?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ALSA: hda-intel: support for period wakeup disabling
[firewire-audio.git] / fs / ext4 / namei.c
blob92203b8a099f076ebbc0d794b3356289096c663f
1 /*
2 * linux/fs/ext4/namei.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
24 * Theodore Ts'o, 2002
25 */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/jbd2.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
39
40 #include "xattr.h"
41 #include "acl.h"
42
43 /*
44 * define how far ahead to read directories while searching them.
45 */
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
50
51 static struct buffer_head *ext4_append(handle_t *handle,
52 struct inode *inode,
53 ext4_lblk_t *block, int *err)
54 {
55 struct buffer_head *bh;
56
57 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
58
59 bh = ext4_bread(handle, inode, *block, 1, err);
60 if (bh) {
61 inode->i_size += inode->i_sb->s_blocksize;
62 EXT4_I(inode)->i_disksize = inode->i_size;
63 *err = ext4_journal_get_write_access(handle, bh);
64 if (*err) {
65 brelse(bh);
66 bh = NULL;
67 }
68 }
69 return bh;
70 }
71
72 #ifndef assert
73 #define assert(test) J_ASSERT(test)
74 #endif
75
76 #ifdef DX_DEBUG
77 #define dxtrace(command) command
78 #else
79 #define dxtrace(command)
80 #endif
81
82 struct fake_dirent
83 {
84 __le32 inode;
85 __le16 rec_len;
86 u8 name_len;
87 u8 file_type;
88 };
89
90 struct dx_countlimit
91 {
92 __le16 limit;
93 __le16 count;
94 };
95
96 struct 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
108 struct 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
126 struct dx_node
127 {
128 struct fake_dirent fake;
129 struct dx_entry entries[0];
130 };
131
132
133 struct dx_frame
134 {
135 struct buffer_head *bh;
136 struct dx_entry *entries;
137 struct dx_entry *at;
138 };
139
140 struct dx_map_entry
141 {
142 u32 hash;
143 u16 offs;
144 u16 size;
145 };
146
147 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
148 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
149 static inline unsigned dx_get_hash(struct dx_entry *entry);
150 static void dx_set_hash(struct dx_entry *entry, unsigned value);
151 static unsigned dx_get_count(struct dx_entry *entries);
152 static unsigned dx_get_limit(struct dx_entry *entries);
153 static void dx_set_count(struct dx_entry *entries, unsigned value);
154 static void dx_set_limit(struct dx_entry *entries, unsigned value);
155 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
156 static unsigned dx_node_limit(struct inode *dir);
157 static struct dx_frame *dx_probe(const struct qstr *d_name,
158 struct inode *dir,
159 struct dx_hash_info *hinfo,
160 struct dx_frame *frame,
161 int *err);
162 static void dx_release(struct dx_frame *frames);
163 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
164 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
166 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
167 struct dx_map_entry *offsets, int count, unsigned blocksize);
168 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
169 static void dx_insert_block(struct dx_frame *frame,
170 u32 hash, ext4_lblk_t block);
171 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
172 struct dx_frame *frame,
173 struct dx_frame *frames,
174 __u32 *start_hash);
175 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
176 const struct qstr *d_name,
177 struct ext4_dir_entry_2 **res_dir,
178 int *err);
179 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
180 struct inode *inode);
181
182 /*
183 * p is at least 6 bytes before the end of page
184 */
185 static inline struct ext4_dir_entry_2 *
186 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
187 {
188 return (struct ext4_dir_entry_2 *)((char *)p +
189 ext4_rec_len_from_disk(p->rec_len, blocksize));
190 }
191
192 /*
193 * Future: use high four bits of block for coalesce-on-delete flags
194 * Mask them off for now.
195 */
196
197 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
198 {
199 return le32_to_cpu(entry->block) & 0x00ffffff;
200 }
201
202 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
203 {
204 entry->block = cpu_to_le32(value);
205 }
206
207 static inline unsigned dx_get_hash(struct dx_entry *entry)
208 {
209 return le32_to_cpu(entry->hash);
210 }
211
212 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
213 {
214 entry->hash = cpu_to_le32(value);
215 }
216
217 static inline unsigned dx_get_count(struct dx_entry *entries)
218 {
219 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
220 }
221
222 static inline unsigned dx_get_limit(struct dx_entry *entries)
223 {
224 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
225 }
226
227 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
228 {
229 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
230 }
231
232 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
233 {
234 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
235 }
236
237 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
238 {
239 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
240 EXT4_DIR_REC_LEN(2) - infosize;
241 return entry_space / sizeof(struct dx_entry);
242 }
243
244 static inline unsigned dx_node_limit(struct inode *dir)
245 {
246 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
247 return entry_space / sizeof(struct dx_entry);
248 }
249
250 /*
251 * Debug
252 */
253 #ifdef DX_DEBUG
254 static void dx_show_index(char * label, struct dx_entry *entries)
255 {
256 int i, n = dx_get_count (entries);
257 printk(KERN_DEBUG "%s index ", label);
258 for (i = 0; i < n; i++) {
259 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
260 0, (unsigned long)dx_get_block(entries + i));
261 }
262 printk("\n");
263 }
264
265 struct stats
266 {
267 unsigned names;
268 unsigned space;
269 unsigned bcount;
270 };
271
272 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
273 int size, int show_names)
274 {
275 unsigned names = 0, space = 0;
276 char *base = (char *) de;
277 struct dx_hash_info h = *hinfo;
278
279 printk("names: ");
280 while ((char *) de < base + size)
281 {
282 if (de->inode)
283 {
284 if (show_names)
285 {
286 int len = de->name_len;
287 char *name = de->name;
288 while (len--) printk("%c", *name++);
289 ext4fs_dirhash(de->name, de->name_len, &h);
290 printk(":%x.%u ", h.hash,
291 ((char *) de - base));
292 }
293 space += EXT4_DIR_REC_LEN(de->name_len);
294 names++;
295 }
296 de = ext4_next_entry(de, size);
297 }
298 printk("(%i)\n", names);
299 return (struct stats) { names, space, 1 };
300 }
301
302 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
303 struct dx_entry *entries, int levels)
304 {
305 unsigned blocksize = dir->i_sb->s_blocksize;
306 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
307 unsigned bcount = 0;
308 struct buffer_head *bh;
309 int err;
310 printk("%i indexed blocks...\n", count);
311 for (i = 0; i < count; i++, entries++)
312 {
313 ext4_lblk_t block = dx_get_block(entries);
314 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
315 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
316 struct stats stats;
317 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
318 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
319 stats = levels?
320 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
321 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
322 names += stats.names;
323 space += stats.space;
324 bcount += stats.bcount;
325 brelse(bh);
326 }
327 if (bcount)
328 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
329 levels ? "" : " ", names, space/bcount,
330 (space/bcount)*100/blocksize);
331 return (struct stats) { names, space, bcount};
332 }
333 #endif /* DX_DEBUG */
334
335 /*
336 * Probe for a directory leaf block to search.
337 *
338 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
339 * error in the directory index, and the caller should fall back to
340 * searching the directory normally. The callers of dx_probe **MUST**
341 * check for this error code, and make sure it never gets reflected
342 * back to userspace.
343 */
344 static struct dx_frame *
345 dx_probe(const struct qstr *d_name, struct inode *dir,
346 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
347 {
348 unsigned count, indirect;
349 struct dx_entry *at, *entries, *p, *q, *m;
350 struct dx_root *root;
351 struct buffer_head *bh;
352 struct dx_frame *frame = frame_in;
353 u32 hash;
354
355 frame->bh = NULL;
356 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
357 goto fail;
358 root = (struct dx_root *) bh->b_data;
359 if (root->info.hash_version != DX_HASH_TEA &&
360 root->info.hash_version != DX_HASH_HALF_MD4 &&
361 root->info.hash_version != DX_HASH_LEGACY) {
362 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
363 root->info.hash_version);
364 brelse(bh);
365 *err = ERR_BAD_DX_DIR;
366 goto fail;
367 }
368 hinfo->hash_version = root->info.hash_version;
369 if (hinfo->hash_version <= DX_HASH_TEA)
370 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
371 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
372 if (d_name)
373 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
374 hash = hinfo->hash;
375
376 if (root->info.unused_flags & 1) {
377 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
378 root->info.unused_flags);
379 brelse(bh);
380 *err = ERR_BAD_DX_DIR;
381 goto fail;
382 }
383
384 if ((indirect = root->info.indirect_levels) > 1) {
385 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
386 root->info.indirect_levels);
387 brelse(bh);
388 *err = ERR_BAD_DX_DIR;
389 goto fail;
390 }
391
392 entries = (struct dx_entry *) (((char *)&root->info) +
393 root->info.info_length);
394
395 if (dx_get_limit(entries) != dx_root_limit(dir,
396 root->info.info_length)) {
397 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
398 brelse(bh);
399 *err = ERR_BAD_DX_DIR;
400 goto fail;
401 }
402
403 dxtrace(printk("Look up %x", hash));
404 while (1)
405 {
406 count = dx_get_count(entries);
407 if (!count || count > dx_get_limit(entries)) {
408 ext4_warning(dir->i_sb,
409 "dx entry: no count or count > limit");
410 brelse(bh);
411 *err = ERR_BAD_DX_DIR;
412 goto fail2;
413 }
414
415 p = entries + 1;
416 q = entries + count - 1;
417 while (p <= q)
418 {
419 m = p + (q - p)/2;
420 dxtrace(printk("."));
421 if (dx_get_hash(m) > hash)
422 q = m - 1;
423 else
424 p = m + 1;
425 }
426
427 if (0) // linear search cross check
428 {
429 unsigned n = count - 1;
430 at = entries;
431 while (n--)
432 {
433 dxtrace(printk(","));
434 if (dx_get_hash(++at) > hash)
435 {
436 at--;
437 break;
438 }
439 }
440 assert (at == p - 1);
441 }
442
443 at = p - 1;
444 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
445 frame->bh = bh;
446 frame->entries = entries;
447 frame->at = at;
448 if (!indirect--) return frame;
449 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
450 goto fail2;
451 at = entries = ((struct dx_node *) bh->b_data)->entries;
452 if (dx_get_limit(entries) != dx_node_limit (dir)) {
453 ext4_warning(dir->i_sb,
454 "dx entry: limit != node limit");
455 brelse(bh);
456 *err = ERR_BAD_DX_DIR;
457 goto fail2;
458 }
459 frame++;
460 frame->bh = NULL;
461 }
462 fail2:
463 while (frame >= frame_in) {
464 brelse(frame->bh);
465 frame--;
466 }
467 fail:
468 if (*err == ERR_BAD_DX_DIR)
469 ext4_warning(dir->i_sb,
470 "Corrupt dir inode %ld, running e2fsck is "
471 "recommended.", dir->i_ino);
472 return NULL;
473 }
474
475 static void dx_release (struct dx_frame *frames)
476 {
477 if (frames[0].bh == NULL)
478 return;
479
480 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
481 brelse(frames[1].bh);
482 brelse(frames[0].bh);
483 }
484
485 /*
486 * This function increments the frame pointer to search the next leaf
487 * block, and reads in the necessary intervening nodes if the search
488 * should be necessary. Whether or not the search is necessary is
489 * controlled by the hash parameter. If the hash value is even, then
490 * the search is only continued if the next block starts with that
491 * hash value. This is used if we are searching for a specific file.
492 *
493 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
494 *
495 * This function returns 1 if the caller should continue to search,
496 * or 0 if it should not. If there is an error reading one of the
497 * index blocks, it will a negative error code.
498 *
499 * If start_hash is non-null, it will be filled in with the starting
500 * hash of the next page.
501 */
502 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
503 struct dx_frame *frame,
504 struct dx_frame *frames,
505 __u32 *start_hash)
506 {
507 struct dx_frame *p;
508 struct buffer_head *bh;
509 int err, num_frames = 0;
510 __u32 bhash;
511
512 p = frame;
513 /*
514 * Find the next leaf page by incrementing the frame pointer.
515 * If we run out of entries in the interior node, loop around and
516 * increment pointer in the parent node. When we break out of
517 * this loop, num_frames indicates the number of interior
518 * nodes need to be read.
519 */
520 while (1) {
521 if (++(p->at) < p->entries + dx_get_count(p->entries))
522 break;
523 if (p == frames)
524 return 0;
525 num_frames++;
526 p--;
527 }
528
529 /*
530 * If the hash is 1, then continue only if the next page has a
531 * continuation hash of any value. This is used for readdir
532 * handling. Otherwise, check to see if the hash matches the
533 * desired contiuation hash. If it doesn't, return since
534 * there's no point to read in the successive index pages.
535 */
536 bhash = dx_get_hash(p->at);
537 if (start_hash)
538 *start_hash = bhash;
539 if ((hash & 1) == 0) {
540 if ((bhash & ~1) != hash)
541 return 0;
542 }
543 /*
544 * If the hash is HASH_NB_ALWAYS, we always go to the next
545 * block so no check is necessary
546 */
547 while (num_frames--) {
548 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
549 0, &err)))
550 return err; /* Failure */
551 p++;
552 brelse(p->bh);
553 p->bh = bh;
554 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
555 }
556 return 1;
557 }
558
559
560 /*
561 * This function fills a red-black tree with information from a
562 * directory block. It returns the number directory entries loaded
563 * into the tree. If there is an error it is returned in err.
564 */
565 static int htree_dirblock_to_tree(struct file *dir_file,
566 struct inode *dir, ext4_lblk_t block,
567 struct dx_hash_info *hinfo,
568 __u32 start_hash, __u32 start_minor_hash)
569 {
570 struct buffer_head *bh;
571 struct ext4_dir_entry_2 *de, *top;
572 int err, count = 0;
573
574 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
575 (unsigned long)block));
576 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
577 return err;
578
579 de = (struct ext4_dir_entry_2 *) bh->b_data;
580 top = (struct ext4_dir_entry_2 *) ((char *) de +
581 dir->i_sb->s_blocksize -
582 EXT4_DIR_REC_LEN(0));
583 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
584 if (!ext4_check_dir_entry(dir, de, bh,
585 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
586 +((char *)de - bh->b_data))) {
587 /* On error, skip the f_pos to the next block. */
588 dir_file->f_pos = (dir_file->f_pos |
589 (dir->i_sb->s_blocksize - 1)) + 1;
590 brelse(bh);
591 return count;
592 }
593 ext4fs_dirhash(de->name, de->name_len, hinfo);
594 if ((hinfo->hash < start_hash) ||
595 ((hinfo->hash == start_hash) &&
596 (hinfo->minor_hash < start_minor_hash)))
597 continue;
598 if (de->inode == 0)
599 continue;
600 if ((err = ext4_htree_store_dirent(dir_file,
601 hinfo->hash, hinfo->minor_hash, de)) != 0) {
602 brelse(bh);
603 return err;
604 }
605 count++;
606 }
607 brelse(bh);
608 return count;
609 }
610
611
612 /*
613 * This function fills a red-black tree with information from a
614 * directory. We start scanning the directory in hash order, starting
615 * at start_hash and start_minor_hash.
616 *
617 * This function returns the number of entries inserted into the tree,
618 * or a negative error code.
619 */
620 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
621 __u32 start_minor_hash, __u32 *next_hash)
622 {
623 struct dx_hash_info hinfo;
624 struct ext4_dir_entry_2 *de;
625 struct dx_frame frames[2], *frame;
626 struct inode *dir;
627 ext4_lblk_t block;
628 int count = 0;
629 int ret, err;
630 __u32 hashval;
631
632 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
633 start_hash, start_minor_hash));
634 dir = dir_file->f_path.dentry->d_inode;
635 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
636 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
637 if (hinfo.hash_version <= DX_HASH_TEA)
638 hinfo.hash_version +=
639 EXT4_SB(dir->i_sb)->s_hash_unsigned;
640 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
641 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
642 start_hash, start_minor_hash);
643 *next_hash = ~0;
644 return count;
645 }
646 hinfo.hash = start_hash;
647 hinfo.minor_hash = 0;
648 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
649 if (!frame)
650 return err;
651
652 /* Add '.' and '..' from the htree header */
653 if (!start_hash && !start_minor_hash) {
654 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
655 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
656 goto errout;
657 count++;
658 }
659 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
660 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
661 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
662 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
663 goto errout;
664 count++;
665 }
666
667 while (1) {
668 block = dx_get_block(frame->at);
669 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
670 start_hash, start_minor_hash);
671 if (ret < 0) {
672 err = ret;
673 goto errout;
674 }
675 count += ret;
676 hashval = ~0;
677 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
678 frame, frames, &hashval);
679 *next_hash = hashval;
680 if (ret < 0) {
681 err = ret;
682 goto errout;
683 }
684 /*
685 * Stop if: (a) there are no more entries, or
686 * (b) we have inserted at least one entry and the
687 * next hash value is not a continuation
688 */
689 if ((ret == 0) ||
690 (count && ((hashval & 1) == 0)))
691 break;
692 }
693 dx_release(frames);
694 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
695 "next hash: %x\n", count, *next_hash));
696 return count;
697 errout:
698 dx_release(frames);
699 return (err);
700 }
701
702
703 /*
704 * Directory block splitting, compacting
705 */
706
707 /*
708 * Create map of hash values, offsets, and sizes, stored at end of block.
709 * Returns number of entries mapped.
710 */
711 static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
712 struct dx_hash_info *hinfo,
713 struct dx_map_entry *map_tail)
714 {
715 int count = 0;
716 char *base = (char *) de;
717 struct dx_hash_info h = *hinfo;
718
719 while ((char *) de < base + blocksize) {
720 if (de->name_len && de->inode) {
721 ext4fs_dirhash(de->name, de->name_len, &h);
722 map_tail--;
723 map_tail->hash = h.hash;
724 map_tail->offs = ((char *) de - base)>>2;
725 map_tail->size = le16_to_cpu(de->rec_len);
726 count++;
727 cond_resched();
728 }
729 /* XXX: do we need to check rec_len == 0 case? -Chris */
730 de = ext4_next_entry(de, blocksize);
731 }
732 return count;
733 }
734
735 /* Sort map by hash value */
736 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
737 {
738 struct dx_map_entry *p, *q, *top = map + count - 1;
739 int more;
740 /* Combsort until bubble sort doesn't suck */
741 while (count > 2) {
742 count = count*10/13;
743 if (count - 9 < 2) /* 9, 10 -> 11 */
744 count = 11;
745 for (p = top, q = p - count; q >= map; p--, q--)
746 if (p->hash < q->hash)
747 swap(*p, *q);
748 }
749 /* Garden variety bubble sort */
750 do {
751 more = 0;
752 q = top;
753 while (q-- > map) {
754 if (q[1].hash >= q[0].hash)
755 continue;
756 swap(*(q+1), *q);
757 more = 1;
758 }
759 } while(more);
760 }
761
762 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
763 {
764 struct dx_entry *entries = frame->entries;
765 struct dx_entry *old = frame->at, *new = old + 1;
766 int count = dx_get_count(entries);
767
768 assert(count < dx_get_limit(entries));
769 assert(old < entries + count);
770 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
771 dx_set_hash(new, hash);
772 dx_set_block(new, block);
773 dx_set_count(entries, count + 1);
774 }
775
776 static void ext4_update_dx_flag(struct inode *inode)
777 {
778 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
779 EXT4_FEATURE_COMPAT_DIR_INDEX))
780 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
781 }
782
783 /*
784 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
785 *
786 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
787 * `de != NULL' is guaranteed by caller.
788 */
789 static inline int ext4_match (int len, const char * const name,
790 struct ext4_dir_entry_2 * de)
791 {
792 if (len != de->name_len)
793 return 0;
794 if (!de->inode)
795 return 0;
796 return !memcmp(name, de->name, len);
797 }
798
799 /*
800 * Returns 0 if not found, -1 on failure, and 1 on success
801 */
802 static inline int search_dirblock(struct buffer_head *bh,
803 struct inode *dir,
804 const struct qstr *d_name,
805 unsigned int offset,
806 struct ext4_dir_entry_2 ** res_dir)
807 {
808 struct ext4_dir_entry_2 * de;
809 char * dlimit;
810 int de_len;
811 const char *name = d_name->name;
812 int namelen = d_name->len;
813
814 de = (struct ext4_dir_entry_2 *) bh->b_data;
815 dlimit = bh->b_data + dir->i_sb->s_blocksize;
816 while ((char *) de < dlimit) {
817 /* this code is executed quadratically often */
818 /* do minimal checking `by hand' */
819
820 if ((char *) de + namelen <= dlimit &&
821 ext4_match (namelen, name, de)) {
822 /* found a match - just to be sure, do a full check */
823 if (!ext4_check_dir_entry(dir, de, bh, offset))
824 return -1;
825 *res_dir = de;
826 return 1;
827 }
828 /* prevent looping on a bad block */
829 de_len = ext4_rec_len_from_disk(de->rec_len,
830 dir->i_sb->s_blocksize);
831 if (de_len <= 0)
832 return -1;
833 offset += de_len;
834 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
835 }
836 return 0;
837 }
838
839
840 /*
841 * ext4_find_entry()
842 *
843 * finds an entry in the specified directory with the wanted name. It
844 * returns the cache buffer in which the entry was found, and the entry
845 * itself (as a parameter - res_dir). It does NOT read the inode of the
846 * entry - you'll have to do that yourself if you want to.
847 *
848 * The returned buffer_head has ->b_count elevated. The caller is expected
849 * to brelse() it when appropriate.
850 */
851 static struct buffer_head * ext4_find_entry (struct inode *dir,
852 const struct qstr *d_name,
853 struct ext4_dir_entry_2 ** res_dir)
854 {
855 struct super_block *sb;
856 struct buffer_head *bh_use[NAMEI_RA_SIZE];
857 struct buffer_head *bh, *ret = NULL;
858 ext4_lblk_t start, block, b;
859 const u8 *name = d_name->name;
860 int ra_max = 0; /* Number of bh's in the readahead
861 buffer, bh_use[] */
862 int ra_ptr = 0; /* Current index into readahead
863 buffer */
864 int num = 0;
865 ext4_lblk_t nblocks;
866 int i, err;
867 int namelen;
868
869 *res_dir = NULL;
870 sb = dir->i_sb;
871 namelen = d_name->len;
872 if (namelen > EXT4_NAME_LEN)
873 return NULL;
874 if ((namelen <= 2) && (name[0] == '.') &&
875 (name[1] == '.' || name[1] == '0')) {
876 /*
877 * "." or ".." will only be in the first block
878 * NFS may look up ".."; "." should be handled by the VFS
879 */
880 block = start = 0;
881 nblocks = 1;
882 goto restart;
883 }
884 if (is_dx(dir)) {
885 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
886 /*
887 * On success, or if the error was file not found,
888 * return. Otherwise, fall back to doing a search the
889 * old fashioned way.
890 */
891 if (bh || (err != ERR_BAD_DX_DIR))
892 return bh;
893 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
894 "falling back\n"));
895 }
896 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
897 start = EXT4_I(dir)->i_dir_start_lookup;
898 if (start >= nblocks)
899 start = 0;
900 block = start;
901 restart:
902 do {
903 /*
904 * We deal with the read-ahead logic here.
905 */
906 if (ra_ptr >= ra_max) {
907 /* Refill the readahead buffer */
908 ra_ptr = 0;
909 b = block;
910 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
911 /*
912 * Terminate if we reach the end of the
913 * directory and must wrap, or if our
914 * search has finished at this block.
915 */
916 if (b >= nblocks || (num && block == start)) {
917 bh_use[ra_max] = NULL;
918 break;
919 }
920 num++;
921 bh = ext4_getblk(NULL, dir, b++, 0, &err);
922 bh_use[ra_max] = bh;
923 if (bh)
924 ll_rw_block(READ_META, 1, &bh);
925 }
926 }
927 if ((bh = bh_use[ra_ptr++]) == NULL)
928 goto next;
929 wait_on_buffer(bh);
930 if (!buffer_uptodate(bh)) {
931 /* read error, skip block & hope for the best */
932 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
933 (unsigned long) block);
934 brelse(bh);
935 goto next;
936 }
937 i = search_dirblock(bh, dir, d_name,
938 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
939 if (i == 1) {
940 EXT4_I(dir)->i_dir_start_lookup = block;
941 ret = bh;
942 goto cleanup_and_exit;
943 } else {
944 brelse(bh);
945 if (i < 0)
946 goto cleanup_and_exit;
947 }
948 next:
949 if (++block >= nblocks)
950 block = 0;
951 } while (block != start);
952
953 /*
954 * If the directory has grown while we were searching, then
955 * search the last part of the directory before giving up.
956 */
957 block = nblocks;
958 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
959 if (block < nblocks) {
960 start = 0;
961 goto restart;
962 }
963
964 cleanup_and_exit:
965 /* Clean up the read-ahead blocks */
966 for (; ra_ptr < ra_max; ra_ptr++)
967 brelse(bh_use[ra_ptr]);
968 return ret;
969 }
970
971 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
972 struct ext4_dir_entry_2 **res_dir, int *err)
973 {
974 struct super_block * sb = dir->i_sb;
975 struct dx_hash_info hinfo;
976 struct dx_frame frames[2], *frame;
977 struct buffer_head *bh;
978 ext4_lblk_t block;
979 int retval;
980
981 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
982 return NULL;
983 do {
984 block = dx_get_block(frame->at);
985 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
986 goto errout;
987
988 retval = search_dirblock(bh, dir, d_name,
989 block << EXT4_BLOCK_SIZE_BITS(sb),
990 res_dir);
991 if (retval == 1) { /* Success! */
992 dx_release(frames);
993 return bh;
994 }
995 brelse(bh);
996 if (retval == -1) {
997 *err = ERR_BAD_DX_DIR;
998 goto errout;
999 }
1000
1001 /* Check to see if we should continue to search */
1002 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1003 frames, NULL);
1004 if (retval < 0) {
1005 ext4_warning(sb,
1006 "error reading index page in directory #%lu",
1007 dir->i_ino);
1008 *err = retval;
1009 goto errout;
1010 }
1011 } while (retval == 1);
1012
1013 *err = -ENOENT;
1014 errout:
1015 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1016 dx_release (frames);
1017 return NULL;
1018 }
1019
1020 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1021 {
1022 struct inode *inode;
1023 struct ext4_dir_entry_2 *de;
1024 struct buffer_head *bh;
1025
1026 if (dentry->d_name.len > EXT4_NAME_LEN)
1027 return ERR_PTR(-ENAMETOOLONG);
1028
1029 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1030 inode = NULL;
1031 if (bh) {
1032 __u32 ino = le32_to_cpu(de->inode);
1033 brelse(bh);
1034 if (!ext4_valid_inum(dir->i_sb, ino)) {
1035 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1036 return ERR_PTR(-EIO);
1037 }
1038 inode = ext4_iget(dir->i_sb, ino);
1039 if (unlikely(IS_ERR(inode))) {
1040 if (PTR_ERR(inode) == -ESTALE) {
1041 EXT4_ERROR_INODE(dir,
1042 "deleted inode referenced: %u",
1043 ino);
1044 return ERR_PTR(-EIO);
1045 } else {
1046 return ERR_CAST(inode);
1047 }
1048 }
1049 }
1050 return d_splice_alias(inode, dentry);
1051 }
1052
1053
1054 struct dentry *ext4_get_parent(struct dentry *child)
1055 {
1056 __u32 ino;
1057 static const struct qstr dotdot = {
1058 .name = "..",
1059 .len = 2,
1060 };
1061 struct ext4_dir_entry_2 * de;
1062 struct buffer_head *bh;
1063
1064 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1065 if (!bh)
1066 return ERR_PTR(-ENOENT);
1067 ino = le32_to_cpu(de->inode);
1068 brelse(bh);
1069
1070 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1071 EXT4_ERROR_INODE(child->d_inode,
1072 "bad parent inode number: %u", ino);
1073 return ERR_PTR(-EIO);
1074 }
1075
1076 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1077 }
1078
1079 #define S_SHIFT 12
1080 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1081 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1082 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1083 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1084 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1085 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1086 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1087 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1088 };
1089
1090 static inline void ext4_set_de_type(struct super_block *sb,
1091 struct ext4_dir_entry_2 *de,
1092 umode_t mode) {
1093 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1094 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1095 }
1096
1097 /*
1098 * Move count entries from end of map between two memory locations.
1099 * Returns pointer to last entry moved.
1100 */
1101 static struct ext4_dir_entry_2 *
1102 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1103 unsigned blocksize)
1104 {
1105 unsigned rec_len = 0;
1106
1107 while (count--) {
1108 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1109 (from + (map->offs<<2));
1110 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1111 memcpy (to, de, rec_len);
1112 ((struct ext4_dir_entry_2 *) to)->rec_len =
1113 ext4_rec_len_to_disk(rec_len, blocksize);
1114 de->inode = 0;
1115 map++;
1116 to += rec_len;
1117 }
1118 return (struct ext4_dir_entry_2 *) (to - rec_len);
1119 }
1120
1121 /*
1122 * Compact each dir entry in the range to the minimal rec_len.
1123 * Returns pointer to last entry in range.
1124 */
1125 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1126 {
1127 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1128 unsigned rec_len = 0;
1129
1130 prev = to = de;
1131 while ((char*)de < base + blocksize) {
1132 next = ext4_next_entry(de, blocksize);
1133 if (de->inode && de->name_len) {
1134 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1135 if (de > to)
1136 memmove(to, de, rec_len);
1137 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1138 prev = to;
1139 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1140 }
1141 de = next;
1142 }
1143 return prev;
1144 }
1145
1146 /*
1147 * Split a full leaf block to make room for a new dir entry.
1148 * Allocate a new block, and move entries so that they are approx. equally full.
1149 * Returns pointer to de in block into which the new entry will be inserted.
1150 */
1151 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1152 struct buffer_head **bh,struct dx_frame *frame,
1153 struct dx_hash_info *hinfo, int *error)
1154 {
1155 unsigned blocksize = dir->i_sb->s_blocksize;
1156 unsigned count, continued;
1157 struct buffer_head *bh2;
1158 ext4_lblk_t newblock;
1159 u32 hash2;
1160 struct dx_map_entry *map;
1161 char *data1 = (*bh)->b_data, *data2;
1162 unsigned split, move, size;
1163 struct ext4_dir_entry_2 *de = NULL, *de2;
1164 int err = 0, i;
1165
1166 bh2 = ext4_append (handle, dir, &newblock, &err);
1167 if (!(bh2)) {
1168 brelse(*bh);
1169 *bh = NULL;
1170 goto errout;
1171 }
1172
1173 BUFFER_TRACE(*bh, "get_write_access");
1174 err = ext4_journal_get_write_access(handle, *bh);
1175 if (err)
1176 goto journal_error;
1177
1178 BUFFER_TRACE(frame->bh, "get_write_access");
1179 err = ext4_journal_get_write_access(handle, frame->bh);
1180 if (err)
1181 goto journal_error;
1182
1183 data2 = bh2->b_data;
1184
1185 /* create map in the end of data2 block */
1186 map = (struct dx_map_entry *) (data2 + blocksize);
1187 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1188 blocksize, hinfo, map);
1189 map -= count;
1190 dx_sort_map(map, count);
1191 /* Split the existing block in the middle, size-wise */
1192 size = 0;
1193 move = 0;
1194 for (i = count-1; i >= 0; i--) {
1195 /* is more than half of this entry in 2nd half of the block? */
1196 if (size + map[i].size/2 > blocksize/2)
1197 break;
1198 size += map[i].size;
1199 move++;
1200 }
1201 /* map index at which we will split */
1202 split = count - move;
1203 hash2 = map[split].hash;
1204 continued = hash2 == map[split - 1].hash;
1205 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1206 (unsigned long)dx_get_block(frame->at),
1207 hash2, split, count-split));
1208
1209 /* Fancy dance to stay within two buffers */
1210 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1211 de = dx_pack_dirents(data1, blocksize);
1212 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1213 blocksize);
1214 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1215 blocksize);
1216 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1217 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1218
1219 /* Which block gets the new entry? */
1220 if (hinfo->hash >= hash2)
1221 {
1222 swap(*bh, bh2);
1223 de = de2;
1224 }
1225 dx_insert_block(frame, hash2 + continued, newblock);
1226 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1227 if (err)
1228 goto journal_error;
1229 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1230 if (err)
1231 goto journal_error;
1232 brelse(bh2);
1233 dxtrace(dx_show_index("frame", frame->entries));
1234 return de;
1235
1236 journal_error:
1237 brelse(*bh);
1238 brelse(bh2);
1239 *bh = NULL;
1240 ext4_std_error(dir->i_sb, err);
1241 errout:
1242 *error = err;
1243 return NULL;
1244 }
1245
1246 /*
1247 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1248 * it points to a directory entry which is guaranteed to be large
1249 * enough for new directory entry. If de is NULL, then
1250 * add_dirent_to_buf will attempt search the directory block for
1251 * space. It will return -ENOSPC if no space is available, and -EIO
1252 * and -EEXIST if directory entry already exists.
1253 */
1254 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1255 struct inode *inode, struct ext4_dir_entry_2 *de,
1256 struct buffer_head *bh)
1257 {
1258 struct inode *dir = dentry->d_parent->d_inode;
1259 const char *name = dentry->d_name.name;
1260 int namelen = dentry->d_name.len;
1261 unsigned int offset = 0;
1262 unsigned int blocksize = dir->i_sb->s_blocksize;
1263 unsigned short reclen;
1264 int nlen, rlen, err;
1265 char *top;
1266
1267 reclen = EXT4_DIR_REC_LEN(namelen);
1268 if (!de) {
1269 de = (struct ext4_dir_entry_2 *)bh->b_data;
1270 top = bh->b_data + blocksize - reclen;
1271 while ((char *) de <= top) {
1272 if (!ext4_check_dir_entry(dir, de, bh, offset))
1273 return -EIO;
1274 if (ext4_match(namelen, name, de))
1275 return -EEXIST;
1276 nlen = EXT4_DIR_REC_LEN(de->name_len);
1277 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1278 if ((de->inode? rlen - nlen: rlen) >= reclen)
1279 break;
1280 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1281 offset += rlen;
1282 }
1283 if ((char *) de > top)
1284 return -ENOSPC;
1285 }
1286 BUFFER_TRACE(bh, "get_write_access");
1287 err = ext4_journal_get_write_access(handle, bh);
1288 if (err) {
1289 ext4_std_error(dir->i_sb, err);
1290 return err;
1291 }
1292
1293 /* By now the buffer is marked for journaling */
1294 nlen = EXT4_DIR_REC_LEN(de->name_len);
1295 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1296 if (de->inode) {
1297 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1298 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1299 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1300 de = de1;
1301 }
1302 de->file_type = EXT4_FT_UNKNOWN;
1303 if (inode) {
1304 de->inode = cpu_to_le32(inode->i_ino);
1305 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1306 } else
1307 de->inode = 0;
1308 de->name_len = namelen;
1309 memcpy(de->name, name, namelen);
1310 /*
1311 * XXX shouldn't update any times until successful
1312 * completion of syscall, but too many callers depend
1313 * on this.
1314 *
1315 * XXX similarly, too many callers depend on
1316 * ext4_new_inode() setting the times, but error
1317 * recovery deletes the inode, so the worst that can
1318 * happen is that the times are slightly out of date
1319 * and/or different from the directory change time.
1320 */
1321 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1322 ext4_update_dx_flag(dir);
1323 dir->i_version++;
1324 ext4_mark_inode_dirty(handle, dir);
1325 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1326 err = ext4_handle_dirty_metadata(handle, dir, bh);
1327 if (err)
1328 ext4_std_error(dir->i_sb, err);
1329 return 0;
1330 }
1331
1332 /*
1333 * This converts a one block unindexed directory to a 3 block indexed
1334 * directory, and adds the dentry to the indexed directory.
1335 */
1336 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1337 struct inode *inode, struct buffer_head *bh)
1338 {
1339 struct inode *dir = dentry->d_parent->d_inode;
1340 const char *name = dentry->d_name.name;
1341 int namelen = dentry->d_name.len;
1342 struct buffer_head *bh2;
1343 struct dx_root *root;
1344 struct dx_frame frames[2], *frame;
1345 struct dx_entry *entries;
1346 struct ext4_dir_entry_2 *de, *de2;
1347 char *data1, *top;
1348 unsigned len;
1349 int retval;
1350 unsigned blocksize;
1351 struct dx_hash_info hinfo;
1352 ext4_lblk_t block;
1353 struct fake_dirent *fde;
1354
1355 blocksize = dir->i_sb->s_blocksize;
1356 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1357 retval = ext4_journal_get_write_access(handle, bh);
1358 if (retval) {
1359 ext4_std_error(dir->i_sb, retval);
1360 brelse(bh);
1361 return retval;
1362 }
1363 root = (struct dx_root *) bh->b_data;
1364
1365 /* The 0th block becomes the root, move the dirents out */
1366 fde = &root->dotdot;
1367 de = (struct ext4_dir_entry_2 *)((char *)fde +
1368 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1369 if ((char *) de >= (((char *) root) + blocksize)) {
1370 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1371 brelse(bh);
1372 return -EIO;
1373 }
1374 len = ((char *) root) + blocksize - (char *) de;
1375
1376 /* Allocate new block for the 0th block's dirents */
1377 bh2 = ext4_append(handle, dir, &block, &retval);
1378 if (!(bh2)) {
1379 brelse(bh);
1380 return retval;
1381 }
1382 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1383 data1 = bh2->b_data;
1384
1385 memcpy (data1, de, len);
1386 de = (struct ext4_dir_entry_2 *) data1;
1387 top = data1 + len;
1388 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1389 de = de2;
1390 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1391 blocksize);
1392 /* Initialize the root; the dot dirents already exist */
1393 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1394 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1395 blocksize);
1396 memset (&root->info, 0, sizeof(root->info));
1397 root->info.info_length = sizeof(root->info);
1398 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1399 entries = root->entries;
1400 dx_set_block(entries, 1);
1401 dx_set_count(entries, 1);
1402 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1403
1404 /* Initialize as for dx_probe */
1405 hinfo.hash_version = root->info.hash_version;
1406 if (hinfo.hash_version <= DX_HASH_TEA)
1407 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1408 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1409 ext4fs_dirhash(name, namelen, &hinfo);
1410 frame = frames;
1411 frame->entries = entries;
1412 frame->at = entries;
1413 frame->bh = bh;
1414 bh = bh2;
1415 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1416 dx_release (frames);
1417 if (!(de))
1418 return retval;
1419
1420 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1421 brelse(bh);
1422 return retval;
1423 }
1424
1425 /*
1426 * ext4_add_entry()
1427 *
1428 * adds a file entry to the specified directory, using the same
1429 * semantics as ext4_find_entry(). It returns NULL if it failed.
1430 *
1431 * NOTE!! The inode part of 'de' is left at 0 - which means you
1432 * may not sleep between calling this and putting something into
1433 * the entry, as someone else might have used it while you slept.
1434 */
1435 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1436 struct inode *inode)
1437 {
1438 struct inode *dir = dentry->d_parent->d_inode;
1439 struct buffer_head *bh;
1440 struct ext4_dir_entry_2 *de;
1441 struct super_block *sb;
1442 int retval;
1443 int dx_fallback=0;
1444 unsigned blocksize;
1445 ext4_lblk_t block, blocks;
1446
1447 sb = dir->i_sb;
1448 blocksize = sb->s_blocksize;
1449 if (!dentry->d_name.len)
1450 return -EINVAL;
1451 if (is_dx(dir)) {
1452 retval = ext4_dx_add_entry(handle, dentry, inode);
1453 if (!retval || (retval != ERR_BAD_DX_DIR))
1454 return retval;
1455 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1456 dx_fallback++;
1457 ext4_mark_inode_dirty(handle, dir);
1458 }
1459 blocks = dir->i_size >> sb->s_blocksize_bits;
1460 for (block = 0; block < blocks; block++) {
1461 bh = ext4_bread(handle, dir, block, 0, &retval);
1462 if(!bh)
1463 return retval;
1464 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1465 if (retval != -ENOSPC) {
1466 brelse(bh);
1467 return retval;
1468 }
1469
1470 if (blocks == 1 && !dx_fallback &&
1471 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1472 return make_indexed_dir(handle, dentry, inode, bh);
1473 brelse(bh);
1474 }
1475 bh = ext4_append(handle, dir, &block, &retval);
1476 if (!bh)
1477 return retval;
1478 de = (struct ext4_dir_entry_2 *) bh->b_data;
1479 de->inode = 0;
1480 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1481 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1482 brelse(bh);
1483 if (retval == 0)
1484 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1485 return retval;
1486 }
1487
1488 /*
1489 * Returns 0 for success, or a negative error value
1490 */
1491 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1492 struct inode *inode)
1493 {
1494 struct dx_frame frames[2], *frame;
1495 struct dx_entry *entries, *at;
1496 struct dx_hash_info hinfo;
1497 struct buffer_head *bh;
1498 struct inode *dir = dentry->d_parent->d_inode;
1499 struct super_block *sb = dir->i_sb;
1500 struct ext4_dir_entry_2 *de;
1501 int err;
1502
1503 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1504 if (!frame)
1505 return err;
1506 entries = frame->entries;
1507 at = frame->at;
1508
1509 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1510 goto cleanup;
1511
1512 BUFFER_TRACE(bh, "get_write_access");
1513 err = ext4_journal_get_write_access(handle, bh);
1514 if (err)
1515 goto journal_error;
1516
1517 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1518 if (err != -ENOSPC)
1519 goto cleanup;
1520
1521 /* Block full, should compress but for now just split */
1522 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1523 dx_get_count(entries), dx_get_limit(entries)));
1524 /* Need to split index? */
1525 if (dx_get_count(entries) == dx_get_limit(entries)) {
1526 ext4_lblk_t newblock;
1527 unsigned icount = dx_get_count(entries);
1528 int levels = frame - frames;
1529 struct dx_entry *entries2;
1530 struct dx_node *node2;
1531 struct buffer_head *bh2;
1532
1533 if (levels && (dx_get_count(frames->entries) ==
1534 dx_get_limit(frames->entries))) {
1535 ext4_warning(sb, "Directory index full!");
1536 err = -ENOSPC;
1537 goto cleanup;
1538 }
1539 bh2 = ext4_append (handle, dir, &newblock, &err);
1540 if (!(bh2))
1541 goto cleanup;
1542 node2 = (struct dx_node *)(bh2->b_data);
1543 entries2 = node2->entries;
1544 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1545 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1546 sb->s_blocksize);
1547 BUFFER_TRACE(frame->bh, "get_write_access");
1548 err = ext4_journal_get_write_access(handle, frame->bh);
1549 if (err)
1550 goto journal_error;
1551 if (levels) {
1552 unsigned icount1 = icount/2, icount2 = icount - icount1;
1553 unsigned hash2 = dx_get_hash(entries + icount1);
1554 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1555 icount1, icount2));
1556
1557 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1558 err = ext4_journal_get_write_access(handle,
1559 frames[0].bh);
1560 if (err)
1561 goto journal_error;
1562
1563 memcpy((char *) entries2, (char *) (entries + icount1),
1564 icount2 * sizeof(struct dx_entry));
1565 dx_set_count(entries, icount1);
1566 dx_set_count(entries2, icount2);
1567 dx_set_limit(entries2, dx_node_limit(dir));
1568
1569 /* Which index block gets the new entry? */
1570 if (at - entries >= icount1) {
1571 frame->at = at = at - entries - icount1 + entries2;
1572 frame->entries = entries = entries2;
1573 swap(frame->bh, bh2);
1574 }
1575 dx_insert_block(frames + 0, hash2, newblock);
1576 dxtrace(dx_show_index("node", frames[1].entries));
1577 dxtrace(dx_show_index("node",
1578 ((struct dx_node *) bh2->b_data)->entries));
1579 err = ext4_handle_dirty_metadata(handle, inode, bh2);
1580 if (err)
1581 goto journal_error;
1582 brelse (bh2);
1583 } else {
1584 dxtrace(printk(KERN_DEBUG
1585 "Creating second level index...\n"));
1586 memcpy((char *) entries2, (char *) entries,
1587 icount * sizeof(struct dx_entry));
1588 dx_set_limit(entries2, dx_node_limit(dir));
1589
1590 /* Set up root */
1591 dx_set_count(entries, 1);
1592 dx_set_block(entries + 0, newblock);
1593 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1594
1595 /* Add new access path frame */
1596 frame = frames + 1;
1597 frame->at = at = at - entries + entries2;
1598 frame->entries = entries = entries2;
1599 frame->bh = bh2;
1600 err = ext4_journal_get_write_access(handle,
1601 frame->bh);
1602 if (err)
1603 goto journal_error;
1604 }
1605 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1606 }
1607 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1608 if (!de)
1609 goto cleanup;
1610 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1611 goto cleanup;
1612
1613 journal_error:
1614 ext4_std_error(dir->i_sb, err);
1615 cleanup:
1616 if (bh)
1617 brelse(bh);
1618 dx_release(frames);
1619 return err;
1620 }
1621
1622 /*
1623 * ext4_delete_entry deletes a directory entry by merging it with the
1624 * previous entry
1625 */
1626 static int ext4_delete_entry(handle_t *handle,
1627 struct inode *dir,
1628 struct ext4_dir_entry_2 *de_del,
1629 struct buffer_head *bh)
1630 {
1631 struct ext4_dir_entry_2 *de, *pde;
1632 unsigned int blocksize = dir->i_sb->s_blocksize;
1633 int i;
1634
1635 i = 0;
1636 pde = NULL;
1637 de = (struct ext4_dir_entry_2 *) bh->b_data;
1638 while (i < bh->b_size) {
1639 if (!ext4_check_dir_entry(dir, de, bh, i))
1640 return -EIO;
1641 if (de == de_del) {
1642 BUFFER_TRACE(bh, "get_write_access");
1643 ext4_journal_get_write_access(handle, bh);
1644 if (pde)
1645 pde->rec_len = ext4_rec_len_to_disk(
1646 ext4_rec_len_from_disk(pde->rec_len,
1647 blocksize) +
1648 ext4_rec_len_from_disk(de->rec_len,
1649 blocksize),
1650 blocksize);
1651 else
1652 de->inode = 0;
1653 dir->i_version++;
1654 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1655 ext4_handle_dirty_metadata(handle, dir, bh);
1656 return 0;
1657 }
1658 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1659 pde = de;
1660 de = ext4_next_entry(de, blocksize);
1661 }
1662 return -ENOENT;
1663 }
1664
1665 /*
1666 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1667 * since this indicates that nlinks count was previously 1.
1668 */
1669 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1670 {
1671 inc_nlink(inode);
1672 if (is_dx(inode) && inode->i_nlink > 1) {
1673 /* limit is 16-bit i_links_count */
1674 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1675 inode->i_nlink = 1;
1676 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1677 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1678 }
1679 }
1680 }
1681
1682 /*
1683 * If a directory had nlink == 1, then we should let it be 1. This indicates
1684 * directory has >EXT4_LINK_MAX subdirs.
1685 */
1686 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1687 {
1688 drop_nlink(inode);
1689 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1690 inc_nlink(inode);
1691 }
1692
1693
1694 static int ext4_add_nondir(handle_t *handle,
1695 struct dentry *dentry, struct inode *inode)
1696 {
1697 int err = ext4_add_entry(handle, dentry, inode);
1698 if (!err) {
1699 ext4_mark_inode_dirty(handle, inode);
1700 d_instantiate(dentry, inode);
1701 unlock_new_inode(inode);
1702 return 0;
1703 }
1704 drop_nlink(inode);
1705 unlock_new_inode(inode);
1706 iput(inode);
1707 return err;
1708 }
1709
1710 /*
1711 * By the time this is called, we already have created
1712 * the directory cache entry for the new file, but it
1713 * is so far negative - it has no inode.
1714 *
1715 * If the create succeeds, we fill in the inode information
1716 * with d_instantiate().
1717 */
1718 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1719 struct nameidata *nd)
1720 {
1721 handle_t *handle;
1722 struct inode *inode;
1723 int err, retries = 0;
1724
1725 dquot_initialize(dir);
1726
1727 retry:
1728 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1729 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1730 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1731 if (IS_ERR(handle))
1732 return PTR_ERR(handle);
1733
1734 if (IS_DIRSYNC(dir))
1735 ext4_handle_sync(handle);
1736
1737 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1738 err = PTR_ERR(inode);
1739 if (!IS_ERR(inode)) {
1740 inode->i_op = &ext4_file_inode_operations;
1741 inode->i_fop = &ext4_file_operations;
1742 ext4_set_aops(inode);
1743 err = ext4_add_nondir(handle, dentry, inode);
1744 }
1745 ext4_journal_stop(handle);
1746 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1747 goto retry;
1748 return err;
1749 }
1750
1751 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1752 int mode, dev_t rdev)
1753 {
1754 handle_t *handle;
1755 struct inode *inode;
1756 int err, retries = 0;
1757
1758 if (!new_valid_dev(rdev))
1759 return -EINVAL;
1760
1761 dquot_initialize(dir);
1762
1763 retry:
1764 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1765 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1766 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1767 if (IS_ERR(handle))
1768 return PTR_ERR(handle);
1769
1770 if (IS_DIRSYNC(dir))
1771 ext4_handle_sync(handle);
1772
1773 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1774 err = PTR_ERR(inode);
1775 if (!IS_ERR(inode)) {
1776 init_special_inode(inode, inode->i_mode, rdev);
1777 #ifdef CONFIG_EXT4_FS_XATTR
1778 inode->i_op = &ext4_special_inode_operations;
1779 #endif
1780 err = ext4_add_nondir(handle, dentry, inode);
1781 }
1782 ext4_journal_stop(handle);
1783 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1784 goto retry;
1785 return err;
1786 }
1787
1788 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1789 {
1790 handle_t *handle;
1791 struct inode *inode;
1792 struct buffer_head *dir_block;
1793 struct ext4_dir_entry_2 *de;
1794 unsigned int blocksize = dir->i_sb->s_blocksize;
1795 int err, retries = 0;
1796
1797 if (EXT4_DIR_LINK_MAX(dir))
1798 return -EMLINK;
1799
1800 dquot_initialize(dir);
1801
1802 retry:
1803 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1804 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1805 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1806 if (IS_ERR(handle))
1807 return PTR_ERR(handle);
1808
1809 if (IS_DIRSYNC(dir))
1810 ext4_handle_sync(handle);
1811
1812 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1813 &dentry->d_name, 0);
1814 err = PTR_ERR(inode);
1815 if (IS_ERR(inode))
1816 goto out_stop;
1817
1818 inode->i_op = &ext4_dir_inode_operations;
1819 inode->i_fop = &ext4_dir_operations;
1820 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1821 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1822 if (!dir_block)
1823 goto out_clear_inode;
1824 BUFFER_TRACE(dir_block, "get_write_access");
1825 ext4_journal_get_write_access(handle, dir_block);
1826 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1827 de->inode = cpu_to_le32(inode->i_ino);
1828 de->name_len = 1;
1829 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1830 blocksize);
1831 strcpy(de->name, ".");
1832 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1833 de = ext4_next_entry(de, blocksize);
1834 de->inode = cpu_to_le32(dir->i_ino);
1835 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1836 blocksize);
1837 de->name_len = 2;
1838 strcpy(de->name, "..");
1839 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1840 inode->i_nlink = 2;
1841 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1842 ext4_handle_dirty_metadata(handle, dir, dir_block);
1843 brelse(dir_block);
1844 ext4_mark_inode_dirty(handle, inode);
1845 err = ext4_add_entry(handle, dentry, inode);
1846 if (err) {
1847 out_clear_inode:
1848 clear_nlink(inode);
1849 unlock_new_inode(inode);
1850 ext4_mark_inode_dirty(handle, inode);
1851 iput(inode);
1852 goto out_stop;
1853 }
1854 ext4_inc_count(handle, dir);
1855 ext4_update_dx_flag(dir);
1856 ext4_mark_inode_dirty(handle, dir);
1857 d_instantiate(dentry, inode);
1858 unlock_new_inode(inode);
1859 out_stop:
1860 ext4_journal_stop(handle);
1861 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1862 goto retry;
1863 return err;
1864 }
1865
1866 /*
1867 * routine to check that the specified directory is empty (for rmdir)
1868 */
1869 static int empty_dir(struct inode *inode)
1870 {
1871 unsigned int offset;
1872 struct buffer_head *bh;
1873 struct ext4_dir_entry_2 *de, *de1;
1874 struct super_block *sb;
1875 int err = 0;
1876
1877 sb = inode->i_sb;
1878 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1879 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1880 if (err)
1881 EXT4_ERROR_INODE(inode,
1882 "error %d reading directory lblock 0", err);
1883 else
1884 ext4_warning(inode->i_sb,
1885 "bad directory (dir #%lu) - no data block",
1886 inode->i_ino);
1887 return 1;
1888 }
1889 de = (struct ext4_dir_entry_2 *) bh->b_data;
1890 de1 = ext4_next_entry(de, sb->s_blocksize);
1891 if (le32_to_cpu(de->inode) != inode->i_ino ||
1892 !le32_to_cpu(de1->inode) ||
1893 strcmp(".", de->name) ||
1894 strcmp("..", de1->name)) {
1895 ext4_warning(inode->i_sb,
1896 "bad directory (dir #%lu) - no `.' or `..'",
1897 inode->i_ino);
1898 brelse(bh);
1899 return 1;
1900 }
1901 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1902 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1903 de = ext4_next_entry(de1, sb->s_blocksize);
1904 while (offset < inode->i_size) {
1905 if (!bh ||
1906 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1907 unsigned int lblock;
1908 err = 0;
1909 brelse(bh);
1910 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1911 bh = ext4_bread(NULL, inode, lblock, 0, &err);
1912 if (!bh) {
1913 if (err)
1914 EXT4_ERROR_INODE(inode,
1915 "error %d reading directory "
1916 "lblock %u", err, lblock);
1917 offset += sb->s_blocksize;
1918 continue;
1919 }
1920 de = (struct ext4_dir_entry_2 *) bh->b_data;
1921 }
1922 if (!ext4_check_dir_entry(inode, de, bh, offset)) {
1923 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1924 sb->s_blocksize);
1925 offset = (offset | (sb->s_blocksize - 1)) + 1;
1926 continue;
1927 }
1928 if (le32_to_cpu(de->inode)) {
1929 brelse(bh);
1930 return 0;
1931 }
1932 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1933 de = ext4_next_entry(de, sb->s_blocksize);
1934 }
1935 brelse(bh);
1936 return 1;
1937 }
1938
1939 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1940 * such inodes, starting at the superblock, in case we crash before the
1941 * file is closed/deleted, or in case the inode truncate spans multiple
1942 * transactions and the last transaction is not recovered after a crash.
1943 *
1944 * At filesystem recovery time, we walk this list deleting unlinked
1945 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1946 */
1947 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1948 {
1949 struct super_block *sb = inode->i_sb;
1950 struct ext4_iloc iloc;
1951 int err = 0, rc;
1952
1953 if (!ext4_handle_valid(handle))
1954 return 0;
1955
1956 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1957 if (!list_empty(&EXT4_I(inode)->i_orphan))
1958 goto out_unlock;
1959
1960 /* Orphan handling is only valid for files with data blocks
1961 * being truncated, or files being unlinked. */
1962
1963 /* @@@ FIXME: Observation from aviro:
1964 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1965 * here (on s_orphan_lock), so race with ext4_link() which might bump
1966 * ->i_nlink. For, say it, character device. Not a regular file,
1967 * not a directory, not a symlink and ->i_nlink > 0.
1968 *
1969 * tytso, 4/25/2009: I'm not sure how that could happen;
1970 * shouldn't the fs core protect us from these sort of
1971 * unlink()/link() races?
1972 */
1973 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1974 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1975
1976 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1977 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1978 if (err)
1979 goto out_unlock;
1980
1981 err = ext4_reserve_inode_write(handle, inode, &iloc);
1982 if (err)
1983 goto out_unlock;
1984 /*
1985 * Due to previous errors inode may be already a part of on-disk
1986 * orphan list. If so skip on-disk list modification.
1987 */
1988 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
1989 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
1990 goto mem_insert;
1991
1992 /* Insert this inode at the head of the on-disk orphan list... */
1993 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1994 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1995 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1996 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1997 if (!err)
1998 err = rc;
1999
2000 /* Only add to the head of the in-memory list if all the
2001 * previous operations succeeded. If the orphan_add is going to
2002 * fail (possibly taking the journal offline), we can't risk
2003 * leaving the inode on the orphan list: stray orphan-list
2004 * entries can cause panics at unmount time.
2005 *
2006 * This is safe: on error we're going to ignore the orphan list
2007 * anyway on the next recovery. */
2008 mem_insert:
2009 if (!err)
2010 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2011
2012 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2013 jbd_debug(4, "orphan inode %lu will point to %d\n",
2014 inode->i_ino, NEXT_ORPHAN(inode));
2015 out_unlock:
2016 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2017 ext4_std_error(inode->i_sb, err);
2018 return err;
2019 }
2020
2021 /*
2022 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2023 * of such inodes stored on disk, because it is finally being cleaned up.
2024 */
2025 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2026 {
2027 struct list_head *prev;
2028 struct ext4_inode_info *ei = EXT4_I(inode);
2029 struct ext4_sb_info *sbi;
2030 __u32 ino_next;
2031 struct ext4_iloc iloc;
2032 int err = 0;
2033
2034 /* ext4_handle_valid() assumes a valid handle_t pointer */
2035 if (handle && !ext4_handle_valid(handle))
2036 return 0;
2037
2038 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2039 if (list_empty(&ei->i_orphan))
2040 goto out;
2041
2042 ino_next = NEXT_ORPHAN(inode);
2043 prev = ei->i_orphan.prev;
2044 sbi = EXT4_SB(inode->i_sb);
2045
2046 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2047
2048 list_del_init(&ei->i_orphan);
2049
2050 /* If we're on an error path, we may not have a valid
2051 * transaction handle with which to update the orphan list on
2052 * disk, but we still need to remove the inode from the linked
2053 * list in memory. */
2054 if (sbi->s_journal && !handle)
2055 goto out;
2056
2057 err = ext4_reserve_inode_write(handle, inode, &iloc);
2058 if (err)
2059 goto out_err;
2060
2061 if (prev == &sbi->s_orphan) {
2062 jbd_debug(4, "superblock will point to %u\n", ino_next);
2063 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2064 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2065 if (err)
2066 goto out_brelse;
2067 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2068 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2069 } else {
2070 struct ext4_iloc iloc2;
2071 struct inode *i_prev =
2072 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2073
2074 jbd_debug(4, "orphan inode %lu will point to %u\n",
2075 i_prev->i_ino, ino_next);
2076 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2077 if (err)
2078 goto out_brelse;
2079 NEXT_ORPHAN(i_prev) = ino_next;
2080 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2081 }
2082 if (err)
2083 goto out_brelse;
2084 NEXT_ORPHAN(inode) = 0;
2085 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2086
2087 out_err:
2088 ext4_std_error(inode->i_sb, err);
2089 out:
2090 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2091 return err;
2092
2093 out_brelse:
2094 brelse(iloc.bh);
2095 goto out_err;
2096 }
2097
2098 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2099 {
2100 int retval;
2101 struct inode *inode;
2102 struct buffer_head *bh;
2103 struct ext4_dir_entry_2 *de;
2104 handle_t *handle;
2105
2106 /* Initialize quotas before so that eventual writes go in
2107 * separate transaction */
2108 dquot_initialize(dir);
2109 dquot_initialize(dentry->d_inode);
2110
2111 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2112 if (IS_ERR(handle))
2113 return PTR_ERR(handle);
2114
2115 retval = -ENOENT;
2116 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2117 if (!bh)
2118 goto end_rmdir;
2119
2120 if (IS_DIRSYNC(dir))
2121 ext4_handle_sync(handle);
2122
2123 inode = dentry->d_inode;
2124
2125 retval = -EIO;
2126 if (le32_to_cpu(de->inode) != inode->i_ino)
2127 goto end_rmdir;
2128
2129 retval = -ENOTEMPTY;
2130 if (!empty_dir(inode))
2131 goto end_rmdir;
2132
2133 retval = ext4_delete_entry(handle, dir, de, bh);
2134 if (retval)
2135 goto end_rmdir;
2136 if (!EXT4_DIR_LINK_EMPTY(inode))
2137 ext4_warning(inode->i_sb,
2138 "empty directory has too many links (%d)",
2139 inode->i_nlink);
2140 inode->i_version++;
2141 clear_nlink(inode);
2142 /* There's no need to set i_disksize: the fact that i_nlink is
2143 * zero will ensure that the right thing happens during any
2144 * recovery. */
2145 inode->i_size = 0;
2146 ext4_orphan_add(handle, inode);
2147 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2148 ext4_mark_inode_dirty(handle, inode);
2149 ext4_dec_count(handle, dir);
2150 ext4_update_dx_flag(dir);
2151 ext4_mark_inode_dirty(handle, dir);
2152
2153 end_rmdir:
2154 ext4_journal_stop(handle);
2155 brelse(bh);
2156 return retval;
2157 }
2158
2159 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2160 {
2161 int retval;
2162 struct inode *inode;
2163 struct buffer_head *bh;
2164 struct ext4_dir_entry_2 *de;
2165 handle_t *handle;
2166
2167 /* Initialize quotas before so that eventual writes go
2168 * in separate transaction */
2169 dquot_initialize(dir);
2170 dquot_initialize(dentry->d_inode);
2171
2172 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2173 if (IS_ERR(handle))
2174 return PTR_ERR(handle);
2175
2176 if (IS_DIRSYNC(dir))
2177 ext4_handle_sync(handle);
2178
2179 retval = -ENOENT;
2180 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2181 if (!bh)
2182 goto end_unlink;
2183
2184 inode = dentry->d_inode;
2185
2186 retval = -EIO;
2187 if (le32_to_cpu(de->inode) != inode->i_ino)
2188 goto end_unlink;
2189
2190 if (!inode->i_nlink) {
2191 ext4_warning(inode->i_sb,
2192 "Deleting nonexistent file (%lu), %d",
2193 inode->i_ino, inode->i_nlink);
2194 inode->i_nlink = 1;
2195 }
2196 retval = ext4_delete_entry(handle, dir, de, bh);
2197 if (retval)
2198 goto end_unlink;
2199 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2200 ext4_update_dx_flag(dir);
2201 ext4_mark_inode_dirty(handle, dir);
2202 drop_nlink(inode);
2203 if (!inode->i_nlink)
2204 ext4_orphan_add(handle, inode);
2205 inode->i_ctime = ext4_current_time(inode);
2206 ext4_mark_inode_dirty(handle, inode);
2207 retval = 0;
2208
2209 end_unlink:
2210 ext4_journal_stop(handle);
2211 brelse(bh);
2212 return retval;
2213 }
2214
2215 static int ext4_symlink(struct inode *dir,
2216 struct dentry *dentry, const char *symname)
2217 {
2218 handle_t *handle;
2219 struct inode *inode;
2220 int l, err, retries = 0;
2221
2222 l = strlen(symname)+1;
2223 if (l > dir->i_sb->s_blocksize)
2224 return -ENAMETOOLONG;
2225
2226 dquot_initialize(dir);
2227
2228 retry:
2229 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2230 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2231 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2232 if (IS_ERR(handle))
2233 return PTR_ERR(handle);
2234
2235 if (IS_DIRSYNC(dir))
2236 ext4_handle_sync(handle);
2237
2238 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2239 &dentry->d_name, 0);
2240 err = PTR_ERR(inode);
2241 if (IS_ERR(inode))
2242 goto out_stop;
2243
2244 if (l > sizeof(EXT4_I(inode)->i_data)) {
2245 inode->i_op = &ext4_symlink_inode_operations;
2246 ext4_set_aops(inode);
2247 /*
2248 * page_symlink() calls into ext4_prepare/commit_write.
2249 * We have a transaction open. All is sweetness. It also sets
2250 * i_size in generic_commit_write().
2251 */
2252 err = __page_symlink(inode, symname, l, 1);
2253 if (err) {
2254 clear_nlink(inode);
2255 unlock_new_inode(inode);
2256 ext4_mark_inode_dirty(handle, inode);
2257 iput(inode);
2258 goto out_stop;
2259 }
2260 } else {
2261 /* clear the extent format for fast symlink */
2262 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2263 inode->i_op = &ext4_fast_symlink_inode_operations;
2264 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2265 inode->i_size = l-1;
2266 }
2267 EXT4_I(inode)->i_disksize = inode->i_size;
2268 err = ext4_add_nondir(handle, dentry, inode);
2269 out_stop:
2270 ext4_journal_stop(handle);
2271 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2272 goto retry;
2273 return err;
2274 }
2275
2276 static int ext4_link(struct dentry *old_dentry,
2277 struct inode *dir, struct dentry *dentry)
2278 {
2279 handle_t *handle;
2280 struct inode *inode = old_dentry->d_inode;
2281 int err, retries = 0;
2282
2283 if (inode->i_nlink >= EXT4_LINK_MAX)
2284 return -EMLINK;
2285
2286 dquot_initialize(dir);
2287
2288 /*
2289 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2290 * otherwise has the potential to corrupt the orphan inode list.
2291 */
2292 if (inode->i_nlink == 0)
2293 return -ENOENT;
2294
2295 retry:
2296 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2297 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2298 if (IS_ERR(handle))
2299 return PTR_ERR(handle);
2300
2301 if (IS_DIRSYNC(dir))
2302 ext4_handle_sync(handle);
2303
2304 inode->i_ctime = ext4_current_time(inode);
2305 ext4_inc_count(handle, inode);
2306 ihold(inode);
2307
2308 err = ext4_add_entry(handle, dentry, inode);
2309 if (!err) {
2310 ext4_mark_inode_dirty(handle, inode);
2311 d_instantiate(dentry, inode);
2312 } else {
2313 drop_nlink(inode);
2314 iput(inode);
2315 }
2316 ext4_journal_stop(handle);
2317 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2318 goto retry;
2319 return err;
2320 }
2321
2322 #define PARENT_INO(buffer, size) \
2323 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2324
2325 /*
2326 * Anybody can rename anything with this: the permission checks are left to the
2327 * higher-level routines.
2328 */
2329 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2330 struct inode *new_dir, struct dentry *new_dentry)
2331 {
2332 handle_t *handle;
2333 struct inode *old_inode, *new_inode;
2334 struct buffer_head *old_bh, *new_bh, *dir_bh;
2335 struct ext4_dir_entry_2 *old_de, *new_de;
2336 int retval, force_da_alloc = 0;
2337
2338 dquot_initialize(old_dir);
2339 dquot_initialize(new_dir);
2340
2341 old_bh = new_bh = dir_bh = NULL;
2342
2343 /* Initialize quotas before so that eventual writes go
2344 * in separate transaction */
2345 if (new_dentry->d_inode)
2346 dquot_initialize(new_dentry->d_inode);
2347 handle = ext4_journal_start(old_dir, 2 *
2348 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2349 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2350 if (IS_ERR(handle))
2351 return PTR_ERR(handle);
2352
2353 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2354 ext4_handle_sync(handle);
2355
2356 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2357 /*
2358 * Check for inode number is _not_ due to possible IO errors.
2359 * We might rmdir the source, keep it as pwd of some process
2360 * and merrily kill the link to whatever was created under the
2361 * same name. Goodbye sticky bit ;-<
2362 */
2363 old_inode = old_dentry->d_inode;
2364 retval = -ENOENT;
2365 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2366 goto end_rename;
2367
2368 new_inode = new_dentry->d_inode;
2369 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2370 if (new_bh) {
2371 if (!new_inode) {
2372 brelse(new_bh);
2373 new_bh = NULL;
2374 }
2375 }
2376 if (S_ISDIR(old_inode->i_mode)) {
2377 if (new_inode) {
2378 retval = -ENOTEMPTY;
2379 if (!empty_dir(new_inode))
2380 goto end_rename;
2381 }
2382 retval = -EIO;
2383 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2384 if (!dir_bh)
2385 goto end_rename;
2386 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2387 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2388 goto end_rename;
2389 retval = -EMLINK;
2390 if (!new_inode && new_dir != old_dir &&
2391 EXT4_DIR_LINK_MAX(new_dir))
2392 goto end_rename;
2393 }
2394 if (!new_bh) {
2395 retval = ext4_add_entry(handle, new_dentry, old_inode);
2396 if (retval)
2397 goto end_rename;
2398 } else {
2399 BUFFER_TRACE(new_bh, "get write access");
2400 ext4_journal_get_write_access(handle, new_bh);
2401 new_de->inode = cpu_to_le32(old_inode->i_ino);
2402 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2403 EXT4_FEATURE_INCOMPAT_FILETYPE))
2404 new_de->file_type = old_de->file_type;
2405 new_dir->i_version++;
2406 new_dir->i_ctime = new_dir->i_mtime =
2407 ext4_current_time(new_dir);
2408 ext4_mark_inode_dirty(handle, new_dir);
2409 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2410 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2411 brelse(new_bh);
2412 new_bh = NULL;
2413 }
2414
2415 /*
2416 * Like most other Unix systems, set the ctime for inodes on a
2417 * rename.
2418 */
2419 old_inode->i_ctime = ext4_current_time(old_inode);
2420 ext4_mark_inode_dirty(handle, old_inode);
2421
2422 /*
2423 * ok, that's it
2424 */
2425 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2426 old_de->name_len != old_dentry->d_name.len ||
2427 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2428 (retval = ext4_delete_entry(handle, old_dir,
2429 old_de, old_bh)) == -ENOENT) {
2430 /* old_de could have moved from under us during htree split, so
2431 * make sure that we are deleting the right entry. We might
2432 * also be pointing to a stale entry in the unused part of
2433 * old_bh so just checking inum and the name isn't enough. */
2434 struct buffer_head *old_bh2;
2435 struct ext4_dir_entry_2 *old_de2;
2436
2437 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2438 if (old_bh2) {
2439 retval = ext4_delete_entry(handle, old_dir,
2440 old_de2, old_bh2);
2441 brelse(old_bh2);
2442 }
2443 }
2444 if (retval) {
2445 ext4_warning(old_dir->i_sb,
2446 "Deleting old file (%lu), %d, error=%d",
2447 old_dir->i_ino, old_dir->i_nlink, retval);
2448 }
2449
2450 if (new_inode) {
2451 ext4_dec_count(handle, new_inode);
2452 new_inode->i_ctime = ext4_current_time(new_inode);
2453 }
2454 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2455 ext4_update_dx_flag(old_dir);
2456 if (dir_bh) {
2457 BUFFER_TRACE(dir_bh, "get_write_access");
2458 ext4_journal_get_write_access(handle, dir_bh);
2459 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2460 cpu_to_le32(new_dir->i_ino);
2461 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2462 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2463 ext4_dec_count(handle, old_dir);
2464 if (new_inode) {
2465 /* checked empty_dir above, can't have another parent,
2466 * ext4_dec_count() won't work for many-linked dirs */
2467 new_inode->i_nlink = 0;
2468 } else {
2469 ext4_inc_count(handle, new_dir);
2470 ext4_update_dx_flag(new_dir);
2471 ext4_mark_inode_dirty(handle, new_dir);
2472 }
2473 }
2474 ext4_mark_inode_dirty(handle, old_dir);
2475 if (new_inode) {
2476 ext4_mark_inode_dirty(handle, new_inode);
2477 if (!new_inode->i_nlink)
2478 ext4_orphan_add(handle, new_inode);
2479 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2480 force_da_alloc = 1;
2481 }
2482 retval = 0;
2483
2484 end_rename:
2485 brelse(dir_bh);
2486 brelse(old_bh);
2487 brelse(new_bh);
2488 ext4_journal_stop(handle);
2489 if (retval == 0 && force_da_alloc)
2490 ext4_alloc_da_blocks(old_inode);
2491 return retval;
2492 }
2493
2494 /*
2495 * directories can handle most operations...
2496 */
2497 const struct inode_operations ext4_dir_inode_operations = {
2498 .create = ext4_create,
2499 .lookup = ext4_lookup,
2500 .link = ext4_link,
2501 .unlink = ext4_unlink,
2502 .symlink = ext4_symlink,
2503 .mkdir = ext4_mkdir,
2504 .rmdir = ext4_rmdir,
2505 .mknod = ext4_mknod,
2506 .rename = ext4_rename,
2507 .setattr = ext4_setattr,
2508 #ifdef CONFIG_EXT4_FS_XATTR
2509 .setxattr = generic_setxattr,
2510 .getxattr = generic_getxattr,
2511 .listxattr = ext4_listxattr,
2512 .removexattr = generic_removexattr,
2513 #endif
2514 .check_acl = ext4_check_acl,
2515 .fiemap = ext4_fiemap,
2516 };
2517
2518 const struct inode_operations ext4_special_inode_operations = {
2519 .setattr = ext4_setattr,
2520 #ifdef CONFIG_EXT4_FS_XATTR
2521 .setxattr = generic_setxattr,
2522 .getxattr = generic_getxattr,
2523 .listxattr = ext4_listxattr,
2524 .removexattr = generic_removexattr,
2525 #endif
2526 .check_acl = ext4_check_acl,
2527 };
AMDTP streaming driver for the Linux FireWire stack (Juju)