Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- 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> | |
ccd979bd MF |
28 | #include <linux/highmem.h> |
29 | #include <linux/pagemap.h> | |
a90714c1 | 30 | #include <linux/quotaops.h> |
cc56c33e | 31 | #include <linux/iversion.h> |
ccd979bd MF |
32 | |
33 | #include <asm/byteorder.h> | |
34 | ||
ccd979bd MF |
35 | #include <cluster/masklog.h> |
36 | ||
37 | #include "ocfs2.h" | |
38 | ||
39 | #include "alloc.h" | |
9b7895ef | 40 | #include "dir.h" |
d6b32bbb | 41 | #include "blockcheck.h" |
ccd979bd MF |
42 | #include "dlmglue.h" |
43 | #include "extent_map.h" | |
44 | #include "file.h" | |
b4df6ed8 | 45 | #include "heartbeat.h" |
ccd979bd MF |
46 | #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" | |
cf1d6c76 | 54 | #include "xattr.h" |
8b2c0dba | 55 | #include "refcounttree.h" |
64f3b269 | 56 | #include "ocfs2_trace.h" |
d56a8f32 | 57 | #include "filecheck.h" |
ccd979bd MF |
58 | |
59 | #include "buffer_head_io.h" | |
60 | ||
ccd979bd MF |
61 | struct ocfs2_find_inode_args |
62 | { | |
63 | u64 fi_blkno; | |
64 | unsigned long fi_ino; | |
65 | unsigned int fi_flags; | |
5fa0613e | 66 | unsigned int fi_sysfile_type; |
ccd979bd MF |
67 | }; |
68 | ||
5fa0613e JK |
69 | static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES]; |
70 | ||
ccd979bd MF |
71 | static int ocfs2_read_locked_inode(struct inode *inode, |
72 | struct ocfs2_find_inode_args *args); | |
73 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); | |
74 | static int ocfs2_find_actor(struct inode *inode, void *opaque); | |
75 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
76 | struct inode *inode, | |
77 | struct buffer_head *fe_bh); | |
78 | ||
d56a8f32 GH |
79 | static int ocfs2_filecheck_read_inode_block_full(struct inode *inode, |
80 | struct buffer_head **bh, | |
81 | int flags, int type); | |
82 | static int ocfs2_filecheck_validate_inode_block(struct super_block *sb, | |
83 | struct buffer_head *bh); | |
84 | static int ocfs2_filecheck_repair_inode_block(struct super_block *sb, | |
85 | struct buffer_head *bh); | |
86 | ||
ca4d147e HP |
87 | void ocfs2_set_inode_flags(struct inode *inode) |
88 | { | |
89 | unsigned int flags = OCFS2_I(inode)->ip_attr; | |
90 | ||
91 | inode->i_flags &= ~(S_IMMUTABLE | | |
92 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); | |
93 | ||
94 | if (flags & OCFS2_IMMUTABLE_FL) | |
95 | inode->i_flags |= S_IMMUTABLE; | |
96 | ||
97 | if (flags & OCFS2_SYNC_FL) | |
98 | inode->i_flags |= S_SYNC; | |
99 | if (flags & OCFS2_APPEND_FL) | |
100 | inode->i_flags |= S_APPEND; | |
101 | if (flags & OCFS2_NOATIME_FL) | |
102 | inode->i_flags |= S_NOATIME; | |
103 | if (flags & OCFS2_DIRSYNC_FL) | |
104 | inode->i_flags |= S_DIRSYNC; | |
105 | } | |
106 | ||
6e4b0d56 JK |
107 | /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */ |
108 | void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi) | |
109 | { | |
110 | unsigned int flags = oi->vfs_inode.i_flags; | |
111 | ||
112 | oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL| | |
113 | OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL); | |
114 | if (flags & S_SYNC) | |
115 | oi->ip_attr |= OCFS2_SYNC_FL; | |
116 | if (flags & S_APPEND) | |
117 | oi->ip_attr |= OCFS2_APPEND_FL; | |
118 | if (flags & S_IMMUTABLE) | |
119 | oi->ip_attr |= OCFS2_IMMUTABLE_FL; | |
120 | if (flags & S_NOATIME) | |
121 | oi->ip_attr |= OCFS2_NOATIME_FL; | |
122 | if (flags & S_DIRSYNC) | |
123 | oi->ip_attr |= OCFS2_DIRSYNC_FL; | |
124 | } | |
125 | ||
6ca497a8 | 126 | struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno) |
127 | { | |
128 | struct ocfs2_find_inode_args args; | |
129 | ||
130 | args.fi_blkno = blkno; | |
131 | args.fi_flags = 0; | |
132 | args.fi_ino = ino_from_blkno(sb, blkno); | |
133 | args.fi_sysfile_type = 0; | |
134 | ||
135 | return ilookup5(sb, blkno, ocfs2_find_actor, &args); | |
136 | } | |
5fa0613e JK |
137 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags, |
138 | int sysfile_type) | |
ccd979bd | 139 | { |
8c4d5a43 | 140 | int rc = -ESTALE; |
ccd979bd MF |
141 | struct inode *inode = NULL; |
142 | struct super_block *sb = osb->sb; | |
143 | struct ocfs2_find_inode_args args; | |
2931cdcb | 144 | journal_t *journal = OCFS2_SB(sb)->journal->j_journal; |
ccd979bd | 145 | |
64f3b269 TM |
146 | trace_ocfs2_iget_begin((unsigned long long)blkno, flags, |
147 | sysfile_type); | |
ccd979bd MF |
148 | |
149 | /* Ok. By now we've either got the offsets passed to us by the | |
150 | * caller, or we just pulled them off the bh. Lets do some | |
151 | * sanity checks to make sure they're OK. */ | |
152 | if (blkno == 0) { | |
153 | inode = ERR_PTR(-EINVAL); | |
154 | mlog_errno(PTR_ERR(inode)); | |
155 | goto bail; | |
156 | } | |
157 | ||
158 | args.fi_blkno = blkno; | |
24c19ef4 | 159 | args.fi_flags = flags; |
ccd979bd | 160 | args.fi_ino = ino_from_blkno(sb, blkno); |
5fa0613e | 161 | args.fi_sysfile_type = sysfile_type; |
ccd979bd MF |
162 | |
163 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, | |
164 | ocfs2_init_locked_inode, &args); | |
165 | /* inode was *not* in the inode cache. 2.6.x requires | |
166 | * us to do our own read_inode call and unlock it | |
167 | * afterwards. */ | |
ccd979bd MF |
168 | if (inode == NULL) { |
169 | inode = ERR_PTR(-ENOMEM); | |
170 | mlog_errno(PTR_ERR(inode)); | |
171 | goto bail; | |
172 | } | |
64f3b269 | 173 | trace_ocfs2_iget5_locked(inode->i_state); |
6218b90e | 174 | if (inode->i_state & I_NEW) { |
d56a8f32 | 175 | rc = ocfs2_read_locked_inode(inode, &args); |
6218b90e TM |
176 | unlock_new_inode(inode); |
177 | } | |
ccd979bd MF |
178 | if (is_bad_inode(inode)) { |
179 | iput(inode); | |
1f3a437f | 180 | inode = ERR_PTR(rc); |
ccd979bd MF |
181 | goto bail; |
182 | } | |
183 | ||
2931cdcb DW |
184 | /* |
185 | * Set transaction id's of transactions that have to be committed | |
186 | * to finish f[data]sync. We set them to currently running transaction | |
187 | * as we cannot be sure that the inode or some of its metadata isn't | |
188 | * part of the transaction - the inode could have been reclaimed and | |
189 | * now it is reread from disk. | |
190 | */ | |
191 | if (journal) { | |
192 | transaction_t *transaction; | |
193 | tid_t tid; | |
194 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
195 | ||
196 | read_lock(&journal->j_state_lock); | |
197 | if (journal->j_running_transaction) | |
198 | transaction = journal->j_running_transaction; | |
199 | else | |
200 | transaction = journal->j_committing_transaction; | |
201 | if (transaction) | |
202 | tid = transaction->t_tid; | |
203 | else | |
204 | tid = journal->j_commit_sequence; | |
205 | read_unlock(&journal->j_state_lock); | |
206 | oi->i_sync_tid = tid; | |
207 | oi->i_datasync_tid = tid; | |
208 | } | |
209 | ||
ccd979bd MF |
210 | bail: |
211 | if (!IS_ERR(inode)) { | |
64f3b269 TM |
212 | trace_ocfs2_iget_end(inode, |
213 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
6a1bd4a5 | 214 | } |
ccd979bd MF |
215 | |
216 | return inode; | |
217 | } | |
218 | ||
219 | ||
220 | /* | |
221 | * here's how inodes get read from disk: | |
222 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR | |
223 | * found? : return the in-memory inode | |
224 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE | |
225 | */ | |
226 | ||
227 | static int ocfs2_find_actor(struct inode *inode, void *opaque) | |
228 | { | |
229 | struct ocfs2_find_inode_args *args = NULL; | |
230 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
231 | int ret = 0; | |
232 | ||
ccd979bd MF |
233 | args = opaque; |
234 | ||
235 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); | |
236 | ||
64f3b269 TM |
237 | trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno); |
238 | ||
ccd979bd MF |
239 | if (oi->ip_blkno != args->fi_blkno) |
240 | goto bail; | |
241 | ||
ccd979bd MF |
242 | ret = 1; |
243 | bail: | |
ccd979bd MF |
244 | return ret; |
245 | } | |
246 | ||
247 | /* | |
248 | * initialize the new inode, but don't do anything that would cause | |
249 | * us to sleep. | |
250 | * return 0 on success, 1 on failure | |
251 | */ | |
252 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) | |
253 | { | |
254 | struct ocfs2_find_inode_args *args = opaque; | |
cb25797d JK |
255 | static struct lock_class_key ocfs2_quota_ip_alloc_sem_key, |
256 | ocfs2_file_ip_alloc_sem_key; | |
ccd979bd | 257 | |
ccd979bd MF |
258 | inode->i_ino = args->fi_ino; |
259 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; | |
5fa0613e | 260 | if (args->fi_sysfile_type != 0) |
9902af79 | 261 | lockdep_set_class(&inode->i_rwsem, |
5fa0613e | 262 | &ocfs2_sysfile_lock_key[args->fi_sysfile_type]); |
cb25797d JK |
263 | if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE || |
264 | args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE || | |
265 | args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE || | |
266 | args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE) | |
267 | lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem, | |
268 | &ocfs2_quota_ip_alloc_sem_key); | |
269 | else | |
270 | lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem, | |
271 | &ocfs2_file_ip_alloc_sem_key); | |
ccd979bd | 272 | |
ccd979bd MF |
273 | return 0; |
274 | } | |
275 | ||
b657c95c JB |
276 | void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, |
277 | int create_ino) | |
ccd979bd MF |
278 | { |
279 | struct super_block *sb; | |
280 | struct ocfs2_super *osb; | |
53da4939 | 281 | int use_plocks = 1; |
ccd979bd | 282 | |
ccd979bd MF |
283 | sb = inode->i_sb; |
284 | osb = OCFS2_SB(sb); | |
285 | ||
53da4939 MF |
286 | if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) || |
287 | ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks()) | |
288 | use_plocks = 0; | |
289 | ||
b657c95c JB |
290 | /* |
291 | * These have all been checked by ocfs2_read_inode_block() or set | |
292 | * by ocfs2_mknod_locked(), so a failure is a code bug. | |
293 | */ | |
294 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode | |
295 | cannot create a superblock | |
296 | inode today. change if | |
297 | that is needed. */ | |
298 | BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))); | |
299 | BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation); | |
ccd979bd | 300 | |
ccd979bd | 301 | |
8110b073 MF |
302 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
303 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); | |
15b1e36b | 304 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
8110b073 | 305 | |
cc56c33e | 306 | inode_set_iversion(inode, 1); |
ccd979bd MF |
307 | inode->i_generation = le32_to_cpu(fe->i_generation); |
308 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
309 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
2c034176 EB |
310 | i_uid_write(inode, le32_to_cpu(fe->i_uid)); |
311 | i_gid_write(inode, le32_to_cpu(fe->i_gid)); | |
ccd979bd MF |
312 | |
313 | /* Fast symlinks will have i_size but no allocated clusters. */ | |
ea022dfb | 314 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) { |
ccd979bd | 315 | inode->i_blocks = 0; |
ea022dfb AV |
316 | inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops; |
317 | } else { | |
8110b073 | 318 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ea022dfb AV |
319 | inode->i_mapping->a_ops = &ocfs2_aops; |
320 | } | |
ccd979bd MF |
321 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
322 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
323 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
324 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
325 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
326 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
327 | ||
328 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) | |
329 | mlog(ML_ERROR, | |
b0697053 MF |
330 | "ip_blkno %llu != i_blkno %llu!\n", |
331 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1ca1a111 | 332 | (unsigned long long)le64_to_cpu(fe->i_blkno)); |
ccd979bd | 333 | |
bfe86848 | 334 | set_nlink(inode, ocfs2_read_links_count(fe)); |
ccd979bd | 335 | |
64f3b269 TM |
336 | trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno, |
337 | le32_to_cpu(fe->i_flags)); | |
bbbd0eb3 | 338 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) { |
24c19ef4 | 339 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; |
bbbd0eb3 JK |
340 | inode->i_flags |= S_NOQUOTA; |
341 | } | |
64f3b269 | 342 | |
ccd979bd MF |
343 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { |
344 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
ccd979bd MF |
345 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { |
346 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
1a224ad1 JK |
347 | } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) { |
348 | inode->i_flags |= S_NOQUOTA; | |
ccd979bd | 349 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { |
ccd979bd MF |
350 | /* we can't actually hit this as read_inode can't |
351 | * handle superblocks today ;-) */ | |
352 | BUG(); | |
353 | } | |
354 | ||
355 | switch (inode->i_mode & S_IFMT) { | |
356 | case S_IFREG: | |
53da4939 MF |
357 | if (use_plocks) |
358 | inode->i_fop = &ocfs2_fops; | |
359 | else | |
360 | inode->i_fop = &ocfs2_fops_no_plocks; | |
ccd979bd MF |
361 | inode->i_op = &ocfs2_file_iops; |
362 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
363 | break; | |
364 | case S_IFDIR: | |
365 | inode->i_op = &ocfs2_dir_iops; | |
53da4939 MF |
366 | if (use_plocks) |
367 | inode->i_fop = &ocfs2_dops; | |
368 | else | |
369 | inode->i_fop = &ocfs2_dops_no_plocks; | |
ccd979bd | 370 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
5e98d492 | 371 | OCFS2_I(inode)->ip_dir_lock_gen = 1; |
ccd979bd MF |
372 | break; |
373 | case S_IFLNK: | |
ea022dfb | 374 | inode->i_op = &ocfs2_symlink_inode_operations; |
21fc61c7 | 375 | inode_nohighmem(inode); |
ccd979bd MF |
376 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
377 | break; | |
378 | default: | |
379 | inode->i_op = &ocfs2_special_file_iops; | |
380 | init_special_inode(inode, inode->i_mode, | |
381 | inode->i_rdev); | |
382 | break; | |
383 | } | |
384 | ||
24c19ef4 MF |
385 | if (create_ino) { |
386 | inode->i_ino = ino_from_blkno(inode->i_sb, | |
387 | le64_to_cpu(fe->i_blkno)); | |
388 | ||
389 | /* | |
390 | * If we ever want to create system files from kernel, | |
391 | * the generation argument to | |
392 | * ocfs2_inode_lock_res_init() will have to change. | |
393 | */ | |
1ca1a111 | 394 | BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL); |
24c19ef4 | 395 | |
e63aecb6 | 396 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres, |
24c19ef4 | 397 | OCFS2_LOCK_TYPE_META, 0, inode); |
50008630 TY |
398 | |
399 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, | |
400 | OCFS2_LOCK_TYPE_OPEN, 0, inode); | |
24c19ef4 MF |
401 | } |
402 | ||
ccd979bd | 403 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, |
24c19ef4 MF |
404 | OCFS2_LOCK_TYPE_RW, inode->i_generation, |
405 | inode); | |
406 | ||
ca4d147e | 407 | ocfs2_set_inode_flags(inode); |
ca4d147e | 408 | |
13821151 TM |
409 | OCFS2_I(inode)->ip_last_used_slot = 0; |
410 | OCFS2_I(inode)->ip_last_used_group = 0; | |
e3b4a97d MF |
411 | |
412 | if (S_ISDIR(inode->i_mode)) | |
413 | ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv, | |
414 | OCFS2_RESV_FLAG_DIR); | |
ccd979bd MF |
415 | } |
416 | ||
417 | static int ocfs2_read_locked_inode(struct inode *inode, | |
418 | struct ocfs2_find_inode_args *args) | |
419 | { | |
420 | struct super_block *sb; | |
421 | struct ocfs2_super *osb; | |
422 | struct ocfs2_dinode *fe; | |
423 | struct buffer_head *bh = NULL; | |
d56a8f32 | 424 | int status, can_lock, lock_level = 0; |
24c19ef4 | 425 | u32 generation = 0; |
ccd979bd | 426 | |
ccd979bd | 427 | status = -EINVAL; |
ccd979bd MF |
428 | sb = inode->i_sb; |
429 | osb = OCFS2_SB(sb); | |
430 | ||
24c19ef4 MF |
431 | /* |
432 | * To improve performance of cold-cache inode stats, we take | |
433 | * the cluster lock here if possible. | |
434 | * | |
435 | * Generally, OCFS2 never trusts the contents of an inode | |
436 | * unless it's holding a cluster lock, so taking it here isn't | |
437 | * a correctness issue as much as it is a performance | |
438 | * improvement. | |
439 | * | |
440 | * There are three times when taking the lock is not a good idea: | |
441 | * | |
442 | * 1) During startup, before we have initialized the DLM. | |
443 | * | |
444 | * 2) If we are reading certain system files which never get | |
445 | * cluster locks (local alloc, truncate log). | |
446 | * | |
447 | * 3) If the process doing the iget() is responsible for | |
448 | * orphan dir recovery. We're holding the orphan dir lock and | |
449 | * can get into a deadlock with another process on another | |
450 | * node in ->delete_inode(). | |
451 | * | |
452 | * #1 and #2 can be simply solved by never taking the lock | |
453 | * here for system files (which are the only type we read | |
454 | * during mount). It's a heavier approach, but our main | |
b595076a | 455 | * concern is user-accessible files anyway. |
24c19ef4 MF |
456 | * |
457 | * #3 works itself out because we'll eventually take the | |
458 | * cluster lock before trusting anything anyway. | |
459 | */ | |
460 | can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
50008630 | 461 | && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) |
c271c5c2 | 462 | && !ocfs2_mount_local(osb); |
24c19ef4 | 463 | |
64f3b269 TM |
464 | trace_ocfs2_read_locked_inode( |
465 | (unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock); | |
466 | ||
24c19ef4 MF |
467 | /* |
468 | * To maintain backwards compatibility with older versions of | |
469 | * ocfs2-tools, we still store the generation value for system | |
470 | * files. The only ones that actually matter to userspace are | |
471 | * the journals, but it's easier and inexpensive to just flag | |
472 | * all system files similarly. | |
473 | */ | |
474 | if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
475 | generation = osb->fs_generation; | |
476 | ||
e63aecb6 | 477 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres, |
24c19ef4 MF |
478 | OCFS2_LOCK_TYPE_META, |
479 | generation, inode); | |
480 | ||
50008630 TY |
481 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, |
482 | OCFS2_LOCK_TYPE_OPEN, | |
483 | 0, inode); | |
484 | ||
24c19ef4 | 485 | if (can_lock) { |
50008630 TY |
486 | status = ocfs2_open_lock(inode); |
487 | if (status) { | |
488 | make_bad_inode(inode); | |
489 | mlog_errno(status); | |
490 | return status; | |
491 | } | |
d56a8f32 | 492 | status = ocfs2_inode_lock(inode, NULL, lock_level); |
24c19ef4 MF |
493 | if (status) { |
494 | make_bad_inode(inode); | |
495 | mlog_errno(status); | |
496 | return status; | |
497 | } | |
ccd979bd MF |
498 | } |
499 | ||
50008630 TY |
500 | if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) { |
501 | status = ocfs2_try_open_lock(inode, 0); | |
502 | if (status) { | |
2bd63216 | 503 | make_bad_inode(inode); |
50008630 TY |
504 | return status; |
505 | } | |
506 | } | |
507 | ||
b657c95c | 508 | if (can_lock) { |
d56a8f32 GH |
509 | if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK) |
510 | status = ocfs2_filecheck_read_inode_block_full(inode, | |
511 | &bh, OCFS2_BH_IGNORE_CACHE, 0); | |
512 | else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX) | |
513 | status = ocfs2_filecheck_read_inode_block_full(inode, | |
514 | &bh, OCFS2_BH_IGNORE_CACHE, 1); | |
515 | else | |
516 | status = ocfs2_read_inode_block_full(inode, | |
517 | &bh, OCFS2_BH_IGNORE_CACHE); | |
b657c95c | 518 | } else { |
da1e9098 | 519 | status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh); |
f5ce5a08 SM |
520 | /* |
521 | * If buffer is in jbd, then its checksum may not have been | |
522 | * computed as yet. | |
523 | */ | |
d56a8f32 GH |
524 | if (!status && !buffer_jbd(bh)) { |
525 | if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK) | |
526 | status = ocfs2_filecheck_validate_inode_block( | |
527 | osb->sb, bh); | |
528 | else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX) | |
529 | status = ocfs2_filecheck_repair_inode_block( | |
530 | osb->sb, bh); | |
531 | else | |
532 | status = ocfs2_validate_inode_block( | |
533 | osb->sb, bh); | |
534 | } | |
b657c95c | 535 | } |
ccd979bd MF |
536 | if (status < 0) { |
537 | mlog_errno(status); | |
ccd979bd MF |
538 | goto bail; |
539 | } | |
540 | ||
24c19ef4 | 541 | status = -EINVAL; |
ccd979bd | 542 | fe = (struct ocfs2_dinode *) bh->b_data; |
ccd979bd | 543 | |
24c19ef4 MF |
544 | /* |
545 | * This is a code bug. Right now the caller needs to | |
546 | * understand whether it is asking for a system file inode or | |
547 | * not so the proper lock names can be built. | |
548 | */ | |
549 | mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) != | |
550 | !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE), | |
551 | "Inode %llu: system file state is ambigous\n", | |
552 | (unsigned long long)args->fi_blkno); | |
ccd979bd MF |
553 | |
554 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || | |
555 | S_ISBLK(le16_to_cpu(fe->i_mode))) | |
b657c95c | 556 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); |
ccd979bd | 557 | |
b657c95c | 558 | ocfs2_populate_inode(inode, fe, 0); |
ccd979bd MF |
559 | |
560 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); | |
561 | ||
d56a8f32 GH |
562 | if (buffer_dirty(bh) && !buffer_jbd(bh)) { |
563 | if (can_lock) { | |
564 | ocfs2_inode_unlock(inode, lock_level); | |
565 | lock_level = 1; | |
566 | ocfs2_inode_lock(inode, NULL, lock_level); | |
567 | } | |
568 | status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); | |
569 | if (status < 0) { | |
570 | mlog_errno(status); | |
571 | goto bail; | |
572 | } | |
573 | } | |
574 | ||
ccd979bd MF |
575 | status = 0; |
576 | ||
577 | bail: | |
24c19ef4 | 578 | if (can_lock) |
d56a8f32 | 579 | ocfs2_inode_unlock(inode, lock_level); |
24c19ef4 MF |
580 | |
581 | if (status < 0) | |
582 | make_bad_inode(inode); | |
583 | ||
4a635a11 | 584 | brelse(bh); |
ccd979bd | 585 | |
ccd979bd MF |
586 | return status; |
587 | } | |
588 | ||
589 | void ocfs2_sync_blockdev(struct super_block *sb) | |
590 | { | |
591 | sync_blockdev(sb->s_bdev); | |
592 | } | |
593 | ||
594 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
595 | struct inode *inode, | |
596 | struct buffer_head *fe_bh) | |
597 | { | |
598 | int status = 0; | |
ccd979bd | 599 | struct ocfs2_dinode *fe; |
60b11392 | 600 | handle_t *handle = NULL; |
ccd979bd | 601 | |
ccd979bd MF |
602 | fe = (struct ocfs2_dinode *) fe_bh->b_data; |
603 | ||
1afc32b9 MF |
604 | /* |
605 | * This check will also skip truncate of inodes with inline | |
606 | * data and fast symlinks. | |
607 | */ | |
3a0782d0 | 608 | if (fe->i_clusters) { |
2b4e30fb JB |
609 | if (ocfs2_should_order_data(inode)) |
610 | ocfs2_begin_ordered_truncate(inode, 0); | |
611 | ||
60b11392 MF |
612 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
613 | if (IS_ERR(handle)) { | |
614 | status = PTR_ERR(handle); | |
0350cb07 | 615 | handle = NULL; |
60b11392 MF |
616 | mlog_errno(status); |
617 | goto out; | |
618 | } | |
619 | ||
0cf2f763 JB |
620 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), |
621 | fe_bh, | |
13723d00 | 622 | OCFS2_JOURNAL_ACCESS_WRITE); |
60b11392 MF |
623 | if (status < 0) { |
624 | mlog_errno(status); | |
625 | goto out; | |
626 | } | |
627 | ||
628 | i_size_write(inode, 0); | |
629 | ||
630 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
631 | if (status < 0) { | |
632 | mlog_errno(status); | |
633 | goto out; | |
634 | } | |
635 | ||
636 | ocfs2_commit_trans(osb, handle); | |
637 | handle = NULL; | |
638 | ||
78f94673 | 639 | status = ocfs2_commit_truncate(osb, inode, fe_bh); |
3a0782d0 MF |
640 | if (status < 0) { |
641 | mlog_errno(status); | |
642 | goto out; | |
643 | } | |
ccd979bd | 644 | } |
ccd979bd | 645 | |
60b11392 MF |
646 | out: |
647 | if (handle) | |
648 | ocfs2_commit_trans(osb, handle); | |
ccd979bd MF |
649 | return status; |
650 | } | |
651 | ||
652 | static int ocfs2_remove_inode(struct inode *inode, | |
653 | struct buffer_head *di_bh, | |
654 | struct inode *orphan_dir_inode, | |
655 | struct buffer_head *orphan_dir_bh) | |
656 | { | |
657 | int status; | |
658 | struct inode *inode_alloc_inode = NULL; | |
659 | struct buffer_head *inode_alloc_bh = NULL; | |
1fabe148 | 660 | handle_t *handle; |
ccd979bd MF |
661 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
662 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
663 | ||
664 | inode_alloc_inode = | |
665 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
666 | le16_to_cpu(di->i_suballoc_slot)); | |
667 | if (!inode_alloc_inode) { | |
62f8b1f0 | 668 | status = -ENOENT; |
ccd979bd MF |
669 | mlog_errno(status); |
670 | goto bail; | |
671 | } | |
672 | ||
5955102c | 673 | inode_lock(inode_alloc_inode); |
e63aecb6 | 674 | status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1); |
ccd979bd | 675 | if (status < 0) { |
5955102c | 676 | inode_unlock(inode_alloc_inode); |
ccd979bd MF |
677 | |
678 | mlog_errno(status); | |
679 | goto bail; | |
680 | } | |
681 | ||
a90714c1 | 682 | handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS + |
9b7895ef | 683 | ocfs2_quota_trans_credits(inode->i_sb)); |
ccd979bd MF |
684 | if (IS_ERR(handle)) { |
685 | status = PTR_ERR(handle); | |
686 | mlog_errno(status); | |
687 | goto bail_unlock; | |
688 | } | |
689 | ||
d4cd1871 LD |
690 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { |
691 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, | |
06ee5c75 | 692 | orphan_dir_bh, false); |
d4cd1871 LD |
693 | if (status < 0) { |
694 | mlog_errno(status); | |
695 | goto bail_commit; | |
696 | } | |
ccd979bd MF |
697 | } |
698 | ||
699 | /* set the inodes dtime */ | |
0cf2f763 | 700 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 701 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
702 | if (status < 0) { |
703 | mlog_errno(status); | |
704 | goto bail_commit; | |
705 | } | |
706 | ||
c62c38f6 | 707 | di->i_dtime = cpu_to_le64(ktime_get_real_seconds()); |
4092d49f | 708 | di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); |
ec20cec7 | 709 | ocfs2_journal_dirty(handle, di_bh); |
ccd979bd | 710 | |
8cb471e8 | 711 | ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh); |
63936dda | 712 | dquot_free_inode(inode); |
ccd979bd MF |
713 | |
714 | status = ocfs2_free_dinode(handle, inode_alloc_inode, | |
715 | inode_alloc_bh, di); | |
716 | if (status < 0) | |
717 | mlog_errno(status); | |
718 | ||
719 | bail_commit: | |
02dc1af4 | 720 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 721 | bail_unlock: |
e63aecb6 | 722 | ocfs2_inode_unlock(inode_alloc_inode, 1); |
5955102c | 723 | inode_unlock(inode_alloc_inode); |
ccd979bd MF |
724 | brelse(inode_alloc_bh); |
725 | bail: | |
726 | iput(inode_alloc_inode); | |
727 | ||
728 | return status; | |
729 | } | |
730 | ||
2bd63216 | 731 | /* |
b4df6ed8 MF |
732 | * Serialize with orphan dir recovery. If the process doing |
733 | * recovery on this orphan dir does an iget() with the dir | |
734 | * i_mutex held, we'll deadlock here. Instead we detect this | |
735 | * and exit early - recovery will wipe this inode for us. | |
736 | */ | |
737 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, | |
738 | int slot) | |
739 | { | |
740 | int ret = 0; | |
741 | ||
742 | spin_lock(&osb->osb_lock); | |
743 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { | |
b4df6ed8 MF |
744 | ret = -EDEADLK; |
745 | goto out; | |
746 | } | |
747 | /* This signals to the orphan recovery process that it should | |
748 | * wait for us to handle the wipe. */ | |
749 | osb->osb_orphan_wipes[slot]++; | |
750 | out: | |
751 | spin_unlock(&osb->osb_lock); | |
64f3b269 | 752 | trace_ocfs2_check_orphan_recovery_state(slot, ret); |
b4df6ed8 MF |
753 | return ret; |
754 | } | |
755 | ||
756 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, | |
757 | int slot) | |
758 | { | |
759 | spin_lock(&osb->osb_lock); | |
760 | osb->osb_orphan_wipes[slot]--; | |
761 | spin_unlock(&osb->osb_lock); | |
762 | ||
763 | wake_up(&osb->osb_wipe_event); | |
764 | } | |
765 | ||
ccd979bd MF |
766 | static int ocfs2_wipe_inode(struct inode *inode, |
767 | struct buffer_head *di_bh) | |
768 | { | |
d577632e | 769 | int status, orphaned_slot = -1; |
ccd979bd MF |
770 | struct inode *orphan_dir_inode = NULL; |
771 | struct buffer_head *orphan_dir_bh = NULL; | |
772 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
d4cd1871 | 773 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; |
ccd979bd | 774 | |
d4cd1871 LD |
775 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { |
776 | orphaned_slot = le16_to_cpu(di->i_orphaned_slot); | |
b4df6ed8 | 777 | |
d4cd1871 LD |
778 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); |
779 | if (status) | |
780 | return status; | |
b4df6ed8 | 781 | |
d4cd1871 LD |
782 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
783 | ORPHAN_DIR_SYSTEM_INODE, | |
784 | orphaned_slot); | |
785 | if (!orphan_dir_inode) { | |
62f8b1f0 | 786 | status = -ENOENT; |
d4cd1871 LD |
787 | mlog_errno(status); |
788 | goto bail; | |
789 | } | |
ccd979bd | 790 | |
d4cd1871 LD |
791 | /* Lock the orphan dir. The lock will be held for the entire |
792 | * delete_inode operation. We do this now to avoid races with | |
793 | * recovery completion on other nodes. */ | |
5955102c | 794 | inode_lock(orphan_dir_inode); |
d4cd1871 LD |
795 | status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1); |
796 | if (status < 0) { | |
5955102c | 797 | inode_unlock(orphan_dir_inode); |
ccd979bd | 798 | |
d4cd1871 LD |
799 | mlog_errno(status); |
800 | goto bail; | |
801 | } | |
ccd979bd MF |
802 | } |
803 | ||
804 | /* we do this while holding the orphan dir lock because we | |
34d024f8 MF |
805 | * don't want recovery being run from another node to try an |
806 | * inode delete underneath us -- this will result in two nodes | |
ccd979bd MF |
807 | * truncating the same file! */ |
808 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); | |
809 | if (status < 0) { | |
810 | mlog_errno(status); | |
811 | goto bail_unlock_dir; | |
812 | } | |
813 | ||
9b7895ef MF |
814 | /* Remove any dir index tree */ |
815 | if (S_ISDIR(inode->i_mode)) { | |
816 | status = ocfs2_dx_dir_truncate(inode, di_bh); | |
817 | if (status) { | |
818 | mlog_errno(status); | |
819 | goto bail_unlock_dir; | |
820 | } | |
821 | } | |
822 | ||
cf1d6c76 TY |
823 | /*Free extended attribute resources associated with this inode.*/ |
824 | status = ocfs2_xattr_remove(inode, di_bh); | |
825 | if (status < 0) { | |
826 | mlog_errno(status); | |
827 | goto bail_unlock_dir; | |
828 | } | |
829 | ||
8b2c0dba TM |
830 | status = ocfs2_remove_refcount_tree(inode, di_bh); |
831 | if (status < 0) { | |
832 | mlog_errno(status); | |
833 | goto bail_unlock_dir; | |
834 | } | |
835 | ||
ccd979bd MF |
836 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, |
837 | orphan_dir_bh); | |
838 | if (status < 0) | |
839 | mlog_errno(status); | |
840 | ||
841 | bail_unlock_dir: | |
d4cd1871 LD |
842 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR) |
843 | return status; | |
844 | ||
e63aecb6 | 845 | ocfs2_inode_unlock(orphan_dir_inode, 1); |
5955102c | 846 | inode_unlock(orphan_dir_inode); |
ccd979bd MF |
847 | brelse(orphan_dir_bh); |
848 | bail: | |
849 | iput(orphan_dir_inode); | |
b4df6ed8 | 850 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
ccd979bd MF |
851 | |
852 | return status; | |
853 | } | |
854 | ||
855 | /* There is a series of simple checks that should be done before a | |
34d024f8 | 856 | * trylock is even considered. Encapsulate those in this function. */ |
ccd979bd MF |
857 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) |
858 | { | |
859 | int ret = 0; | |
860 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
861 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
862 | ||
64f3b269 TM |
863 | trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task, |
864 | (unsigned long long)oi->ip_blkno, | |
865 | oi->ip_flags); | |
866 | ||
ccd979bd MF |
867 | /* We shouldn't be getting here for the root directory |
868 | * inode.. */ | |
869 | if (inode == osb->root_inode) { | |
870 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); | |
871 | goto bail; | |
872 | } | |
873 | ||
bd62ad7a JK |
874 | /* |
875 | * If we're coming from downconvert_thread we can't go into our own | |
876 | * voting [hello, deadlock city!] so we cannot delete the inode. But | |
877 | * since we dropped last inode ref when downconverting dentry lock, | |
878 | * we cannot have the file open and thus the node doing unlink will | |
879 | * take care of deleting the inode. | |
880 | */ | |
64f3b269 | 881 | if (current == osb->dc_task) |
ccd979bd | 882 | goto bail; |
ccd979bd MF |
883 | |
884 | spin_lock(&oi->ip_lock); | |
885 | /* OCFS2 *never* deletes system files. This should technically | |
886 | * never get here as system file inodes should always have a | |
887 | * positive link count. */ | |
888 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { | |
b0697053 MF |
889 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
890 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
891 | goto bail_unlock; |
892 | } | |
893 | ||
ccd979bd MF |
894 | ret = 1; |
895 | bail_unlock: | |
896 | spin_unlock(&oi->ip_lock); | |
897 | bail: | |
898 | return ret; | |
899 | } | |
900 | ||
901 | /* Query the cluster to determine whether we should wipe an inode from | |
902 | * disk or not. | |
903 | * | |
904 | * Requires the inode to have the cluster lock. */ | |
905 | static int ocfs2_query_inode_wipe(struct inode *inode, | |
906 | struct buffer_head *di_bh, | |
907 | int *wipe) | |
908 | { | |
64f3b269 | 909 | int status = 0, reason = 0; |
ccd979bd MF |
910 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
911 | struct ocfs2_dinode *di; | |
912 | ||
913 | *wipe = 0; | |
914 | ||
64f3b269 TM |
915 | trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno, |
916 | inode->i_nlink); | |
917 | ||
ccd979bd MF |
918 | /* While we were waiting for the cluster lock in |
919 | * ocfs2_delete_inode, another node might have asked to delete | |
920 | * the inode. Recheck our flags to catch this. */ | |
921 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
64f3b269 | 922 | reason = 1; |
ccd979bd MF |
923 | goto bail; |
924 | } | |
925 | ||
926 | /* Now that we have an up to date inode, we can double check | |
927 | * the link count. */ | |
64f3b269 | 928 | if (inode->i_nlink) |
ccd979bd | 929 | goto bail; |
ccd979bd MF |
930 | |
931 | /* Do some basic inode verification... */ | |
932 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
d4cd1871 LD |
933 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) && |
934 | !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { | |
b54c2ca4 TY |
935 | /* |
936 | * Inodes in the orphan dir must have ORPHANED_FL. The only | |
937 | * inodes that come back out of the orphan dir are reflink | |
938 | * targets. A reflink target may be moved out of the orphan | |
939 | * dir between the time we scan the directory and the time we | |
940 | * process it. This would lead to HAS_REFCOUNT_FL being set but | |
941 | * ORPHANED_FL not. | |
942 | */ | |
943 | if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) { | |
64f3b269 | 944 | reason = 2; |
b54c2ca4 TY |
945 | goto bail; |
946 | } | |
947 | ||
ccd979bd MF |
948 | /* for lack of a better error? */ |
949 | status = -EEXIST; | |
950 | mlog(ML_ERROR, | |
b0697053 | 951 | "Inode %llu (on-disk %llu) not orphaned! " |
ccd979bd | 952 | "Disk flags 0x%x, inode flags 0x%x\n", |
b0697053 | 953 | (unsigned long long)oi->ip_blkno, |
1ca1a111 MF |
954 | (unsigned long long)le64_to_cpu(di->i_blkno), |
955 | le32_to_cpu(di->i_flags), oi->ip_flags); | |
ccd979bd MF |
956 | goto bail; |
957 | } | |
958 | ||
959 | /* has someone already deleted us?! baaad... */ | |
960 | if (di->i_dtime) { | |
961 | status = -EEXIST; | |
962 | mlog_errno(status); | |
963 | goto bail; | |
964 | } | |
965 | ||
6f16bf65 MF |
966 | /* |
967 | * This is how ocfs2 determines whether an inode is still live | |
968 | * within the cluster. Every node takes a shared read lock on | |
969 | * the inode open lock in ocfs2_read_locked_inode(). When we | |
970 | * get to ->delete_inode(), each node tries to convert it's | |
971 | * lock to an exclusive. Trylocks are serialized by the inode | |
25985edc | 972 | * meta data lock. If the upconvert succeeds, we know the inode |
6f16bf65 MF |
973 | * is no longer live and can be deleted. |
974 | * | |
975 | * Though we call this with the meta data lock held, the | |
976 | * trylock keeps us from ABBA deadlock. | |
977 | */ | |
978 | status = ocfs2_try_open_lock(inode, 1); | |
50008630 | 979 | if (status == -EAGAIN) { |
ccd979bd | 980 | status = 0; |
64f3b269 | 981 | reason = 3; |
ccd979bd MF |
982 | goto bail; |
983 | } | |
984 | if (status < 0) { | |
985 | mlog_errno(status); | |
986 | goto bail; | |
987 | } | |
988 | ||
50008630 | 989 | *wipe = 1; |
64f3b269 | 990 | trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot)); |
ccd979bd MF |
991 | |
992 | bail: | |
64f3b269 | 993 | trace_ocfs2_query_inode_wipe_end(status, reason); |
ccd979bd MF |
994 | return status; |
995 | } | |
996 | ||
997 | /* Support function for ocfs2_delete_inode. Will help us keep the | |
998 | * inode data in a consistent state for clear_inode. Always truncates | |
999 | * pages, optionally sync's them first. */ | |
1000 | static void ocfs2_cleanup_delete_inode(struct inode *inode, | |
1001 | int sync_data) | |
1002 | { | |
64f3b269 TM |
1003 | trace_ocfs2_cleanup_delete_inode( |
1004 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); | |
ccd979bd | 1005 | if (sync_data) |
249ec93c | 1006 | filemap_write_and_wait(inode->i_mapping); |
91b0abe3 | 1007 | truncate_inode_pages_final(&inode->i_data); |
ccd979bd MF |
1008 | } |
1009 | ||
066d92dc | 1010 | static void ocfs2_delete_inode(struct inode *inode) |
ccd979bd MF |
1011 | { |
1012 | int wipe, status; | |
e4b963f1 | 1013 | sigset_t oldset; |
ccd979bd | 1014 | struct buffer_head *di_bh = NULL; |
ad694821 | 1015 | struct ocfs2_dinode *di = NULL; |
ccd979bd | 1016 | |
64f3b269 TM |
1017 | trace_ocfs2_delete_inode(inode->i_ino, |
1018 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1019 | is_bad_inode(inode)); | |
ccd979bd | 1020 | |
a90714c1 JK |
1021 | /* When we fail in read_inode() we mark inode as bad. The second test |
1022 | * catches the case when inode allocation fails before allocating | |
1023 | * a block for inode. */ | |
64f3b269 | 1024 | if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) |
ccd979bd | 1025 | goto bail; |
ccd979bd MF |
1026 | |
1027 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
1028 | /* It's probably not necessary to truncate_inode_pages | |
1029 | * here but we do it for safety anyway (it will most | |
1030 | * likely be a no-op anyway) */ | |
1031 | ocfs2_cleanup_delete_inode(inode, 0); | |
1032 | goto bail; | |
1033 | } | |
1034 | ||
bd62ad7a JK |
1035 | dquot_initialize(inode); |
1036 | ||
ccd979bd MF |
1037 | /* We want to block signals in delete_inode as the lock and |
1038 | * messaging paths may return us -ERESTARTSYS. Which would | |
1039 | * cause us to exit early, resulting in inodes being orphaned | |
1040 | * forever. */ | |
e4b963f1 | 1041 | ocfs2_block_signals(&oldset); |
ccd979bd | 1042 | |
6ca497a8 | 1043 | /* |
1044 | * Synchronize us against ocfs2_get_dentry. We take this in | |
1045 | * shared mode so that all nodes can still concurrently | |
1046 | * process deletes. | |
1047 | */ | |
1048 | status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0); | |
1049 | if (status < 0) { | |
1050 | mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status); | |
1051 | ocfs2_cleanup_delete_inode(inode, 0); | |
1052 | goto bail_unblock; | |
1053 | } | |
ccd979bd MF |
1054 | /* Lock down the inode. This gives us an up to date view of |
1055 | * it's metadata (for verification), and allows us to | |
34d024f8 | 1056 | * serialize delete_inode on multiple nodes. |
ccd979bd MF |
1057 | * |
1058 | * Even though we might be doing a truncate, we don't take the | |
1059 | * allocation lock here as it won't be needed - nobody will | |
1060 | * have the file open. | |
1061 | */ | |
e63aecb6 | 1062 | status = ocfs2_inode_lock(inode, &di_bh, 1); |
ccd979bd MF |
1063 | if (status < 0) { |
1064 | if (status != -ENOENT) | |
1065 | mlog_errno(status); | |
1066 | ocfs2_cleanup_delete_inode(inode, 0); | |
6ca497a8 | 1067 | goto bail_unlock_nfs_sync; |
ccd979bd MF |
1068 | } |
1069 | ||
ad694821 JQ |
1070 | di = (struct ocfs2_dinode *)di_bh->b_data; |
1071 | /* Skip inode deletion and wait for dio orphan entry recovered | |
1072 | * first */ | |
1073 | if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) { | |
1074 | ocfs2_cleanup_delete_inode(inode, 0); | |
1075 | goto bail_unlock_inode; | |
1076 | } | |
1077 | ||
ccd979bd MF |
1078 | /* Query the cluster. This will be the final decision made |
1079 | * before we go ahead and wipe the inode. */ | |
1080 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); | |
1081 | if (!wipe || status < 0) { | |
34d024f8 | 1082 | /* Error and remote inode busy both mean we won't be |
ccd979bd MF |
1083 | * removing the inode, so they take almost the same |
1084 | * path. */ | |
1085 | if (status < 0) | |
1086 | mlog_errno(status); | |
1087 | ||
34d024f8 MF |
1088 | /* Someone in the cluster has disallowed a wipe of |
1089 | * this inode, or it was never completely | |
1090 | * orphaned. Write out the pages and exit now. */ | |
ccd979bd MF |
1091 | ocfs2_cleanup_delete_inode(inode, 1); |
1092 | goto bail_unlock_inode; | |
1093 | } | |
1094 | ||
1095 | ocfs2_cleanup_delete_inode(inode, 0); | |
1096 | ||
1097 | status = ocfs2_wipe_inode(inode, di_bh); | |
1098 | if (status < 0) { | |
b4df6ed8 MF |
1099 | if (status != -EDEADLK) |
1100 | mlog_errno(status); | |
ccd979bd MF |
1101 | goto bail_unlock_inode; |
1102 | } | |
1103 | ||
24c19ef4 MF |
1104 | /* |
1105 | * Mark the inode as successfully deleted. | |
1106 | * | |
1107 | * This is important for ocfs2_clear_inode() as it will check | |
1108 | * this flag and skip any checkpointing work | |
1109 | * | |
1110 | * ocfs2_stuff_meta_lvb() also uses this flag to invalidate | |
1111 | * the LVB for other nodes. | |
1112 | */ | |
ccd979bd MF |
1113 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; |
1114 | ||
1115 | bail_unlock_inode: | |
e63aecb6 | 1116 | ocfs2_inode_unlock(inode, 1); |
ccd979bd | 1117 | brelse(di_bh); |
6ca497a8 | 1118 | |
1119 | bail_unlock_nfs_sync: | |
1120 | ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0); | |
1121 | ||
ccd979bd | 1122 | bail_unblock: |
e4b963f1 | 1123 | ocfs2_unblock_signals(&oldset); |
ccd979bd | 1124 | bail: |
c1e8d35e | 1125 | return; |
ccd979bd MF |
1126 | } |
1127 | ||
066d92dc | 1128 | static void ocfs2_clear_inode(struct inode *inode) |
ccd979bd MF |
1129 | { |
1130 | int status; | |
1131 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
84d86f83 | 1132 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd | 1133 | |
dbd5768f | 1134 | clear_inode(inode); |
64f3b269 TM |
1135 | trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno, |
1136 | inode->i_nlink); | |
ccd979bd MF |
1137 | |
1138 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, | |
1139 | "Inode=%lu\n", inode->i_ino); | |
1140 | ||
9f754758 | 1141 | dquot_drop(inode); |
257ba15c | 1142 | |
34d024f8 MF |
1143 | /* To preven remote deletes we hold open lock before, now it |
1144 | * is time to unlock PR and EX open locks. */ | |
50008630 TY |
1145 | ocfs2_open_unlock(inode); |
1146 | ||
ccd979bd | 1147 | /* Do these before all the other work so that we don't bounce |
34d024f8 | 1148 | * the downconvert thread while waiting to destroy the locks. */ |
84d86f83 JK |
1149 | ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres); |
1150 | ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres); | |
1151 | ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres); | |
ccd979bd | 1152 | |
4fe370af MF |
1153 | ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap, |
1154 | &oi->ip_la_data_resv); | |
1155 | ocfs2_resv_init_once(&oi->ip_la_data_resv); | |
1156 | ||
ccd979bd MF |
1157 | /* We very well may get a clear_inode before all an inodes |
1158 | * metadata has hit disk. Of course, we can't drop any cluster | |
1159 | * locks until the journal has finished with it. The only | |
1160 | * exception here are successfully wiped inodes - their | |
1161 | * metadata can now be considered to be part of the system | |
1162 | * inodes from which it came. */ | |
1163 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) | |
1164 | ocfs2_checkpoint_inode(inode); | |
1165 | ||
1166 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), | |
b0697053 MF |
1167 | "Clear inode of %llu, inode has io markers\n", |
1168 | (unsigned long long)oi->ip_blkno); | |
4506cfb6 RD |
1169 | mlog_bug_on_msg(!list_empty(&oi->ip_unwritten_list), |
1170 | "Clear inode of %llu, inode has unwritten extents\n", | |
1171 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1172 | |
83418978 MF |
1173 | ocfs2_extent_map_trunc(inode, 0); |
1174 | ||
ccd979bd MF |
1175 | status = ocfs2_drop_inode_locks(inode); |
1176 | if (status < 0) | |
1177 | mlog_errno(status); | |
1178 | ||
1179 | ocfs2_lock_res_free(&oi->ip_rw_lockres); | |
e63aecb6 | 1180 | ocfs2_lock_res_free(&oi->ip_inode_lockres); |
50008630 | 1181 | ocfs2_lock_res_free(&oi->ip_open_lockres); |
ccd979bd | 1182 | |
66fb345d | 1183 | ocfs2_metadata_cache_exit(INODE_CACHE(inode)); |
ccd979bd | 1184 | |
8cb471e8 | 1185 | mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached, |
b0697053 | 1186 | "Clear inode of %llu, inode has %u cache items\n", |
8cb471e8 JB |
1187 | (unsigned long long)oi->ip_blkno, |
1188 | INODE_CACHE(inode)->ci_num_cached); | |
ccd979bd | 1189 | |
8cb471e8 | 1190 | mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE), |
b0697053 MF |
1191 | "Clear inode of %llu, inode has a bad flag\n", |
1192 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1193 | |
1194 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), | |
b0697053 MF |
1195 | "Clear inode of %llu, inode is locked\n", |
1196 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1197 | |
251b6ecc | 1198 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
b0697053 MF |
1199 | "Clear inode of %llu, io_mutex is locked\n", |
1200 | (unsigned long long)oi->ip_blkno); | |
251b6ecc | 1201 | mutex_unlock(&oi->ip_io_mutex); |
ccd979bd MF |
1202 | |
1203 | /* | |
1204 | * down_trylock() returns 0, down_write_trylock() returns 1 | |
1205 | * kernel 1, world 0 | |
1206 | */ | |
1207 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), | |
b0697053 MF |
1208 | "Clear inode of %llu, alloc_sem is locked\n", |
1209 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1210 | up_write(&oi->ip_alloc_sem); |
1211 | ||
1212 | mlog_bug_on_msg(oi->ip_open_count, | |
b0697053 MF |
1213 | "Clear inode of %llu has open count %d\n", |
1214 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); | |
ccd979bd MF |
1215 | |
1216 | /* Clear all other flags. */ | |
47460d65 | 1217 | oi->ip_flags = 0; |
ccd979bd MF |
1218 | oi->ip_dir_start_lookup = 0; |
1219 | oi->ip_blkno = 0ULL; | |
ae0dff68 SM |
1220 | |
1221 | /* | |
1222 | * ip_jinode is used to track txns against this inode. We ensure that | |
1223 | * the journal is flushed before journal shutdown. Thus it is safe to | |
1224 | * have inodes get cleaned up after journal shutdown. | |
1225 | */ | |
2b4e30fb JB |
1226 | jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal, |
1227 | &oi->ip_jinode); | |
ccd979bd MF |
1228 | } |
1229 | ||
066d92dc AV |
1230 | void ocfs2_evict_inode(struct inode *inode) |
1231 | { | |
1232 | if (!inode->i_nlink || | |
1233 | (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) { | |
1234 | ocfs2_delete_inode(inode); | |
1235 | } else { | |
91b0abe3 | 1236 | truncate_inode_pages_final(&inode->i_data); |
066d92dc AV |
1237 | } |
1238 | ocfs2_clear_inode(inode); | |
1239 | } | |
1240 | ||
ccd979bd MF |
1241 | /* Called under inode_lock, with no more references on the |
1242 | * struct inode, so it's safe here to check the flags field | |
1243 | * and to manipulate i_nlink without any other locks. */ | |
45321ac5 | 1244 | int ocfs2_drop_inode(struct inode *inode) |
ccd979bd MF |
1245 | { |
1246 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1247 | ||
64f3b269 TM |
1248 | trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno, |
1249 | inode->i_nlink, oi->ip_flags); | |
ccd979bd | 1250 | |
513e2dae X |
1251 | assert_spin_locked(&inode->i_lock); |
1252 | inode->i_state |= I_WILL_FREE; | |
1253 | spin_unlock(&inode->i_lock); | |
1254 | write_inode_now(inode, 1); | |
1255 | spin_lock(&inode->i_lock); | |
1256 | WARN_ON(inode->i_state & I_NEW); | |
1257 | inode->i_state &= ~I_WILL_FREE; | |
ccd979bd | 1258 | |
513e2dae | 1259 | return 1; |
ccd979bd MF |
1260 | } |
1261 | ||
ccd979bd MF |
1262 | /* |
1263 | * This is called from our getattr. | |
1264 | */ | |
1265 | int ocfs2_inode_revalidate(struct dentry *dentry) | |
1266 | { | |
2b0143b5 | 1267 | struct inode *inode = d_inode(dentry); |
ccd979bd MF |
1268 | int status = 0; |
1269 | ||
64f3b269 TM |
1270 | trace_ocfs2_inode_revalidate(inode, |
1271 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL, | |
1272 | inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0); | |
ccd979bd MF |
1273 | |
1274 | if (!inode) { | |
ccd979bd MF |
1275 | status = -ENOENT; |
1276 | goto bail; | |
1277 | } | |
1278 | ||
1279 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1280 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
1281 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
ccd979bd MF |
1282 | status = -ENOENT; |
1283 | goto bail; | |
1284 | } | |
1285 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1286 | ||
e63aecb6 | 1287 | /* Let ocfs2_inode_lock do the work of updating our struct |
ccd979bd | 1288 | * inode for us. */ |
e63aecb6 | 1289 | status = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
1290 | if (status < 0) { |
1291 | if (status != -ENOENT) | |
1292 | mlog_errno(status); | |
1293 | goto bail; | |
1294 | } | |
e63aecb6 | 1295 | ocfs2_inode_unlock(inode, 0); |
ccd979bd | 1296 | bail: |
ccd979bd MF |
1297 | return status; |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Updates a disk inode from a | |
1302 | * struct inode. | |
1303 | * Only takes ip_lock. | |
1304 | */ | |
1fabe148 | 1305 | int ocfs2_mark_inode_dirty(handle_t *handle, |
ccd979bd MF |
1306 | struct inode *inode, |
1307 | struct buffer_head *bh) | |
1308 | { | |
1309 | int status; | |
1310 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
1311 | ||
64f3b269 | 1312 | trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd | 1313 | |
0cf2f763 | 1314 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 1315 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1316 | if (status < 0) { |
1317 | mlog_errno(status); | |
1318 | goto leave; | |
1319 | } | |
1320 | ||
1321 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1322 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
6e4b0d56 | 1323 | ocfs2_get_inode_flags(OCFS2_I(inode)); |
ca4d147e | 1324 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
15b1e36b | 1325 | fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features); |
ccd979bd MF |
1326 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1327 | ||
1328 | fe->i_size = cpu_to_le64(i_size_read(inode)); | |
198a1ca3 | 1329 | ocfs2_set_links_count(fe, inode->i_nlink); |
2c034176 EB |
1330 | fe->i_uid = cpu_to_le32(i_uid_read(inode)); |
1331 | fe->i_gid = cpu_to_le32(i_gid_read(inode)); | |
ccd979bd MF |
1332 | fe->i_mode = cpu_to_le16(inode->i_mode); |
1333 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); | |
1334 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
1335 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
1336 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
1337 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1338 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
1339 | ||
ec20cec7 | 1340 | ocfs2_journal_dirty(handle, bh); |
2931cdcb | 1341 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
ccd979bd | 1342 | leave: |
ccd979bd MF |
1343 | return status; |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * | |
1348 | * Updates a struct inode from a disk inode. | |
1349 | * does no i/o, only takes ip_lock. | |
1350 | */ | |
1351 | void ocfs2_refresh_inode(struct inode *inode, | |
1352 | struct ocfs2_dinode *fe) | |
1353 | { | |
ccd979bd MF |
1354 | spin_lock(&OCFS2_I(inode)->ip_lock); |
1355 | ||
1356 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e | 1357 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
15b1e36b | 1358 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
ca4d147e | 1359 | ocfs2_set_inode_flags(inode); |
ccd979bd | 1360 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
bfe86848 | 1361 | set_nlink(inode, ocfs2_read_links_count(fe)); |
2c034176 EB |
1362 | i_uid_write(inode, le32_to_cpu(fe->i_uid)); |
1363 | i_gid_write(inode, le32_to_cpu(fe->i_gid)); | |
ccd979bd | 1364 | inode->i_mode = le16_to_cpu(fe->i_mode); |
ccd979bd MF |
1365 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) |
1366 | inode->i_blocks = 0; | |
1367 | else | |
8110b073 | 1368 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
1369 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
1370 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
1371 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
1372 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
1373 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
1374 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
1375 | ||
1376 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1377 | } | |
b657c95c JB |
1378 | |
1379 | int ocfs2_validate_inode_block(struct super_block *sb, | |
1380 | struct buffer_head *bh) | |
1381 | { | |
d6b32bbb | 1382 | int rc; |
b657c95c JB |
1383 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; |
1384 | ||
64f3b269 | 1385 | trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr); |
970e4936 | 1386 | |
b657c95c JB |
1387 | BUG_ON(!buffer_uptodate(bh)); |
1388 | ||
d6b32bbb JB |
1389 | /* |
1390 | * If the ecc fails, we return the error but otherwise | |
1391 | * leave the filesystem running. We know any error is | |
1392 | * local to this block. | |
1393 | */ | |
1394 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check); | |
13723d00 JB |
1395 | if (rc) { |
1396 | mlog(ML_ERROR, "Checksum failed for dinode %llu\n", | |
1397 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 1398 | goto bail; |
13723d00 | 1399 | } |
d6b32bbb JB |
1400 | |
1401 | /* | |
1402 | * Errors after here are fatal. | |
1403 | */ | |
1404 | ||
1405 | rc = -EINVAL; | |
1406 | ||
b657c95c | 1407 | if (!OCFS2_IS_VALID_DINODE(di)) { |
17a5b9ab | 1408 | rc = ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n", |
7ecef14a JP |
1409 | (unsigned long long)bh->b_blocknr, 7, |
1410 | di->i_signature); | |
b657c95c JB |
1411 | goto bail; |
1412 | } | |
1413 | ||
1414 | if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) { | |
17a5b9ab | 1415 | rc = ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n", |
7ecef14a JP |
1416 | (unsigned long long)bh->b_blocknr, |
1417 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
b657c95c JB |
1418 | goto bail; |
1419 | } | |
1420 | ||
1421 | if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
17a5b9ab | 1422 | rc = ocfs2_error(sb, |
7ecef14a JP |
1423 | "Invalid dinode #%llu: OCFS2_VALID_FL not set\n", |
1424 | (unsigned long long)bh->b_blocknr); | |
b657c95c JB |
1425 | goto bail; |
1426 | } | |
1427 | ||
1428 | if (le32_to_cpu(di->i_fs_generation) != | |
1429 | OCFS2_SB(sb)->fs_generation) { | |
17a5b9ab | 1430 | rc = ocfs2_error(sb, |
7ecef14a JP |
1431 | "Invalid dinode #%llu: fs_generation is %u\n", |
1432 | (unsigned long long)bh->b_blocknr, | |
1433 | le32_to_cpu(di->i_fs_generation)); | |
b657c95c JB |
1434 | goto bail; |
1435 | } | |
1436 | ||
1437 | rc = 0; | |
1438 | ||
1439 | bail: | |
1440 | return rc; | |
1441 | } | |
1442 | ||
d56a8f32 GH |
1443 | static int ocfs2_filecheck_validate_inode_block(struct super_block *sb, |
1444 | struct buffer_head *bh) | |
1445 | { | |
1446 | int rc = 0; | |
1447 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; | |
1448 | ||
1449 | trace_ocfs2_filecheck_validate_inode_block( | |
1450 | (unsigned long long)bh->b_blocknr); | |
1451 | ||
1452 | BUG_ON(!buffer_uptodate(bh)); | |
1453 | ||
1454 | /* | |
1455 | * Call ocfs2_validate_meta_ecc() first since it has ecc repair | |
1456 | * function, but we should not return error immediately when ecc | |
1457 | * validation fails, because the reason is quite likely the invalid | |
1458 | * inode number inputed. | |
1459 | */ | |
1460 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check); | |
1461 | if (rc) { | |
1462 | mlog(ML_ERROR, | |
1463 | "Filecheck: checksum failed for dinode %llu\n", | |
1464 | (unsigned long long)bh->b_blocknr); | |
1465 | rc = -OCFS2_FILECHECK_ERR_BLOCKECC; | |
1466 | } | |
1467 | ||
1468 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
1469 | mlog(ML_ERROR, | |
1470 | "Filecheck: invalid dinode #%llu: signature = %.*s\n", | |
1471 | (unsigned long long)bh->b_blocknr, 7, di->i_signature); | |
1472 | rc = -OCFS2_FILECHECK_ERR_INVALIDINO; | |
1473 | goto bail; | |
1474 | } else if (rc) | |
1475 | goto bail; | |
1476 | ||
1477 | if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) { | |
1478 | mlog(ML_ERROR, | |
1479 | "Filecheck: invalid dinode #%llu: i_blkno is %llu\n", | |
1480 | (unsigned long long)bh->b_blocknr, | |
1481 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
1482 | rc = -OCFS2_FILECHECK_ERR_BLOCKNO; | |
1483 | goto bail; | |
1484 | } | |
1485 | ||
1486 | if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
1487 | mlog(ML_ERROR, | |
1488 | "Filecheck: invalid dinode #%llu: OCFS2_VALID_FL " | |
1489 | "not set\n", | |
1490 | (unsigned long long)bh->b_blocknr); | |
1491 | rc = -OCFS2_FILECHECK_ERR_VALIDFLAG; | |
1492 | goto bail; | |
1493 | } | |
1494 | ||
1495 | if (le32_to_cpu(di->i_fs_generation) != | |
1496 | OCFS2_SB(sb)->fs_generation) { | |
1497 | mlog(ML_ERROR, | |
1498 | "Filecheck: invalid dinode #%llu: fs_generation is %u\n", | |
1499 | (unsigned long long)bh->b_blocknr, | |
1500 | le32_to_cpu(di->i_fs_generation)); | |
1501 | rc = -OCFS2_FILECHECK_ERR_GENERATION; | |
1502 | goto bail; | |
1503 | } | |
1504 | ||
1505 | bail: | |
1506 | return rc; | |
1507 | } | |
1508 | ||
1509 | static int ocfs2_filecheck_repair_inode_block(struct super_block *sb, | |
1510 | struct buffer_head *bh) | |
1511 | { | |
1512 | int changed = 0; | |
1513 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; | |
1514 | ||
1515 | if (!ocfs2_filecheck_validate_inode_block(sb, bh)) | |
1516 | return 0; | |
1517 | ||
1518 | trace_ocfs2_filecheck_repair_inode_block( | |
1519 | (unsigned long long)bh->b_blocknr); | |
1520 | ||
1521 | if (ocfs2_is_hard_readonly(OCFS2_SB(sb)) || | |
1522 | ocfs2_is_soft_readonly(OCFS2_SB(sb))) { | |
1523 | mlog(ML_ERROR, | |
1524 | "Filecheck: cannot repair dinode #%llu " | |
1525 | "on readonly filesystem\n", | |
1526 | (unsigned long long)bh->b_blocknr); | |
1527 | return -OCFS2_FILECHECK_ERR_READONLY; | |
1528 | } | |
1529 | ||
1530 | if (buffer_jbd(bh)) { | |
1531 | mlog(ML_ERROR, | |
1532 | "Filecheck: cannot repair dinode #%llu, " | |
1533 | "its buffer is in jbd\n", | |
1534 | (unsigned long long)bh->b_blocknr); | |
1535 | return -OCFS2_FILECHECK_ERR_INJBD; | |
1536 | } | |
1537 | ||
1538 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
1539 | /* Cannot fix invalid inode block */ | |
1540 | return -OCFS2_FILECHECK_ERR_INVALIDINO; | |
1541 | } | |
1542 | ||
1543 | if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
1544 | /* Cannot just add VALID_FL flag back as a fix, | |
1545 | * need more things to check here. | |
1546 | */ | |
1547 | return -OCFS2_FILECHECK_ERR_VALIDFLAG; | |
1548 | } | |
1549 | ||
1550 | if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) { | |
1551 | di->i_blkno = cpu_to_le64(bh->b_blocknr); | |
1552 | changed = 1; | |
1553 | mlog(ML_ERROR, | |
1554 | "Filecheck: reset dinode #%llu: i_blkno to %llu\n", | |
1555 | (unsigned long long)bh->b_blocknr, | |
1556 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
1557 | } | |
1558 | ||
1559 | if (le32_to_cpu(di->i_fs_generation) != | |
1560 | OCFS2_SB(sb)->fs_generation) { | |
1561 | di->i_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation); | |
1562 | changed = 1; | |
1563 | mlog(ML_ERROR, | |
1564 | "Filecheck: reset dinode #%llu: fs_generation to %u\n", | |
1565 | (unsigned long long)bh->b_blocknr, | |
1566 | le32_to_cpu(di->i_fs_generation)); | |
1567 | } | |
1568 | ||
1569 | if (changed || ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check)) { | |
1570 | ocfs2_compute_meta_ecc(sb, bh->b_data, &di->i_check); | |
1571 | mark_buffer_dirty(bh); | |
1572 | mlog(ML_ERROR, | |
1573 | "Filecheck: reset dinode #%llu: compute meta ecc\n", | |
1574 | (unsigned long long)bh->b_blocknr); | |
1575 | } | |
1576 | ||
1577 | return 0; | |
1578 | } | |
1579 | ||
1580 | static int | |
1581 | ocfs2_filecheck_read_inode_block_full(struct inode *inode, | |
1582 | struct buffer_head **bh, | |
1583 | int flags, int type) | |
1584 | { | |
1585 | int rc; | |
1586 | struct buffer_head *tmp = *bh; | |
1587 | ||
1588 | if (!type) /* Check inode block */ | |
1589 | rc = ocfs2_read_blocks(INODE_CACHE(inode), | |
1590 | OCFS2_I(inode)->ip_blkno, | |
1591 | 1, &tmp, flags, | |
1592 | ocfs2_filecheck_validate_inode_block); | |
1593 | else /* Repair inode block */ | |
1594 | rc = ocfs2_read_blocks(INODE_CACHE(inode), | |
1595 | OCFS2_I(inode)->ip_blkno, | |
1596 | 1, &tmp, flags, | |
1597 | ocfs2_filecheck_repair_inode_block); | |
1598 | ||
1599 | /* If ocfs2_read_blocks() got us a new bh, pass it up. */ | |
1600 | if (!rc && !*bh) | |
1601 | *bh = tmp; | |
1602 | ||
1603 | return rc; | |
1604 | } | |
1605 | ||
b657c95c JB |
1606 | int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh, |
1607 | int flags) | |
1608 | { | |
1609 | int rc; | |
1610 | struct buffer_head *tmp = *bh; | |
1611 | ||
8cb471e8 JB |
1612 | rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno, |
1613 | 1, &tmp, flags, ocfs2_validate_inode_block); | |
b657c95c JB |
1614 | |
1615 | /* If ocfs2_read_blocks() got us a new bh, pass it up. */ | |
970e4936 | 1616 | if (!rc && !*bh) |
b657c95c JB |
1617 | *bh = tmp; |
1618 | ||
b657c95c JB |
1619 | return rc; |
1620 | } | |
1621 | ||
1622 | int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh) | |
1623 | { | |
1624 | return ocfs2_read_inode_block_full(inode, bh, 0); | |
1625 | } | |
6e5a3d75 | 1626 | |
6e5a3d75 JB |
1627 | |
1628 | static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci) | |
1629 | { | |
1630 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1631 | ||
1632 | return oi->ip_blkno; | |
1633 | } | |
1634 | ||
8cb471e8 JB |
1635 | static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci) |
1636 | { | |
1637 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1638 | ||
1639 | return oi->vfs_inode.i_sb; | |
1640 | } | |
1641 | ||
6e5a3d75 JB |
1642 | static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci) |
1643 | { | |
1644 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1645 | ||
1646 | spin_lock(&oi->ip_lock); | |
1647 | } | |
1648 | ||
1649 | static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci) | |
1650 | { | |
1651 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1652 | ||
1653 | spin_unlock(&oi->ip_lock); | |
1654 | } | |
1655 | ||
1656 | static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci) | |
1657 | { | |
1658 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1659 | ||
1660 | mutex_lock(&oi->ip_io_mutex); | |
1661 | } | |
1662 | ||
1663 | static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci) | |
1664 | { | |
1665 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1666 | ||
1667 | mutex_unlock(&oi->ip_io_mutex); | |
1668 | } | |
1669 | ||
1670 | const struct ocfs2_caching_operations ocfs2_inode_caching_ops = { | |
1671 | .co_owner = ocfs2_inode_cache_owner, | |
8cb471e8 | 1672 | .co_get_super = ocfs2_inode_cache_get_super, |
6e5a3d75 JB |
1673 | .co_cache_lock = ocfs2_inode_cache_lock, |
1674 | .co_cache_unlock = ocfs2_inode_cache_unlock, | |
1675 | .co_io_lock = ocfs2_inode_cache_io_lock, | |
1676 | .co_io_unlock = ocfs2_inode_cache_io_unlock, | |
1677 | }; | |
1678 |