Loading...
Note: File does not exist in v4.6.
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/fs.h>
28#include <linux/pagemap.h>
29#include <linux/jbd.h>
30#include <linux/time.h>
31#include <linux/ext3_fs.h>
32#include <linux/ext3_jbd.h>
33#include <linux/fcntl.h>
34#include <linux/stat.h>
35#include <linux/string.h>
36#include <linux/quotaops.h>
37#include <linux/buffer_head.h>
38#include <linux/bio.h>
39#include <trace/events/ext3.h>
40
41#include "namei.h"
42#include "xattr.h"
43#include "acl.h"
44
45/*
46 * define how far ahead to read directories while searching them.
47 */
48#define NAMEI_RA_CHUNKS 2
49#define NAMEI_RA_BLOCKS 4
50#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
52
53static struct buffer_head *ext3_append(handle_t *handle,
54 struct inode *inode,
55 u32 *block, int *err)
56{
57 struct buffer_head *bh;
58
59 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
60
61 bh = ext3_bread(handle, inode, *block, 1, err);
62 if (bh) {
63 inode->i_size += inode->i_sb->s_blocksize;
64 EXT3_I(inode)->i_disksize = inode->i_size;
65 *err = ext3_journal_get_write_access(handle, bh);
66 if (*err) {
67 brelse(bh);
68 bh = NULL;
69 }
70 }
71 return bh;
72}
73
74#ifndef assert
75#define assert(test) J_ASSERT(test)
76#endif
77
78#ifdef DX_DEBUG
79#define dxtrace(command) command
80#else
81#define dxtrace(command)
82#endif
83
84struct fake_dirent
85{
86 __le32 inode;
87 __le16 rec_len;
88 u8 name_len;
89 u8 file_type;
90};
91
92struct dx_countlimit
93{
94 __le16 limit;
95 __le16 count;
96};
97
98struct dx_entry
99{
100 __le32 hash;
101 __le32 block;
102};
103
104/*
105 * dx_root_info is laid out so that if it should somehow get overlaid by a
106 * dirent the two low bits of the hash version will be zero. Therefore, the
107 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
108 */
109
110struct dx_root
111{
112 struct fake_dirent dot;
113 char dot_name[4];
114 struct fake_dirent dotdot;
115 char dotdot_name[4];
116 struct dx_root_info
117 {
118 __le32 reserved_zero;
119 u8 hash_version;
120 u8 info_length; /* 8 */
121 u8 indirect_levels;
122 u8 unused_flags;
123 }
124 info;
125 struct dx_entry entries[0];
126};
127
128struct dx_node
129{
130 struct fake_dirent fake;
131 struct dx_entry entries[0];
132};
133
134
135struct dx_frame
136{
137 struct buffer_head *bh;
138 struct dx_entry *entries;
139 struct dx_entry *at;
140};
141
142struct dx_map_entry
143{
144 u32 hash;
145 u16 offs;
146 u16 size;
147};
148
149static inline unsigned dx_get_block (struct dx_entry *entry);
150static void dx_set_block (struct dx_entry *entry, unsigned value);
151static inline unsigned dx_get_hash (struct dx_entry *entry);
152static void dx_set_hash (struct dx_entry *entry, unsigned value);
153static unsigned dx_get_count (struct dx_entry *entries);
154static unsigned dx_get_limit (struct dx_entry *entries);
155static void dx_set_count (struct dx_entry *entries, unsigned value);
156static void dx_set_limit (struct dx_entry *entries, unsigned value);
157static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
158static unsigned dx_node_limit (struct inode *dir);
159static struct dx_frame *dx_probe(struct qstr *entry,
160 struct inode *dir,
161 struct dx_hash_info *hinfo,
162 struct dx_frame *frame,
163 int *err);
164static void dx_release (struct dx_frame *frames);
165static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
166 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
167static void dx_sort_map(struct dx_map_entry *map, unsigned count);
168static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
169 struct dx_map_entry *offsets, int count);
170static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize);
171static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
172static int ext3_htree_next_block(struct inode *dir, __u32 hash,
173 struct dx_frame *frame,
174 struct dx_frame *frames,
175 __u32 *start_hash);
176static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
177 struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
178 int *err);
179static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
180 struct inode *inode);
181
182/*
183 * p is at least 6 bytes before the end of page
184 */
185static inline struct ext3_dir_entry_2 *
186ext3_next_entry(struct ext3_dir_entry_2 *p)
187{
188 return (struct ext3_dir_entry_2 *)((char *)p +
189 ext3_rec_len_from_disk(p->rec_len));
190}
191
192/*
193 * Future: use high four bits of block for coalesce-on-delete flags
194 * Mask them off for now.
195 */
196
197static inline unsigned dx_get_block (struct dx_entry *entry)
198{
199 return le32_to_cpu(entry->block) & 0x00ffffff;
200}
201
202static inline void dx_set_block (struct dx_entry *entry, unsigned value)
203{
204 entry->block = cpu_to_le32(value);
205}
206
207static inline unsigned dx_get_hash (struct dx_entry *entry)
208{
209 return le32_to_cpu(entry->hash);
210}
211
212static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
213{
214 entry->hash = cpu_to_le32(value);
215}
216
217static inline unsigned dx_get_count (struct dx_entry *entries)
218{
219 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
220}
221
222static inline unsigned dx_get_limit (struct dx_entry *entries)
223{
224 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
225}
226
227static inline void dx_set_count (struct dx_entry *entries, unsigned value)
228{
229 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
230}
231
232static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
233{
234 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
235}
236
237static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
238{
239 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
240 EXT3_DIR_REC_LEN(2) - infosize;
241 return entry_space / sizeof(struct dx_entry);
242}
243
244static inline unsigned dx_node_limit (struct inode *dir)
245{
246 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
247 return entry_space / sizeof(struct dx_entry);
248}
249
250/*
251 * Debug
252 */
253#ifdef DX_DEBUG
254static void dx_show_index (char * label, struct dx_entry *entries)
255{
256 int i, n = dx_get_count (entries);
257 printk("%s index ", label);
258 for (i = 0; i < n; i++)
259 {
260 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
261 }
262 printk("\n");
263}
264
265struct stats
266{
267 unsigned names;
268 unsigned space;
269 unsigned bcount;
270};
271
272static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
273 int size, int show_names)
274{
275 unsigned names = 0, space = 0;
276 char *base = (char *) de;
277 struct dx_hash_info h = *hinfo;
278
279 printk("names: ");
280 while ((char *) de < base + size)
281 {
282 if (de->inode)
283 {
284 if (show_names)
285 {
286 int len = de->name_len;
287 char *name = de->name;
288 while (len--) printk("%c", *name++);
289 ext3fs_dirhash(de->name, de->name_len, &h);
290 printk(":%x.%u ", h.hash,
291 (unsigned) ((char *) de - base));
292 }
293 space += EXT3_DIR_REC_LEN(de->name_len);
294 names++;
295 }
296 de = ext3_next_entry(de);
297 }
298 printk("(%i)\n", names);
299 return (struct stats) { names, space, 1 };
300}
301
302struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
303 struct dx_entry *entries, int levels)
304{
305 unsigned blocksize = dir->i_sb->s_blocksize;
306 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
307 unsigned bcount = 0;
308 struct buffer_head *bh;
309 int err;
310 printk("%i indexed blocks...\n", count);
311 for (i = 0; i < count; i++, entries++)
312 {
313 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
314 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
315 struct stats stats;
316 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
317 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
318 stats = levels?
319 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
320 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
321 names += stats.names;
322 space += stats.space;
323 bcount += stats.bcount;
324 brelse (bh);
325 }
326 if (bcount)
327 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
328 names, space/bcount,(space/bcount)*100/blocksize);
329 return (struct stats) { names, space, bcount};
330}
331#endif /* DX_DEBUG */
332
333/*
334 * Probe for a directory leaf block to search.
335 *
336 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
337 * error in the directory index, and the caller should fall back to
338 * searching the directory normally. The callers of dx_probe **MUST**
339 * check for this error code, and make sure it never gets reflected
340 * back to userspace.
341 */
342static struct dx_frame *
343dx_probe(struct qstr *entry, struct inode *dir,
344 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
345{
346 unsigned count, indirect;
347 struct dx_entry *at, *entries, *p, *q, *m;
348 struct dx_root *root;
349 struct buffer_head *bh;
350 struct dx_frame *frame = frame_in;
351 u32 hash;
352
353 frame->bh = NULL;
354 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
355 goto fail;
356 root = (struct dx_root *) bh->b_data;
357 if (root->info.hash_version != DX_HASH_TEA &&
358 root->info.hash_version != DX_HASH_HALF_MD4 &&
359 root->info.hash_version != DX_HASH_LEGACY) {
360 ext3_warning(dir->i_sb, __func__,
361 "Unrecognised inode hash code %d",
362 root->info.hash_version);
363 brelse(bh);
364 *err = ERR_BAD_DX_DIR;
365 goto fail;
366 }
367 hinfo->hash_version = root->info.hash_version;
368 if (hinfo->hash_version <= DX_HASH_TEA)
369 hinfo->hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
370 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
371 if (entry)
372 ext3fs_dirhash(entry->name, entry->len, hinfo);
373 hash = hinfo->hash;
374
375 if (root->info.unused_flags & 1) {
376 ext3_warning(dir->i_sb, __func__,
377 "Unimplemented inode hash flags: %#06x",
378 root->info.unused_flags);
379 brelse(bh);
380 *err = ERR_BAD_DX_DIR;
381 goto fail;
382 }
383
384 if ((indirect = root->info.indirect_levels) > 1) {
385 ext3_warning(dir->i_sb, __func__,
386 "Unimplemented inode hash depth: %#06x",
387 root->info.indirect_levels);
388 brelse(bh);
389 *err = ERR_BAD_DX_DIR;
390 goto fail;
391 }
392
393 entries = (struct dx_entry *) (((char *)&root->info) +
394 root->info.info_length);
395
396 if (dx_get_limit(entries) != dx_root_limit(dir,
397 root->info.info_length)) {
398 ext3_warning(dir->i_sb, __func__,
399 "dx entry: limit != root limit");
400 brelse(bh);
401 *err = ERR_BAD_DX_DIR;
402 goto fail;
403 }
404
405 dxtrace (printk("Look up %x", hash));
406 while (1)
407 {
408 count = dx_get_count(entries);
409 if (!count || count > dx_get_limit(entries)) {
410 ext3_warning(dir->i_sb, __func__,
411 "dx entry: no count or count > limit");
412 brelse(bh);
413 *err = ERR_BAD_DX_DIR;
414 goto fail2;
415 }
416
417 p = entries + 1;
418 q = entries + count - 1;
419 while (p <= q)
420 {
421 m = p + (q - p)/2;
422 dxtrace(printk("."));
423 if (dx_get_hash(m) > hash)
424 q = m - 1;
425 else
426 p = m + 1;
427 }
428
429 if (0) // linear search cross check
430 {
431 unsigned n = count - 1;
432 at = entries;
433 while (n--)
434 {
435 dxtrace(printk(","));
436 if (dx_get_hash(++at) > hash)
437 {
438 at--;
439 break;
440 }
441 }
442 assert (at == p - 1);
443 }
444
445 at = p - 1;
446 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
447 frame->bh = bh;
448 frame->entries = entries;
449 frame->at = at;
450 if (!indirect--) return frame;
451 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
452 goto fail2;
453 at = entries = ((struct dx_node *) bh->b_data)->entries;
454 if (dx_get_limit(entries) != dx_node_limit (dir)) {
455 ext3_warning(dir->i_sb, __func__,
456 "dx entry: limit != node limit");
457 brelse(bh);
458 *err = ERR_BAD_DX_DIR;
459 goto fail2;
460 }
461 frame++;
462 frame->bh = NULL;
463 }
464fail2:
465 while (frame >= frame_in) {
466 brelse(frame->bh);
467 frame--;
468 }
469fail:
470 if (*err == ERR_BAD_DX_DIR)
471 ext3_warning(dir->i_sb, __func__,
472 "Corrupt dir inode %ld, running e2fsck is "
473 "recommended.", dir->i_ino);
474 return NULL;
475}
476
477static void dx_release (struct dx_frame *frames)
478{
479 if (frames[0].bh == NULL)
480 return;
481
482 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
483 brelse(frames[1].bh);
484 brelse(frames[0].bh);
485}
486
487/*
488 * This function increments the frame pointer to search the next leaf
489 * block, and reads in the necessary intervening nodes if the search
490 * should be necessary. Whether or not the search is necessary is
491 * controlled by the hash parameter. If the hash value is even, then
492 * the search is only continued if the next block starts with that
493 * hash value. This is used if we are searching for a specific file.
494 *
495 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
496 *
497 * This function returns 1 if the caller should continue to search,
498 * or 0 if it should not. If there is an error reading one of the
499 * index blocks, it will a negative error code.
500 *
501 * If start_hash is non-null, it will be filled in with the starting
502 * hash of the next page.
503 */
504static int ext3_htree_next_block(struct inode *dir, __u32 hash,
505 struct dx_frame *frame,
506 struct dx_frame *frames,
507 __u32 *start_hash)
508{
509 struct dx_frame *p;
510 struct buffer_head *bh;
511 int err, num_frames = 0;
512 __u32 bhash;
513
514 p = frame;
515 /*
516 * Find the next leaf page by incrementing the frame pointer.
517 * If we run out of entries in the interior node, loop around and
518 * increment pointer in the parent node. When we break out of
519 * this loop, num_frames indicates the number of interior
520 * nodes need to be read.
521 */
522 while (1) {
523 if (++(p->at) < p->entries + dx_get_count(p->entries))
524 break;
525 if (p == frames)
526 return 0;
527 num_frames++;
528 p--;
529 }
530
531 /*
532 * If the hash is 1, then continue only if the next page has a
533 * continuation hash of any value. This is used for readdir
534 * handling. Otherwise, check to see if the hash matches the
535 * desired contiuation hash. If it doesn't, return since
536 * there's no point to read in the successive index pages.
537 */
538 bhash = dx_get_hash(p->at);
539 if (start_hash)
540 *start_hash = bhash;
541 if ((hash & 1) == 0) {
542 if ((bhash & ~1) != hash)
543 return 0;
544 }
545 /*
546 * If the hash is HASH_NB_ALWAYS, we always go to the next
547 * block so no check is necessary
548 */
549 while (num_frames--) {
550 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
551 0, &err)))
552 return err; /* Failure */
553 p++;
554 brelse (p->bh);
555 p->bh = bh;
556 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
557 }
558 return 1;
559}
560
561
562/*
563 * This function fills a red-black tree with information from a
564 * directory block. It returns the number directory entries loaded
565 * into the tree. If there is an error it is returned in err.
566 */
567static int htree_dirblock_to_tree(struct file *dir_file,
568 struct inode *dir, int block,
569 struct dx_hash_info *hinfo,
570 __u32 start_hash, __u32 start_minor_hash)
571{
572 struct buffer_head *bh;
573 struct ext3_dir_entry_2 *de, *top;
574 int err, count = 0;
575
576 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
577 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
578 return err;
579
580 de = (struct ext3_dir_entry_2 *) bh->b_data;
581 top = (struct ext3_dir_entry_2 *) ((char *) de +
582 dir->i_sb->s_blocksize -
583 EXT3_DIR_REC_LEN(0));
584 for (; de < top; de = ext3_next_entry(de)) {
585 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
586 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
587 +((char *)de - bh->b_data))) {
588 /* On error, skip the f_pos to the next block. */
589 dir_file->f_pos = (dir_file->f_pos |
590 (dir->i_sb->s_blocksize - 1)) + 1;
591 brelse (bh);
592 return count;
593 }
594 ext3fs_dirhash(de->name, de->name_len, hinfo);
595 if ((hinfo->hash < start_hash) ||
596 ((hinfo->hash == start_hash) &&
597 (hinfo->minor_hash < start_minor_hash)))
598 continue;
599 if (de->inode == 0)
600 continue;
601 if ((err = ext3_htree_store_dirent(dir_file,
602 hinfo->hash, hinfo->minor_hash, de)) != 0) {
603 brelse(bh);
604 return err;
605 }
606 count++;
607 }
608 brelse(bh);
609 return count;
610}
611
612
613/*
614 * This function fills a red-black tree with information from a
615 * directory. We start scanning the directory in hash order, starting
616 * at start_hash and start_minor_hash.
617 *
618 * This function returns the number of entries inserted into the tree,
619 * or a negative error code.
620 */
621int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
622 __u32 start_minor_hash, __u32 *next_hash)
623{
624 struct dx_hash_info hinfo;
625 struct ext3_dir_entry_2 *de;
626 struct dx_frame frames[2], *frame;
627 struct inode *dir;
628 int block, err;
629 int count = 0;
630 int ret;
631 __u32 hashval;
632
633 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
634 start_minor_hash));
635 dir = dir_file->f_path.dentry->d_inode;
636 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
637 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
638 if (hinfo.hash_version <= DX_HASH_TEA)
639 hinfo.hash_version +=
640 EXT3_SB(dir->i_sb)->s_hash_unsigned;
641 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
642 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
643 start_hash, start_minor_hash);
644 *next_hash = ~0;
645 return count;
646 }
647 hinfo.hash = start_hash;
648 hinfo.minor_hash = 0;
649 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
650 if (!frame)
651 return err;
652
653 /* Add '.' and '..' from the htree header */
654 if (!start_hash && !start_minor_hash) {
655 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
656 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
657 goto errout;
658 count++;
659 }
660 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
661 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
662 de = ext3_next_entry(de);
663 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
664 goto errout;
665 count++;
666 }
667
668 while (1) {
669 block = dx_get_block(frame->at);
670 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
671 start_hash, start_minor_hash);
672 if (ret < 0) {
673 err = ret;
674 goto errout;
675 }
676 count += ret;
677 hashval = ~0;
678 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
679 frame, frames, &hashval);
680 *next_hash = hashval;
681 if (ret < 0) {
682 err = ret;
683 goto errout;
684 }
685 /*
686 * Stop if: (a) there are no more entries, or
687 * (b) we have inserted at least one entry and the
688 * next hash value is not a continuation
689 */
690 if ((ret == 0) ||
691 (count && ((hashval & 1) == 0)))
692 break;
693 }
694 dx_release(frames);
695 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
696 count, *next_hash));
697 return count;
698errout:
699 dx_release(frames);
700 return (err);
701}
702
703
704/*
705 * Directory block splitting, compacting
706 */
707
708/*
709 * Create map of hash values, offsets, and sizes, stored at end of block.
710 * Returns number of entries mapped.
711 */
712static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
713 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
714{
715 int count = 0;
716 char *base = (char *) de;
717 struct dx_hash_info h = *hinfo;
718
719 while ((char *) de < base + blocksize)
720 {
721 if (de->name_len && de->inode) {
722 ext3fs_dirhash(de->name, de->name_len, &h);
723 map_tail--;
724 map_tail->hash = h.hash;
725 map_tail->offs = (u16) ((char *) de - base);
726 map_tail->size = le16_to_cpu(de->rec_len);
727 count++;
728 cond_resched();
729 }
730 /* XXX: do we need to check rec_len == 0 case? -Chris */
731 de = ext3_next_entry(de);
732 }
733 return count;
734}
735
736/* Sort map by hash value */
737static void dx_sort_map (struct dx_map_entry *map, unsigned count)
738{
739 struct dx_map_entry *p, *q, *top = map + count - 1;
740 int more;
741 /* Combsort until bubble sort doesn't suck */
742 while (count > 2)
743 {
744 count = count*10/13;
745 if (count - 9 < 2) /* 9, 10 -> 11 */
746 count = 11;
747 for (p = top, q = p - count; q >= map; p--, q--)
748 if (p->hash < q->hash)
749 swap(*p, *q);
750 }
751 /* Garden variety bubble sort */
752 do {
753 more = 0;
754 q = top;
755 while (q-- > map)
756 {
757 if (q[1].hash >= q[0].hash)
758 continue;
759 swap(*(q+1), *q);
760 more = 1;
761 }
762 } while(more);
763}
764
765static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
766{
767 struct dx_entry *entries = frame->entries;
768 struct dx_entry *old = frame->at, *new = old + 1;
769 int count = dx_get_count(entries);
770
771 assert(count < dx_get_limit(entries));
772 assert(old < entries + count);
773 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
774 dx_set_hash(new, hash);
775 dx_set_block(new, block);
776 dx_set_count(entries, count + 1);
777}
778
779static void ext3_update_dx_flag(struct inode *inode)
780{
781 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
782 EXT3_FEATURE_COMPAT_DIR_INDEX))
783 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
784}
785
786/*
787 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
788 *
789 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
790 * `de != NULL' is guaranteed by caller.
791 */
792static inline int ext3_match (int len, const char * const name,
793 struct ext3_dir_entry_2 * de)
794{
795 if (len != de->name_len)
796 return 0;
797 if (!de->inode)
798 return 0;
799 return !memcmp(name, de->name, len);
800}
801
802/*
803 * Returns 0 if not found, -1 on failure, and 1 on success
804 */
805static inline int search_dirblock(struct buffer_head * bh,
806 struct inode *dir,
807 struct qstr *child,
808 unsigned long offset,
809 struct ext3_dir_entry_2 ** res_dir)
810{
811 struct ext3_dir_entry_2 * de;
812 char * dlimit;
813 int de_len;
814 const char *name = child->name;
815 int namelen = child->len;
816
817 de = (struct ext3_dir_entry_2 *) bh->b_data;
818 dlimit = bh->b_data + dir->i_sb->s_blocksize;
819 while ((char *) de < dlimit) {
820 /* this code is executed quadratically often */
821 /* do minimal checking `by hand' */
822
823 if ((char *) de + namelen <= dlimit &&
824 ext3_match (namelen, name, de)) {
825 /* found a match - just to be sure, do a full check */
826 if (!ext3_check_dir_entry("ext3_find_entry",
827 dir, de, bh, offset))
828 return -1;
829 *res_dir = de;
830 return 1;
831 }
832 /* prevent looping on a bad block */
833 de_len = ext3_rec_len_from_disk(de->rec_len);
834 if (de_len <= 0)
835 return -1;
836 offset += de_len;
837 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
838 }
839 return 0;
840}
841
842
843/*
844 * ext3_find_entry()
845 *
846 * finds an entry in the specified directory with the wanted name. It
847 * returns the cache buffer in which the entry was found, and the entry
848 * itself (as a parameter - res_dir). It does NOT read the inode of the
849 * entry - you'll have to do that yourself if you want to.
850 *
851 * The returned buffer_head has ->b_count elevated. The caller is expected
852 * to brelse() it when appropriate.
853 */
854static struct buffer_head *ext3_find_entry(struct inode *dir,
855 struct qstr *entry,
856 struct ext3_dir_entry_2 **res_dir)
857{
858 struct super_block * sb;
859 struct buffer_head * bh_use[NAMEI_RA_SIZE];
860 struct buffer_head * bh, *ret = NULL;
861 unsigned long start, block, b;
862 const u8 *name = entry->name;
863 int ra_max = 0; /* Number of bh's in the readahead
864 buffer, bh_use[] */
865 int ra_ptr = 0; /* Current index into readahead
866 buffer */
867 int num = 0;
868 int nblocks, i, err;
869 int namelen;
870
871 *res_dir = NULL;
872 sb = dir->i_sb;
873 namelen = entry->len;
874 if (namelen > EXT3_NAME_LEN)
875 return NULL;
876 if ((namelen <= 2) && (name[0] == '.') &&
877 (name[1] == '.' || name[1] == 0)) {
878 /*
879 * "." or ".." will only be in the first block
880 * NFS may look up ".."; "." should be handled by the VFS
881 */
882 block = start = 0;
883 nblocks = 1;
884 goto restart;
885 }
886 if (is_dx(dir)) {
887 bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
888 /*
889 * On success, or if the error was file not found,
890 * return. Otherwise, fall back to doing a search the
891 * old fashioned way.
892 */
893 if (bh || (err != ERR_BAD_DX_DIR))
894 return bh;
895 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
896 }
897 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
898 start = EXT3_I(dir)->i_dir_start_lookup;
899 if (start >= nblocks)
900 start = 0;
901 block = start;
902restart:
903 do {
904 /*
905 * We deal with the read-ahead logic here.
906 */
907 if (ra_ptr >= ra_max) {
908 /* Refill the readahead buffer */
909 ra_ptr = 0;
910 b = block;
911 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
912 /*
913 * Terminate if we reach the end of the
914 * directory and must wrap, or if our
915 * search has finished at this block.
916 */
917 if (b >= nblocks || (num && block == start)) {
918 bh_use[ra_max] = NULL;
919 break;
920 }
921 num++;
922 bh = ext3_getblk(NULL, dir, b++, 0, &err);
923 bh_use[ra_max] = bh;
924 if (bh)
925 ll_rw_block(READ | REQ_META | REQ_PRIO,
926 1, &bh);
927 }
928 }
929 if ((bh = bh_use[ra_ptr++]) == NULL)
930 goto next;
931 wait_on_buffer(bh);
932 if (!buffer_uptodate(bh)) {
933 /* read error, skip block & hope for the best */
934 ext3_error(sb, __func__, "reading directory #%lu "
935 "offset %lu", dir->i_ino, block);
936 brelse(bh);
937 goto next;
938 }
939 i = search_dirblock(bh, dir, entry,
940 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
941 if (i == 1) {
942 EXT3_I(dir)->i_dir_start_lookup = block;
943 ret = bh;
944 goto cleanup_and_exit;
945 } else {
946 brelse(bh);
947 if (i < 0)
948 goto cleanup_and_exit;
949 }
950 next:
951 if (++block >= nblocks)
952 block = 0;
953 } while (block != start);
954
955 /*
956 * If the directory has grown while we were searching, then
957 * search the last part of the directory before giving up.
958 */
959 block = nblocks;
960 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
961 if (block < nblocks) {
962 start = 0;
963 goto restart;
964 }
965
966cleanup_and_exit:
967 /* Clean up the read-ahead blocks */
968 for (; ra_ptr < ra_max; ra_ptr++)
969 brelse (bh_use[ra_ptr]);
970 return ret;
971}
972
973static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
974 struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
975 int *err)
976{
977 struct super_block *sb = dir->i_sb;
978 struct dx_hash_info hinfo;
979 struct dx_frame frames[2], *frame;
980 struct buffer_head *bh;
981 unsigned long block;
982 int retval;
983
984 if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
985 return NULL;
986 do {
987 block = dx_get_block(frame->at);
988 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
989 goto errout;
990
991 retval = search_dirblock(bh, dir, entry,
992 block << EXT3_BLOCK_SIZE_BITS(sb),
993 res_dir);
994 if (retval == 1) {
995 dx_release(frames);
996 return bh;
997 }
998 brelse(bh);
999 if (retval == -1) {
1000 *err = ERR_BAD_DX_DIR;
1001 goto errout;
1002 }
1003
1004 /* Check to see if we should continue to search */
1005 retval = ext3_htree_next_block(dir, hinfo.hash, frame,
1006 frames, NULL);
1007 if (retval < 0) {
1008 ext3_warning(sb, __func__,
1009 "error reading index page in directory #%lu",
1010 dir->i_ino);
1011 *err = retval;
1012 goto errout;
1013 }
1014 } while (retval == 1);
1015
1016 *err = -ENOENT;
1017errout:
1018 dxtrace(printk("%s not found\n", entry->name));
1019 dx_release (frames);
1020 return NULL;
1021}
1022
1023static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1024{
1025 struct inode * inode;
1026 struct ext3_dir_entry_2 * de;
1027 struct buffer_head * bh;
1028
1029 if (dentry->d_name.len > EXT3_NAME_LEN)
1030 return ERR_PTR(-ENAMETOOLONG);
1031
1032 bh = ext3_find_entry(dir, &dentry->d_name, &de);
1033 inode = NULL;
1034 if (bh) {
1035 unsigned long ino = le32_to_cpu(de->inode);
1036 brelse (bh);
1037 if (!ext3_valid_inum(dir->i_sb, ino)) {
1038 ext3_error(dir->i_sb, "ext3_lookup",
1039 "bad inode number: %lu", ino);
1040 return ERR_PTR(-EIO);
1041 }
1042 inode = ext3_iget(dir->i_sb, ino);
1043 if (inode == ERR_PTR(-ESTALE)) {
1044 ext3_error(dir->i_sb, __func__,
1045 "deleted inode referenced: %lu",
1046 ino);
1047 return ERR_PTR(-EIO);
1048 }
1049 }
1050 return d_splice_alias(inode, dentry);
1051}
1052
1053
1054struct dentry *ext3_get_parent(struct dentry *child)
1055{
1056 unsigned long ino;
1057 struct qstr dotdot = {.name = "..", .len = 2};
1058 struct ext3_dir_entry_2 * de;
1059 struct buffer_head *bh;
1060
1061 bh = ext3_find_entry(child->d_inode, &dotdot, &de);
1062 if (!bh)
1063 return ERR_PTR(-ENOENT);
1064 ino = le32_to_cpu(de->inode);
1065 brelse(bh);
1066
1067 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1068 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1069 "bad inode number: %lu", ino);
1070 return ERR_PTR(-EIO);
1071 }
1072
1073 return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
1074}
1075
1076#define S_SHIFT 12
1077static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1078 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1079 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1080 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1081 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1082 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1083 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1084 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1085};
1086
1087static inline void ext3_set_de_type(struct super_block *sb,
1088 struct ext3_dir_entry_2 *de,
1089 umode_t mode) {
1090 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1091 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1092}
1093
1094/*
1095 * Move count entries from end of map between two memory locations.
1096 * Returns pointer to last entry moved.
1097 */
1098static struct ext3_dir_entry_2 *
1099dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1100{
1101 unsigned rec_len = 0;
1102
1103 while (count--) {
1104 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1105 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1106 memcpy (to, de, rec_len);
1107 ((struct ext3_dir_entry_2 *) to)->rec_len =
1108 ext3_rec_len_to_disk(rec_len);
1109 de->inode = 0;
1110 map++;
1111 to += rec_len;
1112 }
1113 return (struct ext3_dir_entry_2 *) (to - rec_len);
1114}
1115
1116/*
1117 * Compact each dir entry in the range to the minimal rec_len.
1118 * Returns pointer to last entry in range.
1119 */
1120static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize)
1121{
1122 struct ext3_dir_entry_2 *next, *to, *prev;
1123 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)base;
1124 unsigned rec_len = 0;
1125
1126 prev = to = de;
1127 while ((char *)de < base + blocksize) {
1128 next = ext3_next_entry(de);
1129 if (de->inode && de->name_len) {
1130 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1131 if (de > to)
1132 memmove(to, de, rec_len);
1133 to->rec_len = ext3_rec_len_to_disk(rec_len);
1134 prev = to;
1135 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1136 }
1137 de = next;
1138 }
1139 return prev;
1140}
1141
1142/*
1143 * Split a full leaf block to make room for a new dir entry.
1144 * Allocate a new block, and move entries so that they are approx. equally full.
1145 * Returns pointer to de in block into which the new entry will be inserted.
1146 */
1147static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1148 struct buffer_head **bh,struct dx_frame *frame,
1149 struct dx_hash_info *hinfo, int *error)
1150{
1151 unsigned blocksize = dir->i_sb->s_blocksize;
1152 unsigned count, continued;
1153 struct buffer_head *bh2;
1154 u32 newblock;
1155 u32 hash2;
1156 struct dx_map_entry *map;
1157 char *data1 = (*bh)->b_data, *data2;
1158 unsigned split, move, size;
1159 struct ext3_dir_entry_2 *de = NULL, *de2;
1160 int err = 0, i;
1161
1162 bh2 = ext3_append (handle, dir, &newblock, &err);
1163 if (!(bh2)) {
1164 brelse(*bh);
1165 *bh = NULL;
1166 goto errout;
1167 }
1168
1169 BUFFER_TRACE(*bh, "get_write_access");
1170 err = ext3_journal_get_write_access(handle, *bh);
1171 if (err)
1172 goto journal_error;
1173
1174 BUFFER_TRACE(frame->bh, "get_write_access");
1175 err = ext3_journal_get_write_access(handle, frame->bh);
1176 if (err)
1177 goto journal_error;
1178
1179 data2 = bh2->b_data;
1180
1181 /* create map in the end of data2 block */
1182 map = (struct dx_map_entry *) (data2 + blocksize);
1183 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1184 blocksize, hinfo, map);
1185 map -= count;
1186 dx_sort_map (map, count);
1187 /* Split the existing block in the middle, size-wise */
1188 size = 0;
1189 move = 0;
1190 for (i = count-1; i >= 0; i--) {
1191 /* is more than half of this entry in 2nd half of the block? */
1192 if (size + map[i].size/2 > blocksize/2)
1193 break;
1194 size += map[i].size;
1195 move++;
1196 }
1197 /* map index at which we will split */
1198 split = count - move;
1199 hash2 = map[split].hash;
1200 continued = hash2 == map[split - 1].hash;
1201 dxtrace(printk("Split block %i at %x, %i/%i\n",
1202 dx_get_block(frame->at), hash2, split, count-split));
1203
1204 /* Fancy dance to stay within two buffers */
1205 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1206 de = dx_pack_dirents(data1,blocksize);
1207 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1208 de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
1209 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1210 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1211
1212 /* Which block gets the new entry? */
1213 if (hinfo->hash >= hash2)
1214 {
1215 swap(*bh, bh2);
1216 de = de2;
1217 }
1218 dx_insert_block (frame, hash2 + continued, newblock);
1219 err = ext3_journal_dirty_metadata (handle, bh2);
1220 if (err)
1221 goto journal_error;
1222 err = ext3_journal_dirty_metadata (handle, frame->bh);
1223 if (err)
1224 goto journal_error;
1225 brelse (bh2);
1226 dxtrace(dx_show_index ("frame", frame->entries));
1227 return de;
1228
1229journal_error:
1230 brelse(*bh);
1231 brelse(bh2);
1232 *bh = NULL;
1233 ext3_std_error(dir->i_sb, err);
1234errout:
1235 *error = err;
1236 return NULL;
1237}
1238
1239
1240/*
1241 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1242 * it points to a directory entry which is guaranteed to be large
1243 * enough for new directory entry. If de is NULL, then
1244 * add_dirent_to_buf will attempt search the directory block for
1245 * space. It will return -ENOSPC if no space is available, and -EIO
1246 * and -EEXIST if directory entry already exists.
1247 *
1248 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1249 * all other cases bh is released.
1250 */
1251static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1252 struct inode *inode, struct ext3_dir_entry_2 *de,
1253 struct buffer_head * bh)
1254{
1255 struct inode *dir = dentry->d_parent->d_inode;
1256 const char *name = dentry->d_name.name;
1257 int namelen = dentry->d_name.len;
1258 unsigned long offset = 0;
1259 unsigned short reclen;
1260 int nlen, rlen, err;
1261 char *top;
1262
1263 reclen = EXT3_DIR_REC_LEN(namelen);
1264 if (!de) {
1265 de = (struct ext3_dir_entry_2 *)bh->b_data;
1266 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1267 while ((char *) de <= top) {
1268 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1269 bh, offset)) {
1270 brelse (bh);
1271 return -EIO;
1272 }
1273 if (ext3_match (namelen, name, de)) {
1274 brelse (bh);
1275 return -EEXIST;
1276 }
1277 nlen = EXT3_DIR_REC_LEN(de->name_len);
1278 rlen = ext3_rec_len_from_disk(de->rec_len);
1279 if ((de->inode? rlen - nlen: rlen) >= reclen)
1280 break;
1281 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1282 offset += rlen;
1283 }
1284 if ((char *) de > top)
1285 return -ENOSPC;
1286 }
1287 BUFFER_TRACE(bh, "get_write_access");
1288 err = ext3_journal_get_write_access(handle, bh);
1289 if (err) {
1290 ext3_std_error(dir->i_sb, err);
1291 brelse(bh);
1292 return err;
1293 }
1294
1295 /* By now the buffer is marked for journaling */
1296 nlen = EXT3_DIR_REC_LEN(de->name_len);
1297 rlen = ext3_rec_len_from_disk(de->rec_len);
1298 if (de->inode) {
1299 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1300 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1301 de->rec_len = ext3_rec_len_to_disk(nlen);
1302 de = de1;
1303 }
1304 de->file_type = EXT3_FT_UNKNOWN;
1305 if (inode) {
1306 de->inode = cpu_to_le32(inode->i_ino);
1307 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1308 } else
1309 de->inode = 0;
1310 de->name_len = namelen;
1311 memcpy (de->name, name, namelen);
1312 /*
1313 * XXX shouldn't update any times until successful
1314 * completion of syscall, but too many callers depend
1315 * on this.
1316 *
1317 * XXX similarly, too many callers depend on
1318 * ext3_new_inode() setting the times, but error
1319 * recovery deletes the inode, so the worst that can
1320 * happen is that the times are slightly out of date
1321 * and/or different from the directory change time.
1322 */
1323 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1324 ext3_update_dx_flag(dir);
1325 dir->i_version++;
1326 ext3_mark_inode_dirty(handle, dir);
1327 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1328 err = ext3_journal_dirty_metadata(handle, bh);
1329 if (err)
1330 ext3_std_error(dir->i_sb, err);
1331 brelse(bh);
1332 return 0;
1333}
1334
1335/*
1336 * This converts a one block unindexed directory to a 3 block indexed
1337 * directory, and adds the dentry to the indexed directory.
1338 */
1339static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1340 struct inode *inode, struct buffer_head *bh)
1341{
1342 struct inode *dir = dentry->d_parent->d_inode;
1343 const char *name = dentry->d_name.name;
1344 int namelen = dentry->d_name.len;
1345 struct buffer_head *bh2;
1346 struct dx_root *root;
1347 struct dx_frame frames[2], *frame;
1348 struct dx_entry *entries;
1349 struct ext3_dir_entry_2 *de, *de2;
1350 char *data1, *top;
1351 unsigned len;
1352 int retval;
1353 unsigned blocksize;
1354 struct dx_hash_info hinfo;
1355 u32 block;
1356 struct fake_dirent *fde;
1357
1358 blocksize = dir->i_sb->s_blocksize;
1359 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1360 retval = ext3_journal_get_write_access(handle, bh);
1361 if (retval) {
1362 ext3_std_error(dir->i_sb, retval);
1363 brelse(bh);
1364 return retval;
1365 }
1366 root = (struct dx_root *) bh->b_data;
1367
1368 /* The 0th block becomes the root, move the dirents out */
1369 fde = &root->dotdot;
1370 de = (struct ext3_dir_entry_2 *)((char *)fde +
1371 ext3_rec_len_from_disk(fde->rec_len));
1372 if ((char *) de >= (((char *) root) + blocksize)) {
1373 ext3_error(dir->i_sb, __func__,
1374 "invalid rec_len for '..' in inode %lu",
1375 dir->i_ino);
1376 brelse(bh);
1377 return -EIO;
1378 }
1379 len = ((char *) root) + blocksize - (char *) de;
1380
1381 bh2 = ext3_append (handle, dir, &block, &retval);
1382 if (!(bh2)) {
1383 brelse(bh);
1384 return retval;
1385 }
1386 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1387 data1 = bh2->b_data;
1388
1389 memcpy (data1, de, len);
1390 de = (struct ext3_dir_entry_2 *) data1;
1391 top = data1 + len;
1392 while ((char *)(de2 = ext3_next_entry(de)) < top)
1393 de = de2;
1394 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1395 /* Initialize the root; the dot dirents already exist */
1396 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1397 de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
1398 memset (&root->info, 0, sizeof(root->info));
1399 root->info.info_length = sizeof(root->info);
1400 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1401 entries = root->entries;
1402 dx_set_block (entries, 1);
1403 dx_set_count (entries, 1);
1404 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1405
1406 /* Initialize as for dx_probe */
1407 hinfo.hash_version = root->info.hash_version;
1408 if (hinfo.hash_version <= DX_HASH_TEA)
1409 hinfo.hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
1410 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1411 ext3fs_dirhash(name, namelen, &hinfo);
1412 frame = frames;
1413 frame->entries = entries;
1414 frame->at = entries;
1415 frame->bh = bh;
1416 bh = bh2;
1417 /*
1418 * Mark buffers dirty here so that if do_split() fails we write a
1419 * consistent set of buffers to disk.
1420 */
1421 ext3_journal_dirty_metadata(handle, frame->bh);
1422 ext3_journal_dirty_metadata(handle, bh);
1423 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1424 if (!de) {
1425 ext3_mark_inode_dirty(handle, dir);
1426 dx_release(frames);
1427 return retval;
1428 }
1429 dx_release(frames);
1430
1431 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1432}
1433
1434/*
1435 * ext3_add_entry()
1436 *
1437 * adds a file entry to the specified directory, using the same
1438 * semantics as ext3_find_entry(). It returns NULL if it failed.
1439 *
1440 * NOTE!! The inode part of 'de' is left at 0 - which means you
1441 * may not sleep between calling this and putting something into
1442 * the entry, as someone else might have used it while you slept.
1443 */
1444static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1445 struct inode *inode)
1446{
1447 struct inode *dir = dentry->d_parent->d_inode;
1448 struct buffer_head * bh;
1449 struct ext3_dir_entry_2 *de;
1450 struct super_block * sb;
1451 int retval;
1452 int dx_fallback=0;
1453 unsigned blocksize;
1454 u32 block, blocks;
1455
1456 sb = dir->i_sb;
1457 blocksize = sb->s_blocksize;
1458 if (!dentry->d_name.len)
1459 return -EINVAL;
1460 if (is_dx(dir)) {
1461 retval = ext3_dx_add_entry(handle, dentry, inode);
1462 if (!retval || (retval != ERR_BAD_DX_DIR))
1463 return retval;
1464 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1465 dx_fallback++;
1466 ext3_mark_inode_dirty(handle, dir);
1467 }
1468 blocks = dir->i_size >> sb->s_blocksize_bits;
1469 for (block = 0; block < blocks; block++) {
1470 bh = ext3_bread(handle, dir, block, 0, &retval);
1471 if(!bh)
1472 return retval;
1473 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1474 if (retval != -ENOSPC)
1475 return retval;
1476
1477 if (blocks == 1 && !dx_fallback &&
1478 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1479 return make_indexed_dir(handle, dentry, inode, bh);
1480 brelse(bh);
1481 }
1482 bh = ext3_append(handle, dir, &block, &retval);
1483 if (!bh)
1484 return retval;
1485 de = (struct ext3_dir_entry_2 *) bh->b_data;
1486 de->inode = 0;
1487 de->rec_len = ext3_rec_len_to_disk(blocksize);
1488 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1489}
1490
1491/*
1492 * Returns 0 for success, or a negative error value
1493 */
1494static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1495 struct inode *inode)
1496{
1497 struct dx_frame frames[2], *frame;
1498 struct dx_entry *entries, *at;
1499 struct dx_hash_info hinfo;
1500 struct buffer_head * bh;
1501 struct inode *dir = dentry->d_parent->d_inode;
1502 struct super_block * sb = dir->i_sb;
1503 struct ext3_dir_entry_2 *de;
1504 int err;
1505
1506 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1507 if (!frame)
1508 return err;
1509 entries = frame->entries;
1510 at = frame->at;
1511
1512 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1513 goto cleanup;
1514
1515 BUFFER_TRACE(bh, "get_write_access");
1516 err = ext3_journal_get_write_access(handle, bh);
1517 if (err)
1518 goto journal_error;
1519
1520 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1521 if (err != -ENOSPC) {
1522 bh = NULL;
1523 goto cleanup;
1524 }
1525
1526 /* Block full, should compress but for now just split */
1527 dxtrace(printk("using %u of %u node entries\n",
1528 dx_get_count(entries), dx_get_limit(entries)));
1529 /* Need to split index? */
1530 if (dx_get_count(entries) == dx_get_limit(entries)) {
1531 u32 newblock;
1532 unsigned icount = dx_get_count(entries);
1533 int levels = frame - frames;
1534 struct dx_entry *entries2;
1535 struct dx_node *node2;
1536 struct buffer_head *bh2;
1537
1538 if (levels && (dx_get_count(frames->entries) ==
1539 dx_get_limit(frames->entries))) {
1540 ext3_warning(sb, __func__,
1541 "Directory index full!");
1542 err = -ENOSPC;
1543 goto cleanup;
1544 }
1545 bh2 = ext3_append (handle, dir, &newblock, &err);
1546 if (!(bh2))
1547 goto cleanup;
1548 node2 = (struct dx_node *)(bh2->b_data);
1549 entries2 = node2->entries;
1550 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1551 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
1552 BUFFER_TRACE(frame->bh, "get_write_access");
1553 err = ext3_journal_get_write_access(handle, frame->bh);
1554 if (err)
1555 goto journal_error;
1556 if (levels) {
1557 unsigned icount1 = icount/2, icount2 = icount - icount1;
1558 unsigned hash2 = dx_get_hash(entries + icount1);
1559 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1560
1561 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1562 err = ext3_journal_get_write_access(handle,
1563 frames[0].bh);
1564 if (err)
1565 goto journal_error;
1566
1567 memcpy ((char *) entries2, (char *) (entries + icount1),
1568 icount2 * sizeof(struct dx_entry));
1569 dx_set_count (entries, icount1);
1570 dx_set_count (entries2, icount2);
1571 dx_set_limit (entries2, dx_node_limit(dir));
1572
1573 /* Which index block gets the new entry? */
1574 if (at - entries >= icount1) {
1575 frame->at = at = at - entries - icount1 + entries2;
1576 frame->entries = entries = entries2;
1577 swap(frame->bh, bh2);
1578 }
1579 dx_insert_block (frames + 0, hash2, newblock);
1580 dxtrace(dx_show_index ("node", frames[1].entries));
1581 dxtrace(dx_show_index ("node",
1582 ((struct dx_node *) bh2->b_data)->entries));
1583 err = ext3_journal_dirty_metadata(handle, bh2);
1584 if (err)
1585 goto journal_error;
1586 brelse (bh2);
1587 } else {
1588 dxtrace(printk("Creating second level index...\n"));
1589 memcpy((char *) entries2, (char *) entries,
1590 icount * sizeof(struct dx_entry));
1591 dx_set_limit(entries2, dx_node_limit(dir));
1592
1593 /* Set up root */
1594 dx_set_count(entries, 1);
1595 dx_set_block(entries + 0, newblock);
1596 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1597
1598 /* Add new access path frame */
1599 frame = frames + 1;
1600 frame->at = at = at - entries + entries2;
1601 frame->entries = entries = entries2;
1602 frame->bh = bh2;
1603 err = ext3_journal_get_write_access(handle,
1604 frame->bh);
1605 if (err)
1606 goto journal_error;
1607 }
1608 err = ext3_journal_dirty_metadata(handle, frames[0].bh);
1609 if (err)
1610 goto journal_error;
1611 }
1612 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1613 if (!de)
1614 goto cleanup;
1615 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1616 bh = NULL;
1617 goto cleanup;
1618
1619journal_error:
1620 ext3_std_error(dir->i_sb, err);
1621cleanup:
1622 if (bh)
1623 brelse(bh);
1624 dx_release(frames);
1625 return err;
1626}
1627
1628/*
1629 * ext3_delete_entry deletes a directory entry by merging it with the
1630 * previous entry
1631 */
1632static int ext3_delete_entry (handle_t *handle,
1633 struct inode * dir,
1634 struct ext3_dir_entry_2 * de_del,
1635 struct buffer_head * bh)
1636{
1637 struct ext3_dir_entry_2 * de, * pde;
1638 int i;
1639
1640 i = 0;
1641 pde = NULL;
1642 de = (struct ext3_dir_entry_2 *) bh->b_data;
1643 while (i < bh->b_size) {
1644 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1645 return -EIO;
1646 if (de == de_del) {
1647 int err;
1648
1649 BUFFER_TRACE(bh, "get_write_access");
1650 err = ext3_journal_get_write_access(handle, bh);
1651 if (err)
1652 goto journal_error;
1653
1654 if (pde)
1655 pde->rec_len = ext3_rec_len_to_disk(
1656 ext3_rec_len_from_disk(pde->rec_len) +
1657 ext3_rec_len_from_disk(de->rec_len));
1658 else
1659 de->inode = 0;
1660 dir->i_version++;
1661 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1662 err = ext3_journal_dirty_metadata(handle, bh);
1663 if (err) {
1664journal_error:
1665 ext3_std_error(dir->i_sb, err);
1666 return err;
1667 }
1668 return 0;
1669 }
1670 i += ext3_rec_len_from_disk(de->rec_len);
1671 pde = de;
1672 de = ext3_next_entry(de);
1673 }
1674 return -ENOENT;
1675}
1676
1677static int ext3_add_nondir(handle_t *handle,
1678 struct dentry *dentry, struct inode *inode)
1679{
1680 int err = ext3_add_entry(handle, dentry, inode);
1681 if (!err) {
1682 ext3_mark_inode_dirty(handle, inode);
1683 d_instantiate(dentry, inode);
1684 unlock_new_inode(inode);
1685 return 0;
1686 }
1687 drop_nlink(inode);
1688 unlock_new_inode(inode);
1689 iput(inode);
1690 return err;
1691}
1692
1693/*
1694 * By the time this is called, we already have created
1695 * the directory cache entry for the new file, but it
1696 * is so far negative - it has no inode.
1697 *
1698 * If the create succeeds, we fill in the inode information
1699 * with d_instantiate().
1700 */
1701static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1702 struct nameidata *nd)
1703{
1704 handle_t *handle;
1705 struct inode * inode;
1706 int err, retries = 0;
1707
1708 dquot_initialize(dir);
1709
1710retry:
1711 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1712 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1713 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1714 if (IS_ERR(handle))
1715 return PTR_ERR(handle);
1716
1717 if (IS_DIRSYNC(dir))
1718 handle->h_sync = 1;
1719
1720 inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1721 err = PTR_ERR(inode);
1722 if (!IS_ERR(inode)) {
1723 inode->i_op = &ext3_file_inode_operations;
1724 inode->i_fop = &ext3_file_operations;
1725 ext3_set_aops(inode);
1726 err = ext3_add_nondir(handle, dentry, inode);
1727 }
1728 ext3_journal_stop(handle);
1729 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1730 goto retry;
1731 return err;
1732}
1733
1734static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1735 int mode, dev_t rdev)
1736{
1737 handle_t *handle;
1738 struct inode *inode;
1739 int err, retries = 0;
1740
1741 if (!new_valid_dev(rdev))
1742 return -EINVAL;
1743
1744 dquot_initialize(dir);
1745
1746retry:
1747 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1748 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1749 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1750 if (IS_ERR(handle))
1751 return PTR_ERR(handle);
1752
1753 if (IS_DIRSYNC(dir))
1754 handle->h_sync = 1;
1755
1756 inode = ext3_new_inode (handle, dir, &dentry->d_name, mode);
1757 err = PTR_ERR(inode);
1758 if (!IS_ERR(inode)) {
1759 init_special_inode(inode, inode->i_mode, rdev);
1760#ifdef CONFIG_EXT3_FS_XATTR
1761 inode->i_op = &ext3_special_inode_operations;
1762#endif
1763 err = ext3_add_nondir(handle, dentry, inode);
1764 }
1765 ext3_journal_stop(handle);
1766 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1767 goto retry;
1768 return err;
1769}
1770
1771static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1772{
1773 handle_t *handle;
1774 struct inode * inode;
1775 struct buffer_head * dir_block = NULL;
1776 struct ext3_dir_entry_2 * de;
1777 int err, retries = 0;
1778
1779 if (dir->i_nlink >= EXT3_LINK_MAX)
1780 return -EMLINK;
1781
1782 dquot_initialize(dir);
1783
1784retry:
1785 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
1786 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1787 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1788 if (IS_ERR(handle))
1789 return PTR_ERR(handle);
1790
1791 if (IS_DIRSYNC(dir))
1792 handle->h_sync = 1;
1793
1794 inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFDIR | mode);
1795 err = PTR_ERR(inode);
1796 if (IS_ERR(inode))
1797 goto out_stop;
1798
1799 inode->i_op = &ext3_dir_inode_operations;
1800 inode->i_fop = &ext3_dir_operations;
1801 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1802 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1803 if (!dir_block)
1804 goto out_clear_inode;
1805
1806 BUFFER_TRACE(dir_block, "get_write_access");
1807 err = ext3_journal_get_write_access(handle, dir_block);
1808 if (err)
1809 goto out_clear_inode;
1810
1811 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1812 de->inode = cpu_to_le32(inode->i_ino);
1813 de->name_len = 1;
1814 de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
1815 strcpy (de->name, ".");
1816 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1817 de = ext3_next_entry(de);
1818 de->inode = cpu_to_le32(dir->i_ino);
1819 de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1820 EXT3_DIR_REC_LEN(1));
1821 de->name_len = 2;
1822 strcpy (de->name, "..");
1823 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1824 inode->i_nlink = 2;
1825 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1826 err = ext3_journal_dirty_metadata(handle, dir_block);
1827 if (err)
1828 goto out_clear_inode;
1829
1830 err = ext3_mark_inode_dirty(handle, inode);
1831 if (!err)
1832 err = ext3_add_entry (handle, dentry, inode);
1833
1834 if (err) {
1835out_clear_inode:
1836 inode->i_nlink = 0;
1837 unlock_new_inode(inode);
1838 ext3_mark_inode_dirty(handle, inode);
1839 iput (inode);
1840 goto out_stop;
1841 }
1842 inc_nlink(dir);
1843 ext3_update_dx_flag(dir);
1844 err = ext3_mark_inode_dirty(handle, dir);
1845 if (err)
1846 goto out_clear_inode;
1847
1848 d_instantiate(dentry, inode);
1849 unlock_new_inode(inode);
1850out_stop:
1851 brelse(dir_block);
1852 ext3_journal_stop(handle);
1853 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1854 goto retry;
1855 return err;
1856}
1857
1858/*
1859 * routine to check that the specified directory is empty (for rmdir)
1860 */
1861static int empty_dir (struct inode * inode)
1862{
1863 unsigned long offset;
1864 struct buffer_head * bh;
1865 struct ext3_dir_entry_2 * de, * de1;
1866 struct super_block * sb;
1867 int err = 0;
1868
1869 sb = inode->i_sb;
1870 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1871 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1872 if (err)
1873 ext3_error(inode->i_sb, __func__,
1874 "error %d reading directory #%lu offset 0",
1875 err, inode->i_ino);
1876 else
1877 ext3_warning(inode->i_sb, __func__,
1878 "bad directory (dir #%lu) - no data block",
1879 inode->i_ino);
1880 return 1;
1881 }
1882 de = (struct ext3_dir_entry_2 *) bh->b_data;
1883 de1 = ext3_next_entry(de);
1884 if (le32_to_cpu(de->inode) != inode->i_ino ||
1885 !le32_to_cpu(de1->inode) ||
1886 strcmp (".", de->name) ||
1887 strcmp ("..", de1->name)) {
1888 ext3_warning (inode->i_sb, "empty_dir",
1889 "bad directory (dir #%lu) - no `.' or `..'",
1890 inode->i_ino);
1891 brelse (bh);
1892 return 1;
1893 }
1894 offset = ext3_rec_len_from_disk(de->rec_len) +
1895 ext3_rec_len_from_disk(de1->rec_len);
1896 de = ext3_next_entry(de1);
1897 while (offset < inode->i_size ) {
1898 if (!bh ||
1899 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1900 err = 0;
1901 brelse (bh);
1902 bh = ext3_bread (NULL, inode,
1903 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1904 if (!bh) {
1905 if (err)
1906 ext3_error(sb, __func__,
1907 "error %d reading directory"
1908 " #%lu offset %lu",
1909 err, inode->i_ino, offset);
1910 offset += sb->s_blocksize;
1911 continue;
1912 }
1913 de = (struct ext3_dir_entry_2 *) bh->b_data;
1914 }
1915 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1916 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1917 sb->s_blocksize);
1918 offset = (offset | (sb->s_blocksize - 1)) + 1;
1919 continue;
1920 }
1921 if (le32_to_cpu(de->inode)) {
1922 brelse (bh);
1923 return 0;
1924 }
1925 offset += ext3_rec_len_from_disk(de->rec_len);
1926 de = ext3_next_entry(de);
1927 }
1928 brelse (bh);
1929 return 1;
1930}
1931
1932/* ext3_orphan_add() links an unlinked or truncated inode into a list of
1933 * such inodes, starting at the superblock, in case we crash before the
1934 * file is closed/deleted, or in case the inode truncate spans multiple
1935 * transactions and the last transaction is not recovered after a crash.
1936 *
1937 * At filesystem recovery time, we walk this list deleting unlinked
1938 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1939 */
1940int ext3_orphan_add(handle_t *handle, struct inode *inode)
1941{
1942 struct super_block *sb = inode->i_sb;
1943 struct ext3_iloc iloc;
1944 int err = 0, rc;
1945
1946 mutex_lock(&EXT3_SB(sb)->s_orphan_lock);
1947 if (!list_empty(&EXT3_I(inode)->i_orphan))
1948 goto out_unlock;
1949
1950 /* Orphan handling is only valid for files with data blocks
1951 * being truncated, or files being unlinked. */
1952
1953 /* @@@ FIXME: Observation from aviro:
1954 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
1955 * here (on s_orphan_lock), so race with ext3_link() which might bump
1956 * ->i_nlink. For, say it, character device. Not a regular file,
1957 * not a directory, not a symlink and ->i_nlink > 0.
1958 *
1959 * tytso, 4/25/2009: I'm not sure how that could happen;
1960 * shouldn't the fs core protect us from these sort of
1961 * unlink()/link() races?
1962 */
1963 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1964 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1965
1966 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1967 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1968 if (err)
1969 goto out_unlock;
1970
1971 err = ext3_reserve_inode_write(handle, inode, &iloc);
1972 if (err)
1973 goto out_unlock;
1974
1975 /* Insert this inode at the head of the on-disk orphan list... */
1976 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1977 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1978 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1979 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1980 if (!err)
1981 err = rc;
1982
1983 /* Only add to the head of the in-memory list if all the
1984 * previous operations succeeded. If the orphan_add is going to
1985 * fail (possibly taking the journal offline), we can't risk
1986 * leaving the inode on the orphan list: stray orphan-list
1987 * entries can cause panics at unmount time.
1988 *
1989 * This is safe: on error we're going to ignore the orphan list
1990 * anyway on the next recovery. */
1991 if (!err)
1992 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1993
1994 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1995 jbd_debug(4, "orphan inode %lu will point to %d\n",
1996 inode->i_ino, NEXT_ORPHAN(inode));
1997out_unlock:
1998 mutex_unlock(&EXT3_SB(sb)->s_orphan_lock);
1999 ext3_std_error(inode->i_sb, err);
2000 return err;
2001}
2002
2003/*
2004 * ext3_orphan_del() removes an unlinked or truncated inode from the list
2005 * of such inodes stored on disk, because it is finally being cleaned up.
2006 */
2007int ext3_orphan_del(handle_t *handle, struct inode *inode)
2008{
2009 struct list_head *prev;
2010 struct ext3_inode_info *ei = EXT3_I(inode);
2011 struct ext3_sb_info *sbi;
2012 unsigned long ino_next;
2013 struct ext3_iloc iloc;
2014 int err = 0;
2015
2016 mutex_lock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2017 if (list_empty(&ei->i_orphan))
2018 goto out;
2019
2020 ino_next = NEXT_ORPHAN(inode);
2021 prev = ei->i_orphan.prev;
2022 sbi = EXT3_SB(inode->i_sb);
2023
2024 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2025
2026 list_del_init(&ei->i_orphan);
2027
2028 /* If we're on an error path, we may not have a valid
2029 * transaction handle with which to update the orphan list on
2030 * disk, but we still need to remove the inode from the linked
2031 * list in memory. */
2032 if (!handle)
2033 goto out;
2034
2035 err = ext3_reserve_inode_write(handle, inode, &iloc);
2036 if (err)
2037 goto out_err;
2038
2039 if (prev == &sbi->s_orphan) {
2040 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2041 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2042 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2043 if (err)
2044 goto out_brelse;
2045 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2046 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2047 } else {
2048 struct ext3_iloc iloc2;
2049 struct inode *i_prev =
2050 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2051
2052 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2053 i_prev->i_ino, ino_next);
2054 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2055 if (err)
2056 goto out_brelse;
2057 NEXT_ORPHAN(i_prev) = ino_next;
2058 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2059 }
2060 if (err)
2061 goto out_brelse;
2062 NEXT_ORPHAN(inode) = 0;
2063 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2064
2065out_err:
2066 ext3_std_error(inode->i_sb, err);
2067out:
2068 mutex_unlock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
2069 return err;
2070
2071out_brelse:
2072 brelse(iloc.bh);
2073 goto out_err;
2074}
2075
2076static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2077{
2078 int retval;
2079 struct inode * inode;
2080 struct buffer_head * bh;
2081 struct ext3_dir_entry_2 * de;
2082 handle_t *handle;
2083
2084 /* Initialize quotas before so that eventual writes go in
2085 * separate transaction */
2086 dquot_initialize(dir);
2087 dquot_initialize(dentry->d_inode);
2088
2089 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2090 if (IS_ERR(handle))
2091 return PTR_ERR(handle);
2092
2093 retval = -ENOENT;
2094 bh = ext3_find_entry(dir, &dentry->d_name, &de);
2095 if (!bh)
2096 goto end_rmdir;
2097
2098 if (IS_DIRSYNC(dir))
2099 handle->h_sync = 1;
2100
2101 inode = dentry->d_inode;
2102
2103 retval = -EIO;
2104 if (le32_to_cpu(de->inode) != inode->i_ino)
2105 goto end_rmdir;
2106
2107 retval = -ENOTEMPTY;
2108 if (!empty_dir (inode))
2109 goto end_rmdir;
2110
2111 retval = ext3_delete_entry(handle, dir, de, bh);
2112 if (retval)
2113 goto end_rmdir;
2114 if (inode->i_nlink != 2)
2115 ext3_warning (inode->i_sb, "ext3_rmdir",
2116 "empty directory has nlink!=2 (%d)",
2117 inode->i_nlink);
2118 inode->i_version++;
2119 clear_nlink(inode);
2120 /* There's no need to set i_disksize: the fact that i_nlink is
2121 * zero will ensure that the right thing happens during any
2122 * recovery. */
2123 inode->i_size = 0;
2124 ext3_orphan_add(handle, inode);
2125 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2126 ext3_mark_inode_dirty(handle, inode);
2127 drop_nlink(dir);
2128 ext3_update_dx_flag(dir);
2129 ext3_mark_inode_dirty(handle, dir);
2130
2131end_rmdir:
2132 ext3_journal_stop(handle);
2133 brelse (bh);
2134 return retval;
2135}
2136
2137static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2138{
2139 int retval;
2140 struct inode * inode;
2141 struct buffer_head * bh;
2142 struct ext3_dir_entry_2 * de;
2143 handle_t *handle;
2144
2145 trace_ext3_unlink_enter(dir, dentry);
2146 /* Initialize quotas before so that eventual writes go
2147 * in separate transaction */
2148 dquot_initialize(dir);
2149 dquot_initialize(dentry->d_inode);
2150
2151 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
2152 if (IS_ERR(handle))
2153 return PTR_ERR(handle);
2154
2155 if (IS_DIRSYNC(dir))
2156 handle->h_sync = 1;
2157
2158 retval = -ENOENT;
2159 bh = ext3_find_entry(dir, &dentry->d_name, &de);
2160 if (!bh)
2161 goto end_unlink;
2162
2163 inode = dentry->d_inode;
2164
2165 retval = -EIO;
2166 if (le32_to_cpu(de->inode) != inode->i_ino)
2167 goto end_unlink;
2168
2169 if (!inode->i_nlink) {
2170 ext3_warning (inode->i_sb, "ext3_unlink",
2171 "Deleting nonexistent file (%lu), %d",
2172 inode->i_ino, inode->i_nlink);
2173 inode->i_nlink = 1;
2174 }
2175 retval = ext3_delete_entry(handle, dir, de, bh);
2176 if (retval)
2177 goto end_unlink;
2178 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2179 ext3_update_dx_flag(dir);
2180 ext3_mark_inode_dirty(handle, dir);
2181 drop_nlink(inode);
2182 if (!inode->i_nlink)
2183 ext3_orphan_add(handle, inode);
2184 inode->i_ctime = dir->i_ctime;
2185 ext3_mark_inode_dirty(handle, inode);
2186 retval = 0;
2187
2188end_unlink:
2189 ext3_journal_stop(handle);
2190 brelse (bh);
2191 trace_ext3_unlink_exit(dentry, retval);
2192 return retval;
2193}
2194
2195static int ext3_symlink (struct inode * dir,
2196 struct dentry *dentry, const char * symname)
2197{
2198 handle_t *handle;
2199 struct inode * inode;
2200 int l, err, retries = 0;
2201 int credits;
2202
2203 l = strlen(symname)+1;
2204 if (l > dir->i_sb->s_blocksize)
2205 return -ENAMETOOLONG;
2206
2207 dquot_initialize(dir);
2208
2209 if (l > EXT3_N_BLOCKS * 4) {
2210 /*
2211 * For non-fast symlinks, we just allocate inode and put it on
2212 * orphan list in the first transaction => we need bitmap,
2213 * group descriptor, sb, inode block, quota blocks, and
2214 * possibly selinux xattr blocks.
2215 */
2216 credits = 4 + EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2217 EXT3_XATTR_TRANS_BLOCKS;
2218 } else {
2219 /*
2220 * Fast symlink. We have to add entry to directory
2221 * (EXT3_DATA_TRANS_BLOCKS + EXT3_INDEX_EXTRA_TRANS_BLOCKS),
2222 * allocate new inode (bitmap, group descriptor, inode block,
2223 * quota blocks, sb is already counted in previous macros).
2224 */
2225 credits = EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2226 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2227 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2228 }
2229retry:
2230 handle = ext3_journal_start(dir, credits);
2231 if (IS_ERR(handle))
2232 return PTR_ERR(handle);
2233
2234 if (IS_DIRSYNC(dir))
2235 handle->h_sync = 1;
2236
2237 inode = ext3_new_inode (handle, dir, &dentry->d_name, S_IFLNK|S_IRWXUGO);
2238 err = PTR_ERR(inode);
2239 if (IS_ERR(inode))
2240 goto out_stop;
2241
2242 if (l > EXT3_N_BLOCKS * 4) {
2243 inode->i_op = &ext3_symlink_inode_operations;
2244 ext3_set_aops(inode);
2245 /*
2246 * We cannot call page_symlink() with transaction started
2247 * because it calls into ext3_write_begin() which acquires page
2248 * lock which ranks below transaction start (and it can also
2249 * wait for journal commit if we are running out of space). So
2250 * we have to stop transaction now and restart it when symlink
2251 * contents is written.
2252 *
2253 * To keep fs consistent in case of crash, we have to put inode
2254 * to orphan list in the mean time.
2255 */
2256 drop_nlink(inode);
2257 err = ext3_orphan_add(handle, inode);
2258 ext3_journal_stop(handle);
2259 if (err)
2260 goto err_drop_inode;
2261 err = __page_symlink(inode, symname, l, 1);
2262 if (err)
2263 goto err_drop_inode;
2264 /*
2265 * Now inode is being linked into dir (EXT3_DATA_TRANS_BLOCKS
2266 * + EXT3_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2267 */
2268 handle = ext3_journal_start(dir,
2269 EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2270 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 1);
2271 if (IS_ERR(handle)) {
2272 err = PTR_ERR(handle);
2273 goto err_drop_inode;
2274 }
2275 inc_nlink(inode);
2276 err = ext3_orphan_del(handle, inode);
2277 if (err) {
2278 ext3_journal_stop(handle);
2279 drop_nlink(inode);
2280 goto err_drop_inode;
2281 }
2282 } else {
2283 inode->i_op = &ext3_fast_symlink_inode_operations;
2284 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2285 inode->i_size = l-1;
2286 }
2287 EXT3_I(inode)->i_disksize = inode->i_size;
2288 err = ext3_add_nondir(handle, dentry, inode);
2289out_stop:
2290 ext3_journal_stop(handle);
2291 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2292 goto retry;
2293 return err;
2294err_drop_inode:
2295 unlock_new_inode(inode);
2296 iput(inode);
2297 return err;
2298}
2299
2300static int ext3_link (struct dentry * old_dentry,
2301 struct inode * dir, struct dentry *dentry)
2302{
2303 handle_t *handle;
2304 struct inode *inode = old_dentry->d_inode;
2305 int err, retries = 0;
2306
2307 if (inode->i_nlink >= EXT3_LINK_MAX)
2308 return -EMLINK;
2309
2310 dquot_initialize(dir);
2311
2312retry:
2313 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
2314 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2315 if (IS_ERR(handle))
2316 return PTR_ERR(handle);
2317
2318 if (IS_DIRSYNC(dir))
2319 handle->h_sync = 1;
2320
2321 inode->i_ctime = CURRENT_TIME_SEC;
2322 inc_nlink(inode);
2323 ihold(inode);
2324
2325 err = ext3_add_entry(handle, dentry, inode);
2326 if (!err) {
2327 ext3_mark_inode_dirty(handle, inode);
2328 d_instantiate(dentry, inode);
2329 } else {
2330 drop_nlink(inode);
2331 iput(inode);
2332 }
2333 ext3_journal_stop(handle);
2334 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2335 goto retry;
2336 return err;
2337}
2338
2339#define PARENT_INO(buffer) \
2340 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
2341
2342/*
2343 * Anybody can rename anything with this: the permission checks are left to the
2344 * higher-level routines.
2345 */
2346static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2347 struct inode * new_dir,struct dentry *new_dentry)
2348{
2349 handle_t *handle;
2350 struct inode * old_inode, * new_inode;
2351 struct buffer_head * old_bh, * new_bh, * dir_bh;
2352 struct ext3_dir_entry_2 * old_de, * new_de;
2353 int retval, flush_file = 0;
2354
2355 dquot_initialize(old_dir);
2356 dquot_initialize(new_dir);
2357
2358 old_bh = new_bh = dir_bh = NULL;
2359
2360 /* Initialize quotas before so that eventual writes go
2361 * in separate transaction */
2362 if (new_dentry->d_inode)
2363 dquot_initialize(new_dentry->d_inode);
2364 handle = ext3_journal_start(old_dir, 2 *
2365 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2366 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
2367 if (IS_ERR(handle))
2368 return PTR_ERR(handle);
2369
2370 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2371 handle->h_sync = 1;
2372
2373 old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
2374 /*
2375 * Check for inode number is _not_ due to possible IO errors.
2376 * We might rmdir the source, keep it as pwd of some process
2377 * and merrily kill the link to whatever was created under the
2378 * same name. Goodbye sticky bit ;-<
2379 */
2380 old_inode = old_dentry->d_inode;
2381 retval = -ENOENT;
2382 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2383 goto end_rename;
2384
2385 new_inode = new_dentry->d_inode;
2386 new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
2387 if (new_bh) {
2388 if (!new_inode) {
2389 brelse (new_bh);
2390 new_bh = NULL;
2391 }
2392 }
2393 if (S_ISDIR(old_inode->i_mode)) {
2394 if (new_inode) {
2395 retval = -ENOTEMPTY;
2396 if (!empty_dir (new_inode))
2397 goto end_rename;
2398 }
2399 retval = -EIO;
2400 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2401 if (!dir_bh)
2402 goto end_rename;
2403 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2404 goto end_rename;
2405 retval = -EMLINK;
2406 if (!new_inode && new_dir!=old_dir &&
2407 new_dir->i_nlink >= EXT3_LINK_MAX)
2408 goto end_rename;
2409 }
2410 if (!new_bh) {
2411 retval = ext3_add_entry (handle, new_dentry, old_inode);
2412 if (retval)
2413 goto end_rename;
2414 } else {
2415 BUFFER_TRACE(new_bh, "get write access");
2416 retval = ext3_journal_get_write_access(handle, new_bh);
2417 if (retval)
2418 goto journal_error;
2419 new_de->inode = cpu_to_le32(old_inode->i_ino);
2420 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2421 EXT3_FEATURE_INCOMPAT_FILETYPE))
2422 new_de->file_type = old_de->file_type;
2423 new_dir->i_version++;
2424 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2425 ext3_mark_inode_dirty(handle, new_dir);
2426 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2427 retval = ext3_journal_dirty_metadata(handle, new_bh);
2428 if (retval)
2429 goto journal_error;
2430 brelse(new_bh);
2431 new_bh = NULL;
2432 }
2433
2434 /*
2435 * Like most other Unix systems, set the ctime for inodes on a
2436 * rename.
2437 */
2438 old_inode->i_ctime = CURRENT_TIME_SEC;
2439 ext3_mark_inode_dirty(handle, old_inode);
2440
2441 /*
2442 * ok, that's it
2443 */
2444 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2445 old_de->name_len != old_dentry->d_name.len ||
2446 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2447 (retval = ext3_delete_entry(handle, old_dir,
2448 old_de, old_bh)) == -ENOENT) {
2449 /* old_de could have moved from under us during htree split, so
2450 * make sure that we are deleting the right entry. We might
2451 * also be pointing to a stale entry in the unused part of
2452 * old_bh so just checking inum and the name isn't enough. */
2453 struct buffer_head *old_bh2;
2454 struct ext3_dir_entry_2 *old_de2;
2455
2456 old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2457 &old_de2);
2458 if (old_bh2) {
2459 retval = ext3_delete_entry(handle, old_dir,
2460 old_de2, old_bh2);
2461 brelse(old_bh2);
2462 }
2463 }
2464 if (retval) {
2465 ext3_warning(old_dir->i_sb, "ext3_rename",
2466 "Deleting old file (%lu), %d, error=%d",
2467 old_dir->i_ino, old_dir->i_nlink, retval);
2468 }
2469
2470 if (new_inode) {
2471 drop_nlink(new_inode);
2472 new_inode->i_ctime = CURRENT_TIME_SEC;
2473 }
2474 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2475 ext3_update_dx_flag(old_dir);
2476 if (dir_bh) {
2477 BUFFER_TRACE(dir_bh, "get_write_access");
2478 retval = ext3_journal_get_write_access(handle, dir_bh);
2479 if (retval)
2480 goto journal_error;
2481 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2482 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2483 retval = ext3_journal_dirty_metadata(handle, dir_bh);
2484 if (retval) {
2485journal_error:
2486 ext3_std_error(new_dir->i_sb, retval);
2487 goto end_rename;
2488 }
2489 drop_nlink(old_dir);
2490 if (new_inode) {
2491 drop_nlink(new_inode);
2492 } else {
2493 inc_nlink(new_dir);
2494 ext3_update_dx_flag(new_dir);
2495 ext3_mark_inode_dirty(handle, new_dir);
2496 }
2497 }
2498 ext3_mark_inode_dirty(handle, old_dir);
2499 if (new_inode) {
2500 ext3_mark_inode_dirty(handle, new_inode);
2501 if (!new_inode->i_nlink)
2502 ext3_orphan_add(handle, new_inode);
2503 if (ext3_should_writeback_data(new_inode))
2504 flush_file = 1;
2505 }
2506 retval = 0;
2507
2508end_rename:
2509 brelse (dir_bh);
2510 brelse (old_bh);
2511 brelse (new_bh);
2512 ext3_journal_stop(handle);
2513 if (retval == 0 && flush_file)
2514 filemap_flush(old_inode->i_mapping);
2515 return retval;
2516}
2517
2518/*
2519 * directories can handle most operations...
2520 */
2521const struct inode_operations ext3_dir_inode_operations = {
2522 .create = ext3_create,
2523 .lookup = ext3_lookup,
2524 .link = ext3_link,
2525 .unlink = ext3_unlink,
2526 .symlink = ext3_symlink,
2527 .mkdir = ext3_mkdir,
2528 .rmdir = ext3_rmdir,
2529 .mknod = ext3_mknod,
2530 .rename = ext3_rename,
2531 .setattr = ext3_setattr,
2532#ifdef CONFIG_EXT3_FS_XATTR
2533 .setxattr = generic_setxattr,
2534 .getxattr = generic_getxattr,
2535 .listxattr = ext3_listxattr,
2536 .removexattr = generic_removexattr,
2537#endif
2538 .get_acl = ext3_get_acl,
2539};
2540
2541const struct inode_operations ext3_special_inode_operations = {
2542 .setattr = ext3_setattr,
2543#ifdef CONFIG_EXT3_FS_XATTR
2544 .setxattr = generic_setxattr,
2545 .getxattr = generic_getxattr,
2546 .listxattr = ext3_listxattr,
2547 .removexattr = generic_removexattr,
2548#endif
2549 .get_acl = ext3_get_acl,
2550};