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 buffer_head *bh;
58 int err;
59
60 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
61 ((inode->i_size >> 10) >=
62 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
63 return ERR_PTR(-ENOSPC);
64
65 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66
67 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
68 if (IS_ERR(bh))
69 return bh;
70 inode->i_size += inode->i_sb->s_blocksize;
71 EXT4_I(inode)->i_disksize = inode->i_size;
72 BUFFER_TRACE(bh, "get_write_access");
73 err = ext4_journal_get_write_access(handle, bh);
74 if (err) {
75 brelse(bh);
76 ext4_std_error(inode->i_sb, err);
77 return ERR_PTR(err);
78 }
79 return bh;
80}
81
82static int ext4_dx_csum_verify(struct inode *inode,
83 struct ext4_dir_entry *dirent);
84
85/*
86 * Hints to ext4_read_dirblock regarding whether we expect a directory
87 * block being read to be an index block, or a block containing
88 * directory entries (and if the latter, whether it was found via a
89 * logical block in an htree index block). This is used to control
90 * what sort of sanity checkinig ext4_read_dirblock() will do on the
91 * directory block read from the storage device. EITHER will means
92 * the caller doesn't know what kind of directory block will be read,
93 * so no specific verification will be done.
94 */
95typedef enum {
96 EITHER, INDEX, DIRENT, DIRENT_HTREE
97} dirblock_type_t;
98
99#define ext4_read_dirblock(inode, block, type) \
100 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
101
102static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
103 ext4_lblk_t block,
104 dirblock_type_t type,
105 const char *func,
106 unsigned int line)
107{
108 struct buffer_head *bh;
109 struct ext4_dir_entry *dirent;
110 int is_dx_block = 0;
111
112 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
113 bh = ERR_PTR(-EIO);
114 else
115 bh = ext4_bread(NULL, inode, block, 0);
116 if (IS_ERR(bh)) {
117 __ext4_warning(inode->i_sb, func, line,
118 "inode #%lu: lblock %lu: comm %s: "
119 "error %ld reading directory block",
120 inode->i_ino, (unsigned long)block,
121 current->comm, PTR_ERR(bh));
122
123 return bh;
124 }
125 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
126 ext4_error_inode(inode, func, line, block,
127 "Directory hole found for htree %s block",
128 (type == INDEX) ? "index" : "leaf");
129 return ERR_PTR(-EFSCORRUPTED);
130 }
131 if (!bh)
132 return NULL;
133 dirent = (struct ext4_dir_entry *) bh->b_data;
134 /* Determine whether or not we have an index block */
135 if (is_dx(inode)) {
136 if (block == 0)
137 is_dx_block = 1;
138 else if (ext4_rec_len_from_disk(dirent->rec_len,
139 inode->i_sb->s_blocksize) ==
140 inode->i_sb->s_blocksize)
141 is_dx_block = 1;
142 }
143 if (!is_dx_block && type == INDEX) {
144 ext4_error_inode(inode, func, line, block,
145 "directory leaf block found instead of index block");
146 brelse(bh);
147 return ERR_PTR(-EFSCORRUPTED);
148 }
149 if (!ext4_has_metadata_csum(inode->i_sb) ||
150 buffer_verified(bh))
151 return bh;
152
153 /*
154 * An empty leaf block can get mistaken for a index block; for
155 * this reason, we can only check the index checksum when the
156 * caller is sure it should be an index block.
157 */
158 if (is_dx_block && type == INDEX) {
159 if (ext4_dx_csum_verify(inode, dirent) &&
160 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
161 set_buffer_verified(bh);
162 else {
163 ext4_error_inode_err(inode, func, line, block,
164 EFSBADCRC,
165 "Directory index failed checksum");
166 brelse(bh);
167 return ERR_PTR(-EFSBADCRC);
168 }
169 }
170 if (!is_dx_block) {
171 if (ext4_dirblock_csum_verify(inode, bh) &&
172 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
173 set_buffer_verified(bh);
174 else {
175 ext4_error_inode_err(inode, func, line, block,
176 EFSBADCRC,
177 "Directory block failed checksum");
178 brelse(bh);
179 return ERR_PTR(-EFSBADCRC);
180 }
181 }
182 return bh;
183}
184
185#ifdef DX_DEBUG
186#define dxtrace(command) command
187#else
188#define dxtrace(command)
189#endif
190
191struct fake_dirent
192{
193 __le32 inode;
194 __le16 rec_len;
195 u8 name_len;
196 u8 file_type;
197};
198
199struct dx_countlimit
200{
201 __le16 limit;
202 __le16 count;
203};
204
205struct dx_entry
206{
207 __le32 hash;
208 __le32 block;
209};
210
211/*
212 * dx_root_info is laid out so that if it should somehow get overlaid by a
213 * dirent the two low bits of the hash version will be zero. Therefore, the
214 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
215 */
216
217struct dx_root
218{
219 struct fake_dirent dot;
220 char dot_name[4];
221 struct fake_dirent dotdot;
222 char dotdot_name[4];
223 struct dx_root_info
224 {
225 __le32 reserved_zero;
226 u8 hash_version;
227 u8 info_length; /* 8 */
228 u8 indirect_levels;
229 u8 unused_flags;
230 }
231 info;
232 struct dx_entry entries[];
233};
234
235struct dx_node
236{
237 struct fake_dirent fake;
238 struct dx_entry entries[];
239};
240
241
242struct dx_frame
243{
244 struct buffer_head *bh;
245 struct dx_entry *entries;
246 struct dx_entry *at;
247};
248
249struct dx_map_entry
250{
251 u32 hash;
252 u16 offs;
253 u16 size;
254};
255
256/*
257 * This goes at the end of each htree block.
258 */
259struct dx_tail {
260 u32 dt_reserved;
261 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
262};
263
264static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
265static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
266static inline unsigned dx_get_hash(struct dx_entry *entry);
267static void dx_set_hash(struct dx_entry *entry, unsigned value);
268static unsigned dx_get_count(struct dx_entry *entries);
269static unsigned dx_get_limit(struct dx_entry *entries);
270static void dx_set_count(struct dx_entry *entries, unsigned value);
271static void dx_set_limit(struct dx_entry *entries, unsigned value);
272static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
273static unsigned dx_node_limit(struct inode *dir);
274static struct dx_frame *dx_probe(struct ext4_filename *fname,
275 struct inode *dir,
276 struct dx_hash_info *hinfo,
277 struct dx_frame *frame);
278static void dx_release(struct dx_frame *frames);
279static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
280 unsigned blocksize, struct dx_hash_info *hinfo,
281 struct dx_map_entry map[]);
282static void dx_sort_map(struct dx_map_entry *map, unsigned count);
283static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
284 char *to, struct dx_map_entry *offsets,
285 int count, unsigned int blocksize);
286static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
287 unsigned int blocksize);
288static void dx_insert_block(struct dx_frame *frame,
289 u32 hash, ext4_lblk_t block);
290static int ext4_htree_next_block(struct inode *dir, __u32 hash,
291 struct dx_frame *frame,
292 struct dx_frame *frames,
293 __u32 *start_hash);
294static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
295 struct ext4_filename *fname,
296 struct ext4_dir_entry_2 **res_dir);
297static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
298 struct inode *dir, struct inode *inode);
299
300/* checksumming functions */
301void ext4_initialize_dirent_tail(struct buffer_head *bh,
302 unsigned int blocksize)
303{
304 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
305
306 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
307 t->det_rec_len = ext4_rec_len_to_disk(
308 sizeof(struct ext4_dir_entry_tail), blocksize);
309 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
310}
311
312/* Walk through a dirent block to find a checksum "dirent" at the tail */
313static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
314 struct buffer_head *bh)
315{
316 struct ext4_dir_entry_tail *t;
317
318#ifdef PARANOID
319 struct ext4_dir_entry *d, *top;
320
321 d = (struct ext4_dir_entry *)bh->b_data;
322 top = (struct ext4_dir_entry *)(bh->b_data +
323 (EXT4_BLOCK_SIZE(inode->i_sb) -
324 sizeof(struct ext4_dir_entry_tail)));
325 while (d < top && d->rec_len)
326 d = (struct ext4_dir_entry *)(((void *)d) +
327 le16_to_cpu(d->rec_len));
328
329 if (d != top)
330 return NULL;
331
332 t = (struct ext4_dir_entry_tail *)d;
333#else
334 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
335#endif
336
337 if (t->det_reserved_zero1 ||
338 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
339 t->det_reserved_zero2 ||
340 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
341 return NULL;
342
343 return t;
344}
345
346static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
347{
348 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
349 struct ext4_inode_info *ei = EXT4_I(inode);
350 __u32 csum;
351
352 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
353 return cpu_to_le32(csum);
354}
355
356#define warn_no_space_for_csum(inode) \
357 __warn_no_space_for_csum((inode), __func__, __LINE__)
358
359static void __warn_no_space_for_csum(struct inode *inode, const char *func,
360 unsigned int line)
361{
362 __ext4_warning_inode(inode, func, line,
363 "No space for directory leaf checksum. Please run e2fsck -D.");
364}
365
366int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
367{
368 struct ext4_dir_entry_tail *t;
369
370 if (!ext4_has_metadata_csum(inode->i_sb))
371 return 1;
372
373 t = get_dirent_tail(inode, bh);
374 if (!t) {
375 warn_no_space_for_csum(inode);
376 return 0;
377 }
378
379 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
380 (char *)t - bh->b_data))
381 return 0;
382
383 return 1;
384}
385
386static void ext4_dirblock_csum_set(struct inode *inode,
387 struct buffer_head *bh)
388{
389 struct ext4_dir_entry_tail *t;
390
391 if (!ext4_has_metadata_csum(inode->i_sb))
392 return;
393
394 t = get_dirent_tail(inode, bh);
395 if (!t) {
396 warn_no_space_for_csum(inode);
397 return;
398 }
399
400 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
401 (char *)t - bh->b_data);
402}
403
404int ext4_handle_dirty_dirblock(handle_t *handle,
405 struct inode *inode,
406 struct buffer_head *bh)
407{
408 ext4_dirblock_csum_set(inode, bh);
409 return ext4_handle_dirty_metadata(handle, inode, bh);
410}
411
412static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
413 struct ext4_dir_entry *dirent,
414 int *offset)
415{
416 struct ext4_dir_entry *dp;
417 struct dx_root_info *root;
418 int count_offset;
419
420 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
421 count_offset = 8;
422 else if (le16_to_cpu(dirent->rec_len) == 12) {
423 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
424 if (le16_to_cpu(dp->rec_len) !=
425 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
426 return NULL;
427 root = (struct dx_root_info *)(((void *)dp + 12));
428 if (root->reserved_zero ||
429 root->info_length != sizeof(struct dx_root_info))
430 return NULL;
431 count_offset = 32;
432 } else
433 return NULL;
434
435 if (offset)
436 *offset = count_offset;
437 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
438}
439
440static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
441 int count_offset, int count, struct dx_tail *t)
442{
443 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
444 struct ext4_inode_info *ei = EXT4_I(inode);
445 __u32 csum;
446 int size;
447 __u32 dummy_csum = 0;
448 int offset = offsetof(struct dx_tail, dt_checksum);
449
450 size = count_offset + (count * sizeof(struct dx_entry));
451 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
452 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
453 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
454
455 return cpu_to_le32(csum);
456}
457
458static int ext4_dx_csum_verify(struct inode *inode,
459 struct ext4_dir_entry *dirent)
460{
461 struct dx_countlimit *c;
462 struct dx_tail *t;
463 int count_offset, limit, count;
464
465 if (!ext4_has_metadata_csum(inode->i_sb))
466 return 1;
467
468 c = get_dx_countlimit(inode, dirent, &count_offset);
469 if (!c) {
470 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
471 return 0;
472 }
473 limit = le16_to_cpu(c->limit);
474 count = le16_to_cpu(c->count);
475 if (count_offset + (limit * sizeof(struct dx_entry)) >
476 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
477 warn_no_space_for_csum(inode);
478 return 0;
479 }
480 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
481
482 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
483 count, t))
484 return 0;
485 return 1;
486}
487
488static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
489{
490 struct dx_countlimit *c;
491 struct dx_tail *t;
492 int count_offset, limit, count;
493
494 if (!ext4_has_metadata_csum(inode->i_sb))
495 return;
496
497 c = get_dx_countlimit(inode, dirent, &count_offset);
498 if (!c) {
499 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
500 return;
501 }
502 limit = le16_to_cpu(c->limit);
503 count = le16_to_cpu(c->count);
504 if (count_offset + (limit * sizeof(struct dx_entry)) >
505 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
506 warn_no_space_for_csum(inode);
507 return;
508 }
509 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
510
511 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
512}
513
514static inline int ext4_handle_dirty_dx_node(handle_t *handle,
515 struct inode *inode,
516 struct buffer_head *bh)
517{
518 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
519 return ext4_handle_dirty_metadata(handle, inode, bh);
520}
521
522/*
523 * p is at least 6 bytes before the end of page
524 */
525static inline struct ext4_dir_entry_2 *
526ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
527{
528 return (struct ext4_dir_entry_2 *)((char *)p +
529 ext4_rec_len_from_disk(p->rec_len, blocksize));
530}
531
532/*
533 * Future: use high four bits of block for coalesce-on-delete flags
534 * Mask them off for now.
535 */
536
537static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
538{
539 return le32_to_cpu(entry->block) & 0x0fffffff;
540}
541
542static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
543{
544 entry->block = cpu_to_le32(value);
545}
546
547static inline unsigned dx_get_hash(struct dx_entry *entry)
548{
549 return le32_to_cpu(entry->hash);
550}
551
552static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
553{
554 entry->hash = cpu_to_le32(value);
555}
556
557static inline unsigned dx_get_count(struct dx_entry *entries)
558{
559 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
560}
561
562static inline unsigned dx_get_limit(struct dx_entry *entries)
563{
564 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
565}
566
567static inline void dx_set_count(struct dx_entry *entries, unsigned value)
568{
569 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
570}
571
572static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
573{
574 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
575}
576
577static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
578{
579 unsigned int entry_space = dir->i_sb->s_blocksize -
580 ext4_dir_rec_len(1, NULL) -
581 ext4_dir_rec_len(2, NULL) - infosize;
582
583 if (ext4_has_metadata_csum(dir->i_sb))
584 entry_space -= sizeof(struct dx_tail);
585 return entry_space / sizeof(struct dx_entry);
586}
587
588static inline unsigned dx_node_limit(struct inode *dir)
589{
590 unsigned int entry_space = dir->i_sb->s_blocksize -
591 ext4_dir_rec_len(0, dir);
592
593 if (ext4_has_metadata_csum(dir->i_sb))
594 entry_space -= sizeof(struct dx_tail);
595 return entry_space / sizeof(struct dx_entry);
596}
597
598/*
599 * Debug
600 */
601#ifdef DX_DEBUG
602static void dx_show_index(char * label, struct dx_entry *entries)
603{
604 int i, n = dx_get_count (entries);
605 printk(KERN_DEBUG "%s index", label);
606 for (i = 0; i < n; i++) {
607 printk(KERN_CONT " %x->%lu",
608 i ? dx_get_hash(entries + i) : 0,
609 (unsigned long)dx_get_block(entries + i));
610 }
611 printk(KERN_CONT "\n");
612}
613
614struct stats
615{
616 unsigned names;
617 unsigned space;
618 unsigned bcount;
619};
620
621static struct stats dx_show_leaf(struct inode *dir,
622 struct dx_hash_info *hinfo,
623 struct ext4_dir_entry_2 *de,
624 int size, int show_names)
625{
626 unsigned names = 0, space = 0;
627 char *base = (char *) de;
628 struct dx_hash_info h = *hinfo;
629
630 printk("names: ");
631 while ((char *) de < base + size)
632 {
633 if (de->inode)
634 {
635 if (show_names)
636 {
637#ifdef CONFIG_FS_ENCRYPTION
638 int len;
639 char *name;
640 struct fscrypt_str fname_crypto_str =
641 FSTR_INIT(NULL, 0);
642 int res = 0;
643
644 name = de->name;
645 len = de->name_len;
646 if (!IS_ENCRYPTED(dir)) {
647 /* Directory is not encrypted */
648 ext4fs_dirhash(dir, de->name,
649 de->name_len, &h);
650 printk("%*.s:(U)%x.%u ", len,
651 name, h.hash,
652 (unsigned) ((char *) de
653 - base));
654 } else {
655 struct fscrypt_str de_name =
656 FSTR_INIT(name, len);
657
658 /* Directory is encrypted */
659 res = fscrypt_fname_alloc_buffer(
660 len, &fname_crypto_str);
661 if (res)
662 printk(KERN_WARNING "Error "
663 "allocating crypto "
664 "buffer--skipping "
665 "crypto\n");
666 res = fscrypt_fname_disk_to_usr(dir,
667 0, 0, &de_name,
668 &fname_crypto_str);
669 if (res) {
670 printk(KERN_WARNING "Error "
671 "converting filename "
672 "from disk to usr"
673 "\n");
674 name = "??";
675 len = 2;
676 } else {
677 name = fname_crypto_str.name;
678 len = fname_crypto_str.len;
679 }
680 if (IS_CASEFOLDED(dir))
681 h.hash = EXT4_DIRENT_HASH(de);
682 else
683 ext4fs_dirhash(dir, de->name,
684 de->name_len, &h);
685 printk("%*.s:(E)%x.%u ", len, name,
686 h.hash, (unsigned) ((char *) de
687 - base));
688 fscrypt_fname_free_buffer(
689 &fname_crypto_str);
690 }
691#else
692 int len = de->name_len;
693 char *name = de->name;
694 ext4fs_dirhash(dir, de->name, de->name_len, &h);
695 printk("%*.s:%x.%u ", len, name, h.hash,
696 (unsigned) ((char *) de - base));
697#endif
698 }
699 space += ext4_dir_rec_len(de->name_len, dir);
700 names++;
701 }
702 de = ext4_next_entry(de, size);
703 }
704 printk(KERN_CONT "(%i)\n", names);
705 return (struct stats) { names, space, 1 };
706}
707
708struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
709 struct dx_entry *entries, int levels)
710{
711 unsigned blocksize = dir->i_sb->s_blocksize;
712 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
713 unsigned bcount = 0;
714 struct buffer_head *bh;
715 printk("%i indexed blocks...\n", count);
716 for (i = 0; i < count; i++, entries++)
717 {
718 ext4_lblk_t block = dx_get_block(entries);
719 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
720 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
721 struct stats stats;
722 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
723 bh = ext4_bread(NULL,dir, block, 0);
724 if (!bh || IS_ERR(bh))
725 continue;
726 stats = levels?
727 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
728 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
729 bh->b_data, blocksize, 0);
730 names += stats.names;
731 space += stats.space;
732 bcount += stats.bcount;
733 brelse(bh);
734 }
735 if (bcount)
736 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
737 levels ? "" : " ", names, space/bcount,
738 (space/bcount)*100/blocksize);
739 return (struct stats) { names, space, bcount};
740}
741
742/*
743 * Linear search cross check
744 */
745static inline void htree_rep_invariant_check(struct dx_entry *at,
746 struct dx_entry *target,
747 u32 hash, unsigned int n)
748{
749 while (n--) {
750 dxtrace(printk(KERN_CONT ","));
751 if (dx_get_hash(++at) > hash) {
752 at--;
753 break;
754 }
755 }
756 ASSERT(at == target - 1);
757}
758#else /* DX_DEBUG */
759static inline void htree_rep_invariant_check(struct dx_entry *at,
760 struct dx_entry *target,
761 u32 hash, unsigned int n)
762{
763}
764#endif /* DX_DEBUG */
765
766/*
767 * Probe for a directory leaf block to search.
768 *
769 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
770 * error in the directory index, and the caller should fall back to
771 * searching the directory normally. The callers of dx_probe **MUST**
772 * check for this error code, and make sure it never gets reflected
773 * back to userspace.
774 */
775static struct dx_frame *
776dx_probe(struct ext4_filename *fname, struct inode *dir,
777 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
778{
779 unsigned count, indirect;
780 struct dx_entry *at, *entries, *p, *q, *m;
781 struct dx_root *root;
782 struct dx_frame *frame = frame_in;
783 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
784 u32 hash;
785
786 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
787 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
788 if (IS_ERR(frame->bh))
789 return (struct dx_frame *) frame->bh;
790
791 root = (struct dx_root *) frame->bh->b_data;
792 if (root->info.hash_version != DX_HASH_TEA &&
793 root->info.hash_version != DX_HASH_HALF_MD4 &&
794 root->info.hash_version != DX_HASH_LEGACY &&
795 root->info.hash_version != DX_HASH_SIPHASH) {
796 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
797 root->info.hash_version);
798 goto fail;
799 }
800 if (ext4_hash_in_dirent(dir)) {
801 if (root->info.hash_version != DX_HASH_SIPHASH) {
802 ext4_warning_inode(dir,
803 "Hash in dirent, but hash is not SIPHASH");
804 goto fail;
805 }
806 } else {
807 if (root->info.hash_version == DX_HASH_SIPHASH) {
808 ext4_warning_inode(dir,
809 "Hash code is SIPHASH, but hash not in dirent");
810 goto fail;
811 }
812 }
813 if (fname)
814 hinfo = &fname->hinfo;
815 hinfo->hash_version = root->info.hash_version;
816 if (hinfo->hash_version <= DX_HASH_TEA)
817 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
818 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
819 /* hash is already computed for encrypted casefolded directory */
820 if (fname && fname_name(fname) &&
821 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)))
822 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
823 hash = hinfo->hash;
824
825 if (root->info.unused_flags & 1) {
826 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
827 root->info.unused_flags);
828 goto fail;
829 }
830
831 indirect = root->info.indirect_levels;
832 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
833 ext4_warning(dir->i_sb,
834 "Directory (ino: %lu) htree depth %#06x exceed"
835 "supported value", dir->i_ino,
836 ext4_dir_htree_level(dir->i_sb));
837 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
838 ext4_warning(dir->i_sb, "Enable large directory "
839 "feature to access it");
840 }
841 goto fail;
842 }
843
844 entries = (struct dx_entry *)(((char *)&root->info) +
845 root->info.info_length);
846
847 if (dx_get_limit(entries) != dx_root_limit(dir,
848 root->info.info_length)) {
849 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
850 dx_get_limit(entries),
851 dx_root_limit(dir, root->info.info_length));
852 goto fail;
853 }
854
855 dxtrace(printk("Look up %x", hash));
856 while (1) {
857 count = dx_get_count(entries);
858 if (!count || count > dx_get_limit(entries)) {
859 ext4_warning_inode(dir,
860 "dx entry: count %u beyond limit %u",
861 count, dx_get_limit(entries));
862 goto fail;
863 }
864
865 p = entries + 1;
866 q = entries + count - 1;
867 while (p <= q) {
868 m = p + (q - p) / 2;
869 dxtrace(printk(KERN_CONT "."));
870 if (dx_get_hash(m) > hash)
871 q = m - 1;
872 else
873 p = m + 1;
874 }
875
876 htree_rep_invariant_check(entries, p, hash, count - 1);
877
878 at = p - 1;
879 dxtrace(printk(KERN_CONT " %x->%u\n",
880 at == entries ? 0 : dx_get_hash(at),
881 dx_get_block(at)));
882 frame->entries = entries;
883 frame->at = at;
884 if (!indirect--)
885 return frame;
886 frame++;
887 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
888 if (IS_ERR(frame->bh)) {
889 ret_err = (struct dx_frame *) frame->bh;
890 frame->bh = NULL;
891 goto fail;
892 }
893 entries = ((struct dx_node *) frame->bh->b_data)->entries;
894
895 if (dx_get_limit(entries) != dx_node_limit(dir)) {
896 ext4_warning_inode(dir,
897 "dx entry: limit %u != node limit %u",
898 dx_get_limit(entries), dx_node_limit(dir));
899 goto fail;
900 }
901 }
902fail:
903 while (frame >= frame_in) {
904 brelse(frame->bh);
905 frame--;
906 }
907
908 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
909 ext4_warning_inode(dir,
910 "Corrupt directory, running e2fsck is recommended");
911 return ret_err;
912}
913
914static void dx_release(struct dx_frame *frames)
915{
916 struct dx_root_info *info;
917 int i;
918 unsigned int indirect_levels;
919
920 if (frames[0].bh == NULL)
921 return;
922
923 info = &((struct dx_root *)frames[0].bh->b_data)->info;
924 /* save local copy, "info" may be freed after brelse() */
925 indirect_levels = info->indirect_levels;
926 for (i = 0; i <= indirect_levels; i++) {
927 if (frames[i].bh == NULL)
928 break;
929 brelse(frames[i].bh);
930 frames[i].bh = NULL;
931 }
932}
933
934/*
935 * This function increments the frame pointer to search the next leaf
936 * block, and reads in the necessary intervening nodes if the search
937 * should be necessary. Whether or not the search is necessary is
938 * controlled by the hash parameter. If the hash value is even, then
939 * the search is only continued if the next block starts with that
940 * hash value. This is used if we are searching for a specific file.
941 *
942 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
943 *
944 * This function returns 1 if the caller should continue to search,
945 * or 0 if it should not. If there is an error reading one of the
946 * index blocks, it will a negative error code.
947 *
948 * If start_hash is non-null, it will be filled in with the starting
949 * hash of the next page.
950 */
951static int ext4_htree_next_block(struct inode *dir, __u32 hash,
952 struct dx_frame *frame,
953 struct dx_frame *frames,
954 __u32 *start_hash)
955{
956 struct dx_frame *p;
957 struct buffer_head *bh;
958 int num_frames = 0;
959 __u32 bhash;
960
961 p = frame;
962 /*
963 * Find the next leaf page by incrementing the frame pointer.
964 * If we run out of entries in the interior node, loop around and
965 * increment pointer in the parent node. When we break out of
966 * this loop, num_frames indicates the number of interior
967 * nodes need to be read.
968 */
969 while (1) {
970 if (++(p->at) < p->entries + dx_get_count(p->entries))
971 break;
972 if (p == frames)
973 return 0;
974 num_frames++;
975 p--;
976 }
977
978 /*
979 * If the hash is 1, then continue only if the next page has a
980 * continuation hash of any value. This is used for readdir
981 * handling. Otherwise, check to see if the hash matches the
982 * desired continuation hash. If it doesn't, return since
983 * there's no point to read in the successive index pages.
984 */
985 bhash = dx_get_hash(p->at);
986 if (start_hash)
987 *start_hash = bhash;
988 if ((hash & 1) == 0) {
989 if ((bhash & ~1) != hash)
990 return 0;
991 }
992 /*
993 * If the hash is HASH_NB_ALWAYS, we always go to the next
994 * block so no check is necessary
995 */
996 while (num_frames--) {
997 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
998 if (IS_ERR(bh))
999 return PTR_ERR(bh);
1000 p++;
1001 brelse(p->bh);
1002 p->bh = bh;
1003 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1004 }
1005 return 1;
1006}
1007
1008
1009/*
1010 * This function fills a red-black tree with information from a
1011 * directory block. It returns the number directory entries loaded
1012 * into the tree. If there is an error it is returned in err.
1013 */
1014static int htree_dirblock_to_tree(struct file *dir_file,
1015 struct inode *dir, ext4_lblk_t block,
1016 struct dx_hash_info *hinfo,
1017 __u32 start_hash, __u32 start_minor_hash)
1018{
1019 struct buffer_head *bh;
1020 struct ext4_dir_entry_2 *de, *top;
1021 int err = 0, count = 0;
1022 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1023 int csum = ext4_has_metadata_csum(dir->i_sb);
1024
1025 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1026 (unsigned long)block));
1027 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1028 if (IS_ERR(bh))
1029 return PTR_ERR(bh);
1030
1031 de = (struct ext4_dir_entry_2 *) bh->b_data;
1032 /* csum entries are not larger in the casefolded encrypted case */
1033 top = (struct ext4_dir_entry_2 *) ((char *) de +
1034 dir->i_sb->s_blocksize -
1035 ext4_dir_rec_len(0,
1036 csum ? NULL : dir));
1037 /* Check if the directory is encrypted */
1038 if (IS_ENCRYPTED(dir)) {
1039 err = fscrypt_prepare_readdir(dir);
1040 if (err < 0) {
1041 brelse(bh);
1042 return err;
1043 }
1044 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1045 &fname_crypto_str);
1046 if (err < 0) {
1047 brelse(bh);
1048 return err;
1049 }
1050 }
1051
1052 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1053 if (ext4_check_dir_entry(dir, NULL, de, bh,
1054 bh->b_data, bh->b_size,
1055 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1056 + ((char *)de - bh->b_data))) {
1057 /* silently ignore the rest of the block */
1058 break;
1059 }
1060 if (ext4_hash_in_dirent(dir)) {
1061 if (de->name_len && de->inode) {
1062 hinfo->hash = EXT4_DIRENT_HASH(de);
1063 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1064 } else {
1065 hinfo->hash = 0;
1066 hinfo->minor_hash = 0;
1067 }
1068 } else {
1069 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1070 }
1071 if ((hinfo->hash < start_hash) ||
1072 ((hinfo->hash == start_hash) &&
1073 (hinfo->minor_hash < start_minor_hash)))
1074 continue;
1075 if (de->inode == 0)
1076 continue;
1077 if (!IS_ENCRYPTED(dir)) {
1078 tmp_str.name = de->name;
1079 tmp_str.len = de->name_len;
1080 err = ext4_htree_store_dirent(dir_file,
1081 hinfo->hash, hinfo->minor_hash, de,
1082 &tmp_str);
1083 } else {
1084 int save_len = fname_crypto_str.len;
1085 struct fscrypt_str de_name = FSTR_INIT(de->name,
1086 de->name_len);
1087
1088 /* Directory is encrypted */
1089 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1090 hinfo->minor_hash, &de_name,
1091 &fname_crypto_str);
1092 if (err) {
1093 count = err;
1094 goto errout;
1095 }
1096 err = ext4_htree_store_dirent(dir_file,
1097 hinfo->hash, hinfo->minor_hash, de,
1098 &fname_crypto_str);
1099 fname_crypto_str.len = save_len;
1100 }
1101 if (err != 0) {
1102 count = err;
1103 goto errout;
1104 }
1105 count++;
1106 }
1107errout:
1108 brelse(bh);
1109 fscrypt_fname_free_buffer(&fname_crypto_str);
1110 return count;
1111}
1112
1113
1114/*
1115 * This function fills a red-black tree with information from a
1116 * directory. We start scanning the directory in hash order, starting
1117 * at start_hash and start_minor_hash.
1118 *
1119 * This function returns the number of entries inserted into the tree,
1120 * or a negative error code.
1121 */
1122int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1123 __u32 start_minor_hash, __u32 *next_hash)
1124{
1125 struct dx_hash_info hinfo;
1126 struct ext4_dir_entry_2 *de;
1127 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1128 struct inode *dir;
1129 ext4_lblk_t block;
1130 int count = 0;
1131 int ret, err;
1132 __u32 hashval;
1133 struct fscrypt_str tmp_str;
1134
1135 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1136 start_hash, start_minor_hash));
1137 dir = file_inode(dir_file);
1138 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1139 if (ext4_hash_in_dirent(dir))
1140 hinfo.hash_version = DX_HASH_SIPHASH;
1141 else
1142 hinfo.hash_version =
1143 EXT4_SB(dir->i_sb)->s_def_hash_version;
1144 if (hinfo.hash_version <= DX_HASH_TEA)
1145 hinfo.hash_version +=
1146 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1147 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1148 if (ext4_has_inline_data(dir)) {
1149 int has_inline_data = 1;
1150 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1151 &hinfo, start_hash,
1152 start_minor_hash,
1153 &has_inline_data);
1154 if (has_inline_data) {
1155 *next_hash = ~0;
1156 return count;
1157 }
1158 }
1159 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1160 start_hash, start_minor_hash);
1161 *next_hash = ~0;
1162 return count;
1163 }
1164 hinfo.hash = start_hash;
1165 hinfo.minor_hash = 0;
1166 frame = dx_probe(NULL, dir, &hinfo, frames);
1167 if (IS_ERR(frame))
1168 return PTR_ERR(frame);
1169
1170 /* Add '.' and '..' from the htree header */
1171 if (!start_hash && !start_minor_hash) {
1172 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1173 tmp_str.name = de->name;
1174 tmp_str.len = de->name_len;
1175 err = ext4_htree_store_dirent(dir_file, 0, 0,
1176 de, &tmp_str);
1177 if (err != 0)
1178 goto errout;
1179 count++;
1180 }
1181 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1182 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1183 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1184 tmp_str.name = de->name;
1185 tmp_str.len = de->name_len;
1186 err = ext4_htree_store_dirent(dir_file, 2, 0,
1187 de, &tmp_str);
1188 if (err != 0)
1189 goto errout;
1190 count++;
1191 }
1192
1193 while (1) {
1194 if (fatal_signal_pending(current)) {
1195 err = -ERESTARTSYS;
1196 goto errout;
1197 }
1198 cond_resched();
1199 block = dx_get_block(frame->at);
1200 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1201 start_hash, start_minor_hash);
1202 if (ret < 0) {
1203 err = ret;
1204 goto errout;
1205 }
1206 count += ret;
1207 hashval = ~0;
1208 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1209 frame, frames, &hashval);
1210 *next_hash = hashval;
1211 if (ret < 0) {
1212 err = ret;
1213 goto errout;
1214 }
1215 /*
1216 * Stop if: (a) there are no more entries, or
1217 * (b) we have inserted at least one entry and the
1218 * next hash value is not a continuation
1219 */
1220 if ((ret == 0) ||
1221 (count && ((hashval & 1) == 0)))
1222 break;
1223 }
1224 dx_release(frames);
1225 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1226 "next hash: %x\n", count, *next_hash));
1227 return count;
1228errout:
1229 dx_release(frames);
1230 return (err);
1231}
1232
1233static inline int search_dirblock(struct buffer_head *bh,
1234 struct inode *dir,
1235 struct ext4_filename *fname,
1236 unsigned int offset,
1237 struct ext4_dir_entry_2 **res_dir)
1238{
1239 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1240 fname, offset, res_dir);
1241}
1242
1243/*
1244 * Directory block splitting, compacting
1245 */
1246
1247/*
1248 * Create map of hash values, offsets, and sizes, stored at end of block.
1249 * Returns number of entries mapped.
1250 */
1251static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1252 unsigned blocksize, struct dx_hash_info *hinfo,
1253 struct dx_map_entry *map_tail)
1254{
1255 int count = 0;
1256 char *base = (char *) de;
1257 struct dx_hash_info h = *hinfo;
1258
1259 while ((char *) de < base + blocksize) {
1260 if (de->name_len && de->inode) {
1261 if (ext4_hash_in_dirent(dir))
1262 h.hash = EXT4_DIRENT_HASH(de);
1263 else
1264 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1265 map_tail--;
1266 map_tail->hash = h.hash;
1267 map_tail->offs = ((char *) de - base)>>2;
1268 map_tail->size = le16_to_cpu(de->rec_len);
1269 count++;
1270 cond_resched();
1271 }
1272 /* XXX: do we need to check rec_len == 0 case? -Chris */
1273 de = ext4_next_entry(de, blocksize);
1274 }
1275 return count;
1276}
1277
1278/* Sort map by hash value */
1279static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1280{
1281 struct dx_map_entry *p, *q, *top = map + count - 1;
1282 int more;
1283 /* Combsort until bubble sort doesn't suck */
1284 while (count > 2) {
1285 count = count*10/13;
1286 if (count - 9 < 2) /* 9, 10 -> 11 */
1287 count = 11;
1288 for (p = top, q = p - count; q >= map; p--, q--)
1289 if (p->hash < q->hash)
1290 swap(*p, *q);
1291 }
1292 /* Garden variety bubble sort */
1293 do {
1294 more = 0;
1295 q = top;
1296 while (q-- > map) {
1297 if (q[1].hash >= q[0].hash)
1298 continue;
1299 swap(*(q+1), *q);
1300 more = 1;
1301 }
1302 } while(more);
1303}
1304
1305static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1306{
1307 struct dx_entry *entries = frame->entries;
1308 struct dx_entry *old = frame->at, *new = old + 1;
1309 int count = dx_get_count(entries);
1310
1311 ASSERT(count < dx_get_limit(entries));
1312 ASSERT(old < entries + count);
1313 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1314 dx_set_hash(new, hash);
1315 dx_set_block(new, block);
1316 dx_set_count(entries, count + 1);
1317}
1318
1319#ifdef CONFIG_UNICODE
1320/*
1321 * Test whether a case-insensitive directory entry matches the filename
1322 * being searched for. If quick is set, assume the name being looked up
1323 * is already in the casefolded form.
1324 *
1325 * Returns: 0 if the directory entry matches, more than 0 if it
1326 * doesn't match or less than zero on error.
1327 */
1328static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1329 u8 *de_name, size_t de_name_len, bool quick)
1330{
1331 const struct super_block *sb = parent->i_sb;
1332 const struct unicode_map *um = sb->s_encoding;
1333 struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1334 struct qstr entry = QSTR_INIT(de_name, de_name_len);
1335 int ret;
1336
1337 if (IS_ENCRYPTED(parent)) {
1338 const struct fscrypt_str encrypted_name =
1339 FSTR_INIT(de_name, de_name_len);
1340
1341 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1342 if (!decrypted_name.name)
1343 return -ENOMEM;
1344 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1345 &decrypted_name);
1346 if (ret < 0)
1347 goto out;
1348 entry.name = decrypted_name.name;
1349 entry.len = decrypted_name.len;
1350 }
1351
1352 if (quick)
1353 ret = utf8_strncasecmp_folded(um, name, &entry);
1354 else
1355 ret = utf8_strncasecmp(um, name, &entry);
1356 if (ret < 0) {
1357 /* Handle invalid character sequence as either an error
1358 * or as an opaque byte sequence.
1359 */
1360 if (sb_has_strict_encoding(sb))
1361 ret = -EINVAL;
1362 else if (name->len != entry.len)
1363 ret = 1;
1364 else
1365 ret = !!memcmp(name->name, entry.name, entry.len);
1366 }
1367out:
1368 kfree(decrypted_name.name);
1369 return ret;
1370}
1371
1372int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1373 struct ext4_filename *name)
1374{
1375 struct fscrypt_str *cf_name = &name->cf_name;
1376 struct dx_hash_info *hinfo = &name->hinfo;
1377 int len;
1378
1379 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding ||
1380 (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1381 cf_name->name = NULL;
1382 return 0;
1383 }
1384
1385 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1386 if (!cf_name->name)
1387 return -ENOMEM;
1388
1389 len = utf8_casefold(dir->i_sb->s_encoding,
1390 iname, cf_name->name,
1391 EXT4_NAME_LEN);
1392 if (len <= 0) {
1393 kfree(cf_name->name);
1394 cf_name->name = NULL;
1395 }
1396 cf_name->len = (unsigned) len;
1397 if (!IS_ENCRYPTED(dir))
1398 return 0;
1399
1400 hinfo->hash_version = DX_HASH_SIPHASH;
1401 hinfo->seed = NULL;
1402 if (cf_name->name)
1403 ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1404 else
1405 ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1406 return 0;
1407}
1408#endif
1409
1410/*
1411 * Test whether a directory entry matches the filename being searched for.
1412 *
1413 * Return: %true if the directory entry matches, otherwise %false.
1414 */
1415static bool ext4_match(struct inode *parent,
1416 const struct ext4_filename *fname,
1417 struct ext4_dir_entry_2 *de)
1418{
1419 struct fscrypt_name f;
1420
1421 if (!de->inode)
1422 return false;
1423
1424 f.usr_fname = fname->usr_fname;
1425 f.disk_name = fname->disk_name;
1426#ifdef CONFIG_FS_ENCRYPTION
1427 f.crypto_buf = fname->crypto_buf;
1428#endif
1429
1430#ifdef CONFIG_UNICODE
1431 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent) &&
1432 (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1433 if (fname->cf_name.name) {
1434 struct qstr cf = {.name = fname->cf_name.name,
1435 .len = fname->cf_name.len};
1436 if (IS_ENCRYPTED(parent)) {
1437 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1438 fname->hinfo.minor_hash !=
1439 EXT4_DIRENT_MINOR_HASH(de)) {
1440
1441 return 0;
1442 }
1443 }
1444 return !ext4_ci_compare(parent, &cf, de->name,
1445 de->name_len, true);
1446 }
1447 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1448 de->name_len, false);
1449 }
1450#endif
1451
1452 return fscrypt_match_name(&f, de->name, de->name_len);
1453}
1454
1455/*
1456 * Returns 0 if not found, -1 on failure, and 1 on success
1457 */
1458int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1459 struct inode *dir, struct ext4_filename *fname,
1460 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1461{
1462 struct ext4_dir_entry_2 * de;
1463 char * dlimit;
1464 int de_len;
1465
1466 de = (struct ext4_dir_entry_2 *)search_buf;
1467 dlimit = search_buf + buf_size;
1468 while ((char *) de < dlimit) {
1469 /* this code is executed quadratically often */
1470 /* do minimal checking `by hand' */
1471 if ((char *) de + de->name_len <= dlimit &&
1472 ext4_match(dir, fname, de)) {
1473 /* found a match - just to be sure, do
1474 * a full check */
1475 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1476 buf_size, offset))
1477 return -1;
1478 *res_dir = de;
1479 return 1;
1480 }
1481 /* prevent looping on a bad block */
1482 de_len = ext4_rec_len_from_disk(de->rec_len,
1483 dir->i_sb->s_blocksize);
1484 if (de_len <= 0)
1485 return -1;
1486 offset += de_len;
1487 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1488 }
1489 return 0;
1490}
1491
1492static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1493 struct ext4_dir_entry *de)
1494{
1495 struct super_block *sb = dir->i_sb;
1496
1497 if (!is_dx(dir))
1498 return 0;
1499 if (block == 0)
1500 return 1;
1501 if (de->inode == 0 &&
1502 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1503 sb->s_blocksize)
1504 return 1;
1505 return 0;
1506}
1507
1508/*
1509 * __ext4_find_entry()
1510 *
1511 * finds an entry in the specified directory with the wanted name. It
1512 * returns the cache buffer in which the entry was found, and the entry
1513 * itself (as a parameter - res_dir). It does NOT read the inode of the
1514 * entry - you'll have to do that yourself if you want to.
1515 *
1516 * The returned buffer_head has ->b_count elevated. The caller is expected
1517 * to brelse() it when appropriate.
1518 */
1519static struct buffer_head *__ext4_find_entry(struct inode *dir,
1520 struct ext4_filename *fname,
1521 struct ext4_dir_entry_2 **res_dir,
1522 int *inlined)
1523{
1524 struct super_block *sb;
1525 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1526 struct buffer_head *bh, *ret = NULL;
1527 ext4_lblk_t start, block;
1528 const u8 *name = fname->usr_fname->name;
1529 size_t ra_max = 0; /* Number of bh's in the readahead
1530 buffer, bh_use[] */
1531 size_t ra_ptr = 0; /* Current index into readahead
1532 buffer */
1533 ext4_lblk_t nblocks;
1534 int i, namelen, retval;
1535
1536 *res_dir = NULL;
1537 sb = dir->i_sb;
1538 namelen = fname->usr_fname->len;
1539 if (namelen > EXT4_NAME_LEN)
1540 return NULL;
1541
1542 if (ext4_has_inline_data(dir)) {
1543 int has_inline_data = 1;
1544 ret = ext4_find_inline_entry(dir, fname, res_dir,
1545 &has_inline_data);
1546 if (has_inline_data) {
1547 if (inlined)
1548 *inlined = 1;
1549 goto cleanup_and_exit;
1550 }
1551 }
1552
1553 if ((namelen <= 2) && (name[0] == '.') &&
1554 (name[1] == '.' || name[1] == '\0')) {
1555 /*
1556 * "." or ".." will only be in the first block
1557 * NFS may look up ".."; "." should be handled by the VFS
1558 */
1559 block = start = 0;
1560 nblocks = 1;
1561 goto restart;
1562 }
1563 if (is_dx(dir)) {
1564 ret = ext4_dx_find_entry(dir, fname, res_dir);
1565 /*
1566 * On success, or if the error was file not found,
1567 * return. Otherwise, fall back to doing a search the
1568 * old fashioned way.
1569 */
1570 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1571 goto cleanup_and_exit;
1572 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1573 "falling back\n"));
1574 ret = NULL;
1575 }
1576 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1577 if (!nblocks) {
1578 ret = NULL;
1579 goto cleanup_and_exit;
1580 }
1581 start = EXT4_I(dir)->i_dir_start_lookup;
1582 if (start >= nblocks)
1583 start = 0;
1584 block = start;
1585restart:
1586 do {
1587 /*
1588 * We deal with the read-ahead logic here.
1589 */
1590 cond_resched();
1591 if (ra_ptr >= ra_max) {
1592 /* Refill the readahead buffer */
1593 ra_ptr = 0;
1594 if (block < start)
1595 ra_max = start - block;
1596 else
1597 ra_max = nblocks - block;
1598 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1599 retval = ext4_bread_batch(dir, block, ra_max,
1600 false /* wait */, bh_use);
1601 if (retval) {
1602 ret = ERR_PTR(retval);
1603 ra_max = 0;
1604 goto cleanup_and_exit;
1605 }
1606 }
1607 if ((bh = bh_use[ra_ptr++]) == NULL)
1608 goto next;
1609 wait_on_buffer(bh);
1610 if (!buffer_uptodate(bh)) {
1611 EXT4_ERROR_INODE_ERR(dir, EIO,
1612 "reading directory lblock %lu",
1613 (unsigned long) block);
1614 brelse(bh);
1615 ret = ERR_PTR(-EIO);
1616 goto cleanup_and_exit;
1617 }
1618 if (!buffer_verified(bh) &&
1619 !is_dx_internal_node(dir, block,
1620 (struct ext4_dir_entry *)bh->b_data) &&
1621 !ext4_dirblock_csum_verify(dir, bh)) {
1622 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1623 "checksumming directory "
1624 "block %lu", (unsigned long)block);
1625 brelse(bh);
1626 ret = ERR_PTR(-EFSBADCRC);
1627 goto cleanup_and_exit;
1628 }
1629 set_buffer_verified(bh);
1630 i = search_dirblock(bh, dir, fname,
1631 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1632 if (i == 1) {
1633 EXT4_I(dir)->i_dir_start_lookup = block;
1634 ret = bh;
1635 goto cleanup_and_exit;
1636 } else {
1637 brelse(bh);
1638 if (i < 0)
1639 goto cleanup_and_exit;
1640 }
1641 next:
1642 if (++block >= nblocks)
1643 block = 0;
1644 } while (block != start);
1645
1646 /*
1647 * If the directory has grown while we were searching, then
1648 * search the last part of the directory before giving up.
1649 */
1650 block = nblocks;
1651 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1652 if (block < nblocks) {
1653 start = 0;
1654 goto restart;
1655 }
1656
1657cleanup_and_exit:
1658 /* Clean up the read-ahead blocks */
1659 for (; ra_ptr < ra_max; ra_ptr++)
1660 brelse(bh_use[ra_ptr]);
1661 return ret;
1662}
1663
1664static struct buffer_head *ext4_find_entry(struct inode *dir,
1665 const struct qstr *d_name,
1666 struct ext4_dir_entry_2 **res_dir,
1667 int *inlined)
1668{
1669 int err;
1670 struct ext4_filename fname;
1671 struct buffer_head *bh;
1672
1673 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1674 if (err == -ENOENT)
1675 return NULL;
1676 if (err)
1677 return ERR_PTR(err);
1678
1679 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1680
1681 ext4_fname_free_filename(&fname);
1682 return bh;
1683}
1684
1685static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1686 struct dentry *dentry,
1687 struct ext4_dir_entry_2 **res_dir)
1688{
1689 int err;
1690 struct ext4_filename fname;
1691 struct buffer_head *bh;
1692
1693 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1694 generic_set_encrypted_ci_d_ops(dentry);
1695 if (err == -ENOENT)
1696 return NULL;
1697 if (err)
1698 return ERR_PTR(err);
1699
1700 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1701
1702 ext4_fname_free_filename(&fname);
1703 return bh;
1704}
1705
1706static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1707 struct ext4_filename *fname,
1708 struct ext4_dir_entry_2 **res_dir)
1709{
1710 struct super_block * sb = dir->i_sb;
1711 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1712 struct buffer_head *bh;
1713 ext4_lblk_t block;
1714 int retval;
1715
1716#ifdef CONFIG_FS_ENCRYPTION
1717 *res_dir = NULL;
1718#endif
1719 frame = dx_probe(fname, dir, NULL, frames);
1720 if (IS_ERR(frame))
1721 return (struct buffer_head *) frame;
1722 do {
1723 block = dx_get_block(frame->at);
1724 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1725 if (IS_ERR(bh))
1726 goto errout;
1727
1728 retval = search_dirblock(bh, dir, fname,
1729 block << EXT4_BLOCK_SIZE_BITS(sb),
1730 res_dir);
1731 if (retval == 1)
1732 goto success;
1733 brelse(bh);
1734 if (retval == -1) {
1735 bh = ERR_PTR(ERR_BAD_DX_DIR);
1736 goto errout;
1737 }
1738
1739 /* Check to see if we should continue to search */
1740 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1741 frames, NULL);
1742 if (retval < 0) {
1743 ext4_warning_inode(dir,
1744 "error %d reading directory index block",
1745 retval);
1746 bh = ERR_PTR(retval);
1747 goto errout;
1748 }
1749 } while (retval == 1);
1750
1751 bh = NULL;
1752errout:
1753 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1754success:
1755 dx_release(frames);
1756 return bh;
1757}
1758
1759static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1760{
1761 struct inode *inode;
1762 struct ext4_dir_entry_2 *de;
1763 struct buffer_head *bh;
1764
1765 if (dentry->d_name.len > EXT4_NAME_LEN)
1766 return ERR_PTR(-ENAMETOOLONG);
1767
1768 bh = ext4_lookup_entry(dir, dentry, &de);
1769 if (IS_ERR(bh))
1770 return ERR_CAST(bh);
1771 inode = NULL;
1772 if (bh) {
1773 __u32 ino = le32_to_cpu(de->inode);
1774 brelse(bh);
1775 if (!ext4_valid_inum(dir->i_sb, ino)) {
1776 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1777 return ERR_PTR(-EFSCORRUPTED);
1778 }
1779 if (unlikely(ino == dir->i_ino)) {
1780 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1781 dentry);
1782 return ERR_PTR(-EFSCORRUPTED);
1783 }
1784 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1785 if (inode == ERR_PTR(-ESTALE)) {
1786 EXT4_ERROR_INODE(dir,
1787 "deleted inode referenced: %u",
1788 ino);
1789 return ERR_PTR(-EFSCORRUPTED);
1790 }
1791 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1792 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1793 !fscrypt_has_permitted_context(dir, inode)) {
1794 ext4_warning(inode->i_sb,
1795 "Inconsistent encryption contexts: %lu/%lu",
1796 dir->i_ino, inode->i_ino);
1797 iput(inode);
1798 return ERR_PTR(-EPERM);
1799 }
1800 }
1801
1802#ifdef CONFIG_UNICODE
1803 if (!inode && IS_CASEFOLDED(dir)) {
1804 /* Eventually we want to call d_add_ci(dentry, NULL)
1805 * for negative dentries in the encoding case as
1806 * well. For now, prevent the negative dentry
1807 * from being cached.
1808 */
1809 return NULL;
1810 }
1811#endif
1812 return d_splice_alias(inode, dentry);
1813}
1814
1815
1816struct dentry *ext4_get_parent(struct dentry *child)
1817{
1818 __u32 ino;
1819 struct ext4_dir_entry_2 * de;
1820 struct buffer_head *bh;
1821
1822 bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1823 if (IS_ERR(bh))
1824 return ERR_CAST(bh);
1825 if (!bh)
1826 return ERR_PTR(-ENOENT);
1827 ino = le32_to_cpu(de->inode);
1828 brelse(bh);
1829
1830 if (!ext4_valid_inum(child->d_sb, ino)) {
1831 EXT4_ERROR_INODE(d_inode(child),
1832 "bad parent inode number: %u", ino);
1833 return ERR_PTR(-EFSCORRUPTED);
1834 }
1835
1836 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1837}
1838
1839/*
1840 * Move count entries from end of map between two memory locations.
1841 * Returns pointer to last entry moved.
1842 */
1843static struct ext4_dir_entry_2 *
1844dx_move_dirents(struct inode *dir, char *from, char *to,
1845 struct dx_map_entry *map, int count,
1846 unsigned blocksize)
1847{
1848 unsigned rec_len = 0;
1849
1850 while (count--) {
1851 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1852 (from + (map->offs<<2));
1853 rec_len = ext4_dir_rec_len(de->name_len, dir);
1854
1855 memcpy (to, de, rec_len);
1856 ((struct ext4_dir_entry_2 *) to)->rec_len =
1857 ext4_rec_len_to_disk(rec_len, blocksize);
1858
1859 /* wipe dir_entry excluding the rec_len field */
1860 de->inode = 0;
1861 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1862 blocksize) -
1863 offsetof(struct ext4_dir_entry_2,
1864 name_len));
1865
1866 map++;
1867 to += rec_len;
1868 }
1869 return (struct ext4_dir_entry_2 *) (to - rec_len);
1870}
1871
1872/*
1873 * Compact each dir entry in the range to the minimal rec_len.
1874 * Returns pointer to last entry in range.
1875 */
1876static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1877 unsigned int blocksize)
1878{
1879 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1880 unsigned rec_len = 0;
1881
1882 prev = to = de;
1883 while ((char*)de < base + blocksize) {
1884 next = ext4_next_entry(de, blocksize);
1885 if (de->inode && de->name_len) {
1886 rec_len = ext4_dir_rec_len(de->name_len, dir);
1887 if (de > to)
1888 memmove(to, de, rec_len);
1889 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1890 prev = to;
1891 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1892 }
1893 de = next;
1894 }
1895 return prev;
1896}
1897
1898/*
1899 * Split a full leaf block to make room for a new dir entry.
1900 * Allocate a new block, and move entries so that they are approx. equally full.
1901 * Returns pointer to de in block into which the new entry will be inserted.
1902 */
1903static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1904 struct buffer_head **bh,struct dx_frame *frame,
1905 struct dx_hash_info *hinfo)
1906{
1907 unsigned blocksize = dir->i_sb->s_blocksize;
1908 unsigned count, continued;
1909 struct buffer_head *bh2;
1910 ext4_lblk_t newblock;
1911 u32 hash2;
1912 struct dx_map_entry *map;
1913 char *data1 = (*bh)->b_data, *data2;
1914 unsigned split, move, size;
1915 struct ext4_dir_entry_2 *de = NULL, *de2;
1916 int csum_size = 0;
1917 int err = 0, i;
1918
1919 if (ext4_has_metadata_csum(dir->i_sb))
1920 csum_size = sizeof(struct ext4_dir_entry_tail);
1921
1922 bh2 = ext4_append(handle, dir, &newblock);
1923 if (IS_ERR(bh2)) {
1924 brelse(*bh);
1925 *bh = NULL;
1926 return (struct ext4_dir_entry_2 *) bh2;
1927 }
1928
1929 BUFFER_TRACE(*bh, "get_write_access");
1930 err = ext4_journal_get_write_access(handle, *bh);
1931 if (err)
1932 goto journal_error;
1933
1934 BUFFER_TRACE(frame->bh, "get_write_access");
1935 err = ext4_journal_get_write_access(handle, frame->bh);
1936 if (err)
1937 goto journal_error;
1938
1939 data2 = bh2->b_data;
1940
1941 /* create map in the end of data2 block */
1942 map = (struct dx_map_entry *) (data2 + blocksize);
1943 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1944 blocksize, hinfo, map);
1945 map -= count;
1946 dx_sort_map(map, count);
1947 /* Ensure that neither split block is over half full */
1948 size = 0;
1949 move = 0;
1950 for (i = count-1; i >= 0; i--) {
1951 /* is more than half of this entry in 2nd half of the block? */
1952 if (size + map[i].size/2 > blocksize/2)
1953 break;
1954 size += map[i].size;
1955 move++;
1956 }
1957 /*
1958 * map index at which we will split
1959 *
1960 * If the sum of active entries didn't exceed half the block size, just
1961 * split it in half by count; each resulting block will have at least
1962 * half the space free.
1963 */
1964 if (i > 0)
1965 split = count - move;
1966 else
1967 split = count/2;
1968
1969 hash2 = map[split].hash;
1970 continued = hash2 == map[split - 1].hash;
1971 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1972 (unsigned long)dx_get_block(frame->at),
1973 hash2, split, count-split));
1974
1975 /* Fancy dance to stay within two buffers */
1976 de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
1977 blocksize);
1978 de = dx_pack_dirents(dir, data1, blocksize);
1979 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1980 (char *) de,
1981 blocksize);
1982 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1983 (char *) de2,
1984 blocksize);
1985 if (csum_size) {
1986 ext4_initialize_dirent_tail(*bh, blocksize);
1987 ext4_initialize_dirent_tail(bh2, blocksize);
1988 }
1989
1990 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1991 blocksize, 1));
1992 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1993 blocksize, 1));
1994
1995 /* Which block gets the new entry? */
1996 if (hinfo->hash >= hash2) {
1997 swap(*bh, bh2);
1998 de = de2;
1999 }
2000 dx_insert_block(frame, hash2 + continued, newblock);
2001 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2002 if (err)
2003 goto journal_error;
2004 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2005 if (err)
2006 goto journal_error;
2007 brelse(bh2);
2008 dxtrace(dx_show_index("frame", frame->entries));
2009 return de;
2010
2011journal_error:
2012 brelse(*bh);
2013 brelse(bh2);
2014 *bh = NULL;
2015 ext4_std_error(dir->i_sb, err);
2016 return ERR_PTR(err);
2017}
2018
2019int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2020 struct buffer_head *bh,
2021 void *buf, int buf_size,
2022 struct ext4_filename *fname,
2023 struct ext4_dir_entry_2 **dest_de)
2024{
2025 struct ext4_dir_entry_2 *de;
2026 unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2027 int nlen, rlen;
2028 unsigned int offset = 0;
2029 char *top;
2030
2031 de = (struct ext4_dir_entry_2 *)buf;
2032 top = buf + buf_size - reclen;
2033 while ((char *) de <= top) {
2034 if (ext4_check_dir_entry(dir, NULL, de, bh,
2035 buf, buf_size, offset))
2036 return -EFSCORRUPTED;
2037 if (ext4_match(dir, fname, de))
2038 return -EEXIST;
2039 nlen = ext4_dir_rec_len(de->name_len, dir);
2040 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2041 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2042 break;
2043 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2044 offset += rlen;
2045 }
2046 if ((char *) de > top)
2047 return -ENOSPC;
2048
2049 *dest_de = de;
2050 return 0;
2051}
2052
2053void ext4_insert_dentry(struct inode *dir,
2054 struct inode *inode,
2055 struct ext4_dir_entry_2 *de,
2056 int buf_size,
2057 struct ext4_filename *fname)
2058{
2059
2060 int nlen, rlen;
2061
2062 nlen = ext4_dir_rec_len(de->name_len, dir);
2063 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2064 if (de->inode) {
2065 struct ext4_dir_entry_2 *de1 =
2066 (struct ext4_dir_entry_2 *)((char *)de + nlen);
2067 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2068 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2069 de = de1;
2070 }
2071 de->file_type = EXT4_FT_UNKNOWN;
2072 de->inode = cpu_to_le32(inode->i_ino);
2073 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2074 de->name_len = fname_len(fname);
2075 memcpy(de->name, fname_name(fname), fname_len(fname));
2076 if (ext4_hash_in_dirent(dir)) {
2077 struct dx_hash_info *hinfo = &fname->hinfo;
2078
2079 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2080 EXT4_DIRENT_HASHES(de)->minor_hash =
2081 cpu_to_le32(hinfo->minor_hash);
2082 }
2083}
2084
2085/*
2086 * Add a new entry into a directory (leaf) block. If de is non-NULL,
2087 * it points to a directory entry which is guaranteed to be large
2088 * enough for new directory entry. If de is NULL, then
2089 * add_dirent_to_buf will attempt search the directory block for
2090 * space. It will return -ENOSPC if no space is available, and -EIO
2091 * and -EEXIST if directory entry already exists.
2092 */
2093static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2094 struct inode *dir,
2095 struct inode *inode, struct ext4_dir_entry_2 *de,
2096 struct buffer_head *bh)
2097{
2098 unsigned int blocksize = dir->i_sb->s_blocksize;
2099 int csum_size = 0;
2100 int err, err2;
2101
2102 if (ext4_has_metadata_csum(inode->i_sb))
2103 csum_size = sizeof(struct ext4_dir_entry_tail);
2104
2105 if (!de) {
2106 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2107 blocksize - csum_size, fname, &de);
2108 if (err)
2109 return err;
2110 }
2111 BUFFER_TRACE(bh, "get_write_access");
2112 err = ext4_journal_get_write_access(handle, bh);
2113 if (err) {
2114 ext4_std_error(dir->i_sb, err);
2115 return err;
2116 }
2117
2118 /* By now the buffer is marked for journaling */
2119 ext4_insert_dentry(dir, inode, de, blocksize, fname);
2120
2121 /*
2122 * XXX shouldn't update any times until successful
2123 * completion of syscall, but too many callers depend
2124 * on this.
2125 *
2126 * XXX similarly, too many callers depend on
2127 * ext4_new_inode() setting the times, but error
2128 * recovery deletes the inode, so the worst that can
2129 * happen is that the times are slightly out of date
2130 * and/or different from the directory change time.
2131 */
2132 dir->i_mtime = dir->i_ctime = current_time(dir);
2133 ext4_update_dx_flag(dir);
2134 inode_inc_iversion(dir);
2135 err2 = ext4_mark_inode_dirty(handle, dir);
2136 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2137 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2138 if (err)
2139 ext4_std_error(dir->i_sb, err);
2140 return err ? err : err2;
2141}
2142
2143/*
2144 * This converts a one block unindexed directory to a 3 block indexed
2145 * directory, and adds the dentry to the indexed directory.
2146 */
2147static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2148 struct inode *dir,
2149 struct inode *inode, struct buffer_head *bh)
2150{
2151 struct buffer_head *bh2;
2152 struct dx_root *root;
2153 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2154 struct dx_entry *entries;
2155 struct ext4_dir_entry_2 *de, *de2;
2156 char *data2, *top;
2157 unsigned len;
2158 int retval;
2159 unsigned blocksize;
2160 ext4_lblk_t block;
2161 struct fake_dirent *fde;
2162 int csum_size = 0;
2163
2164 if (ext4_has_metadata_csum(inode->i_sb))
2165 csum_size = sizeof(struct ext4_dir_entry_tail);
2166
2167 blocksize = dir->i_sb->s_blocksize;
2168 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2169 BUFFER_TRACE(bh, "get_write_access");
2170 retval = ext4_journal_get_write_access(handle, bh);
2171 if (retval) {
2172 ext4_std_error(dir->i_sb, retval);
2173 brelse(bh);
2174 return retval;
2175 }
2176 root = (struct dx_root *) bh->b_data;
2177
2178 /* The 0th block becomes the root, move the dirents out */
2179 fde = &root->dotdot;
2180 de = (struct ext4_dir_entry_2 *)((char *)fde +
2181 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2182 if ((char *) de >= (((char *) root) + blocksize)) {
2183 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2184 brelse(bh);
2185 return -EFSCORRUPTED;
2186 }
2187 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2188
2189 /* Allocate new block for the 0th block's dirents */
2190 bh2 = ext4_append(handle, dir, &block);
2191 if (IS_ERR(bh2)) {
2192 brelse(bh);
2193 return PTR_ERR(bh2);
2194 }
2195 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2196 data2 = bh2->b_data;
2197
2198 memcpy(data2, de, len);
2199 memset(de, 0, len); /* wipe old data */
2200 de = (struct ext4_dir_entry_2 *) data2;
2201 top = data2 + len;
2202 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2203 de = de2;
2204 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2205 (char *) de, blocksize);
2206
2207 if (csum_size)
2208 ext4_initialize_dirent_tail(bh2, blocksize);
2209
2210 /* Initialize the root; the dot dirents already exist */
2211 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2212 de->rec_len = ext4_rec_len_to_disk(
2213 blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2214 memset (&root->info, 0, sizeof(root->info));
2215 root->info.info_length = sizeof(root->info);
2216 if (ext4_hash_in_dirent(dir))
2217 root->info.hash_version = DX_HASH_SIPHASH;
2218 else
2219 root->info.hash_version =
2220 EXT4_SB(dir->i_sb)->s_def_hash_version;
2221
2222 entries = root->entries;
2223 dx_set_block(entries, 1);
2224 dx_set_count(entries, 1);
2225 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2226
2227 /* Initialize as for dx_probe */
2228 fname->hinfo.hash_version = root->info.hash_version;
2229 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2230 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2231 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2232
2233 /* casefolded encrypted hashes are computed on fname setup */
2234 if (!ext4_hash_in_dirent(dir))
2235 ext4fs_dirhash(dir, fname_name(fname),
2236 fname_len(fname), &fname->hinfo);
2237
2238 memset(frames, 0, sizeof(frames));
2239 frame = frames;
2240 frame->entries = entries;
2241 frame->at = entries;
2242 frame->bh = bh;
2243
2244 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2245 if (retval)
2246 goto out_frames;
2247 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2248 if (retval)
2249 goto out_frames;
2250
2251 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2252 if (IS_ERR(de)) {
2253 retval = PTR_ERR(de);
2254 goto out_frames;
2255 }
2256
2257 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2258out_frames:
2259 /*
2260 * Even if the block split failed, we have to properly write
2261 * out all the changes we did so far. Otherwise we can end up
2262 * with corrupted filesystem.
2263 */
2264 if (retval)
2265 ext4_mark_inode_dirty(handle, dir);
2266 dx_release(frames);
2267 brelse(bh2);
2268 return retval;
2269}
2270
2271/*
2272 * ext4_add_entry()
2273 *
2274 * adds a file entry to the specified directory, using the same
2275 * semantics as ext4_find_entry(). It returns NULL if it failed.
2276 *
2277 * NOTE!! The inode part of 'de' is left at 0 - which means you
2278 * may not sleep between calling this and putting something into
2279 * the entry, as someone else might have used it while you slept.
2280 */
2281static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2282 struct inode *inode)
2283{
2284 struct inode *dir = d_inode(dentry->d_parent);
2285 struct buffer_head *bh = NULL;
2286 struct ext4_dir_entry_2 *de;
2287 struct super_block *sb;
2288 struct ext4_filename fname;
2289 int retval;
2290 int dx_fallback=0;
2291 unsigned blocksize;
2292 ext4_lblk_t block, blocks;
2293 int csum_size = 0;
2294
2295 if (ext4_has_metadata_csum(inode->i_sb))
2296 csum_size = sizeof(struct ext4_dir_entry_tail);
2297
2298 sb = dir->i_sb;
2299 blocksize = sb->s_blocksize;
2300 if (!dentry->d_name.len)
2301 return -EINVAL;
2302
2303 if (fscrypt_is_nokey_name(dentry))
2304 return -ENOKEY;
2305
2306#ifdef CONFIG_UNICODE
2307 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2308 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2309 return -EINVAL;
2310#endif
2311
2312 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2313 if (retval)
2314 return retval;
2315
2316 if (ext4_has_inline_data(dir)) {
2317 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2318 if (retval < 0)
2319 goto out;
2320 if (retval == 1) {
2321 retval = 0;
2322 goto out;
2323 }
2324 }
2325
2326 if (is_dx(dir)) {
2327 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2328 if (!retval || (retval != ERR_BAD_DX_DIR))
2329 goto out;
2330 /* Can we just ignore htree data? */
2331 if (ext4_has_metadata_csum(sb)) {
2332 EXT4_ERROR_INODE(dir,
2333 "Directory has corrupted htree index.");
2334 retval = -EFSCORRUPTED;
2335 goto out;
2336 }
2337 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2338 dx_fallback++;
2339 retval = ext4_mark_inode_dirty(handle, dir);
2340 if (unlikely(retval))
2341 goto out;
2342 }
2343 blocks = dir->i_size >> sb->s_blocksize_bits;
2344 for (block = 0; block < blocks; block++) {
2345 bh = ext4_read_dirblock(dir, block, DIRENT);
2346 if (bh == NULL) {
2347 bh = ext4_bread(handle, dir, block,
2348 EXT4_GET_BLOCKS_CREATE);
2349 goto add_to_new_block;
2350 }
2351 if (IS_ERR(bh)) {
2352 retval = PTR_ERR(bh);
2353 bh = NULL;
2354 goto out;
2355 }
2356 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2357 NULL, bh);
2358 if (retval != -ENOSPC)
2359 goto out;
2360
2361 if (blocks == 1 && !dx_fallback &&
2362 ext4_has_feature_dir_index(sb)) {
2363 retval = make_indexed_dir(handle, &fname, dir,
2364 inode, bh);
2365 bh = NULL; /* make_indexed_dir releases bh */
2366 goto out;
2367 }
2368 brelse(bh);
2369 }
2370 bh = ext4_append(handle, dir, &block);
2371add_to_new_block:
2372 if (IS_ERR(bh)) {
2373 retval = PTR_ERR(bh);
2374 bh = NULL;
2375 goto out;
2376 }
2377 de = (struct ext4_dir_entry_2 *) bh->b_data;
2378 de->inode = 0;
2379 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2380
2381 if (csum_size)
2382 ext4_initialize_dirent_tail(bh, blocksize);
2383
2384 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2385out:
2386 ext4_fname_free_filename(&fname);
2387 brelse(bh);
2388 if (retval == 0)
2389 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2390 return retval;
2391}
2392
2393/*
2394 * Returns 0 for success, or a negative error value
2395 */
2396static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2397 struct inode *dir, struct inode *inode)
2398{
2399 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2400 struct dx_entry *entries, *at;
2401 struct buffer_head *bh;
2402 struct super_block *sb = dir->i_sb;
2403 struct ext4_dir_entry_2 *de;
2404 int restart;
2405 int err;
2406
2407again:
2408 restart = 0;
2409 frame = dx_probe(fname, dir, NULL, frames);
2410 if (IS_ERR(frame))
2411 return PTR_ERR(frame);
2412 entries = frame->entries;
2413 at = frame->at;
2414 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2415 if (IS_ERR(bh)) {
2416 err = PTR_ERR(bh);
2417 bh = NULL;
2418 goto cleanup;
2419 }
2420
2421 BUFFER_TRACE(bh, "get_write_access");
2422 err = ext4_journal_get_write_access(handle, bh);
2423 if (err)
2424 goto journal_error;
2425
2426 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2427 if (err != -ENOSPC)
2428 goto cleanup;
2429
2430 err = 0;
2431 /* Block full, should compress but for now just split */
2432 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2433 dx_get_count(entries), dx_get_limit(entries)));
2434 /* Need to split index? */
2435 if (dx_get_count(entries) == dx_get_limit(entries)) {
2436 ext4_lblk_t newblock;
2437 int levels = frame - frames + 1;
2438 unsigned int icount;
2439 int add_level = 1;
2440 struct dx_entry *entries2;
2441 struct dx_node *node2;
2442 struct buffer_head *bh2;
2443
2444 while (frame > frames) {
2445 if (dx_get_count((frame - 1)->entries) <
2446 dx_get_limit((frame - 1)->entries)) {
2447 add_level = 0;
2448 break;
2449 }
2450 frame--; /* split higher index block */
2451 at = frame->at;
2452 entries = frame->entries;
2453 restart = 1;
2454 }
2455 if (add_level && levels == ext4_dir_htree_level(sb)) {
2456 ext4_warning(sb, "Directory (ino: %lu) index full, "
2457 "reach max htree level :%d",
2458 dir->i_ino, levels);
2459 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2460 ext4_warning(sb, "Large directory feature is "
2461 "not enabled on this "
2462 "filesystem");
2463 }
2464 err = -ENOSPC;
2465 goto cleanup;
2466 }
2467 icount = dx_get_count(entries);
2468 bh2 = ext4_append(handle, dir, &newblock);
2469 if (IS_ERR(bh2)) {
2470 err = PTR_ERR(bh2);
2471 goto cleanup;
2472 }
2473 node2 = (struct dx_node *)(bh2->b_data);
2474 entries2 = node2->entries;
2475 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2476 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2477 sb->s_blocksize);
2478 BUFFER_TRACE(frame->bh, "get_write_access");
2479 err = ext4_journal_get_write_access(handle, frame->bh);
2480 if (err)
2481 goto journal_error;
2482 if (!add_level) {
2483 unsigned icount1 = icount/2, icount2 = icount - icount1;
2484 unsigned hash2 = dx_get_hash(entries + icount1);
2485 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2486 icount1, icount2));
2487
2488 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2489 err = ext4_journal_get_write_access(handle,
2490 (frame - 1)->bh);
2491 if (err)
2492 goto journal_error;
2493
2494 memcpy((char *) entries2, (char *) (entries + icount1),
2495 icount2 * sizeof(struct dx_entry));
2496 dx_set_count(entries, icount1);
2497 dx_set_count(entries2, icount2);
2498 dx_set_limit(entries2, dx_node_limit(dir));
2499
2500 /* Which index block gets the new entry? */
2501 if (at - entries >= icount1) {
2502 frame->at = at - entries - icount1 + entries2;
2503 frame->entries = entries = entries2;
2504 swap(frame->bh, bh2);
2505 }
2506 dx_insert_block((frame - 1), hash2, newblock);
2507 dxtrace(dx_show_index("node", frame->entries));
2508 dxtrace(dx_show_index("node",
2509 ((struct dx_node *) bh2->b_data)->entries));
2510 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2511 if (err)
2512 goto journal_error;
2513 brelse (bh2);
2514 err = ext4_handle_dirty_dx_node(handle, dir,
2515 (frame - 1)->bh);
2516 if (err)
2517 goto journal_error;
2518 err = ext4_handle_dirty_dx_node(handle, dir,
2519 frame->bh);
2520 if (restart || err)
2521 goto journal_error;
2522 } else {
2523 struct dx_root *dxroot;
2524 memcpy((char *) entries2, (char *) entries,
2525 icount * sizeof(struct dx_entry));
2526 dx_set_limit(entries2, dx_node_limit(dir));
2527
2528 /* Set up root */
2529 dx_set_count(entries, 1);
2530 dx_set_block(entries + 0, newblock);
2531 dxroot = (struct dx_root *)frames[0].bh->b_data;
2532 dxroot->info.indirect_levels += 1;
2533 dxtrace(printk(KERN_DEBUG
2534 "Creating %d level index...\n",
2535 dxroot->info.indirect_levels));
2536 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2537 if (err)
2538 goto journal_error;
2539 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2540 brelse(bh2);
2541 restart = 1;
2542 goto journal_error;
2543 }
2544 }
2545 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2546 if (IS_ERR(de)) {
2547 err = PTR_ERR(de);
2548 goto cleanup;
2549 }
2550 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2551 goto cleanup;
2552
2553journal_error:
2554 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2555cleanup:
2556 brelse(bh);
2557 dx_release(frames);
2558 /* @restart is true means htree-path has been changed, we need to
2559 * repeat dx_probe() to find out valid htree-path
2560 */
2561 if (restart && err == 0)
2562 goto again;
2563 return err;
2564}
2565
2566/*
2567 * ext4_generic_delete_entry deletes a directory entry by merging it
2568 * with the previous entry
2569 */
2570int ext4_generic_delete_entry(struct inode *dir,
2571 struct ext4_dir_entry_2 *de_del,
2572 struct buffer_head *bh,
2573 void *entry_buf,
2574 int buf_size,
2575 int csum_size)
2576{
2577 struct ext4_dir_entry_2 *de, *pde;
2578 unsigned int blocksize = dir->i_sb->s_blocksize;
2579 int i;
2580
2581 i = 0;
2582 pde = NULL;
2583 de = (struct ext4_dir_entry_2 *)entry_buf;
2584 while (i < buf_size - csum_size) {
2585 if (ext4_check_dir_entry(dir, NULL, de, bh,
2586 entry_buf, buf_size, i))
2587 return -EFSCORRUPTED;
2588 if (de == de_del) {
2589 if (pde) {
2590 pde->rec_len = ext4_rec_len_to_disk(
2591 ext4_rec_len_from_disk(pde->rec_len,
2592 blocksize) +
2593 ext4_rec_len_from_disk(de->rec_len,
2594 blocksize),
2595 blocksize);
2596
2597 /* wipe entire dir_entry */
2598 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2599 blocksize));
2600 } else {
2601 /* wipe dir_entry excluding the rec_len field */
2602 de->inode = 0;
2603 memset(&de->name_len, 0,
2604 ext4_rec_len_from_disk(de->rec_len,
2605 blocksize) -
2606 offsetof(struct ext4_dir_entry_2,
2607 name_len));
2608 }
2609
2610 inode_inc_iversion(dir);
2611 return 0;
2612 }
2613 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2614 pde = de;
2615 de = ext4_next_entry(de, blocksize);
2616 }
2617 return -ENOENT;
2618}
2619
2620static int ext4_delete_entry(handle_t *handle,
2621 struct inode *dir,
2622 struct ext4_dir_entry_2 *de_del,
2623 struct buffer_head *bh)
2624{
2625 int err, csum_size = 0;
2626
2627 if (ext4_has_inline_data(dir)) {
2628 int has_inline_data = 1;
2629 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2630 &has_inline_data);
2631 if (has_inline_data)
2632 return err;
2633 }
2634
2635 if (ext4_has_metadata_csum(dir->i_sb))
2636 csum_size = sizeof(struct ext4_dir_entry_tail);
2637
2638 BUFFER_TRACE(bh, "get_write_access");
2639 err = ext4_journal_get_write_access(handle, bh);
2640 if (unlikely(err))
2641 goto out;
2642
2643 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2644 dir->i_sb->s_blocksize, csum_size);
2645 if (err)
2646 goto out;
2647
2648 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2649 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2650 if (unlikely(err))
2651 goto out;
2652
2653 return 0;
2654out:
2655 if (err != -ENOENT)
2656 ext4_std_error(dir->i_sb, err);
2657 return err;
2658}
2659
2660/*
2661 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2662 * since this indicates that nlinks count was previously 1 to avoid overflowing
2663 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2664 * that subdirectory link counts are not being maintained accurately.
2665 *
2666 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2667 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2668 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2669 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2670 */
2671static void ext4_inc_count(struct inode *inode)
2672{
2673 inc_nlink(inode);
2674 if (is_dx(inode) &&
2675 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2676 set_nlink(inode, 1);
2677}
2678
2679/*
2680 * If a directory had nlink == 1, then we should let it be 1. This indicates
2681 * directory has >EXT4_LINK_MAX subdirs.
2682 */
2683static void ext4_dec_count(struct inode *inode)
2684{
2685 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2686 drop_nlink(inode);
2687}
2688
2689
2690/*
2691 * Add non-directory inode to a directory. On success, the inode reference is
2692 * consumed by dentry is instantiation. This is also indicated by clearing of
2693 * *inodep pointer. On failure, the caller is responsible for dropping the
2694 * inode reference in the safe context.
2695 */
2696static int ext4_add_nondir(handle_t *handle,
2697 struct dentry *dentry, struct inode **inodep)
2698{
2699 struct inode *dir = d_inode(dentry->d_parent);
2700 struct inode *inode = *inodep;
2701 int err = ext4_add_entry(handle, dentry, inode);
2702 if (!err) {
2703 err = ext4_mark_inode_dirty(handle, inode);
2704 if (IS_DIRSYNC(dir))
2705 ext4_handle_sync(handle);
2706 d_instantiate_new(dentry, inode);
2707 *inodep = NULL;
2708 return err;
2709 }
2710 drop_nlink(inode);
2711 ext4_orphan_add(handle, inode);
2712 unlock_new_inode(inode);
2713 return err;
2714}
2715
2716/*
2717 * By the time this is called, we already have created
2718 * the directory cache entry for the new file, but it
2719 * is so far negative - it has no inode.
2720 *
2721 * If the create succeeds, we fill in the inode information
2722 * with d_instantiate().
2723 */
2724static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir,
2725 struct dentry *dentry, umode_t mode, bool excl)
2726{
2727 handle_t *handle;
2728 struct inode *inode;
2729 int err, credits, retries = 0;
2730
2731 err = dquot_initialize(dir);
2732 if (err)
2733 return err;
2734
2735 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2736 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2737retry:
2738 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2739 0, NULL, EXT4_HT_DIR, credits);
2740 handle = ext4_journal_current_handle();
2741 err = PTR_ERR(inode);
2742 if (!IS_ERR(inode)) {
2743 inode->i_op = &ext4_file_inode_operations;
2744 inode->i_fop = &ext4_file_operations;
2745 ext4_set_aops(inode);
2746 err = ext4_add_nondir(handle, dentry, &inode);
2747 if (!err)
2748 ext4_fc_track_create(handle, dentry);
2749 }
2750 if (handle)
2751 ext4_journal_stop(handle);
2752 if (!IS_ERR_OR_NULL(inode))
2753 iput(inode);
2754 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2755 goto retry;
2756 return err;
2757}
2758
2759static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2760 struct dentry *dentry, umode_t mode, dev_t rdev)
2761{
2762 handle_t *handle;
2763 struct inode *inode;
2764 int err, credits, retries = 0;
2765
2766 err = dquot_initialize(dir);
2767 if (err)
2768 return err;
2769
2770 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2771 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2772retry:
2773 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2774 0, NULL, EXT4_HT_DIR, credits);
2775 handle = ext4_journal_current_handle();
2776 err = PTR_ERR(inode);
2777 if (!IS_ERR(inode)) {
2778 init_special_inode(inode, inode->i_mode, rdev);
2779 inode->i_op = &ext4_special_inode_operations;
2780 err = ext4_add_nondir(handle, dentry, &inode);
2781 if (!err)
2782 ext4_fc_track_create(handle, dentry);
2783 }
2784 if (handle)
2785 ext4_journal_stop(handle);
2786 if (!IS_ERR_OR_NULL(inode))
2787 iput(inode);
2788 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2789 goto retry;
2790 return err;
2791}
2792
2793static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2794 struct dentry *dentry, umode_t mode)
2795{
2796 handle_t *handle;
2797 struct inode *inode;
2798 int err, retries = 0;
2799
2800 err = dquot_initialize(dir);
2801 if (err)
2802 return err;
2803
2804retry:
2805 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2806 NULL, 0, NULL,
2807 EXT4_HT_DIR,
2808 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2809 4 + EXT4_XATTR_TRANS_BLOCKS);
2810 handle = ext4_journal_current_handle();
2811 err = PTR_ERR(inode);
2812 if (!IS_ERR(inode)) {
2813 inode->i_op = &ext4_file_inode_operations;
2814 inode->i_fop = &ext4_file_operations;
2815 ext4_set_aops(inode);
2816 d_tmpfile(dentry, inode);
2817 err = ext4_orphan_add(handle, inode);
2818 if (err)
2819 goto err_unlock_inode;
2820 mark_inode_dirty(inode);
2821 unlock_new_inode(inode);
2822 }
2823 if (handle)
2824 ext4_journal_stop(handle);
2825 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2826 goto retry;
2827 return err;
2828err_unlock_inode:
2829 ext4_journal_stop(handle);
2830 unlock_new_inode(inode);
2831 return err;
2832}
2833
2834struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2835 struct ext4_dir_entry_2 *de,
2836 int blocksize, int csum_size,
2837 unsigned int parent_ino, int dotdot_real_len)
2838{
2839 de->inode = cpu_to_le32(inode->i_ino);
2840 de->name_len = 1;
2841 de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2842 blocksize);
2843 strcpy(de->name, ".");
2844 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2845
2846 de = ext4_next_entry(de, blocksize);
2847 de->inode = cpu_to_le32(parent_ino);
2848 de->name_len = 2;
2849 if (!dotdot_real_len)
2850 de->rec_len = ext4_rec_len_to_disk(blocksize -
2851 (csum_size + ext4_dir_rec_len(1, NULL)),
2852 blocksize);
2853 else
2854 de->rec_len = ext4_rec_len_to_disk(
2855 ext4_dir_rec_len(de->name_len, NULL),
2856 blocksize);
2857 strcpy(de->name, "..");
2858 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2859
2860 return ext4_next_entry(de, blocksize);
2861}
2862
2863int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2864 struct inode *inode)
2865{
2866 struct buffer_head *dir_block = NULL;
2867 struct ext4_dir_entry_2 *de;
2868 ext4_lblk_t block = 0;
2869 unsigned int blocksize = dir->i_sb->s_blocksize;
2870 int csum_size = 0;
2871 int err;
2872
2873 if (ext4_has_metadata_csum(dir->i_sb))
2874 csum_size = sizeof(struct ext4_dir_entry_tail);
2875
2876 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2877 err = ext4_try_create_inline_dir(handle, dir, inode);
2878 if (err < 0 && err != -ENOSPC)
2879 goto out;
2880 if (!err)
2881 goto out;
2882 }
2883
2884 inode->i_size = 0;
2885 dir_block = ext4_append(handle, inode, &block);
2886 if (IS_ERR(dir_block))
2887 return PTR_ERR(dir_block);
2888 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2889 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2890 set_nlink(inode, 2);
2891 if (csum_size)
2892 ext4_initialize_dirent_tail(dir_block, blocksize);
2893
2894 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2895 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2896 if (err)
2897 goto out;
2898 set_buffer_verified(dir_block);
2899out:
2900 brelse(dir_block);
2901 return err;
2902}
2903
2904static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2905 struct dentry *dentry, umode_t mode)
2906{
2907 handle_t *handle;
2908 struct inode *inode;
2909 int err, err2 = 0, credits, retries = 0;
2910
2911 if (EXT4_DIR_LINK_MAX(dir))
2912 return -EMLINK;
2913
2914 err = dquot_initialize(dir);
2915 if (err)
2916 return err;
2917
2918 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2919 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2920retry:
2921 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2922 &dentry->d_name,
2923 0, NULL, EXT4_HT_DIR, credits);
2924 handle = ext4_journal_current_handle();
2925 err = PTR_ERR(inode);
2926 if (IS_ERR(inode))
2927 goto out_stop;
2928
2929 inode->i_op = &ext4_dir_inode_operations;
2930 inode->i_fop = &ext4_dir_operations;
2931 err = ext4_init_new_dir(handle, dir, inode);
2932 if (err)
2933 goto out_clear_inode;
2934 err = ext4_mark_inode_dirty(handle, inode);
2935 if (!err)
2936 err = ext4_add_entry(handle, dentry, inode);
2937 if (err) {
2938out_clear_inode:
2939 clear_nlink(inode);
2940 ext4_orphan_add(handle, inode);
2941 unlock_new_inode(inode);
2942 err2 = ext4_mark_inode_dirty(handle, inode);
2943 if (unlikely(err2))
2944 err = err2;
2945 ext4_journal_stop(handle);
2946 iput(inode);
2947 goto out_retry;
2948 }
2949 ext4_inc_count(dir);
2950
2951 ext4_update_dx_flag(dir);
2952 err = ext4_mark_inode_dirty(handle, dir);
2953 if (err)
2954 goto out_clear_inode;
2955 d_instantiate_new(dentry, inode);
2956 ext4_fc_track_create(handle, dentry);
2957 if (IS_DIRSYNC(dir))
2958 ext4_handle_sync(handle);
2959
2960out_stop:
2961 if (handle)
2962 ext4_journal_stop(handle);
2963out_retry:
2964 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2965 goto retry;
2966 return err;
2967}
2968
2969/*
2970 * routine to check that the specified directory is empty (for rmdir)
2971 */
2972bool ext4_empty_dir(struct inode *inode)
2973{
2974 unsigned int offset;
2975 struct buffer_head *bh;
2976 struct ext4_dir_entry_2 *de;
2977 struct super_block *sb;
2978
2979 if (ext4_has_inline_data(inode)) {
2980 int has_inline_data = 1;
2981 int ret;
2982
2983 ret = empty_inline_dir(inode, &has_inline_data);
2984 if (has_inline_data)
2985 return ret;
2986 }
2987
2988 sb = inode->i_sb;
2989 if (inode->i_size < ext4_dir_rec_len(1, NULL) +
2990 ext4_dir_rec_len(2, NULL)) {
2991 EXT4_ERROR_INODE(inode, "invalid size");
2992 return true;
2993 }
2994 /* The first directory block must not be a hole,
2995 * so treat it as DIRENT_HTREE
2996 */
2997 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2998 if (IS_ERR(bh))
2999 return true;
3000
3001 de = (struct ext4_dir_entry_2 *) bh->b_data;
3002 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3003 0) ||
3004 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3005 ext4_warning_inode(inode, "directory missing '.'");
3006 brelse(bh);
3007 return true;
3008 }
3009 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3010 de = ext4_next_entry(de, sb->s_blocksize);
3011 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3012 offset) ||
3013 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3014 ext4_warning_inode(inode, "directory missing '..'");
3015 brelse(bh);
3016 return true;
3017 }
3018 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3019 while (offset < inode->i_size) {
3020 if (!(offset & (sb->s_blocksize - 1))) {
3021 unsigned int lblock;
3022 brelse(bh);
3023 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3024 bh = ext4_read_dirblock(inode, lblock, EITHER);
3025 if (bh == NULL) {
3026 offset += sb->s_blocksize;
3027 continue;
3028 }
3029 if (IS_ERR(bh))
3030 return true;
3031 }
3032 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3033 (offset & (sb->s_blocksize - 1)));
3034 if (ext4_check_dir_entry(inode, NULL, de, bh,
3035 bh->b_data, bh->b_size, offset)) {
3036 offset = (offset | (sb->s_blocksize - 1)) + 1;
3037 continue;
3038 }
3039 if (le32_to_cpu(de->inode)) {
3040 brelse(bh);
3041 return false;
3042 }
3043 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3044 }
3045 brelse(bh);
3046 return true;
3047}
3048
3049/*
3050 * ext4_orphan_add() links an unlinked or truncated inode into a list of
3051 * such inodes, starting at the superblock, in case we crash before the
3052 * file is closed/deleted, or in case the inode truncate spans multiple
3053 * transactions and the last transaction is not recovered after a crash.
3054 *
3055 * At filesystem recovery time, we walk this list deleting unlinked
3056 * inodes and truncating linked inodes in ext4_orphan_cleanup().
3057 *
3058 * Orphan list manipulation functions must be called under i_mutex unless
3059 * we are just creating the inode or deleting it.
3060 */
3061int ext4_orphan_add(handle_t *handle, struct inode *inode)
3062{
3063 struct super_block *sb = inode->i_sb;
3064 struct ext4_sb_info *sbi = EXT4_SB(sb);
3065 struct ext4_iloc iloc;
3066 int err = 0, rc;
3067 bool dirty = false;
3068
3069 if (!sbi->s_journal || is_bad_inode(inode))
3070 return 0;
3071
3072 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3073 !inode_is_locked(inode));
3074 /*
3075 * Exit early if inode already is on orphan list. This is a big speedup
3076 * since we don't have to contend on the global s_orphan_lock.
3077 */
3078 if (!list_empty(&EXT4_I(inode)->i_orphan))
3079 return 0;
3080
3081 /*
3082 * Orphan handling is only valid for files with data blocks
3083 * being truncated, or files being unlinked. Note that we either
3084 * hold i_mutex, or the inode can not be referenced from outside,
3085 * so i_nlink should not be bumped due to race
3086 */
3087 ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
3088 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
3089
3090 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3091 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3092 if (err)
3093 goto out;
3094
3095 err = ext4_reserve_inode_write(handle, inode, &iloc);
3096 if (err)
3097 goto out;
3098
3099 mutex_lock(&sbi->s_orphan_lock);
3100 /*
3101 * Due to previous errors inode may be already a part of on-disk
3102 * orphan list. If so skip on-disk list modification.
3103 */
3104 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
3105 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
3106 /* Insert this inode at the head of the on-disk orphan list */
3107 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
3108 lock_buffer(sbi->s_sbh);
3109 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
3110 ext4_superblock_csum_set(sb);
3111 unlock_buffer(sbi->s_sbh);
3112 dirty = true;
3113 }
3114 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
3115 mutex_unlock(&sbi->s_orphan_lock);
3116
3117 if (dirty) {
3118 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3119 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
3120 if (!err)
3121 err = rc;
3122 if (err) {
3123 /*
3124 * We have to remove inode from in-memory list if
3125 * addition to on disk orphan list failed. Stray orphan
3126 * list entries can cause panics at unmount time.
3127 */
3128 mutex_lock(&sbi->s_orphan_lock);
3129 list_del_init(&EXT4_I(inode)->i_orphan);
3130 mutex_unlock(&sbi->s_orphan_lock);
3131 }
3132 } else
3133 brelse(iloc.bh);
3134
3135 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3136 jbd_debug(4, "orphan inode %lu will point to %d\n",
3137 inode->i_ino, NEXT_ORPHAN(inode));
3138out:
3139 ext4_std_error(sb, err);
3140 return err;
3141}
3142
3143/*
3144 * ext4_orphan_del() removes an unlinked or truncated inode from the list
3145 * of such inodes stored on disk, because it is finally being cleaned up.
3146 */
3147int ext4_orphan_del(handle_t *handle, struct inode *inode)
3148{
3149 struct list_head *prev;
3150 struct ext4_inode_info *ei = EXT4_I(inode);
3151 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3152 __u32 ino_next;
3153 struct ext4_iloc iloc;
3154 int err = 0;
3155
3156 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3157 return 0;
3158
3159 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3160 !inode_is_locked(inode));
3161 /* Do this quick check before taking global s_orphan_lock. */
3162 if (list_empty(&ei->i_orphan))
3163 return 0;
3164
3165 if (handle) {
3166 /* Grab inode buffer early before taking global s_orphan_lock */
3167 err = ext4_reserve_inode_write(handle, inode, &iloc);
3168 }
3169
3170 mutex_lock(&sbi->s_orphan_lock);
3171 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3172
3173 prev = ei->i_orphan.prev;
3174 list_del_init(&ei->i_orphan);
3175
3176 /* If we're on an error path, we may not have a valid
3177 * transaction handle with which to update the orphan list on
3178 * disk, but we still need to remove the inode from the linked
3179 * list in memory. */
3180 if (!handle || err) {
3181 mutex_unlock(&sbi->s_orphan_lock);
3182 goto out_err;
3183 }
3184
3185 ino_next = NEXT_ORPHAN(inode);
3186 if (prev == &sbi->s_orphan) {
3187 jbd_debug(4, "superblock will point to %u\n", ino_next);
3188 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3189 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3190 if (err) {
3191 mutex_unlock(&sbi->s_orphan_lock);
3192 goto out_brelse;
3193 }
3194 lock_buffer(sbi->s_sbh);
3195 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3196 ext4_superblock_csum_set(inode->i_sb);
3197 unlock_buffer(sbi->s_sbh);
3198 mutex_unlock(&sbi->s_orphan_lock);
3199 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3200 } else {
3201 struct ext4_iloc iloc2;
3202 struct inode *i_prev =
3203 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3204
3205 jbd_debug(4, "orphan inode %lu will point to %u\n",
3206 i_prev->i_ino, ino_next);
3207 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3208 if (err) {
3209 mutex_unlock(&sbi->s_orphan_lock);
3210 goto out_brelse;
3211 }
3212 NEXT_ORPHAN(i_prev) = ino_next;
3213 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3214 mutex_unlock(&sbi->s_orphan_lock);
3215 }
3216 if (err)
3217 goto out_brelse;
3218 NEXT_ORPHAN(inode) = 0;
3219 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3220out_err:
3221 ext4_std_error(inode->i_sb, err);
3222 return err;
3223
3224out_brelse:
3225 brelse(iloc.bh);
3226 goto out_err;
3227}
3228
3229static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3230{
3231 int retval;
3232 struct inode *inode;
3233 struct buffer_head *bh;
3234 struct ext4_dir_entry_2 *de;
3235 handle_t *handle = NULL;
3236
3237 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3238 return -EIO;
3239
3240 /* Initialize quotas before so that eventual writes go in
3241 * separate transaction */
3242 retval = dquot_initialize(dir);
3243 if (retval)
3244 return retval;
3245 retval = dquot_initialize(d_inode(dentry));
3246 if (retval)
3247 return retval;
3248
3249 retval = -ENOENT;
3250 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3251 if (IS_ERR(bh))
3252 return PTR_ERR(bh);
3253 if (!bh)
3254 goto end_rmdir;
3255
3256 inode = d_inode(dentry);
3257
3258 retval = -EFSCORRUPTED;
3259 if (le32_to_cpu(de->inode) != inode->i_ino)
3260 goto end_rmdir;
3261
3262 retval = -ENOTEMPTY;
3263 if (!ext4_empty_dir(inode))
3264 goto end_rmdir;
3265
3266 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3267 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3268 if (IS_ERR(handle)) {
3269 retval = PTR_ERR(handle);
3270 handle = NULL;
3271 goto end_rmdir;
3272 }
3273
3274 if (IS_DIRSYNC(dir))
3275 ext4_handle_sync(handle);
3276
3277 retval = ext4_delete_entry(handle, dir, de, bh);
3278 if (retval)
3279 goto end_rmdir;
3280 if (!EXT4_DIR_LINK_EMPTY(inode))
3281 ext4_warning_inode(inode,
3282 "empty directory '%.*s' has too many links (%u)",
3283 dentry->d_name.len, dentry->d_name.name,
3284 inode->i_nlink);
3285 inode_inc_iversion(inode);
3286 clear_nlink(inode);
3287 /* There's no need to set i_disksize: the fact that i_nlink is
3288 * zero will ensure that the right thing happens during any
3289 * recovery. */
3290 inode->i_size = 0;
3291 ext4_orphan_add(handle, inode);
3292 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3293 retval = ext4_mark_inode_dirty(handle, inode);
3294 if (retval)
3295 goto end_rmdir;
3296 ext4_dec_count(dir);
3297 ext4_update_dx_flag(dir);
3298 ext4_fc_track_unlink(handle, dentry);
3299 retval = ext4_mark_inode_dirty(handle, dir);
3300
3301#ifdef CONFIG_UNICODE
3302 /* VFS negative dentries are incompatible with Encoding and
3303 * Case-insensitiveness. Eventually we'll want avoid
3304 * invalidating the dentries here, alongside with returning the
3305 * negative dentries at ext4_lookup(), when it is better
3306 * supported by the VFS for the CI case.
3307 */
3308 if (IS_CASEFOLDED(dir))
3309 d_invalidate(dentry);
3310#endif
3311
3312end_rmdir:
3313 brelse(bh);
3314 if (handle)
3315 ext4_journal_stop(handle);
3316 return retval;
3317}
3318
3319int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3320 struct inode *inode)
3321{
3322 int retval = -ENOENT;
3323 struct buffer_head *bh;
3324 struct ext4_dir_entry_2 *de;
3325 int skip_remove_dentry = 0;
3326
3327 bh = ext4_find_entry(dir, d_name, &de, NULL);
3328 if (IS_ERR(bh))
3329 return PTR_ERR(bh);
3330
3331 if (!bh)
3332 return -ENOENT;
3333
3334 if (le32_to_cpu(de->inode) != inode->i_ino) {
3335 /*
3336 * It's okay if we find dont find dentry which matches
3337 * the inode. That's because it might have gotten
3338 * renamed to a different inode number
3339 */
3340 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3341 skip_remove_dentry = 1;
3342 else
3343 goto out;
3344 }
3345
3346 if (IS_DIRSYNC(dir))
3347 ext4_handle_sync(handle);
3348
3349 if (!skip_remove_dentry) {
3350 retval = ext4_delete_entry(handle, dir, de, bh);
3351 if (retval)
3352 goto out;
3353 dir->i_ctime = dir->i_mtime = current_time(dir);
3354 ext4_update_dx_flag(dir);
3355 retval = ext4_mark_inode_dirty(handle, dir);
3356 if (retval)
3357 goto out;
3358 } else {
3359 retval = 0;
3360 }
3361 if (inode->i_nlink == 0)
3362 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3363 d_name->len, d_name->name);
3364 else
3365 drop_nlink(inode);
3366 if (!inode->i_nlink)
3367 ext4_orphan_add(handle, inode);
3368 inode->i_ctime = current_time(inode);
3369 retval = ext4_mark_inode_dirty(handle, inode);
3370
3371out:
3372 brelse(bh);
3373 return retval;
3374}
3375
3376static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3377{
3378 handle_t *handle;
3379 int retval;
3380
3381 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3382 return -EIO;
3383
3384 trace_ext4_unlink_enter(dir, dentry);
3385 /*
3386 * Initialize quotas before so that eventual writes go
3387 * in separate transaction
3388 */
3389 retval = dquot_initialize(dir);
3390 if (retval)
3391 goto out_trace;
3392 retval = dquot_initialize(d_inode(dentry));
3393 if (retval)
3394 goto out_trace;
3395
3396 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3397 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3398 if (IS_ERR(handle)) {
3399 retval = PTR_ERR(handle);
3400 goto out_trace;
3401 }
3402
3403 retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3404 if (!retval)
3405 ext4_fc_track_unlink(handle, dentry);
3406#ifdef CONFIG_UNICODE
3407 /* VFS negative dentries are incompatible with Encoding and
3408 * Case-insensitiveness. Eventually we'll want avoid
3409 * invalidating the dentries here, alongside with returning the
3410 * negative dentries at ext4_lookup(), when it is better
3411 * supported by the VFS for the CI case.
3412 */
3413 if (IS_CASEFOLDED(dir))
3414 d_invalidate(dentry);
3415#endif
3416 if (handle)
3417 ext4_journal_stop(handle);
3418
3419out_trace:
3420 trace_ext4_unlink_exit(dentry, retval);
3421 return retval;
3422}
3423
3424static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3425 struct dentry *dentry, const char *symname)
3426{
3427 handle_t *handle;
3428 struct inode *inode;
3429 int err, len = strlen(symname);
3430 int credits;
3431 struct fscrypt_str disk_link;
3432
3433 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3434 return -EIO;
3435
3436 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3437 &disk_link);
3438 if (err)
3439 return err;
3440
3441 err = dquot_initialize(dir);
3442 if (err)
3443 return err;
3444
3445 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3446 /*
3447 * For non-fast symlinks, we just allocate inode and put it on
3448 * orphan list in the first transaction => we need bitmap,
3449 * group descriptor, sb, inode block, quota blocks, and
3450 * possibly selinux xattr blocks.
3451 */
3452 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3453 EXT4_XATTR_TRANS_BLOCKS;
3454 } else {
3455 /*
3456 * Fast symlink. We have to add entry to directory
3457 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3458 * allocate new inode (bitmap, group descriptor, inode block,
3459 * quota blocks, sb is already counted in previous macros).
3460 */
3461 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3462 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3463 }
3464
3465 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3466 &dentry->d_name, 0, NULL,
3467 EXT4_HT_DIR, credits);
3468 handle = ext4_journal_current_handle();
3469 if (IS_ERR(inode)) {
3470 if (handle)
3471 ext4_journal_stop(handle);
3472 return PTR_ERR(inode);
3473 }
3474
3475 if (IS_ENCRYPTED(inode)) {
3476 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3477 if (err)
3478 goto err_drop_inode;
3479 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3480 }
3481
3482 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3483 if (!IS_ENCRYPTED(inode))
3484 inode->i_op = &ext4_symlink_inode_operations;
3485 inode_nohighmem(inode);
3486 ext4_set_aops(inode);
3487 /*
3488 * We cannot call page_symlink() with transaction started
3489 * because it calls into ext4_write_begin() which can wait
3490 * for transaction commit if we are running out of space
3491 * and thus we deadlock. So we have to stop transaction now
3492 * and restart it when symlink contents is written.
3493 *
3494 * To keep fs consistent in case of crash, we have to put inode
3495 * to orphan list in the mean time.
3496 */
3497 drop_nlink(inode);
3498 err = ext4_orphan_add(handle, inode);
3499 if (handle)
3500 ext4_journal_stop(handle);
3501 handle = NULL;
3502 if (err)
3503 goto err_drop_inode;
3504 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3505 if (err)
3506 goto err_drop_inode;
3507 /*
3508 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3509 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3510 */
3511 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3512 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3513 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3514 if (IS_ERR(handle)) {
3515 err = PTR_ERR(handle);
3516 handle = NULL;
3517 goto err_drop_inode;
3518 }
3519 set_nlink(inode, 1);
3520 err = ext4_orphan_del(handle, inode);
3521 if (err)
3522 goto err_drop_inode;
3523 } else {
3524 /* clear the extent format for fast symlink */
3525 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3526 if (!IS_ENCRYPTED(inode)) {
3527 inode->i_op = &ext4_fast_symlink_inode_operations;
3528 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3529 }
3530 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3531 disk_link.len);
3532 inode->i_size = disk_link.len - 1;
3533 }
3534 EXT4_I(inode)->i_disksize = inode->i_size;
3535 err = ext4_add_nondir(handle, dentry, &inode);
3536 if (handle)
3537 ext4_journal_stop(handle);
3538 if (inode)
3539 iput(inode);
3540 goto out_free_encrypted_link;
3541
3542err_drop_inode:
3543 if (handle)
3544 ext4_journal_stop(handle);
3545 clear_nlink(inode);
3546 unlock_new_inode(inode);
3547 iput(inode);
3548out_free_encrypted_link:
3549 if (disk_link.name != (unsigned char *)symname)
3550 kfree(disk_link.name);
3551 return err;
3552}
3553
3554int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3555{
3556 handle_t *handle;
3557 int err, retries = 0;
3558retry:
3559 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3560 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3561 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3562 if (IS_ERR(handle))
3563 return PTR_ERR(handle);
3564
3565 if (IS_DIRSYNC(dir))
3566 ext4_handle_sync(handle);
3567
3568 inode->i_ctime = current_time(inode);
3569 ext4_inc_count(inode);
3570 ihold(inode);
3571
3572 err = ext4_add_entry(handle, dentry, inode);
3573 if (!err) {
3574 err = ext4_mark_inode_dirty(handle, inode);
3575 /* this can happen only for tmpfile being
3576 * linked the first time
3577 */
3578 if (inode->i_nlink == 1)
3579 ext4_orphan_del(handle, inode);
3580 d_instantiate(dentry, inode);
3581 ext4_fc_track_link(handle, dentry);
3582 } else {
3583 drop_nlink(inode);
3584 iput(inode);
3585 }
3586 ext4_journal_stop(handle);
3587 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3588 goto retry;
3589 return err;
3590}
3591
3592static int ext4_link(struct dentry *old_dentry,
3593 struct inode *dir, struct dentry *dentry)
3594{
3595 struct inode *inode = d_inode(old_dentry);
3596 int err;
3597
3598 if (inode->i_nlink >= EXT4_LINK_MAX)
3599 return -EMLINK;
3600
3601 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3602 if (err)
3603 return err;
3604
3605 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3606 (!projid_eq(EXT4_I(dir)->i_projid,
3607 EXT4_I(old_dentry->d_inode)->i_projid)))
3608 return -EXDEV;
3609
3610 err = dquot_initialize(dir);
3611 if (err)
3612 return err;
3613 return __ext4_link(dir, inode, dentry);
3614}
3615
3616/*
3617 * Try to find buffer head where contains the parent block.
3618 * It should be the inode block if it is inlined or the 1st block
3619 * if it is a normal dir.
3620 */
3621static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3622 struct inode *inode,
3623 int *retval,
3624 struct ext4_dir_entry_2 **parent_de,
3625 int *inlined)
3626{
3627 struct buffer_head *bh;
3628
3629 if (!ext4_has_inline_data(inode)) {
3630 /* The first directory block must not be a hole, so
3631 * treat it as DIRENT_HTREE
3632 */
3633 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3634 if (IS_ERR(bh)) {
3635 *retval = PTR_ERR(bh);
3636 return NULL;
3637 }
3638 *parent_de = ext4_next_entry(
3639 (struct ext4_dir_entry_2 *)bh->b_data,
3640 inode->i_sb->s_blocksize);
3641 return bh;
3642 }
3643
3644 *inlined = 1;
3645 return ext4_get_first_inline_block(inode, parent_de, retval);
3646}
3647
3648struct ext4_renament {
3649 struct inode *dir;
3650 struct dentry *dentry;
3651 struct inode *inode;
3652 bool is_dir;
3653 int dir_nlink_delta;
3654
3655 /* entry for "dentry" */
3656 struct buffer_head *bh;
3657 struct ext4_dir_entry_2 *de;
3658 int inlined;
3659
3660 /* entry for ".." in inode if it's a directory */
3661 struct buffer_head *dir_bh;
3662 struct ext4_dir_entry_2 *parent_de;
3663 int dir_inlined;
3664};
3665
3666static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3667{
3668 int retval;
3669
3670 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3671 &retval, &ent->parent_de,
3672 &ent->dir_inlined);
3673 if (!ent->dir_bh)
3674 return retval;
3675 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3676 return -EFSCORRUPTED;
3677 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3678 return ext4_journal_get_write_access(handle, ent->dir_bh);
3679}
3680
3681static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3682 unsigned dir_ino)
3683{
3684 int retval;
3685
3686 ent->parent_de->inode = cpu_to_le32(dir_ino);
3687 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3688 if (!ent->dir_inlined) {
3689 if (is_dx(ent->inode)) {
3690 retval = ext4_handle_dirty_dx_node(handle,
3691 ent->inode,
3692 ent->dir_bh);
3693 } else {
3694 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3695 ent->dir_bh);
3696 }
3697 } else {
3698 retval = ext4_mark_inode_dirty(handle, ent->inode);
3699 }
3700 if (retval) {
3701 ext4_std_error(ent->dir->i_sb, retval);
3702 return retval;
3703 }
3704 return 0;
3705}
3706
3707static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3708 unsigned ino, unsigned file_type)
3709{
3710 int retval, retval2;
3711
3712 BUFFER_TRACE(ent->bh, "get write access");
3713 retval = ext4_journal_get_write_access(handle, ent->bh);
3714 if (retval)
3715 return retval;
3716 ent->de->inode = cpu_to_le32(ino);
3717 if (ext4_has_feature_filetype(ent->dir->i_sb))
3718 ent->de->file_type = file_type;
3719 inode_inc_iversion(ent->dir);
3720 ent->dir->i_ctime = ent->dir->i_mtime =
3721 current_time(ent->dir);
3722 retval = ext4_mark_inode_dirty(handle, ent->dir);
3723 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3724 if (!ent->inlined) {
3725 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3726 if (unlikely(retval2)) {
3727 ext4_std_error(ent->dir->i_sb, retval2);
3728 return retval2;
3729 }
3730 }
3731 return retval;
3732}
3733
3734static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3735 unsigned ino, unsigned file_type)
3736{
3737 struct ext4_renament old = *ent;
3738 int retval = 0;
3739
3740 /*
3741 * old->de could have moved from under us during make indexed dir,
3742 * so the old->de may no longer valid and need to find it again
3743 * before reset old inode info.
3744 */
3745 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3746 if (IS_ERR(old.bh))
3747 retval = PTR_ERR(old.bh);
3748 if (!old.bh)
3749 retval = -ENOENT;
3750 if (retval) {
3751 ext4_std_error(old.dir->i_sb, retval);
3752 return;
3753 }
3754
3755 ext4_setent(handle, &old, ino, file_type);
3756 brelse(old.bh);
3757}
3758
3759static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3760 const struct qstr *d_name)
3761{
3762 int retval = -ENOENT;
3763 struct buffer_head *bh;
3764 struct ext4_dir_entry_2 *de;
3765
3766 bh = ext4_find_entry(dir, d_name, &de, NULL);
3767 if (IS_ERR(bh))
3768 return PTR_ERR(bh);
3769 if (bh) {
3770 retval = ext4_delete_entry(handle, dir, de, bh);
3771 brelse(bh);
3772 }
3773 return retval;
3774}
3775
3776static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3777 int force_reread)
3778{
3779 int retval;
3780 /*
3781 * ent->de could have moved from under us during htree split, so make
3782 * sure that we are deleting the right entry. We might also be pointing
3783 * to a stale entry in the unused part of ent->bh so just checking inum
3784 * and the name isn't enough.
3785 */
3786 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3787 ent->de->name_len != ent->dentry->d_name.len ||
3788 strncmp(ent->de->name, ent->dentry->d_name.name,
3789 ent->de->name_len) ||
3790 force_reread) {
3791 retval = ext4_find_delete_entry(handle, ent->dir,
3792 &ent->dentry->d_name);
3793 } else {
3794 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3795 if (retval == -ENOENT) {
3796 retval = ext4_find_delete_entry(handle, ent->dir,
3797 &ent->dentry->d_name);
3798 }
3799 }
3800
3801 if (retval) {
3802 ext4_warning_inode(ent->dir,
3803 "Deleting old file: nlink %d, error=%d",
3804 ent->dir->i_nlink, retval);
3805 }
3806}
3807
3808static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3809{
3810 if (ent->dir_nlink_delta) {
3811 if (ent->dir_nlink_delta == -1)
3812 ext4_dec_count(ent->dir);
3813 else
3814 ext4_inc_count(ent->dir);
3815 ext4_mark_inode_dirty(handle, ent->dir);
3816 }
3817}
3818
3819static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3820 struct ext4_renament *ent,
3821 int credits, handle_t **h)
3822{
3823 struct inode *wh;
3824 handle_t *handle;
3825 int retries = 0;
3826
3827 /*
3828 * for inode block, sb block, group summaries,
3829 * and inode bitmap
3830 */
3831 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3832 EXT4_XATTR_TRANS_BLOCKS + 4);
3833retry:
3834 wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3835 S_IFCHR | WHITEOUT_MODE,
3836 &ent->dentry->d_name, 0, NULL,
3837 EXT4_HT_DIR, credits);
3838
3839 handle = ext4_journal_current_handle();
3840 if (IS_ERR(wh)) {
3841 if (handle)
3842 ext4_journal_stop(handle);
3843 if (PTR_ERR(wh) == -ENOSPC &&
3844 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3845 goto retry;
3846 } else {
3847 *h = handle;
3848 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3849 wh->i_op = &ext4_special_inode_operations;
3850 }
3851 return wh;
3852}
3853
3854/*
3855 * Anybody can rename anything with this: the permission checks are left to the
3856 * higher-level routines.
3857 *
3858 * n.b. old_{dentry,inode) refers to the source dentry/inode
3859 * while new_{dentry,inode) refers to the destination dentry/inode
3860 * This comes from rename(const char *oldpath, const char *newpath)
3861 */
3862static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3863 struct dentry *old_dentry, struct inode *new_dir,
3864 struct dentry *new_dentry, unsigned int flags)
3865{
3866 handle_t *handle = NULL;
3867 struct ext4_renament old = {
3868 .dir = old_dir,
3869 .dentry = old_dentry,
3870 .inode = d_inode(old_dentry),
3871 };
3872 struct ext4_renament new = {
3873 .dir = new_dir,
3874 .dentry = new_dentry,
3875 .inode = d_inode(new_dentry),
3876 };
3877 int force_reread;
3878 int retval;
3879 struct inode *whiteout = NULL;
3880 int credits;
3881 u8 old_file_type;
3882
3883 if (new.inode && new.inode->i_nlink == 0) {
3884 EXT4_ERROR_INODE(new.inode,
3885 "target of rename is already freed");
3886 return -EFSCORRUPTED;
3887 }
3888
3889 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3890 (!projid_eq(EXT4_I(new_dir)->i_projid,
3891 EXT4_I(old_dentry->d_inode)->i_projid)))
3892 return -EXDEV;
3893
3894 retval = dquot_initialize(old.dir);
3895 if (retval)
3896 return retval;
3897 retval = dquot_initialize(new.dir);
3898 if (retval)
3899 return retval;
3900
3901 /* Initialize quotas before so that eventual writes go
3902 * in separate transaction */
3903 if (new.inode) {
3904 retval = dquot_initialize(new.inode);
3905 if (retval)
3906 return retval;
3907 }
3908
3909 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3910 if (IS_ERR(old.bh))
3911 return PTR_ERR(old.bh);
3912 /*
3913 * Check for inode number is _not_ due to possible IO errors.
3914 * We might rmdir the source, keep it as pwd of some process
3915 * and merrily kill the link to whatever was created under the
3916 * same name. Goodbye sticky bit ;-<
3917 */
3918 retval = -ENOENT;
3919 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3920 goto release_bh;
3921
3922 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3923 &new.de, &new.inlined);
3924 if (IS_ERR(new.bh)) {
3925 retval = PTR_ERR(new.bh);
3926 new.bh = NULL;
3927 goto release_bh;
3928 }
3929 if (new.bh) {
3930 if (!new.inode) {
3931 brelse(new.bh);
3932 new.bh = NULL;
3933 }
3934 }
3935 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3936 ext4_alloc_da_blocks(old.inode);
3937
3938 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3939 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3940 if (!(flags & RENAME_WHITEOUT)) {
3941 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3942 if (IS_ERR(handle)) {
3943 retval = PTR_ERR(handle);
3944 goto release_bh;
3945 }
3946 } else {
3947 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3948 if (IS_ERR(whiteout)) {
3949 retval = PTR_ERR(whiteout);
3950 goto release_bh;
3951 }
3952 }
3953
3954 old_file_type = old.de->file_type;
3955 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3956 ext4_handle_sync(handle);
3957
3958 if (S_ISDIR(old.inode->i_mode)) {
3959 if (new.inode) {
3960 retval = -ENOTEMPTY;
3961 if (!ext4_empty_dir(new.inode))
3962 goto end_rename;
3963 } else {
3964 retval = -EMLINK;
3965 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3966 goto end_rename;
3967 }
3968 retval = ext4_rename_dir_prepare(handle, &old);
3969 if (retval)
3970 goto end_rename;
3971 }
3972 /*
3973 * If we're renaming a file within an inline_data dir and adding or
3974 * setting the new dirent causes a conversion from inline_data to
3975 * extents/blockmap, we need to force the dirent delete code to
3976 * re-read the directory, or else we end up trying to delete a dirent
3977 * from what is now the extent tree root (or a block map).
3978 */
3979 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3980 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3981
3982 if (whiteout) {
3983 /*
3984 * Do this before adding a new entry, so the old entry is sure
3985 * to be still pointing to the valid old entry.
3986 */
3987 retval = ext4_setent(handle, &old, whiteout->i_ino,
3988 EXT4_FT_CHRDEV);
3989 if (retval)
3990 goto end_rename;
3991 retval = ext4_mark_inode_dirty(handle, whiteout);
3992 if (unlikely(retval))
3993 goto end_rename;
3994
3995 }
3996 if (!new.bh) {
3997 retval = ext4_add_entry(handle, new.dentry, old.inode);
3998 if (retval)
3999 goto end_rename;
4000 } else {
4001 retval = ext4_setent(handle, &new,
4002 old.inode->i_ino, old_file_type);
4003 if (retval)
4004 goto end_rename;
4005 }
4006 if (force_reread)
4007 force_reread = !ext4_test_inode_flag(new.dir,
4008 EXT4_INODE_INLINE_DATA);
4009
4010 /*
4011 * Like most other Unix systems, set the ctime for inodes on a
4012 * rename.
4013 */
4014 old.inode->i_ctime = current_time(old.inode);
4015 retval = ext4_mark_inode_dirty(handle, old.inode);
4016 if (unlikely(retval))
4017 goto end_rename;
4018
4019 if (!whiteout) {
4020 /*
4021 * ok, that's it
4022 */
4023 ext4_rename_delete(handle, &old, force_reread);
4024 }
4025
4026 if (new.inode) {
4027 ext4_dec_count(new.inode);
4028 new.inode->i_ctime = current_time(new.inode);
4029 }
4030 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
4031 ext4_update_dx_flag(old.dir);
4032 if (old.dir_bh) {
4033 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4034 if (retval)
4035 goto end_rename;
4036
4037 ext4_dec_count(old.dir);
4038 if (new.inode) {
4039 /* checked ext4_empty_dir above, can't have another
4040 * parent, ext4_dec_count() won't work for many-linked
4041 * dirs */
4042 clear_nlink(new.inode);
4043 } else {
4044 ext4_inc_count(new.dir);
4045 ext4_update_dx_flag(new.dir);
4046 retval = ext4_mark_inode_dirty(handle, new.dir);
4047 if (unlikely(retval))
4048 goto end_rename;
4049 }
4050 }
4051 retval = ext4_mark_inode_dirty(handle, old.dir);
4052 if (unlikely(retval))
4053 goto end_rename;
4054
4055 if (S_ISDIR(old.inode->i_mode)) {
4056 /*
4057 * We disable fast commits here that's because the
4058 * replay code is not yet capable of changing dot dot
4059 * dirents in directories.
4060 */
4061 ext4_fc_mark_ineligible(old.inode->i_sb,
4062 EXT4_FC_REASON_RENAME_DIR);
4063 } else {
4064 if (new.inode)
4065 ext4_fc_track_unlink(handle, new.dentry);
4066 __ext4_fc_track_link(handle, old.inode, new.dentry);
4067 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4068 if (whiteout)
4069 __ext4_fc_track_create(handle, whiteout, old.dentry);
4070 }
4071
4072 if (new.inode) {
4073 retval = ext4_mark_inode_dirty(handle, new.inode);
4074 if (unlikely(retval))
4075 goto end_rename;
4076 if (!new.inode->i_nlink)
4077 ext4_orphan_add(handle, new.inode);
4078 }
4079 retval = 0;
4080
4081end_rename:
4082 if (whiteout) {
4083 if (retval) {
4084 ext4_resetent(handle, &old,
4085 old.inode->i_ino, old_file_type);
4086 drop_nlink(whiteout);
4087 ext4_orphan_add(handle, whiteout);
4088 }
4089 unlock_new_inode(whiteout);
4090 ext4_journal_stop(handle);
4091 iput(whiteout);
4092 } else {
4093 ext4_journal_stop(handle);
4094 }
4095release_bh:
4096 brelse(old.dir_bh);
4097 brelse(old.bh);
4098 brelse(new.bh);
4099 return retval;
4100}
4101
4102static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4103 struct inode *new_dir, struct dentry *new_dentry)
4104{
4105 handle_t *handle = NULL;
4106 struct ext4_renament old = {
4107 .dir = old_dir,
4108 .dentry = old_dentry,
4109 .inode = d_inode(old_dentry),
4110 };
4111 struct ext4_renament new = {
4112 .dir = new_dir,
4113 .dentry = new_dentry,
4114 .inode = d_inode(new_dentry),
4115 };
4116 u8 new_file_type;
4117 int retval;
4118 struct timespec64 ctime;
4119
4120 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4121 !projid_eq(EXT4_I(new_dir)->i_projid,
4122 EXT4_I(old_dentry->d_inode)->i_projid)) ||
4123 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4124 !projid_eq(EXT4_I(old_dir)->i_projid,
4125 EXT4_I(new_dentry->d_inode)->i_projid)))
4126 return -EXDEV;
4127
4128 retval = dquot_initialize(old.dir);
4129 if (retval)
4130 return retval;
4131 retval = dquot_initialize(new.dir);
4132 if (retval)
4133 return retval;
4134
4135 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4136 &old.de, &old.inlined);
4137 if (IS_ERR(old.bh))
4138 return PTR_ERR(old.bh);
4139 /*
4140 * Check for inode number is _not_ due to possible IO errors.
4141 * We might rmdir the source, keep it as pwd of some process
4142 * and merrily kill the link to whatever was created under the
4143 * same name. Goodbye sticky bit ;-<
4144 */
4145 retval = -ENOENT;
4146 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4147 goto end_rename;
4148
4149 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4150 &new.de, &new.inlined);
4151 if (IS_ERR(new.bh)) {
4152 retval = PTR_ERR(new.bh);
4153 new.bh = NULL;
4154 goto end_rename;
4155 }
4156
4157 /* RENAME_EXCHANGE case: old *and* new must both exist */
4158 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4159 goto end_rename;
4160
4161 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4162 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4163 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4164 if (IS_ERR(handle)) {
4165 retval = PTR_ERR(handle);
4166 handle = NULL;
4167 goto end_rename;
4168 }
4169
4170 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4171 ext4_handle_sync(handle);
4172
4173 if (S_ISDIR(old.inode->i_mode)) {
4174 old.is_dir = true;
4175 retval = ext4_rename_dir_prepare(handle, &old);
4176 if (retval)
4177 goto end_rename;
4178 }
4179 if (S_ISDIR(new.inode->i_mode)) {
4180 new.is_dir = true;
4181 retval = ext4_rename_dir_prepare(handle, &new);
4182 if (retval)
4183 goto end_rename;
4184 }
4185
4186 /*
4187 * Other than the special case of overwriting a directory, parents'
4188 * nlink only needs to be modified if this is a cross directory rename.
4189 */
4190 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4191 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4192 new.dir_nlink_delta = -old.dir_nlink_delta;
4193 retval = -EMLINK;
4194 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4195 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4196 goto end_rename;
4197 }
4198
4199 new_file_type = new.de->file_type;
4200 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4201 if (retval)
4202 goto end_rename;
4203
4204 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4205 if (retval)
4206 goto end_rename;
4207
4208 /*
4209 * Like most other Unix systems, set the ctime for inodes on a
4210 * rename.
4211 */
4212 ctime = current_time(old.inode);
4213 old.inode->i_ctime = ctime;
4214 new.inode->i_ctime = ctime;
4215 retval = ext4_mark_inode_dirty(handle, old.inode);
4216 if (unlikely(retval))
4217 goto end_rename;
4218 retval = ext4_mark_inode_dirty(handle, new.inode);
4219 if (unlikely(retval))
4220 goto end_rename;
4221 ext4_fc_mark_ineligible(new.inode->i_sb,
4222 EXT4_FC_REASON_CROSS_RENAME);
4223 if (old.dir_bh) {
4224 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4225 if (retval)
4226 goto end_rename;
4227 }
4228 if (new.dir_bh) {
4229 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4230 if (retval)
4231 goto end_rename;
4232 }
4233 ext4_update_dir_count(handle, &old);
4234 ext4_update_dir_count(handle, &new);
4235 retval = 0;
4236
4237end_rename:
4238 brelse(old.dir_bh);
4239 brelse(new.dir_bh);
4240 brelse(old.bh);
4241 brelse(new.bh);
4242 if (handle)
4243 ext4_journal_stop(handle);
4244 return retval;
4245}
4246
4247static int ext4_rename2(struct user_namespace *mnt_userns,
4248 struct inode *old_dir, struct dentry *old_dentry,
4249 struct inode *new_dir, struct dentry *new_dentry,
4250 unsigned int flags)
4251{
4252 int err;
4253
4254 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4255 return -EIO;
4256
4257 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4258 return -EINVAL;
4259
4260 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4261 flags);
4262 if (err)
4263 return err;
4264
4265 if (flags & RENAME_EXCHANGE) {
4266 return ext4_cross_rename(old_dir, old_dentry,
4267 new_dir, new_dentry);
4268 }
4269
4270 return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4271}
4272
4273/*
4274 * directories can handle most operations...
4275 */
4276const struct inode_operations ext4_dir_inode_operations = {
4277 .create = ext4_create,
4278 .lookup = ext4_lookup,
4279 .link = ext4_link,
4280 .unlink = ext4_unlink,
4281 .symlink = ext4_symlink,
4282 .mkdir = ext4_mkdir,
4283 .rmdir = ext4_rmdir,
4284 .mknod = ext4_mknod,
4285 .tmpfile = ext4_tmpfile,
4286 .rename = ext4_rename2,
4287 .setattr = ext4_setattr,
4288 .getattr = ext4_getattr,
4289 .listxattr = ext4_listxattr,
4290 .get_acl = ext4_get_acl,
4291 .set_acl = ext4_set_acl,
4292 .fiemap = ext4_fiemap,
4293 .fileattr_get = ext4_fileattr_get,
4294 .fileattr_set = ext4_fileattr_set,
4295};
4296
4297const struct inode_operations ext4_special_inode_operations = {
4298 .setattr = ext4_setattr,
4299 .getattr = ext4_getattr,
4300 .listxattr = ext4_listxattr,
4301 .get_acl = ext4_get_acl,
4302 .set_acl = ext4_set_acl,
4303};