]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/ext3/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 | |
e9ad5620 | 18 | * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998 |
1da177e4 | 19 | * Hash Tree Directory indexing (c) |
e9ad5620 | 20 | * Daniel Phillips, 2001 |
1da177e4 | 21 | * Hash Tree Directory indexing porting |
e9ad5620 | 22 | * Christopher Li, 2002 |
1da177e4 | 23 | * Hash Tree Directory indexing cleanup |
e9ad5620 | 24 | * Theodore Ts'o, 2002 |
1da177e4 LT |
25 | */ |
26 | ||
27 | #include <linux/fs.h> | |
28 | #include <linux/pagemap.h> | |
29 | #include <linux/jbd.h> | |
30 | #include <linux/time.h> | |
31 | #include <linux/ext3_fs.h> | |
32 | #include <linux/ext3_jbd.h> | |
33 | #include <linux/fcntl.h> | |
34 | #include <linux/stat.h> | |
35 | #include <linux/string.h> | |
36 | #include <linux/quotaops.h> | |
37 | #include <linux/buffer_head.h> | |
caa38fb0 | 38 | #include <linux/bio.h> |
381be254 BD |
39 | |
40 | #include "namei.h" | |
1da177e4 LT |
41 | #include "xattr.h" |
42 | #include "acl.h" | |
43 | ||
44 | /* | |
45 | * define how far ahead to read directories while searching them. | |
46 | */ | |
47 | #define NAMEI_RA_CHUNKS 2 | |
48 | #define NAMEI_RA_BLOCKS 4 | |
49 | #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) | |
50 | #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b)) | |
51 | ||
52 | static struct buffer_head *ext3_append(handle_t *handle, | |
53 | struct inode *inode, | |
54 | u32 *block, int *err) | |
55 | { | |
56 | struct buffer_head *bh; | |
57 | ||
58 | *block = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
59 | ||
60 | if ((bh = ext3_bread(handle, inode, *block, 1, err))) { | |
61 | inode->i_size += inode->i_sb->s_blocksize; | |
62 | EXT3_I(inode)->i_disksize = inode->i_size; | |
63 | ext3_journal_get_write_access(handle,bh); | |
64 | } | |
65 | return bh; | |
66 | } | |
67 | ||
68 | #ifndef assert | |
69 | #define assert(test) J_ASSERT(test) | |
70 | #endif | |
71 | ||
72 | #ifndef swap | |
73 | #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0) | |
74 | #endif | |
75 | ||
76 | #ifdef DX_DEBUG | |
77 | #define dxtrace(command) command | |
78 | #else | |
ae6ddcc5 | 79 | #define dxtrace(command) |
1da177e4 LT |
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; | |
ef2b02d3 ES |
143 | u16 offs; |
144 | u16 size; | |
1da177e4 LT |
145 | }; |
146 | ||
1da177e4 LT |
147 | static inline unsigned dx_get_block (struct dx_entry *entry); |
148 | static void dx_set_block (struct dx_entry *entry, unsigned 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(struct dentry *dentry, | |
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 ext3_dir_entry_2 *de, int size, | |
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 ext3_dir_entry_2 *dx_move_dirents (char *from, char *to, | |
167 | struct dx_map_entry *offsets, int count); | |
168 | static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size); | |
169 | static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block); | |
170 | static int ext3_htree_next_block(struct inode *dir, __u32 hash, | |
171 | struct dx_frame *frame, | |
ae6ddcc5 | 172 | struct dx_frame *frames, |
1da177e4 LT |
173 | __u32 *start_hash); |
174 | static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry, | |
175 | struct ext3_dir_entry_2 **res_dir, int *err); | |
176 | static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry, | |
177 | struct inode *inode); | |
178 | ||
7c06a8dc JK |
179 | /* |
180 | * p is at least 6 bytes before the end of page | |
181 | */ | |
182 | static inline struct ext3_dir_entry_2 * | |
183 | ext3_next_entry(struct ext3_dir_entry_2 *p) | |
184 | { | |
185 | return (struct ext3_dir_entry_2 *)((char *)p + | |
186 | ext3_rec_len_from_disk(p->rec_len)); | |
187 | } | |
188 | ||
1da177e4 LT |
189 | /* |
190 | * Future: use high four bits of block for coalesce-on-delete flags | |
191 | * Mask them off for now. | |
192 | */ | |
193 | ||
194 | static inline unsigned dx_get_block (struct dx_entry *entry) | |
195 | { | |
196 | return le32_to_cpu(entry->block) & 0x00ffffff; | |
197 | } | |
198 | ||
199 | static inline void dx_set_block (struct dx_entry *entry, unsigned value) | |
200 | { | |
201 | entry->block = cpu_to_le32(value); | |
202 | } | |
203 | ||
204 | static inline unsigned dx_get_hash (struct dx_entry *entry) | |
205 | { | |
206 | return le32_to_cpu(entry->hash); | |
207 | } | |
208 | ||
209 | static inline void dx_set_hash (struct dx_entry *entry, unsigned value) | |
210 | { | |
211 | entry->hash = cpu_to_le32(value); | |
212 | } | |
213 | ||
214 | static inline unsigned dx_get_count (struct dx_entry *entries) | |
215 | { | |
216 | return le16_to_cpu(((struct dx_countlimit *) entries)->count); | |
217 | } | |
218 | ||
219 | static inline unsigned dx_get_limit (struct dx_entry *entries) | |
220 | { | |
221 | return le16_to_cpu(((struct dx_countlimit *) entries)->limit); | |
222 | } | |
223 | ||
224 | static inline void dx_set_count (struct dx_entry *entries, unsigned value) | |
225 | { | |
226 | ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); | |
227 | } | |
228 | ||
229 | static inline void dx_set_limit (struct dx_entry *entries, unsigned value) | |
230 | { | |
231 | ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); | |
232 | } | |
233 | ||
234 | static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize) | |
235 | { | |
236 | unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) - | |
237 | EXT3_DIR_REC_LEN(2) - infosize; | |
238 | return 0? 20: entry_space / sizeof(struct dx_entry); | |
239 | } | |
240 | ||
241 | static inline unsigned dx_node_limit (struct inode *dir) | |
242 | { | |
243 | unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0); | |
244 | return 0? 22: entry_space / sizeof(struct dx_entry); | |
245 | } | |
246 | ||
247 | /* | |
248 | * Debug | |
249 | */ | |
250 | #ifdef DX_DEBUG | |
251 | static void dx_show_index (char * label, struct dx_entry *entries) | |
252 | { | |
253 | int i, n = dx_get_count (entries); | |
254 | printk("%s index ", label); | |
255 | for (i = 0; i < n; i++) | |
256 | { | |
257 | printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i)); | |
258 | } | |
259 | printk("\n"); | |
260 | } | |
261 | ||
262 | struct stats | |
ae6ddcc5 | 263 | { |
1da177e4 LT |
264 | unsigned names; |
265 | unsigned space; | |
266 | unsigned bcount; | |
267 | }; | |
268 | ||
269 | static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de, | |
270 | int size, int show_names) | |
271 | { | |
272 | unsigned names = 0, space = 0; | |
273 | char *base = (char *) de; | |
274 | struct dx_hash_info h = *hinfo; | |
275 | ||
276 | printk("names: "); | |
277 | while ((char *) de < base + size) | |
278 | { | |
279 | if (de->inode) | |
280 | { | |
281 | if (show_names) | |
282 | { | |
283 | int len = de->name_len; | |
284 | char *name = de->name; | |
285 | while (len--) printk("%c", *name++); | |
286 | ext3fs_dirhash(de->name, de->name_len, &h); | |
287 | printk(":%x.%u ", h.hash, | |
288 | ((char *) de - base)); | |
289 | } | |
290 | space += EXT3_DIR_REC_LEN(de->name_len); | |
e9ad5620 | 291 | names++; |
1da177e4 | 292 | } |
7c06a8dc | 293 | de = ext3_next_entry(de); |
1da177e4 LT |
294 | } |
295 | printk("(%i)\n", names); | |
296 | return (struct stats) { names, space, 1 }; | |
297 | } | |
298 | ||
299 | struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, | |
300 | struct dx_entry *entries, int levels) | |
301 | { | |
302 | unsigned blocksize = dir->i_sb->s_blocksize; | |
303 | unsigned count = dx_get_count (entries), names = 0, space = 0, i; | |
304 | unsigned bcount = 0; | |
305 | struct buffer_head *bh; | |
306 | int err; | |
307 | printk("%i indexed blocks...\n", count); | |
308 | for (i = 0; i < count; i++, entries++) | |
309 | { | |
310 | u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0; | |
311 | u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; | |
312 | struct stats stats; | |
313 | printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); | |
314 | if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue; | |
315 | stats = levels? | |
316 | dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): | |
317 | dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0); | |
318 | names += stats.names; | |
319 | space += stats.space; | |
320 | bcount += stats.bcount; | |
321 | brelse (bh); | |
322 | } | |
323 | if (bcount) | |
324 | printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ", | |
325 | names, space/bcount,(space/bcount)*100/blocksize); | |
326 | return (struct stats) { names, space, bcount}; | |
327 | } | |
328 | #endif /* DX_DEBUG */ | |
329 | ||
330 | /* | |
331 | * Probe for a directory leaf block to search. | |
332 | * | |
333 | * dx_probe can return ERR_BAD_DX_DIR, which means there was a format | |
334 | * error in the directory index, and the caller should fall back to | |
335 | * searching the directory normally. The callers of dx_probe **MUST** | |
336 | * check for this error code, and make sure it never gets reflected | |
337 | * back to userspace. | |
338 | */ | |
339 | static struct dx_frame * | |
340 | dx_probe(struct dentry *dentry, struct inode *dir, | |
341 | struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err) | |
342 | { | |
343 | unsigned count, indirect; | |
344 | struct dx_entry *at, *entries, *p, *q, *m; | |
345 | struct dx_root *root; | |
346 | struct buffer_head *bh; | |
347 | struct dx_frame *frame = frame_in; | |
348 | u32 hash; | |
349 | ||
350 | frame->bh = NULL; | |
351 | if (dentry) | |
352 | dir = dentry->d_parent->d_inode; | |
353 | if (!(bh = ext3_bread (NULL,dir, 0, 0, err))) | |
354 | goto fail; | |
355 | root = (struct dx_root *) bh->b_data; | |
356 | if (root->info.hash_version != DX_HASH_TEA && | |
357 | root->info.hash_version != DX_HASH_HALF_MD4 && | |
358 | root->info.hash_version != DX_HASH_LEGACY) { | |
359 | ext3_warning(dir->i_sb, __FUNCTION__, | |
360 | "Unrecognised inode hash code %d", | |
361 | root->info.hash_version); | |
362 | brelse(bh); | |
363 | *err = ERR_BAD_DX_DIR; | |
364 | goto fail; | |
365 | } | |
366 | hinfo->hash_version = root->info.hash_version; | |
367 | hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed; | |
368 | if (dentry) | |
369 | ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo); | |
370 | hash = hinfo->hash; | |
371 | ||
372 | if (root->info.unused_flags & 1) { | |
373 | ext3_warning(dir->i_sb, __FUNCTION__, | |
374 | "Unimplemented inode hash flags: %#06x", | |
375 | root->info.unused_flags); | |
376 | brelse(bh); | |
377 | *err = ERR_BAD_DX_DIR; | |
378 | goto fail; | |
379 | } | |
380 | ||
381 | if ((indirect = root->info.indirect_levels) > 1) { | |
382 | ext3_warning(dir->i_sb, __FUNCTION__, | |
383 | "Unimplemented inode hash depth: %#06x", | |
384 | root->info.indirect_levels); | |
385 | brelse(bh); | |
386 | *err = ERR_BAD_DX_DIR; | |
387 | goto fail; | |
388 | } | |
389 | ||
390 | entries = (struct dx_entry *) (((char *)&root->info) + | |
391 | root->info.info_length); | |
3d82abae ES |
392 | |
393 | if (dx_get_limit(entries) != dx_root_limit(dir, | |
394 | root->info.info_length)) { | |
395 | ext3_warning(dir->i_sb, __FUNCTION__, | |
396 | "dx entry: limit != root limit"); | |
397 | brelse(bh); | |
398 | *err = ERR_BAD_DX_DIR; | |
399 | goto fail; | |
400 | } | |
401 | ||
1da177e4 LT |
402 | dxtrace (printk("Look up %x", hash)); |
403 | while (1) | |
404 | { | |
405 | count = dx_get_count(entries); | |
3d82abae ES |
406 | if (!count || count > dx_get_limit(entries)) { |
407 | ext3_warning(dir->i_sb, __FUNCTION__, | |
408 | "dx entry: no count or count > limit"); | |
409 | brelse(bh); | |
410 | *err = ERR_BAD_DX_DIR; | |
411 | goto fail2; | |
412 | } | |
413 | ||
1da177e4 LT |
414 | p = entries + 1; |
415 | q = entries + count - 1; | |
416 | while (p <= q) | |
417 | { | |
418 | m = p + (q - p)/2; | |
419 | dxtrace(printk(".")); | |
420 | if (dx_get_hash(m) > hash) | |
421 | q = m - 1; | |
422 | else | |
423 | p = m + 1; | |
424 | } | |
425 | ||
426 | if (0) // linear search cross check | |
427 | { | |
428 | unsigned n = count - 1; | |
429 | at = entries; | |
430 | while (n--) | |
431 | { | |
432 | dxtrace(printk(",")); | |
433 | if (dx_get_hash(++at) > hash) | |
434 | { | |
435 | at--; | |
436 | break; | |
437 | } | |
438 | } | |
439 | assert (at == p - 1); | |
440 | } | |
441 | ||
442 | at = p - 1; | |
443 | dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at))); | |
444 | frame->bh = bh; | |
445 | frame->entries = entries; | |
446 | frame->at = at; | |
447 | if (!indirect--) return frame; | |
448 | if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err))) | |
449 | goto fail2; | |
450 | at = entries = ((struct dx_node *) bh->b_data)->entries; | |
3d82abae ES |
451 | if (dx_get_limit(entries) != dx_node_limit (dir)) { |
452 | ext3_warning(dir->i_sb, __FUNCTION__, | |
453 | "dx entry: limit != node limit"); | |
454 | brelse(bh); | |
455 | *err = ERR_BAD_DX_DIR; | |
456 | goto fail2; | |
457 | } | |
1da177e4 | 458 | frame++; |
3d82abae | 459 | frame->bh = NULL; |
1da177e4 LT |
460 | } |
461 | fail2: | |
462 | while (frame >= frame_in) { | |
463 | brelse(frame->bh); | |
464 | frame--; | |
465 | } | |
466 | fail: | |
3d82abae ES |
467 | if (*err == ERR_BAD_DX_DIR) |
468 | ext3_warning(dir->i_sb, __FUNCTION__, | |
469 | "Corrupt dir inode %ld, running e2fsck is " | |
470 | "recommended.", dir->i_ino); | |
1da177e4 LT |
471 | return NULL; |
472 | } | |
473 | ||
474 | static void dx_release (struct dx_frame *frames) | |
475 | { | |
476 | if (frames[0].bh == NULL) | |
477 | return; | |
478 | ||
479 | if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels) | |
480 | brelse(frames[1].bh); | |
481 | brelse(frames[0].bh); | |
482 | } | |
483 | ||
484 | /* | |
485 | * This function increments the frame pointer to search the next leaf | |
486 | * block, and reads in the necessary intervening nodes if the search | |
487 | * should be necessary. Whether or not the search is necessary is | |
488 | * controlled by the hash parameter. If the hash value is even, then | |
489 | * the search is only continued if the next block starts with that | |
490 | * hash value. This is used if we are searching for a specific file. | |
491 | * | |
492 | * If the hash value is HASH_NB_ALWAYS, then always go to the next block. | |
493 | * | |
494 | * This function returns 1 if the caller should continue to search, | |
495 | * or 0 if it should not. If there is an error reading one of the | |
496 | * index blocks, it will a negative error code. | |
497 | * | |
498 | * If start_hash is non-null, it will be filled in with the starting | |
499 | * hash of the next page. | |
500 | */ | |
501 | static int ext3_htree_next_block(struct inode *dir, __u32 hash, | |
502 | struct dx_frame *frame, | |
ae6ddcc5 | 503 | struct dx_frame *frames, |
1da177e4 LT |
504 | __u32 *start_hash) |
505 | { | |
506 | struct dx_frame *p; | |
507 | struct buffer_head *bh; | |
508 | int err, num_frames = 0; | |
509 | __u32 bhash; | |
510 | ||
511 | p = frame; | |
512 | /* | |
513 | * Find the next leaf page by incrementing the frame pointer. | |
514 | * If we run out of entries in the interior node, loop around and | |
515 | * increment pointer in the parent node. When we break out of | |
516 | * this loop, num_frames indicates the number of interior | |
517 | * nodes need to be read. | |
518 | */ | |
519 | while (1) { | |
520 | if (++(p->at) < p->entries + dx_get_count(p->entries)) | |
521 | break; | |
522 | if (p == frames) | |
523 | return 0; | |
524 | num_frames++; | |
525 | p--; | |
526 | } | |
527 | ||
528 | /* | |
529 | * If the hash is 1, then continue only if the next page has a | |
530 | * continuation hash of any value. This is used for readdir | |
531 | * handling. Otherwise, check to see if the hash matches the | |
532 | * desired contiuation hash. If it doesn't, return since | |
533 | * there's no point to read in the successive index pages. | |
534 | */ | |
535 | bhash = dx_get_hash(p->at); | |
536 | if (start_hash) | |
537 | *start_hash = bhash; | |
538 | if ((hash & 1) == 0) { | |
539 | if ((bhash & ~1) != hash) | |
540 | return 0; | |
541 | } | |
542 | /* | |
543 | * If the hash is HASH_NB_ALWAYS, we always go to the next | |
544 | * block so no check is necessary | |
545 | */ | |
546 | while (num_frames--) { | |
547 | if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at), | |
548 | 0, &err))) | |
549 | return err; /* Failure */ | |
550 | p++; | |
551 | brelse (p->bh); | |
552 | p->bh = bh; | |
553 | p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; | |
554 | } | |
555 | return 1; | |
556 | } | |
557 | ||
558 | ||
1da177e4 LT |
559 | /* |
560 | * This function fills a red-black tree with information from a | |
561 | * directory block. It returns the number directory entries loaded | |
562 | * into the tree. If there is an error it is returned in err. | |
563 | */ | |
564 | static int htree_dirblock_to_tree(struct file *dir_file, | |
565 | struct inode *dir, int block, | |
566 | struct dx_hash_info *hinfo, | |
567 | __u32 start_hash, __u32 start_minor_hash) | |
568 | { | |
569 | struct buffer_head *bh; | |
570 | struct ext3_dir_entry_2 *de, *top; | |
571 | int err, count = 0; | |
572 | ||
573 | dxtrace(printk("In htree dirblock_to_tree: block %d\n", block)); | |
574 | if (!(bh = ext3_bread (NULL, dir, block, 0, &err))) | |
575 | return err; | |
576 | ||
577 | de = (struct ext3_dir_entry_2 *) bh->b_data; | |
578 | top = (struct ext3_dir_entry_2 *) ((char *) de + | |
579 | dir->i_sb->s_blocksize - | |
580 | EXT3_DIR_REC_LEN(0)); | |
581 | for (; de < top; de = ext3_next_entry(de)) { | |
40b85134 ES |
582 | if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh, |
583 | (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb)) | |
584 | +((char *)de - bh->b_data))) { | |
585 | /* On error, skip the f_pos to the next block. */ | |
586 | dir_file->f_pos = (dir_file->f_pos | | |
587 | (dir->i_sb->s_blocksize - 1)) + 1; | |
588 | brelse (bh); | |
589 | return count; | |
590 | } | |
1da177e4 LT |
591 | ext3fs_dirhash(de->name, de->name_len, hinfo); |
592 | if ((hinfo->hash < start_hash) || | |
593 | ((hinfo->hash == start_hash) && | |
594 | (hinfo->minor_hash < start_minor_hash))) | |
595 | continue; | |
596 | if (de->inode == 0) | |
597 | continue; | |
598 | if ((err = ext3_htree_store_dirent(dir_file, | |
599 | hinfo->hash, hinfo->minor_hash, de)) != 0) { | |
600 | brelse(bh); | |
601 | return err; | |
602 | } | |
603 | count++; | |
604 | } | |
605 | brelse(bh); | |
606 | return count; | |
607 | } | |
608 | ||
609 | ||
610 | /* | |
611 | * This function fills a red-black tree with information from a | |
612 | * directory. We start scanning the directory in hash order, starting | |
613 | * at start_hash and start_minor_hash. | |
614 | * | |
615 | * This function returns the number of entries inserted into the tree, | |
616 | * or a negative error code. | |
617 | */ | |
618 | int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash, | |
619 | __u32 start_minor_hash, __u32 *next_hash) | |
620 | { | |
621 | struct dx_hash_info hinfo; | |
622 | struct ext3_dir_entry_2 *de; | |
623 | struct dx_frame frames[2], *frame; | |
624 | struct inode *dir; | |
625 | int block, err; | |
626 | int count = 0; | |
627 | int ret; | |
628 | __u32 hashval; | |
629 | ||
630 | dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash, | |
631 | start_minor_hash)); | |
fe21a693 | 632 | dir = dir_file->f_path.dentry->d_inode; |
1da177e4 LT |
633 | if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) { |
634 | hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version; | |
635 | hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed; | |
636 | count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, | |
637 | start_hash, start_minor_hash); | |
638 | *next_hash = ~0; | |
639 | return count; | |
640 | } | |
641 | hinfo.hash = start_hash; | |
642 | hinfo.minor_hash = 0; | |
fe21a693 | 643 | frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err); |
1da177e4 LT |
644 | if (!frame) |
645 | return err; | |
646 | ||
647 | /* Add '.' and '..' from the htree header */ | |
648 | if (!start_hash && !start_minor_hash) { | |
649 | de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data; | |
650 | if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0) | |
651 | goto errout; | |
652 | count++; | |
653 | } | |
654 | if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { | |
655 | de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data; | |
656 | de = ext3_next_entry(de); | |
657 | if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0) | |
658 | goto errout; | |
659 | count++; | |
660 | } | |
661 | ||
662 | while (1) { | |
663 | block = dx_get_block(frame->at); | |
664 | ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, | |
665 | start_hash, start_minor_hash); | |
666 | if (ret < 0) { | |
667 | err = ret; | |
668 | goto errout; | |
669 | } | |
670 | count += ret; | |
671 | hashval = ~0; | |
ae6ddcc5 | 672 | ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS, |
1da177e4 LT |
673 | frame, frames, &hashval); |
674 | *next_hash = hashval; | |
675 | if (ret < 0) { | |
676 | err = ret; | |
677 | goto errout; | |
678 | } | |
679 | /* | |
680 | * Stop if: (a) there are no more entries, or | |
681 | * (b) we have inserted at least one entry and the | |
682 | * next hash value is not a continuation | |
683 | */ | |
684 | if ((ret == 0) || | |
685 | (count && ((hashval & 1) == 0))) | |
686 | break; | |
687 | } | |
688 | dx_release(frames); | |
ae6ddcc5 | 689 | dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n", |
1da177e4 LT |
690 | count, *next_hash)); |
691 | return count; | |
692 | errout: | |
693 | dx_release(frames); | |
694 | return (err); | |
695 | } | |
696 | ||
697 | ||
698 | /* | |
699 | * Directory block splitting, compacting | |
700 | */ | |
701 | ||
ef2b02d3 ES |
702 | /* |
703 | * Create map of hash values, offsets, and sizes, stored at end of block. | |
704 | * Returns number of entries mapped. | |
705 | */ | |
1da177e4 LT |
706 | static int dx_make_map (struct ext3_dir_entry_2 *de, int size, |
707 | struct dx_hash_info *hinfo, struct dx_map_entry *map_tail) | |
708 | { | |
709 | int count = 0; | |
710 | char *base = (char *) de; | |
711 | struct dx_hash_info h = *hinfo; | |
712 | ||
713 | while ((char *) de < base + size) | |
714 | { | |
715 | if (de->name_len && de->inode) { | |
716 | ext3fs_dirhash(de->name, de->name_len, &h); | |
717 | map_tail--; | |
718 | map_tail->hash = h.hash; | |
ef2b02d3 ES |
719 | map_tail->offs = (u16) ((char *) de - base); |
720 | map_tail->size = le16_to_cpu(de->rec_len); | |
1da177e4 LT |
721 | count++; |
722 | cond_resched(); | |
723 | } | |
724 | /* XXX: do we need to check rec_len == 0 case? -Chris */ | |
7c06a8dc | 725 | de = ext3_next_entry(de); |
1da177e4 LT |
726 | } |
727 | return count; | |
728 | } | |
729 | ||
ef2b02d3 | 730 | /* Sort map by hash value */ |
1da177e4 LT |
731 | static void dx_sort_map (struct dx_map_entry *map, unsigned count) |
732 | { | |
733 | struct dx_map_entry *p, *q, *top = map + count - 1; | |
734 | int more; | |
735 | /* Combsort until bubble sort doesn't suck */ | |
736 | while (count > 2) | |
737 | { | |
738 | count = count*10/13; | |
739 | if (count - 9 < 2) /* 9, 10 -> 11 */ | |
740 | count = 11; | |
741 | for (p = top, q = p - count; q >= map; p--, q--) | |
742 | if (p->hash < q->hash) | |
743 | swap(*p, *q); | |
744 | } | |
745 | /* Garden variety bubble sort */ | |
746 | do { | |
747 | more = 0; | |
748 | q = top; | |
749 | while (q-- > map) | |
750 | { | |
751 | if (q[1].hash >= q[0].hash) | |
752 | continue; | |
753 | swap(*(q+1), *q); | |
754 | more = 1; | |
755 | } | |
756 | } while(more); | |
757 | } | |
758 | ||
759 | static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block) | |
760 | { | |
761 | struct dx_entry *entries = frame->entries; | |
762 | struct dx_entry *old = frame->at, *new = old + 1; | |
763 | int count = dx_get_count(entries); | |
764 | ||
765 | assert(count < dx_get_limit(entries)); | |
766 | assert(old < entries + count); | |
767 | memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); | |
768 | dx_set_hash(new, hash); | |
769 | dx_set_block(new, block); | |
770 | dx_set_count(entries, count + 1); | |
771 | } | |
1da177e4 LT |
772 | |
773 | static void ext3_update_dx_flag(struct inode *inode) | |
774 | { | |
775 | if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb, | |
776 | EXT3_FEATURE_COMPAT_DIR_INDEX)) | |
777 | EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL; | |
778 | } | |
779 | ||
780 | /* | |
781 | * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure. | |
782 | * | |
783 | * `len <= EXT3_NAME_LEN' is guaranteed by caller. | |
784 | * `de != NULL' is guaranteed by caller. | |
785 | */ | |
786 | static inline int ext3_match (int len, const char * const name, | |
787 | struct ext3_dir_entry_2 * de) | |
788 | { | |
789 | if (len != de->name_len) | |
790 | return 0; | |
791 | if (!de->inode) | |
792 | return 0; | |
793 | return !memcmp(name, de->name, len); | |
794 | } | |
795 | ||
796 | /* | |
797 | * Returns 0 if not found, -1 on failure, and 1 on success | |
798 | */ | |
799 | static inline int search_dirblock(struct buffer_head * bh, | |
800 | struct inode *dir, | |
801 | struct dentry *dentry, | |
802 | unsigned long offset, | |
803 | struct ext3_dir_entry_2 ** res_dir) | |
804 | { | |
805 | struct ext3_dir_entry_2 * de; | |
806 | char * dlimit; | |
807 | int de_len; | |
808 | const char *name = dentry->d_name.name; | |
809 | int namelen = dentry->d_name.len; | |
810 | ||
811 | de = (struct ext3_dir_entry_2 *) bh->b_data; | |
812 | dlimit = bh->b_data + dir->i_sb->s_blocksize; | |
813 | while ((char *) de < dlimit) { | |
814 | /* this code is executed quadratically often */ | |
815 | /* do minimal checking `by hand' */ | |
816 | ||
817 | if ((char *) de + namelen <= dlimit && | |
818 | ext3_match (namelen, name, de)) { | |
819 | /* found a match - just to be sure, do a full check */ | |
820 | if (!ext3_check_dir_entry("ext3_find_entry", | |
821 | dir, de, bh, offset)) | |
822 | return -1; | |
823 | *res_dir = de; | |
824 | return 1; | |
825 | } | |
826 | /* prevent looping on a bad block */ | |
7c06a8dc | 827 | de_len = ext3_rec_len_from_disk(de->rec_len); |
1da177e4 LT |
828 | if (de_len <= 0) |
829 | return -1; | |
830 | offset += de_len; | |
831 | de = (struct ext3_dir_entry_2 *) ((char *) de + de_len); | |
832 | } | |
833 | return 0; | |
834 | } | |
835 | ||
836 | ||
837 | /* | |
838 | * ext3_find_entry() | |
839 | * | |
840 | * finds an entry in the specified directory with the wanted name. It | |
841 | * returns the cache buffer in which the entry was found, and the entry | |
842 | * itself (as a parameter - res_dir). It does NOT read the inode of the | |
843 | * entry - you'll have to do that yourself if you want to. | |
844 | * | |
845 | * The returned buffer_head has ->b_count elevated. The caller is expected | |
846 | * to brelse() it when appropriate. | |
847 | */ | |
848 | static struct buffer_head * ext3_find_entry (struct dentry *dentry, | |
849 | struct ext3_dir_entry_2 ** res_dir) | |
850 | { | |
851 | struct super_block * sb; | |
852 | struct buffer_head * bh_use[NAMEI_RA_SIZE]; | |
853 | struct buffer_head * bh, *ret = NULL; | |
854 | unsigned long start, block, b; | |
855 | int ra_max = 0; /* Number of bh's in the readahead | |
856 | buffer, bh_use[] */ | |
857 | int ra_ptr = 0; /* Current index into readahead | |
858 | buffer */ | |
859 | int num = 0; | |
860 | int nblocks, i, err; | |
861 | struct inode *dir = dentry->d_parent->d_inode; | |
862 | int namelen; | |
1da177e4 LT |
863 | |
864 | *res_dir = NULL; | |
865 | sb = dir->i_sb; | |
1da177e4 | 866 | namelen = dentry->d_name.len; |
1da177e4 LT |
867 | if (namelen > EXT3_NAME_LEN) |
868 | return NULL; | |
1da177e4 LT |
869 | if (is_dx(dir)) { |
870 | bh = ext3_dx_find_entry(dentry, res_dir, &err); | |
871 | /* | |
872 | * On success, or if the error was file not found, | |
873 | * return. Otherwise, fall back to doing a search the | |
874 | * old fashioned way. | |
875 | */ | |
876 | if (bh || (err != ERR_BAD_DX_DIR)) | |
877 | return bh; | |
878 | dxtrace(printk("ext3_find_entry: dx failed, falling back\n")); | |
879 | } | |
1da177e4 LT |
880 | nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb); |
881 | start = EXT3_I(dir)->i_dir_start_lookup; | |
882 | if (start >= nblocks) | |
883 | start = 0; | |
884 | block = start; | |
885 | restart: | |
886 | do { | |
887 | /* | |
888 | * We deal with the read-ahead logic here. | |
889 | */ | |
890 | if (ra_ptr >= ra_max) { | |
891 | /* Refill the readahead buffer */ | |
892 | ra_ptr = 0; | |
893 | b = block; | |
894 | for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { | |
895 | /* | |
896 | * Terminate if we reach the end of the | |
897 | * directory and must wrap, or if our | |
898 | * search has finished at this block. | |
899 | */ | |
900 | if (b >= nblocks || (num && block == start)) { | |
901 | bh_use[ra_max] = NULL; | |
902 | break; | |
903 | } | |
904 | num++; | |
905 | bh = ext3_getblk(NULL, dir, b++, 0, &err); | |
906 | bh_use[ra_max] = bh; | |
907 | if (bh) | |
caa38fb0 | 908 | ll_rw_block(READ_META, 1, &bh); |
1da177e4 LT |
909 | } |
910 | } | |
911 | if ((bh = bh_use[ra_ptr++]) == NULL) | |
912 | goto next; | |
913 | wait_on_buffer(bh); | |
914 | if (!buffer_uptodate(bh)) { | |
915 | /* read error, skip block & hope for the best */ | |
916 | ext3_error(sb, __FUNCTION__, "reading directory #%lu " | |
917 | "offset %lu", dir->i_ino, block); | |
918 | brelse(bh); | |
919 | goto next; | |
920 | } | |
921 | i = search_dirblock(bh, dir, dentry, | |
922 | block << EXT3_BLOCK_SIZE_BITS(sb), res_dir); | |
923 | if (i == 1) { | |
924 | EXT3_I(dir)->i_dir_start_lookup = block; | |
925 | ret = bh; | |
926 | goto cleanup_and_exit; | |
927 | } else { | |
928 | brelse(bh); | |
929 | if (i < 0) | |
930 | goto cleanup_and_exit; | |
931 | } | |
932 | next: | |
933 | if (++block >= nblocks) | |
934 | block = 0; | |
935 | } while (block != start); | |
936 | ||
937 | /* | |
938 | * If the directory has grown while we were searching, then | |
939 | * search the last part of the directory before giving up. | |
940 | */ | |
941 | block = nblocks; | |
942 | nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb); | |
943 | if (block < nblocks) { | |
944 | start = 0; | |
945 | goto restart; | |
946 | } | |
947 | ||
948 | cleanup_and_exit: | |
949 | /* Clean up the read-ahead blocks */ | |
950 | for (; ra_ptr < ra_max; ra_ptr++) | |
951 | brelse (bh_use[ra_ptr]); | |
952 | return ret; | |
953 | } | |
954 | ||
1da177e4 LT |
955 | static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry, |
956 | struct ext3_dir_entry_2 **res_dir, int *err) | |
957 | { | |
958 | struct super_block * sb; | |
959 | struct dx_hash_info hinfo; | |
960 | u32 hash; | |
961 | struct dx_frame frames[2], *frame; | |
962 | struct ext3_dir_entry_2 *de, *top; | |
963 | struct buffer_head *bh; | |
964 | unsigned long block; | |
965 | int retval; | |
966 | int namelen = dentry->d_name.len; | |
967 | const u8 *name = dentry->d_name.name; | |
968 | struct inode *dir = dentry->d_parent->d_inode; | |
969 | ||
970 | sb = dir->i_sb; | |
acfa1823 AD |
971 | /* NFS may look up ".." - look at dx_root directory block */ |
972 | if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){ | |
973 | if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err))) | |
974 | return NULL; | |
975 | } else { | |
976 | frame = frames; | |
977 | frame->bh = NULL; /* for dx_release() */ | |
978 | frame->at = (struct dx_entry *)frames; /* hack for zero entry*/ | |
979 | dx_set_block(frame->at, 0); /* dx_root block is 0 */ | |
980 | } | |
1da177e4 LT |
981 | hash = hinfo.hash; |
982 | do { | |
983 | block = dx_get_block(frame->at); | |
984 | if (!(bh = ext3_bread (NULL,dir, block, 0, err))) | |
985 | goto errout; | |
986 | de = (struct ext3_dir_entry_2 *) bh->b_data; | |
987 | top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize - | |
988 | EXT3_DIR_REC_LEN(0)); | |
989 | for (; de < top; de = ext3_next_entry(de)) | |
990 | if (ext3_match (namelen, name, de)) { | |
991 | if (!ext3_check_dir_entry("ext3_find_entry", | |
992 | dir, de, bh, | |
993 | (block<<EXT3_BLOCK_SIZE_BITS(sb)) | |
994 | +((char *)de - bh->b_data))) { | |
995 | brelse (bh); | |
fedee54d | 996 | *err = ERR_BAD_DX_DIR; |
1da177e4 LT |
997 | goto errout; |
998 | } | |
999 | *res_dir = de; | |
1000 | dx_release (frames); | |
1001 | return bh; | |
1002 | } | |
1003 | brelse (bh); | |
1004 | /* Check to see if we should continue to search */ | |
1005 | retval = ext3_htree_next_block(dir, hash, frame, | |
1006 | frames, NULL); | |
1007 | if (retval < 0) { | |
1008 | ext3_warning(sb, __FUNCTION__, | |
1009 | "error reading index page in directory #%lu", | |
1010 | dir->i_ino); | |
1011 | *err = retval; | |
1012 | goto errout; | |
1013 | } | |
1014 | } while (retval == 1); | |
1015 | ||
1016 | *err = -ENOENT; | |
1017 | errout: | |
1018 | dxtrace(printk("%s not found\n", name)); | |
1019 | dx_release (frames); | |
1020 | return NULL; | |
1021 | } | |
1da177e4 LT |
1022 | |
1023 | static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd) | |
1024 | { | |
1025 | struct inode * inode; | |
1026 | struct ext3_dir_entry_2 * de; | |
1027 | struct buffer_head * bh; | |
1028 | ||
1029 | if (dentry->d_name.len > EXT3_NAME_LEN) | |
1030 | return ERR_PTR(-ENAMETOOLONG); | |
1031 | ||
1032 | bh = ext3_find_entry(dentry, &de); | |
1033 | inode = NULL; | |
1034 | if (bh) { | |
1035 | unsigned long ino = le32_to_cpu(de->inode); | |
1036 | brelse (bh); | |
2ccb48eb NB |
1037 | if (!ext3_valid_inum(dir->i_sb, ino)) { |
1038 | ext3_error(dir->i_sb, "ext3_lookup", | |
1039 | "bad inode number: %lu", ino); | |
473043dc | 1040 | return ERR_PTR(-EIO); |
a6c15c2b | 1041 | } |
473043dc DH |
1042 | inode = ext3_iget(dir->i_sb, ino); |
1043 | if (IS_ERR(inode)) | |
1044 | return ERR_CAST(inode); | |
1da177e4 | 1045 | } |
ba7fe369 | 1046 | return d_splice_alias(inode, dentry); |
1da177e4 LT |
1047 | } |
1048 | ||
1049 | ||
1050 | struct dentry *ext3_get_parent(struct dentry *child) | |
1051 | { | |
1052 | unsigned long ino; | |
1053 | struct dentry *parent; | |
1054 | struct inode *inode; | |
1055 | struct dentry dotdot; | |
1056 | struct ext3_dir_entry_2 * de; | |
1057 | struct buffer_head *bh; | |
1058 | ||
1059 | dotdot.d_name.name = ".."; | |
1060 | dotdot.d_name.len = 2; | |
1061 | dotdot.d_parent = child; /* confusing, isn't it! */ | |
1062 | ||
1063 | bh = ext3_find_entry(&dotdot, &de); | |
1064 | inode = NULL; | |
1065 | if (!bh) | |
1066 | return ERR_PTR(-ENOENT); | |
1067 | ino = le32_to_cpu(de->inode); | |
1068 | brelse(bh); | |
2ccb48eb NB |
1069 | |
1070 | if (!ext3_valid_inum(child->d_inode->i_sb, ino)) { | |
1071 | ext3_error(child->d_inode->i_sb, "ext3_get_parent", | |
1072 | "bad inode number: %lu", ino); | |
473043dc | 1073 | return ERR_PTR(-EIO); |
a6c15c2b VA |
1074 | } |
1075 | ||
473043dc DH |
1076 | inode = ext3_iget(child->d_inode->i_sb, ino); |
1077 | if (IS_ERR(inode)) | |
1078 | return ERR_CAST(inode); | |
1079 | ||
1da177e4 LT |
1080 | parent = d_alloc_anon(inode); |
1081 | if (!parent) { | |
1082 | iput(inode); | |
1083 | parent = ERR_PTR(-ENOMEM); | |
1084 | } | |
1085 | return parent; | |
ae6ddcc5 | 1086 | } |
1da177e4 LT |
1087 | |
1088 | #define S_SHIFT 12 | |
1089 | static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = { | |
1090 | [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE, | |
1091 | [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR, | |
1092 | [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV, | |
1093 | [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV, | |
1094 | [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO, | |
1095 | [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK, | |
1096 | [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK, | |
1097 | }; | |
1098 | ||
1099 | static inline void ext3_set_de_type(struct super_block *sb, | |
1100 | struct ext3_dir_entry_2 *de, | |
1101 | umode_t mode) { | |
1102 | if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE)) | |
1103 | de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT]; | |
1104 | } | |
1105 | ||
ef2b02d3 ES |
1106 | /* |
1107 | * Move count entries from end of map between two memory locations. | |
1108 | * Returns pointer to last entry moved. | |
1109 | */ | |
1da177e4 LT |
1110 | static struct ext3_dir_entry_2 * |
1111 | dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count) | |
1112 | { | |
1113 | unsigned rec_len = 0; | |
1114 | ||
1115 | while (count--) { | |
1116 | struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs); | |
1117 | rec_len = EXT3_DIR_REC_LEN(de->name_len); | |
1118 | memcpy (to, de, rec_len); | |
1119 | ((struct ext3_dir_entry_2 *) to)->rec_len = | |
7c06a8dc | 1120 | ext3_rec_len_to_disk(rec_len); |
1da177e4 LT |
1121 | de->inode = 0; |
1122 | map++; | |
1123 | to += rec_len; | |
1124 | } | |
1125 | return (struct ext3_dir_entry_2 *) (to - rec_len); | |
1126 | } | |
1127 | ||
ef2b02d3 ES |
1128 | /* |
1129 | * Compact each dir entry in the range to the minimal rec_len. | |
1130 | * Returns pointer to last entry in range. | |
1131 | */ | |
1da177e4 LT |
1132 | static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size) |
1133 | { | |
1134 | struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base; | |
1135 | unsigned rec_len = 0; | |
1136 | ||
1137 | prev = to = de; | |
1138 | while ((char*)de < base + size) { | |
7c06a8dc | 1139 | next = ext3_next_entry(de); |
1da177e4 LT |
1140 | if (de->inode && de->name_len) { |
1141 | rec_len = EXT3_DIR_REC_LEN(de->name_len); | |
1142 | if (de > to) | |
1143 | memmove(to, de, rec_len); | |
7c06a8dc | 1144 | to->rec_len = ext3_rec_len_to_disk(rec_len); |
1da177e4 LT |
1145 | prev = to; |
1146 | to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len); | |
1147 | } | |
1148 | de = next; | |
1149 | } | |
1150 | return prev; | |
1151 | } | |
1152 | ||
ef2b02d3 ES |
1153 | /* |
1154 | * Split a full leaf block to make room for a new dir entry. | |
1155 | * Allocate a new block, and move entries so that they are approx. equally full. | |
1156 | * Returns pointer to de in block into which the new entry will be inserted. | |
1157 | */ | |
1da177e4 LT |
1158 | static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, |
1159 | struct buffer_head **bh,struct dx_frame *frame, | |
1160 | struct dx_hash_info *hinfo, int *error) | |
1161 | { | |
1162 | unsigned blocksize = dir->i_sb->s_blocksize; | |
1163 | unsigned count, continued; | |
1164 | struct buffer_head *bh2; | |
1165 | u32 newblock; | |
1166 | u32 hash2; | |
1167 | struct dx_map_entry *map; | |
1168 | char *data1 = (*bh)->b_data, *data2; | |
ef2b02d3 | 1169 | unsigned split, move, size, i; |
1da177e4 | 1170 | struct ext3_dir_entry_2 *de = NULL, *de2; |
fedee54d | 1171 | int err = 0; |
1da177e4 | 1172 | |
fedee54d | 1173 | bh2 = ext3_append (handle, dir, &newblock, &err); |
1da177e4 LT |
1174 | if (!(bh2)) { |
1175 | brelse(*bh); | |
1176 | *bh = NULL; | |
1177 | goto errout; | |
1178 | } | |
1179 | ||
1180 | BUFFER_TRACE(*bh, "get_write_access"); | |
1181 | err = ext3_journal_get_write_access(handle, *bh); | |
fedee54d DM |
1182 | if (err) |
1183 | goto journal_error; | |
1184 | ||
1da177e4 LT |
1185 | BUFFER_TRACE(frame->bh, "get_write_access"); |
1186 | err = ext3_journal_get_write_access(handle, frame->bh); | |
1187 | if (err) | |
1188 | goto journal_error; | |
1189 | ||
1190 | data2 = bh2->b_data; | |
1191 | ||
1192 | /* create map in the end of data2 block */ | |
1193 | map = (struct dx_map_entry *) (data2 + blocksize); | |
1194 | count = dx_make_map ((struct ext3_dir_entry_2 *) data1, | |
1195 | blocksize, hinfo, map); | |
1196 | map -= count; | |
1da177e4 | 1197 | dx_sort_map (map, count); |
ef2b02d3 ES |
1198 | /* Split the existing block in the middle, size-wise */ |
1199 | size = 0; | |
1200 | move = 0; | |
1201 | for (i = count-1; i >= 0; i--) { | |
1202 | /* is more than half of this entry in 2nd half of the block? */ | |
1203 | if (size + map[i].size/2 > blocksize/2) | |
1204 | break; | |
1205 | size += map[i].size; | |
1206 | move++; | |
1207 | } | |
1208 | /* map index at which we will split */ | |
1209 | split = count - move; | |
1da177e4 LT |
1210 | hash2 = map[split].hash; |
1211 | continued = hash2 == map[split - 1].hash; | |
1212 | dxtrace(printk("Split block %i at %x, %i/%i\n", | |
1213 | dx_get_block(frame->at), hash2, split, count-split)); | |
1214 | ||
1215 | /* Fancy dance to stay within two buffers */ | |
1216 | de2 = dx_move_dirents(data1, data2, map + split, count - split); | |
1217 | de = dx_pack_dirents(data1,blocksize); | |
7c06a8dc JK |
1218 | de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de); |
1219 | de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2); | |
1da177e4 LT |
1220 | dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1)); |
1221 | dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1)); | |
1222 | ||
1223 | /* Which block gets the new entry? */ | |
1224 | if (hinfo->hash >= hash2) | |
1225 | { | |
1226 | swap(*bh, bh2); | |
1227 | de = de2; | |
1228 | } | |
1229 | dx_insert_block (frame, hash2 + continued, newblock); | |
1230 | err = ext3_journal_dirty_metadata (handle, bh2); | |
1231 | if (err) | |
1232 | goto journal_error; | |
1233 | err = ext3_journal_dirty_metadata (handle, frame->bh); | |
1234 | if (err) | |
1235 | goto journal_error; | |
1236 | brelse (bh2); | |
1237 | dxtrace(dx_show_index ("frame", frame->entries)); | |
1da177e4 | 1238 | return de; |
fedee54d DM |
1239 | |
1240 | journal_error: | |
1241 | brelse(*bh); | |
1242 | brelse(bh2); | |
1243 | *bh = NULL; | |
1244 | ext3_std_error(dir->i_sb, err); | |
1245 | errout: | |
1246 | *error = err; | |
1247 | return NULL; | |
1da177e4 | 1248 | } |
1da177e4 LT |
1249 | |
1250 | ||
1251 | /* | |
1252 | * Add a new entry into a directory (leaf) block. If de is non-NULL, | |
1253 | * it points to a directory entry which is guaranteed to be large | |
1254 | * enough for new directory entry. If de is NULL, then | |
1255 | * add_dirent_to_buf will attempt search the directory block for | |
1256 | * space. It will return -ENOSPC if no space is available, and -EIO | |
1257 | * and -EEXIST if directory entry already exists. | |
ae6ddcc5 | 1258 | * |
1da177e4 LT |
1259 | * NOTE! bh is NOT released in the case where ENOSPC is returned. In |
1260 | * all other cases bh is released. | |
1261 | */ | |
1262 | static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry, | |
1263 | struct inode *inode, struct ext3_dir_entry_2 *de, | |
1264 | struct buffer_head * bh) | |
1265 | { | |
1266 | struct inode *dir = dentry->d_parent->d_inode; | |
1267 | const char *name = dentry->d_name.name; | |
1268 | int namelen = dentry->d_name.len; | |
1269 | unsigned long offset = 0; | |
1270 | unsigned short reclen; | |
1271 | int nlen, rlen, err; | |
1272 | char *top; | |
1273 | ||
1274 | reclen = EXT3_DIR_REC_LEN(namelen); | |
1275 | if (!de) { | |
1276 | de = (struct ext3_dir_entry_2 *)bh->b_data; | |
1277 | top = bh->b_data + dir->i_sb->s_blocksize - reclen; | |
1278 | while ((char *) de <= top) { | |
1279 | if (!ext3_check_dir_entry("ext3_add_entry", dir, de, | |
1280 | bh, offset)) { | |
1281 | brelse (bh); | |
1282 | return -EIO; | |
1283 | } | |
1284 | if (ext3_match (namelen, name, de)) { | |
1285 | brelse (bh); | |
1286 | return -EEXIST; | |
1287 | } | |
1288 | nlen = EXT3_DIR_REC_LEN(de->name_len); | |
7c06a8dc | 1289 | rlen = ext3_rec_len_from_disk(de->rec_len); |
1da177e4 LT |
1290 | if ((de->inode? rlen - nlen: rlen) >= reclen) |
1291 | break; | |
1292 | de = (struct ext3_dir_entry_2 *)((char *)de + rlen); | |
1293 | offset += rlen; | |
1294 | } | |
1295 | if ((char *) de > top) | |
1296 | return -ENOSPC; | |
1297 | } | |
1298 | BUFFER_TRACE(bh, "get_write_access"); | |
1299 | err = ext3_journal_get_write_access(handle, bh); | |
1300 | if (err) { | |
1301 | ext3_std_error(dir->i_sb, err); | |
1302 | brelse(bh); | |
1303 | return err; | |
1304 | } | |
1305 | ||
1306 | /* By now the buffer is marked for journaling */ | |
1307 | nlen = EXT3_DIR_REC_LEN(de->name_len); | |
7c06a8dc | 1308 | rlen = ext3_rec_len_from_disk(de->rec_len); |
1da177e4 LT |
1309 | if (de->inode) { |
1310 | struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen); | |
7c06a8dc JK |
1311 | de1->rec_len = ext3_rec_len_to_disk(rlen - nlen); |
1312 | de->rec_len = ext3_rec_len_to_disk(nlen); | |
1da177e4 LT |
1313 | de = de1; |
1314 | } | |
1315 | de->file_type = EXT3_FT_UNKNOWN; | |
1316 | if (inode) { | |
1317 | de->inode = cpu_to_le32(inode->i_ino); | |
1318 | ext3_set_de_type(dir->i_sb, de, inode->i_mode); | |
1319 | } else | |
1320 | de->inode = 0; | |
1321 | de->name_len = namelen; | |
1322 | memcpy (de->name, name, namelen); | |
1323 | /* | |
1324 | * XXX shouldn't update any times until successful | |
1325 | * completion of syscall, but too many callers depend | |
1326 | * on this. | |
1327 | * | |
1328 | * XXX similarly, too many callers depend on | |
1329 | * ext3_new_inode() setting the times, but error | |
1330 | * recovery deletes the inode, so the worst that can | |
1331 | * happen is that the times are slightly out of date | |
1332 | * and/or different from the directory change time. | |
1333 | */ | |
1334 | dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC; | |
1335 | ext3_update_dx_flag(dir); | |
1336 | dir->i_version++; | |
1337 | ext3_mark_inode_dirty(handle, dir); | |
1338 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
1339 | err = ext3_journal_dirty_metadata(handle, bh); | |
1340 | if (err) | |
1341 | ext3_std_error(dir->i_sb, err); | |
1342 | brelse(bh); | |
1343 | return 0; | |
1344 | } | |
1345 | ||
1da177e4 LT |
1346 | /* |
1347 | * This converts a one block unindexed directory to a 3 block indexed | |
1348 | * directory, and adds the dentry to the indexed directory. | |
1349 | */ | |
1350 | static int make_indexed_dir(handle_t *handle, struct dentry *dentry, | |
1351 | struct inode *inode, struct buffer_head *bh) | |
1352 | { | |
1353 | struct inode *dir = dentry->d_parent->d_inode; | |
1354 | const char *name = dentry->d_name.name; | |
1355 | int namelen = dentry->d_name.len; | |
1356 | struct buffer_head *bh2; | |
1357 | struct dx_root *root; | |
1358 | struct dx_frame frames[2], *frame; | |
1359 | struct dx_entry *entries; | |
1360 | struct ext3_dir_entry_2 *de, *de2; | |
1361 | char *data1, *top; | |
1362 | unsigned len; | |
1363 | int retval; | |
1364 | unsigned blocksize; | |
1365 | struct dx_hash_info hinfo; | |
1366 | u32 block; | |
1367 | struct fake_dirent *fde; | |
1368 | ||
1369 | blocksize = dir->i_sb->s_blocksize; | |
1370 | dxtrace(printk("Creating index\n")); | |
1371 | retval = ext3_journal_get_write_access(handle, bh); | |
1372 | if (retval) { | |
1373 | ext3_std_error(dir->i_sb, retval); | |
1374 | brelse(bh); | |
1375 | return retval; | |
1376 | } | |
1377 | root = (struct dx_root *) bh->b_data; | |
1378 | ||
1379 | bh2 = ext3_append (handle, dir, &block, &retval); | |
1380 | if (!(bh2)) { | |
1381 | brelse(bh); | |
1382 | return retval; | |
1383 | } | |
1384 | EXT3_I(dir)->i_flags |= EXT3_INDEX_FL; | |
1385 | data1 = bh2->b_data; | |
1386 | ||
1387 | /* The 0th block becomes the root, move the dirents out */ | |
1388 | fde = &root->dotdot; | |
7c06a8dc JK |
1389 | de = (struct ext3_dir_entry_2 *)((char *)fde + |
1390 | ext3_rec_len_from_disk(fde->rec_len)); | |
1da177e4 LT |
1391 | len = ((char *) root) + blocksize - (char *) de; |
1392 | memcpy (data1, de, len); | |
1393 | de = (struct ext3_dir_entry_2 *) data1; | |
1394 | top = data1 + len; | |
7c06a8dc | 1395 | while ((char *)(de2 = ext3_next_entry(de)) < top) |
1da177e4 | 1396 | de = de2; |
7c06a8dc | 1397 | de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de); |
1da177e4 LT |
1398 | /* Initialize the root; the dot dirents already exist */ |
1399 | de = (struct ext3_dir_entry_2 *) (&root->dotdot); | |
7c06a8dc | 1400 | de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2)); |
1da177e4 LT |
1401 | memset (&root->info, 0, sizeof(root->info)); |
1402 | root->info.info_length = sizeof(root->info); | |
1403 | root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version; | |
1404 | entries = root->entries; | |
1405 | dx_set_block (entries, 1); | |
1406 | dx_set_count (entries, 1); | |
1407 | dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info))); | |
1408 | ||
1409 | /* Initialize as for dx_probe */ | |
1410 | hinfo.hash_version = root->info.hash_version; | |
1411 | hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed; | |
1412 | ext3fs_dirhash(name, namelen, &hinfo); | |
1413 | frame = frames; | |
1414 | frame->entries = entries; | |
1415 | frame->at = entries; | |
1416 | frame->bh = bh; | |
1417 | bh = bh2; | |
1418 | de = do_split(handle,dir, &bh, frame, &hinfo, &retval); | |
1419 | dx_release (frames); | |
1420 | if (!(de)) | |
1421 | return retval; | |
1422 | ||
1423 | return add_dirent_to_buf(handle, dentry, inode, de, bh); | |
1424 | } | |
1da177e4 LT |
1425 | |
1426 | /* | |
1427 | * ext3_add_entry() | |
1428 | * | |
1429 | * adds a file entry to the specified directory, using the same | |
1430 | * semantics as ext3_find_entry(). It returns NULL if it failed. | |
1431 | * | |
1432 | * NOTE!! The inode part of 'de' is left at 0 - which means you | |
1433 | * may not sleep between calling this and putting something into | |
1434 | * the entry, as someone else might have used it while you slept. | |
1435 | */ | |
1436 | static int ext3_add_entry (handle_t *handle, struct dentry *dentry, | |
1437 | struct inode *inode) | |
1438 | { | |
1439 | struct inode *dir = dentry->d_parent->d_inode; | |
1440 | unsigned long offset; | |
1441 | struct buffer_head * bh; | |
1442 | struct ext3_dir_entry_2 *de; | |
1443 | struct super_block * sb; | |
1444 | int retval; | |
1da177e4 | 1445 | int dx_fallback=0; |
1da177e4 | 1446 | unsigned blocksize; |
1da177e4 LT |
1447 | u32 block, blocks; |
1448 | ||
1449 | sb = dir->i_sb; | |
1450 | blocksize = sb->s_blocksize; | |
1451 | if (!dentry->d_name.len) | |
1452 | return -EINVAL; | |
1da177e4 LT |
1453 | if (is_dx(dir)) { |
1454 | retval = ext3_dx_add_entry(handle, dentry, inode); | |
1455 | if (!retval || (retval != ERR_BAD_DX_DIR)) | |
1456 | return retval; | |
1457 | EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL; | |
1458 | dx_fallback++; | |
1459 | ext3_mark_inode_dirty(handle, dir); | |
1460 | } | |
1da177e4 LT |
1461 | blocks = dir->i_size >> sb->s_blocksize_bits; |
1462 | for (block = 0, offset = 0; block < blocks; block++) { | |
1463 | bh = ext3_bread(handle, dir, block, 0, &retval); | |
1464 | if(!bh) | |
1465 | return retval; | |
1466 | retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
1467 | if (retval != -ENOSPC) | |
1468 | return retval; | |
1469 | ||
1da177e4 LT |
1470 | if (blocks == 1 && !dx_fallback && |
1471 | EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX)) | |
1472 | return make_indexed_dir(handle, dentry, inode, bh); | |
1da177e4 LT |
1473 | brelse(bh); |
1474 | } | |
1475 | bh = ext3_append(handle, dir, &block, &retval); | |
1476 | if (!bh) | |
1477 | return retval; | |
1478 | de = (struct ext3_dir_entry_2 *) bh->b_data; | |
1479 | de->inode = 0; | |
7c06a8dc | 1480 | de->rec_len = ext3_rec_len_to_disk(blocksize); |
1da177e4 LT |
1481 | return add_dirent_to_buf(handle, dentry, inode, de, bh); |
1482 | } | |
1483 | ||
1da177e4 LT |
1484 | /* |
1485 | * Returns 0 for success, or a negative error value | |
1486 | */ | |
1487 | static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry, | |
1488 | struct inode *inode) | |
1489 | { | |
1490 | struct dx_frame frames[2], *frame; | |
1491 | struct dx_entry *entries, *at; | |
1492 | struct dx_hash_info hinfo; | |
1493 | struct buffer_head * bh; | |
1494 | struct inode *dir = dentry->d_parent->d_inode; | |
1495 | struct super_block * sb = dir->i_sb; | |
1496 | struct ext3_dir_entry_2 *de; | |
1497 | int err; | |
1498 | ||
1499 | frame = dx_probe(dentry, NULL, &hinfo, frames, &err); | |
1500 | if (!frame) | |
1501 | return err; | |
1502 | entries = frame->entries; | |
1503 | at = frame->at; | |
1504 | ||
1505 | if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err))) | |
1506 | goto cleanup; | |
1507 | ||
1508 | BUFFER_TRACE(bh, "get_write_access"); | |
1509 | err = ext3_journal_get_write_access(handle, bh); | |
1510 | if (err) | |
1511 | goto journal_error; | |
1512 | ||
1513 | err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
1514 | if (err != -ENOSPC) { | |
1515 | bh = NULL; | |
1516 | goto cleanup; | |
1517 | } | |
1518 | ||
1519 | /* Block full, should compress but for now just split */ | |
1520 | dxtrace(printk("using %u of %u node entries\n", | |
1521 | dx_get_count(entries), dx_get_limit(entries))); | |
1522 | /* Need to split index? */ | |
1523 | if (dx_get_count(entries) == dx_get_limit(entries)) { | |
1524 | u32 newblock; | |
1525 | unsigned icount = dx_get_count(entries); | |
1526 | int levels = frame - frames; | |
1527 | struct dx_entry *entries2; | |
1528 | struct dx_node *node2; | |
1529 | struct buffer_head *bh2; | |
1530 | ||
1531 | if (levels && (dx_get_count(frames->entries) == | |
1532 | dx_get_limit(frames->entries))) { | |
1533 | ext3_warning(sb, __FUNCTION__, | |
9f40668d | 1534 | "Directory index full!"); |
1da177e4 LT |
1535 | err = -ENOSPC; |
1536 | goto cleanup; | |
1537 | } | |
1538 | bh2 = ext3_append (handle, dir, &newblock, &err); | |
1539 | if (!(bh2)) | |
1540 | goto cleanup; | |
1541 | node2 = (struct dx_node *)(bh2->b_data); | |
1542 | entries2 = node2->entries; | |
7c06a8dc | 1543 | node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize); |
1da177e4 LT |
1544 | node2->fake.inode = 0; |
1545 | BUFFER_TRACE(frame->bh, "get_write_access"); | |
1546 | err = ext3_journal_get_write_access(handle, frame->bh); | |
1547 | if (err) | |
1548 | goto journal_error; | |
1549 | if (levels) { | |
1550 | unsigned icount1 = icount/2, icount2 = icount - icount1; | |
1551 | unsigned hash2 = dx_get_hash(entries + icount1); | |
1552 | dxtrace(printk("Split index %i/%i\n", icount1, icount2)); | |
1553 | ||
1554 | BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ | |
1555 | err = ext3_journal_get_write_access(handle, | |
1556 | frames[0].bh); | |
1557 | if (err) | |
1558 | goto journal_error; | |
1559 | ||
1560 | memcpy ((char *) entries2, (char *) (entries + icount1), | |
1561 | icount2 * sizeof(struct dx_entry)); | |
1562 | dx_set_count (entries, icount1); | |
1563 | dx_set_count (entries2, icount2); | |
1564 | dx_set_limit (entries2, dx_node_limit(dir)); | |
1565 | ||
1566 | /* Which index block gets the new entry? */ | |
1567 | if (at - entries >= icount1) { | |
1568 | frame->at = at = at - entries - icount1 + entries2; | |
1569 | frame->entries = entries = entries2; | |
1570 | swap(frame->bh, bh2); | |
1571 | } | |
1572 | dx_insert_block (frames + 0, hash2, newblock); | |
1573 | dxtrace(dx_show_index ("node", frames[1].entries)); | |
1574 | dxtrace(dx_show_index ("node", | |
1575 | ((struct dx_node *) bh2->b_data)->entries)); | |
1576 | err = ext3_journal_dirty_metadata(handle, bh2); | |
1577 | if (err) | |
1578 | goto journal_error; | |
1579 | brelse (bh2); | |
1580 | } else { | |
1581 | dxtrace(printk("Creating second level index...\n")); | |
1582 | memcpy((char *) entries2, (char *) entries, | |
1583 | icount * sizeof(struct dx_entry)); | |
1584 | dx_set_limit(entries2, dx_node_limit(dir)); | |
1585 | ||
1586 | /* Set up root */ | |
1587 | dx_set_count(entries, 1); | |
1588 | dx_set_block(entries + 0, newblock); | |
1589 | ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; | |
1590 | ||
1591 | /* Add new access path frame */ | |
1592 | frame = frames + 1; | |
1593 | frame->at = at = at - entries + entries2; | |
1594 | frame->entries = entries = entries2; | |
1595 | frame->bh = bh2; | |
1596 | err = ext3_journal_get_write_access(handle, | |
1597 | frame->bh); | |
1598 | if (err) | |
1599 | goto journal_error; | |
1600 | } | |
1601 | ext3_journal_dirty_metadata(handle, frames[0].bh); | |
1602 | } | |
1603 | de = do_split(handle, dir, &bh, frame, &hinfo, &err); | |
1604 | if (!de) | |
1605 | goto cleanup; | |
1606 | err = add_dirent_to_buf(handle, dentry, inode, de, bh); | |
1607 | bh = NULL; | |
1608 | goto cleanup; | |
1609 | ||
1610 | journal_error: | |
1611 | ext3_std_error(dir->i_sb, err); | |
1612 | cleanup: | |
1613 | if (bh) | |
1614 | brelse(bh); | |
1615 | dx_release(frames); | |
1616 | return err; | |
1617 | } | |
1da177e4 LT |
1618 | |
1619 | /* | |
1620 | * ext3_delete_entry deletes a directory entry by merging it with the | |
1621 | * previous entry | |
1622 | */ | |
ae6ddcc5 | 1623 | static int ext3_delete_entry (handle_t *handle, |
1da177e4 LT |
1624 | struct inode * dir, |
1625 | struct ext3_dir_entry_2 * de_del, | |
1626 | struct buffer_head * bh) | |
1627 | { | |
1628 | struct ext3_dir_entry_2 * de, * pde; | |
1629 | int i; | |
1630 | ||
1631 | i = 0; | |
1632 | pde = NULL; | |
1633 | de = (struct ext3_dir_entry_2 *) bh->b_data; | |
1634 | while (i < bh->b_size) { | |
1635 | if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i)) | |
1636 | return -EIO; | |
1637 | if (de == de_del) { | |
1638 | BUFFER_TRACE(bh, "get_write_access"); | |
1639 | ext3_journal_get_write_access(handle, bh); | |
1640 | if (pde) | |
7c06a8dc JK |
1641 | pde->rec_len = ext3_rec_len_to_disk( |
1642 | ext3_rec_len_from_disk(pde->rec_len) + | |
1643 | ext3_rec_len_from_disk(de->rec_len)); | |
1da177e4 LT |
1644 | else |
1645 | de->inode = 0; | |
1646 | dir->i_version++; | |
1647 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); | |
1648 | ext3_journal_dirty_metadata(handle, bh); | |
1649 | return 0; | |
1650 | } | |
7c06a8dc | 1651 | i += ext3_rec_len_from_disk(de->rec_len); |
1da177e4 | 1652 | pde = de; |
7c06a8dc | 1653 | de = ext3_next_entry(de); |
1da177e4 LT |
1654 | } |
1655 | return -ENOENT; | |
1656 | } | |
1657 | ||
1da177e4 LT |
1658 | static int ext3_add_nondir(handle_t *handle, |
1659 | struct dentry *dentry, struct inode *inode) | |
1660 | { | |
1661 | int err = ext3_add_entry(handle, dentry, inode); | |
1662 | if (!err) { | |
1663 | ext3_mark_inode_dirty(handle, inode); | |
1664 | d_instantiate(dentry, inode); | |
1665 | return 0; | |
1666 | } | |
731b9a54 | 1667 | drop_nlink(inode); |
1da177e4 LT |
1668 | iput(inode); |
1669 | return err; | |
1670 | } | |
1671 | ||
1672 | /* | |
1673 | * By the time this is called, we already have created | |
1674 | * the directory cache entry for the new file, but it | |
1675 | * is so far negative - it has no inode. | |
1676 | * | |
1677 | * If the create succeeds, we fill in the inode information | |
ae6ddcc5 | 1678 | * with d_instantiate(). |
1da177e4 LT |
1679 | */ |
1680 | static int ext3_create (struct inode * dir, struct dentry * dentry, int mode, | |
1681 | struct nameidata *nd) | |
1682 | { | |
ae6ddcc5 | 1683 | handle_t *handle; |
1da177e4 LT |
1684 | struct inode * inode; |
1685 | int err, retries = 0; | |
1686 | ||
1687 | retry: | |
1f54587b | 1688 | handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + |
1da177e4 | 1689 | EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + |
1f54587b | 1690 | 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); |
1da177e4 LT |
1691 | if (IS_ERR(handle)) |
1692 | return PTR_ERR(handle); | |
1693 | ||
1694 | if (IS_DIRSYNC(dir)) | |
1695 | handle->h_sync = 1; | |
1696 | ||
1697 | inode = ext3_new_inode (handle, dir, mode); | |
1698 | err = PTR_ERR(inode); | |
1699 | if (!IS_ERR(inode)) { | |
1700 | inode->i_op = &ext3_file_inode_operations; | |
1701 | inode->i_fop = &ext3_file_operations; | |
1702 | ext3_set_aops(inode); | |
1703 | err = ext3_add_nondir(handle, dentry, inode); | |
1704 | } | |
1705 | ext3_journal_stop(handle); | |
1706 | if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) | |
1707 | goto retry; | |
1708 | return err; | |
1709 | } | |
1710 | ||
1711 | static int ext3_mknod (struct inode * dir, struct dentry *dentry, | |
1712 | int mode, dev_t rdev) | |
1713 | { | |
1714 | handle_t *handle; | |
1715 | struct inode *inode; | |
1716 | int err, retries = 0; | |
1717 | ||
1718 | if (!new_valid_dev(rdev)) | |
1719 | return -EINVAL; | |
1720 | ||
1721 | retry: | |
1f54587b | 1722 | handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + |
e9ad5620 | 1723 | EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + |
1f54587b | 1724 | 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); |
1da177e4 LT |
1725 | if (IS_ERR(handle)) |
1726 | return PTR_ERR(handle); | |
1727 | ||
1728 | if (IS_DIRSYNC(dir)) | |
1729 | handle->h_sync = 1; | |
1730 | ||
1731 | inode = ext3_new_inode (handle, dir, mode); | |
1732 | err = PTR_ERR(inode); | |
1733 | if (!IS_ERR(inode)) { | |
1734 | init_special_inode(inode, inode->i_mode, rdev); | |
1735 | #ifdef CONFIG_EXT3_FS_XATTR | |
1736 | inode->i_op = &ext3_special_inode_operations; | |
1737 | #endif | |
1738 | err = ext3_add_nondir(handle, dentry, inode); | |
1739 | } | |
1740 | ext3_journal_stop(handle); | |
1741 | if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) | |
1742 | goto retry; | |
1743 | return err; | |
1744 | } | |
1745 | ||
1746 | static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode) | |
1747 | { | |
1748 | handle_t *handle; | |
1749 | struct inode * inode; | |
1750 | struct buffer_head * dir_block; | |
1751 | struct ext3_dir_entry_2 * de; | |
1752 | int err, retries = 0; | |
1753 | ||
1754 | if (dir->i_nlink >= EXT3_LINK_MAX) | |
1755 | return -EMLINK; | |
1756 | ||
1757 | retry: | |
1f54587b | 1758 | handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + |
1da177e4 | 1759 | EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 + |
1f54587b | 1760 | 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); |
1da177e4 LT |
1761 | if (IS_ERR(handle)) |
1762 | return PTR_ERR(handle); | |
1763 | ||
1764 | if (IS_DIRSYNC(dir)) | |
1765 | handle->h_sync = 1; | |
1766 | ||
1767 | inode = ext3_new_inode (handle, dir, S_IFDIR | mode); | |
1768 | err = PTR_ERR(inode); | |
1769 | if (IS_ERR(inode)) | |
1770 | goto out_stop; | |
1771 | ||
1772 | inode->i_op = &ext3_dir_inode_operations; | |
1773 | inode->i_fop = &ext3_dir_operations; | |
1774 | inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize; | |
1775 | dir_block = ext3_bread (handle, inode, 0, 1, &err); | |
1776 | if (!dir_block) { | |
9a53c3a7 | 1777 | drop_nlink(inode); /* is this nlink == 0? */ |
1da177e4 LT |
1778 | ext3_mark_inode_dirty(handle, inode); |
1779 | iput (inode); | |
1780 | goto out_stop; | |
1781 | } | |
1782 | BUFFER_TRACE(dir_block, "get_write_access"); | |
1783 | ext3_journal_get_write_access(handle, dir_block); | |
1784 | de = (struct ext3_dir_entry_2 *) dir_block->b_data; | |
1785 | de->inode = cpu_to_le32(inode->i_ino); | |
1786 | de->name_len = 1; | |
7c06a8dc | 1787 | de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len)); |
1da177e4 LT |
1788 | strcpy (de->name, "."); |
1789 | ext3_set_de_type(dir->i_sb, de, S_IFDIR); | |
7c06a8dc | 1790 | de = ext3_next_entry(de); |
1da177e4 | 1791 | de->inode = cpu_to_le32(dir->i_ino); |
7c06a8dc JK |
1792 | de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize - |
1793 | EXT3_DIR_REC_LEN(1)); | |
1da177e4 LT |
1794 | de->name_len = 2; |
1795 | strcpy (de->name, ".."); | |
1796 | ext3_set_de_type(dir->i_sb, de, S_IFDIR); | |
1797 | inode->i_nlink = 2; | |
1798 | BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata"); | |
1799 | ext3_journal_dirty_metadata(handle, dir_block); | |
1800 | brelse (dir_block); | |
1801 | ext3_mark_inode_dirty(handle, inode); | |
1802 | err = ext3_add_entry (handle, dentry, inode); | |
1803 | if (err) { | |
1804 | inode->i_nlink = 0; | |
1805 | ext3_mark_inode_dirty(handle, inode); | |
1806 | iput (inode); | |
1807 | goto out_stop; | |
1808 | } | |
d8c76e6f | 1809 | inc_nlink(dir); |
1da177e4 LT |
1810 | ext3_update_dx_flag(dir); |
1811 | ext3_mark_inode_dirty(handle, dir); | |
1812 | d_instantiate(dentry, inode); | |
1813 | out_stop: | |
1814 | ext3_journal_stop(handle); | |
1815 | if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) | |
1816 | goto retry; | |
1817 | return err; | |
1818 | } | |
1819 | ||
1820 | /* | |
1821 | * routine to check that the specified directory is empty (for rmdir) | |
1822 | */ | |
1823 | static int empty_dir (struct inode * inode) | |
1824 | { | |
1825 | unsigned long offset; | |
1826 | struct buffer_head * bh; | |
1827 | struct ext3_dir_entry_2 * de, * de1; | |
1828 | struct super_block * sb; | |
1829 | int err = 0; | |
1830 | ||
1831 | sb = inode->i_sb; | |
1832 | if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) || | |
1833 | !(bh = ext3_bread (NULL, inode, 0, 0, &err))) { | |
1834 | if (err) | |
1835 | ext3_error(inode->i_sb, __FUNCTION__, | |
1836 | "error %d reading directory #%lu offset 0", | |
1837 | err, inode->i_ino); | |
1838 | else | |
1839 | ext3_warning(inode->i_sb, __FUNCTION__, | |
1840 | "bad directory (dir #%lu) - no data block", | |
1841 | inode->i_ino); | |
1842 | return 1; | |
1843 | } | |
1844 | de = (struct ext3_dir_entry_2 *) bh->b_data; | |
7c06a8dc | 1845 | de1 = ext3_next_entry(de); |
1da177e4 | 1846 | if (le32_to_cpu(de->inode) != inode->i_ino || |
ae6ddcc5 | 1847 | !le32_to_cpu(de1->inode) || |
1da177e4 LT |
1848 | strcmp (".", de->name) || |
1849 | strcmp ("..", de1->name)) { | |
e9ad5620 | 1850 | ext3_warning (inode->i_sb, "empty_dir", |
1da177e4 LT |
1851 | "bad directory (dir #%lu) - no `.' or `..'", |
1852 | inode->i_ino); | |
1853 | brelse (bh); | |
1854 | return 1; | |
1855 | } | |
7c06a8dc JK |
1856 | offset = ext3_rec_len_from_disk(de->rec_len) + |
1857 | ext3_rec_len_from_disk(de1->rec_len); | |
1858 | de = ext3_next_entry(de1); | |
1da177e4 LT |
1859 | while (offset < inode->i_size ) { |
1860 | if (!bh || | |
1861 | (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { | |
1862 | err = 0; | |
1863 | brelse (bh); | |
1864 | bh = ext3_bread (NULL, inode, | |
1865 | offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err); | |
1866 | if (!bh) { | |
1867 | if (err) | |
1868 | ext3_error(sb, __FUNCTION__, | |
1869 | "error %d reading directory" | |
1870 | " #%lu offset %lu", | |
1871 | err, inode->i_ino, offset); | |
1872 | offset += sb->s_blocksize; | |
1873 | continue; | |
1874 | } | |
1875 | de = (struct ext3_dir_entry_2 *) bh->b_data; | |
1876 | } | |
1877 | if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) { | |
1878 | de = (struct ext3_dir_entry_2 *)(bh->b_data + | |
1879 | sb->s_blocksize); | |
1880 | offset = (offset | (sb->s_blocksize - 1)) + 1; | |
1881 | continue; | |
1882 | } | |
1883 | if (le32_to_cpu(de->inode)) { | |
1884 | brelse (bh); | |
1885 | return 0; | |
1886 | } | |
7c06a8dc JK |
1887 | offset += ext3_rec_len_from_disk(de->rec_len); |
1888 | de = ext3_next_entry(de); | |
1da177e4 LT |
1889 | } |
1890 | brelse (bh); | |
1891 | return 1; | |
1892 | } | |
1893 | ||
1894 | /* ext3_orphan_add() links an unlinked or truncated inode into a list of | |
1895 | * such inodes, starting at the superblock, in case we crash before the | |
1896 | * file is closed/deleted, or in case the inode truncate spans multiple | |
1897 | * transactions and the last transaction is not recovered after a crash. | |
1898 | * | |
1899 | * At filesystem recovery time, we walk this list deleting unlinked | |
1900 | * inodes and truncating linked inodes in ext3_orphan_cleanup(). | |
1901 | */ | |
1902 | int ext3_orphan_add(handle_t *handle, struct inode *inode) | |
1903 | { | |
1904 | struct super_block *sb = inode->i_sb; | |
1905 | struct ext3_iloc iloc; | |
1906 | int err = 0, rc; | |
1907 | ||
1908 | lock_super(sb); | |
1909 | if (!list_empty(&EXT3_I(inode)->i_orphan)) | |
1910 | goto out_unlock; | |
1911 | ||
1912 | /* Orphan handling is only valid for files with data blocks | |
1913 | * being truncated, or files being unlinked. */ | |
1914 | ||
1915 | /* @@@ FIXME: Observation from aviro: | |
ae6ddcc5 | 1916 | * I think I can trigger J_ASSERT in ext3_orphan_add(). We block |
1da177e4 LT |
1917 | * here (on lock_super()), so race with ext3_link() which might bump |
1918 | * ->i_nlink. For, say it, character device. Not a regular file, | |
1919 | * not a directory, not a symlink and ->i_nlink > 0. | |
1920 | */ | |
1921 | J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
1922 | S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); | |
1923 | ||
1924 | BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access"); | |
1925 | err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh); | |
1926 | if (err) | |
1927 | goto out_unlock; | |
1928 | ||
1929 | err = ext3_reserve_inode_write(handle, inode, &iloc); | |
1930 | if (err) | |
1931 | goto out_unlock; | |
1932 | ||
1933 | /* Insert this inode at the head of the on-disk orphan list... */ | |
1934 | NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan); | |
1935 | EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); | |
1936 | err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh); | |
1937 | rc = ext3_mark_iloc_dirty(handle, inode, &iloc); | |
1938 | if (!err) | |
1939 | err = rc; | |
1940 | ||
1941 | /* Only add to the head of the in-memory list if all the | |
1942 | * previous operations succeeded. If the orphan_add is going to | |
1943 | * fail (possibly taking the journal offline), we can't risk | |
1944 | * leaving the inode on the orphan list: stray orphan-list | |
1945 | * entries can cause panics at unmount time. | |
1946 | * | |
1947 | * This is safe: on error we're going to ignore the orphan list | |
1948 | * anyway on the next recovery. */ | |
1949 | if (!err) | |
1950 | list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan); | |
1951 | ||
eee194e7 ES |
1952 | jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); |
1953 | jbd_debug(4, "orphan inode %lu will point to %d\n", | |
1da177e4 LT |
1954 | inode->i_ino, NEXT_ORPHAN(inode)); |
1955 | out_unlock: | |
1956 | unlock_super(sb); | |
1957 | ext3_std_error(inode->i_sb, err); | |
1958 | return err; | |
1959 | } | |
1960 | ||
1961 | /* | |
1962 | * ext3_orphan_del() removes an unlinked or truncated inode from the list | |
1963 | * of such inodes stored on disk, because it is finally being cleaned up. | |
1964 | */ | |
1965 | int ext3_orphan_del(handle_t *handle, struct inode *inode) | |
1966 | { | |
1967 | struct list_head *prev; | |
1968 | struct ext3_inode_info *ei = EXT3_I(inode); | |
1969 | struct ext3_sb_info *sbi; | |
1970 | unsigned long ino_next; | |
1971 | struct ext3_iloc iloc; | |
1972 | int err = 0; | |
1973 | ||
1974 | lock_super(inode->i_sb); | |
1975 | if (list_empty(&ei->i_orphan)) { | |
1976 | unlock_super(inode->i_sb); | |
1977 | return 0; | |
1978 | } | |
1979 | ||
1980 | ino_next = NEXT_ORPHAN(inode); | |
1981 | prev = ei->i_orphan.prev; | |
1982 | sbi = EXT3_SB(inode->i_sb); | |
1983 | ||
1984 | jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); | |
1985 | ||
1986 | list_del_init(&ei->i_orphan); | |
1987 | ||
1988 | /* If we're on an error path, we may not have a valid | |
1989 | * transaction handle with which to update the orphan list on | |
1990 | * disk, but we still need to remove the inode from the linked | |
1991 | * list in memory. */ | |
1992 | if (!handle) | |
1993 | goto out; | |
1994 | ||
1995 | err = ext3_reserve_inode_write(handle, inode, &iloc); | |
1996 | if (err) | |
1997 | goto out_err; | |
1998 | ||
1999 | if (prev == &sbi->s_orphan) { | |
2000 | jbd_debug(4, "superblock will point to %lu\n", ino_next); | |
2001 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); | |
2002 | err = ext3_journal_get_write_access(handle, sbi->s_sbh); | |
2003 | if (err) | |
2004 | goto out_brelse; | |
2005 | sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); | |
2006 | err = ext3_journal_dirty_metadata(handle, sbi->s_sbh); | |
2007 | } else { | |
2008 | struct ext3_iloc iloc2; | |
2009 | struct inode *i_prev = | |
2010 | &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode; | |
2011 | ||
2012 | jbd_debug(4, "orphan inode %lu will point to %lu\n", | |
2013 | i_prev->i_ino, ino_next); | |
2014 | err = ext3_reserve_inode_write(handle, i_prev, &iloc2); | |
2015 | if (err) | |
2016 | goto out_brelse; | |
2017 | NEXT_ORPHAN(i_prev) = ino_next; | |
2018 | err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2); | |
2019 | } | |
2020 | if (err) | |
2021 | goto out_brelse; | |
2022 | NEXT_ORPHAN(inode) = 0; | |
2023 | err = ext3_mark_iloc_dirty(handle, inode, &iloc); | |
2024 | ||
2025 | out_err: | |
2026 | ext3_std_error(inode->i_sb, err); | |
2027 | out: | |
2028 | unlock_super(inode->i_sb); | |
2029 | return err; | |
2030 | ||
2031 | out_brelse: | |
2032 | brelse(iloc.bh); | |
2033 | goto out_err; | |
2034 | } | |
2035 | ||
2036 | static int ext3_rmdir (struct inode * dir, struct dentry *dentry) | |
2037 | { | |
2038 | int retval; | |
2039 | struct inode * inode; | |
2040 | struct buffer_head * bh; | |
2041 | struct ext3_dir_entry_2 * de; | |
2042 | handle_t *handle; | |
2043 | ||
2044 | /* Initialize quotas before so that eventual writes go in | |
2045 | * separate transaction */ | |
2046 | DQUOT_INIT(dentry->d_inode); | |
1f54587b | 2047 | handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb)); |
1da177e4 LT |
2048 | if (IS_ERR(handle)) |
2049 | return PTR_ERR(handle); | |
2050 | ||
2051 | retval = -ENOENT; | |
2052 | bh = ext3_find_entry (dentry, &de); | |
2053 | if (!bh) | |
2054 | goto end_rmdir; | |
2055 | ||
2056 | if (IS_DIRSYNC(dir)) | |
2057 | handle->h_sync = 1; | |
2058 | ||
2059 | inode = dentry->d_inode; | |
2060 | ||
2061 | retval = -EIO; | |
2062 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2063 | goto end_rmdir; | |
2064 | ||
2065 | retval = -ENOTEMPTY; | |
2066 | if (!empty_dir (inode)) | |
2067 | goto end_rmdir; | |
2068 | ||
2069 | retval = ext3_delete_entry(handle, dir, de, bh); | |
2070 | if (retval) | |
2071 | goto end_rmdir; | |
2072 | if (inode->i_nlink != 2) | |
2073 | ext3_warning (inode->i_sb, "ext3_rmdir", | |
2074 | "empty directory has nlink!=2 (%d)", | |
2075 | inode->i_nlink); | |
2076 | inode->i_version++; | |
ce71ec36 | 2077 | clear_nlink(inode); |
1da177e4 LT |
2078 | /* There's no need to set i_disksize: the fact that i_nlink is |
2079 | * zero will ensure that the right thing happens during any | |
2080 | * recovery. */ | |
2081 | inode->i_size = 0; | |
2082 | ext3_orphan_add(handle, inode); | |
2083 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC; | |
2084 | ext3_mark_inode_dirty(handle, inode); | |
9a53c3a7 | 2085 | drop_nlink(dir); |
1da177e4 LT |
2086 | ext3_update_dx_flag(dir); |
2087 | ext3_mark_inode_dirty(handle, dir); | |
2088 | ||
2089 | end_rmdir: | |
2090 | ext3_journal_stop(handle); | |
2091 | brelse (bh); | |
2092 | return retval; | |
2093 | } | |
2094 | ||
2095 | static int ext3_unlink(struct inode * dir, struct dentry *dentry) | |
2096 | { | |
2097 | int retval; | |
2098 | struct inode * inode; | |
2099 | struct buffer_head * bh; | |
2100 | struct ext3_dir_entry_2 * de; | |
2101 | handle_t *handle; | |
2102 | ||
2103 | /* Initialize quotas before so that eventual writes go | |
2104 | * in separate transaction */ | |
2105 | DQUOT_INIT(dentry->d_inode); | |
1f54587b | 2106 | handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb)); |
1da177e4 LT |
2107 | if (IS_ERR(handle)) |
2108 | return PTR_ERR(handle); | |
2109 | ||
2110 | if (IS_DIRSYNC(dir)) | |
2111 | handle->h_sync = 1; | |
2112 | ||
2113 | retval = -ENOENT; | |
2114 | bh = ext3_find_entry (dentry, &de); | |
2115 | if (!bh) | |
2116 | goto end_unlink; | |
2117 | ||
2118 | inode = dentry->d_inode; | |
2119 | ||
2120 | retval = -EIO; | |
2121 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2122 | goto end_unlink; | |
2123 | ||
2124 | if (!inode->i_nlink) { | |
2125 | ext3_warning (inode->i_sb, "ext3_unlink", | |
2126 | "Deleting nonexistent file (%lu), %d", | |
2127 | inode->i_ino, inode->i_nlink); | |
2128 | inode->i_nlink = 1; | |
2129 | } | |
2130 | retval = ext3_delete_entry(handle, dir, de, bh); | |
2131 | if (retval) | |
2132 | goto end_unlink; | |
2133 | dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC; | |
2134 | ext3_update_dx_flag(dir); | |
2135 | ext3_mark_inode_dirty(handle, dir); | |
9a53c3a7 | 2136 | drop_nlink(inode); |
1da177e4 LT |
2137 | if (!inode->i_nlink) |
2138 | ext3_orphan_add(handle, inode); | |
2139 | inode->i_ctime = dir->i_ctime; | |
2140 | ext3_mark_inode_dirty(handle, inode); | |
2141 | retval = 0; | |
2142 | ||
2143 | end_unlink: | |
2144 | ext3_journal_stop(handle); | |
2145 | brelse (bh); | |
2146 | return retval; | |
2147 | } | |
2148 | ||
2149 | static int ext3_symlink (struct inode * dir, | |
2150 | struct dentry *dentry, const char * symname) | |
2151 | { | |
2152 | handle_t *handle; | |
2153 | struct inode * inode; | |
2154 | int l, err, retries = 0; | |
2155 | ||
2156 | l = strlen(symname)+1; | |
2157 | if (l > dir->i_sb->s_blocksize) | |
2158 | return -ENAMETOOLONG; | |
2159 | ||
2160 | retry: | |
1f54587b | 2161 | handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + |
e9ad5620 | 2162 | EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 + |
1f54587b | 2163 | 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb)); |
1da177e4 LT |
2164 | if (IS_ERR(handle)) |
2165 | return PTR_ERR(handle); | |
2166 | ||
2167 | if (IS_DIRSYNC(dir)) | |
2168 | handle->h_sync = 1; | |
2169 | ||
2170 | inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO); | |
2171 | err = PTR_ERR(inode); | |
2172 | if (IS_ERR(inode)) | |
2173 | goto out_stop; | |
2174 | ||
2175 | if (l > sizeof (EXT3_I(inode)->i_data)) { | |
2176 | inode->i_op = &ext3_symlink_inode_operations; | |
2177 | ext3_set_aops(inode); | |
2178 | /* | |
2179 | * page_symlink() calls into ext3_prepare/commit_write. | |
2180 | * We have a transaction open. All is sweetness. It also sets | |
2181 | * i_size in generic_commit_write(). | |
2182 | */ | |
0adb25d2 KK |
2183 | err = __page_symlink(inode, symname, l, |
2184 | mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS); | |
1da177e4 | 2185 | if (err) { |
731b9a54 | 2186 | drop_nlink(inode); |
1da177e4 LT |
2187 | ext3_mark_inode_dirty(handle, inode); |
2188 | iput (inode); | |
2189 | goto out_stop; | |
2190 | } | |
2191 | } else { | |
2192 | inode->i_op = &ext3_fast_symlink_inode_operations; | |
2193 | memcpy((char*)&EXT3_I(inode)->i_data,symname,l); | |
2194 | inode->i_size = l-1; | |
2195 | } | |
2196 | EXT3_I(inode)->i_disksize = inode->i_size; | |
2197 | err = ext3_add_nondir(handle, dentry, inode); | |
2198 | out_stop: | |
2199 | ext3_journal_stop(handle); | |
2200 | if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) | |
2201 | goto retry; | |
2202 | return err; | |
2203 | } | |
2204 | ||
2205 | static int ext3_link (struct dentry * old_dentry, | |
2206 | struct inode * dir, struct dentry *dentry) | |
2207 | { | |
2208 | handle_t *handle; | |
2209 | struct inode *inode = old_dentry->d_inode; | |
2210 | int err, retries = 0; | |
2211 | ||
2212 | if (inode->i_nlink >= EXT3_LINK_MAX) | |
2213 | return -EMLINK; | |
2988a774 ES |
2214 | /* |
2215 | * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing | |
2216 | * otherwise has the potential to corrupt the orphan inode list. | |
2217 | */ | |
2218 | if (inode->i_nlink == 0) | |
2219 | return -ENOENT; | |
1da177e4 LT |
2220 | |
2221 | retry: | |
1f54587b | 2222 | handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) + |
1da177e4 LT |
2223 | EXT3_INDEX_EXTRA_TRANS_BLOCKS); |
2224 | if (IS_ERR(handle)) | |
2225 | return PTR_ERR(handle); | |
2226 | ||
2227 | if (IS_DIRSYNC(dir)) | |
2228 | handle->h_sync = 1; | |
2229 | ||
2230 | inode->i_ctime = CURRENT_TIME_SEC; | |
731b9a54 | 2231 | inc_nlink(inode); |
1da177e4 LT |
2232 | atomic_inc(&inode->i_count); |
2233 | ||
2234 | err = ext3_add_nondir(handle, dentry, inode); | |
2235 | ext3_journal_stop(handle); | |
2236 | if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries)) | |
2237 | goto retry; | |
2238 | return err; | |
2239 | } | |
2240 | ||
2241 | #define PARENT_INO(buffer) \ | |
7c06a8dc | 2242 | (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode) |
1da177e4 LT |
2243 | |
2244 | /* | |
2245 | * Anybody can rename anything with this: the permission checks are left to the | |
2246 | * higher-level routines. | |
2247 | */ | |
2248 | static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry, | |
2249 | struct inode * new_dir,struct dentry *new_dentry) | |
2250 | { | |
2251 | handle_t *handle; | |
2252 | struct inode * old_inode, * new_inode; | |
2253 | struct buffer_head * old_bh, * new_bh, * dir_bh; | |
2254 | struct ext3_dir_entry_2 * old_de, * new_de; | |
2255 | int retval; | |
2256 | ||
2257 | old_bh = new_bh = dir_bh = NULL; | |
2258 | ||
2259 | /* Initialize quotas before so that eventual writes go | |
2260 | * in separate transaction */ | |
2261 | if (new_dentry->d_inode) | |
2262 | DQUOT_INIT(new_dentry->d_inode); | |
1f54587b JK |
2263 | handle = ext3_journal_start(old_dir, 2 * |
2264 | EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) + | |
e9ad5620 | 2265 | EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2); |
1da177e4 LT |
2266 | if (IS_ERR(handle)) |
2267 | return PTR_ERR(handle); | |
2268 | ||
2269 | if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir)) | |
2270 | handle->h_sync = 1; | |
2271 | ||
2272 | old_bh = ext3_find_entry (old_dentry, &old_de); | |
2273 | /* | |
2274 | * Check for inode number is _not_ due to possible IO errors. | |
2275 | * We might rmdir the source, keep it as pwd of some process | |
2276 | * and merrily kill the link to whatever was created under the | |
2277 | * same name. Goodbye sticky bit ;-< | |
2278 | */ | |
2279 | old_inode = old_dentry->d_inode; | |
2280 | retval = -ENOENT; | |
2281 | if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino) | |
2282 | goto end_rename; | |
2283 | ||
2284 | new_inode = new_dentry->d_inode; | |
2285 | new_bh = ext3_find_entry (new_dentry, &new_de); | |
2286 | if (new_bh) { | |
2287 | if (!new_inode) { | |
2288 | brelse (new_bh); | |
2289 | new_bh = NULL; | |
2290 | } | |
2291 | } | |
2292 | if (S_ISDIR(old_inode->i_mode)) { | |
2293 | if (new_inode) { | |
2294 | retval = -ENOTEMPTY; | |
2295 | if (!empty_dir (new_inode)) | |
2296 | goto end_rename; | |
2297 | } | |
2298 | retval = -EIO; | |
2299 | dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval); | |
2300 | if (!dir_bh) | |
2301 | goto end_rename; | |
2302 | if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino) | |
2303 | goto end_rename; | |
2304 | retval = -EMLINK; | |
2305 | if (!new_inode && new_dir!=old_dir && | |
2306 | new_dir->i_nlink >= EXT3_LINK_MAX) | |
2307 | goto end_rename; | |
2308 | } | |
2309 | if (!new_bh) { | |
2310 | retval = ext3_add_entry (handle, new_dentry, old_inode); | |
2311 | if (retval) | |
2312 | goto end_rename; | |
2313 | } else { | |
2314 | BUFFER_TRACE(new_bh, "get write access"); | |
2315 | ext3_journal_get_write_access(handle, new_bh); | |
2316 | new_de->inode = cpu_to_le32(old_inode->i_ino); | |
2317 | if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb, | |
2318 | EXT3_FEATURE_INCOMPAT_FILETYPE)) | |
2319 | new_de->file_type = old_de->file_type; | |
2320 | new_dir->i_version++; | |
2321 | BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata"); | |
2322 | ext3_journal_dirty_metadata(handle, new_bh); | |
2323 | brelse(new_bh); | |
2324 | new_bh = NULL; | |
2325 | } | |
2326 | ||
2327 | /* | |
2328 | * Like most other Unix systems, set the ctime for inodes on a | |
2329 | * rename. | |
2330 | */ | |
2331 | old_inode->i_ctime = CURRENT_TIME_SEC; | |
2332 | ext3_mark_inode_dirty(handle, old_inode); | |
2333 | ||
2334 | /* | |
2335 | * ok, that's it | |
2336 | */ | |
2337 | if (le32_to_cpu(old_de->inode) != old_inode->i_ino || | |
2338 | old_de->name_len != old_dentry->d_name.len || | |
2339 | strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) || | |
2340 | (retval = ext3_delete_entry(handle, old_dir, | |
2341 | old_de, old_bh)) == -ENOENT) { | |
2342 | /* old_de could have moved from under us during htree split, so | |
2343 | * make sure that we are deleting the right entry. We might | |
2344 | * also be pointing to a stale entry in the unused part of | |
2345 | * old_bh so just checking inum and the name isn't enough. */ | |
2346 | struct buffer_head *old_bh2; | |
2347 | struct ext3_dir_entry_2 *old_de2; | |
2348 | ||
2349 | old_bh2 = ext3_find_entry(old_dentry, &old_de2); | |
2350 | if (old_bh2) { | |
2351 | retval = ext3_delete_entry(handle, old_dir, | |
2352 | old_de2, old_bh2); | |
2353 | brelse(old_bh2); | |
2354 | } | |
2355 | } | |
2356 | if (retval) { | |
2357 | ext3_warning(old_dir->i_sb, "ext3_rename", | |
2358 | "Deleting old file (%lu), %d, error=%d", | |
2359 | old_dir->i_ino, old_dir->i_nlink, retval); | |
2360 | } | |
2361 | ||
2362 | if (new_inode) { | |
9a53c3a7 | 2363 | drop_nlink(new_inode); |
1da177e4 LT |
2364 | new_inode->i_ctime = CURRENT_TIME_SEC; |
2365 | } | |
2366 | old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC; | |
2367 | ext3_update_dx_flag(old_dir); | |
2368 | if (dir_bh) { | |
2369 | BUFFER_TRACE(dir_bh, "get_write_access"); | |
2370 | ext3_journal_get_write_access(handle, dir_bh); | |
2371 | PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino); | |
2372 | BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata"); | |
2373 | ext3_journal_dirty_metadata(handle, dir_bh); | |
9a53c3a7 | 2374 | drop_nlink(old_dir); |
1da177e4 | 2375 | if (new_inode) { |
9a53c3a7 | 2376 | drop_nlink(new_inode); |
1da177e4 | 2377 | } else { |
d8c76e6f | 2378 | inc_nlink(new_dir); |
1da177e4 LT |
2379 | ext3_update_dx_flag(new_dir); |
2380 | ext3_mark_inode_dirty(handle, new_dir); | |
2381 | } | |
2382 | } | |
2383 | ext3_mark_inode_dirty(handle, old_dir); | |
2384 | if (new_inode) { | |
2385 | ext3_mark_inode_dirty(handle, new_inode); | |
2386 | if (!new_inode->i_nlink) | |
2387 | ext3_orphan_add(handle, new_inode); | |
2388 | } | |
2389 | retval = 0; | |
2390 | ||
2391 | end_rename: | |
2392 | brelse (dir_bh); | |
2393 | brelse (old_bh); | |
2394 | brelse (new_bh); | |
2395 | ext3_journal_stop(handle); | |
2396 | return retval; | |
2397 | } | |
2398 | ||
2399 | /* | |
2400 | * directories can handle most operations... | |
2401 | */ | |
754661f1 | 2402 | const struct inode_operations ext3_dir_inode_operations = { |
1da177e4 LT |
2403 | .create = ext3_create, |
2404 | .lookup = ext3_lookup, | |
2405 | .link = ext3_link, | |
2406 | .unlink = ext3_unlink, | |
2407 | .symlink = ext3_symlink, | |
2408 | .mkdir = ext3_mkdir, | |
2409 | .rmdir = ext3_rmdir, | |
2410 | .mknod = ext3_mknod, | |
2411 | .rename = ext3_rename, | |
2412 | .setattr = ext3_setattr, | |
2413 | #ifdef CONFIG_EXT3_FS_XATTR | |
2414 | .setxattr = generic_setxattr, | |
2415 | .getxattr = generic_getxattr, | |
2416 | .listxattr = ext3_listxattr, | |
2417 | .removexattr = generic_removexattr, | |
2418 | #endif | |
2419 | .permission = ext3_permission, | |
2420 | }; | |
2421 | ||
754661f1 | 2422 | const struct inode_operations ext3_special_inode_operations = { |
1da177e4 LT |
2423 | .setattr = ext3_setattr, |
2424 | #ifdef CONFIG_EXT3_FS_XATTR | |
2425 | .setxattr = generic_setxattr, | |
2426 | .getxattr = generic_getxattr, | |
2427 | .listxattr = ext3_listxattr, | |
2428 | .removexattr = generic_removexattr, | |
2429 | #endif | |
2430 | .permission = ext3_permission, | |
ae6ddcc5 | 2431 | }; |