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> |
ccd979bd MF |
31 | |
32 | #include <asm/byteorder.h> | |
33 | ||
ccd979bd MF |
34 | #include <cluster/masklog.h> |
35 | ||
36 | #include "ocfs2.h" | |
37 | ||
38 | #include "alloc.h" | |
9b7895ef | 39 | #include "dir.h" |
d6b32bbb | 40 | #include "blockcheck.h" |
ccd979bd MF |
41 | #include "dlmglue.h" |
42 | #include "extent_map.h" | |
43 | #include "file.h" | |
b4df6ed8 | 44 | #include "heartbeat.h" |
ccd979bd MF |
45 | #include "inode.h" |
46 | #include "journal.h" | |
47 | #include "namei.h" | |
48 | #include "suballoc.h" | |
49 | #include "super.h" | |
50 | #include "symlink.h" | |
51 | #include "sysfile.h" | |
52 | #include "uptodate.h" | |
cf1d6c76 | 53 | #include "xattr.h" |
8b2c0dba | 54 | #include "refcounttree.h" |
64f3b269 | 55 | #include "ocfs2_trace.h" |
d56a8f32 | 56 | #include "filecheck.h" |
ccd979bd MF |
57 | |
58 | #include "buffer_head_io.h" | |
59 | ||
ccd979bd MF |
60 | struct ocfs2_find_inode_args |
61 | { | |
62 | u64 fi_blkno; | |
63 | unsigned long fi_ino; | |
64 | unsigned int fi_flags; | |
5fa0613e | 65 | unsigned int fi_sysfile_type; |
ccd979bd MF |
66 | }; |
67 | ||
5fa0613e JK |
68 | static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES]; |
69 | ||
ccd979bd MF |
70 | static int ocfs2_read_locked_inode(struct inode *inode, |
71 | struct ocfs2_find_inode_args *args); | |
72 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); | |
73 | static int ocfs2_find_actor(struct inode *inode, void *opaque); | |
74 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
75 | struct inode *inode, | |
76 | struct buffer_head *fe_bh); | |
77 | ||
d56a8f32 GH |
78 | static int ocfs2_filecheck_read_inode_block_full(struct inode *inode, |
79 | struct buffer_head **bh, | |
80 | int flags, int type); | |
81 | static int ocfs2_filecheck_validate_inode_block(struct super_block *sb, | |
82 | struct buffer_head *bh); | |
83 | static int ocfs2_filecheck_repair_inode_block(struct super_block *sb, | |
84 | struct buffer_head *bh); | |
85 | ||
ca4d147e HP |
86 | void ocfs2_set_inode_flags(struct inode *inode) |
87 | { | |
88 | unsigned int flags = OCFS2_I(inode)->ip_attr; | |
89 | ||
90 | inode->i_flags &= ~(S_IMMUTABLE | | |
91 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); | |
92 | ||
93 | if (flags & OCFS2_IMMUTABLE_FL) | |
94 | inode->i_flags |= S_IMMUTABLE; | |
95 | ||
96 | if (flags & OCFS2_SYNC_FL) | |
97 | inode->i_flags |= S_SYNC; | |
98 | if (flags & OCFS2_APPEND_FL) | |
99 | inode->i_flags |= S_APPEND; | |
100 | if (flags & OCFS2_NOATIME_FL) | |
101 | inode->i_flags |= S_NOATIME; | |
102 | if (flags & OCFS2_DIRSYNC_FL) | |
103 | inode->i_flags |= S_DIRSYNC; | |
104 | } | |
105 | ||
6e4b0d56 JK |
106 | /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */ |
107 | void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi) | |
108 | { | |
109 | unsigned int flags = oi->vfs_inode.i_flags; | |
110 | ||
111 | oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL| | |
112 | OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL); | |
113 | if (flags & S_SYNC) | |
114 | oi->ip_attr |= OCFS2_SYNC_FL; | |
115 | if (flags & S_APPEND) | |
116 | oi->ip_attr |= OCFS2_APPEND_FL; | |
117 | if (flags & S_IMMUTABLE) | |
118 | oi->ip_attr |= OCFS2_IMMUTABLE_FL; | |
119 | if (flags & S_NOATIME) | |
120 | oi->ip_attr |= OCFS2_NOATIME_FL; | |
121 | if (flags & S_DIRSYNC) | |
122 | oi->ip_attr |= OCFS2_DIRSYNC_FL; | |
123 | } | |
124 | ||
6ca497a8 | 125 | struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno) |
126 | { | |
127 | struct ocfs2_find_inode_args args; | |
128 | ||
129 | args.fi_blkno = blkno; | |
130 | args.fi_flags = 0; | |
131 | args.fi_ino = ino_from_blkno(sb, blkno); | |
132 | args.fi_sysfile_type = 0; | |
133 | ||
134 | return ilookup5(sb, blkno, ocfs2_find_actor, &args); | |
135 | } | |
5fa0613e JK |
136 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags, |
137 | int sysfile_type) | |
ccd979bd | 138 | { |
d56a8f32 | 139 | int rc = 0; |
ccd979bd MF |
140 | struct inode *inode = NULL; |
141 | struct super_block *sb = osb->sb; | |
142 | struct ocfs2_find_inode_args args; | |
2931cdcb | 143 | journal_t *journal = OCFS2_SB(sb)->journal->j_journal; |
ccd979bd | 144 | |
64f3b269 TM |
145 | trace_ocfs2_iget_begin((unsigned long long)blkno, flags, |
146 | sysfile_type); | |
ccd979bd MF |
147 | |
148 | /* Ok. By now we've either got the offsets passed to us by the | |
149 | * caller, or we just pulled them off the bh. Lets do some | |
150 | * sanity checks to make sure they're OK. */ | |
151 | if (blkno == 0) { | |
152 | inode = ERR_PTR(-EINVAL); | |
153 | mlog_errno(PTR_ERR(inode)); | |
154 | goto bail; | |
155 | } | |
156 | ||
157 | args.fi_blkno = blkno; | |
24c19ef4 | 158 | args.fi_flags = flags; |
ccd979bd | 159 | args.fi_ino = ino_from_blkno(sb, blkno); |
5fa0613e | 160 | args.fi_sysfile_type = sysfile_type; |
ccd979bd MF |
161 | |
162 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, | |
163 | ocfs2_init_locked_inode, &args); | |
164 | /* inode was *not* in the inode cache. 2.6.x requires | |
165 | * us to do our own read_inode call and unlock it | |
166 | * afterwards. */ | |
ccd979bd MF |
167 | if (inode == NULL) { |
168 | inode = ERR_PTR(-ENOMEM); | |
169 | mlog_errno(PTR_ERR(inode)); | |
170 | goto bail; | |
171 | } | |
64f3b269 | 172 | trace_ocfs2_iget5_locked(inode->i_state); |
6218b90e | 173 | if (inode->i_state & I_NEW) { |
d56a8f32 | 174 | rc = ocfs2_read_locked_inode(inode, &args); |
6218b90e TM |
175 | unlock_new_inode(inode); |
176 | } | |
ccd979bd MF |
177 | if (is_bad_inode(inode)) { |
178 | iput(inode); | |
d56a8f32 GH |
179 | if ((flags & OCFS2_FI_FLAG_FILECHECK_CHK) || |
180 | (flags & OCFS2_FI_FLAG_FILECHECK_FIX)) | |
181 | /* Return OCFS2_FILECHECK_ERR_XXX related errno */ | |
182 | inode = ERR_PTR(rc); | |
183 | else | |
184 | inode = ERR_PTR(-ESTALE); | |
ccd979bd MF |
185 | goto bail; |
186 | } | |
187 | ||
2931cdcb DW |
188 | /* |
189 | * Set transaction id's of transactions that have to be committed | |
190 | * to finish f[data]sync. We set them to currently running transaction | |
191 | * as we cannot be sure that the inode or some of its metadata isn't | |
192 | * part of the transaction - the inode could have been reclaimed and | |
193 | * now it is reread from disk. | |
194 | */ | |
195 | if (journal) { | |
196 | transaction_t *transaction; | |
197 | tid_t tid; | |
198 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
199 | ||
200 | read_lock(&journal->j_state_lock); | |
201 | if (journal->j_running_transaction) | |
202 | transaction = journal->j_running_transaction; | |
203 | else | |
204 | transaction = journal->j_committing_transaction; | |
205 | if (transaction) | |
206 | tid = transaction->t_tid; | |
207 | else | |
208 | tid = journal->j_commit_sequence; | |
209 | read_unlock(&journal->j_state_lock); | |
210 | oi->i_sync_tid = tid; | |
211 | oi->i_datasync_tid = tid; | |
212 | } | |
213 | ||
ccd979bd MF |
214 | bail: |
215 | if (!IS_ERR(inode)) { | |
64f3b269 TM |
216 | trace_ocfs2_iget_end(inode, |
217 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
6a1bd4a5 | 218 | } |
ccd979bd MF |
219 | |
220 | return inode; | |
221 | } | |
222 | ||
223 | ||
224 | /* | |
225 | * here's how inodes get read from disk: | |
226 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR | |
227 | * found? : return the in-memory inode | |
228 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE | |
229 | */ | |
230 | ||
231 | static int ocfs2_find_actor(struct inode *inode, void *opaque) | |
232 | { | |
233 | struct ocfs2_find_inode_args *args = NULL; | |
234 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
235 | int ret = 0; | |
236 | ||
ccd979bd MF |
237 | args = opaque; |
238 | ||
239 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); | |
240 | ||
64f3b269 TM |
241 | trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno); |
242 | ||
ccd979bd MF |
243 | if (oi->ip_blkno != args->fi_blkno) |
244 | goto bail; | |
245 | ||
ccd979bd MF |
246 | ret = 1; |
247 | bail: | |
ccd979bd MF |
248 | return ret; |
249 | } | |
250 | ||
251 | /* | |
252 | * initialize the new inode, but don't do anything that would cause | |
253 | * us to sleep. | |
254 | * return 0 on success, 1 on failure | |
255 | */ | |
256 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) | |
257 | { | |
258 | struct ocfs2_find_inode_args *args = opaque; | |
cb25797d JK |
259 | static struct lock_class_key ocfs2_quota_ip_alloc_sem_key, |
260 | ocfs2_file_ip_alloc_sem_key; | |
ccd979bd | 261 | |
ccd979bd MF |
262 | inode->i_ino = args->fi_ino; |
263 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; | |
5fa0613e | 264 | if (args->fi_sysfile_type != 0) |
9902af79 | 265 | lockdep_set_class(&inode->i_rwsem, |
5fa0613e | 266 | &ocfs2_sysfile_lock_key[args->fi_sysfile_type]); |
cb25797d JK |
267 | if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE || |
268 | args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE || | |
269 | args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE || | |
270 | args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE) | |
271 | lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem, | |
272 | &ocfs2_quota_ip_alloc_sem_key); | |
273 | else | |
274 | lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem, | |
275 | &ocfs2_file_ip_alloc_sem_key); | |
ccd979bd | 276 | |
ccd979bd MF |
277 | return 0; |
278 | } | |
279 | ||
b657c95c JB |
280 | void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, |
281 | int create_ino) | |
ccd979bd MF |
282 | { |
283 | struct super_block *sb; | |
284 | struct ocfs2_super *osb; | |
53da4939 | 285 | int use_plocks = 1; |
ccd979bd | 286 | |
ccd979bd MF |
287 | sb = inode->i_sb; |
288 | osb = OCFS2_SB(sb); | |
289 | ||
53da4939 MF |
290 | if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) || |
291 | ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks()) | |
292 | use_plocks = 0; | |
293 | ||
b657c95c JB |
294 | /* |
295 | * These have all been checked by ocfs2_read_inode_block() or set | |
296 | * by ocfs2_mknod_locked(), so a failure is a code bug. | |
297 | */ | |
298 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode | |
299 | cannot create a superblock | |
300 | inode today. change if | |
301 | that is needed. */ | |
302 | BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))); | |
303 | BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation); | |
ccd979bd | 304 | |
ccd979bd | 305 | |
8110b073 MF |
306 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
307 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); | |
15b1e36b | 308 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
8110b073 | 309 | |
ccd979bd MF |
310 | inode->i_version = 1; |
311 | inode->i_generation = le32_to_cpu(fe->i_generation); | |
312 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
313 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
2c034176 EB |
314 | i_uid_write(inode, le32_to_cpu(fe->i_uid)); |
315 | i_gid_write(inode, le32_to_cpu(fe->i_gid)); | |
ccd979bd MF |
316 | |
317 | /* Fast symlinks will have i_size but no allocated clusters. */ | |
ea022dfb | 318 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) { |
ccd979bd | 319 | inode->i_blocks = 0; |
ea022dfb AV |
320 | inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops; |
321 | } else { | |
8110b073 | 322 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ea022dfb AV |
323 | inode->i_mapping->a_ops = &ocfs2_aops; |
324 | } | |
ccd979bd MF |
325 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
326 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
327 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
328 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
329 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
330 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
331 | ||
332 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) | |
333 | mlog(ML_ERROR, | |
b0697053 MF |
334 | "ip_blkno %llu != i_blkno %llu!\n", |
335 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1ca1a111 | 336 | (unsigned long long)le64_to_cpu(fe->i_blkno)); |
ccd979bd | 337 | |
bfe86848 | 338 | set_nlink(inode, ocfs2_read_links_count(fe)); |
ccd979bd | 339 | |
64f3b269 TM |
340 | trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno, |
341 | le32_to_cpu(fe->i_flags)); | |
bbbd0eb3 | 342 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) { |
24c19ef4 | 343 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; |
bbbd0eb3 JK |
344 | inode->i_flags |= S_NOQUOTA; |
345 | } | |
64f3b269 | 346 | |
ccd979bd MF |
347 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { |
348 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
ccd979bd MF |
349 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { |
350 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
1a224ad1 JK |
351 | } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) { |
352 | inode->i_flags |= S_NOQUOTA; | |
ccd979bd | 353 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { |
ccd979bd MF |
354 | /* we can't actually hit this as read_inode can't |
355 | * handle superblocks today ;-) */ | |
356 | BUG(); | |
357 | } | |
358 | ||
359 | switch (inode->i_mode & S_IFMT) { | |
360 | case S_IFREG: | |
53da4939 MF |
361 | if (use_plocks) |
362 | inode->i_fop = &ocfs2_fops; | |
363 | else | |
364 | inode->i_fop = &ocfs2_fops_no_plocks; | |
ccd979bd MF |
365 | inode->i_op = &ocfs2_file_iops; |
366 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
367 | break; | |
368 | case S_IFDIR: | |
369 | inode->i_op = &ocfs2_dir_iops; | |
53da4939 MF |
370 | if (use_plocks) |
371 | inode->i_fop = &ocfs2_dops; | |
372 | else | |
373 | inode->i_fop = &ocfs2_dops_no_plocks; | |
ccd979bd | 374 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
5e98d492 | 375 | OCFS2_I(inode)->ip_dir_lock_gen = 1; |
ccd979bd MF |
376 | break; |
377 | case S_IFLNK: | |
ea022dfb | 378 | inode->i_op = &ocfs2_symlink_inode_operations; |
21fc61c7 | 379 | inode_nohighmem(inode); |
ccd979bd MF |
380 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
381 | break; | |
382 | default: | |
383 | inode->i_op = &ocfs2_special_file_iops; | |
384 | init_special_inode(inode, inode->i_mode, | |
385 | inode->i_rdev); | |
386 | break; | |
387 | } | |
388 | ||
24c19ef4 MF |
389 | if (create_ino) { |
390 | inode->i_ino = ino_from_blkno(inode->i_sb, | |
391 | le64_to_cpu(fe->i_blkno)); | |
392 | ||
393 | /* | |
394 | * If we ever want to create system files from kernel, | |
395 | * the generation argument to | |
396 | * ocfs2_inode_lock_res_init() will have to change. | |
397 | */ | |
1ca1a111 | 398 | BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL); |
24c19ef4 | 399 | |
e63aecb6 | 400 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres, |
24c19ef4 | 401 | OCFS2_LOCK_TYPE_META, 0, inode); |
50008630 TY |
402 | |
403 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, | |
404 | OCFS2_LOCK_TYPE_OPEN, 0, inode); | |
24c19ef4 MF |
405 | } |
406 | ||
ccd979bd | 407 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, |
24c19ef4 MF |
408 | OCFS2_LOCK_TYPE_RW, inode->i_generation, |
409 | inode); | |
410 | ||
ca4d147e | 411 | ocfs2_set_inode_flags(inode); |
ca4d147e | 412 | |
13821151 TM |
413 | OCFS2_I(inode)->ip_last_used_slot = 0; |
414 | OCFS2_I(inode)->ip_last_used_group = 0; | |
e3b4a97d MF |
415 | |
416 | if (S_ISDIR(inode->i_mode)) | |
417 | ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv, | |
418 | OCFS2_RESV_FLAG_DIR); | |
ccd979bd MF |
419 | } |
420 | ||
421 | static int ocfs2_read_locked_inode(struct inode *inode, | |
422 | struct ocfs2_find_inode_args *args) | |
423 | { | |
424 | struct super_block *sb; | |
425 | struct ocfs2_super *osb; | |
426 | struct ocfs2_dinode *fe; | |
427 | struct buffer_head *bh = NULL; | |
d56a8f32 | 428 | int status, can_lock, lock_level = 0; |
24c19ef4 | 429 | u32 generation = 0; |
ccd979bd | 430 | |
ccd979bd | 431 | status = -EINVAL; |
ccd979bd MF |
432 | sb = inode->i_sb; |
433 | osb = OCFS2_SB(sb); | |
434 | ||
24c19ef4 MF |
435 | /* |
436 | * To improve performance of cold-cache inode stats, we take | |
437 | * the cluster lock here if possible. | |
438 | * | |
439 | * Generally, OCFS2 never trusts the contents of an inode | |
440 | * unless it's holding a cluster lock, so taking it here isn't | |
441 | * a correctness issue as much as it is a performance | |
442 | * improvement. | |
443 | * | |
444 | * There are three times when taking the lock is not a good idea: | |
445 | * | |
446 | * 1) During startup, before we have initialized the DLM. | |
447 | * | |
448 | * 2) If we are reading certain system files which never get | |
449 | * cluster locks (local alloc, truncate log). | |
450 | * | |
451 | * 3) If the process doing the iget() is responsible for | |
452 | * orphan dir recovery. We're holding the orphan dir lock and | |
453 | * can get into a deadlock with another process on another | |
454 | * node in ->delete_inode(). | |
455 | * | |
456 | * #1 and #2 can be simply solved by never taking the lock | |
457 | * here for system files (which are the only type we read | |
458 | * during mount). It's a heavier approach, but our main | |
b595076a | 459 | * concern is user-accessible files anyway. |
24c19ef4 MF |
460 | * |
461 | * #3 works itself out because we'll eventually take the | |
462 | * cluster lock before trusting anything anyway. | |
463 | */ | |
464 | can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
50008630 | 465 | && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) |
c271c5c2 | 466 | && !ocfs2_mount_local(osb); |
24c19ef4 | 467 | |
64f3b269 TM |
468 | trace_ocfs2_read_locked_inode( |
469 | (unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock); | |
470 | ||
24c19ef4 MF |
471 | /* |
472 | * To maintain backwards compatibility with older versions of | |
473 | * ocfs2-tools, we still store the generation value for system | |
474 | * files. The only ones that actually matter to userspace are | |
475 | * the journals, but it's easier and inexpensive to just flag | |
476 | * all system files similarly. | |
477 | */ | |
478 | if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
479 | generation = osb->fs_generation; | |
480 | ||
e63aecb6 | 481 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres, |
24c19ef4 MF |
482 | OCFS2_LOCK_TYPE_META, |
483 | generation, inode); | |
484 | ||
50008630 TY |
485 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, |
486 | OCFS2_LOCK_TYPE_OPEN, | |
487 | 0, inode); | |
488 | ||
24c19ef4 | 489 | if (can_lock) { |
50008630 TY |
490 | status = ocfs2_open_lock(inode); |
491 | if (status) { | |
492 | make_bad_inode(inode); | |
493 | mlog_errno(status); | |
494 | return status; | |
495 | } | |
d56a8f32 | 496 | status = ocfs2_inode_lock(inode, NULL, lock_level); |
24c19ef4 MF |
497 | if (status) { |
498 | make_bad_inode(inode); | |
499 | mlog_errno(status); | |
500 | return status; | |
501 | } | |
ccd979bd MF |
502 | } |
503 | ||
50008630 TY |
504 | if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) { |
505 | status = ocfs2_try_open_lock(inode, 0); | |
506 | if (status) { | |
2bd63216 | 507 | make_bad_inode(inode); |
50008630 TY |
508 | return status; |
509 | } | |
510 | } | |
511 | ||
b657c95c | 512 | if (can_lock) { |
d56a8f32 GH |
513 | if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK) |
514 | status = ocfs2_filecheck_read_inode_block_full(inode, | |
515 | &bh, OCFS2_BH_IGNORE_CACHE, 0); | |
516 | else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX) | |
517 | status = ocfs2_filecheck_read_inode_block_full(inode, | |
518 | &bh, OCFS2_BH_IGNORE_CACHE, 1); | |
519 | else | |
520 | status = ocfs2_read_inode_block_full(inode, | |
521 | &bh, OCFS2_BH_IGNORE_CACHE); | |
b657c95c | 522 | } else { |
da1e9098 | 523 | status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh); |
f5ce5a08 SM |
524 | /* |
525 | * If buffer is in jbd, then its checksum may not have been | |
526 | * computed as yet. | |
527 | */ | |
d56a8f32 GH |
528 | if (!status && !buffer_jbd(bh)) { |
529 | if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_CHK) | |
530 | status = ocfs2_filecheck_validate_inode_block( | |
531 | osb->sb, bh); | |
532 | else if (args->fi_flags & OCFS2_FI_FLAG_FILECHECK_FIX) | |
533 | status = ocfs2_filecheck_repair_inode_block( | |
534 | osb->sb, bh); | |
535 | else | |
536 | status = ocfs2_validate_inode_block( | |
537 | osb->sb, bh); | |
538 | } | |
b657c95c | 539 | } |
ccd979bd MF |
540 | if (status < 0) { |
541 | mlog_errno(status); | |
ccd979bd MF |
542 | goto bail; |
543 | } | |
544 | ||
24c19ef4 | 545 | status = -EINVAL; |
ccd979bd | 546 | fe = (struct ocfs2_dinode *) bh->b_data; |
ccd979bd | 547 | |
24c19ef4 MF |
548 | /* |
549 | * This is a code bug. Right now the caller needs to | |
550 | * understand whether it is asking for a system file inode or | |
551 | * not so the proper lock names can be built. | |
552 | */ | |
553 | mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) != | |
554 | !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE), | |
555 | "Inode %llu: system file state is ambigous\n", | |
556 | (unsigned long long)args->fi_blkno); | |
ccd979bd MF |
557 | |
558 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || | |
559 | S_ISBLK(le16_to_cpu(fe->i_mode))) | |
b657c95c | 560 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); |
ccd979bd | 561 | |
b657c95c | 562 | ocfs2_populate_inode(inode, fe, 0); |
ccd979bd MF |
563 | |
564 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); | |
565 | ||
d56a8f32 GH |
566 | if (buffer_dirty(bh) && !buffer_jbd(bh)) { |
567 | if (can_lock) { | |
568 | ocfs2_inode_unlock(inode, lock_level); | |
569 | lock_level = 1; | |
570 | ocfs2_inode_lock(inode, NULL, lock_level); | |
571 | } | |
572 | status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); | |
573 | if (status < 0) { | |
574 | mlog_errno(status); | |
575 | goto bail; | |
576 | } | |
577 | } | |
578 | ||
ccd979bd MF |
579 | status = 0; |
580 | ||
581 | bail: | |
24c19ef4 | 582 | if (can_lock) |
d56a8f32 | 583 | ocfs2_inode_unlock(inode, lock_level); |
24c19ef4 MF |
584 | |
585 | if (status < 0) | |
586 | make_bad_inode(inode); | |
587 | ||
4a635a11 | 588 | brelse(bh); |
ccd979bd | 589 | |
ccd979bd MF |
590 | return status; |
591 | } | |
592 | ||
593 | void ocfs2_sync_blockdev(struct super_block *sb) | |
594 | { | |
595 | sync_blockdev(sb->s_bdev); | |
596 | } | |
597 | ||
598 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
599 | struct inode *inode, | |
600 | struct buffer_head *fe_bh) | |
601 | { | |
602 | int status = 0; | |
ccd979bd | 603 | struct ocfs2_dinode *fe; |
60b11392 | 604 | handle_t *handle = NULL; |
ccd979bd | 605 | |
ccd979bd MF |
606 | fe = (struct ocfs2_dinode *) fe_bh->b_data; |
607 | ||
1afc32b9 MF |
608 | /* |
609 | * This check will also skip truncate of inodes with inline | |
610 | * data and fast symlinks. | |
611 | */ | |
3a0782d0 | 612 | if (fe->i_clusters) { |
2b4e30fb JB |
613 | if (ocfs2_should_order_data(inode)) |
614 | ocfs2_begin_ordered_truncate(inode, 0); | |
615 | ||
60b11392 MF |
616 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
617 | if (IS_ERR(handle)) { | |
618 | status = PTR_ERR(handle); | |
0350cb07 | 619 | handle = NULL; |
60b11392 MF |
620 | mlog_errno(status); |
621 | goto out; | |
622 | } | |
623 | ||
0cf2f763 JB |
624 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), |
625 | fe_bh, | |
13723d00 | 626 | OCFS2_JOURNAL_ACCESS_WRITE); |
60b11392 MF |
627 | if (status < 0) { |
628 | mlog_errno(status); | |
629 | goto out; | |
630 | } | |
631 | ||
632 | i_size_write(inode, 0); | |
633 | ||
634 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
635 | if (status < 0) { | |
636 | mlog_errno(status); | |
637 | goto out; | |
638 | } | |
639 | ||
640 | ocfs2_commit_trans(osb, handle); | |
641 | handle = NULL; | |
642 | ||
78f94673 | 643 | status = ocfs2_commit_truncate(osb, inode, fe_bh); |
3a0782d0 MF |
644 | if (status < 0) { |
645 | mlog_errno(status); | |
646 | goto out; | |
647 | } | |
ccd979bd | 648 | } |
ccd979bd | 649 | |
60b11392 MF |
650 | out: |
651 | if (handle) | |
652 | ocfs2_commit_trans(osb, handle); | |
ccd979bd MF |
653 | return status; |
654 | } | |
655 | ||
656 | static int ocfs2_remove_inode(struct inode *inode, | |
657 | struct buffer_head *di_bh, | |
658 | struct inode *orphan_dir_inode, | |
659 | struct buffer_head *orphan_dir_bh) | |
660 | { | |
661 | int status; | |
662 | struct inode *inode_alloc_inode = NULL; | |
663 | struct buffer_head *inode_alloc_bh = NULL; | |
1fabe148 | 664 | handle_t *handle; |
ccd979bd MF |
665 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
666 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
667 | ||
668 | inode_alloc_inode = | |
669 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
670 | le16_to_cpu(di->i_suballoc_slot)); | |
671 | if (!inode_alloc_inode) { | |
62f8b1f0 | 672 | status = -ENOENT; |
ccd979bd MF |
673 | mlog_errno(status); |
674 | goto bail; | |
675 | } | |
676 | ||
5955102c | 677 | inode_lock(inode_alloc_inode); |
e63aecb6 | 678 | status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1); |
ccd979bd | 679 | if (status < 0) { |
5955102c | 680 | inode_unlock(inode_alloc_inode); |
ccd979bd MF |
681 | |
682 | mlog_errno(status); | |
683 | goto bail; | |
684 | } | |
685 | ||
a90714c1 | 686 | handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS + |
9b7895ef | 687 | ocfs2_quota_trans_credits(inode->i_sb)); |
ccd979bd MF |
688 | if (IS_ERR(handle)) { |
689 | status = PTR_ERR(handle); | |
690 | mlog_errno(status); | |
691 | goto bail_unlock; | |
692 | } | |
693 | ||
d4cd1871 LD |
694 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { |
695 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, | |
06ee5c75 | 696 | orphan_dir_bh, false); |
d4cd1871 LD |
697 | if (status < 0) { |
698 | mlog_errno(status); | |
699 | goto bail_commit; | |
700 | } | |
ccd979bd MF |
701 | } |
702 | ||
703 | /* set the inodes dtime */ | |
0cf2f763 | 704 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 705 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
706 | if (status < 0) { |
707 | mlog_errno(status); | |
708 | goto bail_commit; | |
709 | } | |
710 | ||
711 | di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec); | |
4092d49f | 712 | di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); |
ec20cec7 | 713 | ocfs2_journal_dirty(handle, di_bh); |
ccd979bd | 714 | |
8cb471e8 | 715 | ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh); |
63936dda | 716 | dquot_free_inode(inode); |
ccd979bd MF |
717 | |
718 | status = ocfs2_free_dinode(handle, inode_alloc_inode, | |
719 | inode_alloc_bh, di); | |
720 | if (status < 0) | |
721 | mlog_errno(status); | |
722 | ||
723 | bail_commit: | |
02dc1af4 | 724 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 725 | bail_unlock: |
e63aecb6 | 726 | ocfs2_inode_unlock(inode_alloc_inode, 1); |
5955102c | 727 | inode_unlock(inode_alloc_inode); |
ccd979bd MF |
728 | brelse(inode_alloc_bh); |
729 | bail: | |
730 | iput(inode_alloc_inode); | |
731 | ||
732 | return status; | |
733 | } | |
734 | ||
2bd63216 | 735 | /* |
b4df6ed8 MF |
736 | * Serialize with orphan dir recovery. If the process doing |
737 | * recovery on this orphan dir does an iget() with the dir | |
738 | * i_mutex held, we'll deadlock here. Instead we detect this | |
739 | * and exit early - recovery will wipe this inode for us. | |
740 | */ | |
741 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, | |
742 | int slot) | |
743 | { | |
744 | int ret = 0; | |
745 | ||
746 | spin_lock(&osb->osb_lock); | |
747 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { | |
b4df6ed8 MF |
748 | ret = -EDEADLK; |
749 | goto out; | |
750 | } | |
751 | /* This signals to the orphan recovery process that it should | |
752 | * wait for us to handle the wipe. */ | |
753 | osb->osb_orphan_wipes[slot]++; | |
754 | out: | |
755 | spin_unlock(&osb->osb_lock); | |
64f3b269 | 756 | trace_ocfs2_check_orphan_recovery_state(slot, ret); |
b4df6ed8 MF |
757 | return ret; |
758 | } | |
759 | ||
760 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, | |
761 | int slot) | |
762 | { | |
763 | spin_lock(&osb->osb_lock); | |
764 | osb->osb_orphan_wipes[slot]--; | |
765 | spin_unlock(&osb->osb_lock); | |
766 | ||
767 | wake_up(&osb->osb_wipe_event); | |
768 | } | |
769 | ||
ccd979bd MF |
770 | static int ocfs2_wipe_inode(struct inode *inode, |
771 | struct buffer_head *di_bh) | |
772 | { | |
d577632e | 773 | int status, orphaned_slot = -1; |
ccd979bd MF |
774 | struct inode *orphan_dir_inode = NULL; |
775 | struct buffer_head *orphan_dir_bh = NULL; | |
776 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
d4cd1871 | 777 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; |
ccd979bd | 778 | |
d4cd1871 LD |
779 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { |
780 | orphaned_slot = le16_to_cpu(di->i_orphaned_slot); | |
b4df6ed8 | 781 | |
d4cd1871 LD |
782 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); |
783 | if (status) | |
784 | return status; | |
b4df6ed8 | 785 | |
d4cd1871 LD |
786 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
787 | ORPHAN_DIR_SYSTEM_INODE, | |
788 | orphaned_slot); | |
789 | if (!orphan_dir_inode) { | |
62f8b1f0 | 790 | status = -ENOENT; |
d4cd1871 LD |
791 | mlog_errno(status); |
792 | goto bail; | |
793 | } | |
ccd979bd | 794 | |
d4cd1871 LD |
795 | /* Lock the orphan dir. The lock will be held for the entire |
796 | * delete_inode operation. We do this now to avoid races with | |
797 | * recovery completion on other nodes. */ | |
5955102c | 798 | inode_lock(orphan_dir_inode); |
d4cd1871 LD |
799 | status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1); |
800 | if (status < 0) { | |
5955102c | 801 | inode_unlock(orphan_dir_inode); |
ccd979bd | 802 | |
d4cd1871 LD |
803 | mlog_errno(status); |
804 | goto bail; | |
805 | } | |
ccd979bd MF |
806 | } |
807 | ||
808 | /* we do this while holding the orphan dir lock because we | |
34d024f8 MF |
809 | * don't want recovery being run from another node to try an |
810 | * inode delete underneath us -- this will result in two nodes | |
ccd979bd MF |
811 | * truncating the same file! */ |
812 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); | |
813 | if (status < 0) { | |
814 | mlog_errno(status); | |
815 | goto bail_unlock_dir; | |
816 | } | |
817 | ||
9b7895ef MF |
818 | /* Remove any dir index tree */ |
819 | if (S_ISDIR(inode->i_mode)) { | |
820 | status = ocfs2_dx_dir_truncate(inode, di_bh); | |
821 | if (status) { | |
822 | mlog_errno(status); | |
823 | goto bail_unlock_dir; | |
824 | } | |
825 | } | |
826 | ||
cf1d6c76 TY |
827 | /*Free extended attribute resources associated with this inode.*/ |
828 | status = ocfs2_xattr_remove(inode, di_bh); | |
829 | if (status < 0) { | |
830 | mlog_errno(status); | |
831 | goto bail_unlock_dir; | |
832 | } | |
833 | ||
8b2c0dba TM |
834 | status = ocfs2_remove_refcount_tree(inode, di_bh); |
835 | if (status < 0) { | |
836 | mlog_errno(status); | |
837 | goto bail_unlock_dir; | |
838 | } | |
839 | ||
ccd979bd MF |
840 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, |
841 | orphan_dir_bh); | |
842 | if (status < 0) | |
843 | mlog_errno(status); | |
844 | ||
845 | bail_unlock_dir: | |
d4cd1871 LD |
846 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR) |
847 | return status; | |
848 | ||
e63aecb6 | 849 | ocfs2_inode_unlock(orphan_dir_inode, 1); |
5955102c | 850 | inode_unlock(orphan_dir_inode); |
ccd979bd MF |
851 | brelse(orphan_dir_bh); |
852 | bail: | |
853 | iput(orphan_dir_inode); | |
b4df6ed8 | 854 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
ccd979bd MF |
855 | |
856 | return status; | |
857 | } | |
858 | ||
859 | /* There is a series of simple checks that should be done before a | |
34d024f8 | 860 | * trylock is even considered. Encapsulate those in this function. */ |
ccd979bd MF |
861 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) |
862 | { | |
863 | int ret = 0; | |
864 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
865 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
866 | ||
64f3b269 TM |
867 | trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task, |
868 | (unsigned long long)oi->ip_blkno, | |
869 | oi->ip_flags); | |
870 | ||
ccd979bd MF |
871 | /* We shouldn't be getting here for the root directory |
872 | * inode.. */ | |
873 | if (inode == osb->root_inode) { | |
874 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); | |
875 | goto bail; | |
876 | } | |
877 | ||
bd62ad7a JK |
878 | /* |
879 | * If we're coming from downconvert_thread we can't go into our own | |
880 | * voting [hello, deadlock city!] so we cannot delete the inode. But | |
881 | * since we dropped last inode ref when downconverting dentry lock, | |
882 | * we cannot have the file open and thus the node doing unlink will | |
883 | * take care of deleting the inode. | |
884 | */ | |
64f3b269 | 885 | if (current == osb->dc_task) |
ccd979bd | 886 | goto bail; |
ccd979bd MF |
887 | |
888 | spin_lock(&oi->ip_lock); | |
889 | /* OCFS2 *never* deletes system files. This should technically | |
890 | * never get here as system file inodes should always have a | |
891 | * positive link count. */ | |
892 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { | |
b0697053 MF |
893 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
894 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
895 | goto bail_unlock; |
896 | } | |
897 | ||
ccd979bd MF |
898 | ret = 1; |
899 | bail_unlock: | |
900 | spin_unlock(&oi->ip_lock); | |
901 | bail: | |
902 | return ret; | |
903 | } | |
904 | ||
905 | /* Query the cluster to determine whether we should wipe an inode from | |
906 | * disk or not. | |
907 | * | |
908 | * Requires the inode to have the cluster lock. */ | |
909 | static int ocfs2_query_inode_wipe(struct inode *inode, | |
910 | struct buffer_head *di_bh, | |
911 | int *wipe) | |
912 | { | |
64f3b269 | 913 | int status = 0, reason = 0; |
ccd979bd MF |
914 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
915 | struct ocfs2_dinode *di; | |
916 | ||
917 | *wipe = 0; | |
918 | ||
64f3b269 TM |
919 | trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno, |
920 | inode->i_nlink); | |
921 | ||
ccd979bd MF |
922 | /* While we were waiting for the cluster lock in |
923 | * ocfs2_delete_inode, another node might have asked to delete | |
924 | * the inode. Recheck our flags to catch this. */ | |
925 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
64f3b269 | 926 | reason = 1; |
ccd979bd MF |
927 | goto bail; |
928 | } | |
929 | ||
930 | /* Now that we have an up to date inode, we can double check | |
931 | * the link count. */ | |
64f3b269 | 932 | if (inode->i_nlink) |
ccd979bd | 933 | goto bail; |
ccd979bd MF |
934 | |
935 | /* Do some basic inode verification... */ | |
936 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
d4cd1871 LD |
937 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) && |
938 | !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) { | |
b54c2ca4 TY |
939 | /* |
940 | * Inodes in the orphan dir must have ORPHANED_FL. The only | |
941 | * inodes that come back out of the orphan dir are reflink | |
942 | * targets. A reflink target may be moved out of the orphan | |
943 | * dir between the time we scan the directory and the time we | |
944 | * process it. This would lead to HAS_REFCOUNT_FL being set but | |
945 | * ORPHANED_FL not. | |
946 | */ | |
947 | if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) { | |
64f3b269 | 948 | reason = 2; |
b54c2ca4 TY |
949 | goto bail; |
950 | } | |
951 | ||
ccd979bd MF |
952 | /* for lack of a better error? */ |
953 | status = -EEXIST; | |
954 | mlog(ML_ERROR, | |
b0697053 | 955 | "Inode %llu (on-disk %llu) not orphaned! " |
ccd979bd | 956 | "Disk flags 0x%x, inode flags 0x%x\n", |
b0697053 | 957 | (unsigned long long)oi->ip_blkno, |
1ca1a111 MF |
958 | (unsigned long long)le64_to_cpu(di->i_blkno), |
959 | le32_to_cpu(di->i_flags), oi->ip_flags); | |
ccd979bd MF |
960 | goto bail; |
961 | } | |
962 | ||
963 | /* has someone already deleted us?! baaad... */ | |
964 | if (di->i_dtime) { | |
965 | status = -EEXIST; | |
966 | mlog_errno(status); | |
967 | goto bail; | |
968 | } | |
969 | ||
6f16bf65 MF |
970 | /* |
971 | * This is how ocfs2 determines whether an inode is still live | |
972 | * within the cluster. Every node takes a shared read lock on | |
973 | * the inode open lock in ocfs2_read_locked_inode(). When we | |
974 | * get to ->delete_inode(), each node tries to convert it's | |
975 | * lock to an exclusive. Trylocks are serialized by the inode | |
25985edc | 976 | * meta data lock. If the upconvert succeeds, we know the inode |
6f16bf65 MF |
977 | * is no longer live and can be deleted. |
978 | * | |
979 | * Though we call this with the meta data lock held, the | |
980 | * trylock keeps us from ABBA deadlock. | |
981 | */ | |
982 | status = ocfs2_try_open_lock(inode, 1); | |
50008630 | 983 | if (status == -EAGAIN) { |
ccd979bd | 984 | status = 0; |
64f3b269 | 985 | reason = 3; |
ccd979bd MF |
986 | goto bail; |
987 | } | |
988 | if (status < 0) { | |
989 | mlog_errno(status); | |
990 | goto bail; | |
991 | } | |
992 | ||
50008630 | 993 | *wipe = 1; |
64f3b269 | 994 | trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot)); |
ccd979bd MF |
995 | |
996 | bail: | |
64f3b269 | 997 | trace_ocfs2_query_inode_wipe_end(status, reason); |
ccd979bd MF |
998 | return status; |
999 | } | |
1000 | ||
1001 | /* Support function for ocfs2_delete_inode. Will help us keep the | |
1002 | * inode data in a consistent state for clear_inode. Always truncates | |
1003 | * pages, optionally sync's them first. */ | |
1004 | static void ocfs2_cleanup_delete_inode(struct inode *inode, | |
1005 | int sync_data) | |
1006 | { | |
64f3b269 TM |
1007 | trace_ocfs2_cleanup_delete_inode( |
1008 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); | |
ccd979bd | 1009 | if (sync_data) |
249ec93c | 1010 | filemap_write_and_wait(inode->i_mapping); |
91b0abe3 | 1011 | truncate_inode_pages_final(&inode->i_data); |
ccd979bd MF |
1012 | } |
1013 | ||
066d92dc | 1014 | static void ocfs2_delete_inode(struct inode *inode) |
ccd979bd MF |
1015 | { |
1016 | int wipe, status; | |
e4b963f1 | 1017 | sigset_t oldset; |
ccd979bd | 1018 | struct buffer_head *di_bh = NULL; |
ad694821 | 1019 | struct ocfs2_dinode *di = NULL; |
ccd979bd | 1020 | |
64f3b269 TM |
1021 | trace_ocfs2_delete_inode(inode->i_ino, |
1022 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1023 | is_bad_inode(inode)); | |
ccd979bd | 1024 | |
a90714c1 JK |
1025 | /* When we fail in read_inode() we mark inode as bad. The second test |
1026 | * catches the case when inode allocation fails before allocating | |
1027 | * a block for inode. */ | |
64f3b269 | 1028 | if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) |
ccd979bd | 1029 | goto bail; |
ccd979bd MF |
1030 | |
1031 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
1032 | /* It's probably not necessary to truncate_inode_pages | |
1033 | * here but we do it for safety anyway (it will most | |
1034 | * likely be a no-op anyway) */ | |
1035 | ocfs2_cleanup_delete_inode(inode, 0); | |
1036 | goto bail; | |
1037 | } | |
1038 | ||
bd62ad7a JK |
1039 | dquot_initialize(inode); |
1040 | ||
ccd979bd MF |
1041 | /* We want to block signals in delete_inode as the lock and |
1042 | * messaging paths may return us -ERESTARTSYS. Which would | |
1043 | * cause us to exit early, resulting in inodes being orphaned | |
1044 | * forever. */ | |
e4b963f1 | 1045 | ocfs2_block_signals(&oldset); |
ccd979bd | 1046 | |
6ca497a8 | 1047 | /* |
1048 | * Synchronize us against ocfs2_get_dentry. We take this in | |
1049 | * shared mode so that all nodes can still concurrently | |
1050 | * process deletes. | |
1051 | */ | |
1052 | status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0); | |
1053 | if (status < 0) { | |
1054 | mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status); | |
1055 | ocfs2_cleanup_delete_inode(inode, 0); | |
1056 | goto bail_unblock; | |
1057 | } | |
ccd979bd MF |
1058 | /* Lock down the inode. This gives us an up to date view of |
1059 | * it's metadata (for verification), and allows us to | |
34d024f8 | 1060 | * serialize delete_inode on multiple nodes. |
ccd979bd MF |
1061 | * |
1062 | * Even though we might be doing a truncate, we don't take the | |
1063 | * allocation lock here as it won't be needed - nobody will | |
1064 | * have the file open. | |
1065 | */ | |
e63aecb6 | 1066 | status = ocfs2_inode_lock(inode, &di_bh, 1); |
ccd979bd MF |
1067 | if (status < 0) { |
1068 | if (status != -ENOENT) | |
1069 | mlog_errno(status); | |
1070 | ocfs2_cleanup_delete_inode(inode, 0); | |
6ca497a8 | 1071 | goto bail_unlock_nfs_sync; |
ccd979bd MF |
1072 | } |
1073 | ||
ad694821 JQ |
1074 | di = (struct ocfs2_dinode *)di_bh->b_data; |
1075 | /* Skip inode deletion and wait for dio orphan entry recovered | |
1076 | * first */ | |
1077 | if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) { | |
1078 | ocfs2_cleanup_delete_inode(inode, 0); | |
1079 | goto bail_unlock_inode; | |
1080 | } | |
1081 | ||
ccd979bd MF |
1082 | /* Query the cluster. This will be the final decision made |
1083 | * before we go ahead and wipe the inode. */ | |
1084 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); | |
1085 | if (!wipe || status < 0) { | |
34d024f8 | 1086 | /* Error and remote inode busy both mean we won't be |
ccd979bd MF |
1087 | * removing the inode, so they take almost the same |
1088 | * path. */ | |
1089 | if (status < 0) | |
1090 | mlog_errno(status); | |
1091 | ||
34d024f8 MF |
1092 | /* Someone in the cluster has disallowed a wipe of |
1093 | * this inode, or it was never completely | |
1094 | * orphaned. Write out the pages and exit now. */ | |
ccd979bd MF |
1095 | ocfs2_cleanup_delete_inode(inode, 1); |
1096 | goto bail_unlock_inode; | |
1097 | } | |
1098 | ||
1099 | ocfs2_cleanup_delete_inode(inode, 0); | |
1100 | ||
1101 | status = ocfs2_wipe_inode(inode, di_bh); | |
1102 | if (status < 0) { | |
b4df6ed8 MF |
1103 | if (status != -EDEADLK) |
1104 | mlog_errno(status); | |
ccd979bd MF |
1105 | goto bail_unlock_inode; |
1106 | } | |
1107 | ||
24c19ef4 MF |
1108 | /* |
1109 | * Mark the inode as successfully deleted. | |
1110 | * | |
1111 | * This is important for ocfs2_clear_inode() as it will check | |
1112 | * this flag and skip any checkpointing work | |
1113 | * | |
1114 | * ocfs2_stuff_meta_lvb() also uses this flag to invalidate | |
1115 | * the LVB for other nodes. | |
1116 | */ | |
ccd979bd MF |
1117 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; |
1118 | ||
1119 | bail_unlock_inode: | |
e63aecb6 | 1120 | ocfs2_inode_unlock(inode, 1); |
ccd979bd | 1121 | brelse(di_bh); |
6ca497a8 | 1122 | |
1123 | bail_unlock_nfs_sync: | |
1124 | ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0); | |
1125 | ||
ccd979bd | 1126 | bail_unblock: |
e4b963f1 | 1127 | ocfs2_unblock_signals(&oldset); |
ccd979bd | 1128 | bail: |
c1e8d35e | 1129 | return; |
ccd979bd MF |
1130 | } |
1131 | ||
066d92dc | 1132 | static void ocfs2_clear_inode(struct inode *inode) |
ccd979bd MF |
1133 | { |
1134 | int status; | |
1135 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
84d86f83 | 1136 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd | 1137 | |
dbd5768f | 1138 | clear_inode(inode); |
64f3b269 TM |
1139 | trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno, |
1140 | inode->i_nlink); | |
ccd979bd MF |
1141 | |
1142 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, | |
1143 | "Inode=%lu\n", inode->i_ino); | |
1144 | ||
9f754758 | 1145 | dquot_drop(inode); |
257ba15c | 1146 | |
34d024f8 MF |
1147 | /* To preven remote deletes we hold open lock before, now it |
1148 | * is time to unlock PR and EX open locks. */ | |
50008630 TY |
1149 | ocfs2_open_unlock(inode); |
1150 | ||
ccd979bd | 1151 | /* Do these before all the other work so that we don't bounce |
34d024f8 | 1152 | * the downconvert thread while waiting to destroy the locks. */ |
84d86f83 JK |
1153 | ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres); |
1154 | ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres); | |
1155 | ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres); | |
ccd979bd | 1156 | |
4fe370af MF |
1157 | ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap, |
1158 | &oi->ip_la_data_resv); | |
1159 | ocfs2_resv_init_once(&oi->ip_la_data_resv); | |
1160 | ||
ccd979bd MF |
1161 | /* We very well may get a clear_inode before all an inodes |
1162 | * metadata has hit disk. Of course, we can't drop any cluster | |
1163 | * locks until the journal has finished with it. The only | |
1164 | * exception here are successfully wiped inodes - their | |
1165 | * metadata can now be considered to be part of the system | |
1166 | * inodes from which it came. */ | |
1167 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) | |
1168 | ocfs2_checkpoint_inode(inode); | |
1169 | ||
1170 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), | |
b0697053 MF |
1171 | "Clear inode of %llu, inode has io markers\n", |
1172 | (unsigned long long)oi->ip_blkno); | |
4506cfb6 RD |
1173 | mlog_bug_on_msg(!list_empty(&oi->ip_unwritten_list), |
1174 | "Clear inode of %llu, inode has unwritten extents\n", | |
1175 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1176 | |
83418978 MF |
1177 | ocfs2_extent_map_trunc(inode, 0); |
1178 | ||
ccd979bd MF |
1179 | status = ocfs2_drop_inode_locks(inode); |
1180 | if (status < 0) | |
1181 | mlog_errno(status); | |
1182 | ||
1183 | ocfs2_lock_res_free(&oi->ip_rw_lockres); | |
e63aecb6 | 1184 | ocfs2_lock_res_free(&oi->ip_inode_lockres); |
50008630 | 1185 | ocfs2_lock_res_free(&oi->ip_open_lockres); |
ccd979bd | 1186 | |
66fb345d | 1187 | ocfs2_metadata_cache_exit(INODE_CACHE(inode)); |
ccd979bd | 1188 | |
8cb471e8 | 1189 | mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached, |
b0697053 | 1190 | "Clear inode of %llu, inode has %u cache items\n", |
8cb471e8 JB |
1191 | (unsigned long long)oi->ip_blkno, |
1192 | INODE_CACHE(inode)->ci_num_cached); | |
ccd979bd | 1193 | |
8cb471e8 | 1194 | mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE), |
b0697053 MF |
1195 | "Clear inode of %llu, inode has a bad flag\n", |
1196 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1197 | |
1198 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), | |
b0697053 MF |
1199 | "Clear inode of %llu, inode is locked\n", |
1200 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1201 | |
251b6ecc | 1202 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
b0697053 MF |
1203 | "Clear inode of %llu, io_mutex is locked\n", |
1204 | (unsigned long long)oi->ip_blkno); | |
251b6ecc | 1205 | mutex_unlock(&oi->ip_io_mutex); |
ccd979bd MF |
1206 | |
1207 | /* | |
1208 | * down_trylock() returns 0, down_write_trylock() returns 1 | |
1209 | * kernel 1, world 0 | |
1210 | */ | |
1211 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), | |
b0697053 MF |
1212 | "Clear inode of %llu, alloc_sem is locked\n", |
1213 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1214 | up_write(&oi->ip_alloc_sem); |
1215 | ||
1216 | mlog_bug_on_msg(oi->ip_open_count, | |
b0697053 MF |
1217 | "Clear inode of %llu has open count %d\n", |
1218 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); | |
ccd979bd MF |
1219 | |
1220 | /* Clear all other flags. */ | |
47460d65 | 1221 | oi->ip_flags = 0; |
ccd979bd MF |
1222 | oi->ip_dir_start_lookup = 0; |
1223 | oi->ip_blkno = 0ULL; | |
ae0dff68 SM |
1224 | |
1225 | /* | |
1226 | * ip_jinode is used to track txns against this inode. We ensure that | |
1227 | * the journal is flushed before journal shutdown. Thus it is safe to | |
1228 | * have inodes get cleaned up after journal shutdown. | |
1229 | */ | |
2b4e30fb JB |
1230 | jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal, |
1231 | &oi->ip_jinode); | |
ccd979bd MF |
1232 | } |
1233 | ||
066d92dc AV |
1234 | void ocfs2_evict_inode(struct inode *inode) |
1235 | { | |
1236 | if (!inode->i_nlink || | |
1237 | (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) { | |
1238 | ocfs2_delete_inode(inode); | |
1239 | } else { | |
91b0abe3 | 1240 | truncate_inode_pages_final(&inode->i_data); |
066d92dc AV |
1241 | } |
1242 | ocfs2_clear_inode(inode); | |
1243 | } | |
1244 | ||
ccd979bd MF |
1245 | /* Called under inode_lock, with no more references on the |
1246 | * struct inode, so it's safe here to check the flags field | |
1247 | * and to manipulate i_nlink without any other locks. */ | |
45321ac5 | 1248 | int ocfs2_drop_inode(struct inode *inode) |
ccd979bd MF |
1249 | { |
1250 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1251 | ||
64f3b269 TM |
1252 | trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno, |
1253 | inode->i_nlink, oi->ip_flags); | |
ccd979bd | 1254 | |
513e2dae X |
1255 | assert_spin_locked(&inode->i_lock); |
1256 | inode->i_state |= I_WILL_FREE; | |
1257 | spin_unlock(&inode->i_lock); | |
1258 | write_inode_now(inode, 1); | |
1259 | spin_lock(&inode->i_lock); | |
1260 | WARN_ON(inode->i_state & I_NEW); | |
1261 | inode->i_state &= ~I_WILL_FREE; | |
ccd979bd | 1262 | |
513e2dae | 1263 | return 1; |
ccd979bd MF |
1264 | } |
1265 | ||
ccd979bd MF |
1266 | /* |
1267 | * This is called from our getattr. | |
1268 | */ | |
1269 | int ocfs2_inode_revalidate(struct dentry *dentry) | |
1270 | { | |
2b0143b5 | 1271 | struct inode *inode = d_inode(dentry); |
ccd979bd MF |
1272 | int status = 0; |
1273 | ||
64f3b269 TM |
1274 | trace_ocfs2_inode_revalidate(inode, |
1275 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL, | |
1276 | inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0); | |
ccd979bd MF |
1277 | |
1278 | if (!inode) { | |
ccd979bd MF |
1279 | status = -ENOENT; |
1280 | goto bail; | |
1281 | } | |
1282 | ||
1283 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1284 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
1285 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
ccd979bd MF |
1286 | status = -ENOENT; |
1287 | goto bail; | |
1288 | } | |
1289 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1290 | ||
e63aecb6 | 1291 | /* Let ocfs2_inode_lock do the work of updating our struct |
ccd979bd | 1292 | * inode for us. */ |
e63aecb6 | 1293 | status = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
1294 | if (status < 0) { |
1295 | if (status != -ENOENT) | |
1296 | mlog_errno(status); | |
1297 | goto bail; | |
1298 | } | |
e63aecb6 | 1299 | ocfs2_inode_unlock(inode, 0); |
ccd979bd | 1300 | bail: |
ccd979bd MF |
1301 | return status; |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Updates a disk inode from a | |
1306 | * struct inode. | |
1307 | * Only takes ip_lock. | |
1308 | */ | |
1fabe148 | 1309 | int ocfs2_mark_inode_dirty(handle_t *handle, |
ccd979bd MF |
1310 | struct inode *inode, |
1311 | struct buffer_head *bh) | |
1312 | { | |
1313 | int status; | |
1314 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
1315 | ||
64f3b269 | 1316 | trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd | 1317 | |
0cf2f763 | 1318 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 1319 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1320 | if (status < 0) { |
1321 | mlog_errno(status); | |
1322 | goto leave; | |
1323 | } | |
1324 | ||
1325 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1326 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
6e4b0d56 | 1327 | ocfs2_get_inode_flags(OCFS2_I(inode)); |
ca4d147e | 1328 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
15b1e36b | 1329 | fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features); |
ccd979bd MF |
1330 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1331 | ||
1332 | fe->i_size = cpu_to_le64(i_size_read(inode)); | |
198a1ca3 | 1333 | ocfs2_set_links_count(fe, inode->i_nlink); |
2c034176 EB |
1334 | fe->i_uid = cpu_to_le32(i_uid_read(inode)); |
1335 | fe->i_gid = cpu_to_le32(i_gid_read(inode)); | |
ccd979bd MF |
1336 | fe->i_mode = cpu_to_le16(inode->i_mode); |
1337 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); | |
1338 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
1339 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
1340 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
1341 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1342 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
1343 | ||
ec20cec7 | 1344 | ocfs2_journal_dirty(handle, bh); |
2931cdcb | 1345 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
ccd979bd | 1346 | leave: |
ccd979bd MF |
1347 | return status; |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * | |
1352 | * Updates a struct inode from a disk inode. | |
1353 | * does no i/o, only takes ip_lock. | |
1354 | */ | |
1355 | void ocfs2_refresh_inode(struct inode *inode, | |
1356 | struct ocfs2_dinode *fe) | |
1357 | { | |
ccd979bd MF |
1358 | spin_lock(&OCFS2_I(inode)->ip_lock); |
1359 | ||
1360 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e | 1361 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
15b1e36b | 1362 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
ca4d147e | 1363 | ocfs2_set_inode_flags(inode); |
ccd979bd | 1364 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
bfe86848 | 1365 | set_nlink(inode, ocfs2_read_links_count(fe)); |
2c034176 EB |
1366 | i_uid_write(inode, le32_to_cpu(fe->i_uid)); |
1367 | i_gid_write(inode, le32_to_cpu(fe->i_gid)); | |
ccd979bd | 1368 | inode->i_mode = le16_to_cpu(fe->i_mode); |
ccd979bd MF |
1369 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) |
1370 | inode->i_blocks = 0; | |
1371 | else | |
8110b073 | 1372 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
1373 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
1374 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
1375 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
1376 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
1377 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
1378 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
1379 | ||
1380 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1381 | } | |
b657c95c JB |
1382 | |
1383 | int ocfs2_validate_inode_block(struct super_block *sb, | |
1384 | struct buffer_head *bh) | |
1385 | { | |
d6b32bbb | 1386 | int rc; |
b657c95c JB |
1387 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; |
1388 | ||
64f3b269 | 1389 | trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr); |
970e4936 | 1390 | |
b657c95c JB |
1391 | BUG_ON(!buffer_uptodate(bh)); |
1392 | ||
d6b32bbb JB |
1393 | /* |
1394 | * If the ecc fails, we return the error but otherwise | |
1395 | * leave the filesystem running. We know any error is | |
1396 | * local to this block. | |
1397 | */ | |
1398 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check); | |
13723d00 JB |
1399 | if (rc) { |
1400 | mlog(ML_ERROR, "Checksum failed for dinode %llu\n", | |
1401 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 1402 | goto bail; |
13723d00 | 1403 | } |
d6b32bbb JB |
1404 | |
1405 | /* | |
1406 | * Errors after here are fatal. | |
1407 | */ | |
1408 | ||
1409 | rc = -EINVAL; | |
1410 | ||
b657c95c | 1411 | if (!OCFS2_IS_VALID_DINODE(di)) { |
17a5b9ab | 1412 | rc = ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n", |
7ecef14a JP |
1413 | (unsigned long long)bh->b_blocknr, 7, |
1414 | di->i_signature); | |
b657c95c JB |
1415 | goto bail; |
1416 | } | |
1417 | ||
1418 | if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) { | |
17a5b9ab | 1419 | rc = ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n", |
7ecef14a JP |
1420 | (unsigned long long)bh->b_blocknr, |
1421 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
b657c95c JB |
1422 | goto bail; |
1423 | } | |
1424 | ||
1425 | if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
17a5b9ab | 1426 | rc = ocfs2_error(sb, |
7ecef14a JP |
1427 | "Invalid dinode #%llu: OCFS2_VALID_FL not set\n", |
1428 | (unsigned long long)bh->b_blocknr); | |
b657c95c JB |
1429 | goto bail; |
1430 | } | |
1431 | ||
1432 | if (le32_to_cpu(di->i_fs_generation) != | |
1433 | OCFS2_SB(sb)->fs_generation) { | |
17a5b9ab | 1434 | rc = ocfs2_error(sb, |
7ecef14a JP |
1435 | "Invalid dinode #%llu: fs_generation is %u\n", |
1436 | (unsigned long long)bh->b_blocknr, | |
1437 | le32_to_cpu(di->i_fs_generation)); | |
b657c95c JB |
1438 | goto bail; |
1439 | } | |
1440 | ||
1441 | rc = 0; | |
1442 | ||
1443 | bail: | |
1444 | return rc; | |
1445 | } | |
1446 | ||
d56a8f32 GH |
1447 | static int ocfs2_filecheck_validate_inode_block(struct super_block *sb, |
1448 | struct buffer_head *bh) | |
1449 | { | |
1450 | int rc = 0; | |
1451 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; | |
1452 | ||
1453 | trace_ocfs2_filecheck_validate_inode_block( | |
1454 | (unsigned long long)bh->b_blocknr); | |
1455 | ||
1456 | BUG_ON(!buffer_uptodate(bh)); | |
1457 | ||
1458 | /* | |
1459 | * Call ocfs2_validate_meta_ecc() first since it has ecc repair | |
1460 | * function, but we should not return error immediately when ecc | |
1461 | * validation fails, because the reason is quite likely the invalid | |
1462 | * inode number inputed. | |
1463 | */ | |
1464 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check); | |
1465 | if (rc) { | |
1466 | mlog(ML_ERROR, | |
1467 | "Filecheck: checksum failed for dinode %llu\n", | |
1468 | (unsigned long long)bh->b_blocknr); | |
1469 | rc = -OCFS2_FILECHECK_ERR_BLOCKECC; | |
1470 | } | |
1471 | ||
1472 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
1473 | mlog(ML_ERROR, | |
1474 | "Filecheck: invalid dinode #%llu: signature = %.*s\n", | |
1475 | (unsigned long long)bh->b_blocknr, 7, di->i_signature); | |
1476 | rc = -OCFS2_FILECHECK_ERR_INVALIDINO; | |
1477 | goto bail; | |
1478 | } else if (rc) | |
1479 | goto bail; | |
1480 | ||
1481 | if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) { | |
1482 | mlog(ML_ERROR, | |
1483 | "Filecheck: invalid dinode #%llu: i_blkno is %llu\n", | |
1484 | (unsigned long long)bh->b_blocknr, | |
1485 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
1486 | rc = -OCFS2_FILECHECK_ERR_BLOCKNO; | |
1487 | goto bail; | |
1488 | } | |
1489 | ||
1490 | if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
1491 | mlog(ML_ERROR, | |
1492 | "Filecheck: invalid dinode #%llu: OCFS2_VALID_FL " | |
1493 | "not set\n", | |
1494 | (unsigned long long)bh->b_blocknr); | |
1495 | rc = -OCFS2_FILECHECK_ERR_VALIDFLAG; | |
1496 | goto bail; | |
1497 | } | |
1498 | ||
1499 | if (le32_to_cpu(di->i_fs_generation) != | |
1500 | OCFS2_SB(sb)->fs_generation) { | |
1501 | mlog(ML_ERROR, | |
1502 | "Filecheck: invalid dinode #%llu: fs_generation is %u\n", | |
1503 | (unsigned long long)bh->b_blocknr, | |
1504 | le32_to_cpu(di->i_fs_generation)); | |
1505 | rc = -OCFS2_FILECHECK_ERR_GENERATION; | |
1506 | goto bail; | |
1507 | } | |
1508 | ||
1509 | bail: | |
1510 | return rc; | |
1511 | } | |
1512 | ||
1513 | static int ocfs2_filecheck_repair_inode_block(struct super_block *sb, | |
1514 | struct buffer_head *bh) | |
1515 | { | |
1516 | int changed = 0; | |
1517 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; | |
1518 | ||
1519 | if (!ocfs2_filecheck_validate_inode_block(sb, bh)) | |
1520 | return 0; | |
1521 | ||
1522 | trace_ocfs2_filecheck_repair_inode_block( | |
1523 | (unsigned long long)bh->b_blocknr); | |
1524 | ||
1525 | if (ocfs2_is_hard_readonly(OCFS2_SB(sb)) || | |
1526 | ocfs2_is_soft_readonly(OCFS2_SB(sb))) { | |
1527 | mlog(ML_ERROR, | |
1528 | "Filecheck: cannot repair dinode #%llu " | |
1529 | "on readonly filesystem\n", | |
1530 | (unsigned long long)bh->b_blocknr); | |
1531 | return -OCFS2_FILECHECK_ERR_READONLY; | |
1532 | } | |
1533 | ||
1534 | if (buffer_jbd(bh)) { | |
1535 | mlog(ML_ERROR, | |
1536 | "Filecheck: cannot repair dinode #%llu, " | |
1537 | "its buffer is in jbd\n", | |
1538 | (unsigned long long)bh->b_blocknr); | |
1539 | return -OCFS2_FILECHECK_ERR_INJBD; | |
1540 | } | |
1541 | ||
1542 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
1543 | /* Cannot fix invalid inode block */ | |
1544 | return -OCFS2_FILECHECK_ERR_INVALIDINO; | |
1545 | } | |
1546 | ||
1547 | if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
1548 | /* Cannot just add VALID_FL flag back as a fix, | |
1549 | * need more things to check here. | |
1550 | */ | |
1551 | return -OCFS2_FILECHECK_ERR_VALIDFLAG; | |
1552 | } | |
1553 | ||
1554 | if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) { | |
1555 | di->i_blkno = cpu_to_le64(bh->b_blocknr); | |
1556 | changed = 1; | |
1557 | mlog(ML_ERROR, | |
1558 | "Filecheck: reset dinode #%llu: i_blkno to %llu\n", | |
1559 | (unsigned long long)bh->b_blocknr, | |
1560 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
1561 | } | |
1562 | ||
1563 | if (le32_to_cpu(di->i_fs_generation) != | |
1564 | OCFS2_SB(sb)->fs_generation) { | |
1565 | di->i_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation); | |
1566 | changed = 1; | |
1567 | mlog(ML_ERROR, | |
1568 | "Filecheck: reset dinode #%llu: fs_generation to %u\n", | |
1569 | (unsigned long long)bh->b_blocknr, | |
1570 | le32_to_cpu(di->i_fs_generation)); | |
1571 | } | |
1572 | ||
1573 | if (changed || ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check)) { | |
1574 | ocfs2_compute_meta_ecc(sb, bh->b_data, &di->i_check); | |
1575 | mark_buffer_dirty(bh); | |
1576 | mlog(ML_ERROR, | |
1577 | "Filecheck: reset dinode #%llu: compute meta ecc\n", | |
1578 | (unsigned long long)bh->b_blocknr); | |
1579 | } | |
1580 | ||
1581 | return 0; | |
1582 | } | |
1583 | ||
1584 | static int | |
1585 | ocfs2_filecheck_read_inode_block_full(struct inode *inode, | |
1586 | struct buffer_head **bh, | |
1587 | int flags, int type) | |
1588 | { | |
1589 | int rc; | |
1590 | struct buffer_head *tmp = *bh; | |
1591 | ||
1592 | if (!type) /* Check inode block */ | |
1593 | rc = ocfs2_read_blocks(INODE_CACHE(inode), | |
1594 | OCFS2_I(inode)->ip_blkno, | |
1595 | 1, &tmp, flags, | |
1596 | ocfs2_filecheck_validate_inode_block); | |
1597 | else /* Repair inode block */ | |
1598 | rc = ocfs2_read_blocks(INODE_CACHE(inode), | |
1599 | OCFS2_I(inode)->ip_blkno, | |
1600 | 1, &tmp, flags, | |
1601 | ocfs2_filecheck_repair_inode_block); | |
1602 | ||
1603 | /* If ocfs2_read_blocks() got us a new bh, pass it up. */ | |
1604 | if (!rc && !*bh) | |
1605 | *bh = tmp; | |
1606 | ||
1607 | return rc; | |
1608 | } | |
1609 | ||
b657c95c JB |
1610 | int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh, |
1611 | int flags) | |
1612 | { | |
1613 | int rc; | |
1614 | struct buffer_head *tmp = *bh; | |
1615 | ||
8cb471e8 JB |
1616 | rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno, |
1617 | 1, &tmp, flags, ocfs2_validate_inode_block); | |
b657c95c JB |
1618 | |
1619 | /* If ocfs2_read_blocks() got us a new bh, pass it up. */ | |
970e4936 | 1620 | if (!rc && !*bh) |
b657c95c JB |
1621 | *bh = tmp; |
1622 | ||
b657c95c JB |
1623 | return rc; |
1624 | } | |
1625 | ||
1626 | int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh) | |
1627 | { | |
1628 | return ocfs2_read_inode_block_full(inode, bh, 0); | |
1629 | } | |
6e5a3d75 | 1630 | |
6e5a3d75 JB |
1631 | |
1632 | static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci) | |
1633 | { | |
1634 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1635 | ||
1636 | return oi->ip_blkno; | |
1637 | } | |
1638 | ||
8cb471e8 JB |
1639 | static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci) |
1640 | { | |
1641 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1642 | ||
1643 | return oi->vfs_inode.i_sb; | |
1644 | } | |
1645 | ||
6e5a3d75 JB |
1646 | static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci) |
1647 | { | |
1648 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1649 | ||
1650 | spin_lock(&oi->ip_lock); | |
1651 | } | |
1652 | ||
1653 | static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci) | |
1654 | { | |
1655 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1656 | ||
1657 | spin_unlock(&oi->ip_lock); | |
1658 | } | |
1659 | ||
1660 | static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci) | |
1661 | { | |
1662 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1663 | ||
1664 | mutex_lock(&oi->ip_io_mutex); | |
1665 | } | |
1666 | ||
1667 | static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci) | |
1668 | { | |
1669 | struct ocfs2_inode_info *oi = cache_info_to_inode(ci); | |
1670 | ||
1671 | mutex_unlock(&oi->ip_io_mutex); | |
1672 | } | |
1673 | ||
1674 | const struct ocfs2_caching_operations ocfs2_inode_caching_ops = { | |
1675 | .co_owner = ocfs2_inode_cache_owner, | |
8cb471e8 | 1676 | .co_get_super = ocfs2_inode_cache_get_super, |
6e5a3d75 JB |
1677 | .co_cache_lock = ocfs2_inode_cache_lock, |
1678 | .co_cache_unlock = ocfs2_inode_cache_unlock, | |
1679 | .co_io_lock = ocfs2_inode_cache_io_lock, | |
1680 | .co_io_unlock = ocfs2_inode_cache_io_unlock, | |
1681 | }; | |
1682 |