blob: 1c9713cceb390279aa4d7c7f49b17bddffd8cc59 [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 * inode.c
5 *
6 * vfs' aops, fops, dops and iops
7 *
8 * Copyright (C) 2002, 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/fs.h>
27#include <linux/types.h>
28#include <linux/slab.h>
29#include <linux/highmem.h>
30#include <linux/pagemap.h>
Jan Karaa90714c2008-10-09 19:38:40 +020031#include <linux/quotaops.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080032
33#include <asm/byteorder.h>
34
35#define MLOG_MASK_PREFIX ML_INODE
36#include <cluster/masklog.h>
37
38#include "ocfs2.h"
39
40#include "alloc.h"
Mark Fasheh9b7895e2008-11-12 16:27:44 -080041#include "dir.h"
Joel Beckerd6b32bb2008-10-17 14:55:01 -070042#include "blockcheck.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080043#include "dlmglue.h"
44#include "extent_map.h"
45#include "file.h"
Mark Fashehb4df6ed2006-02-22 17:35:08 -080046#include "heartbeat.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080047#include "inode.h"
48#include "journal.h"
49#include "namei.h"
50#include "suballoc.h"
51#include "super.h"
52#include "symlink.h"
53#include "sysfile.h"
54#include "uptodate.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080055#include "xattr.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080056
57#include "buffer_head_io.h"
58
Mark Fashehccd979b2005-12-15 14:31:24 -080059struct ocfs2_find_inode_args
60{
61 u64 fi_blkno;
62 unsigned long fi_ino;
63 unsigned int fi_flags;
Jan Kara5fa06132008-01-11 00:11:45 +010064 unsigned int fi_sysfile_type;
Mark Fashehccd979b2005-12-15 14:31:24 -080065};
66
Jan Kara5fa06132008-01-11 00:11:45 +010067static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
68
Mark Fashehccd979b2005-12-15 14:31:24 -080069static int ocfs2_read_locked_inode(struct inode *inode,
70 struct ocfs2_find_inode_args *args);
71static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
72static int ocfs2_find_actor(struct inode *inode, void *opaque);
73static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
74 struct inode *inode,
75 struct buffer_head *fe_bh);
76
Herbert Poetzlca4d1472006-07-03 17:27:12 -070077void ocfs2_set_inode_flags(struct inode *inode)
78{
79 unsigned int flags = OCFS2_I(inode)->ip_attr;
80
81 inode->i_flags &= ~(S_IMMUTABLE |
82 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
83
84 if (flags & OCFS2_IMMUTABLE_FL)
85 inode->i_flags |= S_IMMUTABLE;
86
87 if (flags & OCFS2_SYNC_FL)
88 inode->i_flags |= S_SYNC;
89 if (flags & OCFS2_APPEND_FL)
90 inode->i_flags |= S_APPEND;
91 if (flags & OCFS2_NOATIME_FL)
92 inode->i_flags |= S_NOATIME;
93 if (flags & OCFS2_DIRSYNC_FL)
94 inode->i_flags |= S_DIRSYNC;
95}
96
Jan Kara6e4b0d52007-04-27 11:08:01 -070097/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
98void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
99{
100 unsigned int flags = oi->vfs_inode.i_flags;
101
102 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
103 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
104 if (flags & S_SYNC)
105 oi->ip_attr |= OCFS2_SYNC_FL;
106 if (flags & S_APPEND)
107 oi->ip_attr |= OCFS2_APPEND_FL;
108 if (flags & S_IMMUTABLE)
109 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
110 if (flags & S_NOATIME)
111 oi->ip_attr |= OCFS2_NOATIME_FL;
112 if (flags & S_DIRSYNC)
113 oi->ip_attr |= OCFS2_DIRSYNC_FL;
114}
115
wengang wang6ca497a2009-03-06 21:29:10 +0800116struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
117{
118 struct ocfs2_find_inode_args args;
119
120 args.fi_blkno = blkno;
121 args.fi_flags = 0;
122 args.fi_ino = ino_from_blkno(sb, blkno);
123 args.fi_sysfile_type = 0;
124
125 return ilookup5(sb, blkno, ocfs2_find_actor, &args);
126}
Jan Kara5fa06132008-01-11 00:11:45 +0100127struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
128 int sysfile_type)
Mark Fashehccd979b2005-12-15 14:31:24 -0800129{
130 struct inode *inode = NULL;
131 struct super_block *sb = osb->sb;
132 struct ocfs2_find_inode_args args;
133
Mark Fashehb0697052006-03-03 10:24:33 -0800134 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800135
136 /* Ok. By now we've either got the offsets passed to us by the
137 * caller, or we just pulled them off the bh. Lets do some
138 * sanity checks to make sure they're OK. */
139 if (blkno == 0) {
140 inode = ERR_PTR(-EINVAL);
141 mlog_errno(PTR_ERR(inode));
142 goto bail;
143 }
144
145 args.fi_blkno = blkno;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700146 args.fi_flags = flags;
Mark Fashehccd979b2005-12-15 14:31:24 -0800147 args.fi_ino = ino_from_blkno(sb, blkno);
Jan Kara5fa06132008-01-11 00:11:45 +0100148 args.fi_sysfile_type = sysfile_type;
Mark Fashehccd979b2005-12-15 14:31:24 -0800149
150 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
151 ocfs2_init_locked_inode, &args);
152 /* inode was *not* in the inode cache. 2.6.x requires
153 * us to do our own read_inode call and unlock it
154 * afterwards. */
155 if (inode && inode->i_state & I_NEW) {
156 mlog(0, "Inode was not in inode cache, reading it.\n");
157 ocfs2_read_locked_inode(inode, &args);
158 unlock_new_inode(inode);
159 }
160 if (inode == NULL) {
161 inode = ERR_PTR(-ENOMEM);
162 mlog_errno(PTR_ERR(inode));
163 goto bail;
164 }
165 if (is_bad_inode(inode)) {
166 iput(inode);
167 inode = ERR_PTR(-ESTALE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800168 goto bail;
169 }
170
171bail:
172 if (!IS_ERR(inode)) {
Mark Fashehb0697052006-03-03 10:24:33 -0800173 mlog(0, "returning inode with number %llu\n",
174 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800175 mlog_exit_ptr(inode);
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800176 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800177
178 return inode;
179}
180
181
182/*
183 * here's how inodes get read from disk:
184 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
185 * found? : return the in-memory inode
186 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
187 */
188
189static int ocfs2_find_actor(struct inode *inode, void *opaque)
190{
191 struct ocfs2_find_inode_args *args = NULL;
192 struct ocfs2_inode_info *oi = OCFS2_I(inode);
193 int ret = 0;
194
195 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
196
197 args = opaque;
198
199 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
200
201 if (oi->ip_blkno != args->fi_blkno)
202 goto bail;
203
Mark Fashehccd979b2005-12-15 14:31:24 -0800204 ret = 1;
205bail:
206 mlog_exit(ret);
207 return ret;
208}
209
210/*
211 * initialize the new inode, but don't do anything that would cause
212 * us to sleep.
213 * return 0 on success, 1 on failure
214 */
215static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
216{
217 struct ocfs2_find_inode_args *args = opaque;
Jan Karacb257972009-06-04 15:26:50 +0200218 static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
219 ocfs2_file_ip_alloc_sem_key;
Mark Fashehccd979b2005-12-15 14:31:24 -0800220
221 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
222
223 inode->i_ino = args->fi_ino;
224 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
Jan Kara5fa06132008-01-11 00:11:45 +0100225 if (args->fi_sysfile_type != 0)
226 lockdep_set_class(&inode->i_mutex,
227 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
Jan Karacb257972009-06-04 15:26:50 +0200228 if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
229 args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
230 args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
231 args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
232 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
233 &ocfs2_quota_ip_alloc_sem_key);
234 else
235 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
236 &ocfs2_file_ip_alloc_sem_key);
Mark Fashehccd979b2005-12-15 14:31:24 -0800237
238 mlog_exit(0);
239 return 0;
240}
241
Joel Beckerb657c952008-11-13 14:49:11 -0800242void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
243 int create_ino)
Mark Fashehccd979b2005-12-15 14:31:24 -0800244{
245 struct super_block *sb;
246 struct ocfs2_super *osb;
Mark Fasheh53da4932008-07-21 14:29:16 -0700247 int use_plocks = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800248
Mark Fashehb0697052006-03-03 10:24:33 -0800249 mlog_entry("(0x%p, size:%llu)\n", inode,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700250 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800251
252 sb = inode->i_sb;
253 osb = OCFS2_SB(sb);
254
Mark Fasheh53da4932008-07-21 14:29:16 -0700255 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
256 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
257 use_plocks = 0;
258
Joel Beckerb657c952008-11-13 14:49:11 -0800259 /*
260 * These have all been checked by ocfs2_read_inode_block() or set
261 * by ocfs2_mknod_locked(), so a failure is a code bug.
262 */
263 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
264 cannot create a superblock
265 inode today. change if
266 that is needed. */
267 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
268 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
Mark Fashehccd979b2005-12-15 14:31:24 -0800269
Mark Fashehccd979b2005-12-15 14:31:24 -0800270
Mark Fasheh8110b072007-03-22 16:53:23 -0700271 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
272 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -0700273 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
Mark Fasheh8110b072007-03-22 16:53:23 -0700274
Mark Fashehccd979b2005-12-15 14:31:24 -0800275 inode->i_version = 1;
276 inode->i_generation = le32_to_cpu(fe->i_generation);
277 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
278 inode->i_mode = le16_to_cpu(fe->i_mode);
279 inode->i_uid = le32_to_cpu(fe->i_uid);
280 inode->i_gid = le32_to_cpu(fe->i_gid);
Mark Fashehccd979b2005-12-15 14:31:24 -0800281
282 /* Fast symlinks will have i_size but no allocated clusters. */
283 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
284 inode->i_blocks = 0;
285 else
Mark Fasheh8110b072007-03-22 16:53:23 -0700286 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800287 inode->i_mapping->a_ops = &ocfs2_aops;
Mark Fashehccd979b2005-12-15 14:31:24 -0800288 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
289 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
290 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
291 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
292 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
293 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
294
295 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
296 mlog(ML_ERROR,
Mark Fashehb0697052006-03-03 10:24:33 -0800297 "ip_blkno %llu != i_blkno %llu!\n",
298 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700299 (unsigned long long)le64_to_cpu(fe->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800300
Mark Fasheh198a1ca2008-11-20 17:54:57 -0800301 inode->i_nlink = ocfs2_read_links_count(fe);
Mark Fashehccd979b2005-12-15 14:31:24 -0800302
Jan Karabbbd0eb2008-08-21 18:22:30 +0200303 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700304 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
Jan Karabbbd0eb2008-08-21 18:22:30 +0200305 inode->i_flags |= S_NOQUOTA;
306 }
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700307
Mark Fashehccd979b2005-12-15 14:31:24 -0800308 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
309 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
310 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
311 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
312 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
Jan Kara1a224ad2008-08-20 15:43:36 +0200313 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
314 inode->i_flags |= S_NOQUOTA;
Mark Fashehccd979b2005-12-15 14:31:24 -0800315 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
316 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
317 /* we can't actually hit this as read_inode can't
318 * handle superblocks today ;-) */
319 BUG();
320 }
321
322 switch (inode->i_mode & S_IFMT) {
323 case S_IFREG:
Mark Fasheh53da4932008-07-21 14:29:16 -0700324 if (use_plocks)
325 inode->i_fop = &ocfs2_fops;
326 else
327 inode->i_fop = &ocfs2_fops_no_plocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800328 inode->i_op = &ocfs2_file_iops;
329 i_size_write(inode, le64_to_cpu(fe->i_size));
330 break;
331 case S_IFDIR:
332 inode->i_op = &ocfs2_dir_iops;
Mark Fasheh53da4932008-07-21 14:29:16 -0700333 if (use_plocks)
334 inode->i_fop = &ocfs2_dops;
335 else
336 inode->i_fop = &ocfs2_dops_no_plocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800337 i_size_write(inode, le64_to_cpu(fe->i_size));
338 break;
339 case S_IFLNK:
340 if (ocfs2_inode_is_fast_symlink(inode))
341 inode->i_op = &ocfs2_fast_symlink_inode_operations;
342 else
343 inode->i_op = &ocfs2_symlink_inode_operations;
344 i_size_write(inode, le64_to_cpu(fe->i_size));
345 break;
346 default:
347 inode->i_op = &ocfs2_special_file_iops;
348 init_special_inode(inode, inode->i_mode,
349 inode->i_rdev);
350 break;
351 }
352
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700353 if (create_ino) {
354 inode->i_ino = ino_from_blkno(inode->i_sb,
355 le64_to_cpu(fe->i_blkno));
356
357 /*
358 * If we ever want to create system files from kernel,
359 * the generation argument to
360 * ocfs2_inode_lock_res_init() will have to change.
361 */
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700362 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700363
Mark Fashehe63aecb62007-10-18 15:30:42 -0700364 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700365 OCFS2_LOCK_TYPE_META, 0, inode);
Tiger Yang50008632007-03-20 16:01:38 -0700366
367 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
368 OCFS2_LOCK_TYPE_OPEN, 0, inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700369 }
370
Mark Fashehccd979b2005-12-15 14:31:24 -0800371 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700372 OCFS2_LOCK_TYPE_RW, inode->i_generation,
373 inode);
374
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700375 ocfs2_set_inode_flags(inode);
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700376
Tao Ma13821152009-02-25 00:53:23 +0800377 OCFS2_I(inode)->ip_last_used_slot = 0;
378 OCFS2_I(inode)->ip_last_used_group = 0;
Joel Beckerb657c952008-11-13 14:49:11 -0800379 mlog_exit_void();
Mark Fashehccd979b2005-12-15 14:31:24 -0800380}
381
382static int ocfs2_read_locked_inode(struct inode *inode,
383 struct ocfs2_find_inode_args *args)
384{
385 struct super_block *sb;
386 struct ocfs2_super *osb;
387 struct ocfs2_dinode *fe;
388 struct buffer_head *bh = NULL;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700389 int status, can_lock;
390 u32 generation = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800391
392 mlog_entry("(0x%p, 0x%p)\n", inode, args);
393
394 status = -EINVAL;
395 if (inode == NULL || inode->i_sb == NULL) {
396 mlog(ML_ERROR, "bad inode\n");
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700397 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800398 }
399 sb = inode->i_sb;
400 osb = OCFS2_SB(sb);
401
402 if (!args) {
403 mlog(ML_ERROR, "bad inode args\n");
404 make_bad_inode(inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700405 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800406 }
407
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700408 /*
409 * To improve performance of cold-cache inode stats, we take
410 * the cluster lock here if possible.
411 *
412 * Generally, OCFS2 never trusts the contents of an inode
413 * unless it's holding a cluster lock, so taking it here isn't
414 * a correctness issue as much as it is a performance
415 * improvement.
416 *
417 * There are three times when taking the lock is not a good idea:
418 *
419 * 1) During startup, before we have initialized the DLM.
420 *
421 * 2) If we are reading certain system files which never get
422 * cluster locks (local alloc, truncate log).
423 *
424 * 3) If the process doing the iget() is responsible for
425 * orphan dir recovery. We're holding the orphan dir lock and
426 * can get into a deadlock with another process on another
427 * node in ->delete_inode().
428 *
429 * #1 and #2 can be simply solved by never taking the lock
430 * here for system files (which are the only type we read
431 * during mount). It's a heavier approach, but our main
432 * concern is user-accesible files anyway.
433 *
434 * #3 works itself out because we'll eventually take the
435 * cluster lock before trusting anything anyway.
436 */
437 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
Tiger Yang50008632007-03-20 16:01:38 -0700438 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800439 && !ocfs2_mount_local(osb);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700440
441 /*
442 * To maintain backwards compatibility with older versions of
443 * ocfs2-tools, we still store the generation value for system
444 * files. The only ones that actually matter to userspace are
445 * the journals, but it's easier and inexpensive to just flag
446 * all system files similarly.
447 */
448 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
449 generation = osb->fs_generation;
450
Mark Fashehe63aecb62007-10-18 15:30:42 -0700451 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700452 OCFS2_LOCK_TYPE_META,
453 generation, inode);
454
Tiger Yang50008632007-03-20 16:01:38 -0700455 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
456 OCFS2_LOCK_TYPE_OPEN,
457 0, inode);
458
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700459 if (can_lock) {
Tiger Yang50008632007-03-20 16:01:38 -0700460 status = ocfs2_open_lock(inode);
461 if (status) {
462 make_bad_inode(inode);
463 mlog_errno(status);
464 return status;
465 }
Mark Fashehe63aecb62007-10-18 15:30:42 -0700466 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700467 if (status) {
468 make_bad_inode(inode);
469 mlog_errno(status);
470 return status;
471 }
472 }
473
Tiger Yang50008632007-03-20 16:01:38 -0700474 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
475 status = ocfs2_try_open_lock(inode, 0);
476 if (status) {
477 make_bad_inode(inode);
478 return status;
479 }
480 }
481
Joel Beckerb657c952008-11-13 14:49:11 -0800482 if (can_lock) {
483 status = ocfs2_read_inode_block_full(inode, &bh,
484 OCFS2_BH_IGNORE_CACHE);
485 } else {
Joel Beckerda1e9092008-10-09 17:20:29 -0700486 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
Joel Beckerb657c952008-11-13 14:49:11 -0800487 if (!status)
488 status = ocfs2_validate_inode_block(osb->sb, bh);
489 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800490 if (status < 0) {
491 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800492 goto bail;
493 }
494
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700495 status = -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800496 fe = (struct ocfs2_dinode *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800497
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700498 /*
499 * This is a code bug. Right now the caller needs to
500 * understand whether it is asking for a system file inode or
501 * not so the proper lock names can be built.
502 */
503 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
504 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
505 "Inode %llu: system file state is ambigous\n",
506 (unsigned long long)args->fi_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800507
508 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
509 S_ISBLK(le16_to_cpu(fe->i_mode)))
Joel Beckerb657c952008-11-13 14:49:11 -0800510 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
Mark Fashehccd979b2005-12-15 14:31:24 -0800511
Joel Beckerb657c952008-11-13 14:49:11 -0800512 ocfs2_populate_inode(inode, fe, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800513
514 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
515
Mark Fashehccd979b2005-12-15 14:31:24 -0800516 status = 0;
517
518bail:
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700519 if (can_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -0700520 ocfs2_inode_unlock(inode, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700521
522 if (status < 0)
523 make_bad_inode(inode);
524
Mark Fashehccd979b2005-12-15 14:31:24 -0800525 if (args && bh)
526 brelse(bh);
527
528 mlog_exit(status);
529 return status;
530}
531
532void ocfs2_sync_blockdev(struct super_block *sb)
533{
534 sync_blockdev(sb->s_bdev);
535}
536
537static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
538 struct inode *inode,
539 struct buffer_head *fe_bh)
540{
541 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800542 struct ocfs2_truncate_context *tc = NULL;
543 struct ocfs2_dinode *fe;
Mark Fasheh60b11392007-02-16 11:46:50 -0800544 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800545
546 mlog_entry_void();
547
548 fe = (struct ocfs2_dinode *) fe_bh->b_data;
549
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700550 /*
551 * This check will also skip truncate of inodes with inline
552 * data and fast symlinks.
553 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800554 if (fe->i_clusters) {
Joel Becker2b4e30f2008-09-03 20:03:41 -0700555 if (ocfs2_should_order_data(inode))
556 ocfs2_begin_ordered_truncate(inode, 0);
557
Mark Fasheh60b11392007-02-16 11:46:50 -0800558 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
559 if (IS_ERR(handle)) {
560 status = PTR_ERR(handle);
561 mlog_errno(status);
562 goto out;
563 }
564
Joel Becker13723d02008-10-17 19:25:01 -0700565 status = ocfs2_journal_access_di(handle, inode, fe_bh,
566 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh60b11392007-02-16 11:46:50 -0800567 if (status < 0) {
568 mlog_errno(status);
569 goto out;
570 }
571
572 i_size_write(inode, 0);
573
574 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
575 if (status < 0) {
576 mlog_errno(status);
577 goto out;
578 }
579
580 ocfs2_commit_trans(osb, handle);
581 handle = NULL;
582
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800583 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
584 if (status < 0) {
585 mlog_errno(status);
586 goto out;
587 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800588
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800589 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
590 if (status < 0) {
591 mlog_errno(status);
592 goto out;
593 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800594 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800595
Mark Fasheh60b11392007-02-16 11:46:50 -0800596out:
597 if (handle)
598 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800599 mlog_exit(status);
600 return status;
601}
602
603static int ocfs2_remove_inode(struct inode *inode,
604 struct buffer_head *di_bh,
605 struct inode *orphan_dir_inode,
606 struct buffer_head *orphan_dir_bh)
607{
608 int status;
609 struct inode *inode_alloc_inode = NULL;
610 struct buffer_head *inode_alloc_bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700611 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800612 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
613 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
614
615 inode_alloc_inode =
616 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
617 le16_to_cpu(di->i_suballoc_slot));
618 if (!inode_alloc_inode) {
619 status = -EEXIST;
620 mlog_errno(status);
621 goto bail;
622 }
623
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800624 mutex_lock(&inode_alloc_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700625 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800626 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800627 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800628
629 mlog_errno(status);
630 goto bail;
631 }
632
Jan Karaa90714c2008-10-09 19:38:40 +0200633 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800634 ocfs2_quota_trans_credits(inode->i_sb));
Mark Fashehccd979b2005-12-15 14:31:24 -0800635 if (IS_ERR(handle)) {
636 status = PTR_ERR(handle);
637 mlog_errno(status);
638 goto bail_unlock;
639 }
640
641 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
642 orphan_dir_bh);
643 if (status < 0) {
644 mlog_errno(status);
645 goto bail_commit;
646 }
647
648 /* set the inodes dtime */
Joel Becker13723d02008-10-17 19:25:01 -0700649 status = ocfs2_journal_access_di(handle, inode, di_bh,
650 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800651 if (status < 0) {
652 mlog_errno(status);
653 goto bail_commit;
654 }
655
656 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
Marcin Slusarz4092d492007-12-25 15:52:59 +0100657 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
Mark Fashehccd979b2005-12-15 14:31:24 -0800658
659 status = ocfs2_journal_dirty(handle, di_bh);
660 if (status < 0) {
661 mlog_errno(status);
662 goto bail_commit;
663 }
664
Joel Becker8cb471e2009-02-10 20:00:41 -0800665 ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
Jan Karaa90714c2008-10-09 19:38:40 +0200666 vfs_dq_free_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800667
668 status = ocfs2_free_dinode(handle, inode_alloc_inode,
669 inode_alloc_bh, di);
670 if (status < 0)
671 mlog_errno(status);
672
673bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700674 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800675bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700676 ocfs2_inode_unlock(inode_alloc_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800677 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800678 brelse(inode_alloc_bh);
679bail:
680 iput(inode_alloc_inode);
681
682 return status;
683}
684
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800685/*
686 * Serialize with orphan dir recovery. If the process doing
687 * recovery on this orphan dir does an iget() with the dir
688 * i_mutex held, we'll deadlock here. Instead we detect this
689 * and exit early - recovery will wipe this inode for us.
690 */
691static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
692 int slot)
693{
694 int ret = 0;
695
696 spin_lock(&osb->osb_lock);
697 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
698 mlog(0, "Recovery is happening on orphan dir %d, will skip "
699 "this inode\n", slot);
700 ret = -EDEADLK;
701 goto out;
702 }
703 /* This signals to the orphan recovery process that it should
704 * wait for us to handle the wipe. */
705 osb->osb_orphan_wipes[slot]++;
706out:
707 spin_unlock(&osb->osb_lock);
708 return ret;
709}
710
711static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
712 int slot)
713{
714 spin_lock(&osb->osb_lock);
715 osb->osb_orphan_wipes[slot]--;
716 spin_unlock(&osb->osb_lock);
717
718 wake_up(&osb->osb_wipe_event);
719}
720
Mark Fashehccd979b2005-12-15 14:31:24 -0800721static int ocfs2_wipe_inode(struct inode *inode,
722 struct buffer_head *di_bh)
723{
724 int status, orphaned_slot;
725 struct inode *orphan_dir_inode = NULL;
726 struct buffer_head *orphan_dir_bh = NULL;
727 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tiger Yang50008632007-03-20 16:01:38 -0700728 struct ocfs2_dinode *di;
Mark Fashehccd979b2005-12-15 14:31:24 -0800729
Tiger Yang50008632007-03-20 16:01:38 -0700730 di = (struct ocfs2_dinode *) di_bh->b_data;
731 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800732
733 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
734 if (status)
735 return status;
736
Mark Fashehccd979b2005-12-15 14:31:24 -0800737 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
738 ORPHAN_DIR_SYSTEM_INODE,
739 orphaned_slot);
740 if (!orphan_dir_inode) {
741 status = -EEXIST;
742 mlog_errno(status);
743 goto bail;
744 }
745
746 /* Lock the orphan dir. The lock will be held for the entire
747 * delete_inode operation. We do this now to avoid races with
748 * recovery completion on other nodes. */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800749 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700750 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800751 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800752 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800753
754 mlog_errno(status);
755 goto bail;
756 }
757
758 /* we do this while holding the orphan dir lock because we
Mark Fasheh34d024f2007-09-24 15:56:19 -0700759 * don't want recovery being run from another node to try an
760 * inode delete underneath us -- this will result in two nodes
Mark Fashehccd979b2005-12-15 14:31:24 -0800761 * truncating the same file! */
762 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
763 if (status < 0) {
764 mlog_errno(status);
765 goto bail_unlock_dir;
766 }
767
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800768 /* Remove any dir index tree */
769 if (S_ISDIR(inode->i_mode)) {
770 status = ocfs2_dx_dir_truncate(inode, di_bh);
771 if (status) {
772 mlog_errno(status);
773 goto bail_unlock_dir;
774 }
775 }
776
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800777 /*Free extended attribute resources associated with this inode.*/
778 status = ocfs2_xattr_remove(inode, di_bh);
779 if (status < 0) {
780 mlog_errno(status);
781 goto bail_unlock_dir;
782 }
783
Mark Fashehccd979b2005-12-15 14:31:24 -0800784 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
785 orphan_dir_bh);
786 if (status < 0)
787 mlog_errno(status);
788
789bail_unlock_dir:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700790 ocfs2_inode_unlock(orphan_dir_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800791 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800792 brelse(orphan_dir_bh);
793bail:
794 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800795 ocfs2_signal_wipe_completion(osb, orphaned_slot);
Mark Fashehccd979b2005-12-15 14:31:24 -0800796
797 return status;
798}
799
800/* There is a series of simple checks that should be done before a
Mark Fasheh34d024f2007-09-24 15:56:19 -0700801 * trylock is even considered. Encapsulate those in this function. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800802static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
803{
804 int ret = 0;
805 struct ocfs2_inode_info *oi = OCFS2_I(inode);
806 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
807
808 /* We shouldn't be getting here for the root directory
809 * inode.. */
810 if (inode == osb->root_inode) {
811 mlog(ML_ERROR, "Skipping delete of root inode.\n");
812 goto bail;
813 }
814
Mark Fasheh34d024f2007-09-24 15:56:19 -0700815 /* If we're coming from downconvert_thread we can't go into our own
Mark Fashehccd979b2005-12-15 14:31:24 -0800816 * voting [hello, deadlock city!], so unforuntately we just
817 * have to skip deleting this guy. That's OK though because
818 * the node who's doing the actual deleting should handle it
819 * anyway. */
Mark Fasheh34d024f2007-09-24 15:56:19 -0700820 if (current == osb->dc_task) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800821 mlog(0, "Skipping delete of %lu because we're currently "
Mark Fasheh34d024f2007-09-24 15:56:19 -0700822 "in downconvert\n", inode->i_ino);
Mark Fashehccd979b2005-12-15 14:31:24 -0800823 goto bail;
824 }
825
826 spin_lock(&oi->ip_lock);
827 /* OCFS2 *never* deletes system files. This should technically
828 * never get here as system file inodes should always have a
829 * positive link count. */
830 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
Mark Fashehb0697052006-03-03 10:24:33 -0800831 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
832 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800833 goto bail_unlock;
834 }
835
Mark Fasheh34d024f2007-09-24 15:56:19 -0700836 /* If we have allowd wipe of this inode for another node, it
837 * will be marked here so we can safely skip it. Recovery will
838 * cleanup any inodes we might inadvertantly skip here. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800839 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
840 mlog(0, "Skipping delete of %lu because another node "
841 "has done this for us.\n", inode->i_ino);
842 goto bail_unlock;
843 }
844
845 ret = 1;
846bail_unlock:
847 spin_unlock(&oi->ip_lock);
848bail:
849 return ret;
850}
851
852/* Query the cluster to determine whether we should wipe an inode from
853 * disk or not.
854 *
855 * Requires the inode to have the cluster lock. */
856static int ocfs2_query_inode_wipe(struct inode *inode,
857 struct buffer_head *di_bh,
858 int *wipe)
859{
860 int status = 0;
861 struct ocfs2_inode_info *oi = OCFS2_I(inode);
862 struct ocfs2_dinode *di;
863
864 *wipe = 0;
865
866 /* While we were waiting for the cluster lock in
867 * ocfs2_delete_inode, another node might have asked to delete
868 * the inode. Recheck our flags to catch this. */
869 if (!ocfs2_inode_is_valid_to_delete(inode)) {
Mark Fashehb0697052006-03-03 10:24:33 -0800870 mlog(0, "Skipping delete of %llu because flags changed\n",
871 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800872 goto bail;
873 }
874
875 /* Now that we have an up to date inode, we can double check
876 * the link count. */
877 if (inode->i_nlink) {
Mark Fashehb0697052006-03-03 10:24:33 -0800878 mlog(0, "Skipping delete of %llu because nlink = %u\n",
879 (unsigned long long)oi->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -0800880 goto bail;
881 }
882
883 /* Do some basic inode verification... */
884 di = (struct ocfs2_dinode *) di_bh->b_data;
885 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
886 /* for lack of a better error? */
887 status = -EEXIST;
888 mlog(ML_ERROR,
Mark Fashehb0697052006-03-03 10:24:33 -0800889 "Inode %llu (on-disk %llu) not orphaned! "
Mark Fashehccd979b2005-12-15 14:31:24 -0800890 "Disk flags 0x%x, inode flags 0x%x\n",
Mark Fashehb0697052006-03-03 10:24:33 -0800891 (unsigned long long)oi->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700892 (unsigned long long)le64_to_cpu(di->i_blkno),
893 le32_to_cpu(di->i_flags), oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800894 goto bail;
895 }
896
897 /* has someone already deleted us?! baaad... */
898 if (di->i_dtime) {
899 status = -EEXIST;
900 mlog_errno(status);
901 goto bail;
902 }
903
Mark Fasheh6f16bf62007-03-20 17:17:54 -0700904 /*
905 * This is how ocfs2 determines whether an inode is still live
906 * within the cluster. Every node takes a shared read lock on
907 * the inode open lock in ocfs2_read_locked_inode(). When we
908 * get to ->delete_inode(), each node tries to convert it's
909 * lock to an exclusive. Trylocks are serialized by the inode
910 * meta data lock. If the upconvert suceeds, we know the inode
911 * is no longer live and can be deleted.
912 *
913 * Though we call this with the meta data lock held, the
914 * trylock keeps us from ABBA deadlock.
915 */
916 status = ocfs2_try_open_lock(inode, 1);
Tiger Yang50008632007-03-20 16:01:38 -0700917 if (status == -EAGAIN) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800918 status = 0;
Joe Perches27592362007-11-19 17:53:34 -0800919 mlog(0, "Skipping delete of %llu because it is in use on "
Mark Fashehb0697052006-03-03 10:24:33 -0800920 "other nodes\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800921 goto bail;
922 }
923 if (status < 0) {
924 mlog_errno(status);
925 goto bail;
926 }
927
Tiger Yang50008632007-03-20 16:01:38 -0700928 *wipe = 1;
929 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
930 (unsigned long long)oi->ip_blkno,
931 le16_to_cpu(di->i_orphaned_slot));
Mark Fashehccd979b2005-12-15 14:31:24 -0800932
933bail:
934 return status;
935}
936
937/* Support function for ocfs2_delete_inode. Will help us keep the
938 * inode data in a consistent state for clear_inode. Always truncates
939 * pages, optionally sync's them first. */
940static void ocfs2_cleanup_delete_inode(struct inode *inode,
941 int sync_data)
942{
Mark Fashehb0697052006-03-03 10:24:33 -0800943 mlog(0, "Cleanup inode %llu, sync = %d\n",
944 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
Mark Fashehccd979b2005-12-15 14:31:24 -0800945 if (sync_data)
946 write_inode_now(inode, 1);
947 truncate_inode_pages(&inode->i_data, 0);
948}
949
950void ocfs2_delete_inode(struct inode *inode)
951{
952 int wipe, status;
953 sigset_t blocked, oldset;
954 struct buffer_head *di_bh = NULL;
955
956 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
957
Jan Karaa90714c2008-10-09 19:38:40 +0200958 /* When we fail in read_inode() we mark inode as bad. The second test
959 * catches the case when inode allocation fails before allocating
960 * a block for inode. */
961 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 mlog(0, "Skipping delete of bad inode\n");
963 goto bail;
964 }
965
966 if (!ocfs2_inode_is_valid_to_delete(inode)) {
967 /* It's probably not necessary to truncate_inode_pages
968 * here but we do it for safety anyway (it will most
969 * likely be a no-op anyway) */
970 ocfs2_cleanup_delete_inode(inode, 0);
971 goto bail;
972 }
973
974 /* We want to block signals in delete_inode as the lock and
975 * messaging paths may return us -ERESTARTSYS. Which would
976 * cause us to exit early, resulting in inodes being orphaned
977 * forever. */
978 sigfillset(&blocked);
979 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
980 if (status < 0) {
981 mlog_errno(status);
982 ocfs2_cleanup_delete_inode(inode, 1);
983 goto bail;
984 }
985
wengang wang6ca497a2009-03-06 21:29:10 +0800986 /*
987 * Synchronize us against ocfs2_get_dentry. We take this in
988 * shared mode so that all nodes can still concurrently
989 * process deletes.
990 */
991 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
992 if (status < 0) {
993 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
994 ocfs2_cleanup_delete_inode(inode, 0);
995 goto bail_unblock;
996 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800997 /* Lock down the inode. This gives us an up to date view of
998 * it's metadata (for verification), and allows us to
Mark Fasheh34d024f2007-09-24 15:56:19 -0700999 * serialize delete_inode on multiple nodes.
Mark Fashehccd979b2005-12-15 14:31:24 -08001000 *
1001 * Even though we might be doing a truncate, we don't take the
1002 * allocation lock here as it won't be needed - nobody will
1003 * have the file open.
1004 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001005 status = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001006 if (status < 0) {
1007 if (status != -ENOENT)
1008 mlog_errno(status);
1009 ocfs2_cleanup_delete_inode(inode, 0);
wengang wang6ca497a2009-03-06 21:29:10 +08001010 goto bail_unlock_nfs_sync;
Mark Fashehccd979b2005-12-15 14:31:24 -08001011 }
1012
1013 /* Query the cluster. This will be the final decision made
1014 * before we go ahead and wipe the inode. */
1015 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1016 if (!wipe || status < 0) {
Mark Fasheh34d024f2007-09-24 15:56:19 -07001017 /* Error and remote inode busy both mean we won't be
Mark Fashehccd979b2005-12-15 14:31:24 -08001018 * removing the inode, so they take almost the same
1019 * path. */
1020 if (status < 0)
1021 mlog_errno(status);
1022
Mark Fasheh34d024f2007-09-24 15:56:19 -07001023 /* Someone in the cluster has disallowed a wipe of
1024 * this inode, or it was never completely
1025 * orphaned. Write out the pages and exit now. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001026 ocfs2_cleanup_delete_inode(inode, 1);
1027 goto bail_unlock_inode;
1028 }
1029
1030 ocfs2_cleanup_delete_inode(inode, 0);
1031
1032 status = ocfs2_wipe_inode(inode, di_bh);
1033 if (status < 0) {
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001034 if (status != -EDEADLK)
1035 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001036 goto bail_unlock_inode;
1037 }
1038
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001039 /*
1040 * Mark the inode as successfully deleted.
1041 *
1042 * This is important for ocfs2_clear_inode() as it will check
1043 * this flag and skip any checkpointing work
1044 *
1045 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1046 * the LVB for other nodes.
1047 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001048 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1049
1050bail_unlock_inode:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001051 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001052 brelse(di_bh);
wengang wang6ca497a2009-03-06 21:29:10 +08001053
1054bail_unlock_nfs_sync:
1055 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1056
Mark Fashehccd979b2005-12-15 14:31:24 -08001057bail_unblock:
1058 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1059 if (status < 0)
1060 mlog_errno(status);
1061bail:
1062 clear_inode(inode);
1063 mlog_exit_void();
1064}
1065
1066void ocfs2_clear_inode(struct inode *inode)
1067{
1068 int status;
1069 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1070
1071 mlog_entry_void();
1072
1073 if (!inode)
1074 goto bail;
1075
Mark Fashehb0697052006-03-03 10:24:33 -08001076 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1077 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001078
1079 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1080 "Inode=%lu\n", inode->i_ino);
1081
Mark Fasheh34d024f2007-09-24 15:56:19 -07001082 /* To preven remote deletes we hold open lock before, now it
1083 * is time to unlock PR and EX open locks. */
Tiger Yang50008632007-03-20 16:01:38 -07001084 ocfs2_open_unlock(inode);
1085
Mark Fashehccd979b2005-12-15 14:31:24 -08001086 /* Do these before all the other work so that we don't bounce
Mark Fasheh34d024f2007-09-24 15:56:19 -07001087 * the downconvert thread while waiting to destroy the locks. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001088 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001089 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001090 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001091
1092 /* We very well may get a clear_inode before all an inodes
1093 * metadata has hit disk. Of course, we can't drop any cluster
1094 * locks until the journal has finished with it. The only
1095 * exception here are successfully wiped inodes - their
1096 * metadata can now be considered to be part of the system
1097 * inodes from which it came. */
1098 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1099 ocfs2_checkpoint_inode(inode);
1100
1101 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
Mark Fashehb0697052006-03-03 10:24:33 -08001102 "Clear inode of %llu, inode has io markers\n",
1103 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001104
Mark Fasheh83418972007-04-23 18:53:12 -07001105 ocfs2_extent_map_trunc(inode, 0);
1106
Mark Fashehccd979b2005-12-15 14:31:24 -08001107 status = ocfs2_drop_inode_locks(inode);
1108 if (status < 0)
1109 mlog_errno(status);
1110
1111 ocfs2_lock_res_free(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001112 ocfs2_lock_res_free(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001113 ocfs2_lock_res_free(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001114
Joel Becker8cb471e2009-02-10 20:00:41 -08001115 ocfs2_metadata_cache_purge(INODE_CACHE(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001116
Joel Becker8cb471e2009-02-10 20:00:41 -08001117 mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
Mark Fashehb0697052006-03-03 10:24:33 -08001118 "Clear inode of %llu, inode has %u cache items\n",
Joel Becker8cb471e2009-02-10 20:00:41 -08001119 (unsigned long long)oi->ip_blkno,
1120 INODE_CACHE(inode)->ci_num_cached);
Mark Fashehccd979b2005-12-15 14:31:24 -08001121
Joel Becker8cb471e2009-02-10 20:00:41 -08001122 mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
Mark Fashehb0697052006-03-03 10:24:33 -08001123 "Clear inode of %llu, inode has a bad flag\n",
1124 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001125
1126 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
Mark Fashehb0697052006-03-03 10:24:33 -08001127 "Clear inode of %llu, inode is locked\n",
1128 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001129
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001130 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
Mark Fashehb0697052006-03-03 10:24:33 -08001131 "Clear inode of %llu, io_mutex is locked\n",
1132 (unsigned long long)oi->ip_blkno);
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001133 mutex_unlock(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001134
1135 /*
1136 * down_trylock() returns 0, down_write_trylock() returns 1
1137 * kernel 1, world 0
1138 */
1139 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
Mark Fashehb0697052006-03-03 10:24:33 -08001140 "Clear inode of %llu, alloc_sem is locked\n",
1141 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001142 up_write(&oi->ip_alloc_sem);
1143
1144 mlog_bug_on_msg(oi->ip_open_count,
Mark Fashehb0697052006-03-03 10:24:33 -08001145 "Clear inode of %llu has open count %d\n",
1146 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001147
1148 /* Clear all other flags. */
Joel Becker47460d62009-02-10 16:05:07 -08001149 oi->ip_flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001150 oi->ip_created_trans = 0;
1151 oi->ip_last_trans = 0;
1152 oi->ip_dir_start_lookup = 0;
1153 oi->ip_blkno = 0ULL;
Sunil Mushranae0dff62008-10-22 13:24:29 -07001154
1155 /*
1156 * ip_jinode is used to track txns against this inode. We ensure that
1157 * the journal is flushed before journal shutdown. Thus it is safe to
1158 * have inodes get cleaned up after journal shutdown.
1159 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07001160 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1161 &oi->ip_jinode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001162
1163bail:
1164 mlog_exit_void();
1165}
1166
1167/* Called under inode_lock, with no more references on the
1168 * struct inode, so it's safe here to check the flags field
1169 * and to manipulate i_nlink without any other locks. */
1170void ocfs2_drop_inode(struct inode *inode)
1171{
1172 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1173
1174 mlog_entry_void();
1175
Mark Fashehb0697052006-03-03 10:24:33 -08001176 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1177 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001178
Mark Fasheh379dfe92006-09-08 14:21:03 -07001179 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1180 generic_delete_inode(inode);
1181 else
1182 generic_drop_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001183
1184 mlog_exit_void();
1185}
1186
1187/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001188 * This is called from our getattr.
1189 */
1190int ocfs2_inode_revalidate(struct dentry *dentry)
1191{
1192 struct inode *inode = dentry->d_inode;
1193 int status = 0;
1194
Mark Fashehb0697052006-03-03 10:24:33 -08001195 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1196 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001197
1198 if (!inode) {
1199 mlog(0, "eep, no inode!\n");
1200 status = -ENOENT;
1201 goto bail;
1202 }
1203
1204 spin_lock(&OCFS2_I(inode)->ip_lock);
1205 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1206 spin_unlock(&OCFS2_I(inode)->ip_lock);
1207 mlog(0, "inode deleted!\n");
1208 status = -ENOENT;
1209 goto bail;
1210 }
1211 spin_unlock(&OCFS2_I(inode)->ip_lock);
1212
Mark Fashehe63aecb62007-10-18 15:30:42 -07001213 /* Let ocfs2_inode_lock do the work of updating our struct
Mark Fashehccd979b2005-12-15 14:31:24 -08001214 * inode for us. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001215 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001216 if (status < 0) {
1217 if (status != -ENOENT)
1218 mlog_errno(status);
1219 goto bail;
1220 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07001221 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001222bail:
1223 mlog_exit(status);
1224
1225 return status;
1226}
1227
1228/*
1229 * Updates a disk inode from a
1230 * struct inode.
1231 * Only takes ip_lock.
1232 */
Mark Fasheh1fabe142006-10-09 18:11:45 -07001233int ocfs2_mark_inode_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001234 struct inode *inode,
1235 struct buffer_head *bh)
1236{
1237 int status;
1238 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1239
Mark Fashehb0697052006-03-03 10:24:33 -08001240 mlog_entry("(inode %llu)\n",
1241 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001242
Joel Becker13723d02008-10-17 19:25:01 -07001243 status = ocfs2_journal_access_di(handle, inode, bh,
1244 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001245 if (status < 0) {
1246 mlog_errno(status);
1247 goto leave;
1248 }
1249
1250 spin_lock(&OCFS2_I(inode)->ip_lock);
1251 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
Jan Kara6e4b0d52007-04-27 11:08:01 -07001252 ocfs2_get_inode_flags(OCFS2_I(inode));
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001253 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001254 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
Mark Fashehccd979b2005-12-15 14:31:24 -08001255 spin_unlock(&OCFS2_I(inode)->ip_lock);
1256
1257 fe->i_size = cpu_to_le64(i_size_read(inode));
Mark Fasheh198a1ca2008-11-20 17:54:57 -08001258 ocfs2_set_links_count(fe, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001259 fe->i_uid = cpu_to_le32(inode->i_uid);
1260 fe->i_gid = cpu_to_le32(inode->i_gid);
1261 fe->i_mode = cpu_to_le16(inode->i_mode);
1262 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1263 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1264 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1265 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1266 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1267 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1268
1269 status = ocfs2_journal_dirty(handle, bh);
1270 if (status < 0)
1271 mlog_errno(status);
1272
1273 status = 0;
1274leave:
1275
1276 mlog_exit(status);
1277 return status;
1278}
1279
1280/*
1281 *
1282 * Updates a struct inode from a disk inode.
1283 * does no i/o, only takes ip_lock.
1284 */
1285void ocfs2_refresh_inode(struct inode *inode,
1286 struct ocfs2_dinode *fe)
1287{
Mark Fashehccd979b2005-12-15 14:31:24 -08001288 spin_lock(&OCFS2_I(inode)->ip_lock);
1289
1290 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001291 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001292 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001293 ocfs2_set_inode_flags(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001294 i_size_write(inode, le64_to_cpu(fe->i_size));
Mark Fasheh198a1ca2008-11-20 17:54:57 -08001295 inode->i_nlink = ocfs2_read_links_count(fe);
Mark Fashehccd979b2005-12-15 14:31:24 -08001296 inode->i_uid = le32_to_cpu(fe->i_uid);
1297 inode->i_gid = le32_to_cpu(fe->i_gid);
1298 inode->i_mode = le16_to_cpu(fe->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001299 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1300 inode->i_blocks = 0;
1301 else
Mark Fasheh8110b072007-03-22 16:53:23 -07001302 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001303 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1304 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1305 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1306 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1307 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1308 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1309
1310 spin_unlock(&OCFS2_I(inode)->ip_lock);
1311}
Joel Beckerb657c952008-11-13 14:49:11 -08001312
1313int ocfs2_validate_inode_block(struct super_block *sb,
1314 struct buffer_head *bh)
1315{
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001316 int rc;
Joel Beckerb657c952008-11-13 14:49:11 -08001317 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1318
Joel Becker970e4932008-11-13 14:49:19 -08001319 mlog(0, "Validating dinode %llu\n",
1320 (unsigned long long)bh->b_blocknr);
1321
Joel Beckerb657c952008-11-13 14:49:11 -08001322 BUG_ON(!buffer_uptodate(bh));
1323
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001324 /*
1325 * If the ecc fails, we return the error but otherwise
1326 * leave the filesystem running. We know any error is
1327 * local to this block.
1328 */
1329 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
Joel Becker13723d02008-10-17 19:25:01 -07001330 if (rc) {
1331 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1332 (unsigned long long)bh->b_blocknr);
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001333 goto bail;
Joel Becker13723d02008-10-17 19:25:01 -07001334 }
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001335
1336 /*
1337 * Errors after here are fatal.
1338 */
1339
1340 rc = -EINVAL;
1341
Joel Beckerb657c952008-11-13 14:49:11 -08001342 if (!OCFS2_IS_VALID_DINODE(di)) {
1343 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1344 (unsigned long long)bh->b_blocknr, 7,
1345 di->i_signature);
1346 goto bail;
1347 }
1348
1349 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1350 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1351 (unsigned long long)bh->b_blocknr,
1352 (unsigned long long)le64_to_cpu(di->i_blkno));
1353 goto bail;
1354 }
1355
1356 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1357 ocfs2_error(sb,
1358 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1359 (unsigned long long)bh->b_blocknr);
1360 goto bail;
1361 }
1362
1363 if (le32_to_cpu(di->i_fs_generation) !=
1364 OCFS2_SB(sb)->fs_generation) {
1365 ocfs2_error(sb,
1366 "Invalid dinode #%llu: fs_generation is %u\n",
1367 (unsigned long long)bh->b_blocknr,
1368 le32_to_cpu(di->i_fs_generation));
1369 goto bail;
1370 }
1371
1372 rc = 0;
1373
1374bail:
1375 return rc;
1376}
1377
1378int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1379 int flags)
1380{
1381 int rc;
1382 struct buffer_head *tmp = *bh;
1383
Joel Becker8cb471e2009-02-10 20:00:41 -08001384 rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1385 1, &tmp, flags, ocfs2_validate_inode_block);
Joel Beckerb657c952008-11-13 14:49:11 -08001386
1387 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
Joel Becker970e4932008-11-13 14:49:19 -08001388 if (!rc && !*bh)
Joel Beckerb657c952008-11-13 14:49:11 -08001389 *bh = tmp;
1390
Joel Beckerb657c952008-11-13 14:49:11 -08001391 return rc;
1392}
1393
1394int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1395{
1396 return ocfs2_read_inode_block_full(inode, bh, 0);
1397}
Joel Becker6e5a3d72009-02-10 19:00:37 -08001398
1399static struct ocfs2_inode_info *cache_info_to_inode(struct ocfs2_caching_info *ci)
1400{
1401 return container_of(ci, struct ocfs2_inode_info, ip_metadata_cache);
1402}
1403
1404static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1405{
1406 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1407
1408 return oi->ip_blkno;
1409}
1410
Joel Becker8cb471e2009-02-10 20:00:41 -08001411static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1412{
1413 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1414
1415 return oi->vfs_inode.i_sb;
1416}
1417
Joel Becker6e5a3d72009-02-10 19:00:37 -08001418static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1419{
1420 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1421
1422 spin_lock(&oi->ip_lock);
1423}
1424
1425static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1426{
1427 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1428
1429 spin_unlock(&oi->ip_lock);
1430}
1431
1432static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1433{
1434 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1435
1436 mutex_lock(&oi->ip_io_mutex);
1437}
1438
1439static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1440{
1441 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1442
1443 mutex_unlock(&oi->ip_io_mutex);
1444}
1445
1446const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1447 .co_owner = ocfs2_inode_cache_owner,
Joel Becker8cb471e2009-02-10 20:00:41 -08001448 .co_get_super = ocfs2_inode_cache_get_super,
Joel Becker6e5a3d72009-02-10 19:00:37 -08001449 .co_cache_lock = ocfs2_inode_cache_lock,
1450 .co_cache_unlock = ocfs2_inode_cache_unlock,
1451 .co_io_lock = ocfs2_inode_cache_io_lock,
1452 .co_io_unlock = ocfs2_inode_cache_io_unlock,
1453};
1454