projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ext4: remove unused header files
[deliverable/linux.git] / fs / ext4 / namei.c
1 /*
2 * linux/fs/ext4/namei.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
24 * Theodore Ts'o, 2002
25 */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
36 #include "ext4.h"
37 #include "ext4_jbd2.h"
38
39 #include "xattr.h"
40 #include "acl.h"
41
42 #include <trace/events/ext4.h>
43 /*
44 * define how far ahead to read directories while searching them.
45 */
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49
50 static struct buffer_head *ext4_append(handle_t *handle,
51 struct inode *inode,
52 ext4_lblk_t *block)
53 {
54 struct buffer_head *bh;
55 int err;
56
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
61
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63
64 bh = ext4_bread(handle, inode, *block, 1);
65 if (IS_ERR(bh))
66 return bh;
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
70 err = ext4_journal_get_write_access(handle, bh);
71 if (err) {
72 brelse(bh);
73 ext4_std_error(inode->i_sb, err);
74 return ERR_PTR(err);
75 }
76 return bh;
77 }
78
79 static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
81
82 typedef enum {
83 EITHER, INDEX, DIRENT
84 } dirblock_type_t;
85
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __LINE__)
88
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 ext4_lblk_t block,
91 dirblock_type_t type,
92 unsigned int line)
93 {
94 struct buffer_head *bh;
95 struct ext4_dir_entry *dirent;
96 int is_dx_block = 0;
97
98 bh = ext4_bread(NULL, inode, block, 0);
99 if (IS_ERR(bh)) {
100 __ext4_warning(inode->i_sb, __func__, line,
101 "error %ld reading directory block "
102 "(ino %lu, block %lu)", PTR_ERR(bh), inode->i_ino,
103 (unsigned long) block);
104
105 return bh;
106 }
107 if (!bh) {
108 ext4_error_inode(inode, __func__, line, block, "Directory hole found");
109 return ERR_PTR(-EIO);
110 }
111 dirent = (struct ext4_dir_entry *) bh->b_data;
112 /* Determine whether or not we have an index block */
113 if (is_dx(inode)) {
114 if (block == 0)
115 is_dx_block = 1;
116 else if (ext4_rec_len_from_disk(dirent->rec_len,
117 inode->i_sb->s_blocksize) ==
118 inode->i_sb->s_blocksize)
119 is_dx_block = 1;
120 }
121 if (!is_dx_block && type == INDEX) {
122 ext4_error_inode(inode, __func__, line, block,
123 "directory leaf block found instead of index block");
124 return ERR_PTR(-EIO);
125 }
126 if (!ext4_has_metadata_csum(inode->i_sb) ||
127 buffer_verified(bh))
128 return bh;
129
130 /*
131 * An empty leaf block can get mistaken for a index block; for
132 * this reason, we can only check the index checksum when the
133 * caller is sure it should be an index block.
134 */
135 if (is_dx_block && type == INDEX) {
136 if (ext4_dx_csum_verify(inode, dirent))
137 set_buffer_verified(bh);
138 else {
139 ext4_error_inode(inode, __func__, line, block,
140 "Directory index failed checksum");
141 brelse(bh);
142 return ERR_PTR(-EIO);
143 }
144 }
145 if (!is_dx_block) {
146 if (ext4_dirent_csum_verify(inode, dirent))
147 set_buffer_verified(bh);
148 else {
149 ext4_error_inode(inode, __func__, line, block,
150 "Directory block failed checksum");
151 brelse(bh);
152 return ERR_PTR(-EIO);
153 }
154 }
155 return bh;
156 }
157
158 #ifndef assert
159 #define assert(test) J_ASSERT(test)
160 #endif
161
162 #ifdef DX_DEBUG
163 #define dxtrace(command) command
164 #else
165 #define dxtrace(command)
166 #endif
167
168 struct fake_dirent
169 {
170 __le32 inode;
171 __le16 rec_len;
172 u8 name_len;
173 u8 file_type;
174 };
175
176 struct dx_countlimit
177 {
178 __le16 limit;
179 __le16 count;
180 };
181
182 struct dx_entry
183 {
184 __le32 hash;
185 __le32 block;
186 };
187
188 /*
189 * dx_root_info is laid out so that if it should somehow get overlaid by a
190 * dirent the two low bits of the hash version will be zero. Therefore, the
191 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
192 */
193
194 struct dx_root
195 {
196 struct fake_dirent dot;
197 char dot_name[4];
198 struct fake_dirent dotdot;
199 char dotdot_name[4];
200 struct dx_root_info
201 {
202 __le32 reserved_zero;
203 u8 hash_version;
204 u8 info_length; /* 8 */
205 u8 indirect_levels;
206 u8 unused_flags;
207 }
208 info;
209 struct dx_entry entries[0];
210 };
211
212 struct dx_node
213 {
214 struct fake_dirent fake;
215 struct dx_entry entries[0];
216 };
217
218
219 struct dx_frame
220 {
221 struct buffer_head *bh;
222 struct dx_entry *entries;
223 struct dx_entry *at;
224 };
225
226 struct dx_map_entry
227 {
228 u32 hash;
229 u16 offs;
230 u16 size;
231 };
232
233 /*
234 * This goes at the end of each htree block.
235 */
236 struct dx_tail {
237 u32 dt_reserved;
238 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
239 };
240
241 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
242 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
243 static inline unsigned dx_get_hash(struct dx_entry *entry);
244 static void dx_set_hash(struct dx_entry *entry, unsigned value);
245 static unsigned dx_get_count(struct dx_entry *entries);
246 static unsigned dx_get_limit(struct dx_entry *entries);
247 static void dx_set_count(struct dx_entry *entries, unsigned value);
248 static void dx_set_limit(struct dx_entry *entries, unsigned value);
249 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
250 static unsigned dx_node_limit(struct inode *dir);
251 static struct dx_frame *dx_probe(const struct qstr *d_name,
252 struct inode *dir,
253 struct dx_hash_info *hinfo,
254 struct dx_frame *frame);
255 static void dx_release(struct dx_frame *frames);
256 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
257 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
258 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
259 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
260 struct dx_map_entry *offsets, int count, unsigned blocksize);
261 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
262 static void dx_insert_block(struct dx_frame *frame,
263 u32 hash, ext4_lblk_t block);
264 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
265 struct dx_frame *frame,
266 struct dx_frame *frames,
267 __u32 *start_hash);
268 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
269 const struct qstr *d_name,
270 struct ext4_dir_entry_2 **res_dir);
271 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
272 struct inode *inode);
273
274 /* checksumming functions */
275 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
276 unsigned int blocksize)
277 {
278 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
279 t->det_rec_len = ext4_rec_len_to_disk(
280 sizeof(struct ext4_dir_entry_tail), blocksize);
281 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
282 }
283
284 /* Walk through a dirent block to find a checksum "dirent" at the tail */
285 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
286 struct ext4_dir_entry *de)
287 {
288 struct ext4_dir_entry_tail *t;
289
290 #ifdef PARANOID
291 struct ext4_dir_entry *d, *top;
292
293 d = de;
294 top = (struct ext4_dir_entry *)(((void *)de) +
295 (EXT4_BLOCK_SIZE(inode->i_sb) -
296 sizeof(struct ext4_dir_entry_tail)));
297 while (d < top && d->rec_len)
298 d = (struct ext4_dir_entry *)(((void *)d) +
299 le16_to_cpu(d->rec_len));
300
301 if (d != top)
302 return NULL;
303
304 t = (struct ext4_dir_entry_tail *)d;
305 #else
306 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
307 #endif
308
309 if (t->det_reserved_zero1 ||
310 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
311 t->det_reserved_zero2 ||
312 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
313 return NULL;
314
315 return t;
316 }
317
318 static __le32 ext4_dirent_csum(struct inode *inode,
319 struct ext4_dir_entry *dirent, int size)
320 {
321 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
322 struct ext4_inode_info *ei = EXT4_I(inode);
323 __u32 csum;
324
325 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
326 return cpu_to_le32(csum);
327 }
328
329 static void warn_no_space_for_csum(struct inode *inode)
330 {
331 ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
332 "checksum. Please run e2fsck -D.", inode->i_ino);
333 }
334
335 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
336 {
337 struct ext4_dir_entry_tail *t;
338
339 if (!ext4_has_metadata_csum(inode->i_sb))
340 return 1;
341
342 t = get_dirent_tail(inode, dirent);
343 if (!t) {
344 warn_no_space_for_csum(inode);
345 return 0;
346 }
347
348 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
349 (void *)t - (void *)dirent))
350 return 0;
351
352 return 1;
353 }
354
355 static void ext4_dirent_csum_set(struct inode *inode,
356 struct ext4_dir_entry *dirent)
357 {
358 struct ext4_dir_entry_tail *t;
359
360 if (!ext4_has_metadata_csum(inode->i_sb))
361 return;
362
363 t = get_dirent_tail(inode, dirent);
364 if (!t) {
365 warn_no_space_for_csum(inode);
366 return;
367 }
368
369 t->det_checksum = ext4_dirent_csum(inode, dirent,
370 (void *)t - (void *)dirent);
371 }
372
373 int ext4_handle_dirty_dirent_node(handle_t *handle,
374 struct inode *inode,
375 struct buffer_head *bh)
376 {
377 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
378 return ext4_handle_dirty_metadata(handle, inode, bh);
379 }
380
381 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
382 struct ext4_dir_entry *dirent,
383 int *offset)
384 {
385 struct ext4_dir_entry *dp;
386 struct dx_root_info *root;
387 int count_offset;
388
389 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
390 count_offset = 8;
391 else if (le16_to_cpu(dirent->rec_len) == 12) {
392 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
393 if (le16_to_cpu(dp->rec_len) !=
394 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
395 return NULL;
396 root = (struct dx_root_info *)(((void *)dp + 12));
397 if (root->reserved_zero ||
398 root->info_length != sizeof(struct dx_root_info))
399 return NULL;
400 count_offset = 32;
401 } else
402 return NULL;
403
404 if (offset)
405 *offset = count_offset;
406 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
407 }
408
409 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
410 int count_offset, int count, struct dx_tail *t)
411 {
412 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
413 struct ext4_inode_info *ei = EXT4_I(inode);
414 __u32 csum;
415 __le32 save_csum;
416 int size;
417
418 size = count_offset + (count * sizeof(struct dx_entry));
419 save_csum = t->dt_checksum;
420 t->dt_checksum = 0;
421 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
422 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
423 t->dt_checksum = save_csum;
424
425 return cpu_to_le32(csum);
426 }
427
428 static int ext4_dx_csum_verify(struct inode *inode,
429 struct ext4_dir_entry *dirent)
430 {
431 struct dx_countlimit *c;
432 struct dx_tail *t;
433 int count_offset, limit, count;
434
435 if (!ext4_has_metadata_csum(inode->i_sb))
436 return 1;
437
438 c = get_dx_countlimit(inode, dirent, &count_offset);
439 if (!c) {
440 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
441 return 1;
442 }
443 limit = le16_to_cpu(c->limit);
444 count = le16_to_cpu(c->count);
445 if (count_offset + (limit * sizeof(struct dx_entry)) >
446 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
447 warn_no_space_for_csum(inode);
448 return 1;
449 }
450 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
451
452 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
453 count, t))
454 return 0;
455 return 1;
456 }
457
458 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
459 {
460 struct dx_countlimit *c;
461 struct dx_tail *t;
462 int count_offset, limit, count;
463
464 if (!ext4_has_metadata_csum(inode->i_sb))
465 return;
466
467 c = get_dx_countlimit(inode, dirent, &count_offset);
468 if (!c) {
469 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
470 return;
471 }
472 limit = le16_to_cpu(c->limit);
473 count = le16_to_cpu(c->count);
474 if (count_offset + (limit * sizeof(struct dx_entry)) >
475 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
476 warn_no_space_for_csum(inode);
477 return;
478 }
479 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
480
481 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
482 }
483
484 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
485 struct inode *inode,
486 struct buffer_head *bh)
487 {
488 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
489 return ext4_handle_dirty_metadata(handle, inode, bh);
490 }
491
492 /*
493 * p is at least 6 bytes before the end of page
494 */
495 static inline struct ext4_dir_entry_2 *
496 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
497 {
498 return (struct ext4_dir_entry_2 *)((char *)p +
499 ext4_rec_len_from_disk(p->rec_len, blocksize));
500 }
501
502 /*
503 * Future: use high four bits of block for coalesce-on-delete flags
504 * Mask them off for now.
505 */
506
507 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
508 {
509 return le32_to_cpu(entry->block) & 0x00ffffff;
510 }
511
512 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
513 {
514 entry->block = cpu_to_le32(value);
515 }
516
517 static inline unsigned dx_get_hash(struct dx_entry *entry)
518 {
519 return le32_to_cpu(entry->hash);
520 }
521
522 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
523 {
524 entry->hash = cpu_to_le32(value);
525 }
526
527 static inline unsigned dx_get_count(struct dx_entry *entries)
528 {
529 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
530 }
531
532 static inline unsigned dx_get_limit(struct dx_entry *entries)
533 {
534 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
535 }
536
537 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
538 {
539 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
540 }
541
542 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
543 {
544 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
545 }
546
547 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
548 {
549 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
550 EXT4_DIR_REC_LEN(2) - infosize;
551
552 if (ext4_has_metadata_csum(dir->i_sb))
553 entry_space -= sizeof(struct dx_tail);
554 return entry_space / sizeof(struct dx_entry);
555 }
556
557 static inline unsigned dx_node_limit(struct inode *dir)
558 {
559 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
560
561 if (ext4_has_metadata_csum(dir->i_sb))
562 entry_space -= sizeof(struct dx_tail);
563 return entry_space / sizeof(struct dx_entry);
564 }
565
566 /*
567 * Debug
568 */
569 #ifdef DX_DEBUG
570 static void dx_show_index(char * label, struct dx_entry *entries)
571 {
572 int i, n = dx_get_count (entries);
573 printk(KERN_DEBUG "%s index ", label);
574 for (i = 0; i < n; i++) {
575 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
576 0, (unsigned long)dx_get_block(entries + i));
577 }
578 printk("\n");
579 }
580
581 struct stats
582 {
583 unsigned names;
584 unsigned space;
585 unsigned bcount;
586 };
587
588 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
589 int size, int show_names)
590 {
591 unsigned names = 0, space = 0;
592 char *base = (char *) de;
593 struct dx_hash_info h = *hinfo;
594
595 printk("names: ");
596 while ((char *) de < base + size)
597 {
598 if (de->inode)
599 {
600 if (show_names)
601 {
602 int len = de->name_len;
603 char *name = de->name;
604 while (len--) printk("%c", *name++);
605 ext4fs_dirhash(de->name, de->name_len, &h);
606 printk(":%x.%u ", h.hash,
607 (unsigned) ((char *) de - base));
608 }
609 space += EXT4_DIR_REC_LEN(de->name_len);
610 names++;
611 }
612 de = ext4_next_entry(de, size);
613 }
614 printk("(%i)\n", names);
615 return (struct stats) { names, space, 1 };
616 }
617
618 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
619 struct dx_entry *entries, int levels)
620 {
621 unsigned blocksize = dir->i_sb->s_blocksize;
622 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
623 unsigned bcount = 0;
624 struct buffer_head *bh;
625 int err;
626 printk("%i indexed blocks...\n", count);
627 for (i = 0; i < count; i++, entries++)
628 {
629 ext4_lblk_t block = dx_get_block(entries);
630 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
631 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
632 struct stats stats;
633 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
634 bh = ext4_bread(NULL,dir, block, 0);
635 if (!bh || IS_ERR(bh))
636 continue;
637 stats = levels?
638 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
639 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
640 names += stats.names;
641 space += stats.space;
642 bcount += stats.bcount;
643 brelse(bh);
644 }
645 if (bcount)
646 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
647 levels ? "" : " ", names, space/bcount,
648 (space/bcount)*100/blocksize);
649 return (struct stats) { names, space, bcount};
650 }
651 #endif /* DX_DEBUG */
652
653 /*
654 * Probe for a directory leaf block to search.
655 *
656 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
657 * error in the directory index, and the caller should fall back to
658 * searching the directory normally. The callers of dx_probe **MUST**
659 * check for this error code, and make sure it never gets reflected
660 * back to userspace.
661 */
662 static struct dx_frame *
663 dx_probe(const struct qstr *d_name, struct inode *dir,
664 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
665 {
666 unsigned count, indirect;
667 struct dx_entry *at, *entries, *p, *q, *m;
668 struct dx_root *root;
669 struct dx_frame *frame = frame_in;
670 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
671 u32 hash;
672
673 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
674 if (IS_ERR(frame->bh))
675 return (struct dx_frame *) frame->bh;
676
677 root = (struct dx_root *) frame->bh->b_data;
678 if (root->info.hash_version != DX_HASH_TEA &&
679 root->info.hash_version != DX_HASH_HALF_MD4 &&
680 root->info.hash_version != DX_HASH_LEGACY) {
681 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
682 root->info.hash_version);
683 goto fail;
684 }
685 hinfo->hash_version = root->info.hash_version;
686 if (hinfo->hash_version <= DX_HASH_TEA)
687 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
688 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
689 if (d_name)
690 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
691 hash = hinfo->hash;
692
693 if (root->info.unused_flags & 1) {
694 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
695 root->info.unused_flags);
696 goto fail;
697 }
698
699 if ((indirect = root->info.indirect_levels) > 1) {
700 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
701 root->info.indirect_levels);
702 goto fail;
703 }
704
705 entries = (struct dx_entry *) (((char *)&root->info) +
706 root->info.info_length);
707
708 if (dx_get_limit(entries) != dx_root_limit(dir,
709 root->info.info_length)) {
710 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
711 goto fail;
712 }
713
714 dxtrace(printk("Look up %x", hash));
715 while (1) {
716 count = dx_get_count(entries);
717 if (!count || count > dx_get_limit(entries)) {
718 ext4_warning(dir->i_sb,
719 "dx entry: no count or count > limit");
720 goto fail;
721 }
722
723 p = entries + 1;
724 q = entries + count - 1;
725 while (p <= q) {
726 m = p + (q - p)/2;
727 dxtrace(printk("."));
728 if (dx_get_hash(m) > hash)
729 q = m - 1;
730 else
731 p = m + 1;
732 }
733
734 if (0) { // linear search cross check
735 unsigned n = count - 1;
736 at = entries;
737 while (n--)
738 {
739 dxtrace(printk(","));
740 if (dx_get_hash(++at) > hash)
741 {
742 at--;
743 break;
744 }
745 }
746 assert (at == p - 1);
747 }
748
749 at = p - 1;
750 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
751 frame->entries = entries;
752 frame->at = at;
753 if (!indirect--)
754 return frame;
755 frame++;
756 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
757 if (IS_ERR(frame->bh)) {
758 ret_err = (struct dx_frame *) frame->bh;
759 frame->bh = NULL;
760 goto fail;
761 }
762 entries = ((struct dx_node *) frame->bh->b_data)->entries;
763
764 if (dx_get_limit(entries) != dx_node_limit (dir)) {
765 ext4_warning(dir->i_sb,
766 "dx entry: limit != node limit");
767 goto fail;
768 }
769 }
770 fail:
771 while (frame >= frame_in) {
772 brelse(frame->bh);
773 frame--;
774 }
775 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
776 ext4_warning(dir->i_sb,
777 "Corrupt dir inode %lu, running e2fsck is "
778 "recommended.", dir->i_ino);
779 return ret_err;
780 }
781
782 static void dx_release (struct dx_frame *frames)
783 {
784 if (frames[0].bh == NULL)
785 return;
786
787 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
788 brelse(frames[1].bh);
789 brelse(frames[0].bh);
790 }
791
792 /*
793 * This function increments the frame pointer to search the next leaf
794 * block, and reads in the necessary intervening nodes if the search
795 * should be necessary. Whether or not the search is necessary is
796 * controlled by the hash parameter. If the hash value is even, then
797 * the search is only continued if the next block starts with that
798 * hash value. This is used if we are searching for a specific file.
799 *
800 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
801 *
802 * This function returns 1 if the caller should continue to search,
803 * or 0 if it should not. If there is an error reading one of the
804 * index blocks, it will a negative error code.
805 *
806 * If start_hash is non-null, it will be filled in with the starting
807 * hash of the next page.
808 */
809 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
810 struct dx_frame *frame,
811 struct dx_frame *frames,
812 __u32 *start_hash)
813 {
814 struct dx_frame *p;
815 struct buffer_head *bh;
816 int num_frames = 0;
817 __u32 bhash;
818
819 p = frame;
820 /*
821 * Find the next leaf page by incrementing the frame pointer.
822 * If we run out of entries in the interior node, loop around and
823 * increment pointer in the parent node. When we break out of
824 * this loop, num_frames indicates the number of interior
825 * nodes need to be read.
826 */
827 while (1) {
828 if (++(p->at) < p->entries + dx_get_count(p->entries))
829 break;
830 if (p == frames)
831 return 0;
832 num_frames++;
833 p--;
834 }
835
836 /*
837 * If the hash is 1, then continue only if the next page has a
838 * continuation hash of any value. This is used for readdir
839 * handling. Otherwise, check to see if the hash matches the
840 * desired contiuation hash. If it doesn't, return since
841 * there's no point to read in the successive index pages.
842 */
843 bhash = dx_get_hash(p->at);
844 if (start_hash)
845 *start_hash = bhash;
846 if ((hash & 1) == 0) {
847 if ((bhash & ~1) != hash)
848 return 0;
849 }
850 /*
851 * If the hash is HASH_NB_ALWAYS, we always go to the next
852 * block so no check is necessary
853 */
854 while (num_frames--) {
855 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
856 if (IS_ERR(bh))
857 return PTR_ERR(bh);
858 p++;
859 brelse(p->bh);
860 p->bh = bh;
861 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
862 }
863 return 1;
864 }
865
866
867 /*
868 * This function fills a red-black tree with information from a
869 * directory block. It returns the number directory entries loaded
870 * into the tree. If there is an error it is returned in err.
871 */
872 static int htree_dirblock_to_tree(struct file *dir_file,
873 struct inode *dir, ext4_lblk_t block,
874 struct dx_hash_info *hinfo,
875 __u32 start_hash, __u32 start_minor_hash)
876 {
877 struct buffer_head *bh;
878 struct ext4_dir_entry_2 *de, *top;
879 int err = 0, count = 0;
880
881 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
882 (unsigned long)block));
883 bh = ext4_read_dirblock(dir, block, DIRENT);
884 if (IS_ERR(bh))
885 return PTR_ERR(bh);
886
887 de = (struct ext4_dir_entry_2 *) bh->b_data;
888 top = (struct ext4_dir_entry_2 *) ((char *) de +
889 dir->i_sb->s_blocksize -
890 EXT4_DIR_REC_LEN(0));
891 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
892 if (ext4_check_dir_entry(dir, NULL, de, bh,
893 bh->b_data, bh->b_size,
894 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
895 + ((char *)de - bh->b_data))) {
896 /* silently ignore the rest of the block */
897 break;
898 }
899 ext4fs_dirhash(de->name, de->name_len, hinfo);
900 if ((hinfo->hash < start_hash) ||
901 ((hinfo->hash == start_hash) &&
902 (hinfo->minor_hash < start_minor_hash)))
903 continue;
904 if (de->inode == 0)
905 continue;
906 if ((err = ext4_htree_store_dirent(dir_file,
907 hinfo->hash, hinfo->minor_hash, de)) != 0) {
908 brelse(bh);
909 return err;
910 }
911 count++;
912 }
913 brelse(bh);
914 return count;
915 }
916
917
918 /*
919 * This function fills a red-black tree with information from a
920 * directory. We start scanning the directory in hash order, starting
921 * at start_hash and start_minor_hash.
922 *
923 * This function returns the number of entries inserted into the tree,
924 * or a negative error code.
925 */
926 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
927 __u32 start_minor_hash, __u32 *next_hash)
928 {
929 struct dx_hash_info hinfo;
930 struct ext4_dir_entry_2 *de;
931 struct dx_frame frames[2], *frame;
932 struct inode *dir;
933 ext4_lblk_t block;
934 int count = 0;
935 int ret, err;
936 __u32 hashval;
937
938 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
939 start_hash, start_minor_hash));
940 dir = file_inode(dir_file);
941 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
942 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
943 if (hinfo.hash_version <= DX_HASH_TEA)
944 hinfo.hash_version +=
945 EXT4_SB(dir->i_sb)->s_hash_unsigned;
946 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
947 if (ext4_has_inline_data(dir)) {
948 int has_inline_data = 1;
949 count = htree_inlinedir_to_tree(dir_file, dir, 0,
950 &hinfo, start_hash,
951 start_minor_hash,
952 &has_inline_data);
953 if (has_inline_data) {
954 *next_hash = ~0;
955 return count;
956 }
957 }
958 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
959 start_hash, start_minor_hash);
960 *next_hash = ~0;
961 return count;
962 }
963 hinfo.hash = start_hash;
964 hinfo.minor_hash = 0;
965 frame = dx_probe(NULL, dir, &hinfo, frames);
966 if (IS_ERR(frame))
967 return PTR_ERR(frame);
968
969 /* Add '.' and '..' from the htree header */
970 if (!start_hash && !start_minor_hash) {
971 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
972 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
973 goto errout;
974 count++;
975 }
976 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
977 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
978 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
979 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
980 goto errout;
981 count++;
982 }
983
984 while (1) {
985 block = dx_get_block(frame->at);
986 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
987 start_hash, start_minor_hash);
988 if (ret < 0) {
989 err = ret;
990 goto errout;
991 }
992 count += ret;
993 hashval = ~0;
994 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
995 frame, frames, &hashval);
996 *next_hash = hashval;
997 if (ret < 0) {
998 err = ret;
999 goto errout;
1000 }
1001 /*
1002 * Stop if: (a) there are no more entries, or
1003 * (b) we have inserted at least one entry and the
1004 * next hash value is not a continuation
1005 */
1006 if ((ret == 0) ||
1007 (count && ((hashval & 1) == 0)))
1008 break;
1009 }
1010 dx_release(frames);
1011 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1012 "next hash: %x\n", count, *next_hash));
1013 return count;
1014 errout:
1015 dx_release(frames);
1016 return (err);
1017 }
1018
1019 static inline int search_dirblock(struct buffer_head *bh,
1020 struct inode *dir,
1021 const struct qstr *d_name,
1022 unsigned int offset,
1023 struct ext4_dir_entry_2 **res_dir)
1024 {
1025 return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1026 d_name, offset, res_dir);
1027 }
1028
1029 /*
1030 * Directory block splitting, compacting
1031 */
1032
1033 /*
1034 * Create map of hash values, offsets, and sizes, stored at end of block.
1035 * Returns number of entries mapped.
1036 */
1037 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1038 struct dx_hash_info *hinfo,
1039 struct dx_map_entry *map_tail)
1040 {
1041 int count = 0;
1042 char *base = (char *) de;
1043 struct dx_hash_info h = *hinfo;
1044
1045 while ((char *) de < base + blocksize) {
1046 if (de->name_len && de->inode) {
1047 ext4fs_dirhash(de->name, de->name_len, &h);
1048 map_tail--;
1049 map_tail->hash = h.hash;
1050 map_tail->offs = ((char *) de - base)>>2;
1051 map_tail->size = le16_to_cpu(de->rec_len);
1052 count++;
1053 cond_resched();
1054 }
1055 /* XXX: do we need to check rec_len == 0 case? -Chris */
1056 de = ext4_next_entry(de, blocksize);
1057 }
1058 return count;
1059 }
1060
1061 /* Sort map by hash value */
1062 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1063 {
1064 struct dx_map_entry *p, *q, *top = map + count - 1;
1065 int more;
1066 /* Combsort until bubble sort doesn't suck */
1067 while (count > 2) {
1068 count = count*10/13;
1069 if (count - 9 < 2) /* 9, 10 -> 11 */
1070 count = 11;
1071 for (p = top, q = p - count; q >= map; p--, q--)
1072 if (p->hash < q->hash)
1073 swap(*p, *q);
1074 }
1075 /* Garden variety bubble sort */
1076 do {
1077 more = 0;
1078 q = top;
1079 while (q-- > map) {
1080 if (q[1].hash >= q[0].hash)
1081 continue;
1082 swap(*(q+1), *q);
1083 more = 1;
1084 }
1085 } while(more);
1086 }
1087
1088 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1089 {
1090 struct dx_entry *entries = frame->entries;
1091 struct dx_entry *old = frame->at, *new = old + 1;
1092 int count = dx_get_count(entries);
1093
1094 assert(count < dx_get_limit(entries));
1095 assert(old < entries + count);
1096 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1097 dx_set_hash(new, hash);
1098 dx_set_block(new, block);
1099 dx_set_count(entries, count + 1);
1100 }
1101
1102 /*
1103 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1104 *
1105 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1106 * `de != NULL' is guaranteed by caller.
1107 */
1108 static inline int ext4_match (int len, const char * const name,
1109 struct ext4_dir_entry_2 * de)
1110 {
1111 if (len != de->name_len)
1112 return 0;
1113 if (!de->inode)
1114 return 0;
1115 return !memcmp(name, de->name, len);
1116 }
1117
1118 /*
1119 * Returns 0 if not found, -1 on failure, and 1 on success
1120 */
1121 int search_dir(struct buffer_head *bh,
1122 char *search_buf,
1123 int buf_size,
1124 struct inode *dir,
1125 const struct qstr *d_name,
1126 unsigned int offset,
1127 struct ext4_dir_entry_2 **res_dir)
1128 {
1129 struct ext4_dir_entry_2 * de;
1130 char * dlimit;
1131 int de_len;
1132 const char *name = d_name->name;
1133 int namelen = d_name->len;
1134
1135 de = (struct ext4_dir_entry_2 *)search_buf;
1136 dlimit = search_buf + buf_size;
1137 while ((char *) de < dlimit) {
1138 /* this code is executed quadratically often */
1139 /* do minimal checking `by hand' */
1140
1141 if ((char *) de + namelen <= dlimit &&
1142 ext4_match (namelen, name, de)) {
1143 /* found a match - just to be sure, do a full check */
1144 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1145 bh->b_size, offset))
1146 return -1;
1147 *res_dir = de;
1148 return 1;
1149 }
1150 /* prevent looping on a bad block */
1151 de_len = ext4_rec_len_from_disk(de->rec_len,
1152 dir->i_sb->s_blocksize);
1153 if (de_len <= 0)
1154 return -1;
1155 offset += de_len;
1156 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1157 }
1158 return 0;
1159 }
1160
1161 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1162 struct ext4_dir_entry *de)
1163 {
1164 struct super_block *sb = dir->i_sb;
1165
1166 if (!is_dx(dir))
1167 return 0;
1168 if (block == 0)
1169 return 1;
1170 if (de->inode == 0 &&
1171 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1172 sb->s_blocksize)
1173 return 1;
1174 return 0;
1175 }
1176
1177 /*
1178 * ext4_find_entry()
1179 *
1180 * finds an entry in the specified directory with the wanted name. It
1181 * returns the cache buffer in which the entry was found, and the entry
1182 * itself (as a parameter - res_dir). It does NOT read the inode of the
1183 * entry - you'll have to do that yourself if you want to.
1184 *
1185 * The returned buffer_head has ->b_count elevated. The caller is expected
1186 * to brelse() it when appropriate.
1187 */
1188 static struct buffer_head * ext4_find_entry (struct inode *dir,
1189 const struct qstr *d_name,
1190 struct ext4_dir_entry_2 **res_dir,
1191 int *inlined)
1192 {
1193 struct super_block *sb;
1194 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1195 struct buffer_head *bh, *ret = NULL;
1196 ext4_lblk_t start, block, b;
1197 const u8 *name = d_name->name;
1198 int ra_max = 0; /* Number of bh's in the readahead
1199 buffer, bh_use[] */
1200 int ra_ptr = 0; /* Current index into readahead
1201 buffer */
1202 int num = 0;
1203 ext4_lblk_t nblocks;
1204 int i, namelen;
1205
1206 *res_dir = NULL;
1207 sb = dir->i_sb;
1208 namelen = d_name->len;
1209 if (namelen > EXT4_NAME_LEN)
1210 return NULL;
1211
1212 if (ext4_has_inline_data(dir)) {
1213 int has_inline_data = 1;
1214 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1215 &has_inline_data);
1216 if (has_inline_data) {
1217 if (inlined)
1218 *inlined = 1;
1219 return ret;
1220 }
1221 }
1222
1223 if ((namelen <= 2) && (name[0] == '.') &&
1224 (name[1] == '.' || name[1] == '\0')) {
1225 /*
1226 * "." or ".." will only be in the first block
1227 * NFS may look up ".."; "." should be handled by the VFS
1228 */
1229 block = start = 0;
1230 nblocks = 1;
1231 goto restart;
1232 }
1233 if (is_dx(dir)) {
1234 bh = ext4_dx_find_entry(dir, d_name, res_dir);
1235 /*
1236 * On success, or if the error was file not found,
1237 * return. Otherwise, fall back to doing a search the
1238 * old fashioned way.
1239 */
1240 if (!IS_ERR(bh) || PTR_ERR(bh) != ERR_BAD_DX_DIR)
1241 return bh;
1242 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1243 "falling back\n"));
1244 }
1245 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1246 start = EXT4_I(dir)->i_dir_start_lookup;
1247 if (start >= nblocks)
1248 start = 0;
1249 block = start;
1250 restart:
1251 do {
1252 /*
1253 * We deal with the read-ahead logic here.
1254 */
1255 if (ra_ptr >= ra_max) {
1256 /* Refill the readahead buffer */
1257 ra_ptr = 0;
1258 b = block;
1259 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1260 /*
1261 * Terminate if we reach the end of the
1262 * directory and must wrap, or if our
1263 * search has finished at this block.
1264 */
1265 if (b >= nblocks || (num && block == start)) {
1266 bh_use[ra_max] = NULL;
1267 break;
1268 }
1269 num++;
1270 bh = ext4_getblk(NULL, dir, b++, 0);
1271 if (unlikely(IS_ERR(bh))) {
1272 if (ra_max == 0)
1273 return bh;
1274 break;
1275 }
1276 bh_use[ra_max] = bh;
1277 if (bh)
1278 ll_rw_block(READ | REQ_META | REQ_PRIO,
1279 1, &bh);
1280 }
1281 }
1282 if ((bh = bh_use[ra_ptr++]) == NULL)
1283 goto next;
1284 wait_on_buffer(bh);
1285 if (!buffer_uptodate(bh)) {
1286 /* read error, skip block & hope for the best */
1287 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1288 (unsigned long) block);
1289 brelse(bh);
1290 goto next;
1291 }
1292 if (!buffer_verified(bh) &&
1293 !is_dx_internal_node(dir, block,
1294 (struct ext4_dir_entry *)bh->b_data) &&
1295 !ext4_dirent_csum_verify(dir,
1296 (struct ext4_dir_entry *)bh->b_data)) {
1297 EXT4_ERROR_INODE(dir, "checksumming directory "
1298 "block %lu", (unsigned long)block);
1299 brelse(bh);
1300 goto next;
1301 }
1302 set_buffer_verified(bh);
1303 i = search_dirblock(bh, dir, d_name,
1304 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1305 if (i == 1) {
1306 EXT4_I(dir)->i_dir_start_lookup = block;
1307 ret = bh;
1308 goto cleanup_and_exit;
1309 } else {
1310 brelse(bh);
1311 if (i < 0)
1312 goto cleanup_and_exit;
1313 }
1314 next:
1315 if (++block >= nblocks)
1316 block = 0;
1317 } while (block != start);
1318
1319 /*
1320 * If the directory has grown while we were searching, then
1321 * search the last part of the directory before giving up.
1322 */
1323 block = nblocks;
1324 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1325 if (block < nblocks) {
1326 start = 0;
1327 goto restart;
1328 }
1329
1330 cleanup_and_exit:
1331 /* Clean up the read-ahead blocks */
1332 for (; ra_ptr < ra_max; ra_ptr++)
1333 brelse(bh_use[ra_ptr]);
1334 return ret;
1335 }
1336
1337 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1338 struct ext4_dir_entry_2 **res_dir)
1339 {
1340 struct super_block * sb = dir->i_sb;
1341 struct dx_hash_info hinfo;
1342 struct dx_frame frames[2], *frame;
1343 struct buffer_head *bh;
1344 ext4_lblk_t block;
1345 int retval;
1346
1347 frame = dx_probe(d_name, dir, &hinfo, frames);
1348 if (IS_ERR(frame))
1349 return (struct buffer_head *) frame;
1350 do {
1351 block = dx_get_block(frame->at);
1352 bh = ext4_read_dirblock(dir, block, DIRENT);
1353 if (IS_ERR(bh))
1354 goto errout;
1355
1356 retval = search_dirblock(bh, dir, d_name,
1357 block << EXT4_BLOCK_SIZE_BITS(sb),
1358 res_dir);
1359 if (retval == 1)
1360 goto success;
1361 brelse(bh);
1362 if (retval == -1) {
1363 bh = ERR_PTR(ERR_BAD_DX_DIR);
1364 goto errout;
1365 }
1366
1367 /* Check to see if we should continue to search */
1368 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1369 frames, NULL);
1370 if (retval < 0) {
1371 ext4_warning(sb,
1372 "error %d reading index page in directory #%lu",
1373 retval, dir->i_ino);
1374 bh = ERR_PTR(retval);
1375 goto errout;
1376 }
1377 } while (retval == 1);
1378
1379 bh = NULL;
1380 errout:
1381 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1382 success:
1383 dx_release(frames);
1384 return bh;
1385 }
1386
1387 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1388 {
1389 struct inode *inode;
1390 struct ext4_dir_entry_2 *de;
1391 struct buffer_head *bh;
1392
1393 if (dentry->d_name.len > EXT4_NAME_LEN)
1394 return ERR_PTR(-ENAMETOOLONG);
1395
1396 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1397 if (IS_ERR(bh))
1398 return (struct dentry *) bh;
1399 inode = NULL;
1400 if (bh) {
1401 __u32 ino = le32_to_cpu(de->inode);
1402 brelse(bh);
1403 if (!ext4_valid_inum(dir->i_sb, ino)) {
1404 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1405 return ERR_PTR(-EIO);
1406 }
1407 if (unlikely(ino == dir->i_ino)) {
1408 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1409 dentry);
1410 return ERR_PTR(-EIO);
1411 }
1412 inode = ext4_iget_normal(dir->i_sb, ino);
1413 if (inode == ERR_PTR(-ESTALE)) {
1414 EXT4_ERROR_INODE(dir,
1415 "deleted inode referenced: %u",
1416 ino);
1417 return ERR_PTR(-EIO);
1418 }
1419 }
1420 return d_splice_alias(inode, dentry);
1421 }
1422
1423
1424 struct dentry *ext4_get_parent(struct dentry *child)
1425 {
1426 __u32 ino;
1427 static const struct qstr dotdot = QSTR_INIT("..", 2);
1428 struct ext4_dir_entry_2 * de;
1429 struct buffer_head *bh;
1430
1431 bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1432 if (IS_ERR(bh))
1433 return (struct dentry *) bh;
1434 if (!bh)
1435 return ERR_PTR(-ENOENT);
1436 ino = le32_to_cpu(de->inode);
1437 brelse(bh);
1438
1439 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1440 EXT4_ERROR_INODE(child->d_inode,
1441 "bad parent inode number: %u", ino);
1442 return ERR_PTR(-EIO);
1443 }
1444
1445 return d_obtain_alias(ext4_iget_normal(child->d_inode->i_sb, ino));
1446 }
1447
1448 /*
1449 * Move count entries from end of map between two memory locations.
1450 * Returns pointer to last entry moved.
1451 */
1452 static struct ext4_dir_entry_2 *
1453 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1454 unsigned blocksize)
1455 {
1456 unsigned rec_len = 0;
1457
1458 while (count--) {
1459 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1460 (from + (map->offs<<2));
1461 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1462 memcpy (to, de, rec_len);
1463 ((struct ext4_dir_entry_2 *) to)->rec_len =
1464 ext4_rec_len_to_disk(rec_len, blocksize);
1465 de->inode = 0;
1466 map++;
1467 to += rec_len;
1468 }
1469 return (struct ext4_dir_entry_2 *) (to - rec_len);
1470 }
1471
1472 /*
1473 * Compact each dir entry in the range to the minimal rec_len.
1474 * Returns pointer to last entry in range.
1475 */
1476 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1477 {
1478 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1479 unsigned rec_len = 0;
1480
1481 prev = to = de;
1482 while ((char*)de < base + blocksize) {
1483 next = ext4_next_entry(de, blocksize);
1484 if (de->inode && de->name_len) {
1485 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1486 if (de > to)
1487 memmove(to, de, rec_len);
1488 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1489 prev = to;
1490 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1491 }
1492 de = next;
1493 }
1494 return prev;
1495 }
1496
1497 /*
1498 * Split a full leaf block to make room for a new dir entry.
1499 * Allocate a new block, and move entries so that they are approx. equally full.
1500 * Returns pointer to de in block into which the new entry will be inserted.
1501 */
1502 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1503 struct buffer_head **bh,struct dx_frame *frame,
1504 struct dx_hash_info *hinfo)
1505 {
1506 unsigned blocksize = dir->i_sb->s_blocksize;
1507 unsigned count, continued;
1508 struct buffer_head *bh2;
1509 ext4_lblk_t newblock;
1510 u32 hash2;
1511 struct dx_map_entry *map;
1512 char *data1 = (*bh)->b_data, *data2;
1513 unsigned split, move, size;
1514 struct ext4_dir_entry_2 *de = NULL, *de2;
1515 struct ext4_dir_entry_tail *t;
1516 int csum_size = 0;
1517 int err = 0, i;
1518
1519 if (ext4_has_metadata_csum(dir->i_sb))
1520 csum_size = sizeof(struct ext4_dir_entry_tail);
1521
1522 bh2 = ext4_append(handle, dir, &newblock);
1523 if (IS_ERR(bh2)) {
1524 brelse(*bh);
1525 *bh = NULL;
1526 return (struct ext4_dir_entry_2 *) bh2;
1527 }
1528
1529 BUFFER_TRACE(*bh, "get_write_access");
1530 err = ext4_journal_get_write_access(handle, *bh);
1531 if (err)
1532 goto journal_error;
1533
1534 BUFFER_TRACE(frame->bh, "get_write_access");
1535 err = ext4_journal_get_write_access(handle, frame->bh);
1536 if (err)
1537 goto journal_error;
1538
1539 data2 = bh2->b_data;
1540
1541 /* create map in the end of data2 block */
1542 map = (struct dx_map_entry *) (data2 + blocksize);
1543 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1544 blocksize, hinfo, map);
1545 map -= count;
1546 dx_sort_map(map, count);
1547 /* Split the existing block in the middle, size-wise */
1548 size = 0;
1549 move = 0;
1550 for (i = count-1; i >= 0; i--) {
1551 /* is more than half of this entry in 2nd half of the block? */
1552 if (size + map[i].size/2 > blocksize/2)
1553 break;
1554 size += map[i].size;
1555 move++;
1556 }
1557 /* map index at which we will split */
1558 split = count - move;
1559 hash2 = map[split].hash;
1560 continued = hash2 == map[split - 1].hash;
1561 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1562 (unsigned long)dx_get_block(frame->at),
1563 hash2, split, count-split));
1564
1565 /* Fancy dance to stay within two buffers */
1566 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1567 de = dx_pack_dirents(data1, blocksize);
1568 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1569 (char *) de,
1570 blocksize);
1571 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1572 (char *) de2,
1573 blocksize);
1574 if (csum_size) {
1575 t = EXT4_DIRENT_TAIL(data2, blocksize);
1576 initialize_dirent_tail(t, blocksize);
1577
1578 t = EXT4_DIRENT_TAIL(data1, blocksize);
1579 initialize_dirent_tail(t, blocksize);
1580 }
1581
1582 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1583 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1584
1585 /* Which block gets the new entry? */
1586 if (hinfo->hash >= hash2) {
1587 swap(*bh, bh2);
1588 de = de2;
1589 }
1590 dx_insert_block(frame, hash2 + continued, newblock);
1591 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1592 if (err)
1593 goto journal_error;
1594 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1595 if (err)
1596 goto journal_error;
1597 brelse(bh2);
1598 dxtrace(dx_show_index("frame", frame->entries));
1599 return de;
1600
1601 journal_error:
1602 brelse(*bh);
1603 brelse(bh2);
1604 *bh = NULL;
1605 ext4_std_error(dir->i_sb, err);
1606 return ERR_PTR(err);
1607 }
1608
1609 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1610 struct buffer_head *bh,
1611 void *buf, int buf_size,
1612 const char *name, int namelen,
1613 struct ext4_dir_entry_2 **dest_de)
1614 {
1615 struct ext4_dir_entry_2 *de;
1616 unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1617 int nlen, rlen;
1618 unsigned int offset = 0;
1619 char *top;
1620
1621 de = (struct ext4_dir_entry_2 *)buf;
1622 top = buf + buf_size - reclen;
1623 while ((char *) de <= top) {
1624 if (ext4_check_dir_entry(dir, NULL, de, bh,
1625 buf, buf_size, offset))
1626 return -EIO;
1627 if (ext4_match(namelen, name, de))
1628 return -EEXIST;
1629 nlen = EXT4_DIR_REC_LEN(de->name_len);
1630 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1631 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1632 break;
1633 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1634 offset += rlen;
1635 }
1636 if ((char *) de > top)
1637 return -ENOSPC;
1638
1639 *dest_de = de;
1640 return 0;
1641 }
1642
1643 void ext4_insert_dentry(struct inode *inode,
1644 struct ext4_dir_entry_2 *de,
1645 int buf_size,
1646 const char *name, int namelen)
1647 {
1648
1649 int nlen, rlen;
1650
1651 nlen = EXT4_DIR_REC_LEN(de->name_len);
1652 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1653 if (de->inode) {
1654 struct ext4_dir_entry_2 *de1 =
1655 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1656 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1657 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1658 de = de1;
1659 }
1660 de->file_type = EXT4_FT_UNKNOWN;
1661 de->inode = cpu_to_le32(inode->i_ino);
1662 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1663 de->name_len = namelen;
1664 memcpy(de->name, name, namelen);
1665 }
1666 /*
1667 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1668 * it points to a directory entry which is guaranteed to be large
1669 * enough for new directory entry. If de is NULL, then
1670 * add_dirent_to_buf will attempt search the directory block for
1671 * space. It will return -ENOSPC if no space is available, and -EIO
1672 * and -EEXIST if directory entry already exists.
1673 */
1674 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1675 struct inode *inode, struct ext4_dir_entry_2 *de,
1676 struct buffer_head *bh)
1677 {
1678 struct inode *dir = dentry->d_parent->d_inode;
1679 const char *name = dentry->d_name.name;
1680 int namelen = dentry->d_name.len;
1681 unsigned int blocksize = dir->i_sb->s_blocksize;
1682 int csum_size = 0;
1683 int err;
1684
1685 if (ext4_has_metadata_csum(inode->i_sb))
1686 csum_size = sizeof(struct ext4_dir_entry_tail);
1687
1688 if (!de) {
1689 err = ext4_find_dest_de(dir, inode,
1690 bh, bh->b_data, blocksize - csum_size,
1691 name, namelen, &de);
1692 if (err)
1693 return err;
1694 }
1695 BUFFER_TRACE(bh, "get_write_access");
1696 err = ext4_journal_get_write_access(handle, bh);
1697 if (err) {
1698 ext4_std_error(dir->i_sb, err);
1699 return err;
1700 }
1701
1702 /* By now the buffer is marked for journaling */
1703 ext4_insert_dentry(inode, de, blocksize, name, namelen);
1704
1705 /*
1706 * XXX shouldn't update any times until successful
1707 * completion of syscall, but too many callers depend
1708 * on this.
1709 *
1710 * XXX similarly, too many callers depend on
1711 * ext4_new_inode() setting the times, but error
1712 * recovery deletes the inode, so the worst that can
1713 * happen is that the times are slightly out of date
1714 * and/or different from the directory change time.
1715 */
1716 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1717 ext4_update_dx_flag(dir);
1718 dir->i_version++;
1719 ext4_mark_inode_dirty(handle, dir);
1720 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1721 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1722 if (err)
1723 ext4_std_error(dir->i_sb, err);
1724 return 0;
1725 }
1726
1727 /*
1728 * This converts a one block unindexed directory to a 3 block indexed
1729 * directory, and adds the dentry to the indexed directory.
1730 */
1731 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1732 struct inode *inode, struct buffer_head *bh)
1733 {
1734 struct inode *dir = dentry->d_parent->d_inode;
1735 const char *name = dentry->d_name.name;
1736 int namelen = dentry->d_name.len;
1737 struct buffer_head *bh2;
1738 struct dx_root *root;
1739 struct dx_frame frames[2], *frame;
1740 struct dx_entry *entries;
1741 struct ext4_dir_entry_2 *de, *de2;
1742 struct ext4_dir_entry_tail *t;
1743 char *data1, *top;
1744 unsigned len;
1745 int retval;
1746 unsigned blocksize;
1747 struct dx_hash_info hinfo;
1748 ext4_lblk_t block;
1749 struct fake_dirent *fde;
1750 int csum_size = 0;
1751
1752 if (ext4_has_metadata_csum(inode->i_sb))
1753 csum_size = sizeof(struct ext4_dir_entry_tail);
1754
1755 blocksize = dir->i_sb->s_blocksize;
1756 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1757 BUFFER_TRACE(bh, "get_write_access");
1758 retval = ext4_journal_get_write_access(handle, bh);
1759 if (retval) {
1760 ext4_std_error(dir->i_sb, retval);
1761 brelse(bh);
1762 return retval;
1763 }
1764 root = (struct dx_root *) bh->b_data;
1765
1766 /* The 0th block becomes the root, move the dirents out */
1767 fde = &root->dotdot;
1768 de = (struct ext4_dir_entry_2 *)((char *)fde +
1769 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1770 if ((char *) de >= (((char *) root) + blocksize)) {
1771 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1772 brelse(bh);
1773 return -EIO;
1774 }
1775 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1776
1777 /* Allocate new block for the 0th block's dirents */
1778 bh2 = ext4_append(handle, dir, &block);
1779 if (IS_ERR(bh2)) {
1780 brelse(bh);
1781 return PTR_ERR(bh2);
1782 }
1783 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1784 data1 = bh2->b_data;
1785
1786 memcpy (data1, de, len);
1787 de = (struct ext4_dir_entry_2 *) data1;
1788 top = data1 + len;
1789 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1790 de = de2;
1791 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1792 (char *) de,
1793 blocksize);
1794
1795 if (csum_size) {
1796 t = EXT4_DIRENT_TAIL(data1, blocksize);
1797 initialize_dirent_tail(t, blocksize);
1798 }
1799
1800 /* Initialize the root; the dot dirents already exist */
1801 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1802 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1803 blocksize);
1804 memset (&root->info, 0, sizeof(root->info));
1805 root->info.info_length = sizeof(root->info);
1806 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1807 entries = root->entries;
1808 dx_set_block(entries, 1);
1809 dx_set_count(entries, 1);
1810 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1811
1812 /* Initialize as for dx_probe */
1813 hinfo.hash_version = root->info.hash_version;
1814 if (hinfo.hash_version <= DX_HASH_TEA)
1815 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1816 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1817 ext4fs_dirhash(name, namelen, &hinfo);
1818 memset(frames, 0, sizeof(frames));
1819 frame = frames;
1820 frame->entries = entries;
1821 frame->at = entries;
1822 frame->bh = bh;
1823 bh = bh2;
1824
1825 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1826 if (retval)
1827 goto out_frames;
1828 retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
1829 if (retval)
1830 goto out_frames;
1831
1832 de = do_split(handle,dir, &bh, frame, &hinfo);
1833 if (IS_ERR(de)) {
1834 retval = PTR_ERR(de);
1835 goto out_frames;
1836 }
1837 dx_release(frames);
1838
1839 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1840 brelse(bh);
1841 return retval;
1842 out_frames:
1843 /*
1844 * Even if the block split failed, we have to properly write
1845 * out all the changes we did so far. Otherwise we can end up
1846 * with corrupted filesystem.
1847 */
1848 ext4_mark_inode_dirty(handle, dir);
1849 dx_release(frames);
1850 return retval;
1851 }
1852
1853 /*
1854 * ext4_add_entry()
1855 *
1856 * adds a file entry to the specified directory, using the same
1857 * semantics as ext4_find_entry(). It returns NULL if it failed.
1858 *
1859 * NOTE!! The inode part of 'de' is left at 0 - which means you
1860 * may not sleep between calling this and putting something into
1861 * the entry, as someone else might have used it while you slept.
1862 */
1863 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1864 struct inode *inode)
1865 {
1866 struct inode *dir = dentry->d_parent->d_inode;
1867 struct buffer_head *bh;
1868 struct ext4_dir_entry_2 *de;
1869 struct ext4_dir_entry_tail *t;
1870 struct super_block *sb;
1871 int retval;
1872 int dx_fallback=0;
1873 unsigned blocksize;
1874 ext4_lblk_t block, blocks;
1875 int csum_size = 0;
1876
1877 if (ext4_has_metadata_csum(inode->i_sb))
1878 csum_size = sizeof(struct ext4_dir_entry_tail);
1879
1880 sb = dir->i_sb;
1881 blocksize = sb->s_blocksize;
1882 if (!dentry->d_name.len)
1883 return -EINVAL;
1884
1885 if (ext4_has_inline_data(dir)) {
1886 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1887 if (retval < 0)
1888 return retval;
1889 if (retval == 1) {
1890 retval = 0;
1891 return retval;
1892 }
1893 }
1894
1895 if (is_dx(dir)) {
1896 retval = ext4_dx_add_entry(handle, dentry, inode);
1897 if (!retval || (retval != ERR_BAD_DX_DIR))
1898 return retval;
1899 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1900 dx_fallback++;
1901 ext4_mark_inode_dirty(handle, dir);
1902 }
1903 blocks = dir->i_size >> sb->s_blocksize_bits;
1904 for (block = 0; block < blocks; block++) {
1905 bh = ext4_read_dirblock(dir, block, DIRENT);
1906 if (IS_ERR(bh))
1907 return PTR_ERR(bh);
1908
1909 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1910 if (retval != -ENOSPC) {
1911 brelse(bh);
1912 return retval;
1913 }
1914
1915 if (blocks == 1 && !dx_fallback &&
1916 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1917 return make_indexed_dir(handle, dentry, inode, bh);
1918 brelse(bh);
1919 }
1920 bh = ext4_append(handle, dir, &block);
1921 if (IS_ERR(bh))
1922 return PTR_ERR(bh);
1923 de = (struct ext4_dir_entry_2 *) bh->b_data;
1924 de->inode = 0;
1925 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1926
1927 if (csum_size) {
1928 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1929 initialize_dirent_tail(t, blocksize);
1930 }
1931
1932 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1933 brelse(bh);
1934 if (retval == 0)
1935 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1936 return retval;
1937 }
1938
1939 /*
1940 * Returns 0 for success, or a negative error value
1941 */
1942 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1943 struct inode *inode)
1944 {
1945 struct dx_frame frames[2], *frame;
1946 struct dx_entry *entries, *at;
1947 struct dx_hash_info hinfo;
1948 struct buffer_head *bh;
1949 struct inode *dir = dentry->d_parent->d_inode;
1950 struct super_block *sb = dir->i_sb;
1951 struct ext4_dir_entry_2 *de;
1952 int err;
1953
1954 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames);
1955 if (IS_ERR(frame))
1956 return PTR_ERR(frame);
1957 entries = frame->entries;
1958 at = frame->at;
1959 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1960 if (IS_ERR(bh)) {
1961 err = PTR_ERR(bh);
1962 bh = NULL;
1963 goto cleanup;
1964 }
1965
1966 BUFFER_TRACE(bh, "get_write_access");
1967 err = ext4_journal_get_write_access(handle, bh);
1968 if (err)
1969 goto journal_error;
1970
1971 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1972 if (err != -ENOSPC)
1973 goto cleanup;
1974
1975 /* Block full, should compress but for now just split */
1976 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1977 dx_get_count(entries), dx_get_limit(entries)));
1978 /* Need to split index? */
1979 if (dx_get_count(entries) == dx_get_limit(entries)) {
1980 ext4_lblk_t newblock;
1981 unsigned icount = dx_get_count(entries);
1982 int levels = frame - frames;
1983 struct dx_entry *entries2;
1984 struct dx_node *node2;
1985 struct buffer_head *bh2;
1986
1987 if (levels && (dx_get_count(frames->entries) ==
1988 dx_get_limit(frames->entries))) {
1989 ext4_warning(sb, "Directory index full!");
1990 err = -ENOSPC;
1991 goto cleanup;
1992 }
1993 bh2 = ext4_append(handle, dir, &newblock);
1994 if (IS_ERR(bh2)) {
1995 err = PTR_ERR(bh2);
1996 goto cleanup;
1997 }
1998 node2 = (struct dx_node *)(bh2->b_data);
1999 entries2 = node2->entries;
2000 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2001 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2002 sb->s_blocksize);
2003 BUFFER_TRACE(frame->bh, "get_write_access");
2004 err = ext4_journal_get_write_access(handle, frame->bh);
2005 if (err)
2006 goto journal_error;
2007 if (levels) {
2008 unsigned icount1 = icount/2, icount2 = icount - icount1;
2009 unsigned hash2 = dx_get_hash(entries + icount1);
2010 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2011 icount1, icount2));
2012
2013 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2014 err = ext4_journal_get_write_access(handle,
2015 frames[0].bh);
2016 if (err)
2017 goto journal_error;
2018
2019 memcpy((char *) entries2, (char *) (entries + icount1),
2020 icount2 * sizeof(struct dx_entry));
2021 dx_set_count(entries, icount1);
2022 dx_set_count(entries2, icount2);
2023 dx_set_limit(entries2, dx_node_limit(dir));
2024
2025 /* Which index block gets the new entry? */
2026 if (at - entries >= icount1) {
2027 frame->at = at = at - entries - icount1 + entries2;
2028 frame->entries = entries = entries2;
2029 swap(frame->bh, bh2);
2030 }
2031 dx_insert_block(frames + 0, hash2, newblock);
2032 dxtrace(dx_show_index("node", frames[1].entries));
2033 dxtrace(dx_show_index("node",
2034 ((struct dx_node *) bh2->b_data)->entries));
2035 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2036 if (err)
2037 goto journal_error;
2038 brelse (bh2);
2039 } else {
2040 dxtrace(printk(KERN_DEBUG
2041 "Creating second level index...\n"));
2042 memcpy((char *) entries2, (char *) entries,
2043 icount * sizeof(struct dx_entry));
2044 dx_set_limit(entries2, dx_node_limit(dir));
2045
2046 /* Set up root */
2047 dx_set_count(entries, 1);
2048 dx_set_block(entries + 0, newblock);
2049 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2050
2051 /* Add new access path frame */
2052 frame = frames + 1;
2053 frame->at = at = at - entries + entries2;
2054 frame->entries = entries = entries2;
2055 frame->bh = bh2;
2056 err = ext4_journal_get_write_access(handle,
2057 frame->bh);
2058 if (err)
2059 goto journal_error;
2060 }
2061 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2062 if (err) {
2063 ext4_std_error(inode->i_sb, err);
2064 goto cleanup;
2065 }
2066 }
2067 de = do_split(handle, dir, &bh, frame, &hinfo);
2068 if (IS_ERR(de)) {
2069 err = PTR_ERR(de);
2070 goto cleanup;
2071 }
2072 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2073 goto cleanup;
2074
2075 journal_error:
2076 ext4_std_error(dir->i_sb, err);
2077 cleanup:
2078 brelse(bh);
2079 dx_release(frames);
2080 return err;
2081 }
2082
2083 /*
2084 * ext4_generic_delete_entry deletes a directory entry by merging it
2085 * with the previous entry
2086 */
2087 int ext4_generic_delete_entry(handle_t *handle,
2088 struct inode *dir,
2089 struct ext4_dir_entry_2 *de_del,
2090 struct buffer_head *bh,
2091 void *entry_buf,
2092 int buf_size,
2093 int csum_size)
2094 {
2095 struct ext4_dir_entry_2 *de, *pde;
2096 unsigned int blocksize = dir->i_sb->s_blocksize;
2097 int i;
2098
2099 i = 0;
2100 pde = NULL;
2101 de = (struct ext4_dir_entry_2 *)entry_buf;
2102 while (i < buf_size - csum_size) {
2103 if (ext4_check_dir_entry(dir, NULL, de, bh,
2104 bh->b_data, bh->b_size, i))
2105 return -EIO;
2106 if (de == de_del) {
2107 if (pde)
2108 pde->rec_len = ext4_rec_len_to_disk(
2109 ext4_rec_len_from_disk(pde->rec_len,
2110 blocksize) +
2111 ext4_rec_len_from_disk(de->rec_len,
2112 blocksize),
2113 blocksize);
2114 else
2115 de->inode = 0;
2116 dir->i_version++;
2117 return 0;
2118 }
2119 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2120 pde = de;
2121 de = ext4_next_entry(de, blocksize);
2122 }
2123 return -ENOENT;
2124 }
2125
2126 static int ext4_delete_entry(handle_t *handle,
2127 struct inode *dir,
2128 struct ext4_dir_entry_2 *de_del,
2129 struct buffer_head *bh)
2130 {
2131 int err, csum_size = 0;
2132
2133 if (ext4_has_inline_data(dir)) {
2134 int has_inline_data = 1;
2135 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2136 &has_inline_data);
2137 if (has_inline_data)
2138 return err;
2139 }
2140
2141 if (ext4_has_metadata_csum(dir->i_sb))
2142 csum_size = sizeof(struct ext4_dir_entry_tail);
2143
2144 BUFFER_TRACE(bh, "get_write_access");
2145 err = ext4_journal_get_write_access(handle, bh);
2146 if (unlikely(err))
2147 goto out;
2148
2149 err = ext4_generic_delete_entry(handle, dir, de_del,
2150 bh, bh->b_data,
2151 dir->i_sb->s_blocksize, csum_size);
2152 if (err)
2153 goto out;
2154
2155 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2156 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2157 if (unlikely(err))
2158 goto out;
2159
2160 return 0;
2161 out:
2162 if (err != -ENOENT)
2163 ext4_std_error(dir->i_sb, err);
2164 return err;
2165 }
2166
2167 /*
2168 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2169 * since this indicates that nlinks count was previously 1.
2170 */
2171 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2172 {
2173 inc_nlink(inode);
2174 if (is_dx(inode) && inode->i_nlink > 1) {
2175 /* limit is 16-bit i_links_count */
2176 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2177 set_nlink(inode, 1);
2178 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2179 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2180 }
2181 }
2182 }
2183
2184 /*
2185 * If a directory had nlink == 1, then we should let it be 1. This indicates
2186 * directory has >EXT4_LINK_MAX subdirs.
2187 */
2188 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2189 {
2190 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2191 drop_nlink(inode);
2192 }
2193
2194
2195 static int ext4_add_nondir(handle_t *handle,
2196 struct dentry *dentry, struct inode *inode)
2197 {
2198 int err = ext4_add_entry(handle, dentry, inode);
2199 if (!err) {
2200 ext4_mark_inode_dirty(handle, inode);
2201 unlock_new_inode(inode);
2202 d_instantiate(dentry, inode);
2203 return 0;
2204 }
2205 drop_nlink(inode);
2206 unlock_new_inode(inode);
2207 iput(inode);
2208 return err;
2209 }
2210
2211 /*
2212 * By the time this is called, we already have created
2213 * the directory cache entry for the new file, but it
2214 * is so far negative - it has no inode.
2215 *
2216 * If the create succeeds, we fill in the inode information
2217 * with d_instantiate().
2218 */
2219 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2220 bool excl)
2221 {
2222 handle_t *handle;
2223 struct inode *inode;
2224 int err, credits, retries = 0;
2225
2226 dquot_initialize(dir);
2227
2228 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2229 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2230 retry:
2231 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2232 NULL, EXT4_HT_DIR, credits);
2233 handle = ext4_journal_current_handle();
2234 err = PTR_ERR(inode);
2235 if (!IS_ERR(inode)) {
2236 inode->i_op = &ext4_file_inode_operations;
2237 if (test_opt(inode->i_sb, DAX))
2238 inode->i_fop = &ext4_dax_file_operations;
2239 else
2240 inode->i_fop = &ext4_file_operations;
2241 ext4_set_aops(inode);
2242 err = ext4_add_nondir(handle, dentry, inode);
2243 if (!err && IS_DIRSYNC(dir))
2244 ext4_handle_sync(handle);
2245 }
2246 if (handle)
2247 ext4_journal_stop(handle);
2248 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2249 goto retry;
2250 return err;
2251 }
2252
2253 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2254 umode_t mode, dev_t rdev)
2255 {
2256 handle_t *handle;
2257 struct inode *inode;
2258 int err, credits, retries = 0;
2259
2260 if (!new_valid_dev(rdev))
2261 return -EINVAL;
2262
2263 dquot_initialize(dir);
2264
2265 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2266 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2267 retry:
2268 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2269 NULL, EXT4_HT_DIR, credits);
2270 handle = ext4_journal_current_handle();
2271 err = PTR_ERR(inode);
2272 if (!IS_ERR(inode)) {
2273 init_special_inode(inode, inode->i_mode, rdev);
2274 inode->i_op = &ext4_special_inode_operations;
2275 err = ext4_add_nondir(handle, dentry, inode);
2276 if (!err && IS_DIRSYNC(dir))
2277 ext4_handle_sync(handle);
2278 }
2279 if (handle)
2280 ext4_journal_stop(handle);
2281 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2282 goto retry;
2283 return err;
2284 }
2285
2286 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2287 {
2288 handle_t *handle;
2289 struct inode *inode;
2290 int err, retries = 0;
2291
2292 dquot_initialize(dir);
2293
2294 retry:
2295 inode = ext4_new_inode_start_handle(dir, mode,
2296 NULL, 0, NULL,
2297 EXT4_HT_DIR,
2298 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2299 4 + EXT4_XATTR_TRANS_BLOCKS);
2300 handle = ext4_journal_current_handle();
2301 err = PTR_ERR(inode);
2302 if (!IS_ERR(inode)) {
2303 inode->i_op = &ext4_file_inode_operations;
2304 if (test_opt(inode->i_sb, DAX))
2305 inode->i_fop = &ext4_dax_file_operations;
2306 else
2307 inode->i_fop = &ext4_file_operations;
2308 ext4_set_aops(inode);
2309 d_tmpfile(dentry, inode);
2310 err = ext4_orphan_add(handle, inode);
2311 if (err)
2312 goto err_unlock_inode;
2313 mark_inode_dirty(inode);
2314 unlock_new_inode(inode);
2315 }
2316 if (handle)
2317 ext4_journal_stop(handle);
2318 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2319 goto retry;
2320 return err;
2321 err_unlock_inode:
2322 ext4_journal_stop(handle);
2323 unlock_new_inode(inode);
2324 return err;
2325 }
2326
2327 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2328 struct ext4_dir_entry_2 *de,
2329 int blocksize, int csum_size,
2330 unsigned int parent_ino, int dotdot_real_len)
2331 {
2332 de->inode = cpu_to_le32(inode->i_ino);
2333 de->name_len = 1;
2334 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2335 blocksize);
2336 strcpy(de->name, ".");
2337 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2338
2339 de = ext4_next_entry(de, blocksize);
2340 de->inode = cpu_to_le32(parent_ino);
2341 de->name_len = 2;
2342 if (!dotdot_real_len)
2343 de->rec_len = ext4_rec_len_to_disk(blocksize -
2344 (csum_size + EXT4_DIR_REC_LEN(1)),
2345 blocksize);
2346 else
2347 de->rec_len = ext4_rec_len_to_disk(
2348 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2349 strcpy(de->name, "..");
2350 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2351
2352 return ext4_next_entry(de, blocksize);
2353 }
2354
2355 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2356 struct inode *inode)
2357 {
2358 struct buffer_head *dir_block = NULL;
2359 struct ext4_dir_entry_2 *de;
2360 struct ext4_dir_entry_tail *t;
2361 ext4_lblk_t block = 0;
2362 unsigned int blocksize = dir->i_sb->s_blocksize;
2363 int csum_size = 0;
2364 int err;
2365
2366 if (ext4_has_metadata_csum(dir->i_sb))
2367 csum_size = sizeof(struct ext4_dir_entry_tail);
2368
2369 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2370 err = ext4_try_create_inline_dir(handle, dir, inode);
2371 if (err < 0 && err != -ENOSPC)
2372 goto out;
2373 if (!err)
2374 goto out;
2375 }
2376
2377 inode->i_size = 0;
2378 dir_block = ext4_append(handle, inode, &block);
2379 if (IS_ERR(dir_block))
2380 return PTR_ERR(dir_block);
2381 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2382 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2383 set_nlink(inode, 2);
2384 if (csum_size) {
2385 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2386 initialize_dirent_tail(t, blocksize);
2387 }
2388
2389 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2390 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2391 if (err)
2392 goto out;
2393 set_buffer_verified(dir_block);
2394 out:
2395 brelse(dir_block);
2396 return err;
2397 }
2398
2399 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2400 {
2401 handle_t *handle;
2402 struct inode *inode;
2403 int err, credits, retries = 0;
2404
2405 if (EXT4_DIR_LINK_MAX(dir))
2406 return -EMLINK;
2407
2408 dquot_initialize(dir);
2409
2410 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2411 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2412 retry:
2413 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2414 &dentry->d_name,
2415 0, NULL, EXT4_HT_DIR, credits);
2416 handle = ext4_journal_current_handle();
2417 err = PTR_ERR(inode);
2418 if (IS_ERR(inode))
2419 goto out_stop;
2420
2421 inode->i_op = &ext4_dir_inode_operations;
2422 inode->i_fop = &ext4_dir_operations;
2423 err = ext4_init_new_dir(handle, dir, inode);
2424 if (err)
2425 goto out_clear_inode;
2426 err = ext4_mark_inode_dirty(handle, inode);
2427 if (!err)
2428 err = ext4_add_entry(handle, dentry, inode);
2429 if (err) {
2430 out_clear_inode:
2431 clear_nlink(inode);
2432 unlock_new_inode(inode);
2433 ext4_mark_inode_dirty(handle, inode);
2434 iput(inode);
2435 goto out_stop;
2436 }
2437 ext4_inc_count(handle, dir);
2438 ext4_update_dx_flag(dir);
2439 err = ext4_mark_inode_dirty(handle, dir);
2440 if (err)
2441 goto out_clear_inode;
2442 unlock_new_inode(inode);
2443 d_instantiate(dentry, inode);
2444 if (IS_DIRSYNC(dir))
2445 ext4_handle_sync(handle);
2446
2447 out_stop:
2448 if (handle)
2449 ext4_journal_stop(handle);
2450 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2451 goto retry;
2452 return err;
2453 }
2454
2455 /*
2456 * routine to check that the specified directory is empty (for rmdir)
2457 */
2458 static int empty_dir(struct inode *inode)
2459 {
2460 unsigned int offset;
2461 struct buffer_head *bh;
2462 struct ext4_dir_entry_2 *de, *de1;
2463 struct super_block *sb;
2464 int err = 0;
2465
2466 if (ext4_has_inline_data(inode)) {
2467 int has_inline_data = 1;
2468
2469 err = empty_inline_dir(inode, &has_inline_data);
2470 if (has_inline_data)
2471 return err;
2472 }
2473
2474 sb = inode->i_sb;
2475 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2476 EXT4_ERROR_INODE(inode, "invalid size");
2477 return 1;
2478 }
2479 bh = ext4_read_dirblock(inode, 0, EITHER);
2480 if (IS_ERR(bh))
2481 return 1;
2482
2483 de = (struct ext4_dir_entry_2 *) bh->b_data;
2484 de1 = ext4_next_entry(de, sb->s_blocksize);
2485 if (le32_to_cpu(de->inode) != inode->i_ino ||
2486 !le32_to_cpu(de1->inode) ||
2487 strcmp(".", de->name) ||
2488 strcmp("..", de1->name)) {
2489 ext4_warning(inode->i_sb,
2490 "bad directory (dir #%lu) - no `.' or `..'",
2491 inode->i_ino);
2492 brelse(bh);
2493 return 1;
2494 }
2495 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2496 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2497 de = ext4_next_entry(de1, sb->s_blocksize);
2498 while (offset < inode->i_size) {
2499 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2500 unsigned int lblock;
2501 err = 0;
2502 brelse(bh);
2503 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2504 bh = ext4_read_dirblock(inode, lblock, EITHER);
2505 if (IS_ERR(bh))
2506 return 1;
2507 de = (struct ext4_dir_entry_2 *) bh->b_data;
2508 }
2509 if (ext4_check_dir_entry(inode, NULL, de, bh,
2510 bh->b_data, bh->b_size, offset)) {
2511 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2512 sb->s_blocksize);
2513 offset = (offset | (sb->s_blocksize - 1)) + 1;
2514 continue;
2515 }
2516 if (le32_to_cpu(de->inode)) {
2517 brelse(bh);
2518 return 0;
2519 }
2520 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2521 de = ext4_next_entry(de, sb->s_blocksize);
2522 }
2523 brelse(bh);
2524 return 1;
2525 }
2526
2527 /*
2528 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2529 * such inodes, starting at the superblock, in case we crash before the
2530 * file is closed/deleted, or in case the inode truncate spans multiple
2531 * transactions and the last transaction is not recovered after a crash.
2532 *
2533 * At filesystem recovery time, we walk this list deleting unlinked
2534 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2535 *
2536 * Orphan list manipulation functions must be called under i_mutex unless
2537 * we are just creating the inode or deleting it.
2538 */
2539 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2540 {
2541 struct super_block *sb = inode->i_sb;
2542 struct ext4_sb_info *sbi = EXT4_SB(sb);
2543 struct ext4_iloc iloc;
2544 int err = 0, rc;
2545 bool dirty = false;
2546
2547 if (!sbi->s_journal || is_bad_inode(inode))
2548 return 0;
2549
2550 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2551 !mutex_is_locked(&inode->i_mutex));
2552 /*
2553 * Exit early if inode already is on orphan list. This is a big speedup
2554 * since we don't have to contend on the global s_orphan_lock.
2555 */
2556 if (!list_empty(&EXT4_I(inode)->i_orphan))
2557 return 0;
2558
2559 /*
2560 * Orphan handling is only valid for files with data blocks
2561 * being truncated, or files being unlinked. Note that we either
2562 * hold i_mutex, or the inode can not be referenced from outside,
2563 * so i_nlink should not be bumped due to race
2564 */
2565 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2566 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2567
2568 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2569 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2570 if (err)
2571 goto out;
2572
2573 err = ext4_reserve_inode_write(handle, inode, &iloc);
2574 if (err)
2575 goto out;
2576
2577 mutex_lock(&sbi->s_orphan_lock);
2578 /*
2579 * Due to previous errors inode may be already a part of on-disk
2580 * orphan list. If so skip on-disk list modification.
2581 */
2582 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2583 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2584 /* Insert this inode at the head of the on-disk orphan list */
2585 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2586 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2587 dirty = true;
2588 }
2589 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2590 mutex_unlock(&sbi->s_orphan_lock);
2591
2592 if (dirty) {
2593 err = ext4_handle_dirty_super(handle, sb);
2594 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2595 if (!err)
2596 err = rc;
2597 if (err) {
2598 /*
2599 * We have to remove inode from in-memory list if
2600 * addition to on disk orphan list failed. Stray orphan
2601 * list entries can cause panics at unmount time.
2602 */
2603 mutex_lock(&sbi->s_orphan_lock);
2604 list_del(&EXT4_I(inode)->i_orphan);
2605 mutex_unlock(&sbi->s_orphan_lock);
2606 }
2607 }
2608 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2609 jbd_debug(4, "orphan inode %lu will point to %d\n",
2610 inode->i_ino, NEXT_ORPHAN(inode));
2611 out:
2612 ext4_std_error(sb, err);
2613 return err;
2614 }
2615
2616 /*
2617 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2618 * of such inodes stored on disk, because it is finally being cleaned up.
2619 */
2620 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2621 {
2622 struct list_head *prev;
2623 struct ext4_inode_info *ei = EXT4_I(inode);
2624 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2625 __u32 ino_next;
2626 struct ext4_iloc iloc;
2627 int err = 0;
2628
2629 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2630 return 0;
2631
2632 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2633 !mutex_is_locked(&inode->i_mutex));
2634 /* Do this quick check before taking global s_orphan_lock. */
2635 if (list_empty(&ei->i_orphan))
2636 return 0;
2637
2638 if (handle) {
2639 /* Grab inode buffer early before taking global s_orphan_lock */
2640 err = ext4_reserve_inode_write(handle, inode, &iloc);
2641 }
2642
2643 mutex_lock(&sbi->s_orphan_lock);
2644 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2645
2646 prev = ei->i_orphan.prev;
2647 list_del_init(&ei->i_orphan);
2648
2649 /* If we're on an error path, we may not have a valid
2650 * transaction handle with which to update the orphan list on
2651 * disk, but we still need to remove the inode from the linked
2652 * list in memory. */
2653 if (!handle || err) {
2654 mutex_unlock(&sbi->s_orphan_lock);
2655 goto out_err;
2656 }
2657
2658 ino_next = NEXT_ORPHAN(inode);
2659 if (prev == &sbi->s_orphan) {
2660 jbd_debug(4, "superblock will point to %u\n", ino_next);
2661 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2662 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2663 if (err) {
2664 mutex_unlock(&sbi->s_orphan_lock);
2665 goto out_brelse;
2666 }
2667 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2668 mutex_unlock(&sbi->s_orphan_lock);
2669 err = ext4_handle_dirty_super(handle, inode->i_sb);
2670 } else {
2671 struct ext4_iloc iloc2;
2672 struct inode *i_prev =
2673 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2674
2675 jbd_debug(4, "orphan inode %lu will point to %u\n",
2676 i_prev->i_ino, ino_next);
2677 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2678 if (err) {
2679 mutex_unlock(&sbi->s_orphan_lock);
2680 goto out_brelse;
2681 }
2682 NEXT_ORPHAN(i_prev) = ino_next;
2683 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2684 mutex_unlock(&sbi->s_orphan_lock);
2685 }
2686 if (err)
2687 goto out_brelse;
2688 NEXT_ORPHAN(inode) = 0;
2689 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2690 out_err:
2691 ext4_std_error(inode->i_sb, err);
2692 return err;
2693
2694 out_brelse:
2695 brelse(iloc.bh);
2696 goto out_err;
2697 }
2698
2699 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2700 {
2701 int retval;
2702 struct inode *inode;
2703 struct buffer_head *bh;
2704 struct ext4_dir_entry_2 *de;
2705 handle_t *handle = NULL;
2706
2707 /* Initialize quotas before so that eventual writes go in
2708 * separate transaction */
2709 dquot_initialize(dir);
2710 dquot_initialize(dentry->d_inode);
2711
2712 retval = -ENOENT;
2713 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2714 if (IS_ERR(bh))
2715 return PTR_ERR(bh);
2716 if (!bh)
2717 goto end_rmdir;
2718
2719 inode = dentry->d_inode;
2720
2721 retval = -EIO;
2722 if (le32_to_cpu(de->inode) != inode->i_ino)
2723 goto end_rmdir;
2724
2725 retval = -ENOTEMPTY;
2726 if (!empty_dir(inode))
2727 goto end_rmdir;
2728
2729 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2730 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2731 if (IS_ERR(handle)) {
2732 retval = PTR_ERR(handle);
2733 handle = NULL;
2734 goto end_rmdir;
2735 }
2736
2737 if (IS_DIRSYNC(dir))
2738 ext4_handle_sync(handle);
2739
2740 retval = ext4_delete_entry(handle, dir, de, bh);
2741 if (retval)
2742 goto end_rmdir;
2743 if (!EXT4_DIR_LINK_EMPTY(inode))
2744 ext4_warning(inode->i_sb,
2745 "empty directory has too many links (%d)",
2746 inode->i_nlink);
2747 inode->i_version++;
2748 clear_nlink(inode);
2749 /* There's no need to set i_disksize: the fact that i_nlink is
2750 * zero will ensure that the right thing happens during any
2751 * recovery. */
2752 inode->i_size = 0;
2753 ext4_orphan_add(handle, inode);
2754 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2755 ext4_mark_inode_dirty(handle, inode);
2756 ext4_dec_count(handle, dir);
2757 ext4_update_dx_flag(dir);
2758 ext4_mark_inode_dirty(handle, dir);
2759
2760 end_rmdir:
2761 brelse(bh);
2762 if (handle)
2763 ext4_journal_stop(handle);
2764 return retval;
2765 }
2766
2767 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2768 {
2769 int retval;
2770 struct inode *inode;
2771 struct buffer_head *bh;
2772 struct ext4_dir_entry_2 *de;
2773 handle_t *handle = NULL;
2774
2775 trace_ext4_unlink_enter(dir, dentry);
2776 /* Initialize quotas before so that eventual writes go
2777 * in separate transaction */
2778 dquot_initialize(dir);
2779 dquot_initialize(dentry->d_inode);
2780
2781 retval = -ENOENT;
2782 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2783 if (IS_ERR(bh))
2784 return PTR_ERR(bh);
2785 if (!bh)
2786 goto end_unlink;
2787
2788 inode = dentry->d_inode;
2789
2790 retval = -EIO;
2791 if (le32_to_cpu(de->inode) != inode->i_ino)
2792 goto end_unlink;
2793
2794 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2795 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2796 if (IS_ERR(handle)) {
2797 retval = PTR_ERR(handle);
2798 handle = NULL;
2799 goto end_unlink;
2800 }
2801
2802 if (IS_DIRSYNC(dir))
2803 ext4_handle_sync(handle);
2804
2805 if (!inode->i_nlink) {
2806 ext4_warning(inode->i_sb,
2807 "Deleting nonexistent file (%lu), %d",
2808 inode->i_ino, inode->i_nlink);
2809 set_nlink(inode, 1);
2810 }
2811 retval = ext4_delete_entry(handle, dir, de, bh);
2812 if (retval)
2813 goto end_unlink;
2814 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2815 ext4_update_dx_flag(dir);
2816 ext4_mark_inode_dirty(handle, dir);
2817 drop_nlink(inode);
2818 if (!inode->i_nlink)
2819 ext4_orphan_add(handle, inode);
2820 inode->i_ctime = ext4_current_time(inode);
2821 ext4_mark_inode_dirty(handle, inode);
2822
2823 end_unlink:
2824 brelse(bh);
2825 if (handle)
2826 ext4_journal_stop(handle);
2827 trace_ext4_unlink_exit(dentry, retval);
2828 return retval;
2829 }
2830
2831 static int ext4_symlink(struct inode *dir,
2832 struct dentry *dentry, const char *symname)
2833 {
2834 handle_t *handle;
2835 struct inode *inode;
2836 int l, err, retries = 0;
2837 int credits;
2838
2839 l = strlen(symname)+1;
2840 if (l > dir->i_sb->s_blocksize)
2841 return -ENAMETOOLONG;
2842
2843 dquot_initialize(dir);
2844
2845 if (l > EXT4_N_BLOCKS * 4) {
2846 /*
2847 * For non-fast symlinks, we just allocate inode and put it on
2848 * orphan list in the first transaction => we need bitmap,
2849 * group descriptor, sb, inode block, quota blocks, and
2850 * possibly selinux xattr blocks.
2851 */
2852 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2853 EXT4_XATTR_TRANS_BLOCKS;
2854 } else {
2855 /*
2856 * Fast symlink. We have to add entry to directory
2857 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2858 * allocate new inode (bitmap, group descriptor, inode block,
2859 * quota blocks, sb is already counted in previous macros).
2860 */
2861 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2862 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2863 }
2864 retry:
2865 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2866 &dentry->d_name, 0, NULL,
2867 EXT4_HT_DIR, credits);
2868 handle = ext4_journal_current_handle();
2869 err = PTR_ERR(inode);
2870 if (IS_ERR(inode))
2871 goto out_stop;
2872
2873 if (l > EXT4_N_BLOCKS * 4) {
2874 inode->i_op = &ext4_symlink_inode_operations;
2875 ext4_set_aops(inode);
2876 /*
2877 * We cannot call page_symlink() with transaction started
2878 * because it calls into ext4_write_begin() which can wait
2879 * for transaction commit if we are running out of space
2880 * and thus we deadlock. So we have to stop transaction now
2881 * and restart it when symlink contents is written.
2882 *
2883 * To keep fs consistent in case of crash, we have to put inode
2884 * to orphan list in the mean time.
2885 */
2886 drop_nlink(inode);
2887 err = ext4_orphan_add(handle, inode);
2888 ext4_journal_stop(handle);
2889 if (err)
2890 goto err_drop_inode;
2891 err = __page_symlink(inode, symname, l, 1);
2892 if (err)
2893 goto err_drop_inode;
2894 /*
2895 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2896 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2897 */
2898 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2899 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2900 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2901 if (IS_ERR(handle)) {
2902 err = PTR_ERR(handle);
2903 goto err_drop_inode;
2904 }
2905 set_nlink(inode, 1);
2906 err = ext4_orphan_del(handle, inode);
2907 if (err) {
2908 ext4_journal_stop(handle);
2909 clear_nlink(inode);
2910 goto err_drop_inode;
2911 }
2912 } else {
2913 /* clear the extent format for fast symlink */
2914 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2915 inode->i_op = &ext4_fast_symlink_inode_operations;
2916 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2917 inode->i_size = l-1;
2918 }
2919 EXT4_I(inode)->i_disksize = inode->i_size;
2920 err = ext4_add_nondir(handle, dentry, inode);
2921 if (!err && IS_DIRSYNC(dir))
2922 ext4_handle_sync(handle);
2923
2924 out_stop:
2925 if (handle)
2926 ext4_journal_stop(handle);
2927 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2928 goto retry;
2929 return err;
2930 err_drop_inode:
2931 unlock_new_inode(inode);
2932 iput(inode);
2933 return err;
2934 }
2935
2936 static int ext4_link(struct dentry *old_dentry,
2937 struct inode *dir, struct dentry *dentry)
2938 {
2939 handle_t *handle;
2940 struct inode *inode = old_dentry->d_inode;
2941 int err, retries = 0;
2942
2943 if (inode->i_nlink >= EXT4_LINK_MAX)
2944 return -EMLINK;
2945
2946 dquot_initialize(dir);
2947
2948 retry:
2949 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2950 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2951 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2952 if (IS_ERR(handle))
2953 return PTR_ERR(handle);
2954
2955 if (IS_DIRSYNC(dir))
2956 ext4_handle_sync(handle);
2957
2958 inode->i_ctime = ext4_current_time(inode);
2959 ext4_inc_count(handle, inode);
2960 ihold(inode);
2961
2962 err = ext4_add_entry(handle, dentry, inode);
2963 if (!err) {
2964 ext4_mark_inode_dirty(handle, inode);
2965 /* this can happen only for tmpfile being
2966 * linked the first time
2967 */
2968 if (inode->i_nlink == 1)
2969 ext4_orphan_del(handle, inode);
2970 d_instantiate(dentry, inode);
2971 } else {
2972 drop_nlink(inode);
2973 iput(inode);
2974 }
2975 ext4_journal_stop(handle);
2976 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2977 goto retry;
2978 return err;
2979 }
2980
2981
2982 /*
2983 * Try to find buffer head where contains the parent block.
2984 * It should be the inode block if it is inlined or the 1st block
2985 * if it is a normal dir.
2986 */
2987 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
2988 struct inode *inode,
2989 int *retval,
2990 struct ext4_dir_entry_2 **parent_de,
2991 int *inlined)
2992 {
2993 struct buffer_head *bh;
2994
2995 if (!ext4_has_inline_data(inode)) {
2996 bh = ext4_read_dirblock(inode, 0, EITHER);
2997 if (IS_ERR(bh)) {
2998 *retval = PTR_ERR(bh);
2999 return NULL;
3000 }
3001 *parent_de = ext4_next_entry(
3002 (struct ext4_dir_entry_2 *)bh->b_data,
3003 inode->i_sb->s_blocksize);
3004 return bh;
3005 }
3006
3007 *inlined = 1;
3008 return ext4_get_first_inline_block(inode, parent_de, retval);
3009 }
3010
3011 struct ext4_renament {
3012 struct inode *dir;
3013 struct dentry *dentry;
3014 struct inode *inode;
3015 bool is_dir;
3016 int dir_nlink_delta;
3017
3018 /* entry for "dentry" */
3019 struct buffer_head *bh;
3020 struct ext4_dir_entry_2 *de;
3021 int inlined;
3022
3023 /* entry for ".." in inode if it's a directory */
3024 struct buffer_head *dir_bh;
3025 struct ext4_dir_entry_2 *parent_de;
3026 int dir_inlined;
3027 };
3028
3029 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3030 {
3031 int retval;
3032
3033 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3034 &retval, &ent->parent_de,
3035 &ent->dir_inlined);
3036 if (!ent->dir_bh)
3037 return retval;
3038 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3039 return -EIO;
3040 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3041 return ext4_journal_get_write_access(handle, ent->dir_bh);
3042 }
3043
3044 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3045 unsigned dir_ino)
3046 {
3047 int retval;
3048
3049 ent->parent_de->inode = cpu_to_le32(dir_ino);
3050 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3051 if (!ent->dir_inlined) {
3052 if (is_dx(ent->inode)) {
3053 retval = ext4_handle_dirty_dx_node(handle,
3054 ent->inode,
3055 ent->dir_bh);
3056 } else {
3057 retval = ext4_handle_dirty_dirent_node(handle,
3058 ent->inode,
3059 ent->dir_bh);
3060 }
3061 } else {
3062 retval = ext4_mark_inode_dirty(handle, ent->inode);
3063 }
3064 if (retval) {
3065 ext4_std_error(ent->dir->i_sb, retval);
3066 return retval;
3067 }
3068 return 0;
3069 }
3070
3071 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3072 unsigned ino, unsigned file_type)
3073 {
3074 int retval;
3075
3076 BUFFER_TRACE(ent->bh, "get write access");
3077 retval = ext4_journal_get_write_access(handle, ent->bh);
3078 if (retval)
3079 return retval;
3080 ent->de->inode = cpu_to_le32(ino);
3081 if (EXT4_HAS_INCOMPAT_FEATURE(ent->dir->i_sb,
3082 EXT4_FEATURE_INCOMPAT_FILETYPE))
3083 ent->de->file_type = file_type;
3084 ent->dir->i_version++;
3085 ent->dir->i_ctime = ent->dir->i_mtime =
3086 ext4_current_time(ent->dir);
3087 ext4_mark_inode_dirty(handle, ent->dir);
3088 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3089 if (!ent->inlined) {
3090 retval = ext4_handle_dirty_dirent_node(handle,
3091 ent->dir, ent->bh);
3092 if (unlikely(retval)) {
3093 ext4_std_error(ent->dir->i_sb, retval);
3094 return retval;
3095 }
3096 }
3097 brelse(ent->bh);
3098 ent->bh = NULL;
3099
3100 return 0;
3101 }
3102
3103 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3104 const struct qstr *d_name)
3105 {
3106 int retval = -ENOENT;
3107 struct buffer_head *bh;
3108 struct ext4_dir_entry_2 *de;
3109
3110 bh = ext4_find_entry(dir, d_name, &de, NULL);
3111 if (IS_ERR(bh))
3112 return PTR_ERR(bh);
3113 if (bh) {
3114 retval = ext4_delete_entry(handle, dir, de, bh);
3115 brelse(bh);
3116 }
3117 return retval;
3118 }
3119
3120 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3121 int force_reread)
3122 {
3123 int retval;
3124 /*
3125 * ent->de could have moved from under us during htree split, so make
3126 * sure that we are deleting the right entry. We might also be pointing
3127 * to a stale entry in the unused part of ent->bh so just checking inum
3128 * and the name isn't enough.
3129 */
3130 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3131 ent->de->name_len != ent->dentry->d_name.len ||
3132 strncmp(ent->de->name, ent->dentry->d_name.name,
3133 ent->de->name_len) ||
3134 force_reread) {
3135 retval = ext4_find_delete_entry(handle, ent->dir,
3136 &ent->dentry->d_name);
3137 } else {
3138 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3139 if (retval == -ENOENT) {
3140 retval = ext4_find_delete_entry(handle, ent->dir,
3141 &ent->dentry->d_name);
3142 }
3143 }
3144
3145 if (retval) {
3146 ext4_warning(ent->dir->i_sb,
3147 "Deleting old file (%lu), %d, error=%d",
3148 ent->dir->i_ino, ent->dir->i_nlink, retval);
3149 }
3150 }
3151
3152 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3153 {
3154 if (ent->dir_nlink_delta) {
3155 if (ent->dir_nlink_delta == -1)
3156 ext4_dec_count(handle, ent->dir);
3157 else
3158 ext4_inc_count(handle, ent->dir);
3159 ext4_mark_inode_dirty(handle, ent->dir);
3160 }
3161 }
3162
3163 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3164 int credits, handle_t **h)
3165 {
3166 struct inode *wh;
3167 handle_t *handle;
3168 int retries = 0;
3169
3170 /*
3171 * for inode block, sb block, group summaries,
3172 * and inode bitmap
3173 */
3174 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3175 EXT4_XATTR_TRANS_BLOCKS + 4);
3176 retry:
3177 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3178 &ent->dentry->d_name, 0, NULL,
3179 EXT4_HT_DIR, credits);
3180
3181 handle = ext4_journal_current_handle();
3182 if (IS_ERR(wh)) {
3183 if (handle)
3184 ext4_journal_stop(handle);
3185 if (PTR_ERR(wh) == -ENOSPC &&
3186 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3187 goto retry;
3188 } else {
3189 *h = handle;
3190 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3191 wh->i_op = &ext4_special_inode_operations;
3192 }
3193 return wh;
3194 }
3195
3196 /*
3197 * Anybody can rename anything with this: the permission checks are left to the
3198 * higher-level routines.
3199 *
3200 * n.b. old_{dentry,inode) refers to the source dentry/inode
3201 * while new_{dentry,inode) refers to the destination dentry/inode
3202 * This comes from rename(const char *oldpath, const char *newpath)
3203 */
3204 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3205 struct inode *new_dir, struct dentry *new_dentry,
3206 unsigned int flags)
3207 {
3208 handle_t *handle = NULL;
3209 struct ext4_renament old = {
3210 .dir = old_dir,
3211 .dentry = old_dentry,
3212 .inode = old_dentry->d_inode,
3213 };
3214 struct ext4_renament new = {
3215 .dir = new_dir,
3216 .dentry = new_dentry,
3217 .inode = new_dentry->d_inode,
3218 };
3219 int force_reread;
3220 int retval;
3221 struct inode *whiteout = NULL;
3222 int credits;
3223 u8 old_file_type;
3224
3225 dquot_initialize(old.dir);
3226 dquot_initialize(new.dir);
3227
3228 /* Initialize quotas before so that eventual writes go
3229 * in separate transaction */
3230 if (new.inode)
3231 dquot_initialize(new.inode);
3232
3233 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3234 if (IS_ERR(old.bh))
3235 return PTR_ERR(old.bh);
3236 /*
3237 * Check for inode number is _not_ due to possible IO errors.
3238 * We might rmdir the source, keep it as pwd of some process
3239 * and merrily kill the link to whatever was created under the
3240 * same name. Goodbye sticky bit ;-<
3241 */
3242 retval = -ENOENT;
3243 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3244 goto end_rename;
3245
3246 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3247 &new.de, &new.inlined);
3248 if (IS_ERR(new.bh)) {
3249 retval = PTR_ERR(new.bh);
3250 new.bh = NULL;
3251 goto end_rename;
3252 }
3253 if (new.bh) {
3254 if (!new.inode) {
3255 brelse(new.bh);
3256 new.bh = NULL;
3257 }
3258 }
3259 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3260 ext4_alloc_da_blocks(old.inode);
3261
3262 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3263 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3264 if (!(flags & RENAME_WHITEOUT)) {
3265 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3266 if (IS_ERR(handle)) {
3267 retval = PTR_ERR(handle);
3268 handle = NULL;
3269 goto end_rename;
3270 }
3271 } else {
3272 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3273 if (IS_ERR(whiteout)) {
3274 retval = PTR_ERR(whiteout);
3275 whiteout = NULL;
3276 goto end_rename;
3277 }
3278 }
3279
3280 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3281 ext4_handle_sync(handle);
3282
3283 if (S_ISDIR(old.inode->i_mode)) {
3284 if (new.inode) {
3285 retval = -ENOTEMPTY;
3286 if (!empty_dir(new.inode))
3287 goto end_rename;
3288 } else {
3289 retval = -EMLINK;
3290 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3291 goto end_rename;
3292 }
3293 retval = ext4_rename_dir_prepare(handle, &old);
3294 if (retval)
3295 goto end_rename;
3296 }
3297 /*
3298 * If we're renaming a file within an inline_data dir and adding or
3299 * setting the new dirent causes a conversion from inline_data to
3300 * extents/blockmap, we need to force the dirent delete code to
3301 * re-read the directory, or else we end up trying to delete a dirent
3302 * from what is now the extent tree root (or a block map).
3303 */
3304 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3305 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3306
3307 old_file_type = old.de->file_type;
3308 if (whiteout) {
3309 /*
3310 * Do this before adding a new entry, so the old entry is sure
3311 * to be still pointing to the valid old entry.
3312 */
3313 retval = ext4_setent(handle, &old, whiteout->i_ino,
3314 EXT4_FT_CHRDEV);
3315 if (retval)
3316 goto end_rename;
3317 ext4_mark_inode_dirty(handle, whiteout);
3318 }
3319 if (!new.bh) {
3320 retval = ext4_add_entry(handle, new.dentry, old.inode);
3321 if (retval)
3322 goto end_rename;
3323 } else {
3324 retval = ext4_setent(handle, &new,
3325 old.inode->i_ino, old_file_type);
3326 if (retval)
3327 goto end_rename;
3328 }
3329 if (force_reread)
3330 force_reread = !ext4_test_inode_flag(new.dir,
3331 EXT4_INODE_INLINE_DATA);
3332
3333 /*
3334 * Like most other Unix systems, set the ctime for inodes on a
3335 * rename.
3336 */
3337 old.inode->i_ctime = ext4_current_time(old.inode);
3338 ext4_mark_inode_dirty(handle, old.inode);
3339
3340 if (!whiteout) {
3341 /*
3342 * ok, that's it
3343 */
3344 ext4_rename_delete(handle, &old, force_reread);
3345 }
3346
3347 if (new.inode) {
3348 ext4_dec_count(handle, new.inode);
3349 new.inode->i_ctime = ext4_current_time(new.inode);
3350 }
3351 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3352 ext4_update_dx_flag(old.dir);
3353 if (old.dir_bh) {
3354 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3355 if (retval)
3356 goto end_rename;
3357
3358 ext4_dec_count(handle, old.dir);
3359 if (new.inode) {
3360 /* checked empty_dir above, can't have another parent,
3361 * ext4_dec_count() won't work for many-linked dirs */
3362 clear_nlink(new.inode);
3363 } else {
3364 ext4_inc_count(handle, new.dir);
3365 ext4_update_dx_flag(new.dir);
3366 ext4_mark_inode_dirty(handle, new.dir);
3367 }
3368 }
3369 ext4_mark_inode_dirty(handle, old.dir);
3370 if (new.inode) {
3371 ext4_mark_inode_dirty(handle, new.inode);
3372 if (!new.inode->i_nlink)
3373 ext4_orphan_add(handle, new.inode);
3374 }
3375 retval = 0;
3376
3377 end_rename:
3378 brelse(old.dir_bh);
3379 brelse(old.bh);
3380 brelse(new.bh);
3381 if (whiteout) {
3382 if (retval)
3383 drop_nlink(whiteout);
3384 unlock_new_inode(whiteout);
3385 iput(whiteout);
3386 }
3387 if (handle)
3388 ext4_journal_stop(handle);
3389 return retval;
3390 }
3391
3392 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3393 struct inode *new_dir, struct dentry *new_dentry)
3394 {
3395 handle_t *handle = NULL;
3396 struct ext4_renament old = {
3397 .dir = old_dir,
3398 .dentry = old_dentry,
3399 .inode = old_dentry->d_inode,
3400 };
3401 struct ext4_renament new = {
3402 .dir = new_dir,
3403 .dentry = new_dentry,
3404 .inode = new_dentry->d_inode,
3405 };
3406 u8 new_file_type;
3407 int retval;
3408
3409 dquot_initialize(old.dir);
3410 dquot_initialize(new.dir);
3411
3412 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3413 &old.de, &old.inlined);
3414 if (IS_ERR(old.bh))
3415 return PTR_ERR(old.bh);
3416 /*
3417 * Check for inode number is _not_ due to possible IO errors.
3418 * We might rmdir the source, keep it as pwd of some process
3419 * and merrily kill the link to whatever was created under the
3420 * same name. Goodbye sticky bit ;-<
3421 */
3422 retval = -ENOENT;
3423 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3424 goto end_rename;
3425
3426 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3427 &new.de, &new.inlined);
3428 if (IS_ERR(new.bh)) {
3429 retval = PTR_ERR(new.bh);
3430 new.bh = NULL;
3431 goto end_rename;
3432 }
3433
3434 /* RENAME_EXCHANGE case: old *and* new must both exist */
3435 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3436 goto end_rename;
3437
3438 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3439 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3440 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3441 if (IS_ERR(handle)) {
3442 retval = PTR_ERR(handle);
3443 handle = NULL;
3444 goto end_rename;
3445 }
3446
3447 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3448 ext4_handle_sync(handle);
3449
3450 if (S_ISDIR(old.inode->i_mode)) {
3451 old.is_dir = true;
3452 retval = ext4_rename_dir_prepare(handle, &old);
3453 if (retval)
3454 goto end_rename;
3455 }
3456 if (S_ISDIR(new.inode->i_mode)) {
3457 new.is_dir = true;
3458 retval = ext4_rename_dir_prepare(handle, &new);
3459 if (retval)
3460 goto end_rename;
3461 }
3462
3463 /*
3464 * Other than the special case of overwriting a directory, parents'
3465 * nlink only needs to be modified if this is a cross directory rename.
3466 */
3467 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3468 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3469 new.dir_nlink_delta = -old.dir_nlink_delta;
3470 retval = -EMLINK;
3471 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3472 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3473 goto end_rename;
3474 }
3475
3476 new_file_type = new.de->file_type;
3477 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3478 if (retval)
3479 goto end_rename;
3480
3481 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3482 if (retval)
3483 goto end_rename;
3484
3485 /*
3486 * Like most other Unix systems, set the ctime for inodes on a
3487 * rename.
3488 */
3489 old.inode->i_ctime = ext4_current_time(old.inode);
3490 new.inode->i_ctime = ext4_current_time(new.inode);
3491 ext4_mark_inode_dirty(handle, old.inode);
3492 ext4_mark_inode_dirty(handle, new.inode);
3493
3494 if (old.dir_bh) {
3495 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3496 if (retval)
3497 goto end_rename;
3498 }
3499 if (new.dir_bh) {
3500 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3501 if (retval)
3502 goto end_rename;
3503 }
3504 ext4_update_dir_count(handle, &old);
3505 ext4_update_dir_count(handle, &new);
3506 retval = 0;
3507
3508 end_rename:
3509 brelse(old.dir_bh);
3510 brelse(new.dir_bh);
3511 brelse(old.bh);
3512 brelse(new.bh);
3513 if (handle)
3514 ext4_journal_stop(handle);
3515 return retval;
3516 }
3517
3518 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3519 struct inode *new_dir, struct dentry *new_dentry,
3520 unsigned int flags)
3521 {
3522 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3523 return -EINVAL;
3524
3525 if (flags & RENAME_EXCHANGE) {
3526 return ext4_cross_rename(old_dir, old_dentry,
3527 new_dir, new_dentry);
3528 }
3529
3530 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3531 }
3532
3533 /*
3534 * directories can handle most operations...
3535 */
3536 const struct inode_operations ext4_dir_inode_operations = {
3537 .create = ext4_create,
3538 .lookup = ext4_lookup,
3539 .link = ext4_link,
3540 .unlink = ext4_unlink,
3541 .symlink = ext4_symlink,
3542 .mkdir = ext4_mkdir,
3543 .rmdir = ext4_rmdir,
3544 .mknod = ext4_mknod,
3545 .tmpfile = ext4_tmpfile,
3546 .rename2 = ext4_rename2,
3547 .setattr = ext4_setattr,
3548 .setxattr = generic_setxattr,
3549 .getxattr = generic_getxattr,
3550 .listxattr = ext4_listxattr,
3551 .removexattr = generic_removexattr,
3552 .get_acl = ext4_get_acl,
3553 .set_acl = ext4_set_acl,
3554 .fiemap = ext4_fiemap,
3555 };
3556
3557 const struct inode_operations ext4_special_inode_operations = {
3558 .setattr = ext4_setattr,
3559 .setxattr = generic_setxattr,
3560 .getxattr = generic_getxattr,
3561 .listxattr = ext4_listxattr,
3562 .removexattr = generic_removexattr,
3563 .get_acl = ext4_get_acl,
3564 .set_acl = ext4_set_acl,
3565 };
Linux kernel - Deliverables
This page took 0.1167 seconds and 5 git commands to generate.