]>
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" | |
ccd979bd MF |
52 | |
53 | #include "buffer_head_io.h" | |
54 | ||
ccd979bd MF |
55 | struct ocfs2_find_inode_args |
56 | { | |
57 | u64 fi_blkno; | |
58 | unsigned long fi_ino; | |
59 | unsigned int fi_flags; | |
60 | }; | |
61 | ||
62 | static int ocfs2_read_locked_inode(struct inode *inode, | |
63 | struct ocfs2_find_inode_args *args); | |
64 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); | |
65 | static int ocfs2_find_actor(struct inode *inode, void *opaque); | |
66 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
67 | struct inode *inode, | |
68 | struct buffer_head *fe_bh); | |
69 | ||
ca4d147e HP |
70 | void ocfs2_set_inode_flags(struct inode *inode) |
71 | { | |
72 | unsigned int flags = OCFS2_I(inode)->ip_attr; | |
73 | ||
74 | inode->i_flags &= ~(S_IMMUTABLE | | |
75 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); | |
76 | ||
77 | if (flags & OCFS2_IMMUTABLE_FL) | |
78 | inode->i_flags |= S_IMMUTABLE; | |
79 | ||
80 | if (flags & OCFS2_SYNC_FL) | |
81 | inode->i_flags |= S_SYNC; | |
82 | if (flags & OCFS2_APPEND_FL) | |
83 | inode->i_flags |= S_APPEND; | |
84 | if (flags & OCFS2_NOATIME_FL) | |
85 | inode->i_flags |= S_NOATIME; | |
86 | if (flags & OCFS2_DIRSYNC_FL) | |
87 | inode->i_flags |= S_DIRSYNC; | |
88 | } | |
89 | ||
6e4b0d56 JK |
90 | /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */ |
91 | void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi) | |
92 | { | |
93 | unsigned int flags = oi->vfs_inode.i_flags; | |
94 | ||
95 | oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL| | |
96 | OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL); | |
97 | if (flags & S_SYNC) | |
98 | oi->ip_attr |= OCFS2_SYNC_FL; | |
99 | if (flags & S_APPEND) | |
100 | oi->ip_attr |= OCFS2_APPEND_FL; | |
101 | if (flags & S_IMMUTABLE) | |
102 | oi->ip_attr |= OCFS2_IMMUTABLE_FL; | |
103 | if (flags & S_NOATIME) | |
104 | oi->ip_attr |= OCFS2_NOATIME_FL; | |
105 | if (flags & S_DIRSYNC) | |
106 | oi->ip_attr |= OCFS2_DIRSYNC_FL; | |
107 | } | |
108 | ||
24c19ef4 | 109 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags) |
ccd979bd MF |
110 | { |
111 | struct inode *inode = NULL; | |
112 | struct super_block *sb = osb->sb; | |
113 | struct ocfs2_find_inode_args args; | |
114 | ||
b0697053 | 115 | mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno); |
ccd979bd MF |
116 | |
117 | /* Ok. By now we've either got the offsets passed to us by the | |
118 | * caller, or we just pulled them off the bh. Lets do some | |
119 | * sanity checks to make sure they're OK. */ | |
120 | if (blkno == 0) { | |
121 | inode = ERR_PTR(-EINVAL); | |
122 | mlog_errno(PTR_ERR(inode)); | |
123 | goto bail; | |
124 | } | |
125 | ||
126 | args.fi_blkno = blkno; | |
24c19ef4 | 127 | args.fi_flags = flags; |
ccd979bd MF |
128 | args.fi_ino = ino_from_blkno(sb, blkno); |
129 | ||
130 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, | |
131 | ocfs2_init_locked_inode, &args); | |
132 | /* inode was *not* in the inode cache. 2.6.x requires | |
133 | * us to do our own read_inode call and unlock it | |
134 | * afterwards. */ | |
135 | if (inode && inode->i_state & I_NEW) { | |
136 | mlog(0, "Inode was not in inode cache, reading it.\n"); | |
137 | ocfs2_read_locked_inode(inode, &args); | |
138 | unlock_new_inode(inode); | |
139 | } | |
140 | if (inode == NULL) { | |
141 | inode = ERR_PTR(-ENOMEM); | |
142 | mlog_errno(PTR_ERR(inode)); | |
143 | goto bail; | |
144 | } | |
145 | if (is_bad_inode(inode)) { | |
146 | iput(inode); | |
147 | inode = ERR_PTR(-ESTALE); | |
ccd979bd MF |
148 | goto bail; |
149 | } | |
150 | ||
151 | bail: | |
152 | if (!IS_ERR(inode)) { | |
b0697053 MF |
153 | mlog(0, "returning inode with number %llu\n", |
154 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 155 | mlog_exit_ptr(inode); |
6a1bd4a5 | 156 | } |
ccd979bd MF |
157 | |
158 | return inode; | |
159 | } | |
160 | ||
161 | ||
162 | /* | |
163 | * here's how inodes get read from disk: | |
164 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR | |
165 | * found? : return the in-memory inode | |
166 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE | |
167 | */ | |
168 | ||
169 | static int ocfs2_find_actor(struct inode *inode, void *opaque) | |
170 | { | |
171 | struct ocfs2_find_inode_args *args = NULL; | |
172 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
173 | int ret = 0; | |
174 | ||
175 | mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque); | |
176 | ||
177 | args = opaque; | |
178 | ||
179 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); | |
180 | ||
181 | if (oi->ip_blkno != args->fi_blkno) | |
182 | goto bail; | |
183 | ||
ccd979bd MF |
184 | ret = 1; |
185 | bail: | |
186 | mlog_exit(ret); | |
187 | return ret; | |
188 | } | |
189 | ||
190 | /* | |
191 | * initialize the new inode, but don't do anything that would cause | |
192 | * us to sleep. | |
193 | * return 0 on success, 1 on failure | |
194 | */ | |
195 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) | |
196 | { | |
197 | struct ocfs2_find_inode_args *args = opaque; | |
198 | ||
199 | mlog_entry("inode = %p, opaque = %p\n", inode, opaque); | |
200 | ||
201 | inode->i_ino = args->fi_ino; | |
202 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; | |
203 | ||
204 | mlog_exit(0); | |
205 | return 0; | |
206 | } | |
207 | ||
208 | int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, | |
209 | int create_ino) | |
210 | { | |
211 | struct super_block *sb; | |
212 | struct ocfs2_super *osb; | |
213 | int status = -EINVAL; | |
214 | ||
b0697053 | 215 | mlog_entry("(0x%p, size:%llu)\n", inode, |
1ca1a111 | 216 | (unsigned long long)le64_to_cpu(fe->i_size)); |
ccd979bd MF |
217 | |
218 | sb = inode->i_sb; | |
219 | osb = OCFS2_SB(sb); | |
220 | ||
221 | /* this means that read_inode cannot create a superblock inode | |
222 | * today. change if needed. */ | |
223 | if (!OCFS2_IS_VALID_DINODE(fe) || | |
224 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
6a1bd4a5 | 225 | mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, " |
ccd979bd | 226 | "signature = %.*s, flags = 0x%x\n", |
b0697053 MF |
227 | inode->i_ino, |
228 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, | |
ccd979bd MF |
229 | fe->i_signature, le32_to_cpu(fe->i_flags)); |
230 | goto bail; | |
231 | } | |
232 | ||
233 | if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) { | |
234 | mlog(ML_ERROR, "file entry generation does not match " | |
235 | "superblock! osb->fs_generation=%x, " | |
236 | "fe->i_fs_generation=%x\n", | |
237 | osb->fs_generation, le32_to_cpu(fe->i_fs_generation)); | |
238 | goto bail; | |
239 | } | |
240 | ||
8110b073 MF |
241 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
242 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); | |
15b1e36b | 243 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
8110b073 | 244 | |
ccd979bd MF |
245 | inode->i_version = 1; |
246 | inode->i_generation = le32_to_cpu(fe->i_generation); | |
247 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
248 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
249 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
250 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
ccd979bd MF |
251 | |
252 | /* Fast symlinks will have i_size but no allocated clusters. */ | |
253 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) | |
254 | inode->i_blocks = 0; | |
255 | else | |
8110b073 | 256 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 257 | inode->i_mapping->a_ops = &ocfs2_aops; |
ccd979bd MF |
258 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
259 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
260 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
261 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
262 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
263 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
264 | ||
265 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) | |
266 | mlog(ML_ERROR, | |
b0697053 MF |
267 | "ip_blkno %llu != i_blkno %llu!\n", |
268 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1ca1a111 | 269 | (unsigned long long)le64_to_cpu(fe->i_blkno)); |
ccd979bd | 270 | |
ccd979bd MF |
271 | inode->i_nlink = le16_to_cpu(fe->i_links_count); |
272 | ||
24c19ef4 MF |
273 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) |
274 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; | |
275 | ||
ccd979bd MF |
276 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { |
277 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
278 | mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino); | |
279 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { | |
280 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
281 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { | |
282 | mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino); | |
283 | /* we can't actually hit this as read_inode can't | |
284 | * handle superblocks today ;-) */ | |
285 | BUG(); | |
286 | } | |
287 | ||
288 | switch (inode->i_mode & S_IFMT) { | |
289 | case S_IFREG: | |
290 | inode->i_fop = &ocfs2_fops; | |
291 | inode->i_op = &ocfs2_file_iops; | |
292 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
293 | break; | |
294 | case S_IFDIR: | |
295 | inode->i_op = &ocfs2_dir_iops; | |
296 | inode->i_fop = &ocfs2_dops; | |
297 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
298 | break; | |
299 | case S_IFLNK: | |
300 | if (ocfs2_inode_is_fast_symlink(inode)) | |
301 | inode->i_op = &ocfs2_fast_symlink_inode_operations; | |
302 | else | |
303 | inode->i_op = &ocfs2_symlink_inode_operations; | |
304 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
305 | break; | |
306 | default: | |
307 | inode->i_op = &ocfs2_special_file_iops; | |
308 | init_special_inode(inode, inode->i_mode, | |
309 | inode->i_rdev); | |
310 | break; | |
311 | } | |
312 | ||
24c19ef4 MF |
313 | if (create_ino) { |
314 | inode->i_ino = ino_from_blkno(inode->i_sb, | |
315 | le64_to_cpu(fe->i_blkno)); | |
316 | ||
317 | /* | |
318 | * If we ever want to create system files from kernel, | |
319 | * the generation argument to | |
320 | * ocfs2_inode_lock_res_init() will have to change. | |
321 | */ | |
1ca1a111 | 322 | BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL); |
24c19ef4 MF |
323 | |
324 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
325 | OCFS2_LOCK_TYPE_META, 0, inode); | |
50008630 TY |
326 | |
327 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, | |
328 | OCFS2_LOCK_TYPE_OPEN, 0, inode); | |
24c19ef4 MF |
329 | } |
330 | ||
ccd979bd | 331 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, |
24c19ef4 MF |
332 | OCFS2_LOCK_TYPE_RW, inode->i_generation, |
333 | inode); | |
334 | ||
ca4d147e | 335 | ocfs2_set_inode_flags(inode); |
ca4d147e | 336 | |
ccd979bd MF |
337 | status = 0; |
338 | bail: | |
339 | mlog_exit(status); | |
340 | return status; | |
341 | } | |
342 | ||
343 | static int ocfs2_read_locked_inode(struct inode *inode, | |
344 | struct ocfs2_find_inode_args *args) | |
345 | { | |
346 | struct super_block *sb; | |
347 | struct ocfs2_super *osb; | |
348 | struct ocfs2_dinode *fe; | |
349 | struct buffer_head *bh = NULL; | |
24c19ef4 MF |
350 | int status, can_lock; |
351 | u32 generation = 0; | |
ccd979bd MF |
352 | |
353 | mlog_entry("(0x%p, 0x%p)\n", inode, args); | |
354 | ||
355 | status = -EINVAL; | |
356 | if (inode == NULL || inode->i_sb == NULL) { | |
357 | mlog(ML_ERROR, "bad inode\n"); | |
24c19ef4 | 358 | return status; |
ccd979bd MF |
359 | } |
360 | sb = inode->i_sb; | |
361 | osb = OCFS2_SB(sb); | |
362 | ||
363 | if (!args) { | |
364 | mlog(ML_ERROR, "bad inode args\n"); | |
365 | make_bad_inode(inode); | |
24c19ef4 MF |
366 | return status; |
367 | } | |
368 | ||
369 | /* | |
370 | * To improve performance of cold-cache inode stats, we take | |
371 | * the cluster lock here if possible. | |
372 | * | |
373 | * Generally, OCFS2 never trusts the contents of an inode | |
374 | * unless it's holding a cluster lock, so taking it here isn't | |
375 | * a correctness issue as much as it is a performance | |
376 | * improvement. | |
377 | * | |
378 | * There are three times when taking the lock is not a good idea: | |
379 | * | |
380 | * 1) During startup, before we have initialized the DLM. | |
381 | * | |
382 | * 2) If we are reading certain system files which never get | |
383 | * cluster locks (local alloc, truncate log). | |
384 | * | |
385 | * 3) If the process doing the iget() is responsible for | |
386 | * orphan dir recovery. We're holding the orphan dir lock and | |
387 | * can get into a deadlock with another process on another | |
388 | * node in ->delete_inode(). | |
389 | * | |
390 | * #1 and #2 can be simply solved by never taking the lock | |
391 | * here for system files (which are the only type we read | |
392 | * during mount). It's a heavier approach, but our main | |
393 | * concern is user-accesible files anyway. | |
394 | * | |
395 | * #3 works itself out because we'll eventually take the | |
396 | * cluster lock before trusting anything anyway. | |
397 | */ | |
398 | can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
50008630 | 399 | && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) |
c271c5c2 | 400 | && !ocfs2_mount_local(osb); |
24c19ef4 MF |
401 | |
402 | /* | |
403 | * To maintain backwards compatibility with older versions of | |
404 | * ocfs2-tools, we still store the generation value for system | |
405 | * files. The only ones that actually matter to userspace are | |
406 | * the journals, but it's easier and inexpensive to just flag | |
407 | * all system files similarly. | |
408 | */ | |
409 | if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
410 | generation = osb->fs_generation; | |
411 | ||
412 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
413 | OCFS2_LOCK_TYPE_META, | |
414 | generation, inode); | |
415 | ||
50008630 TY |
416 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, |
417 | OCFS2_LOCK_TYPE_OPEN, | |
418 | 0, inode); | |
419 | ||
24c19ef4 | 420 | if (can_lock) { |
50008630 TY |
421 | status = ocfs2_open_lock(inode); |
422 | if (status) { | |
423 | make_bad_inode(inode); | |
424 | mlog_errno(status); | |
425 | return status; | |
426 | } | |
4bcec184 | 427 | status = ocfs2_meta_lock(inode, NULL, 0); |
24c19ef4 MF |
428 | if (status) { |
429 | make_bad_inode(inode); | |
430 | mlog_errno(status); | |
431 | return status; | |
432 | } | |
ccd979bd MF |
433 | } |
434 | ||
50008630 TY |
435 | if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) { |
436 | status = ocfs2_try_open_lock(inode, 0); | |
437 | if (status) { | |
438 | make_bad_inode(inode); | |
439 | return status; | |
440 | } | |
441 | } | |
442 | ||
24c19ef4 MF |
443 | status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, |
444 | can_lock ? inode : NULL); | |
ccd979bd MF |
445 | if (status < 0) { |
446 | mlog_errno(status); | |
ccd979bd MF |
447 | goto bail; |
448 | } | |
449 | ||
24c19ef4 | 450 | status = -EINVAL; |
ccd979bd MF |
451 | fe = (struct ocfs2_dinode *) bh->b_data; |
452 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
a46043e0 MF |
453 | mlog(0, "Invalid dinode #%llu: signature = %.*s\n", |
454 | (unsigned long long)args->fi_blkno, 7, | |
1ca1a111 | 455 | fe->i_signature); |
ccd979bd MF |
456 | goto bail; |
457 | } | |
458 | ||
24c19ef4 MF |
459 | /* |
460 | * This is a code bug. Right now the caller needs to | |
461 | * understand whether it is asking for a system file inode or | |
462 | * not so the proper lock names can be built. | |
463 | */ | |
464 | mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) != | |
465 | !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE), | |
466 | "Inode %llu: system file state is ambigous\n", | |
467 | (unsigned long long)args->fi_blkno); | |
ccd979bd MF |
468 | |
469 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || | |
470 | S_ISBLK(le16_to_cpu(fe->i_mode))) | |
471 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
472 | ||
6a1bd4a5 | 473 | if (ocfs2_populate_inode(inode, fe, 0) < 0) |
ccd979bd | 474 | goto bail; |
ccd979bd MF |
475 | |
476 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); | |
477 | ||
ccd979bd MF |
478 | status = 0; |
479 | ||
480 | bail: | |
24c19ef4 MF |
481 | if (can_lock) |
482 | ocfs2_meta_unlock(inode, 0); | |
483 | ||
484 | if (status < 0) | |
485 | make_bad_inode(inode); | |
486 | ||
ccd979bd MF |
487 | if (args && bh) |
488 | brelse(bh); | |
489 | ||
490 | mlog_exit(status); | |
491 | return status; | |
492 | } | |
493 | ||
494 | void ocfs2_sync_blockdev(struct super_block *sb) | |
495 | { | |
496 | sync_blockdev(sb->s_bdev); | |
497 | } | |
498 | ||
499 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
500 | struct inode *inode, | |
501 | struct buffer_head *fe_bh) | |
502 | { | |
503 | int status = 0; | |
ccd979bd MF |
504 | struct ocfs2_truncate_context *tc = NULL; |
505 | struct ocfs2_dinode *fe; | |
60b11392 | 506 | handle_t *handle = NULL; |
ccd979bd MF |
507 | |
508 | mlog_entry_void(); | |
509 | ||
510 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
511 | ||
1afc32b9 MF |
512 | /* |
513 | * This check will also skip truncate of inodes with inline | |
514 | * data and fast symlinks. | |
515 | */ | |
3a0782d0 | 516 | if (fe->i_clusters) { |
60b11392 MF |
517 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
518 | if (IS_ERR(handle)) { | |
519 | status = PTR_ERR(handle); | |
520 | mlog_errno(status); | |
521 | goto out; | |
522 | } | |
523 | ||
524 | status = ocfs2_journal_access(handle, inode, fe_bh, | |
525 | OCFS2_JOURNAL_ACCESS_WRITE); | |
526 | if (status < 0) { | |
527 | mlog_errno(status); | |
528 | goto out; | |
529 | } | |
530 | ||
531 | i_size_write(inode, 0); | |
532 | ||
533 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
534 | if (status < 0) { | |
535 | mlog_errno(status); | |
536 | goto out; | |
537 | } | |
538 | ||
539 | ocfs2_commit_trans(osb, handle); | |
540 | handle = NULL; | |
541 | ||
3a0782d0 MF |
542 | status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc); |
543 | if (status < 0) { | |
544 | mlog_errno(status); | |
545 | goto out; | |
546 | } | |
ccd979bd | 547 | |
3a0782d0 MF |
548 | status = ocfs2_commit_truncate(osb, inode, fe_bh, tc); |
549 | if (status < 0) { | |
550 | mlog_errno(status); | |
551 | goto out; | |
552 | } | |
ccd979bd | 553 | } |
ccd979bd | 554 | |
60b11392 MF |
555 | out: |
556 | if (handle) | |
557 | ocfs2_commit_trans(osb, handle); | |
ccd979bd MF |
558 | mlog_exit(status); |
559 | return status; | |
560 | } | |
561 | ||
562 | static int ocfs2_remove_inode(struct inode *inode, | |
563 | struct buffer_head *di_bh, | |
564 | struct inode *orphan_dir_inode, | |
565 | struct buffer_head *orphan_dir_bh) | |
566 | { | |
567 | int status; | |
568 | struct inode *inode_alloc_inode = NULL; | |
569 | struct buffer_head *inode_alloc_bh = NULL; | |
1fabe148 | 570 | handle_t *handle; |
ccd979bd MF |
571 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
572 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
573 | ||
574 | inode_alloc_inode = | |
575 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
576 | le16_to_cpu(di->i_suballoc_slot)); | |
577 | if (!inode_alloc_inode) { | |
578 | status = -EEXIST; | |
579 | mlog_errno(status); | |
580 | goto bail; | |
581 | } | |
582 | ||
1b1dcc1b | 583 | mutex_lock(&inode_alloc_inode->i_mutex); |
4bcec184 | 584 | status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1); |
ccd979bd | 585 | if (status < 0) { |
1b1dcc1b | 586 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
587 | |
588 | mlog_errno(status); | |
589 | goto bail; | |
590 | } | |
591 | ||
65eff9cc | 592 | handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS); |
ccd979bd MF |
593 | if (IS_ERR(handle)) { |
594 | status = PTR_ERR(handle); | |
595 | mlog_errno(status); | |
596 | goto bail_unlock; | |
597 | } | |
598 | ||
599 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, | |
600 | orphan_dir_bh); | |
601 | if (status < 0) { | |
602 | mlog_errno(status); | |
603 | goto bail_commit; | |
604 | } | |
605 | ||
606 | /* set the inodes dtime */ | |
607 | status = ocfs2_journal_access(handle, inode, di_bh, | |
608 | OCFS2_JOURNAL_ACCESS_WRITE); | |
609 | if (status < 0) { | |
610 | mlog_errno(status); | |
611 | goto bail_commit; | |
612 | } | |
613 | ||
614 | di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec); | |
615 | le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); | |
616 | ||
617 | status = ocfs2_journal_dirty(handle, di_bh); | |
618 | if (status < 0) { | |
619 | mlog_errno(status); | |
620 | goto bail_commit; | |
621 | } | |
622 | ||
623 | ocfs2_remove_from_cache(inode, di_bh); | |
624 | ||
625 | status = ocfs2_free_dinode(handle, inode_alloc_inode, | |
626 | inode_alloc_bh, di); | |
627 | if (status < 0) | |
628 | mlog_errno(status); | |
629 | ||
630 | bail_commit: | |
02dc1af4 | 631 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
632 | bail_unlock: |
633 | ocfs2_meta_unlock(inode_alloc_inode, 1); | |
1b1dcc1b | 634 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
635 | brelse(inode_alloc_bh); |
636 | bail: | |
637 | iput(inode_alloc_inode); | |
638 | ||
639 | return status; | |
640 | } | |
641 | ||
b4df6ed8 MF |
642 | /* |
643 | * Serialize with orphan dir recovery. If the process doing | |
644 | * recovery on this orphan dir does an iget() with the dir | |
645 | * i_mutex held, we'll deadlock here. Instead we detect this | |
646 | * and exit early - recovery will wipe this inode for us. | |
647 | */ | |
648 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, | |
649 | int slot) | |
650 | { | |
651 | int ret = 0; | |
652 | ||
653 | spin_lock(&osb->osb_lock); | |
654 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { | |
655 | mlog(0, "Recovery is happening on orphan dir %d, will skip " | |
656 | "this inode\n", slot); | |
657 | ret = -EDEADLK; | |
658 | goto out; | |
659 | } | |
660 | /* This signals to the orphan recovery process that it should | |
661 | * wait for us to handle the wipe. */ | |
662 | osb->osb_orphan_wipes[slot]++; | |
663 | out: | |
664 | spin_unlock(&osb->osb_lock); | |
665 | return ret; | |
666 | } | |
667 | ||
668 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, | |
669 | int slot) | |
670 | { | |
671 | spin_lock(&osb->osb_lock); | |
672 | osb->osb_orphan_wipes[slot]--; | |
673 | spin_unlock(&osb->osb_lock); | |
674 | ||
675 | wake_up(&osb->osb_wipe_event); | |
676 | } | |
677 | ||
ccd979bd MF |
678 | static int ocfs2_wipe_inode(struct inode *inode, |
679 | struct buffer_head *di_bh) | |
680 | { | |
681 | int status, orphaned_slot; | |
682 | struct inode *orphan_dir_inode = NULL; | |
683 | struct buffer_head *orphan_dir_bh = NULL; | |
684 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
50008630 | 685 | struct ocfs2_dinode *di; |
ccd979bd | 686 | |
50008630 TY |
687 | di = (struct ocfs2_dinode *) di_bh->b_data; |
688 | orphaned_slot = le16_to_cpu(di->i_orphaned_slot); | |
b4df6ed8 MF |
689 | |
690 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); | |
691 | if (status) | |
692 | return status; | |
693 | ||
ccd979bd MF |
694 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
695 | ORPHAN_DIR_SYSTEM_INODE, | |
696 | orphaned_slot); | |
697 | if (!orphan_dir_inode) { | |
698 | status = -EEXIST; | |
699 | mlog_errno(status); | |
700 | goto bail; | |
701 | } | |
702 | ||
703 | /* Lock the orphan dir. The lock will be held for the entire | |
704 | * delete_inode operation. We do this now to avoid races with | |
705 | * recovery completion on other nodes. */ | |
1b1dcc1b | 706 | mutex_lock(&orphan_dir_inode->i_mutex); |
4bcec184 | 707 | status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1); |
ccd979bd | 708 | if (status < 0) { |
1b1dcc1b | 709 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
710 | |
711 | mlog_errno(status); | |
712 | goto bail; | |
713 | } | |
714 | ||
715 | /* we do this while holding the orphan dir lock because we | |
34d024f8 MF |
716 | * don't want recovery being run from another node to try an |
717 | * inode delete underneath us -- this will result in two nodes | |
ccd979bd MF |
718 | * truncating the same file! */ |
719 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); | |
720 | if (status < 0) { | |
721 | mlog_errno(status); | |
722 | goto bail_unlock_dir; | |
723 | } | |
724 | ||
725 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, | |
726 | orphan_dir_bh); | |
727 | if (status < 0) | |
728 | mlog_errno(status); | |
729 | ||
730 | bail_unlock_dir: | |
731 | ocfs2_meta_unlock(orphan_dir_inode, 1); | |
1b1dcc1b | 732 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
733 | brelse(orphan_dir_bh); |
734 | bail: | |
735 | iput(orphan_dir_inode); | |
b4df6ed8 | 736 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
ccd979bd MF |
737 | |
738 | return status; | |
739 | } | |
740 | ||
741 | /* There is a series of simple checks that should be done before a | |
34d024f8 | 742 | * trylock is even considered. Encapsulate those in this function. */ |
ccd979bd MF |
743 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) |
744 | { | |
745 | int ret = 0; | |
746 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
747 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
748 | ||
749 | /* We shouldn't be getting here for the root directory | |
750 | * inode.. */ | |
751 | if (inode == osb->root_inode) { | |
752 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); | |
753 | goto bail; | |
754 | } | |
755 | ||
34d024f8 | 756 | /* If we're coming from downconvert_thread we can't go into our own |
ccd979bd MF |
757 | * voting [hello, deadlock city!], so unforuntately we just |
758 | * have to skip deleting this guy. That's OK though because | |
759 | * the node who's doing the actual deleting should handle it | |
760 | * anyway. */ | |
34d024f8 | 761 | if (current == osb->dc_task) { |
ccd979bd | 762 | mlog(0, "Skipping delete of %lu because we're currently " |
34d024f8 | 763 | "in downconvert\n", inode->i_ino); |
ccd979bd MF |
764 | goto bail; |
765 | } | |
766 | ||
767 | spin_lock(&oi->ip_lock); | |
768 | /* OCFS2 *never* deletes system files. This should technically | |
769 | * never get here as system file inodes should always have a | |
770 | * positive link count. */ | |
771 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { | |
b0697053 MF |
772 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
773 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
774 | goto bail_unlock; |
775 | } | |
776 | ||
34d024f8 MF |
777 | /* If we have allowd wipe of this inode for another node, it |
778 | * will be marked here so we can safely skip it. Recovery will | |
779 | * cleanup any inodes we might inadvertantly skip here. */ | |
ccd979bd MF |
780 | if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) { |
781 | mlog(0, "Skipping delete of %lu because another node " | |
782 | "has done this for us.\n", inode->i_ino); | |
783 | goto bail_unlock; | |
784 | } | |
785 | ||
786 | ret = 1; | |
787 | bail_unlock: | |
788 | spin_unlock(&oi->ip_lock); | |
789 | bail: | |
790 | return ret; | |
791 | } | |
792 | ||
793 | /* Query the cluster to determine whether we should wipe an inode from | |
794 | * disk or not. | |
795 | * | |
796 | * Requires the inode to have the cluster lock. */ | |
797 | static int ocfs2_query_inode_wipe(struct inode *inode, | |
798 | struct buffer_head *di_bh, | |
799 | int *wipe) | |
800 | { | |
801 | int status = 0; | |
802 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
803 | struct ocfs2_dinode *di; | |
804 | ||
805 | *wipe = 0; | |
806 | ||
807 | /* While we were waiting for the cluster lock in | |
808 | * ocfs2_delete_inode, another node might have asked to delete | |
809 | * the inode. Recheck our flags to catch this. */ | |
810 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
b0697053 MF |
811 | mlog(0, "Skipping delete of %llu because flags changed\n", |
812 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
813 | goto bail; |
814 | } | |
815 | ||
816 | /* Now that we have an up to date inode, we can double check | |
817 | * the link count. */ | |
818 | if (inode->i_nlink) { | |
b0697053 MF |
819 | mlog(0, "Skipping delete of %llu because nlink = %u\n", |
820 | (unsigned long long)oi->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
821 | goto bail; |
822 | } | |
823 | ||
824 | /* Do some basic inode verification... */ | |
825 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
826 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) { | |
827 | /* for lack of a better error? */ | |
828 | status = -EEXIST; | |
829 | mlog(ML_ERROR, | |
b0697053 | 830 | "Inode %llu (on-disk %llu) not orphaned! " |
ccd979bd | 831 | "Disk flags 0x%x, inode flags 0x%x\n", |
b0697053 | 832 | (unsigned long long)oi->ip_blkno, |
1ca1a111 MF |
833 | (unsigned long long)le64_to_cpu(di->i_blkno), |
834 | le32_to_cpu(di->i_flags), oi->ip_flags); | |
ccd979bd MF |
835 | goto bail; |
836 | } | |
837 | ||
838 | /* has someone already deleted us?! baaad... */ | |
839 | if (di->i_dtime) { | |
840 | status = -EEXIST; | |
841 | mlog_errno(status); | |
842 | goto bail; | |
843 | } | |
844 | ||
6f16bf65 MF |
845 | /* |
846 | * This is how ocfs2 determines whether an inode is still live | |
847 | * within the cluster. Every node takes a shared read lock on | |
848 | * the inode open lock in ocfs2_read_locked_inode(). When we | |
849 | * get to ->delete_inode(), each node tries to convert it's | |
850 | * lock to an exclusive. Trylocks are serialized by the inode | |
851 | * meta data lock. If the upconvert suceeds, we know the inode | |
852 | * is no longer live and can be deleted. | |
853 | * | |
854 | * Though we call this with the meta data lock held, the | |
855 | * trylock keeps us from ABBA deadlock. | |
856 | */ | |
857 | status = ocfs2_try_open_lock(inode, 1); | |
50008630 | 858 | if (status == -EAGAIN) { |
ccd979bd | 859 | status = 0; |
2759236f | 860 | mlog(0, "Skipping delete of %llu because it is in use on " |
b0697053 | 861 | "other nodes\n", (unsigned long long)oi->ip_blkno); |
ccd979bd MF |
862 | goto bail; |
863 | } | |
864 | if (status < 0) { | |
865 | mlog_errno(status); | |
866 | goto bail; | |
867 | } | |
868 | ||
50008630 TY |
869 | *wipe = 1; |
870 | mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n", | |
871 | (unsigned long long)oi->ip_blkno, | |
872 | le16_to_cpu(di->i_orphaned_slot)); | |
ccd979bd MF |
873 | |
874 | bail: | |
875 | return status; | |
876 | } | |
877 | ||
878 | /* Support function for ocfs2_delete_inode. Will help us keep the | |
879 | * inode data in a consistent state for clear_inode. Always truncates | |
880 | * pages, optionally sync's them first. */ | |
881 | static void ocfs2_cleanup_delete_inode(struct inode *inode, | |
882 | int sync_data) | |
883 | { | |
b0697053 MF |
884 | mlog(0, "Cleanup inode %llu, sync = %d\n", |
885 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); | |
ccd979bd MF |
886 | if (sync_data) |
887 | write_inode_now(inode, 1); | |
888 | truncate_inode_pages(&inode->i_data, 0); | |
889 | } | |
890 | ||
891 | void ocfs2_delete_inode(struct inode *inode) | |
892 | { | |
893 | int wipe, status; | |
894 | sigset_t blocked, oldset; | |
895 | struct buffer_head *di_bh = NULL; | |
896 | ||
897 | mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino); | |
898 | ||
899 | if (is_bad_inode(inode)) { | |
900 | mlog(0, "Skipping delete of bad inode\n"); | |
901 | goto bail; | |
902 | } | |
903 | ||
904 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
905 | /* It's probably not necessary to truncate_inode_pages | |
906 | * here but we do it for safety anyway (it will most | |
907 | * likely be a no-op anyway) */ | |
908 | ocfs2_cleanup_delete_inode(inode, 0); | |
909 | goto bail; | |
910 | } | |
911 | ||
912 | /* We want to block signals in delete_inode as the lock and | |
913 | * messaging paths may return us -ERESTARTSYS. Which would | |
914 | * cause us to exit early, resulting in inodes being orphaned | |
915 | * forever. */ | |
916 | sigfillset(&blocked); | |
917 | status = sigprocmask(SIG_BLOCK, &blocked, &oldset); | |
918 | if (status < 0) { | |
919 | mlog_errno(status); | |
920 | ocfs2_cleanup_delete_inode(inode, 1); | |
921 | goto bail; | |
922 | } | |
923 | ||
924 | /* Lock down the inode. This gives us an up to date view of | |
925 | * it's metadata (for verification), and allows us to | |
34d024f8 | 926 | * serialize delete_inode on multiple nodes. |
ccd979bd MF |
927 | * |
928 | * Even though we might be doing a truncate, we don't take the | |
929 | * allocation lock here as it won't be needed - nobody will | |
930 | * have the file open. | |
931 | */ | |
4bcec184 | 932 | status = ocfs2_meta_lock(inode, &di_bh, 1); |
ccd979bd MF |
933 | if (status < 0) { |
934 | if (status != -ENOENT) | |
935 | mlog_errno(status); | |
936 | ocfs2_cleanup_delete_inode(inode, 0); | |
937 | goto bail_unblock; | |
938 | } | |
939 | ||
940 | /* Query the cluster. This will be the final decision made | |
941 | * before we go ahead and wipe the inode. */ | |
942 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); | |
943 | if (!wipe || status < 0) { | |
34d024f8 | 944 | /* Error and remote inode busy both mean we won't be |
ccd979bd MF |
945 | * removing the inode, so they take almost the same |
946 | * path. */ | |
947 | if (status < 0) | |
948 | mlog_errno(status); | |
949 | ||
34d024f8 MF |
950 | /* Someone in the cluster has disallowed a wipe of |
951 | * this inode, or it was never completely | |
952 | * orphaned. Write out the pages and exit now. */ | |
ccd979bd MF |
953 | ocfs2_cleanup_delete_inode(inode, 1); |
954 | goto bail_unlock_inode; | |
955 | } | |
956 | ||
957 | ocfs2_cleanup_delete_inode(inode, 0); | |
958 | ||
959 | status = ocfs2_wipe_inode(inode, di_bh); | |
960 | if (status < 0) { | |
b4df6ed8 MF |
961 | if (status != -EDEADLK) |
962 | mlog_errno(status); | |
ccd979bd MF |
963 | goto bail_unlock_inode; |
964 | } | |
965 | ||
24c19ef4 MF |
966 | /* |
967 | * Mark the inode as successfully deleted. | |
968 | * | |
969 | * This is important for ocfs2_clear_inode() as it will check | |
970 | * this flag and skip any checkpointing work | |
971 | * | |
972 | * ocfs2_stuff_meta_lvb() also uses this flag to invalidate | |
973 | * the LVB for other nodes. | |
974 | */ | |
ccd979bd MF |
975 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; |
976 | ||
977 | bail_unlock_inode: | |
978 | ocfs2_meta_unlock(inode, 1); | |
979 | brelse(di_bh); | |
980 | bail_unblock: | |
981 | status = sigprocmask(SIG_SETMASK, &oldset, NULL); | |
982 | if (status < 0) | |
983 | mlog_errno(status); | |
984 | bail: | |
985 | clear_inode(inode); | |
986 | mlog_exit_void(); | |
987 | } | |
988 | ||
989 | void ocfs2_clear_inode(struct inode *inode) | |
990 | { | |
991 | int status; | |
992 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
993 | ||
994 | mlog_entry_void(); | |
995 | ||
996 | if (!inode) | |
997 | goto bail; | |
998 | ||
b0697053 MF |
999 | mlog(0, "Clearing inode: %llu, nlink = %u\n", |
1000 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
1001 | |
1002 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, | |
1003 | "Inode=%lu\n", inode->i_ino); | |
1004 | ||
34d024f8 MF |
1005 | /* To preven remote deletes we hold open lock before, now it |
1006 | * is time to unlock PR and EX open locks. */ | |
50008630 TY |
1007 | ocfs2_open_unlock(inode); |
1008 | ||
ccd979bd | 1009 | /* Do these before all the other work so that we don't bounce |
34d024f8 | 1010 | * the downconvert thread while waiting to destroy the locks. */ |
ccd979bd MF |
1011 | ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres); |
1012 | ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres); | |
50008630 | 1013 | ocfs2_mark_lockres_freeing(&oi->ip_open_lockres); |
ccd979bd MF |
1014 | |
1015 | /* We very well may get a clear_inode before all an inodes | |
1016 | * metadata has hit disk. Of course, we can't drop any cluster | |
1017 | * locks until the journal has finished with it. The only | |
1018 | * exception here are successfully wiped inodes - their | |
1019 | * metadata can now be considered to be part of the system | |
1020 | * inodes from which it came. */ | |
1021 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) | |
1022 | ocfs2_checkpoint_inode(inode); | |
1023 | ||
1024 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), | |
b0697053 MF |
1025 | "Clear inode of %llu, inode has io markers\n", |
1026 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1027 | |
83418978 MF |
1028 | ocfs2_extent_map_trunc(inode, 0); |
1029 | ||
ccd979bd MF |
1030 | status = ocfs2_drop_inode_locks(inode); |
1031 | if (status < 0) | |
1032 | mlog_errno(status); | |
1033 | ||
1034 | ocfs2_lock_res_free(&oi->ip_rw_lockres); | |
1035 | ocfs2_lock_res_free(&oi->ip_meta_lockres); | |
50008630 | 1036 | ocfs2_lock_res_free(&oi->ip_open_lockres); |
ccd979bd MF |
1037 | |
1038 | ocfs2_metadata_cache_purge(inode); | |
1039 | ||
1040 | mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached, | |
b0697053 MF |
1041 | "Clear inode of %llu, inode has %u cache items\n", |
1042 | (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached); | |
ccd979bd MF |
1043 | |
1044 | mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE), | |
b0697053 MF |
1045 | "Clear inode of %llu, inode has a bad flag\n", |
1046 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1047 | |
1048 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), | |
b0697053 MF |
1049 | "Clear inode of %llu, inode is locked\n", |
1050 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1051 | |
251b6ecc | 1052 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
b0697053 MF |
1053 | "Clear inode of %llu, io_mutex is locked\n", |
1054 | (unsigned long long)oi->ip_blkno); | |
251b6ecc | 1055 | mutex_unlock(&oi->ip_io_mutex); |
ccd979bd MF |
1056 | |
1057 | /* | |
1058 | * down_trylock() returns 0, down_write_trylock() returns 1 | |
1059 | * kernel 1, world 0 | |
1060 | */ | |
1061 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), | |
b0697053 MF |
1062 | "Clear inode of %llu, alloc_sem is locked\n", |
1063 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1064 | up_write(&oi->ip_alloc_sem); |
1065 | ||
1066 | mlog_bug_on_msg(oi->ip_open_count, | |
b0697053 MF |
1067 | "Clear inode of %llu has open count %d\n", |
1068 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); | |
ccd979bd MF |
1069 | |
1070 | /* Clear all other flags. */ | |
1071 | oi->ip_flags = OCFS2_INODE_CACHE_INLINE; | |
1072 | oi->ip_created_trans = 0; | |
1073 | oi->ip_last_trans = 0; | |
1074 | oi->ip_dir_start_lookup = 0; | |
1075 | oi->ip_blkno = 0ULL; | |
1076 | ||
1077 | bail: | |
1078 | mlog_exit_void(); | |
1079 | } | |
1080 | ||
1081 | /* Called under inode_lock, with no more references on the | |
1082 | * struct inode, so it's safe here to check the flags field | |
1083 | * and to manipulate i_nlink without any other locks. */ | |
1084 | void ocfs2_drop_inode(struct inode *inode) | |
1085 | { | |
1086 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1087 | ||
1088 | mlog_entry_void(); | |
1089 | ||
b0697053 MF |
1090 | mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n", |
1091 | (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags); | |
ccd979bd | 1092 | |
379dfe9d MF |
1093 | if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) |
1094 | generic_delete_inode(inode); | |
1095 | else | |
1096 | generic_drop_inode(inode); | |
ccd979bd MF |
1097 | |
1098 | mlog_exit_void(); | |
1099 | } | |
1100 | ||
1101 | /* | |
1102 | * TODO: this should probably be merged into ocfs2_get_block | |
1103 | * | |
1104 | * However, you now need to pay attention to the cont_prepare_write() | |
1105 | * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much | |
1106 | * expects never to extend). | |
1107 | */ | |
1108 | struct buffer_head *ocfs2_bread(struct inode *inode, | |
1109 | int block, int *err, int reada) | |
1110 | { | |
1111 | struct buffer_head *bh = NULL; | |
1112 | int tmperr; | |
1113 | u64 p_blkno; | |
1114 | int readflags = OCFS2_BH_CACHED; | |
1115 | ||
ccd979bd MF |
1116 | if (reada) |
1117 | readflags |= OCFS2_BH_READAHEAD; | |
ccd979bd MF |
1118 | |
1119 | if (((u64)block << inode->i_sb->s_blocksize_bits) >= | |
1120 | i_size_read(inode)) { | |
1121 | BUG_ON(!reada); | |
1122 | return NULL; | |
1123 | } | |
1124 | ||
ee19a779 | 1125 | down_read(&OCFS2_I(inode)->ip_alloc_sem); |
49cb8d2d MF |
1126 | tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, |
1127 | NULL); | |
ee19a779 | 1128 | up_read(&OCFS2_I(inode)->ip_alloc_sem); |
ccd979bd MF |
1129 | if (tmperr < 0) { |
1130 | mlog_errno(tmperr); | |
1131 | goto fail; | |
1132 | } | |
1133 | ||
1134 | tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh, | |
1135 | readflags, inode); | |
1136 | if (tmperr < 0) | |
1137 | goto fail; | |
1138 | ||
1139 | tmperr = 0; | |
1140 | ||
1141 | *err = 0; | |
1142 | return bh; | |
1143 | ||
1144 | fail: | |
1145 | if (bh) { | |
1146 | brelse(bh); | |
1147 | bh = NULL; | |
1148 | } | |
1149 | *err = -EIO; | |
1150 | return NULL; | |
1151 | } | |
1152 | ||
1153 | /* | |
1154 | * This is called from our getattr. | |
1155 | */ | |
1156 | int ocfs2_inode_revalidate(struct dentry *dentry) | |
1157 | { | |
1158 | struct inode *inode = dentry->d_inode; | |
1159 | int status = 0; | |
1160 | ||
b0697053 MF |
1161 | mlog_entry("(inode = 0x%p, ino = %llu)\n", inode, |
1162 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL); | |
ccd979bd MF |
1163 | |
1164 | if (!inode) { | |
1165 | mlog(0, "eep, no inode!\n"); | |
1166 | status = -ENOENT; | |
1167 | goto bail; | |
1168 | } | |
1169 | ||
1170 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1171 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
1172 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1173 | mlog(0, "inode deleted!\n"); | |
1174 | status = -ENOENT; | |
1175 | goto bail; | |
1176 | } | |
1177 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1178 | ||
1179 | /* Let ocfs2_meta_lock do the work of updating our struct | |
1180 | * inode for us. */ | |
4bcec184 | 1181 | status = ocfs2_meta_lock(inode, NULL, 0); |
ccd979bd MF |
1182 | if (status < 0) { |
1183 | if (status != -ENOENT) | |
1184 | mlog_errno(status); | |
1185 | goto bail; | |
1186 | } | |
1187 | ocfs2_meta_unlock(inode, 0); | |
1188 | bail: | |
1189 | mlog_exit(status); | |
1190 | ||
1191 | return status; | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * Updates a disk inode from a | |
1196 | * struct inode. | |
1197 | * Only takes ip_lock. | |
1198 | */ | |
1fabe148 | 1199 | int ocfs2_mark_inode_dirty(handle_t *handle, |
ccd979bd MF |
1200 | struct inode *inode, |
1201 | struct buffer_head *bh) | |
1202 | { | |
1203 | int status; | |
1204 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
1205 | ||
b0697053 MF |
1206 | mlog_entry("(inode %llu)\n", |
1207 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
1208 | |
1209 | status = ocfs2_journal_access(handle, inode, bh, | |
1210 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1211 | if (status < 0) { | |
1212 | mlog_errno(status); | |
1213 | goto leave; | |
1214 | } | |
1215 | ||
1216 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1217 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
6e4b0d56 | 1218 | ocfs2_get_inode_flags(OCFS2_I(inode)); |
ca4d147e | 1219 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
15b1e36b | 1220 | fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features); |
ccd979bd MF |
1221 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1222 | ||
1223 | fe->i_size = cpu_to_le64(i_size_read(inode)); | |
1224 | fe->i_links_count = cpu_to_le16(inode->i_nlink); | |
1225 | fe->i_uid = cpu_to_le32(inode->i_uid); | |
1226 | fe->i_gid = cpu_to_le32(inode->i_gid); | |
1227 | fe->i_mode = cpu_to_le16(inode->i_mode); | |
1228 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); | |
1229 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
1230 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
1231 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
1232 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1233 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
1234 | ||
1235 | status = ocfs2_journal_dirty(handle, bh); | |
1236 | if (status < 0) | |
1237 | mlog_errno(status); | |
1238 | ||
1239 | status = 0; | |
1240 | leave: | |
1241 | ||
1242 | mlog_exit(status); | |
1243 | return status; | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * | |
1248 | * Updates a struct inode from a disk inode. | |
1249 | * does no i/o, only takes ip_lock. | |
1250 | */ | |
1251 | void ocfs2_refresh_inode(struct inode *inode, | |
1252 | struct ocfs2_dinode *fe) | |
1253 | { | |
ccd979bd MF |
1254 | spin_lock(&OCFS2_I(inode)->ip_lock); |
1255 | ||
1256 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e | 1257 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
15b1e36b | 1258 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
ca4d147e | 1259 | ocfs2_set_inode_flags(inode); |
ccd979bd MF |
1260 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
1261 | inode->i_nlink = le16_to_cpu(fe->i_links_count); | |
1262 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
1263 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
1264 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
ccd979bd MF |
1265 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) |
1266 | inode->i_blocks = 0; | |
1267 | else | |
8110b073 | 1268 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
1269 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
1270 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
1271 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
1272 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
1273 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
1274 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
1275 | ||
1276 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1277 | } |