]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/ext4/namei.c
Merge tag 'linux-kselftest-4.13-rc6-fixes' of git://git.kernel.org/pub/scm/linux...
[mirror_ubuntu-artful-kernel.git] / fs / ext4 / namei.c
CommitLineData
ac27a0ec 1/*
617ba13b 2 * linux/fs/ext4/namei.c
ac27a0ec
DK
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
24 * Theodore Ts'o, 2002
25 */
26
27#include <linux/fs.h>
28#include <linux/pagemap.h>
ac27a0ec 29#include <linux/time.h>
ac27a0ec
DK
30#include <linux/fcntl.h>
31#include <linux/stat.h>
32#include <linux/string.h>
33#include <linux/quotaops.h>
34#include <linux/buffer_head.h>
35#include <linux/bio.h>
3dcf5451
CH
36#include "ext4.h"
37#include "ext4_jbd2.h"
ac27a0ec 38
ac27a0ec
DK
39#include "xattr.h"
40#include "acl.h"
41
0562e0ba 42#include <trace/events/ext4.h>
ac27a0ec
DK
43/*
44 * define how far ahead to read directories while searching them.
45 */
46#define NAMEI_RA_CHUNKS 2
47#define NAMEI_RA_BLOCKS 4
8c55e204 48#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
ac27a0ec 49
617ba13b 50static struct buffer_head *ext4_append(handle_t *handle,
ac27a0ec 51 struct inode *inode,
0f70b406 52 ext4_lblk_t *block)
ac27a0ec
DK
53{
54 struct buffer_head *bh;
1c215028 55 int err;
ac27a0ec 56
df981d03
TT
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
0f70b406
TT
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
df981d03 61
ac27a0ec
DK
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63
c5e298ae 64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
1c215028
TT
65 if (IS_ERR(bh))
66 return bh;
0f70b406
TT
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
5d601255 69 BUFFER_TRACE(bh, "get_write_access");
0f70b406
TT
70 err = ext4_journal_get_write_access(handle, bh);
71 if (err) {
72 brelse(bh);
73 ext4_std_error(inode->i_sb, err);
74 return ERR_PTR(err);
6d1ab10e 75 }
ac27a0ec
DK
76 return bh;
77}
78
dc6982ff
TT
79static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
81
82typedef enum {
83 EITHER, INDEX, DIRENT
84} dirblock_type_t;
85
86#define ext4_read_dirblock(inode, block, type) \
b03a2f7e 87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
dc6982ff
TT
88
89static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
b03a2f7e
AD
90 ext4_lblk_t block,
91 dirblock_type_t type,
92 const char *func,
93 unsigned int line)
dc6982ff
TT
94{
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
1c215028 97 int is_dx_block = 0;
dc6982ff 98
1c215028
TT
99 bh = ext4_bread(NULL, inode, block, 0);
100 if (IS_ERR(bh)) {
b03a2f7e
AD
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
1c215028
TT
106
107 return bh;
108 }
109 if (!bh) {
b03a2f7e
AD
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
6a797d27 112 return ERR_PTR(-EFSCORRUPTED);
dc6982ff
TT
113 }
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
116 if (is_dx(inode)) {
117 if (block == 0)
118 is_dx_block = 1;
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
122 is_dx_block = 1;
123 }
124 if (!is_dx_block && type == INDEX) {
b03a2f7e 125 ext4_error_inode(inode, func, line, block,
dc6982ff 126 "directory leaf block found instead of index block");
6a797d27 127 return ERR_PTR(-EFSCORRUPTED);
dc6982ff 128 }
9aa5d32b 129 if (!ext4_has_metadata_csum(inode->i_sb) ||
dc6982ff
TT
130 buffer_verified(bh))
131 return bh;
132
133 /*
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
137 */
138 if (is_dx_block && type == INDEX) {
139 if (ext4_dx_csum_verify(inode, dirent))
140 set_buffer_verified(bh);
141 else {
b03a2f7e
AD
142 ext4_error_inode(inode, func, line, block,
143 "Directory index failed checksum");
a871611b 144 brelse(bh);
6a797d27 145 return ERR_PTR(-EFSBADCRC);
a871611b 146 }
ac27a0ec 147 }
dc6982ff
TT
148 if (!is_dx_block) {
149 if (ext4_dirent_csum_verify(inode, dirent))
150 set_buffer_verified(bh);
151 else {
b03a2f7e
AD
152 ext4_error_inode(inode, func, line, block,
153 "Directory block failed checksum");
dc6982ff 154 brelse(bh);
6a797d27 155 return ERR_PTR(-EFSBADCRC);
dc6982ff 156 }
6d1ab10e 157 }
ac27a0ec
DK
158 return bh;
159}
160
161#ifndef assert
162#define assert(test) J_ASSERT(test)
163#endif
164
ac27a0ec
DK
165#ifdef DX_DEBUG
166#define dxtrace(command) command
167#else
168#define dxtrace(command)
169#endif
170
171struct fake_dirent
172{
173 __le32 inode;
174 __le16 rec_len;
175 u8 name_len;
176 u8 file_type;
177};
178
179struct dx_countlimit
180{
181 __le16 limit;
182 __le16 count;
183};
184
185struct dx_entry
186{
187 __le32 hash;
188 __le32 block;
189};
190
191/*
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
195 */
196
197struct dx_root
198{
199 struct fake_dirent dot;
200 char dot_name[4];
201 struct fake_dirent dotdot;
202 char dotdot_name[4];
203 struct dx_root_info
204 {
205 __le32 reserved_zero;
206 u8 hash_version;
207 u8 info_length; /* 8 */
208 u8 indirect_levels;
209 u8 unused_flags;
210 }
211 info;
212 struct dx_entry entries[0];
213};
214
215struct dx_node
216{
217 struct fake_dirent fake;
218 struct dx_entry entries[0];
219};
220
221
222struct dx_frame
223{
224 struct buffer_head *bh;
225 struct dx_entry *entries;
226 struct dx_entry *at;
227};
228
229struct dx_map_entry
230{
231 u32 hash;
ef2b02d3
ES
232 u16 offs;
233 u16 size;
ac27a0ec
DK
234};
235
e6153918
DW
236/*
237 * This goes at the end of each htree block.
238 */
239struct dx_tail {
240 u32 dt_reserved;
241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
242};
243
725d26d3
AK
244static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
af5bc92d
TT
246static inline unsigned dx_get_hash(struct dx_entry *entry);
247static void dx_set_hash(struct dx_entry *entry, unsigned value);
248static unsigned dx_get_count(struct dx_entry *entries);
249static unsigned dx_get_limit(struct dx_entry *entries);
250static void dx_set_count(struct dx_entry *entries, unsigned value);
251static void dx_set_limit(struct dx_entry *entries, unsigned value);
252static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253static unsigned dx_node_limit(struct inode *dir);
5b643f9c 254static struct dx_frame *dx_probe(struct ext4_filename *fname,
ac27a0ec
DK
255 struct inode *dir,
256 struct dx_hash_info *hinfo,
dd73b5d5 257 struct dx_frame *frame);
af5bc92d 258static void dx_release(struct dx_frame *frames);
1f3862b5
MH
259static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 unsigned blocksize, struct dx_hash_info *hinfo,
261 struct dx_map_entry map[]);
ac27a0ec 262static void dx_sort_map(struct dx_map_entry *map, unsigned count);
af5bc92d 263static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
3d0518f4 264 struct dx_map_entry *offsets, int count, unsigned blocksize);
8bad4597 265static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
725d26d3
AK
266static void dx_insert_block(struct dx_frame *frame,
267 u32 hash, ext4_lblk_t block);
617ba13b 268static int ext4_htree_next_block(struct inode *dir, __u32 hash,
ac27a0ec
DK
269 struct dx_frame *frame,
270 struct dx_frame *frames,
271 __u32 *start_hash);
f702ba0f 272static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
5b643f9c 273 struct ext4_filename *fname,
537d8f93 274 struct ext4_dir_entry_2 **res_dir);
5b643f9c 275static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
56a04915 276 struct inode *dir, struct inode *inode);
ac27a0ec 277
dbe89444 278/* checksumming functions */
3c47d541
TM
279void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 unsigned int blocksize)
b0336e8d
DW
281{
282 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 t->det_rec_len = ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail), blocksize);
285 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
286}
287
288/* Walk through a dirent block to find a checksum "dirent" at the tail */
289static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 struct ext4_dir_entry *de)
291{
292 struct ext4_dir_entry_tail *t;
293
294#ifdef PARANOID
295 struct ext4_dir_entry *d, *top;
296
297 d = de;
298 top = (struct ext4_dir_entry *)(((void *)de) +
299 (EXT4_BLOCK_SIZE(inode->i_sb) -
300 sizeof(struct ext4_dir_entry_tail)));
301 while (d < top && d->rec_len)
302 d = (struct ext4_dir_entry *)(((void *)d) +
303 le16_to_cpu(d->rec_len));
304
305 if (d != top)
306 return NULL;
307
308 t = (struct ext4_dir_entry_tail *)d;
309#else
310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
311#endif
312
313 if (t->det_reserved_zero1 ||
314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 t->det_reserved_zero2 ||
316 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
317 return NULL;
318
319 return t;
320}
321
322static __le32 ext4_dirent_csum(struct inode *inode,
323 struct ext4_dir_entry *dirent, int size)
324{
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct ext4_inode_info *ei = EXT4_I(inode);
327 __u32 csum;
328
329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 return cpu_to_le32(csum);
331}
332
b03a2f7e
AD
333#define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
335
336static void __warn_no_space_for_csum(struct inode *inode, const char *func,
337 unsigned int line)
dffe9d8d 338{
b03a2f7e
AD
339 __ext4_warning_inode(inode, func, line,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
dffe9d8d
TT
341}
342
b0336e8d
DW
343int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
344{
345 struct ext4_dir_entry_tail *t;
346
9aa5d32b 347 if (!ext4_has_metadata_csum(inode->i_sb))
b0336e8d
DW
348 return 1;
349
350 t = get_dirent_tail(inode, dirent);
351 if (!t) {
dffe9d8d 352 warn_no_space_for_csum(inode);
b0336e8d
DW
353 return 0;
354 }
355
356 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 (void *)t - (void *)dirent))
358 return 0;
359
360 return 1;
361}
362
363static void ext4_dirent_csum_set(struct inode *inode,
364 struct ext4_dir_entry *dirent)
365{
366 struct ext4_dir_entry_tail *t;
367
9aa5d32b 368 if (!ext4_has_metadata_csum(inode->i_sb))
b0336e8d
DW
369 return;
370
371 t = get_dirent_tail(inode, dirent);
372 if (!t) {
dffe9d8d 373 warn_no_space_for_csum(inode);
b0336e8d
DW
374 return;
375 }
376
377 t->det_checksum = ext4_dirent_csum(inode, dirent,
378 (void *)t - (void *)dirent);
379}
380
3c47d541
TM
381int ext4_handle_dirty_dirent_node(handle_t *handle,
382 struct inode *inode,
383 struct buffer_head *bh)
b0336e8d
DW
384{
385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 return ext4_handle_dirty_metadata(handle, inode, bh);
387}
388
dbe89444
DW
389static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 struct ext4_dir_entry *dirent,
391 int *offset)
392{
393 struct ext4_dir_entry *dp;
394 struct dx_root_info *root;
395 int count_offset;
396
397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
398 count_offset = 8;
399 else if (le16_to_cpu(dirent->rec_len) == 12) {
400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 if (le16_to_cpu(dp->rec_len) !=
402 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
403 return NULL;
404 root = (struct dx_root_info *)(((void *)dp + 12));
405 if (root->reserved_zero ||
406 root->info_length != sizeof(struct dx_root_info))
407 return NULL;
408 count_offset = 32;
409 } else
410 return NULL;
411
412 if (offset)
413 *offset = count_offset;
414 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
415}
416
417static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 int count_offset, int count, struct dx_tail *t)
419{
420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 struct ext4_inode_info *ei = EXT4_I(inode);
d6a77105 422 __u32 csum;
dbe89444 423 int size;
b47820ed
DJ
424 __u32 dummy_csum = 0;
425 int offset = offsetof(struct dx_tail, dt_checksum);
dbe89444
DW
426
427 size = count_offset + (count * sizeof(struct dx_entry));
dbe89444 428 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
b47820ed
DJ
429 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
430 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
dbe89444
DW
431
432 return cpu_to_le32(csum);
433}
434
435static int ext4_dx_csum_verify(struct inode *inode,
436 struct ext4_dir_entry *dirent)
437{
438 struct dx_countlimit *c;
439 struct dx_tail *t;
440 int count_offset, limit, count;
441
9aa5d32b 442 if (!ext4_has_metadata_csum(inode->i_sb))
dbe89444
DW
443 return 1;
444
445 c = get_dx_countlimit(inode, dirent, &count_offset);
446 if (!c) {
447 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
fa964540 448 return 0;
dbe89444
DW
449 }
450 limit = le16_to_cpu(c->limit);
451 count = le16_to_cpu(c->count);
452 if (count_offset + (limit * sizeof(struct dx_entry)) >
453 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
dffe9d8d 454 warn_no_space_for_csum(inode);
fa964540 455 return 0;
dbe89444
DW
456 }
457 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
458
459 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
460 count, t))
461 return 0;
462 return 1;
463}
464
465static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
466{
467 struct dx_countlimit *c;
468 struct dx_tail *t;
469 int count_offset, limit, count;
470
9aa5d32b 471 if (!ext4_has_metadata_csum(inode->i_sb))
dbe89444
DW
472 return;
473
474 c = get_dx_countlimit(inode, dirent, &count_offset);
475 if (!c) {
476 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
477 return;
478 }
479 limit = le16_to_cpu(c->limit);
480 count = le16_to_cpu(c->count);
481 if (count_offset + (limit * sizeof(struct dx_entry)) >
482 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
dffe9d8d 483 warn_no_space_for_csum(inode);
dbe89444
DW
484 return;
485 }
486 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
487
488 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
489}
490
491static inline int ext4_handle_dirty_dx_node(handle_t *handle,
492 struct inode *inode,
493 struct buffer_head *bh)
494{
495 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 return ext4_handle_dirty_metadata(handle, inode, bh);
497}
498
f795e140
LZ
499/*
500 * p is at least 6 bytes before the end of page
501 */
502static inline struct ext4_dir_entry_2 *
3d0518f4 503ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
f795e140
LZ
504{
505 return (struct ext4_dir_entry_2 *)((char *)p +
3d0518f4 506 ext4_rec_len_from_disk(p->rec_len, blocksize));
f795e140
LZ
507}
508
ac27a0ec
DK
509/*
510 * Future: use high four bits of block for coalesce-on-delete flags
511 * Mask them off for now.
512 */
513
725d26d3 514static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
ac27a0ec 515{
e08ac99f 516 return le32_to_cpu(entry->block) & 0x0fffffff;
ac27a0ec
DK
517}
518
725d26d3 519static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
ac27a0ec
DK
520{
521 entry->block = cpu_to_le32(value);
522}
523
af5bc92d 524static inline unsigned dx_get_hash(struct dx_entry *entry)
ac27a0ec
DK
525{
526 return le32_to_cpu(entry->hash);
527}
528
af5bc92d 529static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
ac27a0ec
DK
530{
531 entry->hash = cpu_to_le32(value);
532}
533
af5bc92d 534static inline unsigned dx_get_count(struct dx_entry *entries)
ac27a0ec
DK
535{
536 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537}
538
af5bc92d 539static inline unsigned dx_get_limit(struct dx_entry *entries)
ac27a0ec
DK
540{
541 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542}
543
af5bc92d 544static inline void dx_set_count(struct dx_entry *entries, unsigned value)
ac27a0ec
DK
545{
546 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547}
548
af5bc92d 549static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
ac27a0ec
DK
550{
551 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552}
553
af5bc92d 554static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
ac27a0ec 555{
617ba13b
MC
556 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 EXT4_DIR_REC_LEN(2) - infosize;
dbe89444 558
9aa5d32b 559 if (ext4_has_metadata_csum(dir->i_sb))
dbe89444 560 entry_space -= sizeof(struct dx_tail);
d9c769b7 561 return entry_space / sizeof(struct dx_entry);
ac27a0ec
DK
562}
563
af5bc92d 564static inline unsigned dx_node_limit(struct inode *dir)
ac27a0ec 565{
617ba13b 566 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
dbe89444 567
9aa5d32b 568 if (ext4_has_metadata_csum(dir->i_sb))
dbe89444 569 entry_space -= sizeof(struct dx_tail);
d9c769b7 570 return entry_space / sizeof(struct dx_entry);
ac27a0ec
DK
571}
572
573/*
574 * Debug
575 */
576#ifdef DX_DEBUG
4776004f 577static void dx_show_index(char * label, struct dx_entry *entries)
ac27a0ec 578{
63f57933 579 int i, n = dx_get_count (entries);
d74f3d25 580 printk(KERN_DEBUG "%s index", label);
63f57933 581 for (i = 0; i < n; i++) {
d74f3d25
JP
582 printk(KERN_CONT " %x->%lu",
583 i ? dx_get_hash(entries + i) : 0,
584 (unsigned long)dx_get_block(entries + i));
63f57933 585 }
d74f3d25 586 printk(KERN_CONT "\n");
ac27a0ec
DK
587}
588
589struct stats
590{
591 unsigned names;
592 unsigned space;
593 unsigned bcount;
594};
595
b3098486
MH
596static struct stats dx_show_leaf(struct inode *dir,
597 struct dx_hash_info *hinfo,
598 struct ext4_dir_entry_2 *de,
599 int size, int show_names)
ac27a0ec
DK
600{
601 unsigned names = 0, space = 0;
602 char *base = (char *) de;
603 struct dx_hash_info h = *hinfo;
604
605 printk("names: ");
606 while ((char *) de < base + size)
607 {
608 if (de->inode)
609 {
610 if (show_names)
611 {
b3098486
MH
612#ifdef CONFIG_EXT4_FS_ENCRYPTION
613 int len;
614 char *name;
a7550b30
JK
615 struct fscrypt_str fname_crypto_str =
616 FSTR_INIT(NULL, 0);
c936e1ec 617 int res = 0;
b3098486
MH
618
619 name = de->name;
620 len = de->name_len;
a7550b30
JK
621 if (ext4_encrypted_inode(dir))
622 res = fscrypt_get_encryption_info(dir);
b7236e21
TT
623 if (res) {
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
b3098486 626 }
a7550b30 627 if (!fscrypt_has_encryption_key(dir)) {
b3098486
MH
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
630 de->name_len, &h);
631 printk("%*.s:(U)%x.%u ", len,
632 name, h.hash,
633 (unsigned) ((char *) de
634 - base));
635 } else {
a7550b30
JK
636 struct fscrypt_str de_name =
637 FSTR_INIT(name, len);
638
b3098486 639 /* Directory is encrypted */
a7550b30
JK
640 res = fscrypt_fname_alloc_buffer(
641 dir, len,
b3098486 642 &fname_crypto_str);
ef1eb3aa 643 if (res)
b3098486
MH
644 printk(KERN_WARNING "Error "
645 "allocating crypto "
646 "buffer--skipping "
647 "crypto\n");
a7550b30
JK
648 res = fscrypt_fname_disk_to_usr(dir,
649 0, 0, &de_name,
650 &fname_crypto_str);
ef1eb3aa 651 if (res) {
b3098486
MH
652 printk(KERN_WARNING "Error "
653 "converting filename "
654 "from disk to usr"
655 "\n");
656 name = "??";
657 len = 2;
658 } else {
659 name = fname_crypto_str.name;
660 len = fname_crypto_str.len;
661 }
5de0b4d0
TT
662 ext4fs_dirhash(de->name, de->name_len,
663 &h);
b3098486
MH
664 printk("%*.s:(E)%x.%u ", len, name,
665 h.hash, (unsigned) ((char *) de
666 - base));
a7550b30
JK
667 fscrypt_fname_free_buffer(
668 &fname_crypto_str);
b3098486
MH
669 }
670#else
ac27a0ec
DK
671 int len = de->name_len;
672 char *name = de->name;
617ba13b 673 ext4fs_dirhash(de->name, de->name_len, &h);
b3098486 674 printk("%*.s:%x.%u ", len, name, h.hash,
265c6a0f 675 (unsigned) ((char *) de - base));
b3098486 676#endif
ac27a0ec 677 }
617ba13b 678 space += EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec
DK
679 names++;
680 }
3d0518f4 681 de = ext4_next_entry(de, size);
ac27a0ec 682 }
d74f3d25 683 printk(KERN_CONT "(%i)\n", names);
ac27a0ec
DK
684 return (struct stats) { names, space, 1 };
685}
686
687struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
688 struct dx_entry *entries, int levels)
689{
690 unsigned blocksize = dir->i_sb->s_blocksize;
af5bc92d 691 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
ac27a0ec
DK
692 unsigned bcount = 0;
693 struct buffer_head *bh;
ac27a0ec
DK
694 printk("%i indexed blocks...\n", count);
695 for (i = 0; i < count; i++, entries++)
696 {
725d26d3
AK
697 ext4_lblk_t block = dx_get_block(entries);
698 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
ac27a0ec
DK
699 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
700 struct stats stats;
701 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
1c215028
TT
702 bh = ext4_bread(NULL,dir, block, 0);
703 if (!bh || IS_ERR(bh))
704 continue;
ac27a0ec
DK
705 stats = levels?
706 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
b3098486
MH
707 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
708 bh->b_data, blocksize, 0);
ac27a0ec
DK
709 names += stats.names;
710 space += stats.space;
711 bcount += stats.bcount;
af5bc92d 712 brelse(bh);
ac27a0ec
DK
713 }
714 if (bcount)
60e6679e 715 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
4776004f
TT
716 levels ? "" : " ", names, space/bcount,
717 (space/bcount)*100/blocksize);
ac27a0ec
DK
718 return (struct stats) { names, space, bcount};
719}
720#endif /* DX_DEBUG */
721
722/*
723 * Probe for a directory leaf block to search.
724 *
725 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
726 * error in the directory index, and the caller should fall back to
727 * searching the directory normally. The callers of dx_probe **MUST**
728 * check for this error code, and make sure it never gets reflected
729 * back to userspace.
730 */
731static struct dx_frame *
5b643f9c 732dx_probe(struct ext4_filename *fname, struct inode *dir,
dd73b5d5 733 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
ac27a0ec
DK
734{
735 unsigned count, indirect;
736 struct dx_entry *at, *entries, *p, *q, *m;
737 struct dx_root *root;
ac27a0ec 738 struct dx_frame *frame = frame_in;
dd73b5d5 739 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
ac27a0ec
DK
740 u32 hash;
741
e08ac99f 742 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
dd73b5d5
TT
743 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
744 if (IS_ERR(frame->bh))
745 return (struct dx_frame *) frame->bh;
746
747 root = (struct dx_root *) frame->bh->b_data;
ac27a0ec
DK
748 if (root->info.hash_version != DX_HASH_TEA &&
749 root->info.hash_version != DX_HASH_HALF_MD4 &&
750 root->info.hash_version != DX_HASH_LEGACY) {
b03a2f7e
AD
751 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
752 root->info.hash_version);
ac27a0ec
DK
753 goto fail;
754 }
5b643f9c
TT
755 if (fname)
756 hinfo = &fname->hinfo;
ac27a0ec 757 hinfo->hash_version = root->info.hash_version;
f99b2589
TT
758 if (hinfo->hash_version <= DX_HASH_TEA)
759 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
617ba13b 760 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
5b643f9c
TT
761 if (fname && fname_name(fname))
762 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
ac27a0ec
DK
763 hash = hinfo->hash;
764
765 if (root->info.unused_flags & 1) {
b03a2f7e
AD
766 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
767 root->info.unused_flags);
ac27a0ec
DK
768 goto fail;
769 }
770
b03a2f7e 771 indirect = root->info.indirect_levels;
e08ac99f
AB
772 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
773 ext4_warning(dir->i_sb,
774 "Directory (ino: %lu) htree depth %#06x exceed"
775 "supported value", dir->i_ino,
776 ext4_dir_htree_level(dir->i_sb));
777 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
778 ext4_warning(dir->i_sb, "Enable large directory "
779 "feature to access it");
780 }
ac27a0ec
DK
781 goto fail;
782 }
783
b03a2f7e
AD
784 entries = (struct dx_entry *)(((char *)&root->info) +
785 root->info.info_length);
3d82abae
ES
786
787 if (dx_get_limit(entries) != dx_root_limit(dir,
788 root->info.info_length)) {
b03a2f7e
AD
789 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
790 dx_get_limit(entries),
791 dx_root_limit(dir, root->info.info_length));
3d82abae
ES
792 goto fail;
793 }
794
af5bc92d 795 dxtrace(printk("Look up %x", hash));
dd73b5d5 796 while (1) {
ac27a0ec 797 count = dx_get_count(entries);
3d82abae 798 if (!count || count > dx_get_limit(entries)) {
b03a2f7e
AD
799 ext4_warning_inode(dir,
800 "dx entry: count %u beyond limit %u",
801 count, dx_get_limit(entries));
dd73b5d5 802 goto fail;
3d82abae
ES
803 }
804
ac27a0ec
DK
805 p = entries + 1;
806 q = entries + count - 1;
dd73b5d5 807 while (p <= q) {
b03a2f7e 808 m = p + (q - p) / 2;
d74f3d25 809 dxtrace(printk(KERN_CONT "."));
ac27a0ec
DK
810 if (dx_get_hash(m) > hash)
811 q = m - 1;
812 else
813 p = m + 1;
814 }
815
dd73b5d5 816 if (0) { // linear search cross check
ac27a0ec
DK
817 unsigned n = count - 1;
818 at = entries;
819 while (n--)
820 {
d74f3d25 821 dxtrace(printk(KERN_CONT ","));
ac27a0ec
DK
822 if (dx_get_hash(++at) > hash)
823 {
824 at--;
825 break;
826 }
827 }
828 assert (at == p - 1);
829 }
830
831 at = p - 1;
d74f3d25
JP
832 dxtrace(printk(KERN_CONT " %x->%u\n",
833 at == entries ? 0 : dx_get_hash(at),
b03a2f7e 834 dx_get_block(at)));
ac27a0ec
DK
835 frame->entries = entries;
836 frame->at = at;
dd73b5d5
TT
837 if (!indirect--)
838 return frame;
839 frame++;
840 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
841 if (IS_ERR(frame->bh)) {
842 ret_err = (struct dx_frame *) frame->bh;
843 frame->bh = NULL;
844 goto fail;
dbe89444 845 }
dd73b5d5 846 entries = ((struct dx_node *) frame->bh->b_data)->entries;
dbe89444 847
b03a2f7e
AD
848 if (dx_get_limit(entries) != dx_node_limit(dir)) {
849 ext4_warning_inode(dir,
850 "dx entry: limit %u != node limit %u",
851 dx_get_limit(entries), dx_node_limit(dir));
dd73b5d5 852 goto fail;
3d82abae 853 }
ac27a0ec 854 }
dd73b5d5 855fail:
ac27a0ec
DK
856 while (frame >= frame_in) {
857 brelse(frame->bh);
858 frame--;
859 }
b3098486 860
dd73b5d5 861 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
b03a2f7e
AD
862 ext4_warning_inode(dir,
863 "Corrupt directory, running e2fsck is recommended");
dd73b5d5 864 return ret_err;
ac27a0ec
DK
865}
866
b03a2f7e 867static void dx_release(struct dx_frame *frames)
ac27a0ec 868{
e08ac99f
AB
869 struct dx_root_info *info;
870 int i;
871
ac27a0ec
DK
872 if (frames[0].bh == NULL)
873 return;
874
e08ac99f
AB
875 info = &((struct dx_root *)frames[0].bh->b_data)->info;
876 for (i = 0; i <= info->indirect_levels; i++) {
877 if (frames[i].bh == NULL)
878 break;
879 brelse(frames[i].bh);
880 frames[i].bh = NULL;
881 }
ac27a0ec
DK
882}
883
884/*
885 * This function increments the frame pointer to search the next leaf
886 * block, and reads in the necessary intervening nodes if the search
887 * should be necessary. Whether or not the search is necessary is
888 * controlled by the hash parameter. If the hash value is even, then
889 * the search is only continued if the next block starts with that
890 * hash value. This is used if we are searching for a specific file.
891 *
892 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
893 *
894 * This function returns 1 if the caller should continue to search,
895 * or 0 if it should not. If there is an error reading one of the
896 * index blocks, it will a negative error code.
897 *
898 * If start_hash is non-null, it will be filled in with the starting
899 * hash of the next page.
900 */
617ba13b 901static int ext4_htree_next_block(struct inode *dir, __u32 hash,
ac27a0ec
DK
902 struct dx_frame *frame,
903 struct dx_frame *frames,
904 __u32 *start_hash)
905{
906 struct dx_frame *p;
907 struct buffer_head *bh;
dc6982ff 908 int num_frames = 0;
ac27a0ec
DK
909 __u32 bhash;
910
911 p = frame;
912 /*
913 * Find the next leaf page by incrementing the frame pointer.
914 * If we run out of entries in the interior node, loop around and
915 * increment pointer in the parent node. When we break out of
916 * this loop, num_frames indicates the number of interior
917 * nodes need to be read.
918 */
919 while (1) {
920 if (++(p->at) < p->entries + dx_get_count(p->entries))
921 break;
922 if (p == frames)
923 return 0;
924 num_frames++;
925 p--;
926 }
927
928 /*
929 * If the hash is 1, then continue only if the next page has a
930 * continuation hash of any value. This is used for readdir
931 * handling. Otherwise, check to see if the hash matches the
932 * desired contiuation hash. If it doesn't, return since
933 * there's no point to read in the successive index pages.
934 */
935 bhash = dx_get_hash(p->at);
936 if (start_hash)
937 *start_hash = bhash;
938 if ((hash & 1) == 0) {
939 if ((bhash & ~1) != hash)
940 return 0;
941 }
942 /*
943 * If the hash is HASH_NB_ALWAYS, we always go to the next
944 * block so no check is necessary
945 */
946 while (num_frames--) {
dc6982ff
TT
947 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
948 if (IS_ERR(bh))
949 return PTR_ERR(bh);
ac27a0ec 950 p++;
af5bc92d 951 brelse(p->bh);
ac27a0ec
DK
952 p->bh = bh;
953 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
954 }
955 return 1;
956}
957
958
ac27a0ec
DK
959/*
960 * This function fills a red-black tree with information from a
961 * directory block. It returns the number directory entries loaded
962 * into the tree. If there is an error it is returned in err.
963 */
964static int htree_dirblock_to_tree(struct file *dir_file,
725d26d3 965 struct inode *dir, ext4_lblk_t block,
ac27a0ec
DK
966 struct dx_hash_info *hinfo,
967 __u32 start_hash, __u32 start_minor_hash)
968{
969 struct buffer_head *bh;
617ba13b 970 struct ext4_dir_entry_2 *de, *top;
90b0a973 971 int err = 0, count = 0;
a7550b30 972 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
ac27a0ec 973
725d26d3
AK
974 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
975 (unsigned long)block));
dc6982ff
TT
976 bh = ext4_read_dirblock(dir, block, DIRENT);
977 if (IS_ERR(bh))
978 return PTR_ERR(bh);
b0336e8d 979
617ba13b
MC
980 de = (struct ext4_dir_entry_2 *) bh->b_data;
981 top = (struct ext4_dir_entry_2 *) ((char *) de +
ac27a0ec 982 dir->i_sb->s_blocksize -
617ba13b 983 EXT4_DIR_REC_LEN(0));
1f3862b5
MH
984#ifdef CONFIG_EXT4_FS_ENCRYPTION
985 /* Check if the directory is encrypted */
b7236e21 986 if (ext4_encrypted_inode(dir)) {
a7550b30 987 err = fscrypt_get_encryption_info(dir);
c936e1ec
TT
988 if (err < 0) {
989 brelse(bh);
990 return err;
991 }
a7550b30 992 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1f3862b5
MH
993 &fname_crypto_str);
994 if (err < 0) {
1f3862b5
MH
995 brelse(bh);
996 return err;
997 }
998 }
999#endif
3d0518f4 1000 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
f7c21177 1001 if (ext4_check_dir_entry(dir, NULL, de, bh,
226ba972 1002 bh->b_data, bh->b_size,
cad3f007
TT
1003 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1004 + ((char *)de - bh->b_data))) {
64cb9273
AV
1005 /* silently ignore the rest of the block */
1006 break;
e6c40211 1007 }
617ba13b 1008 ext4fs_dirhash(de->name, de->name_len, hinfo);
ac27a0ec
DK
1009 if ((hinfo->hash < start_hash) ||
1010 ((hinfo->hash == start_hash) &&
1011 (hinfo->minor_hash < start_minor_hash)))
1012 continue;
1013 if (de->inode == 0)
1014 continue;
b7236e21 1015 if (!ext4_encrypted_inode(dir)) {
1f3862b5
MH
1016 tmp_str.name = de->name;
1017 tmp_str.len = de->name_len;
1018 err = ext4_htree_store_dirent(dir_file,
1019 hinfo->hash, hinfo->minor_hash, de,
1020 &tmp_str);
1021 } else {
d2299590 1022 int save_len = fname_crypto_str.len;
a7550b30
JK
1023 struct fscrypt_str de_name = FSTR_INIT(de->name,
1024 de->name_len);
d2299590 1025
1f3862b5 1026 /* Directory is encrypted */
a7550b30
JK
1027 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1028 hinfo->minor_hash, &de_name,
1029 &fname_crypto_str);
ef1eb3aa 1030 if (err) {
1f3862b5
MH
1031 count = err;
1032 goto errout;
1033 }
1034 err = ext4_htree_store_dirent(dir_file,
1035 hinfo->hash, hinfo->minor_hash, de,
1036 &fname_crypto_str);
d2299590 1037 fname_crypto_str.len = save_len;
1f3862b5 1038 }
2f61830a 1039 if (err != 0) {
1f3862b5
MH
1040 count = err;
1041 goto errout;
ac27a0ec
DK
1042 }
1043 count++;
1044 }
1f3862b5 1045errout:
ac27a0ec 1046 brelse(bh);
1f3862b5 1047#ifdef CONFIG_EXT4_FS_ENCRYPTION
a7550b30 1048 fscrypt_fname_free_buffer(&fname_crypto_str);
1f3862b5 1049#endif
ac27a0ec
DK
1050 return count;
1051}
1052
1053
1054/*
1055 * This function fills a red-black tree with information from a
1056 * directory. We start scanning the directory in hash order, starting
1057 * at start_hash and start_minor_hash.
1058 *
1059 * This function returns the number of entries inserted into the tree,
1060 * or a negative error code.
1061 */
617ba13b 1062int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
ac27a0ec
DK
1063 __u32 start_minor_hash, __u32 *next_hash)
1064{
1065 struct dx_hash_info hinfo;
617ba13b 1066 struct ext4_dir_entry_2 *de;
e08ac99f 1067 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 1068 struct inode *dir;
725d26d3 1069 ext4_lblk_t block;
ac27a0ec 1070 int count = 0;
725d26d3 1071 int ret, err;
ac27a0ec 1072 __u32 hashval;
a7550b30 1073 struct fscrypt_str tmp_str;
ac27a0ec 1074
60e6679e 1075 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
4776004f 1076 start_hash, start_minor_hash));
496ad9aa 1077 dir = file_inode(dir_file);
12e9b892 1078 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
617ba13b 1079 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
f99b2589
TT
1080 if (hinfo.hash_version <= DX_HASH_TEA)
1081 hinfo.hash_version +=
1082 EXT4_SB(dir->i_sb)->s_hash_unsigned;
617ba13b 1083 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
8af0f082
TM
1084 if (ext4_has_inline_data(dir)) {
1085 int has_inline_data = 1;
1086 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1087 &hinfo, start_hash,
1088 start_minor_hash,
1089 &has_inline_data);
1090 if (has_inline_data) {
1091 *next_hash = ~0;
1092 return count;
1093 }
1094 }
ac27a0ec
DK
1095 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1096 start_hash, start_minor_hash);
1097 *next_hash = ~0;
1098 return count;
1099 }
1100 hinfo.hash = start_hash;
1101 hinfo.minor_hash = 0;
dd73b5d5
TT
1102 frame = dx_probe(NULL, dir, &hinfo, frames);
1103 if (IS_ERR(frame))
1104 return PTR_ERR(frame);
ac27a0ec
DK
1105
1106 /* Add '.' and '..' from the htree header */
1107 if (!start_hash && !start_minor_hash) {
617ba13b 1108 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
2f61830a
TT
1109 tmp_str.name = de->name;
1110 tmp_str.len = de->name_len;
1111 err = ext4_htree_store_dirent(dir_file, 0, 0,
1112 de, &tmp_str);
1113 if (err != 0)
ac27a0ec
DK
1114 goto errout;
1115 count++;
1116 }
1117 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
617ba13b 1118 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
3d0518f4 1119 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
2f61830a
TT
1120 tmp_str.name = de->name;
1121 tmp_str.len = de->name_len;
1122 err = ext4_htree_store_dirent(dir_file, 2, 0,
1123 de, &tmp_str);
1124 if (err != 0)
ac27a0ec
DK
1125 goto errout;
1126 count++;
1127 }
1128
1129 while (1) {
1f60fbe7
TT
1130 if (fatal_signal_pending(current)) {
1131 err = -ERESTARTSYS;
1132 goto errout;
1133 }
1134 cond_resched();
ac27a0ec
DK
1135 block = dx_get_block(frame->at);
1136 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1137 start_hash, start_minor_hash);
1138 if (ret < 0) {
1139 err = ret;
1140 goto errout;
1141 }
1142 count += ret;
1143 hashval = ~0;
617ba13b 1144 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
ac27a0ec
DK
1145 frame, frames, &hashval);
1146 *next_hash = hashval;
1147 if (ret < 0) {
1148 err = ret;
1149 goto errout;
1150 }
1151 /*
1152 * Stop if: (a) there are no more entries, or
1153 * (b) we have inserted at least one entry and the
1154 * next hash value is not a continuation
1155 */
1156 if ((ret == 0) ||
1157 (count && ((hashval & 1) == 0)))
1158 break;
1159 }
1160 dx_release(frames);
4776004f
TT
1161 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1162 "next hash: %x\n", count, *next_hash));
ac27a0ec
DK
1163 return count;
1164errout:
1165 dx_release(frames);
1166 return (err);
1167}
1168
7335cd3b
TM
1169static inline int search_dirblock(struct buffer_head *bh,
1170 struct inode *dir,
5b643f9c 1171 struct ext4_filename *fname,
7335cd3b
TM
1172 unsigned int offset,
1173 struct ext4_dir_entry_2 **res_dir)
1174{
5b643f9c 1175 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
d6b97550 1176 fname, offset, res_dir);
7335cd3b
TM
1177}
1178
ac27a0ec
DK
1179/*
1180 * Directory block splitting, compacting
1181 */
1182
ef2b02d3
ES
1183/*
1184 * Create map of hash values, offsets, and sizes, stored at end of block.
1185 * Returns number of entries mapped.
1186 */
1f3862b5
MH
1187static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1188 unsigned blocksize, struct dx_hash_info *hinfo,
8bad4597 1189 struct dx_map_entry *map_tail)
ac27a0ec
DK
1190{
1191 int count = 0;
1192 char *base = (char *) de;
1193 struct dx_hash_info h = *hinfo;
1194
8bad4597 1195 while ((char *) de < base + blocksize) {
ac27a0ec 1196 if (de->name_len && de->inode) {
617ba13b 1197 ext4fs_dirhash(de->name, de->name_len, &h);
ac27a0ec
DK
1198 map_tail--;
1199 map_tail->hash = h.hash;
9aee2286 1200 map_tail->offs = ((char *) de - base)>>2;
ef2b02d3 1201 map_tail->size = le16_to_cpu(de->rec_len);
ac27a0ec
DK
1202 count++;
1203 cond_resched();
1204 }
1205 /* XXX: do we need to check rec_len == 0 case? -Chris */
3d0518f4 1206 de = ext4_next_entry(de, blocksize);
ac27a0ec
DK
1207 }
1208 return count;
1209}
1210
ef2b02d3 1211/* Sort map by hash value */
ac27a0ec
DK
1212static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1213{
63f57933
AM
1214 struct dx_map_entry *p, *q, *top = map + count - 1;
1215 int more;
1216 /* Combsort until bubble sort doesn't suck */
1217 while (count > 2) {
1218 count = count*10/13;
1219 if (count - 9 < 2) /* 9, 10 -> 11 */
1220 count = 11;
1221 for (p = top, q = p - count; q >= map; p--, q--)
1222 if (p->hash < q->hash)
1223 swap(*p, *q);
1224 }
1225 /* Garden variety bubble sort */
1226 do {
1227 more = 0;
1228 q = top;
1229 while (q-- > map) {
1230 if (q[1].hash >= q[0].hash)
ac27a0ec 1231 continue;
63f57933
AM
1232 swap(*(q+1), *q);
1233 more = 1;
ac27a0ec
DK
1234 }
1235 } while(more);
1236}
1237
725d26d3 1238static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
ac27a0ec
DK
1239{
1240 struct dx_entry *entries = frame->entries;
1241 struct dx_entry *old = frame->at, *new = old + 1;
1242 int count = dx_get_count(entries);
1243
1244 assert(count < dx_get_limit(entries));
1245 assert(old < entries + count);
1246 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1247 dx_set_hash(new, hash);
1248 dx_set_block(new, block);
1249 dx_set_count(entries, count + 1);
1250}
ac27a0ec 1251
ac27a0ec 1252/*
d9b9f8d5 1253 * Test whether a directory entry matches the filename being searched for.
ac27a0ec 1254 *
d9b9f8d5 1255 * Return: %true if the directory entry matches, otherwise %false.
ac27a0ec 1256 */
d9b9f8d5
EB
1257static inline bool ext4_match(const struct ext4_filename *fname,
1258 const struct ext4_dir_entry_2 *de)
ac27a0ec 1259{
067d1023 1260 struct fscrypt_name f;
1f3862b5 1261
ac27a0ec 1262 if (!de->inode)
d9b9f8d5 1263 return false;
1f3862b5 1264
067d1023
EB
1265 f.usr_fname = fname->usr_fname;
1266 f.disk_name = fname->disk_name;
1f3862b5 1267#ifdef CONFIG_EXT4_FS_ENCRYPTION
067d1023 1268 f.crypto_buf = fname->crypto_buf;
1f3862b5 1269#endif
067d1023 1270 return fscrypt_match_name(&f, de->name, de->name_len);
ac27a0ec
DK
1271}
1272
1273/*
1274 * Returns 0 if not found, -1 on failure, and 1 on success
1275 */
5b643f9c
TT
1276int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1277 struct inode *dir, struct ext4_filename *fname,
5b643f9c 1278 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
ac27a0ec 1279{
617ba13b 1280 struct ext4_dir_entry_2 * de;
ac27a0ec
DK
1281 char * dlimit;
1282 int de_len;
1f3862b5 1283
7335cd3b
TM
1284 de = (struct ext4_dir_entry_2 *)search_buf;
1285 dlimit = search_buf + buf_size;
ac27a0ec
DK
1286 while ((char *) de < dlimit) {
1287 /* this code is executed quadratically often */
1288 /* do minimal checking `by hand' */
d9b9f8d5
EB
1289 if ((char *) de + de->name_len <= dlimit &&
1290 ext4_match(fname, de)) {
1291 /* found a match - just to be sure, do
1292 * a full check */
1293 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1294 bh->b_size, offset))
1295 return -1;
1296 *res_dir = de;
1297 return 1;
ac27a0ec
DK
1298 }
1299 /* prevent looping on a bad block */
3d0518f4
WY
1300 de_len = ext4_rec_len_from_disk(de->rec_len,
1301 dir->i_sb->s_blocksize);
d9b9f8d5
EB
1302 if (de_len <= 0)
1303 return -1;
ac27a0ec 1304 offset += de_len;
617ba13b 1305 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
ac27a0ec 1306 }
d9b9f8d5 1307 return 0;
ac27a0ec
DK
1308}
1309
c6af8803
DW
1310static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1311 struct ext4_dir_entry *de)
1312{
1313 struct super_block *sb = dir->i_sb;
1314
1315 if (!is_dx(dir))
1316 return 0;
1317 if (block == 0)
1318 return 1;
1319 if (de->inode == 0 &&
1320 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1321 sb->s_blocksize)
1322 return 1;
1323 return 0;
1324}
ac27a0ec
DK
1325
1326/*
617ba13b 1327 * ext4_find_entry()
ac27a0ec
DK
1328 *
1329 * finds an entry in the specified directory with the wanted name. It
1330 * returns the cache buffer in which the entry was found, and the entry
1331 * itself (as a parameter - res_dir). It does NOT read the inode of the
1332 * entry - you'll have to do that yourself if you want to.
1333 *
1334 * The returned buffer_head has ->b_count elevated. The caller is expected
1335 * to brelse() it when appropriate.
1336 */
f702ba0f
TT
1337static struct buffer_head * ext4_find_entry (struct inode *dir,
1338 const struct qstr *d_name,
32f7f22c
TM
1339 struct ext4_dir_entry_2 **res_dir,
1340 int *inlined)
ac27a0ec 1341{
af5bc92d
TT
1342 struct super_block *sb;
1343 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1344 struct buffer_head *bh, *ret = NULL;
9699d4f9 1345 ext4_lblk_t start, block;
8941ec8b 1346 const u8 *name = d_name->name;
9699d4f9 1347 size_t ra_max = 0; /* Number of bh's in the readahead
ac27a0ec 1348 buffer, bh_use[] */
9699d4f9 1349 size_t ra_ptr = 0; /* Current index into readahead
ac27a0ec 1350 buffer */
725d26d3 1351 ext4_lblk_t nblocks;
5b643f9c
TT
1352 int i, namelen, retval;
1353 struct ext4_filename fname;
ac27a0ec
DK
1354
1355 *res_dir = NULL;
1356 sb = dir->i_sb;
f702ba0f 1357 namelen = d_name->len;
617ba13b 1358 if (namelen > EXT4_NAME_LEN)
ac27a0ec 1359 return NULL;
e8e948e7 1360
5b643f9c 1361 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
54475f53
EB
1362 if (retval == -ENOENT)
1363 return NULL;
5b643f9c
TT
1364 if (retval)
1365 return ERR_PTR(retval);
1366
e8e948e7
TM
1367 if (ext4_has_inline_data(dir)) {
1368 int has_inline_data = 1;
d6b97550 1369 ret = ext4_find_inline_entry(dir, &fname, res_dir,
e8e948e7 1370 &has_inline_data);
32f7f22c
TM
1371 if (has_inline_data) {
1372 if (inlined)
1373 *inlined = 1;
5b643f9c 1374 goto cleanup_and_exit;
32f7f22c 1375 }
e8e948e7
TM
1376 }
1377
8941ec8b 1378 if ((namelen <= 2) && (name[0] == '.') &&
6d5c3aa8 1379 (name[1] == '.' || name[1] == '\0')) {
8941ec8b
TT
1380 /*
1381 * "." or ".." will only be in the first block
1382 * NFS may look up ".."; "." should be handled by the VFS
1383 */
1384 block = start = 0;
1385 nblocks = 1;
1386 goto restart;
1387 }
ac27a0ec 1388 if (is_dx(dir)) {
5b643f9c 1389 ret = ext4_dx_find_entry(dir, &fname, res_dir);
ac27a0ec
DK
1390 /*
1391 * On success, or if the error was file not found,
1392 * return. Otherwise, fall back to doing a search the
1393 * old fashioned way.
1394 */
5b643f9c
TT
1395 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1396 goto cleanup_and_exit;
4776004f
TT
1397 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1398 "falling back\n"));
ac27a0ec 1399 }
617ba13b
MC
1400 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1401 start = EXT4_I(dir)->i_dir_start_lookup;
ac27a0ec
DK
1402 if (start >= nblocks)
1403 start = 0;
1404 block = start;
1405restart:
1406 do {
1407 /*
1408 * We deal with the read-ahead logic here.
1409 */
1410 if (ra_ptr >= ra_max) {
1411 /* Refill the readahead buffer */
1412 ra_ptr = 0;
9699d4f9
TE
1413 if (block < start)
1414 ra_max = start - block;
1415 else
1416 ra_max = nblocks - block;
1417 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1418 retval = ext4_bread_batch(dir, block, ra_max,
1419 false /* wait */, bh_use);
1420 if (retval) {
1421 ret = ERR_PTR(retval);
1422 ra_max = 0;
1423 goto cleanup_and_exit;
ac27a0ec
DK
1424 }
1425 }
1426 if ((bh = bh_use[ra_ptr++]) == NULL)
1427 goto next;
1428 wait_on_buffer(bh);
1429 if (!buffer_uptodate(bh)) {
24676da4
TT
1430 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1431 (unsigned long) block);
ac27a0ec 1432 brelse(bh);
6febe6f2
KK
1433 ret = ERR_PTR(-EIO);
1434 goto cleanup_and_exit;
ac27a0ec 1435 }
b0336e8d 1436 if (!buffer_verified(bh) &&
c6af8803
DW
1437 !is_dx_internal_node(dir, block,
1438 (struct ext4_dir_entry *)bh->b_data) &&
b0336e8d
DW
1439 !ext4_dirent_csum_verify(dir,
1440 (struct ext4_dir_entry *)bh->b_data)) {
1441 EXT4_ERROR_INODE(dir, "checksumming directory "
1442 "block %lu", (unsigned long)block);
1443 brelse(bh);
bdddf342
TT
1444 ret = ERR_PTR(-EFSBADCRC);
1445 goto cleanup_and_exit;
b0336e8d
DW
1446 }
1447 set_buffer_verified(bh);
d6b97550 1448 i = search_dirblock(bh, dir, &fname,
617ba13b 1449 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
ac27a0ec 1450 if (i == 1) {
617ba13b 1451 EXT4_I(dir)->i_dir_start_lookup = block;
ac27a0ec
DK
1452 ret = bh;
1453 goto cleanup_and_exit;
1454 } else {
1455 brelse(bh);
1456 if (i < 0)
1457 goto cleanup_and_exit;
1458 }
1459 next:
1460 if (++block >= nblocks)
1461 block = 0;
1462 } while (block != start);
1463
1464 /*
1465 * If the directory has grown while we were searching, then
1466 * search the last part of the directory before giving up.
1467 */
1468 block = nblocks;
617ba13b 1469 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
ac27a0ec
DK
1470 if (block < nblocks) {
1471 start = 0;
1472 goto restart;
1473 }
1474
1475cleanup_and_exit:
1476 /* Clean up the read-ahead blocks */
1477 for (; ra_ptr < ra_max; ra_ptr++)
af5bc92d 1478 brelse(bh_use[ra_ptr]);
5b643f9c 1479 ext4_fname_free_filename(&fname);
ac27a0ec
DK
1480 return ret;
1481}
1482
5b643f9c
TT
1483static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1484 struct ext4_filename *fname,
1485 struct ext4_dir_entry_2 **res_dir)
ac27a0ec 1486{
8941ec8b 1487 struct super_block * sb = dir->i_sb;
e08ac99f 1488 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 1489 struct buffer_head *bh;
725d26d3 1490 ext4_lblk_t block;
ac27a0ec 1491 int retval;
ac27a0ec 1492
1f3862b5
MH
1493#ifdef CONFIG_EXT4_FS_ENCRYPTION
1494 *res_dir = NULL;
1495#endif
5b643f9c 1496 frame = dx_probe(fname, dir, NULL, frames);
dd73b5d5
TT
1497 if (IS_ERR(frame))
1498 return (struct buffer_head *) frame;
ac27a0ec
DK
1499 do {
1500 block = dx_get_block(frame->at);
dc6982ff 1501 bh = ext4_read_dirblock(dir, block, DIRENT);
537d8f93 1502 if (IS_ERR(bh))
b0336e8d 1503 goto errout;
537d8f93 1504
d6b97550 1505 retval = search_dirblock(bh, dir, fname,
7845c049
TT
1506 block << EXT4_BLOCK_SIZE_BITS(sb),
1507 res_dir);
537d8f93
TT
1508 if (retval == 1)
1509 goto success;
af5bc92d 1510 brelse(bh);
7845c049 1511 if (retval == -1) {
537d8f93 1512 bh = ERR_PTR(ERR_BAD_DX_DIR);
7845c049
TT
1513 goto errout;
1514 }
1515
ac27a0ec 1516 /* Check to see if we should continue to search */
5b643f9c 1517 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
ac27a0ec
DK
1518 frames, NULL);
1519 if (retval < 0) {
b03a2f7e
AD
1520 ext4_warning_inode(dir,
1521 "error %d reading directory index block",
1522 retval);
537d8f93 1523 bh = ERR_PTR(retval);
ac27a0ec
DK
1524 goto errout;
1525 }
1526 } while (retval == 1);
1527
537d8f93 1528 bh = NULL;
ac27a0ec 1529errout:
d6b97550 1530 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
537d8f93
TT
1531success:
1532 dx_release(frames);
1533 return bh;
ac27a0ec 1534}
ac27a0ec 1535
00cd8dd3 1536static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
ac27a0ec 1537{
af5bc92d
TT
1538 struct inode *inode;
1539 struct ext4_dir_entry_2 *de;
1540 struct buffer_head *bh;
ac27a0ec 1541
a7550b30
JK
1542 if (ext4_encrypted_inode(dir)) {
1543 int res = fscrypt_get_encryption_info(dir);
28b4c263
TT
1544
1545 /*
a7550b30 1546 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
28b4c263 1547 * created while the directory was encrypted and we
a7550b30 1548 * have access to the key.
28b4c263 1549 */
a7550b30
JK
1550 if (fscrypt_has_encryption_key(dir))
1551 fscrypt_set_encrypted_dentry(dentry);
1552 fscrypt_set_d_op(dentry);
1553 if (res && res != -ENOKEY)
1554 return ERR_PTR(res);
1555 }
28b4c263 1556
a7550b30
JK
1557 if (dentry->d_name.len > EXT4_NAME_LEN)
1558 return ERR_PTR(-ENAMETOOLONG);
ac27a0ec 1559
32f7f22c 1560 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
36de9286
TT
1561 if (IS_ERR(bh))
1562 return (struct dentry *) bh;
ac27a0ec
DK
1563 inode = NULL;
1564 if (bh) {
498e5f24 1565 __u32 ino = le32_to_cpu(de->inode);
af5bc92d 1566 brelse(bh);
617ba13b 1567 if (!ext4_valid_inum(dir->i_sb, ino)) {
24676da4 1568 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
6a797d27 1569 return ERR_PTR(-EFSCORRUPTED);
a6c15c2b 1570 }
7e936b73 1571 if (unlikely(ino == dir->i_ino)) {
a34e15cc
DH
1572 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1573 dentry);
6a797d27 1574 return ERR_PTR(-EFSCORRUPTED);
7e936b73 1575 }
f4bb2981 1576 inode = ext4_iget_normal(dir->i_sb, ino);
a9049376
AV
1577 if (inode == ERR_PTR(-ESTALE)) {
1578 EXT4_ERROR_INODE(dir,
1579 "deleted inode referenced: %u",
1580 ino);
6a797d27 1581 return ERR_PTR(-EFSCORRUPTED);
e6f009b0 1582 }
d9cdc903 1583 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
ff978b09 1584 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
a7550b30 1585 !fscrypt_has_permitted_context(dir, inode)) {
d9cdc903 1586 ext4_warning(inode->i_sb,
8d2ae1cb 1587 "Inconsistent encryption contexts: %lu/%lu",
8c68084b 1588 dir->i_ino, inode->i_ino);
dd01b690 1589 iput(inode);
d9cdc903
TT
1590 return ERR_PTR(-EPERM);
1591 }
ac27a0ec
DK
1592 }
1593 return d_splice_alias(inode, dentry);
1594}
1595
1596
617ba13b 1597struct dentry *ext4_get_parent(struct dentry *child)
ac27a0ec 1598{
498e5f24 1599 __u32 ino;
26fe5750 1600 static const struct qstr dotdot = QSTR_INIT("..", 2);
617ba13b 1601 struct ext4_dir_entry_2 * de;
ac27a0ec
DK
1602 struct buffer_head *bh;
1603
2b0143b5 1604 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
36de9286
TT
1605 if (IS_ERR(bh))
1606 return (struct dentry *) bh;
ac27a0ec
DK
1607 if (!bh)
1608 return ERR_PTR(-ENOENT);
1609 ino = le32_to_cpu(de->inode);
1610 brelse(bh);
1611
fc64005c 1612 if (!ext4_valid_inum(child->d_sb, ino)) {
2b0143b5 1613 EXT4_ERROR_INODE(d_inode(child),
24676da4 1614 "bad parent inode number: %u", ino);
6a797d27 1615 return ERR_PTR(-EFSCORRUPTED);
a6c15c2b
VA
1616 }
1617
fc64005c 1618 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
ac27a0ec
DK
1619}
1620
ef2b02d3
ES
1621/*
1622 * Move count entries from end of map between two memory locations.
1623 * Returns pointer to last entry moved.
1624 */
617ba13b 1625static struct ext4_dir_entry_2 *
3d0518f4
WY
1626dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1627 unsigned blocksize)
ac27a0ec
DK
1628{
1629 unsigned rec_len = 0;
1630
1631 while (count--) {
60e6679e 1632 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
9aee2286 1633 (from + (map->offs<<2));
617ba13b 1634 rec_len = EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec 1635 memcpy (to, de, rec_len);
617ba13b 1636 ((struct ext4_dir_entry_2 *) to)->rec_len =
3d0518f4 1637 ext4_rec_len_to_disk(rec_len, blocksize);
ac27a0ec
DK
1638 de->inode = 0;
1639 map++;
1640 to += rec_len;
1641 }
617ba13b 1642 return (struct ext4_dir_entry_2 *) (to - rec_len);
ac27a0ec
DK
1643}
1644
ef2b02d3
ES
1645/*
1646 * Compact each dir entry in the range to the minimal rec_len.
1647 * Returns pointer to last entry in range.
1648 */
8bad4597 1649static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
ac27a0ec 1650{
617ba13b 1651 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
ac27a0ec
DK
1652 unsigned rec_len = 0;
1653
1654 prev = to = de;
8bad4597 1655 while ((char*)de < base + blocksize) {
3d0518f4 1656 next = ext4_next_entry(de, blocksize);
ac27a0ec 1657 if (de->inode && de->name_len) {
617ba13b 1658 rec_len = EXT4_DIR_REC_LEN(de->name_len);
ac27a0ec
DK
1659 if (de > to)
1660 memmove(to, de, rec_len);
3d0518f4 1661 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
ac27a0ec 1662 prev = to;
617ba13b 1663 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
ac27a0ec
DK
1664 }
1665 de = next;
1666 }
1667 return prev;
1668}
1669
ef2b02d3
ES
1670/*
1671 * Split a full leaf block to make room for a new dir entry.
1672 * Allocate a new block, and move entries so that they are approx. equally full.
1673 * Returns pointer to de in block into which the new entry will be inserted.
1674 */
617ba13b 1675static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
ac27a0ec 1676 struct buffer_head **bh,struct dx_frame *frame,
f8b3b59d 1677 struct dx_hash_info *hinfo)
ac27a0ec
DK
1678{
1679 unsigned blocksize = dir->i_sb->s_blocksize;
1680 unsigned count, continued;
1681 struct buffer_head *bh2;
725d26d3 1682 ext4_lblk_t newblock;
ac27a0ec
DK
1683 u32 hash2;
1684 struct dx_map_entry *map;
1685 char *data1 = (*bh)->b_data, *data2;
59e315b4 1686 unsigned split, move, size;
617ba13b 1687 struct ext4_dir_entry_2 *de = NULL, *de2;
b0336e8d
DW
1688 struct ext4_dir_entry_tail *t;
1689 int csum_size = 0;
59e315b4 1690 int err = 0, i;
ac27a0ec 1691
9aa5d32b 1692 if (ext4_has_metadata_csum(dir->i_sb))
b0336e8d
DW
1693 csum_size = sizeof(struct ext4_dir_entry_tail);
1694
0f70b406
TT
1695 bh2 = ext4_append(handle, dir, &newblock);
1696 if (IS_ERR(bh2)) {
ac27a0ec
DK
1697 brelse(*bh);
1698 *bh = NULL;
f8b3b59d 1699 return (struct ext4_dir_entry_2 *) bh2;
ac27a0ec
DK
1700 }
1701
1702 BUFFER_TRACE(*bh, "get_write_access");
617ba13b 1703 err = ext4_journal_get_write_access(handle, *bh);
fedee54d
DM
1704 if (err)
1705 goto journal_error;
1706
ac27a0ec 1707 BUFFER_TRACE(frame->bh, "get_write_access");
617ba13b 1708 err = ext4_journal_get_write_access(handle, frame->bh);
ac27a0ec
DK
1709 if (err)
1710 goto journal_error;
1711
1712 data2 = bh2->b_data;
1713
1714 /* create map in the end of data2 block */
1715 map = (struct dx_map_entry *) (data2 + blocksize);
1f3862b5 1716 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
ac27a0ec
DK
1717 blocksize, hinfo, map);
1718 map -= count;
af5bc92d 1719 dx_sort_map(map, count);
ef2b02d3
ES
1720 /* Split the existing block in the middle, size-wise */
1721 size = 0;
1722 move = 0;
1723 for (i = count-1; i >= 0; i--) {
1724 /* is more than half of this entry in 2nd half of the block? */
1725 if (size + map[i].size/2 > blocksize/2)
1726 break;
1727 size += map[i].size;
1728 move++;
1729 }
1730 /* map index at which we will split */
1731 split = count - move;
ac27a0ec
DK
1732 hash2 = map[split].hash;
1733 continued = hash2 == map[split - 1].hash;
725d26d3
AK
1734 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1735 (unsigned long)dx_get_block(frame->at),
1736 hash2, split, count-split));
ac27a0ec
DK
1737
1738 /* Fancy dance to stay within two buffers */
1f3862b5
MH
1739 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1740 blocksize);
af5bc92d 1741 de = dx_pack_dirents(data1, blocksize);
b0336e8d
DW
1742 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1743 (char *) de,
3d0518f4 1744 blocksize);
b0336e8d
DW
1745 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1746 (char *) de2,
3d0518f4 1747 blocksize);
b0336e8d
DW
1748 if (csum_size) {
1749 t = EXT4_DIRENT_TAIL(data2, blocksize);
1750 initialize_dirent_tail(t, blocksize);
1751
1752 t = EXT4_DIRENT_TAIL(data1, blocksize);
1753 initialize_dirent_tail(t, blocksize);
1754 }
1755
b3098486
MH
1756 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1757 blocksize, 1));
1758 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1759 blocksize, 1));
ac27a0ec
DK
1760
1761 /* Which block gets the new entry? */
f8b3b59d 1762 if (hinfo->hash >= hash2) {
ac27a0ec
DK
1763 swap(*bh, bh2);
1764 de = de2;
1765 }
af5bc92d 1766 dx_insert_block(frame, hash2 + continued, newblock);
b0336e8d 1767 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
ac27a0ec
DK
1768 if (err)
1769 goto journal_error;
dbe89444 1770 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
ac27a0ec
DK
1771 if (err)
1772 goto journal_error;
af5bc92d
TT
1773 brelse(bh2);
1774 dxtrace(dx_show_index("frame", frame->entries));
ac27a0ec 1775 return de;
fedee54d
DM
1776
1777journal_error:
1778 brelse(*bh);
1779 brelse(bh2);
1780 *bh = NULL;
1781 ext4_std_error(dir->i_sb, err);
f8b3b59d 1782 return ERR_PTR(err);
ac27a0ec 1783}
ac27a0ec 1784
978fef91
TM
1785int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1786 struct buffer_head *bh,
1787 void *buf, int buf_size,
5b643f9c 1788 struct ext4_filename *fname,
978fef91
TM
1789 struct ext4_dir_entry_2 **dest_de)
1790{
1791 struct ext4_dir_entry_2 *de;
5b643f9c 1792 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
978fef91
TM
1793 int nlen, rlen;
1794 unsigned int offset = 0;
1795 char *top;
1f3862b5 1796
978fef91
TM
1797 de = (struct ext4_dir_entry_2 *)buf;
1798 top = buf + buf_size - reclen;
1799 while ((char *) de <= top) {
1800 if (ext4_check_dir_entry(dir, NULL, de, bh,
d9b9f8d5
EB
1801 buf, buf_size, offset))
1802 return -EFSCORRUPTED;
1803 if (ext4_match(fname, de))
1804 return -EEXIST;
978fef91
TM
1805 nlen = EXT4_DIR_REC_LEN(de->name_len);
1806 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1807 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1808 break;
1809 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1810 offset += rlen;
1811 }
1f3862b5 1812 if ((char *) de > top)
d9b9f8d5
EB
1813 return -ENOSPC;
1814
1815 *dest_de = de;
1816 return 0;
978fef91
TM
1817}
1818
1bc0af60
EB
1819void ext4_insert_dentry(struct inode *inode,
1820 struct ext4_dir_entry_2 *de,
1821 int buf_size,
1822 struct ext4_filename *fname)
978fef91
TM
1823{
1824
1825 int nlen, rlen;
1826
1827 nlen = EXT4_DIR_REC_LEN(de->name_len);
1828 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1829 if (de->inode) {
1830 struct ext4_dir_entry_2 *de1 =
4bdfc873 1831 (struct ext4_dir_entry_2 *)((char *)de + nlen);
978fef91
TM
1832 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1833 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1834 de = de1;
1835 }
1836 de->file_type = EXT4_FT_UNKNOWN;
1837 de->inode = cpu_to_le32(inode->i_ino);
1838 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
5b643f9c
TT
1839 de->name_len = fname_len(fname);
1840 memcpy(de->name, fname_name(fname), fname_len(fname));
978fef91 1841}
4bdfc873 1842
ac27a0ec
DK
1843/*
1844 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1845 * it points to a directory entry which is guaranteed to be large
1846 * enough for new directory entry. If de is NULL, then
1847 * add_dirent_to_buf will attempt search the directory block for
1848 * space. It will return -ENOSPC if no space is available, and -EIO
1849 * and -EEXIST if directory entry already exists.
ac27a0ec 1850 */
5b643f9c
TT
1851static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1852 struct inode *dir,
617ba13b 1853 struct inode *inode, struct ext4_dir_entry_2 *de,
af5bc92d 1854 struct buffer_head *bh)
ac27a0ec 1855{
3d0518f4 1856 unsigned int blocksize = dir->i_sb->s_blocksize;
b0336e8d 1857 int csum_size = 0;
978fef91 1858 int err;
b0336e8d 1859
9aa5d32b 1860 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 1861 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec 1862
ac27a0ec 1863 if (!de) {
5b643f9c
TT
1864 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1865 blocksize - csum_size, fname, &de);
978fef91
TM
1866 if (err)
1867 return err;
ac27a0ec
DK
1868 }
1869 BUFFER_TRACE(bh, "get_write_access");
617ba13b 1870 err = ext4_journal_get_write_access(handle, bh);
ac27a0ec 1871 if (err) {
617ba13b 1872 ext4_std_error(dir->i_sb, err);
ac27a0ec
DK
1873 return err;
1874 }
1875
1bc0af60
EB
1876 /* By now the buffer is marked for journaling */
1877 ext4_insert_dentry(inode, de, blocksize, fname);
978fef91 1878
ac27a0ec
DK
1879 /*
1880 * XXX shouldn't update any times until successful
1881 * completion of syscall, but too many callers depend
1882 * on this.
1883 *
1884 * XXX similarly, too many callers depend on
617ba13b 1885 * ext4_new_inode() setting the times, but error
ac27a0ec
DK
1886 * recovery deletes the inode, so the worst that can
1887 * happen is that the times are slightly out of date
1888 * and/or different from the directory change time.
1889 */
eeca7ea1 1890 dir->i_mtime = dir->i_ctime = current_time(dir);
617ba13b 1891 ext4_update_dx_flag(dir);
e08ac99f 1892 inode_inc_iversion(dir);
617ba13b 1893 ext4_mark_inode_dirty(handle, dir);
0390131b 1894 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
b0336e8d 1895 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
ac27a0ec 1896 if (err)
617ba13b 1897 ext4_std_error(dir->i_sb, err);
ac27a0ec
DK
1898 return 0;
1899}
1900
ac27a0ec
DK
1901/*
1902 * This converts a one block unindexed directory to a 3 block indexed
1903 * directory, and adds the dentry to the indexed directory.
1904 */
5b643f9c 1905static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
56a04915 1906 struct inode *dir,
ac27a0ec
DK
1907 struct inode *inode, struct buffer_head *bh)
1908{
ac27a0ec
DK
1909 struct buffer_head *bh2;
1910 struct dx_root *root;
e08ac99f 1911 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 1912 struct dx_entry *entries;
617ba13b 1913 struct ext4_dir_entry_2 *de, *de2;
b0336e8d 1914 struct ext4_dir_entry_tail *t;
ac27a0ec
DK
1915 char *data1, *top;
1916 unsigned len;
1917 int retval;
1918 unsigned blocksize;
725d26d3 1919 ext4_lblk_t block;
ac27a0ec 1920 struct fake_dirent *fde;
4bdfc873
MH
1921 int csum_size = 0;
1922
9aa5d32b 1923 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 1924 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec
DK
1925
1926 blocksize = dir->i_sb->s_blocksize;
e6b8bc09 1927 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
5d601255 1928 BUFFER_TRACE(bh, "get_write_access");
617ba13b 1929 retval = ext4_journal_get_write_access(handle, bh);
ac27a0ec 1930 if (retval) {
617ba13b 1931 ext4_std_error(dir->i_sb, retval);
ac27a0ec
DK
1932 brelse(bh);
1933 return retval;
1934 }
1935 root = (struct dx_root *) bh->b_data;
1936
e6b8bc09
TT
1937 /* The 0th block becomes the root, move the dirents out */
1938 fde = &root->dotdot;
1939 de = (struct ext4_dir_entry_2 *)((char *)fde +
3d0518f4 1940 ext4_rec_len_from_disk(fde->rec_len, blocksize));
e6b8bc09 1941 if ((char *) de >= (((char *) root) + blocksize)) {
24676da4 1942 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
e6b8bc09 1943 brelse(bh);
6a797d27 1944 return -EFSCORRUPTED;
e6b8bc09 1945 }
b0336e8d 1946 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
e6b8bc09
TT
1947
1948 /* Allocate new block for the 0th block's dirents */
0f70b406
TT
1949 bh2 = ext4_append(handle, dir, &block);
1950 if (IS_ERR(bh2)) {
ac27a0ec 1951 brelse(bh);
0f70b406 1952 return PTR_ERR(bh2);
ac27a0ec 1953 }
12e9b892 1954 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
ac27a0ec
DK
1955 data1 = bh2->b_data;
1956
ac27a0ec 1957 memcpy (data1, de, len);
617ba13b 1958 de = (struct ext4_dir_entry_2 *) data1;
ac27a0ec 1959 top = data1 + len;
3d0518f4 1960 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
ac27a0ec 1961 de = de2;
b0336e8d
DW
1962 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1963 (char *) de,
3d0518f4 1964 blocksize);
b0336e8d
DW
1965
1966 if (csum_size) {
1967 t = EXT4_DIRENT_TAIL(data1, blocksize);
1968 initialize_dirent_tail(t, blocksize);
1969 }
1970
ac27a0ec 1971 /* Initialize the root; the dot dirents already exist */
617ba13b 1972 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
3d0518f4
WY
1973 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1974 blocksize);
ac27a0ec
DK
1975 memset (&root->info, 0, sizeof(root->info));
1976 root->info.info_length = sizeof(root->info);
617ba13b 1977 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
ac27a0ec 1978 entries = root->entries;
af5bc92d
TT
1979 dx_set_block(entries, 1);
1980 dx_set_count(entries, 1);
1981 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
ac27a0ec
DK
1982
1983 /* Initialize as for dx_probe */
5b643f9c
TT
1984 fname->hinfo.hash_version = root->info.hash_version;
1985 if (fname->hinfo.hash_version <= DX_HASH_TEA)
1986 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1987 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1988 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1989
6050d47a 1990 memset(frames, 0, sizeof(frames));
ac27a0ec
DK
1991 frame = frames;
1992 frame->entries = entries;
1993 frame->at = entries;
1994 frame->bh = bh;
6976a6f2 1995
6050d47a
JK
1996 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1997 if (retval)
1998 goto out_frames;
e81d4477 1999 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
6050d47a
JK
2000 if (retval)
2001 goto out_frames;
6976a6f2 2002
e81d4477 2003 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
f8b3b59d 2004 if (IS_ERR(de)) {
6050d47a
JK
2005 retval = PTR_ERR(de);
2006 goto out_frames;
7ad8e4e6 2007 }
ac27a0ec 2008
e81d4477 2009 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
6050d47a
JK
2010out_frames:
2011 /*
2012 * Even if the block split failed, we have to properly write
2013 * out all the changes we did so far. Otherwise we can end up
2014 * with corrupted filesystem.
2015 */
e81d4477 2016 if (retval)
2017 ext4_mark_inode_dirty(handle, dir);
6050d47a 2018 dx_release(frames);
e81d4477 2019 brelse(bh2);
6050d47a 2020 return retval;
ac27a0ec 2021}
ac27a0ec
DK
2022
2023/*
617ba13b 2024 * ext4_add_entry()
ac27a0ec
DK
2025 *
2026 * adds a file entry to the specified directory, using the same
617ba13b 2027 * semantics as ext4_find_entry(). It returns NULL if it failed.
ac27a0ec
DK
2028 *
2029 * NOTE!! The inode part of 'de' is left at 0 - which means you
2030 * may not sleep between calling this and putting something into
2031 * the entry, as someone else might have used it while you slept.
2032 */
af5bc92d
TT
2033static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2034 struct inode *inode)
ac27a0ec 2035{
2b0143b5 2036 struct inode *dir = d_inode(dentry->d_parent);
e12fb972 2037 struct buffer_head *bh = NULL;
617ba13b 2038 struct ext4_dir_entry_2 *de;
b0336e8d 2039 struct ext4_dir_entry_tail *t;
af5bc92d 2040 struct super_block *sb;
5b643f9c 2041 struct ext4_filename fname;
ac27a0ec 2042 int retval;
ac27a0ec 2043 int dx_fallback=0;
ac27a0ec 2044 unsigned blocksize;
725d26d3 2045 ext4_lblk_t block, blocks;
b0336e8d
DW
2046 int csum_size = 0;
2047
9aa5d32b 2048 if (ext4_has_metadata_csum(inode->i_sb))
b0336e8d 2049 csum_size = sizeof(struct ext4_dir_entry_tail);
ac27a0ec
DK
2050
2051 sb = dir->i_sb;
2052 blocksize = sb->s_blocksize;
2053 if (!dentry->d_name.len)
2054 return -EINVAL;
3c47d541 2055
5b643f9c
TT
2056 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2057 if (retval)
2058 return retval;
2059
3c47d541 2060 if (ext4_has_inline_data(dir)) {
56a04915 2061 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
3c47d541 2062 if (retval < 0)
5b643f9c 2063 goto out;
3c47d541
TM
2064 if (retval == 1) {
2065 retval = 0;
e12fb972 2066 goto out;
3c47d541
TM
2067 }
2068 }
2069
ac27a0ec 2070 if (is_dx(dir)) {
56a04915 2071 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
ac27a0ec 2072 if (!retval || (retval != ERR_BAD_DX_DIR))
e12fb972 2073 goto out;
12e9b892 2074 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
ac27a0ec 2075 dx_fallback++;
617ba13b 2076 ext4_mark_inode_dirty(handle, dir);
ac27a0ec 2077 }
ac27a0ec 2078 blocks = dir->i_size >> sb->s_blocksize_bits;
498e5f24 2079 for (block = 0; block < blocks; block++) {
dc6982ff 2080 bh = ext4_read_dirblock(dir, block, DIRENT);
5b643f9c
TT
2081 if (IS_ERR(bh)) {
2082 retval = PTR_ERR(bh);
2083 bh = NULL;
2084 goto out;
2085 }
2086 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2087 NULL, bh);
e12fb972
LC
2088 if (retval != -ENOSPC)
2089 goto out;
ac27a0ec 2090
ac27a0ec 2091 if (blocks == 1 && !dx_fallback &&
e2b911c5 2092 ext4_has_feature_dir_index(sb)) {
56a04915 2093 retval = make_indexed_dir(handle, &fname, dir,
5b643f9c 2094 inode, bh);
e12fb972
LC
2095 bh = NULL; /* make_indexed_dir releases bh */
2096 goto out;
2097 }
ac27a0ec
DK
2098 brelse(bh);
2099 }
0f70b406 2100 bh = ext4_append(handle, dir, &block);
5b643f9c
TT
2101 if (IS_ERR(bh)) {
2102 retval = PTR_ERR(bh);
2103 bh = NULL;
2104 goto out;
2105 }
617ba13b 2106 de = (struct ext4_dir_entry_2 *) bh->b_data;
ac27a0ec 2107 de->inode = 0;
b0336e8d
DW
2108 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2109
2110 if (csum_size) {
2111 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2112 initialize_dirent_tail(t, blocksize);
2113 }
2114
5b643f9c 2115 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
e12fb972 2116out:
5b643f9c 2117 ext4_fname_free_filename(&fname);
2de770a4 2118 brelse(bh);
14ece102
FM
2119 if (retval == 0)
2120 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2de770a4 2121 return retval;
ac27a0ec
DK
2122}
2123
ac27a0ec
DK
2124/*
2125 * Returns 0 for success, or a negative error value
2126 */
5b643f9c 2127static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
56a04915 2128 struct inode *dir, struct inode *inode)
ac27a0ec 2129{
e08ac99f 2130 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
ac27a0ec 2131 struct dx_entry *entries, *at;
af5bc92d 2132 struct buffer_head *bh;
af5bc92d 2133 struct super_block *sb = dir->i_sb;
617ba13b 2134 struct ext4_dir_entry_2 *de;
e08ac99f 2135 int restart;
ac27a0ec
DK
2136 int err;
2137
e08ac99f
AB
2138again:
2139 restart = 0;
5b643f9c 2140 frame = dx_probe(fname, dir, NULL, frames);
dd73b5d5
TT
2141 if (IS_ERR(frame))
2142 return PTR_ERR(frame);
ac27a0ec
DK
2143 entries = frame->entries;
2144 at = frame->at;
dc6982ff
TT
2145 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2146 if (IS_ERR(bh)) {
2147 err = PTR_ERR(bh);
2148 bh = NULL;
ac27a0ec 2149 goto cleanup;
6d1ab10e 2150 }
ac27a0ec
DK
2151
2152 BUFFER_TRACE(bh, "get_write_access");
617ba13b 2153 err = ext4_journal_get_write_access(handle, bh);
ac27a0ec
DK
2154 if (err)
2155 goto journal_error;
2156
5b643f9c 2157 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2de770a4 2158 if (err != -ENOSPC)
ac27a0ec 2159 goto cleanup;
ac27a0ec 2160
e08ac99f 2161 err = 0;
ac27a0ec 2162 /* Block full, should compress but for now just split */
4776004f 2163 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
ac27a0ec
DK
2164 dx_get_count(entries), dx_get_limit(entries)));
2165 /* Need to split index? */
2166 if (dx_get_count(entries) == dx_get_limit(entries)) {
725d26d3 2167 ext4_lblk_t newblock;
e08ac99f
AB
2168 int levels = frame - frames + 1;
2169 unsigned int icount;
2170 int add_level = 1;
ac27a0ec
DK
2171 struct dx_entry *entries2;
2172 struct dx_node *node2;
2173 struct buffer_head *bh2;
2174
e08ac99f
AB
2175 while (frame > frames) {
2176 if (dx_get_count((frame - 1)->entries) <
2177 dx_get_limit((frame - 1)->entries)) {
2178 add_level = 0;
2179 break;
2180 }
2181 frame--; /* split higher index block */
2182 at = frame->at;
2183 entries = frame->entries;
2184 restart = 1;
2185 }
2186 if (add_level && levels == ext4_dir_htree_level(sb)) {
2187 ext4_warning(sb, "Directory (ino: %lu) index full, "
2188 "reach max htree level :%d",
2189 dir->i_ino, levels);
2190 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2191 ext4_warning(sb, "Large directory feature is "
2192 "not enabled on this "
2193 "filesystem");
2194 }
ac27a0ec
DK
2195 err = -ENOSPC;
2196 goto cleanup;
2197 }
e08ac99f 2198 icount = dx_get_count(entries);
0f70b406
TT
2199 bh2 = ext4_append(handle, dir, &newblock);
2200 if (IS_ERR(bh2)) {
2201 err = PTR_ERR(bh2);
ac27a0ec 2202 goto cleanup;
0f70b406 2203 }
ac27a0ec
DK
2204 node2 = (struct dx_node *)(bh2->b_data);
2205 entries2 = node2->entries;
1f7bebb9 2206 memset(&node2->fake, 0, sizeof(struct fake_dirent));
3d0518f4
WY
2207 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2208 sb->s_blocksize);
ac27a0ec 2209 BUFFER_TRACE(frame->bh, "get_write_access");
617ba13b 2210 err = ext4_journal_get_write_access(handle, frame->bh);
ac27a0ec
DK
2211 if (err)
2212 goto journal_error;
e08ac99f 2213 if (!add_level) {
ac27a0ec
DK
2214 unsigned icount1 = icount/2, icount2 = icount - icount1;
2215 unsigned hash2 = dx_get_hash(entries + icount1);
4776004f
TT
2216 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2217 icount1, icount2));
ac27a0ec
DK
2218
2219 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
617ba13b 2220 err = ext4_journal_get_write_access(handle,
e08ac99f 2221 (frame - 1)->bh);
ac27a0ec
DK
2222 if (err)
2223 goto journal_error;
2224
af5bc92d
TT
2225 memcpy((char *) entries2, (char *) (entries + icount1),
2226 icount2 * sizeof(struct dx_entry));
2227 dx_set_count(entries, icount1);
2228 dx_set_count(entries2, icount2);
2229 dx_set_limit(entries2, dx_node_limit(dir));
ac27a0ec
DK
2230
2231 /* Which index block gets the new entry? */
2232 if (at - entries >= icount1) {
2233 frame->at = at = at - entries - icount1 + entries2;
2234 frame->entries = entries = entries2;
2235 swap(frame->bh, bh2);
2236 }
e08ac99f
AB
2237 dx_insert_block((frame - 1), hash2, newblock);
2238 dxtrace(dx_show_index("node", frame->entries));
af5bc92d 2239 dxtrace(dx_show_index("node",
ac27a0ec 2240 ((struct dx_node *) bh2->b_data)->entries));
dbe89444 2241 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
ac27a0ec
DK
2242 if (err)
2243 goto journal_error;
2244 brelse (bh2);
e08ac99f
AB
2245 err = ext4_handle_dirty_dx_node(handle, dir,
2246 (frame - 1)->bh);
2247 if (err)
2248 goto journal_error;
2249 if (restart) {
2250 err = ext4_handle_dirty_dx_node(handle, dir,
2251 frame->bh);
2252 goto journal_error;
2253 }
ac27a0ec 2254 } else {
e08ac99f 2255 struct dx_root *dxroot;
ac27a0ec
DK
2256 memcpy((char *) entries2, (char *) entries,
2257 icount * sizeof(struct dx_entry));
2258 dx_set_limit(entries2, dx_node_limit(dir));
2259
2260 /* Set up root */
2261 dx_set_count(entries, 1);
2262 dx_set_block(entries + 0, newblock);
e08ac99f
AB
2263 dxroot = (struct dx_root *)frames[0].bh->b_data;
2264 dxroot->info.indirect_levels += 1;
2265 dxtrace(printk(KERN_DEBUG
2266 "Creating %d level index...\n",
2267 info->indirect_levels));
2268 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
ac27a0ec
DK
2269 if (err)
2270 goto journal_error;
e08ac99f
AB
2271 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2272 brelse(bh2);
2273 restart = 1;
2274 goto journal_error;
b4097142 2275 }
ac27a0ec 2276 }
5b643f9c 2277 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
f8b3b59d
TT
2278 if (IS_ERR(de)) {
2279 err = PTR_ERR(de);
ac27a0ec 2280 goto cleanup;
f8b3b59d 2281 }
5b643f9c 2282 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
ac27a0ec
DK
2283 goto cleanup;
2284
2285journal_error:
e08ac99f 2286 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
ac27a0ec 2287cleanup:
b1deefc9 2288 brelse(bh);
ac27a0ec 2289 dx_release(frames);
e08ac99f
AB
2290 /* @restart is true means htree-path has been changed, we need to
2291 * repeat dx_probe() to find out valid htree-path
2292 */
2293 if (restart && err == 0)
2294 goto again;
ac27a0ec
DK
2295 return err;
2296}
ac27a0ec
DK
2297
2298/*
05019a9e
TM
2299 * ext4_generic_delete_entry deletes a directory entry by merging it
2300 * with the previous entry
ac27a0ec 2301 */
05019a9e
TM
2302int ext4_generic_delete_entry(handle_t *handle,
2303 struct inode *dir,
2304 struct ext4_dir_entry_2 *de_del,
2305 struct buffer_head *bh,
2306 void *entry_buf,
2307 int buf_size,
2308 int csum_size)
ac27a0ec 2309{
af5bc92d 2310 struct ext4_dir_entry_2 *de, *pde;
3d0518f4 2311 unsigned int blocksize = dir->i_sb->s_blocksize;
05019a9e 2312 int i;
b0336e8d 2313
ac27a0ec
DK
2314 i = 0;
2315 pde = NULL;
05019a9e
TM
2316 de = (struct ext4_dir_entry_2 *)entry_buf;
2317 while (i < buf_size - csum_size) {
226ba972
TM
2318 if (ext4_check_dir_entry(dir, NULL, de, bh,
2319 bh->b_data, bh->b_size, i))
6a797d27 2320 return -EFSCORRUPTED;
ac27a0ec 2321 if (de == de_del) {
ac27a0ec 2322 if (pde)
a72d7f83 2323 pde->rec_len = ext4_rec_len_to_disk(
3d0518f4
WY
2324 ext4_rec_len_from_disk(pde->rec_len,
2325 blocksize) +
2326 ext4_rec_len_from_disk(de->rec_len,
2327 blocksize),
2328 blocksize);
ac27a0ec
DK
2329 else
2330 de->inode = 0;
e08ac99f 2331 inode_inc_iversion(dir);
ac27a0ec
DK
2332 return 0;
2333 }
3d0518f4 2334 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
ac27a0ec 2335 pde = de;
3d0518f4 2336 de = ext4_next_entry(de, blocksize);
ac27a0ec
DK
2337 }
2338 return -ENOENT;
2339}
2340
05019a9e
TM
2341static int ext4_delete_entry(handle_t *handle,
2342 struct inode *dir,
2343 struct ext4_dir_entry_2 *de_del,
2344 struct buffer_head *bh)
2345{
2346 int err, csum_size = 0;
2347
9f40fe54
TM
2348 if (ext4_has_inline_data(dir)) {
2349 int has_inline_data = 1;
2350 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2351 &has_inline_data);
2352 if (has_inline_data)
2353 return err;
2354 }
2355
9aa5d32b 2356 if (ext4_has_metadata_csum(dir->i_sb))
05019a9e
TM
2357 csum_size = sizeof(struct ext4_dir_entry_tail);
2358
2359 BUFFER_TRACE(bh, "get_write_access");
2360 err = ext4_journal_get_write_access(handle, bh);
2361 if (unlikely(err))
2362 goto out;
2363
2364 err = ext4_generic_delete_entry(handle, dir, de_del,
2365 bh, bh->b_data,
2366 dir->i_sb->s_blocksize, csum_size);
2367 if (err)
2368 goto out;
2369
2370 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2371 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2372 if (unlikely(err))
2373 goto out;
2374
2375 return 0;
2376out:
2377 if (err != -ENOENT)
2378 ext4_std_error(dir->i_sb, err);
2379 return err;
2380}
2381
f8628a14 2382/*
c7414892
AD
2383 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2384 * since this indicates that nlinks count was previously 1 to avoid overflowing
2385 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2386 * that subdirectory link counts are not being maintained accurately.
2387 *
2388 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2389 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2390 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2391 * on regular files) and to avoid creating huge/slow non-HTREE directories.
f8628a14
AD
2392 */
2393static void ext4_inc_count(handle_t *handle, struct inode *inode)
2394{
2395 inc_nlink(inode);
c7414892
AD
2396 if (is_dx(inode) &&
2397 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2398 set_nlink(inode, 1);
f8628a14
AD
2399}
2400
2401/*
2402 * If a directory had nlink == 1, then we should let it be 1. This indicates
2403 * directory has >EXT4_LINK_MAX subdirs.
2404 */
2405static void ext4_dec_count(handle_t *handle, struct inode *inode)
2406{
909a4cf1
AD
2407 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2408 drop_nlink(inode);
f8628a14
AD
2409}
2410
2411
617ba13b 2412static int ext4_add_nondir(handle_t *handle,
ac27a0ec
DK
2413 struct dentry *dentry, struct inode *inode)
2414{
617ba13b 2415 int err = ext4_add_entry(handle, dentry, inode);
ac27a0ec 2416 if (!err) {
617ba13b 2417 ext4_mark_inode_dirty(handle, inode);
6b38e842 2418 unlock_new_inode(inode);
8fc37ec5 2419 d_instantiate(dentry, inode);
ac27a0ec
DK
2420 return 0;
2421 }
731b9a54 2422 drop_nlink(inode);
6b38e842 2423 unlock_new_inode(inode);
ac27a0ec
DK
2424 iput(inode);
2425 return err;
2426}
2427
2428/*
2429 * By the time this is called, we already have created
2430 * the directory cache entry for the new file, but it
2431 * is so far negative - it has no inode.
2432 *
2433 * If the create succeeds, we fill in the inode information
2434 * with d_instantiate().
2435 */
4acdaf27 2436static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
ebfc3b49 2437 bool excl)
ac27a0ec
DK
2438{
2439 handle_t *handle;
af5bc92d 2440 struct inode *inode;
1139575a 2441 int err, credits, retries = 0;
ac27a0ec 2442
a7cdadee
JK
2443 err = dquot_initialize(dir);
2444 if (err)
2445 return err;
907f4554 2446
1139575a 2447 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2448 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2449retry:
1139575a
TT
2450 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2451 NULL, EXT4_HT_DIR, credits);
2452 handle = ext4_journal_current_handle();
ac27a0ec
DK
2453 err = PTR_ERR(inode);
2454 if (!IS_ERR(inode)) {
617ba13b 2455 inode->i_op = &ext4_file_inode_operations;
be64f884 2456 inode->i_fop = &ext4_file_operations;
617ba13b 2457 ext4_set_aops(inode);
e709e9df 2458 err = ext4_add_nondir(handle, dentry, inode);
1139575a
TT
2459 if (!err && IS_DIRSYNC(dir))
2460 ext4_handle_sync(handle);
ac27a0ec 2461 }
1139575a
TT
2462 if (handle)
2463 ext4_journal_stop(handle);
617ba13b 2464 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK
2465 goto retry;
2466 return err;
2467}
2468
af5bc92d 2469static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1a67aafb 2470 umode_t mode, dev_t rdev)
ac27a0ec
DK
2471{
2472 handle_t *handle;
2473 struct inode *inode;
1139575a 2474 int err, credits, retries = 0;
ac27a0ec 2475
a7cdadee
JK
2476 err = dquot_initialize(dir);
2477 if (err)
2478 return err;
907f4554 2479
1139575a 2480 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2481 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2482retry:
1139575a
TT
2483 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2484 NULL, EXT4_HT_DIR, credits);
2485 handle = ext4_journal_current_handle();
ac27a0ec
DK
2486 err = PTR_ERR(inode);
2487 if (!IS_ERR(inode)) {
2488 init_special_inode(inode, inode->i_mode, rdev);
617ba13b 2489 inode->i_op = &ext4_special_inode_operations;
617ba13b 2490 err = ext4_add_nondir(handle, dentry, inode);
1139575a
TT
2491 if (!err && IS_DIRSYNC(dir))
2492 ext4_handle_sync(handle);
ac27a0ec 2493 }
1139575a
TT
2494 if (handle)
2495 ext4_journal_stop(handle);
617ba13b 2496 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK
2497 goto retry;
2498 return err;
2499}
2500
af51a2ac
AV
2501static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2502{
2503 handle_t *handle;
2504 struct inode *inode;
2505 int err, retries = 0;
2506
a7cdadee
JK
2507 err = dquot_initialize(dir);
2508 if (err)
2509 return err;
af51a2ac
AV
2510
2511retry:
2512 inode = ext4_new_inode_start_handle(dir, mode,
2513 NULL, 0, NULL,
2514 EXT4_HT_DIR,
2515 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2516 4 + EXT4_XATTR_TRANS_BLOCKS);
2517 handle = ext4_journal_current_handle();
2518 err = PTR_ERR(inode);
2519 if (!IS_ERR(inode)) {
2520 inode->i_op = &ext4_file_inode_operations;
be64f884 2521 inode->i_fop = &ext4_file_operations;
af51a2ac 2522 ext4_set_aops(inode);
e94bd349 2523 d_tmpfile(dentry, inode);
af51a2ac
AV
2524 err = ext4_orphan_add(handle, inode);
2525 if (err)
43ae9e3f 2526 goto err_unlock_inode;
af51a2ac 2527 mark_inode_dirty(inode);
af51a2ac
AV
2528 unlock_new_inode(inode);
2529 }
2530 if (handle)
2531 ext4_journal_stop(handle);
2532 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2533 goto retry;
2534 return err;
43ae9e3f 2535err_unlock_inode:
af51a2ac
AV
2536 ext4_journal_stop(handle);
2537 unlock_new_inode(inode);
af51a2ac
AV
2538 return err;
2539}
2540
a774f9c2
TM
2541struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2542 struct ext4_dir_entry_2 *de,
2543 int blocksize, int csum_size,
2544 unsigned int parent_ino, int dotdot_real_len)
2545{
2546 de->inode = cpu_to_le32(inode->i_ino);
2547 de->name_len = 1;
2548 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2549 blocksize);
2550 strcpy(de->name, ".");
2551 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2552
2553 de = ext4_next_entry(de, blocksize);
2554 de->inode = cpu_to_le32(parent_ino);
2555 de->name_len = 2;
2556 if (!dotdot_real_len)
2557 de->rec_len = ext4_rec_len_to_disk(blocksize -
2558 (csum_size + EXT4_DIR_REC_LEN(1)),
2559 blocksize);
2560 else
2561 de->rec_len = ext4_rec_len_to_disk(
2562 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2563 strcpy(de->name, "..");
2564 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2565
2566 return ext4_next_entry(de, blocksize);
2567}
2568
2569static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2570 struct inode *inode)
ac27a0ec 2571{
dabd991f 2572 struct buffer_head *dir_block = NULL;
af5bc92d 2573 struct ext4_dir_entry_2 *de;
b0336e8d 2574 struct ext4_dir_entry_tail *t;
dc6982ff 2575 ext4_lblk_t block = 0;
3d0518f4 2576 unsigned int blocksize = dir->i_sb->s_blocksize;
b0336e8d 2577 int csum_size = 0;
a774f9c2 2578 int err;
ac27a0ec 2579
9aa5d32b 2580 if (ext4_has_metadata_csum(dir->i_sb))
b0336e8d
DW
2581 csum_size = sizeof(struct ext4_dir_entry_tail);
2582
3c47d541
TM
2583 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2584 err = ext4_try_create_inline_dir(handle, dir, inode);
2585 if (err < 0 && err != -ENOSPC)
2586 goto out;
2587 if (!err)
2588 goto out;
2589 }
2590
dc6982ff 2591 inode->i_size = 0;
0f70b406
TT
2592 dir_block = ext4_append(handle, inode, &block);
2593 if (IS_ERR(dir_block))
2594 return PTR_ERR(dir_block);
a774f9c2
TM
2595 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2596 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2597 set_nlink(inode, 2);
2598 if (csum_size) {
2599 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2600 initialize_dirent_tail(t, blocksize);
2601 }
2602
2603 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2604 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2605 if (err)
2606 goto out;
2607 set_buffer_verified(dir_block);
2608out:
2609 brelse(dir_block);
2610 return err;
2611}
2612
2613static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2614{
2615 handle_t *handle;
2616 struct inode *inode;
1139575a 2617 int err, credits, retries = 0;
a774f9c2 2618
f8628a14 2619 if (EXT4_DIR_LINK_MAX(dir))
ac27a0ec
DK
2620 return -EMLINK;
2621
a7cdadee
JK
2622 err = dquot_initialize(dir);
2623 if (err)
2624 return err;
907f4554 2625
1139575a 2626 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 2627 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
ac27a0ec 2628retry:
1139575a
TT
2629 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2630 &dentry->d_name,
2631 0, NULL, EXT4_HT_DIR, credits);
2632 handle = ext4_journal_current_handle();
ac27a0ec
DK
2633 err = PTR_ERR(inode);
2634 if (IS_ERR(inode))
2635 goto out_stop;
2636
617ba13b
MC
2637 inode->i_op = &ext4_dir_inode_operations;
2638 inode->i_fop = &ext4_dir_operations;
a774f9c2 2639 err = ext4_init_new_dir(handle, dir, inode);
dabd991f
NK
2640 if (err)
2641 goto out_clear_inode;
2642 err = ext4_mark_inode_dirty(handle, inode);
2643 if (!err)
2644 err = ext4_add_entry(handle, dentry, inode);
ac27a0ec 2645 if (err) {
4cdeed86
AK
2646out_clear_inode:
2647 clear_nlink(inode);
6b38e842 2648 unlock_new_inode(inode);
617ba13b 2649 ext4_mark_inode_dirty(handle, inode);
af5bc92d 2650 iput(inode);
ac27a0ec
DK
2651 goto out_stop;
2652 }
f8628a14 2653 ext4_inc_count(handle, dir);
617ba13b 2654 ext4_update_dx_flag(dir);
dabd991f
NK
2655 err = ext4_mark_inode_dirty(handle, dir);
2656 if (err)
2657 goto out_clear_inode;
6b38e842 2658 unlock_new_inode(inode);
8fc37ec5 2659 d_instantiate(dentry, inode);
1139575a
TT
2660 if (IS_DIRSYNC(dir))
2661 ext4_handle_sync(handle);
2662
ac27a0ec 2663out_stop:
1139575a
TT
2664 if (handle)
2665 ext4_journal_stop(handle);
617ba13b 2666 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK
2667 goto retry;
2668 return err;
2669}
2670
2671/*
2672 * routine to check that the specified directory is empty (for rmdir)
2673 */
a7550b30 2674bool ext4_empty_dir(struct inode *inode)
ac27a0ec 2675{
498e5f24 2676 unsigned int offset;
af5bc92d
TT
2677 struct buffer_head *bh;
2678 struct ext4_dir_entry_2 *de, *de1;
2679 struct super_block *sb;
ac27a0ec 2680
61f86638
TM
2681 if (ext4_has_inline_data(inode)) {
2682 int has_inline_data = 1;
a7550b30 2683 int ret;
61f86638 2684
a7550b30 2685 ret = empty_inline_dir(inode, &has_inline_data);
61f86638 2686 if (has_inline_data)
a7550b30 2687 return ret;
61f86638
TM
2688 }
2689
ac27a0ec 2690 sb = inode->i_sb;
dc6982ff
TT
2691 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2692 EXT4_ERROR_INODE(inode, "invalid size");
a7550b30 2693 return true;
ac27a0ec 2694 }
dc6982ff
TT
2695 bh = ext4_read_dirblock(inode, 0, EITHER);
2696 if (IS_ERR(bh))
a7550b30 2697 return true;
dc6982ff 2698
617ba13b 2699 de = (struct ext4_dir_entry_2 *) bh->b_data;
3d0518f4 2700 de1 = ext4_next_entry(de, sb->s_blocksize);
ac27a0ec 2701 if (le32_to_cpu(de->inode) != inode->i_ino ||
b03a2f7e
AD
2702 le32_to_cpu(de1->inode) == 0 ||
2703 strcmp(".", de->name) || strcmp("..", de1->name)) {
2704 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
af5bc92d 2705 brelse(bh);
a7550b30 2706 return true;
ac27a0ec 2707 }
3d0518f4
WY
2708 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2709 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2710 de = ext4_next_entry(de1, sb->s_blocksize);
af5bc92d 2711 while (offset < inode->i_size) {
236f5ecb 2712 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
24676da4 2713 unsigned int lblock;
af5bc92d 2714 brelse(bh);
24676da4 2715 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
dc6982ff
TT
2716 bh = ext4_read_dirblock(inode, lblock, EITHER);
2717 if (IS_ERR(bh))
a7550b30 2718 return true;
617ba13b 2719 de = (struct ext4_dir_entry_2 *) bh->b_data;
ac27a0ec 2720 }
226ba972
TM
2721 if (ext4_check_dir_entry(inode, NULL, de, bh,
2722 bh->b_data, bh->b_size, offset)) {
617ba13b 2723 de = (struct ext4_dir_entry_2 *)(bh->b_data +
ac27a0ec
DK
2724 sb->s_blocksize);
2725 offset = (offset | (sb->s_blocksize - 1)) + 1;
2726 continue;
2727 }
2728 if (le32_to_cpu(de->inode)) {
af5bc92d 2729 brelse(bh);
a7550b30 2730 return false;
ac27a0ec 2731 }
3d0518f4
WY
2732 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2733 de = ext4_next_entry(de, sb->s_blocksize);
ac27a0ec 2734 }
af5bc92d 2735 brelse(bh);
a7550b30 2736 return true;
ac27a0ec
DK
2737}
2738
d745a8c2
JK
2739/*
2740 * ext4_orphan_add() links an unlinked or truncated inode into a list of
ac27a0ec
DK
2741 * such inodes, starting at the superblock, in case we crash before the
2742 * file is closed/deleted, or in case the inode truncate spans multiple
2743 * transactions and the last transaction is not recovered after a crash.
2744 *
2745 * At filesystem recovery time, we walk this list deleting unlinked
617ba13b 2746 * inodes and truncating linked inodes in ext4_orphan_cleanup().
d745a8c2
JK
2747 *
2748 * Orphan list manipulation functions must be called under i_mutex unless
2749 * we are just creating the inode or deleting it.
ac27a0ec 2750 */
617ba13b 2751int ext4_orphan_add(handle_t *handle, struct inode *inode)
ac27a0ec
DK
2752{
2753 struct super_block *sb = inode->i_sb;
cd2c080c 2754 struct ext4_sb_info *sbi = EXT4_SB(sb);
617ba13b 2755 struct ext4_iloc iloc;
ac27a0ec 2756 int err = 0, rc;
d745a8c2 2757 bool dirty = false;
ac27a0ec 2758
e2bfb088 2759 if (!sbi->s_journal || is_bad_inode(inode))
0390131b
FM
2760 return 0;
2761
d745a8c2 2762 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
5955102c 2763 !inode_is_locked(inode));
d745a8c2
JK
2764 /*
2765 * Exit early if inode already is on orphan list. This is a big speedup
2766 * since we don't have to contend on the global s_orphan_lock.
2767 */
617ba13b 2768 if (!list_empty(&EXT4_I(inode)->i_orphan))
d745a8c2 2769 return 0;
ac27a0ec 2770
afb86178
LC
2771 /*
2772 * Orphan handling is only valid for files with data blocks
2773 * being truncated, or files being unlinked. Note that we either
2774 * hold i_mutex, or the inode can not be referenced from outside,
2775 * so i_nlink should not be bumped due to race
ac27a0ec 2776 */
af5bc92d
TT
2777 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2778 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
ac27a0ec 2779
cd2c080c
JK
2780 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2781 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
ac27a0ec 2782 if (err)
d745a8c2 2783 goto out;
ac27a0ec 2784
617ba13b 2785 err = ext4_reserve_inode_write(handle, inode, &iloc);
ac27a0ec 2786 if (err)
d745a8c2
JK
2787 goto out;
2788
2789 mutex_lock(&sbi->s_orphan_lock);
6e3617e5
DM
2790 /*
2791 * Due to previous errors inode may be already a part of on-disk
2792 * orphan list. If so skip on-disk list modification.
2793 */
d745a8c2
JK
2794 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2795 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2796 /* Insert this inode at the head of the on-disk orphan list */
2797 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2798 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2799 dirty = true;
2800 }
2801 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2802 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 2803
d745a8c2
JK
2804 if (dirty) {
2805 err = ext4_handle_dirty_super(handle, sb);
2806 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2807 if (!err)
2808 err = rc;
2809 if (err) {
2810 /*
2811 * We have to remove inode from in-memory list if
2812 * addition to on disk orphan list failed. Stray orphan
2813 * list entries can cause panics at unmount time.
2814 */
2815 mutex_lock(&sbi->s_orphan_lock);
74177f55 2816 list_del_init(&EXT4_I(inode)->i_orphan);
d745a8c2
JK
2817 mutex_unlock(&sbi->s_orphan_lock);
2818 }
2819 }
ac27a0ec
DK
2820 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2821 jbd_debug(4, "orphan inode %lu will point to %d\n",
2822 inode->i_ino, NEXT_ORPHAN(inode));
d745a8c2 2823out:
cd2c080c 2824 ext4_std_error(sb, err);
ac27a0ec
DK
2825 return err;
2826}
2827
2828/*
617ba13b 2829 * ext4_orphan_del() removes an unlinked or truncated inode from the list
ac27a0ec
DK
2830 * of such inodes stored on disk, because it is finally being cleaned up.
2831 */
617ba13b 2832int ext4_orphan_del(handle_t *handle, struct inode *inode)
ac27a0ec
DK
2833{
2834 struct list_head *prev;
617ba13b 2835 struct ext4_inode_info *ei = EXT4_I(inode);
cd2c080c 2836 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
498e5f24 2837 __u32 ino_next;
617ba13b 2838 struct ext4_iloc iloc;
ac27a0ec
DK
2839 int err = 0;
2840
cd2c080c 2841 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
0390131b
FM
2842 return 0;
2843
d745a8c2 2844 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
5955102c 2845 !inode_is_locked(inode));
d745a8c2 2846 /* Do this quick check before taking global s_orphan_lock. */
3b9d4ed2 2847 if (list_empty(&ei->i_orphan))
d745a8c2 2848 return 0;
ac27a0ec 2849
d745a8c2
JK
2850 if (handle) {
2851 /* Grab inode buffer early before taking global s_orphan_lock */
2852 err = ext4_reserve_inode_write(handle, inode, &iloc);
2853 }
ac27a0ec 2854
d745a8c2 2855 mutex_lock(&sbi->s_orphan_lock);
ac27a0ec
DK
2856 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2857
d745a8c2 2858 prev = ei->i_orphan.prev;
ac27a0ec
DK
2859 list_del_init(&ei->i_orphan);
2860
2861 /* If we're on an error path, we may not have a valid
2862 * transaction handle with which to update the orphan list on
2863 * disk, but we still need to remove the inode from the linked
2864 * list in memory. */
d745a8c2
JK
2865 if (!handle || err) {
2866 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 2867 goto out_err;
d745a8c2 2868 }
ac27a0ec 2869
d745a8c2 2870 ino_next = NEXT_ORPHAN(inode);
ac27a0ec 2871 if (prev == &sbi->s_orphan) {
498e5f24 2872 jbd_debug(4, "superblock will point to %u\n", ino_next);
ac27a0ec 2873 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
617ba13b 2874 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
d745a8c2
JK
2875 if (err) {
2876 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 2877 goto out_brelse;
d745a8c2 2878 }
ac27a0ec 2879 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
d745a8c2 2880 mutex_unlock(&sbi->s_orphan_lock);
b50924c2 2881 err = ext4_handle_dirty_super(handle, inode->i_sb);
ac27a0ec 2882 } else {
617ba13b 2883 struct ext4_iloc iloc2;
ac27a0ec 2884 struct inode *i_prev =
617ba13b 2885 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
ac27a0ec 2886
498e5f24 2887 jbd_debug(4, "orphan inode %lu will point to %u\n",
ac27a0ec 2888 i_prev->i_ino, ino_next);
617ba13b 2889 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
d745a8c2
JK
2890 if (err) {
2891 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec 2892 goto out_brelse;
d745a8c2 2893 }
ac27a0ec 2894 NEXT_ORPHAN(i_prev) = ino_next;
617ba13b 2895 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
d745a8c2 2896 mutex_unlock(&sbi->s_orphan_lock);
ac27a0ec
DK
2897 }
2898 if (err)
2899 goto out_brelse;
2900 NEXT_ORPHAN(inode) = 0;
617ba13b 2901 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
ac27a0ec 2902out_err:
617ba13b 2903 ext4_std_error(inode->i_sb, err);
ac27a0ec
DK
2904 return err;
2905
2906out_brelse:
2907 brelse(iloc.bh);
2908 goto out_err;
2909}
2910
af5bc92d 2911static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
ac27a0ec
DK
2912{
2913 int retval;
af5bc92d
TT
2914 struct inode *inode;
2915 struct buffer_head *bh;
2916 struct ext4_dir_entry_2 *de;
8dcfaad2 2917 handle_t *handle = NULL;
ac27a0ec 2918
0db1ff22
TT
2919 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2920 return -EIO;
2921
ac27a0ec
DK
2922 /* Initialize quotas before so that eventual writes go in
2923 * separate transaction */
a7cdadee
JK
2924 retval = dquot_initialize(dir);
2925 if (retval)
2926 return retval;
2927 retval = dquot_initialize(d_inode(dentry));
2928 if (retval)
2929 return retval;
907f4554 2930
ac27a0ec 2931 retval = -ENOENT;
32f7f22c 2932 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
36de9286
TT
2933 if (IS_ERR(bh))
2934 return PTR_ERR(bh);
ac27a0ec
DK
2935 if (!bh)
2936 goto end_rmdir;
2937
2b0143b5 2938 inode = d_inode(dentry);
ac27a0ec 2939
6a797d27 2940 retval = -EFSCORRUPTED;
ac27a0ec
DK
2941 if (le32_to_cpu(de->inode) != inode->i_ino)
2942 goto end_rmdir;
2943
2944 retval = -ENOTEMPTY;
e875a2dd 2945 if (!ext4_empty_dir(inode))
ac27a0ec
DK
2946 goto end_rmdir;
2947
8dcfaad2 2948 handle = ext4_journal_start(dir, EXT4_HT_DIR,
64044abf 2949 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
8dcfaad2
TT
2950 if (IS_ERR(handle)) {
2951 retval = PTR_ERR(handle);
2952 handle = NULL;
2953 goto end_rmdir;
2954 }
2955
2956 if (IS_DIRSYNC(dir))
2957 ext4_handle_sync(handle);
2958
617ba13b 2959 retval = ext4_delete_entry(handle, dir, de, bh);
ac27a0ec
DK
2960 if (retval)
2961 goto end_rmdir;
f8628a14 2962 if (!EXT4_DIR_LINK_EMPTY(inode))
b03a2f7e
AD
2963 ext4_warning_inode(inode,
2964 "empty directory '%.*s' has too many links (%u)",
2965 dentry->d_name.len, dentry->d_name.name,
af5bc92d 2966 inode->i_nlink);
ac27a0ec
DK
2967 inode->i_version++;
2968 clear_nlink(inode);
2969 /* There's no need to set i_disksize: the fact that i_nlink is
2970 * zero will ensure that the right thing happens during any
2971 * recovery. */
2972 inode->i_size = 0;
617ba13b 2973 ext4_orphan_add(handle, inode);
eeca7ea1 2974 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
617ba13b 2975 ext4_mark_inode_dirty(handle, inode);
f8628a14 2976 ext4_dec_count(handle, dir);
617ba13b
MC
2977 ext4_update_dx_flag(dir);
2978 ext4_mark_inode_dirty(handle, dir);
ac27a0ec
DK
2979
2980end_rmdir:
af5bc92d 2981 brelse(bh);
8dcfaad2
TT
2982 if (handle)
2983 ext4_journal_stop(handle);
ac27a0ec
DK
2984 return retval;
2985}
2986
af5bc92d 2987static int ext4_unlink(struct inode *dir, struct dentry *dentry)
ac27a0ec
DK
2988{
2989 int retval;
af5bc92d
TT
2990 struct inode *inode;
2991 struct buffer_head *bh;
2992 struct ext4_dir_entry_2 *de;
931b6864 2993 handle_t *handle = NULL;
ac27a0ec 2994
0db1ff22
TT
2995 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2996 return -EIO;
2997
0562e0ba 2998 trace_ext4_unlink_enter(dir, dentry);
ac27a0ec
DK
2999 /* Initialize quotas before so that eventual writes go
3000 * in separate transaction */
a7cdadee
JK
3001 retval = dquot_initialize(dir);
3002 if (retval)
3003 return retval;
3004 retval = dquot_initialize(d_inode(dentry));
3005 if (retval)
3006 return retval;
907f4554 3007
ac27a0ec 3008 retval = -ENOENT;
32f7f22c 3009 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
36de9286
TT
3010 if (IS_ERR(bh))
3011 return PTR_ERR(bh);
ac27a0ec
DK
3012 if (!bh)
3013 goto end_unlink;
3014
2b0143b5 3015 inode = d_inode(dentry);
ac27a0ec 3016
6a797d27 3017 retval = -EFSCORRUPTED;
ac27a0ec
DK
3018 if (le32_to_cpu(de->inode) != inode->i_ino)
3019 goto end_unlink;
3020
931b6864 3021 handle = ext4_journal_start(dir, EXT4_HT_DIR,
64044abf 3022 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
931b6864
TT
3023 if (IS_ERR(handle)) {
3024 retval = PTR_ERR(handle);
3025 handle = NULL;
3026 goto end_unlink;
3027 }
3028
3029 if (IS_DIRSYNC(dir))
3030 ext4_handle_sync(handle);
3031
b03a2f7e
AD
3032 if (inode->i_nlink == 0) {
3033 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3034 dentry->d_name.len, dentry->d_name.name);
bfe86848 3035 set_nlink(inode, 1);
ac27a0ec 3036 }
617ba13b 3037 retval = ext4_delete_entry(handle, dir, de, bh);
ac27a0ec
DK
3038 if (retval)
3039 goto end_unlink;
eeca7ea1 3040 dir->i_ctime = dir->i_mtime = current_time(dir);
617ba13b
MC
3041 ext4_update_dx_flag(dir);
3042 ext4_mark_inode_dirty(handle, dir);
825f1481 3043 drop_nlink(inode);
ac27a0ec 3044 if (!inode->i_nlink)
617ba13b 3045 ext4_orphan_add(handle, inode);
eeca7ea1 3046 inode->i_ctime = current_time(inode);
617ba13b 3047 ext4_mark_inode_dirty(handle, inode);
ac27a0ec
DK
3048
3049end_unlink:
af5bc92d 3050 brelse(bh);
931b6864
TT
3051 if (handle)
3052 ext4_journal_stop(handle);
0562e0ba 3053 trace_ext4_unlink_exit(dentry, retval);
ac27a0ec
DK
3054 return retval;
3055}
3056
af5bc92d
TT
3057static int ext4_symlink(struct inode *dir,
3058 struct dentry *dentry, const char *symname)
ac27a0ec
DK
3059{
3060 handle_t *handle;
af5bc92d 3061 struct inode *inode;
f348c252 3062 int err, len = strlen(symname);
df5e6223 3063 int credits;
f348c252 3064 bool encryption_required;
a7550b30
JK
3065 struct fscrypt_str disk_link;
3066 struct fscrypt_symlink_data *sd = NULL;
ac27a0ec 3067
0db1ff22
TT
3068 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3069 return -EIO;
3070
f348c252
TT
3071 disk_link.len = len + 1;
3072 disk_link.name = (char *) symname;
3073
6ddb2447
TT
3074 encryption_required = (ext4_encrypted_inode(dir) ||
3075 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
4d3c4e5b 3076 if (encryption_required) {
a7550b30 3077 err = fscrypt_get_encryption_info(dir);
4d3c4e5b
TT
3078 if (err)
3079 return err;
a7550b30 3080 if (!fscrypt_has_encryption_key(dir))
54475f53 3081 return -ENOKEY;
a7550b30
JK
3082 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3083 sizeof(struct fscrypt_symlink_data));
4d3c4e5b
TT
3084 sd = kzalloc(disk_link.len, GFP_KERNEL);
3085 if (!sd)
3086 return -ENOMEM;
3087 }
3088
3089 if (disk_link.len > dir->i_sb->s_blocksize) {
3090 err = -ENAMETOOLONG;
3091 goto err_free_sd;
3092 }
ac27a0ec 3093
a7cdadee
JK
3094 err = dquot_initialize(dir);
3095 if (err)
926631c2 3096 goto err_free_sd;
907f4554 3097
f348c252 3098 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
df5e6223
JK
3099 /*
3100 * For non-fast symlinks, we just allocate inode and put it on
3101 * orphan list in the first transaction => we need bitmap,
8c208719
ES
3102 * group descriptor, sb, inode block, quota blocks, and
3103 * possibly selinux xattr blocks.
df5e6223 3104 */
8c208719
ES
3105 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3106 EXT4_XATTR_TRANS_BLOCKS;
df5e6223
JK
3107 } else {
3108 /*
3109 * Fast symlink. We have to add entry to directory
3110 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3111 * allocate new inode (bitmap, group descriptor, inode block,
3112 * quota blocks, sb is already counted in previous macros).
3113 */
3114 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
eb9cc7e1 3115 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
df5e6223 3116 }
f348c252 3117
1139575a
TT
3118 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3119 &dentry->d_name, 0, NULL,
3120 EXT4_HT_DIR, credits);
3121 handle = ext4_journal_current_handle();
f348c252
TT
3122 if (IS_ERR(inode)) {
3123 if (handle)
3124 ext4_journal_stop(handle);
4d3c4e5b
TT
3125 err = PTR_ERR(inode);
3126 goto err_free_sd;
f348c252
TT
3127 }
3128
3129 if (encryption_required) {
f348c252 3130 struct qstr istr;
a7550b30
JK
3131 struct fscrypt_str ostr =
3132 FSTR_INIT(sd->encrypted_path, disk_link.len);
f348c252 3133
f348c252
TT
3134 istr.name = (const unsigned char *) symname;
3135 istr.len = len;
a7550b30 3136 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
ef1eb3aa 3137 if (err)
f348c252
TT
3138 goto err_drop_inode;
3139 sd->len = cpu_to_le16(ostr.len);
3140 disk_link.name = (char *) sd;
a7a67e8a 3141 inode->i_op = &ext4_encrypted_symlink_inode_operations;
f348c252 3142 }
ac27a0ec 3143
f348c252 3144 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
a7a67e8a
AV
3145 if (!encryption_required)
3146 inode->i_op = &ext4_symlink_inode_operations;
21fc61c7 3147 inode_nohighmem(inode);
617ba13b 3148 ext4_set_aops(inode);
ac27a0ec 3149 /*
df5e6223
JK
3150 * We cannot call page_symlink() with transaction started
3151 * because it calls into ext4_write_begin() which can wait
3152 * for transaction commit if we are running out of space
3153 * and thus we deadlock. So we have to stop transaction now
3154 * and restart it when symlink contents is written.
3155 *
3156 * To keep fs consistent in case of crash, we have to put inode
3157 * to orphan list in the mean time.
ac27a0ec 3158 */
df5e6223
JK
3159 drop_nlink(inode);
3160 err = ext4_orphan_add(handle, inode);
3161 ext4_journal_stop(handle);
f348c252 3162 handle = NULL;
df5e6223
JK
3163 if (err)
3164 goto err_drop_inode;
f348c252 3165 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
df5e6223
JK
3166 if (err)
3167 goto err_drop_inode;
3168 /*
3169 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3170 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3171 */
9924a92a 3172 handle = ext4_journal_start(dir, EXT4_HT_DIR,
df5e6223
JK
3173 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3174 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3175 if (IS_ERR(handle)) {
3176 err = PTR_ERR(handle);
f348c252 3177 handle = NULL;
df5e6223
JK
3178 goto err_drop_inode;
3179 }
0ce8c010 3180 set_nlink(inode, 1);
df5e6223 3181 err = ext4_orphan_del(handle, inode);
f348c252 3182 if (err)
df5e6223 3183 goto err_drop_inode;
ac27a0ec 3184 } else {
e65187e6 3185 /* clear the extent format for fast symlink */
12e9b892 3186 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
75e7566b 3187 if (!encryption_required) {
a7a67e8a 3188 inode->i_op = &ext4_fast_symlink_inode_operations;
75e7566b
AV
3189 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3190 }
f348c252
TT
3191 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3192 disk_link.len);
3193 inode->i_size = disk_link.len - 1;
ac27a0ec 3194 }
617ba13b
MC
3195 EXT4_I(inode)->i_disksize = inode->i_size;
3196 err = ext4_add_nondir(handle, dentry, inode);
1139575a
TT
3197 if (!err && IS_DIRSYNC(dir))
3198 ext4_handle_sync(handle);
3199
1139575a
TT
3200 if (handle)
3201 ext4_journal_stop(handle);
f348c252 3202 kfree(sd);
ac27a0ec 3203 return err;
df5e6223 3204err_drop_inode:
f348c252
TT
3205 if (handle)
3206 ext4_journal_stop(handle);
f348c252 3207 clear_nlink(inode);
df5e6223
JK
3208 unlock_new_inode(inode);
3209 iput(inode);
4d3c4e5b
TT
3210err_free_sd:
3211 kfree(sd);
df5e6223 3212 return err;
ac27a0ec
DK
3213}
3214
af5bc92d
TT
3215static int ext4_link(struct dentry *old_dentry,
3216 struct inode *dir, struct dentry *dentry)
ac27a0ec
DK
3217{
3218 handle_t *handle;
2b0143b5 3219 struct inode *inode = d_inode(old_dentry);
ac27a0ec
DK
3220 int err, retries = 0;
3221
b05ab1dc 3222 if (inode->i_nlink >= EXT4_LINK_MAX)
ac27a0ec 3223 return -EMLINK;
d9cdc903 3224 if (ext4_encrypted_inode(dir) &&
a7550b30 3225 !fscrypt_has_permitted_context(dir, inode))
d9cdc903 3226 return -EPERM;
040cb378
LX
3227
3228 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3229 (!projid_eq(EXT4_I(dir)->i_projid,
3230 EXT4_I(old_dentry->d_inode)->i_projid)))
3231 return -EXDEV;
3232
a7cdadee
JK
3233 err = dquot_initialize(dir);
3234 if (err)
3235 return err;
907f4554 3236
ac27a0ec 3237retry:
9924a92a
TT
3238 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3239 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
af51a2ac 3240 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
ac27a0ec
DK
3241 if (IS_ERR(handle))
3242 return PTR_ERR(handle);
3243
3244 if (IS_DIRSYNC(dir))
0390131b 3245 ext4_handle_sync(handle);
ac27a0ec 3246
eeca7ea1 3247 inode->i_ctime = current_time(inode);
f8628a14 3248 ext4_inc_count(handle, inode);
7de9c6ee 3249 ihold(inode);
ac27a0ec 3250
6b38e842
AV
3251 err = ext4_add_entry(handle, dentry, inode);
3252 if (!err) {
3253 ext4_mark_inode_dirty(handle, inode);
af51a2ac
AV
3254 /* this can happen only for tmpfile being
3255 * linked the first time
3256 */
3257 if (inode->i_nlink == 1)
3258 ext4_orphan_del(handle, inode);
6b38e842
AV
3259 d_instantiate(dentry, inode);
3260 } else {
3261 drop_nlink(inode);
3262 iput(inode);
3263 }
617ba13b
MC
3264 ext4_journal_stop(handle);
3265 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
ac27a0ec
DK
3266 goto retry;
3267 return err;
3268}
3269
32f7f22c
TM
3270
3271/*
3272 * Try to find buffer head where contains the parent block.
3273 * It should be the inode block if it is inlined or the 1st block
3274 * if it is a normal dir.
3275 */
3276static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3277 struct inode *inode,
3278 int *retval,
3279 struct ext4_dir_entry_2 **parent_de,
3280 int *inlined)
3281{
3282 struct buffer_head *bh;
3283
3284 if (!ext4_has_inline_data(inode)) {
dc6982ff
TT
3285 bh = ext4_read_dirblock(inode, 0, EITHER);
3286 if (IS_ERR(bh)) {
3287 *retval = PTR_ERR(bh);
32f7f22c
TM
3288 return NULL;
3289 }
3290 *parent_de = ext4_next_entry(
3291 (struct ext4_dir_entry_2 *)bh->b_data,
3292 inode->i_sb->s_blocksize);
3293 return bh;
3294 }
3295
3296 *inlined = 1;
3297 return ext4_get_first_inline_block(inode, parent_de, retval);
3298}
ac27a0ec 3299
c0d268c3
MS
3300struct ext4_renament {
3301 struct inode *dir;
3302 struct dentry *dentry;
3303 struct inode *inode;
bd42998a
MS
3304 bool is_dir;
3305 int dir_nlink_delta;
c0d268c3
MS
3306
3307 /* entry for "dentry" */
3308 struct buffer_head *bh;
3309 struct ext4_dir_entry_2 *de;
3310 int inlined;
3311
3312 /* entry for ".." in inode if it's a directory */
3313 struct buffer_head *dir_bh;
3314 struct ext4_dir_entry_2 *parent_de;
3315 int dir_inlined;
3316};
3317
bd1af145
MS
3318static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3319{
3320 int retval;
3321
3322 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3323 &retval, &ent->parent_de,
3324 &ent->dir_inlined);
3325 if (!ent->dir_bh)
3326 return retval;
3327 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
6a797d27 3328 return -EFSCORRUPTED;
bd1af145
MS
3329 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3330 return ext4_journal_get_write_access(handle, ent->dir_bh);
3331}
3332
3333static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3334 unsigned dir_ino)
3335{
3336 int retval;
3337
3338 ent->parent_de->inode = cpu_to_le32(dir_ino);
3339 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3340 if (!ent->dir_inlined) {
3341 if (is_dx(ent->inode)) {
3342 retval = ext4_handle_dirty_dx_node(handle,
3343 ent->inode,
3344 ent->dir_bh);
3345 } else {
3346 retval = ext4_handle_dirty_dirent_node(handle,
3347 ent->inode,
3348 ent->dir_bh);
3349 }
3350 } else {
3351 retval = ext4_mark_inode_dirty(handle, ent->inode);
3352 }
3353 if (retval) {
3354 ext4_std_error(ent->dir->i_sb, retval);
3355 return retval;
3356 }
3357 return 0;
3358}
3359
3360static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3361 unsigned ino, unsigned file_type)
3362{
3363 int retval;
3364
3365 BUFFER_TRACE(ent->bh, "get write access");
3366 retval = ext4_journal_get_write_access(handle, ent->bh);
3367 if (retval)
3368 return retval;
3369 ent->de->inode = cpu_to_le32(ino);
e2b911c5 3370 if (ext4_has_feature_filetype(ent->dir->i_sb))
bd1af145
MS
3371 ent->de->file_type = file_type;
3372 ent->dir->i_version++;
3373 ent->dir->i_ctime = ent->dir->i_mtime =
eeca7ea1 3374 current_time(ent->dir);
bd1af145
MS
3375 ext4_mark_inode_dirty(handle, ent->dir);
3376 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3377 if (!ent->inlined) {
3378 retval = ext4_handle_dirty_dirent_node(handle,
3379 ent->dir, ent->bh);
3380 if (unlikely(retval)) {
3381 ext4_std_error(ent->dir->i_sb, retval);
3382 return retval;
3383 }
3384 }
3385 brelse(ent->bh);
3386 ent->bh = NULL;
3387
3388 return 0;
3389}
3390
3391static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3392 const struct qstr *d_name)
3393{
3394 int retval = -ENOENT;
3395 struct buffer_head *bh;
3396 struct ext4_dir_entry_2 *de;
3397
3398 bh = ext4_find_entry(dir, d_name, &de, NULL);
36de9286
TT
3399 if (IS_ERR(bh))
3400 return PTR_ERR(bh);
bd1af145
MS
3401 if (bh) {
3402 retval = ext4_delete_entry(handle, dir, de, bh);
3403 brelse(bh);
3404 }
3405 return retval;
3406}
3407
d80d448c
DW
3408static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3409 int force_reread)
bd1af145
MS
3410{
3411 int retval;
3412 /*
3413 * ent->de could have moved from under us during htree split, so make
3414 * sure that we are deleting the right entry. We might also be pointing
3415 * to a stale entry in the unused part of ent->bh so just checking inum
3416 * and the name isn't enough.
3417 */
3418 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3419 ent->de->name_len != ent->dentry->d_name.len ||
3420 strncmp(ent->de->name, ent->dentry->d_name.name,
d80d448c
DW
3421 ent->de->name_len) ||
3422 force_reread) {
bd1af145
MS
3423 retval = ext4_find_delete_entry(handle, ent->dir,
3424 &ent->dentry->d_name);
3425 } else {
3426 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3427 if (retval == -ENOENT) {
3428 retval = ext4_find_delete_entry(handle, ent->dir,
3429 &ent->dentry->d_name);
3430 }
3431 }
3432
3433 if (retval) {
b03a2f7e
AD
3434 ext4_warning_inode(ent->dir,
3435 "Deleting old file: nlink %d, error=%d",
3436 ent->dir->i_nlink, retval);
bd1af145
MS
3437 }
3438}
3439
bd42998a
MS
3440static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3441{
3442 if (ent->dir_nlink_delta) {
3443 if (ent->dir_nlink_delta == -1)
3444 ext4_dec_count(handle, ent->dir);
3445 else
3446 ext4_inc_count(handle, ent->dir);
3447 ext4_mark_inode_dirty(handle, ent->dir);
3448 }
3449}
3450
cd808dec
MS
3451static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3452 int credits, handle_t **h)
3453{
3454 struct inode *wh;
3455 handle_t *handle;
3456 int retries = 0;
3457
3458 /*
3459 * for inode block, sb block, group summaries,
3460 * and inode bitmap
3461 */
3462 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3463 EXT4_XATTR_TRANS_BLOCKS + 4);
3464retry:
3465 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3466 &ent->dentry->d_name, 0, NULL,
3467 EXT4_HT_DIR, credits);
3468
3469 handle = ext4_journal_current_handle();
3470 if (IS_ERR(wh)) {
3471 if (handle)
3472 ext4_journal_stop(handle);
3473 if (PTR_ERR(wh) == -ENOSPC &&
3474 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3475 goto retry;
3476 } else {
3477 *h = handle;
3478 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3479 wh->i_op = &ext4_special_inode_operations;
3480 }
3481 return wh;
3482}
3483
ac27a0ec
DK
3484/*
3485 * Anybody can rename anything with this: the permission checks are left to the
3486 * higher-level routines.
0e202704
TT
3487 *
3488 * n.b. old_{dentry,inode) refers to the source dentry/inode
3489 * while new_{dentry,inode) refers to the destination dentry/inode
3490 * This comes from rename(const char *oldpath, const char *newpath)
ac27a0ec 3491 */
af5bc92d 3492static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
cd808dec
MS
3493 struct inode *new_dir, struct dentry *new_dentry,
3494 unsigned int flags)
ac27a0ec 3495{
5b61de75 3496 handle_t *handle = NULL;
c0d268c3
MS
3497 struct ext4_renament old = {
3498 .dir = old_dir,
3499 .dentry = old_dentry,
2b0143b5 3500 .inode = d_inode(old_dentry),
c0d268c3
MS
3501 };
3502 struct ext4_renament new = {
3503 .dir = new_dir,
3504 .dentry = new_dentry,
2b0143b5 3505 .inode = d_inode(new_dentry),
c0d268c3 3506 };
d80d448c 3507 int force_reread;
0e202704 3508 int retval;
cd808dec
MS
3509 struct inode *whiteout = NULL;
3510 int credits;
3511 u8 old_file_type;
907f4554 3512
040cb378
LX
3513 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3514 (!projid_eq(EXT4_I(new_dir)->i_projid,
3515 EXT4_I(old_dentry->d_inode)->i_projid)))
3516 return -EXDEV;
3517
173b8439
TT
3518 if ((ext4_encrypted_inode(old_dir) &&
3519 !fscrypt_has_encryption_key(old_dir)) ||
3520 (ext4_encrypted_inode(new_dir) &&
3521 !fscrypt_has_encryption_key(new_dir)))
3522 return -ENOKEY;
3523
a7cdadee
JK
3524 retval = dquot_initialize(old.dir);
3525 if (retval)
3526 return retval;
3527 retval = dquot_initialize(new.dir);
3528 if (retval)
3529 return retval;
ac27a0ec
DK
3530
3531 /* Initialize quotas before so that eventual writes go
3532 * in separate transaction */
a7cdadee
JK
3533 if (new.inode) {
3534 retval = dquot_initialize(new.inode);
3535 if (retval)
3536 return retval;
3537 }
ac27a0ec 3538
c0d268c3 3539 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
36de9286
TT
3540 if (IS_ERR(old.bh))
3541 return PTR_ERR(old.bh);
ac27a0ec
DK
3542 /*
3543 * Check for inode number is _not_ due to possible IO errors.
3544 * We might rmdir the source, keep it as pwd of some process
3545 * and merrily kill the link to whatever was created under the
3546 * same name. Goodbye sticky bit ;-<
3547 */
ac27a0ec 3548 retval = -ENOENT;
c0d268c3 3549 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
ac27a0ec
DK
3550 goto end_rename;
3551
d9cdc903
TT
3552 if ((old.dir != new.dir) &&
3553 ext4_encrypted_inode(new.dir) &&
a7550b30 3554 !fscrypt_has_permitted_context(new.dir, old.inode)) {
d9cdc903
TT
3555 retval = -EPERM;
3556 goto end_rename;
3557 }
3558
c0d268c3
MS
3559 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3560 &new.de, &new.inlined);
36de9286
TT
3561 if (IS_ERR(new.bh)) {
3562 retval = PTR_ERR(new.bh);
a9cfcd63 3563 new.bh = NULL;
36de9286
TT
3564 goto end_rename;
3565 }
c0d268c3
MS
3566 if (new.bh) {
3567 if (!new.inode) {
3568 brelse(new.bh);
3569 new.bh = NULL;
ac27a0ec
DK
3570 }
3571 }
c0d268c3
MS
3572 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3573 ext4_alloc_da_blocks(old.inode);
5b61de75 3574
cd808dec
MS
3575 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3576 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3577 if (!(flags & RENAME_WHITEOUT)) {
3578 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
7071b715
KK
3579 if (IS_ERR(handle)) {
3580 retval = PTR_ERR(handle);
3581 handle = NULL;
3582 goto end_rename;
3583 }
cd808dec
MS
3584 } else {
3585 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
7071b715
KK
3586 if (IS_ERR(whiteout)) {
3587 retval = PTR_ERR(whiteout);
3588 whiteout = NULL;
3589 goto end_rename;
3590 }
cd808dec 3591 }
5b61de75 3592
c0d268c3 3593 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
5b61de75
TT
3594 ext4_handle_sync(handle);
3595
c0d268c3
MS
3596 if (S_ISDIR(old.inode->i_mode)) {
3597 if (new.inode) {
ac27a0ec 3598 retval = -ENOTEMPTY;
e875a2dd 3599 if (!ext4_empty_dir(new.inode))
ac27a0ec 3600 goto end_rename;
0d7d5d67
MS
3601 } else {
3602 retval = -EMLINK;
3603 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3604 goto end_rename;
ac27a0ec 3605 }
bd1af145 3606 retval = ext4_rename_dir_prepare(handle, &old);
ef607893
AG
3607 if (retval)
3608 goto end_rename;
ac27a0ec 3609 }
d80d448c
DW
3610 /*
3611 * If we're renaming a file within an inline_data dir and adding or
3612 * setting the new dirent causes a conversion from inline_data to
3613 * extents/blockmap, we need to force the dirent delete code to
3614 * re-read the directory, or else we end up trying to delete a dirent
3615 * from what is now the extent tree root (or a block map).
3616 */
3617 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3618 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
cd808dec
MS
3619
3620 old_file_type = old.de->file_type;
3621 if (whiteout) {
3622 /*
3623 * Do this before adding a new entry, so the old entry is sure
3624 * to be still pointing to the valid old entry.
3625 */
3626 retval = ext4_setent(handle, &old, whiteout->i_ino,
3627 EXT4_FT_CHRDEV);
3628 if (retval)
3629 goto end_rename;
3630 ext4_mark_inode_dirty(handle, whiteout);
3631 }
c0d268c3
MS
3632 if (!new.bh) {
3633 retval = ext4_add_entry(handle, new.dentry, old.inode);
ac27a0ec
DK
3634 if (retval)
3635 goto end_rename;
3636 } else {
bd1af145 3637 retval = ext4_setent(handle, &new,
cd808dec 3638 old.inode->i_ino, old_file_type);
ef607893
AG
3639 if (retval)
3640 goto end_rename;
ac27a0ec 3641 }
d80d448c
DW
3642 if (force_reread)
3643 force_reread = !ext4_test_inode_flag(new.dir,
3644 EXT4_INODE_INLINE_DATA);
ac27a0ec
DK
3645
3646 /*
3647 * Like most other Unix systems, set the ctime for inodes on a
3648 * rename.
3649 */
eeca7ea1 3650 old.inode->i_ctime = current_time(old.inode);
c0d268c3 3651 ext4_mark_inode_dirty(handle, old.inode);
ac27a0ec 3652
cd808dec
MS
3653 if (!whiteout) {
3654 /*
3655 * ok, that's it
3656 */
3657 ext4_rename_delete(handle, &old, force_reread);
3658 }
ac27a0ec 3659
c0d268c3
MS
3660 if (new.inode) {
3661 ext4_dec_count(handle, new.inode);
eeca7ea1 3662 new.inode->i_ctime = current_time(new.inode);
ac27a0ec 3663 }
eeca7ea1 3664 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
c0d268c3
MS
3665 ext4_update_dx_flag(old.dir);
3666 if (old.dir_bh) {
bd1af145
MS
3667 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3668 if (retval)
b4097142 3669 goto end_rename;
bd1af145 3670
c0d268c3
MS
3671 ext4_dec_count(handle, old.dir);
3672 if (new.inode) {
e875a2dd
MH
3673 /* checked ext4_empty_dir above, can't have another
3674 * parent, ext4_dec_count() won't work for many-linked
3675 * dirs */
c0d268c3 3676 clear_nlink(new.inode);
ac27a0ec 3677 } else {
c0d268c3
MS
3678 ext4_inc_count(handle, new.dir);
3679 ext4_update_dx_flag(new.dir);
3680 ext4_mark_inode_dirty(handle, new.dir);
ac27a0ec
DK
3681 }
3682 }
c0d268c3
MS
3683 ext4_mark_inode_dirty(handle, old.dir);
3684 if (new.inode) {
3685 ext4_mark_inode_dirty(handle, new.inode);
3686 if (!new.inode->i_nlink)
3687 ext4_orphan_add(handle, new.inode);
ac27a0ec
DK
3688 }
3689 retval = 0;
3690
3691end_rename:
c0d268c3
MS
3692 brelse(old.dir_bh);
3693 brelse(old.bh);
3694 brelse(new.bh);
cd808dec
MS
3695 if (whiteout) {
3696 if (retval)
3697 drop_nlink(whiteout);
3698 unlock_new_inode(whiteout);
3699 iput(whiteout);
3700 }
5b61de75
TT
3701 if (handle)
3702 ext4_journal_stop(handle);
ac27a0ec
DK
3703 return retval;
3704}
3705
bd42998a
MS
3706static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3707 struct inode *new_dir, struct dentry *new_dentry)
3708{
3709 handle_t *handle = NULL;
3710 struct ext4_renament old = {
3711 .dir = old_dir,
3712 .dentry = old_dentry,
2b0143b5 3713 .inode = d_inode(old_dentry),
bd42998a
MS
3714 };
3715 struct ext4_renament new = {
3716 .dir = new_dir,
3717 .dentry = new_dentry,
2b0143b5 3718 .inode = d_inode(new_dentry),
bd42998a
MS
3719 };
3720 u8 new_file_type;
3721 int retval;
eeca7ea1 3722 struct timespec ctime;
bd42998a 3723
173b8439
TT
3724 if ((ext4_encrypted_inode(old_dir) &&
3725 !fscrypt_has_encryption_key(old_dir)) ||
3726 (ext4_encrypted_inode(new_dir) &&
3727 !fscrypt_has_encryption_key(new_dir)))
3728 return -ENOKEY;
3729
c2faccaf
TT
3730 if ((ext4_encrypted_inode(old_dir) ||
3731 ext4_encrypted_inode(new_dir)) &&
3732 (old_dir != new_dir) &&
a7550b30
JK
3733 (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3734 !fscrypt_has_permitted_context(old_dir, new.inode)))
c2faccaf
TT
3735 return -EPERM;
3736
040cb378
LX
3737 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3738 !projid_eq(EXT4_I(new_dir)->i_projid,
3739 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3740 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3741 !projid_eq(EXT4_I(old_dir)->i_projid,
3742 EXT4_I(new_dentry->d_inode)->i_projid)))
3743 return -EXDEV;
3744
a7cdadee
JK
3745 retval = dquot_initialize(old.dir);
3746 if (retval)
3747 return retval;
3748 retval = dquot_initialize(new.dir);
3749 if (retval)
3750 return retval;
bd42998a
MS
3751
3752 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3753 &old.de, &old.inlined);
36de9286
TT
3754 if (IS_ERR(old.bh))
3755 return PTR_ERR(old.bh);
bd42998a
MS
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 */
3762 retval = -ENOENT;
3763 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3764 goto end_rename;
3765
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 }
bd42998a
MS
3773
3774 /* RENAME_EXCHANGE case: old *and* new must both exist */
3775 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3776 goto end_rename;
3777
3778 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3779 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3780 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
7071b715
KK
3781 if (IS_ERR(handle)) {
3782 retval = PTR_ERR(handle);
3783 handle = NULL;
3784 goto end_rename;
3785 }
bd42998a
MS
3786
3787 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3788 ext4_handle_sync(handle);
3789
3790 if (S_ISDIR(old.inode->i_mode)) {
3791 old.is_dir = true;
3792 retval = ext4_rename_dir_prepare(handle, &old);
3793 if (retval)
3794 goto end_rename;
3795 }
3796 if (S_ISDIR(new.inode->i_mode)) {
3797 new.is_dir = true;
3798 retval = ext4_rename_dir_prepare(handle, &new);
3799 if (retval)
3800 goto end_rename;
3801 }
3802
3803 /*
3804 * Other than the special case of overwriting a directory, parents'
3805 * nlink only needs to be modified if this is a cross directory rename.
3806 */
3807 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3808 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3809 new.dir_nlink_delta = -old.dir_nlink_delta;
3810 retval = -EMLINK;
3811 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3812 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3813 goto end_rename;
3814 }
3815
3816 new_file_type = new.de->file_type;
3817 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3818 if (retval)
3819 goto end_rename;
3820
3821 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3822 if (retval)
3823 goto end_rename;
3824
3825 /*
3826 * Like most other Unix systems, set the ctime for inodes on a
3827 * rename.
3828 */
eeca7ea1
DD
3829 ctime = current_time(old.inode);
3830 old.inode->i_ctime = ctime;
3831 new.inode->i_ctime = ctime;
bd42998a
MS
3832 ext4_mark_inode_dirty(handle, old.inode);
3833 ext4_mark_inode_dirty(handle, new.inode);
3834
3835 if (old.dir_bh) {
3836 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3837 if (retval)
3838 goto end_rename;
3839 }
3840 if (new.dir_bh) {
3841 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3842 if (retval)
3843 goto end_rename;
3844 }
3845 ext4_update_dir_count(handle, &old);
3846 ext4_update_dir_count(handle, &new);
3847 retval = 0;
3848
3849end_rename:
3850 brelse(old.dir_bh);
3851 brelse(new.dir_bh);
3852 brelse(old.bh);
3853 brelse(new.bh);
3854 if (handle)
3855 ext4_journal_stop(handle);
3856 return retval;
3857}
3858
0a7c3937
MS
3859static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3860 struct inode *new_dir, struct dentry *new_dentry,
3861 unsigned int flags)
3862{
0db1ff22
TT
3863 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3864 return -EIO;
3865
cd808dec 3866 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
0a7c3937
MS
3867 return -EINVAL;
3868
bd42998a
MS
3869 if (flags & RENAME_EXCHANGE) {
3870 return ext4_cross_rename(old_dir, old_dentry,
3871 new_dir, new_dentry);
3872 }
cd808dec
MS
3873
3874 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
0a7c3937
MS
3875}
3876
ac27a0ec
DK
3877/*
3878 * directories can handle most operations...
3879 */
754661f1 3880const struct inode_operations ext4_dir_inode_operations = {
617ba13b
MC
3881 .create = ext4_create,
3882 .lookup = ext4_lookup,
3883 .link = ext4_link,
3884 .unlink = ext4_unlink,
3885 .symlink = ext4_symlink,
3886 .mkdir = ext4_mkdir,
3887 .rmdir = ext4_rmdir,
3888 .mknod = ext4_mknod,
af51a2ac 3889 .tmpfile = ext4_tmpfile,
2773bf00 3890 .rename = ext4_rename2,
617ba13b 3891 .setattr = ext4_setattr,
99652ea5 3892 .getattr = ext4_getattr,
617ba13b 3893 .listxattr = ext4_listxattr,
4e34e719 3894 .get_acl = ext4_get_acl,
64e178a7 3895 .set_acl = ext4_set_acl,
abc8746e 3896 .fiemap = ext4_fiemap,
ac27a0ec
DK
3897};
3898
754661f1 3899const struct inode_operations ext4_special_inode_operations = {
617ba13b 3900 .setattr = ext4_setattr,
99652ea5 3901 .getattr = ext4_getattr,
617ba13b 3902 .listxattr = ext4_listxattr,
4e34e719 3903 .get_acl = ext4_get_acl,
64e178a7 3904 .set_acl = ext4_set_acl,
ac27a0ec 3905};