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 "ext4.h"
39#include "ext4_jbd2.h"
40
41#include "xattr.h"
42#include "acl.h"
43
44#include <trace/events/ext4.h>
45/*
46 * define how far ahead to read directories while searching them.
47 */
48#define NAMEI_RA_CHUNKS 2
49#define NAMEI_RA_BLOCKS 4
50#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51
52static struct buffer_head *ext4_append(handle_t *handle,
53 struct inode *inode,
54 ext4_lblk_t *block)
55{
56 struct buffer_head *bh;
57 int err;
58
59 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
60 ((inode->i_size >> 10) >=
61 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
62 return ERR_PTR(-ENOSPC);
63
64 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
65
66 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
67 if (IS_ERR(bh))
68 return bh;
69 inode->i_size += inode->i_sb->s_blocksize;
70 EXT4_I(inode)->i_disksize = inode->i_size;
71 BUFFER_TRACE(bh, "get_write_access");
72 err = ext4_journal_get_write_access(handle, bh);
73 if (err) {
74 brelse(bh);
75 ext4_std_error(inode->i_sb, err);
76 return ERR_PTR(err);
77 }
78 return bh;
79}
80
81static int ext4_dx_csum_verify(struct inode *inode,
82 struct ext4_dir_entry *dirent);
83
84typedef enum {
85 EITHER, INDEX, DIRENT
86} dirblock_type_t;
87
88#define ext4_read_dirblock(inode, block, type) \
89 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
90
91static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
92 ext4_lblk_t block,
93 dirblock_type_t type,
94 const char *func,
95 unsigned int line)
96{
97 struct buffer_head *bh;
98 struct ext4_dir_entry *dirent;
99 int is_dx_block = 0;
100
101 bh = ext4_bread(NULL, inode, block, 0);
102 if (IS_ERR(bh)) {
103 __ext4_warning(inode->i_sb, func, line,
104 "inode #%lu: lblock %lu: comm %s: "
105 "error %ld reading directory block",
106 inode->i_ino, (unsigned long)block,
107 current->comm, PTR_ERR(bh));
108
109 return bh;
110 }
111 if (!bh) {
112 ext4_error_inode(inode, func, line, block,
113 "Directory hole found");
114 return ERR_PTR(-EFSCORRUPTED);
115 }
116 dirent = (struct ext4_dir_entry *) bh->b_data;
117 /* Determine whether or not we have an index block */
118 if (is_dx(inode)) {
119 if (block == 0)
120 is_dx_block = 1;
121 else if (ext4_rec_len_from_disk(dirent->rec_len,
122 inode->i_sb->s_blocksize) ==
123 inode->i_sb->s_blocksize)
124 is_dx_block = 1;
125 }
126 if (!is_dx_block && type == INDEX) {
127 ext4_error_inode(inode, func, line, block,
128 "directory leaf block found instead of index block");
129 return ERR_PTR(-EFSCORRUPTED);
130 }
131 if (!ext4_has_metadata_csum(inode->i_sb) ||
132 buffer_verified(bh))
133 return bh;
134
135 /*
136 * An empty leaf block can get mistaken for a index block; for
137 * this reason, we can only check the index checksum when the
138 * caller is sure it should be an index block.
139 */
140 if (is_dx_block && type == INDEX) {
141 if (ext4_dx_csum_verify(inode, dirent))
142 set_buffer_verified(bh);
143 else {
144 ext4_error_inode(inode, func, line, block,
145 "Directory index failed checksum");
146 brelse(bh);
147 return ERR_PTR(-EFSBADCRC);
148 }
149 }
150 if (!is_dx_block) {
151 if (ext4_dirent_csum_verify(inode, dirent))
152 set_buffer_verified(bh);
153 else {
154 ext4_error_inode(inode, func, line, block,
155 "Directory block failed checksum");
156 brelse(bh);
157 return ERR_PTR(-EFSBADCRC);
158 }
159 }
160 return bh;
161}
162
163#ifndef assert
164#define assert(test) J_ASSERT(test)
165#endif
166
167#ifdef DX_DEBUG
168#define dxtrace(command) command
169#else
170#define dxtrace(command)
171#endif
172
173struct fake_dirent
174{
175 __le32 inode;
176 __le16 rec_len;
177 u8 name_len;
178 u8 file_type;
179};
180
181struct dx_countlimit
182{
183 __le16 limit;
184 __le16 count;
185};
186
187struct dx_entry
188{
189 __le32 hash;
190 __le32 block;
191};
192
193/*
194 * dx_root_info is laid out so that if it should somehow get overlaid by a
195 * dirent the two low bits of the hash version will be zero. Therefore, the
196 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
197 */
198
199struct dx_root
200{
201 struct fake_dirent dot;
202 char dot_name[4];
203 struct fake_dirent dotdot;
204 char dotdot_name[4];
205 struct dx_root_info
206 {
207 __le32 reserved_zero;
208 u8 hash_version;
209 u8 info_length; /* 8 */
210 u8 indirect_levels;
211 u8 unused_flags;
212 }
213 info;
214 struct dx_entry entries[0];
215};
216
217struct dx_node
218{
219 struct fake_dirent fake;
220 struct dx_entry entries[0];
221};
222
223
224struct dx_frame
225{
226 struct buffer_head *bh;
227 struct dx_entry *entries;
228 struct dx_entry *at;
229};
230
231struct dx_map_entry
232{
233 u32 hash;
234 u16 offs;
235 u16 size;
236};
237
238/*
239 * This goes at the end of each htree block.
240 */
241struct dx_tail {
242 u32 dt_reserved;
243 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
244};
245
246static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
247static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
248static inline unsigned dx_get_hash(struct dx_entry *entry);
249static void dx_set_hash(struct dx_entry *entry, unsigned value);
250static unsigned dx_get_count(struct dx_entry *entries);
251static unsigned dx_get_limit(struct dx_entry *entries);
252static void dx_set_count(struct dx_entry *entries, unsigned value);
253static void dx_set_limit(struct dx_entry *entries, unsigned value);
254static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
255static unsigned dx_node_limit(struct inode *dir);
256static struct dx_frame *dx_probe(struct ext4_filename *fname,
257 struct inode *dir,
258 struct dx_hash_info *hinfo,
259 struct dx_frame *frame);
260static void dx_release(struct dx_frame *frames);
261static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
262 unsigned blocksize, struct dx_hash_info *hinfo,
263 struct dx_map_entry map[]);
264static void dx_sort_map(struct dx_map_entry *map, unsigned count);
265static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
266 struct dx_map_entry *offsets, int count, unsigned blocksize);
267static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
268static void dx_insert_block(struct dx_frame *frame,
269 u32 hash, ext4_lblk_t block);
270static int ext4_htree_next_block(struct inode *dir, __u32 hash,
271 struct dx_frame *frame,
272 struct dx_frame *frames,
273 __u32 *start_hash);
274static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
275 struct ext4_filename *fname,
276 struct ext4_dir_entry_2 **res_dir);
277static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
278 struct inode *dir, struct inode *inode);
279
280/* checksumming functions */
281void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
282 unsigned int blocksize)
283{
284 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
285 t->det_rec_len = ext4_rec_len_to_disk(
286 sizeof(struct ext4_dir_entry_tail), blocksize);
287 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
288}
289
290/* Walk through a dirent block to find a checksum "dirent" at the tail */
291static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
292 struct ext4_dir_entry *de)
293{
294 struct ext4_dir_entry_tail *t;
295
296#ifdef PARANOID
297 struct ext4_dir_entry *d, *top;
298
299 d = de;
300 top = (struct ext4_dir_entry *)(((void *)de) +
301 (EXT4_BLOCK_SIZE(inode->i_sb) -
302 sizeof(struct ext4_dir_entry_tail)));
303 while (d < top && d->rec_len)
304 d = (struct ext4_dir_entry *)(((void *)d) +
305 le16_to_cpu(d->rec_len));
306
307 if (d != top)
308 return NULL;
309
310 t = (struct ext4_dir_entry_tail *)d;
311#else
312 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
313#endif
314
315 if (t->det_reserved_zero1 ||
316 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
317 t->det_reserved_zero2 ||
318 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
319 return NULL;
320
321 return t;
322}
323
324static __le32 ext4_dirent_csum(struct inode *inode,
325 struct ext4_dir_entry *dirent, int size)
326{
327 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
328 struct ext4_inode_info *ei = EXT4_I(inode);
329 __u32 csum;
330
331 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
332 return cpu_to_le32(csum);
333}
334
335#define warn_no_space_for_csum(inode) \
336 __warn_no_space_for_csum((inode), __func__, __LINE__)
337
338static void __warn_no_space_for_csum(struct inode *inode, const char *func,
339 unsigned int line)
340{
341 __ext4_warning_inode(inode, func, line,
342 "No space for directory leaf checksum. Please run e2fsck -D.");
343}
344
345int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
346{
347 struct ext4_dir_entry_tail *t;
348
349 if (!ext4_has_metadata_csum(inode->i_sb))
350 return 1;
351
352 t = get_dirent_tail(inode, dirent);
353 if (!t) {
354 warn_no_space_for_csum(inode);
355 return 0;
356 }
357
358 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
359 (void *)t - (void *)dirent))
360 return 0;
361
362 return 1;
363}
364
365static void ext4_dirent_csum_set(struct inode *inode,
366 struct ext4_dir_entry *dirent)
367{
368 struct ext4_dir_entry_tail *t;
369
370 if (!ext4_has_metadata_csum(inode->i_sb))
371 return;
372
373 t = get_dirent_tail(inode, dirent);
374 if (!t) {
375 warn_no_space_for_csum(inode);
376 return;
377 }
378
379 t->det_checksum = ext4_dirent_csum(inode, dirent,
380 (void *)t - (void *)dirent);
381}
382
383int ext4_handle_dirty_dirent_node(handle_t *handle,
384 struct inode *inode,
385 struct buffer_head *bh)
386{
387 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
388 return ext4_handle_dirty_metadata(handle, inode, bh);
389}
390
391static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
392 struct ext4_dir_entry *dirent,
393 int *offset)
394{
395 struct ext4_dir_entry *dp;
396 struct dx_root_info *root;
397 int count_offset;
398
399 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
400 count_offset = 8;
401 else if (le16_to_cpu(dirent->rec_len) == 12) {
402 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
403 if (le16_to_cpu(dp->rec_len) !=
404 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
405 return NULL;
406 root = (struct dx_root_info *)(((void *)dp + 12));
407 if (root->reserved_zero ||
408 root->info_length != sizeof(struct dx_root_info))
409 return NULL;
410 count_offset = 32;
411 } else
412 return NULL;
413
414 if (offset)
415 *offset = count_offset;
416 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
417}
418
419static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
420 int count_offset, int count, struct dx_tail *t)
421{
422 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
423 struct ext4_inode_info *ei = EXT4_I(inode);
424 __u32 csum;
425 int size;
426 __u32 dummy_csum = 0;
427 int offset = offsetof(struct dx_tail, dt_checksum);
428
429 size = count_offset + (count * sizeof(struct dx_entry));
430 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
431 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
432 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
433
434 return cpu_to_le32(csum);
435}
436
437static int ext4_dx_csum_verify(struct inode *inode,
438 struct ext4_dir_entry *dirent)
439{
440 struct dx_countlimit *c;
441 struct dx_tail *t;
442 int count_offset, limit, count;
443
444 if (!ext4_has_metadata_csum(inode->i_sb))
445 return 1;
446
447 c = get_dx_countlimit(inode, dirent, &count_offset);
448 if (!c) {
449 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
450 return 0;
451 }
452 limit = le16_to_cpu(c->limit);
453 count = le16_to_cpu(c->count);
454 if (count_offset + (limit * sizeof(struct dx_entry)) >
455 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
456 warn_no_space_for_csum(inode);
457 return 0;
458 }
459 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
460
461 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
462 count, t))
463 return 0;
464 return 1;
465}
466
467static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
468{
469 struct dx_countlimit *c;
470 struct dx_tail *t;
471 int count_offset, limit, count;
472
473 if (!ext4_has_metadata_csum(inode->i_sb))
474 return;
475
476 c = get_dx_countlimit(inode, dirent, &count_offset);
477 if (!c) {
478 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
479 return;
480 }
481 limit = le16_to_cpu(c->limit);
482 count = le16_to_cpu(c->count);
483 if (count_offset + (limit * sizeof(struct dx_entry)) >
484 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
485 warn_no_space_for_csum(inode);
486 return;
487 }
488 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
489
490 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
491}
492
493static inline int ext4_handle_dirty_dx_node(handle_t *handle,
494 struct inode *inode,
495 struct buffer_head *bh)
496{
497 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
498 return ext4_handle_dirty_metadata(handle, inode, bh);
499}
500
501/*
502 * p is at least 6 bytes before the end of page
503 */
504static inline struct ext4_dir_entry_2 *
505ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
506{
507 return (struct ext4_dir_entry_2 *)((char *)p +
508 ext4_rec_len_from_disk(p->rec_len, blocksize));
509}
510
511/*
512 * Future: use high four bits of block for coalesce-on-delete flags
513 * Mask them off for now.
514 */
515
516static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
517{
518 return le32_to_cpu(entry->block) & 0x0fffffff;
519}
520
521static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
522{
523 entry->block = cpu_to_le32(value);
524}
525
526static inline unsigned dx_get_hash(struct dx_entry *entry)
527{
528 return le32_to_cpu(entry->hash);
529}
530
531static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
532{
533 entry->hash = cpu_to_le32(value);
534}
535
536static inline unsigned dx_get_count(struct dx_entry *entries)
537{
538 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
539}
540
541static inline unsigned dx_get_limit(struct dx_entry *entries)
542{
543 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
544}
545
546static inline void dx_set_count(struct dx_entry *entries, unsigned value)
547{
548 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
549}
550
551static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
552{
553 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
554}
555
556static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
557{
558 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
559 EXT4_DIR_REC_LEN(2) - infosize;
560
561 if (ext4_has_metadata_csum(dir->i_sb))
562 entry_space -= sizeof(struct dx_tail);
563 return entry_space / sizeof(struct dx_entry);
564}
565
566static inline unsigned dx_node_limit(struct inode *dir)
567{
568 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
569
570 if (ext4_has_metadata_csum(dir->i_sb))
571 entry_space -= sizeof(struct dx_tail);
572 return entry_space / sizeof(struct dx_entry);
573}
574
575/*
576 * Debug
577 */
578#ifdef DX_DEBUG
579static void dx_show_index(char * label, struct dx_entry *entries)
580{
581 int i, n = dx_get_count (entries);
582 printk(KERN_DEBUG "%s index", label);
583 for (i = 0; i < n; i++) {
584 printk(KERN_CONT " %x->%lu",
585 i ? dx_get_hash(entries + i) : 0,
586 (unsigned long)dx_get_block(entries + i));
587 }
588 printk(KERN_CONT "\n");
589}
590
591struct stats
592{
593 unsigned names;
594 unsigned space;
595 unsigned bcount;
596};
597
598static struct stats dx_show_leaf(struct inode *dir,
599 struct dx_hash_info *hinfo,
600 struct ext4_dir_entry_2 *de,
601 int size, int show_names)
602{
603 unsigned names = 0, space = 0;
604 char *base = (char *) de;
605 struct dx_hash_info h = *hinfo;
606
607 printk("names: ");
608 while ((char *) de < base + size)
609 {
610 if (de->inode)
611 {
612 if (show_names)
613 {
614#ifdef CONFIG_EXT4_FS_ENCRYPTION
615 int len;
616 char *name;
617 struct fscrypt_str fname_crypto_str =
618 FSTR_INIT(NULL, 0);
619 int res = 0;
620
621 name = de->name;
622 len = de->name_len;
623 if (ext4_encrypted_inode(dir))
624 res = fscrypt_get_encryption_info(dir);
625 if (res) {
626 printk(KERN_WARNING "Error setting up"
627 " fname crypto: %d\n", res);
628 }
629 if (!fscrypt_has_encryption_key(dir)) {
630 /* Directory is not encrypted */
631 ext4fs_dirhash(de->name,
632 de->name_len, &h);
633 printk("%*.s:(U)%x.%u ", len,
634 name, h.hash,
635 (unsigned) ((char *) de
636 - base));
637 } else {
638 struct fscrypt_str de_name =
639 FSTR_INIT(name, len);
640
641 /* Directory is encrypted */
642 res = fscrypt_fname_alloc_buffer(
643 dir, len,
644 &fname_crypto_str);
645 if (res)
646 printk(KERN_WARNING "Error "
647 "allocating crypto "
648 "buffer--skipping "
649 "crypto\n");
650 res = fscrypt_fname_disk_to_usr(dir,
651 0, 0, &de_name,
652 &fname_crypto_str);
653 if (res) {
654 printk(KERN_WARNING "Error "
655 "converting filename "
656 "from disk to usr"
657 "\n");
658 name = "??";
659 len = 2;
660 } else {
661 name = fname_crypto_str.name;
662 len = fname_crypto_str.len;
663 }
664 ext4fs_dirhash(de->name, de->name_len,
665 &h);
666 printk("%*.s:(E)%x.%u ", len, name,
667 h.hash, (unsigned) ((char *) de
668 - base));
669 fscrypt_fname_free_buffer(
670 &fname_crypto_str);
671 }
672#else
673 int len = de->name_len;
674 char *name = de->name;
675 ext4fs_dirhash(de->name, de->name_len, &h);
676 printk("%*.s:%x.%u ", len, name, h.hash,
677 (unsigned) ((char *) de - base));
678#endif
679 }
680 space += EXT4_DIR_REC_LEN(de->name_len);
681 names++;
682 }
683 de = ext4_next_entry(de, size);
684 }
685 printk(KERN_CONT "(%i)\n", names);
686 return (struct stats) { names, space, 1 };
687}
688
689struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
690 struct dx_entry *entries, int levels)
691{
692 unsigned blocksize = dir->i_sb->s_blocksize;
693 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
694 unsigned bcount = 0;
695 struct buffer_head *bh;
696 printk("%i indexed blocks...\n", count);
697 for (i = 0; i < count; i++, entries++)
698 {
699 ext4_lblk_t block = dx_get_block(entries);
700 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
701 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
702 struct stats stats;
703 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
704 bh = ext4_bread(NULL,dir, block, 0);
705 if (!bh || IS_ERR(bh))
706 continue;
707 stats = levels?
708 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
709 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
710 bh->b_data, blocksize, 0);
711 names += stats.names;
712 space += stats.space;
713 bcount += stats.bcount;
714 brelse(bh);
715 }
716 if (bcount)
717 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
718 levels ? "" : " ", names, space/bcount,
719 (space/bcount)*100/blocksize);
720 return (struct stats) { names, space, bcount};
721}
722#endif /* DX_DEBUG */
723
724/*
725 * Probe for a directory leaf block to search.
726 *
727 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
728 * error in the directory index, and the caller should fall back to
729 * searching the directory normally. The callers of dx_probe **MUST**
730 * check for this error code, and make sure it never gets reflected
731 * back to userspace.
732 */
733static struct dx_frame *
734dx_probe(struct ext4_filename *fname, struct inode *dir,
735 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
736{
737 unsigned count, indirect;
738 struct dx_entry *at, *entries, *p, *q, *m;
739 struct dx_root *root;
740 struct dx_frame *frame = frame_in;
741 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
742 u32 hash;
743
744 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
745 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
746 if (IS_ERR(frame->bh))
747 return (struct dx_frame *) frame->bh;
748
749 root = (struct dx_root *) frame->bh->b_data;
750 if (root->info.hash_version != DX_HASH_TEA &&
751 root->info.hash_version != DX_HASH_HALF_MD4 &&
752 root->info.hash_version != DX_HASH_LEGACY) {
753 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
754 root->info.hash_version);
755 goto fail;
756 }
757 if (fname)
758 hinfo = &fname->hinfo;
759 hinfo->hash_version = root->info.hash_version;
760 if (hinfo->hash_version <= DX_HASH_TEA)
761 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
762 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
763 if (fname && fname_name(fname))
764 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
765 hash = hinfo->hash;
766
767 if (root->info.unused_flags & 1) {
768 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
769 root->info.unused_flags);
770 goto fail;
771 }
772
773 indirect = root->info.indirect_levels;
774 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
775 ext4_warning(dir->i_sb,
776 "Directory (ino: %lu) htree depth %#06x exceed"
777 "supported value", dir->i_ino,
778 ext4_dir_htree_level(dir->i_sb));
779 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
780 ext4_warning(dir->i_sb, "Enable large directory "
781 "feature to access it");
782 }
783 goto fail;
784 }
785
786 entries = (struct dx_entry *)(((char *)&root->info) +
787 root->info.info_length);
788
789 if (dx_get_limit(entries) != dx_root_limit(dir,
790 root->info.info_length)) {
791 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
792 dx_get_limit(entries),
793 dx_root_limit(dir, root->info.info_length));
794 goto fail;
795 }
796
797 dxtrace(printk("Look up %x", hash));
798 while (1) {
799 count = dx_get_count(entries);
800 if (!count || count > dx_get_limit(entries)) {
801 ext4_warning_inode(dir,
802 "dx entry: count %u beyond limit %u",
803 count, dx_get_limit(entries));
804 goto fail;
805 }
806
807 p = entries + 1;
808 q = entries + count - 1;
809 while (p <= q) {
810 m = p + (q - p) / 2;
811 dxtrace(printk(KERN_CONT "."));
812 if (dx_get_hash(m) > hash)
813 q = m - 1;
814 else
815 p = m + 1;
816 }
817
818 if (0) { // linear search cross check
819 unsigned n = count - 1;
820 at = entries;
821 while (n--)
822 {
823 dxtrace(printk(KERN_CONT ","));
824 if (dx_get_hash(++at) > hash)
825 {
826 at--;
827 break;
828 }
829 }
830 assert (at == p - 1);
831 }
832
833 at = p - 1;
834 dxtrace(printk(KERN_CONT " %x->%u\n",
835 at == entries ? 0 : dx_get_hash(at),
836 dx_get_block(at)));
837 frame->entries = entries;
838 frame->at = at;
839 if (!indirect--)
840 return frame;
841 frame++;
842 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
843 if (IS_ERR(frame->bh)) {
844 ret_err = (struct dx_frame *) frame->bh;
845 frame->bh = NULL;
846 goto fail;
847 }
848 entries = ((struct dx_node *) frame->bh->b_data)->entries;
849
850 if (dx_get_limit(entries) != dx_node_limit(dir)) {
851 ext4_warning_inode(dir,
852 "dx entry: limit %u != node limit %u",
853 dx_get_limit(entries), dx_node_limit(dir));
854 goto fail;
855 }
856 }
857fail:
858 while (frame >= frame_in) {
859 brelse(frame->bh);
860 frame--;
861 }
862
863 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
864 ext4_warning_inode(dir,
865 "Corrupt directory, running e2fsck is recommended");
866 return ret_err;
867}
868
869static void dx_release(struct dx_frame *frames)
870{
871 struct dx_root_info *info;
872 int i;
873
874 if (frames[0].bh == NULL)
875 return;
876
877 info = &((struct dx_root *)frames[0].bh->b_data)->info;
878 for (i = 0; i <= info->indirect_levels; i++) {
879 if (frames[i].bh == NULL)
880 break;
881 brelse(frames[i].bh);
882 frames[i].bh = NULL;
883 }
884}
885
886/*
887 * This function increments the frame pointer to search the next leaf
888 * block, and reads in the necessary intervening nodes if the search
889 * should be necessary. Whether or not the search is necessary is
890 * controlled by the hash parameter. If the hash value is even, then
891 * the search is only continued if the next block starts with that
892 * hash value. This is used if we are searching for a specific file.
893 *
894 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
895 *
896 * This function returns 1 if the caller should continue to search,
897 * or 0 if it should not. If there is an error reading one of the
898 * index blocks, it will a negative error code.
899 *
900 * If start_hash is non-null, it will be filled in with the starting
901 * hash of the next page.
902 */
903static int ext4_htree_next_block(struct inode *dir, __u32 hash,
904 struct dx_frame *frame,
905 struct dx_frame *frames,
906 __u32 *start_hash)
907{
908 struct dx_frame *p;
909 struct buffer_head *bh;
910 int num_frames = 0;
911 __u32 bhash;
912
913 p = frame;
914 /*
915 * Find the next leaf page by incrementing the frame pointer.
916 * If we run out of entries in the interior node, loop around and
917 * increment pointer in the parent node. When we break out of
918 * this loop, num_frames indicates the number of interior
919 * nodes need to be read.
920 */
921 while (1) {
922 if (++(p->at) < p->entries + dx_get_count(p->entries))
923 break;
924 if (p == frames)
925 return 0;
926 num_frames++;
927 p--;
928 }
929
930 /*
931 * If the hash is 1, then continue only if the next page has a
932 * continuation hash of any value. This is used for readdir
933 * handling. Otherwise, check to see if the hash matches the
934 * desired contiuation hash. If it doesn't, return since
935 * there's no point to read in the successive index pages.
936 */
937 bhash = dx_get_hash(p->at);
938 if (start_hash)
939 *start_hash = bhash;
940 if ((hash & 1) == 0) {
941 if ((bhash & ~1) != hash)
942 return 0;
943 }
944 /*
945 * If the hash is HASH_NB_ALWAYS, we always go to the next
946 * block so no check is necessary
947 */
948 while (num_frames--) {
949 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
950 if (IS_ERR(bh))
951 return PTR_ERR(bh);
952 p++;
953 brelse(p->bh);
954 p->bh = bh;
955 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
956 }
957 return 1;
958}
959
960
961/*
962 * This function fills a red-black tree with information from a
963 * directory block. It returns the number directory entries loaded
964 * into the tree. If there is an error it is returned in err.
965 */
966static int htree_dirblock_to_tree(struct file *dir_file,
967 struct inode *dir, ext4_lblk_t block,
968 struct dx_hash_info *hinfo,
969 __u32 start_hash, __u32 start_minor_hash)
970{
971 struct buffer_head *bh;
972 struct ext4_dir_entry_2 *de, *top;
973 int err = 0, count = 0;
974 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
975
976 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
977 (unsigned long)block));
978 bh = ext4_read_dirblock(dir, block, DIRENT);
979 if (IS_ERR(bh))
980 return PTR_ERR(bh);
981
982 de = (struct ext4_dir_entry_2 *) bh->b_data;
983 top = (struct ext4_dir_entry_2 *) ((char *) de +
984 dir->i_sb->s_blocksize -
985 EXT4_DIR_REC_LEN(0));
986#ifdef CONFIG_EXT4_FS_ENCRYPTION
987 /* Check if the directory is encrypted */
988 if (ext4_encrypted_inode(dir)) {
989 err = fscrypt_get_encryption_info(dir);
990 if (err < 0) {
991 brelse(bh);
992 return err;
993 }
994 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
995 &fname_crypto_str);
996 if (err < 0) {
997 brelse(bh);
998 return err;
999 }
1000 }
1001#endif
1002 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1003 if (ext4_check_dir_entry(dir, NULL, de, bh,
1004 bh->b_data, bh->b_size,
1005 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1006 + ((char *)de - bh->b_data))) {
1007 /* silently ignore the rest of the block */
1008 break;
1009 }
1010 ext4fs_dirhash(de->name, de->name_len, hinfo);
1011 if ((hinfo->hash < start_hash) ||
1012 ((hinfo->hash == start_hash) &&
1013 (hinfo->minor_hash < start_minor_hash)))
1014 continue;
1015 if (de->inode == 0)
1016 continue;
1017 if (!ext4_encrypted_inode(dir)) {
1018 tmp_str.name = de->name;
1019 tmp_str.len = de->name_len;
1020 err = ext4_htree_store_dirent(dir_file,
1021 hinfo->hash, hinfo->minor_hash, de,
1022 &tmp_str);
1023 } else {
1024 int save_len = fname_crypto_str.len;
1025 struct fscrypt_str de_name = FSTR_INIT(de->name,
1026 de->name_len);
1027
1028 /* Directory is encrypted */
1029 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1030 hinfo->minor_hash, &de_name,
1031 &fname_crypto_str);
1032 if (err) {
1033 count = err;
1034 goto errout;
1035 }
1036 err = ext4_htree_store_dirent(dir_file,
1037 hinfo->hash, hinfo->minor_hash, de,
1038 &fname_crypto_str);
1039 fname_crypto_str.len = save_len;
1040 }
1041 if (err != 0) {
1042 count = err;
1043 goto errout;
1044 }
1045 count++;
1046 }
1047errout:
1048 brelse(bh);
1049#ifdef CONFIG_EXT4_FS_ENCRYPTION
1050 fscrypt_fname_free_buffer(&fname_crypto_str);
1051#endif
1052 return count;
1053}
1054
1055
1056/*
1057 * This function fills a red-black tree with information from a
1058 * directory. We start scanning the directory in hash order, starting
1059 * at start_hash and start_minor_hash.
1060 *
1061 * This function returns the number of entries inserted into the tree,
1062 * or a negative error code.
1063 */
1064int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1065 __u32 start_minor_hash, __u32 *next_hash)
1066{
1067 struct dx_hash_info hinfo;
1068 struct ext4_dir_entry_2 *de;
1069 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1070 struct inode *dir;
1071 ext4_lblk_t block;
1072 int count = 0;
1073 int ret, err;
1074 __u32 hashval;
1075 struct fscrypt_str tmp_str;
1076
1077 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1078 start_hash, start_minor_hash));
1079 dir = file_inode(dir_file);
1080 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1081 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1082 if (hinfo.hash_version <= DX_HASH_TEA)
1083 hinfo.hash_version +=
1084 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1085 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1086 if (ext4_has_inline_data(dir)) {
1087 int has_inline_data = 1;
1088 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1089 &hinfo, start_hash,
1090 start_minor_hash,
1091 &has_inline_data);
1092 if (has_inline_data) {
1093 *next_hash = ~0;
1094 return count;
1095 }
1096 }
1097 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1098 start_hash, start_minor_hash);
1099 *next_hash = ~0;
1100 return count;
1101 }
1102 hinfo.hash = start_hash;
1103 hinfo.minor_hash = 0;
1104 frame = dx_probe(NULL, dir, &hinfo, frames);
1105 if (IS_ERR(frame))
1106 return PTR_ERR(frame);
1107
1108 /* Add '.' and '..' from the htree header */
1109 if (!start_hash && !start_minor_hash) {
1110 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1111 tmp_str.name = de->name;
1112 tmp_str.len = de->name_len;
1113 err = ext4_htree_store_dirent(dir_file, 0, 0,
1114 de, &tmp_str);
1115 if (err != 0)
1116 goto errout;
1117 count++;
1118 }
1119 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1120 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1121 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1122 tmp_str.name = de->name;
1123 tmp_str.len = de->name_len;
1124 err = ext4_htree_store_dirent(dir_file, 2, 0,
1125 de, &tmp_str);
1126 if (err != 0)
1127 goto errout;
1128 count++;
1129 }
1130
1131 while (1) {
1132 if (fatal_signal_pending(current)) {
1133 err = -ERESTARTSYS;
1134 goto errout;
1135 }
1136 cond_resched();
1137 block = dx_get_block(frame->at);
1138 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1139 start_hash, start_minor_hash);
1140 if (ret < 0) {
1141 err = ret;
1142 goto errout;
1143 }
1144 count += ret;
1145 hashval = ~0;
1146 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1147 frame, frames, &hashval);
1148 *next_hash = hashval;
1149 if (ret < 0) {
1150 err = ret;
1151 goto errout;
1152 }
1153 /*
1154 * Stop if: (a) there are no more entries, or
1155 * (b) we have inserted at least one entry and the
1156 * next hash value is not a continuation
1157 */
1158 if ((ret == 0) ||
1159 (count && ((hashval & 1) == 0)))
1160 break;
1161 }
1162 dx_release(frames);
1163 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1164 "next hash: %x\n", count, *next_hash));
1165 return count;
1166errout:
1167 dx_release(frames);
1168 return (err);
1169}
1170
1171static inline int search_dirblock(struct buffer_head *bh,
1172 struct inode *dir,
1173 struct ext4_filename *fname,
1174 unsigned int offset,
1175 struct ext4_dir_entry_2 **res_dir)
1176{
1177 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1178 fname, offset, res_dir);
1179}
1180
1181/*
1182 * Directory block splitting, compacting
1183 */
1184
1185/*
1186 * Create map of hash values, offsets, and sizes, stored at end of block.
1187 * Returns number of entries mapped.
1188 */
1189static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1190 unsigned blocksize, struct dx_hash_info *hinfo,
1191 struct dx_map_entry *map_tail)
1192{
1193 int count = 0;
1194 char *base = (char *) de;
1195 struct dx_hash_info h = *hinfo;
1196
1197 while ((char *) de < base + blocksize) {
1198 if (de->name_len && de->inode) {
1199 ext4fs_dirhash(de->name, de->name_len, &h);
1200 map_tail--;
1201 map_tail->hash = h.hash;
1202 map_tail->offs = ((char *) de - base)>>2;
1203 map_tail->size = le16_to_cpu(de->rec_len);
1204 count++;
1205 cond_resched();
1206 }
1207 /* XXX: do we need to check rec_len == 0 case? -Chris */
1208 de = ext4_next_entry(de, blocksize);
1209 }
1210 return count;
1211}
1212
1213/* Sort map by hash value */
1214static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1215{
1216 struct dx_map_entry *p, *q, *top = map + count - 1;
1217 int more;
1218 /* Combsort until bubble sort doesn't suck */
1219 while (count > 2) {
1220 count = count*10/13;
1221 if (count - 9 < 2) /* 9, 10 -> 11 */
1222 count = 11;
1223 for (p = top, q = p - count; q >= map; p--, q--)
1224 if (p->hash < q->hash)
1225 swap(*p, *q);
1226 }
1227 /* Garden variety bubble sort */
1228 do {
1229 more = 0;
1230 q = top;
1231 while (q-- > map) {
1232 if (q[1].hash >= q[0].hash)
1233 continue;
1234 swap(*(q+1), *q);
1235 more = 1;
1236 }
1237 } while(more);
1238}
1239
1240static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1241{
1242 struct dx_entry *entries = frame->entries;
1243 struct dx_entry *old = frame->at, *new = old + 1;
1244 int count = dx_get_count(entries);
1245
1246 assert(count < dx_get_limit(entries));
1247 assert(old < entries + count);
1248 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1249 dx_set_hash(new, hash);
1250 dx_set_block(new, block);
1251 dx_set_count(entries, count + 1);
1252}
1253
1254/*
1255 * Test whether a directory entry matches the filename being searched for.
1256 *
1257 * Return: %true if the directory entry matches, otherwise %false.
1258 */
1259static inline bool ext4_match(const struct ext4_filename *fname,
1260 const struct ext4_dir_entry_2 *de)
1261{
1262 struct fscrypt_name f;
1263
1264 if (!de->inode)
1265 return false;
1266
1267 f.usr_fname = fname->usr_fname;
1268 f.disk_name = fname->disk_name;
1269#ifdef CONFIG_EXT4_FS_ENCRYPTION
1270 f.crypto_buf = fname->crypto_buf;
1271#endif
1272 return fscrypt_match_name(&f, de->name, de->name_len);
1273}
1274
1275/*
1276 * Returns 0 if not found, -1 on failure, and 1 on success
1277 */
1278int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1279 struct inode *dir, struct ext4_filename *fname,
1280 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1281{
1282 struct ext4_dir_entry_2 * de;
1283 char * dlimit;
1284 int de_len;
1285
1286 de = (struct ext4_dir_entry_2 *)search_buf;
1287 dlimit = search_buf + buf_size;
1288 while ((char *) de < dlimit) {
1289 /* this code is executed quadratically often */
1290 /* do minimal checking `by hand' */
1291 if ((char *) de + de->name_len <= dlimit &&
1292 ext4_match(fname, de)) {
1293 /* found a match - just to be sure, do
1294 * a full check */
1295 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1296 bh->b_size, offset))
1297 return -1;
1298 *res_dir = de;
1299 return 1;
1300 }
1301 /* prevent looping on a bad block */
1302 de_len = ext4_rec_len_from_disk(de->rec_len,
1303 dir->i_sb->s_blocksize);
1304 if (de_len <= 0)
1305 return -1;
1306 offset += de_len;
1307 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1308 }
1309 return 0;
1310}
1311
1312static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1313 struct ext4_dir_entry *de)
1314{
1315 struct super_block *sb = dir->i_sb;
1316
1317 if (!is_dx(dir))
1318 return 0;
1319 if (block == 0)
1320 return 1;
1321 if (de->inode == 0 &&
1322 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1323 sb->s_blocksize)
1324 return 1;
1325 return 0;
1326}
1327
1328/*
1329 * ext4_find_entry()
1330 *
1331 * finds an entry in the specified directory with the wanted name. It
1332 * returns the cache buffer in which the entry was found, and the entry
1333 * itself (as a parameter - res_dir). It does NOT read the inode of the
1334 * entry - you'll have to do that yourself if you want to.
1335 *
1336 * The returned buffer_head has ->b_count elevated. The caller is expected
1337 * to brelse() it when appropriate.
1338 */
1339static struct buffer_head * ext4_find_entry (struct inode *dir,
1340 const struct qstr *d_name,
1341 struct ext4_dir_entry_2 **res_dir,
1342 int *inlined)
1343{
1344 struct super_block *sb;
1345 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1346 struct buffer_head *bh, *ret = NULL;
1347 ext4_lblk_t start, block;
1348 const u8 *name = d_name->name;
1349 size_t ra_max = 0; /* Number of bh's in the readahead
1350 buffer, bh_use[] */
1351 size_t ra_ptr = 0; /* Current index into readahead
1352 buffer */
1353 ext4_lblk_t nblocks;
1354 int i, namelen, retval;
1355 struct ext4_filename fname;
1356
1357 *res_dir = NULL;
1358 sb = dir->i_sb;
1359 namelen = d_name->len;
1360 if (namelen > EXT4_NAME_LEN)
1361 return NULL;
1362
1363 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1364 if (retval == -ENOENT)
1365 return NULL;
1366 if (retval)
1367 return ERR_PTR(retval);
1368
1369 if (ext4_has_inline_data(dir)) {
1370 int has_inline_data = 1;
1371 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1372 &has_inline_data);
1373 if (has_inline_data) {
1374 if (inlined)
1375 *inlined = 1;
1376 goto cleanup_and_exit;
1377 }
1378 }
1379
1380 if ((namelen <= 2) && (name[0] == '.') &&
1381 (name[1] == '.' || name[1] == '\0')) {
1382 /*
1383 * "." or ".." will only be in the first block
1384 * NFS may look up ".."; "." should be handled by the VFS
1385 */
1386 block = start = 0;
1387 nblocks = 1;
1388 goto restart;
1389 }
1390 if (is_dx(dir)) {
1391 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1392 /*
1393 * On success, or if the error was file not found,
1394 * return. Otherwise, fall back to doing a search the
1395 * old fashioned way.
1396 */
1397 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1398 goto cleanup_and_exit;
1399 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1400 "falling back\n"));
1401 }
1402 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1403 if (!nblocks) {
1404 ret = NULL;
1405 goto cleanup_and_exit;
1406 }
1407 start = EXT4_I(dir)->i_dir_start_lookup;
1408 if (start >= nblocks)
1409 start = 0;
1410 block = start;
1411restart:
1412 do {
1413 /*
1414 * We deal with the read-ahead logic here.
1415 */
1416 if (ra_ptr >= ra_max) {
1417 /* Refill the readahead buffer */
1418 ra_ptr = 0;
1419 if (block < start)
1420 ra_max = start - block;
1421 else
1422 ra_max = nblocks - block;
1423 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1424 retval = ext4_bread_batch(dir, block, ra_max,
1425 false /* wait */, bh_use);
1426 if (retval) {
1427 ret = ERR_PTR(retval);
1428 ra_max = 0;
1429 goto cleanup_and_exit;
1430 }
1431 }
1432 if ((bh = bh_use[ra_ptr++]) == NULL)
1433 goto next;
1434 wait_on_buffer(bh);
1435 if (!buffer_uptodate(bh)) {
1436 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1437 (unsigned long) block);
1438 brelse(bh);
1439 ret = ERR_PTR(-EIO);
1440 goto cleanup_and_exit;
1441 }
1442 if (!buffer_verified(bh) &&
1443 !is_dx_internal_node(dir, block,
1444 (struct ext4_dir_entry *)bh->b_data) &&
1445 !ext4_dirent_csum_verify(dir,
1446 (struct ext4_dir_entry *)bh->b_data)) {
1447 EXT4_ERROR_INODE(dir, "checksumming directory "
1448 "block %lu", (unsigned long)block);
1449 brelse(bh);
1450 ret = ERR_PTR(-EFSBADCRC);
1451 goto cleanup_and_exit;
1452 }
1453 set_buffer_verified(bh);
1454 i = search_dirblock(bh, dir, &fname,
1455 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1456 if (i == 1) {
1457 EXT4_I(dir)->i_dir_start_lookup = block;
1458 ret = bh;
1459 goto cleanup_and_exit;
1460 } else {
1461 brelse(bh);
1462 if (i < 0)
1463 goto cleanup_and_exit;
1464 }
1465 next:
1466 if (++block >= nblocks)
1467 block = 0;
1468 } while (block != start);
1469
1470 /*
1471 * If the directory has grown while we were searching, then
1472 * search the last part of the directory before giving up.
1473 */
1474 block = nblocks;
1475 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1476 if (block < nblocks) {
1477 start = 0;
1478 goto restart;
1479 }
1480
1481cleanup_and_exit:
1482 /* Clean up the read-ahead blocks */
1483 for (; ra_ptr < ra_max; ra_ptr++)
1484 brelse(bh_use[ra_ptr]);
1485 ext4_fname_free_filename(&fname);
1486 return ret;
1487}
1488
1489static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1490 struct ext4_filename *fname,
1491 struct ext4_dir_entry_2 **res_dir)
1492{
1493 struct super_block * sb = dir->i_sb;
1494 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1495 struct buffer_head *bh;
1496 ext4_lblk_t block;
1497 int retval;
1498
1499#ifdef CONFIG_EXT4_FS_ENCRYPTION
1500 *res_dir = NULL;
1501#endif
1502 frame = dx_probe(fname, dir, NULL, frames);
1503 if (IS_ERR(frame))
1504 return (struct buffer_head *) frame;
1505 do {
1506 block = dx_get_block(frame->at);
1507 bh = ext4_read_dirblock(dir, block, DIRENT);
1508 if (IS_ERR(bh))
1509 goto errout;
1510
1511 retval = search_dirblock(bh, dir, fname,
1512 block << EXT4_BLOCK_SIZE_BITS(sb),
1513 res_dir);
1514 if (retval == 1)
1515 goto success;
1516 brelse(bh);
1517 if (retval == -1) {
1518 bh = ERR_PTR(ERR_BAD_DX_DIR);
1519 goto errout;
1520 }
1521
1522 /* Check to see if we should continue to search */
1523 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1524 frames, NULL);
1525 if (retval < 0) {
1526 ext4_warning_inode(dir,
1527 "error %d reading directory index block",
1528 retval);
1529 bh = ERR_PTR(retval);
1530 goto errout;
1531 }
1532 } while (retval == 1);
1533
1534 bh = NULL;
1535errout:
1536 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1537success:
1538 dx_release(frames);
1539 return bh;
1540}
1541
1542static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1543{
1544 struct inode *inode;
1545 struct ext4_dir_entry_2 *de;
1546 struct buffer_head *bh;
1547 int err;
1548
1549 err = fscrypt_prepare_lookup(dir, dentry, flags);
1550 if (err)
1551 return ERR_PTR(err);
1552
1553 if (dentry->d_name.len > EXT4_NAME_LEN)
1554 return ERR_PTR(-ENAMETOOLONG);
1555
1556 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1557 if (IS_ERR(bh))
1558 return (struct dentry *) bh;
1559 inode = NULL;
1560 if (bh) {
1561 __u32 ino = le32_to_cpu(de->inode);
1562 brelse(bh);
1563 if (!ext4_valid_inum(dir->i_sb, ino)) {
1564 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1565 return ERR_PTR(-EFSCORRUPTED);
1566 }
1567 if (unlikely(ino == dir->i_ino)) {
1568 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1569 dentry);
1570 return ERR_PTR(-EFSCORRUPTED);
1571 }
1572 inode = ext4_iget_normal(dir->i_sb, ino);
1573 if (inode == ERR_PTR(-ESTALE)) {
1574 EXT4_ERROR_INODE(dir,
1575 "deleted inode referenced: %u",
1576 ino);
1577 return ERR_PTR(-EFSCORRUPTED);
1578 }
1579 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1580 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1581 !fscrypt_has_permitted_context(dir, inode)) {
1582 ext4_warning(inode->i_sb,
1583 "Inconsistent encryption contexts: %lu/%lu",
1584 dir->i_ino, inode->i_ino);
1585 iput(inode);
1586 return ERR_PTR(-EPERM);
1587 }
1588 }
1589 return d_splice_alias(inode, dentry);
1590}
1591
1592
1593struct dentry *ext4_get_parent(struct dentry *child)
1594{
1595 __u32 ino;
1596 static const struct qstr dotdot = QSTR_INIT("..", 2);
1597 struct ext4_dir_entry_2 * de;
1598 struct buffer_head *bh;
1599
1600 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1601 if (IS_ERR(bh))
1602 return (struct dentry *) bh;
1603 if (!bh)
1604 return ERR_PTR(-ENOENT);
1605 ino = le32_to_cpu(de->inode);
1606 brelse(bh);
1607
1608 if (!ext4_valid_inum(child->d_sb, ino)) {
1609 EXT4_ERROR_INODE(d_inode(child),
1610 "bad parent inode number: %u", ino);
1611 return ERR_PTR(-EFSCORRUPTED);
1612 }
1613
1614 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1615}
1616
1617/*
1618 * Move count entries from end of map between two memory locations.
1619 * Returns pointer to last entry moved.
1620 */
1621static struct ext4_dir_entry_2 *
1622dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1623 unsigned blocksize)
1624{
1625 unsigned rec_len = 0;
1626
1627 while (count--) {
1628 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1629 (from + (map->offs<<2));
1630 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1631 memcpy (to, de, rec_len);
1632 ((struct ext4_dir_entry_2 *) to)->rec_len =
1633 ext4_rec_len_to_disk(rec_len, blocksize);
1634 de->inode = 0;
1635 map++;
1636 to += rec_len;
1637 }
1638 return (struct ext4_dir_entry_2 *) (to - rec_len);
1639}
1640
1641/*
1642 * Compact each dir entry in the range to the minimal rec_len.
1643 * Returns pointer to last entry in range.
1644 */
1645static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1646{
1647 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1648 unsigned rec_len = 0;
1649
1650 prev = to = de;
1651 while ((char*)de < base + blocksize) {
1652 next = ext4_next_entry(de, blocksize);
1653 if (de->inode && de->name_len) {
1654 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1655 if (de > to)
1656 memmove(to, de, rec_len);
1657 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1658 prev = to;
1659 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1660 }
1661 de = next;
1662 }
1663 return prev;
1664}
1665
1666/*
1667 * Split a full leaf block to make room for a new dir entry.
1668 * Allocate a new block, and move entries so that they are approx. equally full.
1669 * Returns pointer to de in block into which the new entry will be inserted.
1670 */
1671static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1672 struct buffer_head **bh,struct dx_frame *frame,
1673 struct dx_hash_info *hinfo)
1674{
1675 unsigned blocksize = dir->i_sb->s_blocksize;
1676 unsigned count, continued;
1677 struct buffer_head *bh2;
1678 ext4_lblk_t newblock;
1679 u32 hash2;
1680 struct dx_map_entry *map;
1681 char *data1 = (*bh)->b_data, *data2;
1682 unsigned split, move, size;
1683 struct ext4_dir_entry_2 *de = NULL, *de2;
1684 struct ext4_dir_entry_tail *t;
1685 int csum_size = 0;
1686 int err = 0, i;
1687
1688 if (ext4_has_metadata_csum(dir->i_sb))
1689 csum_size = sizeof(struct ext4_dir_entry_tail);
1690
1691 bh2 = ext4_append(handle, dir, &newblock);
1692 if (IS_ERR(bh2)) {
1693 brelse(*bh);
1694 *bh = NULL;
1695 return (struct ext4_dir_entry_2 *) bh2;
1696 }
1697
1698 BUFFER_TRACE(*bh, "get_write_access");
1699 err = ext4_journal_get_write_access(handle, *bh);
1700 if (err)
1701 goto journal_error;
1702
1703 BUFFER_TRACE(frame->bh, "get_write_access");
1704 err = ext4_journal_get_write_access(handle, frame->bh);
1705 if (err)
1706 goto journal_error;
1707
1708 data2 = bh2->b_data;
1709
1710 /* create map in the end of data2 block */
1711 map = (struct dx_map_entry *) (data2 + blocksize);
1712 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1713 blocksize, hinfo, map);
1714 map -= count;
1715 dx_sort_map(map, count);
1716 /* Split the existing block in the middle, size-wise */
1717 size = 0;
1718 move = 0;
1719 for (i = count-1; i >= 0; i--) {
1720 /* is more than half of this entry in 2nd half of the block? */
1721 if (size + map[i].size/2 > blocksize/2)
1722 break;
1723 size += map[i].size;
1724 move++;
1725 }
1726 /* map index at which we will split */
1727 split = count - move;
1728 hash2 = map[split].hash;
1729 continued = hash2 == map[split - 1].hash;
1730 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1731 (unsigned long)dx_get_block(frame->at),
1732 hash2, split, count-split));
1733
1734 /* Fancy dance to stay within two buffers */
1735 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1736 blocksize);
1737 de = dx_pack_dirents(data1, blocksize);
1738 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1739 (char *) de,
1740 blocksize);
1741 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1742 (char *) de2,
1743 blocksize);
1744 if (csum_size) {
1745 t = EXT4_DIRENT_TAIL(data2, blocksize);
1746 initialize_dirent_tail(t, blocksize);
1747
1748 t = EXT4_DIRENT_TAIL(data1, blocksize);
1749 initialize_dirent_tail(t, blocksize);
1750 }
1751
1752 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1753 blocksize, 1));
1754 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1755 blocksize, 1));
1756
1757 /* Which block gets the new entry? */
1758 if (hinfo->hash >= hash2) {
1759 swap(*bh, bh2);
1760 de = de2;
1761 }
1762 dx_insert_block(frame, hash2 + continued, newblock);
1763 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1764 if (err)
1765 goto journal_error;
1766 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1767 if (err)
1768 goto journal_error;
1769 brelse(bh2);
1770 dxtrace(dx_show_index("frame", frame->entries));
1771 return de;
1772
1773journal_error:
1774 brelse(*bh);
1775 brelse(bh2);
1776 *bh = NULL;
1777 ext4_std_error(dir->i_sb, err);
1778 return ERR_PTR(err);
1779}
1780
1781int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1782 struct buffer_head *bh,
1783 void *buf, int buf_size,
1784 struct ext4_filename *fname,
1785 struct ext4_dir_entry_2 **dest_de)
1786{
1787 struct ext4_dir_entry_2 *de;
1788 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1789 int nlen, rlen;
1790 unsigned int offset = 0;
1791 char *top;
1792
1793 de = (struct ext4_dir_entry_2 *)buf;
1794 top = buf + buf_size - reclen;
1795 while ((char *) de <= top) {
1796 if (ext4_check_dir_entry(dir, NULL, de, bh,
1797 buf, buf_size, offset))
1798 return -EFSCORRUPTED;
1799 if (ext4_match(fname, de))
1800 return -EEXIST;
1801 nlen = EXT4_DIR_REC_LEN(de->name_len);
1802 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1803 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1804 break;
1805 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1806 offset += rlen;
1807 }
1808 if ((char *) de > top)
1809 return -ENOSPC;
1810
1811 *dest_de = de;
1812 return 0;
1813}
1814
1815void ext4_insert_dentry(struct inode *inode,
1816 struct ext4_dir_entry_2 *de,
1817 int buf_size,
1818 struct ext4_filename *fname)
1819{
1820
1821 int nlen, rlen;
1822
1823 nlen = EXT4_DIR_REC_LEN(de->name_len);
1824 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1825 if (de->inode) {
1826 struct ext4_dir_entry_2 *de1 =
1827 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1828 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1829 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1830 de = de1;
1831 }
1832 de->file_type = EXT4_FT_UNKNOWN;
1833 de->inode = cpu_to_le32(inode->i_ino);
1834 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1835 de->name_len = fname_len(fname);
1836 memcpy(de->name, fname_name(fname), fname_len(fname));
1837}
1838
1839/*
1840 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1841 * it points to a directory entry which is guaranteed to be large
1842 * enough for new directory entry. If de is NULL, then
1843 * add_dirent_to_buf will attempt search the directory block for
1844 * space. It will return -ENOSPC if no space is available, and -EIO
1845 * and -EEXIST if directory entry already exists.
1846 */
1847static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1848 struct inode *dir,
1849 struct inode *inode, struct ext4_dir_entry_2 *de,
1850 struct buffer_head *bh)
1851{
1852 unsigned int blocksize = dir->i_sb->s_blocksize;
1853 int csum_size = 0;
1854 int err;
1855
1856 if (ext4_has_metadata_csum(inode->i_sb))
1857 csum_size = sizeof(struct ext4_dir_entry_tail);
1858
1859 if (!de) {
1860 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1861 blocksize - csum_size, fname, &de);
1862 if (err)
1863 return err;
1864 }
1865 BUFFER_TRACE(bh, "get_write_access");
1866 err = ext4_journal_get_write_access(handle, bh);
1867 if (err) {
1868 ext4_std_error(dir->i_sb, err);
1869 return err;
1870 }
1871
1872 /* By now the buffer is marked for journaling */
1873 ext4_insert_dentry(inode, de, blocksize, fname);
1874
1875 /*
1876 * XXX shouldn't update any times until successful
1877 * completion of syscall, but too many callers depend
1878 * on this.
1879 *
1880 * XXX similarly, too many callers depend on
1881 * ext4_new_inode() setting the times, but error
1882 * recovery deletes the inode, so the worst that can
1883 * happen is that the times are slightly out of date
1884 * and/or different from the directory change time.
1885 */
1886 dir->i_mtime = dir->i_ctime = current_time(dir);
1887 ext4_update_dx_flag(dir);
1888 inode_inc_iversion(dir);
1889 ext4_mark_inode_dirty(handle, dir);
1890 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1891 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1892 if (err)
1893 ext4_std_error(dir->i_sb, err);
1894 return 0;
1895}
1896
1897/*
1898 * This converts a one block unindexed directory to a 3 block indexed
1899 * directory, and adds the dentry to the indexed directory.
1900 */
1901static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1902 struct inode *dir,
1903 struct inode *inode, struct buffer_head *bh)
1904{
1905 struct buffer_head *bh2;
1906 struct dx_root *root;
1907 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1908 struct dx_entry *entries;
1909 struct ext4_dir_entry_2 *de, *de2;
1910 struct ext4_dir_entry_tail *t;
1911 char *data1, *top;
1912 unsigned len;
1913 int retval;
1914 unsigned blocksize;
1915 ext4_lblk_t block;
1916 struct fake_dirent *fde;
1917 int csum_size = 0;
1918
1919 if (ext4_has_metadata_csum(inode->i_sb))
1920 csum_size = sizeof(struct ext4_dir_entry_tail);
1921
1922 blocksize = dir->i_sb->s_blocksize;
1923 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1924 BUFFER_TRACE(bh, "get_write_access");
1925 retval = ext4_journal_get_write_access(handle, bh);
1926 if (retval) {
1927 ext4_std_error(dir->i_sb, retval);
1928 brelse(bh);
1929 return retval;
1930 }
1931 root = (struct dx_root *) bh->b_data;
1932
1933 /* The 0th block becomes the root, move the dirents out */
1934 fde = &root->dotdot;
1935 de = (struct ext4_dir_entry_2 *)((char *)fde +
1936 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1937 if ((char *) de >= (((char *) root) + blocksize)) {
1938 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1939 brelse(bh);
1940 return -EFSCORRUPTED;
1941 }
1942 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1943
1944 /* Allocate new block for the 0th block's dirents */
1945 bh2 = ext4_append(handle, dir, &block);
1946 if (IS_ERR(bh2)) {
1947 brelse(bh);
1948 return PTR_ERR(bh2);
1949 }
1950 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1951 data1 = bh2->b_data;
1952
1953 memcpy (data1, de, len);
1954 de = (struct ext4_dir_entry_2 *) data1;
1955 top = data1 + len;
1956 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1957 de = de2;
1958 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1959 (char *) de,
1960 blocksize);
1961
1962 if (csum_size) {
1963 t = EXT4_DIRENT_TAIL(data1, blocksize);
1964 initialize_dirent_tail(t, blocksize);
1965 }
1966
1967 /* Initialize the root; the dot dirents already exist */
1968 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1969 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1970 blocksize);
1971 memset (&root->info, 0, sizeof(root->info));
1972 root->info.info_length = sizeof(root->info);
1973 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1974 entries = root->entries;
1975 dx_set_block(entries, 1);
1976 dx_set_count(entries, 1);
1977 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1978
1979 /* Initialize as for dx_probe */
1980 fname->hinfo.hash_version = root->info.hash_version;
1981 if (fname->hinfo.hash_version <= DX_HASH_TEA)
1982 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1983 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1984 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1985
1986 memset(frames, 0, sizeof(frames));
1987 frame = frames;
1988 frame->entries = entries;
1989 frame->at = entries;
1990 frame->bh = bh;
1991
1992 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1993 if (retval)
1994 goto out_frames;
1995 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1996 if (retval)
1997 goto out_frames;
1998
1999 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2000 if (IS_ERR(de)) {
2001 retval = PTR_ERR(de);
2002 goto out_frames;
2003 }
2004
2005 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2006out_frames:
2007 /*
2008 * Even if the block split failed, we have to properly write
2009 * out all the changes we did so far. Otherwise we can end up
2010 * with corrupted filesystem.
2011 */
2012 if (retval)
2013 ext4_mark_inode_dirty(handle, dir);
2014 dx_release(frames);
2015 brelse(bh2);
2016 return retval;
2017}
2018
2019/*
2020 * ext4_add_entry()
2021 *
2022 * adds a file entry to the specified directory, using the same
2023 * semantics as ext4_find_entry(). It returns NULL if it failed.
2024 *
2025 * NOTE!! The inode part of 'de' is left at 0 - which means you
2026 * may not sleep between calling this and putting something into
2027 * the entry, as someone else might have used it while you slept.
2028 */
2029static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2030 struct inode *inode)
2031{
2032 struct inode *dir = d_inode(dentry->d_parent);
2033 struct buffer_head *bh = NULL;
2034 struct ext4_dir_entry_2 *de;
2035 struct ext4_dir_entry_tail *t;
2036 struct super_block *sb;
2037 struct ext4_filename fname;
2038 int retval;
2039 int dx_fallback=0;
2040 unsigned blocksize;
2041 ext4_lblk_t block, blocks;
2042 int csum_size = 0;
2043
2044 if (ext4_has_metadata_csum(inode->i_sb))
2045 csum_size = sizeof(struct ext4_dir_entry_tail);
2046
2047 sb = dir->i_sb;
2048 blocksize = sb->s_blocksize;
2049 if (!dentry->d_name.len)
2050 return -EINVAL;
2051
2052 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2053 if (retval)
2054 return retval;
2055
2056 if (ext4_has_inline_data(dir)) {
2057 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2058 if (retval < 0)
2059 goto out;
2060 if (retval == 1) {
2061 retval = 0;
2062 goto out;
2063 }
2064 }
2065
2066 if (is_dx(dir)) {
2067 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2068 if (!retval || (retval != ERR_BAD_DX_DIR))
2069 goto out;
2070 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2071 dx_fallback++;
2072 ext4_mark_inode_dirty(handle, dir);
2073 }
2074 blocks = dir->i_size >> sb->s_blocksize_bits;
2075 for (block = 0; block < blocks; block++) {
2076 bh = ext4_read_dirblock(dir, block, DIRENT);
2077 if (IS_ERR(bh)) {
2078 retval = PTR_ERR(bh);
2079 bh = NULL;
2080 goto out;
2081 }
2082 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2083 NULL, bh);
2084 if (retval != -ENOSPC)
2085 goto out;
2086
2087 if (blocks == 1 && !dx_fallback &&
2088 ext4_has_feature_dir_index(sb)) {
2089 retval = make_indexed_dir(handle, &fname, dir,
2090 inode, bh);
2091 bh = NULL; /* make_indexed_dir releases bh */
2092 goto out;
2093 }
2094 brelse(bh);
2095 }
2096 bh = ext4_append(handle, dir, &block);
2097 if (IS_ERR(bh)) {
2098 retval = PTR_ERR(bh);
2099 bh = NULL;
2100 goto out;
2101 }
2102 de = (struct ext4_dir_entry_2 *) bh->b_data;
2103 de->inode = 0;
2104 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2105
2106 if (csum_size) {
2107 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2108 initialize_dirent_tail(t, blocksize);
2109 }
2110
2111 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2112out:
2113 ext4_fname_free_filename(&fname);
2114 brelse(bh);
2115 if (retval == 0)
2116 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2117 return retval;
2118}
2119
2120/*
2121 * Returns 0 for success, or a negative error value
2122 */
2123static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2124 struct inode *dir, struct inode *inode)
2125{
2126 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2127 struct dx_entry *entries, *at;
2128 struct buffer_head *bh;
2129 struct super_block *sb = dir->i_sb;
2130 struct ext4_dir_entry_2 *de;
2131 int restart;
2132 int err;
2133
2134again:
2135 restart = 0;
2136 frame = dx_probe(fname, dir, NULL, frames);
2137 if (IS_ERR(frame))
2138 return PTR_ERR(frame);
2139 entries = frame->entries;
2140 at = frame->at;
2141 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2142 if (IS_ERR(bh)) {
2143 err = PTR_ERR(bh);
2144 bh = NULL;
2145 goto cleanup;
2146 }
2147
2148 BUFFER_TRACE(bh, "get_write_access");
2149 err = ext4_journal_get_write_access(handle, bh);
2150 if (err)
2151 goto journal_error;
2152
2153 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2154 if (err != -ENOSPC)
2155 goto cleanup;
2156
2157 err = 0;
2158 /* Block full, should compress but for now just split */
2159 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2160 dx_get_count(entries), dx_get_limit(entries)));
2161 /* Need to split index? */
2162 if (dx_get_count(entries) == dx_get_limit(entries)) {
2163 ext4_lblk_t newblock;
2164 int levels = frame - frames + 1;
2165 unsigned int icount;
2166 int add_level = 1;
2167 struct dx_entry *entries2;
2168 struct dx_node *node2;
2169 struct buffer_head *bh2;
2170
2171 while (frame > frames) {
2172 if (dx_get_count((frame - 1)->entries) <
2173 dx_get_limit((frame - 1)->entries)) {
2174 add_level = 0;
2175 break;
2176 }
2177 frame--; /* split higher index block */
2178 at = frame->at;
2179 entries = frame->entries;
2180 restart = 1;
2181 }
2182 if (add_level && levels == ext4_dir_htree_level(sb)) {
2183 ext4_warning(sb, "Directory (ino: %lu) index full, "
2184 "reach max htree level :%d",
2185 dir->i_ino, levels);
2186 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2187 ext4_warning(sb, "Large directory feature is "
2188 "not enabled on this "
2189 "filesystem");
2190 }
2191 err = -ENOSPC;
2192 goto cleanup;
2193 }
2194 icount = dx_get_count(entries);
2195 bh2 = ext4_append(handle, dir, &newblock);
2196 if (IS_ERR(bh2)) {
2197 err = PTR_ERR(bh2);
2198 goto cleanup;
2199 }
2200 node2 = (struct dx_node *)(bh2->b_data);
2201 entries2 = node2->entries;
2202 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2203 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2204 sb->s_blocksize);
2205 BUFFER_TRACE(frame->bh, "get_write_access");
2206 err = ext4_journal_get_write_access(handle, frame->bh);
2207 if (err)
2208 goto journal_error;
2209 if (!add_level) {
2210 unsigned icount1 = icount/2, icount2 = icount - icount1;
2211 unsigned hash2 = dx_get_hash(entries + icount1);
2212 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2213 icount1, icount2));
2214
2215 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2216 err = ext4_journal_get_write_access(handle,
2217 (frame - 1)->bh);
2218 if (err)
2219 goto journal_error;
2220
2221 memcpy((char *) entries2, (char *) (entries + icount1),
2222 icount2 * sizeof(struct dx_entry));
2223 dx_set_count(entries, icount1);
2224 dx_set_count(entries2, icount2);
2225 dx_set_limit(entries2, dx_node_limit(dir));
2226
2227 /* Which index block gets the new entry? */
2228 if (at - entries >= icount1) {
2229 frame->at = at = at - entries - icount1 + entries2;
2230 frame->entries = entries = entries2;
2231 swap(frame->bh, bh2);
2232 }
2233 dx_insert_block((frame - 1), hash2, newblock);
2234 dxtrace(dx_show_index("node", frame->entries));
2235 dxtrace(dx_show_index("node",
2236 ((struct dx_node *) bh2->b_data)->entries));
2237 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2238 if (err)
2239 goto journal_error;
2240 brelse (bh2);
2241 err = ext4_handle_dirty_dx_node(handle, dir,
2242 (frame - 1)->bh);
2243 if (err)
2244 goto journal_error;
2245 if (restart) {
2246 err = ext4_handle_dirty_dx_node(handle, dir,
2247 frame->bh);
2248 goto journal_error;
2249 }
2250 } else {
2251 struct dx_root *dxroot;
2252 memcpy((char *) entries2, (char *) entries,
2253 icount * sizeof(struct dx_entry));
2254 dx_set_limit(entries2, dx_node_limit(dir));
2255
2256 /* Set up root */
2257 dx_set_count(entries, 1);
2258 dx_set_block(entries + 0, newblock);
2259 dxroot = (struct dx_root *)frames[0].bh->b_data;
2260 dxroot->info.indirect_levels += 1;
2261 dxtrace(printk(KERN_DEBUG
2262 "Creating %d level index...\n",
2263 info->indirect_levels));
2264 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2265 if (err)
2266 goto journal_error;
2267 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2268 brelse(bh2);
2269 restart = 1;
2270 goto journal_error;
2271 }
2272 }
2273 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2274 if (IS_ERR(de)) {
2275 err = PTR_ERR(de);
2276 goto cleanup;
2277 }
2278 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2279 goto cleanup;
2280
2281journal_error:
2282 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2283cleanup:
2284 brelse(bh);
2285 dx_release(frames);
2286 /* @restart is true means htree-path has been changed, we need to
2287 * repeat dx_probe() to find out valid htree-path
2288 */
2289 if (restart && err == 0)
2290 goto again;
2291 return err;
2292}
2293
2294/*
2295 * ext4_generic_delete_entry deletes a directory entry by merging it
2296 * with the previous entry
2297 */
2298int ext4_generic_delete_entry(handle_t *handle,
2299 struct inode *dir,
2300 struct ext4_dir_entry_2 *de_del,
2301 struct buffer_head *bh,
2302 void *entry_buf,
2303 int buf_size,
2304 int csum_size)
2305{
2306 struct ext4_dir_entry_2 *de, *pde;
2307 unsigned int blocksize = dir->i_sb->s_blocksize;
2308 int i;
2309
2310 i = 0;
2311 pde = NULL;
2312 de = (struct ext4_dir_entry_2 *)entry_buf;
2313 while (i < buf_size - csum_size) {
2314 if (ext4_check_dir_entry(dir, NULL, de, bh,
2315 bh->b_data, bh->b_size, i))
2316 return -EFSCORRUPTED;
2317 if (de == de_del) {
2318 if (pde)
2319 pde->rec_len = ext4_rec_len_to_disk(
2320 ext4_rec_len_from_disk(pde->rec_len,
2321 blocksize) +
2322 ext4_rec_len_from_disk(de->rec_len,
2323 blocksize),
2324 blocksize);
2325 else
2326 de->inode = 0;
2327 inode_inc_iversion(dir);
2328 return 0;
2329 }
2330 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2331 pde = de;
2332 de = ext4_next_entry(de, blocksize);
2333 }
2334 return -ENOENT;
2335}
2336
2337static int ext4_delete_entry(handle_t *handle,
2338 struct inode *dir,
2339 struct ext4_dir_entry_2 *de_del,
2340 struct buffer_head *bh)
2341{
2342 int err, csum_size = 0;
2343
2344 if (ext4_has_inline_data(dir)) {
2345 int has_inline_data = 1;
2346 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2347 &has_inline_data);
2348 if (has_inline_data)
2349 return err;
2350 }
2351
2352 if (ext4_has_metadata_csum(dir->i_sb))
2353 csum_size = sizeof(struct ext4_dir_entry_tail);
2354
2355 BUFFER_TRACE(bh, "get_write_access");
2356 err = ext4_journal_get_write_access(handle, bh);
2357 if (unlikely(err))
2358 goto out;
2359
2360 err = ext4_generic_delete_entry(handle, dir, de_del,
2361 bh, bh->b_data,
2362 dir->i_sb->s_blocksize, csum_size);
2363 if (err)
2364 goto out;
2365
2366 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2367 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2368 if (unlikely(err))
2369 goto out;
2370
2371 return 0;
2372out:
2373 if (err != -ENOENT)
2374 ext4_std_error(dir->i_sb, err);
2375 return err;
2376}
2377
2378/*
2379 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2380 * since this indicates that nlinks count was previously 1 to avoid overflowing
2381 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2382 * that subdirectory link counts are not being maintained accurately.
2383 *
2384 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2385 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2386 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2387 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2388 */
2389static void ext4_inc_count(handle_t *handle, struct inode *inode)
2390{
2391 inc_nlink(inode);
2392 if (is_dx(inode) &&
2393 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2394 set_nlink(inode, 1);
2395}
2396
2397/*
2398 * If a directory had nlink == 1, then we should let it be 1. This indicates
2399 * directory has >EXT4_LINK_MAX subdirs.
2400 */
2401static void ext4_dec_count(handle_t *handle, struct inode *inode)
2402{
2403 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2404 drop_nlink(inode);
2405}
2406
2407
2408static int ext4_add_nondir(handle_t *handle,
2409 struct dentry *dentry, struct inode *inode)
2410{
2411 int err = ext4_add_entry(handle, dentry, inode);
2412 if (!err) {
2413 ext4_mark_inode_dirty(handle, inode);
2414 d_instantiate_new(dentry, inode);
2415 return 0;
2416 }
2417 drop_nlink(inode);
2418 unlock_new_inode(inode);
2419 iput(inode);
2420 return err;
2421}
2422
2423/*
2424 * By the time this is called, we already have created
2425 * the directory cache entry for the new file, but it
2426 * is so far negative - it has no inode.
2427 *
2428 * If the create succeeds, we fill in the inode information
2429 * with d_instantiate().
2430 */
2431static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2432 bool excl)
2433{
2434 handle_t *handle;
2435 struct inode *inode;
2436 int err, credits, retries = 0;
2437
2438 err = dquot_initialize(dir);
2439 if (err)
2440 return err;
2441
2442 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2443 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2444retry:
2445 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2446 NULL, EXT4_HT_DIR, credits);
2447 handle = ext4_journal_current_handle();
2448 err = PTR_ERR(inode);
2449 if (!IS_ERR(inode)) {
2450 inode->i_op = &ext4_file_inode_operations;
2451 inode->i_fop = &ext4_file_operations;
2452 ext4_set_aops(inode);
2453 err = ext4_add_nondir(handle, dentry, inode);
2454 if (!err && IS_DIRSYNC(dir))
2455 ext4_handle_sync(handle);
2456 }
2457 if (handle)
2458 ext4_journal_stop(handle);
2459 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2460 goto retry;
2461 return err;
2462}
2463
2464static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2465 umode_t mode, dev_t rdev)
2466{
2467 handle_t *handle;
2468 struct inode *inode;
2469 int err, credits, retries = 0;
2470
2471 err = dquot_initialize(dir);
2472 if (err)
2473 return err;
2474
2475 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2476 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2477retry:
2478 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2479 NULL, EXT4_HT_DIR, credits);
2480 handle = ext4_journal_current_handle();
2481 err = PTR_ERR(inode);
2482 if (!IS_ERR(inode)) {
2483 init_special_inode(inode, inode->i_mode, rdev);
2484 inode->i_op = &ext4_special_inode_operations;
2485 err = ext4_add_nondir(handle, dentry, inode);
2486 if (!err && IS_DIRSYNC(dir))
2487 ext4_handle_sync(handle);
2488 }
2489 if (handle)
2490 ext4_journal_stop(handle);
2491 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2492 goto retry;
2493 return err;
2494}
2495
2496static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2497{
2498 handle_t *handle;
2499 struct inode *inode;
2500 int err, retries = 0;
2501
2502 err = dquot_initialize(dir);
2503 if (err)
2504 return err;
2505
2506retry:
2507 inode = ext4_new_inode_start_handle(dir, mode,
2508 NULL, 0, NULL,
2509 EXT4_HT_DIR,
2510 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2511 4 + EXT4_XATTR_TRANS_BLOCKS);
2512 handle = ext4_journal_current_handle();
2513 err = PTR_ERR(inode);
2514 if (!IS_ERR(inode)) {
2515 inode->i_op = &ext4_file_inode_operations;
2516 inode->i_fop = &ext4_file_operations;
2517 ext4_set_aops(inode);
2518 d_tmpfile(dentry, inode);
2519 err = ext4_orphan_add(handle, inode);
2520 if (err)
2521 goto err_unlock_inode;
2522 mark_inode_dirty(inode);
2523 unlock_new_inode(inode);
2524 }
2525 if (handle)
2526 ext4_journal_stop(handle);
2527 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2528 goto retry;
2529 return err;
2530err_unlock_inode:
2531 ext4_journal_stop(handle);
2532 unlock_new_inode(inode);
2533 return err;
2534}
2535
2536struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2537 struct ext4_dir_entry_2 *de,
2538 int blocksize, int csum_size,
2539 unsigned int parent_ino, int dotdot_real_len)
2540{
2541 de->inode = cpu_to_le32(inode->i_ino);
2542 de->name_len = 1;
2543 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2544 blocksize);
2545 strcpy(de->name, ".");
2546 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2547
2548 de = ext4_next_entry(de, blocksize);
2549 de->inode = cpu_to_le32(parent_ino);
2550 de->name_len = 2;
2551 if (!dotdot_real_len)
2552 de->rec_len = ext4_rec_len_to_disk(blocksize -
2553 (csum_size + EXT4_DIR_REC_LEN(1)),
2554 blocksize);
2555 else
2556 de->rec_len = ext4_rec_len_to_disk(
2557 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2558 strcpy(de->name, "..");
2559 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2560
2561 return ext4_next_entry(de, blocksize);
2562}
2563
2564static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2565 struct inode *inode)
2566{
2567 struct buffer_head *dir_block = NULL;
2568 struct ext4_dir_entry_2 *de;
2569 struct ext4_dir_entry_tail *t;
2570 ext4_lblk_t block = 0;
2571 unsigned int blocksize = dir->i_sb->s_blocksize;
2572 int csum_size = 0;
2573 int err;
2574
2575 if (ext4_has_metadata_csum(dir->i_sb))
2576 csum_size = sizeof(struct ext4_dir_entry_tail);
2577
2578 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2579 err = ext4_try_create_inline_dir(handle, dir, inode);
2580 if (err < 0 && err != -ENOSPC)
2581 goto out;
2582 if (!err)
2583 goto out;
2584 }
2585
2586 inode->i_size = 0;
2587 dir_block = ext4_append(handle, inode, &block);
2588 if (IS_ERR(dir_block))
2589 return PTR_ERR(dir_block);
2590 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2591 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2592 set_nlink(inode, 2);
2593 if (csum_size) {
2594 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2595 initialize_dirent_tail(t, blocksize);
2596 }
2597
2598 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2599 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2600 if (err)
2601 goto out;
2602 set_buffer_verified(dir_block);
2603out:
2604 brelse(dir_block);
2605 return err;
2606}
2607
2608static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2609{
2610 handle_t *handle;
2611 struct inode *inode;
2612 int err, credits, retries = 0;
2613
2614 if (EXT4_DIR_LINK_MAX(dir))
2615 return -EMLINK;
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, S_IFDIR | mode,
2625 &dentry->d_name,
2626 0, NULL, EXT4_HT_DIR, credits);
2627 handle = ext4_journal_current_handle();
2628 err = PTR_ERR(inode);
2629 if (IS_ERR(inode))
2630 goto out_stop;
2631
2632 inode->i_op = &ext4_dir_inode_operations;
2633 inode->i_fop = &ext4_dir_operations;
2634 err = ext4_init_new_dir(handle, dir, inode);
2635 if (err)
2636 goto out_clear_inode;
2637 err = ext4_mark_inode_dirty(handle, inode);
2638 if (!err)
2639 err = ext4_add_entry(handle, dentry, inode);
2640 if (err) {
2641out_clear_inode:
2642 clear_nlink(inode);
2643 unlock_new_inode(inode);
2644 ext4_mark_inode_dirty(handle, inode);
2645 iput(inode);
2646 goto out_stop;
2647 }
2648 ext4_inc_count(handle, dir);
2649 ext4_update_dx_flag(dir);
2650 err = ext4_mark_inode_dirty(handle, dir);
2651 if (err)
2652 goto out_clear_inode;
2653 d_instantiate_new(dentry, inode);
2654 if (IS_DIRSYNC(dir))
2655 ext4_handle_sync(handle);
2656
2657out_stop:
2658 if (handle)
2659 ext4_journal_stop(handle);
2660 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2661 goto retry;
2662 return err;
2663}
2664
2665/*
2666 * routine to check that the specified directory is empty (for rmdir)
2667 */
2668bool ext4_empty_dir(struct inode *inode)
2669{
2670 unsigned int offset;
2671 struct buffer_head *bh;
2672 struct ext4_dir_entry_2 *de, *de1;
2673 struct super_block *sb;
2674
2675 if (ext4_has_inline_data(inode)) {
2676 int has_inline_data = 1;
2677 int ret;
2678
2679 ret = empty_inline_dir(inode, &has_inline_data);
2680 if (has_inline_data)
2681 return ret;
2682 }
2683
2684 sb = inode->i_sb;
2685 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2686 EXT4_ERROR_INODE(inode, "invalid size");
2687 return true;
2688 }
2689 bh = ext4_read_dirblock(inode, 0, EITHER);
2690 if (IS_ERR(bh))
2691 return true;
2692
2693 de = (struct ext4_dir_entry_2 *) bh->b_data;
2694 de1 = ext4_next_entry(de, sb->s_blocksize);
2695 if (le32_to_cpu(de->inode) != inode->i_ino ||
2696 le32_to_cpu(de1->inode) == 0 ||
2697 strcmp(".", de->name) || strcmp("..", de1->name)) {
2698 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2699 brelse(bh);
2700 return true;
2701 }
2702 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2703 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2704 de = ext4_next_entry(de1, sb->s_blocksize);
2705 while (offset < inode->i_size) {
2706 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2707 unsigned int lblock;
2708 brelse(bh);
2709 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2710 bh = ext4_read_dirblock(inode, lblock, EITHER);
2711 if (IS_ERR(bh))
2712 return true;
2713 de = (struct ext4_dir_entry_2 *) bh->b_data;
2714 }
2715 if (ext4_check_dir_entry(inode, NULL, de, bh,
2716 bh->b_data, bh->b_size, offset)) {
2717 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2718 sb->s_blocksize);
2719 offset = (offset | (sb->s_blocksize - 1)) + 1;
2720 continue;
2721 }
2722 if (le32_to_cpu(de->inode)) {
2723 brelse(bh);
2724 return false;
2725 }
2726 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2727 de = ext4_next_entry(de, sb->s_blocksize);
2728 }
2729 brelse(bh);
2730 return true;
2731}
2732
2733/*
2734 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2735 * such inodes, starting at the superblock, in case we crash before the
2736 * file is closed/deleted, or in case the inode truncate spans multiple
2737 * transactions and the last transaction is not recovered after a crash.
2738 *
2739 * At filesystem recovery time, we walk this list deleting unlinked
2740 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2741 *
2742 * Orphan list manipulation functions must be called under i_mutex unless
2743 * we are just creating the inode or deleting it.
2744 */
2745int ext4_orphan_add(handle_t *handle, struct inode *inode)
2746{
2747 struct super_block *sb = inode->i_sb;
2748 struct ext4_sb_info *sbi = EXT4_SB(sb);
2749 struct ext4_iloc iloc;
2750 int err = 0, rc;
2751 bool dirty = false;
2752
2753 if (!sbi->s_journal || is_bad_inode(inode))
2754 return 0;
2755
2756 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2757 !inode_is_locked(inode));
2758 /*
2759 * Exit early if inode already is on orphan list. This is a big speedup
2760 * since we don't have to contend on the global s_orphan_lock.
2761 */
2762 if (!list_empty(&EXT4_I(inode)->i_orphan))
2763 return 0;
2764
2765 /*
2766 * Orphan handling is only valid for files with data blocks
2767 * being truncated, or files being unlinked. Note that we either
2768 * hold i_mutex, or the inode can not be referenced from outside,
2769 * so i_nlink should not be bumped due to race
2770 */
2771 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2772 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2773
2774 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2775 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2776 if (err)
2777 goto out;
2778
2779 err = ext4_reserve_inode_write(handle, inode, &iloc);
2780 if (err)
2781 goto out;
2782
2783 mutex_lock(&sbi->s_orphan_lock);
2784 /*
2785 * Due to previous errors inode may be already a part of on-disk
2786 * orphan list. If so skip on-disk list modification.
2787 */
2788 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2789 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2790 /* Insert this inode at the head of the on-disk orphan list */
2791 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2792 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2793 dirty = true;
2794 }
2795 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2796 mutex_unlock(&sbi->s_orphan_lock);
2797
2798 if (dirty) {
2799 err = ext4_handle_dirty_super(handle, sb);
2800 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2801 if (!err)
2802 err = rc;
2803 if (err) {
2804 /*
2805 * We have to remove inode from in-memory list if
2806 * addition to on disk orphan list failed. Stray orphan
2807 * list entries can cause panics at unmount time.
2808 */
2809 mutex_lock(&sbi->s_orphan_lock);
2810 list_del_init(&EXT4_I(inode)->i_orphan);
2811 mutex_unlock(&sbi->s_orphan_lock);
2812 }
2813 }
2814 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2815 jbd_debug(4, "orphan inode %lu will point to %d\n",
2816 inode->i_ino, NEXT_ORPHAN(inode));
2817out:
2818 ext4_std_error(sb, err);
2819 return err;
2820}
2821
2822/*
2823 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2824 * of such inodes stored on disk, because it is finally being cleaned up.
2825 */
2826int ext4_orphan_del(handle_t *handle, struct inode *inode)
2827{
2828 struct list_head *prev;
2829 struct ext4_inode_info *ei = EXT4_I(inode);
2830 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2831 __u32 ino_next;
2832 struct ext4_iloc iloc;
2833 int err = 0;
2834
2835 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2836 return 0;
2837
2838 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2839 !inode_is_locked(inode));
2840 /* Do this quick check before taking global s_orphan_lock. */
2841 if (list_empty(&ei->i_orphan))
2842 return 0;
2843
2844 if (handle) {
2845 /* Grab inode buffer early before taking global s_orphan_lock */
2846 err = ext4_reserve_inode_write(handle, inode, &iloc);
2847 }
2848
2849 mutex_lock(&sbi->s_orphan_lock);
2850 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2851
2852 prev = ei->i_orphan.prev;
2853 list_del_init(&ei->i_orphan);
2854
2855 /* If we're on an error path, we may not have a valid
2856 * transaction handle with which to update the orphan list on
2857 * disk, but we still need to remove the inode from the linked
2858 * list in memory. */
2859 if (!handle || err) {
2860 mutex_unlock(&sbi->s_orphan_lock);
2861 goto out_err;
2862 }
2863
2864 ino_next = NEXT_ORPHAN(inode);
2865 if (prev == &sbi->s_orphan) {
2866 jbd_debug(4, "superblock will point to %u\n", ino_next);
2867 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2868 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2869 if (err) {
2870 mutex_unlock(&sbi->s_orphan_lock);
2871 goto out_brelse;
2872 }
2873 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2874 mutex_unlock(&sbi->s_orphan_lock);
2875 err = ext4_handle_dirty_super(handle, inode->i_sb);
2876 } else {
2877 struct ext4_iloc iloc2;
2878 struct inode *i_prev =
2879 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2880
2881 jbd_debug(4, "orphan inode %lu will point to %u\n",
2882 i_prev->i_ino, ino_next);
2883 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2884 if (err) {
2885 mutex_unlock(&sbi->s_orphan_lock);
2886 goto out_brelse;
2887 }
2888 NEXT_ORPHAN(i_prev) = ino_next;
2889 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2890 mutex_unlock(&sbi->s_orphan_lock);
2891 }
2892 if (err)
2893 goto out_brelse;
2894 NEXT_ORPHAN(inode) = 0;
2895 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2896out_err:
2897 ext4_std_error(inode->i_sb, err);
2898 return err;
2899
2900out_brelse:
2901 brelse(iloc.bh);
2902 goto out_err;
2903}
2904
2905static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2906{
2907 int retval;
2908 struct inode *inode;
2909 struct buffer_head *bh;
2910 struct ext4_dir_entry_2 *de;
2911 handle_t *handle = NULL;
2912
2913 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2914 return -EIO;
2915
2916 /* Initialize quotas before so that eventual writes go in
2917 * separate transaction */
2918 retval = dquot_initialize(dir);
2919 if (retval)
2920 return retval;
2921 retval = dquot_initialize(d_inode(dentry));
2922 if (retval)
2923 return retval;
2924
2925 retval = -ENOENT;
2926 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2927 if (IS_ERR(bh))
2928 return PTR_ERR(bh);
2929 if (!bh)
2930 goto end_rmdir;
2931
2932 inode = d_inode(dentry);
2933
2934 retval = -EFSCORRUPTED;
2935 if (le32_to_cpu(de->inode) != inode->i_ino)
2936 goto end_rmdir;
2937
2938 retval = -ENOTEMPTY;
2939 if (!ext4_empty_dir(inode))
2940 goto end_rmdir;
2941
2942 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2943 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2944 if (IS_ERR(handle)) {
2945 retval = PTR_ERR(handle);
2946 handle = NULL;
2947 goto end_rmdir;
2948 }
2949
2950 if (IS_DIRSYNC(dir))
2951 ext4_handle_sync(handle);
2952
2953 retval = ext4_delete_entry(handle, dir, de, bh);
2954 if (retval)
2955 goto end_rmdir;
2956 if (!EXT4_DIR_LINK_EMPTY(inode))
2957 ext4_warning_inode(inode,
2958 "empty directory '%.*s' has too many links (%u)",
2959 dentry->d_name.len, dentry->d_name.name,
2960 inode->i_nlink);
2961 inode_inc_iversion(inode);
2962 clear_nlink(inode);
2963 /* There's no need to set i_disksize: the fact that i_nlink is
2964 * zero will ensure that the right thing happens during any
2965 * recovery. */
2966 inode->i_size = 0;
2967 ext4_orphan_add(handle, inode);
2968 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2969 ext4_mark_inode_dirty(handle, inode);
2970 ext4_dec_count(handle, dir);
2971 ext4_update_dx_flag(dir);
2972 ext4_mark_inode_dirty(handle, dir);
2973
2974end_rmdir:
2975 brelse(bh);
2976 if (handle)
2977 ext4_journal_stop(handle);
2978 return retval;
2979}
2980
2981static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2982{
2983 int retval;
2984 struct inode *inode;
2985 struct buffer_head *bh;
2986 struct ext4_dir_entry_2 *de;
2987 handle_t *handle = NULL;
2988
2989 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2990 return -EIO;
2991
2992 trace_ext4_unlink_enter(dir, dentry);
2993 /* Initialize quotas before so that eventual writes go
2994 * in separate transaction */
2995 retval = dquot_initialize(dir);
2996 if (retval)
2997 return retval;
2998 retval = dquot_initialize(d_inode(dentry));
2999 if (retval)
3000 return retval;
3001
3002 retval = -ENOENT;
3003 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3004 if (IS_ERR(bh))
3005 return PTR_ERR(bh);
3006 if (!bh)
3007 goto end_unlink;
3008
3009 inode = d_inode(dentry);
3010
3011 retval = -EFSCORRUPTED;
3012 if (le32_to_cpu(de->inode) != inode->i_ino)
3013 goto end_unlink;
3014
3015 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3016 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3017 if (IS_ERR(handle)) {
3018 retval = PTR_ERR(handle);
3019 handle = NULL;
3020 goto end_unlink;
3021 }
3022
3023 if (IS_DIRSYNC(dir))
3024 ext4_handle_sync(handle);
3025
3026 if (inode->i_nlink == 0) {
3027 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3028 dentry->d_name.len, dentry->d_name.name);
3029 set_nlink(inode, 1);
3030 }
3031 retval = ext4_delete_entry(handle, dir, de, bh);
3032 if (retval)
3033 goto end_unlink;
3034 dir->i_ctime = dir->i_mtime = current_time(dir);
3035 ext4_update_dx_flag(dir);
3036 ext4_mark_inode_dirty(handle, dir);
3037 drop_nlink(inode);
3038 if (!inode->i_nlink)
3039 ext4_orphan_add(handle, inode);
3040 inode->i_ctime = current_time(inode);
3041 ext4_mark_inode_dirty(handle, inode);
3042
3043end_unlink:
3044 brelse(bh);
3045 if (handle)
3046 ext4_journal_stop(handle);
3047 trace_ext4_unlink_exit(dentry, retval);
3048 return retval;
3049}
3050
3051static int ext4_symlink(struct inode *dir,
3052 struct dentry *dentry, const char *symname)
3053{
3054 handle_t *handle;
3055 struct inode *inode;
3056 int err, len = strlen(symname);
3057 int credits;
3058 struct fscrypt_str disk_link;
3059
3060 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3061 return -EIO;
3062
3063 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3064 &disk_link);
3065 if (err)
3066 return err;
3067
3068 err = dquot_initialize(dir);
3069 if (err)
3070 return err;
3071
3072 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3073 /*
3074 * For non-fast symlinks, we just allocate inode and put it on
3075 * orphan list in the first transaction => we need bitmap,
3076 * group descriptor, sb, inode block, quota blocks, and
3077 * possibly selinux xattr blocks.
3078 */
3079 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3080 EXT4_XATTR_TRANS_BLOCKS;
3081 } else {
3082 /*
3083 * Fast symlink. We have to add entry to directory
3084 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3085 * allocate new inode (bitmap, group descriptor, inode block,
3086 * quota blocks, sb is already counted in previous macros).
3087 */
3088 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3089 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3090 }
3091
3092 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3093 &dentry->d_name, 0, NULL,
3094 EXT4_HT_DIR, credits);
3095 handle = ext4_journal_current_handle();
3096 if (IS_ERR(inode)) {
3097 if (handle)
3098 ext4_journal_stop(handle);
3099 return PTR_ERR(inode);
3100 }
3101
3102 if (IS_ENCRYPTED(inode)) {
3103 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3104 if (err)
3105 goto err_drop_inode;
3106 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3107 }
3108
3109 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3110 if (!IS_ENCRYPTED(inode))
3111 inode->i_op = &ext4_symlink_inode_operations;
3112 inode_nohighmem(inode);
3113 ext4_set_aops(inode);
3114 /*
3115 * We cannot call page_symlink() with transaction started
3116 * because it calls into ext4_write_begin() which can wait
3117 * for transaction commit if we are running out of space
3118 * and thus we deadlock. So we have to stop transaction now
3119 * and restart it when symlink contents is written.
3120 *
3121 * To keep fs consistent in case of crash, we have to put inode
3122 * to orphan list in the mean time.
3123 */
3124 drop_nlink(inode);
3125 err = ext4_orphan_add(handle, inode);
3126 ext4_journal_stop(handle);
3127 handle = NULL;
3128 if (err)
3129 goto err_drop_inode;
3130 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3131 if (err)
3132 goto err_drop_inode;
3133 /*
3134 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3135 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3136 */
3137 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3138 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3139 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3140 if (IS_ERR(handle)) {
3141 err = PTR_ERR(handle);
3142 handle = NULL;
3143 goto err_drop_inode;
3144 }
3145 set_nlink(inode, 1);
3146 err = ext4_orphan_del(handle, inode);
3147 if (err)
3148 goto err_drop_inode;
3149 } else {
3150 /* clear the extent format for fast symlink */
3151 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3152 if (!IS_ENCRYPTED(inode)) {
3153 inode->i_op = &ext4_fast_symlink_inode_operations;
3154 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3155 }
3156 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3157 disk_link.len);
3158 inode->i_size = disk_link.len - 1;
3159 }
3160 EXT4_I(inode)->i_disksize = inode->i_size;
3161 err = ext4_add_nondir(handle, dentry, inode);
3162 if (!err && IS_DIRSYNC(dir))
3163 ext4_handle_sync(handle);
3164
3165 if (handle)
3166 ext4_journal_stop(handle);
3167 goto out_free_encrypted_link;
3168
3169err_drop_inode:
3170 if (handle)
3171 ext4_journal_stop(handle);
3172 clear_nlink(inode);
3173 unlock_new_inode(inode);
3174 iput(inode);
3175out_free_encrypted_link:
3176 if (disk_link.name != (unsigned char *)symname)
3177 kfree(disk_link.name);
3178 return err;
3179}
3180
3181static int ext4_link(struct dentry *old_dentry,
3182 struct inode *dir, struct dentry *dentry)
3183{
3184 handle_t *handle;
3185 struct inode *inode = d_inode(old_dentry);
3186 int err, retries = 0;
3187
3188 if (inode->i_nlink >= EXT4_LINK_MAX)
3189 return -EMLINK;
3190
3191 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3192 if (err)
3193 return err;
3194
3195 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3196 (!projid_eq(EXT4_I(dir)->i_projid,
3197 EXT4_I(old_dentry->d_inode)->i_projid)))
3198 return -EXDEV;
3199
3200 err = dquot_initialize(dir);
3201 if (err)
3202 return err;
3203
3204retry:
3205 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3206 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3207 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3208 if (IS_ERR(handle))
3209 return PTR_ERR(handle);
3210
3211 if (IS_DIRSYNC(dir))
3212 ext4_handle_sync(handle);
3213
3214 inode->i_ctime = current_time(inode);
3215 ext4_inc_count(handle, inode);
3216 ihold(inode);
3217
3218 err = ext4_add_entry(handle, dentry, inode);
3219 if (!err) {
3220 ext4_mark_inode_dirty(handle, inode);
3221 /* this can happen only for tmpfile being
3222 * linked the first time
3223 */
3224 if (inode->i_nlink == 1)
3225 ext4_orphan_del(handle, inode);
3226 d_instantiate(dentry, inode);
3227 } else {
3228 drop_nlink(inode);
3229 iput(inode);
3230 }
3231 ext4_journal_stop(handle);
3232 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3233 goto retry;
3234 return err;
3235}
3236
3237
3238/*
3239 * Try to find buffer head where contains the parent block.
3240 * It should be the inode block if it is inlined or the 1st block
3241 * if it is a normal dir.
3242 */
3243static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3244 struct inode *inode,
3245 int *retval,
3246 struct ext4_dir_entry_2 **parent_de,
3247 int *inlined)
3248{
3249 struct buffer_head *bh;
3250
3251 if (!ext4_has_inline_data(inode)) {
3252 bh = ext4_read_dirblock(inode, 0, EITHER);
3253 if (IS_ERR(bh)) {
3254 *retval = PTR_ERR(bh);
3255 return NULL;
3256 }
3257 *parent_de = ext4_next_entry(
3258 (struct ext4_dir_entry_2 *)bh->b_data,
3259 inode->i_sb->s_blocksize);
3260 return bh;
3261 }
3262
3263 *inlined = 1;
3264 return ext4_get_first_inline_block(inode, parent_de, retval);
3265}
3266
3267struct ext4_renament {
3268 struct inode *dir;
3269 struct dentry *dentry;
3270 struct inode *inode;
3271 bool is_dir;
3272 int dir_nlink_delta;
3273
3274 /* entry for "dentry" */
3275 struct buffer_head *bh;
3276 struct ext4_dir_entry_2 *de;
3277 int inlined;
3278
3279 /* entry for ".." in inode if it's a directory */
3280 struct buffer_head *dir_bh;
3281 struct ext4_dir_entry_2 *parent_de;
3282 int dir_inlined;
3283};
3284
3285static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3286{
3287 int retval;
3288
3289 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3290 &retval, &ent->parent_de,
3291 &ent->dir_inlined);
3292 if (!ent->dir_bh)
3293 return retval;
3294 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3295 return -EFSCORRUPTED;
3296 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3297 return ext4_journal_get_write_access(handle, ent->dir_bh);
3298}
3299
3300static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3301 unsigned dir_ino)
3302{
3303 int retval;
3304
3305 ent->parent_de->inode = cpu_to_le32(dir_ino);
3306 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3307 if (!ent->dir_inlined) {
3308 if (is_dx(ent->inode)) {
3309 retval = ext4_handle_dirty_dx_node(handle,
3310 ent->inode,
3311 ent->dir_bh);
3312 } else {
3313 retval = ext4_handle_dirty_dirent_node(handle,
3314 ent->inode,
3315 ent->dir_bh);
3316 }
3317 } else {
3318 retval = ext4_mark_inode_dirty(handle, ent->inode);
3319 }
3320 if (retval) {
3321 ext4_std_error(ent->dir->i_sb, retval);
3322 return retval;
3323 }
3324 return 0;
3325}
3326
3327static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3328 unsigned ino, unsigned file_type)
3329{
3330 int retval;
3331
3332 BUFFER_TRACE(ent->bh, "get write access");
3333 retval = ext4_journal_get_write_access(handle, ent->bh);
3334 if (retval)
3335 return retval;
3336 ent->de->inode = cpu_to_le32(ino);
3337 if (ext4_has_feature_filetype(ent->dir->i_sb))
3338 ent->de->file_type = file_type;
3339 inode_inc_iversion(ent->dir);
3340 ent->dir->i_ctime = ent->dir->i_mtime =
3341 current_time(ent->dir);
3342 ext4_mark_inode_dirty(handle, ent->dir);
3343 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3344 if (!ent->inlined) {
3345 retval = ext4_handle_dirty_dirent_node(handle,
3346 ent->dir, ent->bh);
3347 if (unlikely(retval)) {
3348 ext4_std_error(ent->dir->i_sb, retval);
3349 return retval;
3350 }
3351 }
3352 brelse(ent->bh);
3353 ent->bh = NULL;
3354
3355 return 0;
3356}
3357
3358static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3359 const struct qstr *d_name)
3360{
3361 int retval = -ENOENT;
3362 struct buffer_head *bh;
3363 struct ext4_dir_entry_2 *de;
3364
3365 bh = ext4_find_entry(dir, d_name, &de, NULL);
3366 if (IS_ERR(bh))
3367 return PTR_ERR(bh);
3368 if (bh) {
3369 retval = ext4_delete_entry(handle, dir, de, bh);
3370 brelse(bh);
3371 }
3372 return retval;
3373}
3374
3375static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3376 int force_reread)
3377{
3378 int retval;
3379 /*
3380 * ent->de could have moved from under us during htree split, so make
3381 * sure that we are deleting the right entry. We might also be pointing
3382 * to a stale entry in the unused part of ent->bh so just checking inum
3383 * and the name isn't enough.
3384 */
3385 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3386 ent->de->name_len != ent->dentry->d_name.len ||
3387 strncmp(ent->de->name, ent->dentry->d_name.name,
3388 ent->de->name_len) ||
3389 force_reread) {
3390 retval = ext4_find_delete_entry(handle, ent->dir,
3391 &ent->dentry->d_name);
3392 } else {
3393 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3394 if (retval == -ENOENT) {
3395 retval = ext4_find_delete_entry(handle, ent->dir,
3396 &ent->dentry->d_name);
3397 }
3398 }
3399
3400 if (retval) {
3401 ext4_warning_inode(ent->dir,
3402 "Deleting old file: nlink %d, error=%d",
3403 ent->dir->i_nlink, retval);
3404 }
3405}
3406
3407static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3408{
3409 if (ent->dir_nlink_delta) {
3410 if (ent->dir_nlink_delta == -1)
3411 ext4_dec_count(handle, ent->dir);
3412 else
3413 ext4_inc_count(handle, ent->dir);
3414 ext4_mark_inode_dirty(handle, ent->dir);
3415 }
3416}
3417
3418static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3419 int credits, handle_t **h)
3420{
3421 struct inode *wh;
3422 handle_t *handle;
3423 int retries = 0;
3424
3425 /*
3426 * for inode block, sb block, group summaries,
3427 * and inode bitmap
3428 */
3429 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3430 EXT4_XATTR_TRANS_BLOCKS + 4);
3431retry:
3432 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3433 &ent->dentry->d_name, 0, NULL,
3434 EXT4_HT_DIR, credits);
3435
3436 handle = ext4_journal_current_handle();
3437 if (IS_ERR(wh)) {
3438 if (handle)
3439 ext4_journal_stop(handle);
3440 if (PTR_ERR(wh) == -ENOSPC &&
3441 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3442 goto retry;
3443 } else {
3444 *h = handle;
3445 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3446 wh->i_op = &ext4_special_inode_operations;
3447 }
3448 return wh;
3449}
3450
3451/*
3452 * Anybody can rename anything with this: the permission checks are left to the
3453 * higher-level routines.
3454 *
3455 * n.b. old_{dentry,inode) refers to the source dentry/inode
3456 * while new_{dentry,inode) refers to the destination dentry/inode
3457 * This comes from rename(const char *oldpath, const char *newpath)
3458 */
3459static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3460 struct inode *new_dir, struct dentry *new_dentry,
3461 unsigned int flags)
3462{
3463 handle_t *handle = NULL;
3464 struct ext4_renament old = {
3465 .dir = old_dir,
3466 .dentry = old_dentry,
3467 .inode = d_inode(old_dentry),
3468 };
3469 struct ext4_renament new = {
3470 .dir = new_dir,
3471 .dentry = new_dentry,
3472 .inode = d_inode(new_dentry),
3473 };
3474 int force_reread;
3475 int retval;
3476 struct inode *whiteout = NULL;
3477 int credits;
3478 u8 old_file_type;
3479
3480 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3481 (!projid_eq(EXT4_I(new_dir)->i_projid,
3482 EXT4_I(old_dentry->d_inode)->i_projid)))
3483 return -EXDEV;
3484
3485 retval = dquot_initialize(old.dir);
3486 if (retval)
3487 return retval;
3488 retval = dquot_initialize(new.dir);
3489 if (retval)
3490 return retval;
3491
3492 /* Initialize quotas before so that eventual writes go
3493 * in separate transaction */
3494 if (new.inode) {
3495 retval = dquot_initialize(new.inode);
3496 if (retval)
3497 return retval;
3498 }
3499
3500 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3501 if (IS_ERR(old.bh))
3502 return PTR_ERR(old.bh);
3503 /*
3504 * Check for inode number is _not_ due to possible IO errors.
3505 * We might rmdir the source, keep it as pwd of some process
3506 * and merrily kill the link to whatever was created under the
3507 * same name. Goodbye sticky bit ;-<
3508 */
3509 retval = -ENOENT;
3510 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3511 goto end_rename;
3512
3513 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3514 &new.de, &new.inlined);
3515 if (IS_ERR(new.bh)) {
3516 retval = PTR_ERR(new.bh);
3517 new.bh = NULL;
3518 goto end_rename;
3519 }
3520 if (new.bh) {
3521 if (!new.inode) {
3522 brelse(new.bh);
3523 new.bh = NULL;
3524 }
3525 }
3526 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3527 ext4_alloc_da_blocks(old.inode);
3528
3529 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3530 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3531 if (!(flags & RENAME_WHITEOUT)) {
3532 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3533 if (IS_ERR(handle)) {
3534 retval = PTR_ERR(handle);
3535 handle = NULL;
3536 goto end_rename;
3537 }
3538 } else {
3539 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3540 if (IS_ERR(whiteout)) {
3541 retval = PTR_ERR(whiteout);
3542 whiteout = NULL;
3543 goto end_rename;
3544 }
3545 }
3546
3547 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3548 ext4_handle_sync(handle);
3549
3550 if (S_ISDIR(old.inode->i_mode)) {
3551 if (new.inode) {
3552 retval = -ENOTEMPTY;
3553 if (!ext4_empty_dir(new.inode))
3554 goto end_rename;
3555 } else {
3556 retval = -EMLINK;
3557 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3558 goto end_rename;
3559 }
3560 retval = ext4_rename_dir_prepare(handle, &old);
3561 if (retval)
3562 goto end_rename;
3563 }
3564 /*
3565 * If we're renaming a file within an inline_data dir and adding or
3566 * setting the new dirent causes a conversion from inline_data to
3567 * extents/blockmap, we need to force the dirent delete code to
3568 * re-read the directory, or else we end up trying to delete a dirent
3569 * from what is now the extent tree root (or a block map).
3570 */
3571 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3572 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3573
3574 old_file_type = old.de->file_type;
3575 if (whiteout) {
3576 /*
3577 * Do this before adding a new entry, so the old entry is sure
3578 * to be still pointing to the valid old entry.
3579 */
3580 retval = ext4_setent(handle, &old, whiteout->i_ino,
3581 EXT4_FT_CHRDEV);
3582 if (retval)
3583 goto end_rename;
3584 ext4_mark_inode_dirty(handle, whiteout);
3585 }
3586 if (!new.bh) {
3587 retval = ext4_add_entry(handle, new.dentry, old.inode);
3588 if (retval)
3589 goto end_rename;
3590 } else {
3591 retval = ext4_setent(handle, &new,
3592 old.inode->i_ino, old_file_type);
3593 if (retval)
3594 goto end_rename;
3595 }
3596 if (force_reread)
3597 force_reread = !ext4_test_inode_flag(new.dir,
3598 EXT4_INODE_INLINE_DATA);
3599
3600 /*
3601 * Like most other Unix systems, set the ctime for inodes on a
3602 * rename.
3603 */
3604 old.inode->i_ctime = current_time(old.inode);
3605 ext4_mark_inode_dirty(handle, old.inode);
3606
3607 if (!whiteout) {
3608 /*
3609 * ok, that's it
3610 */
3611 ext4_rename_delete(handle, &old, force_reread);
3612 }
3613
3614 if (new.inode) {
3615 ext4_dec_count(handle, new.inode);
3616 new.inode->i_ctime = current_time(new.inode);
3617 }
3618 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3619 ext4_update_dx_flag(old.dir);
3620 if (old.dir_bh) {
3621 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3622 if (retval)
3623 goto end_rename;
3624
3625 ext4_dec_count(handle, old.dir);
3626 if (new.inode) {
3627 /* checked ext4_empty_dir above, can't have another
3628 * parent, ext4_dec_count() won't work for many-linked
3629 * dirs */
3630 clear_nlink(new.inode);
3631 } else {
3632 ext4_inc_count(handle, new.dir);
3633 ext4_update_dx_flag(new.dir);
3634 ext4_mark_inode_dirty(handle, new.dir);
3635 }
3636 }
3637 ext4_mark_inode_dirty(handle, old.dir);
3638 if (new.inode) {
3639 ext4_mark_inode_dirty(handle, new.inode);
3640 if (!new.inode->i_nlink)
3641 ext4_orphan_add(handle, new.inode);
3642 }
3643 retval = 0;
3644
3645end_rename:
3646 brelse(old.dir_bh);
3647 brelse(old.bh);
3648 brelse(new.bh);
3649 if (whiteout) {
3650 if (retval)
3651 drop_nlink(whiteout);
3652 unlock_new_inode(whiteout);
3653 iput(whiteout);
3654 }
3655 if (handle)
3656 ext4_journal_stop(handle);
3657 return retval;
3658}
3659
3660static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3661 struct inode *new_dir, struct dentry *new_dentry)
3662{
3663 handle_t *handle = NULL;
3664 struct ext4_renament old = {
3665 .dir = old_dir,
3666 .dentry = old_dentry,
3667 .inode = d_inode(old_dentry),
3668 };
3669 struct ext4_renament new = {
3670 .dir = new_dir,
3671 .dentry = new_dentry,
3672 .inode = d_inode(new_dentry),
3673 };
3674 u8 new_file_type;
3675 int retval;
3676 struct timespec ctime;
3677
3678 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3679 !projid_eq(EXT4_I(new_dir)->i_projid,
3680 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3681 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3682 !projid_eq(EXT4_I(old_dir)->i_projid,
3683 EXT4_I(new_dentry->d_inode)->i_projid)))
3684 return -EXDEV;
3685
3686 retval = dquot_initialize(old.dir);
3687 if (retval)
3688 return retval;
3689 retval = dquot_initialize(new.dir);
3690 if (retval)
3691 return retval;
3692
3693 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3694 &old.de, &old.inlined);
3695 if (IS_ERR(old.bh))
3696 return PTR_ERR(old.bh);
3697 /*
3698 * Check for inode number is _not_ due to possible IO errors.
3699 * We might rmdir the source, keep it as pwd of some process
3700 * and merrily kill the link to whatever was created under the
3701 * same name. Goodbye sticky bit ;-<
3702 */
3703 retval = -ENOENT;
3704 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3705 goto end_rename;
3706
3707 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3708 &new.de, &new.inlined);
3709 if (IS_ERR(new.bh)) {
3710 retval = PTR_ERR(new.bh);
3711 new.bh = NULL;
3712 goto end_rename;
3713 }
3714
3715 /* RENAME_EXCHANGE case: old *and* new must both exist */
3716 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3717 goto end_rename;
3718
3719 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3720 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3721 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3722 if (IS_ERR(handle)) {
3723 retval = PTR_ERR(handle);
3724 handle = NULL;
3725 goto end_rename;
3726 }
3727
3728 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3729 ext4_handle_sync(handle);
3730
3731 if (S_ISDIR(old.inode->i_mode)) {
3732 old.is_dir = true;
3733 retval = ext4_rename_dir_prepare(handle, &old);
3734 if (retval)
3735 goto end_rename;
3736 }
3737 if (S_ISDIR(new.inode->i_mode)) {
3738 new.is_dir = true;
3739 retval = ext4_rename_dir_prepare(handle, &new);
3740 if (retval)
3741 goto end_rename;
3742 }
3743
3744 /*
3745 * Other than the special case of overwriting a directory, parents'
3746 * nlink only needs to be modified if this is a cross directory rename.
3747 */
3748 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3749 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3750 new.dir_nlink_delta = -old.dir_nlink_delta;
3751 retval = -EMLINK;
3752 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3753 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3754 goto end_rename;
3755 }
3756
3757 new_file_type = new.de->file_type;
3758 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3759 if (retval)
3760 goto end_rename;
3761
3762 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3763 if (retval)
3764 goto end_rename;
3765
3766 /*
3767 * Like most other Unix systems, set the ctime for inodes on a
3768 * rename.
3769 */
3770 ctime = current_time(old.inode);
3771 old.inode->i_ctime = ctime;
3772 new.inode->i_ctime = ctime;
3773 ext4_mark_inode_dirty(handle, old.inode);
3774 ext4_mark_inode_dirty(handle, new.inode);
3775
3776 if (old.dir_bh) {
3777 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3778 if (retval)
3779 goto end_rename;
3780 }
3781 if (new.dir_bh) {
3782 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3783 if (retval)
3784 goto end_rename;
3785 }
3786 ext4_update_dir_count(handle, &old);
3787 ext4_update_dir_count(handle, &new);
3788 retval = 0;
3789
3790end_rename:
3791 brelse(old.dir_bh);
3792 brelse(new.dir_bh);
3793 brelse(old.bh);
3794 brelse(new.bh);
3795 if (handle)
3796 ext4_journal_stop(handle);
3797 return retval;
3798}
3799
3800static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3801 struct inode *new_dir, struct dentry *new_dentry,
3802 unsigned int flags)
3803{
3804 int err;
3805
3806 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3807 return -EIO;
3808
3809 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3810 return -EINVAL;
3811
3812 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3813 flags);
3814 if (err)
3815 return err;
3816
3817 if (flags & RENAME_EXCHANGE) {
3818 return ext4_cross_rename(old_dir, old_dentry,
3819 new_dir, new_dentry);
3820 }
3821
3822 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3823}
3824
3825/*
3826 * directories can handle most operations...
3827 */
3828const struct inode_operations ext4_dir_inode_operations = {
3829 .create = ext4_create,
3830 .lookup = ext4_lookup,
3831 .link = ext4_link,
3832 .unlink = ext4_unlink,
3833 .symlink = ext4_symlink,
3834 .mkdir = ext4_mkdir,
3835 .rmdir = ext4_rmdir,
3836 .mknod = ext4_mknod,
3837 .tmpfile = ext4_tmpfile,
3838 .rename = ext4_rename2,
3839 .setattr = ext4_setattr,
3840 .getattr = ext4_getattr,
3841 .listxattr = ext4_listxattr,
3842 .get_acl = ext4_get_acl,
3843 .set_acl = ext4_set_acl,
3844 .fiemap = ext4_fiemap,
3845};
3846
3847const struct inode_operations ext4_special_inode_operations = {
3848 .setattr = ext4_setattr,
3849 .getattr = ext4_getattr,
3850 .listxattr = ext4_listxattr,
3851 .get_acl = ext4_get_acl,
3852 .set_acl = ext4_set_acl,
3853};