]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/ext4/namei.c
Merge branch 'linux-next' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes...
[mirror_ubuntu-bionic-kernel.git] / fs / ext4 / namei.c
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 "namei.h"
41 #include "xattr.h"
42 #include "acl.h"
43
44 /*
45 * define how far ahead to read directories while searching them.
46 */
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 #define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51
52 static struct buffer_head *ext4_append(handle_t *handle,
53 struct inode *inode,
54 ext4_lblk_t *block, int *err)
55 {
56 struct buffer_head *bh;
57
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
60 bh = ext4_bread(handle, inode, *block, 1, err);
61 if (bh) {
62 inode->i_size += inode->i_sb->s_blocksize;
63 EXT4_I(inode)->i_disksize = inode->i_size;
64 *err = ext4_journal_get_write_access(handle, bh);
65 if (*err) {
66 brelse(bh);
67 bh = NULL;
68 }
69 }
70 return bh;
71 }
72
73 #ifndef assert
74 #define assert(test) J_ASSERT(test)
75 #endif
76
77 #ifdef DX_DEBUG
78 #define dxtrace(command) command
79 #else
80 #define dxtrace(command)
81 #endif
82
83 struct fake_dirent
84 {
85 __le32 inode;
86 __le16 rec_len;
87 u8 name_len;
88 u8 file_type;
89 };
90
91 struct dx_countlimit
92 {
93 __le16 limit;
94 __le16 count;
95 };
96
97 struct 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
109 struct 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
127 struct dx_node
128 {
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
131 };
132
133
134 struct dx_frame
135 {
136 struct buffer_head *bh;
137 struct dx_entry *entries;
138 struct dx_entry *at;
139 };
140
141 struct dx_map_entry
142 {
143 u32 hash;
144 u16 offs;
145 u16 size;
146 };
147
148 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
149 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
150 static inline unsigned dx_get_hash(struct dx_entry *entry);
151 static void dx_set_hash(struct dx_entry *entry, unsigned value);
152 static unsigned dx_get_count(struct dx_entry *entries);
153 static unsigned dx_get_limit(struct dx_entry *entries);
154 static void dx_set_count(struct dx_entry *entries, unsigned value);
155 static void dx_set_limit(struct dx_entry *entries, unsigned value);
156 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
157 static unsigned dx_node_limit(struct inode *dir);
158 static struct dx_frame *dx_probe(const struct qstr *d_name,
159 struct inode *dir,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
162 int *err);
163 static void dx_release(struct dx_frame *frames);
164 static int dx_make_map(struct ext4_dir_entry_2 *de, int size,
165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
168 struct dx_map_entry *offsets, int count);
169 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size);
170 static void dx_insert_block(struct dx_frame *frame,
171 u32 hash, ext4_lblk_t block);
172 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
173 struct dx_frame *frame,
174 struct dx_frame *frames,
175 __u32 *start_hash);
176 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
177 const struct qstr *d_name,
178 struct ext4_dir_entry_2 **res_dir,
179 int *err);
180 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
181 struct inode *inode);
182
183 /*
184 * p is at least 6 bytes before the end of page
185 */
186 static inline struct ext4_dir_entry_2 *
187 ext4_next_entry(struct ext4_dir_entry_2 *p)
188 {
189 return (struct ext4_dir_entry_2 *)((char *)p +
190 ext4_rec_len_from_disk(p->rec_len));
191 }
192
193 /*
194 * Future: use high four bits of block for coalesce-on-delete flags
195 * Mask them off for now.
196 */
197
198 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
199 {
200 return le32_to_cpu(entry->block) & 0x00ffffff;
201 }
202
203 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
204 {
205 entry->block = cpu_to_le32(value);
206 }
207
208 static inline unsigned dx_get_hash(struct dx_entry *entry)
209 {
210 return le32_to_cpu(entry->hash);
211 }
212
213 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
214 {
215 entry->hash = cpu_to_le32(value);
216 }
217
218 static inline unsigned dx_get_count(struct dx_entry *entries)
219 {
220 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
221 }
222
223 static inline unsigned dx_get_limit(struct dx_entry *entries)
224 {
225 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
226 }
227
228 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
229 {
230 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
231 }
232
233 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
234 {
235 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
236 }
237
238 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
239 {
240 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
241 EXT4_DIR_REC_LEN(2) - infosize;
242 return entry_space / sizeof(struct dx_entry);
243 }
244
245 static inline unsigned dx_node_limit(struct inode *dir)
246 {
247 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
248 return entry_space / sizeof(struct dx_entry);
249 }
250
251 /*
252 * Debug
253 */
254 #ifdef DX_DEBUG
255 static void dx_show_index(char * label, struct dx_entry *entries)
256 {
257 int i, n = dx_get_count (entries);
258 printk(KERN_DEBUG "%s index ", label);
259 for (i = 0; i < n; i++) {
260 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
261 0, (unsigned long)dx_get_block(entries + i));
262 }
263 printk("\n");
264 }
265
266 struct stats
267 {
268 unsigned names;
269 unsigned space;
270 unsigned bcount;
271 };
272
273 static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
274 int size, int show_names)
275 {
276 unsigned names = 0, space = 0;
277 char *base = (char *) de;
278 struct dx_hash_info h = *hinfo;
279
280 printk("names: ");
281 while ((char *) de < base + size)
282 {
283 if (de->inode)
284 {
285 if (show_names)
286 {
287 int len = de->name_len;
288 char *name = de->name;
289 while (len--) printk("%c", *name++);
290 ext4fs_dirhash(de->name, de->name_len, &h);
291 printk(":%x.%u ", h.hash,
292 ((char *) de - base));
293 }
294 space += EXT4_DIR_REC_LEN(de->name_len);
295 names++;
296 }
297 de = ext4_next_entry(de);
298 }
299 printk("(%i)\n", names);
300 return (struct stats) { names, space, 1 };
301 }
302
303 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
304 struct dx_entry *entries, int levels)
305 {
306 unsigned blocksize = dir->i_sb->s_blocksize;
307 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
308 unsigned bcount = 0;
309 struct buffer_head *bh;
310 int err;
311 printk("%i indexed blocks...\n", count);
312 for (i = 0; i < count; i++, entries++)
313 {
314 ext4_lblk_t block = dx_get_block(entries);
315 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
316 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
317 struct stats stats;
318 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
319 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
320 stats = levels?
321 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
322 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
323 names += stats.names;
324 space += stats.space;
325 bcount += stats.bcount;
326 brelse(bh);
327 }
328 if (bcount)
329 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
330 levels ? "" : " ", names, space/bcount,
331 (space/bcount)*100/blocksize);
332 return (struct stats) { names, space, bcount};
333 }
334 #endif /* DX_DEBUG */
335
336 /*
337 * Probe for a directory leaf block to search.
338 *
339 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
340 * error in the directory index, and the caller should fall back to
341 * searching the directory normally. The callers of dx_probe **MUST**
342 * check for this error code, and make sure it never gets reflected
343 * back to userspace.
344 */
345 static struct dx_frame *
346 dx_probe(const struct qstr *d_name, struct inode *dir,
347 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
348 {
349 unsigned count, indirect;
350 struct dx_entry *at, *entries, *p, *q, *m;
351 struct dx_root *root;
352 struct buffer_head *bh;
353 struct dx_frame *frame = frame_in;
354 u32 hash;
355
356 frame->bh = NULL;
357 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
358 goto fail;
359 root = (struct dx_root *) bh->b_data;
360 if (root->info.hash_version != DX_HASH_TEA &&
361 root->info.hash_version != DX_HASH_HALF_MD4 &&
362 root->info.hash_version != DX_HASH_LEGACY) {
363 ext4_warning(dir->i_sb, __func__,
364 "Unrecognised inode hash code %d",
365 root->info.hash_version);
366 brelse(bh);
367 *err = ERR_BAD_DX_DIR;
368 goto fail;
369 }
370 hinfo->hash_version = root->info.hash_version;
371 if (hinfo->hash_version <= DX_HASH_TEA)
372 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
373 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
374 if (d_name)
375 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
376 hash = hinfo->hash;
377
378 if (root->info.unused_flags & 1) {
379 ext4_warning(dir->i_sb, __func__,
380 "Unimplemented inode hash flags: %#06x",
381 root->info.unused_flags);
382 brelse(bh);
383 *err = ERR_BAD_DX_DIR;
384 goto fail;
385 }
386
387 if ((indirect = root->info.indirect_levels) > 1) {
388 ext4_warning(dir->i_sb, __func__,
389 "Unimplemented inode hash depth: %#06x",
390 root->info.indirect_levels);
391 brelse(bh);
392 *err = ERR_BAD_DX_DIR;
393 goto fail;
394 }
395
396 entries = (struct dx_entry *) (((char *)&root->info) +
397 root->info.info_length);
398
399 if (dx_get_limit(entries) != dx_root_limit(dir,
400 root->info.info_length)) {
401 ext4_warning(dir->i_sb, __func__,
402 "dx entry: limit != root limit");
403 brelse(bh);
404 *err = ERR_BAD_DX_DIR;
405 goto fail;
406 }
407
408 dxtrace(printk("Look up %x", hash));
409 while (1)
410 {
411 count = dx_get_count(entries);
412 if (!count || count > dx_get_limit(entries)) {
413 ext4_warning(dir->i_sb, __func__,
414 "dx entry: no count or count > limit");
415 brelse(bh);
416 *err = ERR_BAD_DX_DIR;
417 goto fail2;
418 }
419
420 p = entries + 1;
421 q = entries + count - 1;
422 while (p <= q)
423 {
424 m = p + (q - p)/2;
425 dxtrace(printk("."));
426 if (dx_get_hash(m) > hash)
427 q = m - 1;
428 else
429 p = m + 1;
430 }
431
432 if (0) // linear search cross check
433 {
434 unsigned n = count - 1;
435 at = entries;
436 while (n--)
437 {
438 dxtrace(printk(","));
439 if (dx_get_hash(++at) > hash)
440 {
441 at--;
442 break;
443 }
444 }
445 assert (at == p - 1);
446 }
447
448 at = p - 1;
449 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
450 frame->bh = bh;
451 frame->entries = entries;
452 frame->at = at;
453 if (!indirect--) return frame;
454 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
455 goto fail2;
456 at = entries = ((struct dx_node *) bh->b_data)->entries;
457 if (dx_get_limit(entries) != dx_node_limit (dir)) {
458 ext4_warning(dir->i_sb, __func__,
459 "dx entry: limit != node limit");
460 brelse(bh);
461 *err = ERR_BAD_DX_DIR;
462 goto fail2;
463 }
464 frame++;
465 frame->bh = NULL;
466 }
467 fail2:
468 while (frame >= frame_in) {
469 brelse(frame->bh);
470 frame--;
471 }
472 fail:
473 if (*err == ERR_BAD_DX_DIR)
474 ext4_warning(dir->i_sb, __func__,
475 "Corrupt dir inode %ld, running e2fsck is "
476 "recommended.", dir->i_ino);
477 return NULL;
478 }
479
480 static void dx_release (struct dx_frame *frames)
481 {
482 if (frames[0].bh == NULL)
483 return;
484
485 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
486 brelse(frames[1].bh);
487 brelse(frames[0].bh);
488 }
489
490 /*
491 * This function increments the frame pointer to search the next leaf
492 * block, and reads in the necessary intervening nodes if the search
493 * should be necessary. Whether or not the search is necessary is
494 * controlled by the hash parameter. If the hash value is even, then
495 * the search is only continued if the next block starts with that
496 * hash value. This is used if we are searching for a specific file.
497 *
498 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
499 *
500 * This function returns 1 if the caller should continue to search,
501 * or 0 if it should not. If there is an error reading one of the
502 * index blocks, it will a negative error code.
503 *
504 * If start_hash is non-null, it will be filled in with the starting
505 * hash of the next page.
506 */
507 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
508 struct dx_frame *frame,
509 struct dx_frame *frames,
510 __u32 *start_hash)
511 {
512 struct dx_frame *p;
513 struct buffer_head *bh;
514 int err, num_frames = 0;
515 __u32 bhash;
516
517 p = frame;
518 /*
519 * Find the next leaf page by incrementing the frame pointer.
520 * If we run out of entries in the interior node, loop around and
521 * increment pointer in the parent node. When we break out of
522 * this loop, num_frames indicates the number of interior
523 * nodes need to be read.
524 */
525 while (1) {
526 if (++(p->at) < p->entries + dx_get_count(p->entries))
527 break;
528 if (p == frames)
529 return 0;
530 num_frames++;
531 p--;
532 }
533
534 /*
535 * If the hash is 1, then continue only if the next page has a
536 * continuation hash of any value. This is used for readdir
537 * handling. Otherwise, check to see if the hash matches the
538 * desired contiuation hash. If it doesn't, return since
539 * there's no point to read in the successive index pages.
540 */
541 bhash = dx_get_hash(p->at);
542 if (start_hash)
543 *start_hash = bhash;
544 if ((hash & 1) == 0) {
545 if ((bhash & ~1) != hash)
546 return 0;
547 }
548 /*
549 * If the hash is HASH_NB_ALWAYS, we always go to the next
550 * block so no check is necessary
551 */
552 while (num_frames--) {
553 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
554 0, &err)))
555 return err; /* Failure */
556 p++;
557 brelse(p->bh);
558 p->bh = bh;
559 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
560 }
561 return 1;
562 }
563
564
565 /*
566 * This function fills a red-black tree with information from a
567 * directory block. It returns the number directory entries loaded
568 * into the tree. If there is an error it is returned in err.
569 */
570 static int htree_dirblock_to_tree(struct file *dir_file,
571 struct inode *dir, ext4_lblk_t block,
572 struct dx_hash_info *hinfo,
573 __u32 start_hash, __u32 start_minor_hash)
574 {
575 struct buffer_head *bh;
576 struct ext4_dir_entry_2 *de, *top;
577 int err, count = 0;
578
579 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
580 (unsigned long)block));
581 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
582 return err;
583
584 de = (struct ext4_dir_entry_2 *) bh->b_data;
585 top = (struct ext4_dir_entry_2 *) ((char *) de +
586 dir->i_sb->s_blocksize -
587 EXT4_DIR_REC_LEN(0));
588 for (; de < top; de = ext4_next_entry(de)) {
589 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
590 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
591 +((char *)de - bh->b_data))) {
592 /* On error, skip the f_pos to the next block. */
593 dir_file->f_pos = (dir_file->f_pos |
594 (dir->i_sb->s_blocksize - 1)) + 1;
595 brelse(bh);
596 return count;
597 }
598 ext4fs_dirhash(de->name, de->name_len, hinfo);
599 if ((hinfo->hash < start_hash) ||
600 ((hinfo->hash == start_hash) &&
601 (hinfo->minor_hash < start_minor_hash)))
602 continue;
603 if (de->inode == 0)
604 continue;
605 if ((err = ext4_htree_store_dirent(dir_file,
606 hinfo->hash, hinfo->minor_hash, de)) != 0) {
607 brelse(bh);
608 return err;
609 }
610 count++;
611 }
612 brelse(bh);
613 return count;
614 }
615
616
617 /*
618 * This function fills a red-black tree with information from a
619 * directory. We start scanning the directory in hash order, starting
620 * at start_hash and start_minor_hash.
621 *
622 * This function returns the number of entries inserted into the tree,
623 * or a negative error code.
624 */
625 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
626 __u32 start_minor_hash, __u32 *next_hash)
627 {
628 struct dx_hash_info hinfo;
629 struct ext4_dir_entry_2 *de;
630 struct dx_frame frames[2], *frame;
631 struct inode *dir;
632 ext4_lblk_t block;
633 int count = 0;
634 int ret, err;
635 __u32 hashval;
636
637 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
638 start_hash, start_minor_hash));
639 dir = dir_file->f_path.dentry->d_inode;
640 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
641 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
642 if (hinfo.hash_version <= DX_HASH_TEA)
643 hinfo.hash_version +=
644 EXT4_SB(dir->i_sb)->s_hash_unsigned;
645 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
646 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
647 start_hash, start_minor_hash);
648 *next_hash = ~0;
649 return count;
650 }
651 hinfo.hash = start_hash;
652 hinfo.minor_hash = 0;
653 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
654 if (!frame)
655 return err;
656
657 /* Add '.' and '..' from the htree header */
658 if (!start_hash && !start_minor_hash) {
659 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
660 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
661 goto errout;
662 count++;
663 }
664 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
665 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
666 de = ext4_next_entry(de);
667 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
668 goto errout;
669 count++;
670 }
671
672 while (1) {
673 block = dx_get_block(frame->at);
674 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
675 start_hash, start_minor_hash);
676 if (ret < 0) {
677 err = ret;
678 goto errout;
679 }
680 count += ret;
681 hashval = ~0;
682 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
683 frame, frames, &hashval);
684 *next_hash = hashval;
685 if (ret < 0) {
686 err = ret;
687 goto errout;
688 }
689 /*
690 * Stop if: (a) there are no more entries, or
691 * (b) we have inserted at least one entry and the
692 * next hash value is not a continuation
693 */
694 if ((ret == 0) ||
695 (count && ((hashval & 1) == 0)))
696 break;
697 }
698 dx_release(frames);
699 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
700 "next hash: %x\n", count, *next_hash));
701 return count;
702 errout:
703 dx_release(frames);
704 return (err);
705 }
706
707
708 /*
709 * Directory block splitting, compacting
710 */
711
712 /*
713 * Create map of hash values, offsets, and sizes, stored at end of block.
714 * Returns number of entries mapped.
715 */
716 static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
717 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
718 {
719 int count = 0;
720 char *base = (char *) de;
721 struct dx_hash_info h = *hinfo;
722
723 while ((char *) de < base + size)
724 {
725 if (de->name_len && de->inode) {
726 ext4fs_dirhash(de->name, de->name_len, &h);
727 map_tail--;
728 map_tail->hash = h.hash;
729 map_tail->offs = (u16) ((char *) de - base);
730 map_tail->size = le16_to_cpu(de->rec_len);
731 count++;
732 cond_resched();
733 }
734 /* XXX: do we need to check rec_len == 0 case? -Chris */
735 de = ext4_next_entry(de);
736 }
737 return count;
738 }
739
740 /* Sort map by hash value */
741 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
742 {
743 struct dx_map_entry *p, *q, *top = map + count - 1;
744 int more;
745 /* Combsort until bubble sort doesn't suck */
746 while (count > 2) {
747 count = count*10/13;
748 if (count - 9 < 2) /* 9, 10 -> 11 */
749 count = 11;
750 for (p = top, q = p - count; q >= map; p--, q--)
751 if (p->hash < q->hash)
752 swap(*p, *q);
753 }
754 /* Garden variety bubble sort */
755 do {
756 more = 0;
757 q = top;
758 while (q-- > map) {
759 if (q[1].hash >= q[0].hash)
760 continue;
761 swap(*(q+1), *q);
762 more = 1;
763 }
764 } while(more);
765 }
766
767 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
768 {
769 struct dx_entry *entries = frame->entries;
770 struct dx_entry *old = frame->at, *new = old + 1;
771 int count = dx_get_count(entries);
772
773 assert(count < dx_get_limit(entries));
774 assert(old < entries + count);
775 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
776 dx_set_hash(new, hash);
777 dx_set_block(new, block);
778 dx_set_count(entries, count + 1);
779 }
780
781 static void ext4_update_dx_flag(struct inode *inode)
782 {
783 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
784 EXT4_FEATURE_COMPAT_DIR_INDEX))
785 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
786 }
787
788 /*
789 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
790 *
791 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
792 * `de != NULL' is guaranteed by caller.
793 */
794 static inline int ext4_match (int len, const char * const name,
795 struct ext4_dir_entry_2 * de)
796 {
797 if (len != de->name_len)
798 return 0;
799 if (!de->inode)
800 return 0;
801 return !memcmp(name, de->name, len);
802 }
803
804 /*
805 * Returns 0 if not found, -1 on failure, and 1 on success
806 */
807 static inline int search_dirblock(struct buffer_head *bh,
808 struct inode *dir,
809 const struct qstr *d_name,
810 unsigned int offset,
811 struct ext4_dir_entry_2 ** res_dir)
812 {
813 struct ext4_dir_entry_2 * de;
814 char * dlimit;
815 int de_len;
816 const char *name = d_name->name;
817 int namelen = d_name->len;
818
819 de = (struct ext4_dir_entry_2 *) bh->b_data;
820 dlimit = bh->b_data + dir->i_sb->s_blocksize;
821 while ((char *) de < dlimit) {
822 /* this code is executed quadratically often */
823 /* do minimal checking `by hand' */
824
825 if ((char *) de + namelen <= dlimit &&
826 ext4_match (namelen, name, de)) {
827 /* found a match - just to be sure, do a full check */
828 if (!ext4_check_dir_entry("ext4_find_entry",
829 dir, de, bh, offset))
830 return -1;
831 *res_dir = de;
832 return 1;
833 }
834 /* prevent looping on a bad block */
835 de_len = ext4_rec_len_from_disk(de->rec_len);
836 if (de_len <= 0)
837 return -1;
838 offset += de_len;
839 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
840 }
841 return 0;
842 }
843
844
845 /*
846 * ext4_find_entry()
847 *
848 * finds an entry in the specified directory with the wanted name. It
849 * returns the cache buffer in which the entry was found, and the entry
850 * itself (as a parameter - res_dir). It does NOT read the inode of the
851 * entry - you'll have to do that yourself if you want to.
852 *
853 * The returned buffer_head has ->b_count elevated. The caller is expected
854 * to brelse() it when appropriate.
855 */
856 static struct buffer_head * ext4_find_entry (struct inode *dir,
857 const struct qstr *d_name,
858 struct ext4_dir_entry_2 ** res_dir)
859 {
860 struct super_block *sb;
861 struct buffer_head *bh_use[NAMEI_RA_SIZE];
862 struct buffer_head *bh, *ret = NULL;
863 ext4_lblk_t start, block, b;
864 int ra_max = 0; /* Number of bh's in the readahead
865 buffer, bh_use[] */
866 int ra_ptr = 0; /* Current index into readahead
867 buffer */
868 int num = 0;
869 ext4_lblk_t nblocks;
870 int i, err;
871 int namelen;
872
873 *res_dir = NULL;
874 sb = dir->i_sb;
875 namelen = d_name->len;
876 if (namelen > EXT4_NAME_LEN)
877 return NULL;
878 if (is_dx(dir)) {
879 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
880 /*
881 * On success, or if the error was file not found,
882 * return. Otherwise, fall back to doing a search the
883 * old fashioned way.
884 */
885 if (bh || (err != ERR_BAD_DX_DIR))
886 return bh;
887 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
888 "falling back\n"));
889 }
890 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
891 start = EXT4_I(dir)->i_dir_start_lookup;
892 if (start >= nblocks)
893 start = 0;
894 block = start;
895 restart:
896 do {
897 /*
898 * We deal with the read-ahead logic here.
899 */
900 if (ra_ptr >= ra_max) {
901 /* Refill the readahead buffer */
902 ra_ptr = 0;
903 b = block;
904 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
905 /*
906 * Terminate if we reach the end of the
907 * directory and must wrap, or if our
908 * search has finished at this block.
909 */
910 if (b >= nblocks || (num && block == start)) {
911 bh_use[ra_max] = NULL;
912 break;
913 }
914 num++;
915 bh = ext4_getblk(NULL, dir, b++, 0, &err);
916 bh_use[ra_max] = bh;
917 if (bh)
918 ll_rw_block(READ_META, 1, &bh);
919 }
920 }
921 if ((bh = bh_use[ra_ptr++]) == NULL)
922 goto next;
923 wait_on_buffer(bh);
924 if (!buffer_uptodate(bh)) {
925 /* read error, skip block & hope for the best */
926 ext4_error(sb, __func__, "reading directory #%lu "
927 "offset %lu", dir->i_ino,
928 (unsigned long)block);
929 brelse(bh);
930 goto next;
931 }
932 i = search_dirblock(bh, dir, d_name,
933 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
934 if (i == 1) {
935 EXT4_I(dir)->i_dir_start_lookup = block;
936 ret = bh;
937 goto cleanup_and_exit;
938 } else {
939 brelse(bh);
940 if (i < 0)
941 goto cleanup_and_exit;
942 }
943 next:
944 if (++block >= nblocks)
945 block = 0;
946 } while (block != start);
947
948 /*
949 * If the directory has grown while we were searching, then
950 * search the last part of the directory before giving up.
951 */
952 block = nblocks;
953 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
954 if (block < nblocks) {
955 start = 0;
956 goto restart;
957 }
958
959 cleanup_and_exit:
960 /* Clean up the read-ahead blocks */
961 for (; ra_ptr < ra_max; ra_ptr++)
962 brelse(bh_use[ra_ptr]);
963 return ret;
964 }
965
966 static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
967 struct ext4_dir_entry_2 **res_dir, int *err)
968 {
969 struct super_block * sb;
970 struct dx_hash_info hinfo;
971 u32 hash;
972 struct dx_frame frames[2], *frame;
973 struct ext4_dir_entry_2 *de, *top;
974 struct buffer_head *bh;
975 ext4_lblk_t block;
976 int retval;
977 int namelen = d_name->len;
978 const u8 *name = d_name->name;
979
980 sb = dir->i_sb;
981 /* NFS may look up ".." - look at dx_root directory block */
982 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
983 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
984 return NULL;
985 } else {
986 frame = frames;
987 frame->bh = NULL; /* for dx_release() */
988 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
989 dx_set_block(frame->at, 0); /* dx_root block is 0 */
990 }
991 hash = hinfo.hash;
992 do {
993 block = dx_get_block(frame->at);
994 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
995 goto errout;
996 de = (struct ext4_dir_entry_2 *) bh->b_data;
997 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
998 EXT4_DIR_REC_LEN(0));
999 for (; de < top; de = ext4_next_entry(de)) {
1000 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1001 + ((char *) de - bh->b_data);
1002
1003 if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1004 brelse(bh);
1005 *err = ERR_BAD_DX_DIR;
1006 goto errout;
1007 }
1008
1009 if (ext4_match(namelen, name, de)) {
1010 *res_dir = de;
1011 dx_release(frames);
1012 return bh;
1013 }
1014 }
1015 brelse(bh);
1016 /* Check to see if we should continue to search */
1017 retval = ext4_htree_next_block(dir, hash, frame,
1018 frames, NULL);
1019 if (retval < 0) {
1020 ext4_warning(sb, __func__,
1021 "error reading index page in directory #%lu",
1022 dir->i_ino);
1023 *err = retval;
1024 goto errout;
1025 }
1026 } while (retval == 1);
1027
1028 *err = -ENOENT;
1029 errout:
1030 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
1031 dx_release (frames);
1032 return NULL;
1033 }
1034
1035 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1036 {
1037 struct inode *inode;
1038 struct ext4_dir_entry_2 *de;
1039 struct buffer_head *bh;
1040
1041 if (dentry->d_name.len > EXT4_NAME_LEN)
1042 return ERR_PTR(-ENAMETOOLONG);
1043
1044 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1045 inode = NULL;
1046 if (bh) {
1047 __u32 ino = le32_to_cpu(de->inode);
1048 brelse(bh);
1049 if (!ext4_valid_inum(dir->i_sb, ino)) {
1050 ext4_error(dir->i_sb, "ext4_lookup",
1051 "bad inode number: %u", ino);
1052 return ERR_PTR(-EIO);
1053 }
1054 inode = ext4_iget(dir->i_sb, ino);
1055 if (IS_ERR(inode))
1056 return ERR_CAST(inode);
1057 }
1058 return d_splice_alias(inode, dentry);
1059 }
1060
1061
1062 struct dentry *ext4_get_parent(struct dentry *child)
1063 {
1064 __u32 ino;
1065 struct inode *inode;
1066 static const struct qstr dotdot = {
1067 .name = "..",
1068 .len = 2,
1069 };
1070 struct ext4_dir_entry_2 * de;
1071 struct buffer_head *bh;
1072
1073 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1074 inode = NULL;
1075 if (!bh)
1076 return ERR_PTR(-ENOENT);
1077 ino = le32_to_cpu(de->inode);
1078 brelse(bh);
1079
1080 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1081 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
1082 "bad inode number: %u", ino);
1083 return ERR_PTR(-EIO);
1084 }
1085
1086 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1087 }
1088
1089 #define S_SHIFT 12
1090 static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1091 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1092 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1093 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1094 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1095 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1096 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1097 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1098 };
1099
1100 static inline void ext4_set_de_type(struct super_block *sb,
1101 struct ext4_dir_entry_2 *de,
1102 umode_t mode) {
1103 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1104 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1105 }
1106
1107 /*
1108 * Move count entries from end of map between two memory locations.
1109 * Returns pointer to last entry moved.
1110 */
1111 static struct ext4_dir_entry_2 *
1112 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1113 {
1114 unsigned rec_len = 0;
1115
1116 while (count--) {
1117 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1118 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1119 memcpy (to, de, rec_len);
1120 ((struct ext4_dir_entry_2 *) to)->rec_len =
1121 ext4_rec_len_to_disk(rec_len);
1122 de->inode = 0;
1123 map++;
1124 to += rec_len;
1125 }
1126 return (struct ext4_dir_entry_2 *) (to - rec_len);
1127 }
1128
1129 /*
1130 * Compact each dir entry in the range to the minimal rec_len.
1131 * Returns pointer to last entry in range.
1132 */
1133 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
1134 {
1135 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1136 unsigned rec_len = 0;
1137
1138 prev = to = de;
1139 while ((char*)de < base + size) {
1140 next = ext4_next_entry(de);
1141 if (de->inode && de->name_len) {
1142 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1143 if (de > to)
1144 memmove(to, de, rec_len);
1145 to->rec_len = ext4_rec_len_to_disk(rec_len);
1146 prev = to;
1147 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1148 }
1149 de = next;
1150 }
1151 return prev;
1152 }
1153
1154 /*
1155 * Split a full leaf block to make room for a new dir entry.
1156 * Allocate a new block, and move entries so that they are approx. equally full.
1157 * Returns pointer to de in block into which the new entry will be inserted.
1158 */
1159 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1160 struct buffer_head **bh,struct dx_frame *frame,
1161 struct dx_hash_info *hinfo, int *error)
1162 {
1163 unsigned blocksize = dir->i_sb->s_blocksize;
1164 unsigned count, continued;
1165 struct buffer_head *bh2;
1166 ext4_lblk_t newblock;
1167 u32 hash2;
1168 struct dx_map_entry *map;
1169 char *data1 = (*bh)->b_data, *data2;
1170 unsigned split, move, size;
1171 struct ext4_dir_entry_2 *de = NULL, *de2;
1172 int err = 0, i;
1173
1174 bh2 = ext4_append (handle, dir, &newblock, &err);
1175 if (!(bh2)) {
1176 brelse(*bh);
1177 *bh = NULL;
1178 goto errout;
1179 }
1180
1181 BUFFER_TRACE(*bh, "get_write_access");
1182 err = ext4_journal_get_write_access(handle, *bh);
1183 if (err)
1184 goto journal_error;
1185
1186 BUFFER_TRACE(frame->bh, "get_write_access");
1187 err = ext4_journal_get_write_access(handle, frame->bh);
1188 if (err)
1189 goto journal_error;
1190
1191 data2 = bh2->b_data;
1192
1193 /* create map in the end of data2 block */
1194 map = (struct dx_map_entry *) (data2 + blocksize);
1195 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1196 blocksize, hinfo, map);
1197 map -= count;
1198 dx_sort_map(map, count);
1199 /* Split the existing block in the middle, size-wise */
1200 size = 0;
1201 move = 0;
1202 for (i = count-1; i >= 0; i--) {
1203 /* is more than half of this entry in 2nd half of the block? */
1204 if (size + map[i].size/2 > blocksize/2)
1205 break;
1206 size += map[i].size;
1207 move++;
1208 }
1209 /* map index at which we will split */
1210 split = count - move;
1211 hash2 = map[split].hash;
1212 continued = hash2 == map[split - 1].hash;
1213 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1214 (unsigned long)dx_get_block(frame->at),
1215 hash2, split, count-split));
1216
1217 /* Fancy dance to stay within two buffers */
1218 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1219 de = dx_pack_dirents(data1, blocksize);
1220 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1221 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2);
1222 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1223 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1224
1225 /* Which block gets the new entry? */
1226 if (hinfo->hash >= hash2)
1227 {
1228 swap(*bh, bh2);
1229 de = de2;
1230 }
1231 dx_insert_block(frame, hash2 + continued, newblock);
1232 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1233 if (err)
1234 goto journal_error;
1235 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1236 if (err)
1237 goto journal_error;
1238 brelse(bh2);
1239 dxtrace(dx_show_index("frame", frame->entries));
1240 return de;
1241
1242 journal_error:
1243 brelse(*bh);
1244 brelse(bh2);
1245 *bh = NULL;
1246 ext4_std_error(dir->i_sb, err);
1247 errout:
1248 *error = err;
1249 return NULL;
1250 }
1251
1252 /*
1253 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1254 * it points to a directory entry which is guaranteed to be large
1255 * enough for new directory entry. If de is NULL, then
1256 * add_dirent_to_buf will attempt search the directory block for
1257 * space. It will return -ENOSPC if no space is available, and -EIO
1258 * and -EEXIST if directory entry already exists.
1259 *
1260 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1261 * all other cases bh is released.
1262 */
1263 static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1264 struct inode *inode, struct ext4_dir_entry_2 *de,
1265 struct buffer_head *bh)
1266 {
1267 struct inode *dir = dentry->d_parent->d_inode;
1268 const char *name = dentry->d_name.name;
1269 int namelen = dentry->d_name.len;
1270 unsigned int offset = 0;
1271 unsigned short reclen;
1272 int nlen, rlen, err;
1273 char *top;
1274
1275 reclen = EXT4_DIR_REC_LEN(namelen);
1276 if (!de) {
1277 de = (struct ext4_dir_entry_2 *)bh->b_data;
1278 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1279 while ((char *) de <= top) {
1280 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
1281 bh, offset)) {
1282 brelse(bh);
1283 return -EIO;
1284 }
1285 if (ext4_match(namelen, name, de)) {
1286 brelse(bh);
1287 return -EEXIST;
1288 }
1289 nlen = EXT4_DIR_REC_LEN(de->name_len);
1290 rlen = ext4_rec_len_from_disk(de->rec_len);
1291 if ((de->inode? rlen - nlen: rlen) >= reclen)
1292 break;
1293 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1294 offset += rlen;
1295 }
1296 if ((char *) de > top)
1297 return -ENOSPC;
1298 }
1299 BUFFER_TRACE(bh, "get_write_access");
1300 err = ext4_journal_get_write_access(handle, bh);
1301 if (err) {
1302 ext4_std_error(dir->i_sb, err);
1303 brelse(bh);
1304 return err;
1305 }
1306
1307 /* By now the buffer is marked for journaling */
1308 nlen = EXT4_DIR_REC_LEN(de->name_len);
1309 rlen = ext4_rec_len_from_disk(de->rec_len);
1310 if (de->inode) {
1311 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1312 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen);
1313 de->rec_len = ext4_rec_len_to_disk(nlen);
1314 de = de1;
1315 }
1316 de->file_type = EXT4_FT_UNKNOWN;
1317 if (inode) {
1318 de->inode = cpu_to_le32(inode->i_ino);
1319 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1320 } else
1321 de->inode = 0;
1322 de->name_len = namelen;
1323 memcpy(de->name, name, namelen);
1324 /*
1325 * XXX shouldn't update any times until successful
1326 * completion of syscall, but too many callers depend
1327 * on this.
1328 *
1329 * XXX similarly, too many callers depend on
1330 * ext4_new_inode() setting the times, but error
1331 * recovery deletes the inode, so the worst that can
1332 * happen is that the times are slightly out of date
1333 * and/or different from the directory change time.
1334 */
1335 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1336 ext4_update_dx_flag(dir);
1337 dir->i_version++;
1338 ext4_mark_inode_dirty(handle, dir);
1339 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1340 err = ext4_handle_dirty_metadata(handle, dir, bh);
1341 if (err)
1342 ext4_std_error(dir->i_sb, err);
1343 brelse(bh);
1344 return 0;
1345 }
1346
1347 /*
1348 * This converts a one block unindexed directory to a 3 block indexed
1349 * directory, and adds the dentry to the indexed directory.
1350 */
1351 static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1352 struct inode *inode, struct buffer_head *bh)
1353 {
1354 struct inode *dir = dentry->d_parent->d_inode;
1355 const char *name = dentry->d_name.name;
1356 int namelen = dentry->d_name.len;
1357 struct buffer_head *bh2;
1358 struct dx_root *root;
1359 struct dx_frame frames[2], *frame;
1360 struct dx_entry *entries;
1361 struct ext4_dir_entry_2 *de, *de2;
1362 char *data1, *top;
1363 unsigned len;
1364 int retval;
1365 unsigned blocksize;
1366 struct dx_hash_info hinfo;
1367 ext4_lblk_t block;
1368 struct fake_dirent *fde;
1369
1370 blocksize = dir->i_sb->s_blocksize;
1371 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1372 retval = ext4_journal_get_write_access(handle, bh);
1373 if (retval) {
1374 ext4_std_error(dir->i_sb, retval);
1375 brelse(bh);
1376 return retval;
1377 }
1378 root = (struct dx_root *) bh->b_data;
1379
1380 /* The 0th block becomes the root, move the dirents out */
1381 fde = &root->dotdot;
1382 de = (struct ext4_dir_entry_2 *)((char *)fde +
1383 ext4_rec_len_from_disk(fde->rec_len));
1384 if ((char *) de >= (((char *) root) + blocksize)) {
1385 ext4_error(dir->i_sb, __func__,
1386 "invalid rec_len for '..' in inode %lu",
1387 dir->i_ino);
1388 brelse(bh);
1389 return -EIO;
1390 }
1391 len = ((char *) root) + blocksize - (char *) de;
1392
1393 /* Allocate new block for the 0th block's dirents */
1394 bh2 = ext4_append(handle, dir, &block, &retval);
1395 if (!(bh2)) {
1396 brelse(bh);
1397 return retval;
1398 }
1399 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
1400 data1 = bh2->b_data;
1401
1402 memcpy (data1, de, len);
1403 de = (struct ext4_dir_entry_2 *) data1;
1404 top = data1 + len;
1405 while ((char *)(de2 = ext4_next_entry(de)) < top)
1406 de = de2;
1407 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1408 /* Initialize the root; the dot dirents already exist */
1409 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1410 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2));
1411 memset (&root->info, 0, sizeof(root->info));
1412 root->info.info_length = sizeof(root->info);
1413 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1414 entries = root->entries;
1415 dx_set_block(entries, 1);
1416 dx_set_count(entries, 1);
1417 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1418
1419 /* Initialize as for dx_probe */
1420 hinfo.hash_version = root->info.hash_version;
1421 if (hinfo.hash_version <= DX_HASH_TEA)
1422 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1423 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1424 ext4fs_dirhash(name, namelen, &hinfo);
1425 frame = frames;
1426 frame->entries = entries;
1427 frame->at = entries;
1428 frame->bh = bh;
1429 bh = bh2;
1430 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1431 dx_release (frames);
1432 if (!(de))
1433 return retval;
1434
1435 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1436 }
1437
1438 /*
1439 * ext4_add_entry()
1440 *
1441 * adds a file entry to the specified directory, using the same
1442 * semantics as ext4_find_entry(). It returns NULL if it failed.
1443 *
1444 * NOTE!! The inode part of 'de' is left at 0 - which means you
1445 * may not sleep between calling this and putting something into
1446 * the entry, as someone else might have used it while you slept.
1447 */
1448 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1449 struct inode *inode)
1450 {
1451 struct inode *dir = dentry->d_parent->d_inode;
1452 struct buffer_head *bh;
1453 struct ext4_dir_entry_2 *de;
1454 struct super_block *sb;
1455 int retval;
1456 int dx_fallback=0;
1457 unsigned blocksize;
1458 ext4_lblk_t block, blocks;
1459
1460 sb = dir->i_sb;
1461 blocksize = sb->s_blocksize;
1462 if (!dentry->d_name.len)
1463 return -EINVAL;
1464 if (is_dx(dir)) {
1465 retval = ext4_dx_add_entry(handle, dentry, inode);
1466 if (!retval || (retval != ERR_BAD_DX_DIR))
1467 return retval;
1468 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
1469 dx_fallback++;
1470 ext4_mark_inode_dirty(handle, dir);
1471 }
1472 blocks = dir->i_size >> sb->s_blocksize_bits;
1473 for (block = 0; block < blocks; block++) {
1474 bh = ext4_bread(handle, dir, block, 0, &retval);
1475 if(!bh)
1476 return retval;
1477 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1478 if (retval != -ENOSPC)
1479 return retval;
1480
1481 if (blocks == 1 && !dx_fallback &&
1482 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1483 return make_indexed_dir(handle, dentry, inode, bh);
1484 brelse(bh);
1485 }
1486 bh = ext4_append(handle, dir, &block, &retval);
1487 if (!bh)
1488 return retval;
1489 de = (struct ext4_dir_entry_2 *) bh->b_data;
1490 de->inode = 0;
1491 de->rec_len = ext4_rec_len_to_disk(blocksize);
1492 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1493 }
1494
1495 /*
1496 * Returns 0 for success, or a negative error value
1497 */
1498 static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1499 struct inode *inode)
1500 {
1501 struct dx_frame frames[2], *frame;
1502 struct dx_entry *entries, *at;
1503 struct dx_hash_info hinfo;
1504 struct buffer_head *bh;
1505 struct inode *dir = dentry->d_parent->d_inode;
1506 struct super_block *sb = dir->i_sb;
1507 struct ext4_dir_entry_2 *de;
1508 int err;
1509
1510 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1511 if (!frame)
1512 return err;
1513 entries = frame->entries;
1514 at = frame->at;
1515
1516 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1517 goto cleanup;
1518
1519 BUFFER_TRACE(bh, "get_write_access");
1520 err = ext4_journal_get_write_access(handle, bh);
1521 if (err)
1522 goto journal_error;
1523
1524 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1525 if (err != -ENOSPC) {
1526 bh = NULL;
1527 goto cleanup;
1528 }
1529
1530 /* Block full, should compress but for now just split */
1531 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1532 dx_get_count(entries), dx_get_limit(entries)));
1533 /* Need to split index? */
1534 if (dx_get_count(entries) == dx_get_limit(entries)) {
1535 ext4_lblk_t newblock;
1536 unsigned icount = dx_get_count(entries);
1537 int levels = frame - frames;
1538 struct dx_entry *entries2;
1539 struct dx_node *node2;
1540 struct buffer_head *bh2;
1541
1542 if (levels && (dx_get_count(frames->entries) ==
1543 dx_get_limit(frames->entries))) {
1544 ext4_warning(sb, __func__,
1545 "Directory index full!");
1546 err = -ENOSPC;
1547 goto cleanup;
1548 }
1549 bh2 = ext4_append (handle, dir, &newblock, &err);
1550 if (!(bh2))
1551 goto cleanup;
1552 node2 = (struct dx_node *)(bh2->b_data);
1553 entries2 = node2->entries;
1554 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize);
1555 node2->fake.inode = 0;
1556 BUFFER_TRACE(frame->bh, "get_write_access");
1557 err = ext4_journal_get_write_access(handle, frame->bh);
1558 if (err)
1559 goto journal_error;
1560 if (levels) {
1561 unsigned icount1 = icount/2, icount2 = icount - icount1;
1562 unsigned hash2 = dx_get_hash(entries + icount1);
1563 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1564 icount1, icount2));
1565
1566 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1567 err = ext4_journal_get_write_access(handle,
1568 frames[0].bh);
1569 if (err)
1570 goto journal_error;
1571
1572 memcpy((char *) entries2, (char *) (entries + icount1),
1573 icount2 * sizeof(struct dx_entry));
1574 dx_set_count(entries, icount1);
1575 dx_set_count(entries2, icount2);
1576 dx_set_limit(entries2, dx_node_limit(dir));
1577
1578 /* Which index block gets the new entry? */
1579 if (at - entries >= icount1) {
1580 frame->at = at = at - entries - icount1 + entries2;
1581 frame->entries = entries = entries2;
1582 swap(frame->bh, bh2);
1583 }
1584 dx_insert_block(frames + 0, hash2, newblock);
1585 dxtrace(dx_show_index("node", frames[1].entries));
1586 dxtrace(dx_show_index("node",
1587 ((struct dx_node *) bh2->b_data)->entries));
1588 err = ext4_handle_dirty_metadata(handle, inode, bh2);
1589 if (err)
1590 goto journal_error;
1591 brelse (bh2);
1592 } else {
1593 dxtrace(printk(KERN_DEBUG
1594 "Creating second level index...\n"));
1595 memcpy((char *) entries2, (char *) entries,
1596 icount * sizeof(struct dx_entry));
1597 dx_set_limit(entries2, dx_node_limit(dir));
1598
1599 /* Set up root */
1600 dx_set_count(entries, 1);
1601 dx_set_block(entries + 0, newblock);
1602 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1603
1604 /* Add new access path frame */
1605 frame = frames + 1;
1606 frame->at = at = at - entries + entries2;
1607 frame->entries = entries = entries2;
1608 frame->bh = bh2;
1609 err = ext4_journal_get_write_access(handle,
1610 frame->bh);
1611 if (err)
1612 goto journal_error;
1613 }
1614 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1615 }
1616 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1617 if (!de)
1618 goto cleanup;
1619 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1620 bh = NULL;
1621 goto cleanup;
1622
1623 journal_error:
1624 ext4_std_error(dir->i_sb, err);
1625 cleanup:
1626 if (bh)
1627 brelse(bh);
1628 dx_release(frames);
1629 return err;
1630 }
1631
1632 /*
1633 * ext4_delete_entry deletes a directory entry by merging it with the
1634 * previous entry
1635 */
1636 static int ext4_delete_entry(handle_t *handle,
1637 struct inode *dir,
1638 struct ext4_dir_entry_2 *de_del,
1639 struct buffer_head *bh)
1640 {
1641 struct ext4_dir_entry_2 *de, *pde;
1642 int i;
1643
1644 i = 0;
1645 pde = NULL;
1646 de = (struct ext4_dir_entry_2 *) bh->b_data;
1647 while (i < bh->b_size) {
1648 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
1649 return -EIO;
1650 if (de == de_del) {
1651 BUFFER_TRACE(bh, "get_write_access");
1652 ext4_journal_get_write_access(handle, bh);
1653 if (pde)
1654 pde->rec_len = ext4_rec_len_to_disk(
1655 ext4_rec_len_from_disk(pde->rec_len) +
1656 ext4_rec_len_from_disk(de->rec_len));
1657 else
1658 de->inode = 0;
1659 dir->i_version++;
1660 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1661 ext4_handle_dirty_metadata(handle, dir, bh);
1662 return 0;
1663 }
1664 i += ext4_rec_len_from_disk(de->rec_len);
1665 pde = de;
1666 de = ext4_next_entry(de);
1667 }
1668 return -ENOENT;
1669 }
1670
1671 /*
1672 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1673 * since this indicates that nlinks count was previously 1.
1674 */
1675 static void ext4_inc_count(handle_t *handle, struct inode *inode)
1676 {
1677 inc_nlink(inode);
1678 if (is_dx(inode) && inode->i_nlink > 1) {
1679 /* limit is 16-bit i_links_count */
1680 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1681 inode->i_nlink = 1;
1682 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1683 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1684 }
1685 }
1686 }
1687
1688 /*
1689 * If a directory had nlink == 1, then we should let it be 1. This indicates
1690 * directory has >EXT4_LINK_MAX subdirs.
1691 */
1692 static void ext4_dec_count(handle_t *handle, struct inode *inode)
1693 {
1694 drop_nlink(inode);
1695 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1696 inc_nlink(inode);
1697 }
1698
1699
1700 static int ext4_add_nondir(handle_t *handle,
1701 struct dentry *dentry, struct inode *inode)
1702 {
1703 int err = ext4_add_entry(handle, dentry, inode);
1704 if (!err) {
1705 ext4_mark_inode_dirty(handle, inode);
1706 d_instantiate(dentry, inode);
1707 unlock_new_inode(inode);
1708 return 0;
1709 }
1710 drop_nlink(inode);
1711 unlock_new_inode(inode);
1712 iput(inode);
1713 return err;
1714 }
1715
1716 /*
1717 * By the time this is called, we already have created
1718 * the directory cache entry for the new file, but it
1719 * is so far negative - it has no inode.
1720 *
1721 * If the create succeeds, we fill in the inode information
1722 * with d_instantiate().
1723 */
1724 static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1725 struct nameidata *nd)
1726 {
1727 handle_t *handle;
1728 struct inode *inode;
1729 int err, retries = 0;
1730
1731 retry:
1732 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1733 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1734 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1735 if (IS_ERR(handle))
1736 return PTR_ERR(handle);
1737
1738 if (IS_DIRSYNC(dir))
1739 ext4_handle_sync(handle);
1740
1741 inode = ext4_new_inode (handle, dir, mode);
1742 err = PTR_ERR(inode);
1743 if (!IS_ERR(inode)) {
1744 inode->i_op = &ext4_file_inode_operations;
1745 inode->i_fop = &ext4_file_operations;
1746 ext4_set_aops(inode);
1747 err = ext4_add_nondir(handle, dentry, inode);
1748 }
1749 ext4_journal_stop(handle);
1750 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1751 goto retry;
1752 return err;
1753 }
1754
1755 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1756 int mode, dev_t rdev)
1757 {
1758 handle_t *handle;
1759 struct inode *inode;
1760 int err, retries = 0;
1761
1762 if (!new_valid_dev(rdev))
1763 return -EINVAL;
1764
1765 retry:
1766 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1767 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1768 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1769 if (IS_ERR(handle))
1770 return PTR_ERR(handle);
1771
1772 if (IS_DIRSYNC(dir))
1773 ext4_handle_sync(handle);
1774
1775 inode = ext4_new_inode(handle, dir, mode);
1776 err = PTR_ERR(inode);
1777 if (!IS_ERR(inode)) {
1778 init_special_inode(inode, inode->i_mode, rdev);
1779 #ifdef CONFIG_EXT4_FS_XATTR
1780 inode->i_op = &ext4_special_inode_operations;
1781 #endif
1782 err = ext4_add_nondir(handle, dentry, inode);
1783 }
1784 ext4_journal_stop(handle);
1785 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1786 goto retry;
1787 return err;
1788 }
1789
1790 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1791 {
1792 handle_t *handle;
1793 struct inode *inode;
1794 struct buffer_head *dir_block;
1795 struct ext4_dir_entry_2 *de;
1796 int err, retries = 0;
1797
1798 if (EXT4_DIR_LINK_MAX(dir))
1799 return -EMLINK;
1800
1801 retry:
1802 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1803 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1804 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
1805 if (IS_ERR(handle))
1806 return PTR_ERR(handle);
1807
1808 if (IS_DIRSYNC(dir))
1809 ext4_handle_sync(handle);
1810
1811 inode = ext4_new_inode(handle, dir, S_IFDIR | mode);
1812 err = PTR_ERR(inode);
1813 if (IS_ERR(inode))
1814 goto out_stop;
1815
1816 inode->i_op = &ext4_dir_inode_operations;
1817 inode->i_fop = &ext4_dir_operations;
1818 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1819 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1820 if (!dir_block)
1821 goto out_clear_inode;
1822 BUFFER_TRACE(dir_block, "get_write_access");
1823 ext4_journal_get_write_access(handle, dir_block);
1824 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1825 de->inode = cpu_to_le32(inode->i_ino);
1826 de->name_len = 1;
1827 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len));
1828 strcpy(de->name, ".");
1829 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1830 de = ext4_next_entry(de);
1831 de->inode = cpu_to_le32(dir->i_ino);
1832 de->rec_len = ext4_rec_len_to_disk(inode->i_sb->s_blocksize -
1833 EXT4_DIR_REC_LEN(1));
1834 de->name_len = 2;
1835 strcpy(de->name, "..");
1836 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1837 inode->i_nlink = 2;
1838 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1839 ext4_handle_dirty_metadata(handle, dir, dir_block);
1840 brelse(dir_block);
1841 ext4_mark_inode_dirty(handle, inode);
1842 err = ext4_add_entry(handle, dentry, inode);
1843 if (err) {
1844 out_clear_inode:
1845 clear_nlink(inode);
1846 unlock_new_inode(inode);
1847 ext4_mark_inode_dirty(handle, inode);
1848 iput(inode);
1849 goto out_stop;
1850 }
1851 ext4_inc_count(handle, dir);
1852 ext4_update_dx_flag(dir);
1853 ext4_mark_inode_dirty(handle, dir);
1854 d_instantiate(dentry, inode);
1855 unlock_new_inode(inode);
1856 out_stop:
1857 ext4_journal_stop(handle);
1858 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1859 goto retry;
1860 return err;
1861 }
1862
1863 /*
1864 * routine to check that the specified directory is empty (for rmdir)
1865 */
1866 static int empty_dir(struct inode *inode)
1867 {
1868 unsigned int offset;
1869 struct buffer_head *bh;
1870 struct ext4_dir_entry_2 *de, *de1;
1871 struct super_block *sb;
1872 int err = 0;
1873
1874 sb = inode->i_sb;
1875 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1876 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1877 if (err)
1878 ext4_error(inode->i_sb, __func__,
1879 "error %d reading directory #%lu offset 0",
1880 err, inode->i_ino);
1881 else
1882 ext4_warning(inode->i_sb, __func__,
1883 "bad directory (dir #%lu) - no data block",
1884 inode->i_ino);
1885 return 1;
1886 }
1887 de = (struct ext4_dir_entry_2 *) bh->b_data;
1888 de1 = ext4_next_entry(de);
1889 if (le32_to_cpu(de->inode) != inode->i_ino ||
1890 !le32_to_cpu(de1->inode) ||
1891 strcmp(".", de->name) ||
1892 strcmp("..", de1->name)) {
1893 ext4_warning(inode->i_sb, "empty_dir",
1894 "bad directory (dir #%lu) - no `.' or `..'",
1895 inode->i_ino);
1896 brelse(bh);
1897 return 1;
1898 }
1899 offset = ext4_rec_len_from_disk(de->rec_len) +
1900 ext4_rec_len_from_disk(de1->rec_len);
1901 de = ext4_next_entry(de1);
1902 while (offset < inode->i_size) {
1903 if (!bh ||
1904 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1905 err = 0;
1906 brelse(bh);
1907 bh = ext4_bread(NULL, inode,
1908 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
1909 if (!bh) {
1910 if (err)
1911 ext4_error(sb, __func__,
1912 "error %d reading directory"
1913 " #%lu offset %u",
1914 err, inode->i_ino, offset);
1915 offset += sb->s_blocksize;
1916 continue;
1917 }
1918 de = (struct ext4_dir_entry_2 *) bh->b_data;
1919 }
1920 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1921 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1922 sb->s_blocksize);
1923 offset = (offset | (sb->s_blocksize - 1)) + 1;
1924 continue;
1925 }
1926 if (le32_to_cpu(de->inode)) {
1927 brelse(bh);
1928 return 0;
1929 }
1930 offset += ext4_rec_len_from_disk(de->rec_len);
1931 de = ext4_next_entry(de);
1932 }
1933 brelse(bh);
1934 return 1;
1935 }
1936
1937 /* ext4_orphan_add() links an unlinked or truncated inode into a list of
1938 * such inodes, starting at the superblock, in case we crash before the
1939 * file is closed/deleted, or in case the inode truncate spans multiple
1940 * transactions and the last transaction is not recovered after a crash.
1941 *
1942 * At filesystem recovery time, we walk this list deleting unlinked
1943 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1944 */
1945 int ext4_orphan_add(handle_t *handle, struct inode *inode)
1946 {
1947 struct super_block *sb = inode->i_sb;
1948 struct ext4_iloc iloc;
1949 int err = 0, rc;
1950
1951 if (!ext4_handle_valid(handle))
1952 return 0;
1953
1954 lock_super(sb);
1955 if (!list_empty(&EXT4_I(inode)->i_orphan))
1956 goto out_unlock;
1957
1958 /* Orphan handling is only valid for files with data blocks
1959 * being truncated, or files being unlinked. */
1960
1961 /* @@@ FIXME: Observation from aviro:
1962 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1963 * here (on lock_super()), so race with ext4_link() which might bump
1964 * ->i_nlink. For, say it, character device. Not a regular file,
1965 * not a directory, not a symlink and ->i_nlink > 0.
1966 */
1967 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1968 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1969
1970 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1971 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1972 if (err)
1973 goto out_unlock;
1974
1975 err = ext4_reserve_inode_write(handle, inode, &iloc);
1976 if (err)
1977 goto out_unlock;
1978
1979 /* Insert this inode at the head of the on-disk orphan list... */
1980 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1981 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1982 err = ext4_handle_dirty_metadata(handle, inode, EXT4_SB(sb)->s_sbh);
1983 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
1984 if (!err)
1985 err = rc;
1986
1987 /* Only add to the head of the in-memory list if all the
1988 * previous operations succeeded. If the orphan_add is going to
1989 * fail (possibly taking the journal offline), we can't risk
1990 * leaving the inode on the orphan list: stray orphan-list
1991 * entries can cause panics at unmount time.
1992 *
1993 * This is safe: on error we're going to ignore the orphan list
1994 * anyway on the next recovery. */
1995 if (!err)
1996 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1997
1998 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1999 jbd_debug(4, "orphan inode %lu will point to %d\n",
2000 inode->i_ino, NEXT_ORPHAN(inode));
2001 out_unlock:
2002 unlock_super(sb);
2003 ext4_std_error(inode->i_sb, err);
2004 return err;
2005 }
2006
2007 /*
2008 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2009 * of such inodes stored on disk, because it is finally being cleaned up.
2010 */
2011 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2012 {
2013 struct list_head *prev;
2014 struct ext4_inode_info *ei = EXT4_I(inode);
2015 struct ext4_sb_info *sbi;
2016 __u32 ino_next;
2017 struct ext4_iloc iloc;
2018 int err = 0;
2019
2020 if (!ext4_handle_valid(handle))
2021 return 0;
2022
2023 lock_super(inode->i_sb);
2024 if (list_empty(&ei->i_orphan)) {
2025 unlock_super(inode->i_sb);
2026 return 0;
2027 }
2028
2029 ino_next = NEXT_ORPHAN(inode);
2030 prev = ei->i_orphan.prev;
2031 sbi = EXT4_SB(inode->i_sb);
2032
2033 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2034
2035 list_del_init(&ei->i_orphan);
2036
2037 /* If we're on an error path, we may not have a valid
2038 * transaction handle with which to update the orphan list on
2039 * disk, but we still need to remove the inode from the linked
2040 * list in memory. */
2041 if (sbi->s_journal && !handle)
2042 goto out;
2043
2044 err = ext4_reserve_inode_write(handle, inode, &iloc);
2045 if (err)
2046 goto out_err;
2047
2048 if (prev == &sbi->s_orphan) {
2049 jbd_debug(4, "superblock will point to %u\n", ino_next);
2050 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2051 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2052 if (err)
2053 goto out_brelse;
2054 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2055 err = ext4_handle_dirty_metadata(handle, inode, sbi->s_sbh);
2056 } else {
2057 struct ext4_iloc iloc2;
2058 struct inode *i_prev =
2059 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2060
2061 jbd_debug(4, "orphan inode %lu will point to %u\n",
2062 i_prev->i_ino, ino_next);
2063 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2064 if (err)
2065 goto out_brelse;
2066 NEXT_ORPHAN(i_prev) = ino_next;
2067 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2068 }
2069 if (err)
2070 goto out_brelse;
2071 NEXT_ORPHAN(inode) = 0;
2072 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2073
2074 out_err:
2075 ext4_std_error(inode->i_sb, err);
2076 out:
2077 unlock_super(inode->i_sb);
2078 return err;
2079
2080 out_brelse:
2081 brelse(iloc.bh);
2082 goto out_err;
2083 }
2084
2085 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2086 {
2087 int retval;
2088 struct inode *inode;
2089 struct buffer_head *bh;
2090 struct ext4_dir_entry_2 *de;
2091 handle_t *handle;
2092
2093 /* Initialize quotas before so that eventual writes go in
2094 * separate transaction */
2095 vfs_dq_init(dentry->d_inode);
2096 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2097 if (IS_ERR(handle))
2098 return PTR_ERR(handle);
2099
2100 retval = -ENOENT;
2101 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2102 if (!bh)
2103 goto end_rmdir;
2104
2105 if (IS_DIRSYNC(dir))
2106 ext4_handle_sync(handle);
2107
2108 inode = dentry->d_inode;
2109
2110 retval = -EIO;
2111 if (le32_to_cpu(de->inode) != inode->i_ino)
2112 goto end_rmdir;
2113
2114 retval = -ENOTEMPTY;
2115 if (!empty_dir(inode))
2116 goto end_rmdir;
2117
2118 retval = ext4_delete_entry(handle, dir, de, bh);
2119 if (retval)
2120 goto end_rmdir;
2121 if (!EXT4_DIR_LINK_EMPTY(inode))
2122 ext4_warning(inode->i_sb, "ext4_rmdir",
2123 "empty directory has too many links (%d)",
2124 inode->i_nlink);
2125 inode->i_version++;
2126 clear_nlink(inode);
2127 /* There's no need to set i_disksize: the fact that i_nlink is
2128 * zero will ensure that the right thing happens during any
2129 * recovery. */
2130 inode->i_size = 0;
2131 ext4_orphan_add(handle, inode);
2132 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2133 ext4_mark_inode_dirty(handle, inode);
2134 ext4_dec_count(handle, dir);
2135 ext4_update_dx_flag(dir);
2136 ext4_mark_inode_dirty(handle, dir);
2137
2138 end_rmdir:
2139 ext4_journal_stop(handle);
2140 brelse(bh);
2141 return retval;
2142 }
2143
2144 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2145 {
2146 int retval;
2147 struct inode *inode;
2148 struct buffer_head *bh;
2149 struct ext4_dir_entry_2 *de;
2150 handle_t *handle;
2151
2152 /* Initialize quotas before so that eventual writes go
2153 * in separate transaction */
2154 vfs_dq_init(dentry->d_inode);
2155 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2156 if (IS_ERR(handle))
2157 return PTR_ERR(handle);
2158
2159 if (IS_DIRSYNC(dir))
2160 ext4_handle_sync(handle);
2161
2162 retval = -ENOENT;
2163 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2164 if (!bh)
2165 goto end_unlink;
2166
2167 inode = dentry->d_inode;
2168
2169 retval = -EIO;
2170 if (le32_to_cpu(de->inode) != inode->i_ino)
2171 goto end_unlink;
2172
2173 if (!inode->i_nlink) {
2174 ext4_warning(inode->i_sb, "ext4_unlink",
2175 "Deleting nonexistent file (%lu), %d",
2176 inode->i_ino, inode->i_nlink);
2177 inode->i_nlink = 1;
2178 }
2179 retval = ext4_delete_entry(handle, dir, de, bh);
2180 if (retval)
2181 goto end_unlink;
2182 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2183 ext4_update_dx_flag(dir);
2184 ext4_mark_inode_dirty(handle, dir);
2185 drop_nlink(inode);
2186 if (!inode->i_nlink)
2187 ext4_orphan_add(handle, inode);
2188 inode->i_ctime = ext4_current_time(inode);
2189 ext4_mark_inode_dirty(handle, inode);
2190 retval = 0;
2191
2192 end_unlink:
2193 ext4_journal_stop(handle);
2194 brelse(bh);
2195 return retval;
2196 }
2197
2198 static int ext4_symlink(struct inode *dir,
2199 struct dentry *dentry, const char *symname)
2200 {
2201 handle_t *handle;
2202 struct inode *inode;
2203 int l, err, retries = 0;
2204
2205 l = strlen(symname)+1;
2206 if (l > dir->i_sb->s_blocksize)
2207 return -ENAMETOOLONG;
2208
2209 retry:
2210 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2211 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2212 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
2213 if (IS_ERR(handle))
2214 return PTR_ERR(handle);
2215
2216 if (IS_DIRSYNC(dir))
2217 ext4_handle_sync(handle);
2218
2219 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO);
2220 err = PTR_ERR(inode);
2221 if (IS_ERR(inode))
2222 goto out_stop;
2223
2224 if (l > sizeof(EXT4_I(inode)->i_data)) {
2225 inode->i_op = &ext4_symlink_inode_operations;
2226 ext4_set_aops(inode);
2227 /*
2228 * page_symlink() calls into ext4_prepare/commit_write.
2229 * We have a transaction open. All is sweetness. It also sets
2230 * i_size in generic_commit_write().
2231 */
2232 err = __page_symlink(inode, symname, l, 1);
2233 if (err) {
2234 clear_nlink(inode);
2235 unlock_new_inode(inode);
2236 ext4_mark_inode_dirty(handle, inode);
2237 iput(inode);
2238 goto out_stop;
2239 }
2240 } else {
2241 /* clear the extent format for fast symlink */
2242 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
2243 inode->i_op = &ext4_fast_symlink_inode_operations;
2244 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2245 inode->i_size = l-1;
2246 }
2247 EXT4_I(inode)->i_disksize = inode->i_size;
2248 err = ext4_add_nondir(handle, dentry, inode);
2249 out_stop:
2250 ext4_journal_stop(handle);
2251 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2252 goto retry;
2253 return err;
2254 }
2255
2256 static int ext4_link(struct dentry *old_dentry,
2257 struct inode *dir, struct dentry *dentry)
2258 {
2259 handle_t *handle;
2260 struct inode *inode = old_dentry->d_inode;
2261 int err, retries = 0;
2262
2263 if (EXT4_DIR_LINK_MAX(inode))
2264 return -EMLINK;
2265
2266 /*
2267 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2268 * otherwise has the potential to corrupt the orphan inode list.
2269 */
2270 if (inode->i_nlink == 0)
2271 return -ENOENT;
2272
2273 retry:
2274 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2275 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2276 if (IS_ERR(handle))
2277 return PTR_ERR(handle);
2278
2279 if (IS_DIRSYNC(dir))
2280 ext4_handle_sync(handle);
2281
2282 inode->i_ctime = ext4_current_time(inode);
2283 ext4_inc_count(handle, inode);
2284 atomic_inc(&inode->i_count);
2285
2286 err = ext4_add_entry(handle, dentry, inode);
2287 if (!err) {
2288 ext4_mark_inode_dirty(handle, inode);
2289 d_instantiate(dentry, inode);
2290 } else {
2291 drop_nlink(inode);
2292 iput(inode);
2293 }
2294 ext4_journal_stop(handle);
2295 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2296 goto retry;
2297 return err;
2298 }
2299
2300 #define PARENT_INO(buffer) \
2301 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
2302
2303 /*
2304 * Anybody can rename anything with this: the permission checks are left to the
2305 * higher-level routines.
2306 */
2307 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2308 struct inode *new_dir, struct dentry *new_dentry)
2309 {
2310 handle_t *handle;
2311 struct inode *old_inode, *new_inode;
2312 struct buffer_head *old_bh, *new_bh, *dir_bh;
2313 struct ext4_dir_entry_2 *old_de, *new_de;
2314 int retval;
2315
2316 old_bh = new_bh = dir_bh = NULL;
2317
2318 /* Initialize quotas before so that eventual writes go
2319 * in separate transaction */
2320 if (new_dentry->d_inode)
2321 vfs_dq_init(new_dentry->d_inode);
2322 handle = ext4_journal_start(old_dir, 2 *
2323 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2324 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2325 if (IS_ERR(handle))
2326 return PTR_ERR(handle);
2327
2328 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2329 ext4_handle_sync(handle);
2330
2331 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2332 /*
2333 * Check for inode number is _not_ due to possible IO errors.
2334 * We might rmdir the source, keep it as pwd of some process
2335 * and merrily kill the link to whatever was created under the
2336 * same name. Goodbye sticky bit ;-<
2337 */
2338 old_inode = old_dentry->d_inode;
2339 retval = -ENOENT;
2340 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2341 goto end_rename;
2342
2343 new_inode = new_dentry->d_inode;
2344 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2345 if (new_bh) {
2346 if (!new_inode) {
2347 brelse(new_bh);
2348 new_bh = NULL;
2349 }
2350 }
2351 if (S_ISDIR(old_inode->i_mode)) {
2352 if (new_inode) {
2353 retval = -ENOTEMPTY;
2354 if (!empty_dir(new_inode))
2355 goto end_rename;
2356 }
2357 retval = -EIO;
2358 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2359 if (!dir_bh)
2360 goto end_rename;
2361 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2362 goto end_rename;
2363 retval = -EMLINK;
2364 if (!new_inode && new_dir != old_dir &&
2365 new_dir->i_nlink >= EXT4_LINK_MAX)
2366 goto end_rename;
2367 }
2368 if (!new_bh) {
2369 retval = ext4_add_entry(handle, new_dentry, old_inode);
2370 if (retval)
2371 goto end_rename;
2372 } else {
2373 BUFFER_TRACE(new_bh, "get write access");
2374 ext4_journal_get_write_access(handle, new_bh);
2375 new_de->inode = cpu_to_le32(old_inode->i_ino);
2376 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2377 EXT4_FEATURE_INCOMPAT_FILETYPE))
2378 new_de->file_type = old_de->file_type;
2379 new_dir->i_version++;
2380 new_dir->i_ctime = new_dir->i_mtime =
2381 ext4_current_time(new_dir);
2382 ext4_mark_inode_dirty(handle, new_dir);
2383 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2384 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2385 brelse(new_bh);
2386 new_bh = NULL;
2387 }
2388
2389 /*
2390 * Like most other Unix systems, set the ctime for inodes on a
2391 * rename.
2392 */
2393 old_inode->i_ctime = ext4_current_time(old_inode);
2394 ext4_mark_inode_dirty(handle, old_inode);
2395
2396 /*
2397 * ok, that's it
2398 */
2399 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2400 old_de->name_len != old_dentry->d_name.len ||
2401 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2402 (retval = ext4_delete_entry(handle, old_dir,
2403 old_de, old_bh)) == -ENOENT) {
2404 /* old_de could have moved from under us during htree split, so
2405 * make sure that we are deleting the right entry. We might
2406 * also be pointing to a stale entry in the unused part of
2407 * old_bh so just checking inum and the name isn't enough. */
2408 struct buffer_head *old_bh2;
2409 struct ext4_dir_entry_2 *old_de2;
2410
2411 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2412 if (old_bh2) {
2413 retval = ext4_delete_entry(handle, old_dir,
2414 old_de2, old_bh2);
2415 brelse(old_bh2);
2416 }
2417 }
2418 if (retval) {
2419 ext4_warning(old_dir->i_sb, "ext4_rename",
2420 "Deleting old file (%lu), %d, error=%d",
2421 old_dir->i_ino, old_dir->i_nlink, retval);
2422 }
2423
2424 if (new_inode) {
2425 ext4_dec_count(handle, new_inode);
2426 new_inode->i_ctime = ext4_current_time(new_inode);
2427 }
2428 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2429 ext4_update_dx_flag(old_dir);
2430 if (dir_bh) {
2431 BUFFER_TRACE(dir_bh, "get_write_access");
2432 ext4_journal_get_write_access(handle, dir_bh);
2433 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2434 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2435 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2436 ext4_dec_count(handle, old_dir);
2437 if (new_inode) {
2438 /* checked empty_dir above, can't have another parent,
2439 * ext4_dec_count() won't work for many-linked dirs */
2440 new_inode->i_nlink = 0;
2441 } else {
2442 ext4_inc_count(handle, new_dir);
2443 ext4_update_dx_flag(new_dir);
2444 ext4_mark_inode_dirty(handle, new_dir);
2445 }
2446 }
2447 ext4_mark_inode_dirty(handle, old_dir);
2448 if (new_inode) {
2449 ext4_mark_inode_dirty(handle, new_inode);
2450 if (!new_inode->i_nlink)
2451 ext4_orphan_add(handle, new_inode);
2452 }
2453 retval = 0;
2454
2455 end_rename:
2456 brelse(dir_bh);
2457 brelse(old_bh);
2458 brelse(new_bh);
2459 ext4_journal_stop(handle);
2460 return retval;
2461 }
2462
2463 /*
2464 * directories can handle most operations...
2465 */
2466 const struct inode_operations ext4_dir_inode_operations = {
2467 .create = ext4_create,
2468 .lookup = ext4_lookup,
2469 .link = ext4_link,
2470 .unlink = ext4_unlink,
2471 .symlink = ext4_symlink,
2472 .mkdir = ext4_mkdir,
2473 .rmdir = ext4_rmdir,
2474 .mknod = ext4_mknod,
2475 .rename = ext4_rename,
2476 .setattr = ext4_setattr,
2477 #ifdef CONFIG_EXT4_FS_XATTR
2478 .setxattr = generic_setxattr,
2479 .getxattr = generic_getxattr,
2480 .listxattr = ext4_listxattr,
2481 .removexattr = generic_removexattr,
2482 #endif
2483 .permission = ext4_permission,
2484 };
2485
2486 const struct inode_operations ext4_special_inode_operations = {
2487 .setattr = ext4_setattr,
2488 #ifdef CONFIG_EXT4_FS_XATTR
2489 .setxattr = generic_setxattr,
2490 .getxattr = generic_getxattr,
2491 .listxattr = ext4_listxattr,
2492 .removexattr = generic_removexattr,
2493 #endif
2494 .permission = ext4_permission,
2495 };