Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * inode.c | |
5 | * | |
6 | * vfs' aops, fops, dops and iops | |
7 | * | |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/fs.h> | |
27 | #include <linux/types.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
30 | #include <linux/pagemap.h> | |
31 | #include <linux/smp_lock.h> | |
32 | ||
33 | #include <asm/byteorder.h> | |
34 | ||
35 | #define MLOG_MASK_PREFIX ML_INODE | |
36 | #include <cluster/masklog.h> | |
37 | ||
38 | #include "ocfs2.h" | |
39 | ||
40 | #include "alloc.h" | |
41 | #include "dlmglue.h" | |
42 | #include "extent_map.h" | |
43 | #include "file.h" | |
b4df6ed8 | 44 | #include "heartbeat.h" |
ccd979bd MF |
45 | #include "inode.h" |
46 | #include "journal.h" | |
47 | #include "namei.h" | |
48 | #include "suballoc.h" | |
49 | #include "super.h" | |
50 | #include "symlink.h" | |
51 | #include "sysfile.h" | |
52 | #include "uptodate.h" | |
53 | #include "vote.h" | |
54 | ||
55 | #include "buffer_head_io.h" | |
56 | ||
ccd979bd MF |
57 | struct ocfs2_find_inode_args |
58 | { | |
59 | u64 fi_blkno; | |
60 | unsigned long fi_ino; | |
61 | unsigned int fi_flags; | |
62 | }; | |
63 | ||
64 | static int ocfs2_read_locked_inode(struct inode *inode, | |
65 | struct ocfs2_find_inode_args *args); | |
66 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque); | |
67 | static int ocfs2_find_actor(struct inode *inode, void *opaque); | |
68 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
69 | struct inode *inode, | |
70 | struct buffer_head *fe_bh); | |
71 | ||
ca4d147e HP |
72 | void ocfs2_set_inode_flags(struct inode *inode) |
73 | { | |
74 | unsigned int flags = OCFS2_I(inode)->ip_attr; | |
75 | ||
76 | inode->i_flags &= ~(S_IMMUTABLE | | |
77 | S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC); | |
78 | ||
79 | if (flags & OCFS2_IMMUTABLE_FL) | |
80 | inode->i_flags |= S_IMMUTABLE; | |
81 | ||
82 | if (flags & OCFS2_SYNC_FL) | |
83 | inode->i_flags |= S_SYNC; | |
84 | if (flags & OCFS2_APPEND_FL) | |
85 | inode->i_flags |= S_APPEND; | |
86 | if (flags & OCFS2_NOATIME_FL) | |
87 | inode->i_flags |= S_NOATIME; | |
88 | if (flags & OCFS2_DIRSYNC_FL) | |
89 | inode->i_flags |= S_DIRSYNC; | |
90 | } | |
91 | ||
ccd979bd MF |
92 | struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb, |
93 | u64 blkno, | |
94 | int delete_vote) | |
95 | { | |
96 | struct ocfs2_find_inode_args args; | |
97 | ||
98 | /* ocfs2_ilookup_for_vote should *only* be called from the | |
99 | * vote thread */ | |
100 | BUG_ON(current != osb->vote_task); | |
101 | ||
102 | args.fi_blkno = blkno; | |
103 | args.fi_flags = OCFS2_FI_FLAG_NOWAIT; | |
104 | if (delete_vote) | |
105 | args.fi_flags |= OCFS2_FI_FLAG_DELETE; | |
106 | args.fi_ino = ino_from_blkno(osb->sb, blkno); | |
107 | return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args); | |
108 | } | |
109 | ||
24c19ef4 | 110 | struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags) |
ccd979bd MF |
111 | { |
112 | struct inode *inode = NULL; | |
113 | struct super_block *sb = osb->sb; | |
114 | struct ocfs2_find_inode_args args; | |
115 | ||
b0697053 | 116 | mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno); |
ccd979bd MF |
117 | |
118 | /* Ok. By now we've either got the offsets passed to us by the | |
119 | * caller, or we just pulled them off the bh. Lets do some | |
120 | * sanity checks to make sure they're OK. */ | |
121 | if (blkno == 0) { | |
122 | inode = ERR_PTR(-EINVAL); | |
123 | mlog_errno(PTR_ERR(inode)); | |
124 | goto bail; | |
125 | } | |
126 | ||
127 | args.fi_blkno = blkno; | |
24c19ef4 | 128 | args.fi_flags = flags; |
ccd979bd MF |
129 | args.fi_ino = ino_from_blkno(sb, blkno); |
130 | ||
131 | inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor, | |
132 | ocfs2_init_locked_inode, &args); | |
133 | /* inode was *not* in the inode cache. 2.6.x requires | |
134 | * us to do our own read_inode call and unlock it | |
135 | * afterwards. */ | |
136 | if (inode && inode->i_state & I_NEW) { | |
137 | mlog(0, "Inode was not in inode cache, reading it.\n"); | |
138 | ocfs2_read_locked_inode(inode, &args); | |
139 | unlock_new_inode(inode); | |
140 | } | |
141 | if (inode == NULL) { | |
142 | inode = ERR_PTR(-ENOMEM); | |
143 | mlog_errno(PTR_ERR(inode)); | |
144 | goto bail; | |
145 | } | |
146 | if (is_bad_inode(inode)) { | |
147 | iput(inode); | |
148 | inode = ERR_PTR(-ESTALE); | |
149 | mlog_errno(PTR_ERR(inode)); | |
150 | goto bail; | |
151 | } | |
152 | ||
153 | bail: | |
154 | if (!IS_ERR(inode)) { | |
b0697053 MF |
155 | mlog(0, "returning inode with number %llu\n", |
156 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
157 | mlog_exit_ptr(inode); |
158 | } else | |
159 | mlog_errno(PTR_ERR(inode)); | |
160 | ||
161 | return inode; | |
162 | } | |
163 | ||
164 | ||
165 | /* | |
166 | * here's how inodes get read from disk: | |
167 | * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR | |
168 | * found? : return the in-memory inode | |
169 | * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE | |
170 | */ | |
171 | ||
172 | static int ocfs2_find_actor(struct inode *inode, void *opaque) | |
173 | { | |
174 | struct ocfs2_find_inode_args *args = NULL; | |
175 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
176 | int ret = 0; | |
177 | ||
178 | mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque); | |
179 | ||
180 | args = opaque; | |
181 | ||
182 | mlog_bug_on_msg(!inode, "No inode in find actor!\n"); | |
183 | ||
184 | if (oi->ip_blkno != args->fi_blkno) | |
185 | goto bail; | |
186 | ||
187 | /* OCFS2_FI_FLAG_NOWAIT is *only* set from | |
188 | * ocfs2_ilookup_for_vote which won't create an inode for one | |
189 | * that isn't found. The vote thread which doesn't want to get | |
190 | * an inode which is in the process of going away - otherwise | |
191 | * the call to __wait_on_freeing_inode in find_inode_fast will | |
192 | * cause it to deadlock on an inode which may be waiting on a | |
193 | * vote (or lock release) in delete_inode */ | |
194 | if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) && | |
195 | (inode->i_state & (I_FREEING|I_CLEAR))) { | |
196 | /* As stated above, we're not going to return an | |
197 | * inode. In the case of a delete vote, the voting | |
198 | * code is going to signal the other node to go | |
199 | * ahead. Mark that state here, so this freeing inode | |
200 | * has the state when it gets to delete_inode. */ | |
201 | if (args->fi_flags & OCFS2_FI_FLAG_DELETE) { | |
202 | spin_lock(&oi->ip_lock); | |
203 | ocfs2_mark_inode_remotely_deleted(inode); | |
204 | spin_unlock(&oi->ip_lock); | |
205 | } | |
206 | goto bail; | |
207 | } | |
208 | ||
209 | ret = 1; | |
210 | bail: | |
211 | mlog_exit(ret); | |
212 | return ret; | |
213 | } | |
214 | ||
215 | /* | |
216 | * initialize the new inode, but don't do anything that would cause | |
217 | * us to sleep. | |
218 | * return 0 on success, 1 on failure | |
219 | */ | |
220 | static int ocfs2_init_locked_inode(struct inode *inode, void *opaque) | |
221 | { | |
222 | struct ocfs2_find_inode_args *args = opaque; | |
223 | ||
224 | mlog_entry("inode = %p, opaque = %p\n", inode, opaque); | |
225 | ||
226 | inode->i_ino = args->fi_ino; | |
227 | OCFS2_I(inode)->ip_blkno = args->fi_blkno; | |
228 | ||
229 | mlog_exit(0); | |
230 | return 0; | |
231 | } | |
232 | ||
233 | int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe, | |
234 | int create_ino) | |
235 | { | |
236 | struct super_block *sb; | |
237 | struct ocfs2_super *osb; | |
238 | int status = -EINVAL; | |
239 | ||
b0697053 MF |
240 | mlog_entry("(0x%p, size:%llu)\n", inode, |
241 | (unsigned long long)fe->i_size); | |
ccd979bd MF |
242 | |
243 | sb = inode->i_sb; | |
244 | osb = OCFS2_SB(sb); | |
245 | ||
246 | /* this means that read_inode cannot create a superblock inode | |
247 | * today. change if needed. */ | |
248 | if (!OCFS2_IS_VALID_DINODE(fe) || | |
249 | !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) { | |
b0697053 | 250 | mlog(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%llu, " |
ccd979bd | 251 | "signature = %.*s, flags = 0x%x\n", |
b0697053 MF |
252 | inode->i_ino, |
253 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, | |
ccd979bd MF |
254 | fe->i_signature, le32_to_cpu(fe->i_flags)); |
255 | goto bail; | |
256 | } | |
257 | ||
258 | if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) { | |
259 | mlog(ML_ERROR, "file entry generation does not match " | |
260 | "superblock! osb->fs_generation=%x, " | |
261 | "fe->i_fs_generation=%x\n", | |
262 | osb->fs_generation, le32_to_cpu(fe->i_fs_generation)); | |
263 | goto bail; | |
264 | } | |
265 | ||
266 | inode->i_version = 1; | |
267 | inode->i_generation = le32_to_cpu(fe->i_generation); | |
268 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
269 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
270 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
271 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
ccd979bd MF |
272 | |
273 | /* Fast symlinks will have i_size but no allocated clusters. */ | |
274 | if (S_ISLNK(inode->i_mode) && !fe->i_clusters) | |
275 | inode->i_blocks = 0; | |
276 | else | |
277 | inode->i_blocks = | |
278 | ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size)); | |
279 | inode->i_mapping->a_ops = &ocfs2_aops; | |
ccd979bd MF |
280 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); |
281 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
282 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
283 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
284 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
285 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
286 | ||
287 | if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno)) | |
288 | mlog(ML_ERROR, | |
b0697053 MF |
289 | "ip_blkno %llu != i_blkno %llu!\n", |
290 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
291 | (unsigned long long)fe->i_blkno); | |
ccd979bd MF |
292 | |
293 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
294 | OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT; | |
ca4d147e | 295 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
ccd979bd | 296 | |
ccd979bd MF |
297 | inode->i_nlink = le16_to_cpu(fe->i_links_count); |
298 | ||
24c19ef4 MF |
299 | if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) |
300 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE; | |
301 | ||
ccd979bd MF |
302 | if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) { |
303 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
304 | mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino); | |
305 | } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) { | |
306 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP; | |
307 | } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) { | |
308 | mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino); | |
309 | /* we can't actually hit this as read_inode can't | |
310 | * handle superblocks today ;-) */ | |
311 | BUG(); | |
312 | } | |
313 | ||
314 | switch (inode->i_mode & S_IFMT) { | |
315 | case S_IFREG: | |
316 | inode->i_fop = &ocfs2_fops; | |
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; | |
322 | inode->i_fop = &ocfs2_dops; | |
323 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
324 | break; | |
325 | case S_IFLNK: | |
326 | if (ocfs2_inode_is_fast_symlink(inode)) | |
327 | inode->i_op = &ocfs2_fast_symlink_inode_operations; | |
328 | else | |
329 | inode->i_op = &ocfs2_symlink_inode_operations; | |
330 | i_size_write(inode, le64_to_cpu(fe->i_size)); | |
331 | break; | |
332 | default: | |
333 | inode->i_op = &ocfs2_special_file_iops; | |
334 | init_special_inode(inode, inode->i_mode, | |
335 | inode->i_rdev); | |
336 | break; | |
337 | } | |
338 | ||
24c19ef4 MF |
339 | if (create_ino) { |
340 | inode->i_ino = ino_from_blkno(inode->i_sb, | |
341 | le64_to_cpu(fe->i_blkno)); | |
342 | ||
343 | /* | |
344 | * If we ever want to create system files from kernel, | |
345 | * the generation argument to | |
346 | * ocfs2_inode_lock_res_init() will have to change. | |
347 | */ | |
348 | BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)); | |
349 | ||
350 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
351 | OCFS2_LOCK_TYPE_META, 0, inode); | |
352 | } | |
353 | ||
ccd979bd | 354 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres, |
24c19ef4 MF |
355 | OCFS2_LOCK_TYPE_RW, inode->i_generation, |
356 | inode); | |
357 | ||
ccd979bd | 358 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres, |
24c19ef4 MF |
359 | OCFS2_LOCK_TYPE_DATA, inode->i_generation, |
360 | inode); | |
ccd979bd | 361 | |
ca4d147e | 362 | ocfs2_set_inode_flags(inode); |
ca4d147e | 363 | |
ccd979bd MF |
364 | status = 0; |
365 | bail: | |
366 | mlog_exit(status); | |
367 | return status; | |
368 | } | |
369 | ||
370 | static int ocfs2_read_locked_inode(struct inode *inode, | |
371 | struct ocfs2_find_inode_args *args) | |
372 | { | |
373 | struct super_block *sb; | |
374 | struct ocfs2_super *osb; | |
375 | struct ocfs2_dinode *fe; | |
376 | struct buffer_head *bh = NULL; | |
24c19ef4 MF |
377 | int status, can_lock; |
378 | u32 generation = 0; | |
ccd979bd MF |
379 | |
380 | mlog_entry("(0x%p, 0x%p)\n", inode, args); | |
381 | ||
382 | status = -EINVAL; | |
383 | if (inode == NULL || inode->i_sb == NULL) { | |
384 | mlog(ML_ERROR, "bad inode\n"); | |
24c19ef4 | 385 | return status; |
ccd979bd MF |
386 | } |
387 | sb = inode->i_sb; | |
388 | osb = OCFS2_SB(sb); | |
389 | ||
390 | if (!args) { | |
391 | mlog(ML_ERROR, "bad inode args\n"); | |
392 | make_bad_inode(inode); | |
24c19ef4 MF |
393 | return status; |
394 | } | |
395 | ||
396 | /* | |
397 | * To improve performance of cold-cache inode stats, we take | |
398 | * the cluster lock here if possible. | |
399 | * | |
400 | * Generally, OCFS2 never trusts the contents of an inode | |
401 | * unless it's holding a cluster lock, so taking it here isn't | |
402 | * a correctness issue as much as it is a performance | |
403 | * improvement. | |
404 | * | |
405 | * There are three times when taking the lock is not a good idea: | |
406 | * | |
407 | * 1) During startup, before we have initialized the DLM. | |
408 | * | |
409 | * 2) If we are reading certain system files which never get | |
410 | * cluster locks (local alloc, truncate log). | |
411 | * | |
412 | * 3) If the process doing the iget() is responsible for | |
413 | * orphan dir recovery. We're holding the orphan dir lock and | |
414 | * can get into a deadlock with another process on another | |
415 | * node in ->delete_inode(). | |
416 | * | |
417 | * #1 and #2 can be simply solved by never taking the lock | |
418 | * here for system files (which are the only type we read | |
419 | * during mount). It's a heavier approach, but our main | |
420 | * concern is user-accesible files anyway. | |
421 | * | |
422 | * #3 works itself out because we'll eventually take the | |
423 | * cluster lock before trusting anything anyway. | |
424 | */ | |
425 | can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
c271c5c2 SM |
426 | && !(args->fi_flags & OCFS2_FI_FLAG_NOLOCK) |
427 | && !ocfs2_mount_local(osb); | |
24c19ef4 MF |
428 | |
429 | /* | |
430 | * To maintain backwards compatibility with older versions of | |
431 | * ocfs2-tools, we still store the generation value for system | |
432 | * files. The only ones that actually matter to userspace are | |
433 | * the journals, but it's easier and inexpensive to just flag | |
434 | * all system files similarly. | |
435 | */ | |
436 | if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE) | |
437 | generation = osb->fs_generation; | |
438 | ||
439 | ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres, | |
440 | OCFS2_LOCK_TYPE_META, | |
441 | generation, inode); | |
442 | ||
443 | if (can_lock) { | |
4bcec184 | 444 | status = ocfs2_meta_lock(inode, NULL, 0); |
24c19ef4 MF |
445 | if (status) { |
446 | make_bad_inode(inode); | |
447 | mlog_errno(status); | |
448 | return status; | |
449 | } | |
ccd979bd MF |
450 | } |
451 | ||
24c19ef4 MF |
452 | status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, |
453 | can_lock ? inode : NULL); | |
ccd979bd MF |
454 | if (status < 0) { |
455 | mlog_errno(status); | |
ccd979bd MF |
456 | goto bail; |
457 | } | |
458 | ||
24c19ef4 | 459 | status = -EINVAL; |
ccd979bd MF |
460 | fe = (struct ocfs2_dinode *) bh->b_data; |
461 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
b0697053 MF |
462 | mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n", |
463 | (unsigned long long)fe->i_blkno, 7, fe->i_signature); | |
ccd979bd MF |
464 | goto bail; |
465 | } | |
466 | ||
24c19ef4 MF |
467 | /* |
468 | * This is a code bug. Right now the caller needs to | |
469 | * understand whether it is asking for a system file inode or | |
470 | * not so the proper lock names can be built. | |
471 | */ | |
472 | mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) != | |
473 | !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE), | |
474 | "Inode %llu: system file state is ambigous\n", | |
475 | (unsigned long long)args->fi_blkno); | |
ccd979bd MF |
476 | |
477 | if (S_ISCHR(le16_to_cpu(fe->i_mode)) || | |
478 | S_ISBLK(le16_to_cpu(fe->i_mode))) | |
479 | inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev)); | |
480 | ||
ccd979bd | 481 | if (ocfs2_populate_inode(inode, fe, 0) < 0) { |
b0697053 MF |
482 | mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n", |
483 | (unsigned long long)fe->i_blkno, inode->i_ino); | |
ccd979bd MF |
484 | goto bail; |
485 | } | |
486 | ||
487 | BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno)); | |
488 | ||
ccd979bd MF |
489 | status = 0; |
490 | ||
491 | bail: | |
24c19ef4 MF |
492 | if (can_lock) |
493 | ocfs2_meta_unlock(inode, 0); | |
494 | ||
495 | if (status < 0) | |
496 | make_bad_inode(inode); | |
497 | ||
ccd979bd MF |
498 | if (args && bh) |
499 | brelse(bh); | |
500 | ||
501 | mlog_exit(status); | |
502 | return status; | |
503 | } | |
504 | ||
505 | void ocfs2_sync_blockdev(struct super_block *sb) | |
506 | { | |
507 | sync_blockdev(sb->s_bdev); | |
508 | } | |
509 | ||
510 | static int ocfs2_truncate_for_delete(struct ocfs2_super *osb, | |
511 | struct inode *inode, | |
512 | struct buffer_head *fe_bh) | |
513 | { | |
514 | int status = 0; | |
1fabe148 | 515 | handle_t *handle = NULL; |
ccd979bd MF |
516 | struct ocfs2_truncate_context *tc = NULL; |
517 | struct ocfs2_dinode *fe; | |
518 | ||
519 | mlog_entry_void(); | |
520 | ||
521 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
522 | ||
523 | /* zero allocation, zero truncate :) */ | |
524 | if (!fe->i_clusters) | |
525 | goto bail; | |
526 | ||
65eff9cc | 527 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
528 | if (IS_ERR(handle)) { |
529 | status = PTR_ERR(handle); | |
530 | handle = NULL; | |
531 | mlog_errno(status); | |
532 | goto bail; | |
533 | } | |
534 | ||
535 | status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL); | |
536 | if (status < 0) { | |
537 | mlog_errno(status); | |
538 | goto bail; | |
539 | } | |
540 | ||
02dc1af4 | 541 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
542 | handle = NULL; |
543 | ||
544 | status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc); | |
545 | if (status < 0) { | |
546 | mlog_errno(status); | |
547 | goto bail; | |
548 | } | |
549 | ||
550 | status = ocfs2_commit_truncate(osb, inode, fe_bh, tc); | |
551 | if (status < 0) { | |
552 | mlog_errno(status); | |
553 | goto bail; | |
554 | } | |
555 | bail: | |
556 | if (handle) | |
02dc1af4 | 557 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
558 | |
559 | mlog_exit(status); | |
560 | return status; | |
561 | } | |
562 | ||
563 | static int ocfs2_remove_inode(struct inode *inode, | |
564 | struct buffer_head *di_bh, | |
565 | struct inode *orphan_dir_inode, | |
566 | struct buffer_head *orphan_dir_bh) | |
567 | { | |
568 | int status; | |
569 | struct inode *inode_alloc_inode = NULL; | |
570 | struct buffer_head *inode_alloc_bh = NULL; | |
1fabe148 | 571 | handle_t *handle; |
ccd979bd MF |
572 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
573 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
574 | ||
575 | inode_alloc_inode = | |
576 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
577 | le16_to_cpu(di->i_suballoc_slot)); | |
578 | if (!inode_alloc_inode) { | |
579 | status = -EEXIST; | |
580 | mlog_errno(status); | |
581 | goto bail; | |
582 | } | |
583 | ||
1b1dcc1b | 584 | mutex_lock(&inode_alloc_inode->i_mutex); |
4bcec184 | 585 | status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1); |
ccd979bd | 586 | if (status < 0) { |
1b1dcc1b | 587 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
588 | |
589 | mlog_errno(status); | |
590 | goto bail; | |
591 | } | |
592 | ||
65eff9cc | 593 | handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS); |
ccd979bd MF |
594 | if (IS_ERR(handle)) { |
595 | status = PTR_ERR(handle); | |
596 | mlog_errno(status); | |
597 | goto bail_unlock; | |
598 | } | |
599 | ||
600 | status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode, | |
601 | orphan_dir_bh); | |
602 | if (status < 0) { | |
603 | mlog_errno(status); | |
604 | goto bail_commit; | |
605 | } | |
606 | ||
607 | /* set the inodes dtime */ | |
608 | status = ocfs2_journal_access(handle, inode, di_bh, | |
609 | OCFS2_JOURNAL_ACCESS_WRITE); | |
610 | if (status < 0) { | |
611 | mlog_errno(status); | |
612 | goto bail_commit; | |
613 | } | |
614 | ||
615 | di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec); | |
616 | le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL)); | |
617 | ||
618 | status = ocfs2_journal_dirty(handle, di_bh); | |
619 | if (status < 0) { | |
620 | mlog_errno(status); | |
621 | goto bail_commit; | |
622 | } | |
623 | ||
624 | ocfs2_remove_from_cache(inode, di_bh); | |
625 | ||
626 | status = ocfs2_free_dinode(handle, inode_alloc_inode, | |
627 | inode_alloc_bh, di); | |
628 | if (status < 0) | |
629 | mlog_errno(status); | |
630 | ||
631 | bail_commit: | |
02dc1af4 | 632 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
633 | bail_unlock: |
634 | ocfs2_meta_unlock(inode_alloc_inode, 1); | |
1b1dcc1b | 635 | mutex_unlock(&inode_alloc_inode->i_mutex); |
ccd979bd MF |
636 | brelse(inode_alloc_bh); |
637 | bail: | |
638 | iput(inode_alloc_inode); | |
639 | ||
640 | return status; | |
641 | } | |
642 | ||
b4df6ed8 MF |
643 | /* |
644 | * Serialize with orphan dir recovery. If the process doing | |
645 | * recovery on this orphan dir does an iget() with the dir | |
646 | * i_mutex held, we'll deadlock here. Instead we detect this | |
647 | * and exit early - recovery will wipe this inode for us. | |
648 | */ | |
649 | static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, | |
650 | int slot) | |
651 | { | |
652 | int ret = 0; | |
653 | ||
654 | spin_lock(&osb->osb_lock); | |
655 | if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { | |
656 | mlog(0, "Recovery is happening on orphan dir %d, will skip " | |
657 | "this inode\n", slot); | |
658 | ret = -EDEADLK; | |
659 | goto out; | |
660 | } | |
661 | /* This signals to the orphan recovery process that it should | |
662 | * wait for us to handle the wipe. */ | |
663 | osb->osb_orphan_wipes[slot]++; | |
664 | out: | |
665 | spin_unlock(&osb->osb_lock); | |
666 | return ret; | |
667 | } | |
668 | ||
669 | static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, | |
670 | int slot) | |
671 | { | |
672 | spin_lock(&osb->osb_lock); | |
673 | osb->osb_orphan_wipes[slot]--; | |
674 | spin_unlock(&osb->osb_lock); | |
675 | ||
676 | wake_up(&osb->osb_wipe_event); | |
677 | } | |
678 | ||
ccd979bd MF |
679 | static int ocfs2_wipe_inode(struct inode *inode, |
680 | struct buffer_head *di_bh) | |
681 | { | |
682 | int status, orphaned_slot; | |
683 | struct inode *orphan_dir_inode = NULL; | |
684 | struct buffer_head *orphan_dir_bh = NULL; | |
685 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
686 | ||
687 | /* We've already voted on this so it should be readonly - no | |
688 | * spinlock needed. */ | |
689 | orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot; | |
b4df6ed8 MF |
690 | |
691 | status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); | |
692 | if (status) | |
693 | return status; | |
694 | ||
ccd979bd MF |
695 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
696 | ORPHAN_DIR_SYSTEM_INODE, | |
697 | orphaned_slot); | |
698 | if (!orphan_dir_inode) { | |
699 | status = -EEXIST; | |
700 | mlog_errno(status); | |
701 | goto bail; | |
702 | } | |
703 | ||
704 | /* Lock the orphan dir. The lock will be held for the entire | |
705 | * delete_inode operation. We do this now to avoid races with | |
706 | * recovery completion on other nodes. */ | |
1b1dcc1b | 707 | mutex_lock(&orphan_dir_inode->i_mutex); |
4bcec184 | 708 | status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1); |
ccd979bd | 709 | if (status < 0) { |
1b1dcc1b | 710 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
711 | |
712 | mlog_errno(status); | |
713 | goto bail; | |
714 | } | |
715 | ||
716 | /* we do this while holding the orphan dir lock because we | |
717 | * don't want recovery being run from another node to vote for | |
718 | * an inode delete on us -- this will result in two nodes | |
719 | * truncating the same file! */ | |
720 | status = ocfs2_truncate_for_delete(osb, inode, di_bh); | |
721 | if (status < 0) { | |
722 | mlog_errno(status); | |
723 | goto bail_unlock_dir; | |
724 | } | |
725 | ||
726 | status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode, | |
727 | orphan_dir_bh); | |
728 | if (status < 0) | |
729 | mlog_errno(status); | |
730 | ||
731 | bail_unlock_dir: | |
732 | ocfs2_meta_unlock(orphan_dir_inode, 1); | |
1b1dcc1b | 733 | mutex_unlock(&orphan_dir_inode->i_mutex); |
ccd979bd MF |
734 | brelse(orphan_dir_bh); |
735 | bail: | |
736 | iput(orphan_dir_inode); | |
b4df6ed8 | 737 | ocfs2_signal_wipe_completion(osb, orphaned_slot); |
ccd979bd MF |
738 | |
739 | return status; | |
740 | } | |
741 | ||
742 | /* There is a series of simple checks that should be done before a | |
743 | * vote is even considered. Encapsulate those in this function. */ | |
744 | static int ocfs2_inode_is_valid_to_delete(struct inode *inode) | |
745 | { | |
746 | int ret = 0; | |
747 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
748 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
749 | ||
750 | /* We shouldn't be getting here for the root directory | |
751 | * inode.. */ | |
752 | if (inode == osb->root_inode) { | |
753 | mlog(ML_ERROR, "Skipping delete of root inode.\n"); | |
754 | goto bail; | |
755 | } | |
756 | ||
757 | /* If we're coming from process_vote we can't go into our own | |
758 | * voting [hello, deadlock city!], so unforuntately we just | |
759 | * have to skip deleting this guy. That's OK though because | |
760 | * the node who's doing the actual deleting should handle it | |
761 | * anyway. */ | |
762 | if (current == osb->vote_task) { | |
763 | mlog(0, "Skipping delete of %lu because we're currently " | |
764 | "in process_vote\n", inode->i_ino); | |
765 | goto bail; | |
766 | } | |
767 | ||
768 | spin_lock(&oi->ip_lock); | |
769 | /* OCFS2 *never* deletes system files. This should technically | |
770 | * never get here as system file inodes should always have a | |
771 | * positive link count. */ | |
772 | if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) { | |
b0697053 MF |
773 | mlog(ML_ERROR, "Skipping delete of system file %llu\n", |
774 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
775 | goto bail_unlock; |
776 | } | |
777 | ||
778 | /* If we have voted "yes" on the wipe of this inode for | |
779 | * another node, it will be marked here so we can safely skip | |
780 | * it. Recovery will cleanup any inodes we might inadvertantly | |
781 | * skip here. */ | |
782 | if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) { | |
783 | mlog(0, "Skipping delete of %lu because another node " | |
784 | "has done this for us.\n", inode->i_ino); | |
785 | goto bail_unlock; | |
786 | } | |
787 | ||
788 | ret = 1; | |
789 | bail_unlock: | |
790 | spin_unlock(&oi->ip_lock); | |
791 | bail: | |
792 | return ret; | |
793 | } | |
794 | ||
795 | /* Query the cluster to determine whether we should wipe an inode from | |
796 | * disk or not. | |
797 | * | |
798 | * Requires the inode to have the cluster lock. */ | |
799 | static int ocfs2_query_inode_wipe(struct inode *inode, | |
800 | struct buffer_head *di_bh, | |
801 | int *wipe) | |
802 | { | |
803 | int status = 0; | |
804 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
805 | struct ocfs2_dinode *di; | |
806 | ||
807 | *wipe = 0; | |
808 | ||
809 | /* While we were waiting for the cluster lock in | |
810 | * ocfs2_delete_inode, another node might have asked to delete | |
811 | * the inode. Recheck our flags to catch this. */ | |
812 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
b0697053 MF |
813 | mlog(0, "Skipping delete of %llu because flags changed\n", |
814 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
815 | goto bail; |
816 | } | |
817 | ||
818 | /* Now that we have an up to date inode, we can double check | |
819 | * the link count. */ | |
820 | if (inode->i_nlink) { | |
b0697053 MF |
821 | mlog(0, "Skipping delete of %llu because nlink = %u\n", |
822 | (unsigned long long)oi->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
823 | goto bail; |
824 | } | |
825 | ||
826 | /* Do some basic inode verification... */ | |
827 | di = (struct ocfs2_dinode *) di_bh->b_data; | |
828 | if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) { | |
829 | /* for lack of a better error? */ | |
830 | status = -EEXIST; | |
831 | mlog(ML_ERROR, | |
b0697053 | 832 | "Inode %llu (on-disk %llu) not orphaned! " |
ccd979bd | 833 | "Disk flags 0x%x, inode flags 0x%x\n", |
b0697053 MF |
834 | (unsigned long long)oi->ip_blkno, |
835 | (unsigned long long)di->i_blkno, di->i_flags, | |
836 | oi->ip_flags); | |
ccd979bd MF |
837 | goto bail; |
838 | } | |
839 | ||
840 | /* has someone already deleted us?! baaad... */ | |
841 | if (di->i_dtime) { | |
842 | status = -EEXIST; | |
843 | mlog_errno(status); | |
844 | goto bail; | |
845 | } | |
846 | ||
847 | status = ocfs2_request_delete_vote(inode); | |
848 | /* -EBUSY means that other nodes are still using the | |
849 | * inode. We're done here though, so avoid doing anything on | |
850 | * disk and let them worry about deleting it. */ | |
851 | if (status == -EBUSY) { | |
852 | status = 0; | |
b0697053 MF |
853 | mlog(0, "Skipping delete of %llu because it is in use on" |
854 | "other nodes\n", (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
855 | goto bail; |
856 | } | |
857 | if (status < 0) { | |
858 | mlog_errno(status); | |
859 | goto bail; | |
860 | } | |
861 | ||
862 | spin_lock(&oi->ip_lock); | |
863 | if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) { | |
864 | /* Nobody knew which slot this inode was orphaned | |
865 | * into. This may happen during node death and | |
866 | * recovery knows how to clean it up so we can safely | |
867 | * ignore this inode for now on. */ | |
b0697053 MF |
868 | mlog(0, "Nobody knew where inode %llu was orphaned!\n", |
869 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
870 | } else { |
871 | *wipe = 1; | |
872 | ||
b0697053 MF |
873 | mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n", |
874 | (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot); | |
ccd979bd MF |
875 | } |
876 | spin_unlock(&oi->ip_lock); | |
877 | ||
878 | bail: | |
879 | return status; | |
880 | } | |
881 | ||
882 | /* Support function for ocfs2_delete_inode. Will help us keep the | |
883 | * inode data in a consistent state for clear_inode. Always truncates | |
884 | * pages, optionally sync's them first. */ | |
885 | static void ocfs2_cleanup_delete_inode(struct inode *inode, | |
886 | int sync_data) | |
887 | { | |
b0697053 MF |
888 | mlog(0, "Cleanup inode %llu, sync = %d\n", |
889 | (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data); | |
ccd979bd MF |
890 | if (sync_data) |
891 | write_inode_now(inode, 1); | |
892 | truncate_inode_pages(&inode->i_data, 0); | |
893 | } | |
894 | ||
895 | void ocfs2_delete_inode(struct inode *inode) | |
896 | { | |
897 | int wipe, status; | |
898 | sigset_t blocked, oldset; | |
899 | struct buffer_head *di_bh = NULL; | |
900 | ||
901 | mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino); | |
902 | ||
903 | if (is_bad_inode(inode)) { | |
904 | mlog(0, "Skipping delete of bad inode\n"); | |
905 | goto bail; | |
906 | } | |
907 | ||
908 | if (!ocfs2_inode_is_valid_to_delete(inode)) { | |
909 | /* It's probably not necessary to truncate_inode_pages | |
910 | * here but we do it for safety anyway (it will most | |
911 | * likely be a no-op anyway) */ | |
912 | ocfs2_cleanup_delete_inode(inode, 0); | |
913 | goto bail; | |
914 | } | |
915 | ||
916 | /* We want to block signals in delete_inode as the lock and | |
917 | * messaging paths may return us -ERESTARTSYS. Which would | |
918 | * cause us to exit early, resulting in inodes being orphaned | |
919 | * forever. */ | |
920 | sigfillset(&blocked); | |
921 | status = sigprocmask(SIG_BLOCK, &blocked, &oldset); | |
922 | if (status < 0) { | |
923 | mlog_errno(status); | |
924 | ocfs2_cleanup_delete_inode(inode, 1); | |
925 | goto bail; | |
926 | } | |
927 | ||
928 | /* Lock down the inode. This gives us an up to date view of | |
929 | * it's metadata (for verification), and allows us to | |
930 | * serialize delete_inode votes. | |
931 | * | |
932 | * Even though we might be doing a truncate, we don't take the | |
933 | * allocation lock here as it won't be needed - nobody will | |
934 | * have the file open. | |
935 | */ | |
4bcec184 | 936 | status = ocfs2_meta_lock(inode, &di_bh, 1); |
ccd979bd MF |
937 | if (status < 0) { |
938 | if (status != -ENOENT) | |
939 | mlog_errno(status); | |
940 | ocfs2_cleanup_delete_inode(inode, 0); | |
941 | goto bail_unblock; | |
942 | } | |
943 | ||
944 | /* Query the cluster. This will be the final decision made | |
945 | * before we go ahead and wipe the inode. */ | |
946 | status = ocfs2_query_inode_wipe(inode, di_bh, &wipe); | |
947 | if (!wipe || status < 0) { | |
948 | /* Error and inode busy vote both mean we won't be | |
949 | * removing the inode, so they take almost the same | |
950 | * path. */ | |
951 | if (status < 0) | |
952 | mlog_errno(status); | |
953 | ||
954 | /* Someone in the cluster has voted to not wipe this | |
955 | * inode, or it was never completely orphaned. Write | |
956 | * out the pages and exit now. */ | |
957 | ocfs2_cleanup_delete_inode(inode, 1); | |
958 | goto bail_unlock_inode; | |
959 | } | |
960 | ||
961 | ocfs2_cleanup_delete_inode(inode, 0); | |
962 | ||
963 | status = ocfs2_wipe_inode(inode, di_bh); | |
964 | if (status < 0) { | |
b4df6ed8 MF |
965 | if (status != -EDEADLK) |
966 | mlog_errno(status); | |
ccd979bd MF |
967 | goto bail_unlock_inode; |
968 | } | |
969 | ||
24c19ef4 MF |
970 | /* |
971 | * Mark the inode as successfully deleted. | |
972 | * | |
973 | * This is important for ocfs2_clear_inode() as it will check | |
974 | * this flag and skip any checkpointing work | |
975 | * | |
976 | * ocfs2_stuff_meta_lvb() also uses this flag to invalidate | |
977 | * the LVB for other nodes. | |
978 | */ | |
ccd979bd MF |
979 | OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED; |
980 | ||
981 | bail_unlock_inode: | |
982 | ocfs2_meta_unlock(inode, 1); | |
983 | brelse(di_bh); | |
984 | bail_unblock: | |
985 | status = sigprocmask(SIG_SETMASK, &oldset, NULL); | |
986 | if (status < 0) | |
987 | mlog_errno(status); | |
988 | bail: | |
989 | clear_inode(inode); | |
990 | mlog_exit_void(); | |
991 | } | |
992 | ||
993 | void ocfs2_clear_inode(struct inode *inode) | |
994 | { | |
995 | int status; | |
996 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
997 | ||
998 | mlog_entry_void(); | |
999 | ||
1000 | if (!inode) | |
1001 | goto bail; | |
1002 | ||
b0697053 MF |
1003 | mlog(0, "Clearing inode: %llu, nlink = %u\n", |
1004 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink); | |
ccd979bd MF |
1005 | |
1006 | mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL, | |
1007 | "Inode=%lu\n", inode->i_ino); | |
1008 | ||
1009 | /* Do these before all the other work so that we don't bounce | |
1010 | * the vote thread while waiting to destroy the locks. */ | |
1011 | ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres); | |
1012 | ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres); | |
1013 | ocfs2_mark_lockres_freeing(&oi->ip_data_lockres); | |
1014 | ||
1015 | /* We very well may get a clear_inode before all an inodes | |
1016 | * metadata has hit disk. Of course, we can't drop any cluster | |
1017 | * locks until the journal has finished with it. The only | |
1018 | * exception here are successfully wiped inodes - their | |
1019 | * metadata can now be considered to be part of the system | |
1020 | * inodes from which it came. */ | |
1021 | if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED)) | |
1022 | ocfs2_checkpoint_inode(inode); | |
1023 | ||
1024 | mlog_bug_on_msg(!list_empty(&oi->ip_io_markers), | |
b0697053 MF |
1025 | "Clear inode of %llu, inode has io markers\n", |
1026 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1027 | |
1028 | ocfs2_extent_map_drop(inode, 0); | |
1029 | ocfs2_extent_map_init(inode); | |
1030 | ||
1031 | status = ocfs2_drop_inode_locks(inode); | |
1032 | if (status < 0) | |
1033 | mlog_errno(status); | |
1034 | ||
1035 | ocfs2_lock_res_free(&oi->ip_rw_lockres); | |
1036 | ocfs2_lock_res_free(&oi->ip_meta_lockres); | |
1037 | ocfs2_lock_res_free(&oi->ip_data_lockres); | |
1038 | ||
1039 | ocfs2_metadata_cache_purge(inode); | |
1040 | ||
1041 | mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached, | |
b0697053 MF |
1042 | "Clear inode of %llu, inode has %u cache items\n", |
1043 | (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached); | |
ccd979bd MF |
1044 | |
1045 | mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE), | |
b0697053 MF |
1046 | "Clear inode of %llu, inode has a bad flag\n", |
1047 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1048 | |
1049 | mlog_bug_on_msg(spin_is_locked(&oi->ip_lock), | |
b0697053 MF |
1050 | "Clear inode of %llu, inode is locked\n", |
1051 | (unsigned long long)oi->ip_blkno); | |
ccd979bd | 1052 | |
251b6ecc | 1053 | mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex), |
b0697053 MF |
1054 | "Clear inode of %llu, io_mutex is locked\n", |
1055 | (unsigned long long)oi->ip_blkno); | |
251b6ecc | 1056 | mutex_unlock(&oi->ip_io_mutex); |
ccd979bd MF |
1057 | |
1058 | /* | |
1059 | * down_trylock() returns 0, down_write_trylock() returns 1 | |
1060 | * kernel 1, world 0 | |
1061 | */ | |
1062 | mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem), | |
b0697053 MF |
1063 | "Clear inode of %llu, alloc_sem is locked\n", |
1064 | (unsigned long long)oi->ip_blkno); | |
ccd979bd MF |
1065 | up_write(&oi->ip_alloc_sem); |
1066 | ||
1067 | mlog_bug_on_msg(oi->ip_open_count, | |
b0697053 MF |
1068 | "Clear inode of %llu has open count %d\n", |
1069 | (unsigned long long)oi->ip_blkno, oi->ip_open_count); | |
ccd979bd MF |
1070 | |
1071 | /* Clear all other flags. */ | |
1072 | oi->ip_flags = OCFS2_INODE_CACHE_INLINE; | |
1073 | oi->ip_created_trans = 0; | |
1074 | oi->ip_last_trans = 0; | |
1075 | oi->ip_dir_start_lookup = 0; | |
1076 | oi->ip_blkno = 0ULL; | |
1077 | ||
1078 | bail: | |
1079 | mlog_exit_void(); | |
1080 | } | |
1081 | ||
1082 | /* Called under inode_lock, with no more references on the | |
1083 | * struct inode, so it's safe here to check the flags field | |
1084 | * and to manipulate i_nlink without any other locks. */ | |
1085 | void ocfs2_drop_inode(struct inode *inode) | |
1086 | { | |
1087 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1088 | ||
1089 | mlog_entry_void(); | |
1090 | ||
b0697053 MF |
1091 | mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n", |
1092 | (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags); | |
ccd979bd MF |
1093 | |
1094 | /* Testing ip_orphaned_slot here wouldn't work because we may | |
1095 | * not have gotten a delete_inode vote from any other nodes | |
1096 | * yet. */ | |
379dfe9d MF |
1097 | if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) |
1098 | generic_delete_inode(inode); | |
1099 | else | |
1100 | generic_drop_inode(inode); | |
ccd979bd MF |
1101 | |
1102 | mlog_exit_void(); | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * TODO: this should probably be merged into ocfs2_get_block | |
1107 | * | |
1108 | * However, you now need to pay attention to the cont_prepare_write() | |
1109 | * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much | |
1110 | * expects never to extend). | |
1111 | */ | |
1112 | struct buffer_head *ocfs2_bread(struct inode *inode, | |
1113 | int block, int *err, int reada) | |
1114 | { | |
1115 | struct buffer_head *bh = NULL; | |
1116 | int tmperr; | |
1117 | u64 p_blkno; | |
1118 | int readflags = OCFS2_BH_CACHED; | |
1119 | ||
ccd979bd MF |
1120 | if (reada) |
1121 | readflags |= OCFS2_BH_READAHEAD; | |
ccd979bd MF |
1122 | |
1123 | if (((u64)block << inode->i_sb->s_blocksize_bits) >= | |
1124 | i_size_read(inode)) { | |
1125 | BUG_ON(!reada); | |
1126 | return NULL; | |
1127 | } | |
1128 | ||
1129 | tmperr = ocfs2_extent_map_get_blocks(inode, block, 1, | |
1130 | &p_blkno, NULL); | |
1131 | if (tmperr < 0) { | |
1132 | mlog_errno(tmperr); | |
1133 | goto fail; | |
1134 | } | |
1135 | ||
1136 | tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh, | |
1137 | readflags, inode); | |
1138 | if (tmperr < 0) | |
1139 | goto fail; | |
1140 | ||
1141 | tmperr = 0; | |
1142 | ||
1143 | *err = 0; | |
1144 | return bh; | |
1145 | ||
1146 | fail: | |
1147 | if (bh) { | |
1148 | brelse(bh); | |
1149 | bh = NULL; | |
1150 | } | |
1151 | *err = -EIO; | |
1152 | return NULL; | |
1153 | } | |
1154 | ||
1155 | /* | |
1156 | * This is called from our getattr. | |
1157 | */ | |
1158 | int ocfs2_inode_revalidate(struct dentry *dentry) | |
1159 | { | |
1160 | struct inode *inode = dentry->d_inode; | |
1161 | int status = 0; | |
1162 | ||
b0697053 MF |
1163 | mlog_entry("(inode = 0x%p, ino = %llu)\n", inode, |
1164 | inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL); | |
ccd979bd MF |
1165 | |
1166 | if (!inode) { | |
1167 | mlog(0, "eep, no inode!\n"); | |
1168 | status = -ENOENT; | |
1169 | goto bail; | |
1170 | } | |
1171 | ||
1172 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1173 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
1174 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1175 | mlog(0, "inode deleted!\n"); | |
1176 | status = -ENOENT; | |
1177 | goto bail; | |
1178 | } | |
1179 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1180 | ||
1181 | /* Let ocfs2_meta_lock do the work of updating our struct | |
1182 | * inode for us. */ | |
4bcec184 | 1183 | status = ocfs2_meta_lock(inode, NULL, 0); |
ccd979bd MF |
1184 | if (status < 0) { |
1185 | if (status != -ENOENT) | |
1186 | mlog_errno(status); | |
1187 | goto bail; | |
1188 | } | |
1189 | ocfs2_meta_unlock(inode, 0); | |
1190 | bail: | |
1191 | mlog_exit(status); | |
1192 | ||
1193 | return status; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * Updates a disk inode from a | |
1198 | * struct inode. | |
1199 | * Only takes ip_lock. | |
1200 | */ | |
1fabe148 | 1201 | int ocfs2_mark_inode_dirty(handle_t *handle, |
ccd979bd MF |
1202 | struct inode *inode, |
1203 | struct buffer_head *bh) | |
1204 | { | |
1205 | int status; | |
1206 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
1207 | ||
b0697053 MF |
1208 | mlog_entry("(inode %llu)\n", |
1209 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
1210 | |
1211 | status = ocfs2_journal_access(handle, inode, bh, | |
1212 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1213 | if (status < 0) { | |
1214 | mlog_errno(status); | |
1215 | goto leave; | |
1216 | } | |
1217 | ||
1218 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1219 | fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
ca4d147e | 1220 | fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr); |
ccd979bd MF |
1221 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1222 | ||
1223 | fe->i_size = cpu_to_le64(i_size_read(inode)); | |
1224 | fe->i_links_count = cpu_to_le16(inode->i_nlink); | |
1225 | fe->i_uid = cpu_to_le32(inode->i_uid); | |
1226 | fe->i_gid = cpu_to_le32(inode->i_gid); | |
1227 | fe->i_mode = cpu_to_le16(inode->i_mode); | |
1228 | fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec); | |
1229 | fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
1230 | fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec); | |
1231 | fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
1232 | fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec); | |
1233 | fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
1234 | ||
1235 | status = ocfs2_journal_dirty(handle, bh); | |
1236 | if (status < 0) | |
1237 | mlog_errno(status); | |
1238 | ||
1239 | status = 0; | |
1240 | leave: | |
1241 | ||
1242 | mlog_exit(status); | |
1243 | return status; | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * | |
1248 | * Updates a struct inode from a disk inode. | |
1249 | * does no i/o, only takes ip_lock. | |
1250 | */ | |
1251 | void ocfs2_refresh_inode(struct inode *inode, | |
1252 | struct ocfs2_dinode *fe) | |
1253 | { | |
ccd979bd MF |
1254 | spin_lock(&OCFS2_I(inode)->ip_lock); |
1255 | ||
1256 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
ca4d147e HP |
1257 | OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr); |
1258 | ocfs2_set_inode_flags(inode); | |
ccd979bd MF |
1259 | i_size_write(inode, le64_to_cpu(fe->i_size)); |
1260 | inode->i_nlink = le16_to_cpu(fe->i_links_count); | |
1261 | inode->i_uid = le32_to_cpu(fe->i_uid); | |
1262 | inode->i_gid = le32_to_cpu(fe->i_gid); | |
1263 | inode->i_mode = le16_to_cpu(fe->i_mode); | |
ccd979bd MF |
1264 | if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0) |
1265 | inode->i_blocks = 0; | |
1266 | else | |
1267 | inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode)); | |
1268 | inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime); | |
1269 | inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec); | |
1270 | inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime); | |
1271 | inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec); | |
1272 | inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime); | |
1273 | inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec); | |
1274 | ||
1275 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1276 | } |