projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
exec: handle idmapped mounts
[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
ac27a0ec
DK		 185#ifdef DX_DEBUG
186#define dxtrace(command) command
187#else
188#define dxtrace(command)
189#endif
190
191struct fake_dirent
192{
193 __le32 inode;
194 __le16 rec_len;
195 u8 name_len;
196 u8 file_type;
197};
198
199struct dx_countlimit
200{
201 __le16 limit;
202 __le16 count;
203};
204
205struct dx_entry
206{
207 __le32 hash;
208 __le32 block;
209};
210
211/*
212 * dx_root_info is laid out so that if it should somehow get overlaid by a
213 * dirent the two low bits of the hash version will be zero. Therefore, the
214 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
215 */
216
217struct dx_root
218{
219 struct fake_dirent dot;
220 char dot_name[4];
221 struct fake_dirent dotdot;
222 char dotdot_name[4];
223 struct dx_root_info
224 {
225 __le32 reserved_zero;
226 u8 hash_version;
227 u8 info_length; /* 8 */
228 u8 indirect_levels;
229 u8 unused_flags;
230 }
231 info;
6cfb061f 232 struct dx_entry entries[];
ac27a0ec
DK		 233};
234
235struct dx_node
236{
237 struct fake_dirent fake;
6cfb061f 238 struct dx_entry entries[];
ac27a0ec
DK		 239};
240
241
242struct dx_frame
243{
244 struct buffer_head *bh;
245 struct dx_entry *entries;
246 struct dx_entry *at;
247};
248
249struct dx_map_entry
250{
251 u32 hash;
ef2b02d3
ES		 252 u16 offs;
253 u16 size;
ac27a0ec
DK		 254};
255
e6153918
DW		 256/*
257 * This goes at the end of each htree block.
258 */
259struct dx_tail {
260 u32 dt_reserved;
261 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
262};
263
725d26d3
AK		 264static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
265static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
af5bc92d
TT		 266static inline unsigned dx_get_hash(struct dx_entry *entry);
267static void dx_set_hash(struct dx_entry *entry, unsigned value);
268static unsigned dx_get_count(struct dx_entry *entries);
269static unsigned dx_get_limit(struct dx_entry *entries);
270static void dx_set_count(struct dx_entry *entries, unsigned value);
271static void dx_set_limit(struct dx_entry *entries, unsigned value);
272static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
273static unsigned dx_node_limit(struct inode *dir);
5b643f9c 274static struct dx_frame *dx_probe(struct ext4_filename *fname,
ac27a0ec
DK		 275 struct inode *dir,
276 struct dx_hash_info *hinfo,
dd73b5d5 277 struct dx_frame *frame);
af5bc92d 278static void dx_release(struct dx_frame *frames);
1f3862b5
MH		 279static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
280 unsigned blocksize, struct dx_hash_info *hinfo,
281 struct dx_map_entry map[]);
ac27a0ec 282static void dx_sort_map(struct dx_map_entry *map, unsigned count);
af5bc92d 283static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
3d0518f4 284 struct dx_map_entry *offsets, int count, unsigned blocksize);
8bad4597 285static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
725d26d3
AK		 286static void dx_insert_block(struct dx_frame *frame,
287 u32 hash, ext4_lblk_t block);
617ba13b 288static int ext4_htree_next_block(struct inode *dir, __u32 hash,
ac27a0ec
DK		 289 struct dx_frame *frame,
290 struct dx_frame *frames,
291 __u32 *start_hash);
f702ba0f 292static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
5b643f9c 293 struct ext4_filename *fname,
537d8f93 294 struct ext4_dir_entry_2 **res_dir);
5b643f9c 295static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
56a04915 296 struct inode *dir, struct inode *inode);
ac27a0ec 297
dbe89444 298/* checksumming functions */
ddce3b94
TT		 299void ext4_initialize_dirent_tail(struct buffer_head *bh,
300 unsigned int blocksize)
b0336e8d 301{
ddce3b94
TT		 302 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
303
b0336e8d
DW		 304 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
305 t->det_rec_len = ext4_rec_len_to_disk(
306 sizeof(struct ext4_dir_entry_tail), blocksize);
307 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
308}
309
310/* Walk through a dirent block to find a checksum "dirent" at the tail */
311static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
f036adb3 312 struct buffer_head *bh)
b0336e8d
DW		 313{
314 struct ext4_dir_entry_tail *t;
315
316#ifdef PARANOID
317 struct ext4_dir_entry *d, *top;
318
f036adb3
TT		 319 d = (struct ext4_dir_entry *)bh->b_data;
320 top = (struct ext4_dir_entry *)(bh->b_data +
b0336e8d 321 (EXT4_BLOCK_SIZE(inode->i_sb) -
f036adb3 322 sizeof(struct ext4_dir_entry_tail)));
b0336e8d
DW		 323 while (d < top && d->rec_len)
324 d = (struct ext4_dir_entry *)(((void *)d) +
325 le16_to_cpu(d->rec_len));
326
327 if (d != top)
328 return NULL;
329
330 t = (struct ext4_dir_entry_tail *)d;
331#else
f036adb3 332 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
b0336e8d
DW		 333#endif
334
335 if (t->det_reserved_zero1 ||
336 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
337 t->det_reserved_zero2 ||
338 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
339 return NULL;
340
341 return t;
342}
343
f036adb3 344static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
b0336e8d
DW		 345{
346 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
347 struct ext4_inode_info *ei = EXT4_I(inode);
348 __u32 csum;
349
350 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
351 return cpu_to_le32(csum);
352}
353
b03a2f7e
AD		 354#define warn_no_space_for_csum(inode) \
355 __warn_no_space_for_csum((inode), __func__, __LINE__)
356
357static void __warn_no_space_for_csum(struct inode *inode, const char *func,
358 unsigned int line)
dffe9d8d 359{
b03a2f7e
AD		 360 __ext4_warning_inode(inode, func, line,
361 "No space for directory leaf checksum. Please run e2fsck -D.");
dffe9d8d
TT		 362}
363
f036adb3 364int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
b0336e8d
DW		 365{
366 struct ext4_dir_entry_tail *t;
367
9aa5d32b 368 if (!ext4_has_metadata_csum(inode->i_sb))
b0336e8d
DW		 369 return 1;
370
f036adb3 371 t = get_dirent_tail(inode, bh);
b0336e8d 372 if (!t) {
dffe9d8d 373 warn_no_space_for_csum(inode);
b0336e8d
DW		 374 return 0;
375 }
376
f036adb3 377 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
ddce3b94 378 (char *)t - bh->b_data))
b0336e8d
DW		 379 return 0;
380
381 return 1;
382}
383
f036adb3
TT		 384static void ext4_dirblock_csum_set(struct inode *inode,
385 struct buffer_head *bh)
b0336e8d
DW		 386{
387 struct ext4_dir_entry_tail *t;
388
9aa5d32b 389 if (!ext4_has_metadata_csum(inode->i_sb))
b0336e8d
DW		 390 return;
391
f036adb3 392 t = get_dirent_tail(inode, bh);
b0336e8d 393 if (!t) {
dffe9d8d 394 warn_no_space_for_csum(inode);
b0336e8d
DW		 395 return;
396 }
397
f036adb3 398 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
ddce3b94 399 (char *)t - bh->b_data);
b0336e8d
DW		 400}
401
f036adb3
TT		 402int ext4_handle_dirty_dirblock(handle_t *handle,
403 struct inode *inode,
404 struct buffer_head *bh)
b0336e8d 405{
f036adb3 406 ext4_dirblock_csum_set(inode, bh);
b0336e8d
DW		 407 return ext4_handle_dirty_metadata(handle, inode, bh);
408}
409
dbe89444
DW		 410static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
411 struct ext4_dir_entry *dirent,
412 int *offset)
413{
414 struct ext4_dir_entry *dp;
415 struct dx_root_info *root;
416 int count_offset;
417
418 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
419 count_offset = 8;
420 else if (le16_to_cpu(dirent->rec_len) == 12) {
421 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
422 if (le16_to_cpu(dp->rec_len) !=
423 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
424 return NULL;
425 root = (struct dx_root_info *)(((void *)dp + 12));
426 if (root->reserved_zero ||
427 root->info_length != sizeof(struct dx_root_info))
428 return NULL;
429 count_offset = 32;
430 } else
431 return NULL;
432
433 if (offset)
434 *offset = count_offset;
435 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
436}
437
438static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
439 int count_offset, int count, struct dx_tail *t)
440{
441 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
442 struct ext4_inode_info *ei = EXT4_I(inode);
d6a77105 443 __u32 csum;
dbe89444 444 int size;
b47820ed
DJ		 445 __u32 dummy_csum = 0;
446 int offset = offsetof(struct dx_tail, dt_checksum);
dbe89444
DW		 447
448 size = count_offset + (count * sizeof(struct dx_entry));
dbe89444 449 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
b47820ed
DJ		 450 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
451 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
dbe89444
DW		 452
453 return cpu_to_le32(csum);
454}
455
456static int ext4_dx_csum_verify(struct inode *inode,
457 struct ext4_dir_entry *dirent)
458{
459 struct dx_countlimit *c;
460 struct dx_tail *t;
461 int count_offset, limit, count;
462
9aa5d32b 463 if (!ext4_has_metadata_csum(inode->i_sb))
dbe89444
DW		 464 return 1;
465
466 c = get_dx_countlimit(inode, dirent, &count_offset);
467 if (!c) {
468 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
fa964540 469 return 0;
dbe89444
DW		 470 }
471 limit = le16_to_cpu(c->limit);
472 count = le16_to_cpu(c->count);
473 if (count_offset + (limit * sizeof(struct dx_entry)) >
474 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
dffe9d8d 475 warn_no_space_for_csum(inode);
fa964540 476 return 0;
dbe89444
DW		 477 }
478 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
479
480 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
481 count, t))
482 return 0;
483 return 1;
484}
485
486static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
487{
488 struct dx_countlimit *c;
489 struct dx_tail *t;
490 int count_offset, limit, count;
491
9aa5d32b 492 if (!ext4_has_metadata_csum(inode->i_sb))
dbe89444
DW		 493 return;
494
495 c = get_dx_countlimit(inode, dirent, &count_offset);
496 if (!c) {
497 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
498 return;
499 }
500 limit = le16_to_cpu(c->limit);
501 count = le16_to_cpu(c->count);
502 if (count_offset + (limit * sizeof(struct dx_entry)) >
503 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
dffe9d8d 504 warn_no_space_for_csum(inode);
dbe89444
DW		 505 return;
506 }
507 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
508
509 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
510}
511
512static inline int ext4_handle_dirty_dx_node(handle_t *handle,
513 struct inode *inode,
514 struct buffer_head *bh)
515{
516 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
517 return ext4_handle_dirty_metadata(handle, inode, bh);
518}
519
f795e140
LZ		 520/*
521 * p is at least 6 bytes before the end of page
522 */
523static inline struct ext4_dir_entry_2 *
3d0518f4 524ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
f795e140
LZ		 525{
526 return (struct ext4_dir_entry_2 *)((char *)p +
3d0518f4 527 ext4_rec_len_from_disk(p->rec_len, blocksize));
f795e140
LZ		 528}
529
ac27a0ec
DK		 530/*
531 * Future: use high four bits of block for coalesce-on-delete flags
532 * Mask them off for now.
533 */
534
725d26d3 535static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
ac27a0ec 536{
e08ac99f 537 return le32_to_cpu(entry->block) & 0x0fffffff;
ac27a0ec
DK		 538}
539
725d26d3 540static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
ac27a0ec
DK		 541{
542 entry->block = cpu_to_le32(value);
543}
544
af5bc92d 545static inline unsigned dx_get_hash(struct dx_entry *entry)
ac27a0ec
DK		 546{
547 return le32_to_cpu(entry->hash);
548}
549
af5bc92d 550static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
ac27a0ec
DK		 551{
552 entry->hash = cpu_to_le32(value);
553}
554
af5bc92d 555static inline unsigned dx_get_count(struct dx_entry *entries)
ac27a0ec
DK		 556{
557 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
558}
559
af5bc92d 560static inline unsigned dx_get_limit(struct dx_entry *entries)
ac27a0ec
DK		 561{
562 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
563}
564
af5bc92d 565static inline void dx_set_count(struct dx_entry *entries, unsigned value)
ac27a0ec
DK		 566{
567 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
568}
569
af5bc92d 570static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
ac27a0ec
DK		 571{
572 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
573}
574
af5bc92d 575static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
ac27a0ec 576{
617ba13b
MC		 577 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
578 EXT4_DIR_REC_LEN(2) - infosize;
dbe89444 579
9aa5d32b 580 if (ext4_has_metadata_csum(dir->i_sb))
dbe89444 581 entry_space -= sizeof(struct dx_tail);
d9c769b7 582 return entry_space / sizeof(struct dx_entry);
ac27a0ec
DK		 583}
584
af5bc92d 585static inline unsigned dx_node_limit(struct inode *dir)
ac27a0ec 586{
617ba13b 587 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
dbe89444 588
9aa5d32b 589 if (ext4_has_metadata_csum(dir->i_sb))
dbe89444 590 entry_space -= sizeof(struct dx_tail);
d9c769b7 591 return entry_space / sizeof(struct dx_entry);
ac27a0ec
DK		 592}
593
594/*
595 * Debug
596 */
597#ifdef DX_DEBUG
4776004f 598static void dx_show_index(char * label, struct dx_entry *entries)
ac27a0ec 599{
63f57933 600 int i, n = dx_get_count (entries);
d74f3d25 601 printk(KERN_DEBUG "%s index", label);
63f57933 602 for (i = 0; i < n; i++) {
d74f3d25
JP		 603 printk(KERN_CONT " %x->%lu",
604 i ? dx_get_hash(entries + i) : 0,
605 (unsigned long)dx_get_block(entries + i));
63f57933 606 }
d74f3d25 607 printk(KERN_CONT "\n");
ac27a0ec
DK		 608}
609
610struct stats
611{
612 unsigned names;
613 unsigned space;
614 unsigned bcount;
615};
616
b3098486
MH		 617static struct stats dx_show_leaf(struct inode *dir,
618 struct dx_hash_info *hinfo,
619 struct ext4_dir_entry_2 *de,
620 int size, int show_names)
ac27a0ec
DK		 621{
622 unsigned names = 0, space = 0;
623 char *base = (char *) de;
624 struct dx_hash_info h = *hinfo;
625
626 printk("names: ");
627 while ((char *) de < base + size)
628 {
629 if (de->inode)
630 {
631 if (show_names)
632 {
643fa961 633#ifdef CONFIG_FS_ENCRYPTION
b3098486
MH		 634 int len;
635 char *name;
a7550b30
JK		 636 struct fscrypt_str fname_crypto_str =
637 FSTR_INIT(NULL, 0);
c936e1ec 638 int res = 0;
b3098486
MH		 639
640 name = de->name;
641 len = de->name_len;
91d0d892 642 if (!IS_ENCRYPTED(dir)) {
b3098486 643 /* Directory is not encrypted */
b886ee3e 644 ext4fs_dirhash(dir, de->name,
b3098486
MH		 645 de->name_len, &h);
646 printk("%*.s:(U)%x.%u ", len,
647 name, h.hash,
648 (unsigned) ((char *) de
649 - base));
650 } else {
a7550b30
JK		 651 struct fscrypt_str de_name =
652 FSTR_INIT(name, len);
653
b3098486 654 /* Directory is encrypted */
a7550b30 655 res = fscrypt_fname_alloc_buffer(
8b10fe68 656 len, &fname_crypto_str);
ef1eb3aa 657 if (res)
b3098486
MH		 658 printk(KERN_WARNING "Error "
659 "allocating crypto "
660 "buffer--skipping "
661 "crypto\n");
a7550b30
JK		 662 res = fscrypt_fname_disk_to_usr(dir,
663 0, 0, &de_name,
664 &fname_crypto_str);
ef1eb3aa 665 if (res) {
b3098486
MH		 666 printk(KERN_WARNING "Error "
667 "converting filename "
668 "from disk to usr"
669 "\n");
670 name = "??";
671 len = 2;
672 } else {
673 name = fname_crypto_str.name;
674 len = fname_crypto_str.len;
675 }
b886ee3e
GKB		 676 ext4fs_dirhash(dir, de->name,
677 de->name_len, &h);
b3098486
MH		 678 printk("%*.s:(E)%x.%u ", len, name,
679 h.hash, (unsigned) ((char *) de
680 - base));
a7550b30
JK		 681 fscrypt_fname_free_buffer(
682 &fname_crypto_str);
b3098486
MH		 683 }
684#else
ac27a0ec
DK		 685 int len = de->name_len;
686 char *name = de->name;
b886ee3e 687 ext4fs_dirhash(dir, de->name, de->name_len, &h);
b3098486 688 printk("%*.s:%x.%u ", len, name, h.hash,
265c6a0f 689 (unsigned) ((char *) de - base));
b3098486 690#endif
ac27a0ec 691 }
617ba13b 692 space += EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec
DK		 693 names++;
694 }
3d0518f4 695 de = ext4_next_entry(de, size);
ac27a0ec 696 }
d74f3d25 697 printk(KERN_CONT "(%i)\n", names);
ac27a0ec
DK		 698 return (struct stats) { names, space, 1 };
699}
700
701struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
702 struct dx_entry *entries, int levels)
703{
704 unsigned blocksize = dir->i_sb->s_blocksize;
af5bc92d 705 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
ac27a0ec
DK		 706 unsigned bcount = 0;
707 struct buffer_head *bh;
ac27a0ec
DK		 708 printk("%i indexed blocks...\n", count);
709 for (i = 0; i < count; i++, entries++)
710 {
725d26d3
AK		 711 ext4_lblk_t block = dx_get_block(entries);
712 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
ac27a0ec
DK		 713 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
714 struct stats stats;
715 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
1c215028
TT		 716 bh = ext4_bread(NULL,dir, block, 0);
717 if (!bh || IS_ERR(bh))
718 continue;
ac27a0ec
DK		 719 stats = levels?
720 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
b3098486
MH		 721 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
722 bh->b_data, blocksize, 0);
ac27a0ec
DK		 723 names += stats.names;
724 space += stats.space;
725 bcount += stats.bcount;
af5bc92d 726 brelse(bh);
ac27a0ec
DK		 727 }
728 if (bcount)
60e6679e 729 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
4776004f
TT		 730 levels ? "" : " ", names, space/bcount,
731 (space/bcount)*100/blocksize);
ac27a0ec
DK		 732 return (struct stats) { names, space, bcount};
733}
734#endif /* DX_DEBUG */
735
736/*
737 * Probe for a directory leaf block to search.
738 *
739 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
740 * error in the directory index, and the caller should fall back to
741 * searching the directory normally. The callers of dx_probe **MUST**
742 * check for this error code, and make sure it never gets reflected
743 * back to userspace.
744 */
745static struct dx_frame *
5b643f9c 746dx_probe(struct ext4_filename *fname, struct inode *dir,
dd73b5d5 747 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
ac27a0ec
DK		 748{
749 unsigned count, indirect;
750 struct dx_entry *at, *entries, *p, *q, *m;
751 struct dx_root *root;
ac27a0ec 752 struct dx_frame *frame = frame_in;
dd73b5d5 753 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
ac27a0ec
DK		 754 u32 hash;
755
e08ac99f 756 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
dd73b5d5
TT		 757 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
758 if (IS_ERR(frame->bh))
759 return (struct dx_frame *) frame->bh;
760
761 root = (struct dx_root *) frame->bh->b_data;
ac27a0ec
DK		 762 if (root->info.hash_version != DX_HASH_TEA &&
763 root->info.hash_version != DX_HASH_HALF_MD4 &&
764 root->info.hash_version != DX_HASH_LEGACY) {
b03a2f7e
AD		 765 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
766 root->info.hash_version);
ac27a0ec
DK		 767 goto fail;
768 }
5b643f9c
TT		 769 if (fname)
770 hinfo = &fname->hinfo;
ac27a0ec 771 hinfo->hash_version = root->info.hash_version;
f99b2589
TT		 772 if (hinfo->hash_version <= DX_HASH_TEA)
773 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
617ba13b 774 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
5b643f9c 775 if (fname && fname_name(fname))
b886ee3e 776 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
ac27a0ec
DK		 777 hash = hinfo->hash;
778
779 if (root->info.unused_flags & 1) {
b03a2f7e
AD		 780 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
781 root->info.unused_flags);
ac27a0ec
DK		 782 goto fail;
783 }
784
b03a2f7e 785 indirect = root->info.indirect_levels;
e08ac99f
AB		 786 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
787 ext4_warning(dir->i_sb,
788 "Directory (ino: %lu) htree depth %#06x exceed"
789 "supported value", dir->i_ino,
790 ext4_dir_htree_level(dir->i_sb));
791 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
792 ext4_warning(dir->i_sb, "Enable large directory "
793 "feature to access it");
794 }
ac27a0ec
DK		 795 goto fail;
796 }
797
b03a2f7e
AD		 798 entries = (struct dx_entry *)(((char *)&root->info) +
799 root->info.info_length);
3d82abae
ES		 800
801 if (dx_get_limit(entries) != dx_root_limit(dir,
802 root->info.info_length)) {
b03a2f7e
AD		 803 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
804 dx_get_limit(entries),
805 dx_root_limit(dir, root->info.info_length));
3d82abae
ES		 806 goto fail;
807 }
808
af5bc92d 809 dxtrace(printk("Look up %x", hash));
dd73b5d5 810 while (1) {
ac27a0ec 811 count = dx_get_count(entries);
3d82abae 812 if (!count || count > dx_get_limit(entries)) {
b03a2f7e
AD		 813 ext4_warning_inode(dir,
814 "dx entry: count %u beyond limit %u",
815 count, dx_get_limit(entries));
dd73b5d5 816 goto fail;
3d82abae
ES		 817 }
818
ac27a0ec
DK		 819 p = entries + 1;
820 q = entries + count - 1;
dd73b5d5 821 while (p <= q) {
b03a2f7e 822 m = p + (q - p) / 2;
d74f3d25 823 dxtrace(printk(KERN_CONT "."));
ac27a0ec
DK		 824 if (dx_get_hash(m) > hash)
825 q = m - 1;
826 else
827 p = m + 1;
828 }
829
dd73b5d5 830 if (0) { // linear search cross check
ac27a0ec
DK		 831 unsigned n = count - 1;
832 at = entries;
833 while (n--)
834 {
d74f3d25 835 dxtrace(printk(KERN_CONT ","));
ac27a0ec
DK		 836 if (dx_get_hash(++at) > hash)
837 {
838 at--;
839 break;
840 }
841 }
837c23fb 842 ASSERT(at == p - 1);
ac27a0ec
DK		 843 }
844
845 at = p - 1;
d74f3d25
JP		 846 dxtrace(printk(KERN_CONT " %x->%u\n",
847 at == entries ? 0 : dx_get_hash(at),
b03a2f7e 848 dx_get_block(at)));
ac27a0ec
DK		 849 frame->entries = entries;
850 frame->at = at;
dd73b5d5
TT		 851 if (!indirect--)
852 return frame;
853 frame++;
854 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
855 if (IS_ERR(frame->bh)) {
856 ret_err = (struct dx_frame *) frame->bh;
857 frame->bh = NULL;
858 goto fail;
dbe89444 859 }
dd73b5d5 860 entries = ((struct dx_node *) frame->bh->b_data)->entries;
dbe89444 861
b03a2f7e
AD		 862 if (dx_get_limit(entries) != dx_node_limit(dir)) {
863 ext4_warning_inode(dir,
864 "dx entry: limit %u != node limit %u",
865 dx_get_limit(entries), dx_node_limit(dir));
dd73b5d5 866 goto fail;
3d82abae 867 }
ac27a0ec 868 }
dd73b5d5 869fail:
ac27a0ec
DK		 870 while (frame >= frame_in) {
871 brelse(frame->bh);
872 frame--;
873 }
b3098486 874
dd73b5d5 875 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
b03a2f7e
AD		 876 ext4_warning_inode(dir,
877 "Corrupt directory, running e2fsck is recommended");
dd73b5d5 878 return ret_err;
ac27a0ec
DK		 879}
880
b03a2f7e 881static void dx_release(struct dx_frame *frames)
ac27a0ec 882{
e08ac99f
AB		 883 struct dx_root_info *info;
884 int i;
08fc98a4 885 unsigned int indirect_levels;
e08ac99f 886
ac27a0ec
DK		 887 if (frames[0].bh == NULL)
888 return;
889
e08ac99f 890 info = &((struct dx_root *)frames[0].bh->b_data)->info;
08fc98a4
ST		 891 /* save local copy, "info" may be freed after brelse() */
892 indirect_levels = info->indirect_levels;
893 for (i = 0; i <= indirect_levels; i++) {
e08ac99f
AB		 894 if (frames[i].bh == NULL)
895 break;
896 brelse(frames[i].bh);
897 frames[i].bh = NULL;
898 }
ac27a0ec
DK		 899}
900
901/*
902 * This function increments the frame pointer to search the next leaf
903 * block, and reads in the necessary intervening nodes if the search
904 * should be necessary. Whether or not the search is necessary is
905 * controlled by the hash parameter. If the hash value is even, then
906 * the search is only continued if the next block starts with that
907 * hash value. This is used if we are searching for a specific file.
908 *
909 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
910 *
911 * This function returns 1 if the caller should continue to search,
912 * or 0 if it should not. If there is an error reading one of the
913 * index blocks, it will a negative error code.
914 *
915 * If start_hash is non-null, it will be filled in with the starting
916 * hash of the next page.
917 */
617ba13b 918static int ext4_htree_next_block(struct inode *dir, __u32 hash,
ac27a0ec
DK		 919 struct dx_frame *frame,
920 struct dx_frame *frames,
921 __u32 *start_hash)
922{
923 struct dx_frame *p;
924 struct buffer_head *bh;
dc6982ff 925 int num_frames = 0;
ac27a0ec
DK		 926 __u32 bhash;
927
928 p = frame;
929 /*
930 * Find the next leaf page by incrementing the frame pointer.
931 * If we run out of entries in the interior node, loop around and
932 * increment pointer in the parent node. When we break out of
933 * this loop, num_frames indicates the number of interior
934 * nodes need to be read.
935 */
936 while (1) {
937 if (++(p->at) < p->entries + dx_get_count(p->entries))
938 break;
939 if (p == frames)
940 return 0;
941 num_frames++;
942 p--;
943 }
944
945 /*
946 * If the hash is 1, then continue only if the next page has a
947 * continuation hash of any value. This is used for readdir
948 * handling. Otherwise, check to see if the hash matches the
949 * desired contiuation hash. If it doesn't, return since
950 * there's no point to read in the successive index pages.
951 */
952 bhash = dx_get_hash(p->at);
953 if (start_hash)
954 *start_hash = bhash;
955 if ((hash & 1) == 0) {
956 if ((bhash & ~1) != hash)
957 return 0;
958 }
959 /*
960 * If the hash is HASH_NB_ALWAYS, we always go to the next
961 * block so no check is necessary
962 */
963 while (num_frames--) {
dc6982ff
TT		 964 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
965 if (IS_ERR(bh))
966 return PTR_ERR(bh);
ac27a0ec 967 p++;
af5bc92d 968 brelse(p->bh);
ac27a0ec
DK		 969 p->bh = bh;
970 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
971 }
972 return 1;
973}
974
975
ac27a0ec
DK		 976/*
977 * This function fills a red-black tree with information from a
978 * directory block. It returns the number directory entries loaded
979 * into the tree. If there is an error it is returned in err.
980 */
981static int htree_dirblock_to_tree(struct file *dir_file,
725d26d3 982 struct inode *dir, ext4_lblk_t block,
ac27a0ec
DK		 983 struct dx_hash_info *hinfo,
984 __u32 start_hash, __u32 start_minor_hash)
985{
986 struct buffer_head *bh;
617ba13b 987 struct ext4_dir_entry_2 *de, *top;
90b0a973 988 int err = 0, count = 0;
a7550b30 989 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
ac27a0ec 990
725d26d3
AK		 991 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
992 (unsigned long)block));
4e19d6b6 993 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
dc6982ff
TT		 994 if (IS_ERR(bh))
995 return PTR_ERR(bh);
b0336e8d 996
617ba13b
MC		 997 de = (struct ext4_dir_entry_2 *) bh->b_data;
998 top = (struct ext4_dir_entry_2 *) ((char *) de +
ac27a0ec 999 dir->i_sb->s_blocksize -
617ba13b 1000 EXT4_DIR_REC_LEN(0));
1f3862b5 1001 /* Check if the directory is encrypted */
592ddec7 1002 if (IS_ENCRYPTED(dir)) {
ec0caa97 1003 err = fscrypt_prepare_readdir(dir);
c936e1ec
TT		 1004 if (err < 0) {
1005 brelse(bh);
1006 return err;
1007 }
8b10fe68
JL		 1008 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1009 &fname_crypto_str);
1f3862b5 1010 if (err < 0) {
1f3862b5
MH		 1011 brelse(bh);
1012 return err;
1013 }
1014 }
64c314ff 1015
3d0518f4 1016 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
f7c21177 1017 if (ext4_check_dir_entry(dir, NULL, de, bh,
226ba972 1018 bh->b_data, bh->b_size,
cad3f007
TT		 1019 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1020 + ((char *)de - bh->b_data))) {
64cb9273
AV		 1021 /* silently ignore the rest of the block */
1022 break;
e6c40211 1023 }
b886ee3e 1024 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
ac27a0ec
DK		 1025 if ((hinfo->hash < start_hash) ||
1026 ((hinfo->hash == start_hash) &&
1027 (hinfo->minor_hash < start_minor_hash)))
1028 continue;
1029 if (de->inode == 0)
1030 continue;
592ddec7 1031 if (!IS_ENCRYPTED(dir)) {
1f3862b5
MH		 1032 tmp_str.name = de->name;
1033 tmp_str.len = de->name_len;
1034 err = ext4_htree_store_dirent(dir_file,
1035 hinfo->hash, hinfo->minor_hash, de,
1036 &tmp_str);
1037 } else {
d2299590 1038 int save_len = fname_crypto_str.len;
a7550b30
JK		 1039 struct fscrypt_str de_name = FSTR_INIT(de->name,
1040 de->name_len);
d2299590 1041
1f3862b5 1042 /* Directory is encrypted */
a7550b30
JK		 1043 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1044 hinfo->minor_hash, &de_name,
1045 &fname_crypto_str);
ef1eb3aa 1046 if (err) {
1f3862b5
MH		 1047 count = err;
1048 goto errout;
1049 }
1050 err = ext4_htree_store_dirent(dir_file,
1051 hinfo->hash, hinfo->minor_hash, de,
1052 &fname_crypto_str);
d2299590 1053 fname_crypto_str.len = save_len;
1f3862b5 1054 }
2f61830a 1055 if (err != 0) {
1f3862b5
MH		 1056 count = err;
1057 goto errout;
ac27a0ec
DK		 1058 }
1059 count++;
1060 }
1f3862b5 1061errout:
ac27a0ec 1062 brelse(bh);
a7550b30 1063 fscrypt_fname_free_buffer(&fname_crypto_str);
ac27a0ec
DK		 1064 return count;
1065}
1066
1067
1068/*
1069 * This function fills a red-black tree with information from a
1070 * directory. We start scanning the directory in hash order, starting
1071 * at start_hash and start_minor_hash.
1072 *
1073 * This function returns the number of entries inserted into the tree,
1074 * or a negative error code.
1075 */
617ba13b 1076int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
ac27a0ec
DK		 1077 __u32 start_minor_hash, __u32 *next_hash)
1078{
1079 struct dx_hash_info hinfo;
617ba13b 1080 struct ext4_dir_entry_2 *de;
e08ac99f 1081 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 1082 struct inode *dir;
725d26d3 1083 ext4_lblk_t block;
ac27a0ec 1084 int count = 0;
725d26d3 1085 int ret, err;
ac27a0ec 1086 __u32 hashval;
a7550b30 1087 struct fscrypt_str tmp_str;
ac27a0ec 1088
60e6679e 1089 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
4776004f 1090 start_hash, start_minor_hash));
496ad9aa 1091 dir = file_inode(dir_file);
12e9b892 1092 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
617ba13b 1093 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
f99b2589
TT		 1094 if (hinfo.hash_version <= DX_HASH_TEA)
1095 hinfo.hash_version +=
1096 EXT4_SB(dir->i_sb)->s_hash_unsigned;
617ba13b 1097 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
8af0f082
TM		 1098 if (ext4_has_inline_data(dir)) {
1099 int has_inline_data = 1;
7633b08b
TT		 1100 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1101 &hinfo, start_hash,
1102 start_minor_hash,
1103 &has_inline_data);
8af0f082
TM		 1104 if (has_inline_data) {
1105 *next_hash = ~0;
1106 return count;
1107 }
1108 }
ac27a0ec
DK		 1109 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1110 start_hash, start_minor_hash);
1111 *next_hash = ~0;
1112 return count;
1113 }
1114 hinfo.hash = start_hash;
1115 hinfo.minor_hash = 0;
dd73b5d5
TT		 1116 frame = dx_probe(NULL, dir, &hinfo, frames);
1117 if (IS_ERR(frame))
1118 return PTR_ERR(frame);
ac27a0ec
DK		 1119
1120 /* Add '.' and '..' from the htree header */
1121 if (!start_hash && !start_minor_hash) {
617ba13b 1122 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
2f61830a
TT		 1123 tmp_str.name = de->name;
1124 tmp_str.len = de->name_len;
1125 err = ext4_htree_store_dirent(dir_file, 0, 0,
1126 de, &tmp_str);
1127 if (err != 0)
ac27a0ec
DK		 1128 goto errout;
1129 count++;
1130 }
1131 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
617ba13b 1132 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
3d0518f4 1133 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
2f61830a
TT		 1134 tmp_str.name = de->name;
1135 tmp_str.len = de->name_len;
1136 err = ext4_htree_store_dirent(dir_file, 2, 0,
1137 de, &tmp_str);
1138 if (err != 0)
ac27a0ec
DK		 1139 goto errout;
1140 count++;
1141 }
1142
1143 while (1) {
1f60fbe7
TT		 1144 if (fatal_signal_pending(current)) {
1145 err = -ERESTARTSYS;
1146 goto errout;
1147 }
1148 cond_resched();
ac27a0ec
DK		 1149 block = dx_get_block(frame->at);
1150 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1151 start_hash, start_minor_hash);
1152 if (ret < 0) {
1153 err = ret;
1154 goto errout;
1155 }
1156 count += ret;
1157 hashval = ~0;
617ba13b 1158 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
ac27a0ec
DK		 1159 frame, frames, &hashval);
1160 *next_hash = hashval;
1161 if (ret < 0) {
1162 err = ret;
1163 goto errout;
1164 }
1165 /*
1166 * Stop if: (a) there are no more entries, or
1167 * (b) we have inserted at least one entry and the
1168 * next hash value is not a continuation
1169 */
1170 if ((ret == 0) ||
1171 (count && ((hashval & 1) == 0)))
1172 break;
1173 }
1174 dx_release(frames);
4776004f
TT		 1175 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1176 "next hash: %x\n", count, *next_hash));
ac27a0ec
DK		 1177 return count;
1178errout:
1179 dx_release(frames);
1180 return (err);
1181}
1182
7335cd3b
TM		 1183static inline int search_dirblock(struct buffer_head *bh,
1184 struct inode *dir,
5b643f9c 1185 struct ext4_filename *fname,
7335cd3b
TM		 1186 unsigned int offset,
1187 struct ext4_dir_entry_2 **res_dir)
1188{
5b643f9c 1189 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
d6b97550 1190 fname, offset, res_dir);
7335cd3b
TM		 1191}
1192
ac27a0ec
DK		 1193/*
1194 * Directory block splitting, compacting
1195 */
1196
ef2b02d3
ES		 1197/*
1198 * Create map of hash values, offsets, and sizes, stored at end of block.
1199 * Returns number of entries mapped.
1200 */
1f3862b5
MH		 1201static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1202 unsigned blocksize, struct dx_hash_info *hinfo,
8bad4597 1203 struct dx_map_entry *map_tail)
ac27a0ec
DK		 1204{
1205 int count = 0;
1206 char *base = (char *) de;
1207 struct dx_hash_info h = *hinfo;
1208
8bad4597 1209 while ((char *) de < base + blocksize) {
ac27a0ec 1210 if (de->name_len && de->inode) {
b886ee3e 1211 ext4fs_dirhash(dir, de->name, de->name_len, &h);
ac27a0ec
DK		 1212 map_tail--;
1213 map_tail->hash = h.hash;
9aee2286 1214 map_tail->offs = ((char *) de - base)>>2;
ef2b02d3 1215 map_tail->size = le16_to_cpu(de->rec_len);
ac27a0ec
DK		 1216 count++;
1217 cond_resched();
1218 }
1219 /* XXX: do we need to check rec_len == 0 case? -Chris */
3d0518f4 1220 de = ext4_next_entry(de, blocksize);
ac27a0ec
DK		 1221 }
1222 return count;
1223}
1224
ef2b02d3 1225/* Sort map by hash value */
ac27a0ec
DK		 1226static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1227{
63f57933
AM		 1228 struct dx_map_entry *p, *q, *top = map + count - 1;
1229 int more;
1230 /* Combsort until bubble sort doesn't suck */
1231 while (count > 2) {
1232 count = count*10/13;
1233 if (count - 9 < 2) /* 9, 10 -> 11 */
1234 count = 11;
1235 for (p = top, q = p - count; q >= map; p--, q--)
1236 if (p->hash < q->hash)
1237 swap(*p, *q);
1238 }
1239 /* Garden variety bubble sort */
1240 do {
1241 more = 0;
1242 q = top;
1243 while (q-- > map) {
1244 if (q[1].hash >= q[0].hash)
ac27a0ec 1245 continue;
63f57933
AM		 1246 swap(*(q+1), *q);
1247 more = 1;
ac27a0ec
DK		 1248 }
1249 } while(more);
1250}
1251
725d26d3 1252static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
ac27a0ec
DK		 1253{
1254 struct dx_entry *entries = frame->entries;
1255 struct dx_entry *old = frame->at, *new = old + 1;
1256 int count = dx_get_count(entries);
1257
837c23fb
CX		 1258 ASSERT(count < dx_get_limit(entries));
1259 ASSERT(old < entries + count);
ac27a0ec
DK		 1260 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1261 dx_set_hash(new, hash);
1262 dx_set_block(new, block);
1263 dx_set_count(entries, count + 1);
1264}
ac27a0ec 1265
b886ee3e
GKB		 1266#ifdef CONFIG_UNICODE
1267/*
1268 * Test whether a case-insensitive directory entry matches the filename
3ae72562
GKB		 1269 * being searched for. If quick is set, assume the name being looked up
1270 * is already in the casefolded form.
b886ee3e
GKB		 1271 *
1272 * Returns: 0 if the directory entry matches, more than 0 if it
1273 * doesn't match or less than zero on error.
1274 */
1275int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
3ae72562 1276 const struct qstr *entry, bool quick)
b886ee3e 1277{
f8f4acb6
DR		 1278 const struct super_block *sb = parent->i_sb;
1279 const struct unicode_map *um = sb->s_encoding;
b886ee3e
GKB		 1280 int ret;
1281
3ae72562
GKB		 1282 if (quick)
1283 ret = utf8_strncasecmp_folded(um, name, entry);
1284 else
1285 ret = utf8_strncasecmp(um, name, entry);
1286
b886ee3e
GKB		 1287 if (ret < 0) {
1288 /* Handle invalid character sequence as either an error
1289 * or as an opaque byte sequence.
1290 */
f8f4acb6 1291 if (sb_has_strict_encoding(sb))
b886ee3e
GKB		 1292 return -EINVAL;
1293
1294 if (name->len != entry->len)
1295 return 1;
1296
1297 return !!memcmp(name->name, entry->name, name->len);
1298 }
1299
1300 return ret;
1301}
3ae72562
GKB		 1302
1303void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1304 struct fscrypt_str *cf_name)
1305{
96fcaf86
GKB		 1306 int len;
1307
f8f4acb6 1308 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
3ae72562
GKB		 1309 cf_name->name = NULL;
1310 return;
1311 }
1312
1313 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1314 if (!cf_name->name)
1315 return;
1316
f8f4acb6 1317 len = utf8_casefold(dir->i_sb->s_encoding,
96fcaf86
GKB		 1318 iname, cf_name->name,
1319 EXT4_NAME_LEN);
1320 if (len <= 0) {
3ae72562
GKB		 1321 kfree(cf_name->name);
1322 cf_name->name = NULL;
96fcaf86 1323 return;
3ae72562 1324 }
96fcaf86
GKB		 1325 cf_name->len = (unsigned) len;
1326
3ae72562 1327}
b886ee3e
GKB		 1328#endif
1329
ac27a0ec 1330/*
d9b9f8d5 1331 * Test whether a directory entry matches the filename being searched for.
ac27a0ec 1332 *
d9b9f8d5 1333 * Return: %true if the directory entry matches, otherwise %false.
ac27a0ec 1334 */
b886ee3e
GKB		 1335static inline bool ext4_match(const struct inode *parent,
1336 const struct ext4_filename *fname,
d9b9f8d5 1337 const struct ext4_dir_entry_2 *de)
ac27a0ec 1338{
067d1023 1339 struct fscrypt_name f;
b886ee3e
GKB		 1340#ifdef CONFIG_UNICODE
1341 const struct qstr entry = {.name = de->name, .len = de->name_len};
1342#endif
1f3862b5 1343
ac27a0ec 1344 if (!de->inode)
d9b9f8d5 1345 return false;
1f3862b5 1346
067d1023
EB		 1347 f.usr_fname = fname->usr_fname;
1348 f.disk_name = fname->disk_name;
643fa961 1349#ifdef CONFIG_FS_ENCRYPTION
067d1023 1350 f.crypto_buf = fname->crypto_buf;
1f3862b5 1351#endif
b886ee3e
GKB		 1352
1353#ifdef CONFIG_UNICODE
f8f4acb6 1354 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
3ae72562
GKB		 1355 if (fname->cf_name.name) {
1356 struct qstr cf = {.name = fname->cf_name.name,
1357 .len = fname->cf_name.len};
1358 return !ext4_ci_compare(parent, &cf, &entry, true);
1359 }
1360 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1361 false);
1362 }
b886ee3e
GKB		 1363#endif
1364
067d1023 1365 return fscrypt_match_name(&f, de->name, de->name_len);
ac27a0ec
DK		 1366}
1367
1368/*
1369 * Returns 0 if not found, -1 on failure, and 1 on success
1370 */
5b643f9c
TT		 1371int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1372 struct inode *dir, struct ext4_filename *fname,
5b643f9c 1373 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
ac27a0ec 1374{
617ba13b 1375 struct ext4_dir_entry_2 * de;
ac27a0ec
DK		 1376 char * dlimit;
1377 int de_len;
1f3862b5 1378
7335cd3b
TM		 1379 de = (struct ext4_dir_entry_2 *)search_buf;
1380 dlimit = search_buf + buf_size;
ac27a0ec
DK		 1381 while ((char *) de < dlimit) {
1382 /* this code is executed quadratically often */
1383 /* do minimal checking `by hand' */
d9b9f8d5 1384 if ((char *) de + de->name_len <= dlimit &&
b886ee3e 1385 ext4_match(dir, fname, de)) {
d9b9f8d5
EB		 1386 /* found a match - just to be sure, do
1387 * a full check */
7303cb5b
JK		 1388 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1389 buf_size, offset))
d9b9f8d5
EB		 1390 return -1;
1391 *res_dir = de;
1392 return 1;
ac27a0ec
DK		 1393 }
1394 /* prevent looping on a bad block */
3d0518f4
WY		 1395 de_len = ext4_rec_len_from_disk(de->rec_len,
1396 dir->i_sb->s_blocksize);
d9b9f8d5
EB		 1397 if (de_len <= 0)
1398 return -1;
ac27a0ec 1399 offset += de_len;
617ba13b 1400 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
ac27a0ec 1401 }
d9b9f8d5 1402 return 0;
ac27a0ec
DK		 1403}
1404
c6af8803
DW		 1405static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1406 struct ext4_dir_entry *de)
1407{
1408 struct super_block *sb = dir->i_sb;
1409
1410 if (!is_dx(dir))
1411 return 0;
1412 if (block == 0)
1413 return 1;
1414 if (de->inode == 0 &&
1415 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1416 sb->s_blocksize)
1417 return 1;
1418 return 0;
1419}
ac27a0ec
DK		 1420
1421/*
b01531db 1422 * __ext4_find_entry()
ac27a0ec
DK		 1423 *
1424 * finds an entry in the specified directory with the wanted name. It
1425 * returns the cache buffer in which the entry was found, and the entry
1426 * itself (as a parameter - res_dir). It does NOT read the inode of the
1427 * entry - you'll have to do that yourself if you want to.
1428 *
1429 * The returned buffer_head has ->b_count elevated. The caller is expected
1430 * to brelse() it when appropriate.
1431 */
b01531db
EB		 1432static struct buffer_head *__ext4_find_entry(struct inode *dir,
1433 struct ext4_filename *fname,
1434 struct ext4_dir_entry_2 **res_dir,
1435 int *inlined)
ac27a0ec 1436{
af5bc92d
TT		 1437 struct super_block *sb;
1438 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1439 struct buffer_head *bh, *ret = NULL;
9699d4f9 1440 ext4_lblk_t start, block;
b01531db 1441 const u8 *name = fname->usr_fname->name;
9699d4f9 1442 size_t ra_max = 0; /* Number of bh's in the readahead
ac27a0ec 1443 buffer, bh_use[] */
9699d4f9 1444 size_t ra_ptr = 0; /* Current index into readahead
ac27a0ec 1445 buffer */
725d26d3 1446 ext4_lblk_t nblocks;
5b643f9c 1447 int i, namelen, retval;
ac27a0ec
DK		 1448
1449 *res_dir = NULL;
1450 sb = dir->i_sb;
b01531db 1451 namelen = fname->usr_fname->len;
617ba13b 1452 if (namelen > EXT4_NAME_LEN)
ac27a0ec 1453 return NULL;
e8e948e7
TM		 1454
1455 if (ext4_has_inline_data(dir)) {
1456 int has_inline_data = 1;
b01531db 1457 ret = ext4_find_inline_entry(dir, fname, res_dir,
e8e948e7 1458 &has_inline_data);
32f7f22c
TM		 1459 if (has_inline_data) {
1460 if (inlined)
1461 *inlined = 1;
5b643f9c 1462 goto cleanup_and_exit;
32f7f22c 1463 }
e8e948e7
TM		 1464 }
1465
8941ec8b 1466 if ((namelen <= 2) && (name[0] == '.') &&
6d5c3aa8 1467 (name[1] == '.' || name[1] == '\0')) {
8941ec8b
TT		 1468 /*
1469 * "." or ".." will only be in the first block
1470 * NFS may look up ".."; "." should be handled by the VFS
1471 */
1472 block = start = 0;
1473 nblocks = 1;
1474 goto restart;
1475 }
ac27a0ec 1476 if (is_dx(dir)) {
b01531db 1477 ret = ext4_dx_find_entry(dir, fname, res_dir);
ac27a0ec
DK		 1478 /*
1479 * On success, or if the error was file not found,
1480 * return. Otherwise, fall back to doing a search the
1481 * old fashioned way.
1482 */
5b643f9c
TT		 1483 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1484 goto cleanup_and_exit;
4776004f
TT		 1485 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1486 "falling back\n"));
f39b3f45 1487 ret = NULL;
ac27a0ec 1488 }
617ba13b 1489 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
9d5afec6
CR		 1490 if (!nblocks) {
1491 ret = NULL;
1492 goto cleanup_and_exit;
1493 }
617ba13b 1494 start = EXT4_I(dir)->i_dir_start_lookup;
ac27a0ec
DK		 1495 if (start >= nblocks)
1496 start = 0;
1497 block = start;
1498restart:
1499 do {
1500 /*
1501 * We deal with the read-ahead logic here.
1502 */
9424ef56 1503 cond_resched();
ac27a0ec
DK		 1504 if (ra_ptr >= ra_max) {
1505 /* Refill the readahead buffer */
1506 ra_ptr = 0;
9699d4f9
TE		 1507 if (block < start)
1508 ra_max = start - block;
1509 else
1510 ra_max = nblocks - block;
1511 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1512 retval = ext4_bread_batch(dir, block, ra_max,
1513 false /* wait */, bh_use);
1514 if (retval) {
1515 ret = ERR_PTR(retval);
1516 ra_max = 0;
1517 goto cleanup_and_exit;
ac27a0ec
DK		 1518 }
1519 }
1520 if ((bh = bh_use[ra_ptr++]) == NULL)
1521 goto next;
1522 wait_on_buffer(bh);
1523 if (!buffer_uptodate(bh)) {
54d3adbc
TT		 1524 EXT4_ERROR_INODE_ERR(dir, EIO,
1525 "reading directory lblock %lu",
1526 (unsigned long) block);
ac27a0ec 1527 brelse(bh);
6febe6f2
KK		 1528 ret = ERR_PTR(-EIO);
1529 goto cleanup_and_exit;
ac27a0ec 1530 }
b0336e8d 1531 if (!buffer_verified(bh) &&
c6af8803
DW		 1532 !is_dx_internal_node(dir, block,
1533 (struct ext4_dir_entry *)bh->b_data) &&
f036adb3 1534 !ext4_dirblock_csum_verify(dir, bh)) {
54d3adbc
TT		 1535 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1536 "checksumming directory "
1537 "block %lu", (unsigned long)block);
b0336e8d 1538 brelse(bh);
bdddf342
TT		 1539 ret = ERR_PTR(-EFSBADCRC);
1540 goto cleanup_and_exit;
b0336e8d
DW		 1541 }
1542 set_buffer_verified(bh);
b01531db 1543 i = search_dirblock(bh, dir, fname,
617ba13b 1544 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
ac27a0ec 1545 if (i == 1) {
617ba13b 1546 EXT4_I(dir)->i_dir_start_lookup = block;
ac27a0ec
DK		 1547 ret = bh;
1548 goto cleanup_and_exit;
1549 } else {
1550 brelse(bh);
1551 if (i < 0)
1552 goto cleanup_and_exit;
1553 }
1554 next:
1555 if (++block >= nblocks)
1556 block = 0;
1557 } while (block != start);
1558
1559 /*
1560 * If the directory has grown while we were searching, then
1561 * search the last part of the directory before giving up.
1562 */
1563 block = nblocks;
617ba13b 1564 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
ac27a0ec
DK		 1565 if (block < nblocks) {
1566 start = 0;
1567 goto restart;
1568 }
1569
1570cleanup_and_exit:
1571 /* Clean up the read-ahead blocks */
1572 for (; ra_ptr < ra_max; ra_ptr++)
af5bc92d 1573 brelse(bh_use[ra_ptr]);
ac27a0ec
DK		 1574 return ret;
1575}
1576
b01531db
EB		 1577static struct buffer_head *ext4_find_entry(struct inode *dir,
1578 const struct qstr *d_name,
1579 struct ext4_dir_entry_2 **res_dir,
1580 int *inlined)
1581{
1582 int err;
1583 struct ext4_filename fname;
1584 struct buffer_head *bh;
1585
1586 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1587 if (err == -ENOENT)
1588 return NULL;
1589 if (err)
1590 return ERR_PTR(err);
1591
1592 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1593
1594 ext4_fname_free_filename(&fname);
1595 return bh;
1596}
1597
1598static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1599 struct dentry *dentry,
1600 struct ext4_dir_entry_2 **res_dir)
1601{
1602 int err;
1603 struct ext4_filename fname;
1604 struct buffer_head *bh;
1605
1606 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
bb9cd910 1607 generic_set_encrypted_ci_d_ops(dentry);
b01531db
EB		 1608 if (err == -ENOENT)
1609 return NULL;
1610 if (err)
1611 return ERR_PTR(err);
1612
1613 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1614
1615 ext4_fname_free_filename(&fname);
1616 return bh;
1617}
1618
5b643f9c
TT		 1619static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1620 struct ext4_filename *fname,
1621 struct ext4_dir_entry_2 **res_dir)
ac27a0ec 1622{
8941ec8b 1623 struct super_block * sb = dir->i_sb;
e08ac99f 1624 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 1625 struct buffer_head *bh;
725d26d3 1626 ext4_lblk_t block;
ac27a0ec 1627 int retval;
ac27a0ec 1628
643fa961 1629#ifdef CONFIG_FS_ENCRYPTION
1f3862b5
MH		 1630 *res_dir = NULL;
1631#endif
5b643f9c 1632 frame = dx_probe(fname, dir, NULL, frames);
dd73b5d5
TT		 1633 if (IS_ERR(frame))
1634 return (struct buffer_head *) frame;
ac27a0ec
DK		 1635 do {
1636 block = dx_get_block(frame->at);
4e19d6b6 1637 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
537d8f93 1638 if (IS_ERR(bh))
b0336e8d 1639 goto errout;
537d8f93 1640
d6b97550 1641 retval = search_dirblock(bh, dir, fname,
7845c049
TT		 1642 block << EXT4_BLOCK_SIZE_BITS(sb),
1643 res_dir);
537d8f93
TT		 1644 if (retval == 1)
1645 goto success;
af5bc92d 1646 brelse(bh);
7845c049 1647 if (retval == -1) {
537d8f93 1648 bh = ERR_PTR(ERR_BAD_DX_DIR);
7845c049
TT		 1649 goto errout;
1650 }
1651
ac27a0ec 1652 /* Check to see if we should continue to search */
5b643f9c 1653 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
ac27a0ec
DK		 1654 frames, NULL);
1655 if (retval < 0) {
b03a2f7e
AD		 1656 ext4_warning_inode(dir,
1657 "error %d reading directory index block",
1658 retval);
537d8f93 1659 bh = ERR_PTR(retval);
ac27a0ec
DK		 1660 goto errout;
1661 }
1662 } while (retval == 1);
1663
537d8f93 1664 bh = NULL;
ac27a0ec 1665errout:
d6b97550 1666 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
537d8f93
TT		 1667success:
1668 dx_release(frames);
1669 return bh;
ac27a0ec 1670}
ac27a0ec 1671
00cd8dd3 1672static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
ac27a0ec 1673{
af5bc92d
TT		 1674 struct inode *inode;
1675 struct ext4_dir_entry_2 *de;
1676 struct buffer_head *bh;
28b4c263 1677
89904275
EB		 1678 if (dentry->d_name.len > EXT4_NAME_LEN)
1679 return ERR_PTR(-ENAMETOOLONG);
ac27a0ec 1680
b01531db 1681 bh = ext4_lookup_entry(dir, dentry, &de);
36de9286 1682 if (IS_ERR(bh))
e884bce1 1683 return ERR_CAST(bh);
ac27a0ec
DK		 1684 inode = NULL;
1685 if (bh) {
498e5f24 1686 __u32 ino = le32_to_cpu(de->inode);
af5bc92d 1687 brelse(bh);
617ba13b 1688 if (!ext4_valid_inum(dir->i_sb, ino)) {
24676da4 1689 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
6a797d27 1690 return ERR_PTR(-EFSCORRUPTED);
a6c15c2b 1691 }
7e936b73 1692 if (unlikely(ino == dir->i_ino)) {
a34e15cc
DH		 1693 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1694 dentry);
6a797d27 1695 return ERR_PTR(-EFSCORRUPTED);
7e936b73 1696 }
8a363970 1697 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
a9049376
AV		 1698 if (inode == ERR_PTR(-ESTALE)) {
1699 EXT4_ERROR_INODE(dir,
1700 "deleted inode referenced: %u",
1701 ino);
6a797d27 1702 return ERR_PTR(-EFSCORRUPTED);
e6f009b0 1703 }
592ddec7 1704 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
ff978b09 1705 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
a7550b30 1706 !fscrypt_has_permitted_context(dir, inode)) {
d9cdc903 1707 ext4_warning(inode->i_sb,
8d2ae1cb 1708 "Inconsistent encryption contexts: %lu/%lu",
8c68084b 1709 dir->i_ino, inode->i_ino);
dd01b690 1710 iput(inode);
d9cdc903
TT		 1711 return ERR_PTR(-EPERM);
1712 }
ac27a0ec 1713 }
b886ee3e
GKB		 1714
1715#ifdef CONFIG_UNICODE
1716 if (!inode && IS_CASEFOLDED(dir)) {
1717 /* Eventually we want to call d_add_ci(dentry, NULL)
1718 * for negative dentries in the encoding case as
1719 * well. For now, prevent the negative dentry
1720 * from being cached.
1721 */
1722 return NULL;
1723 }
1724#endif
ac27a0ec
DK		 1725 return d_splice_alias(inode, dentry);
1726}
1727
1728
617ba13b 1729struct dentry *ext4_get_parent(struct dentry *child)
ac27a0ec 1730{
498e5f24 1731 __u32 ino;
26fe5750 1732 static const struct qstr dotdot = QSTR_INIT("..", 2);
617ba13b 1733 struct ext4_dir_entry_2 * de;
ac27a0ec
DK		 1734 struct buffer_head *bh;
1735
2b0143b5 1736 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
36de9286 1737 if (IS_ERR(bh))
e884bce1 1738 return ERR_CAST(bh);
ac27a0ec
DK		 1739 if (!bh)
1740 return ERR_PTR(-ENOENT);
1741 ino = le32_to_cpu(de->inode);
1742 brelse(bh);
1743
fc64005c 1744 if (!ext4_valid_inum(child->d_sb, ino)) {
2b0143b5 1745 EXT4_ERROR_INODE(d_inode(child),
24676da4 1746 "bad parent inode number: %u", ino);
6a797d27 1747 return ERR_PTR(-EFSCORRUPTED);
a6c15c2b
VA		 1748 }
1749
8a363970 1750 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
ac27a0ec
DK		 1751}
1752
ef2b02d3
ES		 1753/*
1754 * Move count entries from end of map between two memory locations.
1755 * Returns pointer to last entry moved.
1756 */
617ba13b 1757static struct ext4_dir_entry_2 *
3d0518f4
WY		 1758dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1759 unsigned blocksize)
ac27a0ec
DK		 1760{
1761 unsigned rec_len = 0;
1762
1763 while (count--) {
60e6679e 1764 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
9aee2286 1765 (from + (map->offs<<2));
617ba13b 1766 rec_len = EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec 1767 memcpy (to, de, rec_len);
617ba13b 1768 ((struct ext4_dir_entry_2 *) to)->rec_len =
3d0518f4 1769 ext4_rec_len_to_disk(rec_len, blocksize);
ac27a0ec
DK		 1770 de->inode = 0;
1771 map++;
1772 to += rec_len;
1773 }
617ba13b 1774 return (struct ext4_dir_entry_2 *) (to - rec_len);
ac27a0ec
DK		 1775}
1776
ef2b02d3
ES		 1777/*
1778 * Compact each dir entry in the range to the minimal rec_len.
1779 * Returns pointer to last entry in range.
1780 */
8bad4597 1781static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
ac27a0ec 1782{
617ba13b 1783 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
ac27a0ec
DK		 1784 unsigned rec_len = 0;
1785
1786 prev = to = de;
8bad4597 1787 while ((char*)de < base + blocksize) {
3d0518f4 1788 next = ext4_next_entry(de, blocksize);
ac27a0ec 1789 if (de->inode && de->name_len) {
617ba13b 1790 rec_len = EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec
DK		 1791 if (de > to)
1792 memmove(to, de, rec_len);
3d0518f4 1793 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
ac27a0ec 1794 prev = to;
617ba13b 1795 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
ac27a0ec
DK		 1796 }
1797 de = next;
1798 }
1799 return prev;
1800}
1801
ef2b02d3
ES		 1802/*
1803 * Split a full leaf block to make room for a new dir entry.
1804 * Allocate a new block, and move entries so that they are approx. equally full.
1805 * Returns pointer to de in block into which the new entry will be inserted.
1806 */
617ba13b 1807static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
ac27a0ec 1808 struct buffer_head **bh,struct dx_frame *frame,
f8b3b59d 1809 struct dx_hash_info *hinfo)
ac27a0ec
DK		 1810{
1811 unsigned blocksize = dir->i_sb->s_blocksize;
1812 unsigned count, continued;
1813 struct buffer_head *bh2;
725d26d3 1814 ext4_lblk_t newblock;
ac27a0ec
DK		 1815 u32 hash2;
1816 struct dx_map_entry *map;
1817 char *data1 = (*bh)->b_data, *data2;
59e315b4 1818 unsigned split, move, size;
617ba13b 1819 struct ext4_dir_entry_2 *de = NULL, *de2;
b0336e8d 1820 int csum_size = 0;
59e315b4 1821 int err = 0, i;
ac27a0ec 1822
9aa5d32b 1823 if (ext4_has_metadata_csum(dir->i_sb))
b0336e8d
DW		 1824 csum_size = sizeof(struct ext4_dir_entry_tail);
1825
0f70b406
TT		 1826 bh2 = ext4_append(handle, dir, &newblock);
1827 if (IS_ERR(bh2)) {
ac27a0ec
DK		 1828 brelse(*bh);
1829 *bh = NULL;
f8b3b59d 1830 return (struct ext4_dir_entry_2 *) bh2;
ac27a0ec
DK		 1831 }
1832
1833 BUFFER_TRACE(*bh, "get_write_access");
617ba13b 1834 err = ext4_journal_get_write_access(handle, *bh);
fedee54d
DM		 1835 if (err)
1836 goto journal_error;
1837
ac27a0ec 1838 BUFFER_TRACE(frame->bh, "get_write_access");
617ba13b 1839 err = ext4_journal_get_write_access(handle, frame->bh);
ac27a0ec
DK		 1840 if (err)
1841 goto journal_error;
1842
1843 data2 = bh2->b_data;
1844
1845 /* create map in the end of data2 block */
1846 map = (struct dx_map_entry *) (data2 + blocksize);
1f3862b5 1847 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
ac27a0ec
DK		 1848 blocksize, hinfo, map);
1849 map -= count;
af5bc92d 1850 dx_sort_map(map, count);
5872331b 1851 /* Ensure that neither split block is over half full */
ef2b02d3
ES		 1852 size = 0;
1853 move = 0;
1854 for (i = count-1; i >= 0; i--) {
1855 /* is more than half of this entry in 2nd half of the block? */
1856 if (size + map[i].size/2 > blocksize/2)
1857 break;
1858 size += map[i].size;
1859 move++;
1860 }
5872331b
ES		 1861 /*
1862 * map index at which we will split
1863 *
1864 * If the sum of active entries didn't exceed half the block size, just
1865 * split it in half by count; each resulting block will have at least
1866 * half the space free.
1867 */
1868 if (i > 0)
1869 split = count - move;
1870 else
1871 split = count/2;
1872
ac27a0ec
DK		 1873 hash2 = map[split].hash;
1874 continued = hash2 == map[split - 1].hash;
725d26d3
AK		 1875 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1876 (unsigned long)dx_get_block(frame->at),
1877 hash2, split, count-split));
ac27a0ec
DK		 1878
1879 /* Fancy dance to stay within two buffers */
1f3862b5
MH		 1880 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1881 blocksize);
af5bc92d 1882 de = dx_pack_dirents(data1, blocksize);
b0336e8d
DW		 1883 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1884 (char *) de,
3d0518f4 1885 blocksize);
b0336e8d
DW		 1886 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1887 (char *) de2,
3d0518f4 1888 blocksize);
b0336e8d 1889 if (csum_size) {
ddce3b94
TT		 1890 ext4_initialize_dirent_tail(*bh, blocksize);
1891 ext4_initialize_dirent_tail(bh2, blocksize);
b0336e8d
DW		 1892 }
1893
b3098486
MH		 1894 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1895 blocksize, 1));
1896 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1897 blocksize, 1));
ac27a0ec
DK		 1898
1899 /* Which block gets the new entry? */
f8b3b59d 1900 if (hinfo->hash >= hash2) {
ac27a0ec
DK		 1901 swap(*bh, bh2);
1902 de = de2;
1903 }
af5bc92d 1904 dx_insert_block(frame, hash2 + continued, newblock);
f036adb3 1905 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
ac27a0ec
DK		 1906 if (err)
1907 goto journal_error;
dbe89444 1908 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
ac27a0ec
DK		 1909 if (err)
1910 goto journal_error;
af5bc92d
TT		 1911 brelse(bh2);
1912 dxtrace(dx_show_index("frame", frame->entries));
ac27a0ec 1913 return de;
fedee54d
DM		 1914
1915journal_error:
1916 brelse(*bh);
1917 brelse(bh2);
1918 *bh = NULL;
1919 ext4_std_error(dir->i_sb, err);
f8b3b59d 1920 return ERR_PTR(err);
ac27a0ec 1921}
ac27a0ec 1922
978fef91
TM		 1923int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1924 struct buffer_head *bh,
1925 void *buf, int buf_size,
5b643f9c 1926 struct ext4_filename *fname,
978fef91
TM		 1927 struct ext4_dir_entry_2 **dest_de)
1928{
1929 struct ext4_dir_entry_2 *de;
5b643f9c 1930 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
978fef91
TM		 1931 int nlen, rlen;
1932 unsigned int offset = 0;
1933 char *top;
1f3862b5 1934
978fef91
TM		 1935 de = (struct ext4_dir_entry_2 *)buf;
1936 top = buf + buf_size - reclen;
1937 while ((char *) de <= top) {
1938 if (ext4_check_dir_entry(dir, NULL, de, bh,
d9b9f8d5
EB		 1939 buf, buf_size, offset))
1940 return -EFSCORRUPTED;
b886ee3e 1941 if (ext4_match(dir, fname, de))
d9b9f8d5 1942 return -EEXIST;
978fef91
TM		 1943 nlen = EXT4_DIR_REC_LEN(de->name_len);
1944 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1945 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1946 break;
1947 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1948 offset += rlen;
1949 }
1f3862b5 1950 if ((char *) de > top)
d9b9f8d5
EB		 1951 return -ENOSPC;
1952
1953 *dest_de = de;
1954 return 0;
978fef91
TM		 1955}
1956
1bc0af60
EB		 1957void ext4_insert_dentry(struct inode *inode,
1958 struct ext4_dir_entry_2 *de,
1959 int buf_size,
1960 struct ext4_filename *fname)
978fef91
TM		 1961{
1962
1963 int nlen, rlen;
1964
1965 nlen = EXT4_DIR_REC_LEN(de->name_len);
1966 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1967 if (de->inode) {
1968 struct ext4_dir_entry_2 *de1 =
4bdfc873 1969 (struct ext4_dir_entry_2 *)((char *)de + nlen);
978fef91
TM		 1970 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1971 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1972 de = de1;
1973 }
1974 de->file_type = EXT4_FT_UNKNOWN;
1975 de->inode = cpu_to_le32(inode->i_ino);
1976 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
5b643f9c
TT		 1977 de->name_len = fname_len(fname);
1978 memcpy(de->name, fname_name(fname), fname_len(fname));
978fef91 1979}
4bdfc873 1980
ac27a0ec
DK		 1981/*
1982 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1983 * it points to a directory entry which is guaranteed to be large
1984 * enough for new directory entry. If de is NULL, then
1985 * add_dirent_to_buf will attempt search the directory block for
1986 * space. It will return -ENOSPC if no space is available, and -EIO
1987 * and -EEXIST if directory entry already exists.
ac27a0ec 1988 */
5b643f9c
TT		 1989static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1990 struct inode *dir,
617ba13b 1991 struct inode *inode, struct ext4_dir_entry_2 *de,
af5bc92d 1992 struct buffer_head *bh)
ac27a0ec 1993{
3d0518f4 1994 unsigned int blocksize = dir->i_sb->s_blocksize;
b0336e8d 1995 int csum_size = 0;
4209ae12 1996 int err, err2;
b0336e8d 1997
9aa5d32b 1998 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 1999 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec 2000
ac27a0ec 2001 if (!de) {
5b643f9c
TT		 2002 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2003 blocksize - csum_size, fname, &de);
978fef91
TM		 2004 if (err)
2005 return err;
ac27a0ec
DK		 2006 }
2007 BUFFER_TRACE(bh, "get_write_access");
617ba13b 2008 err = ext4_journal_get_write_access(handle, bh);
ac27a0ec 2009 if (err) {
617ba13b 2010 ext4_std_error(dir->i_sb, err);
ac27a0ec
DK		 2011 return err;
2012 }
2013
1bc0af60
EB		 2014 /* By now the buffer is marked for journaling */
2015 ext4_insert_dentry(inode, de, blocksize, fname);
978fef91 2016
ac27a0ec
DK		 2017 /*
2018 * XXX shouldn't update any times until successful
2019 * completion of syscall, but too many callers depend
2020 * on this.
2021 *
2022 * XXX similarly, too many callers depend on
617ba13b 2023 * ext4_new_inode() setting the times, but error
ac27a0ec
DK		 2024 * recovery deletes the inode, so the worst that can
2025 * happen is that the times are slightly out of date
2026 * and/or different from the directory change time.
2027 */
eeca7ea1 2028 dir->i_mtime = dir->i_ctime = current_time(dir);
617ba13b 2029 ext4_update_dx_flag(dir);
e08ac99f 2030 inode_inc_iversion(dir);
4209ae12 2031 err2 = ext4_mark_inode_dirty(handle, dir);
0390131b 2032 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
f036adb3 2033 err = ext4_handle_dirty_dirblock(handle, dir, bh);
ac27a0ec 2034 if (err)
617ba13b 2035 ext4_std_error(dir->i_sb, err);
4209ae12 2036 return err ? err : err2;
ac27a0ec
DK		 2037}
2038
ac27a0ec
DK		 2039/*
2040 * This converts a one block unindexed directory to a 3 block indexed
2041 * directory, and adds the dentry to the indexed directory.
2042 */
5b643f9c 2043static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
56a04915 2044 struct inode *dir,
ac27a0ec
DK		 2045 struct inode *inode, struct buffer_head *bh)
2046{
ac27a0ec
DK		 2047 struct buffer_head *bh2;
2048 struct dx_root *root;
e08ac99f 2049 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 2050 struct dx_entry *entries;
617ba13b 2051 struct ext4_dir_entry_2 *de, *de2;
ddce3b94 2052 char *data2, *top;
ac27a0ec
DK		 2053 unsigned len;
2054 int retval;
2055 unsigned blocksize;
725d26d3 2056 ext4_lblk_t block;
ac27a0ec 2057 struct fake_dirent *fde;
4bdfc873
MH		 2058 int csum_size = 0;
2059
9aa5d32b 2060 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 2061 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec
DK		 2062
2063 blocksize = dir->i_sb->s_blocksize;
e6b8bc09 2064 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
5d601255 2065 BUFFER_TRACE(bh, "get_write_access");
617ba13b 2066 retval = ext4_journal_get_write_access(handle, bh);
ac27a0ec 2067 if (retval) {
617ba13b 2068 ext4_std_error(dir->i_sb, retval);
ac27a0ec
DK		 2069 brelse(bh);
2070 return retval;
2071 }
2072 root = (struct dx_root *) bh->b_data;
2073
e6b8bc09
TT		 2074 /* The 0th block becomes the root, move the dirents out */
2075 fde = &root->dotdot;
2076 de = (struct ext4_dir_entry_2 *)((char *)fde +
3d0518f4 2077 ext4_rec_len_from_disk(fde->rec_len, blocksize));
e6b8bc09 2078 if ((char *) de >= (((char *) root) + blocksize)) {
24676da4 2079 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
e6b8bc09 2080 brelse(bh);
6a797d27 2081 return -EFSCORRUPTED;
e6b8bc09 2082 }
b0336e8d 2083 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
e6b8bc09
TT		 2084
2085 /* Allocate new block for the 0th block's dirents */
0f70b406
TT		 2086 bh2 = ext4_append(handle, dir, &block);
2087 if (IS_ERR(bh2)) {
ac27a0ec 2088 brelse(bh);
0f70b406 2089 return PTR_ERR(bh2);
ac27a0ec 2090 }
12e9b892 2091 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
ddce3b94 2092 data2 = bh2->b_data;
ac27a0ec 2093
ddce3b94
TT		 2094 memcpy(data2, de, len);
2095 de = (struct ext4_dir_entry_2 *) data2;
2096 top = data2 + len;
3d0518f4 2097 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
ac27a0ec 2098 de = de2;
ddce3b94
TT		 2099 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2100 (char *) de, blocksize);
b0336e8d 2101
ddce3b94
TT		 2102 if (csum_size)
2103 ext4_initialize_dirent_tail(bh2, blocksize);
b0336e8d 2104
ac27a0ec 2105 /* Initialize the root; the dot dirents already exist */
617ba13b 2106 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
3d0518f4
WY		 2107 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2108 blocksize);
ac27a0ec
DK		 2109 memset (&root->info, 0, sizeof(root->info));
2110 root->info.info_length = sizeof(root->info);
617ba13b 2111 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
ac27a0ec 2112 entries = root->entries;
af5bc92d
TT		 2113 dx_set_block(entries, 1);
2114 dx_set_count(entries, 1);
2115 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
ac27a0ec
DK		 2116
2117 /* Initialize as for dx_probe */
5b643f9c
TT		 2118 fname->hinfo.hash_version = root->info.hash_version;
2119 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2120 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2121 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
b886ee3e 2122 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
5b643f9c 2123
6050d47a 2124 memset(frames, 0, sizeof(frames));
ac27a0ec
DK		 2125 frame = frames;
2126 frame->entries = entries;
2127 frame->at = entries;
2128 frame->bh = bh;
6976a6f2 2129
6050d47a
JK		 2130 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2131 if (retval)
2132 goto out_frames;
f036adb3 2133 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
6050d47a
JK		 2134 if (retval)
2135 goto out_frames;
6976a6f2 2136
e81d4477 2137 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
f8b3b59d 2138 if (IS_ERR(de)) {
6050d47a
JK		 2139 retval = PTR_ERR(de);
2140 goto out_frames;
7ad8e4e6 2141 }
ac27a0ec 2142
e81d4477 2143 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
6050d47a
JK		 2144out_frames:
2145 /*
2146 * Even if the block split failed, we have to properly write
2147 * out all the changes we did so far. Otherwise we can end up
2148 * with corrupted filesystem.
2149 */
e81d4477 2150 if (retval)
2151 ext4_mark_inode_dirty(handle, dir);
6050d47a 2152 dx_release(frames);
e81d4477 2153 brelse(bh2);
6050d47a 2154 return retval;
ac27a0ec 2155}
ac27a0ec
DK		 2156
2157/*
617ba13b 2158 * ext4_add_entry()
ac27a0ec
DK		 2159 *
2160 * adds a file entry to the specified directory, using the same
617ba13b 2161 * semantics as ext4_find_entry(). It returns NULL if it failed.
ac27a0ec
DK		 2162 *
2163 * NOTE!! The inode part of 'de' is left at 0 - which means you
2164 * may not sleep between calling this and putting something into
2165 * the entry, as someone else might have used it while you slept.
2166 */
af5bc92d
TT		 2167static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2168 struct inode *inode)
ac27a0ec 2169{
2b0143b5 2170 struct inode *dir = d_inode(dentry->d_parent);
e12fb972 2171 struct buffer_head *bh = NULL;
617ba13b 2172 struct ext4_dir_entry_2 *de;
af5bc92d 2173 struct super_block *sb;
5b643f9c 2174 struct ext4_filename fname;
ac27a0ec 2175 int retval;
ac27a0ec 2176 int dx_fallback=0;
ac27a0ec 2177 unsigned blocksize;
725d26d3 2178 ext4_lblk_t block, blocks;
b0336e8d
DW		 2179 int csum_size = 0;
2180
9aa5d32b 2181 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 2182 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec
DK		 2183
2184 sb = dir->i_sb;
2185 blocksize = sb->s_blocksize;
2186 if (!dentry->d_name.len)
2187 return -EINVAL;
3c47d541 2188
75d18cd1
EB		 2189 if (fscrypt_is_nokey_name(dentry))
2190 return -ENOKEY;
2191
b886ee3e 2192#ifdef CONFIG_UNICODE
f8f4acb6
DR		 2193 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2194 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
b886ee3e
GKB		 2195 return -EINVAL;
2196#endif
2197
5b643f9c
TT		 2198 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2199 if (retval)
2200 return retval;
2201
3c47d541 2202 if (ext4_has_inline_data(dir)) {
56a04915 2203 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
3c47d541 2204 if (retval < 0)
5b643f9c 2205 goto out;
3c47d541
TM		 2206 if (retval == 1) {
2207 retval = 0;
e12fb972 2208 goto out;
3c47d541
TM		 2209 }
2210 }
2211
ac27a0ec 2212 if (is_dx(dir)) {
56a04915 2213 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
ac27a0ec 2214 if (!retval || (retval != ERR_BAD_DX_DIR))
e12fb972 2215 goto out;
48a34311
JK		 2216 /* Can we just ignore htree data? */
2217 if (ext4_has_metadata_csum(sb)) {
2218 EXT4_ERROR_INODE(dir,
2219 "Directory has corrupted htree index.");
2220 retval = -EFSCORRUPTED;
2221 goto out;
2222 }
12e9b892 2223 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
ac27a0ec 2224 dx_fallback++;
4209ae12
HS		 2225 retval = ext4_mark_inode_dirty(handle, dir);
2226 if (unlikely(retval))
2227 goto out;
ac27a0ec 2228 }
ac27a0ec 2229 blocks = dir->i_size >> sb->s_blocksize_bits;
498e5f24 2230 for (block = 0; block < blocks; block++) {
dc6982ff 2231 bh = ext4_read_dirblock(dir, block, DIRENT);
4e19d6b6
TT		 2232 if (bh == NULL) {
2233 bh = ext4_bread(handle, dir, block,
2234 EXT4_GET_BLOCKS_CREATE);
2235 goto add_to_new_block;
2236 }
5b643f9c
TT		 2237 if (IS_ERR(bh)) {
2238 retval = PTR_ERR(bh);
2239 bh = NULL;
2240 goto out;
2241 }
2242 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2243 NULL, bh);
e12fb972
LC		 2244 if (retval != -ENOSPC)
2245 goto out;
ac27a0ec 2246
ac27a0ec 2247 if (blocks == 1 && !dx_fallback &&
e2b911c5 2248 ext4_has_feature_dir_index(sb)) {
56a04915 2249 retval = make_indexed_dir(handle, &fname, dir,
5b643f9c 2250 inode, bh);
e12fb972
LC		 2251 bh = NULL; /* make_indexed_dir releases bh */
2252 goto out;
2253 }
ac27a0ec
DK		 2254 brelse(bh);
2255 }
0f70b406 2256 bh = ext4_append(handle, dir, &block);
4e19d6b6 2257add_to_new_block:
5b643f9c
TT		 2258 if (IS_ERR(bh)) {
2259 retval = PTR_ERR(bh);
2260 bh = NULL;
2261 goto out;
2262 }
617ba13b 2263 de = (struct ext4_dir_entry_2 *) bh->b_data;
ac27a0ec 2264 de->inode = 0;
b0336e8d
DW		 2265 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2266
ddce3b94
TT		 2267 if (csum_size)
2268 ext4_initialize_dirent_tail(bh, blocksize);
b0336e8d 2269
5b643f9c 2270 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
e12fb972 2271out:
5b643f9c 2272 ext4_fname_free_filename(&fname);
2de770a4 2273 brelse(bh);
14ece102
FM		 2274 if (retval == 0)
2275 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2de770a4 2276 return retval;
ac27a0ec
DK		 2277}
2278
ac27a0ec
DK		 2279/*
2280 * Returns 0 for success, or a negative error value
2281 */
5b643f9c 2282static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
56a04915 2283 struct inode *dir, struct inode *inode)
ac27a0ec 2284{
e08ac99f 2285 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 2286 struct dx_entry *entries, *at;
af5bc92d 2287 struct buffer_head *bh;
af5bc92d 2288 struct super_block *sb = dir->i_sb;
617ba13b 2289 struct ext4_dir_entry_2 *de;
e08ac99f 2290 int restart;
ac27a0ec
DK		 2291 int err;
2292
e08ac99f
AB		 2293again:
2294 restart = 0;
5b643f9c 2295 frame = dx_probe(fname, dir, NULL, frames);
dd73b5d5
TT		 2296 if (IS_ERR(frame))
2297 return PTR_ERR(frame);
ac27a0ec
DK		 2298 entries = frame->entries;
2299 at = frame->at;
4e19d6b6 2300 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
dc6982ff
TT		 2301 if (IS_ERR(bh)) {
2302 err = PTR_ERR(bh);
2303 bh = NULL;
ac27a0ec 2304 goto cleanup;
6d1ab10e 2305 }
ac27a0ec
DK		 2306
2307 BUFFER_TRACE(bh, "get_write_access");
617ba13b 2308 err = ext4_journal_get_write_access(handle, bh);
ac27a0ec
DK		 2309 if (err)
2310 goto journal_error;
2311
5b643f9c 2312 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2de770a4 2313 if (err != -ENOSPC)
ac27a0ec 2314 goto cleanup;
ac27a0ec 2315
e08ac99f 2316 err = 0;
ac27a0ec 2317 /* Block full, should compress but for now just split */
4776004f 2318 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
ac27a0ec
DK		 2319 dx_get_count(entries), dx_get_limit(entries)));
2320 /* Need to split index? */
2321 if (dx_get_count(entries) == dx_get_limit(entries)) {
725d26d3 2322 ext4_lblk_t newblock;
e08ac99f
AB		 2323 int levels = frame - frames + 1;
2324 unsigned int icount;
2325 int add_level = 1;
ac27a0ec
DK		 2326 struct dx_entry *entries2;
2327 struct dx_node *node2;
2328 struct buffer_head *bh2;
2329
e08ac99f
AB		 2330 while (frame > frames) {
2331 if (dx_get_count((frame - 1)->entries) <
2332 dx_get_limit((frame - 1)->entries)) {
2333 add_level = 0;
2334 break;
2335 }
2336 frame--; /* split higher index block */
2337 at = frame->at;
2338 entries = frame->entries;
2339 restart = 1;
2340 }
2341 if (add_level && levels == ext4_dir_htree_level(sb)) {
2342 ext4_warning(sb, "Directory (ino: %lu) index full, "
2343 "reach max htree level :%d",
2344 dir->i_ino, levels);
2345 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2346 ext4_warning(sb, "Large directory feature is "
2347 "not enabled on this "
2348 "filesystem");
2349 }
ac27a0ec
DK		 2350 err = -ENOSPC;
2351 goto cleanup;
2352 }
e08ac99f 2353 icount = dx_get_count(entries);
0f70b406
TT		 2354 bh2 = ext4_append(handle, dir, &newblock);
2355 if (IS_ERR(bh2)) {
2356 err = PTR_ERR(bh2);
ac27a0ec 2357 goto cleanup;
0f70b406 2358 }
ac27a0ec
DK		 2359 node2 = (struct dx_node *)(bh2->b_data);
2360 entries2 = node2->entries;
1f7bebb9 2361 memset(&node2->fake, 0, sizeof(struct fake_dirent));
3d0518f4
WY		 2362 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2363 sb->s_blocksize);
ac27a0ec 2364 BUFFER_TRACE(frame->bh, "get_write_access");
617ba13b 2365 err = ext4_journal_get_write_access(handle, frame->bh);
ac27a0ec
DK		 2366 if (err)
2367 goto journal_error;
e08ac99f 2368 if (!add_level) {
ac27a0ec
DK		 2369 unsigned icount1 = icount/2, icount2 = icount - icount1;
2370 unsigned hash2 = dx_get_hash(entries + icount1);
4776004f
TT		 2371 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2372 icount1, icount2));
ac27a0ec
DK		 2373
2374 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
617ba13b 2375 err = ext4_journal_get_write_access(handle,
e08ac99f 2376 (frame - 1)->bh);
ac27a0ec
DK		 2377 if (err)
2378 goto journal_error;
2379
af5bc92d
TT		 2380 memcpy((char *) entries2, (char *) (entries + icount1),
2381 icount2 * sizeof(struct dx_entry));
2382 dx_set_count(entries, icount1);
2383 dx_set_count(entries2, icount2);
2384 dx_set_limit(entries2, dx_node_limit(dir));
ac27a0ec
DK		 2385
2386 /* Which index block gets the new entry? */
2387 if (at - entries >= icount1) {
2388 frame->at = at = at - entries - icount1 + entries2;
2389 frame->entries = entries = entries2;
2390 swap(frame->bh, bh2);
2391 }
e08ac99f
AB		 2392 dx_insert_block((frame - 1), hash2, newblock);
2393 dxtrace(dx_show_index("node", frame->entries));
af5bc92d 2394 dxtrace(dx_show_index("node",
ac27a0ec 2395 ((struct dx_node *) bh2->b_data)->entries));
dbe89444 2396 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
ac27a0ec
DK		 2397 if (err)
2398 goto journal_error;
2399 brelse (bh2);
e08ac99f
AB		 2400 err = ext4_handle_dirty_dx_node(handle, dir,
2401 (frame - 1)->bh);
2402 if (err)
2403 goto journal_error;
2404 if (restart) {
2405 err = ext4_handle_dirty_dx_node(handle, dir,
2406 frame->bh);
2407 goto journal_error;
2408 }
ac27a0ec 2409 } else {
e08ac99f 2410 struct dx_root *dxroot;
ac27a0ec
DK		 2411 memcpy((char *) entries2, (char *) entries,
2412 icount * sizeof(struct dx_entry));
2413 dx_set_limit(entries2, dx_node_limit(dir));
2414
2415 /* Set up root */
2416 dx_set_count(entries, 1);
2417 dx_set_block(entries + 0, newblock);
e08ac99f
AB		 2418 dxroot = (struct dx_root *)frames[0].bh->b_data;
2419 dxroot->info.indirect_levels += 1;
2420 dxtrace(printk(KERN_DEBUG
2421 "Creating %d level index...\n",
799578ab 2422 dxroot->info.indirect_levels));
e08ac99f 2423 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
ac27a0ec
DK		 2424 if (err)
2425 goto journal_error;
e08ac99f
AB		 2426 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2427 brelse(bh2);
2428 restart = 1;
2429 goto journal_error;
b4097142 2430 }
ac27a0ec 2431 }
5b643f9c 2432 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
f8b3b59d
TT		 2433 if (IS_ERR(de)) {
2434 err = PTR_ERR(de);
ac27a0ec 2435 goto cleanup;
f8b3b59d 2436 }
5b643f9c 2437 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
ac27a0ec
DK		 2438 goto cleanup;
2439
2440journal_error:
e08ac99f 2441 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
ac27a0ec 2442cleanup:
b1deefc9 2443 brelse(bh);
ac27a0ec 2444 dx_release(frames);
e08ac99f
AB		 2445 /* @restart is true means htree-path has been changed, we need to
2446 * repeat dx_probe() to find out valid htree-path
2447 */
2448 if (restart && err == 0)
2449 goto again;
ac27a0ec
DK		 2450 return err;
2451}
ac27a0ec
DK		 2452
2453/*
05019a9e
TM		 2454 * ext4_generic_delete_entry deletes a directory entry by merging it
2455 * with the previous entry
ac27a0ec 2456 */
2fe34d29 2457int ext4_generic_delete_entry(struct inode *dir,
05019a9e
TM		 2458 struct ext4_dir_entry_2 *de_del,
2459 struct buffer_head *bh,
2460 void *entry_buf,
2461 int buf_size,
2462 int csum_size)
ac27a0ec 2463{
af5bc92d 2464 struct ext4_dir_entry_2 *de, *pde;
3d0518f4 2465 unsigned int blocksize = dir->i_sb->s_blocksize;
05019a9e 2466 int i;
b0336e8d 2467
ac27a0ec
DK		 2468 i = 0;
2469 pde = NULL;
05019a9e
TM		 2470 de = (struct ext4_dir_entry_2 *)entry_buf;
2471 while (i < buf_size - csum_size) {
226ba972 2472 if (ext4_check_dir_entry(dir, NULL, de, bh,
7303cb5b 2473 entry_buf, buf_size, i))
6a797d27 2474 return -EFSCORRUPTED;
ac27a0ec 2475 if (de == de_del) {
ac27a0ec 2476 if (pde)
a72d7f83 2477 pde->rec_len = ext4_rec_len_to_disk(
3d0518f4
WY		 2478 ext4_rec_len_from_disk(pde->rec_len,
2479 blocksize) +
2480 ext4_rec_len_from_disk(de->rec_len,
2481 blocksize),
2482 blocksize);
ac27a0ec
DK		 2483 else
2484 de->inode = 0;
e08ac99f 2485 inode_inc_iversion(dir);
ac27a0ec
DK		 2486 return 0;
2487 }
3d0518f4 2488 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
ac27a0ec 2489 pde = de;
3d0518f4 2490 de = ext4_next_entry(de, blocksize);
ac27a0ec
DK		 2491 }
2492 return -ENOENT;
2493}
2494
05019a9e
TM		 2495static int ext4_delete_entry(handle_t *handle,
2496 struct inode *dir,
2497 struct ext4_dir_entry_2 *de_del,
2498 struct buffer_head *bh)
2499{
2500 int err, csum_size = 0;
2501
9f40fe54
TM		 2502 if (ext4_has_inline_data(dir)) {
2503 int has_inline_data = 1;
2504 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2505 &has_inline_data);
2506 if (has_inline_data)
2507 return err;
2508 }
2509
9aa5d32b 2510 if (ext4_has_metadata_csum(dir->i_sb))
05019a9e
TM		 2511 csum_size = sizeof(struct ext4_dir_entry_tail);
2512
2513 BUFFER_TRACE(bh, "get_write_access");
2514 err = ext4_journal_get_write_access(handle, bh);
2515 if (unlikely(err))
2516 goto out;
2517
2fe34d29 2518 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
05019a9e
TM		 2519 dir->i_sb->s_blocksize, csum_size);
2520 if (err)
2521 goto out;
2522
2523 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
f036adb3 2524 err = ext4_handle_dirty_dirblock(handle, dir, bh);
05019a9e
TM		 2525 if (unlikely(err))
2526 goto out;
2527
2528 return 0;
2529out:
2530 if (err != -ENOENT)
2531 ext4_std_error(dir->i_sb, err);
2532 return err;
2533}
2534
f8628a14 2535/*
c7414892
AD		 2536 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2537 * since this indicates that nlinks count was previously 1 to avoid overflowing
2538 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2539 * that subdirectory link counts are not being maintained accurately.
2540 *
2541 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2542 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2543 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2544 * on regular files) and to avoid creating huge/slow non-HTREE directories.
f8628a14 2545 */
15ed2851 2546static void ext4_inc_count(struct inode *inode)
f8628a14
AD		 2547{
2548 inc_nlink(inode);
c7414892
AD		 2549 if (is_dx(inode) &&
2550 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2551 set_nlink(inode, 1);
f8628a14
AD		 2552}
2553
2554/*
2555 * If a directory had nlink == 1, then we should let it be 1. This indicates
2556 * directory has >EXT4_LINK_MAX subdirs.
2557 */
15ed2851 2558static void ext4_dec_count(struct inode *inode)
f8628a14 2559{
909a4cf1
AD		 2560 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2561 drop_nlink(inode);
f8628a14
AD		 2562}
2563
2564
9b88f9fb
JK		 2565/*
2566 * Add non-directory inode to a directory. On success, the inode reference is
2567 * consumed by dentry is instantiation. This is also indicated by clearing of
2568 * *inodep pointer. On failure, the caller is responsible for dropping the
2569 * inode reference in the safe context.
2570 */
617ba13b 2571static int ext4_add_nondir(handle_t *handle,
9b88f9fb 2572 struct dentry *dentry, struct inode **inodep)
ac27a0ec 2573{
a9e26328 2574 struct inode *dir = d_inode(dentry->d_parent);
9b88f9fb 2575 struct inode *inode = *inodep;
617ba13b 2576 int err = ext4_add_entry(handle, dentry, inode);
ac27a0ec 2577 if (!err) {
4209ae12 2578 err = ext4_mark_inode_dirty(handle, inode);
a9e26328
JK		 2579 if (IS_DIRSYNC(dir))
2580 ext4_handle_sync(handle);
1e2e547a 2581 d_instantiate_new(dentry, inode);
9b88f9fb 2582 *inodep = NULL;
4209ae12 2583 return err;
ac27a0ec 2584 }
731b9a54 2585 drop_nlink(inode);
9b88f9fb 2586 ext4_orphan_add(handle, inode);
6b38e842 2587 unlock_new_inode(inode);
ac27a0ec
DK		 2588 return err;
2589}
2590
2591/*
2592 * By the time this is called, we already have created
2593 * the directory cache entry for the new file, but it
2594 * is so far negative - it has no inode.
2595 *
2596 * If the create succeeds, we fill in the inode information
2597 * with d_instantiate().
2598 */
4acdaf27 2599static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
ebfc3b49 2600 bool excl)
ac27a0ec
DK		 2601{
2602 handle_t *handle;
a80f7fcf 2603 struct inode *inode;
1139575a 2604 int err, credits, retries = 0;
ac27a0ec 2605
a7cdadee
JK		 2606 err = dquot_initialize(dir);
2607 if (err)
2608 return err;
907f4554 2609
1139575a 2610 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2611 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2612retry:
1139575a
TT		 2613 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2614 NULL, EXT4_HT_DIR, credits);
2615 handle = ext4_journal_current_handle();
ac27a0ec
DK		 2616 err = PTR_ERR(inode);
2617 if (!IS_ERR(inode)) {
617ba13b 2618 inode->i_op = &ext4_file_inode_operations;
be64f884 2619 inode->i_fop = &ext4_file_operations;
617ba13b 2620 ext4_set_aops(inode);
9b88f9fb 2621 err = ext4_add_nondir(handle, dentry, &inode);
a80f7fcf
HS		 2622 if (!err)
2623 ext4_fc_track_create(handle, dentry);
ac27a0ec 2624 }
1139575a
TT		 2625 if (handle)
2626 ext4_journal_stop(handle);
9b88f9fb
JK		 2627 if (!IS_ERR_OR_NULL(inode))
2628 iput(inode);
617ba13b 2629 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK		 2630 goto retry;
2631 return err;
2632}
2633
af5bc92d 2634static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 2635 umode_t mode, dev_t rdev)
ac27a0ec
DK		 2636{
2637 handle_t *handle;
a80f7fcf 2638 struct inode *inode;
1139575a 2639 int err, credits, retries = 0;
ac27a0ec 2640
a7cdadee
JK		 2641 err = dquot_initialize(dir);
2642 if (err)
2643 return err;
907f4554 2644
1139575a 2645 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2646 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2647retry:
1139575a
TT		 2648 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2649 NULL, EXT4_HT_DIR, credits);
2650 handle = ext4_journal_current_handle();
ac27a0ec
DK		 2651 err = PTR_ERR(inode);
2652 if (!IS_ERR(inode)) {
2653 init_special_inode(inode, inode->i_mode, rdev);
617ba13b 2654 inode->i_op = &ext4_special_inode_operations;
9b88f9fb 2655 err = ext4_add_nondir(handle, dentry, &inode);
aa75f4d3 2656 if (!err)
a80f7fcf 2657 ext4_fc_track_create(handle, dentry);
ac27a0ec 2658 }
1139575a
TT		 2659 if (handle)
2660 ext4_journal_stop(handle);
9b88f9fb
JK		 2661 if (!IS_ERR_OR_NULL(inode))
2662 iput(inode);
617ba13b 2663 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK		 2664 goto retry;
2665 return err;
2666}
2667
af51a2ac
AV		 2668static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2669{
2670 handle_t *handle;
2671 struct inode *inode;
2672 int err, retries = 0;
2673
a7cdadee
JK		 2674 err = dquot_initialize(dir);
2675 if (err)
2676 return err;
af51a2ac
AV		 2677
2678retry:
2679 inode = ext4_new_inode_start_handle(dir, mode,
2680 NULL, 0, NULL,
2681 EXT4_HT_DIR,
2682 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2683 4 + EXT4_XATTR_TRANS_BLOCKS);
2684 handle = ext4_journal_current_handle();
2685 err = PTR_ERR(inode);
2686 if (!IS_ERR(inode)) {
2687 inode->i_op = &ext4_file_inode_operations;
be64f884 2688 inode->i_fop = &ext4_file_operations;
af51a2ac 2689 ext4_set_aops(inode);
e94bd349 2690 d_tmpfile(dentry, inode);
af51a2ac
AV		 2691 err = ext4_orphan_add(handle, inode);
2692 if (err)
43ae9e3f 2693 goto err_unlock_inode;
af51a2ac 2694 mark_inode_dirty(inode);
af51a2ac
AV		 2695 unlock_new_inode(inode);
2696 }
2697 if (handle)
2698 ext4_journal_stop(handle);
2699 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2700 goto retry;
2701 return err;
43ae9e3f 2702err_unlock_inode:
af51a2ac
AV		 2703 ext4_journal_stop(handle);
2704 unlock_new_inode(inode);
af51a2ac
AV		 2705 return err;
2706}
2707
a774f9c2
TM		 2708struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2709 struct ext4_dir_entry_2 *de,
2710 int blocksize, int csum_size,
2711 unsigned int parent_ino, int dotdot_real_len)
2712{
2713 de->inode = cpu_to_le32(inode->i_ino);
2714 de->name_len = 1;
2715 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2716 blocksize);
2717 strcpy(de->name, ".");
2718 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2719
2720 de = ext4_next_entry(de, blocksize);
2721 de->inode = cpu_to_le32(parent_ino);
2722 de->name_len = 2;
2723 if (!dotdot_real_len)
2724 de->rec_len = ext4_rec_len_to_disk(blocksize -
2725 (csum_size + EXT4_DIR_REC_LEN(1)),
2726 blocksize);
2727 else
2728 de->rec_len = ext4_rec_len_to_disk(
2729 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2730 strcpy(de->name, "..");
2731 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2732
2733 return ext4_next_entry(de, blocksize);
2734}
2735
8016e29f 2736int ext4_init_new_dir(handle_t *handle, struct inode *dir,
a774f9c2 2737 struct inode *inode)
ac27a0ec 2738{
dabd991f 2739 struct buffer_head *dir_block = NULL;
af5bc92d 2740 struct ext4_dir_entry_2 *de;
dc6982ff 2741 ext4_lblk_t block = 0;
3d0518f4 2742 unsigned int blocksize = dir->i_sb->s_blocksize;
b0336e8d 2743 int csum_size = 0;
a774f9c2 2744 int err;
ac27a0ec 2745
9aa5d32b 2746 if (ext4_has_metadata_csum(dir->i_sb))
b0336e8d
DW		 2747 csum_size = sizeof(struct ext4_dir_entry_tail);
2748
3c47d541
TM		 2749 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2750 err = ext4_try_create_inline_dir(handle, dir, inode);
2751 if (err < 0 && err != -ENOSPC)
2752 goto out;
2753 if (!err)
2754 goto out;
2755 }
2756
dc6982ff 2757 inode->i_size = 0;
0f70b406
TT		 2758 dir_block = ext4_append(handle, inode, &block);
2759 if (IS_ERR(dir_block))
2760 return PTR_ERR(dir_block);
a774f9c2
TM		 2761 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2762 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2763 set_nlink(inode, 2);
ddce3b94
TT		 2764 if (csum_size)
2765 ext4_initialize_dirent_tail(dir_block, blocksize);
a774f9c2
TM		 2766
2767 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
f036adb3 2768 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
a774f9c2
TM		 2769 if (err)
2770 goto out;
2771 set_buffer_verified(dir_block);
2772out:
2773 brelse(dir_block);
2774 return err;
2775}
2776
2777static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2778{
2779 handle_t *handle;
2780 struct inode *inode;
4209ae12 2781 int err, err2 = 0, credits, retries = 0;
a774f9c2 2782
f8628a14 2783 if (EXT4_DIR_LINK_MAX(dir))
ac27a0ec
DK		 2784 return -EMLINK;
2785
a7cdadee
JK		 2786 err = dquot_initialize(dir);
2787 if (err)
2788 return err;
907f4554 2789
1139575a 2790 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2791 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2792retry:
1139575a
TT		 2793 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2794 &dentry->d_name,
2795 0, NULL, EXT4_HT_DIR, credits);
2796 handle = ext4_journal_current_handle();
ac27a0ec
DK		 2797 err = PTR_ERR(inode);
2798 if (IS_ERR(inode))
2799 goto out_stop;
2800
617ba13b
MC		 2801 inode->i_op = &ext4_dir_inode_operations;
2802 inode->i_fop = &ext4_dir_operations;
a774f9c2 2803 err = ext4_init_new_dir(handle, dir, inode);
dabd991f
NK		 2804 if (err)
2805 goto out_clear_inode;
2806 err = ext4_mark_inode_dirty(handle, inode);
2807 if (!err)
2808 err = ext4_add_entry(handle, dentry, inode);
ac27a0ec 2809 if (err) {
4cdeed86
AK		 2810out_clear_inode:
2811 clear_nlink(inode);
9b88f9fb 2812 ext4_orphan_add(handle, inode);
6b38e842 2813 unlock_new_inode(inode);
4209ae12
HS		 2814 err2 = ext4_mark_inode_dirty(handle, inode);
2815 if (unlikely(err2))
2816 err = err2;
9b88f9fb 2817 ext4_journal_stop(handle);
af5bc92d 2818 iput(inode);
9b88f9fb 2819 goto out_retry;
ac27a0ec 2820 }
15ed2851 2821 ext4_inc_count(dir);
aa75f4d3 2822
617ba13b 2823 ext4_update_dx_flag(dir);
dabd991f
NK		 2824 err = ext4_mark_inode_dirty(handle, dir);
2825 if (err)
2826 goto out_clear_inode;
1e2e547a 2827 d_instantiate_new(dentry, inode);
a80f7fcf 2828 ext4_fc_track_create(handle, dentry);
1139575a
TT		 2829 if (IS_DIRSYNC(dir))
2830 ext4_handle_sync(handle);
2831
ac27a0ec 2832out_stop:
1139575a
TT		 2833 if (handle)
2834 ext4_journal_stop(handle);
9b88f9fb 2835out_retry:
617ba13b 2836 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK		 2837 goto retry;
2838 return err;
2839}
2840
2841/*
2842 * routine to check that the specified directory is empty (for rmdir)
2843 */
a7550b30 2844bool ext4_empty_dir(struct inode *inode)
ac27a0ec 2845{
498e5f24 2846 unsigned int offset;
af5bc92d 2847 struct buffer_head *bh;
64d4ce89 2848 struct ext4_dir_entry_2 *de;
af5bc92d 2849 struct super_block *sb;
ac27a0ec 2850
61f86638
TM		 2851 if (ext4_has_inline_data(inode)) {
2852 int has_inline_data = 1;
a7550b30 2853 int ret;
61f86638 2854
a7550b30 2855 ret = empty_inline_dir(inode, &has_inline_data);
61f86638 2856 if (has_inline_data)
a7550b30 2857 return ret;
61f86638
TM		 2858 }
2859
ac27a0ec 2860 sb = inode->i_sb;
dc6982ff
TT		 2861 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2862 EXT4_ERROR_INODE(inode, "invalid size");
a7550b30 2863 return true;
ac27a0ec 2864 }
4e19d6b6
TT		 2865 /* The first directory block must not be a hole,
2866 * so treat it as DIRENT_HTREE
2867 */
2868 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
dc6982ff 2869 if (IS_ERR(bh))
a7550b30 2870 return true;
dc6982ff 2871
617ba13b 2872 de = (struct ext4_dir_entry_2 *) bh->b_data;
64d4ce89
JK		 2873 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2874 0) ||
2875 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2876 ext4_warning_inode(inode, "directory missing '.'");
2877 brelse(bh);
2878 return true;
2879 }
2880 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2881 de = ext4_next_entry(de, sb->s_blocksize);
2882 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2883 offset) ||
2884 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2885 ext4_warning_inode(inode, "directory missing '..'");
af5bc92d 2886 brelse(bh);
a7550b30 2887 return true;
ac27a0ec 2888 }
64d4ce89 2889 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
af5bc92d 2890 while (offset < inode->i_size) {
64d4ce89 2891 if (!(offset & (sb->s_blocksize - 1))) {
24676da4 2892 unsigned int lblock;
af5bc92d 2893 brelse(bh);
24676da4 2894 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
dc6982ff 2895 bh = ext4_read_dirblock(inode, lblock, EITHER);
4e19d6b6
TT		 2896 if (bh == NULL) {
2897 offset += sb->s_blocksize;
2898 continue;
2899 }
dc6982ff 2900 if (IS_ERR(bh))
a7550b30 2901 return true;
ac27a0ec 2902 }
64d4ce89
JK		 2903 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2904 (offset & (sb->s_blocksize - 1)));
226ba972
TM		 2905 if (ext4_check_dir_entry(inode, NULL, de, bh,
2906 bh->b_data, bh->b_size, offset)) {
ac27a0ec
DK		 2907 offset = (offset | (sb->s_blocksize - 1)) + 1;
2908 continue;
2909 }
2910 if (le32_to_cpu(de->inode)) {
af5bc92d 2911 brelse(bh);
a7550b30 2912 return false;
ac27a0ec 2913 }
3d0518f4 2914 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
ac27a0ec 2915 }
af5bc92d 2916 brelse(bh);
a7550b30 2917 return true;
ac27a0ec
DK		 2918}
2919
d745a8c2
JK		 2920/*
2921 * ext4_orphan_add() links an unlinked or truncated inode into a list of
ac27a0ec
DK		 2922 * such inodes, starting at the superblock, in case we crash before the
2923 * file is closed/deleted, or in case the inode truncate spans multiple
2924 * transactions and the last transaction is not recovered after a crash.
2925 *
2926 * At filesystem recovery time, we walk this list deleting unlinked
617ba13b 2927 * inodes and truncating linked inodes in ext4_orphan_cleanup().
d745a8c2
JK		 2928 *
2929 * Orphan list manipulation functions must be called under i_mutex unless
2930 * we are just creating the inode or deleting it.
ac27a0ec 2931 */
617ba13b 2932int ext4_orphan_add(handle_t *handle, struct inode *inode)
ac27a0ec
DK		 2933{
2934 struct super_block *sb = inode->i_sb;
cd2c080c 2935 struct ext4_sb_info *sbi = EXT4_SB(sb);
617ba13b 2936 struct ext4_iloc iloc;
ac27a0ec 2937 int err = 0, rc;
d745a8c2 2938 bool dirty = false;
ac27a0ec 2939
e2bfb088 2940 if (!sbi->s_journal || is_bad_inode(inode))
0390131b
FM		 2941 return 0;
2942
d745a8c2 2943 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
5955102c 2944 !inode_is_locked(inode));
d745a8c2
JK		 2945 /*
2946 * Exit early if inode already is on orphan list. This is a big speedup
2947 * since we don't have to contend on the global s_orphan_lock.
2948 */
617ba13b 2949 if (!list_empty(&EXT4_I(inode)->i_orphan))
d745a8c2 2950 return 0;
ac27a0ec 2951
afb86178
LC		 2952 /*
2953 * Orphan handling is only valid for files with data blocks
2954 * being truncated, or files being unlinked. Note that we either
2955 * hold i_mutex, or the inode can not be referenced from outside,
2956 * so i_nlink should not be bumped due to race
ac27a0ec 2957 */
837c23fb 2958 ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
af5bc92d 2959 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
ac27a0ec 2960
cd2c080c
JK		 2961 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2962 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
ac27a0ec 2963 if (err)
d745a8c2 2964 goto out;
ac27a0ec 2965
617ba13b 2966 err = ext4_reserve_inode_write(handle, inode, &iloc);
ac27a0ec 2967 if (err)
d745a8c2
JK		 2968 goto out;
2969
2970 mutex_lock(&sbi->s_orphan_lock);
6e3617e5
DM		 2971 /*
2972 * Due to previous errors inode may be already a part of on-disk
2973 * orphan list. If so skip on-disk list modification.
2974 */
d745a8c2
JK		 2975 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2976 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2977 /* Insert this inode at the head of the on-disk orphan list */
2978 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
05c2c00f 2979 lock_buffer(sbi->s_sbh);
d745a8c2 2980 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
05c2c00f
JK		 2981 ext4_superblock_csum_set(sb);
2982 unlock_buffer(sbi->s_sbh);
d745a8c2
JK		 2983 dirty = true;
2984 }
2985 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2986 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 2987
d745a8c2 2988 if (dirty) {
a3f5cf14 2989 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
d745a8c2
JK		 2990 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2991 if (!err)
2992 err = rc;
2993 if (err) {
2994 /*
2995 * We have to remove inode from in-memory list if
2996 * addition to on disk orphan list failed. Stray orphan
2997 * list entries can cause panics at unmount time.
2998 */
2999 mutex_lock(&sbi->s_orphan_lock);
74177f55 3000 list_del_init(&EXT4_I(inode)->i_orphan);
d745a8c2
JK		 3001 mutex_unlock(&sbi->s_orphan_lock);
3002 }
feaf264c
VA		 3003 } else
3004 brelse(iloc.bh);
3005
ac27a0ec
DK		 3006 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3007 jbd_debug(4, "orphan inode %lu will point to %d\n",
3008 inode->i_ino, NEXT_ORPHAN(inode));
d745a8c2 3009out:
cd2c080c 3010 ext4_std_error(sb, err);
ac27a0ec
DK		 3011 return err;
3012}
3013
3014/*
617ba13b 3015 * ext4_orphan_del() removes an unlinked or truncated inode from the list
ac27a0ec
DK		 3016 * of such inodes stored on disk, because it is finally being cleaned up.
3017 */
617ba13b 3018int ext4_orphan_del(handle_t *handle, struct inode *inode)
ac27a0ec
DK		 3019{
3020 struct list_head *prev;
617ba13b 3021 struct ext4_inode_info *ei = EXT4_I(inode);
cd2c080c 3022 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
498e5f24 3023 __u32 ino_next;
617ba13b 3024 struct ext4_iloc iloc;
ac27a0ec
DK		 3025 int err = 0;
3026
cd2c080c 3027 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
0390131b
FM		 3028 return 0;
3029
d745a8c2 3030 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
5955102c 3031 !inode_is_locked(inode));
d745a8c2 3032 /* Do this quick check before taking global s_orphan_lock. */
3b9d4ed2 3033 if (list_empty(&ei->i_orphan))
d745a8c2 3034 return 0;
ac27a0ec 3035
d745a8c2
JK		 3036 if (handle) {
3037 /* Grab inode buffer early before taking global s_orphan_lock */
3038 err = ext4_reserve_inode_write(handle, inode, &iloc);
3039 }
ac27a0ec 3040
d745a8c2 3041 mutex_lock(&sbi->s_orphan_lock);
ac27a0ec
DK		 3042 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3043
d745a8c2 3044 prev = ei->i_orphan.prev;
ac27a0ec
DK		 3045 list_del_init(&ei->i_orphan);
3046
3047 /* If we're on an error path, we may not have a valid
3048 * transaction handle with which to update the orphan list on
3049 * disk, but we still need to remove the inode from the linked
3050 * list in memory. */
d745a8c2
JK		 3051 if (!handle || err) {
3052 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 3053 goto out_err;
d745a8c2 3054 }
ac27a0ec 3055
d745a8c2 3056 ino_next = NEXT_ORPHAN(inode);
ac27a0ec 3057 if (prev == &sbi->s_orphan) {
498e5f24 3058 jbd_debug(4, "superblock will point to %u\n", ino_next);
ac27a0ec 3059 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
617ba13b 3060 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
d745a8c2
JK		 3061 if (err) {
3062 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 3063 goto out_brelse;
d745a8c2 3064 }
05c2c00f 3065 lock_buffer(sbi->s_sbh);
ac27a0ec 3066 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
05c2c00f
JK		 3067 ext4_superblock_csum_set(inode->i_sb);
3068 unlock_buffer(sbi->s_sbh);
d745a8c2 3069 mutex_unlock(&sbi->s_orphan_lock);
a3f5cf14 3070 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
ac27a0ec 3071 } else {
617ba13b 3072 struct ext4_iloc iloc2;
ac27a0ec 3073 struct inode *i_prev =
617ba13b 3074 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
ac27a0ec 3075
498e5f24 3076 jbd_debug(4, "orphan inode %lu will point to %u\n",
ac27a0ec 3077 i_prev->i_ino, ino_next);
617ba13b 3078 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
d745a8c2
JK		 3079 if (err) {
3080 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 3081 goto out_brelse;
d745a8c2 3082 }
ac27a0ec 3083 NEXT_ORPHAN(i_prev) = ino_next;
617ba13b 3084 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
d745a8c2 3085 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec
DK		 3086 }
3087 if (err)
3088 goto out_brelse;
3089 NEXT_ORPHAN(inode) = 0;
617ba13b 3090 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
ac27a0ec 3091out_err:
617ba13b 3092 ext4_std_error(inode->i_sb, err);
ac27a0ec
DK		 3093 return err;
3094
3095out_brelse:
3096 brelse(iloc.bh);
3097 goto out_err;
3098}
3099
af5bc92d 3100static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
ac27a0ec
DK		 3101{
3102 int retval;
af5bc92d
TT		 3103 struct inode *inode;
3104 struct buffer_head *bh;
3105 struct ext4_dir_entry_2 *de;
8dcfaad2 3106 handle_t *handle = NULL;
ac27a0ec 3107
0db1ff22
TT		 3108 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3109 return -EIO;
3110
ac27a0ec
DK		 3111 /* Initialize quotas before so that eventual writes go in
3112 * separate transaction */
a7cdadee
JK		 3113 retval = dquot_initialize(dir);
3114 if (retval)
3115 return retval;
3116 retval = dquot_initialize(d_inode(dentry));
3117 if (retval)
3118 return retval;
907f4554 3119
ac27a0ec 3120 retval = -ENOENT;
32f7f22c 3121 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
36de9286
TT		 3122 if (IS_ERR(bh))
3123 return PTR_ERR(bh);
ac27a0ec
DK		 3124 if (!bh)
3125 goto end_rmdir;
3126
2b0143b5 3127 inode = d_inode(dentry);
ac27a0ec 3128
6a797d27 3129 retval = -EFSCORRUPTED;
ac27a0ec
DK		 3130 if (le32_to_cpu(de->inode) != inode->i_ino)
3131 goto end_rmdir;
3132
3133 retval = -ENOTEMPTY;
e875a2dd 3134 if (!ext4_empty_dir(inode))
ac27a0ec
DK		 3135 goto end_rmdir;
3136
8dcfaad2 3137 handle = ext4_journal_start(dir, EXT4_HT_DIR,
64044abf 3138 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
8dcfaad2
TT		 3139 if (IS_ERR(handle)) {
3140 retval = PTR_ERR(handle);
3141 handle = NULL;
3142 goto end_rmdir;
3143 }
3144
3145 if (IS_DIRSYNC(dir))
3146 ext4_handle_sync(handle);
3147
617ba13b 3148 retval = ext4_delete_entry(handle, dir, de, bh);
ac27a0ec
DK		 3149 if (retval)
3150 goto end_rmdir;
f8628a14 3151 if (!EXT4_DIR_LINK_EMPTY(inode))
b03a2f7e
AD		 3152 ext4_warning_inode(inode,
3153 "empty directory '%.*s' has too many links (%u)",
3154 dentry->d_name.len, dentry->d_name.name,
af5bc92d 3155 inode->i_nlink);
ee73f9a5 3156 inode_inc_iversion(inode);
ac27a0ec
DK		 3157 clear_nlink(inode);
3158 /* There's no need to set i_disksize: the fact that i_nlink is
3159 * zero will ensure that the right thing happens during any
3160 * recovery. */
3161 inode->i_size = 0;
617ba13b 3162 ext4_orphan_add(handle, inode);
eeca7ea1 3163 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
4209ae12
HS		 3164 retval = ext4_mark_inode_dirty(handle, inode);
3165 if (retval)
3166 goto end_rmdir;
15ed2851 3167 ext4_dec_count(dir);
617ba13b 3168 ext4_update_dx_flag(dir);
a80f7fcf 3169 ext4_fc_track_unlink(handle, dentry);
4209ae12 3170 retval = ext4_mark_inode_dirty(handle, dir);
ac27a0ec 3171
b886ee3e
GKB		 3172#ifdef CONFIG_UNICODE
3173 /* VFS negative dentries are incompatible with Encoding and
3174 * Case-insensitiveness. Eventually we'll want avoid
3175 * invalidating the dentries here, alongside with returning the
3176 * negative dentries at ext4_lookup(), when it is better
3177 * supported by the VFS for the CI case.
3178 */
3179 if (IS_CASEFOLDED(dir))
3180 d_invalidate(dentry);
3181#endif
3182
ac27a0ec 3183end_rmdir:
af5bc92d 3184 brelse(bh);
8dcfaad2
TT		 3185 if (handle)
3186 ext4_journal_stop(handle);
ac27a0ec
DK		 3187 return retval;
3188}
3189
a80f7fcf 3190int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
8016e29f 3191 struct inode *inode)
ac27a0ec 3192{
8016e29f 3193 int retval = -ENOENT;
af5bc92d
TT		 3194 struct buffer_head *bh;
3195 struct ext4_dir_entry_2 *de;
8016e29f 3196 int skip_remove_dentry = 0;
ac27a0ec 3197
8016e29f
HS		 3198 bh = ext4_find_entry(dir, d_name, &de, NULL);
3199 if (IS_ERR(bh))
3200 return PTR_ERR(bh);
ac27a0ec 3201
8016e29f
HS		 3202 if (!bh)
3203 return -ENOENT;
ac27a0ec 3204
e5f78159 3205 if (le32_to_cpu(de->inode) != inode->i_ino) {
8016e29f
HS		 3206 /*
3207 * It's okay if we find dont find dentry which matches
3208 * the inode. That's because it might have gotten
3209 * renamed to a different inode number
3210 */
3211 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3212 skip_remove_dentry = 1;
3213 else
a80f7fcf 3214 goto out;
931b6864
TT		 3215 }
3216
3217 if (IS_DIRSYNC(dir))
3218 ext4_handle_sync(handle);
3219
8016e29f
HS		 3220 if (!skip_remove_dentry) {
3221 retval = ext4_delete_entry(handle, dir, de, bh);
3222 if (retval)
a80f7fcf 3223 goto out;
8016e29f
HS		 3224 dir->i_ctime = dir->i_mtime = current_time(dir);
3225 ext4_update_dx_flag(dir);
3226 retval = ext4_mark_inode_dirty(handle, dir);
3227 if (retval)
a80f7fcf 3228 goto out;
8016e29f
HS		 3229 } else {
3230 retval = 0;
3231 }
c7df4a1e
TT		 3232 if (inode->i_nlink == 0)
3233 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
8016e29f 3234 d_name->len, d_name->name);
c7df4a1e
TT		 3235 else
3236 drop_nlink(inode);
ac27a0ec 3237 if (!inode->i_nlink)
617ba13b 3238 ext4_orphan_add(handle, inode);
eeca7ea1 3239 inode->i_ctime = current_time(inode);
4209ae12 3240 retval = ext4_mark_inode_dirty(handle, inode);
ac27a0ec 3241
a80f7fcf 3242out:
8016e29f
HS		 3243 brelse(bh);
3244 return retval;
3245}
3246
3247static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3248{
a80f7fcf 3249 handle_t *handle;
8016e29f
HS		 3250 int retval;
3251
3252 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3253 return -EIO;
3254
3255 trace_ext4_unlink_enter(dir, dentry);
3256 /*
3257 * Initialize quotas before so that eventual writes go
3258 * in separate transaction
3259 */
3260 retval = dquot_initialize(dir);
3261 if (retval)
3262 goto out_trace;
3263 retval = dquot_initialize(d_inode(dentry));
3264 if (retval)
3265 goto out_trace;
3266
a80f7fcf
HS		 3267 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3268 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3269 if (IS_ERR(handle)) {
3270 retval = PTR_ERR(handle);
3271 goto out_trace;
3272 }
3273
3274 retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
aa75f4d3 3275 if (!retval)
a80f7fcf 3276 ext4_fc_track_unlink(handle, dentry);
b886ee3e
GKB		 3277#ifdef CONFIG_UNICODE
3278 /* VFS negative dentries are incompatible with Encoding and
3279 * Case-insensitiveness. Eventually we'll want avoid
3280 * invalidating the dentries here, alongside with returning the
3281 * negative dentries at ext4_lookup(), when it is better
3282 * supported by the VFS for the CI case.
3283 */
3284 if (IS_CASEFOLDED(dir))
3285 d_invalidate(dentry);
3286#endif
a80f7fcf
HS		 3287 if (handle)
3288 ext4_journal_stop(handle);
b886ee3e 3289
e5f78159 3290out_trace:
0562e0ba 3291 trace_ext4_unlink_exit(dentry, retval);
ac27a0ec
DK		 3292 return retval;
3293}
3294
af5bc92d
TT		 3295static int ext4_symlink(struct inode *dir,
3296 struct dentry *dentry, const char *symname)
ac27a0ec
DK		 3297{
3298 handle_t *handle;
af5bc92d 3299 struct inode *inode;
f348c252 3300 int err, len = strlen(symname);
df5e6223 3301 int credits;
a7550b30 3302 struct fscrypt_str disk_link;
ac27a0ec 3303
0db1ff22
TT		 3304 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3305 return -EIO;
3306
78e1060c
EB		 3307 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3308 &disk_link);
3309 if (err)
3310 return err;
ac27a0ec 3311
a7cdadee
JK		 3312 err = dquot_initialize(dir);
3313 if (err)
78e1060c 3314 return err;
907f4554 3315
f348c252 3316 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
df5e6223
JK		 3317 /*
3318 * For non-fast symlinks, we just allocate inode and put it on
3319 * orphan list in the first transaction => we need bitmap,
8c208719
ES		 3320 * group descriptor, sb, inode block, quota blocks, and
3321 * possibly selinux xattr blocks.
df5e6223 3322 */
8c208719
ES		 3323 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3324 EXT4_XATTR_TRANS_BLOCKS;
df5e6223
JK		 3325 } else {
3326 /*
3327 * Fast symlink. We have to add entry to directory
3328 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3329 * allocate new inode (bitmap, group descriptor, inode block,
3330 * quota blocks, sb is already counted in previous macros).
3331 */
3332 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 3333 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
df5e6223 3334 }
f348c252 3335
1139575a
TT		 3336 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3337 &dentry->d_name, 0, NULL,
3338 EXT4_HT_DIR, credits);
3339 handle = ext4_journal_current_handle();
f348c252
TT		 3340 if (IS_ERR(inode)) {
3341 if (handle)
3342 ext4_journal_stop(handle);
78e1060c 3343 return PTR_ERR(inode);
f348c252
TT		 3344 }
3345
78e1060c
EB		 3346 if (IS_ENCRYPTED(inode)) {
3347 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
ef1eb3aa 3348 if (err)
f348c252 3349 goto err_drop_inode;
a7a67e8a 3350 inode->i_op = &ext4_encrypted_symlink_inode_operations;
f348c252 3351 }
ac27a0ec 3352
f348c252 3353 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
78e1060c 3354 if (!IS_ENCRYPTED(inode))
a7a67e8a 3355 inode->i_op = &ext4_symlink_inode_operations;
21fc61c7 3356 inode_nohighmem(inode);
617ba13b 3357 ext4_set_aops(inode);
ac27a0ec 3358 /*
df5e6223
JK		 3359 * We cannot call page_symlink() with transaction started
3360 * because it calls into ext4_write_begin() which can wait
3361 * for transaction commit if we are running out of space
3362 * and thus we deadlock. So we have to stop transaction now
3363 * and restart it when symlink contents is written.
3364 *
3365 * To keep fs consistent in case of crash, we have to put inode
3366 * to orphan list in the mean time.
ac27a0ec 3367 */
df5e6223
JK		 3368 drop_nlink(inode);
3369 err = ext4_orphan_add(handle, inode);
8016e29f
HS		 3370 if (handle)
3371 ext4_journal_stop(handle);
f348c252 3372 handle = NULL;
df5e6223
JK		 3373 if (err)
3374 goto err_drop_inode;
f348c252 3375 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
df5e6223
JK		 3376 if (err)
3377 goto err_drop_inode;
3378 /*
3379 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3380 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3381 */
9924a92a 3382 handle = ext4_journal_start(dir, EXT4_HT_DIR,
df5e6223
JK		 3383 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3384 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3385 if (IS_ERR(handle)) {
3386 err = PTR_ERR(handle);
f348c252 3387 handle = NULL;
df5e6223
JK		 3388 goto err_drop_inode;
3389 }
0ce8c010 3390 set_nlink(inode, 1);
df5e6223 3391 err = ext4_orphan_del(handle, inode);
f348c252 3392 if (err)
df5e6223 3393 goto err_drop_inode;
ac27a0ec 3394 } else {
e65187e6 3395 /* clear the extent format for fast symlink */
12e9b892 3396 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
78e1060c 3397 if (!IS_ENCRYPTED(inode)) {
a7a67e8a 3398 inode->i_op = &ext4_fast_symlink_inode_operations;
75e7566b
AV		 3399 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3400 }
f348c252
TT		 3401 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3402 disk_link.len);
3403 inode->i_size = disk_link.len - 1;
ac27a0ec 3404 }
617ba13b 3405 EXT4_I(inode)->i_disksize = inode->i_size;
9b88f9fb 3406 err = ext4_add_nondir(handle, dentry, &inode);
1139575a
TT		 3407 if (handle)
3408 ext4_journal_stop(handle);
9b88f9fb
JK		 3409 if (inode)
3410 iput(inode);
78e1060c
EB		 3411 goto out_free_encrypted_link;
3412
df5e6223 3413err_drop_inode:
f348c252
TT		 3414 if (handle)
3415 ext4_journal_stop(handle);
f348c252 3416 clear_nlink(inode);
df5e6223
JK		 3417 unlock_new_inode(inode);
3418 iput(inode);
78e1060c
EB		 3419out_free_encrypted_link:
3420 if (disk_link.name != (unsigned char *)symname)
3421 kfree(disk_link.name);
df5e6223 3422 return err;
ac27a0ec
DK		 3423}
3424
8016e29f 3425int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
ac27a0ec
DK		 3426{
3427 handle_t *handle;
ac27a0ec 3428 int err, retries = 0;
ac27a0ec 3429retry:
9924a92a
TT		 3430 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3431 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
af51a2ac 3432 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
ac27a0ec
DK		 3433 if (IS_ERR(handle))
3434 return PTR_ERR(handle);
3435
3436 if (IS_DIRSYNC(dir))
0390131b 3437 ext4_handle_sync(handle);
ac27a0ec 3438
eeca7ea1 3439 inode->i_ctime = current_time(inode);
15ed2851 3440 ext4_inc_count(inode);
7de9c6ee 3441 ihold(inode);
ac27a0ec 3442
6b38e842
AV		 3443 err = ext4_add_entry(handle, dentry, inode);
3444 if (!err) {
4209ae12 3445 err = ext4_mark_inode_dirty(handle, inode);
af51a2ac
AV		 3446 /* this can happen only for tmpfile being
3447 * linked the first time
3448 */
3449 if (inode->i_nlink == 1)
3450 ext4_orphan_del(handle, inode);
6b38e842 3451 d_instantiate(dentry, inode);
a80f7fcf 3452 ext4_fc_track_link(handle, dentry);
6b38e842
AV		 3453 } else {
3454 drop_nlink(inode);
3455 iput(inode);
3456 }
617ba13b
MC		 3457 ext4_journal_stop(handle);
3458 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK		 3459 goto retry;
3460 return err;
3461}
3462
8016e29f
HS		 3463static int ext4_link(struct dentry *old_dentry,
3464 struct inode *dir, struct dentry *dentry)
3465{
3466 struct inode *inode = d_inode(old_dentry);
3467 int err;
3468
3469 if (inode->i_nlink >= EXT4_LINK_MAX)
3470 return -EMLINK;
3471
3472 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3473 if (err)
3474 return err;
3475
3476 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3477 (!projid_eq(EXT4_I(dir)->i_projid,
3478 EXT4_I(old_dentry->d_inode)->i_projid)))
3479 return -EXDEV;
3480
3481 err = dquot_initialize(dir);
3482 if (err)
3483 return err;
3484 return __ext4_link(dir, inode, dentry);
3485}
32f7f22c
TM		 3486
3487/*
3488 * Try to find buffer head where contains the parent block.
3489 * It should be the inode block if it is inlined or the 1st block
3490 * if it is a normal dir.
3491 */
3492static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3493 struct inode *inode,
3494 int *retval,
3495 struct ext4_dir_entry_2 **parent_de,
3496 int *inlined)
3497{
3498 struct buffer_head *bh;
3499
3500 if (!ext4_has_inline_data(inode)) {
4e19d6b6
TT		 3501 /* The first directory block must not be a hole, so
3502 * treat it as DIRENT_HTREE
3503 */
3504 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
dc6982ff
TT		 3505 if (IS_ERR(bh)) {
3506 *retval = PTR_ERR(bh);
32f7f22c
TM		 3507 return NULL;
3508 }
3509 *parent_de = ext4_next_entry(
3510 (struct ext4_dir_entry_2 *)bh->b_data,
3511 inode->i_sb->s_blocksize);
3512 return bh;
3513 }
3514
3515 *inlined = 1;
3516 return ext4_get_first_inline_block(inode, parent_de, retval);
3517}
ac27a0ec 3518
c0d268c3
MS		 3519struct ext4_renament {
3520 struct inode *dir;
3521 struct dentry *dentry;
3522 struct inode *inode;
bd42998a
MS		 3523 bool is_dir;
3524 int dir_nlink_delta;
c0d268c3
MS		 3525
3526 /* entry for "dentry" */
3527 struct buffer_head *bh;
3528 struct ext4_dir_entry_2 *de;
3529 int inlined;
3530
3531 /* entry for ".." in inode if it's a directory */
3532 struct buffer_head *dir_bh;
3533 struct ext4_dir_entry_2 *parent_de;
3534 int dir_inlined;
3535};
3536
bd1af145
MS		 3537static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3538{
3539 int retval;
3540
3541 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3542 &retval, &ent->parent_de,
3543 &ent->dir_inlined);
3544 if (!ent->dir_bh)
3545 return retval;
3546 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
6a797d27 3547 return -EFSCORRUPTED;
bd1af145
MS		 3548 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3549 return ext4_journal_get_write_access(handle, ent->dir_bh);
3550}
3551
3552static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3553 unsigned dir_ino)
3554{
3555 int retval;
3556
3557 ent->parent_de->inode = cpu_to_le32(dir_ino);
3558 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3559 if (!ent->dir_inlined) {
3560 if (is_dx(ent->inode)) {
3561 retval = ext4_handle_dirty_dx_node(handle,
3562 ent->inode,
3563 ent->dir_bh);
3564 } else {
f036adb3
TT		 3565 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3566 ent->dir_bh);
bd1af145
MS		 3567 }
3568 } else {
3569 retval = ext4_mark_inode_dirty(handle, ent->inode);
3570 }
3571 if (retval) {
3572 ext4_std_error(ent->dir->i_sb, retval);
3573 return retval;
3574 }
3575 return 0;
3576}
3577
3578static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3579 unsigned ino, unsigned file_type)
3580{
4209ae12 3581 int retval, retval2;
bd1af145
MS		 3582
3583 BUFFER_TRACE(ent->bh, "get write access");
3584 retval = ext4_journal_get_write_access(handle, ent->bh);
3585 if (retval)
3586 return retval;
3587 ent->de->inode = cpu_to_le32(ino);
e2b911c5 3588 if (ext4_has_feature_filetype(ent->dir->i_sb))
bd1af145 3589 ent->de->file_type = file_type;
ee73f9a5 3590 inode_inc_iversion(ent->dir);
bd1af145 3591 ent->dir->i_ctime = ent->dir->i_mtime =
eeca7ea1 3592 current_time(ent->dir);
4209ae12 3593 retval = ext4_mark_inode_dirty(handle, ent->dir);
bd1af145
MS		 3594 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3595 if (!ent->inlined) {
4209ae12
HS		 3596 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3597 if (unlikely(retval2)) {
3598 ext4_std_error(ent->dir->i_sb, retval2);
3599 return retval2;
bd1af145
MS		 3600 }
3601 }
4209ae12 3602 return retval;
bd1af145
MS		 3603}
3604
3605static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3606 const struct qstr *d_name)
3607{
3608 int retval = -ENOENT;
3609 struct buffer_head *bh;
3610 struct ext4_dir_entry_2 *de;
3611
3612 bh = ext4_find_entry(dir, d_name, &de, NULL);
36de9286
TT		 3613 if (IS_ERR(bh))
3614 return PTR_ERR(bh);
bd1af145
MS		 3615 if (bh) {
3616 retval = ext4_delete_entry(handle, dir, de, bh);
3617 brelse(bh);
3618 }
3619 return retval;
3620}
3621
d80d448c
DW		 3622static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3623 int force_reread)
bd1af145
MS		 3624{
3625 int retval;
3626 /*
3627 * ent->de could have moved from under us during htree split, so make
3628 * sure that we are deleting the right entry. We might also be pointing
3629 * to a stale entry in the unused part of ent->bh so just checking inum
3630 * and the name isn't enough.
3631 */
3632 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3633 ent->de->name_len != ent->dentry->d_name.len ||
3634 strncmp(ent->de->name, ent->dentry->d_name.name,
d80d448c
DW		 3635 ent->de->name_len) ||
3636 force_reread) {
bd1af145
MS		 3637 retval = ext4_find_delete_entry(handle, ent->dir,
3638 &ent->dentry->d_name);
3639 } else {
3640 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3641 if (retval == -ENOENT) {
3642 retval = ext4_find_delete_entry(handle, ent->dir,
3643 &ent->dentry->d_name);
3644 }
3645 }
3646
3647 if (retval) {
b03a2f7e
AD		 3648 ext4_warning_inode(ent->dir,
3649 "Deleting old file: nlink %d, error=%d",
3650 ent->dir->i_nlink, retval);
bd1af145
MS		 3651 }
3652}
3653
bd42998a
MS		 3654static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3655{
3656 if (ent->dir_nlink_delta) {
3657 if (ent->dir_nlink_delta == -1)
15ed2851 3658 ext4_dec_count(ent->dir);
bd42998a 3659 else
15ed2851 3660 ext4_inc_count(ent->dir);
bd42998a
MS		 3661 ext4_mark_inode_dirty(handle, ent->dir);
3662 }
3663}
3664
cd808dec
MS		 3665static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3666 int credits, handle_t **h)
3667{
3668 struct inode *wh;
3669 handle_t *handle;
3670 int retries = 0;
3671
3672 /*
3673 * for inode block, sb block, group summaries,
3674 * and inode bitmap
3675 */
3676 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3677 EXT4_XATTR_TRANS_BLOCKS + 4);
3678retry:
3679 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3680 &ent->dentry->d_name, 0, NULL,
3681 EXT4_HT_DIR, credits);
3682
3683 handle = ext4_journal_current_handle();
3684 if (IS_ERR(wh)) {
3685 if (handle)
3686 ext4_journal_stop(handle);
3687 if (PTR_ERR(wh) == -ENOSPC &&
3688 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3689 goto retry;
3690 } else {
3691 *h = handle;
3692 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3693 wh->i_op = &ext4_special_inode_operations;
3694 }
3695 return wh;
3696}
3697
ac27a0ec
DK		 3698/*
3699 * Anybody can rename anything with this: the permission checks are left to the
3700 * higher-level routines.
0e202704
TT		 3701 *
3702 * n.b. old_{dentry,inode) refers to the source dentry/inode
3703 * while new_{dentry,inode) refers to the destination dentry/inode
3704 * This comes from rename(const char *oldpath, const char *newpath)
ac27a0ec 3705 */
af5bc92d 3706static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
cd808dec
MS		 3707 struct inode *new_dir, struct dentry *new_dentry,
3708 unsigned int flags)
ac27a0ec 3709{
5b61de75 3710 handle_t *handle = NULL;
c0d268c3
MS		 3711 struct ext4_renament old = {
3712 .dir = old_dir,
3713 .dentry = old_dentry,
2b0143b5 3714 .inode = d_inode(old_dentry),
c0d268c3
MS		 3715 };
3716 struct ext4_renament new = {
3717 .dir = new_dir,
3718 .dentry = new_dentry,
2b0143b5 3719 .inode = d_inode(new_dentry),
c0d268c3 3720 };
d80d448c 3721 int force_reread;
0e202704 3722 int retval;
cd808dec
MS		 3723 struct inode *whiteout = NULL;
3724 int credits;
3725 u8 old_file_type;
907f4554 3726
b50282f3
TT		 3727 if (new.inode && new.inode->i_nlink == 0) {
3728 EXT4_ERROR_INODE(new.inode,
3729 "target of rename is already freed");
3730 return -EFSCORRUPTED;
3731 }
3732
040cb378
LX		 3733 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3734 (!projid_eq(EXT4_I(new_dir)->i_projid,
3735 EXT4_I(old_dentry->d_inode)->i_projid)))
3736 return -EXDEV;
3737
a7cdadee
JK		 3738 retval = dquot_initialize(old.dir);
3739 if (retval)
3740 return retval;
3741 retval = dquot_initialize(new.dir);
3742 if (retval)
3743 return retval;
ac27a0ec
DK		 3744
3745 /* Initialize quotas before so that eventual writes go
3746 * in separate transaction */
a7cdadee
JK		 3747 if (new.inode) {
3748 retval = dquot_initialize(new.inode);
3749 if (retval)
3750 return retval;
3751 }
ac27a0ec 3752
c0d268c3 3753 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
36de9286
TT		 3754 if (IS_ERR(old.bh))
3755 return PTR_ERR(old.bh);
ac27a0ec
DK		 3756 /*
3757 * Check for inode number is _not_ due to possible IO errors.
3758 * We might rmdir the source, keep it as pwd of some process
3759 * and merrily kill the link to whatever was created under the
3760 * same name. Goodbye sticky bit ;-<
3761 */
ac27a0ec 3762 retval = -ENOENT;
c0d268c3 3763 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
ac27a0ec
DK		 3764 goto end_rename;
3765
c0d268c3
MS		 3766 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3767 &new.de, &new.inlined);
36de9286
TT		 3768 if (IS_ERR(new.bh)) {
3769 retval = PTR_ERR(new.bh);
a9cfcd63 3770 new.bh = NULL;
36de9286
TT		 3771 goto end_rename;
3772 }
c0d268c3
MS		 3773 if (new.bh) {
3774 if (!new.inode) {
3775 brelse(new.bh);
3776 new.bh = NULL;
ac27a0ec
DK		 3777 }
3778 }
c0d268c3
MS		 3779 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3780 ext4_alloc_da_blocks(old.inode);
5b61de75 3781
cd808dec
MS		 3782 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3783 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3784 if (!(flags & RENAME_WHITEOUT)) {
3785 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
7071b715
KK		 3786 if (IS_ERR(handle)) {
3787 retval = PTR_ERR(handle);
3788 handle = NULL;
3789 goto end_rename;
3790 }
cd808dec
MS		 3791 } else {
3792 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
7071b715
KK		 3793 if (IS_ERR(whiteout)) {
3794 retval = PTR_ERR(whiteout);
3795 whiteout = NULL;
3796 goto end_rename;
3797 }
cd808dec 3798 }
5b61de75 3799
6b4b8e6b 3800 old_file_type = old.de->file_type;
c0d268c3 3801 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
5b61de75
TT		 3802 ext4_handle_sync(handle);
3803
c0d268c3
MS		 3804 if (S_ISDIR(old.inode->i_mode)) {
3805 if (new.inode) {
ac27a0ec 3806 retval = -ENOTEMPTY;
e875a2dd 3807 if (!ext4_empty_dir(new.inode))
ac27a0ec 3808 goto end_rename;
0d7d5d67
MS		 3809 } else {
3810 retval = -EMLINK;
3811 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3812 goto end_rename;
ac27a0ec 3813 }
bd1af145 3814 retval = ext4_rename_dir_prepare(handle, &old);
ef607893
AG		 3815 if (retval)
3816 goto end_rename;
ac27a0ec 3817 }
d80d448c
DW		 3818 /*
3819 * If we're renaming a file within an inline_data dir and adding or
3820 * setting the new dirent causes a conversion from inline_data to
3821 * extents/blockmap, we need to force the dirent delete code to
3822 * re-read the directory, or else we end up trying to delete a dirent
3823 * from what is now the extent tree root (or a block map).
3824 */
3825 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3826 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
cd808dec 3827
cd808dec
MS		 3828 if (whiteout) {
3829 /*
3830 * Do this before adding a new entry, so the old entry is sure
3831 * to be still pointing to the valid old entry.
3832 */
3833 retval = ext4_setent(handle, &old, whiteout->i_ino,
3834 EXT4_FT_CHRDEV);
3835 if (retval)
3836 goto end_rename;
4209ae12
HS		 3837 retval = ext4_mark_inode_dirty(handle, whiteout);
3838 if (unlikely(retval))
3839 goto end_rename;
cd808dec 3840 }
c0d268c3
MS		 3841 if (!new.bh) {
3842 retval = ext4_add_entry(handle, new.dentry, old.inode);
ac27a0ec
DK		 3843 if (retval)
3844 goto end_rename;
3845 } else {
bd1af145 3846 retval = ext4_setent(handle, &new,
cd808dec 3847 old.inode->i_ino, old_file_type);
ef607893
AG		 3848 if (retval)
3849 goto end_rename;
ac27a0ec 3850 }
d80d448c
DW		 3851 if (force_reread)
3852 force_reread = !ext4_test_inode_flag(new.dir,
3853 EXT4_INODE_INLINE_DATA);
ac27a0ec
DK		 3854
3855 /*
3856 * Like most other Unix systems, set the ctime for inodes on a
3857 * rename.
3858 */
eeca7ea1 3859 old.inode->i_ctime = current_time(old.inode);
4209ae12
HS		 3860 retval = ext4_mark_inode_dirty(handle, old.inode);
3861 if (unlikely(retval))
3862 goto end_rename;
ac27a0ec 3863
cd808dec
MS		 3864 if (!whiteout) {
3865 /*
3866 * ok, that's it
3867 */
3868 ext4_rename_delete(handle, &old, force_reread);
3869 }
ac27a0ec 3870
c0d268c3 3871 if (new.inode) {
15ed2851 3872 ext4_dec_count(new.inode);
eeca7ea1 3873 new.inode->i_ctime = current_time(new.inode);
ac27a0ec 3874 }
eeca7ea1 3875 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
c0d268c3
MS		 3876 ext4_update_dx_flag(old.dir);
3877 if (old.dir_bh) {
bd1af145
MS		 3878 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3879 if (retval)
b4097142 3880 goto end_rename;
bd1af145 3881
15ed2851 3882 ext4_dec_count(old.dir);
c0d268c3 3883 if (new.inode) {
e875a2dd
MH		 3884 /* checked ext4_empty_dir above, can't have another
3885 * parent, ext4_dec_count() won't work for many-linked
3886 * dirs */
c0d268c3 3887 clear_nlink(new.inode);
ac27a0ec 3888 } else {
15ed2851 3889 ext4_inc_count(new.dir);
c0d268c3 3890 ext4_update_dx_flag(new.dir);
4209ae12
HS		 3891 retval = ext4_mark_inode_dirty(handle, new.dir);
3892 if (unlikely(retval))
3893 goto end_rename;
ac27a0ec
DK		 3894 }
3895 }
4209ae12
HS		 3896 retval = ext4_mark_inode_dirty(handle, old.dir);
3897 if (unlikely(retval))
3898 goto end_rename;
aa75f4d3
HS		 3899
3900 if (S_ISDIR(old.inode->i_mode)) {
3901 /*
3902 * We disable fast commits here that's because the
3903 * replay code is not yet capable of changing dot dot
3904 * dirents in directories.
3905 */
3906 ext4_fc_mark_ineligible(old.inode->i_sb,
3907 EXT4_FC_REASON_RENAME_DIR);
3908 } else {
3909 if (new.inode)
a80f7fcf
HS		 3910 ext4_fc_track_unlink(handle, new.dentry);
3911 __ext4_fc_track_link(handle, old.inode, new.dentry);
3912 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
aa75f4d3
HS		 3913 }
3914
c0d268c3 3915 if (new.inode) {
4209ae12
HS		 3916 retval = ext4_mark_inode_dirty(handle, new.inode);
3917 if (unlikely(retval))
3918 goto end_rename;
c0d268c3
MS		 3919 if (!new.inode->i_nlink)
3920 ext4_orphan_add(handle, new.inode);
ac27a0ec
DK		 3921 }
3922 retval = 0;
3923
3924end_rename:
cd808dec 3925 if (whiteout) {
6b4b8e6b 3926 if (retval) {
3927 ext4_setent(handle, &old,
3928 old.inode->i_ino, old_file_type);
cd808dec 3929 drop_nlink(whiteout);
6b4b8e6b 3930 }
cd808dec
MS		 3931 unlock_new_inode(whiteout);
3932 iput(whiteout);
6b4b8e6b 3933
cd808dec 3934 }
6b4b8e6b 3935 brelse(old.dir_bh);
3936 brelse(old.bh);
3937 brelse(new.bh);
5b61de75
TT		 3938 if (handle)
3939 ext4_journal_stop(handle);
ac27a0ec
DK		 3940 return retval;
3941}
3942
bd42998a
MS		 3943static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3944 struct inode *new_dir, struct dentry *new_dentry)
3945{
3946 handle_t *handle = NULL;
3947 struct ext4_renament old = {
3948 .dir = old_dir,
3949 .dentry = old_dentry,
2b0143b5 3950 .inode = d_inode(old_dentry),
bd42998a
MS		 3951 };
3952 struct ext4_renament new = {
3953 .dir = new_dir,
3954 .dentry = new_dentry,
2b0143b5 3955 .inode = d_inode(new_dentry),
bd42998a
MS		 3956 };
3957 u8 new_file_type;
3958 int retval;
95582b00 3959 struct timespec64 ctime;
bd42998a 3960
040cb378
LX		 3961 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3962 !projid_eq(EXT4_I(new_dir)->i_projid,
3963 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3964 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3965 !projid_eq(EXT4_I(old_dir)->i_projid,
3966 EXT4_I(new_dentry->d_inode)->i_projid)))
3967 return -EXDEV;
3968
a7cdadee
JK		 3969 retval = dquot_initialize(old.dir);
3970 if (retval)
3971 return retval;
3972 retval = dquot_initialize(new.dir);
3973 if (retval)
3974 return retval;
bd42998a
MS		 3975
3976 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3977 &old.de, &old.inlined);
36de9286
TT		 3978 if (IS_ERR(old.bh))
3979 return PTR_ERR(old.bh);
bd42998a
MS		 3980 /*
3981 * Check for inode number is _not_ due to possible IO errors.
3982 * We might rmdir the source, keep it as pwd of some process
3983 * and merrily kill the link to whatever was created under the
3984 * same name. Goodbye sticky bit ;-<
3985 */
3986 retval = -ENOENT;
3987 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3988 goto end_rename;
3989
3990 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3991 &new.de, &new.inlined);
36de9286
TT		 3992 if (IS_ERR(new.bh)) {
3993 retval = PTR_ERR(new.bh);
a9cfcd63 3994 new.bh = NULL;
36de9286
TT		 3995 goto end_rename;
3996 }
bd42998a
MS		 3997
3998 /* RENAME_EXCHANGE case: old *and* new must both exist */
3999 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4000 goto end_rename;
4001
4002 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4003 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4004 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
7071b715
KK		 4005 if (IS_ERR(handle)) {
4006 retval = PTR_ERR(handle);
4007 handle = NULL;
4008 goto end_rename;
4009 }
bd42998a
MS		 4010
4011 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4012 ext4_handle_sync(handle);
4013
4014 if (S_ISDIR(old.inode->i_mode)) {
4015 old.is_dir = true;
4016 retval = ext4_rename_dir_prepare(handle, &old);
4017 if (retval)
4018 goto end_rename;
4019 }
4020 if (S_ISDIR(new.inode->i_mode)) {
4021 new.is_dir = true;
4022 retval = ext4_rename_dir_prepare(handle, &new);
4023 if (retval)
4024 goto end_rename;
4025 }
4026
4027 /*
4028 * Other than the special case of overwriting a directory, parents'
4029 * nlink only needs to be modified if this is a cross directory rename.
4030 */
4031 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4032 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4033 new.dir_nlink_delta = -old.dir_nlink_delta;
4034 retval = -EMLINK;
4035 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4036 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4037 goto end_rename;
4038 }
4039
4040 new_file_type = new.de->file_type;
4041 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4042 if (retval)
4043 goto end_rename;
4044
4045 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4046 if (retval)
4047 goto end_rename;
4048
4049 /*
4050 * Like most other Unix systems, set the ctime for inodes on a
4051 * rename.
4052 */
eeca7ea1
DD		 4053 ctime = current_time(old.inode);
4054 old.inode->i_ctime = ctime;
4055 new.inode->i_ctime = ctime;
4209ae12
HS		 4056 retval = ext4_mark_inode_dirty(handle, old.inode);
4057 if (unlikely(retval))
4058 goto end_rename;
4059 retval = ext4_mark_inode_dirty(handle, new.inode);
4060 if (unlikely(retval))
4061 goto end_rename;
aa75f4d3
HS		 4062 ext4_fc_mark_ineligible(new.inode->i_sb,
4063 EXT4_FC_REASON_CROSS_RENAME);
bd42998a
MS		 4064 if (old.dir_bh) {
4065 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4066 if (retval)
4067 goto end_rename;
4068 }
4069 if (new.dir_bh) {
4070 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4071 if (retval)
4072 goto end_rename;
4073 }
4074 ext4_update_dir_count(handle, &old);
4075 ext4_update_dir_count(handle, &new);
4076 retval = 0;
4077
4078end_rename:
4079 brelse(old.dir_bh);
4080 brelse(new.dir_bh);
4081 brelse(old.bh);
4082 brelse(new.bh);
4083 if (handle)
4084 ext4_journal_stop(handle);
4085 return retval;
4086}
4087
0a7c3937
MS		 4088static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4089 struct inode *new_dir, struct dentry *new_dentry,
4090 unsigned int flags)
4091{
07543d16
EB		 4092 int err;
4093
0db1ff22
TT		 4094 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4095 return -EIO;
4096
cd808dec 4097 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
0a7c3937
MS		 4098 return -EINVAL;
4099
07543d16
EB		 4100 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4101 flags);
4102 if (err)
4103 return err;
4104
bd42998a
MS		 4105 if (flags & RENAME_EXCHANGE) {
4106 return ext4_cross_rename(old_dir, old_dentry,
4107 new_dir, new_dentry);
4108 }
cd808dec
MS		 4109
4110 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
0a7c3937
MS		 4111}
4112
ac27a0ec
DK		 4113/*
4114 * directories can handle most operations...
4115 */
754661f1 4116const struct inode_operations ext4_dir_inode_operations = {
617ba13b
MC		 4117 .create = ext4_create,
4118 .lookup = ext4_lookup,
4119 .link = ext4_link,
4120 .unlink = ext4_unlink,
4121 .symlink = ext4_symlink,
4122 .mkdir = ext4_mkdir,
4123 .rmdir = ext4_rmdir,
4124 .mknod = ext4_mknod,
af51a2ac 4125 .tmpfile = ext4_tmpfile,
2773bf00 4126 .rename = ext4_rename2,
617ba13b 4127 .setattr = ext4_setattr,
99652ea5 4128 .getattr = ext4_getattr,
617ba13b 4129 .listxattr = ext4_listxattr,
4e34e719 4130 .get_acl = ext4_get_acl,
64e178a7 4131 .set_acl = ext4_set_acl,
abc8746e 4132 .fiemap = ext4_fiemap,
ac27a0ec
DK		 4133};
4134
754661f1 4135const struct inode_operations ext4_special_inode_operations = {
617ba13b 4136 .setattr = ext4_setattr,
99652ea5 4137 .getattr = ext4_getattr,
617ba13b 4138 .listxattr = ext4_listxattr,
4e34e719 4139 .get_acl = ext4_get_acl,
64e178a7 4140 .set_acl = ext4_set_acl,
ac27a0ec 4141};
Linux 6.x block layer and io_uring tree(s)