defconfig: exynos9610: Re-add dropped Wi-Fi AP options lost
[GitHub/LineageOS/android_kernel_motorola_exynos9610.git] / fs / ext4 / namei.c
... / ...
CommitLineData
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 "ext4.h"
38#include "ext4_jbd2.h"
39
40#include "xattr.h"
41#include "acl.h"
42
43#include <trace/events/ext4.h>
44/*
45 * define how far ahead to read directories while searching them.
46 */
47#define NAMEI_RA_CHUNKS 2
48#define NAMEI_RA_BLOCKS 4
49#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50
51static struct buffer_head *ext4_append(handle_t *handle,
52 struct inode *inode,
53 ext4_lblk_t *block)
54{
55 struct buffer_head *bh;
56 int err;
57
58 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 ((inode->i_size >> 10) >=
60 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
61 return ERR_PTR(-ENOSPC);
62
63 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64
65 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
66 if (IS_ERR(bh))
67 return bh;
68 inode->i_size += inode->i_sb->s_blocksize;
69 EXT4_I(inode)->i_disksize = inode->i_size;
70 BUFFER_TRACE(bh, "get_write_access");
71 err = ext4_journal_get_write_access(handle, bh);
72 if (err) {
73 brelse(bh);
74 ext4_std_error(inode->i_sb, err);
75 return ERR_PTR(err);
76 }
77 return bh;
78}
79
80static int ext4_dx_csum_verify(struct inode *inode,
81 struct ext4_dir_entry *dirent);
82
83typedef enum {
84 EITHER, INDEX, DIRENT
85} dirblock_type_t;
86
87#define ext4_read_dirblock(inode, block, type) \
88 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
89
90static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
91 ext4_lblk_t block,
92 dirblock_type_t type,
93 const char *func,
94 unsigned int line)
95{
96 struct buffer_head *bh;
97 struct ext4_dir_entry *dirent;
98 int is_dx_block = 0;
99
100 bh = ext4_bread(NULL, inode, block, 0);
101 if (IS_ERR(bh)) {
102 __ext4_warning(inode->i_sb, func, line,
103 "inode #%lu: lblock %lu: comm %s: "
104 "error %ld reading directory block",
105 inode->i_ino, (unsigned long)block,
106 current->comm, PTR_ERR(bh));
107
108 return bh;
109 }
110 if (!bh) {
111 ext4_error_inode(inode, func, line, block,
112 "Directory hole found");
113 return ERR_PTR(-EFSCORRUPTED);
114 }
115 dirent = (struct ext4_dir_entry *) bh->b_data;
116 /* Determine whether or not we have an index block */
117 if (is_dx(inode)) {
118 if (block == 0)
119 is_dx_block = 1;
120 else if (ext4_rec_len_from_disk(dirent->rec_len,
121 inode->i_sb->s_blocksize) ==
122 inode->i_sb->s_blocksize)
123 is_dx_block = 1;
124 }
125 if (!is_dx_block && type == INDEX) {
126 ext4_error_inode(inode, func, line, block,
127 "directory leaf block found instead of index block");
128 brelse(bh);
129 return ERR_PTR(-EFSCORRUPTED);
130 }
131 if (!ext4_has_metadata_csum(inode->i_sb) ||
132 buffer_verified(bh))
133 return bh;
134
135 /*
136 * An empty leaf block can get mistaken for a index block; for
137 * this reason, we can only check the index checksum when the
138 * caller is sure it should be an index block.
139 */
140 if (is_dx_block && type == INDEX) {
141 if (ext4_dx_csum_verify(inode, dirent))
142 set_buffer_verified(bh);
143 else {
144 ext4_error_inode(inode, func, line, block,
145 "Directory index failed checksum");
146 brelse(bh);
147 return ERR_PTR(-EFSBADCRC);
148 }
149 }
150 if (!is_dx_block) {
151 if (ext4_dirent_csum_verify(inode, dirent))
152 set_buffer_verified(bh);
153 else {
154 ext4_error_inode(inode, func, line, block,
155 "Directory block failed checksum");
156 brelse(bh);
157 return ERR_PTR(-EFSBADCRC);
158 }
159 }
160 return bh;
161}
162
163#ifndef assert
164#define assert(test) J_ASSERT(test)
165#endif
166
167#ifdef DX_DEBUG
168#define dxtrace(command) command
169#else
170#define dxtrace(command)
171#endif
172
173struct fake_dirent
174{
175 __le32 inode;
176 __le16 rec_len;
177 u8 name_len;
178 u8 file_type;
179};
180
181struct dx_countlimit
182{
183 __le16 limit;
184 __le16 count;
185};
186
187struct dx_entry
188{
189 __le32 hash;
190 __le32 block;
191};
192
193/*
194 * dx_root_info is laid out so that if it should somehow get overlaid by a
195 * dirent the two low bits of the hash version will be zero. Therefore, the
196 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
197 */
198
199struct dx_root
200{
201 struct fake_dirent dot;
202 char dot_name[4];
203 struct fake_dirent dotdot;
204 char dotdot_name[4];
205 struct dx_root_info
206 {
207 __le32 reserved_zero;
208 u8 hash_version;
209 u8 info_length; /* 8 */
210 u8 indirect_levels;
211 u8 unused_flags;
212 }
213 info;
214 struct dx_entry entries[0];
215};
216
217struct dx_node
218{
219 struct fake_dirent fake;
220 struct dx_entry entries[0];
221};
222
223
224struct dx_frame
225{
226 struct buffer_head *bh;
227 struct dx_entry *entries;
228 struct dx_entry *at;
229};
230
231struct dx_map_entry
232{
233 u32 hash;
234 u16 offs;
235 u16 size;
236};
237
238/*
239 * This goes at the end of each htree block.
240 */
241struct dx_tail {
242 u32 dt_reserved;
243 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
244};
245
246static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
247static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
248static inline unsigned dx_get_hash(struct dx_entry *entry);
249static void dx_set_hash(struct dx_entry *entry, unsigned value);
250static unsigned dx_get_count(struct dx_entry *entries);
251static unsigned dx_get_limit(struct dx_entry *entries);
252static void dx_set_count(struct dx_entry *entries, unsigned value);
253static void dx_set_limit(struct dx_entry *entries, unsigned value);
254static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
255static unsigned dx_node_limit(struct inode *dir);
256static struct dx_frame *dx_probe(struct ext4_filename *fname,
257 struct inode *dir,
258 struct dx_hash_info *hinfo,
259 struct dx_frame *frame);
260static void dx_release(struct dx_frame *frames);
261static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
262 unsigned blocksize, struct dx_hash_info *hinfo,
263 struct dx_map_entry map[]);
264static void dx_sort_map(struct dx_map_entry *map, unsigned count);
265static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
266 struct dx_map_entry *offsets, int count, unsigned blocksize);
267static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
268static void dx_insert_block(struct dx_frame *frame,
269 u32 hash, ext4_lblk_t block);
270static int ext4_htree_next_block(struct inode *dir, __u32 hash,
271 struct dx_frame *frame,
272 struct dx_frame *frames,
273 __u32 *start_hash);
274static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
275 struct ext4_filename *fname,
276 struct ext4_dir_entry_2 **res_dir);
277static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
278 struct inode *dir, struct inode *inode);
279
280/* checksumming functions */
281void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
282 unsigned int blocksize)
283{
284 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
285 t->det_rec_len = ext4_rec_len_to_disk(
286 sizeof(struct ext4_dir_entry_tail), blocksize);
287 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
288}
289
290/* Walk through a dirent block to find a checksum "dirent" at the tail */
291static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
292 struct ext4_dir_entry *de)
293{
294 struct ext4_dir_entry_tail *t;
295
296#ifdef PARANOID
297 struct ext4_dir_entry *d, *top;
298
299 d = de;
300 top = (struct ext4_dir_entry *)(((void *)de) +
301 (EXT4_BLOCK_SIZE(inode->i_sb) -
302 sizeof(struct ext4_dir_entry_tail)));
303 while (d < top && d->rec_len)
304 d = (struct ext4_dir_entry *)(((void *)d) +
305 le16_to_cpu(d->rec_len));
306
307 if (d != top)
308 return NULL;
309
310 t = (struct ext4_dir_entry_tail *)d;
311#else
312 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
313#endif
314
315 if (t->det_reserved_zero1 ||
316 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
317 t->det_reserved_zero2 ||
318 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
319 return NULL;
320
321 return t;
322}
323
324static __le32 ext4_dirent_csum(struct inode *inode,
325 struct ext4_dir_entry *dirent, int size)
326{
327 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
328 struct ext4_inode_info *ei = EXT4_I(inode);
329 __u32 csum;
330
331 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
332 return cpu_to_le32(csum);
333}
334
335#define warn_no_space_for_csum(inode) \
336 __warn_no_space_for_csum((inode), __func__, __LINE__)
337
338static void __warn_no_space_for_csum(struct inode *inode, const char *func,
339 unsigned int line)
340{
341 __ext4_warning_inode(inode, func, line,
342 "No space for directory leaf checksum. Please run e2fsck -D.");
343}
344
345int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
346{
347 struct ext4_dir_entry_tail *t;
348
349 if (!ext4_has_metadata_csum(inode->i_sb))
350 return 1;
351
352 t = get_dirent_tail(inode, dirent);
353 if (!t) {
354 warn_no_space_for_csum(inode);
355 return 0;
356 }
357
358 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
359 (void *)t - (void *)dirent))
360 return 0;
361
362 return 1;
363}
364
365static void ext4_dirent_csum_set(struct inode *inode,
366 struct ext4_dir_entry *dirent)
367{
368 struct ext4_dir_entry_tail *t;
369
370 if (!ext4_has_metadata_csum(inode->i_sb))
371 return;
372
373 t = get_dirent_tail(inode, dirent);
374 if (!t) {
375 warn_no_space_for_csum(inode);
376 return;
377 }
378
379 t->det_checksum = ext4_dirent_csum(inode, dirent,
380 (void *)t - (void *)dirent);
381}
382
383int ext4_handle_dirty_dirent_node(handle_t *handle,
384 struct inode *inode,
385 struct buffer_head *bh)
386{
387 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
388 return ext4_handle_dirty_metadata(handle, inode, bh);
389}
390
391static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
392 struct ext4_dir_entry *dirent,
393 int *offset)
394{
395 struct ext4_dir_entry *dp;
396 struct dx_root_info *root;
397 int count_offset;
398
399 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
400 count_offset = 8;
401 else if (le16_to_cpu(dirent->rec_len) == 12) {
402 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
403 if (le16_to_cpu(dp->rec_len) !=
404 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
405 return NULL;
406 root = (struct dx_root_info *)(((void *)dp + 12));
407 if (root->reserved_zero ||
408 root->info_length != sizeof(struct dx_root_info))
409 return NULL;
410 count_offset = 32;
411 } else
412 return NULL;
413
414 if (offset)
415 *offset = count_offset;
416 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
417}
418
419static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
420 int count_offset, int count, struct dx_tail *t)
421{
422 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
423 struct ext4_inode_info *ei = EXT4_I(inode);
424 __u32 csum;
425 int size;
426 __u32 dummy_csum = 0;
427 int offset = offsetof(struct dx_tail, dt_checksum);
428
429 size = count_offset + (count * sizeof(struct dx_entry));
430 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
431 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
432 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
433
434 return cpu_to_le32(csum);
435}
436
437static int ext4_dx_csum_verify(struct inode *inode,
438 struct ext4_dir_entry *dirent)
439{
440 struct dx_countlimit *c;
441 struct dx_tail *t;
442 int count_offset, limit, count;
443
444 if (!ext4_has_metadata_csum(inode->i_sb))
445 return 1;
446
447 c = get_dx_countlimit(inode, dirent, &count_offset);
448 if (!c) {
449 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
450 return 0;
451 }
452 limit = le16_to_cpu(c->limit);
453 count = le16_to_cpu(c->count);
454 if (count_offset + (limit * sizeof(struct dx_entry)) >
455 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
456 warn_no_space_for_csum(inode);
457 return 0;
458 }
459 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
460
461 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
462 count, t))
463 return 0;
464 return 1;
465}
466
467static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
468{
469 struct dx_countlimit *c;
470 struct dx_tail *t;
471 int count_offset, limit, count;
472
473 if (!ext4_has_metadata_csum(inode->i_sb))
474 return;
475
476 c = get_dx_countlimit(inode, dirent, &count_offset);
477 if (!c) {
478 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
479 return;
480 }
481 limit = le16_to_cpu(c->limit);
482 count = le16_to_cpu(c->count);
483 if (count_offset + (limit * sizeof(struct dx_entry)) >
484 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
485 warn_no_space_for_csum(inode);
486 return;
487 }
488 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
489
490 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
491}
492
493static inline int ext4_handle_dirty_dx_node(handle_t *handle,
494 struct inode *inode,
495 struct buffer_head *bh)
496{
497 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
498 return ext4_handle_dirty_metadata(handle, inode, bh);
499}
500
501/*
502 * p is at least 6 bytes before the end of page
503 */
504static inline struct ext4_dir_entry_2 *
505ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
506{
507 return (struct ext4_dir_entry_2 *)((char *)p +
508 ext4_rec_len_from_disk(p->rec_len, blocksize));
509}
510
511/*
512 * Future: use high four bits of block for coalesce-on-delete flags
513 * Mask them off for now.
514 */
515
516static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
517{
518 return le32_to_cpu(entry->block) & 0x0fffffff;
519}
520
521static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
522{
523 entry->block = cpu_to_le32(value);
524}
525
526static inline unsigned dx_get_hash(struct dx_entry *entry)
527{
528 return le32_to_cpu(entry->hash);
529}
530
531static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
532{
533 entry->hash = cpu_to_le32(value);
534}
535
536static inline unsigned dx_get_count(struct dx_entry *entries)
537{
538 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
539}
540
541static inline unsigned dx_get_limit(struct dx_entry *entries)
542{
543 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
544}
545
546static inline void dx_set_count(struct dx_entry *entries, unsigned value)
547{
548 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
549}
550
551static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
552{
553 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
554}
555
556static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
557{
558 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
559 EXT4_DIR_REC_LEN(2) - infosize;
560
561 if (ext4_has_metadata_csum(dir->i_sb))
562 entry_space -= sizeof(struct dx_tail);
563 return entry_space / sizeof(struct dx_entry);
564}
565
566static inline unsigned dx_node_limit(struct inode *dir)
567{
568 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
569
570 if (ext4_has_metadata_csum(dir->i_sb))
571 entry_space -= sizeof(struct dx_tail);
572 return entry_space / sizeof(struct dx_entry);
573}
574
575/*
576 * Debug
577 */
578#ifdef DX_DEBUG
579static void dx_show_index(char * label, struct dx_entry *entries)
580{
581 int i, n = dx_get_count (entries);
582 printk(KERN_DEBUG "%s index", label);
583 for (i = 0; i < n; i++) {
584 printk(KERN_CONT " %x->%lu",
585 i ? dx_get_hash(entries + i) : 0,
586 (unsigned long)dx_get_block(entries + i));
587 }
588 printk(KERN_CONT "\n");
589}
590
591struct stats
592{
593 unsigned names;
594 unsigned space;
595 unsigned bcount;
596};
597
598static struct stats dx_show_leaf(struct inode *dir,
599 struct dx_hash_info *hinfo,
600 struct ext4_dir_entry_2 *de,
601 int size, int show_names)
602{
603 unsigned names = 0, space = 0;
604 char *base = (char *) de;
605 struct dx_hash_info h = *hinfo;
606
607 printk("names: ");
608 while ((char *) de < base + size)
609 {
610 if (de->inode)
611 {
612 if (show_names)
613 {
614#ifdef CONFIG_EXT4_FS_ENCRYPTION
615 int len;
616 char *name;
617 struct fscrypt_str fname_crypto_str =
618 FSTR_INIT(NULL, 0);
619 int res = 0;
620
621 name = de->name;
622 len = de->name_len;
623 if (ext4_encrypted_inode(dir))
624 res = fscrypt_get_encryption_info(dir);
625 if (res) {
626 printk(KERN_WARNING "Error setting up"
627 " fname crypto: %d\n", res);
628 }
629 if (!fscrypt_has_encryption_key(dir)) {
630 /* Directory is not encrypted */
631 ext4fs_dirhash(de->name,
632 de->name_len, &h);
633 printk("%*.s:(U)%x.%u ", len,
634 name, h.hash,
635 (unsigned) ((char *) de
636 - base));
637 } else {
638 struct fscrypt_str de_name =
639 FSTR_INIT(name, len);
640
641 /* Directory is encrypted */
642 res = fscrypt_fname_alloc_buffer(
643 dir, len,
644 &fname_crypto_str);
645 if (res)
646 printk(KERN_WARNING "Error "
647 "allocating crypto "
648 "buffer--skipping "
649 "crypto\n");
650 res = fscrypt_fname_disk_to_usr(dir,
651 0, 0, &de_name,
652 &fname_crypto_str);
653 if (res) {
654 printk(KERN_WARNING "Error "
655 "converting filename "
656 "from disk to usr"
657 "\n");
658 name = "??";
659 len = 2;
660 } else {
661 name = fname_crypto_str.name;
662 len = fname_crypto_str.len;
663 }
664 ext4fs_dirhash(de->name, de->name_len,
665 &h);
666 printk("%*.s:(E)%x.%u ", len, name,
667 h.hash, (unsigned) ((char *) de
668 - base));
669 fscrypt_fname_free_buffer(
670 &fname_crypto_str);
671 }
672#else
673 int len = de->name_len;
674 char *name = de->name;
675 ext4fs_dirhash(de->name, de->name_len, &h);
676 printk("%*.s:%x.%u ", len, name, h.hash,
677 (unsigned) ((char *) de - base));
678#endif
679 }
680 space += EXT4_DIR_REC_LEN(de->name_len);
681 names++;
682 }
683 de = ext4_next_entry(de, size);
684 }
685 printk(KERN_CONT "(%i)\n", names);
686 return (struct stats) { names, space, 1 };
687}
688
689struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
690 struct dx_entry *entries, int levels)
691{
692 unsigned blocksize = dir->i_sb->s_blocksize;
693 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
694 unsigned bcount = 0;
695 struct buffer_head *bh;
696 printk("%i indexed blocks...\n", count);
697 for (i = 0; i < count; i++, entries++)
698 {
699 ext4_lblk_t block = dx_get_block(entries);
700 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
701 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
702 struct stats stats;
703 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
704 bh = ext4_bread(NULL,dir, block, 0);
705 if (!bh || IS_ERR(bh))
706 continue;
707 stats = levels?
708 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
709 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
710 bh->b_data, blocksize, 0);
711 names += stats.names;
712 space += stats.space;
713 bcount += stats.bcount;
714 brelse(bh);
715 }
716 if (bcount)
717 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
718 levels ? "" : " ", names, space/bcount,
719 (space/bcount)*100/blocksize);
720 return (struct stats) { names, space, bcount};
721}
722#endif /* DX_DEBUG */
723
724/*
725 * Probe for a directory leaf block to search.
726 *
727 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
728 * error in the directory index, and the caller should fall back to
729 * searching the directory normally. The callers of dx_probe **MUST**
730 * check for this error code, and make sure it never gets reflected
731 * back to userspace.
732 */
733static struct dx_frame *
734dx_probe(struct ext4_filename *fname, struct inode *dir,
735 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
736{
737 unsigned count, indirect;
738 struct dx_entry *at, *entries, *p, *q, *m;
739 struct dx_root *root;
740 struct dx_frame *frame = frame_in;
741 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
742 u32 hash;
743
744 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
745 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
746 if (IS_ERR(frame->bh))
747 return (struct dx_frame *) frame->bh;
748
749 root = (struct dx_root *) frame->bh->b_data;
750 if (root->info.hash_version != DX_HASH_TEA &&
751 root->info.hash_version != DX_HASH_HALF_MD4 &&
752 root->info.hash_version != DX_HASH_LEGACY) {
753 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
754 root->info.hash_version);
755 goto fail;
756 }
757 if (fname)
758 hinfo = &fname->hinfo;
759 hinfo->hash_version = root->info.hash_version;
760 if (hinfo->hash_version <= DX_HASH_TEA)
761 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
762 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
763 if (fname && fname_name(fname))
764 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
765 hash = hinfo->hash;
766
767 if (root->info.unused_flags & 1) {
768 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
769 root->info.unused_flags);
770 goto fail;
771 }
772
773 indirect = root->info.indirect_levels;
774 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
775 ext4_warning(dir->i_sb,
776 "Directory (ino: %lu) htree depth %#06x exceed"
777 "supported value", dir->i_ino,
778 ext4_dir_htree_level(dir->i_sb));
779 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
780 ext4_warning(dir->i_sb, "Enable large directory "
781 "feature to access it");
782 }
783 goto fail;
784 }
785
786 entries = (struct dx_entry *)(((char *)&root->info) +
787 root->info.info_length);
788
789 if (dx_get_limit(entries) != dx_root_limit(dir,
790 root->info.info_length)) {
791 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
792 dx_get_limit(entries),
793 dx_root_limit(dir, root->info.info_length));
794 goto fail;
795 }
796
797 dxtrace(printk("Look up %x", hash));
798 while (1) {
799 count = dx_get_count(entries);
800 if (!count || count > dx_get_limit(entries)) {
801 ext4_warning_inode(dir,
802 "dx entry: count %u beyond limit %u",
803 count, dx_get_limit(entries));
804 goto fail;
805 }
806
807 p = entries + 1;
808 q = entries + count - 1;
809 while (p <= q) {
810 m = p + (q - p) / 2;
811 dxtrace(printk(KERN_CONT "."));
812 if (dx_get_hash(m) > hash)
813 q = m - 1;
814 else
815 p = m + 1;
816 }
817
818 if (0) { // linear search cross check
819 unsigned n = count - 1;
820 at = entries;
821 while (n--)
822 {
823 dxtrace(printk(KERN_CONT ","));
824 if (dx_get_hash(++at) > hash)
825 {
826 at--;
827 break;
828 }
829 }
830 assert (at == p - 1);
831 }
832
833 at = p - 1;
834 dxtrace(printk(KERN_CONT " %x->%u\n",
835 at == entries ? 0 : dx_get_hash(at),
836 dx_get_block(at)));
837 frame->entries = entries;
838 frame->at = at;
839 if (!indirect--)
840 return frame;
841 frame++;
842 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
843 if (IS_ERR(frame->bh)) {
844 ret_err = (struct dx_frame *) frame->bh;
845 frame->bh = NULL;
846 goto fail;
847 }
848 entries = ((struct dx_node *) frame->bh->b_data)->entries;
849
850 if (dx_get_limit(entries) != dx_node_limit(dir)) {
851 ext4_warning_inode(dir,
852 "dx entry: limit %u != node limit %u",
853 dx_get_limit(entries), dx_node_limit(dir));
854 goto fail;
855 }
856 }
857fail:
858 while (frame >= frame_in) {
859 brelse(frame->bh);
860 frame--;
861 }
862
863 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
864 ext4_warning_inode(dir,
865 "Corrupt directory, running e2fsck is recommended");
866 return ret_err;
867}
868
869static void dx_release(struct dx_frame *frames)
870{
871 struct dx_root_info *info;
872 int i;
873
874 if (frames[0].bh == NULL)
875 return;
876
877 info = &((struct dx_root *)frames[0].bh->b_data)->info;
878 for (i = 0; i <= info->indirect_levels; i++) {
879 if (frames[i].bh == NULL)
880 break;
881 brelse(frames[i].bh);
882 frames[i].bh = NULL;
883 }
884}
885
886/*
887 * This function increments the frame pointer to search the next leaf
888 * block, and reads in the necessary intervening nodes if the search
889 * should be necessary. Whether or not the search is necessary is
890 * controlled by the hash parameter. If the hash value is even, then
891 * the search is only continued if the next block starts with that
892 * hash value. This is used if we are searching for a specific file.
893 *
894 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
895 *
896 * This function returns 1 if the caller should continue to search,
897 * or 0 if it should not. If there is an error reading one of the
898 * index blocks, it will a negative error code.
899 *
900 * If start_hash is non-null, it will be filled in with the starting
901 * hash of the next page.
902 */
903static int ext4_htree_next_block(struct inode *dir, __u32 hash,
904 struct dx_frame *frame,
905 struct dx_frame *frames,
906 __u32 *start_hash)
907{
908 struct dx_frame *p;
909 struct buffer_head *bh;
910 int num_frames = 0;
911 __u32 bhash;
912
913 p = frame;
914 /*
915 * Find the next leaf page by incrementing the frame pointer.
916 * If we run out of entries in the interior node, loop around and
917 * increment pointer in the parent node. When we break out of
918 * this loop, num_frames indicates the number of interior
919 * nodes need to be read.
920 */
921 while (1) {
922 if (++(p->at) < p->entries + dx_get_count(p->entries))
923 break;
924 if (p == frames)
925 return 0;
926 num_frames++;
927 p--;
928 }
929
930 /*
931 * If the hash is 1, then continue only if the next page has a
932 * continuation hash of any value. This is used for readdir
933 * handling. Otherwise, check to see if the hash matches the
934 * desired contiuation hash. If it doesn't, return since
935 * there's no point to read in the successive index pages.
936 */
937 bhash = dx_get_hash(p->at);
938 if (start_hash)
939 *start_hash = bhash;
940 if ((hash & 1) == 0) {
941 if ((bhash & ~1) != hash)
942 return 0;
943 }
944 /*
945 * If the hash is HASH_NB_ALWAYS, we always go to the next
946 * block so no check is necessary
947 */
948 while (num_frames--) {
949 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
950 if (IS_ERR(bh))
951 return PTR_ERR(bh);
952 p++;
953 brelse(p->bh);
954 p->bh = bh;
955 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
956 }
957 return 1;
958}
959
960
961/*
962 * This function fills a red-black tree with information from a
963 * directory block. It returns the number directory entries loaded
964 * into the tree. If there is an error it is returned in err.
965 */
966static int htree_dirblock_to_tree(struct file *dir_file,
967 struct inode *dir, ext4_lblk_t block,
968 struct dx_hash_info *hinfo,
969 __u32 start_hash, __u32 start_minor_hash)
970{
971 struct buffer_head *bh;
972 struct ext4_dir_entry_2 *de, *top;
973 int err = 0, count = 0;
974 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
975
976 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
977 (unsigned long)block));
978 bh = ext4_read_dirblock(dir, block, DIRENT);
979 if (IS_ERR(bh))
980 return PTR_ERR(bh);
981
982 de = (struct ext4_dir_entry_2 *) bh->b_data;
983 top = (struct ext4_dir_entry_2 *) ((char *) de +
984 dir->i_sb->s_blocksize -
985 EXT4_DIR_REC_LEN(0));
986#ifdef CONFIG_EXT4_FS_ENCRYPTION
987 /* Check if the directory is encrypted */
988 if (ext4_encrypted_inode(dir)) {
989 err = fscrypt_get_encryption_info(dir);
990 if (err < 0) {
991 brelse(bh);
992 return err;
993 }
994 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
995 &fname_crypto_str);
996 if (err < 0) {
997 brelse(bh);
998 return err;
999 }
1000 }
1001#endif
1002 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1003 if (ext4_check_dir_entry(dir, NULL, de, bh,
1004 bh->b_data, bh->b_size,
1005 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1006 + ((char *)de - bh->b_data))) {
1007 /* silently ignore the rest of the block */
1008 break;
1009 }
1010 ext4fs_dirhash(de->name, de->name_len, hinfo);
1011 if ((hinfo->hash < start_hash) ||
1012 ((hinfo->hash == start_hash) &&
1013 (hinfo->minor_hash < start_minor_hash)))
1014 continue;
1015 if (de->inode == 0)
1016 continue;
1017 if (!ext4_encrypted_inode(dir)) {
1018 tmp_str.name = de->name;
1019 tmp_str.len = de->name_len;
1020 err = ext4_htree_store_dirent(dir_file,
1021 hinfo->hash, hinfo->minor_hash, de,
1022 &tmp_str);
1023 } else {
1024 int save_len = fname_crypto_str.len;
1025 struct fscrypt_str de_name = FSTR_INIT(de->name,
1026 de->name_len);
1027
1028 /* Directory is encrypted */
1029 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1030 hinfo->minor_hash, &de_name,
1031 &fname_crypto_str);
1032 if (err) {
1033 count = err;
1034 goto errout;
1035 }
1036 err = ext4_htree_store_dirent(dir_file,
1037 hinfo->hash, hinfo->minor_hash, de,
1038 &fname_crypto_str);
1039 fname_crypto_str.len = save_len;
1040 }
1041 if (err != 0) {
1042 count = err;
1043 goto errout;
1044 }
1045 count++;
1046 }
1047errout:
1048 brelse(bh);
1049#ifdef CONFIG_EXT4_FS_ENCRYPTION
1050 fscrypt_fname_free_buffer(&fname_crypto_str);
1051#endif
1052 return count;
1053}
1054
1055
1056/*
1057 * This function fills a red-black tree with information from a
1058 * directory. We start scanning the directory in hash order, starting
1059 * at start_hash and start_minor_hash.
1060 *
1061 * This function returns the number of entries inserted into the tree,
1062 * or a negative error code.
1063 */
1064int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1065 __u32 start_minor_hash, __u32 *next_hash)
1066{
1067 struct dx_hash_info hinfo;
1068 struct ext4_dir_entry_2 *de;
1069 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1070 struct inode *dir;
1071 ext4_lblk_t block;
1072 int count = 0;
1073 int ret, err;
1074 __u32 hashval;
1075 struct fscrypt_str tmp_str;
1076
1077 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1078 start_hash, start_minor_hash));
1079 dir = file_inode(dir_file);
1080 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1081 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1082 if (hinfo.hash_version <= DX_HASH_TEA)
1083 hinfo.hash_version +=
1084 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1085 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1086 if (ext4_has_inline_data(dir)) {
1087 int has_inline_data = 1;
1088 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1089 &hinfo, start_hash,
1090 start_minor_hash,
1091 &has_inline_data);
1092 if (has_inline_data) {
1093 *next_hash = ~0;
1094 return count;
1095 }
1096 }
1097 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1098 start_hash, start_minor_hash);
1099 *next_hash = ~0;
1100 return count;
1101 }
1102 hinfo.hash = start_hash;
1103 hinfo.minor_hash = 0;
1104 frame = dx_probe(NULL, dir, &hinfo, frames);
1105 if (IS_ERR(frame))
1106 return PTR_ERR(frame);
1107
1108 /* Add '.' and '..' from the htree header */
1109 if (!start_hash && !start_minor_hash) {
1110 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1111 tmp_str.name = de->name;
1112 tmp_str.len = de->name_len;
1113 err = ext4_htree_store_dirent(dir_file, 0, 0,
1114 de, &tmp_str);
1115 if (err != 0)
1116 goto errout;
1117 count++;
1118 }
1119 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1120 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1121 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1122 tmp_str.name = de->name;
1123 tmp_str.len = de->name_len;
1124 err = ext4_htree_store_dirent(dir_file, 2, 0,
1125 de, &tmp_str);
1126 if (err != 0)
1127 goto errout;
1128 count++;
1129 }
1130
1131 while (1) {
1132 if (fatal_signal_pending(current)) {
1133 err = -ERESTARTSYS;
1134 goto errout;
1135 }
1136 cond_resched();
1137 block = dx_get_block(frame->at);
1138 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1139 start_hash, start_minor_hash);
1140 if (ret < 0) {
1141 err = ret;
1142 goto errout;
1143 }
1144 count += ret;
1145 hashval = ~0;
1146 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1147 frame, frames, &hashval);
1148 *next_hash = hashval;
1149 if (ret < 0) {
1150 err = ret;
1151 goto errout;
1152 }
1153 /*
1154 * Stop if: (a) there are no more entries, or
1155 * (b) we have inserted at least one entry and the
1156 * next hash value is not a continuation
1157 */
1158 if ((ret == 0) ||
1159 (count && ((hashval & 1) == 0)))
1160 break;
1161 }
1162 dx_release(frames);
1163 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1164 "next hash: %x\n", count, *next_hash));
1165 return count;
1166errout:
1167 dx_release(frames);
1168 return (err);
1169}
1170
1171static inline int search_dirblock(struct buffer_head *bh,
1172 struct inode *dir,
1173 struct ext4_filename *fname,
1174 unsigned int offset,
1175 struct ext4_dir_entry_2 **res_dir)
1176{
1177 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1178 fname, offset, res_dir);
1179}
1180
1181/*
1182 * Directory block splitting, compacting
1183 */
1184
1185/*
1186 * Create map of hash values, offsets, and sizes, stored at end of block.
1187 * Returns number of entries mapped.
1188 */
1189static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1190 unsigned blocksize, struct dx_hash_info *hinfo,
1191 struct dx_map_entry *map_tail)
1192{
1193 int count = 0;
1194 char *base = (char *) de;
1195 struct dx_hash_info h = *hinfo;
1196
1197 while ((char *) de < base + blocksize) {
1198 if (de->name_len && de->inode) {
1199 ext4fs_dirhash(de->name, de->name_len, &h);
1200 map_tail--;
1201 map_tail->hash = h.hash;
1202 map_tail->offs = ((char *) de - base)>>2;
1203 map_tail->size = le16_to_cpu(de->rec_len);
1204 count++;
1205 cond_resched();
1206 }
1207 /* XXX: do we need to check rec_len == 0 case? -Chris */
1208 de = ext4_next_entry(de, blocksize);
1209 }
1210 return count;
1211}
1212
1213/* Sort map by hash value */
1214static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1215{
1216 struct dx_map_entry *p, *q, *top = map + count - 1;
1217 int more;
1218 /* Combsort until bubble sort doesn't suck */
1219 while (count > 2) {
1220 count = count*10/13;
1221 if (count - 9 < 2) /* 9, 10 -> 11 */
1222 count = 11;
1223 for (p = top, q = p - count; q >= map; p--, q--)
1224 if (p->hash < q->hash)
1225 swap(*p, *q);
1226 }
1227 /* Garden variety bubble sort */
1228 do {
1229 more = 0;
1230 q = top;
1231 while (q-- > map) {
1232 if (q[1].hash >= q[0].hash)
1233 continue;
1234 swap(*(q+1), *q);
1235 more = 1;
1236 }
1237 } while(more);
1238}
1239
1240static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1241{
1242 struct dx_entry *entries = frame->entries;
1243 struct dx_entry *old = frame->at, *new = old + 1;
1244 int count = dx_get_count(entries);
1245
1246 assert(count < dx_get_limit(entries));
1247 assert(old < entries + count);
1248 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1249 dx_set_hash(new, hash);
1250 dx_set_block(new, block);
1251 dx_set_count(entries, count + 1);
1252}
1253
1254/*
1255 * Test whether a directory entry matches the filename being searched for.
1256 *
1257 * Return: %true if the directory entry matches, otherwise %false.
1258 */
1259static inline bool ext4_match(const struct ext4_filename *fname,
1260 const struct ext4_dir_entry_2 *de)
1261{
1262 struct fscrypt_name f;
1263
1264 if (!de->inode)
1265 return false;
1266
1267 f.usr_fname = fname->usr_fname;
1268 f.disk_name = fname->disk_name;
1269#ifdef CONFIG_EXT4_FS_ENCRYPTION
1270 f.crypto_buf = fname->crypto_buf;
1271#endif
1272 return fscrypt_match_name(&f, de->name, de->name_len);
1273}
1274
1275/*
1276 * Returns 0 if not found, -1 on failure, and 1 on success
1277 */
1278int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1279 struct inode *dir, struct ext4_filename *fname,
1280 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1281{
1282 struct ext4_dir_entry_2 * de;
1283 char * dlimit;
1284 int de_len;
1285
1286 de = (struct ext4_dir_entry_2 *)search_buf;
1287 dlimit = search_buf + buf_size;
1288 while ((char *) de < dlimit) {
1289 /* this code is executed quadratically often */
1290 /* do minimal checking `by hand' */
1291 if ((char *) de + de->name_len <= dlimit &&
1292 ext4_match(fname, de)) {
1293 /* found a match - just to be sure, do
1294 * a full check */
1295 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1296 bh->b_size, offset))
1297 return -1;
1298 *res_dir = de;
1299 return 1;
1300 }
1301 /* prevent looping on a bad block */
1302 de_len = ext4_rec_len_from_disk(de->rec_len,
1303 dir->i_sb->s_blocksize);
1304 if (de_len <= 0)
1305 return -1;
1306 offset += de_len;
1307 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1308 }
1309 return 0;
1310}
1311
1312static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1313 struct ext4_dir_entry *de)
1314{
1315 struct super_block *sb = dir->i_sb;
1316
1317 if (!is_dx(dir))
1318 return 0;
1319 if (block == 0)
1320 return 1;
1321 if (de->inode == 0 &&
1322 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1323 sb->s_blocksize)
1324 return 1;
1325 return 0;
1326}
1327
1328/*
1329 * ext4_find_entry()
1330 *
1331 * finds an entry in the specified directory with the wanted name. It
1332 * returns the cache buffer in which the entry was found, and the entry
1333 * itself (as a parameter - res_dir). It does NOT read the inode of the
1334 * entry - you'll have to do that yourself if you want to.
1335 *
1336 * The returned buffer_head has ->b_count elevated. The caller is expected
1337 * to brelse() it when appropriate.
1338 */
1339static struct buffer_head * ext4_find_entry (struct inode *dir,
1340 const struct qstr *d_name,
1341 struct ext4_dir_entry_2 **res_dir,
1342 int *inlined)
1343{
1344 struct super_block *sb;
1345 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1346 struct buffer_head *bh, *ret = NULL;
1347 ext4_lblk_t start, block;
1348 const u8 *name = d_name->name;
1349 size_t ra_max = 0; /* Number of bh's in the readahead
1350 buffer, bh_use[] */
1351 size_t ra_ptr = 0; /* Current index into readahead
1352 buffer */
1353 ext4_lblk_t nblocks;
1354 int i, namelen, retval;
1355 struct ext4_filename fname;
1356
1357 *res_dir = NULL;
1358 sb = dir->i_sb;
1359 namelen = d_name->len;
1360 if (namelen > EXT4_NAME_LEN)
1361 return NULL;
1362
1363 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1364 if (retval == -ENOENT)
1365 return NULL;
1366 if (retval)
1367 return ERR_PTR(retval);
1368
1369 if (ext4_has_inline_data(dir)) {
1370 int has_inline_data = 1;
1371 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1372 &has_inline_data);
1373 if (has_inline_data) {
1374 if (inlined)
1375 *inlined = 1;
1376 goto cleanup_and_exit;
1377 }
1378 }
1379
1380 if ((namelen <= 2) && (name[0] == '.') &&
1381 (name[1] == '.' || name[1] == '\0')) {
1382 /*
1383 * "." or ".." will only be in the first block
1384 * NFS may look up ".."; "." should be handled by the VFS
1385 */
1386 block = start = 0;
1387 nblocks = 1;
1388 goto restart;
1389 }
1390 if (is_dx(dir)) {
1391 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1392 /*
1393 * On success, or if the error was file not found,
1394 * return. Otherwise, fall back to doing a search the
1395 * old fashioned way.
1396 */
1397 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1398 goto cleanup_and_exit;
1399 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1400 "falling back\n"));
1401 ret = NULL;
1402 }
1403 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1404 if (!nblocks) {
1405 ret = NULL;
1406 goto cleanup_and_exit;
1407 }
1408 start = EXT4_I(dir)->i_dir_start_lookup;
1409 if (start >= nblocks)
1410 start = 0;
1411 block = start;
1412restart:
1413 do {
1414 /*
1415 * We deal with the read-ahead logic here.
1416 */
1417 if (ra_ptr >= ra_max) {
1418 /* Refill the readahead buffer */
1419 ra_ptr = 0;
1420 if (block < start)
1421 ra_max = start - block;
1422 else
1423 ra_max = nblocks - block;
1424 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1425 retval = ext4_bread_batch(dir, block, ra_max,
1426 false /* wait */, bh_use);
1427 if (retval) {
1428 ret = ERR_PTR(retval);
1429 ra_max = 0;
1430 goto cleanup_and_exit;
1431 }
1432 }
1433 if ((bh = bh_use[ra_ptr++]) == NULL)
1434 goto next;
1435 wait_on_buffer(bh);
1436 if (!buffer_uptodate(bh)) {
1437 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1438 (unsigned long) block);
1439 brelse(bh);
1440 ret = ERR_PTR(-EIO);
1441 goto cleanup_and_exit;
1442 }
1443 if (!buffer_verified(bh) &&
1444 !is_dx_internal_node(dir, block,
1445 (struct ext4_dir_entry *)bh->b_data) &&
1446 !ext4_dirent_csum_verify(dir,
1447 (struct ext4_dir_entry *)bh->b_data)) {
1448 EXT4_ERROR_INODE(dir, "checksumming directory "
1449 "block %lu", (unsigned long)block);
1450 brelse(bh);
1451 ret = ERR_PTR(-EFSBADCRC);
1452 goto cleanup_and_exit;
1453 }
1454 set_buffer_verified(bh);
1455 i = search_dirblock(bh, dir, &fname,
1456 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1457 if (i == 1) {
1458 EXT4_I(dir)->i_dir_start_lookup = block;
1459 ret = bh;
1460 goto cleanup_and_exit;
1461 } else {
1462 brelse(bh);
1463 if (i < 0)
1464 goto cleanup_and_exit;
1465 }
1466 next:
1467 if (++block >= nblocks)
1468 block = 0;
1469 } while (block != start);
1470
1471 /*
1472 * If the directory has grown while we were searching, then
1473 * search the last part of the directory before giving up.
1474 */
1475 block = nblocks;
1476 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1477 if (block < nblocks) {
1478 start = 0;
1479 goto restart;
1480 }
1481
1482cleanup_and_exit:
1483 /* Clean up the read-ahead blocks */
1484 for (; ra_ptr < ra_max; ra_ptr++)
1485 brelse(bh_use[ra_ptr]);
1486 ext4_fname_free_filename(&fname);
1487 return ret;
1488}
1489
1490static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1491 struct ext4_filename *fname,
1492 struct ext4_dir_entry_2 **res_dir)
1493{
1494 struct super_block * sb = dir->i_sb;
1495 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1496 struct buffer_head *bh;
1497 ext4_lblk_t block;
1498 int retval;
1499
1500#ifdef CONFIG_EXT4_FS_ENCRYPTION
1501 *res_dir = NULL;
1502#endif
1503 frame = dx_probe(fname, dir, NULL, frames);
1504 if (IS_ERR(frame))
1505 return (struct buffer_head *) frame;
1506 do {
1507 block = dx_get_block(frame->at);
1508 bh = ext4_read_dirblock(dir, block, DIRENT);
1509 if (IS_ERR(bh))
1510 goto errout;
1511
1512 retval = search_dirblock(bh, dir, fname,
1513 block << EXT4_BLOCK_SIZE_BITS(sb),
1514 res_dir);
1515 if (retval == 1)
1516 goto success;
1517 brelse(bh);
1518 if (retval == -1) {
1519 bh = ERR_PTR(ERR_BAD_DX_DIR);
1520 goto errout;
1521 }
1522
1523 /* Check to see if we should continue to search */
1524 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1525 frames, NULL);
1526 if (retval < 0) {
1527 ext4_warning_inode(dir,
1528 "error %d reading directory index block",
1529 retval);
1530 bh = ERR_PTR(retval);
1531 goto errout;
1532 }
1533 } while (retval == 1);
1534
1535 bh = NULL;
1536errout:
1537 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1538success:
1539 dx_release(frames);
1540 return bh;
1541}
1542
1543static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1544{
1545 struct inode *inode;
1546 struct ext4_dir_entry_2 *de;
1547 struct buffer_head *bh;
1548
1549 if (ext4_encrypted_inode(dir)) {
1550 int res = fscrypt_get_encryption_info(dir);
1551
1552 /*
1553 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1554 * created while the directory was encrypted and we
1555 * have access to the key.
1556 */
1557 if (fscrypt_has_encryption_key(dir))
1558 fscrypt_set_encrypted_dentry(dentry);
1559 fscrypt_set_d_op(dentry);
1560 if (res && res != -ENOKEY)
1561 return ERR_PTR(res);
1562 }
1563
1564 if (dentry->d_name.len > EXT4_NAME_LEN)
1565 return ERR_PTR(-ENAMETOOLONG);
1566
1567 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1568 if (IS_ERR(bh))
1569 return (struct dentry *) bh;
1570 inode = NULL;
1571 if (bh) {
1572 __u32 ino = le32_to_cpu(de->inode);
1573 brelse(bh);
1574 if (!ext4_valid_inum(dir->i_sb, ino)) {
1575 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1576 return ERR_PTR(-EFSCORRUPTED);
1577 }
1578 if (unlikely(ino == dir->i_ino)) {
1579 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1580 dentry);
1581 return ERR_PTR(-EFSCORRUPTED);
1582 }
1583 inode = ext4_iget_normal(dir->i_sb, ino);
1584 if (inode == ERR_PTR(-ESTALE)) {
1585 EXT4_ERROR_INODE(dir,
1586 "deleted inode referenced: %u",
1587 ino);
1588 return ERR_PTR(-EFSCORRUPTED);
1589 }
1590 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1591 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1592 !fscrypt_has_permitted_context(dir, inode)) {
1593 ext4_warning(inode->i_sb,
1594 "Inconsistent encryption contexts: %lu/%lu",
1595 dir->i_ino, inode->i_ino);
1596 iput(inode);
1597 return ERR_PTR(-EPERM);
1598 }
1599 }
1600 return d_splice_alias(inode, dentry);
1601}
1602
1603
1604struct dentry *ext4_get_parent(struct dentry *child)
1605{
1606 __u32 ino;
1607 static const struct qstr dotdot = QSTR_INIT("..", 2);
1608 struct ext4_dir_entry_2 * de;
1609 struct buffer_head *bh;
1610
1611 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1612 if (IS_ERR(bh))
1613 return (struct dentry *) bh;
1614 if (!bh)
1615 return ERR_PTR(-ENOENT);
1616 ino = le32_to_cpu(de->inode);
1617 brelse(bh);
1618
1619 if (!ext4_valid_inum(child->d_sb, ino)) {
1620 EXT4_ERROR_INODE(d_inode(child),
1621 "bad parent inode number: %u", ino);
1622 return ERR_PTR(-EFSCORRUPTED);
1623 }
1624
1625 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1626}
1627
1628/*
1629 * Move count entries from end of map between two memory locations.
1630 * Returns pointer to last entry moved.
1631 */
1632static struct ext4_dir_entry_2 *
1633dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1634 unsigned blocksize)
1635{
1636 unsigned rec_len = 0;
1637
1638 while (count--) {
1639 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1640 (from + (map->offs<<2));
1641 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1642 memcpy (to, de, rec_len);
1643 ((struct ext4_dir_entry_2 *) to)->rec_len =
1644 ext4_rec_len_to_disk(rec_len, blocksize);
1645 de->inode = 0;
1646 map++;
1647 to += rec_len;
1648 }
1649 return (struct ext4_dir_entry_2 *) (to - rec_len);
1650}
1651
1652/*
1653 * Compact each dir entry in the range to the minimal rec_len.
1654 * Returns pointer to last entry in range.
1655 */
1656static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1657{
1658 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1659 unsigned rec_len = 0;
1660
1661 prev = to = de;
1662 while ((char*)de < base + blocksize) {
1663 next = ext4_next_entry(de, blocksize);
1664 if (de->inode && de->name_len) {
1665 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1666 if (de > to)
1667 memmove(to, de, rec_len);
1668 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1669 prev = to;
1670 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1671 }
1672 de = next;
1673 }
1674 return prev;
1675}
1676
1677/*
1678 * Split a full leaf block to make room for a new dir entry.
1679 * Allocate a new block, and move entries so that they are approx. equally full.
1680 * Returns pointer to de in block into which the new entry will be inserted.
1681 */
1682static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1683 struct buffer_head **bh,struct dx_frame *frame,
1684 struct dx_hash_info *hinfo)
1685{
1686 unsigned blocksize = dir->i_sb->s_blocksize;
1687 unsigned count, continued;
1688 struct buffer_head *bh2;
1689 ext4_lblk_t newblock;
1690 u32 hash2;
1691 struct dx_map_entry *map;
1692 char *data1 = (*bh)->b_data, *data2;
1693 unsigned split, move, size;
1694 struct ext4_dir_entry_2 *de = NULL, *de2;
1695 struct ext4_dir_entry_tail *t;
1696 int csum_size = 0;
1697 int err = 0, i;
1698
1699 if (ext4_has_metadata_csum(dir->i_sb))
1700 csum_size = sizeof(struct ext4_dir_entry_tail);
1701
1702 bh2 = ext4_append(handle, dir, &newblock);
1703 if (IS_ERR(bh2)) {
1704 brelse(*bh);
1705 *bh = NULL;
1706 return (struct ext4_dir_entry_2 *) bh2;
1707 }
1708
1709 BUFFER_TRACE(*bh, "get_write_access");
1710 err = ext4_journal_get_write_access(handle, *bh);
1711 if (err)
1712 goto journal_error;
1713
1714 BUFFER_TRACE(frame->bh, "get_write_access");
1715 err = ext4_journal_get_write_access(handle, frame->bh);
1716 if (err)
1717 goto journal_error;
1718
1719 data2 = bh2->b_data;
1720
1721 /* create map in the end of data2 block */
1722 map = (struct dx_map_entry *) (data2 + blocksize);
1723 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1724 blocksize, hinfo, map);
1725 map -= count;
1726 dx_sort_map(map, count);
1727 /* Split the existing block in the middle, size-wise */
1728 size = 0;
1729 move = 0;
1730 for (i = count-1; i >= 0; i--) {
1731 /* is more than half of this entry in 2nd half of the block? */
1732 if (size + map[i].size/2 > blocksize/2)
1733 break;
1734 size += map[i].size;
1735 move++;
1736 }
1737 /* map index at which we will split */
1738 split = count - move;
1739 hash2 = map[split].hash;
1740 continued = hash2 == map[split - 1].hash;
1741 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1742 (unsigned long)dx_get_block(frame->at),
1743 hash2, split, count-split));
1744
1745 /* Fancy dance to stay within two buffers */
1746 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1747 blocksize);
1748 de = dx_pack_dirents(data1, blocksize);
1749 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1750 (char *) de,
1751 blocksize);
1752 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1753 (char *) de2,
1754 blocksize);
1755 if (csum_size) {
1756 t = EXT4_DIRENT_TAIL(data2, blocksize);
1757 initialize_dirent_tail(t, blocksize);
1758
1759 t = EXT4_DIRENT_TAIL(data1, blocksize);
1760 initialize_dirent_tail(t, blocksize);
1761 }
1762
1763 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1764 blocksize, 1));
1765 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1766 blocksize, 1));
1767
1768 /* Which block gets the new entry? */
1769 if (hinfo->hash >= hash2) {
1770 swap(*bh, bh2);
1771 de = de2;
1772 }
1773 dx_insert_block(frame, hash2 + continued, newblock);
1774 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1775 if (err)
1776 goto journal_error;
1777 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1778 if (err)
1779 goto journal_error;
1780 brelse(bh2);
1781 dxtrace(dx_show_index("frame", frame->entries));
1782 return de;
1783
1784journal_error:
1785 brelse(*bh);
1786 brelse(bh2);
1787 *bh = NULL;
1788 ext4_std_error(dir->i_sb, err);
1789 return ERR_PTR(err);
1790}
1791
1792int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1793 struct buffer_head *bh,
1794 void *buf, int buf_size,
1795 struct ext4_filename *fname,
1796 struct ext4_dir_entry_2 **dest_de)
1797{
1798 struct ext4_dir_entry_2 *de;
1799 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1800 int nlen, rlen;
1801 unsigned int offset = 0;
1802 char *top;
1803
1804 de = (struct ext4_dir_entry_2 *)buf;
1805 top = buf + buf_size - reclen;
1806 while ((char *) de <= top) {
1807 if (ext4_check_dir_entry(dir, NULL, de, bh,
1808 buf, buf_size, offset))
1809 return -EFSCORRUPTED;
1810 if (ext4_match(fname, de))
1811 return -EEXIST;
1812 nlen = EXT4_DIR_REC_LEN(de->name_len);
1813 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1814 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1815 break;
1816 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1817 offset += rlen;
1818 }
1819 if ((char *) de > top)
1820 return -ENOSPC;
1821
1822 *dest_de = de;
1823 return 0;
1824}
1825
1826void ext4_insert_dentry(struct inode *inode,
1827 struct ext4_dir_entry_2 *de,
1828 int buf_size,
1829 struct ext4_filename *fname)
1830{
1831
1832 int nlen, rlen;
1833
1834 nlen = EXT4_DIR_REC_LEN(de->name_len);
1835 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1836 if (de->inode) {
1837 struct ext4_dir_entry_2 *de1 =
1838 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1839 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1840 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1841 de = de1;
1842 }
1843 de->file_type = EXT4_FT_UNKNOWN;
1844 de->inode = cpu_to_le32(inode->i_ino);
1845 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1846 de->name_len = fname_len(fname);
1847 memcpy(de->name, fname_name(fname), fname_len(fname));
1848}
1849
1850/*
1851 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1852 * it points to a directory entry which is guaranteed to be large
1853 * enough for new directory entry. If de is NULL, then
1854 * add_dirent_to_buf will attempt search the directory block for
1855 * space. It will return -ENOSPC if no space is available, and -EIO
1856 * and -EEXIST if directory entry already exists.
1857 */
1858static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1859 struct inode *dir,
1860 struct inode *inode, struct ext4_dir_entry_2 *de,
1861 struct buffer_head *bh)
1862{
1863 unsigned int blocksize = dir->i_sb->s_blocksize;
1864 int csum_size = 0;
1865 int err;
1866
1867 if (ext4_has_metadata_csum(inode->i_sb))
1868 csum_size = sizeof(struct ext4_dir_entry_tail);
1869
1870 if (!de) {
1871 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1872 blocksize - csum_size, fname, &de);
1873 if (err)
1874 return err;
1875 }
1876 BUFFER_TRACE(bh, "get_write_access");
1877 err = ext4_journal_get_write_access(handle, bh);
1878 if (err) {
1879 ext4_std_error(dir->i_sb, err);
1880 return err;
1881 }
1882
1883 /* By now the buffer is marked for journaling */
1884 ext4_insert_dentry(inode, de, blocksize, fname);
1885
1886 /*
1887 * XXX shouldn't update any times until successful
1888 * completion of syscall, but too many callers depend
1889 * on this.
1890 *
1891 * XXX similarly, too many callers depend on
1892 * ext4_new_inode() setting the times, but error
1893 * recovery deletes the inode, so the worst that can
1894 * happen is that the times are slightly out of date
1895 * and/or different from the directory change time.
1896 */
1897 dir->i_mtime = dir->i_ctime = current_time(dir);
1898 ext4_update_dx_flag(dir);
1899 inode_inc_iversion(dir);
1900 ext4_mark_inode_dirty(handle, dir);
1901 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1902 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1903 if (err)
1904 ext4_std_error(dir->i_sb, err);
1905 return 0;
1906}
1907
1908/*
1909 * This converts a one block unindexed directory to a 3 block indexed
1910 * directory, and adds the dentry to the indexed directory.
1911 */
1912static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1913 struct inode *dir,
1914 struct inode *inode, struct buffer_head *bh)
1915{
1916 struct buffer_head *bh2;
1917 struct dx_root *root;
1918 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1919 struct dx_entry *entries;
1920 struct ext4_dir_entry_2 *de, *de2;
1921 struct ext4_dir_entry_tail *t;
1922 char *data1, *top;
1923 unsigned len;
1924 int retval;
1925 unsigned blocksize;
1926 ext4_lblk_t block;
1927 struct fake_dirent *fde;
1928 int csum_size = 0;
1929
1930 if (ext4_has_metadata_csum(inode->i_sb))
1931 csum_size = sizeof(struct ext4_dir_entry_tail);
1932
1933 blocksize = dir->i_sb->s_blocksize;
1934 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1935 BUFFER_TRACE(bh, "get_write_access");
1936 retval = ext4_journal_get_write_access(handle, bh);
1937 if (retval) {
1938 ext4_std_error(dir->i_sb, retval);
1939 brelse(bh);
1940 return retval;
1941 }
1942 root = (struct dx_root *) bh->b_data;
1943
1944 /* The 0th block becomes the root, move the dirents out */
1945 fde = &root->dotdot;
1946 de = (struct ext4_dir_entry_2 *)((char *)fde +
1947 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1948 if ((char *) de >= (((char *) root) + blocksize)) {
1949 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1950 brelse(bh);
1951 return -EFSCORRUPTED;
1952 }
1953 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1954
1955 /* Allocate new block for the 0th block's dirents */
1956 bh2 = ext4_append(handle, dir, &block);
1957 if (IS_ERR(bh2)) {
1958 brelse(bh);
1959 return PTR_ERR(bh2);
1960 }
1961 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1962 data1 = bh2->b_data;
1963
1964 memcpy (data1, de, len);
1965 de = (struct ext4_dir_entry_2 *) data1;
1966 top = data1 + len;
1967 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1968 de = de2;
1969 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1970 (char *) de,
1971 blocksize);
1972
1973 if (csum_size) {
1974 t = EXT4_DIRENT_TAIL(data1, blocksize);
1975 initialize_dirent_tail(t, blocksize);
1976 }
1977
1978 /* Initialize the root; the dot dirents already exist */
1979 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1980 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1981 blocksize);
1982 memset (&root->info, 0, sizeof(root->info));
1983 root->info.info_length = sizeof(root->info);
1984 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1985 entries = root->entries;
1986 dx_set_block(entries, 1);
1987 dx_set_count(entries, 1);
1988 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1989
1990 /* Initialize as for dx_probe */
1991 fname->hinfo.hash_version = root->info.hash_version;
1992 if (fname->hinfo.hash_version <= DX_HASH_TEA)
1993 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1994 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1995 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1996
1997 memset(frames, 0, sizeof(frames));
1998 frame = frames;
1999 frame->entries = entries;
2000 frame->at = entries;
2001 frame->bh = bh;
2002
2003 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2004 if (retval)
2005 goto out_frames;
2006 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2007 if (retval)
2008 goto out_frames;
2009
2010 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2011 if (IS_ERR(de)) {
2012 retval = PTR_ERR(de);
2013 goto out_frames;
2014 }
2015
2016 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2017out_frames:
2018 /*
2019 * Even if the block split failed, we have to properly write
2020 * out all the changes we did so far. Otherwise we can end up
2021 * with corrupted filesystem.
2022 */
2023 if (retval)
2024 ext4_mark_inode_dirty(handle, dir);
2025 dx_release(frames);
2026 brelse(bh2);
2027 return retval;
2028}
2029
2030/*
2031 * ext4_add_entry()
2032 *
2033 * adds a file entry to the specified directory, using the same
2034 * semantics as ext4_find_entry(). It returns NULL if it failed.
2035 *
2036 * NOTE!! The inode part of 'de' is left at 0 - which means you
2037 * may not sleep between calling this and putting something into
2038 * the entry, as someone else might have used it while you slept.
2039 */
2040static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2041 struct inode *inode)
2042{
2043 struct inode *dir = d_inode(dentry->d_parent);
2044 struct buffer_head *bh = NULL;
2045 struct ext4_dir_entry_2 *de;
2046 struct ext4_dir_entry_tail *t;
2047 struct super_block *sb;
2048 struct ext4_filename fname;
2049 int retval;
2050 int dx_fallback=0;
2051 unsigned blocksize;
2052 ext4_lblk_t block, blocks;
2053 int csum_size = 0;
2054
2055 if (ext4_has_metadata_csum(inode->i_sb))
2056 csum_size = sizeof(struct ext4_dir_entry_tail);
2057
2058 sb = dir->i_sb;
2059 blocksize = sb->s_blocksize;
2060 if (!dentry->d_name.len)
2061 return -EINVAL;
2062
2063 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2064 if (retval)
2065 return retval;
2066
2067 if (ext4_has_inline_data(dir)) {
2068 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2069 if (retval < 0)
2070 goto out;
2071 if (retval == 1) {
2072 retval = 0;
2073 goto out;
2074 }
2075 }
2076
2077 if (is_dx(dir)) {
2078 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2079 if (!retval || (retval != ERR_BAD_DX_DIR))
2080 goto out;
2081 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2082 dx_fallback++;
2083 ext4_mark_inode_dirty(handle, dir);
2084 }
2085 blocks = dir->i_size >> sb->s_blocksize_bits;
2086 for (block = 0; block < blocks; block++) {
2087 bh = ext4_read_dirblock(dir, block, DIRENT);
2088 if (IS_ERR(bh)) {
2089 retval = PTR_ERR(bh);
2090 bh = NULL;
2091 goto out;
2092 }
2093 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2094 NULL, bh);
2095 if (retval != -ENOSPC)
2096 goto out;
2097
2098 if (blocks == 1 && !dx_fallback &&
2099 ext4_has_feature_dir_index(sb)) {
2100 retval = make_indexed_dir(handle, &fname, dir,
2101 inode, bh);
2102 bh = NULL; /* make_indexed_dir releases bh */
2103 goto out;
2104 }
2105 brelse(bh);
2106 }
2107 bh = ext4_append(handle, dir, &block);
2108 if (IS_ERR(bh)) {
2109 retval = PTR_ERR(bh);
2110 bh = NULL;
2111 goto out;
2112 }
2113 de = (struct ext4_dir_entry_2 *) bh->b_data;
2114 de->inode = 0;
2115 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2116
2117 if (csum_size) {
2118 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2119 initialize_dirent_tail(t, blocksize);
2120 }
2121
2122 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2123out:
2124 ext4_fname_free_filename(&fname);
2125 brelse(bh);
2126 if (retval == 0)
2127 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2128 return retval;
2129}
2130
2131/*
2132 * Returns 0 for success, or a negative error value
2133 */
2134static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2135 struct inode *dir, struct inode *inode)
2136{
2137 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2138 struct dx_entry *entries, *at;
2139 struct buffer_head *bh;
2140 struct super_block *sb = dir->i_sb;
2141 struct ext4_dir_entry_2 *de;
2142 int restart;
2143 int err;
2144
2145again:
2146 restart = 0;
2147 frame = dx_probe(fname, dir, NULL, frames);
2148 if (IS_ERR(frame))
2149 return PTR_ERR(frame);
2150 entries = frame->entries;
2151 at = frame->at;
2152 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2153 if (IS_ERR(bh)) {
2154 err = PTR_ERR(bh);
2155 bh = NULL;
2156 goto cleanup;
2157 }
2158
2159 BUFFER_TRACE(bh, "get_write_access");
2160 err = ext4_journal_get_write_access(handle, bh);
2161 if (err)
2162 goto journal_error;
2163
2164 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2165 if (err != -ENOSPC)
2166 goto cleanup;
2167
2168 err = 0;
2169 /* Block full, should compress but for now just split */
2170 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2171 dx_get_count(entries), dx_get_limit(entries)));
2172 /* Need to split index? */
2173 if (dx_get_count(entries) == dx_get_limit(entries)) {
2174 ext4_lblk_t newblock;
2175 int levels = frame - frames + 1;
2176 unsigned int icount;
2177 int add_level = 1;
2178 struct dx_entry *entries2;
2179 struct dx_node *node2;
2180 struct buffer_head *bh2;
2181
2182 while (frame > frames) {
2183 if (dx_get_count((frame - 1)->entries) <
2184 dx_get_limit((frame - 1)->entries)) {
2185 add_level = 0;
2186 break;
2187 }
2188 frame--; /* split higher index block */
2189 at = frame->at;
2190 entries = frame->entries;
2191 restart = 1;
2192 }
2193 if (add_level && levels == ext4_dir_htree_level(sb)) {
2194 ext4_warning(sb, "Directory (ino: %lu) index full, "
2195 "reach max htree level :%d",
2196 dir->i_ino, levels);
2197 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2198 ext4_warning(sb, "Large directory feature is "
2199 "not enabled on this "
2200 "filesystem");
2201 }
2202 err = -ENOSPC;
2203 goto cleanup;
2204 }
2205 icount = dx_get_count(entries);
2206 bh2 = ext4_append(handle, dir, &newblock);
2207 if (IS_ERR(bh2)) {
2208 err = PTR_ERR(bh2);
2209 goto cleanup;
2210 }
2211 node2 = (struct dx_node *)(bh2->b_data);
2212 entries2 = node2->entries;
2213 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2214 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2215 sb->s_blocksize);
2216 BUFFER_TRACE(frame->bh, "get_write_access");
2217 err = ext4_journal_get_write_access(handle, frame->bh);
2218 if (err)
2219 goto journal_error;
2220 if (!add_level) {
2221 unsigned icount1 = icount/2, icount2 = icount - icount1;
2222 unsigned hash2 = dx_get_hash(entries + icount1);
2223 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2224 icount1, icount2));
2225
2226 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2227 err = ext4_journal_get_write_access(handle,
2228 (frame - 1)->bh);
2229 if (err)
2230 goto journal_error;
2231
2232 memcpy((char *) entries2, (char *) (entries + icount1),
2233 icount2 * sizeof(struct dx_entry));
2234 dx_set_count(entries, icount1);
2235 dx_set_count(entries2, icount2);
2236 dx_set_limit(entries2, dx_node_limit(dir));
2237
2238 /* Which index block gets the new entry? */
2239 if (at - entries >= icount1) {
2240 frame->at = at = at - entries - icount1 + entries2;
2241 frame->entries = entries = entries2;
2242 swap(frame->bh, bh2);
2243 }
2244 dx_insert_block((frame - 1), hash2, newblock);
2245 dxtrace(dx_show_index("node", frame->entries));
2246 dxtrace(dx_show_index("node",
2247 ((struct dx_node *) bh2->b_data)->entries));
2248 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2249 if (err)
2250 goto journal_error;
2251 brelse (bh2);
2252 err = ext4_handle_dirty_dx_node(handle, dir,
2253 (frame - 1)->bh);
2254 if (err)
2255 goto journal_error;
2256 if (restart) {
2257 err = ext4_handle_dirty_dx_node(handle, dir,
2258 frame->bh);
2259 goto journal_error;
2260 }
2261 } else {
2262 struct dx_root *dxroot;
2263 memcpy((char *) entries2, (char *) entries,
2264 icount * sizeof(struct dx_entry));
2265 dx_set_limit(entries2, dx_node_limit(dir));
2266
2267 /* Set up root */
2268 dx_set_count(entries, 1);
2269 dx_set_block(entries + 0, newblock);
2270 dxroot = (struct dx_root *)frames[0].bh->b_data;
2271 dxroot->info.indirect_levels += 1;
2272 dxtrace(printk(KERN_DEBUG
2273 "Creating %d level index...\n",
2274 info->indirect_levels));
2275 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2276 if (err)
2277 goto journal_error;
2278 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2279 brelse(bh2);
2280 restart = 1;
2281 goto journal_error;
2282 }
2283 }
2284 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2285 if (IS_ERR(de)) {
2286 err = PTR_ERR(de);
2287 goto cleanup;
2288 }
2289 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2290 goto cleanup;
2291
2292journal_error:
2293 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2294cleanup:
2295 brelse(bh);
2296 dx_release(frames);
2297 /* @restart is true means htree-path has been changed, we need to
2298 * repeat dx_probe() to find out valid htree-path
2299 */
2300 if (restart && err == 0)
2301 goto again;
2302 return err;
2303}
2304
2305/*
2306 * ext4_generic_delete_entry deletes a directory entry by merging it
2307 * with the previous entry
2308 */
2309int ext4_generic_delete_entry(handle_t *handle,
2310 struct inode *dir,
2311 struct ext4_dir_entry_2 *de_del,
2312 struct buffer_head *bh,
2313 void *entry_buf,
2314 int buf_size,
2315 int csum_size)
2316{
2317 struct ext4_dir_entry_2 *de, *pde;
2318 unsigned int blocksize = dir->i_sb->s_blocksize;
2319 int i;
2320
2321 i = 0;
2322 pde = NULL;
2323 de = (struct ext4_dir_entry_2 *)entry_buf;
2324 while (i < buf_size - csum_size) {
2325 if (ext4_check_dir_entry(dir, NULL, de, bh,
2326 bh->b_data, bh->b_size, i))
2327 return -EFSCORRUPTED;
2328 if (de == de_del) {
2329 if (pde)
2330 pde->rec_len = ext4_rec_len_to_disk(
2331 ext4_rec_len_from_disk(pde->rec_len,
2332 blocksize) +
2333 ext4_rec_len_from_disk(de->rec_len,
2334 blocksize),
2335 blocksize);
2336 else
2337 de->inode = 0;
2338 inode_inc_iversion(dir);
2339 return 0;
2340 }
2341 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2342 pde = de;
2343 de = ext4_next_entry(de, blocksize);
2344 }
2345 return -ENOENT;
2346}
2347
2348static int ext4_delete_entry(handle_t *handle,
2349 struct inode *dir,
2350 struct ext4_dir_entry_2 *de_del,
2351 struct buffer_head *bh)
2352{
2353 int err, csum_size = 0;
2354
2355 if (ext4_has_inline_data(dir)) {
2356 int has_inline_data = 1;
2357 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2358 &has_inline_data);
2359 if (has_inline_data)
2360 return err;
2361 }
2362
2363 if (ext4_has_metadata_csum(dir->i_sb))
2364 csum_size = sizeof(struct ext4_dir_entry_tail);
2365
2366 BUFFER_TRACE(bh, "get_write_access");
2367 err = ext4_journal_get_write_access(handle, bh);
2368 if (unlikely(err))
2369 goto out;
2370
2371 err = ext4_generic_delete_entry(handle, dir, de_del,
2372 bh, bh->b_data,
2373 dir->i_sb->s_blocksize, csum_size);
2374 if (err)
2375 goto out;
2376
2377 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2378 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2379 if (unlikely(err))
2380 goto out;
2381
2382 return 0;
2383out:
2384 if (err != -ENOENT)
2385 ext4_std_error(dir->i_sb, err);
2386 return err;
2387}
2388
2389/*
2390 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2391 * since this indicates that nlinks count was previously 1 to avoid overflowing
2392 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2393 * that subdirectory link counts are not being maintained accurately.
2394 *
2395 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2396 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2397 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2398 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2399 */
2400static void ext4_inc_count(handle_t *handle, struct inode *inode)
2401{
2402 inc_nlink(inode);
2403 if (is_dx(inode) &&
2404 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2405 set_nlink(inode, 1);
2406}
2407
2408/*
2409 * If a directory had nlink == 1, then we should let it be 1. This indicates
2410 * directory has >EXT4_LINK_MAX subdirs.
2411 */
2412static void ext4_dec_count(handle_t *handle, struct inode *inode)
2413{
2414 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2415 drop_nlink(inode);
2416}
2417
2418
2419static int ext4_add_nondir(handle_t *handle,
2420 struct dentry *dentry, struct inode *inode)
2421{
2422 int err = ext4_add_entry(handle, dentry, inode);
2423 if (!err) {
2424 ext4_mark_inode_dirty(handle, inode);
2425 d_instantiate_new(dentry, inode);
2426 return 0;
2427 }
2428 drop_nlink(inode);
2429 unlock_new_inode(inode);
2430 iput(inode);
2431 return err;
2432}
2433
2434/*
2435 * By the time this is called, we already have created
2436 * the directory cache entry for the new file, but it
2437 * is so far negative - it has no inode.
2438 *
2439 * If the create succeeds, we fill in the inode information
2440 * with d_instantiate().
2441 */
2442static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2443 bool excl)
2444{
2445 handle_t *handle;
2446 struct inode *inode;
2447 int err, credits, retries = 0;
2448
2449 err = dquot_initialize(dir);
2450 if (err)
2451 return err;
2452
2453 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2454 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2455retry:
2456 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2457 NULL, EXT4_HT_DIR, credits);
2458 handle = ext4_journal_current_handle();
2459 err = PTR_ERR(inode);
2460 if (!IS_ERR(inode)) {
2461 inode->i_op = &ext4_file_inode_operations;
2462 inode->i_fop = &ext4_file_operations;
2463 ext4_set_aops(inode);
2464 err = ext4_add_nondir(handle, dentry, inode);
2465 if (!err && IS_DIRSYNC(dir))
2466 ext4_handle_sync(handle);
2467 }
2468 if (handle)
2469 ext4_journal_stop(handle);
2470 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2471 goto retry;
2472 return err;
2473}
2474
2475static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2476 umode_t mode, dev_t rdev)
2477{
2478 handle_t *handle;
2479 struct inode *inode;
2480 int err, credits, retries = 0;
2481
2482 err = dquot_initialize(dir);
2483 if (err)
2484 return err;
2485
2486 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2487 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2488retry:
2489 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2490 NULL, EXT4_HT_DIR, credits);
2491 handle = ext4_journal_current_handle();
2492 err = PTR_ERR(inode);
2493 if (!IS_ERR(inode)) {
2494 init_special_inode(inode, inode->i_mode, rdev);
2495 inode->i_op = &ext4_special_inode_operations;
2496 err = ext4_add_nondir(handle, dentry, inode);
2497 if (!err && IS_DIRSYNC(dir))
2498 ext4_handle_sync(handle);
2499 }
2500 if (handle)
2501 ext4_journal_stop(handle);
2502 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2503 goto retry;
2504 return err;
2505}
2506
2507static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2508{
2509 handle_t *handle;
2510 struct inode *inode;
2511 int err, retries = 0;
2512
2513 err = dquot_initialize(dir);
2514 if (err)
2515 return err;
2516
2517retry:
2518 inode = ext4_new_inode_start_handle(dir, mode,
2519 NULL, 0, NULL,
2520 EXT4_HT_DIR,
2521 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2522 4 + EXT4_XATTR_TRANS_BLOCKS);
2523 handle = ext4_journal_current_handle();
2524 err = PTR_ERR(inode);
2525 if (!IS_ERR(inode)) {
2526 inode->i_op = &ext4_file_inode_operations;
2527 inode->i_fop = &ext4_file_operations;
2528 ext4_set_aops(inode);
2529 d_tmpfile(dentry, inode);
2530 err = ext4_orphan_add(handle, inode);
2531 if (err)
2532 goto err_unlock_inode;
2533 mark_inode_dirty(inode);
2534 unlock_new_inode(inode);
2535 }
2536 if (handle)
2537 ext4_journal_stop(handle);
2538 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2539 goto retry;
2540 return err;
2541err_unlock_inode:
2542 ext4_journal_stop(handle);
2543 unlock_new_inode(inode);
2544 return err;
2545}
2546
2547struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2548 struct ext4_dir_entry_2 *de,
2549 int blocksize, int csum_size,
2550 unsigned int parent_ino, int dotdot_real_len)
2551{
2552 de->inode = cpu_to_le32(inode->i_ino);
2553 de->name_len = 1;
2554 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2555 blocksize);
2556 strcpy(de->name, ".");
2557 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2558
2559 de = ext4_next_entry(de, blocksize);
2560 de->inode = cpu_to_le32(parent_ino);
2561 de->name_len = 2;
2562 if (!dotdot_real_len)
2563 de->rec_len = ext4_rec_len_to_disk(blocksize -
2564 (csum_size + EXT4_DIR_REC_LEN(1)),
2565 blocksize);
2566 else
2567 de->rec_len = ext4_rec_len_to_disk(
2568 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2569 strcpy(de->name, "..");
2570 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2571
2572 return ext4_next_entry(de, blocksize);
2573}
2574
2575static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2576 struct inode *inode)
2577{
2578 struct buffer_head *dir_block = NULL;
2579 struct ext4_dir_entry_2 *de;
2580 struct ext4_dir_entry_tail *t;
2581 ext4_lblk_t block = 0;
2582 unsigned int blocksize = dir->i_sb->s_blocksize;
2583 int csum_size = 0;
2584 int err;
2585
2586 if (ext4_has_metadata_csum(dir->i_sb))
2587 csum_size = sizeof(struct ext4_dir_entry_tail);
2588
2589 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2590 err = ext4_try_create_inline_dir(handle, dir, inode);
2591 if (err < 0 && err != -ENOSPC)
2592 goto out;
2593 if (!err)
2594 goto out;
2595 }
2596
2597 inode->i_size = 0;
2598 dir_block = ext4_append(handle, inode, &block);
2599 if (IS_ERR(dir_block))
2600 return PTR_ERR(dir_block);
2601 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2602 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2603 set_nlink(inode, 2);
2604 if (csum_size) {
2605 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2606 initialize_dirent_tail(t, blocksize);
2607 }
2608
2609 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2610 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2611 if (err)
2612 goto out;
2613 set_buffer_verified(dir_block);
2614out:
2615 brelse(dir_block);
2616 return err;
2617}
2618
2619static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2620{
2621 handle_t *handle;
2622 struct inode *inode;
2623 int err, credits, retries = 0;
2624
2625 if (EXT4_DIR_LINK_MAX(dir))
2626 return -EMLINK;
2627
2628 err = dquot_initialize(dir);
2629 if (err)
2630 return err;
2631
2632 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2633 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2634retry:
2635 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2636 &dentry->d_name,
2637 0, NULL, EXT4_HT_DIR, credits);
2638 handle = ext4_journal_current_handle();
2639 err = PTR_ERR(inode);
2640 if (IS_ERR(inode))
2641 goto out_stop;
2642
2643 inode->i_op = &ext4_dir_inode_operations;
2644 inode->i_fop = &ext4_dir_operations;
2645 err = ext4_init_new_dir(handle, dir, inode);
2646 if (err)
2647 goto out_clear_inode;
2648 err = ext4_mark_inode_dirty(handle, inode);
2649 if (!err)
2650 err = ext4_add_entry(handle, dentry, inode);
2651 if (err) {
2652out_clear_inode:
2653 clear_nlink(inode);
2654 unlock_new_inode(inode);
2655 ext4_mark_inode_dirty(handle, inode);
2656 iput(inode);
2657 goto out_stop;
2658 }
2659 ext4_inc_count(handle, dir);
2660 ext4_update_dx_flag(dir);
2661 err = ext4_mark_inode_dirty(handle, dir);
2662 if (err)
2663 goto out_clear_inode;
2664 d_instantiate_new(dentry, inode);
2665 if (IS_DIRSYNC(dir))
2666 ext4_handle_sync(handle);
2667
2668out_stop:
2669 if (handle)
2670 ext4_journal_stop(handle);
2671 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2672 goto retry;
2673 return err;
2674}
2675
2676/*
2677 * routine to check that the specified directory is empty (for rmdir)
2678 */
2679bool ext4_empty_dir(struct inode *inode)
2680{
2681 unsigned int offset;
2682 struct buffer_head *bh;
2683 struct ext4_dir_entry_2 *de, *de1;
2684 struct super_block *sb;
2685
2686 if (ext4_has_inline_data(inode)) {
2687 int has_inline_data = 1;
2688 int ret;
2689
2690 ret = empty_inline_dir(inode, &has_inline_data);
2691 if (has_inline_data)
2692 return ret;
2693 }
2694
2695 sb = inode->i_sb;
2696 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2697 EXT4_ERROR_INODE(inode, "invalid size");
2698 return true;
2699 }
2700 bh = ext4_read_dirblock(inode, 0, EITHER);
2701 if (IS_ERR(bh))
2702 return true;
2703
2704 de = (struct ext4_dir_entry_2 *) bh->b_data;
2705 de1 = ext4_next_entry(de, sb->s_blocksize);
2706 if (le32_to_cpu(de->inode) != inode->i_ino ||
2707 le32_to_cpu(de1->inode) == 0 ||
2708 strcmp(".", de->name) || strcmp("..", de1->name)) {
2709 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2710 brelse(bh);
2711 return true;
2712 }
2713 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2714 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2715 de = ext4_next_entry(de1, sb->s_blocksize);
2716 while (offset < inode->i_size) {
2717 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2718 unsigned int lblock;
2719 brelse(bh);
2720 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2721 bh = ext4_read_dirblock(inode, lblock, EITHER);
2722 if (IS_ERR(bh))
2723 return true;
2724 de = (struct ext4_dir_entry_2 *) bh->b_data;
2725 }
2726 if (ext4_check_dir_entry(inode, NULL, de, bh,
2727 bh->b_data, bh->b_size, offset)) {
2728 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2729 sb->s_blocksize);
2730 offset = (offset | (sb->s_blocksize - 1)) + 1;
2731 continue;
2732 }
2733 if (le32_to_cpu(de->inode)) {
2734 brelse(bh);
2735 return false;
2736 }
2737 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2738 de = ext4_next_entry(de, sb->s_blocksize);
2739 }
2740 brelse(bh);
2741 return true;
2742}
2743
2744/*
2745 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2746 * such inodes, starting at the superblock, in case we crash before the
2747 * file is closed/deleted, or in case the inode truncate spans multiple
2748 * transactions and the last transaction is not recovered after a crash.
2749 *
2750 * At filesystem recovery time, we walk this list deleting unlinked
2751 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2752 *
2753 * Orphan list manipulation functions must be called under i_mutex unless
2754 * we are just creating the inode or deleting it.
2755 */
2756int ext4_orphan_add(handle_t *handle, struct inode *inode)
2757{
2758 struct super_block *sb = inode->i_sb;
2759 struct ext4_sb_info *sbi = EXT4_SB(sb);
2760 struct ext4_iloc iloc;
2761 int err = 0, rc;
2762 bool dirty = false;
2763
2764 if (!sbi->s_journal || is_bad_inode(inode))
2765 return 0;
2766
2767 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2768 !inode_is_locked(inode));
2769 /*
2770 * Exit early if inode already is on orphan list. This is a big speedup
2771 * since we don't have to contend on the global s_orphan_lock.
2772 */
2773 if (!list_empty(&EXT4_I(inode)->i_orphan))
2774 return 0;
2775
2776 /*
2777 * Orphan handling is only valid for files with data blocks
2778 * being truncated, or files being unlinked. Note that we either
2779 * hold i_mutex, or the inode can not be referenced from outside,
2780 * so i_nlink should not be bumped due to race
2781 */
2782 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2783 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2784
2785 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2786 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2787 if (err)
2788 goto out;
2789
2790 err = ext4_reserve_inode_write(handle, inode, &iloc);
2791 if (err)
2792 goto out;
2793
2794 mutex_lock(&sbi->s_orphan_lock);
2795 /*
2796 * Due to previous errors inode may be already a part of on-disk
2797 * orphan list. If so skip on-disk list modification.
2798 */
2799 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2800 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2801 /* Insert this inode at the head of the on-disk orphan list */
2802 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2803 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2804 dirty = true;
2805 }
2806 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2807 mutex_unlock(&sbi->s_orphan_lock);
2808
2809 if (dirty) {
2810 err = ext4_handle_dirty_super(handle, sb);
2811 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2812 if (!err)
2813 err = rc;
2814 if (err) {
2815 /*
2816 * We have to remove inode from in-memory list if
2817 * addition to on disk orphan list failed. Stray orphan
2818 * list entries can cause panics at unmount time.
2819 */
2820 mutex_lock(&sbi->s_orphan_lock);
2821 list_del_init(&EXT4_I(inode)->i_orphan);
2822 mutex_unlock(&sbi->s_orphan_lock);
2823 }
2824 } else
2825 brelse(iloc.bh);
2826
2827 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2828 jbd_debug(4, "orphan inode %lu will point to %d\n",
2829 inode->i_ino, NEXT_ORPHAN(inode));
2830out:
2831 ext4_std_error(sb, err);
2832 return err;
2833}
2834
2835/*
2836 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2837 * of such inodes stored on disk, because it is finally being cleaned up.
2838 */
2839int ext4_orphan_del(handle_t *handle, struct inode *inode)
2840{
2841 struct list_head *prev;
2842 struct ext4_inode_info *ei = EXT4_I(inode);
2843 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2844 __u32 ino_next;
2845 struct ext4_iloc iloc;
2846 int err = 0;
2847
2848 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2849 return 0;
2850
2851 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2852 !inode_is_locked(inode));
2853 /* Do this quick check before taking global s_orphan_lock. */
2854 if (list_empty(&ei->i_orphan))
2855 return 0;
2856
2857 if (handle) {
2858 /* Grab inode buffer early before taking global s_orphan_lock */
2859 err = ext4_reserve_inode_write(handle, inode, &iloc);
2860 }
2861
2862 mutex_lock(&sbi->s_orphan_lock);
2863 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2864
2865 prev = ei->i_orphan.prev;
2866 list_del_init(&ei->i_orphan);
2867
2868 /* If we're on an error path, we may not have a valid
2869 * transaction handle with which to update the orphan list on
2870 * disk, but we still need to remove the inode from the linked
2871 * list in memory. */
2872 if (!handle || err) {
2873 mutex_unlock(&sbi->s_orphan_lock);
2874 goto out_err;
2875 }
2876
2877 ino_next = NEXT_ORPHAN(inode);
2878 if (prev == &sbi->s_orphan) {
2879 jbd_debug(4, "superblock will point to %u\n", ino_next);
2880 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2881 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2882 if (err) {
2883 mutex_unlock(&sbi->s_orphan_lock);
2884 goto out_brelse;
2885 }
2886 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2887 mutex_unlock(&sbi->s_orphan_lock);
2888 err = ext4_handle_dirty_super(handle, inode->i_sb);
2889 } else {
2890 struct ext4_iloc iloc2;
2891 struct inode *i_prev =
2892 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2893
2894 jbd_debug(4, "orphan inode %lu will point to %u\n",
2895 i_prev->i_ino, ino_next);
2896 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2897 if (err) {
2898 mutex_unlock(&sbi->s_orphan_lock);
2899 goto out_brelse;
2900 }
2901 NEXT_ORPHAN(i_prev) = ino_next;
2902 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2903 mutex_unlock(&sbi->s_orphan_lock);
2904 }
2905 if (err)
2906 goto out_brelse;
2907 NEXT_ORPHAN(inode) = 0;
2908 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2909out_err:
2910 ext4_std_error(inode->i_sb, err);
2911 return err;
2912
2913out_brelse:
2914 brelse(iloc.bh);
2915 goto out_err;
2916}
2917
2918static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2919{
2920 int retval;
2921 struct inode *inode;
2922 struct buffer_head *bh;
2923 struct ext4_dir_entry_2 *de;
2924 handle_t *handle = NULL;
2925
2926 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2927 return -EIO;
2928
2929 /* Initialize quotas before so that eventual writes go in
2930 * separate transaction */
2931 retval = dquot_initialize(dir);
2932 if (retval)
2933 return retval;
2934 retval = dquot_initialize(d_inode(dentry));
2935 if (retval)
2936 return retval;
2937
2938 retval = -ENOENT;
2939 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2940 if (IS_ERR(bh))
2941 return PTR_ERR(bh);
2942 if (!bh)
2943 goto end_rmdir;
2944
2945 inode = d_inode(dentry);
2946
2947 retval = -EFSCORRUPTED;
2948 if (le32_to_cpu(de->inode) != inode->i_ino)
2949 goto end_rmdir;
2950
2951 retval = -ENOTEMPTY;
2952 if (!ext4_empty_dir(inode))
2953 goto end_rmdir;
2954
2955 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2956 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2957 if (IS_ERR(handle)) {
2958 retval = PTR_ERR(handle);
2959 handle = NULL;
2960 goto end_rmdir;
2961 }
2962
2963 if (IS_DIRSYNC(dir))
2964 ext4_handle_sync(handle);
2965
2966 retval = ext4_delete_entry(handle, dir, de, bh);
2967 if (retval)
2968 goto end_rmdir;
2969 if (!EXT4_DIR_LINK_EMPTY(inode))
2970 ext4_warning_inode(inode,
2971 "empty directory '%.*s' has too many links (%u)",
2972 dentry->d_name.len, dentry->d_name.name,
2973 inode->i_nlink);
2974 inode->i_version++;
2975 clear_nlink(inode);
2976 /* There's no need to set i_disksize: the fact that i_nlink is
2977 * zero will ensure that the right thing happens during any
2978 * recovery. */
2979 inode->i_size = 0;
2980 ext4_orphan_add(handle, inode);
2981 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2982 ext4_mark_inode_dirty(handle, inode);
2983 ext4_dec_count(handle, dir);
2984 ext4_update_dx_flag(dir);
2985 ext4_mark_inode_dirty(handle, dir);
2986
2987end_rmdir:
2988 brelse(bh);
2989 if (handle)
2990 ext4_journal_stop(handle);
2991 return retval;
2992}
2993
2994static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2995{
2996 int retval;
2997 struct inode *inode;
2998 struct buffer_head *bh;
2999 struct ext4_dir_entry_2 *de;
3000 handle_t *handle = NULL;
3001
3002 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3003 return -EIO;
3004
3005 trace_ext4_unlink_enter(dir, dentry);
3006 /* Initialize quotas before so that eventual writes go
3007 * in separate transaction */
3008 retval = dquot_initialize(dir);
3009 if (retval)
3010 return retval;
3011 retval = dquot_initialize(d_inode(dentry));
3012 if (retval)
3013 return retval;
3014
3015 retval = -ENOENT;
3016 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3017 if (IS_ERR(bh))
3018 return PTR_ERR(bh);
3019 if (!bh)
3020 goto end_unlink;
3021
3022 inode = d_inode(dentry);
3023
3024 retval = -EFSCORRUPTED;
3025 if (le32_to_cpu(de->inode) != inode->i_ino)
3026 goto end_unlink;
3027
3028 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3029 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3030 if (IS_ERR(handle)) {
3031 retval = PTR_ERR(handle);
3032 handle = NULL;
3033 goto end_unlink;
3034 }
3035
3036 if (IS_DIRSYNC(dir))
3037 ext4_handle_sync(handle);
3038
3039 if (inode->i_nlink == 0) {
3040 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3041 dentry->d_name.len, dentry->d_name.name);
3042 set_nlink(inode, 1);
3043 }
3044 retval = ext4_delete_entry(handle, dir, de, bh);
3045 if (retval)
3046 goto end_unlink;
3047 dir->i_ctime = dir->i_mtime = current_time(dir);
3048 ext4_update_dx_flag(dir);
3049 ext4_mark_inode_dirty(handle, dir);
3050 drop_nlink(inode);
3051 if (!inode->i_nlink)
3052 ext4_orphan_add(handle, inode);
3053 inode->i_ctime = current_time(inode);
3054 ext4_mark_inode_dirty(handle, inode);
3055
3056end_unlink:
3057 brelse(bh);
3058 if (handle)
3059 ext4_journal_stop(handle);
3060 trace_ext4_unlink_exit(dentry, retval);
3061 return retval;
3062}
3063
3064static int ext4_symlink(struct inode *dir,
3065 struct dentry *dentry, const char *symname)
3066{
3067 handle_t *handle;
3068 struct inode *inode;
3069 int err, len = strlen(symname);
3070 int credits;
3071 struct fscrypt_str disk_link;
3072
3073 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3074 return -EIO;
3075
3076 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3077 &disk_link);
3078 if (err)
3079 return err;
3080
3081 err = dquot_initialize(dir);
3082 if (err)
3083 return err;
3084
3085 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3086 /*
3087 * For non-fast symlinks, we just allocate inode and put it on
3088 * orphan list in the first transaction => we need bitmap,
3089 * group descriptor, sb, inode block, quota blocks, and
3090 * possibly selinux xattr blocks.
3091 */
3092 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3093 EXT4_XATTR_TRANS_BLOCKS;
3094 } else {
3095 /*
3096 * Fast symlink. We have to add entry to directory
3097 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3098 * allocate new inode (bitmap, group descriptor, inode block,
3099 * quota blocks, sb is already counted in previous macros).
3100 */
3101 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3102 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3103 }
3104
3105 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3106 &dentry->d_name, 0, NULL,
3107 EXT4_HT_DIR, credits);
3108 handle = ext4_journal_current_handle();
3109 if (IS_ERR(inode)) {
3110 if (handle)
3111 ext4_journal_stop(handle);
3112 return PTR_ERR(inode);
3113 }
3114
3115 if (IS_ENCRYPTED(inode)) {
3116 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3117 if (err)
3118 goto err_drop_inode;
3119 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3120 }
3121
3122 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3123 if (!IS_ENCRYPTED(inode))
3124 inode->i_op = &ext4_symlink_inode_operations;
3125 inode_nohighmem(inode);
3126 ext4_set_aops(inode);
3127 /*
3128 * We cannot call page_symlink() with transaction started
3129 * because it calls into ext4_write_begin() which can wait
3130 * for transaction commit if we are running out of space
3131 * and thus we deadlock. So we have to stop transaction now
3132 * and restart it when symlink contents is written.
3133 *
3134 * To keep fs consistent in case of crash, we have to put inode
3135 * to orphan list in the mean time.
3136 */
3137 drop_nlink(inode);
3138 err = ext4_orphan_add(handle, inode);
3139 ext4_journal_stop(handle);
3140 handle = NULL;
3141 if (err)
3142 goto err_drop_inode;
3143 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3144 if (err)
3145 goto err_drop_inode;
3146 /*
3147 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3148 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3149 */
3150 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3151 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3152 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3153 if (IS_ERR(handle)) {
3154 err = PTR_ERR(handle);
3155 handle = NULL;
3156 goto err_drop_inode;
3157 }
3158 set_nlink(inode, 1);
3159 err = ext4_orphan_del(handle, inode);
3160 if (err)
3161 goto err_drop_inode;
3162 } else {
3163 /* clear the extent format for fast symlink */
3164 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3165 if (!IS_ENCRYPTED(inode)) {
3166 inode->i_op = &ext4_fast_symlink_inode_operations;
3167 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3168 }
3169 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3170 disk_link.len);
3171 inode->i_size = disk_link.len - 1;
3172 }
3173 EXT4_I(inode)->i_disksize = inode->i_size;
3174 err = ext4_add_nondir(handle, dentry, inode);
3175 if (!err && IS_DIRSYNC(dir))
3176 ext4_handle_sync(handle);
3177
3178 if (handle)
3179 ext4_journal_stop(handle);
3180 goto out_free_encrypted_link;
3181
3182err_drop_inode:
3183 if (handle)
3184 ext4_journal_stop(handle);
3185 clear_nlink(inode);
3186 unlock_new_inode(inode);
3187 iput(inode);
3188out_free_encrypted_link:
3189 if (disk_link.name != (unsigned char *)symname)
3190 kfree(disk_link.name);
3191 return err;
3192}
3193
3194static int ext4_link(struct dentry *old_dentry,
3195 struct inode *dir, struct dentry *dentry)
3196{
3197 handle_t *handle;
3198 struct inode *inode = d_inode(old_dentry);
3199 int err, retries = 0;
3200
3201 if (inode->i_nlink >= EXT4_LINK_MAX)
3202 return -EMLINK;
3203 if (ext4_encrypted_inode(dir) &&
3204 !fscrypt_has_permitted_context(dir, inode))
3205 return -EPERM;
3206
3207 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3208 (!projid_eq(EXT4_I(dir)->i_projid,
3209 EXT4_I(old_dentry->d_inode)->i_projid)))
3210 return -EXDEV;
3211
3212 err = dquot_initialize(dir);
3213 if (err)
3214 return err;
3215
3216retry:
3217 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3218 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3219 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3220 if (IS_ERR(handle))
3221 return PTR_ERR(handle);
3222
3223 if (IS_DIRSYNC(dir))
3224 ext4_handle_sync(handle);
3225
3226 inode->i_ctime = current_time(inode);
3227 ext4_inc_count(handle, inode);
3228 ihold(inode);
3229
3230 err = ext4_add_entry(handle, dentry, inode);
3231 if (!err) {
3232 ext4_mark_inode_dirty(handle, inode);
3233 /* this can happen only for tmpfile being
3234 * linked the first time
3235 */
3236 if (inode->i_nlink == 1)
3237 ext4_orphan_del(handle, inode);
3238 d_instantiate(dentry, inode);
3239 } else {
3240 drop_nlink(inode);
3241 iput(inode);
3242 }
3243 ext4_journal_stop(handle);
3244 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3245 goto retry;
3246 return err;
3247}
3248
3249
3250/*
3251 * Try to find buffer head where contains the parent block.
3252 * It should be the inode block if it is inlined or the 1st block
3253 * if it is a normal dir.
3254 */
3255static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3256 struct inode *inode,
3257 int *retval,
3258 struct ext4_dir_entry_2 **parent_de,
3259 int *inlined)
3260{
3261 struct buffer_head *bh;
3262
3263 if (!ext4_has_inline_data(inode)) {
3264 bh = ext4_read_dirblock(inode, 0, EITHER);
3265 if (IS_ERR(bh)) {
3266 *retval = PTR_ERR(bh);
3267 return NULL;
3268 }
3269 *parent_de = ext4_next_entry(
3270 (struct ext4_dir_entry_2 *)bh->b_data,
3271 inode->i_sb->s_blocksize);
3272 return bh;
3273 }
3274
3275 *inlined = 1;
3276 return ext4_get_first_inline_block(inode, parent_de, retval);
3277}
3278
3279struct ext4_renament {
3280 struct inode *dir;
3281 struct dentry *dentry;
3282 struct inode *inode;
3283 bool is_dir;
3284 int dir_nlink_delta;
3285
3286 /* entry for "dentry" */
3287 struct buffer_head *bh;
3288 struct ext4_dir_entry_2 *de;
3289 int inlined;
3290
3291 /* entry for ".." in inode if it's a directory */
3292 struct buffer_head *dir_bh;
3293 struct ext4_dir_entry_2 *parent_de;
3294 int dir_inlined;
3295};
3296
3297static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3298{
3299 int retval;
3300
3301 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3302 &retval, &ent->parent_de,
3303 &ent->dir_inlined);
3304 if (!ent->dir_bh)
3305 return retval;
3306 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3307 return -EFSCORRUPTED;
3308 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3309 return ext4_journal_get_write_access(handle, ent->dir_bh);
3310}
3311
3312static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3313 unsigned dir_ino)
3314{
3315 int retval;
3316
3317 ent->parent_de->inode = cpu_to_le32(dir_ino);
3318 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3319 if (!ent->dir_inlined) {
3320 if (is_dx(ent->inode)) {
3321 retval = ext4_handle_dirty_dx_node(handle,
3322 ent->inode,
3323 ent->dir_bh);
3324 } else {
3325 retval = ext4_handle_dirty_dirent_node(handle,
3326 ent->inode,
3327 ent->dir_bh);
3328 }
3329 } else {
3330 retval = ext4_mark_inode_dirty(handle, ent->inode);
3331 }
3332 if (retval) {
3333 ext4_std_error(ent->dir->i_sb, retval);
3334 return retval;
3335 }
3336 return 0;
3337}
3338
3339static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3340 unsigned ino, unsigned file_type)
3341{
3342 int retval;
3343
3344 BUFFER_TRACE(ent->bh, "get write access");
3345 retval = ext4_journal_get_write_access(handle, ent->bh);
3346 if (retval)
3347 return retval;
3348 ent->de->inode = cpu_to_le32(ino);
3349 if (ext4_has_feature_filetype(ent->dir->i_sb))
3350 ent->de->file_type = file_type;
3351 ent->dir->i_version++;
3352 ent->dir->i_ctime = ent->dir->i_mtime =
3353 current_time(ent->dir);
3354 ext4_mark_inode_dirty(handle, ent->dir);
3355 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3356 if (!ent->inlined) {
3357 retval = ext4_handle_dirty_dirent_node(handle,
3358 ent->dir, ent->bh);
3359 if (unlikely(retval)) {
3360 ext4_std_error(ent->dir->i_sb, retval);
3361 return retval;
3362 }
3363 }
3364 brelse(ent->bh);
3365 ent->bh = NULL;
3366
3367 return 0;
3368}
3369
3370static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3371 const struct qstr *d_name)
3372{
3373 int retval = -ENOENT;
3374 struct buffer_head *bh;
3375 struct ext4_dir_entry_2 *de;
3376
3377 bh = ext4_find_entry(dir, d_name, &de, NULL);
3378 if (IS_ERR(bh))
3379 return PTR_ERR(bh);
3380 if (bh) {
3381 retval = ext4_delete_entry(handle, dir, de, bh);
3382 brelse(bh);
3383 }
3384 return retval;
3385}
3386
3387static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3388 int force_reread)
3389{
3390 int retval;
3391 /*
3392 * ent->de could have moved from under us during htree split, so make
3393 * sure that we are deleting the right entry. We might also be pointing
3394 * to a stale entry in the unused part of ent->bh so just checking inum
3395 * and the name isn't enough.
3396 */
3397 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3398 ent->de->name_len != ent->dentry->d_name.len ||
3399 strncmp(ent->de->name, ent->dentry->d_name.name,
3400 ent->de->name_len) ||
3401 force_reread) {
3402 retval = ext4_find_delete_entry(handle, ent->dir,
3403 &ent->dentry->d_name);
3404 } else {
3405 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3406 if (retval == -ENOENT) {
3407 retval = ext4_find_delete_entry(handle, ent->dir,
3408 &ent->dentry->d_name);
3409 }
3410 }
3411
3412 if (retval) {
3413 ext4_warning_inode(ent->dir,
3414 "Deleting old file: nlink %d, error=%d",
3415 ent->dir->i_nlink, retval);
3416 }
3417}
3418
3419static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3420{
3421 if (ent->dir_nlink_delta) {
3422 if (ent->dir_nlink_delta == -1)
3423 ext4_dec_count(handle, ent->dir);
3424 else
3425 ext4_inc_count(handle, ent->dir);
3426 ext4_mark_inode_dirty(handle, ent->dir);
3427 }
3428}
3429
3430static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3431 int credits, handle_t **h)
3432{
3433 struct inode *wh;
3434 handle_t *handle;
3435 int retries = 0;
3436
3437 /*
3438 * for inode block, sb block, group summaries,
3439 * and inode bitmap
3440 */
3441 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3442 EXT4_XATTR_TRANS_BLOCKS + 4);
3443retry:
3444 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3445 &ent->dentry->d_name, 0, NULL,
3446 EXT4_HT_DIR, credits);
3447
3448 handle = ext4_journal_current_handle();
3449 if (IS_ERR(wh)) {
3450 if (handle)
3451 ext4_journal_stop(handle);
3452 if (PTR_ERR(wh) == -ENOSPC &&
3453 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3454 goto retry;
3455 } else {
3456 *h = handle;
3457 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3458 wh->i_op = &ext4_special_inode_operations;
3459 }
3460 return wh;
3461}
3462
3463/*
3464 * Anybody can rename anything with this: the permission checks are left to the
3465 * higher-level routines.
3466 *
3467 * n.b. old_{dentry,inode) refers to the source dentry/inode
3468 * while new_{dentry,inode) refers to the destination dentry/inode
3469 * This comes from rename(const char *oldpath, const char *newpath)
3470 */
3471static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3472 struct inode *new_dir, struct dentry *new_dentry,
3473 unsigned int flags)
3474{
3475 handle_t *handle = NULL;
3476 struct ext4_renament old = {
3477 .dir = old_dir,
3478 .dentry = old_dentry,
3479 .inode = d_inode(old_dentry),
3480 };
3481 struct ext4_renament new = {
3482 .dir = new_dir,
3483 .dentry = new_dentry,
3484 .inode = d_inode(new_dentry),
3485 };
3486 int force_reread;
3487 int retval;
3488 struct inode *whiteout = NULL;
3489 int credits;
3490 u8 old_file_type;
3491
3492 if (new.inode && new.inode->i_nlink == 0) {
3493 EXT4_ERROR_INODE(new.inode,
3494 "target of rename is already freed");
3495 return -EFSCORRUPTED;
3496 }
3497
3498 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3499 (!projid_eq(EXT4_I(new_dir)->i_projid,
3500 EXT4_I(old_dentry->d_inode)->i_projid)))
3501 return -EXDEV;
3502
3503 if ((ext4_encrypted_inode(old_dir) &&
3504 !fscrypt_has_encryption_key(old_dir)) ||
3505 (ext4_encrypted_inode(new_dir) &&
3506 !fscrypt_has_encryption_key(new_dir)))
3507 return -ENOKEY;
3508
3509 retval = dquot_initialize(old.dir);
3510 if (retval)
3511 return retval;
3512 retval = dquot_initialize(new.dir);
3513 if (retval)
3514 return retval;
3515
3516 /* Initialize quotas before so that eventual writes go
3517 * in separate transaction */
3518 if (new.inode) {
3519 retval = dquot_initialize(new.inode);
3520 if (retval)
3521 return retval;
3522 }
3523
3524 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3525 if (IS_ERR(old.bh))
3526 return PTR_ERR(old.bh);
3527 /*
3528 * Check for inode number is _not_ due to possible IO errors.
3529 * We might rmdir the source, keep it as pwd of some process
3530 * and merrily kill the link to whatever was created under the
3531 * same name. Goodbye sticky bit ;-<
3532 */
3533 retval = -ENOENT;
3534 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3535 goto end_rename;
3536
3537 if ((old.dir != new.dir) &&
3538 ext4_encrypted_inode(new.dir) &&
3539 !fscrypt_has_permitted_context(new.dir, old.inode)) {
3540 retval = -EPERM;
3541 goto end_rename;
3542 }
3543
3544 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3545 &new.de, &new.inlined);
3546 if (IS_ERR(new.bh)) {
3547 retval = PTR_ERR(new.bh);
3548 new.bh = NULL;
3549 goto end_rename;
3550 }
3551 if (new.bh) {
3552 if (!new.inode) {
3553 brelse(new.bh);
3554 new.bh = NULL;
3555 }
3556 }
3557 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3558 ext4_alloc_da_blocks(old.inode);
3559
3560 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3561 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3562 if (!(flags & RENAME_WHITEOUT)) {
3563 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3564 if (IS_ERR(handle)) {
3565 retval = PTR_ERR(handle);
3566 handle = NULL;
3567 goto end_rename;
3568 }
3569 } else {
3570 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3571 if (IS_ERR(whiteout)) {
3572 retval = PTR_ERR(whiteout);
3573 whiteout = NULL;
3574 goto end_rename;
3575 }
3576 }
3577
3578 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3579 ext4_handle_sync(handle);
3580
3581 if (S_ISDIR(old.inode->i_mode)) {
3582 if (new.inode) {
3583 retval = -ENOTEMPTY;
3584 if (!ext4_empty_dir(new.inode))
3585 goto end_rename;
3586 } else {
3587 retval = -EMLINK;
3588 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3589 goto end_rename;
3590 }
3591 retval = ext4_rename_dir_prepare(handle, &old);
3592 if (retval)
3593 goto end_rename;
3594 }
3595 /*
3596 * If we're renaming a file within an inline_data dir and adding or
3597 * setting the new dirent causes a conversion from inline_data to
3598 * extents/blockmap, we need to force the dirent delete code to
3599 * re-read the directory, or else we end up trying to delete a dirent
3600 * from what is now the extent tree root (or a block map).
3601 */
3602 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3603 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3604
3605 old_file_type = old.de->file_type;
3606 if (whiteout) {
3607 /*
3608 * Do this before adding a new entry, so the old entry is sure
3609 * to be still pointing to the valid old entry.
3610 */
3611 retval = ext4_setent(handle, &old, whiteout->i_ino,
3612 EXT4_FT_CHRDEV);
3613 if (retval)
3614 goto end_rename;
3615 ext4_mark_inode_dirty(handle, whiteout);
3616 }
3617 if (!new.bh) {
3618 retval = ext4_add_entry(handle, new.dentry, old.inode);
3619 if (retval)
3620 goto end_rename;
3621 } else {
3622 retval = ext4_setent(handle, &new,
3623 old.inode->i_ino, old_file_type);
3624 if (retval)
3625 goto end_rename;
3626 }
3627 if (force_reread)
3628 force_reread = !ext4_test_inode_flag(new.dir,
3629 EXT4_INODE_INLINE_DATA);
3630
3631 /*
3632 * Like most other Unix systems, set the ctime for inodes on a
3633 * rename.
3634 */
3635 old.inode->i_ctime = current_time(old.inode);
3636 ext4_mark_inode_dirty(handle, old.inode);
3637
3638 if (!whiteout) {
3639 /*
3640 * ok, that's it
3641 */
3642 ext4_rename_delete(handle, &old, force_reread);
3643 }
3644
3645 if (new.inode) {
3646 ext4_dec_count(handle, new.inode);
3647 new.inode->i_ctime = current_time(new.inode);
3648 }
3649 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3650 ext4_update_dx_flag(old.dir);
3651 if (old.dir_bh) {
3652 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3653 if (retval)
3654 goto end_rename;
3655
3656 ext4_dec_count(handle, old.dir);
3657 if (new.inode) {
3658 /* checked ext4_empty_dir above, can't have another
3659 * parent, ext4_dec_count() won't work for many-linked
3660 * dirs */
3661 clear_nlink(new.inode);
3662 } else {
3663 ext4_inc_count(handle, new.dir);
3664 ext4_update_dx_flag(new.dir);
3665 ext4_mark_inode_dirty(handle, new.dir);
3666 }
3667 }
3668 ext4_mark_inode_dirty(handle, old.dir);
3669 if (new.inode) {
3670 ext4_mark_inode_dirty(handle, new.inode);
3671 if (!new.inode->i_nlink)
3672 ext4_orphan_add(handle, new.inode);
3673 }
3674 retval = 0;
3675
3676end_rename:
3677 brelse(old.dir_bh);
3678 brelse(old.bh);
3679 brelse(new.bh);
3680 if (whiteout) {
3681 if (retval)
3682 drop_nlink(whiteout);
3683 unlock_new_inode(whiteout);
3684 iput(whiteout);
3685 }
3686 if (handle)
3687 ext4_journal_stop(handle);
3688 return retval;
3689}
3690
3691static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3692 struct inode *new_dir, struct dentry *new_dentry)
3693{
3694 handle_t *handle = NULL;
3695 struct ext4_renament old = {
3696 .dir = old_dir,
3697 .dentry = old_dentry,
3698 .inode = d_inode(old_dentry),
3699 };
3700 struct ext4_renament new = {
3701 .dir = new_dir,
3702 .dentry = new_dentry,
3703 .inode = d_inode(new_dentry),
3704 };
3705 u8 new_file_type;
3706 int retval;
3707 struct timespec ctime;
3708
3709 if ((ext4_encrypted_inode(old_dir) &&
3710 !fscrypt_has_encryption_key(old_dir)) ||
3711 (ext4_encrypted_inode(new_dir) &&
3712 !fscrypt_has_encryption_key(new_dir)))
3713 return -ENOKEY;
3714
3715 if ((ext4_encrypted_inode(old_dir) ||
3716 ext4_encrypted_inode(new_dir)) &&
3717 (old_dir != new_dir) &&
3718 (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3719 !fscrypt_has_permitted_context(old_dir, new.inode)))
3720 return -EPERM;
3721
3722 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3723 !projid_eq(EXT4_I(new_dir)->i_projid,
3724 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3725 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3726 !projid_eq(EXT4_I(old_dir)->i_projid,
3727 EXT4_I(new_dentry->d_inode)->i_projid)))
3728 return -EXDEV;
3729
3730 retval = dquot_initialize(old.dir);
3731 if (retval)
3732 return retval;
3733 retval = dquot_initialize(new.dir);
3734 if (retval)
3735 return retval;
3736
3737 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3738 &old.de, &old.inlined);
3739 if (IS_ERR(old.bh))
3740 return PTR_ERR(old.bh);
3741 /*
3742 * Check for inode number is _not_ due to possible IO errors.
3743 * We might rmdir the source, keep it as pwd of some process
3744 * and merrily kill the link to whatever was created under the
3745 * same name. Goodbye sticky bit ;-<
3746 */
3747 retval = -ENOENT;
3748 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3749 goto end_rename;
3750
3751 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3752 &new.de, &new.inlined);
3753 if (IS_ERR(new.bh)) {
3754 retval = PTR_ERR(new.bh);
3755 new.bh = NULL;
3756 goto end_rename;
3757 }
3758
3759 /* RENAME_EXCHANGE case: old *and* new must both exist */
3760 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3761 goto end_rename;
3762
3763 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3764 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3765 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3766 if (IS_ERR(handle)) {
3767 retval = PTR_ERR(handle);
3768 handle = NULL;
3769 goto end_rename;
3770 }
3771
3772 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3773 ext4_handle_sync(handle);
3774
3775 if (S_ISDIR(old.inode->i_mode)) {
3776 old.is_dir = true;
3777 retval = ext4_rename_dir_prepare(handle, &old);
3778 if (retval)
3779 goto end_rename;
3780 }
3781 if (S_ISDIR(new.inode->i_mode)) {
3782 new.is_dir = true;
3783 retval = ext4_rename_dir_prepare(handle, &new);
3784 if (retval)
3785 goto end_rename;
3786 }
3787
3788 /*
3789 * Other than the special case of overwriting a directory, parents'
3790 * nlink only needs to be modified if this is a cross directory rename.
3791 */
3792 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3793 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3794 new.dir_nlink_delta = -old.dir_nlink_delta;
3795 retval = -EMLINK;
3796 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3797 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3798 goto end_rename;
3799 }
3800
3801 new_file_type = new.de->file_type;
3802 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3803 if (retval)
3804 goto end_rename;
3805
3806 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3807 if (retval)
3808 goto end_rename;
3809
3810 /*
3811 * Like most other Unix systems, set the ctime for inodes on a
3812 * rename.
3813 */
3814 ctime = current_time(old.inode);
3815 old.inode->i_ctime = ctime;
3816 new.inode->i_ctime = ctime;
3817 ext4_mark_inode_dirty(handle, old.inode);
3818 ext4_mark_inode_dirty(handle, new.inode);
3819
3820 if (old.dir_bh) {
3821 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3822 if (retval)
3823 goto end_rename;
3824 }
3825 if (new.dir_bh) {
3826 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3827 if (retval)
3828 goto end_rename;
3829 }
3830 ext4_update_dir_count(handle, &old);
3831 ext4_update_dir_count(handle, &new);
3832 retval = 0;
3833
3834end_rename:
3835 brelse(old.dir_bh);
3836 brelse(new.dir_bh);
3837 brelse(old.bh);
3838 brelse(new.bh);
3839 if (handle)
3840 ext4_journal_stop(handle);
3841 return retval;
3842}
3843
3844static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3845 struct inode *new_dir, struct dentry *new_dentry,
3846 unsigned int flags)
3847{
3848 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3849 return -EIO;
3850
3851 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3852 return -EINVAL;
3853
3854 if (flags & RENAME_EXCHANGE) {
3855 return ext4_cross_rename(old_dir, old_dentry,
3856 new_dir, new_dentry);
3857 }
3858
3859 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3860}
3861
3862/*
3863 * directories can handle most operations...
3864 */
3865const struct inode_operations ext4_dir_inode_operations = {
3866 .create = ext4_create,
3867 .lookup = ext4_lookup,
3868 .link = ext4_link,
3869 .unlink = ext4_unlink,
3870 .symlink = ext4_symlink,
3871 .mkdir = ext4_mkdir,
3872 .rmdir = ext4_rmdir,
3873 .mknod = ext4_mknod,
3874 .tmpfile = ext4_tmpfile,
3875 .rename = ext4_rename2,
3876 .setattr = ext4_setattr,
3877 .getattr = ext4_getattr,
3878 .listxattr = ext4_listxattr,
3879 .get_acl = ext4_get_acl,
3880 .set_acl = ext4_set_acl,
3881 .fiemap = ext4_fiemap,
3882};
3883
3884const struct inode_operations ext4_special_inode_operations = {
3885 .setattr = ext4_setattr,
3886 .getattr = ext4_getattr,
3887 .listxattr = ext4_listxattr,
3888 .get_acl = ext4_get_acl,
3889 .set_acl = ext4_set_acl,
3890};