?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
HID: hiddev: Fix slab-out-of-bounds write in hiddev_ioctl_usage()
[linux/fpc-iii.git] / fs / ext4 / namei.c
blob061b026e464c5166bce8dc35f9245d0b4b172b9c
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 brelse(bh);
128 return ERR_PTR(-EFSCORRUPTED);
129 }
130 if (!ext4_has_metadata_csum(inode->i_sb) ||
131 buffer_verified(bh))
132 return bh;
133
134 /*
135 * An empty leaf block can get mistaken for a index block; for
136 * this reason, we can only check the index checksum when the
137 * caller is sure it should be an index block.
138 */
139 if (is_dx_block && type == INDEX) {
140 if (ext4_dx_csum_verify(inode, dirent))
141 set_buffer_verified(bh);
142 else {
143 ext4_error_inode(inode, func, line, block,
144 "Directory index failed checksum");
145 brelse(bh);
146 return ERR_PTR(-EFSBADCRC);
147 }
148 }
149 if (!is_dx_block) {
150 if (ext4_dirent_csum_verify(inode, dirent))
151 set_buffer_verified(bh);
152 else {
153 ext4_error_inode(inode, func, line, block,
154 "Directory block failed checksum");
155 brelse(bh);
156 return ERR_PTR(-EFSBADCRC);
157 }
158 }
159 return bh;
160 }
161
162 #ifndef assert
163 #define assert(test) J_ASSERT(test)
164 #endif
165
166 #ifdef DX_DEBUG
167 #define dxtrace(command) command
168 #else
169 #define dxtrace(command)
170 #endif
171
172 struct fake_dirent
173 {
174 __le32 inode;
175 __le16 rec_len;
176 u8 name_len;
177 u8 file_type;
178 };
179
180 struct dx_countlimit
181 {
182 __le16 limit;
183 __le16 count;
184 };
185
186 struct dx_entry
187 {
188 __le32 hash;
189 __le32 block;
190 };
191
192 /*
193 * dx_root_info is laid out so that if it should somehow get overlaid by a
194 * dirent the two low bits of the hash version will be zero. Therefore, the
195 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
196 */
197
198 struct dx_root
199 {
200 struct fake_dirent dot;
201 char dot_name[4];
202 struct fake_dirent dotdot;
203 char dotdot_name[4];
204 struct dx_root_info
205 {
206 __le32 reserved_zero;
207 u8 hash_version;
208 u8 info_length; /* 8 */
209 u8 indirect_levels;
210 u8 unused_flags;
211 }
212 info;
213 struct dx_entry entries[0];
214 };
215
216 struct dx_node
217 {
218 struct fake_dirent fake;
219 struct dx_entry entries[0];
220 };
221
222
223 struct dx_frame
224 {
225 struct buffer_head *bh;
226 struct dx_entry *entries;
227 struct dx_entry *at;
228 };
229
230 struct dx_map_entry
231 {
232 u32 hash;
233 u16 offs;
234 u16 size;
235 };
236
237 /*
238 * This goes at the end of each htree block.
239 */
240 struct dx_tail {
241 u32 dt_reserved;
242 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
243 };
244
245 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
246 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
247 static inline unsigned dx_get_hash(struct dx_entry *entry);
248 static void dx_set_hash(struct dx_entry *entry, unsigned value);
249 static unsigned dx_get_count(struct dx_entry *entries);
250 static unsigned dx_get_limit(struct dx_entry *entries);
251 static void dx_set_count(struct dx_entry *entries, unsigned value);
252 static void dx_set_limit(struct dx_entry *entries, unsigned value);
253 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
254 static unsigned dx_node_limit(struct inode *dir);
255 static struct dx_frame *dx_probe(struct ext4_filename *fname,
256 struct inode *dir,
257 struct dx_hash_info *hinfo,
258 struct dx_frame *frame);
259 static void dx_release(struct dx_frame *frames);
260 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
261 unsigned blocksize, struct dx_hash_info *hinfo,
262 struct dx_map_entry map[]);
263 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
264 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
265 struct dx_map_entry *offsets, int count, unsigned blocksize);
266 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
267 static void dx_insert_block(struct dx_frame *frame,
268 u32 hash, ext4_lblk_t block);
269 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
270 struct dx_frame *frame,
271 struct dx_frame *frames,
272 __u32 *start_hash);
273 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
274 struct ext4_filename *fname,
275 struct ext4_dir_entry_2 **res_dir);
276 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
277 struct dentry *dentry, struct inode *inode);
278
279 /* checksumming functions */
280 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
281 unsigned int blocksize)
282 {
283 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
284 t->det_rec_len = ext4_rec_len_to_disk(
285 sizeof(struct ext4_dir_entry_tail), blocksize);
286 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
287 }
288
289 /* Walk through a dirent block to find a checksum "dirent" at the tail */
290 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
291 struct ext4_dir_entry *de)
292 {
293 struct ext4_dir_entry_tail *t;
294
295 #ifdef PARANOID
296 struct ext4_dir_entry *d, *top;
297
298 d = de;
299 top = (struct ext4_dir_entry *)(((void *)de) +
300 (EXT4_BLOCK_SIZE(inode->i_sb) -
301 sizeof(struct ext4_dir_entry_tail)));
302 while (d < top && d->rec_len)
303 d = (struct ext4_dir_entry *)(((void *)d) +
304 le16_to_cpu(d->rec_len));
305
306 if (d != top)
307 return NULL;
308
309 t = (struct ext4_dir_entry_tail *)d;
310 #else
311 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
312 #endif
313
314 if (t->det_reserved_zero1 ||
315 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
316 t->det_reserved_zero2 ||
317 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
318 return NULL;
319
320 return t;
321 }
322
323 static __le32 ext4_dirent_csum(struct inode *inode,
324 struct ext4_dir_entry *dirent, int size)
325 {
326 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
327 struct ext4_inode_info *ei = EXT4_I(inode);
328 __u32 csum;
329
330 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
331 return cpu_to_le32(csum);
332 }
333
334 #define warn_no_space_for_csum(inode) \
335 __warn_no_space_for_csum((inode), __func__, __LINE__)
336
337 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
338 unsigned int line)
339 {
340 __ext4_warning_inode(inode, func, line,
341 "No space for directory leaf checksum. Please run e2fsck -D.");
342 }
343
344 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
345 {
346 struct ext4_dir_entry_tail *t;
347
348 if (!ext4_has_metadata_csum(inode->i_sb))
349 return 1;
350
351 t = get_dirent_tail(inode, dirent);
352 if (!t) {
353 warn_no_space_for_csum(inode);
354 return 0;
355 }
356
357 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
358 (void *)t - (void *)dirent))
359 return 0;
360
361 return 1;
362 }
363
364 static void ext4_dirent_csum_set(struct inode *inode,
365 struct ext4_dir_entry *dirent)
366 {
367 struct ext4_dir_entry_tail *t;
368
369 if (!ext4_has_metadata_csum(inode->i_sb))
370 return;
371
372 t = get_dirent_tail(inode, dirent);
373 if (!t) {
374 warn_no_space_for_csum(inode);
375 return;
376 }
377
378 t->det_checksum = ext4_dirent_csum(inode, dirent,
379 (void *)t - (void *)dirent);
380 }
381
382 int ext4_handle_dirty_dirent_node(handle_t *handle,
383 struct inode *inode,
384 struct buffer_head *bh)
385 {
386 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
387 return ext4_handle_dirty_metadata(handle, inode, bh);
388 }
389
390 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
391 struct ext4_dir_entry *dirent,
392 int *offset)
393 {
394 struct ext4_dir_entry *dp;
395 struct dx_root_info *root;
396 int count_offset;
397
398 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
399 count_offset = 8;
400 else if (le16_to_cpu(dirent->rec_len) == 12) {
401 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
402 if (le16_to_cpu(dp->rec_len) !=
403 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
404 return NULL;
405 root = (struct dx_root_info *)(((void *)dp + 12));
406 if (root->reserved_zero ||
407 root->info_length != sizeof(struct dx_root_info))
408 return NULL;
409 count_offset = 32;
410 } else
411 return NULL;
412
413 if (offset)
414 *offset = count_offset;
415 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
416 }
417
418 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
419 int count_offset, int count, struct dx_tail *t)
420 {
421 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
422 struct ext4_inode_info *ei = EXT4_I(inode);
423 __u32 csum;
424 int size;
425 __u32 dummy_csum = 0;
426 int offset = offsetof(struct dx_tail, dt_checksum);
427
428 size = count_offset + (count * sizeof(struct dx_entry));
429 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
431 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
432
433 return cpu_to_le32(csum);
434 }
435
436 static int ext4_dx_csum_verify(struct inode *inode,
437 struct ext4_dir_entry *dirent)
438 {
439 struct dx_countlimit *c;
440 struct dx_tail *t;
441 int count_offset, limit, count;
442
443 if (!ext4_has_metadata_csum(inode->i_sb))
444 return 1;
445
446 c = get_dx_countlimit(inode, dirent, &count_offset);
447 if (!c) {
448 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
449 return 1;
450 }
451 limit = le16_to_cpu(c->limit);
452 count = le16_to_cpu(c->count);
453 if (count_offset + (limit * sizeof(struct dx_entry)) >
454 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
455 warn_no_space_for_csum(inode);
456 return 1;
457 }
458 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
459
460 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
461 count, t))
462 return 0;
463 return 1;
464 }
465
466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
467 {
468 struct dx_countlimit *c;
469 struct dx_tail *t;
470 int count_offset, limit, count;
471
472 if (!ext4_has_metadata_csum(inode->i_sb))
473 return;
474
475 c = get_dx_countlimit(inode, dirent, &count_offset);
476 if (!c) {
477 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
478 return;
479 }
480 limit = le16_to_cpu(c->limit);
481 count = le16_to_cpu(c->count);
482 if (count_offset + (limit * sizeof(struct dx_entry)) >
483 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
484 warn_no_space_for_csum(inode);
485 return;
486 }
487 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
488
489 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
490 }
491
492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493 struct inode *inode,
494 struct buffer_head *bh)
495 {
496 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
497 return ext4_handle_dirty_metadata(handle, inode, bh);
498 }
499
500 /*
501 * p is at least 6 bytes before the end of page
502 */
503 static inline struct ext4_dir_entry_2 *
504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
505 {
506 return (struct ext4_dir_entry_2 *)((char *)p +
507 ext4_rec_len_from_disk(p->rec_len, blocksize));
508 }
509
510 /*
511 * Future: use high four bits of block for coalesce-on-delete flags
512 * Mask them off for now.
513 */
514
515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
516 {
517 return le32_to_cpu(entry->block) & 0x00ffffff;
518 }
519
520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
521 {
522 entry->block = cpu_to_le32(value);
523 }
524
525 static inline unsigned dx_get_hash(struct dx_entry *entry)
526 {
527 return le32_to_cpu(entry->hash);
528 }
529
530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
531 {
532 entry->hash = cpu_to_le32(value);
533 }
534
535 static inline unsigned dx_get_count(struct dx_entry *entries)
536 {
537 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
538 }
539
540 static inline unsigned dx_get_limit(struct dx_entry *entries)
541 {
542 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
543 }
544
545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
546 {
547 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
548 }
549
550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
551 {
552 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
553 }
554
555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
556 {
557 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
558 EXT4_DIR_REC_LEN(2) - infosize;
559
560 if (ext4_has_metadata_csum(dir->i_sb))
561 entry_space -= sizeof(struct dx_tail);
562 return entry_space / sizeof(struct dx_entry);
563 }
564
565 static inline unsigned dx_node_limit(struct inode *dir)
566 {
567 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568
569 if (ext4_has_metadata_csum(dir->i_sb))
570 entry_space -= sizeof(struct dx_tail);
571 return entry_space / sizeof(struct dx_entry);
572 }
573
574 /*
575 * Debug
576 */
577 #ifdef DX_DEBUG
578 static void dx_show_index(char * label, struct dx_entry *entries)
579 {
580 int i, n = dx_get_count (entries);
581 printk(KERN_DEBUG "%s index ", label);
582 for (i = 0; i < n; i++) {
583 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
584 0, (unsigned long)dx_get_block(entries + i));
585 }
586 printk("\n");
587 }
588
589 struct stats
590 {
591 unsigned names;
592 unsigned space;
593 unsigned bcount;
594 };
595
596 static struct stats dx_show_leaf(struct inode *dir,
597 struct dx_hash_info *hinfo,
598 struct ext4_dir_entry_2 *de,
599 int size, int show_names)
600 {
601 unsigned names = 0, space = 0;
602 char *base = (char *) de;
603 struct dx_hash_info h = *hinfo;
604
605 printk("names: ");
606 while ((char *) de < base + size)
607 {
608 if (de->inode)
609 {
610 if (show_names)
611 {
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
613 int len;
614 char *name;
615 struct ext4_str fname_crypto_str
616 = {.name = NULL, .len = 0};
617 int res = 0;
618
619 name = de->name;
620 len = de->name_len;
621 if (ext4_encrypted_inode(inode))
622 res = ext4_get_encryption_info(dir);
623 if (res) {
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
626 }
627 if (ctx == NULL) {
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
630 de->name_len, &h);
631 printk("%*.s:(U)%x.%u ", len,
632 name, h.hash,
633 (unsigned) ((char *) de
634 - base));
635 } else {
636 /* Directory is encrypted */
637 res = ext4_fname_crypto_alloc_buffer(
638 ctx, de->name_len,
639 &fname_crypto_str);
640 if (res < 0) {
641 printk(KERN_WARNING "Error "
642 "allocating crypto "
643 "buffer--skipping "
644 "crypto\n");
645 ctx = NULL;
646 }
647 res = ext4_fname_disk_to_usr(ctx, NULL, de,
648 &fname_crypto_str);
649 if (res < 0) {
650 printk(KERN_WARNING "Error "
651 "converting filename "
652 "from disk to usr"
653 "\n");
654 name = "??";
655 len = 2;
656 } else {
657 name = fname_crypto_str.name;
658 len = fname_crypto_str.len;
659 }
660 ext4fs_dirhash(de->name, de->name_len,
661 &h);
662 printk("%*.s:(E)%x.%u ", len, name,
663 h.hash, (unsigned) ((char *) de
664 - base));
665 ext4_fname_crypto_free_buffer(
666 &fname_crypto_str);
667 }
668 #else
669 int len = de->name_len;
670 char *name = de->name;
671 ext4fs_dirhash(de->name, de->name_len, &h);
672 printk("%*.s:%x.%u ", len, name, h.hash,
673 (unsigned) ((char *) de - base));
674 #endif
675 }
676 space += EXT4_DIR_REC_LEN(de->name_len);
677 names++;
678 }
679 de = ext4_next_entry(de, size);
680 }
681 printk("(%i)\n", names);
682 return (struct stats) { names, space, 1 };
683 }
684
685 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
686 struct dx_entry *entries, int levels)
687 {
688 unsigned blocksize = dir->i_sb->s_blocksize;
689 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
690 unsigned bcount = 0;
691 struct buffer_head *bh;
692 printk("%i indexed blocks...\n", count);
693 for (i = 0; i < count; i++, entries++)
694 {
695 ext4_lblk_t block = dx_get_block(entries);
696 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
697 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
698 struct stats stats;
699 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
700 bh = ext4_bread(NULL,dir, block, 0);
701 if (!bh || IS_ERR(bh))
702 continue;
703 stats = levels?
704 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
705 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
706 bh->b_data, blocksize, 0);
707 names += stats.names;
708 space += stats.space;
709 bcount += stats.bcount;
710 brelse(bh);
711 }
712 if (bcount)
713 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
714 levels ? "" : " ", names, space/bcount,
715 (space/bcount)*100/blocksize);
716 return (struct stats) { names, space, bcount};
717 }
718 #endif /* DX_DEBUG */
719
720 /*
721 * Probe for a directory leaf block to search.
722 *
723 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
724 * error in the directory index, and the caller should fall back to
725 * searching the directory normally. The callers of dx_probe **MUST**
726 * check for this error code, and make sure it never gets reflected
727 * back to userspace.
728 */
729 static struct dx_frame *
730 dx_probe(struct ext4_filename *fname, struct inode *dir,
731 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
732 {
733 unsigned count, indirect;
734 struct dx_entry *at, *entries, *p, *q, *m;
735 struct dx_root *root;
736 struct dx_frame *frame = frame_in;
737 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
738 u32 hash;
739
740 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
741 if (IS_ERR(frame->bh))
742 return (struct dx_frame *) frame->bh;
743
744 root = (struct dx_root *) frame->bh->b_data;
745 if (root->info.hash_version != DX_HASH_TEA &&
746 root->info.hash_version != DX_HASH_HALF_MD4 &&
747 root->info.hash_version != DX_HASH_LEGACY) {
748 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
749 root->info.hash_version);
750 goto fail;
751 }
752 if (fname)
753 hinfo = &fname->hinfo;
754 hinfo->hash_version = root->info.hash_version;
755 if (hinfo->hash_version <= DX_HASH_TEA)
756 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
757 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
758 if (fname && fname_name(fname))
759 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
760 hash = hinfo->hash;
761
762 if (root->info.unused_flags & 1) {
763 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
764 root->info.unused_flags);
765 goto fail;
766 }
767
768 indirect = root->info.indirect_levels;
769 if (indirect > 1) {
770 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
771 root->info.indirect_levels);
772 goto fail;
773 }
774
775 entries = (struct dx_entry *)(((char *)&root->info) +
776 root->info.info_length);
777
778 if (dx_get_limit(entries) != dx_root_limit(dir,
779 root->info.info_length)) {
780 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
781 dx_get_limit(entries),
782 dx_root_limit(dir, root->info.info_length));
783 goto fail;
784 }
785
786 dxtrace(printk("Look up %x", hash));
787 while (1) {
788 count = dx_get_count(entries);
789 if (!count || count > dx_get_limit(entries)) {
790 ext4_warning_inode(dir,
791 "dx entry: count %u beyond limit %u",
792 count, dx_get_limit(entries));
793 goto fail;
794 }
795
796 p = entries + 1;
797 q = entries + count - 1;
798 while (p <= q) {
799 m = p + (q - p) / 2;
800 dxtrace(printk("."));
801 if (dx_get_hash(m) > hash)
802 q = m - 1;
803 else
804 p = m + 1;
805 }
806
807 if (0) { // linear search cross check
808 unsigned n = count - 1;
809 at = entries;
810 while (n--)
811 {
812 dxtrace(printk(","));
813 if (dx_get_hash(++at) > hash)
814 {
815 at--;
816 break;
817 }
818 }
819 assert (at == p - 1);
820 }
821
822 at = p - 1;
823 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
824 dx_get_block(at)));
825 frame->entries = entries;
826 frame->at = at;
827 if (!indirect--)
828 return frame;
829 frame++;
830 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
831 if (IS_ERR(frame->bh)) {
832 ret_err = (struct dx_frame *) frame->bh;
833 frame->bh = NULL;
834 goto fail;
835 }
836 entries = ((struct dx_node *) frame->bh->b_data)->entries;
837
838 if (dx_get_limit(entries) != dx_node_limit(dir)) {
839 ext4_warning_inode(dir,
840 "dx entry: limit %u != node limit %u",
841 dx_get_limit(entries), dx_node_limit(dir));
842 goto fail;
843 }
844 }
845 fail:
846 while (frame >= frame_in) {
847 brelse(frame->bh);
848 frame--;
849 }
850
851 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
852 ext4_warning_inode(dir,
853 "Corrupt directory, running e2fsck is recommended");
854 return ret_err;
855 }
856
857 static void dx_release(struct dx_frame *frames)
858 {
859 if (frames[0].bh == NULL)
860 return;
861
862 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
863 brelse(frames[1].bh);
864 brelse(frames[0].bh);
865 }
866
867 /*
868 * This function increments the frame pointer to search the next leaf
869 * block, and reads in the necessary intervening nodes if the search
870 * should be necessary. Whether or not the search is necessary is
871 * controlled by the hash parameter. If the hash value is even, then
872 * the search is only continued if the next block starts with that
873 * hash value. This is used if we are searching for a specific file.
874 *
875 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
876 *
877 * This function returns 1 if the caller should continue to search,
878 * or 0 if it should not. If there is an error reading one of the
879 * index blocks, it will a negative error code.
880 *
881 * If start_hash is non-null, it will be filled in with the starting
882 * hash of the next page.
883 */
884 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
885 struct dx_frame *frame,
886 struct dx_frame *frames,
887 __u32 *start_hash)
888 {
889 struct dx_frame *p;
890 struct buffer_head *bh;
891 int num_frames = 0;
892 __u32 bhash;
893
894 p = frame;
895 /*
896 * Find the next leaf page by incrementing the frame pointer.
897 * If we run out of entries in the interior node, loop around and
898 * increment pointer in the parent node. When we break out of
899 * this loop, num_frames indicates the number of interior
900 * nodes need to be read.
901 */
902 while (1) {
903 if (++(p->at) < p->entries + dx_get_count(p->entries))
904 break;
905 if (p == frames)
906 return 0;
907 num_frames++;
908 p--;
909 }
910
911 /*
912 * If the hash is 1, then continue only if the next page has a
913 * continuation hash of any value. This is used for readdir
914 * handling. Otherwise, check to see if the hash matches the
915 * desired contiuation hash. If it doesn't, return since
916 * there's no point to read in the successive index pages.
917 */
918 bhash = dx_get_hash(p->at);
919 if (start_hash)
920 *start_hash = bhash;
921 if ((hash & 1) == 0) {
922 if ((bhash & ~1) != hash)
923 return 0;
924 }
925 /*
926 * If the hash is HASH_NB_ALWAYS, we always go to the next
927 * block so no check is necessary
928 */
929 while (num_frames--) {
930 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
931 if (IS_ERR(bh))
932 return PTR_ERR(bh);
933 p++;
934 brelse(p->bh);
935 p->bh = bh;
936 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
937 }
938 return 1;
939 }
940
941
942 /*
943 * This function fills a red-black tree with information from a
944 * directory block. It returns the number directory entries loaded
945 * into the tree. If there is an error it is returned in err.
946 */
947 static int htree_dirblock_to_tree(struct file *dir_file,
948 struct inode *dir, ext4_lblk_t block,
949 struct dx_hash_info *hinfo,
950 __u32 start_hash, __u32 start_minor_hash)
951 {
952 struct buffer_head *bh;
953 struct ext4_dir_entry_2 *de, *top;
954 int err = 0, count = 0;
955 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
956
957 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
958 (unsigned long)block));
959 bh = ext4_read_dirblock(dir, block, DIRENT);
960 if (IS_ERR(bh))
961 return PTR_ERR(bh);
962
963 de = (struct ext4_dir_entry_2 *) bh->b_data;
964 top = (struct ext4_dir_entry_2 *) ((char *) de +
965 dir->i_sb->s_blocksize -
966 EXT4_DIR_REC_LEN(0));
967 #ifdef CONFIG_EXT4_FS_ENCRYPTION
968 /* Check if the directory is encrypted */
969 if (ext4_encrypted_inode(dir)) {
970 err = ext4_get_encryption_info(dir);
971 if (err < 0) {
972 brelse(bh);
973 return err;
974 }
975 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
976 &fname_crypto_str);
977 if (err < 0) {
978 brelse(bh);
979 return err;
980 }
981 }
982 #endif
983 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
984 if (ext4_check_dir_entry(dir, NULL, de, bh,
985 bh->b_data, bh->b_size,
986 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
987 + ((char *)de - bh->b_data))) {
988 /* silently ignore the rest of the block */
989 break;
990 }
991 ext4fs_dirhash(de->name, de->name_len, hinfo);
992 if ((hinfo->hash < start_hash) ||
993 ((hinfo->hash == start_hash) &&
994 (hinfo->minor_hash < start_minor_hash)))
995 continue;
996 if (de->inode == 0)
997 continue;
998 if (!ext4_encrypted_inode(dir)) {
999 tmp_str.name = de->name;
1000 tmp_str.len = de->name_len;
1001 err = ext4_htree_store_dirent(dir_file,
1002 hinfo->hash, hinfo->minor_hash, de,
1003 &tmp_str);
1004 } else {
1005 int save_len = fname_crypto_str.len;
1006
1007 /* Directory is encrypted */
1008 err = ext4_fname_disk_to_usr(dir, hinfo, de,
1009 &fname_crypto_str);
1010 if (err < 0) {
1011 count = err;
1012 goto errout;
1013 }
1014 err = ext4_htree_store_dirent(dir_file,
1015 hinfo->hash, hinfo->minor_hash, de,
1016 &fname_crypto_str);
1017 fname_crypto_str.len = save_len;
1018 }
1019 if (err != 0) {
1020 count = err;
1021 goto errout;
1022 }
1023 count++;
1024 }
1025 errout:
1026 brelse(bh);
1027 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1028 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1029 #endif
1030 return count;
1031 }
1032
1033
1034 /*
1035 * This function fills a red-black tree with information from a
1036 * directory. We start scanning the directory in hash order, starting
1037 * at start_hash and start_minor_hash.
1038 *
1039 * This function returns the number of entries inserted into the tree,
1040 * or a negative error code.
1041 */
1042 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1043 __u32 start_minor_hash, __u32 *next_hash)
1044 {
1045 struct dx_hash_info hinfo;
1046 struct ext4_dir_entry_2 *de;
1047 struct dx_frame frames[2], *frame;
1048 struct inode *dir;
1049 ext4_lblk_t block;
1050 int count = 0;
1051 int ret, err;
1052 __u32 hashval;
1053 struct ext4_str tmp_str;
1054
1055 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1056 start_hash, start_minor_hash));
1057 dir = file_inode(dir_file);
1058 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1059 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1060 if (hinfo.hash_version <= DX_HASH_TEA)
1061 hinfo.hash_version +=
1062 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1063 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1064 if (ext4_has_inline_data(dir)) {
1065 int has_inline_data = 1;
1066 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1067 &hinfo, start_hash,
1068 start_minor_hash,
1069 &has_inline_data);
1070 if (has_inline_data) {
1071 *next_hash = ~0;
1072 return count;
1073 }
1074 }
1075 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1076 start_hash, start_minor_hash);
1077 *next_hash = ~0;
1078 return count;
1079 }
1080 hinfo.hash = start_hash;
1081 hinfo.minor_hash = 0;
1082 frame = dx_probe(NULL, dir, &hinfo, frames);
1083 if (IS_ERR(frame))
1084 return PTR_ERR(frame);
1085
1086 /* Add '.' and '..' from the htree header */
1087 if (!start_hash && !start_minor_hash) {
1088 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1089 tmp_str.name = de->name;
1090 tmp_str.len = de->name_len;
1091 err = ext4_htree_store_dirent(dir_file, 0, 0,
1092 de, &tmp_str);
1093 if (err != 0)
1094 goto errout;
1095 count++;
1096 }
1097 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1098 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1099 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1100 tmp_str.name = de->name;
1101 tmp_str.len = de->name_len;
1102 err = ext4_htree_store_dirent(dir_file, 2, 0,
1103 de, &tmp_str);
1104 if (err != 0)
1105 goto errout;
1106 count++;
1107 }
1108
1109 while (1) {
1110 block = dx_get_block(frame->at);
1111 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1112 start_hash, start_minor_hash);
1113 if (ret < 0) {
1114 err = ret;
1115 goto errout;
1116 }
1117 count += ret;
1118 hashval = ~0;
1119 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1120 frame, frames, &hashval);
1121 *next_hash = hashval;
1122 if (ret < 0) {
1123 err = ret;
1124 goto errout;
1125 }
1126 /*
1127 * Stop if: (a) there are no more entries, or
1128 * (b) we have inserted at least one entry and the
1129 * next hash value is not a continuation
1130 */
1131 if ((ret == 0) ||
1132 (count && ((hashval & 1) == 0)))
1133 break;
1134 }
1135 dx_release(frames);
1136 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1137 "next hash: %x\n", count, *next_hash));
1138 return count;
1139 errout:
1140 dx_release(frames);
1141 return (err);
1142 }
1143
1144 static inline int search_dirblock(struct buffer_head *bh,
1145 struct inode *dir,
1146 struct ext4_filename *fname,
1147 const struct qstr *d_name,
1148 unsigned int offset,
1149 struct ext4_dir_entry_2 **res_dir)
1150 {
1151 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1152 fname, d_name, offset, res_dir);
1153 }
1154
1155 /*
1156 * Directory block splitting, compacting
1157 */
1158
1159 /*
1160 * Create map of hash values, offsets, and sizes, stored at end of block.
1161 * Returns number of entries mapped.
1162 */
1163 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1164 unsigned blocksize, struct dx_hash_info *hinfo,
1165 struct dx_map_entry *map_tail)
1166 {
1167 int count = 0;
1168 char *base = (char *) de;
1169 struct dx_hash_info h = *hinfo;
1170
1171 while ((char *) de < base + blocksize) {
1172 if (de->name_len && de->inode) {
1173 ext4fs_dirhash(de->name, de->name_len, &h);
1174 map_tail--;
1175 map_tail->hash = h.hash;
1176 map_tail->offs = ((char *) de - base)>>2;
1177 map_tail->size = le16_to_cpu(de->rec_len);
1178 count++;
1179 cond_resched();
1180 }
1181 /* XXX: do we need to check rec_len == 0 case? -Chris */
1182 de = ext4_next_entry(de, blocksize);
1183 }
1184 return count;
1185 }
1186
1187 /* Sort map by hash value */
1188 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1189 {
1190 struct dx_map_entry *p, *q, *top = map + count - 1;
1191 int more;
1192 /* Combsort until bubble sort doesn't suck */
1193 while (count > 2) {
1194 count = count*10/13;
1195 if (count - 9 < 2) /* 9, 10 -> 11 */
1196 count = 11;
1197 for (p = top, q = p - count; q >= map; p--, q--)
1198 if (p->hash < q->hash)
1199 swap(*p, *q);
1200 }
1201 /* Garden variety bubble sort */
1202 do {
1203 more = 0;
1204 q = top;
1205 while (q-- > map) {
1206 if (q[1].hash >= q[0].hash)
1207 continue;
1208 swap(*(q+1), *q);
1209 more = 1;
1210 }
1211 } while(more);
1212 }
1213
1214 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1215 {
1216 struct dx_entry *entries = frame->entries;
1217 struct dx_entry *old = frame->at, *new = old + 1;
1218 int count = dx_get_count(entries);
1219
1220 assert(count < dx_get_limit(entries));
1221 assert(old < entries + count);
1222 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1223 dx_set_hash(new, hash);
1224 dx_set_block(new, block);
1225 dx_set_count(entries, count + 1);
1226 }
1227
1228 /*
1229 * Test whether a directory entry matches the filename being searched for.
1230 *
1231 * Return: %true if the directory entry matches, otherwise %false.
1232 */
1233 static inline bool ext4_match(const struct ext4_filename *fname,
1234 const 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 false;
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 <= 32)
1247 return 0;
1248 ret = memcmp(de->name + ((de->name_len - 17) & ~15),
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
1273 de = (struct ext4_dir_entry_2 *)search_buf;
1274 dlimit = search_buf + buf_size;
1275 while ((char *) de < dlimit) {
1276 /* this code is executed quadratically often */
1277 /* do minimal checking `by hand' */
1278 if ((char *) de + de->name_len <= dlimit &&
1279 ext4_match(fname, de)) {
1280 /* found a match - just to be sure, do
1281 * a full check */
1282 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1283 buf_size, offset))
1284 return -1;
1285 *res_dir = de;
1286 return 1;
1287 }
1288 /* prevent looping on a bad block */
1289 de_len = ext4_rec_len_from_disk(de->rec_len,
1290 dir->i_sb->s_blocksize);
1291 if (de_len <= 0)
1292 return -1;
1293 offset += de_len;
1294 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1295 }
1296 return 0;
1297 }
1298
1299 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1300 struct ext4_dir_entry *de)
1301 {
1302 struct super_block *sb = dir->i_sb;
1303
1304 if (!is_dx(dir))
1305 return 0;
1306 if (block == 0)
1307 return 1;
1308 if (de->inode == 0 &&
1309 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1310 sb->s_blocksize)
1311 return 1;
1312 return 0;
1313 }
1314
1315 /*
1316 * ext4_find_entry()
1317 *
1318 * finds an entry in the specified directory with the wanted name. It
1319 * returns the cache buffer in which the entry was found, and the entry
1320 * itself (as a parameter - res_dir). It does NOT read the inode of the
1321 * entry - you'll have to do that yourself if you want to.
1322 *
1323 * The returned buffer_head has ->b_count elevated. The caller is expected
1324 * to brelse() it when appropriate.
1325 */
1326 static struct buffer_head * ext4_find_entry (struct inode *dir,
1327 const struct qstr *d_name,
1328 struct ext4_dir_entry_2 **res_dir,
1329 int *inlined)
1330 {
1331 struct super_block *sb;
1332 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1333 struct buffer_head *bh, *ret = NULL;
1334 ext4_lblk_t start, block, b;
1335 const u8 *name = d_name->name;
1336 int ra_max = 0; /* Number of bh's in the readahead
1337 buffer, bh_use[] */
1338 int ra_ptr = 0; /* Current index into readahead
1339 buffer */
1340 int num = 0;
1341 ext4_lblk_t nblocks;
1342 int i, namelen, retval;
1343 struct ext4_filename fname;
1344
1345 *res_dir = NULL;
1346 sb = dir->i_sb;
1347 namelen = d_name->len;
1348 if (namelen > EXT4_NAME_LEN)
1349 return NULL;
1350
1351 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1352 if (retval)
1353 return ERR_PTR(retval);
1354
1355 if (ext4_has_inline_data(dir)) {
1356 int has_inline_data = 1;
1357 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1358 &has_inline_data);
1359 if (has_inline_data) {
1360 if (inlined)
1361 *inlined = 1;
1362 goto cleanup_and_exit;
1363 }
1364 }
1365
1366 if ((namelen <= 2) && (name[0] == '.') &&
1367 (name[1] == '.' || name[1] == '\0')) {
1368 /*
1369 * "." or ".." will only be in the first block
1370 * NFS may look up ".."; "." should be handled by the VFS
1371 */
1372 block = start = 0;
1373 nblocks = 1;
1374 goto restart;
1375 }
1376 if (is_dx(dir)) {
1377 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1378 /*
1379 * On success, or if the error was file not found,
1380 * return. Otherwise, fall back to doing a search the
1381 * old fashioned way.
1382 */
1383 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1384 goto cleanup_and_exit;
1385 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1386 "falling back\n"));
1387 ret = NULL;
1388 }
1389 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1390 if (!nblocks) {
1391 ret = NULL;
1392 goto cleanup_and_exit;
1393 }
1394 start = EXT4_I(dir)->i_dir_start_lookup;
1395 if (start >= nblocks)
1396 start = 0;
1397 block = start;
1398 restart:
1399 do {
1400 /*
1401 * We deal with the read-ahead logic here.
1402 */
1403 cond_resched();
1404 if (ra_ptr >= ra_max) {
1405 /* Refill the readahead buffer */
1406 ra_ptr = 0;
1407 b = block;
1408 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1409 /*
1410 * Terminate if we reach the end of the
1411 * directory and must wrap, or if our
1412 * search has finished at this block.
1413 */
1414 if (b >= nblocks || (num && block == start)) {
1415 bh_use[ra_max] = NULL;
1416 break;
1417 }
1418 num++;
1419 bh = ext4_getblk(NULL, dir, b++, 0);
1420 if (IS_ERR(bh)) {
1421 if (ra_max == 0) {
1422 ret = bh;
1423 goto cleanup_and_exit;
1424 }
1425 break;
1426 }
1427 bh_use[ra_max] = bh;
1428 if (bh)
1429 ll_rw_block(READ | REQ_META | REQ_PRIO,
1430 1, &bh);
1431 }
1432 }
1433 if ((bh = bh_use[ra_ptr++]) == NULL)
1434 goto next;
1435 wait_on_buffer(bh);
1436 if (!buffer_uptodate(bh)) {
1437 /* read error, skip block & hope for the best */
1438 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1439 (unsigned long) block);
1440 brelse(bh);
1441 goto next;
1442 }
1443 if (!buffer_verified(bh) &&
1444 !is_dx_internal_node(dir, block,
1445 (struct ext4_dir_entry *)bh->b_data) &&
1446 !ext4_dirent_csum_verify(dir,
1447 (struct ext4_dir_entry *)bh->b_data)) {
1448 EXT4_ERROR_INODE(dir, "checksumming directory "
1449 "block %lu", (unsigned long)block);
1450 brelse(bh);
1451 goto next;
1452 }
1453 set_buffer_verified(bh);
1454 i = search_dirblock(bh, dir, &fname, d_name,
1455 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1456 if (i == 1) {
1457 EXT4_I(dir)->i_dir_start_lookup = block;
1458 ret = bh;
1459 goto cleanup_and_exit;
1460 } else {
1461 brelse(bh);
1462 if (i < 0)
1463 goto cleanup_and_exit;
1464 }
1465 next:
1466 if (++block >= nblocks)
1467 block = 0;
1468 } while (block != start);
1469
1470 /*
1471 * If the directory has grown while we were searching, then
1472 * search the last part of the directory before giving up.
1473 */
1474 block = nblocks;
1475 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1476 if (block < nblocks) {
1477 start = 0;
1478 goto restart;
1479 }
1480
1481 cleanup_and_exit:
1482 /* Clean up the read-ahead blocks */
1483 for (; ra_ptr < ra_max; ra_ptr++)
1484 brelse(bh_use[ra_ptr]);
1485 ext4_fname_free_filename(&fname);
1486 return ret;
1487 }
1488
1489 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1490 struct ext4_filename *fname,
1491 struct ext4_dir_entry_2 **res_dir)
1492 {
1493 struct super_block * sb = dir->i_sb;
1494 struct dx_frame frames[2], *frame;
1495 const struct qstr *d_name = fname->usr_fname;
1496 struct buffer_head *bh;
1497 ext4_lblk_t block;
1498 int retval;
1499
1500 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1501 *res_dir = NULL;
1502 #endif
1503 frame = dx_probe(fname, dir, NULL, frames);
1504 if (IS_ERR(frame))
1505 return (struct buffer_head *) frame;
1506 do {
1507 block = dx_get_block(frame->at);
1508 bh = ext4_read_dirblock(dir, block, DIRENT);
1509 if (IS_ERR(bh))
1510 goto errout;
1511
1512 retval = search_dirblock(bh, dir, fname, d_name,
1513 block << EXT4_BLOCK_SIZE_BITS(sb),
1514 res_dir);
1515 if (retval == 1)
1516 goto success;
1517 brelse(bh);
1518 if (retval == -1) {
1519 bh = ERR_PTR(ERR_BAD_DX_DIR);
1520 goto errout;
1521 }
1522
1523 /* Check to see if we should continue to search */
1524 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1525 frames, NULL);
1526 if (retval < 0) {
1527 ext4_warning_inode(dir,
1528 "error %d reading directory index block",
1529 retval);
1530 bh = ERR_PTR(retval);
1531 goto errout;
1532 }
1533 } while (retval == 1);
1534
1535 bh = NULL;
1536 errout:
1537 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1538 success:
1539 dx_release(frames);
1540 return bh;
1541 }
1542
1543 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1544 {
1545 struct inode *inode;
1546 struct ext4_dir_entry_2 *de;
1547 struct buffer_head *bh;
1548
1549 if (ext4_encrypted_inode(dir)) {
1550 int res = ext4_get_encryption_info(dir);
1551
1552 /*
1553 * This should be a properly defined flag for
1554 * dentry->d_flags when we uplift this to the VFS.
1555 * d_fsdata is set to (void *) 1 if if the dentry is
1556 * created while the directory was encrypted and we
1557 * don't have access to the key.
1558 */
1559 dentry->d_fsdata = NULL;
1560 if (ext4_encryption_info(dir))
1561 dentry->d_fsdata = (void *) 1;
1562 d_set_d_op(dentry, &ext4_encrypted_d_ops);
1563 if (res && res != -ENOKEY)
1564 return ERR_PTR(res);
1565 }
1566
1567 if (dentry->d_name.len > EXT4_NAME_LEN)
1568 return ERR_PTR(-ENAMETOOLONG);
1569
1570 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1571 if (IS_ERR(bh))
1572 return (struct dentry *) bh;
1573 inode = NULL;
1574 if (bh) {
1575 __u32 ino = le32_to_cpu(de->inode);
1576 brelse(bh);
1577 if (!ext4_valid_inum(dir->i_sb, ino)) {
1578 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1579 return ERR_PTR(-EFSCORRUPTED);
1580 }
1581 if (unlikely(ino == dir->i_ino)) {
1582 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1583 dentry);
1584 return ERR_PTR(-EFSCORRUPTED);
1585 }
1586 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1587 if (inode == ERR_PTR(-ESTALE)) {
1588 EXT4_ERROR_INODE(dir,
1589 "deleted inode referenced: %u",
1590 ino);
1591 return ERR_PTR(-EFSCORRUPTED);
1592 }
1593 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1594 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1595 S_ISLNK(inode->i_mode)) &&
1596 !ext4_is_child_context_consistent_with_parent(dir,
1597 inode)) {
1598 iput(inode);
1599 ext4_warning(inode->i_sb,
1600 "Inconsistent encryption contexts: %lu/%lu\n",
1601 (unsigned long) dir->i_ino,
1602 (unsigned long) inode->i_ino);
1603 return ERR_PTR(-EPERM);
1604 }
1605 }
1606 return d_splice_alias(inode, dentry);
1607 }
1608
1609
1610 struct dentry *ext4_get_parent(struct dentry *child)
1611 {
1612 __u32 ino;
1613 static const struct qstr dotdot = QSTR_INIT("..", 2);
1614 struct ext4_dir_entry_2 * de;
1615 struct buffer_head *bh;
1616
1617 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1618 if (IS_ERR(bh))
1619 return (struct dentry *) bh;
1620 if (!bh)
1621 return ERR_PTR(-ENOENT);
1622 ino = le32_to_cpu(de->inode);
1623 brelse(bh);
1624
1625 if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1626 EXT4_ERROR_INODE(d_inode(child),
1627 "bad parent inode number: %u", ino);
1628 return ERR_PTR(-EFSCORRUPTED);
1629 }
1630
1631 return d_obtain_alias(ext4_iget(d_inode(child)->i_sb, ino, EXT4_IGET_NORMAL));
1632 }
1633
1634 /*
1635 * Move count entries from end of map between two memory locations.
1636 * Returns pointer to last entry moved.
1637 */
1638 static struct ext4_dir_entry_2 *
1639 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1640 unsigned blocksize)
1641 {
1642 unsigned rec_len = 0;
1643
1644 while (count--) {
1645 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1646 (from + (map->offs<<2));
1647 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1648 memcpy (to, de, rec_len);
1649 ((struct ext4_dir_entry_2 *) to)->rec_len =
1650 ext4_rec_len_to_disk(rec_len, blocksize);
1651 de->inode = 0;
1652 map++;
1653 to += rec_len;
1654 }
1655 return (struct ext4_dir_entry_2 *) (to - rec_len);
1656 }
1657
1658 /*
1659 * Compact each dir entry in the range to the minimal rec_len.
1660 * Returns pointer to last entry in range.
1661 */
1662 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1663 {
1664 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1665 unsigned rec_len = 0;
1666
1667 prev = to = de;
1668 while ((char*)de < base + blocksize) {
1669 next = ext4_next_entry(de, blocksize);
1670 if (de->inode && de->name_len) {
1671 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1672 if (de > to)
1673 memmove(to, de, rec_len);
1674 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1675 prev = to;
1676 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1677 }
1678 de = next;
1679 }
1680 return prev;
1681 }
1682
1683 /*
1684 * Split a full leaf block to make room for a new dir entry.
1685 * Allocate a new block, and move entries so that they are approx. equally full.
1686 * Returns pointer to de in block into which the new entry will be inserted.
1687 */
1688 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1689 struct buffer_head **bh,struct dx_frame *frame,
1690 struct dx_hash_info *hinfo)
1691 {
1692 unsigned blocksize = dir->i_sb->s_blocksize;
1693 unsigned count, continued;
1694 struct buffer_head *bh2;
1695 ext4_lblk_t newblock;
1696 u32 hash2;
1697 struct dx_map_entry *map;
1698 char *data1 = (*bh)->b_data, *data2;
1699 unsigned split, move, size;
1700 struct ext4_dir_entry_2 *de = NULL, *de2;
1701 struct ext4_dir_entry_tail *t;
1702 int csum_size = 0;
1703 int err = 0, i;
1704
1705 if (ext4_has_metadata_csum(dir->i_sb))
1706 csum_size = sizeof(struct ext4_dir_entry_tail);
1707
1708 bh2 = ext4_append(handle, dir, &newblock);
1709 if (IS_ERR(bh2)) {
1710 brelse(*bh);
1711 *bh = NULL;
1712 return (struct ext4_dir_entry_2 *) bh2;
1713 }
1714
1715 BUFFER_TRACE(*bh, "get_write_access");
1716 err = ext4_journal_get_write_access(handle, *bh);
1717 if (err)
1718 goto journal_error;
1719
1720 BUFFER_TRACE(frame->bh, "get_write_access");
1721 err = ext4_journal_get_write_access(handle, frame->bh);
1722 if (err)
1723 goto journal_error;
1724
1725 data2 = bh2->b_data;
1726
1727 /* create map in the end of data2 block */
1728 map = (struct dx_map_entry *) (data2 + blocksize);
1729 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1730 blocksize, hinfo, map);
1731 map -= count;
1732 dx_sort_map(map, count);
1733 /* Ensure that neither split block is over half full */
1734 size = 0;
1735 move = 0;
1736 for (i = count-1; i >= 0; i--) {
1737 /* is more than half of this entry in 2nd half of the block? */
1738 if (size + map[i].size/2 > blocksize/2)
1739 break;
1740 size += map[i].size;
1741 move++;
1742 }
1743 /*
1744 * map index at which we will split
1745 *
1746 * If the sum of active entries didn't exceed half the block size, just
1747 * split it in half by count; each resulting block will have at least
1748 * half the space free.
1749 */
1750 if (i > 0)
1751 split = count - move;
1752 else
1753 split = count/2;
1754
1755 hash2 = map[split].hash;
1756 continued = hash2 == map[split - 1].hash;
1757 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1758 (unsigned long)dx_get_block(frame->at),
1759 hash2, split, count-split));
1760
1761 /* Fancy dance to stay within two buffers */
1762 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1763 blocksize);
1764 de = dx_pack_dirents(data1, blocksize);
1765 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1766 (char *) de,
1767 blocksize);
1768 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1769 (char *) de2,
1770 blocksize);
1771 if (csum_size) {
1772 t = EXT4_DIRENT_TAIL(data2, blocksize);
1773 initialize_dirent_tail(t, blocksize);
1774
1775 t = EXT4_DIRENT_TAIL(data1, blocksize);
1776 initialize_dirent_tail(t, blocksize);
1777 }
1778
1779 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1780 blocksize, 1));
1781 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1782 blocksize, 1));
1783
1784 /* Which block gets the new entry? */
1785 if (hinfo->hash >= hash2) {
1786 swap(*bh, bh2);
1787 de = de2;
1788 }
1789 dx_insert_block(frame, hash2 + continued, newblock);
1790 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1791 if (err)
1792 goto journal_error;
1793 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1794 if (err)
1795 goto journal_error;
1796 brelse(bh2);
1797 dxtrace(dx_show_index("frame", frame->entries));
1798 return de;
1799
1800 journal_error:
1801 brelse(*bh);
1802 brelse(bh2);
1803 *bh = NULL;
1804 ext4_std_error(dir->i_sb, err);
1805 return ERR_PTR(err);
1806 }
1807
1808 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1809 struct buffer_head *bh,
1810 void *buf, int buf_size,
1811 struct ext4_filename *fname,
1812 struct ext4_dir_entry_2 **dest_de)
1813 {
1814 struct ext4_dir_entry_2 *de;
1815 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1816 int nlen, rlen;
1817 unsigned int offset = 0;
1818 char *top;
1819
1820 de = (struct ext4_dir_entry_2 *)buf;
1821 top = buf + buf_size - reclen;
1822 while ((char *) de <= top) {
1823 if (ext4_check_dir_entry(dir, NULL, de, bh,
1824 buf, buf_size, offset))
1825 return -EFSCORRUPTED;
1826 if (ext4_match(fname, de))
1827 return -EEXIST;
1828 nlen = EXT4_DIR_REC_LEN(de->name_len);
1829 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1830 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1831 break;
1832 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1833 offset += rlen;
1834 }
1835 if ((char *) de > top)
1836 return -ENOSPC;
1837
1838 *dest_de = de;
1839 return 0;
1840 }
1841
1842 int ext4_insert_dentry(struct inode *dir,
1843 struct inode *inode,
1844 struct ext4_dir_entry_2 *de,
1845 int buf_size,
1846 struct ext4_filename *fname)
1847 {
1848
1849 int nlen, rlen;
1850
1851 nlen = EXT4_DIR_REC_LEN(de->name_len);
1852 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1853 if (de->inode) {
1854 struct ext4_dir_entry_2 *de1 =
1855 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1856 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1857 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1858 de = de1;
1859 }
1860 de->file_type = EXT4_FT_UNKNOWN;
1861 de->inode = cpu_to_le32(inode->i_ino);
1862 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1863 de->name_len = fname_len(fname);
1864 memcpy(de->name, fname_name(fname), fname_len(fname));
1865 return 0;
1866 }
1867
1868 /*
1869 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1870 * it points to a directory entry which is guaranteed to be large
1871 * enough for new directory entry. If de is NULL, then
1872 * add_dirent_to_buf will attempt search the directory block for
1873 * space. It will return -ENOSPC if no space is available, and -EIO
1874 * and -EEXIST if directory entry already exists.
1875 */
1876 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1877 struct inode *dir,
1878 struct inode *inode, struct ext4_dir_entry_2 *de,
1879 struct buffer_head *bh)
1880 {
1881 unsigned int blocksize = dir->i_sb->s_blocksize;
1882 int csum_size = 0;
1883 int err;
1884
1885 if (ext4_has_metadata_csum(inode->i_sb))
1886 csum_size = sizeof(struct ext4_dir_entry_tail);
1887
1888 if (!de) {
1889 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1890 blocksize - csum_size, fname, &de);
1891 if (err)
1892 return err;
1893 }
1894 BUFFER_TRACE(bh, "get_write_access");
1895 err = ext4_journal_get_write_access(handle, bh);
1896 if (err) {
1897 ext4_std_error(dir->i_sb, err);
1898 return err;
1899 }
1900
1901 /* By now the buffer is marked for journaling. Due to crypto operations,
1902 * the following function call may fail */
1903 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1904 if (err < 0)
1905 return err;
1906
1907 /*
1908 * XXX shouldn't update any times until successful
1909 * completion of syscall, but too many callers depend
1910 * on this.
1911 *
1912 * XXX similarly, too many callers depend on
1913 * ext4_new_inode() setting the times, but error
1914 * recovery deletes the inode, so the worst that can
1915 * happen is that the times are slightly out of date
1916 * and/or different from the directory change time.
1917 */
1918 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1919 ext4_update_dx_flag(dir);
1920 dir->i_version++;
1921 ext4_mark_inode_dirty(handle, dir);
1922 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1923 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1924 if (err)
1925 ext4_std_error(dir->i_sb, err);
1926 return 0;
1927 }
1928
1929 /*
1930 * This converts a one block unindexed directory to a 3 block indexed
1931 * directory, and adds the dentry to the indexed directory.
1932 */
1933 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1934 struct dentry *dentry,
1935 struct inode *inode, struct buffer_head *bh)
1936 {
1937 struct inode *dir = d_inode(dentry->d_parent);
1938 struct buffer_head *bh2;
1939 struct dx_root *root;
1940 struct dx_frame frames[2], *frame;
1941 struct dx_entry *entries;
1942 struct ext4_dir_entry_2 *de, *de2;
1943 struct ext4_dir_entry_tail *t;
1944 char *data1, *top;
1945 unsigned len;
1946 int retval;
1947 unsigned blocksize;
1948 ext4_lblk_t block;
1949 struct fake_dirent *fde;
1950 int csum_size = 0;
1951
1952 if (ext4_has_metadata_csum(inode->i_sb))
1953 csum_size = sizeof(struct ext4_dir_entry_tail);
1954
1955 blocksize = dir->i_sb->s_blocksize;
1956 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1957 BUFFER_TRACE(bh, "get_write_access");
1958 retval = ext4_journal_get_write_access(handle, bh);
1959 if (retval) {
1960 ext4_std_error(dir->i_sb, retval);
1961 brelse(bh);
1962 return retval;
1963 }
1964 root = (struct dx_root *) bh->b_data;
1965
1966 /* The 0th block becomes the root, move the dirents out */
1967 fde = &root->dotdot;
1968 de = (struct ext4_dir_entry_2 *)((char *)fde +
1969 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1970 if ((char *) de >= (((char *) root) + blocksize)) {
1971 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1972 brelse(bh);
1973 return -EFSCORRUPTED;
1974 }
1975 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1976
1977 /* Allocate new block for the 0th block's dirents */
1978 bh2 = ext4_append(handle, dir, &block);
1979 if (IS_ERR(bh2)) {
1980 brelse(bh);
1981 return PTR_ERR(bh2);
1982 }
1983 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1984 data1 = bh2->b_data;
1985
1986 memcpy (data1, de, len);
1987 de = (struct ext4_dir_entry_2 *) data1;
1988 top = data1 + len;
1989 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1990 de = de2;
1991 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1992 (char *) de,
1993 blocksize);
1994
1995 if (csum_size) {
1996 t = EXT4_DIRENT_TAIL(data1, blocksize);
1997 initialize_dirent_tail(t, blocksize);
1998 }
1999
2000 /* Initialize the root; the dot dirents already exist */
2001 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2002 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2003 blocksize);
2004 memset (&root->info, 0, sizeof(root->info));
2005 root->info.info_length = sizeof(root->info);
2006 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2007 entries = root->entries;
2008 dx_set_block(entries, 1);
2009 dx_set_count(entries, 1);
2010 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2011
2012 /* Initialize as for dx_probe */
2013 fname->hinfo.hash_version = root->info.hash_version;
2014 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2015 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2016 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2017 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2018
2019 memset(frames, 0, sizeof(frames));
2020 frame = frames;
2021 frame->entries = entries;
2022 frame->at = entries;
2023 frame->bh = bh;
2024
2025 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2026 if (retval)
2027 goto out_frames;
2028 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2029 if (retval)
2030 goto out_frames;
2031
2032 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2033 if (IS_ERR(de)) {
2034 retval = PTR_ERR(de);
2035 goto out_frames;
2036 }
2037
2038 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2039 out_frames:
2040 /*
2041 * Even if the block split failed, we have to properly write
2042 * out all the changes we did so far. Otherwise we can end up
2043 * with corrupted filesystem.
2044 */
2045 if (retval)
2046 ext4_mark_inode_dirty(handle, dir);
2047 dx_release(frames);
2048 brelse(bh2);
2049 return retval;
2050 }
2051
2052 /*
2053 * ext4_add_entry()
2054 *
2055 * adds a file entry to the specified directory, using the same
2056 * semantics as ext4_find_entry(). It returns NULL if it failed.
2057 *
2058 * NOTE!! The inode part of 'de' is left at 0 - which means you
2059 * may not sleep between calling this and putting something into
2060 * the entry, as someone else might have used it while you slept.
2061 */
2062 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2063 struct inode *inode)
2064 {
2065 struct inode *dir = d_inode(dentry->d_parent);
2066 struct buffer_head *bh = NULL;
2067 struct ext4_dir_entry_2 *de;
2068 struct ext4_dir_entry_tail *t;
2069 struct super_block *sb;
2070 struct ext4_filename fname;
2071 int retval;
2072 int dx_fallback=0;
2073 unsigned blocksize;
2074 ext4_lblk_t block, blocks;
2075 int csum_size = 0;
2076
2077 if (ext4_has_metadata_csum(inode->i_sb))
2078 csum_size = sizeof(struct ext4_dir_entry_tail);
2079
2080 sb = dir->i_sb;
2081 blocksize = sb->s_blocksize;
2082 if (!dentry->d_name.len)
2083 return -EINVAL;
2084
2085 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2086 if (retval)
2087 return retval;
2088
2089 if (ext4_has_inline_data(dir)) {
2090 retval = ext4_try_add_inline_entry(handle, &fname,
2091 dentry, inode);
2092 if (retval < 0)
2093 goto out;
2094 if (retval == 1) {
2095 retval = 0;
2096 goto out;
2097 }
2098 }
2099
2100 if (is_dx(dir)) {
2101 retval = ext4_dx_add_entry(handle, &fname, dentry, inode);
2102 if (!retval || (retval != ERR_BAD_DX_DIR))
2103 goto out;
2104 /* Can we just ignore htree data? */
2105 if (ext4_has_metadata_csum(sb)) {
2106 EXT4_ERROR_INODE(dir,
2107 "Directory has corrupted htree index.");
2108 retval = -EFSCORRUPTED;
2109 goto out;
2110 }
2111 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2112 dx_fallback++;
2113 ext4_mark_inode_dirty(handle, dir);
2114 }
2115 blocks = dir->i_size >> sb->s_blocksize_bits;
2116 for (block = 0; block < blocks; block++) {
2117 bh = ext4_read_dirblock(dir, block, DIRENT);
2118 if (IS_ERR(bh)) {
2119 retval = PTR_ERR(bh);
2120 bh = NULL;
2121 goto out;
2122 }
2123 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2124 NULL, bh);
2125 if (retval != -ENOSPC)
2126 goto out;
2127
2128 if (blocks == 1 && !dx_fallback &&
2129 ext4_has_feature_dir_index(sb)) {
2130 retval = make_indexed_dir(handle, &fname, dentry,
2131 inode, bh);
2132 bh = NULL; /* make_indexed_dir releases bh */
2133 goto out;
2134 }
2135 brelse(bh);
2136 }
2137 bh = ext4_append(handle, dir, &block);
2138 if (IS_ERR(bh)) {
2139 retval = PTR_ERR(bh);
2140 bh = NULL;
2141 goto out;
2142 }
2143 de = (struct ext4_dir_entry_2 *) bh->b_data;
2144 de->inode = 0;
2145 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2146
2147 if (csum_size) {
2148 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2149 initialize_dirent_tail(t, blocksize);
2150 }
2151
2152 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2153 out:
2154 ext4_fname_free_filename(&fname);
2155 brelse(bh);
2156 if (retval == 0)
2157 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2158 return retval;
2159 }
2160
2161 /*
2162 * Returns 0 for success, or a negative error value
2163 */
2164 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2165 struct dentry *dentry, struct inode *inode)
2166 {
2167 struct dx_frame frames[2], *frame;
2168 struct dx_entry *entries, *at;
2169 struct buffer_head *bh;
2170 struct inode *dir = d_inode(dentry->d_parent);
2171 struct super_block *sb = dir->i_sb;
2172 struct ext4_dir_entry_2 *de;
2173 int err;
2174
2175 frame = dx_probe(fname, dir, NULL, frames);
2176 if (IS_ERR(frame))
2177 return PTR_ERR(frame);
2178 entries = frame->entries;
2179 at = frame->at;
2180 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2181 if (IS_ERR(bh)) {
2182 err = PTR_ERR(bh);
2183 bh = NULL;
2184 goto cleanup;
2185 }
2186
2187 BUFFER_TRACE(bh, "get_write_access");
2188 err = ext4_journal_get_write_access(handle, bh);
2189 if (err)
2190 goto journal_error;
2191
2192 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2193 if (err != -ENOSPC)
2194 goto cleanup;
2195
2196 /* Block full, should compress but for now just split */
2197 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2198 dx_get_count(entries), dx_get_limit(entries)));
2199 /* Need to split index? */
2200 if (dx_get_count(entries) == dx_get_limit(entries)) {
2201 ext4_lblk_t newblock;
2202 unsigned icount = dx_get_count(entries);
2203 int levels = frame - frames;
2204 struct dx_entry *entries2;
2205 struct dx_node *node2;
2206 struct buffer_head *bh2;
2207
2208 if (levels && (dx_get_count(frames->entries) ==
2209 dx_get_limit(frames->entries))) {
2210 ext4_warning_inode(dir, "Directory index full!");
2211 err = -ENOSPC;
2212 goto cleanup;
2213 }
2214 bh2 = ext4_append(handle, dir, &newblock);
2215 if (IS_ERR(bh2)) {
2216 err = PTR_ERR(bh2);
2217 goto cleanup;
2218 }
2219 node2 = (struct dx_node *)(bh2->b_data);
2220 entries2 = node2->entries;
2221 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2222 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2223 sb->s_blocksize);
2224 BUFFER_TRACE(frame->bh, "get_write_access");
2225 err = ext4_journal_get_write_access(handle, frame->bh);
2226 if (err)
2227 goto journal_error;
2228 if (levels) {
2229 unsigned icount1 = icount/2, icount2 = icount - icount1;
2230 unsigned hash2 = dx_get_hash(entries + icount1);
2231 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2232 icount1, icount2));
2233
2234 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2235 err = ext4_journal_get_write_access(handle,
2236 frames[0].bh);
2237 if (err)
2238 goto journal_error;
2239
2240 memcpy((char *) entries2, (char *) (entries + icount1),
2241 icount2 * sizeof(struct dx_entry));
2242 dx_set_count(entries, icount1);
2243 dx_set_count(entries2, icount2);
2244 dx_set_limit(entries2, dx_node_limit(dir));
2245
2246 /* Which index block gets the new entry? */
2247 if (at - entries >= icount1) {
2248 frame->at = at = at - entries - icount1 + entries2;
2249 frame->entries = entries = entries2;
2250 swap(frame->bh, bh2);
2251 }
2252 dx_insert_block(frames + 0, hash2, newblock);
2253 dxtrace(dx_show_index("node", frames[1].entries));
2254 dxtrace(dx_show_index("node",
2255 ((struct dx_node *) bh2->b_data)->entries));
2256 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2257 if (err)
2258 goto journal_error;
2259 brelse (bh2);
2260 } else {
2261 dxtrace(printk(KERN_DEBUG
2262 "Creating second level index...\n"));
2263 memcpy((char *) entries2, (char *) entries,
2264 icount * sizeof(struct dx_entry));
2265 dx_set_limit(entries2, dx_node_limit(dir));
2266
2267 /* Set up root */
2268 dx_set_count(entries, 1);
2269 dx_set_block(entries + 0, newblock);
2270 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2271
2272 /* Add new access path frame */
2273 frame = frames + 1;
2274 frame->at = at = at - entries + entries2;
2275 frame->entries = entries = entries2;
2276 frame->bh = bh2;
2277 err = ext4_journal_get_write_access(handle,
2278 frame->bh);
2279 if (err)
2280 goto journal_error;
2281 }
2282 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2283 if (err) {
2284 ext4_std_error(inode->i_sb, err);
2285 goto cleanup;
2286 }
2287 }
2288 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2289 if (IS_ERR(de)) {
2290 err = PTR_ERR(de);
2291 goto cleanup;
2292 }
2293 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2294 goto cleanup;
2295
2296 journal_error:
2297 ext4_std_error(dir->i_sb, err);
2298 cleanup:
2299 brelse(bh);
2300 dx_release(frames);
2301 return err;
2302 }
2303
2304 /*
2305 * ext4_generic_delete_entry deletes a directory entry by merging it
2306 * with the previous entry
2307 */
2308 int ext4_generic_delete_entry(handle_t *handle,
2309 struct inode *dir,
2310 struct ext4_dir_entry_2 *de_del,
2311 struct buffer_head *bh,
2312 void *entry_buf,
2313 int buf_size,
2314 int csum_size)
2315 {
2316 struct ext4_dir_entry_2 *de, *pde;
2317 unsigned int blocksize = dir->i_sb->s_blocksize;
2318 int i;
2319
2320 i = 0;
2321 pde = NULL;
2322 de = (struct ext4_dir_entry_2 *)entry_buf;
2323 while (i < buf_size - csum_size) {
2324 if (ext4_check_dir_entry(dir, NULL, de, bh,
2325 entry_buf, buf_size, i))
2326 return -EFSCORRUPTED;
2327 if (de == de_del) {
2328 if (pde)
2329 pde->rec_len = ext4_rec_len_to_disk(
2330 ext4_rec_len_from_disk(pde->rec_len,
2331 blocksize) +
2332 ext4_rec_len_from_disk(de->rec_len,
2333 blocksize),
2334 blocksize);
2335 else
2336 de->inode = 0;
2337 dir->i_version++;
2338 return 0;
2339 }
2340 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2341 pde = de;
2342 de = ext4_next_entry(de, blocksize);
2343 }
2344 return -ENOENT;
2345 }
2346
2347 static int ext4_delete_entry(handle_t *handle,
2348 struct inode *dir,
2349 struct ext4_dir_entry_2 *de_del,
2350 struct buffer_head *bh)
2351 {
2352 int err, csum_size = 0;
2353
2354 if (ext4_has_inline_data(dir)) {
2355 int has_inline_data = 1;
2356 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2357 &has_inline_data);
2358 if (has_inline_data)
2359 return err;
2360 }
2361
2362 if (ext4_has_metadata_csum(dir->i_sb))
2363 csum_size = sizeof(struct ext4_dir_entry_tail);
2364
2365 BUFFER_TRACE(bh, "get_write_access");
2366 err = ext4_journal_get_write_access(handle, bh);
2367 if (unlikely(err))
2368 goto out;
2369
2370 err = ext4_generic_delete_entry(handle, dir, de_del,
2371 bh, bh->b_data,
2372 dir->i_sb->s_blocksize, csum_size);
2373 if (err)
2374 goto out;
2375
2376 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2377 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2378 if (unlikely(err))
2379 goto out;
2380
2381 return 0;
2382 out:
2383 if (err != -ENOENT)
2384 ext4_std_error(dir->i_sb, err);
2385 return err;
2386 }
2387
2388 /*
2389 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2390 * since this indicates that nlinks count was previously 1.
2391 */
2392 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2393 {
2394 inc_nlink(inode);
2395 if (is_dx(inode) && inode->i_nlink > 1) {
2396 /* limit is 16-bit i_links_count */
2397 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2398 set_nlink(inode, 1);
2399 ext4_set_feature_dir_nlink(inode->i_sb);
2400 }
2401 }
2402 }
2403
2404 /*
2405 * If a directory had nlink == 1, then we should let it be 1. This indicates
2406 * directory has >EXT4_LINK_MAX subdirs.
2407 */
2408 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2409 {
2410 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2411 drop_nlink(inode);
2412 }
2413
2414
2415 static int ext4_add_nondir(handle_t *handle,
2416 struct dentry *dentry, struct inode *inode)
2417 {
2418 int err = ext4_add_entry(handle, dentry, inode);
2419 if (!err) {
2420 ext4_mark_inode_dirty(handle, inode);
2421 d_instantiate_new(dentry, inode);
2422 return 0;
2423 }
2424 drop_nlink(inode);
2425 unlock_new_inode(inode);
2426 iput(inode);
2427 return err;
2428 }
2429
2430 /*
2431 * By the time this is called, we already have created
2432 * the directory cache entry for the new file, but it
2433 * is so far negative - it has no inode.
2434 *
2435 * If the create succeeds, we fill in the inode information
2436 * with d_instantiate().
2437 */
2438 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2439 bool excl)
2440 {
2441 handle_t *handle;
2442 struct inode *inode;
2443 int err, credits, retries = 0;
2444
2445 err = dquot_initialize(dir);
2446 if (err)
2447 return err;
2448
2449 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2450 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2451 retry:
2452 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2453 NULL, EXT4_HT_DIR, credits);
2454 handle = ext4_journal_current_handle();
2455 err = PTR_ERR(inode);
2456 if (!IS_ERR(inode)) {
2457 inode->i_op = &ext4_file_inode_operations;
2458 inode->i_fop = &ext4_file_operations;
2459 ext4_set_aops(inode);
2460 err = ext4_add_nondir(handle, dentry, inode);
2461 if (!err && IS_DIRSYNC(dir))
2462 ext4_handle_sync(handle);
2463 }
2464 if (handle)
2465 ext4_journal_stop(handle);
2466 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2467 goto retry;
2468 return err;
2469 }
2470
2471 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2472 umode_t mode, dev_t rdev)
2473 {
2474 handle_t *handle;
2475 struct inode *inode;
2476 int err, credits, retries = 0;
2477
2478 err = dquot_initialize(dir);
2479 if (err)
2480 return err;
2481
2482 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2483 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2484 retry:
2485 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2486 NULL, EXT4_HT_DIR, credits);
2487 handle = ext4_journal_current_handle();
2488 err = PTR_ERR(inode);
2489 if (!IS_ERR(inode)) {
2490 init_special_inode(inode, inode->i_mode, rdev);
2491 inode->i_op = &ext4_special_inode_operations;
2492 err = ext4_add_nondir(handle, dentry, inode);
2493 if (!err && IS_DIRSYNC(dir))
2494 ext4_handle_sync(handle);
2495 }
2496 if (handle)
2497 ext4_journal_stop(handle);
2498 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2499 goto retry;
2500 return err;
2501 }
2502
2503 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2504 {
2505 handle_t *handle;
2506 struct inode *inode;
2507 int err, retries = 0;
2508
2509 err = dquot_initialize(dir);
2510 if (err)
2511 return err;
2512
2513 retry:
2514 inode = ext4_new_inode_start_handle(dir, mode,
2515 NULL, 0, NULL,
2516 EXT4_HT_DIR,
2517 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2518 4 + EXT4_XATTR_TRANS_BLOCKS);
2519 handle = ext4_journal_current_handle();
2520 err = PTR_ERR(inode);
2521 if (!IS_ERR(inode)) {
2522 inode->i_op = &ext4_file_inode_operations;
2523 inode->i_fop = &ext4_file_operations;
2524 ext4_set_aops(inode);
2525 d_tmpfile(dentry, inode);
2526 err = ext4_orphan_add(handle, inode);
2527 if (err)
2528 goto err_unlock_inode;
2529 mark_inode_dirty(inode);
2530 unlock_new_inode(inode);
2531 }
2532 if (handle)
2533 ext4_journal_stop(handle);
2534 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2535 goto retry;
2536 return err;
2537 err_unlock_inode:
2538 ext4_journal_stop(handle);
2539 unlock_new_inode(inode);
2540 return err;
2541 }
2542
2543 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2544 struct ext4_dir_entry_2 *de,
2545 int blocksize, int csum_size,
2546 unsigned int parent_ino, int dotdot_real_len)
2547 {
2548 de->inode = cpu_to_le32(inode->i_ino);
2549 de->name_len = 1;
2550 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2551 blocksize);
2552 strcpy(de->name, ".");
2553 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2554
2555 de = ext4_next_entry(de, blocksize);
2556 de->inode = cpu_to_le32(parent_ino);
2557 de->name_len = 2;
2558 if (!dotdot_real_len)
2559 de->rec_len = ext4_rec_len_to_disk(blocksize -
2560 (csum_size + EXT4_DIR_REC_LEN(1)),
2561 blocksize);
2562 else
2563 de->rec_len = ext4_rec_len_to_disk(
2564 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2565 strcpy(de->name, "..");
2566 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2567
2568 return ext4_next_entry(de, blocksize);
2569 }
2570
2571 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2572 struct inode *inode)
2573 {
2574 struct buffer_head *dir_block = NULL;
2575 struct ext4_dir_entry_2 *de;
2576 struct ext4_dir_entry_tail *t;
2577 ext4_lblk_t block = 0;
2578 unsigned int blocksize = dir->i_sb->s_blocksize;
2579 int csum_size = 0;
2580 int err;
2581
2582 if (ext4_has_metadata_csum(dir->i_sb))
2583 csum_size = sizeof(struct ext4_dir_entry_tail);
2584
2585 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2586 err = ext4_try_create_inline_dir(handle, dir, inode);
2587 if (err < 0 && err != -ENOSPC)
2588 goto out;
2589 if (!err)
2590 goto out;
2591 }
2592
2593 inode->i_size = 0;
2594 dir_block = ext4_append(handle, inode, &block);
2595 if (IS_ERR(dir_block))
2596 return PTR_ERR(dir_block);
2597 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2598 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2599 set_nlink(inode, 2);
2600 if (csum_size) {
2601 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2602 initialize_dirent_tail(t, blocksize);
2603 }
2604
2605 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2606 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2607 if (err)
2608 goto out;
2609 set_buffer_verified(dir_block);
2610 out:
2611 brelse(dir_block);
2612 return err;
2613 }
2614
2615 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2616 {
2617 handle_t *handle;
2618 struct inode *inode;
2619 int err, credits, retries = 0;
2620
2621 if (EXT4_DIR_LINK_MAX(dir))
2622 return -EMLINK;
2623
2624 err = dquot_initialize(dir);
2625 if (err)
2626 return err;
2627
2628 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2629 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2630 retry:
2631 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2632 &dentry->d_name,
2633 0, NULL, EXT4_HT_DIR, credits);
2634 handle = ext4_journal_current_handle();
2635 err = PTR_ERR(inode);
2636 if (IS_ERR(inode))
2637 goto out_stop;
2638
2639 inode->i_op = &ext4_dir_inode_operations;
2640 inode->i_fop = &ext4_dir_operations;
2641 err = ext4_init_new_dir(handle, dir, inode);
2642 if (err)
2643 goto out_clear_inode;
2644 err = ext4_mark_inode_dirty(handle, inode);
2645 if (!err)
2646 err = ext4_add_entry(handle, dentry, inode);
2647 if (err) {
2648 out_clear_inode:
2649 clear_nlink(inode);
2650 unlock_new_inode(inode);
2651 ext4_mark_inode_dirty(handle, inode);
2652 iput(inode);
2653 goto out_stop;
2654 }
2655 ext4_inc_count(handle, dir);
2656 ext4_update_dx_flag(dir);
2657 err = ext4_mark_inode_dirty(handle, dir);
2658 if (err)
2659 goto out_clear_inode;
2660 d_instantiate_new(dentry, inode);
2661 if (IS_DIRSYNC(dir))
2662 ext4_handle_sync(handle);
2663
2664 out_stop:
2665 if (handle)
2666 ext4_journal_stop(handle);
2667 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2668 goto retry;
2669 return err;
2670 }
2671
2672 /*
2673 * routine to check that the specified directory is empty (for rmdir)
2674 */
2675 int ext4_empty_dir(struct inode *inode)
2676 {
2677 unsigned int offset;
2678 struct buffer_head *bh;
2679 struct ext4_dir_entry_2 *de, *de1;
2680 struct super_block *sb;
2681 int err = 0;
2682
2683 if (ext4_has_inline_data(inode)) {
2684 int has_inline_data = 1;
2685
2686 err = empty_inline_dir(inode, &has_inline_data);
2687 if (has_inline_data)
2688 return err;
2689 }
2690
2691 sb = inode->i_sb;
2692 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2693 EXT4_ERROR_INODE(inode, "invalid size");
2694 return 1;
2695 }
2696 bh = ext4_read_dirblock(inode, 0, EITHER);
2697 if (IS_ERR(bh))
2698 return 1;
2699
2700 de = (struct ext4_dir_entry_2 *) bh->b_data;
2701 de1 = ext4_next_entry(de, sb->s_blocksize);
2702 if (le32_to_cpu(de->inode) != inode->i_ino ||
2703 le32_to_cpu(de1->inode) == 0 ||
2704 strcmp(".", de->name) || strcmp("..", de1->name)) {
2705 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2706 brelse(bh);
2707 return 1;
2708 }
2709 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2710 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2711 de = ext4_next_entry(de1, sb->s_blocksize);
2712 while (offset < inode->i_size) {
2713 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2714 unsigned int lblock;
2715 err = 0;
2716 brelse(bh);
2717 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2718 bh = ext4_read_dirblock(inode, lblock, EITHER);
2719 if (IS_ERR(bh))
2720 return 1;
2721 de = (struct ext4_dir_entry_2 *) bh->b_data;
2722 }
2723 if (ext4_check_dir_entry(inode, NULL, de, bh,
2724 bh->b_data, bh->b_size, offset)) {
2725 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2726 sb->s_blocksize);
2727 offset = (offset | (sb->s_blocksize - 1)) + 1;
2728 continue;
2729 }
2730 if (le32_to_cpu(de->inode)) {
2731 brelse(bh);
2732 return 0;
2733 }
2734 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2735 de = ext4_next_entry(de, sb->s_blocksize);
2736 }
2737 brelse(bh);
2738 return 1;
2739 }
2740
2741 /*
2742 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2743 * such inodes, starting at the superblock, in case we crash before the
2744 * file is closed/deleted, or in case the inode truncate spans multiple
2745 * transactions and the last transaction is not recovered after a crash.
2746 *
2747 * At filesystem recovery time, we walk this list deleting unlinked
2748 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2749 *
2750 * Orphan list manipulation functions must be called under i_mutex unless
2751 * we are just creating the inode or deleting it.
2752 */
2753 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2754 {
2755 struct super_block *sb = inode->i_sb;
2756 struct ext4_sb_info *sbi = EXT4_SB(sb);
2757 struct ext4_iloc iloc;
2758 int err = 0, rc;
2759 bool dirty = false;
2760
2761 if (!sbi->s_journal || is_bad_inode(inode))
2762 return 0;
2763
2764 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2765 !mutex_is_locked(&inode->i_mutex));
2766 /*
2767 * Exit early if inode already is on orphan list. This is a big speedup
2768 * since we don't have to contend on the global s_orphan_lock.
2769 */
2770 if (!list_empty(&EXT4_I(inode)->i_orphan))
2771 return 0;
2772
2773 /*
2774 * Orphan handling is only valid for files with data blocks
2775 * being truncated, or files being unlinked. Note that we either
2776 * hold i_mutex, or the inode can not be referenced from outside,
2777 * so i_nlink should not be bumped due to race
2778 */
2779 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2780 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2781
2782 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2783 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2784 if (err)
2785 goto out;
2786
2787 err = ext4_reserve_inode_write(handle, inode, &iloc);
2788 if (err)
2789 goto out;
2790
2791 mutex_lock(&sbi->s_orphan_lock);
2792 /*
2793 * Due to previous errors inode may be already a part of on-disk
2794 * orphan list. If so skip on-disk list modification.
2795 */
2796 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2797 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2798 /* Insert this inode at the head of the on-disk orphan list */
2799 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2800 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2801 dirty = true;
2802 }
2803 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2804 mutex_unlock(&sbi->s_orphan_lock);
2805
2806 if (dirty) {
2807 err = ext4_handle_dirty_super(handle, sb);
2808 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2809 if (!err)
2810 err = rc;
2811 if (err) {
2812 /*
2813 * We have to remove inode from in-memory list if
2814 * addition to on disk orphan list failed. Stray orphan
2815 * list entries can cause panics at unmount time.
2816 */
2817 mutex_lock(&sbi->s_orphan_lock);
2818 list_del_init(&EXT4_I(inode)->i_orphan);
2819 mutex_unlock(&sbi->s_orphan_lock);
2820 }
2821 } else
2822 brelse(iloc.bh);
2823
2824 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2825 jbd_debug(4, "orphan inode %lu will point to %d\n",
2826 inode->i_ino, NEXT_ORPHAN(inode));
2827 out:
2828 ext4_std_error(sb, err);
2829 return err;
2830 }
2831
2832 /*
2833 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2834 * of such inodes stored on disk, because it is finally being cleaned up.
2835 */
2836 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2837 {
2838 struct list_head *prev;
2839 struct ext4_inode_info *ei = EXT4_I(inode);
2840 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2841 __u32 ino_next;
2842 struct ext4_iloc iloc;
2843 int err = 0;
2844
2845 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2846 return 0;
2847
2848 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2849 !mutex_is_locked(&inode->i_mutex));
2850 /* Do this quick check before taking global s_orphan_lock. */
2851 if (list_empty(&ei->i_orphan))
2852 return 0;
2853
2854 if (handle) {
2855 /* Grab inode buffer early before taking global s_orphan_lock */
2856 err = ext4_reserve_inode_write(handle, inode, &iloc);
2857 }
2858
2859 mutex_lock(&sbi->s_orphan_lock);
2860 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2861
2862 prev = ei->i_orphan.prev;
2863 list_del_init(&ei->i_orphan);
2864
2865 /* If we're on an error path, we may not have a valid
2866 * transaction handle with which to update the orphan list on
2867 * disk, but we still need to remove the inode from the linked
2868 * list in memory. */
2869 if (!handle || err) {
2870 mutex_unlock(&sbi->s_orphan_lock);
2871 goto out_err;
2872 }
2873
2874 ino_next = NEXT_ORPHAN(inode);
2875 if (prev == &sbi->s_orphan) {
2876 jbd_debug(4, "superblock will point to %u\n", ino_next);
2877 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2878 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2879 if (err) {
2880 mutex_unlock(&sbi->s_orphan_lock);
2881 goto out_brelse;
2882 }
2883 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2884 mutex_unlock(&sbi->s_orphan_lock);
2885 err = ext4_handle_dirty_super(handle, inode->i_sb);
2886 } else {
2887 struct ext4_iloc iloc2;
2888 struct inode *i_prev =
2889 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2890
2891 jbd_debug(4, "orphan inode %lu will point to %u\n",
2892 i_prev->i_ino, ino_next);
2893 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2894 if (err) {
2895 mutex_unlock(&sbi->s_orphan_lock);
2896 goto out_brelse;
2897 }
2898 NEXT_ORPHAN(i_prev) = ino_next;
2899 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2900 mutex_unlock(&sbi->s_orphan_lock);
2901 }
2902 if (err)
2903 goto out_brelse;
2904 NEXT_ORPHAN(inode) = 0;
2905 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2906 out_err:
2907 ext4_std_error(inode->i_sb, err);
2908 return err;
2909
2910 out_brelse:
2911 brelse(iloc.bh);
2912 goto out_err;
2913 }
2914
2915 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2916 {
2917 int retval;
2918 struct inode *inode;
2919 struct buffer_head *bh;
2920 struct ext4_dir_entry_2 *de;
2921 handle_t *handle = NULL;
2922
2923 /* Initialize quotas before so that eventual writes go in
2924 * separate transaction */
2925 retval = dquot_initialize(dir);
2926 if (retval)
2927 return retval;
2928 retval = dquot_initialize(d_inode(dentry));
2929 if (retval)
2930 return retval;
2931
2932 retval = -ENOENT;
2933 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2934 if (IS_ERR(bh))
2935 return PTR_ERR(bh);
2936 if (!bh)
2937 goto end_rmdir;
2938
2939 inode = d_inode(dentry);
2940
2941 retval = -EFSCORRUPTED;
2942 if (le32_to_cpu(de->inode) != inode->i_ino)
2943 goto end_rmdir;
2944
2945 retval = -ENOTEMPTY;
2946 if (!ext4_empty_dir(inode))
2947 goto end_rmdir;
2948
2949 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2950 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2951 if (IS_ERR(handle)) {
2952 retval = PTR_ERR(handle);
2953 handle = NULL;
2954 goto end_rmdir;
2955 }
2956
2957 if (IS_DIRSYNC(dir))
2958 ext4_handle_sync(handle);
2959
2960 retval = ext4_delete_entry(handle, dir, de, bh);
2961 if (retval)
2962 goto end_rmdir;
2963 if (!EXT4_DIR_LINK_EMPTY(inode))
2964 ext4_warning_inode(inode,
2965 "empty directory '%.*s' has too many links (%u)",
2966 dentry->d_name.len, dentry->d_name.name,
2967 inode->i_nlink);
2968 inode->i_version++;
2969 clear_nlink(inode);
2970 /* There's no need to set i_disksize: the fact that i_nlink is
2971 * zero will ensure that the right thing happens during any
2972 * recovery. */
2973 inode->i_size = 0;
2974 ext4_orphan_add(handle, inode);
2975 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2976 ext4_mark_inode_dirty(handle, inode);
2977 ext4_dec_count(handle, dir);
2978 ext4_update_dx_flag(dir);
2979 ext4_mark_inode_dirty(handle, dir);
2980
2981 end_rmdir:
2982 brelse(bh);
2983 if (handle)
2984 ext4_journal_stop(handle);
2985 return retval;
2986 }
2987
2988 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2989 {
2990 int retval;
2991 struct inode *inode;
2992 struct buffer_head *bh;
2993 struct ext4_dir_entry_2 *de;
2994 handle_t *handle = NULL;
2995
2996 trace_ext4_unlink_enter(dir, dentry);
2997 /* Initialize quotas before so that eventual writes go
2998 * in separate transaction */
2999 retval = dquot_initialize(dir);
3000 if (retval)
3001 return retval;
3002 retval = dquot_initialize(d_inode(dentry));
3003 if (retval)
3004 return retval;
3005
3006 retval = -ENOENT;
3007 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3008 if (IS_ERR(bh))
3009 return PTR_ERR(bh);
3010 if (!bh)
3011 goto end_unlink;
3012
3013 inode = d_inode(dentry);
3014
3015 retval = -EFSCORRUPTED;
3016 if (le32_to_cpu(de->inode) != inode->i_ino)
3017 goto end_unlink;
3018
3019 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3020 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3021 if (IS_ERR(handle)) {
3022 retval = PTR_ERR(handle);
3023 handle = NULL;
3024 goto end_unlink;
3025 }
3026
3027 if (IS_DIRSYNC(dir))
3028 ext4_handle_sync(handle);
3029
3030 retval = ext4_delete_entry(handle, dir, de, bh);
3031 if (retval)
3032 goto end_unlink;
3033 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3034 ext4_update_dx_flag(dir);
3035 ext4_mark_inode_dirty(handle, dir);
3036 if (inode->i_nlink == 0)
3037 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3038 dentry->d_name.len, dentry->d_name.name);
3039 else
3040 drop_nlink(inode);
3041 if (!inode->i_nlink)
3042 ext4_orphan_add(handle, inode);
3043 inode->i_ctime = ext4_current_time(inode);
3044 ext4_mark_inode_dirty(handle, inode);
3045
3046 end_unlink:
3047 brelse(bh);
3048 if (handle)
3049 ext4_journal_stop(handle);
3050 trace_ext4_unlink_exit(dentry, retval);
3051 return retval;
3052 }
3053
3054 static int ext4_symlink(struct inode *dir,
3055 struct dentry *dentry, const char *symname)
3056 {
3057 handle_t *handle;
3058 struct inode *inode;
3059 int err, len = strlen(symname);
3060 int credits;
3061 bool encryption_required;
3062 struct ext4_str disk_link;
3063 struct ext4_encrypted_symlink_data *sd = NULL;
3064
3065 disk_link.len = len + 1;
3066 disk_link.name = (char *) symname;
3067
3068 encryption_required = (ext4_encrypted_inode(dir) ||
3069 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3070 if (encryption_required) {
3071 err = ext4_get_encryption_info(dir);
3072 if (err)
3073 return err;
3074 if (ext4_encryption_info(dir) == NULL)
3075 return -EPERM;
3076 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3077 sizeof(struct ext4_encrypted_symlink_data));
3078 sd = kzalloc(disk_link.len, GFP_KERNEL);
3079 if (!sd)
3080 return -ENOMEM;
3081 }
3082
3083 if (disk_link.len > dir->i_sb->s_blocksize) {
3084 err = -ENAMETOOLONG;
3085 goto err_free_sd;
3086 }
3087
3088 err = dquot_initialize(dir);
3089 if (err)
3090 goto err_free_sd;
3091
3092 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3093 /*
3094 * For non-fast symlinks, we just allocate inode and put it on
3095 * orphan list in the first transaction => we need bitmap,
3096 * group descriptor, sb, inode block, quota blocks, and
3097 * possibly selinux xattr blocks.
3098 */
3099 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3100 EXT4_XATTR_TRANS_BLOCKS;
3101 } else {
3102 /*
3103 * Fast symlink. We have to add entry to directory
3104 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3105 * allocate new inode (bitmap, group descriptor, inode block,
3106 * quota blocks, sb is already counted in previous macros).
3107 */
3108 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3109 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3110 }
3111
3112 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3113 &dentry->d_name, 0, NULL,
3114 EXT4_HT_DIR, credits);
3115 handle = ext4_journal_current_handle();
3116 if (IS_ERR(inode)) {
3117 if (handle)
3118 ext4_journal_stop(handle);
3119 err = PTR_ERR(inode);
3120 goto err_free_sd;
3121 }
3122
3123 if (encryption_required) {
3124 struct qstr istr;
3125 struct ext4_str ostr;
3126
3127 istr.name = (const unsigned char *) symname;
3128 istr.len = len;
3129 ostr.name = sd->encrypted_path;
3130 ostr.len = disk_link.len;
3131 err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3132 if (err < 0)
3133 goto err_drop_inode;
3134 sd->len = cpu_to_le16(ostr.len);
3135 disk_link.name = (char *) sd;
3136 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3137 }
3138
3139 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3140 if (!encryption_required)
3141 inode->i_op = &ext4_symlink_inode_operations;
3142 inode_nohighmem(inode);
3143 ext4_set_aops(inode);
3144 /*
3145 * We cannot call page_symlink() with transaction started
3146 * because it calls into ext4_write_begin() which can wait
3147 * for transaction commit if we are running out of space
3148 * and thus we deadlock. So we have to stop transaction now
3149 * and restart it when symlink contents is written.
3150 *
3151 * To keep fs consistent in case of crash, we have to put inode
3152 * to orphan list in the mean time.
3153 */
3154 drop_nlink(inode);
3155 err = ext4_orphan_add(handle, inode);
3156 ext4_journal_stop(handle);
3157 handle = NULL;
3158 if (err)
3159 goto err_drop_inode;
3160 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3161 if (err)
3162 goto err_drop_inode;
3163 /*
3164 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3165 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3166 */
3167 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3168 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3169 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3170 if (IS_ERR(handle)) {
3171 err = PTR_ERR(handle);
3172 handle = NULL;
3173 goto err_drop_inode;
3174 }
3175 set_nlink(inode, 1);
3176 err = ext4_orphan_del(handle, inode);
3177 if (err)
3178 goto err_drop_inode;
3179 } else {
3180 /* clear the extent format for fast symlink */
3181 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3182 if (!encryption_required) {
3183 inode->i_op = &ext4_fast_symlink_inode_operations;
3184 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3185 }
3186 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3187 disk_link.len);
3188 inode->i_size = disk_link.len - 1;
3189 }
3190 EXT4_I(inode)->i_disksize = inode->i_size;
3191 err = ext4_add_nondir(handle, dentry, inode);
3192 if (!err && IS_DIRSYNC(dir))
3193 ext4_handle_sync(handle);
3194
3195 if (handle)
3196 ext4_journal_stop(handle);
3197 kfree(sd);
3198 return err;
3199 err_drop_inode:
3200 if (handle)
3201 ext4_journal_stop(handle);
3202 clear_nlink(inode);
3203 unlock_new_inode(inode);
3204 iput(inode);
3205 err_free_sd:
3206 kfree(sd);
3207 return err;
3208 }
3209
3210 static int ext4_link(struct dentry *old_dentry,
3211 struct inode *dir, struct dentry *dentry)
3212 {
3213 handle_t *handle;
3214 struct inode *inode = d_inode(old_dentry);
3215 int err, retries = 0;
3216
3217 if (inode->i_nlink >= EXT4_LINK_MAX)
3218 return -EMLINK;
3219 if (ext4_encrypted_inode(dir) &&
3220 !ext4_is_child_context_consistent_with_parent(dir, inode))
3221 return -EPERM;
3222 err = dquot_initialize(dir);
3223 if (err)
3224 return err;
3225
3226 retry:
3227 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3228 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3229 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3230 if (IS_ERR(handle))
3231 return PTR_ERR(handle);
3232
3233 if (IS_DIRSYNC(dir))
3234 ext4_handle_sync(handle);
3235
3236 inode->i_ctime = ext4_current_time(inode);
3237 ext4_inc_count(handle, inode);
3238 ihold(inode);
3239
3240 err = ext4_add_entry(handle, dentry, inode);
3241 if (!err) {
3242 ext4_mark_inode_dirty(handle, inode);
3243 /* this can happen only for tmpfile being
3244 * linked the first time
3245 */
3246 if (inode->i_nlink == 1)
3247 ext4_orphan_del(handle, inode);
3248 d_instantiate(dentry, inode);
3249 } else {
3250 drop_nlink(inode);
3251 iput(inode);
3252 }
3253 ext4_journal_stop(handle);
3254 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3255 goto retry;
3256 return err;
3257 }
3258
3259
3260 /*
3261 * Try to find buffer head where contains the parent block.
3262 * It should be the inode block if it is inlined or the 1st block
3263 * if it is a normal dir.
3264 */
3265 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3266 struct inode *inode,
3267 int *retval,
3268 struct ext4_dir_entry_2 **parent_de,
3269 int *inlined)
3270 {
3271 struct buffer_head *bh;
3272
3273 if (!ext4_has_inline_data(inode)) {
3274 bh = ext4_read_dirblock(inode, 0, EITHER);
3275 if (IS_ERR(bh)) {
3276 *retval = PTR_ERR(bh);
3277 return NULL;
3278 }
3279 *parent_de = ext4_next_entry(
3280 (struct ext4_dir_entry_2 *)bh->b_data,
3281 inode->i_sb->s_blocksize);
3282 return bh;
3283 }
3284
3285 *inlined = 1;
3286 return ext4_get_first_inline_block(inode, parent_de, retval);
3287 }
3288
3289 struct ext4_renament {
3290 struct inode *dir;
3291 struct dentry *dentry;
3292 struct inode *inode;
3293 bool is_dir;
3294 int dir_nlink_delta;
3295
3296 /* entry for "dentry" */
3297 struct buffer_head *bh;
3298 struct ext4_dir_entry_2 *de;
3299 int inlined;
3300
3301 /* entry for ".." in inode if it's a directory */
3302 struct buffer_head *dir_bh;
3303 struct ext4_dir_entry_2 *parent_de;
3304 int dir_inlined;
3305 };
3306
3307 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3308 {
3309 int retval;
3310
3311 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3312 &retval, &ent->parent_de,
3313 &ent->dir_inlined);
3314 if (!ent->dir_bh)
3315 return retval;
3316 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3317 return -EFSCORRUPTED;
3318 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3319 return ext4_journal_get_write_access(handle, ent->dir_bh);
3320 }
3321
3322 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3323 unsigned dir_ino)
3324 {
3325 int retval;
3326
3327 ent->parent_de->inode = cpu_to_le32(dir_ino);
3328 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3329 if (!ent->dir_inlined) {
3330 if (is_dx(ent->inode)) {
3331 retval = ext4_handle_dirty_dx_node(handle,
3332 ent->inode,
3333 ent->dir_bh);
3334 } else {
3335 retval = ext4_handle_dirty_dirent_node(handle,
3336 ent->inode,
3337 ent->dir_bh);
3338 }
3339 } else {
3340 retval = ext4_mark_inode_dirty(handle, ent->inode);
3341 }
3342 if (retval) {
3343 ext4_std_error(ent->dir->i_sb, retval);
3344 return retval;
3345 }
3346 return 0;
3347 }
3348
3349 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3350 unsigned ino, unsigned file_type)
3351 {
3352 int retval;
3353
3354 BUFFER_TRACE(ent->bh, "get write access");
3355 retval = ext4_journal_get_write_access(handle, ent->bh);
3356 if (retval)
3357 return retval;
3358 ent->de->inode = cpu_to_le32(ino);
3359 if (ext4_has_feature_filetype(ent->dir->i_sb))
3360 ent->de->file_type = file_type;
3361 ent->dir->i_version++;
3362 ent->dir->i_ctime = ent->dir->i_mtime =
3363 ext4_current_time(ent->dir);
3364 ext4_mark_inode_dirty(handle, ent->dir);
3365 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3366 if (!ent->inlined) {
3367 retval = ext4_handle_dirty_dirent_node(handle,
3368 ent->dir, ent->bh);
3369 if (unlikely(retval)) {
3370 ext4_std_error(ent->dir->i_sb, retval);
3371 return retval;
3372 }
3373 }
3374 brelse(ent->bh);
3375 ent->bh = NULL;
3376
3377 return 0;
3378 }
3379
3380 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3381 const struct qstr *d_name)
3382 {
3383 int retval = -ENOENT;
3384 struct buffer_head *bh;
3385 struct ext4_dir_entry_2 *de;
3386
3387 bh = ext4_find_entry(dir, d_name, &de, NULL);
3388 if (IS_ERR(bh))
3389 return PTR_ERR(bh);
3390 if (bh) {
3391 retval = ext4_delete_entry(handle, dir, de, bh);
3392 brelse(bh);
3393 }
3394 return retval;
3395 }
3396
3397 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3398 int force_reread)
3399 {
3400 int retval;
3401 /*
3402 * ent->de could have moved from under us during htree split, so make
3403 * sure that we are deleting the right entry. We might also be pointing
3404 * to a stale entry in the unused part of ent->bh so just checking inum
3405 * and the name isn't enough.
3406 */
3407 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3408 ent->de->name_len != ent->dentry->d_name.len ||
3409 strncmp(ent->de->name, ent->dentry->d_name.name,
3410 ent->de->name_len) ||
3411 force_reread) {
3412 retval = ext4_find_delete_entry(handle, ent->dir,
3413 &ent->dentry->d_name);
3414 } else {
3415 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3416 if (retval == -ENOENT) {
3417 retval = ext4_find_delete_entry(handle, ent->dir,
3418 &ent->dentry->d_name);
3419 }
3420 }
3421
3422 if (retval) {
3423 ext4_warning_inode(ent->dir,
3424 "Deleting old file: nlink %d, error=%d",
3425 ent->dir->i_nlink, retval);
3426 }
3427 }
3428
3429 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3430 {
3431 if (ent->dir_nlink_delta) {
3432 if (ent->dir_nlink_delta == -1)
3433 ext4_dec_count(handle, ent->dir);
3434 else
3435 ext4_inc_count(handle, ent->dir);
3436 ext4_mark_inode_dirty(handle, ent->dir);
3437 }
3438 }
3439
3440 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3441 int credits, handle_t **h)
3442 {
3443 struct inode *wh;
3444 handle_t *handle;
3445 int retries = 0;
3446
3447 /*
3448 * for inode block, sb block, group summaries,
3449 * and inode bitmap
3450 */
3451 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3452 EXT4_XATTR_TRANS_BLOCKS + 4);
3453 retry:
3454 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3455 &ent->dentry->d_name, 0, NULL,
3456 EXT4_HT_DIR, credits);
3457
3458 handle = ext4_journal_current_handle();
3459 if (IS_ERR(wh)) {
3460 if (handle)
3461 ext4_journal_stop(handle);
3462 if (PTR_ERR(wh) == -ENOSPC &&
3463 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3464 goto retry;
3465 } else {
3466 *h = handle;
3467 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3468 wh->i_op = &ext4_special_inode_operations;
3469 }
3470 return wh;
3471 }
3472
3473 /*
3474 * Anybody can rename anything with this: the permission checks are left to the
3475 * higher-level routines.
3476 *
3477 * n.b. old_{dentry,inode) refers to the source dentry/inode
3478 * while new_{dentry,inode) refers to the destination dentry/inode
3479 * This comes from rename(const char *oldpath, const char *newpath)
3480 */
3481 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3482 struct inode *new_dir, struct dentry *new_dentry,
3483 unsigned int flags)
3484 {
3485 handle_t *handle = NULL;
3486 struct ext4_renament old = {
3487 .dir = old_dir,
3488 .dentry = old_dentry,
3489 .inode = d_inode(old_dentry),
3490 };
3491 struct ext4_renament new = {
3492 .dir = new_dir,
3493 .dentry = new_dentry,
3494 .inode = d_inode(new_dentry),
3495 };
3496 int force_reread;
3497 int retval;
3498 struct inode *whiteout = NULL;
3499 int credits;
3500 u8 old_file_type;
3501
3502 if ((ext4_encrypted_inode(old_dir) &&
3503 !ext4_has_encryption_key(old_dir)) ||
3504 (ext4_encrypted_inode(new_dir) &&
3505 !ext4_has_encryption_key(new_dir)))
3506 return -ENOKEY;
3507
3508 retval = dquot_initialize(old.dir);
3509 if (retval)
3510 return retval;
3511 retval = dquot_initialize(new.dir);
3512 if (retval)
3513 return retval;
3514
3515 /* Initialize quotas before so that eventual writes go
3516 * in separate transaction */
3517 if (new.inode) {
3518 retval = dquot_initialize(new.inode);
3519 if (retval)
3520 return retval;
3521 }
3522
3523 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3524 if (IS_ERR(old.bh))
3525 return PTR_ERR(old.bh);
3526 /*
3527 * Check for inode number is _not_ due to possible IO errors.
3528 * We might rmdir the source, keep it as pwd of some process
3529 * and merrily kill the link to whatever was created under the
3530 * same name. Goodbye sticky bit ;-<
3531 */
3532 retval = -ENOENT;
3533 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3534 goto end_rename;
3535
3536 if ((old.dir != new.dir) &&
3537 ext4_encrypted_inode(new.dir) &&
3538 !ext4_is_child_context_consistent_with_parent(new.dir,
3539 old.inode)) {
3540 retval = -EPERM;
3541 goto end_rename;
3542 }
3543
3544 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3545 &new.de, &new.inlined);
3546 if (IS_ERR(new.bh)) {
3547 retval = PTR_ERR(new.bh);
3548 new.bh = NULL;
3549 goto end_rename;
3550 }
3551 if (new.bh) {
3552 if (!new.inode) {
3553 brelse(new.bh);
3554 new.bh = NULL;
3555 }
3556 }
3557 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3558 ext4_alloc_da_blocks(old.inode);
3559
3560 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3561 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3562 if (!(flags & RENAME_WHITEOUT)) {
3563 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3564 if (IS_ERR(handle)) {
3565 retval = PTR_ERR(handle);
3566 handle = NULL;
3567 goto end_rename;
3568 }
3569 } else {
3570 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3571 if (IS_ERR(whiteout)) {
3572 retval = PTR_ERR(whiteout);
3573 whiteout = NULL;
3574 goto end_rename;
3575 }
3576 }
3577
3578 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3579 ext4_handle_sync(handle);
3580
3581 if (S_ISDIR(old.inode->i_mode)) {
3582 if (new.inode) {
3583 retval = -ENOTEMPTY;
3584 if (!ext4_empty_dir(new.inode))
3585 goto end_rename;
3586 } else {
3587 retval = -EMLINK;
3588 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3589 goto end_rename;
3590 }
3591 retval = ext4_rename_dir_prepare(handle, &old);
3592 if (retval)
3593 goto end_rename;
3594 }
3595 /*
3596 * If we're renaming a file within an inline_data dir and adding or
3597 * setting the new dirent causes a conversion from inline_data to
3598 * extents/blockmap, we need to force the dirent delete code to
3599 * re-read the directory, or else we end up trying to delete a dirent
3600 * from what is now the extent tree root (or a block map).
3601 */
3602 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3603 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3604
3605 old_file_type = old.de->file_type;
3606 if (whiteout) {
3607 /*
3608 * Do this before adding a new entry, so the old entry is sure
3609 * to be still pointing to the valid old entry.
3610 */
3611 retval = ext4_setent(handle, &old, whiteout->i_ino,
3612 EXT4_FT_CHRDEV);
3613 if (retval)
3614 goto end_rename;
3615 ext4_mark_inode_dirty(handle, whiteout);
3616 }
3617 if (!new.bh) {
3618 retval = ext4_add_entry(handle, new.dentry, old.inode);
3619 if (retval)
3620 goto end_rename;
3621 } else {
3622 retval = ext4_setent(handle, &new,
3623 old.inode->i_ino, old_file_type);
3624 if (retval)
3625 goto end_rename;
3626 }
3627 if (force_reread)
3628 force_reread = !ext4_test_inode_flag(new.dir,
3629 EXT4_INODE_INLINE_DATA);
3630
3631 /*
3632 * Like most other Unix systems, set the ctime for inodes on a
3633 * rename.
3634 */
3635 old.inode->i_ctime = ext4_current_time(old.inode);
3636 ext4_mark_inode_dirty(handle, old.inode);
3637
3638 if (!whiteout) {
3639 /*
3640 * ok, that's it
3641 */
3642 ext4_rename_delete(handle, &old, force_reread);
3643 }
3644
3645 if (new.inode) {
3646 ext4_dec_count(handle, new.inode);
3647 new.inode->i_ctime = ext4_current_time(new.inode);
3648 }
3649 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3650 ext4_update_dx_flag(old.dir);
3651 if (old.dir_bh) {
3652 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3653 if (retval)
3654 goto end_rename;
3655
3656 ext4_dec_count(handle, old.dir);
3657 if (new.inode) {
3658 /* checked ext4_empty_dir above, can't have another
3659 * parent, ext4_dec_count() won't work for many-linked
3660 * dirs */
3661 clear_nlink(new.inode);
3662 } else {
3663 ext4_inc_count(handle, new.dir);
3664 ext4_update_dx_flag(new.dir);
3665 ext4_mark_inode_dirty(handle, new.dir);
3666 }
3667 }
3668 ext4_mark_inode_dirty(handle, old.dir);
3669 if (new.inode) {
3670 ext4_mark_inode_dirty(handle, new.inode);
3671 if (!new.inode->i_nlink)
3672 ext4_orphan_add(handle, new.inode);
3673 }
3674 retval = 0;
3675
3676 end_rename:
3677 brelse(old.dir_bh);
3678 brelse(old.bh);
3679 brelse(new.bh);
3680 if (whiteout) {
3681 if (retval)
3682 drop_nlink(whiteout);
3683 unlock_new_inode(whiteout);
3684 iput(whiteout);
3685 }
3686 if (handle)
3687 ext4_journal_stop(handle);
3688 return retval;
3689 }
3690
3691 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3692 struct inode *new_dir, struct dentry *new_dentry)
3693 {
3694 handle_t *handle = NULL;
3695 struct ext4_renament old = {
3696 .dir = old_dir,
3697 .dentry = old_dentry,
3698 .inode = d_inode(old_dentry),
3699 };
3700 struct ext4_renament new = {
3701 .dir = new_dir,
3702 .dentry = new_dentry,
3703 .inode = d_inode(new_dentry),
3704 };
3705 u8 new_file_type;
3706 int retval;
3707
3708 if ((ext4_encrypted_inode(old_dir) &&
3709 !ext4_has_encryption_key(old_dir)) ||
3710 (ext4_encrypted_inode(new_dir) &&
3711 !ext4_has_encryption_key(new_dir)))
3712 return -ENOKEY;
3713
3714 if ((ext4_encrypted_inode(old_dir) ||
3715 ext4_encrypted_inode(new_dir)) &&
3716 (old_dir != new_dir) &&
3717 (!ext4_is_child_context_consistent_with_parent(new_dir,
3718 old.inode) ||
3719 !ext4_is_child_context_consistent_with_parent(old_dir,
3720 new.inode)))
3721 return -EPERM;
3722
3723 retval = dquot_initialize(old.dir);
3724 if (retval)
3725 return retval;
3726 retval = dquot_initialize(new.dir);
3727 if (retval)
3728 return retval;
3729
3730 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3731 &old.de, &old.inlined);
3732 if (IS_ERR(old.bh))
3733 return PTR_ERR(old.bh);
3734 /*
3735 * Check for inode number is _not_ due to possible IO errors.
3736 * We might rmdir the source, keep it as pwd of some process
3737 * and merrily kill the link to whatever was created under the
3738 * same name. Goodbye sticky bit ;-<
3739 */
3740 retval = -ENOENT;
3741 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3742 goto end_rename;
3743
3744 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3745 &new.de, &new.inlined);
3746 if (IS_ERR(new.bh)) {
3747 retval = PTR_ERR(new.bh);
3748 new.bh = NULL;
3749 goto end_rename;
3750 }
3751
3752 /* RENAME_EXCHANGE case: old *and* new must both exist */
3753 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3754 goto end_rename;
3755
3756 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3757 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3758 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3759 if (IS_ERR(handle)) {
3760 retval = PTR_ERR(handle);
3761 handle = NULL;
3762 goto end_rename;
3763 }
3764
3765 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3766 ext4_handle_sync(handle);
3767
3768 if (S_ISDIR(old.inode->i_mode)) {
3769 old.is_dir = true;
3770 retval = ext4_rename_dir_prepare(handle, &old);
3771 if (retval)
3772 goto end_rename;
3773 }
3774 if (S_ISDIR(new.inode->i_mode)) {
3775 new.is_dir = true;
3776 retval = ext4_rename_dir_prepare(handle, &new);
3777 if (retval)
3778 goto end_rename;
3779 }
3780
3781 /*
3782 * Other than the special case of overwriting a directory, parents'
3783 * nlink only needs to be modified if this is a cross directory rename.
3784 */
3785 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3786 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3787 new.dir_nlink_delta = -old.dir_nlink_delta;
3788 retval = -EMLINK;
3789 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3790 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3791 goto end_rename;
3792 }
3793
3794 new_file_type = new.de->file_type;
3795 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3796 if (retval)
3797 goto end_rename;
3798
3799 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3800 if (retval)
3801 goto end_rename;
3802
3803 /*
3804 * Like most other Unix systems, set the ctime for inodes on a
3805 * rename.
3806 */
3807 old.inode->i_ctime = ext4_current_time(old.inode);
3808 new.inode->i_ctime = ext4_current_time(new.inode);
3809 ext4_mark_inode_dirty(handle, old.inode);
3810 ext4_mark_inode_dirty(handle, new.inode);
3811
3812 if (old.dir_bh) {
3813 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3814 if (retval)
3815 goto end_rename;
3816 }
3817 if (new.dir_bh) {
3818 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3819 if (retval)
3820 goto end_rename;
3821 }
3822 ext4_update_dir_count(handle, &old);
3823 ext4_update_dir_count(handle, &new);
3824 retval = 0;
3825
3826 end_rename:
3827 brelse(old.dir_bh);
3828 brelse(new.dir_bh);
3829 brelse(old.bh);
3830 brelse(new.bh);
3831 if (handle)
3832 ext4_journal_stop(handle);
3833 return retval;
3834 }
3835
3836 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3837 struct inode *new_dir, struct dentry *new_dentry,
3838 unsigned int flags)
3839 {
3840 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3841 return -EINVAL;
3842
3843 if (flags & RENAME_EXCHANGE) {
3844 return ext4_cross_rename(old_dir, old_dentry,
3845 new_dir, new_dentry);
3846 }
3847
3848 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3849 }
3850
3851 /*
3852 * directories can handle most operations...
3853 */
3854 const struct inode_operations ext4_dir_inode_operations = {
3855 .create = ext4_create,
3856 .lookup = ext4_lookup,
3857 .link = ext4_link,
3858 .unlink = ext4_unlink,
3859 .symlink = ext4_symlink,
3860 .mkdir = ext4_mkdir,
3861 .rmdir = ext4_rmdir,
3862 .mknod = ext4_mknod,
3863 .tmpfile = ext4_tmpfile,
3864 .rename2 = ext4_rename2,
3865 .setattr = ext4_setattr,
3866 .setxattr = generic_setxattr,
3867 .getxattr = generic_getxattr,
3868 .listxattr = ext4_listxattr,
3869 .removexattr = generic_removexattr,
3870 .get_acl = ext4_get_acl,
3871 .set_acl = ext4_set_acl,
3872 .fiemap = ext4_fiemap,
3873 };
3874
3875 const struct inode_operations ext4_special_inode_operations = {
3876 .setattr = ext4_setattr,
3877 .setxattr = generic_setxattr,
3878 .getxattr = generic_getxattr,
3879 .listxattr = ext4_listxattr,
3880 .removexattr = generic_removexattr,
3881 .get_acl = ext4_get_acl,
3882 .set_acl = ext4_set_acl,
3883 };
Linux with added FPC-III support