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 DK |
39 | |
40 | #include "namei.h" | |
41 | #include "xattr.h" | |
42 | #include "acl.h" | |
43 | ||
44 | /* | |
45 | * define how far ahead to read directories while searching them. | |
46 | */ | |
47 | #define NAMEI_RA_CHUNKS 2 | |
48 | #define NAMEI_RA_BLOCKS 4 | |
8c55e204 | 49 | #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS) |
ac27a0ec DK |
50 | #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b)) |
51 | ||
617ba13b | 52 | static struct buffer_head *ext4_append(handle_t *handle, |
ac27a0ec | 53 | struct inode *inode, |
725d26d3 | 54 | ext4_lblk_t *block, int *err) |
ac27a0ec DK |
55 | { |
56 | struct buffer_head *bh; | |
57 | ||
58 | *block = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
59 | ||
a871611b AM |
60 | bh = ext4_bread(handle, inode, *block, 1, err); |
61 | if (bh) { | |
ac27a0ec | 62 | inode->i_size += inode->i_sb->s_blocksize; |
617ba13b | 63 | EXT4_I(inode)->i_disksize = inode->i_size; |
a871611b AM |
64 | *err = ext4_journal_get_write_access(handle, bh); |
65 | if (*err) { | |
66 | brelse(bh); | |
67 | bh = NULL; | |
68 | } | |
ac27a0ec DK |
69 | } |
70 | return bh; | |
71 | } | |
72 | ||
73 | #ifndef assert | |
74 | #define assert(test) J_ASSERT(test) | |
75 | #endif | |
76 | ||
77 | #ifndef swap | |
78 | #define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0) | |
79 | #endif | |
80 | ||
81 | #ifdef DX_DEBUG | |
82 | #define dxtrace(command) command | |
83 | #else | |
84 | #define dxtrace(command) | |
85 | #endif | |
86 | ||
87 | struct fake_dirent | |
88 | { | |
89 | __le32 inode; | |
90 | __le16 rec_len; | |
91 | u8 name_len; | |
92 | u8 file_type; | |
93 | }; | |
94 | ||
95 | struct dx_countlimit | |
96 | { | |
97 | __le16 limit; | |
98 | __le16 count; | |
99 | }; | |
100 | ||
101 | struct dx_entry | |
102 | { | |
103 | __le32 hash; | |
104 | __le32 block; | |
105 | }; | |
106 | ||
107 | /* | |
108 | * dx_root_info is laid out so that if it should somehow get overlaid by a | |
109 | * dirent the two low bits of the hash version will be zero. Therefore, the | |
110 | * hash version mod 4 should never be 0. Sincerely, the paranoia department. | |
111 | */ | |
112 | ||
113 | struct dx_root | |
114 | { | |
115 | struct fake_dirent dot; | |
116 | char dot_name[4]; | |
117 | struct fake_dirent dotdot; | |
118 | char dotdot_name[4]; | |
119 | struct dx_root_info | |
120 | { | |
121 | __le32 reserved_zero; | |
122 | u8 hash_version; | |
123 | u8 info_length; /* 8 */ | |
124 | u8 indirect_levels; | |
125 | u8 unused_flags; | |
126 | } | |
127 | info; | |
128 | struct dx_entry entries[0]; | |
129 | }; | |
130 | ||
131 | struct dx_node | |
132 | { | |
133 | struct fake_dirent fake; | |
134 | struct dx_entry entries[0]; | |
135 | }; | |
136 | ||
137 | ||
138 | struct dx_frame | |
139 | { | |
140 | struct buffer_head *bh; | |
141 | struct dx_entry *entries; | |
142 | struct dx_entry *at; | |
143 | }; | |
144 | ||
145 | struct dx_map_entry | |
146 | { | |
147 | u32 hash; | |
ef2b02d3 ES |
148 | u16 offs; |
149 | u16 size; | |
ac27a0ec DK |
150 | }; |
151 | ||
725d26d3 AK |
152 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry); |
153 | static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value); | |
af5bc92d TT |
154 | static inline unsigned dx_get_hash(struct dx_entry *entry); |
155 | static void dx_set_hash(struct dx_entry *entry, unsigned value); | |
156 | static unsigned dx_get_count(struct dx_entry *entries); | |
157 | static unsigned dx_get_limit(struct dx_entry *entries); | |
158 | static void dx_set_count(struct dx_entry *entries, unsigned value); | |
159 | static void dx_set_limit(struct dx_entry *entries, unsigned value); | |
160 | static unsigned dx_root_limit(struct inode *dir, unsigned infosize); | |
161 | static unsigned dx_node_limit(struct inode *dir); | |
f702ba0f | 162 | static struct dx_frame *dx_probe(const struct qstr *d_name, |
ac27a0ec DK |
163 | struct inode *dir, |
164 | struct dx_hash_info *hinfo, | |
165 | struct dx_frame *frame, | |
166 | int *err); | |
af5bc92d TT |
167 | static void dx_release(struct dx_frame *frames); |
168 | static int dx_make_map(struct ext4_dir_entry_2 *de, int size, | |
169 | struct dx_hash_info *hinfo, struct dx_map_entry map[]); | |
ac27a0ec | 170 | static void dx_sort_map(struct dx_map_entry *map, unsigned count); |
af5bc92d | 171 | static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to, |
ac27a0ec | 172 | struct dx_map_entry *offsets, int count); |
af5bc92d | 173 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size); |
725d26d3 AK |
174 | static void dx_insert_block(struct dx_frame *frame, |
175 | u32 hash, ext4_lblk_t block); | |
617ba13b | 176 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, |
ac27a0ec DK |
177 | struct dx_frame *frame, |
178 | struct dx_frame *frames, | |
179 | __u32 *start_hash); | |
f702ba0f TT |
180 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir, |
181 | const struct qstr *d_name, | |
182 | struct ext4_dir_entry_2 **res_dir, | |
183 | int *err); | |
617ba13b | 184 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, |
ac27a0ec DK |
185 | struct inode *inode); |
186 | ||
f795e140 LZ |
187 | /* |
188 | * p is at least 6 bytes before the end of page | |
189 | */ | |
190 | static inline struct ext4_dir_entry_2 * | |
191 | ext4_next_entry(struct ext4_dir_entry_2 *p) | |
192 | { | |
193 | return (struct ext4_dir_entry_2 *)((char *)p + | |
194 | ext4_rec_len_from_disk(p->rec_len)); | |
195 | } | |
196 | ||
ac27a0ec DK |
197 | /* |
198 | * Future: use high four bits of block for coalesce-on-delete flags | |
199 | * Mask them off for now. | |
200 | */ | |
201 | ||
725d26d3 | 202 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) |
ac27a0ec DK |
203 | { |
204 | return le32_to_cpu(entry->block) & 0x00ffffff; | |
205 | } | |
206 | ||
725d26d3 | 207 | static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) |
ac27a0ec DK |
208 | { |
209 | entry->block = cpu_to_le32(value); | |
210 | } | |
211 | ||
af5bc92d | 212 | static inline unsigned dx_get_hash(struct dx_entry *entry) |
ac27a0ec DK |
213 | { |
214 | return le32_to_cpu(entry->hash); | |
215 | } | |
216 | ||
af5bc92d | 217 | static inline void dx_set_hash(struct dx_entry *entry, unsigned value) |
ac27a0ec DK |
218 | { |
219 | entry->hash = cpu_to_le32(value); | |
220 | } | |
221 | ||
af5bc92d | 222 | static inline unsigned dx_get_count(struct dx_entry *entries) |
ac27a0ec DK |
223 | { |
224 | return le16_to_cpu(((struct dx_countlimit *) entries)->count); | |
225 | } | |
226 | ||
af5bc92d | 227 | static inline unsigned dx_get_limit(struct dx_entry *entries) |
ac27a0ec DK |
228 | { |
229 | return le16_to_cpu(((struct dx_countlimit *) entries)->limit); | |
230 | } | |
231 | ||
af5bc92d | 232 | static inline void dx_set_count(struct dx_entry *entries, unsigned value) |
ac27a0ec DK |
233 | { |
234 | ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); | |
235 | } | |
236 | ||
af5bc92d | 237 | static inline void dx_set_limit(struct dx_entry *entries, unsigned value) |
ac27a0ec DK |
238 | { |
239 | ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); | |
240 | } | |
241 | ||
af5bc92d | 242 | static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) |
ac27a0ec | 243 | { |
617ba13b MC |
244 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - |
245 | EXT4_DIR_REC_LEN(2) - infosize; | |
d9c769b7 | 246 | return entry_space / sizeof(struct dx_entry); |
ac27a0ec DK |
247 | } |
248 | ||
af5bc92d | 249 | static inline unsigned dx_node_limit(struct inode *dir) |
ac27a0ec | 250 | { |
617ba13b | 251 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); |
d9c769b7 | 252 | return entry_space / sizeof(struct dx_entry); |
ac27a0ec DK |
253 | } |
254 | ||
255 | /* | |
256 | * Debug | |
257 | */ | |
258 | #ifdef DX_DEBUG | |
4776004f | 259 | static void dx_show_index(char * label, struct dx_entry *entries) |
ac27a0ec | 260 | { |
63f57933 | 261 | int i, n = dx_get_count (entries); |
4776004f | 262 | printk(KERN_DEBUG "%s index ", label); |
63f57933 | 263 | for (i = 0; i < n; i++) { |
4776004f | 264 | printk("%x->%lu ", i ? dx_get_hash(entries + i) : |
725d26d3 | 265 | 0, (unsigned long)dx_get_block(entries + i)); |
63f57933 AM |
266 | } |
267 | printk("\n"); | |
ac27a0ec DK |
268 | } |
269 | ||
270 | struct stats | |
271 | { | |
272 | unsigned names; | |
273 | unsigned space; | |
274 | unsigned bcount; | |
275 | }; | |
276 | ||
617ba13b | 277 | static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de, |
ac27a0ec DK |
278 | int size, int show_names) |
279 | { | |
280 | unsigned names = 0, space = 0; | |
281 | char *base = (char *) de; | |
282 | struct dx_hash_info h = *hinfo; | |
283 | ||
284 | printk("names: "); | |
285 | while ((char *) de < base + size) | |
286 | { | |
287 | if (de->inode) | |
288 | { | |
289 | if (show_names) | |
290 | { | |
291 | int len = de->name_len; | |
292 | char *name = de->name; | |
293 | while (len--) printk("%c", *name++); | |
617ba13b | 294 | ext4fs_dirhash(de->name, de->name_len, &h); |
ac27a0ec DK |
295 | printk(":%x.%u ", h.hash, |
296 | ((char *) de - base)); | |
297 | } | |
617ba13b | 298 | space += EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec DK |
299 | names++; |
300 | } | |
a72d7f83 | 301 | de = ext4_next_entry(de); |
ac27a0ec DK |
302 | } |
303 | printk("(%i)\n", names); | |
304 | return (struct stats) { names, space, 1 }; | |
305 | } | |
306 | ||
307 | struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, | |
308 | struct dx_entry *entries, int levels) | |
309 | { | |
310 | unsigned blocksize = dir->i_sb->s_blocksize; | |
af5bc92d | 311 | unsigned count = dx_get_count(entries), names = 0, space = 0, i; |
ac27a0ec DK |
312 | unsigned bcount = 0; |
313 | struct buffer_head *bh; | |
314 | int err; | |
315 | printk("%i indexed blocks...\n", count); | |
316 | for (i = 0; i < count; i++, entries++) | |
317 | { | |
725d26d3 AK |
318 | ext4_lblk_t block = dx_get_block(entries); |
319 | ext4_lblk_t hash = i ? dx_get_hash(entries): 0; | |
ac27a0ec DK |
320 | u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; |
321 | struct stats stats; | |
322 | printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); | |
617ba13b | 323 | if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue; |
ac27a0ec DK |
324 | stats = levels? |
325 | dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): | |
617ba13b | 326 | dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0); |
ac27a0ec DK |
327 | names += stats.names; |
328 | space += stats.space; | |
329 | bcount += stats.bcount; | |
af5bc92d | 330 | brelse(bh); |
ac27a0ec DK |
331 | } |
332 | if (bcount) | |
4776004f TT |
333 | printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", |
334 | levels ? "" : " ", names, space/bcount, | |
335 | (space/bcount)*100/blocksize); | |
ac27a0ec DK |
336 | return (struct stats) { names, space, bcount}; |
337 | } | |
338 | #endif /* DX_DEBUG */ | |
339 | ||
340 | /* | |
341 | * Probe for a directory leaf block to search. | |
342 | * | |
343 | * dx_probe can return ERR_BAD_DX_DIR, which means there was a format | |
344 | * error in the directory index, and the caller should fall back to | |
345 | * searching the directory normally. The callers of dx_probe **MUST** | |
346 | * check for this error code, and make sure it never gets reflected | |
347 | * back to userspace. | |
348 | */ | |
349 | static struct dx_frame * | |
f702ba0f | 350 | dx_probe(const struct qstr *d_name, struct inode *dir, |
ac27a0ec DK |
351 | struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err) |
352 | { | |
353 | unsigned count, indirect; | |
354 | struct dx_entry *at, *entries, *p, *q, *m; | |
355 | struct dx_root *root; | |
356 | struct buffer_head *bh; | |
357 | struct dx_frame *frame = frame_in; | |
358 | u32 hash; | |
359 | ||
360 | frame->bh = NULL; | |
617ba13b | 361 | if (!(bh = ext4_bread (NULL,dir, 0, 0, err))) |
ac27a0ec DK |
362 | goto fail; |
363 | root = (struct dx_root *) bh->b_data; | |
364 | if (root->info.hash_version != DX_HASH_TEA && | |
365 | root->info.hash_version != DX_HASH_HALF_MD4 && | |
366 | root->info.hash_version != DX_HASH_LEGACY) { | |
46e665e9 | 367 | ext4_warning(dir->i_sb, __func__, |
ac27a0ec DK |
368 | "Unrecognised inode hash code %d", |
369 | root->info.hash_version); | |
370 | brelse(bh); | |
371 | *err = ERR_BAD_DX_DIR; | |
372 | goto fail; | |
373 | } | |
374 | hinfo->hash_version = root->info.hash_version; | |
617ba13b | 375 | hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
f702ba0f TT |
376 | if (d_name) |
377 | ext4fs_dirhash(d_name->name, d_name->len, hinfo); | |
ac27a0ec DK |
378 | hash = hinfo->hash; |
379 | ||
380 | if (root->info.unused_flags & 1) { | |
46e665e9 | 381 | ext4_warning(dir->i_sb, __func__, |
ac27a0ec DK |
382 | "Unimplemented inode hash flags: %#06x", |
383 | root->info.unused_flags); | |
384 | brelse(bh); | |
385 | *err = ERR_BAD_DX_DIR; | |
386 | goto fail; | |
387 | } | |
388 | ||
389 | if ((indirect = root->info.indirect_levels) > 1) { | |
46e665e9 | 390 | ext4_warning(dir->i_sb, __func__, |
ac27a0ec DK |
391 | "Unimplemented inode hash depth: %#06x", |
392 | root->info.indirect_levels); | |
393 | brelse(bh); | |
394 | *err = ERR_BAD_DX_DIR; | |
395 | goto fail; | |
396 | } | |
397 | ||
398 | entries = (struct dx_entry *) (((char *)&root->info) + | |
399 | root->info.info_length); | |
3d82abae ES |
400 | |
401 | if (dx_get_limit(entries) != dx_root_limit(dir, | |
402 | root->info.info_length)) { | |
46e665e9 | 403 | ext4_warning(dir->i_sb, __func__, |
3d82abae ES |
404 | "dx entry: limit != root limit"); |
405 | brelse(bh); | |
406 | *err = ERR_BAD_DX_DIR; | |
407 | goto fail; | |
408 | } | |
409 | ||
af5bc92d | 410 | dxtrace(printk("Look up %x", hash)); |
ac27a0ec DK |
411 | while (1) |
412 | { | |
413 | count = dx_get_count(entries); | |
3d82abae | 414 | if (!count || count > dx_get_limit(entries)) { |
46e665e9 | 415 | ext4_warning(dir->i_sb, __func__, |
3d82abae ES |
416 | "dx entry: no count or count > limit"); |
417 | brelse(bh); | |
418 | *err = ERR_BAD_DX_DIR; | |
419 | goto fail2; | |
420 | } | |
421 | ||
ac27a0ec DK |
422 | p = entries + 1; |
423 | q = entries + count - 1; | |
424 | while (p <= q) | |
425 | { | |
426 | m = p + (q - p)/2; | |
427 | dxtrace(printk(".")); | |
428 | if (dx_get_hash(m) > hash) | |
429 | q = m - 1; | |
430 | else | |
431 | p = m + 1; | |
432 | } | |
433 | ||
434 | if (0) // linear search cross check | |
435 | { | |
436 | unsigned n = count - 1; | |
437 | at = entries; | |
438 | while (n--) | |
439 | { | |
440 | dxtrace(printk(",")); | |
441 | if (dx_get_hash(++at) > hash) | |
442 | { | |
443 | at--; | |
444 | break; | |
445 | } | |
446 | } | |
447 | assert (at == p - 1); | |
448 | } | |
449 | ||
450 | at = p - 1; | |
451 | dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at))); | |
452 | frame->bh = bh; | |
453 | frame->entries = entries; | |
454 | frame->at = at; | |
455 | if (!indirect--) return frame; | |
617ba13b | 456 | if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err))) |
ac27a0ec DK |
457 | goto fail2; |
458 | at = entries = ((struct dx_node *) bh->b_data)->entries; | |
3d82abae | 459 | if (dx_get_limit(entries) != dx_node_limit (dir)) { |
46e665e9 | 460 | ext4_warning(dir->i_sb, __func__, |
3d82abae ES |
461 | "dx entry: limit != node limit"); |
462 | brelse(bh); | |
463 | *err = ERR_BAD_DX_DIR; | |
464 | goto fail2; | |
465 | } | |
ac27a0ec | 466 | frame++; |
3d82abae | 467 | frame->bh = NULL; |
ac27a0ec DK |
468 | } |
469 | fail2: | |
470 | while (frame >= frame_in) { | |
471 | brelse(frame->bh); | |
472 | frame--; | |
473 | } | |
474 | fail: | |
3d82abae | 475 | if (*err == ERR_BAD_DX_DIR) |
46e665e9 | 476 | ext4_warning(dir->i_sb, __func__, |
3d82abae ES |
477 | "Corrupt dir inode %ld, running e2fsck is " |
478 | "recommended.", dir->i_ino); | |
ac27a0ec DK |
479 | return NULL; |
480 | } | |
481 | ||
482 | static void dx_release (struct dx_frame *frames) | |
483 | { | |
484 | if (frames[0].bh == NULL) | |
485 | return; | |
486 | ||
487 | if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels) | |
488 | brelse(frames[1].bh); | |
489 | brelse(frames[0].bh); | |
490 | } | |
491 | ||
492 | /* | |
493 | * This function increments the frame pointer to search the next leaf | |
494 | * block, and reads in the necessary intervening nodes if the search | |
495 | * should be necessary. Whether or not the search is necessary is | |
496 | * controlled by the hash parameter. If the hash value is even, then | |
497 | * the search is only continued if the next block starts with that | |
498 | * hash value. This is used if we are searching for a specific file. | |
499 | * | |
500 | * If the hash value is HASH_NB_ALWAYS, then always go to the next block. | |
501 | * | |
502 | * This function returns 1 if the caller should continue to search, | |
503 | * or 0 if it should not. If there is an error reading one of the | |
504 | * index blocks, it will a negative error code. | |
505 | * | |
506 | * If start_hash is non-null, it will be filled in with the starting | |
507 | * hash of the next page. | |
508 | */ | |
617ba13b | 509 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, |
ac27a0ec DK |
510 | struct dx_frame *frame, |
511 | struct dx_frame *frames, | |
512 | __u32 *start_hash) | |
513 | { | |
514 | struct dx_frame *p; | |
515 | struct buffer_head *bh; | |
516 | int err, num_frames = 0; | |
517 | __u32 bhash; | |
518 | ||
519 | p = frame; | |
520 | /* | |
521 | * Find the next leaf page by incrementing the frame pointer. | |
522 | * If we run out of entries in the interior node, loop around and | |
523 | * increment pointer in the parent node. When we break out of | |
524 | * this loop, num_frames indicates the number of interior | |
525 | * nodes need to be read. | |
526 | */ | |
527 | while (1) { | |
528 | if (++(p->at) < p->entries + dx_get_count(p->entries)) | |
529 | break; | |
530 | if (p == frames) | |
531 | return 0; | |
532 | num_frames++; | |
533 | p--; | |
534 | } | |
535 | ||
536 | /* | |
537 | * If the hash is 1, then continue only if the next page has a | |
538 | * continuation hash of any value. This is used for readdir | |
539 | * handling. Otherwise, check to see if the hash matches the | |
540 | * desired contiuation hash. If it doesn't, return since | |
541 | * there's no point to read in the successive index pages. | |
542 | */ | |
543 | bhash = dx_get_hash(p->at); | |
544 | if (start_hash) | |
545 | *start_hash = bhash; | |
546 | if ((hash & 1) == 0) { | |
547 | if ((bhash & ~1) != hash) | |
548 | return 0; | |
549 | } | |
550 | /* | |
551 | * If the hash is HASH_NB_ALWAYS, we always go to the next | |
552 | * block so no check is necessary | |
553 | */ | |
554 | while (num_frames--) { | |
617ba13b | 555 | if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at), |
ac27a0ec DK |
556 | 0, &err))) |
557 | return err; /* Failure */ | |
558 | p++; | |
af5bc92d | 559 | brelse(p->bh); |
ac27a0ec DK |
560 | p->bh = bh; |
561 | p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; | |
562 | } | |
563 | return 1; | |
564 | } | |
565 | ||
566 | ||
ac27a0ec DK |
567 | /* |
568 | * This function fills a red-black tree with information from a | |
569 | * directory block. It returns the number directory entries loaded | |
570 | * into the tree. If there is an error it is returned in err. | |
571 | */ | |
572 | static int htree_dirblock_to_tree(struct file *dir_file, | |
725d26d3 | 573 | struct inode *dir, ext4_lblk_t block, |
ac27a0ec DK |
574 | struct dx_hash_info *hinfo, |
575 | __u32 start_hash, __u32 start_minor_hash) | |
576 | { | |
577 | struct buffer_head *bh; | |
617ba13b | 578 | struct ext4_dir_entry_2 *de, *top; |
ac27a0ec DK |
579 | int err, count = 0; |
580 | ||
725d26d3 AK |
581 | dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", |
582 | (unsigned long)block)); | |
617ba13b | 583 | if (!(bh = ext4_bread (NULL, dir, block, 0, &err))) |
ac27a0ec DK |
584 | return err; |
585 | ||
617ba13b MC |
586 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
587 | top = (struct ext4_dir_entry_2 *) ((char *) de + | |
ac27a0ec | 588 | dir->i_sb->s_blocksize - |
617ba13b MC |
589 | EXT4_DIR_REC_LEN(0)); |
590 | for (; de < top; de = ext4_next_entry(de)) { | |
e6c40211 ES |
591 | if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh, |
592 | (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) | |
593 | +((char *)de - bh->b_data))) { | |
594 | /* On error, skip the f_pos to the next block. */ | |
595 | dir_file->f_pos = (dir_file->f_pos | | |
596 | (dir->i_sb->s_blocksize - 1)) + 1; | |
af5bc92d | 597 | brelse(bh); |
e6c40211 ES |
598 | return count; |
599 | } | |
617ba13b | 600 | ext4fs_dirhash(de->name, de->name_len, hinfo); |
ac27a0ec DK |
601 | if ((hinfo->hash < start_hash) || |
602 | ((hinfo->hash == start_hash) && | |
603 | (hinfo->minor_hash < start_minor_hash))) | |
604 | continue; | |
605 | if (de->inode == 0) | |
606 | continue; | |
617ba13b | 607 | if ((err = ext4_htree_store_dirent(dir_file, |
ac27a0ec DK |
608 | hinfo->hash, hinfo->minor_hash, de)) != 0) { |
609 | brelse(bh); | |
610 | return err; | |
611 | } | |
612 | count++; | |
613 | } | |
614 | brelse(bh); | |
615 | return count; | |
616 | } | |
617 | ||
618 | ||
619 | /* | |
620 | * This function fills a red-black tree with information from a | |
621 | * directory. We start scanning the directory in hash order, starting | |
622 | * at start_hash and start_minor_hash. | |
623 | * | |
624 | * This function returns the number of entries inserted into the tree, | |
625 | * or a negative error code. | |
626 | */ | |
617ba13b | 627 | int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, |
ac27a0ec DK |
628 | __u32 start_minor_hash, __u32 *next_hash) |
629 | { | |
630 | struct dx_hash_info hinfo; | |
617ba13b | 631 | struct ext4_dir_entry_2 *de; |
ac27a0ec DK |
632 | struct dx_frame frames[2], *frame; |
633 | struct inode *dir; | |
725d26d3 | 634 | ext4_lblk_t block; |
ac27a0ec | 635 | int count = 0; |
725d26d3 | 636 | int ret, err; |
ac27a0ec DK |
637 | __u32 hashval; |
638 | ||
4776004f TT |
639 | dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", |
640 | start_hash, start_minor_hash)); | |
9d549890 | 641 | dir = dir_file->f_path.dentry->d_inode; |
617ba13b MC |
642 | if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) { |
643 | hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; | |
644 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; | |
ac27a0ec DK |
645 | count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, |
646 | start_hash, start_minor_hash); | |
647 | *next_hash = ~0; | |
648 | return count; | |
649 | } | |
650 | hinfo.hash = start_hash; | |
651 | hinfo.minor_hash = 0; | |
f702ba0f | 652 | frame = dx_probe(NULL, dir, &hinfo, frames, &err); |
ac27a0ec DK |
653 | if (!frame) |
654 | return err; | |
655 | ||
656 | /* Add '.' and '..' from the htree header */ | |
657 | if (!start_hash && !start_minor_hash) { | |
617ba13b MC |
658 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; |
659 | if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0) | |
ac27a0ec DK |
660 | goto errout; |
661 | count++; | |
662 | } | |
663 | if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { | |
617ba13b MC |
664 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; |
665 | de = ext4_next_entry(de); | |
666 | if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0) | |
ac27a0ec DK |
667 | goto errout; |
668 | count++; | |
669 | } | |
670 | ||
671 | while (1) { | |
672 | block = dx_get_block(frame->at); | |
673 | ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, | |
674 | start_hash, start_minor_hash); | |
675 | if (ret < 0) { | |
676 | err = ret; | |
677 | goto errout; | |
678 | } | |
679 | count += ret; | |
680 | hashval = ~0; | |
617ba13b | 681 | ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, |
ac27a0ec DK |
682 | frame, frames, &hashval); |
683 | *next_hash = hashval; | |
684 | if (ret < 0) { | |
685 | err = ret; | |
686 | goto errout; | |
687 | } | |
688 | /* | |
689 | * Stop if: (a) there are no more entries, or | |
690 | * (b) we have inserted at least one entry and the | |
691 | * next hash value is not a continuation | |
692 | */ | |
693 | if ((ret == 0) || | |
694 | (count && ((hashval & 1) == 0))) | |
695 | break; | |
696 | } | |
697 | dx_release(frames); | |
4776004f TT |
698 | dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " |
699 | "next hash: %x\n", count, *next_hash)); | |
ac27a0ec DK |
700 | return count; |
701 | errout: | |
702 | dx_release(frames); | |
703 | return (err); | |
704 | } | |
705 | ||
706 | ||
707 | /* | |
708 | * Directory block splitting, compacting | |
709 | */ | |
710 | ||
ef2b02d3 ES |
711 | /* |
712 | * Create map of hash values, offsets, and sizes, stored at end of block. | |
713 | * Returns number of entries mapped. | |
714 | */ | |
617ba13b | 715 | static int dx_make_map (struct ext4_dir_entry_2 *de, int size, |
ac27a0ec DK |
716 | struct dx_hash_info *hinfo, struct dx_map_entry *map_tail) |
717 | { | |
718 | int count = 0; | |
719 | char *base = (char *) de; | |
720 | struct dx_hash_info h = *hinfo; | |
721 | ||
722 | while ((char *) de < base + size) | |
723 | { | |
724 | if (de->name_len && de->inode) { | |
617ba13b | 725 | ext4fs_dirhash(de->name, de->name_len, &h); |
ac27a0ec DK |
726 | map_tail--; |
727 | map_tail->hash = h.hash; | |
ef2b02d3 ES |
728 | map_tail->offs = (u16) ((char *) de - base); |
729 | map_tail->size = le16_to_cpu(de->rec_len); | |
ac27a0ec DK |
730 | count++; |
731 | cond_resched(); | |
732 | } | |
733 | /* XXX: do we need to check rec_len == 0 case? -Chris */ | |
a72d7f83 | 734 | de = ext4_next_entry(de); |
ac27a0ec DK |
735 | } |
736 | return count; | |
737 | } | |
738 | ||
ef2b02d3 | 739 | /* Sort map by hash value */ |
ac27a0ec DK |
740 | static void dx_sort_map (struct dx_map_entry *map, unsigned count) |
741 | { | |
63f57933 AM |
742 | struct dx_map_entry *p, *q, *top = map + count - 1; |
743 | int more; | |
744 | /* Combsort until bubble sort doesn't suck */ | |
745 | while (count > 2) { | |
746 | count = count*10/13; | |
747 | if (count - 9 < 2) /* 9, 10 -> 11 */ | |
748 | count = 11; | |
749 | for (p = top, q = p - count; q >= map; p--, q--) | |
750 | if (p->hash < q->hash) | |
751 | swap(*p, *q); | |
752 | } | |
753 | /* Garden variety bubble sort */ | |
754 | do { | |
755 | more = 0; | |
756 | q = top; | |
757 | while (q-- > map) { | |
758 | if (q[1].hash >= q[0].hash) | |
ac27a0ec | 759 | continue; |
63f57933 AM |
760 | swap(*(q+1), *q); |
761 | more = 1; | |
ac27a0ec DK |
762 | } |
763 | } while(more); | |
764 | } | |
765 | ||
725d26d3 | 766 | static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) |
ac27a0ec DK |
767 | { |
768 | struct dx_entry *entries = frame->entries; | |
769 | struct dx_entry *old = frame->at, *new = old + 1; | |
770 | int count = dx_get_count(entries); | |
771 | ||
772 | assert(count < dx_get_limit(entries)); | |
773 | assert(old < entries + count); | |
774 | memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); | |
775 | dx_set_hash(new, hash); | |
776 | dx_set_block(new, block); | |
777 | dx_set_count(entries, count + 1); | |
778 | } | |
ac27a0ec | 779 | |
617ba13b | 780 | static void ext4_update_dx_flag(struct inode *inode) |
ac27a0ec | 781 | { |
617ba13b MC |
782 | if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb, |
783 | EXT4_FEATURE_COMPAT_DIR_INDEX)) | |
784 | EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL; | |
ac27a0ec DK |
785 | } |
786 | ||
787 | /* | |
617ba13b | 788 | * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure. |
ac27a0ec | 789 | * |
617ba13b | 790 | * `len <= EXT4_NAME_LEN' is guaranteed by caller. |
ac27a0ec DK |
791 | * `de != NULL' is guaranteed by caller. |
792 | */ | |
617ba13b MC |
793 | static inline int ext4_match (int len, const char * const name, |
794 | struct ext4_dir_entry_2 * de) | |
ac27a0ec DK |
795 | { |
796 | if (len != de->name_len) | |
797 | return 0; | |
798 | if (!de->inode) | |
799 | return 0; | |
800 | return !memcmp(name, de->name, len); | |
801 | } | |
802 | ||
803 | /* | |
804 | * Returns 0 if not found, -1 on failure, and 1 on success | |
805 | */ | |
af5bc92d | 806 | static inline int search_dirblock(struct buffer_head *bh, |
ac27a0ec | 807 | struct inode *dir, |
f702ba0f | 808 | const struct qstr *d_name, |
ac27a0ec | 809 | unsigned long offset, |
617ba13b | 810 | struct ext4_dir_entry_2 ** res_dir) |
ac27a0ec | 811 | { |
617ba13b | 812 | struct ext4_dir_entry_2 * de; |
ac27a0ec DK |
813 | char * dlimit; |
814 | int de_len; | |
f702ba0f TT |
815 | const char *name = d_name->name; |
816 | int namelen = d_name->len; | |
ac27a0ec | 817 | |
617ba13b | 818 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec DK |
819 | dlimit = bh->b_data + dir->i_sb->s_blocksize; |
820 | while ((char *) de < dlimit) { | |
821 | /* this code is executed quadratically often */ | |
822 | /* do minimal checking `by hand' */ | |
823 | ||
824 | if ((char *) de + namelen <= dlimit && | |
617ba13b | 825 | ext4_match (namelen, name, de)) { |
ac27a0ec | 826 | /* found a match - just to be sure, do a full check */ |
617ba13b | 827 | if (!ext4_check_dir_entry("ext4_find_entry", |
ac27a0ec DK |
828 | dir, de, bh, offset)) |
829 | return -1; | |
830 | *res_dir = de; | |
831 | return 1; | |
832 | } | |
833 | /* prevent looping on a bad block */ | |
a72d7f83 | 834 | de_len = ext4_rec_len_from_disk(de->rec_len); |
ac27a0ec DK |
835 | if (de_len <= 0) |
836 | return -1; | |
837 | offset += de_len; | |
617ba13b | 838 | de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); |
ac27a0ec DK |
839 | } |
840 | return 0; | |
841 | } | |
842 | ||
843 | ||
844 | /* | |
617ba13b | 845 | * ext4_find_entry() |
ac27a0ec DK |
846 | * |
847 | * finds an entry in the specified directory with the wanted name. It | |
848 | * returns the cache buffer in which the entry was found, and the entry | |
849 | * itself (as a parameter - res_dir). It does NOT read the inode of the | |
850 | * entry - you'll have to do that yourself if you want to. | |
851 | * | |
852 | * The returned buffer_head has ->b_count elevated. The caller is expected | |
853 | * to brelse() it when appropriate. | |
854 | */ | |
f702ba0f TT |
855 | static struct buffer_head * ext4_find_entry (struct inode *dir, |
856 | const struct qstr *d_name, | |
617ba13b | 857 | struct ext4_dir_entry_2 ** res_dir) |
ac27a0ec | 858 | { |
af5bc92d TT |
859 | struct super_block *sb; |
860 | struct buffer_head *bh_use[NAMEI_RA_SIZE]; | |
861 | struct buffer_head *bh, *ret = NULL; | |
725d26d3 | 862 | ext4_lblk_t start, block, b; |
ac27a0ec DK |
863 | int ra_max = 0; /* Number of bh's in the readahead |
864 | buffer, bh_use[] */ | |
865 | int ra_ptr = 0; /* Current index into readahead | |
866 | buffer */ | |
867 | int num = 0; | |
725d26d3 AK |
868 | ext4_lblk_t nblocks; |
869 | int i, err; | |
ac27a0ec | 870 | int namelen; |
ac27a0ec DK |
871 | |
872 | *res_dir = NULL; | |
873 | sb = dir->i_sb; | |
f702ba0f | 874 | namelen = d_name->len; |
617ba13b | 875 | if (namelen > EXT4_NAME_LEN) |
ac27a0ec | 876 | return NULL; |
ac27a0ec | 877 | if (is_dx(dir)) { |
f702ba0f | 878 | bh = ext4_dx_find_entry(dir, d_name, res_dir, &err); |
ac27a0ec DK |
879 | /* |
880 | * On success, or if the error was file not found, | |
881 | * return. Otherwise, fall back to doing a search the | |
882 | * old fashioned way. | |
883 | */ | |
884 | if (bh || (err != ERR_BAD_DX_DIR)) | |
885 | return bh; | |
4776004f TT |
886 | dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " |
887 | "falling back\n")); | |
ac27a0ec | 888 | } |
617ba13b MC |
889 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); |
890 | start = EXT4_I(dir)->i_dir_start_lookup; | |
ac27a0ec DK |
891 | if (start >= nblocks) |
892 | start = 0; | |
893 | block = start; | |
894 | restart: | |
895 | do { | |
896 | /* | |
897 | * We deal with the read-ahead logic here. | |
898 | */ | |
899 | if (ra_ptr >= ra_max) { | |
900 | /* Refill the readahead buffer */ | |
901 | ra_ptr = 0; | |
902 | b = block; | |
903 | for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { | |
904 | /* | |
905 | * Terminate if we reach the end of the | |
906 | * directory and must wrap, or if our | |
907 | * search has finished at this block. | |
908 | */ | |
909 | if (b >= nblocks || (num && block == start)) { | |
910 | bh_use[ra_max] = NULL; | |
911 | break; | |
912 | } | |
913 | num++; | |
617ba13b | 914 | bh = ext4_getblk(NULL, dir, b++, 0, &err); |
ac27a0ec DK |
915 | bh_use[ra_max] = bh; |
916 | if (bh) | |
917 | ll_rw_block(READ_META, 1, &bh); | |
918 | } | |
919 | } | |
920 | if ((bh = bh_use[ra_ptr++]) == NULL) | |
921 | goto next; | |
922 | wait_on_buffer(bh); | |
923 | if (!buffer_uptodate(bh)) { | |
924 | /* read error, skip block & hope for the best */ | |
46e665e9 | 925 | ext4_error(sb, __func__, "reading directory #%lu " |
725d26d3 AK |
926 | "offset %lu", dir->i_ino, |
927 | (unsigned long)block); | |
ac27a0ec DK |
928 | brelse(bh); |
929 | goto next; | |
930 | } | |
f702ba0f | 931 | i = search_dirblock(bh, dir, d_name, |
617ba13b | 932 | block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); |
ac27a0ec | 933 | if (i == 1) { |
617ba13b | 934 | EXT4_I(dir)->i_dir_start_lookup = block; |
ac27a0ec DK |
935 | ret = bh; |
936 | goto cleanup_and_exit; | |
937 | } else { | |
938 | brelse(bh); | |
939 | if (i < 0) | |
940 | goto cleanup_and_exit; | |
941 | } | |
942 | next: | |
943 | if (++block >= nblocks) | |
944 | block = 0; | |
945 | } while (block != start); | |
946 | ||
947 | /* | |
948 | * If the directory has grown while we were searching, then | |
949 | * search the last part of the directory before giving up. | |
950 | */ | |
951 | block = nblocks; | |
617ba13b | 952 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); |
ac27a0ec DK |
953 | if (block < nblocks) { |
954 | start = 0; | |
955 | goto restart; | |
956 | } | |
957 | ||
958 | cleanup_and_exit: | |
959 | /* Clean up the read-ahead blocks */ | |
960 | for (; ra_ptr < ra_max; ra_ptr++) | |
af5bc92d | 961 | brelse(bh_use[ra_ptr]); |
ac27a0ec DK |
962 | return ret; |
963 | } | |
964 | ||
f702ba0f | 965 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name, |
617ba13b | 966 | struct ext4_dir_entry_2 **res_dir, int *err) |
ac27a0ec DK |
967 | { |
968 | struct super_block * sb; | |
969 | struct dx_hash_info hinfo; | |
970 | u32 hash; | |
971 | struct dx_frame frames[2], *frame; | |
617ba13b | 972 | struct ext4_dir_entry_2 *de, *top; |
ac27a0ec | 973 | struct buffer_head *bh; |
725d26d3 | 974 | ext4_lblk_t block; |
ac27a0ec | 975 | int retval; |
f702ba0f TT |
976 | int namelen = d_name->len; |
977 | const u8 *name = d_name->name; | |
ac27a0ec DK |
978 | |
979 | sb = dir->i_sb; | |
980 | /* NFS may look up ".." - look at dx_root directory block */ | |
981 | if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){ | |
f702ba0f | 982 | if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err))) |
ac27a0ec DK |
983 | return NULL; |
984 | } else { | |
985 | frame = frames; | |
986 | frame->bh = NULL; /* for dx_release() */ | |
987 | frame->at = (struct dx_entry *)frames; /* hack for zero entry*/ | |
988 | dx_set_block(frame->at, 0); /* dx_root block is 0 */ | |
989 | } | |
990 | hash = hinfo.hash; | |
991 | do { | |
992 | block = dx_get_block(frame->at); | |
617ba13b | 993 | if (!(bh = ext4_bread (NULL,dir, block, 0, err))) |
ac27a0ec | 994 | goto errout; |
617ba13b MC |
995 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
996 | top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize - | |
997 | EXT4_DIR_REC_LEN(0)); | |
f3b35f06 DG |
998 | for (; de < top; de = ext4_next_entry(de)) { |
999 | int off = (block << EXT4_BLOCK_SIZE_BITS(sb)) | |
1000 | + ((char *) de - bh->b_data); | |
1001 | ||
1002 | if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) { | |
1003 | brelse(bh); | |
fedee54d | 1004 | *err = ERR_BAD_DX_DIR; |
ac27a0ec DK |
1005 | goto errout; |
1006 | } | |
f3b35f06 DG |
1007 | |
1008 | if (ext4_match(namelen, name, de)) { | |
1009 | *res_dir = de; | |
1010 | dx_release(frames); | |
1011 | return bh; | |
1012 | } | |
ac27a0ec | 1013 | } |
af5bc92d | 1014 | brelse(bh); |
ac27a0ec | 1015 | /* Check to see if we should continue to search */ |
617ba13b | 1016 | retval = ext4_htree_next_block(dir, hash, frame, |
ac27a0ec DK |
1017 | frames, NULL); |
1018 | if (retval < 0) { | |
46e665e9 | 1019 | ext4_warning(sb, __func__, |
ac27a0ec DK |
1020 | "error reading index page in directory #%lu", |
1021 | dir->i_ino); | |
1022 | *err = retval; | |
1023 | goto errout; | |
1024 | } | |
1025 | } while (retval == 1); | |
1026 | ||
1027 | *err = -ENOENT; | |
1028 | errout: | |
4776004f | 1029 | dxtrace(printk(KERN_DEBUG "%s not found\n", name)); |
ac27a0ec DK |
1030 | dx_release (frames); |
1031 | return NULL; | |
1032 | } | |
ac27a0ec | 1033 | |
af5bc92d | 1034 | static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
ac27a0ec | 1035 | { |
af5bc92d TT |
1036 | struct inode *inode; |
1037 | struct ext4_dir_entry_2 *de; | |
1038 | struct buffer_head *bh; | |
ac27a0ec | 1039 | |
617ba13b | 1040 | if (dentry->d_name.len > EXT4_NAME_LEN) |
ac27a0ec DK |
1041 | return ERR_PTR(-ENAMETOOLONG); |
1042 | ||
f702ba0f | 1043 | bh = ext4_find_entry(dir, &dentry->d_name, &de); |
ac27a0ec DK |
1044 | inode = NULL; |
1045 | if (bh) { | |
1046 | unsigned long ino = le32_to_cpu(de->inode); | |
af5bc92d | 1047 | brelse(bh); |
617ba13b MC |
1048 | if (!ext4_valid_inum(dir->i_sb, ino)) { |
1049 | ext4_error(dir->i_sb, "ext4_lookup", | |
ac27a0ec | 1050 | "bad inode number: %lu", ino); |
1d1fe1ee | 1051 | return ERR_PTR(-EIO); |
a6c15c2b | 1052 | } |
1d1fe1ee DH |
1053 | inode = ext4_iget(dir->i_sb, ino); |
1054 | if (IS_ERR(inode)) | |
1055 | return ERR_CAST(inode); | |
ac27a0ec DK |
1056 | } |
1057 | return d_splice_alias(inode, dentry); | |
1058 | } | |
1059 | ||
1060 | ||
617ba13b | 1061 | struct dentry *ext4_get_parent(struct dentry *child) |
ac27a0ec DK |
1062 | { |
1063 | unsigned long ino; | |
ac27a0ec | 1064 | struct inode *inode; |
f702ba0f TT |
1065 | static const struct qstr dotdot = { |
1066 | .name = "..", | |
1067 | .len = 2, | |
1068 | }; | |
617ba13b | 1069 | struct ext4_dir_entry_2 * de; |
ac27a0ec DK |
1070 | struct buffer_head *bh; |
1071 | ||
f702ba0f | 1072 | bh = ext4_find_entry(child->d_inode, &dotdot, &de); |
ac27a0ec DK |
1073 | inode = NULL; |
1074 | if (!bh) | |
1075 | return ERR_PTR(-ENOENT); | |
1076 | ino = le32_to_cpu(de->inode); | |
1077 | brelse(bh); | |
1078 | ||
617ba13b MC |
1079 | if (!ext4_valid_inum(child->d_inode->i_sb, ino)) { |
1080 | ext4_error(child->d_inode->i_sb, "ext4_get_parent", | |
ac27a0ec | 1081 | "bad inode number: %lu", ino); |
1d1fe1ee | 1082 | return ERR_PTR(-EIO); |
a6c15c2b VA |
1083 | } |
1084 | ||
44003728 | 1085 | return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino)); |
ac27a0ec DK |
1086 | } |
1087 | ||
1088 | #define S_SHIFT 12 | |
617ba13b MC |
1089 | static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = { |
1090 | [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE, | |
1091 | [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR, | |
1092 | [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV, | |
1093 | [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV, | |
1094 | [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO, | |
1095 | [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK, | |
1096 | [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK, | |
ac27a0ec DK |
1097 | }; |
1098 | ||
617ba13b MC |
1099 | static inline void ext4_set_de_type(struct super_block *sb, |
1100 | struct ext4_dir_entry_2 *de, | |
ac27a0ec | 1101 | umode_t mode) { |
617ba13b MC |
1102 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE)) |
1103 | de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT]; | |
ac27a0ec DK |
1104 | } |
1105 | ||
ef2b02d3 ES |
1106 | /* |
1107 | * Move count entries from end of map between two memory locations. | |
1108 | * Returns pointer to last entry moved. | |
1109 | */ | |
617ba13b | 1110 | static struct ext4_dir_entry_2 * |
ac27a0ec DK |
1111 | dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count) |
1112 | { | |
1113 | unsigned rec_len = 0; | |
1114 | ||
1115 | while (count--) { | |
617ba13b MC |
1116 | struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs); |
1117 | rec_len = EXT4_DIR_REC_LEN(de->name_len); | |
ac27a0ec | 1118 | memcpy (to, de, rec_len); |
617ba13b | 1119 | ((struct ext4_dir_entry_2 *) to)->rec_len = |
a72d7f83 | 1120 | ext4_rec_len_to_disk(rec_len); |
ac27a0ec DK |
1121 | de->inode = 0; |
1122 | map++; | |
1123 | to += rec_len; | |
1124 | } | |
617ba13b | 1125 | return (struct ext4_dir_entry_2 *) (to - rec_len); |
ac27a0ec DK |
1126 | } |
1127 | ||
ef2b02d3 ES |
1128 | /* |
1129 | * Compact each dir entry in the range to the minimal rec_len. | |
1130 | * Returns pointer to last entry in range. | |
1131 | */ | |
617ba13b | 1132 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size) |
ac27a0ec | 1133 | { |
617ba13b | 1134 | struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; |
ac27a0ec DK |
1135 | unsigned rec_len = 0; |
1136 | ||
1137 | prev = to = de; | |
1138 | while ((char*)de < base + size) { | |
a72d7f83 | 1139 | next = ext4_next_entry(de); |
ac27a0ec | 1140 | if (de->inode && de->name_len) { |
617ba13b | 1141 | rec_len = EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec DK |
1142 | if (de > to) |
1143 | memmove(to, de, rec_len); | |
a72d7f83 | 1144 | to->rec_len = ext4_rec_len_to_disk(rec_len); |
ac27a0ec | 1145 | prev = to; |
617ba13b | 1146 | to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); |
ac27a0ec DK |
1147 | } |
1148 | de = next; | |
1149 | } | |
1150 | return prev; | |
1151 | } | |
1152 | ||
ef2b02d3 ES |
1153 | /* |
1154 | * Split a full leaf block to make room for a new dir entry. | |
1155 | * Allocate a new block, and move entries so that they are approx. equally full. | |
1156 | * Returns pointer to de in block into which the new entry will be inserted. | |
1157 | */ | |
617ba13b | 1158 | static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, |
ac27a0ec DK |
1159 | struct buffer_head **bh,struct dx_frame *frame, |
1160 | struct dx_hash_info *hinfo, int *error) | |
1161 | { | |
1162 | unsigned blocksize = dir->i_sb->s_blocksize; | |
1163 | unsigned count, continued; | |
1164 | struct buffer_head *bh2; | |
725d26d3 | 1165 | ext4_lblk_t newblock; |
ac27a0ec DK |
1166 | u32 hash2; |
1167 | struct dx_map_entry *map; | |
1168 | char *data1 = (*bh)->b_data, *data2; | |
ef2b02d3 | 1169 | unsigned split, move, size, i; |
617ba13b | 1170 | struct ext4_dir_entry_2 *de = NULL, *de2; |
fedee54d | 1171 | int err = 0; |
ac27a0ec | 1172 | |
fedee54d | 1173 | bh2 = ext4_append (handle, dir, &newblock, &err); |
ac27a0ec DK |
1174 | if (!(bh2)) { |
1175 | brelse(*bh); | |
1176 | *bh = NULL; | |
1177 | goto errout; | |
1178 | } | |
1179 | ||
1180 | BUFFER_TRACE(*bh, "get_write_access"); | |
617ba13b | 1181 | err = ext4_journal_get_write_access(handle, *bh); |
fedee54d DM |
1182 | if (err) |
1183 | goto journal_error; | |
1184 | ||
ac27a0ec | 1185 | BUFFER_TRACE(frame->bh, "get_write_access"); |
617ba13b | 1186 | err = ext4_journal_get_write_access(handle, frame->bh); |
ac27a0ec DK |
1187 | if (err) |
1188 | goto journal_error; | |
1189 | ||
1190 | data2 = bh2->b_data; | |
1191 | ||
1192 | /* create map in the end of data2 block */ | |
1193 | map = (struct dx_map_entry *) (data2 + blocksize); | |
af5bc92d | 1194 | count = dx_make_map((struct ext4_dir_entry_2 *) data1, |
ac27a0ec DK |
1195 | blocksize, hinfo, map); |
1196 | map -= count; | |
af5bc92d | 1197 | dx_sort_map(map, count); |
ef2b02d3 ES |
1198 | /* Split the existing block in the middle, size-wise */ |
1199 | size = 0; | |
1200 | move = 0; | |
1201 | for (i = count-1; i >= 0; i--) { | |
1202 | /* is more than half of this entry in 2nd half of the block? */ | |
1203 | if (size + map[i].size/2 > blocksize/2) | |
1204 | break; | |
1205 | size += map[i].size; | |
1206 | move++; | |
1207 | } | |
1208 | /* map index at which we will split */ | |
1209 | split = count - move; | |
ac27a0ec DK |
1210 | hash2 = map[split].hash; |
1211 | continued = hash2 == map[split - 1].hash; | |
725d26d3 AK |
1212 | dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", |
1213 | (unsigned long)dx_get_block(frame->at), | |
1214 | hash2, split, count-split)); | |
ac27a0ec DK |
1215 | |
1216 | /* Fancy dance to stay within two buffers */ | |
1217 | de2 = dx_move_dirents(data1, data2, map + split, count - split); | |
af5bc92d | 1218 | de = dx_pack_dirents(data1, blocksize); |
a72d7f83 JK |
1219 | de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de); |
1220 | de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2); | |
617ba13b MC |
1221 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1)); |
1222 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1)); | |
ac27a0ec DK |
1223 | |
1224 | /* Which block gets the new entry? */ | |
1225 | if (hinfo->hash >= hash2) | |
1226 | { | |
1227 | swap(*bh, bh2); | |
1228 | de = de2; | |
1229 | } | |
af5bc92d TT |
1230 | dx_insert_block(frame, hash2 + continued, newblock); |
1231 | err = ext4_journal_dirty_metadata(handle, bh2); | |
ac27a0ec DK |
1232 | if (err) |
1233 | goto journal_error; | |
af5bc92d | 1234 | err = ext4_journal_dirty_metadata(handle, frame->bh); |
ac27a0ec DK |
1235 | if (err) |
1236 | goto journal_error; | |
af5bc92d TT |
1237 | brelse(bh2); |
1238 | dxtrace(dx_show_index("frame", frame->entries)); | |
ac27a0ec | 1239 | return de; |
fedee54d DM |
1240 | |
1241 | journal_error: | |
1242 | brelse(*bh); | |
1243 | brelse(bh2); | |
1244 | *bh = NULL; | |
1245 | ext4_std_error(dir->i_sb, err); | |
1246 | errout: | |
1247 | *error = err; | |
1248 | return NULL; | |
ac27a0ec | 1249 | } |
ac27a0ec DK |
1250 | |
1251 | /* | |
1252 | * Add a new entry into a directory (leaf) block. If de is non-NULL, | |
1253 | * it points to a directory entry which is guaranteed to be large | |
1254 | * enough for new directory entry. If de is NULL, then | |
1255 | * add_dirent_to_buf will attempt search the directory block for | |
1256 | * space. It will return -ENOSPC if no space is available, and -EIO | |
1257 | * and -EEXIST if directory entry already exists. | |
1258 | * | |
1259 | * NOTE! bh is NOT released in the case where ENOSPC is returned. In | |
1260 | * all other cases bh is released. | |
1261 | */ | |
1262 | static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry, | |
617ba13b | 1263 | struct inode *inode, struct ext4_dir_entry_2 *de, |
af5bc92d | 1264 | struct buffer_head *bh) |
ac27a0ec DK |
1265 | { |
1266 | struct inode *dir = dentry->d_parent->d_inode; | |
1267 | const char *name = dentry->d_name.name; | |
1268 | int namelen = dentry->d_name.len; | |
1269 | unsigned long offset = 0; | |
1270 | unsigned short reclen; | |
1271 | int nlen, rlen, err; | |
1272 | char *top; | |
1273 | ||
617ba13b | 1274 | reclen = EXT4_DIR_REC_LEN(namelen); |
ac27a0ec | 1275 | if (!de) { |
617ba13b | 1276 | de = (struct ext4_dir_entry_2 *)bh->b_data; |
ac27a0ec DK |
1277 | top = bh->b_data + dir->i_sb->s_blocksize - reclen; |
1278 | while ((char *) de <= top) { | |
617ba13b | 1279 | if (!ext4_check_dir_entry("ext4_add_entry", dir, de, |
ac27a0ec | 1280 | bh, offset)) { |
af5bc92d | 1281 | brelse(bh); |
ac27a0ec DK |
1282 | return -EIO; |
1283 | } | |
af5bc92d TT |
1284 | if (ext4_match(namelen, name, de)) { |
1285 | brelse(bh); | |
ac27a0ec DK |
1286 | return -EEXIST; |
1287 | } | |
617ba13b | 1288 | nlen = EXT4_DIR_REC_LEN(de->name_len); |
a72d7f83 | 1289 | rlen = ext4_rec_len_from_disk(de->rec_len); |
ac27a0ec DK |
1290 | if ((de->inode? rlen - nlen: rlen) >= reclen) |
1291 | break; | |
617ba13b | 1292 | de = (struct ext4_dir_entry_2 *)((char *)de + rlen); |
ac27a0ec DK |
1293 | offset += rlen; |
1294 | } | |
1295 | if ((char *) de > top) | |
1296 | return -ENOSPC; | |
1297 | } | |
1298 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1299 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1300 | if (err) { |
617ba13b | 1301 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1302 | brelse(bh); |
1303 | return err; | |
1304 | } | |
1305 | ||
1306 | /* By now the buffer is marked for journaling */ | |
617ba13b | 1307 | nlen = EXT4_DIR_REC_LEN(de->name_len); |
a72d7f83 | 1308 | rlen = ext4_rec_len_from_disk(de->rec_len); |
ac27a0ec | 1309 | if (de->inode) { |
617ba13b | 1310 | struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen); |
a72d7f83 JK |
1311 | de1->rec_len = ext4_rec_len_to_disk(rlen - nlen); |
1312 | de->rec_len = ext4_rec_len_to_disk(nlen); | |
ac27a0ec DK |
1313 | de = de1; |
1314 | } | |
617ba13b | 1315 | de->file_type = EXT4_FT_UNKNOWN; |
ac27a0ec DK |
1316 | if (inode) { |
1317 | de->inode = cpu_to_le32(inode->i_ino); | |
617ba13b | 1318 | ext4_set_de_type(dir->i_sb, de, inode->i_mode); |
ac27a0ec DK |
1319 | } else |
1320 | de->inode = 0; | |
1321 | de->name_len = namelen; | |
af5bc92d | 1322 | memcpy(de->name, name, namelen); |
ac27a0ec DK |
1323 | /* |
1324 | * XXX shouldn't update any times until successful | |
1325 | * completion of syscall, but too many callers depend | |
1326 | * on this. | |
1327 | * | |
1328 | * XXX similarly, too many callers depend on | |
617ba13b | 1329 | * ext4_new_inode() setting the times, but error |
ac27a0ec DK |
1330 | * recovery deletes the inode, so the worst that can |
1331 | * happen is that the times are slightly out of date | |
1332 | * and/or different from the directory change time. | |
1333 | */ | |
ef7f3835 | 1334 | dir->i_mtime = dir->i_ctime = ext4_current_time(dir); |
617ba13b | 1335 | ext4_update_dx_flag(dir); |
ac27a0ec | 1336 | dir->i_version++; |
617ba13b MC |
1337 | ext4_mark_inode_dirty(handle, dir); |
1338 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); | |
1339 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec | 1340 | if (err) |
617ba13b | 1341 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1342 | brelse(bh); |
1343 | return 0; | |
1344 | } | |
1345 | ||
ac27a0ec DK |
1346 | /* |
1347 | * This converts a one block unindexed directory to a 3 block indexed | |
1348 | * directory, and adds the dentry to the indexed directory. | |
1349 | */ | |
1350 | static int make_indexed_dir(handle_t *handle, struct dentry *dentry, | |
1351 | struct inode *inode, struct buffer_head *bh) | |
1352 | { | |
1353 | struct inode *dir = dentry->d_parent->d_inode; | |
1354 | const char *name = dentry->d_name.name; | |
1355 | int namelen = dentry->d_name.len; | |
1356 | struct buffer_head *bh2; | |
1357 | struct dx_root *root; | |
1358 | struct dx_frame frames[2], *frame; | |
1359 | struct dx_entry *entries; | |
617ba13b | 1360 | struct ext4_dir_entry_2 *de, *de2; |
ac27a0ec DK |
1361 | char *data1, *top; |
1362 | unsigned len; | |
1363 | int retval; | |
1364 | unsigned blocksize; | |
1365 | struct dx_hash_info hinfo; | |
725d26d3 | 1366 | ext4_lblk_t block; |
ac27a0ec DK |
1367 | struct fake_dirent *fde; |
1368 | ||
1369 | blocksize = dir->i_sb->s_blocksize; | |
4776004f | 1370 | dxtrace(printk(KERN_DEBUG "Creating index\n")); |
617ba13b | 1371 | retval = ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1372 | if (retval) { |
617ba13b | 1373 | ext4_std_error(dir->i_sb, retval); |
ac27a0ec DK |
1374 | brelse(bh); |
1375 | return retval; | |
1376 | } | |
1377 | root = (struct dx_root *) bh->b_data; | |
1378 | ||
af5bc92d | 1379 | bh2 = ext4_append(handle, dir, &block, &retval); |
ac27a0ec DK |
1380 | if (!(bh2)) { |
1381 | brelse(bh); | |
1382 | return retval; | |
1383 | } | |
617ba13b | 1384 | EXT4_I(dir)->i_flags |= EXT4_INDEX_FL; |
ac27a0ec DK |
1385 | data1 = bh2->b_data; |
1386 | ||
1387 | /* The 0th block becomes the root, move the dirents out */ | |
1388 | fde = &root->dotdot; | |
a72d7f83 JK |
1389 | de = (struct ext4_dir_entry_2 *)((char *)fde + |
1390 | ext4_rec_len_from_disk(fde->rec_len)); | |
ac27a0ec DK |
1391 | len = ((char *) root) + blocksize - (char *) de; |
1392 | memcpy (data1, de, len); | |
617ba13b | 1393 | de = (struct ext4_dir_entry_2 *) data1; |
ac27a0ec | 1394 | top = data1 + len; |
a72d7f83 | 1395 | while ((char *)(de2 = ext4_next_entry(de)) < top) |
ac27a0ec | 1396 | de = de2; |
a72d7f83 | 1397 | de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de); |
ac27a0ec | 1398 | /* Initialize the root; the dot dirents already exist */ |
617ba13b | 1399 | de = (struct ext4_dir_entry_2 *) (&root->dotdot); |
a72d7f83 | 1400 | de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2)); |
ac27a0ec DK |
1401 | memset (&root->info, 0, sizeof(root->info)); |
1402 | root->info.info_length = sizeof(root->info); | |
617ba13b | 1403 | root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; |
ac27a0ec | 1404 | entries = root->entries; |
af5bc92d TT |
1405 | dx_set_block(entries, 1); |
1406 | dx_set_count(entries, 1); | |
1407 | dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); | |
ac27a0ec DK |
1408 | |
1409 | /* Initialize as for dx_probe */ | |
1410 | hinfo.hash_version = root->info.hash_version; | |
617ba13b MC |
1411 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
1412 | ext4fs_dirhash(name, namelen, &hinfo); | |
ac27a0ec DK |
1413 | frame = frames; |
1414 | frame->entries = entries; | |
1415 | frame->at = entries; | |
1416 | frame->bh = bh; | |
1417 | bh = bh2; | |
1418 | de = do_split(handle,dir, &bh, frame, &hinfo, &retval); | |
1419 | dx_release (frames); | |
1420 | if (!(de)) | |
1421 | return retval; | |
1422 | ||
1423 | return add_dirent_to_buf(handle, dentry, inode, de, bh); | |
1424 | } | |
ac27a0ec DK |
1425 | |
1426 | /* | |
617ba13b | 1427 | * ext4_add_entry() |
ac27a0ec DK |
1428 | * |
1429 | * adds a file entry to the specified directory, using the same | |
617ba13b | 1430 | * semantics as ext4_find_entry(). It returns NULL if it failed. |
ac27a0ec DK |
1431 | * |
1432 | * NOTE!! The inode part of 'de' is left at 0 - which means you | |
1433 | * may not sleep between calling this and putting something into | |
1434 | * the entry, as someone else might have used it while you slept. | |
1435 | */ | |
af5bc92d TT |
1436 | static int ext4_add_entry(handle_t *handle, struct dentry *dentry, |
1437 | struct inode *inode) | |
ac27a0ec DK |
1438 | { |
1439 | struct inode *dir = dentry->d_parent->d_inode; | |
1440 | unsigned long offset; | |
af5bc92d | 1441 | struct buffer_head *bh; |
617ba13b | 1442 | struct ext4_dir_entry_2 *de; |
af5bc92d | 1443 | struct super_block *sb; |
ac27a0ec | 1444 | int retval; |
ac27a0ec | 1445 | int dx_fallback=0; |
ac27a0ec | 1446 | unsigned blocksize; |
725d26d3 | 1447 | ext4_lblk_t block, blocks; |
ac27a0ec DK |
1448 | |
1449 | sb = dir->i_sb; | |
1450 | blocksize = sb->s_blocksize; | |
1451 | if (!dentry->d_name.len) | |
1452 | return -EINVAL; | |
ac27a0ec | 1453 | if (is_dx(dir)) { |
617ba13b | 1454 | retval = ext4_dx_add_entry(handle, dentry, inode); |
ac27a0ec DK |
1455 | if (!retval || (retval != ERR_BAD_DX_DIR)) |
1456 | return retval; | |
617ba13b | 1457 | EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL; |
ac27a0ec | 1458 | dx_fallback++; |
617ba13b | 1459 | ext4_mark_inode_dirty(handle, dir); |
ac27a0ec | 1460 | } |
ac27a0ec DK |
1461 | blocks = dir->i_size >> sb->s_blocksize_bits; |
1462 | for (block = 0, offset = 0; block < blocks; block++) { | |
617ba13b | 1463 | bh = ext4_bread(handle, dir, block, 0, &retval); |
ac27a0ec DK |
1464 | if(!bh) |
1465 | return retval; | |
1466 | retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
1467 | if (retval != -ENOSPC) | |
1468 | return retval; | |
1469 | ||
ac27a0ec | 1470 | if (blocks == 1 && !dx_fallback && |
617ba13b | 1471 | EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) |
ac27a0ec | 1472 | return make_indexed_dir(handle, dentry, inode, bh); |
ac27a0ec DK |
1473 | brelse(bh); |
1474 | } | |
617ba13b | 1475 | bh = ext4_append(handle, dir, &block, &retval); |
ac27a0ec DK |
1476 | if (!bh) |
1477 | return retval; | |
617ba13b | 1478 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 1479 | de->inode = 0; |
a72d7f83 | 1480 | de->rec_len = ext4_rec_len_to_disk(blocksize); |
ac27a0ec DK |
1481 | return add_dirent_to_buf(handle, dentry, inode, de, bh); |
1482 | } | |
1483 | ||
ac27a0ec DK |
1484 | /* |
1485 | * Returns 0 for success, or a negative error value | |
1486 | */ | |
617ba13b | 1487 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, |
ac27a0ec DK |
1488 | struct inode *inode) |
1489 | { | |
1490 | struct dx_frame frames[2], *frame; | |
1491 | struct dx_entry *entries, *at; | |
1492 | struct dx_hash_info hinfo; | |
af5bc92d | 1493 | struct buffer_head *bh; |
ac27a0ec | 1494 | struct inode *dir = dentry->d_parent->d_inode; |
af5bc92d | 1495 | struct super_block *sb = dir->i_sb; |
617ba13b | 1496 | struct ext4_dir_entry_2 *de; |
ac27a0ec DK |
1497 | int err; |
1498 | ||
f702ba0f | 1499 | frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err); |
ac27a0ec DK |
1500 | if (!frame) |
1501 | return err; | |
1502 | entries = frame->entries; | |
1503 | at = frame->at; | |
1504 | ||
617ba13b | 1505 | if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err))) |
ac27a0ec DK |
1506 | goto cleanup; |
1507 | ||
1508 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1509 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1510 | if (err) |
1511 | goto journal_error; | |
1512 | ||
1513 | err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
1514 | if (err != -ENOSPC) { | |
1515 | bh = NULL; | |
1516 | goto cleanup; | |
1517 | } | |
1518 | ||
1519 | /* Block full, should compress but for now just split */ | |
4776004f | 1520 | dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", |
ac27a0ec DK |
1521 | dx_get_count(entries), dx_get_limit(entries))); |
1522 | /* Need to split index? */ | |
1523 | if (dx_get_count(entries) == dx_get_limit(entries)) { | |
725d26d3 | 1524 | ext4_lblk_t newblock; |
ac27a0ec DK |
1525 | unsigned icount = dx_get_count(entries); |
1526 | int levels = frame - frames; | |
1527 | struct dx_entry *entries2; | |
1528 | struct dx_node *node2; | |
1529 | struct buffer_head *bh2; | |
1530 | ||
1531 | if (levels && (dx_get_count(frames->entries) == | |
1532 | dx_get_limit(frames->entries))) { | |
46e665e9 | 1533 | ext4_warning(sb, __func__, |
ac27a0ec DK |
1534 | "Directory index full!"); |
1535 | err = -ENOSPC; | |
1536 | goto cleanup; | |
1537 | } | |
617ba13b | 1538 | bh2 = ext4_append (handle, dir, &newblock, &err); |
ac27a0ec DK |
1539 | if (!(bh2)) |
1540 | goto cleanup; | |
1541 | node2 = (struct dx_node *)(bh2->b_data); | |
1542 | entries2 = node2->entries; | |
a72d7f83 | 1543 | node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize); |
ac27a0ec DK |
1544 | node2->fake.inode = 0; |
1545 | BUFFER_TRACE(frame->bh, "get_write_access"); | |
617ba13b | 1546 | err = ext4_journal_get_write_access(handle, frame->bh); |
ac27a0ec DK |
1547 | if (err) |
1548 | goto journal_error; | |
1549 | if (levels) { | |
1550 | unsigned icount1 = icount/2, icount2 = icount - icount1; | |
1551 | unsigned hash2 = dx_get_hash(entries + icount1); | |
4776004f TT |
1552 | dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", |
1553 | icount1, icount2)); | |
ac27a0ec DK |
1554 | |
1555 | BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ | |
617ba13b | 1556 | err = ext4_journal_get_write_access(handle, |
ac27a0ec DK |
1557 | frames[0].bh); |
1558 | if (err) | |
1559 | goto journal_error; | |
1560 | ||
af5bc92d TT |
1561 | memcpy((char *) entries2, (char *) (entries + icount1), |
1562 | icount2 * sizeof(struct dx_entry)); | |
1563 | dx_set_count(entries, icount1); | |
1564 | dx_set_count(entries2, icount2); | |
1565 | dx_set_limit(entries2, dx_node_limit(dir)); | |
ac27a0ec DK |
1566 | |
1567 | /* Which index block gets the new entry? */ | |
1568 | if (at - entries >= icount1) { | |
1569 | frame->at = at = at - entries - icount1 + entries2; | |
1570 | frame->entries = entries = entries2; | |
1571 | swap(frame->bh, bh2); | |
1572 | } | |
af5bc92d TT |
1573 | dx_insert_block(frames + 0, hash2, newblock); |
1574 | dxtrace(dx_show_index("node", frames[1].entries)); | |
1575 | dxtrace(dx_show_index("node", | |
ac27a0ec | 1576 | ((struct dx_node *) bh2->b_data)->entries)); |
617ba13b | 1577 | err = ext4_journal_dirty_metadata(handle, bh2); |
ac27a0ec DK |
1578 | if (err) |
1579 | goto journal_error; | |
1580 | brelse (bh2); | |
1581 | } else { | |
4776004f TT |
1582 | dxtrace(printk(KERN_DEBUG |
1583 | "Creating second level index...\n")); | |
ac27a0ec DK |
1584 | memcpy((char *) entries2, (char *) entries, |
1585 | icount * sizeof(struct dx_entry)); | |
1586 | dx_set_limit(entries2, dx_node_limit(dir)); | |
1587 | ||
1588 | /* Set up root */ | |
1589 | dx_set_count(entries, 1); | |
1590 | dx_set_block(entries + 0, newblock); | |
1591 | ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; | |
1592 | ||
1593 | /* Add new access path frame */ | |
1594 | frame = frames + 1; | |
1595 | frame->at = at = at - entries + entries2; | |
1596 | frame->entries = entries = entries2; | |
1597 | frame->bh = bh2; | |
617ba13b | 1598 | err = ext4_journal_get_write_access(handle, |
ac27a0ec DK |
1599 | frame->bh); |
1600 | if (err) | |
1601 | goto journal_error; | |
1602 | } | |
617ba13b | 1603 | ext4_journal_dirty_metadata(handle, frames[0].bh); |
ac27a0ec DK |
1604 | } |
1605 | de = do_split(handle, dir, &bh, frame, &hinfo, &err); | |
1606 | if (!de) | |
1607 | goto cleanup; | |
1608 | err = add_dirent_to_buf(handle, dentry, inode, de, bh); | |
1609 | bh = NULL; | |
1610 | goto cleanup; | |
1611 | ||
1612 | journal_error: | |
617ba13b | 1613 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1614 | cleanup: |
1615 | if (bh) | |
1616 | brelse(bh); | |
1617 | dx_release(frames); | |
1618 | return err; | |
1619 | } | |
ac27a0ec DK |
1620 | |
1621 | /* | |
617ba13b | 1622 | * ext4_delete_entry deletes a directory entry by merging it with the |
ac27a0ec DK |
1623 | * previous entry |
1624 | */ | |
af5bc92d TT |
1625 | static int ext4_delete_entry(handle_t *handle, |
1626 | struct inode *dir, | |
1627 | struct ext4_dir_entry_2 *de_del, | |
1628 | struct buffer_head *bh) | |
ac27a0ec | 1629 | { |
af5bc92d | 1630 | struct ext4_dir_entry_2 *de, *pde; |
ac27a0ec DK |
1631 | int i; |
1632 | ||
1633 | i = 0; | |
1634 | pde = NULL; | |
617ba13b | 1635 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 1636 | while (i < bh->b_size) { |
617ba13b | 1637 | if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i)) |
ac27a0ec DK |
1638 | return -EIO; |
1639 | if (de == de_del) { | |
1640 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1641 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1642 | if (pde) |
a72d7f83 JK |
1643 | pde->rec_len = ext4_rec_len_to_disk( |
1644 | ext4_rec_len_from_disk(pde->rec_len) + | |
1645 | ext4_rec_len_from_disk(de->rec_len)); | |
ac27a0ec DK |
1646 | else |
1647 | de->inode = 0; | |
1648 | dir->i_version++; | |
617ba13b MC |
1649 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
1650 | ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
1651 | return 0; |
1652 | } | |
a72d7f83 | 1653 | i += ext4_rec_len_from_disk(de->rec_len); |
ac27a0ec | 1654 | pde = de; |
a72d7f83 | 1655 | de = ext4_next_entry(de); |
ac27a0ec DK |
1656 | } |
1657 | return -ENOENT; | |
1658 | } | |
1659 | ||
f8628a14 AD |
1660 | /* |
1661 | * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2, | |
1662 | * since this indicates that nlinks count was previously 1. | |
1663 | */ | |
1664 | static void ext4_inc_count(handle_t *handle, struct inode *inode) | |
1665 | { | |
1666 | inc_nlink(inode); | |
1667 | if (is_dx(inode) && inode->i_nlink > 1) { | |
1668 | /* limit is 16-bit i_links_count */ | |
1669 | if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) { | |
1670 | inode->i_nlink = 1; | |
1671 | EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb, | |
1672 | EXT4_FEATURE_RO_COMPAT_DIR_NLINK); | |
1673 | } | |
1674 | } | |
1675 | } | |
1676 | ||
1677 | /* | |
1678 | * If a directory had nlink == 1, then we should let it be 1. This indicates | |
1679 | * directory has >EXT4_LINK_MAX subdirs. | |
1680 | */ | |
1681 | static void ext4_dec_count(handle_t *handle, struct inode *inode) | |
1682 | { | |
1683 | drop_nlink(inode); | |
1684 | if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0) | |
1685 | inc_nlink(inode); | |
1686 | } | |
1687 | ||
1688 | ||
617ba13b | 1689 | static int ext4_add_nondir(handle_t *handle, |
ac27a0ec DK |
1690 | struct dentry *dentry, struct inode *inode) |
1691 | { | |
617ba13b | 1692 | int err = ext4_add_entry(handle, dentry, inode); |
ac27a0ec | 1693 | if (!err) { |
617ba13b | 1694 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 1695 | d_instantiate(dentry, inode); |
6b38e842 | 1696 | unlock_new_inode(inode); |
ac27a0ec DK |
1697 | return 0; |
1698 | } | |
731b9a54 | 1699 | drop_nlink(inode); |
6b38e842 | 1700 | unlock_new_inode(inode); |
ac27a0ec DK |
1701 | iput(inode); |
1702 | return err; | |
1703 | } | |
1704 | ||
1705 | /* | |
1706 | * By the time this is called, we already have created | |
1707 | * the directory cache entry for the new file, but it | |
1708 | * is so far negative - it has no inode. | |
1709 | * | |
1710 | * If the create succeeds, we fill in the inode information | |
1711 | * with d_instantiate(). | |
1712 | */ | |
af5bc92d TT |
1713 | static int ext4_create(struct inode *dir, struct dentry *dentry, int mode, |
1714 | struct nameidata *nd) | |
ac27a0ec DK |
1715 | { |
1716 | handle_t *handle; | |
af5bc92d | 1717 | struct inode *inode; |
ac27a0ec DK |
1718 | int err, retries = 0; |
1719 | ||
1720 | retry: | |
617ba13b MC |
1721 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
1722 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
1723 | 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb)); | |
ac27a0ec DK |
1724 | if (IS_ERR(handle)) |
1725 | return PTR_ERR(handle); | |
1726 | ||
1727 | if (IS_DIRSYNC(dir)) | |
1728 | handle->h_sync = 1; | |
1729 | ||
617ba13b | 1730 | inode = ext4_new_inode (handle, dir, mode); |
ac27a0ec DK |
1731 | err = PTR_ERR(inode); |
1732 | if (!IS_ERR(inode)) { | |
617ba13b MC |
1733 | inode->i_op = &ext4_file_inode_operations; |
1734 | inode->i_fop = &ext4_file_operations; | |
1735 | ext4_set_aops(inode); | |
1736 | err = ext4_add_nondir(handle, dentry, inode); | |
ac27a0ec | 1737 | } |
617ba13b MC |
1738 | ext4_journal_stop(handle); |
1739 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
1740 | goto retry; |
1741 | return err; | |
1742 | } | |
1743 | ||
af5bc92d TT |
1744 | static int ext4_mknod(struct inode *dir, struct dentry *dentry, |
1745 | int mode, dev_t rdev) | |
ac27a0ec DK |
1746 | { |
1747 | handle_t *handle; | |
1748 | struct inode *inode; | |
1749 | int err, retries = 0; | |
1750 | ||
1751 | if (!new_valid_dev(rdev)) | |
1752 | return -EINVAL; | |
1753 | ||
1754 | retry: | |
617ba13b MC |
1755 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
1756 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
1757 | 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb)); | |
ac27a0ec DK |
1758 | if (IS_ERR(handle)) |
1759 | return PTR_ERR(handle); | |
1760 | ||
1761 | if (IS_DIRSYNC(dir)) | |
1762 | handle->h_sync = 1; | |
1763 | ||
af5bc92d | 1764 | inode = ext4_new_inode(handle, dir, mode); |
ac27a0ec DK |
1765 | err = PTR_ERR(inode); |
1766 | if (!IS_ERR(inode)) { | |
1767 | init_special_inode(inode, inode->i_mode, rdev); | |
03010a33 | 1768 | #ifdef CONFIG_EXT4_FS_XATTR |
617ba13b | 1769 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec | 1770 | #endif |
617ba13b | 1771 | err = ext4_add_nondir(handle, dentry, inode); |
ac27a0ec | 1772 | } |
617ba13b MC |
1773 | ext4_journal_stop(handle); |
1774 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
1775 | goto retry; |
1776 | return err; | |
1777 | } | |
1778 | ||
af5bc92d | 1779 | static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
ac27a0ec DK |
1780 | { |
1781 | handle_t *handle; | |
af5bc92d TT |
1782 | struct inode *inode; |
1783 | struct buffer_head *dir_block; | |
1784 | struct ext4_dir_entry_2 *de; | |
ac27a0ec DK |
1785 | int err, retries = 0; |
1786 | ||
f8628a14 | 1787 | if (EXT4_DIR_LINK_MAX(dir)) |
ac27a0ec DK |
1788 | return -EMLINK; |
1789 | ||
1790 | retry: | |
617ba13b MC |
1791 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
1792 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
1793 | 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb)); | |
ac27a0ec DK |
1794 | if (IS_ERR(handle)) |
1795 | return PTR_ERR(handle); | |
1796 | ||
1797 | if (IS_DIRSYNC(dir)) | |
1798 | handle->h_sync = 1; | |
1799 | ||
af5bc92d | 1800 | inode = ext4_new_inode(handle, dir, S_IFDIR | mode); |
ac27a0ec DK |
1801 | err = PTR_ERR(inode); |
1802 | if (IS_ERR(inode)) | |
1803 | goto out_stop; | |
1804 | ||
617ba13b MC |
1805 | inode->i_op = &ext4_dir_inode_operations; |
1806 | inode->i_fop = &ext4_dir_operations; | |
1807 | inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize; | |
af5bc92d | 1808 | dir_block = ext4_bread(handle, inode, 0, 1, &err); |
4cdeed86 AK |
1809 | if (!dir_block) |
1810 | goto out_clear_inode; | |
ac27a0ec | 1811 | BUFFER_TRACE(dir_block, "get_write_access"); |
617ba13b MC |
1812 | ext4_journal_get_write_access(handle, dir_block); |
1813 | de = (struct ext4_dir_entry_2 *) dir_block->b_data; | |
ac27a0ec DK |
1814 | de->inode = cpu_to_le32(inode->i_ino); |
1815 | de->name_len = 1; | |
a72d7f83 | 1816 | de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len)); |
af5bc92d | 1817 | strcpy(de->name, "."); |
617ba13b | 1818 | ext4_set_de_type(dir->i_sb, de, S_IFDIR); |
a72d7f83 | 1819 | de = ext4_next_entry(de); |
ac27a0ec | 1820 | de->inode = cpu_to_le32(dir->i_ino); |
a72d7f83 JK |
1821 | de->rec_len = ext4_rec_len_to_disk(inode->i_sb->s_blocksize - |
1822 | EXT4_DIR_REC_LEN(1)); | |
ac27a0ec | 1823 | de->name_len = 2; |
af5bc92d | 1824 | strcpy(de->name, ".."); |
617ba13b | 1825 | ext4_set_de_type(dir->i_sb, de, S_IFDIR); |
ac27a0ec | 1826 | inode->i_nlink = 2; |
617ba13b MC |
1827 | BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata"); |
1828 | ext4_journal_dirty_metadata(handle, dir_block); | |
af5bc92d | 1829 | brelse(dir_block); |
617ba13b | 1830 | ext4_mark_inode_dirty(handle, inode); |
af5bc92d | 1831 | err = ext4_add_entry(handle, dentry, inode); |
ac27a0ec | 1832 | if (err) { |
4cdeed86 AK |
1833 | out_clear_inode: |
1834 | clear_nlink(inode); | |
6b38e842 | 1835 | unlock_new_inode(inode); |
617ba13b | 1836 | ext4_mark_inode_dirty(handle, inode); |
af5bc92d | 1837 | iput(inode); |
ac27a0ec DK |
1838 | goto out_stop; |
1839 | } | |
f8628a14 | 1840 | ext4_inc_count(handle, dir); |
617ba13b MC |
1841 | ext4_update_dx_flag(dir); |
1842 | ext4_mark_inode_dirty(handle, dir); | |
ac27a0ec | 1843 | d_instantiate(dentry, inode); |
6b38e842 | 1844 | unlock_new_inode(inode); |
ac27a0ec | 1845 | out_stop: |
617ba13b MC |
1846 | ext4_journal_stop(handle); |
1847 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
1848 | goto retry; |
1849 | return err; | |
1850 | } | |
1851 | ||
1852 | /* | |
1853 | * routine to check that the specified directory is empty (for rmdir) | |
1854 | */ | |
af5bc92d | 1855 | static int empty_dir(struct inode *inode) |
ac27a0ec DK |
1856 | { |
1857 | unsigned long offset; | |
af5bc92d TT |
1858 | struct buffer_head *bh; |
1859 | struct ext4_dir_entry_2 *de, *de1; | |
1860 | struct super_block *sb; | |
ac27a0ec DK |
1861 | int err = 0; |
1862 | ||
1863 | sb = inode->i_sb; | |
617ba13b | 1864 | if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) || |
af5bc92d | 1865 | !(bh = ext4_bread(NULL, inode, 0, 0, &err))) { |
ac27a0ec | 1866 | if (err) |
46e665e9 | 1867 | ext4_error(inode->i_sb, __func__, |
ac27a0ec DK |
1868 | "error %d reading directory #%lu offset 0", |
1869 | err, inode->i_ino); | |
1870 | else | |
46e665e9 | 1871 | ext4_warning(inode->i_sb, __func__, |
ac27a0ec DK |
1872 | "bad directory (dir #%lu) - no data block", |
1873 | inode->i_ino); | |
1874 | return 1; | |
1875 | } | |
617ba13b | 1876 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
a72d7f83 | 1877 | de1 = ext4_next_entry(de); |
ac27a0ec DK |
1878 | if (le32_to_cpu(de->inode) != inode->i_ino || |
1879 | !le32_to_cpu(de1->inode) || | |
af5bc92d TT |
1880 | strcmp(".", de->name) || |
1881 | strcmp("..", de1->name)) { | |
1882 | ext4_warning(inode->i_sb, "empty_dir", | |
1883 | "bad directory (dir #%lu) - no `.' or `..'", | |
1884 | inode->i_ino); | |
1885 | brelse(bh); | |
ac27a0ec DK |
1886 | return 1; |
1887 | } | |
a72d7f83 JK |
1888 | offset = ext4_rec_len_from_disk(de->rec_len) + |
1889 | ext4_rec_len_from_disk(de1->rec_len); | |
1890 | de = ext4_next_entry(de1); | |
af5bc92d | 1891 | while (offset < inode->i_size) { |
ac27a0ec DK |
1892 | if (!bh || |
1893 | (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { | |
1894 | err = 0; | |
af5bc92d TT |
1895 | brelse(bh); |
1896 | bh = ext4_bread(NULL, inode, | |
617ba13b | 1897 | offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err); |
ac27a0ec DK |
1898 | if (!bh) { |
1899 | if (err) | |
46e665e9 | 1900 | ext4_error(sb, __func__, |
ac27a0ec DK |
1901 | "error %d reading directory" |
1902 | " #%lu offset %lu", | |
1903 | err, inode->i_ino, offset); | |
1904 | offset += sb->s_blocksize; | |
1905 | continue; | |
1906 | } | |
617ba13b | 1907 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 1908 | } |
617ba13b MC |
1909 | if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) { |
1910 | de = (struct ext4_dir_entry_2 *)(bh->b_data + | |
ac27a0ec DK |
1911 | sb->s_blocksize); |
1912 | offset = (offset | (sb->s_blocksize - 1)) + 1; | |
1913 | continue; | |
1914 | } | |
1915 | if (le32_to_cpu(de->inode)) { | |
af5bc92d | 1916 | brelse(bh); |
ac27a0ec DK |
1917 | return 0; |
1918 | } | |
a72d7f83 JK |
1919 | offset += ext4_rec_len_from_disk(de->rec_len); |
1920 | de = ext4_next_entry(de); | |
ac27a0ec | 1921 | } |
af5bc92d | 1922 | brelse(bh); |
ac27a0ec DK |
1923 | return 1; |
1924 | } | |
1925 | ||
617ba13b | 1926 | /* ext4_orphan_add() links an unlinked or truncated inode into a list of |
ac27a0ec DK |
1927 | * such inodes, starting at the superblock, in case we crash before the |
1928 | * file is closed/deleted, or in case the inode truncate spans multiple | |
1929 | * transactions and the last transaction is not recovered after a crash. | |
1930 | * | |
1931 | * At filesystem recovery time, we walk this list deleting unlinked | |
617ba13b | 1932 | * inodes and truncating linked inodes in ext4_orphan_cleanup(). |
ac27a0ec | 1933 | */ |
617ba13b | 1934 | int ext4_orphan_add(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
1935 | { |
1936 | struct super_block *sb = inode->i_sb; | |
617ba13b | 1937 | struct ext4_iloc iloc; |
ac27a0ec DK |
1938 | int err = 0, rc; |
1939 | ||
1940 | lock_super(sb); | |
617ba13b | 1941 | if (!list_empty(&EXT4_I(inode)->i_orphan)) |
ac27a0ec DK |
1942 | goto out_unlock; |
1943 | ||
1944 | /* Orphan handling is only valid for files with data blocks | |
1945 | * being truncated, or files being unlinked. */ | |
1946 | ||
1947 | /* @@@ FIXME: Observation from aviro: | |
617ba13b MC |
1948 | * I think I can trigger J_ASSERT in ext4_orphan_add(). We block |
1949 | * here (on lock_super()), so race with ext4_link() which might bump | |
ac27a0ec DK |
1950 | * ->i_nlink. For, say it, character device. Not a regular file, |
1951 | * not a directory, not a symlink and ->i_nlink > 0. | |
1952 | */ | |
af5bc92d TT |
1953 | J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
1954 | S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); | |
ac27a0ec | 1955 | |
617ba13b MC |
1956 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); |
1957 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); | |
ac27a0ec DK |
1958 | if (err) |
1959 | goto out_unlock; | |
1960 | ||
617ba13b | 1961 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
ac27a0ec DK |
1962 | if (err) |
1963 | goto out_unlock; | |
1964 | ||
1965 | /* Insert this inode at the head of the on-disk orphan list... */ | |
617ba13b MC |
1966 | NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan); |
1967 | EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); | |
1968 | err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh); | |
1969 | rc = ext4_mark_iloc_dirty(handle, inode, &iloc); | |
ac27a0ec DK |
1970 | if (!err) |
1971 | err = rc; | |
1972 | ||
1973 | /* Only add to the head of the in-memory list if all the | |
1974 | * previous operations succeeded. If the orphan_add is going to | |
1975 | * fail (possibly taking the journal offline), we can't risk | |
1976 | * leaving the inode on the orphan list: stray orphan-list | |
1977 | * entries can cause panics at unmount time. | |
1978 | * | |
1979 | * This is safe: on error we're going to ignore the orphan list | |
1980 | * anyway on the next recovery. */ | |
1981 | if (!err) | |
617ba13b | 1982 | list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); |
ac27a0ec DK |
1983 | |
1984 | jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); | |
1985 | jbd_debug(4, "orphan inode %lu will point to %d\n", | |
1986 | inode->i_ino, NEXT_ORPHAN(inode)); | |
1987 | out_unlock: | |
1988 | unlock_super(sb); | |
617ba13b | 1989 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
1990 | return err; |
1991 | } | |
1992 | ||
1993 | /* | |
617ba13b | 1994 | * ext4_orphan_del() removes an unlinked or truncated inode from the list |
ac27a0ec DK |
1995 | * of such inodes stored on disk, because it is finally being cleaned up. |
1996 | */ | |
617ba13b | 1997 | int ext4_orphan_del(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
1998 | { |
1999 | struct list_head *prev; | |
617ba13b MC |
2000 | struct ext4_inode_info *ei = EXT4_I(inode); |
2001 | struct ext4_sb_info *sbi; | |
ac27a0ec | 2002 | unsigned long ino_next; |
617ba13b | 2003 | struct ext4_iloc iloc; |
ac27a0ec DK |
2004 | int err = 0; |
2005 | ||
2006 | lock_super(inode->i_sb); | |
2007 | if (list_empty(&ei->i_orphan)) { | |
2008 | unlock_super(inode->i_sb); | |
2009 | return 0; | |
2010 | } | |
2011 | ||
2012 | ino_next = NEXT_ORPHAN(inode); | |
2013 | prev = ei->i_orphan.prev; | |
617ba13b | 2014 | sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
2015 | |
2016 | jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); | |
2017 | ||
2018 | list_del_init(&ei->i_orphan); | |
2019 | ||
2020 | /* If we're on an error path, we may not have a valid | |
2021 | * transaction handle with which to update the orphan list on | |
2022 | * disk, but we still need to remove the inode from the linked | |
2023 | * list in memory. */ | |
2024 | if (!handle) | |
2025 | goto out; | |
2026 | ||
617ba13b | 2027 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
ac27a0ec DK |
2028 | if (err) |
2029 | goto out_err; | |
2030 | ||
2031 | if (prev == &sbi->s_orphan) { | |
2032 | jbd_debug(4, "superblock will point to %lu\n", ino_next); | |
2033 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); | |
617ba13b | 2034 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
ac27a0ec DK |
2035 | if (err) |
2036 | goto out_brelse; | |
2037 | sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); | |
617ba13b | 2038 | err = ext4_journal_dirty_metadata(handle, sbi->s_sbh); |
ac27a0ec | 2039 | } else { |
617ba13b | 2040 | struct ext4_iloc iloc2; |
ac27a0ec | 2041 | struct inode *i_prev = |
617ba13b | 2042 | &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; |
ac27a0ec DK |
2043 | |
2044 | jbd_debug(4, "orphan inode %lu will point to %lu\n", | |
2045 | i_prev->i_ino, ino_next); | |
617ba13b | 2046 | err = ext4_reserve_inode_write(handle, i_prev, &iloc2); |
ac27a0ec DK |
2047 | if (err) |
2048 | goto out_brelse; | |
2049 | NEXT_ORPHAN(i_prev) = ino_next; | |
617ba13b | 2050 | err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); |
ac27a0ec DK |
2051 | } |
2052 | if (err) | |
2053 | goto out_brelse; | |
2054 | NEXT_ORPHAN(inode) = 0; | |
617ba13b | 2055 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
2056 | |
2057 | out_err: | |
617ba13b | 2058 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
2059 | out: |
2060 | unlock_super(inode->i_sb); | |
2061 | return err; | |
2062 | ||
2063 | out_brelse: | |
2064 | brelse(iloc.bh); | |
2065 | goto out_err; | |
2066 | } | |
2067 | ||
af5bc92d | 2068 | static int ext4_rmdir(struct inode *dir, struct dentry *dentry) |
ac27a0ec DK |
2069 | { |
2070 | int retval; | |
af5bc92d TT |
2071 | struct inode *inode; |
2072 | struct buffer_head *bh; | |
2073 | struct ext4_dir_entry_2 *de; | |
ac27a0ec DK |
2074 | handle_t *handle; |
2075 | ||
2076 | /* Initialize quotas before so that eventual writes go in | |
2077 | * separate transaction */ | |
2078 | DQUOT_INIT(dentry->d_inode); | |
617ba13b | 2079 | handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb)); |
ac27a0ec DK |
2080 | if (IS_ERR(handle)) |
2081 | return PTR_ERR(handle); | |
2082 | ||
2083 | retval = -ENOENT; | |
f702ba0f | 2084 | bh = ext4_find_entry(dir, &dentry->d_name, &de); |
ac27a0ec DK |
2085 | if (!bh) |
2086 | goto end_rmdir; | |
2087 | ||
2088 | if (IS_DIRSYNC(dir)) | |
2089 | handle->h_sync = 1; | |
2090 | ||
2091 | inode = dentry->d_inode; | |
2092 | ||
2093 | retval = -EIO; | |
2094 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2095 | goto end_rmdir; | |
2096 | ||
2097 | retval = -ENOTEMPTY; | |
af5bc92d | 2098 | if (!empty_dir(inode)) |
ac27a0ec DK |
2099 | goto end_rmdir; |
2100 | ||
617ba13b | 2101 | retval = ext4_delete_entry(handle, dir, de, bh); |
ac27a0ec DK |
2102 | if (retval) |
2103 | goto end_rmdir; | |
f8628a14 | 2104 | if (!EXT4_DIR_LINK_EMPTY(inode)) |
af5bc92d TT |
2105 | ext4_warning(inode->i_sb, "ext4_rmdir", |
2106 | "empty directory has too many links (%d)", | |
2107 | inode->i_nlink); | |
ac27a0ec DK |
2108 | inode->i_version++; |
2109 | clear_nlink(inode); | |
2110 | /* There's no need to set i_disksize: the fact that i_nlink is | |
2111 | * zero will ensure that the right thing happens during any | |
2112 | * recovery. */ | |
2113 | inode->i_size = 0; | |
617ba13b | 2114 | ext4_orphan_add(handle, inode); |
ef7f3835 | 2115 | inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode); |
617ba13b | 2116 | ext4_mark_inode_dirty(handle, inode); |
f8628a14 | 2117 | ext4_dec_count(handle, dir); |
617ba13b MC |
2118 | ext4_update_dx_flag(dir); |
2119 | ext4_mark_inode_dirty(handle, dir); | |
ac27a0ec DK |
2120 | |
2121 | end_rmdir: | |
617ba13b | 2122 | ext4_journal_stop(handle); |
af5bc92d | 2123 | brelse(bh); |
ac27a0ec DK |
2124 | return retval; |
2125 | } | |
2126 | ||
af5bc92d | 2127 | static int ext4_unlink(struct inode *dir, struct dentry *dentry) |
ac27a0ec DK |
2128 | { |
2129 | int retval; | |
af5bc92d TT |
2130 | struct inode *inode; |
2131 | struct buffer_head *bh; | |
2132 | struct ext4_dir_entry_2 *de; | |
ac27a0ec DK |
2133 | handle_t *handle; |
2134 | ||
2135 | /* Initialize quotas before so that eventual writes go | |
2136 | * in separate transaction */ | |
2137 | DQUOT_INIT(dentry->d_inode); | |
617ba13b | 2138 | handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb)); |
ac27a0ec DK |
2139 | if (IS_ERR(handle)) |
2140 | return PTR_ERR(handle); | |
2141 | ||
2142 | if (IS_DIRSYNC(dir)) | |
2143 | handle->h_sync = 1; | |
2144 | ||
2145 | retval = -ENOENT; | |
f702ba0f | 2146 | bh = ext4_find_entry(dir, &dentry->d_name, &de); |
ac27a0ec DK |
2147 | if (!bh) |
2148 | goto end_unlink; | |
2149 | ||
2150 | inode = dentry->d_inode; | |
2151 | ||
2152 | retval = -EIO; | |
2153 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2154 | goto end_unlink; | |
2155 | ||
2156 | if (!inode->i_nlink) { | |
af5bc92d TT |
2157 | ext4_warning(inode->i_sb, "ext4_unlink", |
2158 | "Deleting nonexistent file (%lu), %d", | |
2159 | inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
2160 | inode->i_nlink = 1; |
2161 | } | |
617ba13b | 2162 | retval = ext4_delete_entry(handle, dir, de, bh); |
ac27a0ec DK |
2163 | if (retval) |
2164 | goto end_unlink; | |
ef7f3835 | 2165 | dir->i_ctime = dir->i_mtime = ext4_current_time(dir); |
617ba13b MC |
2166 | ext4_update_dx_flag(dir); |
2167 | ext4_mark_inode_dirty(handle, dir); | |
825f1481 | 2168 | drop_nlink(inode); |
ac27a0ec | 2169 | if (!inode->i_nlink) |
617ba13b | 2170 | ext4_orphan_add(handle, inode); |
ef7f3835 | 2171 | inode->i_ctime = ext4_current_time(inode); |
617ba13b | 2172 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
2173 | retval = 0; |
2174 | ||
2175 | end_unlink: | |
617ba13b | 2176 | ext4_journal_stop(handle); |
af5bc92d | 2177 | brelse(bh); |
ac27a0ec DK |
2178 | return retval; |
2179 | } | |
2180 | ||
af5bc92d TT |
2181 | static int ext4_symlink(struct inode *dir, |
2182 | struct dentry *dentry, const char *symname) | |
ac27a0ec DK |
2183 | { |
2184 | handle_t *handle; | |
af5bc92d | 2185 | struct inode *inode; |
ac27a0ec DK |
2186 | int l, err, retries = 0; |
2187 | ||
2188 | l = strlen(symname)+1; | |
2189 | if (l > dir->i_sb->s_blocksize) | |
2190 | return -ENAMETOOLONG; | |
2191 | ||
2192 | retry: | |
617ba13b MC |
2193 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2194 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 + | |
2195 | 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb)); | |
ac27a0ec DK |
2196 | if (IS_ERR(handle)) |
2197 | return PTR_ERR(handle); | |
2198 | ||
2199 | if (IS_DIRSYNC(dir)) | |
2200 | handle->h_sync = 1; | |
2201 | ||
af5bc92d | 2202 | inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO); |
ac27a0ec DK |
2203 | err = PTR_ERR(inode); |
2204 | if (IS_ERR(inode)) | |
2205 | goto out_stop; | |
2206 | ||
af5bc92d | 2207 | if (l > sizeof(EXT4_I(inode)->i_data)) { |
617ba13b MC |
2208 | inode->i_op = &ext4_symlink_inode_operations; |
2209 | ext4_set_aops(inode); | |
ac27a0ec | 2210 | /* |
617ba13b | 2211 | * page_symlink() calls into ext4_prepare/commit_write. |
ac27a0ec DK |
2212 | * We have a transaction open. All is sweetness. It also sets |
2213 | * i_size in generic_commit_write(). | |
2214 | */ | |
2215 | err = __page_symlink(inode, symname, l, | |
2216 | mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS); | |
2217 | if (err) { | |
825f1481 | 2218 | clear_nlink(inode); |
6b38e842 | 2219 | unlock_new_inode(inode); |
617ba13b | 2220 | ext4_mark_inode_dirty(handle, inode); |
af5bc92d | 2221 | iput(inode); |
ac27a0ec DK |
2222 | goto out_stop; |
2223 | } | |
2224 | } else { | |
e65187e6 AK |
2225 | /* clear the extent format for fast symlink */ |
2226 | EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL; | |
617ba13b | 2227 | inode->i_op = &ext4_fast_symlink_inode_operations; |
af5bc92d | 2228 | memcpy((char *)&EXT4_I(inode)->i_data, symname, l); |
ac27a0ec DK |
2229 | inode->i_size = l-1; |
2230 | } | |
617ba13b MC |
2231 | EXT4_I(inode)->i_disksize = inode->i_size; |
2232 | err = ext4_add_nondir(handle, dentry, inode); | |
ac27a0ec | 2233 | out_stop: |
617ba13b MC |
2234 | ext4_journal_stop(handle); |
2235 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
2236 | goto retry; |
2237 | return err; | |
2238 | } | |
2239 | ||
af5bc92d TT |
2240 | static int ext4_link(struct dentry *old_dentry, |
2241 | struct inode *dir, struct dentry *dentry) | |
ac27a0ec DK |
2242 | { |
2243 | handle_t *handle; | |
2244 | struct inode *inode = old_dentry->d_inode; | |
2245 | int err, retries = 0; | |
2246 | ||
f8628a14 | 2247 | if (EXT4_DIR_LINK_MAX(inode)) |
ac27a0ec | 2248 | return -EMLINK; |
f8628a14 | 2249 | |
2988a774 ES |
2250 | /* |
2251 | * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing | |
2252 | * otherwise has the potential to corrupt the orphan inode list. | |
2253 | */ | |
2254 | if (inode->i_nlink == 0) | |
2255 | return -ENOENT; | |
ac27a0ec DK |
2256 | |
2257 | retry: | |
617ba13b MC |
2258 | handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2259 | EXT4_INDEX_EXTRA_TRANS_BLOCKS); | |
ac27a0ec DK |
2260 | if (IS_ERR(handle)) |
2261 | return PTR_ERR(handle); | |
2262 | ||
2263 | if (IS_DIRSYNC(dir)) | |
2264 | handle->h_sync = 1; | |
2265 | ||
ef7f3835 | 2266 | inode->i_ctime = ext4_current_time(inode); |
f8628a14 | 2267 | ext4_inc_count(handle, inode); |
ac27a0ec DK |
2268 | atomic_inc(&inode->i_count); |
2269 | ||
6b38e842 AV |
2270 | err = ext4_add_entry(handle, dentry, inode); |
2271 | if (!err) { | |
2272 | ext4_mark_inode_dirty(handle, inode); | |
2273 | d_instantiate(dentry, inode); | |
2274 | } else { | |
2275 | drop_nlink(inode); | |
2276 | iput(inode); | |
2277 | } | |
617ba13b MC |
2278 | ext4_journal_stop(handle); |
2279 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
2280 | goto retry; |
2281 | return err; | |
2282 | } | |
2283 | ||
2284 | #define PARENT_INO(buffer) \ | |
a72d7f83 | 2285 | (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode) |
ac27a0ec DK |
2286 | |
2287 | /* | |
2288 | * Anybody can rename anything with this: the permission checks are left to the | |
2289 | * higher-level routines. | |
2290 | */ | |
af5bc92d TT |
2291 | static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry, |
2292 | struct inode *new_dir, struct dentry *new_dentry) | |
ac27a0ec DK |
2293 | { |
2294 | handle_t *handle; | |
af5bc92d TT |
2295 | struct inode *old_inode, *new_inode; |
2296 | struct buffer_head *old_bh, *new_bh, *dir_bh; | |
2297 | struct ext4_dir_entry_2 *old_de, *new_de; | |
ac27a0ec DK |
2298 | int retval; |
2299 | ||
2300 | old_bh = new_bh = dir_bh = NULL; | |
2301 | ||
2302 | /* Initialize quotas before so that eventual writes go | |
2303 | * in separate transaction */ | |
2304 | if (new_dentry->d_inode) | |
2305 | DQUOT_INIT(new_dentry->d_inode); | |
617ba13b MC |
2306 | handle = ext4_journal_start(old_dir, 2 * |
2307 | EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) + | |
2308 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2); | |
ac27a0ec DK |
2309 | if (IS_ERR(handle)) |
2310 | return PTR_ERR(handle); | |
2311 | ||
2312 | if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir)) | |
2313 | handle->h_sync = 1; | |
2314 | ||
f702ba0f | 2315 | old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de); |
ac27a0ec DK |
2316 | /* |
2317 | * Check for inode number is _not_ due to possible IO errors. | |
2318 | * We might rmdir the source, keep it as pwd of some process | |
2319 | * and merrily kill the link to whatever was created under the | |
2320 | * same name. Goodbye sticky bit ;-< | |
2321 | */ | |
2322 | old_inode = old_dentry->d_inode; | |
2323 | retval = -ENOENT; | |
2324 | if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino) | |
2325 | goto end_rename; | |
2326 | ||
2327 | new_inode = new_dentry->d_inode; | |
f702ba0f | 2328 | new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de); |
ac27a0ec DK |
2329 | if (new_bh) { |
2330 | if (!new_inode) { | |
af5bc92d | 2331 | brelse(new_bh); |
ac27a0ec DK |
2332 | new_bh = NULL; |
2333 | } | |
2334 | } | |
2335 | if (S_ISDIR(old_inode->i_mode)) { | |
2336 | if (new_inode) { | |
2337 | retval = -ENOTEMPTY; | |
af5bc92d | 2338 | if (!empty_dir(new_inode)) |
ac27a0ec DK |
2339 | goto end_rename; |
2340 | } | |
2341 | retval = -EIO; | |
af5bc92d | 2342 | dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval); |
ac27a0ec DK |
2343 | if (!dir_bh) |
2344 | goto end_rename; | |
2345 | if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino) | |
2346 | goto end_rename; | |
2347 | retval = -EMLINK; | |
af5bc92d | 2348 | if (!new_inode && new_dir != old_dir && |
617ba13b | 2349 | new_dir->i_nlink >= EXT4_LINK_MAX) |
ac27a0ec DK |
2350 | goto end_rename; |
2351 | } | |
2352 | if (!new_bh) { | |
af5bc92d | 2353 | retval = ext4_add_entry(handle, new_dentry, old_inode); |
ac27a0ec DK |
2354 | if (retval) |
2355 | goto end_rename; | |
2356 | } else { | |
2357 | BUFFER_TRACE(new_bh, "get write access"); | |
617ba13b | 2358 | ext4_journal_get_write_access(handle, new_bh); |
ac27a0ec | 2359 | new_de->inode = cpu_to_le32(old_inode->i_ino); |
617ba13b MC |
2360 | if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb, |
2361 | EXT4_FEATURE_INCOMPAT_FILETYPE)) | |
ac27a0ec DK |
2362 | new_de->file_type = old_de->file_type; |
2363 | new_dir->i_version++; | |
53b7e9f6 JK |
2364 | new_dir->i_ctime = new_dir->i_mtime = |
2365 | ext4_current_time(new_dir); | |
2366 | ext4_mark_inode_dirty(handle, new_dir); | |
617ba13b MC |
2367 | BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata"); |
2368 | ext4_journal_dirty_metadata(handle, new_bh); | |
ac27a0ec DK |
2369 | brelse(new_bh); |
2370 | new_bh = NULL; | |
2371 | } | |
2372 | ||
2373 | /* | |
2374 | * Like most other Unix systems, set the ctime for inodes on a | |
2375 | * rename. | |
2376 | */ | |
ef7f3835 | 2377 | old_inode->i_ctime = ext4_current_time(old_inode); |
617ba13b | 2378 | ext4_mark_inode_dirty(handle, old_inode); |
ac27a0ec DK |
2379 | |
2380 | /* | |
2381 | * ok, that's it | |
2382 | */ | |
2383 | if (le32_to_cpu(old_de->inode) != old_inode->i_ino || | |
2384 | old_de->name_len != old_dentry->d_name.len || | |
2385 | strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) || | |
617ba13b | 2386 | (retval = ext4_delete_entry(handle, old_dir, |
ac27a0ec DK |
2387 | old_de, old_bh)) == -ENOENT) { |
2388 | /* old_de could have moved from under us during htree split, so | |
2389 | * make sure that we are deleting the right entry. We might | |
2390 | * also be pointing to a stale entry in the unused part of | |
2391 | * old_bh so just checking inum and the name isn't enough. */ | |
2392 | struct buffer_head *old_bh2; | |
617ba13b | 2393 | struct ext4_dir_entry_2 *old_de2; |
ac27a0ec | 2394 | |
f702ba0f | 2395 | old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2); |
ac27a0ec | 2396 | if (old_bh2) { |
617ba13b | 2397 | retval = ext4_delete_entry(handle, old_dir, |
ac27a0ec DK |
2398 | old_de2, old_bh2); |
2399 | brelse(old_bh2); | |
2400 | } | |
2401 | } | |
2402 | if (retval) { | |
617ba13b | 2403 | ext4_warning(old_dir->i_sb, "ext4_rename", |
ac27a0ec DK |
2404 | "Deleting old file (%lu), %d, error=%d", |
2405 | old_dir->i_ino, old_dir->i_nlink, retval); | |
2406 | } | |
2407 | ||
2408 | if (new_inode) { | |
f8628a14 | 2409 | ext4_dec_count(handle, new_inode); |
ef7f3835 | 2410 | new_inode->i_ctime = ext4_current_time(new_inode); |
ac27a0ec | 2411 | } |
ef7f3835 | 2412 | old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir); |
617ba13b | 2413 | ext4_update_dx_flag(old_dir); |
ac27a0ec DK |
2414 | if (dir_bh) { |
2415 | BUFFER_TRACE(dir_bh, "get_write_access"); | |
617ba13b | 2416 | ext4_journal_get_write_access(handle, dir_bh); |
ac27a0ec | 2417 | PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino); |
617ba13b MC |
2418 | BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata"); |
2419 | ext4_journal_dirty_metadata(handle, dir_bh); | |
f8628a14 | 2420 | ext4_dec_count(handle, old_dir); |
ac27a0ec | 2421 | if (new_inode) { |
f8628a14 | 2422 | /* checked empty_dir above, can't have another parent, |
825f1481 | 2423 | * ext4_dec_count() won't work for many-linked dirs */ |
f8628a14 | 2424 | new_inode->i_nlink = 0; |
ac27a0ec | 2425 | } else { |
f8628a14 | 2426 | ext4_inc_count(handle, new_dir); |
617ba13b MC |
2427 | ext4_update_dx_flag(new_dir); |
2428 | ext4_mark_inode_dirty(handle, new_dir); | |
ac27a0ec DK |
2429 | } |
2430 | } | |
617ba13b | 2431 | ext4_mark_inode_dirty(handle, old_dir); |
ac27a0ec | 2432 | if (new_inode) { |
617ba13b | 2433 | ext4_mark_inode_dirty(handle, new_inode); |
ac27a0ec | 2434 | if (!new_inode->i_nlink) |
617ba13b | 2435 | ext4_orphan_add(handle, new_inode); |
ac27a0ec DK |
2436 | } |
2437 | retval = 0; | |
2438 | ||
2439 | end_rename: | |
af5bc92d TT |
2440 | brelse(dir_bh); |
2441 | brelse(old_bh); | |
2442 | brelse(new_bh); | |
617ba13b | 2443 | ext4_journal_stop(handle); |
ac27a0ec DK |
2444 | return retval; |
2445 | } | |
2446 | ||
2447 | /* | |
2448 | * directories can handle most operations... | |
2449 | */ | |
754661f1 | 2450 | const struct inode_operations ext4_dir_inode_operations = { |
617ba13b MC |
2451 | .create = ext4_create, |
2452 | .lookup = ext4_lookup, | |
2453 | .link = ext4_link, | |
2454 | .unlink = ext4_unlink, | |
2455 | .symlink = ext4_symlink, | |
2456 | .mkdir = ext4_mkdir, | |
2457 | .rmdir = ext4_rmdir, | |
2458 | .mknod = ext4_mknod, | |
2459 | .rename = ext4_rename, | |
2460 | .setattr = ext4_setattr, | |
03010a33 | 2461 | #ifdef CONFIG_EXT4_FS_XATTR |
ac27a0ec DK |
2462 | .setxattr = generic_setxattr, |
2463 | .getxattr = generic_getxattr, | |
617ba13b | 2464 | .listxattr = ext4_listxattr, |
ac27a0ec DK |
2465 | .removexattr = generic_removexattr, |
2466 | #endif | |
617ba13b | 2467 | .permission = ext4_permission, |
ac27a0ec DK |
2468 | }; |
2469 | ||
754661f1 | 2470 | const struct inode_operations ext4_special_inode_operations = { |
617ba13b | 2471 | .setattr = ext4_setattr, |
03010a33 | 2472 | #ifdef CONFIG_EXT4_FS_XATTR |
ac27a0ec DK |
2473 | .setxattr = generic_setxattr, |
2474 | .getxattr = generic_getxattr, | |
617ba13b | 2475 | .listxattr = ext4_listxattr, |
ac27a0ec DK |
2476 | .removexattr = generic_removexattr, |
2477 | #endif | |
617ba13b | 2478 | .permission = ext4_permission, |
ac27a0ec | 2479 | }; |