 break;
560 };
561
d680efe9 562 ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
24c19ef4 563 generation, res->l_name);
d680efe9
MF		 564 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
565}
566
54a7e755
MF		 567static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
568{
569 struct inode *inode = ocfs2_lock_res_inode(lockres);
570
571 return OCFS2_SB(inode->i_sb);
572}
573
9e33d69f
JK		 574static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
575{
576 struct ocfs2_mem_dqinfo *info = lockres->l_priv;
577
578 return OCFS2_SB(info->dqi_gi.dqi_sb);
579}
580
cf8e06f1
MF		 581static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
582{
583 struct ocfs2_file_private *fp = lockres->l_priv;
584
585 return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
586}
587
d680efe9
MF		 588static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
589{
590 __be64 inode_blkno_be;
591
592 memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
593 sizeof(__be64));
594
595 return be64_to_cpu(inode_blkno_be);
596}
597
54a7e755
MF		 598static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
599{
600 struct ocfs2_dentry_lock *dl = lockres->l_priv;
601
602 return OCFS2_SB(dl->dl_inode->i_sb);
603}
604
d680efe9
MF		 605void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
606 u64 parent, struct inode *inode)
607{
608 int len;
609 u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
610 __be64 inode_blkno_be = cpu_to_be64(inode_blkno);
611 struct ocfs2_lock_res *lockres = &dl->dl_lockres;
612
613 ocfs2_lock_res_init_once(lockres);
614
615 /*
616 * Unfortunately, the standard lock naming scheme won't work
617 * here because we have two 16 byte values to use. Instead,
618 * we'll stuff the inode number as a binary value. We still
619 * want error prints to show something without garbling the
620 * display, so drop a null byte in there before the inode
621 * number. A future version of OCFS2 will likely use all
622 * binary lock names. The stringified names have been a
623 * tremendous aid in debugging, but now that the debugfs
624 * interface exists, we can mangle things there if need be.
625 *
626 * NOTE: We also drop the standard "pad" value (the total lock
627 * name size stays the same though - the last part is all
628 * zeros due to the memset in ocfs2_lock_res_init_once()
629 */
630 len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
631 "%c%016llx",
632 ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
633 (long long)parent);
634
635 BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
636
637 memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
638 sizeof(__be64));
639
640 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
641 OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
642 dl);
ccd979bd
MF		 643}
644
645static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
646 struct ocfs2_super *osb)
647{
648 /* Superblock lockres doesn't come from a slab so we call init
649 * once on it manually. */
650 ocfs2_lock_res_init_once(res);
d680efe9
MF		 651 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
652 0, res->l_name);
ccd979bd 653 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
ccd979bd
MF		 654 &ocfs2_super_lops, osb);
655}
656
657static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
658 struct ocfs2_super *osb)
659{
660 /* Rename lockres doesn't come from a slab so we call init
661 * once on it manually. */
662 ocfs2_lock_res_init_once(res);
d680efe9
MF		 663 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
664 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
ccd979bd
MF		 665 &ocfs2_rename_lops, osb);
666}
667
6ca497a8 668static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
669 struct ocfs2_super *osb)
670{
671 /* nfs_sync lockres doesn't come from a slab so we call init
672 * once on it manually. */
673 ocfs2_lock_res_init_once(res);
674 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
675 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
676 &ocfs2_nfs_sync_lops, osb);
677}
678
83273932
SE		 679static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
680 struct ocfs2_super *osb)
681{
83273932
SE		 682 ocfs2_lock_res_init_once(res);
683 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
684 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
685 &ocfs2_orphan_scan_lops, osb);
83273932
SE		 686}
687
cf8e06f1
MF		 688void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
689 struct ocfs2_file_private *fp)
690{
691 struct inode *inode = fp->fp_file->f_mapping->host;
692 struct ocfs2_inode_info *oi = OCFS2_I(inode);
693
694 ocfs2_lock_res_init_once(lockres);
695 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
696 inode->i_generation, lockres->l_name);
697 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
698 OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
699 fp);
700 lockres->l_flags |= OCFS2_LOCK_NOCACHE;
701}
702
9e33d69f
JK		 703void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
704 struct ocfs2_mem_dqinfo *info)
705{
706 ocfs2_lock_res_init_once(lockres);
707 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
708 0, lockres->l_name);
709 ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
710 OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
711 info);
712}
713
8dec98ed
TM		 714void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
715 struct ocfs2_super *osb, u64 ref_blkno,
716 unsigned int generation)
717{
718 ocfs2_lock_res_init_once(lockres);
719 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
720 generation, lockres->l_name);
721 ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
722 &ocfs2_refcount_block_lops, osb);
723}
724
ccd979bd
MF		 725void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
726{
727 mlog_entry_void();
728
729 if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
730 return;
731
732 ocfs2_remove_lockres_tracking(res);
733
734 mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
735 "Lockres %s is on the blocked list\n",
736 res->l_name);
737 mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
738 "Lockres %s has mask waiters pending\n",
739 res->l_name);
740 mlog_bug_on_msg(spin_is_locked(&res->l_lock),
741 "Lockres %s is locked\n",
742 res->l_name);
743 mlog_bug_on_msg(res->l_ro_holders,
744 "Lockres %s has %u ro holders\n",
745 res->l_name, res->l_ro_holders);
746 mlog_bug_on_msg(res->l_ex_holders,
747 "Lockres %s has %u ex holders\n",
748 res->l_name, res->l_ex_holders);
749
750 /* Need to clear out the lock status block for the dlm */
751 memset(&res->l_lksb, 0, sizeof(res->l_lksb));
752
753 res->l_flags = 0UL;
754 mlog_exit_void();
755}
756
757static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
758 int level)
759{
760 mlog_entry_void();
761
762 BUG_ON(!lockres);
763
764 switch(level) {
bd3e7610 765 case DLM_LOCK_EX:
ccd979bd
MF		 766 lockres->l_ex_holders++;
767 break;
bd3e7610 768 case DLM_LOCK_PR:
ccd979bd
MF		 769 lockres->l_ro_holders++;
770 break;
771 default:
772 BUG();
773 }
774
775 mlog_exit_void();
776}
777
778static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
779 int level)
780{
781 mlog_entry_void();
782
783 BUG_ON(!lockres);
784
785 switch(level) {
bd3e7610 786 case DLM_LOCK_EX:
ccd979bd
MF		 787 BUG_ON(!lockres->l_ex_holders);
788 lockres->l_ex_holders--;
789 break;
bd3e7610 790 case DLM_LOCK_PR:
ccd979bd
MF		 791 BUG_ON(!lockres->l_ro_holders);
792 lockres->l_ro_holders--;
793 break;
794 default:
795 BUG();
796 }
797 mlog_exit_void();
798}
799
800/* WARNING: This function lives in a world where the only three lock
801 * levels are EX, PR, and NL. It *will* have to be adjusted when more
802 * lock types are added. */
803static inline int ocfs2_highest_compat_lock_level(int level)
804{
bd3e7610 805 int new_level = DLM_LOCK_EX;
ccd979bd 806
bd3e7610
JB		 807 if (level == DLM_LOCK_EX)
808 new_level = DLM_LOCK_NL;
809 else if (level == DLM_LOCK_PR)
810 new_level = DLM_LOCK_PR;
ccd979bd
MF		 811 return new_level;
812}
813
814static void lockres_set_flags(struct ocfs2_lock_res *lockres,
815 unsigned long newflags)
816{
800deef3 817 struct ocfs2_mask_waiter *mw, *tmp;
ccd979bd
MF		 818
819 assert_spin_locked(&lockres->l_lock);
820
821 lockres->l_flags = newflags;
822
800deef3 823 list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
ccd979bd
MF		 824 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
825 continue;
826
827 list_del_init(&mw->mw_item);
828 mw->mw_status = 0;
829 complete(&mw->mw_complete);
830 }
831}
832static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
833{
834 lockres_set_flags(lockres, lockres->l_flags | or);
835}
836static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
837 unsigned long clear)
838{
839 lockres_set_flags(lockres, lockres->l_flags & ~clear);
840}
841
842static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
843{
844 mlog_entry_void();
845
846 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
847 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
848 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
bd3e7610 849 BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
ccd979bd
MF		 850
851 lockres->l_level = lockres->l_requested;
852 if (lockres->l_level <=
853 ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
bd3e7610 854 lockres->l_blocking = DLM_LOCK_NL;
ccd979bd
MF		 855 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
856 }
857 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
858
859 mlog_exit_void();
860}
861
862static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
863{
864 mlog_entry_void();
865
866 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
867 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
868
869 /* Convert from RO to EX doesn't really need anything as our
870 * information is already up to data. Convert from NL to
871 * *anything* however should mark ourselves as needing an
872 * update */
bd3e7610 873 if (lockres->l_level == DLM_LOCK_NL &&
f625c979 874 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
ccd979bd
MF		 875 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
876
877 lockres->l_level = lockres->l_requested;
a1912826
SM		 878
879 /*
880 * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
881 * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
882 * downconverting the lock before the upconvert has fully completed.
883 */
884 lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
885
ccd979bd
MF		 886 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
887
888 mlog_exit_void();
889}
890
891static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
892{
893 mlog_entry_void();
894
3cf0c507 895 BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
ccd979bd
MF		 896 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
897
bd3e7610 898 if (lockres->l_requested > DLM_LOCK_NL &&
f625c979
MF		 899 !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
900 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
ccd979bd
MF		 901 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
902
903 lockres->l_level = lockres->l_requested;
904 lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
905 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
906
907 mlog_exit_void();
908}
909
ccd979bd
MF		 910static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
911 int level)
912{
913 int needs_downconvert = 0;
914 mlog_entry_void();
915
916 assert_spin_locked(&lockres->l_lock);
917
ccd979bd
MF		 918 if (level > lockres->l_blocking) {
919 /* only schedule a downconvert if we haven't already scheduled
920 * one that goes low enough to satisfy the level we're
921 * blocking. this also catches the case where we get
922 * duplicate BASTs */
923 if (ocfs2_highest_compat_lock_level(level) <
924 ocfs2_highest_compat_lock_level(lockres->l_blocking))
925 needs_downconvert = 1;
926
927 lockres->l_blocking = level;
928 }
929
0b94a909
WW		 930 if (needs_downconvert)
931 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
932
ccd979bd
MF		 933 mlog_exit(needs_downconvert);
934 return needs_downconvert;
935}
936
de551246
JB		 937/*
938 * OCFS2_LOCK_PENDING and l_pending_gen.
939 *
940 * Why does OCFS2_LOCK_PENDING exist? To close a race between setting
941 * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock(). See ocfs2_unblock_lock()
942 * for more details on the race.
943 *
944 * OCFS2_LOCK_PENDING closes the race quite nicely. However, it introduces
945 * a race on itself. In o2dlm, we can get the ast before ocfs2_dlm_lock()
946 * returns. The ast clears OCFS2_LOCK_BUSY, and must therefore clear
947 * OCFS2_LOCK_PENDING at the same time. When ocfs2_dlm_lock() returns,
948 * the caller is going to try to clear PENDING again. If nothing else is
949 * happening, __lockres_clear_pending() sees PENDING is unset and does
950 * nothing.
951 *
952 * But what if another path (eg downconvert thread) has just started a
953 * new locking action? The other path has re-set PENDING. Our path
954 * cannot clear PENDING, because that will re-open the original race
955 * window.
956 *
957 * [Example]
958 *
959 * ocfs2_meta_lock()
960 * ocfs2_cluster_lock()
961 * set BUSY
962 * set PENDING
963 * drop l_lock
964 * ocfs2_dlm_lock()
965 * ocfs2_locking_ast() ocfs2_downconvert_thread()
966 * clear PENDING ocfs2_unblock_lock()
967 * take_l_lock
968 * !BUSY
969 * ocfs2_prepare_downconvert()
970 * set BUSY
971 * set PENDING
972 * drop l_lock
973 * take l_lock
974 * clear PENDING
975 * drop l_lock
976 * <window>
977 * ocfs2_dlm_lock()
978 *
979 * So as you can see, we now have a window where l_lock is not held,
980 * PENDING is not set, and ocfs2_dlm_lock() has not been called.
981 *
982 * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
983 * set by ocfs2_prepare_downconvert(). That wasn't nice.
984 *
985 * To solve this we introduce l_pending_gen. A call to
986 * lockres_clear_pending() will only do so when it is passed a generation
987 * number that matches the lockres. lockres_set_pending() will return the
988 * current generation number. When ocfs2_cluster_lock() goes to clear
989 * PENDING, it passes the generation it got from set_pending(). In our
990 * example above, the generation numbers will *not* match. Thus,
991 * ocfs2_cluster_lock() will not clear the PENDING set by
992 * ocfs2_prepare_downconvert().
993 */
994
995/* Unlocked version for ocfs2_locking_ast() */
996static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
997 unsigned int generation,
998 struct ocfs2_super *osb)
999{
1000 assert_spin_locked(&lockres->l_lock);
1001
1002 /*
1003 * The ast and locking functions can race us here. The winner
1004 * will clear pending, the loser will not.
1005 */
1006 if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
1007 (lockres->l_pending_gen != generation))
1008 return;
1009
1010 lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
1011 lockres->l_pending_gen++;
1012
1013 /*
1014 * The downconvert thread may have skipped us because we
1015 * were PENDING. Wake it up.
1016 */
1017 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1018 ocfs2_wake_downconvert_thread(osb);
1019}
1020
1021/* Locked version for callers of ocfs2_dlm_lock() */
1022static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
1023 unsigned int generation,
1024 struct ocfs2_super *osb)
1025{
1026 unsigned long flags;
1027
1028 spin_lock_irqsave(&lockres->l_lock, flags);
1029 __lockres_clear_pending(lockres, generation, osb);
1030 spin_unlock_irqrestore(&lockres->l_lock, flags);
1031}
1032
1033static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
1034{
1035 assert_spin_locked(&lockres->l_lock);
1036 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
1037
1038 lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
1039
1040 return lockres->l_pending_gen;
1041}
1042
1043
aa2623ad 1044static void ocfs2_blocking_ast(void *opaque, int level)
ccd979bd 1045{
aa2623ad
MF		 1046 struct ocfs2_lock_res *lockres = opaque;
1047 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
ccd979bd
MF		 1048 int needs_downconvert;
1049 unsigned long flags;
1050
bd3e7610 1051 BUG_ON(level <= DLM_LOCK_NL);
ccd979bd 1052
aa2623ad
MF		 1053 mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n",
1054 lockres->l_name, level, lockres->l_level,
1055 ocfs2_lock_type_string(lockres->l_type));
1056
cf8e06f1
MF		 1057 /*
1058 * We can skip the bast for locks which don't enable caching -
1059 * they'll be dropped at the earliest possible time anyway.
1060 */
1061 if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
1062 return;
1063
ccd979bd
MF		 1064 spin_lock_irqsave(&lockres->l_lock, flags);
1065 needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
1066 if (needs_downconvert)
1067 ocfs2_schedule_blocked_lock(osb, lockres);
1068 spin_unlock_irqrestore(&lockres->l_lock, flags);
1069
d680efe9
MF		 1070 wake_up(&lockres->l_event);
1071
34d024f8 1072 ocfs2_wake_downconvert_thread(osb);
ccd979bd
MF		 1073}
1074
e92d57df 1075static void ocfs2_locking_ast(void *opaque)
ccd979bd 1076{
e92d57df 1077 struct ocfs2_lock_res *lockres = opaque;
de551246 1078 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
ccd979bd 1079 unsigned long flags;
1693a5c0 1080 int status;
ccd979bd
MF		 1081
1082 spin_lock_irqsave(&lockres->l_lock, flags);
1083
1693a5c0
DT		 1084 status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1085
1086 if (status == -EAGAIN) {
1087 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1088 goto out;
1089 }
1090
1091 if (status) {
8f2c9c1b 1092 mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
1693a5c0 1093 lockres->l_name, status);
ccd979bd
MF		 1094 spin_unlock_irqrestore(&lockres->l_lock, flags);
1095 return;
1096 }
1097
1098 switch(lockres->l_action) {
1099 case OCFS2_AST_ATTACH:
1100 ocfs2_generic_handle_attach_action(lockres);
e92d57df 1101 lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
ccd979bd
MF		 1102 break;
1103 case OCFS2_AST_CONVERT:
1104 ocfs2_generic_handle_convert_action(lockres);
1105 break;
1106 case OCFS2_AST_DOWNCONVERT:
1107 ocfs2_generic_handle_downconvert_action(lockres);
1108 break;
1109 default:
e92d57df
MF		 1110 mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
1111 "lockres flags = 0x%lx, unlock action: %u\n",
1112 lockres->l_name, lockres->l_action, lockres->l_flags,
1113 lockres->l_unlock_action);
ccd979bd
MF		 1114 BUG();
1115 }
1693a5c0 1116out:
ccd979bd
MF		 1117 /* set it to something invalid so if we get called again we
1118 * can catch it. */
1119 lockres->l_action = OCFS2_AST_INVALID;
ccd979bd 1120
de551246
JB		 1121 /* Did we try to cancel this lock? Clear that state */
1122 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1123 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1124
1125 /*
1126 * We may have beaten the locking functions here. We certainly
1127 * know that dlm_lock() has been called :-)
1128 * Because we can't have two lock calls in flight at once, we
1129 * can use lockres->l_pending_gen.
1130 */
1131 __lockres_clear_pending(lockres, lockres->l_pending_gen, osb);
1132
ccd979bd 1133 wake_up(&lockres->l_event);
d680efe9 1134 spin_unlock_irqrestore(&lockres->l_lock, flags);
ccd979bd
MF		 1135}
1136
ccd979bd
MF		 1137static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1138 int convert)
1139{
1140 unsigned long flags;
1141
1142 mlog_entry_void();
1143 spin_lock_irqsave(&lockres->l_lock, flags);
1144 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
a1912826 1145 lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
ccd979bd
MF		 1146 if (convert)
1147 lockres->l_action = OCFS2_AST_INVALID;
1148 else
1149 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1150 spin_unlock_irqrestore(&lockres->l_lock, flags);
1151
1152 wake_up(&lockres->l_event);
1153 mlog_exit_void();
1154}
1155
1156/* Note: If we detect another process working on the lock (i.e.,
1157 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1158 * to do the right thing in that case.
1159 */
1160static int ocfs2_lock_create(struct ocfs2_super *osb,
1161 struct ocfs2_lock_res *lockres,
1162 int level,
bd3e7610 1163 u32 dlm_flags)
ccd979bd
MF		 1164{
1165 int ret = 0;
ccd979bd 1166 unsigned long flags;
de551246 1167 unsigned int gen;
ccd979bd
MF		 1168
1169 mlog_entry_void();
1170
bd3e7610 1171 mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
ccd979bd
MF		 1172 dlm_flags);
1173
1174 spin_lock_irqsave(&lockres->l_lock, flags);
1175 if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1176 (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1177 spin_unlock_irqrestore(&lockres->l_lock, flags);
1178 goto bail;
1179 }
1180
1181 lockres->l_action = OCFS2_AST_ATTACH;
1182 lockres->l_requested = level;
1183 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
de551246 1184 gen = lockres_set_pending(lockres);
ccd979bd
MF		 1185 spin_unlock_irqrestore(&lockres->l_lock, flags);
1186
4670c46d 1187 ret = ocfs2_dlm_lock(osb->cconn,
7431cd7e
JB		 1188 level,
1189 &lockres->l_lksb,
1190 dlm_flags,
1191 lockres->l_name,
1192 OCFS2_LOCK_ID_MAX_LEN - 1,
1193 lockres);
de551246 1194 lockres_clear_pending(lockres, gen, osb);
7431cd7e
JB		 1195 if (ret) {
1196 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
ccd979bd
MF		 1197 ocfs2_recover_from_dlm_error(lockres, 1);
1198 }
1199
7431cd7e 1200 mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
ccd979bd
MF		 1201
1202bail:
1203 mlog_exit(ret);
1204 return ret;
1205}
1206
1207static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1208 int flag)
1209{
1210 unsigned long flags;
1211 int ret;
1212
1213 spin_lock_irqsave(&lockres->l_lock, flags);
1214 ret = lockres->l_flags & flag;
1215 spin_unlock_irqrestore(&lockres->l_lock, flags);
1216
1217 return ret;
1218}
1219
1220static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1221
1222{
1223 wait_event(lockres->l_event,
1224 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1225}
1226
1227static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1228
1229{
1230 wait_event(lockres->l_event,
1231 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1232}
1233
1234/* predict what lock level we'll be dropping down to on behalf
1235 * of another node, and return true if the currently wanted
1236 * level will be compatible with it. */
1237static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1238 int wanted)
1239{
1240 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1241
1242 return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1243}
1244
1245static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1246{
1247 INIT_LIST_HEAD(&mw->mw_item);
1248 init_completion(&mw->mw_complete);
8ddb7b00 1249 ocfs2_init_start_time(mw);
ccd979bd
MF		 1250}
1251
1252static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1253{
1254 wait_for_completion(&mw->mw_complete);
1255 /* Re-arm the completion in case we want to wait on it again */
1256 INIT_COMPLETION(mw->mw_complete);
1257 return mw->mw_status;
1258}
1259
1260static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1261 struct ocfs2_mask_waiter *mw,
1262 unsigned long mask,
1263 unsigned long goal)
1264{
1265 BUG_ON(!list_empty(&mw->mw_item));
1266
1267 assert_spin_locked(&lockres->l_lock);
1268
1269 list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1270 mw->mw_mask = mask;
1271 mw->mw_goal = goal;
1272}
1273
1274/* returns 0 if the mw that was removed was already satisfied, -EBUSY
1275 * if the mask still hadn't reached its goal */
1276static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1277 struct ocfs2_mask_waiter *mw)
1278{
1279 unsigned long flags;
1280 int ret = 0;
1281
1282 spin_lock_irqsave(&lockres->l_lock, flags);
1283 if (!list_empty(&mw->mw_item)) {
1284 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1285 ret = -EBUSY;
1286
1287 list_del_init(&mw->mw_item);
1288 init_completion(&mw->mw_complete);
1289 }
1290 spin_unlock_irqrestore(&lockres->l_lock, flags);
1291
1292 return ret;
1293
1294}
1295
cf8e06f1
MF		 1296static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1297 struct ocfs2_lock_res *lockres)
1298{
1299 int ret;
1300
1301 ret = wait_for_completion_interruptible(&mw->mw_complete);
1302 if (ret)
1303 lockres_remove_mask_waiter(lockres, mw);
1304 else
1305 ret = mw->mw_status;
1306 /* Re-arm the completion in case we want to wait on it again */
1307 INIT_COMPLETION(mw->mw_complete);
1308 return ret;
1309}
1310
cb25797d
JK		 1311static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
1312 struct ocfs2_lock_res *lockres,
1313 int level,
1314 u32 lkm_flags,
1315 int arg_flags,
1316 int l_subclass,
1317 unsigned long caller_ip)
ccd979bd
MF		 1318{
1319 struct ocfs2_mask_waiter mw;
ccd979bd
MF		 1320 int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1321 int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1322 unsigned long flags;
de551246 1323 unsigned int gen;
1693a5c0 1324 int noqueue_attempted = 0;
ccd979bd
MF		 1325
1326 mlog_entry_void();
1327
1328 ocfs2_init_mask_waiter(&mw);
1329
b80fc012 1330 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
bd3e7610 1331 lkm_flags |= DLM_LKF_VALBLK;
b80fc012 1332
ccd979bd
MF		 1333again:
1334 wait = 0;
1335
a1912826
SM		 1336 spin_lock_irqsave(&lockres->l_lock, flags);
1337
ccd979bd
MF		 1338 if (catch_signals && signal_pending(current)) {
1339 ret = -ERESTARTSYS;
a1912826 1340 goto unlock;
ccd979bd
MF		 1341 }
1342
ccd979bd
MF		 1343 mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1344 "Cluster lock called on freeing lockres %s! flags "
1345 "0x%lx\n", lockres->l_name, lockres->l_flags);
1346
1347 /* We only compare against the currently granted level
1348 * here. If the lock is blocked waiting on a downconvert,
1349 * we'll get caught below. */
1350 if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1351 level > lockres->l_level) {
1352 /* is someone sitting in dlm_lock? If so, wait on
1353 * them. */
1354 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1355 wait = 1;
1356 goto unlock;
1357 }
1358
a1912826
SM		 1359 if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
1360 /*
1361 * We've upconverted. If the lock now has a level we can
1362 * work with, we take it. If, however, the lock is not at the
1363 * required level, we go thru the full cycle. One way this could
1364 * happen is if a process requesting an upconvert to PR is
1365 * closely followed by another requesting upconvert to an EX.
1366 * If the process requesting EX lands here, we want it to
1367 * continue attempting to upconvert and let the process
1368 * requesting PR take the lock.
1369 * If multiple processes request upconvert to PR, the first one
1370 * here will take the lock. The others will have to go thru the
1371 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
1372 * downconvert request.
1373 */
1374 if (level <= lockres->l_level)
1375 goto update_holders;
1376 }
1377
ccd979bd
MF		 1378 if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1379 !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1380 /* is the lock is currently blocked on behalf of
1381 * another node */
1382 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1383 wait = 1;
1384 goto unlock;
1385 }
1386
1387 if (level > lockres->l_level) {
1693a5c0
DT		 1388 if (noqueue_attempted > 0) {
1389 ret = -EAGAIN;
1390 goto unlock;
1391 }
1392 if (lkm_flags & DLM_LKF_NOQUEUE)
1393 noqueue_attempted = 1;
1394
ccd979bd
MF		 1395 if (lockres->l_action != OCFS2_AST_INVALID)
1396 mlog(ML_ERROR, "lockres %s has action %u pending\n",
1397 lockres->l_name, lockres->l_action);
1398
019d1b22
MF		 1399 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1400 lockres->l_action = OCFS2_AST_ATTACH;
bd3e7610 1401 lkm_flags &= ~DLM_LKF_CONVERT;
019d1b22
MF		 1402 } else {
1403 lockres->l_action = OCFS2_AST_CONVERT;
bd3e7610 1404 lkm_flags |= DLM_LKF_CONVERT;
019d1b22
MF		 1405 }
1406
ccd979bd
MF		 1407 lockres->l_requested = level;
1408 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
de551246 1409 gen = lockres_set_pending(lockres);
ccd979bd
MF		 1410 spin_unlock_irqrestore(&lockres->l_lock, flags);
1411
bd3e7610
JB		 1412 BUG_ON(level == DLM_LOCK_IV);
1413 BUG_ON(level == DLM_LOCK_NL);
ccd979bd
MF		 1414
1415 mlog(0, "lock %s, convert from %d to level = %d\n",
1416 lockres->l_name, lockres->l_level, level);
1417
1418 /* call dlm_lock to upgrade lock now */
4670c46d 1419 ret = ocfs2_dlm_lock(osb->cconn,
7431cd7e
JB		 1420 level,
1421 &lockres->l_lksb,
1422 lkm_flags,
1423 lockres->l_name,
1424 OCFS2_LOCK_ID_MAX_LEN - 1,
1425 lockres);
de551246 1426 lockres_clear_pending(lockres, gen, osb);
7431cd7e
JB		 1427 if (ret) {
1428 if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1429 (ret != -EAGAIN)) {
24ef1815 1430 ocfs2_log_dlm_error("ocfs2_dlm_lock",
7431cd7e 1431 ret, lockres);
ccd979bd
MF		 1432 }
1433 ocfs2_recover_from_dlm_error(lockres, 1);
1434 goto out;
1435 }
1436
73ac36ea 1437 mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
ccd979bd
MF		 1438 lockres->l_name);
1439
1440 /* At this point we've gone inside the dlm and need to
1441 * complete our work regardless. */
1442 catch_signals = 0;
1443
1444 /* wait for busy to clear and carry on */
1445 goto again;
1446 }
1447
a1912826 1448update_holders:
ccd979bd
MF		 1449 /* Ok, if we get here then we're good to go. */
1450 ocfs2_inc_holders(lockres, level);
1451
1452 ret = 0;
1453unlock:
a1912826
SM		 1454 lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1455
ccd979bd
MF		 1456 spin_unlock_irqrestore(&lockres->l_lock, flags);
1457out:
1458 /*
1459 * This is helping work around a lock inversion between the page lock
1460 * and dlm locks. One path holds the page lock while calling aops
1461 * which block acquiring dlm locks. The voting thread holds dlm
1462 * locks while acquiring page locks while down converting data locks.
1463 * This block is helping an aop path notice the inversion and back
1464 * off to unlock its page lock before trying the dlm lock again.
1465 */
1466 if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1467 mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1468 wait = 0;
1469 if (lockres_remove_mask_waiter(lockres, &mw))
1470 ret = -EAGAIN;
1471 else
1472 goto again;
1473 }
1474 if (wait) {
1475 ret = ocfs2_wait_for_mask(&mw);
1476 if (ret == 0)
1477 goto again;
1478 mlog_errno(ret);
1479 }
8ddb7b00 1480 ocfs2_update_lock_stats(lockres, level, &mw, ret);
ccd979bd 1481
cb25797d
JK		 1482#ifdef CONFIG_DEBUG_LOCK_ALLOC
1483 if (!ret && lockres->l_lockdep_map.key != NULL) {
1484 if (level == DLM_LOCK_PR)
1485 rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
1486 !!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1487 caller_ip);
1488 else
1489 rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
1490 !!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1491 caller_ip);
1492 }
1493#endif
ccd979bd
MF		 1494 mlog_exit(ret);
1495 return ret;
1496}
1497
cb25797d
JK		 1498static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
1499 struct ocfs2_lock_res *lockres,
1500 int level,
1501 u32 lkm_flags,
1502 int arg_flags)
1503{
1504 return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
1505 0, _RET_IP_);
1506}
1507
1508
1509static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
1510 struct ocfs2_lock_res *lockres,
1511 int level,
1512 unsigned long caller_ip)
ccd979bd
MF		 1513{
1514 unsigned long flags;
1515
1516 mlog_entry_void();
1517 spin_lock_irqsave(&lockres->l_lock, flags);
1518 ocfs2_dec_holders(lockres, level);
34d024f8 1519 ocfs2_downconvert_on_unlock(osb, lockres);
ccd979bd 1520 spin_unlock_irqrestore(&lockres->l_lock, flags);
cb25797d
JK		 1521#ifdef CONFIG_DEBUG_LOCK_ALLOC
1522 if (lockres->l_lockdep_map.key != NULL)
1523 rwsem_release(&lockres->l_lockdep_map, 1, caller_ip);
1524#endif
ccd979bd
MF		 1525 mlog_exit_void();
1526}
1527
da66116e
AB		 1528static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1529 struct ocfs2_lock_res *lockres,
1530 int ex,
1531 int local)
ccd979bd 1532{
bd3e7610 1533 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
ccd979bd 1534 unsigned long flags;
bd3e7610 1535 u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
ccd979bd
MF		 1536
1537 spin_lock_irqsave(&lockres->l_lock, flags);
1538 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1539 lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1540 spin_unlock_irqrestore(&lockres->l_lock, flags);
1541
24c19ef4 1542 return ocfs2_lock_create(osb, lockres, level, lkm_flags);
ccd979bd
MF		 1543}
1544
1545/* Grants us an EX lock on the data and metadata resources, skipping
1546 * the normal cluster directory lookup. Use this ONLY on newly created
1547 * inodes which other nodes can't possibly see, and which haven't been
1548 * hashed in the inode hash yet. This can give us a good performance
1549 * increase as it'll skip the network broadcast normally associated
1550 * with creating a new lock resource. */
1551int ocfs2_create_new_inode_locks(struct inode *inode)
1552{
1553 int ret;
d680efe9 1554 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1555
1556 BUG_ON(!inode);
1557 BUG_ON(!ocfs2_inode_is_new(inode));
1558
1559 mlog_entry_void();
1560
b0697053 1561 mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 1562
1563 /* NOTE: That we don't increment any of the holder counts, nor
1564 * do we add anything to a journal handle. Since this is
1565 * supposed to be a new inode which the cluster doesn't know
1566 * about yet, there is no need to. As far as the LVB handling
1567 * is concerned, this is basically like acquiring an EX lock
1568 * on a resource which has an invalid one -- we'll set it
1569 * valid when we release the EX. */
1570
24c19ef4 1571 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
ccd979bd
MF		 1572 if (ret) {
1573 mlog_errno(ret);
1574 goto bail;
1575 }
1576
24c19ef4 1577 /*
bd3e7610 1578 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
24c19ef4
MF		 1579 * don't use a generation in their lock names.
1580 */
e63aecb6 1581 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
ccd979bd
MF		 1582 if (ret) {
1583 mlog_errno(ret);
1584 goto bail;
1585 }
1586
50008630
TY		 1587 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1588 if (ret) {
1589 mlog_errno(ret);
1590 goto bail;
1591 }
1592
ccd979bd
MF		 1593bail:
1594 mlog_exit(ret);
1595 return ret;
1596}
1597
1598int ocfs2_rw_lock(struct inode *inode, int write)
1599{
1600 int status, level;
1601 struct ocfs2_lock_res *lockres;
c271c5c2 1602 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1603
1604 BUG_ON(!inode);
1605
1606 mlog_entry_void();
1607
b0697053
MF		 1608 mlog(0, "inode %llu take %s RW lock\n",
1609 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1610 write ? "EXMODE" : "PRMODE");
1611
d92bc512
CL		 1612 if (ocfs2_mount_local(osb)) {
1613 mlog_exit(0);
c271c5c2 1614 return 0;
d92bc512 1615 }
c271c5c2 1616
ccd979bd
MF		 1617 lockres = &OCFS2_I(inode)->ip_rw_lockres;
1618
bd3e7610 1619 level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
ccd979bd
MF		 1620
1621 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1622 0);
1623 if (status < 0)
1624 mlog_errno(status);
1625
1626 mlog_exit(status);
1627 return status;
1628}
1629
1630void ocfs2_rw_unlock(struct inode *inode, int write)
1631{
bd3e7610 1632 int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
ccd979bd 1633 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
c271c5c2 1634 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 1635
1636 mlog_entry_void();
1637
b0697053
MF		 1638 mlog(0, "inode %llu drop %s RW lock\n",
1639 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 1640 write ? "EXMODE" : "PRMODE");
1641
c271c5c2
SM		 1642 if (!ocfs2_mount_local(osb))
1643 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
ccd979bd
MF		 1644
1645 mlog_exit_void();
1646}
1647
50008630
TY		 1648/*
1649 * ocfs2_open_lock always get PR mode lock.
1650 */
1651int ocfs2_open_lock(struct inode *inode)
1652{
1653 int status = 0;
1654 struct ocfs2_lock_res *lockres;
1655 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1656
1657 BUG_ON(!inode);
1658
1659 mlog_entry_void();
1660
1661 mlog(0, "inode %llu take PRMODE open lock\n",
1662 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1663
1664 if (ocfs2_mount_local(osb))
1665 goto out;
1666
1667 lockres = &OCFS2_I(inode)->ip_open_lockres;
1668
1669 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
bd3e7610 1670 DLM_LOCK_PR, 0, 0);
50008630
TY		 1671 if (status < 0)
1672 mlog_errno(status);
1673
1674out:
1675 mlog_exit(status);
1676 return status;
1677}
1678
1679int ocfs2_try_open_lock(struct inode *inode, int write)
1680{
1681 int status = 0, level;
1682 struct ocfs2_lock_res *lockres;
1683 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1684
1685 BUG_ON(!inode);
1686
1687 mlog_entry_void();
1688
1689 mlog(0, "inode %llu try to take %s open lock\n",
1690 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1691 write ? "EXMODE" : "PRMODE");
1692
1693 if (ocfs2_mount_local(osb))
1694 goto out;
1695
1696 lockres = &OCFS2_I(inode)->ip_open_lockres;
1697
bd3e7610 1698 level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
50008630
TY		 1699
1700 /*
1701 * The file system may already holding a PRMODE/EXMODE open lock.
bd3e7610 1702 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
50008630
TY		 1703 * other nodes and the -EAGAIN will indicate to the caller that
1704 * this inode is still in use.
1705 */
1706 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
bd3e7610 1707 level, DLM_LKF_NOQUEUE, 0);
50008630
TY		 1708
1709out:
1710 mlog_exit(status);
1711 return status;
1712}
1713
1714/*
1715 * ocfs2_open_unlock unlock PR and EX mode open locks.
1716 */
1717void ocfs2_open_unlock(struct inode *inode)
1718{
1719 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1720 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1721
1722 mlog_entry_void();
1723
1724 mlog(0, "inode %llu drop open lock\n",
1725 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1726
1727 if (ocfs2_mount_local(osb))
1728 goto out;
1729
1730 if(lockres->l_ro_holders)
1731 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
bd3e7610 1732 DLM_LOCK_PR);
50008630
TY		 1733 if(lockres->l_ex_holders)
1734 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
bd3e7610 1735 DLM_LOCK_EX);
50008630
TY		 1736
1737out:
1738 mlog_exit_void();
1739}
1740
cf8e06f1
MF		 1741static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1742 int level)
1743{
1744 int ret;
1745 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1746 unsigned long flags;
1747 struct ocfs2_mask_waiter mw;
1748
1749 ocfs2_init_mask_waiter(&mw);
1750
1751retry_cancel:
1752 spin_lock_irqsave(&lockres->l_lock, flags);
1753 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1754 ret = ocfs2_prepare_cancel_convert(osb, lockres);
1755 if (ret) {
1756 spin_unlock_irqrestore(&lockres->l_lock, flags);
1757 ret = ocfs2_cancel_convert(osb, lockres);
1758 if (ret < 0) {
1759 mlog_errno(ret);
1760 goto out;
1761 }
1762 goto retry_cancel;
1763 }
1764 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1765 spin_unlock_irqrestore(&lockres->l_lock, flags);
1766
1767 ocfs2_wait_for_mask(&mw);
1768 goto retry_cancel;
1769 }
1770
1771 ret = -ERESTARTSYS;
1772 /*
1773 * We may still have gotten the lock, in which case there's no
1774 * point to restarting the syscall.
1775 */
1776 if (lockres->l_level == level)
1777 ret = 0;
1778
1779 mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1780 lockres->l_flags, lockres->l_level, lockres->l_action);
1781
1782 spin_unlock_irqrestore(&lockres->l_lock, flags);
1783
1784out:
1785 return ret;
1786}
1787
1788/*
1789 * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1790 * flock() calls. The locking approach this requires is sufficiently
1791 * different from all other cluster lock types that we implement a
1792 * seperate path to the "low-level" dlm calls. In particular:
1793 *
1794 * - No optimization of lock levels is done - we take at exactly
1795 * what's been requested.
1796 *
1797 * - No lock caching is employed. We immediately downconvert to
1798 * no-lock at unlock time. This also means flock locks never go on
1799 * the blocking list).
1800 *
1801 * - Since userspace can trivially deadlock itself with flock, we make
1802 * sure to allow cancellation of a misbehaving applications flock()
1803 * request.
1804 *
1805 * - Access to any flock lockres doesn't require concurrency, so we
1806 * can simplify the code by requiring the caller to guarantee
1807 * serialization of dlmglue flock calls.
1808 */
1809int ocfs2_file_lock(struct file *file, int ex, int trylock)
1810{
e988cf1c
MF		 1811 int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1812 unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
cf8e06f1
MF		 1813 unsigned long flags;
1814 struct ocfs2_file_private *fp = file->private_data;
1815 struct ocfs2_lock_res *lockres = &fp->fp_flock;
1816 struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1817 struct ocfs2_mask_waiter mw;
1818
1819 ocfs2_init_mask_waiter(&mw);
1820
1821 if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
bd3e7610 1822 (lockres->l_level > DLM_LOCK_NL)) {
cf8e06f1
MF		 1823 mlog(ML_ERROR,
1824 "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1825 "level: %u\n", lockres->l_name, lockres->l_flags,
1826 lockres->l_level);
1827 return -EINVAL;
1828 }
1829
1830 spin_lock_irqsave(&lockres->l_lock, flags);
1831 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1832 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1833 spin_unlock_irqrestore(&lockres->l_lock, flags);
1834
1835 /*
1836 * Get the lock at NLMODE to start - that way we
1837 * can cancel the upconvert request if need be.
1838 */
e988cf1c 1839 ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
cf8e06f1
MF		 1840 if (ret < 0) {
1841 mlog_errno(ret);
1842 goto out;
1843 }
1844
1845 ret = ocfs2_wait_for_mask(&mw);
1846 if (ret) {
1847 mlog_errno(ret);
1848 goto out;
1849 }
1850 spin_lock_irqsave(&lockres->l_lock, flags);
1851 }
1852
1853 lockres->l_action = OCFS2_AST_CONVERT;
e988cf1c 1854 lkm_flags |= DLM_LKF_CONVERT;
cf8e06f1
MF		 1855 lockres->l_requested = level;
1856 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1857
1858 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1859 spin_unlock_irqrestore(&lockres->l_lock, flags);
1860
4670c46d 1861 ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
24ef1815
JB		 1862 lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1,
1863 lockres);
7431cd7e
JB		 1864 if (ret) {
1865 if (!trylock || (ret != -EAGAIN)) {
24ef1815 1866 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
cf8e06f1
MF		 1867 ret = -EINVAL;
1868 }
1869
1870 ocfs2_recover_from_dlm_error(lockres, 1);
1871 lockres_remove_mask_waiter(lockres, &mw);
1872 goto out;
1873 }
1874
1875 ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
1876 if (ret == -ERESTARTSYS) {
1877 /*
1878 * Userspace can cause deadlock itself with
1879 * flock(). Current behavior locally is to allow the
1880 * deadlock, but abort the system call if a signal is
1881 * received. We follow this example, otherwise a
1882 * poorly written program could sit in kernel until
1883 * reboot.
1884 *
1885 * Handling this is a bit more complicated for Ocfs2
1886 * though. We can't exit this function with an
1887 * outstanding lock request, so a cancel convert is
1888 * required. We intentionally overwrite 'ret' - if the
1889 * cancel fails and the lock was granted, it's easier
af901ca1 1890 * to just bubble success back up to the user.
cf8e06f1
MF		 1891 */
1892 ret = ocfs2_flock_handle_signal(lockres, level);
1693a5c0
DT		 1893 } else if (!ret && (level > lockres->l_level)) {
1894 /* Trylock failed asynchronously */
1895 BUG_ON(!trylock);
1896 ret = -EAGAIN;
cf8e06f1
MF		 1897 }
1898
1899out:
1900
1901 mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
1902 lockres->l_name, ex, trylock, ret);
1903 return ret;
1904}
1905
1906void ocfs2_file_unlock(struct file *file)
1907{
1908 int ret;
de551246 1909 unsigned int gen;
cf8e06f1
MF		 1910 unsigned long flags;
1911 struct ocfs2_file_private *fp = file->private_data;
1912 struct ocfs2_lock_res *lockres = &fp->fp_flock;
1913 struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1914 struct ocfs2_mask_waiter mw;
1915
1916 ocfs2_init_mask_waiter(&mw);
1917
1918 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
1919 return;
1920
e988cf1c 1921 if (lockres->l_level == DLM_LOCK_NL)
cf8e06f1
MF		 1922 return;
1923
1924 mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
1925 lockres->l_name, lockres->l_flags, lockres->l_level,
1926 lockres->l_action);
1927
1928 spin_lock_irqsave(&lockres->l_lock, flags);
1929 /*
1930 * Fake a blocking ast for the downconvert code.
1931 */
1932 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
bd3e7610 1933 lockres->l_blocking = DLM_LOCK_EX;
cf8e06f1 1934
e988cf1c 1935 gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
cf8e06f1
MF		 1936 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1937 spin_unlock_irqrestore(&lockres->l_lock, flags);
1938
e988cf1c 1939 ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
cf8e06f1
MF		 1940 if (ret) {
1941 mlog_errno(ret);
1942 return;
1943 }
1944
1945 ret = ocfs2_wait_for_mask(&mw);
1946 if (ret)
1947 mlog_errno(ret);
1948}
1949
34d024f8
MF		 1950static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
1951 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 1952{
1953 int kick = 0;
1954
1955 mlog_entry_void();
1956
1957 /* If we know that another node is waiting on our lock, kick
34d024f8 1958 * the downconvert thread * pre-emptively when we reach a release
ccd979bd
MF		 1959 * condition. */
1960 if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1961 switch(lockres->l_blocking) {
bd3e7610 1962 case DLM_LOCK_EX:
ccd979bd
MF		 1963 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1964 kick = 1;
1965 break;
bd3e7610 1966 case DLM_LOCK_PR:
ccd979bd
MF		 1967 if (!lockres->l_ex_holders)
1968 kick = 1;
1969 break;
1970 default:
1971 BUG();
1972 }
1973 }
1974
1975 if (kick)
34d024f8 1976 ocfs2_wake_downconvert_thread(osb);
ccd979bd
MF		 1977
1978 mlog_exit_void();
1979}
1980
ccd979bd
MF		 1981#define OCFS2_SEC_BITS 34
1982#define OCFS2_SEC_SHIFT (64 - 34)
1983#define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1)
1984
1985/* LVB only has room for 64 bits of time here so we pack it for
1986 * now. */
1987static u64 ocfs2_pack_timespec(struct timespec *spec)
1988{
1989 u64 res;
1990 u64 sec = spec->tv_sec;
1991 u32 nsec = spec->tv_nsec;
1992
1993 res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
1994
1995 return res;
1996}
1997
1998/* Call this with the lockres locked. I am reasonably sure we don't
1999 * need ip_lock in this function as anyone who would be changing those
e63aecb6 2000 * values is supposed to be blocked in ocfs2_inode_lock right now. */
ccd979bd
MF		 2001static void __ocfs2_stuff_meta_lvb(struct inode *inode)
2002{
2003 struct ocfs2_inode_info *oi = OCFS2_I(inode);
e63aecb6 2004 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
ccd979bd
MF		 2005 struct ocfs2_meta_lvb *lvb;
2006
2007 mlog_entry_void();
2008
a641dc2a 2009 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
ccd979bd 2010
24c19ef4
MF		 2011 /*
2012 * Invalidate the LVB of a deleted inode - this way other
2013 * nodes are forced to go to disk and discover the new inode
2014 * status.
2015 */
2016 if (oi->ip_flags & OCFS2_INODE_DELETED) {
2017 lvb->lvb_version = 0;
2018 goto out;
2019 }
2020
4d3b83f7 2021 lvb->lvb_version = OCFS2_LVB_VERSION;
ccd979bd
MF		 2022 lvb->lvb_isize = cpu_to_be64(i_size_read(inode));
2023 lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
2024 lvb->lvb_iuid = cpu_to_be32(inode->i_uid);
2025 lvb->lvb_igid = cpu_to_be32(inode->i_gid);
2026 lvb->lvb_imode = cpu_to_be16(inode->i_mode);
2027 lvb->lvb_inlink = cpu_to_be16(inode->i_nlink);
2028 lvb->lvb_iatime_packed =
2029 cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
2030 lvb->lvb_ictime_packed =
2031 cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
2032 lvb->lvb_imtime_packed =
2033 cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
ca4d147e 2034 lvb->lvb_iattr = cpu_to_be32(oi->ip_attr);
15b1e36b 2035 lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
f9e2d82e 2036 lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
ccd979bd 2037
24c19ef4 2038out:
ccd979bd
MF		 2039 mlog_meta_lvb(0, lockres);
2040
2041 mlog_exit_void();
2042}
2043
2044static void ocfs2_unpack_timespec(struct timespec *spec,
2045 u64 packed_time)
2046{
2047 spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
2048 spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
2049}
2050
2051static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
2052{
2053 struct ocfs2_inode_info *oi = OCFS2_I(inode);
e63aecb6 2054 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
ccd979bd
MF		 2055 struct ocfs2_meta_lvb *lvb;
2056
2057 mlog_entry_void();
2058
2059 mlog_meta_lvb(0, lockres);
2060
a641dc2a 2061 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
ccd979bd
MF		 2062
2063 /* We're safe here without the lockres lock... */
2064 spin_lock(&oi->ip_lock);
2065 oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
2066 i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
2067
ca4d147e 2068 oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
15b1e36b 2069 oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
ca4d147e
HP		 2070 ocfs2_set_inode_flags(inode);
2071
ccd979bd
MF		 2072 /* fast-symlinks are a special case */
2073 if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
2074 inode->i_blocks = 0;
2075 else
8110b073 2076 inode->i_blocks = ocfs2_inode_sector_count(inode);
ccd979bd
MF		 2077
2078 inode->i_uid = be32_to_cpu(lvb->lvb_iuid);
2079 inode->i_gid = be32_to_cpu(lvb->lvb_igid);
2080 inode->i_mode = be16_to_cpu(lvb->lvb_imode);
2081 inode->i_nlink = be16_to_cpu(lvb->lvb_inlink);
2082 ocfs2_unpack_timespec(&inode->i_atime,
2083 be64_to_cpu(lvb->lvb_iatime_packed));
2084 ocfs2_unpack_timespec(&inode->i_mtime,
2085 be64_to_cpu(lvb->lvb_imtime_packed));
2086 ocfs2_unpack_timespec(&inode->i_ctime,
2087 be64_to_cpu(lvb->lvb_ictime_packed));
2088 spin_unlock(&oi->ip_lock);
2089
2090 mlog_exit_void();
2091}
2092
f9e2d82e
MF		 2093static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
2094 struct ocfs2_lock_res *lockres)
ccd979bd 2095{
a641dc2a 2096 struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
ccd979bd 2097
1c520dfb
JB		 2098 if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
2099 && lvb->lvb_version == OCFS2_LVB_VERSION
f9e2d82e 2100 && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
ccd979bd
MF		 2101 return 1;
2102 return 0;
2103}
2104
2105/* Determine whether a lock resource needs to be refreshed, and
2106 * arbitrate who gets to refresh it.
2107 *
2108 * 0 means no refresh needed.
2109 *
2110 * > 0 means you need to refresh this and you MUST call
2111 * ocfs2_complete_lock_res_refresh afterwards. */
2112static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
2113{
2114 unsigned long flags;
2115 int status = 0;
2116
2117 mlog_entry_void();
2118
2119refresh_check:
2120 spin_lock_irqsave(&lockres->l_lock, flags);
2121 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2122 spin_unlock_irqrestore(&lockres->l_lock, flags);
2123 goto bail;
2124 }
2125
2126 if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2127 spin_unlock_irqrestore(&lockres->l_lock, flags);
2128
2129 ocfs2_wait_on_refreshing_lock(lockres);
2130 goto refresh_check;
2131 }
2132
2133 /* Ok, I'll be the one to refresh this lock. */
2134 lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2135 spin_unlock_irqrestore(&lockres->l_lock, flags);
2136
2137 status = 1;
2138bail:
2139 mlog_exit(status);
2140 return status;
2141}
2142
2143/* If status is non zero, I'll mark it as not being in refresh
2144 * anymroe, but i won't clear the needs refresh flag. */
2145static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2146 int status)
2147{
2148 unsigned long flags;
2149 mlog_entry_void();
2150
2151 spin_lock_irqsave(&lockres->l_lock, flags);
2152 lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2153 if (!status)
2154 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2155 spin_unlock_irqrestore(&lockres->l_lock, flags);
2156
2157 wake_up(&lockres->l_event);
2158
2159 mlog_exit_void();
2160}
2161
2162/* may or may not return a bh if it went to disk. */
e63aecb6 2163static int ocfs2_inode_lock_update(struct inode *inode,
ccd979bd
MF		 2164 struct buffer_head **bh)
2165{
2166 int status = 0;
2167 struct ocfs2_inode_info *oi = OCFS2_I(inode);
e63aecb6 2168 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
ccd979bd 2169 struct ocfs2_dinode *fe;
c271c5c2 2170 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 2171
2172 mlog_entry_void();
2173
be9e986b
MF		 2174 if (ocfs2_mount_local(osb))
2175 goto bail;
2176
ccd979bd
MF		 2177 spin_lock(&oi->ip_lock);
2178 if (oi->ip_flags & OCFS2_INODE_DELETED) {
b0697053 2179 mlog(0, "Orphaned inode %llu was deleted while we "
ccd979bd 2180 "were waiting on a lock. ip_flags = 0x%x\n",
b0697053 2181 (unsigned long long)oi->ip_blkno, oi->ip_flags);
ccd979bd
MF		 2182 spin_unlock(&oi->ip_lock);
2183 status = -ENOENT;
2184 goto bail;
2185 }
2186 spin_unlock(&oi->ip_lock);
2187
be9e986b
MF		 2188 if (!ocfs2_should_refresh_lock_res(lockres))
2189 goto bail;
ccd979bd
MF		 2190
2191 /* This will discard any caching information we might have had
2192 * for the inode metadata. */
8cb471e8 2193 ocfs2_metadata_cache_purge(INODE_CACHE(inode));
ccd979bd 2194
83418978
MF		 2195 ocfs2_extent_map_trunc(inode, 0);
2196
be9e986b 2197 if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
b0697053
MF		 2198 mlog(0, "Trusting LVB on inode %llu\n",
2199 (unsigned long long)oi->ip_blkno);
ccd979bd
MF		 2200 ocfs2_refresh_inode_from_lvb(inode);
2201 } else {
2202 /* Boo, we have to go to disk. */
2203 /* read bh, cast, ocfs2_refresh_inode */
b657c95c 2204 status = ocfs2_read_inode_block(inode, bh);
ccd979bd
MF		 2205 if (status < 0) {
2206 mlog_errno(status);
2207 goto bail_refresh;
2208 }
2209 fe = (struct ocfs2_dinode *) (*bh)->b_data;
2210
2211 /* This is a good chance to make sure we're not
b657c95c
JB		 2212 * locking an invalid object. ocfs2_read_inode_block()
2213 * already checked that the inode block is sane.
ccd979bd
MF		 2214 *
2215 * We bug on a stale inode here because we checked
2216 * above whether it was wiped from disk. The wiping
2217 * node provides a guarantee that we receive that
2218 * message and can mark the inode before dropping any
2219 * locks associated with it. */
ccd979bd
MF		 2220 mlog_bug_on_msg(inode->i_generation !=
2221 le32_to_cpu(fe->i_generation),
b0697053 2222 "Invalid dinode %llu disk generation: %u "
ccd979bd 2223 "inode->i_generation: %u\n",
b0697053
MF		 2224 (unsigned long long)oi->ip_blkno,
2225 le32_to_cpu(fe->i_generation),
ccd979bd
MF		 2226 inode->i_generation);
2227 mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2228 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
b0697053
MF		 2229 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
2230 (unsigned long long)oi->ip_blkno,
2231 (unsigned long long)le64_to_cpu(fe->i_dtime),
ccd979bd
MF		 2232 le32_to_cpu(fe->i_flags));
2233
2234 ocfs2_refresh_inode(inode, fe);
8ddb7b00 2235 ocfs2_track_lock_refresh(lockres);
ccd979bd
MF		 2236 }
2237
2238 status = 0;
2239bail_refresh:
be9e986b 2240 ocfs2_complete_lock_res_refresh(lockres, status);
ccd979bd
MF		 2241bail:
2242 mlog_exit(status);
2243 return status;
2244}
2245
2246static int ocfs2_assign_bh(struct inode *inode,
2247 struct buffer_head **ret_bh,
2248 struct buffer_head *passed_bh)
2249{
2250 int status;
2251
2252 if (passed_bh) {
2253 /* Ok, the update went to disk for us, use the
2254 * returned bh. */
2255 *ret_bh = passed_bh;
2256 get_bh(*ret_bh);
2257
2258 return 0;
2259 }
2260
b657c95c 2261 status = ocfs2_read_inode_block(inode, ret_bh);
ccd979bd
MF		 2262 if (status < 0)
2263 mlog_errno(status);
2264
2265 return status;
2266}
2267
2268/*
2269 * returns < 0 error if the callback will never be called, otherwise
2270 * the result of the lock will be communicated via the callback.
2271 */
cb25797d
JK		 2272int ocfs2_inode_lock_full_nested(struct inode *inode,
2273 struct buffer_head **ret_bh,
2274 int ex,
2275 int arg_flags,
2276 int subclass)
ccd979bd 2277{
bd3e7610
JB		 2278 int status, level, acquired;
2279 u32 dlm_flags;
c271c5c2 2280 struct ocfs2_lock_res *lockres = NULL;
ccd979bd
MF		 2281 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2282 struct buffer_head *local_bh = NULL;
2283
2284 BUG_ON(!inode);
2285
2286 mlog_entry_void();
2287
b0697053
MF		 2288 mlog(0, "inode %llu, take %s META lock\n",
2289 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 2290 ex ? "EXMODE" : "PRMODE");
2291
2292 status = 0;
2293 acquired = 0;
2294 /* We'll allow faking a readonly metadata lock for
2295 * rodevices. */
2296 if (ocfs2_is_hard_readonly(osb)) {
2297 if (ex)
2298 status = -EROFS;
2299 goto bail;
2300 }
2301
c271c5c2
SM		 2302 if (ocfs2_mount_local(osb))
2303 goto local;
2304
ccd979bd 2305 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
553abd04 2306 ocfs2_wait_for_recovery(osb);
ccd979bd 2307
e63aecb6 2308 lockres = &OCFS2_I(inode)->ip_inode_lockres;
bd3e7610 2309 level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
ccd979bd
MF		 2310 dlm_flags = 0;
2311 if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
bd3e7610 2312 dlm_flags |= DLM_LKF_NOQUEUE;
ccd979bd 2313
cb25797d
JK		 2314 status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
2315 arg_flags, subclass, _RET_IP_);
ccd979bd
MF		 2316 if (status < 0) {
2317 if (status != -EAGAIN && status != -EIOCBRETRY)
2318 mlog_errno(status);
2319 goto bail;
2320 }
2321
2322 /* Notify the error cleanup path to drop the cluster lock. */
2323 acquired = 1;
2324
2325 /* We wait twice because a node may have died while we were in
2326 * the lower dlm layers. The second time though, we've
2327 * committed to owning this lock so we don't allow signals to
2328 * abort the operation. */
2329 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
553abd04 2330 ocfs2_wait_for_recovery(osb);
ccd979bd 2331
c271c5c2 2332local:
24c19ef4
MF		 2333 /*
2334 * We only see this flag if we're being called from
2335 * ocfs2_read_locked_inode(). It means we're locking an inode
2336 * which hasn't been populated yet, so clear the refresh flag
2337 * and let the caller handle it.
2338 */
2339 if (inode->i_state & I_NEW) {
2340 status = 0;
c271c5c2
SM		 2341 if (lockres)
2342 ocfs2_complete_lock_res_refresh(lockres, 0);
24c19ef4
MF		 2343 goto bail;
2344 }
2345
ccd979bd 2346 /* This is fun. The caller may want a bh back, or it may
e63aecb6 2347 * not. ocfs2_inode_lock_update definitely wants one in, but
ccd979bd
MF		 2348 * may or may not read one, depending on what's in the
2349 * LVB. The result of all of this is that we've *only* gone to
2350 * disk if we have to, so the complexity is worthwhile. */
e63aecb6 2351 status = ocfs2_inode_lock_update(inode, &local_bh);
ccd979bd
MF		 2352 if (status < 0) {
2353 if (status != -ENOENT)
2354 mlog_errno(status);
2355 goto bail;
2356 }
2357
2358 if (ret_bh) {
2359 status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2360 if (status < 0) {
2361 mlog_errno(status);
2362 goto bail;
2363 }
2364 }
2365
ccd979bd
MF		 2366bail:
2367 if (status < 0) {
2368 if (ret_bh && (*ret_bh)) {
2369 brelse(*ret_bh);
2370 *ret_bh = NULL;
2371 }
2372 if (acquired)
e63aecb6 2373 ocfs2_inode_unlock(inode, ex);
ccd979bd
MF		 2374 }
2375
2376 if (local_bh)
2377 brelse(local_bh);
2378
2379 mlog_exit(status);
2380 return status;
2381}
2382
2383/*
34d024f8
MF		 2384 * This is working around a lock inversion between tasks acquiring DLM
2385 * locks while holding a page lock and the downconvert thread which
2386 * blocks dlm lock acquiry while acquiring page locks.
ccd979bd
MF		 2387 *
2388 * ** These _with_page variantes are only intended to be called from aop
2389 * methods that hold page locks and return a very specific *positive* error
2390 * code that aop methods pass up to the VFS -- test for errors with != 0. **
2391 *
34d024f8
MF		 2392 * The DLM is called such that it returns -EAGAIN if it would have
2393 * blocked waiting for the downconvert thread. In that case we unlock
2394 * our page so the downconvert thread can make progress. Once we've
2395 * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2396 * that called us can bubble that back up into the VFS who will then
2397 * immediately retry the aop call.
ccd979bd
MF		 2398 *
2399 * We do a blocking lock and immediate unlock before returning, though, so that
2400 * the lock has a great chance of being cached on this node by the time the VFS
2401 * calls back to retry the aop. This has a potential to livelock as nodes
2402 * ping locks back and forth, but that's a risk we're willing to take to avoid
2403 * the lock inversion simply.
2404 */
e63aecb6 2405int ocfs2_inode_lock_with_page(struct inode *inode,
ccd979bd
MF		 2406 struct buffer_head **ret_bh,
2407 int ex,
2408 struct page *page)
2409{
2410 int ret;
2411
e63aecb6 2412 ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
ccd979bd
MF		 2413 if (ret == -EAGAIN) {
2414 unlock_page(page);
e63aecb6
MF		 2415 if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2416 ocfs2_inode_unlock(inode, ex);
ccd979bd
MF		 2417 ret = AOP_TRUNCATED_PAGE;
2418 }
2419
2420 return ret;
2421}
2422
e63aecb6 2423int ocfs2_inode_lock_atime(struct inode *inode,
7f1a37e3
TY		 2424 struct vfsmount *vfsmnt,
2425 int *level)
2426{
2427 int ret;
2428
2429 mlog_entry_void();
e63aecb6 2430 ret = ocfs2_inode_lock(inode, NULL, 0);
7f1a37e3
TY		 2431 if (ret < 0) {
2432 mlog_errno(ret);
2433 return ret;
2434 }
2435
2436 /*
2437 * If we should update atime, we will get EX lock,
2438 * otherwise we just get PR lock.
2439 */
2440 if (ocfs2_should_update_atime(inode, vfsmnt)) {
2441 struct buffer_head *bh = NULL;
2442
e63aecb6
MF		 2443 ocfs2_inode_unlock(inode, 0);
2444 ret = ocfs2_inode_lock(inode, &bh, 1);
7f1a37e3
TY		 2445 if (ret < 0) {
2446 mlog_errno(ret);
2447 return ret;
2448 }
2449 *level = 1;
2450 if (ocfs2_should_update_atime(inode, vfsmnt))
2451 ocfs2_update_inode_atime(inode, bh);
2452 if (bh)
2453 brelse(bh);
2454 } else
2455 *level = 0;
2456
2457 mlog_exit(ret);
2458 return ret;
2459}
2460
e63aecb6 2461void ocfs2_inode_unlock(struct inode *inode,
ccd979bd
MF		 2462 int ex)
2463{
bd3e7610 2464 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
e63aecb6 2465 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
c271c5c2 2466 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
ccd979bd
MF		 2467
2468 mlog_entry_void();
2469
b0697053
MF		 2470 mlog(0, "inode %llu drop %s META lock\n",
2471 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ccd979bd
MF		 2472 ex ? "EXMODE" : "PRMODE");
2473
c271c5c2
SM		 2474 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
2475 !ocfs2_mount_local(osb))
ccd979bd
MF		 2476 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
2477
2478 mlog_exit_void();
2479}
2480
df152c24 2481int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
83273932
SE		 2482{
2483 struct ocfs2_lock_res *lockres;
2484 struct ocfs2_orphan_scan_lvb *lvb;
83273932
SE		 2485 int status = 0;
2486
df152c24
SM		 2487 if (ocfs2_is_hard_readonly(osb))
2488 return -EROFS;
2489
2490 if (ocfs2_mount_local(osb))
2491 return 0;
2492
83273932 2493 lockres = &osb->osb_orphan_scan.os_lockres;
df152c24 2494 status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
83273932
SE		 2495 if (status < 0)
2496 return status;
2497
2498 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
1c520dfb
JB		 2499 if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2500 lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
83273932 2501 *seqno = be32_to_cpu(lvb->lvb_os_seqno);
3211949f
SM		 2502 else
2503 *seqno = osb->osb_orphan_scan.os_seqno + 1;
2504
83273932
SE		 2505 return status;
2506}
2507
df152c24 2508void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
83273932
SE		 2509{
2510 struct ocfs2_lock_res *lockres;
2511 struct ocfs2_orphan_scan_lvb *lvb;
83273932 2512
df152c24
SM		 2513 if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
2514 lockres = &osb->osb_orphan_scan.os_lockres;
2515 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2516 lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
2517 lvb->lvb_os_seqno = cpu_to_be32(seqno);
2518 ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2519 }
83273932
SE		 2520}
2521
ccd979bd
MF		 2522int ocfs2_super_lock(struct ocfs2_super *osb,
2523 int ex)
2524{
c271c5c2 2525 int status = 0;
bd3e7610 2526 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
ccd979bd 2527 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
ccd979bd
MF		 2528
2529 mlog_entry_void();
2530
2531 if (ocfs2_is_hard_readonly(osb))
2532 return -EROFS;
2533
c271c5c2
SM		 2534 if (ocfs2_mount_local(osb))
2535 goto bail;
2536
ccd979bd
MF		 2537 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2538 if (status < 0) {
2539 mlog_errno(status);
2540 goto bail;
2541 }
2542
2543 /* The super block lock path is really in the best position to
2544 * know when resources covered by the lock need to be
2545 * refreshed, so we do it here. Of course, making sense of
2546 * everything is up to the caller :) */
2547 status = ocfs2_should_refresh_lock_res(lockres);
2548 if (status < 0) {
2549 mlog_errno(status);
2550 goto bail;
2551 }
2552 if (status) {
8e8a4603 2553 status = ocfs2_refresh_slot_info(osb);
ccd979bd
MF		 2554
2555 ocfs2_complete_lock_res_refresh(lockres, status);
2556
2557 if (status < 0)
2558 mlog_errno(status);
8ddb7b00 2559 ocfs2_track_lock_refresh(lockres);
ccd979bd
MF		 2560 }
2561bail:
2562 mlog_exit(status);
2563 return status;
2564}
2565
2566void ocfs2_super_unlock(struct ocfs2_super *osb,
2567 int ex)
2568{
bd3e7610 2569 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
ccd979bd
MF		 2570 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2571
c271c5c2
SM		 2572 if (!ocfs2_mount_local(osb))
2573 ocfs2_cluster_unlock(osb, lockres, level);
ccd979bd
MF		 2574}
2575
2576int ocfs2_rename_lock(struct ocfs2_super *osb)
2577{
2578 int status;
2579 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2580
2581 if (ocfs2_is_hard_readonly(osb))
2582 return -EROFS;
2583
c271c5c2
SM		 2584 if (ocfs2_mount_local(osb))
2585 return 0;
2586
bd3e7610 2587 status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
ccd979bd
MF		 2588 if (status < 0)
2589 mlog_errno(status);
2590
2591 return status;
2592}
2593
2594void ocfs2_rename_unlock(struct ocfs2_super *osb)
2595{
2596 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2597
c271c5c2 2598 if (!ocfs2_mount_local(osb))
bd3e7610 2599 ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
ccd979bd
MF		 2600}
2601
6ca497a8 2602int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2603{
2604 int status;
2605 struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2606
2607 if (ocfs2_is_hard_readonly(osb))
2608 return -EROFS;
2609
2610 if (ocfs2_mount_local(osb))
2611 return 0;
2612
2613 status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2614 0, 0);
2615 if (status < 0)
2616 mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2617
2618 return status;
2619}
2620
2621void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2622{
2623 struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2624
2625 if (!ocfs2_mount_local(osb))
2626 ocfs2_cluster_unlock(osb, lockres,
2627 ex ? LKM_EXMODE : LKM_PRMODE);
2628}
2629
d680efe9
MF		 2630int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2631{
2632 int ret;
bd3e7610 2633 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
d680efe9
MF		 2634 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2635 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2636
2637 BUG_ON(!dl);
2638
2639 if (ocfs2_is_hard_readonly(osb))
2640 return -EROFS;
2641
c271c5c2
SM		 2642 if (ocfs2_mount_local(osb))
2643 return 0;
2644
d680efe9
MF		 2645 ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2646 if (ret < 0)
2647 mlog_errno(ret);
2648
2649 return ret;
2650}
2651
2652void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2653{
bd3e7610 2654 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
d680efe9
MF		 2655 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2656 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2657
c271c5c2
SM		 2658 if (!ocfs2_mount_local(osb))
2659 ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
d680efe9
MF		 2660}
2661
ccd979bd
MF		 2662/* Reference counting of the dlm debug structure. We want this because
2663 * open references on the debug inodes can live on after a mount, so
2664 * we can't rely on the ocfs2_super to always exist. */
2665static void ocfs2_dlm_debug_free(struct kref *kref)
2666{
2667 struct ocfs2_dlm_debug *dlm_debug;
2668
2669 dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2670
2671 kfree(dlm_debug);
2672}
2673
2674void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2675{
2676 if (dlm_debug)
2677 kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2678}
2679
2680static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2681{
2682 kref_get(&debug->d_refcnt);
2683}
2684
2685struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2686{
2687 struct ocfs2_dlm_debug *dlm_debug;
2688
2689 dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2690 if (!dlm_debug) {
2691 mlog_errno(-ENOMEM);
2692 goto out;
2693 }
2694
2695 kref_init(&dlm_debug->d_refcnt);
2696 INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2697 dlm_debug->d_locking_state = NULL;
2698out:
2699 return dlm_debug;
2700}
2701
2702/* Access to this is arbitrated for us via seq_file->sem. */
2703struct ocfs2_dlm_seq_priv {
2704 struct ocfs2_dlm_debug *p_dlm_debug;
2705 struct ocfs2_lock_res p_iter_res;
2706 struct ocfs2_lock_res p_tmp_res;
2707};
2708
2709static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2710 struct ocfs2_dlm_seq_priv *priv)
2711{
2712 struct ocfs2_lock_res *iter, *ret = NULL;
2713 struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2714
2715 assert_spin_locked(&ocfs2_dlm_tracking_lock);
2716
2717 list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2718 /* discover the head of the list */
2719 if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2720 mlog(0, "End of list found, %p\n", ret);
2721 break;
2722 }
2723
2724 /* We track our "dummy" iteration lockres' by a NULL
2725 * l_ops field. */
2726 if (iter->l_ops != NULL) {
2727 ret = iter;
2728 break;
2729 }
2730 }
2731
2732 return ret;
2733}
2734
2735static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2736{
2737 struct ocfs2_dlm_seq_priv *priv = m->private;
2738 struct ocfs2_lock_res *iter;
2739
2740 spin_lock(&ocfs2_dlm_tracking_lock);
2741 iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2742 if (iter) {
2743 /* Since lockres' have the lifetime of their container
2744 * (which can be inodes, ocfs2_supers, etc) we want to
2745 * copy this out to a temporary lockres while still
2746 * under the spinlock. Obviously after this we can't
2747 * trust any pointers on the copy returned, but that's
2748 * ok as the information we want isn't typically held
2749 * in them. */
2750 priv->p_tmp_res = *iter;
2751 iter = &priv->p_tmp_res;
2752 }
2753 spin_unlock(&ocfs2_dlm_tracking_lock);
2754
2755 return iter;
2756}
2757
2758static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2759{
2760}
2761
2762static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2763{
2764 struct ocfs2_dlm_seq_priv *priv = m->private;
2765 struct ocfs2_lock_res *iter = v;
2766 struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2767
2768 spin_lock(&ocfs2_dlm_tracking_lock);
2769 iter = ocfs2_dlm_next_res(iter, priv);
2770 list_del_init(&dummy->l_debug_list);
2771 if (iter) {
2772 list_add(&dummy->l_debug_list, &iter->l_debug_list);
2773 priv->p_tmp_res = *iter;
2774 iter = &priv->p_tmp_res;
2775 }
2776 spin_unlock(&ocfs2_dlm_tracking_lock);
2777
2778 return iter;
2779}
2780
2781/* So that debugfs.ocfs2 can determine which format is being used */
8ddb7b00 2782#define OCFS2_DLM_DEBUG_STR_VERSION 2
ccd979bd
MF		 2783static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2784{
2785 int i;
2786 char *lvb;
2787 struct ocfs2_lock_res *lockres = v;
2788
2789 if (!lockres)
2790 return -EINVAL;
2791
d680efe9
MF		 2792 seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2793
2794 if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2795 seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2796 lockres->l_name,
2797 (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2798 else
2799 seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2800
2801 seq_printf(m, "%d\t"
ccd979bd
MF		 2802 "0x%lx\t"
2803 "0x%x\t"
2804 "0x%x\t"
2805 "%u\t"
2806 "%u\t"
2807 "%d\t"
2808 "%d\t",
ccd979bd
MF		 2809 lockres->l_level,
2810 lockres->l_flags,
2811 lockres->l_action,
2812 lockres->l_unlock_action,
2813 lockres->l_ro_holders,
2814 lockres->l_ex_holders,
2815 lockres->l_requested,
2816 lockres->l_blocking);
2817
2818 /* Dump the raw LVB */
8f2c9c1b 2819 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
ccd979bd
MF		 2820 for(i = 0; i < DLM_LVB_LEN; i++)
2821 seq_printf(m, "0x%x\t", lvb[i]);
2822
8ddb7b00
SM		 2823#ifdef CONFIG_OCFS2_FS_STATS
2824# define lock_num_prmode(_l) (_l)->l_lock_num_prmode
2825# define lock_num_exmode(_l) (_l)->l_lock_num_exmode
2826# define lock_num_prmode_failed(_l) (_l)->l_lock_num_prmode_failed
2827# define lock_num_exmode_failed(_l) (_l)->l_lock_num_exmode_failed
2828# define lock_total_prmode(_l) (_l)->l_lock_total_prmode
2829# define lock_total_exmode(_l) (_l)->l_lock_total_exmode
2830# define lock_max_prmode(_l) (_l)->l_lock_max_prmode
2831# define lock_max_exmode(_l) (_l)->l_lock_max_exmode
2832# define lock_refresh(_l) (_l)->l_lock_refresh
2833#else
dd25e55e
RD		 2834# define lock_num_prmode(_l) (0ULL)
2835# define lock_num_exmode(_l) (0ULL)
8ddb7b00
SM		 2836# define lock_num_prmode_failed(_l) (0)
2837# define lock_num_exmode_failed(_l) (0)
dd25e55e
RD		 2838# define lock_total_prmode(_l) (0ULL)
2839# define lock_total_exmode(_l) (0ULL)
8ddb7b00
SM		 2840# define lock_max_prmode(_l) (0)
2841# define lock_max_exmode(_l) (0)
2842# define lock_refresh(_l) (0)
2843#endif
2844 /* The following seq_print was added in version 2 of this output */
2845 seq_printf(m, "%llu\t"
2846 "%llu\t"
2847 "%u\t"
2848 "%u\t"
2849 "%llu\t"
2850 "%llu\t"
2851 "%u\t"
2852 "%u\t"
2853 "%u\t",
2854 lock_num_prmode(lockres),
2855 lock_num_exmode(lockres),
2856 lock_num_prmode_failed(lockres),
2857 lock_num_exmode_failed(lockres),
2858 lock_total_prmode(lockres),
2859 lock_total_exmode(lockres),
2860 lock_max_prmode(lockres),
2861 lock_max_exmode(lockres),
2862 lock_refresh(lockres));
2863
ccd979bd
MF		 2864 /* End the line */
2865 seq_printf(m, "\n");
2866 return 0;
2867}
2868
90d99779 2869static const struct seq_operations ocfs2_dlm_seq_ops = {
ccd979bd
MF		 2870 .start = ocfs2_dlm_seq_start,
2871 .stop = ocfs2_dlm_seq_stop,
2872 .next = ocfs2_dlm_seq_next,
2873 .show = ocfs2_dlm_seq_show,
2874};
2875
2876static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2877{
2878 struct seq_file *seq = (struct seq_file *) file->private_data;
2879 struct ocfs2_dlm_seq_priv *priv = seq->private;
2880 struct ocfs2_lock_res *res = &priv->p_iter_res;
2881
2882 ocfs2_remove_lockres_tracking(res);
2883 ocfs2_put_dlm_debug(priv->p_dlm_debug);
2884 return seq_release_private(inode, file);
2885}
2886
2887static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2888{
2889 int ret;
2890 struct ocfs2_dlm_seq_priv *priv;
2891 struct seq_file *seq;
2892 struct ocfs2_super *osb;
2893
2894 priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2895 if (!priv) {
2896 ret = -ENOMEM;
2897 mlog_errno(ret);
2898 goto out;
2899 }
8e18e294 2900 osb = inode->i_private;
ccd979bd
MF		 2901 ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2902 priv->p_dlm_debug = osb->osb_dlm_debug;
2903 INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2904
2905 ret = seq_open(file, &ocfs2_dlm_seq_ops);
2906 if (ret) {
2907 kfree(priv);
2908 mlog_errno(ret);
2909 goto out;
2910 }
2911
2912 seq = (struct seq_file *) file->private_data;
2913 seq->private = priv;
2914
2915 ocfs2_add_lockres_tracking(&priv->p_iter_res,
2916 priv->p_dlm_debug);
2917
2918out:
2919 return ret;
2920}
2921
4b6f5d20 2922static const struct file_operations ocfs2_dlm_debug_fops = {
ccd979bd
MF		 2923 .open = ocfs2_dlm_debug_open,
2924 .release = ocfs2_dlm_debug_release,
2925 .read = seq_read,
2926 .llseek = seq_lseek,
2927};
2928
2929static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2930{
2931 int ret = 0;
2932 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2933
2934 dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2935 S_IFREG|S_IRUSR,
2936 osb->osb_debug_root,
2937 osb,
2938 &ocfs2_dlm_debug_fops);
2939 if (!dlm_debug->d_locking_state) {
2940 ret = -EINVAL;
2941 mlog(ML_ERROR,
2942 "Unable to create locking state debugfs file.\n");
2943 goto out;
2944 }
2945
2946 ocfs2_get_dlm_debug(dlm_debug);
2947out:
2948 return ret;
2949}
2950
2951static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2952{
2953 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2954
2955 if (dlm_debug) {
2956 debugfs_remove(dlm_debug->d_locking_state);
2957 ocfs2_put_dlm_debug(dlm_debug);
2958 }
2959}
2960
2961int ocfs2_dlm_init(struct ocfs2_super *osb)
2962{
c271c5c2 2963 int status = 0;
4670c46d 2964 struct ocfs2_cluster_connection *conn = NULL;
ccd979bd
MF		 2965
2966 mlog_entry_void();
2967
0abd6d18
MF		 2968 if (ocfs2_mount_local(osb)) {
2969 osb->node_num = 0;
c271c5c2 2970 goto local;
0abd6d18 2971 }
c271c5c2 2972
ccd979bd
MF		 2973 status = ocfs2_dlm_init_debug(osb);
2974 if (status < 0) {
2975 mlog_errno(status);
2976 goto bail;
2977 }
2978
34d024f8
MF		 2979 /* launch downconvert thread */
2980 osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc");
2981 if (IS_ERR(osb->dc_task)) {
2982 status = PTR_ERR(osb->dc_task);
2983 osb->dc_task = NULL;
ccd979bd
MF		 2984 mlog_errno(status);
2985 goto bail;
2986 }
2987
ccd979bd 2988 /* for now, uuid == domain */
9c6c877c
JB		 2989 status = ocfs2_cluster_connect(osb->osb_cluster_stack,
2990 osb->uuid_str,
4670c46d
JB		 2991 strlen(osb->uuid_str),
2992 ocfs2_do_node_down, osb,
2993 &conn);
2994 if (status) {
ccd979bd
MF		 2995 mlog_errno(status);
2996 goto bail;
2997 }
2998
0abd6d18
MF		 2999 status = ocfs2_cluster_this_node(&osb->node_num);
3000 if (status < 0) {
3001 mlog_errno(status);
3002 mlog(ML_ERROR,
3003 "could not find this host's node number\n");
286eaa95 3004 ocfs2_cluster_disconnect(conn, 0);
0abd6d18
MF		 3005 goto bail;
3006 }
3007
c271c5c2 3008local:
ccd979bd
MF		 3009 ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3010 ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
6ca497a8 3011 ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
83273932 3012 ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
ccd979bd 3013
4670c46d 3014 osb->cconn = conn;
ccd979bd
MF		 3015
3016 status = 0;
3017bail:
3018 if (status < 0) {
3019 ocfs2_dlm_shutdown_debug(osb);
34d024f8
MF		 3020 if (osb->dc_task)
3021 kthread_stop(osb->dc_task);
ccd979bd
MF		 3022 }
3023
3024 mlog_exit(status);
3025 return status;
3026}
3027
286eaa95
JB		 3028void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3029 int hangup_pending)
ccd979bd
MF		 3030{
3031 mlog_entry_void();
3032
ccd979bd
MF		 3033 ocfs2_drop_osb_locks(osb);
3034
4670c46d
JB		 3035 /*
3036 * Now that we have dropped all locks and ocfs2_dismount_volume()
3037 * has disabled recovery, the DLM won't be talking to us. It's
3038 * safe to tear things down before disconnecting the cluster.
3039 */
3040
34d024f8
MF		 3041 if (osb->dc_task) {
3042 kthread_stop(osb->dc_task);
3043 osb->dc_task = NULL;
ccd979bd
MF		 3044 }
3045
3046 ocfs2_lock_res_free(&osb->osb_super_lockres);
3047 ocfs2_lock_res_free(&osb->osb_rename_lockres);
6ca497a8 3048 ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
83273932 3049 ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
ccd979bd 3050
286eaa95 3051 ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
4670c46d 3052 osb->cconn = NULL;
ccd979bd
MF		 3053
3054 ocfs2_dlm_shutdown_debug(osb);
3055
3056 mlog_exit_void();
3057}
3058
7431cd7e 3059static void ocfs2_unlock_ast(void *opaque, int error)
ccd979bd
MF		 3060{
3061 struct ocfs2_lock_res *lockres = opaque;
3062 unsigned long flags;
3063
3064 mlog_entry_void();
3065
3066 mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
3067 lockres->l_unlock_action);
3068
3069 spin_lock_irqsave(&lockres->l_lock, flags);
de551246 3070 if (error) {
7431cd7e
JB		 3071 mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
3072 "unlock_action %d\n", error, lockres->l_name,
ccd979bd
MF		 3073 lockres->l_unlock_action);
3074 spin_unlock_irqrestore(&lockres->l_lock, flags);
d92bc512 3075 mlog_exit_void();
ccd979bd
MF		 3076 return;
3077 }
3078
3079 switch(lockres->l_unlock_action) {
3080 case OCFS2_UNLOCK_CANCEL_CONVERT:
3081 mlog(0, "Cancel convert success for %s\n", lockres->l_name);
3082 lockres->l_action = OCFS2_AST_INVALID;
a4b91965
SM		 3083 /* Downconvert thread may have requeued this lock, we
3084 * need to wake it. */
3085 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3086 ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
ccd979bd
MF		 3087 break;
3088 case OCFS2_UNLOCK_DROP_LOCK:
bd3e7610 3089 lockres->l_level = DLM_LOCK_IV;
ccd979bd
MF		 3090 break;
3091 default:
3092 BUG();
3093 }
3094
3095 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
ccd979bd 3096 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
ccd979bd 3097 wake_up(&lockres->l_event);
07f9eebc 3098 spin_unlock_irqrestore(&lockres->l_lock, flags);
ccd979bd
MF		 3099
3100 mlog_exit_void();
3101}
3102
ccd979bd 3103static int ocfs2_drop_lock(struct ocfs2_super *osb,
0d5dc6c2 3104 struct ocfs2_lock_res *lockres)
ccd979bd 3105{
7431cd7e 3106 int ret;
ccd979bd 3107 unsigned long flags;
bd3e7610 3108 u32 lkm_flags = 0;
ccd979bd
MF		 3109
3110 /* We didn't get anywhere near actually using this lockres. */
3111 if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3112 goto out;
3113
b80fc012 3114 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
bd3e7610 3115 lkm_flags |= DLM_LKF_VALBLK;
b80fc012 3116
ccd979bd
MF		 3117 spin_lock_irqsave(&lockres->l_lock, flags);
3118
3119 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3120 "lockres %s, flags 0x%lx\n",
3121 lockres->l_name, lockres->l_flags);
3122
3123 while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3124 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3125 "%u, unlock_action = %u\n",
3126 lockres->l_name, lockres->l_flags, lockres->l_action,
3127 lockres->l_unlock_action);
3128
3129 spin_unlock_irqrestore(&lockres->l_lock, flags);
3130
3131 /* XXX: Today we just wait on any busy
3132 * locks... Perhaps we need to cancel converts in the
3133 * future? */
3134 ocfs2_wait_on_busy_lock(lockres);
3135
3136 spin_lock_irqsave(&lockres->l_lock, flags);
3137 }
3138
0d5dc6c2
MF		 3139 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3140 if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
bd3e7610 3141 lockres->l_level == DLM_LOCK_EX &&
0d5dc6c2
MF		 3142 !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3143 lockres->l_ops->set_lvb(lockres);
3144 }
ccd979bd
MF		 3145
3146 if (lockres->l_flags & OCFS2_LOCK_BUSY)
3147 mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3148 lockres->l_name);
3149 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3150 mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3151
3152 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3153 spin_unlock_irqrestore(&lockres->l_lock, flags);
3154 goto out;
3155 }
3156
3157 lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3158
3159 /* make sure we never get here while waiting for an ast to
3160 * fire. */
3161 BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3162
3163 /* is this necessary? */
3164 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3165 lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3166 spin_unlock_irqrestore(&lockres->l_lock, flags);
3167
3168 mlog(0, "lock %s\n", lockres->l_name);
3169
4670c46d 3170 ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags,
7431cd7e
JB		 3171 lockres);
3172 if (ret) {
3173 ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
ccd979bd 3174 mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
cf0acdcd 3175 ocfs2_dlm_dump_lksb(&lockres->l_lksb);
ccd979bd
MF		 3176 BUG();
3177 }
73ac36ea 3178 mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
ccd979bd
MF		 3179 lockres->l_name);
3180
3181 ocfs2_wait_on_busy_lock(lockres);
3182out:
3183 mlog_exit(0);
3184 return 0;
3185}
3186
3187/* Mark the lockres as being dropped. It will no longer be
3188 * queued if blocking, but we still may have to wait on it
34d024f8 3189 * being dequeued from the downconvert thread before we can consider
2bd63216 3190 * it safe to drop.
ccd979bd
MF		 3191 *
3192 * You can *not* attempt to call cluster_lock on this lockres anymore. */
3193void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
3194{
3195 int status;
3196 struct ocfs2_mask_waiter mw;
3197 unsigned long flags;
3198
3199 ocfs2_init_mask_waiter(&mw);
3200
3201 spin_lock_irqsave(&lockres->l_lock, flags);
3202 lockres->l_flags |= OCFS2_LOCK_FREEING;
3203 while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3204 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3205 spin_unlock_irqrestore(&lockres->l_lock, flags);
3206
3207 mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3208
3209 status = ocfs2_wait_for_mask(&mw);
3210 if (status)
3211 mlog_errno(status);
3212
3213 spin_lock_irqsave(&lockres->l_lock, flags);
3214 }
3215 spin_unlock_irqrestore(&lockres->l_lock, flags);
3216}
3217
d680efe9
MF		 3218void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3219 struct ocfs2_lock_res *lockres)
ccd979bd 3220{
d680efe9 3221 int ret;
ccd979bd 3222
d680efe9 3223 ocfs2_mark_lockres_freeing(lockres);
0d5dc6c2 3224 ret = ocfs2_drop_lock(osb, lockres);
d680efe9
MF		 3225 if (ret)
3226 mlog_errno(ret);
3227}
ccd979bd 3228
d680efe9
MF		 3229static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3230{
3231 ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3232 ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
6ca497a8 3233 ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
83273932 3234 ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
ccd979bd
MF		 3235}
3236
ccd979bd
MF		 3237int ocfs2_drop_inode_locks(struct inode *inode)
3238{
3239 int status, err;
ccd979bd
MF		 3240
3241 mlog_entry_void();
3242
3243 /* No need to call ocfs2_mark_lockres_freeing here -
3244 * ocfs2_clear_inode has done it for us. */
3245
3246 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
50008630 3247 &OCFS2_I(inode)->ip_open_lockres);
ccd979bd
MF		 3248 if (err < 0)
3249 mlog_errno(err);
3250
3251 status = err;
3252
3253 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
e63aecb6 3254 &OCFS2_I(inode)->ip_inode_lockres);
ccd979bd
MF		 3255 if (err < 0)
3256 mlog_errno(err);
3257 if (err < 0 && !status)
3258 status = err;
3259
3260 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
0d5dc6c2 3261 &OCFS2_I(inode)->ip_rw_lockres);
ccd979bd
MF		 3262 if (err < 0)
3263 mlog_errno(err);
3264 if (err < 0 && !status)
3265 status = err;
3266
3267 mlog_exit(status);
3268 return status;
3269}
3270
de551246
JB		 3271static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3272 int new_level)
ccd979bd
MF		 3273{
3274 assert_spin_locked(&lockres->l_lock);
3275
bd3e7610 3276 BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
ccd979bd
MF		 3277
3278 if (lockres->l_level <= new_level) {
bd3e7610 3279 mlog(ML_ERROR, "lockres->l_level (%d) <= new_level (%d)\n",
ccd979bd
MF		 3280 lockres->l_level, new_level);
3281 BUG();
3282 }
3283
3284 mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
3285 lockres->l_name, new_level, lockres->l_blocking);
3286
3287 lockres->l_action = OCFS2_AST_DOWNCONVERT;
3288 lockres->l_requested = new_level;
3289 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
de551246 3290 return lockres_set_pending(lockres);
ccd979bd
MF		 3291}
3292
3293static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3294 struct ocfs2_lock_res *lockres,
3295 int new_level,
de551246
JB		 3296 int lvb,
3297 unsigned int generation)
ccd979bd 3298{
bd3e7610
JB		 3299 int ret;
3300 u32 dlm_flags = DLM_LKF_CONVERT;
ccd979bd
MF		 3301
3302 mlog_entry_void();
3303
3304 if (lvb)
bd3e7610 3305 dlm_flags |= DLM_LKF_VALBLK;
ccd979bd 3306
4670c46d 3307 ret = ocfs2_dlm_lock(osb->cconn,
7431cd7e
JB		 3308 new_level,
3309 &lockres->l_lksb,
3310 dlm_flags,
3311 lockres->l_name,
3312 OCFS2_LOCK_ID_MAX_LEN - 1,
3313 lockres);
de551246 3314 lockres_clear_pending(lockres, generation, osb);
7431cd7e
JB		 3315 if (ret) {
3316 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
ccd979bd
MF		 3317 ocfs2_recover_from_dlm_error(lockres, 1);
3318 goto bail;
3319 }
3320
3321 ret = 0;
3322bail:
3323 mlog_exit(ret);
3324 return ret;
3325}
3326
24ef1815 3327/* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
ccd979bd
MF		 3328static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3329 struct ocfs2_lock_res *lockres)
3330{
3331 assert_spin_locked(&lockres->l_lock);
3332
3333 mlog_entry_void();
3334 mlog(0, "lock %s\n", lockres->l_name);
3335
3336 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3337 /* If we're already trying to cancel a lock conversion
3338 * then just drop the spinlock and allow the caller to
3339 * requeue this lock. */
3340
3341 mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
3342 return 0;
3343 }
3344
3345 /* were we in a convert when we got the bast fire? */
3346 BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3347 lockres->l_action != OCFS2_AST_DOWNCONVERT);
3348 /* set things up for the unlockast to know to just
3349 * clear out the ast_action and unset busy, etc. */
3350 lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3351
3352 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3353 "lock %s, invalid flags: 0x%lx\n",
3354 lockres->l_name, lockres->l_flags);
3355
3356 return 1;
3357}
3358
3359static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3360 struct ocfs2_lock_res *lockres)
3361{
3362 int ret;
ccd979bd
MF		 3363
3364 mlog_entry_void();
3365 mlog(0, "lock %s\n", lockres->l_name);
3366
4670c46d 3367 ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
7431cd7e
JB		 3368 DLM_LKF_CANCEL, lockres);
3369 if (ret) {
3370 ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
ccd979bd
MF		 3371 ocfs2_recover_from_dlm_error(lockres, 0);
3372 }
3373
24ef1815 3374 mlog(0, "lock %s return from ocfs2_dlm_unlock\n", lockres->l_name);
ccd979bd 3375
ccd979bd
MF		 3376 mlog_exit(ret);
3377 return ret;
3378}
3379
b5e500e2
MF		 3380static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3381 struct ocfs2_lock_res *lockres,
3382 struct ocfs2_unblock_ctl *ctl)
ccd979bd
MF		 3383{
3384 unsigned long flags;
3385 int blocking;
3386 int new_level;
079b8057 3387 int level;
ccd979bd 3388 int ret = 0;
5ef0d4ea 3389 int set_lvb = 0;
de551246 3390 unsigned int gen;
ccd979bd
MF		 3391
3392 mlog_entry_void();
3393
3394 spin_lock_irqsave(&lockres->l_lock, flags);
3395
ccd979bd 3396recheck:
db0f6ce6
SM		 3397 /*
3398 * Is it still blocking? If not, we have no more work to do.
3399 */
3400 if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3401 BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3402 spin_unlock_irqrestore(&lockres->l_lock, flags);
3403 ret = 0;
3404 goto leave;
3405 }
3406
ccd979bd 3407 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
de551246
JB		 3408 /* XXX
3409 * This is a *big* race. The OCFS2_LOCK_PENDING flag
3410 * exists entirely for one reason - another thread has set
3411 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3412 *
3413 * If we do ocfs2_cancel_convert() before the other thread
3414 * calls dlm_lock(), our cancel will do nothing. We will
3415 * get no ast, and we will have no way of knowing the
3416 * cancel failed. Meanwhile, the other thread will call
3417 * into dlm_lock() and wait...forever.
3418 *
3419 * Why forever? Because another node has asked for the
3420 * lock first; that's why we're here in unblock_lock().
3421 *
3422 * The solution is OCFS2_LOCK_PENDING. When PENDING is
3423 * set, we just requeue the unblock. Only when the other
3424 * thread has called dlm_lock() and cleared PENDING will
3425 * we then cancel their request.
3426 *
3427 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3428 * at the same time they set OCFS2_DLM_BUSY. They must
3429 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3430 */
3431 if (lockres->l_flags & OCFS2_LOCK_PENDING)
3432 goto leave_requeue;
3433
d680efe9 3434 ctl->requeue = 1;
ccd979bd
MF		 3435 ret = ocfs2_prepare_cancel_convert(osb, lockres);
3436 spin_unlock_irqrestore(&lockres->l_lock, flags);
3437 if (ret) {
3438 ret = ocfs2_cancel_convert(osb, lockres);
3439 if (ret < 0)
3440 mlog_errno(ret);
3441 }
3442 goto leave;
3443 }
3444
a1912826
SM		 3445 /*
3446 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3447 * set when the ast is received for an upconvert just before the
3448 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3449 * on the heels of the ast, we want to delay the downconvert just
3450 * enough to allow the up requestor to do its task. Because this
3451 * lock is in the blocked queue, the lock will be downconverted
3452 * as soon as the requestor is done with the lock.
3453 */
3454 if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3455 goto leave_requeue;
3456
0d74125a
SM		 3457 /*
3458 * How can we block and yet be at NL? We were trying to upconvert
3459 * from NL and got canceled. The code comes back here, and now
3460 * we notice and clear BLOCKING.
3461 */
3462 if (lockres->l_level == DLM_LOCK_NL) {
3463 BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
3464 lockres->l_blocking = DLM_LOCK_NL;
3465 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3466 spin_unlock_irqrestore(&lockres->l_lock, flags);
3467 goto leave;
3468 }
3469
ccd979bd
MF		 3470 /* if we're blocking an exclusive and we have *any* holders,
3471 * then requeue. */
bd3e7610 3472 if ((lockres->l_blocking == DLM_LOCK_EX)
f7fbfdd1
MF		 3473 && (lockres->l_ex_holders || lockres->l_ro_holders))
3474 goto leave_requeue;
ccd979bd
MF		 3475
3476 /* If it's a PR we're blocking, then only
3477 * requeue if we've got any EX holders */
bd3e7610 3478 if (lockres->l_blocking == DLM_LOCK_PR &&
f7fbfdd1
MF		 3479 lockres->l_ex_holders)
3480 goto leave_requeue;
3481
3482 /*
3483 * Can we get a lock in this state if the holder counts are
3484 * zero? The meta data unblock code used to check this.
3485 */
3486 if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3487 && (lockres->l_flags & OCFS2_LOCK_REFRESHING))
3488 goto leave_requeue;
ccd979bd 3489
16d5b956
MF		 3490 new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3491
3492 if (lockres->l_ops->check_downconvert
3493 && !lockres->l_ops->check_downconvert(lockres, new_level))
3494 goto leave_requeue;
3495
ccd979bd
MF		 3496 /* If we get here, then we know that there are no more
3497 * incompatible holders (and anyone asking for an incompatible
3498 * lock is blocked). We can now downconvert the lock */
cc567d89 3499 if (!lockres->l_ops->downconvert_worker)
ccd979bd
MF		 3500 goto downconvert;
3501
3502 /* Some lockres types want to do a bit of work before
3503 * downconverting a lock. Allow that here. The worker function
3504 * may sleep, so we save off a copy of what we're blocking as
3505 * it may change while we're not holding the spin lock. */
3506 blocking = lockres->l_blocking;
079b8057 3507 level = lockres->l_level;
ccd979bd
MF		 3508 spin_unlock_irqrestore(&lockres->l_lock, flags);
3509
cc567d89 3510 ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
d680efe9
MF		 3511
3512 if (ctl->unblock_action == UNBLOCK_STOP_POST)
3513 goto leave;
ccd979bd
MF		 3514
3515 spin_lock_irqsave(&lockres->l_lock, flags);
079b8057 3516 if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
ccd979bd
MF		 3517 /* If this changed underneath us, then we can't drop
3518 * it just yet. */
3519 goto recheck;
3520 }
3521
3522downconvert:
d680efe9 3523 ctl->requeue = 0;
ccd979bd 3524
5ef0d4ea 3525 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
bd3e7610 3526 if (lockres->l_level == DLM_LOCK_EX)
5ef0d4ea
MF		 3527 set_lvb = 1;
3528
3529 /*
3530 * We only set the lvb if the lock has been fully
3531 * refreshed - otherwise we risk setting stale
3532 * data. Otherwise, there's no need to actually clear
3533 * out the lvb here as it's value is still valid.
3534 */
3535 if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3536 lockres->l_ops->set_lvb(lockres);
3537 }
3538
de551246 3539 gen = ocfs2_prepare_downconvert(lockres, new_level);
ccd979bd 3540 spin_unlock_irqrestore(&lockres->l_lock, flags);
de551246
JB		 3541 ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3542 gen);
3543
ccd979bd
MF		 3544leave:
3545 mlog_exit(ret);
3546 return ret;
f7fbfdd1
MF		 3547
3548leave_requeue:
3549 spin_unlock_irqrestore(&lockres->l_lock, flags);
3550 ctl->requeue = 1;
3551
3552 mlog_exit(0);
3553 return 0;
ccd979bd
MF		 3554}
3555
d680efe9
MF		 3556static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3557 int blocking)
ccd979bd
MF		 3558{
3559 struct inode *inode;
3560 struct address_space *mapping;
3561
ccd979bd
MF		 3562 inode = ocfs2_lock_res_inode(lockres);
3563 mapping = inode->i_mapping;
3564
1044e401 3565 if (!S_ISREG(inode->i_mode))
f1f54068
MF		 3566 goto out;
3567
7f4a2a97
MF		 3568 /*
3569 * We need this before the filemap_fdatawrite() so that it can
3570 * transfer the dirty bit from the PTE to the
3571 * page. Unfortunately this means that even for EX->PR
3572 * downconverts, we'll lose our mappings and have to build
3573 * them up again.
3574 */
3575 unmap_mapping_range(mapping, 0, 0, 0);
3576
ccd979bd 3577 if (filemap_fdatawrite(mapping)) {
b0697053
MF		 3578 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3579 (unsigned long long)OCFS2_I(inode)->ip_blkno);
ccd979bd
MF		 3580 }
3581 sync_mapping_buffers(mapping);
bd3e7610 3582 if (blocking == DLM_LOCK_EX) {
ccd979bd 3583 truncate_inode_pages(mapping, 0);
ccd979bd
MF		 3584 } else {
3585 /* We only need to wait on the I/O if we're not also
3586 * truncating pages because truncate_inode_pages waits
3587 * for us above. We don't truncate pages if we're
3588 * blocking anything < EXMODE because we want to keep
3589 * them around in that case. */
3590 filemap_fdatawait(mapping);
3591 }
3592
f1f54068 3593out:
d680efe9 3594 return UNBLOCK_CONTINUE;
ccd979bd
MF		 3595}
3596
a4338481
TM		 3597static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3598 struct ocfs2_lock_res *lockres,
3599 int new_level)
810d5aeb 3600{
a4338481 3601 int checkpointed = ocfs2_ci_fully_checkpointed(ci);
810d5aeb 3602
bd3e7610
JB		 3603 BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3604 BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
810d5aeb
MF		 3605
3606 if (checkpointed)
3607 return 1;
3608
a4338481 3609 ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
810d5aeb
MF		 3610 return 0;
3611}
3612
a4338481
TM		 3613static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3614 int new_level)
3615{
3616 struct inode *inode = ocfs2_lock_res_inode(lockres);
3617
3618 return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3619}
3620
810d5aeb
MF		 3621static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3622{
3623 struct inode *inode = ocfs2_lock_res_inode(lockres);
3624
3625 __ocfs2_stuff_meta_lvb(inode);
3626}
3627
d680efe9
MF		 3628/*
3629 * Does the final reference drop on our dentry lock. Right now this
34d024f8 3630 * happens in the downconvert thread, but we could choose to simplify the
d680efe9
MF		 3631 * dlmglue API and push these off to the ocfs2_wq in the future.
3632 */
3633static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3634 struct ocfs2_lock_res *lockres)
3635{
3636 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3637 ocfs2_dentry_lock_put(osb, dl);
3638}
3639
3640/*
3641 * d_delete() matching dentries before the lock downconvert.
3642 *
3643 * At this point, any process waiting to destroy the
3644 * dentry_lock due to last ref count is stopped by the
3645 * OCFS2_LOCK_QUEUED flag.
3646 *
3647 * We have two potential problems
3648 *
3649 * 1) If we do the last reference drop on our dentry_lock (via dput)
3650 * we'll wind up in ocfs2_release_dentry_lock(), waiting on
3651 * the downconvert to finish. Instead we take an elevated
3652 * reference and push the drop until after we've completed our
3653 * unblock processing.
3654 *
3655 * 2) There might be another process with a final reference,
3656 * waiting on us to finish processing. If this is the case, we
3657 * detect it and exit out - there's no more dentries anyway.
3658 */
3659static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3660 int blocking)
3661{
3662 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3663 struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3664 struct dentry *dentry;
3665 unsigned long flags;
3666 int extra_ref = 0;
3667
3668 /*
3669 * This node is blocking another node from getting a read
3670 * lock. This happens when we've renamed within a
3671 * directory. We've forced the other nodes to d_delete(), but
3672 * we never actually dropped our lock because it's still
3673 * valid. The downconvert code will retain a PR for this node,
3674 * so there's no further work to do.
3675 */
bd3e7610 3676 if (blocking == DLM_LOCK_PR)
d680efe9
MF		 3677 return UNBLOCK_CONTINUE;
3678
3679 /*
3680 * Mark this inode as potentially orphaned. The code in
3681 * ocfs2_delete_inode() will figure out whether it actually
3682 * needs to be freed or not.
3683 */
3684 spin_lock(&oi->ip_lock);
3685 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
3686 spin_unlock(&oi->ip_lock);
3687
3688 /*
3689 * Yuck. We need to make sure however that the check of
3690 * OCFS2_LOCK_FREEING and the extra reference are atomic with
3691 * respect to a reference decrement or the setting of that
3692 * flag.
3693 */
3694 spin_lock_irqsave(&lockres->l_lock, flags);
3695 spin_lock(&dentry_attach_lock);
3696 if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
3697 && dl->dl_count) {
3698 dl->dl_count++;
3699 extra_ref = 1;
3700 }
3701 spin_unlock(&dentry_attach_lock);
3702 spin_unlock_irqrestore(&lockres->l_lock, flags);
3703
3704 mlog(0, "extra_ref = %d\n", extra_ref);
3705
3706 /*
3707 * We have a process waiting on us in ocfs2_dentry_iput(),
3708 * which means we can't have any more outstanding
3709 * aliases. There's no need to do any more work.
3710 */
3711 if (!extra_ref)
3712 return UNBLOCK_CONTINUE;
3713
3714 spin_lock(&dentry_attach_lock);
3715 while (1) {
3716 dentry = ocfs2_find_local_alias(dl->dl_inode,
3717 dl->dl_parent_blkno, 1);
3718 if (!dentry)
3719 break;
3720 spin_unlock(&dentry_attach_lock);
3721
3722 mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
3723 dentry->d_name.name);
3724
3725 /*
3726 * The following dcache calls may do an
3727 * iput(). Normally we don't want that from the
3728 * downconverting thread, but in this case it's ok
3729 * because the requesting node already has an
3730 * exclusive lock on the inode, so it can't be queued
3731 * for a downconvert.
3732 */
3733 d_delete(dentry);
3734 dput(dentry);
3735
3736 spin_lock(&dentry_attach_lock);
3737 }
3738 spin_unlock(&dentry_attach_lock);
3739
3740 /*
3741 * If we are the last holder of this dentry lock, there is no
3742 * reason to downconvert so skip straight to the unlock.
3743 */
3744 if (dl->dl_count == 1)
3745 return UNBLOCK_STOP_POST;
3746
3747 return UNBLOCK_CONTINUE_POST;
3748}
3749
8dec98ed
TM		 3750static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
3751 int new_level)
3752{
3753 struct ocfs2_refcount_tree *tree =
3754 ocfs2_lock_res_refcount_tree(lockres);
3755
3756 return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
3757}
3758
3759static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
3760 int blocking)
3761{
3762 struct ocfs2_refcount_tree *tree =
3763 ocfs2_lock_res_refcount_tree(lockres);
3764
3765 ocfs2_metadata_cache_purge(&tree->rf_ci);
3766
3767 return UNBLOCK_CONTINUE;
3768}
3769
9e33d69f
JK		 3770static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
3771{
3772 struct ocfs2_qinfo_lvb *lvb;
3773 struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
3774 struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3775 oinfo->dqi_gi.dqi_type);
3776
3777 mlog_entry_void();
3778
a641dc2a 3779 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
9e33d69f
JK		 3780 lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
3781 lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
3782 lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
3783 lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
3784 lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
3785 lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
3786 lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
3787
3788 mlog_exit_void();
3789}
3790
3791void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3792{
3793 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3794 struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3795 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3796
3797 mlog_entry_void();
3798 if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
3799 ocfs2_cluster_unlock(osb, lockres, level);
3800 mlog_exit_void();
3801}
3802
3803static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
3804{
3805 struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3806 oinfo->dqi_gi.dqi_type);
3807 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3808 struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
85eb8b73 3809 struct buffer_head *bh = NULL;
9e33d69f
JK		 3810 struct ocfs2_global_disk_dqinfo *gdinfo;
3811 int status = 0;
3812
1c520dfb
JB		 3813 if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
3814 lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
9e33d69f
JK		 3815 info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
3816 info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
3817 oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
3818 oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
3819 oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
3820 oinfo->dqi_gi.dqi_free_entry =
3821 be32_to_cpu(lvb->lvb_free_entry);
3822 } else {
85eb8b73
JB		 3823 status = ocfs2_read_quota_block(oinfo->dqi_gqinode, 0, &bh);
3824 if (status) {
9e33d69f
JK		 3825 mlog_errno(status);
3826 goto bail;
3827 }
3828 gdinfo = (struct ocfs2_global_disk_dqinfo *)
3829 (bh->b_data + OCFS2_GLOBAL_INFO_OFF);
3830 info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
3831 info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
3832 oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
3833 oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
3834 oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
3835 oinfo->dqi_gi.dqi_free_entry =
3836 le32_to_cpu(gdinfo->dqi_free_entry);
3837 brelse(bh);
3838 ocfs2_track_lock_refresh(lockres);
3839 }
3840
3841bail:
3842 return status;
3843}
3844
3845/* Lock quota info, this function expects at least shared lock on the quota file
3846 * so that we can safely refresh quota info from disk. */
3847int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3848{
3849 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3850 struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3851 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3852 int status = 0;
3853
3854 mlog_entry_void();
3855
3856 /* On RO devices, locking really isn't needed... */
3857 if (ocfs2_is_hard_readonly(osb)) {
3858 if (ex)
3859 status = -EROFS;
3860 goto bail;
3861 }
3862 if (ocfs2_mount_local(osb))
3863 goto bail;
3864
3865 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3866 if (status < 0) {
3867 mlog_errno(status);
3868 goto bail;
3869 }
3870 if (!ocfs2_should_refresh_lock_res(lockres))
3871 goto bail;
3872 /* OK, we have the lock but we need to refresh the quota info */
3873 status = ocfs2_refresh_qinfo(oinfo);
3874 if (status)
3875 ocfs2_qinfo_unlock(oinfo, ex);
3876 ocfs2_complete_lock_res_refresh(lockres, status);
3877bail:
3878 mlog_exit(status);
3879 return status;
3880}
3881
8dec98ed
TM		 3882int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
3883{
3884 int status;
3885 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3886 struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
3887 struct ocfs2_super *osb = lockres->l_priv;
3888
3889
3890 if (ocfs2_is_hard_readonly(osb))
3891 return -EROFS;
3892
3893 if (ocfs2_mount_local(osb))
3894 return 0;
3895
3896 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3897 if (status < 0)
3898 mlog_errno(status);
3899
3900 return status;
3901}
3902
3903void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
3904{
3905 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3906 struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
3907 struct ocfs2_super *osb = lockres->l_priv;
3908
3909 if (!ocfs2_mount_local(osb))
3910 ocfs2_cluster_unlock(osb, lockres, level);
3911}
3912
4670c46d
JB		 3913/*
3914 * This is the filesystem locking protocol. It provides the lock handling
3915 * hooks for the underlying DLM. It has a maximum version number.
3916 * The version number allows interoperability with systems running at
3917 * the same major number and an equal or smaller minor number.
3918 *
3919 * Whenever the filesystem does new things with locks (adds or removes a
3920 * lock, orders them differently, does different things underneath a lock),
3921 * the version must be changed. The protocol is negotiated when joining
3922 * the dlm domain. A node may join the domain if its major version is
3923 * identical to all other nodes and its minor version is greater than
3924 * or equal to all other nodes. When its minor version is greater than
3925 * the other nodes, it will run at the minor version specified by the
3926 * other nodes.
3927 *
3928 * If a locking change is made that will not be compatible with older
3929 * versions, the major number must be increased and the minor version set
3930 * to zero. If a change merely adds a behavior that can be disabled when
3931 * speaking to older versions, the minor version must be increased. If a
3932 * change adds a fully backwards compatible change (eg, LVB changes that
3933 * are just ignored by older versions), the version does not need to be
3934 * updated.
3935 */
24ef1815 3936static struct ocfs2_locking_protocol lproto = {
4670c46d
JB		 3937 .lp_max_version = {
3938 .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
3939 .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
3940 },
24ef1815
JB		 3941 .lp_lock_ast = ocfs2_locking_ast,
3942 .lp_blocking_ast = ocfs2_blocking_ast,
3943 .lp_unlock_ast = ocfs2_unlock_ast,
3944};
3945
63e0c48a 3946void ocfs2_set_locking_protocol(void)
24ef1815 3947{
63e0c48a 3948 ocfs2_stack_glue_set_locking_protocol(&lproto);
24ef1815
JB		 3949}
3950
24ef1815 3951
00600056
AB		 3952static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3953 struct ocfs2_lock_res *lockres)
ccd979bd
MF		 3954{
3955 int status;
d680efe9 3956 struct ocfs2_unblock_ctl ctl = {0, 0,};
ccd979bd
MF		 3957 unsigned long flags;
3958
3959 /* Our reference to the lockres in this function can be
3960 * considered valid until we remove the OCFS2_LOCK_QUEUED
3961 * flag. */
3962
3963 mlog_entry_void();
3964
3965 BUG_ON(!lockres);
3966 BUG_ON(!lockres->l_ops);
ccd979bd
MF		 3967
3968 mlog(0, "lockres %s blocked.\n", lockres->l_name);
3969
3970 /* Detect whether a lock has been marked as going away while
34d024f8 3971 * the downconvert thread was processing other things. A lock can
ccd979bd
MF		 3972 * still be marked with OCFS2_LOCK_FREEING after this check,
3973 * but short circuiting here will still save us some
3974 * performance. */
3975 spin_lock_irqsave(&lockres->l_lock, flags);
3976 if (lockres->l_flags & OCFS2_LOCK_FREEING)
3977 goto unqueue;
3978 spin_unlock_irqrestore(&lockres->l_lock, flags);
3979
b5e500e2 3980 status = ocfs2_unblock_lock(osb, lockres, &ctl);
ccd979bd
MF		 3981 if (status < 0)
3982 mlog_errno(status);
3983
3984 spin_lock_irqsave(&lockres->l_lock, flags);
3985unqueue:
d680efe9 3986 if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
ccd979bd
MF		 3987 lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3988 } else
3989 ocfs2_schedule_blocked_lock(osb, lockres);
3990
3991 mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
d680efe9 3992 ctl.requeue ? "yes" : "no");
ccd979bd
MF		 3993 spin_unlock_irqrestore(&lockres->l_lock, flags);
3994
d680efe9
MF		 3995 if (ctl.unblock_action != UNBLOCK_CONTINUE
3996 && lockres->l_ops->post_unlock)
3997 lockres->l_ops->post_unlock(osb, lockres);
3998
ccd979bd
MF		 3999 mlog_exit_void();
4000}
4001
4002static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
4003 struct ocfs2_lock_res *lockres)
4004{
4005 mlog_entry_void();
4006
4007 assert_spin_locked(&lockres->l_lock);
4008
4009 if (lockres->l_flags & OCFS2_LOCK_FREEING) {
4010 /* Do not schedule a lock for downconvert when it's on
4011 * the way to destruction - any nodes wanting access
4012 * to the resource will get it soon. */
4013 mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
4014 lockres->l_name, lockres->l_flags);
4015 return;
4016 }
4017
4018 lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
4019
34d024f8 4020 spin_lock(&osb->dc_task_lock);
ccd979bd
MF		 4021 if (list_empty(&lockres->l_blocked_list)) {
4022 list_add_tail(&lockres->l_blocked_list,
4023 &osb->blocked_lock_list);
4024 osb->blocked_lock_count++;
4025 }
34d024f8 4026 spin_unlock(&osb->dc_task_lock);
ccd979bd
MF		 4027
4028 mlog_exit_void();
4029}
34d024f8
MF		 4030
4031static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
4032{
4033 unsigned long processed;
4034 struct ocfs2_lock_res *lockres;
4035
4036 mlog_entry_void();
4037
4038 spin_lock(&osb->dc_task_lock);
4039 /* grab this early so we know to try again if a state change and
4040 * wake happens part-way through our work */
4041 osb->dc_work_sequence = osb->dc_wake_sequence;
4042
4043 processed = osb->blocked_lock_count;
4044 while (processed) {
4045 BUG_ON(list_empty(&osb->blocked_lock_list));
4046
4047 lockres = list_entry(osb->blocked_lock_list.next,
4048 struct ocfs2_lock_res, l_blocked_list);
4049 list_del_init(&lockres->l_blocked_list);
4050 osb->blocked_lock_count--;
4051 spin_unlock(&osb->dc_task_lock);
4052
4053 BUG_ON(!processed);
4054 processed--;
4055
4056 ocfs2_process_blocked_lock(osb, lockres);
4057
4058 spin_lock(&osb->dc_task_lock);
4059 }
4060 spin_unlock(&osb->dc_task_lock);
4061
4062 mlog_exit_void();
4063}
4064
4065static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
4066{
4067 int empty = 0;
4068
4069 spin_lock(&osb->dc_task_lock);
4070 if (list_empty(&osb->blocked_lock_list))
4071 empty = 1;
4072
4073 spin_unlock(&osb->dc_task_lock);
4074 return empty;
4075}
4076
4077static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4078{
4079 int should_wake = 0;
4080
4081 spin_lock(&osb->dc_task_lock);
4082 if (osb->dc_work_sequence != osb->dc_wake_sequence)
4083 should_wake = 1;
4084 spin_unlock(&osb->dc_task_lock);
4085
4086 return should_wake;
4087}
4088
200bfae3 4089static int ocfs2_downconvert_thread(void *arg)
34d024f8
MF		 4090{
4091 int status = 0;
4092 struct ocfs2_super *osb = arg;
4093
4094 /* only quit once we've been asked to stop and there is no more
4095 * work available */
4096 while (!(kthread_should_stop() &&
4097 ocfs2_downconvert_thread_lists_empty(osb))) {
4098
4099 wait_event_interruptible(osb->dc_event,
4100 ocfs2_downconvert_thread_should_wake(osb) ||
4101 kthread_should_stop());
4102
4103 mlog(0, "downconvert_thread: awoken\n");
4104
4105 ocfs2_downconvert_thread_do_work(osb);
4106 }
4107
4108 osb->dc_task = NULL;
4109 return status;
4110}
4111
4112void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4113{
4114 spin_lock(&osb->dc_task_lock);
4115 /* make sure the voting thread gets a swipe at whatever changes
4116 * the caller may have made to the voting state */
4117 osb->dc_wake_sequence++;
4118 spin_unlock(&osb->dc_task_lock);
4119 wake_up(&osb->dc_event);
4120}
kernel source for telekom puls (alcatel/mediatek)