?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.12.28
[linux/fpc-iii.git] / fs / ext4 / namei.c
blob5a0408d7b1147094c3e82b6d11750b33396b7732
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/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
39
40 #include "xattr.h"
41 #include "acl.h"
42
43 #include <trace/events/ext4.h>
44 /*
45 * define how far ahead to read directories while searching them.
46 */
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50
51 static struct buffer_head *ext4_append(handle_t *handle,
52 struct inode *inode,
53 ext4_lblk_t *block)
54 {
55 struct buffer_head *bh;
56 int err = 0;
57
58 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 ((inode->i_size >> 10) >=
60 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
61 return ERR_PTR(-ENOSPC);
62
63 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64
65 bh = ext4_bread(handle, inode, *block, 1, &err);
66 if (!bh)
67 return ERR_PTR(err);
68 inode->i_size += inode->i_sb->s_blocksize;
69 EXT4_I(inode)->i_disksize = inode->i_size;
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 err = 0, is_dx_block = 0;
97
98 bh = ext4_bread(NULL, inode, block, 0, &err);
99 if (!bh) {
100 if (err == 0) {
101 ext4_error_inode(inode, __func__, line, block,
102 "Directory hole found");
103 return ERR_PTR(-EIO);
104 }
105 __ext4_warning(inode->i_sb, __func__, line,
106 "error reading directory block "
107 "(ino %lu, block %lu)", inode->i_ino,
108 (unsigned long) block);
109 return ERR_PTR(err);
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_RO_COMPAT_FEATURE(inode->i_sb,
127 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) ||
128 buffer_verified(bh))
129 return bh;
130
131 /*
132 * An empty leaf block can get mistaken for a index block; for
133 * this reason, we can only check the index checksum when the
134 * caller is sure it should be an index block.
135 */
136 if (is_dx_block && type == INDEX) {
137 if (ext4_dx_csum_verify(inode, dirent))
138 set_buffer_verified(bh);
139 else {
140 ext4_error_inode(inode, __func__, line, block,
141 "Directory index failed checksum");
142 brelse(bh);
143 return ERR_PTR(-EIO);
144 }
145 }
146 if (!is_dx_block) {
147 if (ext4_dirent_csum_verify(inode, dirent))
148 set_buffer_verified(bh);
149 else {
150 ext4_error_inode(inode, __func__, line, block,
151 "Directory block failed checksum");
152 brelse(bh);
153 return ERR_PTR(-EIO);
154 }
155 }
156 return bh;
157 }
158
159 #ifndef assert
160 #define assert(test) J_ASSERT(test)
161 #endif
162
163 #ifdef DX_DEBUG
164 #define dxtrace(command) command
165 #else
166 #define dxtrace(command)
167 #endif
168
169 struct fake_dirent
170 {
171 __le32 inode;
172 __le16 rec_len;
173 u8 name_len;
174 u8 file_type;
175 };
176
177 struct dx_countlimit
178 {
179 __le16 limit;
180 __le16 count;
181 };
182
183 struct dx_entry
184 {
185 __le32 hash;
186 __le32 block;
187 };
188
189 /*
190 * dx_root_info is laid out so that if it should somehow get overlaid by a
191 * dirent the two low bits of the hash version will be zero. Therefore, the
192 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
193 */
194
195 struct dx_root
196 {
197 struct fake_dirent dot;
198 char dot_name[4];
199 struct fake_dirent dotdot;
200 char dotdot_name[4];
201 struct dx_root_info
202 {
203 __le32 reserved_zero;
204 u8 hash_version;
205 u8 info_length; /* 8 */
206 u8 indirect_levels;
207 u8 unused_flags;
208 }
209 info;
210 struct dx_entry entries[0];
211 };
212
213 struct dx_node
214 {
215 struct fake_dirent fake;
216 struct dx_entry entries[0];
217 };
218
219
220 struct dx_frame
221 {
222 struct buffer_head *bh;
223 struct dx_entry *entries;
224 struct dx_entry *at;
225 };
226
227 struct dx_map_entry
228 {
229 u32 hash;
230 u16 offs;
231 u16 size;
232 };
233
234 /*
235 * This goes at the end of each htree block.
236 */
237 struct dx_tail {
238 u32 dt_reserved;
239 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
240 };
241
242 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
243 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
244 static inline unsigned dx_get_hash(struct dx_entry *entry);
245 static void dx_set_hash(struct dx_entry *entry, unsigned value);
246 static unsigned dx_get_count(struct dx_entry *entries);
247 static unsigned dx_get_limit(struct dx_entry *entries);
248 static void dx_set_count(struct dx_entry *entries, unsigned value);
249 static void dx_set_limit(struct dx_entry *entries, unsigned value);
250 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
251 static unsigned dx_node_limit(struct inode *dir);
252 static struct dx_frame *dx_probe(const struct qstr *d_name,
253 struct inode *dir,
254 struct dx_hash_info *hinfo,
255 struct dx_frame *frame,
256 int *err);
257 static void dx_release(struct dx_frame *frames);
258 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
259 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
260 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
261 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
262 struct dx_map_entry *offsets, int count, unsigned blocksize);
263 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
264 static void dx_insert_block(struct dx_frame *frame,
265 u32 hash, ext4_lblk_t block);
266 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
267 struct dx_frame *frame,
268 struct dx_frame *frames,
269 __u32 *start_hash);
270 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
271 const struct qstr *d_name,
272 struct ext4_dir_entry_2 **res_dir,
273 int *err);
274 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
275 struct inode *inode);
276
277 /* checksumming functions */
278 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
279 unsigned int blocksize)
280 {
281 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
282 t->det_rec_len = ext4_rec_len_to_disk(
283 sizeof(struct ext4_dir_entry_tail), blocksize);
284 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
285 }
286
287 /* Walk through a dirent block to find a checksum "dirent" at the tail */
288 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
289 struct ext4_dir_entry *de)
290 {
291 struct ext4_dir_entry_tail *t;
292
293 #ifdef PARANOID
294 struct ext4_dir_entry *d, *top;
295
296 d = de;
297 top = (struct ext4_dir_entry *)(((void *)de) +
298 (EXT4_BLOCK_SIZE(inode->i_sb) -
299 sizeof(struct ext4_dir_entry_tail)));
300 while (d < top && d->rec_len)
301 d = (struct ext4_dir_entry *)(((void *)d) +
302 le16_to_cpu(d->rec_len));
303
304 if (d != top)
305 return NULL;
306
307 t = (struct ext4_dir_entry_tail *)d;
308 #else
309 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
310 #endif
311
312 if (t->det_reserved_zero1 ||
313 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
314 t->det_reserved_zero2 ||
315 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
316 return NULL;
317
318 return t;
319 }
320
321 static __le32 ext4_dirent_csum(struct inode *inode,
322 struct ext4_dir_entry *dirent, int size)
323 {
324 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
325 struct ext4_inode_info *ei = EXT4_I(inode);
326 __u32 csum;
327
328 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
329 return cpu_to_le32(csum);
330 }
331
332 static void warn_no_space_for_csum(struct inode *inode)
333 {
334 ext4_warning(inode->i_sb, "no space in directory inode %lu leaf for "
335 "checksum. Please run e2fsck -D.", inode->i_ino);
336 }
337
338 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
339 {
340 struct ext4_dir_entry_tail *t;
341
342 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
343 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
344 return 1;
345
346 t = get_dirent_tail(inode, dirent);
347 if (!t) {
348 warn_no_space_for_csum(inode);
349 return 0;
350 }
351
352 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
353 (void *)t - (void *)dirent))
354 return 0;
355
356 return 1;
357 }
358
359 static void ext4_dirent_csum_set(struct inode *inode,
360 struct ext4_dir_entry *dirent)
361 {
362 struct ext4_dir_entry_tail *t;
363
364 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
365 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
366 return;
367
368 t = get_dirent_tail(inode, dirent);
369 if (!t) {
370 warn_no_space_for_csum(inode);
371 return;
372 }
373
374 t->det_checksum = ext4_dirent_csum(inode, dirent,
375 (void *)t - (void *)dirent);
376 }
377
378 int ext4_handle_dirty_dirent_node(handle_t *handle,
379 struct inode *inode,
380 struct buffer_head *bh)
381 {
382 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
383 return ext4_handle_dirty_metadata(handle, inode, bh);
384 }
385
386 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
387 struct ext4_dir_entry *dirent,
388 int *offset)
389 {
390 struct ext4_dir_entry *dp;
391 struct dx_root_info *root;
392 int count_offset;
393
394 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
395 count_offset = 8;
396 else if (le16_to_cpu(dirent->rec_len) == 12) {
397 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
398 if (le16_to_cpu(dp->rec_len) !=
399 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
400 return NULL;
401 root = (struct dx_root_info *)(((void *)dp + 12));
402 if (root->reserved_zero ||
403 root->info_length != sizeof(struct dx_root_info))
404 return NULL;
405 count_offset = 32;
406 } else
407 return NULL;
408
409 if (offset)
410 *offset = count_offset;
411 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
412 }
413
414 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
415 int count_offset, int count, struct dx_tail *t)
416 {
417 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
418 struct ext4_inode_info *ei = EXT4_I(inode);
419 __u32 csum;
420 __le32 save_csum;
421 int size;
422
423 size = count_offset + (count * sizeof(struct dx_entry));
424 save_csum = t->dt_checksum;
425 t->dt_checksum = 0;
426 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
427 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
428 t->dt_checksum = save_csum;
429
430 return cpu_to_le32(csum);
431 }
432
433 static int ext4_dx_csum_verify(struct inode *inode,
434 struct ext4_dir_entry *dirent)
435 {
436 struct dx_countlimit *c;
437 struct dx_tail *t;
438 int count_offset, limit, count;
439
440 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
441 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
442 return 1;
443
444 c = get_dx_countlimit(inode, dirent, &count_offset);
445 if (!c) {
446 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
447 return 1;
448 }
449 limit = le16_to_cpu(c->limit);
450 count = le16_to_cpu(c->count);
451 if (count_offset + (limit * sizeof(struct dx_entry)) >
452 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
453 warn_no_space_for_csum(inode);
454 return 1;
455 }
456 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
457
458 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
459 count, t))
460 return 0;
461 return 1;
462 }
463
464 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
465 {
466 struct dx_countlimit *c;
467 struct dx_tail *t;
468 int count_offset, limit, count;
469
470 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
471 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
472 return;
473
474 c = get_dx_countlimit(inode, dirent, &count_offset);
475 if (!c) {
476 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
477 return;
478 }
479 limit = le16_to_cpu(c->limit);
480 count = le16_to_cpu(c->count);
481 if (count_offset + (limit * sizeof(struct dx_entry)) >
482 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483 warn_no_space_for_csum(inode);
484 return;
485 }
486 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
487
488 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
489 }
490
491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
492 struct inode *inode,
493 struct buffer_head *bh)
494 {
495 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 return ext4_handle_dirty_metadata(handle, inode, bh);
497 }
498
499 /*
500 * p is at least 6 bytes before the end of page
501 */
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
504 {
505 return (struct ext4_dir_entry_2 *)((char *)p +
506 ext4_rec_len_from_disk(p->rec_len, blocksize));
507 }
508
509 /*
510 * Future: use high four bits of block for coalesce-on-delete flags
511 * Mask them off for now.
512 */
513
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
515 {
516 return le32_to_cpu(entry->block) & 0x00ffffff;
517 }
518
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
520 {
521 entry->block = cpu_to_le32(value);
522 }
523
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
525 {
526 return le32_to_cpu(entry->hash);
527 }
528
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
530 {
531 entry->hash = cpu_to_le32(value);
532 }
533
534 static inline unsigned dx_get_count(struct dx_entry *entries)
535 {
536 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537 }
538
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
540 {
541 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542 }
543
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
545 {
546 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547 }
548
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
550 {
551 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552 }
553
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
555 {
556 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 EXT4_DIR_REC_LEN(2) - infosize;
558
559 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
560 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
561 entry_space -= sizeof(struct dx_tail);
562 return entry_space / sizeof(struct dx_entry);
563 }
564
565 static inline unsigned dx_node_limit(struct inode *dir)
566 {
567 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568
569 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
570 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
571 entry_space -= sizeof(struct dx_tail);
572 return entry_space / sizeof(struct dx_entry);
573 }
574
575 /*
576 * Debug
577 */
578 #ifdef DX_DEBUG
579 static void dx_show_index(char * label, struct dx_entry *entries)
580 {
581 int i, n = dx_get_count (entries);
582 printk(KERN_DEBUG "%s index ", label);
583 for (i = 0; i < n; i++) {
584 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
585 0, (unsigned long)dx_get_block(entries + i));
586 }
587 printk("\n");
588 }
589
590 struct stats
591 {
592 unsigned names;
593 unsigned space;
594 unsigned bcount;
595 };
596
597 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
598 int size, int show_names)
599 {
600 unsigned names = 0, space = 0;
601 char *base = (char *) de;
602 struct dx_hash_info h = *hinfo;
603
604 printk("names: ");
605 while ((char *) de < base + size)
606 {
607 if (de->inode)
608 {
609 if (show_names)
610 {
611 int len = de->name_len;
612 char *name = de->name;
613 while (len--) printk("%c", *name++);
614 ext4fs_dirhash(de->name, de->name_len, &h);
615 printk(":%x.%u ", h.hash,
616 (unsigned) ((char *) de - base));
617 }
618 space += EXT4_DIR_REC_LEN(de->name_len);
619 names++;
620 }
621 de = ext4_next_entry(de, size);
622 }
623 printk("(%i)\n", names);
624 return (struct stats) { names, space, 1 };
625 }
626
627 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
628 struct dx_entry *entries, int levels)
629 {
630 unsigned blocksize = dir->i_sb->s_blocksize;
631 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
632 unsigned bcount = 0;
633 struct buffer_head *bh;
634 int err;
635 printk("%i indexed blocks...\n", count);
636 for (i = 0; i < count; i++, entries++)
637 {
638 ext4_lblk_t block = dx_get_block(entries);
639 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
640 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
641 struct stats stats;
642 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
643 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
644 stats = levels?
645 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
646 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
647 names += stats.names;
648 space += stats.space;
649 bcount += stats.bcount;
650 brelse(bh);
651 }
652 if (bcount)
653 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
654 levels ? "" : " ", names, space/bcount,
655 (space/bcount)*100/blocksize);
656 return (struct stats) { names, space, bcount};
657 }
658 #endif /* DX_DEBUG */
659
660 /*
661 * Probe for a directory leaf block to search.
662 *
663 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
664 * error in the directory index, and the caller should fall back to
665 * searching the directory normally. The callers of dx_probe **MUST**
666 * check for this error code, and make sure it never gets reflected
667 * back to userspace.
668 */
669 static struct dx_frame *
670 dx_probe(const struct qstr *d_name, struct inode *dir,
671 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
672 {
673 unsigned count, indirect;
674 struct dx_entry *at, *entries, *p, *q, *m;
675 struct dx_root *root;
676 struct buffer_head *bh;
677 struct dx_frame *frame = frame_in;
678 u32 hash;
679
680 frame->bh = NULL;
681 bh = ext4_read_dirblock(dir, 0, INDEX);
682 if (IS_ERR(bh)) {
683 *err = PTR_ERR(bh);
684 goto fail;
685 }
686 root = (struct dx_root *) bh->b_data;
687 if (root->info.hash_version != DX_HASH_TEA &&
688 root->info.hash_version != DX_HASH_HALF_MD4 &&
689 root->info.hash_version != DX_HASH_LEGACY) {
690 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
691 root->info.hash_version);
692 brelse(bh);
693 *err = ERR_BAD_DX_DIR;
694 goto fail;
695 }
696 hinfo->hash_version = root->info.hash_version;
697 if (hinfo->hash_version <= DX_HASH_TEA)
698 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
699 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
700 if (d_name)
701 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
702 hash = hinfo->hash;
703
704 if (root->info.unused_flags & 1) {
705 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
706 root->info.unused_flags);
707 brelse(bh);
708 *err = ERR_BAD_DX_DIR;
709 goto fail;
710 }
711
712 if ((indirect = root->info.indirect_levels) > 1) {
713 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
714 root->info.indirect_levels);
715 brelse(bh);
716 *err = ERR_BAD_DX_DIR;
717 goto fail;
718 }
719
720 entries = (struct dx_entry *) (((char *)&root->info) +
721 root->info.info_length);
722
723 if (dx_get_limit(entries) != dx_root_limit(dir,
724 root->info.info_length)) {
725 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
726 brelse(bh);
727 *err = ERR_BAD_DX_DIR;
728 goto fail;
729 }
730
731 dxtrace(printk("Look up %x", hash));
732 while (1)
733 {
734 count = dx_get_count(entries);
735 if (!count || count > dx_get_limit(entries)) {
736 ext4_warning(dir->i_sb,
737 "dx entry: no count or count > limit");
738 brelse(bh);
739 *err = ERR_BAD_DX_DIR;
740 goto fail2;
741 }
742
743 p = entries + 1;
744 q = entries + count - 1;
745 while (p <= q)
746 {
747 m = p + (q - p)/2;
748 dxtrace(printk("."));
749 if (dx_get_hash(m) > hash)
750 q = m - 1;
751 else
752 p = m + 1;
753 }
754
755 if (0) // linear search cross check
756 {
757 unsigned n = count - 1;
758 at = entries;
759 while (n--)
760 {
761 dxtrace(printk(","));
762 if (dx_get_hash(++at) > hash)
763 {
764 at--;
765 break;
766 }
767 }
768 assert (at == p - 1);
769 }
770
771 at = p - 1;
772 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
773 frame->bh = bh;
774 frame->entries = entries;
775 frame->at = at;
776 if (!indirect--) return frame;
777 bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
778 if (IS_ERR(bh)) {
779 *err = PTR_ERR(bh);
780 goto fail2;
781 }
782 entries = ((struct dx_node *) bh->b_data)->entries;
783
784 if (dx_get_limit(entries) != dx_node_limit (dir)) {
785 ext4_warning(dir->i_sb,
786 "dx entry: limit != node limit");
787 brelse(bh);
788 *err = ERR_BAD_DX_DIR;
789 goto fail2;
790 }
791 frame++;
792 frame->bh = NULL;
793 }
794 fail2:
795 while (frame >= frame_in) {
796 brelse(frame->bh);
797 frame--;
798 }
799 fail:
800 if (*err == ERR_BAD_DX_DIR)
801 ext4_warning(dir->i_sb,
802 "Corrupt dir inode %lu, running e2fsck is "
803 "recommended.", dir->i_ino);
804 return NULL;
805 }
806
807 static void dx_release (struct dx_frame *frames)
808 {
809 if (frames[0].bh == NULL)
810 return;
811
812 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
813 brelse(frames[1].bh);
814 brelse(frames[0].bh);
815 }
816
817 /*
818 * This function increments the frame pointer to search the next leaf
819 * block, and reads in the necessary intervening nodes if the search
820 * should be necessary. Whether or not the search is necessary is
821 * controlled by the hash parameter. If the hash value is even, then
822 * the search is only continued if the next block starts with that
823 * hash value. This is used if we are searching for a specific file.
824 *
825 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
826 *
827 * This function returns 1 if the caller should continue to search,
828 * or 0 if it should not. If there is an error reading one of the
829 * index blocks, it will a negative error code.
830 *
831 * If start_hash is non-null, it will be filled in with the starting
832 * hash of the next page.
833 */
834 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
835 struct dx_frame *frame,
836 struct dx_frame *frames,
837 __u32 *start_hash)
838 {
839 struct dx_frame *p;
840 struct buffer_head *bh;
841 int num_frames = 0;
842 __u32 bhash;
843
844 p = frame;
845 /*
846 * Find the next leaf page by incrementing the frame pointer.
847 * If we run out of entries in the interior node, loop around and
848 * increment pointer in the parent node. When we break out of
849 * this loop, num_frames indicates the number of interior
850 * nodes need to be read.
851 */
852 while (1) {
853 if (++(p->at) < p->entries + dx_get_count(p->entries))
854 break;
855 if (p == frames)
856 return 0;
857 num_frames++;
858 p--;
859 }
860
861 /*
862 * If the hash is 1, then continue only if the next page has a
863 * continuation hash of any value. This is used for readdir
864 * handling. Otherwise, check to see if the hash matches the
865 * desired contiuation hash. If it doesn't, return since
866 * there's no point to read in the successive index pages.
867 */
868 bhash = dx_get_hash(p->at);
869 if (start_hash)
870 *start_hash = bhash;
871 if ((hash & 1) == 0) {
872 if ((bhash & ~1) != hash)
873 return 0;
874 }
875 /*
876 * If the hash is HASH_NB_ALWAYS, we always go to the next
877 * block so no check is necessary
878 */
879 while (num_frames--) {
880 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
881 if (IS_ERR(bh))
882 return PTR_ERR(bh);
883 p++;
884 brelse(p->bh);
885 p->bh = bh;
886 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
887 }
888 return 1;
889 }
890
891
892 /*
893 * This function fills a red-black tree with information from a
894 * directory block. It returns the number directory entries loaded
895 * into the tree. If there is an error it is returned in err.
896 */
897 static int htree_dirblock_to_tree(struct file *dir_file,
898 struct inode *dir, ext4_lblk_t block,
899 struct dx_hash_info *hinfo,
900 __u32 start_hash, __u32 start_minor_hash)
901 {
902 struct buffer_head *bh;
903 struct ext4_dir_entry_2 *de, *top;
904 int err = 0, count = 0;
905
906 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
907 (unsigned long)block));
908 bh = ext4_read_dirblock(dir, block, DIRENT);
909 if (IS_ERR(bh))
910 return PTR_ERR(bh);
911
912 de = (struct ext4_dir_entry_2 *) bh->b_data;
913 top = (struct ext4_dir_entry_2 *) ((char *) de +
914 dir->i_sb->s_blocksize -
915 EXT4_DIR_REC_LEN(0));
916 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
917 if (ext4_check_dir_entry(dir, NULL, de, bh,
918 bh->b_data, bh->b_size,
919 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
920 + ((char *)de - bh->b_data))) {
921 /* silently ignore the rest of the block */
922 break;
923 }
924 ext4fs_dirhash(de->name, de->name_len, hinfo);
925 if ((hinfo->hash < start_hash) ||
926 ((hinfo->hash == start_hash) &&
927 (hinfo->minor_hash < start_minor_hash)))
928 continue;
929 if (de->inode == 0)
930 continue;
931 if ((err = ext4_htree_store_dirent(dir_file,
932 hinfo->hash, hinfo->minor_hash, de)) != 0) {
933 brelse(bh);
934 return err;
935 }
936 count++;
937 }
938 brelse(bh);
939 return count;
940 }
941
942
943 /*
944 * This function fills a red-black tree with information from a
945 * directory. We start scanning the directory in hash order, starting
946 * at start_hash and start_minor_hash.
947 *
948 * This function returns the number of entries inserted into the tree,
949 * or a negative error code.
950 */
951 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
952 __u32 start_minor_hash, __u32 *next_hash)
953 {
954 struct dx_hash_info hinfo;
955 struct ext4_dir_entry_2 *de;
956 struct dx_frame frames[2], *frame;
957 struct inode *dir;
958 ext4_lblk_t block;
959 int count = 0;
960 int ret, err;
961 __u32 hashval;
962
963 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
964 start_hash, start_minor_hash));
965 dir = file_inode(dir_file);
966 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
967 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
968 if (hinfo.hash_version <= DX_HASH_TEA)
969 hinfo.hash_version +=
970 EXT4_SB(dir->i_sb)->s_hash_unsigned;
971 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
972 if (ext4_has_inline_data(dir)) {
973 int has_inline_data = 1;
974 count = htree_inlinedir_to_tree(dir_file, dir, 0,
975 &hinfo, start_hash,
976 start_minor_hash,
977 &has_inline_data);
978 if (has_inline_data) {
979 *next_hash = ~0;
980 return count;
981 }
982 }
983 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
984 start_hash, start_minor_hash);
985 *next_hash = ~0;
986 return count;
987 }
988 hinfo.hash = start_hash;
989 hinfo.minor_hash = 0;
990 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
991 if (!frame)
992 return err;
993
994 /* Add '.' and '..' from the htree header */
995 if (!start_hash && !start_minor_hash) {
996 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
997 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
998 goto errout;
999 count++;
1000 }
1001 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1002 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1003 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1004 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
1005 goto errout;
1006 count++;
1007 }
1008
1009 while (1) {
1010 block = dx_get_block(frame->at);
1011 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1012 start_hash, start_minor_hash);
1013 if (ret < 0) {
1014 err = ret;
1015 goto errout;
1016 }
1017 count += ret;
1018 hashval = ~0;
1019 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1020 frame, frames, &hashval);
1021 *next_hash = hashval;
1022 if (ret < 0) {
1023 err = ret;
1024 goto errout;
1025 }
1026 /*
1027 * Stop if: (a) there are no more entries, or
1028 * (b) we have inserted at least one entry and the
1029 * next hash value is not a continuation
1030 */
1031 if ((ret == 0) ||
1032 (count && ((hashval & 1) == 0)))
1033 break;
1034 }
1035 dx_release(frames);
1036 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1037 "next hash: %x\n", count, *next_hash));
1038 return count;
1039 errout:
1040 dx_release(frames);
1041 return (err);
1042 }
1043
1044 static inline int search_dirblock(struct buffer_head *bh,
1045 struct inode *dir,
1046 const struct qstr *d_name,
1047 unsigned int offset,
1048 struct ext4_dir_entry_2 **res_dir)
1049 {
1050 return search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1051 d_name, offset, res_dir);
1052 }
1053
1054 /*
1055 * Directory block splitting, compacting
1056 */
1057
1058 /*
1059 * Create map of hash values, offsets, and sizes, stored at end of block.
1060 * Returns number of entries mapped.
1061 */
1062 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
1063 struct dx_hash_info *hinfo,
1064 struct dx_map_entry *map_tail)
1065 {
1066 int count = 0;
1067 char *base = (char *) de;
1068 struct dx_hash_info h = *hinfo;
1069
1070 while ((char *) de < base + blocksize) {
1071 if (de->name_len && de->inode) {
1072 ext4fs_dirhash(de->name, de->name_len, &h);
1073 map_tail--;
1074 map_tail->hash = h.hash;
1075 map_tail->offs = ((char *) de - base)>>2;
1076 map_tail->size = le16_to_cpu(de->rec_len);
1077 count++;
1078 cond_resched();
1079 }
1080 /* XXX: do we need to check rec_len == 0 case? -Chris */
1081 de = ext4_next_entry(de, blocksize);
1082 }
1083 return count;
1084 }
1085
1086 /* Sort map by hash value */
1087 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1088 {
1089 struct dx_map_entry *p, *q, *top = map + count - 1;
1090 int more;
1091 /* Combsort until bubble sort doesn't suck */
1092 while (count > 2) {
1093 count = count*10/13;
1094 if (count - 9 < 2) /* 9, 10 -> 11 */
1095 count = 11;
1096 for (p = top, q = p - count; q >= map; p--, q--)
1097 if (p->hash < q->hash)
1098 swap(*p, *q);
1099 }
1100 /* Garden variety bubble sort */
1101 do {
1102 more = 0;
1103 q = top;
1104 while (q-- > map) {
1105 if (q[1].hash >= q[0].hash)
1106 continue;
1107 swap(*(q+1), *q);
1108 more = 1;
1109 }
1110 } while(more);
1111 }
1112
1113 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1114 {
1115 struct dx_entry *entries = frame->entries;
1116 struct dx_entry *old = frame->at, *new = old + 1;
1117 int count = dx_get_count(entries);
1118
1119 assert(count < dx_get_limit(entries));
1120 assert(old < entries + count);
1121 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1122 dx_set_hash(new, hash);
1123 dx_set_block(new, block);
1124 dx_set_count(entries, count + 1);
1125 }
1126
1127 /*
1128 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1129 *
1130 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1131 * `de != NULL' is guaranteed by caller.
1132 */
1133 static inline int ext4_match (int len, const char * const name,
1134 struct ext4_dir_entry_2 * de)
1135 {
1136 if (len != de->name_len)
1137 return 0;
1138 if (!de->inode)
1139 return 0;
1140 return !memcmp(name, de->name, len);
1141 }
1142
1143 /*
1144 * Returns 0 if not found, -1 on failure, and 1 on success
1145 */
1146 int search_dir(struct buffer_head *bh,
1147 char *search_buf,
1148 int buf_size,
1149 struct inode *dir,
1150 const struct qstr *d_name,
1151 unsigned int offset,
1152 struct ext4_dir_entry_2 **res_dir)
1153 {
1154 struct ext4_dir_entry_2 * de;
1155 char * dlimit;
1156 int de_len;
1157 const char *name = d_name->name;
1158 int namelen = d_name->len;
1159
1160 de = (struct ext4_dir_entry_2 *)search_buf;
1161 dlimit = search_buf + buf_size;
1162 while ((char *) de < dlimit) {
1163 /* this code is executed quadratically often */
1164 /* do minimal checking `by hand' */
1165
1166 if ((char *) de + namelen <= dlimit &&
1167 ext4_match (namelen, name, de)) {
1168 /* found a match - just to be sure, do a full check */
1169 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1170 bh->b_size, offset))
1171 return -1;
1172 *res_dir = de;
1173 return 1;
1174 }
1175 /* prevent looping on a bad block */
1176 de_len = ext4_rec_len_from_disk(de->rec_len,
1177 dir->i_sb->s_blocksize);
1178 if (de_len <= 0)
1179 return -1;
1180 offset += de_len;
1181 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1182 }
1183 return 0;
1184 }
1185
1186 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1187 struct ext4_dir_entry *de)
1188 {
1189 struct super_block *sb = dir->i_sb;
1190
1191 if (!is_dx(dir))
1192 return 0;
1193 if (block == 0)
1194 return 1;
1195 if (de->inode == 0 &&
1196 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1197 sb->s_blocksize)
1198 return 1;
1199 return 0;
1200 }
1201
1202 /*
1203 * ext4_find_entry()
1204 *
1205 * finds an entry in the specified directory with the wanted name. It
1206 * returns the cache buffer in which the entry was found, and the entry
1207 * itself (as a parameter - res_dir). It does NOT read the inode of the
1208 * entry - you'll have to do that yourself if you want to.
1209 *
1210 * The returned buffer_head has ->b_count elevated. The caller is expected
1211 * to brelse() it when appropriate.
1212 */
1213 static struct buffer_head * ext4_find_entry (struct inode *dir,
1214 const struct qstr *d_name,
1215 struct ext4_dir_entry_2 **res_dir,
1216 int *inlined)
1217 {
1218 struct super_block *sb;
1219 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1220 struct buffer_head *bh, *ret = NULL;
1221 ext4_lblk_t start, block, b;
1222 const u8 *name = d_name->name;
1223 int ra_max = 0; /* Number of bh's in the readahead
1224 buffer, bh_use[] */
1225 int ra_ptr = 0; /* Current index into readahead
1226 buffer */
1227 int num = 0;
1228 ext4_lblk_t nblocks;
1229 int i, err;
1230 int namelen;
1231
1232 *res_dir = NULL;
1233 sb = dir->i_sb;
1234 namelen = d_name->len;
1235 if (namelen > EXT4_NAME_LEN)
1236 return NULL;
1237
1238 if (ext4_has_inline_data(dir)) {
1239 int has_inline_data = 1;
1240 ret = ext4_find_inline_entry(dir, d_name, res_dir,
1241 &has_inline_data);
1242 if (has_inline_data) {
1243 if (inlined)
1244 *inlined = 1;
1245 return ret;
1246 }
1247 }
1248
1249 if ((namelen <= 2) && (name[0] == '.') &&
1250 (name[1] == '.' || name[1] == '\0')) {
1251 /*
1252 * "." or ".." will only be in the first block
1253 * NFS may look up ".."; "." should be handled by the VFS
1254 */
1255 block = start = 0;
1256 nblocks = 1;
1257 goto restart;
1258 }
1259 if (is_dx(dir)) {
1260 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1261 /*
1262 * On success, or if the error was file not found,
1263 * return. Otherwise, fall back to doing a search the
1264 * old fashioned way.
1265 */
1266 if (bh || (err != ERR_BAD_DX_DIR))
1267 return bh;
1268 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1269 "falling back\n"));
1270 }
1271 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1272 start = EXT4_I(dir)->i_dir_start_lookup;
1273 if (start >= nblocks)
1274 start = 0;
1275 block = start;
1276 restart:
1277 do {
1278 /*
1279 * We deal with the read-ahead logic here.
1280 */
1281 if (ra_ptr >= ra_max) {
1282 /* Refill the readahead buffer */
1283 ra_ptr = 0;
1284 b = block;
1285 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1286 /*
1287 * Terminate if we reach the end of the
1288 * directory and must wrap, or if our
1289 * search has finished at this block.
1290 */
1291 if (b >= nblocks || (num && block == start)) {
1292 bh_use[ra_max] = NULL;
1293 break;
1294 }
1295 num++;
1296 bh = ext4_getblk(NULL, dir, b++, 0, &err);
1297 bh_use[ra_max] = bh;
1298 if (bh)
1299 ll_rw_block(READ | REQ_META | REQ_PRIO,
1300 1, &bh);
1301 }
1302 }
1303 if ((bh = bh_use[ra_ptr++]) == NULL)
1304 goto next;
1305 wait_on_buffer(bh);
1306 if (!buffer_uptodate(bh)) {
1307 /* read error, skip block & hope for the best */
1308 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1309 (unsigned long) block);
1310 brelse(bh);
1311 goto next;
1312 }
1313 if (!buffer_verified(bh) &&
1314 !is_dx_internal_node(dir, block,
1315 (struct ext4_dir_entry *)bh->b_data) &&
1316 !ext4_dirent_csum_verify(dir,
1317 (struct ext4_dir_entry *)bh->b_data)) {
1318 EXT4_ERROR_INODE(dir, "checksumming directory "
1319 "block %lu", (unsigned long)block);
1320 brelse(bh);
1321 goto next;
1322 }
1323 set_buffer_verified(bh);
1324 i = search_dirblock(bh, dir, d_name,
1325 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1326 if (i == 1) {
1327 EXT4_I(dir)->i_dir_start_lookup = block;
1328 ret = bh;
1329 goto cleanup_and_exit;
1330 } else {
1331 brelse(bh);
1332 if (i < 0)
1333 goto cleanup_and_exit;
1334 }
1335 next:
1336 if (++block >= nblocks)
1337 block = 0;
1338 } while (block != start);
1339
1340 /*
1341 * If the directory has grown while we were searching, then
1342 * search the last part of the directory before giving up.
1343 */
1344 block = nblocks;
1345 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1346 if (block < nblocks) {
1347 start = 0;
1348 goto restart;
1349 }
1350
1351 cleanup_and_exit:
1352 /* Clean up the read-ahead blocks */
1353 for (; ra_ptr < ra_max; ra_ptr++)
1354 brelse(bh_use[ra_ptr]);
1355 return ret;
1356 }
1357
1358 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1359 struct ext4_dir_entry_2 **res_dir, int *err)
1360 {
1361 struct super_block * sb = dir->i_sb;
1362 struct dx_hash_info hinfo;
1363 struct dx_frame frames[2], *frame;
1364 struct buffer_head *bh;
1365 ext4_lblk_t block;
1366 int retval;
1367
1368 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1369 return NULL;
1370 do {
1371 block = dx_get_block(frame->at);
1372 bh = ext4_read_dirblock(dir, block, DIRENT);
1373 if (IS_ERR(bh)) {
1374 *err = PTR_ERR(bh);
1375 goto errout;
1376 }
1377 retval = search_dirblock(bh, dir, d_name,
1378 block << EXT4_BLOCK_SIZE_BITS(sb),
1379 res_dir);
1380 if (retval == 1) { /* Success! */
1381 dx_release(frames);
1382 return bh;
1383 }
1384 brelse(bh);
1385 if (retval == -1) {
1386 *err = ERR_BAD_DX_DIR;
1387 goto errout;
1388 }
1389
1390 /* Check to see if we should continue to search */
1391 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1392 frames, NULL);
1393 if (retval < 0) {
1394 ext4_warning(sb,
1395 "error reading index page in directory #%lu",
1396 dir->i_ino);
1397 *err = retval;
1398 goto errout;
1399 }
1400 } while (retval == 1);
1401
1402 *err = -ENOENT;
1403 errout:
1404 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1405 dx_release (frames);
1406 return NULL;
1407 }
1408
1409 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1410 {
1411 struct inode *inode;
1412 struct ext4_dir_entry_2 *de;
1413 struct buffer_head *bh;
1414
1415 if (dentry->d_name.len > EXT4_NAME_LEN)
1416 return ERR_PTR(-ENAMETOOLONG);
1417
1418 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1419 inode = NULL;
1420 if (bh) {
1421 __u32 ino = le32_to_cpu(de->inode);
1422 brelse(bh);
1423 if (!ext4_valid_inum(dir->i_sb, ino)) {
1424 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1425 return ERR_PTR(-EIO);
1426 }
1427 if (unlikely(ino == dir->i_ino)) {
1428 EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1429 dentry->d_name.len,
1430 dentry->d_name.name);
1431 return ERR_PTR(-EIO);
1432 }
1433 inode = ext4_iget(dir->i_sb, ino);
1434 if (inode == ERR_PTR(-ESTALE)) {
1435 EXT4_ERROR_INODE(dir,
1436 "deleted inode referenced: %u",
1437 ino);
1438 return ERR_PTR(-EIO);
1439 }
1440 }
1441 return d_splice_alias(inode, dentry);
1442 }
1443
1444
1445 struct dentry *ext4_get_parent(struct dentry *child)
1446 {
1447 __u32 ino;
1448 static const struct qstr dotdot = QSTR_INIT("..", 2);
1449 struct ext4_dir_entry_2 * de;
1450 struct buffer_head *bh;
1451
1452 bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
1453 if (!bh)
1454 return ERR_PTR(-ENOENT);
1455 ino = le32_to_cpu(de->inode);
1456 brelse(bh);
1457
1458 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1459 EXT4_ERROR_INODE(child->d_inode,
1460 "bad parent inode number: %u", ino);
1461 return ERR_PTR(-EIO);
1462 }
1463
1464 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1465 }
1466
1467 /*
1468 * Move count entries from end of map between two memory locations.
1469 * Returns pointer to last entry moved.
1470 */
1471 static struct ext4_dir_entry_2 *
1472 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1473 unsigned blocksize)
1474 {
1475 unsigned rec_len = 0;
1476
1477 while (count--) {
1478 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1479 (from + (map->offs<<2));
1480 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1481 memcpy (to, de, rec_len);
1482 ((struct ext4_dir_entry_2 *) to)->rec_len =
1483 ext4_rec_len_to_disk(rec_len, blocksize);
1484 de->inode = 0;
1485 map++;
1486 to += rec_len;
1487 }
1488 return (struct ext4_dir_entry_2 *) (to - rec_len);
1489 }
1490
1491 /*
1492 * Compact each dir entry in the range to the minimal rec_len.
1493 * Returns pointer to last entry in range.
1494 */
1495 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1496 {
1497 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1498 unsigned rec_len = 0;
1499
1500 prev = to = de;
1501 while ((char*)de < base + blocksize) {
1502 next = ext4_next_entry(de, blocksize);
1503 if (de->inode && de->name_len) {
1504 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1505 if (de > to)
1506 memmove(to, de, rec_len);
1507 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1508 prev = to;
1509 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1510 }
1511 de = next;
1512 }
1513 return prev;
1514 }
1515
1516 /*
1517 * Split a full leaf block to make room for a new dir entry.
1518 * Allocate a new block, and move entries so that they are approx. equally full.
1519 * Returns pointer to de in block into which the new entry will be inserted.
1520 */
1521 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1522 struct buffer_head **bh,struct dx_frame *frame,
1523 struct dx_hash_info *hinfo, int *error)
1524 {
1525 unsigned blocksize = dir->i_sb->s_blocksize;
1526 unsigned count, continued;
1527 struct buffer_head *bh2;
1528 ext4_lblk_t newblock;
1529 u32 hash2;
1530 struct dx_map_entry *map;
1531 char *data1 = (*bh)->b_data, *data2;
1532 unsigned split, move, size;
1533 struct ext4_dir_entry_2 *de = NULL, *de2;
1534 struct ext4_dir_entry_tail *t;
1535 int csum_size = 0;
1536 int err = 0, i;
1537
1538 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1539 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1540 csum_size = sizeof(struct ext4_dir_entry_tail);
1541
1542 bh2 = ext4_append(handle, dir, &newblock);
1543 if (IS_ERR(bh2)) {
1544 brelse(*bh);
1545 *bh = NULL;
1546 *error = PTR_ERR(bh2);
1547 return NULL;
1548 }
1549
1550 BUFFER_TRACE(*bh, "get_write_access");
1551 err = ext4_journal_get_write_access(handle, *bh);
1552 if (err)
1553 goto journal_error;
1554
1555 BUFFER_TRACE(frame->bh, "get_write_access");
1556 err = ext4_journal_get_write_access(handle, frame->bh);
1557 if (err)
1558 goto journal_error;
1559
1560 data2 = bh2->b_data;
1561
1562 /* create map in the end of data2 block */
1563 map = (struct dx_map_entry *) (data2 + blocksize);
1564 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1565 blocksize, hinfo, map);
1566 map -= count;
1567 dx_sort_map(map, count);
1568 /* Split the existing block in the middle, size-wise */
1569 size = 0;
1570 move = 0;
1571 for (i = count-1; i >= 0; i--) {
1572 /* is more than half of this entry in 2nd half of the block? */
1573 if (size + map[i].size/2 > blocksize/2)
1574 break;
1575 size += map[i].size;
1576 move++;
1577 }
1578 /* map index at which we will split */
1579 split = count - move;
1580 hash2 = map[split].hash;
1581 continued = hash2 == map[split - 1].hash;
1582 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1583 (unsigned long)dx_get_block(frame->at),
1584 hash2, split, count-split));
1585
1586 /* Fancy dance to stay within two buffers */
1587 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1588 de = dx_pack_dirents(data1, blocksize);
1589 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1590 (char *) de,
1591 blocksize);
1592 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1593 (char *) de2,
1594 blocksize);
1595 if (csum_size) {
1596 t = EXT4_DIRENT_TAIL(data2, blocksize);
1597 initialize_dirent_tail(t, blocksize);
1598
1599 t = EXT4_DIRENT_TAIL(data1, blocksize);
1600 initialize_dirent_tail(t, blocksize);
1601 }
1602
1603 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1604 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1605
1606 /* Which block gets the new entry? */
1607 if (hinfo->hash >= hash2)
1608 {
1609 swap(*bh, bh2);
1610 de = de2;
1611 }
1612 dx_insert_block(frame, hash2 + continued, newblock);
1613 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1614 if (err)
1615 goto journal_error;
1616 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1617 if (err)
1618 goto journal_error;
1619 brelse(bh2);
1620 dxtrace(dx_show_index("frame", frame->entries));
1621 return de;
1622
1623 journal_error:
1624 brelse(*bh);
1625 brelse(bh2);
1626 *bh = NULL;
1627 ext4_std_error(dir->i_sb, err);
1628 *error = err;
1629 return NULL;
1630 }
1631
1632 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1633 struct buffer_head *bh,
1634 void *buf, int buf_size,
1635 const char *name, int namelen,
1636 struct ext4_dir_entry_2 **dest_de)
1637 {
1638 struct ext4_dir_entry_2 *de;
1639 unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
1640 int nlen, rlen;
1641 unsigned int offset = 0;
1642 char *top;
1643
1644 de = (struct ext4_dir_entry_2 *)buf;
1645 top = buf + buf_size - reclen;
1646 while ((char *) de <= top) {
1647 if (ext4_check_dir_entry(dir, NULL, de, bh,
1648 buf, buf_size, offset))
1649 return -EIO;
1650 if (ext4_match(namelen, name, de))
1651 return -EEXIST;
1652 nlen = EXT4_DIR_REC_LEN(de->name_len);
1653 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1654 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1655 break;
1656 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1657 offset += rlen;
1658 }
1659 if ((char *) de > top)
1660 return -ENOSPC;
1661
1662 *dest_de = de;
1663 return 0;
1664 }
1665
1666 void ext4_insert_dentry(struct inode *inode,
1667 struct ext4_dir_entry_2 *de,
1668 int buf_size,
1669 const char *name, int namelen)
1670 {
1671
1672 int nlen, rlen;
1673
1674 nlen = EXT4_DIR_REC_LEN(de->name_len);
1675 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1676 if (de->inode) {
1677 struct ext4_dir_entry_2 *de1 =
1678 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1679 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1680 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1681 de = de1;
1682 }
1683 de->file_type = EXT4_FT_UNKNOWN;
1684 de->inode = cpu_to_le32(inode->i_ino);
1685 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1686 de->name_len = namelen;
1687 memcpy(de->name, name, namelen);
1688 }
1689 /*
1690 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1691 * it points to a directory entry which is guaranteed to be large
1692 * enough for new directory entry. If de is NULL, then
1693 * add_dirent_to_buf will attempt search the directory block for
1694 * space. It will return -ENOSPC if no space is available, and -EIO
1695 * and -EEXIST if directory entry already exists.
1696 */
1697 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1698 struct inode *inode, struct ext4_dir_entry_2 *de,
1699 struct buffer_head *bh)
1700 {
1701 struct inode *dir = dentry->d_parent->d_inode;
1702 const char *name = dentry->d_name.name;
1703 int namelen = dentry->d_name.len;
1704 unsigned int blocksize = dir->i_sb->s_blocksize;
1705 int csum_size = 0;
1706 int err;
1707
1708 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1709 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1710 csum_size = sizeof(struct ext4_dir_entry_tail);
1711
1712 if (!de) {
1713 err = ext4_find_dest_de(dir, inode,
1714 bh, bh->b_data, blocksize - csum_size,
1715 name, namelen, &de);
1716 if (err)
1717 return err;
1718 }
1719 BUFFER_TRACE(bh, "get_write_access");
1720 err = ext4_journal_get_write_access(handle, bh);
1721 if (err) {
1722 ext4_std_error(dir->i_sb, err);
1723 return err;
1724 }
1725
1726 /* By now the buffer is marked for journaling */
1727 ext4_insert_dentry(inode, de, blocksize, name, namelen);
1728
1729 /*
1730 * XXX shouldn't update any times until successful
1731 * completion of syscall, but too many callers depend
1732 * on this.
1733 *
1734 * XXX similarly, too many callers depend on
1735 * ext4_new_inode() setting the times, but error
1736 * recovery deletes the inode, so the worst that can
1737 * happen is that the times are slightly out of date
1738 * and/or different from the directory change time.
1739 */
1740 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1741 ext4_update_dx_flag(dir);
1742 dir->i_version++;
1743 ext4_mark_inode_dirty(handle, dir);
1744 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1745 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1746 if (err)
1747 ext4_std_error(dir->i_sb, err);
1748 return 0;
1749 }
1750
1751 /*
1752 * This converts a one block unindexed directory to a 3 block indexed
1753 * directory, and adds the dentry to the indexed directory.
1754 */
1755 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1756 struct inode *inode, struct buffer_head *bh)
1757 {
1758 struct inode *dir = dentry->d_parent->d_inode;
1759 const char *name = dentry->d_name.name;
1760 int namelen = dentry->d_name.len;
1761 struct buffer_head *bh2;
1762 struct dx_root *root;
1763 struct dx_frame frames[2], *frame;
1764 struct dx_entry *entries;
1765 struct ext4_dir_entry_2 *de, *de2;
1766 struct ext4_dir_entry_tail *t;
1767 char *data1, *top;
1768 unsigned len;
1769 int retval;
1770 unsigned blocksize;
1771 struct dx_hash_info hinfo;
1772 ext4_lblk_t block;
1773 struct fake_dirent *fde;
1774 int csum_size = 0;
1775
1776 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1777 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1778 csum_size = sizeof(struct ext4_dir_entry_tail);
1779
1780 blocksize = dir->i_sb->s_blocksize;
1781 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1782 retval = ext4_journal_get_write_access(handle, bh);
1783 if (retval) {
1784 ext4_std_error(dir->i_sb, retval);
1785 brelse(bh);
1786 return retval;
1787 }
1788 root = (struct dx_root *) bh->b_data;
1789
1790 /* The 0th block becomes the root, move the dirents out */
1791 fde = &root->dotdot;
1792 de = (struct ext4_dir_entry_2 *)((char *)fde +
1793 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1794 if ((char *) de >= (((char *) root) + blocksize)) {
1795 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1796 brelse(bh);
1797 return -EIO;
1798 }
1799 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1800
1801 /* Allocate new block for the 0th block's dirents */
1802 bh2 = ext4_append(handle, dir, &block);
1803 if (IS_ERR(bh2)) {
1804 brelse(bh);
1805 return PTR_ERR(bh2);
1806 }
1807 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1808 data1 = bh2->b_data;
1809
1810 memcpy (data1, de, len);
1811 de = (struct ext4_dir_entry_2 *) data1;
1812 top = data1 + len;
1813 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1814 de = de2;
1815 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1816 (char *) de,
1817 blocksize);
1818
1819 if (csum_size) {
1820 t = EXT4_DIRENT_TAIL(data1, blocksize);
1821 initialize_dirent_tail(t, blocksize);
1822 }
1823
1824 /* Initialize the root; the dot dirents already exist */
1825 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1826 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1827 blocksize);
1828 memset (&root->info, 0, sizeof(root->info));
1829 root->info.info_length = sizeof(root->info);
1830 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1831 entries = root->entries;
1832 dx_set_block(entries, 1);
1833 dx_set_count(entries, 1);
1834 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1835
1836 /* Initialize as for dx_probe */
1837 hinfo.hash_version = root->info.hash_version;
1838 if (hinfo.hash_version <= DX_HASH_TEA)
1839 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1840 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1841 ext4fs_dirhash(name, namelen, &hinfo);
1842 frame = frames;
1843 frame->entries = entries;
1844 frame->at = entries;
1845 frame->bh = bh;
1846 bh = bh2;
1847
1848 ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1849 ext4_handle_dirty_dirent_node(handle, dir, bh);
1850
1851 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1852 if (!de) {
1853 /*
1854 * Even if the block split failed, we have to properly write
1855 * out all the changes we did so far. Otherwise we can end up
1856 * with corrupted filesystem.
1857 */
1858 ext4_mark_inode_dirty(handle, dir);
1859 dx_release(frames);
1860 return retval;
1861 }
1862 dx_release(frames);
1863
1864 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1865 brelse(bh);
1866 return retval;
1867 }
1868
1869 /*
1870 * ext4_add_entry()
1871 *
1872 * adds a file entry to the specified directory, using the same
1873 * semantics as ext4_find_entry(). It returns NULL if it failed.
1874 *
1875 * NOTE!! The inode part of 'de' is left at 0 - which means you
1876 * may not sleep between calling this and putting something into
1877 * the entry, as someone else might have used it while you slept.
1878 */
1879 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1880 struct inode *inode)
1881 {
1882 struct inode *dir = dentry->d_parent->d_inode;
1883 struct buffer_head *bh;
1884 struct ext4_dir_entry_2 *de;
1885 struct ext4_dir_entry_tail *t;
1886 struct super_block *sb;
1887 int retval;
1888 int dx_fallback=0;
1889 unsigned blocksize;
1890 ext4_lblk_t block, blocks;
1891 int csum_size = 0;
1892
1893 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1894 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1895 csum_size = sizeof(struct ext4_dir_entry_tail);
1896
1897 sb = dir->i_sb;
1898 blocksize = sb->s_blocksize;
1899 if (!dentry->d_name.len)
1900 return -EINVAL;
1901
1902 if (ext4_has_inline_data(dir)) {
1903 retval = ext4_try_add_inline_entry(handle, dentry, inode);
1904 if (retval < 0)
1905 return retval;
1906 if (retval == 1) {
1907 retval = 0;
1908 return retval;
1909 }
1910 }
1911
1912 if (is_dx(dir)) {
1913 retval = ext4_dx_add_entry(handle, dentry, inode);
1914 if (!retval || (retval != ERR_BAD_DX_DIR))
1915 return retval;
1916 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1917 dx_fallback++;
1918 ext4_mark_inode_dirty(handle, dir);
1919 }
1920 blocks = dir->i_size >> sb->s_blocksize_bits;
1921 for (block = 0; block < blocks; block++) {
1922 bh = ext4_read_dirblock(dir, block, DIRENT);
1923 if (IS_ERR(bh))
1924 return PTR_ERR(bh);
1925
1926 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1927 if (retval != -ENOSPC) {
1928 brelse(bh);
1929 return retval;
1930 }
1931
1932 if (blocks == 1 && !dx_fallback &&
1933 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1934 return make_indexed_dir(handle, dentry, inode, bh);
1935 brelse(bh);
1936 }
1937 bh = ext4_append(handle, dir, &block);
1938 if (IS_ERR(bh))
1939 return PTR_ERR(bh);
1940 de = (struct ext4_dir_entry_2 *) bh->b_data;
1941 de->inode = 0;
1942 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1943
1944 if (csum_size) {
1945 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1946 initialize_dirent_tail(t, blocksize);
1947 }
1948
1949 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1950 brelse(bh);
1951 if (retval == 0)
1952 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1953 return retval;
1954 }
1955
1956 /*
1957 * Returns 0 for success, or a negative error value
1958 */
1959 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1960 struct inode *inode)
1961 {
1962 struct dx_frame frames[2], *frame;
1963 struct dx_entry *entries, *at;
1964 struct dx_hash_info hinfo;
1965 struct buffer_head *bh;
1966 struct inode *dir = dentry->d_parent->d_inode;
1967 struct super_block *sb = dir->i_sb;
1968 struct ext4_dir_entry_2 *de;
1969 int err;
1970
1971 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1972 if (!frame)
1973 return err;
1974 entries = frame->entries;
1975 at = frame->at;
1976 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
1977 if (IS_ERR(bh)) {
1978 err = PTR_ERR(bh);
1979 bh = NULL;
1980 goto cleanup;
1981 }
1982
1983 BUFFER_TRACE(bh, "get_write_access");
1984 err = ext4_journal_get_write_access(handle, bh);
1985 if (err)
1986 goto journal_error;
1987
1988 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1989 if (err != -ENOSPC)
1990 goto cleanup;
1991
1992 /* Block full, should compress but for now just split */
1993 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1994 dx_get_count(entries), dx_get_limit(entries)));
1995 /* Need to split index? */
1996 if (dx_get_count(entries) == dx_get_limit(entries)) {
1997 ext4_lblk_t newblock;
1998 unsigned icount = dx_get_count(entries);
1999 int levels = frame - frames;
2000 struct dx_entry *entries2;
2001 struct dx_node *node2;
2002 struct buffer_head *bh2;
2003
2004 if (levels && (dx_get_count(frames->entries) ==
2005 dx_get_limit(frames->entries))) {
2006 ext4_warning(sb, "Directory index full!");
2007 err = -ENOSPC;
2008 goto cleanup;
2009 }
2010 bh2 = ext4_append(handle, dir, &newblock);
2011 if (IS_ERR(bh2)) {
2012 err = PTR_ERR(bh2);
2013 goto cleanup;
2014 }
2015 node2 = (struct dx_node *)(bh2->b_data);
2016 entries2 = node2->entries;
2017 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2018 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2019 sb->s_blocksize);
2020 BUFFER_TRACE(frame->bh, "get_write_access");
2021 err = ext4_journal_get_write_access(handle, frame->bh);
2022 if (err)
2023 goto journal_error;
2024 if (levels) {
2025 unsigned icount1 = icount/2, icount2 = icount - icount1;
2026 unsigned hash2 = dx_get_hash(entries + icount1);
2027 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2028 icount1, icount2));
2029
2030 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2031 err = ext4_journal_get_write_access(handle,
2032 frames[0].bh);
2033 if (err)
2034 goto journal_error;
2035
2036 memcpy((char *) entries2, (char *) (entries + icount1),
2037 icount2 * sizeof(struct dx_entry));
2038 dx_set_count(entries, icount1);
2039 dx_set_count(entries2, icount2);
2040 dx_set_limit(entries2, dx_node_limit(dir));
2041
2042 /* Which index block gets the new entry? */
2043 if (at - entries >= icount1) {
2044 frame->at = at = at - entries - icount1 + entries2;
2045 frame->entries = entries = entries2;
2046 swap(frame->bh, bh2);
2047 }
2048 dx_insert_block(frames + 0, hash2, newblock);
2049 dxtrace(dx_show_index("node", frames[1].entries));
2050 dxtrace(dx_show_index("node",
2051 ((struct dx_node *) bh2->b_data)->entries));
2052 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2053 if (err)
2054 goto journal_error;
2055 brelse (bh2);
2056 } else {
2057 dxtrace(printk(KERN_DEBUG
2058 "Creating second level index...\n"));
2059 memcpy((char *) entries2, (char *) entries,
2060 icount * sizeof(struct dx_entry));
2061 dx_set_limit(entries2, dx_node_limit(dir));
2062
2063 /* Set up root */
2064 dx_set_count(entries, 1);
2065 dx_set_block(entries + 0, newblock);
2066 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2067
2068 /* Add new access path frame */
2069 frame = frames + 1;
2070 frame->at = at = at - entries + entries2;
2071 frame->entries = entries = entries2;
2072 frame->bh = bh2;
2073 err = ext4_journal_get_write_access(handle,
2074 frame->bh);
2075 if (err)
2076 goto journal_error;
2077 }
2078 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2079 if (err) {
2080 ext4_std_error(inode->i_sb, err);
2081 goto cleanup;
2082 }
2083 }
2084 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
2085 if (!de)
2086 goto cleanup;
2087 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
2088 goto cleanup;
2089
2090 journal_error:
2091 ext4_std_error(dir->i_sb, err);
2092 cleanup:
2093 brelse(bh);
2094 dx_release(frames);
2095 return err;
2096 }
2097
2098 /*
2099 * ext4_generic_delete_entry deletes a directory entry by merging it
2100 * with the previous entry
2101 */
2102 int ext4_generic_delete_entry(handle_t *handle,
2103 struct inode *dir,
2104 struct ext4_dir_entry_2 *de_del,
2105 struct buffer_head *bh,
2106 void *entry_buf,
2107 int buf_size,
2108 int csum_size)
2109 {
2110 struct ext4_dir_entry_2 *de, *pde;
2111 unsigned int blocksize = dir->i_sb->s_blocksize;
2112 int i;
2113
2114 i = 0;
2115 pde = NULL;
2116 de = (struct ext4_dir_entry_2 *)entry_buf;
2117 while (i < buf_size - csum_size) {
2118 if (ext4_check_dir_entry(dir, NULL, de, bh,
2119 bh->b_data, bh->b_size, i))
2120 return -EIO;
2121 if (de == de_del) {
2122 if (pde)
2123 pde->rec_len = ext4_rec_len_to_disk(
2124 ext4_rec_len_from_disk(pde->rec_len,
2125 blocksize) +
2126 ext4_rec_len_from_disk(de->rec_len,
2127 blocksize),
2128 blocksize);
2129 else
2130 de->inode = 0;
2131 dir->i_version++;
2132 return 0;
2133 }
2134 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2135 pde = de;
2136 de = ext4_next_entry(de, blocksize);
2137 }
2138 return -ENOENT;
2139 }
2140
2141 static int ext4_delete_entry(handle_t *handle,
2142 struct inode *dir,
2143 struct ext4_dir_entry_2 *de_del,
2144 struct buffer_head *bh)
2145 {
2146 int err, csum_size = 0;
2147
2148 if (ext4_has_inline_data(dir)) {
2149 int has_inline_data = 1;
2150 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2151 &has_inline_data);
2152 if (has_inline_data)
2153 return err;
2154 }
2155
2156 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2157 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2158 csum_size = sizeof(struct ext4_dir_entry_tail);
2159
2160 BUFFER_TRACE(bh, "get_write_access");
2161 err = ext4_journal_get_write_access(handle, bh);
2162 if (unlikely(err))
2163 goto out;
2164
2165 err = ext4_generic_delete_entry(handle, dir, de_del,
2166 bh, bh->b_data,
2167 dir->i_sb->s_blocksize, csum_size);
2168 if (err)
2169 goto out;
2170
2171 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2172 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2173 if (unlikely(err))
2174 goto out;
2175
2176 return 0;
2177 out:
2178 if (err != -ENOENT)
2179 ext4_std_error(dir->i_sb, err);
2180 return err;
2181 }
2182
2183 /*
2184 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2185 * since this indicates that nlinks count was previously 1.
2186 */
2187 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2188 {
2189 inc_nlink(inode);
2190 if (is_dx(inode) && inode->i_nlink > 1) {
2191 /* limit is 16-bit i_links_count */
2192 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2193 set_nlink(inode, 1);
2194 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2195 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2196 }
2197 }
2198 }
2199
2200 /*
2201 * If a directory had nlink == 1, then we should let it be 1. This indicates
2202 * directory has >EXT4_LINK_MAX subdirs.
2203 */
2204 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2205 {
2206 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2207 drop_nlink(inode);
2208 }
2209
2210
2211 static int ext4_add_nondir(handle_t *handle,
2212 struct dentry *dentry, struct inode *inode)
2213 {
2214 int err = ext4_add_entry(handle, dentry, inode);
2215 if (!err) {
2216 ext4_mark_inode_dirty(handle, inode);
2217 unlock_new_inode(inode);
2218 d_instantiate(dentry, inode);
2219 return 0;
2220 }
2221 drop_nlink(inode);
2222 unlock_new_inode(inode);
2223 iput(inode);
2224 return err;
2225 }
2226
2227 /*
2228 * By the time this is called, we already have created
2229 * the directory cache entry for the new file, but it
2230 * is so far negative - it has no inode.
2231 *
2232 * If the create succeeds, we fill in the inode information
2233 * with d_instantiate().
2234 */
2235 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2236 bool excl)
2237 {
2238 handle_t *handle;
2239 struct inode *inode;
2240 int err, credits, retries = 0;
2241
2242 dquot_initialize(dir);
2243
2244 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2245 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2246 retry:
2247 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2248 NULL, EXT4_HT_DIR, credits);
2249 handle = ext4_journal_current_handle();
2250 err = PTR_ERR(inode);
2251 if (!IS_ERR(inode)) {
2252 inode->i_op = &ext4_file_inode_operations;
2253 inode->i_fop = &ext4_file_operations;
2254 ext4_set_aops(inode);
2255 err = ext4_add_nondir(handle, dentry, inode);
2256 if (!err && IS_DIRSYNC(dir))
2257 ext4_handle_sync(handle);
2258 }
2259 if (handle)
2260 ext4_journal_stop(handle);
2261 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2262 goto retry;
2263 return err;
2264 }
2265
2266 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2267 umode_t mode, dev_t rdev)
2268 {
2269 handle_t *handle;
2270 struct inode *inode;
2271 int err, credits, retries = 0;
2272
2273 if (!new_valid_dev(rdev))
2274 return -EINVAL;
2275
2276 dquot_initialize(dir);
2277
2278 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2279 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2280 retry:
2281 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2282 NULL, EXT4_HT_DIR, credits);
2283 handle = ext4_journal_current_handle();
2284 err = PTR_ERR(inode);
2285 if (!IS_ERR(inode)) {
2286 init_special_inode(inode, inode->i_mode, rdev);
2287 inode->i_op = &ext4_special_inode_operations;
2288 err = ext4_add_nondir(handle, dentry, inode);
2289 if (!err && IS_DIRSYNC(dir))
2290 ext4_handle_sync(handle);
2291 }
2292 if (handle)
2293 ext4_journal_stop(handle);
2294 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2295 goto retry;
2296 return err;
2297 }
2298
2299 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2300 {
2301 handle_t *handle;
2302 struct inode *inode;
2303 int err, retries = 0;
2304
2305 dquot_initialize(dir);
2306
2307 retry:
2308 inode = ext4_new_inode_start_handle(dir, mode,
2309 NULL, 0, NULL,
2310 EXT4_HT_DIR,
2311 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2312 4 + EXT4_XATTR_TRANS_BLOCKS);
2313 handle = ext4_journal_current_handle();
2314 err = PTR_ERR(inode);
2315 if (!IS_ERR(inode)) {
2316 inode->i_op = &ext4_file_inode_operations;
2317 inode->i_fop = &ext4_file_operations;
2318 ext4_set_aops(inode);
2319 d_tmpfile(dentry, inode);
2320 err = ext4_orphan_add(handle, inode);
2321 if (err)
2322 goto err_unlock_inode;
2323 mark_inode_dirty(inode);
2324 unlock_new_inode(inode);
2325 }
2326 if (handle)
2327 ext4_journal_stop(handle);
2328 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2329 goto retry;
2330 return err;
2331 err_unlock_inode:
2332 ext4_journal_stop(handle);
2333 unlock_new_inode(inode);
2334 return err;
2335 }
2336
2337 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2338 struct ext4_dir_entry_2 *de,
2339 int blocksize, int csum_size,
2340 unsigned int parent_ino, int dotdot_real_len)
2341 {
2342 de->inode = cpu_to_le32(inode->i_ino);
2343 de->name_len = 1;
2344 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2345 blocksize);
2346 strcpy(de->name, ".");
2347 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2348
2349 de = ext4_next_entry(de, blocksize);
2350 de->inode = cpu_to_le32(parent_ino);
2351 de->name_len = 2;
2352 if (!dotdot_real_len)
2353 de->rec_len = ext4_rec_len_to_disk(blocksize -
2354 (csum_size + EXT4_DIR_REC_LEN(1)),
2355 blocksize);
2356 else
2357 de->rec_len = ext4_rec_len_to_disk(
2358 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2359 strcpy(de->name, "..");
2360 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2361
2362 return ext4_next_entry(de, blocksize);
2363 }
2364
2365 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2366 struct inode *inode)
2367 {
2368 struct buffer_head *dir_block = NULL;
2369 struct ext4_dir_entry_2 *de;
2370 struct ext4_dir_entry_tail *t;
2371 ext4_lblk_t block = 0;
2372 unsigned int blocksize = dir->i_sb->s_blocksize;
2373 int csum_size = 0;
2374 int err;
2375
2376 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2377 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2378 csum_size = sizeof(struct ext4_dir_entry_tail);
2379
2380 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2381 err = ext4_try_create_inline_dir(handle, dir, inode);
2382 if (err < 0 && err != -ENOSPC)
2383 goto out;
2384 if (!err)
2385 goto out;
2386 }
2387
2388 inode->i_size = 0;
2389 dir_block = ext4_append(handle, inode, &block);
2390 if (IS_ERR(dir_block))
2391 return PTR_ERR(dir_block);
2392 BUFFER_TRACE(dir_block, "get_write_access");
2393 err = ext4_journal_get_write_access(handle, dir_block);
2394 if (err)
2395 goto out;
2396 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2397 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2398 set_nlink(inode, 2);
2399 if (csum_size) {
2400 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2401 initialize_dirent_tail(t, blocksize);
2402 }
2403
2404 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2405 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2406 if (err)
2407 goto out;
2408 set_buffer_verified(dir_block);
2409 out:
2410 brelse(dir_block);
2411 return err;
2412 }
2413
2414 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2415 {
2416 handle_t *handle;
2417 struct inode *inode;
2418 int err, credits, retries = 0;
2419
2420 if (EXT4_DIR_LINK_MAX(dir))
2421 return -EMLINK;
2422
2423 dquot_initialize(dir);
2424
2425 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2426 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2427 retry:
2428 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2429 &dentry->d_name,
2430 0, NULL, EXT4_HT_DIR, credits);
2431 handle = ext4_journal_current_handle();
2432 err = PTR_ERR(inode);
2433 if (IS_ERR(inode))
2434 goto out_stop;
2435
2436 inode->i_op = &ext4_dir_inode_operations;
2437 inode->i_fop = &ext4_dir_operations;
2438 err = ext4_init_new_dir(handle, dir, inode);
2439 if (err)
2440 goto out_clear_inode;
2441 err = ext4_mark_inode_dirty(handle, inode);
2442 if (!err)
2443 err = ext4_add_entry(handle, dentry, inode);
2444 if (err) {
2445 out_clear_inode:
2446 clear_nlink(inode);
2447 unlock_new_inode(inode);
2448 ext4_mark_inode_dirty(handle, inode);
2449 iput(inode);
2450 goto out_stop;
2451 }
2452 ext4_inc_count(handle, dir);
2453 ext4_update_dx_flag(dir);
2454 err = ext4_mark_inode_dirty(handle, dir);
2455 if (err)
2456 goto out_clear_inode;
2457 unlock_new_inode(inode);
2458 d_instantiate(dentry, inode);
2459 if (IS_DIRSYNC(dir))
2460 ext4_handle_sync(handle);
2461
2462 out_stop:
2463 if (handle)
2464 ext4_journal_stop(handle);
2465 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2466 goto retry;
2467 return err;
2468 }
2469
2470 /*
2471 * routine to check that the specified directory is empty (for rmdir)
2472 */
2473 static int empty_dir(struct inode *inode)
2474 {
2475 unsigned int offset;
2476 struct buffer_head *bh;
2477 struct ext4_dir_entry_2 *de, *de1;
2478 struct super_block *sb;
2479 int err = 0;
2480
2481 if (ext4_has_inline_data(inode)) {
2482 int has_inline_data = 1;
2483
2484 err = empty_inline_dir(inode, &has_inline_data);
2485 if (has_inline_data)
2486 return err;
2487 }
2488
2489 sb = inode->i_sb;
2490 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2491 EXT4_ERROR_INODE(inode, "invalid size");
2492 return 1;
2493 }
2494 bh = ext4_read_dirblock(inode, 0, EITHER);
2495 if (IS_ERR(bh))
2496 return 1;
2497
2498 de = (struct ext4_dir_entry_2 *) bh->b_data;
2499 de1 = ext4_next_entry(de, sb->s_blocksize);
2500 if (le32_to_cpu(de->inode) != inode->i_ino ||
2501 !le32_to_cpu(de1->inode) ||
2502 strcmp(".", de->name) ||
2503 strcmp("..", de1->name)) {
2504 ext4_warning(inode->i_sb,
2505 "bad directory (dir #%lu) - no `.' or `..'",
2506 inode->i_ino);
2507 brelse(bh);
2508 return 1;
2509 }
2510 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2511 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2512 de = ext4_next_entry(de1, sb->s_blocksize);
2513 while (offset < inode->i_size) {
2514 if (!bh ||
2515 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2516 unsigned int lblock;
2517 err = 0;
2518 brelse(bh);
2519 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2520 bh = ext4_read_dirblock(inode, lblock, EITHER);
2521 if (IS_ERR(bh))
2522 return 1;
2523 de = (struct ext4_dir_entry_2 *) bh->b_data;
2524 }
2525 if (ext4_check_dir_entry(inode, NULL, de, bh,
2526 bh->b_data, bh->b_size, offset)) {
2527 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2528 sb->s_blocksize);
2529 offset = (offset | (sb->s_blocksize - 1)) + 1;
2530 continue;
2531 }
2532 if (le32_to_cpu(de->inode)) {
2533 brelse(bh);
2534 return 0;
2535 }
2536 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2537 de = ext4_next_entry(de, sb->s_blocksize);
2538 }
2539 brelse(bh);
2540 return 1;
2541 }
2542
2543 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
2544 * such inodes, starting at the superblock, in case we crash before the
2545 * file is closed/deleted, or in case the inode truncate spans multiple
2546 * transactions and the last transaction is not recovered after a crash.
2547 *
2548 * At filesystem recovery time, we walk this list deleting unlinked
2549 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2550 */
2551 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2552 {
2553 struct super_block *sb = inode->i_sb;
2554 struct ext4_iloc iloc;
2555 int err = 0, rc;
2556
2557 if (!EXT4_SB(sb)->s_journal)
2558 return 0;
2559
2560 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2561 if (!list_empty(&EXT4_I(inode)->i_orphan))
2562 goto out_unlock;
2563
2564 /*
2565 * Orphan handling is only valid for files with data blocks
2566 * being truncated, or files being unlinked. Note that we either
2567 * hold i_mutex, or the inode can not be referenced from outside,
2568 * so i_nlink should not be bumped due to race
2569 */
2570 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2571 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2572
2573 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2574 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2575 if (err)
2576 goto out_unlock;
2577
2578 err = ext4_reserve_inode_write(handle, inode, &iloc);
2579 if (err)
2580 goto out_unlock;
2581 /*
2582 * Due to previous errors inode may be already a part of on-disk
2583 * orphan list. If so skip on-disk list modification.
2584 */
2585 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2586 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2587 goto mem_insert;
2588
2589 /* Insert this inode at the head of the on-disk orphan list... */
2590 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2591 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2592 err = ext4_handle_dirty_super(handle, sb);
2593 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2594 if (!err)
2595 err = rc;
2596
2597 /* Only add to the head of the in-memory list if all the
2598 * previous operations succeeded. If the orphan_add is going to
2599 * fail (possibly taking the journal offline), we can't risk
2600 * leaving the inode on the orphan list: stray orphan-list
2601 * entries can cause panics at unmount time.
2602 *
2603 * This is safe: on error we're going to ignore the orphan list
2604 * anyway on the next recovery. */
2605 mem_insert:
2606 if (!err)
2607 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2608
2609 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2610 jbd_debug(4, "orphan inode %lu will point to %d\n",
2611 inode->i_ino, NEXT_ORPHAN(inode));
2612 out_unlock:
2613 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2614 ext4_std_error(inode->i_sb, err);
2615 return err;
2616 }
2617
2618 /*
2619 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2620 * of such inodes stored on disk, because it is finally being cleaned up.
2621 */
2622 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2623 {
2624 struct list_head *prev;
2625 struct ext4_inode_info *ei = EXT4_I(inode);
2626 struct ext4_sb_info *sbi;
2627 __u32 ino_next;
2628 struct ext4_iloc iloc;
2629 int err = 0;
2630
2631 if ((!EXT4_SB(inode->i_sb)->s_journal) &&
2632 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS))
2633 return 0;
2634
2635 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2636 if (list_empty(&ei->i_orphan))
2637 goto out;
2638
2639 ino_next = NEXT_ORPHAN(inode);
2640 prev = ei->i_orphan.prev;
2641 sbi = EXT4_SB(inode->i_sb);
2642
2643 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2644
2645 list_del_init(&ei->i_orphan);
2646
2647 /* If we're on an error path, we may not have a valid
2648 * transaction handle with which to update the orphan list on
2649 * disk, but we still need to remove the inode from the linked
2650 * list in memory. */
2651 if (!handle)
2652 goto out;
2653
2654 err = ext4_reserve_inode_write(handle, inode, &iloc);
2655 if (err)
2656 goto out_err;
2657
2658 if (prev == &sbi->s_orphan) {
2659 jbd_debug(4, "superblock will point to %u\n", ino_next);
2660 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2661 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2662 if (err)
2663 goto out_brelse;
2664 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2665 err = ext4_handle_dirty_super(handle, inode->i_sb);
2666 } else {
2667 struct ext4_iloc iloc2;
2668 struct inode *i_prev =
2669 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2670
2671 jbd_debug(4, "orphan inode %lu will point to %u\n",
2672 i_prev->i_ino, ino_next);
2673 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2674 if (err)
2675 goto out_brelse;
2676 NEXT_ORPHAN(i_prev) = ino_next;
2677 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2678 }
2679 if (err)
2680 goto out_brelse;
2681 NEXT_ORPHAN(inode) = 0;
2682 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2683
2684 out_err:
2685 ext4_std_error(inode->i_sb, err);
2686 out:
2687 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2688 return err;
2689
2690 out_brelse:
2691 brelse(iloc.bh);
2692 goto out_err;
2693 }
2694
2695 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2696 {
2697 int retval;
2698 struct inode *inode;
2699 struct buffer_head *bh;
2700 struct ext4_dir_entry_2 *de;
2701 handle_t *handle = NULL;
2702
2703 /* Initialize quotas before so that eventual writes go in
2704 * separate transaction */
2705 dquot_initialize(dir);
2706 dquot_initialize(dentry->d_inode);
2707
2708 retval = -ENOENT;
2709 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2710 if (!bh)
2711 goto end_rmdir;
2712
2713 inode = dentry->d_inode;
2714
2715 retval = -EIO;
2716 if (le32_to_cpu(de->inode) != inode->i_ino)
2717 goto end_rmdir;
2718
2719 retval = -ENOTEMPTY;
2720 if (!empty_dir(inode))
2721 goto end_rmdir;
2722
2723 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2724 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2725 if (IS_ERR(handle)) {
2726 retval = PTR_ERR(handle);
2727 handle = NULL;
2728 goto end_rmdir;
2729 }
2730
2731 if (IS_DIRSYNC(dir))
2732 ext4_handle_sync(handle);
2733
2734 retval = ext4_delete_entry(handle, dir, de, bh);
2735 if (retval)
2736 goto end_rmdir;
2737 if (!EXT4_DIR_LINK_EMPTY(inode))
2738 ext4_warning(inode->i_sb,
2739 "empty directory has too many links (%d)",
2740 inode->i_nlink);
2741 inode->i_version++;
2742 clear_nlink(inode);
2743 /* There's no need to set i_disksize: the fact that i_nlink is
2744 * zero will ensure that the right thing happens during any
2745 * recovery. */
2746 inode->i_size = 0;
2747 ext4_orphan_add(handle, inode);
2748 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2749 ext4_mark_inode_dirty(handle, inode);
2750 ext4_dec_count(handle, dir);
2751 ext4_update_dx_flag(dir);
2752 ext4_mark_inode_dirty(handle, dir);
2753
2754 end_rmdir:
2755 brelse(bh);
2756 if (handle)
2757 ext4_journal_stop(handle);
2758 return retval;
2759 }
2760
2761 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2762 {
2763 int retval;
2764 struct inode *inode;
2765 struct buffer_head *bh;
2766 struct ext4_dir_entry_2 *de;
2767 handle_t *handle = NULL;
2768
2769 trace_ext4_unlink_enter(dir, dentry);
2770 /* Initialize quotas before so that eventual writes go
2771 * in separate transaction */
2772 dquot_initialize(dir);
2773 dquot_initialize(dentry->d_inode);
2774
2775 retval = -ENOENT;
2776 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2777 if (!bh)
2778 goto end_unlink;
2779
2780 inode = dentry->d_inode;
2781
2782 retval = -EIO;
2783 if (le32_to_cpu(de->inode) != inode->i_ino)
2784 goto end_unlink;
2785
2786 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2787 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2788 if (IS_ERR(handle)) {
2789 retval = PTR_ERR(handle);
2790 handle = NULL;
2791 goto end_unlink;
2792 }
2793
2794 if (IS_DIRSYNC(dir))
2795 ext4_handle_sync(handle);
2796
2797 if (!inode->i_nlink) {
2798 ext4_warning(inode->i_sb,
2799 "Deleting nonexistent file (%lu), %d",
2800 inode->i_ino, inode->i_nlink);
2801 set_nlink(inode, 1);
2802 }
2803 retval = ext4_delete_entry(handle, dir, de, bh);
2804 if (retval)
2805 goto end_unlink;
2806 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2807 ext4_update_dx_flag(dir);
2808 ext4_mark_inode_dirty(handle, dir);
2809 drop_nlink(inode);
2810 if (!inode->i_nlink)
2811 ext4_orphan_add(handle, inode);
2812 inode->i_ctime = ext4_current_time(inode);
2813 ext4_mark_inode_dirty(handle, inode);
2814 retval = 0;
2815
2816 end_unlink:
2817 brelse(bh);
2818 if (handle)
2819 ext4_journal_stop(handle);
2820 trace_ext4_unlink_exit(dentry, retval);
2821 return retval;
2822 }
2823
2824 static int ext4_symlink(struct inode *dir,
2825 struct dentry *dentry, const char *symname)
2826 {
2827 handle_t *handle;
2828 struct inode *inode;
2829 int l, err, retries = 0;
2830 int credits;
2831
2832 l = strlen(symname)+1;
2833 if (l > dir->i_sb->s_blocksize)
2834 return -ENAMETOOLONG;
2835
2836 dquot_initialize(dir);
2837
2838 if (l > EXT4_N_BLOCKS * 4) {
2839 /*
2840 * For non-fast symlinks, we just allocate inode and put it on
2841 * orphan list in the first transaction => we need bitmap,
2842 * group descriptor, sb, inode block, quota blocks, and
2843 * possibly selinux xattr blocks.
2844 */
2845 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2846 EXT4_XATTR_TRANS_BLOCKS;
2847 } else {
2848 /*
2849 * Fast symlink. We have to add entry to directory
2850 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2851 * allocate new inode (bitmap, group descriptor, inode block,
2852 * quota blocks, sb is already counted in previous macros).
2853 */
2854 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2855 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
2856 }
2857 retry:
2858 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
2859 &dentry->d_name, 0, NULL,
2860 EXT4_HT_DIR, credits);
2861 handle = ext4_journal_current_handle();
2862 err = PTR_ERR(inode);
2863 if (IS_ERR(inode))
2864 goto out_stop;
2865
2866 if (l > EXT4_N_BLOCKS * 4) {
2867 inode->i_op = &ext4_symlink_inode_operations;
2868 ext4_set_aops(inode);
2869 /*
2870 * We cannot call page_symlink() with transaction started
2871 * because it calls into ext4_write_begin() which can wait
2872 * for transaction commit if we are running out of space
2873 * and thus we deadlock. So we have to stop transaction now
2874 * and restart it when symlink contents is written.
2875 *
2876 * To keep fs consistent in case of crash, we have to put inode
2877 * to orphan list in the mean time.
2878 */
2879 drop_nlink(inode);
2880 err = ext4_orphan_add(handle, inode);
2881 ext4_journal_stop(handle);
2882 if (err)
2883 goto err_drop_inode;
2884 err = __page_symlink(inode, symname, l, 1);
2885 if (err)
2886 goto err_drop_inode;
2887 /*
2888 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2889 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2890 */
2891 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2892 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2893 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2894 if (IS_ERR(handle)) {
2895 err = PTR_ERR(handle);
2896 goto err_drop_inode;
2897 }
2898 set_nlink(inode, 1);
2899 err = ext4_orphan_del(handle, inode);
2900 if (err) {
2901 ext4_journal_stop(handle);
2902 clear_nlink(inode);
2903 goto err_drop_inode;
2904 }
2905 } else {
2906 /* clear the extent format for fast symlink */
2907 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2908 inode->i_op = &ext4_fast_symlink_inode_operations;
2909 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2910 inode->i_size = l-1;
2911 }
2912 EXT4_I(inode)->i_disksize = inode->i_size;
2913 err = ext4_add_nondir(handle, dentry, inode);
2914 if (!err && IS_DIRSYNC(dir))
2915 ext4_handle_sync(handle);
2916
2917 out_stop:
2918 if (handle)
2919 ext4_journal_stop(handle);
2920 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2921 goto retry;
2922 return err;
2923 err_drop_inode:
2924 unlock_new_inode(inode);
2925 iput(inode);
2926 return err;
2927 }
2928
2929 static int ext4_link(struct dentry *old_dentry,
2930 struct inode *dir, struct dentry *dentry)
2931 {
2932 handle_t *handle;
2933 struct inode *inode = old_dentry->d_inode;
2934 int err, retries = 0;
2935
2936 if (inode->i_nlink >= EXT4_LINK_MAX)
2937 return -EMLINK;
2938
2939 dquot_initialize(dir);
2940
2941 retry:
2942 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2943 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2944 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
2945 if (IS_ERR(handle))
2946 return PTR_ERR(handle);
2947
2948 if (IS_DIRSYNC(dir))
2949 ext4_handle_sync(handle);
2950
2951 inode->i_ctime = ext4_current_time(inode);
2952 ext4_inc_count(handle, inode);
2953 ihold(inode);
2954
2955 err = ext4_add_entry(handle, dentry, inode);
2956 if (!err) {
2957 ext4_mark_inode_dirty(handle, inode);
2958 /* this can happen only for tmpfile being
2959 * linked the first time
2960 */
2961 if (inode->i_nlink == 1)
2962 ext4_orphan_del(handle, inode);
2963 d_instantiate(dentry, inode);
2964 } else {
2965 drop_nlink(inode);
2966 iput(inode);
2967 }
2968 ext4_journal_stop(handle);
2969 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2970 goto retry;
2971 return err;
2972 }
2973
2974
2975 /*
2976 * Try to find buffer head where contains the parent block.
2977 * It should be the inode block if it is inlined or the 1st block
2978 * if it is a normal dir.
2979 */
2980 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
2981 struct inode *inode,
2982 int *retval,
2983 struct ext4_dir_entry_2 **parent_de,
2984 int *inlined)
2985 {
2986 struct buffer_head *bh;
2987
2988 if (!ext4_has_inline_data(inode)) {
2989 bh = ext4_read_dirblock(inode, 0, EITHER);
2990 if (IS_ERR(bh)) {
2991 *retval = PTR_ERR(bh);
2992 return NULL;
2993 }
2994 *parent_de = ext4_next_entry(
2995 (struct ext4_dir_entry_2 *)bh->b_data,
2996 inode->i_sb->s_blocksize);
2997 return bh;
2998 }
2999
3000 *inlined = 1;
3001 return ext4_get_first_inline_block(inode, parent_de, retval);
3002 }
3003
3004 /*
3005 * Anybody can rename anything with this: the permission checks are left to the
3006 * higher-level routines.
3007 *
3008 * n.b. old_{dentry,inode) refers to the source dentry/inode
3009 * while new_{dentry,inode) refers to the destination dentry/inode
3010 * This comes from rename(const char *oldpath, const char *newpath)
3011 */
3012 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3013 struct inode *new_dir, struct dentry *new_dentry)
3014 {
3015 handle_t *handle = NULL;
3016 struct inode *old_inode, *new_inode;
3017 struct buffer_head *old_bh, *new_bh, *dir_bh;
3018 struct ext4_dir_entry_2 *old_de, *new_de;
3019 int retval;
3020 int inlined = 0, new_inlined = 0;
3021 struct ext4_dir_entry_2 *parent_de;
3022
3023 dquot_initialize(old_dir);
3024 dquot_initialize(new_dir);
3025
3026 old_bh = new_bh = dir_bh = NULL;
3027
3028 /* Initialize quotas before so that eventual writes go
3029 * in separate transaction */
3030 if (new_dentry->d_inode)
3031 dquot_initialize(new_dentry->d_inode);
3032
3033 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de, NULL);
3034 /*
3035 * Check for inode number is _not_ due to possible IO errors.
3036 * We might rmdir the source, keep it as pwd of some process
3037 * and merrily kill the link to whatever was created under the
3038 * same name. Goodbye sticky bit ;-<
3039 */
3040 old_inode = old_dentry->d_inode;
3041 retval = -ENOENT;
3042 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
3043 goto end_rename;
3044
3045 new_inode = new_dentry->d_inode;
3046 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name,
3047 &new_de, &new_inlined);
3048 if (new_bh) {
3049 if (!new_inode) {
3050 brelse(new_bh);
3051 new_bh = NULL;
3052 }
3053 }
3054 if (new_inode && !test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
3055 ext4_alloc_da_blocks(old_inode);
3056
3057 handle = ext4_journal_start(old_dir, EXT4_HT_DIR,
3058 (2 * EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
3059 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3060 if (IS_ERR(handle))
3061 return PTR_ERR(handle);
3062
3063 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
3064 ext4_handle_sync(handle);
3065
3066 if (S_ISDIR(old_inode->i_mode)) {
3067 if (new_inode) {
3068 retval = -ENOTEMPTY;
3069 if (!empty_dir(new_inode))
3070 goto end_rename;
3071 }
3072 retval = -EIO;
3073 dir_bh = ext4_get_first_dir_block(handle, old_inode,
3074 &retval, &parent_de,
3075 &inlined);
3076 if (!dir_bh)
3077 goto end_rename;
3078 if (le32_to_cpu(parent_de->inode) != old_dir->i_ino)
3079 goto end_rename;
3080 retval = -EMLINK;
3081 if (!new_inode && new_dir != old_dir &&
3082 EXT4_DIR_LINK_MAX(new_dir))
3083 goto end_rename;
3084 BUFFER_TRACE(dir_bh, "get_write_access");
3085 retval = ext4_journal_get_write_access(handle, dir_bh);
3086 if (retval)
3087 goto end_rename;
3088 }
3089 if (!new_bh) {
3090 retval = ext4_add_entry(handle, new_dentry, old_inode);
3091 if (retval)
3092 goto end_rename;
3093 } else {
3094 BUFFER_TRACE(new_bh, "get write access");
3095 retval = ext4_journal_get_write_access(handle, new_bh);
3096 if (retval)
3097 goto end_rename;
3098 new_de->inode = cpu_to_le32(old_inode->i_ino);
3099 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
3100 EXT4_FEATURE_INCOMPAT_FILETYPE))
3101 new_de->file_type = old_de->file_type;
3102 new_dir->i_version++;
3103 new_dir->i_ctime = new_dir->i_mtime =
3104 ext4_current_time(new_dir);
3105 ext4_mark_inode_dirty(handle, new_dir);
3106 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
3107 if (!new_inlined) {
3108 retval = ext4_handle_dirty_dirent_node(handle,
3109 new_dir, new_bh);
3110 if (unlikely(retval)) {
3111 ext4_std_error(new_dir->i_sb, retval);
3112 goto end_rename;
3113 }
3114 }
3115 brelse(new_bh);
3116 new_bh = NULL;
3117 }
3118
3119 /*
3120 * Like most other Unix systems, set the ctime for inodes on a
3121 * rename.
3122 */
3123 old_inode->i_ctime = ext4_current_time(old_inode);
3124 ext4_mark_inode_dirty(handle, old_inode);
3125
3126 /*
3127 * ok, that's it
3128 */
3129 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
3130 old_de->name_len != old_dentry->d_name.len ||
3131 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
3132 (retval = ext4_delete_entry(handle, old_dir,
3133 old_de, old_bh)) == -ENOENT) {
3134 /* old_de could have moved from under us during htree split, so
3135 * make sure that we are deleting the right entry. We might
3136 * also be pointing to a stale entry in the unused part of
3137 * old_bh so just checking inum and the name isn't enough. */
3138 struct buffer_head *old_bh2;
3139 struct ext4_dir_entry_2 *old_de2;
3140
3141 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name,
3142 &old_de2, NULL);
3143 if (old_bh2) {
3144 retval = ext4_delete_entry(handle, old_dir,
3145 old_de2, old_bh2);
3146 brelse(old_bh2);
3147 }
3148 }
3149 if (retval) {
3150 ext4_warning(old_dir->i_sb,
3151 "Deleting old file (%lu), %d, error=%d",
3152 old_dir->i_ino, old_dir->i_nlink, retval);
3153 }
3154
3155 if (new_inode) {
3156 ext4_dec_count(handle, new_inode);
3157 new_inode->i_ctime = ext4_current_time(new_inode);
3158 }
3159 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
3160 ext4_update_dx_flag(old_dir);
3161 if (dir_bh) {
3162 parent_de->inode = cpu_to_le32(new_dir->i_ino);
3163 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
3164 if (!inlined) {
3165 if (is_dx(old_inode)) {
3166 retval = ext4_handle_dirty_dx_node(handle,
3167 old_inode,
3168 dir_bh);
3169 } else {
3170 retval = ext4_handle_dirty_dirent_node(handle,
3171 old_inode, dir_bh);
3172 }
3173 } else {
3174 retval = ext4_mark_inode_dirty(handle, old_inode);
3175 }
3176 if (retval) {
3177 ext4_std_error(old_dir->i_sb, retval);
3178 goto end_rename;
3179 }
3180 ext4_dec_count(handle, old_dir);
3181 if (new_inode) {
3182 /* checked empty_dir above, can't have another parent,
3183 * ext4_dec_count() won't work for many-linked dirs */
3184 clear_nlink(new_inode);
3185 } else {
3186 ext4_inc_count(handle, new_dir);
3187 ext4_update_dx_flag(new_dir);
3188 ext4_mark_inode_dirty(handle, new_dir);
3189 }
3190 }
3191 ext4_mark_inode_dirty(handle, old_dir);
3192 if (new_inode) {
3193 ext4_mark_inode_dirty(handle, new_inode);
3194 if (!new_inode->i_nlink)
3195 ext4_orphan_add(handle, new_inode);
3196 }
3197 retval = 0;
3198
3199 end_rename:
3200 brelse(dir_bh);
3201 brelse(old_bh);
3202 brelse(new_bh);
3203 if (handle)
3204 ext4_journal_stop(handle);
3205 return retval;
3206 }
3207
3208 /*
3209 * directories can handle most operations...
3210 */
3211 const struct inode_operations ext4_dir_inode_operations = {
3212 .create = ext4_create,
3213 .lookup = ext4_lookup,
3214 .link = ext4_link,
3215 .unlink = ext4_unlink,
3216 .symlink = ext4_symlink,
3217 .mkdir = ext4_mkdir,
3218 .rmdir = ext4_rmdir,
3219 .mknod = ext4_mknod,
3220 .tmpfile = ext4_tmpfile,
3221 .rename = ext4_rename,
3222 .setattr = ext4_setattr,
3223 .setxattr = generic_setxattr,
3224 .getxattr = generic_getxattr,
3225 .listxattr = ext4_listxattr,
3226 .removexattr = generic_removexattr,
3227 .get_acl = ext4_get_acl,
3228 .fiemap = ext4_fiemap,
3229 };
3230
3231 const struct inode_operations ext4_special_inode_operations = {
3232 .setattr = ext4_setattr,
3233 .setxattr = generic_setxattr,
3234 .getxattr = generic_getxattr,
3235 .listxattr = ext4_listxattr,
3236 .removexattr = generic_removexattr,
3237 .get_acl = ext4_get_acl,
3238 };
Linux with added FPC-III support