blob: 7219a86d34ccc3102b360e10c298ed50fda903a7 [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"
Mark Fashehccd979b2005-12-15 14:31:24 -080056
57#include "buffer_head_io.h"
58
59struct ocfs2_mask_waiter {
60 struct list_head mw_item;
61 int mw_status;
62 struct completion mw_complete;
63 unsigned long mw_mask;
64 unsigned long mw_goal;
Sunil Mushran8ddb7b02008-05-13 13:45:15 -070065#ifdef CONFIG_OCFS2_FS_STATS
66 unsigned long long mw_lock_start;
67#endif
Mark Fashehccd979b2005-12-15 14:31:24 -080068};
69
Mark Fasheh54a7e752006-09-12 21:49:13 -070070static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
71static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
Mark Fashehcf8e06f2007-12-20 16:43:10 -080072static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
Jan Kara9e33d692008-08-25 19:56:50 +020073static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -080074
Mark Fashehd680efe2006-09-08 14:14:34 -070075/*
Mark Fashehcc567d82006-09-13 21:52:21 -070076 * Return value from ->downconvert_worker functions.
Mark Fashehd680efe2006-09-08 14:14:34 -070077 *
Mark Fashehb5e500e2006-09-13 22:01:16 -070078 * These control the precise actions of ocfs2_unblock_lock()
Mark Fashehd680efe2006-09-08 14:14:34 -070079 * and ocfs2_process_blocked_lock()
80 *
81 */
82enum ocfs2_unblock_action {
83 UNBLOCK_CONTINUE = 0, /* Continue downconvert */
84 UNBLOCK_CONTINUE_POST = 1, /* Continue downconvert, fire
85 * ->post_unlock callback */
86 UNBLOCK_STOP_POST = 2, /* Do not downconvert, fire
87 * ->post_unlock() callback. */
88};
89
90struct ocfs2_unblock_ctl {
91 int requeue;
92 enum ocfs2_unblock_action unblock_action;
93};
94
Mark Fasheh810d5ae2006-09-13 21:39:52 -070095static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
96 int new_level);
97static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
98
Mark Fashehcc567d82006-09-13 21:52:21 -070099static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
100 int blocking);
101
Mark Fashehcc567d82006-09-13 21:52:21 -0700102static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
103 int blocking);
Mark Fashehd680efe2006-09-08 14:14:34 -0700104
105static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
106 struct ocfs2_lock_res *lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -0800107
Jan Kara9e33d692008-08-25 19:56:50 +0200108static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700109
110#define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
111
112/* This aids in debugging situations where a bad LVB might be involved. */
113static void ocfs2_dump_meta_lvb_info(u64 level,
114 const char *function,
115 unsigned int line,
116 struct ocfs2_lock_res *lockres)
117{
Mark Fasheha641dc22008-12-24 16:03:48 -0800118 struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700119
120 mlog(level, "LVB information for %s (called from %s:%u):\n",
121 lockres->l_name, function, line);
122 mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
123 lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
124 be32_to_cpu(lvb->lvb_igeneration));
125 mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
126 (unsigned long long)be64_to_cpu(lvb->lvb_isize),
127 be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
128 be16_to_cpu(lvb->lvb_imode));
129 mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
130 "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
131 (long long)be64_to_cpu(lvb->lvb_iatime_packed),
132 (long long)be64_to_cpu(lvb->lvb_ictime_packed),
133 (long long)be64_to_cpu(lvb->lvb_imtime_packed),
134 be32_to_cpu(lvb->lvb_iattr));
135}
136
137
Mark Fashehf625c972006-09-12 21:24:53 -0700138/*
139 * OCFS2 Lock Resource Operations
140 *
141 * These fine tune the behavior of the generic dlmglue locking infrastructure.
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -0700142 *
143 * The most basic of lock types can point ->l_priv to their respective
144 * struct ocfs2_super and allow the default actions to manage things.
145 *
146 * Right now, each lock type also needs to implement an init function,
147 * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
148 * should be called when the lock is no longer needed (i.e., object
149 * destruction time).
Mark Fashehf625c972006-09-12 21:24:53 -0700150 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800151struct ocfs2_lock_res_ops {
Mark Fasheh54a7e752006-09-12 21:49:13 -0700152 /*
153 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
154 * this callback if ->l_priv is not an ocfs2_super pointer
155 */
156 struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
Mark Fashehb5e500e2006-09-13 22:01:16 -0700157
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -0700158 /*
Mark Fasheh34d024f2007-09-24 15:56:19 -0700159 * Optionally called in the downconvert thread after a
160 * successful downconvert. The lockres will not be referenced
161 * after this callback is called, so it is safe to free
162 * memory, etc.
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -0700163 *
164 * The exact semantics of when this is called are controlled
165 * by ->downconvert_worker()
166 */
Mark Fashehd680efe2006-09-08 14:14:34 -0700167 void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
Mark Fashehf625c972006-09-12 21:24:53 -0700168
169 /*
Mark Fasheh16d5b952006-09-13 21:10:12 -0700170 * Allow a lock type to add checks to determine whether it is
171 * safe to downconvert a lock. Return 0 to re-queue the
172 * downconvert at a later time, nonzero to continue.
173 *
174 * For most locks, the default checks that there are no
175 * incompatible holders are sufficient.
176 *
177 * Called with the lockres spinlock held.
178 */
179 int (*check_downconvert)(struct ocfs2_lock_res *, int);
180
181 /*
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -0700182 * Allows a lock type to populate the lock value block. This
183 * is called on downconvert, and when we drop a lock.
184 *
185 * Locks that want to use this should set LOCK_TYPE_USES_LVB
186 * in the flags field.
187 *
188 * Called with the lockres spinlock held.
189 */
190 void (*set_lvb)(struct ocfs2_lock_res *);
191
192 /*
Mark Fashehcc567d82006-09-13 21:52:21 -0700193 * Called from the downconvert thread when it is determined
194 * that a lock will be downconverted. This is called without
195 * any locks held so the function can do work that might
196 * schedule (syncing out data, etc).
197 *
198 * This should return any one of the ocfs2_unblock_action
199 * values, depending on what it wants the thread to do.
200 */
201 int (*downconvert_worker)(struct ocfs2_lock_res *, int);
202
203 /*
Mark Fashehf625c972006-09-12 21:24:53 -0700204 * LOCK_TYPE_* flags which describe the specific requirements
205 * of a lock type. Descriptions of each individual flag follow.
206 */
207 int flags;
Mark Fashehccd979b2005-12-15 14:31:24 -0800208};
209
Mark Fashehf625c972006-09-12 21:24:53 -0700210/*
211 * Some locks want to "refresh" potentially stale data when a
212 * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
213 * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
214 * individual lockres l_flags member from the ast function. It is
215 * expected that the locking wrapper will clear the
216 * OCFS2_LOCK_NEEDS_REFRESH flag when done.
217 */
218#define LOCK_TYPE_REQUIRES_REFRESH 0x1
219
Mark Fashehb80fc012006-09-12 22:08:14 -0700220/*
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -0700221 * Indicate that a lock type makes use of the lock value block. The
222 * ->set_lvb lock type callback must be defined.
Mark Fashehb80fc012006-09-12 22:08:14 -0700223 */
224#define LOCK_TYPE_USES_LVB 0x2
225
Mark Fashehccd979b2005-12-15 14:31:24 -0800226static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
Mark Fasheh54a7e752006-09-12 21:49:13 -0700227 .get_osb = ocfs2_get_inode_osb,
Mark Fashehf625c972006-09-12 21:24:53 -0700228 .flags = 0,
Mark Fashehccd979b2005-12-15 14:31:24 -0800229};
230
Mark Fashehe63aecb62007-10-18 15:30:42 -0700231static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
Mark Fasheh54a7e752006-09-12 21:49:13 -0700232 .get_osb = ocfs2_get_inode_osb,
Mark Fasheh810d5ae2006-09-13 21:39:52 -0700233 .check_downconvert = ocfs2_check_meta_downconvert,
234 .set_lvb = ocfs2_set_meta_lvb,
Mark Fashehf1f54062007-10-18 15:13:59 -0700235 .downconvert_worker = ocfs2_data_convert_worker,
Mark Fashehb80fc012006-09-12 22:08:14 -0700236 .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
Mark Fashehccd979b2005-12-15 14:31:24 -0800237};
238
Mark Fashehccd979b2005-12-15 14:31:24 -0800239static struct ocfs2_lock_res_ops ocfs2_super_lops = {
Mark Fashehf625c972006-09-12 21:24:53 -0700240 .flags = LOCK_TYPE_REQUIRES_REFRESH,
Mark Fashehccd979b2005-12-15 14:31:24 -0800241};
242
243static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
Mark Fashehf625c972006-09-12 21:24:53 -0700244 .flags = 0,
Mark Fashehccd979b2005-12-15 14:31:24 -0800245};
246
Mark Fashehd680efe2006-09-08 14:14:34 -0700247static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
Mark Fasheh54a7e752006-09-12 21:49:13 -0700248 .get_osb = ocfs2_get_dentry_osb,
Mark Fashehd680efe2006-09-08 14:14:34 -0700249 .post_unlock = ocfs2_dentry_post_unlock,
Mark Fashehcc567d82006-09-13 21:52:21 -0700250 .downconvert_worker = ocfs2_dentry_convert_worker,
Mark Fashehf625c972006-09-12 21:24:53 -0700251 .flags = 0,
Mark Fashehd680efe2006-09-08 14:14:34 -0700252};
253
Tiger Yang50008632007-03-20 16:01:38 -0700254static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
255 .get_osb = ocfs2_get_inode_osb,
256 .flags = 0,
257};
258
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800259static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
260 .get_osb = ocfs2_get_file_osb,
261 .flags = 0,
262};
263
Jan Kara9e33d692008-08-25 19:56:50 +0200264static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
265 .set_lvb = ocfs2_set_qinfo_lvb,
266 .get_osb = ocfs2_get_qinfo_osb,
267 .flags = LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
268};
269
Mark Fashehccd979b2005-12-15 14:31:24 -0800270static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
271{
272 return lockres->l_type == OCFS2_LOCK_TYPE_META ||
Tiger Yang50008632007-03-20 16:01:38 -0700273 lockres->l_type == OCFS2_LOCK_TYPE_RW ||
274 lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
Mark Fashehccd979b2005-12-15 14:31:24 -0800275}
276
Mark Fashehccd979b2005-12-15 14:31:24 -0800277static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
278{
279 BUG_ON(!ocfs2_is_inode_lock(lockres));
280
281 return (struct inode *) lockres->l_priv;
282}
283
Mark Fashehd680efe2006-09-08 14:14:34 -0700284static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
285{
286 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
287
288 return (struct ocfs2_dentry_lock *)lockres->l_priv;
289}
290
Jan Kara9e33d692008-08-25 19:56:50 +0200291static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
292{
293 BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
294
295 return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
296}
297
Mark Fasheh54a7e752006-09-12 21:49:13 -0700298static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
299{
300 if (lockres->l_ops->get_osb)
301 return lockres->l_ops->get_osb(lockres);
302
303 return (struct ocfs2_super *)lockres->l_priv;
304}
305
Mark Fashehccd979b2005-12-15 14:31:24 -0800306static int ocfs2_lock_create(struct ocfs2_super *osb,
307 struct ocfs2_lock_res *lockres,
308 int level,
Joel Beckerbd3e7612008-02-01 12:14:57 -0800309 u32 dlm_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800310static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
311 int wanted);
312static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
313 struct ocfs2_lock_res *lockres,
314 int level);
315static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
316static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
317static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
318static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
319static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
320 struct ocfs2_lock_res *lockres);
321static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
322 int convert);
Sunil Mushranc74ff8b2009-02-03 12:37:14 -0800323#define ocfs2_log_dlm_error(_func, _err, _lockres) do { \
324 if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY) \
325 mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n", \
326 _err, _func, _lockres->l_name); \
327 else \
328 mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n", \
329 _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name, \
330 (unsigned int)ocfs2_get_dentry_lock_ino(_lockres)); \
Mark Fashehccd979b2005-12-15 14:31:24 -0800331} while (0)
Mark Fasheh34d024f2007-09-24 15:56:19 -0700332static int ocfs2_downconvert_thread(void *arg);
333static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
334 struct ocfs2_lock_res *lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700335static int ocfs2_inode_lock_update(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -0800336 struct buffer_head **bh);
337static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
338static inline int ocfs2_highest_compat_lock_level(int level);
Joel Beckerde551242008-02-01 14:45:08 -0800339static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
340 int new_level);
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800341static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
342 struct ocfs2_lock_res *lockres,
343 int new_level,
Joel Beckerde551242008-02-01 14:45:08 -0800344 int lvb,
345 unsigned int generation);
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800346static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
347 struct ocfs2_lock_res *lockres);
348static int ocfs2_cancel_convert(struct ocfs2_super *osb,
349 struct ocfs2_lock_res *lockres);
350
Mark Fashehccd979b2005-12-15 14:31:24 -0800351
Mark Fashehccd979b2005-12-15 14:31:24 -0800352static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
353 u64 blkno,
354 u32 generation,
355 char *name)
356{
357 int len;
358
359 mlog_entry_void();
360
361 BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
362
Mark Fashehb0697052006-03-03 10:24:33 -0800363 len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
364 ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
365 (long long)blkno, generation);
Mark Fashehccd979b2005-12-15 14:31:24 -0800366
367 BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
368
369 mlog(0, "built lock resource with name: %s\n", name);
370
371 mlog_exit_void();
372}
373
Ingo Molnar34af9462006-06-27 02:53:55 -0700374static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -0800375
376static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
377 struct ocfs2_dlm_debug *dlm_debug)
378{
379 mlog(0, "Add tracking for lockres %s\n", res->l_name);
380
381 spin_lock(&ocfs2_dlm_tracking_lock);
382 list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
383 spin_unlock(&ocfs2_dlm_tracking_lock);
384}
385
386static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
387{
388 spin_lock(&ocfs2_dlm_tracking_lock);
389 if (!list_empty(&res->l_debug_list))
390 list_del_init(&res->l_debug_list);
391 spin_unlock(&ocfs2_dlm_tracking_lock);
392}
393
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700394#ifdef CONFIG_OCFS2_FS_STATS
395static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
396{
397 res->l_lock_num_prmode = 0;
398 res->l_lock_num_prmode_failed = 0;
399 res->l_lock_total_prmode = 0;
400 res->l_lock_max_prmode = 0;
401 res->l_lock_num_exmode = 0;
402 res->l_lock_num_exmode_failed = 0;
403 res->l_lock_total_exmode = 0;
404 res->l_lock_max_exmode = 0;
405 res->l_lock_refresh = 0;
406}
407
408static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
409 struct ocfs2_mask_waiter *mw, int ret)
410{
411 unsigned long long *num, *sum;
412 unsigned int *max, *failed;
413 struct timespec ts = current_kernel_time();
414 unsigned long long time = timespec_to_ns(&ts) - mw->mw_lock_start;
415
416 if (level == LKM_PRMODE) {
417 num = &res->l_lock_num_prmode;
418 sum = &res->l_lock_total_prmode;
419 max = &res->l_lock_max_prmode;
420 failed = &res->l_lock_num_prmode_failed;
421 } else if (level == LKM_EXMODE) {
422 num = &res->l_lock_num_exmode;
423 sum = &res->l_lock_total_exmode;
424 max = &res->l_lock_max_exmode;
425 failed = &res->l_lock_num_exmode_failed;
426 } else
427 return;
428
429 (*num)++;
430 (*sum) += time;
431 if (time > *max)
432 *max = time;
433 if (ret)
434 (*failed)++;
435}
436
437static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
438{
439 lockres->l_lock_refresh++;
440}
441
442static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
443{
444 struct timespec ts = current_kernel_time();
445 mw->mw_lock_start = timespec_to_ns(&ts);
446}
447#else
448static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
449{
450}
451static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
452 int level, struct ocfs2_mask_waiter *mw, int ret)
453{
454}
455static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
456{
457}
458static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
459{
460}
461#endif
462
Mark Fashehccd979b2005-12-15 14:31:24 -0800463static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
464 struct ocfs2_lock_res *res,
465 enum ocfs2_lock_type type,
Mark Fashehccd979b2005-12-15 14:31:24 -0800466 struct ocfs2_lock_res_ops *ops,
467 void *priv)
468{
Mark Fashehccd979b2005-12-15 14:31:24 -0800469 res->l_type = type;
470 res->l_ops = ops;
471 res->l_priv = priv;
472
Joel Beckerbd3e7612008-02-01 12:14:57 -0800473 res->l_level = DLM_LOCK_IV;
474 res->l_requested = DLM_LOCK_IV;
475 res->l_blocking = DLM_LOCK_IV;
Mark Fashehccd979b2005-12-15 14:31:24 -0800476 res->l_action = OCFS2_AST_INVALID;
477 res->l_unlock_action = OCFS2_UNLOCK_INVALID;
478
479 res->l_flags = OCFS2_LOCK_INITIALIZED;
480
481 ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
Sunil Mushran8ddb7b02008-05-13 13:45:15 -0700482
483 ocfs2_init_lock_stats(res);
Mark Fashehccd979b2005-12-15 14:31:24 -0800484}
485
486void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
487{
488 /* This also clears out the lock status block */
489 memset(res, 0, sizeof(struct ocfs2_lock_res));
490 spin_lock_init(&res->l_lock);
491 init_waitqueue_head(&res->l_event);
492 INIT_LIST_HEAD(&res->l_blocked_list);
493 INIT_LIST_HEAD(&res->l_mask_waiters);
494}
495
496void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
497 enum ocfs2_lock_type type,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700498 unsigned int generation,
Mark Fashehccd979b2005-12-15 14:31:24 -0800499 struct inode *inode)
500{
501 struct ocfs2_lock_res_ops *ops;
502
503 switch(type) {
504 case OCFS2_LOCK_TYPE_RW:
505 ops = &ocfs2_inode_rw_lops;
506 break;
507 case OCFS2_LOCK_TYPE_META:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700508 ops = &ocfs2_inode_inode_lops;
Mark Fashehccd979b2005-12-15 14:31:24 -0800509 break;
Tiger Yang50008632007-03-20 16:01:38 -0700510 case OCFS2_LOCK_TYPE_OPEN:
511 ops = &ocfs2_inode_open_lops;
512 break;
Mark Fashehccd979b2005-12-15 14:31:24 -0800513 default:
514 mlog_bug_on_msg(1, "type: %d\n", type);
515 ops = NULL; /* thanks, gcc */
516 break;
517 };
518
Mark Fashehd680efe2006-09-08 14:14:34 -0700519 ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700520 generation, res->l_name);
Mark Fashehd680efe2006-09-08 14:14:34 -0700521 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
522}
523
Mark Fasheh54a7e752006-09-12 21:49:13 -0700524static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
525{
526 struct inode *inode = ocfs2_lock_res_inode(lockres);
527
528 return OCFS2_SB(inode->i_sb);
529}
530
Jan Kara9e33d692008-08-25 19:56:50 +0200531static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
532{
533 struct ocfs2_mem_dqinfo *info = lockres->l_priv;
534
535 return OCFS2_SB(info->dqi_gi.dqi_sb);
536}
537
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800538static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
539{
540 struct ocfs2_file_private *fp = lockres->l_priv;
541
542 return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
543}
544
Mark Fashehd680efe2006-09-08 14:14:34 -0700545static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
546{
547 __be64 inode_blkno_be;
548
549 memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
550 sizeof(__be64));
551
552 return be64_to_cpu(inode_blkno_be);
553}
554
Mark Fasheh54a7e752006-09-12 21:49:13 -0700555static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
556{
557 struct ocfs2_dentry_lock *dl = lockres->l_priv;
558
559 return OCFS2_SB(dl->dl_inode->i_sb);
560}
561
Mark Fashehd680efe2006-09-08 14:14:34 -0700562void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
563 u64 parent, struct inode *inode)
564{
565 int len;
566 u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
567 __be64 inode_blkno_be = cpu_to_be64(inode_blkno);
568 struct ocfs2_lock_res *lockres = &dl->dl_lockres;
569
570 ocfs2_lock_res_init_once(lockres);
571
572 /*
573 * Unfortunately, the standard lock naming scheme won't work
574 * here because we have two 16 byte values to use. Instead,
575 * we'll stuff the inode number as a binary value. We still
576 * want error prints to show something without garbling the
577 * display, so drop a null byte in there before the inode
578 * number. A future version of OCFS2 will likely use all
579 * binary lock names. The stringified names have been a
580 * tremendous aid in debugging, but now that the debugfs
581 * interface exists, we can mangle things there if need be.
582 *
583 * NOTE: We also drop the standard "pad" value (the total lock
584 * name size stays the same though - the last part is all
585 * zeros due to the memset in ocfs2_lock_res_init_once()
586 */
587 len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
588 "%c%016llx",
589 ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
590 (long long)parent);
591
592 BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
593
594 memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
595 sizeof(__be64));
596
597 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
598 OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
599 dl);
Mark Fashehccd979b2005-12-15 14:31:24 -0800600}
601
602static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
603 struct ocfs2_super *osb)
604{
605 /* Superblock lockres doesn't come from a slab so we call init
606 * once on it manually. */
607 ocfs2_lock_res_init_once(res);
Mark Fashehd680efe2006-09-08 14:14:34 -0700608 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
609 0, res->l_name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800610 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
Mark Fashehccd979b2005-12-15 14:31:24 -0800611 &ocfs2_super_lops, osb);
612}
613
614static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
615 struct ocfs2_super *osb)
616{
617 /* Rename lockres doesn't come from a slab so we call init
618 * once on it manually. */
619 ocfs2_lock_res_init_once(res);
Mark Fashehd680efe2006-09-08 14:14:34 -0700620 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
621 ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
Mark Fashehccd979b2005-12-15 14:31:24 -0800622 &ocfs2_rename_lops, osb);
623}
624
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800625void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
626 struct ocfs2_file_private *fp)
627{
628 struct inode *inode = fp->fp_file->f_mapping->host;
629 struct ocfs2_inode_info *oi = OCFS2_I(inode);
630
631 ocfs2_lock_res_init_once(lockres);
632 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
633 inode->i_generation, lockres->l_name);
634 ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
635 OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
636 fp);
637 lockres->l_flags |= OCFS2_LOCK_NOCACHE;
638}
639
Jan Kara9e33d692008-08-25 19:56:50 +0200640void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
641 struct ocfs2_mem_dqinfo *info)
642{
643 ocfs2_lock_res_init_once(lockres);
644 ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
645 0, lockres->l_name);
646 ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
647 OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
648 info);
649}
650
Mark Fashehccd979b2005-12-15 14:31:24 -0800651void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
652{
653 mlog_entry_void();
654
655 if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
656 return;
657
658 ocfs2_remove_lockres_tracking(res);
659
660 mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
661 "Lockres %s is on the blocked list\n",
662 res->l_name);
663 mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
664 "Lockres %s has mask waiters pending\n",
665 res->l_name);
666 mlog_bug_on_msg(spin_is_locked(&res->l_lock),
667 "Lockres %s is locked\n",
668 res->l_name);
669 mlog_bug_on_msg(res->l_ro_holders,
670 "Lockres %s has %u ro holders\n",
671 res->l_name, res->l_ro_holders);
672 mlog_bug_on_msg(res->l_ex_holders,
673 "Lockres %s has %u ex holders\n",
674 res->l_name, res->l_ex_holders);
675
676 /* Need to clear out the lock status block for the dlm */
677 memset(&res->l_lksb, 0, sizeof(res->l_lksb));
678
679 res->l_flags = 0UL;
680 mlog_exit_void();
681}
682
683static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
684 int level)
685{
686 mlog_entry_void();
687
688 BUG_ON(!lockres);
689
690 switch(level) {
Joel Beckerbd3e7612008-02-01 12:14:57 -0800691 case DLM_LOCK_EX:
Mark Fashehccd979b2005-12-15 14:31:24 -0800692 lockres->l_ex_holders++;
693 break;
Joel Beckerbd3e7612008-02-01 12:14:57 -0800694 case DLM_LOCK_PR:
Mark Fashehccd979b2005-12-15 14:31:24 -0800695 lockres->l_ro_holders++;
696 break;
697 default:
698 BUG();
699 }
700
701 mlog_exit_void();
702}
703
704static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
705 int level)
706{
707 mlog_entry_void();
708
709 BUG_ON(!lockres);
710
711 switch(level) {
Joel Beckerbd3e7612008-02-01 12:14:57 -0800712 case DLM_LOCK_EX:
Mark Fashehccd979b2005-12-15 14:31:24 -0800713 BUG_ON(!lockres->l_ex_holders);
714 lockres->l_ex_holders--;
715 break;
Joel Beckerbd3e7612008-02-01 12:14:57 -0800716 case DLM_LOCK_PR:
Mark Fashehccd979b2005-12-15 14:31:24 -0800717 BUG_ON(!lockres->l_ro_holders);
718 lockres->l_ro_holders--;
719 break;
720 default:
721 BUG();
722 }
723 mlog_exit_void();
724}
725
726/* WARNING: This function lives in a world where the only three lock
727 * levels are EX, PR, and NL. It *will* have to be adjusted when more
728 * lock types are added. */
729static inline int ocfs2_highest_compat_lock_level(int level)
730{
Joel Beckerbd3e7612008-02-01 12:14:57 -0800731 int new_level = DLM_LOCK_EX;
Mark Fashehccd979b2005-12-15 14:31:24 -0800732
Joel Beckerbd3e7612008-02-01 12:14:57 -0800733 if (level == DLM_LOCK_EX)
734 new_level = DLM_LOCK_NL;
735 else if (level == DLM_LOCK_PR)
736 new_level = DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -0800737 return new_level;
738}
739
740static void lockres_set_flags(struct ocfs2_lock_res *lockres,
741 unsigned long newflags)
742{
Christoph Hellwig800deef2007-05-17 16:03:13 +0200743 struct ocfs2_mask_waiter *mw, *tmp;
Mark Fashehccd979b2005-12-15 14:31:24 -0800744
745 assert_spin_locked(&lockres->l_lock);
746
747 lockres->l_flags = newflags;
748
Christoph Hellwig800deef2007-05-17 16:03:13 +0200749 list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800750 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
751 continue;
752
753 list_del_init(&mw->mw_item);
754 mw->mw_status = 0;
755 complete(&mw->mw_complete);
756 }
757}
758static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
759{
760 lockres_set_flags(lockres, lockres->l_flags | or);
761}
762static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
763 unsigned long clear)
764{
765 lockres_set_flags(lockres, lockres->l_flags & ~clear);
766}
767
768static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
769{
770 mlog_entry_void();
771
772 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
773 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
774 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
Joel Beckerbd3e7612008-02-01 12:14:57 -0800775 BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800776
777 lockres->l_level = lockres->l_requested;
778 if (lockres->l_level <=
779 ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
Joel Beckerbd3e7612008-02-01 12:14:57 -0800780 lockres->l_blocking = DLM_LOCK_NL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800781 lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
782 }
783 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
784
785 mlog_exit_void();
786}
787
788static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
789{
790 mlog_entry_void();
791
792 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
793 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
794
795 /* Convert from RO to EX doesn't really need anything as our
796 * information is already up to data. Convert from NL to
797 * *anything* however should mark ourselves as needing an
798 * update */
Joel Beckerbd3e7612008-02-01 12:14:57 -0800799 if (lockres->l_level == DLM_LOCK_NL &&
Mark Fashehf625c972006-09-12 21:24:53 -0700800 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
Mark Fashehccd979b2005-12-15 14:31:24 -0800801 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
802
803 lockres->l_level = lockres->l_requested;
804 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
805
806 mlog_exit_void();
807}
808
809static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
810{
811 mlog_entry_void();
812
Roel Kluin3cf0c502007-10-27 00:20:36 +0200813 BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
Mark Fashehccd979b2005-12-15 14:31:24 -0800814 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
815
Joel Beckerbd3e7612008-02-01 12:14:57 -0800816 if (lockres->l_requested > DLM_LOCK_NL &&
Mark Fashehf625c972006-09-12 21:24:53 -0700817 !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
818 lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
Mark Fashehccd979b2005-12-15 14:31:24 -0800819 lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
820
821 lockres->l_level = lockres->l_requested;
822 lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
823 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
824
825 mlog_exit_void();
826}
827
Mark Fashehccd979b2005-12-15 14:31:24 -0800828static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
829 int level)
830{
831 int needs_downconvert = 0;
832 mlog_entry_void();
833
834 assert_spin_locked(&lockres->l_lock);
835
836 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
837
838 if (level > lockres->l_blocking) {
839 /* only schedule a downconvert if we haven't already scheduled
840 * one that goes low enough to satisfy the level we're
841 * blocking. this also catches the case where we get
842 * duplicate BASTs */
843 if (ocfs2_highest_compat_lock_level(level) <
844 ocfs2_highest_compat_lock_level(lockres->l_blocking))
845 needs_downconvert = 1;
846
847 lockres->l_blocking = level;
848 }
849
850 mlog_exit(needs_downconvert);
851 return needs_downconvert;
852}
853
Joel Beckerde551242008-02-01 14:45:08 -0800854/*
855 * OCFS2_LOCK_PENDING and l_pending_gen.
856 *
857 * Why does OCFS2_LOCK_PENDING exist? To close a race between setting
858 * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock(). See ocfs2_unblock_lock()
859 * for more details on the race.
860 *
861 * OCFS2_LOCK_PENDING closes the race quite nicely. However, it introduces
862 * a race on itself. In o2dlm, we can get the ast before ocfs2_dlm_lock()
863 * returns. The ast clears OCFS2_LOCK_BUSY, and must therefore clear
864 * OCFS2_LOCK_PENDING at the same time. When ocfs2_dlm_lock() returns,
865 * the caller is going to try to clear PENDING again. If nothing else is
866 * happening, __lockres_clear_pending() sees PENDING is unset and does
867 * nothing.
868 *
869 * But what if another path (eg downconvert thread) has just started a
870 * new locking action? The other path has re-set PENDING. Our path
871 * cannot clear PENDING, because that will re-open the original race
872 * window.
873 *
874 * [Example]
875 *
876 * ocfs2_meta_lock()
877 * ocfs2_cluster_lock()
878 * set BUSY
879 * set PENDING
880 * drop l_lock
881 * ocfs2_dlm_lock()
882 * ocfs2_locking_ast() ocfs2_downconvert_thread()
883 * clear PENDING ocfs2_unblock_lock()
884 * take_l_lock
885 * !BUSY
886 * ocfs2_prepare_downconvert()
887 * set BUSY
888 * set PENDING
889 * drop l_lock
890 * take l_lock
891 * clear PENDING
892 * drop l_lock
893 * <window>
894 * ocfs2_dlm_lock()
895 *
896 * So as you can see, we now have a window where l_lock is not held,
897 * PENDING is not set, and ocfs2_dlm_lock() has not been called.
898 *
899 * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
900 * set by ocfs2_prepare_downconvert(). That wasn't nice.
901 *
902 * To solve this we introduce l_pending_gen. A call to
903 * lockres_clear_pending() will only do so when it is passed a generation
904 * number that matches the lockres. lockres_set_pending() will return the
905 * current generation number. When ocfs2_cluster_lock() goes to clear
906 * PENDING, it passes the generation it got from set_pending(). In our
907 * example above, the generation numbers will *not* match. Thus,
908 * ocfs2_cluster_lock() will not clear the PENDING set by
909 * ocfs2_prepare_downconvert().
910 */
911
912/* Unlocked version for ocfs2_locking_ast() */
913static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
914 unsigned int generation,
915 struct ocfs2_super *osb)
916{
917 assert_spin_locked(&lockres->l_lock);
918
919 /*
920 * The ast and locking functions can race us here. The winner
921 * will clear pending, the loser will not.
922 */
923 if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
924 (lockres->l_pending_gen != generation))
925 return;
926
927 lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
928 lockres->l_pending_gen++;
929
930 /*
931 * The downconvert thread may have skipped us because we
932 * were PENDING. Wake it up.
933 */
934 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
935 ocfs2_wake_downconvert_thread(osb);
936}
937
938/* Locked version for callers of ocfs2_dlm_lock() */
939static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
940 unsigned int generation,
941 struct ocfs2_super *osb)
942{
943 unsigned long flags;
944
945 spin_lock_irqsave(&lockres->l_lock, flags);
946 __lockres_clear_pending(lockres, generation, osb);
947 spin_unlock_irqrestore(&lockres->l_lock, flags);
948}
949
950static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
951{
952 assert_spin_locked(&lockres->l_lock);
953 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
954
955 lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
956
957 return lockres->l_pending_gen;
958}
959
960
Mark Fashehaa2623a2006-09-12 21:58:23 -0700961static void ocfs2_blocking_ast(void *opaque, int level)
Mark Fashehccd979b2005-12-15 14:31:24 -0800962{
Mark Fashehaa2623a2006-09-12 21:58:23 -0700963 struct ocfs2_lock_res *lockres = opaque;
964 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -0800965 int needs_downconvert;
966 unsigned long flags;
967
Joel Beckerbd3e7612008-02-01 12:14:57 -0800968 BUG_ON(level <= DLM_LOCK_NL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800969
Mark Fashehaa2623a2006-09-12 21:58:23 -0700970 mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n",
971 lockres->l_name, level, lockres->l_level,
972 ocfs2_lock_type_string(lockres->l_type));
973
Mark Fashehcf8e06f2007-12-20 16:43:10 -0800974 /*
975 * We can skip the bast for locks which don't enable caching -
976 * they'll be dropped at the earliest possible time anyway.
977 */
978 if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
979 return;
980
Mark Fashehccd979b2005-12-15 14:31:24 -0800981 spin_lock_irqsave(&lockres->l_lock, flags);
982 needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
983 if (needs_downconvert)
984 ocfs2_schedule_blocked_lock(osb, lockres);
985 spin_unlock_irqrestore(&lockres->l_lock, flags);
986
Mark Fashehd680efe2006-09-08 14:14:34 -0700987 wake_up(&lockres->l_event);
988
Mark Fasheh34d024f2007-09-24 15:56:19 -0700989 ocfs2_wake_downconvert_thread(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800990}
991
Mark Fashehe92d57d2006-09-12 21:34:35 -0700992static void ocfs2_locking_ast(void *opaque)
Mark Fashehccd979b2005-12-15 14:31:24 -0800993{
Mark Fashehe92d57d2006-09-12 21:34:35 -0700994 struct ocfs2_lock_res *lockres = opaque;
Joel Beckerde551242008-02-01 14:45:08 -0800995 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -0800996 unsigned long flags;
David Teigland1693a5c2008-01-30 16:52:53 -0800997 int status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800998
999 spin_lock_irqsave(&lockres->l_lock, flags);
1000
David Teigland1693a5c2008-01-30 16:52:53 -08001001 status = ocfs2_dlm_lock_status(&lockres->l_lksb);
1002
1003 if (status == -EAGAIN) {
1004 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1005 goto out;
1006 }
1007
1008 if (status) {
Joel Becker8f2c9c12008-02-01 12:16:57 -08001009 mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
David Teigland1693a5c2008-01-30 16:52:53 -08001010 lockres->l_name, status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001011 spin_unlock_irqrestore(&lockres->l_lock, flags);
1012 return;
1013 }
1014
1015 switch(lockres->l_action) {
1016 case OCFS2_AST_ATTACH:
1017 ocfs2_generic_handle_attach_action(lockres);
Mark Fashehe92d57d2006-09-12 21:34:35 -07001018 lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001019 break;
1020 case OCFS2_AST_CONVERT:
1021 ocfs2_generic_handle_convert_action(lockres);
1022 break;
1023 case OCFS2_AST_DOWNCONVERT:
1024 ocfs2_generic_handle_downconvert_action(lockres);
1025 break;
1026 default:
Mark Fashehe92d57d2006-09-12 21:34:35 -07001027 mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u "
1028 "lockres flags = 0x%lx, unlock action: %u\n",
1029 lockres->l_name, lockres->l_action, lockres->l_flags,
1030 lockres->l_unlock_action);
Mark Fashehccd979b2005-12-15 14:31:24 -08001031 BUG();
1032 }
David Teigland1693a5c2008-01-30 16:52:53 -08001033out:
Mark Fashehccd979b2005-12-15 14:31:24 -08001034 /* set it to something invalid so if we get called again we
1035 * can catch it. */
1036 lockres->l_action = OCFS2_AST_INVALID;
Mark Fashehccd979b2005-12-15 14:31:24 -08001037
Joel Beckerde551242008-02-01 14:45:08 -08001038 /* Did we try to cancel this lock? Clear that state */
1039 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
1040 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1041
1042 /*
1043 * We may have beaten the locking functions here. We certainly
1044 * know that dlm_lock() has been called :-)
1045 * Because we can't have two lock calls in flight at once, we
1046 * can use lockres->l_pending_gen.
1047 */
1048 __lockres_clear_pending(lockres, lockres->l_pending_gen, osb);
1049
Mark Fashehccd979b2005-12-15 14:31:24 -08001050 wake_up(&lockres->l_event);
Mark Fashehd680efe2006-09-08 14:14:34 -07001051 spin_unlock_irqrestore(&lockres->l_lock, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001052}
1053
Mark Fashehccd979b2005-12-15 14:31:24 -08001054static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
1055 int convert)
1056{
1057 unsigned long flags;
1058
1059 mlog_entry_void();
1060 spin_lock_irqsave(&lockres->l_lock, flags);
1061 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
1062 if (convert)
1063 lockres->l_action = OCFS2_AST_INVALID;
1064 else
1065 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
1066 spin_unlock_irqrestore(&lockres->l_lock, flags);
1067
1068 wake_up(&lockres->l_event);
1069 mlog_exit_void();
1070}
1071
1072/* Note: If we detect another process working on the lock (i.e.,
1073 * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
1074 * to do the right thing in that case.
1075 */
1076static int ocfs2_lock_create(struct ocfs2_super *osb,
1077 struct ocfs2_lock_res *lockres,
1078 int level,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001079 u32 dlm_flags)
Mark Fashehccd979b2005-12-15 14:31:24 -08001080{
1081 int ret = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001082 unsigned long flags;
Joel Beckerde551242008-02-01 14:45:08 -08001083 unsigned int gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08001084
1085 mlog_entry_void();
1086
Joel Beckerbd3e7612008-02-01 12:14:57 -08001087 mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
Mark Fashehccd979b2005-12-15 14:31:24 -08001088 dlm_flags);
1089
1090 spin_lock_irqsave(&lockres->l_lock, flags);
1091 if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
1092 (lockres->l_flags & OCFS2_LOCK_BUSY)) {
1093 spin_unlock_irqrestore(&lockres->l_lock, flags);
1094 goto bail;
1095 }
1096
1097 lockres->l_action = OCFS2_AST_ATTACH;
1098 lockres->l_requested = level;
1099 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
Joel Beckerde551242008-02-01 14:45:08 -08001100 gen = lockres_set_pending(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001101 spin_unlock_irqrestore(&lockres->l_lock, flags);
1102
Joel Becker4670c462008-02-01 14:39:35 -08001103 ret = ocfs2_dlm_lock(osb->cconn,
Joel Becker7431cd72008-02-01 12:15:37 -08001104 level,
1105 &lockres->l_lksb,
1106 dlm_flags,
1107 lockres->l_name,
1108 OCFS2_LOCK_ID_MAX_LEN - 1,
1109 lockres);
Joel Beckerde551242008-02-01 14:45:08 -08001110 lockres_clear_pending(lockres, gen, osb);
Joel Becker7431cd72008-02-01 12:15:37 -08001111 if (ret) {
1112 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001113 ocfs2_recover_from_dlm_error(lockres, 1);
1114 }
1115
Joel Becker7431cd72008-02-01 12:15:37 -08001116 mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
Mark Fashehccd979b2005-12-15 14:31:24 -08001117
1118bail:
1119 mlog_exit(ret);
1120 return ret;
1121}
1122
1123static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
1124 int flag)
1125{
1126 unsigned long flags;
1127 int ret;
1128
1129 spin_lock_irqsave(&lockres->l_lock, flags);
1130 ret = lockres->l_flags & flag;
1131 spin_unlock_irqrestore(&lockres->l_lock, flags);
1132
1133 return ret;
1134}
1135
1136static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
1137
1138{
1139 wait_event(lockres->l_event,
1140 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
1141}
1142
1143static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
1144
1145{
1146 wait_event(lockres->l_event,
1147 !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
1148}
1149
1150/* predict what lock level we'll be dropping down to on behalf
1151 * of another node, and return true if the currently wanted
1152 * level will be compatible with it. */
1153static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
1154 int wanted)
1155{
1156 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
1157
1158 return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
1159}
1160
1161static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
1162{
1163 INIT_LIST_HEAD(&mw->mw_item);
1164 init_completion(&mw->mw_complete);
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07001165 ocfs2_init_start_time(mw);
Mark Fashehccd979b2005-12-15 14:31:24 -08001166}
1167
1168static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
1169{
1170 wait_for_completion(&mw->mw_complete);
1171 /* Re-arm the completion in case we want to wait on it again */
1172 INIT_COMPLETION(mw->mw_complete);
1173 return mw->mw_status;
1174}
1175
1176static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
1177 struct ocfs2_mask_waiter *mw,
1178 unsigned long mask,
1179 unsigned long goal)
1180{
1181 BUG_ON(!list_empty(&mw->mw_item));
1182
1183 assert_spin_locked(&lockres->l_lock);
1184
1185 list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
1186 mw->mw_mask = mask;
1187 mw->mw_goal = goal;
1188}
1189
1190/* returns 0 if the mw that was removed was already satisfied, -EBUSY
1191 * if the mask still hadn't reached its goal */
1192static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
1193 struct ocfs2_mask_waiter *mw)
1194{
1195 unsigned long flags;
1196 int ret = 0;
1197
1198 spin_lock_irqsave(&lockres->l_lock, flags);
1199 if (!list_empty(&mw->mw_item)) {
1200 if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
1201 ret = -EBUSY;
1202
1203 list_del_init(&mw->mw_item);
1204 init_completion(&mw->mw_complete);
1205 }
1206 spin_unlock_irqrestore(&lockres->l_lock, flags);
1207
1208 return ret;
1209
1210}
1211
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001212static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
1213 struct ocfs2_lock_res *lockres)
1214{
1215 int ret;
1216
1217 ret = wait_for_completion_interruptible(&mw->mw_complete);
1218 if (ret)
1219 lockres_remove_mask_waiter(lockres, mw);
1220 else
1221 ret = mw->mw_status;
1222 /* Re-arm the completion in case we want to wait on it again */
1223 INIT_COMPLETION(mw->mw_complete);
1224 return ret;
1225}
1226
Mark Fashehccd979b2005-12-15 14:31:24 -08001227static int ocfs2_cluster_lock(struct ocfs2_super *osb,
1228 struct ocfs2_lock_res *lockres,
1229 int level,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001230 u32 lkm_flags,
Mark Fashehccd979b2005-12-15 14:31:24 -08001231 int arg_flags)
1232{
1233 struct ocfs2_mask_waiter mw;
Mark Fashehccd979b2005-12-15 14:31:24 -08001234 int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
1235 int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
1236 unsigned long flags;
Joel Beckerde551242008-02-01 14:45:08 -08001237 unsigned int gen;
David Teigland1693a5c2008-01-30 16:52:53 -08001238 int noqueue_attempted = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001239
1240 mlog_entry_void();
1241
1242 ocfs2_init_mask_waiter(&mw);
1243
Mark Fashehb80fc012006-09-12 22:08:14 -07001244 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
Joel Beckerbd3e7612008-02-01 12:14:57 -08001245 lkm_flags |= DLM_LKF_VALBLK;
Mark Fashehb80fc012006-09-12 22:08:14 -07001246
Mark Fashehccd979b2005-12-15 14:31:24 -08001247again:
1248 wait = 0;
1249
1250 if (catch_signals && signal_pending(current)) {
1251 ret = -ERESTARTSYS;
1252 goto out;
1253 }
1254
1255 spin_lock_irqsave(&lockres->l_lock, flags);
1256
1257 mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
1258 "Cluster lock called on freeing lockres %s! flags "
1259 "0x%lx\n", lockres->l_name, lockres->l_flags);
1260
1261 /* We only compare against the currently granted level
1262 * here. If the lock is blocked waiting on a downconvert,
1263 * we'll get caught below. */
1264 if (lockres->l_flags & OCFS2_LOCK_BUSY &&
1265 level > lockres->l_level) {
1266 /* is someone sitting in dlm_lock? If so, wait on
1267 * them. */
1268 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1269 wait = 1;
1270 goto unlock;
1271 }
1272
Mark Fashehccd979b2005-12-15 14:31:24 -08001273 if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
1274 !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
1275 /* is the lock is currently blocked on behalf of
1276 * another node */
1277 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
1278 wait = 1;
1279 goto unlock;
1280 }
1281
1282 if (level > lockres->l_level) {
David Teigland1693a5c2008-01-30 16:52:53 -08001283 if (noqueue_attempted > 0) {
1284 ret = -EAGAIN;
1285 goto unlock;
1286 }
1287 if (lkm_flags & DLM_LKF_NOQUEUE)
1288 noqueue_attempted = 1;
1289
Mark Fashehccd979b2005-12-15 14:31:24 -08001290 if (lockres->l_action != OCFS2_AST_INVALID)
1291 mlog(ML_ERROR, "lockres %s has action %u pending\n",
1292 lockres->l_name, lockres->l_action);
1293
Mark Fasheh019d1b22007-10-05 12:09:05 -07001294 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1295 lockres->l_action = OCFS2_AST_ATTACH;
Joel Beckerbd3e7612008-02-01 12:14:57 -08001296 lkm_flags &= ~DLM_LKF_CONVERT;
Mark Fasheh019d1b22007-10-05 12:09:05 -07001297 } else {
1298 lockres->l_action = OCFS2_AST_CONVERT;
Joel Beckerbd3e7612008-02-01 12:14:57 -08001299 lkm_flags |= DLM_LKF_CONVERT;
Mark Fasheh019d1b22007-10-05 12:09:05 -07001300 }
1301
Mark Fashehccd979b2005-12-15 14:31:24 -08001302 lockres->l_requested = level;
1303 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
Joel Beckerde551242008-02-01 14:45:08 -08001304 gen = lockres_set_pending(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001305 spin_unlock_irqrestore(&lockres->l_lock, flags);
1306
Joel Beckerbd3e7612008-02-01 12:14:57 -08001307 BUG_ON(level == DLM_LOCK_IV);
1308 BUG_ON(level == DLM_LOCK_NL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001309
1310 mlog(0, "lock %s, convert from %d to level = %d\n",
1311 lockres->l_name, lockres->l_level, level);
1312
1313 /* call dlm_lock to upgrade lock now */
Joel Becker4670c462008-02-01 14:39:35 -08001314 ret = ocfs2_dlm_lock(osb->cconn,
Joel Becker7431cd72008-02-01 12:15:37 -08001315 level,
1316 &lockres->l_lksb,
1317 lkm_flags,
1318 lockres->l_name,
1319 OCFS2_LOCK_ID_MAX_LEN - 1,
1320 lockres);
Joel Beckerde551242008-02-01 14:45:08 -08001321 lockres_clear_pending(lockres, gen, osb);
Joel Becker7431cd72008-02-01 12:15:37 -08001322 if (ret) {
1323 if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
1324 (ret != -EAGAIN)) {
Joel Becker24ef1812008-01-29 17:37:32 -08001325 ocfs2_log_dlm_error("ocfs2_dlm_lock",
Joel Becker7431cd72008-02-01 12:15:37 -08001326 ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001327 }
1328 ocfs2_recover_from_dlm_error(lockres, 1);
1329 goto out;
1330 }
1331
Coly Li73ac36e2009-01-07 18:09:16 -08001332 mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
Mark Fashehccd979b2005-12-15 14:31:24 -08001333 lockres->l_name);
1334
1335 /* At this point we've gone inside the dlm and need to
1336 * complete our work regardless. */
1337 catch_signals = 0;
1338
1339 /* wait for busy to clear and carry on */
1340 goto again;
1341 }
1342
1343 /* Ok, if we get here then we're good to go. */
1344 ocfs2_inc_holders(lockres, level);
1345
1346 ret = 0;
1347unlock:
1348 spin_unlock_irqrestore(&lockres->l_lock, flags);
1349out:
1350 /*
1351 * This is helping work around a lock inversion between the page lock
1352 * and dlm locks. One path holds the page lock while calling aops
1353 * which block acquiring dlm locks. The voting thread holds dlm
1354 * locks while acquiring page locks while down converting data locks.
1355 * This block is helping an aop path notice the inversion and back
1356 * off to unlock its page lock before trying the dlm lock again.
1357 */
1358 if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
1359 mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
1360 wait = 0;
1361 if (lockres_remove_mask_waiter(lockres, &mw))
1362 ret = -EAGAIN;
1363 else
1364 goto again;
1365 }
1366 if (wait) {
1367 ret = ocfs2_wait_for_mask(&mw);
1368 if (ret == 0)
1369 goto again;
1370 mlog_errno(ret);
1371 }
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07001372 ocfs2_update_lock_stats(lockres, level, &mw, ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001373
1374 mlog_exit(ret);
1375 return ret;
1376}
1377
1378static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
1379 struct ocfs2_lock_res *lockres,
1380 int level)
1381{
1382 unsigned long flags;
1383
1384 mlog_entry_void();
1385 spin_lock_irqsave(&lockres->l_lock, flags);
1386 ocfs2_dec_holders(lockres, level);
Mark Fasheh34d024f2007-09-24 15:56:19 -07001387 ocfs2_downconvert_on_unlock(osb, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001388 spin_unlock_irqrestore(&lockres->l_lock, flags);
1389 mlog_exit_void();
1390}
1391
Adrian Bunkda661162006-11-20 03:24:28 +01001392static int ocfs2_create_new_lock(struct ocfs2_super *osb,
1393 struct ocfs2_lock_res *lockres,
1394 int ex,
1395 int local)
Mark Fashehccd979b2005-12-15 14:31:24 -08001396{
Joel Beckerbd3e7612008-02-01 12:14:57 -08001397 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08001398 unsigned long flags;
Joel Beckerbd3e7612008-02-01 12:14:57 -08001399 u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001400
1401 spin_lock_irqsave(&lockres->l_lock, flags);
1402 BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
1403 lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
1404 spin_unlock_irqrestore(&lockres->l_lock, flags);
1405
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001406 return ocfs2_lock_create(osb, lockres, level, lkm_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001407}
1408
1409/* Grants us an EX lock on the data and metadata resources, skipping
1410 * the normal cluster directory lookup. Use this ONLY on newly created
1411 * inodes which other nodes can't possibly see, and which haven't been
1412 * hashed in the inode hash yet. This can give us a good performance
1413 * increase as it'll skip the network broadcast normally associated
1414 * with creating a new lock resource. */
1415int ocfs2_create_new_inode_locks(struct inode *inode)
1416{
1417 int ret;
Mark Fashehd680efe2006-09-08 14:14:34 -07001418 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001419
1420 BUG_ON(!inode);
1421 BUG_ON(!ocfs2_inode_is_new(inode));
1422
1423 mlog_entry_void();
1424
Mark Fashehb0697052006-03-03 10:24:33 -08001425 mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001426
1427 /* NOTE: That we don't increment any of the holder counts, nor
1428 * do we add anything to a journal handle. Since this is
1429 * supposed to be a new inode which the cluster doesn't know
1430 * about yet, there is no need to. As far as the LVB handling
1431 * is concerned, this is basically like acquiring an EX lock
1432 * on a resource which has an invalid one -- we'll set it
1433 * valid when we release the EX. */
1434
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001435 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001436 if (ret) {
1437 mlog_errno(ret);
1438 goto bail;
1439 }
1440
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001441 /*
Joel Beckerbd3e7612008-02-01 12:14:57 -08001442 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001443 * don't use a generation in their lock names.
1444 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001445 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001446 if (ret) {
1447 mlog_errno(ret);
1448 goto bail;
1449 }
1450
Tiger Yang50008632007-03-20 16:01:38 -07001451 ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
1452 if (ret) {
1453 mlog_errno(ret);
1454 goto bail;
1455 }
1456
Mark Fashehccd979b2005-12-15 14:31:24 -08001457bail:
1458 mlog_exit(ret);
1459 return ret;
1460}
1461
1462int ocfs2_rw_lock(struct inode *inode, int write)
1463{
1464 int status, level;
1465 struct ocfs2_lock_res *lockres;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001466 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001467
1468 BUG_ON(!inode);
1469
1470 mlog_entry_void();
1471
Mark Fashehb0697052006-03-03 10:24:33 -08001472 mlog(0, "inode %llu take %s RW lock\n",
1473 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08001474 write ? "EXMODE" : "PRMODE");
1475
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001476 if (ocfs2_mount_local(osb))
1477 return 0;
1478
Mark Fashehccd979b2005-12-15 14:31:24 -08001479 lockres = &OCFS2_I(inode)->ip_rw_lockres;
1480
Joel Beckerbd3e7612008-02-01 12:14:57 -08001481 level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08001482
1483 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
1484 0);
1485 if (status < 0)
1486 mlog_errno(status);
1487
1488 mlog_exit(status);
1489 return status;
1490}
1491
1492void ocfs2_rw_unlock(struct inode *inode, int write)
1493{
Joel Beckerbd3e7612008-02-01 12:14:57 -08001494 int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08001495 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001496 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001497
1498 mlog_entry_void();
1499
Mark Fashehb0697052006-03-03 10:24:33 -08001500 mlog(0, "inode %llu drop %s RW lock\n",
1501 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08001502 write ? "EXMODE" : "PRMODE");
1503
Sunil Mushranc271c5c2006-12-05 17:56:35 -08001504 if (!ocfs2_mount_local(osb))
1505 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001506
1507 mlog_exit_void();
1508}
1509
Tiger Yang50008632007-03-20 16:01:38 -07001510/*
1511 * ocfs2_open_lock always get PR mode lock.
1512 */
1513int ocfs2_open_lock(struct inode *inode)
1514{
1515 int status = 0;
1516 struct ocfs2_lock_res *lockres;
1517 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1518
1519 BUG_ON(!inode);
1520
1521 mlog_entry_void();
1522
1523 mlog(0, "inode %llu take PRMODE open lock\n",
1524 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1525
1526 if (ocfs2_mount_local(osb))
1527 goto out;
1528
1529 lockres = &OCFS2_I(inode)->ip_open_lockres;
1530
1531 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001532 DLM_LOCK_PR, 0, 0);
Tiger Yang50008632007-03-20 16:01:38 -07001533 if (status < 0)
1534 mlog_errno(status);
1535
1536out:
1537 mlog_exit(status);
1538 return status;
1539}
1540
1541int ocfs2_try_open_lock(struct inode *inode, int write)
1542{
1543 int status = 0, level;
1544 struct ocfs2_lock_res *lockres;
1545 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1546
1547 BUG_ON(!inode);
1548
1549 mlog_entry_void();
1550
1551 mlog(0, "inode %llu try to take %s open lock\n",
1552 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1553 write ? "EXMODE" : "PRMODE");
1554
1555 if (ocfs2_mount_local(osb))
1556 goto out;
1557
1558 lockres = &OCFS2_I(inode)->ip_open_lockres;
1559
Joel Beckerbd3e7612008-02-01 12:14:57 -08001560 level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
Tiger Yang50008632007-03-20 16:01:38 -07001561
1562 /*
1563 * The file system may already holding a PRMODE/EXMODE open lock.
Joel Beckerbd3e7612008-02-01 12:14:57 -08001564 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
Tiger Yang50008632007-03-20 16:01:38 -07001565 * other nodes and the -EAGAIN will indicate to the caller that
1566 * this inode is still in use.
1567 */
1568 status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001569 level, DLM_LKF_NOQUEUE, 0);
Tiger Yang50008632007-03-20 16:01:38 -07001570
1571out:
1572 mlog_exit(status);
1573 return status;
1574}
1575
1576/*
1577 * ocfs2_open_unlock unlock PR and EX mode open locks.
1578 */
1579void ocfs2_open_unlock(struct inode *inode)
1580{
1581 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
1582 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1583
1584 mlog_entry_void();
1585
1586 mlog(0, "inode %llu drop open lock\n",
1587 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1588
1589 if (ocfs2_mount_local(osb))
1590 goto out;
1591
1592 if(lockres->l_ro_holders)
1593 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001594 DLM_LOCK_PR);
Tiger Yang50008632007-03-20 16:01:38 -07001595 if(lockres->l_ex_holders)
1596 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
Joel Beckerbd3e7612008-02-01 12:14:57 -08001597 DLM_LOCK_EX);
Tiger Yang50008632007-03-20 16:01:38 -07001598
1599out:
1600 mlog_exit_void();
1601}
1602
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001603static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
1604 int level)
1605{
1606 int ret;
1607 struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
1608 unsigned long flags;
1609 struct ocfs2_mask_waiter mw;
1610
1611 ocfs2_init_mask_waiter(&mw);
1612
1613retry_cancel:
1614 spin_lock_irqsave(&lockres->l_lock, flags);
1615 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
1616 ret = ocfs2_prepare_cancel_convert(osb, lockres);
1617 if (ret) {
1618 spin_unlock_irqrestore(&lockres->l_lock, flags);
1619 ret = ocfs2_cancel_convert(osb, lockres);
1620 if (ret < 0) {
1621 mlog_errno(ret);
1622 goto out;
1623 }
1624 goto retry_cancel;
1625 }
1626 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1627 spin_unlock_irqrestore(&lockres->l_lock, flags);
1628
1629 ocfs2_wait_for_mask(&mw);
1630 goto retry_cancel;
1631 }
1632
1633 ret = -ERESTARTSYS;
1634 /*
1635 * We may still have gotten the lock, in which case there's no
1636 * point to restarting the syscall.
1637 */
1638 if (lockres->l_level == level)
1639 ret = 0;
1640
1641 mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
1642 lockres->l_flags, lockres->l_level, lockres->l_action);
1643
1644 spin_unlock_irqrestore(&lockres->l_lock, flags);
1645
1646out:
1647 return ret;
1648}
1649
1650/*
1651 * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
1652 * flock() calls. The locking approach this requires is sufficiently
1653 * different from all other cluster lock types that we implement a
1654 * seperate path to the "low-level" dlm calls. In particular:
1655 *
1656 * - No optimization of lock levels is done - we take at exactly
1657 * what's been requested.
1658 *
1659 * - No lock caching is employed. We immediately downconvert to
1660 * no-lock at unlock time. This also means flock locks never go on
1661 * the blocking list).
1662 *
1663 * - Since userspace can trivially deadlock itself with flock, we make
1664 * sure to allow cancellation of a misbehaving applications flock()
1665 * request.
1666 *
1667 * - Access to any flock lockres doesn't require concurrency, so we
1668 * can simplify the code by requiring the caller to guarantee
1669 * serialization of dlmglue flock calls.
1670 */
1671int ocfs2_file_lock(struct file *file, int ex, int trylock)
1672{
Mark Fashehe988cf12008-07-10 09:25:39 -07001673 int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
1674 unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001675 unsigned long flags;
1676 struct ocfs2_file_private *fp = file->private_data;
1677 struct ocfs2_lock_res *lockres = &fp->fp_flock;
1678 struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1679 struct ocfs2_mask_waiter mw;
1680
1681 ocfs2_init_mask_waiter(&mw);
1682
1683 if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
Joel Beckerbd3e7612008-02-01 12:14:57 -08001684 (lockres->l_level > DLM_LOCK_NL)) {
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001685 mlog(ML_ERROR,
1686 "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
1687 "level: %u\n", lockres->l_name, lockres->l_flags,
1688 lockres->l_level);
1689 return -EINVAL;
1690 }
1691
1692 spin_lock_irqsave(&lockres->l_lock, flags);
1693 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
1694 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1695 spin_unlock_irqrestore(&lockres->l_lock, flags);
1696
1697 /*
1698 * Get the lock at NLMODE to start - that way we
1699 * can cancel the upconvert request if need be.
1700 */
Mark Fashehe988cf12008-07-10 09:25:39 -07001701 ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001702 if (ret < 0) {
1703 mlog_errno(ret);
1704 goto out;
1705 }
1706
1707 ret = ocfs2_wait_for_mask(&mw);
1708 if (ret) {
1709 mlog_errno(ret);
1710 goto out;
1711 }
1712 spin_lock_irqsave(&lockres->l_lock, flags);
1713 }
1714
1715 lockres->l_action = OCFS2_AST_CONVERT;
Mark Fashehe988cf12008-07-10 09:25:39 -07001716 lkm_flags |= DLM_LKF_CONVERT;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001717 lockres->l_requested = level;
1718 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
1719
1720 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1721 spin_unlock_irqrestore(&lockres->l_lock, flags);
1722
Joel Becker4670c462008-02-01 14:39:35 -08001723 ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
Joel Becker24ef1812008-01-29 17:37:32 -08001724 lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1,
1725 lockres);
Joel Becker7431cd72008-02-01 12:15:37 -08001726 if (ret) {
1727 if (!trylock || (ret != -EAGAIN)) {
Joel Becker24ef1812008-01-29 17:37:32 -08001728 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001729 ret = -EINVAL;
1730 }
1731
1732 ocfs2_recover_from_dlm_error(lockres, 1);
1733 lockres_remove_mask_waiter(lockres, &mw);
1734 goto out;
1735 }
1736
1737 ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
1738 if (ret == -ERESTARTSYS) {
1739 /*
1740 * Userspace can cause deadlock itself with
1741 * flock(). Current behavior locally is to allow the
1742 * deadlock, but abort the system call if a signal is
1743 * received. We follow this example, otherwise a
1744 * poorly written program could sit in kernel until
1745 * reboot.
1746 *
1747 * Handling this is a bit more complicated for Ocfs2
1748 * though. We can't exit this function with an
1749 * outstanding lock request, so a cancel convert is
1750 * required. We intentionally overwrite 'ret' - if the
1751 * cancel fails and the lock was granted, it's easier
1752 * to just bubble sucess back up to the user.
1753 */
1754 ret = ocfs2_flock_handle_signal(lockres, level);
David Teigland1693a5c2008-01-30 16:52:53 -08001755 } else if (!ret && (level > lockres->l_level)) {
1756 /* Trylock failed asynchronously */
1757 BUG_ON(!trylock);
1758 ret = -EAGAIN;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001759 }
1760
1761out:
1762
1763 mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
1764 lockres->l_name, ex, trylock, ret);
1765 return ret;
1766}
1767
1768void ocfs2_file_unlock(struct file *file)
1769{
1770 int ret;
Joel Beckerde551242008-02-01 14:45:08 -08001771 unsigned int gen;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001772 unsigned long flags;
1773 struct ocfs2_file_private *fp = file->private_data;
1774 struct ocfs2_lock_res *lockres = &fp->fp_flock;
1775 struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
1776 struct ocfs2_mask_waiter mw;
1777
1778 ocfs2_init_mask_waiter(&mw);
1779
1780 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
1781 return;
1782
Mark Fashehe988cf12008-07-10 09:25:39 -07001783 if (lockres->l_level == DLM_LOCK_NL)
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001784 return;
1785
1786 mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
1787 lockres->l_name, lockres->l_flags, lockres->l_level,
1788 lockres->l_action);
1789
1790 spin_lock_irqsave(&lockres->l_lock, flags);
1791 /*
1792 * Fake a blocking ast for the downconvert code.
1793 */
1794 lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
Joel Beckerbd3e7612008-02-01 12:14:57 -08001795 lockres->l_blocking = DLM_LOCK_EX;
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001796
Mark Fashehe988cf12008-07-10 09:25:39 -07001797 gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001798 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
1799 spin_unlock_irqrestore(&lockres->l_lock, flags);
1800
Mark Fashehe988cf12008-07-10 09:25:39 -07001801 ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
Mark Fashehcf8e06f2007-12-20 16:43:10 -08001802 if (ret) {
1803 mlog_errno(ret);
1804 return;
1805 }
1806
1807 ret = ocfs2_wait_for_mask(&mw);
1808 if (ret)
1809 mlog_errno(ret);
1810}
1811
Mark Fasheh34d024f2007-09-24 15:56:19 -07001812static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
1813 struct ocfs2_lock_res *lockres)
Mark Fashehccd979b2005-12-15 14:31:24 -08001814{
1815 int kick = 0;
1816
1817 mlog_entry_void();
1818
1819 /* If we know that another node is waiting on our lock, kick
Mark Fasheh34d024f2007-09-24 15:56:19 -07001820 * the downconvert thread * pre-emptively when we reach a release
Mark Fashehccd979b2005-12-15 14:31:24 -08001821 * condition. */
1822 if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
1823 switch(lockres->l_blocking) {
Joel Beckerbd3e7612008-02-01 12:14:57 -08001824 case DLM_LOCK_EX:
Mark Fashehccd979b2005-12-15 14:31:24 -08001825 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
1826 kick = 1;
1827 break;
Joel Beckerbd3e7612008-02-01 12:14:57 -08001828 case DLM_LOCK_PR:
Mark Fashehccd979b2005-12-15 14:31:24 -08001829 if (!lockres->l_ex_holders)
1830 kick = 1;
1831 break;
1832 default:
1833 BUG();
1834 }
1835 }
1836
1837 if (kick)
Mark Fasheh34d024f2007-09-24 15:56:19 -07001838 ocfs2_wake_downconvert_thread(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001839
1840 mlog_exit_void();
1841}
1842
Mark Fashehccd979b2005-12-15 14:31:24 -08001843#define OCFS2_SEC_BITS 34
1844#define OCFS2_SEC_SHIFT (64 - 34)
1845#define OCFS2_NSEC_MASK ((1ULL << OCFS2_SEC_SHIFT) - 1)
1846
1847/* LVB only has room for 64 bits of time here so we pack it for
1848 * now. */
1849static u64 ocfs2_pack_timespec(struct timespec *spec)
1850{
1851 u64 res;
1852 u64 sec = spec->tv_sec;
1853 u32 nsec = spec->tv_nsec;
1854
1855 res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
1856
1857 return res;
1858}
1859
1860/* Call this with the lockres locked. I am reasonably sure we don't
1861 * need ip_lock in this function as anyone who would be changing those
Mark Fashehe63aecb62007-10-18 15:30:42 -07001862 * values is supposed to be blocked in ocfs2_inode_lock right now. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001863static void __ocfs2_stuff_meta_lvb(struct inode *inode)
1864{
1865 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001866 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
Mark Fashehccd979b2005-12-15 14:31:24 -08001867 struct ocfs2_meta_lvb *lvb;
1868
1869 mlog_entry_void();
1870
Mark Fasheha641dc22008-12-24 16:03:48 -08001871 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001872
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001873 /*
1874 * Invalidate the LVB of a deleted inode - this way other
1875 * nodes are forced to go to disk and discover the new inode
1876 * status.
1877 */
1878 if (oi->ip_flags & OCFS2_INODE_DELETED) {
1879 lvb->lvb_version = 0;
1880 goto out;
1881 }
1882
Mark Fasheh4d3b83f2006-09-12 15:22:18 -07001883 lvb->lvb_version = OCFS2_LVB_VERSION;
Mark Fashehccd979b2005-12-15 14:31:24 -08001884 lvb->lvb_isize = cpu_to_be64(i_size_read(inode));
1885 lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
1886 lvb->lvb_iuid = cpu_to_be32(inode->i_uid);
1887 lvb->lvb_igid = cpu_to_be32(inode->i_gid);
1888 lvb->lvb_imode = cpu_to_be16(inode->i_mode);
1889 lvb->lvb_inlink = cpu_to_be16(inode->i_nlink);
1890 lvb->lvb_iatime_packed =
1891 cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
1892 lvb->lvb_ictime_packed =
1893 cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
1894 lvb->lvb_imtime_packed =
1895 cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001896 lvb->lvb_iattr = cpu_to_be32(oi->ip_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001897 lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
Mark Fashehf9e2d822006-09-12 15:35:49 -07001898 lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
Mark Fashehccd979b2005-12-15 14:31:24 -08001899
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001900out:
Mark Fashehccd979b2005-12-15 14:31:24 -08001901 mlog_meta_lvb(0, lockres);
1902
1903 mlog_exit_void();
1904}
1905
1906static void ocfs2_unpack_timespec(struct timespec *spec,
1907 u64 packed_time)
1908{
1909 spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
1910 spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
1911}
1912
1913static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
1914{
1915 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001916 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
Mark Fashehccd979b2005-12-15 14:31:24 -08001917 struct ocfs2_meta_lvb *lvb;
1918
1919 mlog_entry_void();
1920
1921 mlog_meta_lvb(0, lockres);
1922
Mark Fasheha641dc22008-12-24 16:03:48 -08001923 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001924
1925 /* We're safe here without the lockres lock... */
1926 spin_lock(&oi->ip_lock);
1927 oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
1928 i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
1929
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001930 oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001931 oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001932 ocfs2_set_inode_flags(inode);
1933
Mark Fashehccd979b2005-12-15 14:31:24 -08001934 /* fast-symlinks are a special case */
1935 if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
1936 inode->i_blocks = 0;
1937 else
Mark Fasheh8110b072007-03-22 16:53:23 -07001938 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001939
1940 inode->i_uid = be32_to_cpu(lvb->lvb_iuid);
1941 inode->i_gid = be32_to_cpu(lvb->lvb_igid);
1942 inode->i_mode = be16_to_cpu(lvb->lvb_imode);
1943 inode->i_nlink = be16_to_cpu(lvb->lvb_inlink);
1944 ocfs2_unpack_timespec(&inode->i_atime,
1945 be64_to_cpu(lvb->lvb_iatime_packed));
1946 ocfs2_unpack_timespec(&inode->i_mtime,
1947 be64_to_cpu(lvb->lvb_imtime_packed));
1948 ocfs2_unpack_timespec(&inode->i_ctime,
1949 be64_to_cpu(lvb->lvb_ictime_packed));
1950 spin_unlock(&oi->ip_lock);
1951
1952 mlog_exit_void();
1953}
1954
Mark Fashehf9e2d822006-09-12 15:35:49 -07001955static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
1956 struct ocfs2_lock_res *lockres)
Mark Fashehccd979b2005-12-15 14:31:24 -08001957{
Mark Fasheha641dc22008-12-24 16:03:48 -08001958 struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08001959
Mark Fashehf9e2d822006-09-12 15:35:49 -07001960 if (lvb->lvb_version == OCFS2_LVB_VERSION
1961 && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
Mark Fashehccd979b2005-12-15 14:31:24 -08001962 return 1;
1963 return 0;
1964}
1965
1966/* Determine whether a lock resource needs to be refreshed, and
1967 * arbitrate who gets to refresh it.
1968 *
1969 * 0 means no refresh needed.
1970 *
1971 * > 0 means you need to refresh this and you MUST call
1972 * ocfs2_complete_lock_res_refresh afterwards. */
1973static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
1974{
1975 unsigned long flags;
1976 int status = 0;
1977
1978 mlog_entry_void();
1979
1980refresh_check:
1981 spin_lock_irqsave(&lockres->l_lock, flags);
1982 if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
1983 spin_unlock_irqrestore(&lockres->l_lock, flags);
1984 goto bail;
1985 }
1986
1987 if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
1988 spin_unlock_irqrestore(&lockres->l_lock, flags);
1989
1990 ocfs2_wait_on_refreshing_lock(lockres);
1991 goto refresh_check;
1992 }
1993
1994 /* Ok, I'll be the one to refresh this lock. */
1995 lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
1996 spin_unlock_irqrestore(&lockres->l_lock, flags);
1997
1998 status = 1;
1999bail:
2000 mlog_exit(status);
2001 return status;
2002}
2003
2004/* If status is non zero, I'll mark it as not being in refresh
2005 * anymroe, but i won't clear the needs refresh flag. */
2006static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
2007 int status)
2008{
2009 unsigned long flags;
2010 mlog_entry_void();
2011
2012 spin_lock_irqsave(&lockres->l_lock, flags);
2013 lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
2014 if (!status)
2015 lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
2016 spin_unlock_irqrestore(&lockres->l_lock, flags);
2017
2018 wake_up(&lockres->l_event);
2019
2020 mlog_exit_void();
2021}
2022
2023/* may or may not return a bh if it went to disk. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002024static int ocfs2_inode_lock_update(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -08002025 struct buffer_head **bh)
2026{
2027 int status = 0;
2028 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002029 struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
Mark Fashehccd979b2005-12-15 14:31:24 -08002030 struct ocfs2_dinode *fe;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002031 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002032
2033 mlog_entry_void();
2034
Mark Fashehbe9e9862007-04-18 15:22:08 -07002035 if (ocfs2_mount_local(osb))
2036 goto bail;
2037
Mark Fashehccd979b2005-12-15 14:31:24 -08002038 spin_lock(&oi->ip_lock);
2039 if (oi->ip_flags & OCFS2_INODE_DELETED) {
Mark Fashehb0697052006-03-03 10:24:33 -08002040 mlog(0, "Orphaned inode %llu was deleted while we "
Mark Fashehccd979b2005-12-15 14:31:24 -08002041 "were waiting on a lock. ip_flags = 0x%x\n",
Mark Fashehb0697052006-03-03 10:24:33 -08002042 (unsigned long long)oi->ip_blkno, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08002043 spin_unlock(&oi->ip_lock);
2044 status = -ENOENT;
2045 goto bail;
2046 }
2047 spin_unlock(&oi->ip_lock);
2048
Mark Fashehbe9e9862007-04-18 15:22:08 -07002049 if (!ocfs2_should_refresh_lock_res(lockres))
2050 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -08002051
2052 /* This will discard any caching information we might have had
2053 * for the inode metadata. */
2054 ocfs2_metadata_cache_purge(inode);
2055
Mark Fasheh83418972007-04-23 18:53:12 -07002056 ocfs2_extent_map_trunc(inode, 0);
2057
Mark Fashehbe9e9862007-04-18 15:22:08 -07002058 if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
Mark Fashehb0697052006-03-03 10:24:33 -08002059 mlog(0, "Trusting LVB on inode %llu\n",
2060 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08002061 ocfs2_refresh_inode_from_lvb(inode);
2062 } else {
2063 /* Boo, we have to go to disk. */
2064 /* read bh, cast, ocfs2_refresh_inode */
Joel Beckerb657c952008-11-13 14:49:11 -08002065 status = ocfs2_read_inode_block(inode, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002066 if (status < 0) {
2067 mlog_errno(status);
2068 goto bail_refresh;
2069 }
2070 fe = (struct ocfs2_dinode *) (*bh)->b_data;
2071
2072 /* This is a good chance to make sure we're not
Joel Beckerb657c952008-11-13 14:49:11 -08002073 * locking an invalid object. ocfs2_read_inode_block()
2074 * already checked that the inode block is sane.
Mark Fashehccd979b2005-12-15 14:31:24 -08002075 *
2076 * We bug on a stale inode here because we checked
2077 * above whether it was wiped from disk. The wiping
2078 * node provides a guarantee that we receive that
2079 * message and can mark the inode before dropping any
2080 * locks associated with it. */
Mark Fashehccd979b2005-12-15 14:31:24 -08002081 mlog_bug_on_msg(inode->i_generation !=
2082 le32_to_cpu(fe->i_generation),
Mark Fashehb0697052006-03-03 10:24:33 -08002083 "Invalid dinode %llu disk generation: %u "
Mark Fashehccd979b2005-12-15 14:31:24 -08002084 "inode->i_generation: %u\n",
Mark Fashehb0697052006-03-03 10:24:33 -08002085 (unsigned long long)oi->ip_blkno,
2086 le32_to_cpu(fe->i_generation),
Mark Fashehccd979b2005-12-15 14:31:24 -08002087 inode->i_generation);
2088 mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
2089 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
Mark Fashehb0697052006-03-03 10:24:33 -08002090 "Stale dinode %llu dtime: %llu flags: 0x%x\n",
2091 (unsigned long long)oi->ip_blkno,
2092 (unsigned long long)le64_to_cpu(fe->i_dtime),
Mark Fashehccd979b2005-12-15 14:31:24 -08002093 le32_to_cpu(fe->i_flags));
2094
2095 ocfs2_refresh_inode(inode, fe);
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002096 ocfs2_track_lock_refresh(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08002097 }
2098
2099 status = 0;
2100bail_refresh:
Mark Fashehbe9e9862007-04-18 15:22:08 -07002101 ocfs2_complete_lock_res_refresh(lockres, status);
Mark Fashehccd979b2005-12-15 14:31:24 -08002102bail:
2103 mlog_exit(status);
2104 return status;
2105}
2106
2107static int ocfs2_assign_bh(struct inode *inode,
2108 struct buffer_head **ret_bh,
2109 struct buffer_head *passed_bh)
2110{
2111 int status;
2112
2113 if (passed_bh) {
2114 /* Ok, the update went to disk for us, use the
2115 * returned bh. */
2116 *ret_bh = passed_bh;
2117 get_bh(*ret_bh);
2118
2119 return 0;
2120 }
2121
Joel Beckerb657c952008-11-13 14:49:11 -08002122 status = ocfs2_read_inode_block(inode, ret_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002123 if (status < 0)
2124 mlog_errno(status);
2125
2126 return status;
2127}
2128
2129/*
2130 * returns < 0 error if the callback will never be called, otherwise
2131 * the result of the lock will be communicated via the callback.
2132 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002133int ocfs2_inode_lock_full(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -08002134 struct buffer_head **ret_bh,
2135 int ex,
2136 int arg_flags)
2137{
Joel Beckerbd3e7612008-02-01 12:14:57 -08002138 int status, level, acquired;
2139 u32 dlm_flags;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002140 struct ocfs2_lock_res *lockres = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002141 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2142 struct buffer_head *local_bh = NULL;
2143
2144 BUG_ON(!inode);
2145
2146 mlog_entry_void();
2147
Mark Fashehb0697052006-03-03 10:24:33 -08002148 mlog(0, "inode %llu, take %s META lock\n",
2149 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08002150 ex ? "EXMODE" : "PRMODE");
2151
2152 status = 0;
2153 acquired = 0;
2154 /* We'll allow faking a readonly metadata lock for
2155 * rodevices. */
2156 if (ocfs2_is_hard_readonly(osb)) {
2157 if (ex)
2158 status = -EROFS;
2159 goto bail;
2160 }
2161
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002162 if (ocfs2_mount_local(osb))
2163 goto local;
2164
Mark Fashehccd979b2005-12-15 14:31:24 -08002165 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
Joel Becker553abd02008-02-01 12:03:57 -08002166 ocfs2_wait_for_recovery(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002167
Mark Fashehe63aecb62007-10-18 15:30:42 -07002168 lockres = &OCFS2_I(inode)->ip_inode_lockres;
Joel Beckerbd3e7612008-02-01 12:14:57 -08002169 level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08002170 dlm_flags = 0;
2171 if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
Joel Beckerbd3e7612008-02-01 12:14:57 -08002172 dlm_flags |= DLM_LKF_NOQUEUE;
Mark Fashehccd979b2005-12-15 14:31:24 -08002173
2174 status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags);
2175 if (status < 0) {
2176 if (status != -EAGAIN && status != -EIOCBRETRY)
2177 mlog_errno(status);
2178 goto bail;
2179 }
2180
2181 /* Notify the error cleanup path to drop the cluster lock. */
2182 acquired = 1;
2183
2184 /* We wait twice because a node may have died while we were in
2185 * the lower dlm layers. The second time though, we've
2186 * committed to owning this lock so we don't allow signals to
2187 * abort the operation. */
2188 if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
Joel Becker553abd02008-02-01 12:03:57 -08002189 ocfs2_wait_for_recovery(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002190
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002191local:
Mark Fasheh24c19ef2006-09-22 17:28:19 -07002192 /*
2193 * We only see this flag if we're being called from
2194 * ocfs2_read_locked_inode(). It means we're locking an inode
2195 * which hasn't been populated yet, so clear the refresh flag
2196 * and let the caller handle it.
2197 */
2198 if (inode->i_state & I_NEW) {
2199 status = 0;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002200 if (lockres)
2201 ocfs2_complete_lock_res_refresh(lockres, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -07002202 goto bail;
2203 }
2204
Mark Fashehccd979b2005-12-15 14:31:24 -08002205 /* This is fun. The caller may want a bh back, or it may
Mark Fashehe63aecb62007-10-18 15:30:42 -07002206 * not. ocfs2_inode_lock_update definitely wants one in, but
Mark Fashehccd979b2005-12-15 14:31:24 -08002207 * may or may not read one, depending on what's in the
2208 * LVB. The result of all of this is that we've *only* gone to
2209 * disk if we have to, so the complexity is worthwhile. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002210 status = ocfs2_inode_lock_update(inode, &local_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08002211 if (status < 0) {
2212 if (status != -ENOENT)
2213 mlog_errno(status);
2214 goto bail;
2215 }
2216
2217 if (ret_bh) {
2218 status = ocfs2_assign_bh(inode, ret_bh, local_bh);
2219 if (status < 0) {
2220 mlog_errno(status);
2221 goto bail;
2222 }
2223 }
2224
Mark Fashehccd979b2005-12-15 14:31:24 -08002225bail:
2226 if (status < 0) {
2227 if (ret_bh && (*ret_bh)) {
2228 brelse(*ret_bh);
2229 *ret_bh = NULL;
2230 }
2231 if (acquired)
Mark Fashehe63aecb62007-10-18 15:30:42 -07002232 ocfs2_inode_unlock(inode, ex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002233 }
2234
2235 if (local_bh)
2236 brelse(local_bh);
2237
2238 mlog_exit(status);
2239 return status;
2240}
2241
2242/*
Mark Fasheh34d024f2007-09-24 15:56:19 -07002243 * This is working around a lock inversion between tasks acquiring DLM
2244 * locks while holding a page lock and the downconvert thread which
2245 * blocks dlm lock acquiry while acquiring page locks.
Mark Fashehccd979b2005-12-15 14:31:24 -08002246 *
2247 * ** These _with_page variantes are only intended to be called from aop
2248 * methods that hold page locks and return a very specific *positive* error
2249 * code that aop methods pass up to the VFS -- test for errors with != 0. **
2250 *
Mark Fasheh34d024f2007-09-24 15:56:19 -07002251 * The DLM is called such that it returns -EAGAIN if it would have
2252 * blocked waiting for the downconvert thread. In that case we unlock
2253 * our page so the downconvert thread can make progress. Once we've
2254 * done this we have to return AOP_TRUNCATED_PAGE so the aop method
2255 * that called us can bubble that back up into the VFS who will then
2256 * immediately retry the aop call.
Mark Fashehccd979b2005-12-15 14:31:24 -08002257 *
2258 * We do a blocking lock and immediate unlock before returning, though, so that
2259 * the lock has a great chance of being cached on this node by the time the VFS
2260 * calls back to retry the aop. This has a potential to livelock as nodes
2261 * ping locks back and forth, but that's a risk we're willing to take to avoid
2262 * the lock inversion simply.
2263 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002264int ocfs2_inode_lock_with_page(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -08002265 struct buffer_head **ret_bh,
2266 int ex,
2267 struct page *page)
2268{
2269 int ret;
2270
Mark Fashehe63aecb62007-10-18 15:30:42 -07002271 ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
Mark Fashehccd979b2005-12-15 14:31:24 -08002272 if (ret == -EAGAIN) {
2273 unlock_page(page);
Mark Fashehe63aecb62007-10-18 15:30:42 -07002274 if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
2275 ocfs2_inode_unlock(inode, ex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002276 ret = AOP_TRUNCATED_PAGE;
2277 }
2278
2279 return ret;
2280}
2281
Mark Fashehe63aecb62007-10-18 15:30:42 -07002282int ocfs2_inode_lock_atime(struct inode *inode,
Tiger Yang7f1a37e2006-11-15 15:48:42 +08002283 struct vfsmount *vfsmnt,
2284 int *level)
2285{
2286 int ret;
2287
2288 mlog_entry_void();
Mark Fashehe63aecb62007-10-18 15:30:42 -07002289 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yang7f1a37e2006-11-15 15:48:42 +08002290 if (ret < 0) {
2291 mlog_errno(ret);
2292 return ret;
2293 }
2294
2295 /*
2296 * If we should update atime, we will get EX lock,
2297 * otherwise we just get PR lock.
2298 */
2299 if (ocfs2_should_update_atime(inode, vfsmnt)) {
2300 struct buffer_head *bh = NULL;
2301
Mark Fashehe63aecb62007-10-18 15:30:42 -07002302 ocfs2_inode_unlock(inode, 0);
2303 ret = ocfs2_inode_lock(inode, &bh, 1);
Tiger Yang7f1a37e2006-11-15 15:48:42 +08002304 if (ret < 0) {
2305 mlog_errno(ret);
2306 return ret;
2307 }
2308 *level = 1;
2309 if (ocfs2_should_update_atime(inode, vfsmnt))
2310 ocfs2_update_inode_atime(inode, bh);
2311 if (bh)
2312 brelse(bh);
2313 } else
2314 *level = 0;
2315
2316 mlog_exit(ret);
2317 return ret;
2318}
2319
Mark Fashehe63aecb62007-10-18 15:30:42 -07002320void ocfs2_inode_unlock(struct inode *inode,
Mark Fashehccd979b2005-12-15 14:31:24 -08002321 int ex)
2322{
Joel Beckerbd3e7612008-02-01 12:14:57 -08002323 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehe63aecb62007-10-18 15:30:42 -07002324 struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002325 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002326
2327 mlog_entry_void();
2328
Mark Fashehb0697052006-03-03 10:24:33 -08002329 mlog(0, "inode %llu drop %s META lock\n",
2330 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -08002331 ex ? "EXMODE" : "PRMODE");
2332
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002333 if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
2334 !ocfs2_mount_local(osb))
Mark Fashehccd979b2005-12-15 14:31:24 -08002335 ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
2336
2337 mlog_exit_void();
2338}
2339
2340int ocfs2_super_lock(struct ocfs2_super *osb,
2341 int ex)
2342{
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002343 int status = 0;
Joel Beckerbd3e7612008-02-01 12:14:57 -08002344 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08002345 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
Mark Fashehccd979b2005-12-15 14:31:24 -08002346
2347 mlog_entry_void();
2348
2349 if (ocfs2_is_hard_readonly(osb))
2350 return -EROFS;
2351
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002352 if (ocfs2_mount_local(osb))
2353 goto bail;
2354
Mark Fashehccd979b2005-12-15 14:31:24 -08002355 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
2356 if (status < 0) {
2357 mlog_errno(status);
2358 goto bail;
2359 }
2360
2361 /* The super block lock path is really in the best position to
2362 * know when resources covered by the lock need to be
2363 * refreshed, so we do it here. Of course, making sense of
2364 * everything is up to the caller :) */
2365 status = ocfs2_should_refresh_lock_res(lockres);
2366 if (status < 0) {
2367 mlog_errno(status);
2368 goto bail;
2369 }
2370 if (status) {
Mark Fasheh8e8a4602008-02-01 11:59:09 -08002371 status = ocfs2_refresh_slot_info(osb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002372
2373 ocfs2_complete_lock_res_refresh(lockres, status);
2374
2375 if (status < 0)
2376 mlog_errno(status);
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002377 ocfs2_track_lock_refresh(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08002378 }
2379bail:
2380 mlog_exit(status);
2381 return status;
2382}
2383
2384void ocfs2_super_unlock(struct ocfs2_super *osb,
2385 int ex)
2386{
Joel Beckerbd3e7612008-02-01 12:14:57 -08002387 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehccd979b2005-12-15 14:31:24 -08002388 struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
2389
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002390 if (!ocfs2_mount_local(osb))
2391 ocfs2_cluster_unlock(osb, lockres, level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002392}
2393
2394int ocfs2_rename_lock(struct ocfs2_super *osb)
2395{
2396 int status;
2397 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2398
2399 if (ocfs2_is_hard_readonly(osb))
2400 return -EROFS;
2401
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002402 if (ocfs2_mount_local(osb))
2403 return 0;
2404
Joel Beckerbd3e7612008-02-01 12:14:57 -08002405 status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08002406 if (status < 0)
2407 mlog_errno(status);
2408
2409 return status;
2410}
2411
2412void ocfs2_rename_unlock(struct ocfs2_super *osb)
2413{
2414 struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
2415
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002416 if (!ocfs2_mount_local(osb))
Joel Beckerbd3e7612008-02-01 12:14:57 -08002417 ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
Mark Fashehccd979b2005-12-15 14:31:24 -08002418}
2419
Mark Fashehd680efe2006-09-08 14:14:34 -07002420int ocfs2_dentry_lock(struct dentry *dentry, int ex)
2421{
2422 int ret;
Joel Beckerbd3e7612008-02-01 12:14:57 -08002423 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehd680efe2006-09-08 14:14:34 -07002424 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2425 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2426
2427 BUG_ON(!dl);
2428
2429 if (ocfs2_is_hard_readonly(osb))
2430 return -EROFS;
2431
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002432 if (ocfs2_mount_local(osb))
2433 return 0;
2434
Mark Fashehd680efe2006-09-08 14:14:34 -07002435 ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
2436 if (ret < 0)
2437 mlog_errno(ret);
2438
2439 return ret;
2440}
2441
2442void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
2443{
Joel Beckerbd3e7612008-02-01 12:14:57 -08002444 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
Mark Fashehd680efe2006-09-08 14:14:34 -07002445 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
2446 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
2447
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002448 if (!ocfs2_mount_local(osb))
2449 ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
Mark Fashehd680efe2006-09-08 14:14:34 -07002450}
2451
Mark Fashehccd979b2005-12-15 14:31:24 -08002452/* Reference counting of the dlm debug structure. We want this because
2453 * open references on the debug inodes can live on after a mount, so
2454 * we can't rely on the ocfs2_super to always exist. */
2455static void ocfs2_dlm_debug_free(struct kref *kref)
2456{
2457 struct ocfs2_dlm_debug *dlm_debug;
2458
2459 dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
2460
2461 kfree(dlm_debug);
2462}
2463
2464void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
2465{
2466 if (dlm_debug)
2467 kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
2468}
2469
2470static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
2471{
2472 kref_get(&debug->d_refcnt);
2473}
2474
2475struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
2476{
2477 struct ocfs2_dlm_debug *dlm_debug;
2478
2479 dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
2480 if (!dlm_debug) {
2481 mlog_errno(-ENOMEM);
2482 goto out;
2483 }
2484
2485 kref_init(&dlm_debug->d_refcnt);
2486 INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
2487 dlm_debug->d_locking_state = NULL;
2488out:
2489 return dlm_debug;
2490}
2491
2492/* Access to this is arbitrated for us via seq_file->sem. */
2493struct ocfs2_dlm_seq_priv {
2494 struct ocfs2_dlm_debug *p_dlm_debug;
2495 struct ocfs2_lock_res p_iter_res;
2496 struct ocfs2_lock_res p_tmp_res;
2497};
2498
2499static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
2500 struct ocfs2_dlm_seq_priv *priv)
2501{
2502 struct ocfs2_lock_res *iter, *ret = NULL;
2503 struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
2504
2505 assert_spin_locked(&ocfs2_dlm_tracking_lock);
2506
2507 list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
2508 /* discover the head of the list */
2509 if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
2510 mlog(0, "End of list found, %p\n", ret);
2511 break;
2512 }
2513
2514 /* We track our "dummy" iteration lockres' by a NULL
2515 * l_ops field. */
2516 if (iter->l_ops != NULL) {
2517 ret = iter;
2518 break;
2519 }
2520 }
2521
2522 return ret;
2523}
2524
2525static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
2526{
2527 struct ocfs2_dlm_seq_priv *priv = m->private;
2528 struct ocfs2_lock_res *iter;
2529
2530 spin_lock(&ocfs2_dlm_tracking_lock);
2531 iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
2532 if (iter) {
2533 /* Since lockres' have the lifetime of their container
2534 * (which can be inodes, ocfs2_supers, etc) we want to
2535 * copy this out to a temporary lockres while still
2536 * under the spinlock. Obviously after this we can't
2537 * trust any pointers on the copy returned, but that's
2538 * ok as the information we want isn't typically held
2539 * in them. */
2540 priv->p_tmp_res = *iter;
2541 iter = &priv->p_tmp_res;
2542 }
2543 spin_unlock(&ocfs2_dlm_tracking_lock);
2544
2545 return iter;
2546}
2547
2548static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
2549{
2550}
2551
2552static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
2553{
2554 struct ocfs2_dlm_seq_priv *priv = m->private;
2555 struct ocfs2_lock_res *iter = v;
2556 struct ocfs2_lock_res *dummy = &priv->p_iter_res;
2557
2558 spin_lock(&ocfs2_dlm_tracking_lock);
2559 iter = ocfs2_dlm_next_res(iter, priv);
2560 list_del_init(&dummy->l_debug_list);
2561 if (iter) {
2562 list_add(&dummy->l_debug_list, &iter->l_debug_list);
2563 priv->p_tmp_res = *iter;
2564 iter = &priv->p_tmp_res;
2565 }
2566 spin_unlock(&ocfs2_dlm_tracking_lock);
2567
2568 return iter;
2569}
2570
2571/* So that debugfs.ocfs2 can determine which format is being used */
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002572#define OCFS2_DLM_DEBUG_STR_VERSION 2
Mark Fashehccd979b2005-12-15 14:31:24 -08002573static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
2574{
2575 int i;
2576 char *lvb;
2577 struct ocfs2_lock_res *lockres = v;
2578
2579 if (!lockres)
2580 return -EINVAL;
2581
Mark Fashehd680efe2006-09-08 14:14:34 -07002582 seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
2583
2584 if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
2585 seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
2586 lockres->l_name,
2587 (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
2588 else
2589 seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
2590
2591 seq_printf(m, "%d\t"
Mark Fashehccd979b2005-12-15 14:31:24 -08002592 "0x%lx\t"
2593 "0x%x\t"
2594 "0x%x\t"
2595 "%u\t"
2596 "%u\t"
2597 "%d\t"
2598 "%d\t",
Mark Fashehccd979b2005-12-15 14:31:24 -08002599 lockres->l_level,
2600 lockres->l_flags,
2601 lockres->l_action,
2602 lockres->l_unlock_action,
2603 lockres->l_ro_holders,
2604 lockres->l_ex_holders,
2605 lockres->l_requested,
2606 lockres->l_blocking);
2607
2608 /* Dump the raw LVB */
Joel Becker8f2c9c12008-02-01 12:16:57 -08002609 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002610 for(i = 0; i < DLM_LVB_LEN; i++)
2611 seq_printf(m, "0x%x\t", lvb[i]);
2612
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002613#ifdef CONFIG_OCFS2_FS_STATS
2614# define lock_num_prmode(_l) (_l)->l_lock_num_prmode
2615# define lock_num_exmode(_l) (_l)->l_lock_num_exmode
2616# define lock_num_prmode_failed(_l) (_l)->l_lock_num_prmode_failed
2617# define lock_num_exmode_failed(_l) (_l)->l_lock_num_exmode_failed
2618# define lock_total_prmode(_l) (_l)->l_lock_total_prmode
2619# define lock_total_exmode(_l) (_l)->l_lock_total_exmode
2620# define lock_max_prmode(_l) (_l)->l_lock_max_prmode
2621# define lock_max_exmode(_l) (_l)->l_lock_max_exmode
2622# define lock_refresh(_l) (_l)->l_lock_refresh
2623#else
Randy Dunlapdd25e552008-05-28 14:41:00 -07002624# define lock_num_prmode(_l) (0ULL)
2625# define lock_num_exmode(_l) (0ULL)
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002626# define lock_num_prmode_failed(_l) (0)
2627# define lock_num_exmode_failed(_l) (0)
Randy Dunlapdd25e552008-05-28 14:41:00 -07002628# define lock_total_prmode(_l) (0ULL)
2629# define lock_total_exmode(_l) (0ULL)
Sunil Mushran8ddb7b02008-05-13 13:45:15 -07002630# define lock_max_prmode(_l) (0)
2631# define lock_max_exmode(_l) (0)
2632# define lock_refresh(_l) (0)
2633#endif
2634 /* The following seq_print was added in version 2 of this output */
2635 seq_printf(m, "%llu\t"
2636 "%llu\t"
2637 "%u\t"
2638 "%u\t"
2639 "%llu\t"
2640 "%llu\t"
2641 "%u\t"
2642 "%u\t"
2643 "%u\t",
2644 lock_num_prmode(lockres),
2645 lock_num_exmode(lockres),
2646 lock_num_prmode_failed(lockres),
2647 lock_num_exmode_failed(lockres),
2648 lock_total_prmode(lockres),
2649 lock_total_exmode(lockres),
2650 lock_max_prmode(lockres),
2651 lock_max_exmode(lockres),
2652 lock_refresh(lockres));
2653
Mark Fashehccd979b2005-12-15 14:31:24 -08002654 /* End the line */
2655 seq_printf(m, "\n");
2656 return 0;
2657}
2658
Jan Engelhardt90d99772008-01-22 20:52:20 +01002659static const struct seq_operations ocfs2_dlm_seq_ops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002660 .start = ocfs2_dlm_seq_start,
2661 .stop = ocfs2_dlm_seq_stop,
2662 .next = ocfs2_dlm_seq_next,
2663 .show = ocfs2_dlm_seq_show,
2664};
2665
2666static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
2667{
2668 struct seq_file *seq = (struct seq_file *) file->private_data;
2669 struct ocfs2_dlm_seq_priv *priv = seq->private;
2670 struct ocfs2_lock_res *res = &priv->p_iter_res;
2671
2672 ocfs2_remove_lockres_tracking(res);
2673 ocfs2_put_dlm_debug(priv->p_dlm_debug);
2674 return seq_release_private(inode, file);
2675}
2676
2677static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
2678{
2679 int ret;
2680 struct ocfs2_dlm_seq_priv *priv;
2681 struct seq_file *seq;
2682 struct ocfs2_super *osb;
2683
2684 priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
2685 if (!priv) {
2686 ret = -ENOMEM;
2687 mlog_errno(ret);
2688 goto out;
2689 }
Theodore Ts'o8e18e292006-09-27 01:50:46 -07002690 osb = inode->i_private;
Mark Fashehccd979b2005-12-15 14:31:24 -08002691 ocfs2_get_dlm_debug(osb->osb_dlm_debug);
2692 priv->p_dlm_debug = osb->osb_dlm_debug;
2693 INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
2694
2695 ret = seq_open(file, &ocfs2_dlm_seq_ops);
2696 if (ret) {
2697 kfree(priv);
2698 mlog_errno(ret);
2699 goto out;
2700 }
2701
2702 seq = (struct seq_file *) file->private_data;
2703 seq->private = priv;
2704
2705 ocfs2_add_lockres_tracking(&priv->p_iter_res,
2706 priv->p_dlm_debug);
2707
2708out:
2709 return ret;
2710}
2711
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002712static const struct file_operations ocfs2_dlm_debug_fops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002713 .open = ocfs2_dlm_debug_open,
2714 .release = ocfs2_dlm_debug_release,
2715 .read = seq_read,
2716 .llseek = seq_lseek,
2717};
2718
2719static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
2720{
2721 int ret = 0;
2722 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2723
2724 dlm_debug->d_locking_state = debugfs_create_file("locking_state",
2725 S_IFREG|S_IRUSR,
2726 osb->osb_debug_root,
2727 osb,
2728 &ocfs2_dlm_debug_fops);
2729 if (!dlm_debug->d_locking_state) {
2730 ret = -EINVAL;
2731 mlog(ML_ERROR,
2732 "Unable to create locking state debugfs file.\n");
2733 goto out;
2734 }
2735
2736 ocfs2_get_dlm_debug(dlm_debug);
2737out:
2738 return ret;
2739}
2740
2741static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
2742{
2743 struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
2744
2745 if (dlm_debug) {
2746 debugfs_remove(dlm_debug->d_locking_state);
2747 ocfs2_put_dlm_debug(dlm_debug);
2748 }
2749}
2750
2751int ocfs2_dlm_init(struct ocfs2_super *osb)
2752{
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002753 int status = 0;
Joel Becker4670c462008-02-01 14:39:35 -08002754 struct ocfs2_cluster_connection *conn = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002755
2756 mlog_entry_void();
2757
Mark Fasheh0abd6d12008-01-29 16:59:56 -08002758 if (ocfs2_mount_local(osb)) {
2759 osb->node_num = 0;
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002760 goto local;
Mark Fasheh0abd6d12008-01-29 16:59:56 -08002761 }
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002762
Mark Fashehccd979b2005-12-15 14:31:24 -08002763 status = ocfs2_dlm_init_debug(osb);
2764 if (status < 0) {
2765 mlog_errno(status);
2766 goto bail;
2767 }
2768
Mark Fasheh34d024f2007-09-24 15:56:19 -07002769 /* launch downconvert thread */
2770 osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc");
2771 if (IS_ERR(osb->dc_task)) {
2772 status = PTR_ERR(osb->dc_task);
2773 osb->dc_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002774 mlog_errno(status);
2775 goto bail;
2776 }
2777
Mark Fashehccd979b2005-12-15 14:31:24 -08002778 /* for now, uuid == domain */
Joel Becker9c6c8772008-02-01 15:17:30 -08002779 status = ocfs2_cluster_connect(osb->osb_cluster_stack,
2780 osb->uuid_str,
Joel Becker4670c462008-02-01 14:39:35 -08002781 strlen(osb->uuid_str),
2782 ocfs2_do_node_down, osb,
2783 &conn);
2784 if (status) {
Mark Fashehccd979b2005-12-15 14:31:24 -08002785 mlog_errno(status);
2786 goto bail;
2787 }
2788
Mark Fasheh0abd6d12008-01-29 16:59:56 -08002789 status = ocfs2_cluster_this_node(&osb->node_num);
2790 if (status < 0) {
2791 mlog_errno(status);
2792 mlog(ML_ERROR,
2793 "could not find this host's node number\n");
Joel Becker286eaa92008-02-01 15:03:57 -08002794 ocfs2_cluster_disconnect(conn, 0);
Mark Fasheh0abd6d12008-01-29 16:59:56 -08002795 goto bail;
2796 }
2797
Sunil Mushranc271c5c2006-12-05 17:56:35 -08002798local:
Mark Fashehccd979b2005-12-15 14:31:24 -08002799 ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
2800 ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
2801
Joel Becker4670c462008-02-01 14:39:35 -08002802 osb->cconn = conn;
Mark Fashehccd979b2005-12-15 14:31:24 -08002803
2804 status = 0;
2805bail:
2806 if (status < 0) {
2807 ocfs2_dlm_shutdown_debug(osb);
Mark Fasheh34d024f2007-09-24 15:56:19 -07002808 if (osb->dc_task)
2809 kthread_stop(osb->dc_task);
Mark Fashehccd979b2005-12-15 14:31:24 -08002810 }
2811
2812 mlog_exit(status);
2813 return status;
2814}
2815
Joel Becker286eaa92008-02-01 15:03:57 -08002816void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
2817 int hangup_pending)
Mark Fashehccd979b2005-12-15 14:31:24 -08002818{
2819 mlog_entry_void();
2820
Mark Fashehccd979b2005-12-15 14:31:24 -08002821 ocfs2_drop_osb_locks(osb);
2822
Joel Becker4670c462008-02-01 14:39:35 -08002823 /*
2824 * Now that we have dropped all locks and ocfs2_dismount_volume()
2825 * has disabled recovery, the DLM won't be talking to us. It's
2826 * safe to tear things down before disconnecting the cluster.
2827 */
2828
Mark Fasheh34d024f2007-09-24 15:56:19 -07002829 if (osb->dc_task) {
2830 kthread_stop(osb->dc_task);
2831 osb->dc_task = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002832 }
2833
2834 ocfs2_lock_res_free(&osb->osb_super_lockres);
2835 ocfs2_lock_res_free(&osb->osb_rename_lockres);
2836
Joel Becker286eaa92008-02-01 15:03:57 -08002837 ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
Joel Becker4670c462008-02-01 14:39:35 -08002838 osb->cconn = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08002839
2840 ocfs2_dlm_shutdown_debug(osb);
2841
2842 mlog_exit_void();
2843}
2844
Joel Becker7431cd72008-02-01 12:15:37 -08002845static void ocfs2_unlock_ast(void *opaque, int error)
Mark Fashehccd979b2005-12-15 14:31:24 -08002846{
2847 struct ocfs2_lock_res *lockres = opaque;
2848 unsigned long flags;
2849
2850 mlog_entry_void();
2851
2852 mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name,
2853 lockres->l_unlock_action);
2854
2855 spin_lock_irqsave(&lockres->l_lock, flags);
Joel Beckerde551242008-02-01 14:45:08 -08002856 if (error) {
Joel Becker7431cd72008-02-01 12:15:37 -08002857 mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
2858 "unlock_action %d\n", error, lockres->l_name,
Mark Fashehccd979b2005-12-15 14:31:24 -08002859 lockres->l_unlock_action);
2860 spin_unlock_irqrestore(&lockres->l_lock, flags);
2861 return;
2862 }
2863
2864 switch(lockres->l_unlock_action) {
2865 case OCFS2_UNLOCK_CANCEL_CONVERT:
2866 mlog(0, "Cancel convert success for %s\n", lockres->l_name);
2867 lockres->l_action = OCFS2_AST_INVALID;
Sunil Mushrana4b91962009-01-29 17:12:31 -08002868 /* Downconvert thread may have requeued this lock, we
2869 * need to wake it. */
2870 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2871 ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
Mark Fashehccd979b2005-12-15 14:31:24 -08002872 break;
2873 case OCFS2_UNLOCK_DROP_LOCK:
Joel Beckerbd3e7612008-02-01 12:14:57 -08002874 lockres->l_level = DLM_LOCK_IV;
Mark Fashehccd979b2005-12-15 14:31:24 -08002875 break;
2876 default:
2877 BUG();
2878 }
2879
2880 lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
Mark Fashehccd979b2005-12-15 14:31:24 -08002881 lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
Mark Fashehccd979b2005-12-15 14:31:24 -08002882 wake_up(&lockres->l_event);
David Teigland07f9eeb2008-11-17 12:28:48 -06002883 spin_unlock_irqrestore(&lockres->l_lock, flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08002884
2885 mlog_exit_void();
2886}
2887
Mark Fashehccd979b2005-12-15 14:31:24 -08002888static int ocfs2_drop_lock(struct ocfs2_super *osb,
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07002889 struct ocfs2_lock_res *lockres)
Mark Fashehccd979b2005-12-15 14:31:24 -08002890{
Joel Becker7431cd72008-02-01 12:15:37 -08002891 int ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08002892 unsigned long flags;
Joel Beckerbd3e7612008-02-01 12:14:57 -08002893 u32 lkm_flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002894
2895 /* We didn't get anywhere near actually using this lockres. */
2896 if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
2897 goto out;
2898
Mark Fashehb80fc012006-09-12 22:08:14 -07002899 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
Joel Beckerbd3e7612008-02-01 12:14:57 -08002900 lkm_flags |= DLM_LKF_VALBLK;
Mark Fashehb80fc012006-09-12 22:08:14 -07002901
Mark Fashehccd979b2005-12-15 14:31:24 -08002902 spin_lock_irqsave(&lockres->l_lock, flags);
2903
2904 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
2905 "lockres %s, flags 0x%lx\n",
2906 lockres->l_name, lockres->l_flags);
2907
2908 while (lockres->l_flags & OCFS2_LOCK_BUSY) {
2909 mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
2910 "%u, unlock_action = %u\n",
2911 lockres->l_name, lockres->l_flags, lockres->l_action,
2912 lockres->l_unlock_action);
2913
2914 spin_unlock_irqrestore(&lockres->l_lock, flags);
2915
2916 /* XXX: Today we just wait on any busy
2917 * locks... Perhaps we need to cancel converts in the
2918 * future? */
2919 ocfs2_wait_on_busy_lock(lockres);
2920
2921 spin_lock_irqsave(&lockres->l_lock, flags);
2922 }
2923
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07002924 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
2925 if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
Joel Beckerbd3e7612008-02-01 12:14:57 -08002926 lockres->l_level == DLM_LOCK_EX &&
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07002927 !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
2928 lockres->l_ops->set_lvb(lockres);
2929 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002930
2931 if (lockres->l_flags & OCFS2_LOCK_BUSY)
2932 mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
2933 lockres->l_name);
2934 if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
2935 mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
2936
2937 if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
2938 spin_unlock_irqrestore(&lockres->l_lock, flags);
2939 goto out;
2940 }
2941
2942 lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
2943
2944 /* make sure we never get here while waiting for an ast to
2945 * fire. */
2946 BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
2947
2948 /* is this necessary? */
2949 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
2950 lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
2951 spin_unlock_irqrestore(&lockres->l_lock, flags);
2952
2953 mlog(0, "lock %s\n", lockres->l_name);
2954
Joel Becker4670c462008-02-01 14:39:35 -08002955 ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags,
Joel Becker7431cd72008-02-01 12:15:37 -08002956 lockres);
2957 if (ret) {
2958 ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08002959 mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
Joel Beckercf0acdc2008-01-29 16:59:55 -08002960 ocfs2_dlm_dump_lksb(&lockres->l_lksb);
Mark Fashehccd979b2005-12-15 14:31:24 -08002961 BUG();
2962 }
Coly Li73ac36e2009-01-07 18:09:16 -08002963 mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
Mark Fashehccd979b2005-12-15 14:31:24 -08002964 lockres->l_name);
2965
2966 ocfs2_wait_on_busy_lock(lockres);
2967out:
2968 mlog_exit(0);
2969 return 0;
2970}
2971
2972/* Mark the lockres as being dropped. It will no longer be
2973 * queued if blocking, but we still may have to wait on it
Mark Fasheh34d024f2007-09-24 15:56:19 -07002974 * being dequeued from the downconvert thread before we can consider
Mark Fashehccd979b2005-12-15 14:31:24 -08002975 * it safe to drop.
2976 *
2977 * You can *not* attempt to call cluster_lock on this lockres anymore. */
2978void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres)
2979{
2980 int status;
2981 struct ocfs2_mask_waiter mw;
2982 unsigned long flags;
2983
2984 ocfs2_init_mask_waiter(&mw);
2985
2986 spin_lock_irqsave(&lockres->l_lock, flags);
2987 lockres->l_flags |= OCFS2_LOCK_FREEING;
2988 while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
2989 lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
2990 spin_unlock_irqrestore(&lockres->l_lock, flags);
2991
2992 mlog(0, "Waiting on lockres %s\n", lockres->l_name);
2993
2994 status = ocfs2_wait_for_mask(&mw);
2995 if (status)
2996 mlog_errno(status);
2997
2998 spin_lock_irqsave(&lockres->l_lock, flags);
2999 }
3000 spin_unlock_irqrestore(&lockres->l_lock, flags);
3001}
3002
Mark Fashehd680efe2006-09-08 14:14:34 -07003003void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
3004 struct ocfs2_lock_res *lockres)
3005{
3006 int ret;
3007
3008 ocfs2_mark_lockres_freeing(lockres);
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07003009 ret = ocfs2_drop_lock(osb, lockres);
Mark Fashehd680efe2006-09-08 14:14:34 -07003010 if (ret)
3011 mlog_errno(ret);
3012}
3013
Mark Fashehccd979b2005-12-15 14:31:24 -08003014static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
3015{
Mark Fashehd680efe2006-09-08 14:14:34 -07003016 ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
3017 ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003018}
3019
Mark Fashehccd979b2005-12-15 14:31:24 -08003020int ocfs2_drop_inode_locks(struct inode *inode)
3021{
3022 int status, err;
Mark Fashehccd979b2005-12-15 14:31:24 -08003023
3024 mlog_entry_void();
3025
3026 /* No need to call ocfs2_mark_lockres_freeing here -
3027 * ocfs2_clear_inode has done it for us. */
3028
3029 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
Tiger Yang50008632007-03-20 16:01:38 -07003030 &OCFS2_I(inode)->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003031 if (err < 0)
3032 mlog_errno(err);
3033
3034 status = err;
3035
3036 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
Mark Fashehe63aecb62007-10-18 15:30:42 -07003037 &OCFS2_I(inode)->ip_inode_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003038 if (err < 0)
3039 mlog_errno(err);
3040 if (err < 0 && !status)
3041 status = err;
3042
3043 err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
Mark Fasheh0d5dc6c2006-09-14 14:44:51 -07003044 &OCFS2_I(inode)->ip_rw_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003045 if (err < 0)
3046 mlog_errno(err);
3047 if (err < 0 && !status)
3048 status = err;
3049
3050 mlog_exit(status);
3051 return status;
3052}
3053
Joel Beckerde551242008-02-01 14:45:08 -08003054static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
3055 int new_level)
Mark Fashehccd979b2005-12-15 14:31:24 -08003056{
3057 assert_spin_locked(&lockres->l_lock);
3058
Joel Beckerbd3e7612008-02-01 12:14:57 -08003059 BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
Mark Fashehccd979b2005-12-15 14:31:24 -08003060
3061 if (lockres->l_level <= new_level) {
Joel Beckerbd3e7612008-02-01 12:14:57 -08003062 mlog(ML_ERROR, "lockres->l_level (%d) <= new_level (%d)\n",
Mark Fashehccd979b2005-12-15 14:31:24 -08003063 lockres->l_level, new_level);
3064 BUG();
3065 }
3066
3067 mlog(0, "lock %s, new_level = %d, l_blocking = %d\n",
3068 lockres->l_name, new_level, lockres->l_blocking);
3069
3070 lockres->l_action = OCFS2_AST_DOWNCONVERT;
3071 lockres->l_requested = new_level;
3072 lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
Joel Beckerde551242008-02-01 14:45:08 -08003073 return lockres_set_pending(lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003074}
3075
3076static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
3077 struct ocfs2_lock_res *lockres,
3078 int new_level,
Joel Beckerde551242008-02-01 14:45:08 -08003079 int lvb,
3080 unsigned int generation)
Mark Fashehccd979b2005-12-15 14:31:24 -08003081{
Joel Beckerbd3e7612008-02-01 12:14:57 -08003082 int ret;
3083 u32 dlm_flags = DLM_LKF_CONVERT;
Mark Fashehccd979b2005-12-15 14:31:24 -08003084
3085 mlog_entry_void();
3086
3087 if (lvb)
Joel Beckerbd3e7612008-02-01 12:14:57 -08003088 dlm_flags |= DLM_LKF_VALBLK;
Mark Fashehccd979b2005-12-15 14:31:24 -08003089
Joel Becker4670c462008-02-01 14:39:35 -08003090 ret = ocfs2_dlm_lock(osb->cconn,
Joel Becker7431cd72008-02-01 12:15:37 -08003091 new_level,
3092 &lockres->l_lksb,
3093 dlm_flags,
3094 lockres->l_name,
3095 OCFS2_LOCK_ID_MAX_LEN - 1,
3096 lockres);
Joel Beckerde551242008-02-01 14:45:08 -08003097 lockres_clear_pending(lockres, generation, osb);
Joel Becker7431cd72008-02-01 12:15:37 -08003098 if (ret) {
3099 ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003100 ocfs2_recover_from_dlm_error(lockres, 1);
3101 goto bail;
3102 }
3103
3104 ret = 0;
3105bail:
3106 mlog_exit(ret);
3107 return ret;
3108}
3109
Joel Becker24ef1812008-01-29 17:37:32 -08003110/* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
Mark Fashehccd979b2005-12-15 14:31:24 -08003111static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
3112 struct ocfs2_lock_res *lockres)
3113{
3114 assert_spin_locked(&lockres->l_lock);
3115
3116 mlog_entry_void();
3117 mlog(0, "lock %s\n", lockres->l_name);
3118
3119 if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
3120 /* If we're already trying to cancel a lock conversion
3121 * then just drop the spinlock and allow the caller to
3122 * requeue this lock. */
3123
3124 mlog(0, "Lockres %s, skip convert\n", lockres->l_name);
3125 return 0;
3126 }
3127
3128 /* were we in a convert when we got the bast fire? */
3129 BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
3130 lockres->l_action != OCFS2_AST_DOWNCONVERT);
3131 /* set things up for the unlockast to know to just
3132 * clear out the ast_action and unset busy, etc. */
3133 lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
3134
3135 mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
3136 "lock %s, invalid flags: 0x%lx\n",
3137 lockres->l_name, lockres->l_flags);
3138
3139 return 1;
3140}
3141
3142static int ocfs2_cancel_convert(struct ocfs2_super *osb,
3143 struct ocfs2_lock_res *lockres)
3144{
3145 int ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08003146
3147 mlog_entry_void();
3148 mlog(0, "lock %s\n", lockres->l_name);
3149
Joel Becker4670c462008-02-01 14:39:35 -08003150 ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
Joel Becker7431cd72008-02-01 12:15:37 -08003151 DLM_LKF_CANCEL, lockres);
3152 if (ret) {
3153 ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08003154 ocfs2_recover_from_dlm_error(lockres, 0);
3155 }
3156
Joel Becker24ef1812008-01-29 17:37:32 -08003157 mlog(0, "lock %s return from ocfs2_dlm_unlock\n", lockres->l_name);
Mark Fashehccd979b2005-12-15 14:31:24 -08003158
3159 mlog_exit(ret);
3160 return ret;
3161}
3162
Mark Fashehb5e500e2006-09-13 22:01:16 -07003163static int ocfs2_unblock_lock(struct ocfs2_super *osb,
3164 struct ocfs2_lock_res *lockres,
3165 struct ocfs2_unblock_ctl *ctl)
Mark Fashehccd979b2005-12-15 14:31:24 -08003166{
3167 unsigned long flags;
3168 int blocking;
3169 int new_level;
3170 int ret = 0;
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -07003171 int set_lvb = 0;
Joel Beckerde551242008-02-01 14:45:08 -08003172 unsigned int gen;
Mark Fashehccd979b2005-12-15 14:31:24 -08003173
3174 mlog_entry_void();
3175
3176 spin_lock_irqsave(&lockres->l_lock, flags);
3177
3178 BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
3179
3180recheck:
3181 if (lockres->l_flags & OCFS2_LOCK_BUSY) {
Joel Beckerde551242008-02-01 14:45:08 -08003182 /* XXX
3183 * This is a *big* race. The OCFS2_LOCK_PENDING flag
3184 * exists entirely for one reason - another thread has set
3185 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
3186 *
3187 * If we do ocfs2_cancel_convert() before the other thread
3188 * calls dlm_lock(), our cancel will do nothing. We will
3189 * get no ast, and we will have no way of knowing the
3190 * cancel failed. Meanwhile, the other thread will call
3191 * into dlm_lock() and wait...forever.
3192 *
3193 * Why forever? Because another node has asked for the
3194 * lock first; that's why we're here in unblock_lock().
3195 *
3196 * The solution is OCFS2_LOCK_PENDING. When PENDING is
3197 * set, we just requeue the unblock. Only when the other
3198 * thread has called dlm_lock() and cleared PENDING will
3199 * we then cancel their request.
3200 *
3201 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
3202 * at the same time they set OCFS2_DLM_BUSY. They must
3203 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
3204 */
3205 if (lockres->l_flags & OCFS2_LOCK_PENDING)
3206 goto leave_requeue;
3207
Mark Fashehd680efe2006-09-08 14:14:34 -07003208 ctl->requeue = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -08003209 ret = ocfs2_prepare_cancel_convert(osb, lockres);
3210 spin_unlock_irqrestore(&lockres->l_lock, flags);
3211 if (ret) {
3212 ret = ocfs2_cancel_convert(osb, lockres);
3213 if (ret < 0)
3214 mlog_errno(ret);
3215 }
3216 goto leave;
3217 }
3218
3219 /* if we're blocking an exclusive and we have *any* holders,
3220 * then requeue. */
Joel Beckerbd3e7612008-02-01 12:14:57 -08003221 if ((lockres->l_blocking == DLM_LOCK_EX)
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003222 && (lockres->l_ex_holders || lockres->l_ro_holders))
3223 goto leave_requeue;
Mark Fashehccd979b2005-12-15 14:31:24 -08003224
3225 /* If it's a PR we're blocking, then only
3226 * requeue if we've got any EX holders */
Joel Beckerbd3e7612008-02-01 12:14:57 -08003227 if (lockres->l_blocking == DLM_LOCK_PR &&
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003228 lockres->l_ex_holders)
3229 goto leave_requeue;
3230
3231 /*
3232 * Can we get a lock in this state if the holder counts are
3233 * zero? The meta data unblock code used to check this.
3234 */
3235 if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
3236 && (lockres->l_flags & OCFS2_LOCK_REFRESHING))
3237 goto leave_requeue;
Mark Fashehccd979b2005-12-15 14:31:24 -08003238
Mark Fasheh16d5b952006-09-13 21:10:12 -07003239 new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
3240
3241 if (lockres->l_ops->check_downconvert
3242 && !lockres->l_ops->check_downconvert(lockres, new_level))
3243 goto leave_requeue;
3244
Mark Fashehccd979b2005-12-15 14:31:24 -08003245 /* If we get here, then we know that there are no more
3246 * incompatible holders (and anyone asking for an incompatible
3247 * lock is blocked). We can now downconvert the lock */
Mark Fashehcc567d82006-09-13 21:52:21 -07003248 if (!lockres->l_ops->downconvert_worker)
Mark Fashehccd979b2005-12-15 14:31:24 -08003249 goto downconvert;
3250
3251 /* Some lockres types want to do a bit of work before
3252 * downconverting a lock. Allow that here. The worker function
3253 * may sleep, so we save off a copy of what we're blocking as
3254 * it may change while we're not holding the spin lock. */
3255 blocking = lockres->l_blocking;
3256 spin_unlock_irqrestore(&lockres->l_lock, flags);
3257
Mark Fashehcc567d82006-09-13 21:52:21 -07003258 ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
Mark Fashehd680efe2006-09-08 14:14:34 -07003259
3260 if (ctl->unblock_action == UNBLOCK_STOP_POST)
3261 goto leave;
Mark Fashehccd979b2005-12-15 14:31:24 -08003262
3263 spin_lock_irqsave(&lockres->l_lock, flags);
3264 if (blocking != lockres->l_blocking) {
3265 /* If this changed underneath us, then we can't drop
3266 * it just yet. */
3267 goto recheck;
3268 }
3269
3270downconvert:
Mark Fashehd680efe2006-09-08 14:14:34 -07003271 ctl->requeue = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08003272
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -07003273 if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
Joel Beckerbd3e7612008-02-01 12:14:57 -08003274 if (lockres->l_level == DLM_LOCK_EX)
Mark Fasheh5ef0d4e2006-09-13 21:21:52 -07003275 set_lvb = 1;
3276
3277 /*
3278 * We only set the lvb if the lock has been fully
3279 * refreshed - otherwise we risk setting stale
3280 * data. Otherwise, there's no need to actually clear
3281 * out the lvb here as it's value is still valid.
3282 */
3283 if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
3284 lockres->l_ops->set_lvb(lockres);
3285 }
3286
Joel Beckerde551242008-02-01 14:45:08 -08003287 gen = ocfs2_prepare_downconvert(lockres, new_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08003288 spin_unlock_irqrestore(&lockres->l_lock, flags);
Joel Beckerde551242008-02-01 14:45:08 -08003289 ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
3290 gen);
3291
Mark Fashehccd979b2005-12-15 14:31:24 -08003292leave:
3293 mlog_exit(ret);
3294 return ret;
Mark Fashehf7fbfdd2006-09-13 21:02:29 -07003295
3296leave_requeue:
3297 spin_unlock_irqrestore(&lockres->l_lock, flags);
3298 ctl->requeue = 1;
3299
3300 mlog_exit(0);
3301 return 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08003302}
3303
Mark Fashehd680efe2006-09-08 14:14:34 -07003304static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
3305 int blocking)
Mark Fashehccd979b2005-12-15 14:31:24 -08003306{
3307 struct inode *inode;
3308 struct address_space *mapping;
3309
Mark Fashehccd979b2005-12-15 14:31:24 -08003310 inode = ocfs2_lock_res_inode(lockres);
3311 mapping = inode->i_mapping;
3312
Mark Fasheh1044e402008-02-28 17:16:03 -08003313 if (!S_ISREG(inode->i_mode))
Mark Fashehf1f54062007-10-18 15:13:59 -07003314 goto out;
3315
Mark Fasheh7f4a2a92006-12-11 11:06:36 -08003316 /*
3317 * We need this before the filemap_fdatawrite() so that it can
3318 * transfer the dirty bit from the PTE to the
3319 * page. Unfortunately this means that even for EX->PR
3320 * downconverts, we'll lose our mappings and have to build
3321 * them up again.
3322 */
3323 unmap_mapping_range(mapping, 0, 0, 0);
3324
Mark Fashehccd979b2005-12-15 14:31:24 -08003325 if (filemap_fdatawrite(mapping)) {
Mark Fashehb0697052006-03-03 10:24:33 -08003326 mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
3327 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08003328 }
3329 sync_mapping_buffers(mapping);
Joel Beckerbd3e7612008-02-01 12:14:57 -08003330 if (blocking == DLM_LOCK_EX) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003331 truncate_inode_pages(mapping, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08003332 } else {
3333 /* We only need to wait on the I/O if we're not also
3334 * truncating pages because truncate_inode_pages waits
3335 * for us above. We don't truncate pages if we're
3336 * blocking anything < EXMODE because we want to keep
3337 * them around in that case. */
3338 filemap_fdatawait(mapping);
3339 }
3340
Mark Fashehf1f54062007-10-18 15:13:59 -07003341out:
Mark Fashehd680efe2006-09-08 14:14:34 -07003342 return UNBLOCK_CONTINUE;
Mark Fashehccd979b2005-12-15 14:31:24 -08003343}
3344
Mark Fasheh810d5ae2006-09-13 21:39:52 -07003345static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
3346 int new_level)
3347{
3348 struct inode *inode = ocfs2_lock_res_inode(lockres);
3349 int checkpointed = ocfs2_inode_fully_checkpointed(inode);
3350
Joel Beckerbd3e7612008-02-01 12:14:57 -08003351 BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
3352 BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
Mark Fasheh810d5ae2006-09-13 21:39:52 -07003353
3354 if (checkpointed)
3355 return 1;
3356
3357 ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb));
3358 return 0;
3359}
3360
3361static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
3362{
3363 struct inode *inode = ocfs2_lock_res_inode(lockres);
3364
3365 __ocfs2_stuff_meta_lvb(inode);
3366}
3367
Mark Fashehd680efe2006-09-08 14:14:34 -07003368/*
3369 * Does the final reference drop on our dentry lock. Right now this
Mark Fasheh34d024f2007-09-24 15:56:19 -07003370 * happens in the downconvert thread, but we could choose to simplify the
Mark Fashehd680efe2006-09-08 14:14:34 -07003371 * dlmglue API and push these off to the ocfs2_wq in the future.
3372 */
3373static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
3374 struct ocfs2_lock_res *lockres)
3375{
3376 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3377 ocfs2_dentry_lock_put(osb, dl);
3378}
3379
3380/*
3381 * d_delete() matching dentries before the lock downconvert.
3382 *
3383 * At this point, any process waiting to destroy the
3384 * dentry_lock due to last ref count is stopped by the
3385 * OCFS2_LOCK_QUEUED flag.
3386 *
3387 * We have two potential problems
3388 *
3389 * 1) If we do the last reference drop on our dentry_lock (via dput)
3390 * we'll wind up in ocfs2_release_dentry_lock(), waiting on
3391 * the downconvert to finish. Instead we take an elevated
3392 * reference and push the drop until after we've completed our
3393 * unblock processing.
3394 *
3395 * 2) There might be another process with a final reference,
3396 * waiting on us to finish processing. If this is the case, we
3397 * detect it and exit out - there's no more dentries anyway.
3398 */
3399static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
3400 int blocking)
3401{
3402 struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
3403 struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
3404 struct dentry *dentry;
3405 unsigned long flags;
3406 int extra_ref = 0;
3407
3408 /*
3409 * This node is blocking another node from getting a read
3410 * lock. This happens when we've renamed within a
3411 * directory. We've forced the other nodes to d_delete(), but
3412 * we never actually dropped our lock because it's still
3413 * valid. The downconvert code will retain a PR for this node,
3414 * so there's no further work to do.
3415 */
Joel Beckerbd3e7612008-02-01 12:14:57 -08003416 if (blocking == DLM_LOCK_PR)
Mark Fashehd680efe2006-09-08 14:14:34 -07003417 return UNBLOCK_CONTINUE;
3418
3419 /*
3420 * Mark this inode as potentially orphaned. The code in
3421 * ocfs2_delete_inode() will figure out whether it actually
3422 * needs to be freed or not.
3423 */
3424 spin_lock(&oi->ip_lock);
3425 oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
3426 spin_unlock(&oi->ip_lock);
3427
3428 /*
3429 * Yuck. We need to make sure however that the check of
3430 * OCFS2_LOCK_FREEING and the extra reference are atomic with
3431 * respect to a reference decrement or the setting of that
3432 * flag.
3433 */
3434 spin_lock_irqsave(&lockres->l_lock, flags);
3435 spin_lock(&dentry_attach_lock);
3436 if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
3437 && dl->dl_count) {
3438 dl->dl_count++;
3439 extra_ref = 1;
3440 }
3441 spin_unlock(&dentry_attach_lock);
3442 spin_unlock_irqrestore(&lockres->l_lock, flags);
3443
3444 mlog(0, "extra_ref = %d\n", extra_ref);
3445
3446 /*
3447 * We have a process waiting on us in ocfs2_dentry_iput(),
3448 * which means we can't have any more outstanding
3449 * aliases. There's no need to do any more work.
3450 */
3451 if (!extra_ref)
3452 return UNBLOCK_CONTINUE;
3453
3454 spin_lock(&dentry_attach_lock);
3455 while (1) {
3456 dentry = ocfs2_find_local_alias(dl->dl_inode,
3457 dl->dl_parent_blkno, 1);
3458 if (!dentry)
3459 break;
3460 spin_unlock(&dentry_attach_lock);
3461
3462 mlog(0, "d_delete(%.*s);\n", dentry->d_name.len,
3463 dentry->d_name.name);
3464
3465 /*
3466 * The following dcache calls may do an
3467 * iput(). Normally we don't want that from the
3468 * downconverting thread, but in this case it's ok
3469 * because the requesting node already has an
3470 * exclusive lock on the inode, so it can't be queued
3471 * for a downconvert.
3472 */
3473 d_delete(dentry);
3474 dput(dentry);
3475
3476 spin_lock(&dentry_attach_lock);
3477 }
3478 spin_unlock(&dentry_attach_lock);
3479
3480 /*
3481 * If we are the last holder of this dentry lock, there is no
3482 * reason to downconvert so skip straight to the unlock.
3483 */
3484 if (dl->dl_count == 1)
3485 return UNBLOCK_STOP_POST;
3486
3487 return UNBLOCK_CONTINUE_POST;
3488}
3489
Jan Kara9e33d692008-08-25 19:56:50 +02003490static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
3491{
3492 struct ocfs2_qinfo_lvb *lvb;
3493 struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
3494 struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3495 oinfo->dqi_gi.dqi_type);
3496
3497 mlog_entry_void();
3498
Mark Fasheha641dc22008-12-24 16:03:48 -08003499 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Jan Kara9e33d692008-08-25 19:56:50 +02003500 lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
3501 lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
3502 lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
3503 lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
3504 lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
3505 lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
3506 lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
3507
3508 mlog_exit_void();
3509}
3510
3511void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3512{
3513 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3514 struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3515 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3516
3517 mlog_entry_void();
3518 if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
3519 ocfs2_cluster_unlock(osb, lockres, level);
3520 mlog_exit_void();
3521}
3522
3523static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
3524{
3525 struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
3526 oinfo->dqi_gi.dqi_type);
3527 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3528 struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
Joel Becker85eb8b72008-11-25 15:31:27 +01003529 struct buffer_head *bh = NULL;
Jan Kara9e33d692008-08-25 19:56:50 +02003530 struct ocfs2_global_disk_dqinfo *gdinfo;
3531 int status = 0;
3532
3533 if (lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
3534 info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
3535 info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
3536 oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
3537 oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
3538 oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
3539 oinfo->dqi_gi.dqi_free_entry =
3540 be32_to_cpu(lvb->lvb_free_entry);
3541 } else {
Joel Becker85eb8b72008-11-25 15:31:27 +01003542 status = ocfs2_read_quota_block(oinfo->dqi_gqinode, 0, &bh);
3543 if (status) {
Jan Kara9e33d692008-08-25 19:56:50 +02003544 mlog_errno(status);
3545 goto bail;
3546 }
3547 gdinfo = (struct ocfs2_global_disk_dqinfo *)
3548 (bh->b_data + OCFS2_GLOBAL_INFO_OFF);
3549 info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
3550 info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
3551 oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
3552 oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
3553 oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
3554 oinfo->dqi_gi.dqi_free_entry =
3555 le32_to_cpu(gdinfo->dqi_free_entry);
3556 brelse(bh);
3557 ocfs2_track_lock_refresh(lockres);
3558 }
3559
3560bail:
3561 return status;
3562}
3563
3564/* Lock quota info, this function expects at least shared lock on the quota file
3565 * so that we can safely refresh quota info from disk. */
3566int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
3567{
3568 struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
3569 struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
3570 int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
3571 int status = 0;
3572
3573 mlog_entry_void();
3574
3575 /* On RO devices, locking really isn't needed... */
3576 if (ocfs2_is_hard_readonly(osb)) {
3577 if (ex)
3578 status = -EROFS;
3579 goto bail;
3580 }
3581 if (ocfs2_mount_local(osb))
3582 goto bail;
3583
3584 status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
3585 if (status < 0) {
3586 mlog_errno(status);
3587 goto bail;
3588 }
3589 if (!ocfs2_should_refresh_lock_res(lockres))
3590 goto bail;
3591 /* OK, we have the lock but we need to refresh the quota info */
3592 status = ocfs2_refresh_qinfo(oinfo);
3593 if (status)
3594 ocfs2_qinfo_unlock(oinfo, ex);
3595 ocfs2_complete_lock_res_refresh(lockres, status);
3596bail:
3597 mlog_exit(status);
3598 return status;
3599}
3600
Joel Becker4670c462008-02-01 14:39:35 -08003601/*
3602 * This is the filesystem locking protocol. It provides the lock handling
3603 * hooks for the underlying DLM. It has a maximum version number.
3604 * The version number allows interoperability with systems running at
3605 * the same major number and an equal or smaller minor number.
3606 *
3607 * Whenever the filesystem does new things with locks (adds or removes a
3608 * lock, orders them differently, does different things underneath a lock),
3609 * the version must be changed. The protocol is negotiated when joining
3610 * the dlm domain. A node may join the domain if its major version is
3611 * identical to all other nodes and its minor version is greater than
3612 * or equal to all other nodes. When its minor version is greater than
3613 * the other nodes, it will run at the minor version specified by the
3614 * other nodes.
3615 *
3616 * If a locking change is made that will not be compatible with older
3617 * versions, the major number must be increased and the minor version set
3618 * to zero. If a change merely adds a behavior that can be disabled when
3619 * speaking to older versions, the minor version must be increased. If a
3620 * change adds a fully backwards compatible change (eg, LVB changes that
3621 * are just ignored by older versions), the version does not need to be
3622 * updated.
3623 */
Joel Becker24ef1812008-01-29 17:37:32 -08003624static struct ocfs2_locking_protocol lproto = {
Joel Becker4670c462008-02-01 14:39:35 -08003625 .lp_max_version = {
3626 .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
3627 .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
3628 },
Joel Becker24ef1812008-01-29 17:37:32 -08003629 .lp_lock_ast = ocfs2_locking_ast,
3630 .lp_blocking_ast = ocfs2_blocking_ast,
3631 .lp_unlock_ast = ocfs2_unlock_ast,
3632};
3633
Joel Becker63e0c482008-01-30 16:58:36 -08003634void ocfs2_set_locking_protocol(void)
Joel Becker24ef1812008-01-29 17:37:32 -08003635{
Joel Becker63e0c482008-01-30 16:58:36 -08003636 ocfs2_stack_glue_set_locking_protocol(&lproto);
Joel Becker24ef1812008-01-29 17:37:32 -08003637}
3638
Joel Becker24ef1812008-01-29 17:37:32 -08003639
Adrian Bunk00600052008-01-29 00:11:41 +02003640static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
3641 struct ocfs2_lock_res *lockres)
Mark Fashehccd979b2005-12-15 14:31:24 -08003642{
3643 int status;
Mark Fashehd680efe2006-09-08 14:14:34 -07003644 struct ocfs2_unblock_ctl ctl = {0, 0,};
Mark Fashehccd979b2005-12-15 14:31:24 -08003645 unsigned long flags;
3646
3647 /* Our reference to the lockres in this function can be
3648 * considered valid until we remove the OCFS2_LOCK_QUEUED
3649 * flag. */
3650
3651 mlog_entry_void();
3652
3653 BUG_ON(!lockres);
3654 BUG_ON(!lockres->l_ops);
Mark Fashehccd979b2005-12-15 14:31:24 -08003655
3656 mlog(0, "lockres %s blocked.\n", lockres->l_name);
3657
3658 /* Detect whether a lock has been marked as going away while
Mark Fasheh34d024f2007-09-24 15:56:19 -07003659 * the downconvert thread was processing other things. A lock can
Mark Fashehccd979b2005-12-15 14:31:24 -08003660 * still be marked with OCFS2_LOCK_FREEING after this check,
3661 * but short circuiting here will still save us some
3662 * performance. */
3663 spin_lock_irqsave(&lockres->l_lock, flags);
3664 if (lockres->l_flags & OCFS2_LOCK_FREEING)
3665 goto unqueue;
3666 spin_unlock_irqrestore(&lockres->l_lock, flags);
3667
Mark Fashehb5e500e2006-09-13 22:01:16 -07003668 status = ocfs2_unblock_lock(osb, lockres, &ctl);
Mark Fashehccd979b2005-12-15 14:31:24 -08003669 if (status < 0)
3670 mlog_errno(status);
3671
3672 spin_lock_irqsave(&lockres->l_lock, flags);
3673unqueue:
Mark Fashehd680efe2006-09-08 14:14:34 -07003674 if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
Mark Fashehccd979b2005-12-15 14:31:24 -08003675 lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
3676 } else
3677 ocfs2_schedule_blocked_lock(osb, lockres);
3678
3679 mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name,
Mark Fashehd680efe2006-09-08 14:14:34 -07003680 ctl.requeue ? "yes" : "no");
Mark Fashehccd979b2005-12-15 14:31:24 -08003681 spin_unlock_irqrestore(&lockres->l_lock, flags);
3682
Mark Fashehd680efe2006-09-08 14:14:34 -07003683 if (ctl.unblock_action != UNBLOCK_CONTINUE
3684 && lockres->l_ops->post_unlock)
3685 lockres->l_ops->post_unlock(osb, lockres);
3686
Mark Fashehccd979b2005-12-15 14:31:24 -08003687 mlog_exit_void();
3688}
3689
3690static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
3691 struct ocfs2_lock_res *lockres)
3692{
3693 mlog_entry_void();
3694
3695 assert_spin_locked(&lockres->l_lock);
3696
3697 if (lockres->l_flags & OCFS2_LOCK_FREEING) {
3698 /* Do not schedule a lock for downconvert when it's on
3699 * the way to destruction - any nodes wanting access
3700 * to the resource will get it soon. */
3701 mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n",
3702 lockres->l_name, lockres->l_flags);
3703 return;
3704 }
3705
3706 lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
3707
Mark Fasheh34d024f2007-09-24 15:56:19 -07003708 spin_lock(&osb->dc_task_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08003709 if (list_empty(&lockres->l_blocked_list)) {
3710 list_add_tail(&lockres->l_blocked_list,
3711 &osb->blocked_lock_list);
3712 osb->blocked_lock_count++;
3713 }
Mark Fasheh34d024f2007-09-24 15:56:19 -07003714 spin_unlock(&osb->dc_task_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08003715
3716 mlog_exit_void();
3717}
Mark Fasheh34d024f2007-09-24 15:56:19 -07003718
3719static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
3720{
3721 unsigned long processed;
3722 struct ocfs2_lock_res *lockres;
3723
3724 mlog_entry_void();
3725
3726 spin_lock(&osb->dc_task_lock);
3727 /* grab this early so we know to try again if a state change and
3728 * wake happens part-way through our work */
3729 osb->dc_work_sequence = osb->dc_wake_sequence;
3730
3731 processed = osb->blocked_lock_count;
3732 while (processed) {
3733 BUG_ON(list_empty(&osb->blocked_lock_list));
3734
3735 lockres = list_entry(osb->blocked_lock_list.next,
3736 struct ocfs2_lock_res, l_blocked_list);
3737 list_del_init(&lockres->l_blocked_list);
3738 osb->blocked_lock_count--;
3739 spin_unlock(&osb->dc_task_lock);
3740
3741 BUG_ON(!processed);
3742 processed--;
3743
3744 ocfs2_process_blocked_lock(osb, lockres);
3745
3746 spin_lock(&osb->dc_task_lock);
3747 }
3748 spin_unlock(&osb->dc_task_lock);
3749
3750 mlog_exit_void();
3751}
3752
3753static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
3754{
3755 int empty = 0;
3756
3757 spin_lock(&osb->dc_task_lock);
3758 if (list_empty(&osb->blocked_lock_list))
3759 empty = 1;
3760
3761 spin_unlock(&osb->dc_task_lock);
3762 return empty;
3763}
3764
3765static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
3766{
3767 int should_wake = 0;
3768
3769 spin_lock(&osb->dc_task_lock);
3770 if (osb->dc_work_sequence != osb->dc_wake_sequence)
3771 should_wake = 1;
3772 spin_unlock(&osb->dc_task_lock);
3773
3774 return should_wake;
3775}
3776
Adrian Bunk200bfae2008-02-17 10:20:38 +02003777static int ocfs2_downconvert_thread(void *arg)
Mark Fasheh34d024f2007-09-24 15:56:19 -07003778{
3779 int status = 0;
3780 struct ocfs2_super *osb = arg;
3781
3782 /* only quit once we've been asked to stop and there is no more
3783 * work available */
3784 while (!(kthread_should_stop() &&
3785 ocfs2_downconvert_thread_lists_empty(osb))) {
3786
3787 wait_event_interruptible(osb->dc_event,
3788 ocfs2_downconvert_thread_should_wake(osb) ||
3789 kthread_should_stop());
3790
3791 mlog(0, "downconvert_thread: awoken\n");
3792
3793 ocfs2_downconvert_thread_do_work(osb);
3794 }
3795
3796 osb->dc_task = NULL;
3797 return status;
3798}
3799
3800void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
3801{
3802 spin_lock(&osb->dc_task_lock);
3803 /* make sure the voting thread gets a swipe at whatever changes
3804 * the caller may have made to the voting state */
3805 osb->dc_wake_sequence++;
3806 spin_unlock(&osb->dc_task_lock);
3807 wake_up(&osb->dc_event);
3808}