]>
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 | ||
57 | #define OCFS2_FI_FLAG_NOWAIT 0x1 | |
58 | #define OCFS2_FI_FLAG_DELETE 0x2 | |
59 | struct ocfs2_find_inode_args | |
60 | { | |
61 | u64 fi_blkno; | |
62 | unsigned long fi_ino; | |
63 | unsigned int fi_flags; | |
64 | }; | |
65 | ||
66 | static int ocfs2_read_locked_inode(struct inode *inode, | |
67 | struct ocfs2_find_inode_args *args); | |
68 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); | |
69 | static int ocfs2_find_actor(struct inode *inode, void *opaque); | |
70 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
71 | struct inode *inode, | |
72 | struct buffer_head *fe_bh); | |
73 | ||
ca4d147e HP |
74 | void ocfs2_set_inode_flags(struct inode *inode) |
75 | { | |
76 | unsigned int flags = OCFS2_I(inode)->ip_attr; | |
77 | ||
78 | inode->i_flags &= ~(S_IMMUTABLE | | |
79 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); | |
80 | ||
81 | if (flags & OCFS2_IMMUTABLE_FL) | |
82 | inode->i_flags |= S_IMMUTABLE; | |
83 | ||
84 | if (flags & OCFS2_SYNC_FL) | |
85 | inode->i_flags |= S_SYNC; | |
86 | if (flags & OCFS2_APPEND_FL) | |
87 | inode->i_flags |= S_APPEND; | |
88 | if (flags & OCFS2_NOATIME_FL) | |
89 | inode->i_flags |= S_NOATIME; | |
90 | if (flags & OCFS2_DIRSYNC_FL) | |
91 | inode->i_flags |= S_DIRSYNC; | |
92 | } | |
93 | ||
ccd979bd MF |
94 | struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb, |
95 | u64 blkno, | |
96 | int delete_vote) | |
97 | { | |
98 | struct ocfs2_find_inode_args args; | |
99 | ||
100 | /* ocfs2_ilookup_for_vote should *only* be called from the | |
101 | * vote thread */ | |
102 | BUG_ON(current != osb->vote_task); | |
103 | ||
104 | args.fi_blkno = blkno; | |
105 | args.fi_flags = OCFS2_FI_FLAG_NOWAIT; | |
106 | if (delete_vote) | |
107 | args.fi_flags |= OCFS2_FI_FLAG_DELETE; | |
108 | args.fi_ino = ino_from_blkno(osb->sb, blkno); | |
109 | return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args); | |
110 | } | |
111 | ||
112 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno) | |
113 | { | |
114 | struct inode *inode = NULL; | |
115 | struct super_block *sb = osb->sb; | |
116 | struct ocfs2_find_inode_args args; | |
117 | ||
b0697053 | 118 | mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno); |
ccd979bd MF |
119 | |
120 | /* Ok. By now we've either got the offsets passed to us by the | |
121 | * caller, or we just pulled them off the bh. Lets do some | |
122 | * sanity checks to make sure they're OK. */ | |
123 | if (blkno == 0) { | |
124 | inode = ERR_PTR(-EINVAL); | |
125 | mlog_errno(PTR_ERR(inode)); | |
126 | goto bail; | |
127 | } | |
128 | ||
129 | args.fi_blkno = blkno; | |
130 | args.fi_flags = 0; | |
131 | args.fi_ino = ino_from_blkno(sb, blkno); | |
132 | ||
133 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, | |
134 | ocfs2_init_locked_inode, &args); | |
135 | /* inode was *not* in the inode cache. 2.6.x requires | |
136 | * us to do our own read_inode call and unlock it | |
137 | * afterwards. */ | |
138 | if (inode && inode->i_state & I_NEW) { | |
139 | mlog(0, "Inode was not in inode cache, reading it.\n"); | |
140 | ocfs2_read_locked_inode(inode, &args); | |
141 | unlock_new_inode(inode); | |
142 | } | |
143 | if (inode == NULL) { | |
144 | inode = ERR_PTR(-ENOMEM); | |
145 | mlog_errno(PTR_ERR(inode)); | |
146 | goto bail; | |
147 | } | |
148 | if (is_bad_inode(inode)) { | |
149 | iput(inode); | |
150 | inode = ERR_PTR(-ESTALE); | |
151 | mlog_errno(PTR_ERR(inode)); | |
152 | goto bail; | |
153 | } | |
154 | ||
155 | bail: | |
156 | if (!IS_ERR(inode)) { | |
b0697053 MF |
157 | mlog(0, "returning inode with number %llu\n", |
158 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
159 | mlog_exit_ptr(inode); |
160 | } else | |
161 | mlog_errno(PTR_ERR(inode)); | |
162 | ||
163 | return inode; | |
164 | } | |
165 | ||
166 | ||
167 | /* | |
168 | * here's how inodes get read from disk: | |
169 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR | |
170 | * found? : return the in-memory inode | |
171 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE | |
172 | */ | |
173 | ||
174 | static int ocfs2_find_actor(struct inode *inode, void *opaque) | |
175 | { | |
176 | struct ocfs2_find_inode_args *args = NULL; | |
177 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
178 | int ret = 0; | |
179 | ||
180 | mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque); | |
181 | ||
182 | args = opaque; | |
183 | ||
184 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); | |
185 | ||
186 | if (oi->ip_blkno != args->fi_blkno) | |
187 | goto bail; | |
188 | ||
189 | /* OCFS2_FI_FLAG_NOWAIT is *only* set from | |
190 | * ocfs2_ilookup_for_vote which won't create an inode for one | |
191 | * that isn't found. The vote thread which doesn't want to get | |
192 | * an inode which is in the process of going away - otherwise | |
193 | * the call to __wait_on_freeing_inode in find_inode_fast will | |
194 | * cause it to deadlock on an inode which may be waiting on a | |
195 | * vote (or lock release) in delete_inode */ | |
196 | if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) && | |
197 | (inode->i_state & (I_FREEING|I_CLEAR))) { | |
198 | /* As stated above, we're not going to return an | |
199 | * inode. In the case of a delete vote, the voting | |
200 | * code is going to signal the other node to go | |
201 | * ahead. Mark that state here, so this freeing inode | |
202 | * has the state when it gets to delete_inode. */ | |
203 | if (args->fi_flags & OCFS2_FI_FLAG_DELETE) { | |
204 | spin_lock(&oi->ip_lock); | |
205 | ocfs2_mark_inode_remotely_deleted(inode); | |
206 | spin_unlock(&oi->ip_lock); | |
207 | } | |
208 | goto bail; | |
209 | } | |
210 | ||
211 | ret = 1; | |
212 | bail: | |
213 | mlog_exit(ret); | |
214 | return ret; | |
215 | } | |
216 | ||
217 | /* | |
218 | * initialize the new inode, but don't do anything that would cause | |
219 | * us to sleep. | |
220 | * return 0 on success, 1 on failure | |
221 | */ | |
222 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) | |
223 | { | |
224 | struct ocfs2_find_inode_args *args = opaque; | |
225 | ||
226 | mlog_entry("inode = %p, opaque = %p\n", inode, opaque); | |
227 | ||
228 | inode->i_ino = args->fi_ino; | |
229 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; | |
230 | ||
231 | mlog_exit(0); | |
232 | return 0; | |
233 | } | |
234 | ||
235 | int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, | |
236 | int create_ino) | |
237 | { | |
238 | struct super_block *sb; | |
239 | struct ocfs2_super *osb; | |
240 | int status = -EINVAL; | |
241 | ||
b0697053 MF |
242 | mlog_entry("(0x%p, size:%llu)\n", inode, |
243 | (unsigned long long)fe->i_size); | |
ccd979bd MF |
244 | |
245 | sb = inode->i_sb; | |
246 | osb = OCFS2_SB(sb); | |
247 | ||
248 | /* this means that read_inode cannot create a superblock inode | |
249 | * today. change if needed. */ | |
250 | if (!OCFS2_IS_VALID_DINODE(fe) || | |
251 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
b0697053 | 252 | mlog(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%llu, " |
ccd979bd | 253 | "signature = %.*s, flags = 0x%x\n", |
b0697053 MF |
254 | inode->i_ino, |
255 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, | |
ccd979bd MF |
256 | fe->i_signature, le32_to_cpu(fe->i_flags)); |
257 | goto bail; | |
258 | } | |
259 | ||
260 | if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) { | |
261 | mlog(ML_ERROR, "file entry generation does not match " | |
262 | "superblock! osb->fs_generation=%x, " | |
263 | "fe->i_fs_generation=%x\n", | |
264 | osb->fs_generation, le32_to_cpu(fe->i_fs_generation)); | |
265 | goto bail; | |
266 | } | |
267 | ||
268 | inode->i_version = 1; | |
269 | inode->i_generation = le32_to_cpu(fe->i_generation); | |
270 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
271 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
272 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
273 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
274 | inode->i_blksize = (u32)osb->s_clustersize; | |
275 | ||
276 | /* Fast symlinks will have i_size but no allocated clusters. */ | |
277 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) | |
278 | inode->i_blocks = 0; | |
279 | else | |
280 | inode->i_blocks = | |
281 | ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size)); | |
282 | inode->i_mapping->a_ops = &ocfs2_aops; | |
ccd979bd MF |
283 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
284 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
285 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
286 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
287 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
288 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
289 | ||
290 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) | |
291 | mlog(ML_ERROR, | |
b0697053 MF |
292 | "ip_blkno %llu != i_blkno %llu!\n", |
293 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
294 | (unsigned long long)fe->i_blkno); | |
ccd979bd MF |
295 | |
296 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
297 | OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT; | |
ca4d147e | 298 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
ccd979bd MF |
299 | |
300 | if (create_ino) | |
301 | inode->i_ino = ino_from_blkno(inode->i_sb, | |
302 | le64_to_cpu(fe->i_blkno)); | |
303 | ||
b0697053 MF |
304 | mlog(0, "blkno = %llu, ino = %lu, create_ino = %s\n", |
305 | (unsigned long long)fe->i_blkno, inode->i_ino, create_ino ? "true" : "false"); | |
ccd979bd MF |
306 | |
307 | inode->i_nlink = le16_to_cpu(fe->i_links_count); | |
308 | ||
309 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { | |
310 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
311 | mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino); | |
312 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { | |
313 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
314 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { | |
315 | mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino); | |
316 | /* we can't actually hit this as read_inode can't | |
317 | * handle superblocks today ;-) */ | |
318 | BUG(); | |
319 | } | |
320 | ||
321 | switch (inode->i_mode & S_IFMT) { | |
322 | case S_IFREG: | |
323 | inode->i_fop = &ocfs2_fops; | |
324 | inode->i_op = &ocfs2_file_iops; | |
325 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
326 | break; | |
327 | case S_IFDIR: | |
328 | inode->i_op = &ocfs2_dir_iops; | |
329 | inode->i_fop = &ocfs2_dops; | |
330 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
331 | break; | |
332 | case S_IFLNK: | |
333 | if (ocfs2_inode_is_fast_symlink(inode)) | |
334 | inode->i_op = &ocfs2_fast_symlink_inode_operations; | |
335 | else | |
336 | inode->i_op = &ocfs2_symlink_inode_operations; | |
337 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
338 | break; | |
339 | default: | |
340 | inode->i_op = &ocfs2_special_file_iops; | |
341 | init_special_inode(inode, inode->i_mode, | |
342 | inode->i_rdev); | |
343 | break; | |
344 | } | |
345 | ||
346 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, | |
347 | OCFS2_LOCK_TYPE_RW, inode); | |
348 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
349 | OCFS2_LOCK_TYPE_META, inode); | |
350 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres, | |
351 | OCFS2_LOCK_TYPE_DATA, inode); | |
352 | ||
ca4d147e HP |
353 | ocfs2_set_inode_flags(inode); |
354 | inode->i_flags |= S_NOATIME; | |
355 | ||
ccd979bd MF |
356 | status = 0; |
357 | bail: | |
358 | mlog_exit(status); | |
359 | return status; | |
360 | } | |
361 | ||
362 | static int ocfs2_read_locked_inode(struct inode *inode, | |
363 | struct ocfs2_find_inode_args *args) | |
364 | { | |
365 | struct super_block *sb; | |
366 | struct ocfs2_super *osb; | |
367 | struct ocfs2_dinode *fe; | |
368 | struct buffer_head *bh = NULL; | |
369 | int status; | |
370 | int sysfile = 0; | |
371 | ||
372 | mlog_entry("(0x%p, 0x%p)\n", inode, args); | |
373 | ||
374 | status = -EINVAL; | |
375 | if (inode == NULL || inode->i_sb == NULL) { | |
376 | mlog(ML_ERROR, "bad inode\n"); | |
377 | goto bail; | |
378 | } | |
379 | sb = inode->i_sb; | |
380 | osb = OCFS2_SB(sb); | |
381 | ||
382 | if (!args) { | |
383 | mlog(ML_ERROR, "bad inode args\n"); | |
384 | make_bad_inode(inode); | |
385 | goto bail; | |
386 | } | |
387 | ||
388 | /* Read the FE off disk. This is safe because the kernel only | |
389 | * does one read_inode2 for a new inode, and if it doesn't | |
390 | * exist yet then nobody can be working on it! */ | |
391 | status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, NULL); | |
392 | if (status < 0) { | |
393 | mlog_errno(status); | |
394 | make_bad_inode(inode); | |
395 | goto bail; | |
396 | } | |
397 | ||
398 | fe = (struct ocfs2_dinode *) bh->b_data; | |
399 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
b0697053 MF |
400 | mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n", |
401 | (unsigned long long)fe->i_blkno, 7, fe->i_signature); | |
ccd979bd MF |
402 | make_bad_inode(inode); |
403 | goto bail; | |
404 | } | |
405 | ||
406 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) | |
407 | sysfile = 1; | |
408 | ||
409 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || | |
410 | S_ISBLK(le16_to_cpu(fe->i_mode))) | |
411 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
412 | ||
413 | status = -EINVAL; | |
414 | if (ocfs2_populate_inode(inode, fe, 0) < 0) { | |
b0697053 MF |
415 | mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n", |
416 | (unsigned long long)fe->i_blkno, inode->i_ino); | |
ccd979bd MF |
417 | make_bad_inode(inode); |
418 | goto bail; | |
419 | } | |
420 | ||
421 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); | |
422 | ||
423 | if (sysfile) | |
424 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; | |
425 | ||
426 | status = 0; | |
427 | ||
428 | bail: | |
429 | if (args && bh) | |
430 | brelse(bh); | |
431 | ||
432 | mlog_exit(status); | |
433 | return status; | |
434 | } | |
435 | ||
436 | void ocfs2_sync_blockdev(struct super_block *sb) | |
437 | { | |
438 | sync_blockdev(sb->s_bdev); | |
439 | } | |
440 | ||
441 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
442 | struct inode *inode, | |
443 | struct buffer_head *fe_bh) | |
444 | { | |
445 | int status = 0; | |
446 | struct ocfs2_journal_handle *handle = NULL; | |
447 | struct ocfs2_truncate_context *tc = NULL; | |
448 | struct ocfs2_dinode *fe; | |
449 | ||
450 | mlog_entry_void(); | |
451 | ||
452 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
453 | ||
454 | /* zero allocation, zero truncate :) */ | |
455 | if (!fe->i_clusters) | |
456 | goto bail; | |
457 | ||
458 | handle = ocfs2_start_trans(osb, handle, OCFS2_INODE_UPDATE_CREDITS); | |
459 | if (IS_ERR(handle)) { | |
460 | status = PTR_ERR(handle); | |
461 | handle = NULL; | |
462 | mlog_errno(status); | |
463 | goto bail; | |
464 | } | |
465 | ||
466 | status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL); | |
467 | if (status < 0) { | |
468 | mlog_errno(status); | |
469 | goto bail; | |
470 | } | |
471 | ||
472 | ocfs2_commit_trans(handle); | |
473 | handle = NULL; | |
474 | ||
475 | status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc); | |
476 | if (status < 0) { | |
477 | mlog_errno(status); | |
478 | goto bail; | |
479 | } | |
480 | ||
481 | status = ocfs2_commit_truncate(osb, inode, fe_bh, tc); | |
482 | if (status < 0) { | |
483 | mlog_errno(status); | |
484 | goto bail; | |
485 | } | |
486 | bail: | |
487 | if (handle) | |
488 | ocfs2_commit_trans(handle); | |
489 | ||
490 | mlog_exit(status); | |
491 | return status; | |
492 | } | |
493 | ||
494 | static int ocfs2_remove_inode(struct inode *inode, | |
495 | struct buffer_head *di_bh, | |
496 | struct inode *orphan_dir_inode, | |
497 | struct buffer_head *orphan_dir_bh) | |
498 | { | |
499 | int status; | |
500 | struct inode *inode_alloc_inode = NULL; | |
501 | struct buffer_head *inode_alloc_bh = NULL; | |
502 | struct ocfs2_journal_handle *handle; | |
503 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
504 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
505 | ||
506 | inode_alloc_inode = | |
507 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
508 | le16_to_cpu(di->i_suballoc_slot)); | |
509 | if (!inode_alloc_inode) { | |
510 | status = -EEXIST; | |
511 | mlog_errno(status); | |
512 | goto bail; | |
513 | } | |
514 | ||
1b1dcc1b | 515 | mutex_lock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
516 | status = ocfs2_meta_lock(inode_alloc_inode, NULL, &inode_alloc_bh, 1); |
517 | if (status < 0) { | |
1b1dcc1b | 518 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
519 | |
520 | mlog_errno(status); | |
521 | goto bail; | |
522 | } | |
523 | ||
524 | handle = ocfs2_start_trans(osb, NULL, OCFS2_DELETE_INODE_CREDITS); | |
525 | if (IS_ERR(handle)) { | |
526 | status = PTR_ERR(handle); | |
527 | mlog_errno(status); | |
528 | goto bail_unlock; | |
529 | } | |
530 | ||
531 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, | |
532 | orphan_dir_bh); | |
533 | if (status < 0) { | |
534 | mlog_errno(status); | |
535 | goto bail_commit; | |
536 | } | |
537 | ||
538 | /* set the inodes dtime */ | |
539 | status = ocfs2_journal_access(handle, inode, di_bh, | |
540 | OCFS2_JOURNAL_ACCESS_WRITE); | |
541 | if (status < 0) { | |
542 | mlog_errno(status); | |
543 | goto bail_commit; | |
544 | } | |
545 | ||
546 | di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec); | |
547 | le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); | |
548 | ||
549 | status = ocfs2_journal_dirty(handle, di_bh); | |
550 | if (status < 0) { | |
551 | mlog_errno(status); | |
552 | goto bail_commit; | |
553 | } | |
554 | ||
555 | ocfs2_remove_from_cache(inode, di_bh); | |
556 | ||
557 | status = ocfs2_free_dinode(handle, inode_alloc_inode, | |
558 | inode_alloc_bh, di); | |
559 | if (status < 0) | |
560 | mlog_errno(status); | |
561 | ||
562 | bail_commit: | |
563 | ocfs2_commit_trans(handle); | |
564 | bail_unlock: | |
565 | ocfs2_meta_unlock(inode_alloc_inode, 1); | |
1b1dcc1b | 566 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
567 | brelse(inode_alloc_bh); |
568 | bail: | |
569 | iput(inode_alloc_inode); | |
570 | ||
571 | return status; | |
572 | } | |
573 | ||
b4df6ed8 MF |
574 | /* |
575 | * Serialize with orphan dir recovery. If the process doing | |
576 | * recovery on this orphan dir does an iget() with the dir | |
577 | * i_mutex held, we'll deadlock here. Instead we detect this | |
578 | * and exit early - recovery will wipe this inode for us. | |
579 | */ | |
580 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, | |
581 | int slot) | |
582 | { | |
583 | int ret = 0; | |
584 | ||
585 | spin_lock(&osb->osb_lock); | |
586 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { | |
587 | mlog(0, "Recovery is happening on orphan dir %d, will skip " | |
588 | "this inode\n", slot); | |
589 | ret = -EDEADLK; | |
590 | goto out; | |
591 | } | |
592 | /* This signals to the orphan recovery process that it should | |
593 | * wait for us to handle the wipe. */ | |
594 | osb->osb_orphan_wipes[slot]++; | |
595 | out: | |
596 | spin_unlock(&osb->osb_lock); | |
597 | return ret; | |
598 | } | |
599 | ||
600 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, | |
601 | int slot) | |
602 | { | |
603 | spin_lock(&osb->osb_lock); | |
604 | osb->osb_orphan_wipes[slot]--; | |
605 | spin_unlock(&osb->osb_lock); | |
606 | ||
607 | wake_up(&osb->osb_wipe_event); | |
608 | } | |
609 | ||
ccd979bd MF |
610 | static int ocfs2_wipe_inode(struct inode *inode, |
611 | struct buffer_head *di_bh) | |
612 | { | |
613 | int status, orphaned_slot; | |
614 | struct inode *orphan_dir_inode = NULL; | |
615 | struct buffer_head *orphan_dir_bh = NULL; | |
616 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
617 | ||
618 | /* We've already voted on this so it should be readonly - no | |
619 | * spinlock needed. */ | |
620 | orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot; | |
b4df6ed8 MF |
621 | |
622 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); | |
623 | if (status) | |
624 | return status; | |
625 | ||
ccd979bd MF |
626 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
627 | ORPHAN_DIR_SYSTEM_INODE, | |
628 | orphaned_slot); | |
629 | if (!orphan_dir_inode) { | |
630 | status = -EEXIST; | |
631 | mlog_errno(status); | |
632 | goto bail; | |
633 | } | |
634 | ||
635 | /* Lock the orphan dir. The lock will be held for the entire | |
636 | * delete_inode operation. We do this now to avoid races with | |
637 | * recovery completion on other nodes. */ | |
1b1dcc1b | 638 | mutex_lock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
639 | status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1); |
640 | if (status < 0) { | |
1b1dcc1b | 641 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
642 | |
643 | mlog_errno(status); | |
644 | goto bail; | |
645 | } | |
646 | ||
647 | /* we do this while holding the orphan dir lock because we | |
648 | * don't want recovery being run from another node to vote for | |
649 | * an inode delete on us -- this will result in two nodes | |
650 | * truncating the same file! */ | |
651 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); | |
652 | if (status < 0) { | |
653 | mlog_errno(status); | |
654 | goto bail_unlock_dir; | |
655 | } | |
656 | ||
657 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, | |
658 | orphan_dir_bh); | |
659 | if (status < 0) | |
660 | mlog_errno(status); | |
661 | ||
662 | bail_unlock_dir: | |
663 | ocfs2_meta_unlock(orphan_dir_inode, 1); | |
1b1dcc1b | 664 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
665 | brelse(orphan_dir_bh); |
666 | bail: | |
667 | iput(orphan_dir_inode); | |
b4df6ed8 | 668 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
ccd979bd MF |
669 | |
670 | return status; | |
671 | } | |
672 | ||
673 | /* There is a series of simple checks that should be done before a | |
674 | * vote is even considered. Encapsulate those in this function. */ | |
675 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) | |
676 | { | |
677 | int ret = 0; | |
678 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
679 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
680 | ||
681 | /* We shouldn't be getting here for the root directory | |
682 | * inode.. */ | |
683 | if (inode == osb->root_inode) { | |
684 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); | |
685 | goto bail; | |
686 | } | |
687 | ||
688 | /* If we're coming from process_vote we can't go into our own | |
689 | * voting [hello, deadlock city!], so unforuntately we just | |
690 | * have to skip deleting this guy. That's OK though because | |
691 | * the node who's doing the actual deleting should handle it | |
692 | * anyway. */ | |
693 | if (current == osb->vote_task) { | |
694 | mlog(0, "Skipping delete of %lu because we're currently " | |
695 | "in process_vote\n", inode->i_ino); | |
696 | goto bail; | |
697 | } | |
698 | ||
699 | spin_lock(&oi->ip_lock); | |
700 | /* OCFS2 *never* deletes system files. This should technically | |
701 | * never get here as system file inodes should always have a | |
702 | * positive link count. */ | |
703 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { | |
b0697053 MF |
704 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
705 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
706 | goto bail_unlock; |
707 | } | |
708 | ||
709 | /* If we have voted "yes" on the wipe of this inode for | |
710 | * another node, it will be marked here so we can safely skip | |
711 | * it. Recovery will cleanup any inodes we might inadvertantly | |
712 | * skip here. */ | |
713 | if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) { | |
714 | mlog(0, "Skipping delete of %lu because another node " | |
715 | "has done this for us.\n", inode->i_ino); | |
716 | goto bail_unlock; | |
717 | } | |
718 | ||
719 | ret = 1; | |
720 | bail_unlock: | |
721 | spin_unlock(&oi->ip_lock); | |
722 | bail: | |
723 | return ret; | |
724 | } | |
725 | ||
726 | /* Query the cluster to determine whether we should wipe an inode from | |
727 | * disk or not. | |
728 | * | |
729 | * Requires the inode to have the cluster lock. */ | |
730 | static int ocfs2_query_inode_wipe(struct inode *inode, | |
731 | struct buffer_head *di_bh, | |
732 | int *wipe) | |
733 | { | |
734 | int status = 0; | |
735 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
736 | struct ocfs2_dinode *di; | |
737 | ||
738 | *wipe = 0; | |
739 | ||
740 | /* While we were waiting for the cluster lock in | |
741 | * ocfs2_delete_inode, another node might have asked to delete | |
742 | * the inode. Recheck our flags to catch this. */ | |
743 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
b0697053 MF |
744 | mlog(0, "Skipping delete of %llu because flags changed\n", |
745 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
746 | goto bail; |
747 | } | |
748 | ||
749 | /* Now that we have an up to date inode, we can double check | |
750 | * the link count. */ | |
751 | if (inode->i_nlink) { | |
b0697053 MF |
752 | mlog(0, "Skipping delete of %llu because nlink = %u\n", |
753 | (unsigned long long)oi->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
754 | goto bail; |
755 | } | |
756 | ||
757 | /* Do some basic inode verification... */ | |
758 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
759 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) { | |
760 | /* for lack of a better error? */ | |
761 | status = -EEXIST; | |
762 | mlog(ML_ERROR, | |
b0697053 | 763 | "Inode %llu (on-disk %llu) not orphaned! " |
ccd979bd | 764 | "Disk flags 0x%x, inode flags 0x%x\n", |
b0697053 MF |
765 | (unsigned long long)oi->ip_blkno, |
766 | (unsigned long long)di->i_blkno, di->i_flags, | |
767 | oi->ip_flags); | |
ccd979bd MF |
768 | goto bail; |
769 | } | |
770 | ||
771 | /* has someone already deleted us?! baaad... */ | |
772 | if (di->i_dtime) { | |
773 | status = -EEXIST; | |
774 | mlog_errno(status); | |
775 | goto bail; | |
776 | } | |
777 | ||
778 | status = ocfs2_request_delete_vote(inode); | |
779 | /* -EBUSY means that other nodes are still using the | |
780 | * inode. We're done here though, so avoid doing anything on | |
781 | * disk and let them worry about deleting it. */ | |
782 | if (status == -EBUSY) { | |
783 | status = 0; | |
b0697053 MF |
784 | mlog(0, "Skipping delete of %llu because it is in use on" |
785 | "other nodes\n", (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
786 | goto bail; |
787 | } | |
788 | if (status < 0) { | |
789 | mlog_errno(status); | |
790 | goto bail; | |
791 | } | |
792 | ||
793 | spin_lock(&oi->ip_lock); | |
794 | if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) { | |
795 | /* Nobody knew which slot this inode was orphaned | |
796 | * into. This may happen during node death and | |
797 | * recovery knows how to clean it up so we can safely | |
798 | * ignore this inode for now on. */ | |
b0697053 MF |
799 | mlog(0, "Nobody knew where inode %llu was orphaned!\n", |
800 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
801 | } else { |
802 | *wipe = 1; | |
803 | ||
b0697053 MF |
804 | mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n", |
805 | (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot); | |
ccd979bd MF |
806 | } |
807 | spin_unlock(&oi->ip_lock); | |
808 | ||
809 | bail: | |
810 | return status; | |
811 | } | |
812 | ||
813 | /* Support function for ocfs2_delete_inode. Will help us keep the | |
814 | * inode data in a consistent state for clear_inode. Always truncates | |
815 | * pages, optionally sync's them first. */ | |
816 | static void ocfs2_cleanup_delete_inode(struct inode *inode, | |
817 | int sync_data) | |
818 | { | |
b0697053 MF |
819 | mlog(0, "Cleanup inode %llu, sync = %d\n", |
820 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); | |
ccd979bd MF |
821 | if (sync_data) |
822 | write_inode_now(inode, 1); | |
823 | truncate_inode_pages(&inode->i_data, 0); | |
824 | } | |
825 | ||
826 | void ocfs2_delete_inode(struct inode *inode) | |
827 | { | |
828 | int wipe, status; | |
829 | sigset_t blocked, oldset; | |
830 | struct buffer_head *di_bh = NULL; | |
831 | ||
832 | mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino); | |
833 | ||
834 | if (is_bad_inode(inode)) { | |
835 | mlog(0, "Skipping delete of bad inode\n"); | |
836 | goto bail; | |
837 | } | |
838 | ||
839 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
840 | /* It's probably not necessary to truncate_inode_pages | |
841 | * here but we do it for safety anyway (it will most | |
842 | * likely be a no-op anyway) */ | |
843 | ocfs2_cleanup_delete_inode(inode, 0); | |
844 | goto bail; | |
845 | } | |
846 | ||
847 | /* We want to block signals in delete_inode as the lock and | |
848 | * messaging paths may return us -ERESTARTSYS. Which would | |
849 | * cause us to exit early, resulting in inodes being orphaned | |
850 | * forever. */ | |
851 | sigfillset(&blocked); | |
852 | status = sigprocmask(SIG_BLOCK, &blocked, &oldset); | |
853 | if (status < 0) { | |
854 | mlog_errno(status); | |
855 | ocfs2_cleanup_delete_inode(inode, 1); | |
856 | goto bail; | |
857 | } | |
858 | ||
859 | /* Lock down the inode. This gives us an up to date view of | |
860 | * it's metadata (for verification), and allows us to | |
861 | * serialize delete_inode votes. | |
862 | * | |
863 | * Even though we might be doing a truncate, we don't take the | |
864 | * allocation lock here as it won't be needed - nobody will | |
865 | * have the file open. | |
866 | */ | |
867 | status = ocfs2_meta_lock(inode, NULL, &di_bh, 1); | |
868 | if (status < 0) { | |
869 | if (status != -ENOENT) | |
870 | mlog_errno(status); | |
871 | ocfs2_cleanup_delete_inode(inode, 0); | |
872 | goto bail_unblock; | |
873 | } | |
874 | ||
875 | /* Query the cluster. This will be the final decision made | |
876 | * before we go ahead and wipe the inode. */ | |
877 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); | |
878 | if (!wipe || status < 0) { | |
879 | /* Error and inode busy vote both mean we won't be | |
880 | * removing the inode, so they take almost the same | |
881 | * path. */ | |
882 | if (status < 0) | |
883 | mlog_errno(status); | |
884 | ||
885 | /* Someone in the cluster has voted to not wipe this | |
886 | * inode, or it was never completely orphaned. Write | |
887 | * out the pages and exit now. */ | |
888 | ocfs2_cleanup_delete_inode(inode, 1); | |
889 | goto bail_unlock_inode; | |
890 | } | |
891 | ||
892 | ocfs2_cleanup_delete_inode(inode, 0); | |
893 | ||
894 | status = ocfs2_wipe_inode(inode, di_bh); | |
895 | if (status < 0) { | |
b4df6ed8 MF |
896 | if (status != -EDEADLK) |
897 | mlog_errno(status); | |
ccd979bd MF |
898 | goto bail_unlock_inode; |
899 | } | |
900 | ||
901 | /* Mark the inode as successfully deleted. This is important | |
902 | * for ocfs2_clear_inode as it will check this flag and skip | |
903 | * any checkpointing work */ | |
904 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; | |
905 | ||
906 | bail_unlock_inode: | |
907 | ocfs2_meta_unlock(inode, 1); | |
908 | brelse(di_bh); | |
909 | bail_unblock: | |
910 | status = sigprocmask(SIG_SETMASK, &oldset, NULL); | |
911 | if (status < 0) | |
912 | mlog_errno(status); | |
913 | bail: | |
914 | clear_inode(inode); | |
915 | mlog_exit_void(); | |
916 | } | |
917 | ||
918 | void ocfs2_clear_inode(struct inode *inode) | |
919 | { | |
920 | int status; | |
921 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
922 | ||
923 | mlog_entry_void(); | |
924 | ||
925 | if (!inode) | |
926 | goto bail; | |
927 | ||
b0697053 MF |
928 | mlog(0, "Clearing inode: %llu, nlink = %u\n", |
929 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
930 | |
931 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, | |
932 | "Inode=%lu\n", inode->i_ino); | |
933 | ||
934 | /* Do these before all the other work so that we don't bounce | |
935 | * the vote thread while waiting to destroy the locks. */ | |
936 | ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres); | |
937 | ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres); | |
938 | ocfs2_mark_lockres_freeing(&oi->ip_data_lockres); | |
939 | ||
940 | /* We very well may get a clear_inode before all an inodes | |
941 | * metadata has hit disk. Of course, we can't drop any cluster | |
942 | * locks until the journal has finished with it. The only | |
943 | * exception here are successfully wiped inodes - their | |
944 | * metadata can now be considered to be part of the system | |
945 | * inodes from which it came. */ | |
946 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) | |
947 | ocfs2_checkpoint_inode(inode); | |
948 | ||
949 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), | |
b0697053 MF |
950 | "Clear inode of %llu, inode has io markers\n", |
951 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
952 | |
953 | ocfs2_extent_map_drop(inode, 0); | |
954 | ocfs2_extent_map_init(inode); | |
955 | ||
956 | status = ocfs2_drop_inode_locks(inode); | |
957 | if (status < 0) | |
958 | mlog_errno(status); | |
959 | ||
960 | ocfs2_lock_res_free(&oi->ip_rw_lockres); | |
961 | ocfs2_lock_res_free(&oi->ip_meta_lockres); | |
962 | ocfs2_lock_res_free(&oi->ip_data_lockres); | |
963 | ||
964 | ocfs2_metadata_cache_purge(inode); | |
965 | ||
966 | mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached, | |
b0697053 MF |
967 | "Clear inode of %llu, inode has %u cache items\n", |
968 | (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached); | |
ccd979bd MF |
969 | |
970 | mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE), | |
b0697053 MF |
971 | "Clear inode of %llu, inode has a bad flag\n", |
972 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
973 | |
974 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), | |
b0697053 MF |
975 | "Clear inode of %llu, inode is locked\n", |
976 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 977 | |
251b6ecc | 978 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
b0697053 MF |
979 | "Clear inode of %llu, io_mutex is locked\n", |
980 | (unsigned long long)oi->ip_blkno); | |
251b6ecc | 981 | mutex_unlock(&oi->ip_io_mutex); |
ccd979bd MF |
982 | |
983 | /* | |
984 | * down_trylock() returns 0, down_write_trylock() returns 1 | |
985 | * kernel 1, world 0 | |
986 | */ | |
987 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), | |
b0697053 MF |
988 | "Clear inode of %llu, alloc_sem is locked\n", |
989 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
990 | up_write(&oi->ip_alloc_sem); |
991 | ||
992 | mlog_bug_on_msg(oi->ip_open_count, | |
b0697053 MF |
993 | "Clear inode of %llu has open count %d\n", |
994 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); | |
ccd979bd | 995 | mlog_bug_on_msg(!list_empty(&oi->ip_handle_list), |
b0697053 MF |
996 | "Clear inode of %llu has non empty handle list\n", |
997 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 998 | mlog_bug_on_msg(oi->ip_handle, |
b0697053 MF |
999 | "Clear inode of %llu has non empty handle pointer\n", |
1000 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1001 | |
1002 | /* Clear all other flags. */ | |
1003 | oi->ip_flags = OCFS2_INODE_CACHE_INLINE; | |
1004 | oi->ip_created_trans = 0; | |
1005 | oi->ip_last_trans = 0; | |
1006 | oi->ip_dir_start_lookup = 0; | |
1007 | oi->ip_blkno = 0ULL; | |
1008 | ||
1009 | bail: | |
1010 | mlog_exit_void(); | |
1011 | } | |
1012 | ||
1013 | /* Called under inode_lock, with no more references on the | |
1014 | * struct inode, so it's safe here to check the flags field | |
1015 | * and to manipulate i_nlink without any other locks. */ | |
1016 | void ocfs2_drop_inode(struct inode *inode) | |
1017 | { | |
1018 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1019 | ||
1020 | mlog_entry_void(); | |
1021 | ||
b0697053 MF |
1022 | mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n", |
1023 | (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags); | |
ccd979bd MF |
1024 | |
1025 | /* Testing ip_orphaned_slot here wouldn't work because we may | |
1026 | * not have gotten a delete_inode vote from any other nodes | |
1027 | * yet. */ | |
1028 | if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) { | |
1029 | mlog(0, "Inode was orphaned on another node, clearing nlink.\n"); | |
1030 | inode->i_nlink = 0; | |
1031 | } | |
1032 | ||
1033 | generic_drop_inode(inode); | |
1034 | ||
1035 | mlog_exit_void(); | |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * TODO: this should probably be merged into ocfs2_get_block | |
1040 | * | |
1041 | * However, you now need to pay attention to the cont_prepare_write() | |
1042 | * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much | |
1043 | * expects never to extend). | |
1044 | */ | |
1045 | struct buffer_head *ocfs2_bread(struct inode *inode, | |
1046 | int block, int *err, int reada) | |
1047 | { | |
1048 | struct buffer_head *bh = NULL; | |
1049 | int tmperr; | |
1050 | u64 p_blkno; | |
1051 | int readflags = OCFS2_BH_CACHED; | |
1052 | ||
ccd979bd MF |
1053 | if (reada) |
1054 | readflags |= OCFS2_BH_READAHEAD; | |
ccd979bd MF |
1055 | |
1056 | if (((u64)block << inode->i_sb->s_blocksize_bits) >= | |
1057 | i_size_read(inode)) { | |
1058 | BUG_ON(!reada); | |
1059 | return NULL; | |
1060 | } | |
1061 | ||
1062 | tmperr = ocfs2_extent_map_get_blocks(inode, block, 1, | |
1063 | &p_blkno, NULL); | |
1064 | if (tmperr < 0) { | |
1065 | mlog_errno(tmperr); | |
1066 | goto fail; | |
1067 | } | |
1068 | ||
1069 | tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh, | |
1070 | readflags, inode); | |
1071 | if (tmperr < 0) | |
1072 | goto fail; | |
1073 | ||
1074 | tmperr = 0; | |
1075 | ||
1076 | *err = 0; | |
1077 | return bh; | |
1078 | ||
1079 | fail: | |
1080 | if (bh) { | |
1081 | brelse(bh); | |
1082 | bh = NULL; | |
1083 | } | |
1084 | *err = -EIO; | |
1085 | return NULL; | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * This is called from our getattr. | |
1090 | */ | |
1091 | int ocfs2_inode_revalidate(struct dentry *dentry) | |
1092 | { | |
1093 | struct inode *inode = dentry->d_inode; | |
1094 | int status = 0; | |
1095 | ||
b0697053 MF |
1096 | mlog_entry("(inode = 0x%p, ino = %llu)\n", inode, |
1097 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL); | |
ccd979bd MF |
1098 | |
1099 | if (!inode) { | |
1100 | mlog(0, "eep, no inode!\n"); | |
1101 | status = -ENOENT; | |
1102 | goto bail; | |
1103 | } | |
1104 | ||
1105 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1106 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
1107 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1108 | mlog(0, "inode deleted!\n"); | |
1109 | status = -ENOENT; | |
1110 | goto bail; | |
1111 | } | |
1112 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1113 | ||
1114 | /* Let ocfs2_meta_lock do the work of updating our struct | |
1115 | * inode for us. */ | |
1116 | status = ocfs2_meta_lock(inode, NULL, NULL, 0); | |
1117 | if (status < 0) { | |
1118 | if (status != -ENOENT) | |
1119 | mlog_errno(status); | |
1120 | goto bail; | |
1121 | } | |
1122 | ocfs2_meta_unlock(inode, 0); | |
1123 | bail: | |
1124 | mlog_exit(status); | |
1125 | ||
1126 | return status; | |
1127 | } | |
1128 | ||
1129 | /* | |
1130 | * Updates a disk inode from a | |
1131 | * struct inode. | |
1132 | * Only takes ip_lock. | |
1133 | */ | |
1134 | int ocfs2_mark_inode_dirty(struct ocfs2_journal_handle *handle, | |
1135 | struct inode *inode, | |
1136 | struct buffer_head *bh) | |
1137 | { | |
1138 | int status; | |
1139 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
1140 | ||
b0697053 MF |
1141 | mlog_entry("(inode %llu)\n", |
1142 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
1143 | |
1144 | status = ocfs2_journal_access(handle, inode, bh, | |
1145 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1146 | if (status < 0) { | |
1147 | mlog_errno(status); | |
1148 | goto leave; | |
1149 | } | |
1150 | ||
1151 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1152 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
ca4d147e | 1153 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
ccd979bd MF |
1154 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1155 | ||
1156 | fe->i_size = cpu_to_le64(i_size_read(inode)); | |
1157 | fe->i_links_count = cpu_to_le16(inode->i_nlink); | |
1158 | fe->i_uid = cpu_to_le32(inode->i_uid); | |
1159 | fe->i_gid = cpu_to_le32(inode->i_gid); | |
1160 | fe->i_mode = cpu_to_le16(inode->i_mode); | |
1161 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); | |
1162 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
1163 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
1164 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
1165 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1166 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
1167 | ||
1168 | status = ocfs2_journal_dirty(handle, bh); | |
1169 | if (status < 0) | |
1170 | mlog_errno(status); | |
1171 | ||
1172 | status = 0; | |
1173 | leave: | |
1174 | ||
1175 | mlog_exit(status); | |
1176 | return status; | |
1177 | } | |
1178 | ||
1179 | /* | |
1180 | * | |
1181 | * Updates a struct inode from a disk inode. | |
1182 | * does no i/o, only takes ip_lock. | |
1183 | */ | |
1184 | void ocfs2_refresh_inode(struct inode *inode, | |
1185 | struct ocfs2_dinode *fe) | |
1186 | { | |
1187 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1188 | ||
1189 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1190 | ||
1191 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e HP |
1192 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
1193 | ocfs2_set_inode_flags(inode); | |
ccd979bd MF |
1194 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
1195 | inode->i_nlink = le16_to_cpu(fe->i_links_count); | |
1196 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
1197 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
1198 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
1199 | inode->i_blksize = (u32) osb->s_clustersize; | |
1200 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) | |
1201 | inode->i_blocks = 0; | |
1202 | else | |
1203 | inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode)); | |
1204 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); | |
1205 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
1206 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
1207 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
1208 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
1209 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
1210 | ||
1211 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1212 | } |