Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/namei.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/namei.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
15 | * Big-endian to little-endian byte-swapping/bitmaps by | |
16 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
17 | * Directory entry file type support and forward compatibility hooks | |
18 | * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998 | |
19 | * Hash Tree Directory indexing (c) | |
20 | * Daniel Phillips, 2001 | |
21 | * Hash Tree Directory indexing porting | |
22 | * Christopher Li, 2002 | |
23 | * Hash Tree Directory indexing cleanup | |
24 | * Theodore Ts'o, 2002 | |
25 | */ | |
26 | ||
27 | #include <linux/fs.h> | |
28 | #include <linux/pagemap.h> | |
dab291af | 29 | #include <linux/jbd2.h> |
ac27a0ec | 30 | #include <linux/time.h> |
ac27a0ec DK |
31 | #include <linux/fcntl.h> |
32 | #include <linux/stat.h> | |
33 | #include <linux/string.h> | |
34 | #include <linux/quotaops.h> | |
35 | #include <linux/buffer_head.h> | |
36 | #include <linux/bio.h> | |
3dcf5451 CH |
37 | #include "ext4.h" |
38 | #include "ext4_jbd2.h" | |
ac27a0ec | 39 | |
ac27a0ec DK |
40 | #include "xattr.h" |
41 | #include "acl.h" | |
42 | ||
0562e0ba | 43 | #include <trace/events/ext4.h> |
ac27a0ec DK |
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 | 50 | |
617ba13b | 51 | static struct buffer_head *ext4_append(handle_t *handle, |
ac27a0ec | 52 | struct inode *inode, |
0f70b406 | 53 | ext4_lblk_t *block) |
ac27a0ec DK |
54 | { |
55 | struct buffer_head *bh; | |
1c215028 | 56 | int err; |
ac27a0ec | 57 | |
df981d03 TT |
58 | if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb && |
59 | ((inode->i_size >> 10) >= | |
0f70b406 TT |
60 | EXT4_SB(inode->i_sb)->s_max_dir_size_kb))) |
61 | return ERR_PTR(-ENOSPC); | |
df981d03 | 62 | |
ac27a0ec DK |
63 | *block = inode->i_size >> inode->i_sb->s_blocksize_bits; |
64 | ||
1c215028 TT |
65 | bh = ext4_bread(handle, inode, *block, 1); |
66 | if (IS_ERR(bh)) | |
67 | return bh; | |
0f70b406 TT |
68 | inode->i_size += inode->i_sb->s_blocksize; |
69 | EXT4_I(inode)->i_disksize = inode->i_size; | |
5d601255 | 70 | BUFFER_TRACE(bh, "get_write_access"); |
0f70b406 TT |
71 | err = ext4_journal_get_write_access(handle, bh); |
72 | if (err) { | |
73 | brelse(bh); | |
74 | ext4_std_error(inode->i_sb, err); | |
75 | return ERR_PTR(err); | |
6d1ab10e | 76 | } |
ac27a0ec DK |
77 | return bh; |
78 | } | |
79 | ||
dc6982ff TT |
80 | static int ext4_dx_csum_verify(struct inode *inode, |
81 | struct ext4_dir_entry *dirent); | |
82 | ||
83 | typedef enum { | |
84 | EITHER, INDEX, DIRENT | |
85 | } dirblock_type_t; | |
86 | ||
87 | #define ext4_read_dirblock(inode, block, type) \ | |
88 | __ext4_read_dirblock((inode), (block), (type), __LINE__) | |
89 | ||
90 | static struct buffer_head *__ext4_read_dirblock(struct inode *inode, | |
91 | ext4_lblk_t block, | |
92 | dirblock_type_t type, | |
93 | unsigned int line) | |
94 | { | |
95 | struct buffer_head *bh; | |
96 | struct ext4_dir_entry *dirent; | |
1c215028 | 97 | int is_dx_block = 0; |
dc6982ff | 98 | |
1c215028 TT |
99 | bh = ext4_bread(NULL, inode, block, 0); |
100 | if (IS_ERR(bh)) { | |
dc6982ff | 101 | __ext4_warning(inode->i_sb, __func__, line, |
1c215028 TT |
102 | "error %ld reading directory block " |
103 | "(ino %lu, block %lu)", PTR_ERR(bh), inode->i_ino, | |
dc6982ff | 104 | (unsigned long) block); |
1c215028 TT |
105 | |
106 | return bh; | |
107 | } | |
108 | if (!bh) { | |
109 | ext4_error_inode(inode, __func__, line, block, "Directory hole found"); | |
110 | return ERR_PTR(-EIO); | |
dc6982ff TT |
111 | } |
112 | dirent = (struct ext4_dir_entry *) bh->b_data; | |
113 | /* Determine whether or not we have an index block */ | |
114 | if (is_dx(inode)) { | |
115 | if (block == 0) | |
116 | is_dx_block = 1; | |
117 | else if (ext4_rec_len_from_disk(dirent->rec_len, | |
118 | inode->i_sb->s_blocksize) == | |
119 | inode->i_sb->s_blocksize) | |
120 | is_dx_block = 1; | |
121 | } | |
122 | if (!is_dx_block && type == INDEX) { | |
123 | ext4_error_inode(inode, __func__, line, block, | |
124 | "directory leaf block found instead of index block"); | |
125 | return ERR_PTR(-EIO); | |
126 | } | |
127 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
128 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) || | |
129 | buffer_verified(bh)) | |
130 | return bh; | |
131 | ||
132 | /* | |
133 | * An empty leaf block can get mistaken for a index block; for | |
134 | * this reason, we can only check the index checksum when the | |
135 | * caller is sure it should be an index block. | |
136 | */ | |
137 | if (is_dx_block && type == INDEX) { | |
138 | if (ext4_dx_csum_verify(inode, dirent)) | |
139 | set_buffer_verified(bh); | |
140 | else { | |
141 | ext4_error_inode(inode, __func__, line, block, | |
142 | "Directory index failed checksum"); | |
a871611b | 143 | brelse(bh); |
dc6982ff | 144 | return ERR_PTR(-EIO); |
a871611b | 145 | } |
ac27a0ec | 146 | } |
dc6982ff TT |
147 | if (!is_dx_block) { |
148 | if (ext4_dirent_csum_verify(inode, dirent)) | |
149 | set_buffer_verified(bh); | |
150 | else { | |
151 | ext4_error_inode(inode, __func__, line, block, | |
152 | "Directory block failed checksum"); | |
153 | brelse(bh); | |
154 | return ERR_PTR(-EIO); | |
155 | } | |
6d1ab10e | 156 | } |
ac27a0ec DK |
157 | return bh; |
158 | } | |
159 | ||
160 | #ifndef assert | |
161 | #define assert(test) J_ASSERT(test) | |
162 | #endif | |
163 | ||
ac27a0ec DK |
164 | #ifdef DX_DEBUG |
165 | #define dxtrace(command) command | |
166 | #else | |
167 | #define dxtrace(command) | |
168 | #endif | |
169 | ||
170 | struct fake_dirent | |
171 | { | |
172 | __le32 inode; | |
173 | __le16 rec_len; | |
174 | u8 name_len; | |
175 | u8 file_type; | |
176 | }; | |
177 | ||
178 | struct dx_countlimit | |
179 | { | |
180 | __le16 limit; | |
181 | __le16 count; | |
182 | }; | |
183 | ||
184 | struct dx_entry | |
185 | { | |
186 | __le32 hash; | |
187 | __le32 block; | |
188 | }; | |
189 | ||
190 | /* | |
191 | * dx_root_info is laid out so that if it should somehow get overlaid by a | |
192 | * dirent the two low bits of the hash version will be zero. Therefore, the | |
193 | * hash version mod 4 should never be 0. Sincerely, the paranoia department. | |
194 | */ | |
195 | ||
196 | struct dx_root | |
197 | { | |
198 | struct fake_dirent dot; | |
199 | char dot_name[4]; | |
200 | struct fake_dirent dotdot; | |
201 | char dotdot_name[4]; | |
202 | struct dx_root_info | |
203 | { | |
204 | __le32 reserved_zero; | |
205 | u8 hash_version; | |
206 | u8 info_length; /* 8 */ | |
207 | u8 indirect_levels; | |
208 | u8 unused_flags; | |
209 | } | |
210 | info; | |
211 | struct dx_entry entries[0]; | |
212 | }; | |
213 | ||
214 | struct dx_node | |
215 | { | |
216 | struct fake_dirent fake; | |
217 | struct dx_entry entries[0]; | |
218 | }; | |
219 | ||
220 | ||
221 | struct dx_frame | |
222 | { | |
223 | struct buffer_head *bh; | |
224 | struct dx_entry *entries; | |
225 | struct dx_entry *at; | |
226 | }; | |
227 | ||
228 | struct dx_map_entry | |
229 | { | |
230 | u32 hash; | |
ef2b02d3 ES |
231 | u16 offs; |
232 | u16 size; | |
ac27a0ec DK |
233 | }; |
234 | ||
e6153918 DW |
235 | /* |
236 | * This goes at the end of each htree block. | |
237 | */ | |
238 | struct dx_tail { | |
239 | u32 dt_reserved; | |
240 | __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */ | |
241 | }; | |
242 | ||
725d26d3 AK |
243 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry); |
244 | static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value); | |
af5bc92d TT |
245 | static inline unsigned dx_get_hash(struct dx_entry *entry); |
246 | static void dx_set_hash(struct dx_entry *entry, unsigned value); | |
247 | static unsigned dx_get_count(struct dx_entry *entries); | |
248 | static unsigned dx_get_limit(struct dx_entry *entries); | |
249 | static void dx_set_count(struct dx_entry *entries, unsigned value); | |
250 | static void dx_set_limit(struct dx_entry *entries, unsigned value); | |
251 | static unsigned dx_root_limit(struct inode *dir, unsigned infosize); | |
252 | static unsigned dx_node_limit(struct inode *dir); | |
f702ba0f | 253 | static struct dx_frame *dx_probe(const struct qstr *d_name, |
ac27a0ec DK |
254 | struct inode *dir, |
255 | struct dx_hash_info *hinfo, | |
dd73b5d5 | 256 | struct dx_frame *frame); |
af5bc92d | 257 | static void dx_release(struct dx_frame *frames); |
8bad4597 | 258 | static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize, |
af5bc92d | 259 | struct dx_hash_info *hinfo, struct dx_map_entry map[]); |
ac27a0ec | 260 | static void dx_sort_map(struct dx_map_entry *map, unsigned count); |
af5bc92d | 261 | static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to, |
3d0518f4 | 262 | struct dx_map_entry *offsets, int count, unsigned blocksize); |
8bad4597 | 263 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize); |
725d26d3 AK |
264 | static void dx_insert_block(struct dx_frame *frame, |
265 | u32 hash, ext4_lblk_t block); | |
617ba13b | 266 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, |
ac27a0ec DK |
267 | struct dx_frame *frame, |
268 | struct dx_frame *frames, | |
269 | __u32 *start_hash); | |
f702ba0f TT |
270 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir, |
271 | const struct qstr *d_name, | |
537d8f93 | 272 | struct ext4_dir_entry_2 **res_dir); |
617ba13b | 273 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, |
ac27a0ec DK |
274 | struct inode *inode); |
275 | ||
dbe89444 | 276 | /* checksumming functions */ |
3c47d541 TM |
277 | void initialize_dirent_tail(struct ext4_dir_entry_tail *t, |
278 | unsigned int blocksize) | |
b0336e8d DW |
279 | { |
280 | memset(t, 0, sizeof(struct ext4_dir_entry_tail)); | |
281 | t->det_rec_len = ext4_rec_len_to_disk( | |
282 | sizeof(struct ext4_dir_entry_tail), blocksize); | |
283 | t->det_reserved_ft = EXT4_FT_DIR_CSUM; | |
284 | } | |
285 | ||
286 | /* Walk through a dirent block to find a checksum "dirent" at the tail */ | |
287 | static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode, | |
288 | struct ext4_dir_entry *de) | |
289 | { | |
290 | struct ext4_dir_entry_tail *t; | |
291 | ||
292 | #ifdef PARANOID | |
293 | struct ext4_dir_entry *d, *top; | |
294 | ||
295 | d = de; | |
296 | top = (struct ext4_dir_entry *)(((void *)de) + | |
297 | (EXT4_BLOCK_SIZE(inode->i_sb) - | |
298 | sizeof(struct ext4_dir_entry_tail))); | |
299 | while (d < top && d->rec_len) | |
300 | d = (struct ext4_dir_entry *)(((void *)d) + | |
301 | le16_to_cpu(d->rec_len)); | |
302 | ||
303 | if (d != top) | |
304 | return NULL; | |
305 | ||
306 | t = (struct ext4_dir_entry_tail *)d; | |
307 | #else | |
308 | t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb)); | |
309 | #endif | |
310 | ||
311 | if (t->det_reserved_zero1 || | |
312 | le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) || | |
313 | t->det_reserved_zero2 || | |
314 | t->det_reserved_ft != EXT4_FT_DIR_CSUM) | |
315 | return NULL; | |
316 | ||
317 | return t; | |
318 | } | |
319 | ||
320 | static __le32 ext4_dirent_csum(struct inode *inode, | |
321 | struct ext4_dir_entry *dirent, int size) | |
322 | { | |
323 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
324 | struct ext4_inode_info *ei = EXT4_I(inode); | |
325 | __u32 csum; | |
326 | ||
327 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); | |
328 | return cpu_to_le32(csum); | |
329 | } | |
330 | ||
dffe9d8d TT |
331 | static void warn_no_space_for_csum(struct inode *inode) |
332 | { | |
333 | ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for " | |
334 | "checksum. Please run e2fsck -D.", inode->i_ino); | |
335 | } | |
336 | ||
b0336e8d DW |
337 | int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent) |
338 | { | |
339 | struct ext4_dir_entry_tail *t; | |
340 | ||
341 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
342 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
343 | return 1; | |
344 | ||
345 | t = get_dirent_tail(inode, dirent); | |
346 | if (!t) { | |
dffe9d8d | 347 | warn_no_space_for_csum(inode); |
b0336e8d DW |
348 | return 0; |
349 | } | |
350 | ||
351 | if (t->det_checksum != ext4_dirent_csum(inode, dirent, | |
352 | (void *)t - (void *)dirent)) | |
353 | return 0; | |
354 | ||
355 | return 1; | |
356 | } | |
357 | ||
358 | static void ext4_dirent_csum_set(struct inode *inode, | |
359 | struct ext4_dir_entry *dirent) | |
360 | { | |
361 | struct ext4_dir_entry_tail *t; | |
362 | ||
363 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
364 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
365 | return; | |
366 | ||
367 | t = get_dirent_tail(inode, dirent); | |
368 | if (!t) { | |
dffe9d8d | 369 | warn_no_space_for_csum(inode); |
b0336e8d DW |
370 | return; |
371 | } | |
372 | ||
373 | t->det_checksum = ext4_dirent_csum(inode, dirent, | |
374 | (void *)t - (void *)dirent); | |
375 | } | |
376 | ||
3c47d541 TM |
377 | int ext4_handle_dirty_dirent_node(handle_t *handle, |
378 | struct inode *inode, | |
379 | struct buffer_head *bh) | |
b0336e8d DW |
380 | { |
381 | ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); | |
382 | return ext4_handle_dirty_metadata(handle, inode, bh); | |
383 | } | |
384 | ||
dbe89444 DW |
385 | static struct dx_countlimit *get_dx_countlimit(struct inode *inode, |
386 | struct ext4_dir_entry *dirent, | |
387 | int *offset) | |
388 | { | |
389 | struct ext4_dir_entry *dp; | |
390 | struct dx_root_info *root; | |
391 | int count_offset; | |
392 | ||
393 | if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb)) | |
394 | count_offset = 8; | |
395 | else if (le16_to_cpu(dirent->rec_len) == 12) { | |
396 | dp = (struct ext4_dir_entry *)(((void *)dirent) + 12); | |
397 | if (le16_to_cpu(dp->rec_len) != | |
398 | EXT4_BLOCK_SIZE(inode->i_sb) - 12) | |
399 | return NULL; | |
400 | root = (struct dx_root_info *)(((void *)dp + 12)); | |
401 | if (root->reserved_zero || | |
402 | root->info_length != sizeof(struct dx_root_info)) | |
403 | return NULL; | |
404 | count_offset = 32; | |
405 | } else | |
406 | return NULL; | |
407 | ||
408 | if (offset) | |
409 | *offset = count_offset; | |
410 | return (struct dx_countlimit *)(((void *)dirent) + count_offset); | |
411 | } | |
412 | ||
413 | static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent, | |
414 | int count_offset, int count, struct dx_tail *t) | |
415 | { | |
416 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
417 | struct ext4_inode_info *ei = EXT4_I(inode); | |
d6a77105 TT |
418 | __u32 csum; |
419 | __le32 save_csum; | |
dbe89444 DW |
420 | int size; |
421 | ||
422 | size = count_offset + (count * sizeof(struct dx_entry)); | |
d6a77105 | 423 | save_csum = t->dt_checksum; |
dbe89444 DW |
424 | t->dt_checksum = 0; |
425 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); | |
426 | csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail)); | |
d6a77105 | 427 | t->dt_checksum = save_csum; |
dbe89444 DW |
428 | |
429 | return cpu_to_le32(csum); | |
430 | } | |
431 | ||
432 | static int ext4_dx_csum_verify(struct inode *inode, | |
433 | struct ext4_dir_entry *dirent) | |
434 | { | |
435 | struct dx_countlimit *c; | |
436 | struct dx_tail *t; | |
437 | int count_offset, limit, count; | |
438 | ||
439 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
440 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
441 | return 1; | |
442 | ||
443 | c = get_dx_countlimit(inode, dirent, &count_offset); | |
444 | if (!c) { | |
445 | EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); | |
446 | return 1; | |
447 | } | |
448 | limit = le16_to_cpu(c->limit); | |
449 | count = le16_to_cpu(c->count); | |
450 | if (count_offset + (limit * sizeof(struct dx_entry)) > | |
451 | EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { | |
dffe9d8d | 452 | warn_no_space_for_csum(inode); |
dbe89444 DW |
453 | return 1; |
454 | } | |
455 | t = (struct dx_tail *)(((struct dx_entry *)c) + limit); | |
456 | ||
457 | if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset, | |
458 | count, t)) | |
459 | return 0; | |
460 | return 1; | |
461 | } | |
462 | ||
463 | static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent) | |
464 | { | |
465 | struct dx_countlimit *c; | |
466 | struct dx_tail *t; | |
467 | int count_offset, limit, count; | |
468 | ||
469 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
470 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
471 | return; | |
472 | ||
473 | c = get_dx_countlimit(inode, dirent, &count_offset); | |
474 | if (!c) { | |
475 | EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); | |
476 | return; | |
477 | } | |
478 | limit = le16_to_cpu(c->limit); | |
479 | count = le16_to_cpu(c->count); | |
480 | if (count_offset + (limit * sizeof(struct dx_entry)) > | |
481 | EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { | |
dffe9d8d | 482 | warn_no_space_for_csum(inode); |
dbe89444 DW |
483 | return; |
484 | } | |
485 | t = (struct dx_tail *)(((struct dx_entry *)c) + limit); | |
486 | ||
487 | t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t); | |
488 | } | |
489 | ||
490 | static inline int ext4_handle_dirty_dx_node(handle_t *handle, | |
491 | struct inode *inode, | |
492 | struct buffer_head *bh) | |
493 | { | |
494 | ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); | |
495 | return ext4_handle_dirty_metadata(handle, inode, bh); | |
496 | } | |
497 | ||
f795e140 LZ |
498 | /* |
499 | * p is at least 6 bytes before the end of page | |
500 | */ | |
501 | static inline struct ext4_dir_entry_2 * | |
3d0518f4 | 502 | ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) |
f795e140 LZ |
503 | { |
504 | return (struct ext4_dir_entry_2 *)((char *)p + | |
3d0518f4 | 505 | ext4_rec_len_from_disk(p->rec_len, blocksize)); |
f795e140 LZ |
506 | } |
507 | ||
ac27a0ec DK |
508 | /* |
509 | * Future: use high four bits of block for coalesce-on-delete flags | |
510 | * Mask them off for now. | |
511 | */ | |
512 | ||
725d26d3 | 513 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) |
ac27a0ec DK |
514 | { |
515 | return le32_to_cpu(entry->block) & 0x00ffffff; | |
516 | } | |
517 | ||
725d26d3 | 518 | static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) |
ac27a0ec DK |
519 | { |
520 | entry->block = cpu_to_le32(value); | |
521 | } | |
522 | ||
af5bc92d | 523 | static inline unsigned dx_get_hash(struct dx_entry *entry) |
ac27a0ec DK |
524 | { |
525 | return le32_to_cpu(entry->hash); | |
526 | } | |
527 | ||
af5bc92d | 528 | static inline void dx_set_hash(struct dx_entry *entry, unsigned value) |
ac27a0ec DK |
529 | { |
530 | entry->hash = cpu_to_le32(value); | |
531 | } | |
532 | ||
af5bc92d | 533 | static inline unsigned dx_get_count(struct dx_entry *entries) |
ac27a0ec DK |
534 | { |
535 | return le16_to_cpu(((struct dx_countlimit *) entries)->count); | |
536 | } | |
537 | ||
af5bc92d | 538 | static inline unsigned dx_get_limit(struct dx_entry *entries) |
ac27a0ec DK |
539 | { |
540 | return le16_to_cpu(((struct dx_countlimit *) entries)->limit); | |
541 | } | |
542 | ||
af5bc92d | 543 | static inline void dx_set_count(struct dx_entry *entries, unsigned value) |
ac27a0ec DK |
544 | { |
545 | ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); | |
546 | } | |
547 | ||
af5bc92d | 548 | static inline void dx_set_limit(struct dx_entry *entries, unsigned value) |
ac27a0ec DK |
549 | { |
550 | ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); | |
551 | } | |
552 | ||
af5bc92d | 553 | static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) |
ac27a0ec | 554 | { |
617ba13b MC |
555 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - |
556 | EXT4_DIR_REC_LEN(2) - infosize; | |
dbe89444 DW |
557 | |
558 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, | |
559 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
560 | entry_space -= sizeof(struct dx_tail); | |
d9c769b7 | 561 | return entry_space / sizeof(struct dx_entry); |
ac27a0ec DK |
562 | } |
563 | ||
af5bc92d | 564 | static inline unsigned dx_node_limit(struct inode *dir) |
ac27a0ec | 565 | { |
617ba13b | 566 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); |
dbe89444 DW |
567 | |
568 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, | |
569 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
570 | entry_space -= sizeof(struct dx_tail); | |
d9c769b7 | 571 | return entry_space / sizeof(struct dx_entry); |
ac27a0ec DK |
572 | } |
573 | ||
574 | /* | |
575 | * Debug | |
576 | */ | |
577 | #ifdef DX_DEBUG | |
4776004f | 578 | static void dx_show_index(char * label, struct dx_entry *entries) |
ac27a0ec | 579 | { |
63f57933 | 580 | int i, n = dx_get_count (entries); |
4776004f | 581 | printk(KERN_DEBUG "%s index ", label); |
63f57933 | 582 | for (i = 0; i < n; i++) { |
4776004f | 583 | printk("%x->%lu ", i ? dx_get_hash(entries + i) : |
725d26d3 | 584 | 0, (unsigned long)dx_get_block(entries + i)); |
63f57933 AM |
585 | } |
586 | printk("\n"); | |
ac27a0ec DK |
587 | } |
588 | ||
589 | struct stats | |
590 | { | |
591 | unsigned names; | |
592 | unsigned space; | |
593 | unsigned bcount; | |
594 | }; | |
595 | ||
617ba13b | 596 | static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de, |
ac27a0ec DK |
597 | int size, int show_names) |
598 | { | |
599 | unsigned names = 0, space = 0; | |
600 | char *base = (char *) de; | |
601 | struct dx_hash_info h = *hinfo; | |
602 | ||
603 | printk("names: "); | |
604 | while ((char *) de < base + size) | |
605 | { | |
606 | if (de->inode) | |
607 | { | |
608 | if (show_names) | |
609 | { | |
610 | int len = de->name_len; | |
611 | char *name = de->name; | |
612 | while (len--) printk("%c", *name++); | |
617ba13b | 613 | ext4fs_dirhash(de->name, de->name_len, &h); |
ac27a0ec | 614 | printk(":%x.%u ", h.hash, |
265c6a0f | 615 | (unsigned) ((char *) de - base)); |
ac27a0ec | 616 | } |
617ba13b | 617 | space += EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec DK |
618 | names++; |
619 | } | |
3d0518f4 | 620 | de = ext4_next_entry(de, size); |
ac27a0ec DK |
621 | } |
622 | printk("(%i)\n", names); | |
623 | return (struct stats) { names, space, 1 }; | |
624 | } | |
625 | ||
626 | struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, | |
627 | struct dx_entry *entries, int levels) | |
628 | { | |
629 | unsigned blocksize = dir->i_sb->s_blocksize; | |
af5bc92d | 630 | unsigned count = dx_get_count(entries), names = 0, space = 0, i; |
ac27a0ec DK |
631 | unsigned bcount = 0; |
632 | struct buffer_head *bh; | |
633 | int err; | |
634 | printk("%i indexed blocks...\n", count); | |
635 | for (i = 0; i < count; i++, entries++) | |
636 | { | |
725d26d3 AK |
637 | ext4_lblk_t block = dx_get_block(entries); |
638 | ext4_lblk_t hash = i ? dx_get_hash(entries): 0; | |
ac27a0ec DK |
639 | u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; |
640 | struct stats stats; | |
641 | printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); | |
1c215028 TT |
642 | bh = ext4_bread(NULL,dir, block, 0); |
643 | if (!bh || IS_ERR(bh)) | |
644 | continue; | |
ac27a0ec DK |
645 | stats = levels? |
646 | dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): | |
617ba13b | 647 | dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0); |
ac27a0ec DK |
648 | names += stats.names; |
649 | space += stats.space; | |
650 | bcount += stats.bcount; | |
af5bc92d | 651 | brelse(bh); |
ac27a0ec DK |
652 | } |
653 | if (bcount) | |
60e6679e | 654 | printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", |
4776004f TT |
655 | levels ? "" : " ", names, space/bcount, |
656 | (space/bcount)*100/blocksize); | |
ac27a0ec DK |
657 | return (struct stats) { names, space, bcount}; |
658 | } | |
659 | #endif /* DX_DEBUG */ | |
660 | ||
661 | /* | |
662 | * Probe for a directory leaf block to search. | |
663 | * | |
664 | * dx_probe can return ERR_BAD_DX_DIR, which means there was a format | |
665 | * error in the directory index, and the caller should fall back to | |
666 | * searching the directory normally. The callers of dx_probe **MUST** | |
667 | * check for this error code, and make sure it never gets reflected | |
668 | * back to userspace. | |
669 | */ | |
670 | static struct dx_frame * | |
f702ba0f | 671 | dx_probe(const struct qstr *d_name, struct inode *dir, |
dd73b5d5 | 672 | struct dx_hash_info *hinfo, struct dx_frame *frame_in) |
ac27a0ec DK |
673 | { |
674 | unsigned count, indirect; | |
675 | struct dx_entry *at, *entries, *p, *q, *m; | |
676 | struct dx_root *root; | |
ac27a0ec | 677 | struct dx_frame *frame = frame_in; |
dd73b5d5 | 678 | struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR); |
ac27a0ec DK |
679 | u32 hash; |
680 | ||
dd73b5d5 TT |
681 | frame->bh = ext4_read_dirblock(dir, 0, INDEX); |
682 | if (IS_ERR(frame->bh)) | |
683 | return (struct dx_frame *) frame->bh; | |
684 | ||
685 | root = (struct dx_root *) frame->bh->b_data; | |
ac27a0ec DK |
686 | if (root->info.hash_version != DX_HASH_TEA && |
687 | root->info.hash_version != DX_HASH_HALF_MD4 && | |
688 | root->info.hash_version != DX_HASH_LEGACY) { | |
12062ddd | 689 | ext4_warning(dir->i_sb, "Unrecognised inode hash code %d", |
ac27a0ec | 690 | root->info.hash_version); |
ac27a0ec DK |
691 | goto fail; |
692 | } | |
693 | hinfo->hash_version = root->info.hash_version; | |
f99b2589 TT |
694 | if (hinfo->hash_version <= DX_HASH_TEA) |
695 | hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b | 696 | hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
f702ba0f TT |
697 | if (d_name) |
698 | ext4fs_dirhash(d_name->name, d_name->len, hinfo); | |
ac27a0ec DK |
699 | hash = hinfo->hash; |
700 | ||
701 | if (root->info.unused_flags & 1) { | |
12062ddd | 702 | ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x", |
ac27a0ec | 703 | root->info.unused_flags); |
ac27a0ec DK |
704 | goto fail; |
705 | } | |
706 | ||
707 | if ((indirect = root->info.indirect_levels) > 1) { | |
12062ddd | 708 | ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x", |
ac27a0ec | 709 | root->info.indirect_levels); |
ac27a0ec DK |
710 | goto fail; |
711 | } | |
712 | ||
713 | entries = (struct dx_entry *) (((char *)&root->info) + | |
714 | root->info.info_length); | |
3d82abae ES |
715 | |
716 | if (dx_get_limit(entries) != dx_root_limit(dir, | |
717 | root->info.info_length)) { | |
12062ddd | 718 | ext4_warning(dir->i_sb, "dx entry: limit != root limit"); |
3d82abae ES |
719 | goto fail; |
720 | } | |
721 | ||
af5bc92d | 722 | dxtrace(printk("Look up %x", hash)); |
dd73b5d5 | 723 | while (1) { |
ac27a0ec | 724 | count = dx_get_count(entries); |
3d82abae | 725 | if (!count || count > dx_get_limit(entries)) { |
12062ddd | 726 | ext4_warning(dir->i_sb, |
3d82abae | 727 | "dx entry: no count or count > limit"); |
dd73b5d5 | 728 | goto fail; |
3d82abae ES |
729 | } |
730 | ||
ac27a0ec DK |
731 | p = entries + 1; |
732 | q = entries + count - 1; | |
dd73b5d5 | 733 | while (p <= q) { |
ac27a0ec DK |
734 | m = p + (q - p)/2; |
735 | dxtrace(printk(".")); | |
736 | if (dx_get_hash(m) > hash) | |
737 | q = m - 1; | |
738 | else | |
739 | p = m + 1; | |
740 | } | |
741 | ||
dd73b5d5 | 742 | if (0) { // linear search cross check |
ac27a0ec DK |
743 | unsigned n = count - 1; |
744 | at = entries; | |
745 | while (n--) | |
746 | { | |
747 | dxtrace(printk(",")); | |
748 | if (dx_get_hash(++at) > hash) | |
749 | { | |
750 | at--; | |
751 | break; | |
752 | } | |
753 | } | |
754 | assert (at == p - 1); | |
755 | } | |
756 | ||
757 | at = p - 1; | |
758 | dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at))); | |
ac27a0ec DK |
759 | frame->entries = entries; |
760 | frame->at = at; | |
dd73b5d5 TT |
761 | if (!indirect--) |
762 | return frame; | |
763 | frame++; | |
764 | frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX); | |
765 | if (IS_ERR(frame->bh)) { | |
766 | ret_err = (struct dx_frame *) frame->bh; | |
767 | frame->bh = NULL; | |
768 | goto fail; | |
dbe89444 | 769 | } |
dd73b5d5 | 770 | entries = ((struct dx_node *) frame->bh->b_data)->entries; |
dbe89444 | 771 | |
3d82abae | 772 | if (dx_get_limit(entries) != dx_node_limit (dir)) { |
12062ddd | 773 | ext4_warning(dir->i_sb, |
3d82abae | 774 | "dx entry: limit != node limit"); |
dd73b5d5 | 775 | goto fail; |
3d82abae | 776 | } |
ac27a0ec | 777 | } |
dd73b5d5 | 778 | fail: |
ac27a0ec DK |
779 | while (frame >= frame_in) { |
780 | brelse(frame->bh); | |
781 | frame--; | |
782 | } | |
dd73b5d5 | 783 | if (ret_err == ERR_PTR(ERR_BAD_DX_DIR)) |
12062ddd | 784 | ext4_warning(dir->i_sb, |
9ee49302 | 785 | "Corrupt dir inode %lu, running e2fsck is " |
3d82abae | 786 | "recommended.", dir->i_ino); |
dd73b5d5 | 787 | return ret_err; |
ac27a0ec DK |
788 | } |
789 | ||
790 | static void dx_release (struct dx_frame *frames) | |
791 | { | |
792 | if (frames[0].bh == NULL) | |
793 | return; | |
794 | ||
795 | if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels) | |
796 | brelse(frames[1].bh); | |
797 | brelse(frames[0].bh); | |
798 | } | |
799 | ||
800 | /* | |
801 | * This function increments the frame pointer to search the next leaf | |
802 | * block, and reads in the necessary intervening nodes if the search | |
803 | * should be necessary. Whether or not the search is necessary is | |
804 | * controlled by the hash parameter. If the hash value is even, then | |
805 | * the search is only continued if the next block starts with that | |
806 | * hash value. This is used if we are searching for a specific file. | |
807 | * | |
808 | * If the hash value is HASH_NB_ALWAYS, then always go to the next block. | |
809 | * | |
810 | * This function returns 1 if the caller should continue to search, | |
811 | * or 0 if it should not. If there is an error reading one of the | |
812 | * index blocks, it will a negative error code. | |
813 | * | |
814 | * If start_hash is non-null, it will be filled in with the starting | |
815 | * hash of the next page. | |
816 | */ | |
617ba13b | 817 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, |
ac27a0ec DK |
818 | struct dx_frame *frame, |
819 | struct dx_frame *frames, | |
820 | __u32 *start_hash) | |
821 | { | |
822 | struct dx_frame *p; | |
823 | struct buffer_head *bh; | |
dc6982ff | 824 | int num_frames = 0; |
ac27a0ec DK |
825 | __u32 bhash; |
826 | ||
827 | p = frame; | |
828 | /* | |
829 | * Find the next leaf page by incrementing the frame pointer. | |
830 | * If we run out of entries in the interior node, loop around and | |
831 | * increment pointer in the parent node. When we break out of | |
832 | * this loop, num_frames indicates the number of interior | |
833 | * nodes need to be read. | |
834 | */ | |
835 | while (1) { | |
836 | if (++(p->at) < p->entries + dx_get_count(p->entries)) | |
837 | break; | |
838 | if (p == frames) | |
839 | return 0; | |
840 | num_frames++; | |
841 | p--; | |
842 | } | |
843 | ||
844 | /* | |
845 | * If the hash is 1, then continue only if the next page has a | |
846 | * continuation hash of any value. This is used for readdir | |
847 | * handling. Otherwise, check to see if the hash matches the | |
848 | * desired contiuation hash. If it doesn't, return since | |
849 | * there's no point to read in the successive index pages. | |
850 | */ | |
851 | bhash = dx_get_hash(p->at); | |
852 | if (start_hash) | |
853 | *start_hash = bhash; | |
854 | if ((hash & 1) == 0) { | |
855 | if ((bhash & ~1) != hash) | |
856 | return 0; | |
857 | } | |
858 | /* | |
859 | * If the hash is HASH_NB_ALWAYS, we always go to the next | |
860 | * block so no check is necessary | |
861 | */ | |
862 | while (num_frames--) { | |
dc6982ff TT |
863 | bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX); |
864 | if (IS_ERR(bh)) | |
865 | return PTR_ERR(bh); | |
ac27a0ec | 866 | p++; |
af5bc92d | 867 | brelse(p->bh); |
ac27a0ec DK |
868 | p->bh = bh; |
869 | p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; | |
870 | } | |
871 | return 1; | |
872 | } | |
873 | ||
874 | ||
ac27a0ec DK |
875 | /* |
876 | * This function fills a red-black tree with information from a | |
877 | * directory block. It returns the number directory entries loaded | |
878 | * into the tree. If there is an error it is returned in err. | |
879 | */ | |
880 | static int htree_dirblock_to_tree(struct file *dir_file, | |
725d26d3 | 881 | struct inode *dir, ext4_lblk_t block, |
ac27a0ec DK |
882 | struct dx_hash_info *hinfo, |
883 | __u32 start_hash, __u32 start_minor_hash) | |
884 | { | |
885 | struct buffer_head *bh; | |
617ba13b | 886 | struct ext4_dir_entry_2 *de, *top; |
90b0a973 | 887 | int err = 0, count = 0; |
ac27a0ec | 888 | |
725d26d3 AK |
889 | dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", |
890 | (unsigned long)block)); | |
dc6982ff TT |
891 | bh = ext4_read_dirblock(dir, block, DIRENT); |
892 | if (IS_ERR(bh)) | |
893 | return PTR_ERR(bh); | |
b0336e8d | 894 | |
617ba13b MC |
895 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
896 | top = (struct ext4_dir_entry_2 *) ((char *) de + | |
ac27a0ec | 897 | dir->i_sb->s_blocksize - |
617ba13b | 898 | EXT4_DIR_REC_LEN(0)); |
3d0518f4 | 899 | for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { |
f7c21177 | 900 | if (ext4_check_dir_entry(dir, NULL, de, bh, |
226ba972 | 901 | bh->b_data, bh->b_size, |
cad3f007 TT |
902 | (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) |
903 | + ((char *)de - bh->b_data))) { | |
64cb9273 AV |
904 | /* silently ignore the rest of the block */ |
905 | break; | |
e6c40211 | 906 | } |
617ba13b | 907 | ext4fs_dirhash(de->name, de->name_len, hinfo); |
ac27a0ec DK |
908 | if ((hinfo->hash < start_hash) || |
909 | ((hinfo->hash == start_hash) && | |
910 | (hinfo->minor_hash < start_minor_hash))) | |
911 | continue; | |
912 | if (de->inode == 0) | |
913 | continue; | |
617ba13b | 914 | if ((err = ext4_htree_store_dirent(dir_file, |
ac27a0ec DK |
915 | hinfo->hash, hinfo->minor_hash, de)) != 0) { |
916 | brelse(bh); | |
917 | return err; | |
918 | } | |
919 | count++; | |
920 | } | |
921 | brelse(bh); | |
922 | return count; | |
923 | } | |
924 | ||
925 | ||
926 | /* | |
927 | * This function fills a red-black tree with information from a | |
928 | * directory. We start scanning the directory in hash order, starting | |
929 | * at start_hash and start_minor_hash. | |
930 | * | |
931 | * This function returns the number of entries inserted into the tree, | |
932 | * or a negative error code. | |
933 | */ | |
617ba13b | 934 | int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, |
ac27a0ec DK |
935 | __u32 start_minor_hash, __u32 *next_hash) |
936 | { | |
937 | struct dx_hash_info hinfo; | |
617ba13b | 938 | struct ext4_dir_entry_2 *de; |
ac27a0ec DK |
939 | struct dx_frame frames[2], *frame; |
940 | struct inode *dir; | |
725d26d3 | 941 | ext4_lblk_t block; |
ac27a0ec | 942 | int count = 0; |
725d26d3 | 943 | int ret, err; |
ac27a0ec DK |
944 | __u32 hashval; |
945 | ||
60e6679e | 946 | dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", |
4776004f | 947 | start_hash, start_minor_hash)); |
496ad9aa | 948 | dir = file_inode(dir_file); |
12e9b892 | 949 | if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) { |
617ba13b | 950 | hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; |
f99b2589 TT |
951 | if (hinfo.hash_version <= DX_HASH_TEA) |
952 | hinfo.hash_version += | |
953 | EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b | 954 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
8af0f082 TM |
955 | if (ext4_has_inline_data(dir)) { |
956 | int has_inline_data = 1; | |
957 | count = htree_inlinedir_to_tree(dir_file, dir, 0, | |
958 | &hinfo, start_hash, | |
959 | start_minor_hash, | |
960 | &has_inline_data); | |
961 | if (has_inline_data) { | |
962 | *next_hash = ~0; | |
963 | return count; | |
964 | } | |
965 | } | |
ac27a0ec DK |
966 | count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, |
967 | start_hash, start_minor_hash); | |
968 | *next_hash = ~0; | |
969 | return count; | |
970 | } | |
971 | hinfo.hash = start_hash; | |
972 | hinfo.minor_hash = 0; | |
dd73b5d5 TT |
973 | frame = dx_probe(NULL, dir, &hinfo, frames); |
974 | if (IS_ERR(frame)) | |
975 | return PTR_ERR(frame); | |
ac27a0ec DK |
976 | |
977 | /* Add '.' and '..' from the htree header */ | |
978 | if (!start_hash && !start_minor_hash) { | |
617ba13b MC |
979 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; |
980 | if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0) | |
ac27a0ec DK |
981 | goto errout; |
982 | count++; | |
983 | } | |
984 | if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { | |
617ba13b | 985 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; |
3d0518f4 | 986 | de = ext4_next_entry(de, dir->i_sb->s_blocksize); |
617ba13b | 987 | if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0) |
ac27a0ec DK |
988 | goto errout; |
989 | count++; | |
990 | } | |
991 | ||
992 | while (1) { | |
993 | block = dx_get_block(frame->at); | |
994 | ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, | |
995 | start_hash, start_minor_hash); | |
996 | if (ret < 0) { | |
997 | err = ret; | |
998 | goto errout; | |
999 | } | |
1000 | count += ret; | |
1001 | hashval = ~0; | |
617ba13b | 1002 | ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, |
ac27a0ec DK |
1003 | frame, frames, &hashval); |
1004 | *next_hash = hashval; | |
1005 | if (ret < 0) { | |
1006 | err = ret; | |
1007 | goto errout; | |
1008 | } | |
1009 | /* | |
1010 | * Stop if: (a) there are no more entries, or | |
1011 | * (b) we have inserted at least one entry and the | |
1012 | * next hash value is not a continuation | |
1013 | */ | |
1014 | if ((ret == 0) || | |
1015 | (count && ((hashval & 1) == 0))) | |
1016 | break; | |
1017 | } | |
1018 | dx_release(frames); | |
4776004f TT |
1019 | dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " |
1020 | "next hash: %x\n", count, *next_hash)); | |
ac27a0ec DK |
1021 | return count; |
1022 | errout: | |
1023 | dx_release(frames); | |
1024 | return (err); | |
1025 | } | |
1026 | ||
7335cd3b TM |
1027 | static inline int search_dirblock(struct buffer_head *bh, |
1028 | struct inode *dir, | |
1029 | const struct qstr *d_name, | |
1030 | unsigned int offset, | |
1031 | struct ext4_dir_entry_2 **res_dir) | |
1032 | { | |
1033 | return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir, | |
1034 | d_name, offset, res_dir); | |
1035 | } | |
1036 | ||
ac27a0ec DK |
1037 | /* |
1038 | * Directory block splitting, compacting | |
1039 | */ | |
1040 | ||
ef2b02d3 ES |
1041 | /* |
1042 | * Create map of hash values, offsets, and sizes, stored at end of block. | |
1043 | * Returns number of entries mapped. | |
1044 | */ | |
8bad4597 TT |
1045 | static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize, |
1046 | struct dx_hash_info *hinfo, | |
1047 | struct dx_map_entry *map_tail) | |
ac27a0ec DK |
1048 | { |
1049 | int count = 0; | |
1050 | char *base = (char *) de; | |
1051 | struct dx_hash_info h = *hinfo; | |
1052 | ||
8bad4597 | 1053 | while ((char *) de < base + blocksize) { |
ac27a0ec | 1054 | if (de->name_len && de->inode) { |
617ba13b | 1055 | ext4fs_dirhash(de->name, de->name_len, &h); |
ac27a0ec DK |
1056 | map_tail--; |
1057 | map_tail->hash = h.hash; | |
9aee2286 | 1058 | map_tail->offs = ((char *) de - base)>>2; |
ef2b02d3 | 1059 | map_tail->size = le16_to_cpu(de->rec_len); |
ac27a0ec DK |
1060 | count++; |
1061 | cond_resched(); | |
1062 | } | |
1063 | /* XXX: do we need to check rec_len == 0 case? -Chris */ | |
3d0518f4 | 1064 | de = ext4_next_entry(de, blocksize); |
ac27a0ec DK |
1065 | } |
1066 | return count; | |
1067 | } | |
1068 | ||
ef2b02d3 | 1069 | /* Sort map by hash value */ |
ac27a0ec DK |
1070 | static void dx_sort_map (struct dx_map_entry *map, unsigned count) |
1071 | { | |
63f57933 AM |
1072 | struct dx_map_entry *p, *q, *top = map + count - 1; |
1073 | int more; | |
1074 | /* Combsort until bubble sort doesn't suck */ | |
1075 | while (count > 2) { | |
1076 | count = count*10/13; | |
1077 | if (count - 9 < 2) /* 9, 10 -> 11 */ | |
1078 | count = 11; | |
1079 | for (p = top, q = p - count; q >= map; p--, q--) | |
1080 | if (p->hash < q->hash) | |
1081 | swap(*p, *q); | |
1082 | } | |
1083 | /* Garden variety bubble sort */ | |
1084 | do { | |
1085 | more = 0; | |
1086 | q = top; | |
1087 | while (q-- > map) { | |
1088 | if (q[1].hash >= q[0].hash) | |
ac27a0ec | 1089 | continue; |
63f57933 AM |
1090 | swap(*(q+1), *q); |
1091 | more = 1; | |
ac27a0ec DK |
1092 | } |
1093 | } while(more); | |
1094 | } | |
1095 | ||
725d26d3 | 1096 | static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) |
ac27a0ec DK |
1097 | { |
1098 | struct dx_entry *entries = frame->entries; | |
1099 | struct dx_entry *old = frame->at, *new = old + 1; | |
1100 | int count = dx_get_count(entries); | |
1101 | ||
1102 | assert(count < dx_get_limit(entries)); | |
1103 | assert(old < entries + count); | |
1104 | memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); | |
1105 | dx_set_hash(new, hash); | |
1106 | dx_set_block(new, block); | |
1107 | dx_set_count(entries, count + 1); | |
1108 | } | |
ac27a0ec | 1109 | |
ac27a0ec | 1110 | /* |
617ba13b | 1111 | * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure. |
ac27a0ec | 1112 | * |
617ba13b | 1113 | * `len <= EXT4_NAME_LEN' is guaranteed by caller. |
ac27a0ec DK |
1114 | * `de != NULL' is guaranteed by caller. |
1115 | */ | |
617ba13b MC |
1116 | static inline int ext4_match (int len, const char * const name, |
1117 | struct ext4_dir_entry_2 * de) | |
ac27a0ec DK |
1118 | { |
1119 | if (len != de->name_len) | |
1120 | return 0; | |
1121 | if (!de->inode) | |
1122 | return 0; | |
1123 | return !memcmp(name, de->name, len); | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * Returns 0 if not found, -1 on failure, and 1 on success | |
1128 | */ | |
7335cd3b TM |
1129 | int search_dir(struct buffer_head *bh, |
1130 | char *search_buf, | |
1131 | int buf_size, | |
1132 | struct inode *dir, | |
1133 | const struct qstr *d_name, | |
1134 | unsigned int offset, | |
1135 | struct ext4_dir_entry_2 **res_dir) | |
ac27a0ec | 1136 | { |
617ba13b | 1137 | struct ext4_dir_entry_2 * de; |
ac27a0ec DK |
1138 | char * dlimit; |
1139 | int de_len; | |
f702ba0f TT |
1140 | const char *name = d_name->name; |
1141 | int namelen = d_name->len; | |
ac27a0ec | 1142 | |
7335cd3b TM |
1143 | de = (struct ext4_dir_entry_2 *)search_buf; |
1144 | dlimit = search_buf + buf_size; | |
ac27a0ec DK |
1145 | while ((char *) de < dlimit) { |
1146 | /* this code is executed quadratically often */ | |
1147 | /* do minimal checking `by hand' */ | |
1148 | ||
1149 | if ((char *) de + namelen <= dlimit && | |
617ba13b | 1150 | ext4_match (namelen, name, de)) { |
ac27a0ec | 1151 | /* found a match - just to be sure, do a full check */ |
226ba972 TM |
1152 | if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data, |
1153 | bh->b_size, offset)) | |
ac27a0ec DK |
1154 | return -1; |
1155 | *res_dir = de; | |
1156 | return 1; | |
1157 | } | |
1158 | /* prevent looping on a bad block */ | |
3d0518f4 WY |
1159 | de_len = ext4_rec_len_from_disk(de->rec_len, |
1160 | dir->i_sb->s_blocksize); | |
ac27a0ec DK |
1161 | if (de_len <= 0) |
1162 | return -1; | |
1163 | offset += de_len; | |
617ba13b | 1164 | de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); |
ac27a0ec DK |
1165 | } |
1166 | return 0; | |
1167 | } | |
1168 | ||
c6af8803 DW |
1169 | static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block, |
1170 | struct ext4_dir_entry *de) | |
1171 | { | |
1172 | struct super_block *sb = dir->i_sb; | |
1173 | ||
1174 | if (!is_dx(dir)) | |
1175 | return 0; | |
1176 | if (block == 0) | |
1177 | return 1; | |
1178 | if (de->inode == 0 && | |
1179 | ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) == | |
1180 | sb->s_blocksize) | |
1181 | return 1; | |
1182 | return 0; | |
1183 | } | |
ac27a0ec DK |
1184 | |
1185 | /* | |
617ba13b | 1186 | * ext4_find_entry() |
ac27a0ec DK |
1187 | * |
1188 | * finds an entry in the specified directory with the wanted name. It | |
1189 | * returns the cache buffer in which the entry was found, and the entry | |
1190 | * itself (as a parameter - res_dir). It does NOT read the inode of the | |
1191 | * entry - you'll have to do that yourself if you want to. | |
1192 | * | |
1193 | * The returned buffer_head has ->b_count elevated. The caller is expected | |
1194 | * to brelse() it when appropriate. | |
1195 | */ | |
f702ba0f TT |
1196 | static struct buffer_head * ext4_find_entry (struct inode *dir, |
1197 | const struct qstr *d_name, | |
32f7f22c TM |
1198 | struct ext4_dir_entry_2 **res_dir, |
1199 | int *inlined) | |
ac27a0ec | 1200 | { |
af5bc92d TT |
1201 | struct super_block *sb; |
1202 | struct buffer_head *bh_use[NAMEI_RA_SIZE]; | |
1203 | struct buffer_head *bh, *ret = NULL; | |
725d26d3 | 1204 | ext4_lblk_t start, block, b; |
8941ec8b | 1205 | const u8 *name = d_name->name; |
ac27a0ec DK |
1206 | int ra_max = 0; /* Number of bh's in the readahead |
1207 | buffer, bh_use[] */ | |
1208 | int ra_ptr = 0; /* Current index into readahead | |
1209 | buffer */ | |
1210 | int num = 0; | |
725d26d3 | 1211 | ext4_lblk_t nblocks; |
10560082 | 1212 | int i, namelen; |
ac27a0ec DK |
1213 | |
1214 | *res_dir = NULL; | |
1215 | sb = dir->i_sb; | |
f702ba0f | 1216 | namelen = d_name->len; |
617ba13b | 1217 | if (namelen > EXT4_NAME_LEN) |
ac27a0ec | 1218 | return NULL; |
e8e948e7 TM |
1219 | |
1220 | if (ext4_has_inline_data(dir)) { | |
1221 | int has_inline_data = 1; | |
1222 | ret = ext4_find_inline_entry(dir, d_name, res_dir, | |
1223 | &has_inline_data); | |
32f7f22c TM |
1224 | if (has_inline_data) { |
1225 | if (inlined) | |
1226 | *inlined = 1; | |
e8e948e7 | 1227 | return ret; |
32f7f22c | 1228 | } |
e8e948e7 TM |
1229 | } |
1230 | ||
8941ec8b | 1231 | if ((namelen <= 2) && (name[0] == '.') && |
6d5c3aa8 | 1232 | (name[1] == '.' || name[1] == '\0')) { |
8941ec8b TT |
1233 | /* |
1234 | * "." or ".." will only be in the first block | |
1235 | * NFS may look up ".."; "." should be handled by the VFS | |
1236 | */ | |
1237 | block = start = 0; | |
1238 | nblocks = 1; | |
1239 | goto restart; | |
1240 | } | |
ac27a0ec | 1241 | if (is_dx(dir)) { |
537d8f93 | 1242 | bh = ext4_dx_find_entry(dir, d_name, res_dir); |
ac27a0ec DK |
1243 | /* |
1244 | * On success, or if the error was file not found, | |
1245 | * return. Otherwise, fall back to doing a search the | |
1246 | * old fashioned way. | |
1247 | */ | |
537d8f93 | 1248 | if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR) |
ac27a0ec | 1249 | return bh; |
4776004f TT |
1250 | dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " |
1251 | "falling back\n")); | |
ac27a0ec | 1252 | } |
617ba13b MC |
1253 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); |
1254 | start = EXT4_I(dir)->i_dir_start_lookup; | |
ac27a0ec DK |
1255 | if (start >= nblocks) |
1256 | start = 0; | |
1257 | block = start; | |
1258 | restart: | |
1259 | do { | |
1260 | /* | |
1261 | * We deal with the read-ahead logic here. | |
1262 | */ | |
1263 | if (ra_ptr >= ra_max) { | |
1264 | /* Refill the readahead buffer */ | |
1265 | ra_ptr = 0; | |
1266 | b = block; | |
1267 | for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { | |
1268 | /* | |
1269 | * Terminate if we reach the end of the | |
1270 | * directory and must wrap, or if our | |
1271 | * search has finished at this block. | |
1272 | */ | |
1273 | if (b >= nblocks || (num && block == start)) { | |
1274 | bh_use[ra_max] = NULL; | |
1275 | break; | |
1276 | } | |
1277 | num++; | |
10560082 TT |
1278 | bh = ext4_getblk(NULL, dir, b++, 0); |
1279 | if (unlikely(IS_ERR(bh))) { | |
36de9286 | 1280 | if (ra_max == 0) |
10560082 | 1281 | return bh; |
36de9286 TT |
1282 | break; |
1283 | } | |
ac27a0ec DK |
1284 | bh_use[ra_max] = bh; |
1285 | if (bh) | |
65299a3b CH |
1286 | ll_rw_block(READ | REQ_META | REQ_PRIO, |
1287 | 1, &bh); | |
ac27a0ec DK |
1288 | } |
1289 | } | |
1290 | if ((bh = bh_use[ra_ptr++]) == NULL) | |
1291 | goto next; | |
1292 | wait_on_buffer(bh); | |
1293 | if (!buffer_uptodate(bh)) { | |
1294 | /* read error, skip block & hope for the best */ | |
24676da4 TT |
1295 | EXT4_ERROR_INODE(dir, "reading directory lblock %lu", |
1296 | (unsigned long) block); | |
ac27a0ec DK |
1297 | brelse(bh); |
1298 | goto next; | |
1299 | } | |
b0336e8d | 1300 | if (!buffer_verified(bh) && |
c6af8803 DW |
1301 | !is_dx_internal_node(dir, block, |
1302 | (struct ext4_dir_entry *)bh->b_data) && | |
b0336e8d DW |
1303 | !ext4_dirent_csum_verify(dir, |
1304 | (struct ext4_dir_entry *)bh->b_data)) { | |
1305 | EXT4_ERROR_INODE(dir, "checksumming directory " | |
1306 | "block %lu", (unsigned long)block); | |
1307 | brelse(bh); | |
1308 | goto next; | |
1309 | } | |
1310 | set_buffer_verified(bh); | |
f702ba0f | 1311 | i = search_dirblock(bh, dir, d_name, |
617ba13b | 1312 | block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); |
ac27a0ec | 1313 | if (i == 1) { |
617ba13b | 1314 | EXT4_I(dir)->i_dir_start_lookup = block; |
ac27a0ec DK |
1315 | ret = bh; |
1316 | goto cleanup_and_exit; | |
1317 | } else { | |
1318 | brelse(bh); | |
1319 | if (i < 0) | |
1320 | goto cleanup_and_exit; | |
1321 | } | |
1322 | next: | |
1323 | if (++block >= nblocks) | |
1324 | block = 0; | |
1325 | } while (block != start); | |
1326 | ||
1327 | /* | |
1328 | * If the directory has grown while we were searching, then | |
1329 | * search the last part of the directory before giving up. | |
1330 | */ | |
1331 | block = nblocks; | |
617ba13b | 1332 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); |
ac27a0ec DK |
1333 | if (block < nblocks) { |
1334 | start = 0; | |
1335 | goto restart; | |
1336 | } | |
1337 | ||
1338 | cleanup_and_exit: | |
1339 | /* Clean up the read-ahead blocks */ | |
1340 | for (; ra_ptr < ra_max; ra_ptr++) | |
af5bc92d | 1341 | brelse(bh_use[ra_ptr]); |
ac27a0ec DK |
1342 | return ret; |
1343 | } | |
1344 | ||
f702ba0f | 1345 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name, |
537d8f93 | 1346 | struct ext4_dir_entry_2 **res_dir) |
ac27a0ec | 1347 | { |
8941ec8b | 1348 | struct super_block * sb = dir->i_sb; |
ac27a0ec | 1349 | struct dx_hash_info hinfo; |
ac27a0ec | 1350 | struct dx_frame frames[2], *frame; |
ac27a0ec | 1351 | struct buffer_head *bh; |
725d26d3 | 1352 | ext4_lblk_t block; |
dd73b5d5 | 1353 | int retval; |
ac27a0ec | 1354 | |
dd73b5d5 TT |
1355 | frame = dx_probe(d_name, dir, &hinfo, frames); |
1356 | if (IS_ERR(frame)) | |
1357 | return (struct buffer_head *) frame; | |
ac27a0ec DK |
1358 | do { |
1359 | block = dx_get_block(frame->at); | |
dc6982ff | 1360 | bh = ext4_read_dirblock(dir, block, DIRENT); |
537d8f93 | 1361 | if (IS_ERR(bh)) |
b0336e8d | 1362 | goto errout; |
537d8f93 | 1363 | |
7845c049 TT |
1364 | retval = search_dirblock(bh, dir, d_name, |
1365 | block << EXT4_BLOCK_SIZE_BITS(sb), | |
1366 | res_dir); | |
537d8f93 TT |
1367 | if (retval == 1) |
1368 | goto success; | |
af5bc92d | 1369 | brelse(bh); |
7845c049 | 1370 | if (retval == -1) { |
537d8f93 | 1371 | bh = ERR_PTR(ERR_BAD_DX_DIR); |
7845c049 TT |
1372 | goto errout; |
1373 | } | |
1374 | ||
ac27a0ec | 1375 | /* Check to see if we should continue to search */ |
8941ec8b | 1376 | retval = ext4_htree_next_block(dir, hinfo.hash, frame, |
ac27a0ec DK |
1377 | frames, NULL); |
1378 | if (retval < 0) { | |
12062ddd | 1379 | ext4_warning(sb, |
537d8f93 TT |
1380 | "error %d reading index page in directory #%lu", |
1381 | retval, dir->i_ino); | |
1382 | bh = ERR_PTR(retval); | |
ac27a0ec DK |
1383 | goto errout; |
1384 | } | |
1385 | } while (retval == 1); | |
1386 | ||
537d8f93 | 1387 | bh = NULL; |
ac27a0ec | 1388 | errout: |
265c6a0f | 1389 | dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name)); |
537d8f93 TT |
1390 | success: |
1391 | dx_release(frames); | |
1392 | return bh; | |
ac27a0ec | 1393 | } |
ac27a0ec | 1394 | |
00cd8dd3 | 1395 | static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) |
ac27a0ec | 1396 | { |
af5bc92d TT |
1397 | struct inode *inode; |
1398 | struct ext4_dir_entry_2 *de; | |
1399 | struct buffer_head *bh; | |
ac27a0ec | 1400 | |
617ba13b | 1401 | if (dentry->d_name.len > EXT4_NAME_LEN) |
ac27a0ec DK |
1402 | return ERR_PTR(-ENAMETOOLONG); |
1403 | ||
32f7f22c | 1404 | bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); |
36de9286 TT |
1405 | if (IS_ERR(bh)) |
1406 | return (struct dentry *) bh; | |
ac27a0ec DK |
1407 | inode = NULL; |
1408 | if (bh) { | |
498e5f24 | 1409 | __u32 ino = le32_to_cpu(de->inode); |
af5bc92d | 1410 | brelse(bh); |
617ba13b | 1411 | if (!ext4_valid_inum(dir->i_sb, ino)) { |
24676da4 | 1412 | EXT4_ERROR_INODE(dir, "bad inode number: %u", ino); |
1d1fe1ee | 1413 | return ERR_PTR(-EIO); |
a6c15c2b | 1414 | } |
7e936b73 | 1415 | if (unlikely(ino == dir->i_ino)) { |
a34e15cc DH |
1416 | EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir", |
1417 | dentry); | |
7e936b73 AD |
1418 | return ERR_PTR(-EIO); |
1419 | } | |
1d1fe1ee | 1420 | inode = ext4_iget(dir->i_sb, ino); |
a9049376 AV |
1421 | if (inode == ERR_PTR(-ESTALE)) { |
1422 | EXT4_ERROR_INODE(dir, | |
1423 | "deleted inode referenced: %u", | |
1424 | ino); | |
1425 | return ERR_PTR(-EIO); | |
e6f009b0 | 1426 | } |
ac27a0ec DK |
1427 | } |
1428 | return d_splice_alias(inode, dentry); | |
1429 | } | |
1430 | ||
1431 | ||
617ba13b | 1432 | struct dentry *ext4_get_parent(struct dentry *child) |
ac27a0ec | 1433 | { |
498e5f24 | 1434 | __u32 ino; |
26fe5750 | 1435 | static const struct qstr dotdot = QSTR_INIT("..", 2); |
617ba13b | 1436 | struct ext4_dir_entry_2 * de; |
ac27a0ec DK |
1437 | struct buffer_head *bh; |
1438 | ||
32f7f22c | 1439 | bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL); |
36de9286 TT |
1440 | if (IS_ERR(bh)) |
1441 | return (struct dentry *) bh; | |
ac27a0ec DK |
1442 | if (!bh) |
1443 | return ERR_PTR(-ENOENT); | |
1444 | ino = le32_to_cpu(de->inode); | |
1445 | brelse(bh); | |
1446 | ||
617ba13b | 1447 | if (!ext4_valid_inum(child->d_inode->i_sb, ino)) { |
24676da4 TT |
1448 | EXT4_ERROR_INODE(child->d_inode, |
1449 | "bad parent inode number: %u", ino); | |
1d1fe1ee | 1450 | return ERR_PTR(-EIO); |
a6c15c2b VA |
1451 | } |
1452 | ||
44003728 | 1453 | return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino)); |
ac27a0ec DK |
1454 | } |
1455 | ||
ef2b02d3 ES |
1456 | /* |
1457 | * Move count entries from end of map between two memory locations. | |
1458 | * Returns pointer to last entry moved. | |
1459 | */ | |
617ba13b | 1460 | static struct ext4_dir_entry_2 * |
3d0518f4 WY |
1461 | dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count, |
1462 | unsigned blocksize) | |
ac27a0ec DK |
1463 | { |
1464 | unsigned rec_len = 0; | |
1465 | ||
1466 | while (count--) { | |
60e6679e | 1467 | struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) |
9aee2286 | 1468 | (from + (map->offs<<2)); |
617ba13b | 1469 | rec_len = EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec | 1470 | memcpy (to, de, rec_len); |
617ba13b | 1471 | ((struct ext4_dir_entry_2 *) to)->rec_len = |
3d0518f4 | 1472 | ext4_rec_len_to_disk(rec_len, blocksize); |
ac27a0ec DK |
1473 | de->inode = 0; |
1474 | map++; | |
1475 | to += rec_len; | |
1476 | } | |
617ba13b | 1477 | return (struct ext4_dir_entry_2 *) (to - rec_len); |
ac27a0ec DK |
1478 | } |
1479 | ||
ef2b02d3 ES |
1480 | /* |
1481 | * Compact each dir entry in the range to the minimal rec_len. | |
1482 | * Returns pointer to last entry in range. | |
1483 | */ | |
8bad4597 | 1484 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize) |
ac27a0ec | 1485 | { |
617ba13b | 1486 | struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; |
ac27a0ec DK |
1487 | unsigned rec_len = 0; |
1488 | ||
1489 | prev = to = de; | |
8bad4597 | 1490 | while ((char*)de < base + blocksize) { |
3d0518f4 | 1491 | next = ext4_next_entry(de, blocksize); |
ac27a0ec | 1492 | if (de->inode && de->name_len) { |
617ba13b | 1493 | rec_len = EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec DK |
1494 | if (de > to) |
1495 | memmove(to, de, rec_len); | |
3d0518f4 | 1496 | to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); |
ac27a0ec | 1497 | prev = to; |
617ba13b | 1498 | to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); |
ac27a0ec DK |
1499 | } |
1500 | de = next; | |
1501 | } | |
1502 | return prev; | |
1503 | } | |
1504 | ||
ef2b02d3 ES |
1505 | /* |
1506 | * Split a full leaf block to make room for a new dir entry. | |
1507 | * Allocate a new block, and move entries so that they are approx. equally full. | |
1508 | * Returns pointer to de in block into which the new entry will be inserted. | |
1509 | */ | |
617ba13b | 1510 | static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, |
ac27a0ec DK |
1511 | struct buffer_head **bh,struct dx_frame *frame, |
1512 | struct dx_hash_info *hinfo, int *error) | |
1513 | { | |
1514 | unsigned blocksize = dir->i_sb->s_blocksize; | |
1515 | unsigned count, continued; | |
1516 | struct buffer_head *bh2; | |
725d26d3 | 1517 | ext4_lblk_t newblock; |
ac27a0ec DK |
1518 | u32 hash2; |
1519 | struct dx_map_entry *map; | |
1520 | char *data1 = (*bh)->b_data, *data2; | |
59e315b4 | 1521 | unsigned split, move, size; |
617ba13b | 1522 | struct ext4_dir_entry_2 *de = NULL, *de2; |
b0336e8d DW |
1523 | struct ext4_dir_entry_tail *t; |
1524 | int csum_size = 0; | |
59e315b4 | 1525 | int err = 0, i; |
ac27a0ec | 1526 | |
b0336e8d DW |
1527 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, |
1528 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
1529 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
1530 | ||
0f70b406 TT |
1531 | bh2 = ext4_append(handle, dir, &newblock); |
1532 | if (IS_ERR(bh2)) { | |
ac27a0ec DK |
1533 | brelse(*bh); |
1534 | *bh = NULL; | |
0f70b406 TT |
1535 | *error = PTR_ERR(bh2); |
1536 | return NULL; | |
ac27a0ec DK |
1537 | } |
1538 | ||
1539 | BUFFER_TRACE(*bh, "get_write_access"); | |
617ba13b | 1540 | err = ext4_journal_get_write_access(handle, *bh); |
fedee54d DM |
1541 | if (err) |
1542 | goto journal_error; | |
1543 | ||
ac27a0ec | 1544 | BUFFER_TRACE(frame->bh, "get_write_access"); |
617ba13b | 1545 | err = ext4_journal_get_write_access(handle, frame->bh); |
ac27a0ec DK |
1546 | if (err) |
1547 | goto journal_error; | |
1548 | ||
1549 | data2 = bh2->b_data; | |
1550 | ||
1551 | /* create map in the end of data2 block */ | |
1552 | map = (struct dx_map_entry *) (data2 + blocksize); | |
af5bc92d | 1553 | count = dx_make_map((struct ext4_dir_entry_2 *) data1, |
ac27a0ec DK |
1554 | blocksize, hinfo, map); |
1555 | map -= count; | |
af5bc92d | 1556 | dx_sort_map(map, count); |
ef2b02d3 ES |
1557 | /* Split the existing block in the middle, size-wise */ |
1558 | size = 0; | |
1559 | move = 0; | |
1560 | for (i = count-1; i >= 0; i--) { | |
1561 | /* is more than half of this entry in 2nd half of the block? */ | |
1562 | if (size + map[i].size/2 > blocksize/2) | |
1563 | break; | |
1564 | size += map[i].size; | |
1565 | move++; | |
1566 | } | |
1567 | /* map index at which we will split */ | |
1568 | split = count - move; | |
ac27a0ec DK |
1569 | hash2 = map[split].hash; |
1570 | continued = hash2 == map[split - 1].hash; | |
725d26d3 AK |
1571 | dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", |
1572 | (unsigned long)dx_get_block(frame->at), | |
1573 | hash2, split, count-split)); | |
ac27a0ec DK |
1574 | |
1575 | /* Fancy dance to stay within two buffers */ | |
3d0518f4 | 1576 | de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize); |
af5bc92d | 1577 | de = dx_pack_dirents(data1, blocksize); |
b0336e8d DW |
1578 | de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - |
1579 | (char *) de, | |
3d0518f4 | 1580 | blocksize); |
b0336e8d DW |
1581 | de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) - |
1582 | (char *) de2, | |
3d0518f4 | 1583 | blocksize); |
b0336e8d DW |
1584 | if (csum_size) { |
1585 | t = EXT4_DIRENT_TAIL(data2, blocksize); | |
1586 | initialize_dirent_tail(t, blocksize); | |
1587 | ||
1588 | t = EXT4_DIRENT_TAIL(data1, blocksize); | |
1589 | initialize_dirent_tail(t, blocksize); | |
1590 | } | |
1591 | ||
617ba13b MC |
1592 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1)); |
1593 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1)); | |
ac27a0ec DK |
1594 | |
1595 | /* Which block gets the new entry? */ | |
1596 | if (hinfo->hash >= hash2) | |
1597 | { | |
1598 | swap(*bh, bh2); | |
1599 | de = de2; | |
1600 | } | |
af5bc92d | 1601 | dx_insert_block(frame, hash2 + continued, newblock); |
b0336e8d | 1602 | err = ext4_handle_dirty_dirent_node(handle, dir, bh2); |
ac27a0ec DK |
1603 | if (err) |
1604 | goto journal_error; | |
dbe89444 | 1605 | err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); |
ac27a0ec DK |
1606 | if (err) |
1607 | goto journal_error; | |
af5bc92d TT |
1608 | brelse(bh2); |
1609 | dxtrace(dx_show_index("frame", frame->entries)); | |
ac27a0ec | 1610 | return de; |
fedee54d DM |
1611 | |
1612 | journal_error: | |
1613 | brelse(*bh); | |
1614 | brelse(bh2); | |
1615 | *bh = NULL; | |
1616 | ext4_std_error(dir->i_sb, err); | |
fedee54d DM |
1617 | *error = err; |
1618 | return NULL; | |
ac27a0ec | 1619 | } |
ac27a0ec | 1620 | |
978fef91 TM |
1621 | int ext4_find_dest_de(struct inode *dir, struct inode *inode, |
1622 | struct buffer_head *bh, | |
1623 | void *buf, int buf_size, | |
1624 | const char *name, int namelen, | |
1625 | struct ext4_dir_entry_2 **dest_de) | |
1626 | { | |
1627 | struct ext4_dir_entry_2 *de; | |
1628 | unsigned short reclen = EXT4_DIR_REC_LEN(namelen); | |
1629 | int nlen, rlen; | |
1630 | unsigned int offset = 0; | |
1631 | char *top; | |
1632 | ||
1633 | de = (struct ext4_dir_entry_2 *)buf; | |
1634 | top = buf + buf_size - reclen; | |
1635 | while ((char *) de <= top) { | |
1636 | if (ext4_check_dir_entry(dir, NULL, de, bh, | |
1637 | buf, buf_size, offset)) | |
1638 | return -EIO; | |
1639 | if (ext4_match(namelen, name, de)) | |
1640 | return -EEXIST; | |
1641 | nlen = EXT4_DIR_REC_LEN(de->name_len); | |
1642 | rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); | |
1643 | if ((de->inode ? rlen - nlen : rlen) >= reclen) | |
1644 | break; | |
1645 | de = (struct ext4_dir_entry_2 *)((char *)de + rlen); | |
1646 | offset += rlen; | |
1647 | } | |
1648 | if ((char *) de > top) | |
1649 | return -ENOSPC; | |
1650 | ||
1651 | *dest_de = de; | |
1652 | return 0; | |
1653 | } | |
1654 | ||
1655 | void ext4_insert_dentry(struct inode *inode, | |
1656 | struct ext4_dir_entry_2 *de, | |
1657 | int buf_size, | |
1658 | const char *name, int namelen) | |
1659 | { | |
1660 | ||
1661 | int nlen, rlen; | |
1662 | ||
1663 | nlen = EXT4_DIR_REC_LEN(de->name_len); | |
1664 | rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); | |
1665 | if (de->inode) { | |
1666 | struct ext4_dir_entry_2 *de1 = | |
1667 | (struct ext4_dir_entry_2 *)((char *)de + nlen); | |
1668 | de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size); | |
1669 | de->rec_len = ext4_rec_len_to_disk(nlen, buf_size); | |
1670 | de = de1; | |
1671 | } | |
1672 | de->file_type = EXT4_FT_UNKNOWN; | |
1673 | de->inode = cpu_to_le32(inode->i_ino); | |
1674 | ext4_set_de_type(inode->i_sb, de, inode->i_mode); | |
1675 | de->name_len = namelen; | |
1676 | memcpy(de->name, name, namelen); | |
1677 | } | |
ac27a0ec DK |
1678 | /* |
1679 | * Add a new entry into a directory (leaf) block. If de is non-NULL, | |
1680 | * it points to a directory entry which is guaranteed to be large | |
1681 | * enough for new directory entry. If de is NULL, then | |
1682 | * add_dirent_to_buf will attempt search the directory block for | |
1683 | * space. It will return -ENOSPC if no space is available, and -EIO | |
1684 | * and -EEXIST if directory entry already exists. | |
ac27a0ec DK |
1685 | */ |
1686 | static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry, | |
617ba13b | 1687 | struct inode *inode, struct ext4_dir_entry_2 *de, |
af5bc92d | 1688 | struct buffer_head *bh) |
ac27a0ec DK |
1689 | { |
1690 | struct inode *dir = dentry->d_parent->d_inode; | |
1691 | const char *name = dentry->d_name.name; | |
1692 | int namelen = dentry->d_name.len; | |
3d0518f4 | 1693 | unsigned int blocksize = dir->i_sb->s_blocksize; |
b0336e8d | 1694 | int csum_size = 0; |
978fef91 | 1695 | int err; |
b0336e8d DW |
1696 | |
1697 | if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
1698 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
1699 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
ac27a0ec | 1700 | |
ac27a0ec | 1701 | if (!de) { |
978fef91 TM |
1702 | err = ext4_find_dest_de(dir, inode, |
1703 | bh, bh->b_data, blocksize - csum_size, | |
1704 | name, namelen, &de); | |
1705 | if (err) | |
1706 | return err; | |
ac27a0ec DK |
1707 | } |
1708 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1709 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1710 | if (err) { |
617ba13b | 1711 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1712 | return err; |
1713 | } | |
1714 | ||
1715 | /* By now the buffer is marked for journaling */ | |
978fef91 TM |
1716 | ext4_insert_dentry(inode, de, blocksize, name, namelen); |
1717 | ||
ac27a0ec DK |
1718 | /* |
1719 | * XXX shouldn't update any times until successful | |
1720 | * completion of syscall, but too many callers depend | |
1721 | * on this. | |
1722 | * | |
1723 | * XXX similarly, too many callers depend on | |
617ba13b | 1724 | * ext4_new_inode() setting the times, but error |
ac27a0ec DK |
1725 | * recovery deletes the inode, so the worst that can |
1726 | * happen is that the times are slightly out of date | |
1727 | * and/or different from the directory change time. | |
1728 | */ | |
ef7f3835 | 1729 | dir->i_mtime = dir->i_ctime = ext4_current_time(dir); |
617ba13b | 1730 | ext4_update_dx_flag(dir); |
ac27a0ec | 1731 | dir->i_version++; |
617ba13b | 1732 | ext4_mark_inode_dirty(handle, dir); |
0390131b | 1733 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
b0336e8d | 1734 | err = ext4_handle_dirty_dirent_node(handle, dir, bh); |
ac27a0ec | 1735 | if (err) |
617ba13b | 1736 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1737 | return 0; |
1738 | } | |
1739 | ||
ac27a0ec DK |
1740 | /* |
1741 | * This converts a one block unindexed directory to a 3 block indexed | |
1742 | * directory, and adds the dentry to the indexed directory. | |
1743 | */ | |
1744 | static int make_indexed_dir(handle_t *handle, struct dentry *dentry, | |
1745 | struct inode *inode, struct buffer_head *bh) | |
1746 | { | |
1747 | struct inode *dir = dentry->d_parent->d_inode; | |
1748 | const char *name = dentry->d_name.name; | |
1749 | int namelen = dentry->d_name.len; | |
1750 | struct buffer_head *bh2; | |
1751 | struct dx_root *root; | |
1752 | struct dx_frame frames[2], *frame; | |
1753 | struct dx_entry *entries; | |
617ba13b | 1754 | struct ext4_dir_entry_2 *de, *de2; |
b0336e8d | 1755 | struct ext4_dir_entry_tail *t; |
ac27a0ec DK |
1756 | char *data1, *top; |
1757 | unsigned len; | |
1758 | int retval; | |
1759 | unsigned blocksize; | |
1760 | struct dx_hash_info hinfo; | |
725d26d3 | 1761 | ext4_lblk_t block; |
ac27a0ec | 1762 | struct fake_dirent *fde; |
b0336e8d DW |
1763 | int csum_size = 0; |
1764 | ||
1765 | if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
1766 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
1767 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
ac27a0ec DK |
1768 | |
1769 | blocksize = dir->i_sb->s_blocksize; | |
e6b8bc09 | 1770 | dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino)); |
5d601255 | 1771 | BUFFER_TRACE(bh, "get_write_access"); |
617ba13b | 1772 | retval = ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1773 | if (retval) { |
617ba13b | 1774 | ext4_std_error(dir->i_sb, retval); |
ac27a0ec DK |
1775 | brelse(bh); |
1776 | return retval; | |
1777 | } | |
1778 | root = (struct dx_root *) bh->b_data; | |
1779 | ||
e6b8bc09 TT |
1780 | /* The 0th block becomes the root, move the dirents out */ |
1781 | fde = &root->dotdot; | |
1782 | de = (struct ext4_dir_entry_2 *)((char *)fde + | |
3d0518f4 | 1783 | ext4_rec_len_from_disk(fde->rec_len, blocksize)); |
e6b8bc09 | 1784 | if ((char *) de >= (((char *) root) + blocksize)) { |
24676da4 | 1785 | EXT4_ERROR_INODE(dir, "invalid rec_len for '..'"); |
e6b8bc09 TT |
1786 | brelse(bh); |
1787 | return -EIO; | |
1788 | } | |
b0336e8d | 1789 | len = ((char *) root) + (blocksize - csum_size) - (char *) de; |
e6b8bc09 TT |
1790 | |
1791 | /* Allocate new block for the 0th block's dirents */ | |
0f70b406 TT |
1792 | bh2 = ext4_append(handle, dir, &block); |
1793 | if (IS_ERR(bh2)) { | |
ac27a0ec | 1794 | brelse(bh); |
0f70b406 | 1795 | return PTR_ERR(bh2); |
ac27a0ec | 1796 | } |
12e9b892 | 1797 | ext4_set_inode_flag(dir, EXT4_INODE_INDEX); |
ac27a0ec DK |
1798 | data1 = bh2->b_data; |
1799 | ||
ac27a0ec | 1800 | memcpy (data1, de, len); |
617ba13b | 1801 | de = (struct ext4_dir_entry_2 *) data1; |
ac27a0ec | 1802 | top = data1 + len; |
3d0518f4 | 1803 | while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) |
ac27a0ec | 1804 | de = de2; |
b0336e8d DW |
1805 | de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - |
1806 | (char *) de, | |
3d0518f4 | 1807 | blocksize); |
b0336e8d DW |
1808 | |
1809 | if (csum_size) { | |
1810 | t = EXT4_DIRENT_TAIL(data1, blocksize); | |
1811 | initialize_dirent_tail(t, blocksize); | |
1812 | } | |
1813 | ||
ac27a0ec | 1814 | /* Initialize the root; the dot dirents already exist */ |
617ba13b | 1815 | de = (struct ext4_dir_entry_2 *) (&root->dotdot); |
3d0518f4 WY |
1816 | de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2), |
1817 | blocksize); | |
ac27a0ec DK |
1818 | memset (&root->info, 0, sizeof(root->info)); |
1819 | root->info.info_length = sizeof(root->info); | |
617ba13b | 1820 | root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; |
ac27a0ec | 1821 | entries = root->entries; |
af5bc92d TT |
1822 | dx_set_block(entries, 1); |
1823 | dx_set_count(entries, 1); | |
1824 | dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); | |
ac27a0ec DK |
1825 | |
1826 | /* Initialize as for dx_probe */ | |
1827 | hinfo.hash_version = root->info.hash_version; | |
f99b2589 TT |
1828 | if (hinfo.hash_version <= DX_HASH_TEA) |
1829 | hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b MC |
1830 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
1831 | ext4fs_dirhash(name, namelen, &hinfo); | |
ac27a0ec DK |
1832 | frame = frames; |
1833 | frame->entries = entries; | |
1834 | frame->at = entries; | |
1835 | frame->bh = bh; | |
1836 | bh = bh2; | |
6976a6f2 | 1837 | |
dbe89444 | 1838 | ext4_handle_dirty_dx_node(handle, dir, frame->bh); |
b0336e8d | 1839 | ext4_handle_dirty_dirent_node(handle, dir, bh); |
6976a6f2 | 1840 | |
ac27a0ec | 1841 | de = do_split(handle,dir, &bh, frame, &hinfo, &retval); |
7ad8e4e6 JK |
1842 | if (!de) { |
1843 | /* | |
1844 | * Even if the block split failed, we have to properly write | |
1845 | * out all the changes we did so far. Otherwise we can end up | |
1846 | * with corrupted filesystem. | |
1847 | */ | |
1848 | ext4_mark_inode_dirty(handle, dir); | |
7ad8e4e6 | 1849 | dx_release(frames); |
ac27a0ec | 1850 | return retval; |
7ad8e4e6 JK |
1851 | } |
1852 | dx_release(frames); | |
ac27a0ec | 1853 | |
2de770a4 TT |
1854 | retval = add_dirent_to_buf(handle, dentry, inode, de, bh); |
1855 | brelse(bh); | |
1856 | return retval; | |
ac27a0ec | 1857 | } |
ac27a0ec DK |
1858 | |
1859 | /* | |
617ba13b | 1860 | * ext4_add_entry() |
ac27a0ec DK |
1861 | * |
1862 | * adds a file entry to the specified directory, using the same | |
617ba13b | 1863 | * semantics as ext4_find_entry(). It returns NULL if it failed. |
ac27a0ec DK |
1864 | * |
1865 | * NOTE!! The inode part of 'de' is left at 0 - which means you | |
1866 | * may not sleep between calling this and putting something into | |
1867 | * the entry, as someone else might have used it while you slept. | |
1868 | */ | |
af5bc92d TT |
1869 | static int ext4_add_entry(handle_t *handle, struct dentry *dentry, |
1870 | struct inode *inode) | |
ac27a0ec DK |
1871 | { |
1872 | struct inode *dir = dentry->d_parent->d_inode; | |
af5bc92d | 1873 | struct buffer_head *bh; |
617ba13b | 1874 | struct ext4_dir_entry_2 *de; |
b0336e8d | 1875 | struct ext4_dir_entry_tail *t; |
af5bc92d | 1876 | struct super_block *sb; |
ac27a0ec | 1877 | int retval; |
ac27a0ec | 1878 | int dx_fallback=0; |
ac27a0ec | 1879 | unsigned blocksize; |
725d26d3 | 1880 | ext4_lblk_t block, blocks; |
b0336e8d DW |
1881 | int csum_size = 0; |
1882 | ||
1883 | if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
1884 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
1885 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
ac27a0ec DK |
1886 | |
1887 | sb = dir->i_sb; | |
1888 | blocksize = sb->s_blocksize; | |
1889 | if (!dentry->d_name.len) | |
1890 | return -EINVAL; | |
3c47d541 TM |
1891 | |
1892 | if (ext4_has_inline_data(dir)) { | |
1893 | retval = ext4_try_add_inline_entry(handle, dentry, inode); | |
1894 | if (retval < 0) | |
1895 | return retval; | |
1896 | if (retval == 1) { | |
1897 | retval = 0; | |
1898 | return retval; | |
1899 | } | |
1900 | } | |
1901 | ||
ac27a0ec | 1902 | if (is_dx(dir)) { |
617ba13b | 1903 | retval = ext4_dx_add_entry(handle, dentry, inode); |
ac27a0ec DK |
1904 | if (!retval || (retval != ERR_BAD_DX_DIR)) |
1905 | return retval; | |
12e9b892 | 1906 | ext4_clear_inode_flag(dir, EXT4_INODE_INDEX); |
ac27a0ec | 1907 | dx_fallback++; |
617ba13b | 1908 | ext4_mark_inode_dirty(handle, dir); |
ac27a0ec | 1909 | } |
ac27a0ec | 1910 | blocks = dir->i_size >> sb->s_blocksize_bits; |
498e5f24 | 1911 | for (block = 0; block < blocks; block++) { |
dc6982ff TT |
1912 | bh = ext4_read_dirblock(dir, block, DIRENT); |
1913 | if (IS_ERR(bh)) | |
1914 | return PTR_ERR(bh); | |
1915 | ||
ac27a0ec | 1916 | retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh); |
2de770a4 TT |
1917 | if (retval != -ENOSPC) { |
1918 | brelse(bh); | |
ac27a0ec | 1919 | return retval; |
2de770a4 | 1920 | } |
ac27a0ec | 1921 | |
ac27a0ec | 1922 | if (blocks == 1 && !dx_fallback && |
1e424a34 TT |
1923 | EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) |
1924 | return make_indexed_dir(handle, dentry, inode, bh); | |
ac27a0ec DK |
1925 | brelse(bh); |
1926 | } | |
0f70b406 TT |
1927 | bh = ext4_append(handle, dir, &block); |
1928 | if (IS_ERR(bh)) | |
1929 | return PTR_ERR(bh); | |
617ba13b | 1930 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 1931 | de->inode = 0; |
b0336e8d DW |
1932 | de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize); |
1933 | ||
1934 | if (csum_size) { | |
1935 | t = EXT4_DIRENT_TAIL(bh->b_data, blocksize); | |
1936 | initialize_dirent_tail(t, blocksize); | |
1937 | } | |
1938 | ||
2de770a4 TT |
1939 | retval = add_dirent_to_buf(handle, dentry, inode, de, bh); |
1940 | brelse(bh); | |
14ece102 FM |
1941 | if (retval == 0) |
1942 | ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); | |
2de770a4 | 1943 | return retval; |
ac27a0ec DK |
1944 | } |
1945 | ||
ac27a0ec DK |
1946 | /* |
1947 | * Returns 0 for success, or a negative error value | |
1948 | */ | |
617ba13b | 1949 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, |
ac27a0ec DK |
1950 | struct inode *inode) |
1951 | { | |
1952 | struct dx_frame frames[2], *frame; | |
1953 | struct dx_entry *entries, *at; | |
1954 | struct dx_hash_info hinfo; | |
af5bc92d | 1955 | struct buffer_head *bh; |
ac27a0ec | 1956 | struct inode *dir = dentry->d_parent->d_inode; |
af5bc92d | 1957 | struct super_block *sb = dir->i_sb; |
617ba13b | 1958 | struct ext4_dir_entry_2 *de; |
ac27a0ec DK |
1959 | int err; |
1960 | ||
dd73b5d5 TT |
1961 | frame = dx_probe(&dentry->d_name, dir, &hinfo, frames); |
1962 | if (IS_ERR(frame)) | |
1963 | return PTR_ERR(frame); | |
ac27a0ec DK |
1964 | entries = frame->entries; |
1965 | at = frame->at; | |
dc6982ff TT |
1966 | bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT); |
1967 | if (IS_ERR(bh)) { | |
1968 | err = PTR_ERR(bh); | |
1969 | bh = NULL; | |
ac27a0ec | 1970 | goto cleanup; |
6d1ab10e | 1971 | } |
ac27a0ec DK |
1972 | |
1973 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1974 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1975 | if (err) |
1976 | goto journal_error; | |
1977 | ||
1978 | err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
2de770a4 | 1979 | if (err != -ENOSPC) |
ac27a0ec | 1980 | goto cleanup; |
ac27a0ec DK |
1981 | |
1982 | /* Block full, should compress but for now just split */ | |
4776004f | 1983 | dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", |
ac27a0ec DK |
1984 | dx_get_count(entries), dx_get_limit(entries))); |
1985 | /* Need to split index? */ | |
1986 | if (dx_get_count(entries) == dx_get_limit(entries)) { | |
725d26d3 | 1987 | ext4_lblk_t newblock; |
ac27a0ec DK |
1988 | unsigned icount = dx_get_count(entries); |
1989 | int levels = frame - frames; | |
1990 | struct dx_entry *entries2; | |
1991 | struct dx_node *node2; | |
1992 | struct buffer_head *bh2; | |
1993 | ||
1994 | if (levels && (dx_get_count(frames->entries) == | |
1995 | dx_get_limit(frames->entries))) { | |
12062ddd | 1996 | ext4_warning(sb, "Directory index full!"); |
ac27a0ec DK |
1997 | err = -ENOSPC; |
1998 | goto cleanup; | |
1999 | } | |
0f70b406 TT |
2000 | bh2 = ext4_append(handle, dir, &newblock); |
2001 | if (IS_ERR(bh2)) { | |
2002 | err = PTR_ERR(bh2); | |
ac27a0ec | 2003 | goto cleanup; |
0f70b406 | 2004 | } |
ac27a0ec DK |
2005 | node2 = (struct dx_node *)(bh2->b_data); |
2006 | entries2 = node2->entries; | |
1f7bebb9 | 2007 | memset(&node2->fake, 0, sizeof(struct fake_dirent)); |
3d0518f4 WY |
2008 | node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize, |
2009 | sb->s_blocksize); | |
ac27a0ec | 2010 | BUFFER_TRACE(frame->bh, "get_write_access"); |
617ba13b | 2011 | err = ext4_journal_get_write_access(handle, frame->bh); |
ac27a0ec DK |
2012 | if (err) |
2013 | goto journal_error; | |
2014 | if (levels) { | |
2015 | unsigned icount1 = icount/2, icount2 = icount - icount1; | |
2016 | unsigned hash2 = dx_get_hash(entries + icount1); | |
4776004f TT |
2017 | dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", |
2018 | icount1, icount2)); | |
ac27a0ec DK |
2019 | |
2020 | BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ | |
617ba13b | 2021 | err = ext4_journal_get_write_access(handle, |
ac27a0ec DK |
2022 | frames[0].bh); |
2023 | if (err) | |
2024 | goto journal_error; | |
2025 | ||
af5bc92d TT |
2026 | memcpy((char *) entries2, (char *) (entries + icount1), |
2027 | icount2 * sizeof(struct dx_entry)); | |
2028 | dx_set_count(entries, icount1); | |
2029 | dx_set_count(entries2, icount2); | |
2030 | dx_set_limit(entries2, dx_node_limit(dir)); | |
ac27a0ec DK |
2031 | |
2032 | /* Which index block gets the new entry? */ | |
2033 | if (at - entries >= icount1) { | |
2034 | frame->at = at = at - entries - icount1 + entries2; | |
2035 | frame->entries = entries = entries2; | |
2036 | swap(frame->bh, bh2); | |
2037 | } | |
af5bc92d TT |
2038 | dx_insert_block(frames + 0, hash2, newblock); |
2039 | dxtrace(dx_show_index("node", frames[1].entries)); | |
2040 | dxtrace(dx_show_index("node", | |
ac27a0ec | 2041 | ((struct dx_node *) bh2->b_data)->entries)); |
dbe89444 | 2042 | err = ext4_handle_dirty_dx_node(handle, dir, bh2); |
ac27a0ec DK |
2043 | if (err) |
2044 | goto journal_error; | |
2045 | brelse (bh2); | |
2046 | } else { | |
4776004f TT |
2047 | dxtrace(printk(KERN_DEBUG |
2048 | "Creating second level index...\n")); | |
ac27a0ec DK |
2049 | memcpy((char *) entries2, (char *) entries, |
2050 | icount * sizeof(struct dx_entry)); | |
2051 | dx_set_limit(entries2, dx_node_limit(dir)); | |
2052 | ||
2053 | /* Set up root */ | |
2054 | dx_set_count(entries, 1); | |
2055 | dx_set_block(entries + 0, newblock); | |
2056 | ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; | |
2057 | ||
2058 | /* Add new access path frame */ | |
2059 | frame = frames + 1; | |
2060 | frame->at = at = at - entries + entries2; | |
2061 | frame->entries = entries = entries2; | |
2062 | frame->bh = bh2; | |
617ba13b | 2063 | err = ext4_journal_get_write_access(handle, |
ac27a0ec DK |
2064 | frame->bh); |
2065 | if (err) | |
2066 | goto journal_error; | |
2067 | } | |
dbe89444 | 2068 | err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh); |
b4097142 TT |
2069 | if (err) { |
2070 | ext4_std_error(inode->i_sb, err); | |
2071 | goto cleanup; | |
2072 | } | |
ac27a0ec DK |
2073 | } |
2074 | de = do_split(handle, dir, &bh, frame, &hinfo, &err); | |
2075 | if (!de) | |
2076 | goto cleanup; | |
2077 | err = add_dirent_to_buf(handle, dentry, inode, de, bh); | |
ac27a0ec DK |
2078 | goto cleanup; |
2079 | ||
2080 | journal_error: | |
617ba13b | 2081 | ext4_std_error(dir->i_sb, err); |
ac27a0ec | 2082 | cleanup: |
b1deefc9 | 2083 | brelse(bh); |
ac27a0ec DK |
2084 | dx_release(frames); |
2085 | return err; | |
2086 | } | |
ac27a0ec DK |
2087 | |
2088 | /* | |
05019a9e TM |
2089 | * ext4_generic_delete_entry deletes a directory entry by merging it |
2090 | * with the previous entry | |
ac27a0ec | 2091 | */ |
05019a9e TM |
2092 | int ext4_generic_delete_entry(handle_t *handle, |
2093 | struct inode *dir, | |
2094 | struct ext4_dir_entry_2 *de_del, | |
2095 | struct buffer_head *bh, | |
2096 | void *entry_buf, | |
2097 | int buf_size, | |
2098 | int csum_size) | |
ac27a0ec | 2099 | { |
af5bc92d | 2100 | struct ext4_dir_entry_2 *de, *pde; |
3d0518f4 | 2101 | unsigned int blocksize = dir->i_sb->s_blocksize; |
05019a9e | 2102 | int i; |
b0336e8d | 2103 | |
ac27a0ec DK |
2104 | i = 0; |
2105 | pde = NULL; | |
05019a9e TM |
2106 | de = (struct ext4_dir_entry_2 *)entry_buf; |
2107 | while (i < buf_size - csum_size) { | |
226ba972 TM |
2108 | if (ext4_check_dir_entry(dir, NULL, de, bh, |
2109 | bh->b_data, bh->b_size, i)) | |
ac27a0ec DK |
2110 | return -EIO; |
2111 | if (de == de_del) { | |
ac27a0ec | 2112 | if (pde) |
a72d7f83 | 2113 | pde->rec_len = ext4_rec_len_to_disk( |
3d0518f4 WY |
2114 | ext4_rec_len_from_disk(pde->rec_len, |
2115 | blocksize) + | |
2116 | ext4_rec_len_from_disk(de->rec_len, | |
2117 | blocksize), | |
2118 | blocksize); | |
ac27a0ec DK |
2119 | else |
2120 | de->inode = 0; | |
2121 | dir->i_version++; | |
ac27a0ec DK |
2122 | return 0; |
2123 | } | |
3d0518f4 | 2124 | i += ext4_rec_len_from_disk(de->rec_len, blocksize); |
ac27a0ec | 2125 | pde = de; |
3d0518f4 | 2126 | de = ext4_next_entry(de, blocksize); |
ac27a0ec DK |
2127 | } |
2128 | return -ENOENT; | |
2129 | } | |
2130 | ||
05019a9e TM |
2131 | static int ext4_delete_entry(handle_t *handle, |
2132 | struct inode *dir, | |
2133 | struct ext4_dir_entry_2 *de_del, | |
2134 | struct buffer_head *bh) | |
2135 | { | |
2136 | int err, csum_size = 0; | |
2137 | ||
9f40fe54 TM |
2138 | if (ext4_has_inline_data(dir)) { |
2139 | int has_inline_data = 1; | |
2140 | err = ext4_delete_inline_entry(handle, dir, de_del, bh, | |
2141 | &has_inline_data); | |
2142 | if (has_inline_data) | |
2143 | return err; | |
2144 | } | |
2145 | ||
05019a9e TM |
2146 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, |
2147 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
2148 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
2149 | ||
2150 | BUFFER_TRACE(bh, "get_write_access"); | |
2151 | err = ext4_journal_get_write_access(handle, bh); | |
2152 | if (unlikely(err)) | |
2153 | goto out; | |
2154 | ||
2155 | err = ext4_generic_delete_entry(handle, dir, de_del, | |
2156 | bh, bh->b_data, | |
2157 | dir->i_sb->s_blocksize, csum_size); | |
2158 | if (err) | |
2159 | goto out; | |
2160 | ||
2161 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
2162 | err = ext4_handle_dirty_dirent_node(handle, dir, bh); | |
2163 | if (unlikely(err)) | |
2164 | goto out; | |
2165 | ||
2166 | return 0; | |
2167 | out: | |
2168 | if (err != -ENOENT) | |
2169 | ext4_std_error(dir->i_sb, err); | |
2170 | return err; | |
2171 | } | |
2172 | ||
f8628a14 AD |
2173 | /* |
2174 | * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2, | |
2175 | * since this indicates that nlinks count was previously 1. | |
2176 | */ | |
2177 | static void ext4_inc_count(handle_t *handle, struct inode *inode) | |
2178 | { | |
2179 | inc_nlink(inode); | |
2180 | if (is_dx(inode) && inode->i_nlink > 1) { | |
2181 | /* limit is 16-bit i_links_count */ | |
2182 | if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) { | |
bfe86848 | 2183 | set_nlink(inode, 1); |
f8628a14 AD |
2184 | EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb, |
2185 | EXT4_FEATURE_RO_COMPAT_DIR_NLINK); | |
2186 | } | |
2187 | } | |
2188 | } | |
2189 | ||
2190 | /* | |
2191 | * If a directory had nlink == 1, then we should let it be 1. This indicates | |
2192 | * directory has >EXT4_LINK_MAX subdirs. | |
2193 | */ | |
2194 | static void ext4_dec_count(handle_t *handle, struct inode *inode) | |
2195 | { | |
909a4cf1 AD |
2196 | if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2) |
2197 | drop_nlink(inode); | |
f8628a14 AD |
2198 | } |
2199 | ||
2200 | ||
617ba13b | 2201 | static int ext4_add_nondir(handle_t *handle, |
ac27a0ec DK |
2202 | struct dentry *dentry, struct inode *inode) |
2203 | { | |
617ba13b | 2204 | int err = ext4_add_entry(handle, dentry, inode); |
ac27a0ec | 2205 | if (!err) { |
617ba13b | 2206 | ext4_mark_inode_dirty(handle, inode); |
6b38e842 | 2207 | unlock_new_inode(inode); |
8fc37ec5 | 2208 | d_instantiate(dentry, inode); |
ac27a0ec DK |
2209 | return 0; |
2210 | } | |
731b9a54 | 2211 | drop_nlink(inode); |
6b38e842 | 2212 | unlock_new_inode(inode); |
ac27a0ec DK |
2213 | iput(inode); |
2214 | return err; | |
2215 | } | |
2216 | ||
2217 | /* | |
2218 | * By the time this is called, we already have created | |
2219 | * the directory cache entry for the new file, but it | |
2220 | * is so far negative - it has no inode. | |
2221 | * | |
2222 | * If the create succeeds, we fill in the inode information | |
2223 | * with d_instantiate(). | |
2224 | */ | |
4acdaf27 | 2225 | static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2226 | bool excl) |
ac27a0ec DK |
2227 | { |
2228 | handle_t *handle; | |
af5bc92d | 2229 | struct inode *inode; |
1139575a | 2230 | int err, credits, retries = 0; |
ac27a0ec | 2231 | |
871a2931 | 2232 | dquot_initialize(dir); |
907f4554 | 2233 | |
1139575a | 2234 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
eb9cc7e1 | 2235 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); |
ac27a0ec | 2236 | retry: |
1139575a TT |
2237 | inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, |
2238 | NULL, EXT4_HT_DIR, credits); | |
2239 | handle = ext4_journal_current_handle(); | |
ac27a0ec DK |
2240 | err = PTR_ERR(inode); |
2241 | if (!IS_ERR(inode)) { | |
617ba13b MC |
2242 | inode->i_op = &ext4_file_inode_operations; |
2243 | inode->i_fop = &ext4_file_operations; | |
2244 | ext4_set_aops(inode); | |
2245 | err = ext4_add_nondir(handle, dentry, inode); | |
1139575a TT |
2246 | if (!err && IS_DIRSYNC(dir)) |
2247 | ext4_handle_sync(handle); | |
ac27a0ec | 2248 | } |
1139575a TT |
2249 | if (handle) |
2250 | ext4_journal_stop(handle); | |
617ba13b | 2251 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) |
ac27a0ec DK |
2252 | goto retry; |
2253 | return err; | |
2254 | } | |
2255 | ||
af5bc92d | 2256 | static int ext4_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 2257 | umode_t mode, dev_t rdev) |
ac27a0ec DK |
2258 | { |
2259 | handle_t *handle; | |
2260 | struct inode *inode; | |
1139575a | 2261 | int err, credits, retries = 0; |
ac27a0ec DK |
2262 | |
2263 | if (!new_valid_dev(rdev)) | |
2264 | return -EINVAL; | |
2265 | ||
871a2931 | 2266 | dquot_initialize(dir); |
907f4554 | 2267 | |
1139575a | 2268 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
eb9cc7e1 | 2269 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); |
ac27a0ec | 2270 | retry: |
1139575a TT |
2271 | inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, |
2272 | NULL, EXT4_HT_DIR, credits); | |
2273 | handle = ext4_journal_current_handle(); | |
ac27a0ec DK |
2274 | err = PTR_ERR(inode); |
2275 | if (!IS_ERR(inode)) { | |
2276 | init_special_inode(inode, inode->i_mode, rdev); | |
617ba13b | 2277 | inode->i_op = &ext4_special_inode_operations; |
617ba13b | 2278 | err = ext4_add_nondir(handle, dentry, inode); |
1139575a TT |
2279 | if (!err && IS_DIRSYNC(dir)) |
2280 | ext4_handle_sync(handle); | |
ac27a0ec | 2281 | } |
1139575a TT |
2282 | if (handle) |
2283 | ext4_journal_stop(handle); | |
617ba13b | 2284 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) |
ac27a0ec DK |
2285 | goto retry; |
2286 | return err; | |
2287 | } | |
2288 | ||
af51a2ac AV |
2289 | static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
2290 | { | |
2291 | handle_t *handle; | |
2292 | struct inode *inode; | |
2293 | int err, retries = 0; | |
2294 | ||
2295 | dquot_initialize(dir); | |
2296 | ||
2297 | retry: | |
2298 | inode = ext4_new_inode_start_handle(dir, mode, | |
2299 | NULL, 0, NULL, | |
2300 | EXT4_HT_DIR, | |
2301 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + | |
2302 | 4 + EXT4_XATTR_TRANS_BLOCKS); | |
2303 | handle = ext4_journal_current_handle(); | |
2304 | err = PTR_ERR(inode); | |
2305 | if (!IS_ERR(inode)) { | |
2306 | inode->i_op = &ext4_file_inode_operations; | |
2307 | inode->i_fop = &ext4_file_operations; | |
2308 | ext4_set_aops(inode); | |
e94bd349 | 2309 | d_tmpfile(dentry, inode); |
af51a2ac AV |
2310 | err = ext4_orphan_add(handle, inode); |
2311 | if (err) | |
43ae9e3f | 2312 | goto err_unlock_inode; |
af51a2ac | 2313 | mark_inode_dirty(inode); |
af51a2ac AV |
2314 | unlock_new_inode(inode); |
2315 | } | |
2316 | if (handle) | |
2317 | ext4_journal_stop(handle); | |
2318 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
2319 | goto retry; | |
2320 | return err; | |
43ae9e3f | 2321 | err_unlock_inode: |
af51a2ac AV |
2322 | ext4_journal_stop(handle); |
2323 | unlock_new_inode(inode); | |
af51a2ac AV |
2324 | return err; |
2325 | } | |
2326 | ||
a774f9c2 TM |
2327 | struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode, |
2328 | struct ext4_dir_entry_2 *de, | |
2329 | int blocksize, int csum_size, | |
2330 | unsigned int parent_ino, int dotdot_real_len) | |
2331 | { | |
2332 | de->inode = cpu_to_le32(inode->i_ino); | |
2333 | de->name_len = 1; | |
2334 | de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len), | |
2335 | blocksize); | |
2336 | strcpy(de->name, "."); | |
2337 | ext4_set_de_type(inode->i_sb, de, S_IFDIR); | |
2338 | ||
2339 | de = ext4_next_entry(de, blocksize); | |
2340 | de->inode = cpu_to_le32(parent_ino); | |
2341 | de->name_len = 2; | |
2342 | if (!dotdot_real_len) | |
2343 | de->rec_len = ext4_rec_len_to_disk(blocksize - | |
2344 | (csum_size + EXT4_DIR_REC_LEN(1)), | |
2345 | blocksize); | |
2346 | else | |
2347 | de->rec_len = ext4_rec_len_to_disk( | |
2348 | EXT4_DIR_REC_LEN(de->name_len), blocksize); | |
2349 | strcpy(de->name, ".."); | |
2350 | ext4_set_de_type(inode->i_sb, de, S_IFDIR); | |
2351 | ||
2352 | return ext4_next_entry(de, blocksize); | |
2353 | } | |
2354 | ||
2355 | static int ext4_init_new_dir(handle_t *handle, struct inode *dir, | |
2356 | struct inode *inode) | |
ac27a0ec | 2357 | { |
dabd991f | 2358 | struct buffer_head *dir_block = NULL; |
af5bc92d | 2359 | struct ext4_dir_entry_2 *de; |
b0336e8d | 2360 | struct ext4_dir_entry_tail *t; |
dc6982ff | 2361 | ext4_lblk_t block = 0; |
3d0518f4 | 2362 | unsigned int blocksize = dir->i_sb->s_blocksize; |
b0336e8d | 2363 | int csum_size = 0; |
a774f9c2 | 2364 | int err; |
ac27a0ec | 2365 | |
b0336e8d DW |
2366 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, |
2367 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
2368 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
2369 | ||
3c47d541 TM |
2370 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
2371 | err = ext4_try_create_inline_dir(handle, dir, inode); | |
2372 | if (err < 0 && err != -ENOSPC) | |
2373 | goto out; | |
2374 | if (!err) | |
2375 | goto out; | |
2376 | } | |
2377 | ||
dc6982ff | 2378 | inode->i_size = 0; |
0f70b406 TT |
2379 | dir_block = ext4_append(handle, inode, &block); |
2380 | if (IS_ERR(dir_block)) | |
2381 | return PTR_ERR(dir_block); | |
a774f9c2 TM |
2382 | BUFFER_TRACE(dir_block, "get_write_access"); |
2383 | err = ext4_journal_get_write_access(handle, dir_block); | |
2384 | if (err) | |
2385 | goto out; | |
2386 | de = (struct ext4_dir_entry_2 *)dir_block->b_data; | |
2387 | ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0); | |
2388 | set_nlink(inode, 2); | |
2389 | if (csum_size) { | |
2390 | t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize); | |
2391 | initialize_dirent_tail(t, blocksize); | |
2392 | } | |
2393 | ||
2394 | BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); | |
2395 | err = ext4_handle_dirty_dirent_node(handle, inode, dir_block); | |
2396 | if (err) | |
2397 | goto out; | |
2398 | set_buffer_verified(dir_block); | |
2399 | out: | |
2400 | brelse(dir_block); | |
2401 | return err; | |
2402 | } | |
2403 | ||
2404 | static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2405 | { | |
2406 | handle_t *handle; | |
2407 | struct inode *inode; | |
1139575a | 2408 | int err, credits, retries = 0; |
a774f9c2 | 2409 | |
f8628a14 | 2410 | if (EXT4_DIR_LINK_MAX(dir)) |
ac27a0ec DK |
2411 | return -EMLINK; |
2412 | ||
871a2931 | 2413 | dquot_initialize(dir); |
907f4554 | 2414 | |
1139575a | 2415 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
eb9cc7e1 | 2416 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3); |
ac27a0ec | 2417 | retry: |
1139575a TT |
2418 | inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode, |
2419 | &dentry->d_name, | |
2420 | 0, NULL, EXT4_HT_DIR, credits); | |
2421 | handle = ext4_journal_current_handle(); | |
ac27a0ec DK |
2422 | err = PTR_ERR(inode); |
2423 | if (IS_ERR(inode)) | |
2424 | goto out_stop; | |
2425 | ||
617ba13b MC |
2426 | inode->i_op = &ext4_dir_inode_operations; |
2427 | inode->i_fop = &ext4_dir_operations; | |
a774f9c2 | 2428 | err = ext4_init_new_dir(handle, dir, inode); |
dabd991f NK |
2429 | if (err) |
2430 | goto out_clear_inode; | |
2431 | err = ext4_mark_inode_dirty(handle, inode); | |
2432 | if (!err) | |
2433 | err = ext4_add_entry(handle, dentry, inode); | |
ac27a0ec | 2434 | if (err) { |
4cdeed86 AK |
2435 | out_clear_inode: |
2436 | clear_nlink(inode); | |
6b38e842 | 2437 | unlock_new_inode(inode); |
617ba13b | 2438 | ext4_mark_inode_dirty(handle, inode); |
af5bc92d | 2439 | iput(inode); |
ac27a0ec DK |
2440 | goto out_stop; |
2441 | } | |
f8628a14 | 2442 | ext4_inc_count(handle, dir); |
617ba13b | 2443 | ext4_update_dx_flag(dir); |
dabd991f NK |
2444 | err = ext4_mark_inode_dirty(handle, dir); |
2445 | if (err) | |
2446 | goto out_clear_inode; | |
6b38e842 | 2447 | unlock_new_inode(inode); |
8fc37ec5 | 2448 | d_instantiate(dentry, inode); |
1139575a TT |
2449 | if (IS_DIRSYNC(dir)) |
2450 | ext4_handle_sync(handle); | |
2451 | ||
ac27a0ec | 2452 | out_stop: |
1139575a TT |
2453 | if (handle) |
2454 | ext4_journal_stop(handle); | |
617ba13b | 2455 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) |
ac27a0ec DK |
2456 | goto retry; |
2457 | return err; | |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * routine to check that the specified directory is empty (for rmdir) | |
2462 | */ | |
af5bc92d | 2463 | static int empty_dir(struct inode *inode) |
ac27a0ec | 2464 | { |
498e5f24 | 2465 | unsigned int offset; |
af5bc92d TT |
2466 | struct buffer_head *bh; |
2467 | struct ext4_dir_entry_2 *de, *de1; | |
2468 | struct super_block *sb; | |
ac27a0ec DK |
2469 | int err = 0; |
2470 | ||
61f86638 TM |
2471 | if (ext4_has_inline_data(inode)) { |
2472 | int has_inline_data = 1; | |
2473 | ||
2474 | err = empty_inline_dir(inode, &has_inline_data); | |
2475 | if (has_inline_data) | |
2476 | return err; | |
2477 | } | |
2478 | ||
ac27a0ec | 2479 | sb = inode->i_sb; |
dc6982ff TT |
2480 | if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) { |
2481 | EXT4_ERROR_INODE(inode, "invalid size"); | |
ac27a0ec DK |
2482 | return 1; |
2483 | } | |
dc6982ff TT |
2484 | bh = ext4_read_dirblock(inode, 0, EITHER); |
2485 | if (IS_ERR(bh)) | |
2486 | return 1; | |
2487 | ||
617ba13b | 2488 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
3d0518f4 | 2489 | de1 = ext4_next_entry(de, sb->s_blocksize); |
ac27a0ec DK |
2490 | if (le32_to_cpu(de->inode) != inode->i_ino || |
2491 | !le32_to_cpu(de1->inode) || | |
af5bc92d TT |
2492 | strcmp(".", de->name) || |
2493 | strcmp("..", de1->name)) { | |
12062ddd | 2494 | ext4_warning(inode->i_sb, |
af5bc92d TT |
2495 | "bad directory (dir #%lu) - no `.' or `..'", |
2496 | inode->i_ino); | |
2497 | brelse(bh); | |
ac27a0ec DK |
2498 | return 1; |
2499 | } | |
3d0518f4 WY |
2500 | offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) + |
2501 | ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize); | |
2502 | de = ext4_next_entry(de1, sb->s_blocksize); | |
af5bc92d | 2503 | while (offset < inode->i_size) { |
236f5ecb | 2504 | if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { |
24676da4 | 2505 | unsigned int lblock; |
ac27a0ec | 2506 | err = 0; |
af5bc92d | 2507 | brelse(bh); |
24676da4 | 2508 | lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb); |
dc6982ff TT |
2509 | bh = ext4_read_dirblock(inode, lblock, EITHER); |
2510 | if (IS_ERR(bh)) | |
2511 | return 1; | |
617ba13b | 2512 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 2513 | } |
226ba972 TM |
2514 | if (ext4_check_dir_entry(inode, NULL, de, bh, |
2515 | bh->b_data, bh->b_size, offset)) { | |
617ba13b | 2516 | de = (struct ext4_dir_entry_2 *)(bh->b_data + |
ac27a0ec DK |
2517 | sb->s_blocksize); |
2518 | offset = (offset | (sb->s_blocksize - 1)) + 1; | |
2519 | continue; | |
2520 | } | |
2521 | if (le32_to_cpu(de->inode)) { | |
af5bc92d | 2522 | brelse(bh); |
ac27a0ec DK |
2523 | return 0; |
2524 | } | |
3d0518f4 WY |
2525 | offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); |
2526 | de = ext4_next_entry(de, sb->s_blocksize); | |
ac27a0ec | 2527 | } |
af5bc92d | 2528 | brelse(bh); |
ac27a0ec DK |
2529 | return 1; |
2530 | } | |
2531 | ||
d745a8c2 JK |
2532 | /* |
2533 | * ext4_orphan_add() links an unlinked or truncated inode into a list of | |
ac27a0ec DK |
2534 | * such inodes, starting at the superblock, in case we crash before the |
2535 | * file is closed/deleted, or in case the inode truncate spans multiple | |
2536 | * transactions and the last transaction is not recovered after a crash. | |
2537 | * | |
2538 | * At filesystem recovery time, we walk this list deleting unlinked | |
617ba13b | 2539 | * inodes and truncating linked inodes in ext4_orphan_cleanup(). |
d745a8c2 JK |
2540 | * |
2541 | * Orphan list manipulation functions must be called under i_mutex unless | |
2542 | * we are just creating the inode or deleting it. | |
ac27a0ec | 2543 | */ |
617ba13b | 2544 | int ext4_orphan_add(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
2545 | { |
2546 | struct super_block *sb = inode->i_sb; | |
cd2c080c | 2547 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
617ba13b | 2548 | struct ext4_iloc iloc; |
ac27a0ec | 2549 | int err = 0, rc; |
d745a8c2 | 2550 | bool dirty = false; |
ac27a0ec | 2551 | |
cd2c080c | 2552 | if (!sbi->s_journal) |
0390131b FM |
2553 | return 0; |
2554 | ||
d745a8c2 JK |
2555 | WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && |
2556 | !mutex_is_locked(&inode->i_mutex)); | |
2557 | /* | |
2558 | * Exit early if inode already is on orphan list. This is a big speedup | |
2559 | * since we don't have to contend on the global s_orphan_lock. | |
2560 | */ | |
617ba13b | 2561 | if (!list_empty(&EXT4_I(inode)->i_orphan)) |
d745a8c2 | 2562 | return 0; |
ac27a0ec | 2563 | |
afb86178 LC |
2564 | /* |
2565 | * Orphan handling is only valid for files with data blocks | |
2566 | * being truncated, or files being unlinked. Note that we either | |
2567 | * hold i_mutex, or the inode can not be referenced from outside, | |
2568 | * so i_nlink should not be bumped due to race | |
ac27a0ec | 2569 | */ |
af5bc92d TT |
2570 | J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
2571 | S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); | |
ac27a0ec | 2572 | |
cd2c080c JK |
2573 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
2574 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); | |
ac27a0ec | 2575 | if (err) |
d745a8c2 | 2576 | goto out; |
ac27a0ec | 2577 | |
617ba13b | 2578 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
ac27a0ec | 2579 | if (err) |
d745a8c2 JK |
2580 | goto out; |
2581 | ||
2582 | mutex_lock(&sbi->s_orphan_lock); | |
6e3617e5 DM |
2583 | /* |
2584 | * Due to previous errors inode may be already a part of on-disk | |
2585 | * orphan list. If so skip on-disk list modification. | |
2586 | */ | |
d745a8c2 JK |
2587 | if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) > |
2588 | (le32_to_cpu(sbi->s_es->s_inodes_count))) { | |
2589 | /* Insert this inode at the head of the on-disk orphan list */ | |
2590 | NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan); | |
2591 | sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); | |
2592 | dirty = true; | |
2593 | } | |
2594 | list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan); | |
2595 | mutex_unlock(&sbi->s_orphan_lock); | |
ac27a0ec | 2596 | |
d745a8c2 JK |
2597 | if (dirty) { |
2598 | err = ext4_handle_dirty_super(handle, sb); | |
2599 | rc = ext4_mark_iloc_dirty(handle, inode, &iloc); | |
2600 | if (!err) | |
2601 | err = rc; | |
2602 | if (err) { | |
2603 | /* | |
2604 | * We have to remove inode from in-memory list if | |
2605 | * addition to on disk orphan list failed. Stray orphan | |
2606 | * list entries can cause panics at unmount time. | |
2607 | */ | |
2608 | mutex_lock(&sbi->s_orphan_lock); | |
2609 | list_del(&EXT4_I(inode)->i_orphan); | |
2610 | mutex_unlock(&sbi->s_orphan_lock); | |
2611 | } | |
2612 | } | |
ac27a0ec DK |
2613 | jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); |
2614 | jbd_debug(4, "orphan inode %lu will point to %d\n", | |
2615 | inode->i_ino, NEXT_ORPHAN(inode)); | |
d745a8c2 | 2616 | out: |
cd2c080c | 2617 | ext4_std_error(sb, err); |
ac27a0ec DK |
2618 | return err; |
2619 | } | |
2620 | ||
2621 | /* | |
617ba13b | 2622 | * ext4_orphan_del() removes an unlinked or truncated inode from the list |
ac27a0ec DK |
2623 | * of such inodes stored on disk, because it is finally being cleaned up. |
2624 | */ | |
617ba13b | 2625 | int ext4_orphan_del(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
2626 | { |
2627 | struct list_head *prev; | |
617ba13b | 2628 | struct ext4_inode_info *ei = EXT4_I(inode); |
cd2c080c | 2629 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
498e5f24 | 2630 | __u32 ino_next; |
617ba13b | 2631 | struct ext4_iloc iloc; |
ac27a0ec DK |
2632 | int err = 0; |
2633 | ||
cd2c080c | 2634 | if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS)) |
0390131b FM |
2635 | return 0; |
2636 | ||
d745a8c2 JK |
2637 | WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) && |
2638 | !mutex_is_locked(&inode->i_mutex)); | |
2639 | /* Do this quick check before taking global s_orphan_lock. */ | |
3b9d4ed2 | 2640 | if (list_empty(&ei->i_orphan)) |
d745a8c2 | 2641 | return 0; |
ac27a0ec | 2642 | |
d745a8c2 JK |
2643 | if (handle) { |
2644 | /* Grab inode buffer early before taking global s_orphan_lock */ | |
2645 | err = ext4_reserve_inode_write(handle, inode, &iloc); | |
2646 | } | |
ac27a0ec | 2647 | |
d745a8c2 | 2648 | mutex_lock(&sbi->s_orphan_lock); |
ac27a0ec DK |
2649 | jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); |
2650 | ||
d745a8c2 | 2651 | prev = ei->i_orphan.prev; |
ac27a0ec DK |
2652 | list_del_init(&ei->i_orphan); |
2653 | ||
2654 | /* If we're on an error path, we may not have a valid | |
2655 | * transaction handle with which to update the orphan list on | |
2656 | * disk, but we still need to remove the inode from the linked | |
2657 | * list in memory. */ | |
d745a8c2 JK |
2658 | if (!handle || err) { |
2659 | mutex_unlock(&sbi->s_orphan_lock); | |
ac27a0ec | 2660 | goto out_err; |
d745a8c2 | 2661 | } |
ac27a0ec | 2662 | |
d745a8c2 | 2663 | ino_next = NEXT_ORPHAN(inode); |
ac27a0ec | 2664 | if (prev == &sbi->s_orphan) { |
498e5f24 | 2665 | jbd_debug(4, "superblock will point to %u\n", ino_next); |
ac27a0ec | 2666 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
617ba13b | 2667 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
d745a8c2 JK |
2668 | if (err) { |
2669 | mutex_unlock(&sbi->s_orphan_lock); | |
ac27a0ec | 2670 | goto out_brelse; |
d745a8c2 | 2671 | } |
ac27a0ec | 2672 | sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); |
d745a8c2 | 2673 | mutex_unlock(&sbi->s_orphan_lock); |
b50924c2 | 2674 | err = ext4_handle_dirty_super(handle, inode->i_sb); |
ac27a0ec | 2675 | } else { |
617ba13b | 2676 | struct ext4_iloc iloc2; |
ac27a0ec | 2677 | struct inode *i_prev = |
617ba13b | 2678 | &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; |
ac27a0ec | 2679 | |
498e5f24 | 2680 | jbd_debug(4, "orphan inode %lu will point to %u\n", |
ac27a0ec | 2681 | i_prev->i_ino, ino_next); |
617ba13b | 2682 | err = ext4_reserve_inode_write(handle, i_prev, &iloc2); |
d745a8c2 JK |
2683 | if (err) { |
2684 | mutex_unlock(&sbi->s_orphan_lock); | |
ac27a0ec | 2685 | goto out_brelse; |
d745a8c2 | 2686 | } |
ac27a0ec | 2687 | NEXT_ORPHAN(i_prev) = ino_next; |
617ba13b | 2688 | err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); |
d745a8c2 | 2689 | mutex_unlock(&sbi->s_orphan_lock); |
ac27a0ec DK |
2690 | } |
2691 | if (err) | |
2692 | goto out_brelse; | |
2693 | NEXT_ORPHAN(inode) = 0; | |
617ba13b | 2694 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec | 2695 | out_err: |
617ba13b | 2696 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
2697 | return err; |
2698 | ||
2699 | out_brelse: | |
2700 | brelse(iloc.bh); | |
2701 | goto out_err; | |
2702 | } | |
2703 | ||
af5bc92d | 2704 | static int ext4_rmdir(struct inode *dir, struct dentry *dentry) |
ac27a0ec DK |
2705 | { |
2706 | int retval; | |
af5bc92d TT |
2707 | struct inode *inode; |
2708 | struct buffer_head *bh; | |
2709 | struct ext4_dir_entry_2 *de; | |
8dcfaad2 | 2710 | handle_t *handle = NULL; |
ac27a0ec DK |
2711 | |
2712 | /* Initialize quotas before so that eventual writes go in | |
2713 | * separate transaction */ | |
871a2931 CH |
2714 | dquot_initialize(dir); |
2715 | dquot_initialize(dentry->d_inode); | |
907f4554 | 2716 | |
ac27a0ec | 2717 | retval = -ENOENT; |
32f7f22c | 2718 | bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); |
36de9286 TT |
2719 | if (IS_ERR(bh)) |
2720 | return PTR_ERR(bh); | |
ac27a0ec DK |
2721 | if (!bh) |
2722 | goto end_rmdir; | |
2723 | ||
ac27a0ec DK |
2724 | inode = dentry->d_inode; |
2725 | ||
2726 | retval = -EIO; | |
2727 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2728 | goto end_rmdir; | |
2729 | ||
2730 | retval = -ENOTEMPTY; | |
af5bc92d | 2731 | if (!empty_dir(inode)) |
ac27a0ec DK |
2732 | goto end_rmdir; |
2733 | ||
8dcfaad2 | 2734 | handle = ext4_journal_start(dir, EXT4_HT_DIR, |
64044abf | 2735 | EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); |
8dcfaad2 TT |
2736 | if (IS_ERR(handle)) { |
2737 | retval = PTR_ERR(handle); | |
2738 | handle = NULL; | |
2739 | goto end_rmdir; | |
2740 | } | |
2741 | ||
2742 | if (IS_DIRSYNC(dir)) | |
2743 | ext4_handle_sync(handle); | |
2744 | ||
617ba13b | 2745 | retval = ext4_delete_entry(handle, dir, de, bh); |
ac27a0ec DK |
2746 | if (retval) |
2747 | goto end_rmdir; | |
f8628a14 | 2748 | if (!EXT4_DIR_LINK_EMPTY(inode)) |
12062ddd | 2749 | ext4_warning(inode->i_sb, |
af5bc92d TT |
2750 | "empty directory has too many links (%d)", |
2751 | inode->i_nlink); | |
ac27a0ec DK |
2752 | inode->i_version++; |
2753 | clear_nlink(inode); | |
2754 | /* There's no need to set i_disksize: the fact that i_nlink is | |
2755 | * zero will ensure that the right thing happens during any | |
2756 | * recovery. */ | |
2757 | inode->i_size = 0; | |
617ba13b | 2758 | ext4_orphan_add(handle, inode); |
ef7f3835 | 2759 | inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode); |
617ba13b | 2760 | ext4_mark_inode_dirty(handle, inode); |
f8628a14 | 2761 | ext4_dec_count(handle, dir); |
617ba13b MC |
2762 | ext4_update_dx_flag(dir); |
2763 | ext4_mark_inode_dirty(handle, dir); | |
ac27a0ec DK |
2764 | |
2765 | end_rmdir: | |
af5bc92d | 2766 | brelse(bh); |
8dcfaad2 TT |
2767 | if (handle) |
2768 | ext4_journal_stop(handle); | |
ac27a0ec DK |
2769 | return retval; |
2770 | } | |
2771 | ||
af5bc92d | 2772 | static int ext4_unlink(struct inode *dir, struct dentry *dentry) |
ac27a0ec DK |
2773 | { |
2774 | int retval; | |
af5bc92d TT |
2775 | struct inode *inode; |
2776 | struct buffer_head *bh; | |
2777 | struct ext4_dir_entry_2 *de; | |
931b6864 | 2778 | handle_t *handle = NULL; |
ac27a0ec | 2779 | |
0562e0ba | 2780 | trace_ext4_unlink_enter(dir, dentry); |
ac27a0ec DK |
2781 | /* Initialize quotas before so that eventual writes go |
2782 | * in separate transaction */ | |
871a2931 CH |
2783 | dquot_initialize(dir); |
2784 | dquot_initialize(dentry->d_inode); | |
907f4554 | 2785 | |
ac27a0ec | 2786 | retval = -ENOENT; |
32f7f22c | 2787 | bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); |
36de9286 TT |
2788 | if (IS_ERR(bh)) |
2789 | return PTR_ERR(bh); | |
ac27a0ec DK |
2790 | if (!bh) |
2791 | goto end_unlink; | |
2792 | ||
2793 | inode = dentry->d_inode; | |
2794 | ||
2795 | retval = -EIO; | |
2796 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2797 | goto end_unlink; | |
2798 | ||
931b6864 | 2799 | handle = ext4_journal_start(dir, EXT4_HT_DIR, |
64044abf | 2800 | EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); |
931b6864 TT |
2801 | if (IS_ERR(handle)) { |
2802 | retval = PTR_ERR(handle); | |
2803 | handle = NULL; | |
2804 | goto end_unlink; | |
2805 | } | |
2806 | ||
2807 | if (IS_DIRSYNC(dir)) | |
2808 | ext4_handle_sync(handle); | |
2809 | ||
ac27a0ec | 2810 | if (!inode->i_nlink) { |
12062ddd | 2811 | ext4_warning(inode->i_sb, |
af5bc92d TT |
2812 | "Deleting nonexistent file (%lu), %d", |
2813 | inode->i_ino, inode->i_nlink); | |
bfe86848 | 2814 | set_nlink(inode, 1); |
ac27a0ec | 2815 | } |
617ba13b | 2816 | retval = ext4_delete_entry(handle, dir, de, bh); |
ac27a0ec DK |
2817 | if (retval) |
2818 | goto end_unlink; | |
ef7f3835 | 2819 | dir->i_ctime = dir->i_mtime = ext4_current_time(dir); |
617ba13b MC |
2820 | ext4_update_dx_flag(dir); |
2821 | ext4_mark_inode_dirty(handle, dir); | |
825f1481 | 2822 | drop_nlink(inode); |
ac27a0ec | 2823 | if (!inode->i_nlink) |
617ba13b | 2824 | ext4_orphan_add(handle, inode); |
ef7f3835 | 2825 | inode->i_ctime = ext4_current_time(inode); |
617ba13b | 2826 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
2827 | retval = 0; |
2828 | ||
2829 | end_unlink: | |
af5bc92d | 2830 | brelse(bh); |
931b6864 TT |
2831 | if (handle) |
2832 | ext4_journal_stop(handle); | |
0562e0ba | 2833 | trace_ext4_unlink_exit(dentry, retval); |
ac27a0ec DK |
2834 | return retval; |
2835 | } | |
2836 | ||
af5bc92d TT |
2837 | static int ext4_symlink(struct inode *dir, |
2838 | struct dentry *dentry, const char *symname) | |
ac27a0ec DK |
2839 | { |
2840 | handle_t *handle; | |
af5bc92d | 2841 | struct inode *inode; |
ac27a0ec | 2842 | int l, err, retries = 0; |
df5e6223 | 2843 | int credits; |
ac27a0ec DK |
2844 | |
2845 | l = strlen(symname)+1; | |
2846 | if (l > dir->i_sb->s_blocksize) | |
2847 | return -ENAMETOOLONG; | |
2848 | ||
871a2931 | 2849 | dquot_initialize(dir); |
907f4554 | 2850 | |
df5e6223 JK |
2851 | if (l > EXT4_N_BLOCKS * 4) { |
2852 | /* | |
2853 | * For non-fast symlinks, we just allocate inode and put it on | |
2854 | * orphan list in the first transaction => we need bitmap, | |
8c208719 ES |
2855 | * group descriptor, sb, inode block, quota blocks, and |
2856 | * possibly selinux xattr blocks. | |
df5e6223 | 2857 | */ |
8c208719 ES |
2858 | credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + |
2859 | EXT4_XATTR_TRANS_BLOCKS; | |
df5e6223 JK |
2860 | } else { |
2861 | /* | |
2862 | * Fast symlink. We have to add entry to directory | |
2863 | * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS), | |
2864 | * allocate new inode (bitmap, group descriptor, inode block, | |
2865 | * quota blocks, sb is already counted in previous macros). | |
2866 | */ | |
2867 | credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
eb9cc7e1 | 2868 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3; |
df5e6223 | 2869 | } |
ac27a0ec | 2870 | retry: |
1139575a TT |
2871 | inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO, |
2872 | &dentry->d_name, 0, NULL, | |
2873 | EXT4_HT_DIR, credits); | |
2874 | handle = ext4_journal_current_handle(); | |
ac27a0ec DK |
2875 | err = PTR_ERR(inode); |
2876 | if (IS_ERR(inode)) | |
2877 | goto out_stop; | |
2878 | ||
df5e6223 | 2879 | if (l > EXT4_N_BLOCKS * 4) { |
617ba13b MC |
2880 | inode->i_op = &ext4_symlink_inode_operations; |
2881 | ext4_set_aops(inode); | |
ac27a0ec | 2882 | /* |
df5e6223 JK |
2883 | * We cannot call page_symlink() with transaction started |
2884 | * because it calls into ext4_write_begin() which can wait | |
2885 | * for transaction commit if we are running out of space | |
2886 | * and thus we deadlock. So we have to stop transaction now | |
2887 | * and restart it when symlink contents is written. | |
2888 | * | |
2889 | * To keep fs consistent in case of crash, we have to put inode | |
2890 | * to orphan list in the mean time. | |
ac27a0ec | 2891 | */ |
df5e6223 JK |
2892 | drop_nlink(inode); |
2893 | err = ext4_orphan_add(handle, inode); | |
2894 | ext4_journal_stop(handle); | |
2895 | if (err) | |
2896 | goto err_drop_inode; | |
54566b2c | 2897 | err = __page_symlink(inode, symname, l, 1); |
df5e6223 JK |
2898 | if (err) |
2899 | goto err_drop_inode; | |
2900 | /* | |
2901 | * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS | |
2902 | * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified | |
2903 | */ | |
9924a92a | 2904 | handle = ext4_journal_start(dir, EXT4_HT_DIR, |
df5e6223 JK |
2905 | EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2906 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1); | |
2907 | if (IS_ERR(handle)) { | |
2908 | err = PTR_ERR(handle); | |
2909 | goto err_drop_inode; | |
2910 | } | |
0ce8c010 | 2911 | set_nlink(inode, 1); |
df5e6223 | 2912 | err = ext4_orphan_del(handle, inode); |
ac27a0ec | 2913 | if (err) { |
df5e6223 | 2914 | ext4_journal_stop(handle); |
825f1481 | 2915 | clear_nlink(inode); |
df5e6223 | 2916 | goto err_drop_inode; |
ac27a0ec DK |
2917 | } |
2918 | } else { | |
e65187e6 | 2919 | /* clear the extent format for fast symlink */ |
12e9b892 | 2920 | ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); |
617ba13b | 2921 | inode->i_op = &ext4_fast_symlink_inode_operations; |
af5bc92d | 2922 | memcpy((char *)&EXT4_I(inode)->i_data, symname, l); |
ac27a0ec DK |
2923 | inode->i_size = l-1; |
2924 | } | |
617ba13b MC |
2925 | EXT4_I(inode)->i_disksize = inode->i_size; |
2926 | err = ext4_add_nondir(handle, dentry, inode); | |
1139575a TT |
2927 | if (!err && IS_DIRSYNC(dir)) |
2928 | ext4_handle_sync(handle); | |
2929 | ||
ac27a0ec | 2930 | out_stop: |
1139575a TT |
2931 | if (handle) |
2932 | ext4_journal_stop(handle); | |
617ba13b | 2933 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) |
ac27a0ec DK |
2934 | goto retry; |
2935 | return err; | |
df5e6223 JK |
2936 | err_drop_inode: |
2937 | unlock_new_inode(inode); | |
2938 | iput(inode); | |
2939 | return err; | |
ac27a0ec DK |
2940 | } |
2941 | ||
af5bc92d TT |
2942 | static int ext4_link(struct dentry *old_dentry, |
2943 | struct inode *dir, struct dentry *dentry) | |
ac27a0ec DK |
2944 | { |
2945 | handle_t *handle; | |
2946 | struct inode *inode = old_dentry->d_inode; | |
2947 | int err, retries = 0; | |
2948 | ||
b05ab1dc | 2949 | if (inode->i_nlink >= EXT4_LINK_MAX) |
ac27a0ec | 2950 | return -EMLINK; |
f8628a14 | 2951 | |
871a2931 | 2952 | dquot_initialize(dir); |
907f4554 | 2953 | |
ac27a0ec | 2954 | retry: |
9924a92a TT |
2955 | handle = ext4_journal_start(dir, EXT4_HT_DIR, |
2956 | (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
af51a2ac | 2957 | EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1); |
ac27a0ec DK |
2958 | if (IS_ERR(handle)) |
2959 | return PTR_ERR(handle); | |
2960 | ||
2961 | if (IS_DIRSYNC(dir)) | |
0390131b | 2962 | ext4_handle_sync(handle); |
ac27a0ec | 2963 | |
ef7f3835 | 2964 | inode->i_ctime = ext4_current_time(inode); |
f8628a14 | 2965 | ext4_inc_count(handle, inode); |
7de9c6ee | 2966 | ihold(inode); |
ac27a0ec | 2967 | |
6b38e842 AV |
2968 | err = ext4_add_entry(handle, dentry, inode); |
2969 | if (!err) { | |
2970 | ext4_mark_inode_dirty(handle, inode); | |
af51a2ac AV |
2971 | /* this can happen only for tmpfile being |
2972 | * linked the first time | |
2973 | */ | |
2974 | if (inode->i_nlink == 1) | |
2975 | ext4_orphan_del(handle, inode); | |
6b38e842 AV |
2976 | d_instantiate(dentry, inode); |
2977 | } else { | |
2978 | drop_nlink(inode); | |
2979 | iput(inode); | |
2980 | } | |
617ba13b MC |
2981 | ext4_journal_stop(handle); |
2982 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
2983 | goto retry; |
2984 | return err; | |
2985 | } | |
2986 | ||
32f7f22c TM |
2987 | |
2988 | /* | |
2989 | * Try to find buffer head where contains the parent block. | |
2990 | * It should be the inode block if it is inlined or the 1st block | |
2991 | * if it is a normal dir. | |
2992 | */ | |
2993 | static struct buffer_head *ext4_get_first_dir_block(handle_t *handle, | |
2994 | struct inode *inode, | |
2995 | int *retval, | |
2996 | struct ext4_dir_entry_2 **parent_de, | |
2997 | int *inlined) | |
2998 | { | |
2999 | struct buffer_head *bh; | |
3000 | ||
3001 | if (!ext4_has_inline_data(inode)) { | |
dc6982ff TT |
3002 | bh = ext4_read_dirblock(inode, 0, EITHER); |
3003 | if (IS_ERR(bh)) { | |
3004 | *retval = PTR_ERR(bh); | |
32f7f22c TM |
3005 | return NULL; |
3006 | } | |
3007 | *parent_de = ext4_next_entry( | |
3008 | (struct ext4_dir_entry_2 *)bh->b_data, | |
3009 | inode->i_sb->s_blocksize); | |
3010 | return bh; | |
3011 | } | |
3012 | ||
3013 | *inlined = 1; | |
3014 | return ext4_get_first_inline_block(inode, parent_de, retval); | |
3015 | } | |
ac27a0ec | 3016 | |
c0d268c3 MS |
3017 | struct ext4_renament { |
3018 | struct inode *dir; | |
3019 | struct dentry *dentry; | |
3020 | struct inode *inode; | |
bd42998a MS |
3021 | bool is_dir; |
3022 | int dir_nlink_delta; | |
c0d268c3 MS |
3023 | |
3024 | /* entry for "dentry" */ | |
3025 | struct buffer_head *bh; | |
3026 | struct ext4_dir_entry_2 *de; | |
3027 | int inlined; | |
3028 | ||
3029 | /* entry for ".." in inode if it's a directory */ | |
3030 | struct buffer_head *dir_bh; | |
3031 | struct ext4_dir_entry_2 *parent_de; | |
3032 | int dir_inlined; | |
3033 | }; | |
3034 | ||
bd1af145 MS |
3035 | static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent) |
3036 | { | |
3037 | int retval; | |
3038 | ||
3039 | ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode, | |
3040 | &retval, &ent->parent_de, | |
3041 | &ent->dir_inlined); | |
3042 | if (!ent->dir_bh) | |
3043 | return retval; | |
3044 | if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino) | |
3045 | return -EIO; | |
3046 | BUFFER_TRACE(ent->dir_bh, "get_write_access"); | |
3047 | return ext4_journal_get_write_access(handle, ent->dir_bh); | |
3048 | } | |
3049 | ||
3050 | static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent, | |
3051 | unsigned dir_ino) | |
3052 | { | |
3053 | int retval; | |
3054 | ||
3055 | ent->parent_de->inode = cpu_to_le32(dir_ino); | |
3056 | BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata"); | |
3057 | if (!ent->dir_inlined) { | |
3058 | if (is_dx(ent->inode)) { | |
3059 | retval = ext4_handle_dirty_dx_node(handle, | |
3060 | ent->inode, | |
3061 | ent->dir_bh); | |
3062 | } else { | |
3063 | retval = ext4_handle_dirty_dirent_node(handle, | |
3064 | ent->inode, | |
3065 | ent->dir_bh); | |
3066 | } | |
3067 | } else { | |
3068 | retval = ext4_mark_inode_dirty(handle, ent->inode); | |
3069 | } | |
3070 | if (retval) { | |
3071 | ext4_std_error(ent->dir->i_sb, retval); | |
3072 | return retval; | |
3073 | } | |
3074 | return 0; | |
3075 | } | |
3076 | ||
3077 | static int ext4_setent(handle_t *handle, struct ext4_renament *ent, | |
3078 | unsigned ino, unsigned file_type) | |
3079 | { | |
3080 | int retval; | |
3081 | ||
3082 | BUFFER_TRACE(ent->bh, "get write access"); | |
3083 | retval = ext4_journal_get_write_access(handle, ent->bh); | |
3084 | if (retval) | |
3085 | return retval; | |
3086 | ent->de->inode = cpu_to_le32(ino); | |
3087 | if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb, | |
3088 | EXT4_FEATURE_INCOMPAT_FILETYPE)) | |
3089 | ent->de->file_type = file_type; | |
3090 | ent->dir->i_version++; | |
3091 | ent->dir->i_ctime = ent->dir->i_mtime = | |
3092 | ext4_current_time(ent->dir); | |
3093 | ext4_mark_inode_dirty(handle, ent->dir); | |
3094 | BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata"); | |
3095 | if (!ent->inlined) { | |
3096 | retval = ext4_handle_dirty_dirent_node(handle, | |
3097 | ent->dir, ent->bh); | |
3098 | if (unlikely(retval)) { | |
3099 | ext4_std_error(ent->dir->i_sb, retval); | |
3100 | return retval; | |
3101 | } | |
3102 | } | |
3103 | brelse(ent->bh); | |
3104 | ent->bh = NULL; | |
3105 | ||
3106 | return 0; | |
3107 | } | |
3108 | ||
3109 | static int ext4_find_delete_entry(handle_t *handle, struct inode *dir, | |
3110 | const struct qstr *d_name) | |
3111 | { | |
3112 | int retval = -ENOENT; | |
3113 | struct buffer_head *bh; | |
3114 | struct ext4_dir_entry_2 *de; | |
3115 | ||
3116 | bh = ext4_find_entry(dir, d_name, &de, NULL); | |
36de9286 TT |
3117 | if (IS_ERR(bh)) |
3118 | return PTR_ERR(bh); | |
bd1af145 MS |
3119 | if (bh) { |
3120 | retval = ext4_delete_entry(handle, dir, de, bh); | |
3121 | brelse(bh); | |
3122 | } | |
3123 | return retval; | |
3124 | } | |
3125 | ||
d80d448c DW |
3126 | static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent, |
3127 | int force_reread) | |
bd1af145 MS |
3128 | { |
3129 | int retval; | |
3130 | /* | |
3131 | * ent->de could have moved from under us during htree split, so make | |
3132 | * sure that we are deleting the right entry. We might also be pointing | |
3133 | * to a stale entry in the unused part of ent->bh so just checking inum | |
3134 | * and the name isn't enough. | |
3135 | */ | |
3136 | if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino || | |
3137 | ent->de->name_len != ent->dentry->d_name.len || | |
3138 | strncmp(ent->de->name, ent->dentry->d_name.name, | |
d80d448c DW |
3139 | ent->de->name_len) || |
3140 | force_reread) { | |
bd1af145 MS |
3141 | retval = ext4_find_delete_entry(handle, ent->dir, |
3142 | &ent->dentry->d_name); | |
3143 | } else { | |
3144 | retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh); | |
3145 | if (retval == -ENOENT) { | |
3146 | retval = ext4_find_delete_entry(handle, ent->dir, | |
3147 | &ent->dentry->d_name); | |
3148 | } | |
3149 | } | |
3150 | ||
3151 | if (retval) { | |
3152 | ext4_warning(ent->dir->i_sb, | |
3153 | "Deleting old file (%lu), %d, error=%d", | |
3154 | ent->dir->i_ino, ent->dir->i_nlink, retval); | |
3155 | } | |
3156 | } | |
3157 | ||
bd42998a MS |
3158 | static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent) |
3159 | { | |
3160 | if (ent->dir_nlink_delta) { | |
3161 | if (ent->dir_nlink_delta == -1) | |
3162 | ext4_dec_count(handle, ent->dir); | |
3163 | else | |
3164 | ext4_inc_count(handle, ent->dir); | |
3165 | ext4_mark_inode_dirty(handle, ent->dir); | |
3166 | } | |
3167 | } | |
3168 | ||
ac27a0ec DK |
3169 | /* |
3170 | * Anybody can rename anything with this: the permission checks are left to the | |
3171 | * higher-level routines. | |
0e202704 TT |
3172 | * |
3173 | * n.b. old_{dentry,inode) refers to the source dentry/inode | |
3174 | * while new_{dentry,inode) refers to the destination dentry/inode | |
3175 | * This comes from rename(const char *oldpath, const char *newpath) | |
ac27a0ec | 3176 | */ |
af5bc92d TT |
3177 | static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry, |
3178 | struct inode *new_dir, struct dentry *new_dentry) | |
ac27a0ec | 3179 | { |
5b61de75 | 3180 | handle_t *handle = NULL; |
c0d268c3 MS |
3181 | struct ext4_renament old = { |
3182 | .dir = old_dir, | |
3183 | .dentry = old_dentry, | |
3184 | .inode = old_dentry->d_inode, | |
3185 | }; | |
3186 | struct ext4_renament new = { | |
3187 | .dir = new_dir, | |
3188 | .dentry = new_dentry, | |
3189 | .inode = new_dentry->d_inode, | |
3190 | }; | |
d80d448c | 3191 | int force_reread; |
0e202704 | 3192 | int retval; |
907f4554 | 3193 | |
c0d268c3 MS |
3194 | dquot_initialize(old.dir); |
3195 | dquot_initialize(new.dir); | |
ac27a0ec DK |
3196 | |
3197 | /* Initialize quotas before so that eventual writes go | |
3198 | * in separate transaction */ | |
c0d268c3 MS |
3199 | if (new.inode) |
3200 | dquot_initialize(new.inode); | |
ac27a0ec | 3201 | |
c0d268c3 | 3202 | old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL); |
36de9286 TT |
3203 | if (IS_ERR(old.bh)) |
3204 | return PTR_ERR(old.bh); | |
ac27a0ec DK |
3205 | /* |
3206 | * Check for inode number is _not_ due to possible IO errors. | |
3207 | * We might rmdir the source, keep it as pwd of some process | |
3208 | * and merrily kill the link to whatever was created under the | |
3209 | * same name. Goodbye sticky bit ;-< | |
3210 | */ | |
ac27a0ec | 3211 | retval = -ENOENT; |
c0d268c3 | 3212 | if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) |
ac27a0ec DK |
3213 | goto end_rename; |
3214 | ||
c0d268c3 MS |
3215 | new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, |
3216 | &new.de, &new.inlined); | |
36de9286 TT |
3217 | if (IS_ERR(new.bh)) { |
3218 | retval = PTR_ERR(new.bh); | |
3219 | goto end_rename; | |
3220 | } | |
c0d268c3 MS |
3221 | if (new.bh) { |
3222 | if (!new.inode) { | |
3223 | brelse(new.bh); | |
3224 | new.bh = NULL; | |
ac27a0ec DK |
3225 | } |
3226 | } | |
c0d268c3 MS |
3227 | if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC)) |
3228 | ext4_alloc_da_blocks(old.inode); | |
5b61de75 | 3229 | |
c0d268c3 MS |
3230 | handle = ext4_journal_start(old.dir, EXT4_HT_DIR, |
3231 | (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + | |
5b61de75 TT |
3232 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2)); |
3233 | if (IS_ERR(handle)) | |
3234 | return PTR_ERR(handle); | |
3235 | ||
c0d268c3 | 3236 | if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) |
5b61de75 TT |
3237 | ext4_handle_sync(handle); |
3238 | ||
c0d268c3 MS |
3239 | if (S_ISDIR(old.inode->i_mode)) { |
3240 | if (new.inode) { | |
ac27a0ec | 3241 | retval = -ENOTEMPTY; |
c0d268c3 | 3242 | if (!empty_dir(new.inode)) |
ac27a0ec | 3243 | goto end_rename; |
0d7d5d67 MS |
3244 | } else { |
3245 | retval = -EMLINK; | |
3246 | if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir)) | |
3247 | goto end_rename; | |
ac27a0ec | 3248 | } |
bd1af145 | 3249 | retval = ext4_rename_dir_prepare(handle, &old); |
ef607893 AG |
3250 | if (retval) |
3251 | goto end_rename; | |
ac27a0ec | 3252 | } |
d80d448c DW |
3253 | /* |
3254 | * If we're renaming a file within an inline_data dir and adding or | |
3255 | * setting the new dirent causes a conversion from inline_data to | |
3256 | * extents/blockmap, we need to force the dirent delete code to | |
3257 | * re-read the directory, or else we end up trying to delete a dirent | |
3258 | * from what is now the extent tree root (or a block map). | |
3259 | */ | |
3260 | force_reread = (new.dir->i_ino == old.dir->i_ino && | |
3261 | ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA)); | |
c0d268c3 MS |
3262 | if (!new.bh) { |
3263 | retval = ext4_add_entry(handle, new.dentry, old.inode); | |
ac27a0ec DK |
3264 | if (retval) |
3265 | goto end_rename; | |
3266 | } else { | |
bd1af145 MS |
3267 | retval = ext4_setent(handle, &new, |
3268 | old.inode->i_ino, old.de->file_type); | |
ef607893 AG |
3269 | if (retval) |
3270 | goto end_rename; | |
ac27a0ec | 3271 | } |
d80d448c DW |
3272 | if (force_reread) |
3273 | force_reread = !ext4_test_inode_flag(new.dir, | |
3274 | EXT4_INODE_INLINE_DATA); | |
ac27a0ec DK |
3275 | |
3276 | /* | |
3277 | * Like most other Unix systems, set the ctime for inodes on a | |
3278 | * rename. | |
3279 | */ | |
c0d268c3 MS |
3280 | old.inode->i_ctime = ext4_current_time(old.inode); |
3281 | ext4_mark_inode_dirty(handle, old.inode); | |
ac27a0ec DK |
3282 | |
3283 | /* | |
3284 | * ok, that's it | |
3285 | */ | |
d80d448c | 3286 | ext4_rename_delete(handle, &old, force_reread); |
ac27a0ec | 3287 | |
c0d268c3 MS |
3288 | if (new.inode) { |
3289 | ext4_dec_count(handle, new.inode); | |
3290 | new.inode->i_ctime = ext4_current_time(new.inode); | |
ac27a0ec | 3291 | } |
c0d268c3 MS |
3292 | old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir); |
3293 | ext4_update_dx_flag(old.dir); | |
3294 | if (old.dir_bh) { | |
bd1af145 MS |
3295 | retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); |
3296 | if (retval) | |
b4097142 | 3297 | goto end_rename; |
bd1af145 | 3298 | |
c0d268c3 MS |
3299 | ext4_dec_count(handle, old.dir); |
3300 | if (new.inode) { | |
f8628a14 | 3301 | /* checked empty_dir above, can't have another parent, |
825f1481 | 3302 | * ext4_dec_count() won't work for many-linked dirs */ |
c0d268c3 | 3303 | clear_nlink(new.inode); |
ac27a0ec | 3304 | } else { |
c0d268c3 MS |
3305 | ext4_inc_count(handle, new.dir); |
3306 | ext4_update_dx_flag(new.dir); | |
3307 | ext4_mark_inode_dirty(handle, new.dir); | |
ac27a0ec DK |
3308 | } |
3309 | } | |
c0d268c3 MS |
3310 | ext4_mark_inode_dirty(handle, old.dir); |
3311 | if (new.inode) { | |
3312 | ext4_mark_inode_dirty(handle, new.inode); | |
3313 | if (!new.inode->i_nlink) | |
3314 | ext4_orphan_add(handle, new.inode); | |
ac27a0ec DK |
3315 | } |
3316 | retval = 0; | |
3317 | ||
3318 | end_rename: | |
c0d268c3 MS |
3319 | brelse(old.dir_bh); |
3320 | brelse(old.bh); | |
3321 | brelse(new.bh); | |
5b61de75 TT |
3322 | if (handle) |
3323 | ext4_journal_stop(handle); | |
ac27a0ec DK |
3324 | return retval; |
3325 | } | |
3326 | ||
bd42998a MS |
3327 | static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry, |
3328 | struct inode *new_dir, struct dentry *new_dentry) | |
3329 | { | |
3330 | handle_t *handle = NULL; | |
3331 | struct ext4_renament old = { | |
3332 | .dir = old_dir, | |
3333 | .dentry = old_dentry, | |
3334 | .inode = old_dentry->d_inode, | |
3335 | }; | |
3336 | struct ext4_renament new = { | |
3337 | .dir = new_dir, | |
3338 | .dentry = new_dentry, | |
3339 | .inode = new_dentry->d_inode, | |
3340 | }; | |
3341 | u8 new_file_type; | |
3342 | int retval; | |
3343 | ||
3344 | dquot_initialize(old.dir); | |
3345 | dquot_initialize(new.dir); | |
3346 | ||
3347 | old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, | |
3348 | &old.de, &old.inlined); | |
36de9286 TT |
3349 | if (IS_ERR(old.bh)) |
3350 | return PTR_ERR(old.bh); | |
bd42998a MS |
3351 | /* |
3352 | * Check for inode number is _not_ due to possible IO errors. | |
3353 | * We might rmdir the source, keep it as pwd of some process | |
3354 | * and merrily kill the link to whatever was created under the | |
3355 | * same name. Goodbye sticky bit ;-< | |
3356 | */ | |
3357 | retval = -ENOENT; | |
3358 | if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino) | |
3359 | goto end_rename; | |
3360 | ||
3361 | new.bh = ext4_find_entry(new.dir, &new.dentry->d_name, | |
3362 | &new.de, &new.inlined); | |
36de9286 TT |
3363 | if (IS_ERR(new.bh)) { |
3364 | retval = PTR_ERR(new.bh); | |
3365 | goto end_rename; | |
3366 | } | |
bd42998a MS |
3367 | |
3368 | /* RENAME_EXCHANGE case: old *and* new must both exist */ | |
3369 | if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino) | |
3370 | goto end_rename; | |
3371 | ||
3372 | handle = ext4_journal_start(old.dir, EXT4_HT_DIR, | |
3373 | (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) + | |
3374 | 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2)); | |
3375 | if (IS_ERR(handle)) | |
3376 | return PTR_ERR(handle); | |
3377 | ||
3378 | if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir)) | |
3379 | ext4_handle_sync(handle); | |
3380 | ||
3381 | if (S_ISDIR(old.inode->i_mode)) { | |
3382 | old.is_dir = true; | |
3383 | retval = ext4_rename_dir_prepare(handle, &old); | |
3384 | if (retval) | |
3385 | goto end_rename; | |
3386 | } | |
3387 | if (S_ISDIR(new.inode->i_mode)) { | |
3388 | new.is_dir = true; | |
3389 | retval = ext4_rename_dir_prepare(handle, &new); | |
3390 | if (retval) | |
3391 | goto end_rename; | |
3392 | } | |
3393 | ||
3394 | /* | |
3395 | * Other than the special case of overwriting a directory, parents' | |
3396 | * nlink only needs to be modified if this is a cross directory rename. | |
3397 | */ | |
3398 | if (old.dir != new.dir && old.is_dir != new.is_dir) { | |
3399 | old.dir_nlink_delta = old.is_dir ? -1 : 1; | |
3400 | new.dir_nlink_delta = -old.dir_nlink_delta; | |
3401 | retval = -EMLINK; | |
3402 | if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) || | |
3403 | (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir))) | |
3404 | goto end_rename; | |
3405 | } | |
3406 | ||
3407 | new_file_type = new.de->file_type; | |
3408 | retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type); | |
3409 | if (retval) | |
3410 | goto end_rename; | |
3411 | ||
3412 | retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type); | |
3413 | if (retval) | |
3414 | goto end_rename; | |
3415 | ||
3416 | /* | |
3417 | * Like most other Unix systems, set the ctime for inodes on a | |
3418 | * rename. | |
3419 | */ | |
3420 | old.inode->i_ctime = ext4_current_time(old.inode); | |
3421 | new.inode->i_ctime = ext4_current_time(new.inode); | |
3422 | ext4_mark_inode_dirty(handle, old.inode); | |
3423 | ext4_mark_inode_dirty(handle, new.inode); | |
3424 | ||
3425 | if (old.dir_bh) { | |
3426 | retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino); | |
3427 | if (retval) | |
3428 | goto end_rename; | |
3429 | } | |
3430 | if (new.dir_bh) { | |
3431 | retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino); | |
3432 | if (retval) | |
3433 | goto end_rename; | |
3434 | } | |
3435 | ext4_update_dir_count(handle, &old); | |
3436 | ext4_update_dir_count(handle, &new); | |
3437 | retval = 0; | |
3438 | ||
3439 | end_rename: | |
3440 | brelse(old.dir_bh); | |
3441 | brelse(new.dir_bh); | |
3442 | brelse(old.bh); | |
3443 | brelse(new.bh); | |
3444 | if (handle) | |
3445 | ext4_journal_stop(handle); | |
3446 | return retval; | |
3447 | } | |
3448 | ||
0a7c3937 MS |
3449 | static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry, |
3450 | struct inode *new_dir, struct dentry *new_dentry, | |
3451 | unsigned int flags) | |
3452 | { | |
bd42998a | 3453 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE)) |
0a7c3937 MS |
3454 | return -EINVAL; |
3455 | ||
bd42998a MS |
3456 | if (flags & RENAME_EXCHANGE) { |
3457 | return ext4_cross_rename(old_dir, old_dentry, | |
3458 | new_dir, new_dentry); | |
3459 | } | |
3460 | /* | |
3461 | * Existence checking was done by the VFS, otherwise "RENAME_NOREPLACE" | |
3462 | * is equivalent to regular rename. | |
3463 | */ | |
0a7c3937 MS |
3464 | return ext4_rename(old_dir, old_dentry, new_dir, new_dentry); |
3465 | } | |
3466 | ||
ac27a0ec DK |
3467 | /* |
3468 | * directories can handle most operations... | |
3469 | */ | |
754661f1 | 3470 | const struct inode_operations ext4_dir_inode_operations = { |
617ba13b MC |
3471 | .create = ext4_create, |
3472 | .lookup = ext4_lookup, | |
3473 | .link = ext4_link, | |
3474 | .unlink = ext4_unlink, | |
3475 | .symlink = ext4_symlink, | |
3476 | .mkdir = ext4_mkdir, | |
3477 | .rmdir = ext4_rmdir, | |
3478 | .mknod = ext4_mknod, | |
af51a2ac | 3479 | .tmpfile = ext4_tmpfile, |
0a7c3937 | 3480 | .rename2 = ext4_rename2, |
617ba13b | 3481 | .setattr = ext4_setattr, |
ac27a0ec DK |
3482 | .setxattr = generic_setxattr, |
3483 | .getxattr = generic_getxattr, | |
617ba13b | 3484 | .listxattr = ext4_listxattr, |
ac27a0ec | 3485 | .removexattr = generic_removexattr, |
4e34e719 | 3486 | .get_acl = ext4_get_acl, |
64e178a7 | 3487 | .set_acl = ext4_set_acl, |
abc8746e | 3488 | .fiemap = ext4_fiemap, |
ac27a0ec DK |
3489 | }; |
3490 | ||
754661f1 | 3491 | const struct inode_operations ext4_special_inode_operations = { |
617ba13b | 3492 | .setattr = ext4_setattr, |
ac27a0ec DK |
3493 | .setxattr = generic_setxattr, |
3494 | .getxattr = generic_getxattr, | |
617ba13b | 3495 | .listxattr = ext4_listxattr, |
ac27a0ec | 3496 | .removexattr = generic_removexattr, |
4e34e719 | 3497 | .get_acl = ext4_get_acl, |
64e178a7 | 3498 | .set_acl = ext4_set_acl, |
ac27a0ec | 3499 | }; |