blob: 81a4cd22f80be84a06eac2b0fbf4348385d76262 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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>
Mark Fashehccd979b2005-12-15 14:31:24 -080030#include <linux/kthread.h>
31#include <linux/pagemap.h>
32#include <linux/debugfs.h>
33#include <linux/seq_file.h>
Sunil Mushran8ddb7b02008-05-13 13:45:15 -070034#include <linux/time.h>
Jan Kara9e33d692008-08-25 19:56:50 +020035#include <linux/quotaops.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080036
Mark Fashehccd979b2005-12-15 14:31:24 -080037#define MLOG_MASK_PREFIX ML_DLM_GLUE
38#include <cluster/masklog.h>
39
40#include "ocfs2.h"
Joel Beckerd24fbcd2008-01-25 17:02:21 -080041#include "ocfs2_lockingver.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080042
43#include "alloc.h"
Mark Fashehd680efe2006-09-08 14:14:34 -070044#include "dcache.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080045#include "dlmglue.h"
46#include "extent_map.h"
Tiger Yang7f1a37e2006-11-15 15:48:42 +080047#include "file.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080048#include "heartbeat.h"
49#include "inode.h"
50#include "journal.h"
Joel Becker24ef1812008-01-29 17:37:32 -080051#include "stackglue.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080052#include "slot_map.h"
53#include "super.h"
54#include "uptodate.h"
Jan Kara9e33d692008-08-25 19:56:50 +020055#include "quota.h"
Tao Ma8dec98e2009-08-18 11:19:58 +080056#include "refcounttree.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080057
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;
Sunil Mushran8ddb7b02008-05-13 13:45:15 -070066#ifdef CONFIG_OCFS2_FS_STATS
Sunil Mushran5bc970e2010-12-28 23:26:03 -080067 ktime_t mw_lock_start;
Sunil Mushran8ddb7b02008-05-13 13:45:15 -070068#endif
Mark Fashehccd979b2005-12-15 14:31:24 -080069};
70
Mark Fasheh54a7e752006-09-12 21:49:13 -070071static 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);
Mark Fashehcf8e06f2007-12-20 16:43:10 -080073static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
Jan Kara9e33d692008-08-25 19:56:50 +020074static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -080075
Mark Fashehd680efe2006-09-08 14:14:34 -070076/*
Mark Fashehcc567d82006-09-13 21:52:21 -070077 * Return value from ->downconvert_worker functions.
Mark Fashehd680efe2006-09-08 14:14:34 -070078 *
Mark Fashehb5e500e2006-09-13 22:01:16 -070079 * These control the precise actions of ocfs2_unblock_lock()
Mark Fashehd680efe2006-09-08 14:14:34 -070080 * 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
Jan Karacb257972009-06-04 15:26:50 +020096/* Lockdep class keys */
97struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES];
98
Mark Fasheh810d5ae2006-09-13 21:39:52 -070099static 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
Mark Fashehcc567d82006-09-13 21:52:21 -0700103static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
104 int blocking);
105
Mark Fashehcc567d82006-09-13 21:52:21 -0700106static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
107 int blocking);
Mark Fashehd680efe2006-09-08 14:14:34 -0700108
109static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
110 struct ocfs2_lock_res *lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -0800111
Jan Kara9e33d692008-08-25 19:56:50 +0200112static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700113
Tao Ma8dec98e2009-08-18 11:19:58 +0800114static 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
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700119#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{
Mark Fasheha641dc22008-12-24 16:03:48 -0800127 struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700128
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
Mark Fashehf625c972006-09-12 21:24:53 -0700147/*
148 * OCFS2 Lock Resource Operations
149 *
150 * These fine tune the behavior of the generic dlmglue locking infrastructure.
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -0700151 *
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).
Mark Fashehf625c972006-09-12 21:24:53 -0700159 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800160struct ocfs2_lock_res_ops {
Mark Fasheh54a7e752006-09-12 21:49:13 -0700161 /*
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 *);
Mark Fashehb5e500e2006-09-13 22:01:16 -0700166
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -0700167 /*
Mark Fasheh34d024f2007-09-24 15:56:19 -0700168 * 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.
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -0700172 *
173 * The exact semantics of when this is called are controlled
174 * by ->downconvert_worker()
175 */
Mark Fashehd680efe2006-09-08 14:14:34 -0700176 void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
Mark Fashehf625c972006-09-12 21:24:53 -0700177
178 /*
Mark Fasheh16d5b9562006-09-13 21:10:12 -0700179 * 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
190 /*
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -0700191 * 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
201 /*
Mark Fashehcc567d82006-09-13 21:52:21 -0700202 * 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
212 /*
Mark Fashehf625c972006-09-12 21:24:53 -0700213 * LOCK_TYPE_* flags which describe the specific requirements
214 * of a lock type. Descriptions of each individual flag follow.
215 */
216 int flags;
Mark Fashehccd979b2005-12-15 14:31:24 -0800217};
218
Mark Fashehf625c972006-09-12 21:24:53 -0700219/*
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
Mark Fashehb80fc012006-09-12 22:08:14 -0700229/*
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -0700230 * Indicate that a lock type makes use of the lock value block. The
231 * ->set_lvb lock type callback must be defined.
Mark Fashehb80fc012006-09-12 22:08:14 -0700232 */
233#define LOCK_TYPE_USES_LVB 0x2
234
Mark Fashehccd979b2005-12-15 14:31:24 -0800235static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
Mark Fasheh54a7e752006-09-12 21:49:13 -0700236 .get_osb = ocfs2_get_inode_osb,
Mark Fashehf625c972006-09-12 21:24:53 -0700237 .flags = 0,
Mark Fashehccd979b2005-12-15 14:31:24 -0800238};
239
Mark Fashehe63aecb62007-10-18 15:30:42 -0700240static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
Mark Fasheh54a7e752006-09-12 21:49:13 -0700241 .get_osb = ocfs2_get_inode_osb,
Mark Fasheh810d5ae2006-09-13 21:39:52 -0700242 .check_downconvert = ocfs2_check_meta_downconvert,
243 .set_lvb = ocfs2_set_meta_lvb,
Mark Fashehf1f54062007-10-18 15:13:59 -0700244 .downconvert_worker = ocfs2_data_convert_worker,
Mark Fashehb80fc012006-09-12 22:08:14 -0700245 .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
Mark Fashehccd979b2005-12-15 14:31:24 -0800246};
247
Mark Fashehccd979b2005-12-15 14:31:24 -0800248static struct ocfs2_lock_res_ops ocfs2_super_lops = {
Mark Fashehf625c972006-09-12 21:24:53 -0700249 .flags = LOCK_TYPE_REQUIRES_REFRESH,
Mark Fashehccd979b2005-12-15 14:31:24 -0800250};
251
252static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
Mark Fashehf625c972006-09-12 21:24:53 -0700253 .flags = 0,
Mark Fashehccd979b2005-12-15 14:31:24 -0800254};
255
wengang wang6ca497a2009-03-06 21:29:10 +0800256static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
257 .flags = 0,
258};
259
Srinivas Eeda83273932009-06-03 17:02:55 -0700260static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = {
261 .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
262};
263
Mark Fashehd680efe2006-09-08 14:14:34 -0700264static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
Mark Fasheh54a7e752006-09-12 21:49:13 -0700265 .get_osb = ocfs2_get_dentry_osb,
Mark Fashehd680efe2006-09-08 14:14:34 -0700266 .post_unlock = ocfs2_dentry_post_unlock,
Mark Fashehcc567d82006-09-13 21:52:21 -0700267 .downconvert_worker = ocfs2_dentry_convert_worker,
Mark Fashehf625c972006-09-12 21:24:53 -0700268 .flags = 0,
Mark Fashehd680efe2006-09-08 14:14:34 -0700269};
270
Tiger Yang50008632007-03-20 16:01:38 -0700271static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
272 .get_osb = ocfs2_get_inode_osb,
273 .flags = 0,
274};
275
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800276static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
277 .get_osb = ocfs2_get_file_osb,
278 .flags = 0,
279};
280
Jan Kara9e33d692008-08-25 19:56:50 +0200281static 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
Tao Ma8dec98e2009-08-18 11:19:58 +0800287static 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
Mark Fashehccd979b2005-12-15 14:31:24 -0800293static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
294{
295 return lockres->l_type == OCFS2_LOCK_TYPE_META ||
Tiger Yang50008632007-03-20 16:01:38 -0700296 lockres->l_type == OCFS2_LOCK_TYPE_RW ||
297 lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
Mark Fashehccd979b2005-12-15 14:31:24 -0800298}
299
Joel Beckerc0e41332010-01-29 14:46:44 -0800300static inline struct ocfs2_lock_res *ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
Joel Beckera796d282010-01-28 19:22:39 -0800301{
302 return container_of(lksb, struct ocfs2_lock_res, l_lksb);
303}
304
Mark Fashehccd979b2005-12-15 14:31:24 -0800305static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
306{
307 BUG_ON(!ocfs2_is_inode_lock(lockres));
308
309 return (struct inode *) lockres->l_priv;
310}
311
Mark Fashehd680efe2006-09-08 14:14:34 -0700312static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
313{
314 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
315
316 return (struct ocfs2_dentry_lock *)lockres->l_priv;
317}
318
Jan Kara9e33d692008-08-25 19:56:50 +0200319static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
320{
321 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
322
323 return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
324}
325
Tao Ma8dec98e2009-08-18 11:19:58 +0800326static inline struct ocfs2_refcount_tree *
327ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
328{
329 return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
330}
331
Mark Fasheh54a7e752006-09-12 21:49:13 -0700332static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
333{
334 if (lockres->l_ops->get_osb)
335 return lockres->l_ops->get_osb(lockres);
336
337 return (struct ocfs2_super *)lockres->l_priv;
338}
339
Mark Fashehccd979b2005-12-15 14:31:24 -0800340static int ocfs2_lock_create(struct ocfs2_super *osb,
341 struct ocfs2_lock_res *lockres,
342 int level,
Joel Beckerbd3e7612008-02-01 12:14:57 -0800343 u32 dlm_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800344static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
345 int wanted);
Jan Karacb257972009-06-04 15:26:50 +0200346static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
347 struct ocfs2_lock_res *lockres,
348 int level, unsigned long caller_ip);
349static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb,
350 struct ocfs2_lock_res *lockres,
351 int level)
352{
353 __ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_);
354}
355
Mark Fashehccd979b2005-12-15 14:31:24 -0800356static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
357static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
358static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
359static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
360static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
361 struct ocfs2_lock_res *lockres);
362static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
363 int convert);
Sunil Mushranc74ff8b2009-02-03 12:37:14 -0800364#define ocfs2_log_dlm_error(_func, _err, _lockres) do { \
365 if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY) \
366 mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n", \
367 _err, _func, _lockres->l_name); \
368 else \
369 mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n", \
370 _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name, \
371 (unsigned int)ocfs2_get_dentry_lock_ino(_lockres)); \
Mark Fashehccd979b2005-12-15 14:31:24 -0800372} while (0)
Mark Fasheh34d024f2007-09-24 15:56:19 -0700373static int ocfs2_downconvert_thread(void *arg);
374static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
375 struct ocfs2_lock_res *lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700376static int ocfs2_inode_lock_update(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -0800377 struct buffer_head **bh);
378static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
379static inline int ocfs2_highest_compat_lock_level(int level);
Joel Beckerde551242008-02-01 14:45:08 -0800380static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
381 int new_level);
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800382static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
383 struct ocfs2_lock_res *lockres,
384 int new_level,
Joel Beckerde551242008-02-01 14:45:08 -0800385 int lvb,
386 unsigned int generation);
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800387static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
388 struct ocfs2_lock_res *lockres);
389static int ocfs2_cancel_convert(struct ocfs2_super *osb,
390 struct ocfs2_lock_res *lockres);
391
Mark Fashehccd979b2005-12-15 14:31:24 -0800392
Mark Fashehccd979b2005-12-15 14:31:24 -0800393static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
394 u64 blkno,
395 u32 generation,
396 char *name)
397{
398 int len;
399
Mark Fashehccd979b2005-12-15 14:31:24 -0800400 BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
401
Mark Fashehb0697052006-03-03 10:24:33 -0800402 len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
403 ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
404 (long long)blkno, generation);
Mark Fashehccd979b2005-12-15 14:31:24 -0800405
406 BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
407
408 mlog(0, "built lock resource with name: %s\n", name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800409}
410
Ingo Molnar34af9462006-06-27 02:53:55 -0700411static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800412
413static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
414 struct ocfs2_dlm_debug *dlm_debug)
415{
416 mlog(0, "Add tracking for lockres %s\n", res->l_name);
417
418 spin_lock(&ocfs2_dlm_tracking_lock);
419 list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
420 spin_unlock(&ocfs2_dlm_tracking_lock);
421}
422
423static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
424{
425 spin_lock(&ocfs2_dlm_tracking_lock);
426 if (!list_empty(&res->l_debug_list))
427 list_del_init(&res->l_debug_list);
428 spin_unlock(&ocfs2_dlm_tracking_lock);
429}
430
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700431#ifdef CONFIG_OCFS2_FS_STATS
432static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
433{
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700434 res->l_lock_refresh = 0;
Sunil Mushran5bc970e2010-12-28 23:26:03 -0800435 memset(&res->l_lock_prmode, 0, sizeof(struct ocfs2_lock_stats));
436 memset(&res->l_lock_exmode, 0, sizeof(struct ocfs2_lock_stats));
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700437}
438
439static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
440 struct ocfs2_mask_waiter *mw, int ret)
441{
Sunil Mushran5bc970e2010-12-28 23:26:03 -0800442 u32 usec;
443 ktime_t kt;
444 struct ocfs2_lock_stats *stats;
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700445
Sunil Mushran5bc970e2010-12-28 23:26:03 -0800446 if (level == LKM_PRMODE)
447 stats = &res->l_lock_prmode;
448 else if (level == LKM_EXMODE)
449 stats = &res->l_lock_exmode;
450 else
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700451 return;
452
Sunil Mushran5bc970e2010-12-28 23:26:03 -0800453 kt = ktime_sub(ktime_get(), mw->mw_lock_start);
454 usec = ktime_to_us(kt);
455
456 stats->ls_gets++;
457 stats->ls_total += ktime_to_ns(kt);
458 /* overflow */
459 if (unlikely(stats->ls_gets) == 0) {
460 stats->ls_gets++;
461 stats->ls_total = ktime_to_ns(kt);
462 }
463
464 if (stats->ls_max < usec)
465 stats->ls_max = usec;
466
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700467 if (ret)
Sunil Mushran5bc970e2010-12-28 23:26:03 -0800468 stats->ls_fail++;
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700469}
470
471static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
472{
473 lockres->l_lock_refresh++;
474}
475
476static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
477{
Sunil Mushran5bc970e2010-12-28 23:26:03 -0800478 mw->mw_lock_start = ktime_get();
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700479}
480#else
481static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
482{
483}
484static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
485 int level, struct ocfs2_mask_waiter *mw, int ret)
486{
487}
488static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
489{
490}
491static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
492{
493}
494#endif
495
Mark Fashehccd979b2005-12-15 14:31:24 -0800496static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
497 struct ocfs2_lock_res *res,
498 enum ocfs2_lock_type type,
Mark Fashehccd979b2005-12-15 14:31:24 -0800499 struct ocfs2_lock_res_ops *ops,
500 void *priv)
501{
Mark Fashehccd979b2005-12-15 14:31:24 -0800502 res->l_type = type;
503 res->l_ops = ops;
504 res->l_priv = priv;
505
Joel Beckerbd3e7612008-02-01 12:14:57 -0800506 res->l_level = DLM_LOCK_IV;
507 res->l_requested = DLM_LOCK_IV;
508 res->l_blocking = DLM_LOCK_IV;
Mark Fashehccd979b2005-12-15 14:31:24 -0800509 res->l_action = OCFS2_AST_INVALID;
510 res->l_unlock_action = OCFS2_UNLOCK_INVALID;
511
512 res->l_flags = OCFS2_LOCK_INITIALIZED;
513
514 ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700515
516 ocfs2_init_lock_stats(res);
Jan Karacb257972009-06-04 15:26:50 +0200517#ifdef CONFIG_DEBUG_LOCK_ALLOC
518 if (type != OCFS2_LOCK_TYPE_OPEN)
519 lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type],
520 &lockdep_keys[type], 0);
521 else
522 res->l_lockdep_map.key = NULL;
523#endif
Mark Fashehccd979b2005-12-15 14:31:24 -0800524}
525
526void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
527{
528 /* This also clears out the lock status block */
529 memset(res, 0, sizeof(struct ocfs2_lock_res));
530 spin_lock_init(&res->l_lock);
531 init_waitqueue_head(&res->l_event);
532 INIT_LIST_HEAD(&res->l_blocked_list);
533 INIT_LIST_HEAD(&res->l_mask_waiters);
534}
535
536void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
537 enum ocfs2_lock_type type,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700538 unsigned int generation,
Mark Fashehccd979b2005-12-15 14:31:24 -0800539 struct inode *inode)
540{
541 struct ocfs2_lock_res_ops *ops;
542
543 switch(type) {
544 case OCFS2_LOCK_TYPE_RW:
545 ops = &ocfs2_inode_rw_lops;
546 break;
547 case OCFS2_LOCK_TYPE_META:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700548 ops = &ocfs2_inode_inode_lops;
Mark Fashehccd979b2005-12-15 14:31:24 -0800549 break;
Tiger Yang50008632007-03-20 16:01:38 -0700550 case OCFS2_LOCK_TYPE_OPEN:
551 ops = &ocfs2_inode_open_lops;
552 break;
Mark Fashehccd979b2005-12-15 14:31:24 -0800553 default:
554 mlog_bug_on_msg(1, "type: %d\n", type);
555 ops = NULL; /* thanks, gcc */
556 break;
557 };
558
Mark Fashehd680efe2006-09-08 14:14:34 -0700559 ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700560 generation, res->l_name);
Mark Fashehd680efe2006-09-08 14:14:34 -0700561 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
562}
563
Mark Fasheh54a7e752006-09-12 21:49:13 -0700564static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
565{
566 struct inode *inode = ocfs2_lock_res_inode(lockres);
567
568 return OCFS2_SB(inode->i_sb);
569}
570
Jan Kara9e33d692008-08-25 19:56:50 +0200571static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
572{
573 struct ocfs2_mem_dqinfo *info = lockres->l_priv;
574
575 return OCFS2_SB(info->dqi_gi.dqi_sb);
576}
577
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800578static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
579{
580 struct ocfs2_file_private *fp = lockres->l_priv;
581
582 return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
583}
584
Mark Fashehd680efe2006-09-08 14:14:34 -0700585static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
586{
587 __be64 inode_blkno_be;
588
589 memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
590 sizeof(__be64));
591
592 return be64_to_cpu(inode_blkno_be);
593}
594
Mark Fasheh54a7e752006-09-12 21:49:13 -0700595static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
596{
597 struct ocfs2_dentry_lock *dl = lockres->l_priv;
598
599 return OCFS2_SB(dl->dl_inode->i_sb);
600}
601
Mark Fashehd680efe2006-09-08 14:14:34 -0700602void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
603 u64 parent, struct inode *inode)
604{
605 int len;
606 u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
607 __be64 inode_blkno_be = cpu_to_be64(inode_blkno);
608 struct ocfs2_lock_res *lockres = &dl->dl_lockres;
609
610 ocfs2_lock_res_init_once(lockres);
611
612 /*
613 * Unfortunately, the standard lock naming scheme won't work
614 * here because we have two 16 byte values to use. Instead,
615 * we'll stuff the inode number as a binary value. We still
616 * want error prints to show something without garbling the
617 * display, so drop a null byte in there before the inode
618 * number. A future version of OCFS2 will likely use all
619 * binary lock names. The stringified names have been a
620 * tremendous aid in debugging, but now that the debugfs
621 * interface exists, we can mangle things there if need be.
622 *
623 * NOTE: We also drop the standard "pad" value (the total lock
624 * name size stays the same though - the last part is all
625 * zeros due to the memset in ocfs2_lock_res_init_once()
626 */
627 len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
628 "%c%016llx",
629 ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
630 (long long)parent);
631
632 BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
633
634 memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
635 sizeof(__be64));
636
637 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
638 OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
639 dl);
Mark Fashehccd979b2005-12-15 14:31:24 -0800640}
641
642static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
643 struct ocfs2_super *osb)
644{
645 /* Superblock lockres doesn't come from a slab so we call init
646 * once on it manually. */
647 ocfs2_lock_res_init_once(res);
Mark Fashehd680efe2006-09-08 14:14:34 -0700648 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
649 0, res->l_name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800650 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
Mark Fashehccd979b2005-12-15 14:31:24 -0800651 &ocfs2_super_lops, osb);
652}
653
654static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
655 struct ocfs2_super *osb)
656{
657 /* Rename lockres doesn't come from a slab so we call init
658 * once on it manually. */
659 ocfs2_lock_res_init_once(res);
Mark Fashehd680efe2006-09-08 14:14:34 -0700660 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
661 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
Mark Fashehccd979b2005-12-15 14:31:24 -0800662 &ocfs2_rename_lops, osb);
663}
664
wengang wang6ca497a2009-03-06 21:29:10 +0800665static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
666 struct ocfs2_super *osb)
667{
668 /* nfs_sync lockres doesn't come from a slab so we call init
669 * once on it manually. */
670 ocfs2_lock_res_init_once(res);
671 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
672 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
673 &ocfs2_nfs_sync_lops, osb);
674}
675
Srinivas Eeda83273932009-06-03 17:02:55 -0700676static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
677 struct ocfs2_super *osb)
678{
Srinivas Eeda83273932009-06-03 17:02:55 -0700679 ocfs2_lock_res_init_once(res);
680 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
681 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
682 &ocfs2_orphan_scan_lops, osb);
Srinivas Eeda83273932009-06-03 17:02:55 -0700683}
684
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800685void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
686 struct ocfs2_file_private *fp)
687{
688 struct inode *inode = fp->fp_file->f_mapping->host;
689 struct ocfs2_inode_info *oi = OCFS2_I(inode);
690
691 ocfs2_lock_res_init_once(lockres);
692 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
693 inode->i_generation, lockres->l_name);
694 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
695 OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
696 fp);
697 lockres->l_flags |= OCFS2_LOCK_NOCACHE;
698}
699
Jan Kara9e33d692008-08-25 19:56:50 +0200700void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
701 struct ocfs2_mem_dqinfo *info)
702{
703 ocfs2_lock_res_init_once(lockres);
704 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
705 0, lockres->l_name);
706 ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
707 OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
708 info);
709}
710
Tao Ma8dec98e2009-08-18 11:19:58 +0800711void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
712 struct ocfs2_super *osb, u64 ref_blkno,
713 unsigned int generation)
714{
715 ocfs2_lock_res_init_once(lockres);
716 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
717 generation, lockres->l_name);
718 ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
719 &ocfs2_refcount_block_lops, osb);
720}
721
Mark Fashehccd979b2005-12-15 14:31:24 -0800722void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
723{
Mark Fashehccd979b2005-12-15 14:31:24 -0800724 if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
725 return;
726
727 ocfs2_remove_lockres_tracking(res);
728
729 mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
730 "Lockres %s is on the blocked list\n",
731 res->l_name);
732 mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
733 "Lockres %s has mask waiters pending\n",
734 res->l_name);
735 mlog_bug_on_msg(spin_is_locked(&res->l_lock),
736 "Lockres %s is locked\n",
737 res->l_name);
738 mlog_bug_on_msg(res->l_ro_holders,
739 "Lockres %s has %u ro holders\n",
740 res->l_name, res->l_ro_holders);
741 mlog_bug_on_msg(res->l_ex_holders,
742 "Lockres %s has %u ex holders\n",
743 res->l_name, res->l_ex_holders);
744
745 /* Need to clear out the lock status block for the dlm */
746 memset(&res->l_lksb, 0, sizeof(res->l_lksb));
747
748 res->l_flags = 0UL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800749}
750
751static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
752 int level)
753{
Mark Fashehccd979b2005-12-15 14:31:24 -0800754 BUG_ON(!lockres);
755
756 switch(level) {
Joel Beckerbd3e7612008-02-01 12:14:57 -0800757 case DLM_LOCK_EX:
Mark Fashehccd979b2005-12-15 14:31:24 -0800758 lockres->l_ex_holders++;
759 break;
Joel Beckerbd3e7612008-02-01 12:14:57 -0800760 case DLM_LOCK_PR:
Mark Fashehccd979b2005-12-15 14:31:24 -0800761 lockres->l_ro_holders++;
762 break;
763 default:
764 BUG();
765 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800766}
767
768static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
769 int level)
770{
Mark Fashehccd979b2005-12-15 14:31:24 -0800771 BUG_ON(!lockres);
772
773 switch(level) {
Joel Beckerbd3e7612008-02-01 12:14:57 -0800774 case DLM_LOCK_EX:
Mark Fashehccd979b2005-12-15 14:31:24 -0800775 BUG_ON(!lockres->l_ex_holders);
776 lockres->l_ex_holders--;
777 break;
Joel Beckerbd3e7612008-02-01 12:14:57 -0800778 case DLM_LOCK_PR:
Mark Fashehccd979b2005-12-15 14:31:24 -0800779 BUG_ON(!lockres->l_ro_holders);
780 lockres->l_ro_holders--;
781 break;
782 default:
783 BUG();
784 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800785}
786
787/* WARNING: This function lives in a world where the only three lock
788 * levels are EX, PR, and NL. It *will* have to be adjusted when more
789 * lock types are added. */
790static inline int ocfs2_highest_compat_lock_level(int level)
791{
Joel Beckerbd3e7612008-02-01 12:14:57 -0800792 int new_level = DLM_LOCK_EX;
Mark Fashehccd979b2005-12-15 14:31:24 -0800793
Joel Beckerbd3e7612008-02-01 12:14:57 -0800794 if (level == DLM_LOCK_EX)
795 new_level = DLM_LOCK_NL;
796 else if (level == DLM_LOCK_PR)
797 new_level = DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -0800798 return new_level;
799}
800
801static void lockres_set_flags(struct ocfs2_lock_res *lockres,
802 unsigned long newflags)
803{
Christoph Hellwig800deef2007-05-17 16:03:13 +0200804 struct ocfs2_mask_waiter *mw, *tmp;
Mark Fashehccd979b2005-12-15 14:31:24 -0800805
806 assert_spin_locked(&lockres->l_lock);
807
808 lockres->l_flags = newflags;
809
Christoph Hellwig800deef2007-05-17 16:03:13 +0200810 list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800811 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
812 continue;
813
814 list_del_init(&mw->mw_item);
815 mw->mw_status = 0;
816 complete(&mw->mw_complete);
817 }
818}
819static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
820{
821 lockres_set_flags(lockres, lockres->l_flags | or);
822}
823static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
824 unsigned long clear)
825{
826 lockres_set_flags(lockres, lockres->l_flags & ~clear);
827}
828
829static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
830{
Mark Fashehccd979b2005-12-15 14:31:24 -0800831 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
832 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
833 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
Joel Beckerbd3e7612008-02-01 12:14:57 -0800834 BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800835
836 lockres->l_level = lockres->l_requested;
837 if (lockres->l_level <=
838 ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
Joel Beckerbd3e7612008-02-01 12:14:57 -0800839 lockres->l_blocking = DLM_LOCK_NL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800840 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
841 }
842 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800843}
844
845static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
846{
Mark Fashehccd979b2005-12-15 14:31:24 -0800847 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
848 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
849
850 /* Convert from RO to EX doesn't really need anything as our
851 * information is already up to data. Convert from NL to
852 * *anything* however should mark ourselves as needing an
853 * update */
Joel Beckerbd3e7612008-02-01 12:14:57 -0800854 if (lockres->l_level == DLM_LOCK_NL &&
Mark Fashehf625c972006-09-12 21:24:53 -0700855 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
Mark Fashehccd979b2005-12-15 14:31:24 -0800856 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
857
858 lockres->l_level = lockres->l_requested;
Sunil Mushrana1912822010-01-21 10:50:03 -0800859
860 /*
861 * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
862 * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
863 * downconverting the lock before the upconvert has fully completed.
864 */
865 lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
866
Mark Fashehccd979b2005-12-15 14:31:24 -0800867 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800868}
869
870static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
871{
Roel Kluin3cf0c502007-10-27 00:20:36 +0200872 BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
Mark Fashehccd979b2005-12-15 14:31:24 -0800873 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
874
Joel Beckerbd3e7612008-02-01 12:14:57 -0800875 if (lockres->l_requested > DLM_LOCK_NL &&
Mark Fashehf625c972006-09-12 21:24:53 -0700876 !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
877 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
Mark Fashehccd979b2005-12-15 14:31:24 -0800878 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
879
880 lockres->l_level = lockres->l_requested;
881 lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
882 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
Mark Fashehccd979b2005-12-15 14:31:24 -0800883}
884
Mark Fashehccd979b2005-12-15 14:31:24 -0800885static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
886 int level)
887{
888 int needs_downconvert = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800889
890 assert_spin_locked(&lockres->l_lock);
891
Mark Fashehccd979b2005-12-15 14:31:24 -0800892 if (level > lockres->l_blocking) {
893 /* only schedule a downconvert if we haven't already scheduled
894 * one that goes low enough to satisfy the level we're
895 * blocking. this also catches the case where we get
896 * duplicate BASTs */
897 if (ocfs2_highest_compat_lock_level(level) <
898 ocfs2_highest_compat_lock_level(lockres->l_blocking))
899 needs_downconvert = 1;
900
901 lockres->l_blocking = level;
902 }
903
Sunil Mushran9b915182010-02-26 19:42:44 -0800904 mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n",
905 lockres->l_name, level, lockres->l_level, lockres->l_blocking,
906 needs_downconvert);
907
Wengang Wang0b94a902010-01-21 10:50:02 -0800908 if (needs_downconvert)
909 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
Tao Mac1e8d352011-03-07 16:43:21 +0800910 mlog(0, "needs_downconvert = %d\n", needs_downconvert);
Mark Fashehccd979b2005-12-15 14:31:24 -0800911 return needs_downconvert;
912}
913
Joel Beckerde551242008-02-01 14:45:08 -0800914/*
915 * OCFS2_LOCK_PENDING and l_pending_gen.
916 *
917 * Why does OCFS2_LOCK_PENDING exist? To close a race between setting
918 * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock(). See ocfs2_unblock_lock()
919 * for more details on the race.
920 *
921 * OCFS2_LOCK_PENDING closes the race quite nicely. However, it introduces
922 * a race on itself. In o2dlm, we can get the ast before ocfs2_dlm_lock()
923 * returns. The ast clears OCFS2_LOCK_BUSY, and must therefore clear
924 * OCFS2_LOCK_PENDING at the same time. When ocfs2_dlm_lock() returns,
925 * the caller is going to try to clear PENDING again. If nothing else is
926 * happening, __lockres_clear_pending() sees PENDING is unset and does
927 * nothing.
928 *
929 * But what if another path (eg downconvert thread) has just started a
930 * new locking action? The other path has re-set PENDING. Our path
931 * cannot clear PENDING, because that will re-open the original race
932 * window.
933 *
934 * [Example]
935 *
936 * ocfs2_meta_lock()
937 * ocfs2_cluster_lock()
938 * set BUSY
939 * set PENDING
940 * drop l_lock
941 * ocfs2_dlm_lock()
942 * ocfs2_locking_ast() ocfs2_downconvert_thread()
943 * clear PENDING ocfs2_unblock_lock()
944 * take_l_lock
945 * !BUSY
946 * ocfs2_prepare_downconvert()
947 * set BUSY
948 * set PENDING
949 * drop l_lock
950 * take l_lock
951 * clear PENDING
952 * drop l_lock
953 * <window>
954 * ocfs2_dlm_lock()
955 *
956 * So as you can see, we now have a window where l_lock is not held,
957 * PENDING is not set, and ocfs2_dlm_lock() has not been called.
958 *
959 * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
960 * set by ocfs2_prepare_downconvert(). That wasn't nice.
961 *
962 * To solve this we introduce l_pending_gen. A call to
963 * lockres_clear_pending() will only do so when it is passed a generation
964 * number that matches the lockres. lockres_set_pending() will return the
965 * current generation number. When ocfs2_cluster_lock() goes to clear
966 * PENDING, it passes the generation it got from set_pending(). In our
967 * example above, the generation numbers will *not* match. Thus,
968 * ocfs2_cluster_lock() will not clear the PENDING set by
969 * ocfs2_prepare_downconvert().
970 */
971
972/* Unlocked version for ocfs2_locking_ast() */
973static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
974 unsigned int generation,
975 struct ocfs2_super *osb)
976{
977 assert_spin_locked(&lockres->l_lock);
978
979 /*
980 * The ast and locking functions can race us here. The winner
981 * will clear pending, the loser will not.
982 */
983 if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
984 (lockres->l_pending_gen != generation))
985 return;
986
987 lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
988 lockres->l_pending_gen++;
989
990 /*
991 * The downconvert thread may have skipped us because we
992 * were PENDING. Wake it up.
993 */
994 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
995 ocfs2_wake_downconvert_thread(osb);
996}
997
998/* Locked version for callers of ocfs2_dlm_lock() */
999static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
1000 unsigned int generation,
1001 struct ocfs2_super *osb)
1002{
1003 unsigned long flags;
1004
1005 spin_lock_irqsave(&lockres->l_lock, flags);
1006 __lockres_clear_pending(lockres, generation, osb);
1007 spin_unlock_irqrestore(&lockres->l_lock, flags);
1008}
1009
1010static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
1011{
1012 assert_spin_locked(&lockres->l_lock);
1013 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
1014
1015 lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
1016
1017 return lockres->l_pending_gen;
1018}
1019
Joel Beckerc0e41332010-01-29 14:46:44 -08001020static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level)
Mark Fashehccd979b2005-12-15 14:31:24 -08001021{
Joel Beckera796d282010-01-28 19:22:39 -08001022 struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
Mark Fashehaa2623a2006-09-12 21:58:23 -07001023 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001024 int needs_downconvert;
1025 unsigned long flags;
1026
Joel Beckerbd3e7612008-02-01 12:14:57 -08001027 BUG_ON(level <= DLM_LOCK_NL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001028
Sunil Mushran9b915182010-02-26 19:42:44 -08001029 mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, "
1030 "type %s\n", lockres->l_name, level, lockres->l_level,
Mark Fashehaa2623a2006-09-12 21:58:23 -07001031 ocfs2_lock_type_string(lockres->l_type));
1032
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001033 /*
1034 * We can skip the bast for locks which don't enable caching -
1035 * they'll be dropped at the earliest possible time anyway.
1036 */
1037 if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
1038 return;
1039
Mark Fashehccd979b2005-12-15 14:31:24 -08001040 spin_lock_irqsave(&lockres->l_lock, flags);
1041 needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
1042 if (needs_downconvert)
1043 ocfs2_schedule_blocked_lock(osb, lockres);
1044 spin_unlock_irqrestore(&lockres->l_lock, flags);
1045
Mark Fashehd680efe2006-09-08 14:14:34 -07001046 wake_up(&lockres->l_event);
1047
Mark Fasheh34d024f2007-09-24 15:56:19 -07001048 ocfs2_wake_downconvert_thread(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001049}
1050
Joel Beckerc0e41332010-01-29 14:46:44 -08001051static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb)
Mark Fashehccd979b2005-12-15 14:31:24 -08001052{
Joel Beckera796d282010-01-28 19:22:39 -08001053 struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
Joel Beckerde551242008-02-01 14:45:08 -08001054 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001055 unsigned long flags;
David Teigland1693a5c2008-01-30 16:52:53 -08001056 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -08001057
1058 spin_lock_irqsave(&lockres->l_lock, flags);
1059
David Teigland1693a5c2008-01-30 16:52:53 -08001060 status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1061
1062 if (status == -EAGAIN) {
1063 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1064 goto out;
1065 }
1066
1067 if (status) {
Joel Becker8f2c9c12008-02-01 12:16:57 -08001068 mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
David Teigland1693a5c2008-01-30 16:52:53 -08001069 lockres->l_name, status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001070 spin_unlock_irqrestore(&lockres->l_lock, flags);
1071 return;
1072 }
1073
Sunil Mushran9b915182010-02-26 19:42:44 -08001074 mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, "
1075 "level %d => %d\n", lockres->l_name, lockres->l_action,
1076 lockres->l_unlock_action, lockres->l_level, lockres->l_requested);
1077
Mark Fashehccd979b2005-12-15 14:31:24 -08001078 switch(lockres->l_action) {
1079 case OCFS2_AST_ATTACH:
1080 ocfs2_generic_handle_attach_action(lockres);
Mark Fashehe92d57d2006-09-12 21:34:35 -07001081 lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001082 break;
1083 case OCFS2_AST_CONVERT:
1084 ocfs2_generic_handle_convert_action(lockres);
1085 break;
1086 case OCFS2_AST_DOWNCONVERT:
1087 ocfs2_generic_handle_downconvert_action(lockres);
1088 break;
1089 default:
Sunil Mushran9b915182010-02-26 19:42:44 -08001090 mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, "
1091 "flags 0x%lx, unlock: %u\n",
Mark Fashehe92d57d2006-09-12 21:34:35 -07001092 lockres->l_name, lockres->l_action, lockres->l_flags,
1093 lockres->l_unlock_action);
Mark Fashehccd979b2005-12-15 14:31:24 -08001094 BUG();
1095 }
David Teigland1693a5c2008-01-30 16:52:53 -08001096out:
Mark Fashehccd979b2005-12-15 14:31:24 -08001097 /* set it to something invalid so if we get called again we
1098 * can catch it. */
1099 lockres->l_action = OCFS2_AST_INVALID;
Mark Fashehccd979b2005-12-15 14:31:24 -08001100
Joel Beckerde551242008-02-01 14:45:08 -08001101 /* Did we try to cancel this lock? Clear that state */
1102 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1103 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1104
1105 /*
1106 * We may have beaten the locking functions here. We certainly
1107 * know that dlm_lock() has been called :-)
1108 * Because we can't have two lock calls in flight at once, we
1109 * can use lockres->l_pending_gen.
1110 */
1111 __lockres_clear_pending(lockres, lockres->l_pending_gen, osb);
1112
Mark Fashehccd979b2005-12-15 14:31:24 -08001113 wake_up(&lockres->l_event);
Mark Fashehd680efe2006-09-08 14:14:34 -07001114 spin_unlock_irqrestore(&lockres->l_lock, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001115}
1116
Joel Becker553b5eb2010-01-29 17:19:06 -08001117static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error)
1118{
1119 struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
1120 unsigned long flags;
1121
Sunil Mushran9b915182010-02-26 19:42:44 -08001122 mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n",
1123 lockres->l_name, lockres->l_unlock_action);
Joel Becker553b5eb2010-01-29 17:19:06 -08001124
1125 spin_lock_irqsave(&lockres->l_lock, flags);
1126 if (error) {
1127 mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
1128 "unlock_action %d\n", error, lockres->l_name,
1129 lockres->l_unlock_action);
1130 spin_unlock_irqrestore(&lockres->l_lock, flags);
Joel Becker553b5eb2010-01-29 17:19:06 -08001131 return;
1132 }
1133
1134 switch(lockres->l_unlock_action) {
1135 case OCFS2_UNLOCK_CANCEL_CONVERT:
1136 mlog(0, "Cancel convert success for %s\n", lockres->l_name);
1137 lockres->l_action = OCFS2_AST_INVALID;
1138 /* Downconvert thread may have requeued this lock, we
1139 * need to wake it. */
1140 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
1141 ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
1142 break;
1143 case OCFS2_UNLOCK_DROP_LOCK:
1144 lockres->l_level = DLM_LOCK_IV;
1145 break;
1146 default:
1147 BUG();
1148 }
1149
1150 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1151 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1152 wake_up(&lockres->l_event);
1153 spin_unlock_irqrestore(&lockres->l_lock, flags);
Joel Becker553b5eb2010-01-29 17:19:06 -08001154}
1155
1156/*
1157 * This is the filesystem locking protocol. It provides the lock handling
1158 * hooks for the underlying DLM. It has a maximum version number.
1159 * The version number allows interoperability with systems running at
1160 * the same major number and an equal or smaller minor number.
1161 *
1162 * Whenever the filesystem does new things with locks (adds or removes a
1163 * lock, orders them differently, does different things underneath a lock),
1164 * the version must be changed. The protocol is negotiated when joining
1165 * the dlm domain. A node may join the domain if its major version is
1166 * identical to all other nodes and its minor version is greater than
1167 * or equal to all other nodes. When its minor version is greater than
1168 * the other nodes, it will run at the minor version specified by the
1169 * other nodes.
1170 *
1171 * If a locking change is made that will not be compatible with older
1172 * versions, the major number must be increased and the minor version set
1173 * to zero. If a change merely adds a behavior that can be disabled when
1174 * speaking to older versions, the minor version must be increased. If a
1175 * change adds a fully backwards compatible change (eg, LVB changes that
1176 * are just ignored by older versions), the version does not need to be
1177 * updated.
1178 */
1179static struct ocfs2_locking_protocol lproto = {
1180 .lp_max_version = {
1181 .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
1182 .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
1183 },
1184 .lp_lock_ast = ocfs2_locking_ast,
1185 .lp_blocking_ast = ocfs2_blocking_ast,
1186 .lp_unlock_ast = ocfs2_unlock_ast,
1187};
1188
1189void ocfs2_set_locking_protocol(void)
1190{
1191 ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version);
1192}
1193
Mark Fashehccd979b2005-12-15 14:31:24 -08001194static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1195 int convert)
1196{
1197 unsigned long flags;
1198
Mark Fashehccd979b2005-12-15 14:31:24 -08001199 spin_lock_irqsave(&lockres->l_lock, flags);
1200 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
Sunil Mushrana1912822010-01-21 10:50:03 -08001201 lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
Mark Fashehccd979b2005-12-15 14:31:24 -08001202 if (convert)
1203 lockres->l_action = OCFS2_AST_INVALID;
1204 else
1205 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1206 spin_unlock_irqrestore(&lockres->l_lock, flags);
1207
1208 wake_up(&lockres->l_event);
Mark Fashehccd979b2005-12-15 14:31:24 -08001209}
1210
1211/* Note: If we detect another process working on the lock (i.e.,
1212 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1213 * to do the right thing in that case.
1214 */
1215static int ocfs2_lock_create(struct ocfs2_super *osb,
1216 struct ocfs2_lock_res *lockres,
1217 int level,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001218 u32 dlm_flags)
Mark Fashehccd979b2005-12-15 14:31:24 -08001219{
1220 int ret = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001221 unsigned long flags;
Joel Beckerde551242008-02-01 14:45:08 -08001222 unsigned int gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08001223
Joel Beckerbd3e7612008-02-01 12:14:57 -08001224 mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
Mark Fashehccd979b2005-12-15 14:31:24 -08001225 dlm_flags);
1226
1227 spin_lock_irqsave(&lockres->l_lock, flags);
1228 if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1229 (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1230 spin_unlock_irqrestore(&lockres->l_lock, flags);
1231 goto bail;
1232 }
1233
1234 lockres->l_action = OCFS2_AST_ATTACH;
1235 lockres->l_requested = level;
1236 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
Joel Beckerde551242008-02-01 14:45:08 -08001237 gen = lockres_set_pending(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001238 spin_unlock_irqrestore(&lockres->l_lock, flags);
1239
Joel Becker4670c462008-02-01 14:39:35 -08001240 ret = ocfs2_dlm_lock(osb->cconn,
Joel Becker7431cd72008-02-01 12:15:37 -08001241 level,
1242 &lockres->l_lksb,
1243 dlm_flags,
1244 lockres->l_name,
Joel Beckera796d282010-01-28 19:22:39 -08001245 OCFS2_LOCK_ID_MAX_LEN - 1);
Joel Beckerde551242008-02-01 14:45:08 -08001246 lockres_clear_pending(lockres, gen, osb);
Joel Becker7431cd72008-02-01 12:15:37 -08001247 if (ret) {
1248 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001249 ocfs2_recover_from_dlm_error(lockres, 1);
1250 }
1251
Joel Becker7431cd72008-02-01 12:15:37 -08001252 mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
Mark Fashehccd979b2005-12-15 14:31:24 -08001253
1254bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001255 return ret;
1256}
1257
1258static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1259 int flag)
1260{
1261 unsigned long flags;
1262 int ret;
1263
1264 spin_lock_irqsave(&lockres->l_lock, flags);
1265 ret = lockres->l_flags & flag;
1266 spin_unlock_irqrestore(&lockres->l_lock, flags);
1267
1268 return ret;
1269}
1270
1271static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1272
1273{
1274 wait_event(lockres->l_event,
1275 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1276}
1277
1278static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1279
1280{
1281 wait_event(lockres->l_event,
1282 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1283}
1284
1285/* predict what lock level we'll be dropping down to on behalf
1286 * of another node, and return true if the currently wanted
1287 * level will be compatible with it. */
1288static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1289 int wanted)
1290{
1291 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1292
1293 return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1294}
1295
1296static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1297{
1298 INIT_LIST_HEAD(&mw->mw_item);
1299 init_completion(&mw->mw_complete);
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07001300 ocfs2_init_start_time(mw);
Mark Fashehccd979b2005-12-15 14:31:24 -08001301}
1302
1303static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1304{
1305 wait_for_completion(&mw->mw_complete);
1306 /* Re-arm the completion in case we want to wait on it again */
1307 INIT_COMPLETION(mw->mw_complete);
1308 return mw->mw_status;
1309}
1310
1311static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1312 struct ocfs2_mask_waiter *mw,
1313 unsigned long mask,
1314 unsigned long goal)
1315{
1316 BUG_ON(!list_empty(&mw->mw_item));
1317
1318 assert_spin_locked(&lockres->l_lock);
1319
1320 list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1321 mw->mw_mask = mask;
1322 mw->mw_goal = goal;
1323}
1324
1325/* returns 0 if the mw that was removed was already satisfied, -EBUSY
1326 * if the mask still hadn't reached its goal */
1327static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1328 struct ocfs2_mask_waiter *mw)
1329{
1330 unsigned long flags;
1331 int ret = 0;
1332
1333 spin_lock_irqsave(&lockres->l_lock, flags);
1334 if (!list_empty(&mw->mw_item)) {
1335 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1336 ret = -EBUSY;
1337
1338 list_del_init(&mw->mw_item);
1339 init_completion(&mw->mw_complete);
1340 }
1341 spin_unlock_irqrestore(&lockres->l_lock, flags);
1342
1343 return ret;
1344
1345}
1346
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001347static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1348 struct ocfs2_lock_res *lockres)
1349{
1350 int ret;
1351
1352 ret = wait_for_completion_interruptible(&mw->mw_complete);
1353 if (ret)
1354 lockres_remove_mask_waiter(lockres, mw);
1355 else
1356 ret = mw->mw_status;
1357 /* Re-arm the completion in case we want to wait on it again */
1358 INIT_COMPLETION(mw->mw_complete);
1359 return ret;
1360}
1361
Jan Karacb257972009-06-04 15:26:50 +02001362static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
1363 struct ocfs2_lock_res *lockres,
1364 int level,
1365 u32 lkm_flags,
1366 int arg_flags,
1367 int l_subclass,
1368 unsigned long caller_ip)
Mark Fashehccd979b2005-12-15 14:31:24 -08001369{
1370 struct ocfs2_mask_waiter mw;
Mark Fashehccd979b2005-12-15 14:31:24 -08001371 int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1372 int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1373 unsigned long flags;
Joel Beckerde551242008-02-01 14:45:08 -08001374 unsigned int gen;
David Teigland1693a5c2008-01-30 16:52:53 -08001375 int noqueue_attempted = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001376
Mark Fashehccd979b2005-12-15 14:31:24 -08001377 ocfs2_init_mask_waiter(&mw);
1378
Mark Fashehb80fc012006-09-12 22:08:14 -07001379 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
Joel Beckerbd3e7612008-02-01 12:14:57 -08001380 lkm_flags |= DLM_LKF_VALBLK;
Mark Fashehb80fc012006-09-12 22:08:14 -07001381
Mark Fashehccd979b2005-12-15 14:31:24 -08001382again:
1383 wait = 0;
1384
Sunil Mushrana1912822010-01-21 10:50:03 -08001385 spin_lock_irqsave(&lockres->l_lock, flags);
1386
Mark Fashehccd979b2005-12-15 14:31:24 -08001387 if (catch_signals && signal_pending(current)) {
1388 ret = -ERESTARTSYS;
Sunil Mushrana1912822010-01-21 10:50:03 -08001389 goto unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001390 }
1391
Mark Fashehccd979b2005-12-15 14:31:24 -08001392 mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1393 "Cluster lock called on freeing lockres %s! flags "
1394 "0x%lx\n", lockres->l_name, lockres->l_flags);
1395
1396 /* We only compare against the currently granted level
1397 * here. If the lock is blocked waiting on a downconvert,
1398 * we'll get caught below. */
1399 if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1400 level > lockres->l_level) {
1401 /* is someone sitting in dlm_lock? If so, wait on
1402 * them. */
1403 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1404 wait = 1;
1405 goto unlock;
1406 }
1407
Sunil Mushrana1912822010-01-21 10:50:03 -08001408 if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
1409 /*
1410 * We've upconverted. If the lock now has a level we can
1411 * work with, we take it. If, however, the lock is not at the
1412 * required level, we go thru the full cycle. One way this could
1413 * happen is if a process requesting an upconvert to PR is
1414 * closely followed by another requesting upconvert to an EX.
1415 * If the process requesting EX lands here, we want it to
1416 * continue attempting to upconvert and let the process
1417 * requesting PR take the lock.
1418 * If multiple processes request upconvert to PR, the first one
1419 * here will take the lock. The others will have to go thru the
1420 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
1421 * downconvert request.
1422 */
1423 if (level <= lockres->l_level)
1424 goto update_holders;
1425 }
1426
Mark Fashehccd979b2005-12-15 14:31:24 -08001427 if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1428 !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1429 /* is the lock is currently blocked on behalf of
1430 * another node */
1431 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1432 wait = 1;
1433 goto unlock;
1434 }
1435
1436 if (level > lockres->l_level) {
David Teigland1693a5c2008-01-30 16:52:53 -08001437 if (noqueue_attempted > 0) {
1438 ret = -EAGAIN;
1439 goto unlock;
1440 }
1441 if (lkm_flags & DLM_LKF_NOQUEUE)
1442 noqueue_attempted = 1;
1443
Mark Fashehccd979b2005-12-15 14:31:24 -08001444 if (lockres->l_action != OCFS2_AST_INVALID)
1445 mlog(ML_ERROR, "lockres %s has action %u pending\n",
1446 lockres->l_name, lockres->l_action);
1447
Mark Fasheh019d1b22007-10-05 12:09:05 -07001448 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1449 lockres->l_action = OCFS2_AST_ATTACH;
Joel Beckerbd3e7612008-02-01 12:14:57 -08001450 lkm_flags &= ~DLM_LKF_CONVERT;
Mark Fasheh019d1b22007-10-05 12:09:05 -07001451 } else {
1452 lockres->l_action = OCFS2_AST_CONVERT;
Joel Beckerbd3e7612008-02-01 12:14:57 -08001453 lkm_flags |= DLM_LKF_CONVERT;
Mark Fasheh019d1b22007-10-05 12:09:05 -07001454 }
1455
Mark Fashehccd979b2005-12-15 14:31:24 -08001456 lockres->l_requested = level;
1457 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
Joel Beckerde551242008-02-01 14:45:08 -08001458 gen = lockres_set_pending(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001459 spin_unlock_irqrestore(&lockres->l_lock, flags);
1460
Joel Beckerbd3e7612008-02-01 12:14:57 -08001461 BUG_ON(level == DLM_LOCK_IV);
1462 BUG_ON(level == DLM_LOCK_NL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001463
Sunil Mushran9b915182010-02-26 19:42:44 -08001464 mlog(ML_BASTS, "lockres %s, convert from %d to %d\n",
Mark Fashehccd979b2005-12-15 14:31:24 -08001465 lockres->l_name, lockres->l_level, level);
1466
1467 /* call dlm_lock to upgrade lock now */
Joel Becker4670c462008-02-01 14:39:35 -08001468 ret = ocfs2_dlm_lock(osb->cconn,
Joel Becker7431cd72008-02-01 12:15:37 -08001469 level,
1470 &lockres->l_lksb,
1471 lkm_flags,
1472 lockres->l_name,
Joel Beckera796d282010-01-28 19:22:39 -08001473 OCFS2_LOCK_ID_MAX_LEN - 1);
Joel Beckerde551242008-02-01 14:45:08 -08001474 lockres_clear_pending(lockres, gen, osb);
Joel Becker7431cd72008-02-01 12:15:37 -08001475 if (ret) {
1476 if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1477 (ret != -EAGAIN)) {
Joel Becker24ef1812008-01-29 17:37:32 -08001478 ocfs2_log_dlm_error("ocfs2_dlm_lock",
Joel Becker7431cd72008-02-01 12:15:37 -08001479 ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001480 }
1481 ocfs2_recover_from_dlm_error(lockres, 1);
1482 goto out;
1483 }
1484
Coly Li73ac36e2009-01-07 18:09:16 -08001485 mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
Mark Fashehccd979b2005-12-15 14:31:24 -08001486 lockres->l_name);
1487
1488 /* At this point we've gone inside the dlm and need to
1489 * complete our work regardless. */
1490 catch_signals = 0;
1491
1492 /* wait for busy to clear and carry on */
1493 goto again;
1494 }
1495
Sunil Mushrana1912822010-01-21 10:50:03 -08001496update_holders:
Mark Fashehccd979b2005-12-15 14:31:24 -08001497 /* Ok, if we get here then we're good to go. */
1498 ocfs2_inc_holders(lockres, level);
1499
1500 ret = 0;
1501unlock:
Sunil Mushrana1912822010-01-21 10:50:03 -08001502 lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
1503
Mark Fashehccd979b2005-12-15 14:31:24 -08001504 spin_unlock_irqrestore(&lockres->l_lock, flags);
1505out:
1506 /*
1507 * This is helping work around a lock inversion between the page lock
1508 * and dlm locks. One path holds the page lock while calling aops
1509 * which block acquiring dlm locks. The voting thread holds dlm
1510 * locks while acquiring page locks while down converting data locks.
1511 * This block is helping an aop path notice the inversion and back
1512 * off to unlock its page lock before trying the dlm lock again.
1513 */
1514 if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1515 mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1516 wait = 0;
1517 if (lockres_remove_mask_waiter(lockres, &mw))
1518 ret = -EAGAIN;
1519 else
1520 goto again;
1521 }
1522 if (wait) {
1523 ret = ocfs2_wait_for_mask(&mw);
1524 if (ret == 0)
1525 goto again;
1526 mlog_errno(ret);
1527 }
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07001528 ocfs2_update_lock_stats(lockres, level, &mw, ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001529
Jan Karacb257972009-06-04 15:26:50 +02001530#ifdef CONFIG_DEBUG_LOCK_ALLOC
1531 if (!ret && lockres->l_lockdep_map.key != NULL) {
1532 if (level == DLM_LOCK_PR)
1533 rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
1534 !!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1535 caller_ip);
1536 else
1537 rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
1538 !!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
1539 caller_ip);
1540 }
1541#endif
Mark Fashehccd979b2005-12-15 14:31:24 -08001542 return ret;
1543}
1544
Jan Karacb257972009-06-04 15:26:50 +02001545static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
1546 struct ocfs2_lock_res *lockres,
1547 int level,
1548 u32 lkm_flags,
1549 int arg_flags)
1550{
1551 return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
1552 0, _RET_IP_);
1553}
1554
1555
1556static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
1557 struct ocfs2_lock_res *lockres,
1558 int level,
1559 unsigned long caller_ip)
Mark Fashehccd979b2005-12-15 14:31:24 -08001560{
1561 unsigned long flags;
1562
Mark Fashehccd979b2005-12-15 14:31:24 -08001563 spin_lock_irqsave(&lockres->l_lock, flags);
1564 ocfs2_dec_holders(lockres, level);
Mark Fasheh34d024f2007-09-24 15:56:19 -07001565 ocfs2_downconvert_on_unlock(osb, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001566 spin_unlock_irqrestore(&lockres->l_lock, flags);
Jan Karacb257972009-06-04 15:26:50 +02001567#ifdef CONFIG_DEBUG_LOCK_ALLOC
1568 if (lockres->l_lockdep_map.key != NULL)
1569 rwsem_release(&lockres->l_lockdep_map, 1, caller_ip);
1570#endif
Mark Fashehccd979b2005-12-15 14:31:24 -08001571}
1572
Adrian Bunkda661162006-11-20 03:24:28 +01001573static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1574 struct ocfs2_lock_res *lockres,
1575 int ex,
1576 int local)
Mark Fashehccd979b2005-12-15 14:31:24 -08001577{
Joel Beckerbd3e7612008-02-01 12:14:57 -08001578 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08001579 unsigned long flags;
Joel Beckerbd3e7612008-02-01 12:14:57 -08001580 u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001581
1582 spin_lock_irqsave(&lockres->l_lock, flags);
1583 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1584 lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1585 spin_unlock_irqrestore(&lockres->l_lock, flags);
1586
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001587 return ocfs2_lock_create(osb, lockres, level, lkm_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001588}
1589
1590/* Grants us an EX lock on the data and metadata resources, skipping
1591 * the normal cluster directory lookup. Use this ONLY on newly created
1592 * inodes which other nodes can't possibly see, and which haven't been
1593 * hashed in the inode hash yet. This can give us a good performance
1594 * increase as it'll skip the network broadcast normally associated
1595 * with creating a new lock resource. */
1596int ocfs2_create_new_inode_locks(struct inode *inode)
1597{
1598 int ret;
Mark Fashehd680efe2006-09-08 14:14:34 -07001599 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001600
1601 BUG_ON(!inode);
1602 BUG_ON(!ocfs2_inode_is_new(inode));
1603
Mark Fashehb0697052006-03-03 10:24:33 -08001604 mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001605
1606 /* NOTE: That we don't increment any of the holder counts, nor
1607 * do we add anything to a journal handle. Since this is
1608 * supposed to be a new inode which the cluster doesn't know
1609 * about yet, there is no need to. As far as the LVB handling
1610 * is concerned, this is basically like acquiring an EX lock
1611 * on a resource which has an invalid one -- we'll set it
1612 * valid when we release the EX. */
1613
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001614 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001615 if (ret) {
1616 mlog_errno(ret);
1617 goto bail;
1618 }
1619
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001620 /*
Joel Beckerbd3e7612008-02-01 12:14:57 -08001621 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001622 * don't use a generation in their lock names.
1623 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001624 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001625 if (ret) {
1626 mlog_errno(ret);
1627 goto bail;
1628 }
1629
Tiger Yang50008632007-03-20 16:01:38 -07001630 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1631 if (ret) {
1632 mlog_errno(ret);
1633 goto bail;
1634 }
1635
Mark Fashehccd979b2005-12-15 14:31:24 -08001636bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001637 return ret;
1638}
1639
1640int ocfs2_rw_lock(struct inode *inode, int write)
1641{
1642 int status, level;
1643 struct ocfs2_lock_res *lockres;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001644 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001645
1646 BUG_ON(!inode);
1647
Mark Fashehb0697052006-03-03 10:24:33 -08001648 mlog(0, "inode %llu take %s RW lock\n",
1649 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08001650 write ? "EXMODE" : "PRMODE");
1651
Tao Mac1e8d352011-03-07 16:43:21 +08001652 if (ocfs2_mount_local(osb))
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001653 return 0;
1654
Mark Fashehccd979b2005-12-15 14:31:24 -08001655 lockres = &OCFS2_I(inode)->ip_rw_lockres;
1656
Joel Beckerbd3e7612008-02-01 12:14:57 -08001657 level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08001658
1659 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1660 0);
1661 if (status < 0)
1662 mlog_errno(status);
1663
Mark Fashehccd979b2005-12-15 14:31:24 -08001664 return status;
1665}
1666
1667void ocfs2_rw_unlock(struct inode *inode, int write)
1668{
Joel Beckerbd3e7612008-02-01 12:14:57 -08001669 int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08001670 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001671 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001672
Mark Fashehb0697052006-03-03 10:24:33 -08001673 mlog(0, "inode %llu drop %s RW lock\n",
1674 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08001675 write ? "EXMODE" : "PRMODE");
1676
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001677 if (!ocfs2_mount_local(osb))
1678 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001679}
1680
Tiger Yang50008632007-03-20 16:01:38 -07001681/*
1682 * ocfs2_open_lock always get PR mode lock.
1683 */
1684int ocfs2_open_lock(struct inode *inode)
1685{
1686 int status = 0;
1687 struct ocfs2_lock_res *lockres;
1688 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1689
1690 BUG_ON(!inode);
1691
Tiger Yang50008632007-03-20 16:01:38 -07001692 mlog(0, "inode %llu take PRMODE open lock\n",
1693 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1694
Tiger Yang03efed82011-05-28 00:34:19 +08001695 if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
Tiger Yang50008632007-03-20 16:01:38 -07001696 goto out;
1697
1698 lockres = &OCFS2_I(inode)->ip_open_lockres;
1699
1700 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001701 DLM_LOCK_PR, 0, 0);
Tiger Yang50008632007-03-20 16:01:38 -07001702 if (status < 0)
1703 mlog_errno(status);
1704
1705out:
Tiger Yang50008632007-03-20 16:01:38 -07001706 return status;
1707}
1708
1709int ocfs2_try_open_lock(struct inode *inode, int write)
1710{
1711 int status = 0, level;
1712 struct ocfs2_lock_res *lockres;
1713 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1714
1715 BUG_ON(!inode);
1716
Tiger Yang50008632007-03-20 16:01:38 -07001717 mlog(0, "inode %llu try to take %s open lock\n",
1718 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1719 write ? "EXMODE" : "PRMODE");
1720
Tiger Yang03efed82011-05-28 00:34:19 +08001721 if (ocfs2_is_hard_readonly(osb)) {
1722 if (write)
1723 status = -EROFS;
1724 goto out;
1725 }
1726
Tiger Yang50008632007-03-20 16:01:38 -07001727 if (ocfs2_mount_local(osb))
1728 goto out;
1729
1730 lockres = &OCFS2_I(inode)->ip_open_lockres;
1731
Joel Beckerbd3e7612008-02-01 12:14:57 -08001732 level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
Tiger Yang50008632007-03-20 16:01:38 -07001733
1734 /*
1735 * The file system may already holding a PRMODE/EXMODE open lock.
Joel Beckerbd3e7612008-02-01 12:14:57 -08001736 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
Tiger Yang50008632007-03-20 16:01:38 -07001737 * other nodes and the -EAGAIN will indicate to the caller that
1738 * this inode is still in use.
1739 */
1740 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001741 level, DLM_LKF_NOQUEUE, 0);
Tiger Yang50008632007-03-20 16:01:38 -07001742
1743out:
Tiger Yang50008632007-03-20 16:01:38 -07001744 return status;
1745}
1746
1747/*
1748 * ocfs2_open_unlock unlock PR and EX mode open locks.
1749 */
1750void ocfs2_open_unlock(struct inode *inode)
1751{
1752 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1753 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1754
Tiger Yang50008632007-03-20 16:01:38 -07001755 mlog(0, "inode %llu drop open lock\n",
1756 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1757
1758 if (ocfs2_mount_local(osb))
1759 goto out;
1760
1761 if(lockres->l_ro_holders)
1762 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001763 DLM_LOCK_PR);
Tiger Yang50008632007-03-20 16:01:38 -07001764 if(lockres->l_ex_holders)
1765 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001766 DLM_LOCK_EX);
Tiger Yang50008632007-03-20 16:01:38 -07001767
1768out:
Tao Mac1e8d352011-03-07 16:43:21 +08001769 return;
Tiger Yang50008632007-03-20 16:01:38 -07001770}
1771
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001772static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1773 int level)
1774{
1775 int ret;
1776 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1777 unsigned long flags;
1778 struct ocfs2_mask_waiter mw;
1779
1780 ocfs2_init_mask_waiter(&mw);
1781
1782retry_cancel:
1783 spin_lock_irqsave(&lockres->l_lock, flags);
1784 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1785 ret = ocfs2_prepare_cancel_convert(osb, lockres);
1786 if (ret) {
1787 spin_unlock_irqrestore(&lockres->l_lock, flags);
1788 ret = ocfs2_cancel_convert(osb, lockres);
1789 if (ret < 0) {
1790 mlog_errno(ret);
1791 goto out;
1792 }
1793 goto retry_cancel;
1794 }
1795 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1796 spin_unlock_irqrestore(&lockres->l_lock, flags);
1797
1798 ocfs2_wait_for_mask(&mw);
1799 goto retry_cancel;
1800 }
1801
1802 ret = -ERESTARTSYS;
1803 /*
1804 * We may still have gotten the lock, in which case there's no
1805 * point to restarting the syscall.
1806 */
1807 if (lockres->l_level == level)
1808 ret = 0;
1809
1810 mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1811 lockres->l_flags, lockres->l_level, lockres->l_action);
1812
1813 spin_unlock_irqrestore(&lockres->l_lock, flags);
1814
1815out:
1816 return ret;
1817}
1818
1819/*
1820 * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1821 * flock() calls. The locking approach this requires is sufficiently
1822 * different from all other cluster lock types that we implement a
Daniel Mack3ad2f3f2010-02-03 08:01:28 +08001823 * separate path to the "low-level" dlm calls. In particular:
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001824 *
1825 * - No optimization of lock levels is done - we take at exactly
1826 * what's been requested.
1827 *
1828 * - No lock caching is employed. We immediately downconvert to
1829 * no-lock at unlock time. This also means flock locks never go on
1830 * the blocking list).
1831 *
1832 * - Since userspace can trivially deadlock itself with flock, we make
1833 * sure to allow cancellation of a misbehaving applications flock()
1834 * request.
1835 *
1836 * - Access to any flock lockres doesn't require concurrency, so we
1837 * can simplify the code by requiring the caller to guarantee
1838 * serialization of dlmglue flock calls.
1839 */
1840int ocfs2_file_lock(struct file *file, int ex, int trylock)
1841{
Mark Fashehe988cf12008-07-10 09:25:39 -07001842 int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1843 unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001844 unsigned long flags;
1845 struct ocfs2_file_private *fp = file->private_data;
1846 struct ocfs2_lock_res *lockres = &fp->fp_flock;
1847 struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1848 struct ocfs2_mask_waiter mw;
1849
1850 ocfs2_init_mask_waiter(&mw);
1851
1852 if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
Joel Beckerbd3e7612008-02-01 12:14:57 -08001853 (lockres->l_level > DLM_LOCK_NL)) {
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001854 mlog(ML_ERROR,
1855 "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1856 "level: %u\n", lockres->l_name, lockres->l_flags,
1857 lockres->l_level);
1858 return -EINVAL;
1859 }
1860
1861 spin_lock_irqsave(&lockres->l_lock, flags);
1862 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1863 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1864 spin_unlock_irqrestore(&lockres->l_lock, flags);
1865
1866 /*
1867 * Get the lock at NLMODE to start - that way we
1868 * can cancel the upconvert request if need be.
1869 */
Mark Fashehe988cf12008-07-10 09:25:39 -07001870 ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001871 if (ret < 0) {
1872 mlog_errno(ret);
1873 goto out;
1874 }
1875
1876 ret = ocfs2_wait_for_mask(&mw);
1877 if (ret) {
1878 mlog_errno(ret);
1879 goto out;
1880 }
1881 spin_lock_irqsave(&lockres->l_lock, flags);
1882 }
1883
1884 lockres->l_action = OCFS2_AST_CONVERT;
Mark Fashehe988cf12008-07-10 09:25:39 -07001885 lkm_flags |= DLM_LKF_CONVERT;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001886 lockres->l_requested = level;
1887 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1888
1889 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1890 spin_unlock_irqrestore(&lockres->l_lock, flags);
1891
Joel Becker4670c462008-02-01 14:39:35 -08001892 ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
Joel Beckera796d282010-01-28 19:22:39 -08001893 lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1);
Joel Becker7431cd72008-02-01 12:15:37 -08001894 if (ret) {
1895 if (!trylock || (ret != -EAGAIN)) {
Joel Becker24ef1812008-01-29 17:37:32 -08001896 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001897 ret = -EINVAL;
1898 }
1899
1900 ocfs2_recover_from_dlm_error(lockres, 1);
1901 lockres_remove_mask_waiter(lockres, &mw);
1902 goto out;
1903 }
1904
1905 ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
1906 if (ret == -ERESTARTSYS) {
1907 /*
1908 * Userspace can cause deadlock itself with
1909 * flock(). Current behavior locally is to allow the
1910 * deadlock, but abort the system call if a signal is
1911 * received. We follow this example, otherwise a
1912 * poorly written program could sit in kernel until
1913 * reboot.
1914 *
1915 * Handling this is a bit more complicated for Ocfs2
1916 * though. We can't exit this function with an
1917 * outstanding lock request, so a cancel convert is
1918 * required. We intentionally overwrite 'ret' - if the
1919 * cancel fails and the lock was granted, it's easier
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001920 * to just bubble success back up to the user.
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001921 */
1922 ret = ocfs2_flock_handle_signal(lockres, level);
David Teigland1693a5c2008-01-30 16:52:53 -08001923 } else if (!ret && (level > lockres->l_level)) {
1924 /* Trylock failed asynchronously */
1925 BUG_ON(!trylock);
1926 ret = -EAGAIN;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001927 }
1928
1929out:
1930
1931 mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
1932 lockres->l_name, ex, trylock, ret);
1933 return ret;
1934}
1935
1936void ocfs2_file_unlock(struct file *file)
1937{
1938 int ret;
Joel Beckerde551242008-02-01 14:45:08 -08001939 unsigned int gen;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001940 unsigned long flags;
1941 struct ocfs2_file_private *fp = file->private_data;
1942 struct ocfs2_lock_res *lockres = &fp->fp_flock;
1943 struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1944 struct ocfs2_mask_waiter mw;
1945
1946 ocfs2_init_mask_waiter(&mw);
1947
1948 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
1949 return;
1950
Mark Fashehe988cf12008-07-10 09:25:39 -07001951 if (lockres->l_level == DLM_LOCK_NL)
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001952 return;
1953
1954 mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
1955 lockres->l_name, lockres->l_flags, lockres->l_level,
1956 lockres->l_action);
1957
1958 spin_lock_irqsave(&lockres->l_lock, flags);
1959 /*
1960 * Fake a blocking ast for the downconvert code.
1961 */
1962 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
Joel Beckerbd3e7612008-02-01 12:14:57 -08001963 lockres->l_blocking = DLM_LOCK_EX;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001964
Mark Fashehe988cf12008-07-10 09:25:39 -07001965 gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001966 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1967 spin_unlock_irqrestore(&lockres->l_lock, flags);
1968
Mark Fashehe988cf12008-07-10 09:25:39 -07001969 ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001970 if (ret) {
1971 mlog_errno(ret);
1972 return;
1973 }
1974
1975 ret = ocfs2_wait_for_mask(&mw);
1976 if (ret)
1977 mlog_errno(ret);
1978}
1979
Mark Fasheh34d024f2007-09-24 15:56:19 -07001980static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
1981 struct ocfs2_lock_res *lockres)
Mark Fashehccd979b2005-12-15 14:31:24 -08001982{
1983 int kick = 0;
1984
Mark Fashehccd979b2005-12-15 14:31:24 -08001985 /* If we know that another node is waiting on our lock, kick
Mark Fasheh34d024f2007-09-24 15:56:19 -07001986 * the downconvert thread * pre-emptively when we reach a release
Mark Fashehccd979b2005-12-15 14:31:24 -08001987 * condition. */
1988 if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1989 switch(lockres->l_blocking) {
Joel Beckerbd3e7612008-02-01 12:14:57 -08001990 case DLM_LOCK_EX:
Mark Fashehccd979b2005-12-15 14:31:24 -08001991 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1992 kick = 1;
1993 break;
Joel Beckerbd3e7612008-02-01 12:14:57 -08001994 case DLM_LOCK_PR:
Mark Fashehccd979b2005-12-15 14:31:24 -08001995 if (!lockres->l_ex_holders)
1996 kick = 1;
1997 break;
1998 default:
1999 BUG();
2000 }
2001 }
2002
2003 if (kick)
Mark Fasheh34d024f2007-09-24 15:56:19 -07002004 ocfs2_wake_downconvert_thread(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002005}
2006
Mark Fashehccd979b2005-12-15 14:31:24 -08002007#define OCFS2_SEC_BITS 34
2008#define OCFS2_SEC_SHIFT (64 - 34)
2009#define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1)
2010
2011/* LVB only has room for 64 bits of time here so we pack it for
2012 * now. */
2013static u64 ocfs2_pack_timespec(struct timespec *spec)
2014{
2015 u64 res;
2016 u64 sec = spec->tv_sec;
2017 u32 nsec = spec->tv_nsec;
2018
2019 res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
2020
2021 return res;
2022}
2023
2024/* Call this with the lockres locked. I am reasonably sure we don't
2025 * need ip_lock in this function as anyone who would be changing those
Mark Fashehe63aecb62007-10-18 15:30:42 -07002026 * values is supposed to be blocked in ocfs2_inode_lock right now. */
Mark Fashehccd979b2005-12-15 14:31:24 -08002027static void __ocfs2_stuff_meta_lvb(struct inode *inode)
2028{
2029 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002030 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
Mark Fashehccd979b2005-12-15 14:31:24 -08002031 struct ocfs2_meta_lvb *lvb;
2032
Mark Fasheha641dc22008-12-24 16:03:48 -08002033 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002034
Mark Fasheh24c19ef2006-09-22 17:28:19 -07002035 /*
2036 * Invalidate the LVB of a deleted inode - this way other
2037 * nodes are forced to go to disk and discover the new inode
2038 * status.
2039 */
2040 if (oi->ip_flags & OCFS2_INODE_DELETED) {
2041 lvb->lvb_version = 0;
2042 goto out;
2043 }
2044
Mark Fasheh4d3b83f2006-09-12 15:22:18 -07002045 lvb->lvb_version = OCFS2_LVB_VERSION;
Mark Fashehccd979b2005-12-15 14:31:24 -08002046 lvb->lvb_isize = cpu_to_be64(i_size_read(inode));
2047 lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
2048 lvb->lvb_iuid = cpu_to_be32(inode->i_uid);
2049 lvb->lvb_igid = cpu_to_be32(inode->i_gid);
2050 lvb->lvb_imode = cpu_to_be16(inode->i_mode);
2051 lvb->lvb_inlink = cpu_to_be16(inode->i_nlink);
2052 lvb->lvb_iatime_packed =
2053 cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
2054 lvb->lvb_ictime_packed =
2055 cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
2056 lvb->lvb_imtime_packed =
2057 cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
Herbert Poetzlca4d1472006-07-03 17:27:12 -07002058 lvb->lvb_iattr = cpu_to_be32(oi->ip_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07002059 lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
Mark Fashehf9e2d822006-09-12 15:35:49 -07002060 lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
Mark Fashehccd979b2005-12-15 14:31:24 -08002061
Mark Fasheh24c19ef2006-09-22 17:28:19 -07002062out:
Mark Fashehccd979b2005-12-15 14:31:24 -08002063 mlog_meta_lvb(0, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08002064}
2065
2066static void ocfs2_unpack_timespec(struct timespec *spec,
2067 u64 packed_time)
2068{
2069 spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
2070 spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
2071}
2072
2073static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
2074{
2075 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002076 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
Mark Fashehccd979b2005-12-15 14:31:24 -08002077 struct ocfs2_meta_lvb *lvb;
2078
Mark Fashehccd979b2005-12-15 14:31:24 -08002079 mlog_meta_lvb(0, lockres);
2080
Mark Fasheha641dc22008-12-24 16:03:48 -08002081 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002082
2083 /* We're safe here without the lockres lock... */
2084 spin_lock(&oi->ip_lock);
2085 oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
2086 i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
2087
Herbert Poetzlca4d1472006-07-03 17:27:12 -07002088 oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07002089 oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07002090 ocfs2_set_inode_flags(inode);
2091
Mark Fashehccd979b2005-12-15 14:31:24 -08002092 /* fast-symlinks are a special case */
2093 if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
2094 inode->i_blocks = 0;
2095 else
Mark Fasheh8110b072007-03-22 16:53:23 -07002096 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08002097
2098 inode->i_uid = be32_to_cpu(lvb->lvb_iuid);
2099 inode->i_gid = be32_to_cpu(lvb->lvb_igid);
2100 inode->i_mode = be16_to_cpu(lvb->lvb_imode);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002101 set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
Mark Fashehccd979b2005-12-15 14:31:24 -08002102 ocfs2_unpack_timespec(&inode->i_atime,
2103 be64_to_cpu(lvb->lvb_iatime_packed));
2104 ocfs2_unpack_timespec(&inode->i_mtime,
2105 be64_to_cpu(lvb->lvb_imtime_packed));
2106 ocfs2_unpack_timespec(&inode->i_ctime,
2107 be64_to_cpu(lvb->lvb_ictime_packed));
2108 spin_unlock(&oi->ip_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08002109}
2110
Mark Fashehf9e2d822006-09-12 15:35:49 -07002111static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
2112 struct ocfs2_lock_res *lockres)
Mark Fashehccd979b2005-12-15 14:31:24 -08002113{
Mark Fasheha641dc22008-12-24 16:03:48 -08002114 struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002115
Joel Becker1c520df2009-06-19 15:14:13 -07002116 if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
2117 && lvb->lvb_version == OCFS2_LVB_VERSION
Mark Fashehf9e2d822006-09-12 15:35:49 -07002118 && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
Mark Fashehccd979b2005-12-15 14:31:24 -08002119 return 1;
2120 return 0;
2121}
2122
2123/* Determine whether a lock resource needs to be refreshed, and
2124 * arbitrate who gets to refresh it.
2125 *
2126 * 0 means no refresh needed.
2127 *
2128 * > 0 means you need to refresh this and you MUST call
2129 * ocfs2_complete_lock_res_refresh afterwards. */
2130static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
2131{
2132 unsigned long flags;
2133 int status = 0;
2134
Mark Fashehccd979b2005-12-15 14:31:24 -08002135refresh_check:
2136 spin_lock_irqsave(&lockres->l_lock, flags);
2137 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
2138 spin_unlock_irqrestore(&lockres->l_lock, flags);
2139 goto bail;
2140 }
2141
2142 if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
2143 spin_unlock_irqrestore(&lockres->l_lock, flags);
2144
2145 ocfs2_wait_on_refreshing_lock(lockres);
2146 goto refresh_check;
2147 }
2148
2149 /* Ok, I'll be the one to refresh this lock. */
2150 lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
2151 spin_unlock_irqrestore(&lockres->l_lock, flags);
2152
2153 status = 1;
2154bail:
Tao Mac1e8d352011-03-07 16:43:21 +08002155 mlog(0, "status %d\n", status);
Mark Fashehccd979b2005-12-15 14:31:24 -08002156 return status;
2157}
2158
2159/* If status is non zero, I'll mark it as not being in refresh
2160 * anymroe, but i won't clear the needs refresh flag. */
2161static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2162 int status)
2163{
2164 unsigned long flags;
Mark Fashehccd979b2005-12-15 14:31:24 -08002165
2166 spin_lock_irqsave(&lockres->l_lock, flags);
2167 lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2168 if (!status)
2169 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2170 spin_unlock_irqrestore(&lockres->l_lock, flags);
2171
2172 wake_up(&lockres->l_event);
Mark Fashehccd979b2005-12-15 14:31:24 -08002173}
2174
2175/* may or may not return a bh if it went to disk. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002176static int ocfs2_inode_lock_update(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -08002177 struct buffer_head **bh)
2178{
2179 int status = 0;
2180 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002181 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
Mark Fashehccd979b2005-12-15 14:31:24 -08002182 struct ocfs2_dinode *fe;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002183 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002184
Mark Fashehbe9e9862007-04-18 15:22:08 -07002185 if (ocfs2_mount_local(osb))
2186 goto bail;
2187
Mark Fashehccd979b2005-12-15 14:31:24 -08002188 spin_lock(&oi->ip_lock);
2189 if (oi->ip_flags & OCFS2_INODE_DELETED) {
Mark Fashehb0697052006-03-03 10:24:33 -08002190 mlog(0, "Orphaned inode %llu was deleted while we "
Mark Fashehccd979b2005-12-15 14:31:24 -08002191 "were waiting on a lock. ip_flags = 0x%x\n",
Mark Fashehb0697052006-03-03 10:24:33 -08002192 (unsigned long long)oi->ip_blkno, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08002193 spin_unlock(&oi->ip_lock);
2194 status = -ENOENT;
2195 goto bail;
2196 }
2197 spin_unlock(&oi->ip_lock);
2198
Mark Fashehbe9e9862007-04-18 15:22:08 -07002199 if (!ocfs2_should_refresh_lock_res(lockres))
2200 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08002201
2202 /* This will discard any caching information we might have had
2203 * for the inode metadata. */
Joel Becker8cb471e2009-02-10 20:00:41 -08002204 ocfs2_metadata_cache_purge(INODE_CACHE(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08002205
Mark Fasheh83418972007-04-23 18:53:12 -07002206 ocfs2_extent_map_trunc(inode, 0);
2207
Mark Fashehbe9e9862007-04-18 15:22:08 -07002208 if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
Mark Fashehb0697052006-03-03 10:24:33 -08002209 mlog(0, "Trusting LVB on inode %llu\n",
2210 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08002211 ocfs2_refresh_inode_from_lvb(inode);
2212 } else {
2213 /* Boo, we have to go to disk. */
2214 /* read bh, cast, ocfs2_refresh_inode */
Joel Beckerb657c952008-11-13 14:49:11 -08002215 status = ocfs2_read_inode_block(inode, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002216 if (status < 0) {
2217 mlog_errno(status);
2218 goto bail_refresh;
2219 }
2220 fe = (struct ocfs2_dinode *) (*bh)->b_data;
2221
2222 /* This is a good chance to make sure we're not
Joel Beckerb657c952008-11-13 14:49:11 -08002223 * locking an invalid object. ocfs2_read_inode_block()
2224 * already checked that the inode block is sane.
Mark Fashehccd979b2005-12-15 14:31:24 -08002225 *
2226 * We bug on a stale inode here because we checked
2227 * above whether it was wiped from disk. The wiping
2228 * node provides a guarantee that we receive that
2229 * message and can mark the inode before dropping any
2230 * locks associated with it. */
Mark Fashehccd979b2005-12-15 14:31:24 -08002231 mlog_bug_on_msg(inode->i_generation !=
2232 le32_to_cpu(fe->i_generation),
Mark Fashehb0697052006-03-03 10:24:33 -08002233 "Invalid dinode %llu disk generation: %u "
Mark Fashehccd979b2005-12-15 14:31:24 -08002234 "inode->i_generation: %u\n",
Mark Fashehb0697052006-03-03 10:24:33 -08002235 (unsigned long long)oi->ip_blkno,
2236 le32_to_cpu(fe->i_generation),
Mark Fashehccd979b2005-12-15 14:31:24 -08002237 inode->i_generation);
2238 mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2239 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
Mark Fashehb0697052006-03-03 10:24:33 -08002240 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
2241 (unsigned long long)oi->ip_blkno,
2242 (unsigned long long)le64_to_cpu(fe->i_dtime),
Mark Fashehccd979b2005-12-15 14:31:24 -08002243 le32_to_cpu(fe->i_flags));
2244
2245 ocfs2_refresh_inode(inode, fe);
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002246 ocfs2_track_lock_refresh(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08002247 }
2248
2249 status = 0;
2250bail_refresh:
Mark Fashehbe9e9862007-04-18 15:22:08 -07002251 ocfs2_complete_lock_res_refresh(lockres, status);
Mark Fashehccd979b2005-12-15 14:31:24 -08002252bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08002253 return status;
2254}
2255
2256static int ocfs2_assign_bh(struct inode *inode,
2257 struct buffer_head **ret_bh,
2258 struct buffer_head *passed_bh)
2259{
2260 int status;
2261
2262 if (passed_bh) {
2263 /* Ok, the update went to disk for us, use the
2264 * returned bh. */
2265 *ret_bh = passed_bh;
2266 get_bh(*ret_bh);
2267
2268 return 0;
2269 }
2270
Joel Beckerb657c952008-11-13 14:49:11 -08002271 status = ocfs2_read_inode_block(inode, ret_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002272 if (status < 0)
2273 mlog_errno(status);
2274
2275 return status;
2276}
2277
2278/*
2279 * returns < 0 error if the callback will never be called, otherwise
2280 * the result of the lock will be communicated via the callback.
2281 */
Jan Karacb257972009-06-04 15:26:50 +02002282int ocfs2_inode_lock_full_nested(struct inode *inode,
2283 struct buffer_head **ret_bh,
2284 int ex,
2285 int arg_flags,
2286 int subclass)
Mark Fashehccd979b2005-12-15 14:31:24 -08002287{
Joel Beckerbd3e7612008-02-01 12:14:57 -08002288 int status, level, acquired;
2289 u32 dlm_flags;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002290 struct ocfs2_lock_res *lockres = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002291 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2292 struct buffer_head *local_bh = NULL;
2293
2294 BUG_ON(!inode);
2295
Mark Fashehb0697052006-03-03 10:24:33 -08002296 mlog(0, "inode %llu, take %s META lock\n",
2297 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08002298 ex ? "EXMODE" : "PRMODE");
2299
2300 status = 0;
2301 acquired = 0;
2302 /* We'll allow faking a readonly metadata lock for
2303 * rodevices. */
2304 if (ocfs2_is_hard_readonly(osb)) {
2305 if (ex)
2306 status = -EROFS;
Tiger Yang03efed82011-05-28 00:34:19 +08002307 goto getbh;
Mark Fashehccd979b2005-12-15 14:31:24 -08002308 }
2309
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002310 if (ocfs2_mount_local(osb))
2311 goto local;
2312
Mark Fashehccd979b2005-12-15 14:31:24 -08002313 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
Joel Becker553abd02008-02-01 12:03:57 -08002314 ocfs2_wait_for_recovery(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002315
Mark Fashehe63aecb62007-10-18 15:30:42 -07002316 lockres = &OCFS2_I(inode)->ip_inode_lockres;
Joel Beckerbd3e7612008-02-01 12:14:57 -08002317 level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08002318 dlm_flags = 0;
2319 if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
Joel Beckerbd3e7612008-02-01 12:14:57 -08002320 dlm_flags |= DLM_LKF_NOQUEUE;
Mark Fashehccd979b2005-12-15 14:31:24 -08002321
Jan Karacb257972009-06-04 15:26:50 +02002322 status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
2323 arg_flags, subclass, _RET_IP_);
Mark Fashehccd979b2005-12-15 14:31:24 -08002324 if (status < 0) {
2325 if (status != -EAGAIN && status != -EIOCBRETRY)
2326 mlog_errno(status);
2327 goto bail;
2328 }
2329
2330 /* Notify the error cleanup path to drop the cluster lock. */
2331 acquired = 1;
2332
2333 /* We wait twice because a node may have died while we were in
2334 * the lower dlm layers. The second time though, we've
2335 * committed to owning this lock so we don't allow signals to
2336 * abort the operation. */
2337 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
Joel Becker553abd02008-02-01 12:03:57 -08002338 ocfs2_wait_for_recovery(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002339
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002340local:
Mark Fasheh24c19ef2006-09-22 17:28:19 -07002341 /*
2342 * We only see this flag if we're being called from
2343 * ocfs2_read_locked_inode(). It means we're locking an inode
2344 * which hasn't been populated yet, so clear the refresh flag
2345 * and let the caller handle it.
2346 */
2347 if (inode->i_state & I_NEW) {
2348 status = 0;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002349 if (lockres)
2350 ocfs2_complete_lock_res_refresh(lockres, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -07002351 goto bail;
2352 }
2353
Mark Fashehccd979b2005-12-15 14:31:24 -08002354 /* This is fun. The caller may want a bh back, or it may
Mark Fashehe63aecb62007-10-18 15:30:42 -07002355 * not. ocfs2_inode_lock_update definitely wants one in, but
Mark Fashehccd979b2005-12-15 14:31:24 -08002356 * may or may not read one, depending on what's in the
2357 * LVB. The result of all of this is that we've *only* gone to
2358 * disk if we have to, so the complexity is worthwhile. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002359 status = ocfs2_inode_lock_update(inode, &local_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002360 if (status < 0) {
2361 if (status != -ENOENT)
2362 mlog_errno(status);
2363 goto bail;
2364 }
Tiger Yang03efed82011-05-28 00:34:19 +08002365getbh:
Mark Fashehccd979b2005-12-15 14:31:24 -08002366 if (ret_bh) {
2367 status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2368 if (status < 0) {
2369 mlog_errno(status);
2370 goto bail;
2371 }
2372 }
2373
Mark Fashehccd979b2005-12-15 14:31:24 -08002374bail:
2375 if (status < 0) {
2376 if (ret_bh && (*ret_bh)) {
2377 brelse(*ret_bh);
2378 *ret_bh = NULL;
2379 }
2380 if (acquired)
Mark Fashehe63aecb62007-10-18 15:30:42 -07002381 ocfs2_inode_unlock(inode, ex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002382 }
2383
2384 if (local_bh)
2385 brelse(local_bh);
2386
Mark Fashehccd979b2005-12-15 14:31:24 -08002387 return status;
2388}
2389
2390/*
Mark Fasheh34d024f2007-09-24 15:56:19 -07002391 * This is working around a lock inversion between tasks acquiring DLM
2392 * locks while holding a page lock and the downconvert thread which
2393 * blocks dlm lock acquiry while acquiring page locks.
Mark Fashehccd979b2005-12-15 14:31:24 -08002394 *
2395 * ** These _with_page variantes are only intended to be called from aop
2396 * methods that hold page locks and return a very specific *positive* error
2397 * code that aop methods pass up to the VFS -- test for errors with != 0. **
2398 *
Mark Fasheh34d024f2007-09-24 15:56:19 -07002399 * The DLM is called such that it returns -EAGAIN if it would have
2400 * blocked waiting for the downconvert thread. In that case we unlock
2401 * our page so the downconvert thread can make progress. Once we've
2402 * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2403 * that called us can bubble that back up into the VFS who will then
2404 * immediately retry the aop call.
Mark Fashehccd979b2005-12-15 14:31:24 -08002405 *
2406 * We do a blocking lock and immediate unlock before returning, though, so that
2407 * the lock has a great chance of being cached on this node by the time the VFS
2408 * calls back to retry the aop. This has a potential to livelock as nodes
2409 * ping locks back and forth, but that's a risk we're willing to take to avoid
2410 * the lock inversion simply.
2411 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002412int ocfs2_inode_lock_with_page(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -08002413 struct buffer_head **ret_bh,
2414 int ex,
2415 struct page *page)
2416{
2417 int ret;
2418
Mark Fashehe63aecb62007-10-18 15:30:42 -07002419 ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
Mark Fashehccd979b2005-12-15 14:31:24 -08002420 if (ret == -EAGAIN) {
2421 unlock_page(page);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002422 if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2423 ocfs2_inode_unlock(inode, ex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002424 ret = AOP_TRUNCATED_PAGE;
2425 }
2426
2427 return ret;
2428}
2429
Mark Fashehe63aecb62007-10-18 15:30:42 -07002430int ocfs2_inode_lock_atime(struct inode *inode,
Tiger Yang7f1a37e2006-11-15 15:48:42 +08002431 struct vfsmount *vfsmnt,
2432 int *level)
2433{
2434 int ret;
2435
Mark Fashehe63aecb62007-10-18 15:30:42 -07002436 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yang7f1a37e2006-11-15 15:48:42 +08002437 if (ret < 0) {
2438 mlog_errno(ret);
2439 return ret;
2440 }
2441
2442 /*
2443 * If we should update atime, we will get EX lock,
2444 * otherwise we just get PR lock.
2445 */
2446 if (ocfs2_should_update_atime(inode, vfsmnt)) {
2447 struct buffer_head *bh = NULL;
2448
Mark Fashehe63aecb62007-10-18 15:30:42 -07002449 ocfs2_inode_unlock(inode, 0);
2450 ret = ocfs2_inode_lock(inode, &bh, 1);
Tiger Yang7f1a37e2006-11-15 15:48:42 +08002451 if (ret < 0) {
2452 mlog_errno(ret);
2453 return ret;
2454 }
2455 *level = 1;
2456 if (ocfs2_should_update_atime(inode, vfsmnt))
2457 ocfs2_update_inode_atime(inode, bh);
2458 if (bh)
2459 brelse(bh);
2460 } else
2461 *level = 0;
2462
Tiger Yang7f1a37e2006-11-15 15:48:42 +08002463 return ret;
2464}
2465
Mark Fashehe63aecb62007-10-18 15:30:42 -07002466void ocfs2_inode_unlock(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -08002467 int ex)
2468{
Joel Beckerbd3e7612008-02-01 12:14:57 -08002469 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehe63aecb62007-10-18 15:30:42 -07002470 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002471 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002472
Mark Fashehb0697052006-03-03 10:24:33 -08002473 mlog(0, "inode %llu drop %s META lock\n",
2474 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08002475 ex ? "EXMODE" : "PRMODE");
2476
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002477 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
2478 !ocfs2_mount_local(osb))
Mark Fashehccd979b2005-12-15 14:31:24 -08002479 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002480}
2481
Sunil Mushrandf152c22009-06-22 11:40:07 -07002482int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
Srinivas Eeda83273932009-06-03 17:02:55 -07002483{
2484 struct ocfs2_lock_res *lockres;
2485 struct ocfs2_orphan_scan_lvb *lvb;
Srinivas Eeda83273932009-06-03 17:02:55 -07002486 int status = 0;
2487
Sunil Mushrandf152c22009-06-22 11:40:07 -07002488 if (ocfs2_is_hard_readonly(osb))
2489 return -EROFS;
2490
2491 if (ocfs2_mount_local(osb))
2492 return 0;
2493
Srinivas Eeda83273932009-06-03 17:02:55 -07002494 lockres = &osb->osb_orphan_scan.os_lockres;
Sunil Mushrandf152c22009-06-22 11:40:07 -07002495 status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
Srinivas Eeda83273932009-06-03 17:02:55 -07002496 if (status < 0)
2497 return status;
2498
2499 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Joel Becker1c520df2009-06-19 15:14:13 -07002500 if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
2501 lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
Srinivas Eeda83273932009-06-03 17:02:55 -07002502 *seqno = be32_to_cpu(lvb->lvb_os_seqno);
Sunil Mushran32119492009-06-19 16:53:18 -07002503 else
2504 *seqno = osb->osb_orphan_scan.os_seqno + 1;
2505
Srinivas Eeda83273932009-06-03 17:02:55 -07002506 return status;
2507}
2508
Sunil Mushrandf152c22009-06-22 11:40:07 -07002509void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
Srinivas Eeda83273932009-06-03 17:02:55 -07002510{
2511 struct ocfs2_lock_res *lockres;
2512 struct ocfs2_orphan_scan_lvb *lvb;
Srinivas Eeda83273932009-06-03 17:02:55 -07002513
Sunil Mushrandf152c22009-06-22 11:40:07 -07002514 if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
2515 lockres = &osb->osb_orphan_scan.os_lockres;
2516 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
2517 lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
2518 lvb->lvb_os_seqno = cpu_to_be32(seqno);
2519 ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
2520 }
Srinivas Eeda83273932009-06-03 17:02:55 -07002521}
2522
Mark Fashehccd979b2005-12-15 14:31:24 -08002523int ocfs2_super_lock(struct ocfs2_super *osb,
2524 int ex)
2525{
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002526 int status = 0;
Joel Beckerbd3e7612008-02-01 12:14:57 -08002527 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08002528 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
Mark Fashehccd979b2005-12-15 14:31:24 -08002529
Mark Fashehccd979b2005-12-15 14:31:24 -08002530 if (ocfs2_is_hard_readonly(osb))
2531 return -EROFS;
2532
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002533 if (ocfs2_mount_local(osb))
2534 goto bail;
2535
Mark Fashehccd979b2005-12-15 14:31:24 -08002536 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2537 if (status < 0) {
2538 mlog_errno(status);
2539 goto bail;
2540 }
2541
2542 /* The super block lock path is really in the best position to
2543 * know when resources covered by the lock need to be
2544 * refreshed, so we do it here. Of course, making sense of
2545 * everything is up to the caller :) */
2546 status = ocfs2_should_refresh_lock_res(lockres);
2547 if (status < 0) {
2548 mlog_errno(status);
2549 goto bail;
2550 }
2551 if (status) {
Mark Fasheh8e8a4602008-02-01 11:59:09 -08002552 status = ocfs2_refresh_slot_info(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002553
2554 ocfs2_complete_lock_res_refresh(lockres, status);
2555
2556 if (status < 0)
2557 mlog_errno(status);
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002558 ocfs2_track_lock_refresh(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08002559 }
2560bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08002561 return status;
2562}
2563
2564void ocfs2_super_unlock(struct ocfs2_super *osb,
2565 int ex)
2566{
Joel Beckerbd3e7612008-02-01 12:14:57 -08002567 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08002568 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2569
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002570 if (!ocfs2_mount_local(osb))
2571 ocfs2_cluster_unlock(osb, lockres, level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002572}
2573
2574int ocfs2_rename_lock(struct ocfs2_super *osb)
2575{
2576 int status;
2577 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2578
2579 if (ocfs2_is_hard_readonly(osb))
2580 return -EROFS;
2581
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002582 if (ocfs2_mount_local(osb))
2583 return 0;
2584
Joel Beckerbd3e7612008-02-01 12:14:57 -08002585 status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08002586 if (status < 0)
2587 mlog_errno(status);
2588
2589 return status;
2590}
2591
2592void ocfs2_rename_unlock(struct ocfs2_super *osb)
2593{
2594 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2595
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002596 if (!ocfs2_mount_local(osb))
Joel Beckerbd3e7612008-02-01 12:14:57 -08002597 ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
Mark Fashehccd979b2005-12-15 14:31:24 -08002598}
2599
wengang wang6ca497a2009-03-06 21:29:10 +08002600int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
2601{
2602 int status;
2603 struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2604
2605 if (ocfs2_is_hard_readonly(osb))
2606 return -EROFS;
2607
2608 if (ocfs2_mount_local(osb))
2609 return 0;
2610
2611 status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
2612 0, 0);
2613 if (status < 0)
2614 mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
2615
2616 return status;
2617}
2618
2619void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
2620{
2621 struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
2622
2623 if (!ocfs2_mount_local(osb))
2624 ocfs2_cluster_unlock(osb, lockres,
2625 ex ? LKM_EXMODE : LKM_PRMODE);
2626}
2627
Mark Fashehd680efe2006-09-08 14:14:34 -07002628int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2629{
2630 int ret;
Joel Beckerbd3e7612008-02-01 12:14:57 -08002631 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehd680efe2006-09-08 14:14:34 -07002632 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2633 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2634
2635 BUG_ON(!dl);
2636
Tiger Yang03efed82011-05-28 00:34:19 +08002637 if (ocfs2_is_hard_readonly(osb)) {
2638 if (ex)
2639 return -EROFS;
2640 return 0;
2641 }
Mark Fashehd680efe2006-09-08 14:14:34 -07002642
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002643 if (ocfs2_mount_local(osb))
2644 return 0;
2645
Mark Fashehd680efe2006-09-08 14:14:34 -07002646 ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2647 if (ret < 0)
2648 mlog_errno(ret);
2649
2650 return ret;
2651}
2652
2653void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2654{
Joel Beckerbd3e7612008-02-01 12:14:57 -08002655 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehd680efe2006-09-08 14:14:34 -07002656 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2657 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2658
Tiger Yang03efed82011-05-28 00:34:19 +08002659 if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002660 ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
Mark Fashehd680efe2006-09-08 14:14:34 -07002661}
2662
Mark Fashehccd979b2005-12-15 14:31:24 -08002663/* Reference counting of the dlm debug structure. We want this because
2664 * open references on the debug inodes can live on after a mount, so
2665 * we can't rely on the ocfs2_super to always exist. */
2666static void ocfs2_dlm_debug_free(struct kref *kref)
2667{
2668 struct ocfs2_dlm_debug *dlm_debug;
2669
2670 dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2671
2672 kfree(dlm_debug);
2673}
2674
2675void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2676{
2677 if (dlm_debug)
2678 kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2679}
2680
2681static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2682{
2683 kref_get(&debug->d_refcnt);
2684}
2685
2686struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2687{
2688 struct ocfs2_dlm_debug *dlm_debug;
2689
2690 dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2691 if (!dlm_debug) {
2692 mlog_errno(-ENOMEM);
2693 goto out;
2694 }
2695
2696 kref_init(&dlm_debug->d_refcnt);
2697 INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2698 dlm_debug->d_locking_state = NULL;
2699out:
2700 return dlm_debug;
2701}
2702
2703/* Access to this is arbitrated for us via seq_file->sem. */
2704struct ocfs2_dlm_seq_priv {
2705 struct ocfs2_dlm_debug *p_dlm_debug;
2706 struct ocfs2_lock_res p_iter_res;
2707 struct ocfs2_lock_res p_tmp_res;
2708};
2709
2710static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2711 struct ocfs2_dlm_seq_priv *priv)
2712{
2713 struct ocfs2_lock_res *iter, *ret = NULL;
2714 struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2715
2716 assert_spin_locked(&ocfs2_dlm_tracking_lock);
2717
2718 list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2719 /* discover the head of the list */
2720 if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2721 mlog(0, "End of list found, %p\n", ret);
2722 break;
2723 }
2724
2725 /* We track our "dummy" iteration lockres' by a NULL
2726 * l_ops field. */
2727 if (iter->l_ops != NULL) {
2728 ret = iter;
2729 break;
2730 }
2731 }
2732
2733 return ret;
2734}
2735
2736static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2737{
2738 struct ocfs2_dlm_seq_priv *priv = m->private;
2739 struct ocfs2_lock_res *iter;
2740
2741 spin_lock(&ocfs2_dlm_tracking_lock);
2742 iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2743 if (iter) {
2744 /* Since lockres' have the lifetime of their container
2745 * (which can be inodes, ocfs2_supers, etc) we want to
2746 * copy this out to a temporary lockres while still
2747 * under the spinlock. Obviously after this we can't
2748 * trust any pointers on the copy returned, but that's
2749 * ok as the information we want isn't typically held
2750 * in them. */
2751 priv->p_tmp_res = *iter;
2752 iter = &priv->p_tmp_res;
2753 }
2754 spin_unlock(&ocfs2_dlm_tracking_lock);
2755
2756 return iter;
2757}
2758
2759static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2760{
2761}
2762
2763static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2764{
2765 struct ocfs2_dlm_seq_priv *priv = m->private;
2766 struct ocfs2_lock_res *iter = v;
2767 struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2768
2769 spin_lock(&ocfs2_dlm_tracking_lock);
2770 iter = ocfs2_dlm_next_res(iter, priv);
2771 list_del_init(&dummy->l_debug_list);
2772 if (iter) {
2773 list_add(&dummy->l_debug_list, &iter->l_debug_list);
2774 priv->p_tmp_res = *iter;
2775 iter = &priv->p_tmp_res;
2776 }
2777 spin_unlock(&ocfs2_dlm_tracking_lock);
2778
2779 return iter;
2780}
2781
Sunil Mushran5bc970e2010-12-28 23:26:03 -08002782/*
2783 * Version is used by debugfs.ocfs2 to determine the format being used
2784 *
2785 * New in version 2
2786 * - Lock stats printed
2787 * New in version 3
2788 * - Max time in lock stats is in usecs (instead of nsecs)
2789 */
2790#define OCFS2_DLM_DEBUG_STR_VERSION 3
Mark Fashehccd979b2005-12-15 14:31:24 -08002791static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2792{
2793 int i;
2794 char *lvb;
2795 struct ocfs2_lock_res *lockres = v;
2796
2797 if (!lockres)
2798 return -EINVAL;
2799
Mark Fashehd680efe2006-09-08 14:14:34 -07002800 seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2801
2802 if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2803 seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2804 lockres->l_name,
2805 (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2806 else
2807 seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2808
2809 seq_printf(m, "%d\t"
Mark Fashehccd979b2005-12-15 14:31:24 -08002810 "0x%lx\t"
2811 "0x%x\t"
2812 "0x%x\t"
2813 "%u\t"
2814 "%u\t"
2815 "%d\t"
2816 "%d\t",
Mark Fashehccd979b2005-12-15 14:31:24 -08002817 lockres->l_level,
2818 lockres->l_flags,
2819 lockres->l_action,
2820 lockres->l_unlock_action,
2821 lockres->l_ro_holders,
2822 lockres->l_ex_holders,
2823 lockres->l_requested,
2824 lockres->l_blocking);
2825
2826 /* Dump the raw LVB */
Joel Becker8f2c9c12008-02-01 12:16:57 -08002827 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002828 for(i = 0; i < DLM_LVB_LEN; i++)
2829 seq_printf(m, "0x%x\t", lvb[i]);
2830
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002831#ifdef CONFIG_OCFS2_FS_STATS
Sunil Mushran5bc970e2010-12-28 23:26:03 -08002832# define lock_num_prmode(_l) ((_l)->l_lock_prmode.ls_gets)
2833# define lock_num_exmode(_l) ((_l)->l_lock_exmode.ls_gets)
2834# define lock_num_prmode_failed(_l) ((_l)->l_lock_prmode.ls_fail)
2835# define lock_num_exmode_failed(_l) ((_l)->l_lock_exmode.ls_fail)
2836# define lock_total_prmode(_l) ((_l)->l_lock_prmode.ls_total)
2837# define lock_total_exmode(_l) ((_l)->l_lock_exmode.ls_total)
2838# define lock_max_prmode(_l) ((_l)->l_lock_prmode.ls_max)
2839# define lock_max_exmode(_l) ((_l)->l_lock_exmode.ls_max)
2840# define lock_refresh(_l) ((_l)->l_lock_refresh)
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002841#else
Sunil Mushran5bc970e2010-12-28 23:26:03 -08002842# define lock_num_prmode(_l) (0)
2843# define lock_num_exmode(_l) (0)
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002844# define lock_num_prmode_failed(_l) (0)
2845# define lock_num_exmode_failed(_l) (0)
Randy Dunlapdd25e552008-05-28 14:41:00 -07002846# define lock_total_prmode(_l) (0ULL)
2847# define lock_total_exmode(_l) (0ULL)
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002848# define lock_max_prmode(_l) (0)
2849# define lock_max_exmode(_l) (0)
2850# define lock_refresh(_l) (0)
2851#endif
2852 /* The following seq_print was added in version 2 of this output */
Sunil Mushran5bc970e2010-12-28 23:26:03 -08002853 seq_printf(m, "%u\t"
2854 "%u\t"
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002855 "%u\t"
2856 "%u\t"
2857 "%llu\t"
2858 "%llu\t"
2859 "%u\t"
2860 "%u\t"
2861 "%u\t",
2862 lock_num_prmode(lockres),
2863 lock_num_exmode(lockres),
2864 lock_num_prmode_failed(lockres),
2865 lock_num_exmode_failed(lockres),
2866 lock_total_prmode(lockres),
2867 lock_total_exmode(lockres),
2868 lock_max_prmode(lockres),
2869 lock_max_exmode(lockres),
2870 lock_refresh(lockres));
2871
Mark Fashehccd979b2005-12-15 14:31:24 -08002872 /* End the line */
2873 seq_printf(m, "\n");
2874 return 0;
2875}
2876
Jan Engelhardt90d99772008-01-22 20:52:20 +01002877static const struct seq_operations ocfs2_dlm_seq_ops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002878 .start = ocfs2_dlm_seq_start,
2879 .stop = ocfs2_dlm_seq_stop,
2880 .next = ocfs2_dlm_seq_next,
2881 .show = ocfs2_dlm_seq_show,
2882};
2883
2884static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2885{
Joe Perches33fa1d92010-07-12 13:50:19 -07002886 struct seq_file *seq = file->private_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08002887 struct ocfs2_dlm_seq_priv *priv = seq->private;
2888 struct ocfs2_lock_res *res = &priv->p_iter_res;
2889
2890 ocfs2_remove_lockres_tracking(res);
2891 ocfs2_put_dlm_debug(priv->p_dlm_debug);
2892 return seq_release_private(inode, file);
2893}
2894
2895static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2896{
2897 int ret;
2898 struct ocfs2_dlm_seq_priv *priv;
2899 struct seq_file *seq;
2900 struct ocfs2_super *osb;
2901
2902 priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2903 if (!priv) {
2904 ret = -ENOMEM;
2905 mlog_errno(ret);
2906 goto out;
2907 }
Theodore Ts'o8e18e292006-09-27 01:50:46 -07002908 osb = inode->i_private;
Mark Fashehccd979b2005-12-15 14:31:24 -08002909 ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2910 priv->p_dlm_debug = osb->osb_dlm_debug;
2911 INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2912
2913 ret = seq_open(file, &ocfs2_dlm_seq_ops);
2914 if (ret) {
2915 kfree(priv);
2916 mlog_errno(ret);
2917 goto out;
2918 }
2919
Joe Perches33fa1d92010-07-12 13:50:19 -07002920 seq = file->private_data;
Mark Fashehccd979b2005-12-15 14:31:24 -08002921 seq->private = priv;
2922
2923 ocfs2_add_lockres_tracking(&priv->p_iter_res,
2924 priv->p_dlm_debug);
2925
2926out:
2927 return ret;
2928}
2929
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002930static const struct file_operations ocfs2_dlm_debug_fops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002931 .open = ocfs2_dlm_debug_open,
2932 .release = ocfs2_dlm_debug_release,
2933 .read = seq_read,
2934 .llseek = seq_lseek,
2935};
2936
2937static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2938{
2939 int ret = 0;
2940 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2941
2942 dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2943 S_IFREG|S_IRUSR,
2944 osb->osb_debug_root,
2945 osb,
2946 &ocfs2_dlm_debug_fops);
2947 if (!dlm_debug->d_locking_state) {
2948 ret = -EINVAL;
2949 mlog(ML_ERROR,
2950 "Unable to create locking state debugfs file.\n");
2951 goto out;
2952 }
2953
2954 ocfs2_get_dlm_debug(dlm_debug);
2955out:
2956 return ret;
2957}
2958
2959static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2960{
2961 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2962
2963 if (dlm_debug) {
2964 debugfs_remove(dlm_debug->d_locking_state);
2965 ocfs2_put_dlm_debug(dlm_debug);
2966 }
2967}
2968
2969int ocfs2_dlm_init(struct ocfs2_super *osb)
2970{
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002971 int status = 0;
Joel Becker4670c462008-02-01 14:39:35 -08002972 struct ocfs2_cluster_connection *conn = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002973
Mark Fasheh0abd6d12008-01-29 16:59:56 -08002974 if (ocfs2_mount_local(osb)) {
2975 osb->node_num = 0;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002976 goto local;
Mark Fasheh0abd6d12008-01-29 16:59:56 -08002977 }
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002978
Mark Fashehccd979b2005-12-15 14:31:24 -08002979 status = ocfs2_dlm_init_debug(osb);
2980 if (status < 0) {
2981 mlog_errno(status);
2982 goto bail;
2983 }
2984
Mark Fasheh34d024f2007-09-24 15:56:19 -07002985 /* launch downconvert thread */
2986 osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc");
2987 if (IS_ERR(osb->dc_task)) {
2988 status = PTR_ERR(osb->dc_task);
2989 osb->dc_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002990 mlog_errno(status);
2991 goto bail;
2992 }
2993
Mark Fashehccd979b2005-12-15 14:31:24 -08002994 /* for now, uuid == domain */
Joel Becker9c6c8772008-02-01 15:17:30 -08002995 status = ocfs2_cluster_connect(osb->osb_cluster_stack,
2996 osb->uuid_str,
Joel Becker4670c462008-02-01 14:39:35 -08002997 strlen(osb->uuid_str),
Joel Becker553b5eb2010-01-29 17:19:06 -08002998 &lproto, ocfs2_do_node_down, osb,
Joel Becker4670c462008-02-01 14:39:35 -08002999 &conn);
3000 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003001 mlog_errno(status);
3002 goto bail;
3003 }
3004
Mark Fasheh0abd6d12008-01-29 16:59:56 -08003005 status = ocfs2_cluster_this_node(&osb->node_num);
3006 if (status < 0) {
3007 mlog_errno(status);
3008 mlog(ML_ERROR,
3009 "could not find this host's node number\n");
Joel Becker286eaa92008-02-01 15:03:57 -08003010 ocfs2_cluster_disconnect(conn, 0);
Mark Fasheh0abd6d12008-01-29 16:59:56 -08003011 goto bail;
3012 }
3013
Sunil Mushranc271c5c2006-12-05 17:56:35 -08003014local:
Mark Fashehccd979b2005-12-15 14:31:24 -08003015 ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
3016 ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
wengang wang6ca497a2009-03-06 21:29:10 +08003017 ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
Srinivas Eeda83273932009-06-03 17:02:55 -07003018 ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08003019
Joel Becker4670c462008-02-01 14:39:35 -08003020 osb->cconn = conn;
Mark Fashehccd979b2005-12-15 14:31:24 -08003021
3022 status = 0;
3023bail:
3024 if (status < 0) {
3025 ocfs2_dlm_shutdown_debug(osb);
Mark Fasheh34d024f2007-09-24 15:56:19 -07003026 if (osb->dc_task)
3027 kthread_stop(osb->dc_task);
Mark Fashehccd979b2005-12-15 14:31:24 -08003028 }
3029
Mark Fashehccd979b2005-12-15 14:31:24 -08003030 return status;
3031}
3032
Joel Becker286eaa92008-02-01 15:03:57 -08003033void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
3034 int hangup_pending)
Mark Fashehccd979b2005-12-15 14:31:24 -08003035{
Mark Fashehccd979b2005-12-15 14:31:24 -08003036 ocfs2_drop_osb_locks(osb);
3037
Joel Becker4670c462008-02-01 14:39:35 -08003038 /*
3039 * Now that we have dropped all locks and ocfs2_dismount_volume()
3040 * has disabled recovery, the DLM won't be talking to us. It's
3041 * safe to tear things down before disconnecting the cluster.
3042 */
3043
Mark Fasheh34d024f2007-09-24 15:56:19 -07003044 if (osb->dc_task) {
3045 kthread_stop(osb->dc_task);
3046 osb->dc_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08003047 }
3048
3049 ocfs2_lock_res_free(&osb->osb_super_lockres);
3050 ocfs2_lock_res_free(&osb->osb_rename_lockres);
wengang wang6ca497a2009-03-06 21:29:10 +08003051 ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
Srinivas Eeda83273932009-06-03 17:02:55 -07003052 ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003053
Joel Becker286eaa92008-02-01 15:03:57 -08003054 ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
Joel Becker4670c462008-02-01 14:39:35 -08003055 osb->cconn = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08003056
3057 ocfs2_dlm_shutdown_debug(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08003058}
3059
Mark Fashehccd979b2005-12-15 14:31:24 -08003060static int ocfs2_drop_lock(struct ocfs2_super *osb,
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07003061 struct ocfs2_lock_res *lockres)
Mark Fashehccd979b2005-12-15 14:31:24 -08003062{
Joel Becker7431cd72008-02-01 12:15:37 -08003063 int ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08003064 unsigned long flags;
Joel Beckerbd3e7612008-02-01 12:14:57 -08003065 u32 lkm_flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08003066
3067 /* We didn't get anywhere near actually using this lockres. */
3068 if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
3069 goto out;
3070
Mark Fashehb80fc012006-09-12 22:08:14 -07003071 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
Joel Beckerbd3e7612008-02-01 12:14:57 -08003072 lkm_flags |= DLM_LKF_VALBLK;
Mark Fashehb80fc012006-09-12 22:08:14 -07003073
Mark Fashehccd979b2005-12-15 14:31:24 -08003074 spin_lock_irqsave(&lockres->l_lock, flags);
3075
3076 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
3077 "lockres %s, flags 0x%lx\n",
3078 lockres->l_name, lockres->l_flags);
3079
3080 while (lockres->l_flags & OCFS2_LOCK_BUSY) {
3081 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
3082 "%u, unlock_action = %u\n",
3083 lockres->l_name, lockres->l_flags, lockres->l_action,
3084 lockres->l_unlock_action);
3085
3086 spin_unlock_irqrestore(&lockres->l_lock, flags);
3087
3088 /* XXX: Today we just wait on any busy
3089 * locks... Perhaps we need to cancel converts in the
3090 * future? */
3091 ocfs2_wait_on_busy_lock(lockres);
3092
3093 spin_lock_irqsave(&lockres->l_lock, flags);
3094 }
3095
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07003096 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
3097 if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
Joel Beckerbd3e7612008-02-01 12:14:57 -08003098 lockres->l_level == DLM_LOCK_EX &&
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07003099 !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3100 lockres->l_ops->set_lvb(lockres);
3101 }
Mark Fashehccd979b2005-12-15 14:31:24 -08003102
3103 if (lockres->l_flags & OCFS2_LOCK_BUSY)
3104 mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
3105 lockres->l_name);
3106 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
3107 mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
3108
3109 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
3110 spin_unlock_irqrestore(&lockres->l_lock, flags);
3111 goto out;
3112 }
3113
3114 lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
3115
3116 /* make sure we never get here while waiting for an ast to
3117 * fire. */
3118 BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
3119
3120 /* is this necessary? */
3121 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
3122 lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
3123 spin_unlock_irqrestore(&lockres->l_lock, flags);
3124
3125 mlog(0, "lock %s\n", lockres->l_name);
3126
Joel Beckera796d282010-01-28 19:22:39 -08003127 ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
Joel Becker7431cd72008-02-01 12:15:37 -08003128 if (ret) {
3129 ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003130 mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
Joel Beckercf0acdc2008-01-29 16:59:55 -08003131 ocfs2_dlm_dump_lksb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08003132 BUG();
3133 }
Coly Li73ac36e2009-01-07 18:09:16 -08003134 mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
Mark Fashehccd979b2005-12-15 14:31:24 -08003135 lockres->l_name);
3136
3137 ocfs2_wait_on_busy_lock(lockres);
3138out:
Mark Fashehccd979b2005-12-15 14:31:24 -08003139 return 0;
3140}
3141
3142/* Mark the lockres as being dropped. It will no longer be
3143 * queued if blocking, but we still may have to wait on it
Mark Fasheh34d024f2007-09-24 15:56:19 -07003144 * being dequeued from the downconvert thread before we can consider
Sunil Mushran2bd63212010-01-25 16:57:38 -08003145 * it safe to drop.
Mark Fashehccd979b2005-12-15 14:31:24 -08003146 *
3147 * You can *not* attempt to call cluster_lock on this lockres anymore. */
3148void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
3149{
3150 int status;
3151 struct ocfs2_mask_waiter mw;
3152 unsigned long flags;
3153
3154 ocfs2_init_mask_waiter(&mw);
3155
3156 spin_lock_irqsave(&lockres->l_lock, flags);
3157 lockres->l_flags |= OCFS2_LOCK_FREEING;
3158 while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
3159 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
3160 spin_unlock_irqrestore(&lockres->l_lock, flags);
3161
3162 mlog(0, "Waiting on lockres %s\n", lockres->l_name);
3163
3164 status = ocfs2_wait_for_mask(&mw);
3165 if (status)
3166 mlog_errno(status);
3167
3168 spin_lock_irqsave(&lockres->l_lock, flags);
3169 }
3170 spin_unlock_irqrestore(&lockres->l_lock, flags);
3171}
3172
Mark Fashehd680efe2006-09-08 14:14:34 -07003173void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3174 struct ocfs2_lock_res *lockres)
3175{
3176 int ret;
3177
3178 ocfs2_mark_lockres_freeing(lockres);
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07003179 ret = ocfs2_drop_lock(osb, lockres);
Mark Fashehd680efe2006-09-08 14:14:34 -07003180 if (ret)
3181 mlog_errno(ret);
3182}
3183
Mark Fashehccd979b2005-12-15 14:31:24 -08003184static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3185{
Mark Fashehd680efe2006-09-08 14:14:34 -07003186 ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3187 ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
wengang wang6ca497a2009-03-06 21:29:10 +08003188 ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
Srinivas Eeda83273932009-06-03 17:02:55 -07003189 ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003190}
3191
Mark Fashehccd979b2005-12-15 14:31:24 -08003192int ocfs2_drop_inode_locks(struct inode *inode)
3193{
3194 int status, err;
Mark Fashehccd979b2005-12-15 14:31:24 -08003195
Mark Fashehccd979b2005-12-15 14:31:24 -08003196 /* No need to call ocfs2_mark_lockres_freeing here -
3197 * ocfs2_clear_inode has done it for us. */
3198
3199 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
Tiger Yang50008632007-03-20 16:01:38 -07003200 &OCFS2_I(inode)->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003201 if (err < 0)
3202 mlog_errno(err);
3203
3204 status = err;
3205
3206 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
Mark Fashehe63aecb62007-10-18 15:30:42 -07003207 &OCFS2_I(inode)->ip_inode_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003208 if (err < 0)
3209 mlog_errno(err);
3210 if (err < 0 && !status)
3211 status = err;
3212
3213 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07003214 &OCFS2_I(inode)->ip_rw_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003215 if (err < 0)
3216 mlog_errno(err);
3217 if (err < 0 && !status)
3218 status = err;
3219
Mark Fashehccd979b2005-12-15 14:31:24 -08003220 return status;
3221}
3222
Joel Beckerde551242008-02-01 14:45:08 -08003223static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3224 int new_level)
Mark Fashehccd979b2005-12-15 14:31:24 -08003225{
3226 assert_spin_locked(&lockres->l_lock);
3227
Joel Beckerbd3e7612008-02-01 12:14:57 -08003228 BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
Mark Fashehccd979b2005-12-15 14:31:24 -08003229
3230 if (lockres->l_level <= new_level) {
Sunil Mushran9b915182010-02-26 19:42:44 -08003231 mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
3232 "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
3233 "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
3234 new_level, list_empty(&lockres->l_blocked_list),
3235 list_empty(&lockres->l_mask_waiters), lockres->l_type,
3236 lockres->l_flags, lockres->l_ro_holders,
3237 lockres->l_ex_holders, lockres->l_action,
3238 lockres->l_unlock_action, lockres->l_requested,
3239 lockres->l_blocking, lockres->l_pending_gen);
Mark Fashehccd979b2005-12-15 14:31:24 -08003240 BUG();
3241 }
3242
Sunil Mushran9b915182010-02-26 19:42:44 -08003243 mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
3244 lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
Mark Fashehccd979b2005-12-15 14:31:24 -08003245
3246 lockres->l_action = OCFS2_AST_DOWNCONVERT;
3247 lockres->l_requested = new_level;
3248 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
Joel Beckerde551242008-02-01 14:45:08 -08003249 return lockres_set_pending(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003250}
3251
3252static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3253 struct ocfs2_lock_res *lockres,
3254 int new_level,
Joel Beckerde551242008-02-01 14:45:08 -08003255 int lvb,
3256 unsigned int generation)
Mark Fashehccd979b2005-12-15 14:31:24 -08003257{
Joel Beckerbd3e7612008-02-01 12:14:57 -08003258 int ret;
3259 u32 dlm_flags = DLM_LKF_CONVERT;
Mark Fashehccd979b2005-12-15 14:31:24 -08003260
Sunil Mushran9b915182010-02-26 19:42:44 -08003261 mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
3262 lockres->l_level, new_level);
3263
Mark Fashehccd979b2005-12-15 14:31:24 -08003264 if (lvb)
Joel Beckerbd3e7612008-02-01 12:14:57 -08003265 dlm_flags |= DLM_LKF_VALBLK;
Mark Fashehccd979b2005-12-15 14:31:24 -08003266
Joel Becker4670c462008-02-01 14:39:35 -08003267 ret = ocfs2_dlm_lock(osb->cconn,
Joel Becker7431cd72008-02-01 12:15:37 -08003268 new_level,
3269 &lockres->l_lksb,
3270 dlm_flags,
3271 lockres->l_name,
Joel Beckera796d282010-01-28 19:22:39 -08003272 OCFS2_LOCK_ID_MAX_LEN - 1);
Joel Beckerde551242008-02-01 14:45:08 -08003273 lockres_clear_pending(lockres, generation, osb);
Joel Becker7431cd72008-02-01 12:15:37 -08003274 if (ret) {
3275 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003276 ocfs2_recover_from_dlm_error(lockres, 1);
3277 goto bail;
3278 }
3279
3280 ret = 0;
3281bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08003282 return ret;
3283}
3284
Joel Becker24ef1812008-01-29 17:37:32 -08003285/* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
Mark Fashehccd979b2005-12-15 14:31:24 -08003286static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3287 struct ocfs2_lock_res *lockres)
3288{
3289 assert_spin_locked(&lockres->l_lock);
3290
Mark Fashehccd979b2005-12-15 14:31:24 -08003291 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3292 /* If we're already trying to cancel a lock conversion
3293 * then just drop the spinlock and allow the caller to
3294 * requeue this lock. */
Sunil Mushran9b915182010-02-26 19:42:44 -08003295 mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
Mark Fashehccd979b2005-12-15 14:31:24 -08003296 return 0;
3297 }
3298
3299 /* were we in a convert when we got the bast fire? */
3300 BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3301 lockres->l_action != OCFS2_AST_DOWNCONVERT);
3302 /* set things up for the unlockast to know to just
3303 * clear out the ast_action and unset busy, etc. */
3304 lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3305
3306 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3307 "lock %s, invalid flags: 0x%lx\n",
3308 lockres->l_name, lockres->l_flags);
3309
Sunil Mushran9b915182010-02-26 19:42:44 -08003310 mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
3311
Mark Fashehccd979b2005-12-15 14:31:24 -08003312 return 1;
3313}
3314
3315static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3316 struct ocfs2_lock_res *lockres)
3317{
3318 int ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08003319
Joel Becker4670c462008-02-01 14:39:35 -08003320 ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
Joel Beckera796d282010-01-28 19:22:39 -08003321 DLM_LKF_CANCEL);
Joel Becker7431cd72008-02-01 12:15:37 -08003322 if (ret) {
3323 ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003324 ocfs2_recover_from_dlm_error(lockres, 0);
3325 }
3326
Sunil Mushran9b915182010-02-26 19:42:44 -08003327 mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
Mark Fashehccd979b2005-12-15 14:31:24 -08003328
Mark Fashehccd979b2005-12-15 14:31:24 -08003329 return ret;
3330}
3331
Mark Fashehb5e500e2006-09-13 22:01:16 -07003332static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3333 struct ocfs2_lock_res *lockres,
3334 struct ocfs2_unblock_ctl *ctl)
Mark Fashehccd979b2005-12-15 14:31:24 -08003335{
3336 unsigned long flags;
3337 int blocking;
3338 int new_level;
Sunil Mushran079b8052010-02-03 10:16:54 -08003339 int level;
Mark Fashehccd979b2005-12-15 14:31:24 -08003340 int ret = 0;
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -07003341 int set_lvb = 0;
Joel Beckerde551242008-02-01 14:45:08 -08003342 unsigned int gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08003343
Mark Fashehccd979b2005-12-15 14:31:24 -08003344 spin_lock_irqsave(&lockres->l_lock, flags);
3345
Mark Fashehccd979b2005-12-15 14:31:24 -08003346recheck:
Sunil Mushrandb0f6ce2010-02-01 16:55:50 -08003347 /*
3348 * Is it still blocking? If not, we have no more work to do.
3349 */
3350 if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
3351 BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
3352 spin_unlock_irqrestore(&lockres->l_lock, flags);
3353 ret = 0;
3354 goto leave;
3355 }
3356
Mark Fashehccd979b2005-12-15 14:31:24 -08003357 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
Joel Beckerde551242008-02-01 14:45:08 -08003358 /* XXX
3359 * This is a *big* race. The OCFS2_LOCK_PENDING flag
3360 * exists entirely for one reason - another thread has set
3361 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3362 *
3363 * If we do ocfs2_cancel_convert() before the other thread
3364 * calls dlm_lock(), our cancel will do nothing. We will
3365 * get no ast, and we will have no way of knowing the
3366 * cancel failed. Meanwhile, the other thread will call
3367 * into dlm_lock() and wait...forever.
3368 *
3369 * Why forever? Because another node has asked for the
3370 * lock first; that's why we're here in unblock_lock().
3371 *
3372 * The solution is OCFS2_LOCK_PENDING. When PENDING is
3373 * set, we just requeue the unblock. Only when the other
3374 * thread has called dlm_lock() and cleared PENDING will
3375 * we then cancel their request.
3376 *
3377 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3378 * at the same time they set OCFS2_DLM_BUSY. They must
3379 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3380 */
Sunil Mushran9b915182010-02-26 19:42:44 -08003381 if (lockres->l_flags & OCFS2_LOCK_PENDING) {
3382 mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
3383 lockres->l_name);
Joel Beckerde551242008-02-01 14:45:08 -08003384 goto leave_requeue;
Sunil Mushran9b915182010-02-26 19:42:44 -08003385 }
Joel Beckerde551242008-02-01 14:45:08 -08003386
Mark Fashehd680efe2006-09-08 14:14:34 -07003387 ctl->requeue = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08003388 ret = ocfs2_prepare_cancel_convert(osb, lockres);
3389 spin_unlock_irqrestore(&lockres->l_lock, flags);
3390 if (ret) {
3391 ret = ocfs2_cancel_convert(osb, lockres);
3392 if (ret < 0)
3393 mlog_errno(ret);
3394 }
3395 goto leave;
3396 }
3397
Sunil Mushrana1912822010-01-21 10:50:03 -08003398 /*
3399 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
3400 * set when the ast is received for an upconvert just before the
3401 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
3402 * on the heels of the ast, we want to delay the downconvert just
3403 * enough to allow the up requestor to do its task. Because this
3404 * lock is in the blocked queue, the lock will be downconverted
3405 * as soon as the requestor is done with the lock.
3406 */
3407 if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
3408 goto leave_requeue;
3409
Sunil Mushran0d741252010-01-29 09:44:11 -08003410 /*
3411 * How can we block and yet be at NL? We were trying to upconvert
3412 * from NL and got canceled. The code comes back here, and now
3413 * we notice and clear BLOCKING.
3414 */
3415 if (lockres->l_level == DLM_LOCK_NL) {
3416 BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
Sunil Mushran9b915182010-02-26 19:42:44 -08003417 mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
Sunil Mushran0d741252010-01-29 09:44:11 -08003418 lockres->l_blocking = DLM_LOCK_NL;
3419 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
3420 spin_unlock_irqrestore(&lockres->l_lock, flags);
3421 goto leave;
3422 }
3423
Mark Fashehccd979b2005-12-15 14:31:24 -08003424 /* if we're blocking an exclusive and we have *any* holders,
3425 * then requeue. */
Joel Beckerbd3e7612008-02-01 12:14:57 -08003426 if ((lockres->l_blocking == DLM_LOCK_EX)
Sunil Mushran9b915182010-02-26 19:42:44 -08003427 && (lockres->l_ex_holders || lockres->l_ro_holders)) {
3428 mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
3429 lockres->l_name, lockres->l_ex_holders,
3430 lockres->l_ro_holders);
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003431 goto leave_requeue;
Sunil Mushran9b915182010-02-26 19:42:44 -08003432 }
Mark Fashehccd979b2005-12-15 14:31:24 -08003433
3434 /* If it's a PR we're blocking, then only
3435 * requeue if we've got any EX holders */
Joel Beckerbd3e7612008-02-01 12:14:57 -08003436 if (lockres->l_blocking == DLM_LOCK_PR &&
Sunil Mushran9b915182010-02-26 19:42:44 -08003437 lockres->l_ex_holders) {
3438 mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
3439 lockres->l_name, lockres->l_ex_holders);
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003440 goto leave_requeue;
Sunil Mushran9b915182010-02-26 19:42:44 -08003441 }
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003442
3443 /*
3444 * Can we get a lock in this state if the holder counts are
3445 * zero? The meta data unblock code used to check this.
3446 */
3447 if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
Sunil Mushran9b915182010-02-26 19:42:44 -08003448 && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
3449 mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
3450 lockres->l_name);
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003451 goto leave_requeue;
Sunil Mushran9b915182010-02-26 19:42:44 -08003452 }
Mark Fashehccd979b2005-12-15 14:31:24 -08003453
Mark Fasheh16d5b9562006-09-13 21:10:12 -07003454 new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3455
3456 if (lockres->l_ops->check_downconvert
Sunil Mushran9b915182010-02-26 19:42:44 -08003457 && !lockres->l_ops->check_downconvert(lockres, new_level)) {
3458 mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
3459 lockres->l_name);
Mark Fasheh16d5b9562006-09-13 21:10:12 -07003460 goto leave_requeue;
Sunil Mushran9b915182010-02-26 19:42:44 -08003461 }
Mark Fasheh16d5b9562006-09-13 21:10:12 -07003462
Mark Fashehccd979b2005-12-15 14:31:24 -08003463 /* If we get here, then we know that there are no more
3464 * incompatible holders (and anyone asking for an incompatible
3465 * lock is blocked). We can now downconvert the lock */
Mark Fashehcc567d82006-09-13 21:52:21 -07003466 if (!lockres->l_ops->downconvert_worker)
Mark Fashehccd979b2005-12-15 14:31:24 -08003467 goto downconvert;
3468
3469 /* Some lockres types want to do a bit of work before
3470 * downconverting a lock. Allow that here. The worker function
3471 * may sleep, so we save off a copy of what we're blocking as
3472 * it may change while we're not holding the spin lock. */
3473 blocking = lockres->l_blocking;
Sunil Mushran079b8052010-02-03 10:16:54 -08003474 level = lockres->l_level;
Mark Fashehccd979b2005-12-15 14:31:24 -08003475 spin_unlock_irqrestore(&lockres->l_lock, flags);
3476
Mark Fashehcc567d82006-09-13 21:52:21 -07003477 ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
Mark Fashehd680efe2006-09-08 14:14:34 -07003478
Sunil Mushran9b915182010-02-26 19:42:44 -08003479 if (ctl->unblock_action == UNBLOCK_STOP_POST) {
3480 mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
3481 lockres->l_name);
Mark Fashehd680efe2006-09-08 14:14:34 -07003482 goto leave;
Sunil Mushran9b915182010-02-26 19:42:44 -08003483 }
Mark Fashehccd979b2005-12-15 14:31:24 -08003484
3485 spin_lock_irqsave(&lockres->l_lock, flags);
Sunil Mushran079b8052010-02-03 10:16:54 -08003486 if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003487 /* If this changed underneath us, then we can't drop
3488 * it just yet. */
Sunil Mushran9b915182010-02-26 19:42:44 -08003489 mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
3490 "Recheck\n", lockres->l_name, blocking,
3491 lockres->l_blocking, level, lockres->l_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08003492 goto recheck;
3493 }
3494
3495downconvert:
Mark Fashehd680efe2006-09-08 14:14:34 -07003496 ctl->requeue = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08003497
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -07003498 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
Joel Beckerbd3e7612008-02-01 12:14:57 -08003499 if (lockres->l_level == DLM_LOCK_EX)
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -07003500 set_lvb = 1;
3501
3502 /*
3503 * We only set the lvb if the lock has been fully
3504 * refreshed - otherwise we risk setting stale
3505 * data. Otherwise, there's no need to actually clear
3506 * out the lvb here as it's value is still valid.
3507 */
3508 if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3509 lockres->l_ops->set_lvb(lockres);
3510 }
3511
Joel Beckerde551242008-02-01 14:45:08 -08003512 gen = ocfs2_prepare_downconvert(lockres, new_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08003513 spin_unlock_irqrestore(&lockres->l_lock, flags);
Joel Beckerde551242008-02-01 14:45:08 -08003514 ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3515 gen);
3516
Mark Fashehccd979b2005-12-15 14:31:24 -08003517leave:
Tao Mac1e8d352011-03-07 16:43:21 +08003518 if (ret)
3519 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08003520 return ret;
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003521
3522leave_requeue:
3523 spin_unlock_irqrestore(&lockres->l_lock, flags);
3524 ctl->requeue = 1;
3525
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003526 return 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08003527}
3528
Mark Fashehd680efe2006-09-08 14:14:34 -07003529static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3530 int blocking)
Mark Fashehccd979b2005-12-15 14:31:24 -08003531{
3532 struct inode *inode;
3533 struct address_space *mapping;
Goldwyn Rodrigues5e98d492010-06-28 10:04:32 -05003534 struct ocfs2_inode_info *oi;
Mark Fashehccd979b2005-12-15 14:31:24 -08003535
Mark Fashehccd979b2005-12-15 14:31:24 -08003536 inode = ocfs2_lock_res_inode(lockres);
3537 mapping = inode->i_mapping;
3538
Goldwyn Rodrigues5e98d492010-06-28 10:04:32 -05003539 if (S_ISDIR(inode->i_mode)) {
3540 oi = OCFS2_I(inode);
3541 oi->ip_dir_lock_gen++;
3542 mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
3543 goto out;
3544 }
3545
Mark Fasheh1044e402008-02-28 17:16:03 -08003546 if (!S_ISREG(inode->i_mode))
Mark Fashehf1f54062007-10-18 15:13:59 -07003547 goto out;
3548
Mark Fasheh7f4a2a92006-12-11 11:06:36 -08003549 /*
3550 * We need this before the filemap_fdatawrite() so that it can
3551 * transfer the dirty bit from the PTE to the
3552 * page. Unfortunately this means that even for EX->PR
3553 * downconverts, we'll lose our mappings and have to build
3554 * them up again.
3555 */
3556 unmap_mapping_range(mapping, 0, 0, 0);
3557
Mark Fashehccd979b2005-12-15 14:31:24 -08003558 if (filemap_fdatawrite(mapping)) {
Mark Fashehb0697052006-03-03 10:24:33 -08003559 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3560 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08003561 }
3562 sync_mapping_buffers(mapping);
Joel Beckerbd3e7612008-02-01 12:14:57 -08003563 if (blocking == DLM_LOCK_EX) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003564 truncate_inode_pages(mapping, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08003565 } else {
3566 /* We only need to wait on the I/O if we're not also
3567 * truncating pages because truncate_inode_pages waits
3568 * for us above. We don't truncate pages if we're
3569 * blocking anything < EXMODE because we want to keep
3570 * them around in that case. */
3571 filemap_fdatawait(mapping);
3572 }
3573
Mark Fashehf1f54062007-10-18 15:13:59 -07003574out:
Mark Fashehd680efe2006-09-08 14:14:34 -07003575 return UNBLOCK_CONTINUE;
Mark Fashehccd979b2005-12-15 14:31:24 -08003576}
3577
Tao Maa4338482009-08-18 11:19:29 +08003578static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
3579 struct ocfs2_lock_res *lockres,
3580 int new_level)
Mark Fasheh810d5ae2006-09-13 21:39:52 -07003581{
Tao Maa4338482009-08-18 11:19:29 +08003582 int checkpointed = ocfs2_ci_fully_checkpointed(ci);
Mark Fasheh810d5ae2006-09-13 21:39:52 -07003583
Joel Beckerbd3e7612008-02-01 12:14:57 -08003584 BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3585 BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
Mark Fasheh810d5ae2006-09-13 21:39:52 -07003586
3587 if (checkpointed)
3588 return 1;
3589
Tao Maa4338482009-08-18 11:19:29 +08003590 ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
Mark Fasheh810d5ae2006-09-13 21:39:52 -07003591 return 0;
3592}
3593
Tao Maa4338482009-08-18 11:19:29 +08003594static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3595 int new_level)
3596{
3597 struct inode *inode = ocfs2_lock_res_inode(lockres);
3598
3599 return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
3600}
3601
Mark Fasheh810d5ae2006-09-13 21:39:52 -07003602static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3603{
3604 struct inode *inode = ocfs2_lock_res_inode(lockres);
3605
3606 __ocfs2_stuff_meta_lvb(inode);
3607}
3608
Mark Fashehd680efe2006-09-08 14:14:34 -07003609/*
3610 * Does the final reference drop on our dentry lock. Right now this
Mark Fasheh34d024f2007-09-24 15:56:19 -07003611 * happens in the downconvert thread, but we could choose to simplify the
Mark Fashehd680efe2006-09-08 14:14:34 -07003612 * dlmglue API and push these off to the ocfs2_wq in the future.
3613 */
3614static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3615 struct ocfs2_lock_res *lockres)
3616{
3617 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3618 ocfs2_dentry_lock_put(osb, dl);
3619}
3620
3621/*
3622 * d_delete() matching dentries before the lock downconvert.
3623 *
3624 * At this point, any process waiting to destroy the
3625 * dentry_lock due to last ref count is stopped by the
3626 * OCFS2_LOCK_QUEUED flag.
3627 *
3628 * We have two potential problems
3629 *
3630 * 1) If we do the last reference drop on our dentry_lock (via dput)
3631 * we'll wind up in ocfs2_release_dentry_lock(), waiting on
3632 * the downconvert to finish. Instead we take an elevated
3633 * reference and push the drop until after we've completed our
3634 * unblock processing.
3635 *
3636 * 2) There might be another process with a final reference,
3637 * waiting on us to finish processing. If this is the case, we
3638 * detect it and exit out - there's no more dentries anyway.
3639 */
3640static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3641 int blocking)
3642{
3643 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3644 struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3645 struct dentry *dentry;
3646 unsigned long flags;
3647 int extra_ref = 0;
3648
3649 /*
3650 * This node is blocking another node from getting a read
3651 * lock. This happens when we've renamed within a
3652 * directory. We've forced the other nodes to d_delete(), but
3653 * we never actually dropped our lock because it's still
3654 * valid. The downconvert code will retain a PR for this node,
3655 * so there's no further work to do.
3656 */
Joel Beckerbd3e7612008-02-01 12:14:57 -08003657 if (blocking == DLM_LOCK_PR)
Mark Fashehd680efe2006-09-08 14:14:34 -07003658 return UNBLOCK_CONTINUE;
3659
3660 /*
3661 * Mark this inode as potentially orphaned. The code in
3662 * ocfs2_delete_inode() will figure out whether it actually
3663 * needs to be freed or not.
3664 */
3665 spin_lock(&oi->ip_lock);
3666 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
3667 spin_unlock(&oi->ip_lock);
3668
3669 /*
3670 * Yuck. We need to make sure however that the check of
3671 * OCFS2_LOCK_FREEING and the extra reference are atomic with
3672 * respect to a reference decrement or the setting of that
3673 * flag.
3674 */
3675 spin_lock_irqsave(&lockres->l_lock, flags);
3676 spin_lock(&dentry_attach_lock);
3677 if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
3678 && dl->dl_count) {
3679 dl->dl_count++;
3680 extra_ref = 1;
3681 }
3682 spin_unlock(&dentry_attach_lock);
3683 spin_unlock_irqrestore(&lockres->l_lock, flags);
3684
3685 mlog(0, "extra_ref = %d\n", extra_ref);
3686
3687 /*
3688 * We have a process waiting on us in ocfs2_dentry_iput(),
3689 * which means we can't have any more outstanding
3690 * aliases. There's no need to do any more work.
3691 */
3692 if (!extra_ref)
3693 return UNBLOCK_CONTINUE;
3694
3695 spin_lock(&dentry_attach_lock);
3696 while (1) {
3697 dentry = ocfs2_find_local_alias(dl->dl_inode,
3698 dl->dl_parent_blkno, 1);
3699 if (!dentry)
3700 break;
3701 spin_unlock(&dentry_attach_lock);
3702
3703 mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
3704 dentry->d_name.name);
3705
3706 /*
3707 * The following dcache calls may do an
3708 * iput(). Normally we don't want that from the
3709 * downconverting thread, but in this case it's ok
3710 * because the requesting node already has an
3711 * exclusive lock on the inode, so it can't be queued
3712 * for a downconvert.
3713 */
3714 d_delete(dentry);
3715 dput(dentry);
3716
3717 spin_lock(&dentry_attach_lock);
3718 }
3719 spin_unlock(&dentry_attach_lock);
3720
3721 /*
3722 * If we are the last holder of this dentry lock, there is no
3723 * reason to downconvert so skip straight to the unlock.
3724 */
3725 if (dl->dl_count == 1)
3726 return UNBLOCK_STOP_POST;
3727
3728 return UNBLOCK_CONTINUE_POST;
3729}
3730
Tao Ma8dec98e2009-08-18 11:19:58 +08003731static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
3732 int new_level)
3733{
3734 struct ocfs2_refcount_tree *tree =
3735 ocfs2_lock_res_refcount_tree(lockres);
3736
3737 return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
3738}
3739
3740static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
3741 int blocking)
3742{
3743 struct ocfs2_refcount_tree *tree =
3744 ocfs2_lock_res_refcount_tree(lockres);
3745
3746 ocfs2_metadata_cache_purge(&tree->rf_ci);
3747
3748 return UNBLOCK_CONTINUE;
3749}
3750
Jan Kara9e33d692008-08-25 19:56:50 +02003751static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
3752{
3753 struct ocfs2_qinfo_lvb *lvb;
3754 struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
3755 struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3756 oinfo->dqi_gi.dqi_type);
3757
Mark Fasheha641dc22008-12-24 16:03:48 -08003758 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Jan Kara9e33d692008-08-25 19:56:50 +02003759 lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
3760 lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
3761 lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
3762 lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
3763 lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
3764 lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
3765 lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
Jan Kara9e33d692008-08-25 19:56:50 +02003766}
3767
3768void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3769{
3770 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3771 struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3772 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3773
Jan Kara9e33d692008-08-25 19:56:50 +02003774 if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
3775 ocfs2_cluster_unlock(osb, lockres, level);
Jan Kara9e33d692008-08-25 19:56:50 +02003776}
3777
3778static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
3779{
3780 struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3781 oinfo->dqi_gi.dqi_type);
3782 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3783 struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Joel Becker85eb8b72008-11-25 15:31:27 +01003784 struct buffer_head *bh = NULL;
Jan Kara9e33d692008-08-25 19:56:50 +02003785 struct ocfs2_global_disk_dqinfo *gdinfo;
3786 int status = 0;
3787
Joel Becker1c520df2009-06-19 15:14:13 -07003788 if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
3789 lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
Jan Kara9e33d692008-08-25 19:56:50 +02003790 info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
3791 info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
3792 oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
3793 oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
3794 oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
3795 oinfo->dqi_gi.dqi_free_entry =
3796 be32_to_cpu(lvb->lvb_free_entry);
3797 } else {
Jan Karaae4f6ef2010-04-28 19:04:29 +02003798 status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
3799 oinfo->dqi_giblk, &bh);
Joel Becker85eb8b72008-11-25 15:31:27 +01003800 if (status) {
Jan Kara9e33d692008-08-25 19:56:50 +02003801 mlog_errno(status);
3802 goto bail;
3803 }
3804 gdinfo = (struct ocfs2_global_disk_dqinfo *)
3805 (bh->b_data + OCFS2_GLOBAL_INFO_OFF);
3806 info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
3807 info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
3808 oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
3809 oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
3810 oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
3811 oinfo->dqi_gi.dqi_free_entry =
3812 le32_to_cpu(gdinfo->dqi_free_entry);
3813 brelse(bh);
3814 ocfs2_track_lock_refresh(lockres);
3815 }
3816
3817bail:
3818 return status;
3819}
3820
3821/* Lock quota info, this function expects at least shared lock on the quota file
3822 * so that we can safely refresh quota info from disk. */
3823int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3824{
3825 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3826 struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3827 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3828 int status = 0;
3829
Jan Kara9e33d692008-08-25 19:56:50 +02003830 /* On RO devices, locking really isn't needed... */
3831 if (ocfs2_is_hard_readonly(osb)) {
3832 if (ex)
3833 status = -EROFS;
3834 goto bail;
3835 }
3836 if (ocfs2_mount_local(osb))
3837 goto bail;
3838
3839 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3840 if (status < 0) {
3841 mlog_errno(status);
3842 goto bail;
3843 }
3844 if (!ocfs2_should_refresh_lock_res(lockres))
3845 goto bail;
3846 /* OK, we have the lock but we need to refresh the quota info */
3847 status = ocfs2_refresh_qinfo(oinfo);
3848 if (status)
3849 ocfs2_qinfo_unlock(oinfo, ex);
3850 ocfs2_complete_lock_res_refresh(lockres, status);
3851bail:
Jan Kara9e33d692008-08-25 19:56:50 +02003852 return status;
3853}
3854
Tao Ma8dec98e2009-08-18 11:19:58 +08003855int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
3856{
3857 int status;
3858 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3859 struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
3860 struct ocfs2_super *osb = lockres->l_priv;
3861
3862
3863 if (ocfs2_is_hard_readonly(osb))
3864 return -EROFS;
3865
3866 if (ocfs2_mount_local(osb))
3867 return 0;
3868
3869 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3870 if (status < 0)
3871 mlog_errno(status);
3872
3873 return status;
3874}
3875
3876void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
3877{
3878 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3879 struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
3880 struct ocfs2_super *osb = lockres->l_priv;
3881
3882 if (!ocfs2_mount_local(osb))
3883 ocfs2_cluster_unlock(osb, lockres, level);
3884}
3885
Adrian Bunk00600052008-01-29 00:11:41 +02003886static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3887 struct ocfs2_lock_res *lockres)
Mark Fashehccd979b2005-12-15 14:31:24 -08003888{
3889 int status;
Mark Fashehd680efe2006-09-08 14:14:34 -07003890 struct ocfs2_unblock_ctl ctl = {0, 0,};
Mark Fashehccd979b2005-12-15 14:31:24 -08003891 unsigned long flags;
3892
3893 /* Our reference to the lockres in this function can be
3894 * considered valid until we remove the OCFS2_LOCK_QUEUED
3895 * flag. */
3896
Mark Fashehccd979b2005-12-15 14:31:24 -08003897 BUG_ON(!lockres);
3898 BUG_ON(!lockres->l_ops);
Mark Fashehccd979b2005-12-15 14:31:24 -08003899
Sunil Mushran9b915182010-02-26 19:42:44 -08003900 mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
Mark Fashehccd979b2005-12-15 14:31:24 -08003901
3902 /* Detect whether a lock has been marked as going away while
Mark Fasheh34d024f2007-09-24 15:56:19 -07003903 * the downconvert thread was processing other things. A lock can
Mark Fashehccd979b2005-12-15 14:31:24 -08003904 * still be marked with OCFS2_LOCK_FREEING after this check,
3905 * but short circuiting here will still save us some
3906 * performance. */
3907 spin_lock_irqsave(&lockres->l_lock, flags);
3908 if (lockres->l_flags & OCFS2_LOCK_FREEING)
3909 goto unqueue;
3910 spin_unlock_irqrestore(&lockres->l_lock, flags);
3911
Mark Fashehb5e500e2006-09-13 22:01:16 -07003912 status = ocfs2_unblock_lock(osb, lockres, &ctl);
Mark Fashehccd979b2005-12-15 14:31:24 -08003913 if (status < 0)
3914 mlog_errno(status);
3915
3916 spin_lock_irqsave(&lockres->l_lock, flags);
3917unqueue:
Mark Fashehd680efe2006-09-08 14:14:34 -07003918 if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003919 lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3920 } else
3921 ocfs2_schedule_blocked_lock(osb, lockres);
3922
Sunil Mushran9b915182010-02-26 19:42:44 -08003923 mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
Mark Fashehd680efe2006-09-08 14:14:34 -07003924 ctl.requeue ? "yes" : "no");
Mark Fashehccd979b2005-12-15 14:31:24 -08003925 spin_unlock_irqrestore(&lockres->l_lock, flags);
3926
Mark Fashehd680efe2006-09-08 14:14:34 -07003927 if (ctl.unblock_action != UNBLOCK_CONTINUE
3928 && lockres->l_ops->post_unlock)
3929 lockres->l_ops->post_unlock(osb, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003930}
3931
3932static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3933 struct ocfs2_lock_res *lockres)
3934{
Mark Fashehccd979b2005-12-15 14:31:24 -08003935 assert_spin_locked(&lockres->l_lock);
3936
3937 if (lockres->l_flags & OCFS2_LOCK_FREEING) {
3938 /* Do not schedule a lock for downconvert when it's on
3939 * the way to destruction - any nodes wanting access
3940 * to the resource will get it soon. */
Sunil Mushran9b915182010-02-26 19:42:44 -08003941 mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
Mark Fashehccd979b2005-12-15 14:31:24 -08003942 lockres->l_name, lockres->l_flags);
3943 return;
3944 }
3945
3946 lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
3947
Mark Fasheh34d024f2007-09-24 15:56:19 -07003948 spin_lock(&osb->dc_task_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08003949 if (list_empty(&lockres->l_blocked_list)) {
3950 list_add_tail(&lockres->l_blocked_list,
3951 &osb->blocked_lock_list);
3952 osb->blocked_lock_count++;
3953 }
Mark Fasheh34d024f2007-09-24 15:56:19 -07003954 spin_unlock(&osb->dc_task_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08003955}
Mark Fasheh34d024f2007-09-24 15:56:19 -07003956
3957static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
3958{
3959 unsigned long processed;
3960 struct ocfs2_lock_res *lockres;
3961
Mark Fasheh34d024f2007-09-24 15:56:19 -07003962 spin_lock(&osb->dc_task_lock);
3963 /* grab this early so we know to try again if a state change and
3964 * wake happens part-way through our work */
3965 osb->dc_work_sequence = osb->dc_wake_sequence;
3966
3967 processed = osb->blocked_lock_count;
3968 while (processed) {
3969 BUG_ON(list_empty(&osb->blocked_lock_list));
3970
3971 lockres = list_entry(osb->blocked_lock_list.next,
3972 struct ocfs2_lock_res, l_blocked_list);
3973 list_del_init(&lockres->l_blocked_list);
3974 osb->blocked_lock_count--;
3975 spin_unlock(&osb->dc_task_lock);
3976
3977 BUG_ON(!processed);
3978 processed--;
3979
3980 ocfs2_process_blocked_lock(osb, lockres);
3981
3982 spin_lock(&osb->dc_task_lock);
3983 }
3984 spin_unlock(&osb->dc_task_lock);
Mark Fasheh34d024f2007-09-24 15:56:19 -07003985}
3986
3987static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
3988{
3989 int empty = 0;
3990
3991 spin_lock(&osb->dc_task_lock);
3992 if (list_empty(&osb->blocked_lock_list))
3993 empty = 1;
3994
3995 spin_unlock(&osb->dc_task_lock);
3996 return empty;
3997}
3998
3999static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
4000{
4001 int should_wake = 0;
4002
4003 spin_lock(&osb->dc_task_lock);
4004 if (osb->dc_work_sequence != osb->dc_wake_sequence)
4005 should_wake = 1;
4006 spin_unlock(&osb->dc_task_lock);
4007
4008 return should_wake;
4009}
4010
Adrian Bunk200bfae2008-02-17 10:20:38 +02004011static int ocfs2_downconvert_thread(void *arg)
Mark Fasheh34d024f2007-09-24 15:56:19 -07004012{
4013 int status = 0;
4014 struct ocfs2_super *osb = arg;
4015
4016 /* only quit once we've been asked to stop and there is no more
4017 * work available */
4018 while (!(kthread_should_stop() &&
4019 ocfs2_downconvert_thread_lists_empty(osb))) {
4020
4021 wait_event_interruptible(osb->dc_event,
4022 ocfs2_downconvert_thread_should_wake(osb) ||
4023 kthread_should_stop());
4024
4025 mlog(0, "downconvert_thread: awoken\n");
4026
4027 ocfs2_downconvert_thread_do_work(osb);
4028 }
4029
4030 osb->dc_task = NULL;
4031 return status;
4032}
4033
4034void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
4035{
4036 spin_lock(&osb->dc_task_lock);
4037 /* make sure the voting thread gets a swipe at whatever changes
4038 * the caller may have made to the voting state */
4039 osb->dc_wake_sequence++;
4040 spin_unlock(&osb->dc_task_lock);
4041 wake_up(&osb->dc_event);
4042}