Loading...
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/jbd2.h>
30#include <linux/time.h>
31#include <linux/fcntl.h>
32#include <linux/stat.h>
33#include <linux/string.h>
34#include <linux/quotaops.h>
35#include <linux/buffer_head.h>
36#include <linux/bio.h>
37#include "ext4.h"
38#include "ext4_jbd2.h"
39
40#include "xattr.h"
41#include "acl.h"
42
43#include <trace/events/ext4.h>
44/*
45 * define how far ahead to read directories while searching them.
46 */
47#define NAMEI_RA_CHUNKS 2
48#define NAMEI_RA_BLOCKS 4
49#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51
52static struct buffer_head *ext4_append(handle_t *handle,
53 struct inode *inode,
54 ext4_lblk_t *block, int *err)
55{
56 struct buffer_head *bh;
57
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60 bh = ext4_bread(handle, inode, *block, 1, err);
61 if (bh) {
62 inode->i_size += inode->i_sb->s_blocksize;
63 EXT4_I(inode)->i_disksize = inode->i_size;
64 *err = ext4_journal_get_write_access(handle, bh);
65 if (*err) {
66 brelse(bh);
67 bh = NULL;
68 }
69 }
70 return bh;
71}
72
73#ifndef assert
74#define assert(test) J_ASSERT(test)
75#endif
76
77#ifdef DX_DEBUG
78#define dxtrace(command) command
79#else
80#define dxtrace(command)
81#endif
82
83struct fake_dirent
84{
85 __le32 inode;
86 __le16 rec_len;
87 u8 name_len;
88 u8 file_type;
89};
90
91struct dx_countlimit
92{
93 __le16 limit;
94 __le16 count;
95};
96
97struct dx_entry
98{
99 __le32 hash;
100 __le32 block;
101};
102
103/*
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
107 */
108
109struct dx_root
110{
111 struct fake_dirent dot;
112 char dot_name[4];
113 struct fake_dirent dotdot;
114 char dotdot_name[4];
115 struct dx_root_info
116 {
117 __le32 reserved_zero;
118 u8 hash_version;
119 u8 info_length; /* 8 */
120 u8 indirect_levels;
121 u8 unused_flags;
122 }
123 info;
124 struct dx_entry entries[0];
125};
126
127struct dx_node
128{
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
131};
132
133
134struct dx_frame
135{
136 struct buffer_head *bh;
137 struct dx_entry *entries;
138 struct dx_entry *at;
139};
140
141struct dx_map_entry
142{
143 u32 hash;
144 u16 offs;
145 u16 size;
146};
147
148/*
149 * This goes at the end of each htree block.
150 */
151struct dx_tail {
152 u32 dt_reserved;
153 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
154};
155
156static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
157static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
158static inline unsigned dx_get_hash(struct dx_entry *entry);
159static void dx_set_hash(struct dx_entry *entry, unsigned value);
160static unsigned dx_get_count(struct dx_entry *entries);
161static unsigned dx_get_limit(struct dx_entry *entries);
162static void dx_set_count(struct dx_entry *entries, unsigned value);
163static void dx_set_limit(struct dx_entry *entries, unsigned value);
164static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
165static unsigned dx_node_limit(struct inode *dir);
166static struct dx_frame *dx_probe(const struct qstr *d_name,
167 struct inode *dir,
168 struct dx_hash_info *hinfo,
169 struct dx_frame *frame,
170 int *err);
171static void dx_release(struct dx_frame *frames);
172static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
173 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
174static void dx_sort_map(struct dx_map_entry *map, unsigned count);
175static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
176 struct dx_map_entry *offsets, int count, unsigned blocksize);
177static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
178static void dx_insert_block(struct dx_frame *frame,
179 u32 hash, ext4_lblk_t block);
180static int ext4_htree_next_block(struct inode *dir, __u32 hash,
181 struct dx_frame *frame,
182 struct dx_frame *frames,
183 __u32 *start_hash);
184static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
185 const struct qstr *d_name,
186 struct ext4_dir_entry_2 **res_dir,
187 int *err);
188static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
189 struct inode *inode);
190
191/* checksumming functions */
192#define EXT4_DIRENT_TAIL(block, blocksize) \
193 ((struct ext4_dir_entry_tail *)(((void *)(block)) + \
194 ((blocksize) - \
195 sizeof(struct ext4_dir_entry_tail))))
196
197static void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
198 unsigned int blocksize)
199{
200 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
201 t->det_rec_len = ext4_rec_len_to_disk(
202 sizeof(struct ext4_dir_entry_tail), blocksize);
203 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
204}
205
206/* Walk through a dirent block to find a checksum "dirent" at the tail */
207static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
208 struct ext4_dir_entry *de)
209{
210 struct ext4_dir_entry_tail *t;
211
212#ifdef PARANOID
213 struct ext4_dir_entry *d, *top;
214
215 d = de;
216 top = (struct ext4_dir_entry *)(((void *)de) +
217 (EXT4_BLOCK_SIZE(inode->i_sb) -
218 sizeof(struct ext4_dir_entry_tail)));
219 while (d < top && d->rec_len)
220 d = (struct ext4_dir_entry *)(((void *)d) +
221 le16_to_cpu(d->rec_len));
222
223 if (d != top)
224 return NULL;
225
226 t = (struct ext4_dir_entry_tail *)d;
227#else
228 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
229#endif
230
231 if (t->det_reserved_zero1 ||
232 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
233 t->det_reserved_zero2 ||
234 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
235 return NULL;
236
237 return t;
238}
239
240static __le32 ext4_dirent_csum(struct inode *inode,
241 struct ext4_dir_entry *dirent, int size)
242{
243 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
244 struct ext4_inode_info *ei = EXT4_I(inode);
245 __u32 csum;
246
247 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
248 return cpu_to_le32(csum);
249}
250
251int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
252{
253 struct ext4_dir_entry_tail *t;
254
255 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
256 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
257 return 1;
258
259 t = get_dirent_tail(inode, dirent);
260 if (!t) {
261 EXT4_ERROR_INODE(inode, "metadata_csum set but no space in dir "
262 "leaf for checksum. Please run e2fsck -D.");
263 return 0;
264 }
265
266 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
267 (void *)t - (void *)dirent))
268 return 0;
269
270 return 1;
271}
272
273static void ext4_dirent_csum_set(struct inode *inode,
274 struct ext4_dir_entry *dirent)
275{
276 struct ext4_dir_entry_tail *t;
277
278 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
279 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
280 return;
281
282 t = get_dirent_tail(inode, dirent);
283 if (!t) {
284 EXT4_ERROR_INODE(inode, "metadata_csum set but no space in dir "
285 "leaf for checksum. Please run e2fsck -D.");
286 return;
287 }
288
289 t->det_checksum = ext4_dirent_csum(inode, dirent,
290 (void *)t - (void *)dirent);
291}
292
293static inline int ext4_handle_dirty_dirent_node(handle_t *handle,
294 struct inode *inode,
295 struct buffer_head *bh)
296{
297 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
298 return ext4_handle_dirty_metadata(handle, inode, bh);
299}
300
301static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
302 struct ext4_dir_entry *dirent,
303 int *offset)
304{
305 struct ext4_dir_entry *dp;
306 struct dx_root_info *root;
307 int count_offset;
308
309 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
310 count_offset = 8;
311 else if (le16_to_cpu(dirent->rec_len) == 12) {
312 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
313 if (le16_to_cpu(dp->rec_len) !=
314 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
315 return NULL;
316 root = (struct dx_root_info *)(((void *)dp + 12));
317 if (root->reserved_zero ||
318 root->info_length != sizeof(struct dx_root_info))
319 return NULL;
320 count_offset = 32;
321 } else
322 return NULL;
323
324 if (offset)
325 *offset = count_offset;
326 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
327}
328
329static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
330 int count_offset, int count, struct dx_tail *t)
331{
332 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
333 struct ext4_inode_info *ei = EXT4_I(inode);
334 __u32 csum, old_csum;
335 int size;
336
337 size = count_offset + (count * sizeof(struct dx_entry));
338 old_csum = t->dt_checksum;
339 t->dt_checksum = 0;
340 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
341 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
342 t->dt_checksum = old_csum;
343
344 return cpu_to_le32(csum);
345}
346
347static int ext4_dx_csum_verify(struct inode *inode,
348 struct ext4_dir_entry *dirent)
349{
350 struct dx_countlimit *c;
351 struct dx_tail *t;
352 int count_offset, limit, count;
353
354 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
355 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
356 return 1;
357
358 c = get_dx_countlimit(inode, dirent, &count_offset);
359 if (!c) {
360 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
361 return 1;
362 }
363 limit = le16_to_cpu(c->limit);
364 count = le16_to_cpu(c->count);
365 if (count_offset + (limit * sizeof(struct dx_entry)) >
366 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
367 EXT4_ERROR_INODE(inode, "metadata_csum set but no space for "
368 "tree checksum found. Run e2fsck -D.");
369 return 1;
370 }
371 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
372
373 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
374 count, t))
375 return 0;
376 return 1;
377}
378
379static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
380{
381 struct dx_countlimit *c;
382 struct dx_tail *t;
383 int count_offset, limit, count;
384
385 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
386 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
387 return;
388
389 c = get_dx_countlimit(inode, dirent, &count_offset);
390 if (!c) {
391 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
392 return;
393 }
394 limit = le16_to_cpu(c->limit);
395 count = le16_to_cpu(c->count);
396 if (count_offset + (limit * sizeof(struct dx_entry)) >
397 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
398 EXT4_ERROR_INODE(inode, "metadata_csum set but no space for "
399 "tree checksum. Run e2fsck -D.");
400 return;
401 }
402 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
403
404 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
405}
406
407static inline int ext4_handle_dirty_dx_node(handle_t *handle,
408 struct inode *inode,
409 struct buffer_head *bh)
410{
411 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
412 return ext4_handle_dirty_metadata(handle, inode, bh);
413}
414
415/*
416 * p is at least 6 bytes before the end of page
417 */
418static inline struct ext4_dir_entry_2 *
419ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
420{
421 return (struct ext4_dir_entry_2 *)((char *)p +
422 ext4_rec_len_from_disk(p->rec_len, blocksize));
423}
424
425/*
426 * Future: use high four bits of block for coalesce-on-delete flags
427 * Mask them off for now.
428 */
429
430static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
431{
432 return le32_to_cpu(entry->block) & 0x00ffffff;
433}
434
435static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
436{
437 entry->block = cpu_to_le32(value);
438}
439
440static inline unsigned dx_get_hash(struct dx_entry *entry)
441{
442 return le32_to_cpu(entry->hash);
443}
444
445static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
446{
447 entry->hash = cpu_to_le32(value);
448}
449
450static inline unsigned dx_get_count(struct dx_entry *entries)
451{
452 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
453}
454
455static inline unsigned dx_get_limit(struct dx_entry *entries)
456{
457 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
458}
459
460static inline void dx_set_count(struct dx_entry *entries, unsigned value)
461{
462 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
463}
464
465static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
466{
467 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
468}
469
470static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
471{
472 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
473 EXT4_DIR_REC_LEN(2) - infosize;
474
475 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
476 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
477 entry_space -= sizeof(struct dx_tail);
478 return entry_space / sizeof(struct dx_entry);
479}
480
481static inline unsigned dx_node_limit(struct inode *dir)
482{
483 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
484
485 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
486 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
487 entry_space -= sizeof(struct dx_tail);
488 return entry_space / sizeof(struct dx_entry);
489}
490
491/*
492 * Debug
493 */
494#ifdef DX_DEBUG
495static void dx_show_index(char * label, struct dx_entry *entries)
496{
497 int i, n = dx_get_count (entries);
498 printk(KERN_DEBUG "%s index ", label);
499 for (i = 0; i < n; i++) {
500 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
501 0, (unsigned long)dx_get_block(entries + i));
502 }
503 printk("\n");
504}
505
506struct stats
507{
508 unsigned names;
509 unsigned space;
510 unsigned bcount;
511};
512
513static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
514 int size, int show_names)
515{
516 unsigned names = 0, space = 0;
517 char *base = (char *) de;
518 struct dx_hash_info h = *hinfo;
519
520 printk("names: ");
521 while ((char *) de < base + size)
522 {
523 if (de->inode)
524 {
525 if (show_names)
526 {
527 int len = de->name_len;
528 char *name = de->name;
529 while (len--) printk("%c", *name++);
530 ext4fs_dirhash(de->name, de->name_len, &h);
531 printk(":%x.%u ", h.hash,
532 (unsigned) ((char *) de - base));
533 }
534 space += EXT4_DIR_REC_LEN(de->name_len);
535 names++;
536 }
537 de = ext4_next_entry(de, size);
538 }
539 printk("(%i)\n", names);
540 return (struct stats) { names, space, 1 };
541}
542
543struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
544 struct dx_entry *entries, int levels)
545{
546 unsigned blocksize = dir->i_sb->s_blocksize;
547 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
548 unsigned bcount = 0;
549 struct buffer_head *bh;
550 int err;
551 printk("%i indexed blocks...\n", count);
552 for (i = 0; i < count; i++, entries++)
553 {
554 ext4_lblk_t block = dx_get_block(entries);
555 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
556 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
557 struct stats stats;
558 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
559 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
560 stats = levels?
561 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
562 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
563 names += stats.names;
564 space += stats.space;
565 bcount += stats.bcount;
566 brelse(bh);
567 }
568 if (bcount)
569 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
570 levels ? "" : " ", names, space/bcount,
571 (space/bcount)*100/blocksize);
572 return (struct stats) { names, space, bcount};
573}
574#endif /* DX_DEBUG */
575
576/*
577 * Probe for a directory leaf block to search.
578 *
579 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
580 * error in the directory index, and the caller should fall back to
581 * searching the directory normally. The callers of dx_probe **MUST**
582 * check for this error code, and make sure it never gets reflected
583 * back to userspace.
584 */
585static struct dx_frame *
586dx_probe(const struct qstr *d_name, struct inode *dir,
587 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
588{
589 unsigned count, indirect;
590 struct dx_entry *at, *entries, *p, *q, *m;
591 struct dx_root *root;
592 struct buffer_head *bh;
593 struct dx_frame *frame = frame_in;
594 u32 hash;
595
596 frame->bh = NULL;
597 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
598 goto fail;
599 root = (struct dx_root *) bh->b_data;
600 if (root->info.hash_version != DX_HASH_TEA &&
601 root->info.hash_version != DX_HASH_HALF_MD4 &&
602 root->info.hash_version != DX_HASH_LEGACY) {
603 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
604 root->info.hash_version);
605 brelse(bh);
606 *err = ERR_BAD_DX_DIR;
607 goto fail;
608 }
609 hinfo->hash_version = root->info.hash_version;
610 if (hinfo->hash_version <= DX_HASH_TEA)
611 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
612 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
613 if (d_name)
614 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
615 hash = hinfo->hash;
616
617 if (root->info.unused_flags & 1) {
618 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
619 root->info.unused_flags);
620 brelse(bh);
621 *err = ERR_BAD_DX_DIR;
622 goto fail;
623 }
624
625 if ((indirect = root->info.indirect_levels) > 1) {
626 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
627 root->info.indirect_levels);
628 brelse(bh);
629 *err = ERR_BAD_DX_DIR;
630 goto fail;
631 }
632
633 if (!buffer_verified(bh) &&
634 !ext4_dx_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data)) {
635 ext4_warning(dir->i_sb, "Root failed checksum");
636 brelse(bh);
637 *err = ERR_BAD_DX_DIR;
638 goto fail;
639 }
640 set_buffer_verified(bh);
641
642 entries = (struct dx_entry *) (((char *)&root->info) +
643 root->info.info_length);
644
645 if (dx_get_limit(entries) != dx_root_limit(dir,
646 root->info.info_length)) {
647 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
648 brelse(bh);
649 *err = ERR_BAD_DX_DIR;
650 goto fail;
651 }
652
653 dxtrace(printk("Look up %x", hash));
654 while (1)
655 {
656 count = dx_get_count(entries);
657 if (!count || count > dx_get_limit(entries)) {
658 ext4_warning(dir->i_sb,
659 "dx entry: no count or count > limit");
660 brelse(bh);
661 *err = ERR_BAD_DX_DIR;
662 goto fail2;
663 }
664
665 p = entries + 1;
666 q = entries + count - 1;
667 while (p <= q)
668 {
669 m = p + (q - p)/2;
670 dxtrace(printk("."));
671 if (dx_get_hash(m) > hash)
672 q = m - 1;
673 else
674 p = m + 1;
675 }
676
677 if (0) // linear search cross check
678 {
679 unsigned n = count - 1;
680 at = entries;
681 while (n--)
682 {
683 dxtrace(printk(","));
684 if (dx_get_hash(++at) > hash)
685 {
686 at--;
687 break;
688 }
689 }
690 assert (at == p - 1);
691 }
692
693 at = p - 1;
694 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
695 frame->bh = bh;
696 frame->entries = entries;
697 frame->at = at;
698 if (!indirect--) return frame;
699 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
700 goto fail2;
701 at = entries = ((struct dx_node *) bh->b_data)->entries;
702
703 if (!buffer_verified(bh) &&
704 !ext4_dx_csum_verify(dir,
705 (struct ext4_dir_entry *)bh->b_data)) {
706 ext4_warning(dir->i_sb, "Node failed checksum");
707 brelse(bh);
708 *err = ERR_BAD_DX_DIR;
709 goto fail;
710 }
711 set_buffer_verified(bh);
712
713 if (dx_get_limit(entries) != dx_node_limit (dir)) {
714 ext4_warning(dir->i_sb,
715 "dx entry: limit != node limit");
716 brelse(bh);
717 *err = ERR_BAD_DX_DIR;
718 goto fail2;
719 }
720 frame++;
721 frame->bh = NULL;
722 }
723fail2:
724 while (frame >= frame_in) {
725 brelse(frame->bh);
726 frame--;
727 }
728fail:
729 if (*err == ERR_BAD_DX_DIR)
730 ext4_warning(dir->i_sb,
731 "Corrupt dir inode %lu, running e2fsck is "
732 "recommended.", dir->i_ino);
733 return NULL;
734}
735
736static void dx_release (struct dx_frame *frames)
737{
738 if (frames[0].bh == NULL)
739 return;
740
741 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
742 brelse(frames[1].bh);
743 brelse(frames[0].bh);
744}
745
746/*
747 * This function increments the frame pointer to search the next leaf
748 * block, and reads in the necessary intervening nodes if the search
749 * should be necessary. Whether or not the search is necessary is
750 * controlled by the hash parameter. If the hash value is even, then
751 * the search is only continued if the next block starts with that
752 * hash value. This is used if we are searching for a specific file.
753 *
754 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
755 *
756 * This function returns 1 if the caller should continue to search,
757 * or 0 if it should not. If there is an error reading one of the
758 * index blocks, it will a negative error code.
759 *
760 * If start_hash is non-null, it will be filled in with the starting
761 * hash of the next page.
762 */
763static int ext4_htree_next_block(struct inode *dir, __u32 hash,
764 struct dx_frame *frame,
765 struct dx_frame *frames,
766 __u32 *start_hash)
767{
768 struct dx_frame *p;
769 struct buffer_head *bh;
770 int err, num_frames = 0;
771 __u32 bhash;
772
773 p = frame;
774 /*
775 * Find the next leaf page by incrementing the frame pointer.
776 * If we run out of entries in the interior node, loop around and
777 * increment pointer in the parent node. When we break out of
778 * this loop, num_frames indicates the number of interior
779 * nodes need to be read.
780 */
781 while (1) {
782 if (++(p->at) < p->entries + dx_get_count(p->entries))
783 break;
784 if (p == frames)
785 return 0;
786 num_frames++;
787 p--;
788 }
789
790 /*
791 * If the hash is 1, then continue only if the next page has a
792 * continuation hash of any value. This is used for readdir
793 * handling. Otherwise, check to see if the hash matches the
794 * desired contiuation hash. If it doesn't, return since
795 * there's no point to read in the successive index pages.
796 */
797 bhash = dx_get_hash(p->at);
798 if (start_hash)
799 *start_hash = bhash;
800 if ((hash & 1) == 0) {
801 if ((bhash & ~1) != hash)
802 return 0;
803 }
804 /*
805 * If the hash is HASH_NB_ALWAYS, we always go to the next
806 * block so no check is necessary
807 */
808 while (num_frames--) {
809 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
810 0, &err)))
811 return err; /* Failure */
812
813 if (!buffer_verified(bh) &&
814 !ext4_dx_csum_verify(dir,
815 (struct ext4_dir_entry *)bh->b_data)) {
816 ext4_warning(dir->i_sb, "Node failed checksum");
817 return -EIO;
818 }
819 set_buffer_verified(bh);
820
821 p++;
822 brelse(p->bh);
823 p->bh = bh;
824 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
825 }
826 return 1;
827}
828
829
830/*
831 * This function fills a red-black tree with information from a
832 * directory block. It returns the number directory entries loaded
833 * into the tree. If there is an error it is returned in err.
834 */
835static int htree_dirblock_to_tree(struct file *dir_file,
836 struct inode *dir, ext4_lblk_t block,
837 struct dx_hash_info *hinfo,
838 __u32 start_hash, __u32 start_minor_hash)
839{
840 struct buffer_head *bh;
841 struct ext4_dir_entry_2 *de, *top;
842 int err, count = 0;
843
844 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
845 (unsigned long)block));
846 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
847 return err;
848
849 if (!buffer_verified(bh) &&
850 !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
851 return -EIO;
852 set_buffer_verified(bh);
853
854 de = (struct ext4_dir_entry_2 *) bh->b_data;
855 top = (struct ext4_dir_entry_2 *) ((char *) de +
856 dir->i_sb->s_blocksize -
857 EXT4_DIR_REC_LEN(0));
858 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
859 if (ext4_check_dir_entry(dir, NULL, de, bh,
860 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
861 + ((char *)de - bh->b_data))) {
862 /* On error, skip the f_pos to the next block. */
863 dir_file->f_pos = (dir_file->f_pos |
864 (dir->i_sb->s_blocksize - 1)) + 1;
865 brelse(bh);
866 return count;
867 }
868 ext4fs_dirhash(de->name, de->name_len, hinfo);
869 if ((hinfo->hash < start_hash) ||
870 ((hinfo->hash == start_hash) &&
871 (hinfo->minor_hash < start_minor_hash)))
872 continue;
873 if (de->inode == 0)
874 continue;
875 if ((err = ext4_htree_store_dirent(dir_file,
876 hinfo->hash, hinfo->minor_hash, de)) != 0) {
877 brelse(bh);
878 return err;
879 }
880 count++;
881 }
882 brelse(bh);
883 return count;
884}
885
886
887/*
888 * This function fills a red-black tree with information from a
889 * directory. We start scanning the directory in hash order, starting
890 * at start_hash and start_minor_hash.
891 *
892 * This function returns the number of entries inserted into the tree,
893 * or a negative error code.
894 */
895int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
896 __u32 start_minor_hash, __u32 *next_hash)
897{
898 struct dx_hash_info hinfo;
899 struct ext4_dir_entry_2 *de;
900 struct dx_frame frames[2], *frame;
901 struct inode *dir;
902 ext4_lblk_t block;
903 int count = 0;
904 int ret, err;
905 __u32 hashval;
906
907 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
908 start_hash, start_minor_hash));
909 dir = dir_file->f_path.dentry->d_inode;
910 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
911 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
912 if (hinfo.hash_version <= DX_HASH_TEA)
913 hinfo.hash_version +=
914 EXT4_SB(dir->i_sb)->s_hash_unsigned;
915 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
916 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
917 start_hash, start_minor_hash);
918 *next_hash = ~0;
919 return count;
920 }
921 hinfo.hash = start_hash;
922 hinfo.minor_hash = 0;
923 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
924 if (!frame)
925 return err;
926
927 /* Add '.' and '..' from the htree header */
928 if (!start_hash && !start_minor_hash) {
929 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
930 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
931 goto errout;
932 count++;
933 }
934 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
935 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
936 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
937 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
938 goto errout;
939 count++;
940 }
941
942 while (1) {
943 block = dx_get_block(frame->at);
944 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
945 start_hash, start_minor_hash);
946 if (ret < 0) {
947 err = ret;
948 goto errout;
949 }
950 count += ret;
951 hashval = ~0;
952 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
953 frame, frames, &hashval);
954 *next_hash = hashval;
955 if (ret < 0) {
956 err = ret;
957 goto errout;
958 }
959 /*
960 * Stop if: (a) there are no more entries, or
961 * (b) we have inserted at least one entry and the
962 * next hash value is not a continuation
963 */
964 if ((ret == 0) ||
965 (count && ((hashval & 1) == 0)))
966 break;
967 }
968 dx_release(frames);
969 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
970 "next hash: %x\n", count, *next_hash));
971 return count;
972errout:
973 dx_release(frames);
974 return (err);
975}
976
977
978/*
979 * Directory block splitting, compacting
980 */
981
982/*
983 * Create map of hash values, offsets, and sizes, stored at end of block.
984 * Returns number of entries mapped.
985 */
986static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
987 struct dx_hash_info *hinfo,
988 struct dx_map_entry *map_tail)
989{
990 int count = 0;
991 char *base = (char *) de;
992 struct dx_hash_info h = *hinfo;
993
994 while ((char *) de < base + blocksize) {
995 if (de->name_len && de->inode) {
996 ext4fs_dirhash(de->name, de->name_len, &h);
997 map_tail--;
998 map_tail->hash = h.hash;
999 map_tail->offs = ((char *) de - base)>>2;
1000 map_tail->size = le16_to_cpu(de->rec_len);
1001 count++;
1002 cond_resched();
1003 }
1004 /* XXX: do we need to check rec_len == 0 case? -Chris */
1005 de = ext4_next_entry(de, blocksize);
1006 }
1007 return count;
1008}
1009
1010/* Sort map by hash value */
1011static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1012{
1013 struct dx_map_entry *p, *q, *top = map + count - 1;
1014 int more;
1015 /* Combsort until bubble sort doesn't suck */
1016 while (count > 2) {
1017 count = count*10/13;
1018 if (count - 9 < 2) /* 9, 10 -> 11 */
1019 count = 11;
1020 for (p = top, q = p - count; q >= map; p--, q--)
1021 if (p->hash < q->hash)
1022 swap(*p, *q);
1023 }
1024 /* Garden variety bubble sort */
1025 do {
1026 more = 0;
1027 q = top;
1028 while (q-- > map) {
1029 if (q[1].hash >= q[0].hash)
1030 continue;
1031 swap(*(q+1), *q);
1032 more = 1;
1033 }
1034 } while(more);
1035}
1036
1037static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1038{
1039 struct dx_entry *entries = frame->entries;
1040 struct dx_entry *old = frame->at, *new = old + 1;
1041 int count = dx_get_count(entries);
1042
1043 assert(count < dx_get_limit(entries));
1044 assert(old < entries + count);
1045 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1046 dx_set_hash(new, hash);
1047 dx_set_block(new, block);
1048 dx_set_count(entries, count + 1);
1049}
1050
1051static void ext4_update_dx_flag(struct inode *inode)
1052{
1053 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
1054 EXT4_FEATURE_COMPAT_DIR_INDEX))
1055 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
1056}
1057
1058/*
1059 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1060 *
1061 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1062 * `de != NULL' is guaranteed by caller.
1063 */
1064static inline int ext4_match (int len, const char * const name,
1065 struct ext4_dir_entry_2 * de)
1066{
1067 if (len != de->name_len)
1068 return 0;
1069 if (!de->inode)
1070 return 0;
1071 return !memcmp(name, de->name, len);
1072}
1073
1074/*
1075 * Returns 0 if not found, -1 on failure, and 1 on success
1076 */
1077static inline int search_dirblock(struct buffer_head *bh,
1078 struct inode *dir,
1079 const struct qstr *d_name,
1080 unsigned int offset,
1081 struct ext4_dir_entry_2 ** res_dir)
1082{
1083 struct ext4_dir_entry_2 * de;
1084 char * dlimit;
1085 int de_len;
1086 const char *name = d_name->name;
1087 int namelen = d_name->len;
1088
1089 de = (struct ext4_dir_entry_2 *) bh->b_data;
1090 dlimit = bh->b_data + dir->i_sb->s_blocksize;
1091 while ((char *) de < dlimit) {
1092 /* this code is executed quadratically often */
1093 /* do minimal checking `by hand' */
1094
1095 if ((char *) de + namelen <= dlimit &&
1096 ext4_match (namelen, name, de)) {
1097 /* found a match - just to be sure, do a full check */
1098 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1099 return -1;
1100 *res_dir = de;
1101 return 1;
1102 }
1103 /* prevent looping on a bad block */
1104 de_len = ext4_rec_len_from_disk(de->rec_len,
1105 dir->i_sb->s_blocksize);
1106 if (de_len <= 0)
1107 return -1;
1108 offset += de_len;
1109 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1110 }
1111 return 0;
1112}
1113
1114
1115/*
1116 * ext4_find_entry()
1117 *
1118 * finds an entry in the specified directory with the wanted name. It
1119 * returns the cache buffer in which the entry was found, and the entry
1120 * itself (as a parameter - res_dir). It does NOT read the inode of the
1121 * entry - you'll have to do that yourself if you want to.
1122 *
1123 * The returned buffer_head has ->b_count elevated. The caller is expected
1124 * to brelse() it when appropriate.
1125 */
1126static struct buffer_head * ext4_find_entry (struct inode *dir,
1127 const struct qstr *d_name,
1128 struct ext4_dir_entry_2 ** res_dir)
1129{
1130 struct super_block *sb;
1131 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1132 struct buffer_head *bh, *ret = NULL;
1133 ext4_lblk_t start, block, b;
1134 const u8 *name = d_name->name;
1135 int ra_max = 0; /* Number of bh's in the readahead
1136 buffer, bh_use[] */
1137 int ra_ptr = 0; /* Current index into readahead
1138 buffer */
1139 int num = 0;
1140 ext4_lblk_t nblocks;
1141 int i, err;
1142 int namelen;
1143
1144 *res_dir = NULL;
1145 sb = dir->i_sb;
1146 namelen = d_name->len;
1147 if (namelen > EXT4_NAME_LEN)
1148 return NULL;
1149 if ((namelen <= 2) && (name[0] == '.') &&
1150 (name[1] == '.' || name[1] == '\0')) {
1151 /*
1152 * "." or ".." will only be in the first block
1153 * NFS may look up ".."; "." should be handled by the VFS
1154 */
1155 block = start = 0;
1156 nblocks = 1;
1157 goto restart;
1158 }
1159 if (is_dx(dir)) {
1160 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1161 /*
1162 * On success, or if the error was file not found,
1163 * return. Otherwise, fall back to doing a search the
1164 * old fashioned way.
1165 */
1166 if (bh || (err != ERR_BAD_DX_DIR))
1167 return bh;
1168 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1169 "falling back\n"));
1170 }
1171 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1172 start = EXT4_I(dir)->i_dir_start_lookup;
1173 if (start >= nblocks)
1174 start = 0;
1175 block = start;
1176restart:
1177 do {
1178 /*
1179 * We deal with the read-ahead logic here.
1180 */
1181 if (ra_ptr >= ra_max) {
1182 /* Refill the readahead buffer */
1183 ra_ptr = 0;
1184 b = block;
1185 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1186 /*
1187 * Terminate if we reach the end of the
1188 * directory and must wrap, or if our
1189 * search has finished at this block.
1190 */
1191 if (b >= nblocks || (num && block == start)) {
1192 bh_use[ra_max] = NULL;
1193 break;
1194 }
1195 num++;
1196 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1197 bh_use[ra_max] = bh;
1198 if (bh)
1199 ll_rw_block(READ | REQ_META | REQ_PRIO,
1200 1, &bh);
1201 }
1202 }
1203 if ((bh = bh_use[ra_ptr++]) == NULL)
1204 goto next;
1205 wait_on_buffer(bh);
1206 if (!buffer_uptodate(bh)) {
1207 /* read error, skip block & hope for the best */
1208 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1209 (unsigned long) block);
1210 brelse(bh);
1211 goto next;
1212 }
1213 if (!buffer_verified(bh) &&
1214 !ext4_dirent_csum_verify(dir,
1215 (struct ext4_dir_entry *)bh->b_data)) {
1216 EXT4_ERROR_INODE(dir, "checksumming directory "
1217 "block %lu", (unsigned long)block);
1218 brelse(bh);
1219 goto next;
1220 }
1221 set_buffer_verified(bh);
1222 i = search_dirblock(bh, dir, d_name,
1223 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1224 if (i == 1) {
1225 EXT4_I(dir)->i_dir_start_lookup = block;
1226 ret = bh;
1227 goto cleanup_and_exit;
1228 } else {
1229 brelse(bh);
1230 if (i < 0)
1231 goto cleanup_and_exit;
1232 }
1233 next:
1234 if (++block >= nblocks)
1235 block = 0;
1236 } while (block != start);
1237
1238 /*
1239 * If the directory has grown while we were searching, then
1240 * search the last part of the directory before giving up.
1241 */
1242 block = nblocks;
1243 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1244 if (block < nblocks) {
1245 start = 0;
1246 goto restart;
1247 }
1248
1249cleanup_and_exit:
1250 /* Clean up the read-ahead blocks */
1251 for (; ra_ptr < ra_max; ra_ptr++)
1252 brelse(bh_use[ra_ptr]);
1253 return ret;
1254}
1255
1256static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1257 struct ext4_dir_entry_2 **res_dir, int *err)
1258{
1259 struct super_block * sb = dir->i_sb;
1260 struct dx_hash_info hinfo;
1261 struct dx_frame frames[2], *frame;
1262 struct buffer_head *bh;
1263 ext4_lblk_t block;
1264 int retval;
1265
1266 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1267 return NULL;
1268 do {
1269 block = dx_get_block(frame->at);
1270 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
1271 goto errout;
1272
1273 if (!buffer_verified(bh) &&
1274 !ext4_dirent_csum_verify(dir,
1275 (struct ext4_dir_entry *)bh->b_data)) {
1276 EXT4_ERROR_INODE(dir, "checksumming directory "
1277 "block %lu", (unsigned long)block);
1278 brelse(bh);
1279 *err = -EIO;
1280 goto errout;
1281 }
1282 set_buffer_verified(bh);
1283 retval = search_dirblock(bh, dir, d_name,
1284 block << EXT4_BLOCK_SIZE_BITS(sb),
1285 res_dir);
1286 if (retval == 1) { /* Success! */
1287 dx_release(frames);
1288 return bh;
1289 }
1290 brelse(bh);
1291 if (retval == -1) {
1292 *err = ERR_BAD_DX_DIR;
1293 goto errout;
1294 }
1295
1296 /* Check to see if we should continue to search */
1297 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1298 frames, NULL);
1299 if (retval < 0) {
1300 ext4_warning(sb,
1301 "error reading index page in directory #%lu",
1302 dir->i_ino);
1303 *err = retval;
1304 goto errout;
1305 }
1306 } while (retval == 1);
1307
1308 *err = -ENOENT;
1309errout:
1310 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1311 dx_release (frames);
1312 return NULL;
1313}
1314
1315static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1316{
1317 struct inode *inode;
1318 struct ext4_dir_entry_2 *de;
1319 struct buffer_head *bh;
1320
1321 if (dentry->d_name.len > EXT4_NAME_LEN)
1322 return ERR_PTR(-ENAMETOOLONG);
1323
1324 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1325 inode = NULL;
1326 if (bh) {
1327 __u32 ino = le32_to_cpu(de->inode);
1328 brelse(bh);
1329 if (!ext4_valid_inum(dir->i_sb, ino)) {
1330 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1331 return ERR_PTR(-EIO);
1332 }
1333 if (unlikely(ino == dir->i_ino)) {
1334 EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1335 dentry->d_name.len,
1336 dentry->d_name.name);
1337 return ERR_PTR(-EIO);
1338 }
1339 inode = ext4_iget(dir->i_sb, ino);
1340 if (inode == ERR_PTR(-ESTALE)) {
1341 EXT4_ERROR_INODE(dir,
1342 "deleted inode referenced: %u",
1343 ino);
1344 return ERR_PTR(-EIO);
1345 }
1346 }
1347 return d_splice_alias(inode, dentry);
1348}
1349
1350
1351struct dentry *ext4_get_parent(struct dentry *child)
1352{
1353 __u32 ino;
1354 static const struct qstr dotdot = QSTR_INIT("..", 2);
1355 struct ext4_dir_entry_2 * de;
1356 struct buffer_head *bh;
1357
1358 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1359 if (!bh)
1360 return ERR_PTR(-ENOENT);
1361 ino = le32_to_cpu(de->inode);
1362 brelse(bh);
1363
1364 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1365 EXT4_ERROR_INODE(child->d_inode,
1366 "bad parent inode number: %u", ino);
1367 return ERR_PTR(-EIO);
1368 }
1369
1370 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1371}
1372
1373#define S_SHIFT 12
1374static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1375 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1376 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1377 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1378 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1379 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1380 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1381 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1382};
1383
1384static inline void ext4_set_de_type(struct super_block *sb,
1385 struct ext4_dir_entry_2 *de,
1386 umode_t mode) {
1387 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1388 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1389}
1390
1391/*
1392 * Move count entries from end of map between two memory locations.
1393 * Returns pointer to last entry moved.
1394 */
1395static struct ext4_dir_entry_2 *
1396dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1397 unsigned blocksize)
1398{
1399 unsigned rec_len = 0;
1400
1401 while (count--) {
1402 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1403 (from + (map->offs<<2));
1404 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1405 memcpy (to, de, rec_len);
1406 ((struct ext4_dir_entry_2 *) to)->rec_len =
1407 ext4_rec_len_to_disk(rec_len, blocksize);
1408 de->inode = 0;
1409 map++;
1410 to += rec_len;
1411 }
1412 return (struct ext4_dir_entry_2 *) (to - rec_len);
1413}
1414
1415/*
1416 * Compact each dir entry in the range to the minimal rec_len.
1417 * Returns pointer to last entry in range.
1418 */
1419static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1420{
1421 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1422 unsigned rec_len = 0;
1423
1424 prev = to = de;
1425 while ((char*)de < base + blocksize) {
1426 next = ext4_next_entry(de, blocksize);
1427 if (de->inode && de->name_len) {
1428 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1429 if (de > to)
1430 memmove(to, de, rec_len);
1431 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1432 prev = to;
1433 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1434 }
1435 de = next;
1436 }
1437 return prev;
1438}
1439
1440/*
1441 * Split a full leaf block to make room for a new dir entry.
1442 * Allocate a new block, and move entries so that they are approx. equally full.
1443 * Returns pointer to de in block into which the new entry will be inserted.
1444 */
1445static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1446 struct buffer_head **bh,struct dx_frame *frame,
1447 struct dx_hash_info *hinfo, int *error)
1448{
1449 unsigned blocksize = dir->i_sb->s_blocksize;
1450 unsigned count, continued;
1451 struct buffer_head *bh2;
1452 ext4_lblk_t newblock;
1453 u32 hash2;
1454 struct dx_map_entry *map;
1455 char *data1 = (*bh)->b_data, *data2;
1456 unsigned split, move, size;
1457 struct ext4_dir_entry_2 *de = NULL, *de2;
1458 struct ext4_dir_entry_tail *t;
1459 int csum_size = 0;
1460 int err = 0, i;
1461
1462 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1463 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1464 csum_size = sizeof(struct ext4_dir_entry_tail);
1465
1466 bh2 = ext4_append (handle, dir, &newblock, &err);
1467 if (!(bh2)) {
1468 brelse(*bh);
1469 *bh = NULL;
1470 goto errout;
1471 }
1472
1473 BUFFER_TRACE(*bh, "get_write_access");
1474 err = ext4_journal_get_write_access(handle, *bh);
1475 if (err)
1476 goto journal_error;
1477
1478 BUFFER_TRACE(frame->bh, "get_write_access");
1479 err = ext4_journal_get_write_access(handle, frame->bh);
1480 if (err)
1481 goto journal_error;
1482
1483 data2 = bh2->b_data;
1484
1485 /* create map in the end of data2 block */
1486 map = (struct dx_map_entry *) (data2 + blocksize);
1487 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1488 blocksize, hinfo, map);
1489 map -= count;
1490 dx_sort_map(map, count);
1491 /* Split the existing block in the middle, size-wise */
1492 size = 0;
1493 move = 0;
1494 for (i = count-1; i >= 0; i--) {
1495 /* is more than half of this entry in 2nd half of the block? */
1496 if (size + map[i].size/2 > blocksize/2)
1497 break;
1498 size += map[i].size;
1499 move++;
1500 }
1501 /* map index at which we will split */
1502 split = count - move;
1503 hash2 = map[split].hash;
1504 continued = hash2 == map[split - 1].hash;
1505 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1506 (unsigned long)dx_get_block(frame->at),
1507 hash2, split, count-split));
1508
1509 /* Fancy dance to stay within two buffers */
1510 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1511 de = dx_pack_dirents(data1, blocksize);
1512 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1513 (char *) de,
1514 blocksize);
1515 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1516 (char *) de2,
1517 blocksize);
1518 if (csum_size) {
1519 t = EXT4_DIRENT_TAIL(data2, blocksize);
1520 initialize_dirent_tail(t, blocksize);
1521
1522 t = EXT4_DIRENT_TAIL(data1, blocksize);
1523 initialize_dirent_tail(t, blocksize);
1524 }
1525
1526 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1527 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1528
1529 /* Which block gets the new entry? */
1530 if (hinfo->hash >= hash2)
1531 {
1532 swap(*bh, bh2);
1533 de = de2;
1534 }
1535 dx_insert_block(frame, hash2 + continued, newblock);
1536 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1537 if (err)
1538 goto journal_error;
1539 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1540 if (err)
1541 goto journal_error;
1542 brelse(bh2);
1543 dxtrace(dx_show_index("frame", frame->entries));
1544 return de;
1545
1546journal_error:
1547 brelse(*bh);
1548 brelse(bh2);
1549 *bh = NULL;
1550 ext4_std_error(dir->i_sb, err);
1551errout:
1552 *error = err;
1553 return NULL;
1554}
1555
1556/*
1557 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1558 * it points to a directory entry which is guaranteed to be large
1559 * enough for new directory entry. If de is NULL, then
1560 * add_dirent_to_buf will attempt search the directory block for
1561 * space. It will return -ENOSPC if no space is available, and -EIO
1562 * and -EEXIST if directory entry already exists.
1563 */
1564static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1565 struct inode *inode, struct ext4_dir_entry_2 *de,
1566 struct buffer_head *bh)
1567{
1568 struct inode *dir = dentry->d_parent->d_inode;
1569 const char *name = dentry->d_name.name;
1570 int namelen = dentry->d_name.len;
1571 unsigned int offset = 0;
1572 unsigned int blocksize = dir->i_sb->s_blocksize;
1573 unsigned short reclen;
1574 int nlen, rlen, err;
1575 char *top;
1576 int csum_size = 0;
1577
1578 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1579 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1580 csum_size = sizeof(struct ext4_dir_entry_tail);
1581
1582 reclen = EXT4_DIR_REC_LEN(namelen);
1583 if (!de) {
1584 de = (struct ext4_dir_entry_2 *)bh->b_data;
1585 top = bh->b_data + (blocksize - csum_size) - reclen;
1586 while ((char *) de <= top) {
1587 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1588 return -EIO;
1589 if (ext4_match(namelen, name, de))
1590 return -EEXIST;
1591 nlen = EXT4_DIR_REC_LEN(de->name_len);
1592 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1593 if ((de->inode? rlen - nlen: rlen) >= reclen)
1594 break;
1595 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1596 offset += rlen;
1597 }
1598 if ((char *) de > top)
1599 return -ENOSPC;
1600 }
1601 BUFFER_TRACE(bh, "get_write_access");
1602 err = ext4_journal_get_write_access(handle, bh);
1603 if (err) {
1604 ext4_std_error(dir->i_sb, err);
1605 return err;
1606 }
1607
1608 /* By now the buffer is marked for journaling */
1609 nlen = EXT4_DIR_REC_LEN(de->name_len);
1610 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1611 if (de->inode) {
1612 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1613 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1614 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1615 de = de1;
1616 }
1617 de->file_type = EXT4_FT_UNKNOWN;
1618 de->inode = cpu_to_le32(inode->i_ino);
1619 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1620 de->name_len = namelen;
1621 memcpy(de->name, name, namelen);
1622 /*
1623 * XXX shouldn't update any times until successful
1624 * completion of syscall, but too many callers depend
1625 * on this.
1626 *
1627 * XXX similarly, too many callers depend on
1628 * ext4_new_inode() setting the times, but error
1629 * recovery deletes the inode, so the worst that can
1630 * happen is that the times are slightly out of date
1631 * and/or different from the directory change time.
1632 */
1633 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1634 ext4_update_dx_flag(dir);
1635 dir->i_version++;
1636 ext4_mark_inode_dirty(handle, dir);
1637 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1638 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1639 if (err)
1640 ext4_std_error(dir->i_sb, err);
1641 return 0;
1642}
1643
1644/*
1645 * This converts a one block unindexed directory to a 3 block indexed
1646 * directory, and adds the dentry to the indexed directory.
1647 */
1648static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1649 struct inode *inode, struct buffer_head *bh)
1650{
1651 struct inode *dir = dentry->d_parent->d_inode;
1652 const char *name = dentry->d_name.name;
1653 int namelen = dentry->d_name.len;
1654 struct buffer_head *bh2;
1655 struct dx_root *root;
1656 struct dx_frame frames[2], *frame;
1657 struct dx_entry *entries;
1658 struct ext4_dir_entry_2 *de, *de2;
1659 struct ext4_dir_entry_tail *t;
1660 char *data1, *top;
1661 unsigned len;
1662 int retval;
1663 unsigned blocksize;
1664 struct dx_hash_info hinfo;
1665 ext4_lblk_t block;
1666 struct fake_dirent *fde;
1667 int csum_size = 0;
1668
1669 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1670 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1671 csum_size = sizeof(struct ext4_dir_entry_tail);
1672
1673 blocksize = dir->i_sb->s_blocksize;
1674 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1675 retval = ext4_journal_get_write_access(handle, bh);
1676 if (retval) {
1677 ext4_std_error(dir->i_sb, retval);
1678 brelse(bh);
1679 return retval;
1680 }
1681 root = (struct dx_root *) bh->b_data;
1682
1683 /* The 0th block becomes the root, move the dirents out */
1684 fde = &root->dotdot;
1685 de = (struct ext4_dir_entry_2 *)((char *)fde +
1686 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1687 if ((char *) de >= (((char *) root) + blocksize)) {
1688 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1689 brelse(bh);
1690 return -EIO;
1691 }
1692 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1693
1694 /* Allocate new block for the 0th block's dirents */
1695 bh2 = ext4_append(handle, dir, &block, &retval);
1696 if (!(bh2)) {
1697 brelse(bh);
1698 return retval;
1699 }
1700 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1701 data1 = bh2->b_data;
1702
1703 memcpy (data1, de, len);
1704 de = (struct ext4_dir_entry_2 *) data1;
1705 top = data1 + len;
1706 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1707 de = de2;
1708 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1709 (char *) de,
1710 blocksize);
1711
1712 if (csum_size) {
1713 t = EXT4_DIRENT_TAIL(data1, blocksize);
1714 initialize_dirent_tail(t, blocksize);
1715 }
1716
1717 /* Initialize the root; the dot dirents already exist */
1718 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1719 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1720 blocksize);
1721 memset (&root->info, 0, sizeof(root->info));
1722 root->info.info_length = sizeof(root->info);
1723 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1724 entries = root->entries;
1725 dx_set_block(entries, 1);
1726 dx_set_count(entries, 1);
1727 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1728
1729 /* Initialize as for dx_probe */
1730 hinfo.hash_version = root->info.hash_version;
1731 if (hinfo.hash_version <= DX_HASH_TEA)
1732 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1733 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1734 ext4fs_dirhash(name, namelen, &hinfo);
1735 frame = frames;
1736 frame->entries = entries;
1737 frame->at = entries;
1738 frame->bh = bh;
1739 bh = bh2;
1740
1741 ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1742 ext4_handle_dirty_dirent_node(handle, dir, bh);
1743
1744 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1745 if (!de) {
1746 /*
1747 * Even if the block split failed, we have to properly write
1748 * out all the changes we did so far. Otherwise we can end up
1749 * with corrupted filesystem.
1750 */
1751 ext4_mark_inode_dirty(handle, dir);
1752 dx_release(frames);
1753 return retval;
1754 }
1755 dx_release(frames);
1756
1757 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1758 brelse(bh);
1759 return retval;
1760}
1761
1762/*
1763 * ext4_add_entry()
1764 *
1765 * adds a file entry to the specified directory, using the same
1766 * semantics as ext4_find_entry(). It returns NULL if it failed.
1767 *
1768 * NOTE!! The inode part of 'de' is left at 0 - which means you
1769 * may not sleep between calling this and putting something into
1770 * the entry, as someone else might have used it while you slept.
1771 */
1772static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1773 struct inode *inode)
1774{
1775 struct inode *dir = dentry->d_parent->d_inode;
1776 struct buffer_head *bh;
1777 struct ext4_dir_entry_2 *de;
1778 struct ext4_dir_entry_tail *t;
1779 struct super_block *sb;
1780 int retval;
1781 int dx_fallback=0;
1782 unsigned blocksize;
1783 ext4_lblk_t block, blocks;
1784 int csum_size = 0;
1785
1786 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1787 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1788 csum_size = sizeof(struct ext4_dir_entry_tail);
1789
1790 sb = dir->i_sb;
1791 blocksize = sb->s_blocksize;
1792 if (!dentry->d_name.len)
1793 return -EINVAL;
1794 if (is_dx(dir)) {
1795 retval = ext4_dx_add_entry(handle, dentry, inode);
1796 if (!retval || (retval != ERR_BAD_DX_DIR))
1797 return retval;
1798 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1799 dx_fallback++;
1800 ext4_mark_inode_dirty(handle, dir);
1801 }
1802 blocks = dir->i_size >> sb->s_blocksize_bits;
1803 for (block = 0; block < blocks; block++) {
1804 bh = ext4_bread(handle, dir, block, 0, &retval);
1805 if(!bh)
1806 return retval;
1807 if (!buffer_verified(bh) &&
1808 !ext4_dirent_csum_verify(dir,
1809 (struct ext4_dir_entry *)bh->b_data))
1810 return -EIO;
1811 set_buffer_verified(bh);
1812 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1813 if (retval != -ENOSPC) {
1814 brelse(bh);
1815 return retval;
1816 }
1817
1818 if (blocks == 1 && !dx_fallback &&
1819 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1820 return make_indexed_dir(handle, dentry, inode, bh);
1821 brelse(bh);
1822 }
1823 bh = ext4_append(handle, dir, &block, &retval);
1824 if (!bh)
1825 return retval;
1826 de = (struct ext4_dir_entry_2 *) bh->b_data;
1827 de->inode = 0;
1828 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1829
1830 if (csum_size) {
1831 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1832 initialize_dirent_tail(t, blocksize);
1833 }
1834
1835 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1836 brelse(bh);
1837 if (retval == 0)
1838 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1839 return retval;
1840}
1841
1842/*
1843 * Returns 0 for success, or a negative error value
1844 */
1845static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1846 struct inode *inode)
1847{
1848 struct dx_frame frames[2], *frame;
1849 struct dx_entry *entries, *at;
1850 struct dx_hash_info hinfo;
1851 struct buffer_head *bh;
1852 struct inode *dir = dentry->d_parent->d_inode;
1853 struct super_block *sb = dir->i_sb;
1854 struct ext4_dir_entry_2 *de;
1855 int err;
1856
1857 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1858 if (!frame)
1859 return err;
1860 entries = frame->entries;
1861 at = frame->at;
1862
1863 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1864 goto cleanup;
1865
1866 if (!buffer_verified(bh) &&
1867 !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
1868 goto journal_error;
1869 set_buffer_verified(bh);
1870
1871 BUFFER_TRACE(bh, "get_write_access");
1872 err = ext4_journal_get_write_access(handle, bh);
1873 if (err)
1874 goto journal_error;
1875
1876 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1877 if (err != -ENOSPC)
1878 goto cleanup;
1879
1880 /* Block full, should compress but for now just split */
1881 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1882 dx_get_count(entries), dx_get_limit(entries)));
1883 /* Need to split index? */
1884 if (dx_get_count(entries) == dx_get_limit(entries)) {
1885 ext4_lblk_t newblock;
1886 unsigned icount = dx_get_count(entries);
1887 int levels = frame - frames;
1888 struct dx_entry *entries2;
1889 struct dx_node *node2;
1890 struct buffer_head *bh2;
1891
1892 if (levels && (dx_get_count(frames->entries) ==
1893 dx_get_limit(frames->entries))) {
1894 ext4_warning(sb, "Directory index full!");
1895 err = -ENOSPC;
1896 goto cleanup;
1897 }
1898 bh2 = ext4_append (handle, dir, &newblock, &err);
1899 if (!(bh2))
1900 goto cleanup;
1901 node2 = (struct dx_node *)(bh2->b_data);
1902 entries2 = node2->entries;
1903 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1904 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1905 sb->s_blocksize);
1906 BUFFER_TRACE(frame->bh, "get_write_access");
1907 err = ext4_journal_get_write_access(handle, frame->bh);
1908 if (err)
1909 goto journal_error;
1910 if (levels) {
1911 unsigned icount1 = icount/2, icount2 = icount - icount1;
1912 unsigned hash2 = dx_get_hash(entries + icount1);
1913 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1914 icount1, icount2));
1915
1916 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1917 err = ext4_journal_get_write_access(handle,
1918 frames[0].bh);
1919 if (err)
1920 goto journal_error;
1921
1922 memcpy((char *) entries2, (char *) (entries + icount1),
1923 icount2 * sizeof(struct dx_entry));
1924 dx_set_count(entries, icount1);
1925 dx_set_count(entries2, icount2);
1926 dx_set_limit(entries2, dx_node_limit(dir));
1927
1928 /* Which index block gets the new entry? */
1929 if (at - entries >= icount1) {
1930 frame->at = at = at - entries - icount1 + entries2;
1931 frame->entries = entries = entries2;
1932 swap(frame->bh, bh2);
1933 }
1934 dx_insert_block(frames + 0, hash2, newblock);
1935 dxtrace(dx_show_index("node", frames[1].entries));
1936 dxtrace(dx_show_index("node",
1937 ((struct dx_node *) bh2->b_data)->entries));
1938 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
1939 if (err)
1940 goto journal_error;
1941 brelse (bh2);
1942 } else {
1943 dxtrace(printk(KERN_DEBUG
1944 "Creating second level index...\n"));
1945 memcpy((char *) entries2, (char *) entries,
1946 icount * sizeof(struct dx_entry));
1947 dx_set_limit(entries2, dx_node_limit(dir));
1948
1949 /* Set up root */
1950 dx_set_count(entries, 1);
1951 dx_set_block(entries + 0, newblock);
1952 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1953
1954 /* Add new access path frame */
1955 frame = frames + 1;
1956 frame->at = at = at - entries + entries2;
1957 frame->entries = entries = entries2;
1958 frame->bh = bh2;
1959 err = ext4_journal_get_write_access(handle,
1960 frame->bh);
1961 if (err)
1962 goto journal_error;
1963 }
1964 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
1965 if (err) {
1966 ext4_std_error(inode->i_sb, err);
1967 goto cleanup;
1968 }
1969 }
1970 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1971 if (!de)
1972 goto cleanup;
1973 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1974 goto cleanup;
1975
1976journal_error:
1977 ext4_std_error(dir->i_sb, err);
1978cleanup:
1979 if (bh)
1980 brelse(bh);
1981 dx_release(frames);
1982 return err;
1983}
1984
1985/*
1986 * ext4_delete_entry deletes a directory entry by merging it with the
1987 * previous entry
1988 */
1989static int ext4_delete_entry(handle_t *handle,
1990 struct inode *dir,
1991 struct ext4_dir_entry_2 *de_del,
1992 struct buffer_head *bh)
1993{
1994 struct ext4_dir_entry_2 *de, *pde;
1995 unsigned int blocksize = dir->i_sb->s_blocksize;
1996 int csum_size = 0;
1997 int i, err;
1998
1999 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2000 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2001 csum_size = sizeof(struct ext4_dir_entry_tail);
2002
2003 i = 0;
2004 pde = NULL;
2005 de = (struct ext4_dir_entry_2 *) bh->b_data;
2006 while (i < bh->b_size - csum_size) {
2007 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
2008 return -EIO;
2009 if (de == de_del) {
2010 BUFFER_TRACE(bh, "get_write_access");
2011 err = ext4_journal_get_write_access(handle, bh);
2012 if (unlikely(err)) {
2013 ext4_std_error(dir->i_sb, err);
2014 return err;
2015 }
2016 if (pde)
2017 pde->rec_len = ext4_rec_len_to_disk(
2018 ext4_rec_len_from_disk(pde->rec_len,
2019 blocksize) +
2020 ext4_rec_len_from_disk(de->rec_len,
2021 blocksize),
2022 blocksize);
2023 else
2024 de->inode = 0;
2025 dir->i_version++;
2026 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2027 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2028 if (unlikely(err)) {
2029 ext4_std_error(dir->i_sb, err);
2030 return err;
2031 }
2032 return 0;
2033 }
2034 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2035 pde = de;
2036 de = ext4_next_entry(de, blocksize);
2037 }
2038 return -ENOENT;
2039}
2040
2041/*
2042 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2043 * since this indicates that nlinks count was previously 1.
2044 */
2045static void ext4_inc_count(handle_t *handle, struct inode *inode)
2046{
2047 inc_nlink(inode);
2048 if (is_dx(inode) && inode->i_nlink > 1) {
2049 /* limit is 16-bit i_links_count */
2050 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2051 set_nlink(inode, 1);
2052 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2053 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2054 }
2055 }
2056}
2057
2058/*
2059 * If a directory had nlink == 1, then we should let it be 1. This indicates
2060 * directory has >EXT4_LINK_MAX subdirs.
2061 */
2062static void ext4_dec_count(handle_t *handle, struct inode *inode)
2063{
2064 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2065 drop_nlink(inode);
2066}
2067
2068
2069static int ext4_add_nondir(handle_t *handle,
2070 struct dentry *dentry, struct inode *inode)
2071{
2072 int err = ext4_add_entry(handle, dentry, inode);
2073 if (!err) {
2074 ext4_mark_inode_dirty(handle, inode);
2075 d_instantiate(dentry, inode);
2076 unlock_new_inode(inode);
2077 return 0;
2078 }
2079 drop_nlink(inode);
2080 unlock_new_inode(inode);
2081 iput(inode);
2082 return err;
2083}
2084
2085/*
2086 * By the time this is called, we already have created
2087 * the directory cache entry for the new file, but it
2088 * is so far negative - it has no inode.
2089 *
2090 * If the create succeeds, we fill in the inode information
2091 * with d_instantiate().
2092 */
2093static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2094 struct nameidata *nd)
2095{
2096 handle_t *handle;
2097 struct inode *inode;
2098 int err, retries = 0;
2099
2100 dquot_initialize(dir);
2101
2102retry:
2103 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2104 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2105 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2106 if (IS_ERR(handle))
2107 return PTR_ERR(handle);
2108
2109 if (IS_DIRSYNC(dir))
2110 ext4_handle_sync(handle);
2111
2112 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2113 err = PTR_ERR(inode);
2114 if (!IS_ERR(inode)) {
2115 inode->i_op = &ext4_file_inode_operations;
2116 inode->i_fop = &ext4_file_operations;
2117 ext4_set_aops(inode);
2118 err = ext4_add_nondir(handle, dentry, inode);
2119 }
2120 ext4_journal_stop(handle);
2121 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2122 goto retry;
2123 return err;
2124}
2125
2126static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2127 umode_t mode, dev_t rdev)
2128{
2129 handle_t *handle;
2130 struct inode *inode;
2131 int err, retries = 0;
2132
2133 if (!new_valid_dev(rdev))
2134 return -EINVAL;
2135
2136 dquot_initialize(dir);
2137
2138retry:
2139 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2140 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2141 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2142 if (IS_ERR(handle))
2143 return PTR_ERR(handle);
2144
2145 if (IS_DIRSYNC(dir))
2146 ext4_handle_sync(handle);
2147
2148 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2149 err = PTR_ERR(inode);
2150 if (!IS_ERR(inode)) {
2151 init_special_inode(inode, inode->i_mode, rdev);
2152#ifdef CONFIG_EXT4_FS_XATTR
2153 inode->i_op = &ext4_special_inode_operations;
2154#endif
2155 err = ext4_add_nondir(handle, dentry, inode);
2156 }
2157 ext4_journal_stop(handle);
2158 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2159 goto retry;
2160 return err;
2161}
2162
2163static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2164{
2165 handle_t *handle;
2166 struct inode *inode;
2167 struct buffer_head *dir_block = NULL;
2168 struct ext4_dir_entry_2 *de;
2169 struct ext4_dir_entry_tail *t;
2170 unsigned int blocksize = dir->i_sb->s_blocksize;
2171 int csum_size = 0;
2172 int err, retries = 0;
2173
2174 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2175 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2176 csum_size = sizeof(struct ext4_dir_entry_tail);
2177
2178 if (EXT4_DIR_LINK_MAX(dir))
2179 return -EMLINK;
2180
2181 dquot_initialize(dir);
2182
2183retry:
2184 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2185 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2186 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2187 if (IS_ERR(handle))
2188 return PTR_ERR(handle);
2189
2190 if (IS_DIRSYNC(dir))
2191 ext4_handle_sync(handle);
2192
2193 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
2194 &dentry->d_name, 0, NULL);
2195 err = PTR_ERR(inode);
2196 if (IS_ERR(inode))
2197 goto out_stop;
2198
2199 inode->i_op = &ext4_dir_inode_operations;
2200 inode->i_fop = &ext4_dir_operations;
2201 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
2202 dir_block = ext4_bread(handle, inode, 0, 1, &err);
2203 if (!dir_block)
2204 goto out_clear_inode;
2205 BUFFER_TRACE(dir_block, "get_write_access");
2206 err = ext4_journal_get_write_access(handle, dir_block);
2207 if (err)
2208 goto out_clear_inode;
2209 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
2210 de->inode = cpu_to_le32(inode->i_ino);
2211 de->name_len = 1;
2212 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2213 blocksize);
2214 strcpy(de->name, ".");
2215 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
2216 de = ext4_next_entry(de, blocksize);
2217 de->inode = cpu_to_le32(dir->i_ino);
2218 de->rec_len = ext4_rec_len_to_disk(blocksize -
2219 (csum_size + EXT4_DIR_REC_LEN(1)),
2220 blocksize);
2221 de->name_len = 2;
2222 strcpy(de->name, "..");
2223 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
2224 set_nlink(inode, 2);
2225
2226 if (csum_size) {
2227 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2228 initialize_dirent_tail(t, blocksize);
2229 }
2230
2231 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2232 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2233 if (err)
2234 goto out_clear_inode;
2235 set_buffer_verified(dir_block);
2236 err = ext4_mark_inode_dirty(handle, inode);
2237 if (!err)
2238 err = ext4_add_entry(handle, dentry, inode);
2239 if (err) {
2240out_clear_inode:
2241 clear_nlink(inode);
2242 unlock_new_inode(inode);
2243 ext4_mark_inode_dirty(handle, inode);
2244 iput(inode);
2245 goto out_stop;
2246 }
2247 ext4_inc_count(handle, dir);
2248 ext4_update_dx_flag(dir);
2249 err = ext4_mark_inode_dirty(handle, dir);
2250 if (err)
2251 goto out_clear_inode;
2252 d_instantiate(dentry, inode);
2253 unlock_new_inode(inode);
2254out_stop:
2255 brelse(dir_block);
2256 ext4_journal_stop(handle);
2257 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2258 goto retry;
2259 return err;
2260}
2261
2262/*
2263 * routine to check that the specified directory is empty (for rmdir)
2264 */
2265static int empty_dir(struct inode *inode)
2266{
2267 unsigned int offset;
2268 struct buffer_head *bh;
2269 struct ext4_dir_entry_2 *de, *de1;
2270 struct super_block *sb;
2271 int err = 0;
2272
2273 sb = inode->i_sb;
2274 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
2275 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
2276 if (err)
2277 EXT4_ERROR_INODE(inode,
2278 "error %d reading directory lblock 0", err);
2279 else
2280 ext4_warning(inode->i_sb,
2281 "bad directory (dir #%lu) - no data block",
2282 inode->i_ino);
2283 return 1;
2284 }
2285 if (!buffer_verified(bh) &&
2286 !ext4_dirent_csum_verify(inode,
2287 (struct ext4_dir_entry *)bh->b_data)) {
2288 EXT4_ERROR_INODE(inode, "checksum error reading directory "
2289 "lblock 0");
2290 return -EIO;
2291 }
2292 set_buffer_verified(bh);
2293 de = (struct ext4_dir_entry_2 *) bh->b_data;
2294 de1 = ext4_next_entry(de, sb->s_blocksize);
2295 if (le32_to_cpu(de->inode) != inode->i_ino ||
2296 !le32_to_cpu(de1->inode) ||
2297 strcmp(".", de->name) ||
2298 strcmp("..", de1->name)) {
2299 ext4_warning(inode->i_sb,
2300 "bad directory (dir #%lu) - no `.' or `..'",
2301 inode->i_ino);
2302 brelse(bh);
2303 return 1;
2304 }
2305 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2306 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2307 de = ext4_next_entry(de1, sb->s_blocksize);
2308 while (offset < inode->i_size) {
2309 if (!bh ||
2310 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2311 unsigned int lblock;
2312 err = 0;
2313 brelse(bh);
2314 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2315 bh = ext4_bread(NULL, inode, lblock, 0, &err);
2316 if (!bh) {
2317 if (err)
2318 EXT4_ERROR_INODE(inode,
2319 "error %d reading directory "
2320 "lblock %u", err, lblock);
2321 offset += sb->s_blocksize;
2322 continue;
2323 }
2324 if (!buffer_verified(bh) &&
2325 !ext4_dirent_csum_verify(inode,
2326 (struct ext4_dir_entry *)bh->b_data)) {
2327 EXT4_ERROR_INODE(inode, "checksum error "
2328 "reading directory lblock 0");
2329 return -EIO;
2330 }
2331 set_buffer_verified(bh);
2332 de = (struct ext4_dir_entry_2 *) bh->b_data;
2333 }
2334 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
2335 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2336 sb->s_blocksize);
2337 offset = (offset | (sb->s_blocksize - 1)) + 1;
2338 continue;
2339 }
2340 if (le32_to_cpu(de->inode)) {
2341 brelse(bh);
2342 return 0;
2343 }
2344 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2345 de = ext4_next_entry(de, sb->s_blocksize);
2346 }
2347 brelse(bh);
2348 return 1;
2349}
2350
2351/* ext4_orphan_add() links an unlinked or truncated inode into a list of
2352 * such inodes, starting at the superblock, in case we crash before the
2353 * file is closed/deleted, or in case the inode truncate spans multiple
2354 * transactions and the last transaction is not recovered after a crash.
2355 *
2356 * At filesystem recovery time, we walk this list deleting unlinked
2357 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2358 */
2359int ext4_orphan_add(handle_t *handle, struct inode *inode)
2360{
2361 struct super_block *sb = inode->i_sb;
2362 struct ext4_iloc iloc;
2363 int err = 0, rc;
2364
2365 if (!ext4_handle_valid(handle))
2366 return 0;
2367
2368 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2369 if (!list_empty(&EXT4_I(inode)->i_orphan))
2370 goto out_unlock;
2371
2372 /*
2373 * Orphan handling is only valid for files with data blocks
2374 * being truncated, or files being unlinked. Note that we either
2375 * hold i_mutex, or the inode can not be referenced from outside,
2376 * so i_nlink should not be bumped due to race
2377 */
2378 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2379 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2380
2381 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2382 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2383 if (err)
2384 goto out_unlock;
2385
2386 err = ext4_reserve_inode_write(handle, inode, &iloc);
2387 if (err)
2388 goto out_unlock;
2389 /*
2390 * Due to previous errors inode may be already a part of on-disk
2391 * orphan list. If so skip on-disk list modification.
2392 */
2393 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2394 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2395 goto mem_insert;
2396
2397 /* Insert this inode at the head of the on-disk orphan list... */
2398 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2399 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2400 err = ext4_handle_dirty_super_now(handle, sb);
2401 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2402 if (!err)
2403 err = rc;
2404
2405 /* Only add to the head of the in-memory list if all the
2406 * previous operations succeeded. If the orphan_add is going to
2407 * fail (possibly taking the journal offline), we can't risk
2408 * leaving the inode on the orphan list: stray orphan-list
2409 * entries can cause panics at unmount time.
2410 *
2411 * This is safe: on error we're going to ignore the orphan list
2412 * anyway on the next recovery. */
2413mem_insert:
2414 if (!err)
2415 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2416
2417 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2418 jbd_debug(4, "orphan inode %lu will point to %d\n",
2419 inode->i_ino, NEXT_ORPHAN(inode));
2420out_unlock:
2421 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2422 ext4_std_error(inode->i_sb, err);
2423 return err;
2424}
2425
2426/*
2427 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2428 * of such inodes stored on disk, because it is finally being cleaned up.
2429 */
2430int ext4_orphan_del(handle_t *handle, struct inode *inode)
2431{
2432 struct list_head *prev;
2433 struct ext4_inode_info *ei = EXT4_I(inode);
2434 struct ext4_sb_info *sbi;
2435 __u32 ino_next;
2436 struct ext4_iloc iloc;
2437 int err = 0;
2438
2439 /* ext4_handle_valid() assumes a valid handle_t pointer */
2440 if (handle && !ext4_handle_valid(handle))
2441 return 0;
2442
2443 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2444 if (list_empty(&ei->i_orphan))
2445 goto out;
2446
2447 ino_next = NEXT_ORPHAN(inode);
2448 prev = ei->i_orphan.prev;
2449 sbi = EXT4_SB(inode->i_sb);
2450
2451 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2452
2453 list_del_init(&ei->i_orphan);
2454
2455 /* If we're on an error path, we may not have a valid
2456 * transaction handle with which to update the orphan list on
2457 * disk, but we still need to remove the inode from the linked
2458 * list in memory. */
2459 if (sbi->s_journal && !handle)
2460 goto out;
2461
2462 err = ext4_reserve_inode_write(handle, inode, &iloc);
2463 if (err)
2464 goto out_err;
2465
2466 if (prev == &sbi->s_orphan) {
2467 jbd_debug(4, "superblock will point to %u\n", ino_next);
2468 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2469 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2470 if (err)
2471 goto out_brelse;
2472 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2473 err = ext4_handle_dirty_super_now(handle, inode->i_sb);
2474 } else {
2475 struct ext4_iloc iloc2;
2476 struct inode *i_prev =
2477 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2478
2479 jbd_debug(4, "orphan inode %lu will point to %u\n",
2480 i_prev->i_ino, ino_next);
2481 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2482 if (err)
2483 goto out_brelse;
2484 NEXT_ORPHAN(i_prev) = ino_next;
2485 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2486 }
2487 if (err)
2488 goto out_brelse;
2489 NEXT_ORPHAN(inode) = 0;
2490 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2491
2492out_err:
2493 ext4_std_error(inode->i_sb, err);
2494out:
2495 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2496 return err;
2497
2498out_brelse:
2499 brelse(iloc.bh);
2500 goto out_err;
2501}
2502
2503static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2504{
2505 int retval;
2506 struct inode *inode;
2507 struct buffer_head *bh;
2508 struct ext4_dir_entry_2 *de;
2509 handle_t *handle;
2510
2511 /* Initialize quotas before so that eventual writes go in
2512 * separate transaction */
2513 dquot_initialize(dir);
2514 dquot_initialize(dentry->d_inode);
2515
2516 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2517 if (IS_ERR(handle))
2518 return PTR_ERR(handle);
2519
2520 retval = -ENOENT;
2521 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2522 if (!bh)
2523 goto end_rmdir;
2524
2525 if (IS_DIRSYNC(dir))
2526 ext4_handle_sync(handle);
2527
2528 inode = dentry->d_inode;
2529
2530 retval = -EIO;
2531 if (le32_to_cpu(de->inode) != inode->i_ino)
2532 goto end_rmdir;
2533
2534 retval = -ENOTEMPTY;
2535 if (!empty_dir(inode))
2536 goto end_rmdir;
2537
2538 retval = ext4_delete_entry(handle, dir, de, bh);
2539 if (retval)
2540 goto end_rmdir;
2541 if (!EXT4_DIR_LINK_EMPTY(inode))
2542 ext4_warning(inode->i_sb,
2543 "empty directory has too many links (%d)",
2544 inode->i_nlink);
2545 inode->i_version++;
2546 clear_nlink(inode);
2547 /* There's no need to set i_disksize: the fact that i_nlink is
2548 * zero will ensure that the right thing happens during any
2549 * recovery. */
2550 inode->i_size = 0;
2551 ext4_orphan_add(handle, inode);
2552 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2553 ext4_mark_inode_dirty(handle, inode);
2554 ext4_dec_count(handle, dir);
2555 ext4_update_dx_flag(dir);
2556 ext4_mark_inode_dirty(handle, dir);
2557
2558end_rmdir:
2559 ext4_journal_stop(handle);
2560 brelse(bh);
2561 return retval;
2562}
2563
2564static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2565{
2566 int retval;
2567 struct inode *inode;
2568 struct buffer_head *bh;
2569 struct ext4_dir_entry_2 *de;
2570 handle_t *handle;
2571
2572 trace_ext4_unlink_enter(dir, dentry);
2573 /* Initialize quotas before so that eventual writes go
2574 * in separate transaction */
2575 dquot_initialize(dir);
2576 dquot_initialize(dentry->d_inode);
2577
2578 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2579 if (IS_ERR(handle))
2580 return PTR_ERR(handle);
2581
2582 if (IS_DIRSYNC(dir))
2583 ext4_handle_sync(handle);
2584
2585 retval = -ENOENT;
2586 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2587 if (!bh)
2588 goto end_unlink;
2589
2590 inode = dentry->d_inode;
2591
2592 retval = -EIO;
2593 if (le32_to_cpu(de->inode) != inode->i_ino)
2594 goto end_unlink;
2595
2596 if (!inode->i_nlink) {
2597 ext4_warning(inode->i_sb,
2598 "Deleting nonexistent file (%lu), %d",
2599 inode->i_ino, inode->i_nlink);
2600 set_nlink(inode, 1);
2601 }
2602 retval = ext4_delete_entry(handle, dir, de, bh);
2603 if (retval)
2604 goto end_unlink;
2605 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2606 ext4_update_dx_flag(dir);
2607 ext4_mark_inode_dirty(handle, dir);
2608 drop_nlink(inode);
2609 if (!inode->i_nlink)
2610 ext4_orphan_add(handle, inode);
2611 inode->i_ctime = ext4_current_time(inode);
2612 ext4_mark_inode_dirty(handle, inode);
2613 retval = 0;
2614
2615end_unlink:
2616 ext4_journal_stop(handle);
2617 brelse(bh);
2618 trace_ext4_unlink_exit(dentry, retval);
2619 return retval;
2620}
2621
2622static int ext4_symlink(struct inode *dir,
2623 struct dentry *dentry, const char *symname)
2624{
2625 handle_t *handle;
2626 struct inode *inode;
2627 int l, err, retries = 0;
2628 int credits;
2629
2630 l = strlen(symname)+1;
2631 if (l > dir->i_sb->s_blocksize)
2632 return -ENAMETOOLONG;
2633
2634 dquot_initialize(dir);
2635
2636 if (l > EXT4_N_BLOCKS * 4) {
2637 /*
2638 * For non-fast symlinks, we just allocate inode and put it on
2639 * orphan list in the first transaction => we need bitmap,
2640 * group descriptor, sb, inode block, quota blocks, and
2641 * possibly selinux xattr blocks.
2642 */
2643 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2644 EXT4_XATTR_TRANS_BLOCKS;
2645 } else {
2646 /*
2647 * Fast symlink. We have to add entry to directory
2648 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2649 * allocate new inode (bitmap, group descriptor, inode block,
2650 * quota blocks, sb is already counted in previous macros).
2651 */
2652 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2653 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2654 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2655 }
2656retry:
2657 handle = ext4_journal_start(dir, credits);
2658 if (IS_ERR(handle))
2659 return PTR_ERR(handle);
2660
2661 if (IS_DIRSYNC(dir))
2662 ext4_handle_sync(handle);
2663
2664 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2665 &dentry->d_name, 0, NULL);
2666 err = PTR_ERR(inode);
2667 if (IS_ERR(inode))
2668 goto out_stop;
2669
2670 if (l > EXT4_N_BLOCKS * 4) {
2671 inode->i_op = &ext4_symlink_inode_operations;
2672 ext4_set_aops(inode);
2673 /*
2674 * We cannot call page_symlink() with transaction started
2675 * because it calls into ext4_write_begin() which can wait
2676 * for transaction commit if we are running out of space
2677 * and thus we deadlock. So we have to stop transaction now
2678 * and restart it when symlink contents is written.
2679 *
2680 * To keep fs consistent in case of crash, we have to put inode
2681 * to orphan list in the mean time.
2682 */
2683 drop_nlink(inode);
2684 err = ext4_orphan_add(handle, inode);
2685 ext4_journal_stop(handle);
2686 if (err)
2687 goto err_drop_inode;
2688 err = __page_symlink(inode, symname, l, 1);
2689 if (err)
2690 goto err_drop_inode;
2691 /*
2692 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2693 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2694 */
2695 handle = ext4_journal_start(dir,
2696 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2697 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2698 if (IS_ERR(handle)) {
2699 err = PTR_ERR(handle);
2700 goto err_drop_inode;
2701 }
2702 set_nlink(inode, 1);
2703 err = ext4_orphan_del(handle, inode);
2704 if (err) {
2705 ext4_journal_stop(handle);
2706 clear_nlink(inode);
2707 goto err_drop_inode;
2708 }
2709 } else {
2710 /* clear the extent format for fast symlink */
2711 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2712 inode->i_op = &ext4_fast_symlink_inode_operations;
2713 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2714 inode->i_size = l-1;
2715 }
2716 EXT4_I(inode)->i_disksize = inode->i_size;
2717 err = ext4_add_nondir(handle, dentry, inode);
2718out_stop:
2719 ext4_journal_stop(handle);
2720 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2721 goto retry;
2722 return err;
2723err_drop_inode:
2724 unlock_new_inode(inode);
2725 iput(inode);
2726 return err;
2727}
2728
2729static int ext4_link(struct dentry *old_dentry,
2730 struct inode *dir, struct dentry *dentry)
2731{
2732 handle_t *handle;
2733 struct inode *inode = old_dentry->d_inode;
2734 int err, retries = 0;
2735
2736 if (inode->i_nlink >= EXT4_LINK_MAX)
2737 return -EMLINK;
2738
2739 dquot_initialize(dir);
2740
2741retry:
2742 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2743 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2744 if (IS_ERR(handle))
2745 return PTR_ERR(handle);
2746
2747 if (IS_DIRSYNC(dir))
2748 ext4_handle_sync(handle);
2749
2750 inode->i_ctime = ext4_current_time(inode);
2751 ext4_inc_count(handle, inode);
2752 ihold(inode);
2753
2754 err = ext4_add_entry(handle, dentry, inode);
2755 if (!err) {
2756 ext4_mark_inode_dirty(handle, inode);
2757 d_instantiate(dentry, inode);
2758 } else {
2759 drop_nlink(inode);
2760 iput(inode);
2761 }
2762 ext4_journal_stop(handle);
2763 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2764 goto retry;
2765 return err;
2766}
2767
2768#define PARENT_INO(buffer, size) \
2769 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2770
2771/*
2772 * Anybody can rename anything with this: the permission checks are left to the
2773 * higher-level routines.
2774 */
2775static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2776 struct inode *new_dir, struct dentry *new_dentry)
2777{
2778 handle_t *handle;
2779 struct inode *old_inode, *new_inode;
2780 struct buffer_head *old_bh, *new_bh, *dir_bh;
2781 struct ext4_dir_entry_2 *old_de, *new_de;
2782 int retval, force_da_alloc = 0;
2783
2784 dquot_initialize(old_dir);
2785 dquot_initialize(new_dir);
2786
2787 old_bh = new_bh = dir_bh = NULL;
2788
2789 /* Initialize quotas before so that eventual writes go
2790 * in separate transaction */
2791 if (new_dentry->d_inode)
2792 dquot_initialize(new_dentry->d_inode);
2793 handle = ext4_journal_start(old_dir, 2 *
2794 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2795 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2796 if (IS_ERR(handle))
2797 return PTR_ERR(handle);
2798
2799 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2800 ext4_handle_sync(handle);
2801
2802 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2803 /*
2804 * Check for inode number is _not_ due to possible IO errors.
2805 * We might rmdir the source, keep it as pwd of some process
2806 * and merrily kill the link to whatever was created under the
2807 * same name. Goodbye sticky bit ;-<
2808 */
2809 old_inode = old_dentry->d_inode;
2810 retval = -ENOENT;
2811 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2812 goto end_rename;
2813
2814 new_inode = new_dentry->d_inode;
2815 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2816 if (new_bh) {
2817 if (!new_inode) {
2818 brelse(new_bh);
2819 new_bh = NULL;
2820 }
2821 }
2822 if (S_ISDIR(old_inode->i_mode)) {
2823 if (new_inode) {
2824 retval = -ENOTEMPTY;
2825 if (!empty_dir(new_inode))
2826 goto end_rename;
2827 }
2828 retval = -EIO;
2829 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2830 if (!dir_bh)
2831 goto end_rename;
2832 if (!buffer_verified(dir_bh) &&
2833 !ext4_dirent_csum_verify(old_inode,
2834 (struct ext4_dir_entry *)dir_bh->b_data))
2835 goto end_rename;
2836 set_buffer_verified(dir_bh);
2837 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2838 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2839 goto end_rename;
2840 retval = -EMLINK;
2841 if (!new_inode && new_dir != old_dir &&
2842 EXT4_DIR_LINK_MAX(new_dir))
2843 goto end_rename;
2844 BUFFER_TRACE(dir_bh, "get_write_access");
2845 retval = ext4_journal_get_write_access(handle, dir_bh);
2846 if (retval)
2847 goto end_rename;
2848 }
2849 if (!new_bh) {
2850 retval = ext4_add_entry(handle, new_dentry, old_inode);
2851 if (retval)
2852 goto end_rename;
2853 } else {
2854 BUFFER_TRACE(new_bh, "get write access");
2855 retval = ext4_journal_get_write_access(handle, new_bh);
2856 if (retval)
2857 goto end_rename;
2858 new_de->inode = cpu_to_le32(old_inode->i_ino);
2859 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2860 EXT4_FEATURE_INCOMPAT_FILETYPE))
2861 new_de->file_type = old_de->file_type;
2862 new_dir->i_version++;
2863 new_dir->i_ctime = new_dir->i_mtime =
2864 ext4_current_time(new_dir);
2865 ext4_mark_inode_dirty(handle, new_dir);
2866 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2867 retval = ext4_handle_dirty_dirent_node(handle, new_dir, new_bh);
2868 if (unlikely(retval)) {
2869 ext4_std_error(new_dir->i_sb, retval);
2870 goto end_rename;
2871 }
2872 brelse(new_bh);
2873 new_bh = NULL;
2874 }
2875
2876 /*
2877 * Like most other Unix systems, set the ctime for inodes on a
2878 * rename.
2879 */
2880 old_inode->i_ctime = ext4_current_time(old_inode);
2881 ext4_mark_inode_dirty(handle, old_inode);
2882
2883 /*
2884 * ok, that's it
2885 */
2886 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2887 old_de->name_len != old_dentry->d_name.len ||
2888 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2889 (retval = ext4_delete_entry(handle, old_dir,
2890 old_de, old_bh)) == -ENOENT) {
2891 /* old_de could have moved from under us during htree split, so
2892 * make sure that we are deleting the right entry. We might
2893 * also be pointing to a stale entry in the unused part of
2894 * old_bh so just checking inum and the name isn't enough. */
2895 struct buffer_head *old_bh2;
2896 struct ext4_dir_entry_2 *old_de2;
2897
2898 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2899 if (old_bh2) {
2900 retval = ext4_delete_entry(handle, old_dir,
2901 old_de2, old_bh2);
2902 brelse(old_bh2);
2903 }
2904 }
2905 if (retval) {
2906 ext4_warning(old_dir->i_sb,
2907 "Deleting old file (%lu), %d, error=%d",
2908 old_dir->i_ino, old_dir->i_nlink, retval);
2909 }
2910
2911 if (new_inode) {
2912 ext4_dec_count(handle, new_inode);
2913 new_inode->i_ctime = ext4_current_time(new_inode);
2914 }
2915 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2916 ext4_update_dx_flag(old_dir);
2917 if (dir_bh) {
2918 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2919 cpu_to_le32(new_dir->i_ino);
2920 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2921 if (is_dx(old_inode)) {
2922 retval = ext4_handle_dirty_dx_node(handle,
2923 old_inode,
2924 dir_bh);
2925 } else {
2926 retval = ext4_handle_dirty_dirent_node(handle,
2927 old_inode,
2928 dir_bh);
2929 }
2930 if (retval) {
2931 ext4_std_error(old_dir->i_sb, retval);
2932 goto end_rename;
2933 }
2934 ext4_dec_count(handle, old_dir);
2935 if (new_inode) {
2936 /* checked empty_dir above, can't have another parent,
2937 * ext4_dec_count() won't work for many-linked dirs */
2938 clear_nlink(new_inode);
2939 } else {
2940 ext4_inc_count(handle, new_dir);
2941 ext4_update_dx_flag(new_dir);
2942 ext4_mark_inode_dirty(handle, new_dir);
2943 }
2944 }
2945 ext4_mark_inode_dirty(handle, old_dir);
2946 if (new_inode) {
2947 ext4_mark_inode_dirty(handle, new_inode);
2948 if (!new_inode->i_nlink)
2949 ext4_orphan_add(handle, new_inode);
2950 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2951 force_da_alloc = 1;
2952 }
2953 retval = 0;
2954
2955end_rename:
2956 brelse(dir_bh);
2957 brelse(old_bh);
2958 brelse(new_bh);
2959 ext4_journal_stop(handle);
2960 if (retval == 0 && force_da_alloc)
2961 ext4_alloc_da_blocks(old_inode);
2962 return retval;
2963}
2964
2965/*
2966 * directories can handle most operations...
2967 */
2968const struct inode_operations ext4_dir_inode_operations = {
2969 .create = ext4_create,
2970 .lookup = ext4_lookup,
2971 .link = ext4_link,
2972 .unlink = ext4_unlink,
2973 .symlink = ext4_symlink,
2974 .mkdir = ext4_mkdir,
2975 .rmdir = ext4_rmdir,
2976 .mknod = ext4_mknod,
2977 .rename = ext4_rename,
2978 .setattr = ext4_setattr,
2979#ifdef CONFIG_EXT4_FS_XATTR
2980 .setxattr = generic_setxattr,
2981 .getxattr = generic_getxattr,
2982 .listxattr = ext4_listxattr,
2983 .removexattr = generic_removexattr,
2984#endif
2985 .get_acl = ext4_get_acl,
2986 .fiemap = ext4_fiemap,
2987};
2988
2989const struct inode_operations ext4_special_inode_operations = {
2990 .setattr = ext4_setattr,
2991#ifdef CONFIG_EXT4_FS_XATTR
2992 .setxattr = generic_setxattr,
2993 .getxattr = generic_getxattr,
2994 .listxattr = ext4_listxattr,
2995 .removexattr = generic_removexattr,
2996#endif
2997 .get_acl = ext4_get_acl,
2998};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/namei.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/namei.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
25 * Theodore Ts'o, 2002
26 */
27
28#include <linux/fs.h>
29#include <linux/pagemap.h>
30#include <linux/time.h>
31#include <linux/fcntl.h>
32#include <linux/stat.h>
33#include <linux/string.h>
34#include <linux/quotaops.h>
35#include <linux/buffer_head.h>
36#include <linux/bio.h>
37#include <linux/iversion.h>
38#include <linux/unicode.h>
39#include "ext4.h"
40#include "ext4_jbd2.h"
41
42#include "xattr.h"
43#include "acl.h"
44
45#include <trace/events/ext4.h>
46/*
47 * define how far ahead to read directories while searching them.
48 */
49#define NAMEI_RA_CHUNKS 2
50#define NAMEI_RA_BLOCKS 4
51#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52
53static struct buffer_head *ext4_append(handle_t *handle,
54 struct inode *inode,
55 ext4_lblk_t *block)
56{
57 struct buffer_head *bh;
58 int err;
59
60 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
61 ((inode->i_size >> 10) >=
62 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
63 return ERR_PTR(-ENOSPC);
64
65 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66
67 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
68 if (IS_ERR(bh))
69 return bh;
70 inode->i_size += inode->i_sb->s_blocksize;
71 EXT4_I(inode)->i_disksize = inode->i_size;
72 BUFFER_TRACE(bh, "get_write_access");
73 err = ext4_journal_get_write_access(handle, bh);
74 if (err) {
75 brelse(bh);
76 ext4_std_error(inode->i_sb, err);
77 return ERR_PTR(err);
78 }
79 return bh;
80}
81
82static int ext4_dx_csum_verify(struct inode *inode,
83 struct ext4_dir_entry *dirent);
84
85/*
86 * Hints to ext4_read_dirblock regarding whether we expect a directory
87 * block being read to be an index block, or a block containing
88 * directory entries (and if the latter, whether it was found via a
89 * logical block in an htree index block). This is used to control
90 * what sort of sanity checkinig ext4_read_dirblock() will do on the
91 * directory block read from the storage device. EITHER will means
92 * the caller doesn't know what kind of directory block will be read,
93 * so no specific verification will be done.
94 */
95typedef enum {
96 EITHER, INDEX, DIRENT, DIRENT_HTREE
97} dirblock_type_t;
98
99#define ext4_read_dirblock(inode, block, type) \
100 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
101
102static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
103 ext4_lblk_t block,
104 dirblock_type_t type,
105 const char *func,
106 unsigned int line)
107{
108 struct buffer_head *bh;
109 struct ext4_dir_entry *dirent;
110 int is_dx_block = 0;
111
112 bh = ext4_bread(NULL, inode, block, 0);
113 if (IS_ERR(bh)) {
114 __ext4_warning(inode->i_sb, func, line,
115 "inode #%lu: lblock %lu: comm %s: "
116 "error %ld reading directory block",
117 inode->i_ino, (unsigned long)block,
118 current->comm, PTR_ERR(bh));
119
120 return bh;
121 }
122 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
123 ext4_error_inode(inode, func, line, block,
124 "Directory hole found for htree %s block",
125 (type == INDEX) ? "index" : "leaf");
126 return ERR_PTR(-EFSCORRUPTED);
127 }
128 if (!bh)
129 return NULL;
130 dirent = (struct ext4_dir_entry *) bh->b_data;
131 /* Determine whether or not we have an index block */
132 if (is_dx(inode)) {
133 if (block == 0)
134 is_dx_block = 1;
135 else if (ext4_rec_len_from_disk(dirent->rec_len,
136 inode->i_sb->s_blocksize) ==
137 inode->i_sb->s_blocksize)
138 is_dx_block = 1;
139 }
140 if (!is_dx_block && type == INDEX) {
141 ext4_error_inode(inode, func, line, block,
142 "directory leaf block found instead of index block");
143 brelse(bh);
144 return ERR_PTR(-EFSCORRUPTED);
145 }
146 if (!ext4_has_metadata_csum(inode->i_sb) ||
147 buffer_verified(bh))
148 return bh;
149
150 /*
151 * An empty leaf block can get mistaken for a index block; for
152 * this reason, we can only check the index checksum when the
153 * caller is sure it should be an index block.
154 */
155 if (is_dx_block && type == INDEX) {
156 if (ext4_dx_csum_verify(inode, dirent))
157 set_buffer_verified(bh);
158 else {
159 ext4_error_inode(inode, func, line, block,
160 "Directory index failed checksum");
161 brelse(bh);
162 return ERR_PTR(-EFSBADCRC);
163 }
164 }
165 if (!is_dx_block) {
166 if (ext4_dirblock_csum_verify(inode, bh))
167 set_buffer_verified(bh);
168 else {
169 ext4_error_inode(inode, func, line, block,
170 "Directory block failed checksum");
171 brelse(bh);
172 return ERR_PTR(-EFSBADCRC);
173 }
174 }
175 return bh;
176}
177
178#ifndef assert
179#define assert(test) J_ASSERT(test)
180#endif
181
182#ifdef DX_DEBUG
183#define dxtrace(command) command
184#else
185#define dxtrace(command)
186#endif
187
188struct fake_dirent
189{
190 __le32 inode;
191 __le16 rec_len;
192 u8 name_len;
193 u8 file_type;
194};
195
196struct dx_countlimit
197{
198 __le16 limit;
199 __le16 count;
200};
201
202struct dx_entry
203{
204 __le32 hash;
205 __le32 block;
206};
207
208/*
209 * dx_root_info is laid out so that if it should somehow get overlaid by a
210 * dirent the two low bits of the hash version will be zero. Therefore, the
211 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
212 */
213
214struct dx_root
215{
216 struct fake_dirent dot;
217 char dot_name[4];
218 struct fake_dirent dotdot;
219 char dotdot_name[4];
220 struct dx_root_info
221 {
222 __le32 reserved_zero;
223 u8 hash_version;
224 u8 info_length; /* 8 */
225 u8 indirect_levels;
226 u8 unused_flags;
227 }
228 info;
229 struct dx_entry entries[0];
230};
231
232struct dx_node
233{
234 struct fake_dirent fake;
235 struct dx_entry entries[0];
236};
237
238
239struct dx_frame
240{
241 struct buffer_head *bh;
242 struct dx_entry *entries;
243 struct dx_entry *at;
244};
245
246struct dx_map_entry
247{
248 u32 hash;
249 u16 offs;
250 u16 size;
251};
252
253/*
254 * This goes at the end of each htree block.
255 */
256struct dx_tail {
257 u32 dt_reserved;
258 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
259};
260
261static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
262static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
263static inline unsigned dx_get_hash(struct dx_entry *entry);
264static void dx_set_hash(struct dx_entry *entry, unsigned value);
265static unsigned dx_get_count(struct dx_entry *entries);
266static unsigned dx_get_limit(struct dx_entry *entries);
267static void dx_set_count(struct dx_entry *entries, unsigned value);
268static void dx_set_limit(struct dx_entry *entries, unsigned value);
269static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
270static unsigned dx_node_limit(struct inode *dir);
271static struct dx_frame *dx_probe(struct ext4_filename *fname,
272 struct inode *dir,
273 struct dx_hash_info *hinfo,
274 struct dx_frame *frame);
275static void dx_release(struct dx_frame *frames);
276static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
277 unsigned blocksize, struct dx_hash_info *hinfo,
278 struct dx_map_entry map[]);
279static void dx_sort_map(struct dx_map_entry *map, unsigned count);
280static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
281 struct dx_map_entry *offsets, int count, unsigned blocksize);
282static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
283static void dx_insert_block(struct dx_frame *frame,
284 u32 hash, ext4_lblk_t block);
285static int ext4_htree_next_block(struct inode *dir, __u32 hash,
286 struct dx_frame *frame,
287 struct dx_frame *frames,
288 __u32 *start_hash);
289static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
290 struct ext4_filename *fname,
291 struct ext4_dir_entry_2 **res_dir);
292static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
293 struct inode *dir, struct inode *inode);
294
295/* checksumming functions */
296void ext4_initialize_dirent_tail(struct buffer_head *bh,
297 unsigned int blocksize)
298{
299 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
300
301 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
302 t->det_rec_len = ext4_rec_len_to_disk(
303 sizeof(struct ext4_dir_entry_tail), blocksize);
304 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
305}
306
307/* Walk through a dirent block to find a checksum "dirent" at the tail */
308static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
309 struct buffer_head *bh)
310{
311 struct ext4_dir_entry_tail *t;
312
313#ifdef PARANOID
314 struct ext4_dir_entry *d, *top;
315
316 d = (struct ext4_dir_entry *)bh->b_data;
317 top = (struct ext4_dir_entry *)(bh->b_data +
318 (EXT4_BLOCK_SIZE(inode->i_sb) -
319 sizeof(struct ext4_dir_entry_tail)));
320 while (d < top && d->rec_len)
321 d = (struct ext4_dir_entry *)(((void *)d) +
322 le16_to_cpu(d->rec_len));
323
324 if (d != top)
325 return NULL;
326
327 t = (struct ext4_dir_entry_tail *)d;
328#else
329 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
330#endif
331
332 if (t->det_reserved_zero1 ||
333 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
334 t->det_reserved_zero2 ||
335 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
336 return NULL;
337
338 return t;
339}
340
341static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
342{
343 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
344 struct ext4_inode_info *ei = EXT4_I(inode);
345 __u32 csum;
346
347 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
348 return cpu_to_le32(csum);
349}
350
351#define warn_no_space_for_csum(inode) \
352 __warn_no_space_for_csum((inode), __func__, __LINE__)
353
354static void __warn_no_space_for_csum(struct inode *inode, const char *func,
355 unsigned int line)
356{
357 __ext4_warning_inode(inode, func, line,
358 "No space for directory leaf checksum. Please run e2fsck -D.");
359}
360
361int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
362{
363 struct ext4_dir_entry_tail *t;
364
365 if (!ext4_has_metadata_csum(inode->i_sb))
366 return 1;
367
368 t = get_dirent_tail(inode, bh);
369 if (!t) {
370 warn_no_space_for_csum(inode);
371 return 0;
372 }
373
374 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
375 (char *)t - bh->b_data))
376 return 0;
377
378 return 1;
379}
380
381static void ext4_dirblock_csum_set(struct inode *inode,
382 struct buffer_head *bh)
383{
384 struct ext4_dir_entry_tail *t;
385
386 if (!ext4_has_metadata_csum(inode->i_sb))
387 return;
388
389 t = get_dirent_tail(inode, bh);
390 if (!t) {
391 warn_no_space_for_csum(inode);
392 return;
393 }
394
395 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
396 (char *)t - bh->b_data);
397}
398
399int ext4_handle_dirty_dirblock(handle_t *handle,
400 struct inode *inode,
401 struct buffer_head *bh)
402{
403 ext4_dirblock_csum_set(inode, bh);
404 return ext4_handle_dirty_metadata(handle, inode, bh);
405}
406
407static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
408 struct ext4_dir_entry *dirent,
409 int *offset)
410{
411 struct ext4_dir_entry *dp;
412 struct dx_root_info *root;
413 int count_offset;
414
415 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
416 count_offset = 8;
417 else if (le16_to_cpu(dirent->rec_len) == 12) {
418 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
419 if (le16_to_cpu(dp->rec_len) !=
420 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
421 return NULL;
422 root = (struct dx_root_info *)(((void *)dp + 12));
423 if (root->reserved_zero ||
424 root->info_length != sizeof(struct dx_root_info))
425 return NULL;
426 count_offset = 32;
427 } else
428 return NULL;
429
430 if (offset)
431 *offset = count_offset;
432 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
433}
434
435static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
436 int count_offset, int count, struct dx_tail *t)
437{
438 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
439 struct ext4_inode_info *ei = EXT4_I(inode);
440 __u32 csum;
441 int size;
442 __u32 dummy_csum = 0;
443 int offset = offsetof(struct dx_tail, dt_checksum);
444
445 size = count_offset + (count * sizeof(struct dx_entry));
446 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
447 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
448 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
449
450 return cpu_to_le32(csum);
451}
452
453static int ext4_dx_csum_verify(struct inode *inode,
454 struct ext4_dir_entry *dirent)
455{
456 struct dx_countlimit *c;
457 struct dx_tail *t;
458 int count_offset, limit, count;
459
460 if (!ext4_has_metadata_csum(inode->i_sb))
461 return 1;
462
463 c = get_dx_countlimit(inode, dirent, &count_offset);
464 if (!c) {
465 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
466 return 0;
467 }
468 limit = le16_to_cpu(c->limit);
469 count = le16_to_cpu(c->count);
470 if (count_offset + (limit * sizeof(struct dx_entry)) >
471 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
472 warn_no_space_for_csum(inode);
473 return 0;
474 }
475 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
476
477 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
478 count, t))
479 return 0;
480 return 1;
481}
482
483static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
484{
485 struct dx_countlimit *c;
486 struct dx_tail *t;
487 int count_offset, limit, count;
488
489 if (!ext4_has_metadata_csum(inode->i_sb))
490 return;
491
492 c = get_dx_countlimit(inode, dirent, &count_offset);
493 if (!c) {
494 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
495 return;
496 }
497 limit = le16_to_cpu(c->limit);
498 count = le16_to_cpu(c->count);
499 if (count_offset + (limit * sizeof(struct dx_entry)) >
500 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
501 warn_no_space_for_csum(inode);
502 return;
503 }
504 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
505
506 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
507}
508
509static inline int ext4_handle_dirty_dx_node(handle_t *handle,
510 struct inode *inode,
511 struct buffer_head *bh)
512{
513 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
514 return ext4_handle_dirty_metadata(handle, inode, bh);
515}
516
517/*
518 * p is at least 6 bytes before the end of page
519 */
520static inline struct ext4_dir_entry_2 *
521ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
522{
523 return (struct ext4_dir_entry_2 *)((char *)p +
524 ext4_rec_len_from_disk(p->rec_len, blocksize));
525}
526
527/*
528 * Future: use high four bits of block for coalesce-on-delete flags
529 * Mask them off for now.
530 */
531
532static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
533{
534 return le32_to_cpu(entry->block) & 0x0fffffff;
535}
536
537static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
538{
539 entry->block = cpu_to_le32(value);
540}
541
542static inline unsigned dx_get_hash(struct dx_entry *entry)
543{
544 return le32_to_cpu(entry->hash);
545}
546
547static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
548{
549 entry->hash = cpu_to_le32(value);
550}
551
552static inline unsigned dx_get_count(struct dx_entry *entries)
553{
554 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
555}
556
557static inline unsigned dx_get_limit(struct dx_entry *entries)
558{
559 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
560}
561
562static inline void dx_set_count(struct dx_entry *entries, unsigned value)
563{
564 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
565}
566
567static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
568{
569 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
570}
571
572static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
573{
574 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
575 EXT4_DIR_REC_LEN(2) - infosize;
576
577 if (ext4_has_metadata_csum(dir->i_sb))
578 entry_space -= sizeof(struct dx_tail);
579 return entry_space / sizeof(struct dx_entry);
580}
581
582static inline unsigned dx_node_limit(struct inode *dir)
583{
584 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
585
586 if (ext4_has_metadata_csum(dir->i_sb))
587 entry_space -= sizeof(struct dx_tail);
588 return entry_space / sizeof(struct dx_entry);
589}
590
591/*
592 * Debug
593 */
594#ifdef DX_DEBUG
595static void dx_show_index(char * label, struct dx_entry *entries)
596{
597 int i, n = dx_get_count (entries);
598 printk(KERN_DEBUG "%s index", label);
599 for (i = 0; i < n; i++) {
600 printk(KERN_CONT " %x->%lu",
601 i ? dx_get_hash(entries + i) : 0,
602 (unsigned long)dx_get_block(entries + i));
603 }
604 printk(KERN_CONT "\n");
605}
606
607struct stats
608{
609 unsigned names;
610 unsigned space;
611 unsigned bcount;
612};
613
614static struct stats dx_show_leaf(struct inode *dir,
615 struct dx_hash_info *hinfo,
616 struct ext4_dir_entry_2 *de,
617 int size, int show_names)
618{
619 unsigned names = 0, space = 0;
620 char *base = (char *) de;
621 struct dx_hash_info h = *hinfo;
622
623 printk("names: ");
624 while ((char *) de < base + size)
625 {
626 if (de->inode)
627 {
628 if (show_names)
629 {
630#ifdef CONFIG_FS_ENCRYPTION
631 int len;
632 char *name;
633 struct fscrypt_str fname_crypto_str =
634 FSTR_INIT(NULL, 0);
635 int res = 0;
636
637 name = de->name;
638 len = de->name_len;
639 if (IS_ENCRYPTED(dir))
640 res = fscrypt_get_encryption_info(dir);
641 if (res) {
642 printk(KERN_WARNING "Error setting up"
643 " fname crypto: %d\n", res);
644 }
645 if (!fscrypt_has_encryption_key(dir)) {
646 /* Directory is not encrypted */
647 ext4fs_dirhash(dir, de->name,
648 de->name_len, &h);
649 printk("%*.s:(U)%x.%u ", len,
650 name, h.hash,
651 (unsigned) ((char *) de
652 - base));
653 } else {
654 struct fscrypt_str de_name =
655 FSTR_INIT(name, len);
656
657 /* Directory is encrypted */
658 res = fscrypt_fname_alloc_buffer(
659 dir, len,
660 &fname_crypto_str);
661 if (res)
662 printk(KERN_WARNING "Error "
663 "allocating crypto "
664 "buffer--skipping "
665 "crypto\n");
666 res = fscrypt_fname_disk_to_usr(dir,
667 0, 0, &de_name,
668 &fname_crypto_str);
669 if (res) {
670 printk(KERN_WARNING "Error "
671 "converting filename "
672 "from disk to usr"
673 "\n");
674 name = "??";
675 len = 2;
676 } else {
677 name = fname_crypto_str.name;
678 len = fname_crypto_str.len;
679 }
680 ext4fs_dirhash(dir, de->name,
681 de->name_len, &h);
682 printk("%*.s:(E)%x.%u ", len, name,
683 h.hash, (unsigned) ((char *) de
684 - base));
685 fscrypt_fname_free_buffer(
686 &fname_crypto_str);
687 }
688#else
689 int len = de->name_len;
690 char *name = de->name;
691 ext4fs_dirhash(dir, de->name, de->name_len, &h);
692 printk("%*.s:%x.%u ", len, name, h.hash,
693 (unsigned) ((char *) de - base));
694#endif
695 }
696 space += EXT4_DIR_REC_LEN(de->name_len);
697 names++;
698 }
699 de = ext4_next_entry(de, size);
700 }
701 printk(KERN_CONT "(%i)\n", names);
702 return (struct stats) { names, space, 1 };
703}
704
705struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
706 struct dx_entry *entries, int levels)
707{
708 unsigned blocksize = dir->i_sb->s_blocksize;
709 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
710 unsigned bcount = 0;
711 struct buffer_head *bh;
712 printk("%i indexed blocks...\n", count);
713 for (i = 0; i < count; i++, entries++)
714 {
715 ext4_lblk_t block = dx_get_block(entries);
716 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
717 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
718 struct stats stats;
719 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
720 bh = ext4_bread(NULL,dir, block, 0);
721 if (!bh || IS_ERR(bh))
722 continue;
723 stats = levels?
724 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
725 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
726 bh->b_data, blocksize, 0);
727 names += stats.names;
728 space += stats.space;
729 bcount += stats.bcount;
730 brelse(bh);
731 }
732 if (bcount)
733 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
734 levels ? "" : " ", names, space/bcount,
735 (space/bcount)*100/blocksize);
736 return (struct stats) { names, space, bcount};
737}
738#endif /* DX_DEBUG */
739
740/*
741 * Probe for a directory leaf block to search.
742 *
743 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
744 * error in the directory index, and the caller should fall back to
745 * searching the directory normally. The callers of dx_probe **MUST**
746 * check for this error code, and make sure it never gets reflected
747 * back to userspace.
748 */
749static struct dx_frame *
750dx_probe(struct ext4_filename *fname, struct inode *dir,
751 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
752{
753 unsigned count, indirect;
754 struct dx_entry *at, *entries, *p, *q, *m;
755 struct dx_root *root;
756 struct dx_frame *frame = frame_in;
757 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
758 u32 hash;
759
760 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
761 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
762 if (IS_ERR(frame->bh))
763 return (struct dx_frame *) frame->bh;
764
765 root = (struct dx_root *) frame->bh->b_data;
766 if (root->info.hash_version != DX_HASH_TEA &&
767 root->info.hash_version != DX_HASH_HALF_MD4 &&
768 root->info.hash_version != DX_HASH_LEGACY) {
769 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
770 root->info.hash_version);
771 goto fail;
772 }
773 if (fname)
774 hinfo = &fname->hinfo;
775 hinfo->hash_version = root->info.hash_version;
776 if (hinfo->hash_version <= DX_HASH_TEA)
777 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
778 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
779 if (fname && fname_name(fname))
780 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
781 hash = hinfo->hash;
782
783 if (root->info.unused_flags & 1) {
784 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
785 root->info.unused_flags);
786 goto fail;
787 }
788
789 indirect = root->info.indirect_levels;
790 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
791 ext4_warning(dir->i_sb,
792 "Directory (ino: %lu) htree depth %#06x exceed"
793 "supported value", dir->i_ino,
794 ext4_dir_htree_level(dir->i_sb));
795 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
796 ext4_warning(dir->i_sb, "Enable large directory "
797 "feature to access it");
798 }
799 goto fail;
800 }
801
802 entries = (struct dx_entry *)(((char *)&root->info) +
803 root->info.info_length);
804
805 if (dx_get_limit(entries) != dx_root_limit(dir,
806 root->info.info_length)) {
807 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
808 dx_get_limit(entries),
809 dx_root_limit(dir, root->info.info_length));
810 goto fail;
811 }
812
813 dxtrace(printk("Look up %x", hash));
814 while (1) {
815 count = dx_get_count(entries);
816 if (!count || count > dx_get_limit(entries)) {
817 ext4_warning_inode(dir,
818 "dx entry: count %u beyond limit %u",
819 count, dx_get_limit(entries));
820 goto fail;
821 }
822
823 p = entries + 1;
824 q = entries + count - 1;
825 while (p <= q) {
826 m = p + (q - p) / 2;
827 dxtrace(printk(KERN_CONT "."));
828 if (dx_get_hash(m) > hash)
829 q = m - 1;
830 else
831 p = m + 1;
832 }
833
834 if (0) { // linear search cross check
835 unsigned n = count - 1;
836 at = entries;
837 while (n--)
838 {
839 dxtrace(printk(KERN_CONT ","));
840 if (dx_get_hash(++at) > hash)
841 {
842 at--;
843 break;
844 }
845 }
846 assert (at == p - 1);
847 }
848
849 at = p - 1;
850 dxtrace(printk(KERN_CONT " %x->%u\n",
851 at == entries ? 0 : dx_get_hash(at),
852 dx_get_block(at)));
853 frame->entries = entries;
854 frame->at = at;
855 if (!indirect--)
856 return frame;
857 frame++;
858 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
859 if (IS_ERR(frame->bh)) {
860 ret_err = (struct dx_frame *) frame->bh;
861 frame->bh = NULL;
862 goto fail;
863 }
864 entries = ((struct dx_node *) frame->bh->b_data)->entries;
865
866 if (dx_get_limit(entries) != dx_node_limit(dir)) {
867 ext4_warning_inode(dir,
868 "dx entry: limit %u != node limit %u",
869 dx_get_limit(entries), dx_node_limit(dir));
870 goto fail;
871 }
872 }
873fail:
874 while (frame >= frame_in) {
875 brelse(frame->bh);
876 frame--;
877 }
878
879 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
880 ext4_warning_inode(dir,
881 "Corrupt directory, running e2fsck is recommended");
882 return ret_err;
883}
884
885static void dx_release(struct dx_frame *frames)
886{
887 struct dx_root_info *info;
888 int i;
889 unsigned int indirect_levels;
890
891 if (frames[0].bh == NULL)
892 return;
893
894 info = &((struct dx_root *)frames[0].bh->b_data)->info;
895 /* save local copy, "info" may be freed after brelse() */
896 indirect_levels = info->indirect_levels;
897 for (i = 0; i <= indirect_levels; i++) {
898 if (frames[i].bh == NULL)
899 break;
900 brelse(frames[i].bh);
901 frames[i].bh = NULL;
902 }
903}
904
905/*
906 * This function increments the frame pointer to search the next leaf
907 * block, and reads in the necessary intervening nodes if the search
908 * should be necessary. Whether or not the search is necessary is
909 * controlled by the hash parameter. If the hash value is even, then
910 * the search is only continued if the next block starts with that
911 * hash value. This is used if we are searching for a specific file.
912 *
913 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
914 *
915 * This function returns 1 if the caller should continue to search,
916 * or 0 if it should not. If there is an error reading one of the
917 * index blocks, it will a negative error code.
918 *
919 * If start_hash is non-null, it will be filled in with the starting
920 * hash of the next page.
921 */
922static int ext4_htree_next_block(struct inode *dir, __u32 hash,
923 struct dx_frame *frame,
924 struct dx_frame *frames,
925 __u32 *start_hash)
926{
927 struct dx_frame *p;
928 struct buffer_head *bh;
929 int num_frames = 0;
930 __u32 bhash;
931
932 p = frame;
933 /*
934 * Find the next leaf page by incrementing the frame pointer.
935 * If we run out of entries in the interior node, loop around and
936 * increment pointer in the parent node. When we break out of
937 * this loop, num_frames indicates the number of interior
938 * nodes need to be read.
939 */
940 while (1) {
941 if (++(p->at) < p->entries + dx_get_count(p->entries))
942 break;
943 if (p == frames)
944 return 0;
945 num_frames++;
946 p--;
947 }
948
949 /*
950 * If the hash is 1, then continue only if the next page has a
951 * continuation hash of any value. This is used for readdir
952 * handling. Otherwise, check to see if the hash matches the
953 * desired contiuation hash. If it doesn't, return since
954 * there's no point to read in the successive index pages.
955 */
956 bhash = dx_get_hash(p->at);
957 if (start_hash)
958 *start_hash = bhash;
959 if ((hash & 1) == 0) {
960 if ((bhash & ~1) != hash)
961 return 0;
962 }
963 /*
964 * If the hash is HASH_NB_ALWAYS, we always go to the next
965 * block so no check is necessary
966 */
967 while (num_frames--) {
968 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
969 if (IS_ERR(bh))
970 return PTR_ERR(bh);
971 p++;
972 brelse(p->bh);
973 p->bh = bh;
974 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
975 }
976 return 1;
977}
978
979
980/*
981 * This function fills a red-black tree with information from a
982 * directory block. It returns the number directory entries loaded
983 * into the tree. If there is an error it is returned in err.
984 */
985static int htree_dirblock_to_tree(struct file *dir_file,
986 struct inode *dir, ext4_lblk_t block,
987 struct dx_hash_info *hinfo,
988 __u32 start_hash, __u32 start_minor_hash)
989{
990 struct buffer_head *bh;
991 struct ext4_dir_entry_2 *de, *top;
992 int err = 0, count = 0;
993 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
994
995 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
996 (unsigned long)block));
997 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
998 if (IS_ERR(bh))
999 return PTR_ERR(bh);
1000
1001 de = (struct ext4_dir_entry_2 *) bh->b_data;
1002 top = (struct ext4_dir_entry_2 *) ((char *) de +
1003 dir->i_sb->s_blocksize -
1004 EXT4_DIR_REC_LEN(0));
1005#ifdef CONFIG_FS_ENCRYPTION
1006 /* Check if the directory is encrypted */
1007 if (IS_ENCRYPTED(dir)) {
1008 err = fscrypt_get_encryption_info(dir);
1009 if (err < 0) {
1010 brelse(bh);
1011 return err;
1012 }
1013 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1014 &fname_crypto_str);
1015 if (err < 0) {
1016 brelse(bh);
1017 return err;
1018 }
1019 }
1020#endif
1021 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1022 if (ext4_check_dir_entry(dir, NULL, de, bh,
1023 bh->b_data, bh->b_size,
1024 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1025 + ((char *)de - bh->b_data))) {
1026 /* silently ignore the rest of the block */
1027 break;
1028 }
1029 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1030 if ((hinfo->hash < start_hash) ||
1031 ((hinfo->hash == start_hash) &&
1032 (hinfo->minor_hash < start_minor_hash)))
1033 continue;
1034 if (de->inode == 0)
1035 continue;
1036 if (!IS_ENCRYPTED(dir)) {
1037 tmp_str.name = de->name;
1038 tmp_str.len = de->name_len;
1039 err = ext4_htree_store_dirent(dir_file,
1040 hinfo->hash, hinfo->minor_hash, de,
1041 &tmp_str);
1042 } else {
1043 int save_len = fname_crypto_str.len;
1044 struct fscrypt_str de_name = FSTR_INIT(de->name,
1045 de->name_len);
1046
1047 /* Directory is encrypted */
1048 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1049 hinfo->minor_hash, &de_name,
1050 &fname_crypto_str);
1051 if (err) {
1052 count = err;
1053 goto errout;
1054 }
1055 err = ext4_htree_store_dirent(dir_file,
1056 hinfo->hash, hinfo->minor_hash, de,
1057 &fname_crypto_str);
1058 fname_crypto_str.len = save_len;
1059 }
1060 if (err != 0) {
1061 count = err;
1062 goto errout;
1063 }
1064 count++;
1065 }
1066errout:
1067 brelse(bh);
1068#ifdef CONFIG_FS_ENCRYPTION
1069 fscrypt_fname_free_buffer(&fname_crypto_str);
1070#endif
1071 return count;
1072}
1073
1074
1075/*
1076 * This function fills a red-black tree with information from a
1077 * directory. We start scanning the directory in hash order, starting
1078 * at start_hash and start_minor_hash.
1079 *
1080 * This function returns the number of entries inserted into the tree,
1081 * or a negative error code.
1082 */
1083int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1084 __u32 start_minor_hash, __u32 *next_hash)
1085{
1086 struct dx_hash_info hinfo;
1087 struct ext4_dir_entry_2 *de;
1088 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1089 struct inode *dir;
1090 ext4_lblk_t block;
1091 int count = 0;
1092 int ret, err;
1093 __u32 hashval;
1094 struct fscrypt_str tmp_str;
1095
1096 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1097 start_hash, start_minor_hash));
1098 dir = file_inode(dir_file);
1099 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1100 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1101 if (hinfo.hash_version <= DX_HASH_TEA)
1102 hinfo.hash_version +=
1103 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1104 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1105 if (ext4_has_inline_data(dir)) {
1106 int has_inline_data = 1;
1107 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1108 &hinfo, start_hash,
1109 start_minor_hash,
1110 &has_inline_data);
1111 if (has_inline_data) {
1112 *next_hash = ~0;
1113 return count;
1114 }
1115 }
1116 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1117 start_hash, start_minor_hash);
1118 *next_hash = ~0;
1119 return count;
1120 }
1121 hinfo.hash = start_hash;
1122 hinfo.minor_hash = 0;
1123 frame = dx_probe(NULL, dir, &hinfo, frames);
1124 if (IS_ERR(frame))
1125 return PTR_ERR(frame);
1126
1127 /* Add '.' and '..' from the htree header */
1128 if (!start_hash && !start_minor_hash) {
1129 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1130 tmp_str.name = de->name;
1131 tmp_str.len = de->name_len;
1132 err = ext4_htree_store_dirent(dir_file, 0, 0,
1133 de, &tmp_str);
1134 if (err != 0)
1135 goto errout;
1136 count++;
1137 }
1138 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1139 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1140 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1141 tmp_str.name = de->name;
1142 tmp_str.len = de->name_len;
1143 err = ext4_htree_store_dirent(dir_file, 2, 0,
1144 de, &tmp_str);
1145 if (err != 0)
1146 goto errout;
1147 count++;
1148 }
1149
1150 while (1) {
1151 if (fatal_signal_pending(current)) {
1152 err = -ERESTARTSYS;
1153 goto errout;
1154 }
1155 cond_resched();
1156 block = dx_get_block(frame->at);
1157 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1158 start_hash, start_minor_hash);
1159 if (ret < 0) {
1160 err = ret;
1161 goto errout;
1162 }
1163 count += ret;
1164 hashval = ~0;
1165 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1166 frame, frames, &hashval);
1167 *next_hash = hashval;
1168 if (ret < 0) {
1169 err = ret;
1170 goto errout;
1171 }
1172 /*
1173 * Stop if: (a) there are no more entries, or
1174 * (b) we have inserted at least one entry and the
1175 * next hash value is not a continuation
1176 */
1177 if ((ret == 0) ||
1178 (count && ((hashval & 1) == 0)))
1179 break;
1180 }
1181 dx_release(frames);
1182 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1183 "next hash: %x\n", count, *next_hash));
1184 return count;
1185errout:
1186 dx_release(frames);
1187 return (err);
1188}
1189
1190static inline int search_dirblock(struct buffer_head *bh,
1191 struct inode *dir,
1192 struct ext4_filename *fname,
1193 unsigned int offset,
1194 struct ext4_dir_entry_2 **res_dir)
1195{
1196 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1197 fname, offset, res_dir);
1198}
1199
1200/*
1201 * Directory block splitting, compacting
1202 */
1203
1204/*
1205 * Create map of hash values, offsets, and sizes, stored at end of block.
1206 * Returns number of entries mapped.
1207 */
1208static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1209 unsigned blocksize, struct dx_hash_info *hinfo,
1210 struct dx_map_entry *map_tail)
1211{
1212 int count = 0;
1213 char *base = (char *) de;
1214 struct dx_hash_info h = *hinfo;
1215
1216 while ((char *) de < base + blocksize) {
1217 if (de->name_len && de->inode) {
1218 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1219 map_tail--;
1220 map_tail->hash = h.hash;
1221 map_tail->offs = ((char *) de - base)>>2;
1222 map_tail->size = le16_to_cpu(de->rec_len);
1223 count++;
1224 cond_resched();
1225 }
1226 /* XXX: do we need to check rec_len == 0 case? -Chris */
1227 de = ext4_next_entry(de, blocksize);
1228 }
1229 return count;
1230}
1231
1232/* Sort map by hash value */
1233static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1234{
1235 struct dx_map_entry *p, *q, *top = map + count - 1;
1236 int more;
1237 /* Combsort until bubble sort doesn't suck */
1238 while (count > 2) {
1239 count = count*10/13;
1240 if (count - 9 < 2) /* 9, 10 -> 11 */
1241 count = 11;
1242 for (p = top, q = p - count; q >= map; p--, q--)
1243 if (p->hash < q->hash)
1244 swap(*p, *q);
1245 }
1246 /* Garden variety bubble sort */
1247 do {
1248 more = 0;
1249 q = top;
1250 while (q-- > map) {
1251 if (q[1].hash >= q[0].hash)
1252 continue;
1253 swap(*(q+1), *q);
1254 more = 1;
1255 }
1256 } while(more);
1257}
1258
1259static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1260{
1261 struct dx_entry *entries = frame->entries;
1262 struct dx_entry *old = frame->at, *new = old + 1;
1263 int count = dx_get_count(entries);
1264
1265 assert(count < dx_get_limit(entries));
1266 assert(old < entries + count);
1267 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1268 dx_set_hash(new, hash);
1269 dx_set_block(new, block);
1270 dx_set_count(entries, count + 1);
1271}
1272
1273#ifdef CONFIG_UNICODE
1274/*
1275 * Test whether a case-insensitive directory entry matches the filename
1276 * being searched for. If quick is set, assume the name being looked up
1277 * is already in the casefolded form.
1278 *
1279 * Returns: 0 if the directory entry matches, more than 0 if it
1280 * doesn't match or less than zero on error.
1281 */
1282int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1283 const struct qstr *entry, bool quick)
1284{
1285 const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1286 const struct unicode_map *um = sbi->s_encoding;
1287 int ret;
1288
1289 if (quick)
1290 ret = utf8_strncasecmp_folded(um, name, entry);
1291 else
1292 ret = utf8_strncasecmp(um, name, entry);
1293
1294 if (ret < 0) {
1295 /* Handle invalid character sequence as either an error
1296 * or as an opaque byte sequence.
1297 */
1298 if (ext4_has_strict_mode(sbi))
1299 return -EINVAL;
1300
1301 if (name->len != entry->len)
1302 return 1;
1303
1304 return !!memcmp(name->name, entry->name, name->len);
1305 }
1306
1307 return ret;
1308}
1309
1310void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1311 struct fscrypt_str *cf_name)
1312{
1313 int len;
1314
1315 if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
1316 cf_name->name = NULL;
1317 return;
1318 }
1319
1320 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1321 if (!cf_name->name)
1322 return;
1323
1324 len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1325 iname, cf_name->name,
1326 EXT4_NAME_LEN);
1327 if (len <= 0) {
1328 kfree(cf_name->name);
1329 cf_name->name = NULL;
1330 return;
1331 }
1332 cf_name->len = (unsigned) len;
1333
1334}
1335#endif
1336
1337/*
1338 * Test whether a directory entry matches the filename being searched for.
1339 *
1340 * Return: %true if the directory entry matches, otherwise %false.
1341 */
1342static inline bool ext4_match(const struct inode *parent,
1343 const struct ext4_filename *fname,
1344 const struct ext4_dir_entry_2 *de)
1345{
1346 struct fscrypt_name f;
1347#ifdef CONFIG_UNICODE
1348 const struct qstr entry = {.name = de->name, .len = de->name_len};
1349#endif
1350
1351 if (!de->inode)
1352 return false;
1353
1354 f.usr_fname = fname->usr_fname;
1355 f.disk_name = fname->disk_name;
1356#ifdef CONFIG_FS_ENCRYPTION
1357 f.crypto_buf = fname->crypto_buf;
1358#endif
1359
1360#ifdef CONFIG_UNICODE
1361 if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1362 if (fname->cf_name.name) {
1363 struct qstr cf = {.name = fname->cf_name.name,
1364 .len = fname->cf_name.len};
1365 return !ext4_ci_compare(parent, &cf, &entry, true);
1366 }
1367 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1368 false);
1369 }
1370#endif
1371
1372 return fscrypt_match_name(&f, de->name, de->name_len);
1373}
1374
1375/*
1376 * Returns 0 if not found, -1 on failure, and 1 on success
1377 */
1378int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1379 struct inode *dir, struct ext4_filename *fname,
1380 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1381{
1382 struct ext4_dir_entry_2 * de;
1383 char * dlimit;
1384 int de_len;
1385
1386 de = (struct ext4_dir_entry_2 *)search_buf;
1387 dlimit = search_buf + buf_size;
1388 while ((char *) de < dlimit) {
1389 /* this code is executed quadratically often */
1390 /* do minimal checking `by hand' */
1391 if ((char *) de + de->name_len <= dlimit &&
1392 ext4_match(dir, fname, de)) {
1393 /* found a match - just to be sure, do
1394 * a full check */
1395 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1396 bh->b_size, offset))
1397 return -1;
1398 *res_dir = de;
1399 return 1;
1400 }
1401 /* prevent looping on a bad block */
1402 de_len = ext4_rec_len_from_disk(de->rec_len,
1403 dir->i_sb->s_blocksize);
1404 if (de_len <= 0)
1405 return -1;
1406 offset += de_len;
1407 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1408 }
1409 return 0;
1410}
1411
1412static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1413 struct ext4_dir_entry *de)
1414{
1415 struct super_block *sb = dir->i_sb;
1416
1417 if (!is_dx(dir))
1418 return 0;
1419 if (block == 0)
1420 return 1;
1421 if (de->inode == 0 &&
1422 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1423 sb->s_blocksize)
1424 return 1;
1425 return 0;
1426}
1427
1428/*
1429 * __ext4_find_entry()
1430 *
1431 * finds an entry in the specified directory with the wanted name. It
1432 * returns the cache buffer in which the entry was found, and the entry
1433 * itself (as a parameter - res_dir). It does NOT read the inode of the
1434 * entry - you'll have to do that yourself if you want to.
1435 *
1436 * The returned buffer_head has ->b_count elevated. The caller is expected
1437 * to brelse() it when appropriate.
1438 */
1439static struct buffer_head *__ext4_find_entry(struct inode *dir,
1440 struct ext4_filename *fname,
1441 struct ext4_dir_entry_2 **res_dir,
1442 int *inlined)
1443{
1444 struct super_block *sb;
1445 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1446 struct buffer_head *bh, *ret = NULL;
1447 ext4_lblk_t start, block;
1448 const u8 *name = fname->usr_fname->name;
1449 size_t ra_max = 0; /* Number of bh's in the readahead
1450 buffer, bh_use[] */
1451 size_t ra_ptr = 0; /* Current index into readahead
1452 buffer */
1453 ext4_lblk_t nblocks;
1454 int i, namelen, retval;
1455
1456 *res_dir = NULL;
1457 sb = dir->i_sb;
1458 namelen = fname->usr_fname->len;
1459 if (namelen > EXT4_NAME_LEN)
1460 return NULL;
1461
1462 if (ext4_has_inline_data(dir)) {
1463 int has_inline_data = 1;
1464 ret = ext4_find_inline_entry(dir, fname, res_dir,
1465 &has_inline_data);
1466 if (has_inline_data) {
1467 if (inlined)
1468 *inlined = 1;
1469 goto cleanup_and_exit;
1470 }
1471 }
1472
1473 if ((namelen <= 2) && (name[0] == '.') &&
1474 (name[1] == '.' || name[1] == '\0')) {
1475 /*
1476 * "." or ".." will only be in the first block
1477 * NFS may look up ".."; "." should be handled by the VFS
1478 */
1479 block = start = 0;
1480 nblocks = 1;
1481 goto restart;
1482 }
1483 if (is_dx(dir)) {
1484 ret = ext4_dx_find_entry(dir, fname, res_dir);
1485 /*
1486 * On success, or if the error was file not found,
1487 * return. Otherwise, fall back to doing a search the
1488 * old fashioned way.
1489 */
1490 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1491 goto cleanup_and_exit;
1492 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1493 "falling back\n"));
1494 ret = NULL;
1495 }
1496 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1497 if (!nblocks) {
1498 ret = NULL;
1499 goto cleanup_and_exit;
1500 }
1501 start = EXT4_I(dir)->i_dir_start_lookup;
1502 if (start >= nblocks)
1503 start = 0;
1504 block = start;
1505restart:
1506 do {
1507 /*
1508 * We deal with the read-ahead logic here.
1509 */
1510 if (ra_ptr >= ra_max) {
1511 /* Refill the readahead buffer */
1512 ra_ptr = 0;
1513 if (block < start)
1514 ra_max = start - block;
1515 else
1516 ra_max = nblocks - block;
1517 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1518 retval = ext4_bread_batch(dir, block, ra_max,
1519 false /* wait */, bh_use);
1520 if (retval) {
1521 ret = ERR_PTR(retval);
1522 ra_max = 0;
1523 goto cleanup_and_exit;
1524 }
1525 }
1526 if ((bh = bh_use[ra_ptr++]) == NULL)
1527 goto next;
1528 wait_on_buffer(bh);
1529 if (!buffer_uptodate(bh)) {
1530 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1531 (unsigned long) block);
1532 brelse(bh);
1533 ret = ERR_PTR(-EIO);
1534 goto cleanup_and_exit;
1535 }
1536 if (!buffer_verified(bh) &&
1537 !is_dx_internal_node(dir, block,
1538 (struct ext4_dir_entry *)bh->b_data) &&
1539 !ext4_dirblock_csum_verify(dir, bh)) {
1540 EXT4_ERROR_INODE(dir, "checksumming directory "
1541 "block %lu", (unsigned long)block);
1542 brelse(bh);
1543 ret = ERR_PTR(-EFSBADCRC);
1544 goto cleanup_and_exit;
1545 }
1546 set_buffer_verified(bh);
1547 i = search_dirblock(bh, dir, fname,
1548 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1549 if (i == 1) {
1550 EXT4_I(dir)->i_dir_start_lookup = block;
1551 ret = bh;
1552 goto cleanup_and_exit;
1553 } else {
1554 brelse(bh);
1555 if (i < 0)
1556 goto cleanup_and_exit;
1557 }
1558 next:
1559 if (++block >= nblocks)
1560 block = 0;
1561 } while (block != start);
1562
1563 /*
1564 * If the directory has grown while we were searching, then
1565 * search the last part of the directory before giving up.
1566 */
1567 block = nblocks;
1568 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1569 if (block < nblocks) {
1570 start = 0;
1571 goto restart;
1572 }
1573
1574cleanup_and_exit:
1575 /* Clean up the read-ahead blocks */
1576 for (; ra_ptr < ra_max; ra_ptr++)
1577 brelse(bh_use[ra_ptr]);
1578 return ret;
1579}
1580
1581static struct buffer_head *ext4_find_entry(struct inode *dir,
1582 const struct qstr *d_name,
1583 struct ext4_dir_entry_2 **res_dir,
1584 int *inlined)
1585{
1586 int err;
1587 struct ext4_filename fname;
1588 struct buffer_head *bh;
1589
1590 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1591 if (err == -ENOENT)
1592 return NULL;
1593 if (err)
1594 return ERR_PTR(err);
1595
1596 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1597
1598 ext4_fname_free_filename(&fname);
1599 return bh;
1600}
1601
1602static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1603 struct dentry *dentry,
1604 struct ext4_dir_entry_2 **res_dir)
1605{
1606 int err;
1607 struct ext4_filename fname;
1608 struct buffer_head *bh;
1609
1610 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1611 if (err == -ENOENT)
1612 return NULL;
1613 if (err)
1614 return ERR_PTR(err);
1615
1616 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1617
1618 ext4_fname_free_filename(&fname);
1619 return bh;
1620}
1621
1622static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1623 struct ext4_filename *fname,
1624 struct ext4_dir_entry_2 **res_dir)
1625{
1626 struct super_block * sb = dir->i_sb;
1627 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1628 struct buffer_head *bh;
1629 ext4_lblk_t block;
1630 int retval;
1631
1632#ifdef CONFIG_FS_ENCRYPTION
1633 *res_dir = NULL;
1634#endif
1635 frame = dx_probe(fname, dir, NULL, frames);
1636 if (IS_ERR(frame))
1637 return (struct buffer_head *) frame;
1638 do {
1639 block = dx_get_block(frame->at);
1640 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1641 if (IS_ERR(bh))
1642 goto errout;
1643
1644 retval = search_dirblock(bh, dir, fname,
1645 block << EXT4_BLOCK_SIZE_BITS(sb),
1646 res_dir);
1647 if (retval == 1)
1648 goto success;
1649 brelse(bh);
1650 if (retval == -1) {
1651 bh = ERR_PTR(ERR_BAD_DX_DIR);
1652 goto errout;
1653 }
1654
1655 /* Check to see if we should continue to search */
1656 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1657 frames, NULL);
1658 if (retval < 0) {
1659 ext4_warning_inode(dir,
1660 "error %d reading directory index block",
1661 retval);
1662 bh = ERR_PTR(retval);
1663 goto errout;
1664 }
1665 } while (retval == 1);
1666
1667 bh = NULL;
1668errout:
1669 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1670success:
1671 dx_release(frames);
1672 return bh;
1673}
1674
1675static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1676{
1677 struct inode *inode;
1678 struct ext4_dir_entry_2 *de;
1679 struct buffer_head *bh;
1680
1681 if (dentry->d_name.len > EXT4_NAME_LEN)
1682 return ERR_PTR(-ENAMETOOLONG);
1683
1684 bh = ext4_lookup_entry(dir, dentry, &de);
1685 if (IS_ERR(bh))
1686 return ERR_CAST(bh);
1687 inode = NULL;
1688 if (bh) {
1689 __u32 ino = le32_to_cpu(de->inode);
1690 brelse(bh);
1691 if (!ext4_valid_inum(dir->i_sb, ino)) {
1692 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1693 return ERR_PTR(-EFSCORRUPTED);
1694 }
1695 if (unlikely(ino == dir->i_ino)) {
1696 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1697 dentry);
1698 return ERR_PTR(-EFSCORRUPTED);
1699 }
1700 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1701 if (inode == ERR_PTR(-ESTALE)) {
1702 EXT4_ERROR_INODE(dir,
1703 "deleted inode referenced: %u",
1704 ino);
1705 return ERR_PTR(-EFSCORRUPTED);
1706 }
1707 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1708 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1709 !fscrypt_has_permitted_context(dir, inode)) {
1710 ext4_warning(inode->i_sb,
1711 "Inconsistent encryption contexts: %lu/%lu",
1712 dir->i_ino, inode->i_ino);
1713 iput(inode);
1714 return ERR_PTR(-EPERM);
1715 }
1716 }
1717
1718#ifdef CONFIG_UNICODE
1719 if (!inode && IS_CASEFOLDED(dir)) {
1720 /* Eventually we want to call d_add_ci(dentry, NULL)
1721 * for negative dentries in the encoding case as
1722 * well. For now, prevent the negative dentry
1723 * from being cached.
1724 */
1725 return NULL;
1726 }
1727#endif
1728 return d_splice_alias(inode, dentry);
1729}
1730
1731
1732struct dentry *ext4_get_parent(struct dentry *child)
1733{
1734 __u32 ino;
1735 static const struct qstr dotdot = QSTR_INIT("..", 2);
1736 struct ext4_dir_entry_2 * de;
1737 struct buffer_head *bh;
1738
1739 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1740 if (IS_ERR(bh))
1741 return ERR_CAST(bh);
1742 if (!bh)
1743 return ERR_PTR(-ENOENT);
1744 ino = le32_to_cpu(de->inode);
1745 brelse(bh);
1746
1747 if (!ext4_valid_inum(child->d_sb, ino)) {
1748 EXT4_ERROR_INODE(d_inode(child),
1749 "bad parent inode number: %u", ino);
1750 return ERR_PTR(-EFSCORRUPTED);
1751 }
1752
1753 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1754}
1755
1756/*
1757 * Move count entries from end of map between two memory locations.
1758 * Returns pointer to last entry moved.
1759 */
1760static struct ext4_dir_entry_2 *
1761dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1762 unsigned blocksize)
1763{
1764 unsigned rec_len = 0;
1765
1766 while (count--) {
1767 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1768 (from + (map->offs<<2));
1769 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1770 memcpy (to, de, rec_len);
1771 ((struct ext4_dir_entry_2 *) to)->rec_len =
1772 ext4_rec_len_to_disk(rec_len, blocksize);
1773 de->inode = 0;
1774 map++;
1775 to += rec_len;
1776 }
1777 return (struct ext4_dir_entry_2 *) (to - rec_len);
1778}
1779
1780/*
1781 * Compact each dir entry in the range to the minimal rec_len.
1782 * Returns pointer to last entry in range.
1783 */
1784static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1785{
1786 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1787 unsigned rec_len = 0;
1788
1789 prev = to = de;
1790 while ((char*)de < base + blocksize) {
1791 next = ext4_next_entry(de, blocksize);
1792 if (de->inode && de->name_len) {
1793 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1794 if (de > to)
1795 memmove(to, de, rec_len);
1796 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1797 prev = to;
1798 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1799 }
1800 de = next;
1801 }
1802 return prev;
1803}
1804
1805/*
1806 * Split a full leaf block to make room for a new dir entry.
1807 * Allocate a new block, and move entries so that they are approx. equally full.
1808 * Returns pointer to de in block into which the new entry will be inserted.
1809 */
1810static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1811 struct buffer_head **bh,struct dx_frame *frame,
1812 struct dx_hash_info *hinfo)
1813{
1814 unsigned blocksize = dir->i_sb->s_blocksize;
1815 unsigned count, continued;
1816 struct buffer_head *bh2;
1817 ext4_lblk_t newblock;
1818 u32 hash2;
1819 struct dx_map_entry *map;
1820 char *data1 = (*bh)->b_data, *data2;
1821 unsigned split, move, size;
1822 struct ext4_dir_entry_2 *de = NULL, *de2;
1823 int csum_size = 0;
1824 int err = 0, i;
1825
1826 if (ext4_has_metadata_csum(dir->i_sb))
1827 csum_size = sizeof(struct ext4_dir_entry_tail);
1828
1829 bh2 = ext4_append(handle, dir, &newblock);
1830 if (IS_ERR(bh2)) {
1831 brelse(*bh);
1832 *bh = NULL;
1833 return (struct ext4_dir_entry_2 *) bh2;
1834 }
1835
1836 BUFFER_TRACE(*bh, "get_write_access");
1837 err = ext4_journal_get_write_access(handle, *bh);
1838 if (err)
1839 goto journal_error;
1840
1841 BUFFER_TRACE(frame->bh, "get_write_access");
1842 err = ext4_journal_get_write_access(handle, frame->bh);
1843 if (err)
1844 goto journal_error;
1845
1846 data2 = bh2->b_data;
1847
1848 /* create map in the end of data2 block */
1849 map = (struct dx_map_entry *) (data2 + blocksize);
1850 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1851 blocksize, hinfo, map);
1852 map -= count;
1853 dx_sort_map(map, count);
1854 /* Split the existing block in the middle, size-wise */
1855 size = 0;
1856 move = 0;
1857 for (i = count-1; i >= 0; i--) {
1858 /* is more than half of this entry in 2nd half of the block? */
1859 if (size + map[i].size/2 > blocksize/2)
1860 break;
1861 size += map[i].size;
1862 move++;
1863 }
1864 /* map index at which we will split */
1865 split = count - move;
1866 hash2 = map[split].hash;
1867 continued = hash2 == map[split - 1].hash;
1868 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1869 (unsigned long)dx_get_block(frame->at),
1870 hash2, split, count-split));
1871
1872 /* Fancy dance to stay within two buffers */
1873 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1874 blocksize);
1875 de = dx_pack_dirents(data1, blocksize);
1876 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1877 (char *) de,
1878 blocksize);
1879 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1880 (char *) de2,
1881 blocksize);
1882 if (csum_size) {
1883 ext4_initialize_dirent_tail(*bh, blocksize);
1884 ext4_initialize_dirent_tail(bh2, blocksize);
1885 }
1886
1887 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1888 blocksize, 1));
1889 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1890 blocksize, 1));
1891
1892 /* Which block gets the new entry? */
1893 if (hinfo->hash >= hash2) {
1894 swap(*bh, bh2);
1895 de = de2;
1896 }
1897 dx_insert_block(frame, hash2 + continued, newblock);
1898 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1899 if (err)
1900 goto journal_error;
1901 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1902 if (err)
1903 goto journal_error;
1904 brelse(bh2);
1905 dxtrace(dx_show_index("frame", frame->entries));
1906 return de;
1907
1908journal_error:
1909 brelse(*bh);
1910 brelse(bh2);
1911 *bh = NULL;
1912 ext4_std_error(dir->i_sb, err);
1913 return ERR_PTR(err);
1914}
1915
1916int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1917 struct buffer_head *bh,
1918 void *buf, int buf_size,
1919 struct ext4_filename *fname,
1920 struct ext4_dir_entry_2 **dest_de)
1921{
1922 struct ext4_dir_entry_2 *de;
1923 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1924 int nlen, rlen;
1925 unsigned int offset = 0;
1926 char *top;
1927
1928 de = (struct ext4_dir_entry_2 *)buf;
1929 top = buf + buf_size - reclen;
1930 while ((char *) de <= top) {
1931 if (ext4_check_dir_entry(dir, NULL, de, bh,
1932 buf, buf_size, offset))
1933 return -EFSCORRUPTED;
1934 if (ext4_match(dir, fname, de))
1935 return -EEXIST;
1936 nlen = EXT4_DIR_REC_LEN(de->name_len);
1937 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1938 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1939 break;
1940 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1941 offset += rlen;
1942 }
1943 if ((char *) de > top)
1944 return -ENOSPC;
1945
1946 *dest_de = de;
1947 return 0;
1948}
1949
1950void ext4_insert_dentry(struct inode *inode,
1951 struct ext4_dir_entry_2 *de,
1952 int buf_size,
1953 struct ext4_filename *fname)
1954{
1955
1956 int nlen, rlen;
1957
1958 nlen = EXT4_DIR_REC_LEN(de->name_len);
1959 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1960 if (de->inode) {
1961 struct ext4_dir_entry_2 *de1 =
1962 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1963 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1964 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1965 de = de1;
1966 }
1967 de->file_type = EXT4_FT_UNKNOWN;
1968 de->inode = cpu_to_le32(inode->i_ino);
1969 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1970 de->name_len = fname_len(fname);
1971 memcpy(de->name, fname_name(fname), fname_len(fname));
1972}
1973
1974/*
1975 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1976 * it points to a directory entry which is guaranteed to be large
1977 * enough for new directory entry. If de is NULL, then
1978 * add_dirent_to_buf will attempt search the directory block for
1979 * space. It will return -ENOSPC if no space is available, and -EIO
1980 * and -EEXIST if directory entry already exists.
1981 */
1982static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1983 struct inode *dir,
1984 struct inode *inode, struct ext4_dir_entry_2 *de,
1985 struct buffer_head *bh)
1986{
1987 unsigned int blocksize = dir->i_sb->s_blocksize;
1988 int csum_size = 0;
1989 int err;
1990
1991 if (ext4_has_metadata_csum(inode->i_sb))
1992 csum_size = sizeof(struct ext4_dir_entry_tail);
1993
1994 if (!de) {
1995 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1996 blocksize - csum_size, fname, &de);
1997 if (err)
1998 return err;
1999 }
2000 BUFFER_TRACE(bh, "get_write_access");
2001 err = ext4_journal_get_write_access(handle, bh);
2002 if (err) {
2003 ext4_std_error(dir->i_sb, err);
2004 return err;
2005 }
2006
2007 /* By now the buffer is marked for journaling */
2008 ext4_insert_dentry(inode, de, blocksize, fname);
2009
2010 /*
2011 * XXX shouldn't update any times until successful
2012 * completion of syscall, but too many callers depend
2013 * on this.
2014 *
2015 * XXX similarly, too many callers depend on
2016 * ext4_new_inode() setting the times, but error
2017 * recovery deletes the inode, so the worst that can
2018 * happen is that the times are slightly out of date
2019 * and/or different from the directory change time.
2020 */
2021 dir->i_mtime = dir->i_ctime = current_time(dir);
2022 ext4_update_dx_flag(dir);
2023 inode_inc_iversion(dir);
2024 ext4_mark_inode_dirty(handle, dir);
2025 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2026 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2027 if (err)
2028 ext4_std_error(dir->i_sb, err);
2029 return 0;
2030}
2031
2032/*
2033 * This converts a one block unindexed directory to a 3 block indexed
2034 * directory, and adds the dentry to the indexed directory.
2035 */
2036static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2037 struct inode *dir,
2038 struct inode *inode, struct buffer_head *bh)
2039{
2040 struct buffer_head *bh2;
2041 struct dx_root *root;
2042 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2043 struct dx_entry *entries;
2044 struct ext4_dir_entry_2 *de, *de2;
2045 char *data2, *top;
2046 unsigned len;
2047 int retval;
2048 unsigned blocksize;
2049 ext4_lblk_t block;
2050 struct fake_dirent *fde;
2051 int csum_size = 0;
2052
2053 if (ext4_has_metadata_csum(inode->i_sb))
2054 csum_size = sizeof(struct ext4_dir_entry_tail);
2055
2056 blocksize = dir->i_sb->s_blocksize;
2057 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2058 BUFFER_TRACE(bh, "get_write_access");
2059 retval = ext4_journal_get_write_access(handle, bh);
2060 if (retval) {
2061 ext4_std_error(dir->i_sb, retval);
2062 brelse(bh);
2063 return retval;
2064 }
2065 root = (struct dx_root *) bh->b_data;
2066
2067 /* The 0th block becomes the root, move the dirents out */
2068 fde = &root->dotdot;
2069 de = (struct ext4_dir_entry_2 *)((char *)fde +
2070 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2071 if ((char *) de >= (((char *) root) + blocksize)) {
2072 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2073 brelse(bh);
2074 return -EFSCORRUPTED;
2075 }
2076 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2077
2078 /* Allocate new block for the 0th block's dirents */
2079 bh2 = ext4_append(handle, dir, &block);
2080 if (IS_ERR(bh2)) {
2081 brelse(bh);
2082 return PTR_ERR(bh2);
2083 }
2084 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2085 data2 = bh2->b_data;
2086
2087 memcpy(data2, de, len);
2088 de = (struct ext4_dir_entry_2 *) data2;
2089 top = data2 + len;
2090 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2091 de = de2;
2092 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2093 (char *) de, blocksize);
2094
2095 if (csum_size)
2096 ext4_initialize_dirent_tail(bh2, blocksize);
2097
2098 /* Initialize the root; the dot dirents already exist */
2099 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2100 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2101 blocksize);
2102 memset (&root->info, 0, sizeof(root->info));
2103 root->info.info_length = sizeof(root->info);
2104 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2105 entries = root->entries;
2106 dx_set_block(entries, 1);
2107 dx_set_count(entries, 1);
2108 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2109
2110 /* Initialize as for dx_probe */
2111 fname->hinfo.hash_version = root->info.hash_version;
2112 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2113 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2114 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2115 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2116
2117 memset(frames, 0, sizeof(frames));
2118 frame = frames;
2119 frame->entries = entries;
2120 frame->at = entries;
2121 frame->bh = bh;
2122
2123 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2124 if (retval)
2125 goto out_frames;
2126 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2127 if (retval)
2128 goto out_frames;
2129
2130 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2131 if (IS_ERR(de)) {
2132 retval = PTR_ERR(de);
2133 goto out_frames;
2134 }
2135
2136 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2137out_frames:
2138 /*
2139 * Even if the block split failed, we have to properly write
2140 * out all the changes we did so far. Otherwise we can end up
2141 * with corrupted filesystem.
2142 */
2143 if (retval)
2144 ext4_mark_inode_dirty(handle, dir);
2145 dx_release(frames);
2146 brelse(bh2);
2147 return retval;
2148}
2149
2150/*
2151 * ext4_add_entry()
2152 *
2153 * adds a file entry to the specified directory, using the same
2154 * semantics as ext4_find_entry(). It returns NULL if it failed.
2155 *
2156 * NOTE!! The inode part of 'de' is left at 0 - which means you
2157 * may not sleep between calling this and putting something into
2158 * the entry, as someone else might have used it while you slept.
2159 */
2160static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2161 struct inode *inode)
2162{
2163 struct inode *dir = d_inode(dentry->d_parent);
2164 struct buffer_head *bh = NULL;
2165 struct ext4_dir_entry_2 *de;
2166 struct super_block *sb;
2167 struct ext4_sb_info *sbi;
2168 struct ext4_filename fname;
2169 int retval;
2170 int dx_fallback=0;
2171 unsigned blocksize;
2172 ext4_lblk_t block, blocks;
2173 int csum_size = 0;
2174
2175 if (ext4_has_metadata_csum(inode->i_sb))
2176 csum_size = sizeof(struct ext4_dir_entry_tail);
2177
2178 sb = dir->i_sb;
2179 sbi = EXT4_SB(sb);
2180 blocksize = sb->s_blocksize;
2181 if (!dentry->d_name.len)
2182 return -EINVAL;
2183
2184#ifdef CONFIG_UNICODE
2185 if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2186 sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2187 return -EINVAL;
2188#endif
2189
2190 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2191 if (retval)
2192 return retval;
2193
2194 if (ext4_has_inline_data(dir)) {
2195 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2196 if (retval < 0)
2197 goto out;
2198 if (retval == 1) {
2199 retval = 0;
2200 goto out;
2201 }
2202 }
2203
2204 if (is_dx(dir)) {
2205 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2206 if (!retval || (retval != ERR_BAD_DX_DIR))
2207 goto out;
2208 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2209 dx_fallback++;
2210 ext4_mark_inode_dirty(handle, dir);
2211 }
2212 blocks = dir->i_size >> sb->s_blocksize_bits;
2213 for (block = 0; block < blocks; block++) {
2214 bh = ext4_read_dirblock(dir, block, DIRENT);
2215 if (bh == NULL) {
2216 bh = ext4_bread(handle, dir, block,
2217 EXT4_GET_BLOCKS_CREATE);
2218 goto add_to_new_block;
2219 }
2220 if (IS_ERR(bh)) {
2221 retval = PTR_ERR(bh);
2222 bh = NULL;
2223 goto out;
2224 }
2225 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2226 NULL, bh);
2227 if (retval != -ENOSPC)
2228 goto out;
2229
2230 if (blocks == 1 && !dx_fallback &&
2231 ext4_has_feature_dir_index(sb)) {
2232 retval = make_indexed_dir(handle, &fname, dir,
2233 inode, bh);
2234 bh = NULL; /* make_indexed_dir releases bh */
2235 goto out;
2236 }
2237 brelse(bh);
2238 }
2239 bh = ext4_append(handle, dir, &block);
2240add_to_new_block:
2241 if (IS_ERR(bh)) {
2242 retval = PTR_ERR(bh);
2243 bh = NULL;
2244 goto out;
2245 }
2246 de = (struct ext4_dir_entry_2 *) bh->b_data;
2247 de->inode = 0;
2248 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2249
2250 if (csum_size)
2251 ext4_initialize_dirent_tail(bh, blocksize);
2252
2253 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2254out:
2255 ext4_fname_free_filename(&fname);
2256 brelse(bh);
2257 if (retval == 0)
2258 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2259 return retval;
2260}
2261
2262/*
2263 * Returns 0 for success, or a negative error value
2264 */
2265static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2266 struct inode *dir, struct inode *inode)
2267{
2268 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2269 struct dx_entry *entries, *at;
2270 struct buffer_head *bh;
2271 struct super_block *sb = dir->i_sb;
2272 struct ext4_dir_entry_2 *de;
2273 int restart;
2274 int err;
2275
2276again:
2277 restart = 0;
2278 frame = dx_probe(fname, dir, NULL, frames);
2279 if (IS_ERR(frame))
2280 return PTR_ERR(frame);
2281 entries = frame->entries;
2282 at = frame->at;
2283 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2284 if (IS_ERR(bh)) {
2285 err = PTR_ERR(bh);
2286 bh = NULL;
2287 goto cleanup;
2288 }
2289
2290 BUFFER_TRACE(bh, "get_write_access");
2291 err = ext4_journal_get_write_access(handle, bh);
2292 if (err)
2293 goto journal_error;
2294
2295 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2296 if (err != -ENOSPC)
2297 goto cleanup;
2298
2299 err = 0;
2300 /* Block full, should compress but for now just split */
2301 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2302 dx_get_count(entries), dx_get_limit(entries)));
2303 /* Need to split index? */
2304 if (dx_get_count(entries) == dx_get_limit(entries)) {
2305 ext4_lblk_t newblock;
2306 int levels = frame - frames + 1;
2307 unsigned int icount;
2308 int add_level = 1;
2309 struct dx_entry *entries2;
2310 struct dx_node *node2;
2311 struct buffer_head *bh2;
2312
2313 while (frame > frames) {
2314 if (dx_get_count((frame - 1)->entries) <
2315 dx_get_limit((frame - 1)->entries)) {
2316 add_level = 0;
2317 break;
2318 }
2319 frame--; /* split higher index block */
2320 at = frame->at;
2321 entries = frame->entries;
2322 restart = 1;
2323 }
2324 if (add_level && levels == ext4_dir_htree_level(sb)) {
2325 ext4_warning(sb, "Directory (ino: %lu) index full, "
2326 "reach max htree level :%d",
2327 dir->i_ino, levels);
2328 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2329 ext4_warning(sb, "Large directory feature is "
2330 "not enabled on this "
2331 "filesystem");
2332 }
2333 err = -ENOSPC;
2334 goto cleanup;
2335 }
2336 icount = dx_get_count(entries);
2337 bh2 = ext4_append(handle, dir, &newblock);
2338 if (IS_ERR(bh2)) {
2339 err = PTR_ERR(bh2);
2340 goto cleanup;
2341 }
2342 node2 = (struct dx_node *)(bh2->b_data);
2343 entries2 = node2->entries;
2344 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2345 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2346 sb->s_blocksize);
2347 BUFFER_TRACE(frame->bh, "get_write_access");
2348 err = ext4_journal_get_write_access(handle, frame->bh);
2349 if (err)
2350 goto journal_error;
2351 if (!add_level) {
2352 unsigned icount1 = icount/2, icount2 = icount - icount1;
2353 unsigned hash2 = dx_get_hash(entries + icount1);
2354 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2355 icount1, icount2));
2356
2357 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2358 err = ext4_journal_get_write_access(handle,
2359 (frame - 1)->bh);
2360 if (err)
2361 goto journal_error;
2362
2363 memcpy((char *) entries2, (char *) (entries + icount1),
2364 icount2 * sizeof(struct dx_entry));
2365 dx_set_count(entries, icount1);
2366 dx_set_count(entries2, icount2);
2367 dx_set_limit(entries2, dx_node_limit(dir));
2368
2369 /* Which index block gets the new entry? */
2370 if (at - entries >= icount1) {
2371 frame->at = at = at - entries - icount1 + entries2;
2372 frame->entries = entries = entries2;
2373 swap(frame->bh, bh2);
2374 }
2375 dx_insert_block((frame - 1), hash2, newblock);
2376 dxtrace(dx_show_index("node", frame->entries));
2377 dxtrace(dx_show_index("node",
2378 ((struct dx_node *) bh2->b_data)->entries));
2379 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2380 if (err)
2381 goto journal_error;
2382 brelse (bh2);
2383 err = ext4_handle_dirty_dx_node(handle, dir,
2384 (frame - 1)->bh);
2385 if (err)
2386 goto journal_error;
2387 if (restart) {
2388 err = ext4_handle_dirty_dx_node(handle, dir,
2389 frame->bh);
2390 goto journal_error;
2391 }
2392 } else {
2393 struct dx_root *dxroot;
2394 memcpy((char *) entries2, (char *) entries,
2395 icount * sizeof(struct dx_entry));
2396 dx_set_limit(entries2, dx_node_limit(dir));
2397
2398 /* Set up root */
2399 dx_set_count(entries, 1);
2400 dx_set_block(entries + 0, newblock);
2401 dxroot = (struct dx_root *)frames[0].bh->b_data;
2402 dxroot->info.indirect_levels += 1;
2403 dxtrace(printk(KERN_DEBUG
2404 "Creating %d level index...\n",
2405 dxroot->info.indirect_levels));
2406 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2407 if (err)
2408 goto journal_error;
2409 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2410 brelse(bh2);
2411 restart = 1;
2412 goto journal_error;
2413 }
2414 }
2415 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2416 if (IS_ERR(de)) {
2417 err = PTR_ERR(de);
2418 goto cleanup;
2419 }
2420 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2421 goto cleanup;
2422
2423journal_error:
2424 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2425cleanup:
2426 brelse(bh);
2427 dx_release(frames);
2428 /* @restart is true means htree-path has been changed, we need to
2429 * repeat dx_probe() to find out valid htree-path
2430 */
2431 if (restart && err == 0)
2432 goto again;
2433 return err;
2434}
2435
2436/*
2437 * ext4_generic_delete_entry deletes a directory entry by merging it
2438 * with the previous entry
2439 */
2440int ext4_generic_delete_entry(handle_t *handle,
2441 struct inode *dir,
2442 struct ext4_dir_entry_2 *de_del,
2443 struct buffer_head *bh,
2444 void *entry_buf,
2445 int buf_size,
2446 int csum_size)
2447{
2448 struct ext4_dir_entry_2 *de, *pde;
2449 unsigned int blocksize = dir->i_sb->s_blocksize;
2450 int i;
2451
2452 i = 0;
2453 pde = NULL;
2454 de = (struct ext4_dir_entry_2 *)entry_buf;
2455 while (i < buf_size - csum_size) {
2456 if (ext4_check_dir_entry(dir, NULL, de, bh,
2457 bh->b_data, bh->b_size, i))
2458 return -EFSCORRUPTED;
2459 if (de == de_del) {
2460 if (pde)
2461 pde->rec_len = ext4_rec_len_to_disk(
2462 ext4_rec_len_from_disk(pde->rec_len,
2463 blocksize) +
2464 ext4_rec_len_from_disk(de->rec_len,
2465 blocksize),
2466 blocksize);
2467 else
2468 de->inode = 0;
2469 inode_inc_iversion(dir);
2470 return 0;
2471 }
2472 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2473 pde = de;
2474 de = ext4_next_entry(de, blocksize);
2475 }
2476 return -ENOENT;
2477}
2478
2479static int ext4_delete_entry(handle_t *handle,
2480 struct inode *dir,
2481 struct ext4_dir_entry_2 *de_del,
2482 struct buffer_head *bh)
2483{
2484 int err, csum_size = 0;
2485
2486 if (ext4_has_inline_data(dir)) {
2487 int has_inline_data = 1;
2488 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2489 &has_inline_data);
2490 if (has_inline_data)
2491 return err;
2492 }
2493
2494 if (ext4_has_metadata_csum(dir->i_sb))
2495 csum_size = sizeof(struct ext4_dir_entry_tail);
2496
2497 BUFFER_TRACE(bh, "get_write_access");
2498 err = ext4_journal_get_write_access(handle, bh);
2499 if (unlikely(err))
2500 goto out;
2501
2502 err = ext4_generic_delete_entry(handle, dir, de_del,
2503 bh, bh->b_data,
2504 dir->i_sb->s_blocksize, csum_size);
2505 if (err)
2506 goto out;
2507
2508 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2509 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2510 if (unlikely(err))
2511 goto out;
2512
2513 return 0;
2514out:
2515 if (err != -ENOENT)
2516 ext4_std_error(dir->i_sb, err);
2517 return err;
2518}
2519
2520/*
2521 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2522 * since this indicates that nlinks count was previously 1 to avoid overflowing
2523 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2524 * that subdirectory link counts are not being maintained accurately.
2525 *
2526 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2527 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2528 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2529 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2530 */
2531static void ext4_inc_count(handle_t *handle, struct inode *inode)
2532{
2533 inc_nlink(inode);
2534 if (is_dx(inode) &&
2535 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2536 set_nlink(inode, 1);
2537}
2538
2539/*
2540 * If a directory had nlink == 1, then we should let it be 1. This indicates
2541 * directory has >EXT4_LINK_MAX subdirs.
2542 */
2543static void ext4_dec_count(handle_t *handle, struct inode *inode)
2544{
2545 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2546 drop_nlink(inode);
2547}
2548
2549
2550static int ext4_add_nondir(handle_t *handle,
2551 struct dentry *dentry, struct inode *inode)
2552{
2553 int err = ext4_add_entry(handle, dentry, inode);
2554 if (!err) {
2555 ext4_mark_inode_dirty(handle, inode);
2556 d_instantiate_new(dentry, inode);
2557 return 0;
2558 }
2559 drop_nlink(inode);
2560 unlock_new_inode(inode);
2561 iput(inode);
2562 return err;
2563}
2564
2565/*
2566 * By the time this is called, we already have created
2567 * the directory cache entry for the new file, but it
2568 * is so far negative - it has no inode.
2569 *
2570 * If the create succeeds, we fill in the inode information
2571 * with d_instantiate().
2572 */
2573static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2574 bool excl)
2575{
2576 handle_t *handle;
2577 struct inode *inode;
2578 int err, credits, retries = 0;
2579
2580 err = dquot_initialize(dir);
2581 if (err)
2582 return err;
2583
2584 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2585 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2586retry:
2587 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2588 NULL, EXT4_HT_DIR, credits);
2589 handle = ext4_journal_current_handle();
2590 err = PTR_ERR(inode);
2591 if (!IS_ERR(inode)) {
2592 inode->i_op = &ext4_file_inode_operations;
2593 inode->i_fop = &ext4_file_operations;
2594 ext4_set_aops(inode);
2595 err = ext4_add_nondir(handle, dentry, inode);
2596 if (!err && IS_DIRSYNC(dir))
2597 ext4_handle_sync(handle);
2598 }
2599 if (handle)
2600 ext4_journal_stop(handle);
2601 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2602 goto retry;
2603 return err;
2604}
2605
2606static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2607 umode_t mode, dev_t rdev)
2608{
2609 handle_t *handle;
2610 struct inode *inode;
2611 int err, credits, retries = 0;
2612
2613 err = dquot_initialize(dir);
2614 if (err)
2615 return err;
2616
2617 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2618 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2619retry:
2620 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2621 NULL, EXT4_HT_DIR, credits);
2622 handle = ext4_journal_current_handle();
2623 err = PTR_ERR(inode);
2624 if (!IS_ERR(inode)) {
2625 init_special_inode(inode, inode->i_mode, rdev);
2626 inode->i_op = &ext4_special_inode_operations;
2627 err = ext4_add_nondir(handle, dentry, inode);
2628 if (!err && IS_DIRSYNC(dir))
2629 ext4_handle_sync(handle);
2630 }
2631 if (handle)
2632 ext4_journal_stop(handle);
2633 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2634 goto retry;
2635 return err;
2636}
2637
2638static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2639{
2640 handle_t *handle;
2641 struct inode *inode;
2642 int err, retries = 0;
2643
2644 err = dquot_initialize(dir);
2645 if (err)
2646 return err;
2647
2648retry:
2649 inode = ext4_new_inode_start_handle(dir, mode,
2650 NULL, 0, NULL,
2651 EXT4_HT_DIR,
2652 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2653 4 + EXT4_XATTR_TRANS_BLOCKS);
2654 handle = ext4_journal_current_handle();
2655 err = PTR_ERR(inode);
2656 if (!IS_ERR(inode)) {
2657 inode->i_op = &ext4_file_inode_operations;
2658 inode->i_fop = &ext4_file_operations;
2659 ext4_set_aops(inode);
2660 d_tmpfile(dentry, inode);
2661 err = ext4_orphan_add(handle, inode);
2662 if (err)
2663 goto err_unlock_inode;
2664 mark_inode_dirty(inode);
2665 unlock_new_inode(inode);
2666 }
2667 if (handle)
2668 ext4_journal_stop(handle);
2669 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2670 goto retry;
2671 return err;
2672err_unlock_inode:
2673 ext4_journal_stop(handle);
2674 unlock_new_inode(inode);
2675 return err;
2676}
2677
2678struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2679 struct ext4_dir_entry_2 *de,
2680 int blocksize, int csum_size,
2681 unsigned int parent_ino, int dotdot_real_len)
2682{
2683 de->inode = cpu_to_le32(inode->i_ino);
2684 de->name_len = 1;
2685 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2686 blocksize);
2687 strcpy(de->name, ".");
2688 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2689
2690 de = ext4_next_entry(de, blocksize);
2691 de->inode = cpu_to_le32(parent_ino);
2692 de->name_len = 2;
2693 if (!dotdot_real_len)
2694 de->rec_len = ext4_rec_len_to_disk(blocksize -
2695 (csum_size + EXT4_DIR_REC_LEN(1)),
2696 blocksize);
2697 else
2698 de->rec_len = ext4_rec_len_to_disk(
2699 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2700 strcpy(de->name, "..");
2701 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2702
2703 return ext4_next_entry(de, blocksize);
2704}
2705
2706static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2707 struct inode *inode)
2708{
2709 struct buffer_head *dir_block = NULL;
2710 struct ext4_dir_entry_2 *de;
2711 ext4_lblk_t block = 0;
2712 unsigned int blocksize = dir->i_sb->s_blocksize;
2713 int csum_size = 0;
2714 int err;
2715
2716 if (ext4_has_metadata_csum(dir->i_sb))
2717 csum_size = sizeof(struct ext4_dir_entry_tail);
2718
2719 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2720 err = ext4_try_create_inline_dir(handle, dir, inode);
2721 if (err < 0 && err != -ENOSPC)
2722 goto out;
2723 if (!err)
2724 goto out;
2725 }
2726
2727 inode->i_size = 0;
2728 dir_block = ext4_append(handle, inode, &block);
2729 if (IS_ERR(dir_block))
2730 return PTR_ERR(dir_block);
2731 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2732 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2733 set_nlink(inode, 2);
2734 if (csum_size)
2735 ext4_initialize_dirent_tail(dir_block, blocksize);
2736
2737 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2738 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2739 if (err)
2740 goto out;
2741 set_buffer_verified(dir_block);
2742out:
2743 brelse(dir_block);
2744 return err;
2745}
2746
2747static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2748{
2749 handle_t *handle;
2750 struct inode *inode;
2751 int err, credits, retries = 0;
2752
2753 if (EXT4_DIR_LINK_MAX(dir))
2754 return -EMLINK;
2755
2756 err = dquot_initialize(dir);
2757 if (err)
2758 return err;
2759
2760 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2761 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2762retry:
2763 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2764 &dentry->d_name,
2765 0, NULL, EXT4_HT_DIR, credits);
2766 handle = ext4_journal_current_handle();
2767 err = PTR_ERR(inode);
2768 if (IS_ERR(inode))
2769 goto out_stop;
2770
2771 inode->i_op = &ext4_dir_inode_operations;
2772 inode->i_fop = &ext4_dir_operations;
2773 err = ext4_init_new_dir(handle, dir, inode);
2774 if (err)
2775 goto out_clear_inode;
2776 err = ext4_mark_inode_dirty(handle, inode);
2777 if (!err)
2778 err = ext4_add_entry(handle, dentry, inode);
2779 if (err) {
2780out_clear_inode:
2781 clear_nlink(inode);
2782 unlock_new_inode(inode);
2783 ext4_mark_inode_dirty(handle, inode);
2784 iput(inode);
2785 goto out_stop;
2786 }
2787 ext4_inc_count(handle, dir);
2788 ext4_update_dx_flag(dir);
2789 err = ext4_mark_inode_dirty(handle, dir);
2790 if (err)
2791 goto out_clear_inode;
2792 d_instantiate_new(dentry, inode);
2793 if (IS_DIRSYNC(dir))
2794 ext4_handle_sync(handle);
2795
2796out_stop:
2797 if (handle)
2798 ext4_journal_stop(handle);
2799 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2800 goto retry;
2801 return err;
2802}
2803
2804/*
2805 * routine to check that the specified directory is empty (for rmdir)
2806 */
2807bool ext4_empty_dir(struct inode *inode)
2808{
2809 unsigned int offset;
2810 struct buffer_head *bh;
2811 struct ext4_dir_entry_2 *de, *de1;
2812 struct super_block *sb;
2813
2814 if (ext4_has_inline_data(inode)) {
2815 int has_inline_data = 1;
2816 int ret;
2817
2818 ret = empty_inline_dir(inode, &has_inline_data);
2819 if (has_inline_data)
2820 return ret;
2821 }
2822
2823 sb = inode->i_sb;
2824 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2825 EXT4_ERROR_INODE(inode, "invalid size");
2826 return true;
2827 }
2828 /* The first directory block must not be a hole,
2829 * so treat it as DIRENT_HTREE
2830 */
2831 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2832 if (IS_ERR(bh))
2833 return true;
2834
2835 de = (struct ext4_dir_entry_2 *) bh->b_data;
2836 de1 = ext4_next_entry(de, sb->s_blocksize);
2837 if (le32_to_cpu(de->inode) != inode->i_ino ||
2838 le32_to_cpu(de1->inode) == 0 ||
2839 strcmp(".", de->name) || strcmp("..", de1->name)) {
2840 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2841 brelse(bh);
2842 return true;
2843 }
2844 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2845 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2846 de = ext4_next_entry(de1, sb->s_blocksize);
2847 while (offset < inode->i_size) {
2848 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2849 unsigned int lblock;
2850 brelse(bh);
2851 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2852 bh = ext4_read_dirblock(inode, lblock, EITHER);
2853 if (bh == NULL) {
2854 offset += sb->s_blocksize;
2855 continue;
2856 }
2857 if (IS_ERR(bh))
2858 return true;
2859 de = (struct ext4_dir_entry_2 *) bh->b_data;
2860 }
2861 if (ext4_check_dir_entry(inode, NULL, de, bh,
2862 bh->b_data, bh->b_size, offset)) {
2863 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2864 sb->s_blocksize);
2865 offset = (offset | (sb->s_blocksize - 1)) + 1;
2866 continue;
2867 }
2868 if (le32_to_cpu(de->inode)) {
2869 brelse(bh);
2870 return false;
2871 }
2872 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2873 de = ext4_next_entry(de, sb->s_blocksize);
2874 }
2875 brelse(bh);
2876 return true;
2877}
2878
2879/*
2880 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2881 * such inodes, starting at the superblock, in case we crash before the
2882 * file is closed/deleted, or in case the inode truncate spans multiple
2883 * transactions and the last transaction is not recovered after a crash.
2884 *
2885 * At filesystem recovery time, we walk this list deleting unlinked
2886 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2887 *
2888 * Orphan list manipulation functions must be called under i_mutex unless
2889 * we are just creating the inode or deleting it.
2890 */
2891int ext4_orphan_add(handle_t *handle, struct inode *inode)
2892{
2893 struct super_block *sb = inode->i_sb;
2894 struct ext4_sb_info *sbi = EXT4_SB(sb);
2895 struct ext4_iloc iloc;
2896 int err = 0, rc;
2897 bool dirty = false;
2898
2899 if (!sbi->s_journal || is_bad_inode(inode))
2900 return 0;
2901
2902 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2903 !inode_is_locked(inode));
2904 /*
2905 * Exit early if inode already is on orphan list. This is a big speedup
2906 * since we don't have to contend on the global s_orphan_lock.
2907 */
2908 if (!list_empty(&EXT4_I(inode)->i_orphan))
2909 return 0;
2910
2911 /*
2912 * Orphan handling is only valid for files with data blocks
2913 * being truncated, or files being unlinked. Note that we either
2914 * hold i_mutex, or the inode can not be referenced from outside,
2915 * so i_nlink should not be bumped due to race
2916 */
2917 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2918 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2919
2920 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2921 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2922 if (err)
2923 goto out;
2924
2925 err = ext4_reserve_inode_write(handle, inode, &iloc);
2926 if (err)
2927 goto out;
2928
2929 mutex_lock(&sbi->s_orphan_lock);
2930 /*
2931 * Due to previous errors inode may be already a part of on-disk
2932 * orphan list. If so skip on-disk list modification.
2933 */
2934 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2935 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2936 /* Insert this inode at the head of the on-disk orphan list */
2937 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2938 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2939 dirty = true;
2940 }
2941 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2942 mutex_unlock(&sbi->s_orphan_lock);
2943
2944 if (dirty) {
2945 err = ext4_handle_dirty_super(handle, sb);
2946 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2947 if (!err)
2948 err = rc;
2949 if (err) {
2950 /*
2951 * We have to remove inode from in-memory list if
2952 * addition to on disk orphan list failed. Stray orphan
2953 * list entries can cause panics at unmount time.
2954 */
2955 mutex_lock(&sbi->s_orphan_lock);
2956 list_del_init(&EXT4_I(inode)->i_orphan);
2957 mutex_unlock(&sbi->s_orphan_lock);
2958 }
2959 } else
2960 brelse(iloc.bh);
2961
2962 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2963 jbd_debug(4, "orphan inode %lu will point to %d\n",
2964 inode->i_ino, NEXT_ORPHAN(inode));
2965out:
2966 ext4_std_error(sb, err);
2967 return err;
2968}
2969
2970/*
2971 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2972 * of such inodes stored on disk, because it is finally being cleaned up.
2973 */
2974int ext4_orphan_del(handle_t *handle, struct inode *inode)
2975{
2976 struct list_head *prev;
2977 struct ext4_inode_info *ei = EXT4_I(inode);
2978 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2979 __u32 ino_next;
2980 struct ext4_iloc iloc;
2981 int err = 0;
2982
2983 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2984 return 0;
2985
2986 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2987 !inode_is_locked(inode));
2988 /* Do this quick check before taking global s_orphan_lock. */
2989 if (list_empty(&ei->i_orphan))
2990 return 0;
2991
2992 if (handle) {
2993 /* Grab inode buffer early before taking global s_orphan_lock */
2994 err = ext4_reserve_inode_write(handle, inode, &iloc);
2995 }
2996
2997 mutex_lock(&sbi->s_orphan_lock);
2998 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2999
3000 prev = ei->i_orphan.prev;
3001 list_del_init(&ei->i_orphan);
3002
3003 /* If we're on an error path, we may not have a valid
3004 * transaction handle with which to update the orphan list on
3005 * disk, but we still need to remove the inode from the linked
3006 * list in memory. */
3007 if (!handle || err) {
3008 mutex_unlock(&sbi->s_orphan_lock);
3009 goto out_err;
3010 }
3011
3012 ino_next = NEXT_ORPHAN(inode);
3013 if (prev == &sbi->s_orphan) {
3014 jbd_debug(4, "superblock will point to %u\n", ino_next);
3015 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3016 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3017 if (err) {
3018 mutex_unlock(&sbi->s_orphan_lock);
3019 goto out_brelse;
3020 }
3021 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3022 mutex_unlock(&sbi->s_orphan_lock);
3023 err = ext4_handle_dirty_super(handle, inode->i_sb);
3024 } else {
3025 struct ext4_iloc iloc2;
3026 struct inode *i_prev =
3027 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3028
3029 jbd_debug(4, "orphan inode %lu will point to %u\n",
3030 i_prev->i_ino, ino_next);
3031 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3032 if (err) {
3033 mutex_unlock(&sbi->s_orphan_lock);
3034 goto out_brelse;
3035 }
3036 NEXT_ORPHAN(i_prev) = ino_next;
3037 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3038 mutex_unlock(&sbi->s_orphan_lock);
3039 }
3040 if (err)
3041 goto out_brelse;
3042 NEXT_ORPHAN(inode) = 0;
3043 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3044out_err:
3045 ext4_std_error(inode->i_sb, err);
3046 return err;
3047
3048out_brelse:
3049 brelse(iloc.bh);
3050 goto out_err;
3051}
3052
3053static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3054{
3055 int retval;
3056 struct inode *inode;
3057 struct buffer_head *bh;
3058 struct ext4_dir_entry_2 *de;
3059 handle_t *handle = NULL;
3060
3061 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3062 return -EIO;
3063
3064 /* Initialize quotas before so that eventual writes go in
3065 * separate transaction */
3066 retval = dquot_initialize(dir);
3067 if (retval)
3068 return retval;
3069 retval = dquot_initialize(d_inode(dentry));
3070 if (retval)
3071 return retval;
3072
3073 retval = -ENOENT;
3074 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3075 if (IS_ERR(bh))
3076 return PTR_ERR(bh);
3077 if (!bh)
3078 goto end_rmdir;
3079
3080 inode = d_inode(dentry);
3081
3082 retval = -EFSCORRUPTED;
3083 if (le32_to_cpu(de->inode) != inode->i_ino)
3084 goto end_rmdir;
3085
3086 retval = -ENOTEMPTY;
3087 if (!ext4_empty_dir(inode))
3088 goto end_rmdir;
3089
3090 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3091 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3092 if (IS_ERR(handle)) {
3093 retval = PTR_ERR(handle);
3094 handle = NULL;
3095 goto end_rmdir;
3096 }
3097
3098 if (IS_DIRSYNC(dir))
3099 ext4_handle_sync(handle);
3100
3101 retval = ext4_delete_entry(handle, dir, de, bh);
3102 if (retval)
3103 goto end_rmdir;
3104 if (!EXT4_DIR_LINK_EMPTY(inode))
3105 ext4_warning_inode(inode,
3106 "empty directory '%.*s' has too many links (%u)",
3107 dentry->d_name.len, dentry->d_name.name,
3108 inode->i_nlink);
3109 inode_inc_iversion(inode);
3110 clear_nlink(inode);
3111 /* There's no need to set i_disksize: the fact that i_nlink is
3112 * zero will ensure that the right thing happens during any
3113 * recovery. */
3114 inode->i_size = 0;
3115 ext4_orphan_add(handle, inode);
3116 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3117 ext4_mark_inode_dirty(handle, inode);
3118 ext4_dec_count(handle, dir);
3119 ext4_update_dx_flag(dir);
3120 ext4_mark_inode_dirty(handle, dir);
3121
3122#ifdef CONFIG_UNICODE
3123 /* VFS negative dentries are incompatible with Encoding and
3124 * Case-insensitiveness. Eventually we'll want avoid
3125 * invalidating the dentries here, alongside with returning the
3126 * negative dentries at ext4_lookup(), when it is better
3127 * supported by the VFS for the CI case.
3128 */
3129 if (IS_CASEFOLDED(dir))
3130 d_invalidate(dentry);
3131#endif
3132
3133end_rmdir:
3134 brelse(bh);
3135 if (handle)
3136 ext4_journal_stop(handle);
3137 return retval;
3138}
3139
3140static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3141{
3142 int retval;
3143 struct inode *inode;
3144 struct buffer_head *bh;
3145 struct ext4_dir_entry_2 *de;
3146 handle_t *handle = NULL;
3147
3148 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3149 return -EIO;
3150
3151 trace_ext4_unlink_enter(dir, dentry);
3152 /* Initialize quotas before so that eventual writes go
3153 * in separate transaction */
3154 retval = dquot_initialize(dir);
3155 if (retval)
3156 return retval;
3157 retval = dquot_initialize(d_inode(dentry));
3158 if (retval)
3159 return retval;
3160
3161 retval = -ENOENT;
3162 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3163 if (IS_ERR(bh))
3164 return PTR_ERR(bh);
3165 if (!bh)
3166 goto end_unlink;
3167
3168 inode = d_inode(dentry);
3169
3170 retval = -EFSCORRUPTED;
3171 if (le32_to_cpu(de->inode) != inode->i_ino)
3172 goto end_unlink;
3173
3174 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3175 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3176 if (IS_ERR(handle)) {
3177 retval = PTR_ERR(handle);
3178 handle = NULL;
3179 goto end_unlink;
3180 }
3181
3182 if (IS_DIRSYNC(dir))
3183 ext4_handle_sync(handle);
3184
3185 if (inode->i_nlink == 0) {
3186 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3187 dentry->d_name.len, dentry->d_name.name);
3188 set_nlink(inode, 1);
3189 }
3190 retval = ext4_delete_entry(handle, dir, de, bh);
3191 if (retval)
3192 goto end_unlink;
3193 dir->i_ctime = dir->i_mtime = current_time(dir);
3194 ext4_update_dx_flag(dir);
3195 ext4_mark_inode_dirty(handle, dir);
3196 drop_nlink(inode);
3197 if (!inode->i_nlink)
3198 ext4_orphan_add(handle, inode);
3199 inode->i_ctime = current_time(inode);
3200 ext4_mark_inode_dirty(handle, inode);
3201
3202#ifdef CONFIG_UNICODE
3203 /* VFS negative dentries are incompatible with Encoding and
3204 * Case-insensitiveness. Eventually we'll want avoid
3205 * invalidating the dentries here, alongside with returning the
3206 * negative dentries at ext4_lookup(), when it is better
3207 * supported by the VFS for the CI case.
3208 */
3209 if (IS_CASEFOLDED(dir))
3210 d_invalidate(dentry);
3211#endif
3212
3213end_unlink:
3214 brelse(bh);
3215 if (handle)
3216 ext4_journal_stop(handle);
3217 trace_ext4_unlink_exit(dentry, retval);
3218 return retval;
3219}
3220
3221static int ext4_symlink(struct inode *dir,
3222 struct dentry *dentry, const char *symname)
3223{
3224 handle_t *handle;
3225 struct inode *inode;
3226 int err, len = strlen(symname);
3227 int credits;
3228 struct fscrypt_str disk_link;
3229
3230 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3231 return -EIO;
3232
3233 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3234 &disk_link);
3235 if (err)
3236 return err;
3237
3238 err = dquot_initialize(dir);
3239 if (err)
3240 return err;
3241
3242 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3243 /*
3244 * For non-fast symlinks, we just allocate inode and put it on
3245 * orphan list in the first transaction => we need bitmap,
3246 * group descriptor, sb, inode block, quota blocks, and
3247 * possibly selinux xattr blocks.
3248 */
3249 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3250 EXT4_XATTR_TRANS_BLOCKS;
3251 } else {
3252 /*
3253 * Fast symlink. We have to add entry to directory
3254 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3255 * allocate new inode (bitmap, group descriptor, inode block,
3256 * quota blocks, sb is already counted in previous macros).
3257 */
3258 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3259 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3260 }
3261
3262 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3263 &dentry->d_name, 0, NULL,
3264 EXT4_HT_DIR, credits);
3265 handle = ext4_journal_current_handle();
3266 if (IS_ERR(inode)) {
3267 if (handle)
3268 ext4_journal_stop(handle);
3269 return PTR_ERR(inode);
3270 }
3271
3272 if (IS_ENCRYPTED(inode)) {
3273 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3274 if (err)
3275 goto err_drop_inode;
3276 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3277 }
3278
3279 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3280 if (!IS_ENCRYPTED(inode))
3281 inode->i_op = &ext4_symlink_inode_operations;
3282 inode_nohighmem(inode);
3283 ext4_set_aops(inode);
3284 /*
3285 * We cannot call page_symlink() with transaction started
3286 * because it calls into ext4_write_begin() which can wait
3287 * for transaction commit if we are running out of space
3288 * and thus we deadlock. So we have to stop transaction now
3289 * and restart it when symlink contents is written.
3290 *
3291 * To keep fs consistent in case of crash, we have to put inode
3292 * to orphan list in the mean time.
3293 */
3294 drop_nlink(inode);
3295 err = ext4_orphan_add(handle, inode);
3296 ext4_journal_stop(handle);
3297 handle = NULL;
3298 if (err)
3299 goto err_drop_inode;
3300 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3301 if (err)
3302 goto err_drop_inode;
3303 /*
3304 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3305 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3306 */
3307 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3308 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3309 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3310 if (IS_ERR(handle)) {
3311 err = PTR_ERR(handle);
3312 handle = NULL;
3313 goto err_drop_inode;
3314 }
3315 set_nlink(inode, 1);
3316 err = ext4_orphan_del(handle, inode);
3317 if (err)
3318 goto err_drop_inode;
3319 } else {
3320 /* clear the extent format for fast symlink */
3321 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3322 if (!IS_ENCRYPTED(inode)) {
3323 inode->i_op = &ext4_fast_symlink_inode_operations;
3324 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3325 }
3326 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3327 disk_link.len);
3328 inode->i_size = disk_link.len - 1;
3329 }
3330 EXT4_I(inode)->i_disksize = inode->i_size;
3331 err = ext4_add_nondir(handle, dentry, inode);
3332 if (!err && IS_DIRSYNC(dir))
3333 ext4_handle_sync(handle);
3334
3335 if (handle)
3336 ext4_journal_stop(handle);
3337 goto out_free_encrypted_link;
3338
3339err_drop_inode:
3340 if (handle)
3341 ext4_journal_stop(handle);
3342 clear_nlink(inode);
3343 unlock_new_inode(inode);
3344 iput(inode);
3345out_free_encrypted_link:
3346 if (disk_link.name != (unsigned char *)symname)
3347 kfree(disk_link.name);
3348 return err;
3349}
3350
3351static int ext4_link(struct dentry *old_dentry,
3352 struct inode *dir, struct dentry *dentry)
3353{
3354 handle_t *handle;
3355 struct inode *inode = d_inode(old_dentry);
3356 int err, retries = 0;
3357
3358 if (inode->i_nlink >= EXT4_LINK_MAX)
3359 return -EMLINK;
3360
3361 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3362 if (err)
3363 return err;
3364
3365 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3366 (!projid_eq(EXT4_I(dir)->i_projid,
3367 EXT4_I(old_dentry->d_inode)->i_projid)))
3368 return -EXDEV;
3369
3370 err = dquot_initialize(dir);
3371 if (err)
3372 return err;
3373
3374retry:
3375 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3376 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3377 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3378 if (IS_ERR(handle))
3379 return PTR_ERR(handle);
3380
3381 if (IS_DIRSYNC(dir))
3382 ext4_handle_sync(handle);
3383
3384 inode->i_ctime = current_time(inode);
3385 ext4_inc_count(handle, inode);
3386 ihold(inode);
3387
3388 err = ext4_add_entry(handle, dentry, inode);
3389 if (!err) {
3390 ext4_mark_inode_dirty(handle, inode);
3391 /* this can happen only for tmpfile being
3392 * linked the first time
3393 */
3394 if (inode->i_nlink == 1)
3395 ext4_orphan_del(handle, inode);
3396 d_instantiate(dentry, inode);
3397 } else {
3398 drop_nlink(inode);
3399 iput(inode);
3400 }
3401 ext4_journal_stop(handle);
3402 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3403 goto retry;
3404 return err;
3405}
3406
3407
3408/*
3409 * Try to find buffer head where contains the parent block.
3410 * It should be the inode block if it is inlined or the 1st block
3411 * if it is a normal dir.
3412 */
3413static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3414 struct inode *inode,
3415 int *retval,
3416 struct ext4_dir_entry_2 **parent_de,
3417 int *inlined)
3418{
3419 struct buffer_head *bh;
3420
3421 if (!ext4_has_inline_data(inode)) {
3422 /* The first directory block must not be a hole, so
3423 * treat it as DIRENT_HTREE
3424 */
3425 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3426 if (IS_ERR(bh)) {
3427 *retval = PTR_ERR(bh);
3428 return NULL;
3429 }
3430 *parent_de = ext4_next_entry(
3431 (struct ext4_dir_entry_2 *)bh->b_data,
3432 inode->i_sb->s_blocksize);
3433 return bh;
3434 }
3435
3436 *inlined = 1;
3437 return ext4_get_first_inline_block(inode, parent_de, retval);
3438}
3439
3440struct ext4_renament {
3441 struct inode *dir;
3442 struct dentry *dentry;
3443 struct inode *inode;
3444 bool is_dir;
3445 int dir_nlink_delta;
3446
3447 /* entry for "dentry" */
3448 struct buffer_head *bh;
3449 struct ext4_dir_entry_2 *de;
3450 int inlined;
3451
3452 /* entry for ".." in inode if it's a directory */
3453 struct buffer_head *dir_bh;
3454 struct ext4_dir_entry_2 *parent_de;
3455 int dir_inlined;
3456};
3457
3458static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3459{
3460 int retval;
3461
3462 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3463 &retval, &ent->parent_de,
3464 &ent->dir_inlined);
3465 if (!ent->dir_bh)
3466 return retval;
3467 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3468 return -EFSCORRUPTED;
3469 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3470 return ext4_journal_get_write_access(handle, ent->dir_bh);
3471}
3472
3473static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3474 unsigned dir_ino)
3475{
3476 int retval;
3477
3478 ent->parent_de->inode = cpu_to_le32(dir_ino);
3479 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3480 if (!ent->dir_inlined) {
3481 if (is_dx(ent->inode)) {
3482 retval = ext4_handle_dirty_dx_node(handle,
3483 ent->inode,
3484 ent->dir_bh);
3485 } else {
3486 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3487 ent->dir_bh);
3488 }
3489 } else {
3490 retval = ext4_mark_inode_dirty(handle, ent->inode);
3491 }
3492 if (retval) {
3493 ext4_std_error(ent->dir->i_sb, retval);
3494 return retval;
3495 }
3496 return 0;
3497}
3498
3499static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3500 unsigned ino, unsigned file_type)
3501{
3502 int retval;
3503
3504 BUFFER_TRACE(ent->bh, "get write access");
3505 retval = ext4_journal_get_write_access(handle, ent->bh);
3506 if (retval)
3507 return retval;
3508 ent->de->inode = cpu_to_le32(ino);
3509 if (ext4_has_feature_filetype(ent->dir->i_sb))
3510 ent->de->file_type = file_type;
3511 inode_inc_iversion(ent->dir);
3512 ent->dir->i_ctime = ent->dir->i_mtime =
3513 current_time(ent->dir);
3514 ext4_mark_inode_dirty(handle, ent->dir);
3515 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3516 if (!ent->inlined) {
3517 retval = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3518 if (unlikely(retval)) {
3519 ext4_std_error(ent->dir->i_sb, retval);
3520 return retval;
3521 }
3522 }
3523 brelse(ent->bh);
3524 ent->bh = NULL;
3525
3526 return 0;
3527}
3528
3529static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3530 const struct qstr *d_name)
3531{
3532 int retval = -ENOENT;
3533 struct buffer_head *bh;
3534 struct ext4_dir_entry_2 *de;
3535
3536 bh = ext4_find_entry(dir, d_name, &de, NULL);
3537 if (IS_ERR(bh))
3538 return PTR_ERR(bh);
3539 if (bh) {
3540 retval = ext4_delete_entry(handle, dir, de, bh);
3541 brelse(bh);
3542 }
3543 return retval;
3544}
3545
3546static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3547 int force_reread)
3548{
3549 int retval;
3550 /*
3551 * ent->de could have moved from under us during htree split, so make
3552 * sure that we are deleting the right entry. We might also be pointing
3553 * to a stale entry in the unused part of ent->bh so just checking inum
3554 * and the name isn't enough.
3555 */
3556 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3557 ent->de->name_len != ent->dentry->d_name.len ||
3558 strncmp(ent->de->name, ent->dentry->d_name.name,
3559 ent->de->name_len) ||
3560 force_reread) {
3561 retval = ext4_find_delete_entry(handle, ent->dir,
3562 &ent->dentry->d_name);
3563 } else {
3564 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3565 if (retval == -ENOENT) {
3566 retval = ext4_find_delete_entry(handle, ent->dir,
3567 &ent->dentry->d_name);
3568 }
3569 }
3570
3571 if (retval) {
3572 ext4_warning_inode(ent->dir,
3573 "Deleting old file: nlink %d, error=%d",
3574 ent->dir->i_nlink, retval);
3575 }
3576}
3577
3578static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3579{
3580 if (ent->dir_nlink_delta) {
3581 if (ent->dir_nlink_delta == -1)
3582 ext4_dec_count(handle, ent->dir);
3583 else
3584 ext4_inc_count(handle, ent->dir);
3585 ext4_mark_inode_dirty(handle, ent->dir);
3586 }
3587}
3588
3589static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3590 int credits, handle_t **h)
3591{
3592 struct inode *wh;
3593 handle_t *handle;
3594 int retries = 0;
3595
3596 /*
3597 * for inode block, sb block, group summaries,
3598 * and inode bitmap
3599 */
3600 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3601 EXT4_XATTR_TRANS_BLOCKS + 4);
3602retry:
3603 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3604 &ent->dentry->d_name, 0, NULL,
3605 EXT4_HT_DIR, credits);
3606
3607 handle = ext4_journal_current_handle();
3608 if (IS_ERR(wh)) {
3609 if (handle)
3610 ext4_journal_stop(handle);
3611 if (PTR_ERR(wh) == -ENOSPC &&
3612 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3613 goto retry;
3614 } else {
3615 *h = handle;
3616 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3617 wh->i_op = &ext4_special_inode_operations;
3618 }
3619 return wh;
3620}
3621
3622/*
3623 * Anybody can rename anything with this: the permission checks are left to the
3624 * higher-level routines.
3625 *
3626 * n.b. old_{dentry,inode) refers to the source dentry/inode
3627 * while new_{dentry,inode) refers to the destination dentry/inode
3628 * This comes from rename(const char *oldpath, const char *newpath)
3629 */
3630static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3631 struct inode *new_dir, struct dentry *new_dentry,
3632 unsigned int flags)
3633{
3634 handle_t *handle = NULL;
3635 struct ext4_renament old = {
3636 .dir = old_dir,
3637 .dentry = old_dentry,
3638 .inode = d_inode(old_dentry),
3639 };
3640 struct ext4_renament new = {
3641 .dir = new_dir,
3642 .dentry = new_dentry,
3643 .inode = d_inode(new_dentry),
3644 };
3645 int force_reread;
3646 int retval;
3647 struct inode *whiteout = NULL;
3648 int credits;
3649 u8 old_file_type;
3650
3651 if (new.inode && new.inode->i_nlink == 0) {
3652 EXT4_ERROR_INODE(new.inode,
3653 "target of rename is already freed");
3654 return -EFSCORRUPTED;
3655 }
3656
3657 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3658 (!projid_eq(EXT4_I(new_dir)->i_projid,
3659 EXT4_I(old_dentry->d_inode)->i_projid)))
3660 return -EXDEV;
3661
3662 retval = dquot_initialize(old.dir);
3663 if (retval)
3664 return retval;
3665 retval = dquot_initialize(new.dir);
3666 if (retval)
3667 return retval;
3668
3669 /* Initialize quotas before so that eventual writes go
3670 * in separate transaction */
3671 if (new.inode) {
3672 retval = dquot_initialize(new.inode);
3673 if (retval)
3674 return retval;
3675 }
3676
3677 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3678 if (IS_ERR(old.bh))
3679 return PTR_ERR(old.bh);
3680 /*
3681 * Check for inode number is _not_ due to possible IO errors.
3682 * We might rmdir the source, keep it as pwd of some process
3683 * and merrily kill the link to whatever was created under the
3684 * same name. Goodbye sticky bit ;-<
3685 */
3686 retval = -ENOENT;
3687 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3688 goto end_rename;
3689
3690 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3691 &new.de, &new.inlined);
3692 if (IS_ERR(new.bh)) {
3693 retval = PTR_ERR(new.bh);
3694 new.bh = NULL;
3695 goto end_rename;
3696 }
3697 if (new.bh) {
3698 if (!new.inode) {
3699 brelse(new.bh);
3700 new.bh = NULL;
3701 }
3702 }
3703 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3704 ext4_alloc_da_blocks(old.inode);
3705
3706 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3707 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3708 if (!(flags & RENAME_WHITEOUT)) {
3709 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3710 if (IS_ERR(handle)) {
3711 retval = PTR_ERR(handle);
3712 handle = NULL;
3713 goto end_rename;
3714 }
3715 } else {
3716 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3717 if (IS_ERR(whiteout)) {
3718 retval = PTR_ERR(whiteout);
3719 whiteout = NULL;
3720 goto end_rename;
3721 }
3722 }
3723
3724 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3725 ext4_handle_sync(handle);
3726
3727 if (S_ISDIR(old.inode->i_mode)) {
3728 if (new.inode) {
3729 retval = -ENOTEMPTY;
3730 if (!ext4_empty_dir(new.inode))
3731 goto end_rename;
3732 } else {
3733 retval = -EMLINK;
3734 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3735 goto end_rename;
3736 }
3737 retval = ext4_rename_dir_prepare(handle, &old);
3738 if (retval)
3739 goto end_rename;
3740 }
3741 /*
3742 * If we're renaming a file within an inline_data dir and adding or
3743 * setting the new dirent causes a conversion from inline_data to
3744 * extents/blockmap, we need to force the dirent delete code to
3745 * re-read the directory, or else we end up trying to delete a dirent
3746 * from what is now the extent tree root (or a block map).
3747 */
3748 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3749 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3750
3751 old_file_type = old.de->file_type;
3752 if (whiteout) {
3753 /*
3754 * Do this before adding a new entry, so the old entry is sure
3755 * to be still pointing to the valid old entry.
3756 */
3757 retval = ext4_setent(handle, &old, whiteout->i_ino,
3758 EXT4_FT_CHRDEV);
3759 if (retval)
3760 goto end_rename;
3761 ext4_mark_inode_dirty(handle, whiteout);
3762 }
3763 if (!new.bh) {
3764 retval = ext4_add_entry(handle, new.dentry, old.inode);
3765 if (retval)
3766 goto end_rename;
3767 } else {
3768 retval = ext4_setent(handle, &new,
3769 old.inode->i_ino, old_file_type);
3770 if (retval)
3771 goto end_rename;
3772 }
3773 if (force_reread)
3774 force_reread = !ext4_test_inode_flag(new.dir,
3775 EXT4_INODE_INLINE_DATA);
3776
3777 /*
3778 * Like most other Unix systems, set the ctime for inodes on a
3779 * rename.
3780 */
3781 old.inode->i_ctime = current_time(old.inode);
3782 ext4_mark_inode_dirty(handle, old.inode);
3783
3784 if (!whiteout) {
3785 /*
3786 * ok, that's it
3787 */
3788 ext4_rename_delete(handle, &old, force_reread);
3789 }
3790
3791 if (new.inode) {
3792 ext4_dec_count(handle, new.inode);
3793 new.inode->i_ctime = current_time(new.inode);
3794 }
3795 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3796 ext4_update_dx_flag(old.dir);
3797 if (old.dir_bh) {
3798 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3799 if (retval)
3800 goto end_rename;
3801
3802 ext4_dec_count(handle, old.dir);
3803 if (new.inode) {
3804 /* checked ext4_empty_dir above, can't have another
3805 * parent, ext4_dec_count() won't work for many-linked
3806 * dirs */
3807 clear_nlink(new.inode);
3808 } else {
3809 ext4_inc_count(handle, new.dir);
3810 ext4_update_dx_flag(new.dir);
3811 ext4_mark_inode_dirty(handle, new.dir);
3812 }
3813 }
3814 ext4_mark_inode_dirty(handle, old.dir);
3815 if (new.inode) {
3816 ext4_mark_inode_dirty(handle, new.inode);
3817 if (!new.inode->i_nlink)
3818 ext4_orphan_add(handle, new.inode);
3819 }
3820 retval = 0;
3821
3822end_rename:
3823 brelse(old.dir_bh);
3824 brelse(old.bh);
3825 brelse(new.bh);
3826 if (whiteout) {
3827 if (retval)
3828 drop_nlink(whiteout);
3829 unlock_new_inode(whiteout);
3830 iput(whiteout);
3831 }
3832 if (handle)
3833 ext4_journal_stop(handle);
3834 return retval;
3835}
3836
3837static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3838 struct inode *new_dir, struct dentry *new_dentry)
3839{
3840 handle_t *handle = NULL;
3841 struct ext4_renament old = {
3842 .dir = old_dir,
3843 .dentry = old_dentry,
3844 .inode = d_inode(old_dentry),
3845 };
3846 struct ext4_renament new = {
3847 .dir = new_dir,
3848 .dentry = new_dentry,
3849 .inode = d_inode(new_dentry),
3850 };
3851 u8 new_file_type;
3852 int retval;
3853 struct timespec64 ctime;
3854
3855 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3856 !projid_eq(EXT4_I(new_dir)->i_projid,
3857 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3858 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3859 !projid_eq(EXT4_I(old_dir)->i_projid,
3860 EXT4_I(new_dentry->d_inode)->i_projid)))
3861 return -EXDEV;
3862
3863 retval = dquot_initialize(old.dir);
3864 if (retval)
3865 return retval;
3866 retval = dquot_initialize(new.dir);
3867 if (retval)
3868 return retval;
3869
3870 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3871 &old.de, &old.inlined);
3872 if (IS_ERR(old.bh))
3873 return PTR_ERR(old.bh);
3874 /*
3875 * Check for inode number is _not_ due to possible IO errors.
3876 * We might rmdir the source, keep it as pwd of some process
3877 * and merrily kill the link to whatever was created under the
3878 * same name. Goodbye sticky bit ;-<
3879 */
3880 retval = -ENOENT;
3881 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3882 goto end_rename;
3883
3884 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3885 &new.de, &new.inlined);
3886 if (IS_ERR(new.bh)) {
3887 retval = PTR_ERR(new.bh);
3888 new.bh = NULL;
3889 goto end_rename;
3890 }
3891
3892 /* RENAME_EXCHANGE case: old *and* new must both exist */
3893 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3894 goto end_rename;
3895
3896 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3897 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3898 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3899 if (IS_ERR(handle)) {
3900 retval = PTR_ERR(handle);
3901 handle = NULL;
3902 goto end_rename;
3903 }
3904
3905 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3906 ext4_handle_sync(handle);
3907
3908 if (S_ISDIR(old.inode->i_mode)) {
3909 old.is_dir = true;
3910 retval = ext4_rename_dir_prepare(handle, &old);
3911 if (retval)
3912 goto end_rename;
3913 }
3914 if (S_ISDIR(new.inode->i_mode)) {
3915 new.is_dir = true;
3916 retval = ext4_rename_dir_prepare(handle, &new);
3917 if (retval)
3918 goto end_rename;
3919 }
3920
3921 /*
3922 * Other than the special case of overwriting a directory, parents'
3923 * nlink only needs to be modified if this is a cross directory rename.
3924 */
3925 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3926 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3927 new.dir_nlink_delta = -old.dir_nlink_delta;
3928 retval = -EMLINK;
3929 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3930 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3931 goto end_rename;
3932 }
3933
3934 new_file_type = new.de->file_type;
3935 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3936 if (retval)
3937 goto end_rename;
3938
3939 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3940 if (retval)
3941 goto end_rename;
3942
3943 /*
3944 * Like most other Unix systems, set the ctime for inodes on a
3945 * rename.
3946 */
3947 ctime = current_time(old.inode);
3948 old.inode->i_ctime = ctime;
3949 new.inode->i_ctime = ctime;
3950 ext4_mark_inode_dirty(handle, old.inode);
3951 ext4_mark_inode_dirty(handle, new.inode);
3952
3953 if (old.dir_bh) {
3954 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3955 if (retval)
3956 goto end_rename;
3957 }
3958 if (new.dir_bh) {
3959 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3960 if (retval)
3961 goto end_rename;
3962 }
3963 ext4_update_dir_count(handle, &old);
3964 ext4_update_dir_count(handle, &new);
3965 retval = 0;
3966
3967end_rename:
3968 brelse(old.dir_bh);
3969 brelse(new.dir_bh);
3970 brelse(old.bh);
3971 brelse(new.bh);
3972 if (handle)
3973 ext4_journal_stop(handle);
3974 return retval;
3975}
3976
3977static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3978 struct inode *new_dir, struct dentry *new_dentry,
3979 unsigned int flags)
3980{
3981 int err;
3982
3983 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3984 return -EIO;
3985
3986 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3987 return -EINVAL;
3988
3989 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3990 flags);
3991 if (err)
3992 return err;
3993
3994 if (flags & RENAME_EXCHANGE) {
3995 return ext4_cross_rename(old_dir, old_dentry,
3996 new_dir, new_dentry);
3997 }
3998
3999 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4000}
4001
4002/*
4003 * directories can handle most operations...
4004 */
4005const struct inode_operations ext4_dir_inode_operations = {
4006 .create = ext4_create,
4007 .lookup = ext4_lookup,
4008 .link = ext4_link,
4009 .unlink = ext4_unlink,
4010 .symlink = ext4_symlink,
4011 .mkdir = ext4_mkdir,
4012 .rmdir = ext4_rmdir,
4013 .mknod = ext4_mknod,
4014 .tmpfile = ext4_tmpfile,
4015 .rename = ext4_rename2,
4016 .setattr = ext4_setattr,
4017 .getattr = ext4_getattr,
4018 .listxattr = ext4_listxattr,
4019 .get_acl = ext4_get_acl,
4020 .set_acl = ext4_set_acl,
4021 .fiemap = ext4_fiemap,
4022};
4023
4024const struct inode_operations ext4_special_inode_operations = {
4025 .setattr = ext4_setattr,
4026 .getattr = ext4_getattr,
4027 .listxattr = ext4_listxattr,
4028 .get_acl = ext4_get_acl,
4029 .set_acl = ext4_set_acl,
4030};