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
347#ifdef PARANOID
348 struct ext4_dir_entry *d, *top;
349
350 d = (struct ext4_dir_entry *)bh->b_data;
351 top = (struct ext4_dir_entry *)(bh->b_data +
352 (EXT4_BLOCK_SIZE(inode->i_sb) -
353 sizeof(struct ext4_dir_entry_tail)));
354 while (d < top && d->rec_len)
355 d = (struct ext4_dir_entry *)(((void *)d) +
356 le16_to_cpu(d->rec_len));
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 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
368 t->det_reserved_zero2 ||
369 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
370 return NULL;
371
372 return t;
373}
374
375static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
376{
377 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
378 struct ext4_inode_info *ei = EXT4_I(inode);
379 __u32 csum;
380
381 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
382 return cpu_to_le32(csum);
383}
384
385#define warn_no_space_for_csum(inode) \
386 __warn_no_space_for_csum((inode), __func__, __LINE__)
387
388static void __warn_no_space_for_csum(struct inode *inode, const char *func,
389 unsigned int line)
390{
391 __ext4_warning_inode(inode, func, line,
392 "No space for directory leaf checksum. Please run e2fsck -D.");
393}
394
395int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
396{
397 struct ext4_dir_entry_tail *t;
398
399 if (!ext4_has_metadata_csum(inode->i_sb))
400 return 1;
401
402 t = get_dirent_tail(inode, bh);
403 if (!t) {
404 warn_no_space_for_csum(inode);
405 return 0;
406 }
407
408 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
409 (char *)t - bh->b_data))
410 return 0;
411
412 return 1;
413}
414
415static void ext4_dirblock_csum_set(struct inode *inode,
416 struct buffer_head *bh)
417{
418 struct ext4_dir_entry_tail *t;
419
420 if (!ext4_has_metadata_csum(inode->i_sb))
421 return;
422
423 t = get_dirent_tail(inode, bh);
424 if (!t) {
425 warn_no_space_for_csum(inode);
426 return;
427 }
428
429 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
430 (char *)t - bh->b_data);
431}
432
433int ext4_handle_dirty_dirblock(handle_t *handle,
434 struct inode *inode,
435 struct buffer_head *bh)
436{
437 ext4_dirblock_csum_set(inode, bh);
438 return ext4_handle_dirty_metadata(handle, inode, bh);
439}
440
441static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
442 struct ext4_dir_entry *dirent,
443 int *offset)
444{
445 struct ext4_dir_entry *dp;
446 struct dx_root_info *root;
447 int count_offset;
448
449 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
450 count_offset = 8;
451 else if (le16_to_cpu(dirent->rec_len) == 12) {
452 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
453 if (le16_to_cpu(dp->rec_len) !=
454 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
455 return NULL;
456 root = (struct dx_root_info *)(((void *)dp + 12));
457 if (root->reserved_zero ||
458 root->info_length != sizeof(struct dx_root_info))
459 return NULL;
460 count_offset = 32;
461 } else
462 return NULL;
463
464 if (offset)
465 *offset = count_offset;
466 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
467}
468
469static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
470 int count_offset, int count, struct dx_tail *t)
471{
472 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
473 struct ext4_inode_info *ei = EXT4_I(inode);
474 __u32 csum;
475 int size;
476 __u32 dummy_csum = 0;
477 int offset = offsetof(struct dx_tail, dt_checksum);
478
479 size = count_offset + (count * sizeof(struct dx_entry));
480 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
481 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
482 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
483
484 return cpu_to_le32(csum);
485}
486
487static int ext4_dx_csum_verify(struct inode *inode,
488 struct ext4_dir_entry *dirent)
489{
490 struct dx_countlimit *c;
491 struct dx_tail *t;
492 int count_offset, limit, count;
493
494 if (!ext4_has_metadata_csum(inode->i_sb))
495 return 1;
496
497 c = get_dx_countlimit(inode, dirent, &count_offset);
498 if (!c) {
499 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
500 return 0;
501 }
502 limit = le16_to_cpu(c->limit);
503 count = le16_to_cpu(c->count);
504 if (count_offset + (limit * sizeof(struct dx_entry)) >
505 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
506 warn_no_space_for_csum(inode);
507 return 0;
508 }
509 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
510
511 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
512 count, t))
513 return 0;
514 return 1;
515}
516
517static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
518{
519 struct dx_countlimit *c;
520 struct dx_tail *t;
521 int count_offset, limit, count;
522
523 if (!ext4_has_metadata_csum(inode->i_sb))
524 return;
525
526 c = get_dx_countlimit(inode, dirent, &count_offset);
527 if (!c) {
528 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
529 return;
530 }
531 limit = le16_to_cpu(c->limit);
532 count = le16_to_cpu(c->count);
533 if (count_offset + (limit * sizeof(struct dx_entry)) >
534 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
535 warn_no_space_for_csum(inode);
536 return;
537 }
538 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
539
540 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
541}
542
543static inline int ext4_handle_dirty_dx_node(handle_t *handle,
544 struct inode *inode,
545 struct buffer_head *bh)
546{
547 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
548 return ext4_handle_dirty_metadata(handle, inode, bh);
549}
550
551/*
552 * p is at least 6 bytes before the end of page
553 */
554static inline struct ext4_dir_entry_2 *
555ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
556{
557 return (struct ext4_dir_entry_2 *)((char *)p +
558 ext4_rec_len_from_disk(p->rec_len, blocksize));
559}
560
561/*
562 * Future: use high four bits of block for coalesce-on-delete flags
563 * Mask them off for now.
564 */
565
566static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
567{
568 return le32_to_cpu(entry->block) & 0x0fffffff;
569}
570
571static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
572{
573 entry->block = cpu_to_le32(value);
574}
575
576static inline unsigned dx_get_hash(struct dx_entry *entry)
577{
578 return le32_to_cpu(entry->hash);
579}
580
581static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
582{
583 entry->hash = cpu_to_le32(value);
584}
585
586static inline unsigned dx_get_count(struct dx_entry *entries)
587{
588 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
589}
590
591static inline unsigned dx_get_limit(struct dx_entry *entries)
592{
593 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
594}
595
596static inline void dx_set_count(struct dx_entry *entries, unsigned value)
597{
598 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
599}
600
601static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
602{
603 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
604}
605
606static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
607{
608 unsigned int entry_space = dir->i_sb->s_blocksize -
609 ext4_dir_rec_len(1, NULL) -
610 ext4_dir_rec_len(2, NULL) - infosize;
611
612 if (ext4_has_metadata_csum(dir->i_sb))
613 entry_space -= sizeof(struct dx_tail);
614 return entry_space / sizeof(struct dx_entry);
615}
616
617static inline unsigned dx_node_limit(struct inode *dir)
618{
619 unsigned int entry_space = dir->i_sb->s_blocksize -
620 ext4_dir_rec_len(0, dir);
621
622 if (ext4_has_metadata_csum(dir->i_sb))
623 entry_space -= sizeof(struct dx_tail);
624 return entry_space / sizeof(struct dx_entry);
625}
626
627/*
628 * Debug
629 */
630#ifdef DX_DEBUG
631static void dx_show_index(char * label, struct dx_entry *entries)
632{
633 int i, n = dx_get_count (entries);
634 printk(KERN_DEBUG "%s index", label);
635 for (i = 0; i < n; i++) {
636 printk(KERN_CONT " %x->%lu",
637 i ? dx_get_hash(entries + i) : 0,
638 (unsigned long)dx_get_block(entries + i));
639 }
640 printk(KERN_CONT "\n");
641}
642
643struct stats
644{
645 unsigned names;
646 unsigned space;
647 unsigned bcount;
648};
649
650static struct stats dx_show_leaf(struct inode *dir,
651 struct dx_hash_info *hinfo,
652 struct ext4_dir_entry_2 *de,
653 int size, int show_names)
654{
655 unsigned names = 0, space = 0;
656 char *base = (char *) de;
657 struct dx_hash_info h = *hinfo;
658
659 printk("names: ");
660 while ((char *) de < base + size)
661 {
662 if (de->inode)
663 {
664 if (show_names)
665 {
666#ifdef CONFIG_FS_ENCRYPTION
667 int len;
668 char *name;
669 struct fscrypt_str fname_crypto_str =
670 FSTR_INIT(NULL, 0);
671 int res = 0;
672
673 name = de->name;
674 len = de->name_len;
675 if (!IS_ENCRYPTED(dir)) {
676 /* Directory is not encrypted */
677 ext4fs_dirhash(dir, de->name,
678 de->name_len, &h);
679 printk("%*.s:(U)%x.%u ", len,
680 name, h.hash,
681 (unsigned) ((char *) de
682 - base));
683 } else {
684 struct fscrypt_str de_name =
685 FSTR_INIT(name, len);
686
687 /* Directory is encrypted */
688 res = fscrypt_fname_alloc_buffer(
689 len, &fname_crypto_str);
690 if (res)
691 printk(KERN_WARNING "Error "
692 "allocating crypto "
693 "buffer--skipping "
694 "crypto\n");
695 res = fscrypt_fname_disk_to_usr(dir,
696 0, 0, &de_name,
697 &fname_crypto_str);
698 if (res) {
699 printk(KERN_WARNING "Error "
700 "converting filename "
701 "from disk to usr"
702 "\n");
703 name = "??";
704 len = 2;
705 } else {
706 name = fname_crypto_str.name;
707 len = fname_crypto_str.len;
708 }
709 if (IS_CASEFOLDED(dir))
710 h.hash = EXT4_DIRENT_HASH(de);
711 else
712 ext4fs_dirhash(dir, de->name,
713 de->name_len, &h);
714 printk("%*.s:(E)%x.%u ", len, name,
715 h.hash, (unsigned) ((char *) de
716 - base));
717 fscrypt_fname_free_buffer(
718 &fname_crypto_str);
719 }
720#else
721 int len = de->name_len;
722 char *name = de->name;
723 ext4fs_dirhash(dir, de->name, de->name_len, &h);
724 printk("%*.s:%x.%u ", len, name, h.hash,
725 (unsigned) ((char *) de - base));
726#endif
727 }
728 space += ext4_dir_rec_len(de->name_len, dir);
729 names++;
730 }
731 de = ext4_next_entry(de, size);
732 }
733 printk(KERN_CONT "(%i)\n", names);
734 return (struct stats) { names, space, 1 };
735}
736
737struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
738 struct dx_entry *entries, int levels)
739{
740 unsigned blocksize = dir->i_sb->s_blocksize;
741 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
742 unsigned bcount = 0;
743 struct buffer_head *bh;
744 printk("%i indexed blocks...\n", count);
745 for (i = 0; i < count; i++, entries++)
746 {
747 ext4_lblk_t block = dx_get_block(entries);
748 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
749 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
750 struct stats stats;
751 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
752 bh = ext4_bread(NULL,dir, block, 0);
753 if (!bh || IS_ERR(bh))
754 continue;
755 stats = levels?
756 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
757 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
758 bh->b_data, blocksize, 0);
759 names += stats.names;
760 space += stats.space;
761 bcount += stats.bcount;
762 brelse(bh);
763 }
764 if (bcount)
765 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
766 levels ? "" : " ", names, space/bcount,
767 (space/bcount)*100/blocksize);
768 return (struct stats) { names, space, bcount};
769}
770
771/*
772 * Linear search cross check
773 */
774static inline void htree_rep_invariant_check(struct dx_entry *at,
775 struct dx_entry *target,
776 u32 hash, unsigned int n)
777{
778 while (n--) {
779 dxtrace(printk(KERN_CONT ","));
780 if (dx_get_hash(++at) > hash) {
781 at--;
782 break;
783 }
784 }
785 ASSERT(at == target - 1);
786}
787#else /* DX_DEBUG */
788static inline void htree_rep_invariant_check(struct dx_entry *at,
789 struct dx_entry *target,
790 u32 hash, unsigned int n)
791{
792}
793#endif /* DX_DEBUG */
794
795/*
796 * Probe for a directory leaf block to search.
797 *
798 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
799 * error in the directory index, and the caller should fall back to
800 * searching the directory normally. The callers of dx_probe **MUST**
801 * check for this error code, and make sure it never gets reflected
802 * back to userspace.
803 */
804static struct dx_frame *
805dx_probe(struct ext4_filename *fname, struct inode *dir,
806 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
807{
808 unsigned count, indirect, level, i;
809 struct dx_entry *at, *entries, *p, *q, *m;
810 struct dx_root *root;
811 struct dx_frame *frame = frame_in;
812 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
813 u32 hash;
814 ext4_lblk_t block;
815 ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
816
817 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
818 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
819 if (IS_ERR(frame->bh))
820 return (struct dx_frame *) frame->bh;
821
822 root = (struct dx_root *) frame->bh->b_data;
823 if (root->info.hash_version != DX_HASH_TEA &&
824 root->info.hash_version != DX_HASH_HALF_MD4 &&
825 root->info.hash_version != DX_HASH_LEGACY &&
826 root->info.hash_version != DX_HASH_SIPHASH) {
827 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
828 root->info.hash_version);
829 goto fail;
830 }
831 if (ext4_hash_in_dirent(dir)) {
832 if (root->info.hash_version != DX_HASH_SIPHASH) {
833 ext4_warning_inode(dir,
834 "Hash in dirent, but hash is not SIPHASH");
835 goto fail;
836 }
837 } else {
838 if (root->info.hash_version == DX_HASH_SIPHASH) {
839 ext4_warning_inode(dir,
840 "Hash code is SIPHASH, but hash not in dirent");
841 goto fail;
842 }
843 }
844 if (fname)
845 hinfo = &fname->hinfo;
846 hinfo->hash_version = root->info.hash_version;
847 if (hinfo->hash_version <= DX_HASH_TEA)
848 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
849 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
850 /* hash is already computed for encrypted casefolded directory */
851 if (fname && fname_name(fname) &&
852 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)))
853 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
854 hash = hinfo->hash;
855
856 if (root->info.unused_flags & 1) {
857 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
858 root->info.unused_flags);
859 goto fail;
860 }
861
862 indirect = root->info.indirect_levels;
863 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
864 ext4_warning(dir->i_sb,
865 "Directory (ino: %lu) htree depth %#06x exceed"
866 "supported value", dir->i_ino,
867 ext4_dir_htree_level(dir->i_sb));
868 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
869 ext4_warning(dir->i_sb, "Enable large directory "
870 "feature to access it");
871 }
872 goto fail;
873 }
874
875 entries = (struct dx_entry *)(((char *)&root->info) +
876 root->info.info_length);
877
878 if (dx_get_limit(entries) != dx_root_limit(dir,
879 root->info.info_length)) {
880 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
881 dx_get_limit(entries),
882 dx_root_limit(dir, root->info.info_length));
883 goto fail;
884 }
885
886 dxtrace(printk("Look up %x", hash));
887 level = 0;
888 blocks[0] = 0;
889 while (1) {
890 count = dx_get_count(entries);
891 if (!count || count > dx_get_limit(entries)) {
892 ext4_warning_inode(dir,
893 "dx entry: count %u beyond limit %u",
894 count, dx_get_limit(entries));
895 goto fail;
896 }
897
898 p = entries + 1;
899 q = entries + count - 1;
900 while (p <= q) {
901 m = p + (q - p) / 2;
902 dxtrace(printk(KERN_CONT "."));
903 if (dx_get_hash(m) > hash)
904 q = m - 1;
905 else
906 p = m + 1;
907 }
908
909 htree_rep_invariant_check(entries, p, hash, count - 1);
910
911 at = p - 1;
912 dxtrace(printk(KERN_CONT " %x->%u\n",
913 at == entries ? 0 : dx_get_hash(at),
914 dx_get_block(at)));
915 frame->entries = entries;
916 frame->at = at;
917
918 block = dx_get_block(at);
919 for (i = 0; i <= level; i++) {
920 if (blocks[i] == block) {
921 ext4_warning_inode(dir,
922 "dx entry: tree cycle block %u points back to block %u",
923 blocks[level], block);
924 goto fail;
925 }
926 }
927 if (++level > indirect)
928 return frame;
929 blocks[level] = block;
930 frame++;
931 frame->bh = ext4_read_dirblock(dir, block, INDEX);
932 if (IS_ERR(frame->bh)) {
933 ret_err = (struct dx_frame *) frame->bh;
934 frame->bh = NULL;
935 goto fail;
936 }
937
938 entries = ((struct dx_node *) frame->bh->b_data)->entries;
939
940 if (dx_get_limit(entries) != dx_node_limit(dir)) {
941 ext4_warning_inode(dir,
942 "dx entry: limit %u != node limit %u",
943 dx_get_limit(entries), dx_node_limit(dir));
944 goto fail;
945 }
946 }
947fail:
948 while (frame >= frame_in) {
949 brelse(frame->bh);
950 frame--;
951 }
952
953 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
954 ext4_warning_inode(dir,
955 "Corrupt directory, running e2fsck is recommended");
956 return ret_err;
957}
958
959static void dx_release(struct dx_frame *frames)
960{
961 struct dx_root_info *info;
962 int i;
963 unsigned int indirect_levels;
964
965 if (frames[0].bh == NULL)
966 return;
967
968 info = &((struct dx_root *)frames[0].bh->b_data)->info;
969 /* save local copy, "info" may be freed after brelse() */
970 indirect_levels = info->indirect_levels;
971 for (i = 0; i <= indirect_levels; i++) {
972 if (frames[i].bh == NULL)
973 break;
974 brelse(frames[i].bh);
975 frames[i].bh = NULL;
976 }
977}
978
979/*
980 * This function increments the frame pointer to search the next leaf
981 * block, and reads in the necessary intervening nodes if the search
982 * should be necessary. Whether or not the search is necessary is
983 * controlled by the hash parameter. If the hash value is even, then
984 * the search is only continued if the next block starts with that
985 * hash value. This is used if we are searching for a specific file.
986 *
987 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
988 *
989 * This function returns 1 if the caller should continue to search,
990 * or 0 if it should not. If there is an error reading one of the
991 * index blocks, it will a negative error code.
992 *
993 * If start_hash is non-null, it will be filled in with the starting
994 * hash of the next page.
995 */
996static int ext4_htree_next_block(struct inode *dir, __u32 hash,
997 struct dx_frame *frame,
998 struct dx_frame *frames,
999 __u32 *start_hash)
1000{
1001 struct dx_frame *p;
1002 struct buffer_head *bh;
1003 int num_frames = 0;
1004 __u32 bhash;
1005
1006 p = frame;
1007 /*
1008 * Find the next leaf page by incrementing the frame pointer.
1009 * If we run out of entries in the interior node, loop around and
1010 * increment pointer in the parent node. When we break out of
1011 * this loop, num_frames indicates the number of interior
1012 * nodes need to be read.
1013 */
1014 while (1) {
1015 if (++(p->at) < p->entries + dx_get_count(p->entries))
1016 break;
1017 if (p == frames)
1018 return 0;
1019 num_frames++;
1020 p--;
1021 }
1022
1023 /*
1024 * If the hash is 1, then continue only if the next page has a
1025 * continuation hash of any value. This is used for readdir
1026 * handling. Otherwise, check to see if the hash matches the
1027 * desired continuation hash. If it doesn't, return since
1028 * there's no point to read in the successive index pages.
1029 */
1030 bhash = dx_get_hash(p->at);
1031 if (start_hash)
1032 *start_hash = bhash;
1033 if ((hash & 1) == 0) {
1034 if ((bhash & ~1) != hash)
1035 return 0;
1036 }
1037 /*
1038 * If the hash is HASH_NB_ALWAYS, we always go to the next
1039 * block so no check is necessary
1040 */
1041 while (num_frames--) {
1042 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1043 if (IS_ERR(bh))
1044 return PTR_ERR(bh);
1045 p++;
1046 brelse(p->bh);
1047 p->bh = bh;
1048 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1049 }
1050 return 1;
1051}
1052
1053
1054/*
1055 * This function fills a red-black tree with information from a
1056 * directory block. It returns the number directory entries loaded
1057 * into the tree. If there is an error it is returned in err.
1058 */
1059static int htree_dirblock_to_tree(struct file *dir_file,
1060 struct inode *dir, ext4_lblk_t block,
1061 struct dx_hash_info *hinfo,
1062 __u32 start_hash, __u32 start_minor_hash)
1063{
1064 struct buffer_head *bh;
1065 struct ext4_dir_entry_2 *de, *top;
1066 int err = 0, count = 0;
1067 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1068 int csum = ext4_has_metadata_csum(dir->i_sb);
1069
1070 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1071 (unsigned long)block));
1072 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1073 if (IS_ERR(bh))
1074 return PTR_ERR(bh);
1075
1076 de = (struct ext4_dir_entry_2 *) bh->b_data;
1077 /* csum entries are not larger in the casefolded encrypted case */
1078 top = (struct ext4_dir_entry_2 *) ((char *) de +
1079 dir->i_sb->s_blocksize -
1080 ext4_dir_rec_len(0,
1081 csum ? NULL : dir));
1082 /* Check if the directory is encrypted */
1083 if (IS_ENCRYPTED(dir)) {
1084 err = fscrypt_prepare_readdir(dir);
1085 if (err < 0) {
1086 brelse(bh);
1087 return err;
1088 }
1089 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1090 &fname_crypto_str);
1091 if (err < 0) {
1092 brelse(bh);
1093 return err;
1094 }
1095 }
1096
1097 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1098 if (ext4_check_dir_entry(dir, NULL, de, bh,
1099 bh->b_data, bh->b_size,
1100 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1101 + ((char *)de - bh->b_data))) {
1102 /* silently ignore the rest of the block */
1103 break;
1104 }
1105 if (ext4_hash_in_dirent(dir)) {
1106 if (de->name_len && de->inode) {
1107 hinfo->hash = EXT4_DIRENT_HASH(de);
1108 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1109 } else {
1110 hinfo->hash = 0;
1111 hinfo->minor_hash = 0;
1112 }
1113 } else {
1114 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1115 }
1116 if ((hinfo->hash < start_hash) ||
1117 ((hinfo->hash == start_hash) &&
1118 (hinfo->minor_hash < start_minor_hash)))
1119 continue;
1120 if (de->inode == 0)
1121 continue;
1122 if (!IS_ENCRYPTED(dir)) {
1123 tmp_str.name = de->name;
1124 tmp_str.len = de->name_len;
1125 err = ext4_htree_store_dirent(dir_file,
1126 hinfo->hash, hinfo->minor_hash, de,
1127 &tmp_str);
1128 } else {
1129 int save_len = fname_crypto_str.len;
1130 struct fscrypt_str de_name = FSTR_INIT(de->name,
1131 de->name_len);
1132
1133 /* Directory is encrypted */
1134 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1135 hinfo->minor_hash, &de_name,
1136 &fname_crypto_str);
1137 if (err) {
1138 count = err;
1139 goto errout;
1140 }
1141 err = ext4_htree_store_dirent(dir_file,
1142 hinfo->hash, hinfo->minor_hash, de,
1143 &fname_crypto_str);
1144 fname_crypto_str.len = save_len;
1145 }
1146 if (err != 0) {
1147 count = err;
1148 goto errout;
1149 }
1150 count++;
1151 }
1152errout:
1153 brelse(bh);
1154 fscrypt_fname_free_buffer(&fname_crypto_str);
1155 return count;
1156}
1157
1158
1159/*
1160 * This function fills a red-black tree with information from a
1161 * directory. We start scanning the directory in hash order, starting
1162 * at start_hash and start_minor_hash.
1163 *
1164 * This function returns the number of entries inserted into the tree,
1165 * or a negative error code.
1166 */
1167int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1168 __u32 start_minor_hash, __u32 *next_hash)
1169{
1170 struct dx_hash_info hinfo;
1171 struct ext4_dir_entry_2 *de;
1172 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1173 struct inode *dir;
1174 ext4_lblk_t block;
1175 int count = 0;
1176 int ret, err;
1177 __u32 hashval;
1178 struct fscrypt_str tmp_str;
1179
1180 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1181 start_hash, start_minor_hash));
1182 dir = file_inode(dir_file);
1183 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1184 if (ext4_hash_in_dirent(dir))
1185 hinfo.hash_version = DX_HASH_SIPHASH;
1186 else
1187 hinfo.hash_version =
1188 EXT4_SB(dir->i_sb)->s_def_hash_version;
1189 if (hinfo.hash_version <= DX_HASH_TEA)
1190 hinfo.hash_version +=
1191 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1192 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1193 if (ext4_has_inline_data(dir)) {
1194 int has_inline_data = 1;
1195 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1196 &hinfo, start_hash,
1197 start_minor_hash,
1198 &has_inline_data);
1199 if (has_inline_data) {
1200 *next_hash = ~0;
1201 return count;
1202 }
1203 }
1204 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1205 start_hash, start_minor_hash);
1206 *next_hash = ~0;
1207 return count;
1208 }
1209 hinfo.hash = start_hash;
1210 hinfo.minor_hash = 0;
1211 frame = dx_probe(NULL, dir, &hinfo, frames);
1212 if (IS_ERR(frame))
1213 return PTR_ERR(frame);
1214
1215 /* Add '.' and '..' from the htree header */
1216 if (!start_hash && !start_minor_hash) {
1217 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1218 tmp_str.name = de->name;
1219 tmp_str.len = de->name_len;
1220 err = ext4_htree_store_dirent(dir_file, 0, 0,
1221 de, &tmp_str);
1222 if (err != 0)
1223 goto errout;
1224 count++;
1225 }
1226 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1227 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1228 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1229 tmp_str.name = de->name;
1230 tmp_str.len = de->name_len;
1231 err = ext4_htree_store_dirent(dir_file, 2, 0,
1232 de, &tmp_str);
1233 if (err != 0)
1234 goto errout;
1235 count++;
1236 }
1237
1238 while (1) {
1239 if (fatal_signal_pending(current)) {
1240 err = -ERESTARTSYS;
1241 goto errout;
1242 }
1243 cond_resched();
1244 block = dx_get_block(frame->at);
1245 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1246 start_hash, start_minor_hash);
1247 if (ret < 0) {
1248 err = ret;
1249 goto errout;
1250 }
1251 count += ret;
1252 hashval = ~0;
1253 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1254 frame, frames, &hashval);
1255 *next_hash = hashval;
1256 if (ret < 0) {
1257 err = ret;
1258 goto errout;
1259 }
1260 /*
1261 * Stop if: (a) there are no more entries, or
1262 * (b) we have inserted at least one entry and the
1263 * next hash value is not a continuation
1264 */
1265 if ((ret == 0) ||
1266 (count && ((hashval & 1) == 0)))
1267 break;
1268 }
1269 dx_release(frames);
1270 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1271 "next hash: %x\n", count, *next_hash));
1272 return count;
1273errout:
1274 dx_release(frames);
1275 return (err);
1276}
1277
1278static inline int search_dirblock(struct buffer_head *bh,
1279 struct inode *dir,
1280 struct ext4_filename *fname,
1281 unsigned int offset,
1282 struct ext4_dir_entry_2 **res_dir)
1283{
1284 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1285 fname, offset, res_dir);
1286}
1287
1288/*
1289 * Directory block splitting, compacting
1290 */
1291
1292/*
1293 * Create map of hash values, offsets, and sizes, stored at end of block.
1294 * Returns number of entries mapped.
1295 */
1296static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1297 struct dx_hash_info *hinfo,
1298 struct dx_map_entry *map_tail)
1299{
1300 int count = 0;
1301 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1302 unsigned int buflen = bh->b_size;
1303 char *base = bh->b_data;
1304 struct dx_hash_info h = *hinfo;
1305
1306 if (ext4_has_metadata_csum(dir->i_sb))
1307 buflen -= sizeof(struct ext4_dir_entry_tail);
1308
1309 while ((char *) de < base + buflen) {
1310 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1311 ((char *)de) - base))
1312 return -EFSCORRUPTED;
1313 if (de->name_len && de->inode) {
1314 if (ext4_hash_in_dirent(dir))
1315 h.hash = EXT4_DIRENT_HASH(de);
1316 else
1317 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1318 map_tail--;
1319 map_tail->hash = h.hash;
1320 map_tail->offs = ((char *) de - base)>>2;
1321 map_tail->size = le16_to_cpu(de->rec_len);
1322 count++;
1323 cond_resched();
1324 }
1325 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1326 }
1327 return count;
1328}
1329
1330/* Sort map by hash value */
1331static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1332{
1333 struct dx_map_entry *p, *q, *top = map + count - 1;
1334 int more;
1335 /* Combsort until bubble sort doesn't suck */
1336 while (count > 2) {
1337 count = count*10/13;
1338 if (count - 9 < 2) /* 9, 10 -> 11 */
1339 count = 11;
1340 for (p = top, q = p - count; q >= map; p--, q--)
1341 if (p->hash < q->hash)
1342 swap(*p, *q);
1343 }
1344 /* Garden variety bubble sort */
1345 do {
1346 more = 0;
1347 q = top;
1348 while (q-- > map) {
1349 if (q[1].hash >= q[0].hash)
1350 continue;
1351 swap(*(q+1), *q);
1352 more = 1;
1353 }
1354 } while(more);
1355}
1356
1357static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1358{
1359 struct dx_entry *entries = frame->entries;
1360 struct dx_entry *old = frame->at, *new = old + 1;
1361 int count = dx_get_count(entries);
1362
1363 ASSERT(count < dx_get_limit(entries));
1364 ASSERT(old < entries + count);
1365 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1366 dx_set_hash(new, hash);
1367 dx_set_block(new, block);
1368 dx_set_count(entries, count + 1);
1369}
1370
1371#if IS_ENABLED(CONFIG_UNICODE)
1372/*
1373 * Test whether a case-insensitive directory entry matches the filename
1374 * being searched for. If quick is set, assume the name being looked up
1375 * is already in the casefolded form.
1376 *
1377 * Returns: 0 if the directory entry matches, more than 0 if it
1378 * doesn't match or less than zero on error.
1379 */
1380static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1381 u8 *de_name, size_t de_name_len, bool quick)
1382{
1383 const struct super_block *sb = parent->i_sb;
1384 const struct unicode_map *um = sb->s_encoding;
1385 struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1386 struct qstr entry = QSTR_INIT(de_name, de_name_len);
1387 int ret;
1388
1389 if (IS_ENCRYPTED(parent)) {
1390 const struct fscrypt_str encrypted_name =
1391 FSTR_INIT(de_name, de_name_len);
1392
1393 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1394 if (!decrypted_name.name)
1395 return -ENOMEM;
1396 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1397 &decrypted_name);
1398 if (ret < 0)
1399 goto out;
1400 entry.name = decrypted_name.name;
1401 entry.len = decrypted_name.len;
1402 }
1403
1404 if (quick)
1405 ret = utf8_strncasecmp_folded(um, name, &entry);
1406 else
1407 ret = utf8_strncasecmp(um, name, &entry);
1408 if (ret < 0) {
1409 /* Handle invalid character sequence as either an error
1410 * or as an opaque byte sequence.
1411 */
1412 if (sb_has_strict_encoding(sb))
1413 ret = -EINVAL;
1414 else if (name->len != entry.len)
1415 ret = 1;
1416 else
1417 ret = !!memcmp(name->name, entry.name, entry.len);
1418 }
1419out:
1420 kfree(decrypted_name.name);
1421 return ret;
1422}
1423
1424int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1425 struct ext4_filename *name)
1426{
1427 struct fscrypt_str *cf_name = &name->cf_name;
1428 struct dx_hash_info *hinfo = &name->hinfo;
1429 int len;
1430
1431 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding ||
1432 (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1433 cf_name->name = NULL;
1434 return 0;
1435 }
1436
1437 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1438 if (!cf_name->name)
1439 return -ENOMEM;
1440
1441 len = utf8_casefold(dir->i_sb->s_encoding,
1442 iname, cf_name->name,
1443 EXT4_NAME_LEN);
1444 if (len <= 0) {
1445 kfree(cf_name->name);
1446 cf_name->name = NULL;
1447 }
1448 cf_name->len = (unsigned) len;
1449 if (!IS_ENCRYPTED(dir))
1450 return 0;
1451
1452 hinfo->hash_version = DX_HASH_SIPHASH;
1453 hinfo->seed = NULL;
1454 if (cf_name->name)
1455 ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1456 else
1457 ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1458 return 0;
1459}
1460#endif
1461
1462/*
1463 * Test whether a directory entry matches the filename being searched for.
1464 *
1465 * Return: %true if the directory entry matches, otherwise %false.
1466 */
1467static bool ext4_match(struct inode *parent,
1468 const struct ext4_filename *fname,
1469 struct ext4_dir_entry_2 *de)
1470{
1471 struct fscrypt_name f;
1472
1473 if (!de->inode)
1474 return false;
1475
1476 f.usr_fname = fname->usr_fname;
1477 f.disk_name = fname->disk_name;
1478#ifdef CONFIG_FS_ENCRYPTION
1479 f.crypto_buf = fname->crypto_buf;
1480#endif
1481
1482#if IS_ENABLED(CONFIG_UNICODE)
1483 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent) &&
1484 (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1485 if (fname->cf_name.name) {
1486 struct qstr cf = {.name = fname->cf_name.name,
1487 .len = fname->cf_name.len};
1488 if (IS_ENCRYPTED(parent)) {
1489 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1490 fname->hinfo.minor_hash !=
1491 EXT4_DIRENT_MINOR_HASH(de)) {
1492
1493 return false;
1494 }
1495 }
1496 return !ext4_ci_compare(parent, &cf, de->name,
1497 de->name_len, true);
1498 }
1499 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1500 de->name_len, false);
1501 }
1502#endif
1503
1504 return fscrypt_match_name(&f, de->name, de->name_len);
1505}
1506
1507/*
1508 * Returns 0 if not found, -1 on failure, and 1 on success
1509 */
1510int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1511 struct inode *dir, struct ext4_filename *fname,
1512 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1513{
1514 struct ext4_dir_entry_2 * de;
1515 char * dlimit;
1516 int de_len;
1517
1518 de = (struct ext4_dir_entry_2 *)search_buf;
1519 dlimit = search_buf + buf_size;
1520 while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1521 /* this code is executed quadratically often */
1522 /* do minimal checking `by hand' */
1523 if (de->name + de->name_len <= dlimit &&
1524 ext4_match(dir, fname, de)) {
1525 /* found a match - just to be sure, do
1526 * a full check */
1527 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1528 buf_size, offset))
1529 return -1;
1530 *res_dir = de;
1531 return 1;
1532 }
1533 /* prevent looping on a bad block */
1534 de_len = ext4_rec_len_from_disk(de->rec_len,
1535 dir->i_sb->s_blocksize);
1536 if (de_len <= 0)
1537 return -1;
1538 offset += de_len;
1539 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1540 }
1541 return 0;
1542}
1543
1544static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1545 struct ext4_dir_entry *de)
1546{
1547 struct super_block *sb = dir->i_sb;
1548
1549 if (!is_dx(dir))
1550 return 0;
1551 if (block == 0)
1552 return 1;
1553 if (de->inode == 0 &&
1554 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1555 sb->s_blocksize)
1556 return 1;
1557 return 0;
1558}
1559
1560/*
1561 * __ext4_find_entry()
1562 *
1563 * finds an entry in the specified directory with the wanted name. It
1564 * returns the cache buffer in which the entry was found, and the entry
1565 * itself (as a parameter - res_dir). It does NOT read the inode of the
1566 * entry - you'll have to do that yourself if you want to.
1567 *
1568 * The returned buffer_head has ->b_count elevated. The caller is expected
1569 * to brelse() it when appropriate.
1570 */
1571static struct buffer_head *__ext4_find_entry(struct inode *dir,
1572 struct ext4_filename *fname,
1573 struct ext4_dir_entry_2 **res_dir,
1574 int *inlined)
1575{
1576 struct super_block *sb;
1577 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1578 struct buffer_head *bh, *ret = NULL;
1579 ext4_lblk_t start, block;
1580 const u8 *name = fname->usr_fname->name;
1581 size_t ra_max = 0; /* Number of bh's in the readahead
1582 buffer, bh_use[] */
1583 size_t ra_ptr = 0; /* Current index into readahead
1584 buffer */
1585 ext4_lblk_t nblocks;
1586 int i, namelen, retval;
1587
1588 *res_dir = NULL;
1589 sb = dir->i_sb;
1590 namelen = fname->usr_fname->len;
1591 if (namelen > EXT4_NAME_LEN)
1592 return NULL;
1593
1594 if (ext4_has_inline_data(dir)) {
1595 int has_inline_data = 1;
1596 ret = ext4_find_inline_entry(dir, fname, res_dir,
1597 &has_inline_data);
1598 if (has_inline_data) {
1599 if (inlined)
1600 *inlined = 1;
1601 goto cleanup_and_exit;
1602 }
1603 }
1604
1605 if ((namelen <= 2) && (name[0] == '.') &&
1606 (name[1] == '.' || name[1] == '\0')) {
1607 /*
1608 * "." or ".." will only be in the first block
1609 * NFS may look up ".."; "." should be handled by the VFS
1610 */
1611 block = start = 0;
1612 nblocks = 1;
1613 goto restart;
1614 }
1615 if (is_dx(dir)) {
1616 ret = ext4_dx_find_entry(dir, fname, res_dir);
1617 /*
1618 * On success, or if the error was file not found,
1619 * return. Otherwise, fall back to doing a search the
1620 * old fashioned way.
1621 */
1622 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1623 goto cleanup_and_exit;
1624 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1625 "falling back\n"));
1626 ret = NULL;
1627 }
1628 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1629 if (!nblocks) {
1630 ret = NULL;
1631 goto cleanup_and_exit;
1632 }
1633 start = EXT4_I(dir)->i_dir_start_lookup;
1634 if (start >= nblocks)
1635 start = 0;
1636 block = start;
1637restart:
1638 do {
1639 /*
1640 * We deal with the read-ahead logic here.
1641 */
1642 cond_resched();
1643 if (ra_ptr >= ra_max) {
1644 /* Refill the readahead buffer */
1645 ra_ptr = 0;
1646 if (block < start)
1647 ra_max = start - block;
1648 else
1649 ra_max = nblocks - block;
1650 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1651 retval = ext4_bread_batch(dir, block, ra_max,
1652 false /* wait */, bh_use);
1653 if (retval) {
1654 ret = ERR_PTR(retval);
1655 ra_max = 0;
1656 goto cleanup_and_exit;
1657 }
1658 }
1659 if ((bh = bh_use[ra_ptr++]) == NULL)
1660 goto next;
1661 wait_on_buffer(bh);
1662 if (!buffer_uptodate(bh)) {
1663 EXT4_ERROR_INODE_ERR(dir, EIO,
1664 "reading directory lblock %lu",
1665 (unsigned long) block);
1666 brelse(bh);
1667 ret = ERR_PTR(-EIO);
1668 goto cleanup_and_exit;
1669 }
1670 if (!buffer_verified(bh) &&
1671 !is_dx_internal_node(dir, block,
1672 (struct ext4_dir_entry *)bh->b_data) &&
1673 !ext4_dirblock_csum_verify(dir, bh)) {
1674 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1675 "checksumming directory "
1676 "block %lu", (unsigned long)block);
1677 brelse(bh);
1678 ret = ERR_PTR(-EFSBADCRC);
1679 goto cleanup_and_exit;
1680 }
1681 set_buffer_verified(bh);
1682 i = search_dirblock(bh, dir, fname,
1683 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1684 if (i == 1) {
1685 EXT4_I(dir)->i_dir_start_lookup = block;
1686 ret = bh;
1687 goto cleanup_and_exit;
1688 } else {
1689 brelse(bh);
1690 if (i < 0)
1691 goto cleanup_and_exit;
1692 }
1693 next:
1694 if (++block >= nblocks)
1695 block = 0;
1696 } while (block != start);
1697
1698 /*
1699 * If the directory has grown while we were searching, then
1700 * search the last part of the directory before giving up.
1701 */
1702 block = nblocks;
1703 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1704 if (block < nblocks) {
1705 start = 0;
1706 goto restart;
1707 }
1708
1709cleanup_and_exit:
1710 /* Clean up the read-ahead blocks */
1711 for (; ra_ptr < ra_max; ra_ptr++)
1712 brelse(bh_use[ra_ptr]);
1713 return ret;
1714}
1715
1716static struct buffer_head *ext4_find_entry(struct inode *dir,
1717 const struct qstr *d_name,
1718 struct ext4_dir_entry_2 **res_dir,
1719 int *inlined)
1720{
1721 int err;
1722 struct ext4_filename fname;
1723 struct buffer_head *bh;
1724
1725 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1726 if (err == -ENOENT)
1727 return NULL;
1728 if (err)
1729 return ERR_PTR(err);
1730
1731 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1732
1733 ext4_fname_free_filename(&fname);
1734 return bh;
1735}
1736
1737static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1738 struct dentry *dentry,
1739 struct ext4_dir_entry_2 **res_dir)
1740{
1741 int err;
1742 struct ext4_filename fname;
1743 struct buffer_head *bh;
1744
1745 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1746 generic_set_encrypted_ci_d_ops(dentry);
1747 if (err == -ENOENT)
1748 return NULL;
1749 if (err)
1750 return ERR_PTR(err);
1751
1752 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1753
1754 ext4_fname_free_filename(&fname);
1755 return bh;
1756}
1757
1758static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1759 struct ext4_filename *fname,
1760 struct ext4_dir_entry_2 **res_dir)
1761{
1762 struct super_block * sb = dir->i_sb;
1763 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1764 struct buffer_head *bh;
1765 ext4_lblk_t block;
1766 int retval;
1767
1768#ifdef CONFIG_FS_ENCRYPTION
1769 *res_dir = NULL;
1770#endif
1771 frame = dx_probe(fname, dir, NULL, frames);
1772 if (IS_ERR(frame))
1773 return (struct buffer_head *) frame;
1774 do {
1775 block = dx_get_block(frame->at);
1776 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1777 if (IS_ERR(bh))
1778 goto errout;
1779
1780 retval = search_dirblock(bh, dir, fname,
1781 block << EXT4_BLOCK_SIZE_BITS(sb),
1782 res_dir);
1783 if (retval == 1)
1784 goto success;
1785 brelse(bh);
1786 if (retval == -1) {
1787 bh = ERR_PTR(ERR_BAD_DX_DIR);
1788 goto errout;
1789 }
1790
1791 /* Check to see if we should continue to search */
1792 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1793 frames, NULL);
1794 if (retval < 0) {
1795 ext4_warning_inode(dir,
1796 "error %d reading directory index block",
1797 retval);
1798 bh = ERR_PTR(retval);
1799 goto errout;
1800 }
1801 } while (retval == 1);
1802
1803 bh = NULL;
1804errout:
1805 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1806success:
1807 dx_release(frames);
1808 return bh;
1809}
1810
1811static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1812{
1813 struct inode *inode;
1814 struct ext4_dir_entry_2 *de;
1815 struct buffer_head *bh;
1816
1817 if (dentry->d_name.len > EXT4_NAME_LEN)
1818 return ERR_PTR(-ENAMETOOLONG);
1819
1820 bh = ext4_lookup_entry(dir, dentry, &de);
1821 if (IS_ERR(bh))
1822 return ERR_CAST(bh);
1823 inode = NULL;
1824 if (bh) {
1825 __u32 ino = le32_to_cpu(de->inode);
1826 brelse(bh);
1827 if (!ext4_valid_inum(dir->i_sb, ino)) {
1828 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1829 return ERR_PTR(-EFSCORRUPTED);
1830 }
1831 if (unlikely(ino == dir->i_ino)) {
1832 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1833 dentry);
1834 return ERR_PTR(-EFSCORRUPTED);
1835 }
1836 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1837 if (inode == ERR_PTR(-ESTALE)) {
1838 EXT4_ERROR_INODE(dir,
1839 "deleted inode referenced: %u",
1840 ino);
1841 return ERR_PTR(-EFSCORRUPTED);
1842 }
1843 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1844 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1845 !fscrypt_has_permitted_context(dir, inode)) {
1846 ext4_warning(inode->i_sb,
1847 "Inconsistent encryption contexts: %lu/%lu",
1848 dir->i_ino, inode->i_ino);
1849 iput(inode);
1850 return ERR_PTR(-EPERM);
1851 }
1852 }
1853
1854#if IS_ENABLED(CONFIG_UNICODE)
1855 if (!inode && IS_CASEFOLDED(dir)) {
1856 /* Eventually we want to call d_add_ci(dentry, NULL)
1857 * for negative dentries in the encoding case as
1858 * well. For now, prevent the negative dentry
1859 * from being cached.
1860 */
1861 return NULL;
1862 }
1863#endif
1864 return d_splice_alias(inode, dentry);
1865}
1866
1867
1868struct dentry *ext4_get_parent(struct dentry *child)
1869{
1870 __u32 ino;
1871 struct ext4_dir_entry_2 * de;
1872 struct buffer_head *bh;
1873
1874 bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1875 if (IS_ERR(bh))
1876 return ERR_CAST(bh);
1877 if (!bh)
1878 return ERR_PTR(-ENOENT);
1879 ino = le32_to_cpu(de->inode);
1880 brelse(bh);
1881
1882 if (!ext4_valid_inum(child->d_sb, ino)) {
1883 EXT4_ERROR_INODE(d_inode(child),
1884 "bad parent inode number: %u", ino);
1885 return ERR_PTR(-EFSCORRUPTED);
1886 }
1887
1888 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1889}
1890
1891/*
1892 * Move count entries from end of map between two memory locations.
1893 * Returns pointer to last entry moved.
1894 */
1895static struct ext4_dir_entry_2 *
1896dx_move_dirents(struct inode *dir, char *from, char *to,
1897 struct dx_map_entry *map, int count,
1898 unsigned blocksize)
1899{
1900 unsigned rec_len = 0;
1901
1902 while (count--) {
1903 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1904 (from + (map->offs<<2));
1905 rec_len = ext4_dir_rec_len(de->name_len, dir);
1906
1907 memcpy (to, de, rec_len);
1908 ((struct ext4_dir_entry_2 *) to)->rec_len =
1909 ext4_rec_len_to_disk(rec_len, blocksize);
1910
1911 /* wipe dir_entry excluding the rec_len field */
1912 de->inode = 0;
1913 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1914 blocksize) -
1915 offsetof(struct ext4_dir_entry_2,
1916 name_len));
1917
1918 map++;
1919 to += rec_len;
1920 }
1921 return (struct ext4_dir_entry_2 *) (to - rec_len);
1922}
1923
1924/*
1925 * Compact each dir entry in the range to the minimal rec_len.
1926 * Returns pointer to last entry in range.
1927 */
1928static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1929 unsigned int blocksize)
1930{
1931 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1932 unsigned rec_len = 0;
1933
1934 prev = to = de;
1935 while ((char*)de < base + blocksize) {
1936 next = ext4_next_entry(de, blocksize);
1937 if (de->inode && de->name_len) {
1938 rec_len = ext4_dir_rec_len(de->name_len, dir);
1939 if (de > to)
1940 memmove(to, de, rec_len);
1941 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1942 prev = to;
1943 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1944 }
1945 de = next;
1946 }
1947 return prev;
1948}
1949
1950/*
1951 * Split a full leaf block to make room for a new dir entry.
1952 * Allocate a new block, and move entries so that they are approx. equally full.
1953 * Returns pointer to de in block into which the new entry will be inserted.
1954 */
1955static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1956 struct buffer_head **bh,struct dx_frame *frame,
1957 struct dx_hash_info *hinfo)
1958{
1959 unsigned blocksize = dir->i_sb->s_blocksize;
1960 unsigned continued;
1961 int count;
1962 struct buffer_head *bh2;
1963 ext4_lblk_t newblock;
1964 u32 hash2;
1965 struct dx_map_entry *map;
1966 char *data1 = (*bh)->b_data, *data2;
1967 unsigned split, move, size;
1968 struct ext4_dir_entry_2 *de = NULL, *de2;
1969 int csum_size = 0;
1970 int err = 0, i;
1971
1972 if (ext4_has_metadata_csum(dir->i_sb))
1973 csum_size = sizeof(struct ext4_dir_entry_tail);
1974
1975 bh2 = ext4_append(handle, dir, &newblock);
1976 if (IS_ERR(bh2)) {
1977 brelse(*bh);
1978 *bh = NULL;
1979 return (struct ext4_dir_entry_2 *) bh2;
1980 }
1981
1982 BUFFER_TRACE(*bh, "get_write_access");
1983 err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
1984 EXT4_JTR_NONE);
1985 if (err)
1986 goto journal_error;
1987
1988 BUFFER_TRACE(frame->bh, "get_write_access");
1989 err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
1990 EXT4_JTR_NONE);
1991 if (err)
1992 goto journal_error;
1993
1994 data2 = bh2->b_data;
1995
1996 /* create map in the end of data2 block */
1997 map = (struct dx_map_entry *) (data2 + blocksize);
1998 count = dx_make_map(dir, *bh, hinfo, map);
1999 if (count < 0) {
2000 err = count;
2001 goto journal_error;
2002 }
2003 map -= count;
2004 dx_sort_map(map, count);
2005 /* Ensure that neither split block is over half full */
2006 size = 0;
2007 move = 0;
2008 for (i = count-1; i >= 0; i--) {
2009 /* is more than half of this entry in 2nd half of the block? */
2010 if (size + map[i].size/2 > blocksize/2)
2011 break;
2012 size += map[i].size;
2013 move++;
2014 }
2015 /*
2016 * map index at which we will split
2017 *
2018 * If the sum of active entries didn't exceed half the block size, just
2019 * split it in half by count; each resulting block will have at least
2020 * half the space free.
2021 */
2022 if (i > 0)
2023 split = count - move;
2024 else
2025 split = count/2;
2026
2027 hash2 = map[split].hash;
2028 continued = hash2 == map[split - 1].hash;
2029 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2030 (unsigned long)dx_get_block(frame->at),
2031 hash2, split, count-split));
2032
2033 /* Fancy dance to stay within two buffers */
2034 de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2035 blocksize);
2036 de = dx_pack_dirents(dir, data1, blocksize);
2037 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2038 (char *) de,
2039 blocksize);
2040 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2041 (char *) de2,
2042 blocksize);
2043 if (csum_size) {
2044 ext4_initialize_dirent_tail(*bh, blocksize);
2045 ext4_initialize_dirent_tail(bh2, blocksize);
2046 }
2047
2048 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2049 blocksize, 1));
2050 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2051 blocksize, 1));
2052
2053 /* Which block gets the new entry? */
2054 if (hinfo->hash >= hash2) {
2055 swap(*bh, bh2);
2056 de = de2;
2057 }
2058 dx_insert_block(frame, hash2 + continued, newblock);
2059 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2060 if (err)
2061 goto journal_error;
2062 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2063 if (err)
2064 goto journal_error;
2065 brelse(bh2);
2066 dxtrace(dx_show_index("frame", frame->entries));
2067 return de;
2068
2069journal_error:
2070 brelse(*bh);
2071 brelse(bh2);
2072 *bh = NULL;
2073 ext4_std_error(dir->i_sb, err);
2074 return ERR_PTR(err);
2075}
2076
2077int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2078 struct buffer_head *bh,
2079 void *buf, int buf_size,
2080 struct ext4_filename *fname,
2081 struct ext4_dir_entry_2 **dest_de)
2082{
2083 struct ext4_dir_entry_2 *de;
2084 unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2085 int nlen, rlen;
2086 unsigned int offset = 0;
2087 char *top;
2088
2089 de = buf;
2090 top = buf + buf_size - reclen;
2091 while ((char *) de <= top) {
2092 if (ext4_check_dir_entry(dir, NULL, de, bh,
2093 buf, buf_size, offset))
2094 return -EFSCORRUPTED;
2095 if (ext4_match(dir, fname, de))
2096 return -EEXIST;
2097 nlen = ext4_dir_rec_len(de->name_len, dir);
2098 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2099 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2100 break;
2101 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2102 offset += rlen;
2103 }
2104 if ((char *) de > top)
2105 return -ENOSPC;
2106
2107 *dest_de = de;
2108 return 0;
2109}
2110
2111void ext4_insert_dentry(struct inode *dir,
2112 struct inode *inode,
2113 struct ext4_dir_entry_2 *de,
2114 int buf_size,
2115 struct ext4_filename *fname)
2116{
2117
2118 int nlen, rlen;
2119
2120 nlen = ext4_dir_rec_len(de->name_len, dir);
2121 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2122 if (de->inode) {
2123 struct ext4_dir_entry_2 *de1 =
2124 (struct ext4_dir_entry_2 *)((char *)de + nlen);
2125 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2126 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2127 de = de1;
2128 }
2129 de->file_type = EXT4_FT_UNKNOWN;
2130 de->inode = cpu_to_le32(inode->i_ino);
2131 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2132 de->name_len = fname_len(fname);
2133 memcpy(de->name, fname_name(fname), fname_len(fname));
2134 if (ext4_hash_in_dirent(dir)) {
2135 struct dx_hash_info *hinfo = &fname->hinfo;
2136
2137 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2138 EXT4_DIRENT_HASHES(de)->minor_hash =
2139 cpu_to_le32(hinfo->minor_hash);
2140 }
2141}
2142
2143/*
2144 * Add a new entry into a directory (leaf) block. If de is non-NULL,
2145 * it points to a directory entry which is guaranteed to be large
2146 * enough for new directory entry. If de is NULL, then
2147 * add_dirent_to_buf will attempt search the directory block for
2148 * space. It will return -ENOSPC if no space is available, and -EIO
2149 * and -EEXIST if directory entry already exists.
2150 */
2151static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2152 struct inode *dir,
2153 struct inode *inode, struct ext4_dir_entry_2 *de,
2154 struct buffer_head *bh)
2155{
2156 unsigned int blocksize = dir->i_sb->s_blocksize;
2157 int csum_size = 0;
2158 int err, err2;
2159
2160 if (ext4_has_metadata_csum(inode->i_sb))
2161 csum_size = sizeof(struct ext4_dir_entry_tail);
2162
2163 if (!de) {
2164 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2165 blocksize - csum_size, fname, &de);
2166 if (err)
2167 return err;
2168 }
2169 BUFFER_TRACE(bh, "get_write_access");
2170 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2171 EXT4_JTR_NONE);
2172 if (err) {
2173 ext4_std_error(dir->i_sb, err);
2174 return err;
2175 }
2176
2177 /* By now the buffer is marked for journaling */
2178 ext4_insert_dentry(dir, inode, de, blocksize, fname);
2179
2180 /*
2181 * XXX shouldn't update any times until successful
2182 * completion of syscall, but too many callers depend
2183 * on this.
2184 *
2185 * XXX similarly, too many callers depend on
2186 * ext4_new_inode() setting the times, but error
2187 * recovery deletes the inode, so the worst that can
2188 * happen is that the times are slightly out of date
2189 * and/or different from the directory change time.
2190 */
2191 dir->i_mtime = dir->i_ctime = current_time(dir);
2192 ext4_update_dx_flag(dir);
2193 inode_inc_iversion(dir);
2194 err2 = ext4_mark_inode_dirty(handle, dir);
2195 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2196 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2197 if (err)
2198 ext4_std_error(dir->i_sb, err);
2199 return err ? err : err2;
2200}
2201
2202/*
2203 * This converts a one block unindexed directory to a 3 block indexed
2204 * directory, and adds the dentry to the indexed directory.
2205 */
2206static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2207 struct inode *dir,
2208 struct inode *inode, struct buffer_head *bh)
2209{
2210 struct buffer_head *bh2;
2211 struct dx_root *root;
2212 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2213 struct dx_entry *entries;
2214 struct ext4_dir_entry_2 *de, *de2;
2215 char *data2, *top;
2216 unsigned len;
2217 int retval;
2218 unsigned blocksize;
2219 ext4_lblk_t block;
2220 struct fake_dirent *fde;
2221 int csum_size = 0;
2222
2223 if (ext4_has_metadata_csum(inode->i_sb))
2224 csum_size = sizeof(struct ext4_dir_entry_tail);
2225
2226 blocksize = dir->i_sb->s_blocksize;
2227 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2228 BUFFER_TRACE(bh, "get_write_access");
2229 retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2230 EXT4_JTR_NONE);
2231 if (retval) {
2232 ext4_std_error(dir->i_sb, retval);
2233 brelse(bh);
2234 return retval;
2235 }
2236 root = (struct dx_root *) bh->b_data;
2237
2238 /* The 0th block becomes the root, move the dirents out */
2239 fde = &root->dotdot;
2240 de = (struct ext4_dir_entry_2 *)((char *)fde +
2241 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2242 if ((char *) de >= (((char *) root) + blocksize)) {
2243 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2244 brelse(bh);
2245 return -EFSCORRUPTED;
2246 }
2247 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2248
2249 /* Allocate new block for the 0th block's dirents */
2250 bh2 = ext4_append(handle, dir, &block);
2251 if (IS_ERR(bh2)) {
2252 brelse(bh);
2253 return PTR_ERR(bh2);
2254 }
2255 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2256 data2 = bh2->b_data;
2257
2258 memcpy(data2, de, len);
2259 memset(de, 0, len); /* wipe old data */
2260 de = (struct ext4_dir_entry_2 *) data2;
2261 top = data2 + len;
2262 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2263 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2264 (data2 + (blocksize - csum_size) -
2265 (char *) de))) {
2266 brelse(bh2);
2267 brelse(bh);
2268 return -EFSCORRUPTED;
2269 }
2270 de = de2;
2271 }
2272 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2273 (char *) de, blocksize);
2274
2275 if (csum_size)
2276 ext4_initialize_dirent_tail(bh2, blocksize);
2277
2278 /* Initialize the root; the dot dirents already exist */
2279 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2280 de->rec_len = ext4_rec_len_to_disk(
2281 blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2282 memset (&root->info, 0, sizeof(root->info));
2283 root->info.info_length = sizeof(root->info);
2284 if (ext4_hash_in_dirent(dir))
2285 root->info.hash_version = DX_HASH_SIPHASH;
2286 else
2287 root->info.hash_version =
2288 EXT4_SB(dir->i_sb)->s_def_hash_version;
2289
2290 entries = root->entries;
2291 dx_set_block(entries, 1);
2292 dx_set_count(entries, 1);
2293 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2294
2295 /* Initialize as for dx_probe */
2296 fname->hinfo.hash_version = root->info.hash_version;
2297 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2298 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2299 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2300
2301 /* casefolded encrypted hashes are computed on fname setup */
2302 if (!ext4_hash_in_dirent(dir))
2303 ext4fs_dirhash(dir, fname_name(fname),
2304 fname_len(fname), &fname->hinfo);
2305
2306 memset(frames, 0, sizeof(frames));
2307 frame = frames;
2308 frame->entries = entries;
2309 frame->at = entries;
2310 frame->bh = bh;
2311
2312 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2313 if (retval)
2314 goto out_frames;
2315 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2316 if (retval)
2317 goto out_frames;
2318
2319 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2320 if (IS_ERR(de)) {
2321 retval = PTR_ERR(de);
2322 goto out_frames;
2323 }
2324
2325 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2326out_frames:
2327 /*
2328 * Even if the block split failed, we have to properly write
2329 * out all the changes we did so far. Otherwise we can end up
2330 * with corrupted filesystem.
2331 */
2332 if (retval)
2333 ext4_mark_inode_dirty(handle, dir);
2334 dx_release(frames);
2335 brelse(bh2);
2336 return retval;
2337}
2338
2339/*
2340 * ext4_add_entry()
2341 *
2342 * adds a file entry to the specified directory, using the same
2343 * semantics as ext4_find_entry(). It returns NULL if it failed.
2344 *
2345 * NOTE!! The inode part of 'de' is left at 0 - which means you
2346 * may not sleep between calling this and putting something into
2347 * the entry, as someone else might have used it while you slept.
2348 */
2349static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2350 struct inode *inode)
2351{
2352 struct inode *dir = d_inode(dentry->d_parent);
2353 struct buffer_head *bh = NULL;
2354 struct ext4_dir_entry_2 *de;
2355 struct super_block *sb;
2356 struct ext4_filename fname;
2357 int retval;
2358 int dx_fallback=0;
2359 unsigned blocksize;
2360 ext4_lblk_t block, blocks;
2361 int csum_size = 0;
2362
2363 if (ext4_has_metadata_csum(inode->i_sb))
2364 csum_size = sizeof(struct ext4_dir_entry_tail);
2365
2366 sb = dir->i_sb;
2367 blocksize = sb->s_blocksize;
2368 if (!dentry->d_name.len)
2369 return -EINVAL;
2370
2371 if (fscrypt_is_nokey_name(dentry))
2372 return -ENOKEY;
2373
2374#if IS_ENABLED(CONFIG_UNICODE)
2375 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2376 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2377 return -EINVAL;
2378#endif
2379
2380 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2381 if (retval)
2382 return retval;
2383
2384 if (ext4_has_inline_data(dir)) {
2385 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2386 if (retval < 0)
2387 goto out;
2388 if (retval == 1) {
2389 retval = 0;
2390 goto out;
2391 }
2392 }
2393
2394 if (is_dx(dir)) {
2395 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2396 if (!retval || (retval != ERR_BAD_DX_DIR))
2397 goto out;
2398 /* Can we just ignore htree data? */
2399 if (ext4_has_metadata_csum(sb)) {
2400 EXT4_ERROR_INODE(dir,
2401 "Directory has corrupted htree index.");
2402 retval = -EFSCORRUPTED;
2403 goto out;
2404 }
2405 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2406 dx_fallback++;
2407 retval = ext4_mark_inode_dirty(handle, dir);
2408 if (unlikely(retval))
2409 goto out;
2410 }
2411 blocks = dir->i_size >> sb->s_blocksize_bits;
2412 for (block = 0; block < blocks; block++) {
2413 bh = ext4_read_dirblock(dir, block, DIRENT);
2414 if (bh == NULL) {
2415 bh = ext4_bread(handle, dir, block,
2416 EXT4_GET_BLOCKS_CREATE);
2417 goto add_to_new_block;
2418 }
2419 if (IS_ERR(bh)) {
2420 retval = PTR_ERR(bh);
2421 bh = NULL;
2422 goto out;
2423 }
2424 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2425 NULL, bh);
2426 if (retval != -ENOSPC)
2427 goto out;
2428
2429 if (blocks == 1 && !dx_fallback &&
2430 ext4_has_feature_dir_index(sb)) {
2431 retval = make_indexed_dir(handle, &fname, dir,
2432 inode, bh);
2433 bh = NULL; /* make_indexed_dir releases bh */
2434 goto out;
2435 }
2436 brelse(bh);
2437 }
2438 bh = ext4_append(handle, dir, &block);
2439add_to_new_block:
2440 if (IS_ERR(bh)) {
2441 retval = PTR_ERR(bh);
2442 bh = NULL;
2443 goto out;
2444 }
2445 de = (struct ext4_dir_entry_2 *) bh->b_data;
2446 de->inode = 0;
2447 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2448
2449 if (csum_size)
2450 ext4_initialize_dirent_tail(bh, blocksize);
2451
2452 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2453out:
2454 ext4_fname_free_filename(&fname);
2455 brelse(bh);
2456 if (retval == 0)
2457 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2458 return retval;
2459}
2460
2461/*
2462 * Returns 0 for success, or a negative error value
2463 */
2464static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2465 struct inode *dir, struct inode *inode)
2466{
2467 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2468 struct dx_entry *entries, *at;
2469 struct buffer_head *bh;
2470 struct super_block *sb = dir->i_sb;
2471 struct ext4_dir_entry_2 *de;
2472 int restart;
2473 int err;
2474
2475again:
2476 restart = 0;
2477 frame = dx_probe(fname, dir, NULL, frames);
2478 if (IS_ERR(frame))
2479 return PTR_ERR(frame);
2480 entries = frame->entries;
2481 at = frame->at;
2482 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2483 if (IS_ERR(bh)) {
2484 err = PTR_ERR(bh);
2485 bh = NULL;
2486 goto cleanup;
2487 }
2488
2489 BUFFER_TRACE(bh, "get_write_access");
2490 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2491 if (err)
2492 goto journal_error;
2493
2494 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2495 if (err != -ENOSPC)
2496 goto cleanup;
2497
2498 err = 0;
2499 /* Block full, should compress but for now just split */
2500 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2501 dx_get_count(entries), dx_get_limit(entries)));
2502 /* Need to split index? */
2503 if (dx_get_count(entries) == dx_get_limit(entries)) {
2504 ext4_lblk_t newblock;
2505 int levels = frame - frames + 1;
2506 unsigned int icount;
2507 int add_level = 1;
2508 struct dx_entry *entries2;
2509 struct dx_node *node2;
2510 struct buffer_head *bh2;
2511
2512 while (frame > frames) {
2513 if (dx_get_count((frame - 1)->entries) <
2514 dx_get_limit((frame - 1)->entries)) {
2515 add_level = 0;
2516 break;
2517 }
2518 frame--; /* split higher index block */
2519 at = frame->at;
2520 entries = frame->entries;
2521 restart = 1;
2522 }
2523 if (add_level && levels == ext4_dir_htree_level(sb)) {
2524 ext4_warning(sb, "Directory (ino: %lu) index full, "
2525 "reach max htree level :%d",
2526 dir->i_ino, levels);
2527 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2528 ext4_warning(sb, "Large directory feature is "
2529 "not enabled on this "
2530 "filesystem");
2531 }
2532 err = -ENOSPC;
2533 goto cleanup;
2534 }
2535 icount = dx_get_count(entries);
2536 bh2 = ext4_append(handle, dir, &newblock);
2537 if (IS_ERR(bh2)) {
2538 err = PTR_ERR(bh2);
2539 goto cleanup;
2540 }
2541 node2 = (struct dx_node *)(bh2->b_data);
2542 entries2 = node2->entries;
2543 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2544 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2545 sb->s_blocksize);
2546 BUFFER_TRACE(frame->bh, "get_write_access");
2547 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2548 EXT4_JTR_NONE);
2549 if (err)
2550 goto journal_error;
2551 if (!add_level) {
2552 unsigned icount1 = icount/2, icount2 = icount - icount1;
2553 unsigned hash2 = dx_get_hash(entries + icount1);
2554 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2555 icount1, icount2));
2556
2557 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2558 err = ext4_journal_get_write_access(handle, sb,
2559 (frame - 1)->bh,
2560 EXT4_JTR_NONE);
2561 if (err)
2562 goto journal_error;
2563
2564 memcpy((char *) entries2, (char *) (entries + icount1),
2565 icount2 * sizeof(struct dx_entry));
2566 dx_set_count(entries, icount1);
2567 dx_set_count(entries2, icount2);
2568 dx_set_limit(entries2, dx_node_limit(dir));
2569
2570 /* Which index block gets the new entry? */
2571 if (at - entries >= icount1) {
2572 frame->at = at - entries - icount1 + entries2;
2573 frame->entries = entries = entries2;
2574 swap(frame->bh, bh2);
2575 }
2576 dx_insert_block((frame - 1), hash2, newblock);
2577 dxtrace(dx_show_index("node", frame->entries));
2578 dxtrace(dx_show_index("node",
2579 ((struct dx_node *) bh2->b_data)->entries));
2580 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2581 if (err)
2582 goto journal_error;
2583 brelse (bh2);
2584 err = ext4_handle_dirty_dx_node(handle, dir,
2585 (frame - 1)->bh);
2586 if (err)
2587 goto journal_error;
2588 err = ext4_handle_dirty_dx_node(handle, dir,
2589 frame->bh);
2590 if (restart || err)
2591 goto journal_error;
2592 } else {
2593 struct dx_root *dxroot;
2594 memcpy((char *) entries2, (char *) entries,
2595 icount * sizeof(struct dx_entry));
2596 dx_set_limit(entries2, dx_node_limit(dir));
2597
2598 /* Set up root */
2599 dx_set_count(entries, 1);
2600 dx_set_block(entries + 0, newblock);
2601 dxroot = (struct dx_root *)frames[0].bh->b_data;
2602 dxroot->info.indirect_levels += 1;
2603 dxtrace(printk(KERN_DEBUG
2604 "Creating %d level index...\n",
2605 dxroot->info.indirect_levels));
2606 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2607 if (err)
2608 goto journal_error;
2609 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2610 brelse(bh2);
2611 restart = 1;
2612 goto journal_error;
2613 }
2614 }
2615 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2616 if (IS_ERR(de)) {
2617 err = PTR_ERR(de);
2618 goto cleanup;
2619 }
2620 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2621 goto cleanup;
2622
2623journal_error:
2624 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2625cleanup:
2626 brelse(bh);
2627 dx_release(frames);
2628 /* @restart is true means htree-path has been changed, we need to
2629 * repeat dx_probe() to find out valid htree-path
2630 */
2631 if (restart && err == 0)
2632 goto again;
2633 return err;
2634}
2635
2636/*
2637 * ext4_generic_delete_entry deletes a directory entry by merging it
2638 * with the previous entry
2639 */
2640int ext4_generic_delete_entry(struct inode *dir,
2641 struct ext4_dir_entry_2 *de_del,
2642 struct buffer_head *bh,
2643 void *entry_buf,
2644 int buf_size,
2645 int csum_size)
2646{
2647 struct ext4_dir_entry_2 *de, *pde;
2648 unsigned int blocksize = dir->i_sb->s_blocksize;
2649 int i;
2650
2651 i = 0;
2652 pde = NULL;
2653 de = entry_buf;
2654 while (i < buf_size - csum_size) {
2655 if (ext4_check_dir_entry(dir, NULL, de, bh,
2656 entry_buf, buf_size, i))
2657 return -EFSCORRUPTED;
2658 if (de == de_del) {
2659 if (pde) {
2660 pde->rec_len = ext4_rec_len_to_disk(
2661 ext4_rec_len_from_disk(pde->rec_len,
2662 blocksize) +
2663 ext4_rec_len_from_disk(de->rec_len,
2664 blocksize),
2665 blocksize);
2666
2667 /* wipe entire dir_entry */
2668 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2669 blocksize));
2670 } else {
2671 /* wipe dir_entry excluding the rec_len field */
2672 de->inode = 0;
2673 memset(&de->name_len, 0,
2674 ext4_rec_len_from_disk(de->rec_len,
2675 blocksize) -
2676 offsetof(struct ext4_dir_entry_2,
2677 name_len));
2678 }
2679
2680 inode_inc_iversion(dir);
2681 return 0;
2682 }
2683 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2684 pde = de;
2685 de = ext4_next_entry(de, blocksize);
2686 }
2687 return -ENOENT;
2688}
2689
2690static int ext4_delete_entry(handle_t *handle,
2691 struct inode *dir,
2692 struct ext4_dir_entry_2 *de_del,
2693 struct buffer_head *bh)
2694{
2695 int err, csum_size = 0;
2696
2697 if (ext4_has_inline_data(dir)) {
2698 int has_inline_data = 1;
2699 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2700 &has_inline_data);
2701 if (has_inline_data)
2702 return err;
2703 }
2704
2705 if (ext4_has_metadata_csum(dir->i_sb))
2706 csum_size = sizeof(struct ext4_dir_entry_tail);
2707
2708 BUFFER_TRACE(bh, "get_write_access");
2709 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2710 EXT4_JTR_NONE);
2711 if (unlikely(err))
2712 goto out;
2713
2714 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2715 dir->i_sb->s_blocksize, csum_size);
2716 if (err)
2717 goto out;
2718
2719 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2720 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2721 if (unlikely(err))
2722 goto out;
2723
2724 return 0;
2725out:
2726 if (err != -ENOENT)
2727 ext4_std_error(dir->i_sb, err);
2728 return err;
2729}
2730
2731/*
2732 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2733 * since this indicates that nlinks count was previously 1 to avoid overflowing
2734 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2735 * that subdirectory link counts are not being maintained accurately.
2736 *
2737 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2738 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2739 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2740 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2741 */
2742static void ext4_inc_count(struct inode *inode)
2743{
2744 inc_nlink(inode);
2745 if (is_dx(inode) &&
2746 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2747 set_nlink(inode, 1);
2748}
2749
2750/*
2751 * If a directory had nlink == 1, then we should let it be 1. This indicates
2752 * directory has >EXT4_LINK_MAX subdirs.
2753 */
2754static void ext4_dec_count(struct inode *inode)
2755{
2756 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2757 drop_nlink(inode);
2758}
2759
2760
2761/*
2762 * Add non-directory inode to a directory. On success, the inode reference is
2763 * consumed by dentry is instantiation. This is also indicated by clearing of
2764 * *inodep pointer. On failure, the caller is responsible for dropping the
2765 * inode reference in the safe context.
2766 */
2767static int ext4_add_nondir(handle_t *handle,
2768 struct dentry *dentry, struct inode **inodep)
2769{
2770 struct inode *dir = d_inode(dentry->d_parent);
2771 struct inode *inode = *inodep;
2772 int err = ext4_add_entry(handle, dentry, inode);
2773 if (!err) {
2774 err = ext4_mark_inode_dirty(handle, inode);
2775 if (IS_DIRSYNC(dir))
2776 ext4_handle_sync(handle);
2777 d_instantiate_new(dentry, inode);
2778 *inodep = NULL;
2779 return err;
2780 }
2781 drop_nlink(inode);
2782 ext4_orphan_add(handle, inode);
2783 unlock_new_inode(inode);
2784 return err;
2785}
2786
2787/*
2788 * By the time this is called, we already have created
2789 * the directory cache entry for the new file, but it
2790 * is so far negative - it has no inode.
2791 *
2792 * If the create succeeds, we fill in the inode information
2793 * with d_instantiate().
2794 */
2795static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir,
2796 struct dentry *dentry, umode_t mode, bool excl)
2797{
2798 handle_t *handle;
2799 struct inode *inode;
2800 int err, credits, retries = 0;
2801
2802 err = dquot_initialize(dir);
2803 if (err)
2804 return err;
2805
2806 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2807 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2808retry:
2809 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2810 0, NULL, EXT4_HT_DIR, credits);
2811 handle = ext4_journal_current_handle();
2812 err = PTR_ERR(inode);
2813 if (!IS_ERR(inode)) {
2814 inode->i_op = &ext4_file_inode_operations;
2815 inode->i_fop = &ext4_file_operations;
2816 ext4_set_aops(inode);
2817 err = ext4_add_nondir(handle, dentry, &inode);
2818 if (!err)
2819 ext4_fc_track_create(handle, dentry);
2820 }
2821 if (handle)
2822 ext4_journal_stop(handle);
2823 if (!IS_ERR_OR_NULL(inode))
2824 iput(inode);
2825 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2826 goto retry;
2827 return err;
2828}
2829
2830static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2831 struct dentry *dentry, umode_t mode, dev_t rdev)
2832{
2833 handle_t *handle;
2834 struct inode *inode;
2835 int err, credits, retries = 0;
2836
2837 err = dquot_initialize(dir);
2838 if (err)
2839 return err;
2840
2841 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2842 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2843retry:
2844 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2845 0, NULL, EXT4_HT_DIR, credits);
2846 handle = ext4_journal_current_handle();
2847 err = PTR_ERR(inode);
2848 if (!IS_ERR(inode)) {
2849 init_special_inode(inode, inode->i_mode, rdev);
2850 inode->i_op = &ext4_special_inode_operations;
2851 err = ext4_add_nondir(handle, dentry, &inode);
2852 if (!err)
2853 ext4_fc_track_create(handle, dentry);
2854 }
2855 if (handle)
2856 ext4_journal_stop(handle);
2857 if (!IS_ERR_OR_NULL(inode))
2858 iput(inode);
2859 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2860 goto retry;
2861 return err;
2862}
2863
2864static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2865 struct file *file, umode_t mode)
2866{
2867 handle_t *handle;
2868 struct inode *inode;
2869 int err, retries = 0;
2870
2871 err = dquot_initialize(dir);
2872 if (err)
2873 return err;
2874
2875retry:
2876 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2877 NULL, 0, NULL,
2878 EXT4_HT_DIR,
2879 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2880 4 + EXT4_XATTR_TRANS_BLOCKS);
2881 handle = ext4_journal_current_handle();
2882 err = PTR_ERR(inode);
2883 if (!IS_ERR(inode)) {
2884 inode->i_op = &ext4_file_inode_operations;
2885 inode->i_fop = &ext4_file_operations;
2886 ext4_set_aops(inode);
2887 d_tmpfile(file, inode);
2888 err = ext4_orphan_add(handle, inode);
2889 if (err)
2890 goto err_unlock_inode;
2891 mark_inode_dirty(inode);
2892 unlock_new_inode(inode);
2893 }
2894 if (handle)
2895 ext4_journal_stop(handle);
2896 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2897 goto retry;
2898 return finish_open_simple(file, err);
2899err_unlock_inode:
2900 ext4_journal_stop(handle);
2901 unlock_new_inode(inode);
2902 return err;
2903}
2904
2905struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2906 struct ext4_dir_entry_2 *de,
2907 int blocksize, int csum_size,
2908 unsigned int parent_ino, int dotdot_real_len)
2909{
2910 de->inode = cpu_to_le32(inode->i_ino);
2911 de->name_len = 1;
2912 de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2913 blocksize);
2914 strcpy(de->name, ".");
2915 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2916
2917 de = ext4_next_entry(de, blocksize);
2918 de->inode = cpu_to_le32(parent_ino);
2919 de->name_len = 2;
2920 if (!dotdot_real_len)
2921 de->rec_len = ext4_rec_len_to_disk(blocksize -
2922 (csum_size + ext4_dir_rec_len(1, NULL)),
2923 blocksize);
2924 else
2925 de->rec_len = ext4_rec_len_to_disk(
2926 ext4_dir_rec_len(de->name_len, NULL),
2927 blocksize);
2928 strcpy(de->name, "..");
2929 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2930
2931 return ext4_next_entry(de, blocksize);
2932}
2933
2934int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2935 struct inode *inode)
2936{
2937 struct buffer_head *dir_block = NULL;
2938 struct ext4_dir_entry_2 *de;
2939 ext4_lblk_t block = 0;
2940 unsigned int blocksize = dir->i_sb->s_blocksize;
2941 int csum_size = 0;
2942 int err;
2943
2944 if (ext4_has_metadata_csum(dir->i_sb))
2945 csum_size = sizeof(struct ext4_dir_entry_tail);
2946
2947 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2948 err = ext4_try_create_inline_dir(handle, dir, inode);
2949 if (err < 0 && err != -ENOSPC)
2950 goto out;
2951 if (!err)
2952 goto out;
2953 }
2954
2955 inode->i_size = 0;
2956 dir_block = ext4_append(handle, inode, &block);
2957 if (IS_ERR(dir_block))
2958 return PTR_ERR(dir_block);
2959 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2960 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2961 set_nlink(inode, 2);
2962 if (csum_size)
2963 ext4_initialize_dirent_tail(dir_block, blocksize);
2964
2965 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2966 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2967 if (err)
2968 goto out;
2969 set_buffer_verified(dir_block);
2970out:
2971 brelse(dir_block);
2972 return err;
2973}
2974
2975static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2976 struct dentry *dentry, umode_t mode)
2977{
2978 handle_t *handle;
2979 struct inode *inode;
2980 int err, err2 = 0, credits, retries = 0;
2981
2982 if (EXT4_DIR_LINK_MAX(dir))
2983 return -EMLINK;
2984
2985 err = dquot_initialize(dir);
2986 if (err)
2987 return err;
2988
2989 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2990 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2991retry:
2992 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2993 &dentry->d_name,
2994 0, NULL, EXT4_HT_DIR, credits);
2995 handle = ext4_journal_current_handle();
2996 err = PTR_ERR(inode);
2997 if (IS_ERR(inode))
2998 goto out_stop;
2999
3000 inode->i_op = &ext4_dir_inode_operations;
3001 inode->i_fop = &ext4_dir_operations;
3002 err = ext4_init_new_dir(handle, dir, inode);
3003 if (err)
3004 goto out_clear_inode;
3005 err = ext4_mark_inode_dirty(handle, inode);
3006 if (!err)
3007 err = ext4_add_entry(handle, dentry, inode);
3008 if (err) {
3009out_clear_inode:
3010 clear_nlink(inode);
3011 ext4_orphan_add(handle, inode);
3012 unlock_new_inode(inode);
3013 err2 = ext4_mark_inode_dirty(handle, inode);
3014 if (unlikely(err2))
3015 err = err2;
3016 ext4_journal_stop(handle);
3017 iput(inode);
3018 goto out_retry;
3019 }
3020 ext4_inc_count(dir);
3021
3022 ext4_update_dx_flag(dir);
3023 err = ext4_mark_inode_dirty(handle, dir);
3024 if (err)
3025 goto out_clear_inode;
3026 d_instantiate_new(dentry, inode);
3027 ext4_fc_track_create(handle, dentry);
3028 if (IS_DIRSYNC(dir))
3029 ext4_handle_sync(handle);
3030
3031out_stop:
3032 if (handle)
3033 ext4_journal_stop(handle);
3034out_retry:
3035 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3036 goto retry;
3037 return err;
3038}
3039
3040/*
3041 * routine to check that the specified directory is empty (for rmdir)
3042 */
3043bool ext4_empty_dir(struct inode *inode)
3044{
3045 unsigned int offset;
3046 struct buffer_head *bh;
3047 struct ext4_dir_entry_2 *de;
3048 struct super_block *sb;
3049
3050 if (ext4_has_inline_data(inode)) {
3051 int has_inline_data = 1;
3052 int ret;
3053
3054 ret = empty_inline_dir(inode, &has_inline_data);
3055 if (has_inline_data)
3056 return ret;
3057 }
3058
3059 sb = inode->i_sb;
3060 if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3061 ext4_dir_rec_len(2, NULL)) {
3062 EXT4_ERROR_INODE(inode, "invalid size");
3063 return false;
3064 }
3065 /* The first directory block must not be a hole,
3066 * so treat it as DIRENT_HTREE
3067 */
3068 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3069 if (IS_ERR(bh))
3070 return false;
3071
3072 de = (struct ext4_dir_entry_2 *) bh->b_data;
3073 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3074 0) ||
3075 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3076 ext4_warning_inode(inode, "directory missing '.'");
3077 brelse(bh);
3078 return false;
3079 }
3080 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3081 de = ext4_next_entry(de, sb->s_blocksize);
3082 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3083 offset) ||
3084 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3085 ext4_warning_inode(inode, "directory missing '..'");
3086 brelse(bh);
3087 return false;
3088 }
3089 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3090 while (offset < inode->i_size) {
3091 if (!(offset & (sb->s_blocksize - 1))) {
3092 unsigned int lblock;
3093 brelse(bh);
3094 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3095 bh = ext4_read_dirblock(inode, lblock, EITHER);
3096 if (bh == NULL) {
3097 offset += sb->s_blocksize;
3098 continue;
3099 }
3100 if (IS_ERR(bh))
3101 return false;
3102 }
3103 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3104 (offset & (sb->s_blocksize - 1)));
3105 if (ext4_check_dir_entry(inode, NULL, de, bh,
3106 bh->b_data, bh->b_size, offset) ||
3107 le32_to_cpu(de->inode)) {
3108 brelse(bh);
3109 return false;
3110 }
3111 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3112 }
3113 brelse(bh);
3114 return true;
3115}
3116
3117static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3118{
3119 int retval;
3120 struct inode *inode;
3121 struct buffer_head *bh;
3122 struct ext4_dir_entry_2 *de;
3123 handle_t *handle = NULL;
3124
3125 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3126 return -EIO;
3127
3128 /* Initialize quotas before so that eventual writes go in
3129 * separate transaction */
3130 retval = dquot_initialize(dir);
3131 if (retval)
3132 return retval;
3133 retval = dquot_initialize(d_inode(dentry));
3134 if (retval)
3135 return retval;
3136
3137 retval = -ENOENT;
3138 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3139 if (IS_ERR(bh))
3140 return PTR_ERR(bh);
3141 if (!bh)
3142 goto end_rmdir;
3143
3144 inode = d_inode(dentry);
3145
3146 retval = -EFSCORRUPTED;
3147 if (le32_to_cpu(de->inode) != inode->i_ino)
3148 goto end_rmdir;
3149
3150 retval = -ENOTEMPTY;
3151 if (!ext4_empty_dir(inode))
3152 goto end_rmdir;
3153
3154 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3155 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3156 if (IS_ERR(handle)) {
3157 retval = PTR_ERR(handle);
3158 handle = NULL;
3159 goto end_rmdir;
3160 }
3161
3162 if (IS_DIRSYNC(dir))
3163 ext4_handle_sync(handle);
3164
3165 retval = ext4_delete_entry(handle, dir, de, bh);
3166 if (retval)
3167 goto end_rmdir;
3168 if (!EXT4_DIR_LINK_EMPTY(inode))
3169 ext4_warning_inode(inode,
3170 "empty directory '%.*s' has too many links (%u)",
3171 dentry->d_name.len, dentry->d_name.name,
3172 inode->i_nlink);
3173 inode_inc_iversion(inode);
3174 clear_nlink(inode);
3175 /* There's no need to set i_disksize: the fact that i_nlink is
3176 * zero will ensure that the right thing happens during any
3177 * recovery. */
3178 inode->i_size = 0;
3179 ext4_orphan_add(handle, inode);
3180 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3181 retval = ext4_mark_inode_dirty(handle, inode);
3182 if (retval)
3183 goto end_rmdir;
3184 ext4_dec_count(dir);
3185 ext4_update_dx_flag(dir);
3186 ext4_fc_track_unlink(handle, dentry);
3187 retval = ext4_mark_inode_dirty(handle, dir);
3188
3189#if IS_ENABLED(CONFIG_UNICODE)
3190 /* VFS negative dentries are incompatible with Encoding and
3191 * Case-insensitiveness. Eventually we'll want avoid
3192 * invalidating the dentries here, alongside with returning the
3193 * negative dentries at ext4_lookup(), when it is better
3194 * supported by the VFS for the CI case.
3195 */
3196 if (IS_CASEFOLDED(dir))
3197 d_invalidate(dentry);
3198#endif
3199
3200end_rmdir:
3201 brelse(bh);
3202 if (handle)
3203 ext4_journal_stop(handle);
3204 return retval;
3205}
3206
3207int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3208 struct inode *inode,
3209 struct dentry *dentry /* NULL during fast_commit recovery */)
3210{
3211 int retval = -ENOENT;
3212 struct buffer_head *bh;
3213 struct ext4_dir_entry_2 *de;
3214 handle_t *handle;
3215 int skip_remove_dentry = 0;
3216
3217 /*
3218 * Keep this outside the transaction; it may have to set up the
3219 * directory's encryption key, which isn't GFP_NOFS-safe.
3220 */
3221 bh = ext4_find_entry(dir, d_name, &de, NULL);
3222 if (IS_ERR(bh))
3223 return PTR_ERR(bh);
3224
3225 if (!bh)
3226 return -ENOENT;
3227
3228 if (le32_to_cpu(de->inode) != inode->i_ino) {
3229 /*
3230 * It's okay if we find dont find dentry which matches
3231 * the inode. That's because it might have gotten
3232 * renamed to a different inode number
3233 */
3234 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3235 skip_remove_dentry = 1;
3236 else
3237 goto out_bh;
3238 }
3239
3240 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3241 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3242 if (IS_ERR(handle)) {
3243 retval = PTR_ERR(handle);
3244 goto out_bh;
3245 }
3246
3247 if (IS_DIRSYNC(dir))
3248 ext4_handle_sync(handle);
3249
3250 if (!skip_remove_dentry) {
3251 retval = ext4_delete_entry(handle, dir, de, bh);
3252 if (retval)
3253 goto out_handle;
3254 dir->i_ctime = dir->i_mtime = current_time(dir);
3255 ext4_update_dx_flag(dir);
3256 retval = ext4_mark_inode_dirty(handle, dir);
3257 if (retval)
3258 goto out_handle;
3259 } else {
3260 retval = 0;
3261 }
3262 if (inode->i_nlink == 0)
3263 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3264 d_name->len, d_name->name);
3265 else
3266 drop_nlink(inode);
3267 if (!inode->i_nlink)
3268 ext4_orphan_add(handle, inode);
3269 inode->i_ctime = current_time(inode);
3270 retval = ext4_mark_inode_dirty(handle, inode);
3271 if (dentry && !retval)
3272 ext4_fc_track_unlink(handle, dentry);
3273out_handle:
3274 ext4_journal_stop(handle);
3275out_bh:
3276 brelse(bh);
3277 return retval;
3278}
3279
3280static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3281{
3282 int retval;
3283
3284 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3285 return -EIO;
3286
3287 trace_ext4_unlink_enter(dir, dentry);
3288 /*
3289 * Initialize quotas before so that eventual writes go
3290 * in separate transaction
3291 */
3292 retval = dquot_initialize(dir);
3293 if (retval)
3294 goto out_trace;
3295 retval = dquot_initialize(d_inode(dentry));
3296 if (retval)
3297 goto out_trace;
3298
3299 retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3300#if IS_ENABLED(CONFIG_UNICODE)
3301 /* VFS negative dentries are incompatible with Encoding and
3302 * Case-insensitiveness. Eventually we'll want avoid
3303 * invalidating the dentries here, alongside with returning the
3304 * negative dentries at ext4_lookup(), when it is better
3305 * supported by the VFS for the CI case.
3306 */
3307 if (IS_CASEFOLDED(dir))
3308 d_invalidate(dentry);
3309#endif
3310
3311out_trace:
3312 trace_ext4_unlink_exit(dentry, retval);
3313 return retval;
3314}
3315
3316static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3317 struct fscrypt_str *disk_link)
3318{
3319 struct buffer_head *bh;
3320 char *kaddr;
3321 int err = 0;
3322
3323 bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3324 if (IS_ERR(bh))
3325 return PTR_ERR(bh);
3326
3327 BUFFER_TRACE(bh, "get_write_access");
3328 err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3329 if (err)
3330 goto out;
3331
3332 kaddr = (char *)bh->b_data;
3333 memcpy(kaddr, disk_link->name, disk_link->len);
3334 inode->i_size = disk_link->len - 1;
3335 EXT4_I(inode)->i_disksize = inode->i_size;
3336 err = ext4_handle_dirty_metadata(handle, inode, bh);
3337out:
3338 brelse(bh);
3339 return err;
3340}
3341
3342static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3343 struct dentry *dentry, const char *symname)
3344{
3345 handle_t *handle;
3346 struct inode *inode;
3347 int err, len = strlen(symname);
3348 int credits;
3349 struct fscrypt_str disk_link;
3350 int retries = 0;
3351
3352 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3353 return -EIO;
3354
3355 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3356 &disk_link);
3357 if (err)
3358 return err;
3359
3360 err = dquot_initialize(dir);
3361 if (err)
3362 return err;
3363
3364 /*
3365 * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3366 * directory. +3 for inode, inode bitmap, group descriptor allocation.
3367 * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3368 * modification.
3369 */
3370 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3371 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3372retry:
3373 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3374 &dentry->d_name, 0, NULL,
3375 EXT4_HT_DIR, credits);
3376 handle = ext4_journal_current_handle();
3377 if (IS_ERR(inode)) {
3378 if (handle)
3379 ext4_journal_stop(handle);
3380 err = PTR_ERR(inode);
3381 goto out_retry;
3382 }
3383
3384 if (IS_ENCRYPTED(inode)) {
3385 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3386 if (err)
3387 goto err_drop_inode;
3388 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3389 } else {
3390 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3391 inode->i_op = &ext4_symlink_inode_operations;
3392 } else {
3393 inode->i_op = &ext4_fast_symlink_inode_operations;
3394 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3395 }
3396 }
3397
3398 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3399 /* alloc symlink block and fill it */
3400 err = ext4_init_symlink_block(handle, inode, &disk_link);
3401 if (err)
3402 goto err_drop_inode;
3403 } else {
3404 /* clear the extent format for fast symlink */
3405 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3406 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3407 disk_link.len);
3408 inode->i_size = disk_link.len - 1;
3409 EXT4_I(inode)->i_disksize = inode->i_size;
3410 }
3411 err = ext4_add_nondir(handle, dentry, &inode);
3412 if (handle)
3413 ext4_journal_stop(handle);
3414 iput(inode);
3415 goto out_retry;
3416
3417err_drop_inode:
3418 clear_nlink(inode);
3419 ext4_orphan_add(handle, inode);
3420 unlock_new_inode(inode);
3421 if (handle)
3422 ext4_journal_stop(handle);
3423 iput(inode);
3424out_retry:
3425 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3426 goto retry;
3427 if (disk_link.name != (unsigned char *)symname)
3428 kfree(disk_link.name);
3429 return err;
3430}
3431
3432int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3433{
3434 handle_t *handle;
3435 int err, retries = 0;
3436retry:
3437 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3438 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3439 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3440 if (IS_ERR(handle))
3441 return PTR_ERR(handle);
3442
3443 if (IS_DIRSYNC(dir))
3444 ext4_handle_sync(handle);
3445
3446 inode->i_ctime = current_time(inode);
3447 ext4_inc_count(inode);
3448 ihold(inode);
3449
3450 err = ext4_add_entry(handle, dentry, inode);
3451 if (!err) {
3452 err = ext4_mark_inode_dirty(handle, inode);
3453 /* this can happen only for tmpfile being
3454 * linked the first time
3455 */
3456 if (inode->i_nlink == 1)
3457 ext4_orphan_del(handle, inode);
3458 d_instantiate(dentry, inode);
3459 ext4_fc_track_link(handle, dentry);
3460 } else {
3461 drop_nlink(inode);
3462 iput(inode);
3463 }
3464 ext4_journal_stop(handle);
3465 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3466 goto retry;
3467 return err;
3468}
3469
3470static int ext4_link(struct dentry *old_dentry,
3471 struct inode *dir, struct dentry *dentry)
3472{
3473 struct inode *inode = d_inode(old_dentry);
3474 int err;
3475
3476 if (inode->i_nlink >= EXT4_LINK_MAX)
3477 return -EMLINK;
3478
3479 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3480 if (err)
3481 return err;
3482
3483 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3484 (!projid_eq(EXT4_I(dir)->i_projid,
3485 EXT4_I(old_dentry->d_inode)->i_projid)))
3486 return -EXDEV;
3487
3488 err = dquot_initialize(dir);
3489 if (err)
3490 return err;
3491 return __ext4_link(dir, inode, dentry);
3492}
3493
3494/*
3495 * Try to find buffer head where contains the parent block.
3496 * It should be the inode block if it is inlined or the 1st block
3497 * if it is a normal dir.
3498 */
3499static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3500 struct inode *inode,
3501 int *retval,
3502 struct ext4_dir_entry_2 **parent_de,
3503 int *inlined)
3504{
3505 struct buffer_head *bh;
3506
3507 if (!ext4_has_inline_data(inode)) {
3508 struct ext4_dir_entry_2 *de;
3509 unsigned int offset;
3510
3511 /* The first directory block must not be a hole, so
3512 * treat it as DIRENT_HTREE
3513 */
3514 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3515 if (IS_ERR(bh)) {
3516 *retval = PTR_ERR(bh);
3517 return NULL;
3518 }
3519
3520 de = (struct ext4_dir_entry_2 *) bh->b_data;
3521 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3522 bh->b_size, 0) ||
3523 le32_to_cpu(de->inode) != inode->i_ino ||
3524 strcmp(".", de->name)) {
3525 EXT4_ERROR_INODE(inode, "directory missing '.'");
3526 brelse(bh);
3527 *retval = -EFSCORRUPTED;
3528 return NULL;
3529 }
3530 offset = ext4_rec_len_from_disk(de->rec_len,
3531 inode->i_sb->s_blocksize);
3532 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3533 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3534 bh->b_size, offset) ||
3535 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3536 EXT4_ERROR_INODE(inode, "directory missing '..'");
3537 brelse(bh);
3538 *retval = -EFSCORRUPTED;
3539 return NULL;
3540 }
3541 *parent_de = de;
3542
3543 return bh;
3544 }
3545
3546 *inlined = 1;
3547 return ext4_get_first_inline_block(inode, parent_de, retval);
3548}
3549
3550struct ext4_renament {
3551 struct inode *dir;
3552 struct dentry *dentry;
3553 struct inode *inode;
3554 bool is_dir;
3555 int dir_nlink_delta;
3556
3557 /* entry for "dentry" */
3558 struct buffer_head *bh;
3559 struct ext4_dir_entry_2 *de;
3560 int inlined;
3561
3562 /* entry for ".." in inode if it's a directory */
3563 struct buffer_head *dir_bh;
3564 struct ext4_dir_entry_2 *parent_de;
3565 int dir_inlined;
3566};
3567
3568static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3569{
3570 int retval;
3571
3572 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3573 &retval, &ent->parent_de,
3574 &ent->dir_inlined);
3575 if (!ent->dir_bh)
3576 return retval;
3577 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3578 return -EFSCORRUPTED;
3579 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3580 return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3581 ent->dir_bh, EXT4_JTR_NONE);
3582}
3583
3584static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3585 unsigned dir_ino)
3586{
3587 int retval;
3588
3589 ent->parent_de->inode = cpu_to_le32(dir_ino);
3590 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3591 if (!ent->dir_inlined) {
3592 if (is_dx(ent->inode)) {
3593 retval = ext4_handle_dirty_dx_node(handle,
3594 ent->inode,
3595 ent->dir_bh);
3596 } else {
3597 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3598 ent->dir_bh);
3599 }
3600 } else {
3601 retval = ext4_mark_inode_dirty(handle, ent->inode);
3602 }
3603 if (retval) {
3604 ext4_std_error(ent->dir->i_sb, retval);
3605 return retval;
3606 }
3607 return 0;
3608}
3609
3610static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3611 unsigned ino, unsigned file_type)
3612{
3613 int retval, retval2;
3614
3615 BUFFER_TRACE(ent->bh, "get write access");
3616 retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3617 EXT4_JTR_NONE);
3618 if (retval)
3619 return retval;
3620 ent->de->inode = cpu_to_le32(ino);
3621 if (ext4_has_feature_filetype(ent->dir->i_sb))
3622 ent->de->file_type = file_type;
3623 inode_inc_iversion(ent->dir);
3624 ent->dir->i_ctime = ent->dir->i_mtime =
3625 current_time(ent->dir);
3626 retval = ext4_mark_inode_dirty(handle, ent->dir);
3627 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3628 if (!ent->inlined) {
3629 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3630 if (unlikely(retval2)) {
3631 ext4_std_error(ent->dir->i_sb, retval2);
3632 return retval2;
3633 }
3634 }
3635 return retval;
3636}
3637
3638static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3639 unsigned ino, unsigned file_type)
3640{
3641 struct ext4_renament old = *ent;
3642 int retval = 0;
3643
3644 /*
3645 * old->de could have moved from under us during make indexed dir,
3646 * so the old->de may no longer valid and need to find it again
3647 * before reset old inode info.
3648 */
3649 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3650 if (IS_ERR(old.bh))
3651 retval = PTR_ERR(old.bh);
3652 if (!old.bh)
3653 retval = -ENOENT;
3654 if (retval) {
3655 ext4_std_error(old.dir->i_sb, retval);
3656 return;
3657 }
3658
3659 ext4_setent(handle, &old, ino, file_type);
3660 brelse(old.bh);
3661}
3662
3663static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3664 const struct qstr *d_name)
3665{
3666 int retval = -ENOENT;
3667 struct buffer_head *bh;
3668 struct ext4_dir_entry_2 *de;
3669
3670 bh = ext4_find_entry(dir, d_name, &de, NULL);
3671 if (IS_ERR(bh))
3672 return PTR_ERR(bh);
3673 if (bh) {
3674 retval = ext4_delete_entry(handle, dir, de, bh);
3675 brelse(bh);
3676 }
3677 return retval;
3678}
3679
3680static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3681 int force_reread)
3682{
3683 int retval;
3684 /*
3685 * ent->de could have moved from under us during htree split, so make
3686 * sure that we are deleting the right entry. We might also be pointing
3687 * to a stale entry in the unused part of ent->bh so just checking inum
3688 * and the name isn't enough.
3689 */
3690 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3691 ent->de->name_len != ent->dentry->d_name.len ||
3692 strncmp(ent->de->name, ent->dentry->d_name.name,
3693 ent->de->name_len) ||
3694 force_reread) {
3695 retval = ext4_find_delete_entry(handle, ent->dir,
3696 &ent->dentry->d_name);
3697 } else {
3698 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3699 if (retval == -ENOENT) {
3700 retval = ext4_find_delete_entry(handle, ent->dir,
3701 &ent->dentry->d_name);
3702 }
3703 }
3704
3705 if (retval) {
3706 ext4_warning_inode(ent->dir,
3707 "Deleting old file: nlink %d, error=%d",
3708 ent->dir->i_nlink, retval);
3709 }
3710}
3711
3712static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3713{
3714 if (ent->dir_nlink_delta) {
3715 if (ent->dir_nlink_delta == -1)
3716 ext4_dec_count(ent->dir);
3717 else
3718 ext4_inc_count(ent->dir);
3719 ext4_mark_inode_dirty(handle, ent->dir);
3720 }
3721}
3722
3723static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3724 struct ext4_renament *ent,
3725 int credits, handle_t **h)
3726{
3727 struct inode *wh;
3728 handle_t *handle;
3729 int retries = 0;
3730
3731 /*
3732 * for inode block, sb block, group summaries,
3733 * and inode bitmap
3734 */
3735 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3736 EXT4_XATTR_TRANS_BLOCKS + 4);
3737retry:
3738 wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3739 S_IFCHR | WHITEOUT_MODE,
3740 &ent->dentry->d_name, 0, NULL,
3741 EXT4_HT_DIR, credits);
3742
3743 handle = ext4_journal_current_handle();
3744 if (IS_ERR(wh)) {
3745 if (handle)
3746 ext4_journal_stop(handle);
3747 if (PTR_ERR(wh) == -ENOSPC &&
3748 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3749 goto retry;
3750 } else {
3751 *h = handle;
3752 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3753 wh->i_op = &ext4_special_inode_operations;
3754 }
3755 return wh;
3756}
3757
3758/*
3759 * Anybody can rename anything with this: the permission checks are left to the
3760 * higher-level routines.
3761 *
3762 * n.b. old_{dentry,inode) refers to the source dentry/inode
3763 * while new_{dentry,inode) refers to the destination dentry/inode
3764 * This comes from rename(const char *oldpath, const char *newpath)
3765 */
3766static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3767 struct dentry *old_dentry, struct inode *new_dir,
3768 struct dentry *new_dentry, unsigned int flags)
3769{
3770 handle_t *handle = NULL;
3771 struct ext4_renament old = {
3772 .dir = old_dir,
3773 .dentry = old_dentry,
3774 .inode = d_inode(old_dentry),
3775 };
3776 struct ext4_renament new = {
3777 .dir = new_dir,
3778 .dentry = new_dentry,
3779 .inode = d_inode(new_dentry),
3780 };
3781 int force_reread;
3782 int retval;
3783 struct inode *whiteout = NULL;
3784 int credits;
3785 u8 old_file_type;
3786
3787 if (new.inode && new.inode->i_nlink == 0) {
3788 EXT4_ERROR_INODE(new.inode,
3789 "target of rename is already freed");
3790 return -EFSCORRUPTED;
3791 }
3792
3793 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3794 (!projid_eq(EXT4_I(new_dir)->i_projid,
3795 EXT4_I(old_dentry->d_inode)->i_projid)))
3796 return -EXDEV;
3797
3798 retval = dquot_initialize(old.dir);
3799 if (retval)
3800 return retval;
3801 retval = dquot_initialize(old.inode);
3802 if (retval)
3803 return retval;
3804 retval = dquot_initialize(new.dir);
3805 if (retval)
3806 return retval;
3807
3808 /* Initialize quotas before so that eventual writes go
3809 * in separate transaction */
3810 if (new.inode) {
3811 retval = dquot_initialize(new.inode);
3812 if (retval)
3813 return retval;
3814 }
3815
3816 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3817 if (IS_ERR(old.bh))
3818 return PTR_ERR(old.bh);
3819 /*
3820 * Check for inode number is _not_ due to possible IO errors.
3821 * We might rmdir the source, keep it as pwd of some process
3822 * and merrily kill the link to whatever was created under the
3823 * same name. Goodbye sticky bit ;-<
3824 */
3825 retval = -ENOENT;
3826 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3827 goto release_bh;
3828
3829 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3830 &new.de, &new.inlined);
3831 if (IS_ERR(new.bh)) {
3832 retval = PTR_ERR(new.bh);
3833 new.bh = NULL;
3834 goto release_bh;
3835 }
3836 if (new.bh) {
3837 if (!new.inode) {
3838 brelse(new.bh);
3839 new.bh = NULL;
3840 }
3841 }
3842 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3843 ext4_alloc_da_blocks(old.inode);
3844
3845 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3846 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3847 if (!(flags & RENAME_WHITEOUT)) {
3848 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3849 if (IS_ERR(handle)) {
3850 retval = PTR_ERR(handle);
3851 goto release_bh;
3852 }
3853 } else {
3854 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3855 if (IS_ERR(whiteout)) {
3856 retval = PTR_ERR(whiteout);
3857 goto release_bh;
3858 }
3859 }
3860
3861 old_file_type = old.de->file_type;
3862 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3863 ext4_handle_sync(handle);
3864
3865 if (S_ISDIR(old.inode->i_mode)) {
3866 if (new.inode) {
3867 retval = -ENOTEMPTY;
3868 if (!ext4_empty_dir(new.inode))
3869 goto end_rename;
3870 } else {
3871 retval = -EMLINK;
3872 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3873 goto end_rename;
3874 }
3875 retval = ext4_rename_dir_prepare(handle, &old);
3876 if (retval)
3877 goto end_rename;
3878 }
3879 /*
3880 * If we're renaming a file within an inline_data dir and adding or
3881 * setting the new dirent causes a conversion from inline_data to
3882 * extents/blockmap, we need to force the dirent delete code to
3883 * re-read the directory, or else we end up trying to delete a dirent
3884 * from what is now the extent tree root (or a block map).
3885 */
3886 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3887 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3888
3889 if (whiteout) {
3890 /*
3891 * Do this before adding a new entry, so the old entry is sure
3892 * to be still pointing to the valid old entry.
3893 */
3894 retval = ext4_setent(handle, &old, whiteout->i_ino,
3895 EXT4_FT_CHRDEV);
3896 if (retval)
3897 goto end_rename;
3898 retval = ext4_mark_inode_dirty(handle, whiteout);
3899 if (unlikely(retval))
3900 goto end_rename;
3901
3902 }
3903 if (!new.bh) {
3904 retval = ext4_add_entry(handle, new.dentry, old.inode);
3905 if (retval)
3906 goto end_rename;
3907 } else {
3908 retval = ext4_setent(handle, &new,
3909 old.inode->i_ino, old_file_type);
3910 if (retval)
3911 goto end_rename;
3912 }
3913 if (force_reread)
3914 force_reread = !ext4_test_inode_flag(new.dir,
3915 EXT4_INODE_INLINE_DATA);
3916
3917 /*
3918 * Like most other Unix systems, set the ctime for inodes on a
3919 * rename.
3920 */
3921 old.inode->i_ctime = current_time(old.inode);
3922 retval = ext4_mark_inode_dirty(handle, old.inode);
3923 if (unlikely(retval))
3924 goto end_rename;
3925
3926 if (!whiteout) {
3927 /*
3928 * ok, that's it
3929 */
3930 ext4_rename_delete(handle, &old, force_reread);
3931 }
3932
3933 if (new.inode) {
3934 ext4_dec_count(new.inode);
3935 new.inode->i_ctime = current_time(new.inode);
3936 }
3937 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3938 ext4_update_dx_flag(old.dir);
3939 if (old.dir_bh) {
3940 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3941 if (retval)
3942 goto end_rename;
3943
3944 ext4_dec_count(old.dir);
3945 if (new.inode) {
3946 /* checked ext4_empty_dir above, can't have another
3947 * parent, ext4_dec_count() won't work for many-linked
3948 * dirs */
3949 clear_nlink(new.inode);
3950 } else {
3951 ext4_inc_count(new.dir);
3952 ext4_update_dx_flag(new.dir);
3953 retval = ext4_mark_inode_dirty(handle, new.dir);
3954 if (unlikely(retval))
3955 goto end_rename;
3956 }
3957 }
3958 retval = ext4_mark_inode_dirty(handle, old.dir);
3959 if (unlikely(retval))
3960 goto end_rename;
3961
3962 if (S_ISDIR(old.inode->i_mode)) {
3963 /*
3964 * We disable fast commits here that's because the
3965 * replay code is not yet capable of changing dot dot
3966 * dirents in directories.
3967 */
3968 ext4_fc_mark_ineligible(old.inode->i_sb,
3969 EXT4_FC_REASON_RENAME_DIR, handle);
3970 } else {
3971 struct super_block *sb = old.inode->i_sb;
3972
3973 if (new.inode)
3974 ext4_fc_track_unlink(handle, new.dentry);
3975 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
3976 !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
3977 !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
3978 __ext4_fc_track_link(handle, old.inode, new.dentry);
3979 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3980 if (whiteout)
3981 __ext4_fc_track_create(handle, whiteout,
3982 old.dentry);
3983 }
3984 }
3985
3986 if (new.inode) {
3987 retval = ext4_mark_inode_dirty(handle, new.inode);
3988 if (unlikely(retval))
3989 goto end_rename;
3990 if (!new.inode->i_nlink)
3991 ext4_orphan_add(handle, new.inode);
3992 }
3993 retval = 0;
3994
3995end_rename:
3996 if (whiteout) {
3997 if (retval) {
3998 ext4_resetent(handle, &old,
3999 old.inode->i_ino, old_file_type);
4000 drop_nlink(whiteout);
4001 ext4_orphan_add(handle, whiteout);
4002 }
4003 unlock_new_inode(whiteout);
4004 ext4_journal_stop(handle);
4005 iput(whiteout);
4006 } else {
4007 ext4_journal_stop(handle);
4008 }
4009release_bh:
4010 brelse(old.dir_bh);
4011 brelse(old.bh);
4012 brelse(new.bh);
4013 return retval;
4014}
4015
4016static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4017 struct inode *new_dir, struct dentry *new_dentry)
4018{
4019 handle_t *handle = NULL;
4020 struct ext4_renament old = {
4021 .dir = old_dir,
4022 .dentry = old_dentry,
4023 .inode = d_inode(old_dentry),
4024 };
4025 struct ext4_renament new = {
4026 .dir = new_dir,
4027 .dentry = new_dentry,
4028 .inode = d_inode(new_dentry),
4029 };
4030 u8 new_file_type;
4031 int retval;
4032 struct timespec64 ctime;
4033
4034 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4035 !projid_eq(EXT4_I(new_dir)->i_projid,
4036 EXT4_I(old_dentry->d_inode)->i_projid)) ||
4037 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4038 !projid_eq(EXT4_I(old_dir)->i_projid,
4039 EXT4_I(new_dentry->d_inode)->i_projid)))
4040 return -EXDEV;
4041
4042 retval = dquot_initialize(old.dir);
4043 if (retval)
4044 return retval;
4045 retval = dquot_initialize(new.dir);
4046 if (retval)
4047 return retval;
4048
4049 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4050 &old.de, &old.inlined);
4051 if (IS_ERR(old.bh))
4052 return PTR_ERR(old.bh);
4053 /*
4054 * Check for inode number is _not_ due to possible IO errors.
4055 * We might rmdir the source, keep it as pwd of some process
4056 * and merrily kill the link to whatever was created under the
4057 * same name. Goodbye sticky bit ;-<
4058 */
4059 retval = -ENOENT;
4060 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4061 goto end_rename;
4062
4063 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4064 &new.de, &new.inlined);
4065 if (IS_ERR(new.bh)) {
4066 retval = PTR_ERR(new.bh);
4067 new.bh = NULL;
4068 goto end_rename;
4069 }
4070
4071 /* RENAME_EXCHANGE case: old *and* new must both exist */
4072 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4073 goto end_rename;
4074
4075 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4076 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4077 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4078 if (IS_ERR(handle)) {
4079 retval = PTR_ERR(handle);
4080 handle = NULL;
4081 goto end_rename;
4082 }
4083
4084 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4085 ext4_handle_sync(handle);
4086
4087 if (S_ISDIR(old.inode->i_mode)) {
4088 old.is_dir = true;
4089 retval = ext4_rename_dir_prepare(handle, &old);
4090 if (retval)
4091 goto end_rename;
4092 }
4093 if (S_ISDIR(new.inode->i_mode)) {
4094 new.is_dir = true;
4095 retval = ext4_rename_dir_prepare(handle, &new);
4096 if (retval)
4097 goto end_rename;
4098 }
4099
4100 /*
4101 * Other than the special case of overwriting a directory, parents'
4102 * nlink only needs to be modified if this is a cross directory rename.
4103 */
4104 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4105 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4106 new.dir_nlink_delta = -old.dir_nlink_delta;
4107 retval = -EMLINK;
4108 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4109 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4110 goto end_rename;
4111 }
4112
4113 new_file_type = new.de->file_type;
4114 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4115 if (retval)
4116 goto end_rename;
4117
4118 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4119 if (retval)
4120 goto end_rename;
4121
4122 /*
4123 * Like most other Unix systems, set the ctime for inodes on a
4124 * rename.
4125 */
4126 ctime = current_time(old.inode);
4127 old.inode->i_ctime = ctime;
4128 new.inode->i_ctime = ctime;
4129 retval = ext4_mark_inode_dirty(handle, old.inode);
4130 if (unlikely(retval))
4131 goto end_rename;
4132 retval = ext4_mark_inode_dirty(handle, new.inode);
4133 if (unlikely(retval))
4134 goto end_rename;
4135 ext4_fc_mark_ineligible(new.inode->i_sb,
4136 EXT4_FC_REASON_CROSS_RENAME, handle);
4137 if (old.dir_bh) {
4138 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4139 if (retval)
4140 goto end_rename;
4141 }
4142 if (new.dir_bh) {
4143 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4144 if (retval)
4145 goto end_rename;
4146 }
4147 ext4_update_dir_count(handle, &old);
4148 ext4_update_dir_count(handle, &new);
4149 retval = 0;
4150
4151end_rename:
4152 brelse(old.dir_bh);
4153 brelse(new.dir_bh);
4154 brelse(old.bh);
4155 brelse(new.bh);
4156 if (handle)
4157 ext4_journal_stop(handle);
4158 return retval;
4159}
4160
4161static int ext4_rename2(struct user_namespace *mnt_userns,
4162 struct inode *old_dir, struct dentry *old_dentry,
4163 struct inode *new_dir, struct dentry *new_dentry,
4164 unsigned int flags)
4165{
4166 int err;
4167
4168 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4169 return -EIO;
4170
4171 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4172 return -EINVAL;
4173
4174 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4175 flags);
4176 if (err)
4177 return err;
4178
4179 if (flags & RENAME_EXCHANGE) {
4180 return ext4_cross_rename(old_dir, old_dentry,
4181 new_dir, new_dentry);
4182 }
4183
4184 return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4185}
4186
4187/*
4188 * directories can handle most operations...
4189 */
4190const struct inode_operations ext4_dir_inode_operations = {
4191 .create = ext4_create,
4192 .lookup = ext4_lookup,
4193 .link = ext4_link,
4194 .unlink = ext4_unlink,
4195 .symlink = ext4_symlink,
4196 .mkdir = ext4_mkdir,
4197 .rmdir = ext4_rmdir,
4198 .mknod = ext4_mknod,
4199 .tmpfile = ext4_tmpfile,
4200 .rename = ext4_rename2,
4201 .setattr = ext4_setattr,
4202 .getattr = ext4_getattr,
4203 .listxattr = ext4_listxattr,
4204 .get_inode_acl = ext4_get_acl,
4205 .set_acl = ext4_set_acl,
4206 .fiemap = ext4_fiemap,
4207 .fileattr_get = ext4_fileattr_get,
4208 .fileattr_set = ext4_fileattr_set,
4209};
4210
4211const struct inode_operations ext4_special_inode_operations = {
4212 .setattr = ext4_setattr,
4213 .getattr = ext4_getattr,
4214 .listxattr = ext4_listxattr,
4215 .get_inode_acl = ext4_get_acl,
4216 .set_acl = ext4_set_acl,
4217};