?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
wlcore: Add RX_BA_WIN_SIZE_CHANGE_EVENT event
[linux/fpc-iii.git] / fs / ext4 / namei.c
blobfafa903ab3c0f2a8b32816b44408225f173a08e7
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 dentry *dentry, 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 1;
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 1;
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("%x->%lu ", i ? dx_get_hash(entries + i) :
583 0, (unsigned long)dx_get_block(entries + i));
584 }
585 printk("\n");
586 }
587
588 struct stats
589 {
590 unsigned names;
591 unsigned space;
592 unsigned bcount;
593 };
594
595 static struct stats dx_show_leaf(struct inode *dir,
596 struct dx_hash_info *hinfo,
597 struct ext4_dir_entry_2 *de,
598 int size, int show_names)
599 {
600 unsigned names = 0, space = 0;
601 char *base = (char *) de;
602 struct dx_hash_info h = *hinfo;
603
604 printk("names: ");
605 while ((char *) de < base + size)
606 {
607 if (de->inode)
608 {
609 if (show_names)
610 {
611 #ifdef CONFIG_EXT4_FS_ENCRYPTION
612 int len;
613 char *name;
614 struct ext4_str fname_crypto_str
615 = {.name = NULL, .len = 0};
616 int res = 0;
617
618 name = de->name;
619 len = de->name_len;
620 if (ext4_encrypted_inode(inode))
621 res = ext4_get_encryption_info(dir);
622 if (res) {
623 printk(KERN_WARNING "Error setting up"
624 " fname crypto: %d\n", res);
625 }
626 if (ctx == NULL) {
627 /* Directory is not encrypted */
628 ext4fs_dirhash(de->name,
629 de->name_len, &h);
630 printk("%*.s:(U)%x.%u ", len,
631 name, h.hash,
632 (unsigned) ((char *) de
633 - base));
634 } else {
635 /* Directory is encrypted */
636 res = ext4_fname_crypto_alloc_buffer(
637 ctx, de->name_len,
638 &fname_crypto_str);
639 if (res < 0) {
640 printk(KERN_WARNING "Error "
641 "allocating crypto "
642 "buffer--skipping "
643 "crypto\n");
644 ctx = NULL;
645 }
646 res = ext4_fname_disk_to_usr(ctx, NULL, de,
647 &fname_crypto_str);
648 if (res < 0) {
649 printk(KERN_WARNING "Error "
650 "converting filename "
651 "from disk to usr"
652 "\n");
653 name = "??";
654 len = 2;
655 } else {
656 name = fname_crypto_str.name;
657 len = fname_crypto_str.len;
658 }
659 ext4fs_dirhash(de->name, de->name_len,
660 &h);
661 printk("%*.s:(E)%x.%u ", len, name,
662 h.hash, (unsigned) ((char *) de
663 - base));
664 ext4_fname_crypto_free_buffer(
665 &fname_crypto_str);
666 }
667 #else
668 int len = de->name_len;
669 char *name = de->name;
670 ext4fs_dirhash(de->name, de->name_len, &h);
671 printk("%*.s:%x.%u ", len, name, h.hash,
672 (unsigned) ((char *) de - base));
673 #endif
674 }
675 space += EXT4_DIR_REC_LEN(de->name_len);
676 names++;
677 }
678 de = ext4_next_entry(de, size);
679 }
680 printk("(%i)\n", names);
681 return (struct stats) { names, space, 1 };
682 }
683
684 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
685 struct dx_entry *entries, int levels)
686 {
687 unsigned blocksize = dir->i_sb->s_blocksize;
688 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
689 unsigned bcount = 0;
690 struct buffer_head *bh;
691 printk("%i indexed blocks...\n", count);
692 for (i = 0; i < count; i++, entries++)
693 {
694 ext4_lblk_t block = dx_get_block(entries);
695 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
696 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
697 struct stats stats;
698 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
699 bh = ext4_bread(NULL,dir, block, 0);
700 if (!bh || IS_ERR(bh))
701 continue;
702 stats = levels?
703 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
704 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
705 bh->b_data, blocksize, 0);
706 names += stats.names;
707 space += stats.space;
708 bcount += stats.bcount;
709 brelse(bh);
710 }
711 if (bcount)
712 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
713 levels ? "" : " ", names, space/bcount,
714 (space/bcount)*100/blocksize);
715 return (struct stats) { names, space, bcount};
716 }
717 #endif /* DX_DEBUG */
718
719 /*
720 * Probe for a directory leaf block to search.
721 *
722 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
723 * error in the directory index, and the caller should fall back to
724 * searching the directory normally. The callers of dx_probe **MUST**
725 * check for this error code, and make sure it never gets reflected
726 * back to userspace.
727 */
728 static struct dx_frame *
729 dx_probe(struct ext4_filename *fname, struct inode *dir,
730 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
731 {
732 unsigned count, indirect;
733 struct dx_entry *at, *entries, *p, *q, *m;
734 struct dx_root *root;
735 struct dx_frame *frame = frame_in;
736 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
737 u32 hash;
738
739 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
740 if (IS_ERR(frame->bh))
741 return (struct dx_frame *) frame->bh;
742
743 root = (struct dx_root *) frame->bh->b_data;
744 if (root->info.hash_version != DX_HASH_TEA &&
745 root->info.hash_version != DX_HASH_HALF_MD4 &&
746 root->info.hash_version != DX_HASH_LEGACY) {
747 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
748 root->info.hash_version);
749 goto fail;
750 }
751 if (fname)
752 hinfo = &fname->hinfo;
753 hinfo->hash_version = root->info.hash_version;
754 if (hinfo->hash_version <= DX_HASH_TEA)
755 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
756 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
757 if (fname && fname_name(fname))
758 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
759 hash = hinfo->hash;
760
761 if (root->info.unused_flags & 1) {
762 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
763 root->info.unused_flags);
764 goto fail;
765 }
766
767 indirect = root->info.indirect_levels;
768 if (indirect > 1) {
769 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
770 root->info.indirect_levels);
771 goto fail;
772 }
773
774 entries = (struct dx_entry *)(((char *)&root->info) +
775 root->info.info_length);
776
777 if (dx_get_limit(entries) != dx_root_limit(dir,
778 root->info.info_length)) {
779 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
780 dx_get_limit(entries),
781 dx_root_limit(dir, root->info.info_length));
782 goto fail;
783 }
784
785 dxtrace(printk("Look up %x", hash));
786 while (1) {
787 count = dx_get_count(entries);
788 if (!count || count > dx_get_limit(entries)) {
789 ext4_warning_inode(dir,
790 "dx entry: count %u beyond limit %u",
791 count, dx_get_limit(entries));
792 goto fail;
793 }
794
795 p = entries + 1;
796 q = entries + count - 1;
797 while (p <= q) {
798 m = p + (q - p) / 2;
799 dxtrace(printk("."));
800 if (dx_get_hash(m) > hash)
801 q = m - 1;
802 else
803 p = m + 1;
804 }
805
806 if (0) { // linear search cross check
807 unsigned n = count - 1;
808 at = entries;
809 while (n--)
810 {
811 dxtrace(printk(","));
812 if (dx_get_hash(++at) > hash)
813 {
814 at--;
815 break;
816 }
817 }
818 assert (at == p - 1);
819 }
820
821 at = p - 1;
822 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
823 dx_get_block(at)));
824 frame->entries = entries;
825 frame->at = at;
826 if (!indirect--)
827 return frame;
828 frame++;
829 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
830 if (IS_ERR(frame->bh)) {
831 ret_err = (struct dx_frame *) frame->bh;
832 frame->bh = NULL;
833 goto fail;
834 }
835 entries = ((struct dx_node *) frame->bh->b_data)->entries;
836
837 if (dx_get_limit(entries) != dx_node_limit(dir)) {
838 ext4_warning_inode(dir,
839 "dx entry: limit %u != node limit %u",
840 dx_get_limit(entries), dx_node_limit(dir));
841 goto fail;
842 }
843 }
844 fail:
845 while (frame >= frame_in) {
846 brelse(frame->bh);
847 frame--;
848 }
849
850 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
851 ext4_warning_inode(dir,
852 "Corrupt directory, running e2fsck is recommended");
853 return ret_err;
854 }
855
856 static void dx_release(struct dx_frame *frames)
857 {
858 if (frames[0].bh == NULL)
859 return;
860
861 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
862 brelse(frames[1].bh);
863 brelse(frames[0].bh);
864 }
865
866 /*
867 * This function increments the frame pointer to search the next leaf
868 * block, and reads in the necessary intervening nodes if the search
869 * should be necessary. Whether or not the search is necessary is
870 * controlled by the hash parameter. If the hash value is even, then
871 * the search is only continued if the next block starts with that
872 * hash value. This is used if we are searching for a specific file.
873 *
874 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
875 *
876 * This function returns 1 if the caller should continue to search,
877 * or 0 if it should not. If there is an error reading one of the
878 * index blocks, it will a negative error code.
879 *
880 * If start_hash is non-null, it will be filled in with the starting
881 * hash of the next page.
882 */
883 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
884 struct dx_frame *frame,
885 struct dx_frame *frames,
886 __u32 *start_hash)
887 {
888 struct dx_frame *p;
889 struct buffer_head *bh;
890 int num_frames = 0;
891 __u32 bhash;
892
893 p = frame;
894 /*
895 * Find the next leaf page by incrementing the frame pointer.
896 * If we run out of entries in the interior node, loop around and
897 * increment pointer in the parent node. When we break out of
898 * this loop, num_frames indicates the number of interior
899 * nodes need to be read.
900 */
901 while (1) {
902 if (++(p->at) < p->entries + dx_get_count(p->entries))
903 break;
904 if (p == frames)
905 return 0;
906 num_frames++;
907 p--;
908 }
909
910 /*
911 * If the hash is 1, then continue only if the next page has a
912 * continuation hash of any value. This is used for readdir
913 * handling. Otherwise, check to see if the hash matches the
914 * desired contiuation hash. If it doesn't, return since
915 * there's no point to read in the successive index pages.
916 */
917 bhash = dx_get_hash(p->at);
918 if (start_hash)
919 *start_hash = bhash;
920 if ((hash & 1) == 0) {
921 if ((bhash & ~1) != hash)
922 return 0;
923 }
924 /*
925 * If the hash is HASH_NB_ALWAYS, we always go to the next
926 * block so no check is necessary
927 */
928 while (num_frames--) {
929 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
930 if (IS_ERR(bh))
931 return PTR_ERR(bh);
932 p++;
933 brelse(p->bh);
934 p->bh = bh;
935 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
936 }
937 return 1;
938 }
939
940
941 /*
942 * This function fills a red-black tree with information from a
943 * directory block. It returns the number directory entries loaded
944 * into the tree. If there is an error it is returned in err.
945 */
946 static int htree_dirblock_to_tree(struct file *dir_file,
947 struct inode *dir, ext4_lblk_t block,
948 struct dx_hash_info *hinfo,
949 __u32 start_hash, __u32 start_minor_hash)
950 {
951 struct buffer_head *bh;
952 struct ext4_dir_entry_2 *de, *top;
953 int err = 0, count = 0;
954 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
955
956 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
957 (unsigned long)block));
958 bh = ext4_read_dirblock(dir, block, DIRENT);
959 if (IS_ERR(bh))
960 return PTR_ERR(bh);
961
962 de = (struct ext4_dir_entry_2 *) bh->b_data;
963 top = (struct ext4_dir_entry_2 *) ((char *) de +
964 dir->i_sb->s_blocksize -
965 EXT4_DIR_REC_LEN(0));
966 #ifdef CONFIG_EXT4_FS_ENCRYPTION
967 /* Check if the directory is encrypted */
968 if (ext4_encrypted_inode(dir)) {
969 err = ext4_get_encryption_info(dir);
970 if (err < 0) {
971 brelse(bh);
972 return err;
973 }
974 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
975 &fname_crypto_str);
976 if (err < 0) {
977 brelse(bh);
978 return err;
979 }
980 }
981 #endif
982 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
983 if (ext4_check_dir_entry(dir, NULL, de, bh,
984 bh->b_data, bh->b_size,
985 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
986 + ((char *)de - bh->b_data))) {
987 /* silently ignore the rest of the block */
988 break;
989 }
990 ext4fs_dirhash(de->name, de->name_len, hinfo);
991 if ((hinfo->hash < start_hash) ||
992 ((hinfo->hash == start_hash) &&
993 (hinfo->minor_hash < start_minor_hash)))
994 continue;
995 if (de->inode == 0)
996 continue;
997 if (!ext4_encrypted_inode(dir)) {
998 tmp_str.name = de->name;
999 tmp_str.len = de->name_len;
1000 err = ext4_htree_store_dirent(dir_file,
1001 hinfo->hash, hinfo->minor_hash, de,
1002 &tmp_str);
1003 } else {
1004 int save_len = fname_crypto_str.len;
1005
1006 /* Directory is encrypted */
1007 err = ext4_fname_disk_to_usr(dir, hinfo, de,
1008 &fname_crypto_str);
1009 if (err < 0) {
1010 count = err;
1011 goto errout;
1012 }
1013 err = ext4_htree_store_dirent(dir_file,
1014 hinfo->hash, hinfo->minor_hash, de,
1015 &fname_crypto_str);
1016 fname_crypto_str.len = save_len;
1017 }
1018 if (err != 0) {
1019 count = err;
1020 goto errout;
1021 }
1022 count++;
1023 }
1024 errout:
1025 brelse(bh);
1026 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1027 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1028 #endif
1029 return count;
1030 }
1031
1032
1033 /*
1034 * This function fills a red-black tree with information from a
1035 * directory. We start scanning the directory in hash order, starting
1036 * at start_hash and start_minor_hash.
1037 *
1038 * This function returns the number of entries inserted into the tree,
1039 * or a negative error code.
1040 */
1041 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1042 __u32 start_minor_hash, __u32 *next_hash)
1043 {
1044 struct dx_hash_info hinfo;
1045 struct ext4_dir_entry_2 *de;
1046 struct dx_frame frames[2], *frame;
1047 struct inode *dir;
1048 ext4_lblk_t block;
1049 int count = 0;
1050 int ret, err;
1051 __u32 hashval;
1052 struct ext4_str tmp_str;
1053
1054 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1055 start_hash, start_minor_hash));
1056 dir = file_inode(dir_file);
1057 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1058 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1059 if (hinfo.hash_version <= DX_HASH_TEA)
1060 hinfo.hash_version +=
1061 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1062 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1063 if (ext4_has_inline_data(dir)) {
1064 int has_inline_data = 1;
1065 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1066 &hinfo, start_hash,
1067 start_minor_hash,
1068 &has_inline_data);
1069 if (has_inline_data) {
1070 *next_hash = ~0;
1071 return count;
1072 }
1073 }
1074 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1075 start_hash, start_minor_hash);
1076 *next_hash = ~0;
1077 return count;
1078 }
1079 hinfo.hash = start_hash;
1080 hinfo.minor_hash = 0;
1081 frame = dx_probe(NULL, dir, &hinfo, frames);
1082 if (IS_ERR(frame))
1083 return PTR_ERR(frame);
1084
1085 /* Add '.' and '..' from the htree header */
1086 if (!start_hash && !start_minor_hash) {
1087 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1088 tmp_str.name = de->name;
1089 tmp_str.len = de->name_len;
1090 err = ext4_htree_store_dirent(dir_file, 0, 0,
1091 de, &tmp_str);
1092 if (err != 0)
1093 goto errout;
1094 count++;
1095 }
1096 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1097 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1098 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1099 tmp_str.name = de->name;
1100 tmp_str.len = de->name_len;
1101 err = ext4_htree_store_dirent(dir_file, 2, 0,
1102 de, &tmp_str);
1103 if (err != 0)
1104 goto errout;
1105 count++;
1106 }
1107
1108 while (1) {
1109 block = dx_get_block(frame->at);
1110 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1111 start_hash, start_minor_hash);
1112 if (ret < 0) {
1113 err = ret;
1114 goto errout;
1115 }
1116 count += ret;
1117 hashval = ~0;
1118 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1119 frame, frames, &hashval);
1120 *next_hash = hashval;
1121 if (ret < 0) {
1122 err = ret;
1123 goto errout;
1124 }
1125 /*
1126 * Stop if: (a) there are no more entries, or
1127 * (b) we have inserted at least one entry and the
1128 * next hash value is not a continuation
1129 */
1130 if ((ret == 0) ||
1131 (count && ((hashval & 1) == 0)))
1132 break;
1133 }
1134 dx_release(frames);
1135 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1136 "next hash: %x\n", count, *next_hash));
1137 return count;
1138 errout:
1139 dx_release(frames);
1140 return (err);
1141 }
1142
1143 static inline int search_dirblock(struct buffer_head *bh,
1144 struct inode *dir,
1145 struct ext4_filename *fname,
1146 const struct qstr *d_name,
1147 unsigned int offset,
1148 struct ext4_dir_entry_2 **res_dir)
1149 {
1150 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1151 fname, d_name, offset, res_dir);
1152 }
1153
1154 /*
1155 * Directory block splitting, compacting
1156 */
1157
1158 /*
1159 * Create map of hash values, offsets, and sizes, stored at end of block.
1160 * Returns number of entries mapped.
1161 */
1162 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1163 unsigned blocksize, struct dx_hash_info *hinfo,
1164 struct dx_map_entry *map_tail)
1165 {
1166 int count = 0;
1167 char *base = (char *) de;
1168 struct dx_hash_info h = *hinfo;
1169
1170 while ((char *) de < base + blocksize) {
1171 if (de->name_len && de->inode) {
1172 ext4fs_dirhash(de->name, de->name_len, &h);
1173 map_tail--;
1174 map_tail->hash = h.hash;
1175 map_tail->offs = ((char *) de - base)>>2;
1176 map_tail->size = le16_to_cpu(de->rec_len);
1177 count++;
1178 cond_resched();
1179 }
1180 /* XXX: do we need to check rec_len == 0 case? -Chris */
1181 de = ext4_next_entry(de, blocksize);
1182 }
1183 return count;
1184 }
1185
1186 /* Sort map by hash value */
1187 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1188 {
1189 struct dx_map_entry *p, *q, *top = map + count - 1;
1190 int more;
1191 /* Combsort until bubble sort doesn't suck */
1192 while (count > 2) {
1193 count = count*10/13;
1194 if (count - 9 < 2) /* 9, 10 -> 11 */
1195 count = 11;
1196 for (p = top, q = p - count; q >= map; p--, q--)
1197 if (p->hash < q->hash)
1198 swap(*p, *q);
1199 }
1200 /* Garden variety bubble sort */
1201 do {
1202 more = 0;
1203 q = top;
1204 while (q-- > map) {
1205 if (q[1].hash >= q[0].hash)
1206 continue;
1207 swap(*(q+1), *q);
1208 more = 1;
1209 }
1210 } while(more);
1211 }
1212
1213 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1214 {
1215 struct dx_entry *entries = frame->entries;
1216 struct dx_entry *old = frame->at, *new = old + 1;
1217 int count = dx_get_count(entries);
1218
1219 assert(count < dx_get_limit(entries));
1220 assert(old < entries + count);
1221 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1222 dx_set_hash(new, hash);
1223 dx_set_block(new, block);
1224 dx_set_count(entries, count + 1);
1225 }
1226
1227 /*
1228 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1229 *
1230 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1231 * `de != NULL' is guaranteed by caller.
1232 */
1233 static inline int ext4_match(struct ext4_filename *fname,
1234 struct ext4_dir_entry_2 *de)
1235 {
1236 const void *name = fname_name(fname);
1237 u32 len = fname_len(fname);
1238
1239 if (!de->inode)
1240 return 0;
1241
1242 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1243 if (unlikely(!name)) {
1244 if (fname->usr_fname->name[0] == '_') {
1245 int ret;
1246 if (de->name_len < 16)
1247 return 0;
1248 ret = memcmp(de->name + de->name_len - 16,
1249 fname->crypto_buf.name + 8, 16);
1250 return (ret == 0) ? 1 : 0;
1251 }
1252 name = fname->crypto_buf.name;
1253 len = fname->crypto_buf.len;
1254 }
1255 #endif
1256 if (de->name_len != len)
1257 return 0;
1258 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1259 }
1260
1261 /*
1262 * Returns 0 if not found, -1 on failure, and 1 on success
1263 */
1264 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1265 struct inode *dir, struct ext4_filename *fname,
1266 const struct qstr *d_name,
1267 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1268 {
1269 struct ext4_dir_entry_2 * de;
1270 char * dlimit;
1271 int de_len;
1272 int res;
1273
1274 de = (struct ext4_dir_entry_2 *)search_buf;
1275 dlimit = search_buf + buf_size;
1276 while ((char *) de < dlimit) {
1277 /* this code is executed quadratically often */
1278 /* do minimal checking `by hand' */
1279 if ((char *) de + de->name_len <= dlimit) {
1280 res = ext4_match(fname, de);
1281 if (res < 0) {
1282 res = -1;
1283 goto return_result;
1284 }
1285 if (res > 0) {
1286 /* found a match - just to be sure, do
1287 * a full check */
1288 if (ext4_check_dir_entry(dir, NULL, de, bh,
1289 bh->b_data,
1290 bh->b_size, offset)) {
1291 res = -1;
1292 goto return_result;
1293 }
1294 *res_dir = de;
1295 res = 1;
1296 goto return_result;
1297 }
1298
1299 }
1300 /* prevent looping on a bad block */
1301 de_len = ext4_rec_len_from_disk(de->rec_len,
1302 dir->i_sb->s_blocksize);
1303 if (de_len <= 0) {
1304 res = -1;
1305 goto return_result;
1306 }
1307 offset += de_len;
1308 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1309 }
1310
1311 res = 0;
1312 return_result:
1313 return res;
1314 }
1315
1316 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1317 struct ext4_dir_entry *de)
1318 {
1319 struct super_block *sb = dir->i_sb;
1320
1321 if (!is_dx(dir))
1322 return 0;
1323 if (block == 0)
1324 return 1;
1325 if (de->inode == 0 &&
1326 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1327 sb->s_blocksize)
1328 return 1;
1329 return 0;
1330 }
1331
1332 /*
1333 * ext4_find_entry()
1334 *
1335 * finds an entry in the specified directory with the wanted name. It
1336 * returns the cache buffer in which the entry was found, and the entry
1337 * itself (as a parameter - res_dir). It does NOT read the inode of the
1338 * entry - you'll have to do that yourself if you want to.
1339 *
1340 * The returned buffer_head has ->b_count elevated. The caller is expected
1341 * to brelse() it when appropriate.
1342 */
1343 static struct buffer_head * ext4_find_entry (struct inode *dir,
1344 const struct qstr *d_name,
1345 struct ext4_dir_entry_2 **res_dir,
1346 int *inlined)
1347 {
1348 struct super_block *sb;
1349 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1350 struct buffer_head *bh, *ret = NULL;
1351 ext4_lblk_t start, block, b;
1352 const u8 *name = d_name->name;
1353 int ra_max = 0; /* Number of bh's in the readahead
1354 buffer, bh_use[] */
1355 int ra_ptr = 0; /* Current index into readahead
1356 buffer */
1357 int num = 0;
1358 ext4_lblk_t nblocks;
1359 int i, namelen, retval;
1360 struct ext4_filename fname;
1361
1362 *res_dir = NULL;
1363 sb = dir->i_sb;
1364 namelen = d_name->len;
1365 if (namelen > EXT4_NAME_LEN)
1366 return NULL;
1367
1368 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1369 if (retval)
1370 return ERR_PTR(retval);
1371
1372 if (ext4_has_inline_data(dir)) {
1373 int has_inline_data = 1;
1374 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1375 &has_inline_data);
1376 if (has_inline_data) {
1377 if (inlined)
1378 *inlined = 1;
1379 goto cleanup_and_exit;
1380 }
1381 }
1382
1383 if ((namelen <= 2) && (name[0] == '.') &&
1384 (name[1] == '.' || name[1] == '\0')) {
1385 /*
1386 * "." or ".." will only be in the first block
1387 * NFS may look up ".."; "." should be handled by the VFS
1388 */
1389 block = start = 0;
1390 nblocks = 1;
1391 goto restart;
1392 }
1393 if (is_dx(dir)) {
1394 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1395 /*
1396 * On success, or if the error was file not found,
1397 * return. Otherwise, fall back to doing a search the
1398 * old fashioned way.
1399 */
1400 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1401 goto cleanup_and_exit;
1402 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1403 "falling back\n"));
1404 }
1405 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1406 start = EXT4_I(dir)->i_dir_start_lookup;
1407 if (start >= nblocks)
1408 start = 0;
1409 block = start;
1410 restart:
1411 do {
1412 /*
1413 * We deal with the read-ahead logic here.
1414 */
1415 if (ra_ptr >= ra_max) {
1416 /* Refill the readahead buffer */
1417 ra_ptr = 0;
1418 b = block;
1419 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1420 /*
1421 * Terminate if we reach the end of the
1422 * directory and must wrap, or if our
1423 * search has finished at this block.
1424 */
1425 if (b >= nblocks || (num && block == start)) {
1426 bh_use[ra_max] = NULL;
1427 break;
1428 }
1429 num++;
1430 bh = ext4_getblk(NULL, dir, b++, 0);
1431 if (IS_ERR(bh)) {
1432 if (ra_max == 0) {
1433 ret = bh;
1434 goto cleanup_and_exit;
1435 }
1436 break;
1437 }
1438 bh_use[ra_max] = bh;
1439 if (bh)
1440 ll_rw_block(READ | REQ_META | REQ_PRIO,
1441 1, &bh);
1442 }
1443 }
1444 if ((bh = bh_use[ra_ptr++]) == NULL)
1445 goto next;
1446 wait_on_buffer(bh);
1447 if (!buffer_uptodate(bh)) {
1448 /* read error, skip block & hope for the best */
1449 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1450 (unsigned long) block);
1451 brelse(bh);
1452 goto next;
1453 }
1454 if (!buffer_verified(bh) &&
1455 !is_dx_internal_node(dir, block,
1456 (struct ext4_dir_entry *)bh->b_data) &&
1457 !ext4_dirent_csum_verify(dir,
1458 (struct ext4_dir_entry *)bh->b_data)) {
1459 EXT4_ERROR_INODE(dir, "checksumming directory "
1460 "block %lu", (unsigned long)block);
1461 brelse(bh);
1462 goto next;
1463 }
1464 set_buffer_verified(bh);
1465 i = search_dirblock(bh, dir, &fname, d_name,
1466 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1467 if (i == 1) {
1468 EXT4_I(dir)->i_dir_start_lookup = block;
1469 ret = bh;
1470 goto cleanup_and_exit;
1471 } else {
1472 brelse(bh);
1473 if (i < 0)
1474 goto cleanup_and_exit;
1475 }
1476 next:
1477 if (++block >= nblocks)
1478 block = 0;
1479 } while (block != start);
1480
1481 /*
1482 * If the directory has grown while we were searching, then
1483 * search the last part of the directory before giving up.
1484 */
1485 block = nblocks;
1486 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1487 if (block < nblocks) {
1488 start = 0;
1489 goto restart;
1490 }
1491
1492 cleanup_and_exit:
1493 /* Clean up the read-ahead blocks */
1494 for (; ra_ptr < ra_max; ra_ptr++)
1495 brelse(bh_use[ra_ptr]);
1496 ext4_fname_free_filename(&fname);
1497 return ret;
1498 }
1499
1500 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1501 struct ext4_filename *fname,
1502 struct ext4_dir_entry_2 **res_dir)
1503 {
1504 struct super_block * sb = dir->i_sb;
1505 struct dx_frame frames[2], *frame;
1506 const struct qstr *d_name = fname->usr_fname;
1507 struct buffer_head *bh;
1508 ext4_lblk_t block;
1509 int retval;
1510
1511 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1512 *res_dir = NULL;
1513 #endif
1514 frame = dx_probe(fname, dir, NULL, frames);
1515 if (IS_ERR(frame))
1516 return (struct buffer_head *) frame;
1517 do {
1518 block = dx_get_block(frame->at);
1519 bh = ext4_read_dirblock(dir, block, DIRENT);
1520 if (IS_ERR(bh))
1521 goto errout;
1522
1523 retval = search_dirblock(bh, dir, fname, d_name,
1524 block << EXT4_BLOCK_SIZE_BITS(sb),
1525 res_dir);
1526 if (retval == 1)
1527 goto success;
1528 brelse(bh);
1529 if (retval == -1) {
1530 bh = ERR_PTR(ERR_BAD_DX_DIR);
1531 goto errout;
1532 }
1533
1534 /* Check to see if we should continue to search */
1535 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1536 frames, NULL);
1537 if (retval < 0) {
1538 ext4_warning_inode(dir,
1539 "error %d reading directory index block",
1540 retval);
1541 bh = ERR_PTR(retval);
1542 goto errout;
1543 }
1544 } while (retval == 1);
1545
1546 bh = NULL;
1547 errout:
1548 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1549 success:
1550 dx_release(frames);
1551 return bh;
1552 }
1553
1554 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1555 {
1556 struct inode *inode;
1557 struct ext4_dir_entry_2 *de;
1558 struct buffer_head *bh;
1559
1560 if (ext4_encrypted_inode(dir)) {
1561 int res = ext4_get_encryption_info(dir);
1562
1563 /*
1564 * This should be a properly defined flag for
1565 * dentry->d_flags when we uplift this to the VFS.
1566 * d_fsdata is set to (void *) 1 if if the dentry is
1567 * created while the directory was encrypted and we
1568 * don't have access to the key.
1569 */
1570 dentry->d_fsdata = NULL;
1571 if (ext4_encryption_info(dir))
1572 dentry->d_fsdata = (void *) 1;
1573 d_set_d_op(dentry, &ext4_encrypted_d_ops);
1574 if (res && res != -ENOKEY)
1575 return ERR_PTR(res);
1576 }
1577
1578 if (dentry->d_name.len > EXT4_NAME_LEN)
1579 return ERR_PTR(-ENAMETOOLONG);
1580
1581 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1582 if (IS_ERR(bh))
1583 return (struct dentry *) bh;
1584 inode = NULL;
1585 if (bh) {
1586 __u32 ino = le32_to_cpu(de->inode);
1587 brelse(bh);
1588 if (!ext4_valid_inum(dir->i_sb, ino)) {
1589 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1590 return ERR_PTR(-EFSCORRUPTED);
1591 }
1592 if (unlikely(ino == dir->i_ino)) {
1593 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1594 dentry);
1595 return ERR_PTR(-EFSCORRUPTED);
1596 }
1597 inode = ext4_iget_normal(dir->i_sb, ino);
1598 if (inode == ERR_PTR(-ESTALE)) {
1599 EXT4_ERROR_INODE(dir,
1600 "deleted inode referenced: %u",
1601 ino);
1602 return ERR_PTR(-EFSCORRUPTED);
1603 }
1604 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1605 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1606 S_ISLNK(inode->i_mode)) &&
1607 !ext4_is_child_context_consistent_with_parent(dir,
1608 inode)) {
1609 iput(inode);
1610 ext4_warning(inode->i_sb,
1611 "Inconsistent encryption contexts: %lu/%lu\n",
1612 (unsigned long) dir->i_ino,
1613 (unsigned long) inode->i_ino);
1614 return ERR_PTR(-EPERM);
1615 }
1616 }
1617 return d_splice_alias(inode, dentry);
1618 }
1619
1620
1621 struct dentry *ext4_get_parent(struct dentry *child)
1622 {
1623 __u32 ino;
1624 static const struct qstr dotdot = QSTR_INIT("..", 2);
1625 struct ext4_dir_entry_2 * de;
1626 struct buffer_head *bh;
1627
1628 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1629 if (IS_ERR(bh))
1630 return (struct dentry *) bh;
1631 if (!bh)
1632 return ERR_PTR(-ENOENT);
1633 ino = le32_to_cpu(de->inode);
1634 brelse(bh);
1635
1636 if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1637 EXT4_ERROR_INODE(d_inode(child),
1638 "bad parent inode number: %u", ino);
1639 return ERR_PTR(-EFSCORRUPTED);
1640 }
1641
1642 return d_obtain_alias(ext4_iget_normal(d_inode(child)->i_sb, ino));
1643 }
1644
1645 /*
1646 * Move count entries from end of map between two memory locations.
1647 * Returns pointer to last entry moved.
1648 */
1649 static struct ext4_dir_entry_2 *
1650 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1651 unsigned blocksize)
1652 {
1653 unsigned rec_len = 0;
1654
1655 while (count--) {
1656 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1657 (from + (map->offs<<2));
1658 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1659 memcpy (to, de, rec_len);
1660 ((struct ext4_dir_entry_2 *) to)->rec_len =
1661 ext4_rec_len_to_disk(rec_len, blocksize);
1662 de->inode = 0;
1663 map++;
1664 to += rec_len;
1665 }
1666 return (struct ext4_dir_entry_2 *) (to - rec_len);
1667 }
1668
1669 /*
1670 * Compact each dir entry in the range to the minimal rec_len.
1671 * Returns pointer to last entry in range.
1672 */
1673 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1674 {
1675 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1676 unsigned rec_len = 0;
1677
1678 prev = to = de;
1679 while ((char*)de < base + blocksize) {
1680 next = ext4_next_entry(de, blocksize);
1681 if (de->inode && de->name_len) {
1682 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1683 if (de > to)
1684 memmove(to, de, rec_len);
1685 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1686 prev = to;
1687 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1688 }
1689 de = next;
1690 }
1691 return prev;
1692 }
1693
1694 /*
1695 * Split a full leaf block to make room for a new dir entry.
1696 * Allocate a new block, and move entries so that they are approx. equally full.
1697 * Returns pointer to de in block into which the new entry will be inserted.
1698 */
1699 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1700 struct buffer_head **bh,struct dx_frame *frame,
1701 struct dx_hash_info *hinfo)
1702 {
1703 unsigned blocksize = dir->i_sb->s_blocksize;
1704 unsigned count, continued;
1705 struct buffer_head *bh2;
1706 ext4_lblk_t newblock;
1707 u32 hash2;
1708 struct dx_map_entry *map;
1709 char *data1 = (*bh)->b_data, *data2;
1710 unsigned split, move, size;
1711 struct ext4_dir_entry_2 *de = NULL, *de2;
1712 struct ext4_dir_entry_tail *t;
1713 int csum_size = 0;
1714 int err = 0, i;
1715
1716 if (ext4_has_metadata_csum(dir->i_sb))
1717 csum_size = sizeof(struct ext4_dir_entry_tail);
1718
1719 bh2 = ext4_append(handle, dir, &newblock);
1720 if (IS_ERR(bh2)) {
1721 brelse(*bh);
1722 *bh = NULL;
1723 return (struct ext4_dir_entry_2 *) bh2;
1724 }
1725
1726 BUFFER_TRACE(*bh, "get_write_access");
1727 err = ext4_journal_get_write_access(handle, *bh);
1728 if (err)
1729 goto journal_error;
1730
1731 BUFFER_TRACE(frame->bh, "get_write_access");
1732 err = ext4_journal_get_write_access(handle, frame->bh);
1733 if (err)
1734 goto journal_error;
1735
1736 data2 = bh2->b_data;
1737
1738 /* create map in the end of data2 block */
1739 map = (struct dx_map_entry *) (data2 + blocksize);
1740 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1741 blocksize, hinfo, map);
1742 map -= count;
1743 dx_sort_map(map, count);
1744 /* Split the existing block in the middle, size-wise */
1745 size = 0;
1746 move = 0;
1747 for (i = count-1; i >= 0; i--) {
1748 /* is more than half of this entry in 2nd half of the block? */
1749 if (size + map[i].size/2 > blocksize/2)
1750 break;
1751 size += map[i].size;
1752 move++;
1753 }
1754 /* map index at which we will split */
1755 split = count - move;
1756 hash2 = map[split].hash;
1757 continued = hash2 == map[split - 1].hash;
1758 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1759 (unsigned long)dx_get_block(frame->at),
1760 hash2, split, count-split));
1761
1762 /* Fancy dance to stay within two buffers */
1763 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1764 blocksize);
1765 de = dx_pack_dirents(data1, blocksize);
1766 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1767 (char *) de,
1768 blocksize);
1769 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1770 (char *) de2,
1771 blocksize);
1772 if (csum_size) {
1773 t = EXT4_DIRENT_TAIL(data2, blocksize);
1774 initialize_dirent_tail(t, blocksize);
1775
1776 t = EXT4_DIRENT_TAIL(data1, blocksize);
1777 initialize_dirent_tail(t, blocksize);
1778 }
1779
1780 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1781 blocksize, 1));
1782 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1783 blocksize, 1));
1784
1785 /* Which block gets the new entry? */
1786 if (hinfo->hash >= hash2) {
1787 swap(*bh, bh2);
1788 de = de2;
1789 }
1790 dx_insert_block(frame, hash2 + continued, newblock);
1791 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1792 if (err)
1793 goto journal_error;
1794 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1795 if (err)
1796 goto journal_error;
1797 brelse(bh2);
1798 dxtrace(dx_show_index("frame", frame->entries));
1799 return de;
1800
1801 journal_error:
1802 brelse(*bh);
1803 brelse(bh2);
1804 *bh = NULL;
1805 ext4_std_error(dir->i_sb, err);
1806 return ERR_PTR(err);
1807 }
1808
1809 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1810 struct buffer_head *bh,
1811 void *buf, int buf_size,
1812 struct ext4_filename *fname,
1813 struct ext4_dir_entry_2 **dest_de)
1814 {
1815 struct ext4_dir_entry_2 *de;
1816 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1817 int nlen, rlen;
1818 unsigned int offset = 0;
1819 char *top;
1820 int res;
1821
1822 de = (struct ext4_dir_entry_2 *)buf;
1823 top = buf + buf_size - reclen;
1824 while ((char *) de <= top) {
1825 if (ext4_check_dir_entry(dir, NULL, de, bh,
1826 buf, buf_size, offset)) {
1827 res = -EFSCORRUPTED;
1828 goto return_result;
1829 }
1830 /* Provide crypto context and crypto buffer to ext4 match */
1831 res = ext4_match(fname, de);
1832 if (res < 0)
1833 goto return_result;
1834 if (res > 0) {
1835 res = -EEXIST;
1836 goto return_result;
1837 }
1838 nlen = EXT4_DIR_REC_LEN(de->name_len);
1839 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1840 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1841 break;
1842 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1843 offset += rlen;
1844 }
1845
1846 if ((char *) de > top)
1847 res = -ENOSPC;
1848 else {
1849 *dest_de = de;
1850 res = 0;
1851 }
1852 return_result:
1853 return res;
1854 }
1855
1856 int ext4_insert_dentry(struct inode *dir,
1857 struct inode *inode,
1858 struct ext4_dir_entry_2 *de,
1859 int buf_size,
1860 struct ext4_filename *fname)
1861 {
1862
1863 int nlen, rlen;
1864
1865 nlen = EXT4_DIR_REC_LEN(de->name_len);
1866 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1867 if (de->inode) {
1868 struct ext4_dir_entry_2 *de1 =
1869 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1870 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1871 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1872 de = de1;
1873 }
1874 de->file_type = EXT4_FT_UNKNOWN;
1875 de->inode = cpu_to_le32(inode->i_ino);
1876 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1877 de->name_len = fname_len(fname);
1878 memcpy(de->name, fname_name(fname), fname_len(fname));
1879 return 0;
1880 }
1881
1882 /*
1883 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1884 * it points to a directory entry which is guaranteed to be large
1885 * enough for new directory entry. If de is NULL, then
1886 * add_dirent_to_buf will attempt search the directory block for
1887 * space. It will return -ENOSPC if no space is available, and -EIO
1888 * and -EEXIST if directory entry already exists.
1889 */
1890 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1891 struct inode *dir,
1892 struct inode *inode, struct ext4_dir_entry_2 *de,
1893 struct buffer_head *bh)
1894 {
1895 unsigned int blocksize = dir->i_sb->s_blocksize;
1896 int csum_size = 0;
1897 int err;
1898
1899 if (ext4_has_metadata_csum(inode->i_sb))
1900 csum_size = sizeof(struct ext4_dir_entry_tail);
1901
1902 if (!de) {
1903 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1904 blocksize - csum_size, fname, &de);
1905 if (err)
1906 return err;
1907 }
1908 BUFFER_TRACE(bh, "get_write_access");
1909 err = ext4_journal_get_write_access(handle, bh);
1910 if (err) {
1911 ext4_std_error(dir->i_sb, err);
1912 return err;
1913 }
1914
1915 /* By now the buffer is marked for journaling. Due to crypto operations,
1916 * the following function call may fail */
1917 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1918 if (err < 0)
1919 return err;
1920
1921 /*
1922 * XXX shouldn't update any times until successful
1923 * completion of syscall, but too many callers depend
1924 * on this.
1925 *
1926 * XXX similarly, too many callers depend on
1927 * ext4_new_inode() setting the times, but error
1928 * recovery deletes the inode, so the worst that can
1929 * happen is that the times are slightly out of date
1930 * and/or different from the directory change time.
1931 */
1932 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1933 ext4_update_dx_flag(dir);
1934 dir->i_version++;
1935 ext4_mark_inode_dirty(handle, dir);
1936 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1937 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1938 if (err)
1939 ext4_std_error(dir->i_sb, err);
1940 return 0;
1941 }
1942
1943 /*
1944 * This converts a one block unindexed directory to a 3 block indexed
1945 * directory, and adds the dentry to the indexed directory.
1946 */
1947 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1948 struct dentry *dentry,
1949 struct inode *inode, struct buffer_head *bh)
1950 {
1951 struct inode *dir = d_inode(dentry->d_parent);
1952 struct buffer_head *bh2;
1953 struct dx_root *root;
1954 struct dx_frame frames[2], *frame;
1955 struct dx_entry *entries;
1956 struct ext4_dir_entry_2 *de, *de2;
1957 struct ext4_dir_entry_tail *t;
1958 char *data1, *top;
1959 unsigned len;
1960 int retval;
1961 unsigned blocksize;
1962 ext4_lblk_t block;
1963 struct fake_dirent *fde;
1964 int csum_size = 0;
1965
1966 if (ext4_has_metadata_csum(inode->i_sb))
1967 csum_size = sizeof(struct ext4_dir_entry_tail);
1968
1969 blocksize = dir->i_sb->s_blocksize;
1970 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1971 BUFFER_TRACE(bh, "get_write_access");
1972 retval = ext4_journal_get_write_access(handle, bh);
1973 if (retval) {
1974 ext4_std_error(dir->i_sb, retval);
1975 brelse(bh);
1976 return retval;
1977 }
1978 root = (struct dx_root *) bh->b_data;
1979
1980 /* The 0th block becomes the root, move the dirents out */
1981 fde = &root->dotdot;
1982 de = (struct ext4_dir_entry_2 *)((char *)fde +
1983 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1984 if ((char *) de >= (((char *) root) + blocksize)) {
1985 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1986 brelse(bh);
1987 return -EFSCORRUPTED;
1988 }
1989 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1990
1991 /* Allocate new block for the 0th block's dirents */
1992 bh2 = ext4_append(handle, dir, &block);
1993 if (IS_ERR(bh2)) {
1994 brelse(bh);
1995 return PTR_ERR(bh2);
1996 }
1997 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1998 data1 = bh2->b_data;
1999
2000 memcpy (data1, de, len);
2001 de = (struct ext4_dir_entry_2 *) data1;
2002 top = data1 + len;
2003 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2004 de = de2;
2005 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2006 (char *) de,
2007 blocksize);
2008
2009 if (csum_size) {
2010 t = EXT4_DIRENT_TAIL(data1, blocksize);
2011 initialize_dirent_tail(t, blocksize);
2012 }
2013
2014 /* Initialize the root; the dot dirents already exist */
2015 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2016 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2017 blocksize);
2018 memset (&root->info, 0, sizeof(root->info));
2019 root->info.info_length = sizeof(root->info);
2020 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2021 entries = root->entries;
2022 dx_set_block(entries, 1);
2023 dx_set_count(entries, 1);
2024 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2025
2026 /* Initialize as for dx_probe */
2027 fname->hinfo.hash_version = root->info.hash_version;
2028 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2029 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2030 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2031 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2032
2033 memset(frames, 0, sizeof(frames));
2034 frame = frames;
2035 frame->entries = entries;
2036 frame->at = entries;
2037 frame->bh = bh;
2038
2039 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2040 if (retval)
2041 goto out_frames;
2042 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2043 if (retval)
2044 goto out_frames;
2045
2046 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2047 if (IS_ERR(de)) {
2048 retval = PTR_ERR(de);
2049 goto out_frames;
2050 }
2051
2052 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2053 out_frames:
2054 /*
2055 * Even if the block split failed, we have to properly write
2056 * out all the changes we did so far. Otherwise we can end up
2057 * with corrupted filesystem.
2058 */
2059 if (retval)
2060 ext4_mark_inode_dirty(handle, dir);
2061 dx_release(frames);
2062 brelse(bh2);
2063 return retval;
2064 }
2065
2066 /*
2067 * ext4_add_entry()
2068 *
2069 * adds a file entry to the specified directory, using the same
2070 * semantics as ext4_find_entry(). It returns NULL if it failed.
2071 *
2072 * NOTE!! The inode part of 'de' is left at 0 - which means you
2073 * may not sleep between calling this and putting something into
2074 * the entry, as someone else might have used it while you slept.
2075 */
2076 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2077 struct inode *inode)
2078 {
2079 struct inode *dir = d_inode(dentry->d_parent);
2080 struct buffer_head *bh = NULL;
2081 struct ext4_dir_entry_2 *de;
2082 struct ext4_dir_entry_tail *t;
2083 struct super_block *sb;
2084 struct ext4_filename fname;
2085 int retval;
2086 int dx_fallback=0;
2087 unsigned blocksize;
2088 ext4_lblk_t block, blocks;
2089 int csum_size = 0;
2090
2091 if (ext4_has_metadata_csum(inode->i_sb))
2092 csum_size = sizeof(struct ext4_dir_entry_tail);
2093
2094 sb = dir->i_sb;
2095 blocksize = sb->s_blocksize;
2096 if (!dentry->d_name.len)
2097 return -EINVAL;
2098
2099 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2100 if (retval)
2101 return retval;
2102
2103 if (ext4_has_inline_data(dir)) {
2104 retval = ext4_try_add_inline_entry(handle, &fname,
2105 dentry, inode);
2106 if (retval < 0)
2107 goto out;
2108 if (retval == 1) {
2109 retval = 0;
2110 goto out;
2111 }
2112 }
2113
2114 if (is_dx(dir)) {
2115 retval = ext4_dx_add_entry(handle, &fname, dentry, inode);
2116 if (!retval || (retval != ERR_BAD_DX_DIR))
2117 goto out;
2118 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2119 dx_fallback++;
2120 ext4_mark_inode_dirty(handle, dir);
2121 }
2122 blocks = dir->i_size >> sb->s_blocksize_bits;
2123 for (block = 0; block < blocks; block++) {
2124 bh = ext4_read_dirblock(dir, block, DIRENT);
2125 if (IS_ERR(bh)) {
2126 retval = PTR_ERR(bh);
2127 bh = NULL;
2128 goto out;
2129 }
2130 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2131 NULL, bh);
2132 if (retval != -ENOSPC)
2133 goto out;
2134
2135 if (blocks == 1 && !dx_fallback &&
2136 ext4_has_feature_dir_index(sb)) {
2137 retval = make_indexed_dir(handle, &fname, dentry,
2138 inode, bh);
2139 bh = NULL; /* make_indexed_dir releases bh */
2140 goto out;
2141 }
2142 brelse(bh);
2143 }
2144 bh = ext4_append(handle, dir, &block);
2145 if (IS_ERR(bh)) {
2146 retval = PTR_ERR(bh);
2147 bh = NULL;
2148 goto out;
2149 }
2150 de = (struct ext4_dir_entry_2 *) bh->b_data;
2151 de->inode = 0;
2152 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2153
2154 if (csum_size) {
2155 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2156 initialize_dirent_tail(t, blocksize);
2157 }
2158
2159 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2160 out:
2161 ext4_fname_free_filename(&fname);
2162 brelse(bh);
2163 if (retval == 0)
2164 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2165 return retval;
2166 }
2167
2168 /*
2169 * Returns 0 for success, or a negative error value
2170 */
2171 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2172 struct dentry *dentry, struct inode *inode)
2173 {
2174 struct dx_frame frames[2], *frame;
2175 struct dx_entry *entries, *at;
2176 struct buffer_head *bh;
2177 struct inode *dir = d_inode(dentry->d_parent);
2178 struct super_block *sb = dir->i_sb;
2179 struct ext4_dir_entry_2 *de;
2180 int err;
2181
2182 frame = dx_probe(fname, dir, NULL, frames);
2183 if (IS_ERR(frame))
2184 return PTR_ERR(frame);
2185 entries = frame->entries;
2186 at = frame->at;
2187 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2188 if (IS_ERR(bh)) {
2189 err = PTR_ERR(bh);
2190 bh = NULL;
2191 goto cleanup;
2192 }
2193
2194 BUFFER_TRACE(bh, "get_write_access");
2195 err = ext4_journal_get_write_access(handle, bh);
2196 if (err)
2197 goto journal_error;
2198
2199 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2200 if (err != -ENOSPC)
2201 goto cleanup;
2202
2203 /* Block full, should compress but for now just split */
2204 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2205 dx_get_count(entries), dx_get_limit(entries)));
2206 /* Need to split index? */
2207 if (dx_get_count(entries) == dx_get_limit(entries)) {
2208 ext4_lblk_t newblock;
2209 unsigned icount = dx_get_count(entries);
2210 int levels = frame - frames;
2211 struct dx_entry *entries2;
2212 struct dx_node *node2;
2213 struct buffer_head *bh2;
2214
2215 if (levels && (dx_get_count(frames->entries) ==
2216 dx_get_limit(frames->entries))) {
2217 ext4_warning_inode(dir, "Directory index full!");
2218 err = -ENOSPC;
2219 goto cleanup;
2220 }
2221 bh2 = ext4_append(handle, dir, &newblock);
2222 if (IS_ERR(bh2)) {
2223 err = PTR_ERR(bh2);
2224 goto cleanup;
2225 }
2226 node2 = (struct dx_node *)(bh2->b_data);
2227 entries2 = node2->entries;
2228 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2229 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2230 sb->s_blocksize);
2231 BUFFER_TRACE(frame->bh, "get_write_access");
2232 err = ext4_journal_get_write_access(handle, frame->bh);
2233 if (err)
2234 goto journal_error;
2235 if (levels) {
2236 unsigned icount1 = icount/2, icount2 = icount - icount1;
2237 unsigned hash2 = dx_get_hash(entries + icount1);
2238 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2239 icount1, icount2));
2240
2241 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2242 err = ext4_journal_get_write_access(handle,
2243 frames[0].bh);
2244 if (err)
2245 goto journal_error;
2246
2247 memcpy((char *) entries2, (char *) (entries + icount1),
2248 icount2 * sizeof(struct dx_entry));
2249 dx_set_count(entries, icount1);
2250 dx_set_count(entries2, icount2);
2251 dx_set_limit(entries2, dx_node_limit(dir));
2252
2253 /* Which index block gets the new entry? */
2254 if (at - entries >= icount1) {
2255 frame->at = at = at - entries - icount1 + entries2;
2256 frame->entries = entries = entries2;
2257 swap(frame->bh, bh2);
2258 }
2259 dx_insert_block(frames + 0, hash2, newblock);
2260 dxtrace(dx_show_index("node", frames[1].entries));
2261 dxtrace(dx_show_index("node",
2262 ((struct dx_node *) bh2->b_data)->entries));
2263 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2264 if (err)
2265 goto journal_error;
2266 brelse (bh2);
2267 } else {
2268 dxtrace(printk(KERN_DEBUG
2269 "Creating second level index...\n"));
2270 memcpy((char *) entries2, (char *) entries,
2271 icount * sizeof(struct dx_entry));
2272 dx_set_limit(entries2, dx_node_limit(dir));
2273
2274 /* Set up root */
2275 dx_set_count(entries, 1);
2276 dx_set_block(entries + 0, newblock);
2277 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2278
2279 /* Add new access path frame */
2280 frame = frames + 1;
2281 frame->at = at = at - entries + entries2;
2282 frame->entries = entries = entries2;
2283 frame->bh = bh2;
2284 err = ext4_journal_get_write_access(handle,
2285 frame->bh);
2286 if (err)
2287 goto journal_error;
2288 }
2289 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2290 if (err) {
2291 ext4_std_error(inode->i_sb, err);
2292 goto cleanup;
2293 }
2294 }
2295 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2296 if (IS_ERR(de)) {
2297 err = PTR_ERR(de);
2298 goto cleanup;
2299 }
2300 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2301 goto cleanup;
2302
2303 journal_error:
2304 ext4_std_error(dir->i_sb, err);
2305 cleanup:
2306 brelse(bh);
2307 dx_release(frames);
2308 return err;
2309 }
2310
2311 /*
2312 * ext4_generic_delete_entry deletes a directory entry by merging it
2313 * with the previous entry
2314 */
2315 int ext4_generic_delete_entry(handle_t *handle,
2316 struct inode *dir,
2317 struct ext4_dir_entry_2 *de_del,
2318 struct buffer_head *bh,
2319 void *entry_buf,
2320 int buf_size,
2321 int csum_size)
2322 {
2323 struct ext4_dir_entry_2 *de, *pde;
2324 unsigned int blocksize = dir->i_sb->s_blocksize;
2325 int i;
2326
2327 i = 0;
2328 pde = NULL;
2329 de = (struct ext4_dir_entry_2 *)entry_buf;
2330 while (i < buf_size - csum_size) {
2331 if (ext4_check_dir_entry(dir, NULL, de, bh,
2332 bh->b_data, bh->b_size, i))
2333 return -EFSCORRUPTED;
2334 if (de == de_del) {
2335 if (pde)
2336 pde->rec_len = ext4_rec_len_to_disk(
2337 ext4_rec_len_from_disk(pde->rec_len,
2338 blocksize) +
2339 ext4_rec_len_from_disk(de->rec_len,
2340 blocksize),
2341 blocksize);
2342 else
2343 de->inode = 0;
2344 dir->i_version++;
2345 return 0;
2346 }
2347 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2348 pde = de;
2349 de = ext4_next_entry(de, blocksize);
2350 }
2351 return -ENOENT;
2352 }
2353
2354 static int ext4_delete_entry(handle_t *handle,
2355 struct inode *dir,
2356 struct ext4_dir_entry_2 *de_del,
2357 struct buffer_head *bh)
2358 {
2359 int err, csum_size = 0;
2360
2361 if (ext4_has_inline_data(dir)) {
2362 int has_inline_data = 1;
2363 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2364 &has_inline_data);
2365 if (has_inline_data)
2366 return err;
2367 }
2368
2369 if (ext4_has_metadata_csum(dir->i_sb))
2370 csum_size = sizeof(struct ext4_dir_entry_tail);
2371
2372 BUFFER_TRACE(bh, "get_write_access");
2373 err = ext4_journal_get_write_access(handle, bh);
2374 if (unlikely(err))
2375 goto out;
2376
2377 err = ext4_generic_delete_entry(handle, dir, de_del,
2378 bh, bh->b_data,
2379 dir->i_sb->s_blocksize, csum_size);
2380 if (err)
2381 goto out;
2382
2383 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2384 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2385 if (unlikely(err))
2386 goto out;
2387
2388 return 0;
2389 out:
2390 if (err != -ENOENT)
2391 ext4_std_error(dir->i_sb, err);
2392 return err;
2393 }
2394
2395 /*
2396 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2397 * since this indicates that nlinks count was previously 1.
2398 */
2399 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2400 {
2401 inc_nlink(inode);
2402 if (is_dx(inode) && inode->i_nlink > 1) {
2403 /* limit is 16-bit i_links_count */
2404 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2405 set_nlink(inode, 1);
2406 ext4_set_feature_dir_nlink(inode->i_sb);
2407 }
2408 }
2409 }
2410
2411 /*
2412 * If a directory had nlink == 1, then we should let it be 1. This indicates
2413 * directory has >EXT4_LINK_MAX subdirs.
2414 */
2415 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2416 {
2417 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2418 drop_nlink(inode);
2419 }
2420
2421
2422 static int ext4_add_nondir(handle_t *handle,
2423 struct dentry *dentry, struct inode *inode)
2424 {
2425 int err = ext4_add_entry(handle, dentry, inode);
2426 if (!err) {
2427 ext4_mark_inode_dirty(handle, inode);
2428 unlock_new_inode(inode);
2429 d_instantiate(dentry, inode);
2430 return 0;
2431 }
2432 drop_nlink(inode);
2433 unlock_new_inode(inode);
2434 iput(inode);
2435 return err;
2436 }
2437
2438 /*
2439 * By the time this is called, we already have created
2440 * the directory cache entry for the new file, but it
2441 * is so far negative - it has no inode.
2442 *
2443 * If the create succeeds, we fill in the inode information
2444 * with d_instantiate().
2445 */
2446 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2447 bool excl)
2448 {
2449 handle_t *handle;
2450 struct inode *inode;
2451 int err, credits, retries = 0;
2452
2453 err = dquot_initialize(dir);
2454 if (err)
2455 return err;
2456
2457 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2458 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2459 retry:
2460 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2461 NULL, EXT4_HT_DIR, credits);
2462 handle = ext4_journal_current_handle();
2463 err = PTR_ERR(inode);
2464 if (!IS_ERR(inode)) {
2465 inode->i_op = &ext4_file_inode_operations;
2466 inode->i_fop = &ext4_file_operations;
2467 ext4_set_aops(inode);
2468 err = ext4_add_nondir(handle, dentry, inode);
2469 if (!err && IS_DIRSYNC(dir))
2470 ext4_handle_sync(handle);
2471 }
2472 if (handle)
2473 ext4_journal_stop(handle);
2474 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2475 goto retry;
2476 return err;
2477 }
2478
2479 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2480 umode_t mode, dev_t rdev)
2481 {
2482 handle_t *handle;
2483 struct inode *inode;
2484 int err, credits, retries = 0;
2485
2486 err = dquot_initialize(dir);
2487 if (err)
2488 return err;
2489
2490 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2491 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2492 retry:
2493 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2494 NULL, EXT4_HT_DIR, credits);
2495 handle = ext4_journal_current_handle();
2496 err = PTR_ERR(inode);
2497 if (!IS_ERR(inode)) {
2498 init_special_inode(inode, inode->i_mode, rdev);
2499 inode->i_op = &ext4_special_inode_operations;
2500 err = ext4_add_nondir(handle, dentry, inode);
2501 if (!err && IS_DIRSYNC(dir))
2502 ext4_handle_sync(handle);
2503 }
2504 if (handle)
2505 ext4_journal_stop(handle);
2506 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2507 goto retry;
2508 return err;
2509 }
2510
2511 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2512 {
2513 handle_t *handle;
2514 struct inode *inode;
2515 int err, retries = 0;
2516
2517 err = dquot_initialize(dir);
2518 if (err)
2519 return err;
2520
2521 retry:
2522 inode = ext4_new_inode_start_handle(dir, mode,
2523 NULL, 0, NULL,
2524 EXT4_HT_DIR,
2525 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2526 4 + EXT4_XATTR_TRANS_BLOCKS);
2527 handle = ext4_journal_current_handle();
2528 err = PTR_ERR(inode);
2529 if (!IS_ERR(inode)) {
2530 inode->i_op = &ext4_file_inode_operations;
2531 inode->i_fop = &ext4_file_operations;
2532 ext4_set_aops(inode);
2533 d_tmpfile(dentry, inode);
2534 err = ext4_orphan_add(handle, inode);
2535 if (err)
2536 goto err_unlock_inode;
2537 mark_inode_dirty(inode);
2538 unlock_new_inode(inode);
2539 }
2540 if (handle)
2541 ext4_journal_stop(handle);
2542 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2543 goto retry;
2544 return err;
2545 err_unlock_inode:
2546 ext4_journal_stop(handle);
2547 unlock_new_inode(inode);
2548 return err;
2549 }
2550
2551 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2552 struct ext4_dir_entry_2 *de,
2553 int blocksize, int csum_size,
2554 unsigned int parent_ino, int dotdot_real_len)
2555 {
2556 de->inode = cpu_to_le32(inode->i_ino);
2557 de->name_len = 1;
2558 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2559 blocksize);
2560 strcpy(de->name, ".");
2561 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2562
2563 de = ext4_next_entry(de, blocksize);
2564 de->inode = cpu_to_le32(parent_ino);
2565 de->name_len = 2;
2566 if (!dotdot_real_len)
2567 de->rec_len = ext4_rec_len_to_disk(blocksize -
2568 (csum_size + EXT4_DIR_REC_LEN(1)),
2569 blocksize);
2570 else
2571 de->rec_len = ext4_rec_len_to_disk(
2572 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2573 strcpy(de->name, "..");
2574 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2575
2576 return ext4_next_entry(de, blocksize);
2577 }
2578
2579 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2580 struct inode *inode)
2581 {
2582 struct buffer_head *dir_block = NULL;
2583 struct ext4_dir_entry_2 *de;
2584 struct ext4_dir_entry_tail *t;
2585 ext4_lblk_t block = 0;
2586 unsigned int blocksize = dir->i_sb->s_blocksize;
2587 int csum_size = 0;
2588 int err;
2589
2590 if (ext4_has_metadata_csum(dir->i_sb))
2591 csum_size = sizeof(struct ext4_dir_entry_tail);
2592
2593 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2594 err = ext4_try_create_inline_dir(handle, dir, inode);
2595 if (err < 0 && err != -ENOSPC)
2596 goto out;
2597 if (!err)
2598 goto out;
2599 }
2600
2601 inode->i_size = 0;
2602 dir_block = ext4_append(handle, inode, &block);
2603 if (IS_ERR(dir_block))
2604 return PTR_ERR(dir_block);
2605 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2606 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2607 set_nlink(inode, 2);
2608 if (csum_size) {
2609 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2610 initialize_dirent_tail(t, blocksize);
2611 }
2612
2613 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2614 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2615 if (err)
2616 goto out;
2617 set_buffer_verified(dir_block);
2618 out:
2619 brelse(dir_block);
2620 return err;
2621 }
2622
2623 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2624 {
2625 handle_t *handle;
2626 struct inode *inode;
2627 int err, credits, retries = 0;
2628
2629 if (EXT4_DIR_LINK_MAX(dir))
2630 return -EMLINK;
2631
2632 err = dquot_initialize(dir);
2633 if (err)
2634 return err;
2635
2636 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2637 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2638 retry:
2639 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2640 &dentry->d_name,
2641 0, NULL, EXT4_HT_DIR, credits);
2642 handle = ext4_journal_current_handle();
2643 err = PTR_ERR(inode);
2644 if (IS_ERR(inode))
2645 goto out_stop;
2646
2647 inode->i_op = &ext4_dir_inode_operations;
2648 inode->i_fop = &ext4_dir_operations;
2649 err = ext4_init_new_dir(handle, dir, inode);
2650 if (err)
2651 goto out_clear_inode;
2652 err = ext4_mark_inode_dirty(handle, inode);
2653 if (!err)
2654 err = ext4_add_entry(handle, dentry, inode);
2655 if (err) {
2656 out_clear_inode:
2657 clear_nlink(inode);
2658 unlock_new_inode(inode);
2659 ext4_mark_inode_dirty(handle, inode);
2660 iput(inode);
2661 goto out_stop;
2662 }
2663 ext4_inc_count(handle, dir);
2664 ext4_update_dx_flag(dir);
2665 err = ext4_mark_inode_dirty(handle, dir);
2666 if (err)
2667 goto out_clear_inode;
2668 unlock_new_inode(inode);
2669 d_instantiate(dentry, inode);
2670 if (IS_DIRSYNC(dir))
2671 ext4_handle_sync(handle);
2672
2673 out_stop:
2674 if (handle)
2675 ext4_journal_stop(handle);
2676 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2677 goto retry;
2678 return err;
2679 }
2680
2681 /*
2682 * routine to check that the specified directory is empty (for rmdir)
2683 */
2684 int ext4_empty_dir(struct inode *inode)
2685 {
2686 unsigned int offset;
2687 struct buffer_head *bh;
2688 struct ext4_dir_entry_2 *de, *de1;
2689 struct super_block *sb;
2690 int err = 0;
2691
2692 if (ext4_has_inline_data(inode)) {
2693 int has_inline_data = 1;
2694
2695 err = empty_inline_dir(inode, &has_inline_data);
2696 if (has_inline_data)
2697 return err;
2698 }
2699
2700 sb = inode->i_sb;
2701 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2702 EXT4_ERROR_INODE(inode, "invalid size");
2703 return 1;
2704 }
2705 bh = ext4_read_dirblock(inode, 0, EITHER);
2706 if (IS_ERR(bh))
2707 return 1;
2708
2709 de = (struct ext4_dir_entry_2 *) bh->b_data;
2710 de1 = ext4_next_entry(de, sb->s_blocksize);
2711 if (le32_to_cpu(de->inode) != inode->i_ino ||
2712 le32_to_cpu(de1->inode) == 0 ||
2713 strcmp(".", de->name) || strcmp("..", de1->name)) {
2714 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2715 brelse(bh);
2716 return 1;
2717 }
2718 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2719 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2720 de = ext4_next_entry(de1, sb->s_blocksize);
2721 while (offset < inode->i_size) {
2722 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2723 unsigned int lblock;
2724 err = 0;
2725 brelse(bh);
2726 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2727 bh = ext4_read_dirblock(inode, lblock, EITHER);
2728 if (IS_ERR(bh))
2729 return 1;
2730 de = (struct ext4_dir_entry_2 *) bh->b_data;
2731 }
2732 if (ext4_check_dir_entry(inode, NULL, de, bh,
2733 bh->b_data, bh->b_size, offset)) {
2734 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2735 sb->s_blocksize);
2736 offset = (offset | (sb->s_blocksize - 1)) + 1;
2737 continue;
2738 }
2739 if (le32_to_cpu(de->inode)) {
2740 brelse(bh);
2741 return 0;
2742 }
2743 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2744 de = ext4_next_entry(de, sb->s_blocksize);
2745 }
2746 brelse(bh);
2747 return 1;
2748 }
2749
2750 /*
2751 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2752 * such inodes, starting at the superblock, in case we crash before the
2753 * file is closed/deleted, or in case the inode truncate spans multiple
2754 * transactions and the last transaction is not recovered after a crash.
2755 *
2756 * At filesystem recovery time, we walk this list deleting unlinked
2757 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2758 *
2759 * Orphan list manipulation functions must be called under i_mutex unless
2760 * we are just creating the inode or deleting it.
2761 */
2762 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2763 {
2764 struct super_block *sb = inode->i_sb;
2765 struct ext4_sb_info *sbi = EXT4_SB(sb);
2766 struct ext4_iloc iloc;
2767 int err = 0, rc;
2768 bool dirty = false;
2769
2770 if (!sbi->s_journal || is_bad_inode(inode))
2771 return 0;
2772
2773 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2774 !mutex_is_locked(&inode->i_mutex));
2775 /*
2776 * Exit early if inode already is on orphan list. This is a big speedup
2777 * since we don't have to contend on the global s_orphan_lock.
2778 */
2779 if (!list_empty(&EXT4_I(inode)->i_orphan))
2780 return 0;
2781
2782 /*
2783 * Orphan handling is only valid for files with data blocks
2784 * being truncated, or files being unlinked. Note that we either
2785 * hold i_mutex, or the inode can not be referenced from outside,
2786 * so i_nlink should not be bumped due to race
2787 */
2788 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2789 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2790
2791 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2792 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2793 if (err)
2794 goto out;
2795
2796 err = ext4_reserve_inode_write(handle, inode, &iloc);
2797 if (err)
2798 goto out;
2799
2800 mutex_lock(&sbi->s_orphan_lock);
2801 /*
2802 * Due to previous errors inode may be already a part of on-disk
2803 * orphan list. If so skip on-disk list modification.
2804 */
2805 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2806 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2807 /* Insert this inode at the head of the on-disk orphan list */
2808 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2809 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2810 dirty = true;
2811 }
2812 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2813 mutex_unlock(&sbi->s_orphan_lock);
2814
2815 if (dirty) {
2816 err = ext4_handle_dirty_super(handle, sb);
2817 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2818 if (!err)
2819 err = rc;
2820 if (err) {
2821 /*
2822 * We have to remove inode from in-memory list if
2823 * addition to on disk orphan list failed. Stray orphan
2824 * list entries can cause panics at unmount time.
2825 */
2826 mutex_lock(&sbi->s_orphan_lock);
2827 list_del_init(&EXT4_I(inode)->i_orphan);
2828 mutex_unlock(&sbi->s_orphan_lock);
2829 }
2830 }
2831 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2832 jbd_debug(4, "orphan inode %lu will point to %d\n",
2833 inode->i_ino, NEXT_ORPHAN(inode));
2834 out:
2835 ext4_std_error(sb, err);
2836 return err;
2837 }
2838
2839 /*
2840 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2841 * of such inodes stored on disk, because it is finally being cleaned up.
2842 */
2843 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2844 {
2845 struct list_head *prev;
2846 struct ext4_inode_info *ei = EXT4_I(inode);
2847 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2848 __u32 ino_next;
2849 struct ext4_iloc iloc;
2850 int err = 0;
2851
2852 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2853 return 0;
2854
2855 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2856 !mutex_is_locked(&inode->i_mutex));
2857 /* Do this quick check before taking global s_orphan_lock. */
2858 if (list_empty(&ei->i_orphan))
2859 return 0;
2860
2861 if (handle) {
2862 /* Grab inode buffer early before taking global s_orphan_lock */
2863 err = ext4_reserve_inode_write(handle, inode, &iloc);
2864 }
2865
2866 mutex_lock(&sbi->s_orphan_lock);
2867 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2868
2869 prev = ei->i_orphan.prev;
2870 list_del_init(&ei->i_orphan);
2871
2872 /* If we're on an error path, we may not have a valid
2873 * transaction handle with which to update the orphan list on
2874 * disk, but we still need to remove the inode from the linked
2875 * list in memory. */
2876 if (!handle || err) {
2877 mutex_unlock(&sbi->s_orphan_lock);
2878 goto out_err;
2879 }
2880
2881 ino_next = NEXT_ORPHAN(inode);
2882 if (prev == &sbi->s_orphan) {
2883 jbd_debug(4, "superblock will point to %u\n", ino_next);
2884 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2885 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2886 if (err) {
2887 mutex_unlock(&sbi->s_orphan_lock);
2888 goto out_brelse;
2889 }
2890 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2891 mutex_unlock(&sbi->s_orphan_lock);
2892 err = ext4_handle_dirty_super(handle, inode->i_sb);
2893 } else {
2894 struct ext4_iloc iloc2;
2895 struct inode *i_prev =
2896 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2897
2898 jbd_debug(4, "orphan inode %lu will point to %u\n",
2899 i_prev->i_ino, ino_next);
2900 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2901 if (err) {
2902 mutex_unlock(&sbi->s_orphan_lock);
2903 goto out_brelse;
2904 }
2905 NEXT_ORPHAN(i_prev) = ino_next;
2906 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2907 mutex_unlock(&sbi->s_orphan_lock);
2908 }
2909 if (err)
2910 goto out_brelse;
2911 NEXT_ORPHAN(inode) = 0;
2912 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2913 out_err:
2914 ext4_std_error(inode->i_sb, err);
2915 return err;
2916
2917 out_brelse:
2918 brelse(iloc.bh);
2919 goto out_err;
2920 }
2921
2922 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2923 {
2924 int retval;
2925 struct inode *inode;
2926 struct buffer_head *bh;
2927 struct ext4_dir_entry_2 *de;
2928 handle_t *handle = NULL;
2929
2930 /* Initialize quotas before so that eventual writes go in
2931 * separate transaction */
2932 retval = dquot_initialize(dir);
2933 if (retval)
2934 return retval;
2935 retval = dquot_initialize(d_inode(dentry));
2936 if (retval)
2937 return retval;
2938
2939 retval = -ENOENT;
2940 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2941 if (IS_ERR(bh))
2942 return PTR_ERR(bh);
2943 if (!bh)
2944 goto end_rmdir;
2945
2946 inode = d_inode(dentry);
2947
2948 retval = -EFSCORRUPTED;
2949 if (le32_to_cpu(de->inode) != inode->i_ino)
2950 goto end_rmdir;
2951
2952 retval = -ENOTEMPTY;
2953 if (!ext4_empty_dir(inode))
2954 goto end_rmdir;
2955
2956 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2957 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2958 if (IS_ERR(handle)) {
2959 retval = PTR_ERR(handle);
2960 handle = NULL;
2961 goto end_rmdir;
2962 }
2963
2964 if (IS_DIRSYNC(dir))
2965 ext4_handle_sync(handle);
2966
2967 retval = ext4_delete_entry(handle, dir, de, bh);
2968 if (retval)
2969 goto end_rmdir;
2970 if (!EXT4_DIR_LINK_EMPTY(inode))
2971 ext4_warning_inode(inode,
2972 "empty directory '%.*s' has too many links (%u)",
2973 dentry->d_name.len, dentry->d_name.name,
2974 inode->i_nlink);
2975 inode->i_version++;
2976 clear_nlink(inode);
2977 /* There's no need to set i_disksize: the fact that i_nlink is
2978 * zero will ensure that the right thing happens during any
2979 * recovery. */
2980 inode->i_size = 0;
2981 ext4_orphan_add(handle, inode);
2982 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2983 ext4_mark_inode_dirty(handle, inode);
2984 ext4_dec_count(handle, dir);
2985 ext4_update_dx_flag(dir);
2986 ext4_mark_inode_dirty(handle, dir);
2987
2988 end_rmdir:
2989 brelse(bh);
2990 if (handle)
2991 ext4_journal_stop(handle);
2992 return retval;
2993 }
2994
2995 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2996 {
2997 int retval;
2998 struct inode *inode;
2999 struct buffer_head *bh;
3000 struct ext4_dir_entry_2 *de;
3001 handle_t *handle = NULL;
3002
3003 trace_ext4_unlink_enter(dir, dentry);
3004 /* Initialize quotas before so that eventual writes go
3005 * in separate transaction */
3006 retval = dquot_initialize(dir);
3007 if (retval)
3008 return retval;
3009 retval = dquot_initialize(d_inode(dentry));
3010 if (retval)
3011 return retval;
3012
3013 retval = -ENOENT;
3014 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3015 if (IS_ERR(bh))
3016 return PTR_ERR(bh);
3017 if (!bh)
3018 goto end_unlink;
3019
3020 inode = d_inode(dentry);
3021
3022 retval = -EFSCORRUPTED;
3023 if (le32_to_cpu(de->inode) != inode->i_ino)
3024 goto end_unlink;
3025
3026 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3027 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3028 if (IS_ERR(handle)) {
3029 retval = PTR_ERR(handle);
3030 handle = NULL;
3031 goto end_unlink;
3032 }
3033
3034 if (IS_DIRSYNC(dir))
3035 ext4_handle_sync(handle);
3036
3037 if (inode->i_nlink == 0) {
3038 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3039 dentry->d_name.len, dentry->d_name.name);
3040 set_nlink(inode, 1);
3041 }
3042 retval = ext4_delete_entry(handle, dir, de, bh);
3043 if (retval)
3044 goto end_unlink;
3045 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3046 ext4_update_dx_flag(dir);
3047 ext4_mark_inode_dirty(handle, dir);
3048 drop_nlink(inode);
3049 if (!inode->i_nlink)
3050 ext4_orphan_add(handle, inode);
3051 inode->i_ctime = ext4_current_time(inode);
3052 ext4_mark_inode_dirty(handle, inode);
3053
3054 end_unlink:
3055 brelse(bh);
3056 if (handle)
3057 ext4_journal_stop(handle);
3058 trace_ext4_unlink_exit(dentry, retval);
3059 return retval;
3060 }
3061
3062 static int ext4_symlink(struct inode *dir,
3063 struct dentry *dentry, const char *symname)
3064 {
3065 handle_t *handle;
3066 struct inode *inode;
3067 int err, len = strlen(symname);
3068 int credits;
3069 bool encryption_required;
3070 struct ext4_str disk_link;
3071 struct ext4_encrypted_symlink_data *sd = NULL;
3072
3073 disk_link.len = len + 1;
3074 disk_link.name = (char *) symname;
3075
3076 encryption_required = (ext4_encrypted_inode(dir) ||
3077 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3078 if (encryption_required) {
3079 err = ext4_get_encryption_info(dir);
3080 if (err)
3081 return err;
3082 if (ext4_encryption_info(dir) == NULL)
3083 return -EPERM;
3084 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3085 sizeof(struct ext4_encrypted_symlink_data));
3086 sd = kzalloc(disk_link.len, GFP_KERNEL);
3087 if (!sd)
3088 return -ENOMEM;
3089 }
3090
3091 if (disk_link.len > dir->i_sb->s_blocksize) {
3092 err = -ENAMETOOLONG;
3093 goto err_free_sd;
3094 }
3095
3096 err = dquot_initialize(dir);
3097 if (err)
3098 goto err_free_sd;
3099
3100 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3101 /*
3102 * For non-fast symlinks, we just allocate inode and put it on
3103 * orphan list in the first transaction => we need bitmap,
3104 * group descriptor, sb, inode block, quota blocks, and
3105 * possibly selinux xattr blocks.
3106 */
3107 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3108 EXT4_XATTR_TRANS_BLOCKS;
3109 } else {
3110 /*
3111 * Fast symlink. We have to add entry to directory
3112 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3113 * allocate new inode (bitmap, group descriptor, inode block,
3114 * quota blocks, sb is already counted in previous macros).
3115 */
3116 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3117 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3118 }
3119
3120 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3121 &dentry->d_name, 0, NULL,
3122 EXT4_HT_DIR, credits);
3123 handle = ext4_journal_current_handle();
3124 if (IS_ERR(inode)) {
3125 if (handle)
3126 ext4_journal_stop(handle);
3127 err = PTR_ERR(inode);
3128 goto err_free_sd;
3129 }
3130
3131 if (encryption_required) {
3132 struct qstr istr;
3133 struct ext4_str ostr;
3134
3135 istr.name = (const unsigned char *) symname;
3136 istr.len = len;
3137 ostr.name = sd->encrypted_path;
3138 ostr.len = disk_link.len;
3139 err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3140 if (err < 0)
3141 goto err_drop_inode;
3142 sd->len = cpu_to_le16(ostr.len);
3143 disk_link.name = (char *) sd;
3144 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3145 }
3146
3147 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3148 if (!encryption_required)
3149 inode->i_op = &ext4_symlink_inode_operations;
3150 ext4_set_aops(inode);
3151 /*
3152 * We cannot call page_symlink() with transaction started
3153 * because it calls into ext4_write_begin() which can wait
3154 * for transaction commit if we are running out of space
3155 * and thus we deadlock. So we have to stop transaction now
3156 * and restart it when symlink contents is written.
3157 *
3158 * To keep fs consistent in case of crash, we have to put inode
3159 * to orphan list in the mean time.
3160 */
3161 drop_nlink(inode);
3162 err = ext4_orphan_add(handle, inode);
3163 ext4_journal_stop(handle);
3164 handle = NULL;
3165 if (err)
3166 goto err_drop_inode;
3167 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3168 if (err)
3169 goto err_drop_inode;
3170 /*
3171 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3172 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3173 */
3174 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3175 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3176 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3177 if (IS_ERR(handle)) {
3178 err = PTR_ERR(handle);
3179 handle = NULL;
3180 goto err_drop_inode;
3181 }
3182 set_nlink(inode, 1);
3183 err = ext4_orphan_del(handle, inode);
3184 if (err)
3185 goto err_drop_inode;
3186 } else {
3187 /* clear the extent format for fast symlink */
3188 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3189 if (!encryption_required) {
3190 inode->i_op = &ext4_fast_symlink_inode_operations;
3191 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3192 }
3193 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3194 disk_link.len);
3195 inode->i_size = disk_link.len - 1;
3196 }
3197 EXT4_I(inode)->i_disksize = inode->i_size;
3198 err = ext4_add_nondir(handle, dentry, inode);
3199 if (!err && IS_DIRSYNC(dir))
3200 ext4_handle_sync(handle);
3201
3202 if (handle)
3203 ext4_journal_stop(handle);
3204 kfree(sd);
3205 return err;
3206 err_drop_inode:
3207 if (handle)
3208 ext4_journal_stop(handle);
3209 clear_nlink(inode);
3210 unlock_new_inode(inode);
3211 iput(inode);
3212 err_free_sd:
3213 kfree(sd);
3214 return err;
3215 }
3216
3217 static int ext4_link(struct dentry *old_dentry,
3218 struct inode *dir, struct dentry *dentry)
3219 {
3220 handle_t *handle;
3221 struct inode *inode = d_inode(old_dentry);
3222 int err, retries = 0;
3223
3224 if (inode->i_nlink >= EXT4_LINK_MAX)
3225 return -EMLINK;
3226 if (ext4_encrypted_inode(dir) &&
3227 !ext4_is_child_context_consistent_with_parent(dir, inode))
3228 return -EPERM;
3229 err = dquot_initialize(dir);
3230 if (err)
3231 return err;
3232
3233 retry:
3234 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3235 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3236 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3237 if (IS_ERR(handle))
3238 return PTR_ERR(handle);
3239
3240 if (IS_DIRSYNC(dir))
3241 ext4_handle_sync(handle);
3242
3243 inode->i_ctime = ext4_current_time(inode);
3244 ext4_inc_count(handle, inode);
3245 ihold(inode);
3246
3247 err = ext4_add_entry(handle, dentry, inode);
3248 if (!err) {
3249 ext4_mark_inode_dirty(handle, inode);
3250 /* this can happen only for tmpfile being
3251 * linked the first time
3252 */
3253 if (inode->i_nlink == 1)
3254 ext4_orphan_del(handle, inode);
3255 d_instantiate(dentry, inode);
3256 } else {
3257 drop_nlink(inode);
3258 iput(inode);
3259 }
3260 ext4_journal_stop(handle);
3261 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3262 goto retry;
3263 return err;
3264 }
3265
3266
3267 /*
3268 * Try to find buffer head where contains the parent block.
3269 * It should be the inode block if it is inlined or the 1st block
3270 * if it is a normal dir.
3271 */
3272 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3273 struct inode *inode,
3274 int *retval,
3275 struct ext4_dir_entry_2 **parent_de,
3276 int *inlined)
3277 {
3278 struct buffer_head *bh;
3279
3280 if (!ext4_has_inline_data(inode)) {
3281 bh = ext4_read_dirblock(inode, 0, EITHER);
3282 if (IS_ERR(bh)) {
3283 *retval = PTR_ERR(bh);
3284 return NULL;
3285 }
3286 *parent_de = ext4_next_entry(
3287 (struct ext4_dir_entry_2 *)bh->b_data,
3288 inode->i_sb->s_blocksize);
3289 return bh;
3290 }
3291
3292 *inlined = 1;
3293 return ext4_get_first_inline_block(inode, parent_de, retval);
3294 }
3295
3296 struct ext4_renament {
3297 struct inode *dir;
3298 struct dentry *dentry;
3299 struct inode *inode;
3300 bool is_dir;
3301 int dir_nlink_delta;
3302
3303 /* entry for "dentry" */
3304 struct buffer_head *bh;
3305 struct ext4_dir_entry_2 *de;
3306 int inlined;
3307
3308 /* entry for ".." in inode if it's a directory */
3309 struct buffer_head *dir_bh;
3310 struct ext4_dir_entry_2 *parent_de;
3311 int dir_inlined;
3312 };
3313
3314 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3315 {
3316 int retval;
3317
3318 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3319 &retval, &ent->parent_de,
3320 &ent->dir_inlined);
3321 if (!ent->dir_bh)
3322 return retval;
3323 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3324 return -EFSCORRUPTED;
3325 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3326 return ext4_journal_get_write_access(handle, ent->dir_bh);
3327 }
3328
3329 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3330 unsigned dir_ino)
3331 {
3332 int retval;
3333
3334 ent->parent_de->inode = cpu_to_le32(dir_ino);
3335 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3336 if (!ent->dir_inlined) {
3337 if (is_dx(ent->inode)) {
3338 retval = ext4_handle_dirty_dx_node(handle,
3339 ent->inode,
3340 ent->dir_bh);
3341 } else {
3342 retval = ext4_handle_dirty_dirent_node(handle,
3343 ent->inode,
3344 ent->dir_bh);
3345 }
3346 } else {
3347 retval = ext4_mark_inode_dirty(handle, ent->inode);
3348 }
3349 if (retval) {
3350 ext4_std_error(ent->dir->i_sb, retval);
3351 return retval;
3352 }
3353 return 0;
3354 }
3355
3356 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3357 unsigned ino, unsigned file_type)
3358 {
3359 int retval;
3360
3361 BUFFER_TRACE(ent->bh, "get write access");
3362 retval = ext4_journal_get_write_access(handle, ent->bh);
3363 if (retval)
3364 return retval;
3365 ent->de->inode = cpu_to_le32(ino);
3366 if (ext4_has_feature_filetype(ent->dir->i_sb))
3367 ent->de->file_type = file_type;
3368 ent->dir->i_version++;
3369 ent->dir->i_ctime = ent->dir->i_mtime =
3370 ext4_current_time(ent->dir);
3371 ext4_mark_inode_dirty(handle, ent->dir);
3372 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3373 if (!ent->inlined) {
3374 retval = ext4_handle_dirty_dirent_node(handle,
3375 ent->dir, ent->bh);
3376 if (unlikely(retval)) {
3377 ext4_std_error(ent->dir->i_sb, retval);
3378 return retval;
3379 }
3380 }
3381 brelse(ent->bh);
3382 ent->bh = NULL;
3383
3384 return 0;
3385 }
3386
3387 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3388 const struct qstr *d_name)
3389 {
3390 int retval = -ENOENT;
3391 struct buffer_head *bh;
3392 struct ext4_dir_entry_2 *de;
3393
3394 bh = ext4_find_entry(dir, d_name, &de, NULL);
3395 if (IS_ERR(bh))
3396 return PTR_ERR(bh);
3397 if (bh) {
3398 retval = ext4_delete_entry(handle, dir, de, bh);
3399 brelse(bh);
3400 }
3401 return retval;
3402 }
3403
3404 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3405 int force_reread)
3406 {
3407 int retval;
3408 /*
3409 * ent->de could have moved from under us during htree split, so make
3410 * sure that we are deleting the right entry. We might also be pointing
3411 * to a stale entry in the unused part of ent->bh so just checking inum
3412 * and the name isn't enough.
3413 */
3414 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3415 ent->de->name_len != ent->dentry->d_name.len ||
3416 strncmp(ent->de->name, ent->dentry->d_name.name,
3417 ent->de->name_len) ||
3418 force_reread) {
3419 retval = ext4_find_delete_entry(handle, ent->dir,
3420 &ent->dentry->d_name);
3421 } else {
3422 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3423 if (retval == -ENOENT) {
3424 retval = ext4_find_delete_entry(handle, ent->dir,
3425 &ent->dentry->d_name);
3426 }
3427 }
3428
3429 if (retval) {
3430 ext4_warning_inode(ent->dir,
3431 "Deleting old file: nlink %d, error=%d",
3432 ent->dir->i_nlink, retval);
3433 }
3434 }
3435
3436 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3437 {
3438 if (ent->dir_nlink_delta) {
3439 if (ent->dir_nlink_delta == -1)
3440 ext4_dec_count(handle, ent->dir);
3441 else
3442 ext4_inc_count(handle, ent->dir);
3443 ext4_mark_inode_dirty(handle, ent->dir);
3444 }
3445 }
3446
3447 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3448 int credits, handle_t **h)
3449 {
3450 struct inode *wh;
3451 handle_t *handle;
3452 int retries = 0;
3453
3454 /*
3455 * for inode block, sb block, group summaries,
3456 * and inode bitmap
3457 */
3458 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3459 EXT4_XATTR_TRANS_BLOCKS + 4);
3460 retry:
3461 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3462 &ent->dentry->d_name, 0, NULL,
3463 EXT4_HT_DIR, credits);
3464
3465 handle = ext4_journal_current_handle();
3466 if (IS_ERR(wh)) {
3467 if (handle)
3468 ext4_journal_stop(handle);
3469 if (PTR_ERR(wh) == -ENOSPC &&
3470 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3471 goto retry;
3472 } else {
3473 *h = handle;
3474 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3475 wh->i_op = &ext4_special_inode_operations;
3476 }
3477 return wh;
3478 }
3479
3480 /*
3481 * Anybody can rename anything with this: the permission checks are left to the
3482 * higher-level routines.
3483 *
3484 * n.b. old_{dentry,inode) refers to the source dentry/inode
3485 * while new_{dentry,inode) refers to the destination dentry/inode
3486 * This comes from rename(const char *oldpath, const char *newpath)
3487 */
3488 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3489 struct inode *new_dir, struct dentry *new_dentry,
3490 unsigned int flags)
3491 {
3492 handle_t *handle = NULL;
3493 struct ext4_renament old = {
3494 .dir = old_dir,
3495 .dentry = old_dentry,
3496 .inode = d_inode(old_dentry),
3497 };
3498 struct ext4_renament new = {
3499 .dir = new_dir,
3500 .dentry = new_dentry,
3501 .inode = d_inode(new_dentry),
3502 };
3503 int force_reread;
3504 int retval;
3505 struct inode *whiteout = NULL;
3506 int credits;
3507 u8 old_file_type;
3508
3509 retval = dquot_initialize(old.dir);
3510 if (retval)
3511 return retval;
3512 retval = dquot_initialize(new.dir);
3513 if (retval)
3514 return retval;
3515
3516 /* Initialize quotas before so that eventual writes go
3517 * in separate transaction */
3518 if (new.inode) {
3519 retval = dquot_initialize(new.inode);
3520 if (retval)
3521 return retval;
3522 }
3523
3524 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3525 if (IS_ERR(old.bh))
3526 return PTR_ERR(old.bh);
3527 /*
3528 * Check for inode number is _not_ due to possible IO errors.
3529 * We might rmdir the source, keep it as pwd of some process
3530 * and merrily kill the link to whatever was created under the
3531 * same name. Goodbye sticky bit ;-<
3532 */
3533 retval = -ENOENT;
3534 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3535 goto end_rename;
3536
3537 if ((old.dir != new.dir) &&
3538 ext4_encrypted_inode(new.dir) &&
3539 !ext4_is_child_context_consistent_with_parent(new.dir,
3540 old.inode)) {
3541 retval = -EPERM;
3542 goto end_rename;
3543 }
3544
3545 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3546 &new.de, &new.inlined);
3547 if (IS_ERR(new.bh)) {
3548 retval = PTR_ERR(new.bh);
3549 new.bh = NULL;
3550 goto end_rename;
3551 }
3552 if (new.bh) {
3553 if (!new.inode) {
3554 brelse(new.bh);
3555 new.bh = NULL;
3556 }
3557 }
3558 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3559 ext4_alloc_da_blocks(old.inode);
3560
3561 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3562 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3563 if (!(flags & RENAME_WHITEOUT)) {
3564 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3565 if (IS_ERR(handle)) {
3566 retval = PTR_ERR(handle);
3567 handle = NULL;
3568 goto end_rename;
3569 }
3570 } else {
3571 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3572 if (IS_ERR(whiteout)) {
3573 retval = PTR_ERR(whiteout);
3574 whiteout = NULL;
3575 goto end_rename;
3576 }
3577 }
3578
3579 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3580 ext4_handle_sync(handle);
3581
3582 if (S_ISDIR(old.inode->i_mode)) {
3583 if (new.inode) {
3584 retval = -ENOTEMPTY;
3585 if (!ext4_empty_dir(new.inode))
3586 goto end_rename;
3587 } else {
3588 retval = -EMLINK;
3589 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3590 goto end_rename;
3591 }
3592 retval = ext4_rename_dir_prepare(handle, &old);
3593 if (retval)
3594 goto end_rename;
3595 }
3596 /*
3597 * If we're renaming a file within an inline_data dir and adding or
3598 * setting the new dirent causes a conversion from inline_data to
3599 * extents/blockmap, we need to force the dirent delete code to
3600 * re-read the directory, or else we end up trying to delete a dirent
3601 * from what is now the extent tree root (or a block map).
3602 */
3603 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3604 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3605
3606 old_file_type = old.de->file_type;
3607 if (whiteout) {
3608 /*
3609 * Do this before adding a new entry, so the old entry is sure
3610 * to be still pointing to the valid old entry.
3611 */
3612 retval = ext4_setent(handle, &old, whiteout->i_ino,
3613 EXT4_FT_CHRDEV);
3614 if (retval)
3615 goto end_rename;
3616 ext4_mark_inode_dirty(handle, whiteout);
3617 }
3618 if (!new.bh) {
3619 retval = ext4_add_entry(handle, new.dentry, old.inode);
3620 if (retval)
3621 goto end_rename;
3622 } else {
3623 retval = ext4_setent(handle, &new,
3624 old.inode->i_ino, old_file_type);
3625 if (retval)
3626 goto end_rename;
3627 }
3628 if (force_reread)
3629 force_reread = !ext4_test_inode_flag(new.dir,
3630 EXT4_INODE_INLINE_DATA);
3631
3632 /*
3633 * Like most other Unix systems, set the ctime for inodes on a
3634 * rename.
3635 */
3636 old.inode->i_ctime = ext4_current_time(old.inode);
3637 ext4_mark_inode_dirty(handle, old.inode);
3638
3639 if (!whiteout) {
3640 /*
3641 * ok, that's it
3642 */
3643 ext4_rename_delete(handle, &old, force_reread);
3644 }
3645
3646 if (new.inode) {
3647 ext4_dec_count(handle, new.inode);
3648 new.inode->i_ctime = ext4_current_time(new.inode);
3649 }
3650 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3651 ext4_update_dx_flag(old.dir);
3652 if (old.dir_bh) {
3653 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3654 if (retval)
3655 goto end_rename;
3656
3657 ext4_dec_count(handle, old.dir);
3658 if (new.inode) {
3659 /* checked ext4_empty_dir above, can't have another
3660 * parent, ext4_dec_count() won't work for many-linked
3661 * dirs */
3662 clear_nlink(new.inode);
3663 } else {
3664 ext4_inc_count(handle, new.dir);
3665 ext4_update_dx_flag(new.dir);
3666 ext4_mark_inode_dirty(handle, new.dir);
3667 }
3668 }
3669 ext4_mark_inode_dirty(handle, old.dir);
3670 if (new.inode) {
3671 ext4_mark_inode_dirty(handle, new.inode);
3672 if (!new.inode->i_nlink)
3673 ext4_orphan_add(handle, new.inode);
3674 }
3675 retval = 0;
3676
3677 end_rename:
3678 brelse(old.dir_bh);
3679 brelse(old.bh);
3680 brelse(new.bh);
3681 if (whiteout) {
3682 if (retval)
3683 drop_nlink(whiteout);
3684 unlock_new_inode(whiteout);
3685 iput(whiteout);
3686 }
3687 if (handle)
3688 ext4_journal_stop(handle);
3689 return retval;
3690 }
3691
3692 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3693 struct inode *new_dir, struct dentry *new_dentry)
3694 {
3695 handle_t *handle = NULL;
3696 struct ext4_renament old = {
3697 .dir = old_dir,
3698 .dentry = old_dentry,
3699 .inode = d_inode(old_dentry),
3700 };
3701 struct ext4_renament new = {
3702 .dir = new_dir,
3703 .dentry = new_dentry,
3704 .inode = d_inode(new_dentry),
3705 };
3706 u8 new_file_type;
3707 int retval;
3708
3709 if ((ext4_encrypted_inode(old_dir) ||
3710 ext4_encrypted_inode(new_dir)) &&
3711 (old_dir != new_dir) &&
3712 (!ext4_is_child_context_consistent_with_parent(new_dir,
3713 old.inode) ||
3714 !ext4_is_child_context_consistent_with_parent(old_dir,
3715 new.inode)))
3716 return -EPERM;
3717
3718 retval = dquot_initialize(old.dir);
3719 if (retval)
3720 return retval;
3721 retval = dquot_initialize(new.dir);
3722 if (retval)
3723 return retval;
3724
3725 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3726 &old.de, &old.inlined);
3727 if (IS_ERR(old.bh))
3728 return PTR_ERR(old.bh);
3729 /*
3730 * Check for inode number is _not_ due to possible IO errors.
3731 * We might rmdir the source, keep it as pwd of some process
3732 * and merrily kill the link to whatever was created under the
3733 * same name. Goodbye sticky bit ;-<
3734 */
3735 retval = -ENOENT;
3736 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3737 goto end_rename;
3738
3739 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3740 &new.de, &new.inlined);
3741 if (IS_ERR(new.bh)) {
3742 retval = PTR_ERR(new.bh);
3743 new.bh = NULL;
3744 goto end_rename;
3745 }
3746
3747 /* RENAME_EXCHANGE case: old *and* new must both exist */
3748 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3749 goto end_rename;
3750
3751 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3752 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3753 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3754 if (IS_ERR(handle)) {
3755 retval = PTR_ERR(handle);
3756 handle = NULL;
3757 goto end_rename;
3758 }
3759
3760 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3761 ext4_handle_sync(handle);
3762
3763 if (S_ISDIR(old.inode->i_mode)) {
3764 old.is_dir = true;
3765 retval = ext4_rename_dir_prepare(handle, &old);
3766 if (retval)
3767 goto end_rename;
3768 }
3769 if (S_ISDIR(new.inode->i_mode)) {
3770 new.is_dir = true;
3771 retval = ext4_rename_dir_prepare(handle, &new);
3772 if (retval)
3773 goto end_rename;
3774 }
3775
3776 /*
3777 * Other than the special case of overwriting a directory, parents'
3778 * nlink only needs to be modified if this is a cross directory rename.
3779 */
3780 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3781 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3782 new.dir_nlink_delta = -old.dir_nlink_delta;
3783 retval = -EMLINK;
3784 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3785 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3786 goto end_rename;
3787 }
3788
3789 new_file_type = new.de->file_type;
3790 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3791 if (retval)
3792 goto end_rename;
3793
3794 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3795 if (retval)
3796 goto end_rename;
3797
3798 /*
3799 * Like most other Unix systems, set the ctime for inodes on a
3800 * rename.
3801 */
3802 old.inode->i_ctime = ext4_current_time(old.inode);
3803 new.inode->i_ctime = ext4_current_time(new.inode);
3804 ext4_mark_inode_dirty(handle, old.inode);
3805 ext4_mark_inode_dirty(handle, new.inode);
3806
3807 if (old.dir_bh) {
3808 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3809 if (retval)
3810 goto end_rename;
3811 }
3812 if (new.dir_bh) {
3813 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3814 if (retval)
3815 goto end_rename;
3816 }
3817 ext4_update_dir_count(handle, &old);
3818 ext4_update_dir_count(handle, &new);
3819 retval = 0;
3820
3821 end_rename:
3822 brelse(old.dir_bh);
3823 brelse(new.dir_bh);
3824 brelse(old.bh);
3825 brelse(new.bh);
3826 if (handle)
3827 ext4_journal_stop(handle);
3828 return retval;
3829 }
3830
3831 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3832 struct inode *new_dir, struct dentry *new_dentry,
3833 unsigned int flags)
3834 {
3835 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3836 return -EINVAL;
3837
3838 if (flags & RENAME_EXCHANGE) {
3839 return ext4_cross_rename(old_dir, old_dentry,
3840 new_dir, new_dentry);
3841 }
3842
3843 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3844 }
3845
3846 /*
3847 * directories can handle most operations...
3848 */
3849 const struct inode_operations ext4_dir_inode_operations = {
3850 .create = ext4_create,
3851 .lookup = ext4_lookup,
3852 .link = ext4_link,
3853 .unlink = ext4_unlink,
3854 .symlink = ext4_symlink,
3855 .mkdir = ext4_mkdir,
3856 .rmdir = ext4_rmdir,
3857 .mknod = ext4_mknod,
3858 .tmpfile = ext4_tmpfile,
3859 .rename2 = ext4_rename2,
3860 .setattr = ext4_setattr,
3861 .setxattr = generic_setxattr,
3862 .getxattr = generic_getxattr,
3863 .listxattr = ext4_listxattr,
3864 .removexattr = generic_removexattr,
3865 .get_acl = ext4_get_acl,
3866 .set_acl = ext4_set_acl,
3867 .fiemap = ext4_fiemap,
3868 };
3869
3870 const struct inode_operations ext4_special_inode_operations = {
3871 .setattr = ext4_setattr,
3872 .setxattr = generic_setxattr,
3873 .getxattr = generic_getxattr,
3874 .listxattr = ext4_listxattr,
3875 .removexattr = generic_removexattr,
3876 .get_acl = ext4_get_acl,
3877 .set_acl = ext4_set_acl,
3878 };
Linux with added FPC-III support