]>
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> | |
31 | #include <linux/smp_lock.h> | |
32 | ||
33 | #include <asm/byteorder.h> | |
34 | ||
35 | #define MLOG_MASK_PREFIX ML_INODE | |
36 | #include <cluster/masklog.h> | |
37 | ||
38 | #include "ocfs2.h" | |
39 | ||
40 | #include "alloc.h" | |
41 | #include "dlmglue.h" | |
42 | #include "extent_map.h" | |
43 | #include "file.h" | |
b4df6ed8 | 44 | #include "heartbeat.h" |
ccd979bd MF |
45 | #include "inode.h" |
46 | #include "journal.h" | |
47 | #include "namei.h" | |
48 | #include "suballoc.h" | |
49 | #include "super.h" | |
50 | #include "symlink.h" | |
51 | #include "sysfile.h" | |
52 | #include "uptodate.h" | |
53 | #include "vote.h" | |
54 | ||
55 | #include "buffer_head_io.h" | |
56 | ||
ccd979bd MF |
57 | struct ocfs2_find_inode_args |
58 | { | |
59 | u64 fi_blkno; | |
60 | unsigned long fi_ino; | |
61 | unsigned int fi_flags; | |
62 | }; | |
63 | ||
64 | static int ocfs2_read_locked_inode(struct inode *inode, | |
65 | struct ocfs2_find_inode_args *args); | |
66 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); | |
67 | static int ocfs2_find_actor(struct inode *inode, void *opaque); | |
68 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
69 | struct inode *inode, | |
70 | struct buffer_head *fe_bh); | |
71 | ||
ca4d147e HP |
72 | void ocfs2_set_inode_flags(struct inode *inode) |
73 | { | |
74 | unsigned int flags = OCFS2_I(inode)->ip_attr; | |
75 | ||
76 | inode->i_flags &= ~(S_IMMUTABLE | | |
77 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); | |
78 | ||
79 | if (flags & OCFS2_IMMUTABLE_FL) | |
80 | inode->i_flags |= S_IMMUTABLE; | |
81 | ||
82 | if (flags & OCFS2_SYNC_FL) | |
83 | inode->i_flags |= S_SYNC; | |
84 | if (flags & OCFS2_APPEND_FL) | |
85 | inode->i_flags |= S_APPEND; | |
86 | if (flags & OCFS2_NOATIME_FL) | |
87 | inode->i_flags |= S_NOATIME; | |
88 | if (flags & OCFS2_DIRSYNC_FL) | |
89 | inode->i_flags |= S_DIRSYNC; | |
90 | } | |
91 | ||
24c19ef4 | 92 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags) |
ccd979bd MF |
93 | { |
94 | struct inode *inode = NULL; | |
95 | struct super_block *sb = osb->sb; | |
96 | struct ocfs2_find_inode_args args; | |
97 | ||
b0697053 | 98 | mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno); |
ccd979bd MF |
99 | |
100 | /* Ok. By now we've either got the offsets passed to us by the | |
101 | * caller, or we just pulled them off the bh. Lets do some | |
102 | * sanity checks to make sure they're OK. */ | |
103 | if (blkno == 0) { | |
104 | inode = ERR_PTR(-EINVAL); | |
105 | mlog_errno(PTR_ERR(inode)); | |
106 | goto bail; | |
107 | } | |
108 | ||
109 | args.fi_blkno = blkno; | |
24c19ef4 | 110 | args.fi_flags = flags; |
ccd979bd MF |
111 | args.fi_ino = ino_from_blkno(sb, blkno); |
112 | ||
113 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, | |
114 | ocfs2_init_locked_inode, &args); | |
115 | /* inode was *not* in the inode cache. 2.6.x requires | |
116 | * us to do our own read_inode call and unlock it | |
117 | * afterwards. */ | |
118 | if (inode && inode->i_state & I_NEW) { | |
119 | mlog(0, "Inode was not in inode cache, reading it.\n"); | |
120 | ocfs2_read_locked_inode(inode, &args); | |
121 | unlock_new_inode(inode); | |
122 | } | |
123 | if (inode == NULL) { | |
124 | inode = ERR_PTR(-ENOMEM); | |
125 | mlog_errno(PTR_ERR(inode)); | |
126 | goto bail; | |
127 | } | |
128 | if (is_bad_inode(inode)) { | |
129 | iput(inode); | |
130 | inode = ERR_PTR(-ESTALE); | |
ccd979bd MF |
131 | goto bail; |
132 | } | |
133 | ||
134 | bail: | |
135 | if (!IS_ERR(inode)) { | |
b0697053 MF |
136 | mlog(0, "returning inode with number %llu\n", |
137 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 138 | mlog_exit_ptr(inode); |
6a1bd4a5 | 139 | } |
ccd979bd MF |
140 | |
141 | return inode; | |
142 | } | |
143 | ||
144 | ||
145 | /* | |
146 | * here's how inodes get read from disk: | |
147 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR | |
148 | * found? : return the in-memory inode | |
149 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE | |
150 | */ | |
151 | ||
152 | static int ocfs2_find_actor(struct inode *inode, void *opaque) | |
153 | { | |
154 | struct ocfs2_find_inode_args *args = NULL; | |
155 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
156 | int ret = 0; | |
157 | ||
158 | mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque); | |
159 | ||
160 | args = opaque; | |
161 | ||
162 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); | |
163 | ||
164 | if (oi->ip_blkno != args->fi_blkno) | |
165 | goto bail; | |
166 | ||
ccd979bd MF |
167 | ret = 1; |
168 | bail: | |
169 | mlog_exit(ret); | |
170 | return ret; | |
171 | } | |
172 | ||
173 | /* | |
174 | * initialize the new inode, but don't do anything that would cause | |
175 | * us to sleep. | |
176 | * return 0 on success, 1 on failure | |
177 | */ | |
178 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) | |
179 | { | |
180 | struct ocfs2_find_inode_args *args = opaque; | |
181 | ||
182 | mlog_entry("inode = %p, opaque = %p\n", inode, opaque); | |
183 | ||
184 | inode->i_ino = args->fi_ino; | |
185 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; | |
186 | ||
187 | mlog_exit(0); | |
188 | return 0; | |
189 | } | |
190 | ||
191 | int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, | |
192 | int create_ino) | |
193 | { | |
194 | struct super_block *sb; | |
195 | struct ocfs2_super *osb; | |
196 | int status = -EINVAL; | |
197 | ||
b0697053 MF |
198 | mlog_entry("(0x%p, size:%llu)\n", inode, |
199 | (unsigned long long)fe->i_size); | |
ccd979bd MF |
200 | |
201 | sb = inode->i_sb; | |
202 | osb = OCFS2_SB(sb); | |
203 | ||
204 | /* this means that read_inode cannot create a superblock inode | |
205 | * today. change if needed. */ | |
206 | if (!OCFS2_IS_VALID_DINODE(fe) || | |
207 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
6a1bd4a5 | 208 | mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, " |
ccd979bd | 209 | "signature = %.*s, flags = 0x%x\n", |
b0697053 MF |
210 | inode->i_ino, |
211 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, | |
ccd979bd MF |
212 | fe->i_signature, le32_to_cpu(fe->i_flags)); |
213 | goto bail; | |
214 | } | |
215 | ||
216 | if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) { | |
217 | mlog(ML_ERROR, "file entry generation does not match " | |
218 | "superblock! osb->fs_generation=%x, " | |
219 | "fe->i_fs_generation=%x\n", | |
220 | osb->fs_generation, le32_to_cpu(fe->i_fs_generation)); | |
221 | goto bail; | |
222 | } | |
223 | ||
224 | inode->i_version = 1; | |
225 | inode->i_generation = le32_to_cpu(fe->i_generation); | |
226 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
227 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
228 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
229 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
ccd979bd MF |
230 | |
231 | /* Fast symlinks will have i_size but no allocated clusters. */ | |
232 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) | |
233 | inode->i_blocks = 0; | |
234 | else | |
235 | inode->i_blocks = | |
236 | ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size)); | |
237 | inode->i_mapping->a_ops = &ocfs2_aops; | |
ccd979bd MF |
238 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
239 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
240 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
241 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
242 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
243 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
244 | ||
245 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) | |
246 | mlog(ML_ERROR, | |
b0697053 MF |
247 | "ip_blkno %llu != i_blkno %llu!\n", |
248 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
249 | (unsigned long long)fe->i_blkno); | |
ccd979bd MF |
250 | |
251 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e | 252 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
ccd979bd | 253 | |
ccd979bd MF |
254 | inode->i_nlink = le16_to_cpu(fe->i_links_count); |
255 | ||
24c19ef4 MF |
256 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) |
257 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; | |
258 | ||
ccd979bd MF |
259 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { |
260 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
261 | mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino); | |
262 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { | |
263 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
264 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { | |
265 | mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino); | |
266 | /* we can't actually hit this as read_inode can't | |
267 | * handle superblocks today ;-) */ | |
268 | BUG(); | |
269 | } | |
270 | ||
271 | switch (inode->i_mode & S_IFMT) { | |
272 | case S_IFREG: | |
273 | inode->i_fop = &ocfs2_fops; | |
274 | inode->i_op = &ocfs2_file_iops; | |
275 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
276 | break; | |
277 | case S_IFDIR: | |
278 | inode->i_op = &ocfs2_dir_iops; | |
279 | inode->i_fop = &ocfs2_dops; | |
280 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
281 | break; | |
282 | case S_IFLNK: | |
283 | if (ocfs2_inode_is_fast_symlink(inode)) | |
284 | inode->i_op = &ocfs2_fast_symlink_inode_operations; | |
285 | else | |
286 | inode->i_op = &ocfs2_symlink_inode_operations; | |
287 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
288 | break; | |
289 | default: | |
290 | inode->i_op = &ocfs2_special_file_iops; | |
291 | init_special_inode(inode, inode->i_mode, | |
292 | inode->i_rdev); | |
293 | break; | |
294 | } | |
295 | ||
24c19ef4 MF |
296 | if (create_ino) { |
297 | inode->i_ino = ino_from_blkno(inode->i_sb, | |
298 | le64_to_cpu(fe->i_blkno)); | |
299 | ||
300 | /* | |
301 | * If we ever want to create system files from kernel, | |
302 | * the generation argument to | |
303 | * ocfs2_inode_lock_res_init() will have to change. | |
304 | */ | |
305 | BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)); | |
306 | ||
307 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
308 | OCFS2_LOCK_TYPE_META, 0, inode); | |
50008630 TY |
309 | |
310 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, | |
311 | OCFS2_LOCK_TYPE_OPEN, 0, inode); | |
24c19ef4 MF |
312 | } |
313 | ||
ccd979bd | 314 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, |
24c19ef4 MF |
315 | OCFS2_LOCK_TYPE_RW, inode->i_generation, |
316 | inode); | |
317 | ||
ccd979bd | 318 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres, |
24c19ef4 MF |
319 | OCFS2_LOCK_TYPE_DATA, inode->i_generation, |
320 | inode); | |
ccd979bd | 321 | |
ca4d147e | 322 | ocfs2_set_inode_flags(inode); |
ca4d147e | 323 | |
ccd979bd MF |
324 | status = 0; |
325 | bail: | |
326 | mlog_exit(status); | |
327 | return status; | |
328 | } | |
329 | ||
330 | static int ocfs2_read_locked_inode(struct inode *inode, | |
331 | struct ocfs2_find_inode_args *args) | |
332 | { | |
333 | struct super_block *sb; | |
334 | struct ocfs2_super *osb; | |
335 | struct ocfs2_dinode *fe; | |
336 | struct buffer_head *bh = NULL; | |
24c19ef4 MF |
337 | int status, can_lock; |
338 | u32 generation = 0; | |
ccd979bd MF |
339 | |
340 | mlog_entry("(0x%p, 0x%p)\n", inode, args); | |
341 | ||
342 | status = -EINVAL; | |
343 | if (inode == NULL || inode->i_sb == NULL) { | |
344 | mlog(ML_ERROR, "bad inode\n"); | |
24c19ef4 | 345 | return status; |
ccd979bd MF |
346 | } |
347 | sb = inode->i_sb; | |
348 | osb = OCFS2_SB(sb); | |
349 | ||
350 | if (!args) { | |
351 | mlog(ML_ERROR, "bad inode args\n"); | |
352 | make_bad_inode(inode); | |
24c19ef4 MF |
353 | return status; |
354 | } | |
355 | ||
356 | /* | |
357 | * To improve performance of cold-cache inode stats, we take | |
358 | * the cluster lock here if possible. | |
359 | * | |
360 | * Generally, OCFS2 never trusts the contents of an inode | |
361 | * unless it's holding a cluster lock, so taking it here isn't | |
362 | * a correctness issue as much as it is a performance | |
363 | * improvement. | |
364 | * | |
365 | * There are three times when taking the lock is not a good idea: | |
366 | * | |
367 | * 1) During startup, before we have initialized the DLM. | |
368 | * | |
369 | * 2) If we are reading certain system files which never get | |
370 | * cluster locks (local alloc, truncate log). | |
371 | * | |
372 | * 3) If the process doing the iget() is responsible for | |
373 | * orphan dir recovery. We're holding the orphan dir lock and | |
374 | * can get into a deadlock with another process on another | |
375 | * node in ->delete_inode(). | |
376 | * | |
377 | * #1 and #2 can be simply solved by never taking the lock | |
378 | * here for system files (which are the only type we read | |
379 | * during mount). It's a heavier approach, but our main | |
380 | * concern is user-accesible files anyway. | |
381 | * | |
382 | * #3 works itself out because we'll eventually take the | |
383 | * cluster lock before trusting anything anyway. | |
384 | */ | |
385 | can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
50008630 | 386 | && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) |
c271c5c2 | 387 | && !ocfs2_mount_local(osb); |
24c19ef4 MF |
388 | |
389 | /* | |
390 | * To maintain backwards compatibility with older versions of | |
391 | * ocfs2-tools, we still store the generation value for system | |
392 | * files. The only ones that actually matter to userspace are | |
393 | * the journals, but it's easier and inexpensive to just flag | |
394 | * all system files similarly. | |
395 | */ | |
396 | if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
397 | generation = osb->fs_generation; | |
398 | ||
399 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
400 | OCFS2_LOCK_TYPE_META, | |
401 | generation, inode); | |
402 | ||
50008630 TY |
403 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, |
404 | OCFS2_LOCK_TYPE_OPEN, | |
405 | 0, inode); | |
406 | ||
24c19ef4 | 407 | if (can_lock) { |
50008630 TY |
408 | status = ocfs2_open_lock(inode); |
409 | if (status) { | |
410 | make_bad_inode(inode); | |
411 | mlog_errno(status); | |
412 | return status; | |
413 | } | |
4bcec184 | 414 | status = ocfs2_meta_lock(inode, NULL, 0); |
24c19ef4 MF |
415 | if (status) { |
416 | make_bad_inode(inode); | |
417 | mlog_errno(status); | |
418 | return status; | |
419 | } | |
ccd979bd MF |
420 | } |
421 | ||
50008630 TY |
422 | if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) { |
423 | status = ocfs2_try_open_lock(inode, 0); | |
424 | if (status) { | |
425 | make_bad_inode(inode); | |
426 | return status; | |
427 | } | |
428 | } | |
429 | ||
24c19ef4 MF |
430 | status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, |
431 | can_lock ? inode : NULL); | |
ccd979bd MF |
432 | if (status < 0) { |
433 | mlog_errno(status); | |
ccd979bd MF |
434 | goto bail; |
435 | } | |
436 | ||
24c19ef4 | 437 | status = -EINVAL; |
ccd979bd MF |
438 | fe = (struct ocfs2_dinode *) bh->b_data; |
439 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
b0697053 MF |
440 | mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n", |
441 | (unsigned long long)fe->i_blkno, 7, fe->i_signature); | |
ccd979bd MF |
442 | goto bail; |
443 | } | |
444 | ||
24c19ef4 MF |
445 | /* |
446 | * This is a code bug. Right now the caller needs to | |
447 | * understand whether it is asking for a system file inode or | |
448 | * not so the proper lock names can be built. | |
449 | */ | |
450 | mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) != | |
451 | !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE), | |
452 | "Inode %llu: system file state is ambigous\n", | |
453 | (unsigned long long)args->fi_blkno); | |
ccd979bd MF |
454 | |
455 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || | |
456 | S_ISBLK(le16_to_cpu(fe->i_mode))) | |
457 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
458 | ||
6a1bd4a5 | 459 | if (ocfs2_populate_inode(inode, fe, 0) < 0) |
ccd979bd | 460 | goto bail; |
ccd979bd MF |
461 | |
462 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); | |
463 | ||
ccd979bd MF |
464 | status = 0; |
465 | ||
466 | bail: | |
24c19ef4 MF |
467 | if (can_lock) |
468 | ocfs2_meta_unlock(inode, 0); | |
469 | ||
470 | if (status < 0) | |
471 | make_bad_inode(inode); | |
472 | ||
ccd979bd MF |
473 | if (args && bh) |
474 | brelse(bh); | |
475 | ||
476 | mlog_exit(status); | |
477 | return status; | |
478 | } | |
479 | ||
480 | void ocfs2_sync_blockdev(struct super_block *sb) | |
481 | { | |
482 | sync_blockdev(sb->s_bdev); | |
483 | } | |
484 | ||
485 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
486 | struct inode *inode, | |
487 | struct buffer_head *fe_bh) | |
488 | { | |
489 | int status = 0; | |
ccd979bd MF |
490 | struct ocfs2_truncate_context *tc = NULL; |
491 | struct ocfs2_dinode *fe; | |
492 | ||
493 | mlog_entry_void(); | |
494 | ||
495 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
496 | ||
3a0782d0 MF |
497 | if (fe->i_clusters) { |
498 | status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc); | |
499 | if (status < 0) { | |
500 | mlog_errno(status); | |
501 | goto out; | |
502 | } | |
ccd979bd | 503 | |
3a0782d0 MF |
504 | status = ocfs2_commit_truncate(osb, inode, fe_bh, tc); |
505 | if (status < 0) { | |
506 | mlog_errno(status); | |
507 | goto out; | |
508 | } | |
ccd979bd | 509 | } |
3a0782d0 | 510 | out: |
ccd979bd MF |
511 | |
512 | mlog_exit(status); | |
513 | return status; | |
514 | } | |
515 | ||
516 | static int ocfs2_remove_inode(struct inode *inode, | |
517 | struct buffer_head *di_bh, | |
518 | struct inode *orphan_dir_inode, | |
519 | struct buffer_head *orphan_dir_bh) | |
520 | { | |
521 | int status; | |
522 | struct inode *inode_alloc_inode = NULL; | |
523 | struct buffer_head *inode_alloc_bh = NULL; | |
1fabe148 | 524 | handle_t *handle; |
ccd979bd MF |
525 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
526 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
527 | ||
528 | inode_alloc_inode = | |
529 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
530 | le16_to_cpu(di->i_suballoc_slot)); | |
531 | if (!inode_alloc_inode) { | |
532 | status = -EEXIST; | |
533 | mlog_errno(status); | |
534 | goto bail; | |
535 | } | |
536 | ||
1b1dcc1b | 537 | mutex_lock(&inode_alloc_inode->i_mutex); |
4bcec184 | 538 | status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1); |
ccd979bd | 539 | if (status < 0) { |
1b1dcc1b | 540 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
541 | |
542 | mlog_errno(status); | |
543 | goto bail; | |
544 | } | |
545 | ||
65eff9cc | 546 | handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS); |
ccd979bd MF |
547 | if (IS_ERR(handle)) { |
548 | status = PTR_ERR(handle); | |
549 | mlog_errno(status); | |
550 | goto bail_unlock; | |
551 | } | |
552 | ||
553 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, | |
554 | orphan_dir_bh); | |
555 | if (status < 0) { | |
556 | mlog_errno(status); | |
557 | goto bail_commit; | |
558 | } | |
559 | ||
560 | /* set the inodes dtime */ | |
561 | status = ocfs2_journal_access(handle, inode, di_bh, | |
562 | OCFS2_JOURNAL_ACCESS_WRITE); | |
563 | if (status < 0) { | |
564 | mlog_errno(status); | |
565 | goto bail_commit; | |
566 | } | |
567 | ||
568 | di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec); | |
569 | le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); | |
570 | ||
571 | status = ocfs2_journal_dirty(handle, di_bh); | |
572 | if (status < 0) { | |
573 | mlog_errno(status); | |
574 | goto bail_commit; | |
575 | } | |
576 | ||
577 | ocfs2_remove_from_cache(inode, di_bh); | |
578 | ||
579 | status = ocfs2_free_dinode(handle, inode_alloc_inode, | |
580 | inode_alloc_bh, di); | |
581 | if (status < 0) | |
582 | mlog_errno(status); | |
583 | ||
584 | bail_commit: | |
02dc1af4 | 585 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
586 | bail_unlock: |
587 | ocfs2_meta_unlock(inode_alloc_inode, 1); | |
1b1dcc1b | 588 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
589 | brelse(inode_alloc_bh); |
590 | bail: | |
591 | iput(inode_alloc_inode); | |
592 | ||
593 | return status; | |
594 | } | |
595 | ||
b4df6ed8 MF |
596 | /* |
597 | * Serialize with orphan dir recovery. If the process doing | |
598 | * recovery on this orphan dir does an iget() with the dir | |
599 | * i_mutex held, we'll deadlock here. Instead we detect this | |
600 | * and exit early - recovery will wipe this inode for us. | |
601 | */ | |
602 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, | |
603 | int slot) | |
604 | { | |
605 | int ret = 0; | |
606 | ||
607 | spin_lock(&osb->osb_lock); | |
608 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { | |
609 | mlog(0, "Recovery is happening on orphan dir %d, will skip " | |
610 | "this inode\n", slot); | |
611 | ret = -EDEADLK; | |
612 | goto out; | |
613 | } | |
614 | /* This signals to the orphan recovery process that it should | |
615 | * wait for us to handle the wipe. */ | |
616 | osb->osb_orphan_wipes[slot]++; | |
617 | out: | |
618 | spin_unlock(&osb->osb_lock); | |
619 | return ret; | |
620 | } | |
621 | ||
622 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, | |
623 | int slot) | |
624 | { | |
625 | spin_lock(&osb->osb_lock); | |
626 | osb->osb_orphan_wipes[slot]--; | |
627 | spin_unlock(&osb->osb_lock); | |
628 | ||
629 | wake_up(&osb->osb_wipe_event); | |
630 | } | |
631 | ||
ccd979bd MF |
632 | static int ocfs2_wipe_inode(struct inode *inode, |
633 | struct buffer_head *di_bh) | |
634 | { | |
635 | int status, orphaned_slot; | |
636 | struct inode *orphan_dir_inode = NULL; | |
637 | struct buffer_head *orphan_dir_bh = NULL; | |
638 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
50008630 | 639 | struct ocfs2_dinode *di; |
ccd979bd | 640 | |
50008630 TY |
641 | di = (struct ocfs2_dinode *) di_bh->b_data; |
642 | orphaned_slot = le16_to_cpu(di->i_orphaned_slot); | |
b4df6ed8 MF |
643 | |
644 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); | |
645 | if (status) | |
646 | return status; | |
647 | ||
ccd979bd MF |
648 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
649 | ORPHAN_DIR_SYSTEM_INODE, | |
650 | orphaned_slot); | |
651 | if (!orphan_dir_inode) { | |
652 | status = -EEXIST; | |
653 | mlog_errno(status); | |
654 | goto bail; | |
655 | } | |
656 | ||
657 | /* Lock the orphan dir. The lock will be held for the entire | |
658 | * delete_inode operation. We do this now to avoid races with | |
659 | * recovery completion on other nodes. */ | |
1b1dcc1b | 660 | mutex_lock(&orphan_dir_inode->i_mutex); |
4bcec184 | 661 | status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1); |
ccd979bd | 662 | if (status < 0) { |
1b1dcc1b | 663 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
664 | |
665 | mlog_errno(status); | |
666 | goto bail; | |
667 | } | |
668 | ||
669 | /* we do this while holding the orphan dir lock because we | |
670 | * don't want recovery being run from another node to vote for | |
671 | * an inode delete on us -- this will result in two nodes | |
672 | * truncating the same file! */ | |
673 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); | |
674 | if (status < 0) { | |
675 | mlog_errno(status); | |
676 | goto bail_unlock_dir; | |
677 | } | |
678 | ||
679 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, | |
680 | orphan_dir_bh); | |
681 | if (status < 0) | |
682 | mlog_errno(status); | |
683 | ||
684 | bail_unlock_dir: | |
685 | ocfs2_meta_unlock(orphan_dir_inode, 1); | |
1b1dcc1b | 686 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
687 | brelse(orphan_dir_bh); |
688 | bail: | |
689 | iput(orphan_dir_inode); | |
b4df6ed8 | 690 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
ccd979bd MF |
691 | |
692 | return status; | |
693 | } | |
694 | ||
695 | /* There is a series of simple checks that should be done before a | |
696 | * vote is even considered. Encapsulate those in this function. */ | |
697 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) | |
698 | { | |
699 | int ret = 0; | |
700 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
701 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
702 | ||
703 | /* We shouldn't be getting here for the root directory | |
704 | * inode.. */ | |
705 | if (inode == osb->root_inode) { | |
706 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); | |
707 | goto bail; | |
708 | } | |
709 | ||
710 | /* If we're coming from process_vote we can't go into our own | |
711 | * voting [hello, deadlock city!], so unforuntately we just | |
712 | * have to skip deleting this guy. That's OK though because | |
713 | * the node who's doing the actual deleting should handle it | |
714 | * anyway. */ | |
715 | if (current == osb->vote_task) { | |
716 | mlog(0, "Skipping delete of %lu because we're currently " | |
717 | "in process_vote\n", inode->i_ino); | |
718 | goto bail; | |
719 | } | |
720 | ||
721 | spin_lock(&oi->ip_lock); | |
722 | /* OCFS2 *never* deletes system files. This should technically | |
723 | * never get here as system file inodes should always have a | |
724 | * positive link count. */ | |
725 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { | |
b0697053 MF |
726 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
727 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
728 | goto bail_unlock; |
729 | } | |
730 | ||
731 | /* If we have voted "yes" on the wipe of this inode for | |
732 | * another node, it will be marked here so we can safely skip | |
733 | * it. Recovery will cleanup any inodes we might inadvertantly | |
734 | * skip here. */ | |
735 | if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) { | |
736 | mlog(0, "Skipping delete of %lu because another node " | |
737 | "has done this for us.\n", inode->i_ino); | |
738 | goto bail_unlock; | |
739 | } | |
740 | ||
741 | ret = 1; | |
742 | bail_unlock: | |
743 | spin_unlock(&oi->ip_lock); | |
744 | bail: | |
745 | return ret; | |
746 | } | |
747 | ||
748 | /* Query the cluster to determine whether we should wipe an inode from | |
749 | * disk or not. | |
750 | * | |
751 | * Requires the inode to have the cluster lock. */ | |
752 | static int ocfs2_query_inode_wipe(struct inode *inode, | |
753 | struct buffer_head *di_bh, | |
754 | int *wipe) | |
755 | { | |
756 | int status = 0; | |
757 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
758 | struct ocfs2_dinode *di; | |
759 | ||
760 | *wipe = 0; | |
761 | ||
762 | /* While we were waiting for the cluster lock in | |
763 | * ocfs2_delete_inode, another node might have asked to delete | |
764 | * the inode. Recheck our flags to catch this. */ | |
765 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
b0697053 MF |
766 | mlog(0, "Skipping delete of %llu because flags changed\n", |
767 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
768 | goto bail; |
769 | } | |
770 | ||
771 | /* Now that we have an up to date inode, we can double check | |
772 | * the link count. */ | |
773 | if (inode->i_nlink) { | |
b0697053 MF |
774 | mlog(0, "Skipping delete of %llu because nlink = %u\n", |
775 | (unsigned long long)oi->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
776 | goto bail; |
777 | } | |
778 | ||
779 | /* Do some basic inode verification... */ | |
780 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
781 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) { | |
782 | /* for lack of a better error? */ | |
783 | status = -EEXIST; | |
784 | mlog(ML_ERROR, | |
b0697053 | 785 | "Inode %llu (on-disk %llu) not orphaned! " |
ccd979bd | 786 | "Disk flags 0x%x, inode flags 0x%x\n", |
b0697053 MF |
787 | (unsigned long long)oi->ip_blkno, |
788 | (unsigned long long)di->i_blkno, di->i_flags, | |
789 | oi->ip_flags); | |
ccd979bd MF |
790 | goto bail; |
791 | } | |
792 | ||
793 | /* has someone already deleted us?! baaad... */ | |
794 | if (di->i_dtime) { | |
795 | status = -EEXIST; | |
796 | mlog_errno(status); | |
797 | goto bail; | |
798 | } | |
799 | ||
6f16bf65 MF |
800 | /* |
801 | * This is how ocfs2 determines whether an inode is still live | |
802 | * within the cluster. Every node takes a shared read lock on | |
803 | * the inode open lock in ocfs2_read_locked_inode(). When we | |
804 | * get to ->delete_inode(), each node tries to convert it's | |
805 | * lock to an exclusive. Trylocks are serialized by the inode | |
806 | * meta data lock. If the upconvert suceeds, we know the inode | |
807 | * is no longer live and can be deleted. | |
808 | * | |
809 | * Though we call this with the meta data lock held, the | |
810 | * trylock keeps us from ABBA deadlock. | |
811 | */ | |
812 | status = ocfs2_try_open_lock(inode, 1); | |
50008630 | 813 | if (status == -EAGAIN) { |
ccd979bd | 814 | status = 0; |
b0697053 MF |
815 | mlog(0, "Skipping delete of %llu because it is in use on" |
816 | "other nodes\n", (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
817 | goto bail; |
818 | } | |
819 | if (status < 0) { | |
820 | mlog_errno(status); | |
821 | goto bail; | |
822 | } | |
823 | ||
50008630 TY |
824 | *wipe = 1; |
825 | mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n", | |
826 | (unsigned long long)oi->ip_blkno, | |
827 | le16_to_cpu(di->i_orphaned_slot)); | |
ccd979bd MF |
828 | |
829 | bail: | |
830 | return status; | |
831 | } | |
832 | ||
833 | /* Support function for ocfs2_delete_inode. Will help us keep the | |
834 | * inode data in a consistent state for clear_inode. Always truncates | |
835 | * pages, optionally sync's them first. */ | |
836 | static void ocfs2_cleanup_delete_inode(struct inode *inode, | |
837 | int sync_data) | |
838 | { | |
b0697053 MF |
839 | mlog(0, "Cleanup inode %llu, sync = %d\n", |
840 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); | |
ccd979bd MF |
841 | if (sync_data) |
842 | write_inode_now(inode, 1); | |
843 | truncate_inode_pages(&inode->i_data, 0); | |
844 | } | |
845 | ||
846 | void ocfs2_delete_inode(struct inode *inode) | |
847 | { | |
848 | int wipe, status; | |
849 | sigset_t blocked, oldset; | |
850 | struct buffer_head *di_bh = NULL; | |
851 | ||
852 | mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino); | |
853 | ||
854 | if (is_bad_inode(inode)) { | |
855 | mlog(0, "Skipping delete of bad inode\n"); | |
856 | goto bail; | |
857 | } | |
858 | ||
859 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
860 | /* It's probably not necessary to truncate_inode_pages | |
861 | * here but we do it for safety anyway (it will most | |
862 | * likely be a no-op anyway) */ | |
863 | ocfs2_cleanup_delete_inode(inode, 0); | |
864 | goto bail; | |
865 | } | |
866 | ||
867 | /* We want to block signals in delete_inode as the lock and | |
868 | * messaging paths may return us -ERESTARTSYS. Which would | |
869 | * cause us to exit early, resulting in inodes being orphaned | |
870 | * forever. */ | |
871 | sigfillset(&blocked); | |
872 | status = sigprocmask(SIG_BLOCK, &blocked, &oldset); | |
873 | if (status < 0) { | |
874 | mlog_errno(status); | |
875 | ocfs2_cleanup_delete_inode(inode, 1); | |
876 | goto bail; | |
877 | } | |
878 | ||
879 | /* Lock down the inode. This gives us an up to date view of | |
880 | * it's metadata (for verification), and allows us to | |
881 | * serialize delete_inode votes. | |
882 | * | |
883 | * Even though we might be doing a truncate, we don't take the | |
884 | * allocation lock here as it won't be needed - nobody will | |
885 | * have the file open. | |
886 | */ | |
4bcec184 | 887 | status = ocfs2_meta_lock(inode, &di_bh, 1); |
ccd979bd MF |
888 | if (status < 0) { |
889 | if (status != -ENOENT) | |
890 | mlog_errno(status); | |
891 | ocfs2_cleanup_delete_inode(inode, 0); | |
892 | goto bail_unblock; | |
893 | } | |
894 | ||
895 | /* Query the cluster. This will be the final decision made | |
896 | * before we go ahead and wipe the inode. */ | |
897 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); | |
898 | if (!wipe || status < 0) { | |
899 | /* Error and inode busy vote both mean we won't be | |
900 | * removing the inode, so they take almost the same | |
901 | * path. */ | |
902 | if (status < 0) | |
903 | mlog_errno(status); | |
904 | ||
905 | /* Someone in the cluster has voted to not wipe this | |
906 | * inode, or it was never completely orphaned. Write | |
907 | * out the pages and exit now. */ | |
908 | ocfs2_cleanup_delete_inode(inode, 1); | |
909 | goto bail_unlock_inode; | |
910 | } | |
911 | ||
912 | ocfs2_cleanup_delete_inode(inode, 0); | |
913 | ||
914 | status = ocfs2_wipe_inode(inode, di_bh); | |
915 | if (status < 0) { | |
b4df6ed8 MF |
916 | if (status != -EDEADLK) |
917 | mlog_errno(status); | |
ccd979bd MF |
918 | goto bail_unlock_inode; |
919 | } | |
920 | ||
24c19ef4 MF |
921 | /* |
922 | * Mark the inode as successfully deleted. | |
923 | * | |
924 | * This is important for ocfs2_clear_inode() as it will check | |
925 | * this flag and skip any checkpointing work | |
926 | * | |
927 | * ocfs2_stuff_meta_lvb() also uses this flag to invalidate | |
928 | * the LVB for other nodes. | |
929 | */ | |
ccd979bd MF |
930 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; |
931 | ||
932 | bail_unlock_inode: | |
933 | ocfs2_meta_unlock(inode, 1); | |
934 | brelse(di_bh); | |
935 | bail_unblock: | |
936 | status = sigprocmask(SIG_SETMASK, &oldset, NULL); | |
937 | if (status < 0) | |
938 | mlog_errno(status); | |
939 | bail: | |
940 | clear_inode(inode); | |
941 | mlog_exit_void(); | |
942 | } | |
943 | ||
944 | void ocfs2_clear_inode(struct inode *inode) | |
945 | { | |
946 | int status; | |
947 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
948 | ||
949 | mlog_entry_void(); | |
950 | ||
951 | if (!inode) | |
952 | goto bail; | |
953 | ||
b0697053 MF |
954 | mlog(0, "Clearing inode: %llu, nlink = %u\n", |
955 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
956 | |
957 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, | |
958 | "Inode=%lu\n", inode->i_ino); | |
959 | ||
50008630 TY |
960 | /* For remove delete_inode vote, we hold open lock before, |
961 | * now it is time to unlock PR and EX open locks. */ | |
962 | ocfs2_open_unlock(inode); | |
963 | ||
ccd979bd MF |
964 | /* Do these before all the other work so that we don't bounce |
965 | * the vote thread while waiting to destroy the locks. */ | |
966 | ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres); | |
967 | ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres); | |
968 | ocfs2_mark_lockres_freeing(&oi->ip_data_lockres); | |
50008630 | 969 | ocfs2_mark_lockres_freeing(&oi->ip_open_lockres); |
ccd979bd MF |
970 | |
971 | /* We very well may get a clear_inode before all an inodes | |
972 | * metadata has hit disk. Of course, we can't drop any cluster | |
973 | * locks until the journal has finished with it. The only | |
974 | * exception here are successfully wiped inodes - their | |
975 | * metadata can now be considered to be part of the system | |
976 | * inodes from which it came. */ | |
977 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) | |
978 | ocfs2_checkpoint_inode(inode); | |
979 | ||
980 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), | |
b0697053 MF |
981 | "Clear inode of %llu, inode has io markers\n", |
982 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 983 | |
ccd979bd MF |
984 | status = ocfs2_drop_inode_locks(inode); |
985 | if (status < 0) | |
986 | mlog_errno(status); | |
987 | ||
988 | ocfs2_lock_res_free(&oi->ip_rw_lockres); | |
989 | ocfs2_lock_res_free(&oi->ip_meta_lockres); | |
990 | ocfs2_lock_res_free(&oi->ip_data_lockres); | |
50008630 | 991 | ocfs2_lock_res_free(&oi->ip_open_lockres); |
ccd979bd MF |
992 | |
993 | ocfs2_metadata_cache_purge(inode); | |
994 | ||
995 | mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached, | |
b0697053 MF |
996 | "Clear inode of %llu, inode has %u cache items\n", |
997 | (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached); | |
ccd979bd MF |
998 | |
999 | mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE), | |
b0697053 MF |
1000 | "Clear inode of %llu, inode has a bad flag\n", |
1001 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1002 | |
1003 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), | |
b0697053 MF |
1004 | "Clear inode of %llu, inode is locked\n", |
1005 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1006 | |
251b6ecc | 1007 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
b0697053 MF |
1008 | "Clear inode of %llu, io_mutex is locked\n", |
1009 | (unsigned long long)oi->ip_blkno); | |
251b6ecc | 1010 | mutex_unlock(&oi->ip_io_mutex); |
ccd979bd MF |
1011 | |
1012 | /* | |
1013 | * down_trylock() returns 0, down_write_trylock() returns 1 | |
1014 | * kernel 1, world 0 | |
1015 | */ | |
1016 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), | |
b0697053 MF |
1017 | "Clear inode of %llu, alloc_sem is locked\n", |
1018 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1019 | up_write(&oi->ip_alloc_sem); |
1020 | ||
1021 | mlog_bug_on_msg(oi->ip_open_count, | |
b0697053 MF |
1022 | "Clear inode of %llu has open count %d\n", |
1023 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); | |
ccd979bd MF |
1024 | |
1025 | /* Clear all other flags. */ | |
1026 | oi->ip_flags = OCFS2_INODE_CACHE_INLINE; | |
1027 | oi->ip_created_trans = 0; | |
1028 | oi->ip_last_trans = 0; | |
1029 | oi->ip_dir_start_lookup = 0; | |
1030 | oi->ip_blkno = 0ULL; | |
1031 | ||
1032 | bail: | |
1033 | mlog_exit_void(); | |
1034 | } | |
1035 | ||
1036 | /* Called under inode_lock, with no more references on the | |
1037 | * struct inode, so it's safe here to check the flags field | |
1038 | * and to manipulate i_nlink without any other locks. */ | |
1039 | void ocfs2_drop_inode(struct inode *inode) | |
1040 | { | |
1041 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1042 | ||
1043 | mlog_entry_void(); | |
1044 | ||
b0697053 MF |
1045 | mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n", |
1046 | (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags); | |
ccd979bd | 1047 | |
379dfe9d MF |
1048 | if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) |
1049 | generic_delete_inode(inode); | |
1050 | else | |
1051 | generic_drop_inode(inode); | |
ccd979bd MF |
1052 | |
1053 | mlog_exit_void(); | |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * TODO: this should probably be merged into ocfs2_get_block | |
1058 | * | |
1059 | * However, you now need to pay attention to the cont_prepare_write() | |
1060 | * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much | |
1061 | * expects never to extend). | |
1062 | */ | |
1063 | struct buffer_head *ocfs2_bread(struct inode *inode, | |
1064 | int block, int *err, int reada) | |
1065 | { | |
1066 | struct buffer_head *bh = NULL; | |
1067 | int tmperr; | |
1068 | u64 p_blkno; | |
1069 | int readflags = OCFS2_BH_CACHED; | |
1070 | ||
ccd979bd MF |
1071 | if (reada) |
1072 | readflags |= OCFS2_BH_READAHEAD; | |
ccd979bd MF |
1073 | |
1074 | if (((u64)block << inode->i_sb->s_blocksize_bits) >= | |
1075 | i_size_read(inode)) { | |
1076 | BUG_ON(!reada); | |
1077 | return NULL; | |
1078 | } | |
1079 | ||
363041a5 | 1080 | tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL); |
ccd979bd MF |
1081 | if (tmperr < 0) { |
1082 | mlog_errno(tmperr); | |
1083 | goto fail; | |
1084 | } | |
1085 | ||
1086 | tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh, | |
1087 | readflags, inode); | |
1088 | if (tmperr < 0) | |
1089 | goto fail; | |
1090 | ||
1091 | tmperr = 0; | |
1092 | ||
1093 | *err = 0; | |
1094 | return bh; | |
1095 | ||
1096 | fail: | |
1097 | if (bh) { | |
1098 | brelse(bh); | |
1099 | bh = NULL; | |
1100 | } | |
1101 | *err = -EIO; | |
1102 | return NULL; | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * This is called from our getattr. | |
1107 | */ | |
1108 | int ocfs2_inode_revalidate(struct dentry *dentry) | |
1109 | { | |
1110 | struct inode *inode = dentry->d_inode; | |
1111 | int status = 0; | |
1112 | ||
b0697053 MF |
1113 | mlog_entry("(inode = 0x%p, ino = %llu)\n", inode, |
1114 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL); | |
ccd979bd MF |
1115 | |
1116 | if (!inode) { | |
1117 | mlog(0, "eep, no inode!\n"); | |
1118 | status = -ENOENT; | |
1119 | goto bail; | |
1120 | } | |
1121 | ||
1122 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1123 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
1124 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1125 | mlog(0, "inode deleted!\n"); | |
1126 | status = -ENOENT; | |
1127 | goto bail; | |
1128 | } | |
1129 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1130 | ||
1131 | /* Let ocfs2_meta_lock do the work of updating our struct | |
1132 | * inode for us. */ | |
4bcec184 | 1133 | status = ocfs2_meta_lock(inode, NULL, 0); |
ccd979bd MF |
1134 | if (status < 0) { |
1135 | if (status != -ENOENT) | |
1136 | mlog_errno(status); | |
1137 | goto bail; | |
1138 | } | |
1139 | ocfs2_meta_unlock(inode, 0); | |
1140 | bail: | |
1141 | mlog_exit(status); | |
1142 | ||
1143 | return status; | |
1144 | } | |
1145 | ||
1146 | /* | |
1147 | * Updates a disk inode from a | |
1148 | * struct inode. | |
1149 | * Only takes ip_lock. | |
1150 | */ | |
1fabe148 | 1151 | int ocfs2_mark_inode_dirty(handle_t *handle, |
ccd979bd MF |
1152 | struct inode *inode, |
1153 | struct buffer_head *bh) | |
1154 | { | |
1155 | int status; | |
1156 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
1157 | ||
b0697053 MF |
1158 | mlog_entry("(inode %llu)\n", |
1159 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
1160 | |
1161 | status = ocfs2_journal_access(handle, inode, bh, | |
1162 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1163 | if (status < 0) { | |
1164 | mlog_errno(status); | |
1165 | goto leave; | |
1166 | } | |
1167 | ||
1168 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1169 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
ca4d147e | 1170 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
ccd979bd MF |
1171 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1172 | ||
1173 | fe->i_size = cpu_to_le64(i_size_read(inode)); | |
1174 | fe->i_links_count = cpu_to_le16(inode->i_nlink); | |
1175 | fe->i_uid = cpu_to_le32(inode->i_uid); | |
1176 | fe->i_gid = cpu_to_le32(inode->i_gid); | |
1177 | fe->i_mode = cpu_to_le16(inode->i_mode); | |
1178 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); | |
1179 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
1180 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
1181 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
1182 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1183 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
1184 | ||
1185 | status = ocfs2_journal_dirty(handle, bh); | |
1186 | if (status < 0) | |
1187 | mlog_errno(status); | |
1188 | ||
1189 | status = 0; | |
1190 | leave: | |
1191 | ||
1192 | mlog_exit(status); | |
1193 | return status; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * | |
1198 | * Updates a struct inode from a disk inode. | |
1199 | * does no i/o, only takes ip_lock. | |
1200 | */ | |
1201 | void ocfs2_refresh_inode(struct inode *inode, | |
1202 | struct ocfs2_dinode *fe) | |
1203 | { | |
ccd979bd MF |
1204 | spin_lock(&OCFS2_I(inode)->ip_lock); |
1205 | ||
1206 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e HP |
1207 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
1208 | ocfs2_set_inode_flags(inode); | |
ccd979bd MF |
1209 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
1210 | inode->i_nlink = le16_to_cpu(fe->i_links_count); | |
1211 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
1212 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
1213 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
ccd979bd MF |
1214 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) |
1215 | inode->i_blocks = 0; | |
1216 | else | |
1217 | inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode)); | |
1218 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); | |
1219 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
1220 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
1221 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
1222 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
1223 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
1224 | ||
1225 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1226 | } |