blob: abb0a95cc71702931c3f31645d09ebd5be4ed545 [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>
Mark Fashehccd979b2005-12-15 14:31:24 -080028#include <linux/highmem.h>
29#include <linux/pagemap.h>
Jan Karaa90714c2008-10-09 19:38:40 +020030#include <linux/quotaops.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080031
32#include <asm/byteorder.h>
33
34#define MLOG_MASK_PREFIX ML_INODE
35#include <cluster/masklog.h>
36
37#include "ocfs2.h"
38
39#include "alloc.h"
Mark Fasheh9b7895e2008-11-12 16:27:44 -080040#include "dir.h"
Joel Beckerd6b32bb2008-10-17 14:55:01 -070041#include "blockcheck.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080042#include "dlmglue.h"
43#include "extent_map.h"
44#include "file.h"
Mark Fashehb4df6ed2006-02-22 17:35:08 -080045#include "heartbeat.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080046#include "inode.h"
47#include "journal.h"
48#include "namei.h"
49#include "suballoc.h"
50#include "super.h"
51#include "symlink.h"
52#include "sysfile.h"
53#include "uptodate.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080054#include "xattr.h"
Tao Ma8b2c0db2009-08-18 11:43:49 +080055#include "refcounttree.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 Fashehb06970532006-03-03 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 Fashehb06970532006-03-03 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 Fashehb06970532006-03-03 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 Fashehb06970532006-03-03 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;
Mark Fashehe3b4a972009-12-07 13:16:07 -0800379
380 if (S_ISDIR(inode->i_mode))
381 ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
382 OCFS2_RESV_FLAG_DIR);
Joel Beckerb657c952008-11-13 14:49:11 -0800383 mlog_exit_void();
Mark Fashehccd979b2005-12-15 14:31:24 -0800384}
385
386static int ocfs2_read_locked_inode(struct inode *inode,
387 struct ocfs2_find_inode_args *args)
388{
389 struct super_block *sb;
390 struct ocfs2_super *osb;
391 struct ocfs2_dinode *fe;
392 struct buffer_head *bh = NULL;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700393 int status, can_lock;
394 u32 generation = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800395
396 mlog_entry("(0x%p, 0x%p)\n", inode, args);
397
398 status = -EINVAL;
399 if (inode == NULL || inode->i_sb == NULL) {
400 mlog(ML_ERROR, "bad inode\n");
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700401 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800402 }
403 sb = inode->i_sb;
404 osb = OCFS2_SB(sb);
405
406 if (!args) {
407 mlog(ML_ERROR, "bad inode args\n");
408 make_bad_inode(inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700409 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800410 }
411
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700412 /*
413 * To improve performance of cold-cache inode stats, we take
414 * the cluster lock here if possible.
415 *
416 * Generally, OCFS2 never trusts the contents of an inode
417 * unless it's holding a cluster lock, so taking it here isn't
418 * a correctness issue as much as it is a performance
419 * improvement.
420 *
421 * There are three times when taking the lock is not a good idea:
422 *
423 * 1) During startup, before we have initialized the DLM.
424 *
425 * 2) If we are reading certain system files which never get
426 * cluster locks (local alloc, truncate log).
427 *
428 * 3) If the process doing the iget() is responsible for
429 * orphan dir recovery. We're holding the orphan dir lock and
430 * can get into a deadlock with another process on another
431 * node in ->delete_inode().
432 *
433 * #1 and #2 can be simply solved by never taking the lock
434 * here for system files (which are the only type we read
435 * during mount). It's a heavier approach, but our main
436 * concern is user-accesible files anyway.
437 *
438 * #3 works itself out because we'll eventually take the
439 * cluster lock before trusting anything anyway.
440 */
441 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
Tiger Yang50008632007-03-20 16:01:38 -0700442 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800443 && !ocfs2_mount_local(osb);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700444
445 /*
446 * To maintain backwards compatibility with older versions of
447 * ocfs2-tools, we still store the generation value for system
448 * files. The only ones that actually matter to userspace are
449 * the journals, but it's easier and inexpensive to just flag
450 * all system files similarly.
451 */
452 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
453 generation = osb->fs_generation;
454
Mark Fashehe63aecb62007-10-18 15:30:42 -0700455 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700456 OCFS2_LOCK_TYPE_META,
457 generation, inode);
458
Tiger Yang50008632007-03-20 16:01:38 -0700459 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
460 OCFS2_LOCK_TYPE_OPEN,
461 0, inode);
462
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700463 if (can_lock) {
Tiger Yang50008632007-03-20 16:01:38 -0700464 status = ocfs2_open_lock(inode);
465 if (status) {
466 make_bad_inode(inode);
467 mlog_errno(status);
468 return status;
469 }
Mark Fashehe63aecb62007-10-18 15:30:42 -0700470 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700471 if (status) {
472 make_bad_inode(inode);
473 mlog_errno(status);
474 return status;
475 }
476 }
477
Tiger Yang50008632007-03-20 16:01:38 -0700478 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
479 status = ocfs2_try_open_lock(inode, 0);
480 if (status) {
Sunil Mushran2bd63212010-01-25 16:57:38 -0800481 make_bad_inode(inode);
Tiger Yang50008632007-03-20 16:01:38 -0700482 return status;
483 }
484 }
485
Joel Beckerb657c952008-11-13 14:49:11 -0800486 if (can_lock) {
487 status = ocfs2_read_inode_block_full(inode, &bh,
488 OCFS2_BH_IGNORE_CACHE);
489 } else {
Joel Beckerda1e9092008-10-09 17:20:29 -0700490 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
Joel Beckerb657c952008-11-13 14:49:11 -0800491 if (!status)
492 status = ocfs2_validate_inode_block(osb->sb, bh);
493 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800494 if (status < 0) {
495 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800496 goto bail;
497 }
498
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700499 status = -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800500 fe = (struct ocfs2_dinode *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800501
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700502 /*
503 * This is a code bug. Right now the caller needs to
504 * understand whether it is asking for a system file inode or
505 * not so the proper lock names can be built.
506 */
507 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
508 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
509 "Inode %llu: system file state is ambigous\n",
510 (unsigned long long)args->fi_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800511
512 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
513 S_ISBLK(le16_to_cpu(fe->i_mode)))
Joel Beckerb657c952008-11-13 14:49:11 -0800514 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
Mark Fashehccd979b2005-12-15 14:31:24 -0800515
Joel Beckerb657c952008-11-13 14:49:11 -0800516 ocfs2_populate_inode(inode, fe, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800517
518 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
519
Mark Fashehccd979b2005-12-15 14:31:24 -0800520 status = 0;
521
522bail:
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700523 if (can_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -0700524 ocfs2_inode_unlock(inode, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700525
526 if (status < 0)
527 make_bad_inode(inode);
528
Mark Fashehccd979b2005-12-15 14:31:24 -0800529 if (args && bh)
530 brelse(bh);
531
532 mlog_exit(status);
533 return status;
534}
535
536void ocfs2_sync_blockdev(struct super_block *sb)
537{
538 sync_blockdev(sb->s_bdev);
539}
540
541static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
542 struct inode *inode,
543 struct buffer_head *fe_bh)
544{
545 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800546 struct ocfs2_dinode *fe;
Mark Fasheh60b11392007-02-16 11:46:50 -0800547 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800548
549 mlog_entry_void();
550
551 fe = (struct ocfs2_dinode *) fe_bh->b_data;
552
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700553 /*
554 * This check will also skip truncate of inodes with inline
555 * data and fast symlinks.
556 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800557 if (fe->i_clusters) {
Joel Becker2b4e30f2008-09-03 20:03:41 -0700558 if (ocfs2_should_order_data(inode))
559 ocfs2_begin_ordered_truncate(inode, 0);
560
Mark Fasheh60b11392007-02-16 11:46:50 -0800561 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
562 if (IS_ERR(handle)) {
563 status = PTR_ERR(handle);
Dan Carpenter0350cb02010-04-22 11:39:29 +0200564 handle = NULL;
Mark Fasheh60b11392007-02-16 11:46:50 -0800565 mlog_errno(status);
566 goto out;
567 }
568
Joel Becker0cf2f762009-02-12 16:41:25 -0800569 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
570 fe_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700571 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh60b11392007-02-16 11:46:50 -0800572 if (status < 0) {
573 mlog_errno(status);
574 goto out;
575 }
576
577 i_size_write(inode, 0);
578
579 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
580 if (status < 0) {
581 mlog_errno(status);
582 goto out;
583 }
584
585 ocfs2_commit_trans(osb, handle);
586 handle = NULL;
587
Tristan Ye78f94672010-05-11 17:54:42 +0800588 status = ocfs2_commit_truncate(osb, inode, fe_bh);
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800589 if (status < 0) {
590 mlog_errno(status);
591 goto out;
592 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800593 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800594
Mark Fasheh60b11392007-02-16 11:46:50 -0800595out:
596 if (handle)
597 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800598 mlog_exit(status);
599 return status;
600}
601
602static int ocfs2_remove_inode(struct inode *inode,
603 struct buffer_head *di_bh,
604 struct inode *orphan_dir_inode,
605 struct buffer_head *orphan_dir_bh)
606{
607 int status;
608 struct inode *inode_alloc_inode = NULL;
609 struct buffer_head *inode_alloc_bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700610 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800611 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
612 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
613
614 inode_alloc_inode =
615 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
616 le16_to_cpu(di->i_suballoc_slot));
617 if (!inode_alloc_inode) {
618 status = -EEXIST;
619 mlog_errno(status);
620 goto bail;
621 }
622
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800623 mutex_lock(&inode_alloc_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700624 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800625 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800626 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800627
628 mlog_errno(status);
629 goto bail;
630 }
631
Jan Karaa90714c2008-10-09 19:38:40 +0200632 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800633 ocfs2_quota_trans_credits(inode->i_sb));
Mark Fashehccd979b2005-12-15 14:31:24 -0800634 if (IS_ERR(handle)) {
635 status = PTR_ERR(handle);
636 mlog_errno(status);
637 goto bail_unlock;
638 }
639
Li Dongyangd4cd1872010-04-22 16:11:19 +0800640 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
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 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800647 }
648
649 /* set the inodes dtime */
Joel Becker0cf2f762009-02-12 16:41:25 -0800650 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700651 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800652 if (status < 0) {
653 mlog_errno(status);
654 goto bail_commit;
655 }
656
657 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
Marcin Slusarz4092d492007-12-25 15:52:59 +0100658 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
Joel Beckerec20cec2010-03-19 14:13:52 -0700659 ocfs2_journal_dirty(handle, di_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800660
Joel Becker8cb471e2009-02-10 20:00:41 -0800661 ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500662 dquot_free_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800663
664 status = ocfs2_free_dinode(handle, inode_alloc_inode,
665 inode_alloc_bh, di);
666 if (status < 0)
667 mlog_errno(status);
668
669bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700670 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800671bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700672 ocfs2_inode_unlock(inode_alloc_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800673 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800674 brelse(inode_alloc_bh);
675bail:
676 iput(inode_alloc_inode);
677
678 return status;
679}
680
Sunil Mushran2bd63212010-01-25 16:57:38 -0800681/*
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800682 * Serialize with orphan dir recovery. If the process doing
683 * recovery on this orphan dir does an iget() with the dir
684 * i_mutex held, we'll deadlock here. Instead we detect this
685 * and exit early - recovery will wipe this inode for us.
686 */
687static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
688 int slot)
689{
690 int ret = 0;
691
692 spin_lock(&osb->osb_lock);
693 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
694 mlog(0, "Recovery is happening on orphan dir %d, will skip "
695 "this inode\n", slot);
696 ret = -EDEADLK;
697 goto out;
698 }
699 /* This signals to the orphan recovery process that it should
700 * wait for us to handle the wipe. */
701 osb->osb_orphan_wipes[slot]++;
702out:
703 spin_unlock(&osb->osb_lock);
704 return ret;
705}
706
707static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
708 int slot)
709{
710 spin_lock(&osb->osb_lock);
711 osb->osb_orphan_wipes[slot]--;
712 spin_unlock(&osb->osb_lock);
713
714 wake_up(&osb->osb_wipe_event);
715}
716
Mark Fashehccd979b2005-12-15 14:31:24 -0800717static int ocfs2_wipe_inode(struct inode *inode,
718 struct buffer_head *di_bh)
719{
Joel Beckerd5776322010-05-03 19:15:49 -0700720 int status, orphaned_slot = -1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800721 struct inode *orphan_dir_inode = NULL;
722 struct buffer_head *orphan_dir_bh = NULL;
723 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Li Dongyangd4cd1872010-04-22 16:11:19 +0800724 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800725
Li Dongyangd4cd1872010-04-22 16:11:19 +0800726 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
727 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800728
Li Dongyangd4cd1872010-04-22 16:11:19 +0800729 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
730 if (status)
731 return status;
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800732
Li Dongyangd4cd1872010-04-22 16:11:19 +0800733 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
734 ORPHAN_DIR_SYSTEM_INODE,
735 orphaned_slot);
736 if (!orphan_dir_inode) {
737 status = -EEXIST;
738 mlog_errno(status);
739 goto bail;
740 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800741
Li Dongyangd4cd1872010-04-22 16:11:19 +0800742 /* Lock the orphan dir. The lock will be held for the entire
743 * delete_inode operation. We do this now to avoid races with
744 * recovery completion on other nodes. */
745 mutex_lock(&orphan_dir_inode->i_mutex);
746 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
747 if (status < 0) {
748 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800749
Li Dongyangd4cd1872010-04-22 16:11:19 +0800750 mlog_errno(status);
751 goto bail;
752 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800753 }
754
755 /* we do this while holding the orphan dir lock because we
Mark Fasheh34d024f2007-09-24 15:56:19 -0700756 * don't want recovery being run from another node to try an
757 * inode delete underneath us -- this will result in two nodes
Mark Fashehccd979b2005-12-15 14:31:24 -0800758 * truncating the same file! */
759 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
760 if (status < 0) {
761 mlog_errno(status);
762 goto bail_unlock_dir;
763 }
764
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800765 /* Remove any dir index tree */
766 if (S_ISDIR(inode->i_mode)) {
767 status = ocfs2_dx_dir_truncate(inode, di_bh);
768 if (status) {
769 mlog_errno(status);
770 goto bail_unlock_dir;
771 }
772 }
773
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800774 /*Free extended attribute resources associated with this inode.*/
775 status = ocfs2_xattr_remove(inode, di_bh);
776 if (status < 0) {
777 mlog_errno(status);
778 goto bail_unlock_dir;
779 }
780
Tao Ma8b2c0db2009-08-18 11:43:49 +0800781 status = ocfs2_remove_refcount_tree(inode, di_bh);
782 if (status < 0) {
783 mlog_errno(status);
784 goto bail_unlock_dir;
785 }
786
Mark Fashehccd979b2005-12-15 14:31:24 -0800787 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
788 orphan_dir_bh);
789 if (status < 0)
790 mlog_errno(status);
791
792bail_unlock_dir:
Li Dongyangd4cd1872010-04-22 16:11:19 +0800793 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
794 return status;
795
Mark Fashehe63aecb62007-10-18 15:30:42 -0700796 ocfs2_inode_unlock(orphan_dir_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800797 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800798 brelse(orphan_dir_bh);
799bail:
800 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800801 ocfs2_signal_wipe_completion(osb, orphaned_slot);
Mark Fashehccd979b2005-12-15 14:31:24 -0800802
803 return status;
804}
805
806/* There is a series of simple checks that should be done before a
Mark Fasheh34d024f2007-09-24 15:56:19 -0700807 * trylock is even considered. Encapsulate those in this function. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800808static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
809{
810 int ret = 0;
811 struct ocfs2_inode_info *oi = OCFS2_I(inode);
812 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
813
814 /* We shouldn't be getting here for the root directory
815 * inode.. */
816 if (inode == osb->root_inode) {
817 mlog(ML_ERROR, "Skipping delete of root inode.\n");
818 goto bail;
819 }
820
Mark Fasheh34d024f2007-09-24 15:56:19 -0700821 /* If we're coming from downconvert_thread we can't go into our own
Mark Fashehccd979b2005-12-15 14:31:24 -0800822 * voting [hello, deadlock city!], so unforuntately we just
823 * have to skip deleting this guy. That's OK though because
824 * the node who's doing the actual deleting should handle it
825 * anyway. */
Mark Fasheh34d024f2007-09-24 15:56:19 -0700826 if (current == osb->dc_task) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800827 mlog(0, "Skipping delete of %lu because we're currently "
Mark Fasheh34d024f2007-09-24 15:56:19 -0700828 "in downconvert\n", inode->i_ino);
Mark Fashehccd979b2005-12-15 14:31:24 -0800829 goto bail;
830 }
831
832 spin_lock(&oi->ip_lock);
833 /* OCFS2 *never* deletes system files. This should technically
834 * never get here as system file inodes should always have a
835 * positive link count. */
836 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
Mark Fashehb06970532006-03-03 10:24:33 -0800837 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
838 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800839 goto bail_unlock;
840 }
841
Mark Fasheh34d024f2007-09-24 15:56:19 -0700842 /* If we have allowd wipe of this inode for another node, it
843 * will be marked here so we can safely skip it. Recovery will
844 * cleanup any inodes we might inadvertantly skip here. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800845 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
846 mlog(0, "Skipping delete of %lu because another node "
847 "has done this for us.\n", inode->i_ino);
848 goto bail_unlock;
849 }
850
851 ret = 1;
852bail_unlock:
853 spin_unlock(&oi->ip_lock);
854bail:
855 return ret;
856}
857
858/* Query the cluster to determine whether we should wipe an inode from
859 * disk or not.
860 *
861 * Requires the inode to have the cluster lock. */
862static int ocfs2_query_inode_wipe(struct inode *inode,
863 struct buffer_head *di_bh,
864 int *wipe)
865{
866 int status = 0;
867 struct ocfs2_inode_info *oi = OCFS2_I(inode);
868 struct ocfs2_dinode *di;
869
870 *wipe = 0;
871
872 /* While we were waiting for the cluster lock in
873 * ocfs2_delete_inode, another node might have asked to delete
874 * the inode. Recheck our flags to catch this. */
875 if (!ocfs2_inode_is_valid_to_delete(inode)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800876 mlog(0, "Skipping delete of %llu because flags changed\n",
877 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800878 goto bail;
879 }
880
881 /* Now that we have an up to date inode, we can double check
882 * the link count. */
883 if (inode->i_nlink) {
Mark Fashehb06970532006-03-03 10:24:33 -0800884 mlog(0, "Skipping delete of %llu because nlink = %u\n",
885 (unsigned long long)oi->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -0800886 goto bail;
887 }
888
889 /* Do some basic inode verification... */
890 di = (struct ocfs2_dinode *) di_bh->b_data;
Li Dongyangd4cd1872010-04-22 16:11:19 +0800891 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
892 !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
Tristan Yeb54c2ca2010-03-19 09:21:10 +0800893 /*
894 * Inodes in the orphan dir must have ORPHANED_FL. The only
895 * inodes that come back out of the orphan dir are reflink
896 * targets. A reflink target may be moved out of the orphan
897 * dir between the time we scan the directory and the time we
898 * process it. This would lead to HAS_REFCOUNT_FL being set but
899 * ORPHANED_FL not.
900 */
901 if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
902 mlog(0, "Reflinked inode %llu is no longer orphaned. "
903 "it shouldn't be deleted\n",
904 (unsigned long long)oi->ip_blkno);
905 goto bail;
906 }
907
Mark Fashehccd979b2005-12-15 14:31:24 -0800908 /* for lack of a better error? */
909 status = -EEXIST;
910 mlog(ML_ERROR,
Mark Fashehb06970532006-03-03 10:24:33 -0800911 "Inode %llu (on-disk %llu) not orphaned! "
Mark Fashehccd979b2005-12-15 14:31:24 -0800912 "Disk flags 0x%x, inode flags 0x%x\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800913 (unsigned long long)oi->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700914 (unsigned long long)le64_to_cpu(di->i_blkno),
915 le32_to_cpu(di->i_flags), oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800916 goto bail;
917 }
918
919 /* has someone already deleted us?! baaad... */
920 if (di->i_dtime) {
921 status = -EEXIST;
922 mlog_errno(status);
923 goto bail;
924 }
925
Mark Fasheh6f16bf62007-03-20 17:17:54 -0700926 /*
927 * This is how ocfs2 determines whether an inode is still live
928 * within the cluster. Every node takes a shared read lock on
929 * the inode open lock in ocfs2_read_locked_inode(). When we
930 * get to ->delete_inode(), each node tries to convert it's
931 * lock to an exclusive. Trylocks are serialized by the inode
932 * meta data lock. If the upconvert suceeds, we know the inode
933 * is no longer live and can be deleted.
934 *
935 * Though we call this with the meta data lock held, the
936 * trylock keeps us from ABBA deadlock.
937 */
938 status = ocfs2_try_open_lock(inode, 1);
Tiger Yang50008632007-03-20 16:01:38 -0700939 if (status == -EAGAIN) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800940 status = 0;
Joe Perches27592362007-11-19 17:53:34 -0800941 mlog(0, "Skipping delete of %llu because it is in use on "
Mark Fashehb06970532006-03-03 10:24:33 -0800942 "other nodes\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800943 goto bail;
944 }
945 if (status < 0) {
946 mlog_errno(status);
947 goto bail;
948 }
949
Tiger Yang50008632007-03-20 16:01:38 -0700950 *wipe = 1;
951 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
952 (unsigned long long)oi->ip_blkno,
953 le16_to_cpu(di->i_orphaned_slot));
Mark Fashehccd979b2005-12-15 14:31:24 -0800954
955bail:
956 return status;
957}
958
959/* Support function for ocfs2_delete_inode. Will help us keep the
960 * inode data in a consistent state for clear_inode. Always truncates
961 * pages, optionally sync's them first. */
962static void ocfs2_cleanup_delete_inode(struct inode *inode,
963 int sync_data)
964{
Mark Fashehb06970532006-03-03 10:24:33 -0800965 mlog(0, "Cleanup inode %llu, sync = %d\n",
966 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
Mark Fashehccd979b2005-12-15 14:31:24 -0800967 if (sync_data)
968 write_inode_now(inode, 1);
969 truncate_inode_pages(&inode->i_data, 0);
970}
971
972void ocfs2_delete_inode(struct inode *inode)
973{
974 int wipe, status;
Joel Beckere4b963f2009-09-02 17:17:36 -0700975 sigset_t oldset;
Mark Fashehccd979b2005-12-15 14:31:24 -0800976 struct buffer_head *di_bh = NULL;
977
978 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
979
Jan Karaa90714c2008-10-09 19:38:40 +0200980 /* When we fail in read_inode() we mark inode as bad. The second test
981 * catches the case when inode allocation fails before allocating
982 * a block for inode. */
983 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800984 mlog(0, "Skipping delete of bad inode\n");
985 goto bail;
986 }
987
Christoph Hellwig871a2932010-03-03 09:05:07 -0500988 dquot_initialize(inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -0500989
Mark Fashehccd979b2005-12-15 14:31:24 -0800990 if (!ocfs2_inode_is_valid_to_delete(inode)) {
991 /* It's probably not necessary to truncate_inode_pages
992 * here but we do it for safety anyway (it will most
993 * likely be a no-op anyway) */
994 ocfs2_cleanup_delete_inode(inode, 0);
995 goto bail;
996 }
997
998 /* We want to block signals in delete_inode as the lock and
999 * messaging paths may return us -ERESTARTSYS. Which would
1000 * cause us to exit early, resulting in inodes being orphaned
1001 * forever. */
Joel Beckere4b963f2009-09-02 17:17:36 -07001002 ocfs2_block_signals(&oldset);
Mark Fashehccd979b2005-12-15 14:31:24 -08001003
wengang wang6ca497a2009-03-06 21:29:10 +08001004 /*
1005 * Synchronize us against ocfs2_get_dentry. We take this in
1006 * shared mode so that all nodes can still concurrently
1007 * process deletes.
1008 */
1009 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1010 if (status < 0) {
1011 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1012 ocfs2_cleanup_delete_inode(inode, 0);
1013 goto bail_unblock;
1014 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001015 /* Lock down the inode. This gives us an up to date view of
1016 * it's metadata (for verification), and allows us to
Mark Fasheh34d024f2007-09-24 15:56:19 -07001017 * serialize delete_inode on multiple nodes.
Mark Fashehccd979b2005-12-15 14:31:24 -08001018 *
1019 * Even though we might be doing a truncate, we don't take the
1020 * allocation lock here as it won't be needed - nobody will
1021 * have the file open.
1022 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001023 status = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001024 if (status < 0) {
1025 if (status != -ENOENT)
1026 mlog_errno(status);
1027 ocfs2_cleanup_delete_inode(inode, 0);
wengang wang6ca497a2009-03-06 21:29:10 +08001028 goto bail_unlock_nfs_sync;
Mark Fashehccd979b2005-12-15 14:31:24 -08001029 }
1030
1031 /* Query the cluster. This will be the final decision made
1032 * before we go ahead and wipe the inode. */
1033 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1034 if (!wipe || status < 0) {
Mark Fasheh34d024f2007-09-24 15:56:19 -07001035 /* Error and remote inode busy both mean we won't be
Mark Fashehccd979b2005-12-15 14:31:24 -08001036 * removing the inode, so they take almost the same
1037 * path. */
1038 if (status < 0)
1039 mlog_errno(status);
1040
Mark Fasheh34d024f2007-09-24 15:56:19 -07001041 /* Someone in the cluster has disallowed a wipe of
1042 * this inode, or it was never completely
1043 * orphaned. Write out the pages and exit now. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001044 ocfs2_cleanup_delete_inode(inode, 1);
1045 goto bail_unlock_inode;
1046 }
1047
1048 ocfs2_cleanup_delete_inode(inode, 0);
1049
1050 status = ocfs2_wipe_inode(inode, di_bh);
1051 if (status < 0) {
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001052 if (status != -EDEADLK)
1053 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001054 goto bail_unlock_inode;
1055 }
1056
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001057 /*
1058 * Mark the inode as successfully deleted.
1059 *
1060 * This is important for ocfs2_clear_inode() as it will check
1061 * this flag and skip any checkpointing work
1062 *
1063 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1064 * the LVB for other nodes.
1065 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001066 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1067
1068bail_unlock_inode:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001069 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001070 brelse(di_bh);
wengang wang6ca497a2009-03-06 21:29:10 +08001071
1072bail_unlock_nfs_sync:
1073 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1074
Mark Fashehccd979b2005-12-15 14:31:24 -08001075bail_unblock:
Joel Beckere4b963f2009-09-02 17:17:36 -07001076 ocfs2_unblock_signals(&oldset);
Mark Fashehccd979b2005-12-15 14:31:24 -08001077bail:
1078 clear_inode(inode);
1079 mlog_exit_void();
1080}
1081
1082void ocfs2_clear_inode(struct inode *inode)
1083{
1084 int status;
1085 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1086
1087 mlog_entry_void();
1088
1089 if (!inode)
1090 goto bail;
1091
Mark Fashehb06970532006-03-03 10:24:33 -08001092 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1093 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001094
1095 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1096 "Inode=%lu\n", inode->i_ino);
1097
Christoph Hellwig9f754752010-03-03 09:05:05 -05001098 dquot_drop(inode);
Christoph Hellwig257ba152010-03-03 09:05:04 -05001099
Mark Fasheh34d024f2007-09-24 15:56:19 -07001100 /* To preven remote deletes we hold open lock before, now it
1101 * is time to unlock PR and EX open locks. */
Tiger Yang50008632007-03-20 16:01:38 -07001102 ocfs2_open_unlock(inode);
1103
Mark Fashehccd979b2005-12-15 14:31:24 -08001104 /* Do these before all the other work so that we don't bounce
Mark Fasheh34d024f2007-09-24 15:56:19 -07001105 * the downconvert thread while waiting to destroy the locks. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001106 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001107 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001108 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001109
Mark Fasheh4fe370a2009-12-07 13:15:40 -08001110 ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
1111 &oi->ip_la_data_resv);
1112 ocfs2_resv_init_once(&oi->ip_la_data_resv);
1113
Mark Fashehccd979b2005-12-15 14:31:24 -08001114 /* We very well may get a clear_inode before all an inodes
1115 * metadata has hit disk. Of course, we can't drop any cluster
1116 * locks until the journal has finished with it. The only
1117 * exception here are successfully wiped inodes - their
1118 * metadata can now be considered to be part of the system
1119 * inodes from which it came. */
1120 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1121 ocfs2_checkpoint_inode(inode);
1122
1123 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
Mark Fashehb06970532006-03-03 10:24:33 -08001124 "Clear inode of %llu, inode has io markers\n",
1125 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001126
Mark Fasheh83418972007-04-23 18:53:12 -07001127 ocfs2_extent_map_trunc(inode, 0);
1128
Mark Fashehccd979b2005-12-15 14:31:24 -08001129 status = ocfs2_drop_inode_locks(inode);
1130 if (status < 0)
1131 mlog_errno(status);
1132
1133 ocfs2_lock_res_free(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001134 ocfs2_lock_res_free(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001135 ocfs2_lock_res_free(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001136
Joel Becker66fb3452009-02-12 15:24:40 -08001137 ocfs2_metadata_cache_exit(INODE_CACHE(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001138
Joel Becker8cb471e2009-02-10 20:00:41 -08001139 mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
Mark Fashehb06970532006-03-03 10:24:33 -08001140 "Clear inode of %llu, inode has %u cache items\n",
Joel Becker8cb471e2009-02-10 20:00:41 -08001141 (unsigned long long)oi->ip_blkno,
1142 INODE_CACHE(inode)->ci_num_cached);
Mark Fashehccd979b2005-12-15 14:31:24 -08001143
Joel Becker8cb471e2009-02-10 20:00:41 -08001144 mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
Mark Fashehb06970532006-03-03 10:24:33 -08001145 "Clear inode of %llu, inode has a bad flag\n",
1146 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001147
1148 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
Mark Fashehb06970532006-03-03 10:24:33 -08001149 "Clear inode of %llu, inode is locked\n",
1150 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001151
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001152 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
Mark Fashehb06970532006-03-03 10:24:33 -08001153 "Clear inode of %llu, io_mutex is locked\n",
1154 (unsigned long long)oi->ip_blkno);
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001155 mutex_unlock(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001156
1157 /*
1158 * down_trylock() returns 0, down_write_trylock() returns 1
1159 * kernel 1, world 0
1160 */
1161 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
Mark Fashehb06970532006-03-03 10:24:33 -08001162 "Clear inode of %llu, alloc_sem is locked\n",
1163 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001164 up_write(&oi->ip_alloc_sem);
1165
1166 mlog_bug_on_msg(oi->ip_open_count,
Mark Fashehb06970532006-03-03 10:24:33 -08001167 "Clear inode of %llu has open count %d\n",
1168 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001169
1170 /* Clear all other flags. */
Joel Becker47460d62009-02-10 16:05:07 -08001171 oi->ip_flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001172 oi->ip_dir_start_lookup = 0;
1173 oi->ip_blkno = 0ULL;
Sunil Mushranae0dff62008-10-22 13:24:29 -07001174
1175 /*
1176 * ip_jinode is used to track txns against this inode. We ensure that
1177 * the journal is flushed before journal shutdown. Thus it is safe to
1178 * have inodes get cleaned up after journal shutdown.
1179 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07001180 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1181 &oi->ip_jinode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001182
1183bail:
1184 mlog_exit_void();
1185}
1186
1187/* Called under inode_lock, with no more references on the
1188 * struct inode, so it's safe here to check the flags field
1189 * and to manipulate i_nlink without any other locks. */
1190void ocfs2_drop_inode(struct inode *inode)
1191{
1192 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1193
1194 mlog_entry_void();
1195
Mark Fashehb06970532006-03-03 10:24:33 -08001196 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1197 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001198
Mark Fasheh379dfe92006-09-08 14:21:03 -07001199 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1200 generic_delete_inode(inode);
1201 else
1202 generic_drop_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001203
1204 mlog_exit_void();
1205}
1206
1207/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001208 * This is called from our getattr.
1209 */
1210int ocfs2_inode_revalidate(struct dentry *dentry)
1211{
1212 struct inode *inode = dentry->d_inode;
1213 int status = 0;
1214
Mark Fashehb06970532006-03-03 10:24:33 -08001215 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1216 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001217
1218 if (!inode) {
1219 mlog(0, "eep, no inode!\n");
1220 status = -ENOENT;
1221 goto bail;
1222 }
1223
1224 spin_lock(&OCFS2_I(inode)->ip_lock);
1225 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1226 spin_unlock(&OCFS2_I(inode)->ip_lock);
1227 mlog(0, "inode deleted!\n");
1228 status = -ENOENT;
1229 goto bail;
1230 }
1231 spin_unlock(&OCFS2_I(inode)->ip_lock);
1232
Mark Fashehe63aecb62007-10-18 15:30:42 -07001233 /* Let ocfs2_inode_lock do the work of updating our struct
Mark Fashehccd979b2005-12-15 14:31:24 -08001234 * inode for us. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001235 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001236 if (status < 0) {
1237 if (status != -ENOENT)
1238 mlog_errno(status);
1239 goto bail;
1240 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07001241 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001242bail:
1243 mlog_exit(status);
1244
1245 return status;
1246}
1247
1248/*
1249 * Updates a disk inode from a
1250 * struct inode.
1251 * Only takes ip_lock.
1252 */
Mark Fasheh1fabe142006-10-09 18:11:45 -07001253int ocfs2_mark_inode_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001254 struct inode *inode,
1255 struct buffer_head *bh)
1256{
1257 int status;
1258 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1259
Mark Fashehb06970532006-03-03 10:24:33 -08001260 mlog_entry("(inode %llu)\n",
1261 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001262
Joel Becker0cf2f762009-02-12 16:41:25 -08001263 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -07001264 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001265 if (status < 0) {
1266 mlog_errno(status);
1267 goto leave;
1268 }
1269
1270 spin_lock(&OCFS2_I(inode)->ip_lock);
1271 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
Jan Kara6e4b0d52007-04-27 11:08:01 -07001272 ocfs2_get_inode_flags(OCFS2_I(inode));
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001273 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001274 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
Mark Fashehccd979b2005-12-15 14:31:24 -08001275 spin_unlock(&OCFS2_I(inode)->ip_lock);
1276
1277 fe->i_size = cpu_to_le64(i_size_read(inode));
Mark Fasheh198a1ca2008-11-20 17:54:57 -08001278 ocfs2_set_links_count(fe, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001279 fe->i_uid = cpu_to_le32(inode->i_uid);
1280 fe->i_gid = cpu_to_le32(inode->i_gid);
1281 fe->i_mode = cpu_to_le16(inode->i_mode);
1282 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1283 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1284 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1285 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1286 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1287 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1288
Joel Beckerec20cec2010-03-19 14:13:52 -07001289 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001290leave:
Mark Fashehccd979b2005-12-15 14:31:24 -08001291 mlog_exit(status);
1292 return status;
1293}
1294
1295/*
1296 *
1297 * Updates a struct inode from a disk inode.
1298 * does no i/o, only takes ip_lock.
1299 */
1300void ocfs2_refresh_inode(struct inode *inode,
1301 struct ocfs2_dinode *fe)
1302{
Mark Fashehccd979b2005-12-15 14:31:24 -08001303 spin_lock(&OCFS2_I(inode)->ip_lock);
1304
1305 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001306 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001307 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001308 ocfs2_set_inode_flags(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001309 i_size_write(inode, le64_to_cpu(fe->i_size));
Mark Fasheh198a1ca2008-11-20 17:54:57 -08001310 inode->i_nlink = ocfs2_read_links_count(fe);
Mark Fashehccd979b2005-12-15 14:31:24 -08001311 inode->i_uid = le32_to_cpu(fe->i_uid);
1312 inode->i_gid = le32_to_cpu(fe->i_gid);
1313 inode->i_mode = le16_to_cpu(fe->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001314 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1315 inode->i_blocks = 0;
1316 else
Mark Fasheh8110b072007-03-22 16:53:23 -07001317 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001318 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1319 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1320 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1321 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1322 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1323 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1324
1325 spin_unlock(&OCFS2_I(inode)->ip_lock);
1326}
Joel Beckerb657c952008-11-13 14:49:11 -08001327
1328int ocfs2_validate_inode_block(struct super_block *sb,
1329 struct buffer_head *bh)
1330{
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001331 int rc;
Joel Beckerb657c952008-11-13 14:49:11 -08001332 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1333
Joel Becker970e4932008-11-13 14:49:19 -08001334 mlog(0, "Validating dinode %llu\n",
1335 (unsigned long long)bh->b_blocknr);
1336
Joel Beckerb657c952008-11-13 14:49:11 -08001337 BUG_ON(!buffer_uptodate(bh));
1338
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001339 /*
1340 * If the ecc fails, we return the error but otherwise
1341 * leave the filesystem running. We know any error is
1342 * local to this block.
1343 */
1344 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
Joel Becker13723d02008-10-17 19:25:01 -07001345 if (rc) {
1346 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1347 (unsigned long long)bh->b_blocknr);
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001348 goto bail;
Joel Becker13723d02008-10-17 19:25:01 -07001349 }
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001350
1351 /*
1352 * Errors after here are fatal.
1353 */
1354
1355 rc = -EINVAL;
1356
Joel Beckerb657c952008-11-13 14:49:11 -08001357 if (!OCFS2_IS_VALID_DINODE(di)) {
1358 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1359 (unsigned long long)bh->b_blocknr, 7,
1360 di->i_signature);
1361 goto bail;
1362 }
1363
1364 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1365 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1366 (unsigned long long)bh->b_blocknr,
1367 (unsigned long long)le64_to_cpu(di->i_blkno));
1368 goto bail;
1369 }
1370
1371 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1372 ocfs2_error(sb,
1373 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1374 (unsigned long long)bh->b_blocknr);
1375 goto bail;
1376 }
1377
1378 if (le32_to_cpu(di->i_fs_generation) !=
1379 OCFS2_SB(sb)->fs_generation) {
1380 ocfs2_error(sb,
1381 "Invalid dinode #%llu: fs_generation is %u\n",
1382 (unsigned long long)bh->b_blocknr,
1383 le32_to_cpu(di->i_fs_generation));
1384 goto bail;
1385 }
1386
1387 rc = 0;
1388
1389bail:
1390 return rc;
1391}
1392
1393int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1394 int flags)
1395{
1396 int rc;
1397 struct buffer_head *tmp = *bh;
1398
Joel Becker8cb471e2009-02-10 20:00:41 -08001399 rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1400 1, &tmp, flags, ocfs2_validate_inode_block);
Joel Beckerb657c952008-11-13 14:49:11 -08001401
1402 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
Joel Becker970e4932008-11-13 14:49:19 -08001403 if (!rc && !*bh)
Joel Beckerb657c952008-11-13 14:49:11 -08001404 *bh = tmp;
1405
Joel Beckerb657c952008-11-13 14:49:11 -08001406 return rc;
1407}
1408
1409int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1410{
1411 return ocfs2_read_inode_block_full(inode, bh, 0);
1412}
Joel Becker6e5a3d72009-02-10 19:00:37 -08001413
Joel Becker6e5a3d72009-02-10 19:00:37 -08001414
1415static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1416{
1417 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1418
1419 return oi->ip_blkno;
1420}
1421
Joel Becker8cb471e2009-02-10 20:00:41 -08001422static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1423{
1424 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1425
1426 return oi->vfs_inode.i_sb;
1427}
1428
Joel Becker6e5a3d72009-02-10 19:00:37 -08001429static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1430{
1431 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1432
1433 spin_lock(&oi->ip_lock);
1434}
1435
1436static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1437{
1438 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1439
1440 spin_unlock(&oi->ip_lock);
1441}
1442
1443static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1444{
1445 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1446
1447 mutex_lock(&oi->ip_io_mutex);
1448}
1449
1450static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1451{
1452 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1453
1454 mutex_unlock(&oi->ip_io_mutex);
1455}
1456
1457const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1458 .co_owner = ocfs2_inode_cache_owner,
Joel Becker8cb471e2009-02-10 20:00:41 -08001459 .co_get_super = ocfs2_inode_cache_get_super,
Joel Becker6e5a3d72009-02-10 19:00:37 -08001460 .co_cache_lock = ocfs2_inode_cache_lock,
1461 .co_cache_unlock = ocfs2_inode_cache_unlock,
1462 .co_io_lock = ocfs2_inode_cache_io_lock,
1463 .co_io_unlock = ocfs2_inode_cache_io_unlock,
1464};
1465