projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
jbd2: fast commit recovery path
[linux-2.6-block.git] / fs / ext4 / namei.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
ac27a0ec 2/*
617ba13b 3 * linux/fs/ext4/namei.c
ac27a0ec
DK		 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>
ac27a0ec 30#include <linux/time.h>
ac27a0ec
DK		 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>
ae5e165d 37#include <linux/iversion.h>
b886ee3e 38#include <linux/unicode.h>
3dcf5451
CH		 39#include "ext4.h"
40#include "ext4_jbd2.h"
ac27a0ec 41
ac27a0ec
DK		 42#include "xattr.h"
43#include "acl.h"
44
0562e0ba 45#include <trace/events/ext4.h>
ac27a0ec
DK		 46/*
47 * define how far ahead to read directories while searching them.
48 */
49#define NAMEI_RA_CHUNKS 2
50#define NAMEI_RA_BLOCKS 4
8c55e204 51#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
ac27a0ec 52
617ba13b 53static struct buffer_head *ext4_append(handle_t *handle,
ac27a0ec 54 struct inode *inode,
0f70b406 55 ext4_lblk_t *block)
ac27a0ec
DK		 56{
57 struct buffer_head *bh;
1c215028 58 int err;
ac27a0ec 59
df981d03
TT		 60 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
61 ((inode->i_size >> 10) >=
0f70b406
TT		 62 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
63 return ERR_PTR(-ENOSPC);
df981d03 64
ac27a0ec
DK		 65 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66
c5e298ae 67 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
1c215028
TT		 68 if (IS_ERR(bh))
69 return bh;
0f70b406
TT		 70 inode->i_size += inode->i_sb->s_blocksize;
71 EXT4_I(inode)->i_disksize = inode->i_size;
5d601255 72 BUFFER_TRACE(bh, "get_write_access");
0f70b406
TT		 73 err = ext4_journal_get_write_access(handle, bh);
74 if (err) {
75 brelse(bh);
76 ext4_std_error(inode->i_sb, err);
77 return ERR_PTR(err);
6d1ab10e 78 }
ac27a0ec
DK		 79 return bh;
80}
81
dc6982ff
TT		 82static int ext4_dx_csum_verify(struct inode *inode,
83 struct ext4_dir_entry *dirent);
84
4e19d6b6
TT		 85/*
86 * Hints to ext4_read_dirblock regarding whether we expect a directory
87 * block being read to be an index block, or a block containing
88 * directory entries (and if the latter, whether it was found via a
89 * logical block in an htree index block). This is used to control
90 * what sort of sanity checkinig ext4_read_dirblock() will do on the
91 * directory block read from the storage device. EITHER will means
92 * the caller doesn't know what kind of directory block will be read,
93 * so no specific verification will be done.
94 */
dc6982ff 95typedef enum {
4e19d6b6 96 EITHER, INDEX, DIRENT, DIRENT_HTREE
dc6982ff
TT		 97} dirblock_type_t;
98
99#define ext4_read_dirblock(inode, block, type) \
b03a2f7e 100 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
dc6982ff
TT		 101
102static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
b03a2f7e
AD		 103 ext4_lblk_t block,
104 dirblock_type_t type,
105 const char *func,
106 unsigned int line)
dc6982ff
TT		 107{
108 struct buffer_head *bh;
109 struct ext4_dir_entry *dirent;
1c215028 110 int is_dx_block = 0;
dc6982ff 111
46f870d6
TT		 112 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
113 bh = ERR_PTR(-EIO);
114 else
115 bh = ext4_bread(NULL, inode, block, 0);
1c215028 116 if (IS_ERR(bh)) {
b03a2f7e
AD		 117 __ext4_warning(inode->i_sb, func, line,
118 "inode #%lu: lblock %lu: comm %s: "
119 "error %ld reading directory block",
120 inode->i_ino, (unsigned long)block,
121 current->comm, PTR_ERR(bh));
1c215028
TT		 122
123 return bh;
124 }
4e19d6b6 125 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
b03a2f7e 126 ext4_error_inode(inode, func, line, block,
4e19d6b6
TT		 127 "Directory hole found for htree %s block",
128 (type == INDEX) ? "index" : "leaf");
6a797d27 129 return ERR_PTR(-EFSCORRUPTED);
dc6982ff 130 }
4e19d6b6
TT		 131 if (!bh)
132 return NULL;
dc6982ff
TT		 133 dirent = (struct ext4_dir_entry *) bh->b_data;
134 /* Determine whether or not we have an index block */
135 if (is_dx(inode)) {
136 if (block == 0)
137 is_dx_block = 1;
138 else if (ext4_rec_len_from_disk(dirent->rec_len,
139 inode->i_sb->s_blocksize) ==
140 inode->i_sb->s_blocksize)
141 is_dx_block = 1;
142 }
143 if (!is_dx_block && type == INDEX) {
b03a2f7e 144 ext4_error_inode(inode, func, line, block,
dc6982ff 145 "directory leaf block found instead of index block");
de59fae0 146 brelse(bh);
6a797d27 147 return ERR_PTR(-EFSCORRUPTED);
dc6982ff 148 }
9aa5d32b 149 if (!ext4_has_metadata_csum(inode->i_sb) ||
dc6982ff
TT		 150 buffer_verified(bh))
151 return bh;
152
153 /*
154 * An empty leaf block can get mistaken for a index block; for
155 * this reason, we can only check the index checksum when the
156 * caller is sure it should be an index block.
157 */
158 if (is_dx_block && type == INDEX) {
46f870d6
TT		 159 if (ext4_dx_csum_verify(inode, dirent) &&
160 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
dc6982ff
TT		 161 set_buffer_verified(bh);
162 else {
54d3adbc
TT		 163 ext4_error_inode_err(inode, func, line, block,
164 EFSBADCRC,
165 "Directory index failed checksum");
a871611b 166 brelse(bh);
6a797d27 167 return ERR_PTR(-EFSBADCRC);
a871611b 168 }
ac27a0ec 169 }
dc6982ff 170 if (!is_dx_block) {
46f870d6
TT		 171 if (ext4_dirblock_csum_verify(inode, bh) &&
172 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
dc6982ff
TT		 173 set_buffer_verified(bh);
174 else {
54d3adbc
TT		 175 ext4_error_inode_err(inode, func, line, block,
176 EFSBADCRC,
177 "Directory block failed checksum");
dc6982ff 178 brelse(bh);
6a797d27 179 return ERR_PTR(-EFSBADCRC);
dc6982ff 180 }
6d1ab10e 181 }
ac27a0ec
DK		 182 return bh;
183}
184
185#ifndef assert
186#define assert(test) J_ASSERT(test)
187#endif
188
ac27a0ec
DK		 189#ifdef DX_DEBUG
190#define dxtrace(command) command
191#else
192#define dxtrace(command)
193#endif
194
195struct fake_dirent
196{
197 __le32 inode;
198 __le16 rec_len;
199 u8 name_len;
200 u8 file_type;
201};
202
203struct dx_countlimit
204{
205 __le16 limit;
206 __le16 count;
207};
208
209struct dx_entry
210{
211 __le32 hash;
212 __le32 block;
213};
214
215/*
216 * dx_root_info is laid out so that if it should somehow get overlaid by a
217 * dirent the two low bits of the hash version will be zero. Therefore, the
218 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
219 */
220
221struct dx_root
222{
223 struct fake_dirent dot;
224 char dot_name[4];
225 struct fake_dirent dotdot;
226 char dotdot_name[4];
227 struct dx_root_info
228 {
229 __le32 reserved_zero;
230 u8 hash_version;
231 u8 info_length; /* 8 */
232 u8 indirect_levels;
233 u8 unused_flags;
234 }
235 info;
6cfb061f 236 struct dx_entry entries[];
ac27a0ec
DK		 237};
238
239struct dx_node
240{
241 struct fake_dirent fake;
6cfb061f 242 struct dx_entry entries[];
ac27a0ec
DK		 243};
244
245
246struct dx_frame
247{
248 struct buffer_head *bh;
249 struct dx_entry *entries;
250 struct dx_entry *at;
251};
252
253struct dx_map_entry
254{
255 u32 hash;
ef2b02d3
ES		 256 u16 offs;
257 u16 size;
ac27a0ec
DK		 258};
259
e6153918
DW		 260/*
261 * This goes at the end of each htree block.
262 */
263struct dx_tail {
264 u32 dt_reserved;
265 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
266};
267
725d26d3
AK		 268static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
269static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
af5bc92d
TT		 270static inline unsigned dx_get_hash(struct dx_entry *entry);
271static void dx_set_hash(struct dx_entry *entry, unsigned value);
272static unsigned dx_get_count(struct dx_entry *entries);
273static unsigned dx_get_limit(struct dx_entry *entries);
274static void dx_set_count(struct dx_entry *entries, unsigned value);
275static void dx_set_limit(struct dx_entry *entries, unsigned value);
276static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
277static unsigned dx_node_limit(struct inode *dir);
5b643f9c 278static struct dx_frame *dx_probe(struct ext4_filename *fname,
ac27a0ec
DK		 279 struct inode *dir,
280 struct dx_hash_info *hinfo,
dd73b5d5 281 struct dx_frame *frame);
af5bc92d 282static void dx_release(struct dx_frame *frames);
1f3862b5
MH		 283static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
284 unsigned blocksize, struct dx_hash_info *hinfo,
285 struct dx_map_entry map[]);
ac27a0ec 286static void dx_sort_map(struct dx_map_entry *map, unsigned count);
af5bc92d 287static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
3d0518f4 288 struct dx_map_entry *offsets, int count, unsigned blocksize);
8bad4597 289static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
725d26d3
AK		 290static void dx_insert_block(struct dx_frame *frame,
291 u32 hash, ext4_lblk_t block);
617ba13b 292static int ext4_htree_next_block(struct inode *dir, __u32 hash,
ac27a0ec
DK		 293 struct dx_frame *frame,
294 struct dx_frame *frames,
295 __u32 *start_hash);
f702ba0f 296static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
5b643f9c 297 struct ext4_filename *fname,
537d8f93 298 struct ext4_dir_entry_2 **res_dir);
5b643f9c 299static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
56a04915 300 struct inode *dir, struct inode *inode);
ac27a0ec 301
dbe89444 302/* checksumming functions */
ddce3b94
TT		 303void ext4_initialize_dirent_tail(struct buffer_head *bh,
304 unsigned int blocksize)
b0336e8d 305{
ddce3b94
TT		 306 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
307
b0336e8d
DW		 308 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
309 t->det_rec_len = ext4_rec_len_to_disk(
310 sizeof(struct ext4_dir_entry_tail), blocksize);
311 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
312}
313
314/* Walk through a dirent block to find a checksum "dirent" at the tail */
315static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
f036adb3 316 struct buffer_head *bh)
b0336e8d
DW		 317{
318 struct ext4_dir_entry_tail *t;
319
320#ifdef PARANOID
321 struct ext4_dir_entry *d, *top;
322
f036adb3
TT		 323 d = (struct ext4_dir_entry *)bh->b_data;
324 top = (struct ext4_dir_entry *)(bh->b_data +
b0336e8d 325 (EXT4_BLOCK_SIZE(inode->i_sb) -
f036adb3 326 sizeof(struct ext4_dir_entry_tail)));
b0336e8d
DW		 327 while (d < top && d->rec_len)
328 d = (struct ext4_dir_entry *)(((void *)d) +
329 le16_to_cpu(d->rec_len));
330
331 if (d != top)
332 return NULL;
333
334 t = (struct ext4_dir_entry_tail *)d;
335#else
f036adb3 336 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
b0336e8d
DW		 337#endif
338
339 if (t->det_reserved_zero1 ||
340 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
341 t->det_reserved_zero2 ||
342 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
343 return NULL;
344
345 return t;
346}
347
f036adb3 348static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
b0336e8d
DW		 349{
350 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
351 struct ext4_inode_info *ei = EXT4_I(inode);
352 __u32 csum;
353
354 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
355 return cpu_to_le32(csum);
356}
357
b03a2f7e
AD		 358#define warn_no_space_for_csum(inode) \
359 __warn_no_space_for_csum((inode), __func__, __LINE__)
360
361static void __warn_no_space_for_csum(struct inode *inode, const char *func,
362 unsigned int line)
dffe9d8d 363{
b03a2f7e
AD		 364 __ext4_warning_inode(inode, func, line,
365 "No space for directory leaf checksum. Please run e2fsck -D.");
dffe9d8d
TT		 366}
367
f036adb3 368int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
b0336e8d
DW		 369{
370 struct ext4_dir_entry_tail *t;
371
9aa5d32b 372 if (!ext4_has_metadata_csum(inode->i_sb))
b0336e8d
DW		 373 return 1;
374
f036adb3 375 t = get_dirent_tail(inode, bh);
b0336e8d 376 if (!t) {
dffe9d8d 377 warn_no_space_for_csum(inode);
b0336e8d
DW		 378 return 0;
379 }
380
f036adb3 381 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
ddce3b94 382 (char *)t - bh->b_data))
b0336e8d
DW		 383 return 0;
384
385 return 1;
386}
387
f036adb3
TT		 388static void ext4_dirblock_csum_set(struct inode *inode,
389 struct buffer_head *bh)
b0336e8d
DW		 390{
391 struct ext4_dir_entry_tail *t;
392
9aa5d32b 393 if (!ext4_has_metadata_csum(inode->i_sb))
b0336e8d
DW		 394 return;
395
f036adb3 396 t = get_dirent_tail(inode, bh);
b0336e8d 397 if (!t) {
dffe9d8d 398 warn_no_space_for_csum(inode);
b0336e8d
DW		 399 return;
400 }
401
f036adb3 402 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
ddce3b94 403 (char *)t - bh->b_data);
b0336e8d
DW		 404}
405
f036adb3
TT		 406int ext4_handle_dirty_dirblock(handle_t *handle,
407 struct inode *inode,
408 struct buffer_head *bh)
b0336e8d 409{
f036adb3 410 ext4_dirblock_csum_set(inode, bh);
b0336e8d
DW		 411 return ext4_handle_dirty_metadata(handle, inode, bh);
412}
413
dbe89444
DW		 414static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
415 struct ext4_dir_entry *dirent,
416 int *offset)
417{
418 struct ext4_dir_entry *dp;
419 struct dx_root_info *root;
420 int count_offset;
421
422 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
423 count_offset = 8;
424 else if (le16_to_cpu(dirent->rec_len) == 12) {
425 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
426 if (le16_to_cpu(dp->rec_len) !=
427 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
428 return NULL;
429 root = (struct dx_root_info *)(((void *)dp + 12));
430 if (root->reserved_zero ||
431 root->info_length != sizeof(struct dx_root_info))
432 return NULL;
433 count_offset = 32;
434 } else
435 return NULL;
436
437 if (offset)
438 *offset = count_offset;
439 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
440}
441
442static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
443 int count_offset, int count, struct dx_tail *t)
444{
445 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
446 struct ext4_inode_info *ei = EXT4_I(inode);
d6a77105 447 __u32 csum;
dbe89444 448 int size;
b47820ed
DJ		 449 __u32 dummy_csum = 0;
450 int offset = offsetof(struct dx_tail, dt_checksum);
dbe89444
DW		 451
452 size = count_offset + (count * sizeof(struct dx_entry));
dbe89444 453 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
b47820ed
DJ		 454 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
455 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
dbe89444
DW		 456
457 return cpu_to_le32(csum);
458}
459
460static int ext4_dx_csum_verify(struct inode *inode,
461 struct ext4_dir_entry *dirent)
462{
463 struct dx_countlimit *c;
464 struct dx_tail *t;
465 int count_offset, limit, count;
466
9aa5d32b 467 if (!ext4_has_metadata_csum(inode->i_sb))
dbe89444
DW		 468 return 1;
469
470 c = get_dx_countlimit(inode, dirent, &count_offset);
471 if (!c) {
472 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
fa964540 473 return 0;
dbe89444
DW		 474 }
475 limit = le16_to_cpu(c->limit);
476 count = le16_to_cpu(c->count);
477 if (count_offset + (limit * sizeof(struct dx_entry)) >
478 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
dffe9d8d 479 warn_no_space_for_csum(inode);
fa964540 480 return 0;
dbe89444
DW		 481 }
482 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
483
484 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
485 count, t))
486 return 0;
487 return 1;
488}
489
490static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
491{
492 struct dx_countlimit *c;
493 struct dx_tail *t;
494 int count_offset, limit, count;
495
9aa5d32b 496 if (!ext4_has_metadata_csum(inode->i_sb))
dbe89444
DW		 497 return;
498
499 c = get_dx_countlimit(inode, dirent, &count_offset);
500 if (!c) {
501 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
502 return;
503 }
504 limit = le16_to_cpu(c->limit);
505 count = le16_to_cpu(c->count);
506 if (count_offset + (limit * sizeof(struct dx_entry)) >
507 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
dffe9d8d 508 warn_no_space_for_csum(inode);
dbe89444
DW		 509 return;
510 }
511 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
512
513 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
514}
515
516static inline int ext4_handle_dirty_dx_node(handle_t *handle,
517 struct inode *inode,
518 struct buffer_head *bh)
519{
520 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
521 return ext4_handle_dirty_metadata(handle, inode, bh);
522}
523
f795e140
LZ		 524/*
525 * p is at least 6 bytes before the end of page
526 */
527static inline struct ext4_dir_entry_2 *
3d0518f4 528ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
f795e140
LZ		 529{
530 return (struct ext4_dir_entry_2 *)((char *)p +
3d0518f4 531 ext4_rec_len_from_disk(p->rec_len, blocksize));
f795e140
LZ		 532}
533
ac27a0ec
DK		 534/*
535 * Future: use high four bits of block for coalesce-on-delete flags
536 * Mask them off for now.
537 */
538
725d26d3 539static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
ac27a0ec 540{
e08ac99f 541 return le32_to_cpu(entry->block) & 0x0fffffff;
ac27a0ec
DK		 542}
543
725d26d3 544static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
ac27a0ec
DK		 545{
546 entry->block = cpu_to_le32(value);
547}
548
af5bc92d 549static inline unsigned dx_get_hash(struct dx_entry *entry)
ac27a0ec
DK		 550{
551 return le32_to_cpu(entry->hash);
552}
553
af5bc92d 554static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
ac27a0ec
DK		 555{
556 entry->hash = cpu_to_le32(value);
557}
558
af5bc92d 559static inline unsigned dx_get_count(struct dx_entry *entries)
ac27a0ec
DK		 560{
561 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
562}
563
af5bc92d 564static inline unsigned dx_get_limit(struct dx_entry *entries)
ac27a0ec
DK		 565{
566 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
567}
568
af5bc92d 569static inline void dx_set_count(struct dx_entry *entries, unsigned value)
ac27a0ec
DK		 570{
571 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
572}
573
af5bc92d 574static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
ac27a0ec
DK		 575{
576 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
577}
578
af5bc92d 579static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
ac27a0ec 580{
617ba13b
MC		 581 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
582 EXT4_DIR_REC_LEN(2) - infosize;
dbe89444 583
9aa5d32b 584 if (ext4_has_metadata_csum(dir->i_sb))
dbe89444 585 entry_space -= sizeof(struct dx_tail);
d9c769b7 586 return entry_space / sizeof(struct dx_entry);
ac27a0ec
DK		 587}
588
af5bc92d 589static inline unsigned dx_node_limit(struct inode *dir)
ac27a0ec 590{
617ba13b 591 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
dbe89444 592
9aa5d32b 593 if (ext4_has_metadata_csum(dir->i_sb))
dbe89444 594 entry_space -= sizeof(struct dx_tail);
d9c769b7 595 return entry_space / sizeof(struct dx_entry);
ac27a0ec
DK		 596}
597
598/*
599 * Debug
600 */
601#ifdef DX_DEBUG
4776004f 602static void dx_show_index(char * label, struct dx_entry *entries)
ac27a0ec 603{
63f57933 604 int i, n = dx_get_count (entries);
d74f3d25 605 printk(KERN_DEBUG "%s index", label);
63f57933 606 for (i = 0; i < n; i++) {
d74f3d25
JP		 607 printk(KERN_CONT " %x->%lu",
608 i ? dx_get_hash(entries + i) : 0,
609 (unsigned long)dx_get_block(entries + i));
63f57933 610 }
d74f3d25 611 printk(KERN_CONT "\n");
ac27a0ec
DK		 612}
613
614struct stats
615{
616 unsigned names;
617 unsigned space;
618 unsigned bcount;
619};
620
b3098486
MH		 621static struct stats dx_show_leaf(struct inode *dir,
622 struct dx_hash_info *hinfo,
623 struct ext4_dir_entry_2 *de,
624 int size, int show_names)
ac27a0ec
DK		 625{
626 unsigned names = 0, space = 0;
627 char *base = (char *) de;
628 struct dx_hash_info h = *hinfo;
629
630 printk("names: ");
631 while ((char *) de < base + size)
632 {
633 if (de->inode)
634 {
635 if (show_names)
636 {
643fa961 637#ifdef CONFIG_FS_ENCRYPTION
b3098486
MH		 638 int len;
639 char *name;
a7550b30
JK		 640 struct fscrypt_str fname_crypto_str =
641 FSTR_INIT(NULL, 0);
c936e1ec 642 int res = 0;
b3098486
MH		 643
644 name = de->name;
645 len = de->name_len;
592ddec7 646 if (IS_ENCRYPTED(dir))
a7550b30 647 res = fscrypt_get_encryption_info(dir);
b7236e21
TT		 648 if (res) {
649 printk(KERN_WARNING "Error setting up"
650 " fname crypto: %d\n", res);
b3098486 651 }
a7550b30 652 if (!fscrypt_has_encryption_key(dir)) {
b3098486 653 /* Directory is not encrypted */
b886ee3e 654 ext4fs_dirhash(dir, de->name,
b3098486
MH		 655 de->name_len, &h);
656 printk("%*.s:(U)%x.%u ", len,
657 name, h.hash,
658 (unsigned) ((char *) de
659 - base));
660 } else {
a7550b30
JK		 661 struct fscrypt_str de_name =
662 FSTR_INIT(name, len);
663
b3098486 664 /* Directory is encrypted */
a7550b30
JK		 665 res = fscrypt_fname_alloc_buffer(
666 dir, len,
b3098486 667 &fname_crypto_str);
ef1eb3aa 668 if (res)
b3098486
MH		 669 printk(KERN_WARNING "Error "
670 "allocating crypto "
671 "buffer--skipping "
672 "crypto\n");
a7550b30
JK		 673 res = fscrypt_fname_disk_to_usr(dir,
674 0, 0, &de_name,
675 &fname_crypto_str);
ef1eb3aa 676 if (res) {
b3098486
MH		 677 printk(KERN_WARNING "Error "
678 "converting filename "
679 "from disk to usr"
680 "\n");
681 name = "??";
682 len = 2;
683 } else {
684 name = fname_crypto_str.name;
685 len = fname_crypto_str.len;
686 }
b886ee3e
GKB		 687 ext4fs_dirhash(dir, de->name,
688 de->name_len, &h);
b3098486
MH		 689 printk("%*.s:(E)%x.%u ", len, name,
690 h.hash, (unsigned) ((char *) de
691 - base));
a7550b30
JK		 692 fscrypt_fname_free_buffer(
693 &fname_crypto_str);
b3098486
MH		 694 }
695#else
ac27a0ec
DK		 696 int len = de->name_len;
697 char *name = de->name;
b886ee3e 698 ext4fs_dirhash(dir, de->name, de->name_len, &h);
b3098486 699 printk("%*.s:%x.%u ", len, name, h.hash,
265c6a0f 700 (unsigned) ((char *) de - base));
b3098486 701#endif
ac27a0ec 702 }
617ba13b 703 space += EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec
DK		 704 names++;
705 }
3d0518f4 706 de = ext4_next_entry(de, size);
ac27a0ec 707 }
d74f3d25 708 printk(KERN_CONT "(%i)\n", names);
ac27a0ec
DK		 709 return (struct stats) { names, space, 1 };
710}
711
712struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
713 struct dx_entry *entries, int levels)
714{
715 unsigned blocksize = dir->i_sb->s_blocksize;
af5bc92d 716 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
ac27a0ec
DK		 717 unsigned bcount = 0;
718 struct buffer_head *bh;
ac27a0ec
DK		 719 printk("%i indexed blocks...\n", count);
720 for (i = 0; i < count; i++, entries++)
721 {
725d26d3
AK		 722 ext4_lblk_t block = dx_get_block(entries);
723 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
ac27a0ec
DK		 724 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
725 struct stats stats;
726 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
1c215028
TT		 727 bh = ext4_bread(NULL,dir, block, 0);
728 if (!bh || IS_ERR(bh))
729 continue;
ac27a0ec
DK		 730 stats = levels?
731 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
b3098486
MH		 732 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
733 bh->b_data, blocksize, 0);
ac27a0ec
DK		 734 names += stats.names;
735 space += stats.space;
736 bcount += stats.bcount;
af5bc92d 737 brelse(bh);
ac27a0ec
DK		 738 }
739 if (bcount)
60e6679e 740 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
4776004f
TT		 741 levels ? "" : " ", names, space/bcount,
742 (space/bcount)*100/blocksize);
ac27a0ec
DK		 743 return (struct stats) { names, space, bcount};
744}
745#endif /* DX_DEBUG */
746
747/*
748 * Probe for a directory leaf block to search.
749 *
750 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
751 * error in the directory index, and the caller should fall back to
752 * searching the directory normally. The callers of dx_probe **MUST**
753 * check for this error code, and make sure it never gets reflected
754 * back to userspace.
755 */
756static struct dx_frame *
5b643f9c 757dx_probe(struct ext4_filename *fname, struct inode *dir,
dd73b5d5 758 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
ac27a0ec
DK		 759{
760 unsigned count, indirect;
761 struct dx_entry *at, *entries, *p, *q, *m;
762 struct dx_root *root;
ac27a0ec 763 struct dx_frame *frame = frame_in;
dd73b5d5 764 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
ac27a0ec
DK		 765 u32 hash;
766
e08ac99f 767 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
dd73b5d5
TT		 768 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
769 if (IS_ERR(frame->bh))
770 return (struct dx_frame *) frame->bh;
771
772 root = (struct dx_root *) frame->bh->b_data;
ac27a0ec
DK		 773 if (root->info.hash_version != DX_HASH_TEA &&
774 root->info.hash_version != DX_HASH_HALF_MD4 &&
775 root->info.hash_version != DX_HASH_LEGACY) {
b03a2f7e
AD		 776 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
777 root->info.hash_version);
ac27a0ec
DK		 778 goto fail;
779 }
5b643f9c
TT		 780 if (fname)
781 hinfo = &fname->hinfo;
ac27a0ec 782 hinfo->hash_version = root->info.hash_version;
f99b2589
TT		 783 if (hinfo->hash_version <= DX_HASH_TEA)
784 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
617ba13b 785 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
5b643f9c 786 if (fname && fname_name(fname))
b886ee3e 787 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
ac27a0ec
DK		 788 hash = hinfo->hash;
789
790 if (root->info.unused_flags & 1) {
b03a2f7e
AD		 791 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
792 root->info.unused_flags);
ac27a0ec
DK		 793 goto fail;
794 }
795
b03a2f7e 796 indirect = root->info.indirect_levels;
e08ac99f
AB		 797 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
798 ext4_warning(dir->i_sb,
799 "Directory (ino: %lu) htree depth %#06x exceed"
800 "supported value", dir->i_ino,
801 ext4_dir_htree_level(dir->i_sb));
802 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
803 ext4_warning(dir->i_sb, "Enable large directory "
804 "feature to access it");
805 }
ac27a0ec
DK		 806 goto fail;
807 }
808
b03a2f7e
AD		 809 entries = (struct dx_entry *)(((char *)&root->info) +
810 root->info.info_length);
3d82abae
ES		 811
812 if (dx_get_limit(entries) != dx_root_limit(dir,
813 root->info.info_length)) {
b03a2f7e
AD		 814 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
815 dx_get_limit(entries),
816 dx_root_limit(dir, root->info.info_length));
3d82abae
ES		 817 goto fail;
818 }
819
af5bc92d 820 dxtrace(printk("Look up %x", hash));
dd73b5d5 821 while (1) {
ac27a0ec 822 count = dx_get_count(entries);
3d82abae 823 if (!count || count > dx_get_limit(entries)) {
b03a2f7e
AD		 824 ext4_warning_inode(dir,
825 "dx entry: count %u beyond limit %u",
826 count, dx_get_limit(entries));
dd73b5d5 827 goto fail;
3d82abae
ES		 828 }
829
ac27a0ec
DK		 830 p = entries + 1;
831 q = entries + count - 1;
dd73b5d5 832 while (p <= q) {
b03a2f7e 833 m = p + (q - p) / 2;
d74f3d25 834 dxtrace(printk(KERN_CONT "."));
ac27a0ec
DK		 835 if (dx_get_hash(m) > hash)
836 q = m - 1;
837 else
838 p = m + 1;
839 }
840
dd73b5d5 841 if (0) { // linear search cross check
ac27a0ec
DK		 842 unsigned n = count - 1;
843 at = entries;
844 while (n--)
845 {
d74f3d25 846 dxtrace(printk(KERN_CONT ","));
ac27a0ec
DK		 847 if (dx_get_hash(++at) > hash)
848 {
849 at--;
850 break;
851 }
852 }
853 assert (at == p - 1);
854 }
855
856 at = p - 1;
d74f3d25
JP		 857 dxtrace(printk(KERN_CONT " %x->%u\n",
858 at == entries ? 0 : dx_get_hash(at),
b03a2f7e 859 dx_get_block(at)));
ac27a0ec
DK		 860 frame->entries = entries;
861 frame->at = at;
dd73b5d5
TT		 862 if (!indirect--)
863 return frame;
864 frame++;
865 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
866 if (IS_ERR(frame->bh)) {
867 ret_err = (struct dx_frame *) frame->bh;
868 frame->bh = NULL;
869 goto fail;
dbe89444 870 }
dd73b5d5 871 entries = ((struct dx_node *) frame->bh->b_data)->entries;
dbe89444 872
b03a2f7e
AD		 873 if (dx_get_limit(entries) != dx_node_limit(dir)) {
874 ext4_warning_inode(dir,
875 "dx entry: limit %u != node limit %u",
876 dx_get_limit(entries), dx_node_limit(dir));
dd73b5d5 877 goto fail;
3d82abae 878 }
ac27a0ec 879 }
dd73b5d5 880fail:
ac27a0ec
DK		 881 while (frame >= frame_in) {
882 brelse(frame->bh);
883 frame--;
884 }
b3098486 885
dd73b5d5 886 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
b03a2f7e
AD		 887 ext4_warning_inode(dir,
888 "Corrupt directory, running e2fsck is recommended");
dd73b5d5 889 return ret_err;
ac27a0ec
DK		 890}
891
b03a2f7e 892static void dx_release(struct dx_frame *frames)
ac27a0ec 893{
e08ac99f
AB		 894 struct dx_root_info *info;
895 int i;
08fc98a4 896 unsigned int indirect_levels;
e08ac99f 897
ac27a0ec
DK		 898 if (frames[0].bh == NULL)
899 return;
900
e08ac99f 901 info = &((struct dx_root *)frames[0].bh->b_data)->info;
08fc98a4
ST		 902 /* save local copy, "info" may be freed after brelse() */
903 indirect_levels = info->indirect_levels;
904 for (i = 0; i <= indirect_levels; i++) {
e08ac99f
AB		 905 if (frames[i].bh == NULL)
906 break;
907 brelse(frames[i].bh);
908 frames[i].bh = NULL;
909 }
ac27a0ec
DK		 910}
911
912/*
913 * This function increments the frame pointer to search the next leaf
914 * block, and reads in the necessary intervening nodes if the search
915 * should be necessary. Whether or not the search is necessary is
916 * controlled by the hash parameter. If the hash value is even, then
917 * the search is only continued if the next block starts with that
918 * hash value. This is used if we are searching for a specific file.
919 *
920 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
921 *
922 * This function returns 1 if the caller should continue to search,
923 * or 0 if it should not. If there is an error reading one of the
924 * index blocks, it will a negative error code.
925 *
926 * If start_hash is non-null, it will be filled in with the starting
927 * hash of the next page.
928 */
617ba13b 929static int ext4_htree_next_block(struct inode *dir, __u32 hash,
ac27a0ec
DK		 930 struct dx_frame *frame,
931 struct dx_frame *frames,
932 __u32 *start_hash)
933{
934 struct dx_frame *p;
935 struct buffer_head *bh;
dc6982ff 936 int num_frames = 0;
ac27a0ec
DK		 937 __u32 bhash;
938
939 p = frame;
940 /*
941 * Find the next leaf page by incrementing the frame pointer.
942 * If we run out of entries in the interior node, loop around and
943 * increment pointer in the parent node. When we break out of
944 * this loop, num_frames indicates the number of interior
945 * nodes need to be read.
946 */
947 while (1) {
948 if (++(p->at) < p->entries + dx_get_count(p->entries))
949 break;
950 if (p == frames)
951 return 0;
952 num_frames++;
953 p--;
954 }
955
956 /*
957 * If the hash is 1, then continue only if the next page has a
958 * continuation hash of any value. This is used for readdir
959 * handling. Otherwise, check to see if the hash matches the
960 * desired contiuation hash. If it doesn't, return since
961 * there's no point to read in the successive index pages.
962 */
963 bhash = dx_get_hash(p->at);
964 if (start_hash)
965 *start_hash = bhash;
966 if ((hash & 1) == 0) {
967 if ((bhash & ~1) != hash)
968 return 0;
969 }
970 /*
971 * If the hash is HASH_NB_ALWAYS, we always go to the next
972 * block so no check is necessary
973 */
974 while (num_frames--) {
dc6982ff
TT		 975 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
976 if (IS_ERR(bh))
977 return PTR_ERR(bh);
ac27a0ec 978 p++;
af5bc92d 979 brelse(p->bh);
ac27a0ec
DK		 980 p->bh = bh;
981 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
982 }
983 return 1;
984}
985
986
ac27a0ec
DK		 987/*
988 * This function fills a red-black tree with information from a
989 * directory block. It returns the number directory entries loaded
990 * into the tree. If there is an error it is returned in err.
991 */
992static int htree_dirblock_to_tree(struct file *dir_file,
725d26d3 993 struct inode *dir, ext4_lblk_t block,
ac27a0ec
DK		 994 struct dx_hash_info *hinfo,
995 __u32 start_hash, __u32 start_minor_hash)
996{
997 struct buffer_head *bh;
617ba13b 998 struct ext4_dir_entry_2 *de, *top;
90b0a973 999 int err = 0, count = 0;
a7550b30 1000 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
ac27a0ec 1001
725d26d3
AK		 1002 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1003 (unsigned long)block));
4e19d6b6 1004 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
dc6982ff
TT		 1005 if (IS_ERR(bh))
1006 return PTR_ERR(bh);
b0336e8d 1007
617ba13b
MC		 1008 de = (struct ext4_dir_entry_2 *) bh->b_data;
1009 top = (struct ext4_dir_entry_2 *) ((char *) de +
ac27a0ec 1010 dir->i_sb->s_blocksize -
617ba13b 1011 EXT4_DIR_REC_LEN(0));
1f3862b5 1012 /* Check if the directory is encrypted */
592ddec7 1013 if (IS_ENCRYPTED(dir)) {
a7550b30 1014 err = fscrypt_get_encryption_info(dir);
c936e1ec
TT		 1015 if (err < 0) {
1016 brelse(bh);
1017 return err;
1018 }
a7550b30 1019 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1f3862b5
MH		 1020 &fname_crypto_str);
1021 if (err < 0) {
1f3862b5
MH		 1022 brelse(bh);
1023 return err;
1024 }
1025 }
64c314ff 1026
3d0518f4 1027 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
f7c21177 1028 if (ext4_check_dir_entry(dir, NULL, de, bh,
226ba972 1029 bh->b_data, bh->b_size,
cad3f007
TT		 1030 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1031 + ((char *)de - bh->b_data))) {
64cb9273
AV		 1032 /* silently ignore the rest of the block */
1033 break;
e6c40211 1034 }
b886ee3e 1035 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
ac27a0ec
DK		 1036 if ((hinfo->hash < start_hash) ||
1037 ((hinfo->hash == start_hash) &&
1038 (hinfo->minor_hash < start_minor_hash)))
1039 continue;
1040 if (de->inode == 0)
1041 continue;
592ddec7 1042 if (!IS_ENCRYPTED(dir)) {
1f3862b5
MH		 1043 tmp_str.name = de->name;
1044 tmp_str.len = de->name_len;
1045 err = ext4_htree_store_dirent(dir_file,
1046 hinfo->hash, hinfo->minor_hash, de,
1047 &tmp_str);
1048 } else {
d2299590 1049 int save_len = fname_crypto_str.len;
a7550b30
JK		 1050 struct fscrypt_str de_name = FSTR_INIT(de->name,
1051 de->name_len);
d2299590 1052
1f3862b5 1053 /* Directory is encrypted */
a7550b30
JK		 1054 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1055 hinfo->minor_hash, &de_name,
1056 &fname_crypto_str);
ef1eb3aa 1057 if (err) {
1f3862b5
MH		 1058 count = err;
1059 goto errout;
1060 }
1061 err = ext4_htree_store_dirent(dir_file,
1062 hinfo->hash, hinfo->minor_hash, de,
1063 &fname_crypto_str);
d2299590 1064 fname_crypto_str.len = save_len;
1f3862b5 1065 }
2f61830a 1066 if (err != 0) {
1f3862b5
MH		 1067 count = err;
1068 goto errout;
ac27a0ec
DK		 1069 }
1070 count++;
1071 }
1f3862b5 1072errout:
ac27a0ec 1073 brelse(bh);
a7550b30 1074 fscrypt_fname_free_buffer(&fname_crypto_str);
ac27a0ec
DK		 1075 return count;
1076}
1077
1078
1079/*
1080 * This function fills a red-black tree with information from a
1081 * directory. We start scanning the directory in hash order, starting
1082 * at start_hash and start_minor_hash.
1083 *
1084 * This function returns the number of entries inserted into the tree,
1085 * or a negative error code.
1086 */
617ba13b 1087int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
ac27a0ec
DK		 1088 __u32 start_minor_hash, __u32 *next_hash)
1089{
1090 struct dx_hash_info hinfo;
617ba13b 1091 struct ext4_dir_entry_2 *de;
e08ac99f 1092 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 1093 struct inode *dir;
725d26d3 1094 ext4_lblk_t block;
ac27a0ec 1095 int count = 0;
725d26d3 1096 int ret, err;
ac27a0ec 1097 __u32 hashval;
a7550b30 1098 struct fscrypt_str tmp_str;
ac27a0ec 1099
60e6679e 1100 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
4776004f 1101 start_hash, start_minor_hash));
496ad9aa 1102 dir = file_inode(dir_file);
12e9b892 1103 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
617ba13b 1104 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
f99b2589
TT		 1105 if (hinfo.hash_version <= DX_HASH_TEA)
1106 hinfo.hash_version +=
1107 EXT4_SB(dir->i_sb)->s_hash_unsigned;
617ba13b 1108 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
8af0f082
TM		 1109 if (ext4_has_inline_data(dir)) {
1110 int has_inline_data = 1;
7633b08b
TT		 1111 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1112 &hinfo, start_hash,
1113 start_minor_hash,
1114 &has_inline_data);
8af0f082
TM		 1115 if (has_inline_data) {
1116 *next_hash = ~0;
1117 return count;
1118 }
1119 }
ac27a0ec
DK		 1120 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1121 start_hash, start_minor_hash);
1122 *next_hash = ~0;
1123 return count;
1124 }
1125 hinfo.hash = start_hash;
1126 hinfo.minor_hash = 0;
dd73b5d5
TT		 1127 frame = dx_probe(NULL, dir, &hinfo, frames);
1128 if (IS_ERR(frame))
1129 return PTR_ERR(frame);
ac27a0ec
DK		 1130
1131 /* Add '.' and '..' from the htree header */
1132 if (!start_hash && !start_minor_hash) {
617ba13b 1133 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
2f61830a
TT		 1134 tmp_str.name = de->name;
1135 tmp_str.len = de->name_len;
1136 err = ext4_htree_store_dirent(dir_file, 0, 0,
1137 de, &tmp_str);
1138 if (err != 0)
ac27a0ec
DK		 1139 goto errout;
1140 count++;
1141 }
1142 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
617ba13b 1143 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
3d0518f4 1144 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
2f61830a
TT		 1145 tmp_str.name = de->name;
1146 tmp_str.len = de->name_len;
1147 err = ext4_htree_store_dirent(dir_file, 2, 0,
1148 de, &tmp_str);
1149 if (err != 0)
ac27a0ec
DK		 1150 goto errout;
1151 count++;
1152 }
1153
1154 while (1) {
1f60fbe7
TT		 1155 if (fatal_signal_pending(current)) {
1156 err = -ERESTARTSYS;
1157 goto errout;
1158 }
1159 cond_resched();
ac27a0ec
DK		 1160 block = dx_get_block(frame->at);
1161 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1162 start_hash, start_minor_hash);
1163 if (ret < 0) {
1164 err = ret;
1165 goto errout;
1166 }
1167 count += ret;
1168 hashval = ~0;
617ba13b 1169 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
ac27a0ec
DK		 1170 frame, frames, &hashval);
1171 *next_hash = hashval;
1172 if (ret < 0) {
1173 err = ret;
1174 goto errout;
1175 }
1176 /*
1177 * Stop if: (a) there are no more entries, or
1178 * (b) we have inserted at least one entry and the
1179 * next hash value is not a continuation
1180 */
1181 if ((ret == 0) ||
1182 (count && ((hashval & 1) == 0)))
1183 break;
1184 }
1185 dx_release(frames);
4776004f
TT		 1186 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1187 "next hash: %x\n", count, *next_hash));
ac27a0ec
DK		 1188 return count;
1189errout:
1190 dx_release(frames);
1191 return (err);
1192}
1193
7335cd3b
TM		 1194static inline int search_dirblock(struct buffer_head *bh,
1195 struct inode *dir,
5b643f9c 1196 struct ext4_filename *fname,
7335cd3b
TM		 1197 unsigned int offset,
1198 struct ext4_dir_entry_2 **res_dir)
1199{
5b643f9c 1200 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
d6b97550 1201 fname, offset, res_dir);
7335cd3b
TM		 1202}
1203
ac27a0ec
DK		 1204/*
1205 * Directory block splitting, compacting
1206 */
1207
ef2b02d3
ES		 1208/*
1209 * Create map of hash values, offsets, and sizes, stored at end of block.
1210 * Returns number of entries mapped.
1211 */
1f3862b5
MH		 1212static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1213 unsigned blocksize, struct dx_hash_info *hinfo,
8bad4597 1214 struct dx_map_entry *map_tail)
ac27a0ec
DK		 1215{
1216 int count = 0;
1217 char *base = (char *) de;
1218 struct dx_hash_info h = *hinfo;
1219
8bad4597 1220 while ((char *) de < base + blocksize) {
ac27a0ec 1221 if (de->name_len && de->inode) {
b886ee3e 1222 ext4fs_dirhash(dir, de->name, de->name_len, &h);
ac27a0ec
DK		 1223 map_tail--;
1224 map_tail->hash = h.hash;
9aee2286 1225 map_tail->offs = ((char *) de - base)>>2;
ef2b02d3 1226 map_tail->size = le16_to_cpu(de->rec_len);
ac27a0ec
DK		 1227 count++;
1228 cond_resched();
1229 }
1230 /* XXX: do we need to check rec_len == 0 case? -Chris */
3d0518f4 1231 de = ext4_next_entry(de, blocksize);
ac27a0ec
DK		 1232 }
1233 return count;
1234}
1235
ef2b02d3 1236/* Sort map by hash value */
ac27a0ec
DK		 1237static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1238{
63f57933
AM		 1239 struct dx_map_entry *p, *q, *top = map + count - 1;
1240 int more;
1241 /* Combsort until bubble sort doesn't suck */
1242 while (count > 2) {
1243 count = count*10/13;
1244 if (count - 9 < 2) /* 9, 10 -> 11 */
1245 count = 11;
1246 for (p = top, q = p - count; q >= map; p--, q--)
1247 if (p->hash < q->hash)
1248 swap(*p, *q);
1249 }
1250 /* Garden variety bubble sort */
1251 do {
1252 more = 0;
1253 q = top;
1254 while (q-- > map) {
1255 if (q[1].hash >= q[0].hash)
ac27a0ec 1256 continue;
63f57933
AM		 1257 swap(*(q+1), *q);
1258 more = 1;
ac27a0ec
DK		 1259 }
1260 } while(more);
1261}
1262
725d26d3 1263static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
ac27a0ec
DK		 1264{
1265 struct dx_entry *entries = frame->entries;
1266 struct dx_entry *old = frame->at, *new = old + 1;
1267 int count = dx_get_count(entries);
1268
1269 assert(count < dx_get_limit(entries));
1270 assert(old < entries + count);
1271 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1272 dx_set_hash(new, hash);
1273 dx_set_block(new, block);
1274 dx_set_count(entries, count + 1);
1275}
ac27a0ec 1276
b886ee3e
GKB		 1277#ifdef CONFIG_UNICODE
1278/*
1279 * Test whether a case-insensitive directory entry matches the filename
3ae72562
GKB		 1280 * being searched for. If quick is set, assume the name being looked up
1281 * is already in the casefolded form.
b886ee3e
GKB		 1282 *
1283 * Returns: 0 if the directory entry matches, more than 0 if it
1284 * doesn't match or less than zero on error.
1285 */
1286int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
3ae72562 1287 const struct qstr *entry, bool quick)
b886ee3e
GKB		 1288{
1289 const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1290 const struct unicode_map *um = sbi->s_encoding;
1291 int ret;
1292
3ae72562
GKB		 1293 if (quick)
1294 ret = utf8_strncasecmp_folded(um, name, entry);
1295 else
1296 ret = utf8_strncasecmp(um, name, entry);
1297
b886ee3e
GKB		 1298 if (ret < 0) {
1299 /* Handle invalid character sequence as either an error
1300 * or as an opaque byte sequence.
1301 */
1302 if (ext4_has_strict_mode(sbi))
1303 return -EINVAL;
1304
1305 if (name->len != entry->len)
1306 return 1;
1307
1308 return !!memcmp(name->name, entry->name, name->len);
1309 }
1310
1311 return ret;
1312}
3ae72562
GKB		 1313
1314void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1315 struct fscrypt_str *cf_name)
1316{
96fcaf86
GKB		 1317 int len;
1318
6456ca65 1319 if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
3ae72562
GKB		 1320 cf_name->name = NULL;
1321 return;
1322 }
1323
1324 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1325 if (!cf_name->name)
1326 return;
1327
96fcaf86
GKB		 1328 len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1329 iname, cf_name->name,
1330 EXT4_NAME_LEN);
1331 if (len <= 0) {
3ae72562
GKB		 1332 kfree(cf_name->name);
1333 cf_name->name = NULL;
96fcaf86 1334 return;
3ae72562 1335 }
96fcaf86
GKB		 1336 cf_name->len = (unsigned) len;
1337
3ae72562 1338}
b886ee3e
GKB		 1339#endif
1340
ac27a0ec 1341/*
d9b9f8d5 1342 * Test whether a directory entry matches the filename being searched for.
ac27a0ec 1343 *
d9b9f8d5 1344 * Return: %true if the directory entry matches, otherwise %false.
ac27a0ec 1345 */
b886ee3e
GKB		 1346static inline bool ext4_match(const struct inode *parent,
1347 const struct ext4_filename *fname,
d9b9f8d5 1348 const struct ext4_dir_entry_2 *de)
ac27a0ec 1349{
067d1023 1350 struct fscrypt_name f;
b886ee3e
GKB		 1351#ifdef CONFIG_UNICODE
1352 const struct qstr entry = {.name = de->name, .len = de->name_len};
1353#endif
1f3862b5 1354
ac27a0ec 1355 if (!de->inode)
d9b9f8d5 1356 return false;
1f3862b5 1357
067d1023
EB		 1358 f.usr_fname = fname->usr_fname;
1359 f.disk_name = fname->disk_name;
643fa961 1360#ifdef CONFIG_FS_ENCRYPTION
067d1023 1361 f.crypto_buf = fname->crypto_buf;
1f3862b5 1362#endif
b886ee3e
GKB		 1363
1364#ifdef CONFIG_UNICODE
3ae72562
GKB		 1365 if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1366 if (fname->cf_name.name) {
1367 struct qstr cf = {.name = fname->cf_name.name,
1368 .len = fname->cf_name.len};
1369 return !ext4_ci_compare(parent, &cf, &entry, true);
1370 }
1371 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1372 false);
1373 }
b886ee3e
GKB		 1374#endif
1375
067d1023 1376 return fscrypt_match_name(&f, de->name, de->name_len);
ac27a0ec
DK		 1377}
1378
1379/*
1380 * Returns 0 if not found, -1 on failure, and 1 on success
1381 */
5b643f9c
TT		 1382int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1383 struct inode *dir, struct ext4_filename *fname,
5b643f9c 1384 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
ac27a0ec 1385{
617ba13b 1386 struct ext4_dir_entry_2 * de;
ac27a0ec
DK		 1387 char * dlimit;
1388 int de_len;
1f3862b5 1389
7335cd3b
TM		 1390 de = (struct ext4_dir_entry_2 *)search_buf;
1391 dlimit = search_buf + buf_size;
ac27a0ec
DK		 1392 while ((char *) de < dlimit) {
1393 /* this code is executed quadratically often */
1394 /* do minimal checking `by hand' */
d9b9f8d5 1395 if ((char *) de + de->name_len <= dlimit &&
b886ee3e 1396 ext4_match(dir, fname, de)) {
d9b9f8d5
EB		 1397 /* found a match - just to be sure, do
1398 * a full check */
7303cb5b
JK		 1399 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1400 buf_size, offset))
d9b9f8d5
EB		 1401 return -1;
1402 *res_dir = de;
1403 return 1;
ac27a0ec
DK		 1404 }
1405 /* prevent looping on a bad block */
3d0518f4
WY		 1406 de_len = ext4_rec_len_from_disk(de->rec_len,
1407 dir->i_sb->s_blocksize);
d9b9f8d5
EB		 1408 if (de_len <= 0)
1409 return -1;
ac27a0ec 1410 offset += de_len;
617ba13b 1411 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
ac27a0ec 1412 }
d9b9f8d5 1413 return 0;
ac27a0ec
DK		 1414}
1415
c6af8803
DW		 1416static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1417 struct ext4_dir_entry *de)
1418{
1419 struct super_block *sb = dir->i_sb;
1420
1421 if (!is_dx(dir))
1422 return 0;
1423 if (block == 0)
1424 return 1;
1425 if (de->inode == 0 &&
1426 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1427 sb->s_blocksize)
1428 return 1;
1429 return 0;
1430}
ac27a0ec
DK		 1431
1432/*
b01531db 1433 * __ext4_find_entry()
ac27a0ec
DK		 1434 *
1435 * finds an entry in the specified directory with the wanted name. It
1436 * returns the cache buffer in which the entry was found, and the entry
1437 * itself (as a parameter - res_dir). It does NOT read the inode of the
1438 * entry - you'll have to do that yourself if you want to.
1439 *
1440 * The returned buffer_head has ->b_count elevated. The caller is expected
1441 * to brelse() it when appropriate.
1442 */
b01531db
EB		 1443static struct buffer_head *__ext4_find_entry(struct inode *dir,
1444 struct ext4_filename *fname,
1445 struct ext4_dir_entry_2 **res_dir,
1446 int *inlined)
ac27a0ec 1447{
af5bc92d
TT		 1448 struct super_block *sb;
1449 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1450 struct buffer_head *bh, *ret = NULL;
9699d4f9 1451 ext4_lblk_t start, block;
b01531db 1452 const u8 *name = fname->usr_fname->name;
9699d4f9 1453 size_t ra_max = 0; /* Number of bh's in the readahead
ac27a0ec 1454 buffer, bh_use[] */
9699d4f9 1455 size_t ra_ptr = 0; /* Current index into readahead
ac27a0ec 1456 buffer */
725d26d3 1457 ext4_lblk_t nblocks;
5b643f9c 1458 int i, namelen, retval;
ac27a0ec
DK		 1459
1460 *res_dir = NULL;
1461 sb = dir->i_sb;
b01531db 1462 namelen = fname->usr_fname->len;
617ba13b 1463 if (namelen > EXT4_NAME_LEN)
ac27a0ec 1464 return NULL;
e8e948e7
TM		 1465
1466 if (ext4_has_inline_data(dir)) {
1467 int has_inline_data = 1;
b01531db 1468 ret = ext4_find_inline_entry(dir, fname, res_dir,
e8e948e7 1469 &has_inline_data);
32f7f22c
TM		 1470 if (has_inline_data) {
1471 if (inlined)
1472 *inlined = 1;
5b643f9c 1473 goto cleanup_and_exit;
32f7f22c 1474 }
e8e948e7
TM		 1475 }
1476
8941ec8b 1477 if ((namelen <= 2) && (name[0] == '.') &&
6d5c3aa8 1478 (name[1] == '.' || name[1] == '\0')) {
8941ec8b
TT		 1479 /*
1480 * "." or ".." will only be in the first block
1481 * NFS may look up ".."; "." should be handled by the VFS
1482 */
1483 block = start = 0;
1484 nblocks = 1;
1485 goto restart;
1486 }
ac27a0ec 1487 if (is_dx(dir)) {
b01531db 1488 ret = ext4_dx_find_entry(dir, fname, res_dir);
ac27a0ec
DK		 1489 /*
1490 * On success, or if the error was file not found,
1491 * return. Otherwise, fall back to doing a search the
1492 * old fashioned way.
1493 */
5b643f9c
TT		 1494 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1495 goto cleanup_and_exit;
4776004f
TT		 1496 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1497 "falling back\n"));
f39b3f45 1498 ret = NULL;
ac27a0ec 1499 }
617ba13b 1500 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
9d5afec6
CR		 1501 if (!nblocks) {
1502 ret = NULL;
1503 goto cleanup_and_exit;
1504 }
617ba13b 1505 start = EXT4_I(dir)->i_dir_start_lookup;
ac27a0ec
DK		 1506 if (start >= nblocks)
1507 start = 0;
1508 block = start;
1509restart:
1510 do {
1511 /*
1512 * We deal with the read-ahead logic here.
1513 */
9424ef56 1514 cond_resched();
ac27a0ec
DK		 1515 if (ra_ptr >= ra_max) {
1516 /* Refill the readahead buffer */
1517 ra_ptr = 0;
9699d4f9
TE		 1518 if (block < start)
1519 ra_max = start - block;
1520 else
1521 ra_max = nblocks - block;
1522 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1523 retval = ext4_bread_batch(dir, block, ra_max,
1524 false /* wait */, bh_use);
1525 if (retval) {
1526 ret = ERR_PTR(retval);
1527 ra_max = 0;
1528 goto cleanup_and_exit;
ac27a0ec
DK		 1529 }
1530 }
1531 if ((bh = bh_use[ra_ptr++]) == NULL)
1532 goto next;
1533 wait_on_buffer(bh);
1534 if (!buffer_uptodate(bh)) {
54d3adbc
TT		 1535 EXT4_ERROR_INODE_ERR(dir, EIO,
1536 "reading directory lblock %lu",
1537 (unsigned long) block);
ac27a0ec 1538 brelse(bh);
6febe6f2
KK		 1539 ret = ERR_PTR(-EIO);
1540 goto cleanup_and_exit;
ac27a0ec 1541 }
b0336e8d 1542 if (!buffer_verified(bh) &&
c6af8803
DW		 1543 !is_dx_internal_node(dir, block,
1544 (struct ext4_dir_entry *)bh->b_data) &&
f036adb3 1545 !ext4_dirblock_csum_verify(dir, bh)) {
54d3adbc
TT		 1546 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1547 "checksumming directory "
1548 "block %lu", (unsigned long)block);
b0336e8d 1549 brelse(bh);
bdddf342
TT		 1550 ret = ERR_PTR(-EFSBADCRC);
1551 goto cleanup_and_exit;
b0336e8d
DW		 1552 }
1553 set_buffer_verified(bh);
b01531db 1554 i = search_dirblock(bh, dir, fname,
617ba13b 1555 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
ac27a0ec 1556 if (i == 1) {
617ba13b 1557 EXT4_I(dir)->i_dir_start_lookup = block;
ac27a0ec
DK		 1558 ret = bh;
1559 goto cleanup_and_exit;
1560 } else {
1561 brelse(bh);
1562 if (i < 0)
1563 goto cleanup_and_exit;
1564 }
1565 next:
1566 if (++block >= nblocks)
1567 block = 0;
1568 } while (block != start);
1569
1570 /*
1571 * If the directory has grown while we were searching, then
1572 * search the last part of the directory before giving up.
1573 */
1574 block = nblocks;
617ba13b 1575 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
ac27a0ec
DK		 1576 if (block < nblocks) {
1577 start = 0;
1578 goto restart;
1579 }
1580
1581cleanup_and_exit:
1582 /* Clean up the read-ahead blocks */
1583 for (; ra_ptr < ra_max; ra_ptr++)
af5bc92d 1584 brelse(bh_use[ra_ptr]);
ac27a0ec
DK		 1585 return ret;
1586}
1587
b01531db
EB		 1588static struct buffer_head *ext4_find_entry(struct inode *dir,
1589 const struct qstr *d_name,
1590 struct ext4_dir_entry_2 **res_dir,
1591 int *inlined)
1592{
1593 int err;
1594 struct ext4_filename fname;
1595 struct buffer_head *bh;
1596
1597 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1598 if (err == -ENOENT)
1599 return NULL;
1600 if (err)
1601 return ERR_PTR(err);
1602
1603 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1604
1605 ext4_fname_free_filename(&fname);
1606 return bh;
1607}
1608
1609static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1610 struct dentry *dentry,
1611 struct ext4_dir_entry_2 **res_dir)
1612{
1613 int err;
1614 struct ext4_filename fname;
1615 struct buffer_head *bh;
1616
1617 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1618 if (err == -ENOENT)
1619 return NULL;
1620 if (err)
1621 return ERR_PTR(err);
1622
1623 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1624
1625 ext4_fname_free_filename(&fname);
1626 return bh;
1627}
1628
5b643f9c
TT		 1629static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1630 struct ext4_filename *fname,
1631 struct ext4_dir_entry_2 **res_dir)
ac27a0ec 1632{
8941ec8b 1633 struct super_block * sb = dir->i_sb;
e08ac99f 1634 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 1635 struct buffer_head *bh;
725d26d3 1636 ext4_lblk_t block;
ac27a0ec 1637 int retval;
ac27a0ec 1638
643fa961 1639#ifdef CONFIG_FS_ENCRYPTION
1f3862b5
MH		 1640 *res_dir = NULL;
1641#endif
5b643f9c 1642 frame = dx_probe(fname, dir, NULL, frames);
dd73b5d5
TT		 1643 if (IS_ERR(frame))
1644 return (struct buffer_head *) frame;
ac27a0ec
DK		 1645 do {
1646 block = dx_get_block(frame->at);
4e19d6b6 1647 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
537d8f93 1648 if (IS_ERR(bh))
b0336e8d 1649 goto errout;
537d8f93 1650
d6b97550 1651 retval = search_dirblock(bh, dir, fname,
7845c049
TT		 1652 block << EXT4_BLOCK_SIZE_BITS(sb),
1653 res_dir);
537d8f93
TT		 1654 if (retval == 1)
1655 goto success;
af5bc92d 1656 brelse(bh);
7845c049 1657 if (retval == -1) {
537d8f93 1658 bh = ERR_PTR(ERR_BAD_DX_DIR);
7845c049
TT		 1659 goto errout;
1660 }
1661
ac27a0ec 1662 /* Check to see if we should continue to search */
5b643f9c 1663 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
ac27a0ec
DK		 1664 frames, NULL);
1665 if (retval < 0) {
b03a2f7e
AD		 1666 ext4_warning_inode(dir,
1667 "error %d reading directory index block",
1668 retval);
537d8f93 1669 bh = ERR_PTR(retval);
ac27a0ec
DK		 1670 goto errout;
1671 }
1672 } while (retval == 1);
1673
537d8f93 1674 bh = NULL;
ac27a0ec 1675errout:
d6b97550 1676 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
537d8f93
TT		 1677success:
1678 dx_release(frames);
1679 return bh;
ac27a0ec 1680}
ac27a0ec 1681
00cd8dd3 1682static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
ac27a0ec 1683{
af5bc92d
TT		 1684 struct inode *inode;
1685 struct ext4_dir_entry_2 *de;
1686 struct buffer_head *bh;
28b4c263 1687
89904275
EB		 1688 if (dentry->d_name.len > EXT4_NAME_LEN)
1689 return ERR_PTR(-ENAMETOOLONG);
ac27a0ec 1690
b01531db 1691 bh = ext4_lookup_entry(dir, dentry, &de);
36de9286 1692 if (IS_ERR(bh))
e884bce1 1693 return ERR_CAST(bh);
ac27a0ec
DK		 1694 inode = NULL;
1695 if (bh) {
498e5f24 1696 __u32 ino = le32_to_cpu(de->inode);
af5bc92d 1697 brelse(bh);
617ba13b 1698 if (!ext4_valid_inum(dir->i_sb, ino)) {
24676da4 1699 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
6a797d27 1700 return ERR_PTR(-EFSCORRUPTED);
a6c15c2b 1701 }
7e936b73 1702 if (unlikely(ino == dir->i_ino)) {
a34e15cc
DH		 1703 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1704 dentry);
6a797d27 1705 return ERR_PTR(-EFSCORRUPTED);
7e936b73 1706 }
8a363970 1707 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
a9049376
AV		 1708 if (inode == ERR_PTR(-ESTALE)) {
1709 EXT4_ERROR_INODE(dir,
1710 "deleted inode referenced: %u",
1711 ino);
6a797d27 1712 return ERR_PTR(-EFSCORRUPTED);
e6f009b0 1713 }
592ddec7 1714 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
ff978b09 1715 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
a7550b30 1716 !fscrypt_has_permitted_context(dir, inode)) {
d9cdc903 1717 ext4_warning(inode->i_sb,
8d2ae1cb 1718 "Inconsistent encryption contexts: %lu/%lu",
8c68084b 1719 dir->i_ino, inode->i_ino);
dd01b690 1720 iput(inode);
d9cdc903
TT		 1721 return ERR_PTR(-EPERM);
1722 }
ac27a0ec 1723 }
b886ee3e
GKB		 1724
1725#ifdef CONFIG_UNICODE
1726 if (!inode && IS_CASEFOLDED(dir)) {
1727 /* Eventually we want to call d_add_ci(dentry, NULL)
1728 * for negative dentries in the encoding case as
1729 * well. For now, prevent the negative dentry
1730 * from being cached.
1731 */
1732 return NULL;
1733 }
1734#endif
ac27a0ec
DK		 1735 return d_splice_alias(inode, dentry);
1736}
1737
1738
617ba13b 1739struct dentry *ext4_get_parent(struct dentry *child)
ac27a0ec 1740{
498e5f24 1741 __u32 ino;
26fe5750 1742 static const struct qstr dotdot = QSTR_INIT("..", 2);
617ba13b 1743 struct ext4_dir_entry_2 * de;
ac27a0ec
DK		 1744 struct buffer_head *bh;
1745
2b0143b5 1746 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
36de9286 1747 if (IS_ERR(bh))
e884bce1 1748 return ERR_CAST(bh);
ac27a0ec
DK		 1749 if (!bh)
1750 return ERR_PTR(-ENOENT);
1751 ino = le32_to_cpu(de->inode);
1752 brelse(bh);
1753
fc64005c 1754 if (!ext4_valid_inum(child->d_sb, ino)) {
2b0143b5 1755 EXT4_ERROR_INODE(d_inode(child),
24676da4 1756 "bad parent inode number: %u", ino);
6a797d27 1757 return ERR_PTR(-EFSCORRUPTED);
a6c15c2b
VA		 1758 }
1759
8a363970 1760 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
ac27a0ec
DK		 1761}
1762
ef2b02d3
ES		 1763/*
1764 * Move count entries from end of map between two memory locations.
1765 * Returns pointer to last entry moved.
1766 */
617ba13b 1767static struct ext4_dir_entry_2 *
3d0518f4
WY		 1768dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1769 unsigned blocksize)
ac27a0ec
DK		 1770{
1771 unsigned rec_len = 0;
1772
1773 while (count--) {
60e6679e 1774 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
9aee2286 1775 (from + (map->offs<<2));
617ba13b 1776 rec_len = EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec 1777 memcpy (to, de, rec_len);
617ba13b 1778 ((struct ext4_dir_entry_2 *) to)->rec_len =
3d0518f4 1779 ext4_rec_len_to_disk(rec_len, blocksize);
ac27a0ec
DK		 1780 de->inode = 0;
1781 map++;
1782 to += rec_len;
1783 }
617ba13b 1784 return (struct ext4_dir_entry_2 *) (to - rec_len);
ac27a0ec
DK		 1785}
1786
ef2b02d3
ES		 1787/*
1788 * Compact each dir entry in the range to the minimal rec_len.
1789 * Returns pointer to last entry in range.
1790 */
8bad4597 1791static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
ac27a0ec 1792{
617ba13b 1793 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
ac27a0ec
DK		 1794 unsigned rec_len = 0;
1795
1796 prev = to = de;
8bad4597 1797 while ((char*)de < base + blocksize) {
3d0518f4 1798 next = ext4_next_entry(de, blocksize);
ac27a0ec 1799 if (de->inode && de->name_len) {
617ba13b 1800 rec_len = EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec
DK		 1801 if (de > to)
1802 memmove(to, de, rec_len);
3d0518f4 1803 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
ac27a0ec 1804 prev = to;
617ba13b 1805 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
ac27a0ec
DK		 1806 }
1807 de = next;
1808 }
1809 return prev;
1810}
1811
ef2b02d3
ES		 1812/*
1813 * Split a full leaf block to make room for a new dir entry.
1814 * Allocate a new block, and move entries so that they are approx. equally full.
1815 * Returns pointer to de in block into which the new entry will be inserted.
1816 */
617ba13b 1817static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
ac27a0ec 1818 struct buffer_head **bh,struct dx_frame *frame,
f8b3b59d 1819 struct dx_hash_info *hinfo)
ac27a0ec
DK		 1820{
1821 unsigned blocksize = dir->i_sb->s_blocksize;
1822 unsigned count, continued;
1823 struct buffer_head *bh2;
725d26d3 1824 ext4_lblk_t newblock;
ac27a0ec
DK		 1825 u32 hash2;
1826 struct dx_map_entry *map;
1827 char *data1 = (*bh)->b_data, *data2;
59e315b4 1828 unsigned split, move, size;
617ba13b 1829 struct ext4_dir_entry_2 *de = NULL, *de2;
b0336e8d 1830 int csum_size = 0;
59e315b4 1831 int err = 0, i;
ac27a0ec 1832
9aa5d32b 1833 if (ext4_has_metadata_csum(dir->i_sb))
b0336e8d
DW		 1834 csum_size = sizeof(struct ext4_dir_entry_tail);
1835
0f70b406
TT		 1836 bh2 = ext4_append(handle, dir, &newblock);
1837 if (IS_ERR(bh2)) {
ac27a0ec
DK		 1838 brelse(*bh);
1839 *bh = NULL;
f8b3b59d 1840 return (struct ext4_dir_entry_2 *) bh2;
ac27a0ec
DK		 1841 }
1842
1843 BUFFER_TRACE(*bh, "get_write_access");
617ba13b 1844 err = ext4_journal_get_write_access(handle, *bh);
fedee54d
DM		 1845 if (err)
1846 goto journal_error;
1847
ac27a0ec 1848 BUFFER_TRACE(frame->bh, "get_write_access");
617ba13b 1849 err = ext4_journal_get_write_access(handle, frame->bh);
ac27a0ec
DK		 1850 if (err)
1851 goto journal_error;
1852
1853 data2 = bh2->b_data;
1854
1855 /* create map in the end of data2 block */
1856 map = (struct dx_map_entry *) (data2 + blocksize);
1f3862b5 1857 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
ac27a0ec
DK		 1858 blocksize, hinfo, map);
1859 map -= count;
af5bc92d 1860 dx_sort_map(map, count);
5872331b 1861 /* Ensure that neither split block is over half full */
ef2b02d3
ES		 1862 size = 0;
1863 move = 0;
1864 for (i = count-1; i >= 0; i--) {
1865 /* is more than half of this entry in 2nd half of the block? */
1866 if (size + map[i].size/2 > blocksize/2)
1867 break;
1868 size += map[i].size;
1869 move++;
1870 }
5872331b
ES		 1871 /*
1872 * map index at which we will split
1873 *
1874 * If the sum of active entries didn't exceed half the block size, just
1875 * split it in half by count; each resulting block will have at least
1876 * half the space free.
1877 */
1878 if (i > 0)
1879 split = count - move;
1880 else
1881 split = count/2;
1882
ac27a0ec
DK		 1883 hash2 = map[split].hash;
1884 continued = hash2 == map[split - 1].hash;
725d26d3
AK		 1885 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1886 (unsigned long)dx_get_block(frame->at),
1887 hash2, split, count-split));
ac27a0ec
DK		 1888
1889 /* Fancy dance to stay within two buffers */
1f3862b5
MH		 1890 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1891 blocksize);
af5bc92d 1892 de = dx_pack_dirents(data1, blocksize);
b0336e8d
DW		 1893 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1894 (char *) de,
3d0518f4 1895 blocksize);
b0336e8d
DW		 1896 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1897 (char *) de2,
3d0518f4 1898 blocksize);
b0336e8d 1899 if (csum_size) {
ddce3b94
TT		 1900 ext4_initialize_dirent_tail(*bh, blocksize);
1901 ext4_initialize_dirent_tail(bh2, blocksize);
b0336e8d
DW		 1902 }
1903
b3098486
MH		 1904 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1905 blocksize, 1));
1906 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1907 blocksize, 1));
ac27a0ec
DK		 1908
1909 /* Which block gets the new entry? */
f8b3b59d 1910 if (hinfo->hash >= hash2) {
ac27a0ec
DK		 1911 swap(*bh, bh2);
1912 de = de2;
1913 }
af5bc92d 1914 dx_insert_block(frame, hash2 + continued, newblock);
f036adb3 1915 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
ac27a0ec
DK		 1916 if (err)
1917 goto journal_error;
dbe89444 1918 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
ac27a0ec
DK		 1919 if (err)
1920 goto journal_error;
af5bc92d
TT		 1921 brelse(bh2);
1922 dxtrace(dx_show_index("frame", frame->entries));
ac27a0ec 1923 return de;
fedee54d
DM		 1924
1925journal_error:
1926 brelse(*bh);
1927 brelse(bh2);
1928 *bh = NULL;
1929 ext4_std_error(dir->i_sb, err);
f8b3b59d 1930 return ERR_PTR(err);
ac27a0ec 1931}
ac27a0ec 1932
978fef91
TM		 1933int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1934 struct buffer_head *bh,
1935 void *buf, int buf_size,
5b643f9c 1936 struct ext4_filename *fname,
978fef91
TM		 1937 struct ext4_dir_entry_2 **dest_de)
1938{
1939 struct ext4_dir_entry_2 *de;
5b643f9c 1940 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
978fef91
TM		 1941 int nlen, rlen;
1942 unsigned int offset = 0;
1943 char *top;
1f3862b5 1944
978fef91
TM		 1945 de = (struct ext4_dir_entry_2 *)buf;
1946 top = buf + buf_size - reclen;
1947 while ((char *) de <= top) {
1948 if (ext4_check_dir_entry(dir, NULL, de, bh,
d9b9f8d5
EB		 1949 buf, buf_size, offset))
1950 return -EFSCORRUPTED;
b886ee3e 1951 if (ext4_match(dir, fname, de))
d9b9f8d5 1952 return -EEXIST;
978fef91
TM		 1953 nlen = EXT4_DIR_REC_LEN(de->name_len);
1954 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1955 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1956 break;
1957 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1958 offset += rlen;
1959 }
1f3862b5 1960 if ((char *) de > top)
d9b9f8d5
EB		 1961 return -ENOSPC;
1962
1963 *dest_de = de;
1964 return 0;
978fef91
TM		 1965}
1966
1bc0af60
EB		 1967void ext4_insert_dentry(struct inode *inode,
1968 struct ext4_dir_entry_2 *de,
1969 int buf_size,
1970 struct ext4_filename *fname)
978fef91
TM		 1971{
1972
1973 int nlen, rlen;
1974
1975 nlen = EXT4_DIR_REC_LEN(de->name_len);
1976 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1977 if (de->inode) {
1978 struct ext4_dir_entry_2 *de1 =
4bdfc873 1979 (struct ext4_dir_entry_2 *)((char *)de + nlen);
978fef91
TM		 1980 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1981 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1982 de = de1;
1983 }
1984 de->file_type = EXT4_FT_UNKNOWN;
1985 de->inode = cpu_to_le32(inode->i_ino);
1986 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
5b643f9c
TT		 1987 de->name_len = fname_len(fname);
1988 memcpy(de->name, fname_name(fname), fname_len(fname));
978fef91 1989}
4bdfc873 1990
ac27a0ec
DK		 1991/*
1992 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1993 * it points to a directory entry which is guaranteed to be large
1994 * enough for new directory entry. If de is NULL, then
1995 * add_dirent_to_buf will attempt search the directory block for
1996 * space. It will return -ENOSPC if no space is available, and -EIO
1997 * and -EEXIST if directory entry already exists.
ac27a0ec 1998 */
5b643f9c
TT		 1999static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2000 struct inode *dir,
617ba13b 2001 struct inode *inode, struct ext4_dir_entry_2 *de,
af5bc92d 2002 struct buffer_head *bh)
ac27a0ec 2003{
3d0518f4 2004 unsigned int blocksize = dir->i_sb->s_blocksize;
b0336e8d 2005 int csum_size = 0;
4209ae12 2006 int err, err2;
b0336e8d 2007
9aa5d32b 2008 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 2009 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec 2010
ac27a0ec 2011 if (!de) {
5b643f9c
TT		 2012 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2013 blocksize - csum_size, fname, &de);
978fef91
TM		 2014 if (err)
2015 return err;
ac27a0ec
DK		 2016 }
2017 BUFFER_TRACE(bh, "get_write_access");
617ba13b 2018 err = ext4_journal_get_write_access(handle, bh);
ac27a0ec 2019 if (err) {
617ba13b 2020 ext4_std_error(dir->i_sb, err);
ac27a0ec
DK		 2021 return err;
2022 }
2023
1bc0af60
EB		 2024 /* By now the buffer is marked for journaling */
2025 ext4_insert_dentry(inode, de, blocksize, fname);
978fef91 2026
ac27a0ec
DK		 2027 /*
2028 * XXX shouldn't update any times until successful
2029 * completion of syscall, but too many callers depend
2030 * on this.
2031 *
2032 * XXX similarly, too many callers depend on
617ba13b 2033 * ext4_new_inode() setting the times, but error
ac27a0ec
DK		 2034 * recovery deletes the inode, so the worst that can
2035 * happen is that the times are slightly out of date
2036 * and/or different from the directory change time.
2037 */
eeca7ea1 2038 dir->i_mtime = dir->i_ctime = current_time(dir);
617ba13b 2039 ext4_update_dx_flag(dir);
e08ac99f 2040 inode_inc_iversion(dir);
4209ae12 2041 err2 = ext4_mark_inode_dirty(handle, dir);
0390131b 2042 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
f036adb3 2043 err = ext4_handle_dirty_dirblock(handle, dir, bh);
ac27a0ec 2044 if (err)
617ba13b 2045 ext4_std_error(dir->i_sb, err);
4209ae12 2046 return err ? err : err2;
ac27a0ec
DK		 2047}
2048
ac27a0ec
DK		 2049/*
2050 * This converts a one block unindexed directory to a 3 block indexed
2051 * directory, and adds the dentry to the indexed directory.
2052 */
5b643f9c 2053static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
56a04915 2054 struct inode *dir,
ac27a0ec
DK		 2055 struct inode *inode, struct buffer_head *bh)
2056{
ac27a0ec
DK		 2057 struct buffer_head *bh2;
2058 struct dx_root *root;
e08ac99f 2059 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 2060 struct dx_entry *entries;
617ba13b 2061 struct ext4_dir_entry_2 *de, *de2;
ddce3b94 2062 char *data2, *top;
ac27a0ec
DK		 2063 unsigned len;
2064 int retval;
2065 unsigned blocksize;
725d26d3 2066 ext4_lblk_t block;
ac27a0ec 2067 struct fake_dirent *fde;
4bdfc873
MH		 2068 int csum_size = 0;
2069
9aa5d32b 2070 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 2071 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec
DK		 2072
2073 blocksize = dir->i_sb->s_blocksize;
e6b8bc09 2074 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
5d601255 2075 BUFFER_TRACE(bh, "get_write_access");
617ba13b 2076 retval = ext4_journal_get_write_access(handle, bh);
ac27a0ec 2077 if (retval) {
617ba13b 2078 ext4_std_error(dir->i_sb, retval);
ac27a0ec
DK		 2079 brelse(bh);
2080 return retval;
2081 }
2082 root = (struct dx_root *) bh->b_data;
2083
e6b8bc09
TT		 2084 /* The 0th block becomes the root, move the dirents out */
2085 fde = &root->dotdot;
2086 de = (struct ext4_dir_entry_2 *)((char *)fde +
3d0518f4 2087 ext4_rec_len_from_disk(fde->rec_len, blocksize));
e6b8bc09 2088 if ((char *) de >= (((char *) root) + blocksize)) {
24676da4 2089 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
e6b8bc09 2090 brelse(bh);
6a797d27 2091 return -EFSCORRUPTED;
e6b8bc09 2092 }
b0336e8d 2093 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
e6b8bc09
TT		 2094
2095 /* Allocate new block for the 0th block's dirents */
0f70b406
TT		 2096 bh2 = ext4_append(handle, dir, &block);
2097 if (IS_ERR(bh2)) {
ac27a0ec 2098 brelse(bh);
0f70b406 2099 return PTR_ERR(bh2);
ac27a0ec 2100 }
12e9b892 2101 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
ddce3b94 2102 data2 = bh2->b_data;
ac27a0ec 2103
ddce3b94
TT		 2104 memcpy(data2, de, len);
2105 de = (struct ext4_dir_entry_2 *) data2;
2106 top = data2 + len;
3d0518f4 2107 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
ac27a0ec 2108 de = de2;
ddce3b94
TT		 2109 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2110 (char *) de, blocksize);
b0336e8d 2111
ddce3b94
TT		 2112 if (csum_size)
2113 ext4_initialize_dirent_tail(bh2, blocksize);
b0336e8d 2114
ac27a0ec 2115 /* Initialize the root; the dot dirents already exist */
617ba13b 2116 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
3d0518f4
WY		 2117 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2118 blocksize);
ac27a0ec
DK		 2119 memset (&root->info, 0, sizeof(root->info));
2120 root->info.info_length = sizeof(root->info);
617ba13b 2121 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
ac27a0ec 2122 entries = root->entries;
af5bc92d
TT		 2123 dx_set_block(entries, 1);
2124 dx_set_count(entries, 1);
2125 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
ac27a0ec
DK		 2126
2127 /* Initialize as for dx_probe */
5b643f9c
TT		 2128 fname->hinfo.hash_version = root->info.hash_version;
2129 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2130 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2131 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
b886ee3e 2132 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
5b643f9c 2133
6050d47a 2134 memset(frames, 0, sizeof(frames));
ac27a0ec
DK		 2135 frame = frames;
2136 frame->entries = entries;
2137 frame->at = entries;
2138 frame->bh = bh;
6976a6f2 2139
6050d47a
JK		 2140 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2141 if (retval)
2142 goto out_frames;
f036adb3 2143 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
6050d47a
JK		 2144 if (retval)
2145 goto out_frames;
6976a6f2 2146
e81d4477 2147 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
f8b3b59d 2148 if (IS_ERR(de)) {
6050d47a
JK		 2149 retval = PTR_ERR(de);
2150 goto out_frames;
7ad8e4e6 2151 }
ac27a0ec 2152
e81d4477 2153 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
6050d47a
JK		 2154out_frames:
2155 /*
2156 * Even if the block split failed, we have to properly write
2157 * out all the changes we did so far. Otherwise we can end up
2158 * with corrupted filesystem.
2159 */
e81d4477 2160 if (retval)
2161 ext4_mark_inode_dirty(handle, dir);
6050d47a 2162 dx_release(frames);
e81d4477 2163 brelse(bh2);
6050d47a 2164 return retval;
ac27a0ec 2165}
ac27a0ec
DK		 2166
2167/*
617ba13b 2168 * ext4_add_entry()
ac27a0ec
DK		 2169 *
2170 * adds a file entry to the specified directory, using the same
617ba13b 2171 * semantics as ext4_find_entry(). It returns NULL if it failed.
ac27a0ec
DK		 2172 *
2173 * NOTE!! The inode part of 'de' is left at 0 - which means you
2174 * may not sleep between calling this and putting something into
2175 * the entry, as someone else might have used it while you slept.
2176 */
af5bc92d
TT		 2177static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2178 struct inode *inode)
ac27a0ec 2179{
2b0143b5 2180 struct inode *dir = d_inode(dentry->d_parent);
e12fb972 2181 struct buffer_head *bh = NULL;
617ba13b 2182 struct ext4_dir_entry_2 *de;
af5bc92d 2183 struct super_block *sb;
68d7b2d8 2184#ifdef CONFIG_UNICODE
b886ee3e 2185 struct ext4_sb_info *sbi;
68d7b2d8 2186#endif
5b643f9c 2187 struct ext4_filename fname;
ac27a0ec 2188 int retval;
ac27a0ec 2189 int dx_fallback=0;
ac27a0ec 2190 unsigned blocksize;
725d26d3 2191 ext4_lblk_t block, blocks;
b0336e8d
DW		 2192 int csum_size = 0;
2193
9aa5d32b 2194 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 2195 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec
DK		 2196
2197 sb = dir->i_sb;
2198 blocksize = sb->s_blocksize;
2199 if (!dentry->d_name.len)
2200 return -EINVAL;
3c47d541 2201
b886ee3e 2202#ifdef CONFIG_UNICODE
68d7b2d8 2203 sbi = EXT4_SB(sb);
b886ee3e 2204 if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
6456ca65 2205 sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
b886ee3e
GKB		 2206 return -EINVAL;
2207#endif
2208
5b643f9c
TT		 2209 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2210 if (retval)
2211 return retval;
2212
3c47d541 2213 if (ext4_has_inline_data(dir)) {
56a04915 2214 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
3c47d541 2215 if (retval < 0)
5b643f9c 2216 goto out;
3c47d541
TM		 2217 if (retval == 1) {
2218 retval = 0;
e12fb972 2219 goto out;
3c47d541
TM		 2220 }
2221 }
2222
ac27a0ec 2223 if (is_dx(dir)) {
56a04915 2224 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
ac27a0ec 2225 if (!retval || (retval != ERR_BAD_DX_DIR))
e12fb972 2226 goto out;
48a34311
JK		 2227 /* Can we just ignore htree data? */
2228 if (ext4_has_metadata_csum(sb)) {
2229 EXT4_ERROR_INODE(dir,
2230 "Directory has corrupted htree index.");
2231 retval = -EFSCORRUPTED;
2232 goto out;
2233 }
12e9b892 2234 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
ac27a0ec 2235 dx_fallback++;
4209ae12
HS		 2236 retval = ext4_mark_inode_dirty(handle, dir);
2237 if (unlikely(retval))
2238 goto out;
ac27a0ec 2239 }
ac27a0ec 2240 blocks = dir->i_size >> sb->s_blocksize_bits;
498e5f24 2241 for (block = 0; block < blocks; block++) {
dc6982ff 2242 bh = ext4_read_dirblock(dir, block, DIRENT);
4e19d6b6
TT		 2243 if (bh == NULL) {
2244 bh = ext4_bread(handle, dir, block,
2245 EXT4_GET_BLOCKS_CREATE);
2246 goto add_to_new_block;
2247 }
5b643f9c
TT		 2248 if (IS_ERR(bh)) {
2249 retval = PTR_ERR(bh);
2250 bh = NULL;
2251 goto out;
2252 }
2253 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2254 NULL, bh);
e12fb972
LC		 2255 if (retval != -ENOSPC)
2256 goto out;
ac27a0ec 2257
ac27a0ec 2258 if (blocks == 1 && !dx_fallback &&
e2b911c5 2259 ext4_has_feature_dir_index(sb)) {
56a04915 2260 retval = make_indexed_dir(handle, &fname, dir,
5b643f9c 2261 inode, bh);
e12fb972
LC		 2262 bh = NULL; /* make_indexed_dir releases bh */
2263 goto out;
2264 }
ac27a0ec
DK		 2265 brelse(bh);
2266 }
0f70b406 2267 bh = ext4_append(handle, dir, &block);
4e19d6b6 2268add_to_new_block:
5b643f9c
TT		 2269 if (IS_ERR(bh)) {
2270 retval = PTR_ERR(bh);
2271 bh = NULL;
2272 goto out;
2273 }
617ba13b 2274 de = (struct ext4_dir_entry_2 *) bh->b_data;
ac27a0ec 2275 de->inode = 0;
b0336e8d
DW		 2276 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2277
ddce3b94
TT		 2278 if (csum_size)
2279 ext4_initialize_dirent_tail(bh, blocksize);
b0336e8d 2280
5b643f9c 2281 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
e12fb972 2282out:
5b643f9c 2283 ext4_fname_free_filename(&fname);
2de770a4 2284 brelse(bh);
14ece102
FM		 2285 if (retval == 0)
2286 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2de770a4 2287 return retval;
ac27a0ec
DK		 2288}
2289
ac27a0ec
DK		 2290/*
2291 * Returns 0 for success, or a negative error value
2292 */
5b643f9c 2293static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
56a04915 2294 struct inode *dir, struct inode *inode)
ac27a0ec 2295{
e08ac99f 2296 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 2297 struct dx_entry *entries, *at;
af5bc92d 2298 struct buffer_head *bh;
af5bc92d 2299 struct super_block *sb = dir->i_sb;
617ba13b 2300 struct ext4_dir_entry_2 *de;
e08ac99f 2301 int restart;
ac27a0ec
DK		 2302 int err;
2303
e08ac99f
AB		 2304again:
2305 restart = 0;
5b643f9c 2306 frame = dx_probe(fname, dir, NULL, frames);
dd73b5d5
TT		 2307 if (IS_ERR(frame))
2308 return PTR_ERR(frame);
ac27a0ec
DK		 2309 entries = frame->entries;
2310 at = frame->at;
4e19d6b6 2311 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
dc6982ff
TT		 2312 if (IS_ERR(bh)) {
2313 err = PTR_ERR(bh);
2314 bh = NULL;
ac27a0ec 2315 goto cleanup;
6d1ab10e 2316 }
ac27a0ec
DK		 2317
2318 BUFFER_TRACE(bh, "get_write_access");
617ba13b 2319 err = ext4_journal_get_write_access(handle, bh);
ac27a0ec
DK		 2320 if (err)
2321 goto journal_error;
2322
5b643f9c 2323 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2de770a4 2324 if (err != -ENOSPC)
ac27a0ec 2325 goto cleanup;
ac27a0ec 2326
e08ac99f 2327 err = 0;
ac27a0ec 2328 /* Block full, should compress but for now just split */
4776004f 2329 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
ac27a0ec
DK		 2330 dx_get_count(entries), dx_get_limit(entries)));
2331 /* Need to split index? */
2332 if (dx_get_count(entries) == dx_get_limit(entries)) {
725d26d3 2333 ext4_lblk_t newblock;
e08ac99f
AB		 2334 int levels = frame - frames + 1;
2335 unsigned int icount;
2336 int add_level = 1;
ac27a0ec
DK		 2337 struct dx_entry *entries2;
2338 struct dx_node *node2;
2339 struct buffer_head *bh2;
2340
e08ac99f
AB		 2341 while (frame > frames) {
2342 if (dx_get_count((frame - 1)->entries) <
2343 dx_get_limit((frame - 1)->entries)) {
2344 add_level = 0;
2345 break;
2346 }
2347 frame--; /* split higher index block */
2348 at = frame->at;
2349 entries = frame->entries;
2350 restart = 1;
2351 }
2352 if (add_level && levels == ext4_dir_htree_level(sb)) {
2353 ext4_warning(sb, "Directory (ino: %lu) index full, "
2354 "reach max htree level :%d",
2355 dir->i_ino, levels);
2356 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2357 ext4_warning(sb, "Large directory feature is "
2358 "not enabled on this "
2359 "filesystem");
2360 }
ac27a0ec
DK		 2361 err = -ENOSPC;
2362 goto cleanup;
2363 }
e08ac99f 2364 icount = dx_get_count(entries);
0f70b406
TT		 2365 bh2 = ext4_append(handle, dir, &newblock);
2366 if (IS_ERR(bh2)) {
2367 err = PTR_ERR(bh2);
ac27a0ec 2368 goto cleanup;
0f70b406 2369 }
ac27a0ec
DK		 2370 node2 = (struct dx_node *)(bh2->b_data);
2371 entries2 = node2->entries;
1f7bebb9 2372 memset(&node2->fake, 0, sizeof(struct fake_dirent));
3d0518f4
WY		 2373 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2374 sb->s_blocksize);
ac27a0ec 2375 BUFFER_TRACE(frame->bh, "get_write_access");
617ba13b 2376 err = ext4_journal_get_write_access(handle, frame->bh);
ac27a0ec
DK		 2377 if (err)
2378 goto journal_error;
e08ac99f 2379 if (!add_level) {
ac27a0ec
DK		 2380 unsigned icount1 = icount/2, icount2 = icount - icount1;
2381 unsigned hash2 = dx_get_hash(entries + icount1);
4776004f
TT		 2382 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2383 icount1, icount2));
ac27a0ec
DK		 2384
2385 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
617ba13b 2386 err = ext4_journal_get_write_access(handle,
e08ac99f 2387 (frame - 1)->bh);
ac27a0ec
DK		 2388 if (err)
2389 goto journal_error;
2390
af5bc92d
TT		 2391 memcpy((char *) entries2, (char *) (entries + icount1),
2392 icount2 * sizeof(struct dx_entry));
2393 dx_set_count(entries, icount1);
2394 dx_set_count(entries2, icount2);
2395 dx_set_limit(entries2, dx_node_limit(dir));
ac27a0ec
DK		 2396
2397 /* Which index block gets the new entry? */
2398 if (at - entries >= icount1) {
2399 frame->at = at = at - entries - icount1 + entries2;
2400 frame->entries = entries = entries2;
2401 swap(frame->bh, bh2);
2402 }
e08ac99f
AB		 2403 dx_insert_block((frame - 1), hash2, newblock);
2404 dxtrace(dx_show_index("node", frame->entries));
af5bc92d 2405 dxtrace(dx_show_index("node",
ac27a0ec 2406 ((struct dx_node *) bh2->b_data)->entries));
dbe89444 2407 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
ac27a0ec
DK		 2408 if (err)
2409 goto journal_error;
2410 brelse (bh2);
e08ac99f
AB		 2411 err = ext4_handle_dirty_dx_node(handle, dir,
2412 (frame - 1)->bh);
2413 if (err)
2414 goto journal_error;
2415 if (restart) {
2416 err = ext4_handle_dirty_dx_node(handle, dir,
2417 frame->bh);
2418 goto journal_error;
2419 }
ac27a0ec 2420 } else {
e08ac99f 2421 struct dx_root *dxroot;
ac27a0ec
DK		 2422 memcpy((char *) entries2, (char *) entries,
2423 icount * sizeof(struct dx_entry));
2424 dx_set_limit(entries2, dx_node_limit(dir));
2425
2426 /* Set up root */
2427 dx_set_count(entries, 1);
2428 dx_set_block(entries + 0, newblock);
e08ac99f
AB		 2429 dxroot = (struct dx_root *)frames[0].bh->b_data;
2430 dxroot->info.indirect_levels += 1;
2431 dxtrace(printk(KERN_DEBUG
2432 "Creating %d level index...\n",
799578ab 2433 dxroot->info.indirect_levels));
e08ac99f 2434 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
ac27a0ec
DK		 2435 if (err)
2436 goto journal_error;
e08ac99f
AB		 2437 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2438 brelse(bh2);
2439 restart = 1;
2440 goto journal_error;
b4097142 2441 }
ac27a0ec 2442 }
5b643f9c 2443 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
f8b3b59d
TT		 2444 if (IS_ERR(de)) {
2445 err = PTR_ERR(de);
ac27a0ec 2446 goto cleanup;
f8b3b59d 2447 }
5b643f9c 2448 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
ac27a0ec
DK		 2449 goto cleanup;
2450
2451journal_error:
e08ac99f 2452 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
ac27a0ec 2453cleanup:
b1deefc9 2454 brelse(bh);
ac27a0ec 2455 dx_release(frames);
e08ac99f
AB		 2456 /* @restart is true means htree-path has been changed, we need to
2457 * repeat dx_probe() to find out valid htree-path
2458 */
2459 if (restart && err == 0)
2460 goto again;
ac27a0ec
DK		 2461 return err;
2462}
ac27a0ec
DK		 2463
2464/*
05019a9e
TM		 2465 * ext4_generic_delete_entry deletes a directory entry by merging it
2466 * with the previous entry
ac27a0ec 2467 */
2fe34d29 2468int ext4_generic_delete_entry(struct inode *dir,
05019a9e
TM		 2469 struct ext4_dir_entry_2 *de_del,
2470 struct buffer_head *bh,
2471 void *entry_buf,
2472 int buf_size,
2473 int csum_size)
ac27a0ec 2474{
af5bc92d 2475 struct ext4_dir_entry_2 *de, *pde;
3d0518f4 2476 unsigned int blocksize = dir->i_sb->s_blocksize;
05019a9e 2477 int i;
b0336e8d 2478
ac27a0ec
DK		 2479 i = 0;
2480 pde = NULL;
05019a9e
TM		 2481 de = (struct ext4_dir_entry_2 *)entry_buf;
2482 while (i < buf_size - csum_size) {
226ba972 2483 if (ext4_check_dir_entry(dir, NULL, de, bh,
7303cb5b 2484 entry_buf, buf_size, i))
6a797d27 2485 return -EFSCORRUPTED;
ac27a0ec 2486 if (de == de_del) {
ac27a0ec 2487 if (pde)
a72d7f83 2488 pde->rec_len = ext4_rec_len_to_disk(
3d0518f4
WY		 2489 ext4_rec_len_from_disk(pde->rec_len,
2490 blocksize) +
2491 ext4_rec_len_from_disk(de->rec_len,
2492 blocksize),
2493 blocksize);
ac27a0ec
DK		 2494 else
2495 de->inode = 0;
e08ac99f 2496 inode_inc_iversion(dir);
ac27a0ec
DK		 2497 return 0;
2498 }
3d0518f4 2499 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
ac27a0ec 2500 pde = de;
3d0518f4 2501 de = ext4_next_entry(de, blocksize);
ac27a0ec
DK		 2502 }
2503 return -ENOENT;
2504}
2505
05019a9e
TM		 2506static int ext4_delete_entry(handle_t *handle,
2507 struct inode *dir,
2508 struct ext4_dir_entry_2 *de_del,
2509 struct buffer_head *bh)
2510{
2511 int err, csum_size = 0;
2512
9f40fe54
TM		 2513 if (ext4_has_inline_data(dir)) {
2514 int has_inline_data = 1;
2515 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2516 &has_inline_data);
2517 if (has_inline_data)
2518 return err;
2519 }
2520
9aa5d32b 2521 if (ext4_has_metadata_csum(dir->i_sb))
05019a9e
TM		 2522 csum_size = sizeof(struct ext4_dir_entry_tail);
2523
2524 BUFFER_TRACE(bh, "get_write_access");
2525 err = ext4_journal_get_write_access(handle, bh);
2526 if (unlikely(err))
2527 goto out;
2528
2fe34d29 2529 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
05019a9e
TM		 2530 dir->i_sb->s_blocksize, csum_size);
2531 if (err)
2532 goto out;
2533
2534 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
f036adb3 2535 err = ext4_handle_dirty_dirblock(handle, dir, bh);
05019a9e
TM		 2536 if (unlikely(err))
2537 goto out;
2538
2539 return 0;
2540out:
2541 if (err != -ENOENT)
2542 ext4_std_error(dir->i_sb, err);
2543 return err;
2544}
2545
f8628a14 2546/*
c7414892
AD		 2547 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2548 * since this indicates that nlinks count was previously 1 to avoid overflowing
2549 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2550 * that subdirectory link counts are not being maintained accurately.
2551 *
2552 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2553 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2554 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2555 * on regular files) and to avoid creating huge/slow non-HTREE directories.
f8628a14 2556 */
15ed2851 2557static void ext4_inc_count(struct inode *inode)
f8628a14
AD		 2558{
2559 inc_nlink(inode);
c7414892
AD		 2560 if (is_dx(inode) &&
2561 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2562 set_nlink(inode, 1);
f8628a14
AD		 2563}
2564
2565/*
2566 * If a directory had nlink == 1, then we should let it be 1. This indicates
2567 * directory has >EXT4_LINK_MAX subdirs.
2568 */
15ed2851 2569static void ext4_dec_count(struct inode *inode)
f8628a14 2570{
909a4cf1
AD		 2571 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2572 drop_nlink(inode);
f8628a14
AD		 2573}
2574
2575
9b88f9fb
JK		 2576/*
2577 * Add non-directory inode to a directory. On success, the inode reference is
2578 * consumed by dentry is instantiation. This is also indicated by clearing of
2579 * *inodep pointer. On failure, the caller is responsible for dropping the
2580 * inode reference in the safe context.
2581 */
617ba13b 2582static int ext4_add_nondir(handle_t *handle,
9b88f9fb 2583 struct dentry *dentry, struct inode **inodep)
ac27a0ec 2584{
a9e26328 2585 struct inode *dir = d_inode(dentry->d_parent);
9b88f9fb 2586 struct inode *inode = *inodep;
617ba13b 2587 int err = ext4_add_entry(handle, dentry, inode);
ac27a0ec 2588 if (!err) {
4209ae12 2589 err = ext4_mark_inode_dirty(handle, inode);
a9e26328
JK		 2590 if (IS_DIRSYNC(dir))
2591 ext4_handle_sync(handle);
1e2e547a 2592 d_instantiate_new(dentry, inode);
9b88f9fb 2593 *inodep = NULL;
4209ae12 2594 return err;
ac27a0ec 2595 }
731b9a54 2596 drop_nlink(inode);
9b88f9fb 2597 ext4_orphan_add(handle, inode);
6b38e842 2598 unlock_new_inode(inode);
ac27a0ec
DK		 2599 return err;
2600}
2601
2602/*
2603 * By the time this is called, we already have created
2604 * the directory cache entry for the new file, but it
2605 * is so far negative - it has no inode.
2606 *
2607 * If the create succeeds, we fill in the inode information
2608 * with d_instantiate().
2609 */
4acdaf27 2610static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
ebfc3b49 2611 bool excl)
ac27a0ec
DK		 2612{
2613 handle_t *handle;
aa75f4d3 2614 struct inode *inode, *inode_save;
1139575a 2615 int err, credits, retries = 0;
ac27a0ec 2616
a7cdadee
JK		 2617 err = dquot_initialize(dir);
2618 if (err)
2619 return err;
907f4554 2620
1139575a 2621 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2622 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2623retry:
1139575a
TT		 2624 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2625 NULL, EXT4_HT_DIR, credits);
2626 handle = ext4_journal_current_handle();
ac27a0ec
DK		 2627 err = PTR_ERR(inode);
2628 if (!IS_ERR(inode)) {
617ba13b 2629 inode->i_op = &ext4_file_inode_operations;
be64f884 2630 inode->i_fop = &ext4_file_operations;
617ba13b 2631 ext4_set_aops(inode);
aa75f4d3
HS		 2632 inode_save = inode;
2633 ihold(inode_save);
9b88f9fb 2634 err = ext4_add_nondir(handle, dentry, &inode);
aa75f4d3
HS		 2635 ext4_fc_track_create(inode_save, dentry);
2636 iput(inode_save);
ac27a0ec 2637 }
1139575a
TT		 2638 if (handle)
2639 ext4_journal_stop(handle);
9b88f9fb
JK		 2640 if (!IS_ERR_OR_NULL(inode))
2641 iput(inode);
617ba13b 2642 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK		 2643 goto retry;
2644 return err;
2645}
2646
af5bc92d 2647static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 2648 umode_t mode, dev_t rdev)
ac27a0ec
DK		 2649{
2650 handle_t *handle;
aa75f4d3 2651 struct inode *inode, *inode_save;
1139575a 2652 int err, credits, retries = 0;
ac27a0ec 2653
a7cdadee
JK		 2654 err = dquot_initialize(dir);
2655 if (err)
2656 return err;
907f4554 2657
1139575a 2658 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2659 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2660retry:
1139575a
TT		 2661 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2662 NULL, EXT4_HT_DIR, credits);
2663 handle = ext4_journal_current_handle();
ac27a0ec
DK		 2664 err = PTR_ERR(inode);
2665 if (!IS_ERR(inode)) {
2666 init_special_inode(inode, inode->i_mode, rdev);
617ba13b 2667 inode->i_op = &ext4_special_inode_operations;
aa75f4d3
HS		 2668 inode_save = inode;
2669 ihold(inode_save);
9b88f9fb 2670 err = ext4_add_nondir(handle, dentry, &inode);
aa75f4d3
HS		 2671 if (!err)
2672 ext4_fc_track_create(inode_save, dentry);
2673 iput(inode_save);
ac27a0ec 2674 }
1139575a
TT		 2675 if (handle)
2676 ext4_journal_stop(handle);
9b88f9fb
JK		 2677 if (!IS_ERR_OR_NULL(inode))
2678 iput(inode);
617ba13b 2679 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK		 2680 goto retry;
2681 return err;
2682}
2683
af51a2ac
AV		 2684static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2685{
2686 handle_t *handle;
2687 struct inode *inode;
2688 int err, retries = 0;
2689
a7cdadee
JK		 2690 err = dquot_initialize(dir);
2691 if (err)
2692 return err;
af51a2ac
AV		 2693
2694retry:
2695 inode = ext4_new_inode_start_handle(dir, mode,
2696 NULL, 0, NULL,
2697 EXT4_HT_DIR,
2698 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2699 4 + EXT4_XATTR_TRANS_BLOCKS);
2700 handle = ext4_journal_current_handle();
2701 err = PTR_ERR(inode);
2702 if (!IS_ERR(inode)) {
2703 inode->i_op = &ext4_file_inode_operations;
be64f884 2704 inode->i_fop = &ext4_file_operations;
af51a2ac 2705 ext4_set_aops(inode);
e94bd349 2706 d_tmpfile(dentry, inode);
af51a2ac
AV		 2707 err = ext4_orphan_add(handle, inode);
2708 if (err)
43ae9e3f 2709 goto err_unlock_inode;
af51a2ac 2710 mark_inode_dirty(inode);
af51a2ac
AV		 2711 unlock_new_inode(inode);
2712 }
2713 if (handle)
2714 ext4_journal_stop(handle);
2715 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2716 goto retry;
2717 return err;
43ae9e3f 2718err_unlock_inode:
af51a2ac
AV		 2719 ext4_journal_stop(handle);
2720 unlock_new_inode(inode);
af51a2ac
AV		 2721 return err;
2722}
2723
a774f9c2
TM		 2724struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2725 struct ext4_dir_entry_2 *de,
2726 int blocksize, int csum_size,
2727 unsigned int parent_ino, int dotdot_real_len)
2728{
2729 de->inode = cpu_to_le32(inode->i_ino);
2730 de->name_len = 1;
2731 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2732 blocksize);
2733 strcpy(de->name, ".");
2734 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2735
2736 de = ext4_next_entry(de, blocksize);
2737 de->inode = cpu_to_le32(parent_ino);
2738 de->name_len = 2;
2739 if (!dotdot_real_len)
2740 de->rec_len = ext4_rec_len_to_disk(blocksize -
2741 (csum_size + EXT4_DIR_REC_LEN(1)),
2742 blocksize);
2743 else
2744 de->rec_len = ext4_rec_len_to_disk(
2745 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2746 strcpy(de->name, "..");
2747 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2748
2749 return ext4_next_entry(de, blocksize);
2750}
2751
2752static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2753 struct inode *inode)
ac27a0ec 2754{
dabd991f 2755 struct buffer_head *dir_block = NULL;
af5bc92d 2756 struct ext4_dir_entry_2 *de;
dc6982ff 2757 ext4_lblk_t block = 0;
3d0518f4 2758 unsigned int blocksize = dir->i_sb->s_blocksize;
b0336e8d 2759 int csum_size = 0;
a774f9c2 2760 int err;
ac27a0ec 2761
9aa5d32b 2762 if (ext4_has_metadata_csum(dir->i_sb))
b0336e8d
DW		 2763 csum_size = sizeof(struct ext4_dir_entry_tail);
2764
3c47d541
TM		 2765 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2766 err = ext4_try_create_inline_dir(handle, dir, inode);
2767 if (err < 0 && err != -ENOSPC)
2768 goto out;
2769 if (!err)
2770 goto out;
2771 }
2772
dc6982ff 2773 inode->i_size = 0;
0f70b406
TT		 2774 dir_block = ext4_append(handle, inode, &block);
2775 if (IS_ERR(dir_block))
2776 return PTR_ERR(dir_block);
a774f9c2
TM		 2777 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2778 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2779 set_nlink(inode, 2);
ddce3b94
TT		 2780 if (csum_size)
2781 ext4_initialize_dirent_tail(dir_block, blocksize);
a774f9c2
TM		 2782
2783 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
f036adb3 2784 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
a774f9c2
TM		 2785 if (err)
2786 goto out;
2787 set_buffer_verified(dir_block);
2788out:
2789 brelse(dir_block);
2790 return err;
2791}
2792
2793static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2794{
2795 handle_t *handle;
2796 struct inode *inode;
4209ae12 2797 int err, err2 = 0, credits, retries = 0;
a774f9c2 2798
f8628a14 2799 if (EXT4_DIR_LINK_MAX(dir))
ac27a0ec
DK		 2800 return -EMLINK;
2801
a7cdadee
JK		 2802 err = dquot_initialize(dir);
2803 if (err)
2804 return err;
907f4554 2805
1139575a 2806 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2807 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2808retry:
1139575a
TT		 2809 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2810 &dentry->d_name,
2811 0, NULL, EXT4_HT_DIR, credits);
2812 handle = ext4_journal_current_handle();
ac27a0ec
DK		 2813 err = PTR_ERR(inode);
2814 if (IS_ERR(inode))
2815 goto out_stop;
2816
617ba13b
MC		 2817 inode->i_op = &ext4_dir_inode_operations;
2818 inode->i_fop = &ext4_dir_operations;
a774f9c2 2819 err = ext4_init_new_dir(handle, dir, inode);
dabd991f
NK		 2820 if (err)
2821 goto out_clear_inode;
2822 err = ext4_mark_inode_dirty(handle, inode);
2823 if (!err)
2824 err = ext4_add_entry(handle, dentry, inode);
ac27a0ec 2825 if (err) {
4cdeed86
AK		 2826out_clear_inode:
2827 clear_nlink(inode);
9b88f9fb 2828 ext4_orphan_add(handle, inode);
6b38e842 2829 unlock_new_inode(inode);
4209ae12
HS		 2830 err2 = ext4_mark_inode_dirty(handle, inode);
2831 if (unlikely(err2))
2832 err = err2;
9b88f9fb 2833 ext4_journal_stop(handle);
af5bc92d 2834 iput(inode);
9b88f9fb 2835 goto out_retry;
ac27a0ec 2836 }
aa75f4d3 2837 ext4_fc_track_create(inode, dentry);
15ed2851 2838 ext4_inc_count(dir);
aa75f4d3 2839
617ba13b 2840 ext4_update_dx_flag(dir);
dabd991f
NK		 2841 err = ext4_mark_inode_dirty(handle, dir);
2842 if (err)
2843 goto out_clear_inode;
1e2e547a 2844 d_instantiate_new(dentry, inode);
1139575a
TT		 2845 if (IS_DIRSYNC(dir))
2846 ext4_handle_sync(handle);
2847
ac27a0ec 2848out_stop:
1139575a
TT		 2849 if (handle)
2850 ext4_journal_stop(handle);
9b88f9fb 2851out_retry:
617ba13b 2852 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK		 2853 goto retry;
2854 return err;
2855}
2856
2857/*
2858 * routine to check that the specified directory is empty (for rmdir)
2859 */
a7550b30 2860bool ext4_empty_dir(struct inode *inode)
ac27a0ec 2861{
498e5f24 2862 unsigned int offset;
af5bc92d 2863 struct buffer_head *bh;
64d4ce89 2864 struct ext4_dir_entry_2 *de;
af5bc92d 2865 struct super_block *sb;
ac27a0ec 2866
61f86638
TM		 2867 if (ext4_has_inline_data(inode)) {
2868 int has_inline_data = 1;
a7550b30 2869 int ret;
61f86638 2870
a7550b30 2871 ret = empty_inline_dir(inode, &has_inline_data);
61f86638 2872 if (has_inline_data)
a7550b30 2873 return ret;
61f86638
TM		 2874 }
2875
ac27a0ec 2876 sb = inode->i_sb;
dc6982ff
TT		 2877 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2878 EXT4_ERROR_INODE(inode, "invalid size");
a7550b30 2879 return true;
ac27a0ec 2880 }
4e19d6b6
TT		 2881 /* The first directory block must not be a hole,
2882 * so treat it as DIRENT_HTREE
2883 */
2884 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
dc6982ff 2885 if (IS_ERR(bh))
a7550b30 2886 return true;
dc6982ff 2887
617ba13b 2888 de = (struct ext4_dir_entry_2 *) bh->b_data;
64d4ce89
JK		 2889 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2890 0) ||
2891 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2892 ext4_warning_inode(inode, "directory missing '.'");
2893 brelse(bh);
2894 return true;
2895 }
2896 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2897 de = ext4_next_entry(de, sb->s_blocksize);
2898 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2899 offset) ||
2900 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2901 ext4_warning_inode(inode, "directory missing '..'");
af5bc92d 2902 brelse(bh);
a7550b30 2903 return true;
ac27a0ec 2904 }
64d4ce89 2905 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
af5bc92d 2906 while (offset < inode->i_size) {
64d4ce89 2907 if (!(offset & (sb->s_blocksize - 1))) {
24676da4 2908 unsigned int lblock;
af5bc92d 2909 brelse(bh);
24676da4 2910 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
dc6982ff 2911 bh = ext4_read_dirblock(inode, lblock, EITHER);
4e19d6b6
TT		 2912 if (bh == NULL) {
2913 offset += sb->s_blocksize;
2914 continue;
2915 }
dc6982ff 2916 if (IS_ERR(bh))
a7550b30 2917 return true;
ac27a0ec 2918 }
64d4ce89
JK		 2919 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2920 (offset & (sb->s_blocksize - 1)));
226ba972
TM		 2921 if (ext4_check_dir_entry(inode, NULL, de, bh,
2922 bh->b_data, bh->b_size, offset)) {
ac27a0ec
DK		 2923 offset = (offset | (sb->s_blocksize - 1)) + 1;
2924 continue;
2925 }
2926 if (le32_to_cpu(de->inode)) {
af5bc92d 2927 brelse(bh);
a7550b30 2928 return false;
ac27a0ec 2929 }
3d0518f4 2930 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
ac27a0ec 2931 }
af5bc92d 2932 brelse(bh);
a7550b30 2933 return true;
ac27a0ec
DK		 2934}
2935
d745a8c2
JK		 2936/*
2937 * ext4_orphan_add() links an unlinked or truncated inode into a list of
ac27a0ec
DK		 2938 * such inodes, starting at the superblock, in case we crash before the
2939 * file is closed/deleted, or in case the inode truncate spans multiple
2940 * transactions and the last transaction is not recovered after a crash.
2941 *
2942 * At filesystem recovery time, we walk this list deleting unlinked
617ba13b 2943 * inodes and truncating linked inodes in ext4_orphan_cleanup().
d745a8c2
JK		 2944 *
2945 * Orphan list manipulation functions must be called under i_mutex unless
2946 * we are just creating the inode or deleting it.
ac27a0ec 2947 */
617ba13b 2948int ext4_orphan_add(handle_t *handle, struct inode *inode)
ac27a0ec
DK		 2949{
2950 struct super_block *sb = inode->i_sb;
cd2c080c 2951 struct ext4_sb_info *sbi = EXT4_SB(sb);
617ba13b 2952 struct ext4_iloc iloc;
ac27a0ec 2953 int err = 0, rc;
d745a8c2 2954 bool dirty = false;
ac27a0ec 2955
e2bfb088 2956 if (!sbi->s_journal || is_bad_inode(inode))
0390131b
FM		 2957 return 0;
2958
d745a8c2 2959 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
5955102c 2960 !inode_is_locked(inode));
d745a8c2
JK		 2961 /*
2962 * Exit early if inode already is on orphan list. This is a big speedup
2963 * since we don't have to contend on the global s_orphan_lock.
2964 */
617ba13b 2965 if (!list_empty(&EXT4_I(inode)->i_orphan))
d745a8c2 2966 return 0;
ac27a0ec 2967
afb86178
LC		 2968 /*
2969 * Orphan handling is only valid for files with data blocks
2970 * being truncated, or files being unlinked. Note that we either
2971 * hold i_mutex, or the inode can not be referenced from outside,
2972 * so i_nlink should not be bumped due to race
ac27a0ec 2973 */
af5bc92d
TT		 2974 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2975 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
ac27a0ec 2976
cd2c080c
JK		 2977 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2978 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
ac27a0ec 2979 if (err)
d745a8c2 2980 goto out;
ac27a0ec 2981
617ba13b 2982 err = ext4_reserve_inode_write(handle, inode, &iloc);
ac27a0ec 2983 if (err)
d745a8c2
JK		 2984 goto out;
2985
2986 mutex_lock(&sbi->s_orphan_lock);
6e3617e5
DM		 2987 /*
2988 * Due to previous errors inode may be already a part of on-disk
2989 * orphan list. If so skip on-disk list modification.
2990 */
d745a8c2
JK		 2991 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2992 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2993 /* Insert this inode at the head of the on-disk orphan list */
2994 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2995 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2996 dirty = true;
2997 }
2998 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2999 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 3000
d745a8c2
JK		 3001 if (dirty) {
3002 err = ext4_handle_dirty_super(handle, sb);
3003 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
3004 if (!err)
3005 err = rc;
3006 if (err) {
3007 /*
3008 * We have to remove inode from in-memory list if
3009 * addition to on disk orphan list failed. Stray orphan
3010 * list entries can cause panics at unmount time.
3011 */
3012 mutex_lock(&sbi->s_orphan_lock);
74177f55 3013 list_del_init(&EXT4_I(inode)->i_orphan);
d745a8c2
JK		 3014 mutex_unlock(&sbi->s_orphan_lock);
3015 }
feaf264c
VA		 3016 } else
3017 brelse(iloc.bh);
3018
ac27a0ec
DK		 3019 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3020 jbd_debug(4, "orphan inode %lu will point to %d\n",
3021 inode->i_ino, NEXT_ORPHAN(inode));
d745a8c2 3022out:
cd2c080c 3023 ext4_std_error(sb, err);
ac27a0ec
DK		 3024 return err;
3025}
3026
3027/*
617ba13b 3028 * ext4_orphan_del() removes an unlinked or truncated inode from the list
ac27a0ec
DK		 3029 * of such inodes stored on disk, because it is finally being cleaned up.
3030 */
617ba13b 3031int ext4_orphan_del(handle_t *handle, struct inode *inode)
ac27a0ec
DK		 3032{
3033 struct list_head *prev;
617ba13b 3034 struct ext4_inode_info *ei = EXT4_I(inode);
cd2c080c 3035 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
498e5f24 3036 __u32 ino_next;
617ba13b 3037 struct ext4_iloc iloc;
ac27a0ec
DK		 3038 int err = 0;
3039
cd2c080c 3040 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
0390131b
FM		 3041 return 0;
3042
d745a8c2 3043 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
5955102c 3044 !inode_is_locked(inode));
d745a8c2 3045 /* Do this quick check before taking global s_orphan_lock. */
3b9d4ed2 3046 if (list_empty(&ei->i_orphan))
d745a8c2 3047 return 0;
ac27a0ec 3048
d745a8c2
JK		 3049 if (handle) {
3050 /* Grab inode buffer early before taking global s_orphan_lock */
3051 err = ext4_reserve_inode_write(handle, inode, &iloc);
3052 }
ac27a0ec 3053
d745a8c2 3054 mutex_lock(&sbi->s_orphan_lock);
ac27a0ec
DK		 3055 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3056
d745a8c2 3057 prev = ei->i_orphan.prev;
ac27a0ec
DK		 3058 list_del_init(&ei->i_orphan);
3059
3060 /* If we're on an error path, we may not have a valid
3061 * transaction handle with which to update the orphan list on
3062 * disk, but we still need to remove the inode from the linked
3063 * list in memory. */
d745a8c2
JK		 3064 if (!handle || err) {
3065 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 3066 goto out_err;
d745a8c2 3067 }
ac27a0ec 3068
d745a8c2 3069 ino_next = NEXT_ORPHAN(inode);
ac27a0ec 3070 if (prev == &sbi->s_orphan) {
498e5f24 3071 jbd_debug(4, "superblock will point to %u\n", ino_next);
ac27a0ec 3072 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
617ba13b 3073 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
d745a8c2
JK		 3074 if (err) {
3075 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 3076 goto out_brelse;
d745a8c2 3077 }
ac27a0ec 3078 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
d745a8c2 3079 mutex_unlock(&sbi->s_orphan_lock);
b50924c2 3080 err = ext4_handle_dirty_super(handle, inode->i_sb);
ac27a0ec 3081 } else {
617ba13b 3082 struct ext4_iloc iloc2;
ac27a0ec 3083 struct inode *i_prev =
617ba13b 3084 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
ac27a0ec 3085
498e5f24 3086 jbd_debug(4, "orphan inode %lu will point to %u\n",
ac27a0ec 3087 i_prev->i_ino, ino_next);
617ba13b 3088 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
d745a8c2
JK		 3089 if (err) {
3090 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 3091 goto out_brelse;
d745a8c2 3092 }
ac27a0ec 3093 NEXT_ORPHAN(i_prev) = ino_next;
617ba13b 3094 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
d745a8c2 3095 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec
DK		 3096 }
3097 if (err)
3098 goto out_brelse;
3099 NEXT_ORPHAN(inode) = 0;
617ba13b 3100 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
ac27a0ec 3101out_err:
617ba13b 3102 ext4_std_error(inode->i_sb, err);
ac27a0ec
DK		 3103 return err;
3104
3105out_brelse:
3106 brelse(iloc.bh);
3107 goto out_err;
3108}
3109
af5bc92d 3110static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
ac27a0ec
DK		 3111{
3112 int retval;
af5bc92d
TT		 3113 struct inode *inode;
3114 struct buffer_head *bh;
3115 struct ext4_dir_entry_2 *de;
8dcfaad2 3116 handle_t *handle = NULL;
ac27a0ec 3117
0db1ff22
TT		 3118 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3119 return -EIO;
3120
ac27a0ec
DK		 3121 /* Initialize quotas before so that eventual writes go in
3122 * separate transaction */
a7cdadee
JK		 3123 retval = dquot_initialize(dir);
3124 if (retval)
3125 return retval;
3126 retval = dquot_initialize(d_inode(dentry));
3127 if (retval)
3128 return retval;
907f4554 3129
ac27a0ec 3130 retval = -ENOENT;
32f7f22c 3131 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
36de9286
TT		 3132 if (IS_ERR(bh))
3133 return PTR_ERR(bh);
ac27a0ec
DK		 3134 if (!bh)
3135 goto end_rmdir;
3136
2b0143b5 3137 inode = d_inode(dentry);
ac27a0ec 3138
6a797d27 3139 retval = -EFSCORRUPTED;
ac27a0ec
DK		 3140 if (le32_to_cpu(de->inode) != inode->i_ino)
3141 goto end_rmdir;
3142
3143 retval = -ENOTEMPTY;
e875a2dd 3144 if (!ext4_empty_dir(inode))
ac27a0ec
DK		 3145 goto end_rmdir;
3146
8dcfaad2 3147 handle = ext4_journal_start(dir, EXT4_HT_DIR,
64044abf 3148 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
8dcfaad2
TT		 3149 if (IS_ERR(handle)) {
3150 retval = PTR_ERR(handle);
3151 handle = NULL;
3152 goto end_rmdir;
3153 }
3154
3155 if (IS_DIRSYNC(dir))
3156 ext4_handle_sync(handle);
3157
617ba13b 3158 retval = ext4_delete_entry(handle, dir, de, bh);
ac27a0ec
DK		 3159 if (retval)
3160 goto end_rmdir;
f8628a14 3161 if (!EXT4_DIR_LINK_EMPTY(inode))
b03a2f7e
AD		 3162 ext4_warning_inode(inode,
3163 "empty directory '%.*s' has too many links (%u)",
3164 dentry->d_name.len, dentry->d_name.name,
af5bc92d 3165 inode->i_nlink);
ee73f9a5 3166 inode_inc_iversion(inode);
ac27a0ec
DK		 3167 clear_nlink(inode);
3168 /* There's no need to set i_disksize: the fact that i_nlink is
3169 * zero will ensure that the right thing happens during any
3170 * recovery. */
3171 inode->i_size = 0;
617ba13b 3172 ext4_orphan_add(handle, inode);
eeca7ea1 3173 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
4209ae12
HS		 3174 retval = ext4_mark_inode_dirty(handle, inode);
3175 if (retval)
3176 goto end_rmdir;
15ed2851 3177 ext4_dec_count(dir);
617ba13b 3178 ext4_update_dx_flag(dir);
aa75f4d3 3179 ext4_fc_track_unlink(inode, dentry);
4209ae12 3180 retval = ext4_mark_inode_dirty(handle, dir);
ac27a0ec 3181
b886ee3e
GKB		 3182#ifdef CONFIG_UNICODE
3183 /* VFS negative dentries are incompatible with Encoding and
3184 * Case-insensitiveness. Eventually we'll want avoid
3185 * invalidating the dentries here, alongside with returning the
3186 * negative dentries at ext4_lookup(), when it is better
3187 * supported by the VFS for the CI case.
3188 */
3189 if (IS_CASEFOLDED(dir))
3190 d_invalidate(dentry);
3191#endif
3192
ac27a0ec 3193end_rmdir:
af5bc92d 3194 brelse(bh);
8dcfaad2
TT		 3195 if (handle)
3196 ext4_journal_stop(handle);
ac27a0ec
DK		 3197 return retval;
3198}
3199
af5bc92d 3200static int ext4_unlink(struct inode *dir, struct dentry *dentry)
ac27a0ec
DK		 3201{
3202 int retval;
af5bc92d
TT		 3203 struct inode *inode;
3204 struct buffer_head *bh;
3205 struct ext4_dir_entry_2 *de;
931b6864 3206 handle_t *handle = NULL;
ac27a0ec 3207
0db1ff22
TT		 3208 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3209 return -EIO;
3210
0562e0ba 3211 trace_ext4_unlink_enter(dir, dentry);
ac27a0ec
DK		 3212 /* Initialize quotas before so that eventual writes go
3213 * in separate transaction */
a7cdadee
JK		 3214 retval = dquot_initialize(dir);
3215 if (retval)
e5f78159 3216 goto out_trace;
a7cdadee
JK		 3217 retval = dquot_initialize(d_inode(dentry));
3218 if (retval)
e5f78159 3219 goto out_trace;
907f4554 3220
32f7f22c 3221 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
e5f78159
YZ		 3222 if (IS_ERR(bh)) {
3223 retval = PTR_ERR(bh);
3224 goto out_trace;
3225 }
3226 if (!bh) {
3227 retval = -ENOENT;
3228 goto out_trace;
3229 }
ac27a0ec 3230
2b0143b5 3231 inode = d_inode(dentry);
ac27a0ec 3232
e5f78159
YZ		 3233 if (le32_to_cpu(de->inode) != inode->i_ino) {
3234 retval = -EFSCORRUPTED;
3235 goto out_bh;
3236 }
ac27a0ec 3237
931b6864 3238 handle = ext4_journal_start(dir, EXT4_HT_DIR,
64044abf 3239 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
931b6864
TT		 3240 if (IS_ERR(handle)) {
3241 retval = PTR_ERR(handle);
e5f78159 3242 goto out_bh;
931b6864
TT		 3243 }
3244
3245 if (IS_DIRSYNC(dir))
3246 ext4_handle_sync(handle);
3247
617ba13b 3248 retval = ext4_delete_entry(handle, dir, de, bh);
ac27a0ec 3249 if (retval)
e5f78159 3250 goto out_handle;
eeca7ea1 3251 dir->i_ctime = dir->i_mtime = current_time(dir);
617ba13b 3252 ext4_update_dx_flag(dir);
4209ae12
HS		 3253 retval = ext4_mark_inode_dirty(handle, dir);
3254 if (retval)
e5f78159 3255 goto out_handle;
c7df4a1e
TT		 3256 if (inode->i_nlink == 0)
3257 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3258 dentry->d_name.len, dentry->d_name.name);
3259 else
3260 drop_nlink(inode);
ac27a0ec 3261 if (!inode->i_nlink)
617ba13b 3262 ext4_orphan_add(handle, inode);
eeca7ea1 3263 inode->i_ctime = current_time(inode);
4209ae12 3264 retval = ext4_mark_inode_dirty(handle, inode);
ac27a0ec 3265
aa75f4d3
HS		 3266 if (!retval)
3267 ext4_fc_track_unlink(d_inode(dentry), dentry);
b886ee3e
GKB		 3268#ifdef CONFIG_UNICODE
3269 /* VFS negative dentries are incompatible with Encoding and
3270 * Case-insensitiveness. Eventually we'll want avoid
3271 * invalidating the dentries here, alongside with returning the
3272 * negative dentries at ext4_lookup(), when it is better
3273 * supported by the VFS for the CI case.
3274 */
3275 if (IS_CASEFOLDED(dir))
3276 d_invalidate(dentry);
3277#endif
3278
e5f78159
YZ		 3279out_handle:
3280 ext4_journal_stop(handle);
3281out_bh:
af5bc92d 3282 brelse(bh);
e5f78159 3283out_trace:
0562e0ba 3284 trace_ext4_unlink_exit(dentry, retval);
ac27a0ec
DK		 3285 return retval;
3286}
3287
af5bc92d
TT		 3288static int ext4_symlink(struct inode *dir,
3289 struct dentry *dentry, const char *symname)
ac27a0ec
DK		 3290{
3291 handle_t *handle;
af5bc92d 3292 struct inode *inode;
f348c252 3293 int err, len = strlen(symname);
df5e6223 3294 int credits;
a7550b30 3295 struct fscrypt_str disk_link;
ac27a0ec 3296
0db1ff22
TT		 3297 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3298 return -EIO;
3299
78e1060c
EB		 3300 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3301 &disk_link);
3302 if (err)
3303 return err;
ac27a0ec 3304
a7cdadee
JK		 3305 err = dquot_initialize(dir);
3306 if (err)
78e1060c 3307 return err;
907f4554 3308
f348c252 3309 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
df5e6223
JK		 3310 /*
3311 * For non-fast symlinks, we just allocate inode and put it on
3312 * orphan list in the first transaction => we need bitmap,
8c208719
ES		 3313 * group descriptor, sb, inode block, quota blocks, and
3314 * possibly selinux xattr blocks.
df5e6223 3315 */
8c208719
ES		 3316 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3317 EXT4_XATTR_TRANS_BLOCKS;
df5e6223
JK		 3318 } else {
3319 /*
3320 * Fast symlink. We have to add entry to directory
3321 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3322 * allocate new inode (bitmap, group descriptor, inode block,
3323 * quota blocks, sb is already counted in previous macros).
3324 */
3325 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 3326 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
df5e6223 3327 }
f348c252 3328
1139575a
TT		 3329 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3330 &dentry->d_name, 0, NULL,
3331 EXT4_HT_DIR, credits);
3332 handle = ext4_journal_current_handle();
f348c252
TT		 3333 if (IS_ERR(inode)) {
3334 if (handle)
3335 ext4_journal_stop(handle);
78e1060c 3336 return PTR_ERR(inode);
f348c252
TT		 3337 }
3338
78e1060c
EB		 3339 if (IS_ENCRYPTED(inode)) {
3340 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
ef1eb3aa 3341 if (err)
f348c252 3342 goto err_drop_inode;
a7a67e8a 3343 inode->i_op = &ext4_encrypted_symlink_inode_operations;
f348c252 3344 }
ac27a0ec 3345
f348c252 3346 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
78e1060c 3347 if (!IS_ENCRYPTED(inode))
a7a67e8a 3348 inode->i_op = &ext4_symlink_inode_operations;
21fc61c7 3349 inode_nohighmem(inode);
617ba13b 3350 ext4_set_aops(inode);
ac27a0ec 3351 /*
df5e6223
JK		 3352 * We cannot call page_symlink() with transaction started
3353 * because it calls into ext4_write_begin() which can wait
3354 * for transaction commit if we are running out of space
3355 * and thus we deadlock. So we have to stop transaction now
3356 * and restart it when symlink contents is written.
3357 *
3358 * To keep fs consistent in case of crash, we have to put inode
3359 * to orphan list in the mean time.
ac27a0ec 3360 */
df5e6223
JK		 3361 drop_nlink(inode);
3362 err = ext4_orphan_add(handle, inode);
3363 ext4_journal_stop(handle);
f348c252 3364 handle = NULL;
df5e6223
JK		 3365 if (err)
3366 goto err_drop_inode;
f348c252 3367 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
df5e6223
JK		 3368 if (err)
3369 goto err_drop_inode;
3370 /*
3371 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3372 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3373 */
9924a92a 3374 handle = ext4_journal_start(dir, EXT4_HT_DIR,
df5e6223
JK		 3375 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3376 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3377 if (IS_ERR(handle)) {
3378 err = PTR_ERR(handle);
f348c252 3379 handle = NULL;
df5e6223
JK		 3380 goto err_drop_inode;
3381 }
0ce8c010 3382 set_nlink(inode, 1);
df5e6223 3383 err = ext4_orphan_del(handle, inode);
f348c252 3384 if (err)
df5e6223 3385 goto err_drop_inode;
ac27a0ec 3386 } else {
e65187e6 3387 /* clear the extent format for fast symlink */
12e9b892 3388 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
78e1060c 3389 if (!IS_ENCRYPTED(inode)) {
a7a67e8a 3390 inode->i_op = &ext4_fast_symlink_inode_operations;
75e7566b
AV		 3391 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3392 }
f348c252
TT		 3393 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3394 disk_link.len);
3395 inode->i_size = disk_link.len - 1;
ac27a0ec 3396 }
617ba13b 3397 EXT4_I(inode)->i_disksize = inode->i_size;
9b88f9fb 3398 err = ext4_add_nondir(handle, dentry, &inode);
1139575a
TT		 3399 if (handle)
3400 ext4_journal_stop(handle);
9b88f9fb
JK		 3401 if (inode)
3402 iput(inode);
78e1060c
EB		 3403 goto out_free_encrypted_link;
3404
df5e6223 3405err_drop_inode:
f348c252
TT		 3406 if (handle)
3407 ext4_journal_stop(handle);
f348c252 3408 clear_nlink(inode);
df5e6223
JK		 3409 unlock_new_inode(inode);
3410 iput(inode);
78e1060c
EB		 3411out_free_encrypted_link:
3412 if (disk_link.name != (unsigned char *)symname)
3413 kfree(disk_link.name);
df5e6223 3414 return err;
ac27a0ec
DK		 3415}
3416
af5bc92d
TT		 3417static int ext4_link(struct dentry *old_dentry,
3418 struct inode *dir, struct dentry *dentry)
ac27a0ec
DK		 3419{
3420 handle_t *handle;
2b0143b5 3421 struct inode *inode = d_inode(old_dentry);
ac27a0ec
DK		 3422 int err, retries = 0;
3423
b05ab1dc 3424 if (inode->i_nlink >= EXT4_LINK_MAX)
ac27a0ec 3425 return -EMLINK;
69725181
EB		 3426
3427 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3428 if (err)
3429 return err;
040cb378 3430
a794df0e
CIK		 3431 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3432 (!projid_eq(EXT4_I(dir)->i_projid,
3433 EXT4_I(old_dentry->d_inode)->i_projid)))
040cb378
LX		 3434 return -EXDEV;
3435
a7cdadee
JK		 3436 err = dquot_initialize(dir);
3437 if (err)
3438 return err;
907f4554 3439
ac27a0ec 3440retry:
9924a92a
TT		 3441 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3442 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
af51a2ac 3443 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
ac27a0ec
DK		 3444 if (IS_ERR(handle))
3445 return PTR_ERR(handle);
3446
3447 if (IS_DIRSYNC(dir))
0390131b 3448 ext4_handle_sync(handle);
ac27a0ec 3449
eeca7ea1 3450 inode->i_ctime = current_time(inode);
15ed2851 3451 ext4_inc_count(inode);
7de9c6ee 3452 ihold(inode);
ac27a0ec 3453
6b38e842
AV		 3454 err = ext4_add_entry(handle, dentry, inode);
3455 if (!err) {
4209ae12 3456 err = ext4_mark_inode_dirty(handle, inode);
af51a2ac
AV		 3457 /* this can happen only for tmpfile being
3458 * linked the first time
3459 */
3460 if (inode->i_nlink == 1)
3461 ext4_orphan_del(handle, inode);
6b38e842
AV		 3462 d_instantiate(dentry, inode);
3463 } else {
3464 drop_nlink(inode);
3465 iput(inode);
3466 }
617ba13b
MC		 3467 ext4_journal_stop(handle);
3468 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK		 3469 goto retry;
3470 return err;
3471}
3472
32f7f22c
TM		 3473
3474/*
3475 * Try to find buffer head where contains the parent block.
3476 * It should be the inode block if it is inlined or the 1st block
3477 * if it is a normal dir.
3478 */
3479static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3480 struct inode *inode,
3481 int *retval,
3482 struct ext4_dir_entry_2 **parent_de,
3483 int *inlined)
3484{
3485 struct buffer_head *bh;
3486
3487 if (!ext4_has_inline_data(inode)) {
4e19d6b6
TT		 3488 /* The first directory block must not be a hole, so
3489 * treat it as DIRENT_HTREE
3490 */
3491 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
dc6982ff
TT		 3492 if (IS_ERR(bh)) {
3493 *retval = PTR_ERR(bh);
32f7f22c
TM		 3494 return NULL;
3495 }
3496 *parent_de = ext4_next_entry(
3497 (struct ext4_dir_entry_2 *)bh->b_data,
3498 inode->i_sb->s_blocksize);
3499 return bh;
3500 }
3501
3502 *inlined = 1;
3503 return ext4_get_first_inline_block(inode, parent_de, retval);
3504}
ac27a0ec 3505
c0d268c3
MS		 3506struct ext4_renament {
3507 struct inode *dir;
3508 struct dentry *dentry;
3509 struct inode *inode;
bd42998a
MS		 3510 bool is_dir;
3511 int dir_nlink_delta;
c0d268c3
MS		 3512
3513 /* entry for "dentry" */
3514 struct buffer_head *bh;
3515 struct ext4_dir_entry_2 *de;
3516 int inlined;
3517
3518 /* entry for ".." in inode if it's a directory */
3519 struct buffer_head *dir_bh;
3520 struct ext4_dir_entry_2 *parent_de;
3521 int dir_inlined;
3522};
3523
bd1af145
MS		 3524static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3525{
3526 int retval;
3527
3528 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3529 &retval, &ent->parent_de,
3530 &ent->dir_inlined);
3531 if (!ent->dir_bh)
3532 return retval;
3533 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
6a797d27 3534 return -EFSCORRUPTED;
bd1af145
MS		 3535 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3536 return ext4_journal_get_write_access(handle, ent->dir_bh);
3537}
3538
3539static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3540 unsigned dir_ino)
3541{
3542 int retval;
3543
3544 ent->parent_de->inode = cpu_to_le32(dir_ino);
3545 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3546 if (!ent->dir_inlined) {
3547 if (is_dx(ent->inode)) {
3548 retval = ext4_handle_dirty_dx_node(handle,
3549 ent->inode,
3550 ent->dir_bh);
3551 } else {
f036adb3
TT		 3552 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3553 ent->dir_bh);
bd1af145
MS		 3554 }
3555 } else {
3556 retval = ext4_mark_inode_dirty(handle, ent->inode);
3557 }
3558 if (retval) {
3559 ext4_std_error(ent->dir->i_sb, retval);
3560 return retval;
3561 }
3562 return 0;
3563}
3564
3565static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3566 unsigned ino, unsigned file_type)
3567{
4209ae12 3568 int retval, retval2;
bd1af145
MS		 3569
3570 BUFFER_TRACE(ent->bh, "get write access");
3571 retval = ext4_journal_get_write_access(handle, ent->bh);
3572 if (retval)
3573 return retval;
3574 ent->de->inode = cpu_to_le32(ino);
e2b911c5 3575 if (ext4_has_feature_filetype(ent->dir->i_sb))
bd1af145 3576 ent->de->file_type = file_type;
ee73f9a5 3577 inode_inc_iversion(ent->dir);
bd1af145 3578 ent->dir->i_ctime = ent->dir->i_mtime =
eeca7ea1 3579 current_time(ent->dir);
4209ae12 3580 retval = ext4_mark_inode_dirty(handle, ent->dir);
bd1af145
MS		 3581 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3582 if (!ent->inlined) {
4209ae12
HS		 3583 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3584 if (unlikely(retval2)) {
3585 ext4_std_error(ent->dir->i_sb, retval2);
3586 return retval2;
bd1af145
MS		 3587 }
3588 }
3589 brelse(ent->bh);
3590 ent->bh = NULL;
3591
4209ae12 3592 return retval;
bd1af145
MS		 3593}
3594
3595static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3596 const struct qstr *d_name)
3597{
3598 int retval = -ENOENT;
3599 struct buffer_head *bh;
3600 struct ext4_dir_entry_2 *de;
3601
3602 bh = ext4_find_entry(dir, d_name, &de, NULL);
36de9286
TT		 3603 if (IS_ERR(bh))
3604 return PTR_ERR(bh);
bd1af145
MS		 3605 if (bh) {
3606 retval = ext4_delete_entry(handle, dir, de, bh);
3607 brelse(bh);
3608 }
3609 return retval;
3610}
3611
d80d448c
DW		 3612static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3613 int force_reread)
bd1af145
MS		 3614{
3615 int retval;
3616 /*
3617 * ent->de could have moved from under us during htree split, so make
3618 * sure that we are deleting the right entry. We might also be pointing
3619 * to a stale entry in the unused part of ent->bh so just checking inum
3620 * and the name isn't enough.
3621 */
3622 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3623 ent->de->name_len != ent->dentry->d_name.len ||
3624 strncmp(ent->de->name, ent->dentry->d_name.name,
d80d448c
DW		 3625 ent->de->name_len) ||
3626 force_reread) {
bd1af145
MS		 3627 retval = ext4_find_delete_entry(handle, ent->dir,
3628 &ent->dentry->d_name);
3629 } else {
3630 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3631 if (retval == -ENOENT) {
3632 retval = ext4_find_delete_entry(handle, ent->dir,
3633 &ent->dentry->d_name);
3634 }
3635 }
3636
3637 if (retval) {
b03a2f7e
AD		 3638 ext4_warning_inode(ent->dir,
3639 "Deleting old file: nlink %d, error=%d",
3640 ent->dir->i_nlink, retval);
bd1af145
MS		 3641 }
3642}
3643
bd42998a
MS		 3644static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3645{
3646 if (ent->dir_nlink_delta) {
3647 if (ent->dir_nlink_delta == -1)
15ed2851 3648 ext4_dec_count(ent->dir);
bd42998a 3649 else
15ed2851 3650 ext4_inc_count(ent->dir);
bd42998a
MS		 3651 ext4_mark_inode_dirty(handle, ent->dir);
3652 }
3653}
3654
cd808dec
MS		 3655static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3656 int credits, handle_t **h)
3657{
3658 struct inode *wh;
3659 handle_t *handle;
3660 int retries = 0;
3661
3662 /*
3663 * for inode block, sb block, group summaries,
3664 * and inode bitmap
3665 */
3666 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3667 EXT4_XATTR_TRANS_BLOCKS + 4);
3668retry:
3669 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3670 &ent->dentry->d_name, 0, NULL,
3671 EXT4_HT_DIR, credits);
3672
3673 handle = ext4_journal_current_handle();
3674 if (IS_ERR(wh)) {
3675 if (handle)
3676 ext4_journal_stop(handle);
3677 if (PTR_ERR(wh) == -ENOSPC &&
3678 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3679 goto retry;
3680 } else {
3681 *h = handle;
3682 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3683 wh->i_op = &ext4_special_inode_operations;
3684 }
3685 return wh;
3686}
3687
ac27a0ec
DK		 3688/*
3689 * Anybody can rename anything with this: the permission checks are left to the
3690 * higher-level routines.
0e202704
TT		 3691 *
3692 * n.b. old_{dentry,inode) refers to the source dentry/inode
3693 * while new_{dentry,inode) refers to the destination dentry/inode
3694 * This comes from rename(const char *oldpath, const char *newpath)
ac27a0ec 3695 */
af5bc92d 3696static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
cd808dec
MS		 3697 struct inode *new_dir, struct dentry *new_dentry,
3698 unsigned int flags)
ac27a0ec 3699{
5b61de75 3700 handle_t *handle = NULL;
c0d268c3
MS		 3701 struct ext4_renament old = {
3702 .dir = old_dir,
3703 .dentry = old_dentry,
2b0143b5 3704 .inode = d_inode(old_dentry),
c0d268c3
MS		 3705 };
3706 struct ext4_renament new = {
3707 .dir = new_dir,
3708 .dentry = new_dentry,
2b0143b5 3709 .inode = d_inode(new_dentry),
c0d268c3 3710 };
d80d448c 3711 int force_reread;
0e202704 3712 int retval;
cd808dec
MS		 3713 struct inode *whiteout = NULL;
3714 int credits;
3715 u8 old_file_type;
907f4554 3716
b50282f3
TT		 3717 if (new.inode && new.inode->i_nlink == 0) {
3718 EXT4_ERROR_INODE(new.inode,
3719 "target of rename is already freed");
3720 return -EFSCORRUPTED;
3721 }
3722
040cb378
LX		 3723 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3724 (!projid_eq(EXT4_I(new_dir)->i_projid,
3725 EXT4_I(old_dentry->d_inode)->i_projid)))
3726 return -EXDEV;
3727
a7cdadee
JK		 3728 retval = dquot_initialize(old.dir);
3729 if (retval)
3730 return retval;
3731 retval = dquot_initialize(new.dir);
3732 if (retval)
3733 return retval;
ac27a0ec
DK		 3734
3735 /* Initialize quotas before so that eventual writes go
3736 * in separate transaction */
a7cdadee
JK		 3737 if (new.inode) {
3738 retval = dquot_initialize(new.inode);
3739 if (retval)
3740 return retval;
3741 }
ac27a0ec 3742
c0d268c3 3743 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
36de9286
TT		 3744 if (IS_ERR(old.bh))
3745 return PTR_ERR(old.bh);
ac27a0ec
DK		 3746 /*
3747 * Check for inode number is _not_ due to possible IO errors.
3748 * We might rmdir the source, keep it as pwd of some process
3749 * and merrily kill the link to whatever was created under the
3750 * same name. Goodbye sticky bit ;-<
3751 */
ac27a0ec 3752 retval = -ENOENT;
c0d268c3 3753 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
ac27a0ec
DK		 3754 goto end_rename;
3755
c0d268c3
MS		 3756 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3757 &new.de, &new.inlined);
36de9286
TT		 3758 if (IS_ERR(new.bh)) {
3759 retval = PTR_ERR(new.bh);
a9cfcd63 3760 new.bh = NULL;
36de9286
TT		 3761 goto end_rename;
3762 }
c0d268c3
MS		 3763 if (new.bh) {
3764 if (!new.inode) {
3765 brelse(new.bh);
3766 new.bh = NULL;
ac27a0ec
DK		 3767 }
3768 }
c0d268c3
MS		 3769 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3770 ext4_alloc_da_blocks(old.inode);
5b61de75 3771
cd808dec
MS		 3772 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3773 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3774 if (!(flags & RENAME_WHITEOUT)) {
3775 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
7071b715
KK		 3776 if (IS_ERR(handle)) {
3777 retval = PTR_ERR(handle);
3778 handle = NULL;
3779 goto end_rename;
3780 }
cd808dec
MS		 3781 } else {
3782 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
7071b715
KK		 3783 if (IS_ERR(whiteout)) {
3784 retval = PTR_ERR(whiteout);
3785 whiteout = NULL;
3786 goto end_rename;
3787 }
cd808dec 3788 }
5b61de75 3789
c0d268c3 3790 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
5b61de75
TT		 3791 ext4_handle_sync(handle);
3792
c0d268c3
MS		 3793 if (S_ISDIR(old.inode->i_mode)) {
3794 if (new.inode) {
ac27a0ec 3795 retval = -ENOTEMPTY;
e875a2dd 3796 if (!ext4_empty_dir(new.inode))
ac27a0ec 3797 goto end_rename;
0d7d5d67
MS		 3798 } else {
3799 retval = -EMLINK;
3800 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3801 goto end_rename;
ac27a0ec 3802 }
bd1af145 3803 retval = ext4_rename_dir_prepare(handle, &old);
ef607893
AG		 3804 if (retval)
3805 goto end_rename;
ac27a0ec 3806 }
d80d448c
DW		 3807 /*
3808 * If we're renaming a file within an inline_data dir and adding or
3809 * setting the new dirent causes a conversion from inline_data to
3810 * extents/blockmap, we need to force the dirent delete code to
3811 * re-read the directory, or else we end up trying to delete a dirent
3812 * from what is now the extent tree root (or a block map).
3813 */
3814 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3815 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
cd808dec
MS		 3816
3817 old_file_type = old.de->file_type;
3818 if (whiteout) {
3819 /*
3820 * Do this before adding a new entry, so the old entry is sure
3821 * to be still pointing to the valid old entry.
3822 */
3823 retval = ext4_setent(handle, &old, whiteout->i_ino,
3824 EXT4_FT_CHRDEV);
3825 if (retval)
3826 goto end_rename;
4209ae12
HS		 3827 retval = ext4_mark_inode_dirty(handle, whiteout);
3828 if (unlikely(retval))
3829 goto end_rename;
cd808dec 3830 }
c0d268c3
MS		 3831 if (!new.bh) {
3832 retval = ext4_add_entry(handle, new.dentry, old.inode);
ac27a0ec
DK		 3833 if (retval)
3834 goto end_rename;
3835 } else {
bd1af145 3836 retval = ext4_setent(handle, &new,
cd808dec 3837 old.inode->i_ino, old_file_type);
ef607893
AG		 3838 if (retval)
3839 goto end_rename;
ac27a0ec 3840 }
d80d448c
DW		 3841 if (force_reread)
3842 force_reread = !ext4_test_inode_flag(new.dir,
3843 EXT4_INODE_INLINE_DATA);
ac27a0ec
DK		 3844
3845 /*
3846 * Like most other Unix systems, set the ctime for inodes on a
3847 * rename.
3848 */
eeca7ea1 3849 old.inode->i_ctime = current_time(old.inode);
4209ae12
HS		 3850 retval = ext4_mark_inode_dirty(handle, old.inode);
3851 if (unlikely(retval))
3852 goto end_rename;
ac27a0ec 3853
cd808dec
MS		 3854 if (!whiteout) {
3855 /*
3856 * ok, that's it
3857 */
3858 ext4_rename_delete(handle, &old, force_reread);
3859 }
ac27a0ec 3860
c0d268c3 3861 if (new.inode) {
15ed2851 3862 ext4_dec_count(new.inode);
eeca7ea1 3863 new.inode->i_ctime = current_time(new.inode);
ac27a0ec 3864 }
eeca7ea1 3865 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
c0d268c3
MS		 3866 ext4_update_dx_flag(old.dir);
3867 if (old.dir_bh) {
bd1af145
MS		 3868 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3869 if (retval)
b4097142 3870 goto end_rename;
bd1af145 3871
15ed2851 3872 ext4_dec_count(old.dir);
c0d268c3 3873 if (new.inode) {
e875a2dd
MH		 3874 /* checked ext4_empty_dir above, can't have another
3875 * parent, ext4_dec_count() won't work for many-linked
3876 * dirs */
c0d268c3 3877 clear_nlink(new.inode);
ac27a0ec 3878 } else {
15ed2851 3879 ext4_inc_count(new.dir);
c0d268c3 3880 ext4_update_dx_flag(new.dir);
4209ae12
HS		 3881 retval = ext4_mark_inode_dirty(handle, new.dir);
3882 if (unlikely(retval))
3883 goto end_rename;
ac27a0ec
DK		 3884 }
3885 }
4209ae12
HS		 3886 retval = ext4_mark_inode_dirty(handle, old.dir);
3887 if (unlikely(retval))
3888 goto end_rename;
aa75f4d3
HS		 3889
3890 if (S_ISDIR(old.inode->i_mode)) {
3891 /*
3892 * We disable fast commits here that's because the
3893 * replay code is not yet capable of changing dot dot
3894 * dirents in directories.
3895 */
3896 ext4_fc_mark_ineligible(old.inode->i_sb,
3897 EXT4_FC_REASON_RENAME_DIR);
3898 } else {
3899 if (new.inode)
3900 ext4_fc_track_unlink(new.inode, new.dentry);
3901 ext4_fc_track_link(old.inode, new.dentry);
3902 ext4_fc_track_unlink(old.inode, old.dentry);
3903 }
3904
c0d268c3 3905 if (new.inode) {
4209ae12
HS		 3906 retval = ext4_mark_inode_dirty(handle, new.inode);
3907 if (unlikely(retval))
3908 goto end_rename;
c0d268c3
MS		 3909 if (!new.inode->i_nlink)
3910 ext4_orphan_add(handle, new.inode);
ac27a0ec
DK		 3911 }
3912 retval = 0;
3913
3914end_rename:
c0d268c3
MS		 3915 brelse(old.dir_bh);
3916 brelse(old.bh);
3917 brelse(new.bh);
cd808dec
MS		 3918 if (whiteout) {
3919 if (retval)
3920 drop_nlink(whiteout);
3921 unlock_new_inode(whiteout);
3922 iput(whiteout);
3923 }
5b61de75
TT		 3924 if (handle)
3925 ext4_journal_stop(handle);
ac27a0ec
DK		 3926 return retval;
3927}
3928
bd42998a
MS		 3929static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3930 struct inode *new_dir, struct dentry *new_dentry)
3931{
3932 handle_t *handle = NULL;
3933 struct ext4_renament old = {
3934 .dir = old_dir,
3935 .dentry = old_dentry,
2b0143b5 3936 .inode = d_inode(old_dentry),
bd42998a
MS		 3937 };
3938 struct ext4_renament new = {
3939 .dir = new_dir,
3940 .dentry = new_dentry,
2b0143b5 3941 .inode = d_inode(new_dentry),
bd42998a
MS		 3942 };
3943 u8 new_file_type;
3944 int retval;
95582b00 3945 struct timespec64 ctime;
bd42998a 3946
040cb378
LX		 3947 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3948 !projid_eq(EXT4_I(new_dir)->i_projid,
3949 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3950 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3951 !projid_eq(EXT4_I(old_dir)->i_projid,
3952 EXT4_I(new_dentry->d_inode)->i_projid)))
3953 return -EXDEV;
3954
a7cdadee
JK		 3955 retval = dquot_initialize(old.dir);
3956 if (retval)
3957 return retval;
3958 retval = dquot_initialize(new.dir);
3959 if (retval)
3960 return retval;
bd42998a
MS		 3961
3962 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3963 &old.de, &old.inlined);
36de9286
TT		 3964 if (IS_ERR(old.bh))
3965 return PTR_ERR(old.bh);
bd42998a
MS		 3966 /*
3967 * Check for inode number is _not_ due to possible IO errors.
3968 * We might rmdir the source, keep it as pwd of some process
3969 * and merrily kill the link to whatever was created under the
3970 * same name. Goodbye sticky bit ;-<
3971 */
3972 retval = -ENOENT;
3973 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3974 goto end_rename;
3975
3976 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3977 &new.de, &new.inlined);
36de9286
TT		 3978 if (IS_ERR(new.bh)) {
3979 retval = PTR_ERR(new.bh);
a9cfcd63 3980 new.bh = NULL;
36de9286
TT		 3981 goto end_rename;
3982 }
bd42998a
MS		 3983
3984 /* RENAME_EXCHANGE case: old *and* new must both exist */
3985 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3986 goto end_rename;
3987
3988 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3989 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3990 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
7071b715
KK		 3991 if (IS_ERR(handle)) {
3992 retval = PTR_ERR(handle);
3993 handle = NULL;
3994 goto end_rename;
3995 }
bd42998a
MS		 3996
3997 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3998 ext4_handle_sync(handle);
3999
4000 if (S_ISDIR(old.inode->i_mode)) {
4001 old.is_dir = true;
4002 retval = ext4_rename_dir_prepare(handle, &old);
4003 if (retval)
4004 goto end_rename;
4005 }
4006 if (S_ISDIR(new.inode->i_mode)) {
4007 new.is_dir = true;
4008 retval = ext4_rename_dir_prepare(handle, &new);
4009 if (retval)
4010 goto end_rename;
4011 }
4012
4013 /*
4014 * Other than the special case of overwriting a directory, parents'
4015 * nlink only needs to be modified if this is a cross directory rename.
4016 */
4017 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4018 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4019 new.dir_nlink_delta = -old.dir_nlink_delta;
4020 retval = -EMLINK;
4021 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4022 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4023 goto end_rename;
4024 }
4025
4026 new_file_type = new.de->file_type;
4027 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4028 if (retval)
4029 goto end_rename;
4030
4031 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4032 if (retval)
4033 goto end_rename;
4034
4035 /*
4036 * Like most other Unix systems, set the ctime for inodes on a
4037 * rename.
4038 */
eeca7ea1
DD		 4039 ctime = current_time(old.inode);
4040 old.inode->i_ctime = ctime;
4041 new.inode->i_ctime = ctime;
4209ae12
HS		 4042 retval = ext4_mark_inode_dirty(handle, old.inode);
4043 if (unlikely(retval))
4044 goto end_rename;
4045 retval = ext4_mark_inode_dirty(handle, new.inode);
4046 if (unlikely(retval))
4047 goto end_rename;
aa75f4d3
HS		 4048 ext4_fc_mark_ineligible(new.inode->i_sb,
4049 EXT4_FC_REASON_CROSS_RENAME);
bd42998a
MS		 4050 if (old.dir_bh) {
4051 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4052 if (retval)
4053 goto end_rename;
4054 }
4055 if (new.dir_bh) {
4056 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4057 if (retval)
4058 goto end_rename;
4059 }
4060 ext4_update_dir_count(handle, &old);
4061 ext4_update_dir_count(handle, &new);
4062 retval = 0;
4063
4064end_rename:
4065 brelse(old.dir_bh);
4066 brelse(new.dir_bh);
4067 brelse(old.bh);
4068 brelse(new.bh);
4069 if (handle)
4070 ext4_journal_stop(handle);
4071 return retval;
4072}
4073
0a7c3937
MS		 4074static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4075 struct inode *new_dir, struct dentry *new_dentry,
4076 unsigned int flags)
4077{
07543d16
EB		 4078 int err;
4079
0db1ff22
TT		 4080 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4081 return -EIO;
4082
cd808dec 4083 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
0a7c3937
MS		 4084 return -EINVAL;
4085
07543d16
EB		 4086 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4087 flags);
4088 if (err)
4089 return err;
4090
bd42998a
MS		 4091 if (flags & RENAME_EXCHANGE) {
4092 return ext4_cross_rename(old_dir, old_dentry,
4093 new_dir, new_dentry);
4094 }
cd808dec
MS		 4095
4096 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
0a7c3937
MS		 4097}
4098
ac27a0ec
DK		 4099/*
4100 * directories can handle most operations...
4101 */
754661f1 4102const struct inode_operations ext4_dir_inode_operations = {
617ba13b
MC		 4103 .create = ext4_create,
4104 .lookup = ext4_lookup,
4105 .link = ext4_link,
4106 .unlink = ext4_unlink,
4107 .symlink = ext4_symlink,
4108 .mkdir = ext4_mkdir,
4109 .rmdir = ext4_rmdir,
4110 .mknod = ext4_mknod,
af51a2ac 4111 .tmpfile = ext4_tmpfile,
2773bf00 4112 .rename = ext4_rename2,
617ba13b 4113 .setattr = ext4_setattr,
99652ea5 4114 .getattr = ext4_getattr,
617ba13b 4115 .listxattr = ext4_listxattr,
4e34e719 4116 .get_acl = ext4_get_acl,
64e178a7 4117 .set_acl = ext4_set_acl,
abc8746e 4118 .fiemap = ext4_fiemap,
ac27a0ec
DK		 4119};
4120
754661f1 4121const struct inode_operations ext4_special_inode_operations = {
617ba13b 4122 .setattr = ext4_setattr,
99652ea5 4123 .getattr = ext4_getattr,
617ba13b 4124 .listxattr = ext4_listxattr,
4e34e719 4125 .get_acl = ext4_get_acl,
64e178a7 4126 .set_acl = ext4_set_acl,
ac27a0ec 4127};
Linux 6.x block layer and io_uring tree(s)