]>
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; | |
0f70b406 | 56 | int err = 0; |
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 | ||
0f70b406 TT |
65 | bh = ext4_bread(handle, inode, *block, 1, &err); |
66 | if (!bh) | |
67 | return ERR_PTR(err); | |
68 | inode->i_size += inode->i_sb->s_blocksize; | |
69 | EXT4_I(inode)->i_disksize = inode->i_size; | |
70 | err = ext4_journal_get_write_access(handle, bh); | |
71 | if (err) { | |
72 | brelse(bh); | |
73 | ext4_std_error(inode->i_sb, err); | |
74 | return ERR_PTR(err); | |
6d1ab10e | 75 | } |
ac27a0ec DK |
76 | return bh; |
77 | } | |
78 | ||
dc6982ff TT |
79 | static int ext4_dx_csum_verify(struct inode *inode, |
80 | struct ext4_dir_entry *dirent); | |
81 | ||
82 | typedef enum { | |
83 | EITHER, INDEX, DIRENT | |
84 | } dirblock_type_t; | |
85 | ||
86 | #define ext4_read_dirblock(inode, block, type) \ | |
87 | __ext4_read_dirblock((inode), (block), (type), __LINE__) | |
88 | ||
89 | static struct buffer_head *__ext4_read_dirblock(struct inode *inode, | |
90 | ext4_lblk_t block, | |
91 | dirblock_type_t type, | |
92 | unsigned int line) | |
93 | { | |
94 | struct buffer_head *bh; | |
95 | struct ext4_dir_entry *dirent; | |
96 | int err = 0, is_dx_block = 0; | |
97 | ||
98 | bh = ext4_bread(NULL, inode, block, 0, &err); | |
99 | if (!bh) { | |
100 | if (err == 0) { | |
101 | ext4_error_inode(inode, __func__, line, block, | |
102 | "Directory hole found"); | |
103 | return ERR_PTR(-EIO); | |
104 | } | |
105 | __ext4_warning(inode->i_sb, __func__, line, | |
106 | "error reading directory block " | |
107 | "(ino %lu, block %lu)", inode->i_ino, | |
108 | (unsigned long) block); | |
109 | return ERR_PTR(err); | |
110 | } | |
111 | dirent = (struct ext4_dir_entry *) bh->b_data; | |
112 | /* Determine whether or not we have an index block */ | |
113 | if (is_dx(inode)) { | |
114 | if (block == 0) | |
115 | is_dx_block = 1; | |
116 | else if (ext4_rec_len_from_disk(dirent->rec_len, | |
117 | inode->i_sb->s_blocksize) == | |
118 | inode->i_sb->s_blocksize) | |
119 | is_dx_block = 1; | |
120 | } | |
121 | if (!is_dx_block && type == INDEX) { | |
122 | ext4_error_inode(inode, __func__, line, block, | |
123 | "directory leaf block found instead of index block"); | |
124 | return ERR_PTR(-EIO); | |
125 | } | |
126 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
127 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) || | |
128 | buffer_verified(bh)) | |
129 | return bh; | |
130 | ||
131 | /* | |
132 | * An empty leaf block can get mistaken for a index block; for | |
133 | * this reason, we can only check the index checksum when the | |
134 | * caller is sure it should be an index block. | |
135 | */ | |
136 | if (is_dx_block && type == INDEX) { | |
137 | if (ext4_dx_csum_verify(inode, dirent)) | |
138 | set_buffer_verified(bh); | |
139 | else { | |
140 | ext4_error_inode(inode, __func__, line, block, | |
141 | "Directory index failed checksum"); | |
a871611b | 142 | brelse(bh); |
dc6982ff | 143 | return ERR_PTR(-EIO); |
a871611b | 144 | } |
ac27a0ec | 145 | } |
dc6982ff TT |
146 | if (!is_dx_block) { |
147 | if (ext4_dirent_csum_verify(inode, dirent)) | |
148 | set_buffer_verified(bh); | |
149 | else { | |
150 | ext4_error_inode(inode, __func__, line, block, | |
151 | "Directory block failed checksum"); | |
152 | brelse(bh); | |
153 | return ERR_PTR(-EIO); | |
154 | } | |
6d1ab10e | 155 | } |
ac27a0ec DK |
156 | return bh; |
157 | } | |
158 | ||
159 | #ifndef assert | |
160 | #define assert(test) J_ASSERT(test) | |
161 | #endif | |
162 | ||
ac27a0ec DK |
163 | #ifdef DX_DEBUG |
164 | #define dxtrace(command) command | |
165 | #else | |
166 | #define dxtrace(command) | |
167 | #endif | |
168 | ||
169 | struct fake_dirent | |
170 | { | |
171 | __le32 inode; | |
172 | __le16 rec_len; | |
173 | u8 name_len; | |
174 | u8 file_type; | |
175 | }; | |
176 | ||
177 | struct dx_countlimit | |
178 | { | |
179 | __le16 limit; | |
180 | __le16 count; | |
181 | }; | |
182 | ||
183 | struct dx_entry | |
184 | { | |
185 | __le32 hash; | |
186 | __le32 block; | |
187 | }; | |
188 | ||
189 | /* | |
190 | * dx_root_info is laid out so that if it should somehow get overlaid by a | |
191 | * dirent the two low bits of the hash version will be zero. Therefore, the | |
192 | * hash version mod 4 should never be 0. Sincerely, the paranoia department. | |
193 | */ | |
194 | ||
195 | struct dx_root | |
196 | { | |
197 | struct fake_dirent dot; | |
198 | char dot_name[4]; | |
199 | struct fake_dirent dotdot; | |
200 | char dotdot_name[4]; | |
201 | struct dx_root_info | |
202 | { | |
203 | __le32 reserved_zero; | |
204 | u8 hash_version; | |
205 | u8 info_length; /* 8 */ | |
206 | u8 indirect_levels; | |
207 | u8 unused_flags; | |
208 | } | |
209 | info; | |
210 | struct dx_entry entries[0]; | |
211 | }; | |
212 | ||
213 | struct dx_node | |
214 | { | |
215 | struct fake_dirent fake; | |
216 | struct dx_entry entries[0]; | |
217 | }; | |
218 | ||
219 | ||
220 | struct dx_frame | |
221 | { | |
222 | struct buffer_head *bh; | |
223 | struct dx_entry *entries; | |
224 | struct dx_entry *at; | |
225 | }; | |
226 | ||
227 | struct dx_map_entry | |
228 | { | |
229 | u32 hash; | |
ef2b02d3 ES |
230 | u16 offs; |
231 | u16 size; | |
ac27a0ec DK |
232 | }; |
233 | ||
e6153918 DW |
234 | /* |
235 | * This goes at the end of each htree block. | |
236 | */ | |
237 | struct dx_tail { | |
238 | u32 dt_reserved; | |
239 | __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */ | |
240 | }; | |
241 | ||
725d26d3 AK |
242 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry); |
243 | static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value); | |
af5bc92d TT |
244 | static inline unsigned dx_get_hash(struct dx_entry *entry); |
245 | static void dx_set_hash(struct dx_entry *entry, unsigned value); | |
246 | static unsigned dx_get_count(struct dx_entry *entries); | |
247 | static unsigned dx_get_limit(struct dx_entry *entries); | |
248 | static void dx_set_count(struct dx_entry *entries, unsigned value); | |
249 | static void dx_set_limit(struct dx_entry *entries, unsigned value); | |
250 | static unsigned dx_root_limit(struct inode *dir, unsigned infosize); | |
251 | static unsigned dx_node_limit(struct inode *dir); | |
f702ba0f | 252 | static struct dx_frame *dx_probe(const struct qstr *d_name, |
ac27a0ec DK |
253 | struct inode *dir, |
254 | struct dx_hash_info *hinfo, | |
255 | struct dx_frame *frame, | |
256 | int *err); | |
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, | |
272 | struct ext4_dir_entry_2 **res_dir, | |
273 | int *err); | |
617ba13b | 274 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, |
ac27a0ec DK |
275 | struct inode *inode); |
276 | ||
dbe89444 | 277 | /* checksumming functions */ |
3c47d541 TM |
278 | void initialize_dirent_tail(struct ext4_dir_entry_tail *t, |
279 | unsigned int blocksize) | |
b0336e8d DW |
280 | { |
281 | memset(t, 0, sizeof(struct ext4_dir_entry_tail)); | |
282 | t->det_rec_len = ext4_rec_len_to_disk( | |
283 | sizeof(struct ext4_dir_entry_tail), blocksize); | |
284 | t->det_reserved_ft = EXT4_FT_DIR_CSUM; | |
285 | } | |
286 | ||
287 | /* Walk through a dirent block to find a checksum "dirent" at the tail */ | |
288 | static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode, | |
289 | struct ext4_dir_entry *de) | |
290 | { | |
291 | struct ext4_dir_entry_tail *t; | |
292 | ||
293 | #ifdef PARANOID | |
294 | struct ext4_dir_entry *d, *top; | |
295 | ||
296 | d = de; | |
297 | top = (struct ext4_dir_entry *)(((void *)de) + | |
298 | (EXT4_BLOCK_SIZE(inode->i_sb) - | |
299 | sizeof(struct ext4_dir_entry_tail))); | |
300 | while (d < top && d->rec_len) | |
301 | d = (struct ext4_dir_entry *)(((void *)d) + | |
302 | le16_to_cpu(d->rec_len)); | |
303 | ||
304 | if (d != top) | |
305 | return NULL; | |
306 | ||
307 | t = (struct ext4_dir_entry_tail *)d; | |
308 | #else | |
309 | t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb)); | |
310 | #endif | |
311 | ||
312 | if (t->det_reserved_zero1 || | |
313 | le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) || | |
314 | t->det_reserved_zero2 || | |
315 | t->det_reserved_ft != EXT4_FT_DIR_CSUM) | |
316 | return NULL; | |
317 | ||
318 | return t; | |
319 | } | |
320 | ||
321 | static __le32 ext4_dirent_csum(struct inode *inode, | |
322 | struct ext4_dir_entry *dirent, int size) | |
323 | { | |
324 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
325 | struct ext4_inode_info *ei = EXT4_I(inode); | |
326 | __u32 csum; | |
327 | ||
328 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); | |
329 | return cpu_to_le32(csum); | |
330 | } | |
331 | ||
dffe9d8d TT |
332 | static void warn_no_space_for_csum(struct inode *inode) |
333 | { | |
334 | ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for " | |
335 | "checksum. Please run e2fsck -D.", inode->i_ino); | |
336 | } | |
337 | ||
b0336e8d DW |
338 | int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent) |
339 | { | |
340 | struct ext4_dir_entry_tail *t; | |
341 | ||
342 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
343 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
344 | return 1; | |
345 | ||
346 | t = get_dirent_tail(inode, dirent); | |
347 | if (!t) { | |
dffe9d8d | 348 | warn_no_space_for_csum(inode); |
b0336e8d DW |
349 | return 0; |
350 | } | |
351 | ||
352 | if (t->det_checksum != ext4_dirent_csum(inode, dirent, | |
353 | (void *)t - (void *)dirent)) | |
354 | return 0; | |
355 | ||
356 | return 1; | |
357 | } | |
358 | ||
359 | static void ext4_dirent_csum_set(struct inode *inode, | |
360 | struct ext4_dir_entry *dirent) | |
361 | { | |
362 | struct ext4_dir_entry_tail *t; | |
363 | ||
364 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
365 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
366 | return; | |
367 | ||
368 | t = get_dirent_tail(inode, dirent); | |
369 | if (!t) { | |
dffe9d8d | 370 | warn_no_space_for_csum(inode); |
b0336e8d DW |
371 | return; |
372 | } | |
373 | ||
374 | t->det_checksum = ext4_dirent_csum(inode, dirent, | |
375 | (void *)t - (void *)dirent); | |
376 | } | |
377 | ||
3c47d541 TM |
378 | int ext4_handle_dirty_dirent_node(handle_t *handle, |
379 | struct inode *inode, | |
380 | struct buffer_head *bh) | |
b0336e8d DW |
381 | { |
382 | ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); | |
383 | return ext4_handle_dirty_metadata(handle, inode, bh); | |
384 | } | |
385 | ||
dbe89444 DW |
386 | static struct dx_countlimit *get_dx_countlimit(struct inode *inode, |
387 | struct ext4_dir_entry *dirent, | |
388 | int *offset) | |
389 | { | |
390 | struct ext4_dir_entry *dp; | |
391 | struct dx_root_info *root; | |
392 | int count_offset; | |
393 | ||
394 | if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb)) | |
395 | count_offset = 8; | |
396 | else if (le16_to_cpu(dirent->rec_len) == 12) { | |
397 | dp = (struct ext4_dir_entry *)(((void *)dirent) + 12); | |
398 | if (le16_to_cpu(dp->rec_len) != | |
399 | EXT4_BLOCK_SIZE(inode->i_sb) - 12) | |
400 | return NULL; | |
401 | root = (struct dx_root_info *)(((void *)dp + 12)); | |
402 | if (root->reserved_zero || | |
403 | root->info_length != sizeof(struct dx_root_info)) | |
404 | return NULL; | |
405 | count_offset = 32; | |
406 | } else | |
407 | return NULL; | |
408 | ||
409 | if (offset) | |
410 | *offset = count_offset; | |
411 | return (struct dx_countlimit *)(((void *)dirent) + count_offset); | |
412 | } | |
413 | ||
414 | static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent, | |
415 | int count_offset, int count, struct dx_tail *t) | |
416 | { | |
417 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
418 | struct ext4_inode_info *ei = EXT4_I(inode); | |
d6a77105 TT |
419 | __u32 csum; |
420 | __le32 save_csum; | |
dbe89444 DW |
421 | int size; |
422 | ||
423 | size = count_offset + (count * sizeof(struct dx_entry)); | |
d6a77105 | 424 | save_csum = t->dt_checksum; |
dbe89444 DW |
425 | t->dt_checksum = 0; |
426 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size); | |
427 | csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail)); | |
d6a77105 | 428 | t->dt_checksum = save_csum; |
dbe89444 DW |
429 | |
430 | return cpu_to_le32(csum); | |
431 | } | |
432 | ||
433 | static int ext4_dx_csum_verify(struct inode *inode, | |
434 | struct ext4_dir_entry *dirent) | |
435 | { | |
436 | struct dx_countlimit *c; | |
437 | struct dx_tail *t; | |
438 | int count_offset, limit, count; | |
439 | ||
440 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
441 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
442 | return 1; | |
443 | ||
444 | c = get_dx_countlimit(inode, dirent, &count_offset); | |
445 | if (!c) { | |
446 | EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); | |
447 | return 1; | |
448 | } | |
449 | limit = le16_to_cpu(c->limit); | |
450 | count = le16_to_cpu(c->count); | |
451 | if (count_offset + (limit * sizeof(struct dx_entry)) > | |
452 | EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { | |
dffe9d8d | 453 | warn_no_space_for_csum(inode); |
dbe89444 DW |
454 | return 1; |
455 | } | |
456 | t = (struct dx_tail *)(((struct dx_entry *)c) + limit); | |
457 | ||
458 | if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset, | |
459 | count, t)) | |
460 | return 0; | |
461 | return 1; | |
462 | } | |
463 | ||
464 | static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent) | |
465 | { | |
466 | struct dx_countlimit *c; | |
467 | struct dx_tail *t; | |
468 | int count_offset, limit, count; | |
469 | ||
470 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
471 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
472 | return; | |
473 | ||
474 | c = get_dx_countlimit(inode, dirent, &count_offset); | |
475 | if (!c) { | |
476 | EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D."); | |
477 | return; | |
478 | } | |
479 | limit = le16_to_cpu(c->limit); | |
480 | count = le16_to_cpu(c->count); | |
481 | if (count_offset + (limit * sizeof(struct dx_entry)) > | |
482 | EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) { | |
dffe9d8d | 483 | warn_no_space_for_csum(inode); |
dbe89444 DW |
484 | return; |
485 | } | |
486 | t = (struct dx_tail *)(((struct dx_entry *)c) + limit); | |
487 | ||
488 | t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t); | |
489 | } | |
490 | ||
491 | static inline int ext4_handle_dirty_dx_node(handle_t *handle, | |
492 | struct inode *inode, | |
493 | struct buffer_head *bh) | |
494 | { | |
495 | ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data); | |
496 | return ext4_handle_dirty_metadata(handle, inode, bh); | |
497 | } | |
498 | ||
f795e140 LZ |
499 | /* |
500 | * p is at least 6 bytes before the end of page | |
501 | */ | |
502 | static inline struct ext4_dir_entry_2 * | |
3d0518f4 | 503 | ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize) |
f795e140 LZ |
504 | { |
505 | return (struct ext4_dir_entry_2 *)((char *)p + | |
3d0518f4 | 506 | ext4_rec_len_from_disk(p->rec_len, blocksize)); |
f795e140 LZ |
507 | } |
508 | ||
ac27a0ec DK |
509 | /* |
510 | * Future: use high four bits of block for coalesce-on-delete flags | |
511 | * Mask them off for now. | |
512 | */ | |
513 | ||
725d26d3 | 514 | static inline ext4_lblk_t dx_get_block(struct dx_entry *entry) |
ac27a0ec DK |
515 | { |
516 | return le32_to_cpu(entry->block) & 0x00ffffff; | |
517 | } | |
518 | ||
725d26d3 | 519 | static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value) |
ac27a0ec DK |
520 | { |
521 | entry->block = cpu_to_le32(value); | |
522 | } | |
523 | ||
af5bc92d | 524 | static inline unsigned dx_get_hash(struct dx_entry *entry) |
ac27a0ec DK |
525 | { |
526 | return le32_to_cpu(entry->hash); | |
527 | } | |
528 | ||
af5bc92d | 529 | static inline void dx_set_hash(struct dx_entry *entry, unsigned value) |
ac27a0ec DK |
530 | { |
531 | entry->hash = cpu_to_le32(value); | |
532 | } | |
533 | ||
af5bc92d | 534 | static inline unsigned dx_get_count(struct dx_entry *entries) |
ac27a0ec DK |
535 | { |
536 | return le16_to_cpu(((struct dx_countlimit *) entries)->count); | |
537 | } | |
538 | ||
af5bc92d | 539 | static inline unsigned dx_get_limit(struct dx_entry *entries) |
ac27a0ec DK |
540 | { |
541 | return le16_to_cpu(((struct dx_countlimit *) entries)->limit); | |
542 | } | |
543 | ||
af5bc92d | 544 | static inline void dx_set_count(struct dx_entry *entries, unsigned value) |
ac27a0ec DK |
545 | { |
546 | ((struct dx_countlimit *) entries)->count = cpu_to_le16(value); | |
547 | } | |
548 | ||
af5bc92d | 549 | static inline void dx_set_limit(struct dx_entry *entries, unsigned value) |
ac27a0ec DK |
550 | { |
551 | ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value); | |
552 | } | |
553 | ||
af5bc92d | 554 | static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize) |
ac27a0ec | 555 | { |
617ba13b MC |
556 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - |
557 | EXT4_DIR_REC_LEN(2) - infosize; | |
dbe89444 DW |
558 | |
559 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, | |
560 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
561 | entry_space -= sizeof(struct dx_tail); | |
d9c769b7 | 562 | return entry_space / sizeof(struct dx_entry); |
ac27a0ec DK |
563 | } |
564 | ||
af5bc92d | 565 | static inline unsigned dx_node_limit(struct inode *dir) |
ac27a0ec | 566 | { |
617ba13b | 567 | unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); |
dbe89444 DW |
568 | |
569 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, | |
570 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
571 | entry_space -= sizeof(struct dx_tail); | |
d9c769b7 | 572 | return entry_space / sizeof(struct dx_entry); |
ac27a0ec DK |
573 | } |
574 | ||
575 | /* | |
576 | * Debug | |
577 | */ | |
578 | #ifdef DX_DEBUG | |
4776004f | 579 | static void dx_show_index(char * label, struct dx_entry *entries) |
ac27a0ec | 580 | { |
63f57933 | 581 | int i, n = dx_get_count (entries); |
4776004f | 582 | printk(KERN_DEBUG "%s index ", label); |
63f57933 | 583 | for (i = 0; i < n; i++) { |
4776004f | 584 | printk("%x->%lu ", i ? dx_get_hash(entries + i) : |
725d26d3 | 585 | 0, (unsigned long)dx_get_block(entries + i)); |
63f57933 AM |
586 | } |
587 | printk("\n"); | |
ac27a0ec DK |
588 | } |
589 | ||
590 | struct stats | |
591 | { | |
592 | unsigned names; | |
593 | unsigned space; | |
594 | unsigned bcount; | |
595 | }; | |
596 | ||
617ba13b | 597 | static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de, |
ac27a0ec DK |
598 | int size, int show_names) |
599 | { | |
600 | unsigned names = 0, space = 0; | |
601 | char *base = (char *) de; | |
602 | struct dx_hash_info h = *hinfo; | |
603 | ||
604 | printk("names: "); | |
605 | while ((char *) de < base + size) | |
606 | { | |
607 | if (de->inode) | |
608 | { | |
609 | if (show_names) | |
610 | { | |
611 | int len = de->name_len; | |
612 | char *name = de->name; | |
613 | while (len--) printk("%c", *name++); | |
617ba13b | 614 | ext4fs_dirhash(de->name, de->name_len, &h); |
ac27a0ec | 615 | printk(":%x.%u ", h.hash, |
265c6a0f | 616 | (unsigned) ((char *) de - base)); |
ac27a0ec | 617 | } |
617ba13b | 618 | space += EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec DK |
619 | names++; |
620 | } | |
3d0518f4 | 621 | de = ext4_next_entry(de, size); |
ac27a0ec DK |
622 | } |
623 | printk("(%i)\n", names); | |
624 | return (struct stats) { names, space, 1 }; | |
625 | } | |
626 | ||
627 | struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir, | |
628 | struct dx_entry *entries, int levels) | |
629 | { | |
630 | unsigned blocksize = dir->i_sb->s_blocksize; | |
af5bc92d | 631 | unsigned count = dx_get_count(entries), names = 0, space = 0, i; |
ac27a0ec DK |
632 | unsigned bcount = 0; |
633 | struct buffer_head *bh; | |
634 | int err; | |
635 | printk("%i indexed blocks...\n", count); | |
636 | for (i = 0; i < count; i++, entries++) | |
637 | { | |
725d26d3 AK |
638 | ext4_lblk_t block = dx_get_block(entries); |
639 | ext4_lblk_t hash = i ? dx_get_hash(entries): 0; | |
ac27a0ec DK |
640 | u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash; |
641 | struct stats stats; | |
642 | printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range); | |
617ba13b | 643 | if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue; |
ac27a0ec DK |
644 | stats = levels? |
645 | dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1): | |
617ba13b | 646 | dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0); |
ac27a0ec DK |
647 | names += stats.names; |
648 | space += stats.space; | |
649 | bcount += stats.bcount; | |
af5bc92d | 650 | brelse(bh); |
ac27a0ec DK |
651 | } |
652 | if (bcount) | |
60e6679e | 653 | printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n", |
4776004f TT |
654 | levels ? "" : " ", names, space/bcount, |
655 | (space/bcount)*100/blocksize); | |
ac27a0ec DK |
656 | return (struct stats) { names, space, bcount}; |
657 | } | |
658 | #endif /* DX_DEBUG */ | |
659 | ||
660 | /* | |
661 | * Probe for a directory leaf block to search. | |
662 | * | |
663 | * dx_probe can return ERR_BAD_DX_DIR, which means there was a format | |
664 | * error in the directory index, and the caller should fall back to | |
665 | * searching the directory normally. The callers of dx_probe **MUST** | |
666 | * check for this error code, and make sure it never gets reflected | |
667 | * back to userspace. | |
668 | */ | |
669 | static struct dx_frame * | |
f702ba0f | 670 | dx_probe(const struct qstr *d_name, struct inode *dir, |
ac27a0ec DK |
671 | struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err) |
672 | { | |
673 | unsigned count, indirect; | |
674 | struct dx_entry *at, *entries, *p, *q, *m; | |
675 | struct dx_root *root; | |
676 | struct buffer_head *bh; | |
677 | struct dx_frame *frame = frame_in; | |
678 | u32 hash; | |
679 | ||
680 | frame->bh = NULL; | |
dc6982ff TT |
681 | bh = ext4_read_dirblock(dir, 0, INDEX); |
682 | if (IS_ERR(bh)) { | |
683 | *err = PTR_ERR(bh); | |
ac27a0ec | 684 | goto fail; |
6d1ab10e | 685 | } |
ac27a0ec DK |
686 | root = (struct dx_root *) bh->b_data; |
687 | if (root->info.hash_version != DX_HASH_TEA && | |
688 | root->info.hash_version != DX_HASH_HALF_MD4 && | |
689 | root->info.hash_version != DX_HASH_LEGACY) { | |
12062ddd | 690 | ext4_warning(dir->i_sb, "Unrecognised inode hash code %d", |
ac27a0ec DK |
691 | root->info.hash_version); |
692 | brelse(bh); | |
693 | *err = ERR_BAD_DX_DIR; | |
694 | goto fail; | |
695 | } | |
696 | hinfo->hash_version = root->info.hash_version; | |
f99b2589 TT |
697 | if (hinfo->hash_version <= DX_HASH_TEA) |
698 | hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b | 699 | hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
f702ba0f TT |
700 | if (d_name) |
701 | ext4fs_dirhash(d_name->name, d_name->len, hinfo); | |
ac27a0ec DK |
702 | hash = hinfo->hash; |
703 | ||
704 | if (root->info.unused_flags & 1) { | |
12062ddd | 705 | ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x", |
ac27a0ec DK |
706 | root->info.unused_flags); |
707 | brelse(bh); | |
708 | *err = ERR_BAD_DX_DIR; | |
709 | goto fail; | |
710 | } | |
711 | ||
712 | if ((indirect = root->info.indirect_levels) > 1) { | |
12062ddd | 713 | ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x", |
ac27a0ec DK |
714 | root->info.indirect_levels); |
715 | brelse(bh); | |
716 | *err = ERR_BAD_DX_DIR; | |
717 | goto fail; | |
718 | } | |
719 | ||
720 | entries = (struct dx_entry *) (((char *)&root->info) + | |
721 | root->info.info_length); | |
3d82abae ES |
722 | |
723 | if (dx_get_limit(entries) != dx_root_limit(dir, | |
724 | root->info.info_length)) { | |
12062ddd | 725 | ext4_warning(dir->i_sb, "dx entry: limit != root limit"); |
3d82abae ES |
726 | brelse(bh); |
727 | *err = ERR_BAD_DX_DIR; | |
728 | goto fail; | |
729 | } | |
730 | ||
af5bc92d | 731 | dxtrace(printk("Look up %x", hash)); |
ac27a0ec DK |
732 | while (1) |
733 | { | |
734 | count = dx_get_count(entries); | |
3d82abae | 735 | if (!count || count > dx_get_limit(entries)) { |
12062ddd | 736 | ext4_warning(dir->i_sb, |
3d82abae ES |
737 | "dx entry: no count or count > limit"); |
738 | brelse(bh); | |
739 | *err = ERR_BAD_DX_DIR; | |
740 | goto fail2; | |
741 | } | |
742 | ||
ac27a0ec DK |
743 | p = entries + 1; |
744 | q = entries + count - 1; | |
745 | while (p <= q) | |
746 | { | |
747 | m = p + (q - p)/2; | |
748 | dxtrace(printk(".")); | |
749 | if (dx_get_hash(m) > hash) | |
750 | q = m - 1; | |
751 | else | |
752 | p = m + 1; | |
753 | } | |
754 | ||
755 | if (0) // linear search cross check | |
756 | { | |
757 | unsigned n = count - 1; | |
758 | at = entries; | |
759 | while (n--) | |
760 | { | |
761 | dxtrace(printk(",")); | |
762 | if (dx_get_hash(++at) > hash) | |
763 | { | |
764 | at--; | |
765 | break; | |
766 | } | |
767 | } | |
768 | assert (at == p - 1); | |
769 | } | |
770 | ||
771 | at = p - 1; | |
772 | dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at))); | |
773 | frame->bh = bh; | |
774 | frame->entries = entries; | |
775 | frame->at = at; | |
776 | if (!indirect--) return frame; | |
dc6982ff TT |
777 | bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX); |
778 | if (IS_ERR(bh)) { | |
779 | *err = PTR_ERR(bh); | |
0ecaef06 | 780 | goto fail2; |
dbe89444 | 781 | } |
41be871f | 782 | entries = ((struct dx_node *) bh->b_data)->entries; |
dbe89444 | 783 | |
3d82abae | 784 | if (dx_get_limit(entries) != dx_node_limit (dir)) { |
12062ddd | 785 | ext4_warning(dir->i_sb, |
3d82abae ES |
786 | "dx entry: limit != node limit"); |
787 | brelse(bh); | |
788 | *err = ERR_BAD_DX_DIR; | |
789 | goto fail2; | |
790 | } | |
ac27a0ec | 791 | frame++; |
3d82abae | 792 | frame->bh = NULL; |
ac27a0ec DK |
793 | } |
794 | fail2: | |
795 | while (frame >= frame_in) { | |
796 | brelse(frame->bh); | |
797 | frame--; | |
798 | } | |
799 | fail: | |
3d82abae | 800 | if (*err == ERR_BAD_DX_DIR) |
12062ddd | 801 | ext4_warning(dir->i_sb, |
9ee49302 | 802 | "Corrupt dir inode %lu, running e2fsck is " |
3d82abae | 803 | "recommended.", dir->i_ino); |
ac27a0ec DK |
804 | return NULL; |
805 | } | |
806 | ||
807 | static void dx_release (struct dx_frame *frames) | |
808 | { | |
809 | if (frames[0].bh == NULL) | |
810 | return; | |
811 | ||
812 | if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels) | |
813 | brelse(frames[1].bh); | |
814 | brelse(frames[0].bh); | |
815 | } | |
816 | ||
817 | /* | |
818 | * This function increments the frame pointer to search the next leaf | |
819 | * block, and reads in the necessary intervening nodes if the search | |
820 | * should be necessary. Whether or not the search is necessary is | |
821 | * controlled by the hash parameter. If the hash value is even, then | |
822 | * the search is only continued if the next block starts with that | |
823 | * hash value. This is used if we are searching for a specific file. | |
824 | * | |
825 | * If the hash value is HASH_NB_ALWAYS, then always go to the next block. | |
826 | * | |
827 | * This function returns 1 if the caller should continue to search, | |
828 | * or 0 if it should not. If there is an error reading one of the | |
829 | * index blocks, it will a negative error code. | |
830 | * | |
831 | * If start_hash is non-null, it will be filled in with the starting | |
832 | * hash of the next page. | |
833 | */ | |
617ba13b | 834 | static int ext4_htree_next_block(struct inode *dir, __u32 hash, |
ac27a0ec DK |
835 | struct dx_frame *frame, |
836 | struct dx_frame *frames, | |
837 | __u32 *start_hash) | |
838 | { | |
839 | struct dx_frame *p; | |
840 | struct buffer_head *bh; | |
dc6982ff | 841 | int num_frames = 0; |
ac27a0ec DK |
842 | __u32 bhash; |
843 | ||
844 | p = frame; | |
845 | /* | |
846 | * Find the next leaf page by incrementing the frame pointer. | |
847 | * If we run out of entries in the interior node, loop around and | |
848 | * increment pointer in the parent node. When we break out of | |
849 | * this loop, num_frames indicates the number of interior | |
850 | * nodes need to be read. | |
851 | */ | |
852 | while (1) { | |
853 | if (++(p->at) < p->entries + dx_get_count(p->entries)) | |
854 | break; | |
855 | if (p == frames) | |
856 | return 0; | |
857 | num_frames++; | |
858 | p--; | |
859 | } | |
860 | ||
861 | /* | |
862 | * If the hash is 1, then continue only if the next page has a | |
863 | * continuation hash of any value. This is used for readdir | |
864 | * handling. Otherwise, check to see if the hash matches the | |
865 | * desired contiuation hash. If it doesn't, return since | |
866 | * there's no point to read in the successive index pages. | |
867 | */ | |
868 | bhash = dx_get_hash(p->at); | |
869 | if (start_hash) | |
870 | *start_hash = bhash; | |
871 | if ((hash & 1) == 0) { | |
872 | if ((bhash & ~1) != hash) | |
873 | return 0; | |
874 | } | |
875 | /* | |
876 | * If the hash is HASH_NB_ALWAYS, we always go to the next | |
877 | * block so no check is necessary | |
878 | */ | |
879 | while (num_frames--) { | |
dc6982ff TT |
880 | bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX); |
881 | if (IS_ERR(bh)) | |
882 | return PTR_ERR(bh); | |
ac27a0ec | 883 | p++; |
af5bc92d | 884 | brelse(p->bh); |
ac27a0ec DK |
885 | p->bh = bh; |
886 | p->at = p->entries = ((struct dx_node *) bh->b_data)->entries; | |
887 | } | |
888 | return 1; | |
889 | } | |
890 | ||
891 | ||
ac27a0ec DK |
892 | /* |
893 | * This function fills a red-black tree with information from a | |
894 | * directory block. It returns the number directory entries loaded | |
895 | * into the tree. If there is an error it is returned in err. | |
896 | */ | |
897 | static int htree_dirblock_to_tree(struct file *dir_file, | |
725d26d3 | 898 | struct inode *dir, ext4_lblk_t block, |
ac27a0ec DK |
899 | struct dx_hash_info *hinfo, |
900 | __u32 start_hash, __u32 start_minor_hash) | |
901 | { | |
902 | struct buffer_head *bh; | |
617ba13b | 903 | struct ext4_dir_entry_2 *de, *top; |
90b0a973 | 904 | int err = 0, count = 0; |
ac27a0ec | 905 | |
725d26d3 AK |
906 | dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n", |
907 | (unsigned long)block)); | |
dc6982ff TT |
908 | bh = ext4_read_dirblock(dir, block, DIRENT); |
909 | if (IS_ERR(bh)) | |
910 | return PTR_ERR(bh); | |
b0336e8d | 911 | |
617ba13b MC |
912 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
913 | top = (struct ext4_dir_entry_2 *) ((char *) de + | |
ac27a0ec | 914 | dir->i_sb->s_blocksize - |
617ba13b | 915 | EXT4_DIR_REC_LEN(0)); |
3d0518f4 | 916 | for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) { |
f7c21177 | 917 | if (ext4_check_dir_entry(dir, NULL, de, bh, |
226ba972 | 918 | bh->b_data, bh->b_size, |
cad3f007 TT |
919 | (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb)) |
920 | + ((char *)de - bh->b_data))) { | |
e6c40211 ES |
921 | /* On error, skip the f_pos to the next block. */ |
922 | dir_file->f_pos = (dir_file->f_pos | | |
923 | (dir->i_sb->s_blocksize - 1)) + 1; | |
af5bc92d | 924 | brelse(bh); |
e6c40211 ES |
925 | return count; |
926 | } | |
617ba13b | 927 | ext4fs_dirhash(de->name, de->name_len, hinfo); |
ac27a0ec DK |
928 | if ((hinfo->hash < start_hash) || |
929 | ((hinfo->hash == start_hash) && | |
930 | (hinfo->minor_hash < start_minor_hash))) | |
931 | continue; | |
932 | if (de->inode == 0) | |
933 | continue; | |
617ba13b | 934 | if ((err = ext4_htree_store_dirent(dir_file, |
ac27a0ec DK |
935 | hinfo->hash, hinfo->minor_hash, de)) != 0) { |
936 | brelse(bh); | |
937 | return err; | |
938 | } | |
939 | count++; | |
940 | } | |
941 | brelse(bh); | |
942 | return count; | |
943 | } | |
944 | ||
945 | ||
946 | /* | |
947 | * This function fills a red-black tree with information from a | |
948 | * directory. We start scanning the directory in hash order, starting | |
949 | * at start_hash and start_minor_hash. | |
950 | * | |
951 | * This function returns the number of entries inserted into the tree, | |
952 | * or a negative error code. | |
953 | */ | |
617ba13b | 954 | int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash, |
ac27a0ec DK |
955 | __u32 start_minor_hash, __u32 *next_hash) |
956 | { | |
957 | struct dx_hash_info hinfo; | |
617ba13b | 958 | struct ext4_dir_entry_2 *de; |
ac27a0ec DK |
959 | struct dx_frame frames[2], *frame; |
960 | struct inode *dir; | |
725d26d3 | 961 | ext4_lblk_t block; |
ac27a0ec | 962 | int count = 0; |
725d26d3 | 963 | int ret, err; |
ac27a0ec DK |
964 | __u32 hashval; |
965 | ||
60e6679e | 966 | dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n", |
4776004f | 967 | start_hash, start_minor_hash)); |
496ad9aa | 968 | dir = file_inode(dir_file); |
12e9b892 | 969 | if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) { |
617ba13b | 970 | hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; |
f99b2589 TT |
971 | if (hinfo.hash_version <= DX_HASH_TEA) |
972 | hinfo.hash_version += | |
973 | EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b | 974 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
ac27a0ec DK |
975 | count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo, |
976 | start_hash, start_minor_hash); | |
977 | *next_hash = ~0; | |
978 | return count; | |
979 | } | |
980 | hinfo.hash = start_hash; | |
981 | hinfo.minor_hash = 0; | |
f702ba0f | 982 | frame = dx_probe(NULL, dir, &hinfo, frames, &err); |
ac27a0ec DK |
983 | if (!frame) |
984 | return err; | |
985 | ||
986 | /* Add '.' and '..' from the htree header */ | |
987 | if (!start_hash && !start_minor_hash) { | |
617ba13b MC |
988 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; |
989 | if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0) | |
ac27a0ec DK |
990 | goto errout; |
991 | count++; | |
992 | } | |
993 | if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) { | |
617ba13b | 994 | de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data; |
3d0518f4 | 995 | de = ext4_next_entry(de, dir->i_sb->s_blocksize); |
617ba13b | 996 | if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0) |
ac27a0ec DK |
997 | goto errout; |
998 | count++; | |
999 | } | |
1000 | ||
1001 | while (1) { | |
1002 | block = dx_get_block(frame->at); | |
1003 | ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo, | |
1004 | start_hash, start_minor_hash); | |
1005 | if (ret < 0) { | |
1006 | err = ret; | |
1007 | goto errout; | |
1008 | } | |
1009 | count += ret; | |
1010 | hashval = ~0; | |
617ba13b | 1011 | ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS, |
ac27a0ec DK |
1012 | frame, frames, &hashval); |
1013 | *next_hash = hashval; | |
1014 | if (ret < 0) { | |
1015 | err = ret; | |
1016 | goto errout; | |
1017 | } | |
1018 | /* | |
1019 | * Stop if: (a) there are no more entries, or | |
1020 | * (b) we have inserted at least one entry and the | |
1021 | * next hash value is not a continuation | |
1022 | */ | |
1023 | if ((ret == 0) || | |
1024 | (count && ((hashval & 1) == 0))) | |
1025 | break; | |
1026 | } | |
1027 | dx_release(frames); | |
4776004f TT |
1028 | dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, " |
1029 | "next hash: %x\n", count, *next_hash)); | |
ac27a0ec DK |
1030 | return count; |
1031 | errout: | |
1032 | dx_release(frames); | |
1033 | return (err); | |
1034 | } | |
1035 | ||
7335cd3b TM |
1036 | static inline int search_dirblock(struct buffer_head *bh, |
1037 | struct inode *dir, | |
1038 | const struct qstr *d_name, | |
1039 | unsigned int offset, | |
1040 | struct ext4_dir_entry_2 **res_dir) | |
1041 | { | |
1042 | return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir, | |
1043 | d_name, offset, res_dir); | |
1044 | } | |
1045 | ||
ac27a0ec DK |
1046 | /* |
1047 | * Directory block splitting, compacting | |
1048 | */ | |
1049 | ||
ef2b02d3 ES |
1050 | /* |
1051 | * Create map of hash values, offsets, and sizes, stored at end of block. | |
1052 | * Returns number of entries mapped. | |
1053 | */ | |
8bad4597 TT |
1054 | static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize, |
1055 | struct dx_hash_info *hinfo, | |
1056 | struct dx_map_entry *map_tail) | |
ac27a0ec DK |
1057 | { |
1058 | int count = 0; | |
1059 | char *base = (char *) de; | |
1060 | struct dx_hash_info h = *hinfo; | |
1061 | ||
8bad4597 | 1062 | while ((char *) de < base + blocksize) { |
ac27a0ec | 1063 | if (de->name_len && de->inode) { |
617ba13b | 1064 | ext4fs_dirhash(de->name, de->name_len, &h); |
ac27a0ec DK |
1065 | map_tail--; |
1066 | map_tail->hash = h.hash; | |
9aee2286 | 1067 | map_tail->offs = ((char *) de - base)>>2; |
ef2b02d3 | 1068 | map_tail->size = le16_to_cpu(de->rec_len); |
ac27a0ec DK |
1069 | count++; |
1070 | cond_resched(); | |
1071 | } | |
1072 | /* XXX: do we need to check rec_len == 0 case? -Chris */ | |
3d0518f4 | 1073 | de = ext4_next_entry(de, blocksize); |
ac27a0ec DK |
1074 | } |
1075 | return count; | |
1076 | } | |
1077 | ||
ef2b02d3 | 1078 | /* Sort map by hash value */ |
ac27a0ec DK |
1079 | static void dx_sort_map (struct dx_map_entry *map, unsigned count) |
1080 | { | |
63f57933 AM |
1081 | struct dx_map_entry *p, *q, *top = map + count - 1; |
1082 | int more; | |
1083 | /* Combsort until bubble sort doesn't suck */ | |
1084 | while (count > 2) { | |
1085 | count = count*10/13; | |
1086 | if (count - 9 < 2) /* 9, 10 -> 11 */ | |
1087 | count = 11; | |
1088 | for (p = top, q = p - count; q >= map; p--, q--) | |
1089 | if (p->hash < q->hash) | |
1090 | swap(*p, *q); | |
1091 | } | |
1092 | /* Garden variety bubble sort */ | |
1093 | do { | |
1094 | more = 0; | |
1095 | q = top; | |
1096 | while (q-- > map) { | |
1097 | if (q[1].hash >= q[0].hash) | |
ac27a0ec | 1098 | continue; |
63f57933 AM |
1099 | swap(*(q+1), *q); |
1100 | more = 1; | |
ac27a0ec DK |
1101 | } |
1102 | } while(more); | |
1103 | } | |
1104 | ||
725d26d3 | 1105 | static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block) |
ac27a0ec DK |
1106 | { |
1107 | struct dx_entry *entries = frame->entries; | |
1108 | struct dx_entry *old = frame->at, *new = old + 1; | |
1109 | int count = dx_get_count(entries); | |
1110 | ||
1111 | assert(count < dx_get_limit(entries)); | |
1112 | assert(old < entries + count); | |
1113 | memmove(new + 1, new, (char *)(entries + count) - (char *)(new)); | |
1114 | dx_set_hash(new, hash); | |
1115 | dx_set_block(new, block); | |
1116 | dx_set_count(entries, count + 1); | |
1117 | } | |
ac27a0ec | 1118 | |
ac27a0ec | 1119 | /* |
617ba13b | 1120 | * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure. |
ac27a0ec | 1121 | * |
617ba13b | 1122 | * `len <= EXT4_NAME_LEN' is guaranteed by caller. |
ac27a0ec DK |
1123 | * `de != NULL' is guaranteed by caller. |
1124 | */ | |
617ba13b MC |
1125 | static inline int ext4_match (int len, const char * const name, |
1126 | struct ext4_dir_entry_2 * de) | |
ac27a0ec DK |
1127 | { |
1128 | if (len != de->name_len) | |
1129 | return 0; | |
1130 | if (!de->inode) | |
1131 | return 0; | |
1132 | return !memcmp(name, de->name, len); | |
1133 | } | |
1134 | ||
1135 | /* | |
1136 | * Returns 0 if not found, -1 on failure, and 1 on success | |
1137 | */ | |
7335cd3b TM |
1138 | int search_dir(struct buffer_head *bh, |
1139 | char *search_buf, | |
1140 | int buf_size, | |
1141 | struct inode *dir, | |
1142 | const struct qstr *d_name, | |
1143 | unsigned int offset, | |
1144 | struct ext4_dir_entry_2 **res_dir) | |
ac27a0ec | 1145 | { |
617ba13b | 1146 | struct ext4_dir_entry_2 * de; |
ac27a0ec DK |
1147 | char * dlimit; |
1148 | int de_len; | |
f702ba0f TT |
1149 | const char *name = d_name->name; |
1150 | int namelen = d_name->len; | |
ac27a0ec | 1151 | |
7335cd3b TM |
1152 | de = (struct ext4_dir_entry_2 *)search_buf; |
1153 | dlimit = search_buf + buf_size; | |
ac27a0ec DK |
1154 | while ((char *) de < dlimit) { |
1155 | /* this code is executed quadratically often */ | |
1156 | /* do minimal checking `by hand' */ | |
1157 | ||
1158 | if ((char *) de + namelen <= dlimit && | |
617ba13b | 1159 | ext4_match (namelen, name, de)) { |
ac27a0ec | 1160 | /* found a match - just to be sure, do a full check */ |
226ba972 TM |
1161 | if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data, |
1162 | bh->b_size, offset)) | |
ac27a0ec DK |
1163 | return -1; |
1164 | *res_dir = de; | |
1165 | return 1; | |
1166 | } | |
1167 | /* prevent looping on a bad block */ | |
3d0518f4 WY |
1168 | de_len = ext4_rec_len_from_disk(de->rec_len, |
1169 | dir->i_sb->s_blocksize); | |
ac27a0ec DK |
1170 | if (de_len <= 0) |
1171 | return -1; | |
1172 | offset += de_len; | |
617ba13b | 1173 | de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); |
ac27a0ec DK |
1174 | } |
1175 | return 0; | |
1176 | } | |
1177 | ||
c6af8803 DW |
1178 | static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block, |
1179 | struct ext4_dir_entry *de) | |
1180 | { | |
1181 | struct super_block *sb = dir->i_sb; | |
1182 | ||
1183 | if (!is_dx(dir)) | |
1184 | return 0; | |
1185 | if (block == 0) | |
1186 | return 1; | |
1187 | if (de->inode == 0 && | |
1188 | ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) == | |
1189 | sb->s_blocksize) | |
1190 | return 1; | |
1191 | return 0; | |
1192 | } | |
ac27a0ec DK |
1193 | |
1194 | /* | |
617ba13b | 1195 | * ext4_find_entry() |
ac27a0ec DK |
1196 | * |
1197 | * finds an entry in the specified directory with the wanted name. It | |
1198 | * returns the cache buffer in which the entry was found, and the entry | |
1199 | * itself (as a parameter - res_dir). It does NOT read the inode of the | |
1200 | * entry - you'll have to do that yourself if you want to. | |
1201 | * | |
1202 | * The returned buffer_head has ->b_count elevated. The caller is expected | |
1203 | * to brelse() it when appropriate. | |
1204 | */ | |
f702ba0f TT |
1205 | static struct buffer_head * ext4_find_entry (struct inode *dir, |
1206 | const struct qstr *d_name, | |
32f7f22c TM |
1207 | struct ext4_dir_entry_2 **res_dir, |
1208 | int *inlined) | |
ac27a0ec | 1209 | { |
af5bc92d TT |
1210 | struct super_block *sb; |
1211 | struct buffer_head *bh_use[NAMEI_RA_SIZE]; | |
1212 | struct buffer_head *bh, *ret = NULL; | |
725d26d3 | 1213 | ext4_lblk_t start, block, b; |
8941ec8b | 1214 | const u8 *name = d_name->name; |
ac27a0ec DK |
1215 | int ra_max = 0; /* Number of bh's in the readahead |
1216 | buffer, bh_use[] */ | |
1217 | int ra_ptr = 0; /* Current index into readahead | |
1218 | buffer */ | |
1219 | int num = 0; | |
725d26d3 AK |
1220 | ext4_lblk_t nblocks; |
1221 | int i, err; | |
ac27a0ec | 1222 | int namelen; |
ac27a0ec DK |
1223 | |
1224 | *res_dir = NULL; | |
1225 | sb = dir->i_sb; | |
f702ba0f | 1226 | namelen = d_name->len; |
617ba13b | 1227 | if (namelen > EXT4_NAME_LEN) |
ac27a0ec | 1228 | return NULL; |
e8e948e7 TM |
1229 | |
1230 | if (ext4_has_inline_data(dir)) { | |
1231 | int has_inline_data = 1; | |
1232 | ret = ext4_find_inline_entry(dir, d_name, res_dir, | |
1233 | &has_inline_data); | |
32f7f22c TM |
1234 | if (has_inline_data) { |
1235 | if (inlined) | |
1236 | *inlined = 1; | |
e8e948e7 | 1237 | return ret; |
32f7f22c | 1238 | } |
e8e948e7 TM |
1239 | } |
1240 | ||
8941ec8b | 1241 | if ((namelen <= 2) && (name[0] == '.') && |
6d5c3aa8 | 1242 | (name[1] == '.' || name[1] == '\0')) { |
8941ec8b TT |
1243 | /* |
1244 | * "." or ".." will only be in the first block | |
1245 | * NFS may look up ".."; "." should be handled by the VFS | |
1246 | */ | |
1247 | block = start = 0; | |
1248 | nblocks = 1; | |
1249 | goto restart; | |
1250 | } | |
ac27a0ec | 1251 | if (is_dx(dir)) { |
f702ba0f | 1252 | bh = ext4_dx_find_entry(dir, d_name, res_dir, &err); |
ac27a0ec DK |
1253 | /* |
1254 | * On success, or if the error was file not found, | |
1255 | * return. Otherwise, fall back to doing a search the | |
1256 | * old fashioned way. | |
1257 | */ | |
1258 | if (bh || (err != ERR_BAD_DX_DIR)) | |
1259 | return bh; | |
4776004f TT |
1260 | dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, " |
1261 | "falling back\n")); | |
ac27a0ec | 1262 | } |
617ba13b MC |
1263 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); |
1264 | start = EXT4_I(dir)->i_dir_start_lookup; | |
ac27a0ec DK |
1265 | if (start >= nblocks) |
1266 | start = 0; | |
1267 | block = start; | |
1268 | restart: | |
1269 | do { | |
1270 | /* | |
1271 | * We deal with the read-ahead logic here. | |
1272 | */ | |
1273 | if (ra_ptr >= ra_max) { | |
1274 | /* Refill the readahead buffer */ | |
1275 | ra_ptr = 0; | |
1276 | b = block; | |
1277 | for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) { | |
1278 | /* | |
1279 | * Terminate if we reach the end of the | |
1280 | * directory and must wrap, or if our | |
1281 | * search has finished at this block. | |
1282 | */ | |
1283 | if (b >= nblocks || (num && block == start)) { | |
1284 | bh_use[ra_max] = NULL; | |
1285 | break; | |
1286 | } | |
1287 | num++; | |
617ba13b | 1288 | bh = ext4_getblk(NULL, dir, b++, 0, &err); |
ac27a0ec DK |
1289 | bh_use[ra_max] = bh; |
1290 | if (bh) | |
65299a3b CH |
1291 | ll_rw_block(READ | REQ_META | REQ_PRIO, |
1292 | 1, &bh); | |
ac27a0ec DK |
1293 | } |
1294 | } | |
1295 | if ((bh = bh_use[ra_ptr++]) == NULL) | |
1296 | goto next; | |
1297 | wait_on_buffer(bh); | |
1298 | if (!buffer_uptodate(bh)) { | |
1299 | /* read error, skip block & hope for the best */ | |
24676da4 TT |
1300 | EXT4_ERROR_INODE(dir, "reading directory lblock %lu", |
1301 | (unsigned long) block); | |
ac27a0ec DK |
1302 | brelse(bh); |
1303 | goto next; | |
1304 | } | |
b0336e8d | 1305 | if (!buffer_verified(bh) && |
c6af8803 DW |
1306 | !is_dx_internal_node(dir, block, |
1307 | (struct ext4_dir_entry *)bh->b_data) && | |
b0336e8d DW |
1308 | !ext4_dirent_csum_verify(dir, |
1309 | (struct ext4_dir_entry *)bh->b_data)) { | |
1310 | EXT4_ERROR_INODE(dir, "checksumming directory " | |
1311 | "block %lu", (unsigned long)block); | |
1312 | brelse(bh); | |
1313 | goto next; | |
1314 | } | |
1315 | set_buffer_verified(bh); | |
f702ba0f | 1316 | i = search_dirblock(bh, dir, d_name, |
617ba13b | 1317 | block << EXT4_BLOCK_SIZE_BITS(sb), res_dir); |
ac27a0ec | 1318 | if (i == 1) { |
617ba13b | 1319 | EXT4_I(dir)->i_dir_start_lookup = block; |
ac27a0ec DK |
1320 | ret = bh; |
1321 | goto cleanup_and_exit; | |
1322 | } else { | |
1323 | brelse(bh); | |
1324 | if (i < 0) | |
1325 | goto cleanup_and_exit; | |
1326 | } | |
1327 | next: | |
1328 | if (++block >= nblocks) | |
1329 | block = 0; | |
1330 | } while (block != start); | |
1331 | ||
1332 | /* | |
1333 | * If the directory has grown while we were searching, then | |
1334 | * search the last part of the directory before giving up. | |
1335 | */ | |
1336 | block = nblocks; | |
617ba13b | 1337 | nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb); |
ac27a0ec DK |
1338 | if (block < nblocks) { |
1339 | start = 0; | |
1340 | goto restart; | |
1341 | } | |
1342 | ||
1343 | cleanup_and_exit: | |
1344 | /* Clean up the read-ahead blocks */ | |
1345 | for (; ra_ptr < ra_max; ra_ptr++) | |
af5bc92d | 1346 | brelse(bh_use[ra_ptr]); |
ac27a0ec DK |
1347 | return ret; |
1348 | } | |
1349 | ||
f702ba0f | 1350 | static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name, |
617ba13b | 1351 | struct ext4_dir_entry_2 **res_dir, int *err) |
ac27a0ec | 1352 | { |
8941ec8b | 1353 | struct super_block * sb = dir->i_sb; |
ac27a0ec | 1354 | struct dx_hash_info hinfo; |
ac27a0ec | 1355 | struct dx_frame frames[2], *frame; |
ac27a0ec | 1356 | struct buffer_head *bh; |
725d26d3 | 1357 | ext4_lblk_t block; |
ac27a0ec | 1358 | int retval; |
ac27a0ec | 1359 | |
8941ec8b TT |
1360 | if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err))) |
1361 | return NULL; | |
ac27a0ec DK |
1362 | do { |
1363 | block = dx_get_block(frame->at); | |
dc6982ff TT |
1364 | bh = ext4_read_dirblock(dir, block, DIRENT); |
1365 | if (IS_ERR(bh)) { | |
1366 | *err = PTR_ERR(bh); | |
b0336e8d DW |
1367 | goto errout; |
1368 | } | |
7845c049 TT |
1369 | retval = search_dirblock(bh, dir, d_name, |
1370 | block << EXT4_BLOCK_SIZE_BITS(sb), | |
1371 | res_dir); | |
1372 | if (retval == 1) { /* Success! */ | |
1373 | dx_release(frames); | |
1374 | return bh; | |
ac27a0ec | 1375 | } |
af5bc92d | 1376 | brelse(bh); |
7845c049 TT |
1377 | if (retval == -1) { |
1378 | *err = ERR_BAD_DX_DIR; | |
1379 | goto errout; | |
1380 | } | |
1381 | ||
ac27a0ec | 1382 | /* Check to see if we should continue to search */ |
8941ec8b | 1383 | retval = ext4_htree_next_block(dir, hinfo.hash, frame, |
ac27a0ec DK |
1384 | frames, NULL); |
1385 | if (retval < 0) { | |
12062ddd | 1386 | ext4_warning(sb, |
ac27a0ec DK |
1387 | "error reading index page in directory #%lu", |
1388 | dir->i_ino); | |
1389 | *err = retval; | |
1390 | goto errout; | |
1391 | } | |
1392 | } while (retval == 1); | |
1393 | ||
1394 | *err = -ENOENT; | |
1395 | errout: | |
265c6a0f | 1396 | dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name)); |
ac27a0ec DK |
1397 | dx_release (frames); |
1398 | return NULL; | |
1399 | } | |
ac27a0ec | 1400 | |
00cd8dd3 | 1401 | static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) |
ac27a0ec | 1402 | { |
af5bc92d TT |
1403 | struct inode *inode; |
1404 | struct ext4_dir_entry_2 *de; | |
1405 | struct buffer_head *bh; | |
ac27a0ec | 1406 | |
617ba13b | 1407 | if (dentry->d_name.len > EXT4_NAME_LEN) |
ac27a0ec DK |
1408 | return ERR_PTR(-ENAMETOOLONG); |
1409 | ||
32f7f22c | 1410 | bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); |
ac27a0ec DK |
1411 | inode = NULL; |
1412 | if (bh) { | |
498e5f24 | 1413 | __u32 ino = le32_to_cpu(de->inode); |
af5bc92d | 1414 | brelse(bh); |
617ba13b | 1415 | if (!ext4_valid_inum(dir->i_sb, ino)) { |
24676da4 | 1416 | EXT4_ERROR_INODE(dir, "bad inode number: %u", ino); |
1d1fe1ee | 1417 | return ERR_PTR(-EIO); |
a6c15c2b | 1418 | } |
7e936b73 AD |
1419 | if (unlikely(ino == dir->i_ino)) { |
1420 | EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir", | |
1421 | dentry->d_name.len, | |
1422 | dentry->d_name.name); | |
1423 | return ERR_PTR(-EIO); | |
1424 | } | |
1d1fe1ee | 1425 | inode = ext4_iget(dir->i_sb, ino); |
a9049376 AV |
1426 | if (inode == ERR_PTR(-ESTALE)) { |
1427 | EXT4_ERROR_INODE(dir, | |
1428 | "deleted inode referenced: %u", | |
1429 | ino); | |
1430 | return ERR_PTR(-EIO); | |
e6f009b0 | 1431 | } |
ac27a0ec DK |
1432 | } |
1433 | return d_splice_alias(inode, dentry); | |
1434 | } | |
1435 | ||
1436 | ||
617ba13b | 1437 | struct dentry *ext4_get_parent(struct dentry *child) |
ac27a0ec | 1438 | { |
498e5f24 | 1439 | __u32 ino; |
26fe5750 | 1440 | static const struct qstr dotdot = QSTR_INIT("..", 2); |
617ba13b | 1441 | struct ext4_dir_entry_2 * de; |
ac27a0ec DK |
1442 | struct buffer_head *bh; |
1443 | ||
32f7f22c | 1444 | bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL); |
ac27a0ec DK |
1445 | if (!bh) |
1446 | return ERR_PTR(-ENOENT); | |
1447 | ino = le32_to_cpu(de->inode); | |
1448 | brelse(bh); | |
1449 | ||
617ba13b | 1450 | if (!ext4_valid_inum(child->d_inode->i_sb, ino)) { |
24676da4 TT |
1451 | EXT4_ERROR_INODE(child->d_inode, |
1452 | "bad parent inode number: %u", ino); | |
1d1fe1ee | 1453 | return ERR_PTR(-EIO); |
a6c15c2b VA |
1454 | } |
1455 | ||
44003728 | 1456 | return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino)); |
ac27a0ec DK |
1457 | } |
1458 | ||
1459 | #define S_SHIFT 12 | |
617ba13b MC |
1460 | static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = { |
1461 | [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE, | |
1462 | [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR, | |
1463 | [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV, | |
1464 | [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV, | |
1465 | [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO, | |
1466 | [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK, | |
1467 | [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK, | |
ac27a0ec DK |
1468 | }; |
1469 | ||
617ba13b MC |
1470 | static inline void ext4_set_de_type(struct super_block *sb, |
1471 | struct ext4_dir_entry_2 *de, | |
ac27a0ec | 1472 | umode_t mode) { |
617ba13b MC |
1473 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE)) |
1474 | de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT]; | |
ac27a0ec DK |
1475 | } |
1476 | ||
ef2b02d3 ES |
1477 | /* |
1478 | * Move count entries from end of map between two memory locations. | |
1479 | * Returns pointer to last entry moved. | |
1480 | */ | |
617ba13b | 1481 | static struct ext4_dir_entry_2 * |
3d0518f4 WY |
1482 | dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count, |
1483 | unsigned blocksize) | |
ac27a0ec DK |
1484 | { |
1485 | unsigned rec_len = 0; | |
1486 | ||
1487 | while (count--) { | |
60e6679e | 1488 | struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) |
9aee2286 | 1489 | (from + (map->offs<<2)); |
617ba13b | 1490 | rec_len = EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec | 1491 | memcpy (to, de, rec_len); |
617ba13b | 1492 | ((struct ext4_dir_entry_2 *) to)->rec_len = |
3d0518f4 | 1493 | ext4_rec_len_to_disk(rec_len, blocksize); |
ac27a0ec DK |
1494 | de->inode = 0; |
1495 | map++; | |
1496 | to += rec_len; | |
1497 | } | |
617ba13b | 1498 | return (struct ext4_dir_entry_2 *) (to - rec_len); |
ac27a0ec DK |
1499 | } |
1500 | ||
ef2b02d3 ES |
1501 | /* |
1502 | * Compact each dir entry in the range to the minimal rec_len. | |
1503 | * Returns pointer to last entry in range. | |
1504 | */ | |
8bad4597 | 1505 | static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize) |
ac27a0ec | 1506 | { |
617ba13b | 1507 | struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base; |
ac27a0ec DK |
1508 | unsigned rec_len = 0; |
1509 | ||
1510 | prev = to = de; | |
8bad4597 | 1511 | while ((char*)de < base + blocksize) { |
3d0518f4 | 1512 | next = ext4_next_entry(de, blocksize); |
ac27a0ec | 1513 | if (de->inode && de->name_len) { |
617ba13b | 1514 | rec_len = EXT4_DIR_REC_LEN(de->name_len); |
ac27a0ec DK |
1515 | if (de > to) |
1516 | memmove(to, de, rec_len); | |
3d0518f4 | 1517 | to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize); |
ac27a0ec | 1518 | prev = to; |
617ba13b | 1519 | to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len); |
ac27a0ec DK |
1520 | } |
1521 | de = next; | |
1522 | } | |
1523 | return prev; | |
1524 | } | |
1525 | ||
ef2b02d3 ES |
1526 | /* |
1527 | * Split a full leaf block to make room for a new dir entry. | |
1528 | * Allocate a new block, and move entries so that they are approx. equally full. | |
1529 | * Returns pointer to de in block into which the new entry will be inserted. | |
1530 | */ | |
617ba13b | 1531 | static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir, |
ac27a0ec DK |
1532 | struct buffer_head **bh,struct dx_frame *frame, |
1533 | struct dx_hash_info *hinfo, int *error) | |
1534 | { | |
1535 | unsigned blocksize = dir->i_sb->s_blocksize; | |
1536 | unsigned count, continued; | |
1537 | struct buffer_head *bh2; | |
725d26d3 | 1538 | ext4_lblk_t newblock; |
ac27a0ec DK |
1539 | u32 hash2; |
1540 | struct dx_map_entry *map; | |
1541 | char *data1 = (*bh)->b_data, *data2; | |
59e315b4 | 1542 | unsigned split, move, size; |
617ba13b | 1543 | struct ext4_dir_entry_2 *de = NULL, *de2; |
b0336e8d DW |
1544 | struct ext4_dir_entry_tail *t; |
1545 | int csum_size = 0; | |
59e315b4 | 1546 | int err = 0, i; |
ac27a0ec | 1547 | |
b0336e8d DW |
1548 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, |
1549 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
1550 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
1551 | ||
0f70b406 TT |
1552 | bh2 = ext4_append(handle, dir, &newblock); |
1553 | if (IS_ERR(bh2)) { | |
ac27a0ec DK |
1554 | brelse(*bh); |
1555 | *bh = NULL; | |
0f70b406 TT |
1556 | *error = PTR_ERR(bh2); |
1557 | return NULL; | |
ac27a0ec DK |
1558 | } |
1559 | ||
1560 | BUFFER_TRACE(*bh, "get_write_access"); | |
617ba13b | 1561 | err = ext4_journal_get_write_access(handle, *bh); |
fedee54d DM |
1562 | if (err) |
1563 | goto journal_error; | |
1564 | ||
ac27a0ec | 1565 | BUFFER_TRACE(frame->bh, "get_write_access"); |
617ba13b | 1566 | err = ext4_journal_get_write_access(handle, frame->bh); |
ac27a0ec DK |
1567 | if (err) |
1568 | goto journal_error; | |
1569 | ||
1570 | data2 = bh2->b_data; | |
1571 | ||
1572 | /* create map in the end of data2 block */ | |
1573 | map = (struct dx_map_entry *) (data2 + blocksize); | |
af5bc92d | 1574 | count = dx_make_map((struct ext4_dir_entry_2 *) data1, |
ac27a0ec DK |
1575 | blocksize, hinfo, map); |
1576 | map -= count; | |
af5bc92d | 1577 | dx_sort_map(map, count); |
ef2b02d3 ES |
1578 | /* Split the existing block in the middle, size-wise */ |
1579 | size = 0; | |
1580 | move = 0; | |
1581 | for (i = count-1; i >= 0; i--) { | |
1582 | /* is more than half of this entry in 2nd half of the block? */ | |
1583 | if (size + map[i].size/2 > blocksize/2) | |
1584 | break; | |
1585 | size += map[i].size; | |
1586 | move++; | |
1587 | } | |
1588 | /* map index at which we will split */ | |
1589 | split = count - move; | |
ac27a0ec DK |
1590 | hash2 = map[split].hash; |
1591 | continued = hash2 == map[split - 1].hash; | |
725d26d3 AK |
1592 | dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n", |
1593 | (unsigned long)dx_get_block(frame->at), | |
1594 | hash2, split, count-split)); | |
ac27a0ec DK |
1595 | |
1596 | /* Fancy dance to stay within two buffers */ | |
3d0518f4 | 1597 | de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize); |
af5bc92d | 1598 | de = dx_pack_dirents(data1, blocksize); |
b0336e8d DW |
1599 | de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - |
1600 | (char *) de, | |
3d0518f4 | 1601 | blocksize); |
b0336e8d DW |
1602 | de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) - |
1603 | (char *) de2, | |
3d0518f4 | 1604 | blocksize); |
b0336e8d DW |
1605 | if (csum_size) { |
1606 | t = EXT4_DIRENT_TAIL(data2, blocksize); | |
1607 | initialize_dirent_tail(t, blocksize); | |
1608 | ||
1609 | t = EXT4_DIRENT_TAIL(data1, blocksize); | |
1610 | initialize_dirent_tail(t, blocksize); | |
1611 | } | |
1612 | ||
617ba13b MC |
1613 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1)); |
1614 | dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1)); | |
ac27a0ec DK |
1615 | |
1616 | /* Which block gets the new entry? */ | |
1617 | if (hinfo->hash >= hash2) | |
1618 | { | |
1619 | swap(*bh, bh2); | |
1620 | de = de2; | |
1621 | } | |
af5bc92d | 1622 | dx_insert_block(frame, hash2 + continued, newblock); |
b0336e8d | 1623 | err = ext4_handle_dirty_dirent_node(handle, dir, bh2); |
ac27a0ec DK |
1624 | if (err) |
1625 | goto journal_error; | |
dbe89444 | 1626 | err = ext4_handle_dirty_dx_node(handle, dir, frame->bh); |
ac27a0ec DK |
1627 | if (err) |
1628 | goto journal_error; | |
af5bc92d TT |
1629 | brelse(bh2); |
1630 | dxtrace(dx_show_index("frame", frame->entries)); | |
ac27a0ec | 1631 | return de; |
fedee54d DM |
1632 | |
1633 | journal_error: | |
1634 | brelse(*bh); | |
1635 | brelse(bh2); | |
1636 | *bh = NULL; | |
1637 | ext4_std_error(dir->i_sb, err); | |
fedee54d DM |
1638 | *error = err; |
1639 | return NULL; | |
ac27a0ec | 1640 | } |
ac27a0ec | 1641 | |
978fef91 TM |
1642 | int ext4_find_dest_de(struct inode *dir, struct inode *inode, |
1643 | struct buffer_head *bh, | |
1644 | void *buf, int buf_size, | |
1645 | const char *name, int namelen, | |
1646 | struct ext4_dir_entry_2 **dest_de) | |
1647 | { | |
1648 | struct ext4_dir_entry_2 *de; | |
1649 | unsigned short reclen = EXT4_DIR_REC_LEN(namelen); | |
1650 | int nlen, rlen; | |
1651 | unsigned int offset = 0; | |
1652 | char *top; | |
1653 | ||
1654 | de = (struct ext4_dir_entry_2 *)buf; | |
1655 | top = buf + buf_size - reclen; | |
1656 | while ((char *) de <= top) { | |
1657 | if (ext4_check_dir_entry(dir, NULL, de, bh, | |
1658 | buf, buf_size, offset)) | |
1659 | return -EIO; | |
1660 | if (ext4_match(namelen, name, de)) | |
1661 | return -EEXIST; | |
1662 | nlen = EXT4_DIR_REC_LEN(de->name_len); | |
1663 | rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); | |
1664 | if ((de->inode ? rlen - nlen : rlen) >= reclen) | |
1665 | break; | |
1666 | de = (struct ext4_dir_entry_2 *)((char *)de + rlen); | |
1667 | offset += rlen; | |
1668 | } | |
1669 | if ((char *) de > top) | |
1670 | return -ENOSPC; | |
1671 | ||
1672 | *dest_de = de; | |
1673 | return 0; | |
1674 | } | |
1675 | ||
1676 | void ext4_insert_dentry(struct inode *inode, | |
1677 | struct ext4_dir_entry_2 *de, | |
1678 | int buf_size, | |
1679 | const char *name, int namelen) | |
1680 | { | |
1681 | ||
1682 | int nlen, rlen; | |
1683 | ||
1684 | nlen = EXT4_DIR_REC_LEN(de->name_len); | |
1685 | rlen = ext4_rec_len_from_disk(de->rec_len, buf_size); | |
1686 | if (de->inode) { | |
1687 | struct ext4_dir_entry_2 *de1 = | |
1688 | (struct ext4_dir_entry_2 *)((char *)de + nlen); | |
1689 | de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size); | |
1690 | de->rec_len = ext4_rec_len_to_disk(nlen, buf_size); | |
1691 | de = de1; | |
1692 | } | |
1693 | de->file_type = EXT4_FT_UNKNOWN; | |
1694 | de->inode = cpu_to_le32(inode->i_ino); | |
1695 | ext4_set_de_type(inode->i_sb, de, inode->i_mode); | |
1696 | de->name_len = namelen; | |
1697 | memcpy(de->name, name, namelen); | |
1698 | } | |
ac27a0ec DK |
1699 | /* |
1700 | * Add a new entry into a directory (leaf) block. If de is non-NULL, | |
1701 | * it points to a directory entry which is guaranteed to be large | |
1702 | * enough for new directory entry. If de is NULL, then | |
1703 | * add_dirent_to_buf will attempt search the directory block for | |
1704 | * space. It will return -ENOSPC if no space is available, and -EIO | |
1705 | * and -EEXIST if directory entry already exists. | |
ac27a0ec DK |
1706 | */ |
1707 | static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry, | |
617ba13b | 1708 | struct inode *inode, struct ext4_dir_entry_2 *de, |
af5bc92d | 1709 | struct buffer_head *bh) |
ac27a0ec DK |
1710 | { |
1711 | struct inode *dir = dentry->d_parent->d_inode; | |
1712 | const char *name = dentry->d_name.name; | |
1713 | int namelen = dentry->d_name.len; | |
3d0518f4 | 1714 | unsigned int blocksize = dir->i_sb->s_blocksize; |
b0336e8d | 1715 | int csum_size = 0; |
978fef91 | 1716 | int err; |
b0336e8d DW |
1717 | |
1718 | if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
1719 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
1720 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
ac27a0ec | 1721 | |
ac27a0ec | 1722 | if (!de) { |
978fef91 TM |
1723 | err = ext4_find_dest_de(dir, inode, |
1724 | bh, bh->b_data, blocksize - csum_size, | |
1725 | name, namelen, &de); | |
1726 | if (err) | |
1727 | return err; | |
ac27a0ec DK |
1728 | } |
1729 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1730 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1731 | if (err) { |
617ba13b | 1732 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1733 | return err; |
1734 | } | |
1735 | ||
1736 | /* By now the buffer is marked for journaling */ | |
978fef91 TM |
1737 | ext4_insert_dentry(inode, de, blocksize, name, namelen); |
1738 | ||
ac27a0ec DK |
1739 | /* |
1740 | * XXX shouldn't update any times until successful | |
1741 | * completion of syscall, but too many callers depend | |
1742 | * on this. | |
1743 | * | |
1744 | * XXX similarly, too many callers depend on | |
617ba13b | 1745 | * ext4_new_inode() setting the times, but error |
ac27a0ec DK |
1746 | * recovery deletes the inode, so the worst that can |
1747 | * happen is that the times are slightly out of date | |
1748 | * and/or different from the directory change time. | |
1749 | */ | |
ef7f3835 | 1750 | dir->i_mtime = dir->i_ctime = ext4_current_time(dir); |
617ba13b | 1751 | ext4_update_dx_flag(dir); |
ac27a0ec | 1752 | dir->i_version++; |
617ba13b | 1753 | ext4_mark_inode_dirty(handle, dir); |
0390131b | 1754 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
b0336e8d | 1755 | err = ext4_handle_dirty_dirent_node(handle, dir, bh); |
ac27a0ec | 1756 | if (err) |
617ba13b | 1757 | ext4_std_error(dir->i_sb, err); |
ac27a0ec DK |
1758 | return 0; |
1759 | } | |
1760 | ||
ac27a0ec DK |
1761 | /* |
1762 | * This converts a one block unindexed directory to a 3 block indexed | |
1763 | * directory, and adds the dentry to the indexed directory. | |
1764 | */ | |
1765 | static int make_indexed_dir(handle_t *handle, struct dentry *dentry, | |
1766 | struct inode *inode, struct buffer_head *bh) | |
1767 | { | |
1768 | struct inode *dir = dentry->d_parent->d_inode; | |
1769 | const char *name = dentry->d_name.name; | |
1770 | int namelen = dentry->d_name.len; | |
1771 | struct buffer_head *bh2; | |
1772 | struct dx_root *root; | |
1773 | struct dx_frame frames[2], *frame; | |
1774 | struct dx_entry *entries; | |
617ba13b | 1775 | struct ext4_dir_entry_2 *de, *de2; |
b0336e8d | 1776 | struct ext4_dir_entry_tail *t; |
ac27a0ec DK |
1777 | char *data1, *top; |
1778 | unsigned len; | |
1779 | int retval; | |
1780 | unsigned blocksize; | |
1781 | struct dx_hash_info hinfo; | |
725d26d3 | 1782 | ext4_lblk_t block; |
ac27a0ec | 1783 | struct fake_dirent *fde; |
b0336e8d DW |
1784 | int csum_size = 0; |
1785 | ||
1786 | if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
1787 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
1788 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
ac27a0ec DK |
1789 | |
1790 | blocksize = dir->i_sb->s_blocksize; | |
e6b8bc09 | 1791 | dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino)); |
617ba13b | 1792 | retval = ext4_journal_get_write_access(handle, bh); |
ac27a0ec | 1793 | if (retval) { |
617ba13b | 1794 | ext4_std_error(dir->i_sb, retval); |
ac27a0ec DK |
1795 | brelse(bh); |
1796 | return retval; | |
1797 | } | |
1798 | root = (struct dx_root *) bh->b_data; | |
1799 | ||
e6b8bc09 TT |
1800 | /* The 0th block becomes the root, move the dirents out */ |
1801 | fde = &root->dotdot; | |
1802 | de = (struct ext4_dir_entry_2 *)((char *)fde + | |
3d0518f4 | 1803 | ext4_rec_len_from_disk(fde->rec_len, blocksize)); |
e6b8bc09 | 1804 | if ((char *) de >= (((char *) root) + blocksize)) { |
24676da4 | 1805 | EXT4_ERROR_INODE(dir, "invalid rec_len for '..'"); |
e6b8bc09 TT |
1806 | brelse(bh); |
1807 | return -EIO; | |
1808 | } | |
b0336e8d | 1809 | len = ((char *) root) + (blocksize - csum_size) - (char *) de; |
e6b8bc09 TT |
1810 | |
1811 | /* Allocate new block for the 0th block's dirents */ | |
0f70b406 TT |
1812 | bh2 = ext4_append(handle, dir, &block); |
1813 | if (IS_ERR(bh2)) { | |
ac27a0ec | 1814 | brelse(bh); |
0f70b406 | 1815 | return PTR_ERR(bh2); |
ac27a0ec | 1816 | } |
12e9b892 | 1817 | ext4_set_inode_flag(dir, EXT4_INODE_INDEX); |
ac27a0ec DK |
1818 | data1 = bh2->b_data; |
1819 | ||
ac27a0ec | 1820 | memcpy (data1, de, len); |
617ba13b | 1821 | de = (struct ext4_dir_entry_2 *) data1; |
ac27a0ec | 1822 | top = data1 + len; |
3d0518f4 | 1823 | while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) |
ac27a0ec | 1824 | de = de2; |
b0336e8d DW |
1825 | de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) - |
1826 | (char *) de, | |
3d0518f4 | 1827 | blocksize); |
b0336e8d DW |
1828 | |
1829 | if (csum_size) { | |
1830 | t = EXT4_DIRENT_TAIL(data1, blocksize); | |
1831 | initialize_dirent_tail(t, blocksize); | |
1832 | } | |
1833 | ||
ac27a0ec | 1834 | /* Initialize the root; the dot dirents already exist */ |
617ba13b | 1835 | de = (struct ext4_dir_entry_2 *) (&root->dotdot); |
3d0518f4 WY |
1836 | de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2), |
1837 | blocksize); | |
ac27a0ec DK |
1838 | memset (&root->info, 0, sizeof(root->info)); |
1839 | root->info.info_length = sizeof(root->info); | |
617ba13b | 1840 | root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version; |
ac27a0ec | 1841 | entries = root->entries; |
af5bc92d TT |
1842 | dx_set_block(entries, 1); |
1843 | dx_set_count(entries, 1); | |
1844 | dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info))); | |
ac27a0ec DK |
1845 | |
1846 | /* Initialize as for dx_probe */ | |
1847 | hinfo.hash_version = root->info.hash_version; | |
f99b2589 TT |
1848 | if (hinfo.hash_version <= DX_HASH_TEA) |
1849 | hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned; | |
617ba13b MC |
1850 | hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed; |
1851 | ext4fs_dirhash(name, namelen, &hinfo); | |
ac27a0ec DK |
1852 | frame = frames; |
1853 | frame->entries = entries; | |
1854 | frame->at = entries; | |
1855 | frame->bh = bh; | |
1856 | bh = bh2; | |
6976a6f2 | 1857 | |
dbe89444 | 1858 | ext4_handle_dirty_dx_node(handle, dir, frame->bh); |
b0336e8d | 1859 | ext4_handle_dirty_dirent_node(handle, dir, bh); |
6976a6f2 | 1860 | |
ac27a0ec | 1861 | de = do_split(handle,dir, &bh, frame, &hinfo, &retval); |
7ad8e4e6 JK |
1862 | if (!de) { |
1863 | /* | |
1864 | * Even if the block split failed, we have to properly write | |
1865 | * out all the changes we did so far. Otherwise we can end up | |
1866 | * with corrupted filesystem. | |
1867 | */ | |
1868 | ext4_mark_inode_dirty(handle, dir); | |
7ad8e4e6 | 1869 | dx_release(frames); |
ac27a0ec | 1870 | return retval; |
7ad8e4e6 JK |
1871 | } |
1872 | dx_release(frames); | |
ac27a0ec | 1873 | |
2de770a4 TT |
1874 | retval = add_dirent_to_buf(handle, dentry, inode, de, bh); |
1875 | brelse(bh); | |
1876 | return retval; | |
ac27a0ec | 1877 | } |
ac27a0ec DK |
1878 | |
1879 | /* | |
617ba13b | 1880 | * ext4_add_entry() |
ac27a0ec DK |
1881 | * |
1882 | * adds a file entry to the specified directory, using the same | |
617ba13b | 1883 | * semantics as ext4_find_entry(). It returns NULL if it failed. |
ac27a0ec DK |
1884 | * |
1885 | * NOTE!! The inode part of 'de' is left at 0 - which means you | |
1886 | * may not sleep between calling this and putting something into | |
1887 | * the entry, as someone else might have used it while you slept. | |
1888 | */ | |
af5bc92d TT |
1889 | static int ext4_add_entry(handle_t *handle, struct dentry *dentry, |
1890 | struct inode *inode) | |
ac27a0ec DK |
1891 | { |
1892 | struct inode *dir = dentry->d_parent->d_inode; | |
af5bc92d | 1893 | struct buffer_head *bh; |
617ba13b | 1894 | struct ext4_dir_entry_2 *de; |
b0336e8d | 1895 | struct ext4_dir_entry_tail *t; |
af5bc92d | 1896 | struct super_block *sb; |
ac27a0ec | 1897 | int retval; |
ac27a0ec | 1898 | int dx_fallback=0; |
ac27a0ec | 1899 | unsigned blocksize; |
725d26d3 | 1900 | ext4_lblk_t block, blocks; |
b0336e8d DW |
1901 | int csum_size = 0; |
1902 | ||
1903 | if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
1904 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
1905 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
ac27a0ec DK |
1906 | |
1907 | sb = dir->i_sb; | |
1908 | blocksize = sb->s_blocksize; | |
1909 | if (!dentry->d_name.len) | |
1910 | return -EINVAL; | |
3c47d541 TM |
1911 | |
1912 | if (ext4_has_inline_data(dir)) { | |
1913 | retval = ext4_try_add_inline_entry(handle, dentry, inode); | |
1914 | if (retval < 0) | |
1915 | return retval; | |
1916 | if (retval == 1) { | |
1917 | retval = 0; | |
1918 | return retval; | |
1919 | } | |
1920 | } | |
1921 | ||
ac27a0ec | 1922 | if (is_dx(dir)) { |
617ba13b | 1923 | retval = ext4_dx_add_entry(handle, dentry, inode); |
ac27a0ec DK |
1924 | if (!retval || (retval != ERR_BAD_DX_DIR)) |
1925 | return retval; | |
12e9b892 | 1926 | ext4_clear_inode_flag(dir, EXT4_INODE_INDEX); |
ac27a0ec | 1927 | dx_fallback++; |
617ba13b | 1928 | ext4_mark_inode_dirty(handle, dir); |
ac27a0ec | 1929 | } |
ac27a0ec | 1930 | blocks = dir->i_size >> sb->s_blocksize_bits; |
498e5f24 | 1931 | for (block = 0; block < blocks; block++) { |
dc6982ff TT |
1932 | bh = ext4_read_dirblock(dir, block, DIRENT); |
1933 | if (IS_ERR(bh)) | |
1934 | return PTR_ERR(bh); | |
1935 | ||
ac27a0ec | 1936 | retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh); |
2de770a4 TT |
1937 | if (retval != -ENOSPC) { |
1938 | brelse(bh); | |
ac27a0ec | 1939 | return retval; |
2de770a4 | 1940 | } |
ac27a0ec | 1941 | |
ac27a0ec | 1942 | if (blocks == 1 && !dx_fallback && |
1e424a34 TT |
1943 | EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) |
1944 | return make_indexed_dir(handle, dentry, inode, bh); | |
ac27a0ec DK |
1945 | brelse(bh); |
1946 | } | |
0f70b406 TT |
1947 | bh = ext4_append(handle, dir, &block); |
1948 | if (IS_ERR(bh)) | |
1949 | return PTR_ERR(bh); | |
617ba13b | 1950 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 1951 | de->inode = 0; |
b0336e8d DW |
1952 | de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize); |
1953 | ||
1954 | if (csum_size) { | |
1955 | t = EXT4_DIRENT_TAIL(bh->b_data, blocksize); | |
1956 | initialize_dirent_tail(t, blocksize); | |
1957 | } | |
1958 | ||
2de770a4 TT |
1959 | retval = add_dirent_to_buf(handle, dentry, inode, de, bh); |
1960 | brelse(bh); | |
14ece102 FM |
1961 | if (retval == 0) |
1962 | ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY); | |
2de770a4 | 1963 | return retval; |
ac27a0ec DK |
1964 | } |
1965 | ||
ac27a0ec DK |
1966 | /* |
1967 | * Returns 0 for success, or a negative error value | |
1968 | */ | |
617ba13b | 1969 | static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry, |
ac27a0ec DK |
1970 | struct inode *inode) |
1971 | { | |
1972 | struct dx_frame frames[2], *frame; | |
1973 | struct dx_entry *entries, *at; | |
1974 | struct dx_hash_info hinfo; | |
af5bc92d | 1975 | struct buffer_head *bh; |
ac27a0ec | 1976 | struct inode *dir = dentry->d_parent->d_inode; |
af5bc92d | 1977 | struct super_block *sb = dir->i_sb; |
617ba13b | 1978 | struct ext4_dir_entry_2 *de; |
ac27a0ec DK |
1979 | int err; |
1980 | ||
f702ba0f | 1981 | frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err); |
ac27a0ec DK |
1982 | if (!frame) |
1983 | return err; | |
1984 | entries = frame->entries; | |
1985 | at = frame->at; | |
dc6982ff TT |
1986 | bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT); |
1987 | if (IS_ERR(bh)) { | |
1988 | err = PTR_ERR(bh); | |
1989 | bh = NULL; | |
ac27a0ec | 1990 | goto cleanup; |
6d1ab10e | 1991 | } |
ac27a0ec DK |
1992 | |
1993 | BUFFER_TRACE(bh, "get_write_access"); | |
617ba13b | 1994 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1995 | if (err) |
1996 | goto journal_error; | |
1997 | ||
1998 | err = add_dirent_to_buf(handle, dentry, inode, NULL, bh); | |
2de770a4 | 1999 | if (err != -ENOSPC) |
ac27a0ec | 2000 | goto cleanup; |
ac27a0ec DK |
2001 | |
2002 | /* Block full, should compress but for now just split */ | |
4776004f | 2003 | dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n", |
ac27a0ec DK |
2004 | dx_get_count(entries), dx_get_limit(entries))); |
2005 | /* Need to split index? */ | |
2006 | if (dx_get_count(entries) == dx_get_limit(entries)) { | |
725d26d3 | 2007 | ext4_lblk_t newblock; |
ac27a0ec DK |
2008 | unsigned icount = dx_get_count(entries); |
2009 | int levels = frame - frames; | |
2010 | struct dx_entry *entries2; | |
2011 | struct dx_node *node2; | |
2012 | struct buffer_head *bh2; | |
2013 | ||
2014 | if (levels && (dx_get_count(frames->entries) == | |
2015 | dx_get_limit(frames->entries))) { | |
12062ddd | 2016 | ext4_warning(sb, "Directory index full!"); |
ac27a0ec DK |
2017 | err = -ENOSPC; |
2018 | goto cleanup; | |
2019 | } | |
0f70b406 TT |
2020 | bh2 = ext4_append(handle, dir, &newblock); |
2021 | if (IS_ERR(bh2)) { | |
2022 | err = PTR_ERR(bh2); | |
ac27a0ec | 2023 | goto cleanup; |
0f70b406 | 2024 | } |
ac27a0ec DK |
2025 | node2 = (struct dx_node *)(bh2->b_data); |
2026 | entries2 = node2->entries; | |
1f7bebb9 | 2027 | memset(&node2->fake, 0, sizeof(struct fake_dirent)); |
3d0518f4 WY |
2028 | node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize, |
2029 | sb->s_blocksize); | |
ac27a0ec | 2030 | BUFFER_TRACE(frame->bh, "get_write_access"); |
617ba13b | 2031 | err = ext4_journal_get_write_access(handle, frame->bh); |
ac27a0ec DK |
2032 | if (err) |
2033 | goto journal_error; | |
2034 | if (levels) { | |
2035 | unsigned icount1 = icount/2, icount2 = icount - icount1; | |
2036 | unsigned hash2 = dx_get_hash(entries + icount1); | |
4776004f TT |
2037 | dxtrace(printk(KERN_DEBUG "Split index %i/%i\n", |
2038 | icount1, icount2)); | |
ac27a0ec DK |
2039 | |
2040 | BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */ | |
617ba13b | 2041 | err = ext4_journal_get_write_access(handle, |
ac27a0ec DK |
2042 | frames[0].bh); |
2043 | if (err) | |
2044 | goto journal_error; | |
2045 | ||
af5bc92d TT |
2046 | memcpy((char *) entries2, (char *) (entries + icount1), |
2047 | icount2 * sizeof(struct dx_entry)); | |
2048 | dx_set_count(entries, icount1); | |
2049 | dx_set_count(entries2, icount2); | |
2050 | dx_set_limit(entries2, dx_node_limit(dir)); | |
ac27a0ec DK |
2051 | |
2052 | /* Which index block gets the new entry? */ | |
2053 | if (at - entries >= icount1) { | |
2054 | frame->at = at = at - entries - icount1 + entries2; | |
2055 | frame->entries = entries = entries2; | |
2056 | swap(frame->bh, bh2); | |
2057 | } | |
af5bc92d TT |
2058 | dx_insert_block(frames + 0, hash2, newblock); |
2059 | dxtrace(dx_show_index("node", frames[1].entries)); | |
2060 | dxtrace(dx_show_index("node", | |
ac27a0ec | 2061 | ((struct dx_node *) bh2->b_data)->entries)); |
dbe89444 | 2062 | err = ext4_handle_dirty_dx_node(handle, dir, bh2); |
ac27a0ec DK |
2063 | if (err) |
2064 | goto journal_error; | |
2065 | brelse (bh2); | |
2066 | } else { | |
4776004f TT |
2067 | dxtrace(printk(KERN_DEBUG |
2068 | "Creating second level index...\n")); | |
ac27a0ec DK |
2069 | memcpy((char *) entries2, (char *) entries, |
2070 | icount * sizeof(struct dx_entry)); | |
2071 | dx_set_limit(entries2, dx_node_limit(dir)); | |
2072 | ||
2073 | /* Set up root */ | |
2074 | dx_set_count(entries, 1); | |
2075 | dx_set_block(entries + 0, newblock); | |
2076 | ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1; | |
2077 | ||
2078 | /* Add new access path frame */ | |
2079 | frame = frames + 1; | |
2080 | frame->at = at = at - entries + entries2; | |
2081 | frame->entries = entries = entries2; | |
2082 | frame->bh = bh2; | |
617ba13b | 2083 | err = ext4_journal_get_write_access(handle, |
ac27a0ec DK |
2084 | frame->bh); |
2085 | if (err) | |
2086 | goto journal_error; | |
2087 | } | |
dbe89444 | 2088 | err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh); |
b4097142 TT |
2089 | if (err) { |
2090 | ext4_std_error(inode->i_sb, err); | |
2091 | goto cleanup; | |
2092 | } | |
ac27a0ec DK |
2093 | } |
2094 | de = do_split(handle, dir, &bh, frame, &hinfo, &err); | |
2095 | if (!de) | |
2096 | goto cleanup; | |
2097 | err = add_dirent_to_buf(handle, dentry, inode, de, bh); | |
ac27a0ec DK |
2098 | goto cleanup; |
2099 | ||
2100 | journal_error: | |
617ba13b | 2101 | ext4_std_error(dir->i_sb, err); |
ac27a0ec | 2102 | cleanup: |
b1deefc9 | 2103 | brelse(bh); |
ac27a0ec DK |
2104 | dx_release(frames); |
2105 | return err; | |
2106 | } | |
ac27a0ec DK |
2107 | |
2108 | /* | |
05019a9e TM |
2109 | * ext4_generic_delete_entry deletes a directory entry by merging it |
2110 | * with the previous entry | |
ac27a0ec | 2111 | */ |
05019a9e TM |
2112 | int ext4_generic_delete_entry(handle_t *handle, |
2113 | struct inode *dir, | |
2114 | struct ext4_dir_entry_2 *de_del, | |
2115 | struct buffer_head *bh, | |
2116 | void *entry_buf, | |
2117 | int buf_size, | |
2118 | int csum_size) | |
ac27a0ec | 2119 | { |
af5bc92d | 2120 | struct ext4_dir_entry_2 *de, *pde; |
3d0518f4 | 2121 | unsigned int blocksize = dir->i_sb->s_blocksize; |
05019a9e | 2122 | int i; |
b0336e8d | 2123 | |
ac27a0ec DK |
2124 | i = 0; |
2125 | pde = NULL; | |
05019a9e TM |
2126 | de = (struct ext4_dir_entry_2 *)entry_buf; |
2127 | while (i < buf_size - csum_size) { | |
226ba972 TM |
2128 | if (ext4_check_dir_entry(dir, NULL, de, bh, |
2129 | bh->b_data, bh->b_size, i)) | |
ac27a0ec DK |
2130 | return -EIO; |
2131 | if (de == de_del) { | |
ac27a0ec | 2132 | if (pde) |
a72d7f83 | 2133 | pde->rec_len = ext4_rec_len_to_disk( |
3d0518f4 WY |
2134 | ext4_rec_len_from_disk(pde->rec_len, |
2135 | blocksize) + | |
2136 | ext4_rec_len_from_disk(de->rec_len, | |
2137 | blocksize), | |
2138 | blocksize); | |
ac27a0ec DK |
2139 | else |
2140 | de->inode = 0; | |
2141 | dir->i_version++; | |
ac27a0ec DK |
2142 | return 0; |
2143 | } | |
3d0518f4 | 2144 | i += ext4_rec_len_from_disk(de->rec_len, blocksize); |
ac27a0ec | 2145 | pde = de; |
3d0518f4 | 2146 | de = ext4_next_entry(de, blocksize); |
ac27a0ec DK |
2147 | } |
2148 | return -ENOENT; | |
2149 | } | |
2150 | ||
05019a9e TM |
2151 | static int ext4_delete_entry(handle_t *handle, |
2152 | struct inode *dir, | |
2153 | struct ext4_dir_entry_2 *de_del, | |
2154 | struct buffer_head *bh) | |
2155 | { | |
2156 | int err, csum_size = 0; | |
2157 | ||
9f40fe54 TM |
2158 | if (ext4_has_inline_data(dir)) { |
2159 | int has_inline_data = 1; | |
2160 | err = ext4_delete_inline_entry(handle, dir, de_del, bh, | |
2161 | &has_inline_data); | |
2162 | if (has_inline_data) | |
2163 | return err; | |
2164 | } | |
2165 | ||
05019a9e TM |
2166 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, |
2167 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
2168 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
2169 | ||
2170 | BUFFER_TRACE(bh, "get_write_access"); | |
2171 | err = ext4_journal_get_write_access(handle, bh); | |
2172 | if (unlikely(err)) | |
2173 | goto out; | |
2174 | ||
2175 | err = ext4_generic_delete_entry(handle, dir, de_del, | |
2176 | bh, bh->b_data, | |
2177 | dir->i_sb->s_blocksize, csum_size); | |
2178 | if (err) | |
2179 | goto out; | |
2180 | ||
2181 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
2182 | err = ext4_handle_dirty_dirent_node(handle, dir, bh); | |
2183 | if (unlikely(err)) | |
2184 | goto out; | |
2185 | ||
2186 | return 0; | |
2187 | out: | |
2188 | if (err != -ENOENT) | |
2189 | ext4_std_error(dir->i_sb, err); | |
2190 | return err; | |
2191 | } | |
2192 | ||
f8628a14 AD |
2193 | /* |
2194 | * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2, | |
2195 | * since this indicates that nlinks count was previously 1. | |
2196 | */ | |
2197 | static void ext4_inc_count(handle_t *handle, struct inode *inode) | |
2198 | { | |
2199 | inc_nlink(inode); | |
2200 | if (is_dx(inode) && inode->i_nlink > 1) { | |
2201 | /* limit is 16-bit i_links_count */ | |
2202 | if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) { | |
bfe86848 | 2203 | set_nlink(inode, 1); |
f8628a14 AD |
2204 | EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb, |
2205 | EXT4_FEATURE_RO_COMPAT_DIR_NLINK); | |
2206 | } | |
2207 | } | |
2208 | } | |
2209 | ||
2210 | /* | |
2211 | * If a directory had nlink == 1, then we should let it be 1. This indicates | |
2212 | * directory has >EXT4_LINK_MAX subdirs. | |
2213 | */ | |
2214 | static void ext4_dec_count(handle_t *handle, struct inode *inode) | |
2215 | { | |
909a4cf1 AD |
2216 | if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2) |
2217 | drop_nlink(inode); | |
f8628a14 AD |
2218 | } |
2219 | ||
2220 | ||
617ba13b | 2221 | static int ext4_add_nondir(handle_t *handle, |
ac27a0ec DK |
2222 | struct dentry *dentry, struct inode *inode) |
2223 | { | |
617ba13b | 2224 | int err = ext4_add_entry(handle, dentry, inode); |
ac27a0ec | 2225 | if (!err) { |
617ba13b | 2226 | ext4_mark_inode_dirty(handle, inode); |
6b38e842 | 2227 | unlock_new_inode(inode); |
8fc37ec5 | 2228 | d_instantiate(dentry, inode); |
ac27a0ec DK |
2229 | return 0; |
2230 | } | |
731b9a54 | 2231 | drop_nlink(inode); |
6b38e842 | 2232 | unlock_new_inode(inode); |
ac27a0ec DK |
2233 | iput(inode); |
2234 | return err; | |
2235 | } | |
2236 | ||
2237 | /* | |
2238 | * By the time this is called, we already have created | |
2239 | * the directory cache entry for the new file, but it | |
2240 | * is so far negative - it has no inode. | |
2241 | * | |
2242 | * If the create succeeds, we fill in the inode information | |
2243 | * with d_instantiate(). | |
2244 | */ | |
4acdaf27 | 2245 | static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2246 | bool excl) |
ac27a0ec DK |
2247 | { |
2248 | handle_t *handle; | |
af5bc92d | 2249 | struct inode *inode; |
1139575a | 2250 | int err, credits, retries = 0; |
ac27a0ec | 2251 | |
871a2931 | 2252 | dquot_initialize(dir); |
907f4554 | 2253 | |
1139575a TT |
2254 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2255 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
2256 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); | |
ac27a0ec | 2257 | retry: |
1139575a TT |
2258 | inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, |
2259 | NULL, EXT4_HT_DIR, credits); | |
2260 | handle = ext4_journal_current_handle(); | |
ac27a0ec DK |
2261 | err = PTR_ERR(inode); |
2262 | if (!IS_ERR(inode)) { | |
617ba13b MC |
2263 | inode->i_op = &ext4_file_inode_operations; |
2264 | inode->i_fop = &ext4_file_operations; | |
2265 | ext4_set_aops(inode); | |
2266 | err = ext4_add_nondir(handle, dentry, inode); | |
1139575a TT |
2267 | if (!err && IS_DIRSYNC(dir)) |
2268 | ext4_handle_sync(handle); | |
ac27a0ec | 2269 | } |
1139575a TT |
2270 | if (handle) |
2271 | ext4_journal_stop(handle); | |
617ba13b | 2272 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) |
ac27a0ec DK |
2273 | goto retry; |
2274 | return err; | |
2275 | } | |
2276 | ||
af5bc92d | 2277 | static int ext4_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 2278 | umode_t mode, dev_t rdev) |
ac27a0ec DK |
2279 | { |
2280 | handle_t *handle; | |
2281 | struct inode *inode; | |
1139575a | 2282 | int err, credits, retries = 0; |
ac27a0ec DK |
2283 | |
2284 | if (!new_valid_dev(rdev)) | |
2285 | return -EINVAL; | |
2286 | ||
871a2931 | 2287 | dquot_initialize(dir); |
907f4554 | 2288 | |
1139575a TT |
2289 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2290 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
2291 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); | |
ac27a0ec | 2292 | retry: |
1139575a TT |
2293 | inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0, |
2294 | NULL, EXT4_HT_DIR, credits); | |
2295 | handle = ext4_journal_current_handle(); | |
ac27a0ec DK |
2296 | err = PTR_ERR(inode); |
2297 | if (!IS_ERR(inode)) { | |
2298 | init_special_inode(inode, inode->i_mode, rdev); | |
617ba13b | 2299 | inode->i_op = &ext4_special_inode_operations; |
617ba13b | 2300 | err = ext4_add_nondir(handle, dentry, inode); |
1139575a TT |
2301 | if (!err && IS_DIRSYNC(dir)) |
2302 | ext4_handle_sync(handle); | |
ac27a0ec | 2303 | } |
1139575a TT |
2304 | if (handle) |
2305 | ext4_journal_stop(handle); | |
617ba13b | 2306 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) |
ac27a0ec DK |
2307 | goto retry; |
2308 | return err; | |
2309 | } | |
2310 | ||
a774f9c2 TM |
2311 | struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode, |
2312 | struct ext4_dir_entry_2 *de, | |
2313 | int blocksize, int csum_size, | |
2314 | unsigned int parent_ino, int dotdot_real_len) | |
2315 | { | |
2316 | de->inode = cpu_to_le32(inode->i_ino); | |
2317 | de->name_len = 1; | |
2318 | de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len), | |
2319 | blocksize); | |
2320 | strcpy(de->name, "."); | |
2321 | ext4_set_de_type(inode->i_sb, de, S_IFDIR); | |
2322 | ||
2323 | de = ext4_next_entry(de, blocksize); | |
2324 | de->inode = cpu_to_le32(parent_ino); | |
2325 | de->name_len = 2; | |
2326 | if (!dotdot_real_len) | |
2327 | de->rec_len = ext4_rec_len_to_disk(blocksize - | |
2328 | (csum_size + EXT4_DIR_REC_LEN(1)), | |
2329 | blocksize); | |
2330 | else | |
2331 | de->rec_len = ext4_rec_len_to_disk( | |
2332 | EXT4_DIR_REC_LEN(de->name_len), blocksize); | |
2333 | strcpy(de->name, ".."); | |
2334 | ext4_set_de_type(inode->i_sb, de, S_IFDIR); | |
2335 | ||
2336 | return ext4_next_entry(de, blocksize); | |
2337 | } | |
2338 | ||
2339 | static int ext4_init_new_dir(handle_t *handle, struct inode *dir, | |
2340 | struct inode *inode) | |
ac27a0ec | 2341 | { |
dabd991f | 2342 | struct buffer_head *dir_block = NULL; |
af5bc92d | 2343 | struct ext4_dir_entry_2 *de; |
b0336e8d | 2344 | struct ext4_dir_entry_tail *t; |
dc6982ff | 2345 | ext4_lblk_t block = 0; |
3d0518f4 | 2346 | unsigned int blocksize = dir->i_sb->s_blocksize; |
b0336e8d | 2347 | int csum_size = 0; |
a774f9c2 | 2348 | int err; |
ac27a0ec | 2349 | |
b0336e8d DW |
2350 | if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb, |
2351 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
2352 | csum_size = sizeof(struct ext4_dir_entry_tail); | |
2353 | ||
3c47d541 TM |
2354 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
2355 | err = ext4_try_create_inline_dir(handle, dir, inode); | |
2356 | if (err < 0 && err != -ENOSPC) | |
2357 | goto out; | |
2358 | if (!err) | |
2359 | goto out; | |
2360 | } | |
2361 | ||
dc6982ff | 2362 | inode->i_size = 0; |
0f70b406 TT |
2363 | dir_block = ext4_append(handle, inode, &block); |
2364 | if (IS_ERR(dir_block)) | |
2365 | return PTR_ERR(dir_block); | |
a774f9c2 TM |
2366 | BUFFER_TRACE(dir_block, "get_write_access"); |
2367 | err = ext4_journal_get_write_access(handle, dir_block); | |
2368 | if (err) | |
2369 | goto out; | |
2370 | de = (struct ext4_dir_entry_2 *)dir_block->b_data; | |
2371 | ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0); | |
2372 | set_nlink(inode, 2); | |
2373 | if (csum_size) { | |
2374 | t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize); | |
2375 | initialize_dirent_tail(t, blocksize); | |
2376 | } | |
2377 | ||
2378 | BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata"); | |
2379 | err = ext4_handle_dirty_dirent_node(handle, inode, dir_block); | |
2380 | if (err) | |
2381 | goto out; | |
2382 | set_buffer_verified(dir_block); | |
2383 | out: | |
2384 | brelse(dir_block); | |
2385 | return err; | |
2386 | } | |
2387 | ||
2388 | static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2389 | { | |
2390 | handle_t *handle; | |
2391 | struct inode *inode; | |
1139575a | 2392 | int err, credits, retries = 0; |
a774f9c2 | 2393 | |
f8628a14 | 2394 | if (EXT4_DIR_LINK_MAX(dir)) |
ac27a0ec DK |
2395 | return -EMLINK; |
2396 | ||
871a2931 | 2397 | dquot_initialize(dir); |
907f4554 | 2398 | |
1139575a TT |
2399 | credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2400 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
2401 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb)); | |
ac27a0ec | 2402 | retry: |
1139575a TT |
2403 | inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode, |
2404 | &dentry->d_name, | |
2405 | 0, NULL, EXT4_HT_DIR, credits); | |
2406 | handle = ext4_journal_current_handle(); | |
ac27a0ec DK |
2407 | err = PTR_ERR(inode); |
2408 | if (IS_ERR(inode)) | |
2409 | goto out_stop; | |
2410 | ||
617ba13b MC |
2411 | inode->i_op = &ext4_dir_inode_operations; |
2412 | inode->i_fop = &ext4_dir_operations; | |
a774f9c2 | 2413 | err = ext4_init_new_dir(handle, dir, inode); |
dabd991f NK |
2414 | if (err) |
2415 | goto out_clear_inode; | |
2416 | err = ext4_mark_inode_dirty(handle, inode); | |
2417 | if (!err) | |
2418 | err = ext4_add_entry(handle, dentry, inode); | |
ac27a0ec | 2419 | if (err) { |
4cdeed86 AK |
2420 | out_clear_inode: |
2421 | clear_nlink(inode); | |
6b38e842 | 2422 | unlock_new_inode(inode); |
617ba13b | 2423 | ext4_mark_inode_dirty(handle, inode); |
af5bc92d | 2424 | iput(inode); |
ac27a0ec DK |
2425 | goto out_stop; |
2426 | } | |
f8628a14 | 2427 | ext4_inc_count(handle, dir); |
617ba13b | 2428 | ext4_update_dx_flag(dir); |
dabd991f NK |
2429 | err = ext4_mark_inode_dirty(handle, dir); |
2430 | if (err) | |
2431 | goto out_clear_inode; | |
6b38e842 | 2432 | unlock_new_inode(inode); |
8fc37ec5 | 2433 | d_instantiate(dentry, inode); |
1139575a TT |
2434 | if (IS_DIRSYNC(dir)) |
2435 | ext4_handle_sync(handle); | |
2436 | ||
ac27a0ec | 2437 | out_stop: |
1139575a TT |
2438 | if (handle) |
2439 | ext4_journal_stop(handle); | |
617ba13b | 2440 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) |
ac27a0ec DK |
2441 | goto retry; |
2442 | return err; | |
2443 | } | |
2444 | ||
2445 | /* | |
2446 | * routine to check that the specified directory is empty (for rmdir) | |
2447 | */ | |
af5bc92d | 2448 | static int empty_dir(struct inode *inode) |
ac27a0ec | 2449 | { |
498e5f24 | 2450 | unsigned int offset; |
af5bc92d TT |
2451 | struct buffer_head *bh; |
2452 | struct ext4_dir_entry_2 *de, *de1; | |
2453 | struct super_block *sb; | |
ac27a0ec DK |
2454 | int err = 0; |
2455 | ||
61f86638 TM |
2456 | if (ext4_has_inline_data(inode)) { |
2457 | int has_inline_data = 1; | |
2458 | ||
2459 | err = empty_inline_dir(inode, &has_inline_data); | |
2460 | if (has_inline_data) | |
2461 | return err; | |
2462 | } | |
2463 | ||
ac27a0ec | 2464 | sb = inode->i_sb; |
dc6982ff TT |
2465 | if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) { |
2466 | EXT4_ERROR_INODE(inode, "invalid size"); | |
ac27a0ec DK |
2467 | return 1; |
2468 | } | |
dc6982ff TT |
2469 | bh = ext4_read_dirblock(inode, 0, EITHER); |
2470 | if (IS_ERR(bh)) | |
2471 | return 1; | |
2472 | ||
617ba13b | 2473 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
3d0518f4 | 2474 | de1 = ext4_next_entry(de, sb->s_blocksize); |
ac27a0ec DK |
2475 | if (le32_to_cpu(de->inode) != inode->i_ino || |
2476 | !le32_to_cpu(de1->inode) || | |
af5bc92d TT |
2477 | strcmp(".", de->name) || |
2478 | strcmp("..", de1->name)) { | |
12062ddd | 2479 | ext4_warning(inode->i_sb, |
af5bc92d TT |
2480 | "bad directory (dir #%lu) - no `.' or `..'", |
2481 | inode->i_ino); | |
2482 | brelse(bh); | |
ac27a0ec DK |
2483 | return 1; |
2484 | } | |
3d0518f4 WY |
2485 | offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) + |
2486 | ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize); | |
2487 | de = ext4_next_entry(de1, sb->s_blocksize); | |
af5bc92d | 2488 | while (offset < inode->i_size) { |
ac27a0ec | 2489 | if (!bh || |
24676da4 TT |
2490 | (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) { |
2491 | unsigned int lblock; | |
ac27a0ec | 2492 | err = 0; |
af5bc92d | 2493 | brelse(bh); |
24676da4 | 2494 | lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb); |
dc6982ff TT |
2495 | bh = ext4_read_dirblock(inode, lblock, EITHER); |
2496 | if (IS_ERR(bh)) | |
2497 | return 1; | |
617ba13b | 2498 | de = (struct ext4_dir_entry_2 *) bh->b_data; |
ac27a0ec | 2499 | } |
226ba972 TM |
2500 | if (ext4_check_dir_entry(inode, NULL, de, bh, |
2501 | bh->b_data, bh->b_size, offset)) { | |
617ba13b | 2502 | de = (struct ext4_dir_entry_2 *)(bh->b_data + |
ac27a0ec DK |
2503 | sb->s_blocksize); |
2504 | offset = (offset | (sb->s_blocksize - 1)) + 1; | |
2505 | continue; | |
2506 | } | |
2507 | if (le32_to_cpu(de->inode)) { | |
af5bc92d | 2508 | brelse(bh); |
ac27a0ec DK |
2509 | return 0; |
2510 | } | |
3d0518f4 WY |
2511 | offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize); |
2512 | de = ext4_next_entry(de, sb->s_blocksize); | |
ac27a0ec | 2513 | } |
af5bc92d | 2514 | brelse(bh); |
ac27a0ec DK |
2515 | return 1; |
2516 | } | |
2517 | ||
617ba13b | 2518 | /* ext4_orphan_add() links an unlinked or truncated inode into a list of |
ac27a0ec DK |
2519 | * such inodes, starting at the superblock, in case we crash before the |
2520 | * file is closed/deleted, or in case the inode truncate spans multiple | |
2521 | * transactions and the last transaction is not recovered after a crash. | |
2522 | * | |
2523 | * At filesystem recovery time, we walk this list deleting unlinked | |
617ba13b | 2524 | * inodes and truncating linked inodes in ext4_orphan_cleanup(). |
ac27a0ec | 2525 | */ |
617ba13b | 2526 | int ext4_orphan_add(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
2527 | { |
2528 | struct super_block *sb = inode->i_sb; | |
617ba13b | 2529 | struct ext4_iloc iloc; |
ac27a0ec DK |
2530 | int err = 0, rc; |
2531 | ||
c9b92530 | 2532 | if (!EXT4_SB(sb)->s_journal) |
0390131b FM |
2533 | return 0; |
2534 | ||
3b9d4ed2 | 2535 | mutex_lock(&EXT4_SB(sb)->s_orphan_lock); |
617ba13b | 2536 | if (!list_empty(&EXT4_I(inode)->i_orphan)) |
ac27a0ec DK |
2537 | goto out_unlock; |
2538 | ||
afb86178 LC |
2539 | /* |
2540 | * Orphan handling is only valid for files with data blocks | |
2541 | * being truncated, or files being unlinked. Note that we either | |
2542 | * hold i_mutex, or the inode can not be referenced from outside, | |
2543 | * so i_nlink should not be bumped due to race | |
ac27a0ec | 2544 | */ |
af5bc92d TT |
2545 | J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
2546 | S_ISLNK(inode->i_mode)) || inode->i_nlink == 0); | |
ac27a0ec | 2547 | |
617ba13b MC |
2548 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access"); |
2549 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); | |
ac27a0ec DK |
2550 | if (err) |
2551 | goto out_unlock; | |
2552 | ||
617ba13b | 2553 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
ac27a0ec DK |
2554 | if (err) |
2555 | goto out_unlock; | |
6e3617e5 DM |
2556 | /* |
2557 | * Due to previous errors inode may be already a part of on-disk | |
2558 | * orphan list. If so skip on-disk list modification. | |
2559 | */ | |
2560 | if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <= | |
2561 | (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) | |
2562 | goto mem_insert; | |
ac27a0ec DK |
2563 | |
2564 | /* Insert this inode at the head of the on-disk orphan list... */ | |
617ba13b MC |
2565 | NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan); |
2566 | EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino); | |
b50924c2 | 2567 | err = ext4_handle_dirty_super(handle, sb); |
617ba13b | 2568 | rc = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
2569 | if (!err) |
2570 | err = rc; | |
2571 | ||
2572 | /* Only add to the head of the in-memory list if all the | |
2573 | * previous operations succeeded. If the orphan_add is going to | |
2574 | * fail (possibly taking the journal offline), we can't risk | |
2575 | * leaving the inode on the orphan list: stray orphan-list | |
2576 | * entries can cause panics at unmount time. | |
2577 | * | |
2578 | * This is safe: on error we're going to ignore the orphan list | |
2579 | * anyway on the next recovery. */ | |
6e3617e5 | 2580 | mem_insert: |
ac27a0ec | 2581 | if (!err) |
617ba13b | 2582 | list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan); |
ac27a0ec DK |
2583 | |
2584 | jbd_debug(4, "superblock will point to %lu\n", inode->i_ino); | |
2585 | jbd_debug(4, "orphan inode %lu will point to %d\n", | |
2586 | inode->i_ino, NEXT_ORPHAN(inode)); | |
2587 | out_unlock: | |
3b9d4ed2 | 2588 | mutex_unlock(&EXT4_SB(sb)->s_orphan_lock); |
617ba13b | 2589 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
2590 | return err; |
2591 | } | |
2592 | ||
2593 | /* | |
617ba13b | 2594 | * ext4_orphan_del() removes an unlinked or truncated inode from the list |
ac27a0ec DK |
2595 | * of such inodes stored on disk, because it is finally being cleaned up. |
2596 | */ | |
617ba13b | 2597 | int ext4_orphan_del(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
2598 | { |
2599 | struct list_head *prev; | |
617ba13b MC |
2600 | struct ext4_inode_info *ei = EXT4_I(inode); |
2601 | struct ext4_sb_info *sbi; | |
498e5f24 | 2602 | __u32 ino_next; |
617ba13b | 2603 | struct ext4_iloc iloc; |
ac27a0ec DK |
2604 | int err = 0; |
2605 | ||
0e9a9a1a TT |
2606 | if ((!EXT4_SB(inode->i_sb)->s_journal) && |
2607 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) | |
0390131b FM |
2608 | return 0; |
2609 | ||
3b9d4ed2 TT |
2610 | mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock); |
2611 | if (list_empty(&ei->i_orphan)) | |
2612 | goto out; | |
ac27a0ec DK |
2613 | |
2614 | ino_next = NEXT_ORPHAN(inode); | |
2615 | prev = ei->i_orphan.prev; | |
617ba13b | 2616 | sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
2617 | |
2618 | jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino); | |
2619 | ||
2620 | list_del_init(&ei->i_orphan); | |
2621 | ||
2622 | /* If we're on an error path, we may not have a valid | |
2623 | * transaction handle with which to update the orphan list on | |
2624 | * disk, but we still need to remove the inode from the linked | |
2625 | * list in memory. */ | |
c9b92530 | 2626 | if (!handle) |
ac27a0ec DK |
2627 | goto out; |
2628 | ||
617ba13b | 2629 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
ac27a0ec DK |
2630 | if (err) |
2631 | goto out_err; | |
2632 | ||
2633 | if (prev == &sbi->s_orphan) { | |
498e5f24 | 2634 | jbd_debug(4, "superblock will point to %u\n", ino_next); |
ac27a0ec | 2635 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
617ba13b | 2636 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); |
ac27a0ec DK |
2637 | if (err) |
2638 | goto out_brelse; | |
2639 | sbi->s_es->s_last_orphan = cpu_to_le32(ino_next); | |
b50924c2 | 2640 | err = ext4_handle_dirty_super(handle, inode->i_sb); |
ac27a0ec | 2641 | } else { |
617ba13b | 2642 | struct ext4_iloc iloc2; |
ac27a0ec | 2643 | struct inode *i_prev = |
617ba13b | 2644 | &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode; |
ac27a0ec | 2645 | |
498e5f24 | 2646 | jbd_debug(4, "orphan inode %lu will point to %u\n", |
ac27a0ec | 2647 | i_prev->i_ino, ino_next); |
617ba13b | 2648 | err = ext4_reserve_inode_write(handle, i_prev, &iloc2); |
ac27a0ec DK |
2649 | if (err) |
2650 | goto out_brelse; | |
2651 | NEXT_ORPHAN(i_prev) = ino_next; | |
617ba13b | 2652 | err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2); |
ac27a0ec DK |
2653 | } |
2654 | if (err) | |
2655 | goto out_brelse; | |
2656 | NEXT_ORPHAN(inode) = 0; | |
617ba13b | 2657 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
2658 | |
2659 | out_err: | |
617ba13b | 2660 | ext4_std_error(inode->i_sb, err); |
ac27a0ec | 2661 | out: |
3b9d4ed2 | 2662 | mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock); |
ac27a0ec DK |
2663 | return err; |
2664 | ||
2665 | out_brelse: | |
2666 | brelse(iloc.bh); | |
2667 | goto out_err; | |
2668 | } | |
2669 | ||
af5bc92d | 2670 | static int ext4_rmdir(struct inode *dir, struct dentry *dentry) |
ac27a0ec DK |
2671 | { |
2672 | int retval; | |
af5bc92d TT |
2673 | struct inode *inode; |
2674 | struct buffer_head *bh; | |
2675 | struct ext4_dir_entry_2 *de; | |
8dcfaad2 | 2676 | handle_t *handle = NULL; |
ac27a0ec DK |
2677 | |
2678 | /* Initialize quotas before so that eventual writes go in | |
2679 | * separate transaction */ | |
871a2931 CH |
2680 | dquot_initialize(dir); |
2681 | dquot_initialize(dentry->d_inode); | |
907f4554 | 2682 | |
ac27a0ec | 2683 | retval = -ENOENT; |
32f7f22c | 2684 | bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); |
ac27a0ec DK |
2685 | if (!bh) |
2686 | goto end_rmdir; | |
2687 | ||
ac27a0ec DK |
2688 | inode = dentry->d_inode; |
2689 | ||
2690 | retval = -EIO; | |
2691 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2692 | goto end_rmdir; | |
2693 | ||
2694 | retval = -ENOTEMPTY; | |
af5bc92d | 2695 | if (!empty_dir(inode)) |
ac27a0ec DK |
2696 | goto end_rmdir; |
2697 | ||
8dcfaad2 | 2698 | handle = ext4_journal_start(dir, EXT4_HT_DIR, |
64044abf | 2699 | EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); |
8dcfaad2 TT |
2700 | if (IS_ERR(handle)) { |
2701 | retval = PTR_ERR(handle); | |
2702 | handle = NULL; | |
2703 | goto end_rmdir; | |
2704 | } | |
2705 | ||
2706 | if (IS_DIRSYNC(dir)) | |
2707 | ext4_handle_sync(handle); | |
2708 | ||
617ba13b | 2709 | retval = ext4_delete_entry(handle, dir, de, bh); |
ac27a0ec DK |
2710 | if (retval) |
2711 | goto end_rmdir; | |
f8628a14 | 2712 | if (!EXT4_DIR_LINK_EMPTY(inode)) |
12062ddd | 2713 | ext4_warning(inode->i_sb, |
af5bc92d TT |
2714 | "empty directory has too many links (%d)", |
2715 | inode->i_nlink); | |
ac27a0ec DK |
2716 | inode->i_version++; |
2717 | clear_nlink(inode); | |
2718 | /* There's no need to set i_disksize: the fact that i_nlink is | |
2719 | * zero will ensure that the right thing happens during any | |
2720 | * recovery. */ | |
2721 | inode->i_size = 0; | |
617ba13b | 2722 | ext4_orphan_add(handle, inode); |
ef7f3835 | 2723 | inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode); |
617ba13b | 2724 | ext4_mark_inode_dirty(handle, inode); |
f8628a14 | 2725 | ext4_dec_count(handle, dir); |
617ba13b MC |
2726 | ext4_update_dx_flag(dir); |
2727 | ext4_mark_inode_dirty(handle, dir); | |
ac27a0ec DK |
2728 | |
2729 | end_rmdir: | |
af5bc92d | 2730 | brelse(bh); |
8dcfaad2 TT |
2731 | if (handle) |
2732 | ext4_journal_stop(handle); | |
ac27a0ec DK |
2733 | return retval; |
2734 | } | |
2735 | ||
af5bc92d | 2736 | static int ext4_unlink(struct inode *dir, struct dentry *dentry) |
ac27a0ec DK |
2737 | { |
2738 | int retval; | |
af5bc92d TT |
2739 | struct inode *inode; |
2740 | struct buffer_head *bh; | |
2741 | struct ext4_dir_entry_2 *de; | |
931b6864 | 2742 | handle_t *handle = NULL; |
ac27a0ec | 2743 | |
0562e0ba | 2744 | trace_ext4_unlink_enter(dir, dentry); |
ac27a0ec DK |
2745 | /* Initialize quotas before so that eventual writes go |
2746 | * in separate transaction */ | |
871a2931 CH |
2747 | dquot_initialize(dir); |
2748 | dquot_initialize(dentry->d_inode); | |
907f4554 | 2749 | |
ac27a0ec | 2750 | retval = -ENOENT; |
32f7f22c | 2751 | bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL); |
ac27a0ec DK |
2752 | if (!bh) |
2753 | goto end_unlink; | |
2754 | ||
2755 | inode = dentry->d_inode; | |
2756 | ||
2757 | retval = -EIO; | |
2758 | if (le32_to_cpu(de->inode) != inode->i_ino) | |
2759 | goto end_unlink; | |
2760 | ||
931b6864 | 2761 | handle = ext4_journal_start(dir, EXT4_HT_DIR, |
64044abf | 2762 | EXT4_DATA_TRANS_BLOCKS(dir->i_sb)); |
931b6864 TT |
2763 | if (IS_ERR(handle)) { |
2764 | retval = PTR_ERR(handle); | |
2765 | handle = NULL; | |
2766 | goto end_unlink; | |
2767 | } | |
2768 | ||
2769 | if (IS_DIRSYNC(dir)) | |
2770 | ext4_handle_sync(handle); | |
2771 | ||
ac27a0ec | 2772 | if (!inode->i_nlink) { |
12062ddd | 2773 | ext4_warning(inode->i_sb, |
af5bc92d TT |
2774 | "Deleting nonexistent file (%lu), %d", |
2775 | inode->i_ino, inode->i_nlink); | |
bfe86848 | 2776 | set_nlink(inode, 1); |
ac27a0ec | 2777 | } |
617ba13b | 2778 | retval = ext4_delete_entry(handle, dir, de, bh); |
ac27a0ec DK |
2779 | if (retval) |
2780 | goto end_unlink; | |
ef7f3835 | 2781 | dir->i_ctime = dir->i_mtime = ext4_current_time(dir); |
617ba13b MC |
2782 | ext4_update_dx_flag(dir); |
2783 | ext4_mark_inode_dirty(handle, dir); | |
825f1481 | 2784 | drop_nlink(inode); |
ac27a0ec | 2785 | if (!inode->i_nlink) |
617ba13b | 2786 | ext4_orphan_add(handle, inode); |
ef7f3835 | 2787 | inode->i_ctime = ext4_current_time(inode); |
617ba13b | 2788 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
2789 | retval = 0; |
2790 | ||
2791 | end_unlink: | |
af5bc92d | 2792 | brelse(bh); |
931b6864 TT |
2793 | if (handle) |
2794 | ext4_journal_stop(handle); | |
0562e0ba | 2795 | trace_ext4_unlink_exit(dentry, retval); |
ac27a0ec DK |
2796 | return retval; |
2797 | } | |
2798 | ||
af5bc92d TT |
2799 | static int ext4_symlink(struct inode *dir, |
2800 | struct dentry *dentry, const char *symname) | |
ac27a0ec DK |
2801 | { |
2802 | handle_t *handle; | |
af5bc92d | 2803 | struct inode *inode; |
ac27a0ec | 2804 | int l, err, retries = 0; |
df5e6223 | 2805 | int credits; |
ac27a0ec DK |
2806 | |
2807 | l = strlen(symname)+1; | |
2808 | if (l > dir->i_sb->s_blocksize) | |
2809 | return -ENAMETOOLONG; | |
2810 | ||
871a2931 | 2811 | dquot_initialize(dir); |
907f4554 | 2812 | |
df5e6223 JK |
2813 | if (l > EXT4_N_BLOCKS * 4) { |
2814 | /* | |
2815 | * For non-fast symlinks, we just allocate inode and put it on | |
2816 | * orphan list in the first transaction => we need bitmap, | |
8c208719 ES |
2817 | * group descriptor, sb, inode block, quota blocks, and |
2818 | * possibly selinux xattr blocks. | |
df5e6223 | 2819 | */ |
8c208719 ES |
2820 | credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) + |
2821 | EXT4_XATTR_TRANS_BLOCKS; | |
df5e6223 JK |
2822 | } else { |
2823 | /* | |
2824 | * Fast symlink. We have to add entry to directory | |
2825 | * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS), | |
2826 | * allocate new inode (bitmap, group descriptor, inode block, | |
2827 | * quota blocks, sb is already counted in previous macros). | |
2828 | */ | |
2829 | credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
2830 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 + | |
2831 | EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb); | |
2832 | } | |
ac27a0ec | 2833 | retry: |
1139575a TT |
2834 | inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO, |
2835 | &dentry->d_name, 0, NULL, | |
2836 | EXT4_HT_DIR, credits); | |
2837 | handle = ext4_journal_current_handle(); | |
ac27a0ec DK |
2838 | err = PTR_ERR(inode); |
2839 | if (IS_ERR(inode)) | |
2840 | goto out_stop; | |
2841 | ||
df5e6223 | 2842 | if (l > EXT4_N_BLOCKS * 4) { |
617ba13b MC |
2843 | inode->i_op = &ext4_symlink_inode_operations; |
2844 | ext4_set_aops(inode); | |
ac27a0ec | 2845 | /* |
df5e6223 JK |
2846 | * We cannot call page_symlink() with transaction started |
2847 | * because it calls into ext4_write_begin() which can wait | |
2848 | * for transaction commit if we are running out of space | |
2849 | * and thus we deadlock. So we have to stop transaction now | |
2850 | * and restart it when symlink contents is written. | |
2851 | * | |
2852 | * To keep fs consistent in case of crash, we have to put inode | |
2853 | * to orphan list in the mean time. | |
ac27a0ec | 2854 | */ |
df5e6223 JK |
2855 | drop_nlink(inode); |
2856 | err = ext4_orphan_add(handle, inode); | |
2857 | ext4_journal_stop(handle); | |
2858 | if (err) | |
2859 | goto err_drop_inode; | |
54566b2c | 2860 | err = __page_symlink(inode, symname, l, 1); |
df5e6223 JK |
2861 | if (err) |
2862 | goto err_drop_inode; | |
2863 | /* | |
2864 | * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS | |
2865 | * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified | |
2866 | */ | |
9924a92a | 2867 | handle = ext4_journal_start(dir, EXT4_HT_DIR, |
df5e6223 JK |
2868 | EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + |
2869 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1); | |
2870 | if (IS_ERR(handle)) { | |
2871 | err = PTR_ERR(handle); | |
2872 | goto err_drop_inode; | |
2873 | } | |
0ce8c010 | 2874 | set_nlink(inode, 1); |
df5e6223 | 2875 | err = ext4_orphan_del(handle, inode); |
ac27a0ec | 2876 | if (err) { |
df5e6223 | 2877 | ext4_journal_stop(handle); |
825f1481 | 2878 | clear_nlink(inode); |
df5e6223 | 2879 | goto err_drop_inode; |
ac27a0ec DK |
2880 | } |
2881 | } else { | |
e65187e6 | 2882 | /* clear the extent format for fast symlink */ |
12e9b892 | 2883 | ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); |
617ba13b | 2884 | inode->i_op = &ext4_fast_symlink_inode_operations; |
af5bc92d | 2885 | memcpy((char *)&EXT4_I(inode)->i_data, symname, l); |
ac27a0ec DK |
2886 | inode->i_size = l-1; |
2887 | } | |
617ba13b MC |
2888 | EXT4_I(inode)->i_disksize = inode->i_size; |
2889 | err = ext4_add_nondir(handle, dentry, inode); | |
1139575a TT |
2890 | if (!err && IS_DIRSYNC(dir)) |
2891 | ext4_handle_sync(handle); | |
2892 | ||
ac27a0ec | 2893 | out_stop: |
1139575a TT |
2894 | if (handle) |
2895 | ext4_journal_stop(handle); | |
617ba13b | 2896 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) |
ac27a0ec DK |
2897 | goto retry; |
2898 | return err; | |
df5e6223 JK |
2899 | err_drop_inode: |
2900 | unlock_new_inode(inode); | |
2901 | iput(inode); | |
2902 | return err; | |
ac27a0ec DK |
2903 | } |
2904 | ||
af5bc92d TT |
2905 | static int ext4_link(struct dentry *old_dentry, |
2906 | struct inode *dir, struct dentry *dentry) | |
ac27a0ec DK |
2907 | { |
2908 | handle_t *handle; | |
2909 | struct inode *inode = old_dentry->d_inode; | |
2910 | int err, retries = 0; | |
2911 | ||
b05ab1dc | 2912 | if (inode->i_nlink >= EXT4_LINK_MAX) |
ac27a0ec | 2913 | return -EMLINK; |
f8628a14 | 2914 | |
871a2931 | 2915 | dquot_initialize(dir); |
907f4554 | 2916 | |
ac27a0ec | 2917 | retry: |
9924a92a TT |
2918 | handle = ext4_journal_start(dir, EXT4_HT_DIR, |
2919 | (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) + | |
2920 | EXT4_INDEX_EXTRA_TRANS_BLOCKS)); | |
ac27a0ec DK |
2921 | if (IS_ERR(handle)) |
2922 | return PTR_ERR(handle); | |
2923 | ||
2924 | if (IS_DIRSYNC(dir)) | |
0390131b | 2925 | ext4_handle_sync(handle); |
ac27a0ec | 2926 | |
ef7f3835 | 2927 | inode->i_ctime = ext4_current_time(inode); |
f8628a14 | 2928 | ext4_inc_count(handle, inode); |
7de9c6ee | 2929 | ihold(inode); |
ac27a0ec | 2930 | |
6b38e842 AV |
2931 | err = ext4_add_entry(handle, dentry, inode); |
2932 | if (!err) { | |
2933 | ext4_mark_inode_dirty(handle, inode); | |
2934 | d_instantiate(dentry, inode); | |
2935 | } else { | |
2936 | drop_nlink(inode); | |
2937 | iput(inode); | |
2938 | } | |
617ba13b MC |
2939 | ext4_journal_stop(handle); |
2940 | if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries)) | |
ac27a0ec DK |
2941 | goto retry; |
2942 | return err; | |
2943 | } | |
2944 | ||
32f7f22c TM |
2945 | |
2946 | /* | |
2947 | * Try to find buffer head where contains the parent block. | |
2948 | * It should be the inode block if it is inlined or the 1st block | |
2949 | * if it is a normal dir. | |
2950 | */ | |
2951 | static struct buffer_head *ext4_get_first_dir_block(handle_t *handle, | |
2952 | struct inode *inode, | |
2953 | int *retval, | |
2954 | struct ext4_dir_entry_2 **parent_de, | |
2955 | int *inlined) | |
2956 | { | |
2957 | struct buffer_head *bh; | |
2958 | ||
2959 | if (!ext4_has_inline_data(inode)) { | |
dc6982ff TT |
2960 | bh = ext4_read_dirblock(inode, 0, EITHER); |
2961 | if (IS_ERR(bh)) { | |
2962 | *retval = PTR_ERR(bh); | |
32f7f22c TM |
2963 | return NULL; |
2964 | } | |
2965 | *parent_de = ext4_next_entry( | |
2966 | (struct ext4_dir_entry_2 *)bh->b_data, | |
2967 | inode->i_sb->s_blocksize); | |
2968 | return bh; | |
2969 | } | |
2970 | ||
2971 | *inlined = 1; | |
2972 | return ext4_get_first_inline_block(inode, parent_de, retval); | |
2973 | } | |
ac27a0ec DK |
2974 | |
2975 | /* | |
2976 | * Anybody can rename anything with this: the permission checks are left to the | |
2977 | * higher-level routines. | |
2978 | */ | |
af5bc92d TT |
2979 | static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry, |
2980 | struct inode *new_dir, struct dentry *new_dentry) | |
ac27a0ec DK |
2981 | { |
2982 | handle_t *handle; | |
af5bc92d TT |
2983 | struct inode *old_inode, *new_inode; |
2984 | struct buffer_head *old_bh, *new_bh, *dir_bh; | |
2985 | struct ext4_dir_entry_2 *old_de, *new_de; | |
8750c6d5 | 2986 | int retval, force_da_alloc = 0; |
32f7f22c TM |
2987 | int inlined = 0, new_inlined = 0; |
2988 | struct ext4_dir_entry_2 *parent_de; | |
ac27a0ec | 2989 | |
871a2931 CH |
2990 | dquot_initialize(old_dir); |
2991 | dquot_initialize(new_dir); | |
907f4554 | 2992 | |
ac27a0ec DK |
2993 | old_bh = new_bh = dir_bh = NULL; |
2994 | ||
2995 | /* Initialize quotas before so that eventual writes go | |
2996 | * in separate transaction */ | |
2997 | if (new_dentry->d_inode) | |
871a2931 | 2998 | dquot_initialize(new_dentry->d_inode); |
9924a92a TT |
2999 | handle = ext4_journal_start(old_dir, EXT4_HT_DIR, |
3000 | (2 * EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) + | |
3001 | EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2)); | |
ac27a0ec DK |
3002 | if (IS_ERR(handle)) |
3003 | return PTR_ERR(handle); | |
3004 | ||
3005 | if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir)) | |
0390131b | 3006 | ext4_handle_sync(handle); |
ac27a0ec | 3007 | |
32f7f22c | 3008 | old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de, NULL); |
ac27a0ec DK |
3009 | /* |
3010 | * Check for inode number is _not_ due to possible IO errors. | |
3011 | * We might rmdir the source, keep it as pwd of some process | |
3012 | * and merrily kill the link to whatever was created under the | |
3013 | * same name. Goodbye sticky bit ;-< | |
3014 | */ | |
3015 | old_inode = old_dentry->d_inode; | |
3016 | retval = -ENOENT; | |
3017 | if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino) | |
3018 | goto end_rename; | |
3019 | ||
3020 | new_inode = new_dentry->d_inode; | |
32f7f22c TM |
3021 | new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, |
3022 | &new_de, &new_inlined); | |
ac27a0ec DK |
3023 | if (new_bh) { |
3024 | if (!new_inode) { | |
af5bc92d | 3025 | brelse(new_bh); |
ac27a0ec DK |
3026 | new_bh = NULL; |
3027 | } | |
3028 | } | |
3029 | if (S_ISDIR(old_inode->i_mode)) { | |
3030 | if (new_inode) { | |
3031 | retval = -ENOTEMPTY; | |
af5bc92d | 3032 | if (!empty_dir(new_inode)) |
ac27a0ec DK |
3033 | goto end_rename; |
3034 | } | |
3035 | retval = -EIO; | |
32f7f22c TM |
3036 | dir_bh = ext4_get_first_dir_block(handle, old_inode, |
3037 | &retval, &parent_de, | |
3038 | &inlined); | |
3039 | if (!dir_bh) | |
ac27a0ec | 3040 | goto end_rename; |
32f7f22c | 3041 | if (le32_to_cpu(parent_de->inode) != old_dir->i_ino) |
ac27a0ec DK |
3042 | goto end_rename; |
3043 | retval = -EMLINK; | |
af5bc92d | 3044 | if (!new_inode && new_dir != old_dir && |
2c94eb86 | 3045 | EXT4_DIR_LINK_MAX(new_dir)) |
ac27a0ec | 3046 | goto end_rename; |
ef607893 AG |
3047 | BUFFER_TRACE(dir_bh, "get_write_access"); |
3048 | retval = ext4_journal_get_write_access(handle, dir_bh); | |
3049 | if (retval) | |
3050 | goto end_rename; | |
ac27a0ec DK |
3051 | } |
3052 | if (!new_bh) { | |
af5bc92d | 3053 | retval = ext4_add_entry(handle, new_dentry, old_inode); |
ac27a0ec DK |
3054 | if (retval) |
3055 | goto end_rename; | |
3056 | } else { | |
3057 | BUFFER_TRACE(new_bh, "get write access"); | |
ef607893 AG |
3058 | retval = ext4_journal_get_write_access(handle, new_bh); |
3059 | if (retval) | |
3060 | goto end_rename; | |
ac27a0ec | 3061 | new_de->inode = cpu_to_le32(old_inode->i_ino); |
617ba13b MC |
3062 | if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb, |
3063 | EXT4_FEATURE_INCOMPAT_FILETYPE)) | |
ac27a0ec DK |
3064 | new_de->file_type = old_de->file_type; |
3065 | new_dir->i_version++; | |
53b7e9f6 JK |
3066 | new_dir->i_ctime = new_dir->i_mtime = |
3067 | ext4_current_time(new_dir); | |
3068 | ext4_mark_inode_dirty(handle, new_dir); | |
0390131b | 3069 | BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata"); |
32f7f22c TM |
3070 | if (!new_inlined) { |
3071 | retval = ext4_handle_dirty_dirent_node(handle, | |
3072 | new_dir, new_bh); | |
3073 | if (unlikely(retval)) { | |
3074 | ext4_std_error(new_dir->i_sb, retval); | |
3075 | goto end_rename; | |
3076 | } | |
b4097142 | 3077 | } |
ac27a0ec DK |
3078 | brelse(new_bh); |
3079 | new_bh = NULL; | |
3080 | } | |
3081 | ||
3082 | /* | |
3083 | * Like most other Unix systems, set the ctime for inodes on a | |
3084 | * rename. | |
3085 | */ | |
ef7f3835 | 3086 | old_inode->i_ctime = ext4_current_time(old_inode); |
617ba13b | 3087 | ext4_mark_inode_dirty(handle, old_inode); |
ac27a0ec DK |
3088 | |
3089 | /* | |
3090 | * ok, that's it | |
3091 | */ | |
3092 | if (le32_to_cpu(old_de->inode) != old_inode->i_ino || | |
3093 | old_de->name_len != old_dentry->d_name.len || | |
3094 | strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) || | |
617ba13b | 3095 | (retval = ext4_delete_entry(handle, old_dir, |
ac27a0ec DK |
3096 | old_de, old_bh)) == -ENOENT) { |
3097 | /* old_de could have moved from under us during htree split, so | |
3098 | * make sure that we are deleting the right entry. We might | |
3099 | * also be pointing to a stale entry in the unused part of | |
3100 | * old_bh so just checking inum and the name isn't enough. */ | |
3101 | struct buffer_head *old_bh2; | |
617ba13b | 3102 | struct ext4_dir_entry_2 *old_de2; |
ac27a0ec | 3103 | |
32f7f22c TM |
3104 | old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, |
3105 | &old_de2, NULL); | |
ac27a0ec | 3106 | if (old_bh2) { |
617ba13b | 3107 | retval = ext4_delete_entry(handle, old_dir, |
ac27a0ec DK |
3108 | old_de2, old_bh2); |
3109 | brelse(old_bh2); | |
3110 | } | |
3111 | } | |
3112 | if (retval) { | |
12062ddd | 3113 | ext4_warning(old_dir->i_sb, |
ac27a0ec DK |
3114 | "Deleting old file (%lu), %d, error=%d", |
3115 | old_dir->i_ino, old_dir->i_nlink, retval); | |
3116 | } | |
3117 | ||
3118 | if (new_inode) { | |
f8628a14 | 3119 | ext4_dec_count(handle, new_inode); |
ef7f3835 | 3120 | new_inode->i_ctime = ext4_current_time(new_inode); |
ac27a0ec | 3121 | } |
ef7f3835 | 3122 | old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir); |
617ba13b | 3123 | ext4_update_dx_flag(old_dir); |
ac27a0ec | 3124 | if (dir_bh) { |
32f7f22c | 3125 | parent_de->inode = cpu_to_le32(new_dir->i_ino); |
0390131b | 3126 | BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata"); |
32f7f22c TM |
3127 | if (!inlined) { |
3128 | if (is_dx(old_inode)) { | |
3129 | retval = ext4_handle_dirty_dx_node(handle, | |
3130 | old_inode, | |
3131 | dir_bh); | |
3132 | } else { | |
3133 | retval = ext4_handle_dirty_dirent_node(handle, | |
3134 | old_inode, dir_bh); | |
3135 | } | |
ef58f69c | 3136 | } else { |
32f7f22c | 3137 | retval = ext4_mark_inode_dirty(handle, old_inode); |
ef58f69c | 3138 | } |
b4097142 TT |
3139 | if (retval) { |
3140 | ext4_std_error(old_dir->i_sb, retval); | |
3141 | goto end_rename; | |
3142 | } | |
f8628a14 | 3143 | ext4_dec_count(handle, old_dir); |
ac27a0ec | 3144 | if (new_inode) { |
f8628a14 | 3145 | /* checked empty_dir above, can't have another parent, |
825f1481 | 3146 | * ext4_dec_count() won't work for many-linked dirs */ |
6d6b77f1 | 3147 | clear_nlink(new_inode); |
ac27a0ec | 3148 | } else { |
f8628a14 | 3149 | ext4_inc_count(handle, new_dir); |
617ba13b MC |
3150 | ext4_update_dx_flag(new_dir); |
3151 | ext4_mark_inode_dirty(handle, new_dir); | |
ac27a0ec DK |
3152 | } |
3153 | } | |
617ba13b | 3154 | ext4_mark_inode_dirty(handle, old_dir); |
ac27a0ec | 3155 | if (new_inode) { |
617ba13b | 3156 | ext4_mark_inode_dirty(handle, new_inode); |
ac27a0ec | 3157 | if (!new_inode->i_nlink) |
617ba13b | 3158 | ext4_orphan_add(handle, new_inode); |
afd4672d TT |
3159 | if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC)) |
3160 | force_da_alloc = 1; | |
ac27a0ec DK |
3161 | } |
3162 | retval = 0; | |
3163 | ||
3164 | end_rename: | |
af5bc92d TT |
3165 | brelse(dir_bh); |
3166 | brelse(old_bh); | |
3167 | brelse(new_bh); | |
617ba13b | 3168 | ext4_journal_stop(handle); |
8750c6d5 TT |
3169 | if (retval == 0 && force_da_alloc) |
3170 | ext4_alloc_da_blocks(old_inode); | |
ac27a0ec DK |
3171 | return retval; |
3172 | } | |
3173 | ||
3174 | /* | |
3175 | * directories can handle most operations... | |
3176 | */ | |
754661f1 | 3177 | const struct inode_operations ext4_dir_inode_operations = { |
617ba13b MC |
3178 | .create = ext4_create, |
3179 | .lookup = ext4_lookup, | |
3180 | .link = ext4_link, | |
3181 | .unlink = ext4_unlink, | |
3182 | .symlink = ext4_symlink, | |
3183 | .mkdir = ext4_mkdir, | |
3184 | .rmdir = ext4_rmdir, | |
3185 | .mknod = ext4_mknod, | |
3186 | .rename = ext4_rename, | |
3187 | .setattr = ext4_setattr, | |
ac27a0ec DK |
3188 | .setxattr = generic_setxattr, |
3189 | .getxattr = generic_getxattr, | |
617ba13b | 3190 | .listxattr = ext4_listxattr, |
ac27a0ec | 3191 | .removexattr = generic_removexattr, |
4e34e719 | 3192 | .get_acl = ext4_get_acl, |
abc8746e | 3193 | .fiemap = ext4_fiemap, |
ac27a0ec DK |
3194 | }; |
3195 | ||
754661f1 | 3196 | const struct inode_operations ext4_special_inode_operations = { |
617ba13b | 3197 | .setattr = ext4_setattr, |
ac27a0ec DK |
3198 | .setxattr = generic_setxattr, |
3199 | .getxattr = generic_getxattr, | |
617ba13b | 3200 | .listxattr = ext4_listxattr, |
ac27a0ec | 3201 | .removexattr = generic_removexattr, |
4e34e719 | 3202 | .get_acl = ext4_get_acl, |
ac27a0ec | 3203 | }; |