blob: 906731a20f1ae37c4f7be8999d72d7986cbabadd [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/smp_lock.h>
Ben Dooks381be252005-10-30 15:02:56 -080040
41#include "namei.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#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 if ((bh = ext3_bread(handle, inode, *block, 1, err))) {
62 inode->i_size += inode->i_sb->s_blocksize;
63 EXT3_I(inode)->i_disksize = inode->i_size;
64 ext3_journal_get_write_access(handle,bh);
65 }
66 return bh;
67}
68
69#ifndef assert
70#define assert(test) J_ASSERT(test)
71#endif
72
73#ifndef swap
74#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
75#endif
76
77#ifdef DX_DEBUG
78#define dxtrace(command) command
79#else
Mingming Caoae6ddcc2006-09-27 01:49:27 -070080#define dxtrace(command)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#endif
82
83struct fake_dirent
84{
85 __le32 inode;
86 __le16 rec_len;
87 u8 name_len;
88 u8 file_type;
89};
90
91struct dx_countlimit
92{
93 __le16 limit;
94 __le16 count;
95};
96
97struct dx_entry
98{
99 __le32 hash;
100 __le32 block;
101};
102
103/*
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
107 */
108
109struct dx_root
110{
111 struct fake_dirent dot;
112 char dot_name[4];
113 struct fake_dirent dotdot;
114 char dotdot_name[4];
115 struct dx_root_info
116 {
117 __le32 reserved_zero;
118 u8 hash_version;
119 u8 info_length; /* 8 */
120 u8 indirect_levels;
121 u8 unused_flags;
122 }
123 info;
124 struct dx_entry entries[0];
125};
126
127struct dx_node
128{
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
131};
132
133
134struct dx_frame
135{
136 struct buffer_head *bh;
137 struct dx_entry *entries;
138 struct dx_entry *at;
139};
140
141struct dx_map_entry
142{
143 u32 hash;
144 u32 offs;
145};
146
147#ifdef CONFIG_EXT3_INDEX
148static inline unsigned dx_get_block (struct dx_entry *entry);
149static void dx_set_block (struct dx_entry *entry, unsigned value);
150static inline unsigned dx_get_hash (struct dx_entry *entry);
151static void dx_set_hash (struct dx_entry *entry, unsigned value);
152static unsigned dx_get_count (struct dx_entry *entries);
153static unsigned dx_get_limit (struct dx_entry *entries);
154static void dx_set_count (struct dx_entry *entries, unsigned value);
155static void dx_set_limit (struct dx_entry *entries, unsigned value);
156static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157static unsigned dx_node_limit (struct inode *dir);
158static struct dx_frame *dx_probe(struct dentry *dentry,
159 struct inode *dir,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
162 int *err);
163static void dx_release (struct dx_frame *frames);
164static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168 struct dx_map_entry *offsets, int count);
169static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
170static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171static int ext3_htree_next_block(struct inode *dir, __u32 hash,
172 struct dx_frame *frame,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700173 struct dx_frame *frames,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 __u32 *start_hash);
175static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
176 struct ext3_dir_entry_2 **res_dir, int *err);
177static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
178 struct inode *inode);
179
180/*
181 * Future: use high four bits of block for coalesce-on-delete flags
182 * Mask them off for now.
183 */
184
185static inline unsigned dx_get_block (struct dx_entry *entry)
186{
187 return le32_to_cpu(entry->block) & 0x00ffffff;
188}
189
190static inline void dx_set_block (struct dx_entry *entry, unsigned value)
191{
192 entry->block = cpu_to_le32(value);
193}
194
195static inline unsigned dx_get_hash (struct dx_entry *entry)
196{
197 return le32_to_cpu(entry->hash);
198}
199
200static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
201{
202 entry->hash = cpu_to_le32(value);
203}
204
205static inline unsigned dx_get_count (struct dx_entry *entries)
206{
207 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
208}
209
210static inline unsigned dx_get_limit (struct dx_entry *entries)
211{
212 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
213}
214
215static inline void dx_set_count (struct dx_entry *entries, unsigned value)
216{
217 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
218}
219
220static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
221{
222 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
223}
224
225static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
226{
227 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
228 EXT3_DIR_REC_LEN(2) - infosize;
229 return 0? 20: entry_space / sizeof(struct dx_entry);
230}
231
232static inline unsigned dx_node_limit (struct inode *dir)
233{
234 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
235 return 0? 22: entry_space / sizeof(struct dx_entry);
236}
237
238/*
239 * Debug
240 */
241#ifdef DX_DEBUG
242static void dx_show_index (char * label, struct dx_entry *entries)
243{
244 int i, n = dx_get_count (entries);
245 printk("%s index ", label);
246 for (i = 0; i < n; i++)
247 {
248 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
249 }
250 printk("\n");
251}
252
253struct stats
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700254{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 unsigned names;
256 unsigned space;
257 unsigned bcount;
258};
259
260static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
261 int size, int show_names)
262{
263 unsigned names = 0, space = 0;
264 char *base = (char *) de;
265 struct dx_hash_info h = *hinfo;
266
267 printk("names: ");
268 while ((char *) de < base + size)
269 {
270 if (de->inode)
271 {
272 if (show_names)
273 {
274 int len = de->name_len;
275 char *name = de->name;
276 while (len--) printk("%c", *name++);
277 ext3fs_dirhash(de->name, de->name_len, &h);
278 printk(":%x.%u ", h.hash,
279 ((char *) de - base));
280 }
281 space += EXT3_DIR_REC_LEN(de->name_len);
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700282 names++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 }
284 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
285 }
286 printk("(%i)\n", names);
287 return (struct stats) { names, space, 1 };
288}
289
290struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291 struct dx_entry *entries, int levels)
292{
293 unsigned blocksize = dir->i_sb->s_blocksize;
294 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
295 unsigned bcount = 0;
296 struct buffer_head *bh;
297 int err;
298 printk("%i indexed blocks...\n", count);
299 for (i = 0; i < count; i++, entries++)
300 {
301 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
302 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
303 struct stats stats;
304 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
305 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
306 stats = levels?
307 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
308 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
309 names += stats.names;
310 space += stats.space;
311 bcount += stats.bcount;
312 brelse (bh);
313 }
314 if (bcount)
315 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
316 names, space/bcount,(space/bcount)*100/blocksize);
317 return (struct stats) { names, space, bcount};
318}
319#endif /* DX_DEBUG */
320
321/*
322 * Probe for a directory leaf block to search.
323 *
324 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
325 * error in the directory index, and the caller should fall back to
326 * searching the directory normally. The callers of dx_probe **MUST**
327 * check for this error code, and make sure it never gets reflected
328 * back to userspace.
329 */
330static struct dx_frame *
331dx_probe(struct dentry *dentry, struct inode *dir,
332 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
333{
334 unsigned count, indirect;
335 struct dx_entry *at, *entries, *p, *q, *m;
336 struct dx_root *root;
337 struct buffer_head *bh;
338 struct dx_frame *frame = frame_in;
339 u32 hash;
340
341 frame->bh = NULL;
342 if (dentry)
343 dir = dentry->d_parent->d_inode;
344 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
345 goto fail;
346 root = (struct dx_root *) bh->b_data;
347 if (root->info.hash_version != DX_HASH_TEA &&
348 root->info.hash_version != DX_HASH_HALF_MD4 &&
349 root->info.hash_version != DX_HASH_LEGACY) {
350 ext3_warning(dir->i_sb, __FUNCTION__,
351 "Unrecognised inode hash code %d",
352 root->info.hash_version);
353 brelse(bh);
354 *err = ERR_BAD_DX_DIR;
355 goto fail;
356 }
357 hinfo->hash_version = root->info.hash_version;
358 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
359 if (dentry)
360 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
361 hash = hinfo->hash;
362
363 if (root->info.unused_flags & 1) {
364 ext3_warning(dir->i_sb, __FUNCTION__,
365 "Unimplemented inode hash flags: %#06x",
366 root->info.unused_flags);
367 brelse(bh);
368 *err = ERR_BAD_DX_DIR;
369 goto fail;
370 }
371
372 if ((indirect = root->info.indirect_levels) > 1) {
373 ext3_warning(dir->i_sb, __FUNCTION__,
374 "Unimplemented inode hash depth: %#06x",
375 root->info.indirect_levels);
376 brelse(bh);
377 *err = ERR_BAD_DX_DIR;
378 goto fail;
379 }
380
381 entries = (struct dx_entry *) (((char *)&root->info) +
382 root->info.info_length);
383 assert(dx_get_limit(entries) == dx_root_limit(dir,
384 root->info.info_length));
385 dxtrace (printk("Look up %x", hash));
386 while (1)
387 {
388 count = dx_get_count(entries);
389 assert (count && count <= dx_get_limit(entries));
390 p = entries + 1;
391 q = entries + count - 1;
392 while (p <= q)
393 {
394 m = p + (q - p)/2;
395 dxtrace(printk("."));
396 if (dx_get_hash(m) > hash)
397 q = m - 1;
398 else
399 p = m + 1;
400 }
401
402 if (0) // linear search cross check
403 {
404 unsigned n = count - 1;
405 at = entries;
406 while (n--)
407 {
408 dxtrace(printk(","));
409 if (dx_get_hash(++at) > hash)
410 {
411 at--;
412 break;
413 }
414 }
415 assert (at == p - 1);
416 }
417
418 at = p - 1;
419 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
420 frame->bh = bh;
421 frame->entries = entries;
422 frame->at = at;
423 if (!indirect--) return frame;
424 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
425 goto fail2;
426 at = entries = ((struct dx_node *) bh->b_data)->entries;
427 assert (dx_get_limit(entries) == dx_node_limit (dir));
428 frame++;
429 }
430fail2:
431 while (frame >= frame_in) {
432 brelse(frame->bh);
433 frame--;
434 }
435fail:
436 return NULL;
437}
438
439static void dx_release (struct dx_frame *frames)
440{
441 if (frames[0].bh == NULL)
442 return;
443
444 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
445 brelse(frames[1].bh);
446 brelse(frames[0].bh);
447}
448
449/*
450 * This function increments the frame pointer to search the next leaf
451 * block, and reads in the necessary intervening nodes if the search
452 * should be necessary. Whether or not the search is necessary is
453 * controlled by the hash parameter. If the hash value is even, then
454 * the search is only continued if the next block starts with that
455 * hash value. This is used if we are searching for a specific file.
456 *
457 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
458 *
459 * This function returns 1 if the caller should continue to search,
460 * or 0 if it should not. If there is an error reading one of the
461 * index blocks, it will a negative error code.
462 *
463 * If start_hash is non-null, it will be filled in with the starting
464 * hash of the next page.
465 */
466static int ext3_htree_next_block(struct inode *dir, __u32 hash,
467 struct dx_frame *frame,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700468 struct dx_frame *frames,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 __u32 *start_hash)
470{
471 struct dx_frame *p;
472 struct buffer_head *bh;
473 int err, num_frames = 0;
474 __u32 bhash;
475
476 p = frame;
477 /*
478 * Find the next leaf page by incrementing the frame pointer.
479 * If we run out of entries in the interior node, loop around and
480 * increment pointer in the parent node. When we break out of
481 * this loop, num_frames indicates the number of interior
482 * nodes need to be read.
483 */
484 while (1) {
485 if (++(p->at) < p->entries + dx_get_count(p->entries))
486 break;
487 if (p == frames)
488 return 0;
489 num_frames++;
490 p--;
491 }
492
493 /*
494 * If the hash is 1, then continue only if the next page has a
495 * continuation hash of any value. This is used for readdir
496 * handling. Otherwise, check to see if the hash matches the
497 * desired contiuation hash. If it doesn't, return since
498 * there's no point to read in the successive index pages.
499 */
500 bhash = dx_get_hash(p->at);
501 if (start_hash)
502 *start_hash = bhash;
503 if ((hash & 1) == 0) {
504 if ((bhash & ~1) != hash)
505 return 0;
506 }
507 /*
508 * If the hash is HASH_NB_ALWAYS, we always go to the next
509 * block so no check is necessary
510 */
511 while (num_frames--) {
512 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
513 0, &err)))
514 return err; /* Failure */
515 p++;
516 brelse (p->bh);
517 p->bh = bh;
518 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
519 }
520 return 1;
521}
522
523
524/*
525 * p is at least 6 bytes before the end of page
526 */
527static inline struct ext3_dir_entry_2 *ext3_next_entry(struct ext3_dir_entry_2 *p)
528{
529 return (struct ext3_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
530}
531
532/*
533 * This function fills a red-black tree with information from a
534 * directory block. It returns the number directory entries loaded
535 * into the tree. If there is an error it is returned in err.
536 */
537static int htree_dirblock_to_tree(struct file *dir_file,
538 struct inode *dir, int block,
539 struct dx_hash_info *hinfo,
540 __u32 start_hash, __u32 start_minor_hash)
541{
542 struct buffer_head *bh;
543 struct ext3_dir_entry_2 *de, *top;
544 int err, count = 0;
545
546 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
547 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
548 return err;
549
550 de = (struct ext3_dir_entry_2 *) bh->b_data;
551 top = (struct ext3_dir_entry_2 *) ((char *) de +
552 dir->i_sb->s_blocksize -
553 EXT3_DIR_REC_LEN(0));
554 for (; de < top; de = ext3_next_entry(de)) {
555 ext3fs_dirhash(de->name, de->name_len, hinfo);
556 if ((hinfo->hash < start_hash) ||
557 ((hinfo->hash == start_hash) &&
558 (hinfo->minor_hash < start_minor_hash)))
559 continue;
560 if (de->inode == 0)
561 continue;
562 if ((err = ext3_htree_store_dirent(dir_file,
563 hinfo->hash, hinfo->minor_hash, de)) != 0) {
564 brelse(bh);
565 return err;
566 }
567 count++;
568 }
569 brelse(bh);
570 return count;
571}
572
573
574/*
575 * This function fills a red-black tree with information from a
576 * directory. We start scanning the directory in hash order, starting
577 * at start_hash and start_minor_hash.
578 *
579 * This function returns the number of entries inserted into the tree,
580 * or a negative error code.
581 */
582int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
583 __u32 start_minor_hash, __u32 *next_hash)
584{
585 struct dx_hash_info hinfo;
586 struct ext3_dir_entry_2 *de;
587 struct dx_frame frames[2], *frame;
588 struct inode *dir;
589 int block, err;
590 int count = 0;
591 int ret;
592 __u32 hashval;
593
594 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
595 start_minor_hash));
596 dir = dir_file->f_dentry->d_inode;
597 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
598 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
599 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
600 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
601 start_hash, start_minor_hash);
602 *next_hash = ~0;
603 return count;
604 }
605 hinfo.hash = start_hash;
606 hinfo.minor_hash = 0;
607 frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
608 if (!frame)
609 return err;
610
611 /* Add '.' and '..' from the htree header */
612 if (!start_hash && !start_minor_hash) {
613 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
614 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
615 goto errout;
616 count++;
617 }
618 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
619 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
620 de = ext3_next_entry(de);
621 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
622 goto errout;
623 count++;
624 }
625
626 while (1) {
627 block = dx_get_block(frame->at);
628 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
629 start_hash, start_minor_hash);
630 if (ret < 0) {
631 err = ret;
632 goto errout;
633 }
634 count += ret;
635 hashval = ~0;
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700636 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 frame, frames, &hashval);
638 *next_hash = hashval;
639 if (ret < 0) {
640 err = ret;
641 goto errout;
642 }
643 /*
644 * Stop if: (a) there are no more entries, or
645 * (b) we have inserted at least one entry and the
646 * next hash value is not a continuation
647 */
648 if ((ret == 0) ||
649 (count && ((hashval & 1) == 0)))
650 break;
651 }
652 dx_release(frames);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700653 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 count, *next_hash));
655 return count;
656errout:
657 dx_release(frames);
658 return (err);
659}
660
661
662/*
663 * Directory block splitting, compacting
664 */
665
666static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
667 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
668{
669 int count = 0;
670 char *base = (char *) de;
671 struct dx_hash_info h = *hinfo;
672
673 while ((char *) de < base + size)
674 {
675 if (de->name_len && de->inode) {
676 ext3fs_dirhash(de->name, de->name_len, &h);
677 map_tail--;
678 map_tail->hash = h.hash;
679 map_tail->offs = (u32) ((char *) de - base);
680 count++;
681 cond_resched();
682 }
683 /* XXX: do we need to check rec_len == 0 case? -Chris */
684 de = (struct ext3_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
685 }
686 return count;
687}
688
689static void dx_sort_map (struct dx_map_entry *map, unsigned count)
690{
691 struct dx_map_entry *p, *q, *top = map + count - 1;
692 int more;
693 /* Combsort until bubble sort doesn't suck */
694 while (count > 2)
695 {
696 count = count*10/13;
697 if (count - 9 < 2) /* 9, 10 -> 11 */
698 count = 11;
699 for (p = top, q = p - count; q >= map; p--, q--)
700 if (p->hash < q->hash)
701 swap(*p, *q);
702 }
703 /* Garden variety bubble sort */
704 do {
705 more = 0;
706 q = top;
707 while (q-- > map)
708 {
709 if (q[1].hash >= q[0].hash)
710 continue;
711 swap(*(q+1), *q);
712 more = 1;
713 }
714 } while(more);
715}
716
717static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
718{
719 struct dx_entry *entries = frame->entries;
720 struct dx_entry *old = frame->at, *new = old + 1;
721 int count = dx_get_count(entries);
722
723 assert(count < dx_get_limit(entries));
724 assert(old < entries + count);
725 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
726 dx_set_hash(new, hash);
727 dx_set_block(new, block);
728 dx_set_count(entries, count + 1);
729}
730#endif
731
732
733static void ext3_update_dx_flag(struct inode *inode)
734{
735 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
736 EXT3_FEATURE_COMPAT_DIR_INDEX))
737 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
738}
739
740/*
741 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
742 *
743 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
744 * `de != NULL' is guaranteed by caller.
745 */
746static inline int ext3_match (int len, const char * const name,
747 struct ext3_dir_entry_2 * de)
748{
749 if (len != de->name_len)
750 return 0;
751 if (!de->inode)
752 return 0;
753 return !memcmp(name, de->name, len);
754}
755
756/*
757 * Returns 0 if not found, -1 on failure, and 1 on success
758 */
759static inline int search_dirblock(struct buffer_head * bh,
760 struct inode *dir,
761 struct dentry *dentry,
762 unsigned long offset,
763 struct ext3_dir_entry_2 ** res_dir)
764{
765 struct ext3_dir_entry_2 * de;
766 char * dlimit;
767 int de_len;
768 const char *name = dentry->d_name.name;
769 int namelen = dentry->d_name.len;
770
771 de = (struct ext3_dir_entry_2 *) bh->b_data;
772 dlimit = bh->b_data + dir->i_sb->s_blocksize;
773 while ((char *) de < dlimit) {
774 /* this code is executed quadratically often */
775 /* do minimal checking `by hand' */
776
777 if ((char *) de + namelen <= dlimit &&
778 ext3_match (namelen, name, de)) {
779 /* found a match - just to be sure, do a full check */
780 if (!ext3_check_dir_entry("ext3_find_entry",
781 dir, de, bh, offset))
782 return -1;
783 *res_dir = de;
784 return 1;
785 }
786 /* prevent looping on a bad block */
787 de_len = le16_to_cpu(de->rec_len);
788 if (de_len <= 0)
789 return -1;
790 offset += de_len;
791 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
792 }
793 return 0;
794}
795
796
797/*
798 * ext3_find_entry()
799 *
800 * finds an entry in the specified directory with the wanted name. It
801 * returns the cache buffer in which the entry was found, and the entry
802 * itself (as a parameter - res_dir). It does NOT read the inode of the
803 * entry - you'll have to do that yourself if you want to.
804 *
805 * The returned buffer_head has ->b_count elevated. The caller is expected
806 * to brelse() it when appropriate.
807 */
808static struct buffer_head * ext3_find_entry (struct dentry *dentry,
809 struct ext3_dir_entry_2 ** res_dir)
810{
811 struct super_block * sb;
812 struct buffer_head * bh_use[NAMEI_RA_SIZE];
813 struct buffer_head * bh, *ret = NULL;
814 unsigned long start, block, b;
815 int ra_max = 0; /* Number of bh's in the readahead
816 buffer, bh_use[] */
817 int ra_ptr = 0; /* Current index into readahead
818 buffer */
819 int num = 0;
820 int nblocks, i, err;
821 struct inode *dir = dentry->d_parent->d_inode;
822 int namelen;
823 const u8 *name;
824 unsigned blocksize;
825
826 *res_dir = NULL;
827 sb = dir->i_sb;
828 blocksize = sb->s_blocksize;
829 namelen = dentry->d_name.len;
830 name = dentry->d_name.name;
831 if (namelen > EXT3_NAME_LEN)
832 return NULL;
833#ifdef CONFIG_EXT3_INDEX
834 if (is_dx(dir)) {
835 bh = ext3_dx_find_entry(dentry, res_dir, &err);
836 /*
837 * On success, or if the error was file not found,
838 * return. Otherwise, fall back to doing a search the
839 * old fashioned way.
840 */
841 if (bh || (err != ERR_BAD_DX_DIR))
842 return bh;
843 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
844 }
845#endif
846 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
847 start = EXT3_I(dir)->i_dir_start_lookup;
848 if (start >= nblocks)
849 start = 0;
850 block = start;
851restart:
852 do {
853 /*
854 * We deal with the read-ahead logic here.
855 */
856 if (ra_ptr >= ra_max) {
857 /* Refill the readahead buffer */
858 ra_ptr = 0;
859 b = block;
860 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
861 /*
862 * Terminate if we reach the end of the
863 * directory and must wrap, or if our
864 * search has finished at this block.
865 */
866 if (b >= nblocks || (num && block == start)) {
867 bh_use[ra_max] = NULL;
868 break;
869 }
870 num++;
871 bh = ext3_getblk(NULL, dir, b++, 0, &err);
872 bh_use[ra_max] = bh;
873 if (bh)
Jens Axboecaa38fb2006-07-23 01:41:26 +0200874 ll_rw_block(READ_META, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 }
876 }
877 if ((bh = bh_use[ra_ptr++]) == NULL)
878 goto next;
879 wait_on_buffer(bh);
880 if (!buffer_uptodate(bh)) {
881 /* read error, skip block & hope for the best */
882 ext3_error(sb, __FUNCTION__, "reading directory #%lu "
883 "offset %lu", dir->i_ino, block);
884 brelse(bh);
885 goto next;
886 }
887 i = search_dirblock(bh, dir, dentry,
888 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
889 if (i == 1) {
890 EXT3_I(dir)->i_dir_start_lookup = block;
891 ret = bh;
892 goto cleanup_and_exit;
893 } else {
894 brelse(bh);
895 if (i < 0)
896 goto cleanup_and_exit;
897 }
898 next:
899 if (++block >= nblocks)
900 block = 0;
901 } while (block != start);
902
903 /*
904 * If the directory has grown while we were searching, then
905 * search the last part of the directory before giving up.
906 */
907 block = nblocks;
908 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
909 if (block < nblocks) {
910 start = 0;
911 goto restart;
912 }
913
914cleanup_and_exit:
915 /* Clean up the read-ahead blocks */
916 for (; ra_ptr < ra_max; ra_ptr++)
917 brelse (bh_use[ra_ptr]);
918 return ret;
919}
920
921#ifdef CONFIG_EXT3_INDEX
922static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
923 struct ext3_dir_entry_2 **res_dir, int *err)
924{
925 struct super_block * sb;
926 struct dx_hash_info hinfo;
927 u32 hash;
928 struct dx_frame frames[2], *frame;
929 struct ext3_dir_entry_2 *de, *top;
930 struct buffer_head *bh;
931 unsigned long block;
932 int retval;
933 int namelen = dentry->d_name.len;
934 const u8 *name = dentry->d_name.name;
935 struct inode *dir = dentry->d_parent->d_inode;
936
937 sb = dir->i_sb;
Andreas Dilgeracfa1823d2005-06-23 00:09:45 -0700938 /* NFS may look up ".." - look at dx_root directory block */
939 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
940 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
941 return NULL;
942 } else {
943 frame = frames;
944 frame->bh = NULL; /* for dx_release() */
945 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
946 dx_set_block(frame->at, 0); /* dx_root block is 0 */
947 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 hash = hinfo.hash;
949 do {
950 block = dx_get_block(frame->at);
951 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
952 goto errout;
953 de = (struct ext3_dir_entry_2 *) bh->b_data;
954 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
955 EXT3_DIR_REC_LEN(0));
956 for (; de < top; de = ext3_next_entry(de))
957 if (ext3_match (namelen, name, de)) {
958 if (!ext3_check_dir_entry("ext3_find_entry",
959 dir, de, bh,
960 (block<<EXT3_BLOCK_SIZE_BITS(sb))
961 +((char *)de - bh->b_data))) {
962 brelse (bh);
963 goto errout;
964 }
965 *res_dir = de;
966 dx_release (frames);
967 return bh;
968 }
969 brelse (bh);
970 /* Check to see if we should continue to search */
971 retval = ext3_htree_next_block(dir, hash, frame,
972 frames, NULL);
973 if (retval < 0) {
974 ext3_warning(sb, __FUNCTION__,
975 "error reading index page in directory #%lu",
976 dir->i_ino);
977 *err = retval;
978 goto errout;
979 }
980 } while (retval == 1);
981
982 *err = -ENOENT;
983errout:
984 dxtrace(printk("%s not found\n", name));
985 dx_release (frames);
986 return NULL;
987}
988#endif
989
990static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
991{
992 struct inode * inode;
993 struct ext3_dir_entry_2 * de;
994 struct buffer_head * bh;
995
996 if (dentry->d_name.len > EXT3_NAME_LEN)
997 return ERR_PTR(-ENAMETOOLONG);
998
999 bh = ext3_find_entry(dentry, &de);
1000 inode = NULL;
1001 if (bh) {
1002 unsigned long ino = le32_to_cpu(de->inode);
1003 brelse (bh);
Neil Brown2ccb48e2006-07-30 03:03:01 -07001004 if (!ext3_valid_inum(dir->i_sb, ino)) {
1005 ext3_error(dir->i_sb, "ext3_lookup",
1006 "bad inode number: %lu", ino);
1007 inode = NULL;
1008 } else
1009 inode = iget(dir->i_sb, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
1011 if (!inode)
1012 return ERR_PTR(-EACCES);
1013 }
Pekka Enbergba7fe362006-01-14 13:21:08 -08001014 return d_splice_alias(inode, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015}
1016
1017
1018struct dentry *ext3_get_parent(struct dentry *child)
1019{
1020 unsigned long ino;
1021 struct dentry *parent;
1022 struct inode *inode;
1023 struct dentry dotdot;
1024 struct ext3_dir_entry_2 * de;
1025 struct buffer_head *bh;
1026
1027 dotdot.d_name.name = "..";
1028 dotdot.d_name.len = 2;
1029 dotdot.d_parent = child; /* confusing, isn't it! */
1030
1031 bh = ext3_find_entry(&dotdot, &de);
1032 inode = NULL;
1033 if (!bh)
1034 return ERR_PTR(-ENOENT);
1035 ino = le32_to_cpu(de->inode);
1036 brelse(bh);
Neil Brown2ccb48e2006-07-30 03:03:01 -07001037
1038 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1039 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1040 "bad inode number: %lu", ino);
1041 inode = NULL;
1042 } else
1043 inode = iget(child->d_inode->i_sb, ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 if (!inode)
1046 return ERR_PTR(-EACCES);
1047
1048 parent = d_alloc_anon(inode);
1049 if (!parent) {
1050 iput(inode);
1051 parent = ERR_PTR(-ENOMEM);
1052 }
1053 return parent;
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001054}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
1056#define S_SHIFT 12
1057static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1058 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1059 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1060 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1061 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1062 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1063 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1064 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1065};
1066
1067static inline void ext3_set_de_type(struct super_block *sb,
1068 struct ext3_dir_entry_2 *de,
1069 umode_t mode) {
1070 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1071 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1072}
1073
1074#ifdef CONFIG_EXT3_INDEX
1075static struct ext3_dir_entry_2 *
1076dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1077{
1078 unsigned rec_len = 0;
1079
1080 while (count--) {
1081 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1082 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1083 memcpy (to, de, rec_len);
1084 ((struct ext3_dir_entry_2 *) to)->rec_len =
1085 cpu_to_le16(rec_len);
1086 de->inode = 0;
1087 map++;
1088 to += rec_len;
1089 }
1090 return (struct ext3_dir_entry_2 *) (to - rec_len);
1091}
1092
1093static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1094{
1095 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1096 unsigned rec_len = 0;
1097
1098 prev = to = de;
1099 while ((char*)de < base + size) {
1100 next = (struct ext3_dir_entry_2 *) ((char *) de +
1101 le16_to_cpu(de->rec_len));
1102 if (de->inode && de->name_len) {
1103 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1104 if (de > to)
1105 memmove(to, de, rec_len);
1106 to->rec_len = cpu_to_le16(rec_len);
1107 prev = to;
1108 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1109 }
1110 de = next;
1111 }
1112 return prev;
1113}
1114
1115static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1116 struct buffer_head **bh,struct dx_frame *frame,
1117 struct dx_hash_info *hinfo, int *error)
1118{
1119 unsigned blocksize = dir->i_sb->s_blocksize;
1120 unsigned count, continued;
1121 struct buffer_head *bh2;
1122 u32 newblock;
1123 u32 hash2;
1124 struct dx_map_entry *map;
1125 char *data1 = (*bh)->b_data, *data2;
1126 unsigned split;
1127 struct ext3_dir_entry_2 *de = NULL, *de2;
1128 int err;
1129
1130 bh2 = ext3_append (handle, dir, &newblock, error);
1131 if (!(bh2)) {
1132 brelse(*bh);
1133 *bh = NULL;
1134 goto errout;
1135 }
1136
1137 BUFFER_TRACE(*bh, "get_write_access");
1138 err = ext3_journal_get_write_access(handle, *bh);
1139 if (err) {
1140 journal_error:
1141 brelse(*bh);
1142 brelse(bh2);
1143 *bh = NULL;
1144 ext3_std_error(dir->i_sb, err);
1145 goto errout;
1146 }
1147 BUFFER_TRACE(frame->bh, "get_write_access");
1148 err = ext3_journal_get_write_access(handle, frame->bh);
1149 if (err)
1150 goto journal_error;
1151
1152 data2 = bh2->b_data;
1153
1154 /* create map in the end of data2 block */
1155 map = (struct dx_map_entry *) (data2 + blocksize);
1156 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1157 blocksize, hinfo, map);
1158 map -= count;
1159 split = count/2; // need to adjust to actual middle
1160 dx_sort_map (map, count);
1161 hash2 = map[split].hash;
1162 continued = hash2 == map[split - 1].hash;
1163 dxtrace(printk("Split block %i at %x, %i/%i\n",
1164 dx_get_block(frame->at), hash2, split, count-split));
1165
1166 /* Fancy dance to stay within two buffers */
1167 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1168 de = dx_pack_dirents(data1,blocksize);
1169 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1170 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
1171 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1172 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1173
1174 /* Which block gets the new entry? */
1175 if (hinfo->hash >= hash2)
1176 {
1177 swap(*bh, bh2);
1178 de = de2;
1179 }
1180 dx_insert_block (frame, hash2 + continued, newblock);
1181 err = ext3_journal_dirty_metadata (handle, bh2);
1182 if (err)
1183 goto journal_error;
1184 err = ext3_journal_dirty_metadata (handle, frame->bh);
1185 if (err)
1186 goto journal_error;
1187 brelse (bh2);
1188 dxtrace(dx_show_index ("frame", frame->entries));
1189errout:
1190 return de;
1191}
1192#endif
1193
1194
1195/*
1196 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1197 * it points to a directory entry which is guaranteed to be large
1198 * enough for new directory entry. If de is NULL, then
1199 * add_dirent_to_buf will attempt search the directory block for
1200 * space. It will return -ENOSPC if no space is available, and -EIO
1201 * and -EEXIST if directory entry already exists.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001202 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1204 * all other cases bh is released.
1205 */
1206static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1207 struct inode *inode, struct ext3_dir_entry_2 *de,
1208 struct buffer_head * bh)
1209{
1210 struct inode *dir = dentry->d_parent->d_inode;
1211 const char *name = dentry->d_name.name;
1212 int namelen = dentry->d_name.len;
1213 unsigned long offset = 0;
1214 unsigned short reclen;
1215 int nlen, rlen, err;
1216 char *top;
1217
1218 reclen = EXT3_DIR_REC_LEN(namelen);
1219 if (!de) {
1220 de = (struct ext3_dir_entry_2 *)bh->b_data;
1221 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1222 while ((char *) de <= top) {
1223 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1224 bh, offset)) {
1225 brelse (bh);
1226 return -EIO;
1227 }
1228 if (ext3_match (namelen, name, de)) {
1229 brelse (bh);
1230 return -EEXIST;
1231 }
1232 nlen = EXT3_DIR_REC_LEN(de->name_len);
1233 rlen = le16_to_cpu(de->rec_len);
1234 if ((de->inode? rlen - nlen: rlen) >= reclen)
1235 break;
1236 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1237 offset += rlen;
1238 }
1239 if ((char *) de > top)
1240 return -ENOSPC;
1241 }
1242 BUFFER_TRACE(bh, "get_write_access");
1243 err = ext3_journal_get_write_access(handle, bh);
1244 if (err) {
1245 ext3_std_error(dir->i_sb, err);
1246 brelse(bh);
1247 return err;
1248 }
1249
1250 /* By now the buffer is marked for journaling */
1251 nlen = EXT3_DIR_REC_LEN(de->name_len);
1252 rlen = le16_to_cpu(de->rec_len);
1253 if (de->inode) {
1254 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
1255 de1->rec_len = cpu_to_le16(rlen - nlen);
1256 de->rec_len = cpu_to_le16(nlen);
1257 de = de1;
1258 }
1259 de->file_type = EXT3_FT_UNKNOWN;
1260 if (inode) {
1261 de->inode = cpu_to_le32(inode->i_ino);
1262 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1263 } else
1264 de->inode = 0;
1265 de->name_len = namelen;
1266 memcpy (de->name, name, namelen);
1267 /*
1268 * XXX shouldn't update any times until successful
1269 * completion of syscall, but too many callers depend
1270 * on this.
1271 *
1272 * XXX similarly, too many callers depend on
1273 * ext3_new_inode() setting the times, but error
1274 * recovery deletes the inode, so the worst that can
1275 * happen is that the times are slightly out of date
1276 * and/or different from the directory change time.
1277 */
1278 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1279 ext3_update_dx_flag(dir);
1280 dir->i_version++;
1281 ext3_mark_inode_dirty(handle, dir);
1282 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1283 err = ext3_journal_dirty_metadata(handle, bh);
1284 if (err)
1285 ext3_std_error(dir->i_sb, err);
1286 brelse(bh);
1287 return 0;
1288}
1289
1290#ifdef CONFIG_EXT3_INDEX
1291/*
1292 * This converts a one block unindexed directory to a 3 block indexed
1293 * directory, and adds the dentry to the indexed directory.
1294 */
1295static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1296 struct inode *inode, struct buffer_head *bh)
1297{
1298 struct inode *dir = dentry->d_parent->d_inode;
1299 const char *name = dentry->d_name.name;
1300 int namelen = dentry->d_name.len;
1301 struct buffer_head *bh2;
1302 struct dx_root *root;
1303 struct dx_frame frames[2], *frame;
1304 struct dx_entry *entries;
1305 struct ext3_dir_entry_2 *de, *de2;
1306 char *data1, *top;
1307 unsigned len;
1308 int retval;
1309 unsigned blocksize;
1310 struct dx_hash_info hinfo;
1311 u32 block;
1312 struct fake_dirent *fde;
1313
1314 blocksize = dir->i_sb->s_blocksize;
1315 dxtrace(printk("Creating index\n"));
1316 retval = ext3_journal_get_write_access(handle, bh);
1317 if (retval) {
1318 ext3_std_error(dir->i_sb, retval);
1319 brelse(bh);
1320 return retval;
1321 }
1322 root = (struct dx_root *) bh->b_data;
1323
1324 bh2 = ext3_append (handle, dir, &block, &retval);
1325 if (!(bh2)) {
1326 brelse(bh);
1327 return retval;
1328 }
1329 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1330 data1 = bh2->b_data;
1331
1332 /* The 0th block becomes the root, move the dirents out */
1333 fde = &root->dotdot;
1334 de = (struct ext3_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
1335 len = ((char *) root) + blocksize - (char *) de;
1336 memcpy (data1, de, len);
1337 de = (struct ext3_dir_entry_2 *) data1;
1338 top = data1 + len;
1339 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1340 de = de2;
1341 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1342 /* Initialize the root; the dot dirents already exist */
1343 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
1344 de->rec_len = cpu_to_le16(blocksize - EXT3_DIR_REC_LEN(2));
1345 memset (&root->info, 0, sizeof(root->info));
1346 root->info.info_length = sizeof(root->info);
1347 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1348 entries = root->entries;
1349 dx_set_block (entries, 1);
1350 dx_set_count (entries, 1);
1351 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1352
1353 /* Initialize as for dx_probe */
1354 hinfo.hash_version = root->info.hash_version;
1355 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1356 ext3fs_dirhash(name, namelen, &hinfo);
1357 frame = frames;
1358 frame->entries = entries;
1359 frame->at = entries;
1360 frame->bh = bh;
1361 bh = bh2;
1362 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1363 dx_release (frames);
1364 if (!(de))
1365 return retval;
1366
1367 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1368}
1369#endif
1370
1371/*
1372 * ext3_add_entry()
1373 *
1374 * adds a file entry to the specified directory, using the same
1375 * semantics as ext3_find_entry(). It returns NULL if it failed.
1376 *
1377 * NOTE!! The inode part of 'de' is left at 0 - which means you
1378 * may not sleep between calling this and putting something into
1379 * the entry, as someone else might have used it while you slept.
1380 */
1381static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1382 struct inode *inode)
1383{
1384 struct inode *dir = dentry->d_parent->d_inode;
1385 unsigned long offset;
1386 struct buffer_head * bh;
1387 struct ext3_dir_entry_2 *de;
1388 struct super_block * sb;
1389 int retval;
1390#ifdef CONFIG_EXT3_INDEX
1391 int dx_fallback=0;
1392#endif
1393 unsigned blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 u32 block, blocks;
1395
1396 sb = dir->i_sb;
1397 blocksize = sb->s_blocksize;
1398 if (!dentry->d_name.len)
1399 return -EINVAL;
1400#ifdef CONFIG_EXT3_INDEX
1401 if (is_dx(dir)) {
1402 retval = ext3_dx_add_entry(handle, dentry, inode);
1403 if (!retval || (retval != ERR_BAD_DX_DIR))
1404 return retval;
1405 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1406 dx_fallback++;
1407 ext3_mark_inode_dirty(handle, dir);
1408 }
1409#endif
1410 blocks = dir->i_size >> sb->s_blocksize_bits;
1411 for (block = 0, offset = 0; block < blocks; block++) {
1412 bh = ext3_bread(handle, dir, block, 0, &retval);
1413 if(!bh)
1414 return retval;
1415 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1416 if (retval != -ENOSPC)
1417 return retval;
1418
1419#ifdef CONFIG_EXT3_INDEX
1420 if (blocks == 1 && !dx_fallback &&
1421 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1422 return make_indexed_dir(handle, dentry, inode, bh);
1423#endif
1424 brelse(bh);
1425 }
1426 bh = ext3_append(handle, dir, &block, &retval);
1427 if (!bh)
1428 return retval;
1429 de = (struct ext3_dir_entry_2 *) bh->b_data;
1430 de->inode = 0;
Johann Lombardi92eeccd2006-06-25 05:48:33 -07001431 de->rec_len = cpu_to_le16(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1433}
1434
1435#ifdef CONFIG_EXT3_INDEX
1436/*
1437 * Returns 0 for success, or a negative error value
1438 */
1439static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1440 struct inode *inode)
1441{
1442 struct dx_frame frames[2], *frame;
1443 struct dx_entry *entries, *at;
1444 struct dx_hash_info hinfo;
1445 struct buffer_head * bh;
1446 struct inode *dir = dentry->d_parent->d_inode;
1447 struct super_block * sb = dir->i_sb;
1448 struct ext3_dir_entry_2 *de;
1449 int err;
1450
1451 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1452 if (!frame)
1453 return err;
1454 entries = frame->entries;
1455 at = frame->at;
1456
1457 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1458 goto cleanup;
1459
1460 BUFFER_TRACE(bh, "get_write_access");
1461 err = ext3_journal_get_write_access(handle, bh);
1462 if (err)
1463 goto journal_error;
1464
1465 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1466 if (err != -ENOSPC) {
1467 bh = NULL;
1468 goto cleanup;
1469 }
1470
1471 /* Block full, should compress but for now just split */
1472 dxtrace(printk("using %u of %u node entries\n",
1473 dx_get_count(entries), dx_get_limit(entries)));
1474 /* Need to split index? */
1475 if (dx_get_count(entries) == dx_get_limit(entries)) {
1476 u32 newblock;
1477 unsigned icount = dx_get_count(entries);
1478 int levels = frame - frames;
1479 struct dx_entry *entries2;
1480 struct dx_node *node2;
1481 struct buffer_head *bh2;
1482
1483 if (levels && (dx_get_count(frames->entries) ==
1484 dx_get_limit(frames->entries))) {
1485 ext3_warning(sb, __FUNCTION__,
Glauber de Oliveira Costa9f406682006-01-08 01:03:22 -08001486 "Directory index full!");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 err = -ENOSPC;
1488 goto cleanup;
1489 }
1490 bh2 = ext3_append (handle, dir, &newblock, &err);
1491 if (!(bh2))
1492 goto cleanup;
1493 node2 = (struct dx_node *)(bh2->b_data);
1494 entries2 = node2->entries;
1495 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1496 node2->fake.inode = 0;
1497 BUFFER_TRACE(frame->bh, "get_write_access");
1498 err = ext3_journal_get_write_access(handle, frame->bh);
1499 if (err)
1500 goto journal_error;
1501 if (levels) {
1502 unsigned icount1 = icount/2, icount2 = icount - icount1;
1503 unsigned hash2 = dx_get_hash(entries + icount1);
1504 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1505
1506 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1507 err = ext3_journal_get_write_access(handle,
1508 frames[0].bh);
1509 if (err)
1510 goto journal_error;
1511
1512 memcpy ((char *) entries2, (char *) (entries + icount1),
1513 icount2 * sizeof(struct dx_entry));
1514 dx_set_count (entries, icount1);
1515 dx_set_count (entries2, icount2);
1516 dx_set_limit (entries2, dx_node_limit(dir));
1517
1518 /* Which index block gets the new entry? */
1519 if (at - entries >= icount1) {
1520 frame->at = at = at - entries - icount1 + entries2;
1521 frame->entries = entries = entries2;
1522 swap(frame->bh, bh2);
1523 }
1524 dx_insert_block (frames + 0, hash2, newblock);
1525 dxtrace(dx_show_index ("node", frames[1].entries));
1526 dxtrace(dx_show_index ("node",
1527 ((struct dx_node *) bh2->b_data)->entries));
1528 err = ext3_journal_dirty_metadata(handle, bh2);
1529 if (err)
1530 goto journal_error;
1531 brelse (bh2);
1532 } else {
1533 dxtrace(printk("Creating second level index...\n"));
1534 memcpy((char *) entries2, (char *) entries,
1535 icount * sizeof(struct dx_entry));
1536 dx_set_limit(entries2, dx_node_limit(dir));
1537
1538 /* Set up root */
1539 dx_set_count(entries, 1);
1540 dx_set_block(entries + 0, newblock);
1541 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1542
1543 /* Add new access path frame */
1544 frame = frames + 1;
1545 frame->at = at = at - entries + entries2;
1546 frame->entries = entries = entries2;
1547 frame->bh = bh2;
1548 err = ext3_journal_get_write_access(handle,
1549 frame->bh);
1550 if (err)
1551 goto journal_error;
1552 }
1553 ext3_journal_dirty_metadata(handle, frames[0].bh);
1554 }
1555 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1556 if (!de)
1557 goto cleanup;
1558 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1559 bh = NULL;
1560 goto cleanup;
1561
1562journal_error:
1563 ext3_std_error(dir->i_sb, err);
1564cleanup:
1565 if (bh)
1566 brelse(bh);
1567 dx_release(frames);
1568 return err;
1569}
1570#endif
1571
1572/*
1573 * ext3_delete_entry deletes a directory entry by merging it with the
1574 * previous entry
1575 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001576static int ext3_delete_entry (handle_t *handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 struct inode * dir,
1578 struct ext3_dir_entry_2 * de_del,
1579 struct buffer_head * bh)
1580{
1581 struct ext3_dir_entry_2 * de, * pde;
1582 int i;
1583
1584 i = 0;
1585 pde = NULL;
1586 de = (struct ext3_dir_entry_2 *) bh->b_data;
1587 while (i < bh->b_size) {
1588 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1589 return -EIO;
1590 if (de == de_del) {
1591 BUFFER_TRACE(bh, "get_write_access");
1592 ext3_journal_get_write_access(handle, bh);
1593 if (pde)
1594 pde->rec_len =
1595 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1596 le16_to_cpu(de->rec_len));
1597 else
1598 de->inode = 0;
1599 dir->i_version++;
1600 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1601 ext3_journal_dirty_metadata(handle, bh);
1602 return 0;
1603 }
1604 i += le16_to_cpu(de->rec_len);
1605 pde = de;
1606 de = (struct ext3_dir_entry_2 *)
1607 ((char *) de + le16_to_cpu(de->rec_len));
1608 }
1609 return -ENOENT;
1610}
1611
1612/*
1613 * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
1614 * do not perform it in these functions. We perform it at the call site,
1615 * if it is needed.
1616 */
1617static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
1618{
Dave Hansend8c76e62006-09-30 23:29:04 -07001619 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620}
1621
1622static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
1623{
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001624 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
1626
1627static int ext3_add_nondir(handle_t *handle,
1628 struct dentry *dentry, struct inode *inode)
1629{
1630 int err = ext3_add_entry(handle, dentry, inode);
1631 if (!err) {
1632 ext3_mark_inode_dirty(handle, inode);
1633 d_instantiate(dentry, inode);
1634 return 0;
1635 }
1636 ext3_dec_count(handle, inode);
1637 iput(inode);
1638 return err;
1639}
1640
1641/*
1642 * By the time this is called, we already have created
1643 * the directory cache entry for the new file, but it
1644 * is so far negative - it has no inode.
1645 *
1646 * If the create succeeds, we fill in the inode information
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001647 * with d_instantiate().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 */
1649static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1650 struct nameidata *nd)
1651{
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001652 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 struct inode * inode;
1654 int err, retries = 0;
1655
1656retry:
Jan Kara1f545872005-06-23 22:01:04 -07001657 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001659 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 if (IS_ERR(handle))
1661 return PTR_ERR(handle);
1662
1663 if (IS_DIRSYNC(dir))
1664 handle->h_sync = 1;
1665
1666 inode = ext3_new_inode (handle, dir, mode);
1667 err = PTR_ERR(inode);
1668 if (!IS_ERR(inode)) {
1669 inode->i_op = &ext3_file_inode_operations;
1670 inode->i_fop = &ext3_file_operations;
1671 ext3_set_aops(inode);
1672 err = ext3_add_nondir(handle, dentry, inode);
1673 }
1674 ext3_journal_stop(handle);
1675 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1676 goto retry;
1677 return err;
1678}
1679
1680static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1681 int mode, dev_t rdev)
1682{
1683 handle_t *handle;
1684 struct inode *inode;
1685 int err, retries = 0;
1686
1687 if (!new_valid_dev(rdev))
1688 return -EINVAL;
1689
1690retry:
Jan Kara1f545872005-06-23 22:01:04 -07001691 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001692 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001693 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 if (IS_ERR(handle))
1695 return PTR_ERR(handle);
1696
1697 if (IS_DIRSYNC(dir))
1698 handle->h_sync = 1;
1699
1700 inode = ext3_new_inode (handle, dir, mode);
1701 err = PTR_ERR(inode);
1702 if (!IS_ERR(inode)) {
1703 init_special_inode(inode, inode->i_mode, rdev);
1704#ifdef CONFIG_EXT3_FS_XATTR
1705 inode->i_op = &ext3_special_inode_operations;
1706#endif
1707 err = ext3_add_nondir(handle, dentry, inode);
1708 }
1709 ext3_journal_stop(handle);
1710 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1711 goto retry;
1712 return err;
1713}
1714
1715static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1716{
1717 handle_t *handle;
1718 struct inode * inode;
1719 struct buffer_head * dir_block;
1720 struct ext3_dir_entry_2 * de;
1721 int err, retries = 0;
1722
1723 if (dir->i_nlink >= EXT3_LINK_MAX)
1724 return -EMLINK;
1725
1726retry:
Jan Kara1f545872005-06-23 22:01:04 -07001727 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001729 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 if (IS_ERR(handle))
1731 return PTR_ERR(handle);
1732
1733 if (IS_DIRSYNC(dir))
1734 handle->h_sync = 1;
1735
1736 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1737 err = PTR_ERR(inode);
1738 if (IS_ERR(inode))
1739 goto out_stop;
1740
1741 inode->i_op = &ext3_dir_inode_operations;
1742 inode->i_fop = &ext3_dir_operations;
1743 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1744 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1745 if (!dir_block) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001746 drop_nlink(inode); /* is this nlink == 0? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 ext3_mark_inode_dirty(handle, inode);
1748 iput (inode);
1749 goto out_stop;
1750 }
1751 BUFFER_TRACE(dir_block, "get_write_access");
1752 ext3_journal_get_write_access(handle, dir_block);
1753 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1754 de->inode = cpu_to_le32(inode->i_ino);
1755 de->name_len = 1;
1756 de->rec_len = cpu_to_le16(EXT3_DIR_REC_LEN(de->name_len));
1757 strcpy (de->name, ".");
1758 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1759 de = (struct ext3_dir_entry_2 *)
1760 ((char *) de + le16_to_cpu(de->rec_len));
1761 de->inode = cpu_to_le32(dir->i_ino);
1762 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT3_DIR_REC_LEN(1));
1763 de->name_len = 2;
1764 strcpy (de->name, "..");
1765 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1766 inode->i_nlink = 2;
1767 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1768 ext3_journal_dirty_metadata(handle, dir_block);
1769 brelse (dir_block);
1770 ext3_mark_inode_dirty(handle, inode);
1771 err = ext3_add_entry (handle, dentry, inode);
1772 if (err) {
1773 inode->i_nlink = 0;
1774 ext3_mark_inode_dirty(handle, inode);
1775 iput (inode);
1776 goto out_stop;
1777 }
Dave Hansend8c76e62006-09-30 23:29:04 -07001778 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 ext3_update_dx_flag(dir);
1780 ext3_mark_inode_dirty(handle, dir);
1781 d_instantiate(dentry, inode);
1782out_stop:
1783 ext3_journal_stop(handle);
1784 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1785 goto retry;
1786 return err;
1787}
1788
1789/*
1790 * routine to check that the specified directory is empty (for rmdir)
1791 */
1792static int empty_dir (struct inode * inode)
1793{
1794 unsigned long offset;
1795 struct buffer_head * bh;
1796 struct ext3_dir_entry_2 * de, * de1;
1797 struct super_block * sb;
1798 int err = 0;
1799
1800 sb = inode->i_sb;
1801 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1802 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1803 if (err)
1804 ext3_error(inode->i_sb, __FUNCTION__,
1805 "error %d reading directory #%lu offset 0",
1806 err, inode->i_ino);
1807 else
1808 ext3_warning(inode->i_sb, __FUNCTION__,
1809 "bad directory (dir #%lu) - no data block",
1810 inode->i_ino);
1811 return 1;
1812 }
1813 de = (struct ext3_dir_entry_2 *) bh->b_data;
1814 de1 = (struct ext3_dir_entry_2 *)
1815 ((char *) de + le16_to_cpu(de->rec_len));
1816 if (le32_to_cpu(de->inode) != inode->i_ino ||
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001817 !le32_to_cpu(de1->inode) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 strcmp (".", de->name) ||
1819 strcmp ("..", de1->name)) {
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001820 ext3_warning (inode->i_sb, "empty_dir",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 "bad directory (dir #%lu) - no `.' or `..'",
1822 inode->i_ino);
1823 brelse (bh);
1824 return 1;
1825 }
1826 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
1827 de = (struct ext3_dir_entry_2 *)
1828 ((char *) de1 + le16_to_cpu(de1->rec_len));
1829 while (offset < inode->i_size ) {
1830 if (!bh ||
1831 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1832 err = 0;
1833 brelse (bh);
1834 bh = ext3_bread (NULL, inode,
1835 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1836 if (!bh) {
1837 if (err)
1838 ext3_error(sb, __FUNCTION__,
1839 "error %d reading directory"
1840 " #%lu offset %lu",
1841 err, inode->i_ino, offset);
1842 offset += sb->s_blocksize;
1843 continue;
1844 }
1845 de = (struct ext3_dir_entry_2 *) bh->b_data;
1846 }
1847 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1848 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1849 sb->s_blocksize);
1850 offset = (offset | (sb->s_blocksize - 1)) + 1;
1851 continue;
1852 }
1853 if (le32_to_cpu(de->inode)) {
1854 brelse (bh);
1855 return 0;
1856 }
1857 offset += le16_to_cpu(de->rec_len);
1858 de = (struct ext3_dir_entry_2 *)
1859 ((char *) de + le16_to_cpu(de->rec_len));
1860 }
1861 brelse (bh);
1862 return 1;
1863}
1864
1865/* ext3_orphan_add() links an unlinked or truncated inode into a list of
1866 * such inodes, starting at the superblock, in case we crash before the
1867 * file is closed/deleted, or in case the inode truncate spans multiple
1868 * transactions and the last transaction is not recovered after a crash.
1869 *
1870 * At filesystem recovery time, we walk this list deleting unlinked
1871 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1872 */
1873int ext3_orphan_add(handle_t *handle, struct inode *inode)
1874{
1875 struct super_block *sb = inode->i_sb;
1876 struct ext3_iloc iloc;
1877 int err = 0, rc;
1878
1879 lock_super(sb);
1880 if (!list_empty(&EXT3_I(inode)->i_orphan))
1881 goto out_unlock;
1882
1883 /* Orphan handling is only valid for files with data blocks
1884 * being truncated, or files being unlinked. */
1885
1886 /* @@@ FIXME: Observation from aviro:
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001887 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 * here (on lock_super()), so race with ext3_link() which might bump
1889 * ->i_nlink. For, say it, character device. Not a regular file,
1890 * not a directory, not a symlink and ->i_nlink > 0.
1891 */
1892 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1893 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1894
1895 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1896 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1897 if (err)
1898 goto out_unlock;
1899
1900 err = ext3_reserve_inode_write(handle, inode, &iloc);
1901 if (err)
1902 goto out_unlock;
1903
1904 /* Insert this inode at the head of the on-disk orphan list... */
1905 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1906 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1907 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1908 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1909 if (!err)
1910 err = rc;
1911
1912 /* Only add to the head of the in-memory list if all the
1913 * previous operations succeeded. If the orphan_add is going to
1914 * fail (possibly taking the journal offline), we can't risk
1915 * leaving the inode on the orphan list: stray orphan-list
1916 * entries can cause panics at unmount time.
1917 *
1918 * This is safe: on error we're going to ignore the orphan list
1919 * anyway on the next recovery. */
1920 if (!err)
1921 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1922
Eric Sandeeneee194e2006-09-27 01:49:30 -07001923 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1924 jbd_debug(4, "orphan inode %lu will point to %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 inode->i_ino, NEXT_ORPHAN(inode));
1926out_unlock:
1927 unlock_super(sb);
1928 ext3_std_error(inode->i_sb, err);
1929 return err;
1930}
1931
1932/*
1933 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1934 * of such inodes stored on disk, because it is finally being cleaned up.
1935 */
1936int ext3_orphan_del(handle_t *handle, struct inode *inode)
1937{
1938 struct list_head *prev;
1939 struct ext3_inode_info *ei = EXT3_I(inode);
1940 struct ext3_sb_info *sbi;
1941 unsigned long ino_next;
1942 struct ext3_iloc iloc;
1943 int err = 0;
1944
1945 lock_super(inode->i_sb);
1946 if (list_empty(&ei->i_orphan)) {
1947 unlock_super(inode->i_sb);
1948 return 0;
1949 }
1950
1951 ino_next = NEXT_ORPHAN(inode);
1952 prev = ei->i_orphan.prev;
1953 sbi = EXT3_SB(inode->i_sb);
1954
1955 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1956
1957 list_del_init(&ei->i_orphan);
1958
1959 /* If we're on an error path, we may not have a valid
1960 * transaction handle with which to update the orphan list on
1961 * disk, but we still need to remove the inode from the linked
1962 * list in memory. */
1963 if (!handle)
1964 goto out;
1965
1966 err = ext3_reserve_inode_write(handle, inode, &iloc);
1967 if (err)
1968 goto out_err;
1969
1970 if (prev == &sbi->s_orphan) {
1971 jbd_debug(4, "superblock will point to %lu\n", ino_next);
1972 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1973 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
1974 if (err)
1975 goto out_brelse;
1976 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
1977 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
1978 } else {
1979 struct ext3_iloc iloc2;
1980 struct inode *i_prev =
1981 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
1982
1983 jbd_debug(4, "orphan inode %lu will point to %lu\n",
1984 i_prev->i_ino, ino_next);
1985 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
1986 if (err)
1987 goto out_brelse;
1988 NEXT_ORPHAN(i_prev) = ino_next;
1989 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
1990 }
1991 if (err)
1992 goto out_brelse;
1993 NEXT_ORPHAN(inode) = 0;
1994 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
1995
1996out_err:
1997 ext3_std_error(inode->i_sb, err);
1998out:
1999 unlock_super(inode->i_sb);
2000 return err;
2001
2002out_brelse:
2003 brelse(iloc.bh);
2004 goto out_err;
2005}
2006
2007static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2008{
2009 int retval;
2010 struct inode * inode;
2011 struct buffer_head * bh;
2012 struct ext3_dir_entry_2 * de;
2013 handle_t *handle;
2014
2015 /* Initialize quotas before so that eventual writes go in
2016 * separate transaction */
2017 DQUOT_INIT(dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002018 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 if (IS_ERR(handle))
2020 return PTR_ERR(handle);
2021
2022 retval = -ENOENT;
2023 bh = ext3_find_entry (dentry, &de);
2024 if (!bh)
2025 goto end_rmdir;
2026
2027 if (IS_DIRSYNC(dir))
2028 handle->h_sync = 1;
2029
2030 inode = dentry->d_inode;
2031
2032 retval = -EIO;
2033 if (le32_to_cpu(de->inode) != inode->i_ino)
2034 goto end_rmdir;
2035
2036 retval = -ENOTEMPTY;
2037 if (!empty_dir (inode))
2038 goto end_rmdir;
2039
2040 retval = ext3_delete_entry(handle, dir, de, bh);
2041 if (retval)
2042 goto end_rmdir;
2043 if (inode->i_nlink != 2)
2044 ext3_warning (inode->i_sb, "ext3_rmdir",
2045 "empty directory has nlink!=2 (%d)",
2046 inode->i_nlink);
2047 inode->i_version++;
Dave Hansence71ec32006-09-30 23:29:06 -07002048 clear_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 /* There's no need to set i_disksize: the fact that i_nlink is
2050 * zero will ensure that the right thing happens during any
2051 * recovery. */
2052 inode->i_size = 0;
2053 ext3_orphan_add(handle, inode);
2054 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2055 ext3_mark_inode_dirty(handle, inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002056 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 ext3_update_dx_flag(dir);
2058 ext3_mark_inode_dirty(handle, dir);
2059
2060end_rmdir:
2061 ext3_journal_stop(handle);
2062 brelse (bh);
2063 return retval;
2064}
2065
2066static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2067{
2068 int retval;
2069 struct inode * inode;
2070 struct buffer_head * bh;
2071 struct ext3_dir_entry_2 * de;
2072 handle_t *handle;
2073
2074 /* Initialize quotas before so that eventual writes go
2075 * in separate transaction */
2076 DQUOT_INIT(dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002077 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 if (IS_ERR(handle))
2079 return PTR_ERR(handle);
2080
2081 if (IS_DIRSYNC(dir))
2082 handle->h_sync = 1;
2083
2084 retval = -ENOENT;
2085 bh = ext3_find_entry (dentry, &de);
2086 if (!bh)
2087 goto end_unlink;
2088
2089 inode = dentry->d_inode;
2090
2091 retval = -EIO;
2092 if (le32_to_cpu(de->inode) != inode->i_ino)
2093 goto end_unlink;
2094
2095 if (!inode->i_nlink) {
2096 ext3_warning (inode->i_sb, "ext3_unlink",
2097 "Deleting nonexistent file (%lu), %d",
2098 inode->i_ino, inode->i_nlink);
2099 inode->i_nlink = 1;
2100 }
2101 retval = ext3_delete_entry(handle, dir, de, bh);
2102 if (retval)
2103 goto end_unlink;
2104 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2105 ext3_update_dx_flag(dir);
2106 ext3_mark_inode_dirty(handle, dir);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002107 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 if (!inode->i_nlink)
2109 ext3_orphan_add(handle, inode);
2110 inode->i_ctime = dir->i_ctime;
2111 ext3_mark_inode_dirty(handle, inode);
2112 retval = 0;
2113
2114end_unlink:
2115 ext3_journal_stop(handle);
2116 brelse (bh);
2117 return retval;
2118}
2119
2120static int ext3_symlink (struct inode * dir,
2121 struct dentry *dentry, const char * symname)
2122{
2123 handle_t *handle;
2124 struct inode * inode;
2125 int l, err, retries = 0;
2126
2127 l = strlen(symname)+1;
2128 if (l > dir->i_sb->s_blocksize)
2129 return -ENAMETOOLONG;
2130
2131retry:
Jan Kara1f545872005-06-23 22:01:04 -07002132 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07002133 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
Jan Kara1f545872005-06-23 22:01:04 -07002134 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 if (IS_ERR(handle))
2136 return PTR_ERR(handle);
2137
2138 if (IS_DIRSYNC(dir))
2139 handle->h_sync = 1;
2140
2141 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2142 err = PTR_ERR(inode);
2143 if (IS_ERR(inode))
2144 goto out_stop;
2145
2146 if (l > sizeof (EXT3_I(inode)->i_data)) {
2147 inode->i_op = &ext3_symlink_inode_operations;
2148 ext3_set_aops(inode);
2149 /*
2150 * page_symlink() calls into ext3_prepare/commit_write.
2151 * We have a transaction open. All is sweetness. It also sets
2152 * i_size in generic_commit_write().
2153 */
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002154 err = __page_symlink(inode, symname, l,
2155 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 if (err) {
2157 ext3_dec_count(handle, inode);
2158 ext3_mark_inode_dirty(handle, inode);
2159 iput (inode);
2160 goto out_stop;
2161 }
2162 } else {
2163 inode->i_op = &ext3_fast_symlink_inode_operations;
2164 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2165 inode->i_size = l-1;
2166 }
2167 EXT3_I(inode)->i_disksize = inode->i_size;
2168 err = ext3_add_nondir(handle, dentry, inode);
2169out_stop:
2170 ext3_journal_stop(handle);
2171 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2172 goto retry;
2173 return err;
2174}
2175
2176static int ext3_link (struct dentry * old_dentry,
2177 struct inode * dir, struct dentry *dentry)
2178{
2179 handle_t *handle;
2180 struct inode *inode = old_dentry->d_inode;
2181 int err, retries = 0;
2182
2183 if (inode->i_nlink >= EXT3_LINK_MAX)
2184 return -EMLINK;
2185
2186retry:
Jan Kara1f545872005-06-23 22:01:04 -07002187 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2189 if (IS_ERR(handle))
2190 return PTR_ERR(handle);
2191
2192 if (IS_DIRSYNC(dir))
2193 handle->h_sync = 1;
2194
2195 inode->i_ctime = CURRENT_TIME_SEC;
2196 ext3_inc_count(handle, inode);
2197 atomic_inc(&inode->i_count);
2198
2199 err = ext3_add_nondir(handle, dentry, inode);
2200 ext3_journal_stop(handle);
2201 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2202 goto retry;
2203 return err;
2204}
2205
2206#define PARENT_INO(buffer) \
2207 ((struct ext3_dir_entry_2 *) ((char *) buffer + \
2208 le16_to_cpu(((struct ext3_dir_entry_2 *) buffer)->rec_len)))->inode
2209
2210/*
2211 * Anybody can rename anything with this: the permission checks are left to the
2212 * higher-level routines.
2213 */
2214static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2215 struct inode * new_dir,struct dentry *new_dentry)
2216{
2217 handle_t *handle;
2218 struct inode * old_inode, * new_inode;
2219 struct buffer_head * old_bh, * new_bh, * dir_bh;
2220 struct ext3_dir_entry_2 * old_de, * new_de;
2221 int retval;
2222
2223 old_bh = new_bh = dir_bh = NULL;
2224
2225 /* Initialize quotas before so that eventual writes go
2226 * in separate transaction */
2227 if (new_dentry->d_inode)
2228 DQUOT_INIT(new_dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002229 handle = ext3_journal_start(old_dir, 2 *
2230 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07002231 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 if (IS_ERR(handle))
2233 return PTR_ERR(handle);
2234
2235 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2236 handle->h_sync = 1;
2237
2238 old_bh = ext3_find_entry (old_dentry, &old_de);
2239 /*
2240 * Check for inode number is _not_ due to possible IO errors.
2241 * We might rmdir the source, keep it as pwd of some process
2242 * and merrily kill the link to whatever was created under the
2243 * same name. Goodbye sticky bit ;-<
2244 */
2245 old_inode = old_dentry->d_inode;
2246 retval = -ENOENT;
2247 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2248 goto end_rename;
2249
2250 new_inode = new_dentry->d_inode;
2251 new_bh = ext3_find_entry (new_dentry, &new_de);
2252 if (new_bh) {
2253 if (!new_inode) {
2254 brelse (new_bh);
2255 new_bh = NULL;
2256 }
2257 }
2258 if (S_ISDIR(old_inode->i_mode)) {
2259 if (new_inode) {
2260 retval = -ENOTEMPTY;
2261 if (!empty_dir (new_inode))
2262 goto end_rename;
2263 }
2264 retval = -EIO;
2265 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2266 if (!dir_bh)
2267 goto end_rename;
2268 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2269 goto end_rename;
2270 retval = -EMLINK;
2271 if (!new_inode && new_dir!=old_dir &&
2272 new_dir->i_nlink >= EXT3_LINK_MAX)
2273 goto end_rename;
2274 }
2275 if (!new_bh) {
2276 retval = ext3_add_entry (handle, new_dentry, old_inode);
2277 if (retval)
2278 goto end_rename;
2279 } else {
2280 BUFFER_TRACE(new_bh, "get write access");
2281 ext3_journal_get_write_access(handle, new_bh);
2282 new_de->inode = cpu_to_le32(old_inode->i_ino);
2283 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2284 EXT3_FEATURE_INCOMPAT_FILETYPE))
2285 new_de->file_type = old_de->file_type;
2286 new_dir->i_version++;
2287 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2288 ext3_journal_dirty_metadata(handle, new_bh);
2289 brelse(new_bh);
2290 new_bh = NULL;
2291 }
2292
2293 /*
2294 * Like most other Unix systems, set the ctime for inodes on a
2295 * rename.
2296 */
2297 old_inode->i_ctime = CURRENT_TIME_SEC;
2298 ext3_mark_inode_dirty(handle, old_inode);
2299
2300 /*
2301 * ok, that's it
2302 */
2303 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2304 old_de->name_len != old_dentry->d_name.len ||
2305 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2306 (retval = ext3_delete_entry(handle, old_dir,
2307 old_de, old_bh)) == -ENOENT) {
2308 /* old_de could have moved from under us during htree split, so
2309 * make sure that we are deleting the right entry. We might
2310 * also be pointing to a stale entry in the unused part of
2311 * old_bh so just checking inum and the name isn't enough. */
2312 struct buffer_head *old_bh2;
2313 struct ext3_dir_entry_2 *old_de2;
2314
2315 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2316 if (old_bh2) {
2317 retval = ext3_delete_entry(handle, old_dir,
2318 old_de2, old_bh2);
2319 brelse(old_bh2);
2320 }
2321 }
2322 if (retval) {
2323 ext3_warning(old_dir->i_sb, "ext3_rename",
2324 "Deleting old file (%lu), %d, error=%d",
2325 old_dir->i_ino, old_dir->i_nlink, retval);
2326 }
2327
2328 if (new_inode) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002329 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 new_inode->i_ctime = CURRENT_TIME_SEC;
2331 }
2332 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2333 ext3_update_dx_flag(old_dir);
2334 if (dir_bh) {
2335 BUFFER_TRACE(dir_bh, "get_write_access");
2336 ext3_journal_get_write_access(handle, dir_bh);
2337 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2338 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2339 ext3_journal_dirty_metadata(handle, dir_bh);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002340 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 if (new_inode) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002342 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 } else {
Dave Hansend8c76e62006-09-30 23:29:04 -07002344 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 ext3_update_dx_flag(new_dir);
2346 ext3_mark_inode_dirty(handle, new_dir);
2347 }
2348 }
2349 ext3_mark_inode_dirty(handle, old_dir);
2350 if (new_inode) {
2351 ext3_mark_inode_dirty(handle, new_inode);
2352 if (!new_inode->i_nlink)
2353 ext3_orphan_add(handle, new_inode);
2354 }
2355 retval = 0;
2356
2357end_rename:
2358 brelse (dir_bh);
2359 brelse (old_bh);
2360 brelse (new_bh);
2361 ext3_journal_stop(handle);
2362 return retval;
2363}
2364
2365/*
2366 * directories can handle most operations...
2367 */
2368struct inode_operations ext3_dir_inode_operations = {
2369 .create = ext3_create,
2370 .lookup = ext3_lookup,
2371 .link = ext3_link,
2372 .unlink = ext3_unlink,
2373 .symlink = ext3_symlink,
2374 .mkdir = ext3_mkdir,
2375 .rmdir = ext3_rmdir,
2376 .mknod = ext3_mknod,
2377 .rename = ext3_rename,
2378 .setattr = ext3_setattr,
2379#ifdef CONFIG_EXT3_FS_XATTR
2380 .setxattr = generic_setxattr,
2381 .getxattr = generic_getxattr,
2382 .listxattr = ext3_listxattr,
2383 .removexattr = generic_removexattr,
2384#endif
2385 .permission = ext3_permission,
2386};
2387
2388struct inode_operations ext3_special_inode_operations = {
2389 .setattr = ext3_setattr,
2390#ifdef CONFIG_EXT3_FS_XATTR
2391 .setxattr = generic_setxattr,
2392 .getxattr = generic_getxattr,
2393 .listxattr = ext3_listxattr,
2394 .removexattr = generic_removexattr,
2395#endif
2396 .permission = ext3_permission,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002397};