1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/namei.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/namei.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
25 * Theodore Ts'o, 2002
26 */
27
28 #include <linux/fs.h>
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include <linux/iversion.h>
38 #include <linux/unicode.h>
39 #include "ext4.h"
40 #include "ext4_jbd2.h"
41
42 #include "xattr.h"
43 #include "acl.h"
44
45 #include <trace/events/ext4.h>
46 /*
47 * define how far ahead to read directories while searching them.
48 */
49 #define NAMEI_RA_CHUNKS 2
50 #define NAMEI_RA_BLOCKS 4
51 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52
ext4_append(handle_t * handle,struct inode * inode,ext4_lblk_t * block)53 static struct buffer_head *ext4_append(handle_t *handle,
54 struct inode *inode,
55 ext4_lblk_t *block)
56 {
57 struct ext4_map_blocks map;
58 struct buffer_head *bh;
59 int err;
60
61 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
62 ((inode->i_size >> 10) >=
63 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
64 return ERR_PTR(-ENOSPC);
65
66 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67 map.m_lblk = *block;
68 map.m_len = 1;
69
70 /*
71 * We're appending new directory block. Make sure the block is not
72 * allocated yet, otherwise we will end up corrupting the
73 * directory.
74 */
75 err = ext4_map_blocks(NULL, inode, &map, 0);
76 if (err < 0)
77 return ERR_PTR(err);
78 if (err) {
79 EXT4_ERROR_INODE(inode, "Logical block already allocated");
80 return ERR_PTR(-EFSCORRUPTED);
81 }
82
83 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
84 if (IS_ERR(bh))
85 return bh;
86 inode->i_size += inode->i_sb->s_blocksize;
87 EXT4_I(inode)->i_disksize = inode->i_size;
88 err = ext4_mark_inode_dirty(handle, inode);
89 if (err)
90 goto out;
91 BUFFER_TRACE(bh, "get_write_access");
92 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
93 EXT4_JTR_NONE);
94 if (err)
95 goto out;
96 return bh;
97
98 out:
99 brelse(bh);
100 ext4_std_error(inode->i_sb, err);
101 return ERR_PTR(err);
102 }
103
104 static int ext4_dx_csum_verify(struct inode *inode,
105 struct ext4_dir_entry *dirent);
106
107 /*
108 * Hints to ext4_read_dirblock regarding whether we expect a directory
109 * block being read to be an index block, or a block containing
110 * directory entries (and if the latter, whether it was found via a
111 * logical block in an htree index block). This is used to control
112 * what sort of sanity checkinig ext4_read_dirblock() will do on the
113 * directory block read from the storage device. EITHER will means
114 * the caller doesn't know what kind of directory block will be read,
115 * so no specific verification will be done.
116 */
117 typedef enum {
118 EITHER, INDEX, DIRENT, DIRENT_HTREE
119 } dirblock_type_t;
120
121 #define ext4_read_dirblock(inode, block, type) \
122 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
123
__ext4_read_dirblock(struct inode * inode,ext4_lblk_t block,dirblock_type_t type,const char * func,unsigned int line)124 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
125 ext4_lblk_t block,
126 dirblock_type_t type,
127 const char *func,
128 unsigned int line)
129 {
130 struct buffer_head *bh;
131 struct ext4_dir_entry *dirent;
132 int is_dx_block = 0;
133
134 if (block >= inode->i_size >> inode->i_blkbits) {
135 ext4_error_inode(inode, func, line, block,
136 "Attempting to read directory block (%u) that is past i_size (%llu)",
137 block, inode->i_size);
138 return ERR_PTR(-EFSCORRUPTED);
139 }
140
141 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
142 bh = ERR_PTR(-EIO);
143 else
144 bh = ext4_bread(NULL, inode, block, 0);
145 if (IS_ERR(bh)) {
146 __ext4_warning(inode->i_sb, func, line,
147 "inode #%lu: lblock %lu: comm %s: "
148 "error %ld reading directory block",
149 inode->i_ino, (unsigned long)block,
150 current->comm, PTR_ERR(bh));
151
152 return bh;
153 }
154 /* The first directory block must not be a hole. */
155 if (!bh && (type == INDEX || type == DIRENT_HTREE || block == 0)) {
156 ext4_error_inode(inode, func, line, block,
157 "Directory hole found for htree %s block %u",
158 (type == INDEX) ? "index" : "leaf", block);
159 return ERR_PTR(-EFSCORRUPTED);
160 }
161 if (!bh)
162 return NULL;
163 dirent = (struct ext4_dir_entry *) bh->b_data;
164 /* Determine whether or not we have an index block */
165 if (is_dx(inode)) {
166 if (block == 0)
167 is_dx_block = 1;
168 else if (ext4_rec_len_from_disk(dirent->rec_len,
169 inode->i_sb->s_blocksize) ==
170 inode->i_sb->s_blocksize)
171 is_dx_block = 1;
172 }
173 if (!is_dx_block && type == INDEX) {
174 ext4_error_inode(inode, func, line, block,
175 "directory leaf block found instead of index block");
176 brelse(bh);
177 return ERR_PTR(-EFSCORRUPTED);
178 }
179 if (!ext4_has_metadata_csum(inode->i_sb) ||
180 buffer_verified(bh))
181 return bh;
182
183 /*
184 * An empty leaf block can get mistaken for a index block; for
185 * this reason, we can only check the index checksum when the
186 * caller is sure it should be an index block.
187 */
188 if (is_dx_block && type == INDEX) {
189 if (ext4_dx_csum_verify(inode, dirent) &&
190 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
191 set_buffer_verified(bh);
192 else {
193 ext4_error_inode_err(inode, func, line, block,
194 EFSBADCRC,
195 "Directory index failed checksum");
196 brelse(bh);
197 return ERR_PTR(-EFSBADCRC);
198 }
199 }
200 if (!is_dx_block) {
201 if (ext4_dirblock_csum_verify(inode, bh) &&
202 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
203 set_buffer_verified(bh);
204 else {
205 ext4_error_inode_err(inode, func, line, block,
206 EFSBADCRC,
207 "Directory block failed checksum");
208 brelse(bh);
209 return ERR_PTR(-EFSBADCRC);
210 }
211 }
212 return bh;
213 }
214
215 #ifdef DX_DEBUG
216 #define dxtrace(command) command
217 #else
218 #define dxtrace(command)
219 #endif
220
221 struct fake_dirent
222 {
223 __le32 inode;
224 __le16 rec_len;
225 u8 name_len;
226 u8 file_type;
227 };
228
229 struct dx_countlimit
230 {
231 __le16 limit;
232 __le16 count;
233 };
234
235 struct dx_entry
236 {
237 __le32 hash;
238 __le32 block;
239 };
240
241 /*
242 * dx_root_info is laid out so that if it should somehow get overlaid by a
243 * dirent the two low bits of the hash version will be zero. Therefore, the
244 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
245 */
246
247 struct dx_root
248 {
249 struct fake_dirent dot;
250 char dot_name[4];
251 struct fake_dirent dotdot;
252 char dotdot_name[4];
253 struct dx_root_info
254 {
255 __le32 reserved_zero;
256 u8 hash_version;
257 u8 info_length; /* 8 */
258 u8 indirect_levels;
259 u8 unused_flags;
260 }
261 info;
262 struct dx_entry entries[];
263 };
264
265 struct dx_node
266 {
267 struct fake_dirent fake;
268 struct dx_entry entries[];
269 };
270
271
272 struct dx_frame
273 {
274 struct buffer_head *bh;
275 struct dx_entry *entries;
276 struct dx_entry *at;
277 };
278
279 struct dx_map_entry
280 {
281 u32 hash;
282 u16 offs;
283 u16 size;
284 };
285
286 /*
287 * This goes at the end of each htree block.
288 */
289 struct dx_tail {
290 u32 dt_reserved;
291 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
292 };
293
294 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
295 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
296 static inline unsigned dx_get_hash(struct dx_entry *entry);
297 static void dx_set_hash(struct dx_entry *entry, unsigned value);
298 static unsigned dx_get_count(struct dx_entry *entries);
299 static unsigned dx_get_limit(struct dx_entry *entries);
300 static void dx_set_count(struct dx_entry *entries, unsigned value);
301 static void dx_set_limit(struct dx_entry *entries, unsigned value);
302 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
303 static unsigned dx_node_limit(struct inode *dir);
304 static struct dx_frame *dx_probe(struct ext4_filename *fname,
305 struct inode *dir,
306 struct dx_hash_info *hinfo,
307 struct dx_frame *frame);
308 static void dx_release(struct dx_frame *frames);
309 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
310 struct dx_hash_info *hinfo,
311 struct dx_map_entry *map_tail);
312 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
313 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
314 char *to, struct dx_map_entry *offsets,
315 int count, unsigned int blocksize);
316 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
317 unsigned int blocksize);
318 static void dx_insert_block(struct dx_frame *frame,
319 u32 hash, ext4_lblk_t block);
320 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
321 struct dx_frame *frame,
322 struct dx_frame *frames,
323 __u32 *start_hash);
324 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
325 struct ext4_filename *fname,
326 struct ext4_dir_entry_2 **res_dir);
327 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
328 struct inode *dir, struct inode *inode);
329
330 /* checksumming functions */
ext4_initialize_dirent_tail(struct buffer_head * bh,unsigned int blocksize)331 void ext4_initialize_dirent_tail(struct buffer_head *bh,
332 unsigned int blocksize)
333 {
334 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
335
336 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
337 t->det_rec_len = ext4_rec_len_to_disk(
338 sizeof(struct ext4_dir_entry_tail), blocksize);
339 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
340 }
341
342 /* Walk through a dirent block to find a checksum "dirent" at the tail */
get_dirent_tail(struct inode * inode,struct buffer_head * bh)343 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
344 struct buffer_head *bh)
345 {
346 struct ext4_dir_entry_tail *t;
347 int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
348
349 #ifdef PARANOID
350 struct ext4_dir_entry *d, *top;
351
352 d = (struct ext4_dir_entry *)bh->b_data;
353 top = (struct ext4_dir_entry *)(bh->b_data +
354 (blocksize - sizeof(struct ext4_dir_entry_tail)));
355 while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
356 d = (struct ext4_dir_entry *)(((void *)d) +
357 ext4_rec_len_from_disk(d->rec_len, blocksize));
358
359 if (d != top)
360 return NULL;
361
362 t = (struct ext4_dir_entry_tail *)d;
363 #else
364 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
365 #endif
366
367 if (t->det_reserved_zero1 ||
368 (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
369 sizeof(struct ext4_dir_entry_tail)) ||
370 t->det_reserved_zero2 ||
371 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
372 return NULL;
373
374 return t;
375 }
376
ext4_dirblock_csum(struct inode * inode,void * dirent,int size)377 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
378 {
379 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
380 struct ext4_inode_info *ei = EXT4_I(inode);
381 __u32 csum;
382
383 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
384 return cpu_to_le32(csum);
385 }
386
387 #define warn_no_space_for_csum(inode) \
388 __warn_no_space_for_csum((inode), __func__, __LINE__)
389
__warn_no_space_for_csum(struct inode * inode,const char * func,unsigned int line)390 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
391 unsigned int line)
392 {
393 __ext4_warning_inode(inode, func, line,
394 "No space for directory leaf checksum. Please run e2fsck -D.");
395 }
396
ext4_dirblock_csum_verify(struct inode * inode,struct buffer_head * bh)397 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
398 {
399 struct ext4_dir_entry_tail *t;
400
401 if (!ext4_has_metadata_csum(inode->i_sb))
402 return 1;
403
404 t = get_dirent_tail(inode, bh);
405 if (!t) {
406 warn_no_space_for_csum(inode);
407 return 0;
408 }
409
410 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
411 (char *)t - bh->b_data))
412 return 0;
413
414 return 1;
415 }
416
ext4_dirblock_csum_set(struct inode * inode,struct buffer_head * bh)417 static void ext4_dirblock_csum_set(struct inode *inode,
418 struct buffer_head *bh)
419 {
420 struct ext4_dir_entry_tail *t;
421
422 if (!ext4_has_metadata_csum(inode->i_sb))
423 return;
424
425 t = get_dirent_tail(inode, bh);
426 if (!t) {
427 warn_no_space_for_csum(inode);
428 return;
429 }
430
431 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
432 (char *)t - bh->b_data);
433 }
434
ext4_handle_dirty_dirblock(handle_t * handle,struct inode * inode,struct buffer_head * bh)435 int ext4_handle_dirty_dirblock(handle_t *handle,
436 struct inode *inode,
437 struct buffer_head *bh)
438 {
439 ext4_dirblock_csum_set(inode, bh);
440 return ext4_handle_dirty_metadata(handle, inode, bh);
441 }
442
get_dx_countlimit(struct inode * inode,struct ext4_dir_entry * dirent,int * offset)443 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
444 struct ext4_dir_entry *dirent,
445 int *offset)
446 {
447 struct ext4_dir_entry *dp;
448 struct dx_root_info *root;
449 int count_offset;
450 int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
451 unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
452
453 if (rlen == blocksize)
454 count_offset = 8;
455 else if (rlen == 12) {
456 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
457 if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
458 return NULL;
459 root = (struct dx_root_info *)(((void *)dp + 12));
460 if (root->reserved_zero ||
461 root->info_length != sizeof(struct dx_root_info))
462 return NULL;
463 count_offset = 32;
464 } else
465 return NULL;
466
467 if (offset)
468 *offset = count_offset;
469 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
470 }
471
ext4_dx_csum(struct inode * inode,struct ext4_dir_entry * dirent,int count_offset,int count,struct dx_tail * t)472 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
473 int count_offset, int count, struct dx_tail *t)
474 {
475 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
476 struct ext4_inode_info *ei = EXT4_I(inode);
477 __u32 csum;
478 int size;
479 __u32 dummy_csum = 0;
480 int offset = offsetof(struct dx_tail, dt_checksum);
481
482 size = count_offset + (count * sizeof(struct dx_entry));
483 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
484 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
485 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
486
487 return cpu_to_le32(csum);
488 }
489
ext4_dx_csum_verify(struct inode * inode,struct ext4_dir_entry * dirent)490 static int ext4_dx_csum_verify(struct inode *inode,
491 struct ext4_dir_entry *dirent)
492 {
493 struct dx_countlimit *c;
494 struct dx_tail *t;
495 int count_offset, limit, count;
496
497 if (!ext4_has_metadata_csum(inode->i_sb))
498 return 1;
499
500 c = get_dx_countlimit(inode, dirent, &count_offset);
501 if (!c) {
502 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
503 return 0;
504 }
505 limit = le16_to_cpu(c->limit);
506 count = le16_to_cpu(c->count);
507 if (count_offset + (limit * sizeof(struct dx_entry)) >
508 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
509 warn_no_space_for_csum(inode);
510 return 0;
511 }
512 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
513
514 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
515 count, t))
516 return 0;
517 return 1;
518 }
519
ext4_dx_csum_set(struct inode * inode,struct ext4_dir_entry * dirent)520 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
521 {
522 struct dx_countlimit *c;
523 struct dx_tail *t;
524 int count_offset, limit, count;
525
526 if (!ext4_has_metadata_csum(inode->i_sb))
527 return;
528
529 c = get_dx_countlimit(inode, dirent, &count_offset);
530 if (!c) {
531 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
532 return;
533 }
534 limit = le16_to_cpu(c->limit);
535 count = le16_to_cpu(c->count);
536 if (count_offset + (limit * sizeof(struct dx_entry)) >
537 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
538 warn_no_space_for_csum(inode);
539 return;
540 }
541 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
542
543 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
544 }
545
ext4_handle_dirty_dx_node(handle_t * handle,struct inode * inode,struct buffer_head * bh)546 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
547 struct inode *inode,
548 struct buffer_head *bh)
549 {
550 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
551 return ext4_handle_dirty_metadata(handle, inode, bh);
552 }
553
554 /*
555 * p is at least 6 bytes before the end of page
556 */
557 static inline struct ext4_dir_entry_2 *
ext4_next_entry(struct ext4_dir_entry_2 * p,unsigned long blocksize)558 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
559 {
560 return (struct ext4_dir_entry_2 *)((char *)p +
561 ext4_rec_len_from_disk(p->rec_len, blocksize));
562 }
563
564 /*
565 * Future: use high four bits of block for coalesce-on-delete flags
566 * Mask them off for now.
567 */
568
dx_get_block(struct dx_entry * entry)569 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
570 {
571 return le32_to_cpu(entry->block) & 0x0fffffff;
572 }
573
dx_set_block(struct dx_entry * entry,ext4_lblk_t value)574 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
575 {
576 entry->block = cpu_to_le32(value);
577 }
578
dx_get_hash(struct dx_entry * entry)579 static inline unsigned dx_get_hash(struct dx_entry *entry)
580 {
581 return le32_to_cpu(entry->hash);
582 }
583
dx_set_hash(struct dx_entry * entry,unsigned value)584 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
585 {
586 entry->hash = cpu_to_le32(value);
587 }
588
dx_get_count(struct dx_entry * entries)589 static inline unsigned dx_get_count(struct dx_entry *entries)
590 {
591 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
592 }
593
dx_get_limit(struct dx_entry * entries)594 static inline unsigned dx_get_limit(struct dx_entry *entries)
595 {
596 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
597 }
598
dx_set_count(struct dx_entry * entries,unsigned value)599 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
600 {
601 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
602 }
603
dx_set_limit(struct dx_entry * entries,unsigned value)604 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
605 {
606 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
607 }
608
dx_root_limit(struct inode * dir,unsigned infosize)609 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
610 {
611 unsigned int entry_space = dir->i_sb->s_blocksize -
612 ext4_dir_rec_len(1, NULL) -
613 ext4_dir_rec_len(2, NULL) - infosize;
614
615 if (ext4_has_metadata_csum(dir->i_sb))
616 entry_space -= sizeof(struct dx_tail);
617 return entry_space / sizeof(struct dx_entry);
618 }
619
dx_node_limit(struct inode * dir)620 static inline unsigned dx_node_limit(struct inode *dir)
621 {
622 unsigned int entry_space = dir->i_sb->s_blocksize -
623 ext4_dir_rec_len(0, dir);
624
625 if (ext4_has_metadata_csum(dir->i_sb))
626 entry_space -= sizeof(struct dx_tail);
627 return entry_space / sizeof(struct dx_entry);
628 }
629
630 /*
631 * Debug
632 */
633 #ifdef DX_DEBUG
dx_show_index(char * label,struct dx_entry * entries)634 static void dx_show_index(char * label, struct dx_entry *entries)
635 {
636 int i, n = dx_get_count (entries);
637 printk(KERN_DEBUG "%s index", label);
638 for (i = 0; i < n; i++) {
639 printk(KERN_CONT " %x->%lu",
640 i ? dx_get_hash(entries + i) : 0,
641 (unsigned long)dx_get_block(entries + i));
642 }
643 printk(KERN_CONT "\n");
644 }
645
646 struct stats
647 {
648 unsigned names;
649 unsigned space;
650 unsigned bcount;
651 };
652
dx_show_leaf(struct inode * dir,struct dx_hash_info * hinfo,struct ext4_dir_entry_2 * de,int size,int show_names)653 static struct stats dx_show_leaf(struct inode *dir,
654 struct dx_hash_info *hinfo,
655 struct ext4_dir_entry_2 *de,
656 int size, int show_names)
657 {
658 unsigned names = 0, space = 0;
659 char *base = (char *) de;
660 struct dx_hash_info h = *hinfo;
661
662 printk("names: ");
663 while ((char *) de < base + size)
664 {
665 if (de->inode)
666 {
667 if (show_names)
668 {
669 #ifdef CONFIG_FS_ENCRYPTION
670 int len;
671 char *name;
672 struct fscrypt_str fname_crypto_str =
673 FSTR_INIT(NULL, 0);
674 int res = 0;
675
676 name = de->name;
677 len = de->name_len;
678 if (!IS_ENCRYPTED(dir)) {
679 /* Directory is not encrypted */
680 (void) ext4fs_dirhash(dir, de->name,
681 de->name_len, &h);
682 printk("%*.s:(U)%x.%u ", len,
683 name, h.hash,
684 (unsigned) ((char *) de
685 - base));
686 } else {
687 struct fscrypt_str de_name =
688 FSTR_INIT(name, len);
689
690 /* Directory is encrypted */
691 res = fscrypt_fname_alloc_buffer(
692 len, &fname_crypto_str);
693 if (res)
694 printk(KERN_WARNING "Error "
695 "allocating crypto "
696 "buffer--skipping "
697 "crypto\n");
698 res = fscrypt_fname_disk_to_usr(dir,
699 0, 0, &de_name,
700 &fname_crypto_str);
701 if (res) {
702 printk(KERN_WARNING "Error "
703 "converting filename "
704 "from disk to usr"
705 "\n");
706 name = "??";
707 len = 2;
708 } else {
709 name = fname_crypto_str.name;
710 len = fname_crypto_str.len;
711 }
712 if (IS_CASEFOLDED(dir))
713 h.hash = EXT4_DIRENT_HASH(de);
714 else
715 (void) ext4fs_dirhash(dir,
716 de->name,
717 de->name_len, &h);
718 printk("%*.s:(E)%x.%u ", len, name,
719 h.hash, (unsigned) ((char *) de
720 - base));
721 fscrypt_fname_free_buffer(
722 &fname_crypto_str);
723 }
724 #else
725 int len = de->name_len;
726 char *name = de->name;
727 (void) ext4fs_dirhash(dir, de->name,
728 de->name_len, &h);
729 printk("%*.s:%x.%u ", len, name, h.hash,
730 (unsigned) ((char *) de - base));
731 #endif
732 }
733 space += ext4_dir_rec_len(de->name_len, dir);
734 names++;
735 }
736 de = ext4_next_entry(de, size);
737 }
738 printk(KERN_CONT "(%i)\n", names);
739 return (struct stats) { names, space, 1 };
740 }
741
dx_show_entries(struct dx_hash_info * hinfo,struct inode * dir,struct dx_entry * entries,int levels)742 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
743 struct dx_entry *entries, int levels)
744 {
745 unsigned blocksize = dir->i_sb->s_blocksize;
746 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
747 unsigned bcount = 0;
748 struct buffer_head *bh;
749 printk("%i indexed blocks...\n", count);
750 for (i = 0; i < count; i++, entries++)
751 {
752 ext4_lblk_t block = dx_get_block(entries);
753 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
754 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
755 struct stats stats;
756 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
757 bh = ext4_bread(NULL,dir, block, 0);
758 if (!bh || IS_ERR(bh))
759 continue;
760 stats = levels?
761 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
762 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
763 bh->b_data, blocksize, 0);
764 names += stats.names;
765 space += stats.space;
766 bcount += stats.bcount;
767 brelse(bh);
768 }
769 if (bcount)
770 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
771 levels ? "" : " ", names, space/bcount,
772 (space/bcount)*100/blocksize);
773 return (struct stats) { names, space, bcount};
774 }
775
776 /*
777 * Linear search cross check
778 */
htree_rep_invariant_check(struct dx_entry * at,struct dx_entry * target,u32 hash,unsigned int n)779 static inline void htree_rep_invariant_check(struct dx_entry *at,
780 struct dx_entry *target,
781 u32 hash, unsigned int n)
782 {
783 while (n--) {
784 dxtrace(printk(KERN_CONT ","));
785 if (dx_get_hash(++at) > hash) {
786 at--;
787 break;
788 }
789 }
790 ASSERT(at == target - 1);
791 }
792 #else /* DX_DEBUG */
htree_rep_invariant_check(struct dx_entry * at,struct dx_entry * target,u32 hash,unsigned int n)793 static inline void htree_rep_invariant_check(struct dx_entry *at,
794 struct dx_entry *target,
795 u32 hash, unsigned int n)
796 {
797 }
798 #endif /* DX_DEBUG */
799
800 /*
801 * Probe for a directory leaf block to search.
802 *
803 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
804 * error in the directory index, and the caller should fall back to
805 * searching the directory normally. The callers of dx_probe **MUST**
806 * check for this error code, and make sure it never gets reflected
807 * back to userspace.
808 */
809 static struct dx_frame *
dx_probe(struct ext4_filename * fname,struct inode * dir,struct dx_hash_info * hinfo,struct dx_frame * frame_in)810 dx_probe(struct ext4_filename *fname, struct inode *dir,
811 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
812 {
813 unsigned count, indirect, level, i;
814 struct dx_entry *at, *entries, *p, *q, *m;
815 struct dx_root *root;
816 struct dx_frame *frame = frame_in;
817 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
818 u32 hash;
819 ext4_lblk_t block;
820 ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
821
822 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
823 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
824 if (IS_ERR(frame->bh))
825 return (struct dx_frame *) frame->bh;
826
827 root = (struct dx_root *) frame->bh->b_data;
828 if (root->info.hash_version != DX_HASH_TEA &&
829 root->info.hash_version != DX_HASH_HALF_MD4 &&
830 root->info.hash_version != DX_HASH_LEGACY &&
831 root->info.hash_version != DX_HASH_SIPHASH) {
832 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
833 root->info.hash_version);
834 goto fail;
835 }
836 if (ext4_hash_in_dirent(dir)) {
837 if (root->info.hash_version != DX_HASH_SIPHASH) {
838 ext4_warning_inode(dir,
839 "Hash in dirent, but hash is not SIPHASH");
840 goto fail;
841 }
842 } else {
843 if (root->info.hash_version == DX_HASH_SIPHASH) {
844 ext4_warning_inode(dir,
845 "Hash code is SIPHASH, but hash not in dirent");
846 goto fail;
847 }
848 }
849 if (fname)
850 hinfo = &fname->hinfo;
851 hinfo->hash_version = root->info.hash_version;
852 if (hinfo->hash_version <= DX_HASH_TEA)
853 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
854 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
855 /* hash is already computed for encrypted casefolded directory */
856 if (fname && fname_name(fname) &&
857 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
858 int ret = ext4fs_dirhash(dir, fname_name(fname),
859 fname_len(fname), hinfo);
860 if (ret < 0) {
861 ret_err = ERR_PTR(ret);
862 goto fail;
863 }
864 }
865 hash = hinfo->hash;
866
867 if (root->info.unused_flags & 1) {
868 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
869 root->info.unused_flags);
870 goto fail;
871 }
872
873 indirect = root->info.indirect_levels;
874 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
875 ext4_warning(dir->i_sb,
876 "Directory (ino: %lu) htree depth %#06x exceed"
877 "supported value", dir->i_ino,
878 ext4_dir_htree_level(dir->i_sb));
879 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
880 ext4_warning(dir->i_sb, "Enable large directory "
881 "feature to access it");
882 }
883 goto fail;
884 }
885
886 entries = (struct dx_entry *)(((char *)&root->info) +
887 root->info.info_length);
888
889 if (dx_get_limit(entries) != dx_root_limit(dir,
890 root->info.info_length)) {
891 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
892 dx_get_limit(entries),
893 dx_root_limit(dir, root->info.info_length));
894 goto fail;
895 }
896
897 dxtrace(printk("Look up %x", hash));
898 level = 0;
899 blocks[0] = 0;
900 while (1) {
901 count = dx_get_count(entries);
902 if (!count || count > dx_get_limit(entries)) {
903 ext4_warning_inode(dir,
904 "dx entry: count %u beyond limit %u",
905 count, dx_get_limit(entries));
906 goto fail;
907 }
908
909 p = entries + 1;
910 q = entries + count - 1;
911 while (p <= q) {
912 m = p + (q - p) / 2;
913 dxtrace(printk(KERN_CONT "."));
914 if (dx_get_hash(m) > hash)
915 q = m - 1;
916 else
917 p = m + 1;
918 }
919
920 htree_rep_invariant_check(entries, p, hash, count - 1);
921
922 at = p - 1;
923 dxtrace(printk(KERN_CONT " %x->%u\n",
924 at == entries ? 0 : dx_get_hash(at),
925 dx_get_block(at)));
926 frame->entries = entries;
927 frame->at = at;
928
929 block = dx_get_block(at);
930 for (i = 0; i <= level; i++) {
931 if (blocks[i] == block) {
932 ext4_warning_inode(dir,
933 "dx entry: tree cycle block %u points back to block %u",
934 blocks[level], block);
935 goto fail;
936 }
937 }
938 if (++level > indirect)
939 return frame;
940 blocks[level] = block;
941 frame++;
942 frame->bh = ext4_read_dirblock(dir, block, INDEX);
943 if (IS_ERR(frame->bh)) {
944 ret_err = (struct dx_frame *) frame->bh;
945 frame->bh = NULL;
946 goto fail;
947 }
948
949 entries = ((struct dx_node *) frame->bh->b_data)->entries;
950
951 if (dx_get_limit(entries) != dx_node_limit(dir)) {
952 ext4_warning_inode(dir,
953 "dx entry: limit %u != node limit %u",
954 dx_get_limit(entries), dx_node_limit(dir));
955 goto fail;
956 }
957 }
958 fail:
959 while (frame >= frame_in) {
960 brelse(frame->bh);
961 frame--;
962 }
963
964 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
965 ext4_warning_inode(dir,
966 "Corrupt directory, running e2fsck is recommended");
967 return ret_err;
968 }
969
dx_release(struct dx_frame * frames)970 static void dx_release(struct dx_frame *frames)
971 {
972 struct dx_root_info *info;
973 int i;
974 unsigned int indirect_levels;
975
976 if (frames[0].bh == NULL)
977 return;
978
979 info = &((struct dx_root *)frames[0].bh->b_data)->info;
980 /* save local copy, "info" may be freed after brelse() */
981 indirect_levels = info->indirect_levels;
982 for (i = 0; i <= indirect_levels; i++) {
983 if (frames[i].bh == NULL)
984 break;
985 brelse(frames[i].bh);
986 frames[i].bh = NULL;
987 }
988 }
989
990 /*
991 * This function increments the frame pointer to search the next leaf
992 * block, and reads in the necessary intervening nodes if the search
993 * should be necessary. Whether or not the search is necessary is
994 * controlled by the hash parameter. If the hash value is even, then
995 * the search is only continued if the next block starts with that
996 * hash value. This is used if we are searching for a specific file.
997 *
998 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
999 *
1000 * This function returns 1 if the caller should continue to search,
1001 * or 0 if it should not. If there is an error reading one of the
1002 * index blocks, it will a negative error code.
1003 *
1004 * If start_hash is non-null, it will be filled in with the starting
1005 * hash of the next page.
1006 */
ext4_htree_next_block(struct inode * dir,__u32 hash,struct dx_frame * frame,struct dx_frame * frames,__u32 * start_hash)1007 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
1008 struct dx_frame *frame,
1009 struct dx_frame *frames,
1010 __u32 *start_hash)
1011 {
1012 struct dx_frame *p;
1013 struct buffer_head *bh;
1014 int num_frames = 0;
1015 __u32 bhash;
1016
1017 p = frame;
1018 /*
1019 * Find the next leaf page by incrementing the frame pointer.
1020 * If we run out of entries in the interior node, loop around and
1021 * increment pointer in the parent node. When we break out of
1022 * this loop, num_frames indicates the number of interior
1023 * nodes need to be read.
1024 */
1025 while (1) {
1026 if (++(p->at) < p->entries + dx_get_count(p->entries))
1027 break;
1028 if (p == frames)
1029 return 0;
1030 num_frames++;
1031 p--;
1032 }
1033
1034 /*
1035 * If the hash is 1, then continue only if the next page has a
1036 * continuation hash of any value. This is used for readdir
1037 * handling. Otherwise, check to see if the hash matches the
1038 * desired continuation hash. If it doesn't, return since
1039 * there's no point to read in the successive index pages.
1040 */
1041 bhash = dx_get_hash(p->at);
1042 if (start_hash)
1043 *start_hash = bhash;
1044 if ((hash & 1) == 0) {
1045 if ((bhash & ~1) != hash)
1046 return 0;
1047 }
1048 /*
1049 * If the hash is HASH_NB_ALWAYS, we always go to the next
1050 * block so no check is necessary
1051 */
1052 while (num_frames--) {
1053 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1054 if (IS_ERR(bh))
1055 return PTR_ERR(bh);
1056 p++;
1057 brelse(p->bh);
1058 p->bh = bh;
1059 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1060 }
1061 return 1;
1062 }
1063
1064
1065 /*
1066 * This function fills a red-black tree with information from a
1067 * directory block. It returns the number directory entries loaded
1068 * into the tree. If there is an error it is returned in err.
1069 */
htree_dirblock_to_tree(struct file * dir_file,struct inode * dir,ext4_lblk_t block,struct dx_hash_info * hinfo,__u32 start_hash,__u32 start_minor_hash)1070 static int htree_dirblock_to_tree(struct file *dir_file,
1071 struct inode *dir, ext4_lblk_t block,
1072 struct dx_hash_info *hinfo,
1073 __u32 start_hash, __u32 start_minor_hash)
1074 {
1075 struct buffer_head *bh;
1076 struct ext4_dir_entry_2 *de, *top;
1077 int err = 0, count = 0;
1078 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1079 int csum = ext4_has_metadata_csum(dir->i_sb);
1080
1081 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1082 (unsigned long)block));
1083 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1084 if (IS_ERR(bh))
1085 return PTR_ERR(bh);
1086
1087 de = (struct ext4_dir_entry_2 *) bh->b_data;
1088 /* csum entries are not larger in the casefolded encrypted case */
1089 top = (struct ext4_dir_entry_2 *) ((char *) de +
1090 dir->i_sb->s_blocksize -
1091 ext4_dir_rec_len(0,
1092 csum ? NULL : dir));
1093 /* Check if the directory is encrypted */
1094 if (IS_ENCRYPTED(dir)) {
1095 err = fscrypt_prepare_readdir(dir);
1096 if (err < 0) {
1097 brelse(bh);
1098 return err;
1099 }
1100 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1101 &fname_crypto_str);
1102 if (err < 0) {
1103 brelse(bh);
1104 return err;
1105 }
1106 }
1107
1108 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1109 if (ext4_check_dir_entry(dir, NULL, de, bh,
1110 bh->b_data, bh->b_size,
1111 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1112 + ((char *)de - bh->b_data))) {
1113 /* silently ignore the rest of the block */
1114 break;
1115 }
1116 if (ext4_hash_in_dirent(dir)) {
1117 if (de->name_len && de->inode) {
1118 hinfo->hash = EXT4_DIRENT_HASH(de);
1119 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1120 } else {
1121 hinfo->hash = 0;
1122 hinfo->minor_hash = 0;
1123 }
1124 } else {
1125 err = ext4fs_dirhash(dir, de->name,
1126 de->name_len, hinfo);
1127 if (err < 0) {
1128 count = err;
1129 goto errout;
1130 }
1131 }
1132 if ((hinfo->hash < start_hash) ||
1133 ((hinfo->hash == start_hash) &&
1134 (hinfo->minor_hash < start_minor_hash)))
1135 continue;
1136 if (de->inode == 0)
1137 continue;
1138 if (!IS_ENCRYPTED(dir)) {
1139 tmp_str.name = de->name;
1140 tmp_str.len = de->name_len;
1141 err = ext4_htree_store_dirent(dir_file,
1142 hinfo->hash, hinfo->minor_hash, de,
1143 &tmp_str);
1144 } else {
1145 int save_len = fname_crypto_str.len;
1146 struct fscrypt_str de_name = FSTR_INIT(de->name,
1147 de->name_len);
1148
1149 /* Directory is encrypted */
1150 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1151 hinfo->minor_hash, &de_name,
1152 &fname_crypto_str);
1153 if (err) {
1154 count = err;
1155 goto errout;
1156 }
1157 err = ext4_htree_store_dirent(dir_file,
1158 hinfo->hash, hinfo->minor_hash, de,
1159 &fname_crypto_str);
1160 fname_crypto_str.len = save_len;
1161 }
1162 if (err != 0) {
1163 count = err;
1164 goto errout;
1165 }
1166 count++;
1167 }
1168 errout:
1169 brelse(bh);
1170 fscrypt_fname_free_buffer(&fname_crypto_str);
1171 return count;
1172 }
1173
1174
1175 /*
1176 * This function fills a red-black tree with information from a
1177 * directory. We start scanning the directory in hash order, starting
1178 * at start_hash and start_minor_hash.
1179 *
1180 * This function returns the number of entries inserted into the tree,
1181 * or a negative error code.
1182 */
ext4_htree_fill_tree(struct file * dir_file,__u32 start_hash,__u32 start_minor_hash,__u32 * next_hash)1183 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1184 __u32 start_minor_hash, __u32 *next_hash)
1185 {
1186 struct dx_hash_info hinfo;
1187 struct ext4_dir_entry_2 *de;
1188 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1189 struct inode *dir;
1190 ext4_lblk_t block;
1191 int count = 0;
1192 int ret, err;
1193 __u32 hashval;
1194 struct fscrypt_str tmp_str;
1195
1196 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1197 start_hash, start_minor_hash));
1198 dir = file_inode(dir_file);
1199 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1200 if (ext4_hash_in_dirent(dir))
1201 hinfo.hash_version = DX_HASH_SIPHASH;
1202 else
1203 hinfo.hash_version =
1204 EXT4_SB(dir->i_sb)->s_def_hash_version;
1205 if (hinfo.hash_version <= DX_HASH_TEA)
1206 hinfo.hash_version +=
1207 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1208 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1209 if (ext4_has_inline_data(dir)) {
1210 int has_inline_data = 1;
1211 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1212 &hinfo, start_hash,
1213 start_minor_hash,
1214 &has_inline_data);
1215 if (has_inline_data) {
1216 *next_hash = ~0;
1217 return count;
1218 }
1219 }
1220 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1221 start_hash, start_minor_hash);
1222 *next_hash = ~0;
1223 return count;
1224 }
1225 hinfo.hash = start_hash;
1226 hinfo.minor_hash = 0;
1227 frame = dx_probe(NULL, dir, &hinfo, frames);
1228 if (IS_ERR(frame))
1229 return PTR_ERR(frame);
1230
1231 /* Add '.' and '..' from the htree header */
1232 if (!start_hash && !start_minor_hash) {
1233 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1234 tmp_str.name = de->name;
1235 tmp_str.len = de->name_len;
1236 err = ext4_htree_store_dirent(dir_file, 0, 0,
1237 de, &tmp_str);
1238 if (err != 0)
1239 goto errout;
1240 count++;
1241 }
1242 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1243 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1244 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1245 tmp_str.name = de->name;
1246 tmp_str.len = de->name_len;
1247 err = ext4_htree_store_dirent(dir_file, 2, 0,
1248 de, &tmp_str);
1249 if (err != 0)
1250 goto errout;
1251 count++;
1252 }
1253
1254 while (1) {
1255 if (fatal_signal_pending(current)) {
1256 err = -ERESTARTSYS;
1257 goto errout;
1258 }
1259 cond_resched();
1260 block = dx_get_block(frame->at);
1261 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1262 start_hash, start_minor_hash);
1263 if (ret < 0) {
1264 err = ret;
1265 goto errout;
1266 }
1267 count += ret;
1268 hashval = ~0;
1269 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1270 frame, frames, &hashval);
1271 *next_hash = hashval;
1272 if (ret < 0) {
1273 err = ret;
1274 goto errout;
1275 }
1276 /*
1277 * Stop if: (a) there are no more entries, or
1278 * (b) we have inserted at least one entry and the
1279 * next hash value is not a continuation
1280 */
1281 if ((ret == 0) ||
1282 (count && ((hashval & 1) == 0)))
1283 break;
1284 }
1285 dx_release(frames);
1286 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1287 "next hash: %x\n", count, *next_hash));
1288 return count;
1289 errout:
1290 dx_release(frames);
1291 return (err);
1292 }
1293
search_dirblock(struct buffer_head * bh,struct inode * dir,struct ext4_filename * fname,unsigned int offset,struct ext4_dir_entry_2 ** res_dir)1294 static inline int search_dirblock(struct buffer_head *bh,
1295 struct inode *dir,
1296 struct ext4_filename *fname,
1297 unsigned int offset,
1298 struct ext4_dir_entry_2 **res_dir)
1299 {
1300 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1301 fname, offset, res_dir);
1302 }
1303
1304 /*
1305 * Directory block splitting, compacting
1306 */
1307
1308 /*
1309 * Create map of hash values, offsets, and sizes, stored at end of block.
1310 * Returns number of entries mapped.
1311 */
dx_make_map(struct inode * dir,struct buffer_head * bh,struct dx_hash_info * hinfo,struct dx_map_entry * map_tail)1312 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1313 struct dx_hash_info *hinfo,
1314 struct dx_map_entry *map_tail)
1315 {
1316 int count = 0;
1317 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1318 unsigned int buflen = bh->b_size;
1319 char *base = bh->b_data;
1320 struct dx_hash_info h = *hinfo;
1321 int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
1322
1323 if (ext4_has_metadata_csum(dir->i_sb))
1324 buflen -= sizeof(struct ext4_dir_entry_tail);
1325
1326 while ((char *) de < base + buflen) {
1327 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1328 ((char *)de) - base))
1329 return -EFSCORRUPTED;
1330 if (de->name_len && de->inode) {
1331 if (ext4_hash_in_dirent(dir))
1332 h.hash = EXT4_DIRENT_HASH(de);
1333 else {
1334 int err = ext4fs_dirhash(dir, de->name,
1335 de->name_len, &h);
1336 if (err < 0)
1337 return err;
1338 }
1339 map_tail--;
1340 map_tail->hash = h.hash;
1341 map_tail->offs = ((char *) de - base)>>2;
1342 map_tail->size = ext4_rec_len_from_disk(de->rec_len,
1343 blocksize);
1344 count++;
1345 cond_resched();
1346 }
1347 de = ext4_next_entry(de, blocksize);
1348 }
1349 return count;
1350 }
1351
1352 /* Sort map by hash value */
dx_sort_map(struct dx_map_entry * map,unsigned count)1353 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1354 {
1355 struct dx_map_entry *p, *q, *top = map + count - 1;
1356 int more;
1357 /* Combsort until bubble sort doesn't suck */
1358 while (count > 2) {
1359 count = count*10/13;
1360 if (count - 9 < 2) /* 9, 10 -> 11 */
1361 count = 11;
1362 for (p = top, q = p - count; q >= map; p--, q--)
1363 if (p->hash < q->hash)
1364 swap(*p, *q);
1365 }
1366 /* Garden variety bubble sort */
1367 do {
1368 more = 0;
1369 q = top;
1370 while (q-- > map) {
1371 if (q[1].hash >= q[0].hash)
1372 continue;
1373 swap(*(q+1), *q);
1374 more = 1;
1375 }
1376 } while(more);
1377 }
1378
dx_insert_block(struct dx_frame * frame,u32 hash,ext4_lblk_t block)1379 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1380 {
1381 struct dx_entry *entries = frame->entries;
1382 struct dx_entry *old = frame->at, *new = old + 1;
1383 int count = dx_get_count(entries);
1384
1385 ASSERT(count < dx_get_limit(entries));
1386 ASSERT(old < entries + count);
1387 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1388 dx_set_hash(new, hash);
1389 dx_set_block(new, block);
1390 dx_set_count(entries, count + 1);
1391 }
1392
1393 #if IS_ENABLED(CONFIG_UNICODE)
1394 /*
1395 * Test whether a case-insensitive directory entry matches the filename
1396 * being searched for. If quick is set, assume the name being looked up
1397 * is already in the casefolded form.
1398 *
1399 * Returns: 0 if the directory entry matches, more than 0 if it
1400 * doesn't match or less than zero on error.
1401 */
ext4_ci_compare(const struct inode * parent,const struct qstr * name,u8 * de_name,size_t de_name_len,bool quick)1402 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1403 u8 *de_name, size_t de_name_len, bool quick)
1404 {
1405 const struct super_block *sb = parent->i_sb;
1406 const struct unicode_map *um = sb->s_encoding;
1407 struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1408 struct qstr entry = QSTR_INIT(de_name, de_name_len);
1409 int ret;
1410
1411 if (IS_ENCRYPTED(parent)) {
1412 const struct fscrypt_str encrypted_name =
1413 FSTR_INIT(de_name, de_name_len);
1414
1415 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1416 if (!decrypted_name.name)
1417 return -ENOMEM;
1418 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1419 &decrypted_name);
1420 if (ret < 0)
1421 goto out;
1422 entry.name = decrypted_name.name;
1423 entry.len = decrypted_name.len;
1424 }
1425
1426 if (quick)
1427 ret = utf8_strncasecmp_folded(um, name, &entry);
1428 else
1429 ret = utf8_strncasecmp(um, name, &entry);
1430 if (ret < 0) {
1431 /* Handle invalid character sequence as either an error
1432 * or as an opaque byte sequence.
1433 */
1434 if (sb_has_strict_encoding(sb))
1435 ret = -EINVAL;
1436 else if (name->len != entry.len)
1437 ret = 1;
1438 else
1439 ret = !!memcmp(name->name, entry.name, entry.len);
1440 }
1441 out:
1442 kfree(decrypted_name.name);
1443 return ret;
1444 }
1445
ext4_fname_setup_ci_filename(struct inode * dir,const struct qstr * iname,struct ext4_filename * name)1446 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1447 struct ext4_filename *name)
1448 {
1449 struct fscrypt_str *cf_name = &name->cf_name;
1450 struct dx_hash_info *hinfo = &name->hinfo;
1451 int len;
1452
1453 if (!IS_CASEFOLDED(dir) ||
1454 (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1455 cf_name->name = NULL;
1456 return 0;
1457 }
1458
1459 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1460 if (!cf_name->name)
1461 return -ENOMEM;
1462
1463 len = utf8_casefold(dir->i_sb->s_encoding,
1464 iname, cf_name->name,
1465 EXT4_NAME_LEN);
1466 if (len <= 0) {
1467 kfree(cf_name->name);
1468 cf_name->name = NULL;
1469 }
1470 cf_name->len = (unsigned) len;
1471 if (!IS_ENCRYPTED(dir))
1472 return 0;
1473
1474 hinfo->hash_version = DX_HASH_SIPHASH;
1475 hinfo->seed = NULL;
1476 if (cf_name->name)
1477 return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1478 else
1479 return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1480 }
1481 #endif
1482
1483 /*
1484 * Test whether a directory entry matches the filename being searched for.
1485 *
1486 * Return: %true if the directory entry matches, otherwise %false.
1487 */
ext4_match(struct inode * parent,const struct ext4_filename * fname,struct ext4_dir_entry_2 * de)1488 static bool ext4_match(struct inode *parent,
1489 const struct ext4_filename *fname,
1490 struct ext4_dir_entry_2 *de)
1491 {
1492 struct fscrypt_name f;
1493
1494 if (!de->inode)
1495 return false;
1496
1497 f.usr_fname = fname->usr_fname;
1498 f.disk_name = fname->disk_name;
1499 #ifdef CONFIG_FS_ENCRYPTION
1500 f.crypto_buf = fname->crypto_buf;
1501 #endif
1502
1503 #if IS_ENABLED(CONFIG_UNICODE)
1504 if (IS_CASEFOLDED(parent) &&
1505 (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1506 if (fname->cf_name.name) {
1507 struct qstr cf = {.name = fname->cf_name.name,
1508 .len = fname->cf_name.len};
1509 if (IS_ENCRYPTED(parent)) {
1510 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1511 fname->hinfo.minor_hash !=
1512 EXT4_DIRENT_MINOR_HASH(de)) {
1513
1514 return false;
1515 }
1516 }
1517 return !ext4_ci_compare(parent, &cf, de->name,
1518 de->name_len, true);
1519 }
1520 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1521 de->name_len, false);
1522 }
1523 #endif
1524
1525 return fscrypt_match_name(&f, de->name, de->name_len);
1526 }
1527
1528 /*
1529 * Returns 0 if not found, -EFSCORRUPTED on failure, and 1 on success
1530 */
ext4_search_dir(struct buffer_head * bh,char * search_buf,int buf_size,struct inode * dir,struct ext4_filename * fname,unsigned int offset,struct ext4_dir_entry_2 ** res_dir)1531 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1532 struct inode *dir, struct ext4_filename *fname,
1533 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1534 {
1535 struct ext4_dir_entry_2 * de;
1536 char * dlimit;
1537 int de_len;
1538
1539 de = (struct ext4_dir_entry_2 *)search_buf;
1540 dlimit = search_buf + buf_size;
1541 while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1542 /* this code is executed quadratically often */
1543 /* do minimal checking `by hand' */
1544 if (de->name + de->name_len <= dlimit &&
1545 ext4_match(dir, fname, de)) {
1546 /* found a match - just to be sure, do
1547 * a full check */
1548 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1549 buf_size, offset))
1550 return -EFSCORRUPTED;
1551 *res_dir = de;
1552 return 1;
1553 }
1554 /* prevent looping on a bad block */
1555 de_len = ext4_rec_len_from_disk(de->rec_len,
1556 dir->i_sb->s_blocksize);
1557 if (de_len <= 0)
1558 return -EFSCORRUPTED;
1559 offset += de_len;
1560 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1561 }
1562 return 0;
1563 }
1564
is_dx_internal_node(struct inode * dir,ext4_lblk_t block,struct ext4_dir_entry * de)1565 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1566 struct ext4_dir_entry *de)
1567 {
1568 struct super_block *sb = dir->i_sb;
1569
1570 if (!is_dx(dir))
1571 return 0;
1572 if (block == 0)
1573 return 1;
1574 if (de->inode == 0 &&
1575 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1576 sb->s_blocksize)
1577 return 1;
1578 return 0;
1579 }
1580
1581 /*
1582 * __ext4_find_entry()
1583 *
1584 * finds an entry in the specified directory with the wanted name. It
1585 * returns the cache buffer in which the entry was found, and the entry
1586 * itself (as a parameter - res_dir). It does NOT read the inode of the
1587 * entry - you'll have to do that yourself if you want to.
1588 *
1589 * The returned buffer_head has ->b_count elevated. The caller is expected
1590 * to brelse() it when appropriate.
1591 */
__ext4_find_entry(struct inode * dir,struct ext4_filename * fname,struct ext4_dir_entry_2 ** res_dir,int * inlined)1592 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1593 struct ext4_filename *fname,
1594 struct ext4_dir_entry_2 **res_dir,
1595 int *inlined)
1596 {
1597 struct super_block *sb;
1598 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1599 struct buffer_head *bh, *ret = NULL;
1600 ext4_lblk_t start, block;
1601 const u8 *name = fname->usr_fname->name;
1602 size_t ra_max = 0; /* Number of bh's in the readahead
1603 buffer, bh_use[] */
1604 size_t ra_ptr = 0; /* Current index into readahead
1605 buffer */
1606 ext4_lblk_t nblocks;
1607 int i, namelen, retval;
1608
1609 *res_dir = NULL;
1610 sb = dir->i_sb;
1611 namelen = fname->usr_fname->len;
1612 if (namelen > EXT4_NAME_LEN)
1613 return NULL;
1614
1615 if (ext4_has_inline_data(dir)) {
1616 int has_inline_data = 1;
1617 ret = ext4_find_inline_entry(dir, fname, res_dir,
1618 &has_inline_data);
1619 if (inlined)
1620 *inlined = has_inline_data;
1621 if (has_inline_data)
1622 goto cleanup_and_exit;
1623 }
1624
1625 if ((namelen <= 2) && (name[0] == '.') &&
1626 (name[1] == '.' || name[1] == '\0')) {
1627 /*
1628 * "." or ".." will only be in the first block
1629 * NFS may look up ".."; "." should be handled by the VFS
1630 */
1631 block = start = 0;
1632 nblocks = 1;
1633 goto restart;
1634 }
1635 if (is_dx(dir)) {
1636 ret = ext4_dx_find_entry(dir, fname, res_dir);
1637 /*
1638 * On success, or if the error was file not found,
1639 * return. Otherwise, fall back to doing a search the
1640 * old fashioned way.
1641 */
1642 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1643 goto cleanup_and_exit;
1644 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1645 "falling back\n"));
1646 ret = NULL;
1647 }
1648 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1649 if (!nblocks) {
1650 ret = NULL;
1651 goto cleanup_and_exit;
1652 }
1653 start = EXT4_I(dir)->i_dir_start_lookup;
1654 if (start >= nblocks)
1655 start = 0;
1656 block = start;
1657 restart:
1658 do {
1659 /*
1660 * We deal with the read-ahead logic here.
1661 */
1662 cond_resched();
1663 if (ra_ptr >= ra_max) {
1664 /* Refill the readahead buffer */
1665 ra_ptr = 0;
1666 if (block < start)
1667 ra_max = start - block;
1668 else
1669 ra_max = nblocks - block;
1670 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1671 retval = ext4_bread_batch(dir, block, ra_max,
1672 false /* wait */, bh_use);
1673 if (retval) {
1674 ret = ERR_PTR(retval);
1675 ra_max = 0;
1676 goto cleanup_and_exit;
1677 }
1678 }
1679 if ((bh = bh_use[ra_ptr++]) == NULL)
1680 goto next;
1681 wait_on_buffer(bh);
1682 if (!buffer_uptodate(bh)) {
1683 EXT4_ERROR_INODE_ERR(dir, EIO,
1684 "reading directory lblock %lu",
1685 (unsigned long) block);
1686 brelse(bh);
1687 ret = ERR_PTR(-EIO);
1688 goto cleanup_and_exit;
1689 }
1690 if (!buffer_verified(bh) &&
1691 !is_dx_internal_node(dir, block,
1692 (struct ext4_dir_entry *)bh->b_data) &&
1693 !ext4_dirblock_csum_verify(dir, bh)) {
1694 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1695 "checksumming directory "
1696 "block %lu", (unsigned long)block);
1697 brelse(bh);
1698 ret = ERR_PTR(-EFSBADCRC);
1699 goto cleanup_and_exit;
1700 }
1701 set_buffer_verified(bh);
1702 i = search_dirblock(bh, dir, fname,
1703 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1704 if (i == 1) {
1705 EXT4_I(dir)->i_dir_start_lookup = block;
1706 ret = bh;
1707 goto cleanup_and_exit;
1708 } else {
1709 brelse(bh);
1710 if (i < 0) {
1711 ret = ERR_PTR(i);
1712 goto cleanup_and_exit;
1713 }
1714 }
1715 next:
1716 if (++block >= nblocks)
1717 block = 0;
1718 } while (block != start);
1719
1720 /*
1721 * If the directory has grown while we were searching, then
1722 * search the last part of the directory before giving up.
1723 */
1724 block = nblocks;
1725 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1726 if (block < nblocks) {
1727 start = 0;
1728 goto restart;
1729 }
1730
1731 cleanup_and_exit:
1732 /* Clean up the read-ahead blocks */
1733 for (; ra_ptr < ra_max; ra_ptr++)
1734 brelse(bh_use[ra_ptr]);
1735 return ret;
1736 }
1737
ext4_find_entry(struct inode * dir,const struct qstr * d_name,struct ext4_dir_entry_2 ** res_dir,int * inlined)1738 static struct buffer_head *ext4_find_entry(struct inode *dir,
1739 const struct qstr *d_name,
1740 struct ext4_dir_entry_2 **res_dir,
1741 int *inlined)
1742 {
1743 int err;
1744 struct ext4_filename fname;
1745 struct buffer_head *bh;
1746
1747 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1748 if (err == -ENOENT)
1749 return NULL;
1750 if (err)
1751 return ERR_PTR(err);
1752
1753 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1754
1755 ext4_fname_free_filename(&fname);
1756 return bh;
1757 }
1758
ext4_lookup_entry(struct inode * dir,struct dentry * dentry,struct ext4_dir_entry_2 ** res_dir)1759 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1760 struct dentry *dentry,
1761 struct ext4_dir_entry_2 **res_dir)
1762 {
1763 int err;
1764 struct ext4_filename fname;
1765 struct buffer_head *bh;
1766
1767 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1768 generic_set_encrypted_ci_d_ops(dentry);
1769 if (err == -ENOENT)
1770 return NULL;
1771 if (err)
1772 return ERR_PTR(err);
1773
1774 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1775
1776 ext4_fname_free_filename(&fname);
1777 return bh;
1778 }
1779
ext4_dx_find_entry(struct inode * dir,struct ext4_filename * fname,struct ext4_dir_entry_2 ** res_dir)1780 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1781 struct ext4_filename *fname,
1782 struct ext4_dir_entry_2 **res_dir)
1783 {
1784 struct super_block * sb = dir->i_sb;
1785 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1786 struct buffer_head *bh;
1787 ext4_lblk_t block;
1788 int retval;
1789
1790 #ifdef CONFIG_FS_ENCRYPTION
1791 *res_dir = NULL;
1792 #endif
1793 frame = dx_probe(fname, dir, NULL, frames);
1794 if (IS_ERR(frame))
1795 return (struct buffer_head *) frame;
1796 do {
1797 block = dx_get_block(frame->at);
1798 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1799 if (IS_ERR(bh))
1800 goto errout;
1801
1802 retval = search_dirblock(bh, dir, fname,
1803 block << EXT4_BLOCK_SIZE_BITS(sb),
1804 res_dir);
1805 if (retval == 1)
1806 goto success;
1807 brelse(bh);
1808 if (retval < 0) {
1809 bh = ERR_PTR(ERR_BAD_DX_DIR);
1810 goto errout;
1811 }
1812
1813 /* Check to see if we should continue to search */
1814 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1815 frames, NULL);
1816 if (retval < 0) {
1817 ext4_warning_inode(dir,
1818 "error %d reading directory index block",
1819 retval);
1820 bh = ERR_PTR(retval);
1821 goto errout;
1822 }
1823 } while (retval == 1);
1824
1825 bh = NULL;
1826 errout:
1827 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1828 success:
1829 dx_release(frames);
1830 return bh;
1831 }
1832
ext4_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)1833 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1834 {
1835 struct inode *inode;
1836 struct ext4_dir_entry_2 *de;
1837 struct buffer_head *bh;
1838
1839 if (dentry->d_name.len > EXT4_NAME_LEN)
1840 return ERR_PTR(-ENAMETOOLONG);
1841
1842 bh = ext4_lookup_entry(dir, dentry, &de);
1843 if (IS_ERR(bh))
1844 return ERR_CAST(bh);
1845 inode = NULL;
1846 if (bh) {
1847 __u32 ino = le32_to_cpu(de->inode);
1848 brelse(bh);
1849 if (!ext4_valid_inum(dir->i_sb, ino)) {
1850 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1851 return ERR_PTR(-EFSCORRUPTED);
1852 }
1853 if (unlikely(ino == dir->i_ino)) {
1854 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1855 dentry);
1856 return ERR_PTR(-EFSCORRUPTED);
1857 }
1858 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1859 if (inode == ERR_PTR(-ESTALE)) {
1860 EXT4_ERROR_INODE(dir,
1861 "deleted inode referenced: %u",
1862 ino);
1863 return ERR_PTR(-EFSCORRUPTED);
1864 }
1865 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1866 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1867 !fscrypt_has_permitted_context(dir, inode)) {
1868 ext4_warning(inode->i_sb,
1869 "Inconsistent encryption contexts: %lu/%lu",
1870 dir->i_ino, inode->i_ino);
1871 iput(inode);
1872 return ERR_PTR(-EPERM);
1873 }
1874 }
1875
1876 #if IS_ENABLED(CONFIG_UNICODE)
1877 if (!inode && IS_CASEFOLDED(dir)) {
1878 /* Eventually we want to call d_add_ci(dentry, NULL)
1879 * for negative dentries in the encoding case as
1880 * well. For now, prevent the negative dentry
1881 * from being cached.
1882 */
1883 return NULL;
1884 }
1885 #endif
1886 return d_splice_alias(inode, dentry);
1887 }
1888
1889
ext4_get_parent(struct dentry * child)1890 struct dentry *ext4_get_parent(struct dentry *child)
1891 {
1892 __u32 ino;
1893 struct ext4_dir_entry_2 * de;
1894 struct buffer_head *bh;
1895
1896 bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1897 if (IS_ERR(bh))
1898 return ERR_CAST(bh);
1899 if (!bh)
1900 return ERR_PTR(-ENOENT);
1901 ino = le32_to_cpu(de->inode);
1902 brelse(bh);
1903
1904 if (!ext4_valid_inum(child->d_sb, ino)) {
1905 EXT4_ERROR_INODE(d_inode(child),
1906 "bad parent inode number: %u", ino);
1907 return ERR_PTR(-EFSCORRUPTED);
1908 }
1909
1910 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1911 }
1912
1913 /*
1914 * Move count entries from end of map between two memory locations.
1915 * Returns pointer to last entry moved.
1916 */
1917 static struct ext4_dir_entry_2 *
dx_move_dirents(struct inode * dir,char * from,char * to,struct dx_map_entry * map,int count,unsigned blocksize)1918 dx_move_dirents(struct inode *dir, char *from, char *to,
1919 struct dx_map_entry *map, int count,
1920 unsigned blocksize)
1921 {
1922 unsigned rec_len = 0;
1923
1924 while (count--) {
1925 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1926 (from + (map->offs<<2));
1927 rec_len = ext4_dir_rec_len(de->name_len, dir);
1928
1929 memcpy (to, de, rec_len);
1930 ((struct ext4_dir_entry_2 *) to)->rec_len =
1931 ext4_rec_len_to_disk(rec_len, blocksize);
1932
1933 /* wipe dir_entry excluding the rec_len field */
1934 de->inode = 0;
1935 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1936 blocksize) -
1937 offsetof(struct ext4_dir_entry_2,
1938 name_len));
1939
1940 map++;
1941 to += rec_len;
1942 }
1943 return (struct ext4_dir_entry_2 *) (to - rec_len);
1944 }
1945
1946 /*
1947 * Compact each dir entry in the range to the minimal rec_len.
1948 * Returns pointer to last entry in range.
1949 */
dx_pack_dirents(struct inode * dir,char * base,unsigned int blocksize)1950 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1951 unsigned int blocksize)
1952 {
1953 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1954 unsigned rec_len = 0;
1955
1956 prev = to = de;
1957 while ((char*)de < base + blocksize) {
1958 next = ext4_next_entry(de, blocksize);
1959 if (de->inode && de->name_len) {
1960 rec_len = ext4_dir_rec_len(de->name_len, dir);
1961 if (de > to)
1962 memmove(to, de, rec_len);
1963 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1964 prev = to;
1965 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1966 }
1967 de = next;
1968 }
1969 return prev;
1970 }
1971
1972 /*
1973 * Split a full leaf block to make room for a new dir entry.
1974 * Allocate a new block, and move entries so that they are approx. equally full.
1975 * Returns pointer to de in block into which the new entry will be inserted.
1976 */
do_split(handle_t * handle,struct inode * dir,struct buffer_head ** bh,struct dx_frame * frame,struct dx_hash_info * hinfo)1977 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1978 struct buffer_head **bh,struct dx_frame *frame,
1979 struct dx_hash_info *hinfo)
1980 {
1981 unsigned blocksize = dir->i_sb->s_blocksize;
1982 unsigned continued;
1983 int count;
1984 struct buffer_head *bh2;
1985 ext4_lblk_t newblock;
1986 u32 hash2;
1987 struct dx_map_entry *map;
1988 char *data1 = (*bh)->b_data, *data2;
1989 unsigned split, move, size;
1990 struct ext4_dir_entry_2 *de = NULL, *de2;
1991 int csum_size = 0;
1992 int err = 0, i;
1993
1994 if (ext4_has_metadata_csum(dir->i_sb))
1995 csum_size = sizeof(struct ext4_dir_entry_tail);
1996
1997 bh2 = ext4_append(handle, dir, &newblock);
1998 if (IS_ERR(bh2)) {
1999 brelse(*bh);
2000 *bh = NULL;
2001 return (struct ext4_dir_entry_2 *) bh2;
2002 }
2003
2004 BUFFER_TRACE(*bh, "get_write_access");
2005 err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
2006 EXT4_JTR_NONE);
2007 if (err)
2008 goto journal_error;
2009
2010 BUFFER_TRACE(frame->bh, "get_write_access");
2011 err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
2012 EXT4_JTR_NONE);
2013 if (err)
2014 goto journal_error;
2015
2016 data2 = bh2->b_data;
2017
2018 /* create map in the end of data2 block */
2019 map = (struct dx_map_entry *) (data2 + blocksize);
2020 count = dx_make_map(dir, *bh, hinfo, map);
2021 if (count < 0) {
2022 err = count;
2023 goto journal_error;
2024 }
2025 map -= count;
2026 dx_sort_map(map, count);
2027 /* Ensure that neither split block is over half full */
2028 size = 0;
2029 move = 0;
2030 for (i = count-1; i >= 0; i--) {
2031 /* is more than half of this entry in 2nd half of the block? */
2032 if (size + map[i].size/2 > blocksize/2)
2033 break;
2034 size += map[i].size;
2035 move++;
2036 }
2037 /*
2038 * map index at which we will split
2039 *
2040 * If the sum of active entries didn't exceed half the block size, just
2041 * split it in half by count; each resulting block will have at least
2042 * half the space free.
2043 */
2044 if (i > 0)
2045 split = count - move;
2046 else
2047 split = count/2;
2048
2049 hash2 = map[split].hash;
2050 continued = split > 0 ? hash2 == map[split - 1].hash : 0;
2051 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2052 (unsigned long)dx_get_block(frame->at),
2053 hash2, split, count-split));
2054
2055 /* Fancy dance to stay within two buffers */
2056 de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2057 blocksize);
2058 de = dx_pack_dirents(dir, data1, blocksize);
2059 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2060 (char *) de,
2061 blocksize);
2062 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2063 (char *) de2,
2064 blocksize);
2065 if (csum_size) {
2066 ext4_initialize_dirent_tail(*bh, blocksize);
2067 ext4_initialize_dirent_tail(bh2, blocksize);
2068 }
2069
2070 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2071 blocksize, 1));
2072 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2073 blocksize, 1));
2074
2075 /* Which block gets the new entry? */
2076 if (hinfo->hash >= hash2) {
2077 swap(*bh, bh2);
2078 de = de2;
2079 }
2080 dx_insert_block(frame, hash2 + continued, newblock);
2081 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2082 if (err)
2083 goto journal_error;
2084 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2085 if (err)
2086 goto journal_error;
2087 brelse(bh2);
2088 dxtrace(dx_show_index("frame", frame->entries));
2089 return de;
2090
2091 journal_error:
2092 brelse(*bh);
2093 brelse(bh2);
2094 *bh = NULL;
2095 ext4_std_error(dir->i_sb, err);
2096 return ERR_PTR(err);
2097 }
2098
ext4_find_dest_de(struct inode * dir,struct inode * inode,struct buffer_head * bh,void * buf,int buf_size,struct ext4_filename * fname,struct ext4_dir_entry_2 ** dest_de)2099 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2100 struct buffer_head *bh,
2101 void *buf, int buf_size,
2102 struct ext4_filename *fname,
2103 struct ext4_dir_entry_2 **dest_de)
2104 {
2105 struct ext4_dir_entry_2 *de;
2106 unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2107 int nlen, rlen;
2108 unsigned int offset = 0;
2109 char *top;
2110
2111 de = buf;
2112 top = buf + buf_size - reclen;
2113 while ((char *) de <= top) {
2114 if (ext4_check_dir_entry(dir, NULL, de, bh,
2115 buf, buf_size, offset))
2116 return -EFSCORRUPTED;
2117 if (ext4_match(dir, fname, de))
2118 return -EEXIST;
2119 nlen = ext4_dir_rec_len(de->name_len, dir);
2120 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2121 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2122 break;
2123 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2124 offset += rlen;
2125 }
2126 if ((char *) de > top)
2127 return -ENOSPC;
2128
2129 *dest_de = de;
2130 return 0;
2131 }
2132
ext4_insert_dentry(struct inode * dir,struct inode * inode,struct ext4_dir_entry_2 * de,int buf_size,struct ext4_filename * fname)2133 void ext4_insert_dentry(struct inode *dir,
2134 struct inode *inode,
2135 struct ext4_dir_entry_2 *de,
2136 int buf_size,
2137 struct ext4_filename *fname)
2138 {
2139
2140 int nlen, rlen;
2141
2142 nlen = ext4_dir_rec_len(de->name_len, dir);
2143 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2144 if (de->inode) {
2145 struct ext4_dir_entry_2 *de1 =
2146 (struct ext4_dir_entry_2 *)((char *)de + nlen);
2147 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2148 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2149 de = de1;
2150 }
2151 de->file_type = EXT4_FT_UNKNOWN;
2152 de->inode = cpu_to_le32(inode->i_ino);
2153 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2154 de->name_len = fname_len(fname);
2155 memcpy(de->name, fname_name(fname), fname_len(fname));
2156 if (ext4_hash_in_dirent(dir)) {
2157 struct dx_hash_info *hinfo = &fname->hinfo;
2158
2159 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2160 EXT4_DIRENT_HASHES(de)->minor_hash =
2161 cpu_to_le32(hinfo->minor_hash);
2162 }
2163 }
2164
2165 /*
2166 * Add a new entry into a directory (leaf) block. If de is non-NULL,
2167 * it points to a directory entry which is guaranteed to be large
2168 * enough for new directory entry. If de is NULL, then
2169 * add_dirent_to_buf will attempt search the directory block for
2170 * space. It will return -ENOSPC if no space is available, and -EIO
2171 * and -EEXIST if directory entry already exists.
2172 */
add_dirent_to_buf(handle_t * handle,struct ext4_filename * fname,struct inode * dir,struct inode * inode,struct ext4_dir_entry_2 * de,struct buffer_head * bh)2173 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2174 struct inode *dir,
2175 struct inode *inode, struct ext4_dir_entry_2 *de,
2176 struct buffer_head *bh)
2177 {
2178 unsigned int blocksize = dir->i_sb->s_blocksize;
2179 int csum_size = 0;
2180 int err, err2;
2181
2182 if (ext4_has_metadata_csum(inode->i_sb))
2183 csum_size = sizeof(struct ext4_dir_entry_tail);
2184
2185 if (!de) {
2186 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2187 blocksize - csum_size, fname, &de);
2188 if (err)
2189 return err;
2190 }
2191 BUFFER_TRACE(bh, "get_write_access");
2192 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2193 EXT4_JTR_NONE);
2194 if (err) {
2195 ext4_std_error(dir->i_sb, err);
2196 return err;
2197 }
2198
2199 /* By now the buffer is marked for journaling */
2200 ext4_insert_dentry(dir, inode, de, blocksize, fname);
2201
2202 /*
2203 * XXX shouldn't update any times until successful
2204 * completion of syscall, but too many callers depend
2205 * on this.
2206 *
2207 * XXX similarly, too many callers depend on
2208 * ext4_new_inode() setting the times, but error
2209 * recovery deletes the inode, so the worst that can
2210 * happen is that the times are slightly out of date
2211 * and/or different from the directory change time.
2212 */
2213 dir->i_mtime = inode_set_ctime_current(dir);
2214 ext4_update_dx_flag(dir);
2215 inode_inc_iversion(dir);
2216 err2 = ext4_mark_inode_dirty(handle, dir);
2217 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2218 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2219 if (err)
2220 ext4_std_error(dir->i_sb, err);
2221 return err ? err : err2;
2222 }
2223
ext4_check_dx_root(struct inode * dir,struct dx_root * root)2224 static bool ext4_check_dx_root(struct inode *dir, struct dx_root *root)
2225 {
2226 struct fake_dirent *fde;
2227 const char *error_msg;
2228 unsigned int rlen;
2229 unsigned int blocksize = dir->i_sb->s_blocksize;
2230 char *blockend = (char *)root + dir->i_sb->s_blocksize;
2231
2232 fde = &root->dot;
2233 if (unlikely(fde->name_len != 1)) {
2234 error_msg = "invalid name_len for '.'";
2235 goto corrupted;
2236 }
2237 if (unlikely(strncmp(root->dot_name, ".", fde->name_len))) {
2238 error_msg = "invalid name for '.'";
2239 goto corrupted;
2240 }
2241 rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
2242 if (unlikely((char *)fde + rlen >= blockend)) {
2243 error_msg = "invalid rec_len for '.'";
2244 goto corrupted;
2245 }
2246
2247 fde = &root->dotdot;
2248 if (unlikely(fde->name_len != 2)) {
2249 error_msg = "invalid name_len for '..'";
2250 goto corrupted;
2251 }
2252 if (unlikely(strncmp(root->dotdot_name, "..", fde->name_len))) {
2253 error_msg = "invalid name for '..'";
2254 goto corrupted;
2255 }
2256 rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
2257 if (unlikely((char *)fde + rlen >= blockend)) {
2258 error_msg = "invalid rec_len for '..'";
2259 goto corrupted;
2260 }
2261
2262 return true;
2263
2264 corrupted:
2265 EXT4_ERROR_INODE(dir, "Corrupt dir, %s, running e2fsck is recommended",
2266 error_msg);
2267 return false;
2268 }
2269
2270 /*
2271 * This converts a one block unindexed directory to a 3 block indexed
2272 * directory, and adds the dentry to the indexed directory.
2273 */
make_indexed_dir(handle_t * handle,struct ext4_filename * fname,struct inode * dir,struct inode * inode,struct buffer_head * bh)2274 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2275 struct inode *dir,
2276 struct inode *inode, struct buffer_head *bh)
2277 {
2278 struct buffer_head *bh2;
2279 struct dx_root *root;
2280 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2281 struct dx_entry *entries;
2282 struct ext4_dir_entry_2 *de, *de2;
2283 char *data2, *top;
2284 unsigned len;
2285 int retval;
2286 unsigned blocksize;
2287 ext4_lblk_t block;
2288 struct fake_dirent *fde;
2289 int csum_size = 0;
2290
2291 if (ext4_has_metadata_csum(inode->i_sb))
2292 csum_size = sizeof(struct ext4_dir_entry_tail);
2293
2294 blocksize = dir->i_sb->s_blocksize;
2295 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2296 BUFFER_TRACE(bh, "get_write_access");
2297 retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2298 EXT4_JTR_NONE);
2299 if (retval) {
2300 ext4_std_error(dir->i_sb, retval);
2301 brelse(bh);
2302 return retval;
2303 }
2304
2305 root = (struct dx_root *) bh->b_data;
2306 if (!ext4_check_dx_root(dir, root)) {
2307 brelse(bh);
2308 return -EFSCORRUPTED;
2309 }
2310
2311 /* The 0th block becomes the root, move the dirents out */
2312 fde = &root->dotdot;
2313 de = (struct ext4_dir_entry_2 *)((char *)fde +
2314 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2315 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2316
2317 /* Allocate new block for the 0th block's dirents */
2318 bh2 = ext4_append(handle, dir, &block);
2319 if (IS_ERR(bh2)) {
2320 brelse(bh);
2321 return PTR_ERR(bh2);
2322 }
2323 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2324 data2 = bh2->b_data;
2325
2326 memcpy(data2, de, len);
2327 memset(de, 0, len); /* wipe old data */
2328 de = (struct ext4_dir_entry_2 *) data2;
2329 top = data2 + len;
2330 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2331 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2332 (data2 + (blocksize - csum_size) -
2333 (char *) de))) {
2334 brelse(bh2);
2335 brelse(bh);
2336 return -EFSCORRUPTED;
2337 }
2338 de = de2;
2339 }
2340 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2341 (char *) de, blocksize);
2342
2343 if (csum_size)
2344 ext4_initialize_dirent_tail(bh2, blocksize);
2345
2346 /* Initialize the root; the dot dirents already exist */
2347 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2348 de->rec_len = ext4_rec_len_to_disk(
2349 blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2350 memset (&root->info, 0, sizeof(root->info));
2351 root->info.info_length = sizeof(root->info);
2352 if (ext4_hash_in_dirent(dir))
2353 root->info.hash_version = DX_HASH_SIPHASH;
2354 else
2355 root->info.hash_version =
2356 EXT4_SB(dir->i_sb)->s_def_hash_version;
2357
2358 entries = root->entries;
2359 dx_set_block(entries, 1);
2360 dx_set_count(entries, 1);
2361 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2362
2363 /* Initialize as for dx_probe */
2364 fname->hinfo.hash_version = root->info.hash_version;
2365 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2366 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2367 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2368
2369 /* casefolded encrypted hashes are computed on fname setup */
2370 if (!ext4_hash_in_dirent(dir)) {
2371 int err = ext4fs_dirhash(dir, fname_name(fname),
2372 fname_len(fname), &fname->hinfo);
2373 if (err < 0) {
2374 brelse(bh2);
2375 brelse(bh);
2376 return err;
2377 }
2378 }
2379 memset(frames, 0, sizeof(frames));
2380 frame = frames;
2381 frame->entries = entries;
2382 frame->at = entries;
2383 frame->bh = bh;
2384
2385 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2386 if (retval)
2387 goto out_frames;
2388 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2389 if (retval)
2390 goto out_frames;
2391
2392 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2393 if (IS_ERR(de)) {
2394 retval = PTR_ERR(de);
2395 goto out_frames;
2396 }
2397
2398 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2399 out_frames:
2400 /*
2401 * Even if the block split failed, we have to properly write
2402 * out all the changes we did so far. Otherwise we can end up
2403 * with corrupted filesystem.
2404 */
2405 if (retval)
2406 ext4_mark_inode_dirty(handle, dir);
2407 dx_release(frames);
2408 brelse(bh2);
2409 return retval;
2410 }
2411
2412 /*
2413 * ext4_add_entry()
2414 *
2415 * adds a file entry to the specified directory, using the same
2416 * semantics as ext4_find_entry(). It returns NULL if it failed.
2417 *
2418 * NOTE!! The inode part of 'de' is left at 0 - which means you
2419 * may not sleep between calling this and putting something into
2420 * the entry, as someone else might have used it while you slept.
2421 */
ext4_add_entry(handle_t * handle,struct dentry * dentry,struct inode * inode)2422 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2423 struct inode *inode)
2424 {
2425 struct inode *dir = d_inode(dentry->d_parent);
2426 struct buffer_head *bh = NULL;
2427 struct ext4_dir_entry_2 *de;
2428 struct super_block *sb;
2429 struct ext4_filename fname;
2430 int retval;
2431 int dx_fallback=0;
2432 unsigned blocksize;
2433 ext4_lblk_t block, blocks;
2434 int csum_size = 0;
2435
2436 if (ext4_has_metadata_csum(inode->i_sb))
2437 csum_size = sizeof(struct ext4_dir_entry_tail);
2438
2439 sb = dir->i_sb;
2440 blocksize = sb->s_blocksize;
2441 if (!dentry->d_name.len)
2442 return -EINVAL;
2443
2444 if (fscrypt_is_nokey_name(dentry))
2445 return -ENOKEY;
2446
2447 #if IS_ENABLED(CONFIG_UNICODE)
2448 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2449 utf8_validate(sb->s_encoding, &dentry->d_name))
2450 return -EINVAL;
2451 #endif
2452
2453 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2454 if (retval)
2455 return retval;
2456
2457 if (ext4_has_inline_data(dir)) {
2458 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2459 if (retval < 0)
2460 goto out;
2461 if (retval == 1) {
2462 retval = 0;
2463 goto out;
2464 }
2465 }
2466
2467 if (is_dx(dir)) {
2468 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2469 if (!retval || (retval != ERR_BAD_DX_DIR))
2470 goto out;
2471 /* Can we just ignore htree data? */
2472 if (ext4_has_metadata_csum(sb)) {
2473 EXT4_ERROR_INODE(dir,
2474 "Directory has corrupted htree index.");
2475 retval = -EFSCORRUPTED;
2476 goto out;
2477 }
2478 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2479 dx_fallback++;
2480 retval = ext4_mark_inode_dirty(handle, dir);
2481 if (unlikely(retval))
2482 goto out;
2483 }
2484 blocks = dir->i_size >> sb->s_blocksize_bits;
2485 for (block = 0; block < blocks; block++) {
2486 bh = ext4_read_dirblock(dir, block, DIRENT);
2487 if (bh == NULL) {
2488 bh = ext4_bread(handle, dir, block,
2489 EXT4_GET_BLOCKS_CREATE);
2490 goto add_to_new_block;
2491 }
2492 if (IS_ERR(bh)) {
2493 retval = PTR_ERR(bh);
2494 bh = NULL;
2495 goto out;
2496 }
2497 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2498 NULL, bh);
2499 if (retval != -ENOSPC)
2500 goto out;
2501
2502 if (blocks == 1 && !dx_fallback &&
2503 ext4_has_feature_dir_index(sb)) {
2504 retval = make_indexed_dir(handle, &fname, dir,
2505 inode, bh);
2506 bh = NULL; /* make_indexed_dir releases bh */
2507 goto out;
2508 }
2509 brelse(bh);
2510 }
2511 bh = ext4_append(handle, dir, &block);
2512 add_to_new_block:
2513 if (IS_ERR(bh)) {
2514 retval = PTR_ERR(bh);
2515 bh = NULL;
2516 goto out;
2517 }
2518 de = (struct ext4_dir_entry_2 *) bh->b_data;
2519 de->inode = 0;
2520 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2521
2522 if (csum_size)
2523 ext4_initialize_dirent_tail(bh, blocksize);
2524
2525 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2526 out:
2527 ext4_fname_free_filename(&fname);
2528 brelse(bh);
2529 if (retval == 0)
2530 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2531 return retval;
2532 }
2533
2534 /*
2535 * Returns 0 for success, or a negative error value
2536 */
ext4_dx_add_entry(handle_t * handle,struct ext4_filename * fname,struct inode * dir,struct inode * inode)2537 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2538 struct inode *dir, struct inode *inode)
2539 {
2540 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2541 struct dx_entry *entries, *at;
2542 struct buffer_head *bh;
2543 struct super_block *sb = dir->i_sb;
2544 struct ext4_dir_entry_2 *de;
2545 int restart;
2546 int err;
2547
2548 again:
2549 restart = 0;
2550 frame = dx_probe(fname, dir, NULL, frames);
2551 if (IS_ERR(frame))
2552 return PTR_ERR(frame);
2553 entries = frame->entries;
2554 at = frame->at;
2555 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2556 if (IS_ERR(bh)) {
2557 err = PTR_ERR(bh);
2558 bh = NULL;
2559 goto cleanup;
2560 }
2561
2562 BUFFER_TRACE(bh, "get_write_access");
2563 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2564 if (err)
2565 goto journal_error;
2566
2567 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2568 if (err != -ENOSPC)
2569 goto cleanup;
2570
2571 err = 0;
2572 /* Block full, should compress but for now just split */
2573 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2574 dx_get_count(entries), dx_get_limit(entries)));
2575 /* Need to split index? */
2576 if (dx_get_count(entries) == dx_get_limit(entries)) {
2577 ext4_lblk_t newblock;
2578 int levels = frame - frames + 1;
2579 unsigned int icount;
2580 int add_level = 1;
2581 struct dx_entry *entries2;
2582 struct dx_node *node2;
2583 struct buffer_head *bh2;
2584
2585 while (frame > frames) {
2586 if (dx_get_count((frame - 1)->entries) <
2587 dx_get_limit((frame - 1)->entries)) {
2588 add_level = 0;
2589 break;
2590 }
2591 frame--; /* split higher index block */
2592 at = frame->at;
2593 entries = frame->entries;
2594 restart = 1;
2595 }
2596 if (add_level && levels == ext4_dir_htree_level(sb)) {
2597 ext4_warning(sb, "Directory (ino: %lu) index full, "
2598 "reach max htree level :%d",
2599 dir->i_ino, levels);
2600 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2601 ext4_warning(sb, "Large directory feature is "
2602 "not enabled on this "
2603 "filesystem");
2604 }
2605 err = -ENOSPC;
2606 goto cleanup;
2607 }
2608 icount = dx_get_count(entries);
2609 bh2 = ext4_append(handle, dir, &newblock);
2610 if (IS_ERR(bh2)) {
2611 err = PTR_ERR(bh2);
2612 goto cleanup;
2613 }
2614 node2 = (struct dx_node *)(bh2->b_data);
2615 entries2 = node2->entries;
2616 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2617 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2618 sb->s_blocksize);
2619 BUFFER_TRACE(frame->bh, "get_write_access");
2620 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2621 EXT4_JTR_NONE);
2622 if (err)
2623 goto journal_error;
2624 if (!add_level) {
2625 unsigned icount1 = icount/2, icount2 = icount - icount1;
2626 unsigned hash2 = dx_get_hash(entries + icount1);
2627 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2628 icount1, icount2));
2629
2630 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2631 err = ext4_journal_get_write_access(handle, sb,
2632 (frame - 1)->bh,
2633 EXT4_JTR_NONE);
2634 if (err)
2635 goto journal_error;
2636
2637 memcpy((char *) entries2, (char *) (entries + icount1),
2638 icount2 * sizeof(struct dx_entry));
2639 dx_set_count(entries, icount1);
2640 dx_set_count(entries2, icount2);
2641 dx_set_limit(entries2, dx_node_limit(dir));
2642
2643 /* Which index block gets the new entry? */
2644 if (at - entries >= icount1) {
2645 frame->at = at - entries - icount1 + entries2;
2646 frame->entries = entries = entries2;
2647 swap(frame->bh, bh2);
2648 }
2649 dx_insert_block((frame - 1), hash2, newblock);
2650 dxtrace(dx_show_index("node", frame->entries));
2651 dxtrace(dx_show_index("node",
2652 ((struct dx_node *) bh2->b_data)->entries));
2653 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2654 if (err)
2655 goto journal_error;
2656 brelse (bh2);
2657 err = ext4_handle_dirty_dx_node(handle, dir,
2658 (frame - 1)->bh);
2659 if (err)
2660 goto journal_error;
2661 err = ext4_handle_dirty_dx_node(handle, dir,
2662 frame->bh);
2663 if (restart || err)
2664 goto journal_error;
2665 } else {
2666 struct dx_root *dxroot;
2667 memcpy((char *) entries2, (char *) entries,
2668 icount * sizeof(struct dx_entry));
2669 dx_set_limit(entries2, dx_node_limit(dir));
2670
2671 /* Set up root */
2672 dx_set_count(entries, 1);
2673 dx_set_block(entries + 0, newblock);
2674 dxroot = (struct dx_root *)frames[0].bh->b_data;
2675 dxroot->info.indirect_levels += 1;
2676 dxtrace(printk(KERN_DEBUG
2677 "Creating %d level index...\n",
2678 dxroot->info.indirect_levels));
2679 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2680 if (err)
2681 goto journal_error;
2682 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2683 brelse(bh2);
2684 restart = 1;
2685 goto journal_error;
2686 }
2687 }
2688 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2689 if (IS_ERR(de)) {
2690 err = PTR_ERR(de);
2691 goto cleanup;
2692 }
2693 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2694 goto cleanup;
2695
2696 journal_error:
2697 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2698 cleanup:
2699 brelse(bh);
2700 dx_release(frames);
2701 /* @restart is true means htree-path has been changed, we need to
2702 * repeat dx_probe() to find out valid htree-path
2703 */
2704 if (restart && err == 0)
2705 goto again;
2706 return err;
2707 }
2708
2709 /*
2710 * ext4_generic_delete_entry deletes a directory entry by merging it
2711 * with the previous entry
2712 */
ext4_generic_delete_entry(struct inode * dir,struct ext4_dir_entry_2 * de_del,struct buffer_head * bh,void * entry_buf,int buf_size,int csum_size)2713 int ext4_generic_delete_entry(struct inode *dir,
2714 struct ext4_dir_entry_2 *de_del,
2715 struct buffer_head *bh,
2716 void *entry_buf,
2717 int buf_size,
2718 int csum_size)
2719 {
2720 struct ext4_dir_entry_2 *de, *pde;
2721 unsigned int blocksize = dir->i_sb->s_blocksize;
2722 int i;
2723
2724 i = 0;
2725 pde = NULL;
2726 de = entry_buf;
2727 while (i < buf_size - csum_size) {
2728 if (ext4_check_dir_entry(dir, NULL, de, bh,
2729 entry_buf, buf_size, i))
2730 return -EFSCORRUPTED;
2731 if (de == de_del) {
2732 if (pde) {
2733 pde->rec_len = ext4_rec_len_to_disk(
2734 ext4_rec_len_from_disk(pde->rec_len,
2735 blocksize) +
2736 ext4_rec_len_from_disk(de->rec_len,
2737 blocksize),
2738 blocksize);
2739
2740 /* wipe entire dir_entry */
2741 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2742 blocksize));
2743 } else {
2744 /* wipe dir_entry excluding the rec_len field */
2745 de->inode = 0;
2746 memset(&de->name_len, 0,
2747 ext4_rec_len_from_disk(de->rec_len,
2748 blocksize) -
2749 offsetof(struct ext4_dir_entry_2,
2750 name_len));
2751 }
2752
2753 inode_inc_iversion(dir);
2754 return 0;
2755 }
2756 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2757 pde = de;
2758 de = ext4_next_entry(de, blocksize);
2759 }
2760 return -ENOENT;
2761 }
2762
ext4_delete_entry(handle_t * handle,struct inode * dir,struct ext4_dir_entry_2 * de_del,struct buffer_head * bh)2763 static int ext4_delete_entry(handle_t *handle,
2764 struct inode *dir,
2765 struct ext4_dir_entry_2 *de_del,
2766 struct buffer_head *bh)
2767 {
2768 int err, csum_size = 0;
2769
2770 if (ext4_has_inline_data(dir)) {
2771 int has_inline_data = 1;
2772 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2773 &has_inline_data);
2774 if (has_inline_data)
2775 return err;
2776 }
2777
2778 if (ext4_has_metadata_csum(dir->i_sb))
2779 csum_size = sizeof(struct ext4_dir_entry_tail);
2780
2781 BUFFER_TRACE(bh, "get_write_access");
2782 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2783 EXT4_JTR_NONE);
2784 if (unlikely(err))
2785 goto out;
2786
2787 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2788 dir->i_sb->s_blocksize, csum_size);
2789 if (err)
2790 goto out;
2791
2792 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2793 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2794 if (unlikely(err))
2795 goto out;
2796
2797 return 0;
2798 out:
2799 if (err != -ENOENT)
2800 ext4_std_error(dir->i_sb, err);
2801 return err;
2802 }
2803
2804 /*
2805 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2806 * since this indicates that nlinks count was previously 1 to avoid overflowing
2807 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2808 * that subdirectory link counts are not being maintained accurately.
2809 *
2810 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2811 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2812 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2813 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2814 */
ext4_inc_count(struct inode * inode)2815 static void ext4_inc_count(struct inode *inode)
2816 {
2817 inc_nlink(inode);
2818 if (is_dx(inode) &&
2819 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2820 set_nlink(inode, 1);
2821 }
2822
2823 /*
2824 * If a directory had nlink == 1, then we should let it be 1. This indicates
2825 * directory has >EXT4_LINK_MAX subdirs.
2826 */
ext4_dec_count(struct inode * inode)2827 static void ext4_dec_count(struct inode *inode)
2828 {
2829 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2830 drop_nlink(inode);
2831 }
2832
2833
2834 /*
2835 * Add non-directory inode to a directory. On success, the inode reference is
2836 * consumed by dentry is instantiation. This is also indicated by clearing of
2837 * *inodep pointer. On failure, the caller is responsible for dropping the
2838 * inode reference in the safe context.
2839 */
ext4_add_nondir(handle_t * handle,struct dentry * dentry,struct inode ** inodep)2840 static int ext4_add_nondir(handle_t *handle,
2841 struct dentry *dentry, struct inode **inodep)
2842 {
2843 struct inode *dir = d_inode(dentry->d_parent);
2844 struct inode *inode = *inodep;
2845 int err = ext4_add_entry(handle, dentry, inode);
2846 if (!err) {
2847 err = ext4_mark_inode_dirty(handle, inode);
2848 if (IS_DIRSYNC(dir))
2849 ext4_handle_sync(handle);
2850 d_instantiate_new(dentry, inode);
2851 *inodep = NULL;
2852 return err;
2853 }
2854 drop_nlink(inode);
2855 ext4_mark_inode_dirty(handle, inode);
2856 ext4_orphan_add(handle, inode);
2857 unlock_new_inode(inode);
2858 return err;
2859 }
2860
2861 /*
2862 * By the time this is called, we already have created
2863 * the directory cache entry for the new file, but it
2864 * is so far negative - it has no inode.
2865 *
2866 * If the create succeeds, we fill in the inode information
2867 * with d_instantiate().
2868 */
ext4_create(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)2869 static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2870 struct dentry *dentry, umode_t mode, bool excl)
2871 {
2872 handle_t *handle;
2873 struct inode *inode;
2874 int err, credits, retries = 0;
2875
2876 err = dquot_initialize(dir);
2877 if (err)
2878 return err;
2879
2880 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2881 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2882 retry:
2883 inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2884 0, NULL, EXT4_HT_DIR, credits);
2885 handle = ext4_journal_current_handle();
2886 err = PTR_ERR(inode);
2887 if (!IS_ERR(inode)) {
2888 inode->i_op = &ext4_file_inode_operations;
2889 inode->i_fop = &ext4_file_operations;
2890 ext4_set_aops(inode);
2891 err = ext4_add_nondir(handle, dentry, &inode);
2892 if (!err)
2893 ext4_fc_track_create(handle, dentry);
2894 }
2895 if (handle)
2896 ext4_journal_stop(handle);
2897 if (!IS_ERR_OR_NULL(inode))
2898 iput(inode);
2899 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2900 goto retry;
2901 return err;
2902 }
2903
ext4_mknod(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)2904 static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2905 struct dentry *dentry, umode_t mode, dev_t rdev)
2906 {
2907 handle_t *handle;
2908 struct inode *inode;
2909 int err, credits, retries = 0;
2910
2911 err = dquot_initialize(dir);
2912 if (err)
2913 return err;
2914
2915 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2916 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2917 retry:
2918 inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2919 0, NULL, EXT4_HT_DIR, credits);
2920 handle = ext4_journal_current_handle();
2921 err = PTR_ERR(inode);
2922 if (!IS_ERR(inode)) {
2923 init_special_inode(inode, inode->i_mode, rdev);
2924 inode->i_op = &ext4_special_inode_operations;
2925 err = ext4_add_nondir(handle, dentry, &inode);
2926 if (!err)
2927 ext4_fc_track_create(handle, dentry);
2928 }
2929 if (handle)
2930 ext4_journal_stop(handle);
2931 if (!IS_ERR_OR_NULL(inode))
2932 iput(inode);
2933 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2934 goto retry;
2935 return err;
2936 }
2937
ext4_tmpfile(struct mnt_idmap * idmap,struct inode * dir,struct file * file,umode_t mode)2938 static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2939 struct file *file, umode_t mode)
2940 {
2941 handle_t *handle;
2942 struct inode *inode;
2943 int err, retries = 0;
2944
2945 err = dquot_initialize(dir);
2946 if (err)
2947 return err;
2948
2949 retry:
2950 inode = ext4_new_inode_start_handle(idmap, dir, mode,
2951 NULL, 0, NULL,
2952 EXT4_HT_DIR,
2953 EXT4_MAXQUOTAS_TRANS_BLOCKS(dir->i_sb) +
2954 4 + EXT4_XATTR_TRANS_BLOCKS);
2955 handle = ext4_journal_current_handle();
2956 err = PTR_ERR(inode);
2957 if (!IS_ERR(inode)) {
2958 inode->i_op = &ext4_file_inode_operations;
2959 inode->i_fop = &ext4_file_operations;
2960 ext4_set_aops(inode);
2961 d_tmpfile(file, inode);
2962 err = ext4_orphan_add(handle, inode);
2963 if (err)
2964 goto err_unlock_inode;
2965 mark_inode_dirty(inode);
2966 unlock_new_inode(inode);
2967 }
2968 if (handle)
2969 ext4_journal_stop(handle);
2970 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2971 goto retry;
2972 return finish_open_simple(file, err);
2973 err_unlock_inode:
2974 ext4_journal_stop(handle);
2975 unlock_new_inode(inode);
2976 return err;
2977 }
2978
ext4_init_dot_dotdot(struct inode * inode,struct ext4_dir_entry_2 * de,int blocksize,int csum_size,unsigned int parent_ino,int dotdot_real_len)2979 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2980 struct ext4_dir_entry_2 *de,
2981 int blocksize, int csum_size,
2982 unsigned int parent_ino, int dotdot_real_len)
2983 {
2984 de->inode = cpu_to_le32(inode->i_ino);
2985 de->name_len = 1;
2986 de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2987 blocksize);
2988 strcpy(de->name, ".");
2989 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2990
2991 de = ext4_next_entry(de, blocksize);
2992 de->inode = cpu_to_le32(parent_ino);
2993 de->name_len = 2;
2994 if (!dotdot_real_len)
2995 de->rec_len = ext4_rec_len_to_disk(blocksize -
2996 (csum_size + ext4_dir_rec_len(1, NULL)),
2997 blocksize);
2998 else
2999 de->rec_len = ext4_rec_len_to_disk(
3000 ext4_dir_rec_len(de->name_len, NULL),
3001 blocksize);
3002 strcpy(de->name, "..");
3003 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
3004
3005 return ext4_next_entry(de, blocksize);
3006 }
3007
ext4_init_new_dir(handle_t * handle,struct inode * dir,struct inode * inode)3008 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
3009 struct inode *inode)
3010 {
3011 struct buffer_head *dir_block = NULL;
3012 struct ext4_dir_entry_2 *de;
3013 ext4_lblk_t block = 0;
3014 unsigned int blocksize = dir->i_sb->s_blocksize;
3015 int csum_size = 0;
3016 int err;
3017
3018 if (ext4_has_metadata_csum(dir->i_sb))
3019 csum_size = sizeof(struct ext4_dir_entry_tail);
3020
3021 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3022 err = ext4_try_create_inline_dir(handle, dir, inode);
3023 if (err < 0 && err != -ENOSPC)
3024 goto out;
3025 if (!err)
3026 goto out;
3027 }
3028
3029 inode->i_size = 0;
3030 dir_block = ext4_append(handle, inode, &block);
3031 if (IS_ERR(dir_block))
3032 return PTR_ERR(dir_block);
3033 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
3034 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
3035 set_nlink(inode, 2);
3036 if (csum_size)
3037 ext4_initialize_dirent_tail(dir_block, blocksize);
3038
3039 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
3040 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
3041 if (err)
3042 goto out;
3043 set_buffer_verified(dir_block);
3044 out:
3045 brelse(dir_block);
3046 return err;
3047 }
3048
ext4_mkdir(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,umode_t mode)3049 static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
3050 struct dentry *dentry, umode_t mode)
3051 {
3052 handle_t *handle;
3053 struct inode *inode;
3054 int err, err2 = 0, credits, retries = 0;
3055
3056 if (EXT4_DIR_LINK_MAX(dir))
3057 return -EMLINK;
3058
3059 err = dquot_initialize(dir);
3060 if (err)
3061 return err;
3062
3063 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3064 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3065 retry:
3066 inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3067 &dentry->d_name,
3068 0, NULL, EXT4_HT_DIR, credits);
3069 handle = ext4_journal_current_handle();
3070 err = PTR_ERR(inode);
3071 if (IS_ERR(inode))
3072 goto out_stop;
3073
3074 inode->i_op = &ext4_dir_inode_operations;
3075 inode->i_fop = &ext4_dir_operations;
3076 err = ext4_init_new_dir(handle, dir, inode);
3077 if (err)
3078 goto out_clear_inode;
3079 err = ext4_mark_inode_dirty(handle, inode);
3080 if (!err)
3081 err = ext4_add_entry(handle, dentry, inode);
3082 if (err) {
3083 out_clear_inode:
3084 clear_nlink(inode);
3085 ext4_orphan_add(handle, inode);
3086 unlock_new_inode(inode);
3087 err2 = ext4_mark_inode_dirty(handle, inode);
3088 if (unlikely(err2))
3089 err = err2;
3090 ext4_journal_stop(handle);
3091 iput(inode);
3092 goto out_retry;
3093 }
3094 ext4_inc_count(dir);
3095
3096 ext4_update_dx_flag(dir);
3097 err = ext4_mark_inode_dirty(handle, dir);
3098 if (err)
3099 goto out_clear_inode;
3100 d_instantiate_new(dentry, inode);
3101 ext4_fc_track_create(handle, dentry);
3102 if (IS_DIRSYNC(dir))
3103 ext4_handle_sync(handle);
3104
3105 out_stop:
3106 if (handle)
3107 ext4_journal_stop(handle);
3108 out_retry:
3109 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3110 goto retry;
3111 return err;
3112 }
3113
3114 /*
3115 * routine to check that the specified directory is empty (for rmdir)
3116 */
ext4_empty_dir(struct inode * inode)3117 bool ext4_empty_dir(struct inode *inode)
3118 {
3119 unsigned int offset;
3120 struct buffer_head *bh;
3121 struct ext4_dir_entry_2 *de;
3122 struct super_block *sb;
3123
3124 if (ext4_has_inline_data(inode)) {
3125 int has_inline_data = 1;
3126 int ret;
3127
3128 ret = empty_inline_dir(inode, &has_inline_data);
3129 if (has_inline_data)
3130 return ret;
3131 }
3132
3133 sb = inode->i_sb;
3134 if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3135 ext4_dir_rec_len(2, NULL)) {
3136 EXT4_ERROR_INODE(inode, "invalid size");
3137 return false;
3138 }
3139 bh = ext4_read_dirblock(inode, 0, EITHER);
3140 if (IS_ERR(bh))
3141 return false;
3142
3143 de = (struct ext4_dir_entry_2 *) bh->b_data;
3144 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3145 0) ||
3146 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3147 ext4_warning_inode(inode, "directory missing '.'");
3148 brelse(bh);
3149 return false;
3150 }
3151 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3152 de = ext4_next_entry(de, sb->s_blocksize);
3153 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3154 offset) ||
3155 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3156 ext4_warning_inode(inode, "directory missing '..'");
3157 brelse(bh);
3158 return false;
3159 }
3160 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3161 while (offset < inode->i_size) {
3162 if (!(offset & (sb->s_blocksize - 1))) {
3163 unsigned int lblock;
3164 brelse(bh);
3165 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3166 bh = ext4_read_dirblock(inode, lblock, EITHER);
3167 if (bh == NULL) {
3168 offset += sb->s_blocksize;
3169 continue;
3170 }
3171 if (IS_ERR(bh))
3172 return false;
3173 }
3174 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3175 (offset & (sb->s_blocksize - 1)));
3176 if (ext4_check_dir_entry(inode, NULL, de, bh,
3177 bh->b_data, bh->b_size, offset) ||
3178 le32_to_cpu(de->inode)) {
3179 brelse(bh);
3180 return false;
3181 }
3182 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3183 }
3184 brelse(bh);
3185 return true;
3186 }
3187
ext4_rmdir(struct inode * dir,struct dentry * dentry)3188 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3189 {
3190 int retval;
3191 struct inode *inode;
3192 struct buffer_head *bh;
3193 struct ext4_dir_entry_2 *de;
3194 handle_t *handle = NULL;
3195
3196 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3197 return -EIO;
3198
3199 /* Initialize quotas before so that eventual writes go in
3200 * separate transaction */
3201 retval = dquot_initialize(dir);
3202 if (retval)
3203 return retval;
3204 retval = dquot_initialize(d_inode(dentry));
3205 if (retval)
3206 return retval;
3207
3208 retval = -ENOENT;
3209 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3210 if (IS_ERR(bh))
3211 return PTR_ERR(bh);
3212 if (!bh)
3213 goto end_rmdir;
3214
3215 inode = d_inode(dentry);
3216
3217 retval = -EFSCORRUPTED;
3218 if (le32_to_cpu(de->inode) != inode->i_ino)
3219 goto end_rmdir;
3220
3221 retval = -ENOTEMPTY;
3222 if (!ext4_empty_dir(inode))
3223 goto end_rmdir;
3224
3225 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3226 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3227 if (IS_ERR(handle)) {
3228 retval = PTR_ERR(handle);
3229 handle = NULL;
3230 goto end_rmdir;
3231 }
3232
3233 if (IS_DIRSYNC(dir))
3234 ext4_handle_sync(handle);
3235
3236 retval = ext4_delete_entry(handle, dir, de, bh);
3237 if (retval)
3238 goto end_rmdir;
3239 if (!EXT4_DIR_LINK_EMPTY(inode))
3240 ext4_warning_inode(inode,
3241 "empty directory '%.*s' has too many links (%u)",
3242 dentry->d_name.len, dentry->d_name.name,
3243 inode->i_nlink);
3244 inode_inc_iversion(inode);
3245 clear_nlink(inode);
3246 /* There's no need to set i_disksize: the fact that i_nlink is
3247 * zero will ensure that the right thing happens during any
3248 * recovery. */
3249 inode->i_size = 0;
3250 ext4_orphan_add(handle, inode);
3251 dir->i_mtime = inode_set_ctime_current(dir);
3252 inode_set_ctime_current(inode);
3253 retval = ext4_mark_inode_dirty(handle, inode);
3254 if (retval)
3255 goto end_rmdir;
3256 ext4_dec_count(dir);
3257 ext4_update_dx_flag(dir);
3258 ext4_fc_track_unlink(handle, dentry);
3259 retval = ext4_mark_inode_dirty(handle, dir);
3260
3261 #if IS_ENABLED(CONFIG_UNICODE)
3262 /* VFS negative dentries are incompatible with Encoding and
3263 * Case-insensitiveness. Eventually we'll want avoid
3264 * invalidating the dentries here, alongside with returning the
3265 * negative dentries at ext4_lookup(), when it is better
3266 * supported by the VFS for the CI case.
3267 */
3268 if (IS_CASEFOLDED(dir))
3269 d_invalidate(dentry);
3270 #endif
3271
3272 end_rmdir:
3273 brelse(bh);
3274 if (handle)
3275 ext4_journal_stop(handle);
3276 return retval;
3277 }
3278
__ext4_unlink(struct inode * dir,const struct qstr * d_name,struct inode * inode,struct dentry * dentry)3279 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3280 struct inode *inode,
3281 struct dentry *dentry /* NULL during fast_commit recovery */)
3282 {
3283 int retval = -ENOENT;
3284 struct buffer_head *bh;
3285 struct ext4_dir_entry_2 *de;
3286 handle_t *handle;
3287 int skip_remove_dentry = 0;
3288
3289 /*
3290 * Keep this outside the transaction; it may have to set up the
3291 * directory's encryption key, which isn't GFP_NOFS-safe.
3292 */
3293 bh = ext4_find_entry(dir, d_name, &de, NULL);
3294 if (IS_ERR(bh))
3295 return PTR_ERR(bh);
3296
3297 if (!bh)
3298 return -ENOENT;
3299
3300 if (le32_to_cpu(de->inode) != inode->i_ino) {
3301 /*
3302 * It's okay if we find dont find dentry which matches
3303 * the inode. That's because it might have gotten
3304 * renamed to a different inode number
3305 */
3306 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3307 skip_remove_dentry = 1;
3308 else
3309 goto out_bh;
3310 }
3311
3312 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3313 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3314 if (IS_ERR(handle)) {
3315 retval = PTR_ERR(handle);
3316 goto out_bh;
3317 }
3318
3319 if (IS_DIRSYNC(dir))
3320 ext4_handle_sync(handle);
3321
3322 if (!skip_remove_dentry) {
3323 retval = ext4_delete_entry(handle, dir, de, bh);
3324 if (retval)
3325 goto out_handle;
3326 dir->i_mtime = inode_set_ctime_current(dir);
3327 ext4_update_dx_flag(dir);
3328 retval = ext4_mark_inode_dirty(handle, dir);
3329 if (retval)
3330 goto out_handle;
3331 } else {
3332 retval = 0;
3333 }
3334 if (inode->i_nlink == 0)
3335 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3336 d_name->len, d_name->name);
3337 else
3338 drop_nlink(inode);
3339 if (!inode->i_nlink)
3340 ext4_orphan_add(handle, inode);
3341 inode_set_ctime_current(inode);
3342 retval = ext4_mark_inode_dirty(handle, inode);
3343 if (dentry && !retval)
3344 ext4_fc_track_unlink(handle, dentry);
3345 out_handle:
3346 ext4_journal_stop(handle);
3347 out_bh:
3348 brelse(bh);
3349 return retval;
3350 }
3351
ext4_unlink(struct inode * dir,struct dentry * dentry)3352 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3353 {
3354 int retval;
3355
3356 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3357 return -EIO;
3358
3359 trace_ext4_unlink_enter(dir, dentry);
3360 /*
3361 * Initialize quotas before so that eventual writes go
3362 * in separate transaction
3363 */
3364 retval = dquot_initialize(dir);
3365 if (retval)
3366 goto out_trace;
3367 retval = dquot_initialize(d_inode(dentry));
3368 if (retval)
3369 goto out_trace;
3370
3371 retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3372 #if IS_ENABLED(CONFIG_UNICODE)
3373 /* VFS negative dentries are incompatible with Encoding and
3374 * Case-insensitiveness. Eventually we'll want avoid
3375 * invalidating the dentries here, alongside with returning the
3376 * negative dentries at ext4_lookup(), when it is better
3377 * supported by the VFS for the CI case.
3378 */
3379 if (IS_CASEFOLDED(dir))
3380 d_invalidate(dentry);
3381 #endif
3382
3383 out_trace:
3384 trace_ext4_unlink_exit(dentry, retval);
3385 return retval;
3386 }
3387
ext4_init_symlink_block(handle_t * handle,struct inode * inode,struct fscrypt_str * disk_link)3388 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3389 struct fscrypt_str *disk_link)
3390 {
3391 struct buffer_head *bh;
3392 char *kaddr;
3393 int err = 0;
3394
3395 bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3396 if (IS_ERR(bh))
3397 return PTR_ERR(bh);
3398
3399 BUFFER_TRACE(bh, "get_write_access");
3400 err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3401 if (err)
3402 goto out;
3403
3404 kaddr = (char *)bh->b_data;
3405 memcpy(kaddr, disk_link->name, disk_link->len);
3406 inode->i_size = disk_link->len - 1;
3407 EXT4_I(inode)->i_disksize = inode->i_size;
3408 err = ext4_handle_dirty_metadata(handle, inode, bh);
3409 out:
3410 brelse(bh);
3411 return err;
3412 }
3413
ext4_symlink(struct mnt_idmap * idmap,struct inode * dir,struct dentry * dentry,const char * symname)3414 static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3415 struct dentry *dentry, const char *symname)
3416 {
3417 handle_t *handle;
3418 struct inode *inode;
3419 int err, len = strlen(symname);
3420 int credits;
3421 struct fscrypt_str disk_link;
3422 int retries = 0;
3423
3424 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3425 return -EIO;
3426
3427 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3428 &disk_link);
3429 if (err)
3430 return err;
3431
3432 err = dquot_initialize(dir);
3433 if (err)
3434 return err;
3435
3436 /*
3437 * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3438 * directory. +3 for inode, inode bitmap, group descriptor allocation.
3439 * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3440 * modification.
3441 */
3442 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3443 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3444 retry:
3445 inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3446 &dentry->d_name, 0, NULL,
3447 EXT4_HT_DIR, credits);
3448 handle = ext4_journal_current_handle();
3449 if (IS_ERR(inode)) {
3450 if (handle)
3451 ext4_journal_stop(handle);
3452 err = PTR_ERR(inode);
3453 goto out_retry;
3454 }
3455
3456 if (IS_ENCRYPTED(inode)) {
3457 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3458 if (err)
3459 goto err_drop_inode;
3460 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3461 } else {
3462 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3463 inode->i_op = &ext4_symlink_inode_operations;
3464 } else {
3465 inode->i_op = &ext4_fast_symlink_inode_operations;
3466 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3467 }
3468 }
3469
3470 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3471 /* alloc symlink block and fill it */
3472 err = ext4_init_symlink_block(handle, inode, &disk_link);
3473 if (err)
3474 goto err_drop_inode;
3475 } else {
3476 /* clear the extent format for fast symlink */
3477 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3478 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3479 disk_link.len);
3480 inode->i_size = disk_link.len - 1;
3481 EXT4_I(inode)->i_disksize = inode->i_size;
3482 }
3483 err = ext4_add_nondir(handle, dentry, &inode);
3484 if (handle)
3485 ext4_journal_stop(handle);
3486 iput(inode);
3487 goto out_retry;
3488
3489 err_drop_inode:
3490 clear_nlink(inode);
3491 ext4_mark_inode_dirty(handle, inode);
3492 ext4_orphan_add(handle, inode);
3493 unlock_new_inode(inode);
3494 if (handle)
3495 ext4_journal_stop(handle);
3496 iput(inode);
3497 out_retry:
3498 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3499 goto retry;
3500 if (disk_link.name != (unsigned char *)symname)
3501 kfree(disk_link.name);
3502 return err;
3503 }
3504
__ext4_link(struct inode * dir,struct inode * inode,struct dentry * dentry)3505 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3506 {
3507 handle_t *handle;
3508 int err, retries = 0;
3509 retry:
3510 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3511 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3512 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3513 if (IS_ERR(handle))
3514 return PTR_ERR(handle);
3515
3516 if (IS_DIRSYNC(dir))
3517 ext4_handle_sync(handle);
3518
3519 inode_set_ctime_current(inode);
3520 ext4_inc_count(inode);
3521 ihold(inode);
3522
3523 err = ext4_add_entry(handle, dentry, inode);
3524 if (!err) {
3525 err = ext4_mark_inode_dirty(handle, inode);
3526 /* this can happen only for tmpfile being
3527 * linked the first time
3528 */
3529 if (inode->i_nlink == 1)
3530 ext4_orphan_del(handle, inode);
3531 d_instantiate(dentry, inode);
3532 ext4_fc_track_link(handle, dentry);
3533 } else {
3534 drop_nlink(inode);
3535 iput(inode);
3536 }
3537 ext4_journal_stop(handle);
3538 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3539 goto retry;
3540 return err;
3541 }
3542
ext4_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)3543 static int ext4_link(struct dentry *old_dentry,
3544 struct inode *dir, struct dentry *dentry)
3545 {
3546 struct inode *inode = d_inode(old_dentry);
3547 int err;
3548
3549 if (inode->i_nlink >= EXT4_LINK_MAX)
3550 return -EMLINK;
3551
3552 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3553 if (err)
3554 return err;
3555
3556 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3557 (!projid_eq(EXT4_I(dir)->i_projid,
3558 EXT4_I(old_dentry->d_inode)->i_projid)))
3559 return -EXDEV;
3560
3561 err = dquot_initialize(dir);
3562 if (err)
3563 return err;
3564 return __ext4_link(dir, inode, dentry);
3565 }
3566
3567 /*
3568 * Try to find buffer head where contains the parent block.
3569 * It should be the inode block if it is inlined or the 1st block
3570 * if it is a normal dir.
3571 */
ext4_get_first_dir_block(handle_t * handle,struct inode * inode,int * retval,struct ext4_dir_entry_2 ** parent_de,int * inlined)3572 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3573 struct inode *inode,
3574 int *retval,
3575 struct ext4_dir_entry_2 **parent_de,
3576 int *inlined)
3577 {
3578 struct buffer_head *bh;
3579
3580 if (!ext4_has_inline_data(inode)) {
3581 struct ext4_dir_entry_2 *de;
3582 unsigned int offset;
3583
3584 bh = ext4_read_dirblock(inode, 0, EITHER);
3585 if (IS_ERR(bh)) {
3586 *retval = PTR_ERR(bh);
3587 return NULL;
3588 }
3589
3590 de = (struct ext4_dir_entry_2 *) bh->b_data;
3591 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3592 bh->b_size, 0) ||
3593 le32_to_cpu(de->inode) != inode->i_ino ||
3594 strcmp(".", de->name)) {
3595 EXT4_ERROR_INODE(inode, "directory missing '.'");
3596 brelse(bh);
3597 *retval = -EFSCORRUPTED;
3598 return NULL;
3599 }
3600 offset = ext4_rec_len_from_disk(de->rec_len,
3601 inode->i_sb->s_blocksize);
3602 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3603 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3604 bh->b_size, offset) ||
3605 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3606 EXT4_ERROR_INODE(inode, "directory missing '..'");
3607 brelse(bh);
3608 *retval = -EFSCORRUPTED;
3609 return NULL;
3610 }
3611 *parent_de = de;
3612
3613 return bh;
3614 }
3615
3616 *inlined = 1;
3617 return ext4_get_first_inline_block(inode, parent_de, retval);
3618 }
3619
3620 struct ext4_renament {
3621 struct inode *dir;
3622 struct dentry *dentry;
3623 struct inode *inode;
3624 bool is_dir;
3625 int dir_nlink_delta;
3626
3627 /* entry for "dentry" */
3628 struct buffer_head *bh;
3629 struct ext4_dir_entry_2 *de;
3630 int inlined;
3631
3632 /* entry for ".." in inode if it's a directory */
3633 struct buffer_head *dir_bh;
3634 struct ext4_dir_entry_2 *parent_de;
3635 int dir_inlined;
3636 };
3637
ext4_rename_dir_prepare(handle_t * handle,struct ext4_renament * ent)3638 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3639 {
3640 int retval;
3641
3642 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3643 &retval, &ent->parent_de,
3644 &ent->dir_inlined);
3645 if (!ent->dir_bh)
3646 return retval;
3647 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3648 return -EFSCORRUPTED;
3649 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3650 return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3651 ent->dir_bh, EXT4_JTR_NONE);
3652 }
3653
ext4_rename_dir_finish(handle_t * handle,struct ext4_renament * ent,unsigned dir_ino)3654 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3655 unsigned dir_ino)
3656 {
3657 int retval;
3658
3659 ent->parent_de->inode = cpu_to_le32(dir_ino);
3660 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3661 if (!ent->dir_inlined) {
3662 if (is_dx(ent->inode)) {
3663 retval = ext4_handle_dirty_dx_node(handle,
3664 ent->inode,
3665 ent->dir_bh);
3666 } else {
3667 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3668 ent->dir_bh);
3669 }
3670 } else {
3671 retval = ext4_mark_inode_dirty(handle, ent->inode);
3672 }
3673 if (retval) {
3674 ext4_std_error(ent->dir->i_sb, retval);
3675 return retval;
3676 }
3677 return 0;
3678 }
3679
ext4_setent(handle_t * handle,struct ext4_renament * ent,unsigned ino,unsigned file_type)3680 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3681 unsigned ino, unsigned file_type)
3682 {
3683 int retval, retval2;
3684
3685 BUFFER_TRACE(ent->bh, "get write access");
3686 retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3687 EXT4_JTR_NONE);
3688 if (retval)
3689 return retval;
3690 ent->de->inode = cpu_to_le32(ino);
3691 if (ext4_has_feature_filetype(ent->dir->i_sb))
3692 ent->de->file_type = file_type;
3693 inode_inc_iversion(ent->dir);
3694 ent->dir->i_mtime = inode_set_ctime_current(ent->dir);
3695 retval = ext4_mark_inode_dirty(handle, ent->dir);
3696 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3697 if (!ent->inlined) {
3698 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3699 if (unlikely(retval2)) {
3700 ext4_std_error(ent->dir->i_sb, retval2);
3701 return retval2;
3702 }
3703 }
3704 return retval;
3705 }
3706
ext4_resetent(handle_t * handle,struct ext4_renament * ent,unsigned ino,unsigned file_type)3707 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3708 unsigned ino, unsigned file_type)
3709 {
3710 struct ext4_renament old = *ent;
3711 int retval = 0;
3712
3713 /*
3714 * old->de could have moved from under us during make indexed dir,
3715 * so the old->de may no longer valid and need to find it again
3716 * before reset old inode info.
3717 */
3718 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3719 &old.inlined);
3720 if (IS_ERR(old.bh))
3721 retval = PTR_ERR(old.bh);
3722 if (!old.bh)
3723 retval = -ENOENT;
3724 if (retval) {
3725 ext4_std_error(old.dir->i_sb, retval);
3726 return;
3727 }
3728
3729 ext4_setent(handle, &old, ino, file_type);
3730 brelse(old.bh);
3731 }
3732
ext4_find_delete_entry(handle_t * handle,struct inode * dir,const struct qstr * d_name)3733 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3734 const struct qstr *d_name)
3735 {
3736 int retval = -ENOENT;
3737 struct buffer_head *bh;
3738 struct ext4_dir_entry_2 *de;
3739
3740 bh = ext4_find_entry(dir, d_name, &de, NULL);
3741 if (IS_ERR(bh))
3742 return PTR_ERR(bh);
3743 if (bh) {
3744 retval = ext4_delete_entry(handle, dir, de, bh);
3745 brelse(bh);
3746 }
3747 return retval;
3748 }
3749
ext4_rename_delete(handle_t * handle,struct ext4_renament * ent,int force_reread)3750 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3751 int force_reread)
3752 {
3753 int retval;
3754 /*
3755 * ent->de could have moved from under us during htree split, so make
3756 * sure that we are deleting the right entry. We might also be pointing
3757 * to a stale entry in the unused part of ent->bh so just checking inum
3758 * and the name isn't enough.
3759 */
3760 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3761 ent->de->name_len != ent->dentry->d_name.len ||
3762 strncmp(ent->de->name, ent->dentry->d_name.name,
3763 ent->de->name_len) ||
3764 force_reread) {
3765 retval = ext4_find_delete_entry(handle, ent->dir,
3766 &ent->dentry->d_name);
3767 } else {
3768 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3769 if (retval == -ENOENT) {
3770 retval = ext4_find_delete_entry(handle, ent->dir,
3771 &ent->dentry->d_name);
3772 }
3773 }
3774
3775 if (retval) {
3776 ext4_warning_inode(ent->dir,
3777 "Deleting old file: nlink %d, error=%d",
3778 ent->dir->i_nlink, retval);
3779 }
3780 }
3781
ext4_update_dir_count(handle_t * handle,struct ext4_renament * ent)3782 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3783 {
3784 if (ent->dir_nlink_delta) {
3785 if (ent->dir_nlink_delta == -1)
3786 ext4_dec_count(ent->dir);
3787 else
3788 ext4_inc_count(ent->dir);
3789 ext4_mark_inode_dirty(handle, ent->dir);
3790 }
3791 }
3792
ext4_whiteout_for_rename(struct mnt_idmap * idmap,struct ext4_renament * ent,int credits,handle_t ** h)3793 static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3794 struct ext4_renament *ent,
3795 int credits, handle_t **h)
3796 {
3797 struct inode *wh;
3798 handle_t *handle;
3799 int retries = 0;
3800
3801 /*
3802 * for inode block, sb block, group summaries,
3803 * and inode bitmap
3804 */
3805 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3806 EXT4_XATTR_TRANS_BLOCKS + 4);
3807 retry:
3808 wh = ext4_new_inode_start_handle(idmap, ent->dir,
3809 S_IFCHR | WHITEOUT_MODE,
3810 &ent->dentry->d_name, 0, NULL,
3811 EXT4_HT_DIR, credits);
3812
3813 handle = ext4_journal_current_handle();
3814 if (IS_ERR(wh)) {
3815 if (handle)
3816 ext4_journal_stop(handle);
3817 if (PTR_ERR(wh) == -ENOSPC &&
3818 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3819 goto retry;
3820 } else {
3821 *h = handle;
3822 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3823 wh->i_op = &ext4_special_inode_operations;
3824 }
3825 return wh;
3826 }
3827
3828 /*
3829 * Anybody can rename anything with this: the permission checks are left to the
3830 * higher-level routines.
3831 *
3832 * n.b. old_{dentry,inode) refers to the source dentry/inode
3833 * while new_{dentry,inode) refers to the destination dentry/inode
3834 * This comes from rename(const char *oldpath, const char *newpath)
3835 */
ext4_rename(struct mnt_idmap * idmap,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)3836 static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3837 struct dentry *old_dentry, struct inode *new_dir,
3838 struct dentry *new_dentry, unsigned int flags)
3839 {
3840 handle_t *handle = NULL;
3841 struct ext4_renament old = {
3842 .dir = old_dir,
3843 .dentry = old_dentry,
3844 .inode = d_inode(old_dentry),
3845 };
3846 struct ext4_renament new = {
3847 .dir = new_dir,
3848 .dentry = new_dentry,
3849 .inode = d_inode(new_dentry),
3850 };
3851 int force_reread;
3852 int retval;
3853 struct inode *whiteout = NULL;
3854 int credits;
3855 u8 old_file_type;
3856
3857 if (new.inode && new.inode->i_nlink == 0) {
3858 EXT4_ERROR_INODE(new.inode,
3859 "target of rename is already freed");
3860 return -EFSCORRUPTED;
3861 }
3862
3863 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3864 (!projid_eq(EXT4_I(new_dir)->i_projid,
3865 EXT4_I(old_dentry->d_inode)->i_projid)))
3866 return -EXDEV;
3867
3868 retval = dquot_initialize(old.dir);
3869 if (retval)
3870 return retval;
3871 retval = dquot_initialize(old.inode);
3872 if (retval)
3873 return retval;
3874 retval = dquot_initialize(new.dir);
3875 if (retval)
3876 return retval;
3877
3878 /* Initialize quotas before so that eventual writes go
3879 * in separate transaction */
3880 if (new.inode) {
3881 retval = dquot_initialize(new.inode);
3882 if (retval)
3883 return retval;
3884 }
3885
3886 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3887 &old.inlined);
3888 if (IS_ERR(old.bh))
3889 return PTR_ERR(old.bh);
3890
3891 /*
3892 * Check for inode number is _not_ due to possible IO errors.
3893 * We might rmdir the source, keep it as pwd of some process
3894 * and merrily kill the link to whatever was created under the
3895 * same name. Goodbye sticky bit ;-<
3896 */
3897 retval = -ENOENT;
3898 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3899 goto release_bh;
3900
3901 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3902 &new.de, &new.inlined);
3903 if (IS_ERR(new.bh)) {
3904 retval = PTR_ERR(new.bh);
3905 new.bh = NULL;
3906 goto release_bh;
3907 }
3908 if (new.bh) {
3909 if (!new.inode) {
3910 brelse(new.bh);
3911 new.bh = NULL;
3912 }
3913 }
3914 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3915 ext4_alloc_da_blocks(old.inode);
3916
3917 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3918 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3919 if (!(flags & RENAME_WHITEOUT)) {
3920 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3921 if (IS_ERR(handle)) {
3922 retval = PTR_ERR(handle);
3923 goto release_bh;
3924 }
3925 } else {
3926 whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3927 if (IS_ERR(whiteout)) {
3928 retval = PTR_ERR(whiteout);
3929 goto release_bh;
3930 }
3931 }
3932
3933 old_file_type = old.de->file_type;
3934 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3935 ext4_handle_sync(handle);
3936
3937 if (S_ISDIR(old.inode->i_mode)) {
3938 if (new.inode) {
3939 retval = -ENOTEMPTY;
3940 if (!ext4_empty_dir(new.inode))
3941 goto end_rename;
3942 } else {
3943 retval = -EMLINK;
3944 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3945 goto end_rename;
3946 }
3947 retval = ext4_rename_dir_prepare(handle, &old);
3948 if (retval)
3949 goto end_rename;
3950 }
3951 /*
3952 * If we're renaming a file within an inline_data dir and adding or
3953 * setting the new dirent causes a conversion from inline_data to
3954 * extents/blockmap, we need to force the dirent delete code to
3955 * re-read the directory, or else we end up trying to delete a dirent
3956 * from what is now the extent tree root (or a block map).
3957 */
3958 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3959 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3960
3961 if (whiteout) {
3962 /*
3963 * Do this before adding a new entry, so the old entry is sure
3964 * to be still pointing to the valid old entry.
3965 */
3966 retval = ext4_setent(handle, &old, whiteout->i_ino,
3967 EXT4_FT_CHRDEV);
3968 if (retval)
3969 goto end_rename;
3970 retval = ext4_mark_inode_dirty(handle, whiteout);
3971 if (unlikely(retval))
3972 goto end_rename;
3973
3974 }
3975 if (!new.bh) {
3976 retval = ext4_add_entry(handle, new.dentry, old.inode);
3977 if (retval)
3978 goto end_rename;
3979 } else {
3980 retval = ext4_setent(handle, &new,
3981 old.inode->i_ino, old_file_type);
3982 if (retval)
3983 goto end_rename;
3984 }
3985 if (force_reread)
3986 force_reread = !ext4_test_inode_flag(new.dir,
3987 EXT4_INODE_INLINE_DATA);
3988
3989 /*
3990 * Like most other Unix systems, set the ctime for inodes on a
3991 * rename.
3992 */
3993 inode_set_ctime_current(old.inode);
3994 retval = ext4_mark_inode_dirty(handle, old.inode);
3995 if (unlikely(retval))
3996 goto end_rename;
3997
3998 if (!whiteout) {
3999 /*
4000 * ok, that's it
4001 */
4002 ext4_rename_delete(handle, &old, force_reread);
4003 }
4004
4005 if (new.inode) {
4006 ext4_dec_count(new.inode);
4007 inode_set_ctime_current(new.inode);
4008 }
4009 old.dir->i_mtime = inode_set_ctime_current(old.dir);
4010 ext4_update_dx_flag(old.dir);
4011 if (old.dir_bh) {
4012 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4013 if (retval)
4014 goto end_rename;
4015
4016 ext4_dec_count(old.dir);
4017 if (new.inode) {
4018 /* checked ext4_empty_dir above, can't have another
4019 * parent, ext4_dec_count() won't work for many-linked
4020 * dirs */
4021 clear_nlink(new.inode);
4022 } else {
4023 ext4_inc_count(new.dir);
4024 ext4_update_dx_flag(new.dir);
4025 retval = ext4_mark_inode_dirty(handle, new.dir);
4026 if (unlikely(retval))
4027 goto end_rename;
4028 }
4029 }
4030 retval = ext4_mark_inode_dirty(handle, old.dir);
4031 if (unlikely(retval))
4032 goto end_rename;
4033
4034 if (S_ISDIR(old.inode->i_mode)) {
4035 /*
4036 * We disable fast commits here that's because the
4037 * replay code is not yet capable of changing dot dot
4038 * dirents in directories.
4039 */
4040 ext4_fc_mark_ineligible(old.inode->i_sb,
4041 EXT4_FC_REASON_RENAME_DIR, handle);
4042 } else {
4043 struct super_block *sb = old.inode->i_sb;
4044
4045 if (new.inode)
4046 ext4_fc_track_unlink(handle, new.dentry);
4047 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4048 !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4049 !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
4050 __ext4_fc_track_link(handle, old.inode, new.dentry);
4051 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4052 if (whiteout)
4053 __ext4_fc_track_create(handle, whiteout,
4054 old.dentry);
4055 }
4056 }
4057
4058 if (new.inode) {
4059 retval = ext4_mark_inode_dirty(handle, new.inode);
4060 if (unlikely(retval))
4061 goto end_rename;
4062 if (!new.inode->i_nlink)
4063 ext4_orphan_add(handle, new.inode);
4064 }
4065 retval = 0;
4066
4067 end_rename:
4068 if (whiteout) {
4069 if (retval) {
4070 ext4_resetent(handle, &old,
4071 old.inode->i_ino, old_file_type);
4072 drop_nlink(whiteout);
4073 ext4_mark_inode_dirty(handle, whiteout);
4074 ext4_orphan_add(handle, whiteout);
4075 }
4076 unlock_new_inode(whiteout);
4077 ext4_journal_stop(handle);
4078 iput(whiteout);
4079 } else {
4080 ext4_journal_stop(handle);
4081 }
4082 release_bh:
4083 brelse(old.dir_bh);
4084 brelse(old.bh);
4085 brelse(new.bh);
4086
4087 return retval;
4088 }
4089
ext4_cross_rename(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry)4090 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4091 struct inode *new_dir, struct dentry *new_dentry)
4092 {
4093 handle_t *handle = NULL;
4094 struct ext4_renament old = {
4095 .dir = old_dir,
4096 .dentry = old_dentry,
4097 .inode = d_inode(old_dentry),
4098 };
4099 struct ext4_renament new = {
4100 .dir = new_dir,
4101 .dentry = new_dentry,
4102 .inode = d_inode(new_dentry),
4103 };
4104 u8 new_file_type;
4105 int retval;
4106
4107 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4108 !projid_eq(EXT4_I(new_dir)->i_projid,
4109 EXT4_I(old_dentry->d_inode)->i_projid)) ||
4110 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4111 !projid_eq(EXT4_I(old_dir)->i_projid,
4112 EXT4_I(new_dentry->d_inode)->i_projid)))
4113 return -EXDEV;
4114
4115 retval = dquot_initialize(old.dir);
4116 if (retval)
4117 return retval;
4118 retval = dquot_initialize(new.dir);
4119 if (retval)
4120 return retval;
4121
4122 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4123 &old.de, &old.inlined);
4124 if (IS_ERR(old.bh))
4125 return PTR_ERR(old.bh);
4126 /*
4127 * Check for inode number is _not_ due to possible IO errors.
4128 * We might rmdir the source, keep it as pwd of some process
4129 * and merrily kill the link to whatever was created under the
4130 * same name. Goodbye sticky bit ;-<
4131 */
4132 retval = -ENOENT;
4133 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4134 goto end_rename;
4135
4136 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4137 &new.de, &new.inlined);
4138 if (IS_ERR(new.bh)) {
4139 retval = PTR_ERR(new.bh);
4140 new.bh = NULL;
4141 goto end_rename;
4142 }
4143
4144 /* RENAME_EXCHANGE case: old *and* new must both exist */
4145 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4146 goto end_rename;
4147
4148 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4149 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4150 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4151 if (IS_ERR(handle)) {
4152 retval = PTR_ERR(handle);
4153 handle = NULL;
4154 goto end_rename;
4155 }
4156
4157 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4158 ext4_handle_sync(handle);
4159
4160 if (S_ISDIR(old.inode->i_mode)) {
4161 old.is_dir = true;
4162 retval = ext4_rename_dir_prepare(handle, &old);
4163 if (retval)
4164 goto end_rename;
4165 }
4166 if (S_ISDIR(new.inode->i_mode)) {
4167 new.is_dir = true;
4168 retval = ext4_rename_dir_prepare(handle, &new);
4169 if (retval)
4170 goto end_rename;
4171 }
4172
4173 /*
4174 * Other than the special case of overwriting a directory, parents'
4175 * nlink only needs to be modified if this is a cross directory rename.
4176 */
4177 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4178 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4179 new.dir_nlink_delta = -old.dir_nlink_delta;
4180 retval = -EMLINK;
4181 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4182 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4183 goto end_rename;
4184 }
4185
4186 new_file_type = new.de->file_type;
4187 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4188 if (retval)
4189 goto end_rename;
4190
4191 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4192 if (retval)
4193 goto end_rename;
4194
4195 /*
4196 * Like most other Unix systems, set the ctime for inodes on a
4197 * rename.
4198 */
4199 inode_set_ctime_current(old.inode);
4200 inode_set_ctime_current(new.inode);
4201 retval = ext4_mark_inode_dirty(handle, old.inode);
4202 if (unlikely(retval))
4203 goto end_rename;
4204 retval = ext4_mark_inode_dirty(handle, new.inode);
4205 if (unlikely(retval))
4206 goto end_rename;
4207 ext4_fc_mark_ineligible(new.inode->i_sb,
4208 EXT4_FC_REASON_CROSS_RENAME, handle);
4209 if (old.dir_bh) {
4210 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4211 if (retval)
4212 goto end_rename;
4213 }
4214 if (new.dir_bh) {
4215 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4216 if (retval)
4217 goto end_rename;
4218 }
4219 ext4_update_dir_count(handle, &old);
4220 ext4_update_dir_count(handle, &new);
4221 retval = 0;
4222
4223 end_rename:
4224 brelse(old.dir_bh);
4225 brelse(new.dir_bh);
4226 brelse(old.bh);
4227 brelse(new.bh);
4228 if (handle)
4229 ext4_journal_stop(handle);
4230 return retval;
4231 }
4232
ext4_rename2(struct mnt_idmap * idmap,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)4233 static int ext4_rename2(struct mnt_idmap *idmap,
4234 struct inode *old_dir, struct dentry *old_dentry,
4235 struct inode *new_dir, struct dentry *new_dentry,
4236 unsigned int flags)
4237 {
4238 int err;
4239
4240 if (unlikely(ext4_forced_shutdown(old_dir->i_sb)))
4241 return -EIO;
4242
4243 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4244 return -EINVAL;
4245
4246 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4247 flags);
4248 if (err)
4249 return err;
4250
4251 if (flags & RENAME_EXCHANGE) {
4252 return ext4_cross_rename(old_dir, old_dentry,
4253 new_dir, new_dentry);
4254 }
4255
4256 return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4257 }
4258
4259 /*
4260 * directories can handle most operations...
4261 */
4262 const struct inode_operations ext4_dir_inode_operations = {
4263 .create = ext4_create,
4264 .lookup = ext4_lookup,
4265 .link = ext4_link,
4266 .unlink = ext4_unlink,
4267 .symlink = ext4_symlink,
4268 .mkdir = ext4_mkdir,
4269 .rmdir = ext4_rmdir,
4270 .mknod = ext4_mknod,
4271 .tmpfile = ext4_tmpfile,
4272 .rename = ext4_rename2,
4273 .setattr = ext4_setattr,
4274 .getattr = ext4_getattr,
4275 .listxattr = ext4_listxattr,
4276 .get_inode_acl = ext4_get_acl,
4277 .set_acl = ext4_set_acl,
4278 .fiemap = ext4_fiemap,
4279 .fileattr_get = ext4_fileattr_get,
4280 .fileattr_set = ext4_fileattr_set,
4281 };
4282
4283 const struct inode_operations ext4_special_inode_operations = {
4284 .setattr = ext4_setattr,
4285 .getattr = ext4_getattr,
4286 .listxattr = ext4_listxattr,
4287 .get_inode_acl = ext4_get_acl,
4288 .set_acl = ext4_set_acl,
4289 };
4290