Loading...
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#include <trace/events/ext4.h>
44/*
45 * define how far ahead to read directories while searching them.
46 */
47#define NAMEI_RA_CHUNKS 2
48#define NAMEI_RA_BLOCKS 4
49#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51
52static struct buffer_head *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
83struct fake_dirent
84{
85 __le32 inode;
86 __le16 rec_len;
87 u8 name_len;
88 u8 file_type;
89};
90
91struct dx_countlimit
92{
93 __le16 limit;
94 __le16 count;
95};
96
97struct dx_entry
98{
99 __le32 hash;
100 __le32 block;
101};
102
103/*
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
107 */
108
109struct dx_root
110{
111 struct fake_dirent dot;
112 char dot_name[4];
113 struct fake_dirent dotdot;
114 char dotdot_name[4];
115 struct dx_root_info
116 {
117 __le32 reserved_zero;
118 u8 hash_version;
119 u8 info_length; /* 8 */
120 u8 indirect_levels;
121 u8 unused_flags;
122 }
123 info;
124 struct dx_entry entries[0];
125};
126
127struct dx_node
128{
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
131};
132
133
134struct dx_frame
135{
136 struct buffer_head *bh;
137 struct dx_entry *entries;
138 struct dx_entry *at;
139};
140
141struct dx_map_entry
142{
143 u32 hash;
144 u16 offs;
145 u16 size;
146};
147
148static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
149static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
150static inline unsigned dx_get_hash(struct dx_entry *entry);
151static void dx_set_hash(struct dx_entry *entry, unsigned value);
152static unsigned dx_get_count(struct dx_entry *entries);
153static unsigned dx_get_limit(struct dx_entry *entries);
154static void dx_set_count(struct dx_entry *entries, unsigned value);
155static void dx_set_limit(struct dx_entry *entries, unsigned value);
156static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
157static unsigned dx_node_limit(struct inode *dir);
158static struct dx_frame *dx_probe(const struct qstr *d_name,
159 struct inode *dir,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
162 int *err);
163static void dx_release(struct dx_frame *frames);
164static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
168 struct dx_map_entry *offsets, int count, unsigned blocksize);
169static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
170static void dx_insert_block(struct dx_frame *frame,
171 u32 hash, ext4_lblk_t block);
172static int ext4_htree_next_block(struct inode *dir, __u32 hash,
173 struct dx_frame *frame,
174 struct dx_frame *frames,
175 __u32 *start_hash);
176static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
177 const struct qstr *d_name,
178 struct ext4_dir_entry_2 **res_dir,
179 int *err);
180static 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 */
186static inline struct ext4_dir_entry_2 *
187ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
188{
189 return (struct ext4_dir_entry_2 *)((char *)p +
190 ext4_rec_len_from_disk(p->rec_len, blocksize));
191}
192
193/*
194 * Future: use high four bits of block for coalesce-on-delete flags
195 * Mask them off for now.
196 */
197
198static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
199{
200 return le32_to_cpu(entry->block) & 0x00ffffff;
201}
202
203static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
204{
205 entry->block = cpu_to_le32(value);
206}
207
208static inline unsigned dx_get_hash(struct dx_entry *entry)
209{
210 return le32_to_cpu(entry->hash);
211}
212
213static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
214{
215 entry->hash = cpu_to_le32(value);
216}
217
218static inline unsigned dx_get_count(struct dx_entry *entries)
219{
220 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
221}
222
223static inline unsigned dx_get_limit(struct dx_entry *entries)
224{
225 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
226}
227
228static inline void dx_set_count(struct dx_entry *entries, unsigned value)
229{
230 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
231}
232
233static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
234{
235 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
236}
237
238static 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
245static 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
255static 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
266struct stats
267{
268 unsigned names;
269 unsigned space;
270 unsigned bcount;
271};
272
273static 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 (unsigned) ((char *) de - base));
293 }
294 space += EXT4_DIR_REC_LEN(de->name_len);
295 names++;
296 }
297 de = ext4_next_entry(de, size);
298 }
299 printk("(%i)\n", names);
300 return (struct stats) { names, space, 1 };
301}
302
303struct 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 */
345static struct dx_frame *
346dx_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, "Unrecognised inode hash code %d",
364 root->info.hash_version);
365 brelse(bh);
366 *err = ERR_BAD_DX_DIR;
367 goto fail;
368 }
369 hinfo->hash_version = root->info.hash_version;
370 if (hinfo->hash_version <= DX_HASH_TEA)
371 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
372 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
373 if (d_name)
374 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
375 hash = hinfo->hash;
376
377 if (root->info.unused_flags & 1) {
378 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
379 root->info.unused_flags);
380 brelse(bh);
381 *err = ERR_BAD_DX_DIR;
382 goto fail;
383 }
384
385 if ((indirect = root->info.indirect_levels) > 1) {
386 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
387 root->info.indirect_levels);
388 brelse(bh);
389 *err = ERR_BAD_DX_DIR;
390 goto fail;
391 }
392
393 entries = (struct dx_entry *) (((char *)&root->info) +
394 root->info.info_length);
395
396 if (dx_get_limit(entries) != dx_root_limit(dir,
397 root->info.info_length)) {
398 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
399 brelse(bh);
400 *err = ERR_BAD_DX_DIR;
401 goto fail;
402 }
403
404 dxtrace(printk("Look up %x", hash));
405 while (1)
406 {
407 count = dx_get_count(entries);
408 if (!count || count > dx_get_limit(entries)) {
409 ext4_warning(dir->i_sb,
410 "dx entry: no count or count > limit");
411 brelse(bh);
412 *err = ERR_BAD_DX_DIR;
413 goto fail2;
414 }
415
416 p = entries + 1;
417 q = entries + count - 1;
418 while (p <= q)
419 {
420 m = p + (q - p)/2;
421 dxtrace(printk("."));
422 if (dx_get_hash(m) > hash)
423 q = m - 1;
424 else
425 p = m + 1;
426 }
427
428 if (0) // linear search cross check
429 {
430 unsigned n = count - 1;
431 at = entries;
432 while (n--)
433 {
434 dxtrace(printk(","));
435 if (dx_get_hash(++at) > hash)
436 {
437 at--;
438 break;
439 }
440 }
441 assert (at == p - 1);
442 }
443
444 at = p - 1;
445 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
446 frame->bh = bh;
447 frame->entries = entries;
448 frame->at = at;
449 if (!indirect--) return frame;
450 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
451 goto fail2;
452 at = entries = ((struct dx_node *) bh->b_data)->entries;
453 if (dx_get_limit(entries) != dx_node_limit (dir)) {
454 ext4_warning(dir->i_sb,
455 "dx entry: limit != node limit");
456 brelse(bh);
457 *err = ERR_BAD_DX_DIR;
458 goto fail2;
459 }
460 frame++;
461 frame->bh = NULL;
462 }
463fail2:
464 while (frame >= frame_in) {
465 brelse(frame->bh);
466 frame--;
467 }
468fail:
469 if (*err == ERR_BAD_DX_DIR)
470 ext4_warning(dir->i_sb,
471 "Corrupt dir inode %ld, running e2fsck is "
472 "recommended.", dir->i_ino);
473 return NULL;
474}
475
476static void dx_release (struct dx_frame *frames)
477{
478 if (frames[0].bh == NULL)
479 return;
480
481 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
482 brelse(frames[1].bh);
483 brelse(frames[0].bh);
484}
485
486/*
487 * This function increments the frame pointer to search the next leaf
488 * block, and reads in the necessary intervening nodes if the search
489 * should be necessary. Whether or not the search is necessary is
490 * controlled by the hash parameter. If the hash value is even, then
491 * the search is only continued if the next block starts with that
492 * hash value. This is used if we are searching for a specific file.
493 *
494 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
495 *
496 * This function returns 1 if the caller should continue to search,
497 * or 0 if it should not. If there is an error reading one of the
498 * index blocks, it will a negative error code.
499 *
500 * If start_hash is non-null, it will be filled in with the starting
501 * hash of the next page.
502 */
503static int ext4_htree_next_block(struct inode *dir, __u32 hash,
504 struct dx_frame *frame,
505 struct dx_frame *frames,
506 __u32 *start_hash)
507{
508 struct dx_frame *p;
509 struct buffer_head *bh;
510 int err, num_frames = 0;
511 __u32 bhash;
512
513 p = frame;
514 /*
515 * Find the next leaf page by incrementing the frame pointer.
516 * If we run out of entries in the interior node, loop around and
517 * increment pointer in the parent node. When we break out of
518 * this loop, num_frames indicates the number of interior
519 * nodes need to be read.
520 */
521 while (1) {
522 if (++(p->at) < p->entries + dx_get_count(p->entries))
523 break;
524 if (p == frames)
525 return 0;
526 num_frames++;
527 p--;
528 }
529
530 /*
531 * If the hash is 1, then continue only if the next page has a
532 * continuation hash of any value. This is used for readdir
533 * handling. Otherwise, check to see if the hash matches the
534 * desired contiuation hash. If it doesn't, return since
535 * there's no point to read in the successive index pages.
536 */
537 bhash = dx_get_hash(p->at);
538 if (start_hash)
539 *start_hash = bhash;
540 if ((hash & 1) == 0) {
541 if ((bhash & ~1) != hash)
542 return 0;
543 }
544 /*
545 * If the hash is HASH_NB_ALWAYS, we always go to the next
546 * block so no check is necessary
547 */
548 while (num_frames--) {
549 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
550 0, &err)))
551 return err; /* Failure */
552 p++;
553 brelse(p->bh);
554 p->bh = bh;
555 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
556 }
557 return 1;
558}
559
560
561/*
562 * This function fills a red-black tree with information from a
563 * directory block. It returns the number directory entries loaded
564 * into the tree. If there is an error it is returned in err.
565 */
566static int htree_dirblock_to_tree(struct file *dir_file,
567 struct inode *dir, ext4_lblk_t block,
568 struct dx_hash_info *hinfo,
569 __u32 start_hash, __u32 start_minor_hash)
570{
571 struct buffer_head *bh;
572 struct ext4_dir_entry_2 *de, *top;
573 int err, count = 0;
574
575 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
576 (unsigned long)block));
577 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
578 return err;
579
580 de = (struct ext4_dir_entry_2 *) bh->b_data;
581 top = (struct ext4_dir_entry_2 *) ((char *) de +
582 dir->i_sb->s_blocksize -
583 EXT4_DIR_REC_LEN(0));
584 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
585 if (ext4_check_dir_entry(dir, NULL, de, bh,
586 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
587 + ((char *)de - bh->b_data))) {
588 /* On error, skip the f_pos to the next block. */
589 dir_file->f_pos = (dir_file->f_pos |
590 (dir->i_sb->s_blocksize - 1)) + 1;
591 brelse(bh);
592 return count;
593 }
594 ext4fs_dirhash(de->name, de->name_len, hinfo);
595 if ((hinfo->hash < start_hash) ||
596 ((hinfo->hash == start_hash) &&
597 (hinfo->minor_hash < start_minor_hash)))
598 continue;
599 if (de->inode == 0)
600 continue;
601 if ((err = ext4_htree_store_dirent(dir_file,
602 hinfo->hash, hinfo->minor_hash, de)) != 0) {
603 brelse(bh);
604 return err;
605 }
606 count++;
607 }
608 brelse(bh);
609 return count;
610}
611
612
613/*
614 * This function fills a red-black tree with information from a
615 * directory. We start scanning the directory in hash order, starting
616 * at start_hash and start_minor_hash.
617 *
618 * This function returns the number of entries inserted into the tree,
619 * or a negative error code.
620 */
621int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
622 __u32 start_minor_hash, __u32 *next_hash)
623{
624 struct dx_hash_info hinfo;
625 struct ext4_dir_entry_2 *de;
626 struct dx_frame frames[2], *frame;
627 struct inode *dir;
628 ext4_lblk_t block;
629 int count = 0;
630 int ret, err;
631 __u32 hashval;
632
633 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
634 start_hash, start_minor_hash));
635 dir = dir_file->f_path.dentry->d_inode;
636 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
637 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
638 if (hinfo.hash_version <= DX_HASH_TEA)
639 hinfo.hash_version +=
640 EXT4_SB(dir->i_sb)->s_hash_unsigned;
641 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
642 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
643 start_hash, start_minor_hash);
644 *next_hash = ~0;
645 return count;
646 }
647 hinfo.hash = start_hash;
648 hinfo.minor_hash = 0;
649 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
650 if (!frame)
651 return err;
652
653 /* Add '.' and '..' from the htree header */
654 if (!start_hash && !start_minor_hash) {
655 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
656 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
657 goto errout;
658 count++;
659 }
660 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
661 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
662 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
663 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
664 goto errout;
665 count++;
666 }
667
668 while (1) {
669 block = dx_get_block(frame->at);
670 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
671 start_hash, start_minor_hash);
672 if (ret < 0) {
673 err = ret;
674 goto errout;
675 }
676 count += ret;
677 hashval = ~0;
678 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
679 frame, frames, &hashval);
680 *next_hash = hashval;
681 if (ret < 0) {
682 err = ret;
683 goto errout;
684 }
685 /*
686 * Stop if: (a) there are no more entries, or
687 * (b) we have inserted at least one entry and the
688 * next hash value is not a continuation
689 */
690 if ((ret == 0) ||
691 (count && ((hashval & 1) == 0)))
692 break;
693 }
694 dx_release(frames);
695 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
696 "next hash: %x\n", count, *next_hash));
697 return count;
698errout:
699 dx_release(frames);
700 return (err);
701}
702
703
704/*
705 * Directory block splitting, compacting
706 */
707
708/*
709 * Create map of hash values, offsets, and sizes, stored at end of block.
710 * Returns number of entries mapped.
711 */
712static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
713 struct dx_hash_info *hinfo,
714 struct dx_map_entry *map_tail)
715{
716 int count = 0;
717 char *base = (char *) de;
718 struct dx_hash_info h = *hinfo;
719
720 while ((char *) de < base + blocksize) {
721 if (de->name_len && de->inode) {
722 ext4fs_dirhash(de->name, de->name_len, &h);
723 map_tail--;
724 map_tail->hash = h.hash;
725 map_tail->offs = ((char *) de - base)>>2;
726 map_tail->size = le16_to_cpu(de->rec_len);
727 count++;
728 cond_resched();
729 }
730 /* XXX: do we need to check rec_len == 0 case? -Chris */
731 de = ext4_next_entry(de, blocksize);
732 }
733 return count;
734}
735
736/* Sort map by hash value */
737static void dx_sort_map (struct dx_map_entry *map, unsigned count)
738{
739 struct dx_map_entry *p, *q, *top = map + count - 1;
740 int more;
741 /* Combsort until bubble sort doesn't suck */
742 while (count > 2) {
743 count = count*10/13;
744 if (count - 9 < 2) /* 9, 10 -> 11 */
745 count = 11;
746 for (p = top, q = p - count; q >= map; p--, q--)
747 if (p->hash < q->hash)
748 swap(*p, *q);
749 }
750 /* Garden variety bubble sort */
751 do {
752 more = 0;
753 q = top;
754 while (q-- > map) {
755 if (q[1].hash >= q[0].hash)
756 continue;
757 swap(*(q+1), *q);
758 more = 1;
759 }
760 } while(more);
761}
762
763static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
764{
765 struct dx_entry *entries = frame->entries;
766 struct dx_entry *old = frame->at, *new = old + 1;
767 int count = dx_get_count(entries);
768
769 assert(count < dx_get_limit(entries));
770 assert(old < entries + count);
771 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
772 dx_set_hash(new, hash);
773 dx_set_block(new, block);
774 dx_set_count(entries, count + 1);
775}
776
777static void ext4_update_dx_flag(struct inode *inode)
778{
779 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
780 EXT4_FEATURE_COMPAT_DIR_INDEX))
781 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
782}
783
784/*
785 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
786 *
787 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
788 * `de != NULL' is guaranteed by caller.
789 */
790static inline int ext4_match (int len, const char * const name,
791 struct ext4_dir_entry_2 * de)
792{
793 if (len != de->name_len)
794 return 0;
795 if (!de->inode)
796 return 0;
797 return !memcmp(name, de->name, len);
798}
799
800/*
801 * Returns 0 if not found, -1 on failure, and 1 on success
802 */
803static inline int search_dirblock(struct buffer_head *bh,
804 struct inode *dir,
805 const struct qstr *d_name,
806 unsigned int offset,
807 struct ext4_dir_entry_2 ** res_dir)
808{
809 struct ext4_dir_entry_2 * de;
810 char * dlimit;
811 int de_len;
812 const char *name = d_name->name;
813 int namelen = d_name->len;
814
815 de = (struct ext4_dir_entry_2 *) bh->b_data;
816 dlimit = bh->b_data + dir->i_sb->s_blocksize;
817 while ((char *) de < dlimit) {
818 /* this code is executed quadratically often */
819 /* do minimal checking `by hand' */
820
821 if ((char *) de + namelen <= dlimit &&
822 ext4_match (namelen, name, de)) {
823 /* found a match - just to be sure, do a full check */
824 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
825 return -1;
826 *res_dir = de;
827 return 1;
828 }
829 /* prevent looping on a bad block */
830 de_len = ext4_rec_len_from_disk(de->rec_len,
831 dir->i_sb->s_blocksize);
832 if (de_len <= 0)
833 return -1;
834 offset += de_len;
835 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
836 }
837 return 0;
838}
839
840
841/*
842 * ext4_find_entry()
843 *
844 * finds an entry in the specified directory with the wanted name. It
845 * returns the cache buffer in which the entry was found, and the entry
846 * itself (as a parameter - res_dir). It does NOT read the inode of the
847 * entry - you'll have to do that yourself if you want to.
848 *
849 * The returned buffer_head has ->b_count elevated. The caller is expected
850 * to brelse() it when appropriate.
851 */
852static struct buffer_head * ext4_find_entry (struct inode *dir,
853 const struct qstr *d_name,
854 struct ext4_dir_entry_2 ** res_dir)
855{
856 struct super_block *sb;
857 struct buffer_head *bh_use[NAMEI_RA_SIZE];
858 struct buffer_head *bh, *ret = NULL;
859 ext4_lblk_t start, block, b;
860 const u8 *name = d_name->name;
861 int ra_max = 0; /* Number of bh's in the readahead
862 buffer, bh_use[] */
863 int ra_ptr = 0; /* Current index into readahead
864 buffer */
865 int num = 0;
866 ext4_lblk_t nblocks;
867 int i, err;
868 int namelen;
869
870 *res_dir = NULL;
871 sb = dir->i_sb;
872 namelen = d_name->len;
873 if (namelen > EXT4_NAME_LEN)
874 return NULL;
875 if ((namelen <= 2) && (name[0] == '.') &&
876 (name[1] == '.' || name[1] == '\0')) {
877 /*
878 * "." or ".." will only be in the first block
879 * NFS may look up ".."; "." should be handled by the VFS
880 */
881 block = start = 0;
882 nblocks = 1;
883 goto restart;
884 }
885 if (is_dx(dir)) {
886 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
887 /*
888 * On success, or if the error was file not found,
889 * return. Otherwise, fall back to doing a search the
890 * old fashioned way.
891 */
892 if (bh || (err != ERR_BAD_DX_DIR))
893 return bh;
894 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
895 "falling back\n"));
896 }
897 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
898 start = EXT4_I(dir)->i_dir_start_lookup;
899 if (start >= nblocks)
900 start = 0;
901 block = start;
902restart:
903 do {
904 /*
905 * We deal with the read-ahead logic here.
906 */
907 if (ra_ptr >= ra_max) {
908 /* Refill the readahead buffer */
909 ra_ptr = 0;
910 b = block;
911 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
912 /*
913 * Terminate if we reach the end of the
914 * directory and must wrap, or if our
915 * search has finished at this block.
916 */
917 if (b >= nblocks || (num && block == start)) {
918 bh_use[ra_max] = NULL;
919 break;
920 }
921 num++;
922 bh = ext4_getblk(NULL, dir, b++, 0, &err);
923 bh_use[ra_max] = bh;
924 if (bh)
925 ll_rw_block(READ | REQ_META | REQ_PRIO,
926 1, &bh);
927 }
928 }
929 if ((bh = bh_use[ra_ptr++]) == NULL)
930 goto next;
931 wait_on_buffer(bh);
932 if (!buffer_uptodate(bh)) {
933 /* read error, skip block & hope for the best */
934 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
935 (unsigned long) block);
936 brelse(bh);
937 goto next;
938 }
939 i = search_dirblock(bh, dir, d_name,
940 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
941 if (i == 1) {
942 EXT4_I(dir)->i_dir_start_lookup = block;
943 ret = bh;
944 goto cleanup_and_exit;
945 } else {
946 brelse(bh);
947 if (i < 0)
948 goto cleanup_and_exit;
949 }
950 next:
951 if (++block >= nblocks)
952 block = 0;
953 } while (block != start);
954
955 /*
956 * If the directory has grown while we were searching, then
957 * search the last part of the directory before giving up.
958 */
959 block = nblocks;
960 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
961 if (block < nblocks) {
962 start = 0;
963 goto restart;
964 }
965
966cleanup_and_exit:
967 /* Clean up the read-ahead blocks */
968 for (; ra_ptr < ra_max; ra_ptr++)
969 brelse(bh_use[ra_ptr]);
970 return ret;
971}
972
973static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
974 struct ext4_dir_entry_2 **res_dir, int *err)
975{
976 struct super_block * sb = dir->i_sb;
977 struct dx_hash_info hinfo;
978 struct dx_frame frames[2], *frame;
979 struct buffer_head *bh;
980 ext4_lblk_t block;
981 int retval;
982
983 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
984 return NULL;
985 do {
986 block = dx_get_block(frame->at);
987 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
988 goto errout;
989
990 retval = search_dirblock(bh, dir, d_name,
991 block << EXT4_BLOCK_SIZE_BITS(sb),
992 res_dir);
993 if (retval == 1) { /* Success! */
994 dx_release(frames);
995 return bh;
996 }
997 brelse(bh);
998 if (retval == -1) {
999 *err = ERR_BAD_DX_DIR;
1000 goto errout;
1001 }
1002
1003 /* Check to see if we should continue to search */
1004 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1005 frames, NULL);
1006 if (retval < 0) {
1007 ext4_warning(sb,
1008 "error reading index page in directory #%lu",
1009 dir->i_ino);
1010 *err = retval;
1011 goto errout;
1012 }
1013 } while (retval == 1);
1014
1015 *err = -ENOENT;
1016errout:
1017 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1018 dx_release (frames);
1019 return NULL;
1020}
1021
1022static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1023{
1024 struct inode *inode;
1025 struct ext4_dir_entry_2 *de;
1026 struct buffer_head *bh;
1027
1028 if (dentry->d_name.len > EXT4_NAME_LEN)
1029 return ERR_PTR(-ENAMETOOLONG);
1030
1031 bh = ext4_find_entry(dir, &dentry->d_name, &de);
1032 inode = NULL;
1033 if (bh) {
1034 __u32 ino = le32_to_cpu(de->inode);
1035 brelse(bh);
1036 if (!ext4_valid_inum(dir->i_sb, ino)) {
1037 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1038 return ERR_PTR(-EIO);
1039 }
1040 inode = ext4_iget(dir->i_sb, ino);
1041 if (inode == ERR_PTR(-ESTALE)) {
1042 EXT4_ERROR_INODE(dir,
1043 "deleted inode referenced: %u",
1044 ino);
1045 return ERR_PTR(-EIO);
1046 }
1047 }
1048 return d_splice_alias(inode, dentry);
1049}
1050
1051
1052struct dentry *ext4_get_parent(struct dentry *child)
1053{
1054 __u32 ino;
1055 static const struct qstr dotdot = {
1056 .name = "..",
1057 .len = 2,
1058 };
1059 struct ext4_dir_entry_2 * de;
1060 struct buffer_head *bh;
1061
1062 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1063 if (!bh)
1064 return ERR_PTR(-ENOENT);
1065 ino = le32_to_cpu(de->inode);
1066 brelse(bh);
1067
1068 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1069 EXT4_ERROR_INODE(child->d_inode,
1070 "bad parent inode number: %u", ino);
1071 return ERR_PTR(-EIO);
1072 }
1073
1074 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1075}
1076
1077#define S_SHIFT 12
1078static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1079 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1080 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1081 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1082 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1083 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1084 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1085 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
1086};
1087
1088static inline void ext4_set_de_type(struct super_block *sb,
1089 struct ext4_dir_entry_2 *de,
1090 umode_t mode) {
1091 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1092 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1093}
1094
1095/*
1096 * Move count entries from end of map between two memory locations.
1097 * Returns pointer to last entry moved.
1098 */
1099static struct ext4_dir_entry_2 *
1100dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1101 unsigned blocksize)
1102{
1103 unsigned rec_len = 0;
1104
1105 while (count--) {
1106 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1107 (from + (map->offs<<2));
1108 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1109 memcpy (to, de, rec_len);
1110 ((struct ext4_dir_entry_2 *) to)->rec_len =
1111 ext4_rec_len_to_disk(rec_len, blocksize);
1112 de->inode = 0;
1113 map++;
1114 to += rec_len;
1115 }
1116 return (struct ext4_dir_entry_2 *) (to - rec_len);
1117}
1118
1119/*
1120 * Compact each dir entry in the range to the minimal rec_len.
1121 * Returns pointer to last entry in range.
1122 */
1123static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1124{
1125 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1126 unsigned rec_len = 0;
1127
1128 prev = to = de;
1129 while ((char*)de < base + blocksize) {
1130 next = ext4_next_entry(de, blocksize);
1131 if (de->inode && de->name_len) {
1132 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1133 if (de > to)
1134 memmove(to, de, rec_len);
1135 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1136 prev = to;
1137 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1138 }
1139 de = next;
1140 }
1141 return prev;
1142}
1143
1144/*
1145 * Split a full leaf block to make room for a new dir entry.
1146 * Allocate a new block, and move entries so that they are approx. equally full.
1147 * Returns pointer to de in block into which the new entry will be inserted.
1148 */
1149static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1150 struct buffer_head **bh,struct dx_frame *frame,
1151 struct dx_hash_info *hinfo, int *error)
1152{
1153 unsigned blocksize = dir->i_sb->s_blocksize;
1154 unsigned count, continued;
1155 struct buffer_head *bh2;
1156 ext4_lblk_t newblock;
1157 u32 hash2;
1158 struct dx_map_entry *map;
1159 char *data1 = (*bh)->b_data, *data2;
1160 unsigned split, move, size;
1161 struct ext4_dir_entry_2 *de = NULL, *de2;
1162 int err = 0, i;
1163
1164 bh2 = ext4_append (handle, dir, &newblock, &err);
1165 if (!(bh2)) {
1166 brelse(*bh);
1167 *bh = NULL;
1168 goto errout;
1169 }
1170
1171 BUFFER_TRACE(*bh, "get_write_access");
1172 err = ext4_journal_get_write_access(handle, *bh);
1173 if (err)
1174 goto journal_error;
1175
1176 BUFFER_TRACE(frame->bh, "get_write_access");
1177 err = ext4_journal_get_write_access(handle, frame->bh);
1178 if (err)
1179 goto journal_error;
1180
1181 data2 = bh2->b_data;
1182
1183 /* create map in the end of data2 block */
1184 map = (struct dx_map_entry *) (data2 + blocksize);
1185 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1186 blocksize, hinfo, map);
1187 map -= count;
1188 dx_sort_map(map, count);
1189 /* Split the existing block in the middle, size-wise */
1190 size = 0;
1191 move = 0;
1192 for (i = count-1; i >= 0; i--) {
1193 /* is more than half of this entry in 2nd half of the block? */
1194 if (size + map[i].size/2 > blocksize/2)
1195 break;
1196 size += map[i].size;
1197 move++;
1198 }
1199 /* map index at which we will split */
1200 split = count - move;
1201 hash2 = map[split].hash;
1202 continued = hash2 == map[split - 1].hash;
1203 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1204 (unsigned long)dx_get_block(frame->at),
1205 hash2, split, count-split));
1206
1207 /* Fancy dance to stay within two buffers */
1208 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1209 de = dx_pack_dirents(data1, blocksize);
1210 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1211 blocksize);
1212 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1213 blocksize);
1214 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1215 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1216
1217 /* Which block gets the new entry? */
1218 if (hinfo->hash >= hash2)
1219 {
1220 swap(*bh, bh2);
1221 de = de2;
1222 }
1223 dx_insert_block(frame, hash2 + continued, newblock);
1224 err = ext4_handle_dirty_metadata(handle, dir, bh2);
1225 if (err)
1226 goto journal_error;
1227 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
1228 if (err)
1229 goto journal_error;
1230 brelse(bh2);
1231 dxtrace(dx_show_index("frame", frame->entries));
1232 return de;
1233
1234journal_error:
1235 brelse(*bh);
1236 brelse(bh2);
1237 *bh = NULL;
1238 ext4_std_error(dir->i_sb, err);
1239errout:
1240 *error = err;
1241 return NULL;
1242}
1243
1244/*
1245 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1246 * it points to a directory entry which is guaranteed to be large
1247 * enough for new directory entry. If de is NULL, then
1248 * add_dirent_to_buf will attempt search the directory block for
1249 * space. It will return -ENOSPC if no space is available, and -EIO
1250 * and -EEXIST if directory entry already exists.
1251 */
1252static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1253 struct inode *inode, struct ext4_dir_entry_2 *de,
1254 struct buffer_head *bh)
1255{
1256 struct inode *dir = dentry->d_parent->d_inode;
1257 const char *name = dentry->d_name.name;
1258 int namelen = dentry->d_name.len;
1259 unsigned int offset = 0;
1260 unsigned int blocksize = dir->i_sb->s_blocksize;
1261 unsigned short reclen;
1262 int nlen, rlen, err;
1263 char *top;
1264
1265 reclen = EXT4_DIR_REC_LEN(namelen);
1266 if (!de) {
1267 de = (struct ext4_dir_entry_2 *)bh->b_data;
1268 top = bh->b_data + blocksize - reclen;
1269 while ((char *) de <= top) {
1270 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1271 return -EIO;
1272 if (ext4_match(namelen, name, de))
1273 return -EEXIST;
1274 nlen = EXT4_DIR_REC_LEN(de->name_len);
1275 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1276 if ((de->inode? rlen - nlen: rlen) >= reclen)
1277 break;
1278 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1279 offset += rlen;
1280 }
1281 if ((char *) de > top)
1282 return -ENOSPC;
1283 }
1284 BUFFER_TRACE(bh, "get_write_access");
1285 err = ext4_journal_get_write_access(handle, bh);
1286 if (err) {
1287 ext4_std_error(dir->i_sb, err);
1288 return err;
1289 }
1290
1291 /* By now the buffer is marked for journaling */
1292 nlen = EXT4_DIR_REC_LEN(de->name_len);
1293 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1294 if (de->inode) {
1295 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1296 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1297 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1298 de = de1;
1299 }
1300 de->file_type = EXT4_FT_UNKNOWN;
1301 if (inode) {
1302 de->inode = cpu_to_le32(inode->i_ino);
1303 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1304 } else
1305 de->inode = 0;
1306 de->name_len = namelen;
1307 memcpy(de->name, name, namelen);
1308 /*
1309 * XXX shouldn't update any times until successful
1310 * completion of syscall, but too many callers depend
1311 * on this.
1312 *
1313 * XXX similarly, too many callers depend on
1314 * ext4_new_inode() setting the times, but error
1315 * recovery deletes the inode, so the worst that can
1316 * happen is that the times are slightly out of date
1317 * and/or different from the directory change time.
1318 */
1319 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1320 ext4_update_dx_flag(dir);
1321 dir->i_version++;
1322 ext4_mark_inode_dirty(handle, dir);
1323 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1324 err = ext4_handle_dirty_metadata(handle, dir, bh);
1325 if (err)
1326 ext4_std_error(dir->i_sb, err);
1327 return 0;
1328}
1329
1330/*
1331 * This converts a one block unindexed directory to a 3 block indexed
1332 * directory, and adds the dentry to the indexed directory.
1333 */
1334static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1335 struct inode *inode, struct buffer_head *bh)
1336{
1337 struct inode *dir = dentry->d_parent->d_inode;
1338 const char *name = dentry->d_name.name;
1339 int namelen = dentry->d_name.len;
1340 struct buffer_head *bh2;
1341 struct dx_root *root;
1342 struct dx_frame frames[2], *frame;
1343 struct dx_entry *entries;
1344 struct ext4_dir_entry_2 *de, *de2;
1345 char *data1, *top;
1346 unsigned len;
1347 int retval;
1348 unsigned blocksize;
1349 struct dx_hash_info hinfo;
1350 ext4_lblk_t block;
1351 struct fake_dirent *fde;
1352
1353 blocksize = dir->i_sb->s_blocksize;
1354 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1355 retval = ext4_journal_get_write_access(handle, bh);
1356 if (retval) {
1357 ext4_std_error(dir->i_sb, retval);
1358 brelse(bh);
1359 return retval;
1360 }
1361 root = (struct dx_root *) bh->b_data;
1362
1363 /* The 0th block becomes the root, move the dirents out */
1364 fde = &root->dotdot;
1365 de = (struct ext4_dir_entry_2 *)((char *)fde +
1366 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1367 if ((char *) de >= (((char *) root) + blocksize)) {
1368 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1369 brelse(bh);
1370 return -EIO;
1371 }
1372 len = ((char *) root) + blocksize - (char *) de;
1373
1374 /* Allocate new block for the 0th block's dirents */
1375 bh2 = ext4_append(handle, dir, &block, &retval);
1376 if (!(bh2)) {
1377 brelse(bh);
1378 return retval;
1379 }
1380 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1381 data1 = bh2->b_data;
1382
1383 memcpy (data1, de, len);
1384 de = (struct ext4_dir_entry_2 *) data1;
1385 top = data1 + len;
1386 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1387 de = de2;
1388 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1389 blocksize);
1390 /* Initialize the root; the dot dirents already exist */
1391 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1392 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1393 blocksize);
1394 memset (&root->info, 0, sizeof(root->info));
1395 root->info.info_length = sizeof(root->info);
1396 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1397 entries = root->entries;
1398 dx_set_block(entries, 1);
1399 dx_set_count(entries, 1);
1400 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1401
1402 /* Initialize as for dx_probe */
1403 hinfo.hash_version = root->info.hash_version;
1404 if (hinfo.hash_version <= DX_HASH_TEA)
1405 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1406 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1407 ext4fs_dirhash(name, namelen, &hinfo);
1408 frame = frames;
1409 frame->entries = entries;
1410 frame->at = entries;
1411 frame->bh = bh;
1412 bh = bh2;
1413
1414 ext4_handle_dirty_metadata(handle, dir, frame->bh);
1415 ext4_handle_dirty_metadata(handle, dir, bh);
1416
1417 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1418 if (!de) {
1419 /*
1420 * Even if the block split failed, we have to properly write
1421 * out all the changes we did so far. Otherwise we can end up
1422 * with corrupted filesystem.
1423 */
1424 ext4_mark_inode_dirty(handle, dir);
1425 dx_release(frames);
1426 return retval;
1427 }
1428 dx_release(frames);
1429
1430 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1431 brelse(bh);
1432 return retval;
1433}
1434
1435/*
1436 * ext4_add_entry()
1437 *
1438 * adds a file entry to the specified directory, using the same
1439 * semantics as ext4_find_entry(). It returns NULL if it failed.
1440 *
1441 * NOTE!! The inode part of 'de' is left at 0 - which means you
1442 * may not sleep between calling this and putting something into
1443 * the entry, as someone else might have used it while you slept.
1444 */
1445static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1446 struct inode *inode)
1447{
1448 struct inode *dir = dentry->d_parent->d_inode;
1449 struct buffer_head *bh;
1450 struct ext4_dir_entry_2 *de;
1451 struct super_block *sb;
1452 int retval;
1453 int dx_fallback=0;
1454 unsigned blocksize;
1455 ext4_lblk_t block, blocks;
1456
1457 sb = dir->i_sb;
1458 blocksize = sb->s_blocksize;
1459 if (!dentry->d_name.len)
1460 return -EINVAL;
1461 if (is_dx(dir)) {
1462 retval = ext4_dx_add_entry(handle, dentry, inode);
1463 if (!retval || (retval != ERR_BAD_DX_DIR))
1464 return retval;
1465 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1466 dx_fallback++;
1467 ext4_mark_inode_dirty(handle, dir);
1468 }
1469 blocks = dir->i_size >> sb->s_blocksize_bits;
1470 for (block = 0; block < blocks; block++) {
1471 bh = ext4_bread(handle, dir, block, 0, &retval);
1472 if(!bh)
1473 return retval;
1474 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1475 if (retval != -ENOSPC) {
1476 brelse(bh);
1477 return retval;
1478 }
1479
1480 if (blocks == 1 && !dx_fallback &&
1481 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1482 return make_indexed_dir(handle, dentry, inode, bh);
1483 brelse(bh);
1484 }
1485 bh = ext4_append(handle, dir, &block, &retval);
1486 if (!bh)
1487 return retval;
1488 de = (struct ext4_dir_entry_2 *) bh->b_data;
1489 de->inode = 0;
1490 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
1491 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1492 brelse(bh);
1493 if (retval == 0)
1494 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1495 return retval;
1496}
1497
1498/*
1499 * Returns 0 for success, or a negative error value
1500 */
1501static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1502 struct inode *inode)
1503{
1504 struct dx_frame frames[2], *frame;
1505 struct dx_entry *entries, *at;
1506 struct dx_hash_info hinfo;
1507 struct buffer_head *bh;
1508 struct inode *dir = dentry->d_parent->d_inode;
1509 struct super_block *sb = dir->i_sb;
1510 struct ext4_dir_entry_2 *de;
1511 int err;
1512
1513 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1514 if (!frame)
1515 return err;
1516 entries = frame->entries;
1517 at = frame->at;
1518
1519 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1520 goto cleanup;
1521
1522 BUFFER_TRACE(bh, "get_write_access");
1523 err = ext4_journal_get_write_access(handle, bh);
1524 if (err)
1525 goto journal_error;
1526
1527 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1528 if (err != -ENOSPC)
1529 goto cleanup;
1530
1531 /* Block full, should compress but for now just split */
1532 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1533 dx_get_count(entries), dx_get_limit(entries)));
1534 /* Need to split index? */
1535 if (dx_get_count(entries) == dx_get_limit(entries)) {
1536 ext4_lblk_t newblock;
1537 unsigned icount = dx_get_count(entries);
1538 int levels = frame - frames;
1539 struct dx_entry *entries2;
1540 struct dx_node *node2;
1541 struct buffer_head *bh2;
1542
1543 if (levels && (dx_get_count(frames->entries) ==
1544 dx_get_limit(frames->entries))) {
1545 ext4_warning(sb, "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 memset(&node2->fake, 0, sizeof(struct fake_dirent));
1555 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1556 sb->s_blocksize);
1557 BUFFER_TRACE(frame->bh, "get_write_access");
1558 err = ext4_journal_get_write_access(handle, frame->bh);
1559 if (err)
1560 goto journal_error;
1561 if (levels) {
1562 unsigned icount1 = icount/2, icount2 = icount - icount1;
1563 unsigned hash2 = dx_get_hash(entries + icount1);
1564 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1565 icount1, icount2));
1566
1567 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1568 err = ext4_journal_get_write_access(handle,
1569 frames[0].bh);
1570 if (err)
1571 goto journal_error;
1572
1573 memcpy((char *) entries2, (char *) (entries + icount1),
1574 icount2 * sizeof(struct dx_entry));
1575 dx_set_count(entries, icount1);
1576 dx_set_count(entries2, icount2);
1577 dx_set_limit(entries2, dx_node_limit(dir));
1578
1579 /* Which index block gets the new entry? */
1580 if (at - entries >= icount1) {
1581 frame->at = at = at - entries - icount1 + entries2;
1582 frame->entries = entries = entries2;
1583 swap(frame->bh, bh2);
1584 }
1585 dx_insert_block(frames + 0, hash2, newblock);
1586 dxtrace(dx_show_index("node", frames[1].entries));
1587 dxtrace(dx_show_index("node",
1588 ((struct dx_node *) bh2->b_data)->entries));
1589 err = ext4_handle_dirty_metadata(handle, inode, bh2);
1590 if (err)
1591 goto journal_error;
1592 brelse (bh2);
1593 } else {
1594 dxtrace(printk(KERN_DEBUG
1595 "Creating second level index...\n"));
1596 memcpy((char *) entries2, (char *) entries,
1597 icount * sizeof(struct dx_entry));
1598 dx_set_limit(entries2, dx_node_limit(dir));
1599
1600 /* Set up root */
1601 dx_set_count(entries, 1);
1602 dx_set_block(entries + 0, newblock);
1603 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1604
1605 /* Add new access path frame */
1606 frame = frames + 1;
1607 frame->at = at = at - entries + entries2;
1608 frame->entries = entries = entries2;
1609 frame->bh = bh2;
1610 err = ext4_journal_get_write_access(handle,
1611 frame->bh);
1612 if (err)
1613 goto journal_error;
1614 }
1615 err = ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1616 if (err) {
1617 ext4_std_error(inode->i_sb, err);
1618 goto cleanup;
1619 }
1620 }
1621 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1622 if (!de)
1623 goto cleanup;
1624 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1625 goto cleanup;
1626
1627journal_error:
1628 ext4_std_error(dir->i_sb, err);
1629cleanup:
1630 if (bh)
1631 brelse(bh);
1632 dx_release(frames);
1633 return err;
1634}
1635
1636/*
1637 * ext4_delete_entry deletes a directory entry by merging it with the
1638 * previous entry
1639 */
1640static int ext4_delete_entry(handle_t *handle,
1641 struct inode *dir,
1642 struct ext4_dir_entry_2 *de_del,
1643 struct buffer_head *bh)
1644{
1645 struct ext4_dir_entry_2 *de, *pde;
1646 unsigned int blocksize = dir->i_sb->s_blocksize;
1647 int i, err;
1648
1649 i = 0;
1650 pde = NULL;
1651 de = (struct ext4_dir_entry_2 *) bh->b_data;
1652 while (i < bh->b_size) {
1653 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
1654 return -EIO;
1655 if (de == de_del) {
1656 BUFFER_TRACE(bh, "get_write_access");
1657 err = ext4_journal_get_write_access(handle, bh);
1658 if (unlikely(err)) {
1659 ext4_std_error(dir->i_sb, err);
1660 return err;
1661 }
1662 if (pde)
1663 pde->rec_len = ext4_rec_len_to_disk(
1664 ext4_rec_len_from_disk(pde->rec_len,
1665 blocksize) +
1666 ext4_rec_len_from_disk(de->rec_len,
1667 blocksize),
1668 blocksize);
1669 else
1670 de->inode = 0;
1671 dir->i_version++;
1672 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1673 err = ext4_handle_dirty_metadata(handle, dir, bh);
1674 if (unlikely(err)) {
1675 ext4_std_error(dir->i_sb, err);
1676 return err;
1677 }
1678 return 0;
1679 }
1680 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
1681 pde = de;
1682 de = ext4_next_entry(de, blocksize);
1683 }
1684 return -ENOENT;
1685}
1686
1687/*
1688 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1689 * since this indicates that nlinks count was previously 1.
1690 */
1691static void ext4_inc_count(handle_t *handle, struct inode *inode)
1692{
1693 inc_nlink(inode);
1694 if (is_dx(inode) && inode->i_nlink > 1) {
1695 /* limit is 16-bit i_links_count */
1696 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1697 inode->i_nlink = 1;
1698 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1699 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1700 }
1701 }
1702}
1703
1704/*
1705 * If a directory had nlink == 1, then we should let it be 1. This indicates
1706 * directory has >EXT4_LINK_MAX subdirs.
1707 */
1708static void ext4_dec_count(handle_t *handle, struct inode *inode)
1709{
1710 drop_nlink(inode);
1711 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1712 inc_nlink(inode);
1713}
1714
1715
1716static int ext4_add_nondir(handle_t *handle,
1717 struct dentry *dentry, struct inode *inode)
1718{
1719 int err = ext4_add_entry(handle, dentry, inode);
1720 if (!err) {
1721 ext4_mark_inode_dirty(handle, inode);
1722 d_instantiate(dentry, inode);
1723 unlock_new_inode(inode);
1724 return 0;
1725 }
1726 drop_nlink(inode);
1727 unlock_new_inode(inode);
1728 iput(inode);
1729 return err;
1730}
1731
1732/*
1733 * By the time this is called, we already have created
1734 * the directory cache entry for the new file, but it
1735 * is so far negative - it has no inode.
1736 *
1737 * If the create succeeds, we fill in the inode information
1738 * with d_instantiate().
1739 */
1740static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1741 struct nameidata *nd)
1742{
1743 handle_t *handle;
1744 struct inode *inode;
1745 int err, retries = 0;
1746
1747 dquot_initialize(dir);
1748
1749retry:
1750 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1751 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1752 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1753 if (IS_ERR(handle))
1754 return PTR_ERR(handle);
1755
1756 if (IS_DIRSYNC(dir))
1757 ext4_handle_sync(handle);
1758
1759 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1760 err = PTR_ERR(inode);
1761 if (!IS_ERR(inode)) {
1762 inode->i_op = &ext4_file_inode_operations;
1763 inode->i_fop = &ext4_file_operations;
1764 ext4_set_aops(inode);
1765 err = ext4_add_nondir(handle, dentry, inode);
1766 }
1767 ext4_journal_stop(handle);
1768 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1769 goto retry;
1770 return err;
1771}
1772
1773static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1774 int mode, dev_t rdev)
1775{
1776 handle_t *handle;
1777 struct inode *inode;
1778 int err, retries = 0;
1779
1780 if (!new_valid_dev(rdev))
1781 return -EINVAL;
1782
1783 dquot_initialize(dir);
1784
1785retry:
1786 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1787 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1788 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1789 if (IS_ERR(handle))
1790 return PTR_ERR(handle);
1791
1792 if (IS_DIRSYNC(dir))
1793 ext4_handle_sync(handle);
1794
1795 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
1796 err = PTR_ERR(inode);
1797 if (!IS_ERR(inode)) {
1798 init_special_inode(inode, inode->i_mode, rdev);
1799#ifdef CONFIG_EXT4_FS_XATTR
1800 inode->i_op = &ext4_special_inode_operations;
1801#endif
1802 err = ext4_add_nondir(handle, dentry, inode);
1803 }
1804 ext4_journal_stop(handle);
1805 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1806 goto retry;
1807 return err;
1808}
1809
1810static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1811{
1812 handle_t *handle;
1813 struct inode *inode;
1814 struct buffer_head *dir_block = NULL;
1815 struct ext4_dir_entry_2 *de;
1816 unsigned int blocksize = dir->i_sb->s_blocksize;
1817 int err, retries = 0;
1818
1819 if (EXT4_DIR_LINK_MAX(dir))
1820 return -EMLINK;
1821
1822 dquot_initialize(dir);
1823
1824retry:
1825 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1826 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1827 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
1828 if (IS_ERR(handle))
1829 return PTR_ERR(handle);
1830
1831 if (IS_DIRSYNC(dir))
1832 ext4_handle_sync(handle);
1833
1834 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1835 &dentry->d_name, 0);
1836 err = PTR_ERR(inode);
1837 if (IS_ERR(inode))
1838 goto out_stop;
1839
1840 inode->i_op = &ext4_dir_inode_operations;
1841 inode->i_fop = &ext4_dir_operations;
1842 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1843 dir_block = ext4_bread(handle, inode, 0, 1, &err);
1844 if (!dir_block)
1845 goto out_clear_inode;
1846 BUFFER_TRACE(dir_block, "get_write_access");
1847 err = ext4_journal_get_write_access(handle, dir_block);
1848 if (err)
1849 goto out_clear_inode;
1850 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
1851 de->inode = cpu_to_le32(inode->i_ino);
1852 de->name_len = 1;
1853 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1854 blocksize);
1855 strcpy(de->name, ".");
1856 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1857 de = ext4_next_entry(de, blocksize);
1858 de->inode = cpu_to_le32(dir->i_ino);
1859 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1860 blocksize);
1861 de->name_len = 2;
1862 strcpy(de->name, "..");
1863 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1864 inode->i_nlink = 2;
1865 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1866 err = ext4_handle_dirty_metadata(handle, dir, dir_block);
1867 if (err)
1868 goto out_clear_inode;
1869 err = ext4_mark_inode_dirty(handle, inode);
1870 if (!err)
1871 err = ext4_add_entry(handle, dentry, inode);
1872 if (err) {
1873out_clear_inode:
1874 clear_nlink(inode);
1875 unlock_new_inode(inode);
1876 ext4_mark_inode_dirty(handle, inode);
1877 iput(inode);
1878 goto out_stop;
1879 }
1880 ext4_inc_count(handle, dir);
1881 ext4_update_dx_flag(dir);
1882 err = ext4_mark_inode_dirty(handle, dir);
1883 if (err)
1884 goto out_clear_inode;
1885 d_instantiate(dentry, inode);
1886 unlock_new_inode(inode);
1887out_stop:
1888 brelse(dir_block);
1889 ext4_journal_stop(handle);
1890 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
1891 goto retry;
1892 return err;
1893}
1894
1895/*
1896 * routine to check that the specified directory is empty (for rmdir)
1897 */
1898static int empty_dir(struct inode *inode)
1899{
1900 unsigned int offset;
1901 struct buffer_head *bh;
1902 struct ext4_dir_entry_2 *de, *de1;
1903 struct super_block *sb;
1904 int err = 0;
1905
1906 sb = inode->i_sb;
1907 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1908 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
1909 if (err)
1910 EXT4_ERROR_INODE(inode,
1911 "error %d reading directory lblock 0", err);
1912 else
1913 ext4_warning(inode->i_sb,
1914 "bad directory (dir #%lu) - no data block",
1915 inode->i_ino);
1916 return 1;
1917 }
1918 de = (struct ext4_dir_entry_2 *) bh->b_data;
1919 de1 = ext4_next_entry(de, sb->s_blocksize);
1920 if (le32_to_cpu(de->inode) != inode->i_ino ||
1921 !le32_to_cpu(de1->inode) ||
1922 strcmp(".", de->name) ||
1923 strcmp("..", de1->name)) {
1924 ext4_warning(inode->i_sb,
1925 "bad directory (dir #%lu) - no `.' or `..'",
1926 inode->i_ino);
1927 brelse(bh);
1928 return 1;
1929 }
1930 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1931 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1932 de = ext4_next_entry(de1, sb->s_blocksize);
1933 while (offset < inode->i_size) {
1934 if (!bh ||
1935 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1936 unsigned int lblock;
1937 err = 0;
1938 brelse(bh);
1939 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1940 bh = ext4_bread(NULL, inode, lblock, 0, &err);
1941 if (!bh) {
1942 if (err)
1943 EXT4_ERROR_INODE(inode,
1944 "error %d reading directory "
1945 "lblock %u", err, lblock);
1946 offset += sb->s_blocksize;
1947 continue;
1948 }
1949 de = (struct ext4_dir_entry_2 *) bh->b_data;
1950 }
1951 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
1952 de = (struct ext4_dir_entry_2 *)(bh->b_data +
1953 sb->s_blocksize);
1954 offset = (offset | (sb->s_blocksize - 1)) + 1;
1955 continue;
1956 }
1957 if (le32_to_cpu(de->inode)) {
1958 brelse(bh);
1959 return 0;
1960 }
1961 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1962 de = ext4_next_entry(de, sb->s_blocksize);
1963 }
1964 brelse(bh);
1965 return 1;
1966}
1967
1968/* ext4_orphan_add() links an unlinked or truncated inode into a list of
1969 * such inodes, starting at the superblock, in case we crash before the
1970 * file is closed/deleted, or in case the inode truncate spans multiple
1971 * transactions and the last transaction is not recovered after a crash.
1972 *
1973 * At filesystem recovery time, we walk this list deleting unlinked
1974 * inodes and truncating linked inodes in ext4_orphan_cleanup().
1975 */
1976int ext4_orphan_add(handle_t *handle, struct inode *inode)
1977{
1978 struct super_block *sb = inode->i_sb;
1979 struct ext4_iloc iloc;
1980 int err = 0, rc;
1981
1982 if (!ext4_handle_valid(handle))
1983 return 0;
1984
1985 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
1986 if (!list_empty(&EXT4_I(inode)->i_orphan))
1987 goto out_unlock;
1988
1989 /*
1990 * Orphan handling is only valid for files with data blocks
1991 * being truncated, or files being unlinked. Note that we either
1992 * hold i_mutex, or the inode can not be referenced from outside,
1993 * so i_nlink should not be bumped due to race
1994 */
1995 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1996 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1997
1998 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1999 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2000 if (err)
2001 goto out_unlock;
2002
2003 err = ext4_reserve_inode_write(handle, inode, &iloc);
2004 if (err)
2005 goto out_unlock;
2006 /*
2007 * Due to previous errors inode may be already a part of on-disk
2008 * orphan list. If so skip on-disk list modification.
2009 */
2010 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2011 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2012 goto mem_insert;
2013
2014 /* Insert this inode at the head of the on-disk orphan list... */
2015 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2016 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2017 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
2018 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2019 if (!err)
2020 err = rc;
2021
2022 /* Only add to the head of the in-memory list if all the
2023 * previous operations succeeded. If the orphan_add is going to
2024 * fail (possibly taking the journal offline), we can't risk
2025 * leaving the inode on the orphan list: stray orphan-list
2026 * entries can cause panics at unmount time.
2027 *
2028 * This is safe: on error we're going to ignore the orphan list
2029 * anyway on the next recovery. */
2030mem_insert:
2031 if (!err)
2032 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2033
2034 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2035 jbd_debug(4, "orphan inode %lu will point to %d\n",
2036 inode->i_ino, NEXT_ORPHAN(inode));
2037out_unlock:
2038 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2039 ext4_std_error(inode->i_sb, err);
2040 return err;
2041}
2042
2043/*
2044 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2045 * of such inodes stored on disk, because it is finally being cleaned up.
2046 */
2047int ext4_orphan_del(handle_t *handle, struct inode *inode)
2048{
2049 struct list_head *prev;
2050 struct ext4_inode_info *ei = EXT4_I(inode);
2051 struct ext4_sb_info *sbi;
2052 __u32 ino_next;
2053 struct ext4_iloc iloc;
2054 int err = 0;
2055
2056 /* ext4_handle_valid() assumes a valid handle_t pointer */
2057 if (handle && !ext4_handle_valid(handle))
2058 return 0;
2059
2060 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2061 if (list_empty(&ei->i_orphan))
2062 goto out;
2063
2064 ino_next = NEXT_ORPHAN(inode);
2065 prev = ei->i_orphan.prev;
2066 sbi = EXT4_SB(inode->i_sb);
2067
2068 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2069
2070 list_del_init(&ei->i_orphan);
2071
2072 /* If we're on an error path, we may not have a valid
2073 * transaction handle with which to update the orphan list on
2074 * disk, but we still need to remove the inode from the linked
2075 * list in memory. */
2076 if (sbi->s_journal && !handle)
2077 goto out;
2078
2079 err = ext4_reserve_inode_write(handle, inode, &iloc);
2080 if (err)
2081 goto out_err;
2082
2083 if (prev == &sbi->s_orphan) {
2084 jbd_debug(4, "superblock will point to %u\n", ino_next);
2085 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2086 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2087 if (err)
2088 goto out_brelse;
2089 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2090 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
2091 } else {
2092 struct ext4_iloc iloc2;
2093 struct inode *i_prev =
2094 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2095
2096 jbd_debug(4, "orphan inode %lu will point to %u\n",
2097 i_prev->i_ino, ino_next);
2098 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2099 if (err)
2100 goto out_brelse;
2101 NEXT_ORPHAN(i_prev) = ino_next;
2102 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2103 }
2104 if (err)
2105 goto out_brelse;
2106 NEXT_ORPHAN(inode) = 0;
2107 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2108
2109out_err:
2110 ext4_std_error(inode->i_sb, err);
2111out:
2112 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2113 return err;
2114
2115out_brelse:
2116 brelse(iloc.bh);
2117 goto out_err;
2118}
2119
2120static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2121{
2122 int retval;
2123 struct inode *inode;
2124 struct buffer_head *bh;
2125 struct ext4_dir_entry_2 *de;
2126 handle_t *handle;
2127
2128 /* Initialize quotas before so that eventual writes go in
2129 * separate transaction */
2130 dquot_initialize(dir);
2131 dquot_initialize(dentry->d_inode);
2132
2133 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2134 if (IS_ERR(handle))
2135 return PTR_ERR(handle);
2136
2137 retval = -ENOENT;
2138 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2139 if (!bh)
2140 goto end_rmdir;
2141
2142 if (IS_DIRSYNC(dir))
2143 ext4_handle_sync(handle);
2144
2145 inode = dentry->d_inode;
2146
2147 retval = -EIO;
2148 if (le32_to_cpu(de->inode) != inode->i_ino)
2149 goto end_rmdir;
2150
2151 retval = -ENOTEMPTY;
2152 if (!empty_dir(inode))
2153 goto end_rmdir;
2154
2155 retval = ext4_delete_entry(handle, dir, de, bh);
2156 if (retval)
2157 goto end_rmdir;
2158 if (!EXT4_DIR_LINK_EMPTY(inode))
2159 ext4_warning(inode->i_sb,
2160 "empty directory has too many links (%d)",
2161 inode->i_nlink);
2162 inode->i_version++;
2163 clear_nlink(inode);
2164 /* There's no need to set i_disksize: the fact that i_nlink is
2165 * zero will ensure that the right thing happens during any
2166 * recovery. */
2167 inode->i_size = 0;
2168 ext4_orphan_add(handle, inode);
2169 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2170 ext4_mark_inode_dirty(handle, inode);
2171 ext4_dec_count(handle, dir);
2172 ext4_update_dx_flag(dir);
2173 ext4_mark_inode_dirty(handle, dir);
2174
2175end_rmdir:
2176 ext4_journal_stop(handle);
2177 brelse(bh);
2178 return retval;
2179}
2180
2181static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2182{
2183 int retval;
2184 struct inode *inode;
2185 struct buffer_head *bh;
2186 struct ext4_dir_entry_2 *de;
2187 handle_t *handle;
2188
2189 trace_ext4_unlink_enter(dir, dentry);
2190 /* Initialize quotas before so that eventual writes go
2191 * in separate transaction */
2192 dquot_initialize(dir);
2193 dquot_initialize(dentry->d_inode);
2194
2195 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2196 if (IS_ERR(handle))
2197 return PTR_ERR(handle);
2198
2199 if (IS_DIRSYNC(dir))
2200 ext4_handle_sync(handle);
2201
2202 retval = -ENOENT;
2203 bh = ext4_find_entry(dir, &dentry->d_name, &de);
2204 if (!bh)
2205 goto end_unlink;
2206
2207 inode = dentry->d_inode;
2208
2209 retval = -EIO;
2210 if (le32_to_cpu(de->inode) != inode->i_ino)
2211 goto end_unlink;
2212
2213 if (!inode->i_nlink) {
2214 ext4_warning(inode->i_sb,
2215 "Deleting nonexistent file (%lu), %d",
2216 inode->i_ino, inode->i_nlink);
2217 inode->i_nlink = 1;
2218 }
2219 retval = ext4_delete_entry(handle, dir, de, bh);
2220 if (retval)
2221 goto end_unlink;
2222 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2223 ext4_update_dx_flag(dir);
2224 ext4_mark_inode_dirty(handle, dir);
2225 drop_nlink(inode);
2226 if (!inode->i_nlink)
2227 ext4_orphan_add(handle, inode);
2228 inode->i_ctime = ext4_current_time(inode);
2229 ext4_mark_inode_dirty(handle, inode);
2230 retval = 0;
2231
2232end_unlink:
2233 ext4_journal_stop(handle);
2234 brelse(bh);
2235 trace_ext4_unlink_exit(dentry, retval);
2236 return retval;
2237}
2238
2239static int ext4_symlink(struct inode *dir,
2240 struct dentry *dentry, const char *symname)
2241{
2242 handle_t *handle;
2243 struct inode *inode;
2244 int l, err, retries = 0;
2245 int credits;
2246
2247 l = strlen(symname)+1;
2248 if (l > dir->i_sb->s_blocksize)
2249 return -ENAMETOOLONG;
2250
2251 dquot_initialize(dir);
2252
2253 if (l > EXT4_N_BLOCKS * 4) {
2254 /*
2255 * For non-fast symlinks, we just allocate inode and put it on
2256 * orphan list in the first transaction => we need bitmap,
2257 * group descriptor, sb, inode block, quota blocks, and
2258 * possibly selinux xattr blocks.
2259 */
2260 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2261 EXT4_XATTR_TRANS_BLOCKS;
2262 } else {
2263 /*
2264 * Fast symlink. We have to add entry to directory
2265 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2266 * allocate new inode (bitmap, group descriptor, inode block,
2267 * quota blocks, sb is already counted in previous macros).
2268 */
2269 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2270 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2271 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2272 }
2273retry:
2274 handle = ext4_journal_start(dir, credits);
2275 if (IS_ERR(handle))
2276 return PTR_ERR(handle);
2277
2278 if (IS_DIRSYNC(dir))
2279 ext4_handle_sync(handle);
2280
2281 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2282 &dentry->d_name, 0);
2283 err = PTR_ERR(inode);
2284 if (IS_ERR(inode))
2285 goto out_stop;
2286
2287 if (l > EXT4_N_BLOCKS * 4) {
2288 inode->i_op = &ext4_symlink_inode_operations;
2289 ext4_set_aops(inode);
2290 /*
2291 * We cannot call page_symlink() with transaction started
2292 * because it calls into ext4_write_begin() which can wait
2293 * for transaction commit if we are running out of space
2294 * and thus we deadlock. So we have to stop transaction now
2295 * and restart it when symlink contents is written.
2296 *
2297 * To keep fs consistent in case of crash, we have to put inode
2298 * to orphan list in the mean time.
2299 */
2300 drop_nlink(inode);
2301 err = ext4_orphan_add(handle, inode);
2302 ext4_journal_stop(handle);
2303 if (err)
2304 goto err_drop_inode;
2305 err = __page_symlink(inode, symname, l, 1);
2306 if (err)
2307 goto err_drop_inode;
2308 /*
2309 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2310 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2311 */
2312 handle = ext4_journal_start(dir,
2313 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2314 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2315 if (IS_ERR(handle)) {
2316 err = PTR_ERR(handle);
2317 goto err_drop_inode;
2318 }
2319 inc_nlink(inode);
2320 err = ext4_orphan_del(handle, inode);
2321 if (err) {
2322 ext4_journal_stop(handle);
2323 clear_nlink(inode);
2324 goto err_drop_inode;
2325 }
2326 } else {
2327 /* clear the extent format for fast symlink */
2328 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2329 inode->i_op = &ext4_fast_symlink_inode_operations;
2330 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2331 inode->i_size = l-1;
2332 }
2333 EXT4_I(inode)->i_disksize = inode->i_size;
2334 err = ext4_add_nondir(handle, dentry, inode);
2335out_stop:
2336 ext4_journal_stop(handle);
2337 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2338 goto retry;
2339 return err;
2340err_drop_inode:
2341 unlock_new_inode(inode);
2342 iput(inode);
2343 return err;
2344}
2345
2346static int ext4_link(struct dentry *old_dentry,
2347 struct inode *dir, struct dentry *dentry)
2348{
2349 handle_t *handle;
2350 struct inode *inode = old_dentry->d_inode;
2351 int err, retries = 0;
2352
2353 if (inode->i_nlink >= EXT4_LINK_MAX)
2354 return -EMLINK;
2355
2356 dquot_initialize(dir);
2357
2358retry:
2359 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2360 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2361 if (IS_ERR(handle))
2362 return PTR_ERR(handle);
2363
2364 if (IS_DIRSYNC(dir))
2365 ext4_handle_sync(handle);
2366
2367 inode->i_ctime = ext4_current_time(inode);
2368 ext4_inc_count(handle, inode);
2369 ihold(inode);
2370
2371 err = ext4_add_entry(handle, dentry, inode);
2372 if (!err) {
2373 ext4_mark_inode_dirty(handle, inode);
2374 d_instantiate(dentry, inode);
2375 } else {
2376 drop_nlink(inode);
2377 iput(inode);
2378 }
2379 ext4_journal_stop(handle);
2380 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2381 goto retry;
2382 return err;
2383}
2384
2385#define PARENT_INO(buffer, size) \
2386 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2387
2388/*
2389 * Anybody can rename anything with this: the permission checks are left to the
2390 * higher-level routines.
2391 */
2392static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2393 struct inode *new_dir, struct dentry *new_dentry)
2394{
2395 handle_t *handle;
2396 struct inode *old_inode, *new_inode;
2397 struct buffer_head *old_bh, *new_bh, *dir_bh;
2398 struct ext4_dir_entry_2 *old_de, *new_de;
2399 int retval, force_da_alloc = 0;
2400
2401 dquot_initialize(old_dir);
2402 dquot_initialize(new_dir);
2403
2404 old_bh = new_bh = dir_bh = NULL;
2405
2406 /* Initialize quotas before so that eventual writes go
2407 * in separate transaction */
2408 if (new_dentry->d_inode)
2409 dquot_initialize(new_dentry->d_inode);
2410 handle = ext4_journal_start(old_dir, 2 *
2411 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2412 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2413 if (IS_ERR(handle))
2414 return PTR_ERR(handle);
2415
2416 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2417 ext4_handle_sync(handle);
2418
2419 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2420 /*
2421 * Check for inode number is _not_ due to possible IO errors.
2422 * We might rmdir the source, keep it as pwd of some process
2423 * and merrily kill the link to whatever was created under the
2424 * same name. Goodbye sticky bit ;-<
2425 */
2426 old_inode = old_dentry->d_inode;
2427 retval = -ENOENT;
2428 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2429 goto end_rename;
2430
2431 new_inode = new_dentry->d_inode;
2432 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2433 if (new_bh) {
2434 if (!new_inode) {
2435 brelse(new_bh);
2436 new_bh = NULL;
2437 }
2438 }
2439 if (S_ISDIR(old_inode->i_mode)) {
2440 if (new_inode) {
2441 retval = -ENOTEMPTY;
2442 if (!empty_dir(new_inode))
2443 goto end_rename;
2444 }
2445 retval = -EIO;
2446 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2447 if (!dir_bh)
2448 goto end_rename;
2449 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2450 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2451 goto end_rename;
2452 retval = -EMLINK;
2453 if (!new_inode && new_dir != old_dir &&
2454 EXT4_DIR_LINK_MAX(new_dir))
2455 goto end_rename;
2456 BUFFER_TRACE(dir_bh, "get_write_access");
2457 retval = ext4_journal_get_write_access(handle, dir_bh);
2458 if (retval)
2459 goto end_rename;
2460 }
2461 if (!new_bh) {
2462 retval = ext4_add_entry(handle, new_dentry, old_inode);
2463 if (retval)
2464 goto end_rename;
2465 } else {
2466 BUFFER_TRACE(new_bh, "get write access");
2467 retval = ext4_journal_get_write_access(handle, new_bh);
2468 if (retval)
2469 goto end_rename;
2470 new_de->inode = cpu_to_le32(old_inode->i_ino);
2471 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2472 EXT4_FEATURE_INCOMPAT_FILETYPE))
2473 new_de->file_type = old_de->file_type;
2474 new_dir->i_version++;
2475 new_dir->i_ctime = new_dir->i_mtime =
2476 ext4_current_time(new_dir);
2477 ext4_mark_inode_dirty(handle, new_dir);
2478 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2479 retval = ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2480 if (unlikely(retval)) {
2481 ext4_std_error(new_dir->i_sb, retval);
2482 goto end_rename;
2483 }
2484 brelse(new_bh);
2485 new_bh = NULL;
2486 }
2487
2488 /*
2489 * Like most other Unix systems, set the ctime for inodes on a
2490 * rename.
2491 */
2492 old_inode->i_ctime = ext4_current_time(old_inode);
2493 ext4_mark_inode_dirty(handle, old_inode);
2494
2495 /*
2496 * ok, that's it
2497 */
2498 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2499 old_de->name_len != old_dentry->d_name.len ||
2500 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2501 (retval = ext4_delete_entry(handle, old_dir,
2502 old_de, old_bh)) == -ENOENT) {
2503 /* old_de could have moved from under us during htree split, so
2504 * make sure that we are deleting the right entry. We might
2505 * also be pointing to a stale entry in the unused part of
2506 * old_bh so just checking inum and the name isn't enough. */
2507 struct buffer_head *old_bh2;
2508 struct ext4_dir_entry_2 *old_de2;
2509
2510 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2511 if (old_bh2) {
2512 retval = ext4_delete_entry(handle, old_dir,
2513 old_de2, old_bh2);
2514 brelse(old_bh2);
2515 }
2516 }
2517 if (retval) {
2518 ext4_warning(old_dir->i_sb,
2519 "Deleting old file (%lu), %d, error=%d",
2520 old_dir->i_ino, old_dir->i_nlink, retval);
2521 }
2522
2523 if (new_inode) {
2524 ext4_dec_count(handle, new_inode);
2525 new_inode->i_ctime = ext4_current_time(new_inode);
2526 }
2527 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2528 ext4_update_dx_flag(old_dir);
2529 if (dir_bh) {
2530 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2531 cpu_to_le32(new_dir->i_ino);
2532 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2533 retval = ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2534 if (retval) {
2535 ext4_std_error(old_dir->i_sb, retval);
2536 goto end_rename;
2537 }
2538 ext4_dec_count(handle, old_dir);
2539 if (new_inode) {
2540 /* checked empty_dir above, can't have another parent,
2541 * ext4_dec_count() won't work for many-linked dirs */
2542 new_inode->i_nlink = 0;
2543 } else {
2544 ext4_inc_count(handle, new_dir);
2545 ext4_update_dx_flag(new_dir);
2546 ext4_mark_inode_dirty(handle, new_dir);
2547 }
2548 }
2549 ext4_mark_inode_dirty(handle, old_dir);
2550 if (new_inode) {
2551 ext4_mark_inode_dirty(handle, new_inode);
2552 if (!new_inode->i_nlink)
2553 ext4_orphan_add(handle, new_inode);
2554 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2555 force_da_alloc = 1;
2556 }
2557 retval = 0;
2558
2559end_rename:
2560 brelse(dir_bh);
2561 brelse(old_bh);
2562 brelse(new_bh);
2563 ext4_journal_stop(handle);
2564 if (retval == 0 && force_da_alloc)
2565 ext4_alloc_da_blocks(old_inode);
2566 return retval;
2567}
2568
2569/*
2570 * directories can handle most operations...
2571 */
2572const struct inode_operations ext4_dir_inode_operations = {
2573 .create = ext4_create,
2574 .lookup = ext4_lookup,
2575 .link = ext4_link,
2576 .unlink = ext4_unlink,
2577 .symlink = ext4_symlink,
2578 .mkdir = ext4_mkdir,
2579 .rmdir = ext4_rmdir,
2580 .mknod = ext4_mknod,
2581 .rename = ext4_rename,
2582 .setattr = ext4_setattr,
2583#ifdef CONFIG_EXT4_FS_XATTR
2584 .setxattr = generic_setxattr,
2585 .getxattr = generic_getxattr,
2586 .listxattr = ext4_listxattr,
2587 .removexattr = generic_removexattr,
2588#endif
2589 .get_acl = ext4_get_acl,
2590 .fiemap = ext4_fiemap,
2591};
2592
2593const struct inode_operations ext4_special_inode_operations = {
2594 .setattr = ext4_setattr,
2595#ifdef CONFIG_EXT4_FS_XATTR
2596 .setxattr = generic_setxattr,
2597 .getxattr = generic_getxattr,
2598 .listxattr = ext4_listxattr,
2599 .removexattr = generic_removexattr,
2600#endif
2601 .get_acl = ext4_get_acl,
2602};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/namei.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/namei.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
25 * Theodore Ts'o, 2002
26 */
27
28#include <linux/fs.h>
29#include <linux/pagemap.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 <linux/iversion.h>
38#include <linux/unicode.h>
39#include "ext4.h"
40#include "ext4_jbd2.h"
41
42#include "xattr.h"
43#include "acl.h"
44
45#include <trace/events/ext4.h>
46/*
47 * define how far ahead to read directories while searching them.
48 */
49#define NAMEI_RA_CHUNKS 2
50#define NAMEI_RA_BLOCKS 4
51#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52
53static struct buffer_head *ext4_append(handle_t *handle,
54 struct inode *inode,
55 ext4_lblk_t *block)
56{
57 struct ext4_map_blocks map;
58 struct buffer_head *bh;
59 int err;
60
61 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
62 ((inode->i_size >> 10) >=
63 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
64 return ERR_PTR(-ENOSPC);
65
66 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67 map.m_lblk = *block;
68 map.m_len = 1;
69
70 /*
71 * We're appending new directory block. Make sure the block is not
72 * allocated yet, otherwise we will end up corrupting the
73 * directory.
74 */
75 err = ext4_map_blocks(NULL, inode, &map, 0);
76 if (err < 0)
77 return ERR_PTR(err);
78 if (err) {
79 EXT4_ERROR_INODE(inode, "Logical block already allocated");
80 return ERR_PTR(-EFSCORRUPTED);
81 }
82
83 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
84 if (IS_ERR(bh))
85 return bh;
86 inode->i_size += inode->i_sb->s_blocksize;
87 EXT4_I(inode)->i_disksize = inode->i_size;
88 err = ext4_mark_inode_dirty(handle, inode);
89 if (err)
90 goto out;
91 BUFFER_TRACE(bh, "get_write_access");
92 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
93 EXT4_JTR_NONE);
94 if (err)
95 goto out;
96 return bh;
97
98out:
99 brelse(bh);
100 ext4_std_error(inode->i_sb, err);
101 return ERR_PTR(err);
102}
103
104static int ext4_dx_csum_verify(struct inode *inode,
105 struct ext4_dir_entry *dirent);
106
107/*
108 * Hints to ext4_read_dirblock regarding whether we expect a directory
109 * block being read to be an index block, or a block containing
110 * directory entries (and if the latter, whether it was found via a
111 * logical block in an htree index block). This is used to control
112 * what sort of sanity checkinig ext4_read_dirblock() will do on the
113 * directory block read from the storage device. EITHER will means
114 * the caller doesn't know what kind of directory block will be read,
115 * so no specific verification will be done.
116 */
117typedef enum {
118 EITHER, INDEX, DIRENT, DIRENT_HTREE
119} dirblock_type_t;
120
121#define ext4_read_dirblock(inode, block, type) \
122 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
123
124static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
125 ext4_lblk_t block,
126 dirblock_type_t type,
127 const char *func,
128 unsigned int line)
129{
130 struct buffer_head *bh;
131 struct ext4_dir_entry *dirent;
132 int is_dx_block = 0;
133
134 if (block >= inode->i_size >> inode->i_blkbits) {
135 ext4_error_inode(inode, func, line, block,
136 "Attempting to read directory block (%u) that is past i_size (%llu)",
137 block, inode->i_size);
138 return ERR_PTR(-EFSCORRUPTED);
139 }
140
141 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
142 bh = ERR_PTR(-EIO);
143 else
144 bh = ext4_bread(NULL, inode, block, 0);
145 if (IS_ERR(bh)) {
146 __ext4_warning(inode->i_sb, func, line,
147 "inode #%lu: lblock %lu: comm %s: "
148 "error %ld reading directory block",
149 inode->i_ino, (unsigned long)block,
150 current->comm, PTR_ERR(bh));
151
152 return bh;
153 }
154 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
155 ext4_error_inode(inode, func, line, block,
156 "Directory hole found for htree %s block",
157 (type == INDEX) ? "index" : "leaf");
158 return ERR_PTR(-EFSCORRUPTED);
159 }
160 if (!bh)
161 return NULL;
162 dirent = (struct ext4_dir_entry *) bh->b_data;
163 /* Determine whether or not we have an index block */
164 if (is_dx(inode)) {
165 if (block == 0)
166 is_dx_block = 1;
167 else if (ext4_rec_len_from_disk(dirent->rec_len,
168 inode->i_sb->s_blocksize) ==
169 inode->i_sb->s_blocksize)
170 is_dx_block = 1;
171 }
172 if (!is_dx_block && type == INDEX) {
173 ext4_error_inode(inode, func, line, block,
174 "directory leaf block found instead of index block");
175 brelse(bh);
176 return ERR_PTR(-EFSCORRUPTED);
177 }
178 if (!ext4_has_metadata_csum(inode->i_sb) ||
179 buffer_verified(bh))
180 return bh;
181
182 /*
183 * An empty leaf block can get mistaken for a index block; for
184 * this reason, we can only check the index checksum when the
185 * caller is sure it should be an index block.
186 */
187 if (is_dx_block && type == INDEX) {
188 if (ext4_dx_csum_verify(inode, dirent) &&
189 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
190 set_buffer_verified(bh);
191 else {
192 ext4_error_inode_err(inode, func, line, block,
193 EFSBADCRC,
194 "Directory index failed checksum");
195 brelse(bh);
196 return ERR_PTR(-EFSBADCRC);
197 }
198 }
199 if (!is_dx_block) {
200 if (ext4_dirblock_csum_verify(inode, bh) &&
201 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
202 set_buffer_verified(bh);
203 else {
204 ext4_error_inode_err(inode, func, line, block,
205 EFSBADCRC,
206 "Directory block failed checksum");
207 brelse(bh);
208 return ERR_PTR(-EFSBADCRC);
209 }
210 }
211 return bh;
212}
213
214#ifdef DX_DEBUG
215#define dxtrace(command) command
216#else
217#define dxtrace(command)
218#endif
219
220struct fake_dirent
221{
222 __le32 inode;
223 __le16 rec_len;
224 u8 name_len;
225 u8 file_type;
226};
227
228struct dx_countlimit
229{
230 __le16 limit;
231 __le16 count;
232};
233
234struct dx_entry
235{
236 __le32 hash;
237 __le32 block;
238};
239
240/*
241 * dx_root_info is laid out so that if it should somehow get overlaid by a
242 * dirent the two low bits of the hash version will be zero. Therefore, the
243 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
244 */
245
246struct dx_root
247{
248 struct fake_dirent dot;
249 char dot_name[4];
250 struct fake_dirent dotdot;
251 char dotdot_name[4];
252 struct dx_root_info
253 {
254 __le32 reserved_zero;
255 u8 hash_version;
256 u8 info_length; /* 8 */
257 u8 indirect_levels;
258 u8 unused_flags;
259 }
260 info;
261 struct dx_entry entries[];
262};
263
264struct dx_node
265{
266 struct fake_dirent fake;
267 struct dx_entry entries[];
268};
269
270
271struct dx_frame
272{
273 struct buffer_head *bh;
274 struct dx_entry *entries;
275 struct dx_entry *at;
276};
277
278struct dx_map_entry
279{
280 u32 hash;
281 u16 offs;
282 u16 size;
283};
284
285/*
286 * This goes at the end of each htree block.
287 */
288struct dx_tail {
289 u32 dt_reserved;
290 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
291};
292
293static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
294static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
295static inline unsigned dx_get_hash(struct dx_entry *entry);
296static void dx_set_hash(struct dx_entry *entry, unsigned value);
297static unsigned dx_get_count(struct dx_entry *entries);
298static unsigned dx_get_limit(struct dx_entry *entries);
299static void dx_set_count(struct dx_entry *entries, unsigned value);
300static void dx_set_limit(struct dx_entry *entries, unsigned value);
301static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
302static unsigned dx_node_limit(struct inode *dir);
303static struct dx_frame *dx_probe(struct ext4_filename *fname,
304 struct inode *dir,
305 struct dx_hash_info *hinfo,
306 struct dx_frame *frame);
307static void dx_release(struct dx_frame *frames);
308static int dx_make_map(struct inode *dir, struct buffer_head *bh,
309 struct dx_hash_info *hinfo,
310 struct dx_map_entry *map_tail);
311static void dx_sort_map(struct dx_map_entry *map, unsigned count);
312static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
313 char *to, struct dx_map_entry *offsets,
314 int count, unsigned int blocksize);
315static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
316 unsigned int blocksize);
317static void dx_insert_block(struct dx_frame *frame,
318 u32 hash, ext4_lblk_t block);
319static int ext4_htree_next_block(struct inode *dir, __u32 hash,
320 struct dx_frame *frame,
321 struct dx_frame *frames,
322 __u32 *start_hash);
323static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
324 struct ext4_filename *fname,
325 struct ext4_dir_entry_2 **res_dir);
326static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
327 struct inode *dir, struct inode *inode);
328
329/* checksumming functions */
330void ext4_initialize_dirent_tail(struct buffer_head *bh,
331 unsigned int blocksize)
332{
333 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
334
335 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
336 t->det_rec_len = ext4_rec_len_to_disk(
337 sizeof(struct ext4_dir_entry_tail), blocksize);
338 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
339}
340
341/* Walk through a dirent block to find a checksum "dirent" at the tail */
342static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
343 struct buffer_head *bh)
344{
345 struct ext4_dir_entry_tail *t;
346 int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
347
348#ifdef PARANOID
349 struct ext4_dir_entry *d, *top;
350
351 d = (struct ext4_dir_entry *)bh->b_data;
352 top = (struct ext4_dir_entry *)(bh->b_data +
353 (blocksize - sizeof(struct ext4_dir_entry_tail)));
354 while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
355 d = (struct ext4_dir_entry *)(((void *)d) +
356 ext4_rec_len_from_disk(d->rec_len, blocksize));
357
358 if (d != top)
359 return NULL;
360
361 t = (struct ext4_dir_entry_tail *)d;
362#else
363 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
364#endif
365
366 if (t->det_reserved_zero1 ||
367 (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
368 sizeof(struct ext4_dir_entry_tail)) ||
369 t->det_reserved_zero2 ||
370 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
371 return NULL;
372
373 return t;
374}
375
376static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
377{
378 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
379 struct ext4_inode_info *ei = EXT4_I(inode);
380 __u32 csum;
381
382 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
383 return cpu_to_le32(csum);
384}
385
386#define warn_no_space_for_csum(inode) \
387 __warn_no_space_for_csum((inode), __func__, __LINE__)
388
389static void __warn_no_space_for_csum(struct inode *inode, const char *func,
390 unsigned int line)
391{
392 __ext4_warning_inode(inode, func, line,
393 "No space for directory leaf checksum. Please run e2fsck -D.");
394}
395
396int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
397{
398 struct ext4_dir_entry_tail *t;
399
400 if (!ext4_has_metadata_csum(inode->i_sb))
401 return 1;
402
403 t = get_dirent_tail(inode, bh);
404 if (!t) {
405 warn_no_space_for_csum(inode);
406 return 0;
407 }
408
409 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
410 (char *)t - bh->b_data))
411 return 0;
412
413 return 1;
414}
415
416static void ext4_dirblock_csum_set(struct inode *inode,
417 struct buffer_head *bh)
418{
419 struct ext4_dir_entry_tail *t;
420
421 if (!ext4_has_metadata_csum(inode->i_sb))
422 return;
423
424 t = get_dirent_tail(inode, bh);
425 if (!t) {
426 warn_no_space_for_csum(inode);
427 return;
428 }
429
430 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
431 (char *)t - bh->b_data);
432}
433
434int ext4_handle_dirty_dirblock(handle_t *handle,
435 struct inode *inode,
436 struct buffer_head *bh)
437{
438 ext4_dirblock_csum_set(inode, bh);
439 return ext4_handle_dirty_metadata(handle, inode, bh);
440}
441
442static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
443 struct ext4_dir_entry *dirent,
444 int *offset)
445{
446 struct ext4_dir_entry *dp;
447 struct dx_root_info *root;
448 int count_offset;
449 int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
450 unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
451
452 if (rlen == blocksize)
453 count_offset = 8;
454 else if (rlen == 12) {
455 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
456 if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
457 return NULL;
458 root = (struct dx_root_info *)(((void *)dp + 12));
459 if (root->reserved_zero ||
460 root->info_length != sizeof(struct dx_root_info))
461 return NULL;
462 count_offset = 32;
463 } else
464 return NULL;
465
466 if (offset)
467 *offset = count_offset;
468 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
469}
470
471static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
472 int count_offset, int count, struct dx_tail *t)
473{
474 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
475 struct ext4_inode_info *ei = EXT4_I(inode);
476 __u32 csum;
477 int size;
478 __u32 dummy_csum = 0;
479 int offset = offsetof(struct dx_tail, dt_checksum);
480
481 size = count_offset + (count * sizeof(struct dx_entry));
482 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
483 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
484 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
485
486 return cpu_to_le32(csum);
487}
488
489static int ext4_dx_csum_verify(struct inode *inode,
490 struct ext4_dir_entry *dirent)
491{
492 struct dx_countlimit *c;
493 struct dx_tail *t;
494 int count_offset, limit, count;
495
496 if (!ext4_has_metadata_csum(inode->i_sb))
497 return 1;
498
499 c = get_dx_countlimit(inode, dirent, &count_offset);
500 if (!c) {
501 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
502 return 0;
503 }
504 limit = le16_to_cpu(c->limit);
505 count = le16_to_cpu(c->count);
506 if (count_offset + (limit * sizeof(struct dx_entry)) >
507 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
508 warn_no_space_for_csum(inode);
509 return 0;
510 }
511 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
512
513 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
514 count, t))
515 return 0;
516 return 1;
517}
518
519static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
520{
521 struct dx_countlimit *c;
522 struct dx_tail *t;
523 int count_offset, limit, count;
524
525 if (!ext4_has_metadata_csum(inode->i_sb))
526 return;
527
528 c = get_dx_countlimit(inode, dirent, &count_offset);
529 if (!c) {
530 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
531 return;
532 }
533 limit = le16_to_cpu(c->limit);
534 count = le16_to_cpu(c->count);
535 if (count_offset + (limit * sizeof(struct dx_entry)) >
536 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
537 warn_no_space_for_csum(inode);
538 return;
539 }
540 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
541
542 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
543}
544
545static inline int ext4_handle_dirty_dx_node(handle_t *handle,
546 struct inode *inode,
547 struct buffer_head *bh)
548{
549 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
550 return ext4_handle_dirty_metadata(handle, inode, bh);
551}
552
553/*
554 * p is at least 6 bytes before the end of page
555 */
556static inline struct ext4_dir_entry_2 *
557ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
558{
559 return (struct ext4_dir_entry_2 *)((char *)p +
560 ext4_rec_len_from_disk(p->rec_len, blocksize));
561}
562
563/*
564 * Future: use high four bits of block for coalesce-on-delete flags
565 * Mask them off for now.
566 */
567
568static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
569{
570 return le32_to_cpu(entry->block) & 0x0fffffff;
571}
572
573static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
574{
575 entry->block = cpu_to_le32(value);
576}
577
578static inline unsigned dx_get_hash(struct dx_entry *entry)
579{
580 return le32_to_cpu(entry->hash);
581}
582
583static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
584{
585 entry->hash = cpu_to_le32(value);
586}
587
588static inline unsigned dx_get_count(struct dx_entry *entries)
589{
590 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
591}
592
593static inline unsigned dx_get_limit(struct dx_entry *entries)
594{
595 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
596}
597
598static inline void dx_set_count(struct dx_entry *entries, unsigned value)
599{
600 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
601}
602
603static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
604{
605 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
606}
607
608static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
609{
610 unsigned int entry_space = dir->i_sb->s_blocksize -
611 ext4_dir_rec_len(1, NULL) -
612 ext4_dir_rec_len(2, NULL) - infosize;
613
614 if (ext4_has_metadata_csum(dir->i_sb))
615 entry_space -= sizeof(struct dx_tail);
616 return entry_space / sizeof(struct dx_entry);
617}
618
619static inline unsigned dx_node_limit(struct inode *dir)
620{
621 unsigned int entry_space = dir->i_sb->s_blocksize -
622 ext4_dir_rec_len(0, dir);
623
624 if (ext4_has_metadata_csum(dir->i_sb))
625 entry_space -= sizeof(struct dx_tail);
626 return entry_space / sizeof(struct dx_entry);
627}
628
629/*
630 * Debug
631 */
632#ifdef DX_DEBUG
633static void dx_show_index(char * label, struct dx_entry *entries)
634{
635 int i, n = dx_get_count (entries);
636 printk(KERN_DEBUG "%s index", label);
637 for (i = 0; i < n; i++) {
638 printk(KERN_CONT " %x->%lu",
639 i ? dx_get_hash(entries + i) : 0,
640 (unsigned long)dx_get_block(entries + i));
641 }
642 printk(KERN_CONT "\n");
643}
644
645struct stats
646{
647 unsigned names;
648 unsigned space;
649 unsigned bcount;
650};
651
652static struct stats dx_show_leaf(struct inode *dir,
653 struct dx_hash_info *hinfo,
654 struct ext4_dir_entry_2 *de,
655 int size, int show_names)
656{
657 unsigned names = 0, space = 0;
658 char *base = (char *) de;
659 struct dx_hash_info h = *hinfo;
660
661 printk("names: ");
662 while ((char *) de < base + size)
663 {
664 if (de->inode)
665 {
666 if (show_names)
667 {
668#ifdef CONFIG_FS_ENCRYPTION
669 int len;
670 char *name;
671 struct fscrypt_str fname_crypto_str =
672 FSTR_INIT(NULL, 0);
673 int res = 0;
674
675 name = de->name;
676 len = de->name_len;
677 if (!IS_ENCRYPTED(dir)) {
678 /* Directory is not encrypted */
679 (void) ext4fs_dirhash(dir, de->name,
680 de->name_len, &h);
681 printk("%*.s:(U)%x.%u ", len,
682 name, h.hash,
683 (unsigned) ((char *) de
684 - base));
685 } else {
686 struct fscrypt_str de_name =
687 FSTR_INIT(name, len);
688
689 /* Directory is encrypted */
690 res = fscrypt_fname_alloc_buffer(
691 len, &fname_crypto_str);
692 if (res)
693 printk(KERN_WARNING "Error "
694 "allocating crypto "
695 "buffer--skipping "
696 "crypto\n");
697 res = fscrypt_fname_disk_to_usr(dir,
698 0, 0, &de_name,
699 &fname_crypto_str);
700 if (res) {
701 printk(KERN_WARNING "Error "
702 "converting filename "
703 "from disk to usr"
704 "\n");
705 name = "??";
706 len = 2;
707 } else {
708 name = fname_crypto_str.name;
709 len = fname_crypto_str.len;
710 }
711 if (IS_CASEFOLDED(dir))
712 h.hash = EXT4_DIRENT_HASH(de);
713 else
714 (void) ext4fs_dirhash(dir,
715 de->name,
716 de->name_len, &h);
717 printk("%*.s:(E)%x.%u ", len, name,
718 h.hash, (unsigned) ((char *) de
719 - base));
720 fscrypt_fname_free_buffer(
721 &fname_crypto_str);
722 }
723#else
724 int len = de->name_len;
725 char *name = de->name;
726 (void) ext4fs_dirhash(dir, de->name,
727 de->name_len, &h);
728 printk("%*.s:%x.%u ", len, name, h.hash,
729 (unsigned) ((char *) de - base));
730#endif
731 }
732 space += ext4_dir_rec_len(de->name_len, dir);
733 names++;
734 }
735 de = ext4_next_entry(de, size);
736 }
737 printk(KERN_CONT "(%i)\n", names);
738 return (struct stats) { names, space, 1 };
739}
740
741struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
742 struct dx_entry *entries, int levels)
743{
744 unsigned blocksize = dir->i_sb->s_blocksize;
745 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
746 unsigned bcount = 0;
747 struct buffer_head *bh;
748 printk("%i indexed blocks...\n", count);
749 for (i = 0; i < count; i++, entries++)
750 {
751 ext4_lblk_t block = dx_get_block(entries);
752 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
753 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
754 struct stats stats;
755 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
756 bh = ext4_bread(NULL,dir, block, 0);
757 if (!bh || IS_ERR(bh))
758 continue;
759 stats = levels?
760 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
761 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
762 bh->b_data, blocksize, 0);
763 names += stats.names;
764 space += stats.space;
765 bcount += stats.bcount;
766 brelse(bh);
767 }
768 if (bcount)
769 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
770 levels ? "" : " ", names, space/bcount,
771 (space/bcount)*100/blocksize);
772 return (struct stats) { names, space, bcount};
773}
774
775/*
776 * Linear search cross check
777 */
778static inline void htree_rep_invariant_check(struct dx_entry *at,
779 struct dx_entry *target,
780 u32 hash, unsigned int n)
781{
782 while (n--) {
783 dxtrace(printk(KERN_CONT ","));
784 if (dx_get_hash(++at) > hash) {
785 at--;
786 break;
787 }
788 }
789 ASSERT(at == target - 1);
790}
791#else /* DX_DEBUG */
792static inline void htree_rep_invariant_check(struct dx_entry *at,
793 struct dx_entry *target,
794 u32 hash, unsigned int n)
795{
796}
797#endif /* DX_DEBUG */
798
799/*
800 * Probe for a directory leaf block to search.
801 *
802 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
803 * error in the directory index, and the caller should fall back to
804 * searching the directory normally. The callers of dx_probe **MUST**
805 * check for this error code, and make sure it never gets reflected
806 * back to userspace.
807 */
808static struct dx_frame *
809dx_probe(struct ext4_filename *fname, struct inode *dir,
810 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
811{
812 unsigned count, indirect, level, i;
813 struct dx_entry *at, *entries, *p, *q, *m;
814 struct dx_root *root;
815 struct dx_frame *frame = frame_in;
816 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
817 u32 hash;
818 ext4_lblk_t block;
819 ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
820
821 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
822 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
823 if (IS_ERR(frame->bh))
824 return (struct dx_frame *) frame->bh;
825
826 root = (struct dx_root *) frame->bh->b_data;
827 if (root->info.hash_version != DX_HASH_TEA &&
828 root->info.hash_version != DX_HASH_HALF_MD4 &&
829 root->info.hash_version != DX_HASH_LEGACY &&
830 root->info.hash_version != DX_HASH_SIPHASH) {
831 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
832 root->info.hash_version);
833 goto fail;
834 }
835 if (ext4_hash_in_dirent(dir)) {
836 if (root->info.hash_version != DX_HASH_SIPHASH) {
837 ext4_warning_inode(dir,
838 "Hash in dirent, but hash is not SIPHASH");
839 goto fail;
840 }
841 } else {
842 if (root->info.hash_version == DX_HASH_SIPHASH) {
843 ext4_warning_inode(dir,
844 "Hash code is SIPHASH, but hash not in dirent");
845 goto fail;
846 }
847 }
848 if (fname)
849 hinfo = &fname->hinfo;
850 hinfo->hash_version = root->info.hash_version;
851 if (hinfo->hash_version <= DX_HASH_TEA)
852 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
853 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
854 /* hash is already computed for encrypted casefolded directory */
855 if (fname && fname_name(fname) &&
856 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
857 int ret = ext4fs_dirhash(dir, fname_name(fname),
858 fname_len(fname), hinfo);
859 if (ret < 0) {
860 ret_err = ERR_PTR(ret);
861 goto fail;
862 }
863 }
864 hash = hinfo->hash;
865
866 if (root->info.unused_flags & 1) {
867 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
868 root->info.unused_flags);
869 goto fail;
870 }
871
872 indirect = root->info.indirect_levels;
873 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
874 ext4_warning(dir->i_sb,
875 "Directory (ino: %lu) htree depth %#06x exceed"
876 "supported value", dir->i_ino,
877 ext4_dir_htree_level(dir->i_sb));
878 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
879 ext4_warning(dir->i_sb, "Enable large directory "
880 "feature to access it");
881 }
882 goto fail;
883 }
884
885 entries = (struct dx_entry *)(((char *)&root->info) +
886 root->info.info_length);
887
888 if (dx_get_limit(entries) != dx_root_limit(dir,
889 root->info.info_length)) {
890 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
891 dx_get_limit(entries),
892 dx_root_limit(dir, root->info.info_length));
893 goto fail;
894 }
895
896 dxtrace(printk("Look up %x", hash));
897 level = 0;
898 blocks[0] = 0;
899 while (1) {
900 count = dx_get_count(entries);
901 if (!count || count > dx_get_limit(entries)) {
902 ext4_warning_inode(dir,
903 "dx entry: count %u beyond limit %u",
904 count, dx_get_limit(entries));
905 goto fail;
906 }
907
908 p = entries + 1;
909 q = entries + count - 1;
910 while (p <= q) {
911 m = p + (q - p) / 2;
912 dxtrace(printk(KERN_CONT "."));
913 if (dx_get_hash(m) > hash)
914 q = m - 1;
915 else
916 p = m + 1;
917 }
918
919 htree_rep_invariant_check(entries, p, hash, count - 1);
920
921 at = p - 1;
922 dxtrace(printk(KERN_CONT " %x->%u\n",
923 at == entries ? 0 : dx_get_hash(at),
924 dx_get_block(at)));
925 frame->entries = entries;
926 frame->at = at;
927
928 block = dx_get_block(at);
929 for (i = 0; i <= level; i++) {
930 if (blocks[i] == block) {
931 ext4_warning_inode(dir,
932 "dx entry: tree cycle block %u points back to block %u",
933 blocks[level], block);
934 goto fail;
935 }
936 }
937 if (++level > indirect)
938 return frame;
939 blocks[level] = block;
940 frame++;
941 frame->bh = ext4_read_dirblock(dir, block, INDEX);
942 if (IS_ERR(frame->bh)) {
943 ret_err = (struct dx_frame *) frame->bh;
944 frame->bh = NULL;
945 goto fail;
946 }
947
948 entries = ((struct dx_node *) frame->bh->b_data)->entries;
949
950 if (dx_get_limit(entries) != dx_node_limit(dir)) {
951 ext4_warning_inode(dir,
952 "dx entry: limit %u != node limit %u",
953 dx_get_limit(entries), dx_node_limit(dir));
954 goto fail;
955 }
956 }
957fail:
958 while (frame >= frame_in) {
959 brelse(frame->bh);
960 frame--;
961 }
962
963 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
964 ext4_warning_inode(dir,
965 "Corrupt directory, running e2fsck is recommended");
966 return ret_err;
967}
968
969static void dx_release(struct dx_frame *frames)
970{
971 struct dx_root_info *info;
972 int i;
973 unsigned int indirect_levels;
974
975 if (frames[0].bh == NULL)
976 return;
977
978 info = &((struct dx_root *)frames[0].bh->b_data)->info;
979 /* save local copy, "info" may be freed after brelse() */
980 indirect_levels = info->indirect_levels;
981 for (i = 0; i <= indirect_levels; i++) {
982 if (frames[i].bh == NULL)
983 break;
984 brelse(frames[i].bh);
985 frames[i].bh = NULL;
986 }
987}
988
989/*
990 * This function increments the frame pointer to search the next leaf
991 * block, and reads in the necessary intervening nodes if the search
992 * should be necessary. Whether or not the search is necessary is
993 * controlled by the hash parameter. If the hash value is even, then
994 * the search is only continued if the next block starts with that
995 * hash value. This is used if we are searching for a specific file.
996 *
997 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
998 *
999 * This function returns 1 if the caller should continue to search,
1000 * or 0 if it should not. If there is an error reading one of the
1001 * index blocks, it will a negative error code.
1002 *
1003 * If start_hash is non-null, it will be filled in with the starting
1004 * hash of the next page.
1005 */
1006static int ext4_htree_next_block(struct inode *dir, __u32 hash,
1007 struct dx_frame *frame,
1008 struct dx_frame *frames,
1009 __u32 *start_hash)
1010{
1011 struct dx_frame *p;
1012 struct buffer_head *bh;
1013 int num_frames = 0;
1014 __u32 bhash;
1015
1016 p = frame;
1017 /*
1018 * Find the next leaf page by incrementing the frame pointer.
1019 * If we run out of entries in the interior node, loop around and
1020 * increment pointer in the parent node. When we break out of
1021 * this loop, num_frames indicates the number of interior
1022 * nodes need to be read.
1023 */
1024 while (1) {
1025 if (++(p->at) < p->entries + dx_get_count(p->entries))
1026 break;
1027 if (p == frames)
1028 return 0;
1029 num_frames++;
1030 p--;
1031 }
1032
1033 /*
1034 * If the hash is 1, then continue only if the next page has a
1035 * continuation hash of any value. This is used for readdir
1036 * handling. Otherwise, check to see if the hash matches the
1037 * desired continuation hash. If it doesn't, return since
1038 * there's no point to read in the successive index pages.
1039 */
1040 bhash = dx_get_hash(p->at);
1041 if (start_hash)
1042 *start_hash = bhash;
1043 if ((hash & 1) == 0) {
1044 if ((bhash & ~1) != hash)
1045 return 0;
1046 }
1047 /*
1048 * If the hash is HASH_NB_ALWAYS, we always go to the next
1049 * block so no check is necessary
1050 */
1051 while (num_frames--) {
1052 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1053 if (IS_ERR(bh))
1054 return PTR_ERR(bh);
1055 p++;
1056 brelse(p->bh);
1057 p->bh = bh;
1058 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1059 }
1060 return 1;
1061}
1062
1063
1064/*
1065 * This function fills a red-black tree with information from a
1066 * directory block. It returns the number directory entries loaded
1067 * into the tree. If there is an error it is returned in err.
1068 */
1069static int htree_dirblock_to_tree(struct file *dir_file,
1070 struct inode *dir, ext4_lblk_t block,
1071 struct dx_hash_info *hinfo,
1072 __u32 start_hash, __u32 start_minor_hash)
1073{
1074 struct buffer_head *bh;
1075 struct ext4_dir_entry_2 *de, *top;
1076 int err = 0, count = 0;
1077 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1078 int csum = ext4_has_metadata_csum(dir->i_sb);
1079
1080 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1081 (unsigned long)block));
1082 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1083 if (IS_ERR(bh))
1084 return PTR_ERR(bh);
1085
1086 de = (struct ext4_dir_entry_2 *) bh->b_data;
1087 /* csum entries are not larger in the casefolded encrypted case */
1088 top = (struct ext4_dir_entry_2 *) ((char *) de +
1089 dir->i_sb->s_blocksize -
1090 ext4_dir_rec_len(0,
1091 csum ? NULL : dir));
1092 /* Check if the directory is encrypted */
1093 if (IS_ENCRYPTED(dir)) {
1094 err = fscrypt_prepare_readdir(dir);
1095 if (err < 0) {
1096 brelse(bh);
1097 return err;
1098 }
1099 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1100 &fname_crypto_str);
1101 if (err < 0) {
1102 brelse(bh);
1103 return err;
1104 }
1105 }
1106
1107 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1108 if (ext4_check_dir_entry(dir, NULL, de, bh,
1109 bh->b_data, bh->b_size,
1110 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1111 + ((char *)de - bh->b_data))) {
1112 /* silently ignore the rest of the block */
1113 break;
1114 }
1115 if (ext4_hash_in_dirent(dir)) {
1116 if (de->name_len && de->inode) {
1117 hinfo->hash = EXT4_DIRENT_HASH(de);
1118 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1119 } else {
1120 hinfo->hash = 0;
1121 hinfo->minor_hash = 0;
1122 }
1123 } else {
1124 err = ext4fs_dirhash(dir, de->name,
1125 de->name_len, hinfo);
1126 if (err < 0) {
1127 count = err;
1128 goto errout;
1129 }
1130 }
1131 if ((hinfo->hash < start_hash) ||
1132 ((hinfo->hash == start_hash) &&
1133 (hinfo->minor_hash < start_minor_hash)))
1134 continue;
1135 if (de->inode == 0)
1136 continue;
1137 if (!IS_ENCRYPTED(dir)) {
1138 tmp_str.name = de->name;
1139 tmp_str.len = de->name_len;
1140 err = ext4_htree_store_dirent(dir_file,
1141 hinfo->hash, hinfo->minor_hash, de,
1142 &tmp_str);
1143 } else {
1144 int save_len = fname_crypto_str.len;
1145 struct fscrypt_str de_name = FSTR_INIT(de->name,
1146 de->name_len);
1147
1148 /* Directory is encrypted */
1149 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1150 hinfo->minor_hash, &de_name,
1151 &fname_crypto_str);
1152 if (err) {
1153 count = err;
1154 goto errout;
1155 }
1156 err = ext4_htree_store_dirent(dir_file,
1157 hinfo->hash, hinfo->minor_hash, de,
1158 &fname_crypto_str);
1159 fname_crypto_str.len = save_len;
1160 }
1161 if (err != 0) {
1162 count = err;
1163 goto errout;
1164 }
1165 count++;
1166 }
1167errout:
1168 brelse(bh);
1169 fscrypt_fname_free_buffer(&fname_crypto_str);
1170 return count;
1171}
1172
1173
1174/*
1175 * This function fills a red-black tree with information from a
1176 * directory. We start scanning the directory in hash order, starting
1177 * at start_hash and start_minor_hash.
1178 *
1179 * This function returns the number of entries inserted into the tree,
1180 * or a negative error code.
1181 */
1182int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1183 __u32 start_minor_hash, __u32 *next_hash)
1184{
1185 struct dx_hash_info hinfo;
1186 struct ext4_dir_entry_2 *de;
1187 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1188 struct inode *dir;
1189 ext4_lblk_t block;
1190 int count = 0;
1191 int ret, err;
1192 __u32 hashval;
1193 struct fscrypt_str tmp_str;
1194
1195 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1196 start_hash, start_minor_hash));
1197 dir = file_inode(dir_file);
1198 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1199 if (ext4_hash_in_dirent(dir))
1200 hinfo.hash_version = DX_HASH_SIPHASH;
1201 else
1202 hinfo.hash_version =
1203 EXT4_SB(dir->i_sb)->s_def_hash_version;
1204 if (hinfo.hash_version <= DX_HASH_TEA)
1205 hinfo.hash_version +=
1206 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1207 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1208 if (ext4_has_inline_data(dir)) {
1209 int has_inline_data = 1;
1210 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1211 &hinfo, start_hash,
1212 start_minor_hash,
1213 &has_inline_data);
1214 if (has_inline_data) {
1215 *next_hash = ~0;
1216 return count;
1217 }
1218 }
1219 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1220 start_hash, start_minor_hash);
1221 *next_hash = ~0;
1222 return count;
1223 }
1224 hinfo.hash = start_hash;
1225 hinfo.minor_hash = 0;
1226 frame = dx_probe(NULL, dir, &hinfo, frames);
1227 if (IS_ERR(frame))
1228 return PTR_ERR(frame);
1229
1230 /* Add '.' and '..' from the htree header */
1231 if (!start_hash && !start_minor_hash) {
1232 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1233 tmp_str.name = de->name;
1234 tmp_str.len = de->name_len;
1235 err = ext4_htree_store_dirent(dir_file, 0, 0,
1236 de, &tmp_str);
1237 if (err != 0)
1238 goto errout;
1239 count++;
1240 }
1241 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1242 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1243 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1244 tmp_str.name = de->name;
1245 tmp_str.len = de->name_len;
1246 err = ext4_htree_store_dirent(dir_file, 2, 0,
1247 de, &tmp_str);
1248 if (err != 0)
1249 goto errout;
1250 count++;
1251 }
1252
1253 while (1) {
1254 if (fatal_signal_pending(current)) {
1255 err = -ERESTARTSYS;
1256 goto errout;
1257 }
1258 cond_resched();
1259 block = dx_get_block(frame->at);
1260 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1261 start_hash, start_minor_hash);
1262 if (ret < 0) {
1263 err = ret;
1264 goto errout;
1265 }
1266 count += ret;
1267 hashval = ~0;
1268 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1269 frame, frames, &hashval);
1270 *next_hash = hashval;
1271 if (ret < 0) {
1272 err = ret;
1273 goto errout;
1274 }
1275 /*
1276 * Stop if: (a) there are no more entries, or
1277 * (b) we have inserted at least one entry and the
1278 * next hash value is not a continuation
1279 */
1280 if ((ret == 0) ||
1281 (count && ((hashval & 1) == 0)))
1282 break;
1283 }
1284 dx_release(frames);
1285 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1286 "next hash: %x\n", count, *next_hash));
1287 return count;
1288errout:
1289 dx_release(frames);
1290 return (err);
1291}
1292
1293static inline int search_dirblock(struct buffer_head *bh,
1294 struct inode *dir,
1295 struct ext4_filename *fname,
1296 unsigned int offset,
1297 struct ext4_dir_entry_2 **res_dir)
1298{
1299 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1300 fname, offset, res_dir);
1301}
1302
1303/*
1304 * Directory block splitting, compacting
1305 */
1306
1307/*
1308 * Create map of hash values, offsets, and sizes, stored at end of block.
1309 * Returns number of entries mapped.
1310 */
1311static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1312 struct dx_hash_info *hinfo,
1313 struct dx_map_entry *map_tail)
1314{
1315 int count = 0;
1316 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1317 unsigned int buflen = bh->b_size;
1318 char *base = bh->b_data;
1319 struct dx_hash_info h = *hinfo;
1320 int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
1321
1322 if (ext4_has_metadata_csum(dir->i_sb))
1323 buflen -= sizeof(struct ext4_dir_entry_tail);
1324
1325 while ((char *) de < base + buflen) {
1326 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1327 ((char *)de) - base))
1328 return -EFSCORRUPTED;
1329 if (de->name_len && de->inode) {
1330 if (ext4_hash_in_dirent(dir))
1331 h.hash = EXT4_DIRENT_HASH(de);
1332 else {
1333 int err = ext4fs_dirhash(dir, de->name,
1334 de->name_len, &h);
1335 if (err < 0)
1336 return err;
1337 }
1338 map_tail--;
1339 map_tail->hash = h.hash;
1340 map_tail->offs = ((char *) de - base)>>2;
1341 map_tail->size = ext4_rec_len_from_disk(de->rec_len,
1342 blocksize);
1343 count++;
1344 cond_resched();
1345 }
1346 de = ext4_next_entry(de, blocksize);
1347 }
1348 return count;
1349}
1350
1351/* Sort map by hash value */
1352static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1353{
1354 struct dx_map_entry *p, *q, *top = map + count - 1;
1355 int more;
1356 /* Combsort until bubble sort doesn't suck */
1357 while (count > 2) {
1358 count = count*10/13;
1359 if (count - 9 < 2) /* 9, 10 -> 11 */
1360 count = 11;
1361 for (p = top, q = p - count; q >= map; p--, q--)
1362 if (p->hash < q->hash)
1363 swap(*p, *q);
1364 }
1365 /* Garden variety bubble sort */
1366 do {
1367 more = 0;
1368 q = top;
1369 while (q-- > map) {
1370 if (q[1].hash >= q[0].hash)
1371 continue;
1372 swap(*(q+1), *q);
1373 more = 1;
1374 }
1375 } while(more);
1376}
1377
1378static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1379{
1380 struct dx_entry *entries = frame->entries;
1381 struct dx_entry *old = frame->at, *new = old + 1;
1382 int count = dx_get_count(entries);
1383
1384 ASSERT(count < dx_get_limit(entries));
1385 ASSERT(old < entries + count);
1386 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1387 dx_set_hash(new, hash);
1388 dx_set_block(new, block);
1389 dx_set_count(entries, count + 1);
1390}
1391
1392#if IS_ENABLED(CONFIG_UNICODE)
1393/*
1394 * Test whether a case-insensitive directory entry matches the filename
1395 * being searched for. If quick is set, assume the name being looked up
1396 * is already in the casefolded form.
1397 *
1398 * Returns: 0 if the directory entry matches, more than 0 if it
1399 * doesn't match or less than zero on error.
1400 */
1401static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1402 u8 *de_name, size_t de_name_len, bool quick)
1403{
1404 const struct super_block *sb = parent->i_sb;
1405 const struct unicode_map *um = sb->s_encoding;
1406 struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1407 struct qstr entry = QSTR_INIT(de_name, de_name_len);
1408 int ret;
1409
1410 if (IS_ENCRYPTED(parent)) {
1411 const struct fscrypt_str encrypted_name =
1412 FSTR_INIT(de_name, de_name_len);
1413
1414 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1415 if (!decrypted_name.name)
1416 return -ENOMEM;
1417 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1418 &decrypted_name);
1419 if (ret < 0)
1420 goto out;
1421 entry.name = decrypted_name.name;
1422 entry.len = decrypted_name.len;
1423 }
1424
1425 if (quick)
1426 ret = utf8_strncasecmp_folded(um, name, &entry);
1427 else
1428 ret = utf8_strncasecmp(um, name, &entry);
1429 if (ret < 0) {
1430 /* Handle invalid character sequence as either an error
1431 * or as an opaque byte sequence.
1432 */
1433 if (sb_has_strict_encoding(sb))
1434 ret = -EINVAL;
1435 else if (name->len != entry.len)
1436 ret = 1;
1437 else
1438 ret = !!memcmp(name->name, entry.name, entry.len);
1439 }
1440out:
1441 kfree(decrypted_name.name);
1442 return ret;
1443}
1444
1445int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1446 struct ext4_filename *name)
1447{
1448 struct fscrypt_str *cf_name = &name->cf_name;
1449 struct dx_hash_info *hinfo = &name->hinfo;
1450 int len;
1451
1452 if (!IS_CASEFOLDED(dir) ||
1453 (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1454 cf_name->name = NULL;
1455 return 0;
1456 }
1457
1458 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1459 if (!cf_name->name)
1460 return -ENOMEM;
1461
1462 len = utf8_casefold(dir->i_sb->s_encoding,
1463 iname, cf_name->name,
1464 EXT4_NAME_LEN);
1465 if (len <= 0) {
1466 kfree(cf_name->name);
1467 cf_name->name = NULL;
1468 }
1469 cf_name->len = (unsigned) len;
1470 if (!IS_ENCRYPTED(dir))
1471 return 0;
1472
1473 hinfo->hash_version = DX_HASH_SIPHASH;
1474 hinfo->seed = NULL;
1475 if (cf_name->name)
1476 return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1477 else
1478 return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1479}
1480#endif
1481
1482/*
1483 * Test whether a directory entry matches the filename being searched for.
1484 *
1485 * Return: %true if the directory entry matches, otherwise %false.
1486 */
1487static bool ext4_match(struct inode *parent,
1488 const struct ext4_filename *fname,
1489 struct ext4_dir_entry_2 *de)
1490{
1491 struct fscrypt_name f;
1492
1493 if (!de->inode)
1494 return false;
1495
1496 f.usr_fname = fname->usr_fname;
1497 f.disk_name = fname->disk_name;
1498#ifdef CONFIG_FS_ENCRYPTION
1499 f.crypto_buf = fname->crypto_buf;
1500#endif
1501
1502#if IS_ENABLED(CONFIG_UNICODE)
1503 if (IS_CASEFOLDED(parent) &&
1504 (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1505 if (fname->cf_name.name) {
1506 struct qstr cf = {.name = fname->cf_name.name,
1507 .len = fname->cf_name.len};
1508 if (IS_ENCRYPTED(parent)) {
1509 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1510 fname->hinfo.minor_hash !=
1511 EXT4_DIRENT_MINOR_HASH(de)) {
1512
1513 return false;
1514 }
1515 }
1516 return !ext4_ci_compare(parent, &cf, de->name,
1517 de->name_len, true);
1518 }
1519 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1520 de->name_len, false);
1521 }
1522#endif
1523
1524 return fscrypt_match_name(&f, de->name, de->name_len);
1525}
1526
1527/*
1528 * Returns 0 if not found, -1 on failure, and 1 on success
1529 */
1530int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1531 struct inode *dir, struct ext4_filename *fname,
1532 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1533{
1534 struct ext4_dir_entry_2 * de;
1535 char * dlimit;
1536 int de_len;
1537
1538 de = (struct ext4_dir_entry_2 *)search_buf;
1539 dlimit = search_buf + buf_size;
1540 while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1541 /* this code is executed quadratically often */
1542 /* do minimal checking `by hand' */
1543 if (de->name + de->name_len <= dlimit &&
1544 ext4_match(dir, fname, de)) {
1545 /* found a match - just to be sure, do
1546 * a full check */
1547 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1548 buf_size, offset))
1549 return -1;
1550 *res_dir = de;
1551 return 1;
1552 }
1553 /* prevent looping on a bad block */
1554 de_len = ext4_rec_len_from_disk(de->rec_len,
1555 dir->i_sb->s_blocksize);
1556 if (de_len <= 0)
1557 return -1;
1558 offset += de_len;
1559 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1560 }
1561 return 0;
1562}
1563
1564static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1565 struct ext4_dir_entry *de)
1566{
1567 struct super_block *sb = dir->i_sb;
1568
1569 if (!is_dx(dir))
1570 return 0;
1571 if (block == 0)
1572 return 1;
1573 if (de->inode == 0 &&
1574 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1575 sb->s_blocksize)
1576 return 1;
1577 return 0;
1578}
1579
1580/*
1581 * __ext4_find_entry()
1582 *
1583 * finds an entry in the specified directory with the wanted name. It
1584 * returns the cache buffer in which the entry was found, and the entry
1585 * itself (as a parameter - res_dir). It does NOT read the inode of the
1586 * entry - you'll have to do that yourself if you want to.
1587 *
1588 * The returned buffer_head has ->b_count elevated. The caller is expected
1589 * to brelse() it when appropriate.
1590 */
1591static struct buffer_head *__ext4_find_entry(struct inode *dir,
1592 struct ext4_filename *fname,
1593 struct ext4_dir_entry_2 **res_dir,
1594 int *inlined)
1595{
1596 struct super_block *sb;
1597 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1598 struct buffer_head *bh, *ret = NULL;
1599 ext4_lblk_t start, block;
1600 const u8 *name = fname->usr_fname->name;
1601 size_t ra_max = 0; /* Number of bh's in the readahead
1602 buffer, bh_use[] */
1603 size_t ra_ptr = 0; /* Current index into readahead
1604 buffer */
1605 ext4_lblk_t nblocks;
1606 int i, namelen, retval;
1607
1608 *res_dir = NULL;
1609 sb = dir->i_sb;
1610 namelen = fname->usr_fname->len;
1611 if (namelen > EXT4_NAME_LEN)
1612 return NULL;
1613
1614 if (ext4_has_inline_data(dir)) {
1615 int has_inline_data = 1;
1616 ret = ext4_find_inline_entry(dir, fname, res_dir,
1617 &has_inline_data);
1618 if (inlined)
1619 *inlined = has_inline_data;
1620 if (has_inline_data)
1621 goto cleanup_and_exit;
1622 }
1623
1624 if ((namelen <= 2) && (name[0] == '.') &&
1625 (name[1] == '.' || name[1] == '\0')) {
1626 /*
1627 * "." or ".." will only be in the first block
1628 * NFS may look up ".."; "." should be handled by the VFS
1629 */
1630 block = start = 0;
1631 nblocks = 1;
1632 goto restart;
1633 }
1634 if (is_dx(dir)) {
1635 ret = ext4_dx_find_entry(dir, fname, res_dir);
1636 /*
1637 * On success, or if the error was file not found,
1638 * return. Otherwise, fall back to doing a search the
1639 * old fashioned way.
1640 */
1641 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1642 goto cleanup_and_exit;
1643 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1644 "falling back\n"));
1645 ret = NULL;
1646 }
1647 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1648 if (!nblocks) {
1649 ret = NULL;
1650 goto cleanup_and_exit;
1651 }
1652 start = EXT4_I(dir)->i_dir_start_lookup;
1653 if (start >= nblocks)
1654 start = 0;
1655 block = start;
1656restart:
1657 do {
1658 /*
1659 * We deal with the read-ahead logic here.
1660 */
1661 cond_resched();
1662 if (ra_ptr >= ra_max) {
1663 /* Refill the readahead buffer */
1664 ra_ptr = 0;
1665 if (block < start)
1666 ra_max = start - block;
1667 else
1668 ra_max = nblocks - block;
1669 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1670 retval = ext4_bread_batch(dir, block, ra_max,
1671 false /* wait */, bh_use);
1672 if (retval) {
1673 ret = ERR_PTR(retval);
1674 ra_max = 0;
1675 goto cleanup_and_exit;
1676 }
1677 }
1678 if ((bh = bh_use[ra_ptr++]) == NULL)
1679 goto next;
1680 wait_on_buffer(bh);
1681 if (!buffer_uptodate(bh)) {
1682 EXT4_ERROR_INODE_ERR(dir, EIO,
1683 "reading directory lblock %lu",
1684 (unsigned long) block);
1685 brelse(bh);
1686 ret = ERR_PTR(-EIO);
1687 goto cleanup_and_exit;
1688 }
1689 if (!buffer_verified(bh) &&
1690 !is_dx_internal_node(dir, block,
1691 (struct ext4_dir_entry *)bh->b_data) &&
1692 !ext4_dirblock_csum_verify(dir, bh)) {
1693 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1694 "checksumming directory "
1695 "block %lu", (unsigned long)block);
1696 brelse(bh);
1697 ret = ERR_PTR(-EFSBADCRC);
1698 goto cleanup_and_exit;
1699 }
1700 set_buffer_verified(bh);
1701 i = search_dirblock(bh, dir, fname,
1702 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1703 if (i == 1) {
1704 EXT4_I(dir)->i_dir_start_lookup = block;
1705 ret = bh;
1706 goto cleanup_and_exit;
1707 } else {
1708 brelse(bh);
1709 if (i < 0)
1710 goto cleanup_and_exit;
1711 }
1712 next:
1713 if (++block >= nblocks)
1714 block = 0;
1715 } while (block != start);
1716
1717 /*
1718 * If the directory has grown while we were searching, then
1719 * search the last part of the directory before giving up.
1720 */
1721 block = nblocks;
1722 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1723 if (block < nblocks) {
1724 start = 0;
1725 goto restart;
1726 }
1727
1728cleanup_and_exit:
1729 /* Clean up the read-ahead blocks */
1730 for (; ra_ptr < ra_max; ra_ptr++)
1731 brelse(bh_use[ra_ptr]);
1732 return ret;
1733}
1734
1735static struct buffer_head *ext4_find_entry(struct inode *dir,
1736 const struct qstr *d_name,
1737 struct ext4_dir_entry_2 **res_dir,
1738 int *inlined)
1739{
1740 int err;
1741 struct ext4_filename fname;
1742 struct buffer_head *bh;
1743
1744 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1745 if (err == -ENOENT)
1746 return NULL;
1747 if (err)
1748 return ERR_PTR(err);
1749
1750 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1751
1752 ext4_fname_free_filename(&fname);
1753 return bh;
1754}
1755
1756static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1757 struct dentry *dentry,
1758 struct ext4_dir_entry_2 **res_dir)
1759{
1760 int err;
1761 struct ext4_filename fname;
1762 struct buffer_head *bh;
1763
1764 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1765 generic_set_encrypted_ci_d_ops(dentry);
1766 if (err == -ENOENT)
1767 return NULL;
1768 if (err)
1769 return ERR_PTR(err);
1770
1771 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1772
1773 ext4_fname_free_filename(&fname);
1774 return bh;
1775}
1776
1777static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1778 struct ext4_filename *fname,
1779 struct ext4_dir_entry_2 **res_dir)
1780{
1781 struct super_block * sb = dir->i_sb;
1782 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1783 struct buffer_head *bh;
1784 ext4_lblk_t block;
1785 int retval;
1786
1787#ifdef CONFIG_FS_ENCRYPTION
1788 *res_dir = NULL;
1789#endif
1790 frame = dx_probe(fname, dir, NULL, frames);
1791 if (IS_ERR(frame))
1792 return (struct buffer_head *) frame;
1793 do {
1794 block = dx_get_block(frame->at);
1795 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1796 if (IS_ERR(bh))
1797 goto errout;
1798
1799 retval = search_dirblock(bh, dir, fname,
1800 block << EXT4_BLOCK_SIZE_BITS(sb),
1801 res_dir);
1802 if (retval == 1)
1803 goto success;
1804 brelse(bh);
1805 if (retval == -1) {
1806 bh = ERR_PTR(ERR_BAD_DX_DIR);
1807 goto errout;
1808 }
1809
1810 /* Check to see if we should continue to search */
1811 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1812 frames, NULL);
1813 if (retval < 0) {
1814 ext4_warning_inode(dir,
1815 "error %d reading directory index block",
1816 retval);
1817 bh = ERR_PTR(retval);
1818 goto errout;
1819 }
1820 } while (retval == 1);
1821
1822 bh = NULL;
1823errout:
1824 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1825success:
1826 dx_release(frames);
1827 return bh;
1828}
1829
1830static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1831{
1832 struct inode *inode;
1833 struct ext4_dir_entry_2 *de;
1834 struct buffer_head *bh;
1835
1836 if (dentry->d_name.len > EXT4_NAME_LEN)
1837 return ERR_PTR(-ENAMETOOLONG);
1838
1839 bh = ext4_lookup_entry(dir, dentry, &de);
1840 if (IS_ERR(bh))
1841 return ERR_CAST(bh);
1842 inode = NULL;
1843 if (bh) {
1844 __u32 ino = le32_to_cpu(de->inode);
1845 brelse(bh);
1846 if (!ext4_valid_inum(dir->i_sb, ino)) {
1847 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1848 return ERR_PTR(-EFSCORRUPTED);
1849 }
1850 if (unlikely(ino == dir->i_ino)) {
1851 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1852 dentry);
1853 return ERR_PTR(-EFSCORRUPTED);
1854 }
1855 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1856 if (inode == ERR_PTR(-ESTALE)) {
1857 EXT4_ERROR_INODE(dir,
1858 "deleted inode referenced: %u",
1859 ino);
1860 return ERR_PTR(-EFSCORRUPTED);
1861 }
1862 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1863 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1864 !fscrypt_has_permitted_context(dir, inode)) {
1865 ext4_warning(inode->i_sb,
1866 "Inconsistent encryption contexts: %lu/%lu",
1867 dir->i_ino, inode->i_ino);
1868 iput(inode);
1869 return ERR_PTR(-EPERM);
1870 }
1871 }
1872
1873#if IS_ENABLED(CONFIG_UNICODE)
1874 if (!inode && IS_CASEFOLDED(dir)) {
1875 /* Eventually we want to call d_add_ci(dentry, NULL)
1876 * for negative dentries in the encoding case as
1877 * well. For now, prevent the negative dentry
1878 * from being cached.
1879 */
1880 return NULL;
1881 }
1882#endif
1883 return d_splice_alias(inode, dentry);
1884}
1885
1886
1887struct dentry *ext4_get_parent(struct dentry *child)
1888{
1889 __u32 ino;
1890 struct ext4_dir_entry_2 * de;
1891 struct buffer_head *bh;
1892
1893 bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1894 if (IS_ERR(bh))
1895 return ERR_CAST(bh);
1896 if (!bh)
1897 return ERR_PTR(-ENOENT);
1898 ino = le32_to_cpu(de->inode);
1899 brelse(bh);
1900
1901 if (!ext4_valid_inum(child->d_sb, ino)) {
1902 EXT4_ERROR_INODE(d_inode(child),
1903 "bad parent inode number: %u", ino);
1904 return ERR_PTR(-EFSCORRUPTED);
1905 }
1906
1907 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1908}
1909
1910/*
1911 * Move count entries from end of map between two memory locations.
1912 * Returns pointer to last entry moved.
1913 */
1914static struct ext4_dir_entry_2 *
1915dx_move_dirents(struct inode *dir, char *from, char *to,
1916 struct dx_map_entry *map, int count,
1917 unsigned blocksize)
1918{
1919 unsigned rec_len = 0;
1920
1921 while (count--) {
1922 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1923 (from + (map->offs<<2));
1924 rec_len = ext4_dir_rec_len(de->name_len, dir);
1925
1926 memcpy (to, de, rec_len);
1927 ((struct ext4_dir_entry_2 *) to)->rec_len =
1928 ext4_rec_len_to_disk(rec_len, blocksize);
1929
1930 /* wipe dir_entry excluding the rec_len field */
1931 de->inode = 0;
1932 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1933 blocksize) -
1934 offsetof(struct ext4_dir_entry_2,
1935 name_len));
1936
1937 map++;
1938 to += rec_len;
1939 }
1940 return (struct ext4_dir_entry_2 *) (to - rec_len);
1941}
1942
1943/*
1944 * Compact each dir entry in the range to the minimal rec_len.
1945 * Returns pointer to last entry in range.
1946 */
1947static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1948 unsigned int blocksize)
1949{
1950 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1951 unsigned rec_len = 0;
1952
1953 prev = to = de;
1954 while ((char*)de < base + blocksize) {
1955 next = ext4_next_entry(de, blocksize);
1956 if (de->inode && de->name_len) {
1957 rec_len = ext4_dir_rec_len(de->name_len, dir);
1958 if (de > to)
1959 memmove(to, de, rec_len);
1960 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1961 prev = to;
1962 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1963 }
1964 de = next;
1965 }
1966 return prev;
1967}
1968
1969/*
1970 * Split a full leaf block to make room for a new dir entry.
1971 * Allocate a new block, and move entries so that they are approx. equally full.
1972 * Returns pointer to de in block into which the new entry will be inserted.
1973 */
1974static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1975 struct buffer_head **bh,struct dx_frame *frame,
1976 struct dx_hash_info *hinfo)
1977{
1978 unsigned blocksize = dir->i_sb->s_blocksize;
1979 unsigned continued;
1980 int count;
1981 struct buffer_head *bh2;
1982 ext4_lblk_t newblock;
1983 u32 hash2;
1984 struct dx_map_entry *map;
1985 char *data1 = (*bh)->b_data, *data2;
1986 unsigned split, move, size;
1987 struct ext4_dir_entry_2 *de = NULL, *de2;
1988 int csum_size = 0;
1989 int err = 0, i;
1990
1991 if (ext4_has_metadata_csum(dir->i_sb))
1992 csum_size = sizeof(struct ext4_dir_entry_tail);
1993
1994 bh2 = ext4_append(handle, dir, &newblock);
1995 if (IS_ERR(bh2)) {
1996 brelse(*bh);
1997 *bh = NULL;
1998 return (struct ext4_dir_entry_2 *) bh2;
1999 }
2000
2001 BUFFER_TRACE(*bh, "get_write_access");
2002 err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
2003 EXT4_JTR_NONE);
2004 if (err)
2005 goto journal_error;
2006
2007 BUFFER_TRACE(frame->bh, "get_write_access");
2008 err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
2009 EXT4_JTR_NONE);
2010 if (err)
2011 goto journal_error;
2012
2013 data2 = bh2->b_data;
2014
2015 /* create map in the end of data2 block */
2016 map = (struct dx_map_entry *) (data2 + blocksize);
2017 count = dx_make_map(dir, *bh, hinfo, map);
2018 if (count < 0) {
2019 err = count;
2020 goto journal_error;
2021 }
2022 map -= count;
2023 dx_sort_map(map, count);
2024 /* Ensure that neither split block is over half full */
2025 size = 0;
2026 move = 0;
2027 for (i = count-1; i >= 0; i--) {
2028 /* is more than half of this entry in 2nd half of the block? */
2029 if (size + map[i].size/2 > blocksize/2)
2030 break;
2031 size += map[i].size;
2032 move++;
2033 }
2034 /*
2035 * map index at which we will split
2036 *
2037 * If the sum of active entries didn't exceed half the block size, just
2038 * split it in half by count; each resulting block will have at least
2039 * half the space free.
2040 */
2041 if (i > 0)
2042 split = count - move;
2043 else
2044 split = count/2;
2045
2046 hash2 = map[split].hash;
2047 continued = hash2 == map[split - 1].hash;
2048 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2049 (unsigned long)dx_get_block(frame->at),
2050 hash2, split, count-split));
2051
2052 /* Fancy dance to stay within two buffers */
2053 de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2054 blocksize);
2055 de = dx_pack_dirents(dir, data1, blocksize);
2056 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2057 (char *) de,
2058 blocksize);
2059 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2060 (char *) de2,
2061 blocksize);
2062 if (csum_size) {
2063 ext4_initialize_dirent_tail(*bh, blocksize);
2064 ext4_initialize_dirent_tail(bh2, blocksize);
2065 }
2066
2067 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2068 blocksize, 1));
2069 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2070 blocksize, 1));
2071
2072 /* Which block gets the new entry? */
2073 if (hinfo->hash >= hash2) {
2074 swap(*bh, bh2);
2075 de = de2;
2076 }
2077 dx_insert_block(frame, hash2 + continued, newblock);
2078 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2079 if (err)
2080 goto journal_error;
2081 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2082 if (err)
2083 goto journal_error;
2084 brelse(bh2);
2085 dxtrace(dx_show_index("frame", frame->entries));
2086 return de;
2087
2088journal_error:
2089 brelse(*bh);
2090 brelse(bh2);
2091 *bh = NULL;
2092 ext4_std_error(dir->i_sb, err);
2093 return ERR_PTR(err);
2094}
2095
2096int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2097 struct buffer_head *bh,
2098 void *buf, int buf_size,
2099 struct ext4_filename *fname,
2100 struct ext4_dir_entry_2 **dest_de)
2101{
2102 struct ext4_dir_entry_2 *de;
2103 unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2104 int nlen, rlen;
2105 unsigned int offset = 0;
2106 char *top;
2107
2108 de = buf;
2109 top = buf + buf_size - reclen;
2110 while ((char *) de <= top) {
2111 if (ext4_check_dir_entry(dir, NULL, de, bh,
2112 buf, buf_size, offset))
2113 return -EFSCORRUPTED;
2114 if (ext4_match(dir, fname, de))
2115 return -EEXIST;
2116 nlen = ext4_dir_rec_len(de->name_len, dir);
2117 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2118 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2119 break;
2120 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2121 offset += rlen;
2122 }
2123 if ((char *) de > top)
2124 return -ENOSPC;
2125
2126 *dest_de = de;
2127 return 0;
2128}
2129
2130void ext4_insert_dentry(struct inode *dir,
2131 struct inode *inode,
2132 struct ext4_dir_entry_2 *de,
2133 int buf_size,
2134 struct ext4_filename *fname)
2135{
2136
2137 int nlen, rlen;
2138
2139 nlen = ext4_dir_rec_len(de->name_len, dir);
2140 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2141 if (de->inode) {
2142 struct ext4_dir_entry_2 *de1 =
2143 (struct ext4_dir_entry_2 *)((char *)de + nlen);
2144 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2145 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2146 de = de1;
2147 }
2148 de->file_type = EXT4_FT_UNKNOWN;
2149 de->inode = cpu_to_le32(inode->i_ino);
2150 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2151 de->name_len = fname_len(fname);
2152 memcpy(de->name, fname_name(fname), fname_len(fname));
2153 if (ext4_hash_in_dirent(dir)) {
2154 struct dx_hash_info *hinfo = &fname->hinfo;
2155
2156 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2157 EXT4_DIRENT_HASHES(de)->minor_hash =
2158 cpu_to_le32(hinfo->minor_hash);
2159 }
2160}
2161
2162/*
2163 * Add a new entry into a directory (leaf) block. If de is non-NULL,
2164 * it points to a directory entry which is guaranteed to be large
2165 * enough for new directory entry. If de is NULL, then
2166 * add_dirent_to_buf will attempt search the directory block for
2167 * space. It will return -ENOSPC if no space is available, and -EIO
2168 * and -EEXIST if directory entry already exists.
2169 */
2170static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2171 struct inode *dir,
2172 struct inode *inode, struct ext4_dir_entry_2 *de,
2173 struct buffer_head *bh)
2174{
2175 unsigned int blocksize = dir->i_sb->s_blocksize;
2176 int csum_size = 0;
2177 int err, err2;
2178
2179 if (ext4_has_metadata_csum(inode->i_sb))
2180 csum_size = sizeof(struct ext4_dir_entry_tail);
2181
2182 if (!de) {
2183 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2184 blocksize - csum_size, fname, &de);
2185 if (err)
2186 return err;
2187 }
2188 BUFFER_TRACE(bh, "get_write_access");
2189 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2190 EXT4_JTR_NONE);
2191 if (err) {
2192 ext4_std_error(dir->i_sb, err);
2193 return err;
2194 }
2195
2196 /* By now the buffer is marked for journaling */
2197 ext4_insert_dentry(dir, inode, de, blocksize, fname);
2198
2199 /*
2200 * XXX shouldn't update any times until successful
2201 * completion of syscall, but too many callers depend
2202 * on this.
2203 *
2204 * XXX similarly, too many callers depend on
2205 * ext4_new_inode() setting the times, but error
2206 * recovery deletes the inode, so the worst that can
2207 * happen is that the times are slightly out of date
2208 * and/or different from the directory change time.
2209 */
2210 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
2211 ext4_update_dx_flag(dir);
2212 inode_inc_iversion(dir);
2213 err2 = ext4_mark_inode_dirty(handle, dir);
2214 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2215 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2216 if (err)
2217 ext4_std_error(dir->i_sb, err);
2218 return err ? err : err2;
2219}
2220
2221/*
2222 * This converts a one block unindexed directory to a 3 block indexed
2223 * directory, and adds the dentry to the indexed directory.
2224 */
2225static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2226 struct inode *dir,
2227 struct inode *inode, struct buffer_head *bh)
2228{
2229 struct buffer_head *bh2;
2230 struct dx_root *root;
2231 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2232 struct dx_entry *entries;
2233 struct ext4_dir_entry_2 *de, *de2;
2234 char *data2, *top;
2235 unsigned len;
2236 int retval;
2237 unsigned blocksize;
2238 ext4_lblk_t block;
2239 struct fake_dirent *fde;
2240 int csum_size = 0;
2241
2242 if (ext4_has_metadata_csum(inode->i_sb))
2243 csum_size = sizeof(struct ext4_dir_entry_tail);
2244
2245 blocksize = dir->i_sb->s_blocksize;
2246 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2247 BUFFER_TRACE(bh, "get_write_access");
2248 retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2249 EXT4_JTR_NONE);
2250 if (retval) {
2251 ext4_std_error(dir->i_sb, retval);
2252 brelse(bh);
2253 return retval;
2254 }
2255 root = (struct dx_root *) bh->b_data;
2256
2257 /* The 0th block becomes the root, move the dirents out */
2258 fde = &root->dotdot;
2259 de = (struct ext4_dir_entry_2 *)((char *)fde +
2260 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2261 if ((char *) de >= (((char *) root) + blocksize)) {
2262 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2263 brelse(bh);
2264 return -EFSCORRUPTED;
2265 }
2266 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2267
2268 /* Allocate new block for the 0th block's dirents */
2269 bh2 = ext4_append(handle, dir, &block);
2270 if (IS_ERR(bh2)) {
2271 brelse(bh);
2272 return PTR_ERR(bh2);
2273 }
2274 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2275 data2 = bh2->b_data;
2276
2277 memcpy(data2, de, len);
2278 memset(de, 0, len); /* wipe old data */
2279 de = (struct ext4_dir_entry_2 *) data2;
2280 top = data2 + len;
2281 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2282 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2283 (char *)de - data2)) {
2284 brelse(bh2);
2285 brelse(bh);
2286 return -EFSCORRUPTED;
2287 }
2288 de = de2;
2289 }
2290 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2291 (char *) de, blocksize);
2292
2293 if (csum_size)
2294 ext4_initialize_dirent_tail(bh2, blocksize);
2295
2296 /* Initialize the root; the dot dirents already exist */
2297 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2298 de->rec_len = ext4_rec_len_to_disk(
2299 blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2300 memset (&root->info, 0, sizeof(root->info));
2301 root->info.info_length = sizeof(root->info);
2302 if (ext4_hash_in_dirent(dir))
2303 root->info.hash_version = DX_HASH_SIPHASH;
2304 else
2305 root->info.hash_version =
2306 EXT4_SB(dir->i_sb)->s_def_hash_version;
2307
2308 entries = root->entries;
2309 dx_set_block(entries, 1);
2310 dx_set_count(entries, 1);
2311 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2312
2313 /* Initialize as for dx_probe */
2314 fname->hinfo.hash_version = root->info.hash_version;
2315 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2316 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2317 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2318
2319 /* casefolded encrypted hashes are computed on fname setup */
2320 if (!ext4_hash_in_dirent(dir)) {
2321 int err = ext4fs_dirhash(dir, fname_name(fname),
2322 fname_len(fname), &fname->hinfo);
2323 if (err < 0) {
2324 brelse(bh2);
2325 brelse(bh);
2326 return err;
2327 }
2328 }
2329 memset(frames, 0, sizeof(frames));
2330 frame = frames;
2331 frame->entries = entries;
2332 frame->at = entries;
2333 frame->bh = bh;
2334
2335 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2336 if (retval)
2337 goto out_frames;
2338 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2339 if (retval)
2340 goto out_frames;
2341
2342 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2343 if (IS_ERR(de)) {
2344 retval = PTR_ERR(de);
2345 goto out_frames;
2346 }
2347
2348 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2349out_frames:
2350 /*
2351 * Even if the block split failed, we have to properly write
2352 * out all the changes we did so far. Otherwise we can end up
2353 * with corrupted filesystem.
2354 */
2355 if (retval)
2356 ext4_mark_inode_dirty(handle, dir);
2357 dx_release(frames);
2358 brelse(bh2);
2359 return retval;
2360}
2361
2362/*
2363 * ext4_add_entry()
2364 *
2365 * adds a file entry to the specified directory, using the same
2366 * semantics as ext4_find_entry(). It returns NULL if it failed.
2367 *
2368 * NOTE!! The inode part of 'de' is left at 0 - which means you
2369 * may not sleep between calling this and putting something into
2370 * the entry, as someone else might have used it while you slept.
2371 */
2372static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2373 struct inode *inode)
2374{
2375 struct inode *dir = d_inode(dentry->d_parent);
2376 struct buffer_head *bh = NULL;
2377 struct ext4_dir_entry_2 *de;
2378 struct super_block *sb;
2379 struct ext4_filename fname;
2380 int retval;
2381 int dx_fallback=0;
2382 unsigned blocksize;
2383 ext4_lblk_t block, blocks;
2384 int csum_size = 0;
2385
2386 if (ext4_has_metadata_csum(inode->i_sb))
2387 csum_size = sizeof(struct ext4_dir_entry_tail);
2388
2389 sb = dir->i_sb;
2390 blocksize = sb->s_blocksize;
2391
2392 if (fscrypt_is_nokey_name(dentry))
2393 return -ENOKEY;
2394
2395#if IS_ENABLED(CONFIG_UNICODE)
2396 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2397 utf8_validate(sb->s_encoding, &dentry->d_name))
2398 return -EINVAL;
2399#endif
2400
2401 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2402 if (retval)
2403 return retval;
2404
2405 if (ext4_has_inline_data(dir)) {
2406 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2407 if (retval < 0)
2408 goto out;
2409 if (retval == 1) {
2410 retval = 0;
2411 goto out;
2412 }
2413 }
2414
2415 if (is_dx(dir)) {
2416 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2417 if (!retval || (retval != ERR_BAD_DX_DIR))
2418 goto out;
2419 /* Can we just ignore htree data? */
2420 if (ext4_has_metadata_csum(sb)) {
2421 EXT4_ERROR_INODE(dir,
2422 "Directory has corrupted htree index.");
2423 retval = -EFSCORRUPTED;
2424 goto out;
2425 }
2426 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2427 dx_fallback++;
2428 retval = ext4_mark_inode_dirty(handle, dir);
2429 if (unlikely(retval))
2430 goto out;
2431 }
2432 blocks = dir->i_size >> sb->s_blocksize_bits;
2433 for (block = 0; block < blocks; block++) {
2434 bh = ext4_read_dirblock(dir, block, DIRENT);
2435 if (bh == NULL) {
2436 bh = ext4_bread(handle, dir, block,
2437 EXT4_GET_BLOCKS_CREATE);
2438 goto add_to_new_block;
2439 }
2440 if (IS_ERR(bh)) {
2441 retval = PTR_ERR(bh);
2442 bh = NULL;
2443 goto out;
2444 }
2445 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2446 NULL, bh);
2447 if (retval != -ENOSPC)
2448 goto out;
2449
2450 if (blocks == 1 && !dx_fallback &&
2451 ext4_has_feature_dir_index(sb)) {
2452 retval = make_indexed_dir(handle, &fname, dir,
2453 inode, bh);
2454 bh = NULL; /* make_indexed_dir releases bh */
2455 goto out;
2456 }
2457 brelse(bh);
2458 }
2459 bh = ext4_append(handle, dir, &block);
2460add_to_new_block:
2461 if (IS_ERR(bh)) {
2462 retval = PTR_ERR(bh);
2463 bh = NULL;
2464 goto out;
2465 }
2466 de = (struct ext4_dir_entry_2 *) bh->b_data;
2467 de->inode = 0;
2468 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2469
2470 if (csum_size)
2471 ext4_initialize_dirent_tail(bh, blocksize);
2472
2473 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2474out:
2475 ext4_fname_free_filename(&fname);
2476 brelse(bh);
2477 if (retval == 0)
2478 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2479 return retval;
2480}
2481
2482/*
2483 * Returns 0 for success, or a negative error value
2484 */
2485static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2486 struct inode *dir, struct inode *inode)
2487{
2488 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2489 struct dx_entry *entries, *at;
2490 struct buffer_head *bh;
2491 struct super_block *sb = dir->i_sb;
2492 struct ext4_dir_entry_2 *de;
2493 int restart;
2494 int err;
2495
2496again:
2497 restart = 0;
2498 frame = dx_probe(fname, dir, NULL, frames);
2499 if (IS_ERR(frame))
2500 return PTR_ERR(frame);
2501 entries = frame->entries;
2502 at = frame->at;
2503 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2504 if (IS_ERR(bh)) {
2505 err = PTR_ERR(bh);
2506 bh = NULL;
2507 goto cleanup;
2508 }
2509
2510 BUFFER_TRACE(bh, "get_write_access");
2511 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2512 if (err)
2513 goto journal_error;
2514
2515 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2516 if (err != -ENOSPC)
2517 goto cleanup;
2518
2519 err = 0;
2520 /* Block full, should compress but for now just split */
2521 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2522 dx_get_count(entries), dx_get_limit(entries)));
2523 /* Need to split index? */
2524 if (dx_get_count(entries) == dx_get_limit(entries)) {
2525 ext4_lblk_t newblock;
2526 int levels = frame - frames + 1;
2527 unsigned int icount;
2528 int add_level = 1;
2529 struct dx_entry *entries2;
2530 struct dx_node *node2;
2531 struct buffer_head *bh2;
2532
2533 while (frame > frames) {
2534 if (dx_get_count((frame - 1)->entries) <
2535 dx_get_limit((frame - 1)->entries)) {
2536 add_level = 0;
2537 break;
2538 }
2539 frame--; /* split higher index block */
2540 at = frame->at;
2541 entries = frame->entries;
2542 restart = 1;
2543 }
2544 if (add_level && levels == ext4_dir_htree_level(sb)) {
2545 ext4_warning(sb, "Directory (ino: %lu) index full, "
2546 "reach max htree level :%d",
2547 dir->i_ino, levels);
2548 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2549 ext4_warning(sb, "Large directory feature is "
2550 "not enabled on this "
2551 "filesystem");
2552 }
2553 err = -ENOSPC;
2554 goto cleanup;
2555 }
2556 icount = dx_get_count(entries);
2557 bh2 = ext4_append(handle, dir, &newblock);
2558 if (IS_ERR(bh2)) {
2559 err = PTR_ERR(bh2);
2560 goto cleanup;
2561 }
2562 node2 = (struct dx_node *)(bh2->b_data);
2563 entries2 = node2->entries;
2564 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2565 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2566 sb->s_blocksize);
2567 BUFFER_TRACE(frame->bh, "get_write_access");
2568 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2569 EXT4_JTR_NONE);
2570 if (err)
2571 goto journal_error;
2572 if (!add_level) {
2573 unsigned icount1 = icount/2, icount2 = icount - icount1;
2574 unsigned hash2 = dx_get_hash(entries + icount1);
2575 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2576 icount1, icount2));
2577
2578 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2579 err = ext4_journal_get_write_access(handle, sb,
2580 (frame - 1)->bh,
2581 EXT4_JTR_NONE);
2582 if (err)
2583 goto journal_error;
2584
2585 memcpy((char *) entries2, (char *) (entries + icount1),
2586 icount2 * sizeof(struct dx_entry));
2587 dx_set_count(entries, icount1);
2588 dx_set_count(entries2, icount2);
2589 dx_set_limit(entries2, dx_node_limit(dir));
2590
2591 /* Which index block gets the new entry? */
2592 if (at - entries >= icount1) {
2593 frame->at = at - entries - icount1 + entries2;
2594 frame->entries = entries = entries2;
2595 swap(frame->bh, bh2);
2596 }
2597 dx_insert_block((frame - 1), hash2, newblock);
2598 dxtrace(dx_show_index("node", frame->entries));
2599 dxtrace(dx_show_index("node",
2600 ((struct dx_node *) bh2->b_data)->entries));
2601 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2602 if (err)
2603 goto journal_error;
2604 brelse (bh2);
2605 err = ext4_handle_dirty_dx_node(handle, dir,
2606 (frame - 1)->bh);
2607 if (err)
2608 goto journal_error;
2609 err = ext4_handle_dirty_dx_node(handle, dir,
2610 frame->bh);
2611 if (restart || err)
2612 goto journal_error;
2613 } else {
2614 struct dx_root *dxroot;
2615 memcpy((char *) entries2, (char *) entries,
2616 icount * sizeof(struct dx_entry));
2617 dx_set_limit(entries2, dx_node_limit(dir));
2618
2619 /* Set up root */
2620 dx_set_count(entries, 1);
2621 dx_set_block(entries + 0, newblock);
2622 dxroot = (struct dx_root *)frames[0].bh->b_data;
2623 dxroot->info.indirect_levels += 1;
2624 dxtrace(printk(KERN_DEBUG
2625 "Creating %d level index...\n",
2626 dxroot->info.indirect_levels));
2627 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2628 if (err)
2629 goto journal_error;
2630 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2631 brelse(bh2);
2632 restart = 1;
2633 goto journal_error;
2634 }
2635 }
2636 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2637 if (IS_ERR(de)) {
2638 err = PTR_ERR(de);
2639 goto cleanup;
2640 }
2641 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2642 goto cleanup;
2643
2644journal_error:
2645 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2646cleanup:
2647 brelse(bh);
2648 dx_release(frames);
2649 /* @restart is true means htree-path has been changed, we need to
2650 * repeat dx_probe() to find out valid htree-path
2651 */
2652 if (restart && err == 0)
2653 goto again;
2654 return err;
2655}
2656
2657/*
2658 * ext4_generic_delete_entry deletes a directory entry by merging it
2659 * with the previous entry
2660 */
2661int ext4_generic_delete_entry(struct inode *dir,
2662 struct ext4_dir_entry_2 *de_del,
2663 struct buffer_head *bh,
2664 void *entry_buf,
2665 int buf_size,
2666 int csum_size)
2667{
2668 struct ext4_dir_entry_2 *de, *pde;
2669 unsigned int blocksize = dir->i_sb->s_blocksize;
2670 int i;
2671
2672 i = 0;
2673 pde = NULL;
2674 de = entry_buf;
2675 while (i < buf_size - csum_size) {
2676 if (ext4_check_dir_entry(dir, NULL, de, bh,
2677 entry_buf, buf_size, i))
2678 return -EFSCORRUPTED;
2679 if (de == de_del) {
2680 if (pde) {
2681 pde->rec_len = ext4_rec_len_to_disk(
2682 ext4_rec_len_from_disk(pde->rec_len,
2683 blocksize) +
2684 ext4_rec_len_from_disk(de->rec_len,
2685 blocksize),
2686 blocksize);
2687
2688 /* wipe entire dir_entry */
2689 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2690 blocksize));
2691 } else {
2692 /* wipe dir_entry excluding the rec_len field */
2693 de->inode = 0;
2694 memset(&de->name_len, 0,
2695 ext4_rec_len_from_disk(de->rec_len,
2696 blocksize) -
2697 offsetof(struct ext4_dir_entry_2,
2698 name_len));
2699 }
2700
2701 inode_inc_iversion(dir);
2702 return 0;
2703 }
2704 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2705 pde = de;
2706 de = ext4_next_entry(de, blocksize);
2707 }
2708 return -ENOENT;
2709}
2710
2711static int ext4_delete_entry(handle_t *handle,
2712 struct inode *dir,
2713 struct ext4_dir_entry_2 *de_del,
2714 struct buffer_head *bh)
2715{
2716 int err, csum_size = 0;
2717
2718 if (ext4_has_inline_data(dir)) {
2719 int has_inline_data = 1;
2720 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2721 &has_inline_data);
2722 if (has_inline_data)
2723 return err;
2724 }
2725
2726 if (ext4_has_metadata_csum(dir->i_sb))
2727 csum_size = sizeof(struct ext4_dir_entry_tail);
2728
2729 BUFFER_TRACE(bh, "get_write_access");
2730 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2731 EXT4_JTR_NONE);
2732 if (unlikely(err))
2733 goto out;
2734
2735 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2736 dir->i_sb->s_blocksize, csum_size);
2737 if (err)
2738 goto out;
2739
2740 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2741 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2742 if (unlikely(err))
2743 goto out;
2744
2745 return 0;
2746out:
2747 if (err != -ENOENT)
2748 ext4_std_error(dir->i_sb, err);
2749 return err;
2750}
2751
2752/*
2753 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2754 * since this indicates that nlinks count was previously 1 to avoid overflowing
2755 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2756 * that subdirectory link counts are not being maintained accurately.
2757 *
2758 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2759 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2760 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2761 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2762 */
2763static void ext4_inc_count(struct inode *inode)
2764{
2765 inc_nlink(inode);
2766 if (is_dx(inode) &&
2767 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2768 set_nlink(inode, 1);
2769}
2770
2771/*
2772 * If a directory had nlink == 1, then we should let it be 1. This indicates
2773 * directory has >EXT4_LINK_MAX subdirs.
2774 */
2775static void ext4_dec_count(struct inode *inode)
2776{
2777 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2778 drop_nlink(inode);
2779}
2780
2781
2782/*
2783 * Add non-directory inode to a directory. On success, the inode reference is
2784 * consumed by dentry is instantiation. This is also indicated by clearing of
2785 * *inodep pointer. On failure, the caller is responsible for dropping the
2786 * inode reference in the safe context.
2787 */
2788static int ext4_add_nondir(handle_t *handle,
2789 struct dentry *dentry, struct inode **inodep)
2790{
2791 struct inode *dir = d_inode(dentry->d_parent);
2792 struct inode *inode = *inodep;
2793 int err = ext4_add_entry(handle, dentry, inode);
2794 if (!err) {
2795 err = ext4_mark_inode_dirty(handle, inode);
2796 if (IS_DIRSYNC(dir))
2797 ext4_handle_sync(handle);
2798 d_instantiate_new(dentry, inode);
2799 *inodep = NULL;
2800 return err;
2801 }
2802 drop_nlink(inode);
2803 ext4_mark_inode_dirty(handle, inode);
2804 ext4_orphan_add(handle, inode);
2805 unlock_new_inode(inode);
2806 return err;
2807}
2808
2809/*
2810 * By the time this is called, we already have created
2811 * the directory cache entry for the new file, but it
2812 * is so far negative - it has no inode.
2813 *
2814 * If the create succeeds, we fill in the inode information
2815 * with d_instantiate().
2816 */
2817static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2818 struct dentry *dentry, umode_t mode, bool excl)
2819{
2820 handle_t *handle;
2821 struct inode *inode;
2822 int err, credits, retries = 0;
2823
2824 err = dquot_initialize(dir);
2825 if (err)
2826 return err;
2827
2828 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2829 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2830retry:
2831 inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2832 0, NULL, EXT4_HT_DIR, credits);
2833 handle = ext4_journal_current_handle();
2834 err = PTR_ERR(inode);
2835 if (!IS_ERR(inode)) {
2836 inode->i_op = &ext4_file_inode_operations;
2837 inode->i_fop = &ext4_file_operations;
2838 ext4_set_aops(inode);
2839 err = ext4_add_nondir(handle, dentry, &inode);
2840 if (!err)
2841 ext4_fc_track_create(handle, dentry);
2842 }
2843 if (handle)
2844 ext4_journal_stop(handle);
2845 if (!IS_ERR_OR_NULL(inode))
2846 iput(inode);
2847 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2848 goto retry;
2849 return err;
2850}
2851
2852static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2853 struct dentry *dentry, umode_t mode, dev_t rdev)
2854{
2855 handle_t *handle;
2856 struct inode *inode;
2857 int err, credits, retries = 0;
2858
2859 err = dquot_initialize(dir);
2860 if (err)
2861 return err;
2862
2863 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2864 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2865retry:
2866 inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2867 0, NULL, EXT4_HT_DIR, credits);
2868 handle = ext4_journal_current_handle();
2869 err = PTR_ERR(inode);
2870 if (!IS_ERR(inode)) {
2871 init_special_inode(inode, inode->i_mode, rdev);
2872 inode->i_op = &ext4_special_inode_operations;
2873 err = ext4_add_nondir(handle, dentry, &inode);
2874 if (!err)
2875 ext4_fc_track_create(handle, dentry);
2876 }
2877 if (handle)
2878 ext4_journal_stop(handle);
2879 if (!IS_ERR_OR_NULL(inode))
2880 iput(inode);
2881 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2882 goto retry;
2883 return err;
2884}
2885
2886static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2887 struct file *file, umode_t mode)
2888{
2889 handle_t *handle;
2890 struct inode *inode;
2891 int err, retries = 0;
2892
2893 err = dquot_initialize(dir);
2894 if (err)
2895 return err;
2896
2897retry:
2898 inode = ext4_new_inode_start_handle(idmap, dir, mode,
2899 NULL, 0, NULL,
2900 EXT4_HT_DIR,
2901 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2902 4 + EXT4_XATTR_TRANS_BLOCKS);
2903 handle = ext4_journal_current_handle();
2904 err = PTR_ERR(inode);
2905 if (!IS_ERR(inode)) {
2906 inode->i_op = &ext4_file_inode_operations;
2907 inode->i_fop = &ext4_file_operations;
2908 ext4_set_aops(inode);
2909 d_tmpfile(file, inode);
2910 err = ext4_orphan_add(handle, inode);
2911 if (err)
2912 goto err_unlock_inode;
2913 mark_inode_dirty(inode);
2914 unlock_new_inode(inode);
2915 }
2916 if (handle)
2917 ext4_journal_stop(handle);
2918 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2919 goto retry;
2920 return finish_open_simple(file, err);
2921err_unlock_inode:
2922 ext4_journal_stop(handle);
2923 unlock_new_inode(inode);
2924 return err;
2925}
2926
2927struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2928 struct ext4_dir_entry_2 *de,
2929 int blocksize, int csum_size,
2930 unsigned int parent_ino, int dotdot_real_len)
2931{
2932 de->inode = cpu_to_le32(inode->i_ino);
2933 de->name_len = 1;
2934 de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2935 blocksize);
2936 strcpy(de->name, ".");
2937 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2938
2939 de = ext4_next_entry(de, blocksize);
2940 de->inode = cpu_to_le32(parent_ino);
2941 de->name_len = 2;
2942 if (!dotdot_real_len)
2943 de->rec_len = ext4_rec_len_to_disk(blocksize -
2944 (csum_size + ext4_dir_rec_len(1, NULL)),
2945 blocksize);
2946 else
2947 de->rec_len = ext4_rec_len_to_disk(
2948 ext4_dir_rec_len(de->name_len, NULL),
2949 blocksize);
2950 strcpy(de->name, "..");
2951 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2952
2953 return ext4_next_entry(de, blocksize);
2954}
2955
2956int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2957 struct inode *inode)
2958{
2959 struct buffer_head *dir_block = NULL;
2960 struct ext4_dir_entry_2 *de;
2961 ext4_lblk_t block = 0;
2962 unsigned int blocksize = dir->i_sb->s_blocksize;
2963 int csum_size = 0;
2964 int err;
2965
2966 if (ext4_has_metadata_csum(dir->i_sb))
2967 csum_size = sizeof(struct ext4_dir_entry_tail);
2968
2969 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2970 err = ext4_try_create_inline_dir(handle, dir, inode);
2971 if (err < 0 && err != -ENOSPC)
2972 goto out;
2973 if (!err)
2974 goto out;
2975 }
2976
2977 inode->i_size = 0;
2978 dir_block = ext4_append(handle, inode, &block);
2979 if (IS_ERR(dir_block))
2980 return PTR_ERR(dir_block);
2981 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2982 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2983 set_nlink(inode, 2);
2984 if (csum_size)
2985 ext4_initialize_dirent_tail(dir_block, blocksize);
2986
2987 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2988 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2989 if (err)
2990 goto out;
2991 set_buffer_verified(dir_block);
2992out:
2993 brelse(dir_block);
2994 return err;
2995}
2996
2997static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2998 struct dentry *dentry, umode_t mode)
2999{
3000 handle_t *handle;
3001 struct inode *inode;
3002 int err, err2 = 0, credits, retries = 0;
3003
3004 if (EXT4_DIR_LINK_MAX(dir))
3005 return -EMLINK;
3006
3007 err = dquot_initialize(dir);
3008 if (err)
3009 return err;
3010
3011 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3012 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3013retry:
3014 inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3015 &dentry->d_name,
3016 0, NULL, EXT4_HT_DIR, credits);
3017 handle = ext4_journal_current_handle();
3018 err = PTR_ERR(inode);
3019 if (IS_ERR(inode))
3020 goto out_stop;
3021
3022 inode->i_op = &ext4_dir_inode_operations;
3023 inode->i_fop = &ext4_dir_operations;
3024 err = ext4_init_new_dir(handle, dir, inode);
3025 if (err)
3026 goto out_clear_inode;
3027 err = ext4_mark_inode_dirty(handle, inode);
3028 if (!err)
3029 err = ext4_add_entry(handle, dentry, inode);
3030 if (err) {
3031out_clear_inode:
3032 clear_nlink(inode);
3033 ext4_orphan_add(handle, inode);
3034 unlock_new_inode(inode);
3035 err2 = ext4_mark_inode_dirty(handle, inode);
3036 if (unlikely(err2))
3037 err = err2;
3038 ext4_journal_stop(handle);
3039 iput(inode);
3040 goto out_retry;
3041 }
3042 ext4_inc_count(dir);
3043
3044 ext4_update_dx_flag(dir);
3045 err = ext4_mark_inode_dirty(handle, dir);
3046 if (err)
3047 goto out_clear_inode;
3048 d_instantiate_new(dentry, inode);
3049 ext4_fc_track_create(handle, dentry);
3050 if (IS_DIRSYNC(dir))
3051 ext4_handle_sync(handle);
3052
3053out_stop:
3054 if (handle)
3055 ext4_journal_stop(handle);
3056out_retry:
3057 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3058 goto retry;
3059 return err;
3060}
3061
3062/*
3063 * routine to check that the specified directory is empty (for rmdir)
3064 */
3065bool ext4_empty_dir(struct inode *inode)
3066{
3067 unsigned int offset;
3068 struct buffer_head *bh;
3069 struct ext4_dir_entry_2 *de;
3070 struct super_block *sb;
3071
3072 if (ext4_has_inline_data(inode)) {
3073 int has_inline_data = 1;
3074 int ret;
3075
3076 ret = empty_inline_dir(inode, &has_inline_data);
3077 if (has_inline_data)
3078 return ret;
3079 }
3080
3081 sb = inode->i_sb;
3082 if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3083 ext4_dir_rec_len(2, NULL)) {
3084 EXT4_ERROR_INODE(inode, "invalid size");
3085 return false;
3086 }
3087 /* The first directory block must not be a hole,
3088 * so treat it as DIRENT_HTREE
3089 */
3090 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3091 if (IS_ERR(bh))
3092 return false;
3093
3094 de = (struct ext4_dir_entry_2 *) bh->b_data;
3095 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3096 0) ||
3097 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3098 ext4_warning_inode(inode, "directory missing '.'");
3099 brelse(bh);
3100 return false;
3101 }
3102 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3103 de = ext4_next_entry(de, sb->s_blocksize);
3104 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3105 offset) ||
3106 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3107 ext4_warning_inode(inode, "directory missing '..'");
3108 brelse(bh);
3109 return false;
3110 }
3111 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3112 while (offset < inode->i_size) {
3113 if (!(offset & (sb->s_blocksize - 1))) {
3114 unsigned int lblock;
3115 brelse(bh);
3116 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3117 bh = ext4_read_dirblock(inode, lblock, EITHER);
3118 if (bh == NULL) {
3119 offset += sb->s_blocksize;
3120 continue;
3121 }
3122 if (IS_ERR(bh))
3123 return false;
3124 }
3125 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3126 (offset & (sb->s_blocksize - 1)));
3127 if (ext4_check_dir_entry(inode, NULL, de, bh,
3128 bh->b_data, bh->b_size, offset) ||
3129 le32_to_cpu(de->inode)) {
3130 brelse(bh);
3131 return false;
3132 }
3133 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3134 }
3135 brelse(bh);
3136 return true;
3137}
3138
3139static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3140{
3141 int retval;
3142 struct inode *inode;
3143 struct buffer_head *bh;
3144 struct ext4_dir_entry_2 *de;
3145 handle_t *handle = NULL;
3146
3147 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3148 return -EIO;
3149
3150 /* Initialize quotas before so that eventual writes go in
3151 * separate transaction */
3152 retval = dquot_initialize(dir);
3153 if (retval)
3154 return retval;
3155 retval = dquot_initialize(d_inode(dentry));
3156 if (retval)
3157 return retval;
3158
3159 retval = -ENOENT;
3160 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3161 if (IS_ERR(bh))
3162 return PTR_ERR(bh);
3163 if (!bh)
3164 goto end_rmdir;
3165
3166 inode = d_inode(dentry);
3167
3168 retval = -EFSCORRUPTED;
3169 if (le32_to_cpu(de->inode) != inode->i_ino)
3170 goto end_rmdir;
3171
3172 retval = -ENOTEMPTY;
3173 if (!ext4_empty_dir(inode))
3174 goto end_rmdir;
3175
3176 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3177 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3178 if (IS_ERR(handle)) {
3179 retval = PTR_ERR(handle);
3180 handle = NULL;
3181 goto end_rmdir;
3182 }
3183
3184 if (IS_DIRSYNC(dir))
3185 ext4_handle_sync(handle);
3186
3187 retval = ext4_delete_entry(handle, dir, de, bh);
3188 if (retval)
3189 goto end_rmdir;
3190 if (!EXT4_DIR_LINK_EMPTY(inode))
3191 ext4_warning_inode(inode,
3192 "empty directory '%.*s' has too many links (%u)",
3193 dentry->d_name.len, dentry->d_name.name,
3194 inode->i_nlink);
3195 inode_inc_iversion(inode);
3196 clear_nlink(inode);
3197 /* There's no need to set i_disksize: the fact that i_nlink is
3198 * zero will ensure that the right thing happens during any
3199 * recovery. */
3200 inode->i_size = 0;
3201 ext4_orphan_add(handle, inode);
3202 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3203 inode_set_ctime_current(inode);
3204 retval = ext4_mark_inode_dirty(handle, inode);
3205 if (retval)
3206 goto end_rmdir;
3207 ext4_dec_count(dir);
3208 ext4_update_dx_flag(dir);
3209 ext4_fc_track_unlink(handle, dentry);
3210 retval = ext4_mark_inode_dirty(handle, dir);
3211
3212#if IS_ENABLED(CONFIG_UNICODE)
3213 /* VFS negative dentries are incompatible with Encoding and
3214 * Case-insensitiveness. Eventually we'll want avoid
3215 * invalidating the dentries here, alongside with returning the
3216 * negative dentries at ext4_lookup(), when it is better
3217 * supported by the VFS for the CI case.
3218 */
3219 if (IS_CASEFOLDED(dir))
3220 d_invalidate(dentry);
3221#endif
3222
3223end_rmdir:
3224 brelse(bh);
3225 if (handle)
3226 ext4_journal_stop(handle);
3227 return retval;
3228}
3229
3230int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3231 struct inode *inode,
3232 struct dentry *dentry /* NULL during fast_commit recovery */)
3233{
3234 int retval = -ENOENT;
3235 struct buffer_head *bh;
3236 struct ext4_dir_entry_2 *de;
3237 handle_t *handle;
3238 int skip_remove_dentry = 0;
3239
3240 /*
3241 * Keep this outside the transaction; it may have to set up the
3242 * directory's encryption key, which isn't GFP_NOFS-safe.
3243 */
3244 bh = ext4_find_entry(dir, d_name, &de, NULL);
3245 if (IS_ERR(bh))
3246 return PTR_ERR(bh);
3247
3248 if (!bh)
3249 return -ENOENT;
3250
3251 if (le32_to_cpu(de->inode) != inode->i_ino) {
3252 /*
3253 * It's okay if we find dont find dentry which matches
3254 * the inode. That's because it might have gotten
3255 * renamed to a different inode number
3256 */
3257 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3258 skip_remove_dentry = 1;
3259 else
3260 goto out_bh;
3261 }
3262
3263 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3264 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3265 if (IS_ERR(handle)) {
3266 retval = PTR_ERR(handle);
3267 goto out_bh;
3268 }
3269
3270 if (IS_DIRSYNC(dir))
3271 ext4_handle_sync(handle);
3272
3273 if (!skip_remove_dentry) {
3274 retval = ext4_delete_entry(handle, dir, de, bh);
3275 if (retval)
3276 goto out_handle;
3277 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3278 ext4_update_dx_flag(dir);
3279 retval = ext4_mark_inode_dirty(handle, dir);
3280 if (retval)
3281 goto out_handle;
3282 } else {
3283 retval = 0;
3284 }
3285 if (inode->i_nlink == 0)
3286 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3287 d_name->len, d_name->name);
3288 else
3289 drop_nlink(inode);
3290 if (!inode->i_nlink)
3291 ext4_orphan_add(handle, inode);
3292 inode_set_ctime_current(inode);
3293 retval = ext4_mark_inode_dirty(handle, inode);
3294 if (dentry && !retval)
3295 ext4_fc_track_unlink(handle, dentry);
3296out_handle:
3297 ext4_journal_stop(handle);
3298out_bh:
3299 brelse(bh);
3300 return retval;
3301}
3302
3303static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3304{
3305 int retval;
3306
3307 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3308 return -EIO;
3309
3310 trace_ext4_unlink_enter(dir, dentry);
3311 /*
3312 * Initialize quotas before so that eventual writes go
3313 * in separate transaction
3314 */
3315 retval = dquot_initialize(dir);
3316 if (retval)
3317 goto out_trace;
3318 retval = dquot_initialize(d_inode(dentry));
3319 if (retval)
3320 goto out_trace;
3321
3322 retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3323#if IS_ENABLED(CONFIG_UNICODE)
3324 /* VFS negative dentries are incompatible with Encoding and
3325 * Case-insensitiveness. Eventually we'll want avoid
3326 * invalidating the dentries here, alongside with returning the
3327 * negative dentries at ext4_lookup(), when it is better
3328 * supported by the VFS for the CI case.
3329 */
3330 if (IS_CASEFOLDED(dir))
3331 d_invalidate(dentry);
3332#endif
3333
3334out_trace:
3335 trace_ext4_unlink_exit(dentry, retval);
3336 return retval;
3337}
3338
3339static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3340 struct fscrypt_str *disk_link)
3341{
3342 struct buffer_head *bh;
3343 char *kaddr;
3344 int err = 0;
3345
3346 bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3347 if (IS_ERR(bh))
3348 return PTR_ERR(bh);
3349
3350 BUFFER_TRACE(bh, "get_write_access");
3351 err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3352 if (err)
3353 goto out;
3354
3355 kaddr = (char *)bh->b_data;
3356 memcpy(kaddr, disk_link->name, disk_link->len);
3357 inode->i_size = disk_link->len - 1;
3358 EXT4_I(inode)->i_disksize = inode->i_size;
3359 err = ext4_handle_dirty_metadata(handle, inode, bh);
3360out:
3361 brelse(bh);
3362 return err;
3363}
3364
3365static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3366 struct dentry *dentry, const char *symname)
3367{
3368 handle_t *handle;
3369 struct inode *inode;
3370 int err, len = strlen(symname);
3371 int credits;
3372 struct fscrypt_str disk_link;
3373 int retries = 0;
3374
3375 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3376 return -EIO;
3377
3378 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3379 &disk_link);
3380 if (err)
3381 return err;
3382
3383 err = dquot_initialize(dir);
3384 if (err)
3385 return err;
3386
3387 /*
3388 * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3389 * directory. +3 for inode, inode bitmap, group descriptor allocation.
3390 * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3391 * modification.
3392 */
3393 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3394 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3395retry:
3396 inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3397 &dentry->d_name, 0, NULL,
3398 EXT4_HT_DIR, credits);
3399 handle = ext4_journal_current_handle();
3400 if (IS_ERR(inode)) {
3401 if (handle)
3402 ext4_journal_stop(handle);
3403 err = PTR_ERR(inode);
3404 goto out_retry;
3405 }
3406
3407 if (IS_ENCRYPTED(inode)) {
3408 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3409 if (err)
3410 goto err_drop_inode;
3411 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3412 } else {
3413 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3414 inode->i_op = &ext4_symlink_inode_operations;
3415 } else {
3416 inode->i_op = &ext4_fast_symlink_inode_operations;
3417 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3418 }
3419 }
3420
3421 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3422 /* alloc symlink block and fill it */
3423 err = ext4_init_symlink_block(handle, inode, &disk_link);
3424 if (err)
3425 goto err_drop_inode;
3426 } else {
3427 /* clear the extent format for fast symlink */
3428 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3429 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3430 disk_link.len);
3431 inode->i_size = disk_link.len - 1;
3432 EXT4_I(inode)->i_disksize = inode->i_size;
3433 }
3434 err = ext4_add_nondir(handle, dentry, &inode);
3435 if (handle)
3436 ext4_journal_stop(handle);
3437 iput(inode);
3438 goto out_retry;
3439
3440err_drop_inode:
3441 clear_nlink(inode);
3442 ext4_mark_inode_dirty(handle, inode);
3443 ext4_orphan_add(handle, inode);
3444 unlock_new_inode(inode);
3445 if (handle)
3446 ext4_journal_stop(handle);
3447 iput(inode);
3448out_retry:
3449 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3450 goto retry;
3451 if (disk_link.name != (unsigned char *)symname)
3452 kfree(disk_link.name);
3453 return err;
3454}
3455
3456int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3457{
3458 handle_t *handle;
3459 int err, retries = 0;
3460retry:
3461 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3462 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3463 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3464 if (IS_ERR(handle))
3465 return PTR_ERR(handle);
3466
3467 if (IS_DIRSYNC(dir))
3468 ext4_handle_sync(handle);
3469
3470 inode_set_ctime_current(inode);
3471 ext4_inc_count(inode);
3472 ihold(inode);
3473
3474 err = ext4_add_entry(handle, dentry, inode);
3475 if (!err) {
3476 err = ext4_mark_inode_dirty(handle, inode);
3477 /* this can happen only for tmpfile being
3478 * linked the first time
3479 */
3480 if (inode->i_nlink == 1)
3481 ext4_orphan_del(handle, inode);
3482 d_instantiate(dentry, inode);
3483 ext4_fc_track_link(handle, dentry);
3484 } else {
3485 drop_nlink(inode);
3486 iput(inode);
3487 }
3488 ext4_journal_stop(handle);
3489 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3490 goto retry;
3491 return err;
3492}
3493
3494static int ext4_link(struct dentry *old_dentry,
3495 struct inode *dir, struct dentry *dentry)
3496{
3497 struct inode *inode = d_inode(old_dentry);
3498 int err;
3499
3500 if (inode->i_nlink >= EXT4_LINK_MAX)
3501 return -EMLINK;
3502
3503 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3504 if (err)
3505 return err;
3506
3507 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3508 (!projid_eq(EXT4_I(dir)->i_projid,
3509 EXT4_I(old_dentry->d_inode)->i_projid)))
3510 return -EXDEV;
3511
3512 err = dquot_initialize(dir);
3513 if (err)
3514 return err;
3515 return __ext4_link(dir, inode, dentry);
3516}
3517
3518/*
3519 * Try to find buffer head where contains the parent block.
3520 * It should be the inode block if it is inlined or the 1st block
3521 * if it is a normal dir.
3522 */
3523static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3524 struct inode *inode,
3525 int *retval,
3526 struct ext4_dir_entry_2 **parent_de,
3527 int *inlined)
3528{
3529 struct buffer_head *bh;
3530
3531 if (!ext4_has_inline_data(inode)) {
3532 struct ext4_dir_entry_2 *de;
3533 unsigned int offset;
3534
3535 /* The first directory block must not be a hole, so
3536 * treat it as DIRENT_HTREE
3537 */
3538 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3539 if (IS_ERR(bh)) {
3540 *retval = PTR_ERR(bh);
3541 return NULL;
3542 }
3543
3544 de = (struct ext4_dir_entry_2 *) bh->b_data;
3545 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3546 bh->b_size, 0) ||
3547 le32_to_cpu(de->inode) != inode->i_ino ||
3548 strcmp(".", de->name)) {
3549 EXT4_ERROR_INODE(inode, "directory missing '.'");
3550 brelse(bh);
3551 *retval = -EFSCORRUPTED;
3552 return NULL;
3553 }
3554 offset = ext4_rec_len_from_disk(de->rec_len,
3555 inode->i_sb->s_blocksize);
3556 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3557 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3558 bh->b_size, offset) ||
3559 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3560 EXT4_ERROR_INODE(inode, "directory missing '..'");
3561 brelse(bh);
3562 *retval = -EFSCORRUPTED;
3563 return NULL;
3564 }
3565 *parent_de = de;
3566
3567 return bh;
3568 }
3569
3570 *inlined = 1;
3571 return ext4_get_first_inline_block(inode, parent_de, retval);
3572}
3573
3574struct ext4_renament {
3575 struct inode *dir;
3576 struct dentry *dentry;
3577 struct inode *inode;
3578 bool is_dir;
3579 int dir_nlink_delta;
3580
3581 /* entry for "dentry" */
3582 struct buffer_head *bh;
3583 struct ext4_dir_entry_2 *de;
3584 int inlined;
3585
3586 /* entry for ".." in inode if it's a directory */
3587 struct buffer_head *dir_bh;
3588 struct ext4_dir_entry_2 *parent_de;
3589 int dir_inlined;
3590};
3591
3592static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent, bool is_cross)
3593{
3594 int retval;
3595
3596 ent->is_dir = true;
3597 if (!is_cross)
3598 return 0;
3599
3600 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3601 &retval, &ent->parent_de,
3602 &ent->dir_inlined);
3603 if (!ent->dir_bh)
3604 return retval;
3605 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3606 return -EFSCORRUPTED;
3607 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3608 return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3609 ent->dir_bh, EXT4_JTR_NONE);
3610}
3611
3612static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3613 unsigned dir_ino)
3614{
3615 int retval;
3616
3617 if (!ent->dir_bh)
3618 return 0;
3619
3620 ent->parent_de->inode = cpu_to_le32(dir_ino);
3621 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3622 if (!ent->dir_inlined) {
3623 if (is_dx(ent->inode)) {
3624 retval = ext4_handle_dirty_dx_node(handle,
3625 ent->inode,
3626 ent->dir_bh);
3627 } else {
3628 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3629 ent->dir_bh);
3630 }
3631 } else {
3632 retval = ext4_mark_inode_dirty(handle, ent->inode);
3633 }
3634 if (retval) {
3635 ext4_std_error(ent->dir->i_sb, retval);
3636 return retval;
3637 }
3638 return 0;
3639}
3640
3641static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3642 unsigned ino, unsigned file_type)
3643{
3644 int retval, retval2;
3645
3646 BUFFER_TRACE(ent->bh, "get write access");
3647 retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3648 EXT4_JTR_NONE);
3649 if (retval)
3650 return retval;
3651 ent->de->inode = cpu_to_le32(ino);
3652 if (ext4_has_feature_filetype(ent->dir->i_sb))
3653 ent->de->file_type = file_type;
3654 inode_inc_iversion(ent->dir);
3655 inode_set_mtime_to_ts(ent->dir, inode_set_ctime_current(ent->dir));
3656 retval = ext4_mark_inode_dirty(handle, ent->dir);
3657 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3658 if (!ent->inlined) {
3659 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3660 if (unlikely(retval2)) {
3661 ext4_std_error(ent->dir->i_sb, retval2);
3662 return retval2;
3663 }
3664 }
3665 return retval;
3666}
3667
3668static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3669 unsigned ino, unsigned file_type)
3670{
3671 struct ext4_renament old = *ent;
3672 int retval = 0;
3673
3674 /*
3675 * old->de could have moved from under us during make indexed dir,
3676 * so the old->de may no longer valid and need to find it again
3677 * before reset old inode info.
3678 */
3679 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3680 &old.inlined);
3681 if (IS_ERR(old.bh))
3682 retval = PTR_ERR(old.bh);
3683 if (!old.bh)
3684 retval = -ENOENT;
3685 if (retval) {
3686 ext4_std_error(old.dir->i_sb, retval);
3687 return;
3688 }
3689
3690 ext4_setent(handle, &old, ino, file_type);
3691 brelse(old.bh);
3692}
3693
3694static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3695 const struct qstr *d_name)
3696{
3697 int retval = -ENOENT;
3698 struct buffer_head *bh;
3699 struct ext4_dir_entry_2 *de;
3700
3701 bh = ext4_find_entry(dir, d_name, &de, NULL);
3702 if (IS_ERR(bh))
3703 return PTR_ERR(bh);
3704 if (bh) {
3705 retval = ext4_delete_entry(handle, dir, de, bh);
3706 brelse(bh);
3707 }
3708 return retval;
3709}
3710
3711static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3712 int force_reread)
3713{
3714 int retval;
3715 /*
3716 * ent->de could have moved from under us during htree split, so make
3717 * sure that we are deleting the right entry. We might also be pointing
3718 * to a stale entry in the unused part of ent->bh so just checking inum
3719 * and the name isn't enough.
3720 */
3721 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3722 ent->de->name_len != ent->dentry->d_name.len ||
3723 strncmp(ent->de->name, ent->dentry->d_name.name,
3724 ent->de->name_len) ||
3725 force_reread) {
3726 retval = ext4_find_delete_entry(handle, ent->dir,
3727 &ent->dentry->d_name);
3728 } else {
3729 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3730 if (retval == -ENOENT) {
3731 retval = ext4_find_delete_entry(handle, ent->dir,
3732 &ent->dentry->d_name);
3733 }
3734 }
3735
3736 if (retval) {
3737 ext4_warning_inode(ent->dir,
3738 "Deleting old file: nlink %d, error=%d",
3739 ent->dir->i_nlink, retval);
3740 }
3741}
3742
3743static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3744{
3745 if (ent->dir_nlink_delta) {
3746 if (ent->dir_nlink_delta == -1)
3747 ext4_dec_count(ent->dir);
3748 else
3749 ext4_inc_count(ent->dir);
3750 ext4_mark_inode_dirty(handle, ent->dir);
3751 }
3752}
3753
3754static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3755 struct ext4_renament *ent,
3756 int credits, handle_t **h)
3757{
3758 struct inode *wh;
3759 handle_t *handle;
3760 int retries = 0;
3761
3762 /*
3763 * for inode block, sb block, group summaries,
3764 * and inode bitmap
3765 */
3766 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3767 EXT4_XATTR_TRANS_BLOCKS + 4);
3768retry:
3769 wh = ext4_new_inode_start_handle(idmap, ent->dir,
3770 S_IFCHR | WHITEOUT_MODE,
3771 &ent->dentry->d_name, 0, NULL,
3772 EXT4_HT_DIR, credits);
3773
3774 handle = ext4_journal_current_handle();
3775 if (IS_ERR(wh)) {
3776 if (handle)
3777 ext4_journal_stop(handle);
3778 if (PTR_ERR(wh) == -ENOSPC &&
3779 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3780 goto retry;
3781 } else {
3782 *h = handle;
3783 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3784 wh->i_op = &ext4_special_inode_operations;
3785 }
3786 return wh;
3787}
3788
3789/*
3790 * Anybody can rename anything with this: the permission checks are left to the
3791 * higher-level routines.
3792 *
3793 * n.b. old_{dentry,inode) refers to the source dentry/inode
3794 * while new_{dentry,inode) refers to the destination dentry/inode
3795 * This comes from rename(const char *oldpath, const char *newpath)
3796 */
3797static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3798 struct dentry *old_dentry, struct inode *new_dir,
3799 struct dentry *new_dentry, unsigned int flags)
3800{
3801 handle_t *handle = NULL;
3802 struct ext4_renament old = {
3803 .dir = old_dir,
3804 .dentry = old_dentry,
3805 .inode = d_inode(old_dentry),
3806 };
3807 struct ext4_renament new = {
3808 .dir = new_dir,
3809 .dentry = new_dentry,
3810 .inode = d_inode(new_dentry),
3811 };
3812 int force_reread;
3813 int retval;
3814 struct inode *whiteout = NULL;
3815 int credits;
3816 u8 old_file_type;
3817
3818 if (new.inode && new.inode->i_nlink == 0) {
3819 EXT4_ERROR_INODE(new.inode,
3820 "target of rename is already freed");
3821 return -EFSCORRUPTED;
3822 }
3823
3824 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3825 (!projid_eq(EXT4_I(new_dir)->i_projid,
3826 EXT4_I(old_dentry->d_inode)->i_projid)))
3827 return -EXDEV;
3828
3829 retval = dquot_initialize(old.dir);
3830 if (retval)
3831 return retval;
3832 retval = dquot_initialize(old.inode);
3833 if (retval)
3834 return retval;
3835 retval = dquot_initialize(new.dir);
3836 if (retval)
3837 return retval;
3838
3839 /* Initialize quotas before so that eventual writes go
3840 * in separate transaction */
3841 if (new.inode) {
3842 retval = dquot_initialize(new.inode);
3843 if (retval)
3844 return retval;
3845 }
3846
3847 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3848 &old.inlined);
3849 if (IS_ERR(old.bh))
3850 return PTR_ERR(old.bh);
3851
3852 /*
3853 * Check for inode number is _not_ due to possible IO errors.
3854 * We might rmdir the source, keep it as pwd of some process
3855 * and merrily kill the link to whatever was created under the
3856 * same name. Goodbye sticky bit ;-<
3857 */
3858 retval = -ENOENT;
3859 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3860 goto release_bh;
3861
3862 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3863 &new.de, &new.inlined);
3864 if (IS_ERR(new.bh)) {
3865 retval = PTR_ERR(new.bh);
3866 new.bh = NULL;
3867 goto release_bh;
3868 }
3869 if (new.bh) {
3870 if (!new.inode) {
3871 brelse(new.bh);
3872 new.bh = NULL;
3873 }
3874 }
3875 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3876 ext4_alloc_da_blocks(old.inode);
3877
3878 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3879 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3880 if (!(flags & RENAME_WHITEOUT)) {
3881 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3882 if (IS_ERR(handle)) {
3883 retval = PTR_ERR(handle);
3884 goto release_bh;
3885 }
3886 } else {
3887 whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3888 if (IS_ERR(whiteout)) {
3889 retval = PTR_ERR(whiteout);
3890 goto release_bh;
3891 }
3892 }
3893
3894 old_file_type = old.de->file_type;
3895 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3896 ext4_handle_sync(handle);
3897
3898 if (S_ISDIR(old.inode->i_mode)) {
3899 if (new.inode) {
3900 retval = -ENOTEMPTY;
3901 if (!ext4_empty_dir(new.inode))
3902 goto end_rename;
3903 } else {
3904 retval = -EMLINK;
3905 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3906 goto end_rename;
3907 }
3908 retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
3909 if (retval)
3910 goto end_rename;
3911 }
3912 /*
3913 * If we're renaming a file within an inline_data dir and adding or
3914 * setting the new dirent causes a conversion from inline_data to
3915 * extents/blockmap, we need to force the dirent delete code to
3916 * re-read the directory, or else we end up trying to delete a dirent
3917 * from what is now the extent tree root (or a block map).
3918 */
3919 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3920 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3921
3922 if (whiteout) {
3923 /*
3924 * Do this before adding a new entry, so the old entry is sure
3925 * to be still pointing to the valid old entry.
3926 */
3927 retval = ext4_setent(handle, &old, whiteout->i_ino,
3928 EXT4_FT_CHRDEV);
3929 if (retval)
3930 goto end_rename;
3931 retval = ext4_mark_inode_dirty(handle, whiteout);
3932 if (unlikely(retval))
3933 goto end_rename;
3934
3935 }
3936 if (!new.bh) {
3937 retval = ext4_add_entry(handle, new.dentry, old.inode);
3938 if (retval)
3939 goto end_rename;
3940 } else {
3941 retval = ext4_setent(handle, &new,
3942 old.inode->i_ino, old_file_type);
3943 if (retval)
3944 goto end_rename;
3945 }
3946 if (force_reread)
3947 force_reread = !ext4_test_inode_flag(new.dir,
3948 EXT4_INODE_INLINE_DATA);
3949
3950 /*
3951 * Like most other Unix systems, set the ctime for inodes on a
3952 * rename.
3953 */
3954 inode_set_ctime_current(old.inode);
3955 retval = ext4_mark_inode_dirty(handle, old.inode);
3956 if (unlikely(retval))
3957 goto end_rename;
3958
3959 if (!whiteout) {
3960 /*
3961 * ok, that's it
3962 */
3963 ext4_rename_delete(handle, &old, force_reread);
3964 }
3965
3966 if (new.inode) {
3967 ext4_dec_count(new.inode);
3968 inode_set_ctime_current(new.inode);
3969 }
3970 inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
3971 ext4_update_dx_flag(old.dir);
3972 if (old.is_dir) {
3973 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3974 if (retval)
3975 goto end_rename;
3976
3977 ext4_dec_count(old.dir);
3978 if (new.inode) {
3979 /* checked ext4_empty_dir above, can't have another
3980 * parent, ext4_dec_count() won't work for many-linked
3981 * dirs */
3982 clear_nlink(new.inode);
3983 } else {
3984 ext4_inc_count(new.dir);
3985 ext4_update_dx_flag(new.dir);
3986 retval = ext4_mark_inode_dirty(handle, new.dir);
3987 if (unlikely(retval))
3988 goto end_rename;
3989 }
3990 }
3991 retval = ext4_mark_inode_dirty(handle, old.dir);
3992 if (unlikely(retval))
3993 goto end_rename;
3994
3995 if (old.is_dir) {
3996 /*
3997 * We disable fast commits here that's because the
3998 * replay code is not yet capable of changing dot dot
3999 * dirents in directories.
4000 */
4001 ext4_fc_mark_ineligible(old.inode->i_sb,
4002 EXT4_FC_REASON_RENAME_DIR, handle);
4003 } else {
4004 struct super_block *sb = old.inode->i_sb;
4005
4006 if (new.inode)
4007 ext4_fc_track_unlink(handle, new.dentry);
4008 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4009 !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4010 !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
4011 __ext4_fc_track_link(handle, old.inode, new.dentry);
4012 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4013 if (whiteout)
4014 __ext4_fc_track_create(handle, whiteout,
4015 old.dentry);
4016 }
4017 }
4018
4019 if (new.inode) {
4020 retval = ext4_mark_inode_dirty(handle, new.inode);
4021 if (unlikely(retval))
4022 goto end_rename;
4023 if (!new.inode->i_nlink)
4024 ext4_orphan_add(handle, new.inode);
4025 }
4026 retval = 0;
4027
4028end_rename:
4029 if (whiteout) {
4030 if (retval) {
4031 ext4_resetent(handle, &old,
4032 old.inode->i_ino, old_file_type);
4033 drop_nlink(whiteout);
4034 ext4_mark_inode_dirty(handle, whiteout);
4035 ext4_orphan_add(handle, whiteout);
4036 }
4037 unlock_new_inode(whiteout);
4038 ext4_journal_stop(handle);
4039 iput(whiteout);
4040 } else {
4041 ext4_journal_stop(handle);
4042 }
4043release_bh:
4044 brelse(old.dir_bh);
4045 brelse(old.bh);
4046 brelse(new.bh);
4047
4048 return retval;
4049}
4050
4051static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4052 struct inode *new_dir, struct dentry *new_dentry)
4053{
4054 handle_t *handle = NULL;
4055 struct ext4_renament old = {
4056 .dir = old_dir,
4057 .dentry = old_dentry,
4058 .inode = d_inode(old_dentry),
4059 };
4060 struct ext4_renament new = {
4061 .dir = new_dir,
4062 .dentry = new_dentry,
4063 .inode = d_inode(new_dentry),
4064 };
4065 u8 new_file_type;
4066 int retval;
4067
4068 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4069 !projid_eq(EXT4_I(new_dir)->i_projid,
4070 EXT4_I(old_dentry->d_inode)->i_projid)) ||
4071 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4072 !projid_eq(EXT4_I(old_dir)->i_projid,
4073 EXT4_I(new_dentry->d_inode)->i_projid)))
4074 return -EXDEV;
4075
4076 retval = dquot_initialize(old.dir);
4077 if (retval)
4078 return retval;
4079 retval = dquot_initialize(new.dir);
4080 if (retval)
4081 return retval;
4082
4083 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4084 &old.de, &old.inlined);
4085 if (IS_ERR(old.bh))
4086 return PTR_ERR(old.bh);
4087 /*
4088 * Check for inode number is _not_ due to possible IO errors.
4089 * We might rmdir the source, keep it as pwd of some process
4090 * and merrily kill the link to whatever was created under the
4091 * same name. Goodbye sticky bit ;-<
4092 */
4093 retval = -ENOENT;
4094 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4095 goto end_rename;
4096
4097 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4098 &new.de, &new.inlined);
4099 if (IS_ERR(new.bh)) {
4100 retval = PTR_ERR(new.bh);
4101 new.bh = NULL;
4102 goto end_rename;
4103 }
4104
4105 /* RENAME_EXCHANGE case: old *and* new must both exist */
4106 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4107 goto end_rename;
4108
4109 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4110 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4111 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4112 if (IS_ERR(handle)) {
4113 retval = PTR_ERR(handle);
4114 handle = NULL;
4115 goto end_rename;
4116 }
4117
4118 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4119 ext4_handle_sync(handle);
4120
4121 if (S_ISDIR(old.inode->i_mode)) {
4122 retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
4123 if (retval)
4124 goto end_rename;
4125 }
4126 if (S_ISDIR(new.inode->i_mode)) {
4127 retval = ext4_rename_dir_prepare(handle, &new, new.dir != old.dir);
4128 if (retval)
4129 goto end_rename;
4130 }
4131
4132 /*
4133 * Other than the special case of overwriting a directory, parents'
4134 * nlink only needs to be modified if this is a cross directory rename.
4135 */
4136 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4137 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4138 new.dir_nlink_delta = -old.dir_nlink_delta;
4139 retval = -EMLINK;
4140 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4141 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4142 goto end_rename;
4143 }
4144
4145 new_file_type = new.de->file_type;
4146 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4147 if (retval)
4148 goto end_rename;
4149
4150 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4151 if (retval)
4152 goto end_rename;
4153
4154 /*
4155 * Like most other Unix systems, set the ctime for inodes on a
4156 * rename.
4157 */
4158 inode_set_ctime_current(old.inode);
4159 inode_set_ctime_current(new.inode);
4160 retval = ext4_mark_inode_dirty(handle, old.inode);
4161 if (unlikely(retval))
4162 goto end_rename;
4163 retval = ext4_mark_inode_dirty(handle, new.inode);
4164 if (unlikely(retval))
4165 goto end_rename;
4166 ext4_fc_mark_ineligible(new.inode->i_sb,
4167 EXT4_FC_REASON_CROSS_RENAME, handle);
4168 if (old.dir_bh) {
4169 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4170 if (retval)
4171 goto end_rename;
4172 }
4173 if (new.dir_bh) {
4174 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4175 if (retval)
4176 goto end_rename;
4177 }
4178 ext4_update_dir_count(handle, &old);
4179 ext4_update_dir_count(handle, &new);
4180 retval = 0;
4181
4182end_rename:
4183 brelse(old.dir_bh);
4184 brelse(new.dir_bh);
4185 brelse(old.bh);
4186 brelse(new.bh);
4187 if (handle)
4188 ext4_journal_stop(handle);
4189 return retval;
4190}
4191
4192static int ext4_rename2(struct mnt_idmap *idmap,
4193 struct inode *old_dir, struct dentry *old_dentry,
4194 struct inode *new_dir, struct dentry *new_dentry,
4195 unsigned int flags)
4196{
4197 int err;
4198
4199 if (unlikely(ext4_forced_shutdown(old_dir->i_sb)))
4200 return -EIO;
4201
4202 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4203 return -EINVAL;
4204
4205 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4206 flags);
4207 if (err)
4208 return err;
4209
4210 if (flags & RENAME_EXCHANGE) {
4211 return ext4_cross_rename(old_dir, old_dentry,
4212 new_dir, new_dentry);
4213 }
4214
4215 return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4216}
4217
4218/*
4219 * directories can handle most operations...
4220 */
4221const struct inode_operations ext4_dir_inode_operations = {
4222 .create = ext4_create,
4223 .lookup = ext4_lookup,
4224 .link = ext4_link,
4225 .unlink = ext4_unlink,
4226 .symlink = ext4_symlink,
4227 .mkdir = ext4_mkdir,
4228 .rmdir = ext4_rmdir,
4229 .mknod = ext4_mknod,
4230 .tmpfile = ext4_tmpfile,
4231 .rename = ext4_rename2,
4232 .setattr = ext4_setattr,
4233 .getattr = ext4_getattr,
4234 .listxattr = ext4_listxattr,
4235 .get_inode_acl = ext4_get_acl,
4236 .set_acl = ext4_set_acl,
4237 .fiemap = ext4_fiemap,
4238 .fileattr_get = ext4_fileattr_get,
4239 .fileattr_set = ext4_fileattr_set,
4240};
4241
4242const struct inode_operations ext4_special_inode_operations = {
4243 .setattr = ext4_setattr,
4244 .getattr = ext4_getattr,
4245 .listxattr = ext4_listxattr,
4246 .get_inode_acl = ext4_get_acl,
4247 .set_acl = ext4_set_acl,
4248};