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,
96 [EXT4_XATTR_INDEX_HURD] = &ext4_xattr_hurd_handler,
99 const struct xattr_handler *ext4_xattr_handlers[] = {
100 &ext4_xattr_user_handler,
101 &ext4_xattr_trusted_handler,
102 #ifdef CONFIG_EXT4_FS_POSIX_ACL
103 &posix_acl_access_xattr_handler,
104 &posix_acl_default_xattr_handler,
106 #ifdef CONFIG_EXT4_FS_SECURITY
107 &ext4_xattr_security_handler,
109 &ext4_xattr_hurd_handler,
113 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
114 inode->i_sb->s_fs_info)->s_ea_block_cache)
116 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
117 inode->i_sb->s_fs_info)->s_ea_inode_cache)
120 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
121 struct inode *inode);
123 #ifdef CONFIG_LOCKDEP
124 void ext4_xattr_inode_set_class(struct inode *ea_inode)
126 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
130 static __le32 ext4_xattr_block_csum(struct inode *inode,
132 struct ext4_xattr_header *hdr)
134 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
136 __le64 dsk_block_nr = cpu_to_le64(block_nr);
137 __u32 dummy_csum = 0;
138 int offset = offsetof(struct ext4_xattr_header, h_checksum);
140 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
141 sizeof(dsk_block_nr));
142 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
143 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
144 offset += sizeof(dummy_csum);
145 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
146 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
148 return cpu_to_le32(csum);
151 static int ext4_xattr_block_csum_verify(struct inode *inode,
152 struct buffer_head *bh)
154 struct ext4_xattr_header *hdr = BHDR(bh);
157 if (ext4_has_metadata_csum(inode->i_sb)) {
159 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
160 bh->b_blocknr, hdr));
166 static void ext4_xattr_block_csum_set(struct inode *inode,
167 struct buffer_head *bh)
169 if (ext4_has_metadata_csum(inode->i_sb))
170 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
171 bh->b_blocknr, BHDR(bh));
174 static inline const struct xattr_handler *
175 ext4_xattr_handler(int name_index)
177 const struct xattr_handler *handler = NULL;
179 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
180 handler = ext4_xattr_handler_map[name_index];
185 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
188 struct ext4_xattr_entry *e = entry;
190 /* Find the end of the names list */
191 while (!IS_LAST_ENTRY(e)) {
192 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
193 if ((void *)next >= end)
194 return -EFSCORRUPTED;
195 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
196 return -EFSCORRUPTED;
200 /* Check the values */
201 while (!IS_LAST_ENTRY(entry)) {
202 u32 size = le32_to_cpu(entry->e_value_size);
204 if (size > EXT4_XATTR_SIZE_MAX)
205 return -EFSCORRUPTED;
207 if (size != 0 && entry->e_value_inum == 0) {
208 u16 offs = le16_to_cpu(entry->e_value_offs);
212 * The value cannot overlap the names, and the value
213 * with padding cannot extend beyond 'end'. Check both
214 * the padded and unpadded sizes, since the size may
215 * overflow to 0 when adding padding.
217 if (offs > end - value_start)
218 return -EFSCORRUPTED;
219 value = value_start + offs;
220 if (value < (void *)e + sizeof(u32) ||
221 size > end - value ||
222 EXT4_XATTR_SIZE(size) > end - value)
223 return -EFSCORRUPTED;
225 entry = EXT4_XATTR_NEXT(entry);
232 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
233 const char *function, unsigned int line)
235 int error = -EFSCORRUPTED;
237 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
238 BHDR(bh)->h_blocks != cpu_to_le32(1))
240 if (buffer_verified(bh))
244 if (!ext4_xattr_block_csum_verify(inode, bh))
246 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
250 __ext4_error_inode(inode, function, line, 0, -error,
251 "corrupted xattr block %llu",
252 (unsigned long long) bh->b_blocknr);
254 set_buffer_verified(bh);
258 #define ext4_xattr_check_block(inode, bh) \
259 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
263 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
264 void *end, const char *function, unsigned int line)
266 int error = -EFSCORRUPTED;
268 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
269 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
271 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
274 __ext4_error_inode(inode, function, line, 0, -error,
275 "corrupted in-inode xattr");
279 #define xattr_check_inode(inode, header, end) \
280 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
283 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
284 void *end, int name_index, const char *name, int sorted)
286 struct ext4_xattr_entry *entry, *next;
292 name_len = strlen(name);
293 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
294 next = EXT4_XATTR_NEXT(entry);
295 if ((void *) next >= end) {
296 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
297 return -EFSCORRUPTED;
299 cmp = name_index - entry->e_name_index;
301 cmp = name_len - entry->e_name_len;
303 cmp = memcmp(name, entry->e_name, name_len);
304 if (cmp <= 0 && (sorted || cmp == 0))
308 return cmp ? -ENODATA : 0;
312 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
314 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
317 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
319 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
320 (u32) inode_peek_iversion_raw(ea_inode);
323 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
325 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
326 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
329 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
331 return (u32)ea_inode->i_atime.tv_sec;
334 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
336 ea_inode->i_atime.tv_sec = hash;
340 * Read the EA value from an inode.
342 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
344 int blocksize = 1 << ea_inode->i_blkbits;
345 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
346 int tail_size = (size % blocksize) ?: blocksize;
347 struct buffer_head *bhs_inline[8];
348 struct buffer_head **bhs = bhs_inline;
351 if (bh_count > ARRAY_SIZE(bhs_inline)) {
352 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
357 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
358 true /* wait */, bhs);
362 for (i = 0; i < bh_count; i++) {
363 /* There shouldn't be any holes in ea_inode. */
368 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
369 i < bh_count - 1 ? blocksize : tail_size);
373 for (i = 0; i < bh_count; i++)
376 if (bhs != bhs_inline)
381 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
383 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
384 u32 ea_inode_hash, struct inode **ea_inode)
389 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
391 err = PTR_ERR(inode);
392 ext4_error(parent->i_sb,
393 "error while reading EA inode %lu err=%d", ea_ino,
398 if (is_bad_inode(inode)) {
399 ext4_error(parent->i_sb,
400 "error while reading EA inode %lu is_bad_inode",
406 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
407 ext4_error(parent->i_sb,
408 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
414 ext4_xattr_inode_set_class(inode);
417 * Check whether this is an old Lustre-style xattr inode. Lustre
418 * implementation does not have hash validation, rather it has a
419 * backpointer from ea_inode to the parent inode.
421 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
422 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
423 inode->i_generation == parent->i_generation) {
424 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
425 ext4_xattr_inode_set_ref(inode, 1);
428 inode->i_flags |= S_NOQUOTA;
439 /* Remove entry from mbcache when EA inode is getting evicted */
440 void ext4_evict_ea_inode(struct inode *inode)
442 struct mb_cache_entry *oe;
444 if (!EA_INODE_CACHE(inode))
446 /* Wait for entry to get unused so that we can remove it */
447 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
448 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
449 mb_cache_entry_wait_unused(oe);
450 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
455 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
456 struct ext4_xattr_entry *entry, void *buffer,
461 /* Verify stored hash matches calculated hash. */
462 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
463 if (hash != ext4_xattr_inode_get_hash(ea_inode))
464 return -EFSCORRUPTED;
467 __le32 e_hash, tmp_data;
469 /* Verify entry hash. */
470 tmp_data = cpu_to_le32(hash);
471 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
473 if (e_hash != entry->e_hash)
474 return -EFSCORRUPTED;
480 * Read xattr value from the EA inode.
483 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
484 void *buffer, size_t size)
486 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
487 struct inode *ea_inode;
490 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
491 le32_to_cpu(entry->e_hash), &ea_inode);
497 if (i_size_read(ea_inode) != size) {
498 ext4_warning_inode(ea_inode,
499 "ea_inode file size=%llu entry size=%zu",
500 i_size_read(ea_inode), size);
505 err = ext4_xattr_inode_read(ea_inode, buffer, size);
509 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
510 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
513 ext4_warning_inode(ea_inode,
514 "EA inode hash validation failed");
519 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
520 ext4_xattr_inode_get_hash(ea_inode),
521 ea_inode->i_ino, true /* reusable */);
529 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
530 void *buffer, size_t buffer_size)
532 struct buffer_head *bh = NULL;
533 struct ext4_xattr_entry *entry;
537 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
539 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
540 name_index, name, buffer, (long)buffer_size);
542 if (!EXT4_I(inode)->i_file_acl)
544 ea_idebug(inode, "reading block %llu",
545 (unsigned long long)EXT4_I(inode)->i_file_acl);
546 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
549 ea_bdebug(bh, "b_count=%d, refcount=%d",
550 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
551 error = ext4_xattr_check_block(inode, bh);
554 ext4_xattr_block_cache_insert(ea_block_cache, bh);
556 end = bh->b_data + bh->b_size;
557 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
560 size = le32_to_cpu(entry->e_value_size);
562 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
565 if (size > buffer_size)
567 if (entry->e_value_inum) {
568 error = ext4_xattr_inode_get(inode, entry, buffer,
573 u16 offset = le16_to_cpu(entry->e_value_offs);
574 void *p = bh->b_data + offset;
576 if (unlikely(p + size > end))
578 memcpy(buffer, p, size);
589 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
590 void *buffer, size_t buffer_size)
592 struct ext4_xattr_ibody_header *header;
593 struct ext4_xattr_entry *entry;
594 struct ext4_inode *raw_inode;
595 struct ext4_iloc iloc;
600 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
602 error = ext4_get_inode_loc(inode, &iloc);
605 raw_inode = ext4_raw_inode(&iloc);
606 header = IHDR(inode, raw_inode);
607 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
608 error = xattr_check_inode(inode, header, end);
611 entry = IFIRST(header);
612 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
615 size = le32_to_cpu(entry->e_value_size);
617 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
620 if (size > buffer_size)
622 if (entry->e_value_inum) {
623 error = ext4_xattr_inode_get(inode, entry, buffer,
628 u16 offset = le16_to_cpu(entry->e_value_offs);
629 void *p = (void *)IFIRST(header) + offset;
631 if (unlikely(p + size > end))
633 memcpy(buffer, p, size);
646 * Copy an extended attribute into the buffer
647 * provided, or compute the buffer size required.
648 * Buffer is NULL to compute the size of the buffer required.
650 * Returns a negative error number on failure, or the number of bytes
651 * used / required on success.
654 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
655 void *buffer, size_t buffer_size)
659 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
662 if (strlen(name) > 255)
665 down_read(&EXT4_I(inode)->xattr_sem);
666 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
668 if (error == -ENODATA)
669 error = ext4_xattr_block_get(inode, name_index, name, buffer,
671 up_read(&EXT4_I(inode)->xattr_sem);
676 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
677 char *buffer, size_t buffer_size)
679 size_t rest = buffer_size;
681 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
682 const struct xattr_handler *handler =
683 ext4_xattr_handler(entry->e_name_index);
685 if (handler && (!handler->list || handler->list(dentry))) {
686 const char *prefix = handler->prefix ?: handler->name;
687 size_t prefix_len = strlen(prefix);
688 size_t size = prefix_len + entry->e_name_len + 1;
693 memcpy(buffer, prefix, prefix_len);
694 buffer += prefix_len;
695 memcpy(buffer, entry->e_name, entry->e_name_len);
696 buffer += entry->e_name_len;
702 return buffer_size - rest; /* total size */
706 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
708 struct inode *inode = d_inode(dentry);
709 struct buffer_head *bh = NULL;
712 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
713 buffer, (long)buffer_size);
715 if (!EXT4_I(inode)->i_file_acl)
717 ea_idebug(inode, "reading block %llu",
718 (unsigned long long)EXT4_I(inode)->i_file_acl);
719 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
722 ea_bdebug(bh, "b_count=%d, refcount=%d",
723 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
724 error = ext4_xattr_check_block(inode, bh);
727 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
728 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
736 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
738 struct inode *inode = d_inode(dentry);
739 struct ext4_xattr_ibody_header *header;
740 struct ext4_inode *raw_inode;
741 struct ext4_iloc iloc;
745 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
747 error = ext4_get_inode_loc(inode, &iloc);
750 raw_inode = ext4_raw_inode(&iloc);
751 header = IHDR(inode, raw_inode);
752 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
753 error = xattr_check_inode(inode, header, end);
756 error = ext4_xattr_list_entries(dentry, IFIRST(header),
757 buffer, buffer_size);
765 * Inode operation listxattr()
767 * d_inode(dentry)->i_rwsem: don't care
769 * Copy a list of attribute names into the buffer
770 * provided, or compute the buffer size required.
771 * Buffer is NULL to compute the size of the buffer required.
773 * Returns a negative error number on failure, or the number of bytes
774 * used / required on success.
777 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
781 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
782 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
789 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
794 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
799 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
802 static void ext4_xattr_update_super_block(handle_t *handle,
803 struct super_block *sb)
805 if (ext4_has_feature_xattr(sb))
808 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
809 if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
810 EXT4_JTR_NONE) == 0) {
811 lock_buffer(EXT4_SB(sb)->s_sbh);
812 ext4_set_feature_xattr(sb);
813 ext4_superblock_csum_set(sb);
814 unlock_buffer(EXT4_SB(sb)->s_sbh);
815 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
819 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
821 struct ext4_iloc iloc = { .bh = NULL };
822 struct buffer_head *bh = NULL;
823 struct ext4_inode *raw_inode;
824 struct ext4_xattr_ibody_header *header;
825 struct ext4_xattr_entry *entry;
826 qsize_t ea_inode_refs = 0;
830 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
832 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
833 ret = ext4_get_inode_loc(inode, &iloc);
836 raw_inode = ext4_raw_inode(&iloc);
837 header = IHDR(inode, raw_inode);
838 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
839 ret = xattr_check_inode(inode, header, end);
843 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
844 entry = EXT4_XATTR_NEXT(entry))
845 if (entry->e_value_inum)
849 if (EXT4_I(inode)->i_file_acl) {
850 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
857 ret = ext4_xattr_check_block(inode, bh);
861 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
862 entry = EXT4_XATTR_NEXT(entry))
863 if (entry->e_value_inum)
866 *usage = ea_inode_refs + 1;
874 static inline size_t round_up_cluster(struct inode *inode, size_t length)
876 struct super_block *sb = inode->i_sb;
877 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
879 size_t mask = ~(cluster_size - 1);
881 return (length + cluster_size - 1) & mask;
884 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
888 err = dquot_alloc_inode(inode);
891 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
893 dquot_free_inode(inode);
897 static void ext4_xattr_inode_free_quota(struct inode *parent,
898 struct inode *ea_inode,
902 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
904 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
905 dquot_free_inode(parent);
908 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
909 struct buffer_head *block_bh, size_t value_len,
916 * 1) Owner inode update
917 * 2) Ref count update on old xattr block
919 * 4) block bitmap update for new xattr block
920 * 5) group descriptor for new xattr block
921 * 6) block bitmap update for old xattr block
922 * 7) group descriptor for old block
924 * 6 & 7 can happen if we have two racing threads T_a and T_b
925 * which are each trying to set an xattr on inodes I_a and I_b
926 * which were both initially sharing an xattr block.
931 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
934 * In case of inline data, we may push out the data to a block,
935 * so we need to reserve credits for this eventuality
937 if (inode && ext4_has_inline_data(inode))
938 credits += ext4_writepage_trans_blocks(inode) + 1;
940 /* We are done if ea_inode feature is not enabled. */
941 if (!ext4_has_feature_ea_inode(sb))
944 /* New ea_inode, inode map, block bitmap, group descriptor. */
948 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
950 /* Indirection block or one level of extent tree. */
953 /* Block bitmap and group descriptor updates for each block. */
954 credits += blocks * 2;
956 /* Blocks themselves. */
960 /* Dereference ea_inode holding old xattr value.
961 * Old ea_inode, inode map, block bitmap, group descriptor.
965 /* Data blocks for old ea_inode. */
966 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
968 /* Indirection block or one level of extent tree for old
973 /* Block bitmap and group descriptor updates for each block. */
974 credits += blocks * 2;
977 /* We may need to clone the existing xattr block in which case we need
978 * to increment ref counts for existing ea_inodes referenced by it.
981 struct ext4_xattr_entry *entry = BFIRST(block_bh);
983 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
984 if (entry->e_value_inum)
985 /* Ref count update on ea_inode. */
991 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
994 struct ext4_iloc iloc;
998 inode_lock(ea_inode);
1000 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1004 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1005 ref_count += ref_change;
1006 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1008 if (ref_change > 0) {
1009 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1010 ea_inode->i_ino, ref_count);
1012 if (ref_count == 1) {
1013 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1014 ea_inode->i_ino, ea_inode->i_nlink);
1016 set_nlink(ea_inode, 1);
1017 ext4_orphan_del(handle, ea_inode);
1020 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1021 ea_inode->i_ino, ref_count);
1023 if (ref_count == 0) {
1024 WARN_ONCE(ea_inode->i_nlink != 1,
1025 "EA inode %lu i_nlink=%u",
1026 ea_inode->i_ino, ea_inode->i_nlink);
1028 clear_nlink(ea_inode);
1029 ext4_orphan_add(handle, ea_inode);
1033 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1035 ext4_warning_inode(ea_inode,
1036 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1038 inode_unlock(ea_inode);
1042 static int ext4_xattr_inode_inc_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_dec_ref(handle_t *handle, struct inode *ea_inode)
1049 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1052 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1053 struct ext4_xattr_entry *first)
1055 struct inode *ea_inode;
1056 struct ext4_xattr_entry *entry;
1057 struct ext4_xattr_entry *failed_entry;
1058 unsigned int ea_ino;
1061 for (entry = first; !IS_LAST_ENTRY(entry);
1062 entry = EXT4_XATTR_NEXT(entry)) {
1063 if (!entry->e_value_inum)
1065 ea_ino = le32_to_cpu(entry->e_value_inum);
1066 err = ext4_xattr_inode_iget(parent, ea_ino,
1067 le32_to_cpu(entry->e_hash),
1071 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1073 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1083 failed_entry = entry;
1085 for (entry = first; entry != failed_entry;
1086 entry = EXT4_XATTR_NEXT(entry)) {
1087 if (!entry->e_value_inum)
1089 ea_ino = le32_to_cpu(entry->e_value_inum);
1090 err = ext4_xattr_inode_iget(parent, ea_ino,
1091 le32_to_cpu(entry->e_hash),
1094 ext4_warning(parent->i_sb,
1095 "cleanup ea_ino %u iget error %d", ea_ino,
1099 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1101 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1108 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1109 struct buffer_head *bh, bool block_csum, bool dirty)
1115 ext4_xattr_block_csum_set(inode, bh);
1116 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1118 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1127 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1128 struct buffer_head *bh,
1129 struct ext4_xattr_entry *first, bool block_csum,
1130 struct ext4_xattr_inode_array **ea_inode_array,
1131 int extra_credits, bool skip_quota)
1133 struct inode *ea_inode;
1134 struct ext4_xattr_entry *entry;
1136 unsigned int ea_ino;
1140 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1141 credits = 2 + extra_credits;
1143 for (entry = first; !IS_LAST_ENTRY(entry);
1144 entry = EXT4_XATTR_NEXT(entry)) {
1145 if (!entry->e_value_inum)
1147 ea_ino = le32_to_cpu(entry->e_value_inum);
1148 err = ext4_xattr_inode_iget(parent, ea_ino,
1149 le32_to_cpu(entry->e_hash),
1154 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1156 ext4_warning_inode(ea_inode,
1157 "Expand inode array err=%d", err);
1162 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1163 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1164 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1167 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1172 err = ext4_journal_get_write_access(handle,
1173 parent->i_sb, bh, EXT4_JTR_NONE);
1175 ext4_warning_inode(ea_inode,
1176 "Re-get write access err=%d",
1182 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1184 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1190 ext4_xattr_inode_free_quota(parent, ea_inode,
1191 le32_to_cpu(entry->e_value_size));
1194 * Forget about ea_inode within the same transaction that
1195 * decrements the ref count. This avoids duplicate decrements in
1196 * case the rest of the work spills over to subsequent
1199 entry->e_value_inum = 0;
1200 entry->e_value_size = 0;
1207 * Note that we are deliberately skipping csum calculation for
1208 * the final update because we do not expect any journal
1209 * restarts until xattr block is freed.
1212 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1214 ext4_warning_inode(parent,
1215 "handle dirty metadata err=%d", err);
1220 * Release the xattr block BH: If the reference count is > 1, decrement it;
1221 * otherwise free the block.
1224 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1225 struct buffer_head *bh,
1226 struct ext4_xattr_inode_array **ea_inode_array,
1229 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1233 BUFFER_TRACE(bh, "get_write_access");
1234 error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1241 hash = le32_to_cpu(BHDR(bh)->h_hash);
1242 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1244 ea_bdebug(bh, "refcount now=0; freeing");
1246 * This must happen under buffer lock for
1247 * ext4_xattr_block_set() to reliably detect freed block
1249 if (ea_block_cache) {
1250 struct mb_cache_entry *oe;
1252 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1256 mb_cache_entry_wait_unused(oe);
1257 mb_cache_entry_put(ea_block_cache, oe);
1264 if (ext4_has_feature_ea_inode(inode->i_sb))
1265 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1267 true /* block_csum */,
1270 true /* skip_quota */);
1271 ext4_free_blocks(handle, inode, bh, 0, 1,
1272 EXT4_FREE_BLOCKS_METADATA |
1273 EXT4_FREE_BLOCKS_FORGET);
1276 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1277 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1278 struct mb_cache_entry *ce;
1280 if (ea_block_cache) {
1281 ce = mb_cache_entry_get(ea_block_cache, hash,
1285 mb_cache_entry_put(ea_block_cache, ce);
1290 ext4_xattr_block_csum_set(inode, bh);
1292 * Beware of this ugliness: Releasing of xattr block references
1293 * from different inodes can race and so we have to protect
1294 * from a race where someone else frees the block (and releases
1295 * its journal_head) before we are done dirtying the buffer. In
1296 * nojournal mode this race is harmless and we actually cannot
1297 * call ext4_handle_dirty_metadata() with locked buffer as
1298 * that function can call sync_dirty_buffer() so for that case
1299 * we handle the dirtying after unlocking the buffer.
1301 if (ext4_handle_valid(handle))
1302 error = ext4_handle_dirty_metadata(handle, inode, bh);
1304 if (!ext4_handle_valid(handle))
1305 error = ext4_handle_dirty_metadata(handle, inode, bh);
1307 ext4_handle_sync(handle);
1308 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1309 ea_bdebug(bh, "refcount now=%d; releasing",
1310 le32_to_cpu(BHDR(bh)->h_refcount));
1313 ext4_std_error(inode->i_sb, error);
1318 * Find the available free space for EAs. This also returns the total number of
1319 * bytes used by EA entries.
1321 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1322 size_t *min_offs, void *base, int *total)
1324 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1325 if (!last->e_value_inum && last->e_value_size) {
1326 size_t offs = le16_to_cpu(last->e_value_offs);
1327 if (offs < *min_offs)
1331 *total += EXT4_XATTR_LEN(last->e_name_len);
1333 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1337 * Write the value of the EA in an inode.
1339 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1340 const void *buf, int bufsize)
1342 struct buffer_head *bh = NULL;
1343 unsigned long block = 0;
1344 int blocksize = ea_inode->i_sb->s_blocksize;
1345 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1346 int csize, wsize = 0;
1347 int ret = 0, ret2 = 0;
1351 while (ret >= 0 && ret < max_blocks) {
1352 struct ext4_map_blocks map;
1353 map.m_lblk = block += ret;
1354 map.m_len = max_blocks -= ret;
1356 ret = ext4_map_blocks(handle, ea_inode, &map,
1357 EXT4_GET_BLOCKS_CREATE);
1359 ext4_mark_inode_dirty(handle, ea_inode);
1360 if (ret == -ENOSPC &&
1361 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1373 while (wsize < bufsize) {
1375 csize = (bufsize - wsize) > blocksize ? blocksize :
1377 bh = ext4_getblk(handle, ea_inode, block, 0);
1382 EXT4_ERROR_INODE(ea_inode,
1383 "ext4_getblk() return bh = NULL");
1384 return -EFSCORRUPTED;
1386 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1391 memcpy(bh->b_data, buf, csize);
1392 set_buffer_uptodate(bh);
1393 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1400 inode_lock(ea_inode);
1401 i_size_write(ea_inode, wsize);
1402 ext4_update_i_disksize(ea_inode, wsize);
1403 inode_unlock(ea_inode);
1405 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1406 if (unlikely(ret2 && !ret))
1416 * Create an inode to store the value of a large EA.
1418 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1419 struct inode *inode, u32 hash)
1421 struct inode *ea_inode = NULL;
1422 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1426 * Let the next inode be the goal, so we try and allocate the EA inode
1427 * in the same group, or nearby one.
1429 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1430 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1432 if (!IS_ERR(ea_inode)) {
1433 ea_inode->i_op = &ext4_file_inode_operations;
1434 ea_inode->i_fop = &ext4_file_operations;
1435 ext4_set_aops(ea_inode);
1436 ext4_xattr_inode_set_class(ea_inode);
1437 unlock_new_inode(ea_inode);
1438 ext4_xattr_inode_set_ref(ea_inode, 1);
1439 ext4_xattr_inode_set_hash(ea_inode, hash);
1440 err = ext4_mark_inode_dirty(handle, ea_inode);
1442 err = ext4_inode_attach_jinode(ea_inode);
1445 return ERR_PTR(err);
1449 * Xattr inodes are shared therefore quota charging is performed
1450 * at a higher level.
1452 dquot_free_inode(ea_inode);
1453 dquot_drop(ea_inode);
1454 inode_lock(ea_inode);
1455 ea_inode->i_flags |= S_NOQUOTA;
1456 inode_unlock(ea_inode);
1462 static struct inode *
1463 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1464 size_t value_len, u32 hash)
1466 struct inode *ea_inode;
1467 struct mb_cache_entry *ce;
1468 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1471 if (!ea_inode_cache)
1474 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1478 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1479 !(current->flags & PF_MEMALLOC_NOFS));
1481 ea_data = kvmalloc(value_len, GFP_KERNEL);
1483 mb_cache_entry_put(ea_inode_cache, ce);
1488 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1490 if (!IS_ERR(ea_inode) &&
1491 !is_bad_inode(ea_inode) &&
1492 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1493 i_size_read(ea_inode) == value_len &&
1494 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1495 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1497 !memcmp(value, ea_data, value_len)) {
1498 mb_cache_entry_touch(ea_inode_cache, ce);
1499 mb_cache_entry_put(ea_inode_cache, ce);
1504 if (!IS_ERR(ea_inode))
1506 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1513 * Add value of the EA in an inode.
1515 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1516 const void *value, size_t value_len,
1517 struct inode **ret_inode)
1519 struct inode *ea_inode;
1523 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1524 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1526 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1532 *ret_inode = ea_inode;
1536 /* Create an inode for the EA value */
1537 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1538 if (IS_ERR(ea_inode))
1539 return PTR_ERR(ea_inode);
1541 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1543 ext4_xattr_inode_dec_ref(handle, ea_inode);
1548 if (EA_INODE_CACHE(inode))
1549 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1550 ea_inode->i_ino, true /* reusable */);
1552 *ret_inode = ea_inode;
1557 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1558 * feature is enabled.
1560 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1562 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1563 struct ext4_xattr_search *s,
1564 handle_t *handle, struct inode *inode,
1567 struct ext4_xattr_entry *last, *next;
1568 struct ext4_xattr_entry *here = s->here;
1569 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1570 int in_inode = i->in_inode;
1571 struct inode *old_ea_inode = NULL;
1572 struct inode *new_ea_inode = NULL;
1573 size_t old_size, new_size;
1576 /* Space used by old and new values. */
1577 old_size = (!s->not_found && !here->e_value_inum) ?
1578 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1579 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1582 * Optimization for the simple case when old and new values have the
1583 * same padded sizes. Not applicable if external inodes are involved.
1585 if (new_size && new_size == old_size) {
1586 size_t offs = le16_to_cpu(here->e_value_offs);
1587 void *val = s->base + offs;
1589 here->e_value_size = cpu_to_le32(i->value_len);
1590 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1591 memset(val, 0, new_size);
1593 memcpy(val, i->value, i->value_len);
1594 /* Clear padding bytes. */
1595 memset(val + i->value_len, 0, new_size - i->value_len);
1600 /* Compute min_offs and last. */
1602 for (; !IS_LAST_ENTRY(last); last = next) {
1603 next = EXT4_XATTR_NEXT(last);
1604 if ((void *)next >= s->end) {
1605 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1606 ret = -EFSCORRUPTED;
1609 if (!last->e_value_inum && last->e_value_size) {
1610 size_t offs = le16_to_cpu(last->e_value_offs);
1611 if (offs < min_offs)
1616 /* Check whether we have enough space. */
1620 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1622 free += EXT4_XATTR_LEN(name_len) + old_size;
1624 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1630 * If storing the value in an external inode is an option,
1631 * reserve space for xattr entries/names in the external
1632 * attribute block so that a long value does not occupy the
1633 * whole space and prevent further entries being added.
1635 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1636 new_size && is_block &&
1637 (min_offs + old_size - new_size) <
1638 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1645 * Getting access to old and new ea inodes is subject to failures.
1646 * Finish that work before doing any modifications to the xattr data.
1648 if (!s->not_found && here->e_value_inum) {
1649 ret = ext4_xattr_inode_iget(inode,
1650 le32_to_cpu(here->e_value_inum),
1651 le32_to_cpu(here->e_hash),
1654 old_ea_inode = NULL;
1658 if (i->value && in_inode) {
1659 WARN_ON_ONCE(!i->value_len);
1661 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1665 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1669 new_ea_inode = NULL;
1670 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1676 /* We are ready to release ref count on the old_ea_inode. */
1677 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1679 /* Release newly required ref count on new_ea_inode. */
1683 err = ext4_xattr_inode_dec_ref(handle,
1686 ext4_warning_inode(new_ea_inode,
1687 "dec ref new_ea_inode err=%d",
1689 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1695 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1696 le32_to_cpu(here->e_value_size));
1699 /* No failures allowed past this point. */
1701 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1702 /* Remove the old value. */
1703 void *first_val = s->base + min_offs;
1704 size_t offs = le16_to_cpu(here->e_value_offs);
1705 void *val = s->base + offs;
1707 memmove(first_val + old_size, first_val, val - first_val);
1708 memset(first_val, 0, old_size);
1709 min_offs += old_size;
1711 /* Adjust all value offsets. */
1713 while (!IS_LAST_ENTRY(last)) {
1714 size_t o = le16_to_cpu(last->e_value_offs);
1716 if (!last->e_value_inum &&
1717 last->e_value_size && o < offs)
1718 last->e_value_offs = cpu_to_le16(o + old_size);
1719 last = EXT4_XATTR_NEXT(last);
1724 /* Remove old name. */
1725 size_t size = EXT4_XATTR_LEN(name_len);
1727 last = ENTRY((void *)last - size);
1728 memmove(here, (void *)here + size,
1729 (void *)last - (void *)here + sizeof(__u32));
1730 memset(last, 0, size);
1731 } else if (s->not_found) {
1732 /* Insert new name. */
1733 size_t size = EXT4_XATTR_LEN(name_len);
1734 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1736 memmove((void *)here + size, here, rest);
1737 memset(here, 0, size);
1738 here->e_name_index = i->name_index;
1739 here->e_name_len = name_len;
1740 memcpy(here->e_name, i->name, name_len);
1742 /* This is an update, reset value info. */
1743 here->e_value_inum = 0;
1744 here->e_value_offs = 0;
1745 here->e_value_size = 0;
1749 /* Insert new value. */
1751 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1752 } else if (i->value_len) {
1753 void *val = s->base + min_offs - new_size;
1755 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1756 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1757 memset(val, 0, new_size);
1759 memcpy(val, i->value, i->value_len);
1760 /* Clear padding bytes. */
1761 memset(val + i->value_len, 0,
1762 new_size - i->value_len);
1765 here->e_value_size = cpu_to_le32(i->value_len);
1772 /* Entry hash calculation. */
1777 * Feed crc32c hash instead of the raw value for entry
1778 * hash calculation. This is to avoid walking
1779 * potentially long value buffer again.
1781 crc32c_hash = cpu_to_le32(
1782 ext4_xattr_inode_get_hash(new_ea_inode));
1783 hash = ext4_xattr_hash_entry(here->e_name,
1786 } else if (is_block) {
1787 __le32 *value = s->base + le16_to_cpu(
1788 here->e_value_offs);
1790 hash = ext4_xattr_hash_entry(here->e_name,
1791 here->e_name_len, value,
1794 here->e_hash = hash;
1798 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1807 struct ext4_xattr_block_find {
1808 struct ext4_xattr_search s;
1809 struct buffer_head *bh;
1813 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1814 struct ext4_xattr_block_find *bs)
1816 struct super_block *sb = inode->i_sb;
1819 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1820 i->name_index, i->name, i->value, (long)i->value_len);
1822 if (EXT4_I(inode)->i_file_acl) {
1823 /* The inode already has an extended attribute block. */
1824 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1825 if (IS_ERR(bs->bh)) {
1826 error = PTR_ERR(bs->bh);
1830 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1831 atomic_read(&(bs->bh->b_count)),
1832 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1833 error = ext4_xattr_check_block(inode, bs->bh);
1836 /* Find the named attribute. */
1837 bs->s.base = BHDR(bs->bh);
1838 bs->s.first = BFIRST(bs->bh);
1839 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1840 bs->s.here = bs->s.first;
1841 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1842 i->name_index, i->name, 1);
1843 if (error && error != -ENODATA)
1845 bs->s.not_found = error;
1851 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1852 struct ext4_xattr_info *i,
1853 struct ext4_xattr_block_find *bs)
1855 struct super_block *sb = inode->i_sb;
1856 struct buffer_head *new_bh = NULL;
1857 struct ext4_xattr_search s_copy = bs->s;
1858 struct ext4_xattr_search *s = &s_copy;
1859 struct mb_cache_entry *ce = NULL;
1861 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1862 struct inode *ea_inode = NULL, *tmp_inode;
1863 size_t old_ea_inode_quota = 0;
1864 unsigned int ea_ino;
1867 #define header(x) ((struct ext4_xattr_header *)(x))
1870 int offset = (char *)s->here - bs->bh->b_data;
1872 BUFFER_TRACE(bs->bh, "get_write_access");
1873 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1877 lock_buffer(bs->bh);
1879 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1880 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1883 * This must happen under buffer lock for
1884 * ext4_xattr_block_set() to reliably detect modified
1887 if (ea_block_cache) {
1888 struct mb_cache_entry *oe;
1890 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1891 hash, bs->bh->b_blocknr);
1894 * Xattr block is getting reused. Leave
1897 mb_cache_entry_put(ea_block_cache, oe);
1901 ea_bdebug(bs->bh, "modifying in-place");
1902 error = ext4_xattr_set_entry(i, s, handle, inode,
1903 true /* is_block */);
1904 ext4_xattr_block_csum_set(inode, bs->bh);
1905 unlock_buffer(bs->bh);
1906 if (error == -EFSCORRUPTED)
1909 error = ext4_handle_dirty_metadata(handle,
1917 unlock_buffer(bs->bh);
1918 ea_bdebug(bs->bh, "cloning");
1919 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1921 if (s->base == NULL)
1923 s->first = ENTRY(header(s->base)+1);
1924 header(s->base)->h_refcount = cpu_to_le32(1);
1925 s->here = ENTRY(s->base + offset);
1926 s->end = s->base + bs->bh->b_size;
1929 * If existing entry points to an xattr inode, we need
1930 * to prevent ext4_xattr_set_entry() from decrementing
1931 * ref count on it because the reference belongs to the
1932 * original block. In this case, make the entry look
1933 * like it has an empty value.
1935 if (!s->not_found && s->here->e_value_inum) {
1936 ea_ino = le32_to_cpu(s->here->e_value_inum);
1937 error = ext4_xattr_inode_iget(inode, ea_ino,
1938 le32_to_cpu(s->here->e_hash),
1943 if (!ext4_test_inode_state(tmp_inode,
1944 EXT4_STATE_LUSTRE_EA_INODE)) {
1946 * Defer quota free call for previous
1947 * inode until success is guaranteed.
1949 old_ea_inode_quota = le32_to_cpu(
1950 s->here->e_value_size);
1954 s->here->e_value_inum = 0;
1955 s->here->e_value_size = 0;
1958 /* Allocate a buffer where we construct the new block. */
1959 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1961 if (s->base == NULL)
1963 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1964 header(s->base)->h_blocks = cpu_to_le32(1);
1965 header(s->base)->h_refcount = cpu_to_le32(1);
1966 s->first = ENTRY(header(s->base)+1);
1967 s->here = ENTRY(header(s->base)+1);
1968 s->end = s->base + sb->s_blocksize;
1971 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1972 if (error == -EFSCORRUPTED)
1977 if (i->value && s->here->e_value_inum) {
1979 * A ref count on ea_inode has been taken as part of the call to
1980 * ext4_xattr_set_entry() above. We would like to drop this
1981 * extra ref but we have to wait until the xattr block is
1982 * initialized and has its own ref count on the ea_inode.
1984 ea_ino = le32_to_cpu(s->here->e_value_inum);
1985 error = ext4_xattr_inode_iget(inode, ea_ino,
1986 le32_to_cpu(s->here->e_hash),
1995 if (!IS_LAST_ENTRY(s->first)) {
1996 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1999 /* We found an identical block in the cache. */
2000 if (new_bh == bs->bh)
2001 ea_bdebug(new_bh, "keeping");
2005 WARN_ON_ONCE(dquot_initialize_needed(inode));
2007 /* The old block is released after updating
2009 error = dquot_alloc_block(inode,
2010 EXT4_C2B(EXT4_SB(sb), 1));
2013 BUFFER_TRACE(new_bh, "get_write_access");
2014 error = ext4_journal_get_write_access(
2019 lock_buffer(new_bh);
2021 * We have to be careful about races with
2022 * adding references to xattr block. Once we
2023 * hold buffer lock xattr block's state is
2024 * stable so we can check the additional
2027 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2028 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2030 * Undo everything and check mbcache
2033 unlock_buffer(new_bh);
2034 dquot_free_block(inode,
2035 EXT4_C2B(EXT4_SB(sb),
2038 mb_cache_entry_put(ea_block_cache, ce);
2043 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2044 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2046 ea_bdebug(new_bh, "reusing; refcount now=%d",
2048 ext4_xattr_block_csum_set(inode, new_bh);
2049 unlock_buffer(new_bh);
2050 error = ext4_handle_dirty_metadata(handle,
2056 mb_cache_entry_touch(ea_block_cache, ce);
2057 mb_cache_entry_put(ea_block_cache, ce);
2059 } else if (bs->bh && s->base == bs->bh->b_data) {
2060 /* We were modifying this block in-place. */
2061 ea_bdebug(bs->bh, "keeping this block");
2062 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2066 /* We need to allocate a new block */
2067 ext4_fsblk_t goal, block;
2069 WARN_ON_ONCE(dquot_initialize_needed(inode));
2071 goal = ext4_group_first_block_no(sb,
2072 EXT4_I(inode)->i_block_group);
2074 /* non-extent files can't have physical blocks past 2^32 */
2075 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2076 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2078 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2083 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2084 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2086 ea_idebug(inode, "creating block %llu",
2087 (unsigned long long)block);
2089 new_bh = sb_getblk(sb, block);
2090 if (unlikely(!new_bh)) {
2093 ext4_free_blocks(handle, inode, NULL, block, 1,
2094 EXT4_FREE_BLOCKS_METADATA);
2097 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2098 ENTRY(header(s->base)+1));
2102 /* Drop the extra ref on ea_inode. */
2103 error = ext4_xattr_inode_dec_ref(handle,
2106 ext4_warning_inode(ea_inode,
2113 lock_buffer(new_bh);
2114 error = ext4_journal_get_create_access(handle, sb,
2115 new_bh, EXT4_JTR_NONE);
2117 unlock_buffer(new_bh);
2121 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2122 ext4_xattr_block_csum_set(inode, new_bh);
2123 set_buffer_uptodate(new_bh);
2124 unlock_buffer(new_bh);
2125 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2126 error = ext4_handle_dirty_metadata(handle, inode,
2133 if (old_ea_inode_quota)
2134 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2136 /* Update the inode. */
2137 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2139 /* Drop the previous xattr block. */
2140 if (bs->bh && bs->bh != new_bh) {
2141 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2143 ext4_xattr_release_block(handle, inode, bs->bh,
2145 0 /* extra_credits */);
2146 ext4_xattr_inode_array_free(ea_inode_array);
2154 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2156 ext4_warning_inode(ea_inode, "dec ref error=%d",
2159 /* If there was an error, revert the quota charge. */
2161 ext4_xattr_inode_free_quota(inode, ea_inode,
2162 i_size_read(ea_inode));
2166 mb_cache_entry_put(ea_block_cache, ce);
2168 if (!(bs->bh && s->base == bs->bh->b_data))
2174 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2178 EXT4_ERROR_INODE(inode, "bad block %llu",
2179 EXT4_I(inode)->i_file_acl);
2185 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2186 struct ext4_xattr_ibody_find *is)
2188 struct ext4_xattr_ibody_header *header;
2189 struct ext4_inode *raw_inode;
2192 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2195 raw_inode = ext4_raw_inode(&is->iloc);
2196 header = IHDR(inode, raw_inode);
2197 is->s.base = is->s.first = IFIRST(header);
2198 is->s.here = is->s.first;
2199 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2200 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2201 error = xattr_check_inode(inode, header, is->s.end);
2204 /* Find the named attribute. */
2205 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2206 i->name_index, i->name, 0);
2207 if (error && error != -ENODATA)
2209 is->s.not_found = error;
2214 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2215 struct ext4_xattr_info *i,
2216 struct ext4_xattr_ibody_find *is)
2218 struct ext4_xattr_ibody_header *header;
2219 struct ext4_xattr_search *s = &is->s;
2222 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2225 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2228 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2229 if (!IS_LAST_ENTRY(s->first)) {
2230 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2231 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2233 header->h_magic = cpu_to_le32(0);
2234 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2239 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2240 struct ext4_xattr_info *i)
2244 /* When e_value_inum is set the value is stored externally. */
2245 if (s->here->e_value_inum)
2247 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2249 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2250 return !memcmp(value, i->value, i->value_len);
2253 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2255 struct buffer_head *bh;
2258 if (!EXT4_I(inode)->i_file_acl)
2260 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2263 error = ext4_xattr_check_block(inode, bh);
2266 return ERR_PTR(error);
2272 * ext4_xattr_set_handle()
2274 * Create, replace or remove an extended attribute for this inode. Value
2275 * is NULL to remove an existing extended attribute, and non-NULL to
2276 * either replace an existing extended attribute, or create a new extended
2277 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2278 * specify that an extended attribute must exist and must not exist
2279 * previous to the call, respectively.
2281 * Returns 0, or a negative error number on failure.
2284 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2285 const char *name, const void *value, size_t value_len,
2288 struct ext4_xattr_info i = {
2289 .name_index = name_index,
2292 .value_len = value_len,
2295 struct ext4_xattr_ibody_find is = {
2296 .s = { .not_found = -ENODATA, },
2298 struct ext4_xattr_block_find bs = {
2299 .s = { .not_found = -ENODATA, },
2306 if (strlen(name) > 255)
2309 ext4_write_lock_xattr(inode, &no_expand);
2311 /* Check journal credits under write lock. */
2312 if (ext4_handle_valid(handle)) {
2313 struct buffer_head *bh;
2316 bh = ext4_xattr_get_block(inode);
2318 error = PTR_ERR(bh);
2322 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2324 flags & XATTR_CREATE);
2327 if (jbd2_handle_buffer_credits(handle) < credits) {
2331 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2334 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2338 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2339 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2340 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2341 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2344 error = ext4_xattr_ibody_find(inode, &i, &is);
2348 error = ext4_xattr_block_find(inode, &i, &bs);
2351 if (is.s.not_found && bs.s.not_found) {
2353 if (flags & XATTR_REPLACE)
2360 if (flags & XATTR_CREATE)
2365 if (!is.s.not_found)
2366 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2367 else if (!bs.s.not_found)
2368 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2371 /* Xattr value did not change? Save us some work and bail out */
2372 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2374 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2377 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2378 (EXT4_XATTR_SIZE(i.value_len) >
2379 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2382 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2383 if (!error && !bs.s.not_found) {
2385 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2386 } else if (error == -ENOSPC) {
2387 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2390 error = ext4_xattr_block_find(inode, &i, &bs);
2394 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2395 if (!error && !is.s.not_found) {
2397 error = ext4_xattr_ibody_set(handle, inode, &i,
2399 } else if (error == -ENOSPC) {
2401 * Xattr does not fit in the block, store at
2402 * external inode if possible.
2404 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2405 i.value_len && !i.in_inode) {
2413 ext4_xattr_update_super_block(handle, inode->i_sb);
2414 inode->i_ctime = current_time(inode);
2415 inode_inc_iversion(inode);
2418 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2420 * The bh is consumed by ext4_mark_iloc_dirty, even with
2425 ext4_handle_sync(handle);
2427 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2432 ext4_write_unlock_xattr(inode, &no_expand);
2436 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2437 bool is_create, int *credits)
2439 struct buffer_head *bh;
2444 if (!EXT4_SB(inode->i_sb)->s_journal)
2447 down_read(&EXT4_I(inode)->xattr_sem);
2449 bh = ext4_xattr_get_block(inode);
2453 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2454 value_len, is_create);
2459 up_read(&EXT4_I(inode)->xattr_sem);
2466 * Like ext4_xattr_set_handle, but start from an inode. This extended
2467 * attribute modification is a filesystem transaction by itself.
2469 * Returns 0, or a negative error number on failure.
2472 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2473 const void *value, size_t value_len, int flags)
2476 struct super_block *sb = inode->i_sb;
2477 int error, retries = 0;
2480 error = dquot_initialize(inode);
2485 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2490 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2491 if (IS_ERR(handle)) {
2492 error = PTR_ERR(handle);
2496 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2497 value, value_len, flags);
2498 error2 = ext4_journal_stop(handle);
2499 if (error == -ENOSPC &&
2500 ext4_should_retry_alloc(sb, &retries))
2505 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2511 * Shift the EA entries in the inode to create space for the increased
2514 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2515 int value_offs_shift, void *to,
2516 void *from, size_t n)
2518 struct ext4_xattr_entry *last = entry;
2521 /* We always shift xattr headers further thus offsets get lower */
2522 BUG_ON(value_offs_shift > 0);
2524 /* Adjust the value offsets of the entries */
2525 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2526 if (!last->e_value_inum && last->e_value_size) {
2527 new_offs = le16_to_cpu(last->e_value_offs) +
2529 last->e_value_offs = cpu_to_le16(new_offs);
2532 /* Shift the entries by n bytes */
2533 memmove(to, from, n);
2537 * Move xattr pointed to by 'entry' from inode into external xattr block
2539 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2540 struct ext4_inode *raw_inode,
2541 struct ext4_xattr_entry *entry)
2543 struct ext4_xattr_ibody_find *is = NULL;
2544 struct ext4_xattr_block_find *bs = NULL;
2545 char *buffer = NULL, *b_entry_name = NULL;
2546 size_t value_size = le32_to_cpu(entry->e_value_size);
2547 struct ext4_xattr_info i = {
2550 .name_index = entry->e_name_index,
2551 .in_inode = !!entry->e_value_inum,
2553 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2556 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2557 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2558 buffer = kmalloc(value_size, GFP_NOFS);
2559 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2560 if (!is || !bs || !buffer || !b_entry_name) {
2565 is->s.not_found = -ENODATA;
2566 bs->s.not_found = -ENODATA;
2570 /* Save the entry name and the entry value */
2571 if (entry->e_value_inum) {
2572 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2576 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2577 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2580 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2581 b_entry_name[entry->e_name_len] = '\0';
2582 i.name = b_entry_name;
2584 error = ext4_get_inode_loc(inode, &is->iloc);
2588 error = ext4_xattr_ibody_find(inode, &i, is);
2592 /* Remove the chosen entry from the inode */
2593 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2598 i.value_len = value_size;
2599 error = ext4_xattr_block_find(inode, &i, bs);
2603 /* Add entry which was removed from the inode into the block */
2604 error = ext4_xattr_block_set(handle, inode, &i, bs);
2609 kfree(b_entry_name);
2612 brelse(is->iloc.bh);
2621 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2622 struct ext4_inode *raw_inode,
2623 int isize_diff, size_t ifree,
2624 size_t bfree, int *total_ino)
2626 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2627 struct ext4_xattr_entry *small_entry;
2628 struct ext4_xattr_entry *entry;
2629 struct ext4_xattr_entry *last;
2630 unsigned int entry_size; /* EA entry size */
2631 unsigned int total_size; /* EA entry size + value size */
2632 unsigned int min_total_size;
2635 while (isize_diff > ifree) {
2638 min_total_size = ~0U;
2639 last = IFIRST(header);
2640 /* Find the entry best suited to be pushed into EA block */
2641 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2642 /* never move system.data out of the inode */
2643 if ((last->e_name_len == 4) &&
2644 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2645 !memcmp(last->e_name, "data", 4))
2647 total_size = EXT4_XATTR_LEN(last->e_name_len);
2648 if (!last->e_value_inum)
2649 total_size += EXT4_XATTR_SIZE(
2650 le32_to_cpu(last->e_value_size));
2651 if (total_size <= bfree &&
2652 total_size < min_total_size) {
2653 if (total_size + ifree < isize_diff) {
2657 min_total_size = total_size;
2662 if (entry == NULL) {
2663 if (small_entry == NULL)
2665 entry = small_entry;
2668 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2669 total_size = entry_size;
2670 if (!entry->e_value_inum)
2671 total_size += EXT4_XATTR_SIZE(
2672 le32_to_cpu(entry->e_value_size));
2673 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2678 *total_ino -= entry_size;
2679 ifree += total_size;
2680 bfree -= total_size;
2687 * Expand an inode by new_extra_isize bytes when EAs are present.
2688 * Returns 0 on success or negative error number on failure.
2690 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2691 struct ext4_inode *raw_inode, handle_t *handle)
2693 struct ext4_xattr_ibody_header *header;
2694 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2695 static unsigned int mnt_count;
2697 size_t ifree, bfree;
2700 int error = 0, tried_min_extra_isize = 0;
2701 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2702 int isize_diff; /* How much do we need to grow i_extra_isize */
2705 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2706 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2709 header = IHDR(inode, raw_inode);
2712 * Check if enough free space is available in the inode to shift the
2713 * entries ahead by new_extra_isize.
2716 base = IFIRST(header);
2717 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2718 min_offs = end - base;
2719 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2721 error = xattr_check_inode(inode, header, end);
2725 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2726 if (ifree >= isize_diff)
2730 * Enough free space isn't available in the inode, check if
2731 * EA block can hold new_extra_isize bytes.
2733 if (EXT4_I(inode)->i_file_acl) {
2734 struct buffer_head *bh;
2736 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2738 error = PTR_ERR(bh);
2741 error = ext4_xattr_check_block(inode, bh);
2747 end = bh->b_data + bh->b_size;
2748 min_offs = end - base;
2749 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2752 if (bfree + ifree < isize_diff) {
2753 if (!tried_min_extra_isize && s_min_extra_isize) {
2754 tried_min_extra_isize++;
2755 new_extra_isize = s_min_extra_isize;
2762 bfree = inode->i_sb->s_blocksize;
2765 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2766 isize_diff, ifree, bfree,
2769 if (error == -ENOSPC && !tried_min_extra_isize &&
2770 s_min_extra_isize) {
2771 tried_min_extra_isize++;
2772 new_extra_isize = s_min_extra_isize;
2778 /* Adjust the offsets and shift the remaining entries ahead */
2779 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2780 - new_extra_isize, (void *)raw_inode +
2781 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2782 (void *)header, total_ino);
2783 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2786 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2787 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2789 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2794 #define EIA_INCR 16 /* must be 2^n */
2795 #define EIA_MASK (EIA_INCR - 1)
2797 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2798 * If @ea_inode_array is new or full it will be grown and the old
2799 * contents copied over.
2802 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2803 struct inode *inode)
2805 if (*ea_inode_array == NULL) {
2807 * Start with 15 inodes, so it fits into a power-of-two size.
2808 * If *ea_inode_array is NULL, this is essentially offsetof()
2811 kmalloc(offsetof(struct ext4_xattr_inode_array,
2814 if (*ea_inode_array == NULL)
2816 (*ea_inode_array)->count = 0;
2817 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2818 /* expand the array once all 15 + n * 16 slots are full */
2819 struct ext4_xattr_inode_array *new_array = NULL;
2820 int count = (*ea_inode_array)->count;
2822 /* if new_array is NULL, this is essentially offsetof() */
2823 new_array = kmalloc(
2824 offsetof(struct ext4_xattr_inode_array,
2825 inodes[count + EIA_INCR]),
2827 if (new_array == NULL)
2829 memcpy(new_array, *ea_inode_array,
2830 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2831 kfree(*ea_inode_array);
2832 *ea_inode_array = new_array;
2834 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2839 * ext4_xattr_delete_inode()
2841 * Free extended attribute resources associated with this inode. Traverse
2842 * all entries and decrement reference on any xattr inodes associated with this
2843 * inode. This is called immediately before an inode is freed. We have exclusive
2844 * access to the inode. If an orphan inode is deleted it will also release its
2845 * references on xattr block and xattr inodes.
2847 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2848 struct ext4_xattr_inode_array **ea_inode_array,
2851 struct buffer_head *bh = NULL;
2852 struct ext4_xattr_ibody_header *header;
2853 struct ext4_iloc iloc = { .bh = NULL };
2854 struct ext4_xattr_entry *entry;
2855 struct inode *ea_inode;
2858 error = ext4_journal_ensure_credits(handle, extra_credits,
2859 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2861 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2865 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2866 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2868 error = ext4_get_inode_loc(inode, &iloc);
2870 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2874 error = ext4_journal_get_write_access(handle, inode->i_sb,
2875 iloc.bh, EXT4_JTR_NONE);
2877 EXT4_ERROR_INODE(inode, "write access (error %d)",
2882 header = IHDR(inode, ext4_raw_inode(&iloc));
2883 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2884 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2886 false /* block_csum */,
2889 false /* skip_quota */);
2892 if (EXT4_I(inode)->i_file_acl) {
2893 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2895 error = PTR_ERR(bh);
2896 if (error == -EIO) {
2897 EXT4_ERROR_INODE_ERR(inode, EIO,
2898 "block %llu read error",
2899 EXT4_I(inode)->i_file_acl);
2904 error = ext4_xattr_check_block(inode, bh);
2908 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2909 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2910 entry = EXT4_XATTR_NEXT(entry)) {
2911 if (!entry->e_value_inum)
2913 error = ext4_xattr_inode_iget(inode,
2914 le32_to_cpu(entry->e_value_inum),
2915 le32_to_cpu(entry->e_hash),
2919 ext4_xattr_inode_free_quota(inode, ea_inode,
2920 le32_to_cpu(entry->e_value_size));
2926 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2929 * Update i_file_acl value in the same transaction that releases
2932 EXT4_I(inode)->i_file_acl = 0;
2933 error = ext4_mark_inode_dirty(handle, inode);
2935 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2939 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2948 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2952 if (ea_inode_array == NULL)
2955 for (idx = 0; idx < ea_inode_array->count; ++idx)
2956 iput(ea_inode_array->inodes[idx]);
2957 kfree(ea_inode_array);
2961 * ext4_xattr_block_cache_insert()
2963 * Create a new entry in the extended attribute block cache, and insert
2964 * it unless such an entry is already in the cache.
2966 * Returns 0, or a negative error number on failure.
2969 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2970 struct buffer_head *bh)
2972 struct ext4_xattr_header *header = BHDR(bh);
2973 __u32 hash = le32_to_cpu(header->h_hash);
2974 int reusable = le32_to_cpu(header->h_refcount) <
2975 EXT4_XATTR_REFCOUNT_MAX;
2978 if (!ea_block_cache)
2980 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2981 bh->b_blocknr, reusable);
2983 if (error == -EBUSY)
2984 ea_bdebug(bh, "already in cache");
2986 ea_bdebug(bh, "inserting [%x]", (int)hash);
2992 * Compare two extended attribute blocks for equality.
2994 * Returns 0 if the blocks are equal, 1 if they differ, and
2995 * a negative error number on errors.
2998 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2999 struct ext4_xattr_header *header2)
3001 struct ext4_xattr_entry *entry1, *entry2;
3003 entry1 = ENTRY(header1+1);
3004 entry2 = ENTRY(header2+1);
3005 while (!IS_LAST_ENTRY(entry1)) {
3006 if (IS_LAST_ENTRY(entry2))
3008 if (entry1->e_hash != entry2->e_hash ||
3009 entry1->e_name_index != entry2->e_name_index ||
3010 entry1->e_name_len != entry2->e_name_len ||
3011 entry1->e_value_size != entry2->e_value_size ||
3012 entry1->e_value_inum != entry2->e_value_inum ||
3013 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3015 if (!entry1->e_value_inum &&
3016 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3017 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3018 le32_to_cpu(entry1->e_value_size)))
3021 entry1 = EXT4_XATTR_NEXT(entry1);
3022 entry2 = EXT4_XATTR_NEXT(entry2);
3024 if (!IS_LAST_ENTRY(entry2))
3030 * ext4_xattr_block_cache_find()
3032 * Find an identical extended attribute block.
3034 * Returns a pointer to the block found, or NULL if such a block was
3035 * not found or an error occurred.
3037 static struct buffer_head *
3038 ext4_xattr_block_cache_find(struct inode *inode,
3039 struct ext4_xattr_header *header,
3040 struct mb_cache_entry **pce)
3042 __u32 hash = le32_to_cpu(header->h_hash);
3043 struct mb_cache_entry *ce;
3044 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3046 if (!ea_block_cache)
3048 if (!header->h_hash)
3049 return NULL; /* never share */
3050 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3051 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3053 struct buffer_head *bh;
3055 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3057 if (PTR_ERR(bh) == -ENOMEM)
3060 EXT4_ERROR_INODE(inode, "block %lu read error",
3061 (unsigned long)ce->e_value);
3062 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3067 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3072 #define NAME_HASH_SHIFT 5
3073 #define VALUE_HASH_SHIFT 16
3076 * ext4_xattr_hash_entry()
3078 * Compute the hash of an extended attribute.
3080 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3085 while (name_len--) {
3086 hash = (hash << NAME_HASH_SHIFT) ^
3087 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3090 while (value_count--) {
3091 hash = (hash << VALUE_HASH_SHIFT) ^
3092 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3093 le32_to_cpu(*value++);
3095 return cpu_to_le32(hash);
3098 #undef NAME_HASH_SHIFT
3099 #undef VALUE_HASH_SHIFT
3101 #define BLOCK_HASH_SHIFT 16
3104 * ext4_xattr_rehash()
3106 * Re-compute the extended attribute hash value after an entry has changed.
3108 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3110 struct ext4_xattr_entry *here;
3113 here = ENTRY(header+1);
3114 while (!IS_LAST_ENTRY(here)) {
3115 if (!here->e_hash) {
3116 /* Block is not shared if an entry's hash value == 0 */
3120 hash = (hash << BLOCK_HASH_SHIFT) ^
3121 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3122 le32_to_cpu(here->e_hash);
3123 here = EXT4_XATTR_NEXT(here);
3125 header->h_hash = cpu_to_le32(hash);
3128 #undef BLOCK_HASH_SHIFT
3130 #define HASH_BUCKET_BITS 10
3133 ext4_xattr_create_cache(void)
3135 return mb_cache_create(HASH_BUCKET_BITS);
3138 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3141 mb_cache_destroy(cache);