]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/namei.c |
ac27a0ec DK |
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> | |
dab291af | 29 | #include <linux/jbd2.h> |
ac27a0ec | 30 | #include <linux/time.h> |
ac27a0ec DK |
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> | |
3dcf5451 CH |
37 | #include "ext4.h" |
38 | #include "ext4_jbd2.h" | |
ac27a0ec | 39 | |
ac27a0ec DK |
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 | |
8c55e204 | 48 | #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) |
ac27a0ec DK |
49 | #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b)) |
50 | ||
617ba13b | 51 | static struct buffer_head *ext4_append(handle_t *handle, |
ac27a0ec | 52 | struct inode *inode, |
725d26d3 | 53 | ext4_lblk_t *block, int *err) |
ac27a0ec DK |
54 | { |
55 | struct buffer_head *bh; | |
56 | ||
57 | *block = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
58 | ||
a871611b AM |
59 | bh = ext4_bread(handle, inode, *block, 1, err); |
60 | if (bh) { | |
ac27a0ec | 61 | inode->i_size += inode->i_sb->s_blocksize; |
617ba13b | 62 | EXT4_I(inode)->i_disksize = inode->i_size; |
a871611b AM |
63 | *err = ext4_journal_get_write_access(handle, bh); |
64 | if (*err) { | |
65 | brelse(bh); | |
66 | bh = NULL; | |
67 | } | |
ac27a0ec DK |
68 | } |
69 | return bh; | |
70 | } | |
71 | ||
72 | #ifndef assert | |
73 | #define assert(test) J_ASSERT(test) | |
74 | #endif | |
75 | ||
ac27a0ec DK |
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; | |
ef2b02d3 ES |
143 | u16 offs; |
144 | u16 size; | |
ac27a0ec DK |
145 | }; |
146 | ||
725d26d3 AK |
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); | |
af5bc92d TT |
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); | |
f702ba0f | 157 | static struct dx_frame *dx_probe(const struct qstr *d_name, |
ac27a0ec DK |
158 | struct inode *dir, |
159 | struct dx_hash_info *hinfo, | |
160 | struct dx_frame *frame, | |
161 | int *err); | |
af5bc92d | 162 | static void dx_release(struct dx_frame *frames); |
8bad4597 | 163 | static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize, |
af5bc92d | 164 | struct dx_hash_info *hinfo, struct dx_map_entry map[]); |
ac27a0ec | 165 | static void dx_sort_map(struct dx_map_entry *map, unsigned count); |
af5bc92d | 166 | static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to, |
3d0518f4 | 167 | struct dx_map_entry *offsets, int count, unsigned blocksize); |
8bad4597 | 168 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize); |
725d26d3 AK |
169 | static void dx_insert_block(struct dx_frame *frame, |
170 | u32 hash, ext4_lblk_t block); | |
617ba13b | 171 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, |
ac27a0ec DK |
172 | struct dx_frame *frame, |
173 | struct dx_frame *frames, | |
174 | __u32 *start_hash); | |
f702ba0f TT |
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); | |
617ba13b | 179 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, |
ac27a0ec DK |
180 | struct inode *inode); |
181 | ||
3d0518f4 WY |
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 | } | |
60e6679e | 190 | |
3d0518f4 WY |
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); | |
60e6679e | 200 | else |
3d0518f4 WY |
201 | return cpu_to_le16(0); |
202 | } | |
203 | return cpu_to_le16((len & 65532) | ((len >> 16) & 3)); | |
204 | } | |
205 | ||
f795e140 LZ |
206 | /* |
207 | * p is at least 6 bytes before the end of page | |
208 | */ | |
209 | static inline struct ext4_dir_entry_2 * | |
3d0518f4 | 210 | ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) |
f795e140 LZ |
211 | { |
212 | return (struct ext4_dir_entry_2 *)((char *)p + | |
3d0518f4 | 213 | ext4_rec_len_from_disk(p->rec_len, blocksize)); |
f795e140 LZ |
214 | } |
215 | ||
ac27a0ec DK |
216 | /* |
217 | * Future: use high four bits of block for coalesce-on-delete flags | |
218 | * Mask them off for now. | |
219 | */ | |
220 | ||
725d26d3 | 221 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) |
ac27a0ec DK |
222 | { |
223 | return le32_to_cpu(entry->block) & 0x00ffffff; | |
224 | } | |
225 | ||
725d26d3 | 226 | static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) |
ac27a0ec DK |
227 | { |
228 | entry->block = cpu_to_le32(value); | |
229 | } | |
230 | ||
af5bc92d | 231 | static inline unsigned dx_get_hash(struct dx_entry *entry) |
ac27a0ec DK |
232 | { |
233 | return le32_to_cpu(entry->hash); | |
234 | } | |
235 | ||
af5bc92d | 236 | static inline void dx_set_hash(struct dx_entry *entry, unsigned value) |
ac27a0ec DK |
237 | { |
238 | entry->hash = cpu_to_le32(value); | |
239 | } | |
240 | ||
af5bc92d | 241 | static inline unsigned dx_get_count(struct dx_entry *entries) |
ac27a0ec DK |
242 | { |
243 | return le16_to_cpu(((struct dx_countlimit *) entries)->count); | |
244 | } | |
245 | ||
af5bc92d | 246 | static inline unsigned dx_get_limit(struct dx_entry *entries) |
ac27a0ec DK |
247 | { |
248 | return le16_to_cpu(((struct dx_countlimit *) entries)->limit); | |
249 | } | |
250 | ||
af5bc92d | 251 | static inline void dx_set_count(struct dx_entry *entries, unsigned value) |
ac27a0ec DK |
252 | { |
253 | ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); | |
254 | } | |
255 | ||
af5bc92d | 256 | static inline void dx_set_limit(struct dx_entry *entries, unsigned value) |
ac27a0ec DK |
257 | { |
258 | ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); | |
259 | } | |
260 | ||
af5bc92d | 261 | static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) |
ac27a0ec | 262 | { |
617ba13b MC |
263 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - |
264 | EXT4_DIR_REC_LEN(2) - infosize; | |
d9c769b7 | 265 | return entry_space / sizeof(struct dx_entry); |
ac27a0ec DK |
266 | } |
267 | ||
af5bc92d | 268 | static inline unsigned dx_node_limit(struct inode *dir) |
ac27a0ec | 269 | { |
617ba13b | 270 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); |
d9c769b7 | 271 | return entry_space / sizeof(struct dx_entry); |
ac27a0ec DK |
272 | } |
273 | ||
274 | /* | |
275 | * Debug | |
276 | */ | |
277 | #ifdef DX_DEBUG | |
4776004f | 278 | static void dx_show_index(char * label, struct dx_entry *entries) |
ac27a0ec | 279 | { |
63f57933 | 280 | int i, n = dx_get_count (entries); |
4776004f | 281 | printk(KERN_DEBUG "%s index ", label); |
63f57933 | 282 | for (i = 0; i < n; i++) { |
4776004f | 283 | printk("%x->%lu ", i ? dx_get_hash(entries + i) : |
725d26d3 | 284 | 0, (unsigned long)dx_get_block(entries + i)); |
63f57933 AM |
285 | } |
286 | printk("\n"); | |
ac27a0ec DK |
287 | } |
288 | ||
289 | struct stats | |
290 | { | |
291 | unsigned names; | |
292 | unsigned space; | |
293 | unsigned bcount; | |
294 | }; | |
295 | ||
617ba13b | 296 | static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de, |
ac27a0ec DK |
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++); | |
617ba13b | 313 | ext4fs_dirhash(de->name, de->name_len, &h); |
ac27a0ec DK |
314 | printk(":%x.%u ", h.hash, |
315 | ((char *) de - base)); | |
316 | } | |
617ba13b | 317 | space += EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec DK |
318 | names++; |
319 | } | |
3d0518f4 | 320 | de = ext4_next_entry(de, size); |
ac27a0ec DK |
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; | |
af5bc92d | 330 | unsigned count = dx_get_count(entries), names = 0, space = 0, i; |
ac27a0ec DK |
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 | { | |
725d26d3 AK |
337 | ext4_lblk_t block = dx_get_block(entries); |
338 | ext4_lblk_t hash = i ? dx_get_hash(entries): 0; | |
ac27a0ec DK |
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); | |
617ba13b | 342 | if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue; |
ac27a0ec DK |
343 | stats = levels? |
344 | dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): | |
617ba13b | 345 | dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0); |
ac27a0ec DK |
346 | names += stats.names; |
347 | space += stats.space; | |
348 | bcount += stats.bcount; | |
af5bc92d | 349 | brelse(bh); |
ac27a0ec DK |
350 | } |
351 | if (bcount) | |
60e6679e | 352 | printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", |
4776004f TT |
353 | levels ? "" : " ", names, space/bcount, |
354 | (space/bcount)*100/blocksize); | |
ac27a0ec DK |
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 * | |
f702ba0f | 369 | dx_probe(const struct qstr *d_name, struct inode *dir, |
ac27a0ec DK |
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; | |
617ba13b | 380 | if (!(bh = ext4_bread (NULL,dir, 0, 0, err))) |
ac27a0ec DK |
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) { | |
12062ddd | 386 | ext4_warning(dir->i_sb, "Unrecognised inode hash code %d", |
ac27a0ec DK |
387 | root->info.hash_version); |
388 | brelse(bh); | |
389 | *err = ERR_BAD_DX_DIR; | |
390 | goto fail; | |
391 | } | |
392 | hinfo->hash_version = root->info.hash_version; | |
f99b2589 TT |
393 | if (hinfo->hash_version <= DX_HASH_TEA) |
394 | hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b | 395 | hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
f702ba0f TT |
396 | if (d_name) |
397 | ext4fs_dirhash(d_name->name, d_name->len, hinfo); | |
ac27a0ec DK |
398 | hash = hinfo->hash; |
399 | ||
400 | if (root->info.unused_flags & 1) { | |
12062ddd | 401 | ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x", |
ac27a0ec DK |
402 | root->info.unused_flags); |
403 | brelse(bh); | |
404 | *err = ERR_BAD_DX_DIR; | |
405 | goto fail; | |
406 | } | |
407 | ||
408 | if ((indirect = root->info.indirect_levels) > 1) { | |
12062ddd | 409 | ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x", |
ac27a0ec DK |
410 | root->info.indirect_levels); |
411 | brelse(bh); | |
412 | *err = ERR_BAD_DX_DIR; | |
413 | goto fail; | |
414 | } | |
415 | ||
416 | entries = (struct dx_entry *) (((char *)&root->info) + | |
417 | root->info.info_length); | |
3d82abae ES |
418 | |
419 | if (dx_get_limit(entries) != dx_root_limit(dir, | |
420 | root->info.info_length)) { | |
12062ddd | 421 | ext4_warning(dir->i_sb, "dx entry: limit != root limit"); |
3d82abae ES |
422 | brelse(bh); |
423 | *err = ERR_BAD_DX_DIR; | |
424 | goto fail; | |
425 | } | |
426 | ||
af5bc92d | 427 | dxtrace(printk("Look up %x", hash)); |
ac27a0ec DK |
428 | while (1) |
429 | { | |
430 | count = dx_get_count(entries); | |
3d82abae | 431 | if (!count || count > dx_get_limit(entries)) { |
12062ddd | 432 | ext4_warning(dir->i_sb, |
3d82abae ES |
433 | "dx entry: no count or count > limit"); |
434 | brelse(bh); | |
435 | *err = ERR_BAD_DX_DIR; | |
436 | goto fail2; | |
437 | } | |
438 | ||
ac27a0ec DK |
439 | p = entries + 1; |
440 | q = entries + count - 1; | |
441 | while (p <= q) | |
442 | { | |
443 | m = p + (q - p)/2; | |
444 | dxtrace(printk(".")); | |
445 | if (dx_get_hash(m) > hash) | |
446 | q = m - 1; | |
447 | else | |
448 | p = m + 1; | |
449 | } | |
450 | ||
451 | if (0) // linear search cross check | |
452 | { | |
453 | unsigned n = count - 1; | |
454 | at = entries; | |
455 | while (n--) | |
456 | { | |
457 | dxtrace(printk(",")); | |
458 | if (dx_get_hash(++at) > hash) | |
459 | { | |
460 | at--; | |
461 | break; | |
462 | } | |
463 | } | |
464 | assert (at == p - 1); | |
465 | } | |
466 | ||
467 | at = p - 1; | |
468 | dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at))); | |
469 | frame->bh = bh; | |
470 | frame->entries = entries; | |
471 | frame->at = at; | |
472 | if (!indirect--) return frame; | |
617ba13b | 473 | if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err))) |
ac27a0ec DK |
474 | goto fail2; |
475 | at = entries = ((struct dx_node *) bh->b_data)->entries; | |
3d82abae | 476 | if (dx_get_limit(entries) != dx_node_limit (dir)) { |
12062ddd | 477 | ext4_warning(dir->i_sb, |
3d82abae ES |
478 | "dx entry: limit != node limit"); |
479 | brelse(bh); | |
480 | *err = ERR_BAD_DX_DIR; | |
481 | goto fail2; | |
482 | } | |
ac27a0ec | 483 | frame++; |
3d82abae | 484 | frame->bh = NULL; |
ac27a0ec DK |
485 | } |
486 | fail2: | |
487 | while (frame >= frame_in) { | |
488 | brelse(frame->bh); | |
489 | frame--; | |
490 | } | |
491 | fail: | |
3d82abae | 492 | if (*err == ERR_BAD_DX_DIR) |
12062ddd | 493 | ext4_warning(dir->i_sb, |
3d82abae ES |
494 | "Corrupt dir inode %ld, running e2fsck is " |
495 | "recommended.", dir->i_ino); | |
ac27a0ec DK |
496 | return NULL; |
497 | } | |
498 | ||
499 | static void dx_release (struct dx_frame *frames) | |
500 | { | |
501 | if (frames[0].bh == NULL) | |
502 | return; | |
503 | ||
504 | if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels) | |
505 | brelse(frames[1].bh); | |
506 | brelse(frames[0].bh); | |
507 | } | |
508 | ||
509 | /* | |
510 | * This function increments the frame pointer to search the next leaf | |
511 | * block, and reads in the necessary intervening nodes if the search | |
512 | * should be necessary. Whether or not the search is necessary is | |
513 | * controlled by the hash parameter. If the hash value is even, then | |
514 | * the search is only continued if the next block starts with that | |
515 | * hash value. This is used if we are searching for a specific file. | |
516 | * | |
517 | * If the hash value is HASH_NB_ALWAYS, then always go to the next block. | |
518 | * | |
519 | * This function returns 1 if the caller should continue to search, | |
520 | * or 0 if it should not. If there is an error reading one of the | |
521 | * index blocks, it will a negative error code. | |
522 | * | |
523 | * If start_hash is non-null, it will be filled in with the starting | |
524 | * hash of the next page. | |
525 | */ | |
617ba13b | 526 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, |
ac27a0ec DK |
527 | struct dx_frame *frame, |
528 | struct dx_frame *frames, | |
529 | __u32 *start_hash) | |
530 | { | |
531 | struct dx_frame *p; | |
532 | struct buffer_head *bh; | |
533 | int err, num_frames = 0; | |
534 | __u32 bhash; | |
535 | ||
536 | p = frame; | |
537 | /* | |
538 | * Find the next leaf page by incrementing the frame pointer. | |
539 | * If we run out of entries in the interior node, loop around and | |
540 | * increment pointer in the parent node. When we break out of | |
541 | * this loop, num_frames indicates the number of interior | |
542 | * nodes need to be read. | |
543 | */ | |
544 | while (1) { | |
545 | if (++(p->at) < p->entries + dx_get_count(p->entries)) | |
546 | break; | |
547 | if (p == frames) | |
548 | return 0; | |
549 | num_frames++; | |
550 | p--; | |
551 | } | |
552 | ||
553 | /* | |
554 | * If the hash is 1, then continue only if the next page has a | |
555 | * continuation hash of any value. This is used for readdir | |
556 | * handling. Otherwise, check to see if the hash matches the | |
557 | * desired contiuation hash. If it doesn't, return since | |
558 | * there's no point to read in the successive index pages. | |
559 | */ | |
560 | bhash = dx_get_hash(p->at); | |
561 | if (start_hash) | |
562 | *start_hash = bhash; | |
563 | if ((hash & 1) == 0) { | |
564 | if ((bhash & ~1) != hash) | |
565 | return 0; | |
566 | } | |
567 | /* | |
568 | * If the hash is HASH_NB_ALWAYS, we always go to the next | |
569 | * block so no check is necessary | |
570 | */ | |
571 | while (num_frames--) { | |
617ba13b | 572 | if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at), |
ac27a0ec DK |
573 | 0, &err))) |
574 | return err; /* Failure */ | |
575 | p++; | |
af5bc92d | 576 | brelse(p->bh); |
ac27a0ec DK |
577 | p->bh = bh; |
578 | p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; | |
579 | } | |
580 | return 1; | |
581 | } | |
582 | ||
583 | ||
ac27a0ec DK |
584 | /* |
585 | * This function fills a red-black tree with information from a | |
586 | * directory block. It returns the number directory entries loaded | |
587 | * into the tree. If there is an error it is returned in err. | |
588 | */ | |
589 | static int htree_dirblock_to_tree(struct file *dir_file, | |
725d26d3 | 590 | struct inode *dir, ext4_lblk_t block, |
ac27a0ec DK |
591 | struct dx_hash_info *hinfo, |
592 | __u32 start_hash, __u32 start_minor_hash) | |
593 | { | |
594 | struct buffer_head *bh; | |
617ba13b | 595 | struct ext4_dir_entry_2 *de, *top; |
ac27a0ec DK |
596 | int err, count = 0; |
597 | ||
725d26d3 AK |
598 | dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", |
599 | (unsigned long)block)); | |
617ba13b | 600 | if (!(bh = ext4_bread (NULL, dir, block, 0, &err))) |
ac27a0ec DK |
601 | return err; |
602 | ||
617ba13b MC |
603 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
604 | top = (struct ext4_dir_entry_2 *) ((char *) de + | |
ac27a0ec | 605 | dir->i_sb->s_blocksize - |
617ba13b | 606 | EXT4_DIR_REC_LEN(0)); |
3d0518f4 | 607 | for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { |
e6c40211 ES |
608 | if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh, |
609 | (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) | |
610 | +((char *)de - bh->b_data))) { | |
611 | /* On error, skip the f_pos to the next block. */ | |
612 | dir_file->f_pos = (dir_file->f_pos | | |
613 | (dir->i_sb->s_blocksize - 1)) + 1; | |
af5bc92d | 614 | brelse(bh); |
e6c40211 ES |
615 | return count; |
616 | } | |
617ba13b | 617 | ext4fs_dirhash(de->name, de->name_len, hinfo); |
ac27a0ec DK |
618 | if ((hinfo->hash < start_hash) || |
619 | ((hinfo->hash == start_hash) && | |
620 | (hinfo->minor_hash < start_minor_hash))) | |
621 | continue; | |
622 | if (de->inode == 0) | |
623 | continue; | |
617ba13b | 624 | if ((err = ext4_htree_store_dirent(dir_file, |
ac27a0ec DK |
625 | hinfo->hash, hinfo->minor_hash, de)) != 0) { |
626 | brelse(bh); | |
627 | return err; | |
628 | } | |
629 | count++; | |
630 | } | |
631 | brelse(bh); | |
632 | return count; | |
633 | } | |
634 | ||
635 | ||
636 | /* | |
637 | * This function fills a red-black tree with information from a | |
638 | * directory. We start scanning the directory in hash order, starting | |
639 | * at start_hash and start_minor_hash. | |
640 | * | |
641 | * This function returns the number of entries inserted into the tree, | |
642 | * or a negative error code. | |
643 | */ | |
617ba13b | 644 | int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, |
ac27a0ec DK |
645 | __u32 start_minor_hash, __u32 *next_hash) |
646 | { | |
647 | struct dx_hash_info hinfo; | |
617ba13b | 648 | struct ext4_dir_entry_2 *de; |
ac27a0ec DK |
649 | struct dx_frame frames[2], *frame; |
650 | struct inode *dir; | |
725d26d3 | 651 | ext4_lblk_t block; |
ac27a0ec | 652 | int count = 0; |
725d26d3 | 653 | int ret, err; |
ac27a0ec DK |
654 | __u32 hashval; |
655 | ||
60e6679e | 656 | dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", |
4776004f | 657 | start_hash, start_minor_hash)); |
9d549890 | 658 | dir = dir_file->f_path.dentry->d_inode; |
12e9b892 | 659 | if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) { |
617ba13b | 660 | hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; |
f99b2589 TT |
661 | if (hinfo.hash_version <= DX_HASH_TEA) |
662 | hinfo.hash_version += | |
663 | EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b | 664 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
ac27a0ec DK |
665 | count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, |
666 | start_hash, start_minor_hash); | |
667 | *next_hash = ~0; | |
668 | return count; | |
669 | } | |
670 | hinfo.hash = start_hash; | |
671 | hinfo.minor_hash = 0; | |
f702ba0f | 672 | frame = dx_probe(NULL, dir, &hinfo, frames, &err); |
ac27a0ec DK |
673 | if (!frame) |
674 | return err; | |
675 | ||
676 | /* Add '.' and '..' from the htree header */ | |
677 | if (!start_hash && !start_minor_hash) { | |
617ba13b MC |
678 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; |
679 | if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0) | |
ac27a0ec DK |
680 | goto errout; |
681 | count++; | |
682 | } | |
683 | if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { | |
617ba13b | 684 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; |
3d0518f4 | 685 | de = ext4_next_entry(de, dir->i_sb->s_blocksize); |
617ba13b | 686 | if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0) |
ac27a0ec DK |
687 | goto errout; |
688 | count++; | |
689 | } | |
690 | ||
691 | while (1) { | |
692 | block = dx_get_block(frame->at); | |
693 | ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, | |
694 | start_hash, start_minor_hash); | |
695 | if (ret < 0) { | |
696 | err = ret; | |
697 | goto errout; | |
698 | } | |
699 | count += ret; | |
700 | hashval = ~0; | |
617ba13b | 701 | ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, |
ac27a0ec DK |
702 | frame, frames, &hashval); |
703 | *next_hash = hashval; | |
704 | if (ret < 0) { | |
705 | err = ret; | |
706 | goto errout; | |
707 | } | |
708 | /* | |
709 | * Stop if: (a) there are no more entries, or | |
710 | * (b) we have inserted at least one entry and the | |
711 | * next hash value is not a continuation | |
712 | */ | |
713 | if ((ret == 0) || | |
714 | (count && ((hashval & 1) == 0))) | |
715 | break; | |
716 | } | |
717 | dx_release(frames); | |
4776004f TT |
718 | dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " |
719 | "next hash: %x\n", count, *next_hash)); | |
ac27a0ec DK |
720 | return count; |
721 | errout: | |
722 | dx_release(frames); | |
723 | return (err); | |
724 | } | |
725 | ||
726 | ||
727 | /* | |
728 | * Directory block splitting, compacting | |
729 | */ | |
730 | ||
ef2b02d3 ES |
731 | /* |
732 | * Create map of hash values, offsets, and sizes, stored at end of block. | |
733 | * Returns number of entries mapped. | |
734 | */ | |
8bad4597 TT |
735 | static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize, |
736 | struct dx_hash_info *hinfo, | |
737 | struct dx_map_entry *map_tail) | |
ac27a0ec DK |
738 | { |
739 | int count = 0; | |
740 | char *base = (char *) de; | |
741 | struct dx_hash_info h = *hinfo; | |
742 | ||
8bad4597 | 743 | while ((char *) de < base + blocksize) { |
ac27a0ec | 744 | if (de->name_len && de->inode) { |
617ba13b | 745 | ext4fs_dirhash(de->name, de->name_len, &h); |
ac27a0ec DK |
746 | map_tail--; |
747 | map_tail->hash = h.hash; | |
9aee2286 | 748 | map_tail->offs = ((char *) de - base)>>2; |
ef2b02d3 | 749 | map_tail->size = le16_to_cpu(de->rec_len); |
ac27a0ec DK |
750 | count++; |
751 | cond_resched(); | |
752 | } | |
753 | /* XXX: do we need to check rec_len == 0 case? -Chris */ | |
3d0518f4 | 754 | de = ext4_next_entry(de, blocksize); |
ac27a0ec DK |
755 | } |
756 | return count; | |
757 | } | |
758 | ||
ef2b02d3 | 759 | /* Sort map by hash value */ |
ac27a0ec DK |
760 | static void dx_sort_map (struct dx_map_entry *map, unsigned count) |
761 | { | |
63f57933 AM |
762 | struct dx_map_entry *p, *q, *top = map + count - 1; |
763 | int more; | |
764 | /* Combsort until bubble sort doesn't suck */ | |
765 | while (count > 2) { | |
766 | count = count*10/13; | |
767 | if (count - 9 < 2) /* 9, 10 -> 11 */ | |
768 | count = 11; | |
769 | for (p = top, q = p - count; q >= map; p--, q--) | |
770 | if (p->hash < q->hash) | |
771 | swap(*p, *q); | |
772 | } | |
773 | /* Garden variety bubble sort */ | |
774 | do { | |
775 | more = 0; | |
776 | q = top; | |
777 | while (q-- > map) { | |
778 | if (q[1].hash >= q[0].hash) | |
ac27a0ec | 779 | continue; |
63f57933 AM |
780 | swap(*(q+1), *q); |
781 | more = 1; | |
ac27a0ec DK |
782 | } |
783 | } while(more); | |
784 | } | |
785 | ||
725d26d3 | 786 | static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) |
ac27a0ec DK |
787 | { |
788 | struct dx_entry *entries = frame->entries; | |
789 | struct dx_entry *old = frame->at, *new = old + 1; | |
790 | int count = dx_get_count(entries); | |
791 | ||
792 | assert(count < dx_get_limit(entries)); | |
793 | assert(old < entries + count); | |
794 | memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); | |
795 | dx_set_hash(new, hash); | |
796 | dx_set_block(new, block); | |
797 | dx_set_count(entries, count + 1); | |
798 | } | |
ac27a0ec | 799 | |
617ba13b | 800 | static void ext4_update_dx_flag(struct inode *inode) |
ac27a0ec | 801 | { |
617ba13b MC |
802 | if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb, |
803 | EXT4_FEATURE_COMPAT_DIR_INDEX)) | |
12e9b892 | 804 | ext4_clear_inode_flag(inode, EXT4_INODE_INDEX); |
ac27a0ec DK |
805 | } |
806 | ||
807 | /* | |
617ba13b | 808 | * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure. |
ac27a0ec | 809 | * |
617ba13b | 810 | * `len <= EXT4_NAME_LEN' is guaranteed by caller. |
ac27a0ec DK |
811 | * `de != NULL' is guaranteed by caller. |
812 | */ | |
617ba13b MC |
813 | static inline int ext4_match (int len, const char * const name, |
814 | struct ext4_dir_entry_2 * de) | |
ac27a0ec DK |
815 | { |
816 | if (len != de->name_len) | |
817 | return 0; | |
818 | if (!de->inode) | |
819 | return 0; | |
820 | return !memcmp(name, de->name, len); | |
821 | } | |
822 | ||
823 | /* | |
824 | * Returns 0 if not found, -1 on failure, and 1 on success | |
825 | */ | |
af5bc92d | 826 | static inline int search_dirblock(struct buffer_head *bh, |
ac27a0ec | 827 | struct inode *dir, |
f702ba0f | 828 | const struct qstr *d_name, |
498e5f24 | 829 | unsigned int offset, |
617ba13b | 830 | struct ext4_dir_entry_2 ** res_dir) |
ac27a0ec | 831 | { |
617ba13b | 832 | struct ext4_dir_entry_2 * de; |
ac27a0ec DK |
833 | char * dlimit; |
834 | int de_len; | |
f702ba0f TT |
835 | const char *name = d_name->name; |
836 | int namelen = d_name->len; | |
ac27a0ec | 837 | |
617ba13b | 838 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec DK |
839 | dlimit = bh->b_data + dir->i_sb->s_blocksize; |
840 | while ((char *) de < dlimit) { | |
841 | /* this code is executed quadratically often */ | |
842 | /* do minimal checking `by hand' */ | |
843 | ||
844 | if ((char *) de + namelen <= dlimit && | |
617ba13b | 845 | ext4_match (namelen, name, de)) { |
ac27a0ec | 846 | /* found a match - just to be sure, do a full check */ |
617ba13b | 847 | if (!ext4_check_dir_entry("ext4_find_entry", |
ac27a0ec DK |
848 | dir, de, bh, offset)) |
849 | return -1; | |
850 | *res_dir = de; | |
851 | return 1; | |
852 | } | |
853 | /* prevent looping on a bad block */ | |
3d0518f4 WY |
854 | de_len = ext4_rec_len_from_disk(de->rec_len, |
855 | dir->i_sb->s_blocksize); | |
ac27a0ec DK |
856 | if (de_len <= 0) |
857 | return -1; | |
858 | offset += de_len; | |
617ba13b | 859 | de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); |
ac27a0ec DK |
860 | } |
861 | return 0; | |
862 | } | |
863 | ||
864 | ||
865 | /* | |
617ba13b | 866 | * ext4_find_entry() |
ac27a0ec DK |
867 | * |
868 | * finds an entry in the specified directory with the wanted name. It | |
869 | * returns the cache buffer in which the entry was found, and the entry | |
870 | * itself (as a parameter - res_dir). It does NOT read the inode of the | |
871 | * entry - you'll have to do that yourself if you want to. | |
872 | * | |
873 | * The returned buffer_head has ->b_count elevated. The caller is expected | |
874 | * to brelse() it when appropriate. | |
875 | */ | |
f702ba0f TT |
876 | static struct buffer_head * ext4_find_entry (struct inode *dir, |
877 | const struct qstr *d_name, | |
617ba13b | 878 | struct ext4_dir_entry_2 ** res_dir) |
ac27a0ec | 879 | { |
af5bc92d TT |
880 | struct super_block *sb; |
881 | struct buffer_head *bh_use[NAMEI_RA_SIZE]; | |
882 | struct buffer_head *bh, *ret = NULL; | |
725d26d3 | 883 | ext4_lblk_t start, block, b; |
ac27a0ec DK |
884 | int ra_max = 0; /* Number of bh's in the readahead |
885 | buffer, bh_use[] */ | |
886 | int ra_ptr = 0; /* Current index into readahead | |
887 | buffer */ | |
888 | int num = 0; | |
725d26d3 AK |
889 | ext4_lblk_t nblocks; |
890 | int i, err; | |
ac27a0ec | 891 | int namelen; |
ac27a0ec DK |
892 | |
893 | *res_dir = NULL; | |
894 | sb = dir->i_sb; | |
f702ba0f | 895 | namelen = d_name->len; |
617ba13b | 896 | if (namelen > EXT4_NAME_LEN) |
ac27a0ec | 897 | return NULL; |
ac27a0ec | 898 | if (is_dx(dir)) { |
f702ba0f | 899 | bh = ext4_dx_find_entry(dir, d_name, res_dir, &err); |
ac27a0ec DK |
900 | /* |
901 | * On success, or if the error was file not found, | |
902 | * return. Otherwise, fall back to doing a search the | |
903 | * old fashioned way. | |
904 | */ | |
905 | if (bh || (err != ERR_BAD_DX_DIR)) | |
906 | return bh; | |
4776004f TT |
907 | dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " |
908 | "falling back\n")); | |
ac27a0ec | 909 | } |
617ba13b MC |
910 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); |
911 | start = EXT4_I(dir)->i_dir_start_lookup; | |
ac27a0ec DK |
912 | if (start >= nblocks) |
913 | start = 0; | |
914 | block = start; | |
915 | restart: | |
916 | do { | |
917 | /* | |
918 | * We deal with the read-ahead logic here. | |
919 | */ | |
920 | if (ra_ptr >= ra_max) { | |
921 | /* Refill the readahead buffer */ | |
922 | ra_ptr = 0; | |
923 | b = block; | |
924 | for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { | |
925 | /* | |
926 | * Terminate if we reach the end of the | |
927 | * directory and must wrap, or if our | |
928 | * search has finished at this block. | |
929 | */ | |
930 | if (b >= nblocks || (num && block == start)) { | |
931 | bh_use[ra_max] = NULL; | |
932 | break; | |
933 | } | |
934 | num++; | |
617ba13b | 935 | bh = ext4_getblk(NULL, dir, b++, 0, &err); |
ac27a0ec DK |
936 | bh_use[ra_max] = bh; |
937 | if (bh) | |
938 | ll_rw_block(READ_META, 1, &bh); | |
939 | } | |
940 | } | |
941 | if ((bh = bh_use[ra_ptr++]) == NULL) | |
942 | goto next; | |
943 | wait_on_buffer(bh); | |
944 | if (!buffer_uptodate(bh)) { | |
945 | /* read error, skip block & hope for the best */ | |
24676da4 TT |
946 | EXT4_ERROR_INODE(dir, "reading directory lblock %lu", |
947 | (unsigned long) block); | |
ac27a0ec DK |
948 | brelse(bh); |
949 | goto next; | |
950 | } | |
f702ba0f | 951 | i = search_dirblock(bh, dir, d_name, |
617ba13b | 952 | block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); |
ac27a0ec | 953 | if (i == 1) { |
617ba13b | 954 | EXT4_I(dir)->i_dir_start_lookup = block; |
ac27a0ec DK |
955 | ret = bh; |
956 | goto cleanup_and_exit; | |
957 | } else { | |
958 | brelse(bh); | |
959 | if (i < 0) | |
960 | goto cleanup_and_exit; | |
961 | } | |
962 | next: | |
963 | if (++block >= nblocks) | |
964 | block = 0; | |
965 | } while (block != start); | |
966 | ||
967 | /* | |
968 | * If the directory has grown while we were searching, then | |
969 | * search the last part of the directory before giving up. | |
970 | */ | |
971 | block = nblocks; | |
617ba13b | 972 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); |
ac27a0ec DK |
973 | if (block < nblocks) { |
974 | start = 0; | |
975 | goto restart; | |
976 | } | |
977 | ||
978 | cleanup_and_exit: | |
979 | /* Clean up the read-ahead blocks */ | |
980 | for (; ra_ptr < ra_max; ra_ptr++) | |
af5bc92d | 981 | brelse(bh_use[ra_ptr]); |
ac27a0ec DK |
982 | return ret; |
983 | } | |
984 | ||
f702ba0f | 985 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name, |
617ba13b | 986 | struct ext4_dir_entry_2 **res_dir, int *err) |
ac27a0ec DK |
987 | { |
988 | struct super_block * sb; | |
989 | struct dx_hash_info hinfo; | |
990 | u32 hash; | |
991 | struct dx_frame frames[2], *frame; | |
617ba13b | 992 | struct ext4_dir_entry_2 *de, *top; |
ac27a0ec | 993 | struct buffer_head *bh; |
725d26d3 | 994 | ext4_lblk_t block; |
ac27a0ec | 995 | int retval; |
f702ba0f TT |
996 | int namelen = d_name->len; |
997 | const u8 *name = d_name->name; | |
ac27a0ec DK |
998 | |
999 | sb = dir->i_sb; | |
1000 | /* NFS may look up ".." - look at dx_root directory block */ | |
1001 | if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){ | |
f702ba0f | 1002 | if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err))) |
ac27a0ec DK |
1003 | return NULL; |
1004 | } else { | |
1005 | frame = frames; | |
1006 | frame->bh = NULL; /* for dx_release() */ | |
1007 | frame->at = (struct dx_entry *)frames; /* hack for zero entry*/ | |
1008 | dx_set_block(frame->at, 0); /* dx_root block is 0 */ | |
1009 | } | |
1010 | hash = hinfo.hash; | |
1011 | do { | |
1012 | block = dx_get_block(frame->at); | |
617ba13b | 1013 | if (!(bh = ext4_bread (NULL,dir, block, 0, err))) |
ac27a0ec | 1014 | goto errout; |
617ba13b MC |
1015 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
1016 | top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize - | |
1017 | EXT4_DIR_REC_LEN(0)); | |
3d0518f4 | 1018 | for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) { |
f3b35f06 DG |
1019 | int off = (block << EXT4_BLOCK_SIZE_BITS(sb)) |
1020 | + ((char *) de - bh->b_data); | |
1021 | ||
1022 | if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) { | |
1023 | brelse(bh); | |
fedee54d | 1024 | *err = ERR_BAD_DX_DIR; |
ac27a0ec DK |
1025 | goto errout; |
1026 | } | |
f3b35f06 DG |
1027 | |
1028 | if (ext4_match(namelen, name, de)) { | |
1029 | *res_dir = de; | |
1030 | dx_release(frames); | |
1031 | return bh; | |
1032 | } | |
ac27a0ec | 1033 | } |
af5bc92d | 1034 | brelse(bh); |
ac27a0ec | 1035 | /* Check to see if we should continue to search */ |
617ba13b | 1036 | retval = ext4_htree_next_block(dir, hash, frame, |
ac27a0ec DK |
1037 | frames, NULL); |
1038 | if (retval < 0) { | |
12062ddd | 1039 | ext4_warning(sb, |
ac27a0ec DK |
1040 | "error reading index page in directory #%lu", |
1041 | dir->i_ino); | |
1042 | *err = retval; | |
1043 | goto errout; | |
1044 | } | |
1045 | } while (retval == 1); | |
1046 | ||
1047 | *err = -ENOENT; | |
1048 | errout: | |
4776004f | 1049 | dxtrace(printk(KERN_DEBUG "%s not found\n", name)); |
ac27a0ec DK |
1050 | dx_release (frames); |
1051 | return NULL; | |
1052 | } | |
ac27a0ec | 1053 | |
af5bc92d | 1054 | static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
ac27a0ec | 1055 | { |
af5bc92d TT |
1056 | struct inode *inode; |
1057 | struct ext4_dir_entry_2 *de; | |
1058 | struct buffer_head *bh; | |
ac27a0ec | 1059 | |
617ba13b | 1060 | if (dentry->d_name.len > EXT4_NAME_LEN) |
ac27a0ec DK |
1061 | return ERR_PTR(-ENAMETOOLONG); |
1062 | ||
f702ba0f | 1063 | bh = ext4_find_entry(dir, &dentry->d_name, &de); |
ac27a0ec DK |
1064 | inode = NULL; |
1065 | if (bh) { | |
498e5f24 | 1066 | __u32 ino = le32_to_cpu(de->inode); |
af5bc92d | 1067 | brelse(bh); |
617ba13b | 1068 | if (!ext4_valid_inum(dir->i_sb, ino)) { |
24676da4 | 1069 | EXT4_ERROR_INODE(dir, "bad inode number: %u", ino); |
1d1fe1ee | 1070 | return ERR_PTR(-EIO); |
a6c15c2b | 1071 | } |
1d1fe1ee | 1072 | inode = ext4_iget(dir->i_sb, ino); |
e6f009b0 BD |
1073 | if (unlikely(IS_ERR(inode))) { |
1074 | if (PTR_ERR(inode) == -ESTALE) { | |
24676da4 TT |
1075 | EXT4_ERROR_INODE(dir, |
1076 | "deleted inode referenced: %u", | |
1077 | ino); | |
e6f009b0 BD |
1078 | return ERR_PTR(-EIO); |
1079 | } else { | |
1080 | return ERR_CAST(inode); | |
1081 | } | |
1082 | } | |
ac27a0ec DK |
1083 | } |
1084 | return d_splice_alias(inode, dentry); | |
1085 | } | |
1086 | ||
1087 | ||
617ba13b | 1088 | struct dentry *ext4_get_parent(struct dentry *child) |
ac27a0ec | 1089 | { |
498e5f24 | 1090 | __u32 ino; |
ac27a0ec | 1091 | struct inode *inode; |
f702ba0f TT |
1092 | static const struct qstr dotdot = { |
1093 | .name = "..", | |
1094 | .len = 2, | |
1095 | }; | |
617ba13b | 1096 | struct ext4_dir_entry_2 * de; |
ac27a0ec DK |
1097 | struct buffer_head *bh; |
1098 | ||
f702ba0f | 1099 | bh = ext4_find_entry(child->d_inode, &dotdot, &de); |
ac27a0ec DK |
1100 | inode = NULL; |
1101 | if (!bh) | |
1102 | return ERR_PTR(-ENOENT); | |
1103 | ino = le32_to_cpu(de->inode); | |
1104 | brelse(bh); | |
1105 | ||
617ba13b | 1106 | if (!ext4_valid_inum(child->d_inode->i_sb, ino)) { |
24676da4 TT |
1107 | EXT4_ERROR_INODE(child->d_inode, |
1108 | "bad parent inode number: %u", ino); | |
1d1fe1ee | 1109 | return ERR_PTR(-EIO); |
a6c15c2b VA |
1110 | } |
1111 | ||
44003728 | 1112 | return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino)); |
ac27a0ec DK |
1113 | } |
1114 | ||
1115 | #define S_SHIFT 12 | |
617ba13b MC |
1116 | static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = { |
1117 | [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE, | |
1118 | [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR, | |
1119 | [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV, | |
1120 | [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV, | |
1121 | [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO, | |
1122 | [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK, | |
1123 | [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK, | |
ac27a0ec DK |
1124 | }; |
1125 | ||
617ba13b MC |
1126 | static inline void ext4_set_de_type(struct super_block *sb, |
1127 | struct ext4_dir_entry_2 *de, | |
ac27a0ec | 1128 | umode_t mode) { |
617ba13b MC |
1129 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE)) |
1130 | de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT]; | |
ac27a0ec DK |
1131 | } |
1132 | ||
ef2b02d3 ES |
1133 | /* |
1134 | * Move count entries from end of map between two memory locations. | |
1135 | * Returns pointer to last entry moved. | |
1136 | */ | |
617ba13b | 1137 | static struct ext4_dir_entry_2 * |
3d0518f4 WY |
1138 | dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count, |
1139 | unsigned blocksize) | |
ac27a0ec DK |
1140 | { |
1141 | unsigned rec_len = 0; | |
1142 | ||
1143 | while (count--) { | |
60e6679e | 1144 | struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) |
9aee2286 | 1145 | (from + (map->offs<<2)); |
617ba13b | 1146 | rec_len = EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec | 1147 | memcpy (to, de, rec_len); |
617ba13b | 1148 | ((struct ext4_dir_entry_2 *) to)->rec_len = |
3d0518f4 | 1149 | ext4_rec_len_to_disk(rec_len, blocksize); |
ac27a0ec DK |
1150 | de->inode = 0; |
1151 | map++; | |
1152 | to += rec_len; | |
1153 | } | |
617ba13b | 1154 | return (struct ext4_dir_entry_2 *) (to - rec_len); |
ac27a0ec DK |
1155 | } |
1156 | ||
ef2b02d3 ES |
1157 | /* |
1158 | * Compact each dir entry in the range to the minimal rec_len. | |
1159 | * Returns pointer to last entry in range. | |
1160 | */ | |
8bad4597 | 1161 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize) |
ac27a0ec | 1162 | { |
617ba13b | 1163 | struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; |
ac27a0ec DK |
1164 | unsigned rec_len = 0; |
1165 | ||
1166 | prev = to = de; | |
8bad4597 | 1167 | while ((char*)de < base + blocksize) { |
3d0518f4 | 1168 | next = ext4_next_entry(de, blocksize); |
ac27a0ec | 1169 | if (de->inode && de->name_len) { |
617ba13b | 1170 | rec_len = EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec DK |
1171 | if (de > to) |
1172 | memmove(to, de, rec_len); | |
3d0518f4 | 1173 | to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); |
ac27a0ec | 1174 | prev = to; |
617ba13b | 1175 | to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); |
ac27a0ec DK |
1176 | } |
1177 | de = next; | |
1178 | } | |
1179 | return prev; | |
1180 | } | |
1181 | ||
ef2b02d3 ES |
1182 | /* |
1183 | * Split a full leaf block to make room for a new dir entry. | |
1184 | * Allocate a new block, and move entries so that they are approx. equally full. | |
1185 | * Returns pointer to de in block into which the new entry will be inserted. | |
1186 | */ | |
617ba13b | 1187 | static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, |
ac27a0ec DK |
1188 | struct buffer_head **bh,struct dx_frame *frame, |
1189 | struct dx_hash_info *hinfo, int *error) | |
1190 | { | |
1191 | unsigned blocksize = dir->i_sb->s_blocksize; | |
1192 | unsigned count, continued; | |
1193 | struct buffer_head *bh2; | |
725d26d3 | 1194 | ext4_lblk_t newblock; |
ac27a0ec DK |
1195 | u32 hash2; |
1196 | struct dx_map_entry *map; | |
1197 | char *data1 = (*bh)->b_data, *data2; | |
59e315b4 | 1198 | unsigned split, move, size; |
617ba13b | 1199 | struct ext4_dir_entry_2 *de = NULL, *de2; |
59e315b4 | 1200 | int err = 0, i; |
ac27a0ec | 1201 | |
fedee54d | 1202 | bh2 = ext4_append (handle, dir, &newblock, &err); |
ac27a0ec DK |
1203 | if (!(bh2)) { |
1204 | brelse(*bh); | |
1205 | *bh = NULL; | |
1206 | goto errout; | |
1207 | } | |
1208 | ||
1209 | BUFFER_TRACE(*bh, "get_write_access"); | |
617ba13b | 1210 | err = ext4_journal_get_write_access(handle, *bh); |
fedee54d DM |
1211 | if (err) |
1212 | goto journal_error; | |
1213 | ||
ac27a0ec | 1214 | BUFFER_TRACE(frame->bh, "get_write_access"); |
617ba13b | 1215 | err = ext4_journal_get_write_access(handle, frame->bh); |
ac27a0ec DK |
1216 | if (err) |
1217 | goto journal_error; | |
1218 | ||
1219 | data2 = bh2->b_data; | |
1220 | ||
1221 | /* create map in the end of data2 block */ | |
1222 | map = (struct dx_map_entry *) (data2 + blocksize); | |
af5bc92d | 1223 | count = dx_make_map((struct ext4_dir_entry_2 *) data1, |
ac27a0ec DK |
1224 | blocksize, hinfo, map); |
1225 | map -= count; | |
af5bc92d | 1226 | dx_sort_map(map, count); |
ef2b02d3 ES |
1227 | /* Split the existing block in the middle, size-wise */ |
1228 | size = 0; | |
1229 | move = 0; | |
1230 | for (i = count-1; i >= 0; i--) { | |
1231 | /* is more than half of this entry in 2nd half of the block? */ | |
1232 | if (size + map[i].size/2 > blocksize/2) | |
1233 | break; | |
1234 | size += map[i].size; | |
1235 | move++; | |
1236 | } | |
1237 | /* map index at which we will split */ | |
1238 | split = count - move; | |
ac27a0ec DK |
1239 | hash2 = map[split].hash; |
1240 | continued = hash2 == map[split - 1].hash; | |
725d26d3 AK |
1241 | dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", |
1242 | (unsigned long)dx_get_block(frame->at), | |
1243 | hash2, split, count-split)); | |
ac27a0ec DK |
1244 | |
1245 | /* Fancy dance to stay within two buffers */ | |
3d0518f4 | 1246 | de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize); |
af5bc92d | 1247 | de = dx_pack_dirents(data1, blocksize); |
3d0518f4 WY |
1248 | de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de, |
1249 | blocksize); | |
1250 | de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2, | |
1251 | blocksize); | |
617ba13b MC |
1252 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1)); |
1253 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1)); | |
ac27a0ec DK |
1254 | |
1255 | /* Which block gets the new entry? */ | |
1256 | if (hinfo->hash >= hash2) | |
1257 | { | |
1258 | swap(*bh, bh2); | |
1259 | de = de2; | |
1260 | } | |
af5bc92d | 1261 | dx_insert_block(frame, hash2 + continued, newblock); |
0390131b | 1262 | err = ext4_handle_dirty_metadata(handle, dir, bh2); |
ac27a0ec DK |
1263 | if (err) |
1264 | goto journal_error; | |
0390131b | 1265 | err = ext4_handle_dirty_metadata(handle, dir, frame->bh); |
ac27a0ec DK |
1266 | if (err) |
1267 | goto journal_error; | |
af5bc92d TT |
1268 | brelse(bh2); |
1269 | dxtrace(dx_show_index("frame", frame->entries)); | |
ac27a0ec | 1270 | return de; |
fedee54d DM |
1271 | |
1272 | journal_error: | |
1273 | brelse(*bh); | |
1274 | brelse(bh2); | |
1275 | *bh = NULL; | |
1276 | ext4_std_error(dir->i_sb, err); | |
1277 | errout: | |
1278 | *error = err; | |
1279 | return NULL; | |
ac27a0ec | 1280 | } |
ac27a0ec DK |
1281 | |
1282 | /* | |
1283 | * Add a new entry into a directory (leaf) block. If de is non-NULL, | |
1284 | * it points to a directory entry which is guaranteed to be large | |
1285 | * enough for new directory entry. If de is NULL, then | |
1286 | * add_dirent_to_buf will attempt search the directory block for | |
1287 | * space. It will return -ENOSPC if no space is available, and -EIO | |
1288 | * and -EEXIST if directory entry already exists. | |
ac27a0ec DK |
1289 | */ |
1290 | static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry, | |
617ba13b | 1291 | struct inode *inode, struct ext4_dir_entry_2 *de, |
af5bc92d | 1292 | struct buffer_head *bh) |
ac27a0ec DK |
1293 | { |
1294 | struct inode *dir = dentry->d_parent->d_inode; | |
1295 | const char *name = dentry->d_name.name; | |
1296 | int namelen = dentry->d_name.len; | |
498e5f24 | 1297 | unsigned int offset = 0; |
3d0518f4 | 1298 | unsigned int blocksize = dir->i_sb->s_blocksize; |
ac27a0ec DK |
1299 | unsigned short reclen; |
1300 | int nlen, rlen, err; | |
1301 | char *top; | |
1302 | ||
617ba13b | 1303 | reclen = EXT4_DIR_REC_LEN(namelen); |
ac27a0ec | 1304 | if (!de) { |
617ba13b | 1305 | de = (struct ext4_dir_entry_2 *)bh->b_data; |
3d0518f4 | 1306 | top = bh->b_data + blocksize - reclen; |
ac27a0ec | 1307 | while ((char *) de <= top) { |
617ba13b | 1308 | if (!ext4_check_dir_entry("ext4_add_entry", dir, de, |
2de770a4 | 1309 | bh, offset)) |
ac27a0ec | 1310 | return -EIO; |
2de770a4 | 1311 | if (ext4_match(namelen, name, de)) |
ac27a0ec | 1312 | return -EEXIST; |
617ba13b | 1313 | nlen = EXT4_DIR_REC_LEN(de->name_len); |
3d0518f4 | 1314 | rlen = ext4_rec_len_from_disk(de->rec_len, blocksize); |
ac27a0ec DK |
1315 | if ((de->inode? rlen - nlen: rlen) >= reclen) |
1316 | break; | |
617ba13b | 1317 | de = (struct ext4_dir_entry_2 *)((char *)de + rlen); |
ac27a0ec DK |
1318 | offset += rlen; |
1319 | } | |
1320 | if ((char *) de > top) | |
1321 | return -ENOSPC; | |
1322 | } | |
1323 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1324 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1325 | if (err) { |
617ba13b | 1326 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1327 | return err; |
1328 | } | |
1329 | ||
1330 | /* By now the buffer is marked for journaling */ | |
617ba13b | 1331 | nlen = EXT4_DIR_REC_LEN(de->name_len); |
3d0518f4 | 1332 | rlen = ext4_rec_len_from_disk(de->rec_len, blocksize); |
ac27a0ec | 1333 | if (de->inode) { |
617ba13b | 1334 | struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen); |
3d0518f4 WY |
1335 | de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize); |
1336 | de->rec_len = ext4_rec_len_to_disk(nlen, blocksize); | |
ac27a0ec DK |
1337 | de = de1; |
1338 | } | |
617ba13b | 1339 | de->file_type = EXT4_FT_UNKNOWN; |
ac27a0ec DK |
1340 | if (inode) { |
1341 | de->inode = cpu_to_le32(inode->i_ino); | |
617ba13b | 1342 | ext4_set_de_type(dir->i_sb, de, inode->i_mode); |
ac27a0ec DK |
1343 | } else |
1344 | de->inode = 0; | |
1345 | de->name_len = namelen; | |
af5bc92d | 1346 | memcpy(de->name, name, namelen); |
ac27a0ec DK |
1347 | /* |
1348 | * XXX shouldn't update any times until successful | |
1349 | * completion of syscall, but too many callers depend | |
1350 | * on this. | |
1351 | * | |
1352 | * XXX similarly, too many callers depend on | |
617ba13b | 1353 | * ext4_new_inode() setting the times, but error |
ac27a0ec DK |
1354 | * recovery deletes the inode, so the worst that can |
1355 | * happen is that the times are slightly out of date | |
1356 | * and/or different from the directory change time. | |
1357 | */ | |
ef7f3835 | 1358 | dir->i_mtime = dir->i_ctime = ext4_current_time(dir); |
617ba13b | 1359 | ext4_update_dx_flag(dir); |
ac27a0ec | 1360 | dir->i_version++; |
617ba13b | 1361 | ext4_mark_inode_dirty(handle, dir); |
0390131b FM |
1362 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
1363 | err = ext4_handle_dirty_metadata(handle, dir, bh); | |
ac27a0ec | 1364 | if (err) |
617ba13b | 1365 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1366 | return 0; |
1367 | } | |
1368 | ||
ac27a0ec DK |
1369 | /* |
1370 | * This converts a one block unindexed directory to a 3 block indexed | |
1371 | * directory, and adds the dentry to the indexed directory. | |
1372 | */ | |
1373 | static int make_indexed_dir(handle_t *handle, struct dentry *dentry, | |
1374 | struct inode *inode, struct buffer_head *bh) | |
1375 | { | |
1376 | struct inode *dir = dentry->d_parent->d_inode; | |
1377 | const char *name = dentry->d_name.name; | |
1378 | int namelen = dentry->d_name.len; | |
1379 | struct buffer_head *bh2; | |
1380 | struct dx_root *root; | |
1381 | struct dx_frame frames[2], *frame; | |
1382 | struct dx_entry *entries; | |
617ba13b | 1383 | struct ext4_dir_entry_2 *de, *de2; |
ac27a0ec DK |
1384 | char *data1, *top; |
1385 | unsigned len; | |
1386 | int retval; | |
1387 | unsigned blocksize; | |
1388 | struct dx_hash_info hinfo; | |
725d26d3 | 1389 | ext4_lblk_t block; |
ac27a0ec DK |
1390 | struct fake_dirent *fde; |
1391 | ||
1392 | blocksize = dir->i_sb->s_blocksize; | |
e6b8bc09 | 1393 | dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino)); |
617ba13b | 1394 | retval = ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1395 | if (retval) { |
617ba13b | 1396 | ext4_std_error(dir->i_sb, retval); |
ac27a0ec DK |
1397 | brelse(bh); |
1398 | return retval; | |
1399 | } | |
1400 | root = (struct dx_root *) bh->b_data; | |
1401 | ||
e6b8bc09 TT |
1402 | /* The 0th block becomes the root, move the dirents out */ |
1403 | fde = &root->dotdot; | |
1404 | de = (struct ext4_dir_entry_2 *)((char *)fde + | |
3d0518f4 | 1405 | ext4_rec_len_from_disk(fde->rec_len, blocksize)); |
e6b8bc09 | 1406 | if ((char *) de >= (((char *) root) + blocksize)) { |
24676da4 | 1407 | EXT4_ERROR_INODE(dir, "invalid rec_len for '..'"); |
e6b8bc09 TT |
1408 | brelse(bh); |
1409 | return -EIO; | |
1410 | } | |
1411 | len = ((char *) root) + blocksize - (char *) de; | |
1412 | ||
1413 | /* Allocate new block for the 0th block's dirents */ | |
af5bc92d | 1414 | bh2 = ext4_append(handle, dir, &block, &retval); |
ac27a0ec DK |
1415 | if (!(bh2)) { |
1416 | brelse(bh); | |
1417 | return retval; | |
1418 | } | |
12e9b892 | 1419 | ext4_set_inode_flag(dir, EXT4_INODE_INDEX); |
ac27a0ec DK |
1420 | data1 = bh2->b_data; |
1421 | ||
ac27a0ec | 1422 | memcpy (data1, de, len); |
617ba13b | 1423 | de = (struct ext4_dir_entry_2 *) data1; |
ac27a0ec | 1424 | top = data1 + len; |
3d0518f4 | 1425 | while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) |
ac27a0ec | 1426 | de = de2; |
3d0518f4 WY |
1427 | de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de, |
1428 | blocksize); | |
ac27a0ec | 1429 | /* Initialize the root; the dot dirents already exist */ |
617ba13b | 1430 | de = (struct ext4_dir_entry_2 *) (&root->dotdot); |
3d0518f4 WY |
1431 | de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2), |
1432 | blocksize); | |
ac27a0ec DK |
1433 | memset (&root->info, 0, sizeof(root->info)); |
1434 | root->info.info_length = sizeof(root->info); | |
617ba13b | 1435 | root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; |
ac27a0ec | 1436 | entries = root->entries; |
af5bc92d TT |
1437 | dx_set_block(entries, 1); |
1438 | dx_set_count(entries, 1); | |
1439 | dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); | |
ac27a0ec DK |
1440 | |
1441 | /* Initialize as for dx_probe */ | |
1442 | hinfo.hash_version = root->info.hash_version; | |
f99b2589 TT |
1443 | if (hinfo.hash_version <= DX_HASH_TEA) |
1444 | hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b MC |
1445 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
1446 | ext4fs_dirhash(name, namelen, &hinfo); | |
ac27a0ec DK |
1447 | frame = frames; |
1448 | frame->entries = entries; | |
1449 | frame->at = entries; | |
1450 | frame->bh = bh; | |
1451 | bh = bh2; | |
1452 | de = do_split(handle,dir, &bh, frame, &hinfo, &retval); | |
1453 | dx_release (frames); | |
1454 | if (!(de)) | |
1455 | return retval; | |
1456 | ||
2de770a4 TT |
1457 | retval = add_dirent_to_buf(handle, dentry, inode, de, bh); |
1458 | brelse(bh); | |
1459 | return retval; | |
ac27a0ec | 1460 | } |
ac27a0ec DK |
1461 | |
1462 | /* | |
617ba13b | 1463 | * ext4_add_entry() |
ac27a0ec DK |
1464 | * |
1465 | * adds a file entry to the specified directory, using the same | |
617ba13b | 1466 | * semantics as ext4_find_entry(). It returns NULL if it failed. |
ac27a0ec DK |
1467 | * |
1468 | * NOTE!! The inode part of 'de' is left at 0 - which means you | |
1469 | * may not sleep between calling this and putting something into | |
1470 | * the entry, as someone else might have used it while you slept. | |
1471 | */ | |
af5bc92d TT |
1472 | static int ext4_add_entry(handle_t *handle, struct dentry *dentry, |
1473 | struct inode *inode) | |
ac27a0ec DK |
1474 | { |
1475 | struct inode *dir = dentry->d_parent->d_inode; | |
af5bc92d | 1476 | struct buffer_head *bh; |
617ba13b | 1477 | struct ext4_dir_entry_2 *de; |
af5bc92d | 1478 | struct super_block *sb; |
ac27a0ec | 1479 | int retval; |
ac27a0ec | 1480 | int dx_fallback=0; |
ac27a0ec | 1481 | unsigned blocksize; |
725d26d3 | 1482 | ext4_lblk_t block, blocks; |
ac27a0ec DK |
1483 | |
1484 | sb = dir->i_sb; | |
1485 | blocksize = sb->s_blocksize; | |
1486 | if (!dentry->d_name.len) | |
1487 | return -EINVAL; | |
ac27a0ec | 1488 | if (is_dx(dir)) { |
617ba13b | 1489 | retval = ext4_dx_add_entry(handle, dentry, inode); |
ac27a0ec DK |
1490 | if (!retval || (retval != ERR_BAD_DX_DIR)) |
1491 | return retval; | |
12e9b892 | 1492 | ext4_clear_inode_flag(dir, EXT4_INODE_INDEX); |
ac27a0ec | 1493 | dx_fallback++; |
617ba13b | 1494 | ext4_mark_inode_dirty(handle, dir); |
ac27a0ec | 1495 | } |
ac27a0ec | 1496 | blocks = dir->i_size >> sb->s_blocksize_bits; |
498e5f24 | 1497 | for (block = 0; block < blocks; block++) { |
617ba13b | 1498 | bh = ext4_bread(handle, dir, block, 0, &retval); |
ac27a0ec DK |
1499 | if(!bh) |
1500 | return retval; | |
1501 | retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
2de770a4 TT |
1502 | if (retval != -ENOSPC) { |
1503 | brelse(bh); | |
ac27a0ec | 1504 | return retval; |
2de770a4 | 1505 | } |
ac27a0ec | 1506 | |
ac27a0ec | 1507 | if (blocks == 1 && !dx_fallback && |
1e424a34 TT |
1508 | EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) |
1509 | return make_indexed_dir(handle, dentry, inode, bh); | |
ac27a0ec DK |
1510 | brelse(bh); |
1511 | } | |
617ba13b | 1512 | bh = ext4_append(handle, dir, &block, &retval); |
ac27a0ec DK |
1513 | if (!bh) |
1514 | return retval; | |
617ba13b | 1515 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 1516 | de->inode = 0; |
3d0518f4 | 1517 | de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize); |
2de770a4 TT |
1518 | retval = add_dirent_to_buf(handle, dentry, inode, de, bh); |
1519 | brelse(bh); | |
14ece102 FM |
1520 | if (retval == 0) |
1521 | ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); | |
2de770a4 | 1522 | return retval; |
ac27a0ec DK |
1523 | } |
1524 | ||
ac27a0ec DK |
1525 | /* |
1526 | * Returns 0 for success, or a negative error value | |
1527 | */ | |
617ba13b | 1528 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, |
ac27a0ec DK |
1529 | struct inode *inode) |
1530 | { | |
1531 | struct dx_frame frames[2], *frame; | |
1532 | struct dx_entry *entries, *at; | |
1533 | struct dx_hash_info hinfo; | |
af5bc92d | 1534 | struct buffer_head *bh; |
ac27a0ec | 1535 | struct inode *dir = dentry->d_parent->d_inode; |
af5bc92d | 1536 | struct super_block *sb = dir->i_sb; |
617ba13b | 1537 | struct ext4_dir_entry_2 *de; |
ac27a0ec DK |
1538 | int err; |
1539 | ||
f702ba0f | 1540 | frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err); |
ac27a0ec DK |
1541 | if (!frame) |
1542 | return err; | |
1543 | entries = frame->entries; | |
1544 | at = frame->at; | |
1545 | ||
617ba13b | 1546 | if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err))) |
ac27a0ec DK |
1547 | goto cleanup; |
1548 | ||
1549 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1550 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1551 | if (err) |
1552 | goto journal_error; | |
1553 | ||
1554 | err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
2de770a4 | 1555 | if (err != -ENOSPC) |
ac27a0ec | 1556 | goto cleanup; |
ac27a0ec DK |
1557 | |
1558 | /* Block full, should compress but for now just split */ | |
4776004f | 1559 | dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", |
ac27a0ec DK |
1560 | dx_get_count(entries), dx_get_limit(entries))); |
1561 | /* Need to split index? */ | |
1562 | if (dx_get_count(entries) == dx_get_limit(entries)) { | |
725d26d3 | 1563 | ext4_lblk_t newblock; |
ac27a0ec DK |
1564 | unsigned icount = dx_get_count(entries); |
1565 | int levels = frame - frames; | |
1566 | struct dx_entry *entries2; | |
1567 | struct dx_node *node2; | |
1568 | struct buffer_head *bh2; | |
1569 | ||
1570 | if (levels && (dx_get_count(frames->entries) == | |
1571 | dx_get_limit(frames->entries))) { | |
12062ddd | 1572 | ext4_warning(sb, "Directory index full!"); |
ac27a0ec DK |
1573 | err = -ENOSPC; |
1574 | goto cleanup; | |
1575 | } | |
617ba13b | 1576 | bh2 = ext4_append (handle, dir, &newblock, &err); |
ac27a0ec DK |
1577 | if (!(bh2)) |
1578 | goto cleanup; | |
1579 | node2 = (struct dx_node *)(bh2->b_data); | |
1580 | entries2 = node2->entries; | |
1f7bebb9 | 1581 | memset(&node2->fake, 0, sizeof(struct fake_dirent)); |
3d0518f4 WY |
1582 | node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize, |
1583 | sb->s_blocksize); | |
ac27a0ec | 1584 | BUFFER_TRACE(frame->bh, "get_write_access"); |
617ba13b | 1585 | err = ext4_journal_get_write_access(handle, frame->bh); |
ac27a0ec DK |
1586 | if (err) |
1587 | goto journal_error; | |
1588 | if (levels) { | |
1589 | unsigned icount1 = icount/2, icount2 = icount - icount1; | |
1590 | unsigned hash2 = dx_get_hash(entries + icount1); | |
4776004f TT |
1591 | dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", |
1592 | icount1, icount2)); | |
ac27a0ec DK |
1593 | |
1594 | BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ | |
617ba13b | 1595 | err = ext4_journal_get_write_access(handle, |
ac27a0ec DK |
1596 | frames[0].bh); |
1597 | if (err) | |
1598 | goto journal_error; | |
1599 | ||
af5bc92d TT |
1600 | memcpy((char *) entries2, (char *) (entries + icount1), |
1601 | icount2 * sizeof(struct dx_entry)); | |
1602 | dx_set_count(entries, icount1); | |
1603 | dx_set_count(entries2, icount2); | |
1604 | dx_set_limit(entries2, dx_node_limit(dir)); | |
ac27a0ec DK |
1605 | |
1606 | /* Which index block gets the new entry? */ | |
1607 | if (at - entries >= icount1) { | |
1608 | frame->at = at = at - entries - icount1 + entries2; | |
1609 | frame->entries = entries = entries2; | |
1610 | swap(frame->bh, bh2); | |
1611 | } | |
af5bc92d TT |
1612 | dx_insert_block(frames + 0, hash2, newblock); |
1613 | dxtrace(dx_show_index("node", frames[1].entries)); | |
1614 | dxtrace(dx_show_index("node", | |
ac27a0ec | 1615 | ((struct dx_node *) bh2->b_data)->entries)); |
0390131b | 1616 | err = ext4_handle_dirty_metadata(handle, inode, bh2); |
ac27a0ec DK |
1617 | if (err) |
1618 | goto journal_error; | |
1619 | brelse (bh2); | |
1620 | } else { | |
4776004f TT |
1621 | dxtrace(printk(KERN_DEBUG |
1622 | "Creating second level index...\n")); | |
ac27a0ec DK |
1623 | memcpy((char *) entries2, (char *) entries, |
1624 | icount * sizeof(struct dx_entry)); | |
1625 | dx_set_limit(entries2, dx_node_limit(dir)); | |
1626 | ||
1627 | /* Set up root */ | |
1628 | dx_set_count(entries, 1); | |
1629 | dx_set_block(entries + 0, newblock); | |
1630 | ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; | |
1631 | ||
1632 | /* Add new access path frame */ | |
1633 | frame = frames + 1; | |
1634 | frame->at = at = at - entries + entries2; | |
1635 | frame->entries = entries = entries2; | |
1636 | frame->bh = bh2; | |
617ba13b | 1637 | err = ext4_journal_get_write_access(handle, |
ac27a0ec DK |
1638 | frame->bh); |
1639 | if (err) | |
1640 | goto journal_error; | |
1641 | } | |
0390131b | 1642 | ext4_handle_dirty_metadata(handle, inode, frames[0].bh); |
ac27a0ec DK |
1643 | } |
1644 | de = do_split(handle, dir, &bh, frame, &hinfo, &err); | |
1645 | if (!de) | |
1646 | goto cleanup; | |
1647 | err = add_dirent_to_buf(handle, dentry, inode, de, bh); | |
ac27a0ec DK |
1648 | goto cleanup; |
1649 | ||
1650 | journal_error: | |
617ba13b | 1651 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1652 | cleanup: |
1653 | if (bh) | |
1654 | brelse(bh); | |
1655 | dx_release(frames); | |
1656 | return err; | |
1657 | } | |
ac27a0ec DK |
1658 | |
1659 | /* | |
617ba13b | 1660 | * ext4_delete_entry deletes a directory entry by merging it with the |
ac27a0ec DK |
1661 | * previous entry |
1662 | */ | |
af5bc92d TT |
1663 | static int ext4_delete_entry(handle_t *handle, |
1664 | struct inode *dir, | |
1665 | struct ext4_dir_entry_2 *de_del, | |
1666 | struct buffer_head *bh) | |
ac27a0ec | 1667 | { |
af5bc92d | 1668 | struct ext4_dir_entry_2 *de, *pde; |
3d0518f4 | 1669 | unsigned int blocksize = dir->i_sb->s_blocksize; |
ac27a0ec DK |
1670 | int i; |
1671 | ||
1672 | i = 0; | |
1673 | pde = NULL; | |
617ba13b | 1674 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 1675 | while (i < bh->b_size) { |
617ba13b | 1676 | if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i)) |
ac27a0ec DK |
1677 | return -EIO; |
1678 | if (de == de_del) { | |
1679 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1680 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1681 | if (pde) |
a72d7f83 | 1682 | pde->rec_len = ext4_rec_len_to_disk( |
3d0518f4 WY |
1683 | ext4_rec_len_from_disk(pde->rec_len, |
1684 | blocksize) + | |
1685 | ext4_rec_len_from_disk(de->rec_len, | |
1686 | blocksize), | |
1687 | blocksize); | |
ac27a0ec DK |
1688 | else |
1689 | de->inode = 0; | |
1690 | dir->i_version++; | |
0390131b FM |
1691 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
1692 | ext4_handle_dirty_metadata(handle, dir, bh); | |
ac27a0ec DK |
1693 | return 0; |
1694 | } | |
3d0518f4 | 1695 | i += ext4_rec_len_from_disk(de->rec_len, blocksize); |
ac27a0ec | 1696 | pde = de; |
3d0518f4 | 1697 | de = ext4_next_entry(de, blocksize); |
ac27a0ec DK |
1698 | } |
1699 | return -ENOENT; | |
1700 | } | |
1701 | ||
f8628a14 AD |
1702 | /* |
1703 | * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2, | |
1704 | * since this indicates that nlinks count was previously 1. | |
1705 | */ | |
1706 | static void ext4_inc_count(handle_t *handle, struct inode *inode) | |
1707 | { | |
1708 | inc_nlink(inode); | |
1709 | if (is_dx(inode) && inode->i_nlink > 1) { | |
1710 | /* limit is 16-bit i_links_count */ | |
1711 | if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) { | |
1712 | inode->i_nlink = 1; | |
1713 | EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb, | |
1714 | EXT4_FEATURE_RO_COMPAT_DIR_NLINK); | |
1715 | } | |
1716 | } | |
1717 | } | |
1718 | ||
1719 | /* | |
1720 | * If a directory had nlink == 1, then we should let it be 1. This indicates | |
1721 | * directory has >EXT4_LINK_MAX subdirs. | |
1722 | */ | |
1723 | static void ext4_dec_count(handle_t *handle, struct inode *inode) | |
1724 | { | |
1725 | drop_nlink(inode); | |
1726 | if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0) | |
1727 | inc_nlink(inode); | |
1728 | } | |
1729 | ||
1730 | ||
617ba13b | 1731 | static int ext4_add_nondir(handle_t *handle, |
ac27a0ec DK |
1732 | struct dentry *dentry, struct inode *inode) |
1733 | { | |
617ba13b | 1734 | int err = ext4_add_entry(handle, dentry, inode); |
ac27a0ec | 1735 | if (!err) { |
617ba13b | 1736 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 1737 | d_instantiate(dentry, inode); |
6b38e842 | 1738 | unlock_new_inode(inode); |
ac27a0ec DK |
1739 | return 0; |
1740 | } | |
731b9a54 | 1741 | drop_nlink(inode); |
6b38e842 | 1742 | unlock_new_inode(inode); |
ac27a0ec DK |
1743 | iput(inode); |
1744 | return err; | |
1745 | } | |
1746 | ||
1747 | /* | |
1748 | * By the time this is called, we already have created | |
1749 | * the directory cache entry for the new file, but it | |
1750 | * is so far negative - it has no inode. | |
1751 | * | |
1752 | * If the create succeeds, we fill in the inode information | |
1753 | * with d_instantiate(). | |
1754 | */ | |
af5bc92d TT |
1755 | static int ext4_create(struct inode *dir, struct dentry *dentry, int mode, |
1756 | struct nameidata *nd) | |
ac27a0ec DK |
1757 | { |
1758 | handle_t *handle; | |
af5bc92d | 1759 | struct inode *inode; |
ac27a0ec DK |
1760 | int err, retries = 0; |
1761 | ||
871a2931 | 1762 | dquot_initialize(dir); |
907f4554 | 1763 | |
ac27a0ec | 1764 | retry: |
617ba13b MC |
1765 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
1766 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
5aca07eb | 1767 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); |
ac27a0ec DK |
1768 | if (IS_ERR(handle)) |
1769 | return PTR_ERR(handle); | |
1770 | ||
1771 | if (IS_DIRSYNC(dir)) | |
0390131b | 1772 | ext4_handle_sync(handle); |
ac27a0ec | 1773 | |
11013911 | 1774 | inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0); |
ac27a0ec DK |
1775 | err = PTR_ERR(inode); |
1776 | if (!IS_ERR(inode)) { | |
617ba13b MC |
1777 | inode->i_op = &ext4_file_inode_operations; |
1778 | inode->i_fop = &ext4_file_operations; | |
1779 | ext4_set_aops(inode); | |
1780 | err = ext4_add_nondir(handle, dentry, inode); | |
ac27a0ec | 1781 | } |
617ba13b MC |
1782 | ext4_journal_stop(handle); |
1783 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
1784 | goto retry; |
1785 | return err; | |
1786 | } | |
1787 | ||
af5bc92d TT |
1788 | static int ext4_mknod(struct inode *dir, struct dentry *dentry, |
1789 | int mode, dev_t rdev) | |
ac27a0ec DK |
1790 | { |
1791 | handle_t *handle; | |
1792 | struct inode *inode; | |
1793 | int err, retries = 0; | |
1794 | ||
1795 | if (!new_valid_dev(rdev)) | |
1796 | return -EINVAL; | |
1797 | ||
871a2931 | 1798 | dquot_initialize(dir); |
907f4554 | 1799 | |
ac27a0ec | 1800 | retry: |
617ba13b MC |
1801 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
1802 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
5aca07eb | 1803 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); |
ac27a0ec DK |
1804 | if (IS_ERR(handle)) |
1805 | return PTR_ERR(handle); | |
1806 | ||
1807 | if (IS_DIRSYNC(dir)) | |
0390131b | 1808 | ext4_handle_sync(handle); |
ac27a0ec | 1809 | |
11013911 | 1810 | inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0); |
ac27a0ec DK |
1811 | err = PTR_ERR(inode); |
1812 | if (!IS_ERR(inode)) { | |
1813 | init_special_inode(inode, inode->i_mode, rdev); | |
03010a33 | 1814 | #ifdef CONFIG_EXT4_FS_XATTR |
617ba13b | 1815 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec | 1816 | #endif |
617ba13b | 1817 | err = ext4_add_nondir(handle, dentry, inode); |
ac27a0ec | 1818 | } |
617ba13b MC |
1819 | ext4_journal_stop(handle); |
1820 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
1821 | goto retry; |
1822 | return err; | |
1823 | } | |
1824 | ||
af5bc92d | 1825 | static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
ac27a0ec DK |
1826 | { |
1827 | handle_t *handle; | |
af5bc92d TT |
1828 | struct inode *inode; |
1829 | struct buffer_head *dir_block; | |
1830 | struct ext4_dir_entry_2 *de; | |
3d0518f4 | 1831 | unsigned int blocksize = dir->i_sb->s_blocksize; |
ac27a0ec DK |
1832 | int err, retries = 0; |
1833 | ||
f8628a14 | 1834 | if (EXT4_DIR_LINK_MAX(dir)) |
ac27a0ec DK |
1835 | return -EMLINK; |
1836 | ||
871a2931 | 1837 | dquot_initialize(dir); |
907f4554 | 1838 | |
ac27a0ec | 1839 | retry: |
617ba13b MC |
1840 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
1841 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
5aca07eb | 1842 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); |
ac27a0ec DK |
1843 | if (IS_ERR(handle)) |
1844 | return PTR_ERR(handle); | |
1845 | ||
1846 | if (IS_DIRSYNC(dir)) | |
0390131b | 1847 | ext4_handle_sync(handle); |
ac27a0ec | 1848 | |
11013911 AD |
1849 | inode = ext4_new_inode(handle, dir, S_IFDIR | mode, |
1850 | &dentry->d_name, 0); | |
ac27a0ec DK |
1851 | err = PTR_ERR(inode); |
1852 | if (IS_ERR(inode)) | |
1853 | goto out_stop; | |
1854 | ||
617ba13b MC |
1855 | inode->i_op = &ext4_dir_inode_operations; |
1856 | inode->i_fop = &ext4_dir_operations; | |
1857 | inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize; | |
af5bc92d | 1858 | dir_block = ext4_bread(handle, inode, 0, 1, &err); |
4cdeed86 AK |
1859 | if (!dir_block) |
1860 | goto out_clear_inode; | |
ac27a0ec | 1861 | BUFFER_TRACE(dir_block, "get_write_access"); |
617ba13b MC |
1862 | ext4_journal_get_write_access(handle, dir_block); |
1863 | de = (struct ext4_dir_entry_2 *) dir_block->b_data; | |
ac27a0ec DK |
1864 | de->inode = cpu_to_le32(inode->i_ino); |
1865 | de->name_len = 1; | |
3d0518f4 WY |
1866 | de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len), |
1867 | blocksize); | |
af5bc92d | 1868 | strcpy(de->name, "."); |
617ba13b | 1869 | ext4_set_de_type(dir->i_sb, de, S_IFDIR); |
3d0518f4 | 1870 | de = ext4_next_entry(de, blocksize); |
ac27a0ec | 1871 | de->inode = cpu_to_le32(dir->i_ino); |
3d0518f4 WY |
1872 | de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1), |
1873 | blocksize); | |
ac27a0ec | 1874 | de->name_len = 2; |
af5bc92d | 1875 | strcpy(de->name, ".."); |
617ba13b | 1876 | ext4_set_de_type(dir->i_sb, de, S_IFDIR); |
ac27a0ec | 1877 | inode->i_nlink = 2; |
0390131b FM |
1878 | BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); |
1879 | ext4_handle_dirty_metadata(handle, dir, dir_block); | |
af5bc92d | 1880 | brelse(dir_block); |
617ba13b | 1881 | ext4_mark_inode_dirty(handle, inode); |
af5bc92d | 1882 | err = ext4_add_entry(handle, dentry, inode); |
ac27a0ec | 1883 | if (err) { |
4cdeed86 AK |
1884 | out_clear_inode: |
1885 | clear_nlink(inode); | |
6b38e842 | 1886 | unlock_new_inode(inode); |
617ba13b | 1887 | ext4_mark_inode_dirty(handle, inode); |
af5bc92d | 1888 | iput(inode); |
ac27a0ec DK |
1889 | goto out_stop; |
1890 | } | |
f8628a14 | 1891 | ext4_inc_count(handle, dir); |
617ba13b MC |
1892 | ext4_update_dx_flag(dir); |
1893 | ext4_mark_inode_dirty(handle, dir); | |
ac27a0ec | 1894 | d_instantiate(dentry, inode); |
6b38e842 | 1895 | unlock_new_inode(inode); |
ac27a0ec | 1896 | out_stop: |
617ba13b MC |
1897 | ext4_journal_stop(handle); |
1898 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
1899 | goto retry; |
1900 | return err; | |
1901 | } | |
1902 | ||
1903 | /* | |
1904 | * routine to check that the specified directory is empty (for rmdir) | |
1905 | */ | |
af5bc92d | 1906 | static int empty_dir(struct inode *inode) |
ac27a0ec | 1907 | { |
498e5f24 | 1908 | unsigned int offset; |
af5bc92d TT |
1909 | struct buffer_head *bh; |
1910 | struct ext4_dir_entry_2 *de, *de1; | |
1911 | struct super_block *sb; | |
ac27a0ec DK |
1912 | int err = 0; |
1913 | ||
1914 | sb = inode->i_sb; | |
617ba13b | 1915 | if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) || |
af5bc92d | 1916 | !(bh = ext4_bread(NULL, inode, 0, 0, &err))) { |
ac27a0ec | 1917 | if (err) |
24676da4 TT |
1918 | EXT4_ERROR_INODE(inode, |
1919 | "error %d reading directory lblock 0", err); | |
ac27a0ec | 1920 | else |
12062ddd | 1921 | ext4_warning(inode->i_sb, |
ac27a0ec DK |
1922 | "bad directory (dir #%lu) - no data block", |
1923 | inode->i_ino); | |
1924 | return 1; | |
1925 | } | |
617ba13b | 1926 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
3d0518f4 | 1927 | de1 = ext4_next_entry(de, sb->s_blocksize); |
ac27a0ec DK |
1928 | if (le32_to_cpu(de->inode) != inode->i_ino || |
1929 | !le32_to_cpu(de1->inode) || | |
af5bc92d TT |
1930 | strcmp(".", de->name) || |
1931 | strcmp("..", de1->name)) { | |
12062ddd | 1932 | ext4_warning(inode->i_sb, |
af5bc92d TT |
1933 | "bad directory (dir #%lu) - no `.' or `..'", |
1934 | inode->i_ino); | |
1935 | brelse(bh); | |
ac27a0ec DK |
1936 | return 1; |
1937 | } | |
3d0518f4 WY |
1938 | offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) + |
1939 | ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize); | |
1940 | de = ext4_next_entry(de1, sb->s_blocksize); | |
af5bc92d | 1941 | while (offset < inode->i_size) { |
ac27a0ec | 1942 | if (!bh || |
24676da4 TT |
1943 | (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { |
1944 | unsigned int lblock; | |
ac27a0ec | 1945 | err = 0; |
af5bc92d | 1946 | brelse(bh); |
24676da4 TT |
1947 | lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb); |
1948 | bh = ext4_bread(NULL, inode, lblock, 0, &err); | |
ac27a0ec DK |
1949 | if (!bh) { |
1950 | if (err) | |
24676da4 TT |
1951 | EXT4_ERROR_INODE(inode, |
1952 | "error %d reading directory " | |
1953 | "lblock %u", err, lblock); | |
ac27a0ec DK |
1954 | offset += sb->s_blocksize; |
1955 | continue; | |
1956 | } | |
617ba13b | 1957 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 1958 | } |
617ba13b MC |
1959 | if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) { |
1960 | de = (struct ext4_dir_entry_2 *)(bh->b_data + | |
ac27a0ec DK |
1961 | sb->s_blocksize); |
1962 | offset = (offset | (sb->s_blocksize - 1)) + 1; | |
1963 | continue; | |
1964 | } | |
1965 | if (le32_to_cpu(de->inode)) { | |
af5bc92d | 1966 | brelse(bh); |
ac27a0ec DK |
1967 | return 0; |
1968 | } | |
3d0518f4 WY |
1969 | offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); |
1970 | de = ext4_next_entry(de, sb->s_blocksize); | |
ac27a0ec | 1971 | } |
af5bc92d | 1972 | brelse(bh); |
ac27a0ec DK |
1973 | return 1; |
1974 | } | |
1975 | ||
617ba13b | 1976 | /* ext4_orphan_add() links an unlinked or truncated inode into a list of |
ac27a0ec DK |
1977 | * such inodes, starting at the superblock, in case we crash before the |
1978 | * file is closed/deleted, or in case the inode truncate spans multiple | |
1979 | * transactions and the last transaction is not recovered after a crash. | |
1980 | * | |
1981 | * At filesystem recovery time, we walk this list deleting unlinked | |
617ba13b | 1982 | * inodes and truncating linked inodes in ext4_orphan_cleanup(). |
ac27a0ec | 1983 | */ |
617ba13b | 1984 | int ext4_orphan_add(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
1985 | { |
1986 | struct super_block *sb = inode->i_sb; | |
617ba13b | 1987 | struct ext4_iloc iloc; |
ac27a0ec DK |
1988 | int err = 0, rc; |
1989 | ||
0390131b FM |
1990 | if (!ext4_handle_valid(handle)) |
1991 | return 0; | |
1992 | ||
3b9d4ed2 | 1993 | mutex_lock(&EXT4_SB(sb)->s_orphan_lock); |
617ba13b | 1994 | if (!list_empty(&EXT4_I(inode)->i_orphan)) |
ac27a0ec DK |
1995 | goto out_unlock; |
1996 | ||
1997 | /* Orphan handling is only valid for files with data blocks | |
1998 | * being truncated, or files being unlinked. */ | |
1999 | ||
2000 | /* @@@ FIXME: Observation from aviro: | |
617ba13b | 2001 | * I think I can trigger J_ASSERT in ext4_orphan_add(). We block |
3b9d4ed2 | 2002 | * here (on s_orphan_lock), so race with ext4_link() which might bump |
ac27a0ec DK |
2003 | * ->i_nlink. For, say it, character device. Not a regular file, |
2004 | * not a directory, not a symlink and ->i_nlink > 0. | |
3b9d4ed2 TT |
2005 | * |
2006 | * tytso, 4/25/2009: I'm not sure how that could happen; | |
2007 | * shouldn't the fs core protect us from these sort of | |
2008 | * unlink()/link() races? | |
ac27a0ec | 2009 | */ |
af5bc92d TT |
2010 | J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
2011 | S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); | |
ac27a0ec | 2012 | |
617ba13b MC |
2013 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); |
2014 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); | |
ac27a0ec DK |
2015 | if (err) |
2016 | goto out_unlock; | |
2017 | ||
617ba13b | 2018 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
ac27a0ec DK |
2019 | if (err) |
2020 | goto out_unlock; | |
6e3617e5 DM |
2021 | /* |
2022 | * Due to previous errors inode may be already a part of on-disk | |
2023 | * orphan list. If so skip on-disk list modification. | |
2024 | */ | |
2025 | if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <= | |
2026 | (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) | |
2027 | goto mem_insert; | |
ac27a0ec DK |
2028 | |
2029 | /* Insert this inode at the head of the on-disk orphan list... */ | |
617ba13b MC |
2030 | NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan); |
2031 | EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); | |
73b50c1c | 2032 | err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); |
617ba13b | 2033 | rc = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
2034 | if (!err) |
2035 | err = rc; | |
2036 | ||
2037 | /* Only add to the head of the in-memory list if all the | |
2038 | * previous operations succeeded. If the orphan_add is going to | |
2039 | * fail (possibly taking the journal offline), we can't risk | |
2040 | * leaving the inode on the orphan list: stray orphan-list | |
2041 | * entries can cause panics at unmount time. | |
2042 | * | |
2043 | * This is safe: on error we're going to ignore the orphan list | |
2044 | * anyway on the next recovery. */ | |
6e3617e5 | 2045 | mem_insert: |
ac27a0ec | 2046 | if (!err) |
617ba13b | 2047 | list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); |
ac27a0ec DK |
2048 | |
2049 | jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); | |
2050 | jbd_debug(4, "orphan inode %lu will point to %d\n", | |
2051 | inode->i_ino, NEXT_ORPHAN(inode)); | |
2052 | out_unlock: | |
3b9d4ed2 | 2053 | mutex_unlock(&EXT4_SB(sb)->s_orphan_lock); |
617ba13b | 2054 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
2055 | return err; |
2056 | } | |
2057 | ||
2058 | /* | |
617ba13b | 2059 | * ext4_orphan_del() removes an unlinked or truncated inode from the list |
ac27a0ec DK |
2060 | * of such inodes stored on disk, because it is finally being cleaned up. |
2061 | */ | |
617ba13b | 2062 | int ext4_orphan_del(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
2063 | { |
2064 | struct list_head *prev; | |
617ba13b MC |
2065 | struct ext4_inode_info *ei = EXT4_I(inode); |
2066 | struct ext4_sb_info *sbi; | |
498e5f24 | 2067 | __u32 ino_next; |
617ba13b | 2068 | struct ext4_iloc iloc; |
ac27a0ec DK |
2069 | int err = 0; |
2070 | ||
d3d1faf6 CW |
2071 | /* ext4_handle_valid() assumes a valid handle_t pointer */ |
2072 | if (handle && !ext4_handle_valid(handle)) | |
0390131b FM |
2073 | return 0; |
2074 | ||
3b9d4ed2 TT |
2075 | mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock); |
2076 | if (list_empty(&ei->i_orphan)) | |
2077 | goto out; | |
ac27a0ec DK |
2078 | |
2079 | ino_next = NEXT_ORPHAN(inode); | |
2080 | prev = ei->i_orphan.prev; | |
617ba13b | 2081 | sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
2082 | |
2083 | jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); | |
2084 | ||
2085 | list_del_init(&ei->i_orphan); | |
2086 | ||
2087 | /* If we're on an error path, we may not have a valid | |
2088 | * transaction handle with which to update the orphan list on | |
2089 | * disk, but we still need to remove the inode from the linked | |
2090 | * list in memory. */ | |
0390131b | 2091 | if (sbi->s_journal && !handle) |
ac27a0ec DK |
2092 | goto out; |
2093 | ||
617ba13b | 2094 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
ac27a0ec DK |
2095 | if (err) |
2096 | goto out_err; | |
2097 | ||
2098 | if (prev == &sbi->s_orphan) { | |
498e5f24 | 2099 | jbd_debug(4, "superblock will point to %u\n", ino_next); |
ac27a0ec | 2100 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
617ba13b | 2101 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
ac27a0ec DK |
2102 | if (err) |
2103 | goto out_brelse; | |
2104 | sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); | |
73b50c1c | 2105 | err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); |
ac27a0ec | 2106 | } else { |
617ba13b | 2107 | struct ext4_iloc iloc2; |
ac27a0ec | 2108 | struct inode *i_prev = |
617ba13b | 2109 | &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; |
ac27a0ec | 2110 | |
498e5f24 | 2111 | jbd_debug(4, "orphan inode %lu will point to %u\n", |
ac27a0ec | 2112 | i_prev->i_ino, ino_next); |
617ba13b | 2113 | err = ext4_reserve_inode_write(handle, i_prev, &iloc2); |
ac27a0ec DK |
2114 | if (err) |
2115 | goto out_brelse; | |
2116 | NEXT_ORPHAN(i_prev) = ino_next; | |
617ba13b | 2117 | err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); |
ac27a0ec DK |
2118 | } |
2119 | if (err) | |
2120 | goto out_brelse; | |
2121 | NEXT_ORPHAN(inode) = 0; | |
617ba13b | 2122 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
2123 | |
2124 | out_err: | |
617ba13b | 2125 | ext4_std_error(inode->i_sb, err); |
ac27a0ec | 2126 | out: |
3b9d4ed2 | 2127 | mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock); |
ac27a0ec DK |
2128 | return err; |
2129 | ||
2130 | out_brelse: | |
2131 | brelse(iloc.bh); | |
2132 | goto out_err; | |
2133 | } | |
2134 | ||
af5bc92d | 2135 | static int ext4_rmdir(struct inode *dir, struct dentry *dentry) |
ac27a0ec DK |
2136 | { |
2137 | int retval; | |
af5bc92d TT |
2138 | struct inode *inode; |
2139 | struct buffer_head *bh; | |
2140 | struct ext4_dir_entry_2 *de; | |
ac27a0ec DK |
2141 | handle_t *handle; |
2142 | ||
2143 | /* Initialize quotas before so that eventual writes go in | |
2144 | * separate transaction */ | |
871a2931 CH |
2145 | dquot_initialize(dir); |
2146 | dquot_initialize(dentry->d_inode); | |
907f4554 | 2147 | |
617ba13b | 2148 | handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb)); |
ac27a0ec DK |
2149 | if (IS_ERR(handle)) |
2150 | return PTR_ERR(handle); | |
2151 | ||
2152 | retval = -ENOENT; | |
f702ba0f | 2153 | bh = ext4_find_entry(dir, &dentry->d_name, &de); |
ac27a0ec DK |
2154 | if (!bh) |
2155 | goto end_rmdir; | |
2156 | ||
2157 | if (IS_DIRSYNC(dir)) | |
0390131b | 2158 | ext4_handle_sync(handle); |
ac27a0ec DK |
2159 | |
2160 | inode = dentry->d_inode; | |
2161 | ||
2162 | retval = -EIO; | |
2163 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2164 | goto end_rmdir; | |
2165 | ||
2166 | retval = -ENOTEMPTY; | |
af5bc92d | 2167 | if (!empty_dir(inode)) |
ac27a0ec DK |
2168 | goto end_rmdir; |
2169 | ||
617ba13b | 2170 | retval = ext4_delete_entry(handle, dir, de, bh); |
ac27a0ec DK |
2171 | if (retval) |
2172 | goto end_rmdir; | |
f8628a14 | 2173 | if (!EXT4_DIR_LINK_EMPTY(inode)) |
12062ddd | 2174 | ext4_warning(inode->i_sb, |
af5bc92d TT |
2175 | "empty directory has too many links (%d)", |
2176 | inode->i_nlink); | |
ac27a0ec DK |
2177 | inode->i_version++; |
2178 | clear_nlink(inode); | |
2179 | /* There's no need to set i_disksize: the fact that i_nlink is | |
2180 | * zero will ensure that the right thing happens during any | |
2181 | * recovery. */ | |
2182 | inode->i_size = 0; | |
617ba13b | 2183 | ext4_orphan_add(handle, inode); |
ef7f3835 | 2184 | inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode); |
617ba13b | 2185 | ext4_mark_inode_dirty(handle, inode); |
f8628a14 | 2186 | ext4_dec_count(handle, dir); |
617ba13b MC |
2187 | ext4_update_dx_flag(dir); |
2188 | ext4_mark_inode_dirty(handle, dir); | |
ac27a0ec DK |
2189 | |
2190 | end_rmdir: | |
617ba13b | 2191 | ext4_journal_stop(handle); |
af5bc92d | 2192 | brelse(bh); |
ac27a0ec DK |
2193 | return retval; |
2194 | } | |
2195 | ||
af5bc92d | 2196 | static int ext4_unlink(struct inode *dir, struct dentry *dentry) |
ac27a0ec DK |
2197 | { |
2198 | int retval; | |
af5bc92d TT |
2199 | struct inode *inode; |
2200 | struct buffer_head *bh; | |
2201 | struct ext4_dir_entry_2 *de; | |
ac27a0ec DK |
2202 | handle_t *handle; |
2203 | ||
2204 | /* Initialize quotas before so that eventual writes go | |
2205 | * in separate transaction */ | |
871a2931 CH |
2206 | dquot_initialize(dir); |
2207 | dquot_initialize(dentry->d_inode); | |
907f4554 | 2208 | |
617ba13b | 2209 | handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb)); |
ac27a0ec DK |
2210 | if (IS_ERR(handle)) |
2211 | return PTR_ERR(handle); | |
2212 | ||
2213 | if (IS_DIRSYNC(dir)) | |
0390131b | 2214 | ext4_handle_sync(handle); |
ac27a0ec DK |
2215 | |
2216 | retval = -ENOENT; | |
f702ba0f | 2217 | bh = ext4_find_entry(dir, &dentry->d_name, &de); |
ac27a0ec DK |
2218 | if (!bh) |
2219 | goto end_unlink; | |
2220 | ||
2221 | inode = dentry->d_inode; | |
2222 | ||
2223 | retval = -EIO; | |
2224 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2225 | goto end_unlink; | |
2226 | ||
2227 | if (!inode->i_nlink) { | |
12062ddd | 2228 | ext4_warning(inode->i_sb, |
af5bc92d TT |
2229 | "Deleting nonexistent file (%lu), %d", |
2230 | inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
2231 | inode->i_nlink = 1; |
2232 | } | |
617ba13b | 2233 | retval = ext4_delete_entry(handle, dir, de, bh); |
ac27a0ec DK |
2234 | if (retval) |
2235 | goto end_unlink; | |
ef7f3835 | 2236 | dir->i_ctime = dir->i_mtime = ext4_current_time(dir); |
617ba13b MC |
2237 | ext4_update_dx_flag(dir); |
2238 | ext4_mark_inode_dirty(handle, dir); | |
825f1481 | 2239 | drop_nlink(inode); |
ac27a0ec | 2240 | if (!inode->i_nlink) |
617ba13b | 2241 | ext4_orphan_add(handle, inode); |
ef7f3835 | 2242 | inode->i_ctime = ext4_current_time(inode); |
617ba13b | 2243 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
2244 | retval = 0; |
2245 | ||
2246 | end_unlink: | |
617ba13b | 2247 | ext4_journal_stop(handle); |
af5bc92d | 2248 | brelse(bh); |
ac27a0ec DK |
2249 | return retval; |
2250 | } | |
2251 | ||
af5bc92d TT |
2252 | static int ext4_symlink(struct inode *dir, |
2253 | struct dentry *dentry, const char *symname) | |
ac27a0ec DK |
2254 | { |
2255 | handle_t *handle; | |
af5bc92d | 2256 | struct inode *inode; |
ac27a0ec DK |
2257 | int l, err, retries = 0; |
2258 | ||
2259 | l = strlen(symname)+1; | |
2260 | if (l > dir->i_sb->s_blocksize) | |
2261 | return -ENAMETOOLONG; | |
2262 | ||
871a2931 | 2263 | dquot_initialize(dir); |
907f4554 | 2264 | |
ac27a0ec | 2265 | retry: |
617ba13b MC |
2266 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2267 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 + | |
5aca07eb | 2268 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); |
ac27a0ec DK |
2269 | if (IS_ERR(handle)) |
2270 | return PTR_ERR(handle); | |
2271 | ||
2272 | if (IS_DIRSYNC(dir)) | |
0390131b | 2273 | ext4_handle_sync(handle); |
ac27a0ec | 2274 | |
11013911 AD |
2275 | inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO, |
2276 | &dentry->d_name, 0); | |
ac27a0ec DK |
2277 | err = PTR_ERR(inode); |
2278 | if (IS_ERR(inode)) | |
2279 | goto out_stop; | |
2280 | ||
af5bc92d | 2281 | if (l > sizeof(EXT4_I(inode)->i_data)) { |
617ba13b MC |
2282 | inode->i_op = &ext4_symlink_inode_operations; |
2283 | ext4_set_aops(inode); | |
ac27a0ec | 2284 | /* |
617ba13b | 2285 | * page_symlink() calls into ext4_prepare/commit_write. |
ac27a0ec DK |
2286 | * We have a transaction open. All is sweetness. It also sets |
2287 | * i_size in generic_commit_write(). | |
2288 | */ | |
54566b2c | 2289 | err = __page_symlink(inode, symname, l, 1); |
ac27a0ec | 2290 | if (err) { |
825f1481 | 2291 | clear_nlink(inode); |
6b38e842 | 2292 | unlock_new_inode(inode); |
617ba13b | 2293 | ext4_mark_inode_dirty(handle, inode); |
af5bc92d | 2294 | iput(inode); |
ac27a0ec DK |
2295 | goto out_stop; |
2296 | } | |
2297 | } else { | |
e65187e6 | 2298 | /* clear the extent format for fast symlink */ |
12e9b892 | 2299 | ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); |
617ba13b | 2300 | inode->i_op = &ext4_fast_symlink_inode_operations; |
af5bc92d | 2301 | memcpy((char *)&EXT4_I(inode)->i_data, symname, l); |
ac27a0ec DK |
2302 | inode->i_size = l-1; |
2303 | } | |
617ba13b MC |
2304 | EXT4_I(inode)->i_disksize = inode->i_size; |
2305 | err = ext4_add_nondir(handle, dentry, inode); | |
ac27a0ec | 2306 | out_stop: |
617ba13b MC |
2307 | ext4_journal_stop(handle); |
2308 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
2309 | goto retry; |
2310 | return err; | |
2311 | } | |
2312 | ||
af5bc92d TT |
2313 | static int ext4_link(struct dentry *old_dentry, |
2314 | struct inode *dir, struct dentry *dentry) | |
ac27a0ec DK |
2315 | { |
2316 | handle_t *handle; | |
2317 | struct inode *inode = old_dentry->d_inode; | |
2318 | int err, retries = 0; | |
2319 | ||
b05ab1dc | 2320 | if (inode->i_nlink >= EXT4_LINK_MAX) |
ac27a0ec | 2321 | return -EMLINK; |
f8628a14 | 2322 | |
871a2931 | 2323 | dquot_initialize(dir); |
907f4554 | 2324 | |
2988a774 ES |
2325 | /* |
2326 | * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing | |
2327 | * otherwise has the potential to corrupt the orphan inode list. | |
2328 | */ | |
2329 | if (inode->i_nlink == 0) | |
2330 | return -ENOENT; | |
ac27a0ec DK |
2331 | |
2332 | retry: | |
617ba13b MC |
2333 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2334 | EXT4_INDEX_EXTRA_TRANS_BLOCKS); | |
ac27a0ec DK |
2335 | if (IS_ERR(handle)) |
2336 | return PTR_ERR(handle); | |
2337 | ||
2338 | if (IS_DIRSYNC(dir)) | |
0390131b | 2339 | ext4_handle_sync(handle); |
ac27a0ec | 2340 | |
ef7f3835 | 2341 | inode->i_ctime = ext4_current_time(inode); |
f8628a14 | 2342 | ext4_inc_count(handle, inode); |
ac27a0ec DK |
2343 | atomic_inc(&inode->i_count); |
2344 | ||
6b38e842 AV |
2345 | err = ext4_add_entry(handle, dentry, inode); |
2346 | if (!err) { | |
2347 | ext4_mark_inode_dirty(handle, inode); | |
2348 | d_instantiate(dentry, inode); | |
2349 | } else { | |
2350 | drop_nlink(inode); | |
2351 | iput(inode); | |
2352 | } | |
617ba13b MC |
2353 | ext4_journal_stop(handle); |
2354 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
2355 | goto retry; |
2356 | return err; | |
2357 | } | |
2358 | ||
3d0518f4 WY |
2359 | #define PARENT_INO(buffer, size) \ |
2360 | (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode) | |
ac27a0ec DK |
2361 | |
2362 | /* | |
2363 | * Anybody can rename anything with this: the permission checks are left to the | |
2364 | * higher-level routines. | |
2365 | */ | |
af5bc92d TT |
2366 | static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry, |
2367 | struct inode *new_dir, struct dentry *new_dentry) | |
ac27a0ec DK |
2368 | { |
2369 | handle_t *handle; | |
af5bc92d TT |
2370 | struct inode *old_inode, *new_inode; |
2371 | struct buffer_head *old_bh, *new_bh, *dir_bh; | |
2372 | struct ext4_dir_entry_2 *old_de, *new_de; | |
8750c6d5 | 2373 | int retval, force_da_alloc = 0; |
ac27a0ec | 2374 | |
871a2931 CH |
2375 | dquot_initialize(old_dir); |
2376 | dquot_initialize(new_dir); | |
907f4554 | 2377 | |
ac27a0ec DK |
2378 | old_bh = new_bh = dir_bh = NULL; |
2379 | ||
2380 | /* Initialize quotas before so that eventual writes go | |
2381 | * in separate transaction */ | |
2382 | if (new_dentry->d_inode) | |
871a2931 | 2383 | dquot_initialize(new_dentry->d_inode); |
617ba13b MC |
2384 | handle = ext4_journal_start(old_dir, 2 * |
2385 | EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) + | |
2386 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2); | |
ac27a0ec DK |
2387 | if (IS_ERR(handle)) |
2388 | return PTR_ERR(handle); | |
2389 | ||
2390 | if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir)) | |
0390131b | 2391 | ext4_handle_sync(handle); |
ac27a0ec | 2392 | |
f702ba0f | 2393 | old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de); |
ac27a0ec DK |
2394 | /* |
2395 | * Check for inode number is _not_ due to possible IO errors. | |
2396 | * We might rmdir the source, keep it as pwd of some process | |
2397 | * and merrily kill the link to whatever was created under the | |
2398 | * same name. Goodbye sticky bit ;-< | |
2399 | */ | |
2400 | old_inode = old_dentry->d_inode; | |
2401 | retval = -ENOENT; | |
2402 | if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino) | |
2403 | goto end_rename; | |
2404 | ||
2405 | new_inode = new_dentry->d_inode; | |
f702ba0f | 2406 | new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de); |
ac27a0ec DK |
2407 | if (new_bh) { |
2408 | if (!new_inode) { | |
af5bc92d | 2409 | brelse(new_bh); |
ac27a0ec DK |
2410 | new_bh = NULL; |
2411 | } | |
2412 | } | |
2413 | if (S_ISDIR(old_inode->i_mode)) { | |
2414 | if (new_inode) { | |
2415 | retval = -ENOTEMPTY; | |
af5bc92d | 2416 | if (!empty_dir(new_inode)) |
ac27a0ec DK |
2417 | goto end_rename; |
2418 | } | |
2419 | retval = -EIO; | |
af5bc92d | 2420 | dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval); |
ac27a0ec DK |
2421 | if (!dir_bh) |
2422 | goto end_rename; | |
3d0518f4 WY |
2423 | if (le32_to_cpu(PARENT_INO(dir_bh->b_data, |
2424 | old_dir->i_sb->s_blocksize)) != old_dir->i_ino) | |
ac27a0ec DK |
2425 | goto end_rename; |
2426 | retval = -EMLINK; | |
af5bc92d | 2427 | if (!new_inode && new_dir != old_dir && |
2c94eb86 | 2428 | EXT4_DIR_LINK_MAX(new_dir)) |
ac27a0ec DK |
2429 | goto end_rename; |
2430 | } | |
2431 | if (!new_bh) { | |
af5bc92d | 2432 | retval = ext4_add_entry(handle, new_dentry, old_inode); |
ac27a0ec DK |
2433 | if (retval) |
2434 | goto end_rename; | |
2435 | } else { | |
2436 | BUFFER_TRACE(new_bh, "get write access"); | |
617ba13b | 2437 | ext4_journal_get_write_access(handle, new_bh); |
ac27a0ec | 2438 | new_de->inode = cpu_to_le32(old_inode->i_ino); |
617ba13b MC |
2439 | if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb, |
2440 | EXT4_FEATURE_INCOMPAT_FILETYPE)) | |
ac27a0ec DK |
2441 | new_de->file_type = old_de->file_type; |
2442 | new_dir->i_version++; | |
53b7e9f6 JK |
2443 | new_dir->i_ctime = new_dir->i_mtime = |
2444 | ext4_current_time(new_dir); | |
2445 | ext4_mark_inode_dirty(handle, new_dir); | |
0390131b FM |
2446 | BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata"); |
2447 | ext4_handle_dirty_metadata(handle, new_dir, new_bh); | |
ac27a0ec DK |
2448 | brelse(new_bh); |
2449 | new_bh = NULL; | |
2450 | } | |
2451 | ||
2452 | /* | |
2453 | * Like most other Unix systems, set the ctime for inodes on a | |
2454 | * rename. | |
2455 | */ | |
ef7f3835 | 2456 | old_inode->i_ctime = ext4_current_time(old_inode); |
617ba13b | 2457 | ext4_mark_inode_dirty(handle, old_inode); |
ac27a0ec DK |
2458 | |
2459 | /* | |
2460 | * ok, that's it | |
2461 | */ | |
2462 | if (le32_to_cpu(old_de->inode) != old_inode->i_ino || | |
2463 | old_de->name_len != old_dentry->d_name.len || | |
2464 | strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) || | |
617ba13b | 2465 | (retval = ext4_delete_entry(handle, old_dir, |
ac27a0ec DK |
2466 | old_de, old_bh)) == -ENOENT) { |
2467 | /* old_de could have moved from under us during htree split, so | |
2468 | * make sure that we are deleting the right entry. We might | |
2469 | * also be pointing to a stale entry in the unused part of | |
2470 | * old_bh so just checking inum and the name isn't enough. */ | |
2471 | struct buffer_head *old_bh2; | |
617ba13b | 2472 | struct ext4_dir_entry_2 *old_de2; |
ac27a0ec | 2473 | |
f702ba0f | 2474 | old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2); |
ac27a0ec | 2475 | if (old_bh2) { |
617ba13b | 2476 | retval = ext4_delete_entry(handle, old_dir, |
ac27a0ec DK |
2477 | old_de2, old_bh2); |
2478 | brelse(old_bh2); | |
2479 | } | |
2480 | } | |
2481 | if (retval) { | |
12062ddd | 2482 | ext4_warning(old_dir->i_sb, |
ac27a0ec DK |
2483 | "Deleting old file (%lu), %d, error=%d", |
2484 | old_dir->i_ino, old_dir->i_nlink, retval); | |
2485 | } | |
2486 | ||
2487 | if (new_inode) { | |
f8628a14 | 2488 | ext4_dec_count(handle, new_inode); |
ef7f3835 | 2489 | new_inode->i_ctime = ext4_current_time(new_inode); |
ac27a0ec | 2490 | } |
ef7f3835 | 2491 | old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir); |
617ba13b | 2492 | ext4_update_dx_flag(old_dir); |
ac27a0ec DK |
2493 | if (dir_bh) { |
2494 | BUFFER_TRACE(dir_bh, "get_write_access"); | |
617ba13b | 2495 | ext4_journal_get_write_access(handle, dir_bh); |
3d0518f4 WY |
2496 | PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) = |
2497 | cpu_to_le32(new_dir->i_ino); | |
0390131b FM |
2498 | BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata"); |
2499 | ext4_handle_dirty_metadata(handle, old_dir, dir_bh); | |
f8628a14 | 2500 | ext4_dec_count(handle, old_dir); |
ac27a0ec | 2501 | if (new_inode) { |
f8628a14 | 2502 | /* checked empty_dir above, can't have another parent, |
825f1481 | 2503 | * ext4_dec_count() won't work for many-linked dirs */ |
f8628a14 | 2504 | new_inode->i_nlink = 0; |
ac27a0ec | 2505 | } else { |
f8628a14 | 2506 | ext4_inc_count(handle, new_dir); |
617ba13b MC |
2507 | ext4_update_dx_flag(new_dir); |
2508 | ext4_mark_inode_dirty(handle, new_dir); | |
ac27a0ec DK |
2509 | } |
2510 | } | |
617ba13b | 2511 | ext4_mark_inode_dirty(handle, old_dir); |
ac27a0ec | 2512 | if (new_inode) { |
617ba13b | 2513 | ext4_mark_inode_dirty(handle, new_inode); |
ac27a0ec | 2514 | if (!new_inode->i_nlink) |
617ba13b | 2515 | ext4_orphan_add(handle, new_inode); |
afd4672d TT |
2516 | if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC)) |
2517 | force_da_alloc = 1; | |
ac27a0ec DK |
2518 | } |
2519 | retval = 0; | |
2520 | ||
2521 | end_rename: | |
af5bc92d TT |
2522 | brelse(dir_bh); |
2523 | brelse(old_bh); | |
2524 | brelse(new_bh); | |
617ba13b | 2525 | ext4_journal_stop(handle); |
8750c6d5 TT |
2526 | if (retval == 0 && force_da_alloc) |
2527 | ext4_alloc_da_blocks(old_inode); | |
ac27a0ec DK |
2528 | return retval; |
2529 | } | |
2530 | ||
2531 | /* | |
2532 | * directories can handle most operations... | |
2533 | */ | |
754661f1 | 2534 | const struct inode_operations ext4_dir_inode_operations = { |
617ba13b MC |
2535 | .create = ext4_create, |
2536 | .lookup = ext4_lookup, | |
2537 | .link = ext4_link, | |
2538 | .unlink = ext4_unlink, | |
2539 | .symlink = ext4_symlink, | |
2540 | .mkdir = ext4_mkdir, | |
2541 | .rmdir = ext4_rmdir, | |
2542 | .mknod = ext4_mknod, | |
2543 | .rename = ext4_rename, | |
2544 | .setattr = ext4_setattr, | |
03010a33 | 2545 | #ifdef CONFIG_EXT4_FS_XATTR |
ac27a0ec DK |
2546 | .setxattr = generic_setxattr, |
2547 | .getxattr = generic_getxattr, | |
617ba13b | 2548 | .listxattr = ext4_listxattr, |
ac27a0ec DK |
2549 | .removexattr = generic_removexattr, |
2550 | #endif | |
1d5ccd1c | 2551 | .check_acl = ext4_check_acl, |
abc8746e | 2552 | .fiemap = ext4_fiemap, |
ac27a0ec DK |
2553 | }; |
2554 | ||
754661f1 | 2555 | const struct inode_operations ext4_special_inode_operations = { |
617ba13b | 2556 | .setattr = ext4_setattr, |
03010a33 | 2557 | #ifdef CONFIG_EXT4_FS_XATTR |
ac27a0ec DK |
2558 | .setxattr = generic_setxattr, |
2559 | .getxattr = generic_getxattr, | |
617ba13b | 2560 | .listxattr = ext4_listxattr, |
ac27a0ec DK |
2561 | .removexattr = generic_removexattr, |
2562 | #endif | |
1d5ccd1c | 2563 | .check_acl = ext4_check_acl, |
ac27a0ec | 2564 | }; |