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
;
195 /* Check the values */
196 while (!IS_LAST_ENTRY(entry
)) {
197 u32 size
= le32_to_cpu(entry
->e_value_size
);
199 if (size
> EXT4_XATTR_SIZE_MAX
)
200 return -EFSCORRUPTED
;
202 if (size
!= 0 && entry
->e_value_inum
== 0) {
203 u16 offs
= le16_to_cpu(entry
->e_value_offs
);
207 * The value cannot overlap the names, and the value
208 * with padding cannot extend beyond 'end'. Check both
209 * the padded and unpadded sizes, since the size may
210 * overflow to 0 when adding padding.
212 if (offs
> end
- value_start
)
213 return -EFSCORRUPTED
;
214 value
= value_start
+ offs
;
215 if (value
< (void *)e
+ sizeof(u32
) ||
216 size
> end
- value
||
217 EXT4_XATTR_SIZE(size
) > end
- value
)
218 return -EFSCORRUPTED
;
220 entry
= EXT4_XATTR_NEXT(entry
);
227 __ext4_xattr_check_block(struct inode
*inode
, struct buffer_head
*bh
,
228 const char *function
, unsigned int line
)
230 int error
= -EFSCORRUPTED
;
232 if (buffer_verified(bh
))
235 if (BHDR(bh
)->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
) ||
236 BHDR(bh
)->h_blocks
!= cpu_to_le32(1))
239 if (!ext4_xattr_block_csum_verify(inode
, bh
))
241 error
= ext4_xattr_check_entries(BFIRST(bh
), bh
->b_data
+ bh
->b_size
,
245 __ext4_error_inode(inode
, function
, line
, 0,
246 "corrupted xattr block %llu",
247 (unsigned long long) bh
->b_blocknr
);
249 set_buffer_verified(bh
);
253 #define ext4_xattr_check_block(inode, bh) \
254 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
258 __xattr_check_inode(struct inode
*inode
, struct ext4_xattr_ibody_header
*header
,
259 void *end
, const char *function
, unsigned int line
)
261 int error
= -EFSCORRUPTED
;
263 if (end
- (void *)header
< sizeof(*header
) + sizeof(u32
) ||
264 (header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)))
266 error
= ext4_xattr_check_entries(IFIRST(header
), end
, IFIRST(header
));
269 __ext4_error_inode(inode
, function
, line
, 0,
270 "corrupted in-inode xattr");
274 #define xattr_check_inode(inode, header, end) \
275 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
278 xattr_find_entry(struct inode
*inode
, struct ext4_xattr_entry
**pentry
,
279 void *end
, int name_index
, const char *name
, int sorted
)
281 struct ext4_xattr_entry
*entry
, *next
;
287 name_len
= strlen(name
);
288 for (entry
= *pentry
; !IS_LAST_ENTRY(entry
); entry
= next
) {
289 next
= EXT4_XATTR_NEXT(entry
);
290 if ((void *) next
>= end
) {
291 EXT4_ERROR_INODE(inode
, "corrupted xattr entries");
292 return -EFSCORRUPTED
;
294 cmp
= name_index
- entry
->e_name_index
;
296 cmp
= name_len
- entry
->e_name_len
;
298 cmp
= memcmp(name
, entry
->e_name
, name_len
);
299 if (cmp
<= 0 && (sorted
|| cmp
== 0))
303 return cmp
? -ENODATA
: 0;
307 ext4_xattr_inode_hash(struct ext4_sb_info
*sbi
, const void *buffer
, size_t size
)
309 return ext4_chksum(sbi
, sbi
->s_csum_seed
, buffer
, size
);
312 static u64
ext4_xattr_inode_get_ref(struct inode
*ea_inode
)
314 return ((u64
)ea_inode
->i_ctime
.tv_sec
<< 32) |
315 ((u32
)ea_inode
->i_version
);
318 static void ext4_xattr_inode_set_ref(struct inode
*ea_inode
, u64 ref_count
)
320 ea_inode
->i_ctime
.tv_sec
= (u32
)(ref_count
>> 32);
321 ea_inode
->i_version
= (u32
)ref_count
;
324 static u32
ext4_xattr_inode_get_hash(struct inode
*ea_inode
)
326 return (u32
)ea_inode
->i_atime
.tv_sec
;
329 static void ext4_xattr_inode_set_hash(struct inode
*ea_inode
, u32 hash
)
331 ea_inode
->i_atime
.tv_sec
= hash
;
335 * Read the EA value from an inode.
337 static int ext4_xattr_inode_read(struct inode
*ea_inode
, void *buf
, size_t size
)
339 int blocksize
= 1 << ea_inode
->i_blkbits
;
340 int bh_count
= (size
+ blocksize
- 1) >> ea_inode
->i_blkbits
;
341 int tail_size
= (size
% blocksize
) ?: blocksize
;
342 struct buffer_head
*bhs_inline
[8];
343 struct buffer_head
**bhs
= bhs_inline
;
346 if (bh_count
> ARRAY_SIZE(bhs_inline
)) {
347 bhs
= kmalloc_array(bh_count
, sizeof(*bhs
), GFP_NOFS
);
352 ret
= ext4_bread_batch(ea_inode
, 0 /* block */, bh_count
,
353 true /* wait */, bhs
);
357 for (i
= 0; i
< bh_count
; i
++) {
358 /* There shouldn't be any holes in ea_inode. */
363 memcpy((char *)buf
+ blocksize
* i
, bhs
[i
]->b_data
,
364 i
< bh_count
- 1 ? blocksize
: tail_size
);
368 for (i
= 0; i
< bh_count
; i
++)
371 if (bhs
!= bhs_inline
)
376 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
378 static int ext4_xattr_inode_iget(struct inode
*parent
, unsigned long ea_ino
,
379 u32 ea_inode_hash
, struct inode
**ea_inode
)
384 inode
= ext4_iget(parent
->i_sb
, ea_ino
);
386 err
= PTR_ERR(inode
);
387 ext4_error(parent
->i_sb
,
388 "error while reading EA inode %lu err=%d", ea_ino
,
393 if (is_bad_inode(inode
)) {
394 ext4_error(parent
->i_sb
,
395 "error while reading EA inode %lu is_bad_inode",
401 if (!(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
)) {
402 ext4_error(parent
->i_sb
,
403 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
409 ext4_xattr_inode_set_class(inode
);
412 * Check whether this is an old Lustre-style xattr inode. Lustre
413 * implementation does not have hash validation, rather it has a
414 * backpointer from ea_inode to the parent inode.
416 if (ea_inode_hash
!= ext4_xattr_inode_get_hash(inode
) &&
417 EXT4_XATTR_INODE_GET_PARENT(inode
) == parent
->i_ino
&&
418 inode
->i_generation
== parent
->i_generation
) {
419 ext4_set_inode_state(inode
, EXT4_STATE_LUSTRE_EA_INODE
);
420 ext4_xattr_inode_set_ref(inode
, 1);
423 inode
->i_flags
|= S_NOQUOTA
;
435 ext4_xattr_inode_verify_hashes(struct inode
*ea_inode
,
436 struct ext4_xattr_entry
*entry
, void *buffer
,
441 /* Verify stored hash matches calculated hash. */
442 hash
= ext4_xattr_inode_hash(EXT4_SB(ea_inode
->i_sb
), buffer
, size
);
443 if (hash
!= ext4_xattr_inode_get_hash(ea_inode
))
444 return -EFSCORRUPTED
;
447 __le32 e_hash
, tmp_data
;
449 /* Verify entry hash. */
450 tmp_data
= cpu_to_le32(hash
);
451 e_hash
= ext4_xattr_hash_entry(entry
->e_name
, entry
->e_name_len
,
453 if (e_hash
!= entry
->e_hash
)
454 return -EFSCORRUPTED
;
460 * Read xattr value from the EA inode.
463 ext4_xattr_inode_get(struct inode
*inode
, struct ext4_xattr_entry
*entry
,
464 void *buffer
, size_t size
)
466 struct mb_cache
*ea_inode_cache
= EA_INODE_CACHE(inode
);
467 struct inode
*ea_inode
;
470 err
= ext4_xattr_inode_iget(inode
, le32_to_cpu(entry
->e_value_inum
),
471 le32_to_cpu(entry
->e_hash
), &ea_inode
);
477 if (i_size_read(ea_inode
) != size
) {
478 ext4_warning_inode(ea_inode
,
479 "ea_inode file size=%llu entry size=%zu",
480 i_size_read(ea_inode
), size
);
485 err
= ext4_xattr_inode_read(ea_inode
, buffer
, size
);
489 if (!ext4_test_inode_state(ea_inode
, EXT4_STATE_LUSTRE_EA_INODE
)) {
490 err
= ext4_xattr_inode_verify_hashes(ea_inode
, entry
, buffer
,
493 ext4_warning_inode(ea_inode
,
494 "EA inode hash validation failed");
499 mb_cache_entry_create(ea_inode_cache
, GFP_NOFS
,
500 ext4_xattr_inode_get_hash(ea_inode
),
501 ea_inode
->i_ino
, true /* reusable */);
509 ext4_xattr_block_get(struct inode
*inode
, int name_index
, const char *name
,
510 void *buffer
, size_t buffer_size
)
512 struct buffer_head
*bh
= NULL
;
513 struct ext4_xattr_entry
*entry
;
517 struct mb_cache
*ea_block_cache
= EA_BLOCK_CACHE(inode
);
519 ea_idebug(inode
, "name=%d.%s, buffer=%p, buffer_size=%ld",
520 name_index
, name
, buffer
, (long)buffer_size
);
523 if (!EXT4_I(inode
)->i_file_acl
)
525 ea_idebug(inode
, "reading block %llu",
526 (unsigned long long)EXT4_I(inode
)->i_file_acl
);
527 bh
= sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
);
530 ea_bdebug(bh
, "b_count=%d, refcount=%d",
531 atomic_read(&(bh
->b_count
)), le32_to_cpu(BHDR(bh
)->h_refcount
));
532 error
= ext4_xattr_check_block(inode
, bh
);
535 ext4_xattr_block_cache_insert(ea_block_cache
, bh
);
537 end
= bh
->b_data
+ bh
->b_size
;
538 error
= xattr_find_entry(inode
, &entry
, end
, name_index
, name
, 1);
541 size
= le32_to_cpu(entry
->e_value_size
);
543 if (unlikely(size
> EXT4_XATTR_SIZE_MAX
))
546 if (size
> buffer_size
)
548 if (entry
->e_value_inum
) {
549 error
= ext4_xattr_inode_get(inode
, entry
, buffer
,
554 u16 offset
= le16_to_cpu(entry
->e_value_offs
);
555 void *p
= bh
->b_data
+ offset
;
557 if (unlikely(p
+ size
> end
))
559 memcpy(buffer
, p
, size
);
570 ext4_xattr_ibody_get(struct inode
*inode
, int name_index
, const char *name
,
571 void *buffer
, size_t buffer_size
)
573 struct ext4_xattr_ibody_header
*header
;
574 struct ext4_xattr_entry
*entry
;
575 struct ext4_inode
*raw_inode
;
576 struct ext4_iloc iloc
;
581 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
))
583 error
= ext4_get_inode_loc(inode
, &iloc
);
586 raw_inode
= ext4_raw_inode(&iloc
);
587 header
= IHDR(inode
, raw_inode
);
588 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
589 error
= xattr_check_inode(inode
, header
, end
);
592 entry
= IFIRST(header
);
593 error
= xattr_find_entry(inode
, &entry
, end
, name_index
, name
, 0);
596 size
= le32_to_cpu(entry
->e_value_size
);
598 if (unlikely(size
> EXT4_XATTR_SIZE_MAX
))
601 if (size
> buffer_size
)
603 if (entry
->e_value_inum
) {
604 error
= ext4_xattr_inode_get(inode
, entry
, buffer
,
609 u16 offset
= le16_to_cpu(entry
->e_value_offs
);
610 void *p
= (void *)IFIRST(header
) + offset
;
612 if (unlikely(p
+ size
> end
))
614 memcpy(buffer
, p
, size
);
627 * Copy an extended attribute into the buffer
628 * provided, or compute the buffer size required.
629 * Buffer is NULL to compute the size of the buffer required.
631 * Returns a negative error number on failure, or the number of bytes
632 * used / required on success.
635 ext4_xattr_get(struct inode
*inode
, int name_index
, const char *name
,
636 void *buffer
, size_t buffer_size
)
640 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
643 if (strlen(name
) > 255)
646 down_read(&EXT4_I(inode
)->xattr_sem
);
647 error
= ext4_xattr_ibody_get(inode
, name_index
, name
, buffer
,
649 if (error
== -ENODATA
)
650 error
= ext4_xattr_block_get(inode
, name_index
, name
, buffer
,
652 up_read(&EXT4_I(inode
)->xattr_sem
);
657 ext4_xattr_list_entries(struct dentry
*dentry
, struct ext4_xattr_entry
*entry
,
658 char *buffer
, size_t buffer_size
)
660 size_t rest
= buffer_size
;
662 for (; !IS_LAST_ENTRY(entry
); entry
= EXT4_XATTR_NEXT(entry
)) {
663 const struct xattr_handler
*handler
=
664 ext4_xattr_handler(entry
->e_name_index
);
666 if (handler
&& (!handler
->list
|| handler
->list(dentry
))) {
667 const char *prefix
= handler
->prefix
?: handler
->name
;
668 size_t prefix_len
= strlen(prefix
);
669 size_t size
= prefix_len
+ entry
->e_name_len
+ 1;
674 memcpy(buffer
, prefix
, prefix_len
);
675 buffer
+= prefix_len
;
676 memcpy(buffer
, entry
->e_name
, entry
->e_name_len
);
677 buffer
+= entry
->e_name_len
;
683 return buffer_size
- rest
; /* total size */
687 ext4_xattr_block_list(struct dentry
*dentry
, char *buffer
, size_t buffer_size
)
689 struct inode
*inode
= d_inode(dentry
);
690 struct buffer_head
*bh
= NULL
;
693 ea_idebug(inode
, "buffer=%p, buffer_size=%ld",
694 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
= sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
);
705 ea_bdebug(bh
, "b_count=%d, refcount=%d",
706 atomic_read(&(bh
->b_count
)), le32_to_cpu(BHDR(bh
)->h_refcount
));
707 error
= ext4_xattr_check_block(inode
, bh
);
710 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode
), bh
);
711 error
= ext4_xattr_list_entries(dentry
, BFIRST(bh
), buffer
, buffer_size
);
720 ext4_xattr_ibody_list(struct dentry
*dentry
, char *buffer
, size_t buffer_size
)
722 struct inode
*inode
= d_inode(dentry
);
723 struct ext4_xattr_ibody_header
*header
;
724 struct ext4_inode
*raw_inode
;
725 struct ext4_iloc iloc
;
729 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
))
731 error
= ext4_get_inode_loc(inode
, &iloc
);
734 raw_inode
= ext4_raw_inode(&iloc
);
735 header
= IHDR(inode
, raw_inode
);
736 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
737 error
= xattr_check_inode(inode
, header
, end
);
740 error
= ext4_xattr_list_entries(dentry
, IFIRST(header
),
741 buffer
, buffer_size
);
749 * Inode operation listxattr()
751 * d_inode(dentry)->i_rwsem: don't care
753 * Copy a list of attribute names into the buffer
754 * provided, or compute the buffer size required.
755 * Buffer is NULL to compute the size of the buffer required.
757 * Returns a negative error number on failure, or the number of bytes
758 * used / required on success.
761 ext4_listxattr(struct dentry
*dentry
, char *buffer
, size_t buffer_size
)
765 down_read(&EXT4_I(d_inode(dentry
))->xattr_sem
);
766 ret
= ret2
= ext4_xattr_ibody_list(dentry
, buffer
, buffer_size
);
773 ret
= ext4_xattr_block_list(dentry
, buffer
, buffer_size
);
778 up_read(&EXT4_I(d_inode(dentry
))->xattr_sem
);
783 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
786 static void ext4_xattr_update_super_block(handle_t
*handle
,
787 struct super_block
*sb
)
789 if (ext4_has_feature_xattr(sb
))
792 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
793 if (ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
) == 0) {
794 ext4_set_feature_xattr(sb
);
795 ext4_handle_dirty_super(handle
, sb
);
799 int ext4_get_inode_usage(struct inode
*inode
, qsize_t
*usage
)
801 struct ext4_iloc iloc
= { .bh
= NULL
};
802 struct buffer_head
*bh
= NULL
;
803 struct ext4_inode
*raw_inode
;
804 struct ext4_xattr_ibody_header
*header
;
805 struct ext4_xattr_entry
*entry
;
806 qsize_t ea_inode_refs
= 0;
810 lockdep_assert_held_read(&EXT4_I(inode
)->xattr_sem
);
812 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
813 ret
= ext4_get_inode_loc(inode
, &iloc
);
816 raw_inode
= ext4_raw_inode(&iloc
);
817 header
= IHDR(inode
, raw_inode
);
818 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
819 ret
= xattr_check_inode(inode
, header
, end
);
823 for (entry
= IFIRST(header
); !IS_LAST_ENTRY(entry
);
824 entry
= EXT4_XATTR_NEXT(entry
))
825 if (entry
->e_value_inum
)
829 if (EXT4_I(inode
)->i_file_acl
) {
830 bh
= sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
);
836 ret
= ext4_xattr_check_block(inode
, bh
);
840 for (entry
= BFIRST(bh
); !IS_LAST_ENTRY(entry
);
841 entry
= EXT4_XATTR_NEXT(entry
))
842 if (entry
->e_value_inum
)
845 *usage
= ea_inode_refs
+ 1;
853 static inline size_t round_up_cluster(struct inode
*inode
, size_t length
)
855 struct super_block
*sb
= inode
->i_sb
;
856 size_t cluster_size
= 1 << (EXT4_SB(sb
)->s_cluster_bits
+
858 size_t mask
= ~(cluster_size
- 1);
860 return (length
+ cluster_size
- 1) & mask
;
863 static int ext4_xattr_inode_alloc_quota(struct inode
*inode
, size_t len
)
867 err
= dquot_alloc_inode(inode
);
870 err
= dquot_alloc_space_nodirty(inode
, round_up_cluster(inode
, len
));
872 dquot_free_inode(inode
);
876 static void ext4_xattr_inode_free_quota(struct inode
*parent
,
877 struct inode
*ea_inode
,
881 ext4_test_inode_state(ea_inode
, EXT4_STATE_LUSTRE_EA_INODE
))
883 dquot_free_space_nodirty(parent
, round_up_cluster(parent
, len
));
884 dquot_free_inode(parent
);
887 int __ext4_xattr_set_credits(struct super_block
*sb
, struct inode
*inode
,
888 struct buffer_head
*block_bh
, size_t value_len
,
895 * 1) Owner inode update
896 * 2) Ref count update on old xattr block
898 * 4) block bitmap update for new xattr block
899 * 5) group descriptor for new xattr block
900 * 6) block bitmap update for old xattr block
901 * 7) group descriptor for old block
903 * 6 & 7 can happen if we have two racing threads T_a and T_b
904 * which are each trying to set an xattr on inodes I_a and I_b
905 * which were both initially sharing an xattr block.
910 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(sb
);
913 * In case of inline data, we may push out the data to a block,
914 * so we need to reserve credits for this eventuality
916 if (inode
&& ext4_has_inline_data(inode
))
917 credits
+= ext4_writepage_trans_blocks(inode
) + 1;
919 /* We are done if ea_inode feature is not enabled. */
920 if (!ext4_has_feature_ea_inode(sb
))
923 /* New ea_inode, inode map, block bitmap, group descriptor. */
927 blocks
= (value_len
+ sb
->s_blocksize
- 1) >> sb
->s_blocksize_bits
;
929 /* Indirection block or one level of extent tree. */
932 /* Block bitmap and group descriptor updates for each block. */
933 credits
+= blocks
* 2;
935 /* Blocks themselves. */
939 /* Dereference ea_inode holding old xattr value.
940 * Old ea_inode, inode map, block bitmap, group descriptor.
944 /* Data blocks for old ea_inode. */
945 blocks
= XATTR_SIZE_MAX
>> sb
->s_blocksize_bits
;
947 /* Indirection block or one level of extent tree for old
952 /* Block bitmap and group descriptor updates for each block. */
953 credits
+= blocks
* 2;
956 /* We may need to clone the existing xattr block in which case we need
957 * to increment ref counts for existing ea_inodes referenced by it.
960 struct ext4_xattr_entry
*entry
= BFIRST(block_bh
);
962 for (; !IS_LAST_ENTRY(entry
); entry
= EXT4_XATTR_NEXT(entry
))
963 if (entry
->e_value_inum
)
964 /* Ref count update on ea_inode. */
970 static int ext4_xattr_ensure_credits(handle_t
*handle
, struct inode
*inode
,
971 int credits
, struct buffer_head
*bh
,
972 bool dirty
, bool block_csum
)
976 if (!ext4_handle_valid(handle
))
979 if (handle
->h_buffer_credits
>= credits
)
982 error
= ext4_journal_extend(handle
, credits
- handle
->h_buffer_credits
);
986 ext4_warning(inode
->i_sb
, "Extend journal (error %d)", error
);
992 ext4_xattr_block_csum_set(inode
, bh
);
993 error
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
995 ext4_warning(inode
->i_sb
, "Handle metadata (error %d)",
1001 error
= ext4_journal_restart(handle
, credits
);
1003 ext4_warning(inode
->i_sb
, "Restart journal (error %d)", error
);
1008 error
= ext4_journal_get_write_access(handle
, bh
);
1010 ext4_warning(inode
->i_sb
,
1011 "Get write access failed (error %d)",
1019 static int ext4_xattr_inode_update_ref(handle_t
*handle
, struct inode
*ea_inode
,
1022 struct mb_cache
*ea_inode_cache
= EA_INODE_CACHE(ea_inode
);
1023 struct ext4_iloc iloc
;
1028 inode_lock(ea_inode
);
1030 ret
= ext4_reserve_inode_write(handle
, ea_inode
, &iloc
);
1036 ref_count
= ext4_xattr_inode_get_ref(ea_inode
);
1037 ref_count
+= ref_change
;
1038 ext4_xattr_inode_set_ref(ea_inode
, ref_count
);
1040 if (ref_change
> 0) {
1041 WARN_ONCE(ref_count
<= 0, "EA inode %lu ref_count=%lld",
1042 ea_inode
->i_ino
, ref_count
);
1044 if (ref_count
== 1) {
1045 WARN_ONCE(ea_inode
->i_nlink
, "EA inode %lu i_nlink=%u",
1046 ea_inode
->i_ino
, ea_inode
->i_nlink
);
1048 set_nlink(ea_inode
, 1);
1049 ext4_orphan_del(handle
, ea_inode
);
1051 if (ea_inode_cache
) {
1052 hash
= ext4_xattr_inode_get_hash(ea_inode
);
1053 mb_cache_entry_create(ea_inode_cache
,
1056 true /* reusable */);
1060 WARN_ONCE(ref_count
< 0, "EA inode %lu ref_count=%lld",
1061 ea_inode
->i_ino
, ref_count
);
1063 if (ref_count
== 0) {
1064 WARN_ONCE(ea_inode
->i_nlink
!= 1,
1065 "EA inode %lu i_nlink=%u",
1066 ea_inode
->i_ino
, ea_inode
->i_nlink
);
1068 clear_nlink(ea_inode
);
1069 ext4_orphan_add(handle
, ea_inode
);
1071 if (ea_inode_cache
) {
1072 hash
= ext4_xattr_inode_get_hash(ea_inode
);
1073 mb_cache_entry_delete(ea_inode_cache
, hash
,
1079 ret
= ext4_mark_iloc_dirty(handle
, ea_inode
, &iloc
);
1082 ext4_warning_inode(ea_inode
,
1083 "ext4_mark_iloc_dirty() failed ret=%d", ret
);
1086 inode_unlock(ea_inode
);
1090 static int ext4_xattr_inode_inc_ref(handle_t
*handle
, struct inode
*ea_inode
)
1092 return ext4_xattr_inode_update_ref(handle
, ea_inode
, 1);
1095 static int ext4_xattr_inode_dec_ref(handle_t
*handle
, struct inode
*ea_inode
)
1097 return ext4_xattr_inode_update_ref(handle
, ea_inode
, -1);
1100 static int ext4_xattr_inode_inc_ref_all(handle_t
*handle
, struct inode
*parent
,
1101 struct ext4_xattr_entry
*first
)
1103 struct inode
*ea_inode
;
1104 struct ext4_xattr_entry
*entry
;
1105 struct ext4_xattr_entry
*failed_entry
;
1106 unsigned int ea_ino
;
1109 for (entry
= first
; !IS_LAST_ENTRY(entry
);
1110 entry
= EXT4_XATTR_NEXT(entry
)) {
1111 if (!entry
->e_value_inum
)
1113 ea_ino
= le32_to_cpu(entry
->e_value_inum
);
1114 err
= ext4_xattr_inode_iget(parent
, ea_ino
,
1115 le32_to_cpu(entry
->e_hash
),
1119 err
= ext4_xattr_inode_inc_ref(handle
, ea_inode
);
1121 ext4_warning_inode(ea_inode
, "inc ref error %d", err
);
1131 failed_entry
= entry
;
1133 for (entry
= first
; entry
!= failed_entry
;
1134 entry
= EXT4_XATTR_NEXT(entry
)) {
1135 if (!entry
->e_value_inum
)
1137 ea_ino
= le32_to_cpu(entry
->e_value_inum
);
1138 err
= ext4_xattr_inode_iget(parent
, ea_ino
,
1139 le32_to_cpu(entry
->e_hash
),
1142 ext4_warning(parent
->i_sb
,
1143 "cleanup ea_ino %u iget error %d", ea_ino
,
1147 err
= ext4_xattr_inode_dec_ref(handle
, ea_inode
);
1149 ext4_warning_inode(ea_inode
, "cleanup dec ref error %d",
1157 ext4_xattr_inode_dec_ref_all(handle_t
*handle
, struct inode
*parent
,
1158 struct buffer_head
*bh
,
1159 struct ext4_xattr_entry
*first
, bool block_csum
,
1160 struct ext4_xattr_inode_array
**ea_inode_array
,
1161 int extra_credits
, bool skip_quota
)
1163 struct inode
*ea_inode
;
1164 struct ext4_xattr_entry
*entry
;
1166 unsigned int ea_ino
;
1170 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1171 credits
= 2 + extra_credits
;
1173 for (entry
= first
; !IS_LAST_ENTRY(entry
);
1174 entry
= EXT4_XATTR_NEXT(entry
)) {
1175 if (!entry
->e_value_inum
)
1177 ea_ino
= le32_to_cpu(entry
->e_value_inum
);
1178 err
= ext4_xattr_inode_iget(parent
, ea_ino
,
1179 le32_to_cpu(entry
->e_hash
),
1184 err
= ext4_expand_inode_array(ea_inode_array
, ea_inode
);
1186 ext4_warning_inode(ea_inode
,
1187 "Expand inode array err=%d", err
);
1192 err
= ext4_xattr_ensure_credits(handle
, parent
, credits
, bh
,
1195 ext4_warning_inode(ea_inode
, "Ensure credits err=%d",
1200 err
= ext4_xattr_inode_dec_ref(handle
, ea_inode
);
1202 ext4_warning_inode(ea_inode
, "ea_inode dec ref err=%d",
1208 ext4_xattr_inode_free_quota(parent
, ea_inode
,
1209 le32_to_cpu(entry
->e_value_size
));
1212 * Forget about ea_inode within the same transaction that
1213 * decrements the ref count. This avoids duplicate decrements in
1214 * case the rest of the work spills over to subsequent
1217 entry
->e_value_inum
= 0;
1218 entry
->e_value_size
= 0;
1225 * Note that we are deliberately skipping csum calculation for
1226 * the final update because we do not expect any journal
1227 * restarts until xattr block is freed.
1230 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1232 ext4_warning_inode(parent
,
1233 "handle dirty metadata err=%d", err
);
1238 * Release the xattr block BH: If the reference count is > 1, decrement it;
1239 * otherwise free the block.
1242 ext4_xattr_release_block(handle_t
*handle
, struct inode
*inode
,
1243 struct buffer_head
*bh
,
1244 struct ext4_xattr_inode_array
**ea_inode_array
,
1247 struct mb_cache
*ea_block_cache
= EA_BLOCK_CACHE(inode
);
1251 BUFFER_TRACE(bh
, "get_write_access");
1252 error
= ext4_journal_get_write_access(handle
, bh
);
1257 hash
= le32_to_cpu(BHDR(bh
)->h_hash
);
1258 ref
= le32_to_cpu(BHDR(bh
)->h_refcount
);
1260 ea_bdebug(bh
, "refcount now=0; freeing");
1262 * This must happen under buffer lock for
1263 * ext4_xattr_block_set() to reliably detect freed block
1266 mb_cache_entry_delete(ea_block_cache
, hash
,
1271 if (ext4_has_feature_ea_inode(inode
->i_sb
))
1272 ext4_xattr_inode_dec_ref_all(handle
, inode
, bh
,
1274 true /* block_csum */,
1277 true /* skip_quota */);
1278 ext4_free_blocks(handle
, inode
, bh
, 0, 1,
1279 EXT4_FREE_BLOCKS_METADATA
|
1280 EXT4_FREE_BLOCKS_FORGET
);
1283 BHDR(bh
)->h_refcount
= cpu_to_le32(ref
);
1284 if (ref
== EXT4_XATTR_REFCOUNT_MAX
- 1) {
1285 struct mb_cache_entry
*ce
;
1287 if (ea_block_cache
) {
1288 ce
= mb_cache_entry_get(ea_block_cache
, hash
,
1292 mb_cache_entry_put(ea_block_cache
, ce
);
1297 ext4_xattr_block_csum_set(inode
, bh
);
1299 * Beware of this ugliness: Releasing of xattr block references
1300 * from different inodes can race and so we have to protect
1301 * from a race where someone else frees the block (and releases
1302 * its journal_head) before we are done dirtying the buffer. In
1303 * nojournal mode this race is harmless and we actually cannot
1304 * call ext4_handle_dirty_metadata() with locked buffer as
1305 * that function can call sync_dirty_buffer() so for that case
1306 * we handle the dirtying after unlocking the buffer.
1308 if (ext4_handle_valid(handle
))
1309 error
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1311 if (!ext4_handle_valid(handle
))
1312 error
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1314 ext4_handle_sync(handle
);
1315 dquot_free_block(inode
, EXT4_C2B(EXT4_SB(inode
->i_sb
), 1));
1316 ea_bdebug(bh
, "refcount now=%d; releasing",
1317 le32_to_cpu(BHDR(bh
)->h_refcount
));
1320 ext4_std_error(inode
->i_sb
, error
);
1325 * Find the available free space for EAs. This also returns the total number of
1326 * bytes used by EA entries.
1328 static size_t ext4_xattr_free_space(struct ext4_xattr_entry
*last
,
1329 size_t *min_offs
, void *base
, int *total
)
1331 for (; !IS_LAST_ENTRY(last
); last
= EXT4_XATTR_NEXT(last
)) {
1332 if (!last
->e_value_inum
&& last
->e_value_size
) {
1333 size_t offs
= le16_to_cpu(last
->e_value_offs
);
1334 if (offs
< *min_offs
)
1338 *total
+= EXT4_XATTR_LEN(last
->e_name_len
);
1340 return (*min_offs
- ((void *)last
- base
) - sizeof(__u32
));
1344 * Write the value of the EA in an inode.
1346 static int ext4_xattr_inode_write(handle_t
*handle
, struct inode
*ea_inode
,
1347 const void *buf
, int bufsize
)
1349 struct buffer_head
*bh
= NULL
;
1350 unsigned long block
= 0;
1351 int blocksize
= ea_inode
->i_sb
->s_blocksize
;
1352 int max_blocks
= (bufsize
+ blocksize
- 1) >> ea_inode
->i_blkbits
;
1353 int csize
, wsize
= 0;
1358 while (ret
>= 0 && ret
< max_blocks
) {
1359 struct ext4_map_blocks map
;
1360 map
.m_lblk
= block
+= ret
;
1361 map
.m_len
= max_blocks
-= ret
;
1363 ret
= ext4_map_blocks(handle
, ea_inode
, &map
,
1364 EXT4_GET_BLOCKS_CREATE
);
1366 ext4_mark_inode_dirty(handle
, ea_inode
);
1367 if (ret
== -ENOSPC
&&
1368 ext4_should_retry_alloc(ea_inode
->i_sb
, &retries
)) {
1380 while (wsize
< bufsize
) {
1383 csize
= (bufsize
- wsize
) > blocksize
? blocksize
:
1385 bh
= ext4_getblk(handle
, ea_inode
, block
, 0);
1388 ret
= ext4_journal_get_write_access(handle
, bh
);
1392 memcpy(bh
->b_data
, buf
, csize
);
1393 set_buffer_uptodate(bh
);
1394 ext4_handle_dirty_metadata(handle
, ea_inode
, bh
);
1401 inode_lock(ea_inode
);
1402 i_size_write(ea_inode
, wsize
);
1403 ext4_update_i_disksize(ea_inode
, wsize
);
1404 inode_unlock(ea_inode
);
1406 ext4_mark_inode_dirty(handle
, ea_inode
);
1415 * Create an inode to store the value of a large EA.
1417 static struct inode
*ext4_xattr_inode_create(handle_t
*handle
,
1418 struct inode
*inode
, u32 hash
)
1420 struct inode
*ea_inode
= NULL
;
1421 uid_t owner
[2] = { i_uid_read(inode
), i_gid_read(inode
) };
1425 * Let the next inode be the goal, so we try and allocate the EA inode
1426 * in the same group, or nearby one.
1428 ea_inode
= ext4_new_inode(handle
, inode
->i_sb
->s_root
->d_inode
,
1429 S_IFREG
| 0600, NULL
, inode
->i_ino
+ 1, owner
,
1431 if (!IS_ERR(ea_inode
)) {
1432 ea_inode
->i_op
= &ext4_file_inode_operations
;
1433 ea_inode
->i_fop
= &ext4_file_operations
;
1434 ext4_set_aops(ea_inode
);
1435 ext4_xattr_inode_set_class(ea_inode
);
1436 unlock_new_inode(ea_inode
);
1437 ext4_xattr_inode_set_ref(ea_inode
, 1);
1438 ext4_xattr_inode_set_hash(ea_inode
, hash
);
1439 err
= ext4_mark_inode_dirty(handle
, ea_inode
);
1441 err
= ext4_inode_attach_jinode(ea_inode
);
1444 return ERR_PTR(err
);
1448 * Xattr inodes are shared therefore quota charging is performed
1449 * at a higher level.
1451 dquot_free_inode(ea_inode
);
1452 dquot_drop(ea_inode
);
1453 inode_lock(ea_inode
);
1454 ea_inode
->i_flags
|= S_NOQUOTA
;
1455 inode_unlock(ea_inode
);
1461 static struct inode
*
1462 ext4_xattr_inode_cache_find(struct inode
*inode
, const void *value
,
1463 size_t value_len
, u32 hash
)
1465 struct inode
*ea_inode
;
1466 struct mb_cache_entry
*ce
;
1467 struct mb_cache
*ea_inode_cache
= EA_INODE_CACHE(inode
);
1470 if (!ea_inode_cache
)
1473 ce
= mb_cache_entry_find_first(ea_inode_cache
, hash
);
1477 ea_data
= ext4_kvmalloc(value_len
, GFP_NOFS
);
1479 mb_cache_entry_put(ea_inode_cache
, ce
);
1484 ea_inode
= ext4_iget(inode
->i_sb
, ce
->e_value
);
1485 if (!IS_ERR(ea_inode
) &&
1486 !is_bad_inode(ea_inode
) &&
1487 (EXT4_I(ea_inode
)->i_flags
& EXT4_EA_INODE_FL
) &&
1488 i_size_read(ea_inode
) == value_len
&&
1489 !ext4_xattr_inode_read(ea_inode
, ea_data
, value_len
) &&
1490 !ext4_xattr_inode_verify_hashes(ea_inode
, NULL
, ea_data
,
1492 !memcmp(value
, ea_data
, value_len
)) {
1493 mb_cache_entry_touch(ea_inode_cache
, ce
);
1494 mb_cache_entry_put(ea_inode_cache
, ce
);
1499 if (!IS_ERR(ea_inode
))
1501 ce
= mb_cache_entry_find_next(ea_inode_cache
, ce
);
1508 * Add value of the EA in an inode.
1510 static int ext4_xattr_inode_lookup_create(handle_t
*handle
, struct inode
*inode
,
1511 const void *value
, size_t value_len
,
1512 struct inode
**ret_inode
)
1514 struct inode
*ea_inode
;
1518 hash
= ext4_xattr_inode_hash(EXT4_SB(inode
->i_sb
), value
, value_len
);
1519 ea_inode
= ext4_xattr_inode_cache_find(inode
, value
, value_len
, hash
);
1521 err
= ext4_xattr_inode_inc_ref(handle
, ea_inode
);
1527 *ret_inode
= ea_inode
;
1531 /* Create an inode for the EA value */
1532 ea_inode
= ext4_xattr_inode_create(handle
, inode
, hash
);
1533 if (IS_ERR(ea_inode
))
1534 return PTR_ERR(ea_inode
);
1536 err
= ext4_xattr_inode_write(handle
, ea_inode
, value
, value_len
);
1538 ext4_xattr_inode_dec_ref(handle
, ea_inode
);
1543 if (EA_INODE_CACHE(inode
))
1544 mb_cache_entry_create(EA_INODE_CACHE(inode
), GFP_NOFS
, hash
,
1545 ea_inode
->i_ino
, true /* reusable */);
1547 *ret_inode
= ea_inode
;
1552 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1553 * feature is enabled.
1555 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1557 static int ext4_xattr_set_entry(struct ext4_xattr_info
*i
,
1558 struct ext4_xattr_search
*s
,
1559 handle_t
*handle
, struct inode
*inode
,
1562 struct ext4_xattr_entry
*last
;
1563 struct ext4_xattr_entry
*here
= s
->here
;
1564 size_t min_offs
= s
->end
- s
->base
, name_len
= strlen(i
->name
);
1565 int in_inode
= i
->in_inode
;
1566 struct inode
*old_ea_inode
= NULL
;
1567 struct inode
*new_ea_inode
= NULL
;
1568 size_t old_size
, new_size
;
1571 /* Space used by old and new values. */
1572 old_size
= (!s
->not_found
&& !here
->e_value_inum
) ?
1573 EXT4_XATTR_SIZE(le32_to_cpu(here
->e_value_size
)) : 0;
1574 new_size
= (i
->value
&& !in_inode
) ? EXT4_XATTR_SIZE(i
->value_len
) : 0;
1577 * Optimization for the simple case when old and new values have the
1578 * same padded sizes. Not applicable if external inodes are involved.
1580 if (new_size
&& new_size
== old_size
) {
1581 size_t offs
= le16_to_cpu(here
->e_value_offs
);
1582 void *val
= s
->base
+ offs
;
1584 here
->e_value_size
= cpu_to_le32(i
->value_len
);
1585 if (i
->value
== EXT4_ZERO_XATTR_VALUE
) {
1586 memset(val
, 0, new_size
);
1588 memcpy(val
, i
->value
, i
->value_len
);
1589 /* Clear padding bytes. */
1590 memset(val
+ i
->value_len
, 0, new_size
- i
->value_len
);
1595 /* Compute min_offs and last. */
1597 for (; !IS_LAST_ENTRY(last
); last
= EXT4_XATTR_NEXT(last
)) {
1598 if (!last
->e_value_inum
&& last
->e_value_size
) {
1599 size_t offs
= le16_to_cpu(last
->e_value_offs
);
1600 if (offs
< min_offs
)
1605 /* Check whether we have enough space. */
1609 free
= min_offs
- ((void *)last
- s
->base
) - sizeof(__u32
);
1611 free
+= EXT4_XATTR_LEN(name_len
) + old_size
;
1613 if (free
< EXT4_XATTR_LEN(name_len
) + new_size
) {
1619 * If storing the value in an external inode is an option,
1620 * reserve space for xattr entries/names in the external
1621 * attribute block so that a long value does not occupy the
1622 * whole space and prevent futher entries being added.
1624 if (ext4_has_feature_ea_inode(inode
->i_sb
) &&
1625 new_size
&& is_block
&&
1626 (min_offs
+ old_size
- new_size
) <
1627 EXT4_XATTR_BLOCK_RESERVE(inode
)) {
1634 * Getting access to old and new ea inodes is subject to failures.
1635 * Finish that work before doing any modifications to the xattr data.
1637 if (!s
->not_found
&& here
->e_value_inum
) {
1638 ret
= ext4_xattr_inode_iget(inode
,
1639 le32_to_cpu(here
->e_value_inum
),
1640 le32_to_cpu(here
->e_hash
),
1643 old_ea_inode
= NULL
;
1647 if (i
->value
&& in_inode
) {
1648 WARN_ON_ONCE(!i
->value_len
);
1650 ret
= ext4_xattr_inode_alloc_quota(inode
, i
->value_len
);
1654 ret
= ext4_xattr_inode_lookup_create(handle
, inode
, i
->value
,
1658 new_ea_inode
= NULL
;
1659 ext4_xattr_inode_free_quota(inode
, NULL
, i
->value_len
);
1665 /* We are ready to release ref count on the old_ea_inode. */
1666 ret
= ext4_xattr_inode_dec_ref(handle
, old_ea_inode
);
1668 /* Release newly required ref count on new_ea_inode. */
1672 err
= ext4_xattr_inode_dec_ref(handle
,
1675 ext4_warning_inode(new_ea_inode
,
1676 "dec ref new_ea_inode err=%d",
1678 ext4_xattr_inode_free_quota(inode
, new_ea_inode
,
1684 ext4_xattr_inode_free_quota(inode
, old_ea_inode
,
1685 le32_to_cpu(here
->e_value_size
));
1688 /* No failures allowed past this point. */
1690 if (!s
->not_found
&& here
->e_value_size
&& here
->e_value_offs
) {
1691 /* Remove the old value. */
1692 void *first_val
= s
->base
+ min_offs
;
1693 size_t offs
= le16_to_cpu(here
->e_value_offs
);
1694 void *val
= s
->base
+ offs
;
1696 memmove(first_val
+ old_size
, first_val
, val
- first_val
);
1697 memset(first_val
, 0, old_size
);
1698 min_offs
+= old_size
;
1700 /* Adjust all value offsets. */
1702 while (!IS_LAST_ENTRY(last
)) {
1703 size_t o
= le16_to_cpu(last
->e_value_offs
);
1705 if (!last
->e_value_inum
&&
1706 last
->e_value_size
&& o
< offs
)
1707 last
->e_value_offs
= cpu_to_le16(o
+ old_size
);
1708 last
= EXT4_XATTR_NEXT(last
);
1713 /* Remove old name. */
1714 size_t size
= EXT4_XATTR_LEN(name_len
);
1716 last
= ENTRY((void *)last
- size
);
1717 memmove(here
, (void *)here
+ size
,
1718 (void *)last
- (void *)here
+ sizeof(__u32
));
1719 memset(last
, 0, size
);
1720 } else if (s
->not_found
) {
1721 /* Insert new name. */
1722 size_t size
= EXT4_XATTR_LEN(name_len
);
1723 size_t rest
= (void *)last
- (void *)here
+ sizeof(__u32
);
1725 memmove((void *)here
+ size
, here
, rest
);
1726 memset(here
, 0, size
);
1727 here
->e_name_index
= i
->name_index
;
1728 here
->e_name_len
= name_len
;
1729 memcpy(here
->e_name
, i
->name
, name_len
);
1731 /* This is an update, reset value info. */
1732 here
->e_value_inum
= 0;
1733 here
->e_value_offs
= 0;
1734 here
->e_value_size
= 0;
1738 /* Insert new value. */
1740 here
->e_value_inum
= cpu_to_le32(new_ea_inode
->i_ino
);
1741 } else if (i
->value_len
) {
1742 void *val
= s
->base
+ min_offs
- new_size
;
1744 here
->e_value_offs
= cpu_to_le16(min_offs
- new_size
);
1745 if (i
->value
== EXT4_ZERO_XATTR_VALUE
) {
1746 memset(val
, 0, new_size
);
1748 memcpy(val
, i
->value
, i
->value_len
);
1749 /* Clear padding bytes. */
1750 memset(val
+ i
->value_len
, 0,
1751 new_size
- i
->value_len
);
1754 here
->e_value_size
= cpu_to_le32(i
->value_len
);
1761 /* Entry hash calculation. */
1766 * Feed crc32c hash instead of the raw value for entry
1767 * hash calculation. This is to avoid walking
1768 * potentially long value buffer again.
1770 crc32c_hash
= cpu_to_le32(
1771 ext4_xattr_inode_get_hash(new_ea_inode
));
1772 hash
= ext4_xattr_hash_entry(here
->e_name
,
1775 } else if (is_block
) {
1776 __le32
*value
= s
->base
+ le16_to_cpu(
1777 here
->e_value_offs
);
1779 hash
= ext4_xattr_hash_entry(here
->e_name
,
1780 here
->e_name_len
, value
,
1783 here
->e_hash
= hash
;
1787 ext4_xattr_rehash((struct ext4_xattr_header
*)s
->base
);
1796 struct ext4_xattr_block_find
{
1797 struct ext4_xattr_search s
;
1798 struct buffer_head
*bh
;
1802 ext4_xattr_block_find(struct inode
*inode
, struct ext4_xattr_info
*i
,
1803 struct ext4_xattr_block_find
*bs
)
1805 struct super_block
*sb
= inode
->i_sb
;
1808 ea_idebug(inode
, "name=%d.%s, value=%p, value_len=%ld",
1809 i
->name_index
, i
->name
, i
->value
, (long)i
->value_len
);
1811 if (EXT4_I(inode
)->i_file_acl
) {
1812 /* The inode already has an extended attribute block. */
1813 bs
->bh
= sb_bread(sb
, EXT4_I(inode
)->i_file_acl
);
1817 ea_bdebug(bs
->bh
, "b_count=%d, refcount=%d",
1818 atomic_read(&(bs
->bh
->b_count
)),
1819 le32_to_cpu(BHDR(bs
->bh
)->h_refcount
));
1820 error
= ext4_xattr_check_block(inode
, bs
->bh
);
1823 /* Find the named attribute. */
1824 bs
->s
.base
= BHDR(bs
->bh
);
1825 bs
->s
.first
= BFIRST(bs
->bh
);
1826 bs
->s
.end
= bs
->bh
->b_data
+ bs
->bh
->b_size
;
1827 bs
->s
.here
= bs
->s
.first
;
1828 error
= xattr_find_entry(inode
, &bs
->s
.here
, bs
->s
.end
,
1829 i
->name_index
, i
->name
, 1);
1830 if (error
&& error
!= -ENODATA
)
1832 bs
->s
.not_found
= error
;
1841 ext4_xattr_block_set(handle_t
*handle
, struct inode
*inode
,
1842 struct ext4_xattr_info
*i
,
1843 struct ext4_xattr_block_find
*bs
)
1845 struct super_block
*sb
= inode
->i_sb
;
1846 struct buffer_head
*new_bh
= NULL
;
1847 struct ext4_xattr_search s_copy
= bs
->s
;
1848 struct ext4_xattr_search
*s
= &s_copy
;
1849 struct mb_cache_entry
*ce
= NULL
;
1851 struct mb_cache
*ea_block_cache
= EA_BLOCK_CACHE(inode
);
1852 struct inode
*ea_inode
= NULL
, *tmp_inode
;
1853 size_t old_ea_inode_quota
= 0;
1854 unsigned int ea_ino
;
1857 #define header(x) ((struct ext4_xattr_header *)(x))
1860 BUFFER_TRACE(bs
->bh
, "get_write_access");
1861 error
= ext4_journal_get_write_access(handle
, bs
->bh
);
1864 lock_buffer(bs
->bh
);
1866 if (header(s
->base
)->h_refcount
== cpu_to_le32(1)) {
1867 __u32 hash
= le32_to_cpu(BHDR(bs
->bh
)->h_hash
);
1870 * This must happen under buffer lock for
1871 * ext4_xattr_block_set() to reliably detect modified
1875 mb_cache_entry_delete(ea_block_cache
, hash
,
1877 ea_bdebug(bs
->bh
, "modifying in-place");
1878 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
,
1879 true /* is_block */);
1880 ext4_xattr_block_csum_set(inode
, bs
->bh
);
1881 unlock_buffer(bs
->bh
);
1882 if (error
== -EFSCORRUPTED
)
1885 error
= ext4_handle_dirty_metadata(handle
,
1892 int offset
= (char *)s
->here
- bs
->bh
->b_data
;
1894 unlock_buffer(bs
->bh
);
1895 ea_bdebug(bs
->bh
, "cloning");
1896 s
->base
= kmalloc(bs
->bh
->b_size
, GFP_NOFS
);
1898 if (s
->base
== NULL
)
1900 memcpy(s
->base
, BHDR(bs
->bh
), bs
->bh
->b_size
);
1901 s
->first
= ENTRY(header(s
->base
)+1);
1902 header(s
->base
)->h_refcount
= cpu_to_le32(1);
1903 s
->here
= ENTRY(s
->base
+ offset
);
1904 s
->end
= s
->base
+ bs
->bh
->b_size
;
1907 * If existing entry points to an xattr inode, we need
1908 * to prevent ext4_xattr_set_entry() from decrementing
1909 * ref count on it because the reference belongs to the
1910 * original block. In this case, make the entry look
1911 * like it has an empty value.
1913 if (!s
->not_found
&& s
->here
->e_value_inum
) {
1914 ea_ino
= le32_to_cpu(s
->here
->e_value_inum
);
1915 error
= ext4_xattr_inode_iget(inode
, ea_ino
,
1916 le32_to_cpu(s
->here
->e_hash
),
1921 if (!ext4_test_inode_state(tmp_inode
,
1922 EXT4_STATE_LUSTRE_EA_INODE
)) {
1924 * Defer quota free call for previous
1925 * inode until success is guaranteed.
1927 old_ea_inode_quota
= le32_to_cpu(
1928 s
->here
->e_value_size
);
1932 s
->here
->e_value_inum
= 0;
1933 s
->here
->e_value_size
= 0;
1937 /* Allocate a buffer where we construct the new block. */
1938 s
->base
= kzalloc(sb
->s_blocksize
, GFP_NOFS
);
1939 /* assert(header == s->base) */
1941 if (s
->base
== NULL
)
1943 header(s
->base
)->h_magic
= cpu_to_le32(EXT4_XATTR_MAGIC
);
1944 header(s
->base
)->h_blocks
= cpu_to_le32(1);
1945 header(s
->base
)->h_refcount
= cpu_to_le32(1);
1946 s
->first
= ENTRY(header(s
->base
)+1);
1947 s
->here
= ENTRY(header(s
->base
)+1);
1948 s
->end
= s
->base
+ sb
->s_blocksize
;
1951 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
, true /* is_block */);
1952 if (error
== -EFSCORRUPTED
)
1957 if (i
->value
&& s
->here
->e_value_inum
) {
1959 * A ref count on ea_inode has been taken as part of the call to
1960 * ext4_xattr_set_entry() above. We would like to drop this
1961 * extra ref but we have to wait until the xattr block is
1962 * initialized and has its own ref count on the ea_inode.
1964 ea_ino
= le32_to_cpu(s
->here
->e_value_inum
);
1965 error
= ext4_xattr_inode_iget(inode
, ea_ino
,
1966 le32_to_cpu(s
->here
->e_hash
),
1975 if (!IS_LAST_ENTRY(s
->first
)) {
1976 new_bh
= ext4_xattr_block_cache_find(inode
, header(s
->base
),
1979 /* We found an identical block in the cache. */
1980 if (new_bh
== bs
->bh
)
1981 ea_bdebug(new_bh
, "keeping");
1985 WARN_ON_ONCE(dquot_initialize_needed(inode
));
1987 /* The old block is released after updating
1989 error
= dquot_alloc_block(inode
,
1990 EXT4_C2B(EXT4_SB(sb
), 1));
1993 BUFFER_TRACE(new_bh
, "get_write_access");
1994 error
= ext4_journal_get_write_access(handle
,
1998 lock_buffer(new_bh
);
2000 * We have to be careful about races with
2001 * freeing, rehashing or adding references to
2002 * xattr block. Once we hold buffer lock xattr
2003 * block's state is stable so we can check
2004 * whether the block got freed / rehashed or
2005 * not. Since we unhash mbcache entry under
2006 * buffer lock when freeing / rehashing xattr
2007 * block, checking whether entry is still
2008 * hashed is reliable. Same rules hold for
2009 * e_reusable handling.
2011 if (hlist_bl_unhashed(&ce
->e_hash_list
) ||
2014 * Undo everything and check mbcache
2017 unlock_buffer(new_bh
);
2018 dquot_free_block(inode
,
2019 EXT4_C2B(EXT4_SB(sb
),
2022 mb_cache_entry_put(ea_block_cache
, ce
);
2027 ref
= le32_to_cpu(BHDR(new_bh
)->h_refcount
) + 1;
2028 BHDR(new_bh
)->h_refcount
= cpu_to_le32(ref
);
2029 if (ref
>= EXT4_XATTR_REFCOUNT_MAX
)
2031 ea_bdebug(new_bh
, "reusing; refcount now=%d",
2033 ext4_xattr_block_csum_set(inode
, new_bh
);
2034 unlock_buffer(new_bh
);
2035 error
= ext4_handle_dirty_metadata(handle
,
2041 mb_cache_entry_touch(ea_block_cache
, ce
);
2042 mb_cache_entry_put(ea_block_cache
, ce
);
2044 } else if (bs
->bh
&& s
->base
== bs
->bh
->b_data
) {
2045 /* We were modifying this block in-place. */
2046 ea_bdebug(bs
->bh
, "keeping this block");
2047 ext4_xattr_block_cache_insert(ea_block_cache
, bs
->bh
);
2051 /* We need to allocate a new block */
2052 ext4_fsblk_t goal
, block
;
2054 WARN_ON_ONCE(dquot_initialize_needed(inode
));
2056 goal
= ext4_group_first_block_no(sb
,
2057 EXT4_I(inode
)->i_block_group
);
2059 /* non-extent files can't have physical blocks past 2^32 */
2060 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
2061 goal
= goal
& EXT4_MAX_BLOCK_FILE_PHYS
;
2063 block
= ext4_new_meta_blocks(handle
, inode
, goal
, 0,
2068 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
2069 BUG_ON(block
> EXT4_MAX_BLOCK_FILE_PHYS
);
2071 ea_idebug(inode
, "creating block %llu",
2072 (unsigned long long)block
);
2074 new_bh
= sb_getblk(sb
, block
);
2075 if (unlikely(!new_bh
)) {
2078 ext4_free_blocks(handle
, inode
, NULL
, block
, 1,
2079 EXT4_FREE_BLOCKS_METADATA
);
2082 error
= ext4_xattr_inode_inc_ref_all(handle
, inode
,
2083 ENTRY(header(s
->base
)+1));
2087 /* Drop the extra ref on ea_inode. */
2088 error
= ext4_xattr_inode_dec_ref(handle
,
2091 ext4_warning_inode(ea_inode
,
2098 lock_buffer(new_bh
);
2099 error
= ext4_journal_get_create_access(handle
, new_bh
);
2101 unlock_buffer(new_bh
);
2105 memcpy(new_bh
->b_data
, s
->base
, new_bh
->b_size
);
2106 ext4_xattr_block_csum_set(inode
, new_bh
);
2107 set_buffer_uptodate(new_bh
);
2108 unlock_buffer(new_bh
);
2109 ext4_xattr_block_cache_insert(ea_block_cache
, new_bh
);
2110 error
= ext4_handle_dirty_metadata(handle
, inode
,
2117 if (old_ea_inode_quota
)
2118 ext4_xattr_inode_free_quota(inode
, NULL
, old_ea_inode_quota
);
2120 /* Update the inode. */
2121 EXT4_I(inode
)->i_file_acl
= new_bh
? new_bh
->b_blocknr
: 0;
2123 /* Drop the previous xattr block. */
2124 if (bs
->bh
&& bs
->bh
!= new_bh
) {
2125 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
2127 ext4_xattr_release_block(handle
, inode
, bs
->bh
,
2129 0 /* extra_credits */);
2130 ext4_xattr_inode_array_free(ea_inode_array
);
2138 error2
= ext4_xattr_inode_dec_ref(handle
, ea_inode
);
2140 ext4_warning_inode(ea_inode
, "dec ref error=%d",
2143 /* If there was an error, revert the quota charge. */
2145 ext4_xattr_inode_free_quota(inode
, ea_inode
,
2146 i_size_read(ea_inode
));
2150 mb_cache_entry_put(ea_block_cache
, ce
);
2152 if (!(bs
->bh
&& s
->base
== bs
->bh
->b_data
))
2158 dquot_free_block(inode
, EXT4_C2B(EXT4_SB(sb
), 1));
2162 EXT4_ERROR_INODE(inode
, "bad block %llu",
2163 EXT4_I(inode
)->i_file_acl
);
2169 int ext4_xattr_ibody_find(struct inode
*inode
, struct ext4_xattr_info
*i
,
2170 struct ext4_xattr_ibody_find
*is
)
2172 struct ext4_xattr_ibody_header
*header
;
2173 struct ext4_inode
*raw_inode
;
2176 if (EXT4_I(inode
)->i_extra_isize
== 0)
2178 raw_inode
= ext4_raw_inode(&is
->iloc
);
2179 header
= IHDR(inode
, raw_inode
);
2180 is
->s
.base
= is
->s
.first
= IFIRST(header
);
2181 is
->s
.here
= is
->s
.first
;
2182 is
->s
.end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
2183 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
2184 error
= xattr_check_inode(inode
, header
, is
->s
.end
);
2187 /* Find the named attribute. */
2188 error
= xattr_find_entry(inode
, &is
->s
.here
, is
->s
.end
,
2189 i
->name_index
, i
->name
, 0);
2190 if (error
&& error
!= -ENODATA
)
2192 is
->s
.not_found
= error
;
2197 int ext4_xattr_ibody_inline_set(handle_t
*handle
, struct inode
*inode
,
2198 struct ext4_xattr_info
*i
,
2199 struct ext4_xattr_ibody_find
*is
)
2201 struct ext4_xattr_ibody_header
*header
;
2202 struct ext4_xattr_search
*s
= &is
->s
;
2205 if (EXT4_I(inode
)->i_extra_isize
== 0)
2207 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
, false /* is_block */);
2209 if (error
== -ENOSPC
&&
2210 ext4_has_inline_data(inode
)) {
2211 error
= ext4_try_to_evict_inline_data(handle
, inode
,
2212 EXT4_XATTR_LEN(strlen(i
->name
) +
2213 EXT4_XATTR_SIZE(i
->value_len
)));
2216 error
= ext4_xattr_ibody_find(inode
, i
, is
);
2219 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
,
2220 false /* is_block */);
2225 header
= IHDR(inode
, ext4_raw_inode(&is
->iloc
));
2226 if (!IS_LAST_ENTRY(s
->first
)) {
2227 header
->h_magic
= cpu_to_le32(EXT4_XATTR_MAGIC
);
2228 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
2230 header
->h_magic
= cpu_to_le32(0);
2231 ext4_clear_inode_state(inode
, EXT4_STATE_XATTR
);
2236 static int ext4_xattr_ibody_set(handle_t
*handle
, struct inode
*inode
,
2237 struct ext4_xattr_info
*i
,
2238 struct ext4_xattr_ibody_find
*is
)
2240 struct ext4_xattr_ibody_header
*header
;
2241 struct ext4_xattr_search
*s
= &is
->s
;
2244 if (EXT4_I(inode
)->i_extra_isize
== 0)
2246 error
= ext4_xattr_set_entry(i
, s
, handle
, inode
, false /* is_block */);
2249 header
= IHDR(inode
, ext4_raw_inode(&is
->iloc
));
2250 if (!IS_LAST_ENTRY(s
->first
)) {
2251 header
->h_magic
= cpu_to_le32(EXT4_XATTR_MAGIC
);
2252 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
2254 header
->h_magic
= cpu_to_le32(0);
2255 ext4_clear_inode_state(inode
, EXT4_STATE_XATTR
);
2260 static int ext4_xattr_value_same(struct ext4_xattr_search
*s
,
2261 struct ext4_xattr_info
*i
)
2265 /* When e_value_inum is set the value is stored externally. */
2266 if (s
->here
->e_value_inum
)
2268 if (le32_to_cpu(s
->here
->e_value_size
) != i
->value_len
)
2270 value
= ((void *)s
->base
) + le16_to_cpu(s
->here
->e_value_offs
);
2271 return !memcmp(value
, i
->value
, i
->value_len
);
2274 static struct buffer_head
*ext4_xattr_get_block(struct inode
*inode
)
2276 struct buffer_head
*bh
;
2279 if (!EXT4_I(inode
)->i_file_acl
)
2281 bh
= sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
);
2283 return ERR_PTR(-EIO
);
2284 error
= ext4_xattr_check_block(inode
, bh
);
2286 return ERR_PTR(error
);
2291 * ext4_xattr_set_handle()
2293 * Create, replace or remove an extended attribute for this inode. Value
2294 * is NULL to remove an existing extended attribute, and non-NULL to
2295 * either replace an existing extended attribute, or create a new extended
2296 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2297 * specify that an extended attribute must exist and must not exist
2298 * previous to the call, respectively.
2300 * Returns 0, or a negative error number on failure.
2303 ext4_xattr_set_handle(handle_t
*handle
, struct inode
*inode
, int name_index
,
2304 const char *name
, const void *value
, size_t value_len
,
2307 struct ext4_xattr_info i
= {
2308 .name_index
= name_index
,
2311 .value_len
= value_len
,
2314 struct ext4_xattr_ibody_find is
= {
2315 .s
= { .not_found
= -ENODATA
, },
2317 struct ext4_xattr_block_find bs
= {
2318 .s
= { .not_found
= -ENODATA
, },
2325 if (strlen(name
) > 255)
2328 ext4_write_lock_xattr(inode
, &no_expand
);
2330 /* Check journal credits under write lock. */
2331 if (ext4_handle_valid(handle
)) {
2332 struct buffer_head
*bh
;
2335 bh
= ext4_xattr_get_block(inode
);
2337 error
= PTR_ERR(bh
);
2341 credits
= __ext4_xattr_set_credits(inode
->i_sb
, inode
, bh
,
2343 flags
& XATTR_CREATE
);
2346 if (!ext4_handle_has_enough_credits(handle
, credits
)) {
2352 error
= ext4_reserve_inode_write(handle
, inode
, &is
.iloc
);
2356 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
)) {
2357 struct ext4_inode
*raw_inode
= ext4_raw_inode(&is
.iloc
);
2358 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
2359 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
2362 error
= ext4_xattr_ibody_find(inode
, &i
, &is
);
2366 error
= ext4_xattr_block_find(inode
, &i
, &bs
);
2369 if (is
.s
.not_found
&& bs
.s
.not_found
) {
2371 if (flags
& XATTR_REPLACE
)
2378 if (flags
& XATTR_CREATE
)
2383 if (!is
.s
.not_found
)
2384 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, &is
);
2385 else if (!bs
.s
.not_found
)
2386 error
= ext4_xattr_block_set(handle
, inode
, &i
, &bs
);
2389 /* Xattr value did not change? Save us some work and bail out */
2390 if (!is
.s
.not_found
&& ext4_xattr_value_same(&is
.s
, &i
))
2392 if (!bs
.s
.not_found
&& ext4_xattr_value_same(&bs
.s
, &i
))
2395 if (ext4_has_feature_ea_inode(inode
->i_sb
) &&
2396 (EXT4_XATTR_SIZE(i
.value_len
) >
2397 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode
->i_sb
->s_blocksize
)))
2400 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, &is
);
2401 if (!error
&& !bs
.s
.not_found
) {
2403 error
= ext4_xattr_block_set(handle
, inode
, &i
, &bs
);
2404 } else if (error
== -ENOSPC
) {
2405 if (EXT4_I(inode
)->i_file_acl
&& !bs
.s
.base
) {
2406 error
= ext4_xattr_block_find(inode
, &i
, &bs
);
2410 error
= ext4_xattr_block_set(handle
, inode
, &i
, &bs
);
2411 if (!error
&& !is
.s
.not_found
) {
2413 error
= ext4_xattr_ibody_set(handle
, inode
, &i
,
2415 } else if (error
== -ENOSPC
) {
2417 * Xattr does not fit in the block, store at
2418 * external inode if possible.
2420 if (ext4_has_feature_ea_inode(inode
->i_sb
) &&
2429 ext4_xattr_update_super_block(handle
, inode
->i_sb
);
2430 inode
->i_ctime
= current_time(inode
);
2433 error
= ext4_mark_iloc_dirty(handle
, inode
, &is
.iloc
);
2435 * The bh is consumed by ext4_mark_iloc_dirty, even with
2440 ext4_handle_sync(handle
);
2446 ext4_write_unlock_xattr(inode
, &no_expand
);
2450 int ext4_xattr_set_credits(struct inode
*inode
, size_t value_len
,
2451 bool is_create
, int *credits
)
2453 struct buffer_head
*bh
;
2458 if (!EXT4_SB(inode
->i_sb
)->s_journal
)
2461 down_read(&EXT4_I(inode
)->xattr_sem
);
2463 bh
= ext4_xattr_get_block(inode
);
2467 *credits
= __ext4_xattr_set_credits(inode
->i_sb
, inode
, bh
,
2468 value_len
, is_create
);
2473 up_read(&EXT4_I(inode
)->xattr_sem
);
2480 * Like ext4_xattr_set_handle, but start from an inode. This extended
2481 * attribute modification is a filesystem transaction by itself.
2483 * Returns 0, or a negative error number on failure.
2486 ext4_xattr_set(struct inode
*inode
, int name_index
, const char *name
,
2487 const void *value
, size_t value_len
, int flags
)
2490 struct super_block
*sb
= inode
->i_sb
;
2491 int error
, retries
= 0;
2494 error
= dquot_initialize(inode
);
2499 error
= ext4_xattr_set_credits(inode
, value_len
, flags
& XATTR_CREATE
,
2504 handle
= ext4_journal_start(inode
, EXT4_HT_XATTR
, credits
);
2505 if (IS_ERR(handle
)) {
2506 error
= PTR_ERR(handle
);
2510 error
= ext4_xattr_set_handle(handle
, inode
, name_index
, name
,
2511 value
, value_len
, flags
);
2512 error2
= ext4_journal_stop(handle
);
2513 if (error
== -ENOSPC
&&
2514 ext4_should_retry_alloc(sb
, &retries
))
2524 * Shift the EA entries in the inode to create space for the increased
2527 static void ext4_xattr_shift_entries(struct ext4_xattr_entry
*entry
,
2528 int value_offs_shift
, void *to
,
2529 void *from
, size_t n
)
2531 struct ext4_xattr_entry
*last
= entry
;
2534 /* We always shift xattr headers further thus offsets get lower */
2535 BUG_ON(value_offs_shift
> 0);
2537 /* Adjust the value offsets of the entries */
2538 for (; !IS_LAST_ENTRY(last
); last
= EXT4_XATTR_NEXT(last
)) {
2539 if (!last
->e_value_inum
&& last
->e_value_size
) {
2540 new_offs
= le16_to_cpu(last
->e_value_offs
) +
2542 last
->e_value_offs
= cpu_to_le16(new_offs
);
2545 /* Shift the entries by n bytes */
2546 memmove(to
, from
, n
);
2550 * Move xattr pointed to by 'entry' from inode into external xattr block
2552 static int ext4_xattr_move_to_block(handle_t
*handle
, struct inode
*inode
,
2553 struct ext4_inode
*raw_inode
,
2554 struct ext4_xattr_entry
*entry
)
2556 struct ext4_xattr_ibody_find
*is
= NULL
;
2557 struct ext4_xattr_block_find
*bs
= NULL
;
2558 char *buffer
= NULL
, *b_entry_name
= NULL
;
2559 size_t value_size
= le32_to_cpu(entry
->e_value_size
);
2560 struct ext4_xattr_info i
= {
2563 .name_index
= entry
->e_name_index
,
2564 .in_inode
= !!entry
->e_value_inum
,
2566 struct ext4_xattr_ibody_header
*header
= IHDR(inode
, raw_inode
);
2569 is
= kzalloc(sizeof(struct ext4_xattr_ibody_find
), GFP_NOFS
);
2570 bs
= kzalloc(sizeof(struct ext4_xattr_block_find
), GFP_NOFS
);
2571 buffer
= kmalloc(value_size
, GFP_NOFS
);
2572 b_entry_name
= kmalloc(entry
->e_name_len
+ 1, GFP_NOFS
);
2573 if (!is
|| !bs
|| !buffer
|| !b_entry_name
) {
2578 is
->s
.not_found
= -ENODATA
;
2579 bs
->s
.not_found
= -ENODATA
;
2583 /* Save the entry name and the entry value */
2584 if (entry
->e_value_inum
) {
2585 error
= ext4_xattr_inode_get(inode
, entry
, buffer
, value_size
);
2589 size_t value_offs
= le16_to_cpu(entry
->e_value_offs
);
2590 memcpy(buffer
, (void *)IFIRST(header
) + value_offs
, value_size
);
2593 memcpy(b_entry_name
, entry
->e_name
, entry
->e_name_len
);
2594 b_entry_name
[entry
->e_name_len
] = '\0';
2595 i
.name
= b_entry_name
;
2597 error
= ext4_get_inode_loc(inode
, &is
->iloc
);
2601 error
= ext4_xattr_ibody_find(inode
, &i
, is
);
2605 /* Remove the chosen entry from the inode */
2606 error
= ext4_xattr_ibody_set(handle
, inode
, &i
, is
);
2611 i
.value_len
= value_size
;
2612 error
= ext4_xattr_block_find(inode
, &i
, bs
);
2616 /* Add entry which was removed from the inode into the block */
2617 error
= ext4_xattr_block_set(handle
, inode
, &i
, bs
);
2622 kfree(b_entry_name
);
2625 brelse(is
->iloc
.bh
);
2632 static int ext4_xattr_make_inode_space(handle_t
*handle
, struct inode
*inode
,
2633 struct ext4_inode
*raw_inode
,
2634 int isize_diff
, size_t ifree
,
2635 size_t bfree
, int *total_ino
)
2637 struct ext4_xattr_ibody_header
*header
= IHDR(inode
, raw_inode
);
2638 struct ext4_xattr_entry
*small_entry
;
2639 struct ext4_xattr_entry
*entry
;
2640 struct ext4_xattr_entry
*last
;
2641 unsigned int entry_size
; /* EA entry size */
2642 unsigned int total_size
; /* EA entry size + value size */
2643 unsigned int min_total_size
;
2646 while (isize_diff
> ifree
) {
2649 min_total_size
= ~0U;
2650 last
= IFIRST(header
);
2651 /* Find the entry best suited to be pushed into EA block */
2652 for (; !IS_LAST_ENTRY(last
); last
= EXT4_XATTR_NEXT(last
)) {
2653 total_size
= EXT4_XATTR_LEN(last
->e_name_len
);
2654 if (!last
->e_value_inum
)
2655 total_size
+= EXT4_XATTR_SIZE(
2656 le32_to_cpu(last
->e_value_size
));
2657 if (total_size
<= bfree
&&
2658 total_size
< min_total_size
) {
2659 if (total_size
+ ifree
< isize_diff
) {
2663 min_total_size
= total_size
;
2668 if (entry
== NULL
) {
2669 if (small_entry
== NULL
)
2671 entry
= small_entry
;
2674 entry_size
= EXT4_XATTR_LEN(entry
->e_name_len
);
2675 total_size
= entry_size
;
2676 if (!entry
->e_value_inum
)
2677 total_size
+= EXT4_XATTR_SIZE(
2678 le32_to_cpu(entry
->e_value_size
));
2679 error
= ext4_xattr_move_to_block(handle
, inode
, raw_inode
,
2684 *total_ino
-= entry_size
;
2685 ifree
+= total_size
;
2686 bfree
-= total_size
;
2693 * Expand an inode by new_extra_isize bytes when EAs are present.
2694 * Returns 0 on success or negative error number on failure.
2696 int ext4_expand_extra_isize_ea(struct inode
*inode
, int new_extra_isize
,
2697 struct ext4_inode
*raw_inode
, handle_t
*handle
)
2699 struct ext4_xattr_ibody_header
*header
;
2700 struct buffer_head
*bh
;
2701 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2702 static unsigned int mnt_count
;
2704 size_t ifree
, bfree
;
2707 int error
= 0, tried_min_extra_isize
= 0;
2708 int s_min_extra_isize
= le16_to_cpu(sbi
->s_es
->s_min_extra_isize
);
2709 int isize_diff
; /* How much do we need to grow i_extra_isize */
2712 isize_diff
= new_extra_isize
- EXT4_I(inode
)->i_extra_isize
;
2713 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
2716 header
= IHDR(inode
, raw_inode
);
2719 * Check if enough free space is available in the inode to shift the
2720 * entries ahead by new_extra_isize.
2723 base
= IFIRST(header
);
2724 end
= (void *)raw_inode
+ EXT4_SB(inode
->i_sb
)->s_inode_size
;
2725 min_offs
= end
- base
;
2726 total_ino
= sizeof(struct ext4_xattr_ibody_header
);
2728 error
= xattr_check_inode(inode
, header
, end
);
2732 ifree
= ext4_xattr_free_space(base
, &min_offs
, base
, &total_ino
);
2733 if (ifree
>= isize_diff
)
2737 * Enough free space isn't available in the inode, check if
2738 * EA block can hold new_extra_isize bytes.
2740 if (EXT4_I(inode
)->i_file_acl
) {
2741 bh
= sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
);
2745 error
= ext4_xattr_check_block(inode
, bh
);
2749 end
= bh
->b_data
+ bh
->b_size
;
2750 min_offs
= end
- base
;
2751 bfree
= ext4_xattr_free_space(BFIRST(bh
), &min_offs
, base
,
2754 if (bfree
+ ifree
< isize_diff
) {
2755 if (!tried_min_extra_isize
&& s_min_extra_isize
) {
2756 tried_min_extra_isize
++;
2757 new_extra_isize
= s_min_extra_isize
;
2764 bfree
= inode
->i_sb
->s_blocksize
;
2767 error
= ext4_xattr_make_inode_space(handle
, inode
, raw_inode
,
2768 isize_diff
, ifree
, bfree
,
2771 if (error
== -ENOSPC
&& !tried_min_extra_isize
&&
2772 s_min_extra_isize
) {
2773 tried_min_extra_isize
++;
2774 new_extra_isize
= s_min_extra_isize
;
2780 /* Adjust the offsets and shift the remaining entries ahead */
2781 ext4_xattr_shift_entries(IFIRST(header
), EXT4_I(inode
)->i_extra_isize
2782 - new_extra_isize
, (void *)raw_inode
+
2783 EXT4_GOOD_OLD_INODE_SIZE
+ new_extra_isize
,
2784 (void *)header
, total_ino
);
2785 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
2788 if (error
&& (mnt_count
!= le16_to_cpu(sbi
->s_es
->s_mnt_count
))) {
2789 ext4_warning(inode
->i_sb
, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2791 mnt_count
= le16_to_cpu(sbi
->s_es
->s_mnt_count
);
2796 #define EIA_INCR 16 /* must be 2^n */
2797 #define EIA_MASK (EIA_INCR - 1)
2799 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2800 * If @ea_inode_array is new or full it will be grown and the old
2801 * contents copied over.
2804 ext4_expand_inode_array(struct ext4_xattr_inode_array
**ea_inode_array
,
2805 struct inode
*inode
)
2807 if (*ea_inode_array
== NULL
) {
2809 * Start with 15 inodes, so it fits into a power-of-two size.
2810 * If *ea_inode_array is NULL, this is essentially offsetof()
2813 kmalloc(offsetof(struct ext4_xattr_inode_array
,
2816 if (*ea_inode_array
== NULL
)
2818 (*ea_inode_array
)->count
= 0;
2819 } else if (((*ea_inode_array
)->count
& EIA_MASK
) == EIA_MASK
) {
2820 /* expand the array once all 15 + n * 16 slots are full */
2821 struct ext4_xattr_inode_array
*new_array
= NULL
;
2822 int count
= (*ea_inode_array
)->count
;
2824 /* if new_array is NULL, this is essentially offsetof() */
2825 new_array
= kmalloc(
2826 offsetof(struct ext4_xattr_inode_array
,
2827 inodes
[count
+ EIA_INCR
]),
2829 if (new_array
== NULL
)
2831 memcpy(new_array
, *ea_inode_array
,
2832 offsetof(struct ext4_xattr_inode_array
, inodes
[count
]));
2833 kfree(*ea_inode_array
);
2834 *ea_inode_array
= new_array
;
2836 (*ea_inode_array
)->inodes
[(*ea_inode_array
)->count
++] = inode
;
2841 * ext4_xattr_delete_inode()
2843 * Free extended attribute resources associated with this inode. Traverse
2844 * all entries and decrement reference on any xattr inodes associated with this
2845 * inode. This is called immediately before an inode is freed. We have exclusive
2846 * access to the inode. If an orphan inode is deleted it will also release its
2847 * references on xattr block and xattr inodes.
2849 int ext4_xattr_delete_inode(handle_t
*handle
, struct inode
*inode
,
2850 struct ext4_xattr_inode_array
**ea_inode_array
,
2853 struct buffer_head
*bh
= NULL
;
2854 struct ext4_xattr_ibody_header
*header
;
2855 struct ext4_iloc iloc
= { .bh
= NULL
};
2856 struct ext4_xattr_entry
*entry
;
2857 struct inode
*ea_inode
;
2860 error
= ext4_xattr_ensure_credits(handle
, inode
, extra_credits
,
2863 false /* block_csum */);
2865 EXT4_ERROR_INODE(inode
, "ensure credits (error %d)", error
);
2869 if (ext4_has_feature_ea_inode(inode
->i_sb
) &&
2870 ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
2872 error
= ext4_get_inode_loc(inode
, &iloc
);
2874 EXT4_ERROR_INODE(inode
, "inode loc (error %d)", error
);
2878 error
= ext4_journal_get_write_access(handle
, iloc
.bh
);
2880 EXT4_ERROR_INODE(inode
, "write access (error %d)",
2885 header
= IHDR(inode
, ext4_raw_inode(&iloc
));
2886 if (header
->h_magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
2887 ext4_xattr_inode_dec_ref_all(handle
, inode
, iloc
.bh
,
2889 false /* block_csum */,
2892 false /* skip_quota */);
2895 if (EXT4_I(inode
)->i_file_acl
) {
2896 bh
= sb_bread(inode
->i_sb
, EXT4_I(inode
)->i_file_acl
);
2898 EXT4_ERROR_INODE(inode
, "block %llu read error",
2899 EXT4_I(inode
)->i_file_acl
);
2903 error
= ext4_xattr_check_block(inode
, bh
);
2907 if (ext4_has_feature_ea_inode(inode
->i_sb
)) {
2908 for (entry
= BFIRST(bh
); !IS_LAST_ENTRY(entry
);
2909 entry
= EXT4_XATTR_NEXT(entry
)) {
2910 if (!entry
->e_value_inum
)
2912 error
= ext4_xattr_inode_iget(inode
,
2913 le32_to_cpu(entry
->e_value_inum
),
2914 le32_to_cpu(entry
->e_hash
),
2918 ext4_xattr_inode_free_quota(inode
, ea_inode
,
2919 le32_to_cpu(entry
->e_value_size
));
2925 ext4_xattr_release_block(handle
, inode
, bh
, ea_inode_array
,
2928 * Update i_file_acl value in the same transaction that releases
2931 EXT4_I(inode
)->i_file_acl
= 0;
2932 error
= ext4_mark_inode_dirty(handle
, inode
);
2934 EXT4_ERROR_INODE(inode
, "mark inode dirty (error %d)",
2946 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array
*ea_inode_array
)
2950 if (ea_inode_array
== NULL
)
2953 for (idx
= 0; idx
< ea_inode_array
->count
; ++idx
)
2954 iput(ea_inode_array
->inodes
[idx
]);
2955 kfree(ea_inode_array
);
2959 * ext4_xattr_block_cache_insert()
2961 * Create a new entry in the extended attribute block cache, and insert
2962 * it unless such an entry is already in the cache.
2964 * Returns 0, or a negative error number on failure.
2967 ext4_xattr_block_cache_insert(struct mb_cache
*ea_block_cache
,
2968 struct buffer_head
*bh
)
2970 struct ext4_xattr_header
*header
= BHDR(bh
);
2971 __u32 hash
= le32_to_cpu(header
->h_hash
);
2972 int reusable
= le32_to_cpu(header
->h_refcount
) <
2973 EXT4_XATTR_REFCOUNT_MAX
;
2976 if (!ea_block_cache
)
2978 error
= mb_cache_entry_create(ea_block_cache
, GFP_NOFS
, hash
,
2979 bh
->b_blocknr
, reusable
);
2981 if (error
== -EBUSY
)
2982 ea_bdebug(bh
, "already in cache");
2984 ea_bdebug(bh
, "inserting [%x]", (int)hash
);
2990 * Compare two extended attribute blocks for equality.
2992 * Returns 0 if the blocks are equal, 1 if they differ, and
2993 * a negative error number on errors.
2996 ext4_xattr_cmp(struct ext4_xattr_header
*header1
,
2997 struct ext4_xattr_header
*header2
)
2999 struct ext4_xattr_entry
*entry1
, *entry2
;
3001 entry1
= ENTRY(header1
+1);
3002 entry2
= ENTRY(header2
+1);
3003 while (!IS_LAST_ENTRY(entry1
)) {
3004 if (IS_LAST_ENTRY(entry2
))
3006 if (entry1
->e_hash
!= entry2
->e_hash
||
3007 entry1
->e_name_index
!= entry2
->e_name_index
||
3008 entry1
->e_name_len
!= entry2
->e_name_len
||
3009 entry1
->e_value_size
!= entry2
->e_value_size
||
3010 entry1
->e_value_inum
!= entry2
->e_value_inum
||
3011 memcmp(entry1
->e_name
, entry2
->e_name
, entry1
->e_name_len
))
3013 if (!entry1
->e_value_inum
&&
3014 memcmp((char *)header1
+ le16_to_cpu(entry1
->e_value_offs
),
3015 (char *)header2
+ le16_to_cpu(entry2
->e_value_offs
),
3016 le32_to_cpu(entry1
->e_value_size
)))
3019 entry1
= EXT4_XATTR_NEXT(entry1
);
3020 entry2
= EXT4_XATTR_NEXT(entry2
);
3022 if (!IS_LAST_ENTRY(entry2
))
3028 * ext4_xattr_block_cache_find()
3030 * Find an identical extended attribute block.
3032 * Returns a pointer to the block found, or NULL if such a block was
3033 * not found or an error occurred.
3035 static struct buffer_head
*
3036 ext4_xattr_block_cache_find(struct inode
*inode
,
3037 struct ext4_xattr_header
*header
,
3038 struct mb_cache_entry
**pce
)
3040 __u32 hash
= le32_to_cpu(header
->h_hash
);
3041 struct mb_cache_entry
*ce
;
3042 struct mb_cache
*ea_block_cache
= EA_BLOCK_CACHE(inode
);
3044 if (!ea_block_cache
)
3046 if (!header
->h_hash
)
3047 return NULL
; /* never share */
3048 ea_idebug(inode
, "looking for cached blocks [%x]", (int)hash
);
3049 ce
= mb_cache_entry_find_first(ea_block_cache
, hash
);
3051 struct buffer_head
*bh
;
3053 bh
= sb_bread(inode
->i_sb
, ce
->e_value
);
3055 EXT4_ERROR_INODE(inode
, "block %lu read error",
3056 (unsigned long)ce
->e_value
);
3057 } else if (ext4_xattr_cmp(header
, BHDR(bh
)) == 0) {
3062 ce
= mb_cache_entry_find_next(ea_block_cache
, ce
);
3067 #define NAME_HASH_SHIFT 5
3068 #define VALUE_HASH_SHIFT 16
3071 * ext4_xattr_hash_entry()
3073 * Compute the hash of an extended attribute.
3075 static __le32
ext4_xattr_hash_entry(char *name
, size_t name_len
, __le32
*value
,
3080 while (name_len
--) {
3081 hash
= (hash
<< NAME_HASH_SHIFT
) ^
3082 (hash
>> (8*sizeof(hash
) - NAME_HASH_SHIFT
)) ^
3085 while (value_count
--) {
3086 hash
= (hash
<< VALUE_HASH_SHIFT
) ^
3087 (hash
>> (8*sizeof(hash
) - VALUE_HASH_SHIFT
)) ^
3088 le32_to_cpu(*value
++);
3090 return cpu_to_le32(hash
);
3093 #undef NAME_HASH_SHIFT
3094 #undef VALUE_HASH_SHIFT
3096 #define BLOCK_HASH_SHIFT 16
3099 * ext4_xattr_rehash()
3101 * Re-compute the extended attribute hash value after an entry has changed.
3103 static void ext4_xattr_rehash(struct ext4_xattr_header
*header
)
3105 struct ext4_xattr_entry
*here
;
3108 here
= ENTRY(header
+1);
3109 while (!IS_LAST_ENTRY(here
)) {
3110 if (!here
->e_hash
) {
3111 /* Block is not shared if an entry's hash value == 0 */
3115 hash
= (hash
<< BLOCK_HASH_SHIFT
) ^
3116 (hash
>> (8*sizeof(hash
) - BLOCK_HASH_SHIFT
)) ^
3117 le32_to_cpu(here
->e_hash
);
3118 here
= EXT4_XATTR_NEXT(here
);
3120 header
->h_hash
= cpu_to_le32(hash
);
3123 #undef BLOCK_HASH_SHIFT
3125 #define HASH_BUCKET_BITS 10
3128 ext4_xattr_create_cache(void)
3130 return mb_cache_create(HASH_BUCKET_BITS
);
3133 void ext4_xattr_destroy_cache(struct mb_cache
*cache
)
3136 mb_cache_destroy(cache
);