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/*
2 * linux/fs/ext4/namei.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
24 * Theodore Ts'o, 2002
25 */
26
27#include <linux/fs.h>
28#include <linux/pagemap.h>
29#include <linux/time.h>
30#include <linux/fcntl.h>
31#include <linux/stat.h>
32#include <linux/string.h>
33#include <linux/quotaops.h>
34#include <linux/buffer_head.h>
35#include <linux/bio.h>
36#include "ext4.h"
37#include "ext4_jbd2.h"
38
39#include "xattr.h"
40#include "acl.h"
41
42#include <trace/events/ext4.h>
43/*
44 * define how far ahead to read directories while searching them.
45 */
46#define NAMEI_RA_CHUNKS 2
47#define NAMEI_RA_BLOCKS 4
48#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49
50static struct buffer_head *ext4_append(handle_t *handle,
51 struct inode *inode,
52 ext4_lblk_t *block)
53{
54 struct buffer_head *bh;
55 int err;
56
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
61
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63
64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
65 if (IS_ERR(bh))
66 return bh;
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
70 err = ext4_journal_get_write_access(handle, bh);
71 if (err) {
72 brelse(bh);
73 ext4_std_error(inode->i_sb, err);
74 return ERR_PTR(err);
75 }
76 return bh;
77}
78
79static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
81
82typedef enum {
83 EITHER, INDEX, DIRENT
84} dirblock_type_t;
85
86#define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
88
89static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 ext4_lblk_t block,
91 dirblock_type_t type,
92 const char *func,
93 unsigned int line)
94{
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
97 int is_dx_block = 0;
98
99 bh = ext4_bread(NULL, inode, block, 0);
100 if (IS_ERR(bh)) {
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
106
107 return bh;
108 }
109 if (!bh) {
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED);
113 }
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
116 if (is_dx(inode)) {
117 if (block == 0)
118 is_dx_block = 1;
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
122 is_dx_block = 1;
123 }
124 if (!is_dx_block && type == INDEX) {
125 ext4_error_inode(inode, func, line, block,
126 "directory leaf block found instead of index block");
127 return ERR_PTR(-EFSCORRUPTED);
128 }
129 if (!ext4_has_metadata_csum(inode->i_sb) ||
130 buffer_verified(bh))
131 return bh;
132
133 /*
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
137 */
138 if (is_dx_block && type == INDEX) {
139 if (ext4_dx_csum_verify(inode, dirent))
140 set_buffer_verified(bh);
141 else {
142 ext4_error_inode(inode, func, line, block,
143 "Directory index failed checksum");
144 brelse(bh);
145 return ERR_PTR(-EFSBADCRC);
146 }
147 }
148 if (!is_dx_block) {
149 if (ext4_dirent_csum_verify(inode, dirent))
150 set_buffer_verified(bh);
151 else {
152 ext4_error_inode(inode, func, line, block,
153 "Directory block failed checksum");
154 brelse(bh);
155 return ERR_PTR(-EFSBADCRC);
156 }
157 }
158 return bh;
159}
160
161#ifndef assert
162#define assert(test) J_ASSERT(test)
163#endif
164
165#ifdef DX_DEBUG
166#define dxtrace(command) command
167#else
168#define dxtrace(command)
169#endif
170
171struct fake_dirent
172{
173 __le32 inode;
174 __le16 rec_len;
175 u8 name_len;
176 u8 file_type;
177};
178
179struct dx_countlimit
180{
181 __le16 limit;
182 __le16 count;
183};
184
185struct dx_entry
186{
187 __le32 hash;
188 __le32 block;
189};
190
191/*
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
195 */
196
197struct dx_root
198{
199 struct fake_dirent dot;
200 char dot_name[4];
201 struct fake_dirent dotdot;
202 char dotdot_name[4];
203 struct dx_root_info
204 {
205 __le32 reserved_zero;
206 u8 hash_version;
207 u8 info_length; /* 8 */
208 u8 indirect_levels;
209 u8 unused_flags;
210 }
211 info;
212 struct dx_entry entries[0];
213};
214
215struct dx_node
216{
217 struct fake_dirent fake;
218 struct dx_entry entries[0];
219};
220
221
222struct dx_frame
223{
224 struct buffer_head *bh;
225 struct dx_entry *entries;
226 struct dx_entry *at;
227};
228
229struct dx_map_entry
230{
231 u32 hash;
232 u16 offs;
233 u16 size;
234};
235
236/*
237 * This goes at the end of each htree block.
238 */
239struct dx_tail {
240 u32 dt_reserved;
241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
242};
243
244static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246static inline unsigned dx_get_hash(struct dx_entry *entry);
247static void dx_set_hash(struct dx_entry *entry, unsigned value);
248static unsigned dx_get_count(struct dx_entry *entries);
249static unsigned dx_get_limit(struct dx_entry *entries);
250static void dx_set_count(struct dx_entry *entries, unsigned value);
251static void dx_set_limit(struct dx_entry *entries, unsigned value);
252static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253static unsigned dx_node_limit(struct inode *dir);
254static struct dx_frame *dx_probe(struct ext4_filename *fname,
255 struct inode *dir,
256 struct dx_hash_info *hinfo,
257 struct dx_frame *frame);
258static void dx_release(struct dx_frame *frames);
259static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 unsigned blocksize, struct dx_hash_info *hinfo,
261 struct dx_map_entry map[]);
262static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 struct dx_map_entry *offsets, int count, unsigned blocksize);
265static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266static void dx_insert_block(struct dx_frame *frame,
267 u32 hash, ext4_lblk_t block);
268static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 struct dx_frame *frame,
270 struct dx_frame *frames,
271 __u32 *start_hash);
272static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 struct ext4_filename *fname,
274 struct ext4_dir_entry_2 **res_dir);
275static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 struct inode *dir, struct inode *inode);
277
278/* checksumming functions */
279void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 unsigned int blocksize)
281{
282 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 t->det_rec_len = ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail), blocksize);
285 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
286}
287
288/* Walk through a dirent block to find a checksum "dirent" at the tail */
289static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 struct ext4_dir_entry *de)
291{
292 struct ext4_dir_entry_tail *t;
293
294#ifdef PARANOID
295 struct ext4_dir_entry *d, *top;
296
297 d = de;
298 top = (struct ext4_dir_entry *)(((void *)de) +
299 (EXT4_BLOCK_SIZE(inode->i_sb) -
300 sizeof(struct ext4_dir_entry_tail)));
301 while (d < top && d->rec_len)
302 d = (struct ext4_dir_entry *)(((void *)d) +
303 le16_to_cpu(d->rec_len));
304
305 if (d != top)
306 return NULL;
307
308 t = (struct ext4_dir_entry_tail *)d;
309#else
310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
311#endif
312
313 if (t->det_reserved_zero1 ||
314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 t->det_reserved_zero2 ||
316 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
317 return NULL;
318
319 return t;
320}
321
322static __le32 ext4_dirent_csum(struct inode *inode,
323 struct ext4_dir_entry *dirent, int size)
324{
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct ext4_inode_info *ei = EXT4_I(inode);
327 __u32 csum;
328
329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 return cpu_to_le32(csum);
331}
332
333#define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
335
336static void __warn_no_space_for_csum(struct inode *inode, const char *func,
337 unsigned int line)
338{
339 __ext4_warning_inode(inode, func, line,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
341}
342
343int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
344{
345 struct ext4_dir_entry_tail *t;
346
347 if (!ext4_has_metadata_csum(inode->i_sb))
348 return 1;
349
350 t = get_dirent_tail(inode, dirent);
351 if (!t) {
352 warn_no_space_for_csum(inode);
353 return 0;
354 }
355
356 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 (void *)t - (void *)dirent))
358 return 0;
359
360 return 1;
361}
362
363static void ext4_dirent_csum_set(struct inode *inode,
364 struct ext4_dir_entry *dirent)
365{
366 struct ext4_dir_entry_tail *t;
367
368 if (!ext4_has_metadata_csum(inode->i_sb))
369 return;
370
371 t = get_dirent_tail(inode, dirent);
372 if (!t) {
373 warn_no_space_for_csum(inode);
374 return;
375 }
376
377 t->det_checksum = ext4_dirent_csum(inode, dirent,
378 (void *)t - (void *)dirent);
379}
380
381int ext4_handle_dirty_dirent_node(handle_t *handle,
382 struct inode *inode,
383 struct buffer_head *bh)
384{
385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 return ext4_handle_dirty_metadata(handle, inode, bh);
387}
388
389static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 struct ext4_dir_entry *dirent,
391 int *offset)
392{
393 struct ext4_dir_entry *dp;
394 struct dx_root_info *root;
395 int count_offset;
396
397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
398 count_offset = 8;
399 else if (le16_to_cpu(dirent->rec_len) == 12) {
400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 if (le16_to_cpu(dp->rec_len) !=
402 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
403 return NULL;
404 root = (struct dx_root_info *)(((void *)dp + 12));
405 if (root->reserved_zero ||
406 root->info_length != sizeof(struct dx_root_info))
407 return NULL;
408 count_offset = 32;
409 } else
410 return NULL;
411
412 if (offset)
413 *offset = count_offset;
414 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
415}
416
417static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 int count_offset, int count, struct dx_tail *t)
419{
420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 struct ext4_inode_info *ei = EXT4_I(inode);
422 __u32 csum;
423 int size;
424 __u32 dummy_csum = 0;
425 int offset = offsetof(struct dx_tail, dt_checksum);
426
427 size = count_offset + (count * sizeof(struct dx_entry));
428 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
429 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
430 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
431
432 return cpu_to_le32(csum);
433}
434
435static int ext4_dx_csum_verify(struct inode *inode,
436 struct ext4_dir_entry *dirent)
437{
438 struct dx_countlimit *c;
439 struct dx_tail *t;
440 int count_offset, limit, count;
441
442 if (!ext4_has_metadata_csum(inode->i_sb))
443 return 1;
444
445 c = get_dx_countlimit(inode, dirent, &count_offset);
446 if (!c) {
447 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
448 return 0;
449 }
450 limit = le16_to_cpu(c->limit);
451 count = le16_to_cpu(c->count);
452 if (count_offset + (limit * sizeof(struct dx_entry)) >
453 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
454 warn_no_space_for_csum(inode);
455 return 0;
456 }
457 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
458
459 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
460 count, t))
461 return 0;
462 return 1;
463}
464
465static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
466{
467 struct dx_countlimit *c;
468 struct dx_tail *t;
469 int count_offset, limit, count;
470
471 if (!ext4_has_metadata_csum(inode->i_sb))
472 return;
473
474 c = get_dx_countlimit(inode, dirent, &count_offset);
475 if (!c) {
476 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
477 return;
478 }
479 limit = le16_to_cpu(c->limit);
480 count = le16_to_cpu(c->count);
481 if (count_offset + (limit * sizeof(struct dx_entry)) >
482 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483 warn_no_space_for_csum(inode);
484 return;
485 }
486 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
487
488 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
489}
490
491static inline int ext4_handle_dirty_dx_node(handle_t *handle,
492 struct inode *inode,
493 struct buffer_head *bh)
494{
495 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 return ext4_handle_dirty_metadata(handle, inode, bh);
497}
498
499/*
500 * p is at least 6 bytes before the end of page
501 */
502static inline struct ext4_dir_entry_2 *
503ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
504{
505 return (struct ext4_dir_entry_2 *)((char *)p +
506 ext4_rec_len_from_disk(p->rec_len, blocksize));
507}
508
509/*
510 * Future: use high four bits of block for coalesce-on-delete flags
511 * Mask them off for now.
512 */
513
514static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
515{
516 return le32_to_cpu(entry->block) & 0x00ffffff;
517}
518
519static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
520{
521 entry->block = cpu_to_le32(value);
522}
523
524static inline unsigned dx_get_hash(struct dx_entry *entry)
525{
526 return le32_to_cpu(entry->hash);
527}
528
529static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
530{
531 entry->hash = cpu_to_le32(value);
532}
533
534static inline unsigned dx_get_count(struct dx_entry *entries)
535{
536 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537}
538
539static inline unsigned dx_get_limit(struct dx_entry *entries)
540{
541 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542}
543
544static inline void dx_set_count(struct dx_entry *entries, unsigned value)
545{
546 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547}
548
549static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
550{
551 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552}
553
554static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
555{
556 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 EXT4_DIR_REC_LEN(2) - infosize;
558
559 if (ext4_has_metadata_csum(dir->i_sb))
560 entry_space -= sizeof(struct dx_tail);
561 return entry_space / sizeof(struct dx_entry);
562}
563
564static inline unsigned dx_node_limit(struct inode *dir)
565{
566 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
567
568 if (ext4_has_metadata_csum(dir->i_sb))
569 entry_space -= sizeof(struct dx_tail);
570 return entry_space / sizeof(struct dx_entry);
571}
572
573/*
574 * Debug
575 */
576#ifdef DX_DEBUG
577static void dx_show_index(char * label, struct dx_entry *entries)
578{
579 int i, n = dx_get_count (entries);
580 printk(KERN_DEBUG "%s index", label);
581 for (i = 0; i < n; i++) {
582 printk(KERN_CONT " %x->%lu",
583 i ? dx_get_hash(entries + i) : 0,
584 (unsigned long)dx_get_block(entries + i));
585 }
586 printk(KERN_CONT "\n");
587}
588
589struct stats
590{
591 unsigned names;
592 unsigned space;
593 unsigned bcount;
594};
595
596static struct stats dx_show_leaf(struct inode *dir,
597 struct dx_hash_info *hinfo,
598 struct ext4_dir_entry_2 *de,
599 int size, int show_names)
600{
601 unsigned names = 0, space = 0;
602 char *base = (char *) de;
603 struct dx_hash_info h = *hinfo;
604
605 printk("names: ");
606 while ((char *) de < base + size)
607 {
608 if (de->inode)
609 {
610 if (show_names)
611 {
612#ifdef CONFIG_EXT4_FS_ENCRYPTION
613 int len;
614 char *name;
615 struct fscrypt_str fname_crypto_str =
616 FSTR_INIT(NULL, 0);
617 int res = 0;
618
619 name = de->name;
620 len = de->name_len;
621 if (ext4_encrypted_inode(dir))
622 res = fscrypt_get_encryption_info(dir);
623 if (res) {
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
626 }
627 if (!fscrypt_has_encryption_key(dir)) {
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
630 de->name_len, &h);
631 printk("%*.s:(U)%x.%u ", len,
632 name, h.hash,
633 (unsigned) ((char *) de
634 - base));
635 } else {
636 struct fscrypt_str de_name =
637 FSTR_INIT(name, len);
638
639 /* Directory is encrypted */
640 res = fscrypt_fname_alloc_buffer(
641 dir, len,
642 &fname_crypto_str);
643 if (res)
644 printk(KERN_WARNING "Error "
645 "allocating crypto "
646 "buffer--skipping "
647 "crypto\n");
648 res = fscrypt_fname_disk_to_usr(dir,
649 0, 0, &de_name,
650 &fname_crypto_str);
651 if (res) {
652 printk(KERN_WARNING "Error "
653 "converting filename "
654 "from disk to usr"
655 "\n");
656 name = "??";
657 len = 2;
658 } else {
659 name = fname_crypto_str.name;
660 len = fname_crypto_str.len;
661 }
662 ext4fs_dirhash(de->name, de->name_len,
663 &h);
664 printk("%*.s:(E)%x.%u ", len, name,
665 h.hash, (unsigned) ((char *) de
666 - base));
667 fscrypt_fname_free_buffer(
668 &fname_crypto_str);
669 }
670#else
671 int len = de->name_len;
672 char *name = de->name;
673 ext4fs_dirhash(de->name, de->name_len, &h);
674 printk("%*.s:%x.%u ", len, name, h.hash,
675 (unsigned) ((char *) de - base));
676#endif
677 }
678 space += EXT4_DIR_REC_LEN(de->name_len);
679 names++;
680 }
681 de = ext4_next_entry(de, size);
682 }
683 printk(KERN_CONT "(%i)\n", names);
684 return (struct stats) { names, space, 1 };
685}
686
687struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
688 struct dx_entry *entries, int levels)
689{
690 unsigned blocksize = dir->i_sb->s_blocksize;
691 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
692 unsigned bcount = 0;
693 struct buffer_head *bh;
694 printk("%i indexed blocks...\n", count);
695 for (i = 0; i < count; i++, entries++)
696 {
697 ext4_lblk_t block = dx_get_block(entries);
698 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
699 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
700 struct stats stats;
701 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
702 bh = ext4_bread(NULL,dir, block, 0);
703 if (!bh || IS_ERR(bh))
704 continue;
705 stats = levels?
706 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
707 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
708 bh->b_data, blocksize, 0);
709 names += stats.names;
710 space += stats.space;
711 bcount += stats.bcount;
712 brelse(bh);
713 }
714 if (bcount)
715 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
716 levels ? "" : " ", names, space/bcount,
717 (space/bcount)*100/blocksize);
718 return (struct stats) { names, space, bcount};
719}
720#endif /* DX_DEBUG */
721
722/*
723 * Probe for a directory leaf block to search.
724 *
725 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
726 * error in the directory index, and the caller should fall back to
727 * searching the directory normally. The callers of dx_probe **MUST**
728 * check for this error code, and make sure it never gets reflected
729 * back to userspace.
730 */
731static struct dx_frame *
732dx_probe(struct ext4_filename *fname, struct inode *dir,
733 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
734{
735 unsigned count, indirect;
736 struct dx_entry *at, *entries, *p, *q, *m;
737 struct dx_root *root;
738 struct dx_frame *frame = frame_in;
739 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
740 u32 hash;
741
742 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
743 if (IS_ERR(frame->bh))
744 return (struct dx_frame *) frame->bh;
745
746 root = (struct dx_root *) frame->bh->b_data;
747 if (root->info.hash_version != DX_HASH_TEA &&
748 root->info.hash_version != DX_HASH_HALF_MD4 &&
749 root->info.hash_version != DX_HASH_LEGACY) {
750 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
751 root->info.hash_version);
752 goto fail;
753 }
754 if (fname)
755 hinfo = &fname->hinfo;
756 hinfo->hash_version = root->info.hash_version;
757 if (hinfo->hash_version <= DX_HASH_TEA)
758 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
759 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
760 if (fname && fname_name(fname))
761 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
762 hash = hinfo->hash;
763
764 if (root->info.unused_flags & 1) {
765 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
766 root->info.unused_flags);
767 goto fail;
768 }
769
770 indirect = root->info.indirect_levels;
771 if (indirect > 1) {
772 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
773 root->info.indirect_levels);
774 goto fail;
775 }
776
777 entries = (struct dx_entry *)(((char *)&root->info) +
778 root->info.info_length);
779
780 if (dx_get_limit(entries) != dx_root_limit(dir,
781 root->info.info_length)) {
782 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
783 dx_get_limit(entries),
784 dx_root_limit(dir, root->info.info_length));
785 goto fail;
786 }
787
788 dxtrace(printk("Look up %x", hash));
789 while (1) {
790 count = dx_get_count(entries);
791 if (!count || count > dx_get_limit(entries)) {
792 ext4_warning_inode(dir,
793 "dx entry: count %u beyond limit %u",
794 count, dx_get_limit(entries));
795 goto fail;
796 }
797
798 p = entries + 1;
799 q = entries + count - 1;
800 while (p <= q) {
801 m = p + (q - p) / 2;
802 dxtrace(printk(KERN_CONT "."));
803 if (dx_get_hash(m) > hash)
804 q = m - 1;
805 else
806 p = m + 1;
807 }
808
809 if (0) { // linear search cross check
810 unsigned n = count - 1;
811 at = entries;
812 while (n--)
813 {
814 dxtrace(printk(KERN_CONT ","));
815 if (dx_get_hash(++at) > hash)
816 {
817 at--;
818 break;
819 }
820 }
821 assert (at == p - 1);
822 }
823
824 at = p - 1;
825 dxtrace(printk(KERN_CONT " %x->%u\n",
826 at == entries ? 0 : dx_get_hash(at),
827 dx_get_block(at)));
828 frame->entries = entries;
829 frame->at = at;
830 if (!indirect--)
831 return frame;
832 frame++;
833 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
834 if (IS_ERR(frame->bh)) {
835 ret_err = (struct dx_frame *) frame->bh;
836 frame->bh = NULL;
837 goto fail;
838 }
839 entries = ((struct dx_node *) frame->bh->b_data)->entries;
840
841 if (dx_get_limit(entries) != dx_node_limit(dir)) {
842 ext4_warning_inode(dir,
843 "dx entry: limit %u != node limit %u",
844 dx_get_limit(entries), dx_node_limit(dir));
845 goto fail;
846 }
847 }
848fail:
849 while (frame >= frame_in) {
850 brelse(frame->bh);
851 frame--;
852 }
853
854 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
855 ext4_warning_inode(dir,
856 "Corrupt directory, running e2fsck is recommended");
857 return ret_err;
858}
859
860static void dx_release(struct dx_frame *frames)
861{
862 if (frames[0].bh == NULL)
863 return;
864
865 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
866 brelse(frames[1].bh);
867 brelse(frames[0].bh);
868}
869
870/*
871 * This function increments the frame pointer to search the next leaf
872 * block, and reads in the necessary intervening nodes if the search
873 * should be necessary. Whether or not the search is necessary is
874 * controlled by the hash parameter. If the hash value is even, then
875 * the search is only continued if the next block starts with that
876 * hash value. This is used if we are searching for a specific file.
877 *
878 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
879 *
880 * This function returns 1 if the caller should continue to search,
881 * or 0 if it should not. If there is an error reading one of the
882 * index blocks, it will a negative error code.
883 *
884 * If start_hash is non-null, it will be filled in with the starting
885 * hash of the next page.
886 */
887static int ext4_htree_next_block(struct inode *dir, __u32 hash,
888 struct dx_frame *frame,
889 struct dx_frame *frames,
890 __u32 *start_hash)
891{
892 struct dx_frame *p;
893 struct buffer_head *bh;
894 int num_frames = 0;
895 __u32 bhash;
896
897 p = frame;
898 /*
899 * Find the next leaf page by incrementing the frame pointer.
900 * If we run out of entries in the interior node, loop around and
901 * increment pointer in the parent node. When we break out of
902 * this loop, num_frames indicates the number of interior
903 * nodes need to be read.
904 */
905 while (1) {
906 if (++(p->at) < p->entries + dx_get_count(p->entries))
907 break;
908 if (p == frames)
909 return 0;
910 num_frames++;
911 p--;
912 }
913
914 /*
915 * If the hash is 1, then continue only if the next page has a
916 * continuation hash of any value. This is used for readdir
917 * handling. Otherwise, check to see if the hash matches the
918 * desired contiuation hash. If it doesn't, return since
919 * there's no point to read in the successive index pages.
920 */
921 bhash = dx_get_hash(p->at);
922 if (start_hash)
923 *start_hash = bhash;
924 if ((hash & 1) == 0) {
925 if ((bhash & ~1) != hash)
926 return 0;
927 }
928 /*
929 * If the hash is HASH_NB_ALWAYS, we always go to the next
930 * block so no check is necessary
931 */
932 while (num_frames--) {
933 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
934 if (IS_ERR(bh))
935 return PTR_ERR(bh);
936 p++;
937 brelse(p->bh);
938 p->bh = bh;
939 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
940 }
941 return 1;
942}
943
944
945/*
946 * This function fills a red-black tree with information from a
947 * directory block. It returns the number directory entries loaded
948 * into the tree. If there is an error it is returned in err.
949 */
950static int htree_dirblock_to_tree(struct file *dir_file,
951 struct inode *dir, ext4_lblk_t block,
952 struct dx_hash_info *hinfo,
953 __u32 start_hash, __u32 start_minor_hash)
954{
955 struct buffer_head *bh;
956 struct ext4_dir_entry_2 *de, *top;
957 int err = 0, count = 0;
958 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
959
960 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
961 (unsigned long)block));
962 bh = ext4_read_dirblock(dir, block, DIRENT);
963 if (IS_ERR(bh))
964 return PTR_ERR(bh);
965
966 de = (struct ext4_dir_entry_2 *) bh->b_data;
967 top = (struct ext4_dir_entry_2 *) ((char *) de +
968 dir->i_sb->s_blocksize -
969 EXT4_DIR_REC_LEN(0));
970#ifdef CONFIG_EXT4_FS_ENCRYPTION
971 /* Check if the directory is encrypted */
972 if (ext4_encrypted_inode(dir)) {
973 err = fscrypt_get_encryption_info(dir);
974 if (err < 0) {
975 brelse(bh);
976 return err;
977 }
978 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
979 &fname_crypto_str);
980 if (err < 0) {
981 brelse(bh);
982 return err;
983 }
984 }
985#endif
986 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
987 if (ext4_check_dir_entry(dir, NULL, de, bh,
988 bh->b_data, bh->b_size,
989 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
990 + ((char *)de - bh->b_data))) {
991 /* silently ignore the rest of the block */
992 break;
993 }
994 ext4fs_dirhash(de->name, de->name_len, hinfo);
995 if ((hinfo->hash < start_hash) ||
996 ((hinfo->hash == start_hash) &&
997 (hinfo->minor_hash < start_minor_hash)))
998 continue;
999 if (de->inode == 0)
1000 continue;
1001 if (!ext4_encrypted_inode(dir)) {
1002 tmp_str.name = de->name;
1003 tmp_str.len = de->name_len;
1004 err = ext4_htree_store_dirent(dir_file,
1005 hinfo->hash, hinfo->minor_hash, de,
1006 &tmp_str);
1007 } else {
1008 int save_len = fname_crypto_str.len;
1009 struct fscrypt_str de_name = FSTR_INIT(de->name,
1010 de->name_len);
1011
1012 /* Directory is encrypted */
1013 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1014 hinfo->minor_hash, &de_name,
1015 &fname_crypto_str);
1016 if (err) {
1017 count = err;
1018 goto errout;
1019 }
1020 err = ext4_htree_store_dirent(dir_file,
1021 hinfo->hash, hinfo->minor_hash, de,
1022 &fname_crypto_str);
1023 fname_crypto_str.len = save_len;
1024 }
1025 if (err != 0) {
1026 count = err;
1027 goto errout;
1028 }
1029 count++;
1030 }
1031errout:
1032 brelse(bh);
1033#ifdef CONFIG_EXT4_FS_ENCRYPTION
1034 fscrypt_fname_free_buffer(&fname_crypto_str);
1035#endif
1036 return count;
1037}
1038
1039
1040/*
1041 * This function fills a red-black tree with information from a
1042 * directory. We start scanning the directory in hash order, starting
1043 * at start_hash and start_minor_hash.
1044 *
1045 * This function returns the number of entries inserted into the tree,
1046 * or a negative error code.
1047 */
1048int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1049 __u32 start_minor_hash, __u32 *next_hash)
1050{
1051 struct dx_hash_info hinfo;
1052 struct ext4_dir_entry_2 *de;
1053 struct dx_frame frames[2], *frame;
1054 struct inode *dir;
1055 ext4_lblk_t block;
1056 int count = 0;
1057 int ret, err;
1058 __u32 hashval;
1059 struct fscrypt_str tmp_str;
1060
1061 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1062 start_hash, start_minor_hash));
1063 dir = file_inode(dir_file);
1064 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1065 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1066 if (hinfo.hash_version <= DX_HASH_TEA)
1067 hinfo.hash_version +=
1068 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1069 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1070 if (ext4_has_inline_data(dir)) {
1071 int has_inline_data = 1;
1072 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1073 &hinfo, start_hash,
1074 start_minor_hash,
1075 &has_inline_data);
1076 if (has_inline_data) {
1077 *next_hash = ~0;
1078 return count;
1079 }
1080 }
1081 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1082 start_hash, start_minor_hash);
1083 *next_hash = ~0;
1084 return count;
1085 }
1086 hinfo.hash = start_hash;
1087 hinfo.minor_hash = 0;
1088 frame = dx_probe(NULL, dir, &hinfo, frames);
1089 if (IS_ERR(frame))
1090 return PTR_ERR(frame);
1091
1092 /* Add '.' and '..' from the htree header */
1093 if (!start_hash && !start_minor_hash) {
1094 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1095 tmp_str.name = de->name;
1096 tmp_str.len = de->name_len;
1097 err = ext4_htree_store_dirent(dir_file, 0, 0,
1098 de, &tmp_str);
1099 if (err != 0)
1100 goto errout;
1101 count++;
1102 }
1103 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1104 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1105 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1106 tmp_str.name = de->name;
1107 tmp_str.len = de->name_len;
1108 err = ext4_htree_store_dirent(dir_file, 2, 0,
1109 de, &tmp_str);
1110 if (err != 0)
1111 goto errout;
1112 count++;
1113 }
1114
1115 while (1) {
1116 if (fatal_signal_pending(current)) {
1117 err = -ERESTARTSYS;
1118 goto errout;
1119 }
1120 cond_resched();
1121 block = dx_get_block(frame->at);
1122 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1123 start_hash, start_minor_hash);
1124 if (ret < 0) {
1125 err = ret;
1126 goto errout;
1127 }
1128 count += ret;
1129 hashval = ~0;
1130 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1131 frame, frames, &hashval);
1132 *next_hash = hashval;
1133 if (ret < 0) {
1134 err = ret;
1135 goto errout;
1136 }
1137 /*
1138 * Stop if: (a) there are no more entries, or
1139 * (b) we have inserted at least one entry and the
1140 * next hash value is not a continuation
1141 */
1142 if ((ret == 0) ||
1143 (count && ((hashval & 1) == 0)))
1144 break;
1145 }
1146 dx_release(frames);
1147 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1148 "next hash: %x\n", count, *next_hash));
1149 return count;
1150errout:
1151 dx_release(frames);
1152 return (err);
1153}
1154
1155static inline int search_dirblock(struct buffer_head *bh,
1156 struct inode *dir,
1157 struct ext4_filename *fname,
1158 const struct qstr *d_name,
1159 unsigned int offset,
1160 struct ext4_dir_entry_2 **res_dir)
1161{
1162 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1163 fname, d_name, offset, res_dir);
1164}
1165
1166/*
1167 * Directory block splitting, compacting
1168 */
1169
1170/*
1171 * Create map of hash values, offsets, and sizes, stored at end of block.
1172 * Returns number of entries mapped.
1173 */
1174static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1175 unsigned blocksize, struct dx_hash_info *hinfo,
1176 struct dx_map_entry *map_tail)
1177{
1178 int count = 0;
1179 char *base = (char *) de;
1180 struct dx_hash_info h = *hinfo;
1181
1182 while ((char *) de < base + blocksize) {
1183 if (de->name_len && de->inode) {
1184 ext4fs_dirhash(de->name, de->name_len, &h);
1185 map_tail--;
1186 map_tail->hash = h.hash;
1187 map_tail->offs = ((char *) de - base)>>2;
1188 map_tail->size = le16_to_cpu(de->rec_len);
1189 count++;
1190 cond_resched();
1191 }
1192 /* XXX: do we need to check rec_len == 0 case? -Chris */
1193 de = ext4_next_entry(de, blocksize);
1194 }
1195 return count;
1196}
1197
1198/* Sort map by hash value */
1199static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1200{
1201 struct dx_map_entry *p, *q, *top = map + count - 1;
1202 int more;
1203 /* Combsort until bubble sort doesn't suck */
1204 while (count > 2) {
1205 count = count*10/13;
1206 if (count - 9 < 2) /* 9, 10 -> 11 */
1207 count = 11;
1208 for (p = top, q = p - count; q >= map; p--, q--)
1209 if (p->hash < q->hash)
1210 swap(*p, *q);
1211 }
1212 /* Garden variety bubble sort */
1213 do {
1214 more = 0;
1215 q = top;
1216 while (q-- > map) {
1217 if (q[1].hash >= q[0].hash)
1218 continue;
1219 swap(*(q+1), *q);
1220 more = 1;
1221 }
1222 } while(more);
1223}
1224
1225static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1226{
1227 struct dx_entry *entries = frame->entries;
1228 struct dx_entry *old = frame->at, *new = old + 1;
1229 int count = dx_get_count(entries);
1230
1231 assert(count < dx_get_limit(entries));
1232 assert(old < entries + count);
1233 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1234 dx_set_hash(new, hash);
1235 dx_set_block(new, block);
1236 dx_set_count(entries, count + 1);
1237}
1238
1239/*
1240 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1241 *
1242 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1243 * `de != NULL' is guaranteed by caller.
1244 */
1245static inline int ext4_match(struct ext4_filename *fname,
1246 struct ext4_dir_entry_2 *de)
1247{
1248 const void *name = fname_name(fname);
1249 u32 len = fname_len(fname);
1250
1251 if (!de->inode)
1252 return 0;
1253
1254#ifdef CONFIG_EXT4_FS_ENCRYPTION
1255 if (unlikely(!name)) {
1256 if (fname->usr_fname->name[0] == '_') {
1257 int ret;
1258 if (de->name_len < 16)
1259 return 0;
1260 ret = memcmp(de->name + de->name_len - 16,
1261 fname->crypto_buf.name + 8, 16);
1262 return (ret == 0) ? 1 : 0;
1263 }
1264 name = fname->crypto_buf.name;
1265 len = fname->crypto_buf.len;
1266 }
1267#endif
1268 if (de->name_len != len)
1269 return 0;
1270 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1271}
1272
1273/*
1274 * Returns 0 if not found, -1 on failure, and 1 on success
1275 */
1276int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1277 struct inode *dir, struct ext4_filename *fname,
1278 const struct qstr *d_name,
1279 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1280{
1281 struct ext4_dir_entry_2 * de;
1282 char * dlimit;
1283 int de_len;
1284 int res;
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 res = ext4_match(fname, de);
1293 if (res < 0) {
1294 res = -1;
1295 goto return_result;
1296 }
1297 if (res > 0) {
1298 /* found a match - just to be sure, do
1299 * a full check */
1300 if (ext4_check_dir_entry(dir, NULL, de, bh,
1301 bh->b_data,
1302 bh->b_size, offset)) {
1303 res = -1;
1304 goto return_result;
1305 }
1306 *res_dir = de;
1307 res = 1;
1308 goto return_result;
1309 }
1310
1311 }
1312 /* prevent looping on a bad block */
1313 de_len = ext4_rec_len_from_disk(de->rec_len,
1314 dir->i_sb->s_blocksize);
1315 if (de_len <= 0) {
1316 res = -1;
1317 goto return_result;
1318 }
1319 offset += de_len;
1320 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1321 }
1322
1323 res = 0;
1324return_result:
1325 return res;
1326}
1327
1328static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1329 struct ext4_dir_entry *de)
1330{
1331 struct super_block *sb = dir->i_sb;
1332
1333 if (!is_dx(dir))
1334 return 0;
1335 if (block == 0)
1336 return 1;
1337 if (de->inode == 0 &&
1338 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1339 sb->s_blocksize)
1340 return 1;
1341 return 0;
1342}
1343
1344/*
1345 * ext4_find_entry()
1346 *
1347 * finds an entry in the specified directory with the wanted name. It
1348 * returns the cache buffer in which the entry was found, and the entry
1349 * itself (as a parameter - res_dir). It does NOT read the inode of the
1350 * entry - you'll have to do that yourself if you want to.
1351 *
1352 * The returned buffer_head has ->b_count elevated. The caller is expected
1353 * to brelse() it when appropriate.
1354 */
1355static struct buffer_head * ext4_find_entry (struct inode *dir,
1356 const struct qstr *d_name,
1357 struct ext4_dir_entry_2 **res_dir,
1358 int *inlined)
1359{
1360 struct super_block *sb;
1361 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1362 struct buffer_head *bh, *ret = NULL;
1363 ext4_lblk_t start, block, b;
1364 const u8 *name = d_name->name;
1365 int ra_max = 0; /* Number of bh's in the readahead
1366 buffer, bh_use[] */
1367 int ra_ptr = 0; /* Current index into readahead
1368 buffer */
1369 int num = 0;
1370 ext4_lblk_t nblocks;
1371 int i, namelen, retval;
1372 struct ext4_filename fname;
1373
1374 *res_dir = NULL;
1375 sb = dir->i_sb;
1376 namelen = d_name->len;
1377 if (namelen > EXT4_NAME_LEN)
1378 return NULL;
1379
1380 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1381 if (retval)
1382 return ERR_PTR(retval);
1383
1384 if (ext4_has_inline_data(dir)) {
1385 int has_inline_data = 1;
1386 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1387 &has_inline_data);
1388 if (has_inline_data) {
1389 if (inlined)
1390 *inlined = 1;
1391 goto cleanup_and_exit;
1392 }
1393 }
1394
1395 if ((namelen <= 2) && (name[0] == '.') &&
1396 (name[1] == '.' || name[1] == '\0')) {
1397 /*
1398 * "." or ".." will only be in the first block
1399 * NFS may look up ".."; "." should be handled by the VFS
1400 */
1401 block = start = 0;
1402 nblocks = 1;
1403 goto restart;
1404 }
1405 if (is_dx(dir)) {
1406 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1407 /*
1408 * On success, or if the error was file not found,
1409 * return. Otherwise, fall back to doing a search the
1410 * old fashioned way.
1411 */
1412 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1413 goto cleanup_and_exit;
1414 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1415 "falling back\n"));
1416 }
1417 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1418 start = EXT4_I(dir)->i_dir_start_lookup;
1419 if (start >= nblocks)
1420 start = 0;
1421 block = start;
1422restart:
1423 do {
1424 /*
1425 * We deal with the read-ahead logic here.
1426 */
1427 if (ra_ptr >= ra_max) {
1428 /* Refill the readahead buffer */
1429 ra_ptr = 0;
1430 b = block;
1431 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1432 /*
1433 * Terminate if we reach the end of the
1434 * directory and must wrap, or if our
1435 * search has finished at this block.
1436 */
1437 if (b >= nblocks || (num && block == start)) {
1438 bh_use[ra_max] = NULL;
1439 break;
1440 }
1441 num++;
1442 bh = ext4_getblk(NULL, dir, b++, 0);
1443 if (IS_ERR(bh)) {
1444 if (ra_max == 0) {
1445 ret = bh;
1446 goto cleanup_and_exit;
1447 }
1448 break;
1449 }
1450 bh_use[ra_max] = bh;
1451 if (bh)
1452 ll_rw_block(REQ_OP_READ,
1453 REQ_META | REQ_PRIO,
1454 1, &bh);
1455 }
1456 }
1457 if ((bh = bh_use[ra_ptr++]) == NULL)
1458 goto next;
1459 wait_on_buffer(bh);
1460 if (!buffer_uptodate(bh)) {
1461 /* read error, skip block & hope for the best */
1462 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1463 (unsigned long) block);
1464 brelse(bh);
1465 goto next;
1466 }
1467 if (!buffer_verified(bh) &&
1468 !is_dx_internal_node(dir, block,
1469 (struct ext4_dir_entry *)bh->b_data) &&
1470 !ext4_dirent_csum_verify(dir,
1471 (struct ext4_dir_entry *)bh->b_data)) {
1472 EXT4_ERROR_INODE(dir, "checksumming directory "
1473 "block %lu", (unsigned long)block);
1474 brelse(bh);
1475 goto next;
1476 }
1477 set_buffer_verified(bh);
1478 i = search_dirblock(bh, dir, &fname, d_name,
1479 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1480 if (i == 1) {
1481 EXT4_I(dir)->i_dir_start_lookup = block;
1482 ret = bh;
1483 goto cleanup_and_exit;
1484 } else {
1485 brelse(bh);
1486 if (i < 0)
1487 goto cleanup_and_exit;
1488 }
1489 next:
1490 if (++block >= nblocks)
1491 block = 0;
1492 } while (block != start);
1493
1494 /*
1495 * If the directory has grown while we were searching, then
1496 * search the last part of the directory before giving up.
1497 */
1498 block = nblocks;
1499 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1500 if (block < nblocks) {
1501 start = 0;
1502 goto restart;
1503 }
1504
1505cleanup_and_exit:
1506 /* Clean up the read-ahead blocks */
1507 for (; ra_ptr < ra_max; ra_ptr++)
1508 brelse(bh_use[ra_ptr]);
1509 ext4_fname_free_filename(&fname);
1510 return ret;
1511}
1512
1513static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1514 struct ext4_filename *fname,
1515 struct ext4_dir_entry_2 **res_dir)
1516{
1517 struct super_block * sb = dir->i_sb;
1518 struct dx_frame frames[2], *frame;
1519 const struct qstr *d_name = fname->usr_fname;
1520 struct buffer_head *bh;
1521 ext4_lblk_t block;
1522 int retval;
1523
1524#ifdef CONFIG_EXT4_FS_ENCRYPTION
1525 *res_dir = NULL;
1526#endif
1527 frame = dx_probe(fname, dir, NULL, frames);
1528 if (IS_ERR(frame))
1529 return (struct buffer_head *) frame;
1530 do {
1531 block = dx_get_block(frame->at);
1532 bh = ext4_read_dirblock(dir, block, DIRENT);
1533 if (IS_ERR(bh))
1534 goto errout;
1535
1536 retval = search_dirblock(bh, dir, fname, d_name,
1537 block << EXT4_BLOCK_SIZE_BITS(sb),
1538 res_dir);
1539 if (retval == 1)
1540 goto success;
1541 brelse(bh);
1542 if (retval == -1) {
1543 bh = ERR_PTR(ERR_BAD_DX_DIR);
1544 goto errout;
1545 }
1546
1547 /* Check to see if we should continue to search */
1548 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1549 frames, NULL);
1550 if (retval < 0) {
1551 ext4_warning_inode(dir,
1552 "error %d reading directory index block",
1553 retval);
1554 bh = ERR_PTR(retval);
1555 goto errout;
1556 }
1557 } while (retval == 1);
1558
1559 bh = NULL;
1560errout:
1561 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1562success:
1563 dx_release(frames);
1564 return bh;
1565}
1566
1567static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1568{
1569 struct inode *inode;
1570 struct ext4_dir_entry_2 *de;
1571 struct buffer_head *bh;
1572
1573 if (ext4_encrypted_inode(dir)) {
1574 int res = fscrypt_get_encryption_info(dir);
1575
1576 /*
1577 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1578 * created while the directory was encrypted and we
1579 * have access to the key.
1580 */
1581 if (fscrypt_has_encryption_key(dir))
1582 fscrypt_set_encrypted_dentry(dentry);
1583 fscrypt_set_d_op(dentry);
1584 if (res && res != -ENOKEY)
1585 return ERR_PTR(res);
1586 }
1587
1588 if (dentry->d_name.len > EXT4_NAME_LEN)
1589 return ERR_PTR(-ENAMETOOLONG);
1590
1591 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1592 if (IS_ERR(bh))
1593 return (struct dentry *) bh;
1594 inode = NULL;
1595 if (bh) {
1596 __u32 ino = le32_to_cpu(de->inode);
1597 brelse(bh);
1598 if (!ext4_valid_inum(dir->i_sb, ino)) {
1599 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1600 return ERR_PTR(-EFSCORRUPTED);
1601 }
1602 if (unlikely(ino == dir->i_ino)) {
1603 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1604 dentry);
1605 return ERR_PTR(-EFSCORRUPTED);
1606 }
1607 inode = ext4_iget_normal(dir->i_sb, ino);
1608 if (inode == ERR_PTR(-ESTALE)) {
1609 EXT4_ERROR_INODE(dir,
1610 "deleted inode referenced: %u",
1611 ino);
1612 return ERR_PTR(-EFSCORRUPTED);
1613 }
1614 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1615 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1616 !fscrypt_has_permitted_context(dir, inode)) {
1617 int nokey = ext4_encrypted_inode(inode) &&
1618 !fscrypt_has_encryption_key(inode);
1619 if (nokey) {
1620 iput(inode);
1621 return ERR_PTR(-ENOKEY);
1622 }
1623 ext4_warning(inode->i_sb,
1624 "Inconsistent encryption contexts: %lu/%lu",
1625 (unsigned long) dir->i_ino,
1626 (unsigned long) inode->i_ino);
1627 iput(inode);
1628 return ERR_PTR(-EPERM);
1629 }
1630 }
1631 return d_splice_alias(inode, dentry);
1632}
1633
1634
1635struct dentry *ext4_get_parent(struct dentry *child)
1636{
1637 __u32 ino;
1638 static const struct qstr dotdot = QSTR_INIT("..", 2);
1639 struct ext4_dir_entry_2 * de;
1640 struct buffer_head *bh;
1641
1642 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1643 if (IS_ERR(bh))
1644 return (struct dentry *) bh;
1645 if (!bh)
1646 return ERR_PTR(-ENOENT);
1647 ino = le32_to_cpu(de->inode);
1648 brelse(bh);
1649
1650 if (!ext4_valid_inum(child->d_sb, ino)) {
1651 EXT4_ERROR_INODE(d_inode(child),
1652 "bad parent inode number: %u", ino);
1653 return ERR_PTR(-EFSCORRUPTED);
1654 }
1655
1656 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1657}
1658
1659/*
1660 * Move count entries from end of map between two memory locations.
1661 * Returns pointer to last entry moved.
1662 */
1663static struct ext4_dir_entry_2 *
1664dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1665 unsigned blocksize)
1666{
1667 unsigned rec_len = 0;
1668
1669 while (count--) {
1670 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1671 (from + (map->offs<<2));
1672 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1673 memcpy (to, de, rec_len);
1674 ((struct ext4_dir_entry_2 *) to)->rec_len =
1675 ext4_rec_len_to_disk(rec_len, blocksize);
1676 de->inode = 0;
1677 map++;
1678 to += rec_len;
1679 }
1680 return (struct ext4_dir_entry_2 *) (to - rec_len);
1681}
1682
1683/*
1684 * Compact each dir entry in the range to the minimal rec_len.
1685 * Returns pointer to last entry in range.
1686 */
1687static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1688{
1689 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1690 unsigned rec_len = 0;
1691
1692 prev = to = de;
1693 while ((char*)de < base + blocksize) {
1694 next = ext4_next_entry(de, blocksize);
1695 if (de->inode && de->name_len) {
1696 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1697 if (de > to)
1698 memmove(to, de, rec_len);
1699 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1700 prev = to;
1701 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1702 }
1703 de = next;
1704 }
1705 return prev;
1706}
1707
1708/*
1709 * Split a full leaf block to make room for a new dir entry.
1710 * Allocate a new block, and move entries so that they are approx. equally full.
1711 * Returns pointer to de in block into which the new entry will be inserted.
1712 */
1713static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1714 struct buffer_head **bh,struct dx_frame *frame,
1715 struct dx_hash_info *hinfo)
1716{
1717 unsigned blocksize = dir->i_sb->s_blocksize;
1718 unsigned count, continued;
1719 struct buffer_head *bh2;
1720 ext4_lblk_t newblock;
1721 u32 hash2;
1722 struct dx_map_entry *map;
1723 char *data1 = (*bh)->b_data, *data2;
1724 unsigned split, move, size;
1725 struct ext4_dir_entry_2 *de = NULL, *de2;
1726 struct ext4_dir_entry_tail *t;
1727 int csum_size = 0;
1728 int err = 0, i;
1729
1730 if (ext4_has_metadata_csum(dir->i_sb))
1731 csum_size = sizeof(struct ext4_dir_entry_tail);
1732
1733 bh2 = ext4_append(handle, dir, &newblock);
1734 if (IS_ERR(bh2)) {
1735 brelse(*bh);
1736 *bh = NULL;
1737 return (struct ext4_dir_entry_2 *) bh2;
1738 }
1739
1740 BUFFER_TRACE(*bh, "get_write_access");
1741 err = ext4_journal_get_write_access(handle, *bh);
1742 if (err)
1743 goto journal_error;
1744
1745 BUFFER_TRACE(frame->bh, "get_write_access");
1746 err = ext4_journal_get_write_access(handle, frame->bh);
1747 if (err)
1748 goto journal_error;
1749
1750 data2 = bh2->b_data;
1751
1752 /* create map in the end of data2 block */
1753 map = (struct dx_map_entry *) (data2 + blocksize);
1754 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1755 blocksize, hinfo, map);
1756 map -= count;
1757 dx_sort_map(map, count);
1758 /* Split the existing block in the middle, size-wise */
1759 size = 0;
1760 move = 0;
1761 for (i = count-1; i >= 0; i--) {
1762 /* is more than half of this entry in 2nd half of the block? */
1763 if (size + map[i].size/2 > blocksize/2)
1764 break;
1765 size += map[i].size;
1766 move++;
1767 }
1768 /* map index at which we will split */
1769 split = count - move;
1770 hash2 = map[split].hash;
1771 continued = hash2 == map[split - 1].hash;
1772 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1773 (unsigned long)dx_get_block(frame->at),
1774 hash2, split, count-split));
1775
1776 /* Fancy dance to stay within two buffers */
1777 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1778 blocksize);
1779 de = dx_pack_dirents(data1, blocksize);
1780 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1781 (char *) de,
1782 blocksize);
1783 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1784 (char *) de2,
1785 blocksize);
1786 if (csum_size) {
1787 t = EXT4_DIRENT_TAIL(data2, blocksize);
1788 initialize_dirent_tail(t, blocksize);
1789
1790 t = EXT4_DIRENT_TAIL(data1, blocksize);
1791 initialize_dirent_tail(t, blocksize);
1792 }
1793
1794 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1795 blocksize, 1));
1796 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1797 blocksize, 1));
1798
1799 /* Which block gets the new entry? */
1800 if (hinfo->hash >= hash2) {
1801 swap(*bh, bh2);
1802 de = de2;
1803 }
1804 dx_insert_block(frame, hash2 + continued, newblock);
1805 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1806 if (err)
1807 goto journal_error;
1808 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1809 if (err)
1810 goto journal_error;
1811 brelse(bh2);
1812 dxtrace(dx_show_index("frame", frame->entries));
1813 return de;
1814
1815journal_error:
1816 brelse(*bh);
1817 brelse(bh2);
1818 *bh = NULL;
1819 ext4_std_error(dir->i_sb, err);
1820 return ERR_PTR(err);
1821}
1822
1823int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1824 struct buffer_head *bh,
1825 void *buf, int buf_size,
1826 struct ext4_filename *fname,
1827 struct ext4_dir_entry_2 **dest_de)
1828{
1829 struct ext4_dir_entry_2 *de;
1830 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1831 int nlen, rlen;
1832 unsigned int offset = 0;
1833 char *top;
1834 int res;
1835
1836 de = (struct ext4_dir_entry_2 *)buf;
1837 top = buf + buf_size - reclen;
1838 while ((char *) de <= top) {
1839 if (ext4_check_dir_entry(dir, NULL, de, bh,
1840 buf, buf_size, offset)) {
1841 res = -EFSCORRUPTED;
1842 goto return_result;
1843 }
1844 /* Provide crypto context and crypto buffer to ext4 match */
1845 res = ext4_match(fname, de);
1846 if (res < 0)
1847 goto return_result;
1848 if (res > 0) {
1849 res = -EEXIST;
1850 goto return_result;
1851 }
1852 nlen = EXT4_DIR_REC_LEN(de->name_len);
1853 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1854 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1855 break;
1856 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1857 offset += rlen;
1858 }
1859
1860 if ((char *) de > top)
1861 res = -ENOSPC;
1862 else {
1863 *dest_de = de;
1864 res = 0;
1865 }
1866return_result:
1867 return res;
1868}
1869
1870int ext4_insert_dentry(struct inode *dir,
1871 struct inode *inode,
1872 struct ext4_dir_entry_2 *de,
1873 int buf_size,
1874 struct ext4_filename *fname)
1875{
1876
1877 int nlen, rlen;
1878
1879 nlen = EXT4_DIR_REC_LEN(de->name_len);
1880 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1881 if (de->inode) {
1882 struct ext4_dir_entry_2 *de1 =
1883 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1884 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1885 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1886 de = de1;
1887 }
1888 de->file_type = EXT4_FT_UNKNOWN;
1889 de->inode = cpu_to_le32(inode->i_ino);
1890 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1891 de->name_len = fname_len(fname);
1892 memcpy(de->name, fname_name(fname), fname_len(fname));
1893 return 0;
1894}
1895
1896/*
1897 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1898 * it points to a directory entry which is guaranteed to be large
1899 * enough for new directory entry. If de is NULL, then
1900 * add_dirent_to_buf will attempt search the directory block for
1901 * space. It will return -ENOSPC if no space is available, and -EIO
1902 * and -EEXIST if directory entry already exists.
1903 */
1904static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1905 struct inode *dir,
1906 struct inode *inode, struct ext4_dir_entry_2 *de,
1907 struct buffer_head *bh)
1908{
1909 unsigned int blocksize = dir->i_sb->s_blocksize;
1910 int csum_size = 0;
1911 int err;
1912
1913 if (ext4_has_metadata_csum(inode->i_sb))
1914 csum_size = sizeof(struct ext4_dir_entry_tail);
1915
1916 if (!de) {
1917 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1918 blocksize - csum_size, fname, &de);
1919 if (err)
1920 return err;
1921 }
1922 BUFFER_TRACE(bh, "get_write_access");
1923 err = ext4_journal_get_write_access(handle, bh);
1924 if (err) {
1925 ext4_std_error(dir->i_sb, err);
1926 return err;
1927 }
1928
1929 /* By now the buffer is marked for journaling. Due to crypto operations,
1930 * the following function call may fail */
1931 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1932 if (err < 0)
1933 return err;
1934
1935 /*
1936 * XXX shouldn't update any times until successful
1937 * completion of syscall, but too many callers depend
1938 * on this.
1939 *
1940 * XXX similarly, too many callers depend on
1941 * ext4_new_inode() setting the times, but error
1942 * recovery deletes the inode, so the worst that can
1943 * happen is that the times are slightly out of date
1944 * and/or different from the directory change time.
1945 */
1946 dir->i_mtime = dir->i_ctime = current_time(dir);
1947 ext4_update_dx_flag(dir);
1948 dir->i_version++;
1949 ext4_mark_inode_dirty(handle, dir);
1950 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1951 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1952 if (err)
1953 ext4_std_error(dir->i_sb, err);
1954 return 0;
1955}
1956
1957/*
1958 * This converts a one block unindexed directory to a 3 block indexed
1959 * directory, and adds the dentry to the indexed directory.
1960 */
1961static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1962 struct inode *dir,
1963 struct inode *inode, struct buffer_head *bh)
1964{
1965 struct buffer_head *bh2;
1966 struct dx_root *root;
1967 struct dx_frame frames[2], *frame;
1968 struct dx_entry *entries;
1969 struct ext4_dir_entry_2 *de, *de2;
1970 struct ext4_dir_entry_tail *t;
1971 char *data1, *top;
1972 unsigned len;
1973 int retval;
1974 unsigned blocksize;
1975 ext4_lblk_t block;
1976 struct fake_dirent *fde;
1977 int csum_size = 0;
1978
1979 if (ext4_has_metadata_csum(inode->i_sb))
1980 csum_size = sizeof(struct ext4_dir_entry_tail);
1981
1982 blocksize = dir->i_sb->s_blocksize;
1983 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1984 BUFFER_TRACE(bh, "get_write_access");
1985 retval = ext4_journal_get_write_access(handle, bh);
1986 if (retval) {
1987 ext4_std_error(dir->i_sb, retval);
1988 brelse(bh);
1989 return retval;
1990 }
1991 root = (struct dx_root *) bh->b_data;
1992
1993 /* The 0th block becomes the root, move the dirents out */
1994 fde = &root->dotdot;
1995 de = (struct ext4_dir_entry_2 *)((char *)fde +
1996 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1997 if ((char *) de >= (((char *) root) + blocksize)) {
1998 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1999 brelse(bh);
2000 return -EFSCORRUPTED;
2001 }
2002 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2003
2004 /* Allocate new block for the 0th block's dirents */
2005 bh2 = ext4_append(handle, dir, &block);
2006 if (IS_ERR(bh2)) {
2007 brelse(bh);
2008 return PTR_ERR(bh2);
2009 }
2010 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2011 data1 = bh2->b_data;
2012
2013 memcpy (data1, de, len);
2014 de = (struct ext4_dir_entry_2 *) data1;
2015 top = data1 + len;
2016 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2017 de = de2;
2018 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2019 (char *) de,
2020 blocksize);
2021
2022 if (csum_size) {
2023 t = EXT4_DIRENT_TAIL(data1, blocksize);
2024 initialize_dirent_tail(t, blocksize);
2025 }
2026
2027 /* Initialize the root; the dot dirents already exist */
2028 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2029 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2030 blocksize);
2031 memset (&root->info, 0, sizeof(root->info));
2032 root->info.info_length = sizeof(root->info);
2033 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2034 entries = root->entries;
2035 dx_set_block(entries, 1);
2036 dx_set_count(entries, 1);
2037 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2038
2039 /* Initialize as for dx_probe */
2040 fname->hinfo.hash_version = root->info.hash_version;
2041 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2042 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2043 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2044 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2045
2046 memset(frames, 0, sizeof(frames));
2047 frame = frames;
2048 frame->entries = entries;
2049 frame->at = entries;
2050 frame->bh = bh;
2051
2052 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2053 if (retval)
2054 goto out_frames;
2055 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2056 if (retval)
2057 goto out_frames;
2058
2059 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2060 if (IS_ERR(de)) {
2061 retval = PTR_ERR(de);
2062 goto out_frames;
2063 }
2064
2065 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2066out_frames:
2067 /*
2068 * Even if the block split failed, we have to properly write
2069 * out all the changes we did so far. Otherwise we can end up
2070 * with corrupted filesystem.
2071 */
2072 if (retval)
2073 ext4_mark_inode_dirty(handle, dir);
2074 dx_release(frames);
2075 brelse(bh2);
2076 return retval;
2077}
2078
2079/*
2080 * ext4_add_entry()
2081 *
2082 * adds a file entry to the specified directory, using the same
2083 * semantics as ext4_find_entry(). It returns NULL if it failed.
2084 *
2085 * NOTE!! The inode part of 'de' is left at 0 - which means you
2086 * may not sleep between calling this and putting something into
2087 * the entry, as someone else might have used it while you slept.
2088 */
2089static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2090 struct inode *inode)
2091{
2092 struct inode *dir = d_inode(dentry->d_parent);
2093 struct buffer_head *bh = NULL;
2094 struct ext4_dir_entry_2 *de;
2095 struct ext4_dir_entry_tail *t;
2096 struct super_block *sb;
2097 struct ext4_filename fname;
2098 int retval;
2099 int dx_fallback=0;
2100 unsigned blocksize;
2101 ext4_lblk_t block, blocks;
2102 int csum_size = 0;
2103
2104 if (ext4_has_metadata_csum(inode->i_sb))
2105 csum_size = sizeof(struct ext4_dir_entry_tail);
2106
2107 sb = dir->i_sb;
2108 blocksize = sb->s_blocksize;
2109 if (!dentry->d_name.len)
2110 return -EINVAL;
2111
2112 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2113 if (retval)
2114 return retval;
2115
2116 if (ext4_has_inline_data(dir)) {
2117 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2118 if (retval < 0)
2119 goto out;
2120 if (retval == 1) {
2121 retval = 0;
2122 goto out;
2123 }
2124 }
2125
2126 if (is_dx(dir)) {
2127 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2128 if (!retval || (retval != ERR_BAD_DX_DIR))
2129 goto out;
2130 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2131 dx_fallback++;
2132 ext4_mark_inode_dirty(handle, dir);
2133 }
2134 blocks = dir->i_size >> sb->s_blocksize_bits;
2135 for (block = 0; block < blocks; block++) {
2136 bh = ext4_read_dirblock(dir, block, DIRENT);
2137 if (IS_ERR(bh)) {
2138 retval = PTR_ERR(bh);
2139 bh = NULL;
2140 goto out;
2141 }
2142 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2143 NULL, bh);
2144 if (retval != -ENOSPC)
2145 goto out;
2146
2147 if (blocks == 1 && !dx_fallback &&
2148 ext4_has_feature_dir_index(sb)) {
2149 retval = make_indexed_dir(handle, &fname, dir,
2150 inode, bh);
2151 bh = NULL; /* make_indexed_dir releases bh */
2152 goto out;
2153 }
2154 brelse(bh);
2155 }
2156 bh = ext4_append(handle, dir, &block);
2157 if (IS_ERR(bh)) {
2158 retval = PTR_ERR(bh);
2159 bh = NULL;
2160 goto out;
2161 }
2162 de = (struct ext4_dir_entry_2 *) bh->b_data;
2163 de->inode = 0;
2164 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2165
2166 if (csum_size) {
2167 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2168 initialize_dirent_tail(t, blocksize);
2169 }
2170
2171 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2172out:
2173 ext4_fname_free_filename(&fname);
2174 brelse(bh);
2175 if (retval == 0)
2176 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2177 return retval;
2178}
2179
2180/*
2181 * Returns 0 for success, or a negative error value
2182 */
2183static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2184 struct inode *dir, struct inode *inode)
2185{
2186 struct dx_frame frames[2], *frame;
2187 struct dx_entry *entries, *at;
2188 struct buffer_head *bh;
2189 struct super_block *sb = dir->i_sb;
2190 struct ext4_dir_entry_2 *de;
2191 int err;
2192
2193 frame = dx_probe(fname, dir, NULL, frames);
2194 if (IS_ERR(frame))
2195 return PTR_ERR(frame);
2196 entries = frame->entries;
2197 at = frame->at;
2198 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2199 if (IS_ERR(bh)) {
2200 err = PTR_ERR(bh);
2201 bh = NULL;
2202 goto cleanup;
2203 }
2204
2205 BUFFER_TRACE(bh, "get_write_access");
2206 err = ext4_journal_get_write_access(handle, bh);
2207 if (err)
2208 goto journal_error;
2209
2210 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2211 if (err != -ENOSPC)
2212 goto cleanup;
2213
2214 /* Block full, should compress but for now just split */
2215 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2216 dx_get_count(entries), dx_get_limit(entries)));
2217 /* Need to split index? */
2218 if (dx_get_count(entries) == dx_get_limit(entries)) {
2219 ext4_lblk_t newblock;
2220 unsigned icount = dx_get_count(entries);
2221 int levels = frame - frames;
2222 struct dx_entry *entries2;
2223 struct dx_node *node2;
2224 struct buffer_head *bh2;
2225
2226 if (levels && (dx_get_count(frames->entries) ==
2227 dx_get_limit(frames->entries))) {
2228 ext4_warning_inode(dir, "Directory index full!");
2229 err = -ENOSPC;
2230 goto cleanup;
2231 }
2232 bh2 = ext4_append(handle, dir, &newblock);
2233 if (IS_ERR(bh2)) {
2234 err = PTR_ERR(bh2);
2235 goto cleanup;
2236 }
2237 node2 = (struct dx_node *)(bh2->b_data);
2238 entries2 = node2->entries;
2239 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2240 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2241 sb->s_blocksize);
2242 BUFFER_TRACE(frame->bh, "get_write_access");
2243 err = ext4_journal_get_write_access(handle, frame->bh);
2244 if (err)
2245 goto journal_error;
2246 if (levels) {
2247 unsigned icount1 = icount/2, icount2 = icount - icount1;
2248 unsigned hash2 = dx_get_hash(entries + icount1);
2249 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2250 icount1, icount2));
2251
2252 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2253 err = ext4_journal_get_write_access(handle,
2254 frames[0].bh);
2255 if (err)
2256 goto journal_error;
2257
2258 memcpy((char *) entries2, (char *) (entries + icount1),
2259 icount2 * sizeof(struct dx_entry));
2260 dx_set_count(entries, icount1);
2261 dx_set_count(entries2, icount2);
2262 dx_set_limit(entries2, dx_node_limit(dir));
2263
2264 /* Which index block gets the new entry? */
2265 if (at - entries >= icount1) {
2266 frame->at = at = at - entries - icount1 + entries2;
2267 frame->entries = entries = entries2;
2268 swap(frame->bh, bh2);
2269 }
2270 dx_insert_block(frames + 0, hash2, newblock);
2271 dxtrace(dx_show_index("node", frames[1].entries));
2272 dxtrace(dx_show_index("node",
2273 ((struct dx_node *) bh2->b_data)->entries));
2274 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2275 if (err)
2276 goto journal_error;
2277 brelse (bh2);
2278 } else {
2279 dxtrace(printk(KERN_DEBUG
2280 "Creating second level index...\n"));
2281 memcpy((char *) entries2, (char *) entries,
2282 icount * sizeof(struct dx_entry));
2283 dx_set_limit(entries2, dx_node_limit(dir));
2284
2285 /* Set up root */
2286 dx_set_count(entries, 1);
2287 dx_set_block(entries + 0, newblock);
2288 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2289
2290 /* Add new access path frame */
2291 frame = frames + 1;
2292 frame->at = at = at - entries + entries2;
2293 frame->entries = entries = entries2;
2294 frame->bh = bh2;
2295 err = ext4_journal_get_write_access(handle,
2296 frame->bh);
2297 if (err)
2298 goto journal_error;
2299 }
2300 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2301 if (err) {
2302 ext4_std_error(inode->i_sb, err);
2303 goto cleanup;
2304 }
2305 }
2306 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2307 if (IS_ERR(de)) {
2308 err = PTR_ERR(de);
2309 goto cleanup;
2310 }
2311 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2312 goto cleanup;
2313
2314journal_error:
2315 ext4_std_error(dir->i_sb, err);
2316cleanup:
2317 brelse(bh);
2318 dx_release(frames);
2319 return err;
2320}
2321
2322/*
2323 * ext4_generic_delete_entry deletes a directory entry by merging it
2324 * with the previous entry
2325 */
2326int ext4_generic_delete_entry(handle_t *handle,
2327 struct inode *dir,
2328 struct ext4_dir_entry_2 *de_del,
2329 struct buffer_head *bh,
2330 void *entry_buf,
2331 int buf_size,
2332 int csum_size)
2333{
2334 struct ext4_dir_entry_2 *de, *pde;
2335 unsigned int blocksize = dir->i_sb->s_blocksize;
2336 int i;
2337
2338 i = 0;
2339 pde = NULL;
2340 de = (struct ext4_dir_entry_2 *)entry_buf;
2341 while (i < buf_size - csum_size) {
2342 if (ext4_check_dir_entry(dir, NULL, de, bh,
2343 bh->b_data, bh->b_size, i))
2344 return -EFSCORRUPTED;
2345 if (de == de_del) {
2346 if (pde)
2347 pde->rec_len = ext4_rec_len_to_disk(
2348 ext4_rec_len_from_disk(pde->rec_len,
2349 blocksize) +
2350 ext4_rec_len_from_disk(de->rec_len,
2351 blocksize),
2352 blocksize);
2353 else
2354 de->inode = 0;
2355 dir->i_version++;
2356 return 0;
2357 }
2358 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2359 pde = de;
2360 de = ext4_next_entry(de, blocksize);
2361 }
2362 return -ENOENT;
2363}
2364
2365static int ext4_delete_entry(handle_t *handle,
2366 struct inode *dir,
2367 struct ext4_dir_entry_2 *de_del,
2368 struct buffer_head *bh)
2369{
2370 int err, csum_size = 0;
2371
2372 if (ext4_has_inline_data(dir)) {
2373 int has_inline_data = 1;
2374 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2375 &has_inline_data);
2376 if (has_inline_data)
2377 return err;
2378 }
2379
2380 if (ext4_has_metadata_csum(dir->i_sb))
2381 csum_size = sizeof(struct ext4_dir_entry_tail);
2382
2383 BUFFER_TRACE(bh, "get_write_access");
2384 err = ext4_journal_get_write_access(handle, bh);
2385 if (unlikely(err))
2386 goto out;
2387
2388 err = ext4_generic_delete_entry(handle, dir, de_del,
2389 bh, bh->b_data,
2390 dir->i_sb->s_blocksize, csum_size);
2391 if (err)
2392 goto out;
2393
2394 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2395 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2396 if (unlikely(err))
2397 goto out;
2398
2399 return 0;
2400out:
2401 if (err != -ENOENT)
2402 ext4_std_error(dir->i_sb, err);
2403 return err;
2404}
2405
2406/*
2407 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2408 * since this indicates that nlinks count was previously 1.
2409 */
2410static void ext4_inc_count(handle_t *handle, struct inode *inode)
2411{
2412 inc_nlink(inode);
2413 if (is_dx(inode) && inode->i_nlink > 1) {
2414 /* limit is 16-bit i_links_count */
2415 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2416 set_nlink(inode, 1);
2417 ext4_set_feature_dir_nlink(inode->i_sb);
2418 }
2419 }
2420}
2421
2422/*
2423 * If a directory had nlink == 1, then we should let it be 1. This indicates
2424 * directory has >EXT4_LINK_MAX subdirs.
2425 */
2426static void ext4_dec_count(handle_t *handle, struct inode *inode)
2427{
2428 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2429 drop_nlink(inode);
2430}
2431
2432
2433static int ext4_add_nondir(handle_t *handle,
2434 struct dentry *dentry, struct inode *inode)
2435{
2436 int err = ext4_add_entry(handle, dentry, inode);
2437 if (!err) {
2438 ext4_mark_inode_dirty(handle, inode);
2439 unlock_new_inode(inode);
2440 d_instantiate(dentry, inode);
2441 return 0;
2442 }
2443 drop_nlink(inode);
2444 unlock_new_inode(inode);
2445 iput(inode);
2446 return err;
2447}
2448
2449/*
2450 * By the time this is called, we already have created
2451 * the directory cache entry for the new file, but it
2452 * is so far negative - it has no inode.
2453 *
2454 * If the create succeeds, we fill in the inode information
2455 * with d_instantiate().
2456 */
2457static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2458 bool excl)
2459{
2460 handle_t *handle;
2461 struct inode *inode;
2462 int err, credits, retries = 0;
2463
2464 err = dquot_initialize(dir);
2465 if (err)
2466 return err;
2467
2468 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2469 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2470retry:
2471 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2472 NULL, EXT4_HT_DIR, credits);
2473 handle = ext4_journal_current_handle();
2474 err = PTR_ERR(inode);
2475 if (!IS_ERR(inode)) {
2476 inode->i_op = &ext4_file_inode_operations;
2477 inode->i_fop = &ext4_file_operations;
2478 ext4_set_aops(inode);
2479 err = ext4_add_nondir(handle, dentry, inode);
2480 if (!err && IS_DIRSYNC(dir))
2481 ext4_handle_sync(handle);
2482 }
2483 if (handle)
2484 ext4_journal_stop(handle);
2485 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2486 goto retry;
2487 return err;
2488}
2489
2490static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2491 umode_t mode, dev_t rdev)
2492{
2493 handle_t *handle;
2494 struct inode *inode;
2495 int err, credits, retries = 0;
2496
2497 err = dquot_initialize(dir);
2498 if (err)
2499 return err;
2500
2501 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2502 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2503retry:
2504 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2505 NULL, EXT4_HT_DIR, credits);
2506 handle = ext4_journal_current_handle();
2507 err = PTR_ERR(inode);
2508 if (!IS_ERR(inode)) {
2509 init_special_inode(inode, inode->i_mode, rdev);
2510 inode->i_op = &ext4_special_inode_operations;
2511 err = ext4_add_nondir(handle, dentry, inode);
2512 if (!err && IS_DIRSYNC(dir))
2513 ext4_handle_sync(handle);
2514 }
2515 if (handle)
2516 ext4_journal_stop(handle);
2517 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2518 goto retry;
2519 return err;
2520}
2521
2522static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2523{
2524 handle_t *handle;
2525 struct inode *inode;
2526 int err, retries = 0;
2527
2528 err = dquot_initialize(dir);
2529 if (err)
2530 return err;
2531
2532retry:
2533 inode = ext4_new_inode_start_handle(dir, mode,
2534 NULL, 0, NULL,
2535 EXT4_HT_DIR,
2536 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2537 4 + EXT4_XATTR_TRANS_BLOCKS);
2538 handle = ext4_journal_current_handle();
2539 err = PTR_ERR(inode);
2540 if (!IS_ERR(inode)) {
2541 inode->i_op = &ext4_file_inode_operations;
2542 inode->i_fop = &ext4_file_operations;
2543 ext4_set_aops(inode);
2544 d_tmpfile(dentry, inode);
2545 err = ext4_orphan_add(handle, inode);
2546 if (err)
2547 goto err_unlock_inode;
2548 mark_inode_dirty(inode);
2549 unlock_new_inode(inode);
2550 }
2551 if (handle)
2552 ext4_journal_stop(handle);
2553 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2554 goto retry;
2555 return err;
2556err_unlock_inode:
2557 ext4_journal_stop(handle);
2558 unlock_new_inode(inode);
2559 return err;
2560}
2561
2562struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2563 struct ext4_dir_entry_2 *de,
2564 int blocksize, int csum_size,
2565 unsigned int parent_ino, int dotdot_real_len)
2566{
2567 de->inode = cpu_to_le32(inode->i_ino);
2568 de->name_len = 1;
2569 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2570 blocksize);
2571 strcpy(de->name, ".");
2572 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2573
2574 de = ext4_next_entry(de, blocksize);
2575 de->inode = cpu_to_le32(parent_ino);
2576 de->name_len = 2;
2577 if (!dotdot_real_len)
2578 de->rec_len = ext4_rec_len_to_disk(blocksize -
2579 (csum_size + EXT4_DIR_REC_LEN(1)),
2580 blocksize);
2581 else
2582 de->rec_len = ext4_rec_len_to_disk(
2583 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2584 strcpy(de->name, "..");
2585 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2586
2587 return ext4_next_entry(de, blocksize);
2588}
2589
2590static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2591 struct inode *inode)
2592{
2593 struct buffer_head *dir_block = NULL;
2594 struct ext4_dir_entry_2 *de;
2595 struct ext4_dir_entry_tail *t;
2596 ext4_lblk_t block = 0;
2597 unsigned int blocksize = dir->i_sb->s_blocksize;
2598 int csum_size = 0;
2599 int err;
2600
2601 if (ext4_has_metadata_csum(dir->i_sb))
2602 csum_size = sizeof(struct ext4_dir_entry_tail);
2603
2604 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2605 err = ext4_try_create_inline_dir(handle, dir, inode);
2606 if (err < 0 && err != -ENOSPC)
2607 goto out;
2608 if (!err)
2609 goto out;
2610 }
2611
2612 inode->i_size = 0;
2613 dir_block = ext4_append(handle, inode, &block);
2614 if (IS_ERR(dir_block))
2615 return PTR_ERR(dir_block);
2616 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2617 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2618 set_nlink(inode, 2);
2619 if (csum_size) {
2620 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2621 initialize_dirent_tail(t, blocksize);
2622 }
2623
2624 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2625 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2626 if (err)
2627 goto out;
2628 set_buffer_verified(dir_block);
2629out:
2630 brelse(dir_block);
2631 return err;
2632}
2633
2634static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2635{
2636 handle_t *handle;
2637 struct inode *inode;
2638 int err, credits, retries = 0;
2639
2640 if (EXT4_DIR_LINK_MAX(dir))
2641 return -EMLINK;
2642
2643 err = dquot_initialize(dir);
2644 if (err)
2645 return err;
2646
2647 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2648 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2649retry:
2650 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2651 &dentry->d_name,
2652 0, NULL, EXT4_HT_DIR, credits);
2653 handle = ext4_journal_current_handle();
2654 err = PTR_ERR(inode);
2655 if (IS_ERR(inode))
2656 goto out_stop;
2657
2658 inode->i_op = &ext4_dir_inode_operations;
2659 inode->i_fop = &ext4_dir_operations;
2660 err = ext4_init_new_dir(handle, dir, inode);
2661 if (err)
2662 goto out_clear_inode;
2663 err = ext4_mark_inode_dirty(handle, inode);
2664 if (!err)
2665 err = ext4_add_entry(handle, dentry, inode);
2666 if (err) {
2667out_clear_inode:
2668 clear_nlink(inode);
2669 unlock_new_inode(inode);
2670 ext4_mark_inode_dirty(handle, inode);
2671 iput(inode);
2672 goto out_stop;
2673 }
2674 ext4_inc_count(handle, dir);
2675 ext4_update_dx_flag(dir);
2676 err = ext4_mark_inode_dirty(handle, dir);
2677 if (err)
2678 goto out_clear_inode;
2679 unlock_new_inode(inode);
2680 d_instantiate(dentry, inode);
2681 if (IS_DIRSYNC(dir))
2682 ext4_handle_sync(handle);
2683
2684out_stop:
2685 if (handle)
2686 ext4_journal_stop(handle);
2687 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2688 goto retry;
2689 return err;
2690}
2691
2692/*
2693 * routine to check that the specified directory is empty (for rmdir)
2694 */
2695bool ext4_empty_dir(struct inode *inode)
2696{
2697 unsigned int offset;
2698 struct buffer_head *bh;
2699 struct ext4_dir_entry_2 *de, *de1;
2700 struct super_block *sb;
2701
2702 if (ext4_has_inline_data(inode)) {
2703 int has_inline_data = 1;
2704 int ret;
2705
2706 ret = empty_inline_dir(inode, &has_inline_data);
2707 if (has_inline_data)
2708 return ret;
2709 }
2710
2711 sb = inode->i_sb;
2712 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2713 EXT4_ERROR_INODE(inode, "invalid size");
2714 return true;
2715 }
2716 bh = ext4_read_dirblock(inode, 0, EITHER);
2717 if (IS_ERR(bh))
2718 return true;
2719
2720 de = (struct ext4_dir_entry_2 *) bh->b_data;
2721 de1 = ext4_next_entry(de, sb->s_blocksize);
2722 if (le32_to_cpu(de->inode) != inode->i_ino ||
2723 le32_to_cpu(de1->inode) == 0 ||
2724 strcmp(".", de->name) || strcmp("..", de1->name)) {
2725 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2726 brelse(bh);
2727 return true;
2728 }
2729 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2730 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2731 de = ext4_next_entry(de1, sb->s_blocksize);
2732 while (offset < inode->i_size) {
2733 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2734 unsigned int lblock;
2735 brelse(bh);
2736 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2737 bh = ext4_read_dirblock(inode, lblock, EITHER);
2738 if (IS_ERR(bh))
2739 return true;
2740 de = (struct ext4_dir_entry_2 *) bh->b_data;
2741 }
2742 if (ext4_check_dir_entry(inode, NULL, de, bh,
2743 bh->b_data, bh->b_size, offset)) {
2744 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2745 sb->s_blocksize);
2746 offset = (offset | (sb->s_blocksize - 1)) + 1;
2747 continue;
2748 }
2749 if (le32_to_cpu(de->inode)) {
2750 brelse(bh);
2751 return false;
2752 }
2753 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2754 de = ext4_next_entry(de, sb->s_blocksize);
2755 }
2756 brelse(bh);
2757 return true;
2758}
2759
2760/*
2761 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2762 * such inodes, starting at the superblock, in case we crash before the
2763 * file is closed/deleted, or in case the inode truncate spans multiple
2764 * transactions and the last transaction is not recovered after a crash.
2765 *
2766 * At filesystem recovery time, we walk this list deleting unlinked
2767 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2768 *
2769 * Orphan list manipulation functions must be called under i_mutex unless
2770 * we are just creating the inode or deleting it.
2771 */
2772int ext4_orphan_add(handle_t *handle, struct inode *inode)
2773{
2774 struct super_block *sb = inode->i_sb;
2775 struct ext4_sb_info *sbi = EXT4_SB(sb);
2776 struct ext4_iloc iloc;
2777 int err = 0, rc;
2778 bool dirty = false;
2779
2780 if (!sbi->s_journal || is_bad_inode(inode))
2781 return 0;
2782
2783 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2784 !inode_is_locked(inode));
2785 /*
2786 * Exit early if inode already is on orphan list. This is a big speedup
2787 * since we don't have to contend on the global s_orphan_lock.
2788 */
2789 if (!list_empty(&EXT4_I(inode)->i_orphan))
2790 return 0;
2791
2792 /*
2793 * Orphan handling is only valid for files with data blocks
2794 * being truncated, or files being unlinked. Note that we either
2795 * hold i_mutex, or the inode can not be referenced from outside,
2796 * so i_nlink should not be bumped due to race
2797 */
2798 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2799 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2800
2801 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2802 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2803 if (err)
2804 goto out;
2805
2806 err = ext4_reserve_inode_write(handle, inode, &iloc);
2807 if (err)
2808 goto out;
2809
2810 mutex_lock(&sbi->s_orphan_lock);
2811 /*
2812 * Due to previous errors inode may be already a part of on-disk
2813 * orphan list. If so skip on-disk list modification.
2814 */
2815 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2816 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2817 /* Insert this inode at the head of the on-disk orphan list */
2818 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2819 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2820 dirty = true;
2821 }
2822 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2823 mutex_unlock(&sbi->s_orphan_lock);
2824
2825 if (dirty) {
2826 err = ext4_handle_dirty_super(handle, sb);
2827 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2828 if (!err)
2829 err = rc;
2830 if (err) {
2831 /*
2832 * We have to remove inode from in-memory list if
2833 * addition to on disk orphan list failed. Stray orphan
2834 * list entries can cause panics at unmount time.
2835 */
2836 mutex_lock(&sbi->s_orphan_lock);
2837 list_del_init(&EXT4_I(inode)->i_orphan);
2838 mutex_unlock(&sbi->s_orphan_lock);
2839 }
2840 }
2841 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2842 jbd_debug(4, "orphan inode %lu will point to %d\n",
2843 inode->i_ino, NEXT_ORPHAN(inode));
2844out:
2845 ext4_std_error(sb, err);
2846 return err;
2847}
2848
2849/*
2850 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2851 * of such inodes stored on disk, because it is finally being cleaned up.
2852 */
2853int ext4_orphan_del(handle_t *handle, struct inode *inode)
2854{
2855 struct list_head *prev;
2856 struct ext4_inode_info *ei = EXT4_I(inode);
2857 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2858 __u32 ino_next;
2859 struct ext4_iloc iloc;
2860 int err = 0;
2861
2862 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2863 return 0;
2864
2865 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2866 !inode_is_locked(inode));
2867 /* Do this quick check before taking global s_orphan_lock. */
2868 if (list_empty(&ei->i_orphan))
2869 return 0;
2870
2871 if (handle) {
2872 /* Grab inode buffer early before taking global s_orphan_lock */
2873 err = ext4_reserve_inode_write(handle, inode, &iloc);
2874 }
2875
2876 mutex_lock(&sbi->s_orphan_lock);
2877 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2878
2879 prev = ei->i_orphan.prev;
2880 list_del_init(&ei->i_orphan);
2881
2882 /* If we're on an error path, we may not have a valid
2883 * transaction handle with which to update the orphan list on
2884 * disk, but we still need to remove the inode from the linked
2885 * list in memory. */
2886 if (!handle || err) {
2887 mutex_unlock(&sbi->s_orphan_lock);
2888 goto out_err;
2889 }
2890
2891 ino_next = NEXT_ORPHAN(inode);
2892 if (prev == &sbi->s_orphan) {
2893 jbd_debug(4, "superblock will point to %u\n", ino_next);
2894 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2895 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2896 if (err) {
2897 mutex_unlock(&sbi->s_orphan_lock);
2898 goto out_brelse;
2899 }
2900 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2901 mutex_unlock(&sbi->s_orphan_lock);
2902 err = ext4_handle_dirty_super(handle, inode->i_sb);
2903 } else {
2904 struct ext4_iloc iloc2;
2905 struct inode *i_prev =
2906 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2907
2908 jbd_debug(4, "orphan inode %lu will point to %u\n",
2909 i_prev->i_ino, ino_next);
2910 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2911 if (err) {
2912 mutex_unlock(&sbi->s_orphan_lock);
2913 goto out_brelse;
2914 }
2915 NEXT_ORPHAN(i_prev) = ino_next;
2916 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2917 mutex_unlock(&sbi->s_orphan_lock);
2918 }
2919 if (err)
2920 goto out_brelse;
2921 NEXT_ORPHAN(inode) = 0;
2922 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2923out_err:
2924 ext4_std_error(inode->i_sb, err);
2925 return err;
2926
2927out_brelse:
2928 brelse(iloc.bh);
2929 goto out_err;
2930}
2931
2932static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2933{
2934 int retval;
2935 struct inode *inode;
2936 struct buffer_head *bh;
2937 struct ext4_dir_entry_2 *de;
2938 handle_t *handle = NULL;
2939
2940 /* Initialize quotas before so that eventual writes go in
2941 * separate transaction */
2942 retval = dquot_initialize(dir);
2943 if (retval)
2944 return retval;
2945 retval = dquot_initialize(d_inode(dentry));
2946 if (retval)
2947 return retval;
2948
2949 retval = -ENOENT;
2950 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2951 if (IS_ERR(bh))
2952 return PTR_ERR(bh);
2953 if (!bh)
2954 goto end_rmdir;
2955
2956 inode = d_inode(dentry);
2957
2958 retval = -EFSCORRUPTED;
2959 if (le32_to_cpu(de->inode) != inode->i_ino)
2960 goto end_rmdir;
2961
2962 retval = -ENOTEMPTY;
2963 if (!ext4_empty_dir(inode))
2964 goto end_rmdir;
2965
2966 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2967 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2968 if (IS_ERR(handle)) {
2969 retval = PTR_ERR(handle);
2970 handle = NULL;
2971 goto end_rmdir;
2972 }
2973
2974 if (IS_DIRSYNC(dir))
2975 ext4_handle_sync(handle);
2976
2977 retval = ext4_delete_entry(handle, dir, de, bh);
2978 if (retval)
2979 goto end_rmdir;
2980 if (!EXT4_DIR_LINK_EMPTY(inode))
2981 ext4_warning_inode(inode,
2982 "empty directory '%.*s' has too many links (%u)",
2983 dentry->d_name.len, dentry->d_name.name,
2984 inode->i_nlink);
2985 inode->i_version++;
2986 clear_nlink(inode);
2987 /* There's no need to set i_disksize: the fact that i_nlink is
2988 * zero will ensure that the right thing happens during any
2989 * recovery. */
2990 inode->i_size = 0;
2991 ext4_orphan_add(handle, inode);
2992 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2993 ext4_mark_inode_dirty(handle, inode);
2994 ext4_dec_count(handle, dir);
2995 ext4_update_dx_flag(dir);
2996 ext4_mark_inode_dirty(handle, dir);
2997
2998end_rmdir:
2999 brelse(bh);
3000 if (handle)
3001 ext4_journal_stop(handle);
3002 return retval;
3003}
3004
3005static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3006{
3007 int retval;
3008 struct inode *inode;
3009 struct buffer_head *bh;
3010 struct ext4_dir_entry_2 *de;
3011 handle_t *handle = NULL;
3012
3013 trace_ext4_unlink_enter(dir, dentry);
3014 /* Initialize quotas before so that eventual writes go
3015 * in separate transaction */
3016 retval = dquot_initialize(dir);
3017 if (retval)
3018 return retval;
3019 retval = dquot_initialize(d_inode(dentry));
3020 if (retval)
3021 return retval;
3022
3023 retval = -ENOENT;
3024 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3025 if (IS_ERR(bh))
3026 return PTR_ERR(bh);
3027 if (!bh)
3028 goto end_unlink;
3029
3030 inode = d_inode(dentry);
3031
3032 retval = -EFSCORRUPTED;
3033 if (le32_to_cpu(de->inode) != inode->i_ino)
3034 goto end_unlink;
3035
3036 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3037 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3038 if (IS_ERR(handle)) {
3039 retval = PTR_ERR(handle);
3040 handle = NULL;
3041 goto end_unlink;
3042 }
3043
3044 if (IS_DIRSYNC(dir))
3045 ext4_handle_sync(handle);
3046
3047 if (inode->i_nlink == 0) {
3048 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3049 dentry->d_name.len, dentry->d_name.name);
3050 set_nlink(inode, 1);
3051 }
3052 retval = ext4_delete_entry(handle, dir, de, bh);
3053 if (retval)
3054 goto end_unlink;
3055 dir->i_ctime = dir->i_mtime = current_time(dir);
3056 ext4_update_dx_flag(dir);
3057 ext4_mark_inode_dirty(handle, dir);
3058 drop_nlink(inode);
3059 if (!inode->i_nlink)
3060 ext4_orphan_add(handle, inode);
3061 inode->i_ctime = current_time(inode);
3062 ext4_mark_inode_dirty(handle, inode);
3063
3064end_unlink:
3065 brelse(bh);
3066 if (handle)
3067 ext4_journal_stop(handle);
3068 trace_ext4_unlink_exit(dentry, retval);
3069 return retval;
3070}
3071
3072static int ext4_symlink(struct inode *dir,
3073 struct dentry *dentry, const char *symname)
3074{
3075 handle_t *handle;
3076 struct inode *inode;
3077 int err, len = strlen(symname);
3078 int credits;
3079 bool encryption_required;
3080 struct fscrypt_str disk_link;
3081 struct fscrypt_symlink_data *sd = NULL;
3082
3083 disk_link.len = len + 1;
3084 disk_link.name = (char *) symname;
3085
3086 encryption_required = (ext4_encrypted_inode(dir) ||
3087 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3088 if (encryption_required) {
3089 err = fscrypt_get_encryption_info(dir);
3090 if (err)
3091 return err;
3092 if (!fscrypt_has_encryption_key(dir))
3093 return -EPERM;
3094 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3095 sizeof(struct fscrypt_symlink_data));
3096 sd = kzalloc(disk_link.len, GFP_KERNEL);
3097 if (!sd)
3098 return -ENOMEM;
3099 }
3100
3101 if (disk_link.len > dir->i_sb->s_blocksize) {
3102 err = -ENAMETOOLONG;
3103 goto err_free_sd;
3104 }
3105
3106 err = dquot_initialize(dir);
3107 if (err)
3108 goto err_free_sd;
3109
3110 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3111 /*
3112 * For non-fast symlinks, we just allocate inode and put it on
3113 * orphan list in the first transaction => we need bitmap,
3114 * group descriptor, sb, inode block, quota blocks, and
3115 * possibly selinux xattr blocks.
3116 */
3117 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3118 EXT4_XATTR_TRANS_BLOCKS;
3119 } else {
3120 /*
3121 * Fast symlink. We have to add entry to directory
3122 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3123 * allocate new inode (bitmap, group descriptor, inode block,
3124 * quota blocks, sb is already counted in previous macros).
3125 */
3126 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3127 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3128 }
3129
3130 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3131 &dentry->d_name, 0, NULL,
3132 EXT4_HT_DIR, credits);
3133 handle = ext4_journal_current_handle();
3134 if (IS_ERR(inode)) {
3135 if (handle)
3136 ext4_journal_stop(handle);
3137 err = PTR_ERR(inode);
3138 goto err_free_sd;
3139 }
3140
3141 if (encryption_required) {
3142 struct qstr istr;
3143 struct fscrypt_str ostr =
3144 FSTR_INIT(sd->encrypted_path, disk_link.len);
3145
3146 istr.name = (const unsigned char *) symname;
3147 istr.len = len;
3148 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3149 if (err)
3150 goto err_drop_inode;
3151 sd->len = cpu_to_le16(ostr.len);
3152 disk_link.name = (char *) sd;
3153 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3154 }
3155
3156 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3157 if (!encryption_required)
3158 inode->i_op = &ext4_symlink_inode_operations;
3159 inode_nohighmem(inode);
3160 ext4_set_aops(inode);
3161 /*
3162 * We cannot call page_symlink() with transaction started
3163 * because it calls into ext4_write_begin() which can wait
3164 * for transaction commit if we are running out of space
3165 * and thus we deadlock. So we have to stop transaction now
3166 * and restart it when symlink contents is written.
3167 *
3168 * To keep fs consistent in case of crash, we have to put inode
3169 * to orphan list in the mean time.
3170 */
3171 drop_nlink(inode);
3172 err = ext4_orphan_add(handle, inode);
3173 ext4_journal_stop(handle);
3174 handle = NULL;
3175 if (err)
3176 goto err_drop_inode;
3177 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3178 if (err)
3179 goto err_drop_inode;
3180 /*
3181 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3182 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3183 */
3184 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3185 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3186 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3187 if (IS_ERR(handle)) {
3188 err = PTR_ERR(handle);
3189 handle = NULL;
3190 goto err_drop_inode;
3191 }
3192 set_nlink(inode, 1);
3193 err = ext4_orphan_del(handle, inode);
3194 if (err)
3195 goto err_drop_inode;
3196 } else {
3197 /* clear the extent format for fast symlink */
3198 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3199 if (!encryption_required) {
3200 inode->i_op = &ext4_fast_symlink_inode_operations;
3201 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3202 }
3203 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3204 disk_link.len);
3205 inode->i_size = disk_link.len - 1;
3206 }
3207 EXT4_I(inode)->i_disksize = inode->i_size;
3208 err = ext4_add_nondir(handle, dentry, inode);
3209 if (!err && IS_DIRSYNC(dir))
3210 ext4_handle_sync(handle);
3211
3212 if (handle)
3213 ext4_journal_stop(handle);
3214 kfree(sd);
3215 return err;
3216err_drop_inode:
3217 if (handle)
3218 ext4_journal_stop(handle);
3219 clear_nlink(inode);
3220 unlock_new_inode(inode);
3221 iput(inode);
3222err_free_sd:
3223 kfree(sd);
3224 return err;
3225}
3226
3227static int ext4_link(struct dentry *old_dentry,
3228 struct inode *dir, struct dentry *dentry)
3229{
3230 handle_t *handle;
3231 struct inode *inode = d_inode(old_dentry);
3232 int err, retries = 0;
3233
3234 if (inode->i_nlink >= EXT4_LINK_MAX)
3235 return -EMLINK;
3236 if (ext4_encrypted_inode(dir) &&
3237 !fscrypt_has_permitted_context(dir, inode))
3238 return -EPERM;
3239
3240 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3241 (!projid_eq(EXT4_I(dir)->i_projid,
3242 EXT4_I(old_dentry->d_inode)->i_projid)))
3243 return -EXDEV;
3244
3245 err = dquot_initialize(dir);
3246 if (err)
3247 return err;
3248
3249retry:
3250 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3251 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3252 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3253 if (IS_ERR(handle))
3254 return PTR_ERR(handle);
3255
3256 if (IS_DIRSYNC(dir))
3257 ext4_handle_sync(handle);
3258
3259 inode->i_ctime = current_time(inode);
3260 ext4_inc_count(handle, inode);
3261 ihold(inode);
3262
3263 err = ext4_add_entry(handle, dentry, inode);
3264 if (!err) {
3265 ext4_mark_inode_dirty(handle, inode);
3266 /* this can happen only for tmpfile being
3267 * linked the first time
3268 */
3269 if (inode->i_nlink == 1)
3270 ext4_orphan_del(handle, inode);
3271 d_instantiate(dentry, inode);
3272 } else {
3273 drop_nlink(inode);
3274 iput(inode);
3275 }
3276 ext4_journal_stop(handle);
3277 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3278 goto retry;
3279 return err;
3280}
3281
3282
3283/*
3284 * Try to find buffer head where contains the parent block.
3285 * It should be the inode block if it is inlined or the 1st block
3286 * if it is a normal dir.
3287 */
3288static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3289 struct inode *inode,
3290 int *retval,
3291 struct ext4_dir_entry_2 **parent_de,
3292 int *inlined)
3293{
3294 struct buffer_head *bh;
3295
3296 if (!ext4_has_inline_data(inode)) {
3297 bh = ext4_read_dirblock(inode, 0, EITHER);
3298 if (IS_ERR(bh)) {
3299 *retval = PTR_ERR(bh);
3300 return NULL;
3301 }
3302 *parent_de = ext4_next_entry(
3303 (struct ext4_dir_entry_2 *)bh->b_data,
3304 inode->i_sb->s_blocksize);
3305 return bh;
3306 }
3307
3308 *inlined = 1;
3309 return ext4_get_first_inline_block(inode, parent_de, retval);
3310}
3311
3312struct ext4_renament {
3313 struct inode *dir;
3314 struct dentry *dentry;
3315 struct inode *inode;
3316 bool is_dir;
3317 int dir_nlink_delta;
3318
3319 /* entry for "dentry" */
3320 struct buffer_head *bh;
3321 struct ext4_dir_entry_2 *de;
3322 int inlined;
3323
3324 /* entry for ".." in inode if it's a directory */
3325 struct buffer_head *dir_bh;
3326 struct ext4_dir_entry_2 *parent_de;
3327 int dir_inlined;
3328};
3329
3330static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3331{
3332 int retval;
3333
3334 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3335 &retval, &ent->parent_de,
3336 &ent->dir_inlined);
3337 if (!ent->dir_bh)
3338 return retval;
3339 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3340 return -EFSCORRUPTED;
3341 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3342 return ext4_journal_get_write_access(handle, ent->dir_bh);
3343}
3344
3345static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3346 unsigned dir_ino)
3347{
3348 int retval;
3349
3350 ent->parent_de->inode = cpu_to_le32(dir_ino);
3351 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3352 if (!ent->dir_inlined) {
3353 if (is_dx(ent->inode)) {
3354 retval = ext4_handle_dirty_dx_node(handle,
3355 ent->inode,
3356 ent->dir_bh);
3357 } else {
3358 retval = ext4_handle_dirty_dirent_node(handle,
3359 ent->inode,
3360 ent->dir_bh);
3361 }
3362 } else {
3363 retval = ext4_mark_inode_dirty(handle, ent->inode);
3364 }
3365 if (retval) {
3366 ext4_std_error(ent->dir->i_sb, retval);
3367 return retval;
3368 }
3369 return 0;
3370}
3371
3372static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3373 unsigned ino, unsigned file_type)
3374{
3375 int retval;
3376
3377 BUFFER_TRACE(ent->bh, "get write access");
3378 retval = ext4_journal_get_write_access(handle, ent->bh);
3379 if (retval)
3380 return retval;
3381 ent->de->inode = cpu_to_le32(ino);
3382 if (ext4_has_feature_filetype(ent->dir->i_sb))
3383 ent->de->file_type = file_type;
3384 ent->dir->i_version++;
3385 ent->dir->i_ctime = ent->dir->i_mtime =
3386 current_time(ent->dir);
3387 ext4_mark_inode_dirty(handle, ent->dir);
3388 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3389 if (!ent->inlined) {
3390 retval = ext4_handle_dirty_dirent_node(handle,
3391 ent->dir, ent->bh);
3392 if (unlikely(retval)) {
3393 ext4_std_error(ent->dir->i_sb, retval);
3394 return retval;
3395 }
3396 }
3397 brelse(ent->bh);
3398 ent->bh = NULL;
3399
3400 return 0;
3401}
3402
3403static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3404 const struct qstr *d_name)
3405{
3406 int retval = -ENOENT;
3407 struct buffer_head *bh;
3408 struct ext4_dir_entry_2 *de;
3409
3410 bh = ext4_find_entry(dir, d_name, &de, NULL);
3411 if (IS_ERR(bh))
3412 return PTR_ERR(bh);
3413 if (bh) {
3414 retval = ext4_delete_entry(handle, dir, de, bh);
3415 brelse(bh);
3416 }
3417 return retval;
3418}
3419
3420static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3421 int force_reread)
3422{
3423 int retval;
3424 /*
3425 * ent->de could have moved from under us during htree split, so make
3426 * sure that we are deleting the right entry. We might also be pointing
3427 * to a stale entry in the unused part of ent->bh so just checking inum
3428 * and the name isn't enough.
3429 */
3430 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3431 ent->de->name_len != ent->dentry->d_name.len ||
3432 strncmp(ent->de->name, ent->dentry->d_name.name,
3433 ent->de->name_len) ||
3434 force_reread) {
3435 retval = ext4_find_delete_entry(handle, ent->dir,
3436 &ent->dentry->d_name);
3437 } else {
3438 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3439 if (retval == -ENOENT) {
3440 retval = ext4_find_delete_entry(handle, ent->dir,
3441 &ent->dentry->d_name);
3442 }
3443 }
3444
3445 if (retval) {
3446 ext4_warning_inode(ent->dir,
3447 "Deleting old file: nlink %d, error=%d",
3448 ent->dir->i_nlink, retval);
3449 }
3450}
3451
3452static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3453{
3454 if (ent->dir_nlink_delta) {
3455 if (ent->dir_nlink_delta == -1)
3456 ext4_dec_count(handle, ent->dir);
3457 else
3458 ext4_inc_count(handle, ent->dir);
3459 ext4_mark_inode_dirty(handle, ent->dir);
3460 }
3461}
3462
3463static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3464 int credits, handle_t **h)
3465{
3466 struct inode *wh;
3467 handle_t *handle;
3468 int retries = 0;
3469
3470 /*
3471 * for inode block, sb block, group summaries,
3472 * and inode bitmap
3473 */
3474 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3475 EXT4_XATTR_TRANS_BLOCKS + 4);
3476retry:
3477 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3478 &ent->dentry->d_name, 0, NULL,
3479 EXT4_HT_DIR, credits);
3480
3481 handle = ext4_journal_current_handle();
3482 if (IS_ERR(wh)) {
3483 if (handle)
3484 ext4_journal_stop(handle);
3485 if (PTR_ERR(wh) == -ENOSPC &&
3486 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3487 goto retry;
3488 } else {
3489 *h = handle;
3490 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3491 wh->i_op = &ext4_special_inode_operations;
3492 }
3493 return wh;
3494}
3495
3496/*
3497 * Anybody can rename anything with this: the permission checks are left to the
3498 * higher-level routines.
3499 *
3500 * n.b. old_{dentry,inode) refers to the source dentry/inode
3501 * while new_{dentry,inode) refers to the destination dentry/inode
3502 * This comes from rename(const char *oldpath, const char *newpath)
3503 */
3504static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3505 struct inode *new_dir, struct dentry *new_dentry,
3506 unsigned int flags)
3507{
3508 handle_t *handle = NULL;
3509 struct ext4_renament old = {
3510 .dir = old_dir,
3511 .dentry = old_dentry,
3512 .inode = d_inode(old_dentry),
3513 };
3514 struct ext4_renament new = {
3515 .dir = new_dir,
3516 .dentry = new_dentry,
3517 .inode = d_inode(new_dentry),
3518 };
3519 int force_reread;
3520 int retval;
3521 struct inode *whiteout = NULL;
3522 int credits;
3523 u8 old_file_type;
3524
3525 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3526 (!projid_eq(EXT4_I(new_dir)->i_projid,
3527 EXT4_I(old_dentry->d_inode)->i_projid)))
3528 return -EXDEV;
3529
3530 retval = dquot_initialize(old.dir);
3531 if (retval)
3532 return retval;
3533 retval = dquot_initialize(new.dir);
3534 if (retval)
3535 return retval;
3536
3537 /* Initialize quotas before so that eventual writes go
3538 * in separate transaction */
3539 if (new.inode) {
3540 retval = dquot_initialize(new.inode);
3541 if (retval)
3542 return retval;
3543 }
3544
3545 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3546 if (IS_ERR(old.bh))
3547 return PTR_ERR(old.bh);
3548 /*
3549 * Check for inode number is _not_ due to possible IO errors.
3550 * We might rmdir the source, keep it as pwd of some process
3551 * and merrily kill the link to whatever was created under the
3552 * same name. Goodbye sticky bit ;-<
3553 */
3554 retval = -ENOENT;
3555 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3556 goto end_rename;
3557
3558 if ((old.dir != new.dir) &&
3559 ext4_encrypted_inode(new.dir) &&
3560 !fscrypt_has_permitted_context(new.dir, old.inode)) {
3561 retval = -EPERM;
3562 goto end_rename;
3563 }
3564
3565 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3566 &new.de, &new.inlined);
3567 if (IS_ERR(new.bh)) {
3568 retval = PTR_ERR(new.bh);
3569 new.bh = NULL;
3570 goto end_rename;
3571 }
3572 if (new.bh) {
3573 if (!new.inode) {
3574 brelse(new.bh);
3575 new.bh = NULL;
3576 }
3577 }
3578 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3579 ext4_alloc_da_blocks(old.inode);
3580
3581 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3582 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3583 if (!(flags & RENAME_WHITEOUT)) {
3584 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3585 if (IS_ERR(handle)) {
3586 retval = PTR_ERR(handle);
3587 handle = NULL;
3588 goto end_rename;
3589 }
3590 } else {
3591 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3592 if (IS_ERR(whiteout)) {
3593 retval = PTR_ERR(whiteout);
3594 whiteout = NULL;
3595 goto end_rename;
3596 }
3597 }
3598
3599 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3600 ext4_handle_sync(handle);
3601
3602 if (S_ISDIR(old.inode->i_mode)) {
3603 if (new.inode) {
3604 retval = -ENOTEMPTY;
3605 if (!ext4_empty_dir(new.inode))
3606 goto end_rename;
3607 } else {
3608 retval = -EMLINK;
3609 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3610 goto end_rename;
3611 }
3612 retval = ext4_rename_dir_prepare(handle, &old);
3613 if (retval)
3614 goto end_rename;
3615 }
3616 /*
3617 * If we're renaming a file within an inline_data dir and adding or
3618 * setting the new dirent causes a conversion from inline_data to
3619 * extents/blockmap, we need to force the dirent delete code to
3620 * re-read the directory, or else we end up trying to delete a dirent
3621 * from what is now the extent tree root (or a block map).
3622 */
3623 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3624 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3625
3626 old_file_type = old.de->file_type;
3627 if (whiteout) {
3628 /*
3629 * Do this before adding a new entry, so the old entry is sure
3630 * to be still pointing to the valid old entry.
3631 */
3632 retval = ext4_setent(handle, &old, whiteout->i_ino,
3633 EXT4_FT_CHRDEV);
3634 if (retval)
3635 goto end_rename;
3636 ext4_mark_inode_dirty(handle, whiteout);
3637 }
3638 if (!new.bh) {
3639 retval = ext4_add_entry(handle, new.dentry, old.inode);
3640 if (retval)
3641 goto end_rename;
3642 } else {
3643 retval = ext4_setent(handle, &new,
3644 old.inode->i_ino, old_file_type);
3645 if (retval)
3646 goto end_rename;
3647 }
3648 if (force_reread)
3649 force_reread = !ext4_test_inode_flag(new.dir,
3650 EXT4_INODE_INLINE_DATA);
3651
3652 /*
3653 * Like most other Unix systems, set the ctime for inodes on a
3654 * rename.
3655 */
3656 old.inode->i_ctime = current_time(old.inode);
3657 ext4_mark_inode_dirty(handle, old.inode);
3658
3659 if (!whiteout) {
3660 /*
3661 * ok, that's it
3662 */
3663 ext4_rename_delete(handle, &old, force_reread);
3664 }
3665
3666 if (new.inode) {
3667 ext4_dec_count(handle, new.inode);
3668 new.inode->i_ctime = current_time(new.inode);
3669 }
3670 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3671 ext4_update_dx_flag(old.dir);
3672 if (old.dir_bh) {
3673 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3674 if (retval)
3675 goto end_rename;
3676
3677 ext4_dec_count(handle, old.dir);
3678 if (new.inode) {
3679 /* checked ext4_empty_dir above, can't have another
3680 * parent, ext4_dec_count() won't work for many-linked
3681 * dirs */
3682 clear_nlink(new.inode);
3683 } else {
3684 ext4_inc_count(handle, new.dir);
3685 ext4_update_dx_flag(new.dir);
3686 ext4_mark_inode_dirty(handle, new.dir);
3687 }
3688 }
3689 ext4_mark_inode_dirty(handle, old.dir);
3690 if (new.inode) {
3691 ext4_mark_inode_dirty(handle, new.inode);
3692 if (!new.inode->i_nlink)
3693 ext4_orphan_add(handle, new.inode);
3694 }
3695 retval = 0;
3696
3697end_rename:
3698 brelse(old.dir_bh);
3699 brelse(old.bh);
3700 brelse(new.bh);
3701 if (whiteout) {
3702 if (retval)
3703 drop_nlink(whiteout);
3704 unlock_new_inode(whiteout);
3705 iput(whiteout);
3706 }
3707 if (handle)
3708 ext4_journal_stop(handle);
3709 return retval;
3710}
3711
3712static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3713 struct inode *new_dir, struct dentry *new_dentry)
3714{
3715 handle_t *handle = NULL;
3716 struct ext4_renament old = {
3717 .dir = old_dir,
3718 .dentry = old_dentry,
3719 .inode = d_inode(old_dentry),
3720 };
3721 struct ext4_renament new = {
3722 .dir = new_dir,
3723 .dentry = new_dentry,
3724 .inode = d_inode(new_dentry),
3725 };
3726 u8 new_file_type;
3727 int retval;
3728 struct timespec ctime;
3729
3730 if ((ext4_encrypted_inode(old_dir) ||
3731 ext4_encrypted_inode(new_dir)) &&
3732 (old_dir != new_dir) &&
3733 (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3734 !fscrypt_has_permitted_context(old_dir, new.inode)))
3735 return -EPERM;
3736
3737 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3738 !projid_eq(EXT4_I(new_dir)->i_projid,
3739 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3740 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3741 !projid_eq(EXT4_I(old_dir)->i_projid,
3742 EXT4_I(new_dentry->d_inode)->i_projid)))
3743 return -EXDEV;
3744
3745 retval = dquot_initialize(old.dir);
3746 if (retval)
3747 return retval;
3748 retval = dquot_initialize(new.dir);
3749 if (retval)
3750 return retval;
3751
3752 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3753 &old.de, &old.inlined);
3754 if (IS_ERR(old.bh))
3755 return PTR_ERR(old.bh);
3756 /*
3757 * Check for inode number is _not_ due to possible IO errors.
3758 * We might rmdir the source, keep it as pwd of some process
3759 * and merrily kill the link to whatever was created under the
3760 * same name. Goodbye sticky bit ;-<
3761 */
3762 retval = -ENOENT;
3763 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3764 goto end_rename;
3765
3766 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3767 &new.de, &new.inlined);
3768 if (IS_ERR(new.bh)) {
3769 retval = PTR_ERR(new.bh);
3770 new.bh = NULL;
3771 goto end_rename;
3772 }
3773
3774 /* RENAME_EXCHANGE case: old *and* new must both exist */
3775 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3776 goto end_rename;
3777
3778 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3779 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3780 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3781 if (IS_ERR(handle)) {
3782 retval = PTR_ERR(handle);
3783 handle = NULL;
3784 goto end_rename;
3785 }
3786
3787 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3788 ext4_handle_sync(handle);
3789
3790 if (S_ISDIR(old.inode->i_mode)) {
3791 old.is_dir = true;
3792 retval = ext4_rename_dir_prepare(handle, &old);
3793 if (retval)
3794 goto end_rename;
3795 }
3796 if (S_ISDIR(new.inode->i_mode)) {
3797 new.is_dir = true;
3798 retval = ext4_rename_dir_prepare(handle, &new);
3799 if (retval)
3800 goto end_rename;
3801 }
3802
3803 /*
3804 * Other than the special case of overwriting a directory, parents'
3805 * nlink only needs to be modified if this is a cross directory rename.
3806 */
3807 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3808 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3809 new.dir_nlink_delta = -old.dir_nlink_delta;
3810 retval = -EMLINK;
3811 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3812 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3813 goto end_rename;
3814 }
3815
3816 new_file_type = new.de->file_type;
3817 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3818 if (retval)
3819 goto end_rename;
3820
3821 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3822 if (retval)
3823 goto end_rename;
3824
3825 /*
3826 * Like most other Unix systems, set the ctime for inodes on a
3827 * rename.
3828 */
3829 ctime = current_time(old.inode);
3830 old.inode->i_ctime = ctime;
3831 new.inode->i_ctime = ctime;
3832 ext4_mark_inode_dirty(handle, old.inode);
3833 ext4_mark_inode_dirty(handle, new.inode);
3834
3835 if (old.dir_bh) {
3836 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3837 if (retval)
3838 goto end_rename;
3839 }
3840 if (new.dir_bh) {
3841 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3842 if (retval)
3843 goto end_rename;
3844 }
3845 ext4_update_dir_count(handle, &old);
3846 ext4_update_dir_count(handle, &new);
3847 retval = 0;
3848
3849end_rename:
3850 brelse(old.dir_bh);
3851 brelse(new.dir_bh);
3852 brelse(old.bh);
3853 brelse(new.bh);
3854 if (handle)
3855 ext4_journal_stop(handle);
3856 return retval;
3857}
3858
3859static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3860 struct inode *new_dir, struct dentry *new_dentry,
3861 unsigned int flags)
3862{
3863 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3864 return -EINVAL;
3865
3866 if (flags & RENAME_EXCHANGE) {
3867 return ext4_cross_rename(old_dir, old_dentry,
3868 new_dir, new_dentry);
3869 }
3870
3871 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3872}
3873
3874/*
3875 * directories can handle most operations...
3876 */
3877const struct inode_operations ext4_dir_inode_operations = {
3878 .create = ext4_create,
3879 .lookup = ext4_lookup,
3880 .link = ext4_link,
3881 .unlink = ext4_unlink,
3882 .symlink = ext4_symlink,
3883 .mkdir = ext4_mkdir,
3884 .rmdir = ext4_rmdir,
3885 .mknod = ext4_mknod,
3886 .tmpfile = ext4_tmpfile,
3887 .rename = ext4_rename2,
3888 .setattr = ext4_setattr,
3889 .listxattr = ext4_listxattr,
3890 .get_acl = ext4_get_acl,
3891 .set_acl = ext4_set_acl,
3892 .fiemap = ext4_fiemap,
3893};
3894
3895const struct inode_operations ext4_special_inode_operations = {
3896 .setattr = ext4_setattr,
3897 .listxattr = ext4_listxattr,
3898 .get_acl = ext4_get_acl,
3899 .set_acl = ext4_set_acl,
3900};