blob: 1586807b81779ef9c02008e6828c9df50d4f000c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Dave Kleikampe9ad5622006-09-27 01:49:35 -070018 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 * Hash Tree Directory indexing (c)
Dave Kleikampe9ad5622006-09-27 01:49:35 -070020 * Daniel Phillips, 2001
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * Hash Tree Directory indexing porting
Dave Kleikampe9ad5622006-09-27 01:49:35 -070022 * Christopher Li, 2002
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 * Hash Tree Directory indexing cleanup
Dave Kleikampe9ad5622006-09-27 01:49:35 -070024 * Theodore Ts'o, 2002
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 */
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>
Jens Axboecaa38fb2006-07-23 01:41:26 +020038#include <linux/bio.h>
Ben Dooks381be252005-10-30 15:02:56 -080039
40#include "namei.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xattr.h"
42#include "acl.h"
43
44/*
45 * define how far ahead to read directories while searching them.
46 */
47#define NAMEI_RA_CHUNKS 2
48#define NAMEI_RA_BLOCKS 4
49#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51
52static struct buffer_head *ext3_append(handle_t *handle,
53 struct inode *inode,
54 u32 *block, int *err)
55{
56 struct buffer_head *bh;
57
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60 if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
61 inode->i_size += inode->i_sb->s_blocksize;
62 EXT3_I(inode)->i_disksize = inode->i_size;
63 ext3_journal_get_write_access(handle,bh);
64 }
65 return bh;
66}
67
68#ifndef assert
69#define assert(test) J_ASSERT(test)
70#endif
71
72#ifndef swap
73#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74#endif
75
76#ifdef DX_DEBUG
77#define dxtrace(command) command
78#else
Mingming Caoae6ddcc2006-09-27 01:49:27 -070079#define dxtrace(command)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#endif
81
82struct fake_dirent
83{
84 __le32 inode;
85 __le16 rec_len;
86 u8 name_len;
87 u8 file_type;
88};
89
90struct dx_countlimit
91{
92 __le16 limit;
93 __le16 count;
94};
95
96struct dx_entry
97{
98 __le32 hash;
99 __le32 block;
100};
101
102/*
103 * dx_root_info is laid out so that if it should somehow get overlaid by a
104 * dirent the two low bits of the hash version will be zero. Therefore, the
105 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
106 */
107
108struct dx_root
109{
110 struct fake_dirent dot;
111 char dot_name[4];
112 struct fake_dirent dotdot;
113 char dotdot_name[4];
114 struct dx_root_info
115 {
116 __le32 reserved_zero;
117 u8 hash_version;
118 u8 info_length; /* 8 */
119 u8 indirect_levels;
120 u8 unused_flags;
121 }
122 info;
123 struct dx_entry entries[0];
124};
125
126struct dx_node
127{
128 struct fake_dirent fake;
129 struct dx_entry entries[0];
130};
131
132
133struct dx_frame
134{
135 struct buffer_head *bh;
136 struct dx_entry *entries;
137 struct dx_entry *at;
138};
139
140struct dx_map_entry
141{
142 u32 hash;
143 u32 offs;
144};
145
146#ifdef CONFIG_EXT3_INDEX
147static inline unsigned dx_get_block (struct dx_entry *entry);
148static void dx_set_block (struct dx_entry *entry, unsigned value);
149static inline unsigned dx_get_hash (struct dx_entry *entry);
150static void dx_set_hash (struct dx_entry *entry, unsigned value);
151static unsigned dx_get_count (struct dx_entry *entries);
152static unsigned dx_get_limit (struct dx_entry *entries);
153static void dx_set_count (struct dx_entry *entries, unsigned value);
154static void dx_set_limit (struct dx_entry *entries, unsigned value);
155static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
156static unsigned dx_node_limit (struct inode *dir);
157static struct dx_frame *dx_probe(struct dentry *dentry,
158 struct inode *dir,
159 struct dx_hash_info *hinfo,
160 struct dx_frame *frame,
161 int *err);
162static void dx_release (struct dx_frame *frames);
163static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
164 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
167 struct dx_map_entry *offsets, int count);
168static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
169static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
170static int ext3_htree_next_block(struct inode *dir, __u32 hash,
171 struct dx_frame *frame,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700172 struct dx_frame *frames,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 __u32 *start_hash);
174static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
175 struct ext3_dir_entry_2 **res_dir, int *err);
176static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
177 struct inode *inode);
178
179/*
180 * Future: use high four bits of block for coalesce-on-delete flags
181 * Mask them off for now.
182 */
183
184static inline unsigned dx_get_block (struct dx_entry *entry)
185{
186 return le32_to_cpu(entry->block) & 0x00ffffff;
187}
188
189static inline void dx_set_block (struct dx_entry *entry, unsigned value)
190{
191 entry->block = cpu_to_le32(value);
192}
193
194static inline unsigned dx_get_hash (struct dx_entry *entry)
195{
196 return le32_to_cpu(entry->hash);
197}
198
199static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
200{
201 entry->hash = cpu_to_le32(value);
202}
203
204static inline unsigned dx_get_count (struct dx_entry *entries)
205{
206 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
207}
208
209static inline unsigned dx_get_limit (struct dx_entry *entries)
210{
211 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
212}
213
214static inline void dx_set_count (struct dx_entry *entries, unsigned value)
215{
216 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
217}
218
219static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
220{
221 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
222}
223
224static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
225{
226 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
227 EXT3_DIR_REC_LEN(2) - infosize;
228 return 0? 20: entry_space / sizeof(struct dx_entry);
229}
230
231static inline unsigned dx_node_limit (struct inode *dir)
232{
233 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
234 return 0? 22: entry_space / sizeof(struct dx_entry);
235}
236
237/*
238 * Debug
239 */
240#ifdef DX_DEBUG
241static void dx_show_index (char * label, struct dx_entry *entries)
242{
243 int i, n = dx_get_count (entries);
244 printk("%s index ", label);
245 for (i = 0; i < n; i++)
246 {
247 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
248 }
249 printk("\n");
250}
251
252struct stats
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700253{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 unsigned names;
255 unsigned space;
256 unsigned bcount;
257};
258
259static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
260 int size, int show_names)
261{
262 unsigned names = 0, space = 0;
263 char *base = (char *) de;
264 struct dx_hash_info h = *hinfo;
265
266 printk("names: ");
267 while ((char *) de < base + size)
268 {
269 if (de->inode)
270 {
271 if (show_names)
272 {
273 int len = de->name_len;
274 char *name = de->name;
275 while (len--) printk("%c", *name++);
276 ext3fs_dirhash(de->name, de->name_len, &h);
277 printk(":%x.%u ", h.hash,
278 ((char *) de - base));
279 }
280 space += EXT3_DIR_REC_LEN(de->name_len);
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700281 names++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 }
283 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
284 }
285 printk("(%i)\n", names);
286 return (struct stats) { names, space, 1 };
287}
288
289struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
290 struct dx_entry *entries, int levels)
291{
292 unsigned blocksize = dir->i_sb->s_blocksize;
293 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
294 unsigned bcount = 0;
295 struct buffer_head *bh;
296 int err;
297 printk("%i indexed blocks...\n", count);
298 for (i = 0; i < count; i++, entries++)
299 {
300 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
301 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
302 struct stats stats;
303 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
304 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
305 stats = levels?
306 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
307 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
308 names += stats.names;
309 space += stats.space;
310 bcount += stats.bcount;
311 brelse (bh);
312 }
313 if (bcount)
314 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
315 names, space/bcount,(space/bcount)*100/blocksize);
316 return (struct stats) { names, space, bcount};
317}
318#endif /* DX_DEBUG */
319
320/*
321 * Probe for a directory leaf block to search.
322 *
323 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
324 * error in the directory index, and the caller should fall back to
325 * searching the directory normally. The callers of dx_probe **MUST**
326 * check for this error code, and make sure it never gets reflected
327 * back to userspace.
328 */
329static struct dx_frame *
330dx_probe(struct dentry *dentry, struct inode *dir,
331 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
332{
333 unsigned count, indirect;
334 struct dx_entry *at, *entries, *p, *q, *m;
335 struct dx_root *root;
336 struct buffer_head *bh;
337 struct dx_frame *frame = frame_in;
338 u32 hash;
339
340 frame->bh = NULL;
341 if (dentry)
342 dir = dentry->d_parent->d_inode;
343 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
344 goto fail;
345 root = (struct dx_root *) bh->b_data;
346 if (root->info.hash_version != DX_HASH_TEA &&
347 root->info.hash_version != DX_HASH_HALF_MD4 &&
348 root->info.hash_version != DX_HASH_LEGACY) {
349 ext3_warning(dir->i_sb, __FUNCTION__,
350 "Unrecognised inode hash code %d",
351 root->info.hash_version);
352 brelse(bh);
353 *err = ERR_BAD_DX_DIR;
354 goto fail;
355 }
356 hinfo->hash_version = root->info.hash_version;
357 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
358 if (dentry)
359 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
360 hash = hinfo->hash;
361
362 if (root->info.unused_flags & 1) {
363 ext3_warning(dir->i_sb, __FUNCTION__,
364 "Unimplemented inode hash flags: %#06x",
365 root->info.unused_flags);
366 brelse(bh);
367 *err = ERR_BAD_DX_DIR;
368 goto fail;
369 }
370
371 if ((indirect = root->info.indirect_levels) > 1) {
372 ext3_warning(dir->i_sb, __FUNCTION__,
373 "Unimplemented inode hash depth: %#06x",
374 root->info.indirect_levels);
375 brelse(bh);
376 *err = ERR_BAD_DX_DIR;
377 goto fail;
378 }
379
380 entries = (struct dx_entry *) (((char *)&root->info) +
381 root->info.info_length);
382 assert(dx_get_limit(entries) == dx_root_limit(dir,
383 root->info.info_length));
384 dxtrace (printk("Look up %x", hash));
385 while (1)
386 {
387 count = dx_get_count(entries);
388 assert (count && count <= dx_get_limit(entries));
389 p = entries + 1;
390 q = entries + count - 1;
391 while (p <= q)
392 {
393 m = p + (q - p)/2;
394 dxtrace(printk("."));
395 if (dx_get_hash(m) > hash)
396 q = m - 1;
397 else
398 p = m + 1;
399 }
400
401 if (0) // linear search cross check
402 {
403 unsigned n = count - 1;
404 at = entries;
405 while (n--)
406 {
407 dxtrace(printk(","));
408 if (dx_get_hash(++at) > hash)
409 {
410 at--;
411 break;
412 }
413 }
414 assert (at == p - 1);
415 }
416
417 at = p - 1;
418 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
419 frame->bh = bh;
420 frame->entries = entries;
421 frame->at = at;
422 if (!indirect--) return frame;
423 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
424 goto fail2;
425 at = entries = ((struct dx_node *) bh->b_data)->entries;
426 assert (dx_get_limit(entries) == dx_node_limit (dir));
427 frame++;
428 }
429fail2:
430 while (frame >= frame_in) {
431 brelse(frame->bh);
432 frame--;
433 }
434fail:
435 return NULL;
436}
437
438static void dx_release (struct dx_frame *frames)
439{
440 if (frames[0].bh == NULL)
441 return;
442
443 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
444 brelse(frames[1].bh);
445 brelse(frames[0].bh);
446}
447
448/*
449 * This function increments the frame pointer to search the next leaf
450 * block, and reads in the necessary intervening nodes if the search
451 * should be necessary. Whether or not the search is necessary is
452 * controlled by the hash parameter. If the hash value is even, then
453 * the search is only continued if the next block starts with that
454 * hash value. This is used if we are searching for a specific file.
455 *
456 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
457 *
458 * This function returns 1 if the caller should continue to search,
459 * or 0 if it should not. If there is an error reading one of the
460 * index blocks, it will a negative error code.
461 *
462 * If start_hash is non-null, it will be filled in with the starting
463 * hash of the next page.
464 */
465static int ext3_htree_next_block(struct inode *dir, __u32 hash,
466 struct dx_frame *frame,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700467 struct dx_frame *frames,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 __u32 *start_hash)
469{
470 struct dx_frame *p;
471 struct buffer_head *bh;
472 int err, num_frames = 0;
473 __u32 bhash;
474
475 p = frame;
476 /*
477 * Find the next leaf page by incrementing the frame pointer.
478 * If we run out of entries in the interior node, loop around and
479 * increment pointer in the parent node. When we break out of
480 * this loop, num_frames indicates the number of interior
481 * nodes need to be read.
482 */
483 while (1) {
484 if (++(p->at) < p->entries + dx_get_count(p->entries))
485 break;
486 if (p == frames)
487 return 0;
488 num_frames++;
489 p--;
490 }
491
492 /*
493 * If the hash is 1, then continue only if the next page has a
494 * continuation hash of any value. This is used for readdir
495 * handling. Otherwise, check to see if the hash matches the
496 * desired contiuation hash. If it doesn't, return since
497 * there's no point to read in the successive index pages.
498 */
499 bhash = dx_get_hash(p->at);
500 if (start_hash)
501 *start_hash = bhash;
502 if ((hash & 1) == 0) {
503 if ((bhash & ~1) != hash)
504 return 0;
505 }
506 /*
507 * If the hash is HASH_NB_ALWAYS, we always go to the next
508 * block so no check is necessary
509 */
510 while (num_frames--) {
511 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
512 0, &err)))
513 return err; /* Failure */
514 p++;
515 brelse (p->bh);
516 p->bh = bh;
517 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
518 }
519 return 1;
520}
521
522
523/*
524 * p is at least 6 bytes before the end of page
525 */
526static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
527{
528 return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
529}
530
531/*
532 * This function fills a red-black tree with information from a
533 * directory block. It returns the number directory entries loaded
534 * into the tree. If there is an error it is returned in err.
535 */
536static int htree_dirblock_to_tree(struct file *dir_file,
537 struct inode *dir, int block,
538 struct dx_hash_info *hinfo,
539 __u32 start_hash, __u32 start_minor_hash)
540{
541 struct buffer_head *bh;
542 struct ext3_dir_entry_2 *de, *top;
543 int err, count = 0;
544
545 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
546 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
547 return err;
548
549 de = (struct ext3_dir_entry_2 *) bh->b_data;
550 top = (struct ext3_dir_entry_2 *) ((char *) de +
551 dir->i_sb->s_blocksize -
552 EXT3_DIR_REC_LEN(0));
553 for (; de < top; de = ext3_next_entry(de)) {
Eric Sandeen40b85132006-12-06 20:36:26 -0800554 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
555 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
556 +((char *)de - bh->b_data))) {
557 /* On error, skip the f_pos to the next block. */
558 dir_file->f_pos = (dir_file->f_pos |
559 (dir->i_sb->s_blocksize - 1)) + 1;
560 brelse (bh);
561 return count;
562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 ext3fs_dirhash(de->name, de->name_len, hinfo);
564 if ((hinfo->hash < start_hash) ||
565 ((hinfo->hash == start_hash) &&
566 (hinfo->minor_hash < start_minor_hash)))
567 continue;
568 if (de->inode == 0)
569 continue;
570 if ((err = ext3_htree_store_dirent(dir_file,
571 hinfo->hash, hinfo->minor_hash, de)) != 0) {
572 brelse(bh);
573 return err;
574 }
575 count++;
576 }
577 brelse(bh);
578 return count;
579}
580
581
582/*
583 * This function fills a red-black tree with information from a
584 * directory. We start scanning the directory in hash order, starting
585 * at start_hash and start_minor_hash.
586 *
587 * This function returns the number of entries inserted into the tree,
588 * or a negative error code.
589 */
590int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
591 __u32 start_minor_hash, __u32 *next_hash)
592{
593 struct dx_hash_info hinfo;
594 struct ext3_dir_entry_2 *de;
595 struct dx_frame frames[2], *frame;
596 struct inode *dir;
597 int block, err;
598 int count = 0;
599 int ret;
600 __u32 hashval;
601
602 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
603 start_minor_hash));
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800604 dir = dir_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
606 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
607 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
608 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
609 start_hash, start_minor_hash);
610 *next_hash = ~0;
611 return count;
612 }
613 hinfo.hash = start_hash;
614 hinfo.minor_hash = 0;
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800615 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (!frame)
617 return err;
618
619 /* Add '.' and '..' from the htree header */
620 if (!start_hash && !start_minor_hash) {
621 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
622 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
623 goto errout;
624 count++;
625 }
626 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
627 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
628 de = ext3_next_entry(de);
629 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
630 goto errout;
631 count++;
632 }
633
634 while (1) {
635 block = dx_get_block(frame->at);
636 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
637 start_hash, start_minor_hash);
638 if (ret < 0) {
639 err = ret;
640 goto errout;
641 }
642 count += ret;
643 hashval = ~0;
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700644 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 frame, frames, &hashval);
646 *next_hash = hashval;
647 if (ret < 0) {
648 err = ret;
649 goto errout;
650 }
651 /*
652 * Stop if: (a) there are no more entries, or
653 * (b) we have inserted at least one entry and the
654 * next hash value is not a continuation
655 */
656 if ((ret == 0) ||
657 (count && ((hashval & 1) == 0)))
658 break;
659 }
660 dx_release(frames);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700661 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 count, *next_hash));
663 return count;
664errout:
665 dx_release(frames);
666 return (err);
667}
668
669
670/*
671 * Directory block splitting, compacting
672 */
673
674static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
675 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
676{
677 int count = 0;
678 char *base = (char *) de;
679 struct dx_hash_info h = *hinfo;
680
681 while ((char *) de < base + size)
682 {
683 if (de->name_len && de->inode) {
684 ext3fs_dirhash(de->name, de->name_len, &h);
685 map_tail--;
686 map_tail->hash = h.hash;
687 map_tail->offs = (u32) ((char *) de - base);
688 count++;
689 cond_resched();
690 }
691 /* XXX: do we need to check rec_len == 0 case? -Chris */
692 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
693 }
694 return count;
695}
696
697static void dx_sort_map (struct dx_map_entry *map, unsigned count)
698{
699 struct dx_map_entry *p, *q, *top = map + count - 1;
700 int more;
701 /* Combsort until bubble sort doesn't suck */
702 while (count > 2)
703 {
704 count = count*10/13;
705 if (count - 9 < 2) /* 9, 10 -> 11 */
706 count = 11;
707 for (p = top, q = p - count; q >= map; p--, q--)
708 if (p->hash < q->hash)
709 swap(*p, *q);
710 }
711 /* Garden variety bubble sort */
712 do {
713 more = 0;
714 q = top;
715 while (q-- > map)
716 {
717 if (q[1].hash >= q[0].hash)
718 continue;
719 swap(*(q+1), *q);
720 more = 1;
721 }
722 } while(more);
723}
724
725static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
726{
727 struct dx_entry *entries = frame->entries;
728 struct dx_entry *old = frame->at, *new = old + 1;
729 int count = dx_get_count(entries);
730
731 assert(count < dx_get_limit(entries));
732 assert(old < entries + count);
733 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
734 dx_set_hash(new, hash);
735 dx_set_block(new, block);
736 dx_set_count(entries, count + 1);
737}
738#endif
739
740
741static void ext3_update_dx_flag(struct inode *inode)
742{
743 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
744 EXT3_FEATURE_COMPAT_DIR_INDEX))
745 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
746}
747
748/*
749 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
750 *
751 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
752 * `de != NULL' is guaranteed by caller.
753 */
754static inline int ext3_match (int len, const char * const name,
755 struct ext3_dir_entry_2 * de)
756{
757 if (len != de->name_len)
758 return 0;
759 if (!de->inode)
760 return 0;
761 return !memcmp(name, de->name, len);
762}
763
764/*
765 * Returns 0 if not found, -1 on failure, and 1 on success
766 */
767static inline int search_dirblock(struct buffer_head * bh,
768 struct inode *dir,
769 struct dentry *dentry,
770 unsigned long offset,
771 struct ext3_dir_entry_2 ** res_dir)
772{
773 struct ext3_dir_entry_2 * de;
774 char * dlimit;
775 int de_len;
776 const char *name = dentry->d_name.name;
777 int namelen = dentry->d_name.len;
778
779 de = (struct ext3_dir_entry_2 *) bh->b_data;
780 dlimit = bh->b_data + dir->i_sb->s_blocksize;
781 while ((char *) de < dlimit) {
782 /* this code is executed quadratically often */
783 /* do minimal checking `by hand' */
784
785 if ((char *) de + namelen <= dlimit &&
786 ext3_match (namelen, name, de)) {
787 /* found a match - just to be sure, do a full check */
788 if (!ext3_check_dir_entry("ext3_find_entry",
789 dir, de, bh, offset))
790 return -1;
791 *res_dir = de;
792 return 1;
793 }
794 /* prevent looping on a bad block */
795 de_len = le16_to_cpu(de->rec_len);
796 if (de_len <= 0)
797 return -1;
798 offset += de_len;
799 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
800 }
801 return 0;
802}
803
804
805/*
806 * ext3_find_entry()
807 *
808 * finds an entry in the specified directory with the wanted name. It
809 * returns the cache buffer in which the entry was found, and the entry
810 * itself (as a parameter - res_dir). It does NOT read the inode of the
811 * entry - you'll have to do that yourself if you want to.
812 *
813 * The returned buffer_head has ->b_count elevated. The caller is expected
814 * to brelse() it when appropriate.
815 */
816static struct buffer_head * ext3_find_entry (struct dentry *dentry,
817 struct ext3_dir_entry_2 ** res_dir)
818{
819 struct super_block * sb;
820 struct buffer_head * bh_use[NAMEI_RA_SIZE];
821 struct buffer_head * bh, *ret = NULL;
822 unsigned long start, block, b;
823 int ra_max = 0; /* Number of bh's in the readahead
824 buffer, bh_use[] */
825 int ra_ptr = 0; /* Current index into readahead
826 buffer */
827 int num = 0;
828 int nblocks, i, err;
829 struct inode *dir = dentry->d_parent->d_inode;
830 int namelen;
831 const u8 *name;
832 unsigned blocksize;
833
834 *res_dir = NULL;
835 sb = dir->i_sb;
836 blocksize = sb->s_blocksize;
837 namelen = dentry->d_name.len;
838 name = dentry->d_name.name;
839 if (namelen > EXT3_NAME_LEN)
840 return NULL;
841#ifdef CONFIG_EXT3_INDEX
842 if (is_dx(dir)) {
843 bh = ext3_dx_find_entry(dentry, res_dir, &err);
844 /*
845 * On success, or if the error was file not found,
846 * return. Otherwise, fall back to doing a search the
847 * old fashioned way.
848 */
849 if (bh || (err != ERR_BAD_DX_DIR))
850 return bh;
851 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
852 }
853#endif
854 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
855 start = EXT3_I(dir)->i_dir_start_lookup;
856 if (start >= nblocks)
857 start = 0;
858 block = start;
859restart:
860 do {
861 /*
862 * We deal with the read-ahead logic here.
863 */
864 if (ra_ptr >= ra_max) {
865 /* Refill the readahead buffer */
866 ra_ptr = 0;
867 b = block;
868 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
869 /*
870 * Terminate if we reach the end of the
871 * directory and must wrap, or if our
872 * search has finished at this block.
873 */
874 if (b >= nblocks || (num && block == start)) {
875 bh_use[ra_max] = NULL;
876 break;
877 }
878 num++;
879 bh = ext3_getblk(NULL, dir, b++, 0, &err);
880 bh_use[ra_max] = bh;
881 if (bh)
Jens Axboecaa38fb2006-07-23 01:41:26 +0200882 ll_rw_block(READ_META, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 }
884 }
885 if ((bh = bh_use[ra_ptr++]) == NULL)
886 goto next;
887 wait_on_buffer(bh);
888 if (!buffer_uptodate(bh)) {
889 /* read error, skip block & hope for the best */
890 ext3_error(sb, __FUNCTION__, "reading directory #%lu "
891 "offset %lu", dir->i_ino, block);
892 brelse(bh);
893 goto next;
894 }
895 i = search_dirblock(bh, dir, dentry,
896 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
897 if (i == 1) {
898 EXT3_I(dir)->i_dir_start_lookup = block;
899 ret = bh;
900 goto cleanup_and_exit;
901 } else {
902 brelse(bh);
903 if (i < 0)
904 goto cleanup_and_exit;
905 }
906 next:
907 if (++block >= nblocks)
908 block = 0;
909 } while (block != start);
910
911 /*
912 * If the directory has grown while we were searching, then
913 * search the last part of the directory before giving up.
914 */
915 block = nblocks;
916 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
917 if (block < nblocks) {
918 start = 0;
919 goto restart;
920 }
921
922cleanup_and_exit:
923 /* Clean up the read-ahead blocks */
924 for (; ra_ptr < ra_max; ra_ptr++)
925 brelse (bh_use[ra_ptr]);
926 return ret;
927}
928
929#ifdef CONFIG_EXT3_INDEX
930static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
931 struct ext3_dir_entry_2 **res_dir, int *err)
932{
933 struct super_block * sb;
934 struct dx_hash_info hinfo;
935 u32 hash;
936 struct dx_frame frames[2], *frame;
937 struct ext3_dir_entry_2 *de, *top;
938 struct buffer_head *bh;
939 unsigned long block;
940 int retval;
941 int namelen = dentry->d_name.len;
942 const u8 *name = dentry->d_name.name;
943 struct inode *dir = dentry->d_parent->d_inode;
944
945 sb = dir->i_sb;
Andreas Dilgeracfa1823d2005-06-23 00:09:45 -0700946 /* NFS may look up ".." - look at dx_root directory block */
947 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
948 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
949 return NULL;
950 } else {
951 frame = frames;
952 frame->bh = NULL; /* for dx_release() */
953 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
954 dx_set_block(frame->at, 0); /* dx_root block is 0 */
955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 hash = hinfo.hash;
957 do {
958 block = dx_get_block(frame->at);
959 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
960 goto errout;
961 de = (struct ext3_dir_entry_2 *) bh->b_data;
962 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
963 EXT3_DIR_REC_LEN(0));
964 for (; de < top; de = ext3_next_entry(de))
965 if (ext3_match (namelen, name, de)) {
966 if (!ext3_check_dir_entry("ext3_find_entry",
967 dir, de, bh,
968 (block<<EXT3_BLOCK_SIZE_BITS(sb))
969 +((char *)de - bh->b_data))) {
970 brelse (bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -0700971 *err = ERR_BAD_DX_DIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 goto errout;
973 }
974 *res_dir = de;
975 dx_release (frames);
976 return bh;
977 }
978 brelse (bh);
979 /* Check to see if we should continue to search */
980 retval = ext3_htree_next_block(dir, hash, frame,
981 frames, NULL);
982 if (retval < 0) {
983 ext3_warning(sb, __FUNCTION__,
984 "error reading index page in directory #%lu",
985 dir->i_ino);
986 *err = retval;
987 goto errout;
988 }
989 } while (retval == 1);
990
991 *err = -ENOENT;
992errout:
993 dxtrace(printk("%s not found\n", name));
994 dx_release (frames);
995 return NULL;
996}
997#endif
998
999static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1000{
1001 struct inode * inode;
1002 struct ext3_dir_entry_2 * de;
1003 struct buffer_head * bh;
1004
1005 if (dentry->d_name.len > EXT3_NAME_LEN)
1006 return ERR_PTR(-ENAMETOOLONG);
1007
1008 bh = ext3_find_entry(dentry, &de);
1009 inode = NULL;
1010 if (bh) {
1011 unsigned long ino = le32_to_cpu(de->inode);
1012 brelse (bh);
Neil Brown2ccb48e2006-07-30 03:03:01 -07001013 if (!ext3_valid_inum(dir->i_sb, ino)) {
1014 ext3_error(dir->i_sb, "ext3_lookup",
1015 "bad inode number: %lu", ino);
1016 inode = NULL;
1017 } else
1018 inode = iget(dir->i_sb, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 if (!inode)
1021 return ERR_PTR(-EACCES);
Vasily Averina6c15c22007-07-15 23:40:46 -07001022
1023 if (is_bad_inode(inode)) {
1024 iput(inode);
1025 return ERR_PTR(-ENOENT);
1026 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 }
Pekka Enbergba7fe362006-01-14 13:21:08 -08001028 return d_splice_alias(inode, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
1031
1032struct dentry *ext3_get_parent(struct dentry *child)
1033{
1034 unsigned long ino;
1035 struct dentry *parent;
1036 struct inode *inode;
1037 struct dentry dotdot;
1038 struct ext3_dir_entry_2 * de;
1039 struct buffer_head *bh;
1040
1041 dotdot.d_name.name = "..";
1042 dotdot.d_name.len = 2;
1043 dotdot.d_parent = child; /* confusing, isn't it! */
1044
1045 bh = ext3_find_entry(&dotdot, &de);
1046 inode = NULL;
1047 if (!bh)
1048 return ERR_PTR(-ENOENT);
1049 ino = le32_to_cpu(de->inode);
1050 brelse(bh);
Neil Brown2ccb48e2006-07-30 03:03:01 -07001051
1052 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1053 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1054 "bad inode number: %lu", ino);
1055 inode = NULL;
1056 } else
1057 inode = iget(child->d_inode->i_sb, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
1059 if (!inode)
1060 return ERR_PTR(-EACCES);
1061
Vasily Averina6c15c22007-07-15 23:40:46 -07001062 if (is_bad_inode(inode)) {
1063 iput(inode);
1064 return ERR_PTR(-ENOENT);
1065 }
1066
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 parent = d_alloc_anon(inode);
1068 if (!parent) {
1069 iput(inode);
1070 parent = ERR_PTR(-ENOMEM);
1071 }
1072 return parent;
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001073}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
1075#define S_SHIFT 12
1076static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1077 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1078 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1079 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1080 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1081 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1082 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1083 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1084};
1085
1086static inline void ext3_set_de_type(struct super_block *sb,
1087 struct ext3_dir_entry_2 *de,
1088 umode_t mode) {
1089 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1090 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1091}
1092
1093#ifdef CONFIG_EXT3_INDEX
1094static struct ext3_dir_entry_2 *
1095dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1096{
1097 unsigned rec_len = 0;
1098
1099 while (count--) {
1100 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1101 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1102 memcpy (to, de, rec_len);
1103 ((struct ext3_dir_entry_2 *) to)->rec_len =
1104 cpu_to_le16(rec_len);
1105 de->inode = 0;
1106 map++;
1107 to += rec_len;
1108 }
1109 return (struct ext3_dir_entry_2 *) (to - rec_len);
1110}
1111
1112static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1113{
1114 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1115 unsigned rec_len = 0;
1116
1117 prev = to = de;
1118 while ((char*)de < base + size) {
1119 next = (struct ext3_dir_entry_2 *) ((char *) de +
1120 le16_to_cpu(de->rec_len));
1121 if (de->inode && de->name_len) {
1122 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1123 if (de > to)
1124 memmove(to, de, rec_len);
1125 to->rec_len = cpu_to_le16(rec_len);
1126 prev = to;
1127 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1128 }
1129 de = next;
1130 }
1131 return prev;
1132}
1133
1134static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1135 struct buffer_head **bh,struct dx_frame *frame,
1136 struct dx_hash_info *hinfo, int *error)
1137{
1138 unsigned blocksize = dir->i_sb->s_blocksize;
1139 unsigned count, continued;
1140 struct buffer_head *bh2;
1141 u32 newblock;
1142 u32 hash2;
1143 struct dx_map_entry *map;
1144 char *data1 = (*bh)->b_data, *data2;
1145 unsigned split;
1146 struct ext3_dir_entry_2 *de = NULL, *de2;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001147 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001149 bh2 = ext3_append (handle, dir, &newblock, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (!(bh2)) {
1151 brelse(*bh);
1152 *bh = NULL;
1153 goto errout;
1154 }
1155
1156 BUFFER_TRACE(*bh, "get_write_access");
1157 err = ext3_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001158 if (err)
1159 goto journal_error;
1160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 BUFFER_TRACE(frame->bh, "get_write_access");
1162 err = ext3_journal_get_write_access(handle, frame->bh);
1163 if (err)
1164 goto journal_error;
1165
1166 data2 = bh2->b_data;
1167
1168 /* create map in the end of data2 block */
1169 map = (struct dx_map_entry *) (data2 + blocksize);
1170 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1171 blocksize, hinfo, map);
1172 map -= count;
1173 split = count/2; // need to adjust to actual middle
1174 dx_sort_map (map, count);
1175 hash2 = map[split].hash;
1176 continued = hash2 == map[split - 1].hash;
1177 dxtrace(printk("Split block %i at %x, %i/%i\n",
1178 dx_get_block(frame->at), hash2, split, count-split));
1179
1180 /* Fancy dance to stay within two buffers */
1181 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1182 de = dx_pack_dirents(data1,blocksize);
1183 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1184 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1185 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1186 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1187
1188 /* Which block gets the new entry? */
1189 if (hinfo->hash >= hash2)
1190 {
1191 swap(*bh, bh2);
1192 de = de2;
1193 }
1194 dx_insert_block (frame, hash2 + continued, newblock);
1195 err = ext3_journal_dirty_metadata (handle, bh2);
1196 if (err)
1197 goto journal_error;
1198 err = ext3_journal_dirty_metadata (handle, frame->bh);
1199 if (err)
1200 goto journal_error;
1201 brelse (bh2);
1202 dxtrace(dx_show_index ("frame", frame->entries));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001204
1205journal_error:
1206 brelse(*bh);
1207 brelse(bh2);
1208 *bh = NULL;
1209 ext3_std_error(dir->i_sb, err);
1210errout:
1211 *error = err;
1212 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213}
1214#endif
1215
1216
1217/*
1218 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1219 * it points to a directory entry which is guaranteed to be large
1220 * enough for new directory entry. If de is NULL, then
1221 * add_dirent_to_buf will attempt search the directory block for
1222 * space. It will return -ENOSPC if no space is available, and -EIO
1223 * and -EEXIST if directory entry already exists.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001224 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1226 * all other cases bh is released.
1227 */
1228static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1229 struct inode *inode, struct ext3_dir_entry_2 *de,
1230 struct buffer_head * bh)
1231{
1232 struct inode *dir = dentry->d_parent->d_inode;
1233 const char *name = dentry->d_name.name;
1234 int namelen = dentry->d_name.len;
1235 unsigned long offset = 0;
1236 unsigned short reclen;
1237 int nlen, rlen, err;
1238 char *top;
1239
1240 reclen = EXT3_DIR_REC_LEN(namelen);
1241 if (!de) {
1242 de = (struct ext3_dir_entry_2 *)bh->b_data;
1243 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1244 while ((char *) de <= top) {
1245 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1246 bh, offset)) {
1247 brelse (bh);
1248 return -EIO;
1249 }
1250 if (ext3_match (namelen, name, de)) {
1251 brelse (bh);
1252 return -EEXIST;
1253 }
1254 nlen = EXT3_DIR_REC_LEN(de->name_len);
1255 rlen = le16_to_cpu(de->rec_len);
1256 if ((de->inode? rlen - nlen: rlen) >= reclen)
1257 break;
1258 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1259 offset += rlen;
1260 }
1261 if ((char *) de > top)
1262 return -ENOSPC;
1263 }
1264 BUFFER_TRACE(bh, "get_write_access");
1265 err = ext3_journal_get_write_access(handle, bh);
1266 if (err) {
1267 ext3_std_error(dir->i_sb, err);
1268 brelse(bh);
1269 return err;
1270 }
1271
1272 /* By now the buffer is marked for journaling */
1273 nlen = EXT3_DIR_REC_LEN(de->name_len);
1274 rlen = le16_to_cpu(de->rec_len);
1275 if (de->inode) {
1276 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1277 de1->rec_len = cpu_to_le16(rlen - nlen);
1278 de->rec_len = cpu_to_le16(nlen);
1279 de = de1;
1280 }
1281 de->file_type = EXT3_FT_UNKNOWN;
1282 if (inode) {
1283 de->inode = cpu_to_le32(inode->i_ino);
1284 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1285 } else
1286 de->inode = 0;
1287 de->name_len = namelen;
1288 memcpy (de->name, name, namelen);
1289 /*
1290 * XXX shouldn't update any times until successful
1291 * completion of syscall, but too many callers depend
1292 * on this.
1293 *
1294 * XXX similarly, too many callers depend on
1295 * ext3_new_inode() setting the times, but error
1296 * recovery deletes the inode, so the worst that can
1297 * happen is that the times are slightly out of date
1298 * and/or different from the directory change time.
1299 */
1300 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1301 ext3_update_dx_flag(dir);
1302 dir->i_version++;
1303 ext3_mark_inode_dirty(handle, dir);
1304 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1305 err = ext3_journal_dirty_metadata(handle, bh);
1306 if (err)
1307 ext3_std_error(dir->i_sb, err);
1308 brelse(bh);
1309 return 0;
1310}
1311
1312#ifdef CONFIG_EXT3_INDEX
1313/*
1314 * This converts a one block unindexed directory to a 3 block indexed
1315 * directory, and adds the dentry to the indexed directory.
1316 */
1317static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1318 struct inode *inode, struct buffer_head *bh)
1319{
1320 struct inode *dir = dentry->d_parent->d_inode;
1321 const char *name = dentry->d_name.name;
1322 int namelen = dentry->d_name.len;
1323 struct buffer_head *bh2;
1324 struct dx_root *root;
1325 struct dx_frame frames[2], *frame;
1326 struct dx_entry *entries;
1327 struct ext3_dir_entry_2 *de, *de2;
1328 char *data1, *top;
1329 unsigned len;
1330 int retval;
1331 unsigned blocksize;
1332 struct dx_hash_info hinfo;
1333 u32 block;
1334 struct fake_dirent *fde;
1335
1336 blocksize = dir->i_sb->s_blocksize;
1337 dxtrace(printk("Creating index\n"));
1338 retval = ext3_journal_get_write_access(handle, bh);
1339 if (retval) {
1340 ext3_std_error(dir->i_sb, retval);
1341 brelse(bh);
1342 return retval;
1343 }
1344 root = (struct dx_root *) bh->b_data;
1345
1346 bh2 = ext3_append (handle, dir, &block, &retval);
1347 if (!(bh2)) {
1348 brelse(bh);
1349 return retval;
1350 }
1351 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1352 data1 = bh2->b_data;
1353
1354 /* The 0th block becomes the root, move the dirents out */
1355 fde = &root->dotdot;
1356 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1357 len = ((char *) root) + blocksize - (char *) de;
1358 memcpy (data1, de, len);
1359 de = (struct ext3_dir_entry_2 *) data1;
1360 top = data1 + len;
1361 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1362 de = de2;
1363 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1364 /* Initialize the root; the dot dirents already exist */
1365 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1366 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1367 memset (&root->info, 0, sizeof(root->info));
1368 root->info.info_length = sizeof(root->info);
1369 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1370 entries = root->entries;
1371 dx_set_block (entries, 1);
1372 dx_set_count (entries, 1);
1373 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1374
1375 /* Initialize as for dx_probe */
1376 hinfo.hash_version = root->info.hash_version;
1377 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1378 ext3fs_dirhash(name, namelen, &hinfo);
1379 frame = frames;
1380 frame->entries = entries;
1381 frame->at = entries;
1382 frame->bh = bh;
1383 bh = bh2;
1384 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1385 dx_release (frames);
1386 if (!(de))
1387 return retval;
1388
1389 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1390}
1391#endif
1392
1393/*
1394 * ext3_add_entry()
1395 *
1396 * adds a file entry to the specified directory, using the same
1397 * semantics as ext3_find_entry(). It returns NULL if it failed.
1398 *
1399 * NOTE!! The inode part of 'de' is left at 0 - which means you
1400 * may not sleep between calling this and putting something into
1401 * the entry, as someone else might have used it while you slept.
1402 */
1403static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1404 struct inode *inode)
1405{
1406 struct inode *dir = dentry->d_parent->d_inode;
1407 unsigned long offset;
1408 struct buffer_head * bh;
1409 struct ext3_dir_entry_2 *de;
1410 struct super_block * sb;
1411 int retval;
1412#ifdef CONFIG_EXT3_INDEX
1413 int dx_fallback=0;
1414#endif
1415 unsigned blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 u32 block, blocks;
1417
1418 sb = dir->i_sb;
1419 blocksize = sb->s_blocksize;
1420 if (!dentry->d_name.len)
1421 return -EINVAL;
1422#ifdef CONFIG_EXT3_INDEX
1423 if (is_dx(dir)) {
1424 retval = ext3_dx_add_entry(handle, dentry, inode);
1425 if (!retval || (retval != ERR_BAD_DX_DIR))
1426 return retval;
1427 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1428 dx_fallback++;
1429 ext3_mark_inode_dirty(handle, dir);
1430 }
1431#endif
1432 blocks = dir->i_size >> sb->s_blocksize_bits;
1433 for (block = 0, offset = 0; block < blocks; block++) {
1434 bh = ext3_bread(handle, dir, block, 0, &retval);
1435 if(!bh)
1436 return retval;
1437 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1438 if (retval != -ENOSPC)
1439 return retval;
1440
1441#ifdef CONFIG_EXT3_INDEX
1442 if (blocks == 1 && !dx_fallback &&
1443 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1444 return make_indexed_dir(handle, dentry, inode, bh);
1445#endif
1446 brelse(bh);
1447 }
1448 bh = ext3_append(handle, dir, &block, &retval);
1449 if (!bh)
1450 return retval;
1451 de = (struct ext3_dir_entry_2 *) bh->b_data;
1452 de->inode = 0;
Johann Lombardi92eeccd2006-06-25 05:48:33 -07001453 de->rec_len = cpu_to_le16(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1455}
1456
1457#ifdef CONFIG_EXT3_INDEX
1458/*
1459 * Returns 0 for success, or a negative error value
1460 */
1461static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1462 struct inode *inode)
1463{
1464 struct dx_frame frames[2], *frame;
1465 struct dx_entry *entries, *at;
1466 struct dx_hash_info hinfo;
1467 struct buffer_head * bh;
1468 struct inode *dir = dentry->d_parent->d_inode;
1469 struct super_block * sb = dir->i_sb;
1470 struct ext3_dir_entry_2 *de;
1471 int err;
1472
1473 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1474 if (!frame)
1475 return err;
1476 entries = frame->entries;
1477 at = frame->at;
1478
1479 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1480 goto cleanup;
1481
1482 BUFFER_TRACE(bh, "get_write_access");
1483 err = ext3_journal_get_write_access(handle, bh);
1484 if (err)
1485 goto journal_error;
1486
1487 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1488 if (err != -ENOSPC) {
1489 bh = NULL;
1490 goto cleanup;
1491 }
1492
1493 /* Block full, should compress but for now just split */
1494 dxtrace(printk("using %u of %u node entries\n",
1495 dx_get_count(entries), dx_get_limit(entries)));
1496 /* Need to split index? */
1497 if (dx_get_count(entries) == dx_get_limit(entries)) {
1498 u32 newblock;
1499 unsigned icount = dx_get_count(entries);
1500 int levels = frame - frames;
1501 struct dx_entry *entries2;
1502 struct dx_node *node2;
1503 struct buffer_head *bh2;
1504
1505 if (levels && (dx_get_count(frames->entries) ==
1506 dx_get_limit(frames->entries))) {
1507 ext3_warning(sb, __FUNCTION__,
Glauber de Oliveira Costa9f406682006-01-08 01:03:22 -08001508 "Directory index full!");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 err = -ENOSPC;
1510 goto cleanup;
1511 }
1512 bh2 = ext3_append (handle, dir, &newblock, &err);
1513 if (!(bh2))
1514 goto cleanup;
1515 node2 = (struct dx_node *)(bh2->b_data);
1516 entries2 = node2->entries;
1517 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1518 node2->fake.inode = 0;
1519 BUFFER_TRACE(frame->bh, "get_write_access");
1520 err = ext3_journal_get_write_access(handle, frame->bh);
1521 if (err)
1522 goto journal_error;
1523 if (levels) {
1524 unsigned icount1 = icount/2, icount2 = icount - icount1;
1525 unsigned hash2 = dx_get_hash(entries + icount1);
1526 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1527
1528 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1529 err = ext3_journal_get_write_access(handle,
1530 frames[0].bh);
1531 if (err)
1532 goto journal_error;
1533
1534 memcpy ((char *) entries2, (char *) (entries + icount1),
1535 icount2 * sizeof(struct dx_entry));
1536 dx_set_count (entries, icount1);
1537 dx_set_count (entries2, icount2);
1538 dx_set_limit (entries2, dx_node_limit(dir));
1539
1540 /* Which index block gets the new entry? */
1541 if (at - entries >= icount1) {
1542 frame->at = at = at - entries - icount1 + entries2;
1543 frame->entries = entries = entries2;
1544 swap(frame->bh, bh2);
1545 }
1546 dx_insert_block (frames + 0, hash2, newblock);
1547 dxtrace(dx_show_index ("node", frames[1].entries));
1548 dxtrace(dx_show_index ("node",
1549 ((struct dx_node *) bh2->b_data)->entries));
1550 err = ext3_journal_dirty_metadata(handle, bh2);
1551 if (err)
1552 goto journal_error;
1553 brelse (bh2);
1554 } else {
1555 dxtrace(printk("Creating second level index...\n"));
1556 memcpy((char *) entries2, (char *) entries,
1557 icount * sizeof(struct dx_entry));
1558 dx_set_limit(entries2, dx_node_limit(dir));
1559
1560 /* Set up root */
1561 dx_set_count(entries, 1);
1562 dx_set_block(entries + 0, newblock);
1563 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1564
1565 /* Add new access path frame */
1566 frame = frames + 1;
1567 frame->at = at = at - entries + entries2;
1568 frame->entries = entries = entries2;
1569 frame->bh = bh2;
1570 err = ext3_journal_get_write_access(handle,
1571 frame->bh);
1572 if (err)
1573 goto journal_error;
1574 }
1575 ext3_journal_dirty_metadata(handle, frames[0].bh);
1576 }
1577 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1578 if (!de)
1579 goto cleanup;
1580 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1581 bh = NULL;
1582 goto cleanup;
1583
1584journal_error:
1585 ext3_std_error(dir->i_sb, err);
1586cleanup:
1587 if (bh)
1588 brelse(bh);
1589 dx_release(frames);
1590 return err;
1591}
1592#endif
1593
1594/*
1595 * ext3_delete_entry deletes a directory entry by merging it with the
1596 * previous entry
1597 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001598static int ext3_delete_entry (handle_t *handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 struct inode * dir,
1600 struct ext3_dir_entry_2 * de_del,
1601 struct buffer_head * bh)
1602{
1603 struct ext3_dir_entry_2 * de, * pde;
1604 int i;
1605
1606 i = 0;
1607 pde = NULL;
1608 de = (struct ext3_dir_entry_2 *) bh->b_data;
1609 while (i < bh->b_size) {
1610 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1611 return -EIO;
1612 if (de == de_del) {
1613 BUFFER_TRACE(bh, "get_write_access");
1614 ext3_journal_get_write_access(handle, bh);
1615 if (pde)
1616 pde->rec_len =
1617 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1618 le16_to_cpu(de->rec_len));
1619 else
1620 de->inode = 0;
1621 dir->i_version++;
1622 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1623 ext3_journal_dirty_metadata(handle, bh);
1624 return 0;
1625 }
1626 i += le16_to_cpu(de->rec_len);
1627 pde = de;
1628 de = (struct ext3_dir_entry_2 *)
1629 ((char *) de + le16_to_cpu(de->rec_len));
1630 }
1631 return -ENOENT;
1632}
1633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634static int ext3_add_nondir(handle_t *handle,
1635 struct dentry *dentry, struct inode *inode)
1636{
1637 int err = ext3_add_entry(handle, dentry, inode);
1638 if (!err) {
1639 ext3_mark_inode_dirty(handle, inode);
1640 d_instantiate(dentry, inode);
1641 return 0;
1642 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001643 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 iput(inode);
1645 return err;
1646}
1647
1648/*
1649 * By the time this is called, we already have created
1650 * the directory cache entry for the new file, but it
1651 * is so far negative - it has no inode.
1652 *
1653 * If the create succeeds, we fill in the inode information
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001654 * with d_instantiate().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 */
1656static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1657 struct nameidata *nd)
1658{
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001659 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 struct inode * inode;
1661 int err, retries = 0;
1662
1663retry:
Jan Kara1f545872005-06-23 22:01:04 -07001664 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001666 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 if (IS_ERR(handle))
1668 return PTR_ERR(handle);
1669
1670 if (IS_DIRSYNC(dir))
1671 handle->h_sync = 1;
1672
1673 inode = ext3_new_inode (handle, dir, mode);
1674 err = PTR_ERR(inode);
1675 if (!IS_ERR(inode)) {
1676 inode->i_op = &ext3_file_inode_operations;
1677 inode->i_fop = &ext3_file_operations;
1678 ext3_set_aops(inode);
1679 err = ext3_add_nondir(handle, dentry, inode);
1680 }
1681 ext3_journal_stop(handle);
1682 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1683 goto retry;
1684 return err;
1685}
1686
1687static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1688 int mode, dev_t rdev)
1689{
1690 handle_t *handle;
1691 struct inode *inode;
1692 int err, retries = 0;
1693
1694 if (!new_valid_dev(rdev))
1695 return -EINVAL;
1696
1697retry:
Jan Kara1f545872005-06-23 22:01:04 -07001698 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001699 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001700 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 if (IS_ERR(handle))
1702 return PTR_ERR(handle);
1703
1704 if (IS_DIRSYNC(dir))
1705 handle->h_sync = 1;
1706
1707 inode = ext3_new_inode (handle, dir, mode);
1708 err = PTR_ERR(inode);
1709 if (!IS_ERR(inode)) {
1710 init_special_inode(inode, inode->i_mode, rdev);
1711#ifdef CONFIG_EXT3_FS_XATTR
1712 inode->i_op = &ext3_special_inode_operations;
1713#endif
1714 err = ext3_add_nondir(handle, dentry, inode);
1715 }
1716 ext3_journal_stop(handle);
1717 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1718 goto retry;
1719 return err;
1720}
1721
1722static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1723{
1724 handle_t *handle;
1725 struct inode * inode;
1726 struct buffer_head * dir_block;
1727 struct ext3_dir_entry_2 * de;
1728 int err, retries = 0;
1729
1730 if (dir->i_nlink >= EXT3_LINK_MAX)
1731 return -EMLINK;
1732
1733retry:
Jan Kara1f545872005-06-23 22:01:04 -07001734 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001736 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 if (IS_ERR(handle))
1738 return PTR_ERR(handle);
1739
1740 if (IS_DIRSYNC(dir))
1741 handle->h_sync = 1;
1742
1743 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1744 err = PTR_ERR(inode);
1745 if (IS_ERR(inode))
1746 goto out_stop;
1747
1748 inode->i_op = &ext3_dir_inode_operations;
1749 inode->i_fop = &ext3_dir_operations;
1750 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1751 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1752 if (!dir_block) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001753 drop_nlink(inode); /* is this nlink == 0? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 ext3_mark_inode_dirty(handle, inode);
1755 iput (inode);
1756 goto out_stop;
1757 }
1758 BUFFER_TRACE(dir_block, "get_write_access");
1759 ext3_journal_get_write_access(handle, dir_block);
1760 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1761 de->inode = cpu_to_le32(inode->i_ino);
1762 de->name_len = 1;
1763 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1764 strcpy (de->name, ".");
1765 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1766 de = (struct ext3_dir_entry_2 *)
1767 ((char *) de + le16_to_cpu(de->rec_len));
1768 de->inode = cpu_to_le32(dir->i_ino);
1769 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1770 de->name_len = 2;
1771 strcpy (de->name, "..");
1772 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1773 inode->i_nlink = 2;
1774 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1775 ext3_journal_dirty_metadata(handle, dir_block);
1776 brelse (dir_block);
1777 ext3_mark_inode_dirty(handle, inode);
1778 err = ext3_add_entry (handle, dentry, inode);
1779 if (err) {
1780 inode->i_nlink = 0;
1781 ext3_mark_inode_dirty(handle, inode);
1782 iput (inode);
1783 goto out_stop;
1784 }
Dave Hansend8c76e62006-09-30 23:29:04 -07001785 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 ext3_update_dx_flag(dir);
1787 ext3_mark_inode_dirty(handle, dir);
1788 d_instantiate(dentry, inode);
1789out_stop:
1790 ext3_journal_stop(handle);
1791 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1792 goto retry;
1793 return err;
1794}
1795
1796/*
1797 * routine to check that the specified directory is empty (for rmdir)
1798 */
1799static int empty_dir (struct inode * inode)
1800{
1801 unsigned long offset;
1802 struct buffer_head * bh;
1803 struct ext3_dir_entry_2 * de, * de1;
1804 struct super_block * sb;
1805 int err = 0;
1806
1807 sb = inode->i_sb;
1808 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1809 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1810 if (err)
1811 ext3_error(inode->i_sb, __FUNCTION__,
1812 "error %d reading directory #%lu offset 0",
1813 err, inode->i_ino);
1814 else
1815 ext3_warning(inode->i_sb, __FUNCTION__,
1816 "bad directory (dir #%lu) - no data block",
1817 inode->i_ino);
1818 return 1;
1819 }
1820 de = (struct ext3_dir_entry_2 *) bh->b_data;
1821 de1 = (struct ext3_dir_entry_2 *)
1822 ((char *) de + le16_to_cpu(de->rec_len));
1823 if (le32_to_cpu(de->inode) != inode->i_ino ||
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001824 !le32_to_cpu(de1->inode) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 strcmp (".", de->name) ||
1826 strcmp ("..", de1->name)) {
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001827 ext3_warning (inode->i_sb, "empty_dir",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 "bad directory (dir #%lu) - no `.' or `..'",
1829 inode->i_ino);
1830 brelse (bh);
1831 return 1;
1832 }
1833 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1834 de = (struct ext3_dir_entry_2 *)
1835 ((char *) de1 + le16_to_cpu(de1->rec_len));
1836 while (offset < inode->i_size ) {
1837 if (!bh ||
1838 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1839 err = 0;
1840 brelse (bh);
1841 bh = ext3_bread (NULL, inode,
1842 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1843 if (!bh) {
1844 if (err)
1845 ext3_error(sb, __FUNCTION__,
1846 "error %d reading directory"
1847 " #%lu offset %lu",
1848 err, inode->i_ino, offset);
1849 offset += sb->s_blocksize;
1850 continue;
1851 }
1852 de = (struct ext3_dir_entry_2 *) bh->b_data;
1853 }
1854 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1855 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1856 sb->s_blocksize);
1857 offset = (offset | (sb->s_blocksize - 1)) + 1;
1858 continue;
1859 }
1860 if (le32_to_cpu(de->inode)) {
1861 brelse (bh);
1862 return 0;
1863 }
1864 offset += le16_to_cpu(de->rec_len);
1865 de = (struct ext3_dir_entry_2 *)
1866 ((char *) de + le16_to_cpu(de->rec_len));
1867 }
1868 brelse (bh);
1869 return 1;
1870}
1871
1872/* ext3_orphan_add() links an unlinked or truncated inode into a list of
1873 * such inodes, starting at the superblock, in case we crash before the
1874 * file is closed/deleted, or in case the inode truncate spans multiple
1875 * transactions and the last transaction is not recovered after a crash.
1876 *
1877 * At filesystem recovery time, we walk this list deleting unlinked
1878 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1879 */
1880int ext3_orphan_add(handle_t *handle, struct inode *inode)
1881{
1882 struct super_block *sb = inode->i_sb;
1883 struct ext3_iloc iloc;
1884 int err = 0, rc;
1885
1886 lock_super(sb);
1887 if (!list_empty(&EXT3_I(inode)->i_orphan))
1888 goto out_unlock;
1889
1890 /* Orphan handling is only valid for files with data blocks
1891 * being truncated, or files being unlinked. */
1892
1893 /* @@@ FIXME: Observation from aviro:
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001894 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 * here (on lock_super()), so race with ext3_link() which might bump
1896 * ->i_nlink. For, say it, character device. Not a regular file,
1897 * not a directory, not a symlink and ->i_nlink > 0.
1898 */
1899 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1900 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1901
1902 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1903 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1904 if (err)
1905 goto out_unlock;
1906
1907 err = ext3_reserve_inode_write(handle, inode, &iloc);
1908 if (err)
1909 goto out_unlock;
1910
1911 /* Insert this inode at the head of the on-disk orphan list... */
1912 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1913 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1914 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1915 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1916 if (!err)
1917 err = rc;
1918
1919 /* Only add to the head of the in-memory list if all the
1920 * previous operations succeeded. If the orphan_add is going to
1921 * fail (possibly taking the journal offline), we can't risk
1922 * leaving the inode on the orphan list: stray orphan-list
1923 * entries can cause panics at unmount time.
1924 *
1925 * This is safe: on error we're going to ignore the orphan list
1926 * anyway on the next recovery. */
1927 if (!err)
1928 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1929
Eric Sandeeneee194e2006-09-27 01:49:30 -07001930 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1931 jbd_debug(4, "orphan inode %lu will point to %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 inode->i_ino, NEXT_ORPHAN(inode));
1933out_unlock:
1934 unlock_super(sb);
1935 ext3_std_error(inode->i_sb, err);
1936 return err;
1937}
1938
1939/*
1940 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1941 * of such inodes stored on disk, because it is finally being cleaned up.
1942 */
1943int ext3_orphan_del(handle_t *handle, struct inode *inode)
1944{
1945 struct list_head *prev;
1946 struct ext3_inode_info *ei = EXT3_I(inode);
1947 struct ext3_sb_info *sbi;
1948 unsigned long ino_next;
1949 struct ext3_iloc iloc;
1950 int err = 0;
1951
1952 lock_super(inode->i_sb);
1953 if (list_empty(&ei->i_orphan)) {
1954 unlock_super(inode->i_sb);
1955 return 0;
1956 }
1957
1958 ino_next = NEXT_ORPHAN(inode);
1959 prev = ei->i_orphan.prev;
1960 sbi = EXT3_SB(inode->i_sb);
1961
1962 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1963
1964 list_del_init(&ei->i_orphan);
1965
1966 /* If we're on an error path, we may not have a valid
1967 * transaction handle with which to update the orphan list on
1968 * disk, but we still need to remove the inode from the linked
1969 * list in memory. */
1970 if (!handle)
1971 goto out;
1972
1973 err = ext3_reserve_inode_write(handle, inode, &iloc);
1974 if (err)
1975 goto out_err;
1976
1977 if (prev == &sbi->s_orphan) {
1978 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1979 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1980 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1981 if (err)
1982 goto out_brelse;
1983 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1984 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1985 } else {
1986 struct ext3_iloc iloc2;
1987 struct inode *i_prev =
1988 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1989
1990 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1991 i_prev->i_ino, ino_next);
1992 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1993 if (err)
1994 goto out_brelse;
1995 NEXT_ORPHAN(i_prev) = ino_next;
1996 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1997 }
1998 if (err)
1999 goto out_brelse;
2000 NEXT_ORPHAN(inode) = 0;
2001 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2002
2003out_err:
2004 ext3_std_error(inode->i_sb, err);
2005out:
2006 unlock_super(inode->i_sb);
2007 return err;
2008
2009out_brelse:
2010 brelse(iloc.bh);
2011 goto out_err;
2012}
2013
2014static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2015{
2016 int retval;
2017 struct inode * inode;
2018 struct buffer_head * bh;
2019 struct ext3_dir_entry_2 * de;
2020 handle_t *handle;
2021
2022 /* Initialize quotas before so that eventual writes go in
2023 * separate transaction */
2024 DQUOT_INIT(dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002025 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 if (IS_ERR(handle))
2027 return PTR_ERR(handle);
2028
2029 retval = -ENOENT;
2030 bh = ext3_find_entry (dentry, &de);
2031 if (!bh)
2032 goto end_rmdir;
2033
2034 if (IS_DIRSYNC(dir))
2035 handle->h_sync = 1;
2036
2037 inode = dentry->d_inode;
2038
2039 retval = -EIO;
2040 if (le32_to_cpu(de->inode) != inode->i_ino)
2041 goto end_rmdir;
2042
2043 retval = -ENOTEMPTY;
2044 if (!empty_dir (inode))
2045 goto end_rmdir;
2046
2047 retval = ext3_delete_entry(handle, dir, de, bh);
2048 if (retval)
2049 goto end_rmdir;
2050 if (inode->i_nlink != 2)
2051 ext3_warning (inode->i_sb, "ext3_rmdir",
2052 "empty directory has nlink!=2 (%d)",
2053 inode->i_nlink);
2054 inode->i_version++;
Dave Hansence71ec32006-09-30 23:29:06 -07002055 clear_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 /* There's no need to set i_disksize: the fact that i_nlink is
2057 * zero will ensure that the right thing happens during any
2058 * recovery. */
2059 inode->i_size = 0;
2060 ext3_orphan_add(handle, inode);
2061 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2062 ext3_mark_inode_dirty(handle, inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002063 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 ext3_update_dx_flag(dir);
2065 ext3_mark_inode_dirty(handle, dir);
2066
2067end_rmdir:
2068 ext3_journal_stop(handle);
2069 brelse (bh);
2070 return retval;
2071}
2072
2073static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2074{
2075 int retval;
2076 struct inode * inode;
2077 struct buffer_head * bh;
2078 struct ext3_dir_entry_2 * de;
2079 handle_t *handle;
2080
2081 /* Initialize quotas before so that eventual writes go
2082 * in separate transaction */
2083 DQUOT_INIT(dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002084 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 if (IS_ERR(handle))
2086 return PTR_ERR(handle);
2087
2088 if (IS_DIRSYNC(dir))
2089 handle->h_sync = 1;
2090
2091 retval = -ENOENT;
2092 bh = ext3_find_entry (dentry, &de);
2093 if (!bh)
2094 goto end_unlink;
2095
2096 inode = dentry->d_inode;
2097
2098 retval = -EIO;
2099 if (le32_to_cpu(de->inode) != inode->i_ino)
2100 goto end_unlink;
2101
2102 if (!inode->i_nlink) {
2103 ext3_warning (inode->i_sb, "ext3_unlink",
2104 "Deleting nonexistent file (%lu), %d",
2105 inode->i_ino, inode->i_nlink);
2106 inode->i_nlink = 1;
2107 }
2108 retval = ext3_delete_entry(handle, dir, de, bh);
2109 if (retval)
2110 goto end_unlink;
2111 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2112 ext3_update_dx_flag(dir);
2113 ext3_mark_inode_dirty(handle, dir);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002114 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 if (!inode->i_nlink)
2116 ext3_orphan_add(handle, inode);
2117 inode->i_ctime = dir->i_ctime;
2118 ext3_mark_inode_dirty(handle, inode);
2119 retval = 0;
2120
2121end_unlink:
2122 ext3_journal_stop(handle);
2123 brelse (bh);
2124 return retval;
2125}
2126
2127static int ext3_symlink (struct inode * dir,
2128 struct dentry *dentry, const char * symname)
2129{
2130 handle_t *handle;
2131 struct inode * inode;
2132 int l, err, retries = 0;
2133
2134 l = strlen(symname)+1;
2135 if (l > dir->i_sb->s_blocksize)
2136 return -ENAMETOOLONG;
2137
2138retry:
Jan Kara1f545872005-06-23 22:01:04 -07002139 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07002140 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
Jan Kara1f545872005-06-23 22:01:04 -07002141 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 if (IS_ERR(handle))
2143 return PTR_ERR(handle);
2144
2145 if (IS_DIRSYNC(dir))
2146 handle->h_sync = 1;
2147
2148 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2149 err = PTR_ERR(inode);
2150 if (IS_ERR(inode))
2151 goto out_stop;
2152
2153 if (l > sizeof (EXT3_I(inode)->i_data)) {
2154 inode->i_op = &ext3_symlink_inode_operations;
2155 ext3_set_aops(inode);
2156 /*
2157 * page_symlink() calls into ext3_prepare/commit_write.
2158 * We have a transaction open. All is sweetness. It also sets
2159 * i_size in generic_commit_write().
2160 */
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002161 err = __page_symlink(inode, symname, l,
2162 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 if (err) {
Eric Sandeen731b9a52007-02-10 01:46:16 -08002164 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 ext3_mark_inode_dirty(handle, inode);
2166 iput (inode);
2167 goto out_stop;
2168 }
2169 } else {
2170 inode->i_op = &ext3_fast_symlink_inode_operations;
2171 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2172 inode->i_size = l-1;
2173 }
2174 EXT3_I(inode)->i_disksize = inode->i_size;
2175 err = ext3_add_nondir(handle, dentry, inode);
2176out_stop:
2177 ext3_journal_stop(handle);
2178 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2179 goto retry;
2180 return err;
2181}
2182
2183static int ext3_link (struct dentry * old_dentry,
2184 struct inode * dir, struct dentry *dentry)
2185{
2186 handle_t *handle;
2187 struct inode *inode = old_dentry->d_inode;
2188 int err, retries = 0;
2189
2190 if (inode->i_nlink >= EXT3_LINK_MAX)
2191 return -EMLINK;
Eric Sandeen2988a772007-02-10 01:46:16 -08002192 /*
2193 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2194 * otherwise has the potential to corrupt the orphan inode list.
2195 */
2196 if (inode->i_nlink == 0)
2197 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
2199retry:
Jan Kara1f545872005-06-23 22:01:04 -07002200 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2202 if (IS_ERR(handle))
2203 return PTR_ERR(handle);
2204
2205 if (IS_DIRSYNC(dir))
2206 handle->h_sync = 1;
2207
2208 inode->i_ctime = CURRENT_TIME_SEC;
Eric Sandeen731b9a52007-02-10 01:46:16 -08002209 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 atomic_inc(&inode->i_count);
2211
2212 err = ext3_add_nondir(handle, dentry, inode);
2213 ext3_journal_stop(handle);
2214 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2215 goto retry;
2216 return err;
2217}
2218
2219#define PARENT_INO(buffer) \
2220 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2221 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2222
2223/*
2224 * Anybody can rename anything with this: the permission checks are left to the
2225 * higher-level routines.
2226 */
2227static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2228 struct inode * new_dir,struct dentry *new_dentry)
2229{
2230 handle_t *handle;
2231 struct inode * old_inode, * new_inode;
2232 struct buffer_head * old_bh, * new_bh, * dir_bh;
2233 struct ext3_dir_entry_2 * old_de, * new_de;
2234 int retval;
2235
2236 old_bh = new_bh = dir_bh = NULL;
2237
2238 /* Initialize quotas before so that eventual writes go
2239 * in separate transaction */
2240 if (new_dentry->d_inode)
2241 DQUOT_INIT(new_dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002242 handle = ext3_journal_start(old_dir, 2 *
2243 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07002244 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 if (IS_ERR(handle))
2246 return PTR_ERR(handle);
2247
2248 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2249 handle->h_sync = 1;
2250
2251 old_bh = ext3_find_entry (old_dentry, &old_de);
2252 /*
2253 * Check for inode number is _not_ due to possible IO errors.
2254 * We might rmdir the source, keep it as pwd of some process
2255 * and merrily kill the link to whatever was created under the
2256 * same name. Goodbye sticky bit ;-<
2257 */
2258 old_inode = old_dentry->d_inode;
2259 retval = -ENOENT;
2260 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2261 goto end_rename;
2262
2263 new_inode = new_dentry->d_inode;
2264 new_bh = ext3_find_entry (new_dentry, &new_de);
2265 if (new_bh) {
2266 if (!new_inode) {
2267 brelse (new_bh);
2268 new_bh = NULL;
2269 }
2270 }
2271 if (S_ISDIR(old_inode->i_mode)) {
2272 if (new_inode) {
2273 retval = -ENOTEMPTY;
2274 if (!empty_dir (new_inode))
2275 goto end_rename;
2276 }
2277 retval = -EIO;
2278 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2279 if (!dir_bh)
2280 goto end_rename;
2281 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2282 goto end_rename;
2283 retval = -EMLINK;
2284 if (!new_inode && new_dir!=old_dir &&
2285 new_dir->i_nlink >= EXT3_LINK_MAX)
2286 goto end_rename;
2287 }
2288 if (!new_bh) {
2289 retval = ext3_add_entry (handle, new_dentry, old_inode);
2290 if (retval)
2291 goto end_rename;
2292 } else {
2293 BUFFER_TRACE(new_bh, "get write access");
2294 ext3_journal_get_write_access(handle, new_bh);
2295 new_de->inode = cpu_to_le32(old_inode->i_ino);
2296 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2297 EXT3_FEATURE_INCOMPAT_FILETYPE))
2298 new_de->file_type = old_de->file_type;
2299 new_dir->i_version++;
2300 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2301 ext3_journal_dirty_metadata(handle, new_bh);
2302 brelse(new_bh);
2303 new_bh = NULL;
2304 }
2305
2306 /*
2307 * Like most other Unix systems, set the ctime for inodes on a
2308 * rename.
2309 */
2310 old_inode->i_ctime = CURRENT_TIME_SEC;
2311 ext3_mark_inode_dirty(handle, old_inode);
2312
2313 /*
2314 * ok, that's it
2315 */
2316 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2317 old_de->name_len != old_dentry->d_name.len ||
2318 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2319 (retval = ext3_delete_entry(handle, old_dir,
2320 old_de, old_bh)) == -ENOENT) {
2321 /* old_de could have moved from under us during htree split, so
2322 * make sure that we are deleting the right entry. We might
2323 * also be pointing to a stale entry in the unused part of
2324 * old_bh so just checking inum and the name isn't enough. */
2325 struct buffer_head *old_bh2;
2326 struct ext3_dir_entry_2 *old_de2;
2327
2328 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2329 if (old_bh2) {
2330 retval = ext3_delete_entry(handle, old_dir,
2331 old_de2, old_bh2);
2332 brelse(old_bh2);
2333 }
2334 }
2335 if (retval) {
2336 ext3_warning(old_dir->i_sb, "ext3_rename",
2337 "Deleting old file (%lu), %d, error=%d",
2338 old_dir->i_ino, old_dir->i_nlink, retval);
2339 }
2340
2341 if (new_inode) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002342 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 new_inode->i_ctime = CURRENT_TIME_SEC;
2344 }
2345 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2346 ext3_update_dx_flag(old_dir);
2347 if (dir_bh) {
2348 BUFFER_TRACE(dir_bh, "get_write_access");
2349 ext3_journal_get_write_access(handle, dir_bh);
2350 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2351 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2352 ext3_journal_dirty_metadata(handle, dir_bh);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002353 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 if (new_inode) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002355 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 } else {
Dave Hansend8c76e62006-09-30 23:29:04 -07002357 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 ext3_update_dx_flag(new_dir);
2359 ext3_mark_inode_dirty(handle, new_dir);
2360 }
2361 }
2362 ext3_mark_inode_dirty(handle, old_dir);
2363 if (new_inode) {
2364 ext3_mark_inode_dirty(handle, new_inode);
2365 if (!new_inode->i_nlink)
2366 ext3_orphan_add(handle, new_inode);
2367 }
2368 retval = 0;
2369
2370end_rename:
2371 brelse (dir_bh);
2372 brelse (old_bh);
2373 brelse (new_bh);
2374 ext3_journal_stop(handle);
2375 return retval;
2376}
2377
2378/*
2379 * directories can handle most operations...
2380 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002381const struct inode_operations ext3_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 .create = ext3_create,
2383 .lookup = ext3_lookup,
2384 .link = ext3_link,
2385 .unlink = ext3_unlink,
2386 .symlink = ext3_symlink,
2387 .mkdir = ext3_mkdir,
2388 .rmdir = ext3_rmdir,
2389 .mknod = ext3_mknod,
2390 .rename = ext3_rename,
2391 .setattr = ext3_setattr,
2392#ifdef CONFIG_EXT3_FS_XATTR
2393 .setxattr = generic_setxattr,
2394 .getxattr = generic_getxattr,
2395 .listxattr = ext3_listxattr,
2396 .removexattr = generic_removexattr,
2397#endif
2398 .permission = ext3_permission,
2399};
2400
Arjan van de Ven754661f2007-02-12 00:55:38 -08002401const struct inode_operations ext3_special_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 .setattr = ext3_setattr,
2403#ifdef CONFIG_EXT3_FS_XATTR
2404 .setxattr = generic_setxattr,
2405 .getxattr = generic_getxattr,
2406 .listxattr = ext3_listxattr,
2407 .removexattr = generic_removexattr,
2408#endif
2409 .permission = ext3_permission,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002410};