?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
pvrusb2: reduce stack usage pvr2_eeprom_analyze()
[linux/fpc-iii.git] / fs / ext4 / namei.c
blob423a21cd077cd25a43fc13657502f6f25f252427
1 /*
2 * linux/fs/ext4/namei.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
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>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
36 #include "ext4.h"
37 #include "ext4_jbd2.h"
38
39 #include "xattr.h"
40 #include "acl.h"
41
42 #include <trace/events/ext4.h>
43 /*
44 * define how far ahead to read directories while searching them.
45 */
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49
50 static struct buffer_head *ext4_append(handle_t *handle,
51 struct inode *inode,
52 ext4_lblk_t *block)
53 {
54 struct buffer_head *bh;
55 int err;
56
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
61
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63
64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
65 if (IS_ERR(bh))
66 return bh;
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
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);
75 }
76 return bh;
77 }
78
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), __func__, __LINE__)
88
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 ext4_lblk_t block,
91 dirblock_type_t type,
92 const char *func,
93 unsigned int line)
94 {
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
97 int is_dx_block = 0;
98
99 bh = ext4_bread(NULL, inode, block, 0);
100 if (IS_ERR(bh)) {
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
106
107 return bh;
108 }
109 if (!bh) {
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED);
113 }
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
116 if (is_dx(inode)) {
117 if (block == 0)
118 is_dx_block = 1;
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
122 is_dx_block = 1;
123 }
124 if (!is_dx_block && type == INDEX) {
125 ext4_error_inode(inode, func, line, block,
126 "directory leaf block found instead of index block");
127 return ERR_PTR(-EFSCORRUPTED);
128 }
129 if (!ext4_has_metadata_csum(inode->i_sb) ||
130 buffer_verified(bh))
131 return bh;
132
133 /*
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
137 */
138 if (is_dx_block && type == INDEX) {
139 if (ext4_dx_csum_verify(inode, dirent))
140 set_buffer_verified(bh);
141 else {
142 ext4_error_inode(inode, func, line, block,
143 "Directory index failed checksum");
144 brelse(bh);
145 return ERR_PTR(-EFSBADCRC);
146 }
147 }
148 if (!is_dx_block) {
149 if (ext4_dirent_csum_verify(inode, dirent))
150 set_buffer_verified(bh);
151 else {
152 ext4_error_inode(inode, func, line, block,
153 "Directory block failed checksum");
154 brelse(bh);
155 return ERR_PTR(-EFSBADCRC);
156 }
157 }
158 return bh;
159 }
160
161 #ifndef assert
162 #define assert(test) J_ASSERT(test)
163 #endif
164
165 #ifdef DX_DEBUG
166 #define dxtrace(command) command
167 #else
168 #define dxtrace(command)
169 #endif
170
171 struct fake_dirent
172 {
173 __le32 inode;
174 __le16 rec_len;
175 u8 name_len;
176 u8 file_type;
177 };
178
179 struct dx_countlimit
180 {
181 __le16 limit;
182 __le16 count;
183 };
184
185 struct dx_entry
186 {
187 __le32 hash;
188 __le32 block;
189 };
190
191 /*
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
195 */
196
197 struct dx_root
198 {
199 struct fake_dirent dot;
200 char dot_name[4];
201 struct fake_dirent dotdot;
202 char dotdot_name[4];
203 struct dx_root_info
204 {
205 __le32 reserved_zero;
206 u8 hash_version;
207 u8 info_length; /* 8 */
208 u8 indirect_levels;
209 u8 unused_flags;
210 }
211 info;
212 struct dx_entry entries[0];
213 };
214
215 struct dx_node
216 {
217 struct fake_dirent fake;
218 struct dx_entry entries[0];
219 };
220
221
222 struct dx_frame
223 {
224 struct buffer_head *bh;
225 struct dx_entry *entries;
226 struct dx_entry *at;
227 };
228
229 struct dx_map_entry
230 {
231 u32 hash;
232 u16 offs;
233 u16 size;
234 };
235
236 /*
237 * This goes at the end of each htree block.
238 */
239 struct dx_tail {
240 u32 dt_reserved;
241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
242 };
243
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
255 struct inode *dir,
256 struct dx_hash_info *hinfo,
257 struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 unsigned blocksize, struct dx_hash_info *hinfo,
261 struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267 u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 struct dx_frame *frame,
270 struct dx_frame *frames,
271 __u32 *start_hash);
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 struct ext4_filename *fname,
274 struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 struct inode *dir, struct inode *inode);
277
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 unsigned int blocksize)
281 {
282 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 t->det_rec_len = ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail), blocksize);
285 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
286 }
287
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 struct ext4_dir_entry *de)
291 {
292 struct ext4_dir_entry_tail *t;
293
294 #ifdef PARANOID
295 struct ext4_dir_entry *d, *top;
296
297 d = de;
298 top = (struct ext4_dir_entry *)(((void *)de) +
299 (EXT4_BLOCK_SIZE(inode->i_sb) -
300 sizeof(struct ext4_dir_entry_tail)));
301 while (d < top && d->rec_len)
302 d = (struct ext4_dir_entry *)(((void *)d) +
303 le16_to_cpu(d->rec_len));
304
305 if (d != top)
306 return NULL;
307
308 t = (struct ext4_dir_entry_tail *)d;
309 #else
310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
311 #endif
312
313 if (t->det_reserved_zero1 ||
314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 t->det_reserved_zero2 ||
316 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
317 return NULL;
318
319 return t;
320 }
321
322 static __le32 ext4_dirent_csum(struct inode *inode,
323 struct ext4_dir_entry *dirent, int size)
324 {
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct ext4_inode_info *ei = EXT4_I(inode);
327 __u32 csum;
328
329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 return cpu_to_le32(csum);
331 }
332
333 #define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
335
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
337 unsigned int line)
338 {
339 __ext4_warning_inode(inode, func, line,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
341 }
342
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
344 {
345 struct ext4_dir_entry_tail *t;
346
347 if (!ext4_has_metadata_csum(inode->i_sb))
348 return 1;
349
350 t = get_dirent_tail(inode, dirent);
351 if (!t) {
352 warn_no_space_for_csum(inode);
353 return 0;
354 }
355
356 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 (void *)t - (void *)dirent))
358 return 0;
359
360 return 1;
361 }
362
363 static void ext4_dirent_csum_set(struct inode *inode,
364 struct ext4_dir_entry *dirent)
365 {
366 struct ext4_dir_entry_tail *t;
367
368 if (!ext4_has_metadata_csum(inode->i_sb))
369 return;
370
371 t = get_dirent_tail(inode, dirent);
372 if (!t) {
373 warn_no_space_for_csum(inode);
374 return;
375 }
376
377 t->det_checksum = ext4_dirent_csum(inode, dirent,
378 (void *)t - (void *)dirent);
379 }
380
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
382 struct inode *inode,
383 struct buffer_head *bh)
384 {
385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 return ext4_handle_dirty_metadata(handle, inode, bh);
387 }
388
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 struct ext4_dir_entry *dirent,
391 int *offset)
392 {
393 struct ext4_dir_entry *dp;
394 struct dx_root_info *root;
395 int count_offset;
396
397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
398 count_offset = 8;
399 else if (le16_to_cpu(dirent->rec_len) == 12) {
400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 if (le16_to_cpu(dp->rec_len) !=
402 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
403 return NULL;
404 root = (struct dx_root_info *)(((void *)dp + 12));
405 if (root->reserved_zero ||
406 root->info_length != sizeof(struct dx_root_info))
407 return NULL;
408 count_offset = 32;
409 } else
410 return NULL;
411
412 if (offset)
413 *offset = count_offset;
414 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
415 }
416
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 int count_offset, int count, struct dx_tail *t)
419 {
420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 struct ext4_inode_info *ei = EXT4_I(inode);
422 __u32 csum;
423 int size;
424 __u32 dummy_csum = 0;
425 int offset = offsetof(struct dx_tail, dt_checksum);
426
427 size = count_offset + (count * sizeof(struct dx_entry));
428 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
429 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
430 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
431
432 return cpu_to_le32(csum);
433 }
434
435 static int ext4_dx_csum_verify(struct inode *inode,
436 struct ext4_dir_entry *dirent)
437 {
438 struct dx_countlimit *c;
439 struct dx_tail *t;
440 int count_offset, limit, count;
441
442 if (!ext4_has_metadata_csum(inode->i_sb))
443 return 1;
444
445 c = get_dx_countlimit(inode, dirent, &count_offset);
446 if (!c) {
447 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
448 return 0;
449 }
450 limit = le16_to_cpu(c->limit);
451 count = le16_to_cpu(c->count);
452 if (count_offset + (limit * sizeof(struct dx_entry)) >
453 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
454 warn_no_space_for_csum(inode);
455 return 0;
456 }
457 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
458
459 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
460 count, t))
461 return 0;
462 return 1;
463 }
464
465 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
466 {
467 struct dx_countlimit *c;
468 struct dx_tail *t;
469 int count_offset, limit, count;
470
471 if (!ext4_has_metadata_csum(inode->i_sb))
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)) {
483 warn_no_space_for_csum(inode);
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
499 /*
500 * p is at least 6 bytes before the end of page
501 */
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
504 {
505 return (struct ext4_dir_entry_2 *)((char *)p +
506 ext4_rec_len_from_disk(p->rec_len, blocksize));
507 }
508
509 /*
510 * Future: use high four bits of block for coalesce-on-delete flags
511 * Mask them off for now.
512 */
513
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
515 {
516 return le32_to_cpu(entry->block) & 0x00ffffff;
517 }
518
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
520 {
521 entry->block = cpu_to_le32(value);
522 }
523
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
525 {
526 return le32_to_cpu(entry->hash);
527 }
528
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
530 {
531 entry->hash = cpu_to_le32(value);
532 }
533
534 static inline unsigned dx_get_count(struct dx_entry *entries)
535 {
536 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537 }
538
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
540 {
541 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542 }
543
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
545 {
546 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547 }
548
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
550 {
551 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552 }
553
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
555 {
556 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 EXT4_DIR_REC_LEN(2) - infosize;
558
559 if (ext4_has_metadata_csum(dir->i_sb))
560 entry_space -= sizeof(struct dx_tail);
561 return entry_space / sizeof(struct dx_entry);
562 }
563
564 static inline unsigned dx_node_limit(struct inode *dir)
565 {
566 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
567
568 if (ext4_has_metadata_csum(dir->i_sb))
569 entry_space -= sizeof(struct dx_tail);
570 return entry_space / sizeof(struct dx_entry);
571 }
572
573 /*
574 * Debug
575 */
576 #ifdef DX_DEBUG
577 static void dx_show_index(char * label, struct dx_entry *entries)
578 {
579 int i, n = dx_get_count (entries);
580 printk(KERN_DEBUG "%s index", label);
581 for (i = 0; i < n; i++) {
582 printk(KERN_CONT " %x->%lu",
583 i ? dx_get_hash(entries + i) : 0,
584 (unsigned long)dx_get_block(entries + i));
585 }
586 printk(KERN_CONT "\n");
587 }
588
589 struct stats
590 {
591 unsigned names;
592 unsigned space;
593 unsigned bcount;
594 };
595
596 static struct stats dx_show_leaf(struct inode *dir,
597 struct dx_hash_info *hinfo,
598 struct ext4_dir_entry_2 *de,
599 int size, int show_names)
600 {
601 unsigned names = 0, space = 0;
602 char *base = (char *) de;
603 struct dx_hash_info h = *hinfo;
604
605 printk("names: ");
606 while ((char *) de < base + size)
607 {
608 if (de->inode)
609 {
610 if (show_names)
611 {
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
613 int len;
614 char *name;
615 struct fscrypt_str fname_crypto_str =
616 FSTR_INIT(NULL, 0);
617 int res = 0;
618
619 name = de->name;
620 len = de->name_len;
621 if (ext4_encrypted_inode(dir))
622 res = fscrypt_get_encryption_info(dir);
623 if (res) {
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
626 }
627 if (!fscrypt_has_encryption_key(dir)) {
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
630 de->name_len, &h);
631 printk("%*.s:(U)%x.%u ", len,
632 name, h.hash,
633 (unsigned) ((char *) de
634 - base));
635 } else {
636 struct fscrypt_str de_name =
637 FSTR_INIT(name, len);
638
639 /* Directory is encrypted */
640 res = fscrypt_fname_alloc_buffer(
641 dir, len,
642 &fname_crypto_str);
643 if (res)
644 printk(KERN_WARNING "Error "
645 "allocating crypto "
646 "buffer--skipping "
647 "crypto\n");
648 res = fscrypt_fname_disk_to_usr(dir,
649 0, 0, &de_name,
650 &fname_crypto_str);
651 if (res) {
652 printk(KERN_WARNING "Error "
653 "converting filename "
654 "from disk to usr"
655 "\n");
656 name = "??";
657 len = 2;
658 } else {
659 name = fname_crypto_str.name;
660 len = fname_crypto_str.len;
661 }
662 ext4fs_dirhash(de->name, de->name_len,
663 &h);
664 printk("%*.s:(E)%x.%u ", len, name,
665 h.hash, (unsigned) ((char *) de
666 - base));
667 fscrypt_fname_free_buffer(
668 &fname_crypto_str);
669 }
670 #else
671 int len = de->name_len;
672 char *name = de->name;
673 ext4fs_dirhash(de->name, de->name_len, &h);
674 printk("%*.s:%x.%u ", len, name, h.hash,
675 (unsigned) ((char *) de - base));
676 #endif
677 }
678 space += EXT4_DIR_REC_LEN(de->name_len);
679 names++;
680 }
681 de = ext4_next_entry(de, size);
682 }
683 printk(KERN_CONT "(%i)\n", names);
684 return (struct stats) { names, space, 1 };
685 }
686
687 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
688 struct dx_entry *entries, int levels)
689 {
690 unsigned blocksize = dir->i_sb->s_blocksize;
691 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
692 unsigned bcount = 0;
693 struct buffer_head *bh;
694 printk("%i indexed blocks...\n", count);
695 for (i = 0; i < count; i++, entries++)
696 {
697 ext4_lblk_t block = dx_get_block(entries);
698 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
699 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
700 struct stats stats;
701 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
702 bh = ext4_bread(NULL,dir, block, 0);
703 if (!bh || IS_ERR(bh))
704 continue;
705 stats = levels?
706 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
707 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
708 bh->b_data, blocksize, 0);
709 names += stats.names;
710 space += stats.space;
711 bcount += stats.bcount;
712 brelse(bh);
713 }
714 if (bcount)
715 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
716 levels ? "" : " ", names, space/bcount,
717 (space/bcount)*100/blocksize);
718 return (struct stats) { names, space, bcount};
719 }
720 #endif /* DX_DEBUG */
721
722 /*
723 * Probe for a directory leaf block to search.
724 *
725 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
726 * error in the directory index, and the caller should fall back to
727 * searching the directory normally. The callers of dx_probe **MUST**
728 * check for this error code, and make sure it never gets reflected
729 * back to userspace.
730 */
731 static struct dx_frame *
732 dx_probe(struct ext4_filename *fname, struct inode *dir,
733 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
734 {
735 unsigned count, indirect;
736 struct dx_entry *at, *entries, *p, *q, *m;
737 struct dx_root *root;
738 struct dx_frame *frame = frame_in;
739 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
740 u32 hash;
741
742 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
743 if (IS_ERR(frame->bh))
744 return (struct dx_frame *) frame->bh;
745
746 root = (struct dx_root *) frame->bh->b_data;
747 if (root->info.hash_version != DX_HASH_TEA &&
748 root->info.hash_version != DX_HASH_HALF_MD4 &&
749 root->info.hash_version != DX_HASH_LEGACY) {
750 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
751 root->info.hash_version);
752 goto fail;
753 }
754 if (fname)
755 hinfo = &fname->hinfo;
756 hinfo->hash_version = root->info.hash_version;
757 if (hinfo->hash_version <= DX_HASH_TEA)
758 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
759 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
760 if (fname && fname_name(fname))
761 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
762 hash = hinfo->hash;
763
764 if (root->info.unused_flags & 1) {
765 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
766 root->info.unused_flags);
767 goto fail;
768 }
769
770 indirect = root->info.indirect_levels;
771 if (indirect > 1) {
772 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
773 root->info.indirect_levels);
774 goto fail;
775 }
776
777 entries = (struct dx_entry *)(((char *)&root->info) +
778 root->info.info_length);
779
780 if (dx_get_limit(entries) != dx_root_limit(dir,
781 root->info.info_length)) {
782 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
783 dx_get_limit(entries),
784 dx_root_limit(dir, root->info.info_length));
785 goto fail;
786 }
787
788 dxtrace(printk("Look up %x", hash));
789 while (1) {
790 count = dx_get_count(entries);
791 if (!count || count > dx_get_limit(entries)) {
792 ext4_warning_inode(dir,
793 "dx entry: count %u beyond limit %u",
794 count, dx_get_limit(entries));
795 goto fail;
796 }
797
798 p = entries + 1;
799 q = entries + count - 1;
800 while (p <= q) {
801 m = p + (q - p) / 2;
802 dxtrace(printk(KERN_CONT "."));
803 if (dx_get_hash(m) > hash)
804 q = m - 1;
805 else
806 p = m + 1;
807 }
808
809 if (0) { // linear search cross check
810 unsigned n = count - 1;
811 at = entries;
812 while (n--)
813 {
814 dxtrace(printk(KERN_CONT ","));
815 if (dx_get_hash(++at) > hash)
816 {
817 at--;
818 break;
819 }
820 }
821 assert (at == p - 1);
822 }
823
824 at = p - 1;
825 dxtrace(printk(KERN_CONT " %x->%u\n",
826 at == entries ? 0 : dx_get_hash(at),
827 dx_get_block(at)));
828 frame->entries = entries;
829 frame->at = at;
830 if (!indirect--)
831 return frame;
832 frame++;
833 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
834 if (IS_ERR(frame->bh)) {
835 ret_err = (struct dx_frame *) frame->bh;
836 frame->bh = NULL;
837 goto fail;
838 }
839 entries = ((struct dx_node *) frame->bh->b_data)->entries;
840
841 if (dx_get_limit(entries) != dx_node_limit(dir)) {
842 ext4_warning_inode(dir,
843 "dx entry: limit %u != node limit %u",
844 dx_get_limit(entries), dx_node_limit(dir));
845 goto fail;
846 }
847 }
848 fail:
849 while (frame >= frame_in) {
850 brelse(frame->bh);
851 frame--;
852 }
853
854 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
855 ext4_warning_inode(dir,
856 "Corrupt directory, running e2fsck is recommended");
857 return ret_err;
858 }
859
860 static void dx_release(struct dx_frame *frames)
861 {
862 if (frames[0].bh == NULL)
863 return;
864
865 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
866 brelse(frames[1].bh);
867 brelse(frames[0].bh);
868 }
869
870 /*
871 * This function increments the frame pointer to search the next leaf
872 * block, and reads in the necessary intervening nodes if the search
873 * should be necessary. Whether or not the search is necessary is
874 * controlled by the hash parameter. If the hash value is even, then
875 * the search is only continued if the next block starts with that
876 * hash value. This is used if we are searching for a specific file.
877 *
878 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
879 *
880 * This function returns 1 if the caller should continue to search,
881 * or 0 if it should not. If there is an error reading one of the
882 * index blocks, it will a negative error code.
883 *
884 * If start_hash is non-null, it will be filled in with the starting
885 * hash of the next page.
886 */
887 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
888 struct dx_frame *frame,
889 struct dx_frame *frames,
890 __u32 *start_hash)
891 {
892 struct dx_frame *p;
893 struct buffer_head *bh;
894 int num_frames = 0;
895 __u32 bhash;
896
897 p = frame;
898 /*
899 * Find the next leaf page by incrementing the frame pointer.
900 * If we run out of entries in the interior node, loop around and
901 * increment pointer in the parent node. When we break out of
902 * this loop, num_frames indicates the number of interior
903 * nodes need to be read.
904 */
905 while (1) {
906 if (++(p->at) < p->entries + dx_get_count(p->entries))
907 break;
908 if (p == frames)
909 return 0;
910 num_frames++;
911 p--;
912 }
913
914 /*
915 * If the hash is 1, then continue only if the next page has a
916 * continuation hash of any value. This is used for readdir
917 * handling. Otherwise, check to see if the hash matches the
918 * desired contiuation hash. If it doesn't, return since
919 * there's no point to read in the successive index pages.
920 */
921 bhash = dx_get_hash(p->at);
922 if (start_hash)
923 *start_hash = bhash;
924 if ((hash & 1) == 0) {
925 if ((bhash & ~1) != hash)
926 return 0;
927 }
928 /*
929 * If the hash is HASH_NB_ALWAYS, we always go to the next
930 * block so no check is necessary
931 */
932 while (num_frames--) {
933 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
934 if (IS_ERR(bh))
935 return PTR_ERR(bh);
936 p++;
937 brelse(p->bh);
938 p->bh = bh;
939 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
940 }
941 return 1;
942 }
943
944
945 /*
946 * This function fills a red-black tree with information from a
947 * directory block. It returns the number directory entries loaded
948 * into the tree. If there is an error it is returned in err.
949 */
950 static int htree_dirblock_to_tree(struct file *dir_file,
951 struct inode *dir, ext4_lblk_t block,
952 struct dx_hash_info *hinfo,
953 __u32 start_hash, __u32 start_minor_hash)
954 {
955 struct buffer_head *bh;
956 struct ext4_dir_entry_2 *de, *top;
957 int err = 0, count = 0;
958 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
959
960 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
961 (unsigned long)block));
962 bh = ext4_read_dirblock(dir, block, DIRENT);
963 if (IS_ERR(bh))
964 return PTR_ERR(bh);
965
966 de = (struct ext4_dir_entry_2 *) bh->b_data;
967 top = (struct ext4_dir_entry_2 *) ((char *) de +
968 dir->i_sb->s_blocksize -
969 EXT4_DIR_REC_LEN(0));
970 #ifdef CONFIG_EXT4_FS_ENCRYPTION
971 /* Check if the directory is encrypted */
972 if (ext4_encrypted_inode(dir)) {
973 err = fscrypt_get_encryption_info(dir);
974 if (err < 0) {
975 brelse(bh);
976 return err;
977 }
978 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
979 &fname_crypto_str);
980 if (err < 0) {
981 brelse(bh);
982 return err;
983 }
984 }
985 #endif
986 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
987 if (ext4_check_dir_entry(dir, NULL, de, bh,
988 bh->b_data, bh->b_size,
989 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
990 + ((char *)de - bh->b_data))) {
991 /* silently ignore the rest of the block */
992 break;
993 }
994 ext4fs_dirhash(de->name, de->name_len, hinfo);
995 if ((hinfo->hash < start_hash) ||
996 ((hinfo->hash == start_hash) &&
997 (hinfo->minor_hash < start_minor_hash)))
998 continue;
999 if (de->inode == 0)
1000 continue;
1001 if (!ext4_encrypted_inode(dir)) {
1002 tmp_str.name = de->name;
1003 tmp_str.len = de->name_len;
1004 err = ext4_htree_store_dirent(dir_file,
1005 hinfo->hash, hinfo->minor_hash, de,
1006 &tmp_str);
1007 } else {
1008 int save_len = fname_crypto_str.len;
1009 struct fscrypt_str de_name = FSTR_INIT(de->name,
1010 de->name_len);
1011
1012 /* Directory is encrypted */
1013 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1014 hinfo->minor_hash, &de_name,
1015 &fname_crypto_str);
1016 if (err) {
1017 count = err;
1018 goto errout;
1019 }
1020 err = ext4_htree_store_dirent(dir_file,
1021 hinfo->hash, hinfo->minor_hash, de,
1022 &fname_crypto_str);
1023 fname_crypto_str.len = save_len;
1024 }
1025 if (err != 0) {
1026 count = err;
1027 goto errout;
1028 }
1029 count++;
1030 }
1031 errout:
1032 brelse(bh);
1033 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1034 fscrypt_fname_free_buffer(&fname_crypto_str);
1035 #endif
1036 return count;
1037 }
1038
1039
1040 /*
1041 * This function fills a red-black tree with information from a
1042 * directory. We start scanning the directory in hash order, starting
1043 * at start_hash and start_minor_hash.
1044 *
1045 * This function returns the number of entries inserted into the tree,
1046 * or a negative error code.
1047 */
1048 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1049 __u32 start_minor_hash, __u32 *next_hash)
1050 {
1051 struct dx_hash_info hinfo;
1052 struct ext4_dir_entry_2 *de;
1053 struct dx_frame frames[2], *frame;
1054 struct inode *dir;
1055 ext4_lblk_t block;
1056 int count = 0;
1057 int ret, err;
1058 __u32 hashval;
1059 struct fscrypt_str tmp_str;
1060
1061 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1062 start_hash, start_minor_hash));
1063 dir = file_inode(dir_file);
1064 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1065 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1066 if (hinfo.hash_version <= DX_HASH_TEA)
1067 hinfo.hash_version +=
1068 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1069 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1070 if (ext4_has_inline_data(dir)) {
1071 int has_inline_data = 1;
1072 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1073 &hinfo, start_hash,
1074 start_minor_hash,
1075 &has_inline_data);
1076 if (has_inline_data) {
1077 *next_hash = ~0;
1078 return count;
1079 }
1080 }
1081 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1082 start_hash, start_minor_hash);
1083 *next_hash = ~0;
1084 return count;
1085 }
1086 hinfo.hash = start_hash;
1087 hinfo.minor_hash = 0;
1088 frame = dx_probe(NULL, dir, &hinfo, frames);
1089 if (IS_ERR(frame))
1090 return PTR_ERR(frame);
1091
1092 /* Add '.' and '..' from the htree header */
1093 if (!start_hash && !start_minor_hash) {
1094 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1095 tmp_str.name = de->name;
1096 tmp_str.len = de->name_len;
1097 err = ext4_htree_store_dirent(dir_file, 0, 0,
1098 de, &tmp_str);
1099 if (err != 0)
1100 goto errout;
1101 count++;
1102 }
1103 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1104 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1105 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1106 tmp_str.name = de->name;
1107 tmp_str.len = de->name_len;
1108 err = ext4_htree_store_dirent(dir_file, 2, 0,
1109 de, &tmp_str);
1110 if (err != 0)
1111 goto errout;
1112 count++;
1113 }
1114
1115 while (1) {
1116 if (fatal_signal_pending(current)) {
1117 err = -ERESTARTSYS;
1118 goto errout;
1119 }
1120 cond_resched();
1121 block = dx_get_block(frame->at);
1122 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1123 start_hash, start_minor_hash);
1124 if (ret < 0) {
1125 err = ret;
1126 goto errout;
1127 }
1128 count += ret;
1129 hashval = ~0;
1130 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1131 frame, frames, &hashval);
1132 *next_hash = hashval;
1133 if (ret < 0) {
1134 err = ret;
1135 goto errout;
1136 }
1137 /*
1138 * Stop if: (a) there are no more entries, or
1139 * (b) we have inserted at least one entry and the
1140 * next hash value is not a continuation
1141 */
1142 if ((ret == 0) ||
1143 (count && ((hashval & 1) == 0)))
1144 break;
1145 }
1146 dx_release(frames);
1147 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1148 "next hash: %x\n", count, *next_hash));
1149 return count;
1150 errout:
1151 dx_release(frames);
1152 return (err);
1153 }
1154
1155 static inline int search_dirblock(struct buffer_head *bh,
1156 struct inode *dir,
1157 struct ext4_filename *fname,
1158 const struct qstr *d_name,
1159 unsigned int offset,
1160 struct ext4_dir_entry_2 **res_dir)
1161 {
1162 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1163 fname, d_name, offset, res_dir);
1164 }
1165
1166 /*
1167 * Directory block splitting, compacting
1168 */
1169
1170 /*
1171 * Create map of hash values, offsets, and sizes, stored at end of block.
1172 * Returns number of entries mapped.
1173 */
1174 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1175 unsigned blocksize, struct dx_hash_info *hinfo,
1176 struct dx_map_entry *map_tail)
1177 {
1178 int count = 0;
1179 char *base = (char *) de;
1180 struct dx_hash_info h = *hinfo;
1181
1182 while ((char *) de < base + blocksize) {
1183 if (de->name_len && de->inode) {
1184 ext4fs_dirhash(de->name, de->name_len, &h);
1185 map_tail--;
1186 map_tail->hash = h.hash;
1187 map_tail->offs = ((char *) de - base)>>2;
1188 map_tail->size = le16_to_cpu(de->rec_len);
1189 count++;
1190 cond_resched();
1191 }
1192 /* XXX: do we need to check rec_len == 0 case? -Chris */
1193 de = ext4_next_entry(de, blocksize);
1194 }
1195 return count;
1196 }
1197
1198 /* Sort map by hash value */
1199 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1200 {
1201 struct dx_map_entry *p, *q, *top = map + count - 1;
1202 int more;
1203 /* Combsort until bubble sort doesn't suck */
1204 while (count > 2) {
1205 count = count*10/13;
1206 if (count - 9 < 2) /* 9, 10 -> 11 */
1207 count = 11;
1208 for (p = top, q = p - count; q >= map; p--, q--)
1209 if (p->hash < q->hash)
1210 swap(*p, *q);
1211 }
1212 /* Garden variety bubble sort */
1213 do {
1214 more = 0;
1215 q = top;
1216 while (q-- > map) {
1217 if (q[1].hash >= q[0].hash)
1218 continue;
1219 swap(*(q+1), *q);
1220 more = 1;
1221 }
1222 } while(more);
1223 }
1224
1225 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1226 {
1227 struct dx_entry *entries = frame->entries;
1228 struct dx_entry *old = frame->at, *new = old + 1;
1229 int count = dx_get_count(entries);
1230
1231 assert(count < dx_get_limit(entries));
1232 assert(old < entries + count);
1233 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1234 dx_set_hash(new, hash);
1235 dx_set_block(new, block);
1236 dx_set_count(entries, count + 1);
1237 }
1238
1239 /*
1240 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1241 *
1242 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1243 * `de != NULL' is guaranteed by caller.
1244 */
1245 static inline int ext4_match(struct ext4_filename *fname,
1246 struct ext4_dir_entry_2 *de)
1247 {
1248 const void *name = fname_name(fname);
1249 u32 len = fname_len(fname);
1250
1251 if (!de->inode)
1252 return 0;
1253
1254 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1255 if (unlikely(!name)) {
1256 if (fname->usr_fname->name[0] == '_') {
1257 int ret;
1258 if (de->name_len <= 32)
1259 return 0;
1260 ret = memcmp(de->name + ((de->name_len - 17) & ~15),
1261 fname->crypto_buf.name + 8, 16);
1262 return (ret == 0) ? 1 : 0;
1263 }
1264 name = fname->crypto_buf.name;
1265 len = fname->crypto_buf.len;
1266 }
1267 #endif
1268 if (de->name_len != len)
1269 return 0;
1270 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1271 }
1272
1273 /*
1274 * Returns 0 if not found, -1 on failure, and 1 on success
1275 */
1276 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1277 struct inode *dir, struct ext4_filename *fname,
1278 const struct qstr *d_name,
1279 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1280 {
1281 struct ext4_dir_entry_2 * de;
1282 char * dlimit;
1283 int de_len;
1284 int res;
1285
1286 de = (struct ext4_dir_entry_2 *)search_buf;
1287 dlimit = search_buf + buf_size;
1288 while ((char *) de < dlimit) {
1289 /* this code is executed quadratically often */
1290 /* do minimal checking `by hand' */
1291 if ((char *) de + de->name_len <= dlimit) {
1292 res = ext4_match(fname, de);
1293 if (res < 0) {
1294 res = -1;
1295 goto return_result;
1296 }
1297 if (res > 0) {
1298 /* found a match - just to be sure, do
1299 * a full check */
1300 if (ext4_check_dir_entry(dir, NULL, de, bh,
1301 bh->b_data,
1302 bh->b_size, offset)) {
1303 res = -1;
1304 goto return_result;
1305 }
1306 *res_dir = de;
1307 res = 1;
1308 goto return_result;
1309 }
1310
1311 }
1312 /* prevent looping on a bad block */
1313 de_len = ext4_rec_len_from_disk(de->rec_len,
1314 dir->i_sb->s_blocksize);
1315 if (de_len <= 0) {
1316 res = -1;
1317 goto return_result;
1318 }
1319 offset += de_len;
1320 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1321 }
1322
1323 res = 0;
1324 return_result:
1325 return res;
1326 }
1327
1328 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1329 struct ext4_dir_entry *de)
1330 {
1331 struct super_block *sb = dir->i_sb;
1332
1333 if (!is_dx(dir))
1334 return 0;
1335 if (block == 0)
1336 return 1;
1337 if (de->inode == 0 &&
1338 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1339 sb->s_blocksize)
1340 return 1;
1341 return 0;
1342 }
1343
1344 /*
1345 * ext4_find_entry()
1346 *
1347 * finds an entry in the specified directory with the wanted name. It
1348 * returns the cache buffer in which the entry was found, and the entry
1349 * itself (as a parameter - res_dir). It does NOT read the inode of the
1350 * entry - you'll have to do that yourself if you want to.
1351 *
1352 * The returned buffer_head has ->b_count elevated. The caller is expected
1353 * to brelse() it when appropriate.
1354 */
1355 static struct buffer_head * ext4_find_entry (struct inode *dir,
1356 const struct qstr *d_name,
1357 struct ext4_dir_entry_2 **res_dir,
1358 int *inlined)
1359 {
1360 struct super_block *sb;
1361 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1362 struct buffer_head *bh, *ret = NULL;
1363 ext4_lblk_t start, block, b;
1364 const u8 *name = d_name->name;
1365 int ra_max = 0; /* Number of bh's in the readahead
1366 buffer, bh_use[] */
1367 int ra_ptr = 0; /* Current index into readahead
1368 buffer */
1369 int num = 0;
1370 ext4_lblk_t nblocks;
1371 int i, namelen, retval;
1372 struct ext4_filename fname;
1373
1374 *res_dir = NULL;
1375 sb = dir->i_sb;
1376 namelen = d_name->len;
1377 if (namelen > EXT4_NAME_LEN)
1378 return NULL;
1379
1380 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1381 if (retval)
1382 return ERR_PTR(retval);
1383
1384 if (ext4_has_inline_data(dir)) {
1385 int has_inline_data = 1;
1386 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1387 &has_inline_data);
1388 if (has_inline_data) {
1389 if (inlined)
1390 *inlined = 1;
1391 goto cleanup_and_exit;
1392 }
1393 }
1394
1395 if ((namelen <= 2) && (name[0] == '.') &&
1396 (name[1] == '.' || name[1] == '\0')) {
1397 /*
1398 * "." or ".." will only be in the first block
1399 * NFS may look up ".."; "." should be handled by the VFS
1400 */
1401 block = start = 0;
1402 nblocks = 1;
1403 goto restart;
1404 }
1405 if (is_dx(dir)) {
1406 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1407 /*
1408 * On success, or if the error was file not found,
1409 * return. Otherwise, fall back to doing a search the
1410 * old fashioned way.
1411 */
1412 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1413 goto cleanup_and_exit;
1414 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1415 "falling back\n"));
1416 }
1417 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1418 start = EXT4_I(dir)->i_dir_start_lookup;
1419 if (start >= nblocks)
1420 start = 0;
1421 block = start;
1422 restart:
1423 do {
1424 /*
1425 * We deal with the read-ahead logic here.
1426 */
1427 if (ra_ptr >= ra_max) {
1428 /* Refill the readahead buffer */
1429 ra_ptr = 0;
1430 b = block;
1431 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1432 /*
1433 * Terminate if we reach the end of the
1434 * directory and must wrap, or if our
1435 * search has finished at this block.
1436 */
1437 if (b >= nblocks || (num && block == start)) {
1438 bh_use[ra_max] = NULL;
1439 break;
1440 }
1441 num++;
1442 bh = ext4_getblk(NULL, dir, b++, 0);
1443 if (IS_ERR(bh)) {
1444 if (ra_max == 0) {
1445 ret = bh;
1446 goto cleanup_and_exit;
1447 }
1448 break;
1449 }
1450 bh_use[ra_max] = bh;
1451 if (bh)
1452 ll_rw_block(REQ_OP_READ,
1453 REQ_META | REQ_PRIO,
1454 1, &bh);
1455 }
1456 }
1457 if ((bh = bh_use[ra_ptr++]) == NULL)
1458 goto next;
1459 wait_on_buffer(bh);
1460 if (!buffer_uptodate(bh)) {
1461 /* read error, skip block & hope for the best */
1462 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1463 (unsigned long) block);
1464 brelse(bh);
1465 goto next;
1466 }
1467 if (!buffer_verified(bh) &&
1468 !is_dx_internal_node(dir, block,
1469 (struct ext4_dir_entry *)bh->b_data) &&
1470 !ext4_dirent_csum_verify(dir,
1471 (struct ext4_dir_entry *)bh->b_data)) {
1472 EXT4_ERROR_INODE(dir, "checksumming directory "
1473 "block %lu", (unsigned long)block);
1474 brelse(bh);
1475 goto next;
1476 }
1477 set_buffer_verified(bh);
1478 i = search_dirblock(bh, dir, &fname, d_name,
1479 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1480 if (i == 1) {
1481 EXT4_I(dir)->i_dir_start_lookup = block;
1482 ret = bh;
1483 goto cleanup_and_exit;
1484 } else {
1485 brelse(bh);
1486 if (i < 0)
1487 goto cleanup_and_exit;
1488 }
1489 next:
1490 if (++block >= nblocks)
1491 block = 0;
1492 } while (block != start);
1493
1494 /*
1495 * If the directory has grown while we were searching, then
1496 * search the last part of the directory before giving up.
1497 */
1498 block = nblocks;
1499 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1500 if (block < nblocks) {
1501 start = 0;
1502 goto restart;
1503 }
1504
1505 cleanup_and_exit:
1506 /* Clean up the read-ahead blocks */
1507 for (; ra_ptr < ra_max; ra_ptr++)
1508 brelse(bh_use[ra_ptr]);
1509 ext4_fname_free_filename(&fname);
1510 return ret;
1511 }
1512
1513 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1514 struct ext4_filename *fname,
1515 struct ext4_dir_entry_2 **res_dir)
1516 {
1517 struct super_block * sb = dir->i_sb;
1518 struct dx_frame frames[2], *frame;
1519 const struct qstr *d_name = fname->usr_fname;
1520 struct buffer_head *bh;
1521 ext4_lblk_t block;
1522 int retval;
1523
1524 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1525 *res_dir = NULL;
1526 #endif
1527 frame = dx_probe(fname, dir, NULL, frames);
1528 if (IS_ERR(frame))
1529 return (struct buffer_head *) frame;
1530 do {
1531 block = dx_get_block(frame->at);
1532 bh = ext4_read_dirblock(dir, block, DIRENT);
1533 if (IS_ERR(bh))
1534 goto errout;
1535
1536 retval = search_dirblock(bh, dir, fname, d_name,
1537 block << EXT4_BLOCK_SIZE_BITS(sb),
1538 res_dir);
1539 if (retval == 1)
1540 goto success;
1541 brelse(bh);
1542 if (retval == -1) {
1543 bh = ERR_PTR(ERR_BAD_DX_DIR);
1544 goto errout;
1545 }
1546
1547 /* Check to see if we should continue to search */
1548 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1549 frames, NULL);
1550 if (retval < 0) {
1551 ext4_warning_inode(dir,
1552 "error %d reading directory index block",
1553 retval);
1554 bh = ERR_PTR(retval);
1555 goto errout;
1556 }
1557 } while (retval == 1);
1558
1559 bh = NULL;
1560 errout:
1561 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1562 success:
1563 dx_release(frames);
1564 return bh;
1565 }
1566
1567 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1568 {
1569 struct inode *inode;
1570 struct ext4_dir_entry_2 *de;
1571 struct buffer_head *bh;
1572
1573 if (ext4_encrypted_inode(dir)) {
1574 int res = fscrypt_get_encryption_info(dir);
1575
1576 /*
1577 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1578 * created while the directory was encrypted and we
1579 * have access to the key.
1580 */
1581 if (fscrypt_has_encryption_key(dir))
1582 fscrypt_set_encrypted_dentry(dentry);
1583 fscrypt_set_d_op(dentry);
1584 if (res && res != -ENOKEY)
1585 return ERR_PTR(res);
1586 }
1587
1588 if (dentry->d_name.len > EXT4_NAME_LEN)
1589 return ERR_PTR(-ENAMETOOLONG);
1590
1591 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1592 if (IS_ERR(bh))
1593 return (struct dentry *) bh;
1594 inode = NULL;
1595 if (bh) {
1596 __u32 ino = le32_to_cpu(de->inode);
1597 brelse(bh);
1598 if (!ext4_valid_inum(dir->i_sb, ino)) {
1599 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1600 return ERR_PTR(-EFSCORRUPTED);
1601 }
1602 if (unlikely(ino == dir->i_ino)) {
1603 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1604 dentry);
1605 return ERR_PTR(-EFSCORRUPTED);
1606 }
1607 inode = ext4_iget_normal(dir->i_sb, ino);
1608 if (inode == ERR_PTR(-ESTALE)) {
1609 EXT4_ERROR_INODE(dir,
1610 "deleted inode referenced: %u",
1611 ino);
1612 return ERR_PTR(-EFSCORRUPTED);
1613 }
1614 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1615 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1616 !fscrypt_has_permitted_context(dir, inode)) {
1617 int nokey = ext4_encrypted_inode(inode) &&
1618 !fscrypt_has_encryption_key(inode);
1619 if (nokey) {
1620 iput(inode);
1621 return ERR_PTR(-ENOKEY);
1622 }
1623 ext4_warning(inode->i_sb,
1624 "Inconsistent encryption contexts: %lu/%lu",
1625 (unsigned long) dir->i_ino,
1626 (unsigned long) inode->i_ino);
1627 iput(inode);
1628 return ERR_PTR(-EPERM);
1629 }
1630 }
1631 return d_splice_alias(inode, dentry);
1632 }
1633
1634
1635 struct dentry *ext4_get_parent(struct dentry *child)
1636 {
1637 __u32 ino;
1638 static const struct qstr dotdot = QSTR_INIT("..", 2);
1639 struct ext4_dir_entry_2 * de;
1640 struct buffer_head *bh;
1641
1642 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1643 if (IS_ERR(bh))
1644 return (struct dentry *) bh;
1645 if (!bh)
1646 return ERR_PTR(-ENOENT);
1647 ino = le32_to_cpu(de->inode);
1648 brelse(bh);
1649
1650 if (!ext4_valid_inum(child->d_sb, ino)) {
1651 EXT4_ERROR_INODE(d_inode(child),
1652 "bad parent inode number: %u", ino);
1653 return ERR_PTR(-EFSCORRUPTED);
1654 }
1655
1656 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1657 }
1658
1659 /*
1660 * Move count entries from end of map between two memory locations.
1661 * Returns pointer to last entry moved.
1662 */
1663 static struct ext4_dir_entry_2 *
1664 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1665 unsigned blocksize)
1666 {
1667 unsigned rec_len = 0;
1668
1669 while (count--) {
1670 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1671 (from + (map->offs<<2));
1672 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1673 memcpy (to, de, rec_len);
1674 ((struct ext4_dir_entry_2 *) to)->rec_len =
1675 ext4_rec_len_to_disk(rec_len, blocksize);
1676 de->inode = 0;
1677 map++;
1678 to += rec_len;
1679 }
1680 return (struct ext4_dir_entry_2 *) (to - rec_len);
1681 }
1682
1683 /*
1684 * Compact each dir entry in the range to the minimal rec_len.
1685 * Returns pointer to last entry in range.
1686 */
1687 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1688 {
1689 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1690 unsigned rec_len = 0;
1691
1692 prev = to = de;
1693 while ((char*)de < base + blocksize) {
1694 next = ext4_next_entry(de, blocksize);
1695 if (de->inode && de->name_len) {
1696 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1697 if (de > to)
1698 memmove(to, de, rec_len);
1699 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1700 prev = to;
1701 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1702 }
1703 de = next;
1704 }
1705 return prev;
1706 }
1707
1708 /*
1709 * Split a full leaf block to make room for a new dir entry.
1710 * Allocate a new block, and move entries so that they are approx. equally full.
1711 * Returns pointer to de in block into which the new entry will be inserted.
1712 */
1713 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1714 struct buffer_head **bh,struct dx_frame *frame,
1715 struct dx_hash_info *hinfo)
1716 {
1717 unsigned blocksize = dir->i_sb->s_blocksize;
1718 unsigned count, continued;
1719 struct buffer_head *bh2;
1720 ext4_lblk_t newblock;
1721 u32 hash2;
1722 struct dx_map_entry *map;
1723 char *data1 = (*bh)->b_data, *data2;
1724 unsigned split, move, size;
1725 struct ext4_dir_entry_2 *de = NULL, *de2;
1726 struct ext4_dir_entry_tail *t;
1727 int csum_size = 0;
1728 int err = 0, i;
1729
1730 if (ext4_has_metadata_csum(dir->i_sb))
1731 csum_size = sizeof(struct ext4_dir_entry_tail);
1732
1733 bh2 = ext4_append(handle, dir, &newblock);
1734 if (IS_ERR(bh2)) {
1735 brelse(*bh);
1736 *bh = NULL;
1737 return (struct ext4_dir_entry_2 *) bh2;
1738 }
1739
1740 BUFFER_TRACE(*bh, "get_write_access");
1741 err = ext4_journal_get_write_access(handle, *bh);
1742 if (err)
1743 goto journal_error;
1744
1745 BUFFER_TRACE(frame->bh, "get_write_access");
1746 err = ext4_journal_get_write_access(handle, frame->bh);
1747 if (err)
1748 goto journal_error;
1749
1750 data2 = bh2->b_data;
1751
1752 /* create map in the end of data2 block */
1753 map = (struct dx_map_entry *) (data2 + blocksize);
1754 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1755 blocksize, hinfo, map);
1756 map -= count;
1757 dx_sort_map(map, count);
1758 /* Split the existing block in the middle, size-wise */
1759 size = 0;
1760 move = 0;
1761 for (i = count-1; i >= 0; i--) {
1762 /* is more than half of this entry in 2nd half of the block? */
1763 if (size + map[i].size/2 > blocksize/2)
1764 break;
1765 size += map[i].size;
1766 move++;
1767 }
1768 /* map index at which we will split */
1769 split = count - move;
1770 hash2 = map[split].hash;
1771 continued = hash2 == map[split - 1].hash;
1772 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1773 (unsigned long)dx_get_block(frame->at),
1774 hash2, split, count-split));
1775
1776 /* Fancy dance to stay within two buffers */
1777 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1778 blocksize);
1779 de = dx_pack_dirents(data1, blocksize);
1780 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1781 (char *) de,
1782 blocksize);
1783 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1784 (char *) de2,
1785 blocksize);
1786 if (csum_size) {
1787 t = EXT4_DIRENT_TAIL(data2, blocksize);
1788 initialize_dirent_tail(t, blocksize);
1789
1790 t = EXT4_DIRENT_TAIL(data1, blocksize);
1791 initialize_dirent_tail(t, blocksize);
1792 }
1793
1794 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1795 blocksize, 1));
1796 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1797 blocksize, 1));
1798
1799 /* Which block gets the new entry? */
1800 if (hinfo->hash >= hash2) {
1801 swap(*bh, bh2);
1802 de = de2;
1803 }
1804 dx_insert_block(frame, hash2 + continued, newblock);
1805 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1806 if (err)
1807 goto journal_error;
1808 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1809 if (err)
1810 goto journal_error;
1811 brelse(bh2);
1812 dxtrace(dx_show_index("frame", frame->entries));
1813 return de;
1814
1815 journal_error:
1816 brelse(*bh);
1817 brelse(bh2);
1818 *bh = NULL;
1819 ext4_std_error(dir->i_sb, err);
1820 return ERR_PTR(err);
1821 }
1822
1823 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1824 struct buffer_head *bh,
1825 void *buf, int buf_size,
1826 struct ext4_filename *fname,
1827 struct ext4_dir_entry_2 **dest_de)
1828 {
1829 struct ext4_dir_entry_2 *de;
1830 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1831 int nlen, rlen;
1832 unsigned int offset = 0;
1833 char *top;
1834 int res;
1835
1836 de = (struct ext4_dir_entry_2 *)buf;
1837 top = buf + buf_size - reclen;
1838 while ((char *) de <= top) {
1839 if (ext4_check_dir_entry(dir, NULL, de, bh,
1840 buf, buf_size, offset)) {
1841 res = -EFSCORRUPTED;
1842 goto return_result;
1843 }
1844 /* Provide crypto context and crypto buffer to ext4 match */
1845 res = ext4_match(fname, de);
1846 if (res < 0)
1847 goto return_result;
1848 if (res > 0) {
1849 res = -EEXIST;
1850 goto return_result;
1851 }
1852 nlen = EXT4_DIR_REC_LEN(de->name_len);
1853 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1854 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1855 break;
1856 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1857 offset += rlen;
1858 }
1859
1860 if ((char *) de > top)
1861 res = -ENOSPC;
1862 else {
1863 *dest_de = de;
1864 res = 0;
1865 }
1866 return_result:
1867 return res;
1868 }
1869
1870 int ext4_insert_dentry(struct inode *dir,
1871 struct inode *inode,
1872 struct ext4_dir_entry_2 *de,
1873 int buf_size,
1874 struct ext4_filename *fname)
1875 {
1876
1877 int nlen, rlen;
1878
1879 nlen = EXT4_DIR_REC_LEN(de->name_len);
1880 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1881 if (de->inode) {
1882 struct ext4_dir_entry_2 *de1 =
1883 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1884 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1885 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1886 de = de1;
1887 }
1888 de->file_type = EXT4_FT_UNKNOWN;
1889 de->inode = cpu_to_le32(inode->i_ino);
1890 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1891 de->name_len = fname_len(fname);
1892 memcpy(de->name, fname_name(fname), fname_len(fname));
1893 return 0;
1894 }
1895
1896 /*
1897 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1898 * it points to a directory entry which is guaranteed to be large
1899 * enough for new directory entry. If de is NULL, then
1900 * add_dirent_to_buf will attempt search the directory block for
1901 * space. It will return -ENOSPC if no space is available, and -EIO
1902 * and -EEXIST if directory entry already exists.
1903 */
1904 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1905 struct inode *dir,
1906 struct inode *inode, struct ext4_dir_entry_2 *de,
1907 struct buffer_head *bh)
1908 {
1909 unsigned int blocksize = dir->i_sb->s_blocksize;
1910 int csum_size = 0;
1911 int err;
1912
1913 if (ext4_has_metadata_csum(inode->i_sb))
1914 csum_size = sizeof(struct ext4_dir_entry_tail);
1915
1916 if (!de) {
1917 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1918 blocksize - csum_size, fname, &de);
1919 if (err)
1920 return err;
1921 }
1922 BUFFER_TRACE(bh, "get_write_access");
1923 err = ext4_journal_get_write_access(handle, bh);
1924 if (err) {
1925 ext4_std_error(dir->i_sb, err);
1926 return err;
1927 }
1928
1929 /* By now the buffer is marked for journaling. Due to crypto operations,
1930 * the following function call may fail */
1931 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1932 if (err < 0)
1933 return err;
1934
1935 /*
1936 * XXX shouldn't update any times until successful
1937 * completion of syscall, but too many callers depend
1938 * on this.
1939 *
1940 * XXX similarly, too many callers depend on
1941 * ext4_new_inode() setting the times, but error
1942 * recovery deletes the inode, so the worst that can
1943 * happen is that the times are slightly out of date
1944 * and/or different from the directory change time.
1945 */
1946 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1947 ext4_update_dx_flag(dir);
1948 dir->i_version++;
1949 ext4_mark_inode_dirty(handle, dir);
1950 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1951 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1952 if (err)
1953 ext4_std_error(dir->i_sb, err);
1954 return 0;
1955 }
1956
1957 /*
1958 * This converts a one block unindexed directory to a 3 block indexed
1959 * directory, and adds the dentry to the indexed directory.
1960 */
1961 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1962 struct inode *dir,
1963 struct inode *inode, struct buffer_head *bh)
1964 {
1965 struct buffer_head *bh2;
1966 struct dx_root *root;
1967 struct dx_frame frames[2], *frame;
1968 struct dx_entry *entries;
1969 struct ext4_dir_entry_2 *de, *de2;
1970 struct ext4_dir_entry_tail *t;
1971 char *data1, *top;
1972 unsigned len;
1973 int retval;
1974 unsigned blocksize;
1975 ext4_lblk_t block;
1976 struct fake_dirent *fde;
1977 int csum_size = 0;
1978
1979 if (ext4_has_metadata_csum(inode->i_sb))
1980 csum_size = sizeof(struct ext4_dir_entry_tail);
1981
1982 blocksize = dir->i_sb->s_blocksize;
1983 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1984 BUFFER_TRACE(bh, "get_write_access");
1985 retval = ext4_journal_get_write_access(handle, bh);
1986 if (retval) {
1987 ext4_std_error(dir->i_sb, retval);
1988 brelse(bh);
1989 return retval;
1990 }
1991 root = (struct dx_root *) bh->b_data;
1992
1993 /* The 0th block becomes the root, move the dirents out */
1994 fde = &root->dotdot;
1995 de = (struct ext4_dir_entry_2 *)((char *)fde +
1996 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1997 if ((char *) de >= (((char *) root) + blocksize)) {
1998 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1999 brelse(bh);
2000 return -EFSCORRUPTED;
2001 }
2002 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2003
2004 /* Allocate new block for the 0th block's dirents */
2005 bh2 = ext4_append(handle, dir, &block);
2006 if (IS_ERR(bh2)) {
2007 brelse(bh);
2008 return PTR_ERR(bh2);
2009 }
2010 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2011 data1 = bh2->b_data;
2012
2013 memcpy (data1, de, len);
2014 de = (struct ext4_dir_entry_2 *) data1;
2015 top = data1 + len;
2016 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2017 de = de2;
2018 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2019 (char *) de,
2020 blocksize);
2021
2022 if (csum_size) {
2023 t = EXT4_DIRENT_TAIL(data1, blocksize);
2024 initialize_dirent_tail(t, blocksize);
2025 }
2026
2027 /* Initialize the root; the dot dirents already exist */
2028 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2029 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2030 blocksize);
2031 memset (&root->info, 0, sizeof(root->info));
2032 root->info.info_length = sizeof(root->info);
2033 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2034 entries = root->entries;
2035 dx_set_block(entries, 1);
2036 dx_set_count(entries, 1);
2037 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2038
2039 /* Initialize as for dx_probe */
2040 fname->hinfo.hash_version = root->info.hash_version;
2041 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2042 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2043 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2044 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2045
2046 memset(frames, 0, sizeof(frames));
2047 frame = frames;
2048 frame->entries = entries;
2049 frame->at = entries;
2050 frame->bh = bh;
2051
2052 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2053 if (retval)
2054 goto out_frames;
2055 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2056 if (retval)
2057 goto out_frames;
2058
2059 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2060 if (IS_ERR(de)) {
2061 retval = PTR_ERR(de);
2062 goto out_frames;
2063 }
2064
2065 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2066 out_frames:
2067 /*
2068 * Even if the block split failed, we have to properly write
2069 * out all the changes we did so far. Otherwise we can end up
2070 * with corrupted filesystem.
2071 */
2072 if (retval)
2073 ext4_mark_inode_dirty(handle, dir);
2074 dx_release(frames);
2075 brelse(bh2);
2076 return retval;
2077 }
2078
2079 /*
2080 * ext4_add_entry()
2081 *
2082 * adds a file entry to the specified directory, using the same
2083 * semantics as ext4_find_entry(). It returns NULL if it failed.
2084 *
2085 * NOTE!! The inode part of 'de' is left at 0 - which means you
2086 * may not sleep between calling this and putting something into
2087 * the entry, as someone else might have used it while you slept.
2088 */
2089 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2090 struct inode *inode)
2091 {
2092 struct inode *dir = d_inode(dentry->d_parent);
2093 struct buffer_head *bh = NULL;
2094 struct ext4_dir_entry_2 *de;
2095 struct ext4_dir_entry_tail *t;
2096 struct super_block *sb;
2097 struct ext4_filename fname;
2098 int retval;
2099 int dx_fallback=0;
2100 unsigned blocksize;
2101 ext4_lblk_t block, blocks;
2102 int csum_size = 0;
2103
2104 if (ext4_has_metadata_csum(inode->i_sb))
2105 csum_size = sizeof(struct ext4_dir_entry_tail);
2106
2107 sb = dir->i_sb;
2108 blocksize = sb->s_blocksize;
2109 if (!dentry->d_name.len)
2110 return -EINVAL;
2111
2112 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2113 if (retval)
2114 return retval;
2115
2116 if (ext4_has_inline_data(dir)) {
2117 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2118 if (retval < 0)
2119 goto out;
2120 if (retval == 1) {
2121 retval = 0;
2122 goto out;
2123 }
2124 }
2125
2126 if (is_dx(dir)) {
2127 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2128 if (!retval || (retval != ERR_BAD_DX_DIR))
2129 goto out;
2130 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2131 dx_fallback++;
2132 ext4_mark_inode_dirty(handle, dir);
2133 }
2134 blocks = dir->i_size >> sb->s_blocksize_bits;
2135 for (block = 0; block < blocks; block++) {
2136 bh = ext4_read_dirblock(dir, block, DIRENT);
2137 if (IS_ERR(bh)) {
2138 retval = PTR_ERR(bh);
2139 bh = NULL;
2140 goto out;
2141 }
2142 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2143 NULL, bh);
2144 if (retval != -ENOSPC)
2145 goto out;
2146
2147 if (blocks == 1 && !dx_fallback &&
2148 ext4_has_feature_dir_index(sb)) {
2149 retval = make_indexed_dir(handle, &fname, dir,
2150 inode, bh);
2151 bh = NULL; /* make_indexed_dir releases bh */
2152 goto out;
2153 }
2154 brelse(bh);
2155 }
2156 bh = ext4_append(handle, dir, &block);
2157 if (IS_ERR(bh)) {
2158 retval = PTR_ERR(bh);
2159 bh = NULL;
2160 goto out;
2161 }
2162 de = (struct ext4_dir_entry_2 *) bh->b_data;
2163 de->inode = 0;
2164 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2165
2166 if (csum_size) {
2167 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2168 initialize_dirent_tail(t, blocksize);
2169 }
2170
2171 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2172 out:
2173 ext4_fname_free_filename(&fname);
2174 brelse(bh);
2175 if (retval == 0)
2176 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2177 return retval;
2178 }
2179
2180 /*
2181 * Returns 0 for success, or a negative error value
2182 */
2183 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2184 struct inode *dir, struct inode *inode)
2185 {
2186 struct dx_frame frames[2], *frame;
2187 struct dx_entry *entries, *at;
2188 struct buffer_head *bh;
2189 struct super_block *sb = dir->i_sb;
2190 struct ext4_dir_entry_2 *de;
2191 int err;
2192
2193 frame = dx_probe(fname, dir, NULL, frames);
2194 if (IS_ERR(frame))
2195 return PTR_ERR(frame);
2196 entries = frame->entries;
2197 at = frame->at;
2198 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2199 if (IS_ERR(bh)) {
2200 err = PTR_ERR(bh);
2201 bh = NULL;
2202 goto cleanup;
2203 }
2204
2205 BUFFER_TRACE(bh, "get_write_access");
2206 err = ext4_journal_get_write_access(handle, bh);
2207 if (err)
2208 goto journal_error;
2209
2210 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2211 if (err != -ENOSPC)
2212 goto cleanup;
2213
2214 /* Block full, should compress but for now just split */
2215 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2216 dx_get_count(entries), dx_get_limit(entries)));
2217 /* Need to split index? */
2218 if (dx_get_count(entries) == dx_get_limit(entries)) {
2219 ext4_lblk_t newblock;
2220 unsigned icount = dx_get_count(entries);
2221 int levels = frame - frames;
2222 struct dx_entry *entries2;
2223 struct dx_node *node2;
2224 struct buffer_head *bh2;
2225
2226 if (levels && (dx_get_count(frames->entries) ==
2227 dx_get_limit(frames->entries))) {
2228 ext4_warning_inode(dir, "Directory index full!");
2229 err = -ENOSPC;
2230 goto cleanup;
2231 }
2232 bh2 = ext4_append(handle, dir, &newblock);
2233 if (IS_ERR(bh2)) {
2234 err = PTR_ERR(bh2);
2235 goto cleanup;
2236 }
2237 node2 = (struct dx_node *)(bh2->b_data);
2238 entries2 = node2->entries;
2239 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2240 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2241 sb->s_blocksize);
2242 BUFFER_TRACE(frame->bh, "get_write_access");
2243 err = ext4_journal_get_write_access(handle, frame->bh);
2244 if (err)
2245 goto journal_error;
2246 if (levels) {
2247 unsigned icount1 = icount/2, icount2 = icount - icount1;
2248 unsigned hash2 = dx_get_hash(entries + icount1);
2249 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2250 icount1, icount2));
2251
2252 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2253 err = ext4_journal_get_write_access(handle,
2254 frames[0].bh);
2255 if (err)
2256 goto journal_error;
2257
2258 memcpy((char *) entries2, (char *) (entries + icount1),
2259 icount2 * sizeof(struct dx_entry));
2260 dx_set_count(entries, icount1);
2261 dx_set_count(entries2, icount2);
2262 dx_set_limit(entries2, dx_node_limit(dir));
2263
2264 /* Which index block gets the new entry? */
2265 if (at - entries >= icount1) {
2266 frame->at = at = at - entries - icount1 + entries2;
2267 frame->entries = entries = entries2;
2268 swap(frame->bh, bh2);
2269 }
2270 dx_insert_block(frames + 0, hash2, newblock);
2271 dxtrace(dx_show_index("node", frames[1].entries));
2272 dxtrace(dx_show_index("node",
2273 ((struct dx_node *) bh2->b_data)->entries));
2274 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2275 if (err)
2276 goto journal_error;
2277 brelse (bh2);
2278 } else {
2279 dxtrace(printk(KERN_DEBUG
2280 "Creating second level index...\n"));
2281 memcpy((char *) entries2, (char *) entries,
2282 icount * sizeof(struct dx_entry));
2283 dx_set_limit(entries2, dx_node_limit(dir));
2284
2285 /* Set up root */
2286 dx_set_count(entries, 1);
2287 dx_set_block(entries + 0, newblock);
2288 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2289
2290 /* Add new access path frame */
2291 frame = frames + 1;
2292 frame->at = at = at - entries + entries2;
2293 frame->entries = entries = entries2;
2294 frame->bh = bh2;
2295 err = ext4_journal_get_write_access(handle,
2296 frame->bh);
2297 if (err)
2298 goto journal_error;
2299 }
2300 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2301 if (err) {
2302 ext4_std_error(inode->i_sb, err);
2303 goto cleanup;
2304 }
2305 }
2306 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2307 if (IS_ERR(de)) {
2308 err = PTR_ERR(de);
2309 goto cleanup;
2310 }
2311 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2312 goto cleanup;
2313
2314 journal_error:
2315 ext4_std_error(dir->i_sb, err);
2316 cleanup:
2317 brelse(bh);
2318 dx_release(frames);
2319 return err;
2320 }
2321
2322 /*
2323 * ext4_generic_delete_entry deletes a directory entry by merging it
2324 * with the previous entry
2325 */
2326 int ext4_generic_delete_entry(handle_t *handle,
2327 struct inode *dir,
2328 struct ext4_dir_entry_2 *de_del,
2329 struct buffer_head *bh,
2330 void *entry_buf,
2331 int buf_size,
2332 int csum_size)
2333 {
2334 struct ext4_dir_entry_2 *de, *pde;
2335 unsigned int blocksize = dir->i_sb->s_blocksize;
2336 int i;
2337
2338 i = 0;
2339 pde = NULL;
2340 de = (struct ext4_dir_entry_2 *)entry_buf;
2341 while (i < buf_size - csum_size) {
2342 if (ext4_check_dir_entry(dir, NULL, de, bh,
2343 bh->b_data, bh->b_size, i))
2344 return -EFSCORRUPTED;
2345 if (de == de_del) {
2346 if (pde)
2347 pde->rec_len = ext4_rec_len_to_disk(
2348 ext4_rec_len_from_disk(pde->rec_len,
2349 blocksize) +
2350 ext4_rec_len_from_disk(de->rec_len,
2351 blocksize),
2352 blocksize);
2353 else
2354 de->inode = 0;
2355 dir->i_version++;
2356 return 0;
2357 }
2358 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2359 pde = de;
2360 de = ext4_next_entry(de, blocksize);
2361 }
2362 return -ENOENT;
2363 }
2364
2365 static int ext4_delete_entry(handle_t *handle,
2366 struct inode *dir,
2367 struct ext4_dir_entry_2 *de_del,
2368 struct buffer_head *bh)
2369 {
2370 int err, csum_size = 0;
2371
2372 if (ext4_has_inline_data(dir)) {
2373 int has_inline_data = 1;
2374 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2375 &has_inline_data);
2376 if (has_inline_data)
2377 return err;
2378 }
2379
2380 if (ext4_has_metadata_csum(dir->i_sb))
2381 csum_size = sizeof(struct ext4_dir_entry_tail);
2382
2383 BUFFER_TRACE(bh, "get_write_access");
2384 err = ext4_journal_get_write_access(handle, bh);
2385 if (unlikely(err))
2386 goto out;
2387
2388 err = ext4_generic_delete_entry(handle, dir, de_del,
2389 bh, bh->b_data,
2390 dir->i_sb->s_blocksize, csum_size);
2391 if (err)
2392 goto out;
2393
2394 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2395 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2396 if (unlikely(err))
2397 goto out;
2398
2399 return 0;
2400 out:
2401 if (err != -ENOENT)
2402 ext4_std_error(dir->i_sb, err);
2403 return err;
2404 }
2405
2406 /*
2407 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2408 * since this indicates that nlinks count was previously 1.
2409 */
2410 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2411 {
2412 inc_nlink(inode);
2413 if (is_dx(inode) && inode->i_nlink > 1) {
2414 /* limit is 16-bit i_links_count */
2415 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2416 set_nlink(inode, 1);
2417 ext4_set_feature_dir_nlink(inode->i_sb);
2418 }
2419 }
2420 }
2421
2422 /*
2423 * If a directory had nlink == 1, then we should let it be 1. This indicates
2424 * directory has >EXT4_LINK_MAX subdirs.
2425 */
2426 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2427 {
2428 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2429 drop_nlink(inode);
2430 }
2431
2432
2433 static int ext4_add_nondir(handle_t *handle,
2434 struct dentry *dentry, struct inode *inode)
2435 {
2436 int err = ext4_add_entry(handle, dentry, inode);
2437 if (!err) {
2438 ext4_mark_inode_dirty(handle, inode);
2439 unlock_new_inode(inode);
2440 d_instantiate(dentry, inode);
2441 return 0;
2442 }
2443 drop_nlink(inode);
2444 unlock_new_inode(inode);
2445 iput(inode);
2446 return err;
2447 }
2448
2449 /*
2450 * By the time this is called, we already have created
2451 * the directory cache entry for the new file, but it
2452 * is so far negative - it has no inode.
2453 *
2454 * If the create succeeds, we fill in the inode information
2455 * with d_instantiate().
2456 */
2457 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2458 bool excl)
2459 {
2460 handle_t *handle;
2461 struct inode *inode;
2462 int err, credits, retries = 0;
2463
2464 err = dquot_initialize(dir);
2465 if (err)
2466 return err;
2467
2468 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2469 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2470 retry:
2471 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2472 NULL, EXT4_HT_DIR, credits);
2473 handle = ext4_journal_current_handle();
2474 err = PTR_ERR(inode);
2475 if (!IS_ERR(inode)) {
2476 inode->i_op = &ext4_file_inode_operations;
2477 inode->i_fop = &ext4_file_operations;
2478 ext4_set_aops(inode);
2479 err = ext4_add_nondir(handle, dentry, inode);
2480 if (!err && IS_DIRSYNC(dir))
2481 ext4_handle_sync(handle);
2482 }
2483 if (handle)
2484 ext4_journal_stop(handle);
2485 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2486 goto retry;
2487 return err;
2488 }
2489
2490 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2491 umode_t mode, dev_t rdev)
2492 {
2493 handle_t *handle;
2494 struct inode *inode;
2495 int err, credits, retries = 0;
2496
2497 err = dquot_initialize(dir);
2498 if (err)
2499 return err;
2500
2501 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2502 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2503 retry:
2504 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2505 NULL, EXT4_HT_DIR, credits);
2506 handle = ext4_journal_current_handle();
2507 err = PTR_ERR(inode);
2508 if (!IS_ERR(inode)) {
2509 init_special_inode(inode, inode->i_mode, rdev);
2510 inode->i_op = &ext4_special_inode_operations;
2511 err = ext4_add_nondir(handle, dentry, inode);
2512 if (!err && IS_DIRSYNC(dir))
2513 ext4_handle_sync(handle);
2514 }
2515 if (handle)
2516 ext4_journal_stop(handle);
2517 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2518 goto retry;
2519 return err;
2520 }
2521
2522 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2523 {
2524 handle_t *handle;
2525 struct inode *inode;
2526 int err, retries = 0;
2527
2528 err = dquot_initialize(dir);
2529 if (err)
2530 return err;
2531
2532 retry:
2533 inode = ext4_new_inode_start_handle(dir, mode,
2534 NULL, 0, NULL,
2535 EXT4_HT_DIR,
2536 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2537 4 + EXT4_XATTR_TRANS_BLOCKS);
2538 handle = ext4_journal_current_handle();
2539 err = PTR_ERR(inode);
2540 if (!IS_ERR(inode)) {
2541 inode->i_op = &ext4_file_inode_operations;
2542 inode->i_fop = &ext4_file_operations;
2543 ext4_set_aops(inode);
2544 d_tmpfile(dentry, inode);
2545 err = ext4_orphan_add(handle, inode);
2546 if (err)
2547 goto err_unlock_inode;
2548 mark_inode_dirty(inode);
2549 unlock_new_inode(inode);
2550 }
2551 if (handle)
2552 ext4_journal_stop(handle);
2553 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2554 goto retry;
2555 return err;
2556 err_unlock_inode:
2557 ext4_journal_stop(handle);
2558 unlock_new_inode(inode);
2559 return err;
2560 }
2561
2562 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2563 struct ext4_dir_entry_2 *de,
2564 int blocksize, int csum_size,
2565 unsigned int parent_ino, int dotdot_real_len)
2566 {
2567 de->inode = cpu_to_le32(inode->i_ino);
2568 de->name_len = 1;
2569 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2570 blocksize);
2571 strcpy(de->name, ".");
2572 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2573
2574 de = ext4_next_entry(de, blocksize);
2575 de->inode = cpu_to_le32(parent_ino);
2576 de->name_len = 2;
2577 if (!dotdot_real_len)
2578 de->rec_len = ext4_rec_len_to_disk(blocksize -
2579 (csum_size + EXT4_DIR_REC_LEN(1)),
2580 blocksize);
2581 else
2582 de->rec_len = ext4_rec_len_to_disk(
2583 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2584 strcpy(de->name, "..");
2585 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2586
2587 return ext4_next_entry(de, blocksize);
2588 }
2589
2590 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2591 struct inode *inode)
2592 {
2593 struct buffer_head *dir_block = NULL;
2594 struct ext4_dir_entry_2 *de;
2595 struct ext4_dir_entry_tail *t;
2596 ext4_lblk_t block = 0;
2597 unsigned int blocksize = dir->i_sb->s_blocksize;
2598 int csum_size = 0;
2599 int err;
2600
2601 if (ext4_has_metadata_csum(dir->i_sb))
2602 csum_size = sizeof(struct ext4_dir_entry_tail);
2603
2604 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2605 err = ext4_try_create_inline_dir(handle, dir, inode);
2606 if (err < 0 && err != -ENOSPC)
2607 goto out;
2608 if (!err)
2609 goto out;
2610 }
2611
2612 inode->i_size = 0;
2613 dir_block = ext4_append(handle, inode, &block);
2614 if (IS_ERR(dir_block))
2615 return PTR_ERR(dir_block);
2616 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2617 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2618 set_nlink(inode, 2);
2619 if (csum_size) {
2620 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2621 initialize_dirent_tail(t, blocksize);
2622 }
2623
2624 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2625 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2626 if (err)
2627 goto out;
2628 set_buffer_verified(dir_block);
2629 out:
2630 brelse(dir_block);
2631 return err;
2632 }
2633
2634 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2635 {
2636 handle_t *handle;
2637 struct inode *inode;
2638 int err, credits, retries = 0;
2639
2640 if (EXT4_DIR_LINK_MAX(dir))
2641 return -EMLINK;
2642
2643 err = dquot_initialize(dir);
2644 if (err)
2645 return err;
2646
2647 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2648 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2649 retry:
2650 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2651 &dentry->d_name,
2652 0, NULL, EXT4_HT_DIR, credits);
2653 handle = ext4_journal_current_handle();
2654 err = PTR_ERR(inode);
2655 if (IS_ERR(inode))
2656 goto out_stop;
2657
2658 inode->i_op = &ext4_dir_inode_operations;
2659 inode->i_fop = &ext4_dir_operations;
2660 err = ext4_init_new_dir(handle, dir, inode);
2661 if (err)
2662 goto out_clear_inode;
2663 err = ext4_mark_inode_dirty(handle, inode);
2664 if (!err)
2665 err = ext4_add_entry(handle, dentry, inode);
2666 if (err) {
2667 out_clear_inode:
2668 clear_nlink(inode);
2669 unlock_new_inode(inode);
2670 ext4_mark_inode_dirty(handle, inode);
2671 iput(inode);
2672 goto out_stop;
2673 }
2674 ext4_inc_count(handle, dir);
2675 ext4_update_dx_flag(dir);
2676 err = ext4_mark_inode_dirty(handle, dir);
2677 if (err)
2678 goto out_clear_inode;
2679 unlock_new_inode(inode);
2680 d_instantiate(dentry, inode);
2681 if (IS_DIRSYNC(dir))
2682 ext4_handle_sync(handle);
2683
2684 out_stop:
2685 if (handle)
2686 ext4_journal_stop(handle);
2687 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2688 goto retry;
2689 return err;
2690 }
2691
2692 /*
2693 * routine to check that the specified directory is empty (for rmdir)
2694 */
2695 bool ext4_empty_dir(struct inode *inode)
2696 {
2697 unsigned int offset;
2698 struct buffer_head *bh;
2699 struct ext4_dir_entry_2 *de, *de1;
2700 struct super_block *sb;
2701
2702 if (ext4_has_inline_data(inode)) {
2703 int has_inline_data = 1;
2704 int ret;
2705
2706 ret = empty_inline_dir(inode, &has_inline_data);
2707 if (has_inline_data)
2708 return ret;
2709 }
2710
2711 sb = inode->i_sb;
2712 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2713 EXT4_ERROR_INODE(inode, "invalid size");
2714 return true;
2715 }
2716 bh = ext4_read_dirblock(inode, 0, EITHER);
2717 if (IS_ERR(bh))
2718 return true;
2719
2720 de = (struct ext4_dir_entry_2 *) bh->b_data;
2721 de1 = ext4_next_entry(de, sb->s_blocksize);
2722 if (le32_to_cpu(de->inode) != inode->i_ino ||
2723 le32_to_cpu(de1->inode) == 0 ||
2724 strcmp(".", de->name) || strcmp("..", de1->name)) {
2725 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2726 brelse(bh);
2727 return true;
2728 }
2729 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2730 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2731 de = ext4_next_entry(de1, sb->s_blocksize);
2732 while (offset < inode->i_size) {
2733 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2734 unsigned int lblock;
2735 brelse(bh);
2736 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2737 bh = ext4_read_dirblock(inode, lblock, EITHER);
2738 if (IS_ERR(bh))
2739 return true;
2740 de = (struct ext4_dir_entry_2 *) bh->b_data;
2741 }
2742 if (ext4_check_dir_entry(inode, NULL, de, bh,
2743 bh->b_data, bh->b_size, offset)) {
2744 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2745 sb->s_blocksize);
2746 offset = (offset | (sb->s_blocksize - 1)) + 1;
2747 continue;
2748 }
2749 if (le32_to_cpu(de->inode)) {
2750 brelse(bh);
2751 return false;
2752 }
2753 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2754 de = ext4_next_entry(de, sb->s_blocksize);
2755 }
2756 brelse(bh);
2757 return true;
2758 }
2759
2760 /*
2761 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2762 * such inodes, starting at the superblock, in case we crash before the
2763 * file is closed/deleted, or in case the inode truncate spans multiple
2764 * transactions and the last transaction is not recovered after a crash.
2765 *
2766 * At filesystem recovery time, we walk this list deleting unlinked
2767 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2768 *
2769 * Orphan list manipulation functions must be called under i_mutex unless
2770 * we are just creating the inode or deleting it.
2771 */
2772 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2773 {
2774 struct super_block *sb = inode->i_sb;
2775 struct ext4_sb_info *sbi = EXT4_SB(sb);
2776 struct ext4_iloc iloc;
2777 int err = 0, rc;
2778 bool dirty = false;
2779
2780 if (!sbi->s_journal || is_bad_inode(inode))
2781 return 0;
2782
2783 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2784 !inode_is_locked(inode));
2785 /*
2786 * Exit early if inode already is on orphan list. This is a big speedup
2787 * since we don't have to contend on the global s_orphan_lock.
2788 */
2789 if (!list_empty(&EXT4_I(inode)->i_orphan))
2790 return 0;
2791
2792 /*
2793 * Orphan handling is only valid for files with data blocks
2794 * being truncated, or files being unlinked. Note that we either
2795 * hold i_mutex, or the inode can not be referenced from outside,
2796 * so i_nlink should not be bumped due to race
2797 */
2798 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2799 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2800
2801 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2802 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2803 if (err)
2804 goto out;
2805
2806 err = ext4_reserve_inode_write(handle, inode, &iloc);
2807 if (err)
2808 goto out;
2809
2810 mutex_lock(&sbi->s_orphan_lock);
2811 /*
2812 * Due to previous errors inode may be already a part of on-disk
2813 * orphan list. If so skip on-disk list modification.
2814 */
2815 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2816 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2817 /* Insert this inode at the head of the on-disk orphan list */
2818 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2819 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2820 dirty = true;
2821 }
2822 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2823 mutex_unlock(&sbi->s_orphan_lock);
2824
2825 if (dirty) {
2826 err = ext4_handle_dirty_super(handle, sb);
2827 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2828 if (!err)
2829 err = rc;
2830 if (err) {
2831 /*
2832 * We have to remove inode from in-memory list if
2833 * addition to on disk orphan list failed. Stray orphan
2834 * list entries can cause panics at unmount time.
2835 */
2836 mutex_lock(&sbi->s_orphan_lock);
2837 list_del_init(&EXT4_I(inode)->i_orphan);
2838 mutex_unlock(&sbi->s_orphan_lock);
2839 }
2840 }
2841 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2842 jbd_debug(4, "orphan inode %lu will point to %d\n",
2843 inode->i_ino, NEXT_ORPHAN(inode));
2844 out:
2845 ext4_std_error(sb, err);
2846 return err;
2847 }
2848
2849 /*
2850 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2851 * of such inodes stored on disk, because it is finally being cleaned up.
2852 */
2853 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2854 {
2855 struct list_head *prev;
2856 struct ext4_inode_info *ei = EXT4_I(inode);
2857 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2858 __u32 ino_next;
2859 struct ext4_iloc iloc;
2860 int err = 0;
2861
2862 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2863 return 0;
2864
2865 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2866 !inode_is_locked(inode));
2867 /* Do this quick check before taking global s_orphan_lock. */
2868 if (list_empty(&ei->i_orphan))
2869 return 0;
2870
2871 if (handle) {
2872 /* Grab inode buffer early before taking global s_orphan_lock */
2873 err = ext4_reserve_inode_write(handle, inode, &iloc);
2874 }
2875
2876 mutex_lock(&sbi->s_orphan_lock);
2877 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2878
2879 prev = ei->i_orphan.prev;
2880 list_del_init(&ei->i_orphan);
2881
2882 /* If we're on an error path, we may not have a valid
2883 * transaction handle with which to update the orphan list on
2884 * disk, but we still need to remove the inode from the linked
2885 * list in memory. */
2886 if (!handle || err) {
2887 mutex_unlock(&sbi->s_orphan_lock);
2888 goto out_err;
2889 }
2890
2891 ino_next = NEXT_ORPHAN(inode);
2892 if (prev == &sbi->s_orphan) {
2893 jbd_debug(4, "superblock will point to %u\n", ino_next);
2894 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2895 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2896 if (err) {
2897 mutex_unlock(&sbi->s_orphan_lock);
2898 goto out_brelse;
2899 }
2900 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2901 mutex_unlock(&sbi->s_orphan_lock);
2902 err = ext4_handle_dirty_super(handle, inode->i_sb);
2903 } else {
2904 struct ext4_iloc iloc2;
2905 struct inode *i_prev =
2906 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2907
2908 jbd_debug(4, "orphan inode %lu will point to %u\n",
2909 i_prev->i_ino, ino_next);
2910 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2911 if (err) {
2912 mutex_unlock(&sbi->s_orphan_lock);
2913 goto out_brelse;
2914 }
2915 NEXT_ORPHAN(i_prev) = ino_next;
2916 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2917 mutex_unlock(&sbi->s_orphan_lock);
2918 }
2919 if (err)
2920 goto out_brelse;
2921 NEXT_ORPHAN(inode) = 0;
2922 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2923 out_err:
2924 ext4_std_error(inode->i_sb, err);
2925 return err;
2926
2927 out_brelse:
2928 brelse(iloc.bh);
2929 goto out_err;
2930 }
2931
2932 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2933 {
2934 int retval;
2935 struct inode *inode;
2936 struct buffer_head *bh;
2937 struct ext4_dir_entry_2 *de;
2938 handle_t *handle = NULL;
2939
2940 /* Initialize quotas before so that eventual writes go in
2941 * separate transaction */
2942 retval = dquot_initialize(dir);
2943 if (retval)
2944 return retval;
2945 retval = dquot_initialize(d_inode(dentry));
2946 if (retval)
2947 return retval;
2948
2949 retval = -ENOENT;
2950 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2951 if (IS_ERR(bh))
2952 return PTR_ERR(bh);
2953 if (!bh)
2954 goto end_rmdir;
2955
2956 inode = d_inode(dentry);
2957
2958 retval = -EFSCORRUPTED;
2959 if (le32_to_cpu(de->inode) != inode->i_ino)
2960 goto end_rmdir;
2961
2962 retval = -ENOTEMPTY;
2963 if (!ext4_empty_dir(inode))
2964 goto end_rmdir;
2965
2966 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2967 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2968 if (IS_ERR(handle)) {
2969 retval = PTR_ERR(handle);
2970 handle = NULL;
2971 goto end_rmdir;
2972 }
2973
2974 if (IS_DIRSYNC(dir))
2975 ext4_handle_sync(handle);
2976
2977 retval = ext4_delete_entry(handle, dir, de, bh);
2978 if (retval)
2979 goto end_rmdir;
2980 if (!EXT4_DIR_LINK_EMPTY(inode))
2981 ext4_warning_inode(inode,
2982 "empty directory '%.*s' has too many links (%u)",
2983 dentry->d_name.len, dentry->d_name.name,
2984 inode->i_nlink);
2985 inode->i_version++;
2986 clear_nlink(inode);
2987 /* There's no need to set i_disksize: the fact that i_nlink is
2988 * zero will ensure that the right thing happens during any
2989 * recovery. */
2990 inode->i_size = 0;
2991 ext4_orphan_add(handle, inode);
2992 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2993 ext4_mark_inode_dirty(handle, inode);
2994 ext4_dec_count(handle, dir);
2995 ext4_update_dx_flag(dir);
2996 ext4_mark_inode_dirty(handle, dir);
2997
2998 end_rmdir:
2999 brelse(bh);
3000 if (handle)
3001 ext4_journal_stop(handle);
3002 return retval;
3003 }
3004
3005 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3006 {
3007 int retval;
3008 struct inode *inode;
3009 struct buffer_head *bh;
3010 struct ext4_dir_entry_2 *de;
3011 handle_t *handle = NULL;
3012
3013 trace_ext4_unlink_enter(dir, dentry);
3014 /* Initialize quotas before so that eventual writes go
3015 * in separate transaction */
3016 retval = dquot_initialize(dir);
3017 if (retval)
3018 return retval;
3019 retval = dquot_initialize(d_inode(dentry));
3020 if (retval)
3021 return retval;
3022
3023 retval = -ENOENT;
3024 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3025 if (IS_ERR(bh))
3026 return PTR_ERR(bh);
3027 if (!bh)
3028 goto end_unlink;
3029
3030 inode = d_inode(dentry);
3031
3032 retval = -EFSCORRUPTED;
3033 if (le32_to_cpu(de->inode) != inode->i_ino)
3034 goto end_unlink;
3035
3036 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3037 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3038 if (IS_ERR(handle)) {
3039 retval = PTR_ERR(handle);
3040 handle = NULL;
3041 goto end_unlink;
3042 }
3043
3044 if (IS_DIRSYNC(dir))
3045 ext4_handle_sync(handle);
3046
3047 if (inode->i_nlink == 0) {
3048 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3049 dentry->d_name.len, dentry->d_name.name);
3050 set_nlink(inode, 1);
3051 }
3052 retval = ext4_delete_entry(handle, dir, de, bh);
3053 if (retval)
3054 goto end_unlink;
3055 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3056 ext4_update_dx_flag(dir);
3057 ext4_mark_inode_dirty(handle, dir);
3058 drop_nlink(inode);
3059 if (!inode->i_nlink)
3060 ext4_orphan_add(handle, inode);
3061 inode->i_ctime = ext4_current_time(inode);
3062 ext4_mark_inode_dirty(handle, inode);
3063
3064 end_unlink:
3065 brelse(bh);
3066 if (handle)
3067 ext4_journal_stop(handle);
3068 trace_ext4_unlink_exit(dentry, retval);
3069 return retval;
3070 }
3071
3072 static int ext4_symlink(struct inode *dir,
3073 struct dentry *dentry, const char *symname)
3074 {
3075 handle_t *handle;
3076 struct inode *inode;
3077 int err, len = strlen(symname);
3078 int credits;
3079 bool encryption_required;
3080 struct fscrypt_str disk_link;
3081 struct fscrypt_symlink_data *sd = NULL;
3082
3083 disk_link.len = len + 1;
3084 disk_link.name = (char *) symname;
3085
3086 encryption_required = (ext4_encrypted_inode(dir) ||
3087 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3088 if (encryption_required) {
3089 err = fscrypt_get_encryption_info(dir);
3090 if (err)
3091 return err;
3092 if (!fscrypt_has_encryption_key(dir))
3093 return -EPERM;
3094 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3095 sizeof(struct fscrypt_symlink_data));
3096 sd = kzalloc(disk_link.len, GFP_KERNEL);
3097 if (!sd)
3098 return -ENOMEM;
3099 }
3100
3101 if (disk_link.len > dir->i_sb->s_blocksize) {
3102 err = -ENAMETOOLONG;
3103 goto err_free_sd;
3104 }
3105
3106 err = dquot_initialize(dir);
3107 if (err)
3108 goto err_free_sd;
3109
3110 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3111 /*
3112 * For non-fast symlinks, we just allocate inode and put it on
3113 * orphan list in the first transaction => we need bitmap,
3114 * group descriptor, sb, inode block, quota blocks, and
3115 * possibly selinux xattr blocks.
3116 */
3117 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3118 EXT4_XATTR_TRANS_BLOCKS;
3119 } else {
3120 /*
3121 * Fast symlink. We have to add entry to directory
3122 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3123 * allocate new inode (bitmap, group descriptor, inode block,
3124 * quota blocks, sb is already counted in previous macros).
3125 */
3126 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3127 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3128 }
3129
3130 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3131 &dentry->d_name, 0, NULL,
3132 EXT4_HT_DIR, credits);
3133 handle = ext4_journal_current_handle();
3134 if (IS_ERR(inode)) {
3135 if (handle)
3136 ext4_journal_stop(handle);
3137 err = PTR_ERR(inode);
3138 goto err_free_sd;
3139 }
3140
3141 if (encryption_required) {
3142 struct qstr istr;
3143 struct fscrypt_str ostr =
3144 FSTR_INIT(sd->encrypted_path, disk_link.len);
3145
3146 istr.name = (const unsigned char *) symname;
3147 istr.len = len;
3148 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3149 if (err)
3150 goto err_drop_inode;
3151 sd->len = cpu_to_le16(ostr.len);
3152 disk_link.name = (char *) sd;
3153 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3154 }
3155
3156 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3157 if (!encryption_required)
3158 inode->i_op = &ext4_symlink_inode_operations;
3159 inode_nohighmem(inode);
3160 ext4_set_aops(inode);
3161 /*
3162 * We cannot call page_symlink() with transaction started
3163 * because it calls into ext4_write_begin() which can wait
3164 * for transaction commit if we are running out of space
3165 * and thus we deadlock. So we have to stop transaction now
3166 * and restart it when symlink contents is written.
3167 *
3168 * To keep fs consistent in case of crash, we have to put inode
3169 * to orphan list in the mean time.
3170 */
3171 drop_nlink(inode);
3172 err = ext4_orphan_add(handle, inode);
3173 ext4_journal_stop(handle);
3174 handle = NULL;
3175 if (err)
3176 goto err_drop_inode;
3177 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3178 if (err)
3179 goto err_drop_inode;
3180 /*
3181 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3182 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3183 */
3184 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3185 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3186 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3187 if (IS_ERR(handle)) {
3188 err = PTR_ERR(handle);
3189 handle = NULL;
3190 goto err_drop_inode;
3191 }
3192 set_nlink(inode, 1);
3193 err = ext4_orphan_del(handle, inode);
3194 if (err)
3195 goto err_drop_inode;
3196 } else {
3197 /* clear the extent format for fast symlink */
3198 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3199 if (!encryption_required) {
3200 inode->i_op = &ext4_fast_symlink_inode_operations;
3201 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3202 }
3203 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3204 disk_link.len);
3205 inode->i_size = disk_link.len - 1;
3206 }
3207 EXT4_I(inode)->i_disksize = inode->i_size;
3208 err = ext4_add_nondir(handle, dentry, inode);
3209 if (!err && IS_DIRSYNC(dir))
3210 ext4_handle_sync(handle);
3211
3212 if (handle)
3213 ext4_journal_stop(handle);
3214 kfree(sd);
3215 return err;
3216 err_drop_inode:
3217 if (handle)
3218 ext4_journal_stop(handle);
3219 clear_nlink(inode);
3220 unlock_new_inode(inode);
3221 iput(inode);
3222 err_free_sd:
3223 kfree(sd);
3224 return err;
3225 }
3226
3227 static int ext4_link(struct dentry *old_dentry,
3228 struct inode *dir, struct dentry *dentry)
3229 {
3230 handle_t *handle;
3231 struct inode *inode = d_inode(old_dentry);
3232 int err, retries = 0;
3233
3234 if (inode->i_nlink >= EXT4_LINK_MAX)
3235 return -EMLINK;
3236 if (ext4_encrypted_inode(dir) &&
3237 !fscrypt_has_permitted_context(dir, inode))
3238 return -EPERM;
3239
3240 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3241 (!projid_eq(EXT4_I(dir)->i_projid,
3242 EXT4_I(old_dentry->d_inode)->i_projid)))
3243 return -EXDEV;
3244
3245 err = dquot_initialize(dir);
3246 if (err)
3247 return err;
3248
3249 retry:
3250 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3251 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3252 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3253 if (IS_ERR(handle))
3254 return PTR_ERR(handle);
3255
3256 if (IS_DIRSYNC(dir))
3257 ext4_handle_sync(handle);
3258
3259 inode->i_ctime = ext4_current_time(inode);
3260 ext4_inc_count(handle, inode);
3261 ihold(inode);
3262
3263 err = ext4_add_entry(handle, dentry, inode);
3264 if (!err) {
3265 ext4_mark_inode_dirty(handle, inode);
3266 /* this can happen only for tmpfile being
3267 * linked the first time
3268 */
3269 if (inode->i_nlink == 1)
3270 ext4_orphan_del(handle, inode);
3271 d_instantiate(dentry, inode);
3272 } else {
3273 drop_nlink(inode);
3274 iput(inode);
3275 }
3276 ext4_journal_stop(handle);
3277 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3278 goto retry;
3279 return err;
3280 }
3281
3282
3283 /*
3284 * Try to find buffer head where contains the parent block.
3285 * It should be the inode block if it is inlined or the 1st block
3286 * if it is a normal dir.
3287 */
3288 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3289 struct inode *inode,
3290 int *retval,
3291 struct ext4_dir_entry_2 **parent_de,
3292 int *inlined)
3293 {
3294 struct buffer_head *bh;
3295
3296 if (!ext4_has_inline_data(inode)) {
3297 bh = ext4_read_dirblock(inode, 0, EITHER);
3298 if (IS_ERR(bh)) {
3299 *retval = PTR_ERR(bh);
3300 return NULL;
3301 }
3302 *parent_de = ext4_next_entry(
3303 (struct ext4_dir_entry_2 *)bh->b_data,
3304 inode->i_sb->s_blocksize);
3305 return bh;
3306 }
3307
3308 *inlined = 1;
3309 return ext4_get_first_inline_block(inode, parent_de, retval);
3310 }
3311
3312 struct ext4_renament {
3313 struct inode *dir;
3314 struct dentry *dentry;
3315 struct inode *inode;
3316 bool is_dir;
3317 int dir_nlink_delta;
3318
3319 /* entry for "dentry" */
3320 struct buffer_head *bh;
3321 struct ext4_dir_entry_2 *de;
3322 int inlined;
3323
3324 /* entry for ".." in inode if it's a directory */
3325 struct buffer_head *dir_bh;
3326 struct ext4_dir_entry_2 *parent_de;
3327 int dir_inlined;
3328 };
3329
3330 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3331 {
3332 int retval;
3333
3334 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3335 &retval, &ent->parent_de,
3336 &ent->dir_inlined);
3337 if (!ent->dir_bh)
3338 return retval;
3339 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3340 return -EFSCORRUPTED;
3341 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3342 return ext4_journal_get_write_access(handle, ent->dir_bh);
3343 }
3344
3345 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3346 unsigned dir_ino)
3347 {
3348 int retval;
3349
3350 ent->parent_de->inode = cpu_to_le32(dir_ino);
3351 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3352 if (!ent->dir_inlined) {
3353 if (is_dx(ent->inode)) {
3354 retval = ext4_handle_dirty_dx_node(handle,
3355 ent->inode,
3356 ent->dir_bh);
3357 } else {
3358 retval = ext4_handle_dirty_dirent_node(handle,
3359 ent->inode,
3360 ent->dir_bh);
3361 }
3362 } else {
3363 retval = ext4_mark_inode_dirty(handle, ent->inode);
3364 }
3365 if (retval) {
3366 ext4_std_error(ent->dir->i_sb, retval);
3367 return retval;
3368 }
3369 return 0;
3370 }
3371
3372 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3373 unsigned ino, unsigned file_type)
3374 {
3375 int retval;
3376
3377 BUFFER_TRACE(ent->bh, "get write access");
3378 retval = ext4_journal_get_write_access(handle, ent->bh);
3379 if (retval)
3380 return retval;
3381 ent->de->inode = cpu_to_le32(ino);
3382 if (ext4_has_feature_filetype(ent->dir->i_sb))
3383 ent->de->file_type = file_type;
3384 ent->dir->i_version++;
3385 ent->dir->i_ctime = ent->dir->i_mtime =
3386 ext4_current_time(ent->dir);
3387 ext4_mark_inode_dirty(handle, ent->dir);
3388 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3389 if (!ent->inlined) {
3390 retval = ext4_handle_dirty_dirent_node(handle,
3391 ent->dir, ent->bh);
3392 if (unlikely(retval)) {
3393 ext4_std_error(ent->dir->i_sb, retval);
3394 return retval;
3395 }
3396 }
3397 brelse(ent->bh);
3398 ent->bh = NULL;
3399
3400 return 0;
3401 }
3402
3403 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3404 const struct qstr *d_name)
3405 {
3406 int retval = -ENOENT;
3407 struct buffer_head *bh;
3408 struct ext4_dir_entry_2 *de;
3409
3410 bh = ext4_find_entry(dir, d_name, &de, NULL);
3411 if (IS_ERR(bh))
3412 return PTR_ERR(bh);
3413 if (bh) {
3414 retval = ext4_delete_entry(handle, dir, de, bh);
3415 brelse(bh);
3416 }
3417 return retval;
3418 }
3419
3420 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3421 int force_reread)
3422 {
3423 int retval;
3424 /*
3425 * ent->de could have moved from under us during htree split, so make
3426 * sure that we are deleting the right entry. We might also be pointing
3427 * to a stale entry in the unused part of ent->bh so just checking inum
3428 * and the name isn't enough.
3429 */
3430 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3431 ent->de->name_len != ent->dentry->d_name.len ||
3432 strncmp(ent->de->name, ent->dentry->d_name.name,
3433 ent->de->name_len) ||
3434 force_reread) {
3435 retval = ext4_find_delete_entry(handle, ent->dir,
3436 &ent->dentry->d_name);
3437 } else {
3438 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3439 if (retval == -ENOENT) {
3440 retval = ext4_find_delete_entry(handle, ent->dir,
3441 &ent->dentry->d_name);
3442 }
3443 }
3444
3445 if (retval) {
3446 ext4_warning_inode(ent->dir,
3447 "Deleting old file: nlink %d, error=%d",
3448 ent->dir->i_nlink, retval);
3449 }
3450 }
3451
3452 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3453 {
3454 if (ent->dir_nlink_delta) {
3455 if (ent->dir_nlink_delta == -1)
3456 ext4_dec_count(handle, ent->dir);
3457 else
3458 ext4_inc_count(handle, ent->dir);
3459 ext4_mark_inode_dirty(handle, ent->dir);
3460 }
3461 }
3462
3463 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3464 int credits, handle_t **h)
3465 {
3466 struct inode *wh;
3467 handle_t *handle;
3468 int retries = 0;
3469
3470 /*
3471 * for inode block, sb block, group summaries,
3472 * and inode bitmap
3473 */
3474 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3475 EXT4_XATTR_TRANS_BLOCKS + 4);
3476 retry:
3477 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3478 &ent->dentry->d_name, 0, NULL,
3479 EXT4_HT_DIR, credits);
3480
3481 handle = ext4_journal_current_handle();
3482 if (IS_ERR(wh)) {
3483 if (handle)
3484 ext4_journal_stop(handle);
3485 if (PTR_ERR(wh) == -ENOSPC &&
3486 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3487 goto retry;
3488 } else {
3489 *h = handle;
3490 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3491 wh->i_op = &ext4_special_inode_operations;
3492 }
3493 return wh;
3494 }
3495
3496 /*
3497 * Anybody can rename anything with this: the permission checks are left to the
3498 * higher-level routines.
3499 *
3500 * n.b. old_{dentry,inode) refers to the source dentry/inode
3501 * while new_{dentry,inode) refers to the destination dentry/inode
3502 * This comes from rename(const char *oldpath, const char *newpath)
3503 */
3504 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3505 struct inode *new_dir, struct dentry *new_dentry,
3506 unsigned int flags)
3507 {
3508 handle_t *handle = NULL;
3509 struct ext4_renament old = {
3510 .dir = old_dir,
3511 .dentry = old_dentry,
3512 .inode = d_inode(old_dentry),
3513 };
3514 struct ext4_renament new = {
3515 .dir = new_dir,
3516 .dentry = new_dentry,
3517 .inode = d_inode(new_dentry),
3518 };
3519 int force_reread;
3520 int retval;
3521 struct inode *whiteout = NULL;
3522 int credits;
3523 u8 old_file_type;
3524
3525 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3526 (!projid_eq(EXT4_I(new_dir)->i_projid,
3527 EXT4_I(old_dentry->d_inode)->i_projid)))
3528 return -EXDEV;
3529
3530 retval = dquot_initialize(old.dir);
3531 if (retval)
3532 return retval;
3533 retval = dquot_initialize(new.dir);
3534 if (retval)
3535 return retval;
3536
3537 /* Initialize quotas before so that eventual writes go
3538 * in separate transaction */
3539 if (new.inode) {
3540 retval = dquot_initialize(new.inode);
3541 if (retval)
3542 return retval;
3543 }
3544
3545 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3546 if (IS_ERR(old.bh))
3547 return PTR_ERR(old.bh);
3548 /*
3549 * Check for inode number is _not_ due to possible IO errors.
3550 * We might rmdir the source, keep it as pwd of some process
3551 * and merrily kill the link to whatever was created under the
3552 * same name. Goodbye sticky bit ;-<
3553 */
3554 retval = -ENOENT;
3555 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3556 goto end_rename;
3557
3558 if ((old.dir != new.dir) &&
3559 ext4_encrypted_inode(new.dir) &&
3560 !fscrypt_has_permitted_context(new.dir, old.inode)) {
3561 retval = -EPERM;
3562 goto end_rename;
3563 }
3564
3565 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3566 &new.de, &new.inlined);
3567 if (IS_ERR(new.bh)) {
3568 retval = PTR_ERR(new.bh);
3569 new.bh = NULL;
3570 goto end_rename;
3571 }
3572 if (new.bh) {
3573 if (!new.inode) {
3574 brelse(new.bh);
3575 new.bh = NULL;
3576 }
3577 }
3578 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3579 ext4_alloc_da_blocks(old.inode);
3580
3581 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3582 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3583 if (!(flags & RENAME_WHITEOUT)) {
3584 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3585 if (IS_ERR(handle)) {
3586 retval = PTR_ERR(handle);
3587 handle = NULL;
3588 goto end_rename;
3589 }
3590 } else {
3591 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3592 if (IS_ERR(whiteout)) {
3593 retval = PTR_ERR(whiteout);
3594 whiteout = NULL;
3595 goto end_rename;
3596 }
3597 }
3598
3599 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3600 ext4_handle_sync(handle);
3601
3602 if (S_ISDIR(old.inode->i_mode)) {
3603 if (new.inode) {
3604 retval = -ENOTEMPTY;
3605 if (!ext4_empty_dir(new.inode))
3606 goto end_rename;
3607 } else {
3608 retval = -EMLINK;
3609 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3610 goto end_rename;
3611 }
3612 retval = ext4_rename_dir_prepare(handle, &old);
3613 if (retval)
3614 goto end_rename;
3615 }
3616 /*
3617 * If we're renaming a file within an inline_data dir and adding or
3618 * setting the new dirent causes a conversion from inline_data to
3619 * extents/blockmap, we need to force the dirent delete code to
3620 * re-read the directory, or else we end up trying to delete a dirent
3621 * from what is now the extent tree root (or a block map).
3622 */
3623 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3624 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3625
3626 old_file_type = old.de->file_type;
3627 if (whiteout) {
3628 /*
3629 * Do this before adding a new entry, so the old entry is sure
3630 * to be still pointing to the valid old entry.
3631 */
3632 retval = ext4_setent(handle, &old, whiteout->i_ino,
3633 EXT4_FT_CHRDEV);
3634 if (retval)
3635 goto end_rename;
3636 ext4_mark_inode_dirty(handle, whiteout);
3637 }
3638 if (!new.bh) {
3639 retval = ext4_add_entry(handle, new.dentry, old.inode);
3640 if (retval)
3641 goto end_rename;
3642 } else {
3643 retval = ext4_setent(handle, &new,
3644 old.inode->i_ino, old_file_type);
3645 if (retval)
3646 goto end_rename;
3647 }
3648 if (force_reread)
3649 force_reread = !ext4_test_inode_flag(new.dir,
3650 EXT4_INODE_INLINE_DATA);
3651
3652 /*
3653 * Like most other Unix systems, set the ctime for inodes on a
3654 * rename.
3655 */
3656 old.inode->i_ctime = ext4_current_time(old.inode);
3657 ext4_mark_inode_dirty(handle, old.inode);
3658
3659 if (!whiteout) {
3660 /*
3661 * ok, that's it
3662 */
3663 ext4_rename_delete(handle, &old, force_reread);
3664 }
3665
3666 if (new.inode) {
3667 ext4_dec_count(handle, new.inode);
3668 new.inode->i_ctime = ext4_current_time(new.inode);
3669 }
3670 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3671 ext4_update_dx_flag(old.dir);
3672 if (old.dir_bh) {
3673 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3674 if (retval)
3675 goto end_rename;
3676
3677 ext4_dec_count(handle, old.dir);
3678 if (new.inode) {
3679 /* checked ext4_empty_dir above, can't have another
3680 * parent, ext4_dec_count() won't work for many-linked
3681 * dirs */
3682 clear_nlink(new.inode);
3683 } else {
3684 ext4_inc_count(handle, new.dir);
3685 ext4_update_dx_flag(new.dir);
3686 ext4_mark_inode_dirty(handle, new.dir);
3687 }
3688 }
3689 ext4_mark_inode_dirty(handle, old.dir);
3690 if (new.inode) {
3691 ext4_mark_inode_dirty(handle, new.inode);
3692 if (!new.inode->i_nlink)
3693 ext4_orphan_add(handle, new.inode);
3694 }
3695 retval = 0;
3696
3697 end_rename:
3698 brelse(old.dir_bh);
3699 brelse(old.bh);
3700 brelse(new.bh);
3701 if (whiteout) {
3702 if (retval)
3703 drop_nlink(whiteout);
3704 unlock_new_inode(whiteout);
3705 iput(whiteout);
3706 }
3707 if (handle)
3708 ext4_journal_stop(handle);
3709 return retval;
3710 }
3711
3712 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3713 struct inode *new_dir, struct dentry *new_dentry)
3714 {
3715 handle_t *handle = NULL;
3716 struct ext4_renament old = {
3717 .dir = old_dir,
3718 .dentry = old_dentry,
3719 .inode = d_inode(old_dentry),
3720 };
3721 struct ext4_renament new = {
3722 .dir = new_dir,
3723 .dentry = new_dentry,
3724 .inode = d_inode(new_dentry),
3725 };
3726 u8 new_file_type;
3727 int retval;
3728
3729 if ((ext4_encrypted_inode(old_dir) ||
3730 ext4_encrypted_inode(new_dir)) &&
3731 (old_dir != new_dir) &&
3732 (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3733 !fscrypt_has_permitted_context(old_dir, new.inode)))
3734 return -EPERM;
3735
3736 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3737 !projid_eq(EXT4_I(new_dir)->i_projid,
3738 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3739 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3740 !projid_eq(EXT4_I(old_dir)->i_projid,
3741 EXT4_I(new_dentry->d_inode)->i_projid)))
3742 return -EXDEV;
3743
3744 retval = dquot_initialize(old.dir);
3745 if (retval)
3746 return retval;
3747 retval = dquot_initialize(new.dir);
3748 if (retval)
3749 return retval;
3750
3751 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3752 &old.de, &old.inlined);
3753 if (IS_ERR(old.bh))
3754 return PTR_ERR(old.bh);
3755 /*
3756 * Check for inode number is _not_ due to possible IO errors.
3757 * We might rmdir the source, keep it as pwd of some process
3758 * and merrily kill the link to whatever was created under the
3759 * same name. Goodbye sticky bit ;-<
3760 */
3761 retval = -ENOENT;
3762 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3763 goto end_rename;
3764
3765 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3766 &new.de, &new.inlined);
3767 if (IS_ERR(new.bh)) {
3768 retval = PTR_ERR(new.bh);
3769 new.bh = NULL;
3770 goto end_rename;
3771 }
3772
3773 /* RENAME_EXCHANGE case: old *and* new must both exist */
3774 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3775 goto end_rename;
3776
3777 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3778 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3779 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3780 if (IS_ERR(handle)) {
3781 retval = PTR_ERR(handle);
3782 handle = NULL;
3783 goto end_rename;
3784 }
3785
3786 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3787 ext4_handle_sync(handle);
3788
3789 if (S_ISDIR(old.inode->i_mode)) {
3790 old.is_dir = true;
3791 retval = ext4_rename_dir_prepare(handle, &old);
3792 if (retval)
3793 goto end_rename;
3794 }
3795 if (S_ISDIR(new.inode->i_mode)) {
3796 new.is_dir = true;
3797 retval = ext4_rename_dir_prepare(handle, &new);
3798 if (retval)
3799 goto end_rename;
3800 }
3801
3802 /*
3803 * Other than the special case of overwriting a directory, parents'
3804 * nlink only needs to be modified if this is a cross directory rename.
3805 */
3806 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3807 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3808 new.dir_nlink_delta = -old.dir_nlink_delta;
3809 retval = -EMLINK;
3810 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3811 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3812 goto end_rename;
3813 }
3814
3815 new_file_type = new.de->file_type;
3816 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3817 if (retval)
3818 goto end_rename;
3819
3820 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3821 if (retval)
3822 goto end_rename;
3823
3824 /*
3825 * Like most other Unix systems, set the ctime for inodes on a
3826 * rename.
3827 */
3828 old.inode->i_ctime = ext4_current_time(old.inode);
3829 new.inode->i_ctime = ext4_current_time(new.inode);
3830 ext4_mark_inode_dirty(handle, old.inode);
3831 ext4_mark_inode_dirty(handle, new.inode);
3832
3833 if (old.dir_bh) {
3834 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3835 if (retval)
3836 goto end_rename;
3837 }
3838 if (new.dir_bh) {
3839 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3840 if (retval)
3841 goto end_rename;
3842 }
3843 ext4_update_dir_count(handle, &old);
3844 ext4_update_dir_count(handle, &new);
3845 retval = 0;
3846
3847 end_rename:
3848 brelse(old.dir_bh);
3849 brelse(new.dir_bh);
3850 brelse(old.bh);
3851 brelse(new.bh);
3852 if (handle)
3853 ext4_journal_stop(handle);
3854 return retval;
3855 }
3856
3857 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3858 struct inode *new_dir, struct dentry *new_dentry,
3859 unsigned int flags)
3860 {
3861 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3862 return -EINVAL;
3863
3864 if (flags & RENAME_EXCHANGE) {
3865 return ext4_cross_rename(old_dir, old_dentry,
3866 new_dir, new_dentry);
3867 }
3868
3869 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3870 }
3871
3872 /*
3873 * directories can handle most operations...
3874 */
3875 const struct inode_operations ext4_dir_inode_operations = {
3876 .create = ext4_create,
3877 .lookup = ext4_lookup,
3878 .link = ext4_link,
3879 .unlink = ext4_unlink,
3880 .symlink = ext4_symlink,
3881 .mkdir = ext4_mkdir,
3882 .rmdir = ext4_rmdir,
3883 .mknod = ext4_mknod,
3884 .tmpfile = ext4_tmpfile,
3885 .rename = ext4_rename2,
3886 .setattr = ext4_setattr,
3887 .listxattr = ext4_listxattr,
3888 .get_acl = ext4_get_acl,
3889 .set_acl = ext4_set_acl,
3890 .fiemap = ext4_fiemap,
3891 };
3892
3893 const struct inode_operations ext4_special_inode_operations = {
3894 .setattr = ext4_setattr,
3895 .listxattr = ext4_listxattr,
3896 .get_acl = ext4_get_acl,
3897 .set_acl = ext4_set_acl,
3898 };
Linux with added FPC-III support