1 /*
2 * linux/fs/ext3/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/quotaops.h>
28 #include "ext3.h"
29 #include "namei.h"
30 #include "xattr.h"
31 #include "acl.h"
32
33 /*
34 * define how far ahead to read directories while searching them.
35 */
36 #define NAMEI_RA_CHUNKS 2
37 #define NAMEI_RA_BLOCKS 4
38 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
39 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
40
41 static struct buffer_head *ext3_append(handle_t *handle,
42 struct inode *inode,
43 u32 *block, int *err)
44 {
45 struct buffer_head *bh;
46
47 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
48
49 bh = ext3_bread(handle, inode, *block, 1, err);
50 if (bh) {
51 inode->i_size += inode->i_sb->s_blocksize;
52 EXT3_I(inode)->i_disksize = inode->i_size;
53 *err = ext3_journal_get_write_access(handle, bh);
54 if (*err) {
55 brelse(bh);
56 bh = NULL;
57 }
58 }
59 return bh;
60 }
61
62 #ifndef assert
63 #define assert(test) J_ASSERT(test)
64 #endif
65
66 #ifdef DX_DEBUG
67 #define dxtrace(command) command
68 #else
69 #define dxtrace(command)
70 #endif
71
72 struct fake_dirent
73 {
74 __le32 inode;
75 __le16 rec_len;
76 u8 name_len;
77 u8 file_type;
78 };
79
80 struct dx_countlimit
81 {
82 __le16 limit;
83 __le16 count;
84 };
85
86 struct dx_entry
87 {
88 __le32 hash;
89 __le32 block;
90 };
91
92 /*
93 * dx_root_info is laid out so that if it should somehow get overlaid by a
94 * dirent the two low bits of the hash version will be zero. Therefore, the
95 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
96 */
97
98 struct dx_root
99 {
100 struct fake_dirent dot;
101 char dot_name[4];
102 struct fake_dirent dotdot;
103 char dotdot_name[4];
104 struct dx_root_info
105 {
106 __le32 reserved_zero;
107 u8 hash_version;
108 u8 info_length; /* 8 */
109 u8 indirect_levels;
110 u8 unused_flags;
111 }
112 info;
113 struct dx_entry entries[0];
114 };
115
116 struct dx_node
117 {
118 struct fake_dirent fake;
119 struct dx_entry entries[0];
120 };
121
122
123 struct dx_frame
124 {
125 struct buffer_head *bh;
126 struct dx_entry *entries;
127 struct dx_entry *at;
128 };
129
130 struct dx_map_entry
131 {
132 u32 hash;
133 u16 offs;
134 u16 size;
135 };
136
137 static inline unsigned dx_get_block (struct dx_entry *entry);
138 static void dx_set_block (struct dx_entry *entry, unsigned value);
139 static inline unsigned dx_get_hash (struct dx_entry *entry);
140 static void dx_set_hash (struct dx_entry *entry, unsigned value);
141 static unsigned dx_get_count (struct dx_entry *entries);
142 static unsigned dx_get_limit (struct dx_entry *entries);
143 static void dx_set_count (struct dx_entry *entries, unsigned value);
144 static void dx_set_limit (struct dx_entry *entries, unsigned value);
145 static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
146 static unsigned dx_node_limit (struct inode *dir);
147 static struct dx_frame *dx_probe(struct qstr *entry,
148 struct inode *dir,
149 struct dx_hash_info *hinfo,
150 struct dx_frame *frame,
151 int *err);
152 static void dx_release (struct dx_frame *frames);
153 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
154 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
155 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
156 static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
157 struct dx_map_entry *offsets, int count);
158 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize);
159 static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
160 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
161 struct dx_frame *frame,
162 struct dx_frame *frames,
163 __u32 *start_hash);
164 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
165 struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
166 int *err);
167 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
168 struct inode *inode);
169
170 /*
171 * p is at least 6 bytes before the end of page
172 */
173 static inline struct ext3_dir_entry_2 *
174 ext3_next_entry(struct ext3_dir_entry_2 *p)
175 {
176 return (struct ext3_dir_entry_2 *)((char *)p +
177 ext3_rec_len_from_disk(p->rec_len));
178 }
179
180 /*
181 * Future: use high four bits of block for coalesce-on-delete flags
182 * Mask them off for now.
183 */
184
185 static inline unsigned dx_get_block (struct dx_entry *entry)
186 {
187 return le32_to_cpu(entry->block) & 0x00ffffff;
188 }
189
190 static inline void dx_set_block (struct dx_entry *entry, unsigned value)
191 {
192 entry->block = cpu_to_le32(value);
193 }
194
195 static inline unsigned dx_get_hash (struct dx_entry *entry)
196 {
197 return le32_to_cpu(entry->hash);
198 }
199
200 static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
201 {
202 entry->hash = cpu_to_le32(value);
203 }
204
205 static inline unsigned dx_get_count (struct dx_entry *entries)
206 {
207 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
208 }
209
210 static inline unsigned dx_get_limit (struct dx_entry *entries)
211 {
212 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
213 }
214
215 static inline void dx_set_count (struct dx_entry *entries, unsigned value)
216 {
217 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
218 }
219
220 static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
221 {
222 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
223 }
224
225 static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
226 {
227 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
228 EXT3_DIR_REC_LEN(2) - infosize;
229 return entry_space / sizeof(struct dx_entry);
230 }
231
232 static inline unsigned dx_node_limit (struct inode *dir)
233 {
234 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
235 return entry_space / sizeof(struct dx_entry);
236 }
237
238 /*
239 * Debug
240 */
241 #ifdef DX_DEBUG
242 static void dx_show_index (char * label, struct dx_entry *entries)
243 {
244 int i, n = dx_get_count (entries);
245 printk("%s index ", label);
246 for (i = 0; i < n; i++)
247 {
248 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
249 }
250 printk("\n");
251 }
252
253 struct stats
254 {
255 unsigned names;
256 unsigned space;
257 unsigned bcount;
258 };
259
260 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
261 int size, int show_names)
262 {
263 unsigned names = 0, space = 0;
264 char *base = (char *) de;
265 struct dx_hash_info h = *hinfo;
266
267 printk("names: ");
268 while ((char *) de < base + size)
269 {
270 if (de->inode)
271 {
272 if (show_names)
273 {
274 int len = de->name_len;
275 char *name = de->name;
276 while (len--) printk("%c", *name++);
277 ext3fs_dirhash(de->name, de->name_len, &h);
278 printk(":%x.%u ", h.hash,
279 (unsigned) ((char *) de - base));
280 }
281 space += EXT3_DIR_REC_LEN(de->name_len);
282 names++;
283 }
284 de = ext3_next_entry(de);
285 }
286 printk("(%i)\n", names);
287 return (struct stats) { names, space, 1 };
288 }
289
290 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291 struct dx_entry *entries, int levels)
292 {
293 unsigned blocksize = dir->i_sb->s_blocksize;
294 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
295 unsigned bcount = 0;
296 struct buffer_head *bh;
297 int err;
298 printk("%i indexed blocks...\n", count);
299 for (i = 0; i < count; i++, entries++)
300 {
301 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
302 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
303 struct stats stats;
304 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
305 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
306 stats = levels?
307 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
308 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
309 names += stats.names;
310 space += stats.space;
311 bcount += stats.bcount;
312 brelse (bh);
313 }
314 if (bcount)
315 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
316 names, space/bcount,(space/bcount)*100/blocksize);
317 return (struct stats) { names, space, bcount};
318 }
319 #endif /* DX_DEBUG */
320
321 /*
322 * Probe for a directory leaf block to search.
323 *
324 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
325 * error in the directory index, and the caller should fall back to
326 * searching the directory normally. The callers of dx_probe **MUST**
327 * check for this error code, and make sure it never gets reflected
328 * back to userspace.
329 */
330 static struct dx_frame *
331 dx_probe(struct qstr *entry, struct inode *dir,
332 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
333 {
334 unsigned count, indirect;
335 struct dx_entry *at, *entries, *p, *q, *m;
336 struct dx_root *root;
337 struct buffer_head *bh;
338 struct dx_frame *frame = frame_in;
339 u32 hash;
340
341 frame->bh = NULL;
342 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
343 goto fail;
344 root = (struct dx_root *) bh->b_data;
345 if (root->info.hash_version != DX_HASH_TEA &&
346 root->info.hash_version != DX_HASH_HALF_MD4 &&
347 root->info.hash_version != DX_HASH_LEGACY) {
348 ext3_warning(dir->i_sb, __func__,
349 "Unrecognised inode hash code %d",
350 root->info.hash_version);
351 brelse(bh);
352 *err = ERR_BAD_DX_DIR;
353 goto fail;
354 }
355 hinfo->hash_version = root->info.hash_version;
356 if (hinfo->hash_version <= DX_HASH_TEA)
357 hinfo->hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
358 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
359 if (entry)
360 ext3fs_dirhash(entry->name, entry->len, hinfo);
361 hash = hinfo->hash;
362
363 if (root->info.unused_flags & 1) {
364 ext3_warning(dir->i_sb, __func__,
365 "Unimplemented inode hash flags: %#06x",
366 root->info.unused_flags);
367 brelse(bh);
368 *err = ERR_BAD_DX_DIR;
369 goto fail;
370 }
371
372 if ((indirect = root->info.indirect_levels) > 1) {
373 ext3_warning(dir->i_sb, __func__,
374 "Unimplemented inode hash depth: %#06x",
375 root->info.indirect_levels);
376 brelse(bh);
377 *err = ERR_BAD_DX_DIR;
378 goto fail;
379 }
380
381 entries = (struct dx_entry *) (((char *)&root->info) +
382 root->info.info_length);
383
384 if (dx_get_limit(entries) != dx_root_limit(dir,
385 root->info.info_length)) {
386 ext3_warning(dir->i_sb, __func__,
387 "dx entry: limit != root limit");
388 brelse(bh);
389 *err = ERR_BAD_DX_DIR;
390 goto fail;
391 }
392
393 dxtrace (printk("Look up %x", hash));
394 while (1)
395 {
396 count = dx_get_count(entries);
397 if (!count || count > dx_get_limit(entries)) {
398 ext3_warning(dir->i_sb, __func__,
399 "dx entry: no count or count > limit");
400 brelse(bh);
401 *err = ERR_BAD_DX_DIR;
402 goto fail2;
403 }
404
405 p = entries + 1;
406 q = entries + count - 1;
407 while (p <= q)
408 {
409 m = p + (q - p)/2;
410 dxtrace(printk("."));
411 if (dx_get_hash(m) > hash)
412 q = m - 1;
413 else
414 p = m + 1;
415 }
416
417 if (0) // linear search cross check
418 {
419 unsigned n = count - 1;
420 at = entries;
421 while (n--)
422 {
423 dxtrace(printk(","));
424 if (dx_get_hash(++at) > hash)
425 {
426 at--;
427 break;
428 }
429 }
430 assert (at == p - 1);
431 }
432
433 at = p - 1;
434 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
435 frame->bh = bh;
436 frame->entries = entries;
437 frame->at = at;
438 if (!indirect--) return frame;
439 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
440 goto fail2;
441 at = entries = ((struct dx_node *) bh->b_data)->entries;
442 if (dx_get_limit(entries) != dx_node_limit (dir)) {
443 ext3_warning(dir->i_sb, __func__,
444 "dx entry: limit != node limit");
445 brelse(bh);
446 *err = ERR_BAD_DX_DIR;
447 goto fail2;
448 }
449 frame++;
450 frame->bh = NULL;
451 }
452 fail2:
453 while (frame >= frame_in) {
454 brelse(frame->bh);
455 frame--;
456 }
457 fail:
458 if (*err == ERR_BAD_DX_DIR)
459 ext3_warning(dir->i_sb, __func__,
460 "Corrupt dir inode %ld, running e2fsck is "
461 "recommended.", dir->i_ino);
462 return NULL;
463 }
464
465 static void dx_release (struct dx_frame *frames)
466 {
467 if (frames[0].bh == NULL)
468 return;
469
470 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
471 brelse(frames[1].bh);
472 brelse(frames[0].bh);
473 }
474
475 /*
476 * This function increments the frame pointer to search the next leaf
477 * block, and reads in the necessary intervening nodes if the search
478 * should be necessary. Whether or not the search is necessary is
479 * controlled by the hash parameter. If the hash value is even, then
480 * the search is only continued if the next block starts with that
481 * hash value. This is used if we are searching for a specific file.
482 *
483 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
484 *
485 * This function returns 1 if the caller should continue to search,
486 * or 0 if it should not. If there is an error reading one of the
487 * index blocks, it will a negative error code.
488 *
489 * If start_hash is non-null, it will be filled in with the starting
490 * hash of the next page.
491 */
492 static int ext3_htree_next_block(struct inode *dir, __u32 hash,
493 struct dx_frame *frame,
494 struct dx_frame *frames,
495 __u32 *start_hash)
496 {
497 struct dx_frame *p;
498 struct buffer_head *bh;
499 int err, num_frames = 0;
500 __u32 bhash;
501
502 p = frame;
503 /*
504 * Find the next leaf page by incrementing the frame pointer.
505 * If we run out of entries in the interior node, loop around and
506 * increment pointer in the parent node. When we break out of
507 * this loop, num_frames indicates the number of interior
508 * nodes need to be read.
509 */
510 while (1) {
511 if (++(p->at) < p->entries + dx_get_count(p->entries))
512 break;
513 if (p == frames)
514 return 0;
515 num_frames++;
516 p--;
517 }
518
519 /*
520 * If the hash is 1, then continue only if the next page has a
521 * continuation hash of any value. This is used for readdir
522 * handling. Otherwise, check to see if the hash matches the
523 * desired contiuation hash. If it doesn't, return since
524 * there's no point to read in the successive index pages.
525 */
526 bhash = dx_get_hash(p->at);
527 if (start_hash)
528 *start_hash = bhash;
529 if ((hash & 1) == 0) {
530 if ((bhash & ~1) != hash)
531 return 0;
532 }
533 /*
534 * If the hash is HASH_NB_ALWAYS, we always go to the next
535 * block so no check is necessary
536 */
537 while (num_frames--) {
538 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
539 0, &err)))
540 return err; /* Failure */
541 p++;
542 brelse (p->bh);
543 p->bh = bh;
544 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
545 }
546 return 1;
547 }
548
549
550 /*
551 * This function fills a red-black tree with information from a
552 * directory block. It returns the number directory entries loaded
553 * into the tree. If there is an error it is returned in err.
554 */
555 static int htree_dirblock_to_tree(struct file *dir_file,
556 struct inode *dir, int block,
557 struct dx_hash_info *hinfo,
558 __u32 start_hash, __u32 start_minor_hash)
559 {
560 struct buffer_head *bh;
561 struct ext3_dir_entry_2 *de, *top;
562 int err, count = 0;
563
564 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
565 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
566 return err;
567
568 de = (struct ext3_dir_entry_2 *) bh->b_data;
569 top = (struct ext3_dir_entry_2 *) ((char *) de +
570 dir->i_sb->s_blocksize -
571 EXT3_DIR_REC_LEN(0));
572 for (; de < top; de = ext3_next_entry(de)) {
573 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
574 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
575 +((char *)de - bh->b_data))) {
576 /* On error, skip the f_pos to the next block. */
577 dir_file->f_pos = (dir_file->f_pos |
578 (dir->i_sb->s_blocksize - 1)) + 1;
579 brelse (bh);
580 return count;
581 }
582 ext3fs_dirhash(de->name, de->name_len, hinfo);
583 if ((hinfo->hash < start_hash) ||
584 ((hinfo->hash == start_hash) &&
585 (hinfo->minor_hash < start_minor_hash)))
586 continue;
587 if (de->inode == 0)
588 continue;
589 if ((err = ext3_htree_store_dirent(dir_file,
590 hinfo->hash, hinfo->minor_hash, de)) != 0) {
591 brelse(bh);
592 return err;
593 }
594 count++;
595 }
596 brelse(bh);
597 return count;
598 }
599
600
601 /*
602 * This function fills a red-black tree with information from a
603 * directory. We start scanning the directory in hash order, starting
604 * at start_hash and start_minor_hash.
605 *
606 * This function returns the number of entries inserted into the tree,
607 * or a negative error code.
608 */
609 int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
610 __u32 start_minor_hash, __u32 *next_hash)
611 {
612 struct dx_hash_info hinfo;
613 struct ext3_dir_entry_2 *de;
614 struct dx_frame frames[2], *frame;
615 struct inode *dir;
616 int block, err;
617 int count = 0;
618 int ret;
619 __u32 hashval;
620
621 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
622 start_minor_hash));
623 dir = dir_file->f_path.dentry->d_inode;
624 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
625 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
626 if (hinfo.hash_version <= DX_HASH_TEA)
627 hinfo.hash_version +=
628 EXT3_SB(dir->i_sb)->s_hash_unsigned;
629 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
630 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
631 start_hash, start_minor_hash);
632 *next_hash = ~0;
633 return count;
634 }
635 hinfo.hash = start_hash;
636 hinfo.minor_hash = 0;
637 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
638 if (!frame)
639 return err;
640
641 /* Add '.' and '..' from the htree header */
642 if (!start_hash && !start_minor_hash) {
643 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
644 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
645 goto errout;
646 count++;
647 }
648 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
649 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
650 de = ext3_next_entry(de);
651 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
652 goto errout;
653 count++;
654 }
655
656 while (1) {
657 block = dx_get_block(frame->at);
658 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
659 start_hash, start_minor_hash);
660 if (ret < 0) {
661 err = ret;
662 goto errout;
663 }
664 count += ret;
665 hashval = ~0;
666 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
667 frame, frames, &hashval);
668 *next_hash = hashval;
669 if (ret < 0) {
670 err = ret;
671 goto errout;
672 }
673 /*
674 * Stop if: (a) there are no more entries, or
675 * (b) we have inserted at least one entry and the
676 * next hash value is not a continuation
677 */
678 if ((ret == 0) ||
679 (count && ((hashval & 1) == 0)))
680 break;
681 }
682 dx_release(frames);
683 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
684 count, *next_hash));
685 return count;
686 errout:
687 dx_release(frames);
688 return (err);
689 }
690
691
692 /*
693 * Directory block splitting, compacting
694 */
695
696 /*
697 * Create map of hash values, offsets, and sizes, stored at end of block.
698 * Returns number of entries mapped.
699 */
700 static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
701 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
702 {
703 int count = 0;
704 char *base = (char *) de;
705 struct dx_hash_info h = *hinfo;
706
707 while ((char *) de < base + blocksize)
708 {
709 if (de->name_len && de->inode) {
710 ext3fs_dirhash(de->name, de->name_len, &h);
711 map_tail--;
712 map_tail->hash = h.hash;
713 map_tail->offs = (u16) ((char *) de - base);
714 map_tail->size = le16_to_cpu(de->rec_len);
715 count++;
716 cond_resched();
717 }
718 /* XXX: do we need to check rec_len == 0 case? -Chris */
719 de = ext3_next_entry(de);
720 }
721 return count;
722 }
723
724 /* Sort map by hash value */
725 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
726 {
727 struct dx_map_entry *p, *q, *top = map + count - 1;
728 int more;
729 /* Combsort until bubble sort doesn't suck */
730 while (count > 2)
731 {
732 count = count*10/13;
733 if (count - 9 < 2) /* 9, 10 -> 11 */
734 count = 11;
735 for (p = top, q = p - count; q >= map; p--, q--)
736 if (p->hash < q->hash)
737 swap(*p, *q);
738 }
739 /* Garden variety bubble sort */
740 do {
741 more = 0;
742 q = top;
743 while (q-- > map)
744 {
745 if (q[1].hash >= q[0].hash)
746 continue;
747 swap(*(q+1), *q);
748 more = 1;
749 }
750 } while(more);
751 }
752
753 static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
754 {
755 struct dx_entry *entries = frame->entries;
756 struct dx_entry *old = frame->at, *new = old + 1;
757 int count = dx_get_count(entries);
758
759 assert(count < dx_get_limit(entries));
760 assert(old < entries + count);
761 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
762 dx_set_hash(new, hash);
763 dx_set_block(new, block);
764 dx_set_count(entries, count + 1);
765 }
766
767 static void ext3_update_dx_flag(struct inode *inode)
768 {
769 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
770 EXT3_FEATURE_COMPAT_DIR_INDEX))
771 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
772 }
773
774 /*
775 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
776 *
777 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
778 * `de != NULL' is guaranteed by caller.
779 */
780 static inline int ext3_match (int len, const char * const name,
781 struct ext3_dir_entry_2 * de)
782 {
783 if (len != de->name_len)
784 return 0;
785 if (!de->inode)
786 return 0;
787 return !memcmp(name, de->name, len);
788 }
789
790 /*
791 * Returns 0 if not found, -1 on failure, and 1 on success
792 */
793 static inline int search_dirblock(struct buffer_head * bh,
794 struct inode *dir,
795 struct qstr *child,
796 unsigned long offset,
797 struct ext3_dir_entry_2 ** res_dir)
798 {
799 struct ext3_dir_entry_2 * de;
800 char * dlimit;
801 int de_len;
802 const char *name = child->name;
803 int namelen = child->len;
804
805 de = (struct ext3_dir_entry_2 *) bh->b_data;
806 dlimit = bh->b_data + dir->i_sb->s_blocksize;
807 while ((char *) de < dlimit) {
808 /* this code is executed quadratically often */
809 /* do minimal checking `by hand' */
810
811 if ((char *) de + namelen <= dlimit &&
812 ext3_match (namelen, name, de)) {
813 /* found a match - just to be sure, do a full check */
814 if (!ext3_check_dir_entry("ext3_find_entry",
815 dir, de, bh, offset))
816 return -1;
817 *res_dir = de;
818 return 1;
819 }
820 /* prevent looping on a bad block */
821 de_len = ext3_rec_len_from_disk(de->rec_len);
822 if (de_len <= 0)
823 return -1;
824 offset += de_len;
825 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
826 }
827 return 0;
828 }
829
830
831 /*
832 * ext3_find_entry()
833 *
834 * finds an entry in the specified directory with the wanted name. It
835 * returns the cache buffer in which the entry was found, and the entry
836 * itself (as a parameter - res_dir). It does NOT read the inode of the
837 * entry - you'll have to do that yourself if you want to.
838 *
839 * The returned buffer_head has ->b_count elevated. The caller is expected
840 * to brelse() it when appropriate.
841 */
842 static struct buffer_head *ext3_find_entry(struct inode *dir,
843 struct qstr *entry,
844 struct ext3_dir_entry_2 **res_dir)
845 {
846 struct super_block * sb;
847 struct buffer_head * bh_use[NAMEI_RA_SIZE];
848 struct buffer_head * bh, *ret = NULL;
849 unsigned long start, block, b;
850 const u8 *name = entry->name;
851 int ra_max = 0; /* Number of bh's in the readahead
852 buffer, bh_use[] */
853 int ra_ptr = 0; /* Current index into readahead
854 buffer */
855 int num = 0;
856 int nblocks, i, err;
857 int namelen;
858
859 *res_dir = NULL;
860 sb = dir->i_sb;
861 namelen = entry->len;
862 if (namelen > EXT3_NAME_LEN)
863 return NULL;
864 if ((namelen <= 2) && (name[0] == '.') &&
865 (name[1] == '.' || name[1] == 0)) {
866 /*
867 * "." or ".." will only be in the first block
868 * NFS may look up ".."; "." should be handled by the VFS
869 */
870 block = start = 0;
871 nblocks = 1;
872 goto restart;
873 }
874 if (is_dx(dir)) {
875 bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
876 /*
877 * On success, or if the error was file not found,
878 * return. Otherwise, fall back to doing a search the
879 * old fashioned way.
880 */
881 if (bh || (err != ERR_BAD_DX_DIR))
882 return bh;
883 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
884 }
885 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
886 start = EXT3_I(dir)->i_dir_start_lookup;
887 if (start >= nblocks)
888 start = 0;
889 block = start;
890 restart:
891 do {
892 /*
893 * We deal with the read-ahead logic here.
894 */
895 if (ra_ptr >= ra_max) {
896 /* Refill the readahead buffer */
897 ra_ptr = 0;
898 b = block;
899 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
900 /*
901 * Terminate if we reach the end of the
902 * directory and must wrap, or if our
903 * search has finished at this block.
904 */
905 if (b >= nblocks || (num && block == start)) {
906 bh_use[ra_max] = NULL;
907 break;
908 }
909 num++;
910 bh = ext3_getblk(NULL, dir, b++, 0, &err);
911 bh_use[ra_max] = bh;
912 if (bh && !bh_uptodate_or_lock(bh)) {
913 get_bh(bh);
914 bh->b_end_io = end_buffer_read_sync;
915 submit_bh(READ | REQ_META | REQ_PRIO,
916 bh);
917 }
918 }
919 }
920 if ((bh = bh_use[ra_ptr++]) == NULL)
921 goto next;
922 wait_on_buffer(bh);
923 if (!buffer_uptodate(bh)) {
924 /* read error, skip block & hope for the best */
925 ext3_error(sb, __func__, "reading directory #%lu "
926 "offset %lu", dir->i_ino, block);
927 brelse(bh);
928 goto next;
929 }
930 i = search_dirblock(bh, dir, entry,
931 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
932 if (i == 1) {
933 EXT3_I(dir)->i_dir_start_lookup = block;
934 ret = bh;
935 goto cleanup_and_exit;
936 } else {
937 brelse(bh);
938 if (i < 0)
939 goto cleanup_and_exit;
940 }
941 next:
942 if (++block >= nblocks)
943 block = 0;
944 } while (block != start);
945
946 /*
947 * If the directory has grown while we were searching, then
948 * search the last part of the directory before giving up.
949 */
950 block = nblocks;
951 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
952 if (block < nblocks) {
953 start = 0;
954 goto restart;
955 }
956
957 cleanup_and_exit:
958 /* Clean up the read-ahead blocks */
959 for (; ra_ptr < ra_max; ra_ptr++)
960 brelse (bh_use[ra_ptr]);
961 return ret;
962 }
963
964 static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
965 struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
966 int *err)
967 {
968 struct super_block *sb = dir->i_sb;
969 struct dx_hash_info hinfo;
970 struct dx_frame frames[2], *frame;
971 struct buffer_head *bh;
972 unsigned long block;
973 int retval;
974
975 if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
976 return NULL;
977 do {
978 block = dx_get_block(frame->at);
979 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
980 goto errout;
981
982 retval = search_dirblock(bh, dir, entry,
983 block << EXT3_BLOCK_SIZE_BITS(sb),
984 res_dir);
985 if (retval == 1) {
986 dx_release(frames);
987 return bh;
988 }
989 brelse(bh);
990 if (retval == -1) {
991 *err = ERR_BAD_DX_DIR;
992 goto errout;
993 }
994
995 /* Check to see if we should continue to search */
996 retval = ext3_htree_next_block(dir, hinfo.hash, frame,
997 frames, NULL);
998 if (retval < 0) {
999 ext3_warning(sb, __func__,
1000 "error reading index page in directory #%lu",
1001 dir->i_ino);
1002 *err = retval;
1003 goto errout;
1004 }
1005 } while (retval == 1);
1006
1007 *err = -ENOENT;
1008 errout:
1009 dxtrace(printk("%s not found\n", entry->name));
1010 dx_release (frames);
1011 return NULL;
1012 }
1013
1014 static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, unsigned int flags)
1015 {
1016 struct inode * inode;
1017 struct ext3_dir_entry_2 * de;
1018 struct buffer_head * bh;
1019
1020 if (dentry->d_name.len > EXT3_NAME_LEN)
1021 return ERR_PTR(-ENAMETOOLONG);
1022
1023 bh = ext3_find_entry(dir, &dentry->d_name, &de);
1024 inode = NULL;
1025 if (bh) {
1026 unsigned long ino = le32_to_cpu(de->inode);
1027 brelse (bh);
1028 if (!ext3_valid_inum(dir->i_sb, ino)) {
1029 ext3_error(dir->i_sb, "ext3_lookup",
1030 "bad inode number: %lu", ino);
1031 return ERR_PTR(-EIO);
1032 }
1033 inode = ext3_iget(dir->i_sb, ino);
1034 if (inode == ERR_PTR(-ESTALE)) {
1035 ext3_error(dir->i_sb, __func__,
1036 "deleted inode referenced: %lu",
1037 ino);
1038 return ERR_PTR(-EIO);
1039 }
1040 }
1041 return d_splice_alias(inode, dentry);
1042 }
1043
1044
1045 struct dentry *ext3_get_parent(struct dentry *child)
1046 {
1047 unsigned long ino;
1048 struct qstr dotdot = QSTR_INIT("..", 2);
1049 struct ext3_dir_entry_2 * de;
1050 struct buffer_head *bh;
1051
1052 bh = ext3_find_entry(child->d_inode, &dotdot, &de);
1053 if (!bh)
1054 return ERR_PTR(-ENOENT);
1055 ino = le32_to_cpu(de->inode);
1056 brelse(bh);
1057
1058 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1059 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1060 "bad inode number: %lu", ino);
1061 return ERR_PTR(-EIO);
1062 }
1063
1064 return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1065 }
1066
1067 #define S_SHIFT 12
1068 static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1069 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1070 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1071 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1072 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1073 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1074 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1075 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1076 };
1077
1078 static inline void ext3_set_de_type(struct super_block *sb,
1079 struct ext3_dir_entry_2 *de,
1080 umode_t mode) {
1081 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1082 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1083 }
1084
1085 /*
1086 * Move count entries from end of map between two memory locations.
1087 * Returns pointer to last entry moved.
1088 */
1089 static struct ext3_dir_entry_2 *
1090 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1091 {
1092 unsigned rec_len = 0;
1093
1094 while (count--) {
1095 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1096 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1097 memcpy (to, de, rec_len);
1098 ((struct ext3_dir_entry_2 *) to)->rec_len =
1099 ext3_rec_len_to_disk(rec_len);
1100 de->inode = 0;
1101 map++;
1102 to += rec_len;
1103 }
1104 return (struct ext3_dir_entry_2 *) (to - rec_len);
1105 }
1106
1107 /*
1108 * Compact each dir entry in the range to the minimal rec_len.
1109 * Returns pointer to last entry in range.
1110 */
1111 static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize)
1112 {
1113 struct ext3_dir_entry_2 *next, *to, *prev;
1114 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)base;
1115 unsigned rec_len = 0;
1116
1117 prev = to = de;
1118 while ((char *)de < base + blocksize) {
1119 next = ext3_next_entry(de);
1120 if (de->inode && de->name_len) {
1121 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1122 if (de > to)
1123 memmove(to, de, rec_len);
1124 to->rec_len = ext3_rec_len_to_disk(rec_len);
1125 prev = to;
1126 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1127 }
1128 de = next;
1129 }
1130 return prev;
1131 }
1132
1133 /*
1134 * Split a full leaf block to make room for a new dir entry.
1135 * Allocate a new block, and move entries so that they are approx. equally full.
1136 * Returns pointer to de in block into which the new entry will be inserted.
1137 */
1138 static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1139 struct buffer_head **bh,struct dx_frame *frame,
1140 struct dx_hash_info *hinfo, int *error)
1141 {
1142 unsigned blocksize = dir->i_sb->s_blocksize;
1143 unsigned count, continued;
1144 struct buffer_head *bh2;
1145 u32 newblock;
1146 u32 hash2;
1147 struct dx_map_entry *map;
1148 char *data1 = (*bh)->b_data, *data2;
1149 unsigned split, move, size;
1150 struct ext3_dir_entry_2 *de = NULL, *de2;
1151 int err = 0, i;
1152
1153 bh2 = ext3_append (handle, dir, &newblock, &err);
1154 if (!(bh2)) {
1155 brelse(*bh);
1156 *bh = NULL;
1157 goto errout;
1158 }
1159
1160 BUFFER_TRACE(*bh, "get_write_access");
1161 err = ext3_journal_get_write_access(handle, *bh);
1162 if (err)
1163 goto journal_error;
1164
1165 BUFFER_TRACE(frame->bh, "get_write_access");
1166 err = ext3_journal_get_write_access(handle, frame->bh);
1167 if (err)
1168 goto journal_error;
1169
1170 data2 = bh2->b_data;
1171
1172 /* create map in the end of data2 block */
1173 map = (struct dx_map_entry *) (data2 + blocksize);
1174 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1175 blocksize, hinfo, map);
1176 map -= count;
1177 dx_sort_map (map, count);
1178 /* Split the existing block in the middle, size-wise */
1179 size = 0;
1180 move = 0;
1181 for (i = count-1; i >= 0; i--) {
1182 /* is more than half of this entry in 2nd half of the block? */
1183 if (size + map[i].size/2 > blocksize/2)
1184 break;
1185 size += map[i].size;
1186 move++;
1187 }
1188 /* map index at which we will split */
1189 split = count - move;
1190 hash2 = map[split].hash;
1191 continued = hash2 == map[split - 1].hash;
1192 dxtrace(printk("Split block %i at %x, %i/%i\n",
1193 dx_get_block(frame->at), hash2, split, count-split));
1194
1195 /* Fancy dance to stay within two buffers */
1196 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1197 de = dx_pack_dirents(data1,blocksize);
1198 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1199 de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1200 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1201 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1202
1203 /* Which block gets the new entry? */
1204 if (hinfo->hash >= hash2)
1205 {
1206 swap(*bh, bh2);
1207 de = de2;
1208 }
1209 dx_insert_block (frame, hash2 + continued, newblock);
1210 err = ext3_journal_dirty_metadata (handle, bh2);
1211 if (err)
1212 goto journal_error;
1213 err = ext3_journal_dirty_metadata (handle, frame->bh);
1214 if (err)
1215 goto journal_error;
1216 brelse (bh2);
1217 dxtrace(dx_show_index ("frame", frame->entries));
1218 return de;
1219
1220 journal_error:
1221 brelse(*bh);
1222 brelse(bh2);
1223 *bh = NULL;
1224 ext3_std_error(dir->i_sb, err);
1225 errout:
1226 *error = err;
1227 return NULL;
1228 }
1229
1230
1231 /*
1232 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1233 * it points to a directory entry which is guaranteed to be large
1234 * enough for new directory entry. If de is NULL, then
1235 * add_dirent_to_buf will attempt search the directory block for
1236 * space. It will return -ENOSPC if no space is available, and -EIO
1237 * and -EEXIST if directory entry already exists.
1238 *
1239 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1240 * all other cases bh is released.
1241 */
1242 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1243 struct inode *inode, struct ext3_dir_entry_2 *de,
1244 struct buffer_head * bh)
1245 {
1246 struct inode *dir = dentry->d_parent->d_inode;
1247 const char *name = dentry->d_name.name;
1248 int namelen = dentry->d_name.len;
1249 unsigned long offset = 0;
1250 unsigned short reclen;
1251 int nlen, rlen, err;
1252 char *top;
1253
1254 reclen = EXT3_DIR_REC_LEN(namelen);
1255 if (!de) {
1256 de = (struct ext3_dir_entry_2 *)bh->b_data;
1257 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1258 while ((char *) de <= top) {
1259 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1260 bh, offset)) {
1261 brelse (bh);
1262 return -EIO;
1263 }
1264 if (ext3_match (namelen, name, de)) {
1265 brelse (bh);
1266 return -EEXIST;
1267 }
1268 nlen = EXT3_DIR_REC_LEN(de->name_len);
1269 rlen = ext3_rec_len_from_disk(de->rec_len);
1270 if ((de->inode? rlen - nlen: rlen) >= reclen)
1271 break;
1272 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1273 offset += rlen;
1274 }
1275 if ((char *) de > top)
1276 return -ENOSPC;
1277 }
1278 BUFFER_TRACE(bh, "get_write_access");
1279 err = ext3_journal_get_write_access(handle, bh);
1280 if (err) {
1281 ext3_std_error(dir->i_sb, err);
1282 brelse(bh);
1283 return err;
1284 }
1285
1286 /* By now the buffer is marked for journaling */
1287 nlen = EXT3_DIR_REC_LEN(de->name_len);
1288 rlen = ext3_rec_len_from_disk(de->rec_len);
1289 if (de->inode) {
1290 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1291 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1292 de->rec_len = ext3_rec_len_to_disk(nlen);
1293 de = de1;
1294 }
1295 de->file_type = EXT3_FT_UNKNOWN;
1296 if (inode) {
1297 de->inode = cpu_to_le32(inode->i_ino);
1298 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1299 } else
1300 de->inode = 0;
1301 de->name_len = namelen;
1302 memcpy (de->name, name, namelen);
1303 /*
1304 * XXX shouldn't update any times until successful
1305 * completion of syscall, but too many callers depend
1306 * on this.
1307 *
1308 * XXX similarly, too many callers depend on
1309 * ext3_new_inode() setting the times, but error
1310 * recovery deletes the inode, so the worst that can
1311 * happen is that the times are slightly out of date
1312 * and/or different from the directory change time.
1313 */
1314 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1315 ext3_update_dx_flag(dir);
1316 dir->i_version++;
1317 ext3_mark_inode_dirty(handle, dir);
1318 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1319 err = ext3_journal_dirty_metadata(handle, bh);
1320 if (err)
1321 ext3_std_error(dir->i_sb, err);
1322 brelse(bh);
1323 return 0;
1324 }
1325
1326 /*
1327 * This converts a one block unindexed directory to a 3 block indexed
1328 * directory, and adds the dentry to the indexed directory.
1329 */
1330 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1331 struct inode *inode, struct buffer_head *bh)
1332 {
1333 struct inode *dir = dentry->d_parent->d_inode;
1334 const char *name = dentry->d_name.name;
1335 int namelen = dentry->d_name.len;
1336 struct buffer_head *bh2;
1337 struct dx_root *root;
1338 struct dx_frame frames[2], *frame;
1339 struct dx_entry *entries;
1340 struct ext3_dir_entry_2 *de, *de2;
1341 char *data1, *top;
1342 unsigned len;
1343 int retval;
1344 unsigned blocksize;
1345 struct dx_hash_info hinfo;
1346 u32 block;
1347 struct fake_dirent *fde;
1348
1349 blocksize = dir->i_sb->s_blocksize;
1350 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1351 retval = ext3_journal_get_write_access(handle, bh);
1352 if (retval) {
1353 ext3_std_error(dir->i_sb, retval);
1354 brelse(bh);
1355 return retval;
1356 }
1357 root = (struct dx_root *) bh->b_data;
1358
1359 /* The 0th block becomes the root, move the dirents out */
1360 fde = &root->dotdot;
1361 de = (struct ext3_dir_entry_2 *)((char *)fde +
1362 ext3_rec_len_from_disk(fde->rec_len));
1363 if ((char *) de >= (((char *) root) + blocksize)) {
1364 ext3_error(dir->i_sb, __func__,
1365 "invalid rec_len for '..' in inode %lu",
1366 dir->i_ino);
1367 brelse(bh);
1368 return -EIO;
1369 }
1370 len = ((char *) root) + blocksize - (char *) de;
1371
1372 bh2 = ext3_append (handle, dir, &block, &retval);
1373 if (!(bh2)) {
1374 brelse(bh);
1375 return retval;
1376 }
1377 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1378 data1 = bh2->b_data;
1379
1380 memcpy (data1, de, len);
1381 de = (struct ext3_dir_entry_2 *) data1;
1382 top = data1 + len;
1383 while ((char *)(de2 = ext3_next_entry(de)) < top)
1384 de = de2;
1385 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1386 /* Initialize the root; the dot dirents already exist */
1387 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1388 de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1389 memset (&root->info, 0, sizeof(root->info));
1390 root->info.info_length = sizeof(root->info);
1391 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1392 entries = root->entries;
1393 dx_set_block (entries, 1);
1394 dx_set_count (entries, 1);
1395 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1396
1397 /* Initialize as for dx_probe */
1398 hinfo.hash_version = root->info.hash_version;
1399 if (hinfo.hash_version <= DX_HASH_TEA)
1400 hinfo.hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
1401 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1402 ext3fs_dirhash(name, namelen, &hinfo);
1403 frame = frames;
1404 frame->entries = entries;
1405 frame->at = entries;
1406 frame->bh = bh;
1407 bh = bh2;
1408 /*
1409 * Mark buffers dirty here so that if do_split() fails we write a
1410 * consistent set of buffers to disk.
1411 */
1412 ext3_journal_dirty_metadata(handle, frame->bh);
1413 ext3_journal_dirty_metadata(handle, bh);
1414 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1415 if (!de) {
1416 ext3_mark_inode_dirty(handle, dir);
1417 dx_release(frames);
1418 return retval;
1419 }
1420 dx_release(frames);
1421
1422 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1423 }
1424
1425 /*
1426 * ext3_add_entry()
1427 *
1428 * adds a file entry to the specified directory, using the same
1429 * semantics as ext3_find_entry(). It returns NULL if it failed.
1430 *
1431 * NOTE!! The inode part of 'de' is left at 0 - which means you
1432 * may not sleep between calling this and putting something into
1433 * the entry, as someone else might have used it while you slept.
1434 */
1435 static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1436 struct inode *inode)
1437 {
1438 struct inode *dir = dentry->d_parent->d_inode;
1439 struct buffer_head * bh;
1440 struct ext3_dir_entry_2 *de;
1441 struct super_block * sb;
1442 int retval;
1443 int dx_fallback=0;
1444 unsigned blocksize;
1445 u32 block, blocks;
1446
1447 sb = dir->i_sb;
1448 blocksize = sb->s_blocksize;
1449 if (!dentry->d_name.len)
1450 return -EINVAL;
1451 if (is_dx(dir)) {
1452 retval = ext3_dx_add_entry(handle, dentry, inode);
1453 if (!retval || (retval != ERR_BAD_DX_DIR))
1454 return retval;
1455 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1456 dx_fallback++;
1457 ext3_mark_inode_dirty(handle, dir);
1458 }
1459 blocks = dir->i_size >> sb->s_blocksize_bits;
1460 for (block = 0; block < blocks; block++) {
1461 bh = ext3_bread(handle, dir, block, 0, &retval);
1462 if(!bh)
1463 return retval;
1464 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1465 if (retval != -ENOSPC)
1466 return retval;
1467
1468 if (blocks == 1 && !dx_fallback &&
1469 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1470 return make_indexed_dir(handle, dentry, inode, bh);
1471 brelse(bh);
1472 }
1473 bh = ext3_append(handle, dir, &block, &retval);
1474 if (!bh)
1475 return retval;
1476 de = (struct ext3_dir_entry_2 *) bh->b_data;
1477 de->inode = 0;
1478 de->rec_len = ext3_rec_len_to_disk(blocksize);
1479 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1480 }
1481
1482 /*
1483 * Returns 0 for success, or a negative error value
1484 */
1485 static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1486 struct inode *inode)
1487 {
1488 struct dx_frame frames[2], *frame;
1489 struct dx_entry *entries, *at;
1490 struct dx_hash_info hinfo;
1491 struct buffer_head * bh;
1492 struct inode *dir = dentry->d_parent->d_inode;
1493 struct super_block * sb = dir->i_sb;
1494 struct ext3_dir_entry_2 *de;
1495 int err;
1496
1497 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1498 if (!frame)
1499 return err;
1500 entries = frame->entries;
1501 at = frame->at;
1502
1503 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1504 goto cleanup;
1505
1506 BUFFER_TRACE(bh, "get_write_access");
1507 err = ext3_journal_get_write_access(handle, bh);
1508 if (err)
1509 goto journal_error;
1510
1511 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1512 if (err != -ENOSPC) {
1513 bh = NULL;
1514 goto cleanup;
1515 }
1516
1517 /* Block full, should compress but for now just split */
1518 dxtrace(printk("using %u of %u node entries\n",
1519 dx_get_count(entries), dx_get_limit(entries)));
1520 /* Need to split index? */
1521 if (dx_get_count(entries) == dx_get_limit(entries)) {
1522 u32 newblock;
1523 unsigned icount = dx_get_count(entries);
1524 int levels = frame - frames;
1525 struct dx_entry *entries2;
1526 struct dx_node *node2;
1527 struct buffer_head *bh2;
1528
1529 if (levels && (dx_get_count(frames->entries) ==
1530 dx_get_limit(frames->entries))) {
1531 ext3_warning(sb, __func__,
1532 "Directory index full!");
1533 err = -ENOSPC;
1534 goto cleanup;
1535 }
1536 bh2 = ext3_append (handle, dir, &newblock, &err);
1537 if (!(bh2))
1538 goto cleanup;
1539 node2 = (struct dx_node *)(bh2->b_data);
1540 entries2 = node2->entries;
1541 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1542 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1543 BUFFER_TRACE(frame->bh, "get_write_access");
1544 err = ext3_journal_get_write_access(handle, frame->bh);
1545 if (err)
1546 goto journal_error;
1547 if (levels) {
1548 unsigned icount1 = icount/2, icount2 = icount - icount1;
1549 unsigned hash2 = dx_get_hash(entries + icount1);
1550 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1551
1552 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1553 err = ext3_journal_get_write_access(handle,
1554 frames[0].bh);
1555 if (err)
1556 goto journal_error;
1557
1558 memcpy ((char *) entries2, (char *) (entries + icount1),
1559 icount2 * sizeof(struct dx_entry));
1560 dx_set_count (entries, icount1);
1561 dx_set_count (entries2, icount2);
1562 dx_set_limit (entries2, dx_node_limit(dir));
1563
1564 /* Which index block gets the new entry? */
1565 if (at - entries >= icount1) {
1566 frame->at = at = at - entries - icount1 + entries2;
1567 frame->entries = entries = entries2;
1568 swap(frame->bh, bh2);
1569 }
1570 dx_insert_block (frames + 0, hash2, newblock);
1571 dxtrace(dx_show_index ("node", frames[1].entries));
1572 dxtrace(dx_show_index ("node",
1573 ((struct dx_node *) bh2->b_data)->entries));
1574 err = ext3_journal_dirty_metadata(handle, bh2);
1575 if (err)
1576 goto journal_error;
1577 brelse (bh2);
1578 } else {
1579 dxtrace(printk("Creating second level index...\n"));
1580 memcpy((char *) entries2, (char *) entries,
1581 icount * sizeof(struct dx_entry));
1582 dx_set_limit(entries2, dx_node_limit(dir));
1583
1584 /* Set up root */
1585 dx_set_count(entries, 1);
1586 dx_set_block(entries + 0, newblock);
1587 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1588
1589 /* Add new access path frame */
1590 frame = frames + 1;
1591 frame->at = at = at - entries + entries2;
1592 frame->entries = entries = entries2;
1593 frame->bh = bh2;
1594 err = ext3_journal_get_write_access(handle,
1595 frame->bh);
1596 if (err)
1597 goto journal_error;
1598 }
1599 err = ext3_journal_dirty_metadata(handle, frames[0].bh);
1600 if (err)
1601 goto journal_error;
1602 }
1603 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1604 if (!de)
1605 goto cleanup;
1606 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1607 bh = NULL;
1608 goto cleanup;
1609
1610 journal_error:
1611 ext3_std_error(dir->i_sb, err);
1612 cleanup:
1613 if (bh)
1614 brelse(bh);
1615 dx_release(frames);
1616 return err;
1617 }
1618
1619 /*
1620 * ext3_delete_entry deletes a directory entry by merging it with the
1621 * previous entry
1622 */
1623 static int ext3_delete_entry (handle_t *handle,
1624 struct inode * dir,
1625 struct ext3_dir_entry_2 * de_del,
1626 struct buffer_head * bh)
1627 {
1628 struct ext3_dir_entry_2 * de, * pde;
1629 int i;
1630
1631 i = 0;
1632 pde = NULL;
1633 de = (struct ext3_dir_entry_2 *) bh->b_data;
1634 while (i < bh->b_size) {
1635 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1636 return -EIO;
1637 if (de == de_del) {
1638 int err;
1639
1640 BUFFER_TRACE(bh, "get_write_access");
1641 err = ext3_journal_get_write_access(handle, bh);
1642 if (err)
1643 goto journal_error;
1644
1645 if (pde)
1646 pde->rec_len = ext3_rec_len_to_disk(
1647 ext3_rec_len_from_disk(pde->rec_len) +
1648 ext3_rec_len_from_disk(de->rec_len));
1649 else
1650 de->inode = 0;
1651 dir->i_version++;
1652 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1653 err = ext3_journal_dirty_metadata(handle, bh);
1654 if (err) {
1655 journal_error:
1656 ext3_std_error(dir->i_sb, err);
1657 return err;
1658 }
1659 return 0;
1660 }
1661 i += ext3_rec_len_from_disk(de->rec_len);
1662 pde = de;
1663 de = ext3_next_entry(de);
1664 }
1665 return -ENOENT;
1666 }
1667
1668 static int ext3_add_nondir(handle_t *handle,
1669 struct dentry *dentry, struct inode *inode)
1670 {
1671 int err = ext3_add_entry(handle, dentry, inode);
1672 if (!err) {
1673 ext3_mark_inode_dirty(handle, inode);
1674 unlock_new_inode(inode);
1675 d_instantiate(dentry, inode);
1676 return 0;
1677 }
1678 drop_nlink(inode);
1679 unlock_new_inode(inode);
1680 iput(inode);
1681 return err;
1682 }
1683
1684 /*
1685 * By the time this is called, we already have created
1686 * the directory cache entry for the new file, but it
1687 * is so far negative - it has no inode.
1688 *
1689 * If the create succeeds, we fill in the inode information
1690 * with d_instantiate().
1691 */
1692 static int ext3_create (struct inode * dir, struct dentry * dentry, umode_t mode,
1693 bool excl)
1694 {
1695 handle_t *handle;
1696 struct inode * inode;
1697 int err, retries = 0;
1698
1699 dquot_initialize(dir);
1700
1701 retry:
1702 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1703 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1704 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1705 if (IS_ERR(handle))
1706 return PTR_ERR(handle);
1707
1708 if (IS_DIRSYNC(dir))
1709 handle->h_sync = 1;
1710
1711 inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1712 err = PTR_ERR(inode);
1713 if (!IS_ERR(inode)) {
1714 inode->i_op = &ext3_file_inode_operations;
1715 inode->i_fop = &ext3_file_operations;
1716 ext3_set_aops(inode);
1717 err = ext3_add_nondir(handle, dentry, inode);
1718 }
1719 ext3_journal_stop(handle);
1720 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1721 goto retry;
1722 return err;
1723 }
1724
1725 static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1726 umode_t mode, dev_t rdev)
1727 {
1728 handle_t *handle;
1729 struct inode *inode;
1730 int err, retries = 0;
1731
1732 if (!new_valid_dev(rdev))
1733 return -EINVAL;
1734
1735 dquot_initialize(dir);
1736
1737 retry:
1738 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1739 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1740 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1741 if (IS_ERR(handle))
1742 return PTR_ERR(handle);
1743
1744 if (IS_DIRSYNC(dir))
1745 handle->h_sync = 1;
1746
1747 inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1748 err = PTR_ERR(inode);
1749 if (!IS_ERR(inode)) {
1750 init_special_inode(inode, inode->i_mode, rdev);
1751 #ifdef CONFIG_EXT3_FS_XATTR
1752 inode->i_op = &ext3_special_inode_operations;
1753 #endif
1754 err = ext3_add_nondir(handle, dentry, inode);
1755 }
1756 ext3_journal_stop(handle);
1757 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1758 goto retry;
1759 return err;
1760 }
1761
1762 static int ext3_mkdir(struct inode * dir, struct dentry * dentry, umode_t mode)
1763 {
1764 handle_t *handle;
1765 struct inode * inode;
1766 struct buffer_head * dir_block = NULL;
1767 struct ext3_dir_entry_2 * de;
1768 int err, retries = 0;
1769
1770 if (dir->i_nlink >= EXT3_LINK_MAX)
1771 return -EMLINK;
1772
1773 dquot_initialize(dir);
1774
1775 retry:
1776 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1777 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1778 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1779 if (IS_ERR(handle))
1780 return PTR_ERR(handle);
1781
1782 if (IS_DIRSYNC(dir))
1783 handle->h_sync = 1;
1784
1785 inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFDIR | mode);
1786 err = PTR_ERR(inode);
1787 if (IS_ERR(inode))
1788 goto out_stop;
1789
1790 inode->i_op = &ext3_dir_inode_operations;
1791 inode->i_fop = &ext3_dir_operations;
1792 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1793 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1794 if (!dir_block)
1795 goto out_clear_inode;
1796
1797 BUFFER_TRACE(dir_block, "get_write_access");
1798 err = ext3_journal_get_write_access(handle, dir_block);
1799 if (err)
1800 goto out_clear_inode;
1801
1802 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1803 de->inode = cpu_to_le32(inode->i_ino);
1804 de->name_len = 1;
1805 de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1806 strcpy (de->name, ".");
1807 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1808 de = ext3_next_entry(de);
1809 de->inode = cpu_to_le32(dir->i_ino);
1810 de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1811 EXT3_DIR_REC_LEN(1));
1812 de->name_len = 2;
1813 strcpy (de->name, "..");
1814 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1815 set_nlink(inode, 2);
1816 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1817 err = ext3_journal_dirty_metadata(handle, dir_block);
1818 if (err)
1819 goto out_clear_inode;
1820
1821 err = ext3_mark_inode_dirty(handle, inode);
1822 if (!err)
1823 err = ext3_add_entry (handle, dentry, inode);
1824
1825 if (err) {
1826 out_clear_inode:
1827 clear_nlink(inode);
1828 unlock_new_inode(inode);
1829 ext3_mark_inode_dirty(handle, inode);
1830 iput (inode);
1831 goto out_stop;
1832 }
1833 inc_nlink(dir);
1834 ext3_update_dx_flag(dir);
1835 err = ext3_mark_inode_dirty(handle, dir);
1836 if (err)
1837 goto out_clear_inode;
1838
1839 unlock_new_inode(inode);
1840 d_instantiate(dentry, inode);
1841 out_stop:
1842 brelse(dir_block);
1843 ext3_journal_stop(handle);
1844 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1845 goto retry;
1846 return err;
1847 }
1848
1849 /*
1850 * routine to check that the specified directory is empty (for rmdir)
1851 */
1852 static int empty_dir (struct inode * inode)
1853 {
1854 unsigned long offset;
1855 struct buffer_head * bh;
1856 struct ext3_dir_entry_2 * de, * de1;
1857 struct super_block * sb;
1858 int err = 0;
1859
1860 sb = inode->i_sb;
1861 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1862 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1863 if (err)
1864 ext3_error(inode->i_sb, __func__,
1865 "error %d reading directory #%lu offset 0",
1866 err, inode->i_ino);
1867 else
1868 ext3_warning(inode->i_sb, __func__,
1869 "bad directory (dir #%lu) - no data block",
1870 inode->i_ino);
1871 return 1;
1872 }
1873 de = (struct ext3_dir_entry_2 *) bh->b_data;
1874 de1 = ext3_next_entry(de);
1875 if (le32_to_cpu(de->inode) != inode->i_ino ||
1876 !le32_to_cpu(de1->inode) ||
1877 strcmp (".", de->name) ||
1878 strcmp ("..", de1->name)) {
1879 ext3_warning (inode->i_sb, "empty_dir",
1880 "bad directory (dir #%lu) - no `.' or `..'",
1881 inode->i_ino);
1882 brelse (bh);
1883 return 1;
1884 }
1885 offset = ext3_rec_len_from_disk(de->rec_len) +
1886 ext3_rec_len_from_disk(de1->rec_len);
1887 de = ext3_next_entry(de1);
1888 while (offset < inode->i_size ) {
1889 if (!bh ||
1890 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1891 err = 0;
1892 brelse (bh);
1893 bh = ext3_bread (NULL, inode,
1894 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1895 if (!bh) {
1896 if (err)
1897 ext3_error(sb, __func__,
1898 "error %d reading directory"
1899 " #%lu offset %lu",
1900 err, inode->i_ino, offset);
1901 offset += sb->s_blocksize;
1902 continue;
1903 }
1904 de = (struct ext3_dir_entry_2 *) bh->b_data;
1905 }
1906 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1907 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1908 sb->s_blocksize);
1909 offset = (offset | (sb->s_blocksize - 1)) + 1;
1910 continue;
1911 }
1912 if (le32_to_cpu(de->inode)) {
1913 brelse (bh);
1914 return 0;
1915 }
1916 offset += ext3_rec_len_from_disk(de->rec_len);
1917 de = ext3_next_entry(de);
1918 }
1919 brelse (bh);
1920 return 1;
1921 }
1922
1923 /* ext3_orphan_add() links an unlinked or truncated inode into a list of
1924 * such inodes, starting at the superblock, in case we crash before the
1925 * file is closed/deleted, or in case the inode truncate spans multiple
1926 * transactions and the last transaction is not recovered after a crash.
1927 *
1928 * At filesystem recovery time, we walk this list deleting unlinked
1929 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1930 */
1931 int ext3_orphan_add(handle_t *handle, struct inode *inode)
1932 {
1933 struct super_block *sb = inode->i_sb;
1934 struct ext3_iloc iloc;
1935 int err = 0, rc;
1936
1937 mutex_lock(&EXT3_SB(sb)->s_orphan_lock);
1938 if (!list_empty(&EXT3_I(inode)->i_orphan))
1939 goto out_unlock;
1940
1941 /* Orphan handling is only valid for files with data blocks
1942 * being truncated, or files being unlinked. */
1943
1944 /* @@@ FIXME: Observation from aviro:
1945 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1946 * here (on s_orphan_lock), so race with ext3_link() which might bump
1947 * ->i_nlink. For, say it, character device. Not a regular file,
1948 * not a directory, not a symlink and ->i_nlink > 0.
1949 *
1950 * tytso, 4/25/2009: I'm not sure how that could happen;
1951 * shouldn't the fs core protect us from these sort of
1952 * unlink()/link() races?
1953 */
1954 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1955 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1956
1957 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1958 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1959 if (err)
1960 goto out_unlock;
1961
1962 err = ext3_reserve_inode_write(handle, inode, &iloc);
1963 if (err)
1964 goto out_unlock;
1965
1966 /* Insert this inode at the head of the on-disk orphan list... */
1967 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1968 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1969 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1970 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1971 if (!err)
1972 err = rc;
1973
1974 /* Only add to the head of the in-memory list if all the
1975 * previous operations succeeded. If the orphan_add is going to
1976 * fail (possibly taking the journal offline), we can't risk
1977 * leaving the inode on the orphan list: stray orphan-list
1978 * entries can cause panics at unmount time.
1979 *
1980 * This is safe: on error we're going to ignore the orphan list
1981 * anyway on the next recovery. */
1982 if (!err)
1983 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1984
1985 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1986 jbd_debug(4, "orphan inode %lu will point to %d\n",
1987 inode->i_ino, NEXT_ORPHAN(inode));
1988 out_unlock:
1989 mutex_unlock(&EXT3_SB(sb)->s_orphan_lock);
1990 ext3_std_error(inode->i_sb, err);
1991 return err;
1992 }
1993
1994 /*
1995 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1996 * of such inodes stored on disk, because it is finally being cleaned up.
1997 */
1998 int ext3_orphan_del(handle_t *handle, struct inode *inode)
1999 {
2000 struct list_head *prev;
2001 struct ext3_inode_info *ei = EXT3_I(inode);
2002 struct ext3_sb_info *sbi;
2003 unsigned long ino_next;
2004 struct ext3_iloc iloc;
2005 int err = 0;
2006
2007 mutex_lock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2008 if (list_empty(&ei->i_orphan))
2009 goto out;
2010
2011 ino_next = NEXT_ORPHAN(inode);
2012 prev = ei->i_orphan.prev;
2013 sbi = EXT3_SB(inode->i_sb);
2014
2015 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2016
2017 list_del_init(&ei->i_orphan);
2018
2019 /* If we're on an error path, we may not have a valid
2020 * transaction handle with which to update the orphan list on
2021 * disk, but we still need to remove the inode from the linked
2022 * list in memory. */
2023 if (!handle)
2024 goto out;
2025
2026 err = ext3_reserve_inode_write(handle, inode, &iloc);
2027 if (err)
2028 goto out_err;
2029
2030 if (prev == &sbi->s_orphan) {
2031 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2032 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2033 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2034 if (err)
2035 goto out_brelse;
2036 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2037 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2038 } else {
2039 struct ext3_iloc iloc2;
2040 struct inode *i_prev =
2041 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2042
2043 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2044 i_prev->i_ino, ino_next);
2045 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2046 if (err)
2047 goto out_brelse;
2048 NEXT_ORPHAN(i_prev) = ino_next;
2049 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2050 }
2051 if (err)
2052 goto out_brelse;
2053 NEXT_ORPHAN(inode) = 0;
2054 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2055
2056 out_err:
2057 ext3_std_error(inode->i_sb, err);
2058 out:
2059 mutex_unlock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2060 return err;
2061
2062 out_brelse:
2063 brelse(iloc.bh);
2064 goto out_err;
2065 }
2066
2067 static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2068 {
2069 int retval;
2070 struct inode * inode;
2071 struct buffer_head * bh;
2072 struct ext3_dir_entry_2 * de;
2073 handle_t *handle;
2074
2075 /* Initialize quotas before so that eventual writes go in
2076 * separate transaction */
2077 dquot_initialize(dir);
2078 dquot_initialize(dentry->d_inode);
2079
2080 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2081 if (IS_ERR(handle))
2082 return PTR_ERR(handle);
2083
2084 retval = -ENOENT;
2085 bh = ext3_find_entry(dir, &dentry->d_name, &de);
2086 if (!bh)
2087 goto end_rmdir;
2088
2089 if (IS_DIRSYNC(dir))
2090 handle->h_sync = 1;
2091
2092 inode = dentry->d_inode;
2093
2094 retval = -EIO;
2095 if (le32_to_cpu(de->inode) != inode->i_ino)
2096 goto end_rmdir;
2097
2098 retval = -ENOTEMPTY;
2099 if (!empty_dir (inode))
2100 goto end_rmdir;
2101
2102 retval = ext3_delete_entry(handle, dir, de, bh);
2103 if (retval)
2104 goto end_rmdir;
2105 if (inode->i_nlink != 2)
2106 ext3_warning (inode->i_sb, "ext3_rmdir",
2107 "empty directory has nlink!=2 (%d)",
2108 inode->i_nlink);
2109 inode->i_version++;
2110 clear_nlink(inode);
2111 /* There's no need to set i_disksize: the fact that i_nlink is
2112 * zero will ensure that the right thing happens during any
2113 * recovery. */
2114 inode->i_size = 0;
2115 ext3_orphan_add(handle, inode);
2116 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2117 ext3_mark_inode_dirty(handle, inode);
2118 drop_nlink(dir);
2119 ext3_update_dx_flag(dir);
2120 ext3_mark_inode_dirty(handle, dir);
2121
2122 end_rmdir:
2123 ext3_journal_stop(handle);
2124 brelse (bh);
2125 return retval;
2126 }
2127
2128 static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2129 {
2130 int retval;
2131 struct inode * inode;
2132 struct buffer_head * bh;
2133 struct ext3_dir_entry_2 * de;
2134 handle_t *handle;
2135
2136 trace_ext3_unlink_enter(dir, dentry);
2137 /* Initialize quotas before so that eventual writes go
2138 * in separate transaction */
2139 dquot_initialize(dir);
2140 dquot_initialize(dentry->d_inode);
2141
2142 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2143 if (IS_ERR(handle))
2144 return PTR_ERR(handle);
2145
2146 if (IS_DIRSYNC(dir))
2147 handle->h_sync = 1;
2148
2149 retval = -ENOENT;
2150 bh = ext3_find_entry(dir, &dentry->d_name, &de);
2151 if (!bh)
2152 goto end_unlink;
2153
2154 inode = dentry->d_inode;
2155
2156 retval = -EIO;
2157 if (le32_to_cpu(de->inode) != inode->i_ino)
2158 goto end_unlink;
2159
2160 if (!inode->i_nlink) {
2161 ext3_warning (inode->i_sb, "ext3_unlink",
2162 "Deleting nonexistent file (%lu), %d",
2163 inode->i_ino, inode->i_nlink);
2164 set_nlink(inode, 1);
2165 }
2166 retval = ext3_delete_entry(handle, dir, de, bh);
2167 if (retval)
2168 goto end_unlink;
2169 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2170 ext3_update_dx_flag(dir);
2171 ext3_mark_inode_dirty(handle, dir);
2172 drop_nlink(inode);
2173 if (!inode->i_nlink)
2174 ext3_orphan_add(handle, inode);
2175 inode->i_ctime = dir->i_ctime;
2176 ext3_mark_inode_dirty(handle, inode);
2177 retval = 0;
2178
2179 end_unlink:
2180 ext3_journal_stop(handle);
2181 brelse (bh);
2182 trace_ext3_unlink_exit(dentry, retval);
2183 return retval;
2184 }
2185
2186 static int ext3_symlink (struct inode * dir,
2187 struct dentry *dentry, const char * symname)
2188 {
2189 handle_t *handle;
2190 struct inode * inode;
2191 int l, err, retries = 0;
2192 int credits;
2193
2194 l = strlen(symname)+1;
2195 if (l > dir->i_sb->s_blocksize)
2196 return -ENAMETOOLONG;
2197
2198 dquot_initialize(dir);
2199
2200 if (l > EXT3_N_BLOCKS * 4) {
2201 /*
2202 * For non-fast symlinks, we just allocate inode and put it on
2203 * orphan list in the first transaction => we need bitmap,
2204 * group descriptor, sb, inode block, quota blocks, and
2205 * possibly selinux xattr blocks.
2206 */
2207 credits = 4 + EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2208 EXT3_XATTR_TRANS_BLOCKS;
2209 } else {
2210 /*
2211 * Fast symlink. We have to add entry to directory
2212 * (EXT3_DATA_TRANS_BLOCKS + EXT3_INDEX_EXTRA_TRANS_BLOCKS),
2213 * allocate new inode (bitmap, group descriptor, inode block,
2214 * quota blocks, sb is already counted in previous macros).
2215 */
2216 credits = EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2217 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2218 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2219 }
2220 retry:
2221 handle = ext3_journal_start(dir, credits);
2222 if (IS_ERR(handle))
2223 return PTR_ERR(handle);
2224
2225 if (IS_DIRSYNC(dir))
2226 handle->h_sync = 1;
2227
2228 inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFLNK|S_IRWXUGO);
2229 err = PTR_ERR(inode);
2230 if (IS_ERR(inode))
2231 goto out_stop;
2232
2233 if (l > EXT3_N_BLOCKS * 4) {
2234 inode->i_op = &ext3_symlink_inode_operations;
2235 ext3_set_aops(inode);
2236 /*
2237 * We cannot call page_symlink() with transaction started
2238 * because it calls into ext3_write_begin() which acquires page
2239 * lock which ranks below transaction start (and it can also
2240 * wait for journal commit if we are running out of space). So
2241 * we have to stop transaction now and restart it when symlink
2242 * contents is written.
2243 *
2244 * To keep fs consistent in case of crash, we have to put inode
2245 * to orphan list in the mean time.
2246 */
2247 drop_nlink(inode);
2248 err = ext3_orphan_add(handle, inode);
2249 ext3_journal_stop(handle);
2250 if (err)
2251 goto err_drop_inode;
2252 err = __page_symlink(inode, symname, l, 1);
2253 if (err)
2254 goto err_drop_inode;
2255 /*
2256 * Now inode is being linked into dir (EXT3_DATA_TRANS_BLOCKS
2257 * + EXT3_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2258 */
2259 handle = ext3_journal_start(dir,
2260 EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2261 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 1);
2262 if (IS_ERR(handle)) {
2263 err = PTR_ERR(handle);
2264 goto err_drop_inode;
2265 }
2266 set_nlink(inode, 1);
2267 err = ext3_orphan_del(handle, inode);
2268 if (err) {
2269 ext3_journal_stop(handle);
2270 drop_nlink(inode);
2271 goto err_drop_inode;
2272 }
2273 } else {
2274 inode->i_op = &ext3_fast_symlink_inode_operations;
2275 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2276 inode->i_size = l-1;
2277 }
2278 EXT3_I(inode)->i_disksize = inode->i_size;
2279 err = ext3_add_nondir(handle, dentry, inode);
2280 out_stop:
2281 ext3_journal_stop(handle);
2282 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2283 goto retry;
2284 return err;
2285 err_drop_inode:
2286 unlock_new_inode(inode);
2287 iput(inode);
2288 return err;
2289 }
2290
2291 static int ext3_link (struct dentry * old_dentry,
2292 struct inode * dir, struct dentry *dentry)
2293 {
2294 handle_t *handle;
2295 struct inode *inode = old_dentry->d_inode;
2296 int err, retries = 0;
2297
2298 if (inode->i_nlink >= EXT3_LINK_MAX)
2299 return -EMLINK;
2300
2301 dquot_initialize(dir);
2302
2303 retry:
2304 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2305 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2306 if (IS_ERR(handle))
2307 return PTR_ERR(handle);
2308
2309 if (IS_DIRSYNC(dir))
2310 handle->h_sync = 1;
2311
2312 inode->i_ctime = CURRENT_TIME_SEC;
2313 inc_nlink(inode);
2314 ihold(inode);
2315
2316 err = ext3_add_entry(handle, dentry, inode);
2317 if (!err) {
2318 ext3_mark_inode_dirty(handle, inode);
2319 d_instantiate(dentry, inode);
2320 } else {
2321 drop_nlink(inode);
2322 iput(inode);
2323 }
2324 ext3_journal_stop(handle);
2325 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2326 goto retry;
2327 return err;
2328 }
2329
2330 #define PARENT_INO(buffer) \
2331 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2332
2333 /*
2334 * Anybody can rename anything with this: the permission checks are left to the
2335 * higher-level routines.
2336 */
2337 static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2338 struct inode * new_dir,struct dentry *new_dentry)
2339 {
2340 handle_t *handle;
2341 struct inode * old_inode, * new_inode;
2342 struct buffer_head * old_bh, * new_bh, * dir_bh;
2343 struct ext3_dir_entry_2 * old_de, * new_de;
2344 int retval, flush_file = 0;
2345
2346 dquot_initialize(old_dir);
2347 dquot_initialize(new_dir);
2348
2349 old_bh = new_bh = dir_bh = NULL;
2350
2351 /* Initialize quotas before so that eventual writes go
2352 * in separate transaction */
2353 if (new_dentry->d_inode)
2354 dquot_initialize(new_dentry->d_inode);
2355 handle = ext3_journal_start(old_dir, 2 *
2356 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2357 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2358 if (IS_ERR(handle))
2359 return PTR_ERR(handle);
2360
2361 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2362 handle->h_sync = 1;
2363
2364 old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
2365 /*
2366 * Check for inode number is _not_ due to possible IO errors.
2367 * We might rmdir the source, keep it as pwd of some process
2368 * and merrily kill the link to whatever was created under the
2369 * same name. Goodbye sticky bit ;-<
2370 */
2371 old_inode = old_dentry->d_inode;
2372 retval = -ENOENT;
2373 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2374 goto end_rename;
2375
2376 new_inode = new_dentry->d_inode;
2377 new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
2378 if (new_bh) {
2379 if (!new_inode) {
2380 brelse (new_bh);
2381 new_bh = NULL;
2382 }
2383 }
2384 if (S_ISDIR(old_inode->i_mode)) {
2385 if (new_inode) {
2386 retval = -ENOTEMPTY;
2387 if (!empty_dir (new_inode))
2388 goto end_rename;
2389 }
2390 retval = -EIO;
2391 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2392 if (!dir_bh)
2393 goto end_rename;
2394 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2395 goto end_rename;
2396 retval = -EMLINK;
2397 if (!new_inode && new_dir!=old_dir &&
2398 new_dir->i_nlink >= EXT3_LINK_MAX)
2399 goto end_rename;
2400 }
2401 if (!new_bh) {
2402 retval = ext3_add_entry (handle, new_dentry, old_inode);
2403 if (retval)
2404 goto end_rename;
2405 } else {
2406 BUFFER_TRACE(new_bh, "get write access");
2407 retval = ext3_journal_get_write_access(handle, new_bh);
2408 if (retval)
2409 goto journal_error;
2410 new_de->inode = cpu_to_le32(old_inode->i_ino);
2411 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2412 EXT3_FEATURE_INCOMPAT_FILETYPE))
2413 new_de->file_type = old_de->file_type;
2414 new_dir->i_version++;
2415 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2416 ext3_mark_inode_dirty(handle, new_dir);
2417 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2418 retval = ext3_journal_dirty_metadata(handle, new_bh);
2419 if (retval)
2420 goto journal_error;
2421 brelse(new_bh);
2422 new_bh = NULL;
2423 }
2424
2425 /*
2426 * Like most other Unix systems, set the ctime for inodes on a
2427 * rename.
2428 */
2429 old_inode->i_ctime = CURRENT_TIME_SEC;
2430 ext3_mark_inode_dirty(handle, old_inode);
2431
2432 /*
2433 * ok, that's it
2434 */
2435 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2436 old_de->name_len != old_dentry->d_name.len ||
2437 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2438 (retval = ext3_delete_entry(handle, old_dir,
2439 old_de, old_bh)) == -ENOENT) {
2440 /* old_de could have moved from under us during htree split, so
2441 * make sure that we are deleting the right entry. We might
2442 * also be pointing to a stale entry in the unused part of
2443 * old_bh so just checking inum and the name isn't enough. */
2444 struct buffer_head *old_bh2;
2445 struct ext3_dir_entry_2 *old_de2;
2446
2447 old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2448 &old_de2);
2449 if (old_bh2) {
2450 retval = ext3_delete_entry(handle, old_dir,
2451 old_de2, old_bh2);
2452 brelse(old_bh2);
2453 }
2454 }
2455 if (retval) {
2456 ext3_warning(old_dir->i_sb, "ext3_rename",
2457 "Deleting old file (%lu), %d, error=%d",
2458 old_dir->i_ino, old_dir->i_nlink, retval);
2459 }
2460
2461 if (new_inode) {
2462 drop_nlink(new_inode);
2463 new_inode->i_ctime = CURRENT_TIME_SEC;
2464 }
2465 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2466 ext3_update_dx_flag(old_dir);
2467 if (dir_bh) {
2468 BUFFER_TRACE(dir_bh, "get_write_access");
2469 retval = ext3_journal_get_write_access(handle, dir_bh);
2470 if (retval)
2471 goto journal_error;
2472 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2473 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2474 retval = ext3_journal_dirty_metadata(handle, dir_bh);
2475 if (retval) {
2476 journal_error:
2477 ext3_std_error(new_dir->i_sb, retval);
2478 goto end_rename;
2479 }
2480 drop_nlink(old_dir);
2481 if (new_inode) {
2482 drop_nlink(new_inode);
2483 } else {
2484 inc_nlink(new_dir);
2485 ext3_update_dx_flag(new_dir);
2486 ext3_mark_inode_dirty(handle, new_dir);
2487 }
2488 }
2489 ext3_mark_inode_dirty(handle, old_dir);
2490 if (new_inode) {
2491 ext3_mark_inode_dirty(handle, new_inode);
2492 if (!new_inode->i_nlink)
2493 ext3_orphan_add(handle, new_inode);
2494 if (ext3_should_writeback_data(new_inode))
2495 flush_file = 1;
2496 }
2497 retval = 0;
2498
2499 end_rename:
2500 brelse (dir_bh);
2501 brelse (old_bh);
2502 brelse (new_bh);
2503 ext3_journal_stop(handle);
2504 if (retval == 0 && flush_file)
2505 filemap_flush(old_inode->i_mapping);
2506 return retval;
2507 }
2508
2509 /*
2510 * directories can handle most operations...
2511 */
2512 const struct inode_operations ext3_dir_inode_operations = {
2513 .create = ext3_create,
2514 .lookup = ext3_lookup,
2515 .link = ext3_link,
2516 .unlink = ext3_unlink,
2517 .symlink = ext3_symlink,
2518 .mkdir = ext3_mkdir,
2519 .rmdir = ext3_rmdir,
2520 .mknod = ext3_mknod,
2521 .rename = ext3_rename,
2522 .setattr = ext3_setattr,
2523 #ifdef CONFIG_EXT3_FS_XATTR
2524 .setxattr = generic_setxattr,
2525 .getxattr = generic_getxattr,
2526 .listxattr = ext3_listxattr,
2527 .removexattr = generic_removexattr,
2528 #endif
2529 .get_acl = ext3_get_acl,
2530 };
2531
2532 const struct inode_operations ext3_special_inode_operations = {
2533 .setattr = ext3_setattr,
2534 #ifdef CONFIG_EXT3_FS_XATTR
2535 .setxattr = generic_setxattr,
2536 .getxattr = generic_getxattr,
2537 .listxattr = ext3_listxattr,
2538 .removexattr = generic_removexattr,
2539 #endif
2540 .get_acl = ext3_get_acl,
2541 };
2542
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.