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 void ext4_xattr_rehash(struct ext4_xattr_header *);
86 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
87 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
88 #ifdef CONFIG_EXT4_FS_POSIX_ACL
89 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
90 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
92 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
93 #ifdef CONFIG_EXT4_FS_SECURITY
94 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
98 const struct xattr_handler *ext4_xattr_handlers[] = {
99 &ext4_xattr_user_handler,
100 &ext4_xattr_trusted_handler,
101 #ifdef CONFIG_EXT4_FS_POSIX_ACL
102 &posix_acl_access_xattr_handler,
103 &posix_acl_default_xattr_handler,
105 #ifdef CONFIG_EXT4_FS_SECURITY
106 &ext4_xattr_security_handler,
111 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
112 inode->i_sb->s_fs_info)->s_ea_block_cache)
114 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
115 inode->i_sb->s_fs_info)->s_ea_inode_cache)
118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119 struct inode *inode);
121 #ifdef CONFIG_LOCKDEP
122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
124 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
128 static __le32 ext4_xattr_block_csum(struct inode *inode,
130 struct ext4_xattr_header *hdr)
132 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
134 __le64 dsk_block_nr = cpu_to_le64(block_nr);
135 __u32 dummy_csum = 0;
136 int offset = offsetof(struct ext4_xattr_header, h_checksum);
138 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
139 sizeof(dsk_block_nr));
140 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
141 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
142 offset += sizeof(dummy_csum);
143 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
144 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
146 return cpu_to_le32(csum);
149 static int ext4_xattr_block_csum_verify(struct inode *inode,
150 struct buffer_head *bh)
152 struct ext4_xattr_header *hdr = BHDR(bh);
155 if (ext4_has_metadata_csum(inode->i_sb)) {
157 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
158 bh->b_blocknr, hdr));
164 static void ext4_xattr_block_csum_set(struct inode *inode,
165 struct buffer_head *bh)
167 if (ext4_has_metadata_csum(inode->i_sb))
168 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
169 bh->b_blocknr, BHDR(bh));
172 static inline const struct xattr_handler *
173 ext4_xattr_handler(int name_index)
175 const struct xattr_handler *handler = NULL;
177 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
178 handler = ext4_xattr_handler_map[name_index];
183 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
186 struct ext4_xattr_entry *e = entry;
188 /* Find the end of the names list */
189 while (!IS_LAST_ENTRY(e)) {
190 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
191 if ((void *)next >= end)
192 return -EFSCORRUPTED;
193 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
194 return -EFSCORRUPTED;
198 /* Check the values */
199 while (!IS_LAST_ENTRY(entry)) {
200 u32 size = le32_to_cpu(entry->e_value_size);
202 if (size > EXT4_XATTR_SIZE_MAX)
203 return -EFSCORRUPTED;
205 if (size != 0 && entry->e_value_inum == 0) {
206 u16 offs = le16_to_cpu(entry->e_value_offs);
210 * The value cannot overlap the names, and the value
211 * with padding cannot extend beyond 'end'. Check both
212 * the padded and unpadded sizes, since the size may
213 * overflow to 0 when adding padding.
215 if (offs > end - value_start)
216 return -EFSCORRUPTED;
217 value = value_start + offs;
218 if (value < (void *)e + sizeof(u32) ||
219 size > end - value ||
220 EXT4_XATTR_SIZE(size) > end - value)
221 return -EFSCORRUPTED;
223 entry = EXT4_XATTR_NEXT(entry);
230 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
231 const char *function, unsigned int line)
233 int error = -EFSCORRUPTED;
235 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
236 BHDR(bh)->h_blocks != cpu_to_le32(1))
238 if (buffer_verified(bh))
242 if (!ext4_xattr_block_csum_verify(inode, bh))
244 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
248 __ext4_error_inode(inode, function, line, 0,
249 "corrupted xattr block %llu",
250 (unsigned long long) bh->b_blocknr);
252 set_buffer_verified(bh);
256 #define ext4_xattr_check_block(inode, bh) \
257 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
261 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
262 void *end, const char *function, unsigned int line)
264 int error = -EFSCORRUPTED;
266 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
267 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
269 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
272 __ext4_error_inode(inode, function, line, 0,
273 "corrupted in-inode xattr");
277 #define xattr_check_inode(inode, header, end) \
278 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
281 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
282 void *end, int name_index, const char *name, int sorted)
284 struct ext4_xattr_entry *entry, *next;
290 name_len = strlen(name);
291 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
292 next = EXT4_XATTR_NEXT(entry);
293 if ((void *) next >= end) {
294 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
295 return -EFSCORRUPTED;
297 cmp = name_index - entry->e_name_index;
299 cmp = name_len - entry->e_name_len;
301 cmp = memcmp(name, entry->e_name, name_len);
302 if (cmp <= 0 && (sorted || cmp == 0))
306 return cmp ? -ENODATA : 0;
310 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
312 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
315 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
317 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
318 (u32) inode_peek_iversion_raw(ea_inode);
321 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
323 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
324 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
327 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
329 return (u32)ea_inode->i_atime.tv_sec;
332 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
334 ea_inode->i_atime.tv_sec = hash;
338 * Read the EA value from an inode.
340 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
342 int blocksize = 1 << ea_inode->i_blkbits;
343 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
344 int tail_size = (size % blocksize) ?: blocksize;
345 struct buffer_head *bhs_inline[8];
346 struct buffer_head **bhs = bhs_inline;
349 if (bh_count > ARRAY_SIZE(bhs_inline)) {
350 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
355 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
356 true /* wait */, bhs);
360 for (i = 0; i < bh_count; i++) {
361 /* There shouldn't be any holes in ea_inode. */
366 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
367 i < bh_count - 1 ? blocksize : tail_size);
371 for (i = 0; i < bh_count; i++)
374 if (bhs != bhs_inline)
379 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
381 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
382 u32 ea_inode_hash, struct inode **ea_inode)
387 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
389 err = PTR_ERR(inode);
390 ext4_error(parent->i_sb,
391 "error while reading EA inode %lu err=%d", ea_ino,
396 if (is_bad_inode(inode)) {
397 ext4_error(parent->i_sb,
398 "error while reading EA inode %lu is_bad_inode",
404 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
405 ext4_error(parent->i_sb,
406 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
412 ext4_xattr_inode_set_class(inode);
415 * Check whether this is an old Lustre-style xattr inode. Lustre
416 * implementation does not have hash validation, rather it has a
417 * backpointer from ea_inode to the parent inode.
419 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
420 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
421 inode->i_generation == parent->i_generation) {
422 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
423 ext4_xattr_inode_set_ref(inode, 1);
426 inode->i_flags |= S_NOQUOTA;
438 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
439 struct ext4_xattr_entry *entry, void *buffer,
444 /* Verify stored hash matches calculated hash. */
445 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
446 if (hash != ext4_xattr_inode_get_hash(ea_inode))
447 return -EFSCORRUPTED;
450 __le32 e_hash, tmp_data;
452 /* Verify entry hash. */
453 tmp_data = cpu_to_le32(hash);
454 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
456 if (e_hash != entry->e_hash)
457 return -EFSCORRUPTED;
463 * Read xattr value from the EA inode.
466 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
467 void *buffer, size_t size)
469 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
470 struct inode *ea_inode;
473 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
474 le32_to_cpu(entry->e_hash), &ea_inode);
480 if (i_size_read(ea_inode) != size) {
481 ext4_warning_inode(ea_inode,
482 "ea_inode file size=%llu entry size=%zu",
483 i_size_read(ea_inode), size);
488 err = ext4_xattr_inode_read(ea_inode, buffer, size);
492 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
493 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
496 ext4_warning_inode(ea_inode,
497 "EA inode hash validation failed");
502 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
503 ext4_xattr_inode_get_hash(ea_inode),
504 ea_inode->i_ino, true /* reusable */);
512 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
513 void *buffer, size_t buffer_size)
515 struct buffer_head *bh = NULL;
516 struct ext4_xattr_entry *entry;
520 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
522 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
523 name_index, name, buffer, (long)buffer_size);
525 if (!EXT4_I(inode)->i_file_acl)
527 ea_idebug(inode, "reading block %llu",
528 (unsigned long long)EXT4_I(inode)->i_file_acl);
529 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
532 ea_bdebug(bh, "b_count=%d, refcount=%d",
533 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
534 error = ext4_xattr_check_block(inode, bh);
537 ext4_xattr_block_cache_insert(ea_block_cache, bh);
539 end = bh->b_data + bh->b_size;
540 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
543 size = le32_to_cpu(entry->e_value_size);
545 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
548 if (size > buffer_size)
550 if (entry->e_value_inum) {
551 error = ext4_xattr_inode_get(inode, entry, buffer,
556 u16 offset = le16_to_cpu(entry->e_value_offs);
557 void *p = bh->b_data + offset;
559 if (unlikely(p + size > end))
561 memcpy(buffer, p, size);
572 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
573 void *buffer, size_t buffer_size)
575 struct ext4_xattr_ibody_header *header;
576 struct ext4_xattr_entry *entry;
577 struct ext4_inode *raw_inode;
578 struct ext4_iloc iloc;
583 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
585 error = ext4_get_inode_loc(inode, &iloc);
588 raw_inode = ext4_raw_inode(&iloc);
589 header = IHDR(inode, raw_inode);
590 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
591 error = xattr_check_inode(inode, header, end);
594 entry = IFIRST(header);
595 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
598 size = le32_to_cpu(entry->e_value_size);
600 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
603 if (size > buffer_size)
605 if (entry->e_value_inum) {
606 error = ext4_xattr_inode_get(inode, entry, buffer,
611 u16 offset = le16_to_cpu(entry->e_value_offs);
612 void *p = (void *)IFIRST(header) + offset;
614 if (unlikely(p + size > end))
616 memcpy(buffer, p, size);
629 * Copy an extended attribute into the buffer
630 * provided, or compute the buffer size required.
631 * Buffer is NULL to compute the size of the buffer required.
633 * Returns a negative error number on failure, or the number of bytes
634 * used / required on success.
637 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
638 void *buffer, size_t buffer_size)
642 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
645 if (strlen(name) > 255)
648 down_read(&EXT4_I(inode)->xattr_sem);
649 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
651 if (error == -ENODATA)
652 error = ext4_xattr_block_get(inode, name_index, name, buffer,
654 up_read(&EXT4_I(inode)->xattr_sem);
659 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
660 char *buffer, size_t buffer_size)
662 size_t rest = buffer_size;
664 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
665 const struct xattr_handler *handler =
666 ext4_xattr_handler(entry->e_name_index);
668 if (handler && (!handler->list || handler->list(dentry))) {
669 const char *prefix = handler->prefix ?: handler->name;
670 size_t prefix_len = strlen(prefix);
671 size_t size = prefix_len + entry->e_name_len + 1;
676 memcpy(buffer, prefix, prefix_len);
677 buffer += prefix_len;
678 memcpy(buffer, entry->e_name, entry->e_name_len);
679 buffer += entry->e_name_len;
685 return buffer_size - rest; /* total size */
689 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
691 struct inode *inode = d_inode(dentry);
692 struct buffer_head *bh = NULL;
695 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
696 buffer, (long)buffer_size);
698 if (!EXT4_I(inode)->i_file_acl)
700 ea_idebug(inode, "reading block %llu",
701 (unsigned long long)EXT4_I(inode)->i_file_acl);
702 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
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,
719 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
721 struct inode *inode = d_inode(dentry);
722 struct ext4_xattr_ibody_header *header;
723 struct ext4_inode *raw_inode;
724 struct ext4_iloc iloc;
728 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
730 error = ext4_get_inode_loc(inode, &iloc);
733 raw_inode = ext4_raw_inode(&iloc);
734 header = IHDR(inode, raw_inode);
735 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
736 error = xattr_check_inode(inode, header, end);
739 error = ext4_xattr_list_entries(dentry, IFIRST(header),
740 buffer, buffer_size);
748 * Inode operation listxattr()
750 * d_inode(dentry)->i_rwsem: don't care
752 * Copy a list of attribute names into the buffer
753 * provided, or compute the buffer size required.
754 * Buffer is NULL to compute the size of the buffer required.
756 * Returns a negative error number on failure, or the number of bytes
757 * used / required on success.
760 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
764 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
765 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
772 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
777 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
782 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
785 static void ext4_xattr_update_super_block(handle_t *handle,
786 struct super_block *sb)
788 if (ext4_has_feature_xattr(sb))
791 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
792 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
793 ext4_set_feature_xattr(sb);
794 ext4_handle_dirty_super(handle, sb);
798 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
800 struct ext4_iloc iloc = { .bh = NULL };
801 struct buffer_head *bh = NULL;
802 struct ext4_inode *raw_inode;
803 struct ext4_xattr_ibody_header *header;
804 struct ext4_xattr_entry *entry;
805 qsize_t ea_inode_refs = 0;
809 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
811 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
812 ret = ext4_get_inode_loc(inode, &iloc);
815 raw_inode = ext4_raw_inode(&iloc);
816 header = IHDR(inode, raw_inode);
817 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
818 ret = xattr_check_inode(inode, header, end);
822 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
823 entry = EXT4_XATTR_NEXT(entry))
824 if (entry->e_value_inum)
828 if (EXT4_I(inode)->i_file_acl) {
829 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
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_inode_update_ref(handle_t *handle, struct inode *ea_inode,
973 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
974 struct ext4_iloc iloc;
979 inode_lock(ea_inode);
981 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
985 ref_count = ext4_xattr_inode_get_ref(ea_inode);
986 ref_count += ref_change;
987 ext4_xattr_inode_set_ref(ea_inode, ref_count);
989 if (ref_change > 0) {
990 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
991 ea_inode->i_ino, ref_count);
993 if (ref_count == 1) {
994 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
995 ea_inode->i_ino, ea_inode->i_nlink);
997 set_nlink(ea_inode, 1);
998 ext4_orphan_del(handle, ea_inode);
1000 if (ea_inode_cache) {
1001 hash = ext4_xattr_inode_get_hash(ea_inode);
1002 mb_cache_entry_create(ea_inode_cache,
1005 true /* reusable */);
1009 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1010 ea_inode->i_ino, ref_count);
1012 if (ref_count == 0) {
1013 WARN_ONCE(ea_inode->i_nlink != 1,
1014 "EA inode %lu i_nlink=%u",
1015 ea_inode->i_ino, ea_inode->i_nlink);
1017 clear_nlink(ea_inode);
1018 ext4_orphan_add(handle, ea_inode);
1020 if (ea_inode_cache) {
1021 hash = ext4_xattr_inode_get_hash(ea_inode);
1022 mb_cache_entry_delete(ea_inode_cache, hash,
1028 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1030 ext4_warning_inode(ea_inode,
1031 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1033 inode_unlock(ea_inode);
1037 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1039 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1042 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1044 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1047 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1048 struct ext4_xattr_entry *first)
1050 struct inode *ea_inode;
1051 struct ext4_xattr_entry *entry;
1052 struct ext4_xattr_entry *failed_entry;
1053 unsigned int ea_ino;
1056 for (entry = first; !IS_LAST_ENTRY(entry);
1057 entry = EXT4_XATTR_NEXT(entry)) {
1058 if (!entry->e_value_inum)
1060 ea_ino = le32_to_cpu(entry->e_value_inum);
1061 err = ext4_xattr_inode_iget(parent, ea_ino,
1062 le32_to_cpu(entry->e_hash),
1066 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1068 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1078 failed_entry = entry;
1080 for (entry = first; entry != failed_entry;
1081 entry = EXT4_XATTR_NEXT(entry)) {
1082 if (!entry->e_value_inum)
1084 ea_ino = le32_to_cpu(entry->e_value_inum);
1085 err = ext4_xattr_inode_iget(parent, ea_ino,
1086 le32_to_cpu(entry->e_hash),
1089 ext4_warning(parent->i_sb,
1090 "cleanup ea_ino %u iget error %d", ea_ino,
1094 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1096 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1103 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1104 struct buffer_head *bh, bool block_csum, bool dirty)
1110 ext4_xattr_block_csum_set(inode, bh);
1111 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1113 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1122 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1123 struct buffer_head *bh,
1124 struct ext4_xattr_entry *first, bool block_csum,
1125 struct ext4_xattr_inode_array **ea_inode_array,
1126 int extra_credits, bool skip_quota)
1128 struct inode *ea_inode;
1129 struct ext4_xattr_entry *entry;
1131 unsigned int ea_ino;
1135 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1136 credits = 2 + extra_credits;
1138 for (entry = first; !IS_LAST_ENTRY(entry);
1139 entry = EXT4_XATTR_NEXT(entry)) {
1140 if (!entry->e_value_inum)
1142 ea_ino = le32_to_cpu(entry->e_value_inum);
1143 err = ext4_xattr_inode_iget(parent, ea_ino,
1144 le32_to_cpu(entry->e_hash),
1149 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1151 ext4_warning_inode(ea_inode,
1152 "Expand inode array err=%d", err);
1157 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1158 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1159 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1162 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1167 err = ext4_journal_get_write_access(handle, bh);
1169 ext4_warning_inode(ea_inode,
1170 "Re-get write access err=%d",
1176 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1178 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1184 ext4_xattr_inode_free_quota(parent, ea_inode,
1185 le32_to_cpu(entry->e_value_size));
1188 * Forget about ea_inode within the same transaction that
1189 * decrements the ref count. This avoids duplicate decrements in
1190 * case the rest of the work spills over to subsequent
1193 entry->e_value_inum = 0;
1194 entry->e_value_size = 0;
1201 * Note that we are deliberately skipping csum calculation for
1202 * the final update because we do not expect any journal
1203 * restarts until xattr block is freed.
1206 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1208 ext4_warning_inode(parent,
1209 "handle dirty metadata err=%d", err);
1214 * Release the xattr block BH: If the reference count is > 1, decrement it;
1215 * otherwise free the block.
1218 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1219 struct buffer_head *bh,
1220 struct ext4_xattr_inode_array **ea_inode_array,
1223 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1227 BUFFER_TRACE(bh, "get_write_access");
1228 error = ext4_journal_get_write_access(handle, bh);
1233 hash = le32_to_cpu(BHDR(bh)->h_hash);
1234 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1236 ea_bdebug(bh, "refcount now=0; freeing");
1238 * This must happen under buffer lock for
1239 * ext4_xattr_block_set() to reliably detect freed block
1242 mb_cache_entry_delete(ea_block_cache, hash,
1247 if (ext4_has_feature_ea_inode(inode->i_sb))
1248 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1250 true /* block_csum */,
1253 true /* skip_quota */);
1254 ext4_free_blocks(handle, inode, bh, 0, 1,
1255 EXT4_FREE_BLOCKS_METADATA |
1256 EXT4_FREE_BLOCKS_FORGET);
1259 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1260 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1261 struct mb_cache_entry *ce;
1263 if (ea_block_cache) {
1264 ce = mb_cache_entry_get(ea_block_cache, hash,
1268 mb_cache_entry_put(ea_block_cache, ce);
1273 ext4_xattr_block_csum_set(inode, bh);
1275 * Beware of this ugliness: Releasing of xattr block references
1276 * from different inodes can race and so we have to protect
1277 * from a race where someone else frees the block (and releases
1278 * its journal_head) before we are done dirtying the buffer. In
1279 * nojournal mode this race is harmless and we actually cannot
1280 * call ext4_handle_dirty_metadata() with locked buffer as
1281 * that function can call sync_dirty_buffer() so for that case
1282 * we handle the dirtying after unlocking the buffer.
1284 if (ext4_handle_valid(handle))
1285 error = ext4_handle_dirty_metadata(handle, inode, bh);
1287 if (!ext4_handle_valid(handle))
1288 error = ext4_handle_dirty_metadata(handle, inode, bh);
1290 ext4_handle_sync(handle);
1291 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1292 ea_bdebug(bh, "refcount now=%d; releasing",
1293 le32_to_cpu(BHDR(bh)->h_refcount));
1296 ext4_std_error(inode->i_sb, error);
1301 * Find the available free space for EAs. This also returns the total number of
1302 * bytes used by EA entries.
1304 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1305 size_t *min_offs, void *base, int *total)
1307 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1308 if (!last->e_value_inum && last->e_value_size) {
1309 size_t offs = le16_to_cpu(last->e_value_offs);
1310 if (offs < *min_offs)
1314 *total += EXT4_XATTR_LEN(last->e_name_len);
1316 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1320 * Write the value of the EA in an inode.
1322 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1323 const void *buf, int bufsize)
1325 struct buffer_head *bh = NULL;
1326 unsigned long block = 0;
1327 int blocksize = ea_inode->i_sb->s_blocksize;
1328 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1329 int csize, wsize = 0;
1334 while (ret >= 0 && ret < max_blocks) {
1335 struct ext4_map_blocks map;
1336 map.m_lblk = block += ret;
1337 map.m_len = max_blocks -= ret;
1339 ret = ext4_map_blocks(handle, ea_inode, &map,
1340 EXT4_GET_BLOCKS_CREATE);
1342 ext4_mark_inode_dirty(handle, ea_inode);
1343 if (ret == -ENOSPC &&
1344 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1356 while (wsize < bufsize) {
1359 csize = (bufsize - wsize) > blocksize ? blocksize :
1361 bh = ext4_getblk(handle, ea_inode, block, 0);
1366 EXT4_ERROR_INODE(ea_inode,
1367 "ext4_getblk() return bh = NULL");
1368 return -EFSCORRUPTED;
1370 ret = ext4_journal_get_write_access(handle, bh);
1374 memcpy(bh->b_data, buf, csize);
1375 set_buffer_uptodate(bh);
1376 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1383 inode_lock(ea_inode);
1384 i_size_write(ea_inode, wsize);
1385 ext4_update_i_disksize(ea_inode, wsize);
1386 inode_unlock(ea_inode);
1388 ext4_mark_inode_dirty(handle, ea_inode);
1397 * Create an inode to store the value of a large EA.
1399 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1400 struct inode *inode, u32 hash)
1402 struct inode *ea_inode = NULL;
1403 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1407 * Let the next inode be the goal, so we try and allocate the EA inode
1408 * in the same group, or nearby one.
1410 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1411 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1413 if (!IS_ERR(ea_inode)) {
1414 ea_inode->i_op = &ext4_file_inode_operations;
1415 ea_inode->i_fop = &ext4_file_operations;
1416 ext4_set_aops(ea_inode);
1417 ext4_xattr_inode_set_class(ea_inode);
1418 unlock_new_inode(ea_inode);
1419 ext4_xattr_inode_set_ref(ea_inode, 1);
1420 ext4_xattr_inode_set_hash(ea_inode, hash);
1421 err = ext4_mark_inode_dirty(handle, ea_inode);
1423 err = ext4_inode_attach_jinode(ea_inode);
1426 return ERR_PTR(err);
1430 * Xattr inodes are shared therefore quota charging is performed
1431 * at a higher level.
1433 dquot_free_inode(ea_inode);
1434 dquot_drop(ea_inode);
1435 inode_lock(ea_inode);
1436 ea_inode->i_flags |= S_NOQUOTA;
1437 inode_unlock(ea_inode);
1443 static struct inode *
1444 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1445 size_t value_len, u32 hash)
1447 struct inode *ea_inode;
1448 struct mb_cache_entry *ce;
1449 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1452 if (!ea_inode_cache)
1455 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1459 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1461 mb_cache_entry_put(ea_inode_cache, ce);
1466 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1468 if (!IS_ERR(ea_inode) &&
1469 !is_bad_inode(ea_inode) &&
1470 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1471 i_size_read(ea_inode) == value_len &&
1472 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1473 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1475 !memcmp(value, ea_data, value_len)) {
1476 mb_cache_entry_touch(ea_inode_cache, ce);
1477 mb_cache_entry_put(ea_inode_cache, ce);
1482 if (!IS_ERR(ea_inode))
1484 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1491 * Add value of the EA in an inode.
1493 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1494 const void *value, size_t value_len,
1495 struct inode **ret_inode)
1497 struct inode *ea_inode;
1501 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1502 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1504 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1510 *ret_inode = ea_inode;
1514 /* Create an inode for the EA value */
1515 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1516 if (IS_ERR(ea_inode))
1517 return PTR_ERR(ea_inode);
1519 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1521 ext4_xattr_inode_dec_ref(handle, ea_inode);
1526 if (EA_INODE_CACHE(inode))
1527 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1528 ea_inode->i_ino, true /* reusable */);
1530 *ret_inode = ea_inode;
1535 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1536 * feature is enabled.
1538 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1540 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1541 struct ext4_xattr_search *s,
1542 handle_t *handle, struct inode *inode,
1545 struct ext4_xattr_entry *last, *next;
1546 struct ext4_xattr_entry *here = s->here;
1547 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1548 int in_inode = i->in_inode;
1549 struct inode *old_ea_inode = NULL;
1550 struct inode *new_ea_inode = NULL;
1551 size_t old_size, new_size;
1554 /* Space used by old and new values. */
1555 old_size = (!s->not_found && !here->e_value_inum) ?
1556 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1557 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1560 * Optimization for the simple case when old and new values have the
1561 * same padded sizes. Not applicable if external inodes are involved.
1563 if (new_size && new_size == old_size) {
1564 size_t offs = le16_to_cpu(here->e_value_offs);
1565 void *val = s->base + offs;
1567 here->e_value_size = cpu_to_le32(i->value_len);
1568 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1569 memset(val, 0, new_size);
1571 memcpy(val, i->value, i->value_len);
1572 /* Clear padding bytes. */
1573 memset(val + i->value_len, 0, new_size - i->value_len);
1578 /* Compute min_offs and last. */
1580 for (; !IS_LAST_ENTRY(last); last = next) {
1581 next = EXT4_XATTR_NEXT(last);
1582 if ((void *)next >= s->end) {
1583 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1584 ret = -EFSCORRUPTED;
1587 if (!last->e_value_inum && last->e_value_size) {
1588 size_t offs = le16_to_cpu(last->e_value_offs);
1589 if (offs < min_offs)
1594 /* Check whether we have enough space. */
1598 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1600 free += EXT4_XATTR_LEN(name_len) + old_size;
1602 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1608 * If storing the value in an external inode is an option,
1609 * reserve space for xattr entries/names in the external
1610 * attribute block so that a long value does not occupy the
1611 * whole space and prevent futher entries being added.
1613 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1614 new_size && is_block &&
1615 (min_offs + old_size - new_size) <
1616 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1623 * Getting access to old and new ea inodes is subject to failures.
1624 * Finish that work before doing any modifications to the xattr data.
1626 if (!s->not_found && here->e_value_inum) {
1627 ret = ext4_xattr_inode_iget(inode,
1628 le32_to_cpu(here->e_value_inum),
1629 le32_to_cpu(here->e_hash),
1632 old_ea_inode = NULL;
1636 if (i->value && in_inode) {
1637 WARN_ON_ONCE(!i->value_len);
1639 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1643 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1647 new_ea_inode = NULL;
1648 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1654 /* We are ready to release ref count on the old_ea_inode. */
1655 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1657 /* Release newly required ref count on new_ea_inode. */
1661 err = ext4_xattr_inode_dec_ref(handle,
1664 ext4_warning_inode(new_ea_inode,
1665 "dec ref new_ea_inode err=%d",
1667 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1673 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1674 le32_to_cpu(here->e_value_size));
1677 /* No failures allowed past this point. */
1679 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1680 /* Remove the old value. */
1681 void *first_val = s->base + min_offs;
1682 size_t offs = le16_to_cpu(here->e_value_offs);
1683 void *val = s->base + offs;
1685 memmove(first_val + old_size, first_val, val - first_val);
1686 memset(first_val, 0, old_size);
1687 min_offs += old_size;
1689 /* Adjust all value offsets. */
1691 while (!IS_LAST_ENTRY(last)) {
1692 size_t o = le16_to_cpu(last->e_value_offs);
1694 if (!last->e_value_inum &&
1695 last->e_value_size && o < offs)
1696 last->e_value_offs = cpu_to_le16(o + old_size);
1697 last = EXT4_XATTR_NEXT(last);
1702 /* Remove old name. */
1703 size_t size = EXT4_XATTR_LEN(name_len);
1705 last = ENTRY((void *)last - size);
1706 memmove(here, (void *)here + size,
1707 (void *)last - (void *)here + sizeof(__u32));
1708 memset(last, 0, size);
1709 } else if (s->not_found) {
1710 /* Insert new name. */
1711 size_t size = EXT4_XATTR_LEN(name_len);
1712 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1714 memmove((void *)here + size, here, rest);
1715 memset(here, 0, size);
1716 here->e_name_index = i->name_index;
1717 here->e_name_len = name_len;
1718 memcpy(here->e_name, i->name, name_len);
1720 /* This is an update, reset value info. */
1721 here->e_value_inum = 0;
1722 here->e_value_offs = 0;
1723 here->e_value_size = 0;
1727 /* Insert new value. */
1729 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1730 } else if (i->value_len) {
1731 void *val = s->base + min_offs - new_size;
1733 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1734 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1735 memset(val, 0, new_size);
1737 memcpy(val, i->value, i->value_len);
1738 /* Clear padding bytes. */
1739 memset(val + i->value_len, 0,
1740 new_size - i->value_len);
1743 here->e_value_size = cpu_to_le32(i->value_len);
1750 /* Entry hash calculation. */
1755 * Feed crc32c hash instead of the raw value for entry
1756 * hash calculation. This is to avoid walking
1757 * potentially long value buffer again.
1759 crc32c_hash = cpu_to_le32(
1760 ext4_xattr_inode_get_hash(new_ea_inode));
1761 hash = ext4_xattr_hash_entry(here->e_name,
1764 } else if (is_block) {
1765 __le32 *value = s->base + le16_to_cpu(
1766 here->e_value_offs);
1768 hash = ext4_xattr_hash_entry(here->e_name,
1769 here->e_name_len, value,
1772 here->e_hash = hash;
1776 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1785 struct ext4_xattr_block_find {
1786 struct ext4_xattr_search s;
1787 struct buffer_head *bh;
1791 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1792 struct ext4_xattr_block_find *bs)
1794 struct super_block *sb = inode->i_sb;
1797 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1798 i->name_index, i->name, i->value, (long)i->value_len);
1800 if (EXT4_I(inode)->i_file_acl) {
1801 /* The inode already has an extended attribute block. */
1802 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1804 return PTR_ERR(bs->bh);
1805 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1806 atomic_read(&(bs->bh->b_count)),
1807 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1808 error = ext4_xattr_check_block(inode, bs->bh);
1811 /* Find the named attribute. */
1812 bs->s.base = BHDR(bs->bh);
1813 bs->s.first = BFIRST(bs->bh);
1814 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1815 bs->s.here = bs->s.first;
1816 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1817 i->name_index, i->name, 1);
1818 if (error && error != -ENODATA)
1820 bs->s.not_found = error;
1826 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1827 struct ext4_xattr_info *i,
1828 struct ext4_xattr_block_find *bs)
1830 struct super_block *sb = inode->i_sb;
1831 struct buffer_head *new_bh = NULL;
1832 struct ext4_xattr_search s_copy = bs->s;
1833 struct ext4_xattr_search *s = &s_copy;
1834 struct mb_cache_entry *ce = NULL;
1836 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1837 struct inode *ea_inode = NULL, *tmp_inode;
1838 size_t old_ea_inode_quota = 0;
1839 unsigned int ea_ino;
1842 #define header(x) ((struct ext4_xattr_header *)(x))
1845 BUFFER_TRACE(bs->bh, "get_write_access");
1846 error = ext4_journal_get_write_access(handle, bs->bh);
1849 lock_buffer(bs->bh);
1851 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1852 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1855 * This must happen under buffer lock for
1856 * ext4_xattr_block_set() to reliably detect modified
1860 mb_cache_entry_delete(ea_block_cache, hash,
1862 ea_bdebug(bs->bh, "modifying in-place");
1863 error = ext4_xattr_set_entry(i, s, handle, inode,
1864 true /* is_block */);
1865 ext4_xattr_block_csum_set(inode, bs->bh);
1866 unlock_buffer(bs->bh);
1867 if (error == -EFSCORRUPTED)
1870 error = ext4_handle_dirty_metadata(handle,
1877 int offset = (char *)s->here - bs->bh->b_data;
1879 unlock_buffer(bs->bh);
1880 ea_bdebug(bs->bh, "cloning");
1881 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1883 if (s->base == NULL)
1885 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1886 s->first = ENTRY(header(s->base)+1);
1887 header(s->base)->h_refcount = cpu_to_le32(1);
1888 s->here = ENTRY(s->base + offset);
1889 s->end = s->base + bs->bh->b_size;
1892 * If existing entry points to an xattr inode, we need
1893 * to prevent ext4_xattr_set_entry() from decrementing
1894 * ref count on it because the reference belongs to the
1895 * original block. In this case, make the entry look
1896 * like it has an empty value.
1898 if (!s->not_found && s->here->e_value_inum) {
1899 ea_ino = le32_to_cpu(s->here->e_value_inum);
1900 error = ext4_xattr_inode_iget(inode, ea_ino,
1901 le32_to_cpu(s->here->e_hash),
1906 if (!ext4_test_inode_state(tmp_inode,
1907 EXT4_STATE_LUSTRE_EA_INODE)) {
1909 * Defer quota free call for previous
1910 * inode until success is guaranteed.
1912 old_ea_inode_quota = le32_to_cpu(
1913 s->here->e_value_size);
1917 s->here->e_value_inum = 0;
1918 s->here->e_value_size = 0;
1922 /* Allocate a buffer where we construct the new block. */
1923 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1924 /* assert(header == s->base) */
1926 if (s->base == NULL)
1928 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1929 header(s->base)->h_blocks = cpu_to_le32(1);
1930 header(s->base)->h_refcount = cpu_to_le32(1);
1931 s->first = ENTRY(header(s->base)+1);
1932 s->here = ENTRY(header(s->base)+1);
1933 s->end = s->base + sb->s_blocksize;
1936 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1937 if (error == -EFSCORRUPTED)
1942 if (i->value && s->here->e_value_inum) {
1944 * A ref count on ea_inode has been taken as part of the call to
1945 * ext4_xattr_set_entry() above. We would like to drop this
1946 * extra ref but we have to wait until the xattr block is
1947 * initialized and has its own ref count on the ea_inode.
1949 ea_ino = le32_to_cpu(s->here->e_value_inum);
1950 error = ext4_xattr_inode_iget(inode, ea_ino,
1951 le32_to_cpu(s->here->e_hash),
1960 if (!IS_LAST_ENTRY(s->first)) {
1961 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1964 /* We found an identical block in the cache. */
1965 if (new_bh == bs->bh)
1966 ea_bdebug(new_bh, "keeping");
1970 WARN_ON_ONCE(dquot_initialize_needed(inode));
1972 /* The old block is released after updating
1974 error = dquot_alloc_block(inode,
1975 EXT4_C2B(EXT4_SB(sb), 1));
1978 BUFFER_TRACE(new_bh, "get_write_access");
1979 error = ext4_journal_get_write_access(handle,
1983 lock_buffer(new_bh);
1985 * We have to be careful about races with
1986 * freeing, rehashing or adding references to
1987 * xattr block. Once we hold buffer lock xattr
1988 * block's state is stable so we can check
1989 * whether the block got freed / rehashed or
1990 * not. Since we unhash mbcache entry under
1991 * buffer lock when freeing / rehashing xattr
1992 * block, checking whether entry is still
1993 * hashed is reliable. Same rules hold for
1994 * e_reusable handling.
1996 if (hlist_bl_unhashed(&ce->e_hash_list) ||
1999 * Undo everything and check mbcache
2002 unlock_buffer(new_bh);
2003 dquot_free_block(inode,
2004 EXT4_C2B(EXT4_SB(sb),
2007 mb_cache_entry_put(ea_block_cache, ce);
2012 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2013 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2014 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2016 ea_bdebug(new_bh, "reusing; refcount now=%d",
2018 ext4_xattr_block_csum_set(inode, new_bh);
2019 unlock_buffer(new_bh);
2020 error = ext4_handle_dirty_metadata(handle,
2026 mb_cache_entry_touch(ea_block_cache, ce);
2027 mb_cache_entry_put(ea_block_cache, ce);
2029 } else if (bs->bh && s->base == bs->bh->b_data) {
2030 /* We were modifying this block in-place. */
2031 ea_bdebug(bs->bh, "keeping this block");
2032 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2036 /* We need to allocate a new block */
2037 ext4_fsblk_t goal, block;
2039 WARN_ON_ONCE(dquot_initialize_needed(inode));
2041 goal = ext4_group_first_block_no(sb,
2042 EXT4_I(inode)->i_block_group);
2044 /* non-extent files can't have physical blocks past 2^32 */
2045 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2046 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2048 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2053 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2054 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2056 ea_idebug(inode, "creating block %llu",
2057 (unsigned long long)block);
2059 new_bh = sb_getblk(sb, block);
2060 if (unlikely(!new_bh)) {
2063 ext4_free_blocks(handle, inode, NULL, block, 1,
2064 EXT4_FREE_BLOCKS_METADATA);
2067 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2068 ENTRY(header(s->base)+1));
2072 /* Drop the extra ref on ea_inode. */
2073 error = ext4_xattr_inode_dec_ref(handle,
2076 ext4_warning_inode(ea_inode,
2083 lock_buffer(new_bh);
2084 error = ext4_journal_get_create_access(handle, new_bh);
2086 unlock_buffer(new_bh);
2090 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2091 ext4_xattr_block_csum_set(inode, new_bh);
2092 set_buffer_uptodate(new_bh);
2093 unlock_buffer(new_bh);
2094 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2095 error = ext4_handle_dirty_metadata(handle, inode,
2102 if (old_ea_inode_quota)
2103 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2105 /* Update the inode. */
2106 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2108 /* Drop the previous xattr block. */
2109 if (bs->bh && bs->bh != new_bh) {
2110 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2112 ext4_xattr_release_block(handle, inode, bs->bh,
2114 0 /* extra_credits */);
2115 ext4_xattr_inode_array_free(ea_inode_array);
2123 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2125 ext4_warning_inode(ea_inode, "dec ref error=%d",
2128 /* If there was an error, revert the quota charge. */
2130 ext4_xattr_inode_free_quota(inode, ea_inode,
2131 i_size_read(ea_inode));
2135 mb_cache_entry_put(ea_block_cache, ce);
2137 if (!(bs->bh && s->base == bs->bh->b_data))
2143 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2147 EXT4_ERROR_INODE(inode, "bad block %llu",
2148 EXT4_I(inode)->i_file_acl);
2154 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2155 struct ext4_xattr_ibody_find *is)
2157 struct ext4_xattr_ibody_header *header;
2158 struct ext4_inode *raw_inode;
2161 if (EXT4_I(inode)->i_extra_isize == 0)
2163 raw_inode = ext4_raw_inode(&is->iloc);
2164 header = IHDR(inode, raw_inode);
2165 is->s.base = is->s.first = IFIRST(header);
2166 is->s.here = is->s.first;
2167 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2168 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2169 error = xattr_check_inode(inode, header, is->s.end);
2172 /* Find the named attribute. */
2173 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2174 i->name_index, i->name, 0);
2175 if (error && error != -ENODATA)
2177 is->s.not_found = error;
2182 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2183 struct ext4_xattr_info *i,
2184 struct ext4_xattr_ibody_find *is)
2186 struct ext4_xattr_ibody_header *header;
2187 struct ext4_xattr_search *s = &is->s;
2190 if (EXT4_I(inode)->i_extra_isize == 0)
2192 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2195 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2196 if (!IS_LAST_ENTRY(s->first)) {
2197 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2198 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2200 header->h_magic = cpu_to_le32(0);
2201 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2206 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2207 struct ext4_xattr_info *i,
2208 struct ext4_xattr_ibody_find *is)
2210 struct ext4_xattr_ibody_header *header;
2211 struct ext4_xattr_search *s = &is->s;
2214 if (EXT4_I(inode)->i_extra_isize == 0)
2216 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2219 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2220 if (!IS_LAST_ENTRY(s->first)) {
2221 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2222 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2224 header->h_magic = cpu_to_le32(0);
2225 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2230 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2231 struct ext4_xattr_info *i)
2235 /* When e_value_inum is set the value is stored externally. */
2236 if (s->here->e_value_inum)
2238 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2240 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2241 return !memcmp(value, i->value, i->value_len);
2244 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2246 struct buffer_head *bh;
2249 if (!EXT4_I(inode)->i_file_acl)
2251 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2254 error = ext4_xattr_check_block(inode, bh);
2257 return ERR_PTR(error);
2263 * ext4_xattr_set_handle()
2265 * Create, replace or remove an extended attribute for this inode. Value
2266 * is NULL to remove an existing extended attribute, and non-NULL to
2267 * either replace an existing extended attribute, or create a new extended
2268 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2269 * specify that an extended attribute must exist and must not exist
2270 * previous to the call, respectively.
2272 * Returns 0, or a negative error number on failure.
2275 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2276 const char *name, const void *value, size_t value_len,
2279 struct ext4_xattr_info i = {
2280 .name_index = name_index,
2283 .value_len = value_len,
2286 struct ext4_xattr_ibody_find is = {
2287 .s = { .not_found = -ENODATA, },
2289 struct ext4_xattr_block_find bs = {
2290 .s = { .not_found = -ENODATA, },
2297 if (strlen(name) > 255)
2300 ext4_write_lock_xattr(inode, &no_expand);
2302 /* Check journal credits under write lock. */
2303 if (ext4_handle_valid(handle)) {
2304 struct buffer_head *bh;
2307 bh = ext4_xattr_get_block(inode);
2309 error = PTR_ERR(bh);
2313 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2315 flags & XATTR_CREATE);
2318 if (jbd2_handle_buffer_credits(handle) < credits) {
2324 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2328 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2329 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2330 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2331 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2334 error = ext4_xattr_ibody_find(inode, &i, &is);
2338 error = ext4_xattr_block_find(inode, &i, &bs);
2341 if (is.s.not_found && bs.s.not_found) {
2343 if (flags & XATTR_REPLACE)
2350 if (flags & XATTR_CREATE)
2355 if (!is.s.not_found)
2356 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2357 else if (!bs.s.not_found)
2358 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2361 /* Xattr value did not change? Save us some work and bail out */
2362 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2364 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2367 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2368 (EXT4_XATTR_SIZE(i.value_len) >
2369 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2372 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2373 if (!error && !bs.s.not_found) {
2375 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2376 } else if (error == -ENOSPC) {
2377 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2380 error = ext4_xattr_block_find(inode, &i, &bs);
2384 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2385 if (!error && !is.s.not_found) {
2387 error = ext4_xattr_ibody_set(handle, inode, &i,
2389 } else if (error == -ENOSPC) {
2391 * Xattr does not fit in the block, store at
2392 * external inode if possible.
2394 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2403 ext4_xattr_update_super_block(handle, inode->i_sb);
2404 inode->i_ctime = current_time(inode);
2407 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2409 * The bh is consumed by ext4_mark_iloc_dirty, even with
2414 ext4_handle_sync(handle);
2420 ext4_write_unlock_xattr(inode, &no_expand);
2424 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2425 bool is_create, int *credits)
2427 struct buffer_head *bh;
2432 if (!EXT4_SB(inode->i_sb)->s_journal)
2435 down_read(&EXT4_I(inode)->xattr_sem);
2437 bh = ext4_xattr_get_block(inode);
2441 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2442 value_len, is_create);
2447 up_read(&EXT4_I(inode)->xattr_sem);
2454 * Like ext4_xattr_set_handle, but start from an inode. This extended
2455 * attribute modification is a filesystem transaction by itself.
2457 * Returns 0, or a negative error number on failure.
2460 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2461 const void *value, size_t value_len, int flags)
2464 struct super_block *sb = inode->i_sb;
2465 int error, retries = 0;
2468 error = dquot_initialize(inode);
2473 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2478 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2479 if (IS_ERR(handle)) {
2480 error = PTR_ERR(handle);
2484 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2485 value, value_len, flags);
2486 error2 = ext4_journal_stop(handle);
2487 if (error == -ENOSPC &&
2488 ext4_should_retry_alloc(sb, &retries))
2498 * Shift the EA entries in the inode to create space for the increased
2501 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2502 int value_offs_shift, void *to,
2503 void *from, size_t n)
2505 struct ext4_xattr_entry *last = entry;
2508 /* We always shift xattr headers further thus offsets get lower */
2509 BUG_ON(value_offs_shift > 0);
2511 /* Adjust the value offsets of the entries */
2512 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2513 if (!last->e_value_inum && last->e_value_size) {
2514 new_offs = le16_to_cpu(last->e_value_offs) +
2516 last->e_value_offs = cpu_to_le16(new_offs);
2519 /* Shift the entries by n bytes */
2520 memmove(to, from, n);
2524 * Move xattr pointed to by 'entry' from inode into external xattr block
2526 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2527 struct ext4_inode *raw_inode,
2528 struct ext4_xattr_entry *entry)
2530 struct ext4_xattr_ibody_find *is = NULL;
2531 struct ext4_xattr_block_find *bs = NULL;
2532 char *buffer = NULL, *b_entry_name = NULL;
2533 size_t value_size = le32_to_cpu(entry->e_value_size);
2534 struct ext4_xattr_info i = {
2537 .name_index = entry->e_name_index,
2538 .in_inode = !!entry->e_value_inum,
2540 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2543 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2544 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2545 buffer = kmalloc(value_size, GFP_NOFS);
2546 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2547 if (!is || !bs || !buffer || !b_entry_name) {
2552 is->s.not_found = -ENODATA;
2553 bs->s.not_found = -ENODATA;
2557 /* Save the entry name and the entry value */
2558 if (entry->e_value_inum) {
2559 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2563 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2564 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2567 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2568 b_entry_name[entry->e_name_len] = '\0';
2569 i.name = b_entry_name;
2571 error = ext4_get_inode_loc(inode, &is->iloc);
2575 error = ext4_xattr_ibody_find(inode, &i, is);
2579 /* Remove the chosen entry from the inode */
2580 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2585 i.value_len = value_size;
2586 error = ext4_xattr_block_find(inode, &i, bs);
2590 /* Add entry which was removed from the inode into the block */
2591 error = ext4_xattr_block_set(handle, inode, &i, bs);
2596 kfree(b_entry_name);
2599 brelse(is->iloc.bh);
2608 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2609 struct ext4_inode *raw_inode,
2610 int isize_diff, size_t ifree,
2611 size_t bfree, int *total_ino)
2613 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2614 struct ext4_xattr_entry *small_entry;
2615 struct ext4_xattr_entry *entry;
2616 struct ext4_xattr_entry *last;
2617 unsigned int entry_size; /* EA entry size */
2618 unsigned int total_size; /* EA entry size + value size */
2619 unsigned int min_total_size;
2622 while (isize_diff > ifree) {
2625 min_total_size = ~0U;
2626 last = IFIRST(header);
2627 /* Find the entry best suited to be pushed into EA block */
2628 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2629 /* never move system.data out of the inode */
2630 if ((last->e_name_len == 4) &&
2631 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2632 !memcmp(last->e_name, "data", 4))
2634 total_size = EXT4_XATTR_LEN(last->e_name_len);
2635 if (!last->e_value_inum)
2636 total_size += EXT4_XATTR_SIZE(
2637 le32_to_cpu(last->e_value_size));
2638 if (total_size <= bfree &&
2639 total_size < min_total_size) {
2640 if (total_size + ifree < isize_diff) {
2644 min_total_size = total_size;
2649 if (entry == NULL) {
2650 if (small_entry == NULL)
2652 entry = small_entry;
2655 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2656 total_size = entry_size;
2657 if (!entry->e_value_inum)
2658 total_size += EXT4_XATTR_SIZE(
2659 le32_to_cpu(entry->e_value_size));
2660 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2665 *total_ino -= entry_size;
2666 ifree += total_size;
2667 bfree -= total_size;
2674 * Expand an inode by new_extra_isize bytes when EAs are present.
2675 * Returns 0 on success or negative error number on failure.
2677 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2678 struct ext4_inode *raw_inode, handle_t *handle)
2680 struct ext4_xattr_ibody_header *header;
2681 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2682 static unsigned int mnt_count;
2684 size_t ifree, bfree;
2687 int error = 0, tried_min_extra_isize = 0;
2688 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2689 int isize_diff; /* How much do we need to grow i_extra_isize */
2692 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2693 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2696 header = IHDR(inode, raw_inode);
2699 * Check if enough free space is available in the inode to shift the
2700 * entries ahead by new_extra_isize.
2703 base = IFIRST(header);
2704 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2705 min_offs = end - base;
2706 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2708 error = xattr_check_inode(inode, header, end);
2712 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2713 if (ifree >= isize_diff)
2717 * Enough free space isn't available in the inode, check if
2718 * EA block can hold new_extra_isize bytes.
2720 if (EXT4_I(inode)->i_file_acl) {
2721 struct buffer_head *bh;
2723 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2725 error = PTR_ERR(bh);
2728 error = ext4_xattr_check_block(inode, bh);
2734 end = bh->b_data + bh->b_size;
2735 min_offs = end - base;
2736 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2739 if (bfree + ifree < isize_diff) {
2740 if (!tried_min_extra_isize && s_min_extra_isize) {
2741 tried_min_extra_isize++;
2742 new_extra_isize = s_min_extra_isize;
2749 bfree = inode->i_sb->s_blocksize;
2752 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2753 isize_diff, ifree, bfree,
2756 if (error == -ENOSPC && !tried_min_extra_isize &&
2757 s_min_extra_isize) {
2758 tried_min_extra_isize++;
2759 new_extra_isize = s_min_extra_isize;
2765 /* Adjust the offsets and shift the remaining entries ahead */
2766 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2767 - new_extra_isize, (void *)raw_inode +
2768 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2769 (void *)header, total_ino);
2770 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2773 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2774 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2776 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2781 #define EIA_INCR 16 /* must be 2^n */
2782 #define EIA_MASK (EIA_INCR - 1)
2784 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2785 * If @ea_inode_array is new or full it will be grown and the old
2786 * contents copied over.
2789 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2790 struct inode *inode)
2792 if (*ea_inode_array == NULL) {
2794 * Start with 15 inodes, so it fits into a power-of-two size.
2795 * If *ea_inode_array is NULL, this is essentially offsetof()
2798 kmalloc(offsetof(struct ext4_xattr_inode_array,
2801 if (*ea_inode_array == NULL)
2803 (*ea_inode_array)->count = 0;
2804 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2805 /* expand the array once all 15 + n * 16 slots are full */
2806 struct ext4_xattr_inode_array *new_array = NULL;
2807 int count = (*ea_inode_array)->count;
2809 /* if new_array is NULL, this is essentially offsetof() */
2810 new_array = kmalloc(
2811 offsetof(struct ext4_xattr_inode_array,
2812 inodes[count + EIA_INCR]),
2814 if (new_array == NULL)
2816 memcpy(new_array, *ea_inode_array,
2817 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2818 kfree(*ea_inode_array);
2819 *ea_inode_array = new_array;
2821 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2826 * ext4_xattr_delete_inode()
2828 * Free extended attribute resources associated with this inode. Traverse
2829 * all entries and decrement reference on any xattr inodes associated with this
2830 * inode. This is called immediately before an inode is freed. We have exclusive
2831 * access to the inode. If an orphan inode is deleted it will also release its
2832 * references on xattr block and xattr inodes.
2834 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2835 struct ext4_xattr_inode_array **ea_inode_array,
2838 struct buffer_head *bh = NULL;
2839 struct ext4_xattr_ibody_header *header;
2840 struct ext4_iloc iloc = { .bh = NULL };
2841 struct ext4_xattr_entry *entry;
2842 struct inode *ea_inode;
2845 error = ext4_journal_ensure_credits(handle, extra_credits,
2846 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2848 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2852 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2853 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2855 error = ext4_get_inode_loc(inode, &iloc);
2857 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2861 error = ext4_journal_get_write_access(handle, iloc.bh);
2863 EXT4_ERROR_INODE(inode, "write access (error %d)",
2868 header = IHDR(inode, ext4_raw_inode(&iloc));
2869 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2870 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2872 false /* block_csum */,
2875 false /* skip_quota */);
2878 if (EXT4_I(inode)->i_file_acl) {
2879 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2881 error = PTR_ERR(bh);
2883 EXT4_ERROR_INODE(inode, "block %llu read error",
2884 EXT4_I(inode)->i_file_acl);
2888 error = ext4_xattr_check_block(inode, bh);
2892 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2893 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2894 entry = EXT4_XATTR_NEXT(entry)) {
2895 if (!entry->e_value_inum)
2897 error = ext4_xattr_inode_iget(inode,
2898 le32_to_cpu(entry->e_value_inum),
2899 le32_to_cpu(entry->e_hash),
2903 ext4_xattr_inode_free_quota(inode, ea_inode,
2904 le32_to_cpu(entry->e_value_size));
2910 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2913 * Update i_file_acl value in the same transaction that releases
2916 EXT4_I(inode)->i_file_acl = 0;
2917 error = ext4_mark_inode_dirty(handle, inode);
2919 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2931 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2935 if (ea_inode_array == NULL)
2938 for (idx = 0; idx < ea_inode_array->count; ++idx)
2939 iput(ea_inode_array->inodes[idx]);
2940 kfree(ea_inode_array);
2944 * ext4_xattr_block_cache_insert()
2946 * Create a new entry in the extended attribute block cache, and insert
2947 * it unless such an entry is already in the cache.
2949 * Returns 0, or a negative error number on failure.
2952 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2953 struct buffer_head *bh)
2955 struct ext4_xattr_header *header = BHDR(bh);
2956 __u32 hash = le32_to_cpu(header->h_hash);
2957 int reusable = le32_to_cpu(header->h_refcount) <
2958 EXT4_XATTR_REFCOUNT_MAX;
2961 if (!ea_block_cache)
2963 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2964 bh->b_blocknr, reusable);
2966 if (error == -EBUSY)
2967 ea_bdebug(bh, "already in cache");
2969 ea_bdebug(bh, "inserting [%x]", (int)hash);
2975 * Compare two extended attribute blocks for equality.
2977 * Returns 0 if the blocks are equal, 1 if they differ, and
2978 * a negative error number on errors.
2981 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2982 struct ext4_xattr_header *header2)
2984 struct ext4_xattr_entry *entry1, *entry2;
2986 entry1 = ENTRY(header1+1);
2987 entry2 = ENTRY(header2+1);
2988 while (!IS_LAST_ENTRY(entry1)) {
2989 if (IS_LAST_ENTRY(entry2))
2991 if (entry1->e_hash != entry2->e_hash ||
2992 entry1->e_name_index != entry2->e_name_index ||
2993 entry1->e_name_len != entry2->e_name_len ||
2994 entry1->e_value_size != entry2->e_value_size ||
2995 entry1->e_value_inum != entry2->e_value_inum ||
2996 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
2998 if (!entry1->e_value_inum &&
2999 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3000 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3001 le32_to_cpu(entry1->e_value_size)))
3004 entry1 = EXT4_XATTR_NEXT(entry1);
3005 entry2 = EXT4_XATTR_NEXT(entry2);
3007 if (!IS_LAST_ENTRY(entry2))
3013 * ext4_xattr_block_cache_find()
3015 * Find an identical extended attribute block.
3017 * Returns a pointer to the block found, or NULL if such a block was
3018 * not found or an error occurred.
3020 static struct buffer_head *
3021 ext4_xattr_block_cache_find(struct inode *inode,
3022 struct ext4_xattr_header *header,
3023 struct mb_cache_entry **pce)
3025 __u32 hash = le32_to_cpu(header->h_hash);
3026 struct mb_cache_entry *ce;
3027 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3029 if (!ea_block_cache)
3031 if (!header->h_hash)
3032 return NULL; /* never share */
3033 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3034 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3036 struct buffer_head *bh;
3038 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3040 if (PTR_ERR(bh) == -ENOMEM)
3043 EXT4_ERROR_INODE(inode, "block %lu read error",
3044 (unsigned long)ce->e_value);
3045 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3050 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3055 #define NAME_HASH_SHIFT 5
3056 #define VALUE_HASH_SHIFT 16
3059 * ext4_xattr_hash_entry()
3061 * Compute the hash of an extended attribute.
3063 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3068 while (name_len--) {
3069 hash = (hash << NAME_HASH_SHIFT) ^
3070 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3073 while (value_count--) {
3074 hash = (hash << VALUE_HASH_SHIFT) ^
3075 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3076 le32_to_cpu(*value++);
3078 return cpu_to_le32(hash);
3081 #undef NAME_HASH_SHIFT
3082 #undef VALUE_HASH_SHIFT
3084 #define BLOCK_HASH_SHIFT 16
3087 * ext4_xattr_rehash()
3089 * Re-compute the extended attribute hash value after an entry has changed.
3091 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3093 struct ext4_xattr_entry *here;
3096 here = ENTRY(header+1);
3097 while (!IS_LAST_ENTRY(here)) {
3098 if (!here->e_hash) {
3099 /* Block is not shared if an entry's hash value == 0 */
3103 hash = (hash << BLOCK_HASH_SHIFT) ^
3104 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3105 le32_to_cpu(here->e_hash);
3106 here = EXT4_XATTR_NEXT(here);
3108 header->h_hash = cpu_to_le32(hash);
3111 #undef BLOCK_HASH_SHIFT
3113 #define HASH_BUCKET_BITS 10
3116 ext4_xattr_create_cache(void)
3118 return mb_cache_create(HASH_BUCKET_BITS);
3121 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3124 mb_cache_destroy(cache);