]>
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> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
30 | #include <linux/pagemap.h> | |
ccd979bd MF |
31 | |
32 | #include <asm/byteorder.h> | |
33 | ||
34 | #define MLOG_MASK_PREFIX ML_INODE | |
35 | #include <cluster/masklog.h> | |
36 | ||
37 | #include "ocfs2.h" | |
38 | ||
39 | #include "alloc.h" | |
40 | #include "dlmglue.h" | |
41 | #include "extent_map.h" | |
42 | #include "file.h" | |
b4df6ed8 | 43 | #include "heartbeat.h" |
ccd979bd MF |
44 | #include "inode.h" |
45 | #include "journal.h" | |
46 | #include "namei.h" | |
47 | #include "suballoc.h" | |
48 | #include "super.h" | |
49 | #include "symlink.h" | |
50 | #include "sysfile.h" | |
51 | #include "uptodate.h" | |
52 | #include "vote.h" | |
53 | ||
54 | #include "buffer_head_io.h" | |
55 | ||
ccd979bd MF |
56 | struct ocfs2_find_inode_args |
57 | { | |
58 | u64 fi_blkno; | |
59 | unsigned long fi_ino; | |
60 | unsigned int fi_flags; | |
61 | }; | |
62 | ||
63 | static int ocfs2_read_locked_inode(struct inode *inode, | |
64 | struct ocfs2_find_inode_args *args); | |
65 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); | |
66 | static int ocfs2_find_actor(struct inode *inode, void *opaque); | |
67 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
68 | struct inode *inode, | |
69 | struct buffer_head *fe_bh); | |
70 | ||
ca4d147e HP |
71 | void ocfs2_set_inode_flags(struct inode *inode) |
72 | { | |
73 | unsigned int flags = OCFS2_I(inode)->ip_attr; | |
74 | ||
75 | inode->i_flags &= ~(S_IMMUTABLE | | |
76 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); | |
77 | ||
78 | if (flags & OCFS2_IMMUTABLE_FL) | |
79 | inode->i_flags |= S_IMMUTABLE; | |
80 | ||
81 | if (flags & OCFS2_SYNC_FL) | |
82 | inode->i_flags |= S_SYNC; | |
83 | if (flags & OCFS2_APPEND_FL) | |
84 | inode->i_flags |= S_APPEND; | |
85 | if (flags & OCFS2_NOATIME_FL) | |
86 | inode->i_flags |= S_NOATIME; | |
87 | if (flags & OCFS2_DIRSYNC_FL) | |
88 | inode->i_flags |= S_DIRSYNC; | |
89 | } | |
90 | ||
6e4b0d56 JK |
91 | /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */ |
92 | void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi) | |
93 | { | |
94 | unsigned int flags = oi->vfs_inode.i_flags; | |
95 | ||
96 | oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL| | |
97 | OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL); | |
98 | if (flags & S_SYNC) | |
99 | oi->ip_attr |= OCFS2_SYNC_FL; | |
100 | if (flags & S_APPEND) | |
101 | oi->ip_attr |= OCFS2_APPEND_FL; | |
102 | if (flags & S_IMMUTABLE) | |
103 | oi->ip_attr |= OCFS2_IMMUTABLE_FL; | |
104 | if (flags & S_NOATIME) | |
105 | oi->ip_attr |= OCFS2_NOATIME_FL; | |
106 | if (flags & S_DIRSYNC) | |
107 | oi->ip_attr |= OCFS2_DIRSYNC_FL; | |
108 | } | |
109 | ||
24c19ef4 | 110 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags) |
ccd979bd MF |
111 | { |
112 | struct inode *inode = NULL; | |
113 | struct super_block *sb = osb->sb; | |
114 | struct ocfs2_find_inode_args args; | |
115 | ||
b0697053 | 116 | mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno); |
ccd979bd MF |
117 | |
118 | /* Ok. By now we've either got the offsets passed to us by the | |
119 | * caller, or we just pulled them off the bh. Lets do some | |
120 | * sanity checks to make sure they're OK. */ | |
121 | if (blkno == 0) { | |
122 | inode = ERR_PTR(-EINVAL); | |
123 | mlog_errno(PTR_ERR(inode)); | |
124 | goto bail; | |
125 | } | |
126 | ||
127 | args.fi_blkno = blkno; | |
24c19ef4 | 128 | args.fi_flags = flags; |
ccd979bd MF |
129 | args.fi_ino = ino_from_blkno(sb, blkno); |
130 | ||
131 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, | |
132 | ocfs2_init_locked_inode, &args); | |
133 | /* inode was *not* in the inode cache. 2.6.x requires | |
134 | * us to do our own read_inode call and unlock it | |
135 | * afterwards. */ | |
136 | if (inode && inode->i_state & I_NEW) { | |
137 | mlog(0, "Inode was not in inode cache, reading it.\n"); | |
138 | ocfs2_read_locked_inode(inode, &args); | |
139 | unlock_new_inode(inode); | |
140 | } | |
141 | if (inode == NULL) { | |
142 | inode = ERR_PTR(-ENOMEM); | |
143 | mlog_errno(PTR_ERR(inode)); | |
144 | goto bail; | |
145 | } | |
146 | if (is_bad_inode(inode)) { | |
147 | iput(inode); | |
148 | inode = ERR_PTR(-ESTALE); | |
ccd979bd MF |
149 | goto bail; |
150 | } | |
151 | ||
152 | bail: | |
153 | if (!IS_ERR(inode)) { | |
b0697053 MF |
154 | mlog(0, "returning inode with number %llu\n", |
155 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 156 | mlog_exit_ptr(inode); |
6a1bd4a5 | 157 | } |
ccd979bd MF |
158 | |
159 | return inode; | |
160 | } | |
161 | ||
162 | ||
163 | /* | |
164 | * here's how inodes get read from disk: | |
165 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR | |
166 | * found? : return the in-memory inode | |
167 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE | |
168 | */ | |
169 | ||
170 | static int ocfs2_find_actor(struct inode *inode, void *opaque) | |
171 | { | |
172 | struct ocfs2_find_inode_args *args = NULL; | |
173 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
174 | int ret = 0; | |
175 | ||
176 | mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque); | |
177 | ||
178 | args = opaque; | |
179 | ||
180 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); | |
181 | ||
182 | if (oi->ip_blkno != args->fi_blkno) | |
183 | goto bail; | |
184 | ||
ccd979bd MF |
185 | ret = 1; |
186 | bail: | |
187 | mlog_exit(ret); | |
188 | return ret; | |
189 | } | |
190 | ||
191 | /* | |
192 | * initialize the new inode, but don't do anything that would cause | |
193 | * us to sleep. | |
194 | * return 0 on success, 1 on failure | |
195 | */ | |
196 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) | |
197 | { | |
198 | struct ocfs2_find_inode_args *args = opaque; | |
199 | ||
200 | mlog_entry("inode = %p, opaque = %p\n", inode, opaque); | |
201 | ||
202 | inode->i_ino = args->fi_ino; | |
203 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; | |
204 | ||
205 | mlog_exit(0); | |
206 | return 0; | |
207 | } | |
208 | ||
209 | int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, | |
210 | int create_ino) | |
211 | { | |
212 | struct super_block *sb; | |
213 | struct ocfs2_super *osb; | |
214 | int status = -EINVAL; | |
215 | ||
b0697053 | 216 | mlog_entry("(0x%p, size:%llu)\n", inode, |
1ca1a111 | 217 | (unsigned long long)le64_to_cpu(fe->i_size)); |
ccd979bd MF |
218 | |
219 | sb = inode->i_sb; | |
220 | osb = OCFS2_SB(sb); | |
221 | ||
222 | /* this means that read_inode cannot create a superblock inode | |
223 | * today. change if needed. */ | |
224 | if (!OCFS2_IS_VALID_DINODE(fe) || | |
225 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
6a1bd4a5 | 226 | mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, " |
ccd979bd | 227 | "signature = %.*s, flags = 0x%x\n", |
b0697053 MF |
228 | inode->i_ino, |
229 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, | |
ccd979bd MF |
230 | fe->i_signature, le32_to_cpu(fe->i_flags)); |
231 | goto bail; | |
232 | } | |
233 | ||
234 | if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) { | |
235 | mlog(ML_ERROR, "file entry generation does not match " | |
236 | "superblock! osb->fs_generation=%x, " | |
237 | "fe->i_fs_generation=%x\n", | |
238 | osb->fs_generation, le32_to_cpu(fe->i_fs_generation)); | |
239 | goto bail; | |
240 | } | |
241 | ||
8110b073 MF |
242 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
243 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); | |
15b1e36b | 244 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
8110b073 | 245 | |
ccd979bd MF |
246 | inode->i_version = 1; |
247 | inode->i_generation = le32_to_cpu(fe->i_generation); | |
248 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
249 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
250 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
251 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
ccd979bd MF |
252 | |
253 | /* Fast symlinks will have i_size but no allocated clusters. */ | |
254 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) | |
255 | inode->i_blocks = 0; | |
256 | else | |
8110b073 | 257 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 258 | inode->i_mapping->a_ops = &ocfs2_aops; |
ccd979bd MF |
259 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
260 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
261 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
262 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
263 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
264 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
265 | ||
266 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) | |
267 | mlog(ML_ERROR, | |
b0697053 MF |
268 | "ip_blkno %llu != i_blkno %llu!\n", |
269 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1ca1a111 | 270 | (unsigned long long)le64_to_cpu(fe->i_blkno)); |
ccd979bd | 271 | |
ccd979bd MF |
272 | inode->i_nlink = le16_to_cpu(fe->i_links_count); |
273 | ||
24c19ef4 MF |
274 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) |
275 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; | |
276 | ||
ccd979bd MF |
277 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { |
278 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
279 | mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino); | |
280 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { | |
281 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
282 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { | |
283 | mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino); | |
284 | /* we can't actually hit this as read_inode can't | |
285 | * handle superblocks today ;-) */ | |
286 | BUG(); | |
287 | } | |
288 | ||
289 | switch (inode->i_mode & S_IFMT) { | |
290 | case S_IFREG: | |
291 | inode->i_fop = &ocfs2_fops; | |
292 | inode->i_op = &ocfs2_file_iops; | |
293 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
294 | break; | |
295 | case S_IFDIR: | |
296 | inode->i_op = &ocfs2_dir_iops; | |
297 | inode->i_fop = &ocfs2_dops; | |
298 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
299 | break; | |
300 | case S_IFLNK: | |
301 | if (ocfs2_inode_is_fast_symlink(inode)) | |
302 | inode->i_op = &ocfs2_fast_symlink_inode_operations; | |
303 | else | |
304 | inode->i_op = &ocfs2_symlink_inode_operations; | |
305 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
306 | break; | |
307 | default: | |
308 | inode->i_op = &ocfs2_special_file_iops; | |
309 | init_special_inode(inode, inode->i_mode, | |
310 | inode->i_rdev); | |
311 | break; | |
312 | } | |
313 | ||
24c19ef4 MF |
314 | if (create_ino) { |
315 | inode->i_ino = ino_from_blkno(inode->i_sb, | |
316 | le64_to_cpu(fe->i_blkno)); | |
317 | ||
318 | /* | |
319 | * If we ever want to create system files from kernel, | |
320 | * the generation argument to | |
321 | * ocfs2_inode_lock_res_init() will have to change. | |
322 | */ | |
1ca1a111 | 323 | BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL); |
24c19ef4 MF |
324 | |
325 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
326 | OCFS2_LOCK_TYPE_META, 0, inode); | |
50008630 TY |
327 | |
328 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, | |
329 | OCFS2_LOCK_TYPE_OPEN, 0, inode); | |
24c19ef4 MF |
330 | } |
331 | ||
ccd979bd | 332 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, |
24c19ef4 MF |
333 | OCFS2_LOCK_TYPE_RW, inode->i_generation, |
334 | inode); | |
335 | ||
ccd979bd | 336 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres, |
24c19ef4 MF |
337 | OCFS2_LOCK_TYPE_DATA, inode->i_generation, |
338 | inode); | |
ccd979bd | 339 | |
ca4d147e | 340 | ocfs2_set_inode_flags(inode); |
ca4d147e | 341 | |
ccd979bd MF |
342 | status = 0; |
343 | bail: | |
344 | mlog_exit(status); | |
345 | return status; | |
346 | } | |
347 | ||
348 | static int ocfs2_read_locked_inode(struct inode *inode, | |
349 | struct ocfs2_find_inode_args *args) | |
350 | { | |
351 | struct super_block *sb; | |
352 | struct ocfs2_super *osb; | |
353 | struct ocfs2_dinode *fe; | |
354 | struct buffer_head *bh = NULL; | |
24c19ef4 MF |
355 | int status, can_lock; |
356 | u32 generation = 0; | |
ccd979bd MF |
357 | |
358 | mlog_entry("(0x%p, 0x%p)\n", inode, args); | |
359 | ||
360 | status = -EINVAL; | |
361 | if (inode == NULL || inode->i_sb == NULL) { | |
362 | mlog(ML_ERROR, "bad inode\n"); | |
24c19ef4 | 363 | return status; |
ccd979bd MF |
364 | } |
365 | sb = inode->i_sb; | |
366 | osb = OCFS2_SB(sb); | |
367 | ||
368 | if (!args) { | |
369 | mlog(ML_ERROR, "bad inode args\n"); | |
370 | make_bad_inode(inode); | |
24c19ef4 MF |
371 | return status; |
372 | } | |
373 | ||
374 | /* | |
375 | * To improve performance of cold-cache inode stats, we take | |
376 | * the cluster lock here if possible. | |
377 | * | |
378 | * Generally, OCFS2 never trusts the contents of an inode | |
379 | * unless it's holding a cluster lock, so taking it here isn't | |
380 | * a correctness issue as much as it is a performance | |
381 | * improvement. | |
382 | * | |
383 | * There are three times when taking the lock is not a good idea: | |
384 | * | |
385 | * 1) During startup, before we have initialized the DLM. | |
386 | * | |
387 | * 2) If we are reading certain system files which never get | |
388 | * cluster locks (local alloc, truncate log). | |
389 | * | |
390 | * 3) If the process doing the iget() is responsible for | |
391 | * orphan dir recovery. We're holding the orphan dir lock and | |
392 | * can get into a deadlock with another process on another | |
393 | * node in ->delete_inode(). | |
394 | * | |
395 | * #1 and #2 can be simply solved by never taking the lock | |
396 | * here for system files (which are the only type we read | |
397 | * during mount). It's a heavier approach, but our main | |
398 | * concern is user-accesible files anyway. | |
399 | * | |
400 | * #3 works itself out because we'll eventually take the | |
401 | * cluster lock before trusting anything anyway. | |
402 | */ | |
403 | can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
50008630 | 404 | && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) |
c271c5c2 | 405 | && !ocfs2_mount_local(osb); |
24c19ef4 MF |
406 | |
407 | /* | |
408 | * To maintain backwards compatibility with older versions of | |
409 | * ocfs2-tools, we still store the generation value for system | |
410 | * files. The only ones that actually matter to userspace are | |
411 | * the journals, but it's easier and inexpensive to just flag | |
412 | * all system files similarly. | |
413 | */ | |
414 | if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
415 | generation = osb->fs_generation; | |
416 | ||
417 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
418 | OCFS2_LOCK_TYPE_META, | |
419 | generation, inode); | |
420 | ||
50008630 TY |
421 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, |
422 | OCFS2_LOCK_TYPE_OPEN, | |
423 | 0, inode); | |
424 | ||
24c19ef4 | 425 | if (can_lock) { |
50008630 TY |
426 | status = ocfs2_open_lock(inode); |
427 | if (status) { | |
428 | make_bad_inode(inode); | |
429 | mlog_errno(status); | |
430 | return status; | |
431 | } | |
4bcec184 | 432 | status = ocfs2_meta_lock(inode, NULL, 0); |
24c19ef4 MF |
433 | if (status) { |
434 | make_bad_inode(inode); | |
435 | mlog_errno(status); | |
436 | return status; | |
437 | } | |
ccd979bd MF |
438 | } |
439 | ||
50008630 TY |
440 | if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) { |
441 | status = ocfs2_try_open_lock(inode, 0); | |
442 | if (status) { | |
443 | make_bad_inode(inode); | |
444 | return status; | |
445 | } | |
446 | } | |
447 | ||
24c19ef4 MF |
448 | status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, |
449 | can_lock ? inode : NULL); | |
ccd979bd MF |
450 | if (status < 0) { |
451 | mlog_errno(status); | |
ccd979bd MF |
452 | goto bail; |
453 | } | |
454 | ||
24c19ef4 | 455 | status = -EINVAL; |
ccd979bd MF |
456 | fe = (struct ocfs2_dinode *) bh->b_data; |
457 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
a46043e0 MF |
458 | mlog(0, "Invalid dinode #%llu: signature = %.*s\n", |
459 | (unsigned long long)args->fi_blkno, 7, | |
1ca1a111 | 460 | fe->i_signature); |
ccd979bd MF |
461 | goto bail; |
462 | } | |
463 | ||
24c19ef4 MF |
464 | /* |
465 | * This is a code bug. Right now the caller needs to | |
466 | * understand whether it is asking for a system file inode or | |
467 | * not so the proper lock names can be built. | |
468 | */ | |
469 | mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) != | |
470 | !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE), | |
471 | "Inode %llu: system file state is ambigous\n", | |
472 | (unsigned long long)args->fi_blkno); | |
ccd979bd MF |
473 | |
474 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || | |
475 | S_ISBLK(le16_to_cpu(fe->i_mode))) | |
476 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
477 | ||
6a1bd4a5 | 478 | if (ocfs2_populate_inode(inode, fe, 0) < 0) |
ccd979bd | 479 | goto bail; |
ccd979bd MF |
480 | |
481 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); | |
482 | ||
ccd979bd MF |
483 | status = 0; |
484 | ||
485 | bail: | |
24c19ef4 MF |
486 | if (can_lock) |
487 | ocfs2_meta_unlock(inode, 0); | |
488 | ||
489 | if (status < 0) | |
490 | make_bad_inode(inode); | |
491 | ||
ccd979bd MF |
492 | if (args && bh) |
493 | brelse(bh); | |
494 | ||
495 | mlog_exit(status); | |
496 | return status; | |
497 | } | |
498 | ||
499 | void ocfs2_sync_blockdev(struct super_block *sb) | |
500 | { | |
501 | sync_blockdev(sb->s_bdev); | |
502 | } | |
503 | ||
504 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
505 | struct inode *inode, | |
506 | struct buffer_head *fe_bh) | |
507 | { | |
508 | int status = 0; | |
ccd979bd MF |
509 | struct ocfs2_truncate_context *tc = NULL; |
510 | struct ocfs2_dinode *fe; | |
60b11392 | 511 | handle_t *handle = NULL; |
ccd979bd MF |
512 | |
513 | mlog_entry_void(); | |
514 | ||
515 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
516 | ||
1afc32b9 MF |
517 | /* |
518 | * This check will also skip truncate of inodes with inline | |
519 | * data and fast symlinks. | |
520 | */ | |
3a0782d0 | 521 | if (fe->i_clusters) { |
60b11392 MF |
522 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
523 | if (IS_ERR(handle)) { | |
524 | status = PTR_ERR(handle); | |
525 | mlog_errno(status); | |
526 | goto out; | |
527 | } | |
528 | ||
529 | status = ocfs2_journal_access(handle, inode, fe_bh, | |
530 | OCFS2_JOURNAL_ACCESS_WRITE); | |
531 | if (status < 0) { | |
532 | mlog_errno(status); | |
533 | goto out; | |
534 | } | |
535 | ||
536 | i_size_write(inode, 0); | |
537 | ||
538 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
539 | if (status < 0) { | |
540 | mlog_errno(status); | |
541 | goto out; | |
542 | } | |
543 | ||
544 | ocfs2_commit_trans(osb, handle); | |
545 | handle = NULL; | |
546 | ||
3a0782d0 MF |
547 | status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc); |
548 | if (status < 0) { | |
549 | mlog_errno(status); | |
550 | goto out; | |
551 | } | |
ccd979bd | 552 | |
3a0782d0 MF |
553 | status = ocfs2_commit_truncate(osb, inode, fe_bh, tc); |
554 | if (status < 0) { | |
555 | mlog_errno(status); | |
556 | goto out; | |
557 | } | |
ccd979bd | 558 | } |
ccd979bd | 559 | |
60b11392 MF |
560 | out: |
561 | if (handle) | |
562 | ocfs2_commit_trans(osb, handle); | |
ccd979bd MF |
563 | mlog_exit(status); |
564 | return status; | |
565 | } | |
566 | ||
567 | static int ocfs2_remove_inode(struct inode *inode, | |
568 | struct buffer_head *di_bh, | |
569 | struct inode *orphan_dir_inode, | |
570 | struct buffer_head *orphan_dir_bh) | |
571 | { | |
572 | int status; | |
573 | struct inode *inode_alloc_inode = NULL; | |
574 | struct buffer_head *inode_alloc_bh = NULL; | |
1fabe148 | 575 | handle_t *handle; |
ccd979bd MF |
576 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
577 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
578 | ||
579 | inode_alloc_inode = | |
580 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
581 | le16_to_cpu(di->i_suballoc_slot)); | |
582 | if (!inode_alloc_inode) { | |
583 | status = -EEXIST; | |
584 | mlog_errno(status); | |
585 | goto bail; | |
586 | } | |
587 | ||
1b1dcc1b | 588 | mutex_lock(&inode_alloc_inode->i_mutex); |
4bcec184 | 589 | status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1); |
ccd979bd | 590 | if (status < 0) { |
1b1dcc1b | 591 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
592 | |
593 | mlog_errno(status); | |
594 | goto bail; | |
595 | } | |
596 | ||
65eff9cc | 597 | handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS); |
ccd979bd MF |
598 | if (IS_ERR(handle)) { |
599 | status = PTR_ERR(handle); | |
600 | mlog_errno(status); | |
601 | goto bail_unlock; | |
602 | } | |
603 | ||
604 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, | |
605 | orphan_dir_bh); | |
606 | if (status < 0) { | |
607 | mlog_errno(status); | |
608 | goto bail_commit; | |
609 | } | |
610 | ||
611 | /* set the inodes dtime */ | |
612 | status = ocfs2_journal_access(handle, inode, di_bh, | |
613 | OCFS2_JOURNAL_ACCESS_WRITE); | |
614 | if (status < 0) { | |
615 | mlog_errno(status); | |
616 | goto bail_commit; | |
617 | } | |
618 | ||
619 | di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec); | |
620 | le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); | |
621 | ||
622 | status = ocfs2_journal_dirty(handle, di_bh); | |
623 | if (status < 0) { | |
624 | mlog_errno(status); | |
625 | goto bail_commit; | |
626 | } | |
627 | ||
628 | ocfs2_remove_from_cache(inode, di_bh); | |
629 | ||
630 | status = ocfs2_free_dinode(handle, inode_alloc_inode, | |
631 | inode_alloc_bh, di); | |
632 | if (status < 0) | |
633 | mlog_errno(status); | |
634 | ||
635 | bail_commit: | |
02dc1af4 | 636 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
637 | bail_unlock: |
638 | ocfs2_meta_unlock(inode_alloc_inode, 1); | |
1b1dcc1b | 639 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
640 | brelse(inode_alloc_bh); |
641 | bail: | |
642 | iput(inode_alloc_inode); | |
643 | ||
644 | return status; | |
645 | } | |
646 | ||
b4df6ed8 MF |
647 | /* |
648 | * Serialize with orphan dir recovery. If the process doing | |
649 | * recovery on this orphan dir does an iget() with the dir | |
650 | * i_mutex held, we'll deadlock here. Instead we detect this | |
651 | * and exit early - recovery will wipe this inode for us. | |
652 | */ | |
653 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, | |
654 | int slot) | |
655 | { | |
656 | int ret = 0; | |
657 | ||
658 | spin_lock(&osb->osb_lock); | |
659 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { | |
660 | mlog(0, "Recovery is happening on orphan dir %d, will skip " | |
661 | "this inode\n", slot); | |
662 | ret = -EDEADLK; | |
663 | goto out; | |
664 | } | |
665 | /* This signals to the orphan recovery process that it should | |
666 | * wait for us to handle the wipe. */ | |
667 | osb->osb_orphan_wipes[slot]++; | |
668 | out: | |
669 | spin_unlock(&osb->osb_lock); | |
670 | return ret; | |
671 | } | |
672 | ||
673 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, | |
674 | int slot) | |
675 | { | |
676 | spin_lock(&osb->osb_lock); | |
677 | osb->osb_orphan_wipes[slot]--; | |
678 | spin_unlock(&osb->osb_lock); | |
679 | ||
680 | wake_up(&osb->osb_wipe_event); | |
681 | } | |
682 | ||
ccd979bd MF |
683 | static int ocfs2_wipe_inode(struct inode *inode, |
684 | struct buffer_head *di_bh) | |
685 | { | |
686 | int status, orphaned_slot; | |
687 | struct inode *orphan_dir_inode = NULL; | |
688 | struct buffer_head *orphan_dir_bh = NULL; | |
689 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
50008630 | 690 | struct ocfs2_dinode *di; |
ccd979bd | 691 | |
50008630 TY |
692 | di = (struct ocfs2_dinode *) di_bh->b_data; |
693 | orphaned_slot = le16_to_cpu(di->i_orphaned_slot); | |
b4df6ed8 MF |
694 | |
695 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); | |
696 | if (status) | |
697 | return status; | |
698 | ||
ccd979bd MF |
699 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
700 | ORPHAN_DIR_SYSTEM_INODE, | |
701 | orphaned_slot); | |
702 | if (!orphan_dir_inode) { | |
703 | status = -EEXIST; | |
704 | mlog_errno(status); | |
705 | goto bail; | |
706 | } | |
707 | ||
708 | /* Lock the orphan dir. The lock will be held for the entire | |
709 | * delete_inode operation. We do this now to avoid races with | |
710 | * recovery completion on other nodes. */ | |
1b1dcc1b | 711 | mutex_lock(&orphan_dir_inode->i_mutex); |
4bcec184 | 712 | status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1); |
ccd979bd | 713 | if (status < 0) { |
1b1dcc1b | 714 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
715 | |
716 | mlog_errno(status); | |
717 | goto bail; | |
718 | } | |
719 | ||
720 | /* we do this while holding the orphan dir lock because we | |
721 | * don't want recovery being run from another node to vote for | |
722 | * an inode delete on us -- this will result in two nodes | |
723 | * truncating the same file! */ | |
724 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); | |
725 | if (status < 0) { | |
726 | mlog_errno(status); | |
727 | goto bail_unlock_dir; | |
728 | } | |
729 | ||
730 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, | |
731 | orphan_dir_bh); | |
732 | if (status < 0) | |
733 | mlog_errno(status); | |
734 | ||
735 | bail_unlock_dir: | |
736 | ocfs2_meta_unlock(orphan_dir_inode, 1); | |
1b1dcc1b | 737 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
738 | brelse(orphan_dir_bh); |
739 | bail: | |
740 | iput(orphan_dir_inode); | |
b4df6ed8 | 741 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
ccd979bd MF |
742 | |
743 | return status; | |
744 | } | |
745 | ||
746 | /* There is a series of simple checks that should be done before a | |
747 | * vote is even considered. Encapsulate those in this function. */ | |
748 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) | |
749 | { | |
750 | int ret = 0; | |
751 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
752 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
753 | ||
754 | /* We shouldn't be getting here for the root directory | |
755 | * inode.. */ | |
756 | if (inode == osb->root_inode) { | |
757 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); | |
758 | goto bail; | |
759 | } | |
760 | ||
761 | /* If we're coming from process_vote we can't go into our own | |
762 | * voting [hello, deadlock city!], so unforuntately we just | |
763 | * have to skip deleting this guy. That's OK though because | |
764 | * the node who's doing the actual deleting should handle it | |
765 | * anyway. */ | |
766 | if (current == osb->vote_task) { | |
767 | mlog(0, "Skipping delete of %lu because we're currently " | |
768 | "in process_vote\n", inode->i_ino); | |
769 | goto bail; | |
770 | } | |
771 | ||
772 | spin_lock(&oi->ip_lock); | |
773 | /* OCFS2 *never* deletes system files. This should technically | |
774 | * never get here as system file inodes should always have a | |
775 | * positive link count. */ | |
776 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { | |
b0697053 MF |
777 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
778 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
779 | goto bail_unlock; |
780 | } | |
781 | ||
782 | /* If we have voted "yes" on the wipe of this inode for | |
783 | * another node, it will be marked here so we can safely skip | |
784 | * it. Recovery will cleanup any inodes we might inadvertantly | |
785 | * skip here. */ | |
786 | if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) { | |
787 | mlog(0, "Skipping delete of %lu because another node " | |
788 | "has done this for us.\n", inode->i_ino); | |
789 | goto bail_unlock; | |
790 | } | |
791 | ||
792 | ret = 1; | |
793 | bail_unlock: | |
794 | spin_unlock(&oi->ip_lock); | |
795 | bail: | |
796 | return ret; | |
797 | } | |
798 | ||
799 | /* Query the cluster to determine whether we should wipe an inode from | |
800 | * disk or not. | |
801 | * | |
802 | * Requires the inode to have the cluster lock. */ | |
803 | static int ocfs2_query_inode_wipe(struct inode *inode, | |
804 | struct buffer_head *di_bh, | |
805 | int *wipe) | |
806 | { | |
807 | int status = 0; | |
808 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
809 | struct ocfs2_dinode *di; | |
810 | ||
811 | *wipe = 0; | |
812 | ||
813 | /* While we were waiting for the cluster lock in | |
814 | * ocfs2_delete_inode, another node might have asked to delete | |
815 | * the inode. Recheck our flags to catch this. */ | |
816 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
b0697053 MF |
817 | mlog(0, "Skipping delete of %llu because flags changed\n", |
818 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
819 | goto bail; |
820 | } | |
821 | ||
822 | /* Now that we have an up to date inode, we can double check | |
823 | * the link count. */ | |
824 | if (inode->i_nlink) { | |
b0697053 MF |
825 | mlog(0, "Skipping delete of %llu because nlink = %u\n", |
826 | (unsigned long long)oi->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
827 | goto bail; |
828 | } | |
829 | ||
830 | /* Do some basic inode verification... */ | |
831 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
832 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) { | |
833 | /* for lack of a better error? */ | |
834 | status = -EEXIST; | |
835 | mlog(ML_ERROR, | |
b0697053 | 836 | "Inode %llu (on-disk %llu) not orphaned! " |
ccd979bd | 837 | "Disk flags 0x%x, inode flags 0x%x\n", |
b0697053 | 838 | (unsigned long long)oi->ip_blkno, |
1ca1a111 MF |
839 | (unsigned long long)le64_to_cpu(di->i_blkno), |
840 | le32_to_cpu(di->i_flags), oi->ip_flags); | |
ccd979bd MF |
841 | goto bail; |
842 | } | |
843 | ||
844 | /* has someone already deleted us?! baaad... */ | |
845 | if (di->i_dtime) { | |
846 | status = -EEXIST; | |
847 | mlog_errno(status); | |
848 | goto bail; | |
849 | } | |
850 | ||
6f16bf65 MF |
851 | /* |
852 | * This is how ocfs2 determines whether an inode is still live | |
853 | * within the cluster. Every node takes a shared read lock on | |
854 | * the inode open lock in ocfs2_read_locked_inode(). When we | |
855 | * get to ->delete_inode(), each node tries to convert it's | |
856 | * lock to an exclusive. Trylocks are serialized by the inode | |
857 | * meta data lock. If the upconvert suceeds, we know the inode | |
858 | * is no longer live and can be deleted. | |
859 | * | |
860 | * Though we call this with the meta data lock held, the | |
861 | * trylock keeps us from ABBA deadlock. | |
862 | */ | |
863 | status = ocfs2_try_open_lock(inode, 1); | |
50008630 | 864 | if (status == -EAGAIN) { |
ccd979bd | 865 | status = 0; |
2759236f | 866 | mlog(0, "Skipping delete of %llu because it is in use on " |
b0697053 | 867 | "other nodes\n", (unsigned long long)oi->ip_blkno); |
ccd979bd MF |
868 | goto bail; |
869 | } | |
870 | if (status < 0) { | |
871 | mlog_errno(status); | |
872 | goto bail; | |
873 | } | |
874 | ||
50008630 TY |
875 | *wipe = 1; |
876 | mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n", | |
877 | (unsigned long long)oi->ip_blkno, | |
878 | le16_to_cpu(di->i_orphaned_slot)); | |
ccd979bd MF |
879 | |
880 | bail: | |
881 | return status; | |
882 | } | |
883 | ||
884 | /* Support function for ocfs2_delete_inode. Will help us keep the | |
885 | * inode data in a consistent state for clear_inode. Always truncates | |
886 | * pages, optionally sync's them first. */ | |
887 | static void ocfs2_cleanup_delete_inode(struct inode *inode, | |
888 | int sync_data) | |
889 | { | |
b0697053 MF |
890 | mlog(0, "Cleanup inode %llu, sync = %d\n", |
891 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); | |
ccd979bd MF |
892 | if (sync_data) |
893 | write_inode_now(inode, 1); | |
894 | truncate_inode_pages(&inode->i_data, 0); | |
895 | } | |
896 | ||
897 | void ocfs2_delete_inode(struct inode *inode) | |
898 | { | |
899 | int wipe, status; | |
900 | sigset_t blocked, oldset; | |
901 | struct buffer_head *di_bh = NULL; | |
902 | ||
903 | mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino); | |
904 | ||
905 | if (is_bad_inode(inode)) { | |
906 | mlog(0, "Skipping delete of bad inode\n"); | |
907 | goto bail; | |
908 | } | |
909 | ||
910 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
911 | /* It's probably not necessary to truncate_inode_pages | |
912 | * here but we do it for safety anyway (it will most | |
913 | * likely be a no-op anyway) */ | |
914 | ocfs2_cleanup_delete_inode(inode, 0); | |
915 | goto bail; | |
916 | } | |
917 | ||
918 | /* We want to block signals in delete_inode as the lock and | |
919 | * messaging paths may return us -ERESTARTSYS. Which would | |
920 | * cause us to exit early, resulting in inodes being orphaned | |
921 | * forever. */ | |
922 | sigfillset(&blocked); | |
923 | status = sigprocmask(SIG_BLOCK, &blocked, &oldset); | |
924 | if (status < 0) { | |
925 | mlog_errno(status); | |
926 | ocfs2_cleanup_delete_inode(inode, 1); | |
927 | goto bail; | |
928 | } | |
929 | ||
930 | /* Lock down the inode. This gives us an up to date view of | |
931 | * it's metadata (for verification), and allows us to | |
932 | * serialize delete_inode votes. | |
933 | * | |
934 | * Even though we might be doing a truncate, we don't take the | |
935 | * allocation lock here as it won't be needed - nobody will | |
936 | * have the file open. | |
937 | */ | |
4bcec184 | 938 | status = ocfs2_meta_lock(inode, &di_bh, 1); |
ccd979bd MF |
939 | if (status < 0) { |
940 | if (status != -ENOENT) | |
941 | mlog_errno(status); | |
942 | ocfs2_cleanup_delete_inode(inode, 0); | |
943 | goto bail_unblock; | |
944 | } | |
945 | ||
946 | /* Query the cluster. This will be the final decision made | |
947 | * before we go ahead and wipe the inode. */ | |
948 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); | |
949 | if (!wipe || status < 0) { | |
950 | /* Error and inode busy vote both mean we won't be | |
951 | * removing the inode, so they take almost the same | |
952 | * path. */ | |
953 | if (status < 0) | |
954 | mlog_errno(status); | |
955 | ||
956 | /* Someone in the cluster has voted to not wipe this | |
957 | * inode, or it was never completely orphaned. Write | |
958 | * out the pages and exit now. */ | |
959 | ocfs2_cleanup_delete_inode(inode, 1); | |
960 | goto bail_unlock_inode; | |
961 | } | |
962 | ||
963 | ocfs2_cleanup_delete_inode(inode, 0); | |
964 | ||
965 | status = ocfs2_wipe_inode(inode, di_bh); | |
966 | if (status < 0) { | |
b4df6ed8 MF |
967 | if (status != -EDEADLK) |
968 | mlog_errno(status); | |
ccd979bd MF |
969 | goto bail_unlock_inode; |
970 | } | |
971 | ||
24c19ef4 MF |
972 | /* |
973 | * Mark the inode as successfully deleted. | |
974 | * | |
975 | * This is important for ocfs2_clear_inode() as it will check | |
976 | * this flag and skip any checkpointing work | |
977 | * | |
978 | * ocfs2_stuff_meta_lvb() also uses this flag to invalidate | |
979 | * the LVB for other nodes. | |
980 | */ | |
ccd979bd MF |
981 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; |
982 | ||
983 | bail_unlock_inode: | |
984 | ocfs2_meta_unlock(inode, 1); | |
985 | brelse(di_bh); | |
986 | bail_unblock: | |
987 | status = sigprocmask(SIG_SETMASK, &oldset, NULL); | |
988 | if (status < 0) | |
989 | mlog_errno(status); | |
990 | bail: | |
991 | clear_inode(inode); | |
992 | mlog_exit_void(); | |
993 | } | |
994 | ||
995 | void ocfs2_clear_inode(struct inode *inode) | |
996 | { | |
997 | int status; | |
998 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
999 | ||
1000 | mlog_entry_void(); | |
1001 | ||
1002 | if (!inode) | |
1003 | goto bail; | |
1004 | ||
b0697053 MF |
1005 | mlog(0, "Clearing inode: %llu, nlink = %u\n", |
1006 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
1007 | |
1008 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, | |
1009 | "Inode=%lu\n", inode->i_ino); | |
1010 | ||
50008630 TY |
1011 | /* For remove delete_inode vote, we hold open lock before, |
1012 | * now it is time to unlock PR and EX open locks. */ | |
1013 | ocfs2_open_unlock(inode); | |
1014 | ||
ccd979bd MF |
1015 | /* Do these before all the other work so that we don't bounce |
1016 | * the vote thread while waiting to destroy the locks. */ | |
1017 | ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres); | |
1018 | ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres); | |
1019 | ocfs2_mark_lockres_freeing(&oi->ip_data_lockres); | |
50008630 | 1020 | ocfs2_mark_lockres_freeing(&oi->ip_open_lockres); |
ccd979bd MF |
1021 | |
1022 | /* We very well may get a clear_inode before all an inodes | |
1023 | * metadata has hit disk. Of course, we can't drop any cluster | |
1024 | * locks until the journal has finished with it. The only | |
1025 | * exception here are successfully wiped inodes - their | |
1026 | * metadata can now be considered to be part of the system | |
1027 | * inodes from which it came. */ | |
1028 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) | |
1029 | ocfs2_checkpoint_inode(inode); | |
1030 | ||
1031 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), | |
b0697053 MF |
1032 | "Clear inode of %llu, inode has io markers\n", |
1033 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1034 | |
83418978 MF |
1035 | ocfs2_extent_map_trunc(inode, 0); |
1036 | ||
ccd979bd MF |
1037 | status = ocfs2_drop_inode_locks(inode); |
1038 | if (status < 0) | |
1039 | mlog_errno(status); | |
1040 | ||
1041 | ocfs2_lock_res_free(&oi->ip_rw_lockres); | |
1042 | ocfs2_lock_res_free(&oi->ip_meta_lockres); | |
1043 | ocfs2_lock_res_free(&oi->ip_data_lockres); | |
50008630 | 1044 | ocfs2_lock_res_free(&oi->ip_open_lockres); |
ccd979bd MF |
1045 | |
1046 | ocfs2_metadata_cache_purge(inode); | |
1047 | ||
1048 | mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached, | |
b0697053 MF |
1049 | "Clear inode of %llu, inode has %u cache items\n", |
1050 | (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached); | |
ccd979bd MF |
1051 | |
1052 | mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE), | |
b0697053 MF |
1053 | "Clear inode of %llu, inode has a bad flag\n", |
1054 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1055 | |
1056 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), | |
b0697053 MF |
1057 | "Clear inode of %llu, inode is locked\n", |
1058 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1059 | |
251b6ecc | 1060 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
b0697053 MF |
1061 | "Clear inode of %llu, io_mutex is locked\n", |
1062 | (unsigned long long)oi->ip_blkno); | |
251b6ecc | 1063 | mutex_unlock(&oi->ip_io_mutex); |
ccd979bd MF |
1064 | |
1065 | /* | |
1066 | * down_trylock() returns 0, down_write_trylock() returns 1 | |
1067 | * kernel 1, world 0 | |
1068 | */ | |
1069 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), | |
b0697053 MF |
1070 | "Clear inode of %llu, alloc_sem is locked\n", |
1071 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1072 | up_write(&oi->ip_alloc_sem); |
1073 | ||
1074 | mlog_bug_on_msg(oi->ip_open_count, | |
b0697053 MF |
1075 | "Clear inode of %llu has open count %d\n", |
1076 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); | |
ccd979bd MF |
1077 | |
1078 | /* Clear all other flags. */ | |
1079 | oi->ip_flags = OCFS2_INODE_CACHE_INLINE; | |
1080 | oi->ip_created_trans = 0; | |
1081 | oi->ip_last_trans = 0; | |
1082 | oi->ip_dir_start_lookup = 0; | |
1083 | oi->ip_blkno = 0ULL; | |
1084 | ||
1085 | bail: | |
1086 | mlog_exit_void(); | |
1087 | } | |
1088 | ||
1089 | /* Called under inode_lock, with no more references on the | |
1090 | * struct inode, so it's safe here to check the flags field | |
1091 | * and to manipulate i_nlink without any other locks. */ | |
1092 | void ocfs2_drop_inode(struct inode *inode) | |
1093 | { | |
1094 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1095 | ||
1096 | mlog_entry_void(); | |
1097 | ||
b0697053 MF |
1098 | mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n", |
1099 | (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags); | |
ccd979bd | 1100 | |
379dfe9d MF |
1101 | if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) |
1102 | generic_delete_inode(inode); | |
1103 | else | |
1104 | generic_drop_inode(inode); | |
ccd979bd MF |
1105 | |
1106 | mlog_exit_void(); | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * TODO: this should probably be merged into ocfs2_get_block | |
1111 | * | |
1112 | * However, you now need to pay attention to the cont_prepare_write() | |
1113 | * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much | |
1114 | * expects never to extend). | |
1115 | */ | |
1116 | struct buffer_head *ocfs2_bread(struct inode *inode, | |
1117 | int block, int *err, int reada) | |
1118 | { | |
1119 | struct buffer_head *bh = NULL; | |
1120 | int tmperr; | |
1121 | u64 p_blkno; | |
1122 | int readflags = OCFS2_BH_CACHED; | |
1123 | ||
ccd979bd MF |
1124 | if (reada) |
1125 | readflags |= OCFS2_BH_READAHEAD; | |
ccd979bd MF |
1126 | |
1127 | if (((u64)block << inode->i_sb->s_blocksize_bits) >= | |
1128 | i_size_read(inode)) { | |
1129 | BUG_ON(!reada); | |
1130 | return NULL; | |
1131 | } | |
1132 | ||
ee19a779 | 1133 | down_read(&OCFS2_I(inode)->ip_alloc_sem); |
49cb8d2d MF |
1134 | tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, |
1135 | NULL); | |
ee19a779 | 1136 | up_read(&OCFS2_I(inode)->ip_alloc_sem); |
ccd979bd MF |
1137 | if (tmperr < 0) { |
1138 | mlog_errno(tmperr); | |
1139 | goto fail; | |
1140 | } | |
1141 | ||
1142 | tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh, | |
1143 | readflags, inode); | |
1144 | if (tmperr < 0) | |
1145 | goto fail; | |
1146 | ||
1147 | tmperr = 0; | |
1148 | ||
1149 | *err = 0; | |
1150 | return bh; | |
1151 | ||
1152 | fail: | |
1153 | if (bh) { | |
1154 | brelse(bh); | |
1155 | bh = NULL; | |
1156 | } | |
1157 | *err = -EIO; | |
1158 | return NULL; | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * This is called from our getattr. | |
1163 | */ | |
1164 | int ocfs2_inode_revalidate(struct dentry *dentry) | |
1165 | { | |
1166 | struct inode *inode = dentry->d_inode; | |
1167 | int status = 0; | |
1168 | ||
b0697053 MF |
1169 | mlog_entry("(inode = 0x%p, ino = %llu)\n", inode, |
1170 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL); | |
ccd979bd MF |
1171 | |
1172 | if (!inode) { | |
1173 | mlog(0, "eep, no inode!\n"); | |
1174 | status = -ENOENT; | |
1175 | goto bail; | |
1176 | } | |
1177 | ||
1178 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1179 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
1180 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1181 | mlog(0, "inode deleted!\n"); | |
1182 | status = -ENOENT; | |
1183 | goto bail; | |
1184 | } | |
1185 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1186 | ||
1187 | /* Let ocfs2_meta_lock do the work of updating our struct | |
1188 | * inode for us. */ | |
4bcec184 | 1189 | status = ocfs2_meta_lock(inode, NULL, 0); |
ccd979bd MF |
1190 | if (status < 0) { |
1191 | if (status != -ENOENT) | |
1192 | mlog_errno(status); | |
1193 | goto bail; | |
1194 | } | |
1195 | ocfs2_meta_unlock(inode, 0); | |
1196 | bail: | |
1197 | mlog_exit(status); | |
1198 | ||
1199 | return status; | |
1200 | } | |
1201 | ||
1202 | /* | |
1203 | * Updates a disk inode from a | |
1204 | * struct inode. | |
1205 | * Only takes ip_lock. | |
1206 | */ | |
1fabe148 | 1207 | int ocfs2_mark_inode_dirty(handle_t *handle, |
ccd979bd MF |
1208 | struct inode *inode, |
1209 | struct buffer_head *bh) | |
1210 | { | |
1211 | int status; | |
1212 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
1213 | ||
b0697053 MF |
1214 | mlog_entry("(inode %llu)\n", |
1215 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
1216 | |
1217 | status = ocfs2_journal_access(handle, inode, bh, | |
1218 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1219 | if (status < 0) { | |
1220 | mlog_errno(status); | |
1221 | goto leave; | |
1222 | } | |
1223 | ||
1224 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1225 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
6e4b0d56 | 1226 | ocfs2_get_inode_flags(OCFS2_I(inode)); |
ca4d147e | 1227 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
15b1e36b | 1228 | fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features); |
ccd979bd MF |
1229 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1230 | ||
1231 | fe->i_size = cpu_to_le64(i_size_read(inode)); | |
1232 | fe->i_links_count = cpu_to_le16(inode->i_nlink); | |
1233 | fe->i_uid = cpu_to_le32(inode->i_uid); | |
1234 | fe->i_gid = cpu_to_le32(inode->i_gid); | |
1235 | fe->i_mode = cpu_to_le16(inode->i_mode); | |
1236 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); | |
1237 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
1238 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
1239 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
1240 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1241 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
1242 | ||
1243 | status = ocfs2_journal_dirty(handle, bh); | |
1244 | if (status < 0) | |
1245 | mlog_errno(status); | |
1246 | ||
1247 | status = 0; | |
1248 | leave: | |
1249 | ||
1250 | mlog_exit(status); | |
1251 | return status; | |
1252 | } | |
1253 | ||
1254 | /* | |
1255 | * | |
1256 | * Updates a struct inode from a disk inode. | |
1257 | * does no i/o, only takes ip_lock. | |
1258 | */ | |
1259 | void ocfs2_refresh_inode(struct inode *inode, | |
1260 | struct ocfs2_dinode *fe) | |
1261 | { | |
ccd979bd MF |
1262 | spin_lock(&OCFS2_I(inode)->ip_lock); |
1263 | ||
1264 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e | 1265 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
15b1e36b | 1266 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
ca4d147e | 1267 | ocfs2_set_inode_flags(inode); |
ccd979bd MF |
1268 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
1269 | inode->i_nlink = le16_to_cpu(fe->i_links_count); | |
1270 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
1271 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
1272 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
ccd979bd MF |
1273 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) |
1274 | inode->i_blocks = 0; | |
1275 | else | |
8110b073 | 1276 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
1277 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
1278 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
1279 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
1280 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
1281 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
1282 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
1283 | ||
1284 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1285 | } |