spi: pl022: handle EPROBE_DEFER for dma
[linux/fpc-iii.git] / fs / ext4 / namei.c
bloba969ab39f3026187242dbea0881d5ebdb991d978
1 /*
2 * linux/fs/ext4/namei.c
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)
9 * from
11 * linux/fs/minix/namei.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
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
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
36 #include "ext4.h"
37 #include "ext4_jbd2.h"
39 #include "xattr.h"
40 #include "acl.h"
42 #include <trace/events/ext4.h>
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 static struct buffer_head *ext4_append(handle_t *handle,
51 struct inode *inode,
52 ext4_lblk_t *block)
54 struct buffer_head *bh;
55 int err;
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
65 if (IS_ERR(bh))
66 return bh;
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
70 err = ext4_journal_get_write_access(handle, bh);
71 if (err) {
72 brelse(bh);
73 ext4_std_error(inode->i_sb, err);
74 return ERR_PTR(err);
76 return bh;
79 static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
82 typedef enum {
83 EITHER, INDEX, DIRENT
84 } dirblock_type_t;
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 ext4_lblk_t block,
91 dirblock_type_t type,
92 const char *func,
93 unsigned int line)
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
97 int is_dx_block = 0;
99 bh = ext4_bread(NULL, inode, block, 0);
100 if (IS_ERR(bh)) {
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
107 return bh;
109 if (!bh) {
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED);
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
116 if (is_dx(inode)) {
117 if (block == 0)
118 is_dx_block = 1;
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
122 is_dx_block = 1;
124 if (!is_dx_block && type == INDEX) {
125 ext4_error_inode(inode, func, line, block,
126 "directory leaf block found instead of index block");
127 return ERR_PTR(-EFSCORRUPTED);
129 if (!ext4_has_metadata_csum(inode->i_sb) ||
130 buffer_verified(bh))
131 return bh;
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
138 if (is_dx_block && type == INDEX) {
139 if (ext4_dx_csum_verify(inode, dirent))
140 set_buffer_verified(bh);
141 else {
142 ext4_error_inode(inode, func, line, block,
143 "Directory index failed checksum");
144 brelse(bh);
145 return ERR_PTR(-EFSBADCRC);
148 if (!is_dx_block) {
149 if (ext4_dirent_csum_verify(inode, dirent))
150 set_buffer_verified(bh);
151 else {
152 ext4_error_inode(inode, func, line, block,
153 "Directory block failed checksum");
154 brelse(bh);
155 return ERR_PTR(-EFSBADCRC);
158 return bh;
161 #ifndef assert
162 #define assert(test) J_ASSERT(test)
163 #endif
165 #ifdef DX_DEBUG
166 #define dxtrace(command) command
167 #else
168 #define dxtrace(command)
169 #endif
171 struct fake_dirent
173 __le32 inode;
174 __le16 rec_len;
175 u8 name_len;
176 u8 file_type;
179 struct dx_countlimit
181 __le16 limit;
182 __le16 count;
185 struct dx_entry
187 __le32 hash;
188 __le32 block;
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
197 struct dx_root
199 struct fake_dirent dot;
200 char dot_name[4];
201 struct fake_dirent dotdot;
202 char dotdot_name[4];
203 struct dx_root_info
205 __le32 reserved_zero;
206 u8 hash_version;
207 u8 info_length; /* 8 */
208 u8 indirect_levels;
209 u8 unused_flags;
211 info;
212 struct dx_entry entries[0];
215 struct dx_node
217 struct fake_dirent fake;
218 struct dx_entry entries[0];
222 struct dx_frame
224 struct buffer_head *bh;
225 struct dx_entry *entries;
226 struct dx_entry *at;
229 struct dx_map_entry
231 u32 hash;
232 u16 offs;
233 u16 size;
237 * This goes at the end of each htree block.
239 struct dx_tail {
240 u32 dt_reserved;
241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
255 struct inode *dir,
256 struct dx_hash_info *hinfo,
257 struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 unsigned blocksize, struct dx_hash_info *hinfo,
261 struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267 u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 struct dx_frame *frame,
270 struct dx_frame *frames,
271 __u32 *start_hash);
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 struct ext4_filename *fname,
274 struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 struct dentry *dentry, struct inode *inode);
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 unsigned int blocksize)
282 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 t->det_rec_len = ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail), blocksize);
285 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 struct ext4_dir_entry *de)
292 struct ext4_dir_entry_tail *t;
294 #ifdef PARANOID
295 struct ext4_dir_entry *d, *top;
297 d = de;
298 top = (struct ext4_dir_entry *)(((void *)de) +
299 (EXT4_BLOCK_SIZE(inode->i_sb) -
300 sizeof(struct ext4_dir_entry_tail)));
301 while (d < top && d->rec_len)
302 d = (struct ext4_dir_entry *)(((void *)d) +
303 le16_to_cpu(d->rec_len));
305 if (d != top)
306 return NULL;
308 t = (struct ext4_dir_entry_tail *)d;
309 #else
310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
311 #endif
313 if (t->det_reserved_zero1 ||
314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 t->det_reserved_zero2 ||
316 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
317 return NULL;
319 return t;
322 static __le32 ext4_dirent_csum(struct inode *inode,
323 struct ext4_dir_entry *dirent, int size)
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct ext4_inode_info *ei = EXT4_I(inode);
327 __u32 csum;
329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 return cpu_to_le32(csum);
333 #define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
337 unsigned int line)
339 __ext4_warning_inode(inode, func, line,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
345 struct ext4_dir_entry_tail *t;
347 if (!ext4_has_metadata_csum(inode->i_sb))
348 return 1;
350 t = get_dirent_tail(inode, dirent);
351 if (!t) {
352 warn_no_space_for_csum(inode);
353 return 0;
356 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 (void *)t - (void *)dirent))
358 return 0;
360 return 1;
363 static void ext4_dirent_csum_set(struct inode *inode,
364 struct ext4_dir_entry *dirent)
366 struct ext4_dir_entry_tail *t;
368 if (!ext4_has_metadata_csum(inode->i_sb))
369 return;
371 t = get_dirent_tail(inode, dirent);
372 if (!t) {
373 warn_no_space_for_csum(inode);
374 return;
377 t->det_checksum = ext4_dirent_csum(inode, dirent,
378 (void *)t - (void *)dirent);
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
382 struct inode *inode,
383 struct buffer_head *bh)
385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 return ext4_handle_dirty_metadata(handle, inode, bh);
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 struct ext4_dir_entry *dirent,
391 int *offset)
393 struct ext4_dir_entry *dp;
394 struct dx_root_info *root;
395 int count_offset;
397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
398 count_offset = 8;
399 else if (le16_to_cpu(dirent->rec_len) == 12) {
400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 if (le16_to_cpu(dp->rec_len) !=
402 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
403 return NULL;
404 root = (struct dx_root_info *)(((void *)dp + 12));
405 if (root->reserved_zero ||
406 root->info_length != sizeof(struct dx_root_info))
407 return NULL;
408 count_offset = 32;
409 } else
410 return NULL;
412 if (offset)
413 *offset = count_offset;
414 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 int count_offset, int count, struct dx_tail *t)
420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 struct ext4_inode_info *ei = EXT4_I(inode);
422 __u32 csum;
423 __le32 save_csum;
424 int size;
426 size = count_offset + (count * sizeof(struct dx_entry));
427 save_csum = t->dt_checksum;
428 t->dt_checksum = 0;
429 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
431 t->dt_checksum = save_csum;
433 return cpu_to_le32(csum);
436 static int ext4_dx_csum_verify(struct inode *inode,
437 struct ext4_dir_entry *dirent)
439 struct dx_countlimit *c;
440 struct dx_tail *t;
441 int count_offset, limit, count;
443 if (!ext4_has_metadata_csum(inode->i_sb))
444 return 1;
446 c = get_dx_countlimit(inode, dirent, &count_offset);
447 if (!c) {
448 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
449 return 1;
451 limit = le16_to_cpu(c->limit);
452 count = le16_to_cpu(c->count);
453 if (count_offset + (limit * sizeof(struct dx_entry)) >
454 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
455 warn_no_space_for_csum(inode);
456 return 1;
458 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
460 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
461 count, t))
462 return 0;
463 return 1;
466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
468 struct dx_countlimit *c;
469 struct dx_tail *t;
470 int count_offset, limit, count;
472 if (!ext4_has_metadata_csum(inode->i_sb))
473 return;
475 c = get_dx_countlimit(inode, dirent, &count_offset);
476 if (!c) {
477 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
478 return;
480 limit = le16_to_cpu(c->limit);
481 count = le16_to_cpu(c->count);
482 if (count_offset + (limit * sizeof(struct dx_entry)) >
483 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
484 warn_no_space_for_csum(inode);
485 return;
487 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
489 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493 struct inode *inode,
494 struct buffer_head *bh)
496 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
497 return ext4_handle_dirty_metadata(handle, inode, bh);
501 * p is at least 6 bytes before the end of page
503 static inline struct ext4_dir_entry_2 *
504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
506 return (struct ext4_dir_entry_2 *)((char *)p +
507 ext4_rec_len_from_disk(p->rec_len, blocksize));
511 * Future: use high four bits of block for coalesce-on-delete flags
512 * Mask them off for now.
515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
517 return le32_to_cpu(entry->block) & 0x00ffffff;
520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
522 entry->block = cpu_to_le32(value);
525 static inline unsigned dx_get_hash(struct dx_entry *entry)
527 return le32_to_cpu(entry->hash);
530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
532 entry->hash = cpu_to_le32(value);
535 static inline unsigned dx_get_count(struct dx_entry *entries)
537 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
540 static inline unsigned dx_get_limit(struct dx_entry *entries)
542 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
547 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
552 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
557 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
558 EXT4_DIR_REC_LEN(2) - infosize;
560 if (ext4_has_metadata_csum(dir->i_sb))
561 entry_space -= sizeof(struct dx_tail);
562 return entry_space / sizeof(struct dx_entry);
565 static inline unsigned dx_node_limit(struct inode *dir)
567 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
569 if (ext4_has_metadata_csum(dir->i_sb))
570 entry_space -= sizeof(struct dx_tail);
571 return entry_space / sizeof(struct dx_entry);
575 * Debug
577 #ifdef DX_DEBUG
578 static void dx_show_index(char * label, struct dx_entry *entries)
580 int i, n = dx_get_count (entries);
581 printk(KERN_DEBUG "%s index ", label);
582 for (i = 0; i < n; i++) {
583 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
584 0, (unsigned long)dx_get_block(entries + i));
586 printk("\n");
589 struct stats
591 unsigned names;
592 unsigned space;
593 unsigned bcount;
596 static struct stats dx_show_leaf(struct inode *dir,
597 struct dx_hash_info *hinfo,
598 struct ext4_dir_entry_2 *de,
599 int size, int show_names)
601 unsigned names = 0, space = 0;
602 char *base = (char *) de;
603 struct dx_hash_info h = *hinfo;
605 printk("names: ");
606 while ((char *) de < base + size)
608 if (de->inode)
610 if (show_names)
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
613 int len;
614 char *name;
615 struct ext4_str fname_crypto_str
616 = {.name = NULL, .len = 0};
617 int res = 0;
619 name = de->name;
620 len = de->name_len;
621 if (ext4_encrypted_inode(inode))
622 res = ext4_get_encryption_info(dir);
623 if (res) {
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
627 if (ctx == NULL) {
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
630 de->name_len, &h);
631 printk("%*.s:(U)%x.%u ", len,
632 name, h.hash,
633 (unsigned) ((char *) de
634 - base));
635 } else {
636 /* Directory is encrypted */
637 res = ext4_fname_crypto_alloc_buffer(
638 ctx, de->name_len,
639 &fname_crypto_str);
640 if (res < 0) {
641 printk(KERN_WARNING "Error "
642 "allocating crypto "
643 "buffer--skipping "
644 "crypto\n");
645 ctx = NULL;
647 res = ext4_fname_disk_to_usr(ctx, NULL, de,
648 &fname_crypto_str);
649 if (res < 0) {
650 printk(KERN_WARNING "Error "
651 "converting filename "
652 "from disk to usr"
653 "\n");
654 name = "??";
655 len = 2;
656 } else {
657 name = fname_crypto_str.name;
658 len = fname_crypto_str.len;
660 ext4fs_dirhash(de->name, de->name_len,
661 &h);
662 printk("%*.s:(E)%x.%u ", len, name,
663 h.hash, (unsigned) ((char *) de
664 - base));
665 ext4_fname_crypto_free_buffer(
666 &fname_crypto_str);
668 #else
669 int len = de->name_len;
670 char *name = de->name;
671 ext4fs_dirhash(de->name, de->name_len, &h);
672 printk("%*.s:%x.%u ", len, name, h.hash,
673 (unsigned) ((char *) de - base));
674 #endif
676 space += EXT4_DIR_REC_LEN(de->name_len);
677 names++;
679 de = ext4_next_entry(de, size);
681 printk("(%i)\n", names);
682 return (struct stats) { names, space, 1 };
685 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
686 struct dx_entry *entries, int levels)
688 unsigned blocksize = dir->i_sb->s_blocksize;
689 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
690 unsigned bcount = 0;
691 struct buffer_head *bh;
692 printk("%i indexed blocks...\n", count);
693 for (i = 0; i < count; i++, entries++)
695 ext4_lblk_t block = dx_get_block(entries);
696 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
697 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
698 struct stats stats;
699 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
700 bh = ext4_bread(NULL,dir, block, 0);
701 if (!bh || IS_ERR(bh))
702 continue;
703 stats = levels?
704 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
705 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
706 bh->b_data, blocksize, 0);
707 names += stats.names;
708 space += stats.space;
709 bcount += stats.bcount;
710 brelse(bh);
712 if (bcount)
713 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
714 levels ? "" : " ", names, space/bcount,
715 (space/bcount)*100/blocksize);
716 return (struct stats) { names, space, bcount};
718 #endif /* DX_DEBUG */
721 * Probe for a directory leaf block to search.
723 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
724 * error in the directory index, and the caller should fall back to
725 * searching the directory normally. The callers of dx_probe **MUST**
726 * check for this error code, and make sure it never gets reflected
727 * back to userspace.
729 static struct dx_frame *
730 dx_probe(struct ext4_filename *fname, struct inode *dir,
731 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
733 unsigned count, indirect;
734 struct dx_entry *at, *entries, *p, *q, *m;
735 struct dx_root *root;
736 struct dx_frame *frame = frame_in;
737 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
738 u32 hash;
740 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
741 if (IS_ERR(frame->bh))
742 return (struct dx_frame *) frame->bh;
744 root = (struct dx_root *) frame->bh->b_data;
745 if (root->info.hash_version != DX_HASH_TEA &&
746 root->info.hash_version != DX_HASH_HALF_MD4 &&
747 root->info.hash_version != DX_HASH_LEGACY) {
748 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
749 root->info.hash_version);
750 goto fail;
752 if (fname)
753 hinfo = &fname->hinfo;
754 hinfo->hash_version = root->info.hash_version;
755 if (hinfo->hash_version <= DX_HASH_TEA)
756 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
757 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
758 if (fname && fname_name(fname))
759 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
760 hash = hinfo->hash;
762 if (root->info.unused_flags & 1) {
763 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
764 root->info.unused_flags);
765 goto fail;
768 indirect = root->info.indirect_levels;
769 if (indirect > 1) {
770 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
771 root->info.indirect_levels);
772 goto fail;
775 entries = (struct dx_entry *)(((char *)&root->info) +
776 root->info.info_length);
778 if (dx_get_limit(entries) != dx_root_limit(dir,
779 root->info.info_length)) {
780 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
781 dx_get_limit(entries),
782 dx_root_limit(dir, root->info.info_length));
783 goto fail;
786 dxtrace(printk("Look up %x", hash));
787 while (1) {
788 count = dx_get_count(entries);
789 if (!count || count > dx_get_limit(entries)) {
790 ext4_warning_inode(dir,
791 "dx entry: count %u beyond limit %u",
792 count, dx_get_limit(entries));
793 goto fail;
796 p = entries + 1;
797 q = entries + count - 1;
798 while (p <= q) {
799 m = p + (q - p) / 2;
800 dxtrace(printk("."));
801 if (dx_get_hash(m) > hash)
802 q = m - 1;
803 else
804 p = m + 1;
807 if (0) { // linear search cross check
808 unsigned n = count - 1;
809 at = entries;
810 while (n--)
812 dxtrace(printk(","));
813 if (dx_get_hash(++at) > hash)
815 at--;
816 break;
819 assert (at == p - 1);
822 at = p - 1;
823 dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
824 dx_get_block(at)));
825 frame->entries = entries;
826 frame->at = at;
827 if (!indirect--)
828 return frame;
829 frame++;
830 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
831 if (IS_ERR(frame->bh)) {
832 ret_err = (struct dx_frame *) frame->bh;
833 frame->bh = NULL;
834 goto fail;
836 entries = ((struct dx_node *) frame->bh->b_data)->entries;
838 if (dx_get_limit(entries) != dx_node_limit(dir)) {
839 ext4_warning_inode(dir,
840 "dx entry: limit %u != node limit %u",
841 dx_get_limit(entries), dx_node_limit(dir));
842 goto fail;
845 fail:
846 while (frame >= frame_in) {
847 brelse(frame->bh);
848 frame--;
851 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
852 ext4_warning_inode(dir,
853 "Corrupt directory, running e2fsck is recommended");
854 return ret_err;
857 static void dx_release(struct dx_frame *frames)
859 if (frames[0].bh == NULL)
860 return;
862 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
863 brelse(frames[1].bh);
864 brelse(frames[0].bh);
868 * This function increments the frame pointer to search the next leaf
869 * block, and reads in the necessary intervening nodes if the search
870 * should be necessary. Whether or not the search is necessary is
871 * controlled by the hash parameter. If the hash value is even, then
872 * the search is only continued if the next block starts with that
873 * hash value. This is used if we are searching for a specific file.
875 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
877 * This function returns 1 if the caller should continue to search,
878 * or 0 if it should not. If there is an error reading one of the
879 * index blocks, it will a negative error code.
881 * If start_hash is non-null, it will be filled in with the starting
882 * hash of the next page.
884 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
885 struct dx_frame *frame,
886 struct dx_frame *frames,
887 __u32 *start_hash)
889 struct dx_frame *p;
890 struct buffer_head *bh;
891 int num_frames = 0;
892 __u32 bhash;
894 p = frame;
896 * Find the next leaf page by incrementing the frame pointer.
897 * If we run out of entries in the interior node, loop around and
898 * increment pointer in the parent node. When we break out of
899 * this loop, num_frames indicates the number of interior
900 * nodes need to be read.
902 while (1) {
903 if (++(p->at) < p->entries + dx_get_count(p->entries))
904 break;
905 if (p == frames)
906 return 0;
907 num_frames++;
908 p--;
912 * If the hash is 1, then continue only if the next page has a
913 * continuation hash of any value. This is used for readdir
914 * handling. Otherwise, check to see if the hash matches the
915 * desired contiuation hash. If it doesn't, return since
916 * there's no point to read in the successive index pages.
918 bhash = dx_get_hash(p->at);
919 if (start_hash)
920 *start_hash = bhash;
921 if ((hash & 1) == 0) {
922 if ((bhash & ~1) != hash)
923 return 0;
926 * If the hash is HASH_NB_ALWAYS, we always go to the next
927 * block so no check is necessary
929 while (num_frames--) {
930 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
931 if (IS_ERR(bh))
932 return PTR_ERR(bh);
933 p++;
934 brelse(p->bh);
935 p->bh = bh;
936 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
938 return 1;
943 * This function fills a red-black tree with information from a
944 * directory block. It returns the number directory entries loaded
945 * into the tree. If there is an error it is returned in err.
947 static int htree_dirblock_to_tree(struct file *dir_file,
948 struct inode *dir, ext4_lblk_t block,
949 struct dx_hash_info *hinfo,
950 __u32 start_hash, __u32 start_minor_hash)
952 struct buffer_head *bh;
953 struct ext4_dir_entry_2 *de, *top;
954 int err = 0, count = 0;
955 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0}, tmp_str;
957 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
958 (unsigned long)block));
959 bh = ext4_read_dirblock(dir, block, DIRENT);
960 if (IS_ERR(bh))
961 return PTR_ERR(bh);
963 de = (struct ext4_dir_entry_2 *) bh->b_data;
964 top = (struct ext4_dir_entry_2 *) ((char *) de +
965 dir->i_sb->s_blocksize -
966 EXT4_DIR_REC_LEN(0));
967 #ifdef CONFIG_EXT4_FS_ENCRYPTION
968 /* Check if the directory is encrypted */
969 if (ext4_encrypted_inode(dir)) {
970 err = ext4_get_encryption_info(dir);
971 if (err < 0) {
972 brelse(bh);
973 return err;
975 err = ext4_fname_crypto_alloc_buffer(dir, EXT4_NAME_LEN,
976 &fname_crypto_str);
977 if (err < 0) {
978 brelse(bh);
979 return err;
982 #endif
983 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
984 if (ext4_check_dir_entry(dir, NULL, de, bh,
985 bh->b_data, bh->b_size,
986 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
987 + ((char *)de - bh->b_data))) {
988 /* silently ignore the rest of the block */
989 break;
991 ext4fs_dirhash(de->name, de->name_len, hinfo);
992 if ((hinfo->hash < start_hash) ||
993 ((hinfo->hash == start_hash) &&
994 (hinfo->minor_hash < start_minor_hash)))
995 continue;
996 if (de->inode == 0)
997 continue;
998 if (!ext4_encrypted_inode(dir)) {
999 tmp_str.name = de->name;
1000 tmp_str.len = de->name_len;
1001 err = ext4_htree_store_dirent(dir_file,
1002 hinfo->hash, hinfo->minor_hash, de,
1003 &tmp_str);
1004 } else {
1005 int save_len = fname_crypto_str.len;
1007 /* Directory is encrypted */
1008 err = ext4_fname_disk_to_usr(dir, hinfo, de,
1009 &fname_crypto_str);
1010 if (err < 0) {
1011 count = err;
1012 goto errout;
1014 err = ext4_htree_store_dirent(dir_file,
1015 hinfo->hash, hinfo->minor_hash, de,
1016 &fname_crypto_str);
1017 fname_crypto_str.len = save_len;
1019 if (err != 0) {
1020 count = err;
1021 goto errout;
1023 count++;
1025 errout:
1026 brelse(bh);
1027 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1028 ext4_fname_crypto_free_buffer(&fname_crypto_str);
1029 #endif
1030 return count;
1035 * This function fills a red-black tree with information from a
1036 * directory. We start scanning the directory in hash order, starting
1037 * at start_hash and start_minor_hash.
1039 * This function returns the number of entries inserted into the tree,
1040 * or a negative error code.
1042 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1043 __u32 start_minor_hash, __u32 *next_hash)
1045 struct dx_hash_info hinfo;
1046 struct ext4_dir_entry_2 *de;
1047 struct dx_frame frames[2], *frame;
1048 struct inode *dir;
1049 ext4_lblk_t block;
1050 int count = 0;
1051 int ret, err;
1052 __u32 hashval;
1053 struct ext4_str tmp_str;
1055 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1056 start_hash, start_minor_hash));
1057 dir = file_inode(dir_file);
1058 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1059 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1060 if (hinfo.hash_version <= DX_HASH_TEA)
1061 hinfo.hash_version +=
1062 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1063 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1064 if (ext4_has_inline_data(dir)) {
1065 int has_inline_data = 1;
1066 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1067 &hinfo, start_hash,
1068 start_minor_hash,
1069 &has_inline_data);
1070 if (has_inline_data) {
1071 *next_hash = ~0;
1072 return count;
1075 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1076 start_hash, start_minor_hash);
1077 *next_hash = ~0;
1078 return count;
1080 hinfo.hash = start_hash;
1081 hinfo.minor_hash = 0;
1082 frame = dx_probe(NULL, dir, &hinfo, frames);
1083 if (IS_ERR(frame))
1084 return PTR_ERR(frame);
1086 /* Add '.' and '..' from the htree header */
1087 if (!start_hash && !start_minor_hash) {
1088 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1089 tmp_str.name = de->name;
1090 tmp_str.len = de->name_len;
1091 err = ext4_htree_store_dirent(dir_file, 0, 0,
1092 de, &tmp_str);
1093 if (err != 0)
1094 goto errout;
1095 count++;
1097 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1098 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1099 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1100 tmp_str.name = de->name;
1101 tmp_str.len = de->name_len;
1102 err = ext4_htree_store_dirent(dir_file, 2, 0,
1103 de, &tmp_str);
1104 if (err != 0)
1105 goto errout;
1106 count++;
1109 while (1) {
1110 block = dx_get_block(frame->at);
1111 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1112 start_hash, start_minor_hash);
1113 if (ret < 0) {
1114 err = ret;
1115 goto errout;
1117 count += ret;
1118 hashval = ~0;
1119 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1120 frame, frames, &hashval);
1121 *next_hash = hashval;
1122 if (ret < 0) {
1123 err = ret;
1124 goto errout;
1127 * Stop if: (a) there are no more entries, or
1128 * (b) we have inserted at least one entry and the
1129 * next hash value is not a continuation
1131 if ((ret == 0) ||
1132 (count && ((hashval & 1) == 0)))
1133 break;
1135 dx_release(frames);
1136 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1137 "next hash: %x\n", count, *next_hash));
1138 return count;
1139 errout:
1140 dx_release(frames);
1141 return (err);
1144 static inline int search_dirblock(struct buffer_head *bh,
1145 struct inode *dir,
1146 struct ext4_filename *fname,
1147 const struct qstr *d_name,
1148 unsigned int offset,
1149 struct ext4_dir_entry_2 **res_dir)
1151 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1152 fname, d_name, offset, res_dir);
1156 * Directory block splitting, compacting
1160 * Create map of hash values, offsets, and sizes, stored at end of block.
1161 * Returns number of entries mapped.
1163 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1164 unsigned blocksize, struct dx_hash_info *hinfo,
1165 struct dx_map_entry *map_tail)
1167 int count = 0;
1168 char *base = (char *) de;
1169 struct dx_hash_info h = *hinfo;
1171 while ((char *) de < base + blocksize) {
1172 if (de->name_len && de->inode) {
1173 ext4fs_dirhash(de->name, de->name_len, &h);
1174 map_tail--;
1175 map_tail->hash = h.hash;
1176 map_tail->offs = ((char *) de - base)>>2;
1177 map_tail->size = le16_to_cpu(de->rec_len);
1178 count++;
1179 cond_resched();
1181 /* XXX: do we need to check rec_len == 0 case? -Chris */
1182 de = ext4_next_entry(de, blocksize);
1184 return count;
1187 /* Sort map by hash value */
1188 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1190 struct dx_map_entry *p, *q, *top = map + count - 1;
1191 int more;
1192 /* Combsort until bubble sort doesn't suck */
1193 while (count > 2) {
1194 count = count*10/13;
1195 if (count - 9 < 2) /* 9, 10 -> 11 */
1196 count = 11;
1197 for (p = top, q = p - count; q >= map; p--, q--)
1198 if (p->hash < q->hash)
1199 swap(*p, *q);
1201 /* Garden variety bubble sort */
1202 do {
1203 more = 0;
1204 q = top;
1205 while (q-- > map) {
1206 if (q[1].hash >= q[0].hash)
1207 continue;
1208 swap(*(q+1), *q);
1209 more = 1;
1211 } while(more);
1214 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1216 struct dx_entry *entries = frame->entries;
1217 struct dx_entry *old = frame->at, *new = old + 1;
1218 int count = dx_get_count(entries);
1220 assert(count < dx_get_limit(entries));
1221 assert(old < entries + count);
1222 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1223 dx_set_hash(new, hash);
1224 dx_set_block(new, block);
1225 dx_set_count(entries, count + 1);
1229 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1231 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1232 * `de != NULL' is guaranteed by caller.
1234 static inline int ext4_match(struct ext4_filename *fname,
1235 struct ext4_dir_entry_2 *de)
1237 const void *name = fname_name(fname);
1238 u32 len = fname_len(fname);
1240 if (!de->inode)
1241 return 0;
1243 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1244 if (unlikely(!name)) {
1245 if (fname->usr_fname->name[0] == '_') {
1246 int ret;
1247 if (de->name_len < 16)
1248 return 0;
1249 ret = memcmp(de->name + de->name_len - 16,
1250 fname->crypto_buf.name + 8, 16);
1251 return (ret == 0) ? 1 : 0;
1253 name = fname->crypto_buf.name;
1254 len = fname->crypto_buf.len;
1256 #endif
1257 if (de->name_len != len)
1258 return 0;
1259 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1263 * Returns 0 if not found, -1 on failure, and 1 on success
1265 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1266 struct inode *dir, struct ext4_filename *fname,
1267 const struct qstr *d_name,
1268 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1270 struct ext4_dir_entry_2 * de;
1271 char * dlimit;
1272 int de_len;
1273 int res;
1275 de = (struct ext4_dir_entry_2 *)search_buf;
1276 dlimit = search_buf + buf_size;
1277 while ((char *) de < dlimit) {
1278 /* this code is executed quadratically often */
1279 /* do minimal checking `by hand' */
1280 if ((char *) de + de->name_len <= dlimit) {
1281 res = ext4_match(fname, de);
1282 if (res < 0) {
1283 res = -1;
1284 goto return_result;
1286 if (res > 0) {
1287 /* found a match - just to be sure, do
1288 * a full check */
1289 if (ext4_check_dir_entry(dir, NULL, de, bh,
1290 bh->b_data,
1291 bh->b_size, offset)) {
1292 res = -1;
1293 goto return_result;
1295 *res_dir = de;
1296 res = 1;
1297 goto return_result;
1301 /* prevent looping on a bad block */
1302 de_len = ext4_rec_len_from_disk(de->rec_len,
1303 dir->i_sb->s_blocksize);
1304 if (de_len <= 0) {
1305 res = -1;
1306 goto return_result;
1308 offset += de_len;
1309 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1312 res = 0;
1313 return_result:
1314 return res;
1317 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1318 struct ext4_dir_entry *de)
1320 struct super_block *sb = dir->i_sb;
1322 if (!is_dx(dir))
1323 return 0;
1324 if (block == 0)
1325 return 1;
1326 if (de->inode == 0 &&
1327 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1328 sb->s_blocksize)
1329 return 1;
1330 return 0;
1334 * ext4_find_entry()
1336 * finds an entry in the specified directory with the wanted name. It
1337 * returns the cache buffer in which the entry was found, and the entry
1338 * itself (as a parameter - res_dir). It does NOT read the inode of the
1339 * entry - you'll have to do that yourself if you want to.
1341 * The returned buffer_head has ->b_count elevated. The caller is expected
1342 * to brelse() it when appropriate.
1344 static struct buffer_head * ext4_find_entry (struct inode *dir,
1345 const struct qstr *d_name,
1346 struct ext4_dir_entry_2 **res_dir,
1347 int *inlined)
1349 struct super_block *sb;
1350 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1351 struct buffer_head *bh, *ret = NULL;
1352 ext4_lblk_t start, block, b;
1353 const u8 *name = d_name->name;
1354 int ra_max = 0; /* Number of bh's in the readahead
1355 buffer, bh_use[] */
1356 int ra_ptr = 0; /* Current index into readahead
1357 buffer */
1358 int num = 0;
1359 ext4_lblk_t nblocks;
1360 int i, namelen, retval;
1361 struct ext4_filename fname;
1363 *res_dir = NULL;
1364 sb = dir->i_sb;
1365 namelen = d_name->len;
1366 if (namelen > EXT4_NAME_LEN)
1367 return NULL;
1369 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1370 if (retval)
1371 return ERR_PTR(retval);
1373 if (ext4_has_inline_data(dir)) {
1374 int has_inline_data = 1;
1375 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1376 &has_inline_data);
1377 if (has_inline_data) {
1378 if (inlined)
1379 *inlined = 1;
1380 goto cleanup_and_exit;
1384 if ((namelen <= 2) && (name[0] == '.') &&
1385 (name[1] == '.' || name[1] == '\0')) {
1387 * "." or ".." will only be in the first block
1388 * NFS may look up ".."; "." should be handled by the VFS
1390 block = start = 0;
1391 nblocks = 1;
1392 goto restart;
1394 if (is_dx(dir)) {
1395 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1397 * On success, or if the error was file not found,
1398 * return. Otherwise, fall back to doing a search the
1399 * old fashioned way.
1401 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1402 goto cleanup_and_exit;
1403 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1404 "falling back\n"));
1406 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1407 start = EXT4_I(dir)->i_dir_start_lookup;
1408 if (start >= nblocks)
1409 start = 0;
1410 block = start;
1411 restart:
1412 do {
1414 * We deal with the read-ahead logic here.
1416 if (ra_ptr >= ra_max) {
1417 /* Refill the readahead buffer */
1418 ra_ptr = 0;
1419 b = block;
1420 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1422 * Terminate if we reach the end of the
1423 * directory and must wrap, or if our
1424 * search has finished at this block.
1426 if (b >= nblocks || (num && block == start)) {
1427 bh_use[ra_max] = NULL;
1428 break;
1430 num++;
1431 bh = ext4_getblk(NULL, dir, b++, 0);
1432 if (IS_ERR(bh)) {
1433 if (ra_max == 0) {
1434 ret = bh;
1435 goto cleanup_and_exit;
1437 break;
1439 bh_use[ra_max] = bh;
1440 if (bh)
1441 ll_rw_block(READ | REQ_META | REQ_PRIO,
1442 1, &bh);
1445 if ((bh = bh_use[ra_ptr++]) == NULL)
1446 goto next;
1447 wait_on_buffer(bh);
1448 if (!buffer_uptodate(bh)) {
1449 /* read error, skip block & hope for the best */
1450 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1451 (unsigned long) block);
1452 brelse(bh);
1453 goto next;
1455 if (!buffer_verified(bh) &&
1456 !is_dx_internal_node(dir, block,
1457 (struct ext4_dir_entry *)bh->b_data) &&
1458 !ext4_dirent_csum_verify(dir,
1459 (struct ext4_dir_entry *)bh->b_data)) {
1460 EXT4_ERROR_INODE(dir, "checksumming directory "
1461 "block %lu", (unsigned long)block);
1462 brelse(bh);
1463 goto next;
1465 set_buffer_verified(bh);
1466 i = search_dirblock(bh, dir, &fname, d_name,
1467 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1468 if (i == 1) {
1469 EXT4_I(dir)->i_dir_start_lookup = block;
1470 ret = bh;
1471 goto cleanup_and_exit;
1472 } else {
1473 brelse(bh);
1474 if (i < 0)
1475 goto cleanup_and_exit;
1477 next:
1478 if (++block >= nblocks)
1479 block = 0;
1480 } while (block != start);
1483 * If the directory has grown while we were searching, then
1484 * search the last part of the directory before giving up.
1486 block = nblocks;
1487 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1488 if (block < nblocks) {
1489 start = 0;
1490 goto restart;
1493 cleanup_and_exit:
1494 /* Clean up the read-ahead blocks */
1495 for (; ra_ptr < ra_max; ra_ptr++)
1496 brelse(bh_use[ra_ptr]);
1497 ext4_fname_free_filename(&fname);
1498 return ret;
1501 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1502 struct ext4_filename *fname,
1503 struct ext4_dir_entry_2 **res_dir)
1505 struct super_block * sb = dir->i_sb;
1506 struct dx_frame frames[2], *frame;
1507 const struct qstr *d_name = fname->usr_fname;
1508 struct buffer_head *bh;
1509 ext4_lblk_t block;
1510 int retval;
1512 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1513 *res_dir = NULL;
1514 #endif
1515 frame = dx_probe(fname, dir, NULL, frames);
1516 if (IS_ERR(frame))
1517 return (struct buffer_head *) frame;
1518 do {
1519 block = dx_get_block(frame->at);
1520 bh = ext4_read_dirblock(dir, block, DIRENT);
1521 if (IS_ERR(bh))
1522 goto errout;
1524 retval = search_dirblock(bh, dir, fname, d_name,
1525 block << EXT4_BLOCK_SIZE_BITS(sb),
1526 res_dir);
1527 if (retval == 1)
1528 goto success;
1529 brelse(bh);
1530 if (retval == -1) {
1531 bh = ERR_PTR(ERR_BAD_DX_DIR);
1532 goto errout;
1535 /* Check to see if we should continue to search */
1536 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1537 frames, NULL);
1538 if (retval < 0) {
1539 ext4_warning_inode(dir,
1540 "error %d reading directory index block",
1541 retval);
1542 bh = ERR_PTR(retval);
1543 goto errout;
1545 } while (retval == 1);
1547 bh = NULL;
1548 errout:
1549 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1550 success:
1551 dx_release(frames);
1552 return bh;
1555 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1557 struct inode *inode;
1558 struct ext4_dir_entry_2 *de;
1559 struct buffer_head *bh;
1561 if (dentry->d_name.len > EXT4_NAME_LEN)
1562 return ERR_PTR(-ENAMETOOLONG);
1564 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1565 if (IS_ERR(bh))
1566 return (struct dentry *) bh;
1567 inode = NULL;
1568 if (bh) {
1569 __u32 ino = le32_to_cpu(de->inode);
1570 brelse(bh);
1571 if (!ext4_valid_inum(dir->i_sb, ino)) {
1572 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1573 return ERR_PTR(-EFSCORRUPTED);
1575 if (unlikely(ino == dir->i_ino)) {
1576 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1577 dentry);
1578 return ERR_PTR(-EFSCORRUPTED);
1580 inode = ext4_iget_normal(dir->i_sb, ino);
1581 if (inode == ERR_PTR(-ESTALE)) {
1582 EXT4_ERROR_INODE(dir,
1583 "deleted inode referenced: %u",
1584 ino);
1585 return ERR_PTR(-EFSCORRUPTED);
1587 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1588 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1589 S_ISLNK(inode->i_mode)) &&
1590 !ext4_is_child_context_consistent_with_parent(dir,
1591 inode)) {
1592 iput(inode);
1593 ext4_warning(inode->i_sb,
1594 "Inconsistent encryption contexts: %lu/%lu\n",
1595 (unsigned long) dir->i_ino,
1596 (unsigned long) inode->i_ino);
1597 return ERR_PTR(-EPERM);
1600 return d_splice_alias(inode, dentry);
1604 struct dentry *ext4_get_parent(struct dentry *child)
1606 __u32 ino;
1607 static const struct qstr dotdot = QSTR_INIT("..", 2);
1608 struct ext4_dir_entry_2 * de;
1609 struct buffer_head *bh;
1611 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1612 if (IS_ERR(bh))
1613 return (struct dentry *) bh;
1614 if (!bh)
1615 return ERR_PTR(-ENOENT);
1616 ino = le32_to_cpu(de->inode);
1617 brelse(bh);
1619 if (!ext4_valid_inum(d_inode(child)->i_sb, ino)) {
1620 EXT4_ERROR_INODE(d_inode(child),
1621 "bad parent inode number: %u", ino);
1622 return ERR_PTR(-EFSCORRUPTED);
1625 return d_obtain_alias(ext4_iget_normal(d_inode(child)->i_sb, ino));
1629 * Move count entries from end of map between two memory locations.
1630 * Returns pointer to last entry moved.
1632 static struct ext4_dir_entry_2 *
1633 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1634 unsigned blocksize)
1636 unsigned rec_len = 0;
1638 while (count--) {
1639 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1640 (from + (map->offs<<2));
1641 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1642 memcpy (to, de, rec_len);
1643 ((struct ext4_dir_entry_2 *) to)->rec_len =
1644 ext4_rec_len_to_disk(rec_len, blocksize);
1645 de->inode = 0;
1646 map++;
1647 to += rec_len;
1649 return (struct ext4_dir_entry_2 *) (to - rec_len);
1653 * Compact each dir entry in the range to the minimal rec_len.
1654 * Returns pointer to last entry in range.
1656 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1658 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1659 unsigned rec_len = 0;
1661 prev = to = de;
1662 while ((char*)de < base + blocksize) {
1663 next = ext4_next_entry(de, blocksize);
1664 if (de->inode && de->name_len) {
1665 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1666 if (de > to)
1667 memmove(to, de, rec_len);
1668 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1669 prev = to;
1670 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1672 de = next;
1674 return prev;
1678 * Split a full leaf block to make room for a new dir entry.
1679 * Allocate a new block, and move entries so that they are approx. equally full.
1680 * Returns pointer to de in block into which the new entry will be inserted.
1682 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1683 struct buffer_head **bh,struct dx_frame *frame,
1684 struct dx_hash_info *hinfo)
1686 unsigned blocksize = dir->i_sb->s_blocksize;
1687 unsigned count, continued;
1688 struct buffer_head *bh2;
1689 ext4_lblk_t newblock;
1690 u32 hash2;
1691 struct dx_map_entry *map;
1692 char *data1 = (*bh)->b_data, *data2;
1693 unsigned split, move, size;
1694 struct ext4_dir_entry_2 *de = NULL, *de2;
1695 struct ext4_dir_entry_tail *t;
1696 int csum_size = 0;
1697 int err = 0, i;
1699 if (ext4_has_metadata_csum(dir->i_sb))
1700 csum_size = sizeof(struct ext4_dir_entry_tail);
1702 bh2 = ext4_append(handle, dir, &newblock);
1703 if (IS_ERR(bh2)) {
1704 brelse(*bh);
1705 *bh = NULL;
1706 return (struct ext4_dir_entry_2 *) bh2;
1709 BUFFER_TRACE(*bh, "get_write_access");
1710 err = ext4_journal_get_write_access(handle, *bh);
1711 if (err)
1712 goto journal_error;
1714 BUFFER_TRACE(frame->bh, "get_write_access");
1715 err = ext4_journal_get_write_access(handle, frame->bh);
1716 if (err)
1717 goto journal_error;
1719 data2 = bh2->b_data;
1721 /* create map in the end of data2 block */
1722 map = (struct dx_map_entry *) (data2 + blocksize);
1723 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1724 blocksize, hinfo, map);
1725 map -= count;
1726 dx_sort_map(map, count);
1727 /* Split the existing block in the middle, size-wise */
1728 size = 0;
1729 move = 0;
1730 for (i = count-1; i >= 0; i--) {
1731 /* is more than half of this entry in 2nd half of the block? */
1732 if (size + map[i].size/2 > blocksize/2)
1733 break;
1734 size += map[i].size;
1735 move++;
1737 /* map index at which we will split */
1738 split = count - move;
1739 hash2 = map[split].hash;
1740 continued = hash2 == map[split - 1].hash;
1741 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1742 (unsigned long)dx_get_block(frame->at),
1743 hash2, split, count-split));
1745 /* Fancy dance to stay within two buffers */
1746 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1747 blocksize);
1748 de = dx_pack_dirents(data1, blocksize);
1749 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1750 (char *) de,
1751 blocksize);
1752 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1753 (char *) de2,
1754 blocksize);
1755 if (csum_size) {
1756 t = EXT4_DIRENT_TAIL(data2, blocksize);
1757 initialize_dirent_tail(t, blocksize);
1759 t = EXT4_DIRENT_TAIL(data1, blocksize);
1760 initialize_dirent_tail(t, blocksize);
1763 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1764 blocksize, 1));
1765 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1766 blocksize, 1));
1768 /* Which block gets the new entry? */
1769 if (hinfo->hash >= hash2) {
1770 swap(*bh, bh2);
1771 de = de2;
1773 dx_insert_block(frame, hash2 + continued, newblock);
1774 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1775 if (err)
1776 goto journal_error;
1777 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1778 if (err)
1779 goto journal_error;
1780 brelse(bh2);
1781 dxtrace(dx_show_index("frame", frame->entries));
1782 return de;
1784 journal_error:
1785 brelse(*bh);
1786 brelse(bh2);
1787 *bh = NULL;
1788 ext4_std_error(dir->i_sb, err);
1789 return ERR_PTR(err);
1792 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1793 struct buffer_head *bh,
1794 void *buf, int buf_size,
1795 struct ext4_filename *fname,
1796 struct ext4_dir_entry_2 **dest_de)
1798 struct ext4_dir_entry_2 *de;
1799 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1800 int nlen, rlen;
1801 unsigned int offset = 0;
1802 char *top;
1803 int res;
1805 de = (struct ext4_dir_entry_2 *)buf;
1806 top = buf + buf_size - reclen;
1807 while ((char *) de <= top) {
1808 if (ext4_check_dir_entry(dir, NULL, de, bh,
1809 buf, buf_size, offset)) {
1810 res = -EFSCORRUPTED;
1811 goto return_result;
1813 /* Provide crypto context and crypto buffer to ext4 match */
1814 res = ext4_match(fname, de);
1815 if (res < 0)
1816 goto return_result;
1817 if (res > 0) {
1818 res = -EEXIST;
1819 goto return_result;
1821 nlen = EXT4_DIR_REC_LEN(de->name_len);
1822 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1823 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1824 break;
1825 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1826 offset += rlen;
1829 if ((char *) de > top)
1830 res = -ENOSPC;
1831 else {
1832 *dest_de = de;
1833 res = 0;
1835 return_result:
1836 return res;
1839 int ext4_insert_dentry(struct inode *dir,
1840 struct inode *inode,
1841 struct ext4_dir_entry_2 *de,
1842 int buf_size,
1843 struct ext4_filename *fname)
1846 int nlen, rlen;
1848 nlen = EXT4_DIR_REC_LEN(de->name_len);
1849 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1850 if (de->inode) {
1851 struct ext4_dir_entry_2 *de1 =
1852 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1853 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1854 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1855 de = de1;
1857 de->file_type = EXT4_FT_UNKNOWN;
1858 de->inode = cpu_to_le32(inode->i_ino);
1859 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1860 de->name_len = fname_len(fname);
1861 memcpy(de->name, fname_name(fname), fname_len(fname));
1862 return 0;
1866 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1867 * it points to a directory entry which is guaranteed to be large
1868 * enough for new directory entry. If de is NULL, then
1869 * add_dirent_to_buf will attempt search the directory block for
1870 * space. It will return -ENOSPC if no space is available, and -EIO
1871 * and -EEXIST if directory entry already exists.
1873 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1874 struct inode *dir,
1875 struct inode *inode, struct ext4_dir_entry_2 *de,
1876 struct buffer_head *bh)
1878 unsigned int blocksize = dir->i_sb->s_blocksize;
1879 int csum_size = 0;
1880 int err;
1882 if (ext4_has_metadata_csum(inode->i_sb))
1883 csum_size = sizeof(struct ext4_dir_entry_tail);
1885 if (!de) {
1886 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1887 blocksize - csum_size, fname, &de);
1888 if (err)
1889 return err;
1891 BUFFER_TRACE(bh, "get_write_access");
1892 err = ext4_journal_get_write_access(handle, bh);
1893 if (err) {
1894 ext4_std_error(dir->i_sb, err);
1895 return err;
1898 /* By now the buffer is marked for journaling. Due to crypto operations,
1899 * the following function call may fail */
1900 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1901 if (err < 0)
1902 return err;
1905 * XXX shouldn't update any times until successful
1906 * completion of syscall, but too many callers depend
1907 * on this.
1909 * XXX similarly, too many callers depend on
1910 * ext4_new_inode() setting the times, but error
1911 * recovery deletes the inode, so the worst that can
1912 * happen is that the times are slightly out of date
1913 * and/or different from the directory change time.
1915 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1916 ext4_update_dx_flag(dir);
1917 dir->i_version++;
1918 ext4_mark_inode_dirty(handle, dir);
1919 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1920 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1921 if (err)
1922 ext4_std_error(dir->i_sb, err);
1923 return 0;
1927 * This converts a one block unindexed directory to a 3 block indexed
1928 * directory, and adds the dentry to the indexed directory.
1930 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1931 struct dentry *dentry,
1932 struct inode *inode, struct buffer_head *bh)
1934 struct inode *dir = d_inode(dentry->d_parent);
1935 struct buffer_head *bh2;
1936 struct dx_root *root;
1937 struct dx_frame frames[2], *frame;
1938 struct dx_entry *entries;
1939 struct ext4_dir_entry_2 *de, *de2;
1940 struct ext4_dir_entry_tail *t;
1941 char *data1, *top;
1942 unsigned len;
1943 int retval;
1944 unsigned blocksize;
1945 ext4_lblk_t block;
1946 struct fake_dirent *fde;
1947 int csum_size = 0;
1949 if (ext4_has_metadata_csum(inode->i_sb))
1950 csum_size = sizeof(struct ext4_dir_entry_tail);
1952 blocksize = dir->i_sb->s_blocksize;
1953 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1954 BUFFER_TRACE(bh, "get_write_access");
1955 retval = ext4_journal_get_write_access(handle, bh);
1956 if (retval) {
1957 ext4_std_error(dir->i_sb, retval);
1958 brelse(bh);
1959 return retval;
1961 root = (struct dx_root *) bh->b_data;
1963 /* The 0th block becomes the root, move the dirents out */
1964 fde = &root->dotdot;
1965 de = (struct ext4_dir_entry_2 *)((char *)fde +
1966 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1967 if ((char *) de >= (((char *) root) + blocksize)) {
1968 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1969 brelse(bh);
1970 return -EFSCORRUPTED;
1972 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1974 /* Allocate new block for the 0th block's dirents */
1975 bh2 = ext4_append(handle, dir, &block);
1976 if (IS_ERR(bh2)) {
1977 brelse(bh);
1978 return PTR_ERR(bh2);
1980 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1981 data1 = bh2->b_data;
1983 memcpy (data1, de, len);
1984 de = (struct ext4_dir_entry_2 *) data1;
1985 top = data1 + len;
1986 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1987 de = de2;
1988 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1989 (char *) de,
1990 blocksize);
1992 if (csum_size) {
1993 t = EXT4_DIRENT_TAIL(data1, blocksize);
1994 initialize_dirent_tail(t, blocksize);
1997 /* Initialize the root; the dot dirents already exist */
1998 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1999 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2000 blocksize);
2001 memset (&root->info, 0, sizeof(root->info));
2002 root->info.info_length = sizeof(root->info);
2003 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2004 entries = root->entries;
2005 dx_set_block(entries, 1);
2006 dx_set_count(entries, 1);
2007 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2009 /* Initialize as for dx_probe */
2010 fname->hinfo.hash_version = root->info.hash_version;
2011 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2012 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2013 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2014 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2016 memset(frames, 0, sizeof(frames));
2017 frame = frames;
2018 frame->entries = entries;
2019 frame->at = entries;
2020 frame->bh = bh;
2021 bh = bh2;
2023 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2024 if (retval)
2025 goto out_frames;
2026 retval = ext4_handle_dirty_dirent_node(handle, dir, bh);
2027 if (retval)
2028 goto out_frames;
2030 de = do_split(handle,dir, &bh, frame, &fname->hinfo);
2031 if (IS_ERR(de)) {
2032 retval = PTR_ERR(de);
2033 goto out_frames;
2035 dx_release(frames);
2037 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2038 brelse(bh);
2039 return retval;
2040 out_frames:
2042 * Even if the block split failed, we have to properly write
2043 * out all the changes we did so far. Otherwise we can end up
2044 * with corrupted filesystem.
2046 ext4_mark_inode_dirty(handle, dir);
2047 dx_release(frames);
2048 return retval;
2052 * ext4_add_entry()
2054 * adds a file entry to the specified directory, using the same
2055 * semantics as ext4_find_entry(). It returns NULL if it failed.
2057 * NOTE!! The inode part of 'de' is left at 0 - which means you
2058 * may not sleep between calling this and putting something into
2059 * the entry, as someone else might have used it while you slept.
2061 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2062 struct inode *inode)
2064 struct inode *dir = d_inode(dentry->d_parent);
2065 struct buffer_head *bh = NULL;
2066 struct ext4_dir_entry_2 *de;
2067 struct ext4_dir_entry_tail *t;
2068 struct super_block *sb;
2069 struct ext4_filename fname;
2070 int retval;
2071 int dx_fallback=0;
2072 unsigned blocksize;
2073 ext4_lblk_t block, blocks;
2074 int csum_size = 0;
2076 if (ext4_has_metadata_csum(inode->i_sb))
2077 csum_size = sizeof(struct ext4_dir_entry_tail);
2079 sb = dir->i_sb;
2080 blocksize = sb->s_blocksize;
2081 if (!dentry->d_name.len)
2082 return -EINVAL;
2084 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2085 if (retval)
2086 return retval;
2088 if (ext4_has_inline_data(dir)) {
2089 retval = ext4_try_add_inline_entry(handle, &fname,
2090 dentry, inode);
2091 if (retval < 0)
2092 goto out;
2093 if (retval == 1) {
2094 retval = 0;
2095 goto out;
2099 if (is_dx(dir)) {
2100 retval = ext4_dx_add_entry(handle, &fname, dentry, inode);
2101 if (!retval || (retval != ERR_BAD_DX_DIR))
2102 goto out;
2103 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2104 dx_fallback++;
2105 ext4_mark_inode_dirty(handle, dir);
2107 blocks = dir->i_size >> sb->s_blocksize_bits;
2108 for (block = 0; block < blocks; block++) {
2109 bh = ext4_read_dirblock(dir, block, DIRENT);
2110 if (IS_ERR(bh)) {
2111 retval = PTR_ERR(bh);
2112 bh = NULL;
2113 goto out;
2115 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2116 NULL, bh);
2117 if (retval != -ENOSPC)
2118 goto out;
2120 if (blocks == 1 && !dx_fallback &&
2121 ext4_has_feature_dir_index(sb)) {
2122 retval = make_indexed_dir(handle, &fname, dentry,
2123 inode, bh);
2124 bh = NULL; /* make_indexed_dir releases bh */
2125 goto out;
2127 brelse(bh);
2129 bh = ext4_append(handle, dir, &block);
2130 if (IS_ERR(bh)) {
2131 retval = PTR_ERR(bh);
2132 bh = NULL;
2133 goto out;
2135 de = (struct ext4_dir_entry_2 *) bh->b_data;
2136 de->inode = 0;
2137 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2139 if (csum_size) {
2140 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2141 initialize_dirent_tail(t, blocksize);
2144 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2145 out:
2146 ext4_fname_free_filename(&fname);
2147 brelse(bh);
2148 if (retval == 0)
2149 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2150 return retval;
2154 * Returns 0 for success, or a negative error value
2156 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2157 struct dentry *dentry, struct inode *inode)
2159 struct dx_frame frames[2], *frame;
2160 struct dx_entry *entries, *at;
2161 struct buffer_head *bh;
2162 struct inode *dir = d_inode(dentry->d_parent);
2163 struct super_block *sb = dir->i_sb;
2164 struct ext4_dir_entry_2 *de;
2165 int err;
2167 frame = dx_probe(fname, dir, NULL, frames);
2168 if (IS_ERR(frame))
2169 return PTR_ERR(frame);
2170 entries = frame->entries;
2171 at = frame->at;
2172 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2173 if (IS_ERR(bh)) {
2174 err = PTR_ERR(bh);
2175 bh = NULL;
2176 goto cleanup;
2179 BUFFER_TRACE(bh, "get_write_access");
2180 err = ext4_journal_get_write_access(handle, bh);
2181 if (err)
2182 goto journal_error;
2184 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2185 if (err != -ENOSPC)
2186 goto cleanup;
2188 /* Block full, should compress but for now just split */
2189 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2190 dx_get_count(entries), dx_get_limit(entries)));
2191 /* Need to split index? */
2192 if (dx_get_count(entries) == dx_get_limit(entries)) {
2193 ext4_lblk_t newblock;
2194 unsigned icount = dx_get_count(entries);
2195 int levels = frame - frames;
2196 struct dx_entry *entries2;
2197 struct dx_node *node2;
2198 struct buffer_head *bh2;
2200 if (levels && (dx_get_count(frames->entries) ==
2201 dx_get_limit(frames->entries))) {
2202 ext4_warning_inode(dir, "Directory index full!");
2203 err = -ENOSPC;
2204 goto cleanup;
2206 bh2 = ext4_append(handle, dir, &newblock);
2207 if (IS_ERR(bh2)) {
2208 err = PTR_ERR(bh2);
2209 goto cleanup;
2211 node2 = (struct dx_node *)(bh2->b_data);
2212 entries2 = node2->entries;
2213 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2214 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2215 sb->s_blocksize);
2216 BUFFER_TRACE(frame->bh, "get_write_access");
2217 err = ext4_journal_get_write_access(handle, frame->bh);
2218 if (err)
2219 goto journal_error;
2220 if (levels) {
2221 unsigned icount1 = icount/2, icount2 = icount - icount1;
2222 unsigned hash2 = dx_get_hash(entries + icount1);
2223 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2224 icount1, icount2));
2226 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2227 err = ext4_journal_get_write_access(handle,
2228 frames[0].bh);
2229 if (err)
2230 goto journal_error;
2232 memcpy((char *) entries2, (char *) (entries + icount1),
2233 icount2 * sizeof(struct dx_entry));
2234 dx_set_count(entries, icount1);
2235 dx_set_count(entries2, icount2);
2236 dx_set_limit(entries2, dx_node_limit(dir));
2238 /* Which index block gets the new entry? */
2239 if (at - entries >= icount1) {
2240 frame->at = at = at - entries - icount1 + entries2;
2241 frame->entries = entries = entries2;
2242 swap(frame->bh, bh2);
2244 dx_insert_block(frames + 0, hash2, newblock);
2245 dxtrace(dx_show_index("node", frames[1].entries));
2246 dxtrace(dx_show_index("node",
2247 ((struct dx_node *) bh2->b_data)->entries));
2248 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2249 if (err)
2250 goto journal_error;
2251 brelse (bh2);
2252 } else {
2253 dxtrace(printk(KERN_DEBUG
2254 "Creating second level index...\n"));
2255 memcpy((char *) entries2, (char *) entries,
2256 icount * sizeof(struct dx_entry));
2257 dx_set_limit(entries2, dx_node_limit(dir));
2259 /* Set up root */
2260 dx_set_count(entries, 1);
2261 dx_set_block(entries + 0, newblock);
2262 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2264 /* Add new access path frame */
2265 frame = frames + 1;
2266 frame->at = at = at - entries + entries2;
2267 frame->entries = entries = entries2;
2268 frame->bh = bh2;
2269 err = ext4_journal_get_write_access(handle,
2270 frame->bh);
2271 if (err)
2272 goto journal_error;
2274 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2275 if (err) {
2276 ext4_std_error(inode->i_sb, err);
2277 goto cleanup;
2280 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2281 if (IS_ERR(de)) {
2282 err = PTR_ERR(de);
2283 goto cleanup;
2285 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2286 goto cleanup;
2288 journal_error:
2289 ext4_std_error(dir->i_sb, err);
2290 cleanup:
2291 brelse(bh);
2292 dx_release(frames);
2293 return err;
2297 * ext4_generic_delete_entry deletes a directory entry by merging it
2298 * with the previous entry
2300 int ext4_generic_delete_entry(handle_t *handle,
2301 struct inode *dir,
2302 struct ext4_dir_entry_2 *de_del,
2303 struct buffer_head *bh,
2304 void *entry_buf,
2305 int buf_size,
2306 int csum_size)
2308 struct ext4_dir_entry_2 *de, *pde;
2309 unsigned int blocksize = dir->i_sb->s_blocksize;
2310 int i;
2312 i = 0;
2313 pde = NULL;
2314 de = (struct ext4_dir_entry_2 *)entry_buf;
2315 while (i < buf_size - csum_size) {
2316 if (ext4_check_dir_entry(dir, NULL, de, bh,
2317 bh->b_data, bh->b_size, i))
2318 return -EFSCORRUPTED;
2319 if (de == de_del) {
2320 if (pde)
2321 pde->rec_len = ext4_rec_len_to_disk(
2322 ext4_rec_len_from_disk(pde->rec_len,
2323 blocksize) +
2324 ext4_rec_len_from_disk(de->rec_len,
2325 blocksize),
2326 blocksize);
2327 else
2328 de->inode = 0;
2329 dir->i_version++;
2330 return 0;
2332 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2333 pde = de;
2334 de = ext4_next_entry(de, blocksize);
2336 return -ENOENT;
2339 static int ext4_delete_entry(handle_t *handle,
2340 struct inode *dir,
2341 struct ext4_dir_entry_2 *de_del,
2342 struct buffer_head *bh)
2344 int err, csum_size = 0;
2346 if (ext4_has_inline_data(dir)) {
2347 int has_inline_data = 1;
2348 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2349 &has_inline_data);
2350 if (has_inline_data)
2351 return err;
2354 if (ext4_has_metadata_csum(dir->i_sb))
2355 csum_size = sizeof(struct ext4_dir_entry_tail);
2357 BUFFER_TRACE(bh, "get_write_access");
2358 err = ext4_journal_get_write_access(handle, bh);
2359 if (unlikely(err))
2360 goto out;
2362 err = ext4_generic_delete_entry(handle, dir, de_del,
2363 bh, bh->b_data,
2364 dir->i_sb->s_blocksize, csum_size);
2365 if (err)
2366 goto out;
2368 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2369 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2370 if (unlikely(err))
2371 goto out;
2373 return 0;
2374 out:
2375 if (err != -ENOENT)
2376 ext4_std_error(dir->i_sb, err);
2377 return err;
2381 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2382 * since this indicates that nlinks count was previously 1.
2384 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2386 inc_nlink(inode);
2387 if (is_dx(inode) && inode->i_nlink > 1) {
2388 /* limit is 16-bit i_links_count */
2389 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2390 set_nlink(inode, 1);
2391 ext4_set_feature_dir_nlink(inode->i_sb);
2397 * If a directory had nlink == 1, then we should let it be 1. This indicates
2398 * directory has >EXT4_LINK_MAX subdirs.
2400 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2402 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2403 drop_nlink(inode);
2407 static int ext4_add_nondir(handle_t *handle,
2408 struct dentry *dentry, struct inode *inode)
2410 int err = ext4_add_entry(handle, dentry, inode);
2411 if (!err) {
2412 ext4_mark_inode_dirty(handle, inode);
2413 unlock_new_inode(inode);
2414 d_instantiate(dentry, inode);
2415 return 0;
2417 drop_nlink(inode);
2418 unlock_new_inode(inode);
2419 iput(inode);
2420 return err;
2424 * By the time this is called, we already have created
2425 * the directory cache entry for the new file, but it
2426 * is so far negative - it has no inode.
2428 * If the create succeeds, we fill in the inode information
2429 * with d_instantiate().
2431 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2432 bool excl)
2434 handle_t *handle;
2435 struct inode *inode;
2436 int err, credits, retries = 0;
2438 err = dquot_initialize(dir);
2439 if (err)
2440 return err;
2442 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2443 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2444 retry:
2445 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2446 NULL, EXT4_HT_DIR, credits);
2447 handle = ext4_journal_current_handle();
2448 err = PTR_ERR(inode);
2449 if (!IS_ERR(inode)) {
2450 inode->i_op = &ext4_file_inode_operations;
2451 inode->i_fop = &ext4_file_operations;
2452 ext4_set_aops(inode);
2453 err = ext4_add_nondir(handle, dentry, inode);
2454 if (!err && IS_DIRSYNC(dir))
2455 ext4_handle_sync(handle);
2457 if (handle)
2458 ext4_journal_stop(handle);
2459 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2460 goto retry;
2461 return err;
2464 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2465 umode_t mode, dev_t rdev)
2467 handle_t *handle;
2468 struct inode *inode;
2469 int err, credits, retries = 0;
2471 err = dquot_initialize(dir);
2472 if (err)
2473 return err;
2475 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2476 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2477 retry:
2478 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2479 NULL, EXT4_HT_DIR, credits);
2480 handle = ext4_journal_current_handle();
2481 err = PTR_ERR(inode);
2482 if (!IS_ERR(inode)) {
2483 init_special_inode(inode, inode->i_mode, rdev);
2484 inode->i_op = &ext4_special_inode_operations;
2485 err = ext4_add_nondir(handle, dentry, inode);
2486 if (!err && IS_DIRSYNC(dir))
2487 ext4_handle_sync(handle);
2489 if (handle)
2490 ext4_journal_stop(handle);
2491 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2492 goto retry;
2493 return err;
2496 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2498 handle_t *handle;
2499 struct inode *inode;
2500 int err, retries = 0;
2502 err = dquot_initialize(dir);
2503 if (err)
2504 return err;
2506 retry:
2507 inode = ext4_new_inode_start_handle(dir, mode,
2508 NULL, 0, NULL,
2509 EXT4_HT_DIR,
2510 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2511 4 + EXT4_XATTR_TRANS_BLOCKS);
2512 handle = ext4_journal_current_handle();
2513 err = PTR_ERR(inode);
2514 if (!IS_ERR(inode)) {
2515 inode->i_op = &ext4_file_inode_operations;
2516 inode->i_fop = &ext4_file_operations;
2517 ext4_set_aops(inode);
2518 d_tmpfile(dentry, inode);
2519 err = ext4_orphan_add(handle, inode);
2520 if (err)
2521 goto err_unlock_inode;
2522 mark_inode_dirty(inode);
2523 unlock_new_inode(inode);
2525 if (handle)
2526 ext4_journal_stop(handle);
2527 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2528 goto retry;
2529 return err;
2530 err_unlock_inode:
2531 ext4_journal_stop(handle);
2532 unlock_new_inode(inode);
2533 return err;
2536 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2537 struct ext4_dir_entry_2 *de,
2538 int blocksize, int csum_size,
2539 unsigned int parent_ino, int dotdot_real_len)
2541 de->inode = cpu_to_le32(inode->i_ino);
2542 de->name_len = 1;
2543 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2544 blocksize);
2545 strcpy(de->name, ".");
2546 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2548 de = ext4_next_entry(de, blocksize);
2549 de->inode = cpu_to_le32(parent_ino);
2550 de->name_len = 2;
2551 if (!dotdot_real_len)
2552 de->rec_len = ext4_rec_len_to_disk(blocksize -
2553 (csum_size + EXT4_DIR_REC_LEN(1)),
2554 blocksize);
2555 else
2556 de->rec_len = ext4_rec_len_to_disk(
2557 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2558 strcpy(de->name, "..");
2559 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2561 return ext4_next_entry(de, blocksize);
2564 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2565 struct inode *inode)
2567 struct buffer_head *dir_block = NULL;
2568 struct ext4_dir_entry_2 *de;
2569 struct ext4_dir_entry_tail *t;
2570 ext4_lblk_t block = 0;
2571 unsigned int blocksize = dir->i_sb->s_blocksize;
2572 int csum_size = 0;
2573 int err;
2575 if (ext4_has_metadata_csum(dir->i_sb))
2576 csum_size = sizeof(struct ext4_dir_entry_tail);
2578 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2579 err = ext4_try_create_inline_dir(handle, dir, inode);
2580 if (err < 0 && err != -ENOSPC)
2581 goto out;
2582 if (!err)
2583 goto out;
2586 inode->i_size = 0;
2587 dir_block = ext4_append(handle, inode, &block);
2588 if (IS_ERR(dir_block))
2589 return PTR_ERR(dir_block);
2590 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2591 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2592 set_nlink(inode, 2);
2593 if (csum_size) {
2594 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2595 initialize_dirent_tail(t, blocksize);
2598 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2599 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2600 if (err)
2601 goto out;
2602 set_buffer_verified(dir_block);
2603 out:
2604 brelse(dir_block);
2605 return err;
2608 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2610 handle_t *handle;
2611 struct inode *inode;
2612 int err, credits, retries = 0;
2614 if (EXT4_DIR_LINK_MAX(dir))
2615 return -EMLINK;
2617 err = dquot_initialize(dir);
2618 if (err)
2619 return err;
2621 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2622 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2623 retry:
2624 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2625 &dentry->d_name,
2626 0, NULL, EXT4_HT_DIR, credits);
2627 handle = ext4_journal_current_handle();
2628 err = PTR_ERR(inode);
2629 if (IS_ERR(inode))
2630 goto out_stop;
2632 inode->i_op = &ext4_dir_inode_operations;
2633 inode->i_fop = &ext4_dir_operations;
2634 err = ext4_init_new_dir(handle, dir, inode);
2635 if (err)
2636 goto out_clear_inode;
2637 err = ext4_mark_inode_dirty(handle, inode);
2638 if (!err)
2639 err = ext4_add_entry(handle, dentry, inode);
2640 if (err) {
2641 out_clear_inode:
2642 clear_nlink(inode);
2643 unlock_new_inode(inode);
2644 ext4_mark_inode_dirty(handle, inode);
2645 iput(inode);
2646 goto out_stop;
2648 ext4_inc_count(handle, dir);
2649 ext4_update_dx_flag(dir);
2650 err = ext4_mark_inode_dirty(handle, dir);
2651 if (err)
2652 goto out_clear_inode;
2653 unlock_new_inode(inode);
2654 d_instantiate(dentry, inode);
2655 if (IS_DIRSYNC(dir))
2656 ext4_handle_sync(handle);
2658 out_stop:
2659 if (handle)
2660 ext4_journal_stop(handle);
2661 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2662 goto retry;
2663 return err;
2667 * routine to check that the specified directory is empty (for rmdir)
2669 int ext4_empty_dir(struct inode *inode)
2671 unsigned int offset;
2672 struct buffer_head *bh;
2673 struct ext4_dir_entry_2 *de, *de1;
2674 struct super_block *sb;
2675 int err = 0;
2677 if (ext4_has_inline_data(inode)) {
2678 int has_inline_data = 1;
2680 err = empty_inline_dir(inode, &has_inline_data);
2681 if (has_inline_data)
2682 return err;
2685 sb = inode->i_sb;
2686 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2687 EXT4_ERROR_INODE(inode, "invalid size");
2688 return 1;
2690 bh = ext4_read_dirblock(inode, 0, EITHER);
2691 if (IS_ERR(bh))
2692 return 1;
2694 de = (struct ext4_dir_entry_2 *) bh->b_data;
2695 de1 = ext4_next_entry(de, sb->s_blocksize);
2696 if (le32_to_cpu(de->inode) != inode->i_ino ||
2697 le32_to_cpu(de1->inode) == 0 ||
2698 strcmp(".", de->name) || strcmp("..", de1->name)) {
2699 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2700 brelse(bh);
2701 return 1;
2703 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2704 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2705 de = ext4_next_entry(de1, sb->s_blocksize);
2706 while (offset < inode->i_size) {
2707 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2708 unsigned int lblock;
2709 err = 0;
2710 brelse(bh);
2711 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2712 bh = ext4_read_dirblock(inode, lblock, EITHER);
2713 if (IS_ERR(bh))
2714 return 1;
2715 de = (struct ext4_dir_entry_2 *) bh->b_data;
2717 if (ext4_check_dir_entry(inode, NULL, de, bh,
2718 bh->b_data, bh->b_size, offset)) {
2719 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2720 sb->s_blocksize);
2721 offset = (offset | (sb->s_blocksize - 1)) + 1;
2722 continue;
2724 if (le32_to_cpu(de->inode)) {
2725 brelse(bh);
2726 return 0;
2728 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2729 de = ext4_next_entry(de, sb->s_blocksize);
2731 brelse(bh);
2732 return 1;
2736 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2737 * such inodes, starting at the superblock, in case we crash before the
2738 * file is closed/deleted, or in case the inode truncate spans multiple
2739 * transactions and the last transaction is not recovered after a crash.
2741 * At filesystem recovery time, we walk this list deleting unlinked
2742 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2744 * Orphan list manipulation functions must be called under i_mutex unless
2745 * we are just creating the inode or deleting it.
2747 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2749 struct super_block *sb = inode->i_sb;
2750 struct ext4_sb_info *sbi = EXT4_SB(sb);
2751 struct ext4_iloc iloc;
2752 int err = 0, rc;
2753 bool dirty = false;
2755 if (!sbi->s_journal || is_bad_inode(inode))
2756 return 0;
2758 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2759 !mutex_is_locked(&inode->i_mutex));
2761 * Exit early if inode already is on orphan list. This is a big speedup
2762 * since we don't have to contend on the global s_orphan_lock.
2764 if (!list_empty(&EXT4_I(inode)->i_orphan))
2765 return 0;
2768 * Orphan handling is only valid for files with data blocks
2769 * being truncated, or files being unlinked. Note that we either
2770 * hold i_mutex, or the inode can not be referenced from outside,
2771 * so i_nlink should not be bumped due to race
2773 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2774 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2776 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2777 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2778 if (err)
2779 goto out;
2781 err = ext4_reserve_inode_write(handle, inode, &iloc);
2782 if (err)
2783 goto out;
2785 mutex_lock(&sbi->s_orphan_lock);
2787 * Due to previous errors inode may be already a part of on-disk
2788 * orphan list. If so skip on-disk list modification.
2790 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2791 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2792 /* Insert this inode at the head of the on-disk orphan list */
2793 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2794 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2795 dirty = true;
2797 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2798 mutex_unlock(&sbi->s_orphan_lock);
2800 if (dirty) {
2801 err = ext4_handle_dirty_super(handle, sb);
2802 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2803 if (!err)
2804 err = rc;
2805 if (err) {
2807 * We have to remove inode from in-memory list if
2808 * addition to on disk orphan list failed. Stray orphan
2809 * list entries can cause panics at unmount time.
2811 mutex_lock(&sbi->s_orphan_lock);
2812 list_del(&EXT4_I(inode)->i_orphan);
2813 mutex_unlock(&sbi->s_orphan_lock);
2816 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2817 jbd_debug(4, "orphan inode %lu will point to %d\n",
2818 inode->i_ino, NEXT_ORPHAN(inode));
2819 out:
2820 ext4_std_error(sb, err);
2821 return err;
2825 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2826 * of such inodes stored on disk, because it is finally being cleaned up.
2828 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2830 struct list_head *prev;
2831 struct ext4_inode_info *ei = EXT4_I(inode);
2832 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2833 __u32 ino_next;
2834 struct ext4_iloc iloc;
2835 int err = 0;
2837 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2838 return 0;
2840 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2841 !mutex_is_locked(&inode->i_mutex));
2842 /* Do this quick check before taking global s_orphan_lock. */
2843 if (list_empty(&ei->i_orphan))
2844 return 0;
2846 if (handle) {
2847 /* Grab inode buffer early before taking global s_orphan_lock */
2848 err = ext4_reserve_inode_write(handle, inode, &iloc);
2851 mutex_lock(&sbi->s_orphan_lock);
2852 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2854 prev = ei->i_orphan.prev;
2855 list_del_init(&ei->i_orphan);
2857 /* If we're on an error path, we may not have a valid
2858 * transaction handle with which to update the orphan list on
2859 * disk, but we still need to remove the inode from the linked
2860 * list in memory. */
2861 if (!handle || err) {
2862 mutex_unlock(&sbi->s_orphan_lock);
2863 goto out_err;
2866 ino_next = NEXT_ORPHAN(inode);
2867 if (prev == &sbi->s_orphan) {
2868 jbd_debug(4, "superblock will point to %u\n", ino_next);
2869 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2870 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2871 if (err) {
2872 mutex_unlock(&sbi->s_orphan_lock);
2873 goto out_brelse;
2875 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2876 mutex_unlock(&sbi->s_orphan_lock);
2877 err = ext4_handle_dirty_super(handle, inode->i_sb);
2878 } else {
2879 struct ext4_iloc iloc2;
2880 struct inode *i_prev =
2881 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2883 jbd_debug(4, "orphan inode %lu will point to %u\n",
2884 i_prev->i_ino, ino_next);
2885 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2886 if (err) {
2887 mutex_unlock(&sbi->s_orphan_lock);
2888 goto out_brelse;
2890 NEXT_ORPHAN(i_prev) = ino_next;
2891 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2892 mutex_unlock(&sbi->s_orphan_lock);
2894 if (err)
2895 goto out_brelse;
2896 NEXT_ORPHAN(inode) = 0;
2897 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2898 out_err:
2899 ext4_std_error(inode->i_sb, err);
2900 return err;
2902 out_brelse:
2903 brelse(iloc.bh);
2904 goto out_err;
2907 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2909 int retval;
2910 struct inode *inode;
2911 struct buffer_head *bh;
2912 struct ext4_dir_entry_2 *de;
2913 handle_t *handle = NULL;
2915 /* Initialize quotas before so that eventual writes go in
2916 * separate transaction */
2917 retval = dquot_initialize(dir);
2918 if (retval)
2919 return retval;
2920 retval = dquot_initialize(d_inode(dentry));
2921 if (retval)
2922 return retval;
2924 retval = -ENOENT;
2925 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2926 if (IS_ERR(bh))
2927 return PTR_ERR(bh);
2928 if (!bh)
2929 goto end_rmdir;
2931 inode = d_inode(dentry);
2933 retval = -EFSCORRUPTED;
2934 if (le32_to_cpu(de->inode) != inode->i_ino)
2935 goto end_rmdir;
2937 retval = -ENOTEMPTY;
2938 if (!ext4_empty_dir(inode))
2939 goto end_rmdir;
2941 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2942 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2943 if (IS_ERR(handle)) {
2944 retval = PTR_ERR(handle);
2945 handle = NULL;
2946 goto end_rmdir;
2949 if (IS_DIRSYNC(dir))
2950 ext4_handle_sync(handle);
2952 retval = ext4_delete_entry(handle, dir, de, bh);
2953 if (retval)
2954 goto end_rmdir;
2955 if (!EXT4_DIR_LINK_EMPTY(inode))
2956 ext4_warning_inode(inode,
2957 "empty directory '%.*s' has too many links (%u)",
2958 dentry->d_name.len, dentry->d_name.name,
2959 inode->i_nlink);
2960 inode->i_version++;
2961 clear_nlink(inode);
2962 /* There's no need to set i_disksize: the fact that i_nlink is
2963 * zero will ensure that the right thing happens during any
2964 * recovery. */
2965 inode->i_size = 0;
2966 ext4_orphan_add(handle, inode);
2967 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2968 ext4_mark_inode_dirty(handle, inode);
2969 ext4_dec_count(handle, dir);
2970 ext4_update_dx_flag(dir);
2971 ext4_mark_inode_dirty(handle, dir);
2973 end_rmdir:
2974 brelse(bh);
2975 if (handle)
2976 ext4_journal_stop(handle);
2977 return retval;
2980 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2982 int retval;
2983 struct inode *inode;
2984 struct buffer_head *bh;
2985 struct ext4_dir_entry_2 *de;
2986 handle_t *handle = NULL;
2988 trace_ext4_unlink_enter(dir, dentry);
2989 /* Initialize quotas before so that eventual writes go
2990 * in separate transaction */
2991 retval = dquot_initialize(dir);
2992 if (retval)
2993 return retval;
2994 retval = dquot_initialize(d_inode(dentry));
2995 if (retval)
2996 return retval;
2998 retval = -ENOENT;
2999 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3000 if (IS_ERR(bh))
3001 return PTR_ERR(bh);
3002 if (!bh)
3003 goto end_unlink;
3005 inode = d_inode(dentry);
3007 retval = -EFSCORRUPTED;
3008 if (le32_to_cpu(de->inode) != inode->i_ino)
3009 goto end_unlink;
3011 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3012 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3013 if (IS_ERR(handle)) {
3014 retval = PTR_ERR(handle);
3015 handle = NULL;
3016 goto end_unlink;
3019 if (IS_DIRSYNC(dir))
3020 ext4_handle_sync(handle);
3022 if (inode->i_nlink == 0) {
3023 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3024 dentry->d_name.len, dentry->d_name.name);
3025 set_nlink(inode, 1);
3027 retval = ext4_delete_entry(handle, dir, de, bh);
3028 if (retval)
3029 goto end_unlink;
3030 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3031 ext4_update_dx_flag(dir);
3032 ext4_mark_inode_dirty(handle, dir);
3033 drop_nlink(inode);
3034 if (!inode->i_nlink)
3035 ext4_orphan_add(handle, inode);
3036 inode->i_ctime = ext4_current_time(inode);
3037 ext4_mark_inode_dirty(handle, inode);
3039 end_unlink:
3040 brelse(bh);
3041 if (handle)
3042 ext4_journal_stop(handle);
3043 trace_ext4_unlink_exit(dentry, retval);
3044 return retval;
3047 static int ext4_symlink(struct inode *dir,
3048 struct dentry *dentry, const char *symname)
3050 handle_t *handle;
3051 struct inode *inode;
3052 int err, len = strlen(symname);
3053 int credits;
3054 bool encryption_required;
3055 struct ext4_str disk_link;
3056 struct ext4_encrypted_symlink_data *sd = NULL;
3058 disk_link.len = len + 1;
3059 disk_link.name = (char *) symname;
3061 encryption_required = (ext4_encrypted_inode(dir) ||
3062 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3063 if (encryption_required) {
3064 err = ext4_get_encryption_info(dir);
3065 if (err)
3066 return err;
3067 if (ext4_encryption_info(dir) == NULL)
3068 return -EPERM;
3069 disk_link.len = (ext4_fname_encrypted_size(dir, len) +
3070 sizeof(struct ext4_encrypted_symlink_data));
3071 sd = kzalloc(disk_link.len, GFP_KERNEL);
3072 if (!sd)
3073 return -ENOMEM;
3076 if (disk_link.len > dir->i_sb->s_blocksize) {
3077 err = -ENAMETOOLONG;
3078 goto err_free_sd;
3081 err = dquot_initialize(dir);
3082 if (err)
3083 goto err_free_sd;
3085 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3087 * For non-fast symlinks, we just allocate inode and put it on
3088 * orphan list in the first transaction => we need bitmap,
3089 * group descriptor, sb, inode block, quota blocks, and
3090 * possibly selinux xattr blocks.
3092 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3093 EXT4_XATTR_TRANS_BLOCKS;
3094 } else {
3096 * Fast symlink. We have to add entry to directory
3097 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3098 * allocate new inode (bitmap, group descriptor, inode block,
3099 * quota blocks, sb is already counted in previous macros).
3101 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3102 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3105 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3106 &dentry->d_name, 0, NULL,
3107 EXT4_HT_DIR, credits);
3108 handle = ext4_journal_current_handle();
3109 if (IS_ERR(inode)) {
3110 if (handle)
3111 ext4_journal_stop(handle);
3112 err = PTR_ERR(inode);
3113 goto err_free_sd;
3116 if (encryption_required) {
3117 struct qstr istr;
3118 struct ext4_str ostr;
3120 istr.name = (const unsigned char *) symname;
3121 istr.len = len;
3122 ostr.name = sd->encrypted_path;
3123 ostr.len = disk_link.len;
3124 err = ext4_fname_usr_to_disk(inode, &istr, &ostr);
3125 if (err < 0)
3126 goto err_drop_inode;
3127 sd->len = cpu_to_le16(ostr.len);
3128 disk_link.name = (char *) sd;
3129 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3132 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3133 if (!encryption_required)
3134 inode->i_op = &ext4_symlink_inode_operations;
3135 ext4_set_aops(inode);
3137 * We cannot call page_symlink() with transaction started
3138 * because it calls into ext4_write_begin() which can wait
3139 * for transaction commit if we are running out of space
3140 * and thus we deadlock. So we have to stop transaction now
3141 * and restart it when symlink contents is written.
3143 * To keep fs consistent in case of crash, we have to put inode
3144 * to orphan list in the mean time.
3146 drop_nlink(inode);
3147 err = ext4_orphan_add(handle, inode);
3148 ext4_journal_stop(handle);
3149 handle = NULL;
3150 if (err)
3151 goto err_drop_inode;
3152 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3153 if (err)
3154 goto err_drop_inode;
3156 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3157 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3159 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3160 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3161 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3162 if (IS_ERR(handle)) {
3163 err = PTR_ERR(handle);
3164 handle = NULL;
3165 goto err_drop_inode;
3167 set_nlink(inode, 1);
3168 err = ext4_orphan_del(handle, inode);
3169 if (err)
3170 goto err_drop_inode;
3171 } else {
3172 /* clear the extent format for fast symlink */
3173 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3174 if (!encryption_required) {
3175 inode->i_op = &ext4_fast_symlink_inode_operations;
3176 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3178 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3179 disk_link.len);
3180 inode->i_size = disk_link.len - 1;
3182 EXT4_I(inode)->i_disksize = inode->i_size;
3183 err = ext4_add_nondir(handle, dentry, inode);
3184 if (!err && IS_DIRSYNC(dir))
3185 ext4_handle_sync(handle);
3187 if (handle)
3188 ext4_journal_stop(handle);
3189 kfree(sd);
3190 return err;
3191 err_drop_inode:
3192 if (handle)
3193 ext4_journal_stop(handle);
3194 clear_nlink(inode);
3195 unlock_new_inode(inode);
3196 iput(inode);
3197 err_free_sd:
3198 kfree(sd);
3199 return err;
3202 static int ext4_link(struct dentry *old_dentry,
3203 struct inode *dir, struct dentry *dentry)
3205 handle_t *handle;
3206 struct inode *inode = d_inode(old_dentry);
3207 int err, retries = 0;
3209 if (inode->i_nlink >= EXT4_LINK_MAX)
3210 return -EMLINK;
3211 if (ext4_encrypted_inode(dir) &&
3212 !ext4_is_child_context_consistent_with_parent(dir, inode))
3213 return -EPERM;
3214 err = dquot_initialize(dir);
3215 if (err)
3216 return err;
3218 retry:
3219 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3220 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3221 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3222 if (IS_ERR(handle))
3223 return PTR_ERR(handle);
3225 if (IS_DIRSYNC(dir))
3226 ext4_handle_sync(handle);
3228 inode->i_ctime = ext4_current_time(inode);
3229 ext4_inc_count(handle, inode);
3230 ihold(inode);
3232 err = ext4_add_entry(handle, dentry, inode);
3233 if (!err) {
3234 ext4_mark_inode_dirty(handle, inode);
3235 /* this can happen only for tmpfile being
3236 * linked the first time
3238 if (inode->i_nlink == 1)
3239 ext4_orphan_del(handle, inode);
3240 d_instantiate(dentry, inode);
3241 } else {
3242 drop_nlink(inode);
3243 iput(inode);
3245 ext4_journal_stop(handle);
3246 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3247 goto retry;
3248 return err;
3253 * Try to find buffer head where contains the parent block.
3254 * It should be the inode block if it is inlined or the 1st block
3255 * if it is a normal dir.
3257 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3258 struct inode *inode,
3259 int *retval,
3260 struct ext4_dir_entry_2 **parent_de,
3261 int *inlined)
3263 struct buffer_head *bh;
3265 if (!ext4_has_inline_data(inode)) {
3266 bh = ext4_read_dirblock(inode, 0, EITHER);
3267 if (IS_ERR(bh)) {
3268 *retval = PTR_ERR(bh);
3269 return NULL;
3271 *parent_de = ext4_next_entry(
3272 (struct ext4_dir_entry_2 *)bh->b_data,
3273 inode->i_sb->s_blocksize);
3274 return bh;
3277 *inlined = 1;
3278 return ext4_get_first_inline_block(inode, parent_de, retval);
3281 struct ext4_renament {
3282 struct inode *dir;
3283 struct dentry *dentry;
3284 struct inode *inode;
3285 bool is_dir;
3286 int dir_nlink_delta;
3288 /* entry for "dentry" */
3289 struct buffer_head *bh;
3290 struct ext4_dir_entry_2 *de;
3291 int inlined;
3293 /* entry for ".." in inode if it's a directory */
3294 struct buffer_head *dir_bh;
3295 struct ext4_dir_entry_2 *parent_de;
3296 int dir_inlined;
3299 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3301 int retval;
3303 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3304 &retval, &ent->parent_de,
3305 &ent->dir_inlined);
3306 if (!ent->dir_bh)
3307 return retval;
3308 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3309 return -EFSCORRUPTED;
3310 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3311 return ext4_journal_get_write_access(handle, ent->dir_bh);
3314 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3315 unsigned dir_ino)
3317 int retval;
3319 ent->parent_de->inode = cpu_to_le32(dir_ino);
3320 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3321 if (!ent->dir_inlined) {
3322 if (is_dx(ent->inode)) {
3323 retval = ext4_handle_dirty_dx_node(handle,
3324 ent->inode,
3325 ent->dir_bh);
3326 } else {
3327 retval = ext4_handle_dirty_dirent_node(handle,
3328 ent->inode,
3329 ent->dir_bh);
3331 } else {
3332 retval = ext4_mark_inode_dirty(handle, ent->inode);
3334 if (retval) {
3335 ext4_std_error(ent->dir->i_sb, retval);
3336 return retval;
3338 return 0;
3341 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3342 unsigned ino, unsigned file_type)
3344 int retval;
3346 BUFFER_TRACE(ent->bh, "get write access");
3347 retval = ext4_journal_get_write_access(handle, ent->bh);
3348 if (retval)
3349 return retval;
3350 ent->de->inode = cpu_to_le32(ino);
3351 if (ext4_has_feature_filetype(ent->dir->i_sb))
3352 ent->de->file_type = file_type;
3353 ent->dir->i_version++;
3354 ent->dir->i_ctime = ent->dir->i_mtime =
3355 ext4_current_time(ent->dir);
3356 ext4_mark_inode_dirty(handle, ent->dir);
3357 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3358 if (!ent->inlined) {
3359 retval = ext4_handle_dirty_dirent_node(handle,
3360 ent->dir, ent->bh);
3361 if (unlikely(retval)) {
3362 ext4_std_error(ent->dir->i_sb, retval);
3363 return retval;
3366 brelse(ent->bh);
3367 ent->bh = NULL;
3369 return 0;
3372 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3373 const struct qstr *d_name)
3375 int retval = -ENOENT;
3376 struct buffer_head *bh;
3377 struct ext4_dir_entry_2 *de;
3379 bh = ext4_find_entry(dir, d_name, &de, NULL);
3380 if (IS_ERR(bh))
3381 return PTR_ERR(bh);
3382 if (bh) {
3383 retval = ext4_delete_entry(handle, dir, de, bh);
3384 brelse(bh);
3386 return retval;
3389 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3390 int force_reread)
3392 int retval;
3394 * ent->de could have moved from under us during htree split, so make
3395 * sure that we are deleting the right entry. We might also be pointing
3396 * to a stale entry in the unused part of ent->bh so just checking inum
3397 * and the name isn't enough.
3399 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3400 ent->de->name_len != ent->dentry->d_name.len ||
3401 strncmp(ent->de->name, ent->dentry->d_name.name,
3402 ent->de->name_len) ||
3403 force_reread) {
3404 retval = ext4_find_delete_entry(handle, ent->dir,
3405 &ent->dentry->d_name);
3406 } else {
3407 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3408 if (retval == -ENOENT) {
3409 retval = ext4_find_delete_entry(handle, ent->dir,
3410 &ent->dentry->d_name);
3414 if (retval) {
3415 ext4_warning_inode(ent->dir,
3416 "Deleting old file: nlink %d, error=%d",
3417 ent->dir->i_nlink, retval);
3421 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3423 if (ent->dir_nlink_delta) {
3424 if (ent->dir_nlink_delta == -1)
3425 ext4_dec_count(handle, ent->dir);
3426 else
3427 ext4_inc_count(handle, ent->dir);
3428 ext4_mark_inode_dirty(handle, ent->dir);
3432 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3433 int credits, handle_t **h)
3435 struct inode *wh;
3436 handle_t *handle;
3437 int retries = 0;
3440 * for inode block, sb block, group summaries,
3441 * and inode bitmap
3443 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3444 EXT4_XATTR_TRANS_BLOCKS + 4);
3445 retry:
3446 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3447 &ent->dentry->d_name, 0, NULL,
3448 EXT4_HT_DIR, credits);
3450 handle = ext4_journal_current_handle();
3451 if (IS_ERR(wh)) {
3452 if (handle)
3453 ext4_journal_stop(handle);
3454 if (PTR_ERR(wh) == -ENOSPC &&
3455 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3456 goto retry;
3457 } else {
3458 *h = handle;
3459 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3460 wh->i_op = &ext4_special_inode_operations;
3462 return wh;
3466 * Anybody can rename anything with this: the permission checks are left to the
3467 * higher-level routines.
3469 * n.b. old_{dentry,inode) refers to the source dentry/inode
3470 * while new_{dentry,inode) refers to the destination dentry/inode
3471 * This comes from rename(const char *oldpath, const char *newpath)
3473 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3474 struct inode *new_dir, struct dentry *new_dentry,
3475 unsigned int flags)
3477 handle_t *handle = NULL;
3478 struct ext4_renament old = {
3479 .dir = old_dir,
3480 .dentry = old_dentry,
3481 .inode = d_inode(old_dentry),
3483 struct ext4_renament new = {
3484 .dir = new_dir,
3485 .dentry = new_dentry,
3486 .inode = d_inode(new_dentry),
3488 int force_reread;
3489 int retval;
3490 struct inode *whiteout = NULL;
3491 int credits;
3492 u8 old_file_type;
3494 retval = dquot_initialize(old.dir);
3495 if (retval)
3496 return retval;
3497 retval = dquot_initialize(new.dir);
3498 if (retval)
3499 return retval;
3501 /* Initialize quotas before so that eventual writes go
3502 * in separate transaction */
3503 if (new.inode) {
3504 retval = dquot_initialize(new.inode);
3505 if (retval)
3506 return retval;
3509 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3510 if (IS_ERR(old.bh))
3511 return PTR_ERR(old.bh);
3513 * Check for inode number is _not_ due to possible IO errors.
3514 * We might rmdir the source, keep it as pwd of some process
3515 * and merrily kill the link to whatever was created under the
3516 * same name. Goodbye sticky bit ;-<
3518 retval = -ENOENT;
3519 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3520 goto end_rename;
3522 if ((old.dir != new.dir) &&
3523 ext4_encrypted_inode(new.dir) &&
3524 !ext4_is_child_context_consistent_with_parent(new.dir,
3525 old.inode)) {
3526 retval = -EPERM;
3527 goto end_rename;
3530 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3531 &new.de, &new.inlined);
3532 if (IS_ERR(new.bh)) {
3533 retval = PTR_ERR(new.bh);
3534 new.bh = NULL;
3535 goto end_rename;
3537 if (new.bh) {
3538 if (!new.inode) {
3539 brelse(new.bh);
3540 new.bh = NULL;
3543 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3544 ext4_alloc_da_blocks(old.inode);
3546 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3547 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3548 if (!(flags & RENAME_WHITEOUT)) {
3549 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3550 if (IS_ERR(handle)) {
3551 retval = PTR_ERR(handle);
3552 handle = NULL;
3553 goto end_rename;
3555 } else {
3556 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3557 if (IS_ERR(whiteout)) {
3558 retval = PTR_ERR(whiteout);
3559 whiteout = NULL;
3560 goto end_rename;
3564 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3565 ext4_handle_sync(handle);
3567 if (S_ISDIR(old.inode->i_mode)) {
3568 if (new.inode) {
3569 retval = -ENOTEMPTY;
3570 if (!ext4_empty_dir(new.inode))
3571 goto end_rename;
3572 } else {
3573 retval = -EMLINK;
3574 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3575 goto end_rename;
3577 retval = ext4_rename_dir_prepare(handle, &old);
3578 if (retval)
3579 goto end_rename;
3582 * If we're renaming a file within an inline_data dir and adding or
3583 * setting the new dirent causes a conversion from inline_data to
3584 * extents/blockmap, we need to force the dirent delete code to
3585 * re-read the directory, or else we end up trying to delete a dirent
3586 * from what is now the extent tree root (or a block map).
3588 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3589 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3591 old_file_type = old.de->file_type;
3592 if (whiteout) {
3594 * Do this before adding a new entry, so the old entry is sure
3595 * to be still pointing to the valid old entry.
3597 retval = ext4_setent(handle, &old, whiteout->i_ino,
3598 EXT4_FT_CHRDEV);
3599 if (retval)
3600 goto end_rename;
3601 ext4_mark_inode_dirty(handle, whiteout);
3603 if (!new.bh) {
3604 retval = ext4_add_entry(handle, new.dentry, old.inode);
3605 if (retval)
3606 goto end_rename;
3607 } else {
3608 retval = ext4_setent(handle, &new,
3609 old.inode->i_ino, old_file_type);
3610 if (retval)
3611 goto end_rename;
3613 if (force_reread)
3614 force_reread = !ext4_test_inode_flag(new.dir,
3615 EXT4_INODE_INLINE_DATA);
3618 * Like most other Unix systems, set the ctime for inodes on a
3619 * rename.
3621 old.inode->i_ctime = ext4_current_time(old.inode);
3622 ext4_mark_inode_dirty(handle, old.inode);
3624 if (!whiteout) {
3626 * ok, that's it
3628 ext4_rename_delete(handle, &old, force_reread);
3631 if (new.inode) {
3632 ext4_dec_count(handle, new.inode);
3633 new.inode->i_ctime = ext4_current_time(new.inode);
3635 old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3636 ext4_update_dx_flag(old.dir);
3637 if (old.dir_bh) {
3638 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3639 if (retval)
3640 goto end_rename;
3642 ext4_dec_count(handle, old.dir);
3643 if (new.inode) {
3644 /* checked ext4_empty_dir above, can't have another
3645 * parent, ext4_dec_count() won't work for many-linked
3646 * dirs */
3647 clear_nlink(new.inode);
3648 } else {
3649 ext4_inc_count(handle, new.dir);
3650 ext4_update_dx_flag(new.dir);
3651 ext4_mark_inode_dirty(handle, new.dir);
3654 ext4_mark_inode_dirty(handle, old.dir);
3655 if (new.inode) {
3656 ext4_mark_inode_dirty(handle, new.inode);
3657 if (!new.inode->i_nlink)
3658 ext4_orphan_add(handle, new.inode);
3660 retval = 0;
3662 end_rename:
3663 brelse(old.dir_bh);
3664 brelse(old.bh);
3665 brelse(new.bh);
3666 if (whiteout) {
3667 if (retval)
3668 drop_nlink(whiteout);
3669 unlock_new_inode(whiteout);
3670 iput(whiteout);
3672 if (handle)
3673 ext4_journal_stop(handle);
3674 return retval;
3677 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3678 struct inode *new_dir, struct dentry *new_dentry)
3680 handle_t *handle = NULL;
3681 struct ext4_renament old = {
3682 .dir = old_dir,
3683 .dentry = old_dentry,
3684 .inode = d_inode(old_dentry),
3686 struct ext4_renament new = {
3687 .dir = new_dir,
3688 .dentry = new_dentry,
3689 .inode = d_inode(new_dentry),
3691 u8 new_file_type;
3692 int retval;
3694 if ((ext4_encrypted_inode(old_dir) ||
3695 ext4_encrypted_inode(new_dir)) &&
3696 (old_dir != new_dir) &&
3697 (!ext4_is_child_context_consistent_with_parent(new_dir,
3698 old.inode) ||
3699 !ext4_is_child_context_consistent_with_parent(old_dir,
3700 new.inode)))
3701 return -EPERM;
3703 retval = dquot_initialize(old.dir);
3704 if (retval)
3705 return retval;
3706 retval = dquot_initialize(new.dir);
3707 if (retval)
3708 return retval;
3710 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3711 &old.de, &old.inlined);
3712 if (IS_ERR(old.bh))
3713 return PTR_ERR(old.bh);
3715 * Check for inode number is _not_ due to possible IO errors.
3716 * We might rmdir the source, keep it as pwd of some process
3717 * and merrily kill the link to whatever was created under the
3718 * same name. Goodbye sticky bit ;-<
3720 retval = -ENOENT;
3721 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3722 goto end_rename;
3724 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3725 &new.de, &new.inlined);
3726 if (IS_ERR(new.bh)) {
3727 retval = PTR_ERR(new.bh);
3728 new.bh = NULL;
3729 goto end_rename;
3732 /* RENAME_EXCHANGE case: old *and* new must both exist */
3733 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3734 goto end_rename;
3736 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3737 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3738 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3739 if (IS_ERR(handle)) {
3740 retval = PTR_ERR(handle);
3741 handle = NULL;
3742 goto end_rename;
3745 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3746 ext4_handle_sync(handle);
3748 if (S_ISDIR(old.inode->i_mode)) {
3749 old.is_dir = true;
3750 retval = ext4_rename_dir_prepare(handle, &old);
3751 if (retval)
3752 goto end_rename;
3754 if (S_ISDIR(new.inode->i_mode)) {
3755 new.is_dir = true;
3756 retval = ext4_rename_dir_prepare(handle, &new);
3757 if (retval)
3758 goto end_rename;
3762 * Other than the special case of overwriting a directory, parents'
3763 * nlink only needs to be modified if this is a cross directory rename.
3765 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3766 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3767 new.dir_nlink_delta = -old.dir_nlink_delta;
3768 retval = -EMLINK;
3769 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3770 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3771 goto end_rename;
3774 new_file_type = new.de->file_type;
3775 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3776 if (retval)
3777 goto end_rename;
3779 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3780 if (retval)
3781 goto end_rename;
3784 * Like most other Unix systems, set the ctime for inodes on a
3785 * rename.
3787 old.inode->i_ctime = ext4_current_time(old.inode);
3788 new.inode->i_ctime = ext4_current_time(new.inode);
3789 ext4_mark_inode_dirty(handle, old.inode);
3790 ext4_mark_inode_dirty(handle, new.inode);
3792 if (old.dir_bh) {
3793 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3794 if (retval)
3795 goto end_rename;
3797 if (new.dir_bh) {
3798 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3799 if (retval)
3800 goto end_rename;
3802 ext4_update_dir_count(handle, &old);
3803 ext4_update_dir_count(handle, &new);
3804 retval = 0;
3806 end_rename:
3807 brelse(old.dir_bh);
3808 brelse(new.dir_bh);
3809 brelse(old.bh);
3810 brelse(new.bh);
3811 if (handle)
3812 ext4_journal_stop(handle);
3813 return retval;
3816 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3817 struct inode *new_dir, struct dentry *new_dentry,
3818 unsigned int flags)
3820 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3821 return -EINVAL;
3823 if (flags & RENAME_EXCHANGE) {
3824 return ext4_cross_rename(old_dir, old_dentry,
3825 new_dir, new_dentry);
3828 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3832 * directories can handle most operations...
3834 const struct inode_operations ext4_dir_inode_operations = {
3835 .create = ext4_create,
3836 .lookup = ext4_lookup,
3837 .link = ext4_link,
3838 .unlink = ext4_unlink,
3839 .symlink = ext4_symlink,
3840 .mkdir = ext4_mkdir,
3841 .rmdir = ext4_rmdir,
3842 .mknod = ext4_mknod,
3843 .tmpfile = ext4_tmpfile,
3844 .rename2 = ext4_rename2,
3845 .setattr = ext4_setattr,
3846 .setxattr = generic_setxattr,
3847 .getxattr = generic_getxattr,
3848 .listxattr = ext4_listxattr,
3849 .removexattr = generic_removexattr,
3850 .get_acl = ext4_get_acl,
3851 .set_acl = ext4_set_acl,
3852 .fiemap = ext4_fiemap,
3855 const struct inode_operations ext4_special_inode_operations = {
3856 .setattr = ext4_setattr,
3857 .setxattr = generic_setxattr,
3858 .getxattr = generic_getxattr,
3859 .listxattr = ext4_listxattr,
3860 .removexattr = generic_removexattr,
3861 .get_acl = ext4_get_acl,
3862 .set_acl = ext4_set_acl,