1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/xattr.c
9 * Extended attributes for symlinks and special files added per
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 <linux/iversion.h>
60 #include "ext4_jbd2.h"
65 #ifdef EXT4_XATTR_DEBUG
66 # define ea_idebug(inode, fmt, ...) \
67 printk(KERN_DEBUG "inode %s:%lu: " fmt "\n", \
68 inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
69 # define ea_bdebug(bh, fmt, ...) \
70 printk(KERN_DEBUG "block %pg:%lu: " fmt "\n", \
71 bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
73 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
74 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
77 static void ext4_xattr_block_cache_insert(struct mb_cache *,
78 struct buffer_head *);
79 static struct buffer_head *
80 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
81 struct mb_cache_entry **);
82 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
84 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value,
86 static void ext4_xattr_rehash(struct ext4_xattr_header *);
88 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
89 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
90 #ifdef CONFIG_EXT4_FS_POSIX_ACL
91 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
92 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
94 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
95 #ifdef CONFIG_EXT4_FS_SECURITY
96 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
98 [EXT4_XATTR_INDEX_HURD] = &ext4_xattr_hurd_handler,
101 const struct xattr_handler *ext4_xattr_handlers[] = {
102 &ext4_xattr_user_handler,
103 &ext4_xattr_trusted_handler,
104 #ifdef CONFIG_EXT4_FS_POSIX_ACL
105 &posix_acl_access_xattr_handler,
106 &posix_acl_default_xattr_handler,
108 #ifdef CONFIG_EXT4_FS_SECURITY
109 &ext4_xattr_security_handler,
111 &ext4_xattr_hurd_handler,
115 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
116 inode->i_sb->s_fs_info)->s_ea_block_cache)
118 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
119 inode->i_sb->s_fs_info)->s_ea_inode_cache)
122 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
123 struct inode *inode);
125 #ifdef CONFIG_LOCKDEP
126 void ext4_xattr_inode_set_class(struct inode *ea_inode)
128 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
132 static __le32 ext4_xattr_block_csum(struct inode *inode,
134 struct ext4_xattr_header *hdr)
136 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
138 __le64 dsk_block_nr = cpu_to_le64(block_nr);
139 __u32 dummy_csum = 0;
140 int offset = offsetof(struct ext4_xattr_header, h_checksum);
142 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
143 sizeof(dsk_block_nr));
144 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
145 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
146 offset += sizeof(dummy_csum);
147 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
148 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
150 return cpu_to_le32(csum);
153 static int ext4_xattr_block_csum_verify(struct inode *inode,
154 struct buffer_head *bh)
156 struct ext4_xattr_header *hdr = BHDR(bh);
159 if (ext4_has_metadata_csum(inode->i_sb)) {
161 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
162 bh->b_blocknr, hdr));
168 static void ext4_xattr_block_csum_set(struct inode *inode,
169 struct buffer_head *bh)
171 if (ext4_has_metadata_csum(inode->i_sb))
172 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
173 bh->b_blocknr, BHDR(bh));
176 static inline const struct xattr_handler *
177 ext4_xattr_handler(int name_index)
179 const struct xattr_handler *handler = NULL;
181 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
182 handler = ext4_xattr_handler_map[name_index];
187 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
190 struct ext4_xattr_entry *e = entry;
192 /* Find the end of the names list */
193 while (!IS_LAST_ENTRY(e)) {
194 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
195 if ((void *)next >= end)
196 return -EFSCORRUPTED;
197 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
198 return -EFSCORRUPTED;
202 /* Check the values */
203 while (!IS_LAST_ENTRY(entry)) {
204 u32 size = le32_to_cpu(entry->e_value_size);
206 if (size > EXT4_XATTR_SIZE_MAX)
207 return -EFSCORRUPTED;
209 if (size != 0 && entry->e_value_inum == 0) {
210 u16 offs = le16_to_cpu(entry->e_value_offs);
214 * The value cannot overlap the names, and the value
215 * with padding cannot extend beyond 'end'. Check both
216 * the padded and unpadded sizes, since the size may
217 * overflow to 0 when adding padding.
219 if (offs > end - value_start)
220 return -EFSCORRUPTED;
221 value = value_start + offs;
222 if (value < (void *)e + sizeof(u32) ||
223 size > end - value ||
224 EXT4_XATTR_SIZE(size) > end - value)
225 return -EFSCORRUPTED;
227 entry = EXT4_XATTR_NEXT(entry);
234 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
235 const char *function, unsigned int line)
237 int error = -EFSCORRUPTED;
239 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
240 BHDR(bh)->h_blocks != cpu_to_le32(1))
242 if (buffer_verified(bh))
246 if (!ext4_xattr_block_csum_verify(inode, bh))
248 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
252 __ext4_error_inode(inode, function, line, 0, -error,
253 "corrupted xattr block %llu",
254 (unsigned long long) bh->b_blocknr);
256 set_buffer_verified(bh);
260 #define ext4_xattr_check_block(inode, bh) \
261 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
265 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
266 void *end, const char *function, unsigned int line)
268 int error = -EFSCORRUPTED;
270 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
271 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
273 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
276 __ext4_error_inode(inode, function, line, 0, -error,
277 "corrupted in-inode xattr");
281 #define xattr_check_inode(inode, header, end) \
282 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
285 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
286 void *end, int name_index, const char *name, int sorted)
288 struct ext4_xattr_entry *entry, *next;
294 name_len = strlen(name);
295 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
296 next = EXT4_XATTR_NEXT(entry);
297 if ((void *) next >= end) {
298 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
299 return -EFSCORRUPTED;
301 cmp = name_index - entry->e_name_index;
303 cmp = name_len - entry->e_name_len;
305 cmp = memcmp(name, entry->e_name, name_len);
306 if (cmp <= 0 && (sorted || cmp == 0))
310 return cmp ? -ENODATA : 0;
314 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
316 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
319 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
321 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
322 (u32) inode_peek_iversion_raw(ea_inode);
325 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
327 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
328 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
331 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
333 return (u32)ea_inode->i_atime.tv_sec;
336 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
338 ea_inode->i_atime.tv_sec = hash;
342 * Read the EA value from an inode.
344 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
346 int blocksize = 1 << ea_inode->i_blkbits;
347 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
348 int tail_size = (size % blocksize) ?: blocksize;
349 struct buffer_head *bhs_inline[8];
350 struct buffer_head **bhs = bhs_inline;
353 if (bh_count > ARRAY_SIZE(bhs_inline)) {
354 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
359 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
360 true /* wait */, bhs);
364 for (i = 0; i < bh_count; i++) {
365 /* There shouldn't be any holes in ea_inode. */
370 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
371 i < bh_count - 1 ? blocksize : tail_size);
375 for (i = 0; i < bh_count; i++)
378 if (bhs != bhs_inline)
383 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
385 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
386 u32 ea_inode_hash, struct inode **ea_inode)
391 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
393 err = PTR_ERR(inode);
394 ext4_error(parent->i_sb,
395 "error while reading EA inode %lu err=%d", ea_ino,
400 if (is_bad_inode(inode)) {
401 ext4_error(parent->i_sb,
402 "error while reading EA inode %lu is_bad_inode",
408 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
409 ext4_error(parent->i_sb,
410 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
416 ext4_xattr_inode_set_class(inode);
419 * Check whether this is an old Lustre-style xattr inode. Lustre
420 * implementation does not have hash validation, rather it has a
421 * backpointer from ea_inode to the parent inode.
423 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
424 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
425 inode->i_generation == parent->i_generation) {
426 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
427 ext4_xattr_inode_set_ref(inode, 1);
430 inode->i_flags |= S_NOQUOTA;
441 /* Remove entry from mbcache when EA inode is getting evicted */
442 void ext4_evict_ea_inode(struct inode *inode)
444 struct mb_cache_entry *oe;
446 if (!EA_INODE_CACHE(inode))
448 /* Wait for entry to get unused so that we can remove it */
449 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
450 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
451 mb_cache_entry_wait_unused(oe);
452 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
457 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
458 struct ext4_xattr_entry *entry, void *buffer,
463 /* Verify stored hash matches calculated hash. */
464 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
465 if (hash != ext4_xattr_inode_get_hash(ea_inode))
466 return -EFSCORRUPTED;
469 __le32 e_hash, tmp_data;
471 /* Verify entry hash. */
472 tmp_data = cpu_to_le32(hash);
473 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
476 if (e_hash == entry->e_hash)
480 * Not good. Maybe the entry hash was calculated
481 * using the buggy signed char version?
483 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
485 if (e_hash == entry->e_hash)
488 /* Still no match - bad */
489 return -EFSCORRUPTED;
495 * Read xattr value from the EA inode.
498 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
499 void *buffer, size_t size)
501 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
502 struct inode *ea_inode;
505 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
506 le32_to_cpu(entry->e_hash), &ea_inode);
512 if (i_size_read(ea_inode) != size) {
513 ext4_warning_inode(ea_inode,
514 "ea_inode file size=%llu entry size=%zu",
515 i_size_read(ea_inode), size);
520 err = ext4_xattr_inode_read(ea_inode, buffer, size);
524 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
525 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
528 ext4_warning_inode(ea_inode,
529 "EA inode hash validation failed");
534 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
535 ext4_xattr_inode_get_hash(ea_inode),
536 ea_inode->i_ino, true /* reusable */);
544 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
545 void *buffer, size_t buffer_size)
547 struct buffer_head *bh = NULL;
548 struct ext4_xattr_entry *entry;
552 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
554 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
555 name_index, name, buffer, (long)buffer_size);
557 if (!EXT4_I(inode)->i_file_acl)
559 ea_idebug(inode, "reading block %llu",
560 (unsigned long long)EXT4_I(inode)->i_file_acl);
561 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
564 ea_bdebug(bh, "b_count=%d, refcount=%d",
565 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
566 error = ext4_xattr_check_block(inode, bh);
569 ext4_xattr_block_cache_insert(ea_block_cache, bh);
571 end = bh->b_data + bh->b_size;
572 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
575 size = le32_to_cpu(entry->e_value_size);
577 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
580 if (size > buffer_size)
582 if (entry->e_value_inum) {
583 error = ext4_xattr_inode_get(inode, entry, buffer,
588 u16 offset = le16_to_cpu(entry->e_value_offs);
589 void *p = bh->b_data + offset;
591 if (unlikely(p + size > end))
593 memcpy(buffer, p, size);
604 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
605 void *buffer, size_t buffer_size)
607 struct ext4_xattr_ibody_header *header;
608 struct ext4_xattr_entry *entry;
609 struct ext4_inode *raw_inode;
610 struct ext4_iloc iloc;
615 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
617 error = ext4_get_inode_loc(inode, &iloc);
620 raw_inode = ext4_raw_inode(&iloc);
621 header = IHDR(inode, raw_inode);
622 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
623 error = xattr_check_inode(inode, header, end);
626 entry = IFIRST(header);
627 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
630 size = le32_to_cpu(entry->e_value_size);
632 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
635 if (size > buffer_size)
637 if (entry->e_value_inum) {
638 error = ext4_xattr_inode_get(inode, entry, buffer,
643 u16 offset = le16_to_cpu(entry->e_value_offs);
644 void *p = (void *)IFIRST(header) + offset;
646 if (unlikely(p + size > end))
648 memcpy(buffer, p, size);
661 * Copy an extended attribute into the buffer
662 * provided, or compute the buffer size required.
663 * Buffer is NULL to compute the size of the buffer required.
665 * Returns a negative error number on failure, or the number of bytes
666 * used / required on success.
669 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
670 void *buffer, size_t buffer_size)
674 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
677 if (strlen(name) > 255)
680 down_read(&EXT4_I(inode)->xattr_sem);
681 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
683 if (error == -ENODATA)
684 error = ext4_xattr_block_get(inode, name_index, name, buffer,
686 up_read(&EXT4_I(inode)->xattr_sem);
691 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
692 char *buffer, size_t buffer_size)
694 size_t rest = buffer_size;
696 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
697 const struct xattr_handler *handler =
698 ext4_xattr_handler(entry->e_name_index);
700 if (handler && (!handler->list || handler->list(dentry))) {
701 const char *prefix = handler->prefix ?: handler->name;
702 size_t prefix_len = strlen(prefix);
703 size_t size = prefix_len + entry->e_name_len + 1;
708 memcpy(buffer, prefix, prefix_len);
709 buffer += prefix_len;
710 memcpy(buffer, entry->e_name, entry->e_name_len);
711 buffer += entry->e_name_len;
717 return buffer_size - rest; /* total size */
721 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
723 struct inode *inode = d_inode(dentry);
724 struct buffer_head *bh = NULL;
727 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
728 buffer, (long)buffer_size);
730 if (!EXT4_I(inode)->i_file_acl)
732 ea_idebug(inode, "reading block %llu",
733 (unsigned long long)EXT4_I(inode)->i_file_acl);
734 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
737 ea_bdebug(bh, "b_count=%d, refcount=%d",
738 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
739 error = ext4_xattr_check_block(inode, bh);
742 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
743 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
751 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
753 struct inode *inode = d_inode(dentry);
754 struct ext4_xattr_ibody_header *header;
755 struct ext4_inode *raw_inode;
756 struct ext4_iloc iloc;
760 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
762 error = ext4_get_inode_loc(inode, &iloc);
765 raw_inode = ext4_raw_inode(&iloc);
766 header = IHDR(inode, raw_inode);
767 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
768 error = xattr_check_inode(inode, header, end);
771 error = ext4_xattr_list_entries(dentry, IFIRST(header),
772 buffer, buffer_size);
780 * Inode operation listxattr()
782 * d_inode(dentry)->i_rwsem: don't care
784 * Copy a list of attribute names into the buffer
785 * provided, or compute the buffer size required.
786 * Buffer is NULL to compute the size of the buffer required.
788 * Returns a negative error number on failure, or the number of bytes
789 * used / required on success.
792 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
796 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
797 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
804 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
809 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
814 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
817 static void ext4_xattr_update_super_block(handle_t *handle,
818 struct super_block *sb)
820 if (ext4_has_feature_xattr(sb))
823 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
824 if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
825 EXT4_JTR_NONE) == 0) {
826 lock_buffer(EXT4_SB(sb)->s_sbh);
827 ext4_set_feature_xattr(sb);
828 ext4_superblock_csum_set(sb);
829 unlock_buffer(EXT4_SB(sb)->s_sbh);
830 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
834 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
836 struct ext4_iloc iloc = { .bh = NULL };
837 struct buffer_head *bh = NULL;
838 struct ext4_inode *raw_inode;
839 struct ext4_xattr_ibody_header *header;
840 struct ext4_xattr_entry *entry;
841 qsize_t ea_inode_refs = 0;
845 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
847 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
848 ret = ext4_get_inode_loc(inode, &iloc);
851 raw_inode = ext4_raw_inode(&iloc);
852 header = IHDR(inode, raw_inode);
853 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
854 ret = xattr_check_inode(inode, header, end);
858 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
859 entry = EXT4_XATTR_NEXT(entry))
860 if (entry->e_value_inum)
864 if (EXT4_I(inode)->i_file_acl) {
865 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
872 ret = ext4_xattr_check_block(inode, bh);
876 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
877 entry = EXT4_XATTR_NEXT(entry))
878 if (entry->e_value_inum)
881 *usage = ea_inode_refs + 1;
889 static inline size_t round_up_cluster(struct inode *inode, size_t length)
891 struct super_block *sb = inode->i_sb;
892 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
894 size_t mask = ~(cluster_size - 1);
896 return (length + cluster_size - 1) & mask;
899 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
903 err = dquot_alloc_inode(inode);
906 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
908 dquot_free_inode(inode);
912 static void ext4_xattr_inode_free_quota(struct inode *parent,
913 struct inode *ea_inode,
917 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
919 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
920 dquot_free_inode(parent);
923 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
924 struct buffer_head *block_bh, size_t value_len,
931 * 1) Owner inode update
932 * 2) Ref count update on old xattr block
934 * 4) block bitmap update for new xattr block
935 * 5) group descriptor for new xattr block
936 * 6) block bitmap update for old xattr block
937 * 7) group descriptor for old block
939 * 6 & 7 can happen if we have two racing threads T_a and T_b
940 * which are each trying to set an xattr on inodes I_a and I_b
941 * which were both initially sharing an xattr block.
946 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
949 * In case of inline data, we may push out the data to a block,
950 * so we need to reserve credits for this eventuality
952 if (inode && ext4_has_inline_data(inode))
953 credits += ext4_writepage_trans_blocks(inode) + 1;
955 /* We are done if ea_inode feature is not enabled. */
956 if (!ext4_has_feature_ea_inode(sb))
959 /* New ea_inode, inode map, block bitmap, group descriptor. */
963 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
965 /* Indirection block or one level of extent tree. */
968 /* Block bitmap and group descriptor updates for each block. */
969 credits += blocks * 2;
971 /* Blocks themselves. */
975 /* Dereference ea_inode holding old xattr value.
976 * Old ea_inode, inode map, block bitmap, group descriptor.
980 /* Data blocks for old ea_inode. */
981 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
983 /* Indirection block or one level of extent tree for old
988 /* Block bitmap and group descriptor updates for each block. */
989 credits += blocks * 2;
992 /* We may need to clone the existing xattr block in which case we need
993 * to increment ref counts for existing ea_inodes referenced by it.
996 struct ext4_xattr_entry *entry = BFIRST(block_bh);
998 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
999 if (entry->e_value_inum)
1000 /* Ref count update on ea_inode. */
1006 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1009 struct ext4_iloc iloc;
1013 inode_lock(ea_inode);
1015 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1019 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1020 ref_count += ref_change;
1021 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1023 if (ref_change > 0) {
1024 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1025 ea_inode->i_ino, ref_count);
1027 if (ref_count == 1) {
1028 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1029 ea_inode->i_ino, ea_inode->i_nlink);
1031 set_nlink(ea_inode, 1);
1032 ext4_orphan_del(handle, ea_inode);
1035 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1036 ea_inode->i_ino, ref_count);
1038 if (ref_count == 0) {
1039 WARN_ONCE(ea_inode->i_nlink != 1,
1040 "EA inode %lu i_nlink=%u",
1041 ea_inode->i_ino, ea_inode->i_nlink);
1043 clear_nlink(ea_inode);
1044 ext4_orphan_add(handle, ea_inode);
1048 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1050 ext4_warning_inode(ea_inode,
1051 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1053 inode_unlock(ea_inode);
1057 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1059 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1062 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1064 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1067 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1068 struct ext4_xattr_entry *first)
1070 struct inode *ea_inode;
1071 struct ext4_xattr_entry *entry;
1072 struct ext4_xattr_entry *failed_entry;
1073 unsigned int ea_ino;
1076 for (entry = first; !IS_LAST_ENTRY(entry);
1077 entry = EXT4_XATTR_NEXT(entry)) {
1078 if (!entry->e_value_inum)
1080 ea_ino = le32_to_cpu(entry->e_value_inum);
1081 err = ext4_xattr_inode_iget(parent, ea_ino,
1082 le32_to_cpu(entry->e_hash),
1086 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1088 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1098 failed_entry = entry;
1100 for (entry = first; entry != failed_entry;
1101 entry = EXT4_XATTR_NEXT(entry)) {
1102 if (!entry->e_value_inum)
1104 ea_ino = le32_to_cpu(entry->e_value_inum);
1105 err = ext4_xattr_inode_iget(parent, ea_ino,
1106 le32_to_cpu(entry->e_hash),
1109 ext4_warning(parent->i_sb,
1110 "cleanup ea_ino %u iget error %d", ea_ino,
1114 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1116 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1123 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1124 struct buffer_head *bh, bool block_csum, bool dirty)
1130 ext4_xattr_block_csum_set(inode, bh);
1131 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1133 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1142 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1143 struct buffer_head *bh,
1144 struct ext4_xattr_entry *first, bool block_csum,
1145 struct ext4_xattr_inode_array **ea_inode_array,
1146 int extra_credits, bool skip_quota)
1148 struct inode *ea_inode;
1149 struct ext4_xattr_entry *entry;
1151 unsigned int ea_ino;
1155 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1156 credits = 2 + extra_credits;
1158 for (entry = first; !IS_LAST_ENTRY(entry);
1159 entry = EXT4_XATTR_NEXT(entry)) {
1160 if (!entry->e_value_inum)
1162 ea_ino = le32_to_cpu(entry->e_value_inum);
1163 err = ext4_xattr_inode_iget(parent, ea_ino,
1164 le32_to_cpu(entry->e_hash),
1169 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1171 ext4_warning_inode(ea_inode,
1172 "Expand inode array err=%d", err);
1177 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1178 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1179 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1182 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1187 err = ext4_journal_get_write_access(handle,
1188 parent->i_sb, bh, EXT4_JTR_NONE);
1190 ext4_warning_inode(ea_inode,
1191 "Re-get write access err=%d",
1197 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1199 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1205 ext4_xattr_inode_free_quota(parent, ea_inode,
1206 le32_to_cpu(entry->e_value_size));
1209 * Forget about ea_inode within the same transaction that
1210 * decrements the ref count. This avoids duplicate decrements in
1211 * case the rest of the work spills over to subsequent
1214 entry->e_value_inum = 0;
1215 entry->e_value_size = 0;
1222 * Note that we are deliberately skipping csum calculation for
1223 * the final update because we do not expect any journal
1224 * restarts until xattr block is freed.
1227 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1229 ext4_warning_inode(parent,
1230 "handle dirty metadata err=%d", err);
1235 * Release the xattr block BH: If the reference count is > 1, decrement it;
1236 * otherwise free the block.
1239 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1240 struct buffer_head *bh,
1241 struct ext4_xattr_inode_array **ea_inode_array,
1244 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1248 BUFFER_TRACE(bh, "get_write_access");
1249 error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1256 hash = le32_to_cpu(BHDR(bh)->h_hash);
1257 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1259 ea_bdebug(bh, "refcount now=0; freeing");
1261 * This must happen under buffer lock for
1262 * ext4_xattr_block_set() to reliably detect freed block
1264 if (ea_block_cache) {
1265 struct mb_cache_entry *oe;
1267 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1271 mb_cache_entry_wait_unused(oe);
1272 mb_cache_entry_put(ea_block_cache, oe);
1279 if (ext4_has_feature_ea_inode(inode->i_sb))
1280 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1282 true /* block_csum */,
1285 true /* skip_quota */);
1286 ext4_free_blocks(handle, inode, bh, 0, 1,
1287 EXT4_FREE_BLOCKS_METADATA |
1288 EXT4_FREE_BLOCKS_FORGET);
1291 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1292 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1293 struct mb_cache_entry *ce;
1295 if (ea_block_cache) {
1296 ce = mb_cache_entry_get(ea_block_cache, hash,
1299 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1300 mb_cache_entry_put(ea_block_cache, ce);
1305 ext4_xattr_block_csum_set(inode, bh);
1307 * Beware of this ugliness: Releasing of xattr block references
1308 * from different inodes can race and so we have to protect
1309 * from a race where someone else frees the block (and releases
1310 * its journal_head) before we are done dirtying the buffer. In
1311 * nojournal mode this race is harmless and we actually cannot
1312 * call ext4_handle_dirty_metadata() with locked buffer as
1313 * that function can call sync_dirty_buffer() so for that case
1314 * we handle the dirtying after unlocking the buffer.
1316 if (ext4_handle_valid(handle))
1317 error = ext4_handle_dirty_metadata(handle, inode, bh);
1319 if (!ext4_handle_valid(handle))
1320 error = ext4_handle_dirty_metadata(handle, inode, bh);
1322 ext4_handle_sync(handle);
1323 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1324 ea_bdebug(bh, "refcount now=%d; releasing",
1325 le32_to_cpu(BHDR(bh)->h_refcount));
1328 ext4_std_error(inode->i_sb, error);
1333 * Find the available free space for EAs. This also returns the total number of
1334 * bytes used by EA entries.
1336 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1337 size_t *min_offs, void *base, int *total)
1339 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1340 if (!last->e_value_inum && last->e_value_size) {
1341 size_t offs = le16_to_cpu(last->e_value_offs);
1342 if (offs < *min_offs)
1346 *total += EXT4_XATTR_LEN(last->e_name_len);
1348 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1352 * Write the value of the EA in an inode.
1354 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1355 const void *buf, int bufsize)
1357 struct buffer_head *bh = NULL;
1358 unsigned long block = 0;
1359 int blocksize = ea_inode->i_sb->s_blocksize;
1360 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1361 int csize, wsize = 0;
1362 int ret = 0, ret2 = 0;
1366 while (ret >= 0 && ret < max_blocks) {
1367 struct ext4_map_blocks map;
1368 map.m_lblk = block += ret;
1369 map.m_len = max_blocks -= ret;
1371 ret = ext4_map_blocks(handle, ea_inode, &map,
1372 EXT4_GET_BLOCKS_CREATE);
1374 ext4_mark_inode_dirty(handle, ea_inode);
1375 if (ret == -ENOSPC &&
1376 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1388 while (wsize < bufsize) {
1390 csize = (bufsize - wsize) > blocksize ? blocksize :
1392 bh = ext4_getblk(handle, ea_inode, block, 0);
1397 EXT4_ERROR_INODE(ea_inode,
1398 "ext4_getblk() return bh = NULL");
1399 return -EFSCORRUPTED;
1401 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1406 memcpy(bh->b_data, buf, csize);
1407 set_buffer_uptodate(bh);
1408 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1415 inode_lock(ea_inode);
1416 i_size_write(ea_inode, wsize);
1417 ext4_update_i_disksize(ea_inode, wsize);
1418 inode_unlock(ea_inode);
1420 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1421 if (unlikely(ret2 && !ret))
1431 * Create an inode to store the value of a large EA.
1433 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1434 struct inode *inode, u32 hash)
1436 struct inode *ea_inode = NULL;
1437 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1441 * Let the next inode be the goal, so we try and allocate the EA inode
1442 * in the same group, or nearby one.
1444 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1445 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1447 if (!IS_ERR(ea_inode)) {
1448 ea_inode->i_op = &ext4_file_inode_operations;
1449 ea_inode->i_fop = &ext4_file_operations;
1450 ext4_set_aops(ea_inode);
1451 ext4_xattr_inode_set_class(ea_inode);
1452 unlock_new_inode(ea_inode);
1453 ext4_xattr_inode_set_ref(ea_inode, 1);
1454 ext4_xattr_inode_set_hash(ea_inode, hash);
1455 err = ext4_mark_inode_dirty(handle, ea_inode);
1457 err = ext4_inode_attach_jinode(ea_inode);
1459 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1460 ext4_warning_inode(ea_inode,
1461 "cleanup dec ref error %d", err);
1463 return ERR_PTR(err);
1467 * Xattr inodes are shared therefore quota charging is performed
1468 * at a higher level.
1470 dquot_free_inode(ea_inode);
1471 dquot_drop(ea_inode);
1472 inode_lock(ea_inode);
1473 ea_inode->i_flags |= S_NOQUOTA;
1474 inode_unlock(ea_inode);
1480 static struct inode *
1481 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1482 size_t value_len, u32 hash)
1484 struct inode *ea_inode;
1485 struct mb_cache_entry *ce;
1486 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1489 if (!ea_inode_cache)
1492 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1496 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1497 !(current->flags & PF_MEMALLOC_NOFS));
1499 ea_data = kvmalloc(value_len, GFP_KERNEL);
1501 mb_cache_entry_put(ea_inode_cache, ce);
1506 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1508 if (!IS_ERR(ea_inode) &&
1509 !is_bad_inode(ea_inode) &&
1510 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1511 i_size_read(ea_inode) == value_len &&
1512 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1513 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1515 !memcmp(value, ea_data, value_len)) {
1516 mb_cache_entry_touch(ea_inode_cache, ce);
1517 mb_cache_entry_put(ea_inode_cache, ce);
1522 if (!IS_ERR(ea_inode))
1524 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1531 * Add value of the EA in an inode.
1533 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1534 const void *value, size_t value_len,
1535 struct inode **ret_inode)
1537 struct inode *ea_inode;
1541 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1542 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1544 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1550 *ret_inode = ea_inode;
1554 /* Create an inode for the EA value */
1555 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1556 if (IS_ERR(ea_inode))
1557 return PTR_ERR(ea_inode);
1559 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1561 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1562 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1567 if (EA_INODE_CACHE(inode))
1568 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1569 ea_inode->i_ino, true /* reusable */);
1571 *ret_inode = ea_inode;
1576 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1577 * feature is enabled.
1579 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1581 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1582 struct ext4_xattr_search *s,
1583 handle_t *handle, struct inode *inode,
1586 struct ext4_xattr_entry *last, *next;
1587 struct ext4_xattr_entry *here = s->here;
1588 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1589 int in_inode = i->in_inode;
1590 struct inode *old_ea_inode = NULL;
1591 struct inode *new_ea_inode = NULL;
1592 size_t old_size, new_size;
1595 /* Space used by old and new values. */
1596 old_size = (!s->not_found && !here->e_value_inum) ?
1597 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1598 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1601 * Optimization for the simple case when old and new values have the
1602 * same padded sizes. Not applicable if external inodes are involved.
1604 if (new_size && new_size == old_size) {
1605 size_t offs = le16_to_cpu(here->e_value_offs);
1606 void *val = s->base + offs;
1608 here->e_value_size = cpu_to_le32(i->value_len);
1609 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1610 memset(val, 0, new_size);
1612 memcpy(val, i->value, i->value_len);
1613 /* Clear padding bytes. */
1614 memset(val + i->value_len, 0, new_size - i->value_len);
1619 /* Compute min_offs and last. */
1621 for (; !IS_LAST_ENTRY(last); last = next) {
1622 next = EXT4_XATTR_NEXT(last);
1623 if ((void *)next >= s->end) {
1624 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1625 ret = -EFSCORRUPTED;
1628 if (!last->e_value_inum && last->e_value_size) {
1629 size_t offs = le16_to_cpu(last->e_value_offs);
1630 if (offs < min_offs)
1635 /* Check whether we have enough space. */
1639 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1641 free += EXT4_XATTR_LEN(name_len) + old_size;
1643 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1649 * If storing the value in an external inode is an option,
1650 * reserve space for xattr entries/names in the external
1651 * attribute block so that a long value does not occupy the
1652 * whole space and prevent further entries being added.
1654 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1655 new_size && is_block &&
1656 (min_offs + old_size - new_size) <
1657 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1664 * Getting access to old and new ea inodes is subject to failures.
1665 * Finish that work before doing any modifications to the xattr data.
1667 if (!s->not_found && here->e_value_inum) {
1668 ret = ext4_xattr_inode_iget(inode,
1669 le32_to_cpu(here->e_value_inum),
1670 le32_to_cpu(here->e_hash),
1673 old_ea_inode = NULL;
1677 if (i->value && in_inode) {
1678 WARN_ON_ONCE(!i->value_len);
1680 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1684 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1688 new_ea_inode = NULL;
1689 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1695 /* We are ready to release ref count on the old_ea_inode. */
1696 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1698 /* Release newly required ref count on new_ea_inode. */
1702 err = ext4_xattr_inode_dec_ref(handle,
1705 ext4_warning_inode(new_ea_inode,
1706 "dec ref new_ea_inode err=%d",
1708 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1714 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1715 le32_to_cpu(here->e_value_size));
1718 /* No failures allowed past this point. */
1720 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1721 /* Remove the old value. */
1722 void *first_val = s->base + min_offs;
1723 size_t offs = le16_to_cpu(here->e_value_offs);
1724 void *val = s->base + offs;
1726 memmove(first_val + old_size, first_val, val - first_val);
1727 memset(first_val, 0, old_size);
1728 min_offs += old_size;
1730 /* Adjust all value offsets. */
1732 while (!IS_LAST_ENTRY(last)) {
1733 size_t o = le16_to_cpu(last->e_value_offs);
1735 if (!last->e_value_inum &&
1736 last->e_value_size && o < offs)
1737 last->e_value_offs = cpu_to_le16(o + old_size);
1738 last = EXT4_XATTR_NEXT(last);
1743 /* Remove old name. */
1744 size_t size = EXT4_XATTR_LEN(name_len);
1746 last = ENTRY((void *)last - size);
1747 memmove(here, (void *)here + size,
1748 (void *)last - (void *)here + sizeof(__u32));
1749 memset(last, 0, size);
1750 } else if (s->not_found) {
1751 /* Insert new name. */
1752 size_t size = EXT4_XATTR_LEN(name_len);
1753 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1755 memmove((void *)here + size, here, rest);
1756 memset(here, 0, size);
1757 here->e_name_index = i->name_index;
1758 here->e_name_len = name_len;
1759 memcpy(here->e_name, i->name, name_len);
1761 /* This is an update, reset value info. */
1762 here->e_value_inum = 0;
1763 here->e_value_offs = 0;
1764 here->e_value_size = 0;
1768 /* Insert new value. */
1770 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1771 } else if (i->value_len) {
1772 void *val = s->base + min_offs - new_size;
1774 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1775 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1776 memset(val, 0, new_size);
1778 memcpy(val, i->value, i->value_len);
1779 /* Clear padding bytes. */
1780 memset(val + i->value_len, 0,
1781 new_size - i->value_len);
1784 here->e_value_size = cpu_to_le32(i->value_len);
1791 /* Entry hash calculation. */
1796 * Feed crc32c hash instead of the raw value for entry
1797 * hash calculation. This is to avoid walking
1798 * potentially long value buffer again.
1800 crc32c_hash = cpu_to_le32(
1801 ext4_xattr_inode_get_hash(new_ea_inode));
1802 hash = ext4_xattr_hash_entry(here->e_name,
1805 } else if (is_block) {
1806 __le32 *value = s->base + le16_to_cpu(
1807 here->e_value_offs);
1809 hash = ext4_xattr_hash_entry(here->e_name,
1810 here->e_name_len, value,
1813 here->e_hash = hash;
1817 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1826 struct ext4_xattr_block_find {
1827 struct ext4_xattr_search s;
1828 struct buffer_head *bh;
1832 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1833 struct ext4_xattr_block_find *bs)
1835 struct super_block *sb = inode->i_sb;
1838 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1839 i->name_index, i->name, i->value, (long)i->value_len);
1841 if (EXT4_I(inode)->i_file_acl) {
1842 /* The inode already has an extended attribute block. */
1843 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1844 if (IS_ERR(bs->bh)) {
1845 error = PTR_ERR(bs->bh);
1849 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1850 atomic_read(&(bs->bh->b_count)),
1851 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1852 error = ext4_xattr_check_block(inode, bs->bh);
1855 /* Find the named attribute. */
1856 bs->s.base = BHDR(bs->bh);
1857 bs->s.first = BFIRST(bs->bh);
1858 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1859 bs->s.here = bs->s.first;
1860 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1861 i->name_index, i->name, 1);
1862 if (error && error != -ENODATA)
1864 bs->s.not_found = error;
1870 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1871 struct ext4_xattr_info *i,
1872 struct ext4_xattr_block_find *bs)
1874 struct super_block *sb = inode->i_sb;
1875 struct buffer_head *new_bh = NULL;
1876 struct ext4_xattr_search s_copy = bs->s;
1877 struct ext4_xattr_search *s = &s_copy;
1878 struct mb_cache_entry *ce = NULL;
1880 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1881 struct inode *ea_inode = NULL, *tmp_inode;
1882 size_t old_ea_inode_quota = 0;
1883 unsigned int ea_ino;
1886 #define header(x) ((struct ext4_xattr_header *)(x))
1889 int offset = (char *)s->here - bs->bh->b_data;
1891 BUFFER_TRACE(bs->bh, "get_write_access");
1892 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1896 lock_buffer(bs->bh);
1898 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1899 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1902 * This must happen under buffer lock for
1903 * ext4_xattr_block_set() to reliably detect modified
1906 if (ea_block_cache) {
1907 struct mb_cache_entry *oe;
1909 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1910 hash, bs->bh->b_blocknr);
1913 * Xattr block is getting reused. Leave
1916 mb_cache_entry_put(ea_block_cache, oe);
1920 ea_bdebug(bs->bh, "modifying in-place");
1921 error = ext4_xattr_set_entry(i, s, handle, inode,
1922 true /* is_block */);
1923 ext4_xattr_block_csum_set(inode, bs->bh);
1924 unlock_buffer(bs->bh);
1925 if (error == -EFSCORRUPTED)
1928 error = ext4_handle_dirty_metadata(handle,
1936 unlock_buffer(bs->bh);
1937 ea_bdebug(bs->bh, "cloning");
1938 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1940 if (s->base == NULL)
1942 s->first = ENTRY(header(s->base)+1);
1943 header(s->base)->h_refcount = cpu_to_le32(1);
1944 s->here = ENTRY(s->base + offset);
1945 s->end = s->base + bs->bh->b_size;
1948 * If existing entry points to an xattr inode, we need
1949 * to prevent ext4_xattr_set_entry() from decrementing
1950 * ref count on it because the reference belongs to the
1951 * original block. In this case, make the entry look
1952 * like it has an empty value.
1954 if (!s->not_found && s->here->e_value_inum) {
1955 ea_ino = le32_to_cpu(s->here->e_value_inum);
1956 error = ext4_xattr_inode_iget(inode, ea_ino,
1957 le32_to_cpu(s->here->e_hash),
1962 if (!ext4_test_inode_state(tmp_inode,
1963 EXT4_STATE_LUSTRE_EA_INODE)) {
1965 * Defer quota free call for previous
1966 * inode until success is guaranteed.
1968 old_ea_inode_quota = le32_to_cpu(
1969 s->here->e_value_size);
1973 s->here->e_value_inum = 0;
1974 s->here->e_value_size = 0;
1977 /* Allocate a buffer where we construct the new block. */
1978 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1980 if (s->base == NULL)
1982 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1983 header(s->base)->h_blocks = cpu_to_le32(1);
1984 header(s->base)->h_refcount = cpu_to_le32(1);
1985 s->first = ENTRY(header(s->base)+1);
1986 s->here = ENTRY(header(s->base)+1);
1987 s->end = s->base + sb->s_blocksize;
1990 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1991 if (error == -EFSCORRUPTED)
1996 if (i->value && s->here->e_value_inum) {
1998 * A ref count on ea_inode has been taken as part of the call to
1999 * ext4_xattr_set_entry() above. We would like to drop this
2000 * extra ref but we have to wait until the xattr block is
2001 * initialized and has its own ref count on the ea_inode.
2003 ea_ino = le32_to_cpu(s->here->e_value_inum);
2004 error = ext4_xattr_inode_iget(inode, ea_ino,
2005 le32_to_cpu(s->here->e_hash),
2014 if (!IS_LAST_ENTRY(s->first)) {
2015 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2018 /* We found an identical block in the cache. */
2019 if (new_bh == bs->bh)
2020 ea_bdebug(new_bh, "keeping");
2024 WARN_ON_ONCE(dquot_initialize_needed(inode));
2026 /* The old block is released after updating
2028 error = dquot_alloc_block(inode,
2029 EXT4_C2B(EXT4_SB(sb), 1));
2032 BUFFER_TRACE(new_bh, "get_write_access");
2033 error = ext4_journal_get_write_access(
2038 lock_buffer(new_bh);
2040 * We have to be careful about races with
2041 * adding references to xattr block. Once we
2042 * hold buffer lock xattr block's state is
2043 * stable so we can check the additional
2046 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2047 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2049 * Undo everything and check mbcache
2052 unlock_buffer(new_bh);
2053 dquot_free_block(inode,
2054 EXT4_C2B(EXT4_SB(sb),
2057 mb_cache_entry_put(ea_block_cache, ce);
2062 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2063 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2064 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2065 ea_bdebug(new_bh, "reusing; refcount now=%d",
2067 ext4_xattr_block_csum_set(inode, new_bh);
2068 unlock_buffer(new_bh);
2069 error = ext4_handle_dirty_metadata(handle,
2075 mb_cache_entry_touch(ea_block_cache, ce);
2076 mb_cache_entry_put(ea_block_cache, ce);
2078 } else if (bs->bh && s->base == bs->bh->b_data) {
2079 /* We were modifying this block in-place. */
2080 ea_bdebug(bs->bh, "keeping this block");
2081 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2085 /* We need to allocate a new block */
2086 ext4_fsblk_t goal, block;
2088 WARN_ON_ONCE(dquot_initialize_needed(inode));
2090 goal = ext4_group_first_block_no(sb,
2091 EXT4_I(inode)->i_block_group);
2092 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2097 ea_idebug(inode, "creating block %llu",
2098 (unsigned long long)block);
2100 new_bh = sb_getblk(sb, block);
2101 if (unlikely(!new_bh)) {
2104 ext4_free_blocks(handle, inode, NULL, block, 1,
2105 EXT4_FREE_BLOCKS_METADATA);
2108 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2109 ENTRY(header(s->base)+1));
2113 /* Drop the extra ref on ea_inode. */
2114 error = ext4_xattr_inode_dec_ref(handle,
2117 ext4_warning_inode(ea_inode,
2124 lock_buffer(new_bh);
2125 error = ext4_journal_get_create_access(handle, sb,
2126 new_bh, EXT4_JTR_NONE);
2128 unlock_buffer(new_bh);
2132 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2133 ext4_xattr_block_csum_set(inode, new_bh);
2134 set_buffer_uptodate(new_bh);
2135 unlock_buffer(new_bh);
2136 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2137 error = ext4_handle_dirty_metadata(handle, inode,
2144 if (old_ea_inode_quota)
2145 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2147 /* Update the inode. */
2148 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2150 /* Drop the previous xattr block. */
2151 if (bs->bh && bs->bh != new_bh) {
2152 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2154 ext4_xattr_release_block(handle, inode, bs->bh,
2156 0 /* extra_credits */);
2157 ext4_xattr_inode_array_free(ea_inode_array);
2165 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2167 ext4_warning_inode(ea_inode, "dec ref error=%d",
2170 /* If there was an error, revert the quota charge. */
2172 ext4_xattr_inode_free_quota(inode, ea_inode,
2173 i_size_read(ea_inode));
2177 mb_cache_entry_put(ea_block_cache, ce);
2179 if (!(bs->bh && s->base == bs->bh->b_data))
2185 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2189 EXT4_ERROR_INODE(inode, "bad block %llu",
2190 EXT4_I(inode)->i_file_acl);
2196 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2197 struct ext4_xattr_ibody_find *is)
2199 struct ext4_xattr_ibody_header *header;
2200 struct ext4_inode *raw_inode;
2203 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2206 raw_inode = ext4_raw_inode(&is->iloc);
2207 header = IHDR(inode, raw_inode);
2208 is->s.base = is->s.first = IFIRST(header);
2209 is->s.here = is->s.first;
2210 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2211 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2212 error = xattr_check_inode(inode, header, is->s.end);
2215 /* Find the named attribute. */
2216 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2217 i->name_index, i->name, 0);
2218 if (error && error != -ENODATA)
2220 is->s.not_found = error;
2225 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2226 struct ext4_xattr_info *i,
2227 struct ext4_xattr_ibody_find *is)
2229 struct ext4_xattr_ibody_header *header;
2230 struct ext4_xattr_search *s = &is->s;
2233 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2236 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2239 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2240 if (!IS_LAST_ENTRY(s->first)) {
2241 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2242 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2244 header->h_magic = cpu_to_le32(0);
2245 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2250 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2251 struct ext4_xattr_info *i)
2255 /* When e_value_inum is set the value is stored externally. */
2256 if (s->here->e_value_inum)
2258 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2260 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2261 return !memcmp(value, i->value, i->value_len);
2264 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2266 struct buffer_head *bh;
2269 if (!EXT4_I(inode)->i_file_acl)
2271 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2274 error = ext4_xattr_check_block(inode, bh);
2277 return ERR_PTR(error);
2283 * ext4_xattr_set_handle()
2285 * Create, replace or remove an extended attribute for this inode. Value
2286 * is NULL to remove an existing extended attribute, and non-NULL to
2287 * either replace an existing extended attribute, or create a new extended
2288 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2289 * specify that an extended attribute must exist and must not exist
2290 * previous to the call, respectively.
2292 * Returns 0, or a negative error number on failure.
2295 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2296 const char *name, const void *value, size_t value_len,
2299 struct ext4_xattr_info i = {
2300 .name_index = name_index,
2303 .value_len = value_len,
2306 struct ext4_xattr_ibody_find is = {
2307 .s = { .not_found = -ENODATA, },
2309 struct ext4_xattr_block_find bs = {
2310 .s = { .not_found = -ENODATA, },
2317 if (strlen(name) > 255)
2320 ext4_write_lock_xattr(inode, &no_expand);
2322 /* Check journal credits under write lock. */
2323 if (ext4_handle_valid(handle)) {
2324 struct buffer_head *bh;
2327 bh = ext4_xattr_get_block(inode);
2329 error = PTR_ERR(bh);
2333 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2335 flags & XATTR_CREATE);
2338 if (jbd2_handle_buffer_credits(handle) < credits) {
2342 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
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) &&
2416 i.value_len && !i.in_inode) {
2424 ext4_xattr_update_super_block(handle, inode->i_sb);
2425 inode->i_ctime = current_time(inode);
2426 inode_inc_iversion(inode);
2429 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2431 * The bh is consumed by ext4_mark_iloc_dirty, even with
2436 ext4_handle_sync(handle);
2438 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2443 ext4_write_unlock_xattr(inode, &no_expand);
2447 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2448 bool is_create, int *credits)
2450 struct buffer_head *bh;
2455 if (!EXT4_SB(inode->i_sb)->s_journal)
2458 down_read(&EXT4_I(inode)->xattr_sem);
2460 bh = ext4_xattr_get_block(inode);
2464 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2465 value_len, is_create);
2470 up_read(&EXT4_I(inode)->xattr_sem);
2477 * Like ext4_xattr_set_handle, but start from an inode. This extended
2478 * attribute modification is a filesystem transaction by itself.
2480 * Returns 0, or a negative error number on failure.
2483 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2484 const void *value, size_t value_len, int flags)
2487 struct super_block *sb = inode->i_sb;
2488 int error, retries = 0;
2491 error = dquot_initialize(inode);
2496 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2501 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2502 if (IS_ERR(handle)) {
2503 error = PTR_ERR(handle);
2507 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2508 value, value_len, flags);
2509 error2 = ext4_journal_stop(handle);
2510 if (error == -ENOSPC &&
2511 ext4_should_retry_alloc(sb, &retries))
2516 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2522 * Shift the EA entries in the inode to create space for the increased
2525 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2526 int value_offs_shift, void *to,
2527 void *from, size_t n)
2529 struct ext4_xattr_entry *last = entry;
2532 /* We always shift xattr headers further thus offsets get lower */
2533 BUG_ON(value_offs_shift > 0);
2535 /* Adjust the value offsets of the entries */
2536 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2537 if (!last->e_value_inum && last->e_value_size) {
2538 new_offs = le16_to_cpu(last->e_value_offs) +
2540 last->e_value_offs = cpu_to_le16(new_offs);
2543 /* Shift the entries by n bytes */
2544 memmove(to, from, n);
2548 * Move xattr pointed to by 'entry' from inode into external xattr block
2550 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2551 struct ext4_inode *raw_inode,
2552 struct ext4_xattr_entry *entry)
2554 struct ext4_xattr_ibody_find *is = NULL;
2555 struct ext4_xattr_block_find *bs = NULL;
2556 char *buffer = NULL, *b_entry_name = NULL;
2557 size_t value_size = le32_to_cpu(entry->e_value_size);
2558 struct ext4_xattr_info i = {
2561 .name_index = entry->e_name_index,
2562 .in_inode = !!entry->e_value_inum,
2564 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2567 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2568 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2569 buffer = kvmalloc(value_size, GFP_NOFS);
2570 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2571 if (!is || !bs || !buffer || !b_entry_name) {
2576 is->s.not_found = -ENODATA;
2577 bs->s.not_found = -ENODATA;
2581 /* Save the entry name and the entry value */
2582 if (entry->e_value_inum) {
2583 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2587 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2588 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2591 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2592 b_entry_name[entry->e_name_len] = '\0';
2593 i.name = b_entry_name;
2595 error = ext4_get_inode_loc(inode, &is->iloc);
2599 error = ext4_xattr_ibody_find(inode, &i, is);
2603 /* Remove the chosen entry from the inode */
2604 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2609 i.value_len = value_size;
2610 error = ext4_xattr_block_find(inode, &i, bs);
2614 /* Add entry which was removed from the inode into the block */
2615 error = ext4_xattr_block_set(handle, inode, &i, bs);
2620 kfree(b_entry_name);
2623 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 /* never move system.data out of the inode */
2654 if ((last->e_name_len == 4) &&
2655 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2656 !memcmp(last->e_name, "data", 4))
2658 total_size = EXT4_XATTR_LEN(last->e_name_len);
2659 if (!last->e_value_inum)
2660 total_size += EXT4_XATTR_SIZE(
2661 le32_to_cpu(last->e_value_size));
2662 if (total_size <= bfree &&
2663 total_size < min_total_size) {
2664 if (total_size + ifree < isize_diff) {
2668 min_total_size = total_size;
2673 if (entry == NULL) {
2674 if (small_entry == NULL)
2676 entry = small_entry;
2679 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2680 total_size = entry_size;
2681 if (!entry->e_value_inum)
2682 total_size += EXT4_XATTR_SIZE(
2683 le32_to_cpu(entry->e_value_size));
2684 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2689 *total_ino -= entry_size;
2690 ifree += total_size;
2691 bfree -= total_size;
2698 * Expand an inode by new_extra_isize bytes when EAs are present.
2699 * Returns 0 on success or negative error number on failure.
2701 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2702 struct ext4_inode *raw_inode, handle_t *handle)
2704 struct ext4_xattr_ibody_header *header;
2705 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2706 static unsigned int mnt_count;
2708 size_t ifree, bfree;
2711 int error = 0, tried_min_extra_isize = 0;
2712 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2713 int isize_diff; /* How much do we need to grow i_extra_isize */
2716 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2717 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2720 header = IHDR(inode, raw_inode);
2723 * Check if enough free space is available in the inode to shift the
2724 * entries ahead by new_extra_isize.
2727 base = IFIRST(header);
2728 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2729 min_offs = end - base;
2730 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2732 error = xattr_check_inode(inode, header, end);
2736 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2737 if (ifree >= isize_diff)
2741 * Enough free space isn't available in the inode, check if
2742 * EA block can hold new_extra_isize bytes.
2744 if (EXT4_I(inode)->i_file_acl) {
2745 struct buffer_head *bh;
2747 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2749 error = PTR_ERR(bh);
2752 error = ext4_xattr_check_block(inode, bh);
2758 end = bh->b_data + bh->b_size;
2759 min_offs = end - base;
2760 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2763 if (bfree + ifree < isize_diff) {
2764 if (!tried_min_extra_isize && s_min_extra_isize) {
2765 tried_min_extra_isize++;
2766 new_extra_isize = s_min_extra_isize;
2773 bfree = inode->i_sb->s_blocksize;
2776 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2777 isize_diff, ifree, bfree,
2780 if (error == -ENOSPC && !tried_min_extra_isize &&
2781 s_min_extra_isize) {
2782 tried_min_extra_isize++;
2783 new_extra_isize = s_min_extra_isize;
2789 /* Adjust the offsets and shift the remaining entries ahead */
2790 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2791 - new_extra_isize, (void *)raw_inode +
2792 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2793 (void *)header, total_ino);
2794 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2797 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2798 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2800 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2805 #define EIA_INCR 16 /* must be 2^n */
2806 #define EIA_MASK (EIA_INCR - 1)
2808 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2809 * If @ea_inode_array is new or full it will be grown and the old
2810 * contents copied over.
2813 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2814 struct inode *inode)
2816 if (*ea_inode_array == NULL) {
2818 * Start with 15 inodes, so it fits into a power-of-two size.
2819 * If *ea_inode_array is NULL, this is essentially offsetof()
2822 kmalloc(offsetof(struct ext4_xattr_inode_array,
2825 if (*ea_inode_array == NULL)
2827 (*ea_inode_array)->count = 0;
2828 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2829 /* expand the array once all 15 + n * 16 slots are full */
2830 struct ext4_xattr_inode_array *new_array = NULL;
2831 int count = (*ea_inode_array)->count;
2833 /* if new_array is NULL, this is essentially offsetof() */
2834 new_array = kmalloc(
2835 offsetof(struct ext4_xattr_inode_array,
2836 inodes[count + EIA_INCR]),
2838 if (new_array == NULL)
2840 memcpy(new_array, *ea_inode_array,
2841 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2842 kfree(*ea_inode_array);
2843 *ea_inode_array = new_array;
2845 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2850 * ext4_xattr_delete_inode()
2852 * Free extended attribute resources associated with this inode. Traverse
2853 * all entries and decrement reference on any xattr inodes associated with this
2854 * inode. This is called immediately before an inode is freed. We have exclusive
2855 * access to the inode. If an orphan inode is deleted it will also release its
2856 * references on xattr block and xattr inodes.
2858 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2859 struct ext4_xattr_inode_array **ea_inode_array,
2862 struct buffer_head *bh = NULL;
2863 struct ext4_xattr_ibody_header *header;
2864 struct ext4_iloc iloc = { .bh = NULL };
2865 struct ext4_xattr_entry *entry;
2866 struct inode *ea_inode;
2869 error = ext4_journal_ensure_credits(handle, extra_credits,
2870 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2872 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2876 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2877 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2879 error = ext4_get_inode_loc(inode, &iloc);
2881 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2885 error = ext4_journal_get_write_access(handle, inode->i_sb,
2886 iloc.bh, EXT4_JTR_NONE);
2888 EXT4_ERROR_INODE(inode, "write access (error %d)",
2893 header = IHDR(inode, ext4_raw_inode(&iloc));
2894 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2895 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2897 false /* block_csum */,
2900 false /* skip_quota */);
2903 if (EXT4_I(inode)->i_file_acl) {
2904 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2906 error = PTR_ERR(bh);
2907 if (error == -EIO) {
2908 EXT4_ERROR_INODE_ERR(inode, EIO,
2909 "block %llu read error",
2910 EXT4_I(inode)->i_file_acl);
2915 error = ext4_xattr_check_block(inode, bh);
2919 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2920 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2921 entry = EXT4_XATTR_NEXT(entry)) {
2922 if (!entry->e_value_inum)
2924 error = ext4_xattr_inode_iget(inode,
2925 le32_to_cpu(entry->e_value_inum),
2926 le32_to_cpu(entry->e_hash),
2930 ext4_xattr_inode_free_quota(inode, ea_inode,
2931 le32_to_cpu(entry->e_value_size));
2937 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2940 * Update i_file_acl value in the same transaction that releases
2943 EXT4_I(inode)->i_file_acl = 0;
2944 error = ext4_mark_inode_dirty(handle, inode);
2946 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2950 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2959 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2963 if (ea_inode_array == NULL)
2966 for (idx = 0; idx < ea_inode_array->count; ++idx)
2967 iput(ea_inode_array->inodes[idx]);
2968 kfree(ea_inode_array);
2972 * ext4_xattr_block_cache_insert()
2974 * Create a new entry in the extended attribute block cache, and insert
2975 * it unless such an entry is already in the cache.
2977 * Returns 0, or a negative error number on failure.
2980 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2981 struct buffer_head *bh)
2983 struct ext4_xattr_header *header = BHDR(bh);
2984 __u32 hash = le32_to_cpu(header->h_hash);
2985 int reusable = le32_to_cpu(header->h_refcount) <
2986 EXT4_XATTR_REFCOUNT_MAX;
2989 if (!ea_block_cache)
2991 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2992 bh->b_blocknr, reusable);
2994 if (error == -EBUSY)
2995 ea_bdebug(bh, "already in cache");
2997 ea_bdebug(bh, "inserting [%x]", (int)hash);
3003 * Compare two extended attribute blocks for equality.
3005 * Returns 0 if the blocks are equal, 1 if they differ, and
3006 * a negative error number on errors.
3009 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3010 struct ext4_xattr_header *header2)
3012 struct ext4_xattr_entry *entry1, *entry2;
3014 entry1 = ENTRY(header1+1);
3015 entry2 = ENTRY(header2+1);
3016 while (!IS_LAST_ENTRY(entry1)) {
3017 if (IS_LAST_ENTRY(entry2))
3019 if (entry1->e_hash != entry2->e_hash ||
3020 entry1->e_name_index != entry2->e_name_index ||
3021 entry1->e_name_len != entry2->e_name_len ||
3022 entry1->e_value_size != entry2->e_value_size ||
3023 entry1->e_value_inum != entry2->e_value_inum ||
3024 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3026 if (!entry1->e_value_inum &&
3027 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3028 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3029 le32_to_cpu(entry1->e_value_size)))
3032 entry1 = EXT4_XATTR_NEXT(entry1);
3033 entry2 = EXT4_XATTR_NEXT(entry2);
3035 if (!IS_LAST_ENTRY(entry2))
3041 * ext4_xattr_block_cache_find()
3043 * Find an identical extended attribute block.
3045 * Returns a pointer to the block found, or NULL if such a block was
3046 * not found or an error occurred.
3048 static struct buffer_head *
3049 ext4_xattr_block_cache_find(struct inode *inode,
3050 struct ext4_xattr_header *header,
3051 struct mb_cache_entry **pce)
3053 __u32 hash = le32_to_cpu(header->h_hash);
3054 struct mb_cache_entry *ce;
3055 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3057 if (!ea_block_cache)
3059 if (!header->h_hash)
3060 return NULL; /* never share */
3061 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3062 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3064 struct buffer_head *bh;
3066 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3068 if (PTR_ERR(bh) == -ENOMEM)
3071 EXT4_ERROR_INODE(inode, "block %lu read error",
3072 (unsigned long)ce->e_value);
3073 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3078 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3083 #define NAME_HASH_SHIFT 5
3084 #define VALUE_HASH_SHIFT 16
3087 * ext4_xattr_hash_entry()
3089 * Compute the hash of an extended attribute.
3091 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3096 while (name_len--) {
3097 hash = (hash << NAME_HASH_SHIFT) ^
3098 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3101 while (value_count--) {
3102 hash = (hash << VALUE_HASH_SHIFT) ^
3103 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3104 le32_to_cpu(*value++);
3106 return cpu_to_le32(hash);
3110 * ext4_xattr_hash_entry_signed()
3112 * Compute the hash of an extended attribute incorrectly.
3114 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3118 while (name_len--) {
3119 hash = (hash << NAME_HASH_SHIFT) ^
3120 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3121 (signed char)*name++;
3123 while (value_count--) {
3124 hash = (hash << VALUE_HASH_SHIFT) ^
3125 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3126 le32_to_cpu(*value++);
3128 return cpu_to_le32(hash);
3131 #undef NAME_HASH_SHIFT
3132 #undef VALUE_HASH_SHIFT
3134 #define BLOCK_HASH_SHIFT 16
3137 * ext4_xattr_rehash()
3139 * Re-compute the extended attribute hash value after an entry has changed.
3141 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3143 struct ext4_xattr_entry *here;
3146 here = ENTRY(header+1);
3147 while (!IS_LAST_ENTRY(here)) {
3148 if (!here->e_hash) {
3149 /* Block is not shared if an entry's hash value == 0 */
3153 hash = (hash << BLOCK_HASH_SHIFT) ^
3154 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3155 le32_to_cpu(here->e_hash);
3156 here = EXT4_XATTR_NEXT(here);
3158 header->h_hash = cpu_to_le32(hash);
3161 #undef BLOCK_HASH_SHIFT
3163 #define HASH_BUCKET_BITS 10
3166 ext4_xattr_create_cache(void)
3168 return mb_cache_create(HASH_BUCKET_BITS);
3171 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3174 mb_cache_destroy(cache);