]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/fs/ext4/namei.c | |
3 | * | |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/namei.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
15 | * Big-endian to little-endian byte-swapping/bitmaps by | |
16 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
17 | * Directory entry file type support and forward compatibility hooks | |
18 | * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998 | |
19 | * Hash Tree Directory indexing (c) | |
20 | * Daniel Phillips, 2001 | |
21 | * Hash Tree Directory indexing porting | |
22 | * Christopher Li, 2002 | |
23 | * Hash Tree Directory indexing cleanup | |
24 | * Theodore Ts'o, 2002 | |
25 | */ | |
26 | ||
27 | #include <linux/fs.h> | |
28 | #include <linux/pagemap.h> | |
29 | #include <linux/jbd2.h> | |
30 | #include <linux/time.h> | |
31 | #include <linux/fcntl.h> | |
32 | #include <linux/stat.h> | |
33 | #include <linux/string.h> | |
34 | #include <linux/quotaops.h> | |
35 | #include <linux/buffer_head.h> | |
36 | #include <linux/bio.h> | |
37 | #include "ext4.h" | |
38 | #include "ext4_jbd2.h" | |
39 | ||
40 | #include "xattr.h" | |
41 | #include "acl.h" | |
42 | ||
43 | /* | |
44 | * define how far ahead to read directories while searching them. | |
45 | */ | |
46 | #define NAMEI_RA_CHUNKS 2 | |
47 | #define NAMEI_RA_BLOCKS 4 | |
48 | #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) | |
49 | #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b)) | |
50 | ||
51 | static struct buffer_head *ext4_append(handle_t *handle, | |
52 | struct inode *inode, | |
53 | ext4_lblk_t *block, int *err) | |
54 | { | |
55 | struct buffer_head *bh; | |
56 | ||
57 | *block = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
58 | ||
59 | bh = ext4_bread(handle, inode, *block, 1, err); | |
60 | if (bh) { | |
61 | inode->i_size += inode->i_sb->s_blocksize; | |
62 | EXT4_I(inode)->i_disksize = inode->i_size; | |
63 | *err = ext4_journal_get_write_access(handle, bh); | |
64 | if (*err) { | |
65 | brelse(bh); | |
66 | bh = NULL; | |
67 | } | |
68 | } | |
69 | return bh; | |
70 | } | |
71 | ||
72 | #ifndef assert | |
73 | #define assert(test) J_ASSERT(test) | |
74 | #endif | |
75 | ||
76 | #ifdef DX_DEBUG | |
77 | #define dxtrace(command) command | |
78 | #else | |
79 | #define dxtrace(command) | |
80 | #endif | |
81 | ||
82 | struct fake_dirent | |
83 | { | |
84 | __le32 inode; | |
85 | __le16 rec_len; | |
86 | u8 name_len; | |
87 | u8 file_type; | |
88 | }; | |
89 | ||
90 | struct dx_countlimit | |
91 | { | |
92 | __le16 limit; | |
93 | __le16 count; | |
94 | }; | |
95 | ||
96 | struct dx_entry | |
97 | { | |
98 | __le32 hash; | |
99 | __le32 block; | |
100 | }; | |
101 | ||
102 | /* | |
103 | * dx_root_info is laid out so that if it should somehow get overlaid by a | |
104 | * dirent the two low bits of the hash version will be zero. Therefore, the | |
105 | * hash version mod 4 should never be 0. Sincerely, the paranoia department. | |
106 | */ | |
107 | ||
108 | struct dx_root | |
109 | { | |
110 | struct fake_dirent dot; | |
111 | char dot_name[4]; | |
112 | struct fake_dirent dotdot; | |
113 | char dotdot_name[4]; | |
114 | struct dx_root_info | |
115 | { | |
116 | __le32 reserved_zero; | |
117 | u8 hash_version; | |
118 | u8 info_length; /* 8 */ | |
119 | u8 indirect_levels; | |
120 | u8 unused_flags; | |
121 | } | |
122 | info; | |
123 | struct dx_entry entries[0]; | |
124 | }; | |
125 | ||
126 | struct dx_node | |
127 | { | |
128 | struct fake_dirent fake; | |
129 | struct dx_entry entries[0]; | |
130 | }; | |
131 | ||
132 | ||
133 | struct dx_frame | |
134 | { | |
135 | struct buffer_head *bh; | |
136 | struct dx_entry *entries; | |
137 | struct dx_entry *at; | |
138 | }; | |
139 | ||
140 | struct dx_map_entry | |
141 | { | |
142 | u32 hash; | |
143 | u16 offs; | |
144 | u16 size; | |
145 | }; | |
146 | ||
147 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry); | |
148 | static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value); | |
149 | static inline unsigned dx_get_hash(struct dx_entry *entry); | |
150 | static void dx_set_hash(struct dx_entry *entry, unsigned value); | |
151 | static unsigned dx_get_count(struct dx_entry *entries); | |
152 | static unsigned dx_get_limit(struct dx_entry *entries); | |
153 | static void dx_set_count(struct dx_entry *entries, unsigned value); | |
154 | static void dx_set_limit(struct dx_entry *entries, unsigned value); | |
155 | static unsigned dx_root_limit(struct inode *dir, unsigned infosize); | |
156 | static unsigned dx_node_limit(struct inode *dir); | |
157 | static struct dx_frame *dx_probe(const struct qstr *d_name, | |
158 | struct inode *dir, | |
159 | struct dx_hash_info *hinfo, | |
160 | struct dx_frame *frame, | |
161 | int *err); | |
162 | static void dx_release(struct dx_frame *frames); | |
163 | static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize, | |
164 | struct dx_hash_info *hinfo, struct dx_map_entry map[]); | |
165 | static void dx_sort_map(struct dx_map_entry *map, unsigned count); | |
166 | static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to, | |
167 | struct dx_map_entry *offsets, int count, unsigned blocksize); | |
168 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize); | |
169 | static void dx_insert_block(struct dx_frame *frame, | |
170 | u32 hash, ext4_lblk_t block); | |
171 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, | |
172 | struct dx_frame *frame, | |
173 | struct dx_frame *frames, | |
174 | __u32 *start_hash); | |
175 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir, | |
176 | const struct qstr *d_name, | |
177 | struct ext4_dir_entry_2 **res_dir, | |
178 | int *err); | |
179 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, | |
180 | struct inode *inode); | |
181 | ||
182 | unsigned int ext4_rec_len_from_disk(__le16 dlen, unsigned blocksize) | |
183 | { | |
184 | unsigned len = le16_to_cpu(dlen); | |
185 | ||
186 | if (len == EXT4_MAX_REC_LEN || len == 0) | |
187 | return blocksize; | |
188 | return (len & 65532) | ((len & 3) << 16); | |
189 | } | |
190 | ||
191 | __le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize) | |
192 | { | |
193 | if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3)) | |
194 | BUG(); | |
195 | if (len < 65536) | |
196 | return cpu_to_le16(len); | |
197 | if (len == blocksize) { | |
198 | if (blocksize == 65536) | |
199 | return cpu_to_le16(EXT4_MAX_REC_LEN); | |
200 | else | |
201 | return cpu_to_le16(0); | |
202 | } | |
203 | return cpu_to_le16((len & 65532) | ((len >> 16) & 3)); | |
204 | } | |
205 | ||
206 | /* | |
207 | * p is at least 6 bytes before the end of page | |
208 | */ | |
209 | static inline struct ext4_dir_entry_2 * | |
210 | ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) | |
211 | { | |
212 | return (struct ext4_dir_entry_2 *)((char *)p + | |
213 | ext4_rec_len_from_disk(p->rec_len, blocksize)); | |
214 | } | |
215 | ||
216 | /* | |
217 | * Future: use high four bits of block for coalesce-on-delete flags | |
218 | * Mask them off for now. | |
219 | */ | |
220 | ||
221 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) | |
222 | { | |
223 | return le32_to_cpu(entry->block) & 0x00ffffff; | |
224 | } | |
225 | ||
226 | static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) | |
227 | { | |
228 | entry->block = cpu_to_le32(value); | |
229 | } | |
230 | ||
231 | static inline unsigned dx_get_hash(struct dx_entry *entry) | |
232 | { | |
233 | return le32_to_cpu(entry->hash); | |
234 | } | |
235 | ||
236 | static inline void dx_set_hash(struct dx_entry *entry, unsigned value) | |
237 | { | |
238 | entry->hash = cpu_to_le32(value); | |
239 | } | |
240 | ||
241 | static inline unsigned dx_get_count(struct dx_entry *entries) | |
242 | { | |
243 | return le16_to_cpu(((struct dx_countlimit *) entries)->count); | |
244 | } | |
245 | ||
246 | static inline unsigned dx_get_limit(struct dx_entry *entries) | |
247 | { | |
248 | return le16_to_cpu(((struct dx_countlimit *) entries)->limit); | |
249 | } | |
250 | ||
251 | static inline void dx_set_count(struct dx_entry *entries, unsigned value) | |
252 | { | |
253 | ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); | |
254 | } | |
255 | ||
256 | static inline void dx_set_limit(struct dx_entry *entries, unsigned value) | |
257 | { | |
258 | ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); | |
259 | } | |
260 | ||
261 | static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) | |
262 | { | |
263 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - | |
264 | EXT4_DIR_REC_LEN(2) - infosize; | |
265 | return entry_space / sizeof(struct dx_entry); | |
266 | } | |
267 | ||
268 | static inline unsigned dx_node_limit(struct inode *dir) | |
269 | { | |
270 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); | |
271 | return entry_space / sizeof(struct dx_entry); | |
272 | } | |
273 | ||
274 | /* | |
275 | * Debug | |
276 | */ | |
277 | #ifdef DX_DEBUG | |
278 | static void dx_show_index(char * label, struct dx_entry *entries) | |
279 | { | |
280 | int i, n = dx_get_count (entries); | |
281 | printk(KERN_DEBUG "%s index ", label); | |
282 | for (i = 0; i < n; i++) { | |
283 | printk("%x->%lu ", i ? dx_get_hash(entries + i) : | |
284 | 0, (unsigned long)dx_get_block(entries + i)); | |
285 | } | |
286 | printk("\n"); | |
287 | } | |
288 | ||
289 | struct stats | |
290 | { | |
291 | unsigned names; | |
292 | unsigned space; | |
293 | unsigned bcount; | |
294 | }; | |
295 | ||
296 | static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de, | |
297 | int size, int show_names) | |
298 | { | |
299 | unsigned names = 0, space = 0; | |
300 | char *base = (char *) de; | |
301 | struct dx_hash_info h = *hinfo; | |
302 | ||
303 | printk("names: "); | |
304 | while ((char *) de < base + size) | |
305 | { | |
306 | if (de->inode) | |
307 | { | |
308 | if (show_names) | |
309 | { | |
310 | int len = de->name_len; | |
311 | char *name = de->name; | |
312 | while (len--) printk("%c", *name++); | |
313 | ext4fs_dirhash(de->name, de->name_len, &h); | |
314 | printk(":%x.%u ", h.hash, | |
315 | ((char *) de - base)); | |
316 | } | |
317 | space += EXT4_DIR_REC_LEN(de->name_len); | |
318 | names++; | |
319 | } | |
320 | de = ext4_next_entry(de, size); | |
321 | } | |
322 | printk("(%i)\n", names); | |
323 | return (struct stats) { names, space, 1 }; | |
324 | } | |
325 | ||
326 | struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, | |
327 | struct dx_entry *entries, int levels) | |
328 | { | |
329 | unsigned blocksize = dir->i_sb->s_blocksize; | |
330 | unsigned count = dx_get_count(entries), names = 0, space = 0, i; | |
331 | unsigned bcount = 0; | |
332 | struct buffer_head *bh; | |
333 | int err; | |
334 | printk("%i indexed blocks...\n", count); | |
335 | for (i = 0; i < count; i++, entries++) | |
336 | { | |
337 | ext4_lblk_t block = dx_get_block(entries); | |
338 | ext4_lblk_t hash = i ? dx_get_hash(entries): 0; | |
339 | u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; | |
340 | struct stats stats; | |
341 | printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); | |
342 | if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue; | |
343 | stats = levels? | |
344 | dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): | |
345 | dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0); | |
346 | names += stats.names; | |
347 | space += stats.space; | |
348 | bcount += stats.bcount; | |
349 | brelse(bh); | |
350 | } | |
351 | if (bcount) | |
352 | printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", | |
353 | levels ? "" : " ", names, space/bcount, | |
354 | (space/bcount)*100/blocksize); | |
355 | return (struct stats) { names, space, bcount}; | |
356 | } | |
357 | #endif /* DX_DEBUG */ | |
358 | ||
359 | /* | |
360 | * Probe for a directory leaf block to search. | |
361 | * | |
362 | * dx_probe can return ERR_BAD_DX_DIR, which means there was a format | |
363 | * error in the directory index, and the caller should fall back to | |
364 | * searching the directory normally. The callers of dx_probe **MUST** | |
365 | * check for this error code, and make sure it never gets reflected | |
366 | * back to userspace. | |
367 | */ | |
368 | static struct dx_frame * | |
369 | dx_probe(const struct qstr *d_name, struct inode *dir, | |
370 | struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err) | |
371 | { | |
372 | unsigned count, indirect; | |
373 | struct dx_entry *at, *entries, *p, *q, *m; | |
374 | struct dx_root *root; | |
375 | struct buffer_head *bh; | |
376 | struct dx_frame *frame = frame_in; | |
377 | u32 hash; | |
378 | ||
379 | frame->bh = NULL; | |
380 | if (!(bh = ext4_bread (NULL,dir, 0, 0, err))) | |
381 | goto fail; | |
382 | root = (struct dx_root *) bh->b_data; | |
383 | if (root->info.hash_version != DX_HASH_TEA && | |
384 | root->info.hash_version != DX_HASH_HALF_MD4 && | |
385 | root->info.hash_version != DX_HASH_LEGACY) { | |
386 | ext4_warning(dir->i_sb, __func__, | |
387 | "Unrecognised inode hash code %d", | |
388 | root->info.hash_version); | |
389 | brelse(bh); | |
390 | *err = ERR_BAD_DX_DIR; | |
391 | goto fail; | |
392 | } | |
393 | hinfo->hash_version = root->info.hash_version; | |
394 | if (hinfo->hash_version <= DX_HASH_TEA) | |
395 | hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
396 | hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; | |
397 | if (d_name) | |
398 | ext4fs_dirhash(d_name->name, d_name->len, hinfo); | |
399 | hash = hinfo->hash; | |
400 | ||
401 | if (root->info.unused_flags & 1) { | |
402 | ext4_warning(dir->i_sb, __func__, | |
403 | "Unimplemented inode hash flags: %#06x", | |
404 | root->info.unused_flags); | |
405 | brelse(bh); | |
406 | *err = ERR_BAD_DX_DIR; | |
407 | goto fail; | |
408 | } | |
409 | ||
410 | if ((indirect = root->info.indirect_levels) > 1) { | |
411 | ext4_warning(dir->i_sb, __func__, | |
412 | "Unimplemented inode hash depth: %#06x", | |
413 | root->info.indirect_levels); | |
414 | brelse(bh); | |
415 | *err = ERR_BAD_DX_DIR; | |
416 | goto fail; | |
417 | } | |
418 | ||
419 | entries = (struct dx_entry *) (((char *)&root->info) + | |
420 | root->info.info_length); | |
421 | ||
422 | if (dx_get_limit(entries) != dx_root_limit(dir, | |
423 | root->info.info_length)) { | |
424 | ext4_warning(dir->i_sb, __func__, | |
425 | "dx entry: limit != root limit"); | |
426 | brelse(bh); | |
427 | *err = ERR_BAD_DX_DIR; | |
428 | goto fail; | |
429 | } | |
430 | ||
431 | dxtrace(printk("Look up %x", hash)); | |
432 | while (1) | |
433 | { | |
434 | count = dx_get_count(entries); | |
435 | if (!count || count > dx_get_limit(entries)) { | |
436 | ext4_warning(dir->i_sb, __func__, | |
437 | "dx entry: no count or count > limit"); | |
438 | brelse(bh); | |
439 | *err = ERR_BAD_DX_DIR; | |
440 | goto fail2; | |
441 | } | |
442 | ||
443 | p = entries + 1; | |
444 | q = entries + count - 1; | |
445 | while (p <= q) | |
446 | { | |
447 | m = p + (q - p)/2; | |
448 | dxtrace(printk(".")); | |
449 | if (dx_get_hash(m) > hash) | |
450 | q = m - 1; | |
451 | else | |
452 | p = m + 1; | |
453 | } | |
454 | ||
455 | if (0) // linear search cross check | |
456 | { | |
457 | unsigned n = count - 1; | |
458 | at = entries; | |
459 | while (n--) | |
460 | { | |
461 | dxtrace(printk(",")); | |
462 | if (dx_get_hash(++at) > hash) | |
463 | { | |
464 | at--; | |
465 | break; | |
466 | } | |
467 | } | |
468 | assert (at == p - 1); | |
469 | } | |
470 | ||
471 | at = p - 1; | |
472 | dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at))); | |
473 | frame->bh = bh; | |
474 | frame->entries = entries; | |
475 | frame->at = at; | |
476 | if (!indirect--) return frame; | |
477 | if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err))) | |
478 | goto fail2; | |
479 | at = entries = ((struct dx_node *) bh->b_data)->entries; | |
480 | if (dx_get_limit(entries) != dx_node_limit (dir)) { | |
481 | ext4_warning(dir->i_sb, __func__, | |
482 | "dx entry: limit != node limit"); | |
483 | brelse(bh); | |
484 | *err = ERR_BAD_DX_DIR; | |
485 | goto fail2; | |
486 | } | |
487 | frame++; | |
488 | frame->bh = NULL; | |
489 | } | |
490 | fail2: | |
491 | while (frame >= frame_in) { | |
492 | brelse(frame->bh); | |
493 | frame--; | |
494 | } | |
495 | fail: | |
496 | if (*err == ERR_BAD_DX_DIR) | |
497 | ext4_warning(dir->i_sb, __func__, | |
498 | "Corrupt dir inode %ld, running e2fsck is " | |
499 | "recommended.", dir->i_ino); | |
500 | return NULL; | |
501 | } | |
502 | ||
503 | static void dx_release (struct dx_frame *frames) | |
504 | { | |
505 | if (frames[0].bh == NULL) | |
506 | return; | |
507 | ||
508 | if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels) | |
509 | brelse(frames[1].bh); | |
510 | brelse(frames[0].bh); | |
511 | } | |
512 | ||
513 | /* | |
514 | * This function increments the frame pointer to search the next leaf | |
515 | * block, and reads in the necessary intervening nodes if the search | |
516 | * should be necessary. Whether or not the search is necessary is | |
517 | * controlled by the hash parameter. If the hash value is even, then | |
518 | * the search is only continued if the next block starts with that | |
519 | * hash value. This is used if we are searching for a specific file. | |
520 | * | |
521 | * If the hash value is HASH_NB_ALWAYS, then always go to the next block. | |
522 | * | |
523 | * This function returns 1 if the caller should continue to search, | |
524 | * or 0 if it should not. If there is an error reading one of the | |
525 | * index blocks, it will a negative error code. | |
526 | * | |
527 | * If start_hash is non-null, it will be filled in with the starting | |
528 | * hash of the next page. | |
529 | */ | |
530 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, | |
531 | struct dx_frame *frame, | |
532 | struct dx_frame *frames, | |
533 | __u32 *start_hash) | |
534 | { | |
535 | struct dx_frame *p; | |
536 | struct buffer_head *bh; | |
537 | int err, num_frames = 0; | |
538 | __u32 bhash; | |
539 | ||
540 | p = frame; | |
541 | /* | |
542 | * Find the next leaf page by incrementing the frame pointer. | |
543 | * If we run out of entries in the interior node, loop around and | |
544 | * increment pointer in the parent node. When we break out of | |
545 | * this loop, num_frames indicates the number of interior | |
546 | * nodes need to be read. | |
547 | */ | |
548 | while (1) { | |
549 | if (++(p->at) < p->entries + dx_get_count(p->entries)) | |
550 | break; | |
551 | if (p == frames) | |
552 | return 0; | |
553 | num_frames++; | |
554 | p--; | |
555 | } | |
556 | ||
557 | /* | |
558 | * If the hash is 1, then continue only if the next page has a | |
559 | * continuation hash of any value. This is used for readdir | |
560 | * handling. Otherwise, check to see if the hash matches the | |
561 | * desired contiuation hash. If it doesn't, return since | |
562 | * there's no point to read in the successive index pages. | |
563 | */ | |
564 | bhash = dx_get_hash(p->at); | |
565 | if (start_hash) | |
566 | *start_hash = bhash; | |
567 | if ((hash & 1) == 0) { | |
568 | if ((bhash & ~1) != hash) | |
569 | return 0; | |
570 | } | |
571 | /* | |
572 | * If the hash is HASH_NB_ALWAYS, we always go to the next | |
573 | * block so no check is necessary | |
574 | */ | |
575 | while (num_frames--) { | |
576 | if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at), | |
577 | 0, &err))) | |
578 | return err; /* Failure */ | |
579 | p++; | |
580 | brelse(p->bh); | |
581 | p->bh = bh; | |
582 | p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; | |
583 | } | |
584 | return 1; | |
585 | } | |
586 | ||
587 | ||
588 | /* | |
589 | * This function fills a red-black tree with information from a | |
590 | * directory block. It returns the number directory entries loaded | |
591 | * into the tree. If there is an error it is returned in err. | |
592 | */ | |
593 | static int htree_dirblock_to_tree(struct file *dir_file, | |
594 | struct inode *dir, ext4_lblk_t block, | |
595 | struct dx_hash_info *hinfo, | |
596 | __u32 start_hash, __u32 start_minor_hash) | |
597 | { | |
598 | struct buffer_head *bh; | |
599 | struct ext4_dir_entry_2 *de, *top; | |
600 | int err, count = 0; | |
601 | ||
602 | dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", | |
603 | (unsigned long)block)); | |
604 | if (!(bh = ext4_bread (NULL, dir, block, 0, &err))) | |
605 | return err; | |
606 | ||
607 | de = (struct ext4_dir_entry_2 *) bh->b_data; | |
608 | top = (struct ext4_dir_entry_2 *) ((char *) de + | |
609 | dir->i_sb->s_blocksize - | |
610 | EXT4_DIR_REC_LEN(0)); | |
611 | for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { | |
612 | if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh, | |
613 | (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) | |
614 | +((char *)de - bh->b_data))) { | |
615 | /* On error, skip the f_pos to the next block. */ | |
616 | dir_file->f_pos = (dir_file->f_pos | | |
617 | (dir->i_sb->s_blocksize - 1)) + 1; | |
618 | brelse(bh); | |
619 | return count; | |
620 | } | |
621 | ext4fs_dirhash(de->name, de->name_len, hinfo); | |
622 | if ((hinfo->hash < start_hash) || | |
623 | ((hinfo->hash == start_hash) && | |
624 | (hinfo->minor_hash < start_minor_hash))) | |
625 | continue; | |
626 | if (de->inode == 0) | |
627 | continue; | |
628 | if ((err = ext4_htree_store_dirent(dir_file, | |
629 | hinfo->hash, hinfo->minor_hash, de)) != 0) { | |
630 | brelse(bh); | |
631 | return err; | |
632 | } | |
633 | count++; | |
634 | } | |
635 | brelse(bh); | |
636 | return count; | |
637 | } | |
638 | ||
639 | ||
640 | /* | |
641 | * This function fills a red-black tree with information from a | |
642 | * directory. We start scanning the directory in hash order, starting | |
643 | * at start_hash and start_minor_hash. | |
644 | * | |
645 | * This function returns the number of entries inserted into the tree, | |
646 | * or a negative error code. | |
647 | */ | |
648 | int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, | |
649 | __u32 start_minor_hash, __u32 *next_hash) | |
650 | { | |
651 | struct dx_hash_info hinfo; | |
652 | struct ext4_dir_entry_2 *de; | |
653 | struct dx_frame frames[2], *frame; | |
654 | struct inode *dir; | |
655 | ext4_lblk_t block; | |
656 | int count = 0; | |
657 | int ret, err; | |
658 | __u32 hashval; | |
659 | ||
660 | dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", | |
661 | start_hash, start_minor_hash)); | |
662 | dir = dir_file->f_path.dentry->d_inode; | |
663 | if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) { | |
664 | hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; | |
665 | if (hinfo.hash_version <= DX_HASH_TEA) | |
666 | hinfo.hash_version += | |
667 | EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
668 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; | |
669 | count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, | |
670 | start_hash, start_minor_hash); | |
671 | *next_hash = ~0; | |
672 | return count; | |
673 | } | |
674 | hinfo.hash = start_hash; | |
675 | hinfo.minor_hash = 0; | |
676 | frame = dx_probe(NULL, dir, &hinfo, frames, &err); | |
677 | if (!frame) | |
678 | return err; | |
679 | ||
680 | /* Add '.' and '..' from the htree header */ | |
681 | if (!start_hash && !start_minor_hash) { | |
682 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; | |
683 | if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0) | |
684 | goto errout; | |
685 | count++; | |
686 | } | |
687 | if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { | |
688 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; | |
689 | de = ext4_next_entry(de, dir->i_sb->s_blocksize); | |
690 | if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0) | |
691 | goto errout; | |
692 | count++; | |
693 | } | |
694 | ||
695 | while (1) { | |
696 | block = dx_get_block(frame->at); | |
697 | ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, | |
698 | start_hash, start_minor_hash); | |
699 | if (ret < 0) { | |
700 | err = ret; | |
701 | goto errout; | |
702 | } | |
703 | count += ret; | |
704 | hashval = ~0; | |
705 | ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, | |
706 | frame, frames, &hashval); | |
707 | *next_hash = hashval; | |
708 | if (ret < 0) { | |
709 | err = ret; | |
710 | goto errout; | |
711 | } | |
712 | /* | |
713 | * Stop if: (a) there are no more entries, or | |
714 | * (b) we have inserted at least one entry and the | |
715 | * next hash value is not a continuation | |
716 | */ | |
717 | if ((ret == 0) || | |
718 | (count && ((hashval & 1) == 0))) | |
719 | break; | |
720 | } | |
721 | dx_release(frames); | |
722 | dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " | |
723 | "next hash: %x\n", count, *next_hash)); | |
724 | return count; | |
725 | errout: | |
726 | dx_release(frames); | |
727 | return (err); | |
728 | } | |
729 | ||
730 | ||
731 | /* | |
732 | * Directory block splitting, compacting | |
733 | */ | |
734 | ||
735 | /* | |
736 | * Create map of hash values, offsets, and sizes, stored at end of block. | |
737 | * Returns number of entries mapped. | |
738 | */ | |
739 | static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize, | |
740 | struct dx_hash_info *hinfo, | |
741 | struct dx_map_entry *map_tail) | |
742 | { | |
743 | int count = 0; | |
744 | char *base = (char *) de; | |
745 | struct dx_hash_info h = *hinfo; | |
746 | ||
747 | while ((char *) de < base + blocksize) { | |
748 | if (de->name_len && de->inode) { | |
749 | ext4fs_dirhash(de->name, de->name_len, &h); | |
750 | map_tail--; | |
751 | map_tail->hash = h.hash; | |
752 | map_tail->offs = ((char *) de - base)>>2; | |
753 | map_tail->size = le16_to_cpu(de->rec_len); | |
754 | count++; | |
755 | cond_resched(); | |
756 | } | |
757 | /* XXX: do we need to check rec_len == 0 case? -Chris */ | |
758 | de = ext4_next_entry(de, blocksize); | |
759 | } | |
760 | return count; | |
761 | } | |
762 | ||
763 | /* Sort map by hash value */ | |
764 | static void dx_sort_map (struct dx_map_entry *map, unsigned count) | |
765 | { | |
766 | struct dx_map_entry *p, *q, *top = map + count - 1; | |
767 | int more; | |
768 | /* Combsort until bubble sort doesn't suck */ | |
769 | while (count > 2) { | |
770 | count = count*10/13; | |
771 | if (count - 9 < 2) /* 9, 10 -> 11 */ | |
772 | count = 11; | |
773 | for (p = top, q = p - count; q >= map; p--, q--) | |
774 | if (p->hash < q->hash) | |
775 | swap(*p, *q); | |
776 | } | |
777 | /* Garden variety bubble sort */ | |
778 | do { | |
779 | more = 0; | |
780 | q = top; | |
781 | while (q-- > map) { | |
782 | if (q[1].hash >= q[0].hash) | |
783 | continue; | |
784 | swap(*(q+1), *q); | |
785 | more = 1; | |
786 | } | |
787 | } while(more); | |
788 | } | |
789 | ||
790 | static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) | |
791 | { | |
792 | struct dx_entry *entries = frame->entries; | |
793 | struct dx_entry *old = frame->at, *new = old + 1; | |
794 | int count = dx_get_count(entries); | |
795 | ||
796 | assert(count < dx_get_limit(entries)); | |
797 | assert(old < entries + count); | |
798 | memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); | |
799 | dx_set_hash(new, hash); | |
800 | dx_set_block(new, block); | |
801 | dx_set_count(entries, count + 1); | |
802 | } | |
803 | ||
804 | static void ext4_update_dx_flag(struct inode *inode) | |
805 | { | |
806 | if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb, | |
807 | EXT4_FEATURE_COMPAT_DIR_INDEX)) | |
808 | EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL; | |
809 | } | |
810 | ||
811 | /* | |
812 | * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure. | |
813 | * | |
814 | * `len <= EXT4_NAME_LEN' is guaranteed by caller. | |
815 | * `de != NULL' is guaranteed by caller. | |
816 | */ | |
817 | static inline int ext4_match (int len, const char * const name, | |
818 | struct ext4_dir_entry_2 * de) | |
819 | { | |
820 | if (len != de->name_len) | |
821 | return 0; | |
822 | if (!de->inode) | |
823 | return 0; | |
824 | return !memcmp(name, de->name, len); | |
825 | } | |
826 | ||
827 | /* | |
828 | * Returns 0 if not found, -1 on failure, and 1 on success | |
829 | */ | |
830 | static inline int search_dirblock(struct buffer_head *bh, | |
831 | struct inode *dir, | |
832 | const struct qstr *d_name, | |
833 | unsigned int offset, | |
834 | struct ext4_dir_entry_2 ** res_dir) | |
835 | { | |
836 | struct ext4_dir_entry_2 * de; | |
837 | char * dlimit; | |
838 | int de_len; | |
839 | const char *name = d_name->name; | |
840 | int namelen = d_name->len; | |
841 | ||
842 | de = (struct ext4_dir_entry_2 *) bh->b_data; | |
843 | dlimit = bh->b_data + dir->i_sb->s_blocksize; | |
844 | while ((char *) de < dlimit) { | |
845 | /* this code is executed quadratically often */ | |
846 | /* do minimal checking `by hand' */ | |
847 | ||
848 | if ((char *) de + namelen <= dlimit && | |
849 | ext4_match (namelen, name, de)) { | |
850 | /* found a match - just to be sure, do a full check */ | |
851 | if (!ext4_check_dir_entry("ext4_find_entry", | |
852 | dir, de, bh, offset)) | |
853 | return -1; | |
854 | *res_dir = de; | |
855 | return 1; | |
856 | } | |
857 | /* prevent looping on a bad block */ | |
858 | de_len = ext4_rec_len_from_disk(de->rec_len, | |
859 | dir->i_sb->s_blocksize); | |
860 | if (de_len <= 0) | |
861 | return -1; | |
862 | offset += de_len; | |
863 | de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); | |
864 | } | |
865 | return 0; | |
866 | } | |
867 | ||
868 | ||
869 | /* | |
870 | * ext4_find_entry() | |
871 | * | |
872 | * finds an entry in the specified directory with the wanted name. It | |
873 | * returns the cache buffer in which the entry was found, and the entry | |
874 | * itself (as a parameter - res_dir). It does NOT read the inode of the | |
875 | * entry - you'll have to do that yourself if you want to. | |
876 | * | |
877 | * The returned buffer_head has ->b_count elevated. The caller is expected | |
878 | * to brelse() it when appropriate. | |
879 | */ | |
880 | static struct buffer_head * ext4_find_entry (struct inode *dir, | |
881 | const struct qstr *d_name, | |
882 | struct ext4_dir_entry_2 ** res_dir) | |
883 | { | |
884 | struct super_block *sb; | |
885 | struct buffer_head *bh_use[NAMEI_RA_SIZE]; | |
886 | struct buffer_head *bh, *ret = NULL; | |
887 | ext4_lblk_t start, block, b; | |
888 | int ra_max = 0; /* Number of bh's in the readahead | |
889 | buffer, bh_use[] */ | |
890 | int ra_ptr = 0; /* Current index into readahead | |
891 | buffer */ | |
892 | int num = 0; | |
893 | ext4_lblk_t nblocks; | |
894 | int i, err; | |
895 | int namelen; | |
896 | ||
897 | *res_dir = NULL; | |
898 | sb = dir->i_sb; | |
899 | namelen = d_name->len; | |
900 | if (namelen > EXT4_NAME_LEN) | |
901 | return NULL; | |
902 | if (is_dx(dir)) { | |
903 | bh = ext4_dx_find_entry(dir, d_name, res_dir, &err); | |
904 | /* | |
905 | * On success, or if the error was file not found, | |
906 | * return. Otherwise, fall back to doing a search the | |
907 | * old fashioned way. | |
908 | */ | |
909 | if (bh || (err != ERR_BAD_DX_DIR)) | |
910 | return bh; | |
911 | dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " | |
912 | "falling back\n")); | |
913 | } | |
914 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); | |
915 | start = EXT4_I(dir)->i_dir_start_lookup; | |
916 | if (start >= nblocks) | |
917 | start = 0; | |
918 | block = start; | |
919 | restart: | |
920 | do { | |
921 | /* | |
922 | * We deal with the read-ahead logic here. | |
923 | */ | |
924 | if (ra_ptr >= ra_max) { | |
925 | /* Refill the readahead buffer */ | |
926 | ra_ptr = 0; | |
927 | b = block; | |
928 | for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { | |
929 | /* | |
930 | * Terminate if we reach the end of the | |
931 | * directory and must wrap, or if our | |
932 | * search has finished at this block. | |
933 | */ | |
934 | if (b >= nblocks || (num && block == start)) { | |
935 | bh_use[ra_max] = NULL; | |
936 | break; | |
937 | } | |
938 | num++; | |
939 | bh = ext4_getblk(NULL, dir, b++, 0, &err); | |
940 | bh_use[ra_max] = bh; | |
941 | if (bh) | |
942 | ll_rw_block(READ_META, 1, &bh); | |
943 | } | |
944 | } | |
945 | if ((bh = bh_use[ra_ptr++]) == NULL) | |
946 | goto next; | |
947 | wait_on_buffer(bh); | |
948 | if (!buffer_uptodate(bh)) { | |
949 | /* read error, skip block & hope for the best */ | |
950 | ext4_error(sb, __func__, "reading directory #%lu " | |
951 | "offset %lu", dir->i_ino, | |
952 | (unsigned long)block); | |
953 | brelse(bh); | |
954 | goto next; | |
955 | } | |
956 | i = search_dirblock(bh, dir, d_name, | |
957 | block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); | |
958 | if (i == 1) { | |
959 | EXT4_I(dir)->i_dir_start_lookup = block; | |
960 | ret = bh; | |
961 | goto cleanup_and_exit; | |
962 | } else { | |
963 | brelse(bh); | |
964 | if (i < 0) | |
965 | goto cleanup_and_exit; | |
966 | } | |
967 | next: | |
968 | if (++block >= nblocks) | |
969 | block = 0; | |
970 | } while (block != start); | |
971 | ||
972 | /* | |
973 | * If the directory has grown while we were searching, then | |
974 | * search the last part of the directory before giving up. | |
975 | */ | |
976 | block = nblocks; | |
977 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); | |
978 | if (block < nblocks) { | |
979 | start = 0; | |
980 | goto restart; | |
981 | } | |
982 | ||
983 | cleanup_and_exit: | |
984 | /* Clean up the read-ahead blocks */ | |
985 | for (; ra_ptr < ra_max; ra_ptr++) | |
986 | brelse(bh_use[ra_ptr]); | |
987 | return ret; | |
988 | } | |
989 | ||
990 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name, | |
991 | struct ext4_dir_entry_2 **res_dir, int *err) | |
992 | { | |
993 | struct super_block * sb; | |
994 | struct dx_hash_info hinfo; | |
995 | u32 hash; | |
996 | struct dx_frame frames[2], *frame; | |
997 | struct ext4_dir_entry_2 *de, *top; | |
998 | struct buffer_head *bh; | |
999 | ext4_lblk_t block; | |
1000 | int retval; | |
1001 | int namelen = d_name->len; | |
1002 | const u8 *name = d_name->name; | |
1003 | ||
1004 | sb = dir->i_sb; | |
1005 | /* NFS may look up ".." - look at dx_root directory block */ | |
1006 | if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){ | |
1007 | if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err))) | |
1008 | return NULL; | |
1009 | } else { | |
1010 | frame = frames; | |
1011 | frame->bh = NULL; /* for dx_release() */ | |
1012 | frame->at = (struct dx_entry *)frames; /* hack for zero entry*/ | |
1013 | dx_set_block(frame->at, 0); /* dx_root block is 0 */ | |
1014 | } | |
1015 | hash = hinfo.hash; | |
1016 | do { | |
1017 | block = dx_get_block(frame->at); | |
1018 | if (!(bh = ext4_bread (NULL,dir, block, 0, err))) | |
1019 | goto errout; | |
1020 | de = (struct ext4_dir_entry_2 *) bh->b_data; | |
1021 | top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize - | |
1022 | EXT4_DIR_REC_LEN(0)); | |
1023 | for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) { | |
1024 | int off = (block << EXT4_BLOCK_SIZE_BITS(sb)) | |
1025 | + ((char *) de - bh->b_data); | |
1026 | ||
1027 | if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) { | |
1028 | brelse(bh); | |
1029 | *err = ERR_BAD_DX_DIR; | |
1030 | goto errout; | |
1031 | } | |
1032 | ||
1033 | if (ext4_match(namelen, name, de)) { | |
1034 | *res_dir = de; | |
1035 | dx_release(frames); | |
1036 | return bh; | |
1037 | } | |
1038 | } | |
1039 | brelse(bh); | |
1040 | /* Check to see if we should continue to search */ | |
1041 | retval = ext4_htree_next_block(dir, hash, frame, | |
1042 | frames, NULL); | |
1043 | if (retval < 0) { | |
1044 | ext4_warning(sb, __func__, | |
1045 | "error reading index page in directory #%lu", | |
1046 | dir->i_ino); | |
1047 | *err = retval; | |
1048 | goto errout; | |
1049 | } | |
1050 | } while (retval == 1); | |
1051 | ||
1052 | *err = -ENOENT; | |
1053 | errout: | |
1054 | dxtrace(printk(KERN_DEBUG "%s not found\n", name)); | |
1055 | dx_release (frames); | |
1056 | return NULL; | |
1057 | } | |
1058 | ||
1059 | static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) | |
1060 | { | |
1061 | struct inode *inode; | |
1062 | struct ext4_dir_entry_2 *de; | |
1063 | struct buffer_head *bh; | |
1064 | ||
1065 | if (dentry->d_name.len > EXT4_NAME_LEN) | |
1066 | return ERR_PTR(-ENAMETOOLONG); | |
1067 | ||
1068 | bh = ext4_find_entry(dir, &dentry->d_name, &de); | |
1069 | inode = NULL; | |
1070 | if (bh) { | |
1071 | __u32 ino = le32_to_cpu(de->inode); | |
1072 | brelse(bh); | |
1073 | if (!ext4_valid_inum(dir->i_sb, ino)) { | |
1074 | ext4_error(dir->i_sb, "ext4_lookup", | |
1075 | "bad inode number: %u", ino); | |
1076 | return ERR_PTR(-EIO); | |
1077 | } | |
1078 | inode = ext4_iget(dir->i_sb, ino); | |
1079 | if (unlikely(IS_ERR(inode))) { | |
1080 | if (PTR_ERR(inode) == -ESTALE) { | |
1081 | ext4_error(dir->i_sb, __func__, | |
1082 | "deleted inode referenced: %u", | |
1083 | ino); | |
1084 | return ERR_PTR(-EIO); | |
1085 | } else { | |
1086 | return ERR_CAST(inode); | |
1087 | } | |
1088 | } | |
1089 | } | |
1090 | return d_splice_alias(inode, dentry); | |
1091 | } | |
1092 | ||
1093 | ||
1094 | struct dentry *ext4_get_parent(struct dentry *child) | |
1095 | { | |
1096 | __u32 ino; | |
1097 | struct inode *inode; | |
1098 | static const struct qstr dotdot = { | |
1099 | .name = "..", | |
1100 | .len = 2, | |
1101 | }; | |
1102 | struct ext4_dir_entry_2 * de; | |
1103 | struct buffer_head *bh; | |
1104 | ||
1105 | bh = ext4_find_entry(child->d_inode, &dotdot, &de); | |
1106 | inode = NULL; | |
1107 | if (!bh) | |
1108 | return ERR_PTR(-ENOENT); | |
1109 | ino = le32_to_cpu(de->inode); | |
1110 | brelse(bh); | |
1111 | ||
1112 | if (!ext4_valid_inum(child->d_inode->i_sb, ino)) { | |
1113 | ext4_error(child->d_inode->i_sb, "ext4_get_parent", | |
1114 | "bad inode number: %u", ino); | |
1115 | return ERR_PTR(-EIO); | |
1116 | } | |
1117 | ||
1118 | return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino)); | |
1119 | } | |
1120 | ||
1121 | #define S_SHIFT 12 | |
1122 | static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = { | |
1123 | [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE, | |
1124 | [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR, | |
1125 | [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV, | |
1126 | [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV, | |
1127 | [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO, | |
1128 | [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK, | |
1129 | [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK, | |
1130 | }; | |
1131 | ||
1132 | static inline void ext4_set_de_type(struct super_block *sb, | |
1133 | struct ext4_dir_entry_2 *de, | |
1134 | umode_t mode) { | |
1135 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE)) | |
1136 | de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT]; | |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * Move count entries from end of map between two memory locations. | |
1141 | * Returns pointer to last entry moved. | |
1142 | */ | |
1143 | static struct ext4_dir_entry_2 * | |
1144 | dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count, | |
1145 | unsigned blocksize) | |
1146 | { | |
1147 | unsigned rec_len = 0; | |
1148 | ||
1149 | while (count--) { | |
1150 | struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) | |
1151 | (from + (map->offs<<2)); | |
1152 | rec_len = EXT4_DIR_REC_LEN(de->name_len); | |
1153 | memcpy (to, de, rec_len); | |
1154 | ((struct ext4_dir_entry_2 *) to)->rec_len = | |
1155 | ext4_rec_len_to_disk(rec_len, blocksize); | |
1156 | de->inode = 0; | |
1157 | map++; | |
1158 | to += rec_len; | |
1159 | } | |
1160 | return (struct ext4_dir_entry_2 *) (to - rec_len); | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * Compact each dir entry in the range to the minimal rec_len. | |
1165 | * Returns pointer to last entry in range. | |
1166 | */ | |
1167 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize) | |
1168 | { | |
1169 | struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; | |
1170 | unsigned rec_len = 0; | |
1171 | ||
1172 | prev = to = de; | |
1173 | while ((char*)de < base + blocksize) { | |
1174 | next = ext4_next_entry(de, blocksize); | |
1175 | if (de->inode && de->name_len) { | |
1176 | rec_len = EXT4_DIR_REC_LEN(de->name_len); | |
1177 | if (de > to) | |
1178 | memmove(to, de, rec_len); | |
1179 | to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); | |
1180 | prev = to; | |
1181 | to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); | |
1182 | } | |
1183 | de = next; | |
1184 | } | |
1185 | return prev; | |
1186 | } | |
1187 | ||
1188 | /* | |
1189 | * Split a full leaf block to make room for a new dir entry. | |
1190 | * Allocate a new block, and move entries so that they are approx. equally full. | |
1191 | * Returns pointer to de in block into which the new entry will be inserted. | |
1192 | */ | |
1193 | static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, | |
1194 | struct buffer_head **bh,struct dx_frame *frame, | |
1195 | struct dx_hash_info *hinfo, int *error) | |
1196 | { | |
1197 | unsigned blocksize = dir->i_sb->s_blocksize; | |
1198 | unsigned count, continued; | |
1199 | struct buffer_head *bh2; | |
1200 | ext4_lblk_t newblock; | |
1201 | u32 hash2; | |
1202 | struct dx_map_entry *map; | |
1203 | char *data1 = (*bh)->b_data, *data2; | |
1204 | unsigned split, move, size; | |
1205 | struct ext4_dir_entry_2 *de = NULL, *de2; | |
1206 | int err = 0, i; | |
1207 | ||
1208 | bh2 = ext4_append (handle, dir, &newblock, &err); | |
1209 | if (!(bh2)) { | |
1210 | brelse(*bh); | |
1211 | *bh = NULL; | |
1212 | goto errout; | |
1213 | } | |
1214 | ||
1215 | BUFFER_TRACE(*bh, "get_write_access"); | |
1216 | err = ext4_journal_get_write_access(handle, *bh); | |
1217 | if (err) | |
1218 | goto journal_error; | |
1219 | ||
1220 | BUFFER_TRACE(frame->bh, "get_write_access"); | |
1221 | err = ext4_journal_get_write_access(handle, frame->bh); | |
1222 | if (err) | |
1223 | goto journal_error; | |
1224 | ||
1225 | data2 = bh2->b_data; | |
1226 | ||
1227 | /* create map in the end of data2 block */ | |
1228 | map = (struct dx_map_entry *) (data2 + blocksize); | |
1229 | count = dx_make_map((struct ext4_dir_entry_2 *) data1, | |
1230 | blocksize, hinfo, map); | |
1231 | map -= count; | |
1232 | dx_sort_map(map, count); | |
1233 | /* Split the existing block in the middle, size-wise */ | |
1234 | size = 0; | |
1235 | move = 0; | |
1236 | for (i = count-1; i >= 0; i--) { | |
1237 | /* is more than half of this entry in 2nd half of the block? */ | |
1238 | if (size + map[i].size/2 > blocksize/2) | |
1239 | break; | |
1240 | size += map[i].size; | |
1241 | move++; | |
1242 | } | |
1243 | /* map index at which we will split */ | |
1244 | split = count - move; | |
1245 | hash2 = map[split].hash; | |
1246 | continued = hash2 == map[split - 1].hash; | |
1247 | dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", | |
1248 | (unsigned long)dx_get_block(frame->at), | |
1249 | hash2, split, count-split)); | |
1250 | ||
1251 | /* Fancy dance to stay within two buffers */ | |
1252 | de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize); | |
1253 | de = dx_pack_dirents(data1, blocksize); | |
1254 | de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de, | |
1255 | blocksize); | |
1256 | de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2, | |
1257 | blocksize); | |
1258 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1)); | |
1259 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1)); | |
1260 | ||
1261 | /* Which block gets the new entry? */ | |
1262 | if (hinfo->hash >= hash2) | |
1263 | { | |
1264 | swap(*bh, bh2); | |
1265 | de = de2; | |
1266 | } | |
1267 | dx_insert_block(frame, hash2 + continued, newblock); | |
1268 | err = ext4_handle_dirty_metadata(handle, dir, bh2); | |
1269 | if (err) | |
1270 | goto journal_error; | |
1271 | err = ext4_handle_dirty_metadata(handle, dir, frame->bh); | |
1272 | if (err) | |
1273 | goto journal_error; | |
1274 | brelse(bh2); | |
1275 | dxtrace(dx_show_index("frame", frame->entries)); | |
1276 | return de; | |
1277 | ||
1278 | journal_error: | |
1279 | brelse(*bh); | |
1280 | brelse(bh2); | |
1281 | *bh = NULL; | |
1282 | ext4_std_error(dir->i_sb, err); | |
1283 | errout: | |
1284 | *error = err; | |
1285 | return NULL; | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * Add a new entry into a directory (leaf) block. If de is non-NULL, | |
1290 | * it points to a directory entry which is guaranteed to be large | |
1291 | * enough for new directory entry. If de is NULL, then | |
1292 | * add_dirent_to_buf will attempt search the directory block for | |
1293 | * space. It will return -ENOSPC if no space is available, and -EIO | |
1294 | * and -EEXIST if directory entry already exists. | |
1295 | */ | |
1296 | static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry, | |
1297 | struct inode *inode, struct ext4_dir_entry_2 *de, | |
1298 | struct buffer_head *bh) | |
1299 | { | |
1300 | struct inode *dir = dentry->d_parent->d_inode; | |
1301 | const char *name = dentry->d_name.name; | |
1302 | int namelen = dentry->d_name.len; | |
1303 | unsigned int offset = 0; | |
1304 | unsigned int blocksize = dir->i_sb->s_blocksize; | |
1305 | unsigned short reclen; | |
1306 | int nlen, rlen, err; | |
1307 | char *top; | |
1308 | ||
1309 | reclen = EXT4_DIR_REC_LEN(namelen); | |
1310 | if (!de) { | |
1311 | de = (struct ext4_dir_entry_2 *)bh->b_data; | |
1312 | top = bh->b_data + blocksize - reclen; | |
1313 | while ((char *) de <= top) { | |
1314 | if (!ext4_check_dir_entry("ext4_add_entry", dir, de, | |
1315 | bh, offset)) | |
1316 | return -EIO; | |
1317 | if (ext4_match(namelen, name, de)) | |
1318 | return -EEXIST; | |
1319 | nlen = EXT4_DIR_REC_LEN(de->name_len); | |
1320 | rlen = ext4_rec_len_from_disk(de->rec_len, blocksize); | |
1321 | if ((de->inode? rlen - nlen: rlen) >= reclen) | |
1322 | break; | |
1323 | de = (struct ext4_dir_entry_2 *)((char *)de + rlen); | |
1324 | offset += rlen; | |
1325 | } | |
1326 | if ((char *) de > top) | |
1327 | return -ENOSPC; | |
1328 | } | |
1329 | BUFFER_TRACE(bh, "get_write_access"); | |
1330 | err = ext4_journal_get_write_access(handle, bh); | |
1331 | if (err) { | |
1332 | ext4_std_error(dir->i_sb, err); | |
1333 | return err; | |
1334 | } | |
1335 | ||
1336 | /* By now the buffer is marked for journaling */ | |
1337 | nlen = EXT4_DIR_REC_LEN(de->name_len); | |
1338 | rlen = ext4_rec_len_from_disk(de->rec_len, blocksize); | |
1339 | if (de->inode) { | |
1340 | struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen); | |
1341 | de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize); | |
1342 | de->rec_len = ext4_rec_len_to_disk(nlen, blocksize); | |
1343 | de = de1; | |
1344 | } | |
1345 | de->file_type = EXT4_FT_UNKNOWN; | |
1346 | if (inode) { | |
1347 | de->inode = cpu_to_le32(inode->i_ino); | |
1348 | ext4_set_de_type(dir->i_sb, de, inode->i_mode); | |
1349 | } else | |
1350 | de->inode = 0; | |
1351 | de->name_len = namelen; | |
1352 | memcpy(de->name, name, namelen); | |
1353 | /* | |
1354 | * XXX shouldn't update any times until successful | |
1355 | * completion of syscall, but too many callers depend | |
1356 | * on this. | |
1357 | * | |
1358 | * XXX similarly, too many callers depend on | |
1359 | * ext4_new_inode() setting the times, but error | |
1360 | * recovery deletes the inode, so the worst that can | |
1361 | * happen is that the times are slightly out of date | |
1362 | * and/or different from the directory change time. | |
1363 | */ | |
1364 | dir->i_mtime = dir->i_ctime = ext4_current_time(dir); | |
1365 | ext4_update_dx_flag(dir); | |
1366 | dir->i_version++; | |
1367 | ext4_mark_inode_dirty(handle, dir); | |
1368 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
1369 | err = ext4_handle_dirty_metadata(handle, dir, bh); | |
1370 | if (err) | |
1371 | ext4_std_error(dir->i_sb, err); | |
1372 | return 0; | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * This converts a one block unindexed directory to a 3 block indexed | |
1377 | * directory, and adds the dentry to the indexed directory. | |
1378 | */ | |
1379 | static int make_indexed_dir(handle_t *handle, struct dentry *dentry, | |
1380 | struct inode *inode, struct buffer_head *bh) | |
1381 | { | |
1382 | struct inode *dir = dentry->d_parent->d_inode; | |
1383 | const char *name = dentry->d_name.name; | |
1384 | int namelen = dentry->d_name.len; | |
1385 | struct buffer_head *bh2; | |
1386 | struct dx_root *root; | |
1387 | struct dx_frame frames[2], *frame; | |
1388 | struct dx_entry *entries; | |
1389 | struct ext4_dir_entry_2 *de, *de2; | |
1390 | char *data1, *top; | |
1391 | unsigned len; | |
1392 | int retval; | |
1393 | unsigned blocksize; | |
1394 | struct dx_hash_info hinfo; | |
1395 | ext4_lblk_t block; | |
1396 | struct fake_dirent *fde; | |
1397 | ||
1398 | blocksize = dir->i_sb->s_blocksize; | |
1399 | dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino)); | |
1400 | retval = ext4_journal_get_write_access(handle, bh); | |
1401 | if (retval) { | |
1402 | ext4_std_error(dir->i_sb, retval); | |
1403 | brelse(bh); | |
1404 | return retval; | |
1405 | } | |
1406 | root = (struct dx_root *) bh->b_data; | |
1407 | ||
1408 | /* The 0th block becomes the root, move the dirents out */ | |
1409 | fde = &root->dotdot; | |
1410 | de = (struct ext4_dir_entry_2 *)((char *)fde + | |
1411 | ext4_rec_len_from_disk(fde->rec_len, blocksize)); | |
1412 | if ((char *) de >= (((char *) root) + blocksize)) { | |
1413 | ext4_error(dir->i_sb, __func__, | |
1414 | "invalid rec_len for '..' in inode %lu", | |
1415 | dir->i_ino); | |
1416 | brelse(bh); | |
1417 | return -EIO; | |
1418 | } | |
1419 | len = ((char *) root) + blocksize - (char *) de; | |
1420 | ||
1421 | /* Allocate new block for the 0th block's dirents */ | |
1422 | bh2 = ext4_append(handle, dir, &block, &retval); | |
1423 | if (!(bh2)) { | |
1424 | brelse(bh); | |
1425 | return retval; | |
1426 | } | |
1427 | EXT4_I(dir)->i_flags |= EXT4_INDEX_FL; | |
1428 | data1 = bh2->b_data; | |
1429 | ||
1430 | memcpy (data1, de, len); | |
1431 | de = (struct ext4_dir_entry_2 *) data1; | |
1432 | top = data1 + len; | |
1433 | while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) | |
1434 | de = de2; | |
1435 | de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de, | |
1436 | blocksize); | |
1437 | /* Initialize the root; the dot dirents already exist */ | |
1438 | de = (struct ext4_dir_entry_2 *) (&root->dotdot); | |
1439 | de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2), | |
1440 | blocksize); | |
1441 | memset (&root->info, 0, sizeof(root->info)); | |
1442 | root->info.info_length = sizeof(root->info); | |
1443 | root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; | |
1444 | entries = root->entries; | |
1445 | dx_set_block(entries, 1); | |
1446 | dx_set_count(entries, 1); | |
1447 | dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); | |
1448 | ||
1449 | /* Initialize as for dx_probe */ | |
1450 | hinfo.hash_version = root->info.hash_version; | |
1451 | if (hinfo.hash_version <= DX_HASH_TEA) | |
1452 | hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
1453 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; | |
1454 | ext4fs_dirhash(name, namelen, &hinfo); | |
1455 | frame = frames; | |
1456 | frame->entries = entries; | |
1457 | frame->at = entries; | |
1458 | frame->bh = bh; | |
1459 | bh = bh2; | |
1460 | de = do_split(handle,dir, &bh, frame, &hinfo, &retval); | |
1461 | dx_release (frames); | |
1462 | if (!(de)) | |
1463 | return retval; | |
1464 | ||
1465 | retval = add_dirent_to_buf(handle, dentry, inode, de, bh); | |
1466 | brelse(bh); | |
1467 | return retval; | |
1468 | } | |
1469 | ||
1470 | /* | |
1471 | * ext4_add_entry() | |
1472 | * | |
1473 | * adds a file entry to the specified directory, using the same | |
1474 | * semantics as ext4_find_entry(). It returns NULL if it failed. | |
1475 | * | |
1476 | * NOTE!! The inode part of 'de' is left at 0 - which means you | |
1477 | * may not sleep between calling this and putting something into | |
1478 | * the entry, as someone else might have used it while you slept. | |
1479 | */ | |
1480 | static int ext4_add_entry(handle_t *handle, struct dentry *dentry, | |
1481 | struct inode *inode) | |
1482 | { | |
1483 | struct inode *dir = dentry->d_parent->d_inode; | |
1484 | struct buffer_head *bh; | |
1485 | struct ext4_dir_entry_2 *de; | |
1486 | struct super_block *sb; | |
1487 | int retval; | |
1488 | int dx_fallback=0; | |
1489 | unsigned blocksize; | |
1490 | ext4_lblk_t block, blocks; | |
1491 | ||
1492 | sb = dir->i_sb; | |
1493 | blocksize = sb->s_blocksize; | |
1494 | if (!dentry->d_name.len) | |
1495 | return -EINVAL; | |
1496 | if (is_dx(dir)) { | |
1497 | retval = ext4_dx_add_entry(handle, dentry, inode); | |
1498 | if (!retval || (retval != ERR_BAD_DX_DIR)) | |
1499 | return retval; | |
1500 | EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL; | |
1501 | dx_fallback++; | |
1502 | ext4_mark_inode_dirty(handle, dir); | |
1503 | } | |
1504 | blocks = dir->i_size >> sb->s_blocksize_bits; | |
1505 | for (block = 0; block < blocks; block++) { | |
1506 | bh = ext4_bread(handle, dir, block, 0, &retval); | |
1507 | if(!bh) | |
1508 | return retval; | |
1509 | retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
1510 | if (retval != -ENOSPC) { | |
1511 | brelse(bh); | |
1512 | return retval; | |
1513 | } | |
1514 | ||
1515 | if (blocks == 1 && !dx_fallback && | |
1516 | EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) | |
1517 | return make_indexed_dir(handle, dentry, inode, bh); | |
1518 | brelse(bh); | |
1519 | } | |
1520 | bh = ext4_append(handle, dir, &block, &retval); | |
1521 | if (!bh) | |
1522 | return retval; | |
1523 | de = (struct ext4_dir_entry_2 *) bh->b_data; | |
1524 | de->inode = 0; | |
1525 | de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize); | |
1526 | retval = add_dirent_to_buf(handle, dentry, inode, de, bh); | |
1527 | brelse(bh); | |
1528 | return retval; | |
1529 | } | |
1530 | ||
1531 | /* | |
1532 | * Returns 0 for success, or a negative error value | |
1533 | */ | |
1534 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, | |
1535 | struct inode *inode) | |
1536 | { | |
1537 | struct dx_frame frames[2], *frame; | |
1538 | struct dx_entry *entries, *at; | |
1539 | struct dx_hash_info hinfo; | |
1540 | struct buffer_head *bh; | |
1541 | struct inode *dir = dentry->d_parent->d_inode; | |
1542 | struct super_block *sb = dir->i_sb; | |
1543 | struct ext4_dir_entry_2 *de; | |
1544 | int err; | |
1545 | ||
1546 | frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err); | |
1547 | if (!frame) | |
1548 | return err; | |
1549 | entries = frame->entries; | |
1550 | at = frame->at; | |
1551 | ||
1552 | if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err))) | |
1553 | goto cleanup; | |
1554 | ||
1555 | BUFFER_TRACE(bh, "get_write_access"); | |
1556 | err = ext4_journal_get_write_access(handle, bh); | |
1557 | if (err) | |
1558 | goto journal_error; | |
1559 | ||
1560 | err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
1561 | if (err != -ENOSPC) | |
1562 | goto cleanup; | |
1563 | ||
1564 | /* Block full, should compress but for now just split */ | |
1565 | dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", | |
1566 | dx_get_count(entries), dx_get_limit(entries))); | |
1567 | /* Need to split index? */ | |
1568 | if (dx_get_count(entries) == dx_get_limit(entries)) { | |
1569 | ext4_lblk_t newblock; | |
1570 | unsigned icount = dx_get_count(entries); | |
1571 | int levels = frame - frames; | |
1572 | struct dx_entry *entries2; | |
1573 | struct dx_node *node2; | |
1574 | struct buffer_head *bh2; | |
1575 | ||
1576 | if (levels && (dx_get_count(frames->entries) == | |
1577 | dx_get_limit(frames->entries))) { | |
1578 | ext4_warning(sb, __func__, | |
1579 | "Directory index full!"); | |
1580 | err = -ENOSPC; | |
1581 | goto cleanup; | |
1582 | } | |
1583 | bh2 = ext4_append (handle, dir, &newblock, &err); | |
1584 | if (!(bh2)) | |
1585 | goto cleanup; | |
1586 | node2 = (struct dx_node *)(bh2->b_data); | |
1587 | entries2 = node2->entries; | |
1588 | memset(&node2->fake, 0, sizeof(struct fake_dirent)); | |
1589 | node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize, | |
1590 | sb->s_blocksize); | |
1591 | BUFFER_TRACE(frame->bh, "get_write_access"); | |
1592 | err = ext4_journal_get_write_access(handle, frame->bh); | |
1593 | if (err) | |
1594 | goto journal_error; | |
1595 | if (levels) { | |
1596 | unsigned icount1 = icount/2, icount2 = icount - icount1; | |
1597 | unsigned hash2 = dx_get_hash(entries + icount1); | |
1598 | dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", | |
1599 | icount1, icount2)); | |
1600 | ||
1601 | BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ | |
1602 | err = ext4_journal_get_write_access(handle, | |
1603 | frames[0].bh); | |
1604 | if (err) | |
1605 | goto journal_error; | |
1606 | ||
1607 | memcpy((char *) entries2, (char *) (entries + icount1), | |
1608 | icount2 * sizeof(struct dx_entry)); | |
1609 | dx_set_count(entries, icount1); | |
1610 | dx_set_count(entries2, icount2); | |
1611 | dx_set_limit(entries2, dx_node_limit(dir)); | |
1612 | ||
1613 | /* Which index block gets the new entry? */ | |
1614 | if (at - entries >= icount1) { | |
1615 | frame->at = at = at - entries - icount1 + entries2; | |
1616 | frame->entries = entries = entries2; | |
1617 | swap(frame->bh, bh2); | |
1618 | } | |
1619 | dx_insert_block(frames + 0, hash2, newblock); | |
1620 | dxtrace(dx_show_index("node", frames[1].entries)); | |
1621 | dxtrace(dx_show_index("node", | |
1622 | ((struct dx_node *) bh2->b_data)->entries)); | |
1623 | err = ext4_handle_dirty_metadata(handle, inode, bh2); | |
1624 | if (err) | |
1625 | goto journal_error; | |
1626 | brelse (bh2); | |
1627 | } else { | |
1628 | dxtrace(printk(KERN_DEBUG | |
1629 | "Creating second level index...\n")); | |
1630 | memcpy((char *) entries2, (char *) entries, | |
1631 | icount * sizeof(struct dx_entry)); | |
1632 | dx_set_limit(entries2, dx_node_limit(dir)); | |
1633 | ||
1634 | /* Set up root */ | |
1635 | dx_set_count(entries, 1); | |
1636 | dx_set_block(entries + 0, newblock); | |
1637 | ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; | |
1638 | ||
1639 | /* Add new access path frame */ | |
1640 | frame = frames + 1; | |
1641 | frame->at = at = at - entries + entries2; | |
1642 | frame->entries = entries = entries2; | |
1643 | frame->bh = bh2; | |
1644 | err = ext4_journal_get_write_access(handle, | |
1645 | frame->bh); | |
1646 | if (err) | |
1647 | goto journal_error; | |
1648 | } | |
1649 | ext4_handle_dirty_metadata(handle, inode, frames[0].bh); | |
1650 | } | |
1651 | de = do_split(handle, dir, &bh, frame, &hinfo, &err); | |
1652 | if (!de) | |
1653 | goto cleanup; | |
1654 | err = add_dirent_to_buf(handle, dentry, inode, de, bh); | |
1655 | goto cleanup; | |
1656 | ||
1657 | journal_error: | |
1658 | ext4_std_error(dir->i_sb, err); | |
1659 | cleanup: | |
1660 | if (bh) | |
1661 | brelse(bh); | |
1662 | dx_release(frames); | |
1663 | return err; | |
1664 | } | |
1665 | ||
1666 | /* | |
1667 | * ext4_delete_entry deletes a directory entry by merging it with the | |
1668 | * previous entry | |
1669 | */ | |
1670 | static int ext4_delete_entry(handle_t *handle, | |
1671 | struct inode *dir, | |
1672 | struct ext4_dir_entry_2 *de_del, | |
1673 | struct buffer_head *bh) | |
1674 | { | |
1675 | struct ext4_dir_entry_2 *de, *pde; | |
1676 | unsigned int blocksize = dir->i_sb->s_blocksize; | |
1677 | int i; | |
1678 | ||
1679 | i = 0; | |
1680 | pde = NULL; | |
1681 | de = (struct ext4_dir_entry_2 *) bh->b_data; | |
1682 | while (i < bh->b_size) { | |
1683 | if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i)) | |
1684 | return -EIO; | |
1685 | if (de == de_del) { | |
1686 | BUFFER_TRACE(bh, "get_write_access"); | |
1687 | ext4_journal_get_write_access(handle, bh); | |
1688 | if (pde) | |
1689 | pde->rec_len = ext4_rec_len_to_disk( | |
1690 | ext4_rec_len_from_disk(pde->rec_len, | |
1691 | blocksize) + | |
1692 | ext4_rec_len_from_disk(de->rec_len, | |
1693 | blocksize), | |
1694 | blocksize); | |
1695 | else | |
1696 | de->inode = 0; | |
1697 | dir->i_version++; | |
1698 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
1699 | ext4_handle_dirty_metadata(handle, dir, bh); | |
1700 | return 0; | |
1701 | } | |
1702 | i += ext4_rec_len_from_disk(de->rec_len, blocksize); | |
1703 | pde = de; | |
1704 | de = ext4_next_entry(de, blocksize); | |
1705 | } | |
1706 | return -ENOENT; | |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2, | |
1711 | * since this indicates that nlinks count was previously 1. | |
1712 | */ | |
1713 | static void ext4_inc_count(handle_t *handle, struct inode *inode) | |
1714 | { | |
1715 | inc_nlink(inode); | |
1716 | if (is_dx(inode) && inode->i_nlink > 1) { | |
1717 | /* limit is 16-bit i_links_count */ | |
1718 | if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) { | |
1719 | inode->i_nlink = 1; | |
1720 | EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb, | |
1721 | EXT4_FEATURE_RO_COMPAT_DIR_NLINK); | |
1722 | } | |
1723 | } | |
1724 | } | |
1725 | ||
1726 | /* | |
1727 | * If a directory had nlink == 1, then we should let it be 1. This indicates | |
1728 | * directory has >EXT4_LINK_MAX subdirs. | |
1729 | */ | |
1730 | static void ext4_dec_count(handle_t *handle, struct inode *inode) | |
1731 | { | |
1732 | drop_nlink(inode); | |
1733 | if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0) | |
1734 | inc_nlink(inode); | |
1735 | } | |
1736 | ||
1737 | ||
1738 | static int ext4_add_nondir(handle_t *handle, | |
1739 | struct dentry *dentry, struct inode *inode) | |
1740 | { | |
1741 | int err = ext4_add_entry(handle, dentry, inode); | |
1742 | if (!err) { | |
1743 | ext4_mark_inode_dirty(handle, inode); | |
1744 | d_instantiate(dentry, inode); | |
1745 | unlock_new_inode(inode); | |
1746 | return 0; | |
1747 | } | |
1748 | drop_nlink(inode); | |
1749 | unlock_new_inode(inode); | |
1750 | iput(inode); | |
1751 | return err; | |
1752 | } | |
1753 | ||
1754 | /* | |
1755 | * By the time this is called, we already have created | |
1756 | * the directory cache entry for the new file, but it | |
1757 | * is so far negative - it has no inode. | |
1758 | * | |
1759 | * If the create succeeds, we fill in the inode information | |
1760 | * with d_instantiate(). | |
1761 | */ | |
1762 | static int ext4_create(struct inode *dir, struct dentry *dentry, int mode, | |
1763 | struct nameidata *nd) | |
1764 | { | |
1765 | handle_t *handle; | |
1766 | struct inode *inode; | |
1767 | int err, retries = 0; | |
1768 | ||
1769 | retry: | |
1770 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
1771 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
1772 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); | |
1773 | if (IS_ERR(handle)) | |
1774 | return PTR_ERR(handle); | |
1775 | ||
1776 | if (IS_DIRSYNC(dir)) | |
1777 | ext4_handle_sync(handle); | |
1778 | ||
1779 | inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0); | |
1780 | err = PTR_ERR(inode); | |
1781 | if (!IS_ERR(inode)) { | |
1782 | inode->i_op = &ext4_file_inode_operations; | |
1783 | inode->i_fop = &ext4_file_operations; | |
1784 | ext4_set_aops(inode); | |
1785 | err = ext4_add_nondir(handle, dentry, inode); | |
1786 | } | |
1787 | ext4_journal_stop(handle); | |
1788 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
1789 | goto retry; | |
1790 | return err; | |
1791 | } | |
1792 | ||
1793 | static int ext4_mknod(struct inode *dir, struct dentry *dentry, | |
1794 | int mode, dev_t rdev) | |
1795 | { | |
1796 | handle_t *handle; | |
1797 | struct inode *inode; | |
1798 | int err, retries = 0; | |
1799 | ||
1800 | if (!new_valid_dev(rdev)) | |
1801 | return -EINVAL; | |
1802 | ||
1803 | retry: | |
1804 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
1805 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
1806 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); | |
1807 | if (IS_ERR(handle)) | |
1808 | return PTR_ERR(handle); | |
1809 | ||
1810 | if (IS_DIRSYNC(dir)) | |
1811 | ext4_handle_sync(handle); | |
1812 | ||
1813 | inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0); | |
1814 | err = PTR_ERR(inode); | |
1815 | if (!IS_ERR(inode)) { | |
1816 | init_special_inode(inode, inode->i_mode, rdev); | |
1817 | #ifdef CONFIG_EXT4_FS_XATTR | |
1818 | inode->i_op = &ext4_special_inode_operations; | |
1819 | #endif | |
1820 | err = ext4_add_nondir(handle, dentry, inode); | |
1821 | } | |
1822 | ext4_journal_stop(handle); | |
1823 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
1824 | goto retry; | |
1825 | return err; | |
1826 | } | |
1827 | ||
1828 | static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1829 | { | |
1830 | handle_t *handle; | |
1831 | struct inode *inode; | |
1832 | struct buffer_head *dir_block; | |
1833 | struct ext4_dir_entry_2 *de; | |
1834 | unsigned int blocksize = dir->i_sb->s_blocksize; | |
1835 | int err, retries = 0; | |
1836 | ||
1837 | if (EXT4_DIR_LINK_MAX(dir)) | |
1838 | return -EMLINK; | |
1839 | ||
1840 | retry: | |
1841 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
1842 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
1843 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); | |
1844 | if (IS_ERR(handle)) | |
1845 | return PTR_ERR(handle); | |
1846 | ||
1847 | if (IS_DIRSYNC(dir)) | |
1848 | ext4_handle_sync(handle); | |
1849 | ||
1850 | inode = ext4_new_inode(handle, dir, S_IFDIR | mode, | |
1851 | &dentry->d_name, 0); | |
1852 | err = PTR_ERR(inode); | |
1853 | if (IS_ERR(inode)) | |
1854 | goto out_stop; | |
1855 | ||
1856 | inode->i_op = &ext4_dir_inode_operations; | |
1857 | inode->i_fop = &ext4_dir_operations; | |
1858 | inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize; | |
1859 | dir_block = ext4_bread(handle, inode, 0, 1, &err); | |
1860 | if (!dir_block) | |
1861 | goto out_clear_inode; | |
1862 | BUFFER_TRACE(dir_block, "get_write_access"); | |
1863 | ext4_journal_get_write_access(handle, dir_block); | |
1864 | de = (struct ext4_dir_entry_2 *) dir_block->b_data; | |
1865 | de->inode = cpu_to_le32(inode->i_ino); | |
1866 | de->name_len = 1; | |
1867 | de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len), | |
1868 | blocksize); | |
1869 | strcpy(de->name, "."); | |
1870 | ext4_set_de_type(dir->i_sb, de, S_IFDIR); | |
1871 | de = ext4_next_entry(de, blocksize); | |
1872 | de->inode = cpu_to_le32(dir->i_ino); | |
1873 | de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1), | |
1874 | blocksize); | |
1875 | de->name_len = 2; | |
1876 | strcpy(de->name, ".."); | |
1877 | ext4_set_de_type(dir->i_sb, de, S_IFDIR); | |
1878 | inode->i_nlink = 2; | |
1879 | BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); | |
1880 | ext4_handle_dirty_metadata(handle, dir, dir_block); | |
1881 | brelse(dir_block); | |
1882 | ext4_mark_inode_dirty(handle, inode); | |
1883 | err = ext4_add_entry(handle, dentry, inode); | |
1884 | if (err) { | |
1885 | out_clear_inode: | |
1886 | clear_nlink(inode); | |
1887 | unlock_new_inode(inode); | |
1888 | ext4_mark_inode_dirty(handle, inode); | |
1889 | iput(inode); | |
1890 | goto out_stop; | |
1891 | } | |
1892 | ext4_inc_count(handle, dir); | |
1893 | ext4_update_dx_flag(dir); | |
1894 | ext4_mark_inode_dirty(handle, dir); | |
1895 | d_instantiate(dentry, inode); | |
1896 | unlock_new_inode(inode); | |
1897 | out_stop: | |
1898 | ext4_journal_stop(handle); | |
1899 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
1900 | goto retry; | |
1901 | return err; | |
1902 | } | |
1903 | ||
1904 | /* | |
1905 | * routine to check that the specified directory is empty (for rmdir) | |
1906 | */ | |
1907 | static int empty_dir(struct inode *inode) | |
1908 | { | |
1909 | unsigned int offset; | |
1910 | struct buffer_head *bh; | |
1911 | struct ext4_dir_entry_2 *de, *de1; | |
1912 | struct super_block *sb; | |
1913 | int err = 0; | |
1914 | ||
1915 | sb = inode->i_sb; | |
1916 | if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) || | |
1917 | !(bh = ext4_bread(NULL, inode, 0, 0, &err))) { | |
1918 | if (err) | |
1919 | ext4_error(inode->i_sb, __func__, | |
1920 | "error %d reading directory #%lu offset 0", | |
1921 | err, inode->i_ino); | |
1922 | else | |
1923 | ext4_warning(inode->i_sb, __func__, | |
1924 | "bad directory (dir #%lu) - no data block", | |
1925 | inode->i_ino); | |
1926 | return 1; | |
1927 | } | |
1928 | de = (struct ext4_dir_entry_2 *) bh->b_data; | |
1929 | de1 = ext4_next_entry(de, sb->s_blocksize); | |
1930 | if (le32_to_cpu(de->inode) != inode->i_ino || | |
1931 | !le32_to_cpu(de1->inode) || | |
1932 | strcmp(".", de->name) || | |
1933 | strcmp("..", de1->name)) { | |
1934 | ext4_warning(inode->i_sb, "empty_dir", | |
1935 | "bad directory (dir #%lu) - no `.' or `..'", | |
1936 | inode->i_ino); | |
1937 | brelse(bh); | |
1938 | return 1; | |
1939 | } | |
1940 | offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) + | |
1941 | ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize); | |
1942 | de = ext4_next_entry(de1, sb->s_blocksize); | |
1943 | while (offset < inode->i_size) { | |
1944 | if (!bh || | |
1945 | (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { | |
1946 | err = 0; | |
1947 | brelse(bh); | |
1948 | bh = ext4_bread(NULL, inode, | |
1949 | offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err); | |
1950 | if (!bh) { | |
1951 | if (err) | |
1952 | ext4_error(sb, __func__, | |
1953 | "error %d reading directory" | |
1954 | " #%lu offset %u", | |
1955 | err, inode->i_ino, offset); | |
1956 | offset += sb->s_blocksize; | |
1957 | continue; | |
1958 | } | |
1959 | de = (struct ext4_dir_entry_2 *) bh->b_data; | |
1960 | } | |
1961 | if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) { | |
1962 | de = (struct ext4_dir_entry_2 *)(bh->b_data + | |
1963 | sb->s_blocksize); | |
1964 | offset = (offset | (sb->s_blocksize - 1)) + 1; | |
1965 | continue; | |
1966 | } | |
1967 | if (le32_to_cpu(de->inode)) { | |
1968 | brelse(bh); | |
1969 | return 0; | |
1970 | } | |
1971 | offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); | |
1972 | de = ext4_next_entry(de, sb->s_blocksize); | |
1973 | } | |
1974 | brelse(bh); | |
1975 | return 1; | |
1976 | } | |
1977 | ||
1978 | /* ext4_orphan_add() links an unlinked or truncated inode into a list of | |
1979 | * such inodes, starting at the superblock, in case we crash before the | |
1980 | * file is closed/deleted, or in case the inode truncate spans multiple | |
1981 | * transactions and the last transaction is not recovered after a crash. | |
1982 | * | |
1983 | * At filesystem recovery time, we walk this list deleting unlinked | |
1984 | * inodes and truncating linked inodes in ext4_orphan_cleanup(). | |
1985 | */ | |
1986 | int ext4_orphan_add(handle_t *handle, struct inode *inode) | |
1987 | { | |
1988 | struct super_block *sb = inode->i_sb; | |
1989 | struct ext4_iloc iloc; | |
1990 | int err = 0, rc; | |
1991 | ||
1992 | if (!ext4_handle_valid(handle)) | |
1993 | return 0; | |
1994 | ||
1995 | mutex_lock(&EXT4_SB(sb)->s_orphan_lock); | |
1996 | if (!list_empty(&EXT4_I(inode)->i_orphan)) | |
1997 | goto out_unlock; | |
1998 | ||
1999 | /* Orphan handling is only valid for files with data blocks | |
2000 | * being truncated, or files being unlinked. */ | |
2001 | ||
2002 | /* @@@ FIXME: Observation from aviro: | |
2003 | * I think I can trigger J_ASSERT in ext4_orphan_add(). We block | |
2004 | * here (on s_orphan_lock), so race with ext4_link() which might bump | |
2005 | * ->i_nlink. For, say it, character device. Not a regular file, | |
2006 | * not a directory, not a symlink and ->i_nlink > 0. | |
2007 | * | |
2008 | * tytso, 4/25/2009: I'm not sure how that could happen; | |
2009 | * shouldn't the fs core protect us from these sort of | |
2010 | * unlink()/link() races? | |
2011 | */ | |
2012 | J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
2013 | S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); | |
2014 | ||
2015 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); | |
2016 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); | |
2017 | if (err) | |
2018 | goto out_unlock; | |
2019 | ||
2020 | err = ext4_reserve_inode_write(handle, inode, &iloc); | |
2021 | if (err) | |
2022 | goto out_unlock; | |
2023 | ||
2024 | /* Insert this inode at the head of the on-disk orphan list... */ | |
2025 | NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan); | |
2026 | EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); | |
2027 | err = ext4_handle_dirty_metadata(handle, inode, EXT4_SB(sb)->s_sbh); | |
2028 | rc = ext4_mark_iloc_dirty(handle, inode, &iloc); | |
2029 | if (!err) | |
2030 | err = rc; | |
2031 | ||
2032 | /* Only add to the head of the in-memory list if all the | |
2033 | * previous operations succeeded. If the orphan_add is going to | |
2034 | * fail (possibly taking the journal offline), we can't risk | |
2035 | * leaving the inode on the orphan list: stray orphan-list | |
2036 | * entries can cause panics at unmount time. | |
2037 | * | |
2038 | * This is safe: on error we're going to ignore the orphan list | |
2039 | * anyway on the next recovery. */ | |
2040 | if (!err) | |
2041 | list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); | |
2042 | ||
2043 | jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); | |
2044 | jbd_debug(4, "orphan inode %lu will point to %d\n", | |
2045 | inode->i_ino, NEXT_ORPHAN(inode)); | |
2046 | out_unlock: | |
2047 | mutex_unlock(&EXT4_SB(sb)->s_orphan_lock); | |
2048 | ext4_std_error(inode->i_sb, err); | |
2049 | return err; | |
2050 | } | |
2051 | ||
2052 | /* | |
2053 | * ext4_orphan_del() removes an unlinked or truncated inode from the list | |
2054 | * of such inodes stored on disk, because it is finally being cleaned up. | |
2055 | */ | |
2056 | int ext4_orphan_del(handle_t *handle, struct inode *inode) | |
2057 | { | |
2058 | struct list_head *prev; | |
2059 | struct ext4_inode_info *ei = EXT4_I(inode); | |
2060 | struct ext4_sb_info *sbi; | |
2061 | __u32 ino_next; | |
2062 | struct ext4_iloc iloc; | |
2063 | int err = 0; | |
2064 | ||
2065 | /* ext4_handle_valid() assumes a valid handle_t pointer */ | |
2066 | if (handle && !ext4_handle_valid(handle)) | |
2067 | return 0; | |
2068 | ||
2069 | mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock); | |
2070 | if (list_empty(&ei->i_orphan)) | |
2071 | goto out; | |
2072 | ||
2073 | ino_next = NEXT_ORPHAN(inode); | |
2074 | prev = ei->i_orphan.prev; | |
2075 | sbi = EXT4_SB(inode->i_sb); | |
2076 | ||
2077 | jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); | |
2078 | ||
2079 | list_del_init(&ei->i_orphan); | |
2080 | ||
2081 | /* If we're on an error path, we may not have a valid | |
2082 | * transaction handle with which to update the orphan list on | |
2083 | * disk, but we still need to remove the inode from the linked | |
2084 | * list in memory. */ | |
2085 | if (sbi->s_journal && !handle) | |
2086 | goto out; | |
2087 | ||
2088 | err = ext4_reserve_inode_write(handle, inode, &iloc); | |
2089 | if (err) | |
2090 | goto out_err; | |
2091 | ||
2092 | if (prev == &sbi->s_orphan) { | |
2093 | jbd_debug(4, "superblock will point to %u\n", ino_next); | |
2094 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); | |
2095 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); | |
2096 | if (err) | |
2097 | goto out_brelse; | |
2098 | sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); | |
2099 | err = ext4_handle_dirty_metadata(handle, inode, sbi->s_sbh); | |
2100 | } else { | |
2101 | struct ext4_iloc iloc2; | |
2102 | struct inode *i_prev = | |
2103 | &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; | |
2104 | ||
2105 | jbd_debug(4, "orphan inode %lu will point to %u\n", | |
2106 | i_prev->i_ino, ino_next); | |
2107 | err = ext4_reserve_inode_write(handle, i_prev, &iloc2); | |
2108 | if (err) | |
2109 | goto out_brelse; | |
2110 | NEXT_ORPHAN(i_prev) = ino_next; | |
2111 | err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); | |
2112 | } | |
2113 | if (err) | |
2114 | goto out_brelse; | |
2115 | NEXT_ORPHAN(inode) = 0; | |
2116 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); | |
2117 | ||
2118 | out_err: | |
2119 | ext4_std_error(inode->i_sb, err); | |
2120 | out: | |
2121 | mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock); | |
2122 | return err; | |
2123 | ||
2124 | out_brelse: | |
2125 | brelse(iloc.bh); | |
2126 | goto out_err; | |
2127 | } | |
2128 | ||
2129 | static int ext4_rmdir(struct inode *dir, struct dentry *dentry) | |
2130 | { | |
2131 | int retval; | |
2132 | struct inode *inode; | |
2133 | struct buffer_head *bh; | |
2134 | struct ext4_dir_entry_2 *de; | |
2135 | handle_t *handle; | |
2136 | ||
2137 | /* Initialize quotas before so that eventual writes go in | |
2138 | * separate transaction */ | |
2139 | vfs_dq_init(dentry->d_inode); | |
2140 | handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb)); | |
2141 | if (IS_ERR(handle)) | |
2142 | return PTR_ERR(handle); | |
2143 | ||
2144 | retval = -ENOENT; | |
2145 | bh = ext4_find_entry(dir, &dentry->d_name, &de); | |
2146 | if (!bh) | |
2147 | goto end_rmdir; | |
2148 | ||
2149 | if (IS_DIRSYNC(dir)) | |
2150 | ext4_handle_sync(handle); | |
2151 | ||
2152 | inode = dentry->d_inode; | |
2153 | ||
2154 | retval = -EIO; | |
2155 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2156 | goto end_rmdir; | |
2157 | ||
2158 | retval = -ENOTEMPTY; | |
2159 | if (!empty_dir(inode)) | |
2160 | goto end_rmdir; | |
2161 | ||
2162 | retval = ext4_delete_entry(handle, dir, de, bh); | |
2163 | if (retval) | |
2164 | goto end_rmdir; | |
2165 | if (!EXT4_DIR_LINK_EMPTY(inode)) | |
2166 | ext4_warning(inode->i_sb, "ext4_rmdir", | |
2167 | "empty directory has too many links (%d)", | |
2168 | inode->i_nlink); | |
2169 | inode->i_version++; | |
2170 | clear_nlink(inode); | |
2171 | /* There's no need to set i_disksize: the fact that i_nlink is | |
2172 | * zero will ensure that the right thing happens during any | |
2173 | * recovery. */ | |
2174 | inode->i_size = 0; | |
2175 | ext4_orphan_add(handle, inode); | |
2176 | inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode); | |
2177 | ext4_mark_inode_dirty(handle, inode); | |
2178 | ext4_dec_count(handle, dir); | |
2179 | ext4_update_dx_flag(dir); | |
2180 | ext4_mark_inode_dirty(handle, dir); | |
2181 | ||
2182 | end_rmdir: | |
2183 | ext4_journal_stop(handle); | |
2184 | brelse(bh); | |
2185 | return retval; | |
2186 | } | |
2187 | ||
2188 | static int ext4_unlink(struct inode *dir, struct dentry *dentry) | |
2189 | { | |
2190 | int retval; | |
2191 | struct inode *inode; | |
2192 | struct buffer_head *bh; | |
2193 | struct ext4_dir_entry_2 *de; | |
2194 | handle_t *handle; | |
2195 | ||
2196 | /* Initialize quotas before so that eventual writes go | |
2197 | * in separate transaction */ | |
2198 | vfs_dq_init(dentry->d_inode); | |
2199 | handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb)); | |
2200 | if (IS_ERR(handle)) | |
2201 | return PTR_ERR(handle); | |
2202 | ||
2203 | if (IS_DIRSYNC(dir)) | |
2204 | ext4_handle_sync(handle); | |
2205 | ||
2206 | retval = -ENOENT; | |
2207 | bh = ext4_find_entry(dir, &dentry->d_name, &de); | |
2208 | if (!bh) | |
2209 | goto end_unlink; | |
2210 | ||
2211 | inode = dentry->d_inode; | |
2212 | ||
2213 | retval = -EIO; | |
2214 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2215 | goto end_unlink; | |
2216 | ||
2217 | if (!inode->i_nlink) { | |
2218 | ext4_warning(inode->i_sb, "ext4_unlink", | |
2219 | "Deleting nonexistent file (%lu), %d", | |
2220 | inode->i_ino, inode->i_nlink); | |
2221 | inode->i_nlink = 1; | |
2222 | } | |
2223 | retval = ext4_delete_entry(handle, dir, de, bh); | |
2224 | if (retval) | |
2225 | goto end_unlink; | |
2226 | dir->i_ctime = dir->i_mtime = ext4_current_time(dir); | |
2227 | ext4_update_dx_flag(dir); | |
2228 | ext4_mark_inode_dirty(handle, dir); | |
2229 | drop_nlink(inode); | |
2230 | if (!inode->i_nlink) | |
2231 | ext4_orphan_add(handle, inode); | |
2232 | inode->i_ctime = ext4_current_time(inode); | |
2233 | ext4_mark_inode_dirty(handle, inode); | |
2234 | retval = 0; | |
2235 | ||
2236 | end_unlink: | |
2237 | ext4_journal_stop(handle); | |
2238 | brelse(bh); | |
2239 | return retval; | |
2240 | } | |
2241 | ||
2242 | static int ext4_symlink(struct inode *dir, | |
2243 | struct dentry *dentry, const char *symname) | |
2244 | { | |
2245 | handle_t *handle; | |
2246 | struct inode *inode; | |
2247 | int l, err, retries = 0; | |
2248 | ||
2249 | l = strlen(symname)+1; | |
2250 | if (l > dir->i_sb->s_blocksize) | |
2251 | return -ENAMETOOLONG; | |
2252 | ||
2253 | retry: | |
2254 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
2255 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 + | |
2256 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); | |
2257 | if (IS_ERR(handle)) | |
2258 | return PTR_ERR(handle); | |
2259 | ||
2260 | if (IS_DIRSYNC(dir)) | |
2261 | ext4_handle_sync(handle); | |
2262 | ||
2263 | inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO, | |
2264 | &dentry->d_name, 0); | |
2265 | err = PTR_ERR(inode); | |
2266 | if (IS_ERR(inode)) | |
2267 | goto out_stop; | |
2268 | ||
2269 | if (l > sizeof(EXT4_I(inode)->i_data)) { | |
2270 | inode->i_op = &ext4_symlink_inode_operations; | |
2271 | ext4_set_aops(inode); | |
2272 | /* | |
2273 | * page_symlink() calls into ext4_prepare/commit_write. | |
2274 | * We have a transaction open. All is sweetness. It also sets | |
2275 | * i_size in generic_commit_write(). | |
2276 | */ | |
2277 | err = __page_symlink(inode, symname, l, 1); | |
2278 | if (err) { | |
2279 | clear_nlink(inode); | |
2280 | unlock_new_inode(inode); | |
2281 | ext4_mark_inode_dirty(handle, inode); | |
2282 | iput(inode); | |
2283 | goto out_stop; | |
2284 | } | |
2285 | } else { | |
2286 | /* clear the extent format for fast symlink */ | |
2287 | EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL; | |
2288 | inode->i_op = &ext4_fast_symlink_inode_operations; | |
2289 | memcpy((char *)&EXT4_I(inode)->i_data, symname, l); | |
2290 | inode->i_size = l-1; | |
2291 | } | |
2292 | EXT4_I(inode)->i_disksize = inode->i_size; | |
2293 | err = ext4_add_nondir(handle, dentry, inode); | |
2294 | out_stop: | |
2295 | ext4_journal_stop(handle); | |
2296 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
2297 | goto retry; | |
2298 | return err; | |
2299 | } | |
2300 | ||
2301 | static int ext4_link(struct dentry *old_dentry, | |
2302 | struct inode *dir, struct dentry *dentry) | |
2303 | { | |
2304 | handle_t *handle; | |
2305 | struct inode *inode = old_dentry->d_inode; | |
2306 | int err, retries = 0; | |
2307 | ||
2308 | if (inode->i_nlink >= EXT4_LINK_MAX) | |
2309 | return -EMLINK; | |
2310 | ||
2311 | /* | |
2312 | * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing | |
2313 | * otherwise has the potential to corrupt the orphan inode list. | |
2314 | */ | |
2315 | if (inode->i_nlink == 0) | |
2316 | return -ENOENT; | |
2317 | ||
2318 | retry: | |
2319 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
2320 | EXT4_INDEX_EXTRA_TRANS_BLOCKS); | |
2321 | if (IS_ERR(handle)) | |
2322 | return PTR_ERR(handle); | |
2323 | ||
2324 | if (IS_DIRSYNC(dir)) | |
2325 | ext4_handle_sync(handle); | |
2326 | ||
2327 | inode->i_ctime = ext4_current_time(inode); | |
2328 | ext4_inc_count(handle, inode); | |
2329 | atomic_inc(&inode->i_count); | |
2330 | ||
2331 | err = ext4_add_entry(handle, dentry, inode); | |
2332 | if (!err) { | |
2333 | ext4_mark_inode_dirty(handle, inode); | |
2334 | d_instantiate(dentry, inode); | |
2335 | } else { | |
2336 | drop_nlink(inode); | |
2337 | iput(inode); | |
2338 | } | |
2339 | ext4_journal_stop(handle); | |
2340 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
2341 | goto retry; | |
2342 | return err; | |
2343 | } | |
2344 | ||
2345 | #define PARENT_INO(buffer, size) \ | |
2346 | (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode) | |
2347 | ||
2348 | /* | |
2349 | * Anybody can rename anything with this: the permission checks are left to the | |
2350 | * higher-level routines. | |
2351 | */ | |
2352 | static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry, | |
2353 | struct inode *new_dir, struct dentry *new_dentry) | |
2354 | { | |
2355 | handle_t *handle; | |
2356 | struct inode *old_inode, *new_inode; | |
2357 | struct buffer_head *old_bh, *new_bh, *dir_bh; | |
2358 | struct ext4_dir_entry_2 *old_de, *new_de; | |
2359 | int retval, force_da_alloc = 0; | |
2360 | ||
2361 | old_bh = new_bh = dir_bh = NULL; | |
2362 | ||
2363 | /* Initialize quotas before so that eventual writes go | |
2364 | * in separate transaction */ | |
2365 | if (new_dentry->d_inode) | |
2366 | vfs_dq_init(new_dentry->d_inode); | |
2367 | handle = ext4_journal_start(old_dir, 2 * | |
2368 | EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) + | |
2369 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2); | |
2370 | if (IS_ERR(handle)) | |
2371 | return PTR_ERR(handle); | |
2372 | ||
2373 | if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir)) | |
2374 | ext4_handle_sync(handle); | |
2375 | ||
2376 | old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de); | |
2377 | /* | |
2378 | * Check for inode number is _not_ due to possible IO errors. | |
2379 | * We might rmdir the source, keep it as pwd of some process | |
2380 | * and merrily kill the link to whatever was created under the | |
2381 | * same name. Goodbye sticky bit ;-< | |
2382 | */ | |
2383 | old_inode = old_dentry->d_inode; | |
2384 | retval = -ENOENT; | |
2385 | if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino) | |
2386 | goto end_rename; | |
2387 | ||
2388 | new_inode = new_dentry->d_inode; | |
2389 | new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de); | |
2390 | if (new_bh) { | |
2391 | if (!new_inode) { | |
2392 | brelse(new_bh); | |
2393 | new_bh = NULL; | |
2394 | } | |
2395 | } | |
2396 | if (S_ISDIR(old_inode->i_mode)) { | |
2397 | if (new_inode) { | |
2398 | retval = -ENOTEMPTY; | |
2399 | if (!empty_dir(new_inode)) | |
2400 | goto end_rename; | |
2401 | } | |
2402 | retval = -EIO; | |
2403 | dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval); | |
2404 | if (!dir_bh) | |
2405 | goto end_rename; | |
2406 | if (le32_to_cpu(PARENT_INO(dir_bh->b_data, | |
2407 | old_dir->i_sb->s_blocksize)) != old_dir->i_ino) | |
2408 | goto end_rename; | |
2409 | retval = -EMLINK; | |
2410 | if (!new_inode && new_dir != old_dir && | |
2411 | EXT4_DIR_LINK_MAX(new_dir)) | |
2412 | goto end_rename; | |
2413 | } | |
2414 | if (!new_bh) { | |
2415 | retval = ext4_add_entry(handle, new_dentry, old_inode); | |
2416 | if (retval) | |
2417 | goto end_rename; | |
2418 | } else { | |
2419 | BUFFER_TRACE(new_bh, "get write access"); | |
2420 | ext4_journal_get_write_access(handle, new_bh); | |
2421 | new_de->inode = cpu_to_le32(old_inode->i_ino); | |
2422 | if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb, | |
2423 | EXT4_FEATURE_INCOMPAT_FILETYPE)) | |
2424 | new_de->file_type = old_de->file_type; | |
2425 | new_dir->i_version++; | |
2426 | new_dir->i_ctime = new_dir->i_mtime = | |
2427 | ext4_current_time(new_dir); | |
2428 | ext4_mark_inode_dirty(handle, new_dir); | |
2429 | BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata"); | |
2430 | ext4_handle_dirty_metadata(handle, new_dir, new_bh); | |
2431 | brelse(new_bh); | |
2432 | new_bh = NULL; | |
2433 | } | |
2434 | ||
2435 | /* | |
2436 | * Like most other Unix systems, set the ctime for inodes on a | |
2437 | * rename. | |
2438 | */ | |
2439 | old_inode->i_ctime = ext4_current_time(old_inode); | |
2440 | ext4_mark_inode_dirty(handle, old_inode); | |
2441 | ||
2442 | /* | |
2443 | * ok, that's it | |
2444 | */ | |
2445 | if (le32_to_cpu(old_de->inode) != old_inode->i_ino || | |
2446 | old_de->name_len != old_dentry->d_name.len || | |
2447 | strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) || | |
2448 | (retval = ext4_delete_entry(handle, old_dir, | |
2449 | old_de, old_bh)) == -ENOENT) { | |
2450 | /* old_de could have moved from under us during htree split, so | |
2451 | * make sure that we are deleting the right entry. We might | |
2452 | * also be pointing to a stale entry in the unused part of | |
2453 | * old_bh so just checking inum and the name isn't enough. */ | |
2454 | struct buffer_head *old_bh2; | |
2455 | struct ext4_dir_entry_2 *old_de2; | |
2456 | ||
2457 | old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2); | |
2458 | if (old_bh2) { | |
2459 | retval = ext4_delete_entry(handle, old_dir, | |
2460 | old_de2, old_bh2); | |
2461 | brelse(old_bh2); | |
2462 | } | |
2463 | } | |
2464 | if (retval) { | |
2465 | ext4_warning(old_dir->i_sb, "ext4_rename", | |
2466 | "Deleting old file (%lu), %d, error=%d", | |
2467 | old_dir->i_ino, old_dir->i_nlink, retval); | |
2468 | } | |
2469 | ||
2470 | if (new_inode) { | |
2471 | ext4_dec_count(handle, new_inode); | |
2472 | new_inode->i_ctime = ext4_current_time(new_inode); | |
2473 | } | |
2474 | old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir); | |
2475 | ext4_update_dx_flag(old_dir); | |
2476 | if (dir_bh) { | |
2477 | BUFFER_TRACE(dir_bh, "get_write_access"); | |
2478 | ext4_journal_get_write_access(handle, dir_bh); | |
2479 | PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) = | |
2480 | cpu_to_le32(new_dir->i_ino); | |
2481 | BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata"); | |
2482 | ext4_handle_dirty_metadata(handle, old_dir, dir_bh); | |
2483 | ext4_dec_count(handle, old_dir); | |
2484 | if (new_inode) { | |
2485 | /* checked empty_dir above, can't have another parent, | |
2486 | * ext4_dec_count() won't work for many-linked dirs */ | |
2487 | new_inode->i_nlink = 0; | |
2488 | } else { | |
2489 | ext4_inc_count(handle, new_dir); | |
2490 | ext4_update_dx_flag(new_dir); | |
2491 | ext4_mark_inode_dirty(handle, new_dir); | |
2492 | } | |
2493 | } | |
2494 | ext4_mark_inode_dirty(handle, old_dir); | |
2495 | if (new_inode) { | |
2496 | ext4_mark_inode_dirty(handle, new_inode); | |
2497 | if (!new_inode->i_nlink) | |
2498 | ext4_orphan_add(handle, new_inode); | |
2499 | if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC)) | |
2500 | force_da_alloc = 1; | |
2501 | } | |
2502 | retval = 0; | |
2503 | ||
2504 | end_rename: | |
2505 | brelse(dir_bh); | |
2506 | brelse(old_bh); | |
2507 | brelse(new_bh); | |
2508 | ext4_journal_stop(handle); | |
2509 | if (retval == 0 && force_da_alloc) | |
2510 | ext4_alloc_da_blocks(old_inode); | |
2511 | return retval; | |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | * directories can handle most operations... | |
2516 | */ | |
2517 | const struct inode_operations ext4_dir_inode_operations = { | |
2518 | .create = ext4_create, | |
2519 | .lookup = ext4_lookup, | |
2520 | .link = ext4_link, | |
2521 | .unlink = ext4_unlink, | |
2522 | .symlink = ext4_symlink, | |
2523 | .mkdir = ext4_mkdir, | |
2524 | .rmdir = ext4_rmdir, | |
2525 | .mknod = ext4_mknod, | |
2526 | .rename = ext4_rename, | |
2527 | .setattr = ext4_setattr, | |
2528 | #ifdef CONFIG_EXT4_FS_XATTR | |
2529 | .setxattr = generic_setxattr, | |
2530 | .getxattr = generic_getxattr, | |
2531 | .listxattr = ext4_listxattr, | |
2532 | .removexattr = generic_removexattr, | |
2533 | #endif | |
2534 | .check_acl = ext4_check_acl, | |
2535 | .fiemap = ext4_fiemap, | |
2536 | }; | |
2537 | ||
2538 | const struct inode_operations ext4_special_inode_operations = { | |
2539 | .setattr = ext4_setattr, | |
2540 | #ifdef CONFIG_EXT4_FS_XATTR | |
2541 | .setxattr = generic_setxattr, | |
2542 | .getxattr = generic_getxattr, | |
2543 | .listxattr = ext4_listxattr, | |
2544 | .removexattr = generic_removexattr, | |
2545 | #endif | |
2546 | .check_acl = ext4_check_acl, | |
2547 | }; |