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 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
113 bh = ERR_PTR(-EIO);
114 else
115 bh = ext4_bread(NULL, inode, block, 0);
116 if (IS_ERR(bh)) {
117 __ext4_warning(inode->i_sb, func, line,
118 "inode #%lu: lblock %lu: comm %s: "
119 "error %ld reading directory block",
120 inode->i_ino, (unsigned long)block,
121 current->comm, PTR_ERR(bh));
122
123 return bh;
124 }
125 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
126 ext4_error_inode(inode, func, line, block,
127 "Directory hole found for htree %s block",
128 (type == INDEX) ? "index" : "leaf");
129 return ERR_PTR(-EFSCORRUPTED);
130 }
131 if (!bh)
132 return NULL;
133 dirent = (struct ext4_dir_entry *) bh->b_data;
134 /* Determine whether or not we have an index block */
135 if (is_dx(inode)) {
136 if (block == 0)
137 is_dx_block = 1;
138 else if (ext4_rec_len_from_disk(dirent->rec_len,
139 inode->i_sb->s_blocksize) ==
140 inode->i_sb->s_blocksize)
141 is_dx_block = 1;
142 }
143 if (!is_dx_block && type == INDEX) {
144 ext4_error_inode(inode, func, line, block,
145 "directory leaf block found instead of index block");
146 brelse(bh);
147 return ERR_PTR(-EFSCORRUPTED);
148 }
149 if (!ext4_has_metadata_csum(inode->i_sb) ||
150 buffer_verified(bh))
151 return bh;
152
153 /*
154 * An empty leaf block can get mistaken for a index block; for
155 * this reason, we can only check the index checksum when the
156 * caller is sure it should be an index block.
157 */
158 if (is_dx_block && type == INDEX) {
159 if (ext4_dx_csum_verify(inode, dirent) &&
160 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
161 set_buffer_verified(bh);
162 else {
163 ext4_error_inode_err(inode, func, line, block,
164 EFSBADCRC,
165 "Directory index failed checksum");
166 brelse(bh);
167 return ERR_PTR(-EFSBADCRC);
168 }
169 }
170 if (!is_dx_block) {
171 if (ext4_dirblock_csum_verify(inode, bh) &&
172 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
173 set_buffer_verified(bh);
174 else {
175 ext4_error_inode_err(inode, func, line, block,
176 EFSBADCRC,
177 "Directory block failed checksum");
178 brelse(bh);
179 return ERR_PTR(-EFSBADCRC);
180 }
181 }
182 return bh;
183}
184
185#ifndef assert
186#define assert(test) J_ASSERT(test)
187#endif
188
189#ifdef DX_DEBUG
190#define dxtrace(command) command
191#else
192#define dxtrace(command)
193#endif
194
195struct fake_dirent
196{
197 __le32 inode;
198 __le16 rec_len;
199 u8 name_len;
200 u8 file_type;
201};
202
203struct dx_countlimit
204{
205 __le16 limit;
206 __le16 count;
207};
208
209struct dx_entry
210{
211 __le32 hash;
212 __le32 block;
213};
214
215/*
216 * dx_root_info is laid out so that if it should somehow get overlaid by a
217 * dirent the two low bits of the hash version will be zero. Therefore, the
218 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
219 */
220
221struct dx_root
222{
223 struct fake_dirent dot;
224 char dot_name[4];
225 struct fake_dirent dotdot;
226 char dotdot_name[4];
227 struct dx_root_info
228 {
229 __le32 reserved_zero;
230 u8 hash_version;
231 u8 info_length; /* 8 */
232 u8 indirect_levels;
233 u8 unused_flags;
234 }
235 info;
236 struct dx_entry entries[];
237};
238
239struct dx_node
240{
241 struct fake_dirent fake;
242 struct dx_entry entries[];
243};
244
245
246struct dx_frame
247{
248 struct buffer_head *bh;
249 struct dx_entry *entries;
250 struct dx_entry *at;
251};
252
253struct dx_map_entry
254{
255 u32 hash;
256 u16 offs;
257 u16 size;
258};
259
260/*
261 * This goes at the end of each htree block.
262 */
263struct dx_tail {
264 u32 dt_reserved;
265 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
266};
267
268static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
269static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
270static inline unsigned dx_get_hash(struct dx_entry *entry);
271static void dx_set_hash(struct dx_entry *entry, unsigned value);
272static unsigned dx_get_count(struct dx_entry *entries);
273static unsigned dx_get_limit(struct dx_entry *entries);
274static void dx_set_count(struct dx_entry *entries, unsigned value);
275static void dx_set_limit(struct dx_entry *entries, unsigned value);
276static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
277static unsigned dx_node_limit(struct inode *dir);
278static struct dx_frame *dx_probe(struct ext4_filename *fname,
279 struct inode *dir,
280 struct dx_hash_info *hinfo,
281 struct dx_frame *frame);
282static void dx_release(struct dx_frame *frames);
283static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
284 unsigned blocksize, struct dx_hash_info *hinfo,
285 struct dx_map_entry map[]);
286static void dx_sort_map(struct dx_map_entry *map, unsigned count);
287static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
288 struct dx_map_entry *offsets, int count, unsigned blocksize);
289static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
290static void dx_insert_block(struct dx_frame *frame,
291 u32 hash, ext4_lblk_t block);
292static int ext4_htree_next_block(struct inode *dir, __u32 hash,
293 struct dx_frame *frame,
294 struct dx_frame *frames,
295 __u32 *start_hash);
296static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
297 struct ext4_filename *fname,
298 struct ext4_dir_entry_2 **res_dir);
299static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
300 struct inode *dir, struct inode *inode);
301
302/* checksumming functions */
303void ext4_initialize_dirent_tail(struct buffer_head *bh,
304 unsigned int blocksize)
305{
306 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
307
308 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
309 t->det_rec_len = ext4_rec_len_to_disk(
310 sizeof(struct ext4_dir_entry_tail), blocksize);
311 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
312}
313
314/* Walk through a dirent block to find a checksum "dirent" at the tail */
315static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
316 struct buffer_head *bh)
317{
318 struct ext4_dir_entry_tail *t;
319
320#ifdef PARANOID
321 struct ext4_dir_entry *d, *top;
322
323 d = (struct ext4_dir_entry *)bh->b_data;
324 top = (struct ext4_dir_entry *)(bh->b_data +
325 (EXT4_BLOCK_SIZE(inode->i_sb) -
326 sizeof(struct ext4_dir_entry_tail)));
327 while (d < top && d->rec_len)
328 d = (struct ext4_dir_entry *)(((void *)d) +
329 le16_to_cpu(d->rec_len));
330
331 if (d != top)
332 return NULL;
333
334 t = (struct ext4_dir_entry_tail *)d;
335#else
336 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
337#endif
338
339 if (t->det_reserved_zero1 ||
340 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
341 t->det_reserved_zero2 ||
342 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
343 return NULL;
344
345 return t;
346}
347
348static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
349{
350 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
351 struct ext4_inode_info *ei = EXT4_I(inode);
352 __u32 csum;
353
354 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
355 return cpu_to_le32(csum);
356}
357
358#define warn_no_space_for_csum(inode) \
359 __warn_no_space_for_csum((inode), __func__, __LINE__)
360
361static void __warn_no_space_for_csum(struct inode *inode, const char *func,
362 unsigned int line)
363{
364 __ext4_warning_inode(inode, func, line,
365 "No space for directory leaf checksum. Please run e2fsck -D.");
366}
367
368int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
369{
370 struct ext4_dir_entry_tail *t;
371
372 if (!ext4_has_metadata_csum(inode->i_sb))
373 return 1;
374
375 t = get_dirent_tail(inode, bh);
376 if (!t) {
377 warn_no_space_for_csum(inode);
378 return 0;
379 }
380
381 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
382 (char *)t - bh->b_data))
383 return 0;
384
385 return 1;
386}
387
388static void ext4_dirblock_csum_set(struct inode *inode,
389 struct buffer_head *bh)
390{
391 struct ext4_dir_entry_tail *t;
392
393 if (!ext4_has_metadata_csum(inode->i_sb))
394 return;
395
396 t = get_dirent_tail(inode, bh);
397 if (!t) {
398 warn_no_space_for_csum(inode);
399 return;
400 }
401
402 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
403 (char *)t - bh->b_data);
404}
405
406int ext4_handle_dirty_dirblock(handle_t *handle,
407 struct inode *inode,
408 struct buffer_head *bh)
409{
410 ext4_dirblock_csum_set(inode, bh);
411 return ext4_handle_dirty_metadata(handle, inode, bh);
412}
413
414static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
415 struct ext4_dir_entry *dirent,
416 int *offset)
417{
418 struct ext4_dir_entry *dp;
419 struct dx_root_info *root;
420 int count_offset;
421
422 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
423 count_offset = 8;
424 else if (le16_to_cpu(dirent->rec_len) == 12) {
425 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
426 if (le16_to_cpu(dp->rec_len) !=
427 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
428 return NULL;
429 root = (struct dx_root_info *)(((void *)dp + 12));
430 if (root->reserved_zero ||
431 root->info_length != sizeof(struct dx_root_info))
432 return NULL;
433 count_offset = 32;
434 } else
435 return NULL;
436
437 if (offset)
438 *offset = count_offset;
439 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
440}
441
442static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
443 int count_offset, int count, struct dx_tail *t)
444{
445 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
446 struct ext4_inode_info *ei = EXT4_I(inode);
447 __u32 csum;
448 int size;
449 __u32 dummy_csum = 0;
450 int offset = offsetof(struct dx_tail, dt_checksum);
451
452 size = count_offset + (count * sizeof(struct dx_entry));
453 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
454 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
455 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
456
457 return cpu_to_le32(csum);
458}
459
460static int ext4_dx_csum_verify(struct inode *inode,
461 struct ext4_dir_entry *dirent)
462{
463 struct dx_countlimit *c;
464 struct dx_tail *t;
465 int count_offset, limit, count;
466
467 if (!ext4_has_metadata_csum(inode->i_sb))
468 return 1;
469
470 c = get_dx_countlimit(inode, dirent, &count_offset);
471 if (!c) {
472 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
473 return 0;
474 }
475 limit = le16_to_cpu(c->limit);
476 count = le16_to_cpu(c->count);
477 if (count_offset + (limit * sizeof(struct dx_entry)) >
478 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
479 warn_no_space_for_csum(inode);
480 return 0;
481 }
482 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
483
484 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
485 count, t))
486 return 0;
487 return 1;
488}
489
490static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
491{
492 struct dx_countlimit *c;
493 struct dx_tail *t;
494 int count_offset, limit, count;
495
496 if (!ext4_has_metadata_csum(inode->i_sb))
497 return;
498
499 c = get_dx_countlimit(inode, dirent, &count_offset);
500 if (!c) {
501 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
502 return;
503 }
504 limit = le16_to_cpu(c->limit);
505 count = le16_to_cpu(c->count);
506 if (count_offset + (limit * sizeof(struct dx_entry)) >
507 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
508 warn_no_space_for_csum(inode);
509 return;
510 }
511 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
512
513 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
514}
515
516static inline int ext4_handle_dirty_dx_node(handle_t *handle,
517 struct inode *inode,
518 struct buffer_head *bh)
519{
520 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
521 return ext4_handle_dirty_metadata(handle, inode, bh);
522}
523
524/*
525 * p is at least 6 bytes before the end of page
526 */
527static inline struct ext4_dir_entry_2 *
528ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
529{
530 return (struct ext4_dir_entry_2 *)((char *)p +
531 ext4_rec_len_from_disk(p->rec_len, blocksize));
532}
533
534/*
535 * Future: use high four bits of block for coalesce-on-delete flags
536 * Mask them off for now.
537 */
538
539static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
540{
541 return le32_to_cpu(entry->block) & 0x0fffffff;
542}
543
544static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
545{
546 entry->block = cpu_to_le32(value);
547}
548
549static inline unsigned dx_get_hash(struct dx_entry *entry)
550{
551 return le32_to_cpu(entry->hash);
552}
553
554static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
555{
556 entry->hash = cpu_to_le32(value);
557}
558
559static inline unsigned dx_get_count(struct dx_entry *entries)
560{
561 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
562}
563
564static inline unsigned dx_get_limit(struct dx_entry *entries)
565{
566 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
567}
568
569static inline void dx_set_count(struct dx_entry *entries, unsigned value)
570{
571 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
572}
573
574static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
575{
576 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
577}
578
579static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
580{
581 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
582 EXT4_DIR_REC_LEN(2) - infosize;
583
584 if (ext4_has_metadata_csum(dir->i_sb))
585 entry_space -= sizeof(struct dx_tail);
586 return entry_space / sizeof(struct dx_entry);
587}
588
589static inline unsigned dx_node_limit(struct inode *dir)
590{
591 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
592
593 if (ext4_has_metadata_csum(dir->i_sb))
594 entry_space -= sizeof(struct dx_tail);
595 return entry_space / sizeof(struct dx_entry);
596}
597
598/*
599 * Debug
600 */
601#ifdef DX_DEBUG
602static void dx_show_index(char * label, struct dx_entry *entries)
603{
604 int i, n = dx_get_count (entries);
605 printk(KERN_DEBUG "%s index", label);
606 for (i = 0; i < n; i++) {
607 printk(KERN_CONT " %x->%lu",
608 i ? dx_get_hash(entries + i) : 0,
609 (unsigned long)dx_get_block(entries + i));
610 }
611 printk(KERN_CONT "\n");
612}
613
614struct stats
615{
616 unsigned names;
617 unsigned space;
618 unsigned bcount;
619};
620
621static struct stats dx_show_leaf(struct inode *dir,
622 struct dx_hash_info *hinfo,
623 struct ext4_dir_entry_2 *de,
624 int size, int show_names)
625{
626 unsigned names = 0, space = 0;
627 char *base = (char *) de;
628 struct dx_hash_info h = *hinfo;
629
630 printk("names: ");
631 while ((char *) de < base + size)
632 {
633 if (de->inode)
634 {
635 if (show_names)
636 {
637#ifdef CONFIG_FS_ENCRYPTION
638 int len;
639 char *name;
640 struct fscrypt_str fname_crypto_str =
641 FSTR_INIT(NULL, 0);
642 int res = 0;
643
644 name = de->name;
645 len = de->name_len;
646 if (IS_ENCRYPTED(dir))
647 res = fscrypt_get_encryption_info(dir);
648 if (res) {
649 printk(KERN_WARNING "Error setting up"
650 " fname crypto: %d\n", res);
651 }
652 if (!fscrypt_has_encryption_key(dir)) {
653 /* Directory is not encrypted */
654 ext4fs_dirhash(dir, de->name,
655 de->name_len, &h);
656 printk("%*.s:(U)%x.%u ", len,
657 name, h.hash,
658 (unsigned) ((char *) de
659 - base));
660 } else {
661 struct fscrypt_str de_name =
662 FSTR_INIT(name, len);
663
664 /* Directory is encrypted */
665 res = fscrypt_fname_alloc_buffer(
666 dir, len,
667 &fname_crypto_str);
668 if (res)
669 printk(KERN_WARNING "Error "
670 "allocating crypto "
671 "buffer--skipping "
672 "crypto\n");
673 res = fscrypt_fname_disk_to_usr(dir,
674 0, 0, &de_name,
675 &fname_crypto_str);
676 if (res) {
677 printk(KERN_WARNING "Error "
678 "converting filename "
679 "from disk to usr"
680 "\n");
681 name = "??";
682 len = 2;
683 } else {
684 name = fname_crypto_str.name;
685 len = fname_crypto_str.len;
686 }
687 ext4fs_dirhash(dir, de->name,
688 de->name_len, &h);
689 printk("%*.s:(E)%x.%u ", len, name,
690 h.hash, (unsigned) ((char *) de
691 - base));
692 fscrypt_fname_free_buffer(
693 &fname_crypto_str);
694 }
695#else
696 int len = de->name_len;
697 char *name = de->name;
698 ext4fs_dirhash(dir, de->name, de->name_len, &h);
699 printk("%*.s:%x.%u ", len, name, h.hash,
700 (unsigned) ((char *) de - base));
701#endif
702 }
703 space += EXT4_DIR_REC_LEN(de->name_len);
704 names++;
705 }
706 de = ext4_next_entry(de, size);
707 }
708 printk(KERN_CONT "(%i)\n", names);
709 return (struct stats) { names, space, 1 };
710}
711
712struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
713 struct dx_entry *entries, int levels)
714{
715 unsigned blocksize = dir->i_sb->s_blocksize;
716 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
717 unsigned bcount = 0;
718 struct buffer_head *bh;
719 printk("%i indexed blocks...\n", count);
720 for (i = 0; i < count; i++, entries++)
721 {
722 ext4_lblk_t block = dx_get_block(entries);
723 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
724 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
725 struct stats stats;
726 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
727 bh = ext4_bread(NULL,dir, block, 0);
728 if (!bh || IS_ERR(bh))
729 continue;
730 stats = levels?
731 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
732 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
733 bh->b_data, blocksize, 0);
734 names += stats.names;
735 space += stats.space;
736 bcount += stats.bcount;
737 brelse(bh);
738 }
739 if (bcount)
740 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
741 levels ? "" : " ", names, space/bcount,
742 (space/bcount)*100/blocksize);
743 return (struct stats) { names, space, bcount};
744}
745#endif /* DX_DEBUG */
746
747/*
748 * Probe for a directory leaf block to search.
749 *
750 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
751 * error in the directory index, and the caller should fall back to
752 * searching the directory normally. The callers of dx_probe **MUST**
753 * check for this error code, and make sure it never gets reflected
754 * back to userspace.
755 */
756static struct dx_frame *
757dx_probe(struct ext4_filename *fname, struct inode *dir,
758 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
759{
760 unsigned count, indirect;
761 struct dx_entry *at, *entries, *p, *q, *m;
762 struct dx_root *root;
763 struct dx_frame *frame = frame_in;
764 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
765 u32 hash;
766
767 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
768 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
769 if (IS_ERR(frame->bh))
770 return (struct dx_frame *) frame->bh;
771
772 root = (struct dx_root *) frame->bh->b_data;
773 if (root->info.hash_version != DX_HASH_TEA &&
774 root->info.hash_version != DX_HASH_HALF_MD4 &&
775 root->info.hash_version != DX_HASH_LEGACY) {
776 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
777 root->info.hash_version);
778 goto fail;
779 }
780 if (fname)
781 hinfo = &fname->hinfo;
782 hinfo->hash_version = root->info.hash_version;
783 if (hinfo->hash_version <= DX_HASH_TEA)
784 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
785 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
786 if (fname && fname_name(fname))
787 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
788 hash = hinfo->hash;
789
790 if (root->info.unused_flags & 1) {
791 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
792 root->info.unused_flags);
793 goto fail;
794 }
795
796 indirect = root->info.indirect_levels;
797 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
798 ext4_warning(dir->i_sb,
799 "Directory (ino: %lu) htree depth %#06x exceed"
800 "supported value", dir->i_ino,
801 ext4_dir_htree_level(dir->i_sb));
802 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
803 ext4_warning(dir->i_sb, "Enable large directory "
804 "feature to access it");
805 }
806 goto fail;
807 }
808
809 entries = (struct dx_entry *)(((char *)&root->info) +
810 root->info.info_length);
811
812 if (dx_get_limit(entries) != dx_root_limit(dir,
813 root->info.info_length)) {
814 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
815 dx_get_limit(entries),
816 dx_root_limit(dir, root->info.info_length));
817 goto fail;
818 }
819
820 dxtrace(printk("Look up %x", hash));
821 while (1) {
822 count = dx_get_count(entries);
823 if (!count || count > dx_get_limit(entries)) {
824 ext4_warning_inode(dir,
825 "dx entry: count %u beyond limit %u",
826 count, dx_get_limit(entries));
827 goto fail;
828 }
829
830 p = entries + 1;
831 q = entries + count - 1;
832 while (p <= q) {
833 m = p + (q - p) / 2;
834 dxtrace(printk(KERN_CONT "."));
835 if (dx_get_hash(m) > hash)
836 q = m - 1;
837 else
838 p = m + 1;
839 }
840
841 if (0) { // linear search cross check
842 unsigned n = count - 1;
843 at = entries;
844 while (n--)
845 {
846 dxtrace(printk(KERN_CONT ","));
847 if (dx_get_hash(++at) > hash)
848 {
849 at--;
850 break;
851 }
852 }
853 assert (at == p - 1);
854 }
855
856 at = p - 1;
857 dxtrace(printk(KERN_CONT " %x->%u\n",
858 at == entries ? 0 : dx_get_hash(at),
859 dx_get_block(at)));
860 frame->entries = entries;
861 frame->at = at;
862 if (!indirect--)
863 return frame;
864 frame++;
865 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
866 if (IS_ERR(frame->bh)) {
867 ret_err = (struct dx_frame *) frame->bh;
868 frame->bh = NULL;
869 goto fail;
870 }
871 entries = ((struct dx_node *) frame->bh->b_data)->entries;
872
873 if (dx_get_limit(entries) != dx_node_limit(dir)) {
874 ext4_warning_inode(dir,
875 "dx entry: limit %u != node limit %u",
876 dx_get_limit(entries), dx_node_limit(dir));
877 goto fail;
878 }
879 }
880fail:
881 while (frame >= frame_in) {
882 brelse(frame->bh);
883 frame--;
884 }
885
886 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
887 ext4_warning_inode(dir,
888 "Corrupt directory, running e2fsck is recommended");
889 return ret_err;
890}
891
892static void dx_release(struct dx_frame *frames)
893{
894 struct dx_root_info *info;
895 int i;
896 unsigned int indirect_levels;
897
898 if (frames[0].bh == NULL)
899 return;
900
901 info = &((struct dx_root *)frames[0].bh->b_data)->info;
902 /* save local copy, "info" may be freed after brelse() */
903 indirect_levels = info->indirect_levels;
904 for (i = 0; i <= indirect_levels; i++) {
905 if (frames[i].bh == NULL)
906 break;
907 brelse(frames[i].bh);
908 frames[i].bh = NULL;
909 }
910}
911
912/*
913 * This function increments the frame pointer to search the next leaf
914 * block, and reads in the necessary intervening nodes if the search
915 * should be necessary. Whether or not the search is necessary is
916 * controlled by the hash parameter. If the hash value is even, then
917 * the search is only continued if the next block starts with that
918 * hash value. This is used if we are searching for a specific file.
919 *
920 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
921 *
922 * This function returns 1 if the caller should continue to search,
923 * or 0 if it should not. If there is an error reading one of the
924 * index blocks, it will a negative error code.
925 *
926 * If start_hash is non-null, it will be filled in with the starting
927 * hash of the next page.
928 */
929static int ext4_htree_next_block(struct inode *dir, __u32 hash,
930 struct dx_frame *frame,
931 struct dx_frame *frames,
932 __u32 *start_hash)
933{
934 struct dx_frame *p;
935 struct buffer_head *bh;
936 int num_frames = 0;
937 __u32 bhash;
938
939 p = frame;
940 /*
941 * Find the next leaf page by incrementing the frame pointer.
942 * If we run out of entries in the interior node, loop around and
943 * increment pointer in the parent node. When we break out of
944 * this loop, num_frames indicates the number of interior
945 * nodes need to be read.
946 */
947 while (1) {
948 if (++(p->at) < p->entries + dx_get_count(p->entries))
949 break;
950 if (p == frames)
951 return 0;
952 num_frames++;
953 p--;
954 }
955
956 /*
957 * If the hash is 1, then continue only if the next page has a
958 * continuation hash of any value. This is used for readdir
959 * handling. Otherwise, check to see if the hash matches the
960 * desired contiuation hash. If it doesn't, return since
961 * there's no point to read in the successive index pages.
962 */
963 bhash = dx_get_hash(p->at);
964 if (start_hash)
965 *start_hash = bhash;
966 if ((hash & 1) == 0) {
967 if ((bhash & ~1) != hash)
968 return 0;
969 }
970 /*
971 * If the hash is HASH_NB_ALWAYS, we always go to the next
972 * block so no check is necessary
973 */
974 while (num_frames--) {
975 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
976 if (IS_ERR(bh))
977 return PTR_ERR(bh);
978 p++;
979 brelse(p->bh);
980 p->bh = bh;
981 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
982 }
983 return 1;
984}
985
986
987/*
988 * This function fills a red-black tree with information from a
989 * directory block. It returns the number directory entries loaded
990 * into the tree. If there is an error it is returned in err.
991 */
992static int htree_dirblock_to_tree(struct file *dir_file,
993 struct inode *dir, ext4_lblk_t block,
994 struct dx_hash_info *hinfo,
995 __u32 start_hash, __u32 start_minor_hash)
996{
997 struct buffer_head *bh;
998 struct ext4_dir_entry_2 *de, *top;
999 int err = 0, count = 0;
1000 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1001
1002 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1003 (unsigned long)block));
1004 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1005 if (IS_ERR(bh))
1006 return PTR_ERR(bh);
1007
1008 de = (struct ext4_dir_entry_2 *) bh->b_data;
1009 top = (struct ext4_dir_entry_2 *) ((char *) de +
1010 dir->i_sb->s_blocksize -
1011 EXT4_DIR_REC_LEN(0));
1012 /* Check if the directory is encrypted */
1013 if (IS_ENCRYPTED(dir)) {
1014 err = fscrypt_get_encryption_info(dir);
1015 if (err < 0) {
1016 brelse(bh);
1017 return err;
1018 }
1019 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1020 &fname_crypto_str);
1021 if (err < 0) {
1022 brelse(bh);
1023 return err;
1024 }
1025 }
1026
1027 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1028 if (ext4_check_dir_entry(dir, NULL, de, bh,
1029 bh->b_data, bh->b_size,
1030 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1031 + ((char *)de - bh->b_data))) {
1032 /* silently ignore the rest of the block */
1033 break;
1034 }
1035 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1036 if ((hinfo->hash < start_hash) ||
1037 ((hinfo->hash == start_hash) &&
1038 (hinfo->minor_hash < start_minor_hash)))
1039 continue;
1040 if (de->inode == 0)
1041 continue;
1042 if (!IS_ENCRYPTED(dir)) {
1043 tmp_str.name = de->name;
1044 tmp_str.len = de->name_len;
1045 err = ext4_htree_store_dirent(dir_file,
1046 hinfo->hash, hinfo->minor_hash, de,
1047 &tmp_str);
1048 } else {
1049 int save_len = fname_crypto_str.len;
1050 struct fscrypt_str de_name = FSTR_INIT(de->name,
1051 de->name_len);
1052
1053 /* Directory is encrypted */
1054 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1055 hinfo->minor_hash, &de_name,
1056 &fname_crypto_str);
1057 if (err) {
1058 count = err;
1059 goto errout;
1060 }
1061 err = ext4_htree_store_dirent(dir_file,
1062 hinfo->hash, hinfo->minor_hash, de,
1063 &fname_crypto_str);
1064 fname_crypto_str.len = save_len;
1065 }
1066 if (err != 0) {
1067 count = err;
1068 goto errout;
1069 }
1070 count++;
1071 }
1072errout:
1073 brelse(bh);
1074 fscrypt_fname_free_buffer(&fname_crypto_str);
1075 return count;
1076}
1077
1078
1079/*
1080 * This function fills a red-black tree with information from a
1081 * directory. We start scanning the directory in hash order, starting
1082 * at start_hash and start_minor_hash.
1083 *
1084 * This function returns the number of entries inserted into the tree,
1085 * or a negative error code.
1086 */
1087int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1088 __u32 start_minor_hash, __u32 *next_hash)
1089{
1090 struct dx_hash_info hinfo;
1091 struct ext4_dir_entry_2 *de;
1092 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1093 struct inode *dir;
1094 ext4_lblk_t block;
1095 int count = 0;
1096 int ret, err;
1097 __u32 hashval;
1098 struct fscrypt_str tmp_str;
1099
1100 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1101 start_hash, start_minor_hash));
1102 dir = file_inode(dir_file);
1103 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1104 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1105 if (hinfo.hash_version <= DX_HASH_TEA)
1106 hinfo.hash_version +=
1107 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1108 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1109 if (ext4_has_inline_data(dir)) {
1110 int has_inline_data = 1;
1111 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1112 &hinfo, start_hash,
1113 start_minor_hash,
1114 &has_inline_data);
1115 if (has_inline_data) {
1116 *next_hash = ~0;
1117 return count;
1118 }
1119 }
1120 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1121 start_hash, start_minor_hash);
1122 *next_hash = ~0;
1123 return count;
1124 }
1125 hinfo.hash = start_hash;
1126 hinfo.minor_hash = 0;
1127 frame = dx_probe(NULL, dir, &hinfo, frames);
1128 if (IS_ERR(frame))
1129 return PTR_ERR(frame);
1130
1131 /* Add '.' and '..' from the htree header */
1132 if (!start_hash && !start_minor_hash) {
1133 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1134 tmp_str.name = de->name;
1135 tmp_str.len = de->name_len;
1136 err = ext4_htree_store_dirent(dir_file, 0, 0,
1137 de, &tmp_str);
1138 if (err != 0)
1139 goto errout;
1140 count++;
1141 }
1142 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1143 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1144 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1145 tmp_str.name = de->name;
1146 tmp_str.len = de->name_len;
1147 err = ext4_htree_store_dirent(dir_file, 2, 0,
1148 de, &tmp_str);
1149 if (err != 0)
1150 goto errout;
1151 count++;
1152 }
1153
1154 while (1) {
1155 if (fatal_signal_pending(current)) {
1156 err = -ERESTARTSYS;
1157 goto errout;
1158 }
1159 cond_resched();
1160 block = dx_get_block(frame->at);
1161 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1162 start_hash, start_minor_hash);
1163 if (ret < 0) {
1164 err = ret;
1165 goto errout;
1166 }
1167 count += ret;
1168 hashval = ~0;
1169 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1170 frame, frames, &hashval);
1171 *next_hash = hashval;
1172 if (ret < 0) {
1173 err = ret;
1174 goto errout;
1175 }
1176 /*
1177 * Stop if: (a) there are no more entries, or
1178 * (b) we have inserted at least one entry and the
1179 * next hash value is not a continuation
1180 */
1181 if ((ret == 0) ||
1182 (count && ((hashval & 1) == 0)))
1183 break;
1184 }
1185 dx_release(frames);
1186 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1187 "next hash: %x\n", count, *next_hash));
1188 return count;
1189errout:
1190 dx_release(frames);
1191 return (err);
1192}
1193
1194static inline int search_dirblock(struct buffer_head *bh,
1195 struct inode *dir,
1196 struct ext4_filename *fname,
1197 unsigned int offset,
1198 struct ext4_dir_entry_2 **res_dir)
1199{
1200 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1201 fname, offset, res_dir);
1202}
1203
1204/*
1205 * Directory block splitting, compacting
1206 */
1207
1208/*
1209 * Create map of hash values, offsets, and sizes, stored at end of block.
1210 * Returns number of entries mapped.
1211 */
1212static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1213 unsigned blocksize, struct dx_hash_info *hinfo,
1214 struct dx_map_entry *map_tail)
1215{
1216 int count = 0;
1217 char *base = (char *) de;
1218 struct dx_hash_info h = *hinfo;
1219
1220 while ((char *) de < base + blocksize) {
1221 if (de->name_len && de->inode) {
1222 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1223 map_tail--;
1224 map_tail->hash = h.hash;
1225 map_tail->offs = ((char *) de - base)>>2;
1226 map_tail->size = le16_to_cpu(de->rec_len);
1227 count++;
1228 cond_resched();
1229 }
1230 /* XXX: do we need to check rec_len == 0 case? -Chris */
1231 de = ext4_next_entry(de, blocksize);
1232 }
1233 return count;
1234}
1235
1236/* Sort map by hash value */
1237static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1238{
1239 struct dx_map_entry *p, *q, *top = map + count - 1;
1240 int more;
1241 /* Combsort until bubble sort doesn't suck */
1242 while (count > 2) {
1243 count = count*10/13;
1244 if (count - 9 < 2) /* 9, 10 -> 11 */
1245 count = 11;
1246 for (p = top, q = p - count; q >= map; p--, q--)
1247 if (p->hash < q->hash)
1248 swap(*p, *q);
1249 }
1250 /* Garden variety bubble sort */
1251 do {
1252 more = 0;
1253 q = top;
1254 while (q-- > map) {
1255 if (q[1].hash >= q[0].hash)
1256 continue;
1257 swap(*(q+1), *q);
1258 more = 1;
1259 }
1260 } while(more);
1261}
1262
1263static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1264{
1265 struct dx_entry *entries = frame->entries;
1266 struct dx_entry *old = frame->at, *new = old + 1;
1267 int count = dx_get_count(entries);
1268
1269 assert(count < dx_get_limit(entries));
1270 assert(old < entries + count);
1271 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1272 dx_set_hash(new, hash);
1273 dx_set_block(new, block);
1274 dx_set_count(entries, count + 1);
1275}
1276
1277#ifdef CONFIG_UNICODE
1278/*
1279 * Test whether a case-insensitive directory entry matches the filename
1280 * being searched for. If quick is set, assume the name being looked up
1281 * is already in the casefolded form.
1282 *
1283 * Returns: 0 if the directory entry matches, more than 0 if it
1284 * doesn't match or less than zero on error.
1285 */
1286int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1287 const struct qstr *entry, bool quick)
1288{
1289 const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1290 const struct unicode_map *um = sbi->s_encoding;
1291 int ret;
1292
1293 if (quick)
1294 ret = utf8_strncasecmp_folded(um, name, entry);
1295 else
1296 ret = utf8_strncasecmp(um, name, entry);
1297
1298 if (ret < 0) {
1299 /* Handle invalid character sequence as either an error
1300 * or as an opaque byte sequence.
1301 */
1302 if (ext4_has_strict_mode(sbi))
1303 return -EINVAL;
1304
1305 if (name->len != entry->len)
1306 return 1;
1307
1308 return !!memcmp(name->name, entry->name, name->len);
1309 }
1310
1311 return ret;
1312}
1313
1314void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1315 struct fscrypt_str *cf_name)
1316{
1317 int len;
1318
1319 if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
1320 cf_name->name = NULL;
1321 return;
1322 }
1323
1324 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1325 if (!cf_name->name)
1326 return;
1327
1328 len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1329 iname, cf_name->name,
1330 EXT4_NAME_LEN);
1331 if (len <= 0) {
1332 kfree(cf_name->name);
1333 cf_name->name = NULL;
1334 return;
1335 }
1336 cf_name->len = (unsigned) len;
1337
1338}
1339#endif
1340
1341/*
1342 * Test whether a directory entry matches the filename being searched for.
1343 *
1344 * Return: %true if the directory entry matches, otherwise %false.
1345 */
1346static inline bool ext4_match(const struct inode *parent,
1347 const struct ext4_filename *fname,
1348 const struct ext4_dir_entry_2 *de)
1349{
1350 struct fscrypt_name f;
1351#ifdef CONFIG_UNICODE
1352 const struct qstr entry = {.name = de->name, .len = de->name_len};
1353#endif
1354
1355 if (!de->inode)
1356 return false;
1357
1358 f.usr_fname = fname->usr_fname;
1359 f.disk_name = fname->disk_name;
1360#ifdef CONFIG_FS_ENCRYPTION
1361 f.crypto_buf = fname->crypto_buf;
1362#endif
1363
1364#ifdef CONFIG_UNICODE
1365 if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1366 if (fname->cf_name.name) {
1367 struct qstr cf = {.name = fname->cf_name.name,
1368 .len = fname->cf_name.len};
1369 return !ext4_ci_compare(parent, &cf, &entry, true);
1370 }
1371 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1372 false);
1373 }
1374#endif
1375
1376 return fscrypt_match_name(&f, de->name, de->name_len);
1377}
1378
1379/*
1380 * Returns 0 if not found, -1 on failure, and 1 on success
1381 */
1382int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1383 struct inode *dir, struct ext4_filename *fname,
1384 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1385{
1386 struct ext4_dir_entry_2 * de;
1387 char * dlimit;
1388 int de_len;
1389
1390 de = (struct ext4_dir_entry_2 *)search_buf;
1391 dlimit = search_buf + buf_size;
1392 while ((char *) de < dlimit) {
1393 /* this code is executed quadratically often */
1394 /* do minimal checking `by hand' */
1395 if ((char *) de + de->name_len <= dlimit &&
1396 ext4_match(dir, fname, de)) {
1397 /* found a match - just to be sure, do
1398 * a full check */
1399 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1400 buf_size, offset))
1401 return -1;
1402 *res_dir = de;
1403 return 1;
1404 }
1405 /* prevent looping on a bad block */
1406 de_len = ext4_rec_len_from_disk(de->rec_len,
1407 dir->i_sb->s_blocksize);
1408 if (de_len <= 0)
1409 return -1;
1410 offset += de_len;
1411 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1412 }
1413 return 0;
1414}
1415
1416static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1417 struct ext4_dir_entry *de)
1418{
1419 struct super_block *sb = dir->i_sb;
1420
1421 if (!is_dx(dir))
1422 return 0;
1423 if (block == 0)
1424 return 1;
1425 if (de->inode == 0 &&
1426 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1427 sb->s_blocksize)
1428 return 1;
1429 return 0;
1430}
1431
1432/*
1433 * __ext4_find_entry()
1434 *
1435 * finds an entry in the specified directory with the wanted name. It
1436 * returns the cache buffer in which the entry was found, and the entry
1437 * itself (as a parameter - res_dir). It does NOT read the inode of the
1438 * entry - you'll have to do that yourself if you want to.
1439 *
1440 * The returned buffer_head has ->b_count elevated. The caller is expected
1441 * to brelse() it when appropriate.
1442 */
1443static struct buffer_head *__ext4_find_entry(struct inode *dir,
1444 struct ext4_filename *fname,
1445 struct ext4_dir_entry_2 **res_dir,
1446 int *inlined)
1447{
1448 struct super_block *sb;
1449 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1450 struct buffer_head *bh, *ret = NULL;
1451 ext4_lblk_t start, block;
1452 const u8 *name = fname->usr_fname->name;
1453 size_t ra_max = 0; /* Number of bh's in the readahead
1454 buffer, bh_use[] */
1455 size_t ra_ptr = 0; /* Current index into readahead
1456 buffer */
1457 ext4_lblk_t nblocks;
1458 int i, namelen, retval;
1459
1460 *res_dir = NULL;
1461 sb = dir->i_sb;
1462 namelen = fname->usr_fname->len;
1463 if (namelen > EXT4_NAME_LEN)
1464 return NULL;
1465
1466 if (ext4_has_inline_data(dir)) {
1467 int has_inline_data = 1;
1468 ret = ext4_find_inline_entry(dir, fname, res_dir,
1469 &has_inline_data);
1470 if (has_inline_data) {
1471 if (inlined)
1472 *inlined = 1;
1473 goto cleanup_and_exit;
1474 }
1475 }
1476
1477 if ((namelen <= 2) && (name[0] == '.') &&
1478 (name[1] == '.' || name[1] == '\0')) {
1479 /*
1480 * "." or ".." will only be in the first block
1481 * NFS may look up ".."; "." should be handled by the VFS
1482 */
1483 block = start = 0;
1484 nblocks = 1;
1485 goto restart;
1486 }
1487 if (is_dx(dir)) {
1488 ret = ext4_dx_find_entry(dir, fname, res_dir);
1489 /*
1490 * On success, or if the error was file not found,
1491 * return. Otherwise, fall back to doing a search the
1492 * old fashioned way.
1493 */
1494 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1495 goto cleanup_and_exit;
1496 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1497 "falling back\n"));
1498 ret = NULL;
1499 }
1500 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1501 if (!nblocks) {
1502 ret = NULL;
1503 goto cleanup_and_exit;
1504 }
1505 start = EXT4_I(dir)->i_dir_start_lookup;
1506 if (start >= nblocks)
1507 start = 0;
1508 block = start;
1509restart:
1510 do {
1511 /*
1512 * We deal with the read-ahead logic here.
1513 */
1514 cond_resched();
1515 if (ra_ptr >= ra_max) {
1516 /* Refill the readahead buffer */
1517 ra_ptr = 0;
1518 if (block < start)
1519 ra_max = start - block;
1520 else
1521 ra_max = nblocks - block;
1522 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1523 retval = ext4_bread_batch(dir, block, ra_max,
1524 false /* wait */, bh_use);
1525 if (retval) {
1526 ret = ERR_PTR(retval);
1527 ra_max = 0;
1528 goto cleanup_and_exit;
1529 }
1530 }
1531 if ((bh = bh_use[ra_ptr++]) == NULL)
1532 goto next;
1533 wait_on_buffer(bh);
1534 if (!buffer_uptodate(bh)) {
1535 EXT4_ERROR_INODE_ERR(dir, EIO,
1536 "reading directory lblock %lu",
1537 (unsigned long) block);
1538 brelse(bh);
1539 ret = ERR_PTR(-EIO);
1540 goto cleanup_and_exit;
1541 }
1542 if (!buffer_verified(bh) &&
1543 !is_dx_internal_node(dir, block,
1544 (struct ext4_dir_entry *)bh->b_data) &&
1545 !ext4_dirblock_csum_verify(dir, bh)) {
1546 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1547 "checksumming directory "
1548 "block %lu", (unsigned long)block);
1549 brelse(bh);
1550 ret = ERR_PTR(-EFSBADCRC);
1551 goto cleanup_and_exit;
1552 }
1553 set_buffer_verified(bh);
1554 i = search_dirblock(bh, dir, fname,
1555 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1556 if (i == 1) {
1557 EXT4_I(dir)->i_dir_start_lookup = block;
1558 ret = bh;
1559 goto cleanup_and_exit;
1560 } else {
1561 brelse(bh);
1562 if (i < 0)
1563 goto cleanup_and_exit;
1564 }
1565 next:
1566 if (++block >= nblocks)
1567 block = 0;
1568 } while (block != start);
1569
1570 /*
1571 * If the directory has grown while we were searching, then
1572 * search the last part of the directory before giving up.
1573 */
1574 block = nblocks;
1575 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1576 if (block < nblocks) {
1577 start = 0;
1578 goto restart;
1579 }
1580
1581cleanup_and_exit:
1582 /* Clean up the read-ahead blocks */
1583 for (; ra_ptr < ra_max; ra_ptr++)
1584 brelse(bh_use[ra_ptr]);
1585 return ret;
1586}
1587
1588static struct buffer_head *ext4_find_entry(struct inode *dir,
1589 const struct qstr *d_name,
1590 struct ext4_dir_entry_2 **res_dir,
1591 int *inlined)
1592{
1593 int err;
1594 struct ext4_filename fname;
1595 struct buffer_head *bh;
1596
1597 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1598 if (err == -ENOENT)
1599 return NULL;
1600 if (err)
1601 return ERR_PTR(err);
1602
1603 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1604
1605 ext4_fname_free_filename(&fname);
1606 return bh;
1607}
1608
1609static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1610 struct dentry *dentry,
1611 struct ext4_dir_entry_2 **res_dir)
1612{
1613 int err;
1614 struct ext4_filename fname;
1615 struct buffer_head *bh;
1616
1617 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1618 if (err == -ENOENT)
1619 return NULL;
1620 if (err)
1621 return ERR_PTR(err);
1622
1623 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1624
1625 ext4_fname_free_filename(&fname);
1626 return bh;
1627}
1628
1629static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1630 struct ext4_filename *fname,
1631 struct ext4_dir_entry_2 **res_dir)
1632{
1633 struct super_block * sb = dir->i_sb;
1634 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1635 struct buffer_head *bh;
1636 ext4_lblk_t block;
1637 int retval;
1638
1639#ifdef CONFIG_FS_ENCRYPTION
1640 *res_dir = NULL;
1641#endif
1642 frame = dx_probe(fname, dir, NULL, frames);
1643 if (IS_ERR(frame))
1644 return (struct buffer_head *) frame;
1645 do {
1646 block = dx_get_block(frame->at);
1647 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1648 if (IS_ERR(bh))
1649 goto errout;
1650
1651 retval = search_dirblock(bh, dir, fname,
1652 block << EXT4_BLOCK_SIZE_BITS(sb),
1653 res_dir);
1654 if (retval == 1)
1655 goto success;
1656 brelse(bh);
1657 if (retval == -1) {
1658 bh = ERR_PTR(ERR_BAD_DX_DIR);
1659 goto errout;
1660 }
1661
1662 /* Check to see if we should continue to search */
1663 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1664 frames, NULL);
1665 if (retval < 0) {
1666 ext4_warning_inode(dir,
1667 "error %d reading directory index block",
1668 retval);
1669 bh = ERR_PTR(retval);
1670 goto errout;
1671 }
1672 } while (retval == 1);
1673
1674 bh = NULL;
1675errout:
1676 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1677success:
1678 dx_release(frames);
1679 return bh;
1680}
1681
1682static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1683{
1684 struct inode *inode;
1685 struct ext4_dir_entry_2 *de;
1686 struct buffer_head *bh;
1687
1688 if (dentry->d_name.len > EXT4_NAME_LEN)
1689 return ERR_PTR(-ENAMETOOLONG);
1690
1691 bh = ext4_lookup_entry(dir, dentry, &de);
1692 if (IS_ERR(bh))
1693 return ERR_CAST(bh);
1694 inode = NULL;
1695 if (bh) {
1696 __u32 ino = le32_to_cpu(de->inode);
1697 brelse(bh);
1698 if (!ext4_valid_inum(dir->i_sb, ino)) {
1699 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1700 return ERR_PTR(-EFSCORRUPTED);
1701 }
1702 if (unlikely(ino == dir->i_ino)) {
1703 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1704 dentry);
1705 return ERR_PTR(-EFSCORRUPTED);
1706 }
1707 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1708 if (inode == ERR_PTR(-ESTALE)) {
1709 EXT4_ERROR_INODE(dir,
1710 "deleted inode referenced: %u",
1711 ino);
1712 return ERR_PTR(-EFSCORRUPTED);
1713 }
1714 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1715 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1716 !fscrypt_has_permitted_context(dir, inode)) {
1717 ext4_warning(inode->i_sb,
1718 "Inconsistent encryption contexts: %lu/%lu",
1719 dir->i_ino, inode->i_ino);
1720 iput(inode);
1721 return ERR_PTR(-EPERM);
1722 }
1723 }
1724
1725#ifdef CONFIG_UNICODE
1726 if (!inode && IS_CASEFOLDED(dir)) {
1727 /* Eventually we want to call d_add_ci(dentry, NULL)
1728 * for negative dentries in the encoding case as
1729 * well. For now, prevent the negative dentry
1730 * from being cached.
1731 */
1732 return NULL;
1733 }
1734#endif
1735 return d_splice_alias(inode, dentry);
1736}
1737
1738
1739struct dentry *ext4_get_parent(struct dentry *child)
1740{
1741 __u32 ino;
1742 static const struct qstr dotdot = QSTR_INIT("..", 2);
1743 struct ext4_dir_entry_2 * de;
1744 struct buffer_head *bh;
1745
1746 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1747 if (IS_ERR(bh))
1748 return ERR_CAST(bh);
1749 if (!bh)
1750 return ERR_PTR(-ENOENT);
1751 ino = le32_to_cpu(de->inode);
1752 brelse(bh);
1753
1754 if (!ext4_valid_inum(child->d_sb, ino)) {
1755 EXT4_ERROR_INODE(d_inode(child),
1756 "bad parent inode number: %u", ino);
1757 return ERR_PTR(-EFSCORRUPTED);
1758 }
1759
1760 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1761}
1762
1763/*
1764 * Move count entries from end of map between two memory locations.
1765 * Returns pointer to last entry moved.
1766 */
1767static struct ext4_dir_entry_2 *
1768dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1769 unsigned blocksize)
1770{
1771 unsigned rec_len = 0;
1772
1773 while (count--) {
1774 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1775 (from + (map->offs<<2));
1776 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1777 memcpy (to, de, rec_len);
1778 ((struct ext4_dir_entry_2 *) to)->rec_len =
1779 ext4_rec_len_to_disk(rec_len, blocksize);
1780 de->inode = 0;
1781 map++;
1782 to += rec_len;
1783 }
1784 return (struct ext4_dir_entry_2 *) (to - rec_len);
1785}
1786
1787/*
1788 * Compact each dir entry in the range to the minimal rec_len.
1789 * Returns pointer to last entry in range.
1790 */
1791static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1792{
1793 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1794 unsigned rec_len = 0;
1795
1796 prev = to = de;
1797 while ((char*)de < base + blocksize) {
1798 next = ext4_next_entry(de, blocksize);
1799 if (de->inode && de->name_len) {
1800 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1801 if (de > to)
1802 memmove(to, de, rec_len);
1803 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1804 prev = to;
1805 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1806 }
1807 de = next;
1808 }
1809 return prev;
1810}
1811
1812/*
1813 * Split a full leaf block to make room for a new dir entry.
1814 * Allocate a new block, and move entries so that they are approx. equally full.
1815 * Returns pointer to de in block into which the new entry will be inserted.
1816 */
1817static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1818 struct buffer_head **bh,struct dx_frame *frame,
1819 struct dx_hash_info *hinfo)
1820{
1821 unsigned blocksize = dir->i_sb->s_blocksize;
1822 unsigned count, continued;
1823 struct buffer_head *bh2;
1824 ext4_lblk_t newblock;
1825 u32 hash2;
1826 struct dx_map_entry *map;
1827 char *data1 = (*bh)->b_data, *data2;
1828 unsigned split, move, size;
1829 struct ext4_dir_entry_2 *de = NULL, *de2;
1830 int csum_size = 0;
1831 int err = 0, i;
1832
1833 if (ext4_has_metadata_csum(dir->i_sb))
1834 csum_size = sizeof(struct ext4_dir_entry_tail);
1835
1836 bh2 = ext4_append(handle, dir, &newblock);
1837 if (IS_ERR(bh2)) {
1838 brelse(*bh);
1839 *bh = NULL;
1840 return (struct ext4_dir_entry_2 *) bh2;
1841 }
1842
1843 BUFFER_TRACE(*bh, "get_write_access");
1844 err = ext4_journal_get_write_access(handle, *bh);
1845 if (err)
1846 goto journal_error;
1847
1848 BUFFER_TRACE(frame->bh, "get_write_access");
1849 err = ext4_journal_get_write_access(handle, frame->bh);
1850 if (err)
1851 goto journal_error;
1852
1853 data2 = bh2->b_data;
1854
1855 /* create map in the end of data2 block */
1856 map = (struct dx_map_entry *) (data2 + blocksize);
1857 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1858 blocksize, hinfo, map);
1859 map -= count;
1860 dx_sort_map(map, count);
1861 /* Ensure that neither split block is over half full */
1862 size = 0;
1863 move = 0;
1864 for (i = count-1; i >= 0; i--) {
1865 /* is more than half of this entry in 2nd half of the block? */
1866 if (size + map[i].size/2 > blocksize/2)
1867 break;
1868 size += map[i].size;
1869 move++;
1870 }
1871 /*
1872 * map index at which we will split
1873 *
1874 * If the sum of active entries didn't exceed half the block size, just
1875 * split it in half by count; each resulting block will have at least
1876 * half the space free.
1877 */
1878 if (i > 0)
1879 split = count - move;
1880 else
1881 split = count/2;
1882
1883 hash2 = map[split].hash;
1884 continued = hash2 == map[split - 1].hash;
1885 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1886 (unsigned long)dx_get_block(frame->at),
1887 hash2, split, count-split));
1888
1889 /* Fancy dance to stay within two buffers */
1890 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1891 blocksize);
1892 de = dx_pack_dirents(data1, blocksize);
1893 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1894 (char *) de,
1895 blocksize);
1896 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1897 (char *) de2,
1898 blocksize);
1899 if (csum_size) {
1900 ext4_initialize_dirent_tail(*bh, blocksize);
1901 ext4_initialize_dirent_tail(bh2, blocksize);
1902 }
1903
1904 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1905 blocksize, 1));
1906 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1907 blocksize, 1));
1908
1909 /* Which block gets the new entry? */
1910 if (hinfo->hash >= hash2) {
1911 swap(*bh, bh2);
1912 de = de2;
1913 }
1914 dx_insert_block(frame, hash2 + continued, newblock);
1915 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1916 if (err)
1917 goto journal_error;
1918 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1919 if (err)
1920 goto journal_error;
1921 brelse(bh2);
1922 dxtrace(dx_show_index("frame", frame->entries));
1923 return de;
1924
1925journal_error:
1926 brelse(*bh);
1927 brelse(bh2);
1928 *bh = NULL;
1929 ext4_std_error(dir->i_sb, err);
1930 return ERR_PTR(err);
1931}
1932
1933int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1934 struct buffer_head *bh,
1935 void *buf, int buf_size,
1936 struct ext4_filename *fname,
1937 struct ext4_dir_entry_2 **dest_de)
1938{
1939 struct ext4_dir_entry_2 *de;
1940 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1941 int nlen, rlen;
1942 unsigned int offset = 0;
1943 char *top;
1944
1945 de = (struct ext4_dir_entry_2 *)buf;
1946 top = buf + buf_size - reclen;
1947 while ((char *) de <= top) {
1948 if (ext4_check_dir_entry(dir, NULL, de, bh,
1949 buf, buf_size, offset))
1950 return -EFSCORRUPTED;
1951 if (ext4_match(dir, fname, de))
1952 return -EEXIST;
1953 nlen = EXT4_DIR_REC_LEN(de->name_len);
1954 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1955 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1956 break;
1957 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1958 offset += rlen;
1959 }
1960 if ((char *) de > top)
1961 return -ENOSPC;
1962
1963 *dest_de = de;
1964 return 0;
1965}
1966
1967void ext4_insert_dentry(struct inode *inode,
1968 struct ext4_dir_entry_2 *de,
1969 int buf_size,
1970 struct ext4_filename *fname)
1971{
1972
1973 int nlen, rlen;
1974
1975 nlen = EXT4_DIR_REC_LEN(de->name_len);
1976 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1977 if (de->inode) {
1978 struct ext4_dir_entry_2 *de1 =
1979 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1980 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1981 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1982 de = de1;
1983 }
1984 de->file_type = EXT4_FT_UNKNOWN;
1985 de->inode = cpu_to_le32(inode->i_ino);
1986 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1987 de->name_len = fname_len(fname);
1988 memcpy(de->name, fname_name(fname), fname_len(fname));
1989}
1990
1991/*
1992 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1993 * it points to a directory entry which is guaranteed to be large
1994 * enough for new directory entry. If de is NULL, then
1995 * add_dirent_to_buf will attempt search the directory block for
1996 * space. It will return -ENOSPC if no space is available, and -EIO
1997 * and -EEXIST if directory entry already exists.
1998 */
1999static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2000 struct inode *dir,
2001 struct inode *inode, struct ext4_dir_entry_2 *de,
2002 struct buffer_head *bh)
2003{
2004 unsigned int blocksize = dir->i_sb->s_blocksize;
2005 int csum_size = 0;
2006 int err, err2;
2007
2008 if (ext4_has_metadata_csum(inode->i_sb))
2009 csum_size = sizeof(struct ext4_dir_entry_tail);
2010
2011 if (!de) {
2012 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2013 blocksize - csum_size, fname, &de);
2014 if (err)
2015 return err;
2016 }
2017 BUFFER_TRACE(bh, "get_write_access");
2018 err = ext4_journal_get_write_access(handle, bh);
2019 if (err) {
2020 ext4_std_error(dir->i_sb, err);
2021 return err;
2022 }
2023
2024 /* By now the buffer is marked for journaling */
2025 ext4_insert_dentry(inode, de, blocksize, fname);
2026
2027 /*
2028 * XXX shouldn't update any times until successful
2029 * completion of syscall, but too many callers depend
2030 * on this.
2031 *
2032 * XXX similarly, too many callers depend on
2033 * ext4_new_inode() setting the times, but error
2034 * recovery deletes the inode, so the worst that can
2035 * happen is that the times are slightly out of date
2036 * and/or different from the directory change time.
2037 */
2038 dir->i_mtime = dir->i_ctime = current_time(dir);
2039 ext4_update_dx_flag(dir);
2040 inode_inc_iversion(dir);
2041 err2 = ext4_mark_inode_dirty(handle, dir);
2042 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2043 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2044 if (err)
2045 ext4_std_error(dir->i_sb, err);
2046 return err ? err : err2;
2047}
2048
2049/*
2050 * This converts a one block unindexed directory to a 3 block indexed
2051 * directory, and adds the dentry to the indexed directory.
2052 */
2053static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2054 struct inode *dir,
2055 struct inode *inode, struct buffer_head *bh)
2056{
2057 struct buffer_head *bh2;
2058 struct dx_root *root;
2059 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2060 struct dx_entry *entries;
2061 struct ext4_dir_entry_2 *de, *de2;
2062 char *data2, *top;
2063 unsigned len;
2064 int retval;
2065 unsigned blocksize;
2066 ext4_lblk_t block;
2067 struct fake_dirent *fde;
2068 int csum_size = 0;
2069
2070 if (ext4_has_metadata_csum(inode->i_sb))
2071 csum_size = sizeof(struct ext4_dir_entry_tail);
2072
2073 blocksize = dir->i_sb->s_blocksize;
2074 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2075 BUFFER_TRACE(bh, "get_write_access");
2076 retval = ext4_journal_get_write_access(handle, bh);
2077 if (retval) {
2078 ext4_std_error(dir->i_sb, retval);
2079 brelse(bh);
2080 return retval;
2081 }
2082 root = (struct dx_root *) bh->b_data;
2083
2084 /* The 0th block becomes the root, move the dirents out */
2085 fde = &root->dotdot;
2086 de = (struct ext4_dir_entry_2 *)((char *)fde +
2087 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2088 if ((char *) de >= (((char *) root) + blocksize)) {
2089 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2090 brelse(bh);
2091 return -EFSCORRUPTED;
2092 }
2093 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2094
2095 /* Allocate new block for the 0th block's dirents */
2096 bh2 = ext4_append(handle, dir, &block);
2097 if (IS_ERR(bh2)) {
2098 brelse(bh);
2099 return PTR_ERR(bh2);
2100 }
2101 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2102 data2 = bh2->b_data;
2103
2104 memcpy(data2, de, len);
2105 de = (struct ext4_dir_entry_2 *) data2;
2106 top = data2 + len;
2107 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2108 de = de2;
2109 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2110 (char *) de, blocksize);
2111
2112 if (csum_size)
2113 ext4_initialize_dirent_tail(bh2, blocksize);
2114
2115 /* Initialize the root; the dot dirents already exist */
2116 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2117 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2118 blocksize);
2119 memset (&root->info, 0, sizeof(root->info));
2120 root->info.info_length = sizeof(root->info);
2121 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2122 entries = root->entries;
2123 dx_set_block(entries, 1);
2124 dx_set_count(entries, 1);
2125 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2126
2127 /* Initialize as for dx_probe */
2128 fname->hinfo.hash_version = root->info.hash_version;
2129 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2130 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2131 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2132 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2133
2134 memset(frames, 0, sizeof(frames));
2135 frame = frames;
2136 frame->entries = entries;
2137 frame->at = entries;
2138 frame->bh = bh;
2139
2140 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2141 if (retval)
2142 goto out_frames;
2143 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2144 if (retval)
2145 goto out_frames;
2146
2147 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2148 if (IS_ERR(de)) {
2149 retval = PTR_ERR(de);
2150 goto out_frames;
2151 }
2152
2153 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2154out_frames:
2155 /*
2156 * Even if the block split failed, we have to properly write
2157 * out all the changes we did so far. Otherwise we can end up
2158 * with corrupted filesystem.
2159 */
2160 if (retval)
2161 ext4_mark_inode_dirty(handle, dir);
2162 dx_release(frames);
2163 brelse(bh2);
2164 return retval;
2165}
2166
2167/*
2168 * ext4_add_entry()
2169 *
2170 * adds a file entry to the specified directory, using the same
2171 * semantics as ext4_find_entry(). It returns NULL if it failed.
2172 *
2173 * NOTE!! The inode part of 'de' is left at 0 - which means you
2174 * may not sleep between calling this and putting something into
2175 * the entry, as someone else might have used it while you slept.
2176 */
2177static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2178 struct inode *inode)
2179{
2180 struct inode *dir = d_inode(dentry->d_parent);
2181 struct buffer_head *bh = NULL;
2182 struct ext4_dir_entry_2 *de;
2183 struct super_block *sb;
2184#ifdef CONFIG_UNICODE
2185 struct ext4_sb_info *sbi;
2186#endif
2187 struct ext4_filename fname;
2188 int retval;
2189 int dx_fallback=0;
2190 unsigned blocksize;
2191 ext4_lblk_t block, blocks;
2192 int csum_size = 0;
2193
2194 if (ext4_has_metadata_csum(inode->i_sb))
2195 csum_size = sizeof(struct ext4_dir_entry_tail);
2196
2197 sb = dir->i_sb;
2198 blocksize = sb->s_blocksize;
2199 if (!dentry->d_name.len)
2200 return -EINVAL;
2201
2202#ifdef CONFIG_UNICODE
2203 sbi = EXT4_SB(sb);
2204 if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2205 sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2206 return -EINVAL;
2207#endif
2208
2209 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2210 if (retval)
2211 return retval;
2212
2213 if (ext4_has_inline_data(dir)) {
2214 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2215 if (retval < 0)
2216 goto out;
2217 if (retval == 1) {
2218 retval = 0;
2219 goto out;
2220 }
2221 }
2222
2223 if (is_dx(dir)) {
2224 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2225 if (!retval || (retval != ERR_BAD_DX_DIR))
2226 goto out;
2227 /* Can we just ignore htree data? */
2228 if (ext4_has_metadata_csum(sb)) {
2229 EXT4_ERROR_INODE(dir,
2230 "Directory has corrupted htree index.");
2231 retval = -EFSCORRUPTED;
2232 goto out;
2233 }
2234 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2235 dx_fallback++;
2236 retval = ext4_mark_inode_dirty(handle, dir);
2237 if (unlikely(retval))
2238 goto out;
2239 }
2240 blocks = dir->i_size >> sb->s_blocksize_bits;
2241 for (block = 0; block < blocks; block++) {
2242 bh = ext4_read_dirblock(dir, block, DIRENT);
2243 if (bh == NULL) {
2244 bh = ext4_bread(handle, dir, block,
2245 EXT4_GET_BLOCKS_CREATE);
2246 goto add_to_new_block;
2247 }
2248 if (IS_ERR(bh)) {
2249 retval = PTR_ERR(bh);
2250 bh = NULL;
2251 goto out;
2252 }
2253 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2254 NULL, bh);
2255 if (retval != -ENOSPC)
2256 goto out;
2257
2258 if (blocks == 1 && !dx_fallback &&
2259 ext4_has_feature_dir_index(sb)) {
2260 retval = make_indexed_dir(handle, &fname, dir,
2261 inode, bh);
2262 bh = NULL; /* make_indexed_dir releases bh */
2263 goto out;
2264 }
2265 brelse(bh);
2266 }
2267 bh = ext4_append(handle, dir, &block);
2268add_to_new_block:
2269 if (IS_ERR(bh)) {
2270 retval = PTR_ERR(bh);
2271 bh = NULL;
2272 goto out;
2273 }
2274 de = (struct ext4_dir_entry_2 *) bh->b_data;
2275 de->inode = 0;
2276 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2277
2278 if (csum_size)
2279 ext4_initialize_dirent_tail(bh, blocksize);
2280
2281 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2282out:
2283 ext4_fname_free_filename(&fname);
2284 brelse(bh);
2285 if (retval == 0)
2286 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2287 return retval;
2288}
2289
2290/*
2291 * Returns 0 for success, or a negative error value
2292 */
2293static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2294 struct inode *dir, struct inode *inode)
2295{
2296 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2297 struct dx_entry *entries, *at;
2298 struct buffer_head *bh;
2299 struct super_block *sb = dir->i_sb;
2300 struct ext4_dir_entry_2 *de;
2301 int restart;
2302 int err;
2303
2304again:
2305 restart = 0;
2306 frame = dx_probe(fname, dir, NULL, frames);
2307 if (IS_ERR(frame))
2308 return PTR_ERR(frame);
2309 entries = frame->entries;
2310 at = frame->at;
2311 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2312 if (IS_ERR(bh)) {
2313 err = PTR_ERR(bh);
2314 bh = NULL;
2315 goto cleanup;
2316 }
2317
2318 BUFFER_TRACE(bh, "get_write_access");
2319 err = ext4_journal_get_write_access(handle, bh);
2320 if (err)
2321 goto journal_error;
2322
2323 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2324 if (err != -ENOSPC)
2325 goto cleanup;
2326
2327 err = 0;
2328 /* Block full, should compress but for now just split */
2329 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2330 dx_get_count(entries), dx_get_limit(entries)));
2331 /* Need to split index? */
2332 if (dx_get_count(entries) == dx_get_limit(entries)) {
2333 ext4_lblk_t newblock;
2334 int levels = frame - frames + 1;
2335 unsigned int icount;
2336 int add_level = 1;
2337 struct dx_entry *entries2;
2338 struct dx_node *node2;
2339 struct buffer_head *bh2;
2340
2341 while (frame > frames) {
2342 if (dx_get_count((frame - 1)->entries) <
2343 dx_get_limit((frame - 1)->entries)) {
2344 add_level = 0;
2345 break;
2346 }
2347 frame--; /* split higher index block */
2348 at = frame->at;
2349 entries = frame->entries;
2350 restart = 1;
2351 }
2352 if (add_level && levels == ext4_dir_htree_level(sb)) {
2353 ext4_warning(sb, "Directory (ino: %lu) index full, "
2354 "reach max htree level :%d",
2355 dir->i_ino, levels);
2356 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2357 ext4_warning(sb, "Large directory feature is "
2358 "not enabled on this "
2359 "filesystem");
2360 }
2361 err = -ENOSPC;
2362 goto cleanup;
2363 }
2364 icount = dx_get_count(entries);
2365 bh2 = ext4_append(handle, dir, &newblock);
2366 if (IS_ERR(bh2)) {
2367 err = PTR_ERR(bh2);
2368 goto cleanup;
2369 }
2370 node2 = (struct dx_node *)(bh2->b_data);
2371 entries2 = node2->entries;
2372 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2373 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2374 sb->s_blocksize);
2375 BUFFER_TRACE(frame->bh, "get_write_access");
2376 err = ext4_journal_get_write_access(handle, frame->bh);
2377 if (err)
2378 goto journal_error;
2379 if (!add_level) {
2380 unsigned icount1 = icount/2, icount2 = icount - icount1;
2381 unsigned hash2 = dx_get_hash(entries + icount1);
2382 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2383 icount1, icount2));
2384
2385 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2386 err = ext4_journal_get_write_access(handle,
2387 (frame - 1)->bh);
2388 if (err)
2389 goto journal_error;
2390
2391 memcpy((char *) entries2, (char *) (entries + icount1),
2392 icount2 * sizeof(struct dx_entry));
2393 dx_set_count(entries, icount1);
2394 dx_set_count(entries2, icount2);
2395 dx_set_limit(entries2, dx_node_limit(dir));
2396
2397 /* Which index block gets the new entry? */
2398 if (at - entries >= icount1) {
2399 frame->at = at = at - entries - icount1 + entries2;
2400 frame->entries = entries = entries2;
2401 swap(frame->bh, bh2);
2402 }
2403 dx_insert_block((frame - 1), hash2, newblock);
2404 dxtrace(dx_show_index("node", frame->entries));
2405 dxtrace(dx_show_index("node",
2406 ((struct dx_node *) bh2->b_data)->entries));
2407 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2408 if (err)
2409 goto journal_error;
2410 brelse (bh2);
2411 err = ext4_handle_dirty_dx_node(handle, dir,
2412 (frame - 1)->bh);
2413 if (err)
2414 goto journal_error;
2415 if (restart) {
2416 err = ext4_handle_dirty_dx_node(handle, dir,
2417 frame->bh);
2418 goto journal_error;
2419 }
2420 } else {
2421 struct dx_root *dxroot;
2422 memcpy((char *) entries2, (char *) entries,
2423 icount * sizeof(struct dx_entry));
2424 dx_set_limit(entries2, dx_node_limit(dir));
2425
2426 /* Set up root */
2427 dx_set_count(entries, 1);
2428 dx_set_block(entries + 0, newblock);
2429 dxroot = (struct dx_root *)frames[0].bh->b_data;
2430 dxroot->info.indirect_levels += 1;
2431 dxtrace(printk(KERN_DEBUG
2432 "Creating %d level index...\n",
2433 dxroot->info.indirect_levels));
2434 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2435 if (err)
2436 goto journal_error;
2437 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2438 brelse(bh2);
2439 restart = 1;
2440 goto journal_error;
2441 }
2442 }
2443 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2444 if (IS_ERR(de)) {
2445 err = PTR_ERR(de);
2446 goto cleanup;
2447 }
2448 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2449 goto cleanup;
2450
2451journal_error:
2452 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2453cleanup:
2454 brelse(bh);
2455 dx_release(frames);
2456 /* @restart is true means htree-path has been changed, we need to
2457 * repeat dx_probe() to find out valid htree-path
2458 */
2459 if (restart && err == 0)
2460 goto again;
2461 return err;
2462}
2463
2464/*
2465 * ext4_generic_delete_entry deletes a directory entry by merging it
2466 * with the previous entry
2467 */
2468int ext4_generic_delete_entry(struct inode *dir,
2469 struct ext4_dir_entry_2 *de_del,
2470 struct buffer_head *bh,
2471 void *entry_buf,
2472 int buf_size,
2473 int csum_size)
2474{
2475 struct ext4_dir_entry_2 *de, *pde;
2476 unsigned int blocksize = dir->i_sb->s_blocksize;
2477 int i;
2478
2479 i = 0;
2480 pde = NULL;
2481 de = (struct ext4_dir_entry_2 *)entry_buf;
2482 while (i < buf_size - csum_size) {
2483 if (ext4_check_dir_entry(dir, NULL, de, bh,
2484 entry_buf, buf_size, i))
2485 return -EFSCORRUPTED;
2486 if (de == de_del) {
2487 if (pde)
2488 pde->rec_len = ext4_rec_len_to_disk(
2489 ext4_rec_len_from_disk(pde->rec_len,
2490 blocksize) +
2491 ext4_rec_len_from_disk(de->rec_len,
2492 blocksize),
2493 blocksize);
2494 else
2495 de->inode = 0;
2496 inode_inc_iversion(dir);
2497 return 0;
2498 }
2499 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2500 pde = de;
2501 de = ext4_next_entry(de, blocksize);
2502 }
2503 return -ENOENT;
2504}
2505
2506static int ext4_delete_entry(handle_t *handle,
2507 struct inode *dir,
2508 struct ext4_dir_entry_2 *de_del,
2509 struct buffer_head *bh)
2510{
2511 int err, csum_size = 0;
2512
2513 if (ext4_has_inline_data(dir)) {
2514 int has_inline_data = 1;
2515 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2516 &has_inline_data);
2517 if (has_inline_data)
2518 return err;
2519 }
2520
2521 if (ext4_has_metadata_csum(dir->i_sb))
2522 csum_size = sizeof(struct ext4_dir_entry_tail);
2523
2524 BUFFER_TRACE(bh, "get_write_access");
2525 err = ext4_journal_get_write_access(handle, bh);
2526 if (unlikely(err))
2527 goto out;
2528
2529 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2530 dir->i_sb->s_blocksize, csum_size);
2531 if (err)
2532 goto out;
2533
2534 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2535 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2536 if (unlikely(err))
2537 goto out;
2538
2539 return 0;
2540out:
2541 if (err != -ENOENT)
2542 ext4_std_error(dir->i_sb, err);
2543 return err;
2544}
2545
2546/*
2547 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2548 * since this indicates that nlinks count was previously 1 to avoid overflowing
2549 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2550 * that subdirectory link counts are not being maintained accurately.
2551 *
2552 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2553 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2554 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2555 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2556 */
2557static void ext4_inc_count(handle_t *handle, struct inode *inode)
2558{
2559 inc_nlink(inode);
2560 if (is_dx(inode) &&
2561 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2562 set_nlink(inode, 1);
2563}
2564
2565/*
2566 * If a directory had nlink == 1, then we should let it be 1. This indicates
2567 * directory has >EXT4_LINK_MAX subdirs.
2568 */
2569static void ext4_dec_count(handle_t *handle, struct inode *inode)
2570{
2571 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2572 drop_nlink(inode);
2573}
2574
2575
2576/*
2577 * Add non-directory inode to a directory. On success, the inode reference is
2578 * consumed by dentry is instantiation. This is also indicated by clearing of
2579 * *inodep pointer. On failure, the caller is responsible for dropping the
2580 * inode reference in the safe context.
2581 */
2582static int ext4_add_nondir(handle_t *handle,
2583 struct dentry *dentry, struct inode **inodep)
2584{
2585 struct inode *dir = d_inode(dentry->d_parent);
2586 struct inode *inode = *inodep;
2587 int err = ext4_add_entry(handle, dentry, inode);
2588 if (!err) {
2589 err = ext4_mark_inode_dirty(handle, inode);
2590 if (IS_DIRSYNC(dir))
2591 ext4_handle_sync(handle);
2592 d_instantiate_new(dentry, inode);
2593 *inodep = NULL;
2594 return err;
2595 }
2596 drop_nlink(inode);
2597 ext4_orphan_add(handle, inode);
2598 unlock_new_inode(inode);
2599 return err;
2600}
2601
2602/*
2603 * By the time this is called, we already have created
2604 * the directory cache entry for the new file, but it
2605 * is so far negative - it has no inode.
2606 *
2607 * If the create succeeds, we fill in the inode information
2608 * with d_instantiate().
2609 */
2610static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2611 bool excl)
2612{
2613 handle_t *handle;
2614 struct inode *inode;
2615 int err, credits, retries = 0;
2616
2617 err = dquot_initialize(dir);
2618 if (err)
2619 return err;
2620
2621 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2622 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2623retry:
2624 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2625 NULL, EXT4_HT_DIR, credits);
2626 handle = ext4_journal_current_handle();
2627 err = PTR_ERR(inode);
2628 if (!IS_ERR(inode)) {
2629 inode->i_op = &ext4_file_inode_operations;
2630 inode->i_fop = &ext4_file_operations;
2631 ext4_set_aops(inode);
2632 err = ext4_add_nondir(handle, dentry, &inode);
2633 }
2634 if (handle)
2635 ext4_journal_stop(handle);
2636 if (!IS_ERR_OR_NULL(inode))
2637 iput(inode);
2638 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2639 goto retry;
2640 return err;
2641}
2642
2643static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2644 umode_t mode, dev_t rdev)
2645{
2646 handle_t *handle;
2647 struct inode *inode;
2648 int err, credits, retries = 0;
2649
2650 err = dquot_initialize(dir);
2651 if (err)
2652 return err;
2653
2654 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2655 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2656retry:
2657 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2658 NULL, EXT4_HT_DIR, credits);
2659 handle = ext4_journal_current_handle();
2660 err = PTR_ERR(inode);
2661 if (!IS_ERR(inode)) {
2662 init_special_inode(inode, inode->i_mode, rdev);
2663 inode->i_op = &ext4_special_inode_operations;
2664 err = ext4_add_nondir(handle, dentry, &inode);
2665 }
2666 if (handle)
2667 ext4_journal_stop(handle);
2668 if (!IS_ERR_OR_NULL(inode))
2669 iput(inode);
2670 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2671 goto retry;
2672 return err;
2673}
2674
2675static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2676{
2677 handle_t *handle;
2678 struct inode *inode;
2679 int err, retries = 0;
2680
2681 err = dquot_initialize(dir);
2682 if (err)
2683 return err;
2684
2685retry:
2686 inode = ext4_new_inode_start_handle(dir, mode,
2687 NULL, 0, NULL,
2688 EXT4_HT_DIR,
2689 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2690 4 + EXT4_XATTR_TRANS_BLOCKS);
2691 handle = ext4_journal_current_handle();
2692 err = PTR_ERR(inode);
2693 if (!IS_ERR(inode)) {
2694 inode->i_op = &ext4_file_inode_operations;
2695 inode->i_fop = &ext4_file_operations;
2696 ext4_set_aops(inode);
2697 d_tmpfile(dentry, inode);
2698 err = ext4_orphan_add(handle, inode);
2699 if (err)
2700 goto err_unlock_inode;
2701 mark_inode_dirty(inode);
2702 unlock_new_inode(inode);
2703 }
2704 if (handle)
2705 ext4_journal_stop(handle);
2706 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2707 goto retry;
2708 return err;
2709err_unlock_inode:
2710 ext4_journal_stop(handle);
2711 unlock_new_inode(inode);
2712 return err;
2713}
2714
2715struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2716 struct ext4_dir_entry_2 *de,
2717 int blocksize, int csum_size,
2718 unsigned int parent_ino, int dotdot_real_len)
2719{
2720 de->inode = cpu_to_le32(inode->i_ino);
2721 de->name_len = 1;
2722 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2723 blocksize);
2724 strcpy(de->name, ".");
2725 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2726
2727 de = ext4_next_entry(de, blocksize);
2728 de->inode = cpu_to_le32(parent_ino);
2729 de->name_len = 2;
2730 if (!dotdot_real_len)
2731 de->rec_len = ext4_rec_len_to_disk(blocksize -
2732 (csum_size + EXT4_DIR_REC_LEN(1)),
2733 blocksize);
2734 else
2735 de->rec_len = ext4_rec_len_to_disk(
2736 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2737 strcpy(de->name, "..");
2738 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2739
2740 return ext4_next_entry(de, blocksize);
2741}
2742
2743static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2744 struct inode *inode)
2745{
2746 struct buffer_head *dir_block = NULL;
2747 struct ext4_dir_entry_2 *de;
2748 ext4_lblk_t block = 0;
2749 unsigned int blocksize = dir->i_sb->s_blocksize;
2750 int csum_size = 0;
2751 int err;
2752
2753 if (ext4_has_metadata_csum(dir->i_sb))
2754 csum_size = sizeof(struct ext4_dir_entry_tail);
2755
2756 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2757 err = ext4_try_create_inline_dir(handle, dir, inode);
2758 if (err < 0 && err != -ENOSPC)
2759 goto out;
2760 if (!err)
2761 goto out;
2762 }
2763
2764 inode->i_size = 0;
2765 dir_block = ext4_append(handle, inode, &block);
2766 if (IS_ERR(dir_block))
2767 return PTR_ERR(dir_block);
2768 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2769 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2770 set_nlink(inode, 2);
2771 if (csum_size)
2772 ext4_initialize_dirent_tail(dir_block, blocksize);
2773
2774 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2775 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2776 if (err)
2777 goto out;
2778 set_buffer_verified(dir_block);
2779out:
2780 brelse(dir_block);
2781 return err;
2782}
2783
2784static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2785{
2786 handle_t *handle;
2787 struct inode *inode;
2788 int err, err2 = 0, credits, retries = 0;
2789
2790 if (EXT4_DIR_LINK_MAX(dir))
2791 return -EMLINK;
2792
2793 err = dquot_initialize(dir);
2794 if (err)
2795 return err;
2796
2797 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2798 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2799retry:
2800 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2801 &dentry->d_name,
2802 0, NULL, EXT4_HT_DIR, credits);
2803 handle = ext4_journal_current_handle();
2804 err = PTR_ERR(inode);
2805 if (IS_ERR(inode))
2806 goto out_stop;
2807
2808 inode->i_op = &ext4_dir_inode_operations;
2809 inode->i_fop = &ext4_dir_operations;
2810 err = ext4_init_new_dir(handle, dir, inode);
2811 if (err)
2812 goto out_clear_inode;
2813 err = ext4_mark_inode_dirty(handle, inode);
2814 if (!err)
2815 err = ext4_add_entry(handle, dentry, inode);
2816 if (err) {
2817out_clear_inode:
2818 clear_nlink(inode);
2819 ext4_orphan_add(handle, inode);
2820 unlock_new_inode(inode);
2821 err2 = ext4_mark_inode_dirty(handle, inode);
2822 if (unlikely(err2))
2823 err = err2;
2824 ext4_journal_stop(handle);
2825 iput(inode);
2826 goto out_retry;
2827 }
2828 ext4_inc_count(handle, dir);
2829 ext4_update_dx_flag(dir);
2830 err = ext4_mark_inode_dirty(handle, dir);
2831 if (err)
2832 goto out_clear_inode;
2833 d_instantiate_new(dentry, inode);
2834 if (IS_DIRSYNC(dir))
2835 ext4_handle_sync(handle);
2836
2837out_stop:
2838 if (handle)
2839 ext4_journal_stop(handle);
2840out_retry:
2841 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2842 goto retry;
2843 return err;
2844}
2845
2846/*
2847 * routine to check that the specified directory is empty (for rmdir)
2848 */
2849bool ext4_empty_dir(struct inode *inode)
2850{
2851 unsigned int offset;
2852 struct buffer_head *bh;
2853 struct ext4_dir_entry_2 *de;
2854 struct super_block *sb;
2855
2856 if (ext4_has_inline_data(inode)) {
2857 int has_inline_data = 1;
2858 int ret;
2859
2860 ret = empty_inline_dir(inode, &has_inline_data);
2861 if (has_inline_data)
2862 return ret;
2863 }
2864
2865 sb = inode->i_sb;
2866 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2867 EXT4_ERROR_INODE(inode, "invalid size");
2868 return true;
2869 }
2870 /* The first directory block must not be a hole,
2871 * so treat it as DIRENT_HTREE
2872 */
2873 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2874 if (IS_ERR(bh))
2875 return true;
2876
2877 de = (struct ext4_dir_entry_2 *) bh->b_data;
2878 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2879 0) ||
2880 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2881 ext4_warning_inode(inode, "directory missing '.'");
2882 brelse(bh);
2883 return true;
2884 }
2885 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2886 de = ext4_next_entry(de, sb->s_blocksize);
2887 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2888 offset) ||
2889 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2890 ext4_warning_inode(inode, "directory missing '..'");
2891 brelse(bh);
2892 return true;
2893 }
2894 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2895 while (offset < inode->i_size) {
2896 if (!(offset & (sb->s_blocksize - 1))) {
2897 unsigned int lblock;
2898 brelse(bh);
2899 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2900 bh = ext4_read_dirblock(inode, lblock, EITHER);
2901 if (bh == NULL) {
2902 offset += sb->s_blocksize;
2903 continue;
2904 }
2905 if (IS_ERR(bh))
2906 return true;
2907 }
2908 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2909 (offset & (sb->s_blocksize - 1)));
2910 if (ext4_check_dir_entry(inode, NULL, de, bh,
2911 bh->b_data, bh->b_size, offset)) {
2912 offset = (offset | (sb->s_blocksize - 1)) + 1;
2913 continue;
2914 }
2915 if (le32_to_cpu(de->inode)) {
2916 brelse(bh);
2917 return false;
2918 }
2919 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2920 }
2921 brelse(bh);
2922 return true;
2923}
2924
2925/*
2926 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2927 * such inodes, starting at the superblock, in case we crash before the
2928 * file is closed/deleted, or in case the inode truncate spans multiple
2929 * transactions and the last transaction is not recovered after a crash.
2930 *
2931 * At filesystem recovery time, we walk this list deleting unlinked
2932 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2933 *
2934 * Orphan list manipulation functions must be called under i_mutex unless
2935 * we are just creating the inode or deleting it.
2936 */
2937int ext4_orphan_add(handle_t *handle, struct inode *inode)
2938{
2939 struct super_block *sb = inode->i_sb;
2940 struct ext4_sb_info *sbi = EXT4_SB(sb);
2941 struct ext4_iloc iloc;
2942 int err = 0, rc;
2943 bool dirty = false;
2944
2945 if (!sbi->s_journal || is_bad_inode(inode))
2946 return 0;
2947
2948 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2949 !inode_is_locked(inode));
2950 /*
2951 * Exit early if inode already is on orphan list. This is a big speedup
2952 * since we don't have to contend on the global s_orphan_lock.
2953 */
2954 if (!list_empty(&EXT4_I(inode)->i_orphan))
2955 return 0;
2956
2957 /*
2958 * Orphan handling is only valid for files with data blocks
2959 * being truncated, or files being unlinked. Note that we either
2960 * hold i_mutex, or the inode can not be referenced from outside,
2961 * so i_nlink should not be bumped due to race
2962 */
2963 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2964 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2965
2966 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2967 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2968 if (err)
2969 goto out;
2970
2971 err = ext4_reserve_inode_write(handle, inode, &iloc);
2972 if (err)
2973 goto out;
2974
2975 mutex_lock(&sbi->s_orphan_lock);
2976 /*
2977 * Due to previous errors inode may be already a part of on-disk
2978 * orphan list. If so skip on-disk list modification.
2979 */
2980 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2981 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2982 /* Insert this inode at the head of the on-disk orphan list */
2983 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2984 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2985 dirty = true;
2986 }
2987 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2988 mutex_unlock(&sbi->s_orphan_lock);
2989
2990 if (dirty) {
2991 err = ext4_handle_dirty_super(handle, sb);
2992 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2993 if (!err)
2994 err = rc;
2995 if (err) {
2996 /*
2997 * We have to remove inode from in-memory list if
2998 * addition to on disk orphan list failed. Stray orphan
2999 * list entries can cause panics at unmount time.
3000 */
3001 mutex_lock(&sbi->s_orphan_lock);
3002 list_del_init(&EXT4_I(inode)->i_orphan);
3003 mutex_unlock(&sbi->s_orphan_lock);
3004 }
3005 } else
3006 brelse(iloc.bh);
3007
3008 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3009 jbd_debug(4, "orphan inode %lu will point to %d\n",
3010 inode->i_ino, NEXT_ORPHAN(inode));
3011out:
3012 ext4_std_error(sb, err);
3013 return err;
3014}
3015
3016/*
3017 * ext4_orphan_del() removes an unlinked or truncated inode from the list
3018 * of such inodes stored on disk, because it is finally being cleaned up.
3019 */
3020int ext4_orphan_del(handle_t *handle, struct inode *inode)
3021{
3022 struct list_head *prev;
3023 struct ext4_inode_info *ei = EXT4_I(inode);
3024 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3025 __u32 ino_next;
3026 struct ext4_iloc iloc;
3027 int err = 0;
3028
3029 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3030 return 0;
3031
3032 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3033 !inode_is_locked(inode));
3034 /* Do this quick check before taking global s_orphan_lock. */
3035 if (list_empty(&ei->i_orphan))
3036 return 0;
3037
3038 if (handle) {
3039 /* Grab inode buffer early before taking global s_orphan_lock */
3040 err = ext4_reserve_inode_write(handle, inode, &iloc);
3041 }
3042
3043 mutex_lock(&sbi->s_orphan_lock);
3044 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3045
3046 prev = ei->i_orphan.prev;
3047 list_del_init(&ei->i_orphan);
3048
3049 /* If we're on an error path, we may not have a valid
3050 * transaction handle with which to update the orphan list on
3051 * disk, but we still need to remove the inode from the linked
3052 * list in memory. */
3053 if (!handle || err) {
3054 mutex_unlock(&sbi->s_orphan_lock);
3055 goto out_err;
3056 }
3057
3058 ino_next = NEXT_ORPHAN(inode);
3059 if (prev == &sbi->s_orphan) {
3060 jbd_debug(4, "superblock will point to %u\n", ino_next);
3061 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3062 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3063 if (err) {
3064 mutex_unlock(&sbi->s_orphan_lock);
3065 goto out_brelse;
3066 }
3067 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3068 mutex_unlock(&sbi->s_orphan_lock);
3069 err = ext4_handle_dirty_super(handle, inode->i_sb);
3070 } else {
3071 struct ext4_iloc iloc2;
3072 struct inode *i_prev =
3073 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3074
3075 jbd_debug(4, "orphan inode %lu will point to %u\n",
3076 i_prev->i_ino, ino_next);
3077 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3078 if (err) {
3079 mutex_unlock(&sbi->s_orphan_lock);
3080 goto out_brelse;
3081 }
3082 NEXT_ORPHAN(i_prev) = ino_next;
3083 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3084 mutex_unlock(&sbi->s_orphan_lock);
3085 }
3086 if (err)
3087 goto out_brelse;
3088 NEXT_ORPHAN(inode) = 0;
3089 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3090out_err:
3091 ext4_std_error(inode->i_sb, err);
3092 return err;
3093
3094out_brelse:
3095 brelse(iloc.bh);
3096 goto out_err;
3097}
3098
3099static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3100{
3101 int retval;
3102 struct inode *inode;
3103 struct buffer_head *bh;
3104 struct ext4_dir_entry_2 *de;
3105 handle_t *handle = NULL;
3106
3107 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3108 return -EIO;
3109
3110 /* Initialize quotas before so that eventual writes go in
3111 * separate transaction */
3112 retval = dquot_initialize(dir);
3113 if (retval)
3114 return retval;
3115 retval = dquot_initialize(d_inode(dentry));
3116 if (retval)
3117 return retval;
3118
3119 retval = -ENOENT;
3120 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3121 if (IS_ERR(bh))
3122 return PTR_ERR(bh);
3123 if (!bh)
3124 goto end_rmdir;
3125
3126 inode = d_inode(dentry);
3127
3128 retval = -EFSCORRUPTED;
3129 if (le32_to_cpu(de->inode) != inode->i_ino)
3130 goto end_rmdir;
3131
3132 retval = -ENOTEMPTY;
3133 if (!ext4_empty_dir(inode))
3134 goto end_rmdir;
3135
3136 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3137 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3138 if (IS_ERR(handle)) {
3139 retval = PTR_ERR(handle);
3140 handle = NULL;
3141 goto end_rmdir;
3142 }
3143
3144 if (IS_DIRSYNC(dir))
3145 ext4_handle_sync(handle);
3146
3147 retval = ext4_delete_entry(handle, dir, de, bh);
3148 if (retval)
3149 goto end_rmdir;
3150 if (!EXT4_DIR_LINK_EMPTY(inode))
3151 ext4_warning_inode(inode,
3152 "empty directory '%.*s' has too many links (%u)",
3153 dentry->d_name.len, dentry->d_name.name,
3154 inode->i_nlink);
3155 inode_inc_iversion(inode);
3156 clear_nlink(inode);
3157 /* There's no need to set i_disksize: the fact that i_nlink is
3158 * zero will ensure that the right thing happens during any
3159 * recovery. */
3160 inode->i_size = 0;
3161 ext4_orphan_add(handle, inode);
3162 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3163 retval = ext4_mark_inode_dirty(handle, inode);
3164 if (retval)
3165 goto end_rmdir;
3166 ext4_dec_count(handle, dir);
3167 ext4_update_dx_flag(dir);
3168 retval = ext4_mark_inode_dirty(handle, dir);
3169
3170#ifdef CONFIG_UNICODE
3171 /* VFS negative dentries are incompatible with Encoding and
3172 * Case-insensitiveness. Eventually we'll want avoid
3173 * invalidating the dentries here, alongside with returning the
3174 * negative dentries at ext4_lookup(), when it is better
3175 * supported by the VFS for the CI case.
3176 */
3177 if (IS_CASEFOLDED(dir))
3178 d_invalidate(dentry);
3179#endif
3180
3181end_rmdir:
3182 brelse(bh);
3183 if (handle)
3184 ext4_journal_stop(handle);
3185 return retval;
3186}
3187
3188static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3189{
3190 int retval;
3191 struct inode *inode;
3192 struct buffer_head *bh;
3193 struct ext4_dir_entry_2 *de;
3194 handle_t *handle = NULL;
3195
3196 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3197 return -EIO;
3198
3199 trace_ext4_unlink_enter(dir, dentry);
3200 /* Initialize quotas before so that eventual writes go
3201 * in separate transaction */
3202 retval = dquot_initialize(dir);
3203 if (retval)
3204 goto out_trace;
3205 retval = dquot_initialize(d_inode(dentry));
3206 if (retval)
3207 goto out_trace;
3208
3209 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3210 if (IS_ERR(bh)) {
3211 retval = PTR_ERR(bh);
3212 goto out_trace;
3213 }
3214 if (!bh) {
3215 retval = -ENOENT;
3216 goto out_trace;
3217 }
3218
3219 inode = d_inode(dentry);
3220
3221 if (le32_to_cpu(de->inode) != inode->i_ino) {
3222 retval = -EFSCORRUPTED;
3223 goto out_bh;
3224 }
3225
3226 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3227 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3228 if (IS_ERR(handle)) {
3229 retval = PTR_ERR(handle);
3230 goto out_bh;
3231 }
3232
3233 if (IS_DIRSYNC(dir))
3234 ext4_handle_sync(handle);
3235
3236 retval = ext4_delete_entry(handle, dir, de, bh);
3237 if (retval)
3238 goto out_handle;
3239 dir->i_ctime = dir->i_mtime = current_time(dir);
3240 ext4_update_dx_flag(dir);
3241 retval = ext4_mark_inode_dirty(handle, dir);
3242 if (retval)
3243 goto out_handle;
3244 if (inode->i_nlink == 0)
3245 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3246 dentry->d_name.len, dentry->d_name.name);
3247 else
3248 drop_nlink(inode);
3249 if (!inode->i_nlink)
3250 ext4_orphan_add(handle, inode);
3251 inode->i_ctime = current_time(inode);
3252 retval = ext4_mark_inode_dirty(handle, inode);
3253
3254#ifdef CONFIG_UNICODE
3255 /* VFS negative dentries are incompatible with Encoding and
3256 * Case-insensitiveness. Eventually we'll want avoid
3257 * invalidating the dentries here, alongside with returning the
3258 * negative dentries at ext4_lookup(), when it is better
3259 * supported by the VFS for the CI case.
3260 */
3261 if (IS_CASEFOLDED(dir))
3262 d_invalidate(dentry);
3263#endif
3264
3265out_handle:
3266 ext4_journal_stop(handle);
3267out_bh:
3268 brelse(bh);
3269out_trace:
3270 trace_ext4_unlink_exit(dentry, retval);
3271 return retval;
3272}
3273
3274static int ext4_symlink(struct inode *dir,
3275 struct dentry *dentry, const char *symname)
3276{
3277 handle_t *handle;
3278 struct inode *inode;
3279 int err, len = strlen(symname);
3280 int credits;
3281 struct fscrypt_str disk_link;
3282
3283 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3284 return -EIO;
3285
3286 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3287 &disk_link);
3288 if (err)
3289 return err;
3290
3291 err = dquot_initialize(dir);
3292 if (err)
3293 return err;
3294
3295 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3296 /*
3297 * For non-fast symlinks, we just allocate inode and put it on
3298 * orphan list in the first transaction => we need bitmap,
3299 * group descriptor, sb, inode block, quota blocks, and
3300 * possibly selinux xattr blocks.
3301 */
3302 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3303 EXT4_XATTR_TRANS_BLOCKS;
3304 } else {
3305 /*
3306 * Fast symlink. We have to add entry to directory
3307 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3308 * allocate new inode (bitmap, group descriptor, inode block,
3309 * quota blocks, sb is already counted in previous macros).
3310 */
3311 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3312 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3313 }
3314
3315 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3316 &dentry->d_name, 0, NULL,
3317 EXT4_HT_DIR, credits);
3318 handle = ext4_journal_current_handle();
3319 if (IS_ERR(inode)) {
3320 if (handle)
3321 ext4_journal_stop(handle);
3322 return PTR_ERR(inode);
3323 }
3324
3325 if (IS_ENCRYPTED(inode)) {
3326 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3327 if (err)
3328 goto err_drop_inode;
3329 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3330 }
3331
3332 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3333 if (!IS_ENCRYPTED(inode))
3334 inode->i_op = &ext4_symlink_inode_operations;
3335 inode_nohighmem(inode);
3336 ext4_set_aops(inode);
3337 /*
3338 * We cannot call page_symlink() with transaction started
3339 * because it calls into ext4_write_begin() which can wait
3340 * for transaction commit if we are running out of space
3341 * and thus we deadlock. So we have to stop transaction now
3342 * and restart it when symlink contents is written.
3343 *
3344 * To keep fs consistent in case of crash, we have to put inode
3345 * to orphan list in the mean time.
3346 */
3347 drop_nlink(inode);
3348 err = ext4_orphan_add(handle, inode);
3349 ext4_journal_stop(handle);
3350 handle = NULL;
3351 if (err)
3352 goto err_drop_inode;
3353 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3354 if (err)
3355 goto err_drop_inode;
3356 /*
3357 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3358 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3359 */
3360 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3361 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3362 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3363 if (IS_ERR(handle)) {
3364 err = PTR_ERR(handle);
3365 handle = NULL;
3366 goto err_drop_inode;
3367 }
3368 set_nlink(inode, 1);
3369 err = ext4_orphan_del(handle, inode);
3370 if (err)
3371 goto err_drop_inode;
3372 } else {
3373 /* clear the extent format for fast symlink */
3374 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3375 if (!IS_ENCRYPTED(inode)) {
3376 inode->i_op = &ext4_fast_symlink_inode_operations;
3377 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3378 }
3379 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3380 disk_link.len);
3381 inode->i_size = disk_link.len - 1;
3382 }
3383 EXT4_I(inode)->i_disksize = inode->i_size;
3384 err = ext4_add_nondir(handle, dentry, &inode);
3385 if (handle)
3386 ext4_journal_stop(handle);
3387 if (inode)
3388 iput(inode);
3389 goto out_free_encrypted_link;
3390
3391err_drop_inode:
3392 if (handle)
3393 ext4_journal_stop(handle);
3394 clear_nlink(inode);
3395 unlock_new_inode(inode);
3396 iput(inode);
3397out_free_encrypted_link:
3398 if (disk_link.name != (unsigned char *)symname)
3399 kfree(disk_link.name);
3400 return err;
3401}
3402
3403static int ext4_link(struct dentry *old_dentry,
3404 struct inode *dir, struct dentry *dentry)
3405{
3406 handle_t *handle;
3407 struct inode *inode = d_inode(old_dentry);
3408 int err, retries = 0;
3409
3410 if (inode->i_nlink >= EXT4_LINK_MAX)
3411 return -EMLINK;
3412
3413 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3414 if (err)
3415 return err;
3416
3417 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3418 (!projid_eq(EXT4_I(dir)->i_projid,
3419 EXT4_I(old_dentry->d_inode)->i_projid)))
3420 return -EXDEV;
3421
3422 err = dquot_initialize(dir);
3423 if (err)
3424 return err;
3425
3426retry:
3427 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3428 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3429 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3430 if (IS_ERR(handle))
3431 return PTR_ERR(handle);
3432
3433 if (IS_DIRSYNC(dir))
3434 ext4_handle_sync(handle);
3435
3436 inode->i_ctime = current_time(inode);
3437 ext4_inc_count(handle, inode);
3438 ihold(inode);
3439
3440 err = ext4_add_entry(handle, dentry, inode);
3441 if (!err) {
3442 err = ext4_mark_inode_dirty(handle, inode);
3443 /* this can happen only for tmpfile being
3444 * linked the first time
3445 */
3446 if (inode->i_nlink == 1)
3447 ext4_orphan_del(handle, inode);
3448 d_instantiate(dentry, inode);
3449 } else {
3450 drop_nlink(inode);
3451 iput(inode);
3452 }
3453 ext4_journal_stop(handle);
3454 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3455 goto retry;
3456 return err;
3457}
3458
3459
3460/*
3461 * Try to find buffer head where contains the parent block.
3462 * It should be the inode block if it is inlined or the 1st block
3463 * if it is a normal dir.
3464 */
3465static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3466 struct inode *inode,
3467 int *retval,
3468 struct ext4_dir_entry_2 **parent_de,
3469 int *inlined)
3470{
3471 struct buffer_head *bh;
3472
3473 if (!ext4_has_inline_data(inode)) {
3474 /* The first directory block must not be a hole, so
3475 * treat it as DIRENT_HTREE
3476 */
3477 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3478 if (IS_ERR(bh)) {
3479 *retval = PTR_ERR(bh);
3480 return NULL;
3481 }
3482 *parent_de = ext4_next_entry(
3483 (struct ext4_dir_entry_2 *)bh->b_data,
3484 inode->i_sb->s_blocksize);
3485 return bh;
3486 }
3487
3488 *inlined = 1;
3489 return ext4_get_first_inline_block(inode, parent_de, retval);
3490}
3491
3492struct ext4_renament {
3493 struct inode *dir;
3494 struct dentry *dentry;
3495 struct inode *inode;
3496 bool is_dir;
3497 int dir_nlink_delta;
3498
3499 /* entry for "dentry" */
3500 struct buffer_head *bh;
3501 struct ext4_dir_entry_2 *de;
3502 int inlined;
3503
3504 /* entry for ".." in inode if it's a directory */
3505 struct buffer_head *dir_bh;
3506 struct ext4_dir_entry_2 *parent_de;
3507 int dir_inlined;
3508};
3509
3510static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3511{
3512 int retval;
3513
3514 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3515 &retval, &ent->parent_de,
3516 &ent->dir_inlined);
3517 if (!ent->dir_bh)
3518 return retval;
3519 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3520 return -EFSCORRUPTED;
3521 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3522 return ext4_journal_get_write_access(handle, ent->dir_bh);
3523}
3524
3525static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3526 unsigned dir_ino)
3527{
3528 int retval;
3529
3530 ent->parent_de->inode = cpu_to_le32(dir_ino);
3531 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3532 if (!ent->dir_inlined) {
3533 if (is_dx(ent->inode)) {
3534 retval = ext4_handle_dirty_dx_node(handle,
3535 ent->inode,
3536 ent->dir_bh);
3537 } else {
3538 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3539 ent->dir_bh);
3540 }
3541 } else {
3542 retval = ext4_mark_inode_dirty(handle, ent->inode);
3543 }
3544 if (retval) {
3545 ext4_std_error(ent->dir->i_sb, retval);
3546 return retval;
3547 }
3548 return 0;
3549}
3550
3551static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3552 unsigned ino, unsigned file_type)
3553{
3554 int retval, retval2;
3555
3556 BUFFER_TRACE(ent->bh, "get write access");
3557 retval = ext4_journal_get_write_access(handle, ent->bh);
3558 if (retval)
3559 return retval;
3560 ent->de->inode = cpu_to_le32(ino);
3561 if (ext4_has_feature_filetype(ent->dir->i_sb))
3562 ent->de->file_type = file_type;
3563 inode_inc_iversion(ent->dir);
3564 ent->dir->i_ctime = ent->dir->i_mtime =
3565 current_time(ent->dir);
3566 retval = ext4_mark_inode_dirty(handle, ent->dir);
3567 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3568 if (!ent->inlined) {
3569 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3570 if (unlikely(retval2)) {
3571 ext4_std_error(ent->dir->i_sb, retval2);
3572 return retval2;
3573 }
3574 }
3575 brelse(ent->bh);
3576 ent->bh = NULL;
3577
3578 return retval;
3579}
3580
3581static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3582 const struct qstr *d_name)
3583{
3584 int retval = -ENOENT;
3585 struct buffer_head *bh;
3586 struct ext4_dir_entry_2 *de;
3587
3588 bh = ext4_find_entry(dir, d_name, &de, NULL);
3589 if (IS_ERR(bh))
3590 return PTR_ERR(bh);
3591 if (bh) {
3592 retval = ext4_delete_entry(handle, dir, de, bh);
3593 brelse(bh);
3594 }
3595 return retval;
3596}
3597
3598static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3599 int force_reread)
3600{
3601 int retval;
3602 /*
3603 * ent->de could have moved from under us during htree split, so make
3604 * sure that we are deleting the right entry. We might also be pointing
3605 * to a stale entry in the unused part of ent->bh so just checking inum
3606 * and the name isn't enough.
3607 */
3608 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3609 ent->de->name_len != ent->dentry->d_name.len ||
3610 strncmp(ent->de->name, ent->dentry->d_name.name,
3611 ent->de->name_len) ||
3612 force_reread) {
3613 retval = ext4_find_delete_entry(handle, ent->dir,
3614 &ent->dentry->d_name);
3615 } else {
3616 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3617 if (retval == -ENOENT) {
3618 retval = ext4_find_delete_entry(handle, ent->dir,
3619 &ent->dentry->d_name);
3620 }
3621 }
3622
3623 if (retval) {
3624 ext4_warning_inode(ent->dir,
3625 "Deleting old file: nlink %d, error=%d",
3626 ent->dir->i_nlink, retval);
3627 }
3628}
3629
3630static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3631{
3632 if (ent->dir_nlink_delta) {
3633 if (ent->dir_nlink_delta == -1)
3634 ext4_dec_count(handle, ent->dir);
3635 else
3636 ext4_inc_count(handle, ent->dir);
3637 ext4_mark_inode_dirty(handle, ent->dir);
3638 }
3639}
3640
3641static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3642 int credits, handle_t **h)
3643{
3644 struct inode *wh;
3645 handle_t *handle;
3646 int retries = 0;
3647
3648 /*
3649 * for inode block, sb block, group summaries,
3650 * and inode bitmap
3651 */
3652 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3653 EXT4_XATTR_TRANS_BLOCKS + 4);
3654retry:
3655 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3656 &ent->dentry->d_name, 0, NULL,
3657 EXT4_HT_DIR, credits);
3658
3659 handle = ext4_journal_current_handle();
3660 if (IS_ERR(wh)) {
3661 if (handle)
3662 ext4_journal_stop(handle);
3663 if (PTR_ERR(wh) == -ENOSPC &&
3664 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3665 goto retry;
3666 } else {
3667 *h = handle;
3668 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3669 wh->i_op = &ext4_special_inode_operations;
3670 }
3671 return wh;
3672}
3673
3674/*
3675 * Anybody can rename anything with this: the permission checks are left to the
3676 * higher-level routines.
3677 *
3678 * n.b. old_{dentry,inode) refers to the source dentry/inode
3679 * while new_{dentry,inode) refers to the destination dentry/inode
3680 * This comes from rename(const char *oldpath, const char *newpath)
3681 */
3682static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3683 struct inode *new_dir, struct dentry *new_dentry,
3684 unsigned int flags)
3685{
3686 handle_t *handle = NULL;
3687 struct ext4_renament old = {
3688 .dir = old_dir,
3689 .dentry = old_dentry,
3690 .inode = d_inode(old_dentry),
3691 };
3692 struct ext4_renament new = {
3693 .dir = new_dir,
3694 .dentry = new_dentry,
3695 .inode = d_inode(new_dentry),
3696 };
3697 int force_reread;
3698 int retval;
3699 struct inode *whiteout = NULL;
3700 int credits;
3701 u8 old_file_type;
3702
3703 if (new.inode && new.inode->i_nlink == 0) {
3704 EXT4_ERROR_INODE(new.inode,
3705 "target of rename is already freed");
3706 return -EFSCORRUPTED;
3707 }
3708
3709 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3710 (!projid_eq(EXT4_I(new_dir)->i_projid,
3711 EXT4_I(old_dentry->d_inode)->i_projid)))
3712 return -EXDEV;
3713
3714 retval = dquot_initialize(old.dir);
3715 if (retval)
3716 return retval;
3717 retval = dquot_initialize(new.dir);
3718 if (retval)
3719 return retval;
3720
3721 /* Initialize quotas before so that eventual writes go
3722 * in separate transaction */
3723 if (new.inode) {
3724 retval = dquot_initialize(new.inode);
3725 if (retval)
3726 return retval;
3727 }
3728
3729 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3730 if (IS_ERR(old.bh))
3731 return PTR_ERR(old.bh);
3732 /*
3733 * Check for inode number is _not_ due to possible IO errors.
3734 * We might rmdir the source, keep it as pwd of some process
3735 * and merrily kill the link to whatever was created under the
3736 * same name. Goodbye sticky bit ;-<
3737 */
3738 retval = -ENOENT;
3739 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3740 goto end_rename;
3741
3742 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3743 &new.de, &new.inlined);
3744 if (IS_ERR(new.bh)) {
3745 retval = PTR_ERR(new.bh);
3746 new.bh = NULL;
3747 goto end_rename;
3748 }
3749 if (new.bh) {
3750 if (!new.inode) {
3751 brelse(new.bh);
3752 new.bh = NULL;
3753 }
3754 }
3755 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3756 ext4_alloc_da_blocks(old.inode);
3757
3758 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3759 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3760 if (!(flags & RENAME_WHITEOUT)) {
3761 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3762 if (IS_ERR(handle)) {
3763 retval = PTR_ERR(handle);
3764 handle = NULL;
3765 goto end_rename;
3766 }
3767 } else {
3768 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3769 if (IS_ERR(whiteout)) {
3770 retval = PTR_ERR(whiteout);
3771 whiteout = NULL;
3772 goto end_rename;
3773 }
3774 }
3775
3776 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3777 ext4_handle_sync(handle);
3778
3779 if (S_ISDIR(old.inode->i_mode)) {
3780 if (new.inode) {
3781 retval = -ENOTEMPTY;
3782 if (!ext4_empty_dir(new.inode))
3783 goto end_rename;
3784 } else {
3785 retval = -EMLINK;
3786 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3787 goto end_rename;
3788 }
3789 retval = ext4_rename_dir_prepare(handle, &old);
3790 if (retval)
3791 goto end_rename;
3792 }
3793 /*
3794 * If we're renaming a file within an inline_data dir and adding or
3795 * setting the new dirent causes a conversion from inline_data to
3796 * extents/blockmap, we need to force the dirent delete code to
3797 * re-read the directory, or else we end up trying to delete a dirent
3798 * from what is now the extent tree root (or a block map).
3799 */
3800 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3801 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3802
3803 old_file_type = old.de->file_type;
3804 if (whiteout) {
3805 /*
3806 * Do this before adding a new entry, so the old entry is sure
3807 * to be still pointing to the valid old entry.
3808 */
3809 retval = ext4_setent(handle, &old, whiteout->i_ino,
3810 EXT4_FT_CHRDEV);
3811 if (retval)
3812 goto end_rename;
3813 retval = ext4_mark_inode_dirty(handle, whiteout);
3814 if (unlikely(retval))
3815 goto end_rename;
3816 }
3817 if (!new.bh) {
3818 retval = ext4_add_entry(handle, new.dentry, old.inode);
3819 if (retval)
3820 goto end_rename;
3821 } else {
3822 retval = ext4_setent(handle, &new,
3823 old.inode->i_ino, old_file_type);
3824 if (retval)
3825 goto end_rename;
3826 }
3827 if (force_reread)
3828 force_reread = !ext4_test_inode_flag(new.dir,
3829 EXT4_INODE_INLINE_DATA);
3830
3831 /*
3832 * Like most other Unix systems, set the ctime for inodes on a
3833 * rename.
3834 */
3835 old.inode->i_ctime = current_time(old.inode);
3836 retval = ext4_mark_inode_dirty(handle, old.inode);
3837 if (unlikely(retval))
3838 goto end_rename;
3839
3840 if (!whiteout) {
3841 /*
3842 * ok, that's it
3843 */
3844 ext4_rename_delete(handle, &old, force_reread);
3845 }
3846
3847 if (new.inode) {
3848 ext4_dec_count(handle, new.inode);
3849 new.inode->i_ctime = current_time(new.inode);
3850 }
3851 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3852 ext4_update_dx_flag(old.dir);
3853 if (old.dir_bh) {
3854 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3855 if (retval)
3856 goto end_rename;
3857
3858 ext4_dec_count(handle, old.dir);
3859 if (new.inode) {
3860 /* checked ext4_empty_dir above, can't have another
3861 * parent, ext4_dec_count() won't work for many-linked
3862 * dirs */
3863 clear_nlink(new.inode);
3864 } else {
3865 ext4_inc_count(handle, new.dir);
3866 ext4_update_dx_flag(new.dir);
3867 retval = ext4_mark_inode_dirty(handle, new.dir);
3868 if (unlikely(retval))
3869 goto end_rename;
3870 }
3871 }
3872 retval = ext4_mark_inode_dirty(handle, old.dir);
3873 if (unlikely(retval))
3874 goto end_rename;
3875 if (new.inode) {
3876 retval = ext4_mark_inode_dirty(handle, new.inode);
3877 if (unlikely(retval))
3878 goto end_rename;
3879 if (!new.inode->i_nlink)
3880 ext4_orphan_add(handle, new.inode);
3881 }
3882 retval = 0;
3883
3884end_rename:
3885 brelse(old.dir_bh);
3886 brelse(old.bh);
3887 brelse(new.bh);
3888 if (whiteout) {
3889 if (retval)
3890 drop_nlink(whiteout);
3891 unlock_new_inode(whiteout);
3892 iput(whiteout);
3893 }
3894 if (handle)
3895 ext4_journal_stop(handle);
3896 return retval;
3897}
3898
3899static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3900 struct inode *new_dir, struct dentry *new_dentry)
3901{
3902 handle_t *handle = NULL;
3903 struct ext4_renament old = {
3904 .dir = old_dir,
3905 .dentry = old_dentry,
3906 .inode = d_inode(old_dentry),
3907 };
3908 struct ext4_renament new = {
3909 .dir = new_dir,
3910 .dentry = new_dentry,
3911 .inode = d_inode(new_dentry),
3912 };
3913 u8 new_file_type;
3914 int retval;
3915 struct timespec64 ctime;
3916
3917 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3918 !projid_eq(EXT4_I(new_dir)->i_projid,
3919 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3920 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3921 !projid_eq(EXT4_I(old_dir)->i_projid,
3922 EXT4_I(new_dentry->d_inode)->i_projid)))
3923 return -EXDEV;
3924
3925 retval = dquot_initialize(old.dir);
3926 if (retval)
3927 return retval;
3928 retval = dquot_initialize(new.dir);
3929 if (retval)
3930 return retval;
3931
3932 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3933 &old.de, &old.inlined);
3934 if (IS_ERR(old.bh))
3935 return PTR_ERR(old.bh);
3936 /*
3937 * Check for inode number is _not_ due to possible IO errors.
3938 * We might rmdir the source, keep it as pwd of some process
3939 * and merrily kill the link to whatever was created under the
3940 * same name. Goodbye sticky bit ;-<
3941 */
3942 retval = -ENOENT;
3943 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3944 goto end_rename;
3945
3946 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3947 &new.de, &new.inlined);
3948 if (IS_ERR(new.bh)) {
3949 retval = PTR_ERR(new.bh);
3950 new.bh = NULL;
3951 goto end_rename;
3952 }
3953
3954 /* RENAME_EXCHANGE case: old *and* new must both exist */
3955 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3956 goto end_rename;
3957
3958 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3959 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3960 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3961 if (IS_ERR(handle)) {
3962 retval = PTR_ERR(handle);
3963 handle = NULL;
3964 goto end_rename;
3965 }
3966
3967 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3968 ext4_handle_sync(handle);
3969
3970 if (S_ISDIR(old.inode->i_mode)) {
3971 old.is_dir = true;
3972 retval = ext4_rename_dir_prepare(handle, &old);
3973 if (retval)
3974 goto end_rename;
3975 }
3976 if (S_ISDIR(new.inode->i_mode)) {
3977 new.is_dir = true;
3978 retval = ext4_rename_dir_prepare(handle, &new);
3979 if (retval)
3980 goto end_rename;
3981 }
3982
3983 /*
3984 * Other than the special case of overwriting a directory, parents'
3985 * nlink only needs to be modified if this is a cross directory rename.
3986 */
3987 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3988 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3989 new.dir_nlink_delta = -old.dir_nlink_delta;
3990 retval = -EMLINK;
3991 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3992 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3993 goto end_rename;
3994 }
3995
3996 new_file_type = new.de->file_type;
3997 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3998 if (retval)
3999 goto end_rename;
4000
4001 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4002 if (retval)
4003 goto end_rename;
4004
4005 /*
4006 * Like most other Unix systems, set the ctime for inodes on a
4007 * rename.
4008 */
4009 ctime = current_time(old.inode);
4010 old.inode->i_ctime = ctime;
4011 new.inode->i_ctime = ctime;
4012 retval = ext4_mark_inode_dirty(handle, old.inode);
4013 if (unlikely(retval))
4014 goto end_rename;
4015 retval = ext4_mark_inode_dirty(handle, new.inode);
4016 if (unlikely(retval))
4017 goto end_rename;
4018
4019 if (old.dir_bh) {
4020 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4021 if (retval)
4022 goto end_rename;
4023 }
4024 if (new.dir_bh) {
4025 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4026 if (retval)
4027 goto end_rename;
4028 }
4029 ext4_update_dir_count(handle, &old);
4030 ext4_update_dir_count(handle, &new);
4031 retval = 0;
4032
4033end_rename:
4034 brelse(old.dir_bh);
4035 brelse(new.dir_bh);
4036 brelse(old.bh);
4037 brelse(new.bh);
4038 if (handle)
4039 ext4_journal_stop(handle);
4040 return retval;
4041}
4042
4043static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4044 struct inode *new_dir, struct dentry *new_dentry,
4045 unsigned int flags)
4046{
4047 int err;
4048
4049 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4050 return -EIO;
4051
4052 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4053 return -EINVAL;
4054
4055 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4056 flags);
4057 if (err)
4058 return err;
4059
4060 if (flags & RENAME_EXCHANGE) {
4061 return ext4_cross_rename(old_dir, old_dentry,
4062 new_dir, new_dentry);
4063 }
4064
4065 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4066}
4067
4068/*
4069 * directories can handle most operations...
4070 */
4071const struct inode_operations ext4_dir_inode_operations = {
4072 .create = ext4_create,
4073 .lookup = ext4_lookup,
4074 .link = ext4_link,
4075 .unlink = ext4_unlink,
4076 .symlink = ext4_symlink,
4077 .mkdir = ext4_mkdir,
4078 .rmdir = ext4_rmdir,
4079 .mknod = ext4_mknod,
4080 .tmpfile = ext4_tmpfile,
4081 .rename = ext4_rename2,
4082 .setattr = ext4_setattr,
4083 .getattr = ext4_getattr,
4084 .listxattr = ext4_listxattr,
4085 .get_acl = ext4_get_acl,
4086 .set_acl = ext4_set_acl,
4087 .fiemap = ext4_fiemap,
4088};
4089
4090const struct inode_operations ext4_special_inode_operations = {
4091 .setattr = ext4_setattr,
4092 .getattr = ext4_getattr,
4093 .listxattr = ext4_listxattr,
4094 .get_acl = ext4_get_acl,
4095 .set_acl = ext4_set_acl,
4096};