1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/xattr.c
5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9 * Extended attributes for symlinks and special files added per
10 * suggestion of Luka Renko <luka.renko@hermes.si>.
11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14 * and Andreas Gruenbacher <agruen@suse.de>.
18 * Extended attributes are stored directly in inodes (on file systems with
19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
20 * field contains the block number if an inode uses an additional block. All
21 * attributes must fit in the inode and one additional block. Blocks that
22 * contain the identical set of attributes may be shared among several inodes.
23 * Identical blocks are detected by keeping a cache of blocks that have
24 * recently been accessed.
26 * The attributes in inodes and on blocks have a different header; the entries
27 * are stored in the same format:
29 * +------------------+
32 * | entry 2 | | growing downwards
37 * | value 3 | | growing upwards
39 * +------------------+
41 * The header is followed by multiple entry descriptors. In disk blocks, the
42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
43 * attribute values are aligned to the end of the block in no specific order.
47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
48 * EA blocks are only changed if they are exclusive to an inode, so
49 * holding xattr_sem also means that nothing but the EA block's reference
50 * count can change. Multiple writers to the same block are synchronized
54 #include <linux/init.h>
56 #include <linux/slab.h>
57 #include <linux/mbcache.h>
58 #include <linux/quotaops.h>
59 #include "ext4_jbd2.h"
64 #ifdef EXT4_XATTR_DEBUG
65 # define ea_idebug(inode, fmt, ...) \
66 printk(KERN_DEBUG "inode %s:%lu: " fmt "\n", \
67 inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
68 # define ea_bdebug(bh, fmt, ...) \
69 printk(KERN_DEBUG "block %pg:%lu: " fmt "\n", \
70 bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
72 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
73 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
76 static void ext4_xattr_block_cache_insert(struct mb_cache
*,
77 struct buffer_head
*);
78 static struct buffer_head
*
79 ext4_xattr_block_cache_find(struct inode
*, struct ext4_xattr_header
*,
80 struct mb_cache_entry
**);
81 static __le32
ext4_xattr_hash_entry(char *name
, size_t name_len
, __le32
*value
,
83 static void ext4_xattr_rehash(struct ext4_xattr_header
*);
85 static const struct xattr_handler
* const ext4_xattr_handler_map
[] = {
86 [EXT4_XATTR_INDEX_USER
] = &ext4_xattr_user_handler
,
87 #ifdef CONFIG_EXT4_FS_POSIX_ACL
88 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS
] = &posix_acl_access_xattr_handler
,
89 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT
] = &posix_acl_default_xattr_handler
,
91 [EXT4_XATTR_INDEX_TRUSTED
] = &ext4_xattr_trusted_handler
,
92 #ifdef CONFIG_EXT4_FS_SECURITY
93 [EXT4_XATTR_INDEX_SECURITY
] = &ext4_xattr_security_handler
,
97 const struct xattr_handler
*ext4_xattr_handlers
[] = {
98 &ext4_xattr_user_handler
,
99 &ext4_xattr_trusted_handler
,
100 #ifdef CONFIG_EXT4_FS_POSIX_ACL
101 &posix_acl_access_xattr_handler
,
102 &posix_acl_default_xattr_handler
,
104 #ifdef CONFIG_EXT4_FS_SECURITY
105 &ext4_xattr_security_handler
,
110 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
111 inode->i_sb->s_fs_info)->s_ea_block_cache)
113 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
114 inode->i_sb->s_fs_info)->s_ea_inode_cache)
117 ext4_expand_inode_array(struct ext4_xattr_inode_array
**ea_inode_array
,
118 struct inode
*inode
);
120 #ifdef CONFIG_LOCKDEP
121 void ext4_xattr_inode_set_class(struct inode
*ea_inode
)
123 lockdep_set_subclass(&ea_inode
->i_rwsem
, 1);
127 static __le32
ext4_xattr_block_csum(struct inode
*inode
,
129 struct ext4_xattr_header
*hdr
)
131 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
133 __le64 dsk_block_nr
= cpu_to_le64(block_nr
);
134 __u32 dummy_csum
= 0;
135 int offset
= offsetof(struct ext4_xattr_header
, h_checksum
);
137 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&dsk_block_nr
,
138 sizeof(dsk_block_nr
));
139 csum
= ext4_chksum(sbi
, csum
, (__u8
*)hdr
, offset
);
140 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, sizeof(dummy_csum
));
141 offset
+= sizeof(dummy_csum
);
142 csum
= ext4_chksum(sbi
, csum
, (__u8
*)hdr
+ offset
,
143 EXT4_BLOCK_SIZE(inode
->i_sb
) - offset
);
145 return cpu_to_le32(csum
);
148 static int ext4_xattr_block_csum_verify(struct inode
*inode
,
149 struct buffer_head
*bh
)
151 struct ext4_xattr_header
*hdr
= BHDR(bh
);
154 if (ext4_has_metadata_csum(inode
->i_sb
)) {
156 ret
= (hdr
->h_checksum
== ext4_xattr_block_csum(inode
,
157 bh
->b_blocknr
, hdr
));
163 static void ext4_xattr_block_csum_set(struct inode
*inode
,
164 struct buffer_head
*bh
)
166 if (ext4_has_metadata_csum(inode
->i_sb
))
167 BHDR(bh
)->h_checksum
= ext4_xattr_block_csum(inode
,
168 bh
->b_blocknr
, BHDR(bh
));
171 static inline const struct xattr_handler
*
172 ext4_xattr_handler(int name_index
)
174 const struct xattr_handler
*handler
= NULL
;
176 if (name_index
> 0 && name_index
< ARRAY_SIZE(ext4_xattr_handler_map
))
177 handler
= ext4_xattr_handler_map
[name_index
];
182 ext4_xattr_check_entries(struct ext4_xattr_entry
*entry
, void *end
,
185 struct ext4_xattr_entry
*e
= entry
;
187 /* Find the end of the names list */
188 while (!IS_LAST_ENTRY(e
)) {
189 struct ext4_xattr_entry
*next
= EXT4_XATTR_NEXT(e
);
190 if ((void *)next
>= end
)
191 return -EFSCORRUPTED
;
192 if (strnlen(e
->e_name
, e
->e_name_len
) != e
->e_name_len
)
193 return -EFSCORRUPTED
;
197 /* Check the values */
198 while (!IS_LAST_ENTRY(entry
)) {
199 u32 size
= le32_to_cpu(entry
->e_value_size
);
201 if (size
> EXT4_XATTR_SIZE_MAX
)
202 return -EFSCORRUPTED
;
204 if (size
!= 0 && entry
->e_value_inum
== 0) {
205 u16 offs
= le16_to_cpu(entry
->e_value_offs
);
209 * The value cannot overlap the names, and the value
210 * with padding cannot extend beyond 'end'. Check both
211 * the padded and unpadded sizes, since the size may
212 * overflow to 0 when adding padding.
214 if (offs
> end
- value_start
)
215 return -EFSCORRUPTED
;
216 value
= value_start
+ offs
;
217 if (value
< (void *)e
+ sizeof(u32
) ||
218 size
> end
- value
||
219 EXT4_XATTR_SIZE(size
) > end
- value
)
220 return -EFSCORRUPTED
;
222 entry
= EXT4_XATTR_NEXT(entry
);
229 __ext4_xattr_check_block(struct inode
*inode
, struct buffer_head
*bh
,
230 const char *function
, unsigned int line
)
232 int error
= -EFSCORRUPTED
;
234 if (BHDR(bh
)->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
) ||
235 BHDR(bh
)->h_blocks
!= cpu_to_le32(1))
237 if (buffer_verified(bh
))
241 if (!ext4_xattr_block_csum_verify(inode
, bh
))
243 error
= ext4_xattr_check_entries(BFIRST(bh
), bh
->b_data
+ bh
->b_size
,
247 __ext4_error_inode(inode
, function
, line
, 0,
248 "corrupted xattr block %llu",
249 (unsigned long long) bh
->b_blocknr
);
251 set_buffer_verified(bh
);
255 #define ext4_xattr_check_block(inode, bh) \
256 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
260 __xattr_check_inode(struct inode
*inode
, struct ext4_xattr_ibody_header
*header
,
261 void *end
, const char *function
, unsigned int line
)
263 int error
= -EFSCORRUPTED
;
265 if (end
- (void *)header
< sizeof(*header
) + sizeof(u32
) ||
266 (header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)))
268 error
= ext4_xattr_check_entries(IFIRST(header
), end
, IFIRST(header
));
271 __ext4_error_inode(inode
, function
, line
, 0,
272 "corrupted in-inode xattr");
276 #define xattr_check_inode(inode, header, end) \
277 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
280 xattr_find_entry(struct inode
*inode
, struct ext4_xattr_entry
**pentry
,
281 void *end
, int name_index
, const char *name
, int sorted
)
283 struct ext4_xattr_entry
*entry
, *next
;
289 name_len
= strlen(name
);
290 for (entry
= *pentry
; !IS_LAST_ENTRY(entry
); entry
= next
) {
291 next
= EXT4_XATTR_NEXT(entry
);
292 if ((void *) next
>= end
) {
293 EXT4_ERROR_INODE(inode
, "corrupted xattr entries");
294 return -EFSCORRUPTED
;
296 cmp
= name_index
- entry
->e_name_index
;
298 cmp
= name_len
- entry
->e_name_len
;
300 cmp
= memcmp(name
, entry
->e_name
, name_len
);
301 if (cmp
<= 0 && (sorted
|| cmp
== 0))
305 return cmp
? -ENODATA
: 0;
309 ext4_xattr_inode_hash(struct ext4_sb_info
*sbi
, const void *buffer
, size_t size
)
311 return ext4_chksum(sbi
, sbi
->s_csum_seed
, buffer
, size
);
314 static u64
ext4_xattr_inode_get_ref(struct inode
*ea_inode
)
316 return ((u64
)ea_inode
->i_ctime
.tv_sec
<< 32) |
317 ((u32
)ea_inode
->i_version
);
320 static void ext4_xattr_inode_set_ref(struct inode
*ea_inode
, u64 ref_count
)
322 ea_inode
->i_ctime
.tv_sec
= (u32
)(ref_count
>> 32);
323 ea_inode
->i_version
= (u32
)ref_count
;
326 static u32
ext4_xattr_inode_get_hash(struct inode
*ea_inode
)
328 return (u32
)ea_inode
->i_atime
.tv_sec
;
331 static void ext4_xattr_inode_set_hash(struct inode
*ea_inode
, u32 hash
)
333 ea_inode
->i_atime
.tv_sec
= hash
;
337 * Read the EA value from an inode.
339 static int ext4_xattr_inode_read(struct inode
*ea_inode
, void *buf
, size_t size
)
341 int blocksize
= 1 << ea_inode
->i_blkbits
;
342 int bh_count
= (size
+ blocksize
- 1) >> ea_inode
->i_blkbits
;
343 int tail_size
= (size
% blocksize
) ?: blocksize
;
344 struct buffer_head
*bhs_inline
[8];
345 struct buffer_head
**bhs
= bhs_inline
;
348 if (bh_count
> ARRAY_SIZE(bhs_inline
)) {
349 bhs
= kmalloc_array(bh_count
, sizeof(*bhs
), GFP_NOFS
);
354 ret
= ext4_bread_batch(ea_inode
, 0 /* block */, bh_count
,
355 true /* wait */, bhs
);
359 for (i
= 0; i
< bh_count
; i
++) {
360 /* There shouldn't be any holes in ea_inode. */
365 memcpy((char *)buf
+ blocksize
* i
, bhs
[i
]->b_data
,
366 i
< bh_count
- 1 ? blocksize
: tail_size
);
370 for (i
= 0; i
< bh_count
; i
++)
373 if (bhs
!= bhs_inline
)
378 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
380 static int ext4_xattr_inode_iget(struct inode
*parent
, unsigned long ea_ino
,
381 u32 ea_inode_hash
, struct inode
**ea_inode
)
386 inode
= ext4_iget(parent
->i_sb
, ea_ino
);
388 err
= PTR_ERR(inode
);
389 ext4_error(parent
->i_sb
,
390 "error while reading EA inode %lu err=%d", ea_ino
,
395 if (is_bad_inode(inode
)) {
396 ext4_error(parent
->i_sb
,
397 "error while reading EA inode %lu is_bad_inode",
403 if (!(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
)) {
404 ext4_error(parent
->i_sb
,
405 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
411 ext4_xattr_inode_set_class(inode
);
414 * Check whether this is an old Lustre-style xattr inode. Lustre
415 * implementation does not have hash validation, rather it has a
416 * backpointer from ea_inode to the parent inode.
418 if (ea_inode_hash
!= ext4_xattr_inode_get_hash(inode
) &&
419 EXT4_XATTR_INODE_GET_PARENT(inode
) == parent
->i_ino
&&
420 inode
->i_generation
== parent
->i_generation
) {
421 ext4_set_inode_state(inode
, EXT4_STATE_LUSTRE_EA_INODE
);
422 ext4_xattr_inode_set_ref(inode
, 1);
425 inode
->i_flags
|= S_NOQUOTA
;
437 ext4_xattr_inode_verify_hashes(struct inode
*ea_inode
,
438 struct ext4_xattr_entry
*entry
, void *buffer
,
443 /* Verify stored hash matches calculated hash. */
444 hash
= ext4_xattr_inode_hash(EXT4_SB(ea_inode
->i_sb
), buffer
, size
);
445 if (hash
!= ext4_xattr_inode_get_hash(ea_inode
))
446 return -EFSCORRUPTED
;
449 __le32 e_hash
, tmp_data
;
451 /* Verify entry hash. */
452 tmp_data
= cpu_to_le32(hash
);
453 e_hash
= ext4_xattr_hash_entry(entry
->e_name
, entry
->e_name_len
,
455 if (e_hash
!= entry
->e_hash
)
456 return -EFSCORRUPTED
;
462 * Read xattr value from the EA inode.
465 ext4_xattr_inode_get(struct inode
*inode
, struct ext4_xattr_entry
*entry
,
466 void *buffer
, size_t size
)
468 struct mb_cache
*ea_inode_cache
= EA_INODE_CACHE(inode
);
469 struct inode
*ea_inode
;
472 err
= ext4_xattr_inode_iget(inode
, le32_to_cpu(entry
->e_value_inum
),
473 le32_to_cpu(entry
->e_hash
), &ea_inode
);
479 if (i_size_read(ea_inode
) != size
) {
480 ext4_warning_inode(ea_inode
,
481 "ea_inode file size=%llu entry size=%zu",
482 i_size_read(ea_inode
), size
);
487 err
= ext4_xattr_inode_read(ea_inode
, buffer
, size
);
491 if (!ext4_test_inode_state(ea_inode
, EXT4_STATE_LUSTRE_EA_INODE
)) {
492 err
= ext4_xattr_inode_verify_hashes(ea_inode
, entry
, buffer
,
495 ext4_warning_inode(ea_inode
,
496 "EA inode hash validation failed");
501 mb_cache_entry_create(ea_inode_cache
, GFP_NOFS
,
502 ext4_xattr_inode_get_hash(ea_inode
),
503 ea_inode
->i_ino
, true /* reusable */);
511 ext4_xattr_block_get(struct inode
*inode
, int name_index
, const char *name
,
512 void *buffer
, size_t buffer_size
)
514 struct buffer_head
*bh
= NULL
;
515 struct ext4_xattr_entry
*entry
;
519 struct mb_cache
*ea_block_cache
= EA_BLOCK_CACHE(inode
);
521 ea_idebug(inode
, "name=%d.%s, buffer=%p, buffer_size=%ld",
522 name_index
, name
, buffer
, (long)buffer_size
);
524 if (!EXT4_I(inode
)->i_file_acl
)
526 ea_idebug(inode
, "reading block %llu",
527 (unsigned long long)EXT4_I(inode
)->i_file_acl
);
528 bh
= ext4_sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
, REQ_PRIO
);
531 ea_bdebug(bh
, "b_count=%d, refcount=%d",
532 atomic_read(&(bh
->b_count
)), le32_to_cpu(BHDR(bh
)->h_refcount
));
533 error
= ext4_xattr_check_block(inode
, bh
);
536 ext4_xattr_block_cache_insert(ea_block_cache
, bh
);
538 end
= bh
->b_data
+ bh
->b_size
;
539 error
= xattr_find_entry(inode
, &entry
, end
, name_index
, name
, 1);
542 size
= le32_to_cpu(entry
->e_value_size
);
544 if (unlikely(size
> EXT4_XATTR_SIZE_MAX
))
547 if (size
> buffer_size
)
549 if (entry
->e_value_inum
) {
550 error
= ext4_xattr_inode_get(inode
, entry
, buffer
,
555 u16 offset
= le16_to_cpu(entry
->e_value_offs
);
556 void *p
= bh
->b_data
+ offset
;
558 if (unlikely(p
+ size
> end
))
560 memcpy(buffer
, p
, size
);
571 ext4_xattr_ibody_get(struct inode
*inode
, int name_index
, const char *name
,
572 void *buffer
, size_t buffer_size
)
574 struct ext4_xattr_ibody_header
*header
;
575 struct ext4_xattr_entry
*entry
;
576 struct ext4_inode
*raw_inode
;
577 struct ext4_iloc iloc
;
582 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
))
584 error
= ext4_get_inode_loc(inode
, &iloc
);
587 raw_inode
= ext4_raw_inode(&iloc
);
588 header
= IHDR(inode
, raw_inode
);
589 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
590 error
= xattr_check_inode(inode
, header
, end
);
593 entry
= IFIRST(header
);
594 error
= xattr_find_entry(inode
, &entry
, end
, name_index
, name
, 0);
597 size
= le32_to_cpu(entry
->e_value_size
);
599 if (unlikely(size
> EXT4_XATTR_SIZE_MAX
))
602 if (size
> buffer_size
)
604 if (entry
->e_value_inum
) {
605 error
= ext4_xattr_inode_get(inode
, entry
, buffer
,
610 u16 offset
= le16_to_cpu(entry
->e_value_offs
);
611 void *p
= (void *)IFIRST(header
) + offset
;
613 if (unlikely(p
+ size
> end
))
615 memcpy(buffer
, p
, size
);
628 * Copy an extended attribute into the buffer
629 * provided, or compute the buffer size required.
630 * Buffer is NULL to compute the size of the buffer required.
632 * Returns a negative error number on failure, or the number of bytes
633 * used / required on success.
636 ext4_xattr_get(struct inode
*inode
, int name_index
, const char *name
,
637 void *buffer
, size_t buffer_size
)
641 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
644 if (strlen(name
) > 255)
647 down_read(&EXT4_I(inode
)->xattr_sem
);
648 error
= ext4_xattr_ibody_get(inode
, name_index
, name
, buffer
,
650 if (error
== -ENODATA
)
651 error
= ext4_xattr_block_get(inode
, name_index
, name
, buffer
,
653 up_read(&EXT4_I(inode
)->xattr_sem
);
658 ext4_xattr_list_entries(struct dentry
*dentry
, struct ext4_xattr_entry
*entry
,
659 char *buffer
, size_t buffer_size
)
661 size_t rest
= buffer_size
;
663 for (; !IS_LAST_ENTRY(entry
); entry
= EXT4_XATTR_NEXT(entry
)) {
664 const struct xattr_handler
*handler
=
665 ext4_xattr_handler(entry
->e_name_index
);
667 if (handler
&& (!handler
->list
|| handler
->list(dentry
))) {
668 const char *prefix
= handler
->prefix
?: handler
->name
;
669 size_t prefix_len
= strlen(prefix
);
670 size_t size
= prefix_len
+ entry
->e_name_len
+ 1;
675 memcpy(buffer
, prefix
, prefix_len
);
676 buffer
+= prefix_len
;
677 memcpy(buffer
, entry
->e_name
, entry
->e_name_len
);
678 buffer
+= entry
->e_name_len
;
684 return buffer_size
- rest
; /* total size */
688 ext4_xattr_block_list(struct dentry
*dentry
, char *buffer
, size_t buffer_size
)
690 struct inode
*inode
= d_inode(dentry
);
691 struct buffer_head
*bh
= NULL
;
694 ea_idebug(inode
, "buffer=%p, buffer_size=%ld",
695 buffer
, (long)buffer_size
);
697 if (!EXT4_I(inode
)->i_file_acl
)
699 ea_idebug(inode
, "reading block %llu",
700 (unsigned long long)EXT4_I(inode
)->i_file_acl
);
701 bh
= ext4_sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
, REQ_PRIO
);
704 ea_bdebug(bh
, "b_count=%d, refcount=%d",
705 atomic_read(&(bh
->b_count
)), le32_to_cpu(BHDR(bh
)->h_refcount
));
706 error
= ext4_xattr_check_block(inode
, bh
);
709 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode
), bh
);
710 error
= ext4_xattr_list_entries(dentry
, BFIRST(bh
), buffer
,
718 ext4_xattr_ibody_list(struct dentry
*dentry
, char *buffer
, size_t buffer_size
)
720 struct inode
*inode
= d_inode(dentry
);
721 struct ext4_xattr_ibody_header
*header
;
722 struct ext4_inode
*raw_inode
;
723 struct ext4_iloc iloc
;
727 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
))
729 error
= ext4_get_inode_loc(inode
, &iloc
);
732 raw_inode
= ext4_raw_inode(&iloc
);
733 header
= IHDR(inode
, raw_inode
);
734 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
735 error
= xattr_check_inode(inode
, header
, end
);
738 error
= ext4_xattr_list_entries(dentry
, IFIRST(header
),
739 buffer
, buffer_size
);
747 * Inode operation listxattr()
749 * d_inode(dentry)->i_rwsem: don't care
751 * Copy a list of attribute names into the buffer
752 * provided, or compute the buffer size required.
753 * Buffer is NULL to compute the size of the buffer required.
755 * Returns a negative error number on failure, or the number of bytes
756 * used / required on success.
759 ext4_listxattr(struct dentry
*dentry
, char *buffer
, size_t buffer_size
)
763 down_read(&EXT4_I(d_inode(dentry
))->xattr_sem
);
764 ret
= ret2
= ext4_xattr_ibody_list(dentry
, buffer
, buffer_size
);
771 ret
= ext4_xattr_block_list(dentry
, buffer
, buffer_size
);
776 up_read(&EXT4_I(d_inode(dentry
))->xattr_sem
);
781 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
784 static void ext4_xattr_update_super_block(handle_t
*handle
,
785 struct super_block
*sb
)
787 if (ext4_has_feature_xattr(sb
))
790 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
791 if (ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
) == 0) {
792 ext4_set_feature_xattr(sb
);
793 ext4_handle_dirty_super(handle
, sb
);
797 int ext4_get_inode_usage(struct inode
*inode
, qsize_t
*usage
)
799 struct ext4_iloc iloc
= { .bh
= NULL
};
800 struct buffer_head
*bh
= NULL
;
801 struct ext4_inode
*raw_inode
;
802 struct ext4_xattr_ibody_header
*header
;
803 struct ext4_xattr_entry
*entry
;
804 qsize_t ea_inode_refs
= 0;
808 lockdep_assert_held_read(&EXT4_I(inode
)->xattr_sem
);
810 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
811 ret
= ext4_get_inode_loc(inode
, &iloc
);
814 raw_inode
= ext4_raw_inode(&iloc
);
815 header
= IHDR(inode
, raw_inode
);
816 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
817 ret
= xattr_check_inode(inode
, header
, end
);
821 for (entry
= IFIRST(header
); !IS_LAST_ENTRY(entry
);
822 entry
= EXT4_XATTR_NEXT(entry
))
823 if (entry
->e_value_inum
)
827 if (EXT4_I(inode
)->i_file_acl
) {
828 bh
= ext4_sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
, REQ_PRIO
);
834 ret
= ext4_xattr_check_block(inode
, bh
);
838 for (entry
= BFIRST(bh
); !IS_LAST_ENTRY(entry
);
839 entry
= EXT4_XATTR_NEXT(entry
))
840 if (entry
->e_value_inum
)
843 *usage
= ea_inode_refs
+ 1;
851 static inline size_t round_up_cluster(struct inode
*inode
, size_t length
)
853 struct super_block
*sb
= inode
->i_sb
;
854 size_t cluster_size
= 1 << (EXT4_SB(sb
)->s_cluster_bits
+
856 size_t mask
= ~(cluster_size
- 1);
858 return (length
+ cluster_size
- 1) & mask
;
861 static int ext4_xattr_inode_alloc_quota(struct inode
*inode
, size_t len
)
865 err
= dquot_alloc_inode(inode
);
868 err
= dquot_alloc_space_nodirty(inode
, round_up_cluster(inode
, len
));
870 dquot_free_inode(inode
);
874 static void ext4_xattr_inode_free_quota(struct inode
*parent
,
875 struct inode
*ea_inode
,
879 ext4_test_inode_state(ea_inode
, EXT4_STATE_LUSTRE_EA_INODE
))
881 dquot_free_space_nodirty(parent
, round_up_cluster(parent
, len
));
882 dquot_free_inode(parent
);
885 int __ext4_xattr_set_credits(struct super_block
*sb
, struct inode
*inode
,
886 struct buffer_head
*block_bh
, size_t value_len
,
893 * 1) Owner inode update
894 * 2) Ref count update on old xattr block
896 * 4) block bitmap update for new xattr block
897 * 5) group descriptor for new xattr block
898 * 6) block bitmap update for old xattr block
899 * 7) group descriptor for old block
901 * 6 & 7 can happen if we have two racing threads T_a and T_b
902 * which are each trying to set an xattr on inodes I_a and I_b
903 * which were both initially sharing an xattr block.
908 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(sb
);
911 * In case of inline data, we may push out the data to a block,
912 * so we need to reserve credits for this eventuality
914 if (inode
&& ext4_has_inline_data(inode
))
915 credits
+= ext4_writepage_trans_blocks(inode
) + 1;
917 /* We are done if ea_inode feature is not enabled. */
918 if (!ext4_has_feature_ea_inode(sb
))
921 /* New ea_inode, inode map, block bitmap, group descriptor. */
925 blocks
= (value_len
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
927 /* Indirection block or one level of extent tree. */
930 /* Block bitmap and group descriptor updates for each block. */
931 credits
+= blocks
* 2;
933 /* Blocks themselves. */
937 /* Dereference ea_inode holding old xattr value.
938 * Old ea_inode, inode map, block bitmap, group descriptor.
942 /* Data blocks for old ea_inode. */
943 blocks
= XATTR_SIZE_MAX
>> sb
->s_blocksize_bits
;
945 /* Indirection block or one level of extent tree for old
950 /* Block bitmap and group descriptor updates for each block. */
951 credits
+= blocks
* 2;
954 /* We may need to clone the existing xattr block in which case we need
955 * to increment ref counts for existing ea_inodes referenced by it.
958 struct ext4_xattr_entry
*entry
= BFIRST(block_bh
);
960 for (; !IS_LAST_ENTRY(entry
); entry
= EXT4_XATTR_NEXT(entry
))
961 if (entry
->e_value_inum
)
962 /* Ref count update on ea_inode. */
968 static int ext4_xattr_ensure_credits(handle_t
*handle
, struct inode
*inode
,
969 int credits
, struct buffer_head
*bh
,
970 bool dirty
, bool block_csum
)
974 if (!ext4_handle_valid(handle
))
977 if (handle
->h_buffer_credits
>= credits
)
980 error
= ext4_journal_extend(handle
, credits
- handle
->h_buffer_credits
);
984 ext4_warning(inode
->i_sb
, "Extend journal (error %d)", error
);
990 ext4_xattr_block_csum_set(inode
, bh
);
991 error
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
993 ext4_warning(inode
->i_sb
, "Handle metadata (error %d)",
999 error
= ext4_journal_restart(handle
, credits
);
1001 ext4_warning(inode
->i_sb
, "Restart journal (error %d)", error
);
1006 error
= ext4_journal_get_write_access(handle
, bh
);
1008 ext4_warning(inode
->i_sb
,
1009 "Get write access failed (error %d)",
1017 static int ext4_xattr_inode_update_ref(handle_t
*handle
, struct inode
*ea_inode
,
1020 struct mb_cache
*ea_inode_cache
= EA_INODE_CACHE(ea_inode
);
1021 struct ext4_iloc iloc
;
1026 inode_lock(ea_inode
);
1028 ret
= ext4_reserve_inode_write(handle
, ea_inode
, &iloc
);
1034 ref_count
= ext4_xattr_inode_get_ref(ea_inode
);
1035 ref_count
+= ref_change
;
1036 ext4_xattr_inode_set_ref(ea_inode
, ref_count
);
1038 if (ref_change
> 0) {
1039 WARN_ONCE(ref_count
<= 0, "EA inode %lu ref_count=%lld",
1040 ea_inode
->i_ino
, ref_count
);
1042 if (ref_count
== 1) {
1043 WARN_ONCE(ea_inode
->i_nlink
, "EA inode %lu i_nlink=%u",
1044 ea_inode
->i_ino
, ea_inode
->i_nlink
);
1046 set_nlink(ea_inode
, 1);
1047 ext4_orphan_del(handle
, ea_inode
);
1049 if (ea_inode_cache
) {
1050 hash
= ext4_xattr_inode_get_hash(ea_inode
);
1051 mb_cache_entry_create(ea_inode_cache
,
1054 true /* reusable */);
1058 WARN_ONCE(ref_count
< 0, "EA inode %lu ref_count=%lld",
1059 ea_inode
->i_ino
, ref_count
);
1061 if (ref_count
== 0) {
1062 WARN_ONCE(ea_inode
->i_nlink
!= 1,
1063 "EA inode %lu i_nlink=%u",
1064 ea_inode
->i_ino
, ea_inode
->i_nlink
);
1066 clear_nlink(ea_inode
);
1067 ext4_orphan_add(handle
, ea_inode
);
1069 if (ea_inode_cache
) {
1070 hash
= ext4_xattr_inode_get_hash(ea_inode
);
1071 mb_cache_entry_delete(ea_inode_cache
, hash
,
1077 ret
= ext4_mark_iloc_dirty(handle
, ea_inode
, &iloc
);
1080 ext4_warning_inode(ea_inode
,
1081 "ext4_mark_iloc_dirty() failed ret=%d", ret
);
1084 inode_unlock(ea_inode
);
1088 static int ext4_xattr_inode_inc_ref(handle_t
*handle
, struct inode
*ea_inode
)
1090 return ext4_xattr_inode_update_ref(handle
, ea_inode
, 1);
1093 static int ext4_xattr_inode_dec_ref(handle_t
*handle
, struct inode
*ea_inode
)
1095 return ext4_xattr_inode_update_ref(handle
, ea_inode
, -1);
1098 static int ext4_xattr_inode_inc_ref_all(handle_t
*handle
, struct inode
*parent
,
1099 struct ext4_xattr_entry
*first
)
1101 struct inode
*ea_inode
;
1102 struct ext4_xattr_entry
*entry
;
1103 struct ext4_xattr_entry
*failed_entry
;
1104 unsigned int ea_ino
;
1107 for (entry
= first
; !IS_LAST_ENTRY(entry
);
1108 entry
= EXT4_XATTR_NEXT(entry
)) {
1109 if (!entry
->e_value_inum
)
1111 ea_ino
= le32_to_cpu(entry
->e_value_inum
);
1112 err
= ext4_xattr_inode_iget(parent
, ea_ino
,
1113 le32_to_cpu(entry
->e_hash
),
1117 err
= ext4_xattr_inode_inc_ref(handle
, ea_inode
);
1119 ext4_warning_inode(ea_inode
, "inc ref error %d", err
);
1129 failed_entry
= entry
;
1131 for (entry
= first
; entry
!= failed_entry
;
1132 entry
= EXT4_XATTR_NEXT(entry
)) {
1133 if (!entry
->e_value_inum
)
1135 ea_ino
= le32_to_cpu(entry
->e_value_inum
);
1136 err
= ext4_xattr_inode_iget(parent
, ea_ino
,
1137 le32_to_cpu(entry
->e_hash
),
1140 ext4_warning(parent
->i_sb
,
1141 "cleanup ea_ino %u iget error %d", ea_ino
,
1145 err
= ext4_xattr_inode_dec_ref(handle
, ea_inode
);
1147 ext4_warning_inode(ea_inode
, "cleanup dec ref error %d",
1155 ext4_xattr_inode_dec_ref_all(handle_t
*handle
, struct inode
*parent
,
1156 struct buffer_head
*bh
,
1157 struct ext4_xattr_entry
*first
, bool block_csum
,
1158 struct ext4_xattr_inode_array
**ea_inode_array
,
1159 int extra_credits
, bool skip_quota
)
1161 struct inode
*ea_inode
;
1162 struct ext4_xattr_entry
*entry
;
1164 unsigned int ea_ino
;
1168 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1169 credits
= 2 + extra_credits
;
1171 for (entry
= first
; !IS_LAST_ENTRY(entry
);
1172 entry
= EXT4_XATTR_NEXT(entry
)) {
1173 if (!entry
->e_value_inum
)
1175 ea_ino
= le32_to_cpu(entry
->e_value_inum
);
1176 err
= ext4_xattr_inode_iget(parent
, ea_ino
,
1177 le32_to_cpu(entry
->e_hash
),
1182 err
= ext4_expand_inode_array(ea_inode_array
, ea_inode
);
1184 ext4_warning_inode(ea_inode
,
1185 "Expand inode array err=%d", err
);
1190 err
= ext4_xattr_ensure_credits(handle
, parent
, credits
, bh
,
1193 ext4_warning_inode(ea_inode
, "Ensure credits err=%d",
1198 err
= ext4_xattr_inode_dec_ref(handle
, ea_inode
);
1200 ext4_warning_inode(ea_inode
, "ea_inode dec ref err=%d",
1206 ext4_xattr_inode_free_quota(parent
, ea_inode
,
1207 le32_to_cpu(entry
->e_value_size
));
1210 * Forget about ea_inode within the same transaction that
1211 * decrements the ref count. This avoids duplicate decrements in
1212 * case the rest of the work spills over to subsequent
1215 entry
->e_value_inum
= 0;
1216 entry
->e_value_size
= 0;
1223 * Note that we are deliberately skipping csum calculation for
1224 * the final update because we do not expect any journal
1225 * restarts until xattr block is freed.
1228 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1230 ext4_warning_inode(parent
,
1231 "handle dirty metadata err=%d", err
);
1236 * Release the xattr block BH: If the reference count is > 1, decrement it;
1237 * otherwise free the block.
1240 ext4_xattr_release_block(handle_t
*handle
, struct inode
*inode
,
1241 struct buffer_head
*bh
,
1242 struct ext4_xattr_inode_array
**ea_inode_array
,
1245 struct mb_cache
*ea_block_cache
= EA_BLOCK_CACHE(inode
);
1249 BUFFER_TRACE(bh
, "get_write_access");
1250 error
= ext4_journal_get_write_access(handle
, bh
);
1255 hash
= le32_to_cpu(BHDR(bh
)->h_hash
);
1256 ref
= le32_to_cpu(BHDR(bh
)->h_refcount
);
1258 ea_bdebug(bh
, "refcount now=0; freeing");
1260 * This must happen under buffer lock for
1261 * ext4_xattr_block_set() to reliably detect freed block
1264 mb_cache_entry_delete(ea_block_cache
, hash
,
1269 if (ext4_has_feature_ea_inode(inode
->i_sb
))
1270 ext4_xattr_inode_dec_ref_all(handle
, inode
, bh
,
1272 true /* block_csum */,
1275 true /* skip_quota */);
1276 ext4_free_blocks(handle
, inode
, bh
, 0, 1,
1277 EXT4_FREE_BLOCKS_METADATA
|
1278 EXT4_FREE_BLOCKS_FORGET
);
1281 BHDR(bh
)->h_refcount
= cpu_to_le32(ref
);
1282 if (ref
== EXT4_XATTR_REFCOUNT_MAX
- 1) {
1283 struct mb_cache_entry
*ce
;
1285 if (ea_block_cache
) {
1286 ce
= mb_cache_entry_get(ea_block_cache
, hash
,
1290 mb_cache_entry_put(ea_block_cache
, ce
);
1295 ext4_xattr_block_csum_set(inode
, bh
);
1297 * Beware of this ugliness: Releasing of xattr block references
1298 * from different inodes can race and so we have to protect
1299 * from a race where someone else frees the block (and releases
1300 * its journal_head) before we are done dirtying the buffer. In
1301 * nojournal mode this race is harmless and we actually cannot
1302 * call ext4_handle_dirty_metadata() with locked buffer as
1303 * that function can call sync_dirty_buffer() so for that case
1304 * we handle the dirtying after unlocking the buffer.
1306 if (ext4_handle_valid(handle
))
1307 error
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1309 if (!ext4_handle_valid(handle
))
1310 error
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1312 ext4_handle_sync(handle
);
1313 dquot_free_block(inode
, EXT4_C2B(EXT4_SB(inode
->i_sb
), 1));
1314 ea_bdebug(bh
, "refcount now=%d; releasing",
1315 le32_to_cpu(BHDR(bh
)->h_refcount
));
1318 ext4_std_error(inode
->i_sb
, error
);
1323 * Find the available free space for EAs. This also returns the total number of
1324 * bytes used by EA entries.
1326 static size_t ext4_xattr_free_space(struct ext4_xattr_entry
*last
,
1327 size_t *min_offs
, void *base
, int *total
)
1329 for (; !IS_LAST_ENTRY(last
); last
= EXT4_XATTR_NEXT(last
)) {
1330 if (!last
->e_value_inum
&& last
->e_value_size
) {
1331 size_t offs
= le16_to_cpu(last
->e_value_offs
);
1332 if (offs
< *min_offs
)
1336 *total
+= EXT4_XATTR_LEN(last
->e_name_len
);
1338 return (*min_offs
- ((void *)last
- base
) - sizeof(__u32
));
1342 * Write the value of the EA in an inode.
1344 static int ext4_xattr_inode_write(handle_t
*handle
, struct inode
*ea_inode
,
1345 const void *buf
, int bufsize
)
1347 struct buffer_head
*bh
= NULL
;
1348 unsigned long block
= 0;
1349 int blocksize
= ea_inode
->i_sb
->s_blocksize
;
1350 int max_blocks
= (bufsize
+ blocksize
- 1) >> ea_inode
->i_blkbits
;
1351 int csize
, wsize
= 0;
1356 while (ret
>= 0 && ret
< max_blocks
) {
1357 struct ext4_map_blocks map
;
1358 map
.m_lblk
= block
+= ret
;
1359 map
.m_len
= max_blocks
-= ret
;
1361 ret
= ext4_map_blocks(handle
, ea_inode
, &map
,
1362 EXT4_GET_BLOCKS_CREATE
);
1364 ext4_mark_inode_dirty(handle
, ea_inode
);
1365 if (ret
== -ENOSPC
&&
1366 ext4_should_retry_alloc(ea_inode
->i_sb
, &retries
)) {
1378 while (wsize
< bufsize
) {
1381 csize
= (bufsize
- wsize
) > blocksize
? blocksize
:
1383 bh
= ext4_getblk(handle
, ea_inode
, block
, 0);
1388 EXT4_ERROR_INODE(ea_inode
,
1389 "ext4_getblk() return bh = NULL");
1390 return -EFSCORRUPTED
;
1392 ret
= ext4_journal_get_write_access(handle
, bh
);
1396 memcpy(bh
->b_data
, buf
, csize
);
1397 set_buffer_uptodate(bh
);
1398 ext4_handle_dirty_metadata(handle
, ea_inode
, bh
);
1405 inode_lock(ea_inode
);
1406 i_size_write(ea_inode
, wsize
);
1407 ext4_update_i_disksize(ea_inode
, wsize
);
1408 inode_unlock(ea_inode
);
1410 ext4_mark_inode_dirty(handle
, ea_inode
);
1419 * Create an inode to store the value of a large EA.
1421 static struct inode
*ext4_xattr_inode_create(handle_t
*handle
,
1422 struct inode
*inode
, u32 hash
)
1424 struct inode
*ea_inode
= NULL
;
1425 uid_t owner
[2] = { i_uid_read(inode
), i_gid_read(inode
) };
1429 * Let the next inode be the goal, so we try and allocate the EA inode
1430 * in the same group, or nearby one.
1432 ea_inode
= ext4_new_inode(handle
, inode
->i_sb
->s_root
->d_inode
,
1433 S_IFREG
| 0600, NULL
, inode
->i_ino
+ 1, owner
,
1435 if (!IS_ERR(ea_inode
)) {
1436 ea_inode
->i_op
= &ext4_file_inode_operations
;
1437 ea_inode
->i_fop
= &ext4_file_operations
;
1438 ext4_set_aops(ea_inode
);
1439 ext4_xattr_inode_set_class(ea_inode
);
1440 unlock_new_inode(ea_inode
);
1441 ext4_xattr_inode_set_ref(ea_inode
, 1);
1442 ext4_xattr_inode_set_hash(ea_inode
, hash
);
1443 err
= ext4_mark_inode_dirty(handle
, ea_inode
);
1445 err
= ext4_inode_attach_jinode(ea_inode
);
1448 return ERR_PTR(err
);
1452 * Xattr inodes are shared therefore quota charging is performed
1453 * at a higher level.
1455 dquot_free_inode(ea_inode
);
1456 dquot_drop(ea_inode
);
1457 inode_lock(ea_inode
);
1458 ea_inode
->i_flags
|= S_NOQUOTA
;
1459 inode_unlock(ea_inode
);
1465 static struct inode
*
1466 ext4_xattr_inode_cache_find(struct inode
*inode
, const void *value
,
1467 size_t value_len
, u32 hash
)
1469 struct inode
*ea_inode
;
1470 struct mb_cache_entry
*ce
;
1471 struct mb_cache
*ea_inode_cache
= EA_INODE_CACHE(inode
);
1474 if (!ea_inode_cache
)
1477 ce
= mb_cache_entry_find_first(ea_inode_cache
, hash
);
1481 ea_data
= ext4_kvmalloc(value_len
, GFP_NOFS
);
1483 mb_cache_entry_put(ea_inode_cache
, ce
);
1488 ea_inode
= ext4_iget(inode
->i_sb
, ce
->e_value
);
1489 if (!IS_ERR(ea_inode
) &&
1490 !is_bad_inode(ea_inode
) &&
1491 (EXT4_I(ea_inode
)->i_flags
& EXT4_EA_INODE_FL
) &&
1492 i_size_read(ea_inode
) == value_len
&&
1493 !ext4_xattr_inode_read(ea_inode
, ea_data
, value_len
) &&
1494 !ext4_xattr_inode_verify_hashes(ea_inode
, NULL
, ea_data
,
1496 !memcmp(value
, ea_data
, value_len
)) {
1497 mb_cache_entry_touch(ea_inode_cache
, ce
);
1498 mb_cache_entry_put(ea_inode_cache
, ce
);
1503 if (!IS_ERR(ea_inode
))
1505 ce
= mb_cache_entry_find_next(ea_inode_cache
, ce
);
1512 * Add value of the EA in an inode.
1514 static int ext4_xattr_inode_lookup_create(handle_t
*handle
, struct inode
*inode
,
1515 const void *value
, size_t value_len
,
1516 struct inode
**ret_inode
)
1518 struct inode
*ea_inode
;
1522 hash
= ext4_xattr_inode_hash(EXT4_SB(inode
->i_sb
), value
, value_len
);
1523 ea_inode
= ext4_xattr_inode_cache_find(inode
, value
, value_len
, hash
);
1525 err
= ext4_xattr_inode_inc_ref(handle
, ea_inode
);
1531 *ret_inode
= ea_inode
;
1535 /* Create an inode for the EA value */
1536 ea_inode
= ext4_xattr_inode_create(handle
, inode
, hash
);
1537 if (IS_ERR(ea_inode
))
1538 return PTR_ERR(ea_inode
);
1540 err
= ext4_xattr_inode_write(handle
, ea_inode
, value
, value_len
);
1542 ext4_xattr_inode_dec_ref(handle
, ea_inode
);
1547 if (EA_INODE_CACHE(inode
))
1548 mb_cache_entry_create(EA_INODE_CACHE(inode
), GFP_NOFS
, hash
,
1549 ea_inode
->i_ino
, true /* reusable */);
1551 *ret_inode
= ea_inode
;
1556 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1557 * feature is enabled.
1559 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1561 static int ext4_xattr_set_entry(struct ext4_xattr_info
*i
,
1562 struct ext4_xattr_search
*s
,
1563 handle_t
*handle
, struct inode
*inode
,
1566 struct ext4_xattr_entry
*last
, *next
;
1567 struct ext4_xattr_entry
*here
= s
->here
;
1568 size_t min_offs
= s
->end
- s
->base
, name_len
= strlen(i
->name
);
1569 int in_inode
= i
->in_inode
;
1570 struct inode
*old_ea_inode
= NULL
;
1571 struct inode
*new_ea_inode
= NULL
;
1572 size_t old_size
, new_size
;
1575 /* Space used by old and new values. */
1576 old_size
= (!s
->not_found
&& !here
->e_value_inum
) ?
1577 EXT4_XATTR_SIZE(le32_to_cpu(here
->e_value_size
)) : 0;
1578 new_size
= (i
->value
&& !in_inode
) ? EXT4_XATTR_SIZE(i
->value_len
) : 0;
1581 * Optimization for the simple case when old and new values have the
1582 * same padded sizes. Not applicable if external inodes are involved.
1584 if (new_size
&& new_size
== old_size
) {
1585 size_t offs
= le16_to_cpu(here
->e_value_offs
);
1586 void *val
= s
->base
+ offs
;
1588 here
->e_value_size
= cpu_to_le32(i
->value_len
);
1589 if (i
->value
== EXT4_ZERO_XATTR_VALUE
) {
1590 memset(val
, 0, new_size
);
1592 memcpy(val
, i
->value
, i
->value_len
);
1593 /* Clear padding bytes. */
1594 memset(val
+ i
->value_len
, 0, new_size
- i
->value_len
);
1599 /* Compute min_offs and last. */
1601 for (; !IS_LAST_ENTRY(last
); last
= next
) {
1602 next
= EXT4_XATTR_NEXT(last
);
1603 if ((void *)next
>= s
->end
) {
1604 EXT4_ERROR_INODE(inode
, "corrupted xattr entries");
1605 ret
= -EFSCORRUPTED
;
1608 if (!last
->e_value_inum
&& last
->e_value_size
) {
1609 size_t offs
= le16_to_cpu(last
->e_value_offs
);
1610 if (offs
< min_offs
)
1615 /* Check whether we have enough space. */
1619 free
= min_offs
- ((void *)last
- s
->base
) - sizeof(__u32
);
1621 free
+= EXT4_XATTR_LEN(name_len
) + old_size
;
1623 if (free
< EXT4_XATTR_LEN(name_len
) + new_size
) {
1629 * If storing the value in an external inode is an option,
1630 * reserve space for xattr entries/names in the external
1631 * attribute block so that a long value does not occupy the
1632 * whole space and prevent futher entries being added.
1634 if (ext4_has_feature_ea_inode(inode
->i_sb
) &&
1635 new_size
&& is_block
&&
1636 (min_offs
+ old_size
- new_size
) <
1637 EXT4_XATTR_BLOCK_RESERVE(inode
)) {
1644 * Getting access to old and new ea inodes is subject to failures.
1645 * Finish that work before doing any modifications to the xattr data.
1647 if (!s
->not_found
&& here
->e_value_inum
) {
1648 ret
= ext4_xattr_inode_iget(inode
,
1649 le32_to_cpu(here
->e_value_inum
),
1650 le32_to_cpu(here
->e_hash
),
1653 old_ea_inode
= NULL
;
1657 if (i
->value
&& in_inode
) {
1658 WARN_ON_ONCE(!i
->value_len
);
1660 ret
= ext4_xattr_inode_alloc_quota(inode
, i
->value_len
);
1664 ret
= ext4_xattr_inode_lookup_create(handle
, inode
, i
->value
,
1668 new_ea_inode
= NULL
;
1669 ext4_xattr_inode_free_quota(inode
, NULL
, i
->value_len
);
1675 /* We are ready to release ref count on the old_ea_inode. */
1676 ret
= ext4_xattr_inode_dec_ref(handle
, old_ea_inode
);
1678 /* Release newly required ref count on new_ea_inode. */
1682 err
= ext4_xattr_inode_dec_ref(handle
,
1685 ext4_warning_inode(new_ea_inode
,
1686 "dec ref new_ea_inode err=%d",
1688 ext4_xattr_inode_free_quota(inode
, new_ea_inode
,
1694 ext4_xattr_inode_free_quota(inode
, old_ea_inode
,
1695 le32_to_cpu(here
->e_value_size
));
1698 /* No failures allowed past this point. */
1700 if (!s
->not_found
&& here
->e_value_size
&& here
->e_value_offs
) {
1701 /* Remove the old value. */
1702 void *first_val
= s
->base
+ min_offs
;
1703 size_t offs
= le16_to_cpu(here
->e_value_offs
);
1704 void *val
= s
->base
+ offs
;
1706 memmove(first_val
+ old_size
, first_val
, val
- first_val
);
1707 memset(first_val
, 0, old_size
);
1708 min_offs
+= old_size
;
1710 /* Adjust all value offsets. */
1712 while (!IS_LAST_ENTRY(last
)) {
1713 size_t o
= le16_to_cpu(last
->e_value_offs
);
1715 if (!last
->e_value_inum
&&
1716 last
->e_value_size
&& o
< offs
)
1717 last
->e_value_offs
= cpu_to_le16(o
+ old_size
);
1718 last
= EXT4_XATTR_NEXT(last
);
1723 /* Remove old name. */
1724 size_t size
= EXT4_XATTR_LEN(name_len
);
1726 last
= ENTRY((void *)last
- size
);
1727 memmove(here
, (void *)here
+ size
,
1728 (void *)last
- (void *)here
+ sizeof(__u32
));
1729 memset(last
, 0, size
);
1730 } else if (s
->not_found
) {
1731 /* Insert new name. */
1732 size_t size
= EXT4_XATTR_LEN(name_len
);
1733 size_t rest
= (void *)last
- (void *)here
+ sizeof(__u32
);
1735 memmove((void *)here
+ size
, here
, rest
);
1736 memset(here
, 0, size
);
1737 here
->e_name_index
= i
->name_index
;
1738 here
->e_name_len
= name_len
;
1739 memcpy(here
->e_name
, i
->name
, name_len
);
1741 /* This is an update, reset value info. */
1742 here
->e_value_inum
= 0;
1743 here
->e_value_offs
= 0;
1744 here
->e_value_size
= 0;
1748 /* Insert new value. */
1750 here
->e_value_inum
= cpu_to_le32(new_ea_inode
->i_ino
);
1751 } else if (i
->value_len
) {
1752 void *val
= s
->base
+ min_offs
- new_size
;
1754 here
->e_value_offs
= cpu_to_le16(min_offs
- new_size
);
1755 if (i
->value
== EXT4_ZERO_XATTR_VALUE
) {
1756 memset(val
, 0, new_size
);
1758 memcpy(val
, i
->value
, i
->value_len
);
1759 /* Clear padding bytes. */
1760 memset(val
+ i
->value_len
, 0,
1761 new_size
- i
->value_len
);
1764 here
->e_value_size
= cpu_to_le32(i
->value_len
);
1771 /* Entry hash calculation. */
1776 * Feed crc32c hash instead of the raw value for entry
1777 * hash calculation. This is to avoid walking
1778 * potentially long value buffer again.
1780 crc32c_hash
= cpu_to_le32(
1781 ext4_xattr_inode_get_hash(new_ea_inode
));
1782 hash
= ext4_xattr_hash_entry(here
->e_name
,
1785 } else if (is_block
) {
1786 __le32
*value
= s
->base
+ le16_to_cpu(
1787 here
->e_value_offs
);
1789 hash
= ext4_xattr_hash_entry(here
->e_name
,
1790 here
->e_name_len
, value
,
1793 here
->e_hash
= hash
;
1797 ext4_xattr_rehash((struct ext4_xattr_header
*)s
->base
);
1806 struct ext4_xattr_block_find
{
1807 struct ext4_xattr_search s
;
1808 struct buffer_head
*bh
;
1812 ext4_xattr_block_find(struct inode
*inode
, struct ext4_xattr_info
*i
,
1813 struct ext4_xattr_block_find
*bs
)
1815 struct super_block
*sb
= inode
->i_sb
;
1818 ea_idebug(inode
, "name=%d.%s, value=%p, value_len=%ld",
1819 i
->name_index
, i
->name
, i
->value
, (long)i
->value_len
);
1821 if (EXT4_I(inode
)->i_file_acl
) {
1822 /* The inode already has an extended attribute block. */
1823 bs
->bh
= ext4_sb_bread(sb
, EXT4_I(inode
)->i_file_acl
, REQ_PRIO
);
1825 return PTR_ERR(bs
->bh
);
1826 ea_bdebug(bs
->bh
, "b_count=%d, refcount=%d",
1827 atomic_read(&(bs
->bh
->b_count
)),
1828 le32_to_cpu(BHDR(bs
->bh
)->h_refcount
));
1829 error
= ext4_xattr_check_block(inode
, bs
->bh
);
1832 /* Find the named attribute. */
1833 bs
->s
.base
= BHDR(bs
->bh
);
1834 bs
->s
.first
= BFIRST(bs
->bh
);
1835 bs
->s
.end
= bs
->bh
->b_data
+ bs
->bh
->b_size
;
1836 bs
->s
.here
= bs
->s
.first
;
1837 error
= xattr_find_entry(inode
, &bs
->s
.here
, bs
->s
.end
,
1838 i
->name_index
, i
->name
, 1);
1839 if (error
&& error
!= -ENODATA
)
1841 bs
->s
.not_found
= error
;
1847 ext4_xattr_block_set(handle_t
*handle
, struct inode
*inode
,
1848 struct ext4_xattr_info
*i
,
1849 struct ext4_xattr_block_find
*bs
)
1851 struct super_block
*sb
= inode
->i_sb
;
1852 struct buffer_head
*new_bh
= NULL
;
1853 struct ext4_xattr_search s_copy
= bs
->s
;
1854 struct ext4_xattr_search
*s
= &s_copy
;
1855 struct mb_cache_entry
*ce
= NULL
;
1857 struct mb_cache
*ea_block_cache
= EA_BLOCK_CACHE(inode
);
1858 struct inode
*ea_inode
= NULL
, *tmp_inode
;
1859 size_t old_ea_inode_quota
= 0;
1860 unsigned int ea_ino
;
1863 #define header(x) ((struct ext4_xattr_header *)(x))
1866 BUFFER_TRACE(bs
->bh
, "get_write_access");
1867 error
= ext4_journal_get_write_access(handle
, bs
->bh
);
1870 lock_buffer(bs
->bh
);
1872 if (header(s
->base
)->h_refcount
== cpu_to_le32(1)) {
1873 __u32 hash
= le32_to_cpu(BHDR(bs
->bh
)->h_hash
);
1876 * This must happen under buffer lock for
1877 * ext4_xattr_block_set() to reliably detect modified
1881 mb_cache_entry_delete(ea_block_cache
, hash
,
1883 ea_bdebug(bs
->bh
, "modifying in-place");
1884 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
,
1885 true /* is_block */);
1886 ext4_xattr_block_csum_set(inode
, bs
->bh
);
1887 unlock_buffer(bs
->bh
);
1888 if (error
== -EFSCORRUPTED
)
1891 error
= ext4_handle_dirty_metadata(handle
,
1898 int offset
= (char *)s
->here
- bs
->bh
->b_data
;
1900 unlock_buffer(bs
->bh
);
1901 ea_bdebug(bs
->bh
, "cloning");
1902 s
->base
= kmalloc(bs
->bh
->b_size
, GFP_NOFS
);
1904 if (s
->base
== NULL
)
1906 memcpy(s
->base
, BHDR(bs
->bh
), bs
->bh
->b_size
);
1907 s
->first
= ENTRY(header(s
->base
)+1);
1908 header(s
->base
)->h_refcount
= cpu_to_le32(1);
1909 s
->here
= ENTRY(s
->base
+ offset
);
1910 s
->end
= s
->base
+ bs
->bh
->b_size
;
1913 * If existing entry points to an xattr inode, we need
1914 * to prevent ext4_xattr_set_entry() from decrementing
1915 * ref count on it because the reference belongs to the
1916 * original block. In this case, make the entry look
1917 * like it has an empty value.
1919 if (!s
->not_found
&& s
->here
->e_value_inum
) {
1920 ea_ino
= le32_to_cpu(s
->here
->e_value_inum
);
1921 error
= ext4_xattr_inode_iget(inode
, ea_ino
,
1922 le32_to_cpu(s
->here
->e_hash
),
1927 if (!ext4_test_inode_state(tmp_inode
,
1928 EXT4_STATE_LUSTRE_EA_INODE
)) {
1930 * Defer quota free call for previous
1931 * inode until success is guaranteed.
1933 old_ea_inode_quota
= le32_to_cpu(
1934 s
->here
->e_value_size
);
1938 s
->here
->e_value_inum
= 0;
1939 s
->here
->e_value_size
= 0;
1943 /* Allocate a buffer where we construct the new block. */
1944 s
->base
= kzalloc(sb
->s_blocksize
, GFP_NOFS
);
1945 /* assert(header == s->base) */
1947 if (s
->base
== NULL
)
1949 header(s
->base
)->h_magic
= cpu_to_le32(EXT4_XATTR_MAGIC
);
1950 header(s
->base
)->h_blocks
= cpu_to_le32(1);
1951 header(s
->base
)->h_refcount
= cpu_to_le32(1);
1952 s
->first
= ENTRY(header(s
->base
)+1);
1953 s
->here
= ENTRY(header(s
->base
)+1);
1954 s
->end
= s
->base
+ sb
->s_blocksize
;
1957 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
, true /* is_block */);
1958 if (error
== -EFSCORRUPTED
)
1963 if (i
->value
&& s
->here
->e_value_inum
) {
1965 * A ref count on ea_inode has been taken as part of the call to
1966 * ext4_xattr_set_entry() above. We would like to drop this
1967 * extra ref but we have to wait until the xattr block is
1968 * initialized and has its own ref count on the ea_inode.
1970 ea_ino
= le32_to_cpu(s
->here
->e_value_inum
);
1971 error
= ext4_xattr_inode_iget(inode
, ea_ino
,
1972 le32_to_cpu(s
->here
->e_hash
),
1981 if (!IS_LAST_ENTRY(s
->first
)) {
1982 new_bh
= ext4_xattr_block_cache_find(inode
, header(s
->base
),
1985 /* We found an identical block in the cache. */
1986 if (new_bh
== bs
->bh
)
1987 ea_bdebug(new_bh
, "keeping");
1991 WARN_ON_ONCE(dquot_initialize_needed(inode
));
1993 /* The old block is released after updating
1995 error
= dquot_alloc_block(inode
,
1996 EXT4_C2B(EXT4_SB(sb
), 1));
1999 BUFFER_TRACE(new_bh
, "get_write_access");
2000 error
= ext4_journal_get_write_access(handle
,
2004 lock_buffer(new_bh
);
2006 * We have to be careful about races with
2007 * freeing, rehashing or adding references to
2008 * xattr block. Once we hold buffer lock xattr
2009 * block's state is stable so we can check
2010 * whether the block got freed / rehashed or
2011 * not. Since we unhash mbcache entry under
2012 * buffer lock when freeing / rehashing xattr
2013 * block, checking whether entry is still
2014 * hashed is reliable. Same rules hold for
2015 * e_reusable handling.
2017 if (hlist_bl_unhashed(&ce
->e_hash_list
) ||
2020 * Undo everything and check mbcache
2023 unlock_buffer(new_bh
);
2024 dquot_free_block(inode
,
2025 EXT4_C2B(EXT4_SB(sb
),
2028 mb_cache_entry_put(ea_block_cache
, ce
);
2033 ref
= le32_to_cpu(BHDR(new_bh
)->h_refcount
) + 1;
2034 BHDR(new_bh
)->h_refcount
= cpu_to_le32(ref
);
2035 if (ref
>= EXT4_XATTR_REFCOUNT_MAX
)
2037 ea_bdebug(new_bh
, "reusing; refcount now=%d",
2039 ext4_xattr_block_csum_set(inode
, new_bh
);
2040 unlock_buffer(new_bh
);
2041 error
= ext4_handle_dirty_metadata(handle
,
2047 mb_cache_entry_touch(ea_block_cache
, ce
);
2048 mb_cache_entry_put(ea_block_cache
, ce
);
2050 } else if (bs
->bh
&& s
->base
== bs
->bh
->b_data
) {
2051 /* We were modifying this block in-place. */
2052 ea_bdebug(bs
->bh
, "keeping this block");
2053 ext4_xattr_block_cache_insert(ea_block_cache
, bs
->bh
);
2057 /* We need to allocate a new block */
2058 ext4_fsblk_t goal
, block
;
2060 WARN_ON_ONCE(dquot_initialize_needed(inode
));
2062 goal
= ext4_group_first_block_no(sb
,
2063 EXT4_I(inode
)->i_block_group
);
2065 /* non-extent files can't have physical blocks past 2^32 */
2066 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
2067 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
2069 block
= ext4_new_meta_blocks(handle
, inode
, goal
, 0,
2074 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
2075 BUG_ON(block
> EXT4_MAX_BLOCK_FILE_PHYS
);
2077 ea_idebug(inode
, "creating block %llu",
2078 (unsigned long long)block
);
2080 new_bh
= sb_getblk(sb
, block
);
2081 if (unlikely(!new_bh
)) {
2084 ext4_free_blocks(handle
, inode
, NULL
, block
, 1,
2085 EXT4_FREE_BLOCKS_METADATA
);
2088 error
= ext4_xattr_inode_inc_ref_all(handle
, inode
,
2089 ENTRY(header(s
->base
)+1));
2093 /* Drop the extra ref on ea_inode. */
2094 error
= ext4_xattr_inode_dec_ref(handle
,
2097 ext4_warning_inode(ea_inode
,
2104 lock_buffer(new_bh
);
2105 error
= ext4_journal_get_create_access(handle
, new_bh
);
2107 unlock_buffer(new_bh
);
2111 memcpy(new_bh
->b_data
, s
->base
, new_bh
->b_size
);
2112 ext4_xattr_block_csum_set(inode
, new_bh
);
2113 set_buffer_uptodate(new_bh
);
2114 unlock_buffer(new_bh
);
2115 ext4_xattr_block_cache_insert(ea_block_cache
, new_bh
);
2116 error
= ext4_handle_dirty_metadata(handle
, inode
,
2123 if (old_ea_inode_quota
)
2124 ext4_xattr_inode_free_quota(inode
, NULL
, old_ea_inode_quota
);
2126 /* Update the inode. */
2127 EXT4_I(inode
)->i_file_acl
= new_bh
? new_bh
->b_blocknr
: 0;
2129 /* Drop the previous xattr block. */
2130 if (bs
->bh
&& bs
->bh
!= new_bh
) {
2131 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
2133 ext4_xattr_release_block(handle
, inode
, bs
->bh
,
2135 0 /* extra_credits */);
2136 ext4_xattr_inode_array_free(ea_inode_array
);
2144 error2
= ext4_xattr_inode_dec_ref(handle
, ea_inode
);
2146 ext4_warning_inode(ea_inode
, "dec ref error=%d",
2149 /* If there was an error, revert the quota charge. */
2151 ext4_xattr_inode_free_quota(inode
, ea_inode
,
2152 i_size_read(ea_inode
));
2156 mb_cache_entry_put(ea_block_cache
, ce
);
2158 if (!(bs
->bh
&& s
->base
== bs
->bh
->b_data
))
2164 dquot_free_block(inode
, EXT4_C2B(EXT4_SB(sb
), 1));
2168 EXT4_ERROR_INODE(inode
, "bad block %llu",
2169 EXT4_I(inode
)->i_file_acl
);
2175 int ext4_xattr_ibody_find(struct inode
*inode
, struct ext4_xattr_info
*i
,
2176 struct ext4_xattr_ibody_find
*is
)
2178 struct ext4_xattr_ibody_header
*header
;
2179 struct ext4_inode
*raw_inode
;
2182 if (EXT4_I(inode
)->i_extra_isize
== 0)
2184 raw_inode
= ext4_raw_inode(&is
->iloc
);
2185 header
= IHDR(inode
, raw_inode
);
2186 is
->s
.base
= is
->s
.first
= IFIRST(header
);
2187 is
->s
.here
= is
->s
.first
;
2188 is
->s
.end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
2189 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
2190 error
= xattr_check_inode(inode
, header
, is
->s
.end
);
2193 /* Find the named attribute. */
2194 error
= xattr_find_entry(inode
, &is
->s
.here
, is
->s
.end
,
2195 i
->name_index
, i
->name
, 0);
2196 if (error
&& error
!= -ENODATA
)
2198 is
->s
.not_found
= error
;
2203 int ext4_xattr_ibody_inline_set(handle_t
*handle
, struct inode
*inode
,
2204 struct ext4_xattr_info
*i
,
2205 struct ext4_xattr_ibody_find
*is
)
2207 struct ext4_xattr_ibody_header
*header
;
2208 struct ext4_xattr_search
*s
= &is
->s
;
2211 if (EXT4_I(inode
)->i_extra_isize
== 0)
2213 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
, false /* is_block */);
2216 header
= IHDR(inode
, ext4_raw_inode(&is
->iloc
));
2217 if (!IS_LAST_ENTRY(s
->first
)) {
2218 header
->h_magic
= cpu_to_le32(EXT4_XATTR_MAGIC
);
2219 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
2221 header
->h_magic
= cpu_to_le32(0);
2222 ext4_clear_inode_state(inode
, EXT4_STATE_XATTR
);
2227 static int ext4_xattr_ibody_set(handle_t
*handle
, struct inode
*inode
,
2228 struct ext4_xattr_info
*i
,
2229 struct ext4_xattr_ibody_find
*is
)
2231 struct ext4_xattr_ibody_header
*header
;
2232 struct ext4_xattr_search
*s
= &is
->s
;
2235 if (EXT4_I(inode
)->i_extra_isize
== 0)
2237 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
, false /* is_block */);
2240 header
= IHDR(inode
, ext4_raw_inode(&is
->iloc
));
2241 if (!IS_LAST_ENTRY(s
->first
)) {
2242 header
->h_magic
= cpu_to_le32(EXT4_XATTR_MAGIC
);
2243 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
2245 header
->h_magic
= cpu_to_le32(0);
2246 ext4_clear_inode_state(inode
, EXT4_STATE_XATTR
);
2251 static int ext4_xattr_value_same(struct ext4_xattr_search
*s
,
2252 struct ext4_xattr_info
*i
)
2256 /* When e_value_inum is set the value is stored externally. */
2257 if (s
->here
->e_value_inum
)
2259 if (le32_to_cpu(s
->here
->e_value_size
) != i
->value_len
)
2261 value
= ((void *)s
->base
) + le16_to_cpu(s
->here
->e_value_offs
);
2262 return !memcmp(value
, i
->value
, i
->value_len
);
2265 static struct buffer_head
*ext4_xattr_get_block(struct inode
*inode
)
2267 struct buffer_head
*bh
;
2270 if (!EXT4_I(inode
)->i_file_acl
)
2272 bh
= ext4_sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
, REQ_PRIO
);
2275 error
= ext4_xattr_check_block(inode
, bh
);
2278 return ERR_PTR(error
);
2284 * ext4_xattr_set_handle()
2286 * Create, replace or remove an extended attribute for this inode. Value
2287 * is NULL to remove an existing extended attribute, and non-NULL to
2288 * either replace an existing extended attribute, or create a new extended
2289 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2290 * specify that an extended attribute must exist and must not exist
2291 * previous to the call, respectively.
2293 * Returns 0, or a negative error number on failure.
2296 ext4_xattr_set_handle(handle_t
*handle
, struct inode
*inode
, int name_index
,
2297 const char *name
, const void *value
, size_t value_len
,
2300 struct ext4_xattr_info i
= {
2301 .name_index
= name_index
,
2304 .value_len
= value_len
,
2307 struct ext4_xattr_ibody_find is
= {
2308 .s
= { .not_found
= -ENODATA
, },
2310 struct ext4_xattr_block_find bs
= {
2311 .s
= { .not_found
= -ENODATA
, },
2318 if (strlen(name
) > 255)
2321 ext4_write_lock_xattr(inode
, &no_expand
);
2323 /* Check journal credits under write lock. */
2324 if (ext4_handle_valid(handle
)) {
2325 struct buffer_head
*bh
;
2328 bh
= ext4_xattr_get_block(inode
);
2330 error
= PTR_ERR(bh
);
2334 credits
= __ext4_xattr_set_credits(inode
->i_sb
, inode
, bh
,
2336 flags
& XATTR_CREATE
);
2339 if (!ext4_handle_has_enough_credits(handle
, credits
)) {
2345 error
= ext4_reserve_inode_write(handle
, inode
, &is
.iloc
);
2349 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
)) {
2350 struct ext4_inode
*raw_inode
= ext4_raw_inode(&is
.iloc
);
2351 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
2352 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
2355 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
2359 error
= ext4_xattr_block_find(inode
, &i
, &bs
);
2362 if (is
.s
.not_found
&& bs
.s
.not_found
) {
2364 if (flags
& XATTR_REPLACE
)
2371 if (flags
& XATTR_CREATE
)
2376 if (!is
.s
.not_found
)
2377 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, &is
);
2378 else if (!bs
.s
.not_found
)
2379 error
= ext4_xattr_block_set(handle
, inode
, &i
, &bs
);
2382 /* Xattr value did not change? Save us some work and bail out */
2383 if (!is
.s
.not_found
&& ext4_xattr_value_same(&is
.s
, &i
))
2385 if (!bs
.s
.not_found
&& ext4_xattr_value_same(&bs
.s
, &i
))
2388 if (ext4_has_feature_ea_inode(inode
->i_sb
) &&
2389 (EXT4_XATTR_SIZE(i
.value_len
) >
2390 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode
->i_sb
->s_blocksize
)))
2393 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, &is
);
2394 if (!error
&& !bs
.s
.not_found
) {
2396 error
= ext4_xattr_block_set(handle
, inode
, &i
, &bs
);
2397 } else if (error
== -ENOSPC
) {
2398 if (EXT4_I(inode
)->i_file_acl
&& !bs
.s
.base
) {
2401 error
= ext4_xattr_block_find(inode
, &i
, &bs
);
2405 error
= ext4_xattr_block_set(handle
, inode
, &i
, &bs
);
2406 if (!error
&& !is
.s
.not_found
) {
2408 error
= ext4_xattr_ibody_set(handle
, inode
, &i
,
2410 } else if (error
== -ENOSPC
) {
2412 * Xattr does not fit in the block, store at
2413 * external inode if possible.
2415 if (ext4_has_feature_ea_inode(inode
->i_sb
) &&
2424 ext4_xattr_update_super_block(handle
, inode
->i_sb
);
2425 inode
->i_ctime
= current_time(inode
);
2428 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
2430 * The bh is consumed by ext4_mark_iloc_dirty, even with
2435 ext4_handle_sync(handle
);
2441 ext4_write_unlock_xattr(inode
, &no_expand
);
2445 int ext4_xattr_set_credits(struct inode
*inode
, size_t value_len
,
2446 bool is_create
, int *credits
)
2448 struct buffer_head
*bh
;
2453 if (!EXT4_SB(inode
->i_sb
)->s_journal
)
2456 down_read(&EXT4_I(inode
)->xattr_sem
);
2458 bh
= ext4_xattr_get_block(inode
);
2462 *credits
= __ext4_xattr_set_credits(inode
->i_sb
, inode
, bh
,
2463 value_len
, is_create
);
2468 up_read(&EXT4_I(inode
)->xattr_sem
);
2475 * Like ext4_xattr_set_handle, but start from an inode. This extended
2476 * attribute modification is a filesystem transaction by itself.
2478 * Returns 0, or a negative error number on failure.
2481 ext4_xattr_set(struct inode
*inode
, int name_index
, const char *name
,
2482 const void *value
, size_t value_len
, int flags
)
2485 struct super_block
*sb
= inode
->i_sb
;
2486 int error
, retries
= 0;
2489 error
= dquot_initialize(inode
);
2494 error
= ext4_xattr_set_credits(inode
, value_len
, flags
& XATTR_CREATE
,
2499 handle
= ext4_journal_start(inode
, EXT4_HT_XATTR
, credits
);
2500 if (IS_ERR(handle
)) {
2501 error
= PTR_ERR(handle
);
2505 error
= ext4_xattr_set_handle(handle
, inode
, name_index
, name
,
2506 value
, value_len
, flags
);
2507 error2
= ext4_journal_stop(handle
);
2508 if (error
== -ENOSPC
&&
2509 ext4_should_retry_alloc(sb
, &retries
))
2519 * Shift the EA entries in the inode to create space for the increased
2522 static void ext4_xattr_shift_entries(struct ext4_xattr_entry
*entry
,
2523 int value_offs_shift
, void *to
,
2524 void *from
, size_t n
)
2526 struct ext4_xattr_entry
*last
= entry
;
2529 /* We always shift xattr headers further thus offsets get lower */
2530 BUG_ON(value_offs_shift
> 0);
2532 /* Adjust the value offsets of the entries */
2533 for (; !IS_LAST_ENTRY(last
); last
= EXT4_XATTR_NEXT(last
)) {
2534 if (!last
->e_value_inum
&& last
->e_value_size
) {
2535 new_offs
= le16_to_cpu(last
->e_value_offs
) +
2537 last
->e_value_offs
= cpu_to_le16(new_offs
);
2540 /* Shift the entries by n bytes */
2541 memmove(to
, from
, n
);
2545 * Move xattr pointed to by 'entry' from inode into external xattr block
2547 static int ext4_xattr_move_to_block(handle_t
*handle
, struct inode
*inode
,
2548 struct ext4_inode
*raw_inode
,
2549 struct ext4_xattr_entry
*entry
)
2551 struct ext4_xattr_ibody_find
*is
= NULL
;
2552 struct ext4_xattr_block_find
*bs
= NULL
;
2553 char *buffer
= NULL
, *b_entry_name
= NULL
;
2554 size_t value_size
= le32_to_cpu(entry
->e_value_size
);
2555 struct ext4_xattr_info i
= {
2558 .name_index
= entry
->e_name_index
,
2559 .in_inode
= !!entry
->e_value_inum
,
2561 struct ext4_xattr_ibody_header
*header
= IHDR(inode
, raw_inode
);
2564 is
= kzalloc(sizeof(struct ext4_xattr_ibody_find
), GFP_NOFS
);
2565 bs
= kzalloc(sizeof(struct ext4_xattr_block_find
), GFP_NOFS
);
2566 buffer
= kmalloc(value_size
, GFP_NOFS
);
2567 b_entry_name
= kmalloc(entry
->e_name_len
+ 1, GFP_NOFS
);
2568 if (!is
|| !bs
|| !buffer
|| !b_entry_name
) {
2573 is
->s
.not_found
= -ENODATA
;
2574 bs
->s
.not_found
= -ENODATA
;
2578 /* Save the entry name and the entry value */
2579 if (entry
->e_value_inum
) {
2580 error
= ext4_xattr_inode_get(inode
, entry
, buffer
, value_size
);
2584 size_t value_offs
= le16_to_cpu(entry
->e_value_offs
);
2585 memcpy(buffer
, (void *)IFIRST(header
) + value_offs
, value_size
);
2588 memcpy(b_entry_name
, entry
->e_name
, entry
->e_name_len
);
2589 b_entry_name
[entry
->e_name_len
] = '\0';
2590 i
.name
= b_entry_name
;
2592 error
= ext4_get_inode_loc(inode
, &is
->iloc
);
2596 error
= ext4_xattr_ibody_find(inode
, &i
, is
);
2600 /* Remove the chosen entry from the inode */
2601 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, is
);
2606 i
.value_len
= value_size
;
2607 error
= ext4_xattr_block_find(inode
, &i
, bs
);
2611 /* Add entry which was removed from the inode into the block */
2612 error
= ext4_xattr_block_set(handle
, inode
, &i
, bs
);
2617 kfree(b_entry_name
);
2620 brelse(is
->iloc
.bh
);
2629 static int ext4_xattr_make_inode_space(handle_t
*handle
, struct inode
*inode
,
2630 struct ext4_inode
*raw_inode
,
2631 int isize_diff
, size_t ifree
,
2632 size_t bfree
, int *total_ino
)
2634 struct ext4_xattr_ibody_header
*header
= IHDR(inode
, raw_inode
);
2635 struct ext4_xattr_entry
*small_entry
;
2636 struct ext4_xattr_entry
*entry
;
2637 struct ext4_xattr_entry
*last
;
2638 unsigned int entry_size
; /* EA entry size */
2639 unsigned int total_size
; /* EA entry size + value size */
2640 unsigned int min_total_size
;
2643 while (isize_diff
> ifree
) {
2646 min_total_size
= ~0U;
2647 last
= IFIRST(header
);
2648 /* Find the entry best suited to be pushed into EA block */
2649 for (; !IS_LAST_ENTRY(last
); last
= EXT4_XATTR_NEXT(last
)) {
2650 /* never move system.data out of the inode */
2651 if ((last
->e_name_len
== 4) &&
2652 (last
->e_name_index
== EXT4_XATTR_INDEX_SYSTEM
) &&
2653 !memcmp(last
->e_name
, "data", 4))
2655 total_size
= EXT4_XATTR_LEN(last
->e_name_len
);
2656 if (!last
->e_value_inum
)
2657 total_size
+= EXT4_XATTR_SIZE(
2658 le32_to_cpu(last
->e_value_size
));
2659 if (total_size
<= bfree
&&
2660 total_size
< min_total_size
) {
2661 if (total_size
+ ifree
< isize_diff
) {
2665 min_total_size
= total_size
;
2670 if (entry
== NULL
) {
2671 if (small_entry
== NULL
)
2673 entry
= small_entry
;
2676 entry_size
= EXT4_XATTR_LEN(entry
->e_name_len
);
2677 total_size
= entry_size
;
2678 if (!entry
->e_value_inum
)
2679 total_size
+= EXT4_XATTR_SIZE(
2680 le32_to_cpu(entry
->e_value_size
));
2681 error
= ext4_xattr_move_to_block(handle
, inode
, raw_inode
,
2686 *total_ino
-= entry_size
;
2687 ifree
+= total_size
;
2688 bfree
-= total_size
;
2695 * Expand an inode by new_extra_isize bytes when EAs are present.
2696 * Returns 0 on success or negative error number on failure.
2698 int ext4_expand_extra_isize_ea(struct inode
*inode
, int new_extra_isize
,
2699 struct ext4_inode
*raw_inode
, handle_t
*handle
)
2701 struct ext4_xattr_ibody_header
*header
;
2702 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2703 static unsigned int mnt_count
;
2705 size_t ifree
, bfree
;
2708 int error
= 0, tried_min_extra_isize
= 0;
2709 int s_min_extra_isize
= le16_to_cpu(sbi
->s_es
->s_min_extra_isize
);
2710 int isize_diff
; /* How much do we need to grow i_extra_isize */
2713 isize_diff
= new_extra_isize
- EXT4_I(inode
)->i_extra_isize
;
2714 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
2717 header
= IHDR(inode
, raw_inode
);
2720 * Check if enough free space is available in the inode to shift the
2721 * entries ahead by new_extra_isize.
2724 base
= IFIRST(header
);
2725 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
2726 min_offs
= end
- base
;
2727 total_ino
= sizeof(struct ext4_xattr_ibody_header
) + sizeof(u32
);
2729 error
= xattr_check_inode(inode
, header
, end
);
2733 ifree
= ext4_xattr_free_space(base
, &min_offs
, base
, &total_ino
);
2734 if (ifree
>= isize_diff
)
2738 * Enough free space isn't available in the inode, check if
2739 * EA block can hold new_extra_isize bytes.
2741 if (EXT4_I(inode
)->i_file_acl
) {
2742 struct buffer_head
*bh
;
2744 bh
= ext4_sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
, REQ_PRIO
);
2746 error
= PTR_ERR(bh
);
2749 error
= ext4_xattr_check_block(inode
, bh
);
2755 end
= bh
->b_data
+ bh
->b_size
;
2756 min_offs
= end
- base
;
2757 bfree
= ext4_xattr_free_space(BFIRST(bh
), &min_offs
, base
,
2760 if (bfree
+ ifree
< isize_diff
) {
2761 if (!tried_min_extra_isize
&& s_min_extra_isize
) {
2762 tried_min_extra_isize
++;
2763 new_extra_isize
= s_min_extra_isize
;
2770 bfree
= inode
->i_sb
->s_blocksize
;
2773 error
= ext4_xattr_make_inode_space(handle
, inode
, raw_inode
,
2774 isize_diff
, ifree
, bfree
,
2777 if (error
== -ENOSPC
&& !tried_min_extra_isize
&&
2778 s_min_extra_isize
) {
2779 tried_min_extra_isize
++;
2780 new_extra_isize
= s_min_extra_isize
;
2786 /* Adjust the offsets and shift the remaining entries ahead */
2787 ext4_xattr_shift_entries(IFIRST(header
), EXT4_I(inode
)->i_extra_isize
2788 - new_extra_isize
, (void *)raw_inode
+
2789 EXT4_GOOD_OLD_INODE_SIZE
+ new_extra_isize
,
2790 (void *)header
, total_ino
);
2791 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
2794 if (error
&& (mnt_count
!= le16_to_cpu(sbi
->s_es
->s_mnt_count
))) {
2795 ext4_warning(inode
->i_sb
, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2797 mnt_count
= le16_to_cpu(sbi
->s_es
->s_mnt_count
);
2802 #define EIA_INCR 16 /* must be 2^n */
2803 #define EIA_MASK (EIA_INCR - 1)
2805 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2806 * If @ea_inode_array is new or full it will be grown and the old
2807 * contents copied over.
2810 ext4_expand_inode_array(struct ext4_xattr_inode_array
**ea_inode_array
,
2811 struct inode
*inode
)
2813 if (*ea_inode_array
== NULL
) {
2815 * Start with 15 inodes, so it fits into a power-of-two size.
2816 * If *ea_inode_array is NULL, this is essentially offsetof()
2819 kmalloc(offsetof(struct ext4_xattr_inode_array
,
2822 if (*ea_inode_array
== NULL
)
2824 (*ea_inode_array
)->count
= 0;
2825 } else if (((*ea_inode_array
)->count
& EIA_MASK
) == EIA_MASK
) {
2826 /* expand the array once all 15 + n * 16 slots are full */
2827 struct ext4_xattr_inode_array
*new_array
= NULL
;
2828 int count
= (*ea_inode_array
)->count
;
2830 /* if new_array is NULL, this is essentially offsetof() */
2831 new_array
= kmalloc(
2832 offsetof(struct ext4_xattr_inode_array
,
2833 inodes
[count
+ EIA_INCR
]),
2835 if (new_array
== NULL
)
2837 memcpy(new_array
, *ea_inode_array
,
2838 offsetof(struct ext4_xattr_inode_array
, inodes
[count
]));
2839 kfree(*ea_inode_array
);
2840 *ea_inode_array
= new_array
;
2842 (*ea_inode_array
)->inodes
[(*ea_inode_array
)->count
++] = inode
;
2847 * ext4_xattr_delete_inode()
2849 * Free extended attribute resources associated with this inode. Traverse
2850 * all entries and decrement reference on any xattr inodes associated with this
2851 * inode. This is called immediately before an inode is freed. We have exclusive
2852 * access to the inode. If an orphan inode is deleted it will also release its
2853 * references on xattr block and xattr inodes.
2855 int ext4_xattr_delete_inode(handle_t
*handle
, struct inode
*inode
,
2856 struct ext4_xattr_inode_array
**ea_inode_array
,
2859 struct buffer_head
*bh
= NULL
;
2860 struct ext4_xattr_ibody_header
*header
;
2861 struct ext4_iloc iloc
= { .bh
= NULL
};
2862 struct ext4_xattr_entry
*entry
;
2863 struct inode
*ea_inode
;
2866 error
= ext4_xattr_ensure_credits(handle
, inode
, extra_credits
,
2869 false /* block_csum */);
2871 EXT4_ERROR_INODE(inode
, "ensure credits (error %d)", error
);
2875 if (ext4_has_feature_ea_inode(inode
->i_sb
) &&
2876 ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
2878 error
= ext4_get_inode_loc(inode
, &iloc
);
2880 EXT4_ERROR_INODE(inode
, "inode loc (error %d)", error
);
2884 error
= ext4_journal_get_write_access(handle
, iloc
.bh
);
2886 EXT4_ERROR_INODE(inode
, "write access (error %d)",
2891 header
= IHDR(inode
, ext4_raw_inode(&iloc
));
2892 if (header
->h_magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
2893 ext4_xattr_inode_dec_ref_all(handle
, inode
, iloc
.bh
,
2895 false /* block_csum */,
2898 false /* skip_quota */);
2901 if (EXT4_I(inode
)->i_file_acl
) {
2902 bh
= ext4_sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
, REQ_PRIO
);
2904 error
= PTR_ERR(bh
);
2906 EXT4_ERROR_INODE(inode
, "block %llu read error",
2907 EXT4_I(inode
)->i_file_acl
);
2910 error
= ext4_xattr_check_block(inode
, bh
);
2914 if (ext4_has_feature_ea_inode(inode
->i_sb
)) {
2915 for (entry
= BFIRST(bh
); !IS_LAST_ENTRY(entry
);
2916 entry
= EXT4_XATTR_NEXT(entry
)) {
2917 if (!entry
->e_value_inum
)
2919 error
= ext4_xattr_inode_iget(inode
,
2920 le32_to_cpu(entry
->e_value_inum
),
2921 le32_to_cpu(entry
->e_hash
),
2925 ext4_xattr_inode_free_quota(inode
, ea_inode
,
2926 le32_to_cpu(entry
->e_value_size
));
2932 ext4_xattr_release_block(handle
, inode
, bh
, ea_inode_array
,
2935 * Update i_file_acl value in the same transaction that releases
2938 EXT4_I(inode
)->i_file_acl
= 0;
2939 error
= ext4_mark_inode_dirty(handle
, inode
);
2941 EXT4_ERROR_INODE(inode
, "mark inode dirty (error %d)",
2953 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array
*ea_inode_array
)
2957 if (ea_inode_array
== NULL
)
2960 for (idx
= 0; idx
< ea_inode_array
->count
; ++idx
)
2961 iput(ea_inode_array
->inodes
[idx
]);
2962 kfree(ea_inode_array
);
2966 * ext4_xattr_block_cache_insert()
2968 * Create a new entry in the extended attribute block cache, and insert
2969 * it unless such an entry is already in the cache.
2971 * Returns 0, or a negative error number on failure.
2974 ext4_xattr_block_cache_insert(struct mb_cache
*ea_block_cache
,
2975 struct buffer_head
*bh
)
2977 struct ext4_xattr_header
*header
= BHDR(bh
);
2978 __u32 hash
= le32_to_cpu(header
->h_hash
);
2979 int reusable
= le32_to_cpu(header
->h_refcount
) <
2980 EXT4_XATTR_REFCOUNT_MAX
;
2983 if (!ea_block_cache
)
2985 error
= mb_cache_entry_create(ea_block_cache
, GFP_NOFS
, hash
,
2986 bh
->b_blocknr
, reusable
);
2988 if (error
== -EBUSY
)
2989 ea_bdebug(bh
, "already in cache");
2991 ea_bdebug(bh
, "inserting [%x]", (int)hash
);
2997 * Compare two extended attribute blocks for equality.
2999 * Returns 0 if the blocks are equal, 1 if they differ, and
3000 * a negative error number on errors.
3003 ext4_xattr_cmp(struct ext4_xattr_header
*header1
,
3004 struct ext4_xattr_header
*header2
)
3006 struct ext4_xattr_entry
*entry1
, *entry2
;
3008 entry1
= ENTRY(header1
+1);
3009 entry2
= ENTRY(header2
+1);
3010 while (!IS_LAST_ENTRY(entry1
)) {
3011 if (IS_LAST_ENTRY(entry2
))
3013 if (entry1
->e_hash
!= entry2
->e_hash
||
3014 entry1
->e_name_index
!= entry2
->e_name_index
||
3015 entry1
->e_name_len
!= entry2
->e_name_len
||
3016 entry1
->e_value_size
!= entry2
->e_value_size
||
3017 entry1
->e_value_inum
!= entry2
->e_value_inum
||
3018 memcmp(entry1
->e_name
, entry2
->e_name
, entry1
->e_name_len
))
3020 if (!entry1
->e_value_inum
&&
3021 memcmp((char *)header1
+ le16_to_cpu(entry1
->e_value_offs
),
3022 (char *)header2
+ le16_to_cpu(entry2
->e_value_offs
),
3023 le32_to_cpu(entry1
->e_value_size
)))
3026 entry1
= EXT4_XATTR_NEXT(entry1
);
3027 entry2
= EXT4_XATTR_NEXT(entry2
);
3029 if (!IS_LAST_ENTRY(entry2
))
3035 * ext4_xattr_block_cache_find()
3037 * Find an identical extended attribute block.
3039 * Returns a pointer to the block found, or NULL if such a block was
3040 * not found or an error occurred.
3042 static struct buffer_head
*
3043 ext4_xattr_block_cache_find(struct inode
*inode
,
3044 struct ext4_xattr_header
*header
,
3045 struct mb_cache_entry
**pce
)
3047 __u32 hash
= le32_to_cpu(header
->h_hash
);
3048 struct mb_cache_entry
*ce
;
3049 struct mb_cache
*ea_block_cache
= EA_BLOCK_CACHE(inode
);
3051 if (!ea_block_cache
)
3053 if (!header
->h_hash
)
3054 return NULL
; /* never share */
3055 ea_idebug(inode
, "looking for cached blocks [%x]", (int)hash
);
3056 ce
= mb_cache_entry_find_first(ea_block_cache
, hash
);
3058 struct buffer_head
*bh
;
3060 bh
= ext4_sb_bread(inode
->i_sb
, ce
->e_value
, REQ_PRIO
);
3062 if (PTR_ERR(bh
) == -ENOMEM
)
3064 EXT4_ERROR_INODE(inode
, "block %lu read error",
3065 (unsigned long)ce
->e_value
);
3066 } else if (ext4_xattr_cmp(header
, BHDR(bh
)) == 0) {
3071 ce
= mb_cache_entry_find_next(ea_block_cache
, ce
);
3076 #define NAME_HASH_SHIFT 5
3077 #define VALUE_HASH_SHIFT 16
3080 * ext4_xattr_hash_entry()
3082 * Compute the hash of an extended attribute.
3084 static __le32
ext4_xattr_hash_entry(char *name
, size_t name_len
, __le32
*value
,
3089 while (name_len
--) {
3090 hash
= (hash
<< NAME_HASH_SHIFT
) ^
3091 (hash
>> (8*sizeof(hash
) - NAME_HASH_SHIFT
)) ^
3094 while (value_count
--) {
3095 hash
= (hash
<< VALUE_HASH_SHIFT
) ^
3096 (hash
>> (8*sizeof(hash
) - VALUE_HASH_SHIFT
)) ^
3097 le32_to_cpu(*value
++);
3099 return cpu_to_le32(hash
);
3102 #undef NAME_HASH_SHIFT
3103 #undef VALUE_HASH_SHIFT
3105 #define BLOCK_HASH_SHIFT 16
3108 * ext4_xattr_rehash()
3110 * Re-compute the extended attribute hash value after an entry has changed.
3112 static void ext4_xattr_rehash(struct ext4_xattr_header
*header
)
3114 struct ext4_xattr_entry
*here
;
3117 here
= ENTRY(header
+1);
3118 while (!IS_LAST_ENTRY(here
)) {
3119 if (!here
->e_hash
) {
3120 /* Block is not shared if an entry's hash value == 0 */
3124 hash
= (hash
<< BLOCK_HASH_SHIFT
) ^
3125 (hash
>> (8*sizeof(hash
) - BLOCK_HASH_SHIFT
)) ^
3126 le32_to_cpu(here
->e_hash
);
3127 here
= EXT4_XATTR_NEXT(here
);
3129 header
->h_hash
= cpu_to_le32(hash
);
3132 #undef BLOCK_HASH_SHIFT
3134 #define HASH_BUCKET_BITS 10
3137 ext4_xattr_create_cache(void)
3139 return mb_cache_create(HASH_BUCKET_BITS
);
3142 void ext4_xattr_destroy_cache(struct mb_cache
*cache
)
3145 mb_cache_destroy(cache
);