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> | |
a90714c1 | 31 | #include <linux/quotaops.h> |
ccd979bd MF |
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" | |
9b7895ef | 41 | #include "dir.h" |
d6b32bbb | 42 | #include "blockcheck.h" |
ccd979bd MF |
43 | #include "dlmglue.h" |
44 | #include "extent_map.h" | |
45 | #include "file.h" | |
b4df6ed8 | 46 | #include "heartbeat.h" |
ccd979bd MF |
47 | #include "inode.h" |
48 | #include "journal.h" | |
49 | #include "namei.h" | |
50 | #include "suballoc.h" | |
51 | #include "super.h" | |
52 | #include "symlink.h" | |
53 | #include "sysfile.h" | |
54 | #include "uptodate.h" | |
cf1d6c76 | 55 | #include "xattr.h" |
ccd979bd MF |
56 | |
57 | #include "buffer_head_io.h" | |
58 | ||
ccd979bd MF |
59 | struct ocfs2_find_inode_args |
60 | { | |
61 | u64 fi_blkno; | |
62 | unsigned long fi_ino; | |
63 | unsigned int fi_flags; | |
5fa0613e | 64 | unsigned int fi_sysfile_type; |
ccd979bd MF |
65 | }; |
66 | ||
5fa0613e JK |
67 | static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES]; |
68 | ||
ccd979bd MF |
69 | static int ocfs2_read_locked_inode(struct inode *inode, |
70 | struct ocfs2_find_inode_args *args); | |
71 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); | |
72 | static int ocfs2_find_actor(struct inode *inode, void *opaque); | |
73 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
74 | struct inode *inode, | |
75 | struct buffer_head *fe_bh); | |
76 | ||
ca4d147e HP |
77 | void ocfs2_set_inode_flags(struct inode *inode) |
78 | { | |
79 | unsigned int flags = OCFS2_I(inode)->ip_attr; | |
80 | ||
81 | inode->i_flags &= ~(S_IMMUTABLE | | |
82 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); | |
83 | ||
84 | if (flags & OCFS2_IMMUTABLE_FL) | |
85 | inode->i_flags |= S_IMMUTABLE; | |
86 | ||
87 | if (flags & OCFS2_SYNC_FL) | |
88 | inode->i_flags |= S_SYNC; | |
89 | if (flags & OCFS2_APPEND_FL) | |
90 | inode->i_flags |= S_APPEND; | |
91 | if (flags & OCFS2_NOATIME_FL) | |
92 | inode->i_flags |= S_NOATIME; | |
93 | if (flags & OCFS2_DIRSYNC_FL) | |
94 | inode->i_flags |= S_DIRSYNC; | |
95 | } | |
96 | ||
6e4b0d56 JK |
97 | /* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */ |
98 | void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi) | |
99 | { | |
100 | unsigned int flags = oi->vfs_inode.i_flags; | |
101 | ||
102 | oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL| | |
103 | OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL); | |
104 | if (flags & S_SYNC) | |
105 | oi->ip_attr |= OCFS2_SYNC_FL; | |
106 | if (flags & S_APPEND) | |
107 | oi->ip_attr |= OCFS2_APPEND_FL; | |
108 | if (flags & S_IMMUTABLE) | |
109 | oi->ip_attr |= OCFS2_IMMUTABLE_FL; | |
110 | if (flags & S_NOATIME) | |
111 | oi->ip_attr |= OCFS2_NOATIME_FL; | |
112 | if (flags & S_DIRSYNC) | |
113 | oi->ip_attr |= OCFS2_DIRSYNC_FL; | |
114 | } | |
115 | ||
6ca497a8 | 116 | struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno) |
117 | { | |
118 | struct ocfs2_find_inode_args args; | |
119 | ||
120 | args.fi_blkno = blkno; | |
121 | args.fi_flags = 0; | |
122 | args.fi_ino = ino_from_blkno(sb, blkno); | |
123 | args.fi_sysfile_type = 0; | |
124 | ||
125 | return ilookup5(sb, blkno, ocfs2_find_actor, &args); | |
126 | } | |
5fa0613e JK |
127 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags, |
128 | int sysfile_type) | |
ccd979bd MF |
129 | { |
130 | struct inode *inode = NULL; | |
131 | struct super_block *sb = osb->sb; | |
132 | struct ocfs2_find_inode_args args; | |
133 | ||
b0697053 | 134 | mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno); |
ccd979bd MF |
135 | |
136 | /* Ok. By now we've either got the offsets passed to us by the | |
137 | * caller, or we just pulled them off the bh. Lets do some | |
138 | * sanity checks to make sure they're OK. */ | |
139 | if (blkno == 0) { | |
140 | inode = ERR_PTR(-EINVAL); | |
141 | mlog_errno(PTR_ERR(inode)); | |
142 | goto bail; | |
143 | } | |
144 | ||
145 | args.fi_blkno = blkno; | |
24c19ef4 | 146 | args.fi_flags = flags; |
ccd979bd | 147 | args.fi_ino = ino_from_blkno(sb, blkno); |
5fa0613e | 148 | args.fi_sysfile_type = sysfile_type; |
ccd979bd MF |
149 | |
150 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, | |
151 | ocfs2_init_locked_inode, &args); | |
152 | /* inode was *not* in the inode cache. 2.6.x requires | |
153 | * us to do our own read_inode call and unlock it | |
154 | * afterwards. */ | |
155 | if (inode && inode->i_state & I_NEW) { | |
156 | mlog(0, "Inode was not in inode cache, reading it.\n"); | |
157 | ocfs2_read_locked_inode(inode, &args); | |
158 | unlock_new_inode(inode); | |
159 | } | |
160 | if (inode == NULL) { | |
161 | inode = ERR_PTR(-ENOMEM); | |
162 | mlog_errno(PTR_ERR(inode)); | |
163 | goto bail; | |
164 | } | |
165 | if (is_bad_inode(inode)) { | |
166 | iput(inode); | |
167 | inode = ERR_PTR(-ESTALE); | |
ccd979bd MF |
168 | goto bail; |
169 | } | |
170 | ||
171 | bail: | |
172 | if (!IS_ERR(inode)) { | |
b0697053 MF |
173 | mlog(0, "returning inode with number %llu\n", |
174 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 175 | mlog_exit_ptr(inode); |
6a1bd4a5 | 176 | } |
ccd979bd MF |
177 | |
178 | return inode; | |
179 | } | |
180 | ||
181 | ||
182 | /* | |
183 | * here's how inodes get read from disk: | |
184 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR | |
185 | * found? : return the in-memory inode | |
186 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE | |
187 | */ | |
188 | ||
189 | static int ocfs2_find_actor(struct inode *inode, void *opaque) | |
190 | { | |
191 | struct ocfs2_find_inode_args *args = NULL; | |
192 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
193 | int ret = 0; | |
194 | ||
195 | mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque); | |
196 | ||
197 | args = opaque; | |
198 | ||
199 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); | |
200 | ||
201 | if (oi->ip_blkno != args->fi_blkno) | |
202 | goto bail; | |
203 | ||
ccd979bd MF |
204 | ret = 1; |
205 | bail: | |
206 | mlog_exit(ret); | |
207 | return ret; | |
208 | } | |
209 | ||
210 | /* | |
211 | * initialize the new inode, but don't do anything that would cause | |
212 | * us to sleep. | |
213 | * return 0 on success, 1 on failure | |
214 | */ | |
215 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) | |
216 | { | |
217 | struct ocfs2_find_inode_args *args = opaque; | |
218 | ||
219 | mlog_entry("inode = %p, opaque = %p\n", inode, opaque); | |
220 | ||
221 | inode->i_ino = args->fi_ino; | |
222 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; | |
5fa0613e JK |
223 | if (args->fi_sysfile_type != 0) |
224 | lockdep_set_class(&inode->i_mutex, | |
225 | &ocfs2_sysfile_lock_key[args->fi_sysfile_type]); | |
ccd979bd MF |
226 | |
227 | mlog_exit(0); | |
228 | return 0; | |
229 | } | |
230 | ||
b657c95c JB |
231 | void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, |
232 | int create_ino) | |
ccd979bd MF |
233 | { |
234 | struct super_block *sb; | |
235 | struct ocfs2_super *osb; | |
53da4939 | 236 | int use_plocks = 1; |
ccd979bd | 237 | |
b0697053 | 238 | mlog_entry("(0x%p, size:%llu)\n", inode, |
1ca1a111 | 239 | (unsigned long long)le64_to_cpu(fe->i_size)); |
ccd979bd MF |
240 | |
241 | sb = inode->i_sb; | |
242 | osb = OCFS2_SB(sb); | |
243 | ||
53da4939 MF |
244 | if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) || |
245 | ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks()) | |
246 | use_plocks = 0; | |
247 | ||
b657c95c JB |
248 | /* |
249 | * These have all been checked by ocfs2_read_inode_block() or set | |
250 | * by ocfs2_mknod_locked(), so a failure is a code bug. | |
251 | */ | |
252 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode | |
253 | cannot create a superblock | |
254 | inode today. change if | |
255 | that is needed. */ | |
256 | BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))); | |
257 | BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation); | |
ccd979bd | 258 | |
ccd979bd | 259 | |
8110b073 MF |
260 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
261 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); | |
15b1e36b | 262 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
8110b073 | 263 | |
ccd979bd MF |
264 | inode->i_version = 1; |
265 | inode->i_generation = le32_to_cpu(fe->i_generation); | |
266 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
267 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
268 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
269 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
ccd979bd MF |
270 | |
271 | /* Fast symlinks will have i_size but no allocated clusters. */ | |
272 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) | |
273 | inode->i_blocks = 0; | |
274 | else | |
8110b073 | 275 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 276 | inode->i_mapping->a_ops = &ocfs2_aops; |
ccd979bd MF |
277 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
278 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
279 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
280 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
281 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
282 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
283 | ||
284 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) | |
285 | mlog(ML_ERROR, | |
b0697053 MF |
286 | "ip_blkno %llu != i_blkno %llu!\n", |
287 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1ca1a111 | 288 | (unsigned long long)le64_to_cpu(fe->i_blkno)); |
ccd979bd | 289 | |
198a1ca3 | 290 | inode->i_nlink = ocfs2_read_links_count(fe); |
ccd979bd | 291 | |
bbbd0eb3 | 292 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) { |
24c19ef4 | 293 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; |
bbbd0eb3 JK |
294 | inode->i_flags |= S_NOQUOTA; |
295 | } | |
24c19ef4 | 296 | |
ccd979bd MF |
297 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { |
298 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
299 | mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino); | |
300 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { | |
301 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
1a224ad1 JK |
302 | } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) { |
303 | inode->i_flags |= S_NOQUOTA; | |
ccd979bd MF |
304 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { |
305 | mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino); | |
306 | /* we can't actually hit this as read_inode can't | |
307 | * handle superblocks today ;-) */ | |
308 | BUG(); | |
309 | } | |
310 | ||
311 | switch (inode->i_mode & S_IFMT) { | |
312 | case S_IFREG: | |
53da4939 MF |
313 | if (use_plocks) |
314 | inode->i_fop = &ocfs2_fops; | |
315 | else | |
316 | inode->i_fop = &ocfs2_fops_no_plocks; | |
ccd979bd MF |
317 | inode->i_op = &ocfs2_file_iops; |
318 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
319 | break; | |
320 | case S_IFDIR: | |
321 | inode->i_op = &ocfs2_dir_iops; | |
53da4939 MF |
322 | if (use_plocks) |
323 | inode->i_fop = &ocfs2_dops; | |
324 | else | |
325 | inode->i_fop = &ocfs2_dops_no_plocks; | |
ccd979bd MF |
326 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
327 | break; | |
328 | case S_IFLNK: | |
329 | if (ocfs2_inode_is_fast_symlink(inode)) | |
330 | inode->i_op = &ocfs2_fast_symlink_inode_operations; | |
331 | else | |
332 | inode->i_op = &ocfs2_symlink_inode_operations; | |
333 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
334 | break; | |
335 | default: | |
336 | inode->i_op = &ocfs2_special_file_iops; | |
337 | init_special_inode(inode, inode->i_mode, | |
338 | inode->i_rdev); | |
339 | break; | |
340 | } | |
341 | ||
24c19ef4 MF |
342 | if (create_ino) { |
343 | inode->i_ino = ino_from_blkno(inode->i_sb, | |
344 | le64_to_cpu(fe->i_blkno)); | |
345 | ||
346 | /* | |
347 | * If we ever want to create system files from kernel, | |
348 | * the generation argument to | |
349 | * ocfs2_inode_lock_res_init() will have to change. | |
350 | */ | |
1ca1a111 | 351 | BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL); |
24c19ef4 | 352 | |
e63aecb6 | 353 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres, |
24c19ef4 | 354 | OCFS2_LOCK_TYPE_META, 0, inode); |
50008630 TY |
355 | |
356 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, | |
357 | OCFS2_LOCK_TYPE_OPEN, 0, inode); | |
24c19ef4 MF |
358 | } |
359 | ||
ccd979bd | 360 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, |
24c19ef4 MF |
361 | OCFS2_LOCK_TYPE_RW, inode->i_generation, |
362 | inode); | |
363 | ||
ca4d147e | 364 | ocfs2_set_inode_flags(inode); |
ca4d147e | 365 | |
13821151 TM |
366 | OCFS2_I(inode)->ip_last_used_slot = 0; |
367 | OCFS2_I(inode)->ip_last_used_group = 0; | |
b657c95c | 368 | mlog_exit_void(); |
ccd979bd MF |
369 | } |
370 | ||
371 | static int ocfs2_read_locked_inode(struct inode *inode, | |
372 | struct ocfs2_find_inode_args *args) | |
373 | { | |
374 | struct super_block *sb; | |
375 | struct ocfs2_super *osb; | |
376 | struct ocfs2_dinode *fe; | |
377 | struct buffer_head *bh = NULL; | |
24c19ef4 MF |
378 | int status, can_lock; |
379 | u32 generation = 0; | |
ccd979bd MF |
380 | |
381 | mlog_entry("(0x%p, 0x%p)\n", inode, args); | |
382 | ||
383 | status = -EINVAL; | |
384 | if (inode == NULL || inode->i_sb == NULL) { | |
385 | mlog(ML_ERROR, "bad inode\n"); | |
24c19ef4 | 386 | return status; |
ccd979bd MF |
387 | } |
388 | sb = inode->i_sb; | |
389 | osb = OCFS2_SB(sb); | |
390 | ||
391 | if (!args) { | |
392 | mlog(ML_ERROR, "bad inode args\n"); | |
393 | make_bad_inode(inode); | |
24c19ef4 MF |
394 | return status; |
395 | } | |
396 | ||
397 | /* | |
398 | * To improve performance of cold-cache inode stats, we take | |
399 | * the cluster lock here if possible. | |
400 | * | |
401 | * Generally, OCFS2 never trusts the contents of an inode | |
402 | * unless it's holding a cluster lock, so taking it here isn't | |
403 | * a correctness issue as much as it is a performance | |
404 | * improvement. | |
405 | * | |
406 | * There are three times when taking the lock is not a good idea: | |
407 | * | |
408 | * 1) During startup, before we have initialized the DLM. | |
409 | * | |
410 | * 2) If we are reading certain system files which never get | |
411 | * cluster locks (local alloc, truncate log). | |
412 | * | |
413 | * 3) If the process doing the iget() is responsible for | |
414 | * orphan dir recovery. We're holding the orphan dir lock and | |
415 | * can get into a deadlock with another process on another | |
416 | * node in ->delete_inode(). | |
417 | * | |
418 | * #1 and #2 can be simply solved by never taking the lock | |
419 | * here for system files (which are the only type we read | |
420 | * during mount). It's a heavier approach, but our main | |
421 | * concern is user-accesible files anyway. | |
422 | * | |
423 | * #3 works itself out because we'll eventually take the | |
424 | * cluster lock before trusting anything anyway. | |
425 | */ | |
426 | can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
50008630 | 427 | && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) |
c271c5c2 | 428 | && !ocfs2_mount_local(osb); |
24c19ef4 MF |
429 | |
430 | /* | |
431 | * To maintain backwards compatibility with older versions of | |
432 | * ocfs2-tools, we still store the generation value for system | |
433 | * files. The only ones that actually matter to userspace are | |
434 | * the journals, but it's easier and inexpensive to just flag | |
435 | * all system files similarly. | |
436 | */ | |
437 | if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
438 | generation = osb->fs_generation; | |
439 | ||
e63aecb6 | 440 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres, |
24c19ef4 MF |
441 | OCFS2_LOCK_TYPE_META, |
442 | generation, inode); | |
443 | ||
50008630 TY |
444 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres, |
445 | OCFS2_LOCK_TYPE_OPEN, | |
446 | 0, inode); | |
447 | ||
24c19ef4 | 448 | if (can_lock) { |
50008630 TY |
449 | status = ocfs2_open_lock(inode); |
450 | if (status) { | |
451 | make_bad_inode(inode); | |
452 | mlog_errno(status); | |
453 | return status; | |
454 | } | |
e63aecb6 | 455 | status = ocfs2_inode_lock(inode, NULL, 0); |
24c19ef4 MF |
456 | if (status) { |
457 | make_bad_inode(inode); | |
458 | mlog_errno(status); | |
459 | return status; | |
460 | } | |
ccd979bd MF |
461 | } |
462 | ||
50008630 TY |
463 | if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) { |
464 | status = ocfs2_try_open_lock(inode, 0); | |
465 | if (status) { | |
466 | make_bad_inode(inode); | |
467 | return status; | |
468 | } | |
469 | } | |
470 | ||
b657c95c JB |
471 | if (can_lock) { |
472 | status = ocfs2_read_inode_block_full(inode, &bh, | |
473 | OCFS2_BH_IGNORE_CACHE); | |
474 | } else { | |
da1e9098 | 475 | status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh); |
b657c95c JB |
476 | if (!status) |
477 | status = ocfs2_validate_inode_block(osb->sb, bh); | |
478 | } | |
ccd979bd MF |
479 | if (status < 0) { |
480 | mlog_errno(status); | |
ccd979bd MF |
481 | goto bail; |
482 | } | |
483 | ||
24c19ef4 | 484 | status = -EINVAL; |
ccd979bd | 485 | fe = (struct ocfs2_dinode *) bh->b_data; |
ccd979bd | 486 | |
24c19ef4 MF |
487 | /* |
488 | * This is a code bug. Right now the caller needs to | |
489 | * understand whether it is asking for a system file inode or | |
490 | * not so the proper lock names can be built. | |
491 | */ | |
492 | mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) != | |
493 | !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE), | |
494 | "Inode %llu: system file state is ambigous\n", | |
495 | (unsigned long long)args->fi_blkno); | |
ccd979bd MF |
496 | |
497 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || | |
498 | S_ISBLK(le16_to_cpu(fe->i_mode))) | |
b657c95c | 499 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); |
ccd979bd | 500 | |
b657c95c | 501 | ocfs2_populate_inode(inode, fe, 0); |
ccd979bd MF |
502 | |
503 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); | |
504 | ||
ccd979bd MF |
505 | status = 0; |
506 | ||
507 | bail: | |
24c19ef4 | 508 | if (can_lock) |
e63aecb6 | 509 | ocfs2_inode_unlock(inode, 0); |
24c19ef4 MF |
510 | |
511 | if (status < 0) | |
512 | make_bad_inode(inode); | |
513 | ||
ccd979bd MF |
514 | if (args && bh) |
515 | brelse(bh); | |
516 | ||
517 | mlog_exit(status); | |
518 | return status; | |
519 | } | |
520 | ||
521 | void ocfs2_sync_blockdev(struct super_block *sb) | |
522 | { | |
523 | sync_blockdev(sb->s_bdev); | |
524 | } | |
525 | ||
526 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
527 | struct inode *inode, | |
528 | struct buffer_head *fe_bh) | |
529 | { | |
530 | int status = 0; | |
ccd979bd MF |
531 | struct ocfs2_truncate_context *tc = NULL; |
532 | struct ocfs2_dinode *fe; | |
60b11392 | 533 | handle_t *handle = NULL; |
ccd979bd MF |
534 | |
535 | mlog_entry_void(); | |
536 | ||
537 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
538 | ||
1afc32b9 MF |
539 | /* |
540 | * This check will also skip truncate of inodes with inline | |
541 | * data and fast symlinks. | |
542 | */ | |
3a0782d0 | 543 | if (fe->i_clusters) { |
2b4e30fb JB |
544 | if (ocfs2_should_order_data(inode)) |
545 | ocfs2_begin_ordered_truncate(inode, 0); | |
546 | ||
60b11392 MF |
547 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
548 | if (IS_ERR(handle)) { | |
549 | status = PTR_ERR(handle); | |
550 | mlog_errno(status); | |
551 | goto out; | |
552 | } | |
553 | ||
13723d00 JB |
554 | status = ocfs2_journal_access_di(handle, inode, fe_bh, |
555 | OCFS2_JOURNAL_ACCESS_WRITE); | |
60b11392 MF |
556 | if (status < 0) { |
557 | mlog_errno(status); | |
558 | goto out; | |
559 | } | |
560 | ||
561 | i_size_write(inode, 0); | |
562 | ||
563 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
564 | if (status < 0) { | |
565 | mlog_errno(status); | |
566 | goto out; | |
567 | } | |
568 | ||
569 | ocfs2_commit_trans(osb, handle); | |
570 | handle = NULL; | |
571 | ||
3a0782d0 MF |
572 | status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc); |
573 | if (status < 0) { | |
574 | mlog_errno(status); | |
575 | goto out; | |
576 | } | |
ccd979bd | 577 | |
3a0782d0 MF |
578 | status = ocfs2_commit_truncate(osb, inode, fe_bh, tc); |
579 | if (status < 0) { | |
580 | mlog_errno(status); | |
581 | goto out; | |
582 | } | |
ccd979bd | 583 | } |
ccd979bd | 584 | |
60b11392 MF |
585 | out: |
586 | if (handle) | |
587 | ocfs2_commit_trans(osb, handle); | |
ccd979bd MF |
588 | mlog_exit(status); |
589 | return status; | |
590 | } | |
591 | ||
592 | static int ocfs2_remove_inode(struct inode *inode, | |
593 | struct buffer_head *di_bh, | |
594 | struct inode *orphan_dir_inode, | |
595 | struct buffer_head *orphan_dir_bh) | |
596 | { | |
597 | int status; | |
598 | struct inode *inode_alloc_inode = NULL; | |
599 | struct buffer_head *inode_alloc_bh = NULL; | |
1fabe148 | 600 | handle_t *handle; |
ccd979bd MF |
601 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
602 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
603 | ||
604 | inode_alloc_inode = | |
605 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
606 | le16_to_cpu(di->i_suballoc_slot)); | |
607 | if (!inode_alloc_inode) { | |
608 | status = -EEXIST; | |
609 | mlog_errno(status); | |
610 | goto bail; | |
611 | } | |
612 | ||
1b1dcc1b | 613 | mutex_lock(&inode_alloc_inode->i_mutex); |
e63aecb6 | 614 | status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1); |
ccd979bd | 615 | if (status < 0) { |
1b1dcc1b | 616 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
617 | |
618 | mlog_errno(status); | |
619 | goto bail; | |
620 | } | |
621 | ||
a90714c1 | 622 | handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS + |
9b7895ef | 623 | ocfs2_quota_trans_credits(inode->i_sb)); |
ccd979bd MF |
624 | if (IS_ERR(handle)) { |
625 | status = PTR_ERR(handle); | |
626 | mlog_errno(status); | |
627 | goto bail_unlock; | |
628 | } | |
629 | ||
630 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, | |
631 | orphan_dir_bh); | |
632 | if (status < 0) { | |
633 | mlog_errno(status); | |
634 | goto bail_commit; | |
635 | } | |
636 | ||
637 | /* set the inodes dtime */ | |
13723d00 JB |
638 | status = ocfs2_journal_access_di(handle, inode, di_bh, |
639 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
640 | if (status < 0) { |
641 | mlog_errno(status); | |
642 | goto bail_commit; | |
643 | } | |
644 | ||
645 | di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec); | |
4092d49f | 646 | di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); |
ccd979bd MF |
647 | |
648 | status = ocfs2_journal_dirty(handle, di_bh); | |
649 | if (status < 0) { | |
650 | mlog_errno(status); | |
651 | goto bail_commit; | |
652 | } | |
653 | ||
654 | ocfs2_remove_from_cache(inode, di_bh); | |
a90714c1 | 655 | vfs_dq_free_inode(inode); |
ccd979bd MF |
656 | |
657 | status = ocfs2_free_dinode(handle, inode_alloc_inode, | |
658 | inode_alloc_bh, di); | |
659 | if (status < 0) | |
660 | mlog_errno(status); | |
661 | ||
662 | bail_commit: | |
02dc1af4 | 663 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 664 | bail_unlock: |
e63aecb6 | 665 | ocfs2_inode_unlock(inode_alloc_inode, 1); |
1b1dcc1b | 666 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
667 | brelse(inode_alloc_bh); |
668 | bail: | |
669 | iput(inode_alloc_inode); | |
670 | ||
671 | return status; | |
672 | } | |
673 | ||
b4df6ed8 MF |
674 | /* |
675 | * Serialize with orphan dir recovery. If the process doing | |
676 | * recovery on this orphan dir does an iget() with the dir | |
677 | * i_mutex held, we'll deadlock here. Instead we detect this | |
678 | * and exit early - recovery will wipe this inode for us. | |
679 | */ | |
680 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, | |
681 | int slot) | |
682 | { | |
683 | int ret = 0; | |
684 | ||
685 | spin_lock(&osb->osb_lock); | |
686 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { | |
687 | mlog(0, "Recovery is happening on orphan dir %d, will skip " | |
688 | "this inode\n", slot); | |
689 | ret = -EDEADLK; | |
690 | goto out; | |
691 | } | |
692 | /* This signals to the orphan recovery process that it should | |
693 | * wait for us to handle the wipe. */ | |
694 | osb->osb_orphan_wipes[slot]++; | |
695 | out: | |
696 | spin_unlock(&osb->osb_lock); | |
697 | return ret; | |
698 | } | |
699 | ||
700 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, | |
701 | int slot) | |
702 | { | |
703 | spin_lock(&osb->osb_lock); | |
704 | osb->osb_orphan_wipes[slot]--; | |
705 | spin_unlock(&osb->osb_lock); | |
706 | ||
707 | wake_up(&osb->osb_wipe_event); | |
708 | } | |
709 | ||
ccd979bd MF |
710 | static int ocfs2_wipe_inode(struct inode *inode, |
711 | struct buffer_head *di_bh) | |
712 | { | |
713 | int status, orphaned_slot; | |
714 | struct inode *orphan_dir_inode = NULL; | |
715 | struct buffer_head *orphan_dir_bh = NULL; | |
716 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
50008630 | 717 | struct ocfs2_dinode *di; |
ccd979bd | 718 | |
50008630 TY |
719 | di = (struct ocfs2_dinode *) di_bh->b_data; |
720 | orphaned_slot = le16_to_cpu(di->i_orphaned_slot); | |
b4df6ed8 MF |
721 | |
722 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); | |
723 | if (status) | |
724 | return status; | |
725 | ||
ccd979bd MF |
726 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
727 | ORPHAN_DIR_SYSTEM_INODE, | |
728 | orphaned_slot); | |
729 | if (!orphan_dir_inode) { | |
730 | status = -EEXIST; | |
731 | mlog_errno(status); | |
732 | goto bail; | |
733 | } | |
734 | ||
735 | /* Lock the orphan dir. The lock will be held for the entire | |
736 | * delete_inode operation. We do this now to avoid races with | |
737 | * recovery completion on other nodes. */ | |
1b1dcc1b | 738 | mutex_lock(&orphan_dir_inode->i_mutex); |
e63aecb6 | 739 | status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1); |
ccd979bd | 740 | if (status < 0) { |
1b1dcc1b | 741 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
742 | |
743 | mlog_errno(status); | |
744 | goto bail; | |
745 | } | |
746 | ||
747 | /* we do this while holding the orphan dir lock because we | |
34d024f8 MF |
748 | * don't want recovery being run from another node to try an |
749 | * inode delete underneath us -- this will result in two nodes | |
ccd979bd MF |
750 | * truncating the same file! */ |
751 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); | |
752 | if (status < 0) { | |
753 | mlog_errno(status); | |
754 | goto bail_unlock_dir; | |
755 | } | |
756 | ||
9b7895ef MF |
757 | /* Remove any dir index tree */ |
758 | if (S_ISDIR(inode->i_mode)) { | |
759 | status = ocfs2_dx_dir_truncate(inode, di_bh); | |
760 | if (status) { | |
761 | mlog_errno(status); | |
762 | goto bail_unlock_dir; | |
763 | } | |
764 | } | |
765 | ||
cf1d6c76 TY |
766 | /*Free extended attribute resources associated with this inode.*/ |
767 | status = ocfs2_xattr_remove(inode, di_bh); | |
768 | if (status < 0) { | |
769 | mlog_errno(status); | |
770 | goto bail_unlock_dir; | |
771 | } | |
772 | ||
ccd979bd MF |
773 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, |
774 | orphan_dir_bh); | |
775 | if (status < 0) | |
776 | mlog_errno(status); | |
777 | ||
778 | bail_unlock_dir: | |
e63aecb6 | 779 | ocfs2_inode_unlock(orphan_dir_inode, 1); |
1b1dcc1b | 780 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
781 | brelse(orphan_dir_bh); |
782 | bail: | |
783 | iput(orphan_dir_inode); | |
b4df6ed8 | 784 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
ccd979bd MF |
785 | |
786 | return status; | |
787 | } | |
788 | ||
789 | /* There is a series of simple checks that should be done before a | |
34d024f8 | 790 | * trylock is even considered. Encapsulate those in this function. */ |
ccd979bd MF |
791 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) |
792 | { | |
793 | int ret = 0; | |
794 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
795 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
796 | ||
797 | /* We shouldn't be getting here for the root directory | |
798 | * inode.. */ | |
799 | if (inode == osb->root_inode) { | |
800 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); | |
801 | goto bail; | |
802 | } | |
803 | ||
34d024f8 | 804 | /* If we're coming from downconvert_thread we can't go into our own |
ccd979bd MF |
805 | * voting [hello, deadlock city!], so unforuntately we just |
806 | * have to skip deleting this guy. That's OK though because | |
807 | * the node who's doing the actual deleting should handle it | |
808 | * anyway. */ | |
34d024f8 | 809 | if (current == osb->dc_task) { |
ccd979bd | 810 | mlog(0, "Skipping delete of %lu because we're currently " |
34d024f8 | 811 | "in downconvert\n", inode->i_ino); |
ccd979bd MF |
812 | goto bail; |
813 | } | |
814 | ||
815 | spin_lock(&oi->ip_lock); | |
816 | /* OCFS2 *never* deletes system files. This should technically | |
817 | * never get here as system file inodes should always have a | |
818 | * positive link count. */ | |
819 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { | |
b0697053 MF |
820 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
821 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
822 | goto bail_unlock; |
823 | } | |
824 | ||
34d024f8 MF |
825 | /* If we have allowd wipe of this inode for another node, it |
826 | * will be marked here so we can safely skip it. Recovery will | |
827 | * cleanup any inodes we might inadvertantly skip here. */ | |
ccd979bd MF |
828 | if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) { |
829 | mlog(0, "Skipping delete of %lu because another node " | |
830 | "has done this for us.\n", inode->i_ino); | |
831 | goto bail_unlock; | |
832 | } | |
833 | ||
834 | ret = 1; | |
835 | bail_unlock: | |
836 | spin_unlock(&oi->ip_lock); | |
837 | bail: | |
838 | return ret; | |
839 | } | |
840 | ||
841 | /* Query the cluster to determine whether we should wipe an inode from | |
842 | * disk or not. | |
843 | * | |
844 | * Requires the inode to have the cluster lock. */ | |
845 | static int ocfs2_query_inode_wipe(struct inode *inode, | |
846 | struct buffer_head *di_bh, | |
847 | int *wipe) | |
848 | { | |
849 | int status = 0; | |
850 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
851 | struct ocfs2_dinode *di; | |
852 | ||
853 | *wipe = 0; | |
854 | ||
855 | /* While we were waiting for the cluster lock in | |
856 | * ocfs2_delete_inode, another node might have asked to delete | |
857 | * the inode. Recheck our flags to catch this. */ | |
858 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
b0697053 MF |
859 | mlog(0, "Skipping delete of %llu because flags changed\n", |
860 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
861 | goto bail; |
862 | } | |
863 | ||
864 | /* Now that we have an up to date inode, we can double check | |
865 | * the link count. */ | |
866 | if (inode->i_nlink) { | |
b0697053 MF |
867 | mlog(0, "Skipping delete of %llu because nlink = %u\n", |
868 | (unsigned long long)oi->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
869 | goto bail; |
870 | } | |
871 | ||
872 | /* Do some basic inode verification... */ | |
873 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
874 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) { | |
875 | /* for lack of a better error? */ | |
876 | status = -EEXIST; | |
877 | mlog(ML_ERROR, | |
b0697053 | 878 | "Inode %llu (on-disk %llu) not orphaned! " |
ccd979bd | 879 | "Disk flags 0x%x, inode flags 0x%x\n", |
b0697053 | 880 | (unsigned long long)oi->ip_blkno, |
1ca1a111 MF |
881 | (unsigned long long)le64_to_cpu(di->i_blkno), |
882 | le32_to_cpu(di->i_flags), oi->ip_flags); | |
ccd979bd MF |
883 | goto bail; |
884 | } | |
885 | ||
886 | /* has someone already deleted us?! baaad... */ | |
887 | if (di->i_dtime) { | |
888 | status = -EEXIST; | |
889 | mlog_errno(status); | |
890 | goto bail; | |
891 | } | |
892 | ||
6f16bf65 MF |
893 | /* |
894 | * This is how ocfs2 determines whether an inode is still live | |
895 | * within the cluster. Every node takes a shared read lock on | |
896 | * the inode open lock in ocfs2_read_locked_inode(). When we | |
897 | * get to ->delete_inode(), each node tries to convert it's | |
898 | * lock to an exclusive. Trylocks are serialized by the inode | |
899 | * meta data lock. If the upconvert suceeds, we know the inode | |
900 | * is no longer live and can be deleted. | |
901 | * | |
902 | * Though we call this with the meta data lock held, the | |
903 | * trylock keeps us from ABBA deadlock. | |
904 | */ | |
905 | status = ocfs2_try_open_lock(inode, 1); | |
50008630 | 906 | if (status == -EAGAIN) { |
ccd979bd | 907 | status = 0; |
2759236f | 908 | mlog(0, "Skipping delete of %llu because it is in use on " |
b0697053 | 909 | "other nodes\n", (unsigned long long)oi->ip_blkno); |
ccd979bd MF |
910 | goto bail; |
911 | } | |
912 | if (status < 0) { | |
913 | mlog_errno(status); | |
914 | goto bail; | |
915 | } | |
916 | ||
50008630 TY |
917 | *wipe = 1; |
918 | mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n", | |
919 | (unsigned long long)oi->ip_blkno, | |
920 | le16_to_cpu(di->i_orphaned_slot)); | |
ccd979bd MF |
921 | |
922 | bail: | |
923 | return status; | |
924 | } | |
925 | ||
926 | /* Support function for ocfs2_delete_inode. Will help us keep the | |
927 | * inode data in a consistent state for clear_inode. Always truncates | |
928 | * pages, optionally sync's them first. */ | |
929 | static void ocfs2_cleanup_delete_inode(struct inode *inode, | |
930 | int sync_data) | |
931 | { | |
b0697053 MF |
932 | mlog(0, "Cleanup inode %llu, sync = %d\n", |
933 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); | |
ccd979bd MF |
934 | if (sync_data) |
935 | write_inode_now(inode, 1); | |
936 | truncate_inode_pages(&inode->i_data, 0); | |
937 | } | |
938 | ||
939 | void ocfs2_delete_inode(struct inode *inode) | |
940 | { | |
941 | int wipe, status; | |
942 | sigset_t blocked, oldset; | |
943 | struct buffer_head *di_bh = NULL; | |
944 | ||
945 | mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino); | |
946 | ||
a90714c1 JK |
947 | /* When we fail in read_inode() we mark inode as bad. The second test |
948 | * catches the case when inode allocation fails before allocating | |
949 | * a block for inode. */ | |
950 | if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) { | |
ccd979bd MF |
951 | mlog(0, "Skipping delete of bad inode\n"); |
952 | goto bail; | |
953 | } | |
954 | ||
955 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
956 | /* It's probably not necessary to truncate_inode_pages | |
957 | * here but we do it for safety anyway (it will most | |
958 | * likely be a no-op anyway) */ | |
959 | ocfs2_cleanup_delete_inode(inode, 0); | |
960 | goto bail; | |
961 | } | |
962 | ||
963 | /* We want to block signals in delete_inode as the lock and | |
964 | * messaging paths may return us -ERESTARTSYS. Which would | |
965 | * cause us to exit early, resulting in inodes being orphaned | |
966 | * forever. */ | |
967 | sigfillset(&blocked); | |
968 | status = sigprocmask(SIG_BLOCK, &blocked, &oldset); | |
969 | if (status < 0) { | |
970 | mlog_errno(status); | |
971 | ocfs2_cleanup_delete_inode(inode, 1); | |
972 | goto bail; | |
973 | } | |
974 | ||
6ca497a8 | 975 | /* |
976 | * Synchronize us against ocfs2_get_dentry. We take this in | |
977 | * shared mode so that all nodes can still concurrently | |
978 | * process deletes. | |
979 | */ | |
980 | status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0); | |
981 | if (status < 0) { | |
982 | mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status); | |
983 | ocfs2_cleanup_delete_inode(inode, 0); | |
984 | goto bail_unblock; | |
985 | } | |
ccd979bd MF |
986 | /* Lock down the inode. This gives us an up to date view of |
987 | * it's metadata (for verification), and allows us to | |
34d024f8 | 988 | * serialize delete_inode on multiple nodes. |
ccd979bd MF |
989 | * |
990 | * Even though we might be doing a truncate, we don't take the | |
991 | * allocation lock here as it won't be needed - nobody will | |
992 | * have the file open. | |
993 | */ | |
e63aecb6 | 994 | status = ocfs2_inode_lock(inode, &di_bh, 1); |
ccd979bd MF |
995 | if (status < 0) { |
996 | if (status != -ENOENT) | |
997 | mlog_errno(status); | |
998 | ocfs2_cleanup_delete_inode(inode, 0); | |
6ca497a8 | 999 | goto bail_unlock_nfs_sync; |
ccd979bd MF |
1000 | } |
1001 | ||
1002 | /* Query the cluster. This will be the final decision made | |
1003 | * before we go ahead and wipe the inode. */ | |
1004 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); | |
1005 | if (!wipe || status < 0) { | |
34d024f8 | 1006 | /* Error and remote inode busy both mean we won't be |
ccd979bd MF |
1007 | * removing the inode, so they take almost the same |
1008 | * path. */ | |
1009 | if (status < 0) | |
1010 | mlog_errno(status); | |
1011 | ||
34d024f8 MF |
1012 | /* Someone in the cluster has disallowed a wipe of |
1013 | * this inode, or it was never completely | |
1014 | * orphaned. Write out the pages and exit now. */ | |
ccd979bd MF |
1015 | ocfs2_cleanup_delete_inode(inode, 1); |
1016 | goto bail_unlock_inode; | |
1017 | } | |
1018 | ||
1019 | ocfs2_cleanup_delete_inode(inode, 0); | |
1020 | ||
1021 | status = ocfs2_wipe_inode(inode, di_bh); | |
1022 | if (status < 0) { | |
b4df6ed8 MF |
1023 | if (status != -EDEADLK) |
1024 | mlog_errno(status); | |
ccd979bd MF |
1025 | goto bail_unlock_inode; |
1026 | } | |
1027 | ||
24c19ef4 MF |
1028 | /* |
1029 | * Mark the inode as successfully deleted. | |
1030 | * | |
1031 | * This is important for ocfs2_clear_inode() as it will check | |
1032 | * this flag and skip any checkpointing work | |
1033 | * | |
1034 | * ocfs2_stuff_meta_lvb() also uses this flag to invalidate | |
1035 | * the LVB for other nodes. | |
1036 | */ | |
ccd979bd MF |
1037 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; |
1038 | ||
1039 | bail_unlock_inode: | |
e63aecb6 | 1040 | ocfs2_inode_unlock(inode, 1); |
ccd979bd | 1041 | brelse(di_bh); |
6ca497a8 | 1042 | |
1043 | bail_unlock_nfs_sync: | |
1044 | ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0); | |
1045 | ||
ccd979bd MF |
1046 | bail_unblock: |
1047 | status = sigprocmask(SIG_SETMASK, &oldset, NULL); | |
1048 | if (status < 0) | |
1049 | mlog_errno(status); | |
1050 | bail: | |
1051 | clear_inode(inode); | |
1052 | mlog_exit_void(); | |
1053 | } | |
1054 | ||
1055 | void ocfs2_clear_inode(struct inode *inode) | |
1056 | { | |
1057 | int status; | |
1058 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1059 | ||
1060 | mlog_entry_void(); | |
1061 | ||
1062 | if (!inode) | |
1063 | goto bail; | |
1064 | ||
b0697053 MF |
1065 | mlog(0, "Clearing inode: %llu, nlink = %u\n", |
1066 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
1067 | |
1068 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, | |
1069 | "Inode=%lu\n", inode->i_ino); | |
1070 | ||
34d024f8 MF |
1071 | /* To preven remote deletes we hold open lock before, now it |
1072 | * is time to unlock PR and EX open locks. */ | |
50008630 TY |
1073 | ocfs2_open_unlock(inode); |
1074 | ||
ccd979bd | 1075 | /* Do these before all the other work so that we don't bounce |
34d024f8 | 1076 | * the downconvert thread while waiting to destroy the locks. */ |
ccd979bd | 1077 | ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres); |
e63aecb6 | 1078 | ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres); |
50008630 | 1079 | ocfs2_mark_lockres_freeing(&oi->ip_open_lockres); |
ccd979bd MF |
1080 | |
1081 | /* We very well may get a clear_inode before all an inodes | |
1082 | * metadata has hit disk. Of course, we can't drop any cluster | |
1083 | * locks until the journal has finished with it. The only | |
1084 | * exception here are successfully wiped inodes - their | |
1085 | * metadata can now be considered to be part of the system | |
1086 | * inodes from which it came. */ | |
1087 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) | |
1088 | ocfs2_checkpoint_inode(inode); | |
1089 | ||
1090 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), | |
b0697053 MF |
1091 | "Clear inode of %llu, inode has io markers\n", |
1092 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1093 | |
83418978 MF |
1094 | ocfs2_extent_map_trunc(inode, 0); |
1095 | ||
ccd979bd MF |
1096 | status = ocfs2_drop_inode_locks(inode); |
1097 | if (status < 0) | |
1098 | mlog_errno(status); | |
1099 | ||
1100 | ocfs2_lock_res_free(&oi->ip_rw_lockres); | |
e63aecb6 | 1101 | ocfs2_lock_res_free(&oi->ip_inode_lockres); |
50008630 | 1102 | ocfs2_lock_res_free(&oi->ip_open_lockres); |
ccd979bd MF |
1103 | |
1104 | ocfs2_metadata_cache_purge(inode); | |
1105 | ||
1106 | mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached, | |
b0697053 MF |
1107 | "Clear inode of %llu, inode has %u cache items\n", |
1108 | (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached); | |
ccd979bd MF |
1109 | |
1110 | mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE), | |
b0697053 MF |
1111 | "Clear inode of %llu, inode has a bad flag\n", |
1112 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1113 | |
1114 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), | |
b0697053 MF |
1115 | "Clear inode of %llu, inode is locked\n", |
1116 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1117 | |
251b6ecc | 1118 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
b0697053 MF |
1119 | "Clear inode of %llu, io_mutex is locked\n", |
1120 | (unsigned long long)oi->ip_blkno); | |
251b6ecc | 1121 | mutex_unlock(&oi->ip_io_mutex); |
ccd979bd MF |
1122 | |
1123 | /* | |
1124 | * down_trylock() returns 0, down_write_trylock() returns 1 | |
1125 | * kernel 1, world 0 | |
1126 | */ | |
1127 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), | |
b0697053 MF |
1128 | "Clear inode of %llu, alloc_sem is locked\n", |
1129 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1130 | up_write(&oi->ip_alloc_sem); |
1131 | ||
1132 | mlog_bug_on_msg(oi->ip_open_count, | |
b0697053 MF |
1133 | "Clear inode of %llu has open count %d\n", |
1134 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); | |
ccd979bd MF |
1135 | |
1136 | /* Clear all other flags. */ | |
1137 | oi->ip_flags = OCFS2_INODE_CACHE_INLINE; | |
1138 | oi->ip_created_trans = 0; | |
1139 | oi->ip_last_trans = 0; | |
1140 | oi->ip_dir_start_lookup = 0; | |
1141 | oi->ip_blkno = 0ULL; | |
ae0dff68 SM |
1142 | |
1143 | /* | |
1144 | * ip_jinode is used to track txns against this inode. We ensure that | |
1145 | * the journal is flushed before journal shutdown. Thus it is safe to | |
1146 | * have inodes get cleaned up after journal shutdown. | |
1147 | */ | |
2b4e30fb JB |
1148 | jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal, |
1149 | &oi->ip_jinode); | |
ccd979bd MF |
1150 | |
1151 | bail: | |
1152 | mlog_exit_void(); | |
1153 | } | |
1154 | ||
1155 | /* Called under inode_lock, with no more references on the | |
1156 | * struct inode, so it's safe here to check the flags field | |
1157 | * and to manipulate i_nlink without any other locks. */ | |
1158 | void ocfs2_drop_inode(struct inode *inode) | |
1159 | { | |
1160 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1161 | ||
1162 | mlog_entry_void(); | |
1163 | ||
b0697053 MF |
1164 | mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n", |
1165 | (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags); | |
ccd979bd | 1166 | |
379dfe9d MF |
1167 | if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) |
1168 | generic_delete_inode(inode); | |
1169 | else | |
1170 | generic_drop_inode(inode); | |
ccd979bd MF |
1171 | |
1172 | mlog_exit_void(); | |
1173 | } | |
1174 | ||
ccd979bd MF |
1175 | /* |
1176 | * This is called from our getattr. | |
1177 | */ | |
1178 | int ocfs2_inode_revalidate(struct dentry *dentry) | |
1179 | { | |
1180 | struct inode *inode = dentry->d_inode; | |
1181 | int status = 0; | |
1182 | ||
b0697053 MF |
1183 | mlog_entry("(inode = 0x%p, ino = %llu)\n", inode, |
1184 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL); | |
ccd979bd MF |
1185 | |
1186 | if (!inode) { | |
1187 | mlog(0, "eep, no inode!\n"); | |
1188 | status = -ENOENT; | |
1189 | goto bail; | |
1190 | } | |
1191 | ||
1192 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1193 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
1194 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1195 | mlog(0, "inode deleted!\n"); | |
1196 | status = -ENOENT; | |
1197 | goto bail; | |
1198 | } | |
1199 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1200 | ||
e63aecb6 | 1201 | /* Let ocfs2_inode_lock do the work of updating our struct |
ccd979bd | 1202 | * inode for us. */ |
e63aecb6 | 1203 | status = ocfs2_inode_lock(inode, NULL, 0); |
ccd979bd MF |
1204 | if (status < 0) { |
1205 | if (status != -ENOENT) | |
1206 | mlog_errno(status); | |
1207 | goto bail; | |
1208 | } | |
e63aecb6 | 1209 | ocfs2_inode_unlock(inode, 0); |
ccd979bd MF |
1210 | bail: |
1211 | mlog_exit(status); | |
1212 | ||
1213 | return status; | |
1214 | } | |
1215 | ||
1216 | /* | |
1217 | * Updates a disk inode from a | |
1218 | * struct inode. | |
1219 | * Only takes ip_lock. | |
1220 | */ | |
1fabe148 | 1221 | int ocfs2_mark_inode_dirty(handle_t *handle, |
ccd979bd MF |
1222 | struct inode *inode, |
1223 | struct buffer_head *bh) | |
1224 | { | |
1225 | int status; | |
1226 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
1227 | ||
b0697053 MF |
1228 | mlog_entry("(inode %llu)\n", |
1229 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd | 1230 | |
13723d00 JB |
1231 | status = ocfs2_journal_access_di(handle, inode, bh, |
1232 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
1233 | if (status < 0) { |
1234 | mlog_errno(status); | |
1235 | goto leave; | |
1236 | } | |
1237 | ||
1238 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1239 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
6e4b0d56 | 1240 | ocfs2_get_inode_flags(OCFS2_I(inode)); |
ca4d147e | 1241 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
15b1e36b | 1242 | fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features); |
ccd979bd MF |
1243 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1244 | ||
1245 | fe->i_size = cpu_to_le64(i_size_read(inode)); | |
198a1ca3 | 1246 | ocfs2_set_links_count(fe, inode->i_nlink); |
ccd979bd MF |
1247 | fe->i_uid = cpu_to_le32(inode->i_uid); |
1248 | fe->i_gid = cpu_to_le32(inode->i_gid); | |
1249 | fe->i_mode = cpu_to_le16(inode->i_mode); | |
1250 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); | |
1251 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
1252 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
1253 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
1254 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1255 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
1256 | ||
1257 | status = ocfs2_journal_dirty(handle, bh); | |
1258 | if (status < 0) | |
1259 | mlog_errno(status); | |
1260 | ||
1261 | status = 0; | |
1262 | leave: | |
1263 | ||
1264 | mlog_exit(status); | |
1265 | return status; | |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * | |
1270 | * Updates a struct inode from a disk inode. | |
1271 | * does no i/o, only takes ip_lock. | |
1272 | */ | |
1273 | void ocfs2_refresh_inode(struct inode *inode, | |
1274 | struct ocfs2_dinode *fe) | |
1275 | { | |
ccd979bd MF |
1276 | spin_lock(&OCFS2_I(inode)->ip_lock); |
1277 | ||
1278 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e | 1279 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
15b1e36b | 1280 | OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features); |
ca4d147e | 1281 | ocfs2_set_inode_flags(inode); |
ccd979bd | 1282 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
198a1ca3 | 1283 | inode->i_nlink = ocfs2_read_links_count(fe); |
ccd979bd MF |
1284 | inode->i_uid = le32_to_cpu(fe->i_uid); |
1285 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
1286 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
ccd979bd MF |
1287 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) |
1288 | inode->i_blocks = 0; | |
1289 | else | |
8110b073 | 1290 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
1291 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
1292 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
1293 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
1294 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
1295 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
1296 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
1297 | ||
1298 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1299 | } | |
b657c95c JB |
1300 | |
1301 | int ocfs2_validate_inode_block(struct super_block *sb, | |
1302 | struct buffer_head *bh) | |
1303 | { | |
d6b32bbb | 1304 | int rc; |
b657c95c JB |
1305 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; |
1306 | ||
970e4936 JB |
1307 | mlog(0, "Validating dinode %llu\n", |
1308 | (unsigned long long)bh->b_blocknr); | |
1309 | ||
b657c95c JB |
1310 | BUG_ON(!buffer_uptodate(bh)); |
1311 | ||
d6b32bbb JB |
1312 | /* |
1313 | * If the ecc fails, we return the error but otherwise | |
1314 | * leave the filesystem running. We know any error is | |
1315 | * local to this block. | |
1316 | */ | |
1317 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check); | |
13723d00 JB |
1318 | if (rc) { |
1319 | mlog(ML_ERROR, "Checksum failed for dinode %llu\n", | |
1320 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 1321 | goto bail; |
13723d00 | 1322 | } |
d6b32bbb JB |
1323 | |
1324 | /* | |
1325 | * Errors after here are fatal. | |
1326 | */ | |
1327 | ||
1328 | rc = -EINVAL; | |
1329 | ||
b657c95c JB |
1330 | if (!OCFS2_IS_VALID_DINODE(di)) { |
1331 | ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n", | |
1332 | (unsigned long long)bh->b_blocknr, 7, | |
1333 | di->i_signature); | |
1334 | goto bail; | |
1335 | } | |
1336 | ||
1337 | if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) { | |
1338 | ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n", | |
1339 | (unsigned long long)bh->b_blocknr, | |
1340 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
1341 | goto bail; | |
1342 | } | |
1343 | ||
1344 | if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
1345 | ocfs2_error(sb, | |
1346 | "Invalid dinode #%llu: OCFS2_VALID_FL not set\n", | |
1347 | (unsigned long long)bh->b_blocknr); | |
1348 | goto bail; | |
1349 | } | |
1350 | ||
1351 | if (le32_to_cpu(di->i_fs_generation) != | |
1352 | OCFS2_SB(sb)->fs_generation) { | |
1353 | ocfs2_error(sb, | |
1354 | "Invalid dinode #%llu: fs_generation is %u\n", | |
1355 | (unsigned long long)bh->b_blocknr, | |
1356 | le32_to_cpu(di->i_fs_generation)); | |
1357 | goto bail; | |
1358 | } | |
1359 | ||
1360 | rc = 0; | |
1361 | ||
1362 | bail: | |
1363 | return rc; | |
1364 | } | |
1365 | ||
1366 | int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh, | |
1367 | int flags) | |
1368 | { | |
1369 | int rc; | |
1370 | struct buffer_head *tmp = *bh; | |
1371 | ||
1372 | rc = ocfs2_read_blocks(inode, OCFS2_I(inode)->ip_blkno, 1, &tmp, | |
970e4936 | 1373 | flags, ocfs2_validate_inode_block); |
b657c95c JB |
1374 | |
1375 | /* If ocfs2_read_blocks() got us a new bh, pass it up. */ | |
970e4936 | 1376 | if (!rc && !*bh) |
b657c95c JB |
1377 | *bh = tmp; |
1378 | ||
b657c95c JB |
1379 | return rc; |
1380 | } | |
1381 | ||
1382 | int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh) | |
1383 | { | |
1384 | return ocfs2_read_inode_block_full(inode, bh, 0); | |
1385 | } |