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/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 __le32 save_csum;
424 int size;
425
426 size = count_offset + (count * sizeof(struct dx_entry));
427 save_csum = t->dt_checksum;
428 t->dt_checksum = 0;
429 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
431 t->dt_checksum = save_csum;
432
433 return cpu_to_le32(csum);
434}
435
436static int ext4_dx_csum_verify(struct inode *inode,
437 struct ext4_dir_entry *dirent)
438{
439 struct dx_countlimit *c;
440 struct dx_tail *t;
441 int count_offset, limit, count;
442
443 if (!ext4_has_metadata_csum(inode->i_sb))
444 return 1;
445
446 c = get_dx_countlimit(inode, dirent, &count_offset);
447 if (!c) {
448 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
449 return 1;
450 }
451 limit = le16_to_cpu(c->limit);
452 count = le16_to_cpu(c->count);
453 if (count_offset + (limit * sizeof(struct dx_entry)) >
454 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
455 warn_no_space_for_csum(inode);
456 return 1;
457 }
458 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
459
460 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
461 count, t))
462 return 0;
463 return 1;
464}
465
466static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
467{
468 struct dx_countlimit *c;
469 struct dx_tail *t;
470 int count_offset, limit, count;
471
472 if (!ext4_has_metadata_csum(inode->i_sb))
473 return;
474
475 c = get_dx_countlimit(inode, dirent, &count_offset);
476 if (!c) {
477 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
478 return;
479 }
480 limit = le16_to_cpu(c->limit);
481 count = le16_to_cpu(c->count);
482 if (count_offset + (limit * sizeof(struct dx_entry)) >
483 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
484 warn_no_space_for_csum(inode);
485 return;
486 }
487 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
488
489 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
490}
491
492static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493 struct inode *inode,
494 struct buffer_head *bh)
495{
496 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
497 return ext4_handle_dirty_metadata(handle, inode, bh);
498}
499
500/*
501 * p is at least 6 bytes before the end of page
502 */
503static inline struct ext4_dir_entry_2 *
504ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
505{
506 return (struct ext4_dir_entry_2 *)((char *)p +
507 ext4_rec_len_from_disk(p->rec_len, blocksize));
508}
509
510/*
511 * Future: use high four bits of block for coalesce-on-delete flags
512 * Mask them off for now.
513 */
514
515static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
516{
517 return le32_to_cpu(entry->block) & 0x00ffffff;
518}
519
520static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
521{
522 entry->block = cpu_to_le32(value);
523}
524
525static inline unsigned dx_get_hash(struct dx_entry *entry)
526{
527 return le32_to_cpu(entry->hash);
528}
529
530static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
531{
532 entry->hash = cpu_to_le32(value);
533}
534
535static inline unsigned dx_get_count(struct dx_entry *entries)
536{
537 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
538}
539
540static inline unsigned dx_get_limit(struct dx_entry *entries)
541{
542 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
543}
544
545static inline void dx_set_count(struct dx_entry *entries, unsigned value)
546{
547 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
548}
549
550static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
551{
552 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
553}
554
555static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
556{
557 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
558 EXT4_DIR_REC_LEN(2) - infosize;
559
560 if (ext4_has_metadata_csum(dir->i_sb))
561 entry_space -= sizeof(struct dx_tail);
562 return entry_space / sizeof(struct dx_entry);
563}
564
565static inline unsigned dx_node_limit(struct inode *dir)
566{
567 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568
569 if (ext4_has_metadata_csum(dir->i_sb))
570 entry_space -= sizeof(struct dx_tail);
571 return entry_space / sizeof(struct dx_entry);
572}
573
574/*
575 * Debug
576 */
577#ifdef DX_DEBUG
578static void dx_show_index(char * label, struct dx_entry *entries)
579{
580 int i, n = dx_get_count (entries);
581 printk(KERN_DEBUG "%s index ", label);
582 for (i = 0; i < n; i++) {
583 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
584 0, (unsigned long)dx_get_block(entries + i));
585 }
586 printk("\n");
587}
588
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 ext4_str fname_crypto_str
616 = {.name = NULL, .len = 0};
617 int res = 0;
618
619 name = de->name;
620 len = de->name_len;
621 if (ext4_encrypted_inode(inode))
622 res = ext4_get_encryption_info(dir);
623 if (res) {
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
626 }
627 if (ctx == NULL) {
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
630 de->name_len, &h);
631 printk("%*.s:(U)%x.%u ", len,
632 name, h.hash,
633 (unsigned) ((char *) de
634 - base));
635 } else {
636 /* Directory is encrypted */
637 res = ext4_fname_crypto_alloc_buffer(
638 ctx, de->name_len,
639 &fname_crypto_str);
640 if (res < 0) {
641 printk(KERN_WARNING "Error "
642 "allocating crypto "
643 "buffer--skipping "
644 "crypto\n");
645 ctx = NULL;
646 }
647 res = ext4_fname_disk_to_usr(ctx, NULL, de,
648 &fname_crypto_str);
649 if (res < 0) {
650 printk(KERN_WARNING "Error "
651 "converting filename "
652 "from disk to usr"
653 "\n");
654 name = "??";
655 len = 2;
656 } else {
657 name = fname_crypto_str.name;
658 len = fname_crypto_str.len;
659 }
660 ext4fs_dirhash(de->name, de->name_len,
661 &h);
662 printk("%*.s:(E)%x.%u ", len, name,
663 h.hash, (unsigned) ((char *) de
664 - base));
665 ext4_fname_crypto_free_buffer(
666 &fname_crypto_str);
667 }
668#else
669 int len = de->name_len;
670 char *name = de->name;
671 ext4fs_dirhash(de->name, de->name_len, &h);
672 printk("%*.s:%x.%u ", len, name, h.hash,
673 (unsigned) ((char *) de - base));
674#endif
675 }
676 space += EXT4_DIR_REC_LEN(de->name_len);
677 names++;
678 }
679 de = ext4_next_entry(de, size);
680 }
681 printk("(%i)\n", names);
682 return (struct stats) { names, space, 1 };
683}
684
685struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
686 struct dx_entry *entries, int levels)
687{
688 unsigned blocksize = dir->i_sb->s_blocksize;
689 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
690 unsigned bcount = 0;
691 struct buffer_head *bh;
692 printk("%i indexed blocks...\n", count);
693 for (i = 0; i < count; i++, entries++)
694 {
695 ext4_lblk_t block = dx_get_block(entries);
696 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
697 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
698 struct stats stats;
699 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
700 bh = ext4_bread(NULL,dir, block, 0);
701 if (!bh || IS_ERR(bh))
702 continue;
703 stats = levels?
704 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
705 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
706 bh->b_data, blocksize, 0);
707 names += stats.names;
708 space += stats.space;
709 bcount += stats.bcount;
710 brelse(bh);
711 }
712 if (bcount)
713 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
714 levels ? "" : " ", names, space/bcount,
715 (space/bcount)*100/blocksize);
716 return (struct stats) { names, space, bcount};
717}
718#endif /* DX_DEBUG */
719
720/*
721 * Probe for a directory leaf block to search.
722 *
723 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
724 * error in the directory index, and the caller should fall back to
725 * searching the directory normally. The callers of dx_probe **MUST**
726 * check for this error code, and make sure it never gets reflected
727 * back to userspace.
728 */
729static struct dx_frame *
730dx_probe(struct ext4_filename *fname, struct inode *dir,
731 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
732{
733 unsigned count, indirect;
734 struct dx_entry *at, *entries, *p, *q, *m;
735 struct dx_root *root;
736 struct dx_frame *frame = frame_in;
737 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
738 u32 hash;
739
740 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
741 if (IS_ERR(frame->bh))
742 return (struct dx_frame *) frame->bh;
743
744 root = (struct dx_root *) frame->bh->b_data;
745 if (root->info.hash_version != DX_HASH_TEA &&
746 root->info.hash_version != DX_HASH_HALF_MD4 &&
747 root->info.hash_version != DX_HASH_LEGACY) {
748 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
749 root->info.hash_version);
750 goto fail;
751 }
752 if (fname)
753 hinfo = &fname->hinfo;
754 hinfo->hash_version = root->info.hash_version;
755 if (hinfo->hash_version <= DX_HASH_TEA)
756 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
757 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
758 if (fname && fname_name(fname))
759 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
760 hash = hinfo->hash;
761
762 if (root->info.unused_flags & 1) {
763 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
764 root->info.unused_flags);
765 goto fail;
766 }
767
768 indirect = root->info.indirect_levels;
769 if (indirect > 1) {
770 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
771 root->info.indirect_levels);
772 goto fail;
773 }
774
775 entries = (struct dx_entry *)(((char *)&root->info) +
776 root->info.info_length);
777
778 if (dx_get_limit(entries) != dx_root_limit(dir,
779 root->info.info_length)) {
780 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
781 dx_get_limit(entries),
782 dx_root_limit(dir, root->info.info_length));
783 goto fail;
784 }
785
786 dxtrace(printk("Look up %x", hash));
787 while (1) {
788 count = dx_get_count(entries);
789 if (!count || count > dx_get_limit(entries)) {
790 ext4_warning_inode(dir,
791 "dx entry: count %u beyond limit %u",
792 count, dx_get_limit(entries));
793 goto fail;
794 }
795
796 p = entries + 1;
797 q = entries + count - 1;
798 while (p <= q) {
799 m = p + (q - p) / 2;
800 dxtrace(printk("."));
801 if (dx_get_hash(m) > hash)
802 q = m - 1;
803 else
804 p = m + 1;
805 }
806
807 if (0) { // linear search cross check
808 unsigned n = count - 1;
809 at = entries;
810 while (n--)
811 {
812 dxtrace(printk(","));
813 if (dx_get_hash(++at) > hash)
814 {
815 at--;
816 break;
817 }
818 }
819 assert (at == p - 1);
820 }
821
822 at = p - 1;
823 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
824 dx_get_block(at)));
825 frame->entries = entries;
826 frame->at = at;
827 if (!indirect--)
828 return frame;
829 frame++;
830 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
831 if (IS_ERR(frame->bh)) {
832 ret_err = (struct dx_frame *) frame->bh;
833 frame->bh = NULL;
834 goto fail;
835 }
836 entries = ((struct dx_node *) frame->bh->b_data)->entries;
837
838 if (dx_get_limit(entries) != dx_node_limit(dir)) {
839 ext4_warning_inode(dir,
840 "dx entry: limit %u != node limit %u",
841 dx_get_limit(entries), dx_node_limit(dir));
842 goto fail;
843 }
844 }
845fail:
846 while (frame >= frame_in) {
847 brelse(frame->bh);
848 frame--;
849 }
850
851 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
852 ext4_warning_inode(dir,
853 "Corrupt directory, running e2fsck is recommended");
854 return ret_err;
855}
856
857static void dx_release(struct dx_frame *frames)
858{
859 if (frames[0].bh == NULL)
860 return;
861
862 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
863 brelse(frames[1].bh);
864 brelse(frames[0].bh);
865}
866
867/*
868 * This function increments the frame pointer to search the next leaf
869 * block, and reads in the necessary intervening nodes if the search
870 * should be necessary. Whether or not the search is necessary is
871 * controlled by the hash parameter. If the hash value is even, then
872 * the search is only continued if the next block starts with that
873 * hash value. This is used if we are searching for a specific file.
874 *
875 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
876 *
877 * This function returns 1 if the caller should continue to search,
878 * or 0 if it should not. If there is an error reading one of the
879 * index blocks, it will a negative error code.
880 *
881 * If start_hash is non-null, it will be filled in with the starting
882 * hash of the next page.
883 */
884static int ext4_htree_next_block(struct inode *dir, __u32 hash,
885 struct dx_frame *frame,
886 struct dx_frame *frames,
887 __u32 *start_hash)
888{
889 struct dx_frame *p;
890 struct buffer_head *bh;
891 int num_frames = 0;
892 __u32 bhash;
893
894 p = frame;
895 /*
896 * Find the next leaf page by incrementing the frame pointer.
897 * If we run out of entries in the interior node, loop around and
898 * increment pointer in the parent node. When we break out of
899 * this loop, num_frames indicates the number of interior
900 * nodes need to be read.
901 */
902 while (1) {
903 if (++(p->at) < p->entries + dx_get_count(p->entries))
904 break;
905 if (p == frames)
906 return 0;
907 num_frames++;
908 p--;
909 }
910
911 /*
912 * If the hash is 1, then continue only if the next page has a
913 * continuation hash of any value. This is used for readdir
914 * handling. Otherwise, check to see if the hash matches the
915 * desired contiuation hash. If it doesn't, return since
916 * there's no point to read in the successive index pages.
917 */
918 bhash = dx_get_hash(p->at);
919 if (start_hash)
920 *start_hash = bhash;
921 if ((hash & 1) == 0) {
922 if ((bhash & ~1) != hash)
923 return 0;
924 }
925 /*
926 * If the hash is HASH_NB_ALWAYS, we always go to the next
927 * block so no check is necessary
928 */
929 while (num_frames--) {
930 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
931 if (IS_ERR(bh))
932 return PTR_ERR(bh);
933 p++;
934 brelse(p->bh);
935 p->bh = bh;
936 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
937 }
938 return 1;
939}
940
941
942/*
943 * This function fills a red-black tree with information from a
944 * directory block. It returns the number directory entries loaded
945 * into the tree. If there is an error it is returned in err.
946 */
947static int htree_dirblock_to_tree(struct file *dir_file,
948 struct inode *dir, ext4_lblk_t block,
949 struct dx_hash_info *hinfo,
950 __u32 start_hash, __u32 start_minor_hash)
951{
952 struct buffer_head *bh;
953 struct ext4_dir_entry_2 *de, *top;
954 int err = 0, count = 0;
955 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
956
957 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
958 (unsigned long)block));
959 bh = ext4_read_dirblock(dir, block, DIRENT);
960 if (IS_ERR(bh))
961 return PTR_ERR(bh);
962
963 de = (struct ext4_dir_entry_2 *) bh->b_data;
964 top = (struct ext4_dir_entry_2 *) ((char *) de +
965 dir->i_sb->s_blocksize -
966 EXT4_DIR_REC_LEN(0));
967#ifdef CONFIG_EXT4_FS_ENCRYPTION
968 /* Check if the directory is encrypted */
969 if (ext4_encrypted_inode(dir)) {
970 err = ext4_get_encryption_info(dir);
971 if (err < 0) {
972 brelse(bh);
973 return err;
974 }
975 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
976 &fname_crypto_str);
977 if (err < 0) {
978 brelse(bh);
979 return err;
980 }
981 }
982#endif
983 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
984 if (ext4_check_dir_entry(dir, NULL, de, bh,
985 bh->b_data, bh->b_size,
986 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
987 + ((char *)de - bh->b_data))) {
988 /* silently ignore the rest of the block */
989 break;
990 }
991 ext4fs_dirhash(de->name, de->name_len, hinfo);
992 if ((hinfo->hash < start_hash) ||
993 ((hinfo->hash == start_hash) &&
994 (hinfo->minor_hash < start_minor_hash)))
995 continue;
996 if (de->inode == 0)
997 continue;
998 if (!ext4_encrypted_inode(dir)) {
999 tmp_str.name = de->name;
1000 tmp_str.len = de->name_len;
1001 err = ext4_htree_store_dirent(dir_file,
1002 hinfo->hash, hinfo->minor_hash, de,
1003 &tmp_str);
1004 } else {
1005 int save_len = fname_crypto_str.len;
1006
1007 /* Directory is encrypted */
1008 err = ext4_fname_disk_to_usr(dir, hinfo, de,
1009 &fname_crypto_str);
1010 if (err < 0) {
1011 count = err;
1012 goto errout;
1013 }
1014 err = ext4_htree_store_dirent(dir_file,
1015 hinfo->hash, hinfo->minor_hash, de,
1016 &fname_crypto_str);
1017 fname_crypto_str.len = save_len;
1018 }
1019 if (err != 0) {
1020 count = err;
1021 goto errout;
1022 }
1023 count++;
1024 }
1025errout:
1026 brelse(bh);
1027#ifdef CONFIG_EXT4_FS_ENCRYPTION
1028 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1029#endif
1030 return count;
1031}
1032
1033
1034/*
1035 * This function fills a red-black tree with information from a
1036 * directory. We start scanning the directory in hash order, starting
1037 * at start_hash and start_minor_hash.
1038 *
1039 * This function returns the number of entries inserted into the tree,
1040 * or a negative error code.
1041 */
1042int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1043 __u32 start_minor_hash, __u32 *next_hash)
1044{
1045 struct dx_hash_info hinfo;
1046 struct ext4_dir_entry_2 *de;
1047 struct dx_frame frames[2], *frame;
1048 struct inode *dir;
1049 ext4_lblk_t block;
1050 int count = 0;
1051 int ret, err;
1052 __u32 hashval;
1053 struct ext4_str tmp_str;
1054
1055 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1056 start_hash, start_minor_hash));
1057 dir = file_inode(dir_file);
1058 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1059 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1060 if (hinfo.hash_version <= DX_HASH_TEA)
1061 hinfo.hash_version +=
1062 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1063 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1064 if (ext4_has_inline_data(dir)) {
1065 int has_inline_data = 1;
1066 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1067 &hinfo, start_hash,
1068 start_minor_hash,
1069 &has_inline_data);
1070 if (has_inline_data) {
1071 *next_hash = ~0;
1072 return count;
1073 }
1074 }
1075 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1076 start_hash, start_minor_hash);
1077 *next_hash = ~0;
1078 return count;
1079 }
1080 hinfo.hash = start_hash;
1081 hinfo.minor_hash = 0;
1082 frame = dx_probe(NULL, dir, &hinfo, frames);
1083 if (IS_ERR(frame))
1084 return PTR_ERR(frame);
1085
1086 /* Add '.' and '..' from the htree header */
1087 if (!start_hash && !start_minor_hash) {
1088 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1089 tmp_str.name = de->name;
1090 tmp_str.len = de->name_len;
1091 err = ext4_htree_store_dirent(dir_file, 0, 0,
1092 de, &tmp_str);
1093 if (err != 0)
1094 goto errout;
1095 count++;
1096 }
1097 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1098 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1099 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1100 tmp_str.name = de->name;
1101 tmp_str.len = de->name_len;
1102 err = ext4_htree_store_dirent(dir_file, 2, 0,
1103 de, &tmp_str);
1104 if (err != 0)
1105 goto errout;
1106 count++;
1107 }
1108
1109 while (1) {
1110 block = dx_get_block(frame->at);
1111 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1112 start_hash, start_minor_hash);
1113 if (ret < 0) {
1114 err = ret;
1115 goto errout;
1116 }
1117 count += ret;
1118 hashval = ~0;
1119 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1120 frame, frames, &hashval);
1121 *next_hash = hashval;
1122 if (ret < 0) {
1123 err = ret;
1124 goto errout;
1125 }
1126 /*
1127 * Stop if: (a) there are no more entries, or
1128 * (b) we have inserted at least one entry and the
1129 * next hash value is not a continuation
1130 */
1131 if ((ret == 0) ||
1132 (count && ((hashval & 1) == 0)))
1133 break;
1134 }
1135 dx_release(frames);
1136 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1137 "next hash: %x\n", count, *next_hash));
1138 return count;
1139errout:
1140 dx_release(frames);
1141 return (err);
1142}
1143
1144static inline int search_dirblock(struct buffer_head *bh,
1145 struct inode *dir,
1146 struct ext4_filename *fname,
1147 const struct qstr *d_name,
1148 unsigned int offset,
1149 struct ext4_dir_entry_2 **res_dir)
1150{
1151 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1152 fname, d_name, offset, res_dir);
1153}
1154
1155/*
1156 * Directory block splitting, compacting
1157 */
1158
1159/*
1160 * Create map of hash values, offsets, and sizes, stored at end of block.
1161 * Returns number of entries mapped.
1162 */
1163static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1164 unsigned blocksize, struct dx_hash_info *hinfo,
1165 struct dx_map_entry *map_tail)
1166{
1167 int count = 0;
1168 char *base = (char *) de;
1169 struct dx_hash_info h = *hinfo;
1170
1171 while ((char *) de < base + blocksize) {
1172 if (de->name_len && de->inode) {
1173 ext4fs_dirhash(de->name, de->name_len, &h);
1174 map_tail--;
1175 map_tail->hash = h.hash;
1176 map_tail->offs = ((char *) de - base)>>2;
1177 map_tail->size = le16_to_cpu(de->rec_len);
1178 count++;
1179 cond_resched();
1180 }
1181 /* XXX: do we need to check rec_len == 0 case? -Chris */
1182 de = ext4_next_entry(de, blocksize);
1183 }
1184 return count;
1185}
1186
1187/* Sort map by hash value */
1188static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1189{
1190 struct dx_map_entry *p, *q, *top = map + count - 1;
1191 int more;
1192 /* Combsort until bubble sort doesn't suck */
1193 while (count > 2) {
1194 count = count*10/13;
1195 if (count - 9 < 2) /* 9, 10 -> 11 */
1196 count = 11;
1197 for (p = top, q = p - count; q >= map; p--, q--)
1198 if (p->hash < q->hash)
1199 swap(*p, *q);
1200 }
1201 /* Garden variety bubble sort */
1202 do {
1203 more = 0;
1204 q = top;
1205 while (q-- > map) {
1206 if (q[1].hash >= q[0].hash)
1207 continue;
1208 swap(*(q+1), *q);
1209 more = 1;
1210 }
1211 } while(more);
1212}
1213
1214static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1215{
1216 struct dx_entry *entries = frame->entries;
1217 struct dx_entry *old = frame->at, *new = old + 1;
1218 int count = dx_get_count(entries);
1219
1220 assert(count < dx_get_limit(entries));
1221 assert(old < entries + count);
1222 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1223 dx_set_hash(new, hash);
1224 dx_set_block(new, block);
1225 dx_set_count(entries, count + 1);
1226}
1227
1228/*
1229 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1230 *
1231 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1232 * `de != NULL' is guaranteed by caller.
1233 */
1234static inline int ext4_match(struct ext4_filename *fname,
1235 struct ext4_dir_entry_2 *de)
1236{
1237 const void *name = fname_name(fname);
1238 u32 len = fname_len(fname);
1239
1240 if (!de->inode)
1241 return 0;
1242
1243#ifdef CONFIG_EXT4_FS_ENCRYPTION
1244 if (unlikely(!name)) {
1245 if (fname->usr_fname->name[0] == '_') {
1246 int ret;
1247 if (de->name_len < 16)
1248 return 0;
1249 ret = memcmp(de->name + de->name_len - 16,
1250 fname->crypto_buf.name + 8, 16);
1251 return (ret == 0) ? 1 : 0;
1252 }
1253 name = fname->crypto_buf.name;
1254 len = fname->crypto_buf.len;
1255 }
1256#endif
1257 if (de->name_len != len)
1258 return 0;
1259 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1260}
1261
1262/*
1263 * Returns 0 if not found, -1 on failure, and 1 on success
1264 */
1265int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1266 struct inode *dir, struct ext4_filename *fname,
1267 const struct qstr *d_name,
1268 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1269{
1270 struct ext4_dir_entry_2 * de;
1271 char * dlimit;
1272 int de_len;
1273 int res;
1274
1275 de = (struct ext4_dir_entry_2 *)search_buf;
1276 dlimit = search_buf + buf_size;
1277 while ((char *) de < dlimit) {
1278 /* this code is executed quadratically often */
1279 /* do minimal checking `by hand' */
1280 if ((char *) de + de->name_len <= dlimit) {
1281 res = ext4_match(fname, de);
1282 if (res < 0) {
1283 res = -1;
1284 goto return_result;
1285 }
1286 if (res > 0) {
1287 /* found a match - just to be sure, do
1288 * a full check */
1289 if (ext4_check_dir_entry(dir, NULL, de, bh,
1290 bh->b_data,
1291 bh->b_size, offset)) {
1292 res = -1;
1293 goto return_result;
1294 }
1295 *res_dir = de;
1296 res = 1;
1297 goto return_result;
1298 }
1299
1300 }
1301 /* prevent looping on a bad block */
1302 de_len = ext4_rec_len_from_disk(de->rec_len,
1303 dir->i_sb->s_blocksize);
1304 if (de_len <= 0) {
1305 res = -1;
1306 goto return_result;
1307 }
1308 offset += de_len;
1309 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1310 }
1311
1312 res = 0;
1313return_result:
1314 return res;
1315}
1316
1317static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1318 struct ext4_dir_entry *de)
1319{
1320 struct super_block *sb = dir->i_sb;
1321
1322 if (!is_dx(dir))
1323 return 0;
1324 if (block == 0)
1325 return 1;
1326 if (de->inode == 0 &&
1327 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1328 sb->s_blocksize)
1329 return 1;
1330 return 0;
1331}
1332
1333/*
1334 * ext4_find_entry()
1335 *
1336 * finds an entry in the specified directory with the wanted name. It
1337 * returns the cache buffer in which the entry was found, and the entry
1338 * itself (as a parameter - res_dir). It does NOT read the inode of the
1339 * entry - you'll have to do that yourself if you want to.
1340 *
1341 * The returned buffer_head has ->b_count elevated. The caller is expected
1342 * to brelse() it when appropriate.
1343 */
1344static struct buffer_head * ext4_find_entry (struct inode *dir,
1345 const struct qstr *d_name,
1346 struct ext4_dir_entry_2 **res_dir,
1347 int *inlined)
1348{
1349 struct super_block *sb;
1350 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1351 struct buffer_head *bh, *ret = NULL;
1352 ext4_lblk_t start, block, b;
1353 const u8 *name = d_name->name;
1354 int ra_max = 0; /* Number of bh's in the readahead
1355 buffer, bh_use[] */
1356 int ra_ptr = 0; /* Current index into readahead
1357 buffer */
1358 int num = 0;
1359 ext4_lblk_t nblocks;
1360 int i, namelen, retval;
1361 struct ext4_filename fname;
1362
1363 *res_dir = NULL;
1364 sb = dir->i_sb;
1365 namelen = d_name->len;
1366 if (namelen > EXT4_NAME_LEN)
1367 return NULL;
1368
1369 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1370 if (retval)
1371 return ERR_PTR(retval);
1372
1373 if (ext4_has_inline_data(dir)) {
1374 int has_inline_data = 1;
1375 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1376 &has_inline_data);
1377 if (has_inline_data) {
1378 if (inlined)
1379 *inlined = 1;
1380 goto cleanup_and_exit;
1381 }
1382 }
1383
1384 if ((namelen <= 2) && (name[0] == '.') &&
1385 (name[1] == '.' || name[1] == '\0')) {
1386 /*
1387 * "." or ".." will only be in the first block
1388 * NFS may look up ".."; "." should be handled by the VFS
1389 */
1390 block = start = 0;
1391 nblocks = 1;
1392 goto restart;
1393 }
1394 if (is_dx(dir)) {
1395 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1396 /*
1397 * On success, or if the error was file not found,
1398 * return. Otherwise, fall back to doing a search the
1399 * old fashioned way.
1400 */
1401 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1402 goto cleanup_and_exit;
1403 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1404 "falling back\n"));
1405 }
1406 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1407 start = EXT4_I(dir)->i_dir_start_lookup;
1408 if (start >= nblocks)
1409 start = 0;
1410 block = start;
1411restart:
1412 do {
1413 /*
1414 * We deal with the read-ahead logic here.
1415 */
1416 if (ra_ptr >= ra_max) {
1417 /* Refill the readahead buffer */
1418 ra_ptr = 0;
1419 b = block;
1420 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1421 /*
1422 * Terminate if we reach the end of the
1423 * directory and must wrap, or if our
1424 * search has finished at this block.
1425 */
1426 if (b >= nblocks || (num && block == start)) {
1427 bh_use[ra_max] = NULL;
1428 break;
1429 }
1430 num++;
1431 bh = ext4_getblk(NULL, dir, b++, 0);
1432 if (IS_ERR(bh)) {
1433 if (ra_max == 0) {
1434 ret = bh;
1435 goto cleanup_and_exit;
1436 }
1437 break;
1438 }
1439 bh_use[ra_max] = bh;
1440 if (bh)
1441 ll_rw_block(READ | REQ_META | REQ_PRIO,
1442 1, &bh);
1443 }
1444 }
1445 if ((bh = bh_use[ra_ptr++]) == NULL)
1446 goto next;
1447 wait_on_buffer(bh);
1448 if (!buffer_uptodate(bh)) {
1449 /* read error, skip block & hope for the best */
1450 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1451 (unsigned long) block);
1452 brelse(bh);
1453 goto next;
1454 }
1455 if (!buffer_verified(bh) &&
1456 !is_dx_internal_node(dir, block,
1457 (struct ext4_dir_entry *)bh->b_data) &&
1458 !ext4_dirent_csum_verify(dir,
1459 (struct ext4_dir_entry *)bh->b_data)) {
1460 EXT4_ERROR_INODE(dir, "checksumming directory "
1461 "block %lu", (unsigned long)block);
1462 brelse(bh);
1463 goto next;
1464 }
1465 set_buffer_verified(bh);
1466 i = search_dirblock(bh, dir, &fname, d_name,
1467 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1468 if (i == 1) {
1469 EXT4_I(dir)->i_dir_start_lookup = block;
1470 ret = bh;
1471 goto cleanup_and_exit;
1472 } else {
1473 brelse(bh);
1474 if (i < 0)
1475 goto cleanup_and_exit;
1476 }
1477 next:
1478 if (++block >= nblocks)
1479 block = 0;
1480 } while (block != start);
1481
1482 /*
1483 * If the directory has grown while we were searching, then
1484 * search the last part of the directory before giving up.
1485 */
1486 block = nblocks;
1487 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1488 if (block < nblocks) {
1489 start = 0;
1490 goto restart;
1491 }
1492
1493cleanup_and_exit:
1494 /* Clean up the read-ahead blocks */
1495 for (; ra_ptr < ra_max; ra_ptr++)
1496 brelse(bh_use[ra_ptr]);
1497 ext4_fname_free_filename(&fname);
1498 return ret;
1499}
1500
1501static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1502 struct ext4_filename *fname,
1503 struct ext4_dir_entry_2 **res_dir)
1504{
1505 struct super_block * sb = dir->i_sb;
1506 struct dx_frame frames[2], *frame;
1507 const struct qstr *d_name = fname->usr_fname;
1508 struct buffer_head *bh;
1509 ext4_lblk_t block;
1510 int retval;
1511
1512#ifdef CONFIG_EXT4_FS_ENCRYPTION
1513 *res_dir = NULL;
1514#endif
1515 frame = dx_probe(fname, dir, NULL, frames);
1516 if (IS_ERR(frame))
1517 return (struct buffer_head *) frame;
1518 do {
1519 block = dx_get_block(frame->at);
1520 bh = ext4_read_dirblock(dir, block, DIRENT);
1521 if (IS_ERR(bh))
1522 goto errout;
1523
1524 retval = search_dirblock(bh, dir, fname, d_name,
1525 block << EXT4_BLOCK_SIZE_BITS(sb),
1526 res_dir);
1527 if (retval == 1)
1528 goto success;
1529 brelse(bh);
1530 if (retval == -1) {
1531 bh = ERR_PTR(ERR_BAD_DX_DIR);
1532 goto errout;
1533 }
1534
1535 /* Check to see if we should continue to search */
1536 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1537 frames, NULL);
1538 if (retval < 0) {
1539 ext4_warning_inode(dir,
1540 "error %d reading directory index block",
1541 retval);
1542 bh = ERR_PTR(retval);
1543 goto errout;
1544 }
1545 } while (retval == 1);
1546
1547 bh = NULL;
1548errout:
1549 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1550success:
1551 dx_release(frames);
1552 return bh;
1553}
1554
1555static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1556{
1557 struct inode *inode;
1558 struct ext4_dir_entry_2 *de;
1559 struct buffer_head *bh;
1560
1561 if (ext4_encrypted_inode(dir)) {
1562 int res = ext4_get_encryption_info(dir);
1563
1564 /*
1565 * This should be a properly defined flag for
1566 * dentry->d_flags when we uplift this to the VFS.
1567 * d_fsdata is set to (void *) 1 if if the dentry is
1568 * created while the directory was encrypted and we
1569 * don't have access to the key.
1570 */
1571 dentry->d_fsdata = NULL;
1572 if (ext4_encryption_info(dir))
1573 dentry->d_fsdata = (void *) 1;
1574 d_set_d_op(dentry, &ext4_encrypted_d_ops);
1575 if (res && res != -ENOKEY)
1576 return ERR_PTR(res);
1577 }
1578
1579 if (dentry->d_name.len > EXT4_NAME_LEN)
1580 return ERR_PTR(-ENAMETOOLONG);
1581
1582 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1583 if (IS_ERR(bh))
1584 return (struct dentry *) bh;
1585 inode = NULL;
1586 if (bh) {
1587 __u32 ino = le32_to_cpu(de->inode);
1588 brelse(bh);
1589 if (!ext4_valid_inum(dir->i_sb, ino)) {
1590 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1591 return ERR_PTR(-EFSCORRUPTED);
1592 }
1593 if (unlikely(ino == dir->i_ino)) {
1594 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1595 dentry);
1596 return ERR_PTR(-EFSCORRUPTED);
1597 }
1598 inode = ext4_iget_normal(dir->i_sb, ino);
1599 if (inode == ERR_PTR(-ESTALE)) {
1600 EXT4_ERROR_INODE(dir,
1601 "deleted inode referenced: %u",
1602 ino);
1603 return ERR_PTR(-EFSCORRUPTED);
1604 }
1605 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1606 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1607 !ext4_is_child_context_consistent_with_parent(dir,
1608 inode)) {
1609 int nokey = ext4_encrypted_inode(inode) &&
1610 !ext4_encryption_info(inode);
1611
1612 iput(inode);
1613 if (nokey)
1614 return ERR_PTR(-ENOKEY);
1615 ext4_warning(inode->i_sb,
1616 "Inconsistent encryption contexts: %lu/%lu\n",
1617 (unsigned long) dir->i_ino,
1618 (unsigned long) inode->i_ino);
1619 return ERR_PTR(-EPERM);
1620 }
1621 }
1622 return d_splice_alias(inode, dentry);
1623}
1624
1625
1626struct dentry *ext4_get_parent(struct dentry *child)
1627{
1628 __u32 ino;
1629 static const struct qstr dotdot = QSTR_INIT("..", 2);
1630 struct ext4_dir_entry_2 * de;
1631 struct buffer_head *bh;
1632
1633 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1634 if (IS_ERR(bh))
1635 return (struct dentry *) bh;
1636 if (!bh)
1637 return ERR_PTR(-ENOENT);
1638 ino = le32_to_cpu(de->inode);
1639 brelse(bh);
1640
1641 if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1642 EXT4_ERROR_INODE(d_inode(child),
1643 "bad parent inode number: %u", ino);
1644 return ERR_PTR(-EFSCORRUPTED);
1645 }
1646
1647 return d_obtain_alias(ext4_iget_normal(d_inode(child)->i_sb, ino));
1648}
1649
1650/*
1651 * Move count entries from end of map between two memory locations.
1652 * Returns pointer to last entry moved.
1653 */
1654static struct ext4_dir_entry_2 *
1655dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1656 unsigned blocksize)
1657{
1658 unsigned rec_len = 0;
1659
1660 while (count--) {
1661 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1662 (from + (map->offs<<2));
1663 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1664 memcpy (to, de, rec_len);
1665 ((struct ext4_dir_entry_2 *) to)->rec_len =
1666 ext4_rec_len_to_disk(rec_len, blocksize);
1667 de->inode = 0;
1668 map++;
1669 to += rec_len;
1670 }
1671 return (struct ext4_dir_entry_2 *) (to - rec_len);
1672}
1673
1674/*
1675 * Compact each dir entry in the range to the minimal rec_len.
1676 * Returns pointer to last entry in range.
1677 */
1678static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1679{
1680 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1681 unsigned rec_len = 0;
1682
1683 prev = to = de;
1684 while ((char*)de < base + blocksize) {
1685 next = ext4_next_entry(de, blocksize);
1686 if (de->inode && de->name_len) {
1687 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1688 if (de > to)
1689 memmove(to, de, rec_len);
1690 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1691 prev = to;
1692 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1693 }
1694 de = next;
1695 }
1696 return prev;
1697}
1698
1699/*
1700 * Split a full leaf block to make room for a new dir entry.
1701 * Allocate a new block, and move entries so that they are approx. equally full.
1702 * Returns pointer to de in block into which the new entry will be inserted.
1703 */
1704static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1705 struct buffer_head **bh,struct dx_frame *frame,
1706 struct dx_hash_info *hinfo)
1707{
1708 unsigned blocksize = dir->i_sb->s_blocksize;
1709 unsigned count, continued;
1710 struct buffer_head *bh2;
1711 ext4_lblk_t newblock;
1712 u32 hash2;
1713 struct dx_map_entry *map;
1714 char *data1 = (*bh)->b_data, *data2;
1715 unsigned split, move, size;
1716 struct ext4_dir_entry_2 *de = NULL, *de2;
1717 struct ext4_dir_entry_tail *t;
1718 int csum_size = 0;
1719 int err = 0, i;
1720
1721 if (ext4_has_metadata_csum(dir->i_sb))
1722 csum_size = sizeof(struct ext4_dir_entry_tail);
1723
1724 bh2 = ext4_append(handle, dir, &newblock);
1725 if (IS_ERR(bh2)) {
1726 brelse(*bh);
1727 *bh = NULL;
1728 return (struct ext4_dir_entry_2 *) bh2;
1729 }
1730
1731 BUFFER_TRACE(*bh, "get_write_access");
1732 err = ext4_journal_get_write_access(handle, *bh);
1733 if (err)
1734 goto journal_error;
1735
1736 BUFFER_TRACE(frame->bh, "get_write_access");
1737 err = ext4_journal_get_write_access(handle, frame->bh);
1738 if (err)
1739 goto journal_error;
1740
1741 data2 = bh2->b_data;
1742
1743 /* create map in the end of data2 block */
1744 map = (struct dx_map_entry *) (data2 + blocksize);
1745 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1746 blocksize, hinfo, map);
1747 map -= count;
1748 dx_sort_map(map, count);
1749 /* Split the existing block in the middle, size-wise */
1750 size = 0;
1751 move = 0;
1752 for (i = count-1; i >= 0; i--) {
1753 /* is more than half of this entry in 2nd half of the block? */
1754 if (size + map[i].size/2 > blocksize/2)
1755 break;
1756 size += map[i].size;
1757 move++;
1758 }
1759 /* map index at which we will split */
1760 split = count - move;
1761 hash2 = map[split].hash;
1762 continued = hash2 == map[split - 1].hash;
1763 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1764 (unsigned long)dx_get_block(frame->at),
1765 hash2, split, count-split));
1766
1767 /* Fancy dance to stay within two buffers */
1768 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1769 blocksize);
1770 de = dx_pack_dirents(data1, blocksize);
1771 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1772 (char *) de,
1773 blocksize);
1774 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1775 (char *) de2,
1776 blocksize);
1777 if (csum_size) {
1778 t = EXT4_DIRENT_TAIL(data2, blocksize);
1779 initialize_dirent_tail(t, blocksize);
1780
1781 t = EXT4_DIRENT_TAIL(data1, blocksize);
1782 initialize_dirent_tail(t, blocksize);
1783 }
1784
1785 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1786 blocksize, 1));
1787 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1788 blocksize, 1));
1789
1790 /* Which block gets the new entry? */
1791 if (hinfo->hash >= hash2) {
1792 swap(*bh, bh2);
1793 de = de2;
1794 }
1795 dx_insert_block(frame, hash2 + continued, newblock);
1796 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1797 if (err)
1798 goto journal_error;
1799 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1800 if (err)
1801 goto journal_error;
1802 brelse(bh2);
1803 dxtrace(dx_show_index("frame", frame->entries));
1804 return de;
1805
1806journal_error:
1807 brelse(*bh);
1808 brelse(bh2);
1809 *bh = NULL;
1810 ext4_std_error(dir->i_sb, err);
1811 return ERR_PTR(err);
1812}
1813
1814int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1815 struct buffer_head *bh,
1816 void *buf, int buf_size,
1817 struct ext4_filename *fname,
1818 struct ext4_dir_entry_2 **dest_de)
1819{
1820 struct ext4_dir_entry_2 *de;
1821 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1822 int nlen, rlen;
1823 unsigned int offset = 0;
1824 char *top;
1825 int res;
1826
1827 de = (struct ext4_dir_entry_2 *)buf;
1828 top = buf + buf_size - reclen;
1829 while ((char *) de <= top) {
1830 if (ext4_check_dir_entry(dir, NULL, de, bh,
1831 buf, buf_size, offset)) {
1832 res = -EFSCORRUPTED;
1833 goto return_result;
1834 }
1835 /* Provide crypto context and crypto buffer to ext4 match */
1836 res = ext4_match(fname, de);
1837 if (res < 0)
1838 goto return_result;
1839 if (res > 0) {
1840 res = -EEXIST;
1841 goto return_result;
1842 }
1843 nlen = EXT4_DIR_REC_LEN(de->name_len);
1844 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1845 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1846 break;
1847 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1848 offset += rlen;
1849 }
1850
1851 if ((char *) de > top)
1852 res = -ENOSPC;
1853 else {
1854 *dest_de = de;
1855 res = 0;
1856 }
1857return_result:
1858 return res;
1859}
1860
1861int ext4_insert_dentry(struct inode *dir,
1862 struct inode *inode,
1863 struct ext4_dir_entry_2 *de,
1864 int buf_size,
1865 struct ext4_filename *fname)
1866{
1867
1868 int nlen, rlen;
1869
1870 nlen = EXT4_DIR_REC_LEN(de->name_len);
1871 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1872 if (de->inode) {
1873 struct ext4_dir_entry_2 *de1 =
1874 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1875 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1876 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1877 de = de1;
1878 }
1879 de->file_type = EXT4_FT_UNKNOWN;
1880 de->inode = cpu_to_le32(inode->i_ino);
1881 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1882 de->name_len = fname_len(fname);
1883 memcpy(de->name, fname_name(fname), fname_len(fname));
1884 return 0;
1885}
1886
1887/*
1888 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1889 * it points to a directory entry which is guaranteed to be large
1890 * enough for new directory entry. If de is NULL, then
1891 * add_dirent_to_buf will attempt search the directory block for
1892 * space. It will return -ENOSPC if no space is available, and -EIO
1893 * and -EEXIST if directory entry already exists.
1894 */
1895static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1896 struct inode *dir,
1897 struct inode *inode, struct ext4_dir_entry_2 *de,
1898 struct buffer_head *bh)
1899{
1900 unsigned int blocksize = dir->i_sb->s_blocksize;
1901 int csum_size = 0;
1902 int err;
1903
1904 if (ext4_has_metadata_csum(inode->i_sb))
1905 csum_size = sizeof(struct ext4_dir_entry_tail);
1906
1907 if (!de) {
1908 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1909 blocksize - csum_size, fname, &de);
1910 if (err)
1911 return err;
1912 }
1913 BUFFER_TRACE(bh, "get_write_access");
1914 err = ext4_journal_get_write_access(handle, bh);
1915 if (err) {
1916 ext4_std_error(dir->i_sb, err);
1917 return err;
1918 }
1919
1920 /* By now the buffer is marked for journaling. Due to crypto operations,
1921 * the following function call may fail */
1922 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1923 if (err < 0)
1924 return err;
1925
1926 /*
1927 * XXX shouldn't update any times until successful
1928 * completion of syscall, but too many callers depend
1929 * on this.
1930 *
1931 * XXX similarly, too many callers depend on
1932 * ext4_new_inode() setting the times, but error
1933 * recovery deletes the inode, so the worst that can
1934 * happen is that the times are slightly out of date
1935 * and/or different from the directory change time.
1936 */
1937 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1938 ext4_update_dx_flag(dir);
1939 dir->i_version++;
1940 ext4_mark_inode_dirty(handle, dir);
1941 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1942 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1943 if (err)
1944 ext4_std_error(dir->i_sb, err);
1945 return 0;
1946}
1947
1948/*
1949 * This converts a one block unindexed directory to a 3 block indexed
1950 * directory, and adds the dentry to the indexed directory.
1951 */
1952static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1953 struct inode *dir,
1954 struct inode *inode, struct buffer_head *bh)
1955{
1956 struct buffer_head *bh2;
1957 struct dx_root *root;
1958 struct dx_frame frames[2], *frame;
1959 struct dx_entry *entries;
1960 struct ext4_dir_entry_2 *de, *de2;
1961 struct ext4_dir_entry_tail *t;
1962 char *data1, *top;
1963 unsigned len;
1964 int retval;
1965 unsigned blocksize;
1966 ext4_lblk_t block;
1967 struct fake_dirent *fde;
1968 int csum_size = 0;
1969
1970 if (ext4_has_metadata_csum(inode->i_sb))
1971 csum_size = sizeof(struct ext4_dir_entry_tail);
1972
1973 blocksize = dir->i_sb->s_blocksize;
1974 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1975 BUFFER_TRACE(bh, "get_write_access");
1976 retval = ext4_journal_get_write_access(handle, bh);
1977 if (retval) {
1978 ext4_std_error(dir->i_sb, retval);
1979 brelse(bh);
1980 return retval;
1981 }
1982 root = (struct dx_root *) bh->b_data;
1983
1984 /* The 0th block becomes the root, move the dirents out */
1985 fde = &root->dotdot;
1986 de = (struct ext4_dir_entry_2 *)((char *)fde +
1987 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1988 if ((char *) de >= (((char *) root) + blocksize)) {
1989 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1990 brelse(bh);
1991 return -EFSCORRUPTED;
1992 }
1993 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1994
1995 /* Allocate new block for the 0th block's dirents */
1996 bh2 = ext4_append(handle, dir, &block);
1997 if (IS_ERR(bh2)) {
1998 brelse(bh);
1999 return PTR_ERR(bh2);
2000 }
2001 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2002 data1 = bh2->b_data;
2003
2004 memcpy (data1, de, len);
2005 de = (struct ext4_dir_entry_2 *) data1;
2006 top = data1 + len;
2007 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2008 de = de2;
2009 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2010 (char *) de,
2011 blocksize);
2012
2013 if (csum_size) {
2014 t = EXT4_DIRENT_TAIL(data1, blocksize);
2015 initialize_dirent_tail(t, blocksize);
2016 }
2017
2018 /* Initialize the root; the dot dirents already exist */
2019 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2020 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2021 blocksize);
2022 memset (&root->info, 0, sizeof(root->info));
2023 root->info.info_length = sizeof(root->info);
2024 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2025 entries = root->entries;
2026 dx_set_block(entries, 1);
2027 dx_set_count(entries, 1);
2028 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2029
2030 /* Initialize as for dx_probe */
2031 fname->hinfo.hash_version = root->info.hash_version;
2032 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2033 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2034 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2035 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2036
2037 memset(frames, 0, sizeof(frames));
2038 frame = frames;
2039 frame->entries = entries;
2040 frame->at = entries;
2041 frame->bh = bh;
2042 bh = bh2;
2043
2044 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2045 if (retval)
2046 goto out_frames;
2047 retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2048 if (retval)
2049 goto out_frames;
2050
2051 de = do_split(handle,dir, &bh, frame, &fname->hinfo);
2052 if (IS_ERR(de)) {
2053 retval = PTR_ERR(de);
2054 goto out_frames;
2055 }
2056 dx_release(frames);
2057
2058 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2059 brelse(bh);
2060 return retval;
2061out_frames:
2062 /*
2063 * Even if the block split failed, we have to properly write
2064 * out all the changes we did so far. Otherwise we can end up
2065 * with corrupted filesystem.
2066 */
2067 ext4_mark_inode_dirty(handle, dir);
2068 dx_release(frames);
2069 return retval;
2070}
2071
2072/*
2073 * ext4_add_entry()
2074 *
2075 * adds a file entry to the specified directory, using the same
2076 * semantics as ext4_find_entry(). It returns NULL if it failed.
2077 *
2078 * NOTE!! The inode part of 'de' is left at 0 - which means you
2079 * may not sleep between calling this and putting something into
2080 * the entry, as someone else might have used it while you slept.
2081 */
2082static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2083 struct inode *inode)
2084{
2085 struct inode *dir = d_inode(dentry->d_parent);
2086 struct buffer_head *bh = NULL;
2087 struct ext4_dir_entry_2 *de;
2088 struct ext4_dir_entry_tail *t;
2089 struct super_block *sb;
2090 struct ext4_filename fname;
2091 int retval;
2092 int dx_fallback=0;
2093 unsigned blocksize;
2094 ext4_lblk_t block, blocks;
2095 int csum_size = 0;
2096
2097 if (ext4_has_metadata_csum(inode->i_sb))
2098 csum_size = sizeof(struct ext4_dir_entry_tail);
2099
2100 sb = dir->i_sb;
2101 blocksize = sb->s_blocksize;
2102 if (!dentry->d_name.len)
2103 return -EINVAL;
2104
2105 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2106 if (retval)
2107 return retval;
2108
2109 if (ext4_has_inline_data(dir)) {
2110 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2111 if (retval < 0)
2112 goto out;
2113 if (retval == 1) {
2114 retval = 0;
2115 goto out;
2116 }
2117 }
2118
2119 if (is_dx(dir)) {
2120 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2121 if (!retval || (retval != ERR_BAD_DX_DIR))
2122 goto out;
2123 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2124 dx_fallback++;
2125 ext4_mark_inode_dirty(handle, dir);
2126 }
2127 blocks = dir->i_size >> sb->s_blocksize_bits;
2128 for (block = 0; block < blocks; block++) {
2129 bh = ext4_read_dirblock(dir, block, DIRENT);
2130 if (IS_ERR(bh)) {
2131 retval = PTR_ERR(bh);
2132 bh = NULL;
2133 goto out;
2134 }
2135 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2136 NULL, bh);
2137 if (retval != -ENOSPC)
2138 goto out;
2139
2140 if (blocks == 1 && !dx_fallback &&
2141 ext4_has_feature_dir_index(sb)) {
2142 retval = make_indexed_dir(handle, &fname, dir,
2143 inode, bh);
2144 bh = NULL; /* make_indexed_dir releases bh */
2145 goto out;
2146 }
2147 brelse(bh);
2148 }
2149 bh = ext4_append(handle, dir, &block);
2150 if (IS_ERR(bh)) {
2151 retval = PTR_ERR(bh);
2152 bh = NULL;
2153 goto out;
2154 }
2155 de = (struct ext4_dir_entry_2 *) bh->b_data;
2156 de->inode = 0;
2157 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2158
2159 if (csum_size) {
2160 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2161 initialize_dirent_tail(t, blocksize);
2162 }
2163
2164 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2165out:
2166 ext4_fname_free_filename(&fname);
2167 brelse(bh);
2168 if (retval == 0)
2169 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2170 return retval;
2171}
2172
2173/*
2174 * Returns 0 for success, or a negative error value
2175 */
2176static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2177 struct inode *dir, struct inode *inode)
2178{
2179 struct dx_frame frames[2], *frame;
2180 struct dx_entry *entries, *at;
2181 struct buffer_head *bh;
2182 struct super_block *sb = dir->i_sb;
2183 struct ext4_dir_entry_2 *de;
2184 int err;
2185
2186 frame = dx_probe(fname, dir, NULL, frames);
2187 if (IS_ERR(frame))
2188 return PTR_ERR(frame);
2189 entries = frame->entries;
2190 at = frame->at;
2191 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2192 if (IS_ERR(bh)) {
2193 err = PTR_ERR(bh);
2194 bh = NULL;
2195 goto cleanup;
2196 }
2197
2198 BUFFER_TRACE(bh, "get_write_access");
2199 err = ext4_journal_get_write_access(handle, bh);
2200 if (err)
2201 goto journal_error;
2202
2203 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2204 if (err != -ENOSPC)
2205 goto cleanup;
2206
2207 /* Block full, should compress but for now just split */
2208 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2209 dx_get_count(entries), dx_get_limit(entries)));
2210 /* Need to split index? */
2211 if (dx_get_count(entries) == dx_get_limit(entries)) {
2212 ext4_lblk_t newblock;
2213 unsigned icount = dx_get_count(entries);
2214 int levels = frame - frames;
2215 struct dx_entry *entries2;
2216 struct dx_node *node2;
2217 struct buffer_head *bh2;
2218
2219 if (levels && (dx_get_count(frames->entries) ==
2220 dx_get_limit(frames->entries))) {
2221 ext4_warning_inode(dir, "Directory index full!");
2222 err = -ENOSPC;
2223 goto cleanup;
2224 }
2225 bh2 = ext4_append(handle, dir, &newblock);
2226 if (IS_ERR(bh2)) {
2227 err = PTR_ERR(bh2);
2228 goto cleanup;
2229 }
2230 node2 = (struct dx_node *)(bh2->b_data);
2231 entries2 = node2->entries;
2232 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2233 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2234 sb->s_blocksize);
2235 BUFFER_TRACE(frame->bh, "get_write_access");
2236 err = ext4_journal_get_write_access(handle, frame->bh);
2237 if (err)
2238 goto journal_error;
2239 if (levels) {
2240 unsigned icount1 = icount/2, icount2 = icount - icount1;
2241 unsigned hash2 = dx_get_hash(entries + icount1);
2242 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2243 icount1, icount2));
2244
2245 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2246 err = ext4_journal_get_write_access(handle,
2247 frames[0].bh);
2248 if (err)
2249 goto journal_error;
2250
2251 memcpy((char *) entries2, (char *) (entries + icount1),
2252 icount2 * sizeof(struct dx_entry));
2253 dx_set_count(entries, icount1);
2254 dx_set_count(entries2, icount2);
2255 dx_set_limit(entries2, dx_node_limit(dir));
2256
2257 /* Which index block gets the new entry? */
2258 if (at - entries >= icount1) {
2259 frame->at = at = at - entries - icount1 + entries2;
2260 frame->entries = entries = entries2;
2261 swap(frame->bh, bh2);
2262 }
2263 dx_insert_block(frames + 0, hash2, newblock);
2264 dxtrace(dx_show_index("node", frames[1].entries));
2265 dxtrace(dx_show_index("node",
2266 ((struct dx_node *) bh2->b_data)->entries));
2267 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2268 if (err)
2269 goto journal_error;
2270 brelse (bh2);
2271 } else {
2272 dxtrace(printk(KERN_DEBUG
2273 "Creating second level index...\n"));
2274 memcpy((char *) entries2, (char *) entries,
2275 icount * sizeof(struct dx_entry));
2276 dx_set_limit(entries2, dx_node_limit(dir));
2277
2278 /* Set up root */
2279 dx_set_count(entries, 1);
2280 dx_set_block(entries + 0, newblock);
2281 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2282
2283 /* Add new access path frame */
2284 frame = frames + 1;
2285 frame->at = at = at - entries + entries2;
2286 frame->entries = entries = entries2;
2287 frame->bh = bh2;
2288 err = ext4_journal_get_write_access(handle,
2289 frame->bh);
2290 if (err)
2291 goto journal_error;
2292 }
2293 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2294 if (err) {
2295 ext4_std_error(inode->i_sb, err);
2296 goto cleanup;
2297 }
2298 }
2299 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2300 if (IS_ERR(de)) {
2301 err = PTR_ERR(de);
2302 goto cleanup;
2303 }
2304 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2305 goto cleanup;
2306
2307journal_error:
2308 ext4_std_error(dir->i_sb, err);
2309cleanup:
2310 brelse(bh);
2311 dx_release(frames);
2312 return err;
2313}
2314
2315/*
2316 * ext4_generic_delete_entry deletes a directory entry by merging it
2317 * with the previous entry
2318 */
2319int ext4_generic_delete_entry(handle_t *handle,
2320 struct inode *dir,
2321 struct ext4_dir_entry_2 *de_del,
2322 struct buffer_head *bh,
2323 void *entry_buf,
2324 int buf_size,
2325 int csum_size)
2326{
2327 struct ext4_dir_entry_2 *de, *pde;
2328 unsigned int blocksize = dir->i_sb->s_blocksize;
2329 int i;
2330
2331 i = 0;
2332 pde = NULL;
2333 de = (struct ext4_dir_entry_2 *)entry_buf;
2334 while (i < buf_size - csum_size) {
2335 if (ext4_check_dir_entry(dir, NULL, de, bh,
2336 bh->b_data, bh->b_size, i))
2337 return -EFSCORRUPTED;
2338 if (de == de_del) {
2339 if (pde)
2340 pde->rec_len = ext4_rec_len_to_disk(
2341 ext4_rec_len_from_disk(pde->rec_len,
2342 blocksize) +
2343 ext4_rec_len_from_disk(de->rec_len,
2344 blocksize),
2345 blocksize);
2346 else
2347 de->inode = 0;
2348 dir->i_version++;
2349 return 0;
2350 }
2351 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2352 pde = de;
2353 de = ext4_next_entry(de, blocksize);
2354 }
2355 return -ENOENT;
2356}
2357
2358static int ext4_delete_entry(handle_t *handle,
2359 struct inode *dir,
2360 struct ext4_dir_entry_2 *de_del,
2361 struct buffer_head *bh)
2362{
2363 int err, csum_size = 0;
2364
2365 if (ext4_has_inline_data(dir)) {
2366 int has_inline_data = 1;
2367 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2368 &has_inline_data);
2369 if (has_inline_data)
2370 return err;
2371 }
2372
2373 if (ext4_has_metadata_csum(dir->i_sb))
2374 csum_size = sizeof(struct ext4_dir_entry_tail);
2375
2376 BUFFER_TRACE(bh, "get_write_access");
2377 err = ext4_journal_get_write_access(handle, bh);
2378 if (unlikely(err))
2379 goto out;
2380
2381 err = ext4_generic_delete_entry(handle, dir, de_del,
2382 bh, bh->b_data,
2383 dir->i_sb->s_blocksize, csum_size);
2384 if (err)
2385 goto out;
2386
2387 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2388 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2389 if (unlikely(err))
2390 goto out;
2391
2392 return 0;
2393out:
2394 if (err != -ENOENT)
2395 ext4_std_error(dir->i_sb, err);
2396 return err;
2397}
2398
2399/*
2400 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2401 * since this indicates that nlinks count was previously 1.
2402 */
2403static void ext4_inc_count(handle_t *handle, struct inode *inode)
2404{
2405 inc_nlink(inode);
2406 if (is_dx(inode) && inode->i_nlink > 1) {
2407 /* limit is 16-bit i_links_count */
2408 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2409 set_nlink(inode, 1);
2410 ext4_set_feature_dir_nlink(inode->i_sb);
2411 }
2412 }
2413}
2414
2415/*
2416 * If a directory had nlink == 1, then we should let it be 1. This indicates
2417 * directory has >EXT4_LINK_MAX subdirs.
2418 */
2419static void ext4_dec_count(handle_t *handle, struct inode *inode)
2420{
2421 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2422 drop_nlink(inode);
2423}
2424
2425
2426static int ext4_add_nondir(handle_t *handle,
2427 struct dentry *dentry, struct inode *inode)
2428{
2429 int err = ext4_add_entry(handle, dentry, inode);
2430 if (!err) {
2431 ext4_mark_inode_dirty(handle, inode);
2432 unlock_new_inode(inode);
2433 d_instantiate(dentry, inode);
2434 return 0;
2435 }
2436 drop_nlink(inode);
2437 unlock_new_inode(inode);
2438 iput(inode);
2439 return err;
2440}
2441
2442/*
2443 * By the time this is called, we already have created
2444 * the directory cache entry for the new file, but it
2445 * is so far negative - it has no inode.
2446 *
2447 * If the create succeeds, we fill in the inode information
2448 * with d_instantiate().
2449 */
2450static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2451 bool excl)
2452{
2453 handle_t *handle;
2454 struct inode *inode;
2455 int err, credits, retries = 0;
2456
2457 err = dquot_initialize(dir);
2458 if (err)
2459 return err;
2460
2461 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2462 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2463retry:
2464 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2465 NULL, EXT4_HT_DIR, credits);
2466 handle = ext4_journal_current_handle();
2467 err = PTR_ERR(inode);
2468 if (!IS_ERR(inode)) {
2469 inode->i_op = &ext4_file_inode_operations;
2470 inode->i_fop = &ext4_file_operations;
2471 ext4_set_aops(inode);
2472 err = ext4_add_nondir(handle, dentry, inode);
2473 if (!err && IS_DIRSYNC(dir))
2474 ext4_handle_sync(handle);
2475 }
2476 if (handle)
2477 ext4_journal_stop(handle);
2478 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2479 goto retry;
2480 return err;
2481}
2482
2483static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2484 umode_t mode, dev_t rdev)
2485{
2486 handle_t *handle;
2487 struct inode *inode;
2488 int err, credits, retries = 0;
2489
2490 err = dquot_initialize(dir);
2491 if (err)
2492 return err;
2493
2494 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2495 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2496retry:
2497 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2498 NULL, EXT4_HT_DIR, credits);
2499 handle = ext4_journal_current_handle();
2500 err = PTR_ERR(inode);
2501 if (!IS_ERR(inode)) {
2502 init_special_inode(inode, inode->i_mode, rdev);
2503 inode->i_op = &ext4_special_inode_operations;
2504 err = ext4_add_nondir(handle, dentry, inode);
2505 if (!err && IS_DIRSYNC(dir))
2506 ext4_handle_sync(handle);
2507 }
2508 if (handle)
2509 ext4_journal_stop(handle);
2510 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2511 goto retry;
2512 return err;
2513}
2514
2515static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2516{
2517 handle_t *handle;
2518 struct inode *inode;
2519 int err, retries = 0;
2520
2521 err = dquot_initialize(dir);
2522 if (err)
2523 return err;
2524
2525retry:
2526 inode = ext4_new_inode_start_handle(dir, mode,
2527 NULL, 0, NULL,
2528 EXT4_HT_DIR,
2529 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2530 4 + EXT4_XATTR_TRANS_BLOCKS);
2531 handle = ext4_journal_current_handle();
2532 err = PTR_ERR(inode);
2533 if (!IS_ERR(inode)) {
2534 inode->i_op = &ext4_file_inode_operations;
2535 inode->i_fop = &ext4_file_operations;
2536 ext4_set_aops(inode);
2537 d_tmpfile(dentry, inode);
2538 err = ext4_orphan_add(handle, inode);
2539 if (err)
2540 goto err_unlock_inode;
2541 mark_inode_dirty(inode);
2542 unlock_new_inode(inode);
2543 }
2544 if (handle)
2545 ext4_journal_stop(handle);
2546 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2547 goto retry;
2548 return err;
2549err_unlock_inode:
2550 ext4_journal_stop(handle);
2551 unlock_new_inode(inode);
2552 return err;
2553}
2554
2555struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2556 struct ext4_dir_entry_2 *de,
2557 int blocksize, int csum_size,
2558 unsigned int parent_ino, int dotdot_real_len)
2559{
2560 de->inode = cpu_to_le32(inode->i_ino);
2561 de->name_len = 1;
2562 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2563 blocksize);
2564 strcpy(de->name, ".");
2565 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2566
2567 de = ext4_next_entry(de, blocksize);
2568 de->inode = cpu_to_le32(parent_ino);
2569 de->name_len = 2;
2570 if (!dotdot_real_len)
2571 de->rec_len = ext4_rec_len_to_disk(blocksize -
2572 (csum_size + EXT4_DIR_REC_LEN(1)),
2573 blocksize);
2574 else
2575 de->rec_len = ext4_rec_len_to_disk(
2576 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2577 strcpy(de->name, "..");
2578 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2579
2580 return ext4_next_entry(de, blocksize);
2581}
2582
2583static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2584 struct inode *inode)
2585{
2586 struct buffer_head *dir_block = NULL;
2587 struct ext4_dir_entry_2 *de;
2588 struct ext4_dir_entry_tail *t;
2589 ext4_lblk_t block = 0;
2590 unsigned int blocksize = dir->i_sb->s_blocksize;
2591 int csum_size = 0;
2592 int err;
2593
2594 if (ext4_has_metadata_csum(dir->i_sb))
2595 csum_size = sizeof(struct ext4_dir_entry_tail);
2596
2597 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2598 err = ext4_try_create_inline_dir(handle, dir, inode);
2599 if (err < 0 && err != -ENOSPC)
2600 goto out;
2601 if (!err)
2602 goto out;
2603 }
2604
2605 inode->i_size = 0;
2606 dir_block = ext4_append(handle, inode, &block);
2607 if (IS_ERR(dir_block))
2608 return PTR_ERR(dir_block);
2609 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2610 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2611 set_nlink(inode, 2);
2612 if (csum_size) {
2613 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2614 initialize_dirent_tail(t, blocksize);
2615 }
2616
2617 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2618 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2619 if (err)
2620 goto out;
2621 set_buffer_verified(dir_block);
2622out:
2623 brelse(dir_block);
2624 return err;
2625}
2626
2627static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2628{
2629 handle_t *handle;
2630 struct inode *inode;
2631 int err, credits, retries = 0;
2632
2633 if (EXT4_DIR_LINK_MAX(dir))
2634 return -EMLINK;
2635
2636 err = dquot_initialize(dir);
2637 if (err)
2638 return err;
2639
2640 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2641 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2642retry:
2643 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2644 &dentry->d_name,
2645 0, NULL, EXT4_HT_DIR, credits);
2646 handle = ext4_journal_current_handle();
2647 err = PTR_ERR(inode);
2648 if (IS_ERR(inode))
2649 goto out_stop;
2650
2651 inode->i_op = &ext4_dir_inode_operations;
2652 inode->i_fop = &ext4_dir_operations;
2653 err = ext4_init_new_dir(handle, dir, inode);
2654 if (err)
2655 goto out_clear_inode;
2656 err = ext4_mark_inode_dirty(handle, inode);
2657 if (!err)
2658 err = ext4_add_entry(handle, dentry, inode);
2659 if (err) {
2660out_clear_inode:
2661 clear_nlink(inode);
2662 unlock_new_inode(inode);
2663 ext4_mark_inode_dirty(handle, inode);
2664 iput(inode);
2665 goto out_stop;
2666 }
2667 ext4_inc_count(handle, dir);
2668 ext4_update_dx_flag(dir);
2669 err = ext4_mark_inode_dirty(handle, dir);
2670 if (err)
2671 goto out_clear_inode;
2672 unlock_new_inode(inode);
2673 d_instantiate(dentry, inode);
2674 if (IS_DIRSYNC(dir))
2675 ext4_handle_sync(handle);
2676
2677out_stop:
2678 if (handle)
2679 ext4_journal_stop(handle);
2680 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2681 goto retry;
2682 return err;
2683}
2684
2685/*
2686 * routine to check that the specified directory is empty (for rmdir)
2687 */
2688int ext4_empty_dir(struct inode *inode)
2689{
2690 unsigned int offset;
2691 struct buffer_head *bh;
2692 struct ext4_dir_entry_2 *de, *de1;
2693 struct super_block *sb;
2694 int err = 0;
2695
2696 if (ext4_has_inline_data(inode)) {
2697 int has_inline_data = 1;
2698
2699 err = empty_inline_dir(inode, &has_inline_data);
2700 if (has_inline_data)
2701 return err;
2702 }
2703
2704 sb = inode->i_sb;
2705 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2706 EXT4_ERROR_INODE(inode, "invalid size");
2707 return 1;
2708 }
2709 bh = ext4_read_dirblock(inode, 0, EITHER);
2710 if (IS_ERR(bh))
2711 return 1;
2712
2713 de = (struct ext4_dir_entry_2 *) bh->b_data;
2714 de1 = ext4_next_entry(de, sb->s_blocksize);
2715 if (le32_to_cpu(de->inode) != inode->i_ino ||
2716 le32_to_cpu(de1->inode) == 0 ||
2717 strcmp(".", de->name) || strcmp("..", de1->name)) {
2718 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2719 brelse(bh);
2720 return 1;
2721 }
2722 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2723 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2724 de = ext4_next_entry(de1, sb->s_blocksize);
2725 while (offset < inode->i_size) {
2726 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2727 unsigned int lblock;
2728 err = 0;
2729 brelse(bh);
2730 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2731 bh = ext4_read_dirblock(inode, lblock, EITHER);
2732 if (IS_ERR(bh))
2733 return 1;
2734 de = (struct ext4_dir_entry_2 *) bh->b_data;
2735 }
2736 if (ext4_check_dir_entry(inode, NULL, de, bh,
2737 bh->b_data, bh->b_size, offset)) {
2738 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2739 sb->s_blocksize);
2740 offset = (offset | (sb->s_blocksize - 1)) + 1;
2741 continue;
2742 }
2743 if (le32_to_cpu(de->inode)) {
2744 brelse(bh);
2745 return 0;
2746 }
2747 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2748 de = ext4_next_entry(de, sb->s_blocksize);
2749 }
2750 brelse(bh);
2751 return 1;
2752}
2753
2754/*
2755 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2756 * such inodes, starting at the superblock, in case we crash before the
2757 * file is closed/deleted, or in case the inode truncate spans multiple
2758 * transactions and the last transaction is not recovered after a crash.
2759 *
2760 * At filesystem recovery time, we walk this list deleting unlinked
2761 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2762 *
2763 * Orphan list manipulation functions must be called under i_mutex unless
2764 * we are just creating the inode or deleting it.
2765 */
2766int ext4_orphan_add(handle_t *handle, struct inode *inode)
2767{
2768 struct super_block *sb = inode->i_sb;
2769 struct ext4_sb_info *sbi = EXT4_SB(sb);
2770 struct ext4_iloc iloc;
2771 int err = 0, rc;
2772 bool dirty = false;
2773
2774 if (!sbi->s_journal || is_bad_inode(inode))
2775 return 0;
2776
2777 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2778 !inode_is_locked(inode));
2779 /*
2780 * Exit early if inode already is on orphan list. This is a big speedup
2781 * since we don't have to contend on the global s_orphan_lock.
2782 */
2783 if (!list_empty(&EXT4_I(inode)->i_orphan))
2784 return 0;
2785
2786 /*
2787 * Orphan handling is only valid for files with data blocks
2788 * being truncated, or files being unlinked. Note that we either
2789 * hold i_mutex, or the inode can not be referenced from outside,
2790 * so i_nlink should not be bumped due to race
2791 */
2792 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2793 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2794
2795 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2796 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2797 if (err)
2798 goto out;
2799
2800 err = ext4_reserve_inode_write(handle, inode, &iloc);
2801 if (err)
2802 goto out;
2803
2804 mutex_lock(&sbi->s_orphan_lock);
2805 /*
2806 * Due to previous errors inode may be already a part of on-disk
2807 * orphan list. If so skip on-disk list modification.
2808 */
2809 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2810 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2811 /* Insert this inode at the head of the on-disk orphan list */
2812 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2813 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2814 dirty = true;
2815 }
2816 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2817 mutex_unlock(&sbi->s_orphan_lock);
2818
2819 if (dirty) {
2820 err = ext4_handle_dirty_super(handle, sb);
2821 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2822 if (!err)
2823 err = rc;
2824 if (err) {
2825 /*
2826 * We have to remove inode from in-memory list if
2827 * addition to on disk orphan list failed. Stray orphan
2828 * list entries can cause panics at unmount time.
2829 */
2830 mutex_lock(&sbi->s_orphan_lock);
2831 list_del(&EXT4_I(inode)->i_orphan);
2832 mutex_unlock(&sbi->s_orphan_lock);
2833 }
2834 }
2835 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2836 jbd_debug(4, "orphan inode %lu will point to %d\n",
2837 inode->i_ino, NEXT_ORPHAN(inode));
2838out:
2839 ext4_std_error(sb, err);
2840 return err;
2841}
2842
2843/*
2844 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2845 * of such inodes stored on disk, because it is finally being cleaned up.
2846 */
2847int ext4_orphan_del(handle_t *handle, struct inode *inode)
2848{
2849 struct list_head *prev;
2850 struct ext4_inode_info *ei = EXT4_I(inode);
2851 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2852 __u32 ino_next;
2853 struct ext4_iloc iloc;
2854 int err = 0;
2855
2856 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2857 return 0;
2858
2859 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2860 !inode_is_locked(inode));
2861 /* Do this quick check before taking global s_orphan_lock. */
2862 if (list_empty(&ei->i_orphan))
2863 return 0;
2864
2865 if (handle) {
2866 /* Grab inode buffer early before taking global s_orphan_lock */
2867 err = ext4_reserve_inode_write(handle, inode, &iloc);
2868 }
2869
2870 mutex_lock(&sbi->s_orphan_lock);
2871 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2872
2873 prev = ei->i_orphan.prev;
2874 list_del_init(&ei->i_orphan);
2875
2876 /* If we're on an error path, we may not have a valid
2877 * transaction handle with which to update the orphan list on
2878 * disk, but we still need to remove the inode from the linked
2879 * list in memory. */
2880 if (!handle || err) {
2881 mutex_unlock(&sbi->s_orphan_lock);
2882 goto out_err;
2883 }
2884
2885 ino_next = NEXT_ORPHAN(inode);
2886 if (prev == &sbi->s_orphan) {
2887 jbd_debug(4, "superblock will point to %u\n", ino_next);
2888 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2889 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2890 if (err) {
2891 mutex_unlock(&sbi->s_orphan_lock);
2892 goto out_brelse;
2893 }
2894 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2895 mutex_unlock(&sbi->s_orphan_lock);
2896 err = ext4_handle_dirty_super(handle, inode->i_sb);
2897 } else {
2898 struct ext4_iloc iloc2;
2899 struct inode *i_prev =
2900 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2901
2902 jbd_debug(4, "orphan inode %lu will point to %u\n",
2903 i_prev->i_ino, ino_next);
2904 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2905 if (err) {
2906 mutex_unlock(&sbi->s_orphan_lock);
2907 goto out_brelse;
2908 }
2909 NEXT_ORPHAN(i_prev) = ino_next;
2910 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2911 mutex_unlock(&sbi->s_orphan_lock);
2912 }
2913 if (err)
2914 goto out_brelse;
2915 NEXT_ORPHAN(inode) = 0;
2916 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2917out_err:
2918 ext4_std_error(inode->i_sb, err);
2919 return err;
2920
2921out_brelse:
2922 brelse(iloc.bh);
2923 goto out_err;
2924}
2925
2926static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2927{
2928 int retval;
2929 struct inode *inode;
2930 struct buffer_head *bh;
2931 struct ext4_dir_entry_2 *de;
2932 handle_t *handle = NULL;
2933
2934 /* Initialize quotas before so that eventual writes go in
2935 * separate transaction */
2936 retval = dquot_initialize(dir);
2937 if (retval)
2938 return retval;
2939 retval = dquot_initialize(d_inode(dentry));
2940 if (retval)
2941 return retval;
2942
2943 retval = -ENOENT;
2944 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2945 if (IS_ERR(bh))
2946 return PTR_ERR(bh);
2947 if (!bh)
2948 goto end_rmdir;
2949
2950 inode = d_inode(dentry);
2951
2952 retval = -EFSCORRUPTED;
2953 if (le32_to_cpu(de->inode) != inode->i_ino)
2954 goto end_rmdir;
2955
2956 retval = -ENOTEMPTY;
2957 if (!ext4_empty_dir(inode))
2958 goto end_rmdir;
2959
2960 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2961 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2962 if (IS_ERR(handle)) {
2963 retval = PTR_ERR(handle);
2964 handle = NULL;
2965 goto end_rmdir;
2966 }
2967
2968 if (IS_DIRSYNC(dir))
2969 ext4_handle_sync(handle);
2970
2971 retval = ext4_delete_entry(handle, dir, de, bh);
2972 if (retval)
2973 goto end_rmdir;
2974 if (!EXT4_DIR_LINK_EMPTY(inode))
2975 ext4_warning_inode(inode,
2976 "empty directory '%.*s' has too many links (%u)",
2977 dentry->d_name.len, dentry->d_name.name,
2978 inode->i_nlink);
2979 inode->i_version++;
2980 clear_nlink(inode);
2981 /* There's no need to set i_disksize: the fact that i_nlink is
2982 * zero will ensure that the right thing happens during any
2983 * recovery. */
2984 inode->i_size = 0;
2985 ext4_orphan_add(handle, inode);
2986 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2987 ext4_mark_inode_dirty(handle, inode);
2988 ext4_dec_count(handle, dir);
2989 ext4_update_dx_flag(dir);
2990 ext4_mark_inode_dirty(handle, dir);
2991
2992end_rmdir:
2993 brelse(bh);
2994 if (handle)
2995 ext4_journal_stop(handle);
2996 return retval;
2997}
2998
2999static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3000{
3001 int retval;
3002 struct inode *inode;
3003 struct buffer_head *bh;
3004 struct ext4_dir_entry_2 *de;
3005 handle_t *handle = NULL;
3006
3007 trace_ext4_unlink_enter(dir, dentry);
3008 /* Initialize quotas before so that eventual writes go
3009 * in separate transaction */
3010 retval = dquot_initialize(dir);
3011 if (retval)
3012 return retval;
3013 retval = dquot_initialize(d_inode(dentry));
3014 if (retval)
3015 return retval;
3016
3017 retval = -ENOENT;
3018 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3019 if (IS_ERR(bh))
3020 return PTR_ERR(bh);
3021 if (!bh)
3022 goto end_unlink;
3023
3024 inode = d_inode(dentry);
3025
3026 retval = -EFSCORRUPTED;
3027 if (le32_to_cpu(de->inode) != inode->i_ino)
3028 goto end_unlink;
3029
3030 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3031 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3032 if (IS_ERR(handle)) {
3033 retval = PTR_ERR(handle);
3034 handle = NULL;
3035 goto end_unlink;
3036 }
3037
3038 if (IS_DIRSYNC(dir))
3039 ext4_handle_sync(handle);
3040
3041 if (inode->i_nlink == 0) {
3042 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3043 dentry->d_name.len, dentry->d_name.name);
3044 set_nlink(inode, 1);
3045 }
3046 retval = ext4_delete_entry(handle, dir, de, bh);
3047 if (retval)
3048 goto end_unlink;
3049 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3050 ext4_update_dx_flag(dir);
3051 ext4_mark_inode_dirty(handle, dir);
3052 drop_nlink(inode);
3053 if (!inode->i_nlink)
3054 ext4_orphan_add(handle, inode);
3055 inode->i_ctime = ext4_current_time(inode);
3056 ext4_mark_inode_dirty(handle, inode);
3057
3058end_unlink:
3059 brelse(bh);
3060 if (handle)
3061 ext4_journal_stop(handle);
3062 trace_ext4_unlink_exit(dentry, retval);
3063 return retval;
3064}
3065
3066static int ext4_symlink(struct inode *dir,
3067 struct dentry *dentry, const char *symname)
3068{
3069 handle_t *handle;
3070 struct inode *inode;
3071 int err, len = strlen(symname);
3072 int credits;
3073 bool encryption_required;
3074 struct ext4_str disk_link;
3075 struct ext4_encrypted_symlink_data *sd = NULL;
3076
3077 disk_link.len = len + 1;
3078 disk_link.name = (char *) symname;
3079
3080 encryption_required = (ext4_encrypted_inode(dir) ||
3081 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3082 if (encryption_required) {
3083 err = ext4_get_encryption_info(dir);
3084 if (err)
3085 return err;
3086 if (ext4_encryption_info(dir) == NULL)
3087 return -EPERM;
3088 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3089 sizeof(struct ext4_encrypted_symlink_data));
3090 sd = kzalloc(disk_link.len, GFP_KERNEL);
3091 if (!sd)
3092 return -ENOMEM;
3093 }
3094
3095 if (disk_link.len > dir->i_sb->s_blocksize) {
3096 err = -ENAMETOOLONG;
3097 goto err_free_sd;
3098 }
3099
3100 err = dquot_initialize(dir);
3101 if (err)
3102 goto err_free_sd;
3103
3104 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3105 /*
3106 * For non-fast symlinks, we just allocate inode and put it on
3107 * orphan list in the first transaction => we need bitmap,
3108 * group descriptor, sb, inode block, quota blocks, and
3109 * possibly selinux xattr blocks.
3110 */
3111 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3112 EXT4_XATTR_TRANS_BLOCKS;
3113 } else {
3114 /*
3115 * Fast symlink. We have to add entry to directory
3116 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3117 * allocate new inode (bitmap, group descriptor, inode block,
3118 * quota blocks, sb is already counted in previous macros).
3119 */
3120 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3121 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3122 }
3123
3124 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3125 &dentry->d_name, 0, NULL,
3126 EXT4_HT_DIR, credits);
3127 handle = ext4_journal_current_handle();
3128 if (IS_ERR(inode)) {
3129 if (handle)
3130 ext4_journal_stop(handle);
3131 err = PTR_ERR(inode);
3132 goto err_free_sd;
3133 }
3134
3135 if (encryption_required) {
3136 struct qstr istr;
3137 struct ext4_str ostr;
3138
3139 istr.name = (const unsigned char *) symname;
3140 istr.len = len;
3141 ostr.name = sd->encrypted_path;
3142 ostr.len = disk_link.len;
3143 err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3144 if (err < 0)
3145 goto err_drop_inode;
3146 sd->len = cpu_to_le16(ostr.len);
3147 disk_link.name = (char *) sd;
3148 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3149 }
3150
3151 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3152 if (!encryption_required)
3153 inode->i_op = &ext4_symlink_inode_operations;
3154 inode_nohighmem(inode);
3155 ext4_set_aops(inode);
3156 /*
3157 * We cannot call page_symlink() with transaction started
3158 * because it calls into ext4_write_begin() which can wait
3159 * for transaction commit if we are running out of space
3160 * and thus we deadlock. So we have to stop transaction now
3161 * and restart it when symlink contents is written.
3162 *
3163 * To keep fs consistent in case of crash, we have to put inode
3164 * to orphan list in the mean time.
3165 */
3166 drop_nlink(inode);
3167 err = ext4_orphan_add(handle, inode);
3168 ext4_journal_stop(handle);
3169 handle = NULL;
3170 if (err)
3171 goto err_drop_inode;
3172 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3173 if (err)
3174 goto err_drop_inode;
3175 /*
3176 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3177 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3178 */
3179 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3180 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3181 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3182 if (IS_ERR(handle)) {
3183 err = PTR_ERR(handle);
3184 handle = NULL;
3185 goto err_drop_inode;
3186 }
3187 set_nlink(inode, 1);
3188 err = ext4_orphan_del(handle, inode);
3189 if (err)
3190 goto err_drop_inode;
3191 } else {
3192 /* clear the extent format for fast symlink */
3193 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3194 if (!encryption_required) {
3195 inode->i_op = &ext4_fast_symlink_inode_operations;
3196 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3197 }
3198 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3199 disk_link.len);
3200 inode->i_size = disk_link.len - 1;
3201 }
3202 EXT4_I(inode)->i_disksize = inode->i_size;
3203 err = ext4_add_nondir(handle, dentry, inode);
3204 if (!err && IS_DIRSYNC(dir))
3205 ext4_handle_sync(handle);
3206
3207 if (handle)
3208 ext4_journal_stop(handle);
3209 kfree(sd);
3210 return err;
3211err_drop_inode:
3212 if (handle)
3213 ext4_journal_stop(handle);
3214 clear_nlink(inode);
3215 unlock_new_inode(inode);
3216 iput(inode);
3217err_free_sd:
3218 kfree(sd);
3219 return err;
3220}
3221
3222static int ext4_link(struct dentry *old_dentry,
3223 struct inode *dir, struct dentry *dentry)
3224{
3225 handle_t *handle;
3226 struct inode *inode = d_inode(old_dentry);
3227 int err, retries = 0;
3228
3229 if (inode->i_nlink >= EXT4_LINK_MAX)
3230 return -EMLINK;
3231 if (ext4_encrypted_inode(dir) &&
3232 !ext4_is_child_context_consistent_with_parent(dir, inode))
3233 return -EPERM;
3234
3235 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3236 (!projid_eq(EXT4_I(dir)->i_projid,
3237 EXT4_I(old_dentry->d_inode)->i_projid)))
3238 return -EXDEV;
3239
3240 err = dquot_initialize(dir);
3241 if (err)
3242 return err;
3243
3244retry:
3245 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3246 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3247 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3248 if (IS_ERR(handle))
3249 return PTR_ERR(handle);
3250
3251 if (IS_DIRSYNC(dir))
3252 ext4_handle_sync(handle);
3253
3254 inode->i_ctime = ext4_current_time(inode);
3255 ext4_inc_count(handle, inode);
3256 ihold(inode);
3257
3258 err = ext4_add_entry(handle, dentry, inode);
3259 if (!err) {
3260 ext4_mark_inode_dirty(handle, inode);
3261 /* this can happen only for tmpfile being
3262 * linked the first time
3263 */
3264 if (inode->i_nlink == 1)
3265 ext4_orphan_del(handle, inode);
3266 d_instantiate(dentry, inode);
3267 } else {
3268 drop_nlink(inode);
3269 iput(inode);
3270 }
3271 ext4_journal_stop(handle);
3272 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3273 goto retry;
3274 return err;
3275}
3276
3277
3278/*
3279 * Try to find buffer head where contains the parent block.
3280 * It should be the inode block if it is inlined or the 1st block
3281 * if it is a normal dir.
3282 */
3283static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3284 struct inode *inode,
3285 int *retval,
3286 struct ext4_dir_entry_2 **parent_de,
3287 int *inlined)
3288{
3289 struct buffer_head *bh;
3290
3291 if (!ext4_has_inline_data(inode)) {
3292 bh = ext4_read_dirblock(inode, 0, EITHER);
3293 if (IS_ERR(bh)) {
3294 *retval = PTR_ERR(bh);
3295 return NULL;
3296 }
3297 *parent_de = ext4_next_entry(
3298 (struct ext4_dir_entry_2 *)bh->b_data,
3299 inode->i_sb->s_blocksize);
3300 return bh;
3301 }
3302
3303 *inlined = 1;
3304 return ext4_get_first_inline_block(inode, parent_de, retval);
3305}
3306
3307struct ext4_renament {
3308 struct inode *dir;
3309 struct dentry *dentry;
3310 struct inode *inode;
3311 bool is_dir;
3312 int dir_nlink_delta;
3313
3314 /* entry for "dentry" */
3315 struct buffer_head *bh;
3316 struct ext4_dir_entry_2 *de;
3317 int inlined;
3318
3319 /* entry for ".." in inode if it's a directory */
3320 struct buffer_head *dir_bh;
3321 struct ext4_dir_entry_2 *parent_de;
3322 int dir_inlined;
3323};
3324
3325static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3326{
3327 int retval;
3328
3329 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3330 &retval, &ent->parent_de,
3331 &ent->dir_inlined);
3332 if (!ent->dir_bh)
3333 return retval;
3334 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3335 return -EFSCORRUPTED;
3336 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3337 return ext4_journal_get_write_access(handle, ent->dir_bh);
3338}
3339
3340static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3341 unsigned dir_ino)
3342{
3343 int retval;
3344
3345 ent->parent_de->inode = cpu_to_le32(dir_ino);
3346 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3347 if (!ent->dir_inlined) {
3348 if (is_dx(ent->inode)) {
3349 retval = ext4_handle_dirty_dx_node(handle,
3350 ent->inode,
3351 ent->dir_bh);
3352 } else {
3353 retval = ext4_handle_dirty_dirent_node(handle,
3354 ent->inode,
3355 ent->dir_bh);
3356 }
3357 } else {
3358 retval = ext4_mark_inode_dirty(handle, ent->inode);
3359 }
3360 if (retval) {
3361 ext4_std_error(ent->dir->i_sb, retval);
3362 return retval;
3363 }
3364 return 0;
3365}
3366
3367static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3368 unsigned ino, unsigned file_type)
3369{
3370 int retval;
3371
3372 BUFFER_TRACE(ent->bh, "get write access");
3373 retval = ext4_journal_get_write_access(handle, ent->bh);
3374 if (retval)
3375 return retval;
3376 ent->de->inode = cpu_to_le32(ino);
3377 if (ext4_has_feature_filetype(ent->dir->i_sb))
3378 ent->de->file_type = file_type;
3379 ent->dir->i_version++;
3380 ent->dir->i_ctime = ent->dir->i_mtime =
3381 ext4_current_time(ent->dir);
3382 ext4_mark_inode_dirty(handle, ent->dir);
3383 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3384 if (!ent->inlined) {
3385 retval = ext4_handle_dirty_dirent_node(handle,
3386 ent->dir, ent->bh);
3387 if (unlikely(retval)) {
3388 ext4_std_error(ent->dir->i_sb, retval);
3389 return retval;
3390 }
3391 }
3392 brelse(ent->bh);
3393 ent->bh = NULL;
3394
3395 return 0;
3396}
3397
3398static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3399 const struct qstr *d_name)
3400{
3401 int retval = -ENOENT;
3402 struct buffer_head *bh;
3403 struct ext4_dir_entry_2 *de;
3404
3405 bh = ext4_find_entry(dir, d_name, &de, NULL);
3406 if (IS_ERR(bh))
3407 return PTR_ERR(bh);
3408 if (bh) {
3409 retval = ext4_delete_entry(handle, dir, de, bh);
3410 brelse(bh);
3411 }
3412 return retval;
3413}
3414
3415static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3416 int force_reread)
3417{
3418 int retval;
3419 /*
3420 * ent->de could have moved from under us during htree split, so make
3421 * sure that we are deleting the right entry. We might also be pointing
3422 * to a stale entry in the unused part of ent->bh so just checking inum
3423 * and the name isn't enough.
3424 */
3425 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3426 ent->de->name_len != ent->dentry->d_name.len ||
3427 strncmp(ent->de->name, ent->dentry->d_name.name,
3428 ent->de->name_len) ||
3429 force_reread) {
3430 retval = ext4_find_delete_entry(handle, ent->dir,
3431 &ent->dentry->d_name);
3432 } else {
3433 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3434 if (retval == -ENOENT) {
3435 retval = ext4_find_delete_entry(handle, ent->dir,
3436 &ent->dentry->d_name);
3437 }
3438 }
3439
3440 if (retval) {
3441 ext4_warning_inode(ent->dir,
3442 "Deleting old file: nlink %d, error=%d",
3443 ent->dir->i_nlink, retval);
3444 }
3445}
3446
3447static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3448{
3449 if (ent->dir_nlink_delta) {
3450 if (ent->dir_nlink_delta == -1)
3451 ext4_dec_count(handle, ent->dir);
3452 else
3453 ext4_inc_count(handle, ent->dir);
3454 ext4_mark_inode_dirty(handle, ent->dir);
3455 }
3456}
3457
3458static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3459 int credits, handle_t **h)
3460{
3461 struct inode *wh;
3462 handle_t *handle;
3463 int retries = 0;
3464
3465 /*
3466 * for inode block, sb block, group summaries,
3467 * and inode bitmap
3468 */
3469 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3470 EXT4_XATTR_TRANS_BLOCKS + 4);
3471retry:
3472 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3473 &ent->dentry->d_name, 0, NULL,
3474 EXT4_HT_DIR, credits);
3475
3476 handle = ext4_journal_current_handle();
3477 if (IS_ERR(wh)) {
3478 if (handle)
3479 ext4_journal_stop(handle);
3480 if (PTR_ERR(wh) == -ENOSPC &&
3481 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3482 goto retry;
3483 } else {
3484 *h = handle;
3485 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3486 wh->i_op = &ext4_special_inode_operations;
3487 }
3488 return wh;
3489}
3490
3491/*
3492 * Anybody can rename anything with this: the permission checks are left to the
3493 * higher-level routines.
3494 *
3495 * n.b. old_{dentry,inode) refers to the source dentry/inode
3496 * while new_{dentry,inode) refers to the destination dentry/inode
3497 * This comes from rename(const char *oldpath, const char *newpath)
3498 */
3499static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3500 struct inode *new_dir, struct dentry *new_dentry,
3501 unsigned int flags)
3502{
3503 handle_t *handle = NULL;
3504 struct ext4_renament old = {
3505 .dir = old_dir,
3506 .dentry = old_dentry,
3507 .inode = d_inode(old_dentry),
3508 };
3509 struct ext4_renament new = {
3510 .dir = new_dir,
3511 .dentry = new_dentry,
3512 .inode = d_inode(new_dentry),
3513 };
3514 int force_reread;
3515 int retval;
3516 struct inode *whiteout = NULL;
3517 int credits;
3518 u8 old_file_type;
3519
3520 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3521 (!projid_eq(EXT4_I(new_dir)->i_projid,
3522 EXT4_I(old_dentry->d_inode)->i_projid)))
3523 return -EXDEV;
3524
3525 retval = dquot_initialize(old.dir);
3526 if (retval)
3527 return retval;
3528 retval = dquot_initialize(new.dir);
3529 if (retval)
3530 return retval;
3531
3532 /* Initialize quotas before so that eventual writes go
3533 * in separate transaction */
3534 if (new.inode) {
3535 retval = dquot_initialize(new.inode);
3536 if (retval)
3537 return retval;
3538 }
3539
3540 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3541 if (IS_ERR(old.bh))
3542 return PTR_ERR(old.bh);
3543 /*
3544 * Check for inode number is _not_ due to possible IO errors.
3545 * We might rmdir the source, keep it as pwd of some process
3546 * and merrily kill the link to whatever was created under the
3547 * same name. Goodbye sticky bit ;-<
3548 */
3549 retval = -ENOENT;
3550 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3551 goto end_rename;
3552
3553 if ((old.dir != new.dir) &&
3554 ext4_encrypted_inode(new.dir) &&
3555 !ext4_is_child_context_consistent_with_parent(new.dir,
3556 old.inode)) {
3557 retval = -EPERM;
3558 goto end_rename;
3559 }
3560
3561 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3562 &new.de, &new.inlined);
3563 if (IS_ERR(new.bh)) {
3564 retval = PTR_ERR(new.bh);
3565 new.bh = NULL;
3566 goto end_rename;
3567 }
3568 if (new.bh) {
3569 if (!new.inode) {
3570 brelse(new.bh);
3571 new.bh = NULL;
3572 }
3573 }
3574 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3575 ext4_alloc_da_blocks(old.inode);
3576
3577 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3578 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3579 if (!(flags & RENAME_WHITEOUT)) {
3580 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3581 if (IS_ERR(handle)) {
3582 retval = PTR_ERR(handle);
3583 handle = NULL;
3584 goto end_rename;
3585 }
3586 } else {
3587 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3588 if (IS_ERR(whiteout)) {
3589 retval = PTR_ERR(whiteout);
3590 whiteout = NULL;
3591 goto end_rename;
3592 }
3593 }
3594
3595 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3596 ext4_handle_sync(handle);
3597
3598 if (S_ISDIR(old.inode->i_mode)) {
3599 if (new.inode) {
3600 retval = -ENOTEMPTY;
3601 if (!ext4_empty_dir(new.inode))
3602 goto end_rename;
3603 } else {
3604 retval = -EMLINK;
3605 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3606 goto end_rename;
3607 }
3608 retval = ext4_rename_dir_prepare(handle, &old);
3609 if (retval)
3610 goto end_rename;
3611 }
3612 /*
3613 * If we're renaming a file within an inline_data dir and adding or
3614 * setting the new dirent causes a conversion from inline_data to
3615 * extents/blockmap, we need to force the dirent delete code to
3616 * re-read the directory, or else we end up trying to delete a dirent
3617 * from what is now the extent tree root (or a block map).
3618 */
3619 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3620 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3621
3622 old_file_type = old.de->file_type;
3623 if (whiteout) {
3624 /*
3625 * Do this before adding a new entry, so the old entry is sure
3626 * to be still pointing to the valid old entry.
3627 */
3628 retval = ext4_setent(handle, &old, whiteout->i_ino,
3629 EXT4_FT_CHRDEV);
3630 if (retval)
3631 goto end_rename;
3632 ext4_mark_inode_dirty(handle, whiteout);
3633 }
3634 if (!new.bh) {
3635 retval = ext4_add_entry(handle, new.dentry, old.inode);
3636 if (retval)
3637 goto end_rename;
3638 } else {
3639 retval = ext4_setent(handle, &new,
3640 old.inode->i_ino, old_file_type);
3641 if (retval)
3642 goto end_rename;
3643 }
3644 if (force_reread)
3645 force_reread = !ext4_test_inode_flag(new.dir,
3646 EXT4_INODE_INLINE_DATA);
3647
3648 /*
3649 * Like most other Unix systems, set the ctime for inodes on a
3650 * rename.
3651 */
3652 old.inode->i_ctime = ext4_current_time(old.inode);
3653 ext4_mark_inode_dirty(handle, old.inode);
3654
3655 if (!whiteout) {
3656 /*
3657 * ok, that's it
3658 */
3659 ext4_rename_delete(handle, &old, force_reread);
3660 }
3661
3662 if (new.inode) {
3663 ext4_dec_count(handle, new.inode);
3664 new.inode->i_ctime = ext4_current_time(new.inode);
3665 }
3666 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3667 ext4_update_dx_flag(old.dir);
3668 if (old.dir_bh) {
3669 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3670 if (retval)
3671 goto end_rename;
3672
3673 ext4_dec_count(handle, old.dir);
3674 if (new.inode) {
3675 /* checked ext4_empty_dir above, can't have another
3676 * parent, ext4_dec_count() won't work for many-linked
3677 * dirs */
3678 clear_nlink(new.inode);
3679 } else {
3680 ext4_inc_count(handle, new.dir);
3681 ext4_update_dx_flag(new.dir);
3682 ext4_mark_inode_dirty(handle, new.dir);
3683 }
3684 }
3685 ext4_mark_inode_dirty(handle, old.dir);
3686 if (new.inode) {
3687 ext4_mark_inode_dirty(handle, new.inode);
3688 if (!new.inode->i_nlink)
3689 ext4_orphan_add(handle, new.inode);
3690 }
3691 retval = 0;
3692
3693end_rename:
3694 brelse(old.dir_bh);
3695 brelse(old.bh);
3696 brelse(new.bh);
3697 if (whiteout) {
3698 if (retval)
3699 drop_nlink(whiteout);
3700 unlock_new_inode(whiteout);
3701 iput(whiteout);
3702 }
3703 if (handle)
3704 ext4_journal_stop(handle);
3705 return retval;
3706}
3707
3708static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3709 struct inode *new_dir, struct dentry *new_dentry)
3710{
3711 handle_t *handle = NULL;
3712 struct ext4_renament old = {
3713 .dir = old_dir,
3714 .dentry = old_dentry,
3715 .inode = d_inode(old_dentry),
3716 };
3717 struct ext4_renament new = {
3718 .dir = new_dir,
3719 .dentry = new_dentry,
3720 .inode = d_inode(new_dentry),
3721 };
3722 u8 new_file_type;
3723 int retval;
3724
3725 if ((ext4_encrypted_inode(old_dir) ||
3726 ext4_encrypted_inode(new_dir)) &&
3727 (old_dir != new_dir) &&
3728 (!ext4_is_child_context_consistent_with_parent(new_dir,
3729 old.inode) ||
3730 !ext4_is_child_context_consistent_with_parent(old_dir,
3731 new.inode)))
3732 return -EPERM;
3733
3734 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3735 !projid_eq(EXT4_I(new_dir)->i_projid,
3736 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3737 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3738 !projid_eq(EXT4_I(old_dir)->i_projid,
3739 EXT4_I(new_dentry->d_inode)->i_projid)))
3740 return -EXDEV;
3741
3742 retval = dquot_initialize(old.dir);
3743 if (retval)
3744 return retval;
3745 retval = dquot_initialize(new.dir);
3746 if (retval)
3747 return retval;
3748
3749 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3750 &old.de, &old.inlined);
3751 if (IS_ERR(old.bh))
3752 return PTR_ERR(old.bh);
3753 /*
3754 * Check for inode number is _not_ due to possible IO errors.
3755 * We might rmdir the source, keep it as pwd of some process
3756 * and merrily kill the link to whatever was created under the
3757 * same name. Goodbye sticky bit ;-<
3758 */
3759 retval = -ENOENT;
3760 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3761 goto end_rename;
3762
3763 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3764 &new.de, &new.inlined);
3765 if (IS_ERR(new.bh)) {
3766 retval = PTR_ERR(new.bh);
3767 new.bh = NULL;
3768 goto end_rename;
3769 }
3770
3771 /* RENAME_EXCHANGE case: old *and* new must both exist */
3772 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3773 goto end_rename;
3774
3775 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3776 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3777 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3778 if (IS_ERR(handle)) {
3779 retval = PTR_ERR(handle);
3780 handle = NULL;
3781 goto end_rename;
3782 }
3783
3784 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3785 ext4_handle_sync(handle);
3786
3787 if (S_ISDIR(old.inode->i_mode)) {
3788 old.is_dir = true;
3789 retval = ext4_rename_dir_prepare(handle, &old);
3790 if (retval)
3791 goto end_rename;
3792 }
3793 if (S_ISDIR(new.inode->i_mode)) {
3794 new.is_dir = true;
3795 retval = ext4_rename_dir_prepare(handle, &new);
3796 if (retval)
3797 goto end_rename;
3798 }
3799
3800 /*
3801 * Other than the special case of overwriting a directory, parents'
3802 * nlink only needs to be modified if this is a cross directory rename.
3803 */
3804 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3805 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3806 new.dir_nlink_delta = -old.dir_nlink_delta;
3807 retval = -EMLINK;
3808 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3809 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3810 goto end_rename;
3811 }
3812
3813 new_file_type = new.de->file_type;
3814 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3815 if (retval)
3816 goto end_rename;
3817
3818 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3819 if (retval)
3820 goto end_rename;
3821
3822 /*
3823 * Like most other Unix systems, set the ctime for inodes on a
3824 * rename.
3825 */
3826 old.inode->i_ctime = ext4_current_time(old.inode);
3827 new.inode->i_ctime = ext4_current_time(new.inode);
3828 ext4_mark_inode_dirty(handle, old.inode);
3829 ext4_mark_inode_dirty(handle, new.inode);
3830
3831 if (old.dir_bh) {
3832 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3833 if (retval)
3834 goto end_rename;
3835 }
3836 if (new.dir_bh) {
3837 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3838 if (retval)
3839 goto end_rename;
3840 }
3841 ext4_update_dir_count(handle, &old);
3842 ext4_update_dir_count(handle, &new);
3843 retval = 0;
3844
3845end_rename:
3846 brelse(old.dir_bh);
3847 brelse(new.dir_bh);
3848 brelse(old.bh);
3849 brelse(new.bh);
3850 if (handle)
3851 ext4_journal_stop(handle);
3852 return retval;
3853}
3854
3855static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3856 struct inode *new_dir, struct dentry *new_dentry,
3857 unsigned int flags)
3858{
3859 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3860 return -EINVAL;
3861
3862 if (flags & RENAME_EXCHANGE) {
3863 return ext4_cross_rename(old_dir, old_dentry,
3864 new_dir, new_dentry);
3865 }
3866
3867 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3868}
3869
3870/*
3871 * directories can handle most operations...
3872 */
3873const struct inode_operations ext4_dir_inode_operations = {
3874 .create = ext4_create,
3875 .lookup = ext4_lookup,
3876 .link = ext4_link,
3877 .unlink = ext4_unlink,
3878 .symlink = ext4_symlink,
3879 .mkdir = ext4_mkdir,
3880 .rmdir = ext4_rmdir,
3881 .mknod = ext4_mknod,
3882 .tmpfile = ext4_tmpfile,
3883 .rename2 = ext4_rename2,
3884 .setattr = ext4_setattr,
3885 .setxattr = generic_setxattr,
3886 .getxattr = generic_getxattr,
3887 .listxattr = ext4_listxattr,
3888 .removexattr = generic_removexattr,
3889 .get_acl = ext4_get_acl,
3890 .set_acl = ext4_set_acl,
3891 .fiemap = ext4_fiemap,
3892};
3893
3894const struct inode_operations ext4_special_inode_operations = {
3895 .setattr = ext4_setattr,
3896 .setxattr = generic_setxattr,
3897 .getxattr = generic_getxattr,
3898 .listxattr = ext4_listxattr,
3899 .removexattr = generic_removexattr,
3900 .get_acl = ext4_get_acl,
3901 .set_acl = ext4_set_acl,
3902};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/namei.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/namei.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
25 * Theodore Ts'o, 2002
26 */
27
28#include <linux/fs.h>
29#include <linux/pagemap.h>
30#include <linux/time.h>
31#include <linux/fcntl.h>
32#include <linux/stat.h>
33#include <linux/string.h>
34#include <linux/quotaops.h>
35#include <linux/buffer_head.h>
36#include <linux/bio.h>
37#include <linux/iversion.h>
38#include <linux/unicode.h>
39#include "ext4.h"
40#include "ext4_jbd2.h"
41
42#include "xattr.h"
43#include "acl.h"
44
45#include <trace/events/ext4.h>
46/*
47 * define how far ahead to read directories while searching them.
48 */
49#define NAMEI_RA_CHUNKS 2
50#define NAMEI_RA_BLOCKS 4
51#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52
53static struct buffer_head *ext4_append(handle_t *handle,
54 struct inode *inode,
55 ext4_lblk_t *block)
56{
57 struct ext4_map_blocks map;
58 struct buffer_head *bh;
59 int err;
60
61 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
62 ((inode->i_size >> 10) >=
63 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
64 return ERR_PTR(-ENOSPC);
65
66 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67 map.m_lblk = *block;
68 map.m_len = 1;
69
70 /*
71 * We're appending new directory block. Make sure the block is not
72 * allocated yet, otherwise we will end up corrupting the
73 * directory.
74 */
75 err = ext4_map_blocks(NULL, inode, &map, 0);
76 if (err < 0)
77 return ERR_PTR(err);
78 if (err) {
79 EXT4_ERROR_INODE(inode, "Logical block already allocated");
80 return ERR_PTR(-EFSCORRUPTED);
81 }
82
83 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
84 if (IS_ERR(bh))
85 return bh;
86 inode->i_size += inode->i_sb->s_blocksize;
87 EXT4_I(inode)->i_disksize = inode->i_size;
88 err = ext4_mark_inode_dirty(handle, inode);
89 if (err)
90 goto out;
91 BUFFER_TRACE(bh, "get_write_access");
92 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
93 EXT4_JTR_NONE);
94 if (err)
95 goto out;
96 return bh;
97
98out:
99 brelse(bh);
100 ext4_std_error(inode->i_sb, err);
101 return ERR_PTR(err);
102}
103
104static int ext4_dx_csum_verify(struct inode *inode,
105 struct ext4_dir_entry *dirent);
106
107/*
108 * Hints to ext4_read_dirblock regarding whether we expect a directory
109 * block being read to be an index block, or a block containing
110 * directory entries (and if the latter, whether it was found via a
111 * logical block in an htree index block). This is used to control
112 * what sort of sanity checkinig ext4_read_dirblock() will do on the
113 * directory block read from the storage device. EITHER will means
114 * the caller doesn't know what kind of directory block will be read,
115 * so no specific verification will be done.
116 */
117typedef enum {
118 EITHER, INDEX, DIRENT, DIRENT_HTREE
119} dirblock_type_t;
120
121#define ext4_read_dirblock(inode, block, type) \
122 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
123
124static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
125 ext4_lblk_t block,
126 dirblock_type_t type,
127 const char *func,
128 unsigned int line)
129{
130 struct buffer_head *bh;
131 struct ext4_dir_entry *dirent;
132 int is_dx_block = 0;
133
134 if (block >= inode->i_size >> inode->i_blkbits) {
135 ext4_error_inode(inode, func, line, block,
136 "Attempting to read directory block (%u) that is past i_size (%llu)",
137 block, inode->i_size);
138 return ERR_PTR(-EFSCORRUPTED);
139 }
140
141 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
142 bh = ERR_PTR(-EIO);
143 else
144 bh = ext4_bread(NULL, inode, block, 0);
145 if (IS_ERR(bh)) {
146 __ext4_warning(inode->i_sb, func, line,
147 "inode #%lu: lblock %lu: comm %s: "
148 "error %ld reading directory block",
149 inode->i_ino, (unsigned long)block,
150 current->comm, PTR_ERR(bh));
151
152 return bh;
153 }
154 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
155 ext4_error_inode(inode, func, line, block,
156 "Directory hole found for htree %s block",
157 (type == INDEX) ? "index" : "leaf");
158 return ERR_PTR(-EFSCORRUPTED);
159 }
160 if (!bh)
161 return NULL;
162 dirent = (struct ext4_dir_entry *) bh->b_data;
163 /* Determine whether or not we have an index block */
164 if (is_dx(inode)) {
165 if (block == 0)
166 is_dx_block = 1;
167 else if (ext4_rec_len_from_disk(dirent->rec_len,
168 inode->i_sb->s_blocksize) ==
169 inode->i_sb->s_blocksize)
170 is_dx_block = 1;
171 }
172 if (!is_dx_block && type == INDEX) {
173 ext4_error_inode(inode, func, line, block,
174 "directory leaf block found instead of index block");
175 brelse(bh);
176 return ERR_PTR(-EFSCORRUPTED);
177 }
178 if (!ext4_has_metadata_csum(inode->i_sb) ||
179 buffer_verified(bh))
180 return bh;
181
182 /*
183 * An empty leaf block can get mistaken for a index block; for
184 * this reason, we can only check the index checksum when the
185 * caller is sure it should be an index block.
186 */
187 if (is_dx_block && type == INDEX) {
188 if (ext4_dx_csum_verify(inode, dirent) &&
189 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
190 set_buffer_verified(bh);
191 else {
192 ext4_error_inode_err(inode, func, line, block,
193 EFSBADCRC,
194 "Directory index failed checksum");
195 brelse(bh);
196 return ERR_PTR(-EFSBADCRC);
197 }
198 }
199 if (!is_dx_block) {
200 if (ext4_dirblock_csum_verify(inode, bh) &&
201 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
202 set_buffer_verified(bh);
203 else {
204 ext4_error_inode_err(inode, func, line, block,
205 EFSBADCRC,
206 "Directory block failed checksum");
207 brelse(bh);
208 return ERR_PTR(-EFSBADCRC);
209 }
210 }
211 return bh;
212}
213
214#ifdef DX_DEBUG
215#define dxtrace(command) command
216#else
217#define dxtrace(command)
218#endif
219
220struct fake_dirent
221{
222 __le32 inode;
223 __le16 rec_len;
224 u8 name_len;
225 u8 file_type;
226};
227
228struct dx_countlimit
229{
230 __le16 limit;
231 __le16 count;
232};
233
234struct dx_entry
235{
236 __le32 hash;
237 __le32 block;
238};
239
240/*
241 * dx_root_info is laid out so that if it should somehow get overlaid by a
242 * dirent the two low bits of the hash version will be zero. Therefore, the
243 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
244 */
245
246struct dx_root
247{
248 struct fake_dirent dot;
249 char dot_name[4];
250 struct fake_dirent dotdot;
251 char dotdot_name[4];
252 struct dx_root_info
253 {
254 __le32 reserved_zero;
255 u8 hash_version;
256 u8 info_length; /* 8 */
257 u8 indirect_levels;
258 u8 unused_flags;
259 }
260 info;
261 struct dx_entry entries[];
262};
263
264struct dx_node
265{
266 struct fake_dirent fake;
267 struct dx_entry entries[];
268};
269
270
271struct dx_frame
272{
273 struct buffer_head *bh;
274 struct dx_entry *entries;
275 struct dx_entry *at;
276};
277
278struct dx_map_entry
279{
280 u32 hash;
281 u16 offs;
282 u16 size;
283};
284
285/*
286 * This goes at the end of each htree block.
287 */
288struct dx_tail {
289 u32 dt_reserved;
290 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
291};
292
293static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
294static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
295static inline unsigned dx_get_hash(struct dx_entry *entry);
296static void dx_set_hash(struct dx_entry *entry, unsigned value);
297static unsigned dx_get_count(struct dx_entry *entries);
298static unsigned dx_get_limit(struct dx_entry *entries);
299static void dx_set_count(struct dx_entry *entries, unsigned value);
300static void dx_set_limit(struct dx_entry *entries, unsigned value);
301static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
302static unsigned dx_node_limit(struct inode *dir);
303static struct dx_frame *dx_probe(struct ext4_filename *fname,
304 struct inode *dir,
305 struct dx_hash_info *hinfo,
306 struct dx_frame *frame);
307static void dx_release(struct dx_frame *frames);
308static int dx_make_map(struct inode *dir, struct buffer_head *bh,
309 struct dx_hash_info *hinfo,
310 struct dx_map_entry *map_tail);
311static void dx_sort_map(struct dx_map_entry *map, unsigned count);
312static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
313 char *to, struct dx_map_entry *offsets,
314 int count, unsigned int blocksize);
315static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
316 unsigned int blocksize);
317static void dx_insert_block(struct dx_frame *frame,
318 u32 hash, ext4_lblk_t block);
319static int ext4_htree_next_block(struct inode *dir, __u32 hash,
320 struct dx_frame *frame,
321 struct dx_frame *frames,
322 __u32 *start_hash);
323static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
324 struct ext4_filename *fname,
325 struct ext4_dir_entry_2 **res_dir);
326static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
327 struct inode *dir, struct inode *inode);
328
329/* checksumming functions */
330void ext4_initialize_dirent_tail(struct buffer_head *bh,
331 unsigned int blocksize)
332{
333 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
334
335 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
336 t->det_rec_len = ext4_rec_len_to_disk(
337 sizeof(struct ext4_dir_entry_tail), blocksize);
338 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
339}
340
341/* Walk through a dirent block to find a checksum "dirent" at the tail */
342static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
343 struct buffer_head *bh)
344{
345 struct ext4_dir_entry_tail *t;
346 int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
347
348#ifdef PARANOID
349 struct ext4_dir_entry *d, *top;
350
351 d = (struct ext4_dir_entry *)bh->b_data;
352 top = (struct ext4_dir_entry *)(bh->b_data +
353 (blocksize - sizeof(struct ext4_dir_entry_tail)));
354 while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
355 d = (struct ext4_dir_entry *)(((void *)d) +
356 ext4_rec_len_from_disk(d->rec_len, blocksize));
357
358 if (d != top)
359 return NULL;
360
361 t = (struct ext4_dir_entry_tail *)d;
362#else
363 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
364#endif
365
366 if (t->det_reserved_zero1 ||
367 (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
368 sizeof(struct ext4_dir_entry_tail)) ||
369 t->det_reserved_zero2 ||
370 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
371 return NULL;
372
373 return t;
374}
375
376static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
377{
378 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
379 struct ext4_inode_info *ei = EXT4_I(inode);
380 __u32 csum;
381
382 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
383 return cpu_to_le32(csum);
384}
385
386#define warn_no_space_for_csum(inode) \
387 __warn_no_space_for_csum((inode), __func__, __LINE__)
388
389static void __warn_no_space_for_csum(struct inode *inode, const char *func,
390 unsigned int line)
391{
392 __ext4_warning_inode(inode, func, line,
393 "No space for directory leaf checksum. Please run e2fsck -D.");
394}
395
396int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
397{
398 struct ext4_dir_entry_tail *t;
399
400 if (!ext4_has_metadata_csum(inode->i_sb))
401 return 1;
402
403 t = get_dirent_tail(inode, bh);
404 if (!t) {
405 warn_no_space_for_csum(inode);
406 return 0;
407 }
408
409 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
410 (char *)t - bh->b_data))
411 return 0;
412
413 return 1;
414}
415
416static void ext4_dirblock_csum_set(struct inode *inode,
417 struct buffer_head *bh)
418{
419 struct ext4_dir_entry_tail *t;
420
421 if (!ext4_has_metadata_csum(inode->i_sb))
422 return;
423
424 t = get_dirent_tail(inode, bh);
425 if (!t) {
426 warn_no_space_for_csum(inode);
427 return;
428 }
429
430 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
431 (char *)t - bh->b_data);
432}
433
434int ext4_handle_dirty_dirblock(handle_t *handle,
435 struct inode *inode,
436 struct buffer_head *bh)
437{
438 ext4_dirblock_csum_set(inode, bh);
439 return ext4_handle_dirty_metadata(handle, inode, bh);
440}
441
442static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
443 struct ext4_dir_entry *dirent,
444 int *offset)
445{
446 struct ext4_dir_entry *dp;
447 struct dx_root_info *root;
448 int count_offset;
449 int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
450 unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
451
452 if (rlen == blocksize)
453 count_offset = 8;
454 else if (rlen == 12) {
455 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
456 if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
457 return NULL;
458 root = (struct dx_root_info *)(((void *)dp + 12));
459 if (root->reserved_zero ||
460 root->info_length != sizeof(struct dx_root_info))
461 return NULL;
462 count_offset = 32;
463 } else
464 return NULL;
465
466 if (offset)
467 *offset = count_offset;
468 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
469}
470
471static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
472 int count_offset, int count, struct dx_tail *t)
473{
474 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
475 struct ext4_inode_info *ei = EXT4_I(inode);
476 __u32 csum;
477 int size;
478 __u32 dummy_csum = 0;
479 int offset = offsetof(struct dx_tail, dt_checksum);
480
481 size = count_offset + (count * sizeof(struct dx_entry));
482 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
483 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
484 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
485
486 return cpu_to_le32(csum);
487}
488
489static int ext4_dx_csum_verify(struct inode *inode,
490 struct ext4_dir_entry *dirent)
491{
492 struct dx_countlimit *c;
493 struct dx_tail *t;
494 int count_offset, limit, count;
495
496 if (!ext4_has_metadata_csum(inode->i_sb))
497 return 1;
498
499 c = get_dx_countlimit(inode, dirent, &count_offset);
500 if (!c) {
501 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
502 return 0;
503 }
504 limit = le16_to_cpu(c->limit);
505 count = le16_to_cpu(c->count);
506 if (count_offset + (limit * sizeof(struct dx_entry)) >
507 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
508 warn_no_space_for_csum(inode);
509 return 0;
510 }
511 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
512
513 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
514 count, t))
515 return 0;
516 return 1;
517}
518
519static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
520{
521 struct dx_countlimit *c;
522 struct dx_tail *t;
523 int count_offset, limit, count;
524
525 if (!ext4_has_metadata_csum(inode->i_sb))
526 return;
527
528 c = get_dx_countlimit(inode, dirent, &count_offset);
529 if (!c) {
530 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
531 return;
532 }
533 limit = le16_to_cpu(c->limit);
534 count = le16_to_cpu(c->count);
535 if (count_offset + (limit * sizeof(struct dx_entry)) >
536 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
537 warn_no_space_for_csum(inode);
538 return;
539 }
540 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
541
542 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
543}
544
545static inline int ext4_handle_dirty_dx_node(handle_t *handle,
546 struct inode *inode,
547 struct buffer_head *bh)
548{
549 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
550 return ext4_handle_dirty_metadata(handle, inode, bh);
551}
552
553/*
554 * p is at least 6 bytes before the end of page
555 */
556static inline struct ext4_dir_entry_2 *
557ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
558{
559 return (struct ext4_dir_entry_2 *)((char *)p +
560 ext4_rec_len_from_disk(p->rec_len, blocksize));
561}
562
563/*
564 * Future: use high four bits of block for coalesce-on-delete flags
565 * Mask them off for now.
566 */
567
568static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
569{
570 return le32_to_cpu(entry->block) & 0x0fffffff;
571}
572
573static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
574{
575 entry->block = cpu_to_le32(value);
576}
577
578static inline unsigned dx_get_hash(struct dx_entry *entry)
579{
580 return le32_to_cpu(entry->hash);
581}
582
583static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
584{
585 entry->hash = cpu_to_le32(value);
586}
587
588static inline unsigned dx_get_count(struct dx_entry *entries)
589{
590 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
591}
592
593static inline unsigned dx_get_limit(struct dx_entry *entries)
594{
595 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
596}
597
598static inline void dx_set_count(struct dx_entry *entries, unsigned value)
599{
600 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
601}
602
603static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
604{
605 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
606}
607
608static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
609{
610 unsigned int entry_space = dir->i_sb->s_blocksize -
611 ext4_dir_rec_len(1, NULL) -
612 ext4_dir_rec_len(2, NULL) - infosize;
613
614 if (ext4_has_metadata_csum(dir->i_sb))
615 entry_space -= sizeof(struct dx_tail);
616 return entry_space / sizeof(struct dx_entry);
617}
618
619static inline unsigned dx_node_limit(struct inode *dir)
620{
621 unsigned int entry_space = dir->i_sb->s_blocksize -
622 ext4_dir_rec_len(0, dir);
623
624 if (ext4_has_metadata_csum(dir->i_sb))
625 entry_space -= sizeof(struct dx_tail);
626 return entry_space / sizeof(struct dx_entry);
627}
628
629/*
630 * Debug
631 */
632#ifdef DX_DEBUG
633static void dx_show_index(char * label, struct dx_entry *entries)
634{
635 int i, n = dx_get_count (entries);
636 printk(KERN_DEBUG "%s index", label);
637 for (i = 0; i < n; i++) {
638 printk(KERN_CONT " %x->%lu",
639 i ? dx_get_hash(entries + i) : 0,
640 (unsigned long)dx_get_block(entries + i));
641 }
642 printk(KERN_CONT "\n");
643}
644
645struct stats
646{
647 unsigned names;
648 unsigned space;
649 unsigned bcount;
650};
651
652static struct stats dx_show_leaf(struct inode *dir,
653 struct dx_hash_info *hinfo,
654 struct ext4_dir_entry_2 *de,
655 int size, int show_names)
656{
657 unsigned names = 0, space = 0;
658 char *base = (char *) de;
659 struct dx_hash_info h = *hinfo;
660
661 printk("names: ");
662 while ((char *) de < base + size)
663 {
664 if (de->inode)
665 {
666 if (show_names)
667 {
668#ifdef CONFIG_FS_ENCRYPTION
669 int len;
670 char *name;
671 struct fscrypt_str fname_crypto_str =
672 FSTR_INIT(NULL, 0);
673 int res = 0;
674
675 name = de->name;
676 len = de->name_len;
677 if (!IS_ENCRYPTED(dir)) {
678 /* Directory is not encrypted */
679 (void) ext4fs_dirhash(dir, de->name,
680 de->name_len, &h);
681 printk("%*.s:(U)%x.%u ", len,
682 name, h.hash,
683 (unsigned) ((char *) de
684 - base));
685 } else {
686 struct fscrypt_str de_name =
687 FSTR_INIT(name, len);
688
689 /* Directory is encrypted */
690 res = fscrypt_fname_alloc_buffer(
691 len, &fname_crypto_str);
692 if (res)
693 printk(KERN_WARNING "Error "
694 "allocating crypto "
695 "buffer--skipping "
696 "crypto\n");
697 res = fscrypt_fname_disk_to_usr(dir,
698 0, 0, &de_name,
699 &fname_crypto_str);
700 if (res) {
701 printk(KERN_WARNING "Error "
702 "converting filename "
703 "from disk to usr"
704 "\n");
705 name = "??";
706 len = 2;
707 } else {
708 name = fname_crypto_str.name;
709 len = fname_crypto_str.len;
710 }
711 if (IS_CASEFOLDED(dir))
712 h.hash = EXT4_DIRENT_HASH(de);
713 else
714 (void) ext4fs_dirhash(dir,
715 de->name,
716 de->name_len, &h);
717 printk("%*.s:(E)%x.%u ", len, name,
718 h.hash, (unsigned) ((char *) de
719 - base));
720 fscrypt_fname_free_buffer(
721 &fname_crypto_str);
722 }
723#else
724 int len = de->name_len;
725 char *name = de->name;
726 (void) ext4fs_dirhash(dir, de->name,
727 de->name_len, &h);
728 printk("%*.s:%x.%u ", len, name, h.hash,
729 (unsigned) ((char *) de - base));
730#endif
731 }
732 space += ext4_dir_rec_len(de->name_len, dir);
733 names++;
734 }
735 de = ext4_next_entry(de, size);
736 }
737 printk(KERN_CONT "(%i)\n", names);
738 return (struct stats) { names, space, 1 };
739}
740
741struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
742 struct dx_entry *entries, int levels)
743{
744 unsigned blocksize = dir->i_sb->s_blocksize;
745 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
746 unsigned bcount = 0;
747 struct buffer_head *bh;
748 printk("%i indexed blocks...\n", count);
749 for (i = 0; i < count; i++, entries++)
750 {
751 ext4_lblk_t block = dx_get_block(entries);
752 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
753 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
754 struct stats stats;
755 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
756 bh = ext4_bread(NULL,dir, block, 0);
757 if (!bh || IS_ERR(bh))
758 continue;
759 stats = levels?
760 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
761 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
762 bh->b_data, blocksize, 0);
763 names += stats.names;
764 space += stats.space;
765 bcount += stats.bcount;
766 brelse(bh);
767 }
768 if (bcount)
769 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
770 levels ? "" : " ", names, space/bcount,
771 (space/bcount)*100/blocksize);
772 return (struct stats) { names, space, bcount};
773}
774
775/*
776 * Linear search cross check
777 */
778static inline void htree_rep_invariant_check(struct dx_entry *at,
779 struct dx_entry *target,
780 u32 hash, unsigned int n)
781{
782 while (n--) {
783 dxtrace(printk(KERN_CONT ","));
784 if (dx_get_hash(++at) > hash) {
785 at--;
786 break;
787 }
788 }
789 ASSERT(at == target - 1);
790}
791#else /* DX_DEBUG */
792static inline void htree_rep_invariant_check(struct dx_entry *at,
793 struct dx_entry *target,
794 u32 hash, unsigned int n)
795{
796}
797#endif /* DX_DEBUG */
798
799/*
800 * Probe for a directory leaf block to search.
801 *
802 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
803 * error in the directory index, and the caller should fall back to
804 * searching the directory normally. The callers of dx_probe **MUST**
805 * check for this error code, and make sure it never gets reflected
806 * back to userspace.
807 */
808static struct dx_frame *
809dx_probe(struct ext4_filename *fname, struct inode *dir,
810 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
811{
812 unsigned count, indirect, level, i;
813 struct dx_entry *at, *entries, *p, *q, *m;
814 struct dx_root *root;
815 struct dx_frame *frame = frame_in;
816 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
817 u32 hash;
818 ext4_lblk_t block;
819 ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
820
821 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
822 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
823 if (IS_ERR(frame->bh))
824 return (struct dx_frame *) frame->bh;
825
826 root = (struct dx_root *) frame->bh->b_data;
827 if (root->info.hash_version != DX_HASH_TEA &&
828 root->info.hash_version != DX_HASH_HALF_MD4 &&
829 root->info.hash_version != DX_HASH_LEGACY &&
830 root->info.hash_version != DX_HASH_SIPHASH) {
831 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
832 root->info.hash_version);
833 goto fail;
834 }
835 if (ext4_hash_in_dirent(dir)) {
836 if (root->info.hash_version != DX_HASH_SIPHASH) {
837 ext4_warning_inode(dir,
838 "Hash in dirent, but hash is not SIPHASH");
839 goto fail;
840 }
841 } else {
842 if (root->info.hash_version == DX_HASH_SIPHASH) {
843 ext4_warning_inode(dir,
844 "Hash code is SIPHASH, but hash not in dirent");
845 goto fail;
846 }
847 }
848 if (fname)
849 hinfo = &fname->hinfo;
850 hinfo->hash_version = root->info.hash_version;
851 if (hinfo->hash_version <= DX_HASH_TEA)
852 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
853 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
854 /* hash is already computed for encrypted casefolded directory */
855 if (fname && fname_name(fname) &&
856 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
857 int ret = ext4fs_dirhash(dir, fname_name(fname),
858 fname_len(fname), hinfo);
859 if (ret < 0) {
860 ret_err = ERR_PTR(ret);
861 goto fail;
862 }
863 }
864 hash = hinfo->hash;
865
866 if (root->info.unused_flags & 1) {
867 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
868 root->info.unused_flags);
869 goto fail;
870 }
871
872 indirect = root->info.indirect_levels;
873 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
874 ext4_warning(dir->i_sb,
875 "Directory (ino: %lu) htree depth %#06x exceed"
876 "supported value", dir->i_ino,
877 ext4_dir_htree_level(dir->i_sb));
878 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
879 ext4_warning(dir->i_sb, "Enable large directory "
880 "feature to access it");
881 }
882 goto fail;
883 }
884
885 entries = (struct dx_entry *)(((char *)&root->info) +
886 root->info.info_length);
887
888 if (dx_get_limit(entries) != dx_root_limit(dir,
889 root->info.info_length)) {
890 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
891 dx_get_limit(entries),
892 dx_root_limit(dir, root->info.info_length));
893 goto fail;
894 }
895
896 dxtrace(printk("Look up %x", hash));
897 level = 0;
898 blocks[0] = 0;
899 while (1) {
900 count = dx_get_count(entries);
901 if (!count || count > dx_get_limit(entries)) {
902 ext4_warning_inode(dir,
903 "dx entry: count %u beyond limit %u",
904 count, dx_get_limit(entries));
905 goto fail;
906 }
907
908 p = entries + 1;
909 q = entries + count - 1;
910 while (p <= q) {
911 m = p + (q - p) / 2;
912 dxtrace(printk(KERN_CONT "."));
913 if (dx_get_hash(m) > hash)
914 q = m - 1;
915 else
916 p = m + 1;
917 }
918
919 htree_rep_invariant_check(entries, p, hash, count - 1);
920
921 at = p - 1;
922 dxtrace(printk(KERN_CONT " %x->%u\n",
923 at == entries ? 0 : dx_get_hash(at),
924 dx_get_block(at)));
925 frame->entries = entries;
926 frame->at = at;
927
928 block = dx_get_block(at);
929 for (i = 0; i <= level; i++) {
930 if (blocks[i] == block) {
931 ext4_warning_inode(dir,
932 "dx entry: tree cycle block %u points back to block %u",
933 blocks[level], block);
934 goto fail;
935 }
936 }
937 if (++level > indirect)
938 return frame;
939 blocks[level] = block;
940 frame++;
941 frame->bh = ext4_read_dirblock(dir, block, INDEX);
942 if (IS_ERR(frame->bh)) {
943 ret_err = (struct dx_frame *) frame->bh;
944 frame->bh = NULL;
945 goto fail;
946 }
947
948 entries = ((struct dx_node *) frame->bh->b_data)->entries;
949
950 if (dx_get_limit(entries) != dx_node_limit(dir)) {
951 ext4_warning_inode(dir,
952 "dx entry: limit %u != node limit %u",
953 dx_get_limit(entries), dx_node_limit(dir));
954 goto fail;
955 }
956 }
957fail:
958 while (frame >= frame_in) {
959 brelse(frame->bh);
960 frame--;
961 }
962
963 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
964 ext4_warning_inode(dir,
965 "Corrupt directory, running e2fsck is recommended");
966 return ret_err;
967}
968
969static void dx_release(struct dx_frame *frames)
970{
971 struct dx_root_info *info;
972 int i;
973 unsigned int indirect_levels;
974
975 if (frames[0].bh == NULL)
976 return;
977
978 info = &((struct dx_root *)frames[0].bh->b_data)->info;
979 /* save local copy, "info" may be freed after brelse() */
980 indirect_levels = info->indirect_levels;
981 for (i = 0; i <= indirect_levels; i++) {
982 if (frames[i].bh == NULL)
983 break;
984 brelse(frames[i].bh);
985 frames[i].bh = NULL;
986 }
987}
988
989/*
990 * This function increments the frame pointer to search the next leaf
991 * block, and reads in the necessary intervening nodes if the search
992 * should be necessary. Whether or not the search is necessary is
993 * controlled by the hash parameter. If the hash value is even, then
994 * the search is only continued if the next block starts with that
995 * hash value. This is used if we are searching for a specific file.
996 *
997 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
998 *
999 * This function returns 1 if the caller should continue to search,
1000 * or 0 if it should not. If there is an error reading one of the
1001 * index blocks, it will a negative error code.
1002 *
1003 * If start_hash is non-null, it will be filled in with the starting
1004 * hash of the next page.
1005 */
1006static int ext4_htree_next_block(struct inode *dir, __u32 hash,
1007 struct dx_frame *frame,
1008 struct dx_frame *frames,
1009 __u32 *start_hash)
1010{
1011 struct dx_frame *p;
1012 struct buffer_head *bh;
1013 int num_frames = 0;
1014 __u32 bhash;
1015
1016 p = frame;
1017 /*
1018 * Find the next leaf page by incrementing the frame pointer.
1019 * If we run out of entries in the interior node, loop around and
1020 * increment pointer in the parent node. When we break out of
1021 * this loop, num_frames indicates the number of interior
1022 * nodes need to be read.
1023 */
1024 while (1) {
1025 if (++(p->at) < p->entries + dx_get_count(p->entries))
1026 break;
1027 if (p == frames)
1028 return 0;
1029 num_frames++;
1030 p--;
1031 }
1032
1033 /*
1034 * If the hash is 1, then continue only if the next page has a
1035 * continuation hash of any value. This is used for readdir
1036 * handling. Otherwise, check to see if the hash matches the
1037 * desired continuation hash. If it doesn't, return since
1038 * there's no point to read in the successive index pages.
1039 */
1040 bhash = dx_get_hash(p->at);
1041 if (start_hash)
1042 *start_hash = bhash;
1043 if ((hash & 1) == 0) {
1044 if ((bhash & ~1) != hash)
1045 return 0;
1046 }
1047 /*
1048 * If the hash is HASH_NB_ALWAYS, we always go to the next
1049 * block so no check is necessary
1050 */
1051 while (num_frames--) {
1052 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1053 if (IS_ERR(bh))
1054 return PTR_ERR(bh);
1055 p++;
1056 brelse(p->bh);
1057 p->bh = bh;
1058 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1059 }
1060 return 1;
1061}
1062
1063
1064/*
1065 * This function fills a red-black tree with information from a
1066 * directory block. It returns the number directory entries loaded
1067 * into the tree. If there is an error it is returned in err.
1068 */
1069static int htree_dirblock_to_tree(struct file *dir_file,
1070 struct inode *dir, ext4_lblk_t block,
1071 struct dx_hash_info *hinfo,
1072 __u32 start_hash, __u32 start_minor_hash)
1073{
1074 struct buffer_head *bh;
1075 struct ext4_dir_entry_2 *de, *top;
1076 int err = 0, count = 0;
1077 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1078 int csum = ext4_has_metadata_csum(dir->i_sb);
1079
1080 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1081 (unsigned long)block));
1082 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1083 if (IS_ERR(bh))
1084 return PTR_ERR(bh);
1085
1086 de = (struct ext4_dir_entry_2 *) bh->b_data;
1087 /* csum entries are not larger in the casefolded encrypted case */
1088 top = (struct ext4_dir_entry_2 *) ((char *) de +
1089 dir->i_sb->s_blocksize -
1090 ext4_dir_rec_len(0,
1091 csum ? NULL : dir));
1092 /* Check if the directory is encrypted */
1093 if (IS_ENCRYPTED(dir)) {
1094 err = fscrypt_prepare_readdir(dir);
1095 if (err < 0) {
1096 brelse(bh);
1097 return err;
1098 }
1099 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1100 &fname_crypto_str);
1101 if (err < 0) {
1102 brelse(bh);
1103 return err;
1104 }
1105 }
1106
1107 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1108 if (ext4_check_dir_entry(dir, NULL, de, bh,
1109 bh->b_data, bh->b_size,
1110 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1111 + ((char *)de - bh->b_data))) {
1112 /* silently ignore the rest of the block */
1113 break;
1114 }
1115 if (ext4_hash_in_dirent(dir)) {
1116 if (de->name_len && de->inode) {
1117 hinfo->hash = EXT4_DIRENT_HASH(de);
1118 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1119 } else {
1120 hinfo->hash = 0;
1121 hinfo->minor_hash = 0;
1122 }
1123 } else {
1124 err = ext4fs_dirhash(dir, de->name,
1125 de->name_len, hinfo);
1126 if (err < 0) {
1127 count = err;
1128 goto errout;
1129 }
1130 }
1131 if ((hinfo->hash < start_hash) ||
1132 ((hinfo->hash == start_hash) &&
1133 (hinfo->minor_hash < start_minor_hash)))
1134 continue;
1135 if (de->inode == 0)
1136 continue;
1137 if (!IS_ENCRYPTED(dir)) {
1138 tmp_str.name = de->name;
1139 tmp_str.len = de->name_len;
1140 err = ext4_htree_store_dirent(dir_file,
1141 hinfo->hash, hinfo->minor_hash, de,
1142 &tmp_str);
1143 } else {
1144 int save_len = fname_crypto_str.len;
1145 struct fscrypt_str de_name = FSTR_INIT(de->name,
1146 de->name_len);
1147
1148 /* Directory is encrypted */
1149 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1150 hinfo->minor_hash, &de_name,
1151 &fname_crypto_str);
1152 if (err) {
1153 count = err;
1154 goto errout;
1155 }
1156 err = ext4_htree_store_dirent(dir_file,
1157 hinfo->hash, hinfo->minor_hash, de,
1158 &fname_crypto_str);
1159 fname_crypto_str.len = save_len;
1160 }
1161 if (err != 0) {
1162 count = err;
1163 goto errout;
1164 }
1165 count++;
1166 }
1167errout:
1168 brelse(bh);
1169 fscrypt_fname_free_buffer(&fname_crypto_str);
1170 return count;
1171}
1172
1173
1174/*
1175 * This function fills a red-black tree with information from a
1176 * directory. We start scanning the directory in hash order, starting
1177 * at start_hash and start_minor_hash.
1178 *
1179 * This function returns the number of entries inserted into the tree,
1180 * or a negative error code.
1181 */
1182int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1183 __u32 start_minor_hash, __u32 *next_hash)
1184{
1185 struct dx_hash_info hinfo;
1186 struct ext4_dir_entry_2 *de;
1187 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1188 struct inode *dir;
1189 ext4_lblk_t block;
1190 int count = 0;
1191 int ret, err;
1192 __u32 hashval;
1193 struct fscrypt_str tmp_str;
1194
1195 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1196 start_hash, start_minor_hash));
1197 dir = file_inode(dir_file);
1198 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1199 if (ext4_hash_in_dirent(dir))
1200 hinfo.hash_version = DX_HASH_SIPHASH;
1201 else
1202 hinfo.hash_version =
1203 EXT4_SB(dir->i_sb)->s_def_hash_version;
1204 if (hinfo.hash_version <= DX_HASH_TEA)
1205 hinfo.hash_version +=
1206 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1207 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1208 if (ext4_has_inline_data(dir)) {
1209 int has_inline_data = 1;
1210 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1211 &hinfo, start_hash,
1212 start_minor_hash,
1213 &has_inline_data);
1214 if (has_inline_data) {
1215 *next_hash = ~0;
1216 return count;
1217 }
1218 }
1219 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1220 start_hash, start_minor_hash);
1221 *next_hash = ~0;
1222 return count;
1223 }
1224 hinfo.hash = start_hash;
1225 hinfo.minor_hash = 0;
1226 frame = dx_probe(NULL, dir, &hinfo, frames);
1227 if (IS_ERR(frame))
1228 return PTR_ERR(frame);
1229
1230 /* Add '.' and '..' from the htree header */
1231 if (!start_hash && !start_minor_hash) {
1232 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1233 tmp_str.name = de->name;
1234 tmp_str.len = de->name_len;
1235 err = ext4_htree_store_dirent(dir_file, 0, 0,
1236 de, &tmp_str);
1237 if (err != 0)
1238 goto errout;
1239 count++;
1240 }
1241 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1242 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1243 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1244 tmp_str.name = de->name;
1245 tmp_str.len = de->name_len;
1246 err = ext4_htree_store_dirent(dir_file, 2, 0,
1247 de, &tmp_str);
1248 if (err != 0)
1249 goto errout;
1250 count++;
1251 }
1252
1253 while (1) {
1254 if (fatal_signal_pending(current)) {
1255 err = -ERESTARTSYS;
1256 goto errout;
1257 }
1258 cond_resched();
1259 block = dx_get_block(frame->at);
1260 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1261 start_hash, start_minor_hash);
1262 if (ret < 0) {
1263 err = ret;
1264 goto errout;
1265 }
1266 count += ret;
1267 hashval = ~0;
1268 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1269 frame, frames, &hashval);
1270 *next_hash = hashval;
1271 if (ret < 0) {
1272 err = ret;
1273 goto errout;
1274 }
1275 /*
1276 * Stop if: (a) there are no more entries, or
1277 * (b) we have inserted at least one entry and the
1278 * next hash value is not a continuation
1279 */
1280 if ((ret == 0) ||
1281 (count && ((hashval & 1) == 0)))
1282 break;
1283 }
1284 dx_release(frames);
1285 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1286 "next hash: %x\n", count, *next_hash));
1287 return count;
1288errout:
1289 dx_release(frames);
1290 return (err);
1291}
1292
1293static inline int search_dirblock(struct buffer_head *bh,
1294 struct inode *dir,
1295 struct ext4_filename *fname,
1296 unsigned int offset,
1297 struct ext4_dir_entry_2 **res_dir)
1298{
1299 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1300 fname, offset, res_dir);
1301}
1302
1303/*
1304 * Directory block splitting, compacting
1305 */
1306
1307/*
1308 * Create map of hash values, offsets, and sizes, stored at end of block.
1309 * Returns number of entries mapped.
1310 */
1311static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1312 struct dx_hash_info *hinfo,
1313 struct dx_map_entry *map_tail)
1314{
1315 int count = 0;
1316 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1317 unsigned int buflen = bh->b_size;
1318 char *base = bh->b_data;
1319 struct dx_hash_info h = *hinfo;
1320 int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
1321
1322 if (ext4_has_metadata_csum(dir->i_sb))
1323 buflen -= sizeof(struct ext4_dir_entry_tail);
1324
1325 while ((char *) de < base + buflen) {
1326 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1327 ((char *)de) - base))
1328 return -EFSCORRUPTED;
1329 if (de->name_len && de->inode) {
1330 if (ext4_hash_in_dirent(dir))
1331 h.hash = EXT4_DIRENT_HASH(de);
1332 else {
1333 int err = ext4fs_dirhash(dir, de->name,
1334 de->name_len, &h);
1335 if (err < 0)
1336 return err;
1337 }
1338 map_tail--;
1339 map_tail->hash = h.hash;
1340 map_tail->offs = ((char *) de - base)>>2;
1341 map_tail->size = ext4_rec_len_from_disk(de->rec_len,
1342 blocksize);
1343 count++;
1344 cond_resched();
1345 }
1346 de = ext4_next_entry(de, blocksize);
1347 }
1348 return count;
1349}
1350
1351/* Sort map by hash value */
1352static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1353{
1354 struct dx_map_entry *p, *q, *top = map + count - 1;
1355 int more;
1356 /* Combsort until bubble sort doesn't suck */
1357 while (count > 2) {
1358 count = count*10/13;
1359 if (count - 9 < 2) /* 9, 10 -> 11 */
1360 count = 11;
1361 for (p = top, q = p - count; q >= map; p--, q--)
1362 if (p->hash < q->hash)
1363 swap(*p, *q);
1364 }
1365 /* Garden variety bubble sort */
1366 do {
1367 more = 0;
1368 q = top;
1369 while (q-- > map) {
1370 if (q[1].hash >= q[0].hash)
1371 continue;
1372 swap(*(q+1), *q);
1373 more = 1;
1374 }
1375 } while(more);
1376}
1377
1378static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1379{
1380 struct dx_entry *entries = frame->entries;
1381 struct dx_entry *old = frame->at, *new = old + 1;
1382 int count = dx_get_count(entries);
1383
1384 ASSERT(count < dx_get_limit(entries));
1385 ASSERT(old < entries + count);
1386 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1387 dx_set_hash(new, hash);
1388 dx_set_block(new, block);
1389 dx_set_count(entries, count + 1);
1390}
1391
1392#if IS_ENABLED(CONFIG_UNICODE)
1393/*
1394 * Test whether a case-insensitive directory entry matches the filename
1395 * being searched for. If quick is set, assume the name being looked up
1396 * is already in the casefolded form.
1397 *
1398 * Returns: 0 if the directory entry matches, more than 0 if it
1399 * doesn't match or less than zero on error.
1400 */
1401static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1402 u8 *de_name, size_t de_name_len, bool quick)
1403{
1404 const struct super_block *sb = parent->i_sb;
1405 const struct unicode_map *um = sb->s_encoding;
1406 struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1407 struct qstr entry = QSTR_INIT(de_name, de_name_len);
1408 int ret;
1409
1410 if (IS_ENCRYPTED(parent)) {
1411 const struct fscrypt_str encrypted_name =
1412 FSTR_INIT(de_name, de_name_len);
1413
1414 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1415 if (!decrypted_name.name)
1416 return -ENOMEM;
1417 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1418 &decrypted_name);
1419 if (ret < 0)
1420 goto out;
1421 entry.name = decrypted_name.name;
1422 entry.len = decrypted_name.len;
1423 }
1424
1425 if (quick)
1426 ret = utf8_strncasecmp_folded(um, name, &entry);
1427 else
1428 ret = utf8_strncasecmp(um, name, &entry);
1429 if (ret < 0) {
1430 /* Handle invalid character sequence as either an error
1431 * or as an opaque byte sequence.
1432 */
1433 if (sb_has_strict_encoding(sb))
1434 ret = -EINVAL;
1435 else if (name->len != entry.len)
1436 ret = 1;
1437 else
1438 ret = !!memcmp(name->name, entry.name, entry.len);
1439 }
1440out:
1441 kfree(decrypted_name.name);
1442 return ret;
1443}
1444
1445int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1446 struct ext4_filename *name)
1447{
1448 struct fscrypt_str *cf_name = &name->cf_name;
1449 struct dx_hash_info *hinfo = &name->hinfo;
1450 int len;
1451
1452 if (!IS_CASEFOLDED(dir) ||
1453 (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1454 cf_name->name = NULL;
1455 return 0;
1456 }
1457
1458 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1459 if (!cf_name->name)
1460 return -ENOMEM;
1461
1462 len = utf8_casefold(dir->i_sb->s_encoding,
1463 iname, cf_name->name,
1464 EXT4_NAME_LEN);
1465 if (len <= 0) {
1466 kfree(cf_name->name);
1467 cf_name->name = NULL;
1468 }
1469 cf_name->len = (unsigned) len;
1470 if (!IS_ENCRYPTED(dir))
1471 return 0;
1472
1473 hinfo->hash_version = DX_HASH_SIPHASH;
1474 hinfo->seed = NULL;
1475 if (cf_name->name)
1476 return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1477 else
1478 return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1479}
1480#endif
1481
1482/*
1483 * Test whether a directory entry matches the filename being searched for.
1484 *
1485 * Return: %true if the directory entry matches, otherwise %false.
1486 */
1487static bool ext4_match(struct inode *parent,
1488 const struct ext4_filename *fname,
1489 struct ext4_dir_entry_2 *de)
1490{
1491 struct fscrypt_name f;
1492
1493 if (!de->inode)
1494 return false;
1495
1496 f.usr_fname = fname->usr_fname;
1497 f.disk_name = fname->disk_name;
1498#ifdef CONFIG_FS_ENCRYPTION
1499 f.crypto_buf = fname->crypto_buf;
1500#endif
1501
1502#if IS_ENABLED(CONFIG_UNICODE)
1503 if (IS_CASEFOLDED(parent) &&
1504 (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1505 if (fname->cf_name.name) {
1506 struct qstr cf = {.name = fname->cf_name.name,
1507 .len = fname->cf_name.len};
1508 if (IS_ENCRYPTED(parent)) {
1509 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1510 fname->hinfo.minor_hash !=
1511 EXT4_DIRENT_MINOR_HASH(de)) {
1512
1513 return false;
1514 }
1515 }
1516 return !ext4_ci_compare(parent, &cf, de->name,
1517 de->name_len, true);
1518 }
1519 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1520 de->name_len, false);
1521 }
1522#endif
1523
1524 return fscrypt_match_name(&f, de->name, de->name_len);
1525}
1526
1527/*
1528 * Returns 0 if not found, -1 on failure, and 1 on success
1529 */
1530int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1531 struct inode *dir, struct ext4_filename *fname,
1532 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1533{
1534 struct ext4_dir_entry_2 * de;
1535 char * dlimit;
1536 int de_len;
1537
1538 de = (struct ext4_dir_entry_2 *)search_buf;
1539 dlimit = search_buf + buf_size;
1540 while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1541 /* this code is executed quadratically often */
1542 /* do minimal checking `by hand' */
1543 if (de->name + de->name_len <= dlimit &&
1544 ext4_match(dir, fname, de)) {
1545 /* found a match - just to be sure, do
1546 * a full check */
1547 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1548 buf_size, offset))
1549 return -1;
1550 *res_dir = de;
1551 return 1;
1552 }
1553 /* prevent looping on a bad block */
1554 de_len = ext4_rec_len_from_disk(de->rec_len,
1555 dir->i_sb->s_blocksize);
1556 if (de_len <= 0)
1557 return -1;
1558 offset += de_len;
1559 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1560 }
1561 return 0;
1562}
1563
1564static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1565 struct ext4_dir_entry *de)
1566{
1567 struct super_block *sb = dir->i_sb;
1568
1569 if (!is_dx(dir))
1570 return 0;
1571 if (block == 0)
1572 return 1;
1573 if (de->inode == 0 &&
1574 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1575 sb->s_blocksize)
1576 return 1;
1577 return 0;
1578}
1579
1580/*
1581 * __ext4_find_entry()
1582 *
1583 * finds an entry in the specified directory with the wanted name. It
1584 * returns the cache buffer in which the entry was found, and the entry
1585 * itself (as a parameter - res_dir). It does NOT read the inode of the
1586 * entry - you'll have to do that yourself if you want to.
1587 *
1588 * The returned buffer_head has ->b_count elevated. The caller is expected
1589 * to brelse() it when appropriate.
1590 */
1591static struct buffer_head *__ext4_find_entry(struct inode *dir,
1592 struct ext4_filename *fname,
1593 struct ext4_dir_entry_2 **res_dir,
1594 int *inlined)
1595{
1596 struct super_block *sb;
1597 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1598 struct buffer_head *bh, *ret = NULL;
1599 ext4_lblk_t start, block;
1600 const u8 *name = fname->usr_fname->name;
1601 size_t ra_max = 0; /* Number of bh's in the readahead
1602 buffer, bh_use[] */
1603 size_t ra_ptr = 0; /* Current index into readahead
1604 buffer */
1605 ext4_lblk_t nblocks;
1606 int i, namelen, retval;
1607
1608 *res_dir = NULL;
1609 sb = dir->i_sb;
1610 namelen = fname->usr_fname->len;
1611 if (namelen > EXT4_NAME_LEN)
1612 return NULL;
1613
1614 if (ext4_has_inline_data(dir)) {
1615 int has_inline_data = 1;
1616 ret = ext4_find_inline_entry(dir, fname, res_dir,
1617 &has_inline_data);
1618 if (inlined)
1619 *inlined = has_inline_data;
1620 if (has_inline_data)
1621 goto cleanup_and_exit;
1622 }
1623
1624 if ((namelen <= 2) && (name[0] == '.') &&
1625 (name[1] == '.' || name[1] == '\0')) {
1626 /*
1627 * "." or ".." will only be in the first block
1628 * NFS may look up ".."; "." should be handled by the VFS
1629 */
1630 block = start = 0;
1631 nblocks = 1;
1632 goto restart;
1633 }
1634 if (is_dx(dir)) {
1635 ret = ext4_dx_find_entry(dir, fname, res_dir);
1636 /*
1637 * On success, or if the error was file not found,
1638 * return. Otherwise, fall back to doing a search the
1639 * old fashioned way.
1640 */
1641 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1642 goto cleanup_and_exit;
1643 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1644 "falling back\n"));
1645 ret = NULL;
1646 }
1647 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1648 if (!nblocks) {
1649 ret = NULL;
1650 goto cleanup_and_exit;
1651 }
1652 start = EXT4_I(dir)->i_dir_start_lookup;
1653 if (start >= nblocks)
1654 start = 0;
1655 block = start;
1656restart:
1657 do {
1658 /*
1659 * We deal with the read-ahead logic here.
1660 */
1661 cond_resched();
1662 if (ra_ptr >= ra_max) {
1663 /* Refill the readahead buffer */
1664 ra_ptr = 0;
1665 if (block < start)
1666 ra_max = start - block;
1667 else
1668 ra_max = nblocks - block;
1669 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1670 retval = ext4_bread_batch(dir, block, ra_max,
1671 false /* wait */, bh_use);
1672 if (retval) {
1673 ret = ERR_PTR(retval);
1674 ra_max = 0;
1675 goto cleanup_and_exit;
1676 }
1677 }
1678 if ((bh = bh_use[ra_ptr++]) == NULL)
1679 goto next;
1680 wait_on_buffer(bh);
1681 if (!buffer_uptodate(bh)) {
1682 EXT4_ERROR_INODE_ERR(dir, EIO,
1683 "reading directory lblock %lu",
1684 (unsigned long) block);
1685 brelse(bh);
1686 ret = ERR_PTR(-EIO);
1687 goto cleanup_and_exit;
1688 }
1689 if (!buffer_verified(bh) &&
1690 !is_dx_internal_node(dir, block,
1691 (struct ext4_dir_entry *)bh->b_data) &&
1692 !ext4_dirblock_csum_verify(dir, bh)) {
1693 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1694 "checksumming directory "
1695 "block %lu", (unsigned long)block);
1696 brelse(bh);
1697 ret = ERR_PTR(-EFSBADCRC);
1698 goto cleanup_and_exit;
1699 }
1700 set_buffer_verified(bh);
1701 i = search_dirblock(bh, dir, fname,
1702 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1703 if (i == 1) {
1704 EXT4_I(dir)->i_dir_start_lookup = block;
1705 ret = bh;
1706 goto cleanup_and_exit;
1707 } else {
1708 brelse(bh);
1709 if (i < 0)
1710 goto cleanup_and_exit;
1711 }
1712 next:
1713 if (++block >= nblocks)
1714 block = 0;
1715 } while (block != start);
1716
1717 /*
1718 * If the directory has grown while we were searching, then
1719 * search the last part of the directory before giving up.
1720 */
1721 block = nblocks;
1722 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1723 if (block < nblocks) {
1724 start = 0;
1725 goto restart;
1726 }
1727
1728cleanup_and_exit:
1729 /* Clean up the read-ahead blocks */
1730 for (; ra_ptr < ra_max; ra_ptr++)
1731 brelse(bh_use[ra_ptr]);
1732 return ret;
1733}
1734
1735static struct buffer_head *ext4_find_entry(struct inode *dir,
1736 const struct qstr *d_name,
1737 struct ext4_dir_entry_2 **res_dir,
1738 int *inlined)
1739{
1740 int err;
1741 struct ext4_filename fname;
1742 struct buffer_head *bh;
1743
1744 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1745 if (err == -ENOENT)
1746 return NULL;
1747 if (err)
1748 return ERR_PTR(err);
1749
1750 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1751
1752 ext4_fname_free_filename(&fname);
1753 return bh;
1754}
1755
1756static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1757 struct dentry *dentry,
1758 struct ext4_dir_entry_2 **res_dir)
1759{
1760 int err;
1761 struct ext4_filename fname;
1762 struct buffer_head *bh;
1763
1764 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1765 if (err == -ENOENT)
1766 return NULL;
1767 if (err)
1768 return ERR_PTR(err);
1769
1770 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1771
1772 ext4_fname_free_filename(&fname);
1773 return bh;
1774}
1775
1776static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1777 struct ext4_filename *fname,
1778 struct ext4_dir_entry_2 **res_dir)
1779{
1780 struct super_block * sb = dir->i_sb;
1781 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1782 struct buffer_head *bh;
1783 ext4_lblk_t block;
1784 int retval;
1785
1786#ifdef CONFIG_FS_ENCRYPTION
1787 *res_dir = NULL;
1788#endif
1789 frame = dx_probe(fname, dir, NULL, frames);
1790 if (IS_ERR(frame))
1791 return (struct buffer_head *) frame;
1792 do {
1793 block = dx_get_block(frame->at);
1794 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1795 if (IS_ERR(bh))
1796 goto errout;
1797
1798 retval = search_dirblock(bh, dir, fname,
1799 block << EXT4_BLOCK_SIZE_BITS(sb),
1800 res_dir);
1801 if (retval == 1)
1802 goto success;
1803 brelse(bh);
1804 if (retval == -1) {
1805 bh = ERR_PTR(ERR_BAD_DX_DIR);
1806 goto errout;
1807 }
1808
1809 /* Check to see if we should continue to search */
1810 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1811 frames, NULL);
1812 if (retval < 0) {
1813 ext4_warning_inode(dir,
1814 "error %d reading directory index block",
1815 retval);
1816 bh = ERR_PTR(retval);
1817 goto errout;
1818 }
1819 } while (retval == 1);
1820
1821 bh = NULL;
1822errout:
1823 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1824success:
1825 dx_release(frames);
1826 return bh;
1827}
1828
1829static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1830{
1831 struct inode *inode;
1832 struct ext4_dir_entry_2 *de;
1833 struct buffer_head *bh;
1834
1835 if (dentry->d_name.len > EXT4_NAME_LEN)
1836 return ERR_PTR(-ENAMETOOLONG);
1837
1838 bh = ext4_lookup_entry(dir, dentry, &de);
1839 if (IS_ERR(bh))
1840 return ERR_CAST(bh);
1841 inode = NULL;
1842 if (bh) {
1843 __u32 ino = le32_to_cpu(de->inode);
1844 brelse(bh);
1845 if (!ext4_valid_inum(dir->i_sb, ino)) {
1846 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1847 return ERR_PTR(-EFSCORRUPTED);
1848 }
1849 if (unlikely(ino == dir->i_ino)) {
1850 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1851 dentry);
1852 return ERR_PTR(-EFSCORRUPTED);
1853 }
1854 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1855 if (inode == ERR_PTR(-ESTALE)) {
1856 EXT4_ERROR_INODE(dir,
1857 "deleted inode referenced: %u",
1858 ino);
1859 return ERR_PTR(-EFSCORRUPTED);
1860 }
1861 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1862 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1863 !fscrypt_has_permitted_context(dir, inode)) {
1864 ext4_warning(inode->i_sb,
1865 "Inconsistent encryption contexts: %lu/%lu",
1866 dir->i_ino, inode->i_ino);
1867 iput(inode);
1868 return ERR_PTR(-EPERM);
1869 }
1870 }
1871
1872#if IS_ENABLED(CONFIG_UNICODE)
1873 if (!inode && IS_CASEFOLDED(dir)) {
1874 /* Eventually we want to call d_add_ci(dentry, NULL)
1875 * for negative dentries in the encoding case as
1876 * well. For now, prevent the negative dentry
1877 * from being cached.
1878 */
1879 return NULL;
1880 }
1881#endif
1882 return d_splice_alias(inode, dentry);
1883}
1884
1885
1886struct dentry *ext4_get_parent(struct dentry *child)
1887{
1888 __u32 ino;
1889 struct ext4_dir_entry_2 * de;
1890 struct buffer_head *bh;
1891
1892 bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1893 if (IS_ERR(bh))
1894 return ERR_CAST(bh);
1895 if (!bh)
1896 return ERR_PTR(-ENOENT);
1897 ino = le32_to_cpu(de->inode);
1898 brelse(bh);
1899
1900 if (!ext4_valid_inum(child->d_sb, ino)) {
1901 EXT4_ERROR_INODE(d_inode(child),
1902 "bad parent inode number: %u", ino);
1903 return ERR_PTR(-EFSCORRUPTED);
1904 }
1905
1906 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1907}
1908
1909/*
1910 * Move count entries from end of map between two memory locations.
1911 * Returns pointer to last entry moved.
1912 */
1913static struct ext4_dir_entry_2 *
1914dx_move_dirents(struct inode *dir, char *from, char *to,
1915 struct dx_map_entry *map, int count,
1916 unsigned blocksize)
1917{
1918 unsigned rec_len = 0;
1919
1920 while (count--) {
1921 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1922 (from + (map->offs<<2));
1923 rec_len = ext4_dir_rec_len(de->name_len, dir);
1924
1925 memcpy (to, de, rec_len);
1926 ((struct ext4_dir_entry_2 *) to)->rec_len =
1927 ext4_rec_len_to_disk(rec_len, blocksize);
1928
1929 /* wipe dir_entry excluding the rec_len field */
1930 de->inode = 0;
1931 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1932 blocksize) -
1933 offsetof(struct ext4_dir_entry_2,
1934 name_len));
1935
1936 map++;
1937 to += rec_len;
1938 }
1939 return (struct ext4_dir_entry_2 *) (to - rec_len);
1940}
1941
1942/*
1943 * Compact each dir entry in the range to the minimal rec_len.
1944 * Returns pointer to last entry in range.
1945 */
1946static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1947 unsigned int blocksize)
1948{
1949 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1950 unsigned rec_len = 0;
1951
1952 prev = to = de;
1953 while ((char*)de < base + blocksize) {
1954 next = ext4_next_entry(de, blocksize);
1955 if (de->inode && de->name_len) {
1956 rec_len = ext4_dir_rec_len(de->name_len, dir);
1957 if (de > to)
1958 memmove(to, de, rec_len);
1959 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1960 prev = to;
1961 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1962 }
1963 de = next;
1964 }
1965 return prev;
1966}
1967
1968/*
1969 * Split a full leaf block to make room for a new dir entry.
1970 * Allocate a new block, and move entries so that they are approx. equally full.
1971 * Returns pointer to de in block into which the new entry will be inserted.
1972 */
1973static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1974 struct buffer_head **bh,struct dx_frame *frame,
1975 struct dx_hash_info *hinfo)
1976{
1977 unsigned blocksize = dir->i_sb->s_blocksize;
1978 unsigned continued;
1979 int count;
1980 struct buffer_head *bh2;
1981 ext4_lblk_t newblock;
1982 u32 hash2;
1983 struct dx_map_entry *map;
1984 char *data1 = (*bh)->b_data, *data2;
1985 unsigned split, move, size;
1986 struct ext4_dir_entry_2 *de = NULL, *de2;
1987 int csum_size = 0;
1988 int err = 0, i;
1989
1990 if (ext4_has_metadata_csum(dir->i_sb))
1991 csum_size = sizeof(struct ext4_dir_entry_tail);
1992
1993 bh2 = ext4_append(handle, dir, &newblock);
1994 if (IS_ERR(bh2)) {
1995 brelse(*bh);
1996 *bh = NULL;
1997 return (struct ext4_dir_entry_2 *) bh2;
1998 }
1999
2000 BUFFER_TRACE(*bh, "get_write_access");
2001 err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
2002 EXT4_JTR_NONE);
2003 if (err)
2004 goto journal_error;
2005
2006 BUFFER_TRACE(frame->bh, "get_write_access");
2007 err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
2008 EXT4_JTR_NONE);
2009 if (err)
2010 goto journal_error;
2011
2012 data2 = bh2->b_data;
2013
2014 /* create map in the end of data2 block */
2015 map = (struct dx_map_entry *) (data2 + blocksize);
2016 count = dx_make_map(dir, *bh, hinfo, map);
2017 if (count < 0) {
2018 err = count;
2019 goto journal_error;
2020 }
2021 map -= count;
2022 dx_sort_map(map, count);
2023 /* Ensure that neither split block is over half full */
2024 size = 0;
2025 move = 0;
2026 for (i = count-1; i >= 0; i--) {
2027 /* is more than half of this entry in 2nd half of the block? */
2028 if (size + map[i].size/2 > blocksize/2)
2029 break;
2030 size += map[i].size;
2031 move++;
2032 }
2033 /*
2034 * map index at which we will split
2035 *
2036 * If the sum of active entries didn't exceed half the block size, just
2037 * split it in half by count; each resulting block will have at least
2038 * half the space free.
2039 */
2040 if (i > 0)
2041 split = count - move;
2042 else
2043 split = count/2;
2044
2045 hash2 = map[split].hash;
2046 continued = hash2 == map[split - 1].hash;
2047 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2048 (unsigned long)dx_get_block(frame->at),
2049 hash2, split, count-split));
2050
2051 /* Fancy dance to stay within two buffers */
2052 de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2053 blocksize);
2054 de = dx_pack_dirents(dir, data1, blocksize);
2055 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2056 (char *) de,
2057 blocksize);
2058 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2059 (char *) de2,
2060 blocksize);
2061 if (csum_size) {
2062 ext4_initialize_dirent_tail(*bh, blocksize);
2063 ext4_initialize_dirent_tail(bh2, blocksize);
2064 }
2065
2066 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2067 blocksize, 1));
2068 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2069 blocksize, 1));
2070
2071 /* Which block gets the new entry? */
2072 if (hinfo->hash >= hash2) {
2073 swap(*bh, bh2);
2074 de = de2;
2075 }
2076 dx_insert_block(frame, hash2 + continued, newblock);
2077 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2078 if (err)
2079 goto journal_error;
2080 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2081 if (err)
2082 goto journal_error;
2083 brelse(bh2);
2084 dxtrace(dx_show_index("frame", frame->entries));
2085 return de;
2086
2087journal_error:
2088 brelse(*bh);
2089 brelse(bh2);
2090 *bh = NULL;
2091 ext4_std_error(dir->i_sb, err);
2092 return ERR_PTR(err);
2093}
2094
2095int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2096 struct buffer_head *bh,
2097 void *buf, int buf_size,
2098 struct ext4_filename *fname,
2099 struct ext4_dir_entry_2 **dest_de)
2100{
2101 struct ext4_dir_entry_2 *de;
2102 unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2103 int nlen, rlen;
2104 unsigned int offset = 0;
2105 char *top;
2106
2107 de = buf;
2108 top = buf + buf_size - reclen;
2109 while ((char *) de <= top) {
2110 if (ext4_check_dir_entry(dir, NULL, de, bh,
2111 buf, buf_size, offset))
2112 return -EFSCORRUPTED;
2113 if (ext4_match(dir, fname, de))
2114 return -EEXIST;
2115 nlen = ext4_dir_rec_len(de->name_len, dir);
2116 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2117 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2118 break;
2119 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2120 offset += rlen;
2121 }
2122 if ((char *) de > top)
2123 return -ENOSPC;
2124
2125 *dest_de = de;
2126 return 0;
2127}
2128
2129void ext4_insert_dentry(struct inode *dir,
2130 struct inode *inode,
2131 struct ext4_dir_entry_2 *de,
2132 int buf_size,
2133 struct ext4_filename *fname)
2134{
2135
2136 int nlen, rlen;
2137
2138 nlen = ext4_dir_rec_len(de->name_len, dir);
2139 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2140 if (de->inode) {
2141 struct ext4_dir_entry_2 *de1 =
2142 (struct ext4_dir_entry_2 *)((char *)de + nlen);
2143 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2144 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2145 de = de1;
2146 }
2147 de->file_type = EXT4_FT_UNKNOWN;
2148 de->inode = cpu_to_le32(inode->i_ino);
2149 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2150 de->name_len = fname_len(fname);
2151 memcpy(de->name, fname_name(fname), fname_len(fname));
2152 if (ext4_hash_in_dirent(dir)) {
2153 struct dx_hash_info *hinfo = &fname->hinfo;
2154
2155 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2156 EXT4_DIRENT_HASHES(de)->minor_hash =
2157 cpu_to_le32(hinfo->minor_hash);
2158 }
2159}
2160
2161/*
2162 * Add a new entry into a directory (leaf) block. If de is non-NULL,
2163 * it points to a directory entry which is guaranteed to be large
2164 * enough for new directory entry. If de is NULL, then
2165 * add_dirent_to_buf will attempt search the directory block for
2166 * space. It will return -ENOSPC if no space is available, and -EIO
2167 * and -EEXIST if directory entry already exists.
2168 */
2169static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2170 struct inode *dir,
2171 struct inode *inode, struct ext4_dir_entry_2 *de,
2172 struct buffer_head *bh)
2173{
2174 unsigned int blocksize = dir->i_sb->s_blocksize;
2175 int csum_size = 0;
2176 int err, err2;
2177
2178 if (ext4_has_metadata_csum(inode->i_sb))
2179 csum_size = sizeof(struct ext4_dir_entry_tail);
2180
2181 if (!de) {
2182 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2183 blocksize - csum_size, fname, &de);
2184 if (err)
2185 return err;
2186 }
2187 BUFFER_TRACE(bh, "get_write_access");
2188 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2189 EXT4_JTR_NONE);
2190 if (err) {
2191 ext4_std_error(dir->i_sb, err);
2192 return err;
2193 }
2194
2195 /* By now the buffer is marked for journaling */
2196 ext4_insert_dentry(dir, inode, de, blocksize, fname);
2197
2198 /*
2199 * XXX shouldn't update any times until successful
2200 * completion of syscall, but too many callers depend
2201 * on this.
2202 *
2203 * XXX similarly, too many callers depend on
2204 * ext4_new_inode() setting the times, but error
2205 * recovery deletes the inode, so the worst that can
2206 * happen is that the times are slightly out of date
2207 * and/or different from the directory change time.
2208 */
2209 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
2210 ext4_update_dx_flag(dir);
2211 inode_inc_iversion(dir);
2212 err2 = ext4_mark_inode_dirty(handle, dir);
2213 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2214 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2215 if (err)
2216 ext4_std_error(dir->i_sb, err);
2217 return err ? err : err2;
2218}
2219
2220/*
2221 * This converts a one block unindexed directory to a 3 block indexed
2222 * directory, and adds the dentry to the indexed directory.
2223 */
2224static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2225 struct inode *dir,
2226 struct inode *inode, struct buffer_head *bh)
2227{
2228 struct buffer_head *bh2;
2229 struct dx_root *root;
2230 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2231 struct dx_entry *entries;
2232 struct ext4_dir_entry_2 *de, *de2;
2233 char *data2, *top;
2234 unsigned len;
2235 int retval;
2236 unsigned blocksize;
2237 ext4_lblk_t block;
2238 struct fake_dirent *fde;
2239 int csum_size = 0;
2240
2241 if (ext4_has_metadata_csum(inode->i_sb))
2242 csum_size = sizeof(struct ext4_dir_entry_tail);
2243
2244 blocksize = dir->i_sb->s_blocksize;
2245 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2246 BUFFER_TRACE(bh, "get_write_access");
2247 retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2248 EXT4_JTR_NONE);
2249 if (retval) {
2250 ext4_std_error(dir->i_sb, retval);
2251 brelse(bh);
2252 return retval;
2253 }
2254 root = (struct dx_root *) bh->b_data;
2255
2256 /* The 0th block becomes the root, move the dirents out */
2257 fde = &root->dotdot;
2258 de = (struct ext4_dir_entry_2 *)((char *)fde +
2259 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2260 if ((char *) de >= (((char *) root) + blocksize)) {
2261 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2262 brelse(bh);
2263 return -EFSCORRUPTED;
2264 }
2265 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2266
2267 /* Allocate new block for the 0th block's dirents */
2268 bh2 = ext4_append(handle, dir, &block);
2269 if (IS_ERR(bh2)) {
2270 brelse(bh);
2271 return PTR_ERR(bh2);
2272 }
2273 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2274 data2 = bh2->b_data;
2275
2276 memcpy(data2, de, len);
2277 memset(de, 0, len); /* wipe old data */
2278 de = (struct ext4_dir_entry_2 *) data2;
2279 top = data2 + len;
2280 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2281 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2282 (char *)de - data2)) {
2283 brelse(bh2);
2284 brelse(bh);
2285 return -EFSCORRUPTED;
2286 }
2287 de = de2;
2288 }
2289 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2290 (char *) de, blocksize);
2291
2292 if (csum_size)
2293 ext4_initialize_dirent_tail(bh2, blocksize);
2294
2295 /* Initialize the root; the dot dirents already exist */
2296 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2297 de->rec_len = ext4_rec_len_to_disk(
2298 blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2299 memset (&root->info, 0, sizeof(root->info));
2300 root->info.info_length = sizeof(root->info);
2301 if (ext4_hash_in_dirent(dir))
2302 root->info.hash_version = DX_HASH_SIPHASH;
2303 else
2304 root->info.hash_version =
2305 EXT4_SB(dir->i_sb)->s_def_hash_version;
2306
2307 entries = root->entries;
2308 dx_set_block(entries, 1);
2309 dx_set_count(entries, 1);
2310 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2311
2312 /* Initialize as for dx_probe */
2313 fname->hinfo.hash_version = root->info.hash_version;
2314 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2315 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2316 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2317
2318 /* casefolded encrypted hashes are computed on fname setup */
2319 if (!ext4_hash_in_dirent(dir)) {
2320 int err = ext4fs_dirhash(dir, fname_name(fname),
2321 fname_len(fname), &fname->hinfo);
2322 if (err < 0) {
2323 brelse(bh2);
2324 brelse(bh);
2325 return err;
2326 }
2327 }
2328 memset(frames, 0, sizeof(frames));
2329 frame = frames;
2330 frame->entries = entries;
2331 frame->at = entries;
2332 frame->bh = bh;
2333
2334 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2335 if (retval)
2336 goto out_frames;
2337 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2338 if (retval)
2339 goto out_frames;
2340
2341 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2342 if (IS_ERR(de)) {
2343 retval = PTR_ERR(de);
2344 goto out_frames;
2345 }
2346
2347 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2348out_frames:
2349 /*
2350 * Even if the block split failed, we have to properly write
2351 * out all the changes we did so far. Otherwise we can end up
2352 * with corrupted filesystem.
2353 */
2354 if (retval)
2355 ext4_mark_inode_dirty(handle, dir);
2356 dx_release(frames);
2357 brelse(bh2);
2358 return retval;
2359}
2360
2361/*
2362 * ext4_add_entry()
2363 *
2364 * adds a file entry to the specified directory, using the same
2365 * semantics as ext4_find_entry(). It returns NULL if it failed.
2366 *
2367 * NOTE!! The inode part of 'de' is left at 0 - which means you
2368 * may not sleep between calling this and putting something into
2369 * the entry, as someone else might have used it while you slept.
2370 */
2371static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2372 struct inode *inode)
2373{
2374 struct inode *dir = d_inode(dentry->d_parent);
2375 struct buffer_head *bh = NULL;
2376 struct ext4_dir_entry_2 *de;
2377 struct super_block *sb;
2378 struct ext4_filename fname;
2379 int retval;
2380 int dx_fallback=0;
2381 unsigned blocksize;
2382 ext4_lblk_t block, blocks;
2383 int csum_size = 0;
2384
2385 if (ext4_has_metadata_csum(inode->i_sb))
2386 csum_size = sizeof(struct ext4_dir_entry_tail);
2387
2388 sb = dir->i_sb;
2389 blocksize = sb->s_blocksize;
2390
2391 if (fscrypt_is_nokey_name(dentry))
2392 return -ENOKEY;
2393
2394#if IS_ENABLED(CONFIG_UNICODE)
2395 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2396 utf8_validate(sb->s_encoding, &dentry->d_name))
2397 return -EINVAL;
2398#endif
2399
2400 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2401 if (retval)
2402 return retval;
2403
2404 if (ext4_has_inline_data(dir)) {
2405 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2406 if (retval < 0)
2407 goto out;
2408 if (retval == 1) {
2409 retval = 0;
2410 goto out;
2411 }
2412 }
2413
2414 if (is_dx(dir)) {
2415 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2416 if (!retval || (retval != ERR_BAD_DX_DIR))
2417 goto out;
2418 /* Can we just ignore htree data? */
2419 if (ext4_has_metadata_csum(sb)) {
2420 EXT4_ERROR_INODE(dir,
2421 "Directory has corrupted htree index.");
2422 retval = -EFSCORRUPTED;
2423 goto out;
2424 }
2425 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2426 dx_fallback++;
2427 retval = ext4_mark_inode_dirty(handle, dir);
2428 if (unlikely(retval))
2429 goto out;
2430 }
2431 blocks = dir->i_size >> sb->s_blocksize_bits;
2432 for (block = 0; block < blocks; block++) {
2433 bh = ext4_read_dirblock(dir, block, DIRENT);
2434 if (bh == NULL) {
2435 bh = ext4_bread(handle, dir, block,
2436 EXT4_GET_BLOCKS_CREATE);
2437 goto add_to_new_block;
2438 }
2439 if (IS_ERR(bh)) {
2440 retval = PTR_ERR(bh);
2441 bh = NULL;
2442 goto out;
2443 }
2444 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2445 NULL, bh);
2446 if (retval != -ENOSPC)
2447 goto out;
2448
2449 if (blocks == 1 && !dx_fallback &&
2450 ext4_has_feature_dir_index(sb)) {
2451 retval = make_indexed_dir(handle, &fname, dir,
2452 inode, bh);
2453 bh = NULL; /* make_indexed_dir releases bh */
2454 goto out;
2455 }
2456 brelse(bh);
2457 }
2458 bh = ext4_append(handle, dir, &block);
2459add_to_new_block:
2460 if (IS_ERR(bh)) {
2461 retval = PTR_ERR(bh);
2462 bh = NULL;
2463 goto out;
2464 }
2465 de = (struct ext4_dir_entry_2 *) bh->b_data;
2466 de->inode = 0;
2467 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2468
2469 if (csum_size)
2470 ext4_initialize_dirent_tail(bh, blocksize);
2471
2472 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2473out:
2474 ext4_fname_free_filename(&fname);
2475 brelse(bh);
2476 if (retval == 0)
2477 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2478 return retval;
2479}
2480
2481/*
2482 * Returns 0 for success, or a negative error value
2483 */
2484static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2485 struct inode *dir, struct inode *inode)
2486{
2487 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2488 struct dx_entry *entries, *at;
2489 struct buffer_head *bh;
2490 struct super_block *sb = dir->i_sb;
2491 struct ext4_dir_entry_2 *de;
2492 int restart;
2493 int err;
2494
2495again:
2496 restart = 0;
2497 frame = dx_probe(fname, dir, NULL, frames);
2498 if (IS_ERR(frame))
2499 return PTR_ERR(frame);
2500 entries = frame->entries;
2501 at = frame->at;
2502 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2503 if (IS_ERR(bh)) {
2504 err = PTR_ERR(bh);
2505 bh = NULL;
2506 goto cleanup;
2507 }
2508
2509 BUFFER_TRACE(bh, "get_write_access");
2510 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2511 if (err)
2512 goto journal_error;
2513
2514 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2515 if (err != -ENOSPC)
2516 goto cleanup;
2517
2518 err = 0;
2519 /* Block full, should compress but for now just split */
2520 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2521 dx_get_count(entries), dx_get_limit(entries)));
2522 /* Need to split index? */
2523 if (dx_get_count(entries) == dx_get_limit(entries)) {
2524 ext4_lblk_t newblock;
2525 int levels = frame - frames + 1;
2526 unsigned int icount;
2527 int add_level = 1;
2528 struct dx_entry *entries2;
2529 struct dx_node *node2;
2530 struct buffer_head *bh2;
2531
2532 while (frame > frames) {
2533 if (dx_get_count((frame - 1)->entries) <
2534 dx_get_limit((frame - 1)->entries)) {
2535 add_level = 0;
2536 break;
2537 }
2538 frame--; /* split higher index block */
2539 at = frame->at;
2540 entries = frame->entries;
2541 restart = 1;
2542 }
2543 if (add_level && levels == ext4_dir_htree_level(sb)) {
2544 ext4_warning(sb, "Directory (ino: %lu) index full, "
2545 "reach max htree level :%d",
2546 dir->i_ino, levels);
2547 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2548 ext4_warning(sb, "Large directory feature is "
2549 "not enabled on this "
2550 "filesystem");
2551 }
2552 err = -ENOSPC;
2553 goto cleanup;
2554 }
2555 icount = dx_get_count(entries);
2556 bh2 = ext4_append(handle, dir, &newblock);
2557 if (IS_ERR(bh2)) {
2558 err = PTR_ERR(bh2);
2559 goto cleanup;
2560 }
2561 node2 = (struct dx_node *)(bh2->b_data);
2562 entries2 = node2->entries;
2563 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2564 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2565 sb->s_blocksize);
2566 BUFFER_TRACE(frame->bh, "get_write_access");
2567 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2568 EXT4_JTR_NONE);
2569 if (err)
2570 goto journal_error;
2571 if (!add_level) {
2572 unsigned icount1 = icount/2, icount2 = icount - icount1;
2573 unsigned hash2 = dx_get_hash(entries + icount1);
2574 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2575 icount1, icount2));
2576
2577 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2578 err = ext4_journal_get_write_access(handle, sb,
2579 (frame - 1)->bh,
2580 EXT4_JTR_NONE);
2581 if (err)
2582 goto journal_error;
2583
2584 memcpy((char *) entries2, (char *) (entries + icount1),
2585 icount2 * sizeof(struct dx_entry));
2586 dx_set_count(entries, icount1);
2587 dx_set_count(entries2, icount2);
2588 dx_set_limit(entries2, dx_node_limit(dir));
2589
2590 /* Which index block gets the new entry? */
2591 if (at - entries >= icount1) {
2592 frame->at = at - entries - icount1 + entries2;
2593 frame->entries = entries = entries2;
2594 swap(frame->bh, bh2);
2595 }
2596 dx_insert_block((frame - 1), hash2, newblock);
2597 dxtrace(dx_show_index("node", frame->entries));
2598 dxtrace(dx_show_index("node",
2599 ((struct dx_node *) bh2->b_data)->entries));
2600 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2601 if (err)
2602 goto journal_error;
2603 brelse (bh2);
2604 err = ext4_handle_dirty_dx_node(handle, dir,
2605 (frame - 1)->bh);
2606 if (err)
2607 goto journal_error;
2608 err = ext4_handle_dirty_dx_node(handle, dir,
2609 frame->bh);
2610 if (restart || err)
2611 goto journal_error;
2612 } else {
2613 struct dx_root *dxroot;
2614 memcpy((char *) entries2, (char *) entries,
2615 icount * sizeof(struct dx_entry));
2616 dx_set_limit(entries2, dx_node_limit(dir));
2617
2618 /* Set up root */
2619 dx_set_count(entries, 1);
2620 dx_set_block(entries + 0, newblock);
2621 dxroot = (struct dx_root *)frames[0].bh->b_data;
2622 dxroot->info.indirect_levels += 1;
2623 dxtrace(printk(KERN_DEBUG
2624 "Creating %d level index...\n",
2625 dxroot->info.indirect_levels));
2626 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2627 if (err)
2628 goto journal_error;
2629 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2630 brelse(bh2);
2631 restart = 1;
2632 goto journal_error;
2633 }
2634 }
2635 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2636 if (IS_ERR(de)) {
2637 err = PTR_ERR(de);
2638 goto cleanup;
2639 }
2640 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2641 goto cleanup;
2642
2643journal_error:
2644 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2645cleanup:
2646 brelse(bh);
2647 dx_release(frames);
2648 /* @restart is true means htree-path has been changed, we need to
2649 * repeat dx_probe() to find out valid htree-path
2650 */
2651 if (restart && err == 0)
2652 goto again;
2653 return err;
2654}
2655
2656/*
2657 * ext4_generic_delete_entry deletes a directory entry by merging it
2658 * with the previous entry
2659 */
2660int ext4_generic_delete_entry(struct inode *dir,
2661 struct ext4_dir_entry_2 *de_del,
2662 struct buffer_head *bh,
2663 void *entry_buf,
2664 int buf_size,
2665 int csum_size)
2666{
2667 struct ext4_dir_entry_2 *de, *pde;
2668 unsigned int blocksize = dir->i_sb->s_blocksize;
2669 int i;
2670
2671 i = 0;
2672 pde = NULL;
2673 de = entry_buf;
2674 while (i < buf_size - csum_size) {
2675 if (ext4_check_dir_entry(dir, NULL, de, bh,
2676 entry_buf, buf_size, i))
2677 return -EFSCORRUPTED;
2678 if (de == de_del) {
2679 if (pde) {
2680 pde->rec_len = ext4_rec_len_to_disk(
2681 ext4_rec_len_from_disk(pde->rec_len,
2682 blocksize) +
2683 ext4_rec_len_from_disk(de->rec_len,
2684 blocksize),
2685 blocksize);
2686
2687 /* wipe entire dir_entry */
2688 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2689 blocksize));
2690 } else {
2691 /* wipe dir_entry excluding the rec_len field */
2692 de->inode = 0;
2693 memset(&de->name_len, 0,
2694 ext4_rec_len_from_disk(de->rec_len,
2695 blocksize) -
2696 offsetof(struct ext4_dir_entry_2,
2697 name_len));
2698 }
2699
2700 inode_inc_iversion(dir);
2701 return 0;
2702 }
2703 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2704 pde = de;
2705 de = ext4_next_entry(de, blocksize);
2706 }
2707 return -ENOENT;
2708}
2709
2710static int ext4_delete_entry(handle_t *handle,
2711 struct inode *dir,
2712 struct ext4_dir_entry_2 *de_del,
2713 struct buffer_head *bh)
2714{
2715 int err, csum_size = 0;
2716
2717 if (ext4_has_inline_data(dir)) {
2718 int has_inline_data = 1;
2719 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2720 &has_inline_data);
2721 if (has_inline_data)
2722 return err;
2723 }
2724
2725 if (ext4_has_metadata_csum(dir->i_sb))
2726 csum_size = sizeof(struct ext4_dir_entry_tail);
2727
2728 BUFFER_TRACE(bh, "get_write_access");
2729 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2730 EXT4_JTR_NONE);
2731 if (unlikely(err))
2732 goto out;
2733
2734 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2735 dir->i_sb->s_blocksize, csum_size);
2736 if (err)
2737 goto out;
2738
2739 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2740 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2741 if (unlikely(err))
2742 goto out;
2743
2744 return 0;
2745out:
2746 if (err != -ENOENT)
2747 ext4_std_error(dir->i_sb, err);
2748 return err;
2749}
2750
2751/*
2752 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2753 * since this indicates that nlinks count was previously 1 to avoid overflowing
2754 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2755 * that subdirectory link counts are not being maintained accurately.
2756 *
2757 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2758 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2759 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2760 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2761 */
2762static void ext4_inc_count(struct inode *inode)
2763{
2764 inc_nlink(inode);
2765 if (is_dx(inode) &&
2766 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2767 set_nlink(inode, 1);
2768}
2769
2770/*
2771 * If a directory had nlink == 1, then we should let it be 1. This indicates
2772 * directory has >EXT4_LINK_MAX subdirs.
2773 */
2774static void ext4_dec_count(struct inode *inode)
2775{
2776 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2777 drop_nlink(inode);
2778}
2779
2780
2781/*
2782 * Add non-directory inode to a directory. On success, the inode reference is
2783 * consumed by dentry is instantiation. This is also indicated by clearing of
2784 * *inodep pointer. On failure, the caller is responsible for dropping the
2785 * inode reference in the safe context.
2786 */
2787static int ext4_add_nondir(handle_t *handle,
2788 struct dentry *dentry, struct inode **inodep)
2789{
2790 struct inode *dir = d_inode(dentry->d_parent);
2791 struct inode *inode = *inodep;
2792 int err = ext4_add_entry(handle, dentry, inode);
2793 if (!err) {
2794 err = ext4_mark_inode_dirty(handle, inode);
2795 if (IS_DIRSYNC(dir))
2796 ext4_handle_sync(handle);
2797 d_instantiate_new(dentry, inode);
2798 *inodep = NULL;
2799 return err;
2800 }
2801 drop_nlink(inode);
2802 ext4_mark_inode_dirty(handle, inode);
2803 ext4_orphan_add(handle, inode);
2804 unlock_new_inode(inode);
2805 return err;
2806}
2807
2808/*
2809 * By the time this is called, we already have created
2810 * the directory cache entry for the new file, but it
2811 * is so far negative - it has no inode.
2812 *
2813 * If the create succeeds, we fill in the inode information
2814 * with d_instantiate().
2815 */
2816static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2817 struct dentry *dentry, umode_t mode, bool excl)
2818{
2819 handle_t *handle;
2820 struct inode *inode;
2821 int err, credits, retries = 0;
2822
2823 err = dquot_initialize(dir);
2824 if (err)
2825 return err;
2826
2827 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2828 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2829retry:
2830 inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2831 0, NULL, EXT4_HT_DIR, credits);
2832 handle = ext4_journal_current_handle();
2833 err = PTR_ERR(inode);
2834 if (!IS_ERR(inode)) {
2835 inode->i_op = &ext4_file_inode_operations;
2836 inode->i_fop = &ext4_file_operations;
2837 ext4_set_aops(inode);
2838 err = ext4_add_nondir(handle, dentry, &inode);
2839 if (!err)
2840 ext4_fc_track_create(handle, dentry);
2841 }
2842 if (handle)
2843 ext4_journal_stop(handle);
2844 if (!IS_ERR_OR_NULL(inode))
2845 iput(inode);
2846 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2847 goto retry;
2848 return err;
2849}
2850
2851static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2852 struct dentry *dentry, umode_t mode, dev_t rdev)
2853{
2854 handle_t *handle;
2855 struct inode *inode;
2856 int err, credits, retries = 0;
2857
2858 err = dquot_initialize(dir);
2859 if (err)
2860 return err;
2861
2862 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2863 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2864retry:
2865 inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2866 0, NULL, EXT4_HT_DIR, credits);
2867 handle = ext4_journal_current_handle();
2868 err = PTR_ERR(inode);
2869 if (!IS_ERR(inode)) {
2870 init_special_inode(inode, inode->i_mode, rdev);
2871 inode->i_op = &ext4_special_inode_operations;
2872 err = ext4_add_nondir(handle, dentry, &inode);
2873 if (!err)
2874 ext4_fc_track_create(handle, dentry);
2875 }
2876 if (handle)
2877 ext4_journal_stop(handle);
2878 if (!IS_ERR_OR_NULL(inode))
2879 iput(inode);
2880 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2881 goto retry;
2882 return err;
2883}
2884
2885static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2886 struct file *file, umode_t mode)
2887{
2888 handle_t *handle;
2889 struct inode *inode;
2890 int err, retries = 0;
2891
2892 err = dquot_initialize(dir);
2893 if (err)
2894 return err;
2895
2896retry:
2897 inode = ext4_new_inode_start_handle(idmap, dir, mode,
2898 NULL, 0, NULL,
2899 EXT4_HT_DIR,
2900 EXT4_MAXQUOTAS_TRANS_BLOCKS(dir->i_sb) +
2901 4 + EXT4_XATTR_TRANS_BLOCKS);
2902 handle = ext4_journal_current_handle();
2903 err = PTR_ERR(inode);
2904 if (!IS_ERR(inode)) {
2905 inode->i_op = &ext4_file_inode_operations;
2906 inode->i_fop = &ext4_file_operations;
2907 ext4_set_aops(inode);
2908 d_tmpfile(file, inode);
2909 err = ext4_orphan_add(handle, inode);
2910 if (err)
2911 goto err_unlock_inode;
2912 mark_inode_dirty(inode);
2913 unlock_new_inode(inode);
2914 }
2915 if (handle)
2916 ext4_journal_stop(handle);
2917 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2918 goto retry;
2919 return finish_open_simple(file, err);
2920err_unlock_inode:
2921 ext4_journal_stop(handle);
2922 unlock_new_inode(inode);
2923 return err;
2924}
2925
2926struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2927 struct ext4_dir_entry_2 *de,
2928 int blocksize, int csum_size,
2929 unsigned int parent_ino, int dotdot_real_len)
2930{
2931 de->inode = cpu_to_le32(inode->i_ino);
2932 de->name_len = 1;
2933 de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2934 blocksize);
2935 strcpy(de->name, ".");
2936 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2937
2938 de = ext4_next_entry(de, blocksize);
2939 de->inode = cpu_to_le32(parent_ino);
2940 de->name_len = 2;
2941 if (!dotdot_real_len)
2942 de->rec_len = ext4_rec_len_to_disk(blocksize -
2943 (csum_size + ext4_dir_rec_len(1, NULL)),
2944 blocksize);
2945 else
2946 de->rec_len = ext4_rec_len_to_disk(
2947 ext4_dir_rec_len(de->name_len, NULL),
2948 blocksize);
2949 strcpy(de->name, "..");
2950 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2951
2952 return ext4_next_entry(de, blocksize);
2953}
2954
2955int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2956 struct inode *inode)
2957{
2958 struct buffer_head *dir_block = NULL;
2959 struct ext4_dir_entry_2 *de;
2960 ext4_lblk_t block = 0;
2961 unsigned int blocksize = dir->i_sb->s_blocksize;
2962 int csum_size = 0;
2963 int err;
2964
2965 if (ext4_has_metadata_csum(dir->i_sb))
2966 csum_size = sizeof(struct ext4_dir_entry_tail);
2967
2968 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2969 err = ext4_try_create_inline_dir(handle, dir, inode);
2970 if (err < 0 && err != -ENOSPC)
2971 goto out;
2972 if (!err)
2973 goto out;
2974 }
2975
2976 inode->i_size = 0;
2977 dir_block = ext4_append(handle, inode, &block);
2978 if (IS_ERR(dir_block))
2979 return PTR_ERR(dir_block);
2980 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2981 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2982 set_nlink(inode, 2);
2983 if (csum_size)
2984 ext4_initialize_dirent_tail(dir_block, blocksize);
2985
2986 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2987 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2988 if (err)
2989 goto out;
2990 set_buffer_verified(dir_block);
2991out:
2992 brelse(dir_block);
2993 return err;
2994}
2995
2996static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2997 struct dentry *dentry, umode_t mode)
2998{
2999 handle_t *handle;
3000 struct inode *inode;
3001 int err, err2 = 0, credits, retries = 0;
3002
3003 if (EXT4_DIR_LINK_MAX(dir))
3004 return -EMLINK;
3005
3006 err = dquot_initialize(dir);
3007 if (err)
3008 return err;
3009
3010 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3011 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3012retry:
3013 inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3014 &dentry->d_name,
3015 0, NULL, EXT4_HT_DIR, credits);
3016 handle = ext4_journal_current_handle();
3017 err = PTR_ERR(inode);
3018 if (IS_ERR(inode))
3019 goto out_stop;
3020
3021 inode->i_op = &ext4_dir_inode_operations;
3022 inode->i_fop = &ext4_dir_operations;
3023 err = ext4_init_new_dir(handle, dir, inode);
3024 if (err)
3025 goto out_clear_inode;
3026 err = ext4_mark_inode_dirty(handle, inode);
3027 if (!err)
3028 err = ext4_add_entry(handle, dentry, inode);
3029 if (err) {
3030out_clear_inode:
3031 clear_nlink(inode);
3032 ext4_orphan_add(handle, inode);
3033 unlock_new_inode(inode);
3034 err2 = ext4_mark_inode_dirty(handle, inode);
3035 if (unlikely(err2))
3036 err = err2;
3037 ext4_journal_stop(handle);
3038 iput(inode);
3039 goto out_retry;
3040 }
3041 ext4_inc_count(dir);
3042
3043 ext4_update_dx_flag(dir);
3044 err = ext4_mark_inode_dirty(handle, dir);
3045 if (err)
3046 goto out_clear_inode;
3047 d_instantiate_new(dentry, inode);
3048 ext4_fc_track_create(handle, dentry);
3049 if (IS_DIRSYNC(dir))
3050 ext4_handle_sync(handle);
3051
3052out_stop:
3053 if (handle)
3054 ext4_journal_stop(handle);
3055out_retry:
3056 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3057 goto retry;
3058 return err;
3059}
3060
3061/*
3062 * routine to check that the specified directory is empty (for rmdir)
3063 */
3064bool ext4_empty_dir(struct inode *inode)
3065{
3066 unsigned int offset;
3067 struct buffer_head *bh;
3068 struct ext4_dir_entry_2 *de;
3069 struct super_block *sb;
3070
3071 if (ext4_has_inline_data(inode)) {
3072 int has_inline_data = 1;
3073 int ret;
3074
3075 ret = empty_inline_dir(inode, &has_inline_data);
3076 if (has_inline_data)
3077 return ret;
3078 }
3079
3080 sb = inode->i_sb;
3081 if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3082 ext4_dir_rec_len(2, NULL)) {
3083 EXT4_ERROR_INODE(inode, "invalid size");
3084 return false;
3085 }
3086 /* The first directory block must not be a hole,
3087 * so treat it as DIRENT_HTREE
3088 */
3089 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3090 if (IS_ERR(bh))
3091 return false;
3092
3093 de = (struct ext4_dir_entry_2 *) bh->b_data;
3094 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3095 0) ||
3096 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3097 ext4_warning_inode(inode, "directory missing '.'");
3098 brelse(bh);
3099 return false;
3100 }
3101 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3102 de = ext4_next_entry(de, sb->s_blocksize);
3103 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3104 offset) ||
3105 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3106 ext4_warning_inode(inode, "directory missing '..'");
3107 brelse(bh);
3108 return false;
3109 }
3110 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3111 while (offset < inode->i_size) {
3112 if (!(offset & (sb->s_blocksize - 1))) {
3113 unsigned int lblock;
3114 brelse(bh);
3115 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3116 bh = ext4_read_dirblock(inode, lblock, EITHER);
3117 if (bh == NULL) {
3118 offset += sb->s_blocksize;
3119 continue;
3120 }
3121 if (IS_ERR(bh))
3122 return false;
3123 }
3124 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3125 (offset & (sb->s_blocksize - 1)));
3126 if (ext4_check_dir_entry(inode, NULL, de, bh,
3127 bh->b_data, bh->b_size, offset) ||
3128 le32_to_cpu(de->inode)) {
3129 brelse(bh);
3130 return false;
3131 }
3132 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3133 }
3134 brelse(bh);
3135 return true;
3136}
3137
3138static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3139{
3140 int retval;
3141 struct inode *inode;
3142 struct buffer_head *bh;
3143 struct ext4_dir_entry_2 *de;
3144 handle_t *handle = NULL;
3145
3146 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3147 return -EIO;
3148
3149 /* Initialize quotas before so that eventual writes go in
3150 * separate transaction */
3151 retval = dquot_initialize(dir);
3152 if (retval)
3153 return retval;
3154 retval = dquot_initialize(d_inode(dentry));
3155 if (retval)
3156 return retval;
3157
3158 retval = -ENOENT;
3159 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3160 if (IS_ERR(bh))
3161 return PTR_ERR(bh);
3162 if (!bh)
3163 goto end_rmdir;
3164
3165 inode = d_inode(dentry);
3166
3167 retval = -EFSCORRUPTED;
3168 if (le32_to_cpu(de->inode) != inode->i_ino)
3169 goto end_rmdir;
3170
3171 retval = -ENOTEMPTY;
3172 if (!ext4_empty_dir(inode))
3173 goto end_rmdir;
3174
3175 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3176 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3177 if (IS_ERR(handle)) {
3178 retval = PTR_ERR(handle);
3179 handle = NULL;
3180 goto end_rmdir;
3181 }
3182
3183 if (IS_DIRSYNC(dir))
3184 ext4_handle_sync(handle);
3185
3186 retval = ext4_delete_entry(handle, dir, de, bh);
3187 if (retval)
3188 goto end_rmdir;
3189 if (!EXT4_DIR_LINK_EMPTY(inode))
3190 ext4_warning_inode(inode,
3191 "empty directory '%.*s' has too many links (%u)",
3192 dentry->d_name.len, dentry->d_name.name,
3193 inode->i_nlink);
3194 inode_inc_iversion(inode);
3195 clear_nlink(inode);
3196 /* There's no need to set i_disksize: the fact that i_nlink is
3197 * zero will ensure that the right thing happens during any
3198 * recovery. */
3199 inode->i_size = 0;
3200 ext4_orphan_add(handle, inode);
3201 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3202 inode_set_ctime_current(inode);
3203 retval = ext4_mark_inode_dirty(handle, inode);
3204 if (retval)
3205 goto end_rmdir;
3206 ext4_dec_count(dir);
3207 ext4_update_dx_flag(dir);
3208 ext4_fc_track_unlink(handle, dentry);
3209 retval = ext4_mark_inode_dirty(handle, dir);
3210
3211#if IS_ENABLED(CONFIG_UNICODE)
3212 /* VFS negative dentries are incompatible with Encoding and
3213 * Case-insensitiveness. Eventually we'll want avoid
3214 * invalidating the dentries here, alongside with returning the
3215 * negative dentries at ext4_lookup(), when it is better
3216 * supported by the VFS for the CI case.
3217 */
3218 if (IS_CASEFOLDED(dir))
3219 d_invalidate(dentry);
3220#endif
3221
3222end_rmdir:
3223 brelse(bh);
3224 if (handle)
3225 ext4_journal_stop(handle);
3226 return retval;
3227}
3228
3229int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3230 struct inode *inode,
3231 struct dentry *dentry /* NULL during fast_commit recovery */)
3232{
3233 int retval = -ENOENT;
3234 struct buffer_head *bh;
3235 struct ext4_dir_entry_2 *de;
3236 handle_t *handle;
3237 int skip_remove_dentry = 0;
3238
3239 /*
3240 * Keep this outside the transaction; it may have to set up the
3241 * directory's encryption key, which isn't GFP_NOFS-safe.
3242 */
3243 bh = ext4_find_entry(dir, d_name, &de, NULL);
3244 if (IS_ERR(bh))
3245 return PTR_ERR(bh);
3246
3247 if (!bh)
3248 return -ENOENT;
3249
3250 if (le32_to_cpu(de->inode) != inode->i_ino) {
3251 /*
3252 * It's okay if we find dont find dentry which matches
3253 * the inode. That's because it might have gotten
3254 * renamed to a different inode number
3255 */
3256 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3257 skip_remove_dentry = 1;
3258 else
3259 goto out_bh;
3260 }
3261
3262 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3263 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3264 if (IS_ERR(handle)) {
3265 retval = PTR_ERR(handle);
3266 goto out_bh;
3267 }
3268
3269 if (IS_DIRSYNC(dir))
3270 ext4_handle_sync(handle);
3271
3272 if (!skip_remove_dentry) {
3273 retval = ext4_delete_entry(handle, dir, de, bh);
3274 if (retval)
3275 goto out_handle;
3276 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3277 ext4_update_dx_flag(dir);
3278 retval = ext4_mark_inode_dirty(handle, dir);
3279 if (retval)
3280 goto out_handle;
3281 } else {
3282 retval = 0;
3283 }
3284 if (inode->i_nlink == 0)
3285 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3286 d_name->len, d_name->name);
3287 else
3288 drop_nlink(inode);
3289 if (!inode->i_nlink)
3290 ext4_orphan_add(handle, inode);
3291 inode_set_ctime_current(inode);
3292 retval = ext4_mark_inode_dirty(handle, inode);
3293 if (dentry && !retval)
3294 ext4_fc_track_unlink(handle, dentry);
3295out_handle:
3296 ext4_journal_stop(handle);
3297out_bh:
3298 brelse(bh);
3299 return retval;
3300}
3301
3302static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3303{
3304 int retval;
3305
3306 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3307 return -EIO;
3308
3309 trace_ext4_unlink_enter(dir, dentry);
3310 /*
3311 * Initialize quotas before so that eventual writes go
3312 * in separate transaction
3313 */
3314 retval = dquot_initialize(dir);
3315 if (retval)
3316 goto out_trace;
3317 retval = dquot_initialize(d_inode(dentry));
3318 if (retval)
3319 goto out_trace;
3320
3321 retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3322#if IS_ENABLED(CONFIG_UNICODE)
3323 /* VFS negative dentries are incompatible with Encoding and
3324 * Case-insensitiveness. Eventually we'll want avoid
3325 * invalidating the dentries here, alongside with returning the
3326 * negative dentries at ext4_lookup(), when it is better
3327 * supported by the VFS for the CI case.
3328 */
3329 if (IS_CASEFOLDED(dir))
3330 d_invalidate(dentry);
3331#endif
3332
3333out_trace:
3334 trace_ext4_unlink_exit(dentry, retval);
3335 return retval;
3336}
3337
3338static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3339 struct fscrypt_str *disk_link)
3340{
3341 struct buffer_head *bh;
3342 char *kaddr;
3343 int err = 0;
3344
3345 bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3346 if (IS_ERR(bh))
3347 return PTR_ERR(bh);
3348
3349 BUFFER_TRACE(bh, "get_write_access");
3350 err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3351 if (err)
3352 goto out;
3353
3354 kaddr = (char *)bh->b_data;
3355 memcpy(kaddr, disk_link->name, disk_link->len);
3356 inode->i_size = disk_link->len - 1;
3357 EXT4_I(inode)->i_disksize = inode->i_size;
3358 err = ext4_handle_dirty_metadata(handle, inode, bh);
3359out:
3360 brelse(bh);
3361 return err;
3362}
3363
3364static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3365 struct dentry *dentry, const char *symname)
3366{
3367 handle_t *handle;
3368 struct inode *inode;
3369 int err, len = strlen(symname);
3370 int credits;
3371 struct fscrypt_str disk_link;
3372 int retries = 0;
3373
3374 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3375 return -EIO;
3376
3377 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3378 &disk_link);
3379 if (err)
3380 return err;
3381
3382 err = dquot_initialize(dir);
3383 if (err)
3384 return err;
3385
3386 /*
3387 * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3388 * directory. +3 for inode, inode bitmap, group descriptor allocation.
3389 * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3390 * modification.
3391 */
3392 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3393 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3394retry:
3395 inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3396 &dentry->d_name, 0, NULL,
3397 EXT4_HT_DIR, credits);
3398 handle = ext4_journal_current_handle();
3399 if (IS_ERR(inode)) {
3400 if (handle)
3401 ext4_journal_stop(handle);
3402 err = PTR_ERR(inode);
3403 goto out_retry;
3404 }
3405
3406 if (IS_ENCRYPTED(inode)) {
3407 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3408 if (err)
3409 goto err_drop_inode;
3410 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3411 } else {
3412 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3413 inode->i_op = &ext4_symlink_inode_operations;
3414 } else {
3415 inode->i_op = &ext4_fast_symlink_inode_operations;
3416 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3417 }
3418 }
3419
3420 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3421 /* alloc symlink block and fill it */
3422 err = ext4_init_symlink_block(handle, inode, &disk_link);
3423 if (err)
3424 goto err_drop_inode;
3425 } else {
3426 /* clear the extent format for fast symlink */
3427 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3428 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3429 disk_link.len);
3430 inode->i_size = disk_link.len - 1;
3431 EXT4_I(inode)->i_disksize = inode->i_size;
3432 }
3433 err = ext4_add_nondir(handle, dentry, &inode);
3434 if (handle)
3435 ext4_journal_stop(handle);
3436 iput(inode);
3437 goto out_retry;
3438
3439err_drop_inode:
3440 clear_nlink(inode);
3441 ext4_mark_inode_dirty(handle, inode);
3442 ext4_orphan_add(handle, inode);
3443 unlock_new_inode(inode);
3444 if (handle)
3445 ext4_journal_stop(handle);
3446 iput(inode);
3447out_retry:
3448 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3449 goto retry;
3450 if (disk_link.name != (unsigned char *)symname)
3451 kfree(disk_link.name);
3452 return err;
3453}
3454
3455int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3456{
3457 handle_t *handle;
3458 int err, retries = 0;
3459retry:
3460 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3461 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3462 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3463 if (IS_ERR(handle))
3464 return PTR_ERR(handle);
3465
3466 if (IS_DIRSYNC(dir))
3467 ext4_handle_sync(handle);
3468
3469 inode_set_ctime_current(inode);
3470 ext4_inc_count(inode);
3471 ihold(inode);
3472
3473 err = ext4_add_entry(handle, dentry, inode);
3474 if (!err) {
3475 err = ext4_mark_inode_dirty(handle, inode);
3476 /* this can happen only for tmpfile being
3477 * linked the first time
3478 */
3479 if (inode->i_nlink == 1)
3480 ext4_orphan_del(handle, inode);
3481 d_instantiate(dentry, inode);
3482 ext4_fc_track_link(handle, dentry);
3483 } else {
3484 drop_nlink(inode);
3485 iput(inode);
3486 }
3487 ext4_journal_stop(handle);
3488 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3489 goto retry;
3490 return err;
3491}
3492
3493static int ext4_link(struct dentry *old_dentry,
3494 struct inode *dir, struct dentry *dentry)
3495{
3496 struct inode *inode = d_inode(old_dentry);
3497 int err;
3498
3499 if (inode->i_nlink >= EXT4_LINK_MAX)
3500 return -EMLINK;
3501
3502 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3503 if (err)
3504 return err;
3505
3506 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3507 (!projid_eq(EXT4_I(dir)->i_projid,
3508 EXT4_I(old_dentry->d_inode)->i_projid)))
3509 return -EXDEV;
3510
3511 err = dquot_initialize(dir);
3512 if (err)
3513 return err;
3514 return __ext4_link(dir, inode, dentry);
3515}
3516
3517/*
3518 * Try to find buffer head where contains the parent block.
3519 * It should be the inode block if it is inlined or the 1st block
3520 * if it is a normal dir.
3521 */
3522static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3523 struct inode *inode,
3524 int *retval,
3525 struct ext4_dir_entry_2 **parent_de,
3526 int *inlined)
3527{
3528 struct buffer_head *bh;
3529
3530 if (!ext4_has_inline_data(inode)) {
3531 struct ext4_dir_entry_2 *de;
3532 unsigned int offset;
3533
3534 /* The first directory block must not be a hole, so
3535 * treat it as DIRENT_HTREE
3536 */
3537 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3538 if (IS_ERR(bh)) {
3539 *retval = PTR_ERR(bh);
3540 return NULL;
3541 }
3542
3543 de = (struct ext4_dir_entry_2 *) bh->b_data;
3544 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3545 bh->b_size, 0) ||
3546 le32_to_cpu(de->inode) != inode->i_ino ||
3547 strcmp(".", de->name)) {
3548 EXT4_ERROR_INODE(inode, "directory missing '.'");
3549 brelse(bh);
3550 *retval = -EFSCORRUPTED;
3551 return NULL;
3552 }
3553 offset = ext4_rec_len_from_disk(de->rec_len,
3554 inode->i_sb->s_blocksize);
3555 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3556 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3557 bh->b_size, offset) ||
3558 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3559 EXT4_ERROR_INODE(inode, "directory missing '..'");
3560 brelse(bh);
3561 *retval = -EFSCORRUPTED;
3562 return NULL;
3563 }
3564 *parent_de = de;
3565
3566 return bh;
3567 }
3568
3569 *inlined = 1;
3570 return ext4_get_first_inline_block(inode, parent_de, retval);
3571}
3572
3573struct ext4_renament {
3574 struct inode *dir;
3575 struct dentry *dentry;
3576 struct inode *inode;
3577 bool is_dir;
3578 int dir_nlink_delta;
3579
3580 /* entry for "dentry" */
3581 struct buffer_head *bh;
3582 struct ext4_dir_entry_2 *de;
3583 int inlined;
3584
3585 /* entry for ".." in inode if it's a directory */
3586 struct buffer_head *dir_bh;
3587 struct ext4_dir_entry_2 *parent_de;
3588 int dir_inlined;
3589};
3590
3591static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent, bool is_cross)
3592{
3593 int retval;
3594
3595 ent->is_dir = true;
3596 if (!is_cross)
3597 return 0;
3598
3599 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3600 &retval, &ent->parent_de,
3601 &ent->dir_inlined);
3602 if (!ent->dir_bh)
3603 return retval;
3604 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3605 return -EFSCORRUPTED;
3606 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3607 return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3608 ent->dir_bh, EXT4_JTR_NONE);
3609}
3610
3611static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3612 unsigned dir_ino)
3613{
3614 int retval;
3615
3616 if (!ent->dir_bh)
3617 return 0;
3618
3619 ent->parent_de->inode = cpu_to_le32(dir_ino);
3620 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3621 if (!ent->dir_inlined) {
3622 if (is_dx(ent->inode)) {
3623 retval = ext4_handle_dirty_dx_node(handle,
3624 ent->inode,
3625 ent->dir_bh);
3626 } else {
3627 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3628 ent->dir_bh);
3629 }
3630 } else {
3631 retval = ext4_mark_inode_dirty(handle, ent->inode);
3632 }
3633 if (retval) {
3634 ext4_std_error(ent->dir->i_sb, retval);
3635 return retval;
3636 }
3637 return 0;
3638}
3639
3640static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3641 unsigned ino, unsigned file_type)
3642{
3643 int retval, retval2;
3644
3645 BUFFER_TRACE(ent->bh, "get write access");
3646 retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3647 EXT4_JTR_NONE);
3648 if (retval)
3649 return retval;
3650 ent->de->inode = cpu_to_le32(ino);
3651 if (ext4_has_feature_filetype(ent->dir->i_sb))
3652 ent->de->file_type = file_type;
3653 inode_inc_iversion(ent->dir);
3654 inode_set_mtime_to_ts(ent->dir, inode_set_ctime_current(ent->dir));
3655 retval = ext4_mark_inode_dirty(handle, ent->dir);
3656 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3657 if (!ent->inlined) {
3658 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3659 if (unlikely(retval2)) {
3660 ext4_std_error(ent->dir->i_sb, retval2);
3661 return retval2;
3662 }
3663 }
3664 return retval;
3665}
3666
3667static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3668 unsigned ino, unsigned file_type)
3669{
3670 struct ext4_renament old = *ent;
3671 int retval = 0;
3672
3673 /*
3674 * old->de could have moved from under us during make indexed dir,
3675 * so the old->de may no longer valid and need to find it again
3676 * before reset old inode info.
3677 */
3678 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3679 &old.inlined);
3680 if (IS_ERR(old.bh))
3681 retval = PTR_ERR(old.bh);
3682 if (!old.bh)
3683 retval = -ENOENT;
3684 if (retval) {
3685 ext4_std_error(old.dir->i_sb, retval);
3686 return;
3687 }
3688
3689 ext4_setent(handle, &old, ino, file_type);
3690 brelse(old.bh);
3691}
3692
3693static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3694 const struct qstr *d_name)
3695{
3696 int retval = -ENOENT;
3697 struct buffer_head *bh;
3698 struct ext4_dir_entry_2 *de;
3699
3700 bh = ext4_find_entry(dir, d_name, &de, NULL);
3701 if (IS_ERR(bh))
3702 return PTR_ERR(bh);
3703 if (bh) {
3704 retval = ext4_delete_entry(handle, dir, de, bh);
3705 brelse(bh);
3706 }
3707 return retval;
3708}
3709
3710static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3711 int force_reread)
3712{
3713 int retval;
3714 /*
3715 * ent->de could have moved from under us during htree split, so make
3716 * sure that we are deleting the right entry. We might also be pointing
3717 * to a stale entry in the unused part of ent->bh so just checking inum
3718 * and the name isn't enough.
3719 */
3720 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3721 ent->de->name_len != ent->dentry->d_name.len ||
3722 strncmp(ent->de->name, ent->dentry->d_name.name,
3723 ent->de->name_len) ||
3724 force_reread) {
3725 retval = ext4_find_delete_entry(handle, ent->dir,
3726 &ent->dentry->d_name);
3727 } else {
3728 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3729 if (retval == -ENOENT) {
3730 retval = ext4_find_delete_entry(handle, ent->dir,
3731 &ent->dentry->d_name);
3732 }
3733 }
3734
3735 if (retval) {
3736 ext4_warning_inode(ent->dir,
3737 "Deleting old file: nlink %d, error=%d",
3738 ent->dir->i_nlink, retval);
3739 }
3740}
3741
3742static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3743{
3744 if (ent->dir_nlink_delta) {
3745 if (ent->dir_nlink_delta == -1)
3746 ext4_dec_count(ent->dir);
3747 else
3748 ext4_inc_count(ent->dir);
3749 ext4_mark_inode_dirty(handle, ent->dir);
3750 }
3751}
3752
3753static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3754 struct ext4_renament *ent,
3755 int credits, handle_t **h)
3756{
3757 struct inode *wh;
3758 handle_t *handle;
3759 int retries = 0;
3760
3761 /*
3762 * for inode block, sb block, group summaries,
3763 * and inode bitmap
3764 */
3765 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3766 EXT4_XATTR_TRANS_BLOCKS + 4);
3767retry:
3768 wh = ext4_new_inode_start_handle(idmap, ent->dir,
3769 S_IFCHR | WHITEOUT_MODE,
3770 &ent->dentry->d_name, 0, NULL,
3771 EXT4_HT_DIR, credits);
3772
3773 handle = ext4_journal_current_handle();
3774 if (IS_ERR(wh)) {
3775 if (handle)
3776 ext4_journal_stop(handle);
3777 if (PTR_ERR(wh) == -ENOSPC &&
3778 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3779 goto retry;
3780 } else {
3781 *h = handle;
3782 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3783 wh->i_op = &ext4_special_inode_operations;
3784 }
3785 return wh;
3786}
3787
3788/*
3789 * Anybody can rename anything with this: the permission checks are left to the
3790 * higher-level routines.
3791 *
3792 * n.b. old_{dentry,inode) refers to the source dentry/inode
3793 * while new_{dentry,inode) refers to the destination dentry/inode
3794 * This comes from rename(const char *oldpath, const char *newpath)
3795 */
3796static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3797 struct dentry *old_dentry, struct inode *new_dir,
3798 struct dentry *new_dentry, unsigned int flags)
3799{
3800 handle_t *handle = NULL;
3801 struct ext4_renament old = {
3802 .dir = old_dir,
3803 .dentry = old_dentry,
3804 .inode = d_inode(old_dentry),
3805 };
3806 struct ext4_renament new = {
3807 .dir = new_dir,
3808 .dentry = new_dentry,
3809 .inode = d_inode(new_dentry),
3810 };
3811 int force_reread;
3812 int retval;
3813 struct inode *whiteout = NULL;
3814 int credits;
3815 u8 old_file_type;
3816
3817 if (new.inode && new.inode->i_nlink == 0) {
3818 EXT4_ERROR_INODE(new.inode,
3819 "target of rename is already freed");
3820 return -EFSCORRUPTED;
3821 }
3822
3823 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3824 (!projid_eq(EXT4_I(new_dir)->i_projid,
3825 EXT4_I(old_dentry->d_inode)->i_projid)))
3826 return -EXDEV;
3827
3828 retval = dquot_initialize(old.dir);
3829 if (retval)
3830 return retval;
3831 retval = dquot_initialize(old.inode);
3832 if (retval)
3833 return retval;
3834 retval = dquot_initialize(new.dir);
3835 if (retval)
3836 return retval;
3837
3838 /* Initialize quotas before so that eventual writes go
3839 * in separate transaction */
3840 if (new.inode) {
3841 retval = dquot_initialize(new.inode);
3842 if (retval)
3843 return retval;
3844 }
3845
3846 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3847 &old.inlined);
3848 if (IS_ERR(old.bh))
3849 return PTR_ERR(old.bh);
3850
3851 /*
3852 * Check for inode number is _not_ due to possible IO errors.
3853 * We might rmdir the source, keep it as pwd of some process
3854 * and merrily kill the link to whatever was created under the
3855 * same name. Goodbye sticky bit ;-<
3856 */
3857 retval = -ENOENT;
3858 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3859 goto release_bh;
3860
3861 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3862 &new.de, &new.inlined);
3863 if (IS_ERR(new.bh)) {
3864 retval = PTR_ERR(new.bh);
3865 new.bh = NULL;
3866 goto release_bh;
3867 }
3868 if (new.bh) {
3869 if (!new.inode) {
3870 brelse(new.bh);
3871 new.bh = NULL;
3872 }
3873 }
3874 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3875 ext4_alloc_da_blocks(old.inode);
3876
3877 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3878 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3879 if (!(flags & RENAME_WHITEOUT)) {
3880 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3881 if (IS_ERR(handle)) {
3882 retval = PTR_ERR(handle);
3883 goto release_bh;
3884 }
3885 } else {
3886 whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3887 if (IS_ERR(whiteout)) {
3888 retval = PTR_ERR(whiteout);
3889 goto release_bh;
3890 }
3891 }
3892
3893 old_file_type = old.de->file_type;
3894 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3895 ext4_handle_sync(handle);
3896
3897 if (S_ISDIR(old.inode->i_mode)) {
3898 if (new.inode) {
3899 retval = -ENOTEMPTY;
3900 if (!ext4_empty_dir(new.inode))
3901 goto end_rename;
3902 } else {
3903 retval = -EMLINK;
3904 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3905 goto end_rename;
3906 }
3907 retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
3908 if (retval)
3909 goto end_rename;
3910 }
3911 /*
3912 * If we're renaming a file within an inline_data dir and adding or
3913 * setting the new dirent causes a conversion from inline_data to
3914 * extents/blockmap, we need to force the dirent delete code to
3915 * re-read the directory, or else we end up trying to delete a dirent
3916 * from what is now the extent tree root (or a block map).
3917 */
3918 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3919 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3920
3921 if (whiteout) {
3922 /*
3923 * Do this before adding a new entry, so the old entry is sure
3924 * to be still pointing to the valid old entry.
3925 */
3926 retval = ext4_setent(handle, &old, whiteout->i_ino,
3927 EXT4_FT_CHRDEV);
3928 if (retval)
3929 goto end_rename;
3930 retval = ext4_mark_inode_dirty(handle, whiteout);
3931 if (unlikely(retval))
3932 goto end_rename;
3933
3934 }
3935 if (!new.bh) {
3936 retval = ext4_add_entry(handle, new.dentry, old.inode);
3937 if (retval)
3938 goto end_rename;
3939 } else {
3940 retval = ext4_setent(handle, &new,
3941 old.inode->i_ino, old_file_type);
3942 if (retval)
3943 goto end_rename;
3944 }
3945 if (force_reread)
3946 force_reread = !ext4_test_inode_flag(new.dir,
3947 EXT4_INODE_INLINE_DATA);
3948
3949 /*
3950 * Like most other Unix systems, set the ctime for inodes on a
3951 * rename.
3952 */
3953 inode_set_ctime_current(old.inode);
3954 retval = ext4_mark_inode_dirty(handle, old.inode);
3955 if (unlikely(retval))
3956 goto end_rename;
3957
3958 if (!whiteout) {
3959 /*
3960 * ok, that's it
3961 */
3962 ext4_rename_delete(handle, &old, force_reread);
3963 }
3964
3965 if (new.inode) {
3966 ext4_dec_count(new.inode);
3967 inode_set_ctime_current(new.inode);
3968 }
3969 inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
3970 ext4_update_dx_flag(old.dir);
3971 if (old.is_dir) {
3972 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3973 if (retval)
3974 goto end_rename;
3975
3976 ext4_dec_count(old.dir);
3977 if (new.inode) {
3978 /* checked ext4_empty_dir above, can't have another
3979 * parent, ext4_dec_count() won't work for many-linked
3980 * dirs */
3981 clear_nlink(new.inode);
3982 } else {
3983 ext4_inc_count(new.dir);
3984 ext4_update_dx_flag(new.dir);
3985 retval = ext4_mark_inode_dirty(handle, new.dir);
3986 if (unlikely(retval))
3987 goto end_rename;
3988 }
3989 }
3990 retval = ext4_mark_inode_dirty(handle, old.dir);
3991 if (unlikely(retval))
3992 goto end_rename;
3993
3994 if (old.is_dir) {
3995 /*
3996 * We disable fast commits here that's because the
3997 * replay code is not yet capable of changing dot dot
3998 * dirents in directories.
3999 */
4000 ext4_fc_mark_ineligible(old.inode->i_sb,
4001 EXT4_FC_REASON_RENAME_DIR, handle);
4002 } else {
4003 struct super_block *sb = old.inode->i_sb;
4004
4005 if (new.inode)
4006 ext4_fc_track_unlink(handle, new.dentry);
4007 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4008 !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4009 !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
4010 __ext4_fc_track_link(handle, old.inode, new.dentry);
4011 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4012 if (whiteout)
4013 __ext4_fc_track_create(handle, whiteout,
4014 old.dentry);
4015 }
4016 }
4017
4018 if (new.inode) {
4019 retval = ext4_mark_inode_dirty(handle, new.inode);
4020 if (unlikely(retval))
4021 goto end_rename;
4022 if (!new.inode->i_nlink)
4023 ext4_orphan_add(handle, new.inode);
4024 }
4025 retval = 0;
4026
4027end_rename:
4028 if (whiteout) {
4029 if (retval) {
4030 ext4_resetent(handle, &old,
4031 old.inode->i_ino, old_file_type);
4032 drop_nlink(whiteout);
4033 ext4_mark_inode_dirty(handle, whiteout);
4034 ext4_orphan_add(handle, whiteout);
4035 }
4036 unlock_new_inode(whiteout);
4037 ext4_journal_stop(handle);
4038 iput(whiteout);
4039 } else {
4040 ext4_journal_stop(handle);
4041 }
4042release_bh:
4043 brelse(old.dir_bh);
4044 brelse(old.bh);
4045 brelse(new.bh);
4046
4047 return retval;
4048}
4049
4050static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4051 struct inode *new_dir, struct dentry *new_dentry)
4052{
4053 handle_t *handle = NULL;
4054 struct ext4_renament old = {
4055 .dir = old_dir,
4056 .dentry = old_dentry,
4057 .inode = d_inode(old_dentry),
4058 };
4059 struct ext4_renament new = {
4060 .dir = new_dir,
4061 .dentry = new_dentry,
4062 .inode = d_inode(new_dentry),
4063 };
4064 u8 new_file_type;
4065 int retval;
4066
4067 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4068 !projid_eq(EXT4_I(new_dir)->i_projid,
4069 EXT4_I(old_dentry->d_inode)->i_projid)) ||
4070 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4071 !projid_eq(EXT4_I(old_dir)->i_projid,
4072 EXT4_I(new_dentry->d_inode)->i_projid)))
4073 return -EXDEV;
4074
4075 retval = dquot_initialize(old.dir);
4076 if (retval)
4077 return retval;
4078 retval = dquot_initialize(new.dir);
4079 if (retval)
4080 return retval;
4081
4082 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4083 &old.de, &old.inlined);
4084 if (IS_ERR(old.bh))
4085 return PTR_ERR(old.bh);
4086 /*
4087 * Check for inode number is _not_ due to possible IO errors.
4088 * We might rmdir the source, keep it as pwd of some process
4089 * and merrily kill the link to whatever was created under the
4090 * same name. Goodbye sticky bit ;-<
4091 */
4092 retval = -ENOENT;
4093 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4094 goto end_rename;
4095
4096 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4097 &new.de, &new.inlined);
4098 if (IS_ERR(new.bh)) {
4099 retval = PTR_ERR(new.bh);
4100 new.bh = NULL;
4101 goto end_rename;
4102 }
4103
4104 /* RENAME_EXCHANGE case: old *and* new must both exist */
4105 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4106 goto end_rename;
4107
4108 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4109 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4110 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4111 if (IS_ERR(handle)) {
4112 retval = PTR_ERR(handle);
4113 handle = NULL;
4114 goto end_rename;
4115 }
4116
4117 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4118 ext4_handle_sync(handle);
4119
4120 if (S_ISDIR(old.inode->i_mode)) {
4121 retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
4122 if (retval)
4123 goto end_rename;
4124 }
4125 if (S_ISDIR(new.inode->i_mode)) {
4126 retval = ext4_rename_dir_prepare(handle, &new, new.dir != old.dir);
4127 if (retval)
4128 goto end_rename;
4129 }
4130
4131 /*
4132 * Other than the special case of overwriting a directory, parents'
4133 * nlink only needs to be modified if this is a cross directory rename.
4134 */
4135 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4136 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4137 new.dir_nlink_delta = -old.dir_nlink_delta;
4138 retval = -EMLINK;
4139 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4140 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4141 goto end_rename;
4142 }
4143
4144 new_file_type = new.de->file_type;
4145 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4146 if (retval)
4147 goto end_rename;
4148
4149 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4150 if (retval)
4151 goto end_rename;
4152
4153 /*
4154 * Like most other Unix systems, set the ctime for inodes on a
4155 * rename.
4156 */
4157 inode_set_ctime_current(old.inode);
4158 inode_set_ctime_current(new.inode);
4159 retval = ext4_mark_inode_dirty(handle, old.inode);
4160 if (unlikely(retval))
4161 goto end_rename;
4162 retval = ext4_mark_inode_dirty(handle, new.inode);
4163 if (unlikely(retval))
4164 goto end_rename;
4165 ext4_fc_mark_ineligible(new.inode->i_sb,
4166 EXT4_FC_REASON_CROSS_RENAME, handle);
4167 if (old.dir_bh) {
4168 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4169 if (retval)
4170 goto end_rename;
4171 }
4172 if (new.dir_bh) {
4173 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4174 if (retval)
4175 goto end_rename;
4176 }
4177 ext4_update_dir_count(handle, &old);
4178 ext4_update_dir_count(handle, &new);
4179 retval = 0;
4180
4181end_rename:
4182 brelse(old.dir_bh);
4183 brelse(new.dir_bh);
4184 brelse(old.bh);
4185 brelse(new.bh);
4186 if (handle)
4187 ext4_journal_stop(handle);
4188 return retval;
4189}
4190
4191static int ext4_rename2(struct mnt_idmap *idmap,
4192 struct inode *old_dir, struct dentry *old_dentry,
4193 struct inode *new_dir, struct dentry *new_dentry,
4194 unsigned int flags)
4195{
4196 int err;
4197
4198 if (unlikely(ext4_forced_shutdown(old_dir->i_sb)))
4199 return -EIO;
4200
4201 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4202 return -EINVAL;
4203
4204 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4205 flags);
4206 if (err)
4207 return err;
4208
4209 if (flags & RENAME_EXCHANGE) {
4210 return ext4_cross_rename(old_dir, old_dentry,
4211 new_dir, new_dentry);
4212 }
4213
4214 return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4215}
4216
4217/*
4218 * directories can handle most operations...
4219 */
4220const struct inode_operations ext4_dir_inode_operations = {
4221 .create = ext4_create,
4222 .lookup = ext4_lookup,
4223 .link = ext4_link,
4224 .unlink = ext4_unlink,
4225 .symlink = ext4_symlink,
4226 .mkdir = ext4_mkdir,
4227 .rmdir = ext4_rmdir,
4228 .mknod = ext4_mknod,
4229 .tmpfile = ext4_tmpfile,
4230 .rename = ext4_rename2,
4231 .setattr = ext4_setattr,
4232 .getattr = ext4_getattr,
4233 .listxattr = ext4_listxattr,
4234 .get_inode_acl = ext4_get_acl,
4235 .set_acl = ext4_set_acl,
4236 .fiemap = ext4_fiemap,
4237 .fileattr_get = ext4_fileattr_get,
4238 .fileattr_set = ext4_fileattr_set,
4239};
4240
4241const struct inode_operations ext4_special_inode_operations = {
4242 .setattr = ext4_setattr,
4243 .getattr = ext4_getattr,
4244 .listxattr = ext4_listxattr,
4245 .get_inode_acl = ext4_get_acl,
4246 .set_acl = ext4_set_acl,
4247};