1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/xattr.c
5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9 * Extended attributes for symlinks and special files added per
10 * suggestion of Luka Renko <luka.renko@hermes.si>.
11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14 * and Andreas Gruenbacher <agruen@suse.de>.
18 * Extended attributes are stored directly in inodes (on file systems with
19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
20 * field contains the block number if an inode uses an additional block. All
21 * attributes must fit in the inode and one additional block. Blocks that
22 * contain the identical set of attributes may be shared among several inodes.
23 * Identical blocks are detected by keeping a cache of blocks that have
24 * recently been accessed.
26 * The attributes in inodes and on blocks have a different header; the entries
27 * are stored in the same format:
29 * +------------------+
32 * | entry 2 | | growing downwards
37 * | value 3 | | growing upwards
39 * +------------------+
41 * The header is followed by multiple entry descriptors. In disk blocks, the
42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
43 * attribute values are aligned to the end of the block in no specific order.
47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
48 * EA blocks are only changed if they are exclusive to an inode, so
49 * holding xattr_sem also means that nothing but the EA block's reference
50 * count can change. Multiple writers to the same block are synchronized
54 #include <linux/init.h>
56 #include <linux/slab.h>
57 #include <linux/mbcache.h>
58 #include <linux/quotaops.h>
59 #include <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;
196 /* Check the values */
197 while (!IS_LAST_ENTRY(entry)) {
198 u32 size = le32_to_cpu(entry->e_value_size);
200 if (size > EXT4_XATTR_SIZE_MAX)
201 return -EFSCORRUPTED;
203 if (size != 0 && entry->e_value_inum == 0) {
204 u16 offs = le16_to_cpu(entry->e_value_offs);
208 * The value cannot overlap the names, and the value
209 * with padding cannot extend beyond 'end'. Check both
210 * the padded and unpadded sizes, since the size may
211 * overflow to 0 when adding padding.
213 if (offs > end - value_start)
214 return -EFSCORRUPTED;
215 value = value_start + offs;
216 if (value < (void *)e + sizeof(u32) ||
217 size > end - value ||
218 EXT4_XATTR_SIZE(size) > end - value)
219 return -EFSCORRUPTED;
221 entry = EXT4_XATTR_NEXT(entry);
228 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
229 const char *function, unsigned int line)
231 int error = -EFSCORRUPTED;
233 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
234 BHDR(bh)->h_blocks != cpu_to_le32(1))
236 if (buffer_verified(bh))
240 if (!ext4_xattr_block_csum_verify(inode, bh))
242 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
246 __ext4_error_inode(inode, function, line, 0,
247 "corrupted xattr block %llu",
248 (unsigned long long) bh->b_blocknr);
250 set_buffer_verified(bh);
254 #define ext4_xattr_check_block(inode, bh) \
255 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
259 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
260 void *end, const char *function, unsigned int line)
262 int error = -EFSCORRUPTED;
264 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
265 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
267 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
270 __ext4_error_inode(inode, function, line, 0,
271 "corrupted in-inode xattr");
275 #define xattr_check_inode(inode, header, end) \
276 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
279 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
280 void *end, int name_index, const char *name, int sorted)
282 struct ext4_xattr_entry *entry, *next;
288 name_len = strlen(name);
289 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
290 next = EXT4_XATTR_NEXT(entry);
291 if ((void *) next >= end) {
292 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
293 return -EFSCORRUPTED;
295 cmp = name_index - entry->e_name_index;
297 cmp = name_len - entry->e_name_len;
299 cmp = memcmp(name, entry->e_name, name_len);
300 if (cmp <= 0 && (sorted || cmp == 0))
304 return cmp ? -ENODATA : 0;
308 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
310 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
313 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
315 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
316 (u32) inode_peek_iversion_raw(ea_inode);
319 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
321 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
322 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
325 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
327 return (u32)ea_inode->i_atime.tv_sec;
330 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
332 ea_inode->i_atime.tv_sec = hash;
336 * Read the EA value from an inode.
338 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
340 int blocksize = 1 << ea_inode->i_blkbits;
341 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
342 int tail_size = (size % blocksize) ?: blocksize;
343 struct buffer_head *bhs_inline[8];
344 struct buffer_head **bhs = bhs_inline;
347 if (bh_count > ARRAY_SIZE(bhs_inline)) {
348 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
353 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
354 true /* wait */, bhs);
358 for (i = 0; i < bh_count; i++) {
359 /* There shouldn't be any holes in ea_inode. */
364 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
365 i < bh_count - 1 ? blocksize : tail_size);
369 for (i = 0; i < bh_count; i++)
372 if (bhs != bhs_inline)
377 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
379 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
380 u32 ea_inode_hash, struct inode **ea_inode)
385 inode = ext4_iget(parent->i_sb, ea_ino);
387 err = PTR_ERR(inode);
388 ext4_error(parent->i_sb,
389 "error while reading EA inode %lu err=%d", ea_ino,
394 if (is_bad_inode(inode)) {
395 ext4_error(parent->i_sb,
396 "error while reading EA inode %lu is_bad_inode",
402 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
403 ext4_error(parent->i_sb,
404 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
410 ext4_xattr_inode_set_class(inode);
413 * Check whether this is an old Lustre-style xattr inode. Lustre
414 * implementation does not have hash validation, rather it has a
415 * backpointer from ea_inode to the parent inode.
417 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
418 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
419 inode->i_generation == parent->i_generation) {
420 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
421 ext4_xattr_inode_set_ref(inode, 1);
424 inode->i_flags |= S_NOQUOTA;
436 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
437 struct ext4_xattr_entry *entry, void *buffer,
442 /* Verify stored hash matches calculated hash. */
443 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
444 if (hash != ext4_xattr_inode_get_hash(ea_inode))
445 return -EFSCORRUPTED;
448 __le32 e_hash, tmp_data;
450 /* Verify entry hash. */
451 tmp_data = cpu_to_le32(hash);
452 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
454 if (e_hash != entry->e_hash)
455 return -EFSCORRUPTED;
461 * Read xattr value from the EA inode.
464 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
465 void *buffer, size_t size)
467 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
468 struct inode *ea_inode;
471 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
472 le32_to_cpu(entry->e_hash), &ea_inode);
478 if (i_size_read(ea_inode) != size) {
479 ext4_warning_inode(ea_inode,
480 "ea_inode file size=%llu entry size=%zu",
481 i_size_read(ea_inode), size);
486 err = ext4_xattr_inode_read(ea_inode, buffer, size);
490 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
491 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
494 ext4_warning_inode(ea_inode,
495 "EA inode hash validation failed");
500 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
501 ext4_xattr_inode_get_hash(ea_inode),
502 ea_inode->i_ino, true /* reusable */);
510 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
511 void *buffer, size_t buffer_size)
513 struct buffer_head *bh = NULL;
514 struct ext4_xattr_entry *entry;
518 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
520 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
521 name_index, name, buffer, (long)buffer_size);
524 if (!EXT4_I(inode)->i_file_acl)
526 ea_idebug(inode, "reading block %llu",
527 (unsigned long long)EXT4_I(inode)->i_file_acl);
528 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
531 ea_bdebug(bh, "b_count=%d, refcount=%d",
532 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
533 error = ext4_xattr_check_block(inode, bh);
536 ext4_xattr_block_cache_insert(ea_block_cache, bh);
538 end = bh->b_data + bh->b_size;
539 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
542 size = le32_to_cpu(entry->e_value_size);
544 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
547 if (size > buffer_size)
549 if (entry->e_value_inum) {
550 error = ext4_xattr_inode_get(inode, entry, buffer,
555 u16 offset = le16_to_cpu(entry->e_value_offs);
556 void *p = bh->b_data + offset;
558 if (unlikely(p + size > end))
560 memcpy(buffer, p, size);
571 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
572 void *buffer, size_t buffer_size)
574 struct ext4_xattr_ibody_header *header;
575 struct ext4_xattr_entry *entry;
576 struct ext4_inode *raw_inode;
577 struct ext4_iloc iloc;
582 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
584 error = ext4_get_inode_loc(inode, &iloc);
587 raw_inode = ext4_raw_inode(&iloc);
588 header = IHDR(inode, raw_inode);
589 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
590 error = xattr_check_inode(inode, header, end);
593 entry = IFIRST(header);
594 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
597 size = le32_to_cpu(entry->e_value_size);
599 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
602 if (size > buffer_size)
604 if (entry->e_value_inum) {
605 error = ext4_xattr_inode_get(inode, entry, buffer,
610 u16 offset = le16_to_cpu(entry->e_value_offs);
611 void *p = (void *)IFIRST(header) + offset;
613 if (unlikely(p + size > end))
615 memcpy(buffer, p, size);
628 * Copy an extended attribute into the buffer
629 * provided, or compute the buffer size required.
630 * Buffer is NULL to compute the size of the buffer required.
632 * Returns a negative error number on failure, or the number of bytes
633 * used / required on success.
636 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
637 void *buffer, size_t buffer_size)
641 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
644 if (strlen(name) > 255)
647 down_read(&EXT4_I(inode)->xattr_sem);
648 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
650 if (error == -ENODATA)
651 error = ext4_xattr_block_get(inode, name_index, name, buffer,
653 up_read(&EXT4_I(inode)->xattr_sem);
658 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
659 char *buffer, size_t buffer_size)
661 size_t rest = buffer_size;
663 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
664 const struct xattr_handler *handler =
665 ext4_xattr_handler(entry->e_name_index);
667 if (handler && (!handler->list || handler->list(dentry))) {
668 const char *prefix = handler->prefix ?: handler->name;
669 size_t prefix_len = strlen(prefix);
670 size_t size = prefix_len + entry->e_name_len + 1;
675 memcpy(buffer, prefix, prefix_len);
676 buffer += prefix_len;
677 memcpy(buffer, entry->e_name, entry->e_name_len);
678 buffer += entry->e_name_len;
684 return buffer_size - rest; /* total size */
688 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
690 struct inode *inode = d_inode(dentry);
691 struct buffer_head *bh = NULL;
694 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
695 buffer, (long)buffer_size);
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 = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
706 ea_bdebug(bh, "b_count=%d, refcount=%d",
707 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
708 error = ext4_xattr_check_block(inode, bh);
711 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
712 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
721 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
723 struct inode *inode = d_inode(dentry);
724 struct ext4_xattr_ibody_header *header;
725 struct ext4_inode *raw_inode;
726 struct ext4_iloc iloc;
730 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
732 error = ext4_get_inode_loc(inode, &iloc);
735 raw_inode = ext4_raw_inode(&iloc);
736 header = IHDR(inode, raw_inode);
737 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
738 error = xattr_check_inode(inode, header, end);
741 error = ext4_xattr_list_entries(dentry, IFIRST(header),
742 buffer, buffer_size);
750 * Inode operation listxattr()
752 * d_inode(dentry)->i_rwsem: don't care
754 * Copy a list of attribute names into the buffer
755 * provided, or compute the buffer size required.
756 * Buffer is NULL to compute the size of the buffer required.
758 * Returns a negative error number on failure, or the number of bytes
759 * used / required on success.
762 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
766 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
767 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
774 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
779 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
784 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
787 static void ext4_xattr_update_super_block(handle_t *handle,
788 struct super_block *sb)
790 if (ext4_has_feature_xattr(sb))
793 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
794 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
795 ext4_set_feature_xattr(sb);
796 ext4_handle_dirty_super(handle, sb);
800 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
802 struct ext4_iloc iloc = { .bh = NULL };
803 struct buffer_head *bh = NULL;
804 struct ext4_inode *raw_inode;
805 struct ext4_xattr_ibody_header *header;
806 struct ext4_xattr_entry *entry;
807 qsize_t ea_inode_refs = 0;
811 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
813 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
814 ret = ext4_get_inode_loc(inode, &iloc);
817 raw_inode = ext4_raw_inode(&iloc);
818 header = IHDR(inode, raw_inode);
819 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
820 ret = xattr_check_inode(inode, header, end);
824 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
825 entry = EXT4_XATTR_NEXT(entry))
826 if (entry->e_value_inum)
830 if (EXT4_I(inode)->i_file_acl) {
831 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
837 ret = ext4_xattr_check_block(inode, bh);
841 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
842 entry = EXT4_XATTR_NEXT(entry))
843 if (entry->e_value_inum)
846 *usage = ea_inode_refs + 1;
854 static inline size_t round_up_cluster(struct inode *inode, size_t length)
856 struct super_block *sb = inode->i_sb;
857 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
859 size_t mask = ~(cluster_size - 1);
861 return (length + cluster_size - 1) & mask;
864 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
868 err = dquot_alloc_inode(inode);
871 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
873 dquot_free_inode(inode);
877 static void ext4_xattr_inode_free_quota(struct inode *parent,
878 struct inode *ea_inode,
882 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
884 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
885 dquot_free_inode(parent);
888 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
889 struct buffer_head *block_bh, size_t value_len,
896 * 1) Owner inode update
897 * 2) Ref count update on old xattr block
899 * 4) block bitmap update for new xattr block
900 * 5) group descriptor for new xattr block
901 * 6) block bitmap update for old xattr block
902 * 7) group descriptor for old block
904 * 6 & 7 can happen if we have two racing threads T_a and T_b
905 * which are each trying to set an xattr on inodes I_a and I_b
906 * which were both initially sharing an xattr block.
911 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
914 * In case of inline data, we may push out the data to a block,
915 * so we need to reserve credits for this eventuality
917 if (inode && ext4_has_inline_data(inode))
918 credits += ext4_writepage_trans_blocks(inode) + 1;
920 /* We are done if ea_inode feature is not enabled. */
921 if (!ext4_has_feature_ea_inode(sb))
924 /* New ea_inode, inode map, block bitmap, group descriptor. */
928 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
930 /* Indirection block or one level of extent tree. */
933 /* Block bitmap and group descriptor updates for each block. */
934 credits += blocks * 2;
936 /* Blocks themselves. */
940 /* Dereference ea_inode holding old xattr value.
941 * Old ea_inode, inode map, block bitmap, group descriptor.
945 /* Data blocks for old ea_inode. */
946 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
948 /* Indirection block or one level of extent tree for old
953 /* Block bitmap and group descriptor updates for each block. */
954 credits += blocks * 2;
957 /* We may need to clone the existing xattr block in which case we need
958 * to increment ref counts for existing ea_inodes referenced by it.
961 struct ext4_xattr_entry *entry = BFIRST(block_bh);
963 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
964 if (entry->e_value_inum)
965 /* Ref count update on ea_inode. */
971 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
972 int credits, struct buffer_head *bh,
973 bool dirty, bool block_csum)
977 if (!ext4_handle_valid(handle))
980 if (handle->h_buffer_credits >= credits)
983 error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
987 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
993 ext4_xattr_block_csum_set(inode, bh);
994 error = ext4_handle_dirty_metadata(handle, NULL, bh);
996 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1002 error = ext4_journal_restart(handle, credits);
1004 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1009 error = ext4_journal_get_write_access(handle, bh);
1011 ext4_warning(inode->i_sb,
1012 "Get write access failed (error %d)",
1020 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1023 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1024 struct ext4_iloc iloc;
1029 inode_lock(ea_inode);
1031 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1037 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1038 ref_count += ref_change;
1039 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1041 if (ref_change > 0) {
1042 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1043 ea_inode->i_ino, ref_count);
1045 if (ref_count == 1) {
1046 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1047 ea_inode->i_ino, ea_inode->i_nlink);
1049 set_nlink(ea_inode, 1);
1050 ext4_orphan_del(handle, ea_inode);
1052 if (ea_inode_cache) {
1053 hash = ext4_xattr_inode_get_hash(ea_inode);
1054 mb_cache_entry_create(ea_inode_cache,
1057 true /* reusable */);
1061 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1062 ea_inode->i_ino, ref_count);
1064 if (ref_count == 0) {
1065 WARN_ONCE(ea_inode->i_nlink != 1,
1066 "EA inode %lu i_nlink=%u",
1067 ea_inode->i_ino, ea_inode->i_nlink);
1069 clear_nlink(ea_inode);
1070 ext4_orphan_add(handle, ea_inode);
1072 if (ea_inode_cache) {
1073 hash = ext4_xattr_inode_get_hash(ea_inode);
1074 mb_cache_entry_delete(ea_inode_cache, hash,
1080 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1083 ext4_warning_inode(ea_inode,
1084 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1087 inode_unlock(ea_inode);
1091 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1093 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1096 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1098 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1101 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1102 struct ext4_xattr_entry *first)
1104 struct inode *ea_inode;
1105 struct ext4_xattr_entry *entry;
1106 struct ext4_xattr_entry *failed_entry;
1107 unsigned int ea_ino;
1110 for (entry = first; !IS_LAST_ENTRY(entry);
1111 entry = EXT4_XATTR_NEXT(entry)) {
1112 if (!entry->e_value_inum)
1114 ea_ino = le32_to_cpu(entry->e_value_inum);
1115 err = ext4_xattr_inode_iget(parent, ea_ino,
1116 le32_to_cpu(entry->e_hash),
1120 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1122 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1132 failed_entry = entry;
1134 for (entry = first; entry != failed_entry;
1135 entry = EXT4_XATTR_NEXT(entry)) {
1136 if (!entry->e_value_inum)
1138 ea_ino = le32_to_cpu(entry->e_value_inum);
1139 err = ext4_xattr_inode_iget(parent, ea_ino,
1140 le32_to_cpu(entry->e_hash),
1143 ext4_warning(parent->i_sb,
1144 "cleanup ea_ino %u iget error %d", ea_ino,
1148 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1150 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1158 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1159 struct buffer_head *bh,
1160 struct ext4_xattr_entry *first, bool block_csum,
1161 struct ext4_xattr_inode_array **ea_inode_array,
1162 int extra_credits, bool skip_quota)
1164 struct inode *ea_inode;
1165 struct ext4_xattr_entry *entry;
1167 unsigned int ea_ino;
1171 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1172 credits = 2 + extra_credits;
1174 for (entry = first; !IS_LAST_ENTRY(entry);
1175 entry = EXT4_XATTR_NEXT(entry)) {
1176 if (!entry->e_value_inum)
1178 ea_ino = le32_to_cpu(entry->e_value_inum);
1179 err = ext4_xattr_inode_iget(parent, ea_ino,
1180 le32_to_cpu(entry->e_hash),
1185 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1187 ext4_warning_inode(ea_inode,
1188 "Expand inode array err=%d", err);
1193 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1196 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1201 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1203 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1209 ext4_xattr_inode_free_quota(parent, ea_inode,
1210 le32_to_cpu(entry->e_value_size));
1213 * Forget about ea_inode within the same transaction that
1214 * decrements the ref count. This avoids duplicate decrements in
1215 * case the rest of the work spills over to subsequent
1218 entry->e_value_inum = 0;
1219 entry->e_value_size = 0;
1226 * Note that we are deliberately skipping csum calculation for
1227 * the final update because we do not expect any journal
1228 * restarts until xattr block is freed.
1231 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1233 ext4_warning_inode(parent,
1234 "handle dirty metadata err=%d", err);
1239 * Release the xattr block BH: If the reference count is > 1, decrement it;
1240 * otherwise free the block.
1243 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1244 struct buffer_head *bh,
1245 struct ext4_xattr_inode_array **ea_inode_array,
1248 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1252 BUFFER_TRACE(bh, "get_write_access");
1253 error = ext4_journal_get_write_access(handle, bh);
1258 hash = le32_to_cpu(BHDR(bh)->h_hash);
1259 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1261 ea_bdebug(bh, "refcount now=0; freeing");
1263 * This must happen under buffer lock for
1264 * ext4_xattr_block_set() to reliably detect freed block
1267 mb_cache_entry_delete(ea_block_cache, hash,
1272 if (ext4_has_feature_ea_inode(inode->i_sb))
1273 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1275 true /* block_csum */,
1278 true /* skip_quota */);
1279 ext4_free_blocks(handle, inode, bh, 0, 1,
1280 EXT4_FREE_BLOCKS_METADATA |
1281 EXT4_FREE_BLOCKS_FORGET);
1284 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1285 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1286 struct mb_cache_entry *ce;
1288 if (ea_block_cache) {
1289 ce = mb_cache_entry_get(ea_block_cache, hash,
1293 mb_cache_entry_put(ea_block_cache, ce);
1298 ext4_xattr_block_csum_set(inode, bh);
1300 * Beware of this ugliness: Releasing of xattr block references
1301 * from different inodes can race and so we have to protect
1302 * from a race where someone else frees the block (and releases
1303 * its journal_head) before we are done dirtying the buffer. In
1304 * nojournal mode this race is harmless and we actually cannot
1305 * call ext4_handle_dirty_metadata() with locked buffer as
1306 * that function can call sync_dirty_buffer() so for that case
1307 * we handle the dirtying after unlocking the buffer.
1309 if (ext4_handle_valid(handle))
1310 error = ext4_handle_dirty_metadata(handle, inode, bh);
1312 if (!ext4_handle_valid(handle))
1313 error = ext4_handle_dirty_metadata(handle, inode, bh);
1315 ext4_handle_sync(handle);
1316 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1317 ea_bdebug(bh, "refcount now=%d; releasing",
1318 le32_to_cpu(BHDR(bh)->h_refcount));
1321 ext4_std_error(inode->i_sb, error);
1326 * Find the available free space for EAs. This also returns the total number of
1327 * bytes used by EA entries.
1329 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1330 size_t *min_offs, void *base, int *total)
1332 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1333 if (!last->e_value_inum && last->e_value_size) {
1334 size_t offs = le16_to_cpu(last->e_value_offs);
1335 if (offs < *min_offs)
1339 *total += EXT4_XATTR_LEN(last->e_name_len);
1341 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1345 * Write the value of the EA in an inode.
1347 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1348 const void *buf, int bufsize)
1350 struct buffer_head *bh = NULL;
1351 unsigned long block = 0;
1352 int blocksize = ea_inode->i_sb->s_blocksize;
1353 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1354 int csize, wsize = 0;
1359 while (ret >= 0 && ret < max_blocks) {
1360 struct ext4_map_blocks map;
1361 map.m_lblk = block += ret;
1362 map.m_len = max_blocks -= ret;
1364 ret = ext4_map_blocks(handle, ea_inode, &map,
1365 EXT4_GET_BLOCKS_CREATE);
1367 ext4_mark_inode_dirty(handle, ea_inode);
1368 if (ret == -ENOSPC &&
1369 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1381 while (wsize < bufsize) {
1384 csize = (bufsize - wsize) > blocksize ? blocksize :
1386 bh = ext4_getblk(handle, ea_inode, block, 0);
1389 ret = ext4_journal_get_write_access(handle, bh);
1393 memcpy(bh->b_data, buf, csize);
1394 set_buffer_uptodate(bh);
1395 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1402 inode_lock(ea_inode);
1403 i_size_write(ea_inode, wsize);
1404 ext4_update_i_disksize(ea_inode, wsize);
1405 inode_unlock(ea_inode);
1407 ext4_mark_inode_dirty(handle, ea_inode);
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 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1480 mb_cache_entry_put(ea_inode_cache, ce);
1485 ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1486 if (!IS_ERR(ea_inode) &&
1487 !is_bad_inode(ea_inode) &&
1488 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1489 i_size_read(ea_inode) == value_len &&
1490 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1491 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1493 !memcmp(value, ea_data, value_len)) {
1494 mb_cache_entry_touch(ea_inode_cache, ce);
1495 mb_cache_entry_put(ea_inode_cache, ce);
1500 if (!IS_ERR(ea_inode))
1502 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1509 * Add value of the EA in an inode.
1511 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1512 const void *value, size_t value_len,
1513 struct inode **ret_inode)
1515 struct inode *ea_inode;
1519 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1520 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1522 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1528 *ret_inode = ea_inode;
1532 /* Create an inode for the EA value */
1533 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1534 if (IS_ERR(ea_inode))
1535 return PTR_ERR(ea_inode);
1537 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1539 ext4_xattr_inode_dec_ref(handle, ea_inode);
1544 if (EA_INODE_CACHE(inode))
1545 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1546 ea_inode->i_ino, true /* reusable */);
1548 *ret_inode = ea_inode;
1553 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1554 * feature is enabled.
1556 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1558 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1559 struct ext4_xattr_search *s,
1560 handle_t *handle, struct inode *inode,
1563 struct ext4_xattr_entry *last, *next;
1564 struct ext4_xattr_entry *here = s->here;
1565 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1566 int in_inode = i->in_inode;
1567 struct inode *old_ea_inode = NULL;
1568 struct inode *new_ea_inode = NULL;
1569 size_t old_size, new_size;
1572 /* Space used by old and new values. */
1573 old_size = (!s->not_found && !here->e_value_inum) ?
1574 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1575 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1578 * Optimization for the simple case when old and new values have the
1579 * same padded sizes. Not applicable if external inodes are involved.
1581 if (new_size && new_size == old_size) {
1582 size_t offs = le16_to_cpu(here->e_value_offs);
1583 void *val = s->base + offs;
1585 here->e_value_size = cpu_to_le32(i->value_len);
1586 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1587 memset(val, 0, new_size);
1589 memcpy(val, i->value, i->value_len);
1590 /* Clear padding bytes. */
1591 memset(val + i->value_len, 0, new_size - i->value_len);
1596 /* Compute min_offs and last. */
1598 for (; !IS_LAST_ENTRY(last); last = next) {
1599 next = EXT4_XATTR_NEXT(last);
1600 if ((void *)next >= s->end) {
1601 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1602 ret = -EFSCORRUPTED;
1605 if (!last->e_value_inum && last->e_value_size) {
1606 size_t offs = le16_to_cpu(last->e_value_offs);
1607 if (offs < min_offs)
1612 /* Check whether we have enough space. */
1616 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1618 free += EXT4_XATTR_LEN(name_len) + old_size;
1620 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1626 * If storing the value in an external inode is an option,
1627 * reserve space for xattr entries/names in the external
1628 * attribute block so that a long value does not occupy the
1629 * whole space and prevent futher entries being added.
1631 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1632 new_size && is_block &&
1633 (min_offs + old_size - new_size) <
1634 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1641 * Getting access to old and new ea inodes is subject to failures.
1642 * Finish that work before doing any modifications to the xattr data.
1644 if (!s->not_found && here->e_value_inum) {
1645 ret = ext4_xattr_inode_iget(inode,
1646 le32_to_cpu(here->e_value_inum),
1647 le32_to_cpu(here->e_hash),
1650 old_ea_inode = NULL;
1654 if (i->value && in_inode) {
1655 WARN_ON_ONCE(!i->value_len);
1657 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1661 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1665 new_ea_inode = NULL;
1666 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1672 /* We are ready to release ref count on the old_ea_inode. */
1673 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1675 /* Release newly required ref count on new_ea_inode. */
1679 err = ext4_xattr_inode_dec_ref(handle,
1682 ext4_warning_inode(new_ea_inode,
1683 "dec ref new_ea_inode err=%d",
1685 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1691 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1692 le32_to_cpu(here->e_value_size));
1695 /* No failures allowed past this point. */
1697 if (!s->not_found && here->e_value_size && here->e_value_offs) {
1698 /* Remove the old value. */
1699 void *first_val = s->base + min_offs;
1700 size_t offs = le16_to_cpu(here->e_value_offs);
1701 void *val = s->base + offs;
1703 memmove(first_val + old_size, first_val, val - first_val);
1704 memset(first_val, 0, old_size);
1705 min_offs += old_size;
1707 /* Adjust all value offsets. */
1709 while (!IS_LAST_ENTRY(last)) {
1710 size_t o = le16_to_cpu(last->e_value_offs);
1712 if (!last->e_value_inum &&
1713 last->e_value_size && o < offs)
1714 last->e_value_offs = cpu_to_le16(o + old_size);
1715 last = EXT4_XATTR_NEXT(last);
1720 /* Remove old name. */
1721 size_t size = EXT4_XATTR_LEN(name_len);
1723 last = ENTRY((void *)last - size);
1724 memmove(here, (void *)here + size,
1725 (void *)last - (void *)here + sizeof(__u32));
1726 memset(last, 0, size);
1727 } else if (s->not_found) {
1728 /* Insert new name. */
1729 size_t size = EXT4_XATTR_LEN(name_len);
1730 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1732 memmove((void *)here + size, here, rest);
1733 memset(here, 0, size);
1734 here->e_name_index = i->name_index;
1735 here->e_name_len = name_len;
1736 memcpy(here->e_name, i->name, name_len);
1738 /* This is an update, reset value info. */
1739 here->e_value_inum = 0;
1740 here->e_value_offs = 0;
1741 here->e_value_size = 0;
1745 /* Insert new value. */
1747 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1748 } else if (i->value_len) {
1749 void *val = s->base + min_offs - new_size;
1751 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1752 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1753 memset(val, 0, new_size);
1755 memcpy(val, i->value, i->value_len);
1756 /* Clear padding bytes. */
1757 memset(val + i->value_len, 0,
1758 new_size - i->value_len);
1761 here->e_value_size = cpu_to_le32(i->value_len);
1768 /* Entry hash calculation. */
1773 * Feed crc32c hash instead of the raw value for entry
1774 * hash calculation. This is to avoid walking
1775 * potentially long value buffer again.
1777 crc32c_hash = cpu_to_le32(
1778 ext4_xattr_inode_get_hash(new_ea_inode));
1779 hash = ext4_xattr_hash_entry(here->e_name,
1782 } else if (is_block) {
1783 __le32 *value = s->base + le16_to_cpu(
1784 here->e_value_offs);
1786 hash = ext4_xattr_hash_entry(here->e_name,
1787 here->e_name_len, value,
1790 here->e_hash = hash;
1794 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1803 struct ext4_xattr_block_find {
1804 struct ext4_xattr_search s;
1805 struct buffer_head *bh;
1809 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1810 struct ext4_xattr_block_find *bs)
1812 struct super_block *sb = inode->i_sb;
1815 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1816 i->name_index, i->name, i->value, (long)i->value_len);
1818 if (EXT4_I(inode)->i_file_acl) {
1819 /* The inode already has an extended attribute block. */
1820 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1824 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1825 atomic_read(&(bs->bh->b_count)),
1826 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1827 error = ext4_xattr_check_block(inode, bs->bh);
1830 /* Find the named attribute. */
1831 bs->s.base = BHDR(bs->bh);
1832 bs->s.first = BFIRST(bs->bh);
1833 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1834 bs->s.here = bs->s.first;
1835 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1836 i->name_index, i->name, 1);
1837 if (error && error != -ENODATA)
1839 bs->s.not_found = error;
1848 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1849 struct ext4_xattr_info *i,
1850 struct ext4_xattr_block_find *bs)
1852 struct super_block *sb = inode->i_sb;
1853 struct buffer_head *new_bh = NULL;
1854 struct ext4_xattr_search s_copy = bs->s;
1855 struct ext4_xattr_search *s = &s_copy;
1856 struct mb_cache_entry *ce = NULL;
1858 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1859 struct inode *ea_inode = NULL, *tmp_inode;
1860 size_t old_ea_inode_quota = 0;
1861 unsigned int ea_ino;
1864 #define header(x) ((struct ext4_xattr_header *)(x))
1867 BUFFER_TRACE(bs->bh, "get_write_access");
1868 error = ext4_journal_get_write_access(handle, bs->bh);
1871 lock_buffer(bs->bh);
1873 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1874 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1877 * This must happen under buffer lock for
1878 * ext4_xattr_block_set() to reliably detect modified
1882 mb_cache_entry_delete(ea_block_cache, hash,
1884 ea_bdebug(bs->bh, "modifying in-place");
1885 error = ext4_xattr_set_entry(i, s, handle, inode,
1886 true /* is_block */);
1887 ext4_xattr_block_csum_set(inode, bs->bh);
1888 unlock_buffer(bs->bh);
1889 if (error == -EFSCORRUPTED)
1892 error = ext4_handle_dirty_metadata(handle,
1899 int offset = (char *)s->here - bs->bh->b_data;
1901 unlock_buffer(bs->bh);
1902 ea_bdebug(bs->bh, "cloning");
1903 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1905 if (s->base == NULL)
1907 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1908 s->first = ENTRY(header(s->base)+1);
1909 header(s->base)->h_refcount = cpu_to_le32(1);
1910 s->here = ENTRY(s->base + offset);
1911 s->end = s->base + bs->bh->b_size;
1914 * If existing entry points to an xattr inode, we need
1915 * to prevent ext4_xattr_set_entry() from decrementing
1916 * ref count on it because the reference belongs to the
1917 * original block. In this case, make the entry look
1918 * like it has an empty value.
1920 if (!s->not_found && s->here->e_value_inum) {
1921 ea_ino = le32_to_cpu(s->here->e_value_inum);
1922 error = ext4_xattr_inode_iget(inode, ea_ino,
1923 le32_to_cpu(s->here->e_hash),
1928 if (!ext4_test_inode_state(tmp_inode,
1929 EXT4_STATE_LUSTRE_EA_INODE)) {
1931 * Defer quota free call for previous
1932 * inode until success is guaranteed.
1934 old_ea_inode_quota = le32_to_cpu(
1935 s->here->e_value_size);
1939 s->here->e_value_inum = 0;
1940 s->here->e_value_size = 0;
1944 /* Allocate a buffer where we construct the new block. */
1945 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1946 /* assert(header == s->base) */
1948 if (s->base == NULL)
1950 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1951 header(s->base)->h_blocks = cpu_to_le32(1);
1952 header(s->base)->h_refcount = cpu_to_le32(1);
1953 s->first = ENTRY(header(s->base)+1);
1954 s->here = ENTRY(header(s->base)+1);
1955 s->end = s->base + sb->s_blocksize;
1958 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1959 if (error == -EFSCORRUPTED)
1964 if (i->value && s->here->e_value_inum) {
1966 * A ref count on ea_inode has been taken as part of the call to
1967 * ext4_xattr_set_entry() above. We would like to drop this
1968 * extra ref but we have to wait until the xattr block is
1969 * initialized and has its own ref count on the ea_inode.
1971 ea_ino = le32_to_cpu(s->here->e_value_inum);
1972 error = ext4_xattr_inode_iget(inode, ea_ino,
1973 le32_to_cpu(s->here->e_hash),
1982 if (!IS_LAST_ENTRY(s->first)) {
1983 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1986 /* We found an identical block in the cache. */
1987 if (new_bh == bs->bh)
1988 ea_bdebug(new_bh, "keeping");
1992 WARN_ON_ONCE(dquot_initialize_needed(inode));
1994 /* The old block is released after updating
1996 error = dquot_alloc_block(inode,
1997 EXT4_C2B(EXT4_SB(sb), 1));
2000 BUFFER_TRACE(new_bh, "get_write_access");
2001 error = ext4_journal_get_write_access(handle,
2005 lock_buffer(new_bh);
2007 * We have to be careful about races with
2008 * freeing, rehashing or adding references to
2009 * xattr block. Once we hold buffer lock xattr
2010 * block's state is stable so we can check
2011 * whether the block got freed / rehashed or
2012 * not. Since we unhash mbcache entry under
2013 * buffer lock when freeing / rehashing xattr
2014 * block, checking whether entry is still
2015 * hashed is reliable. Same rules hold for
2016 * e_reusable handling.
2018 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2021 * Undo everything and check mbcache
2024 unlock_buffer(new_bh);
2025 dquot_free_block(inode,
2026 EXT4_C2B(EXT4_SB(sb),
2029 mb_cache_entry_put(ea_block_cache, ce);
2034 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2035 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2036 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2038 ea_bdebug(new_bh, "reusing; refcount now=%d",
2040 ext4_xattr_block_csum_set(inode, new_bh);
2041 unlock_buffer(new_bh);
2042 error = ext4_handle_dirty_metadata(handle,
2048 mb_cache_entry_touch(ea_block_cache, ce);
2049 mb_cache_entry_put(ea_block_cache, ce);
2051 } else if (bs->bh && s->base == bs->bh->b_data) {
2052 /* We were modifying this block in-place. */
2053 ea_bdebug(bs->bh, "keeping this block");
2054 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2058 /* We need to allocate a new block */
2059 ext4_fsblk_t goal, block;
2061 WARN_ON_ONCE(dquot_initialize_needed(inode));
2063 goal = ext4_group_first_block_no(sb,
2064 EXT4_I(inode)->i_block_group);
2066 /* non-extent files can't have physical blocks past 2^32 */
2067 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2068 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2070 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2075 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2076 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2078 ea_idebug(inode, "creating block %llu",
2079 (unsigned long long)block);
2081 new_bh = sb_getblk(sb, block);
2082 if (unlikely(!new_bh)) {
2085 ext4_free_blocks(handle, inode, NULL, block, 1,
2086 EXT4_FREE_BLOCKS_METADATA);
2089 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2090 ENTRY(header(s->base)+1));
2094 /* Drop the extra ref on ea_inode. */
2095 error = ext4_xattr_inode_dec_ref(handle,
2098 ext4_warning_inode(ea_inode,
2105 lock_buffer(new_bh);
2106 error = ext4_journal_get_create_access(handle, new_bh);
2108 unlock_buffer(new_bh);
2112 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2113 ext4_xattr_block_csum_set(inode, new_bh);
2114 set_buffer_uptodate(new_bh);
2115 unlock_buffer(new_bh);
2116 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2117 error = ext4_handle_dirty_metadata(handle, inode,
2124 if (old_ea_inode_quota)
2125 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2127 /* Update the inode. */
2128 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2130 /* Drop the previous xattr block. */
2131 if (bs->bh && bs->bh != new_bh) {
2132 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2134 ext4_xattr_release_block(handle, inode, bs->bh,
2136 0 /* extra_credits */);
2137 ext4_xattr_inode_array_free(ea_inode_array);
2145 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2147 ext4_warning_inode(ea_inode, "dec ref error=%d",
2150 /* If there was an error, revert the quota charge. */
2152 ext4_xattr_inode_free_quota(inode, ea_inode,
2153 i_size_read(ea_inode));
2157 mb_cache_entry_put(ea_block_cache, ce);
2159 if (!(bs->bh && s->base == bs->bh->b_data))
2165 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2169 EXT4_ERROR_INODE(inode, "bad block %llu",
2170 EXT4_I(inode)->i_file_acl);
2176 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2177 struct ext4_xattr_ibody_find *is)
2179 struct ext4_xattr_ibody_header *header;
2180 struct ext4_inode *raw_inode;
2183 if (EXT4_I(inode)->i_extra_isize == 0)
2185 raw_inode = ext4_raw_inode(&is->iloc);
2186 header = IHDR(inode, raw_inode);
2187 is->s.base = is->s.first = IFIRST(header);
2188 is->s.here = is->s.first;
2189 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2190 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2191 error = xattr_check_inode(inode, header, is->s.end);
2194 /* Find the named attribute. */
2195 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2196 i->name_index, i->name, 0);
2197 if (error && error != -ENODATA)
2199 is->s.not_found = error;
2204 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2205 struct ext4_xattr_info *i,
2206 struct ext4_xattr_ibody_find *is)
2208 struct ext4_xattr_ibody_header *header;
2209 struct ext4_xattr_search *s = &is->s;
2212 if (EXT4_I(inode)->i_extra_isize == 0)
2214 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2217 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2218 if (!IS_LAST_ENTRY(s->first)) {
2219 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2220 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2222 header->h_magic = cpu_to_le32(0);
2223 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2228 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2229 struct ext4_xattr_info *i,
2230 struct ext4_xattr_ibody_find *is)
2232 struct ext4_xattr_ibody_header *header;
2233 struct ext4_xattr_search *s = &is->s;
2236 if (EXT4_I(inode)->i_extra_isize == 0)
2238 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2241 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2242 if (!IS_LAST_ENTRY(s->first)) {
2243 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2244 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2246 header->h_magic = cpu_to_le32(0);
2247 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2252 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2253 struct ext4_xattr_info *i)
2257 /* When e_value_inum is set the value is stored externally. */
2258 if (s->here->e_value_inum)
2260 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2262 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2263 return !memcmp(value, i->value, i->value_len);
2266 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2268 struct buffer_head *bh;
2271 if (!EXT4_I(inode)->i_file_acl)
2273 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2275 return ERR_PTR(-EIO);
2276 error = ext4_xattr_check_block(inode, bh);
2278 return ERR_PTR(error);
2283 * ext4_xattr_set_handle()
2285 * Create, replace or remove an extended attribute for this inode. Value
2286 * is NULL to remove an existing extended attribute, and non-NULL to
2287 * either replace an existing extended attribute, or create a new extended
2288 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2289 * specify that an extended attribute must exist and must not exist
2290 * previous to the call, respectively.
2292 * Returns 0, or a negative error number on failure.
2295 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2296 const char *name, const void *value, size_t value_len,
2299 struct ext4_xattr_info i = {
2300 .name_index = name_index,
2303 .value_len = value_len,
2306 struct ext4_xattr_ibody_find is = {
2307 .s = { .not_found = -ENODATA, },
2309 struct ext4_xattr_block_find bs = {
2310 .s = { .not_found = -ENODATA, },
2317 if (strlen(name) > 255)
2320 ext4_write_lock_xattr(inode, &no_expand);
2322 /* Check journal credits under write lock. */
2323 if (ext4_handle_valid(handle)) {
2324 struct buffer_head *bh;
2327 bh = ext4_xattr_get_block(inode);
2329 error = PTR_ERR(bh);
2333 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2335 flags & XATTR_CREATE);
2338 if (!ext4_handle_has_enough_credits(handle, credits)) {
2344 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2348 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2349 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2350 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2351 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2354 error = ext4_xattr_ibody_find(inode, &i, &is);
2358 error = ext4_xattr_block_find(inode, &i, &bs);
2361 if (is.s.not_found && bs.s.not_found) {
2363 if (flags & XATTR_REPLACE)
2370 if (flags & XATTR_CREATE)
2375 if (!is.s.not_found)
2376 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2377 else if (!bs.s.not_found)
2378 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2381 /* Xattr value did not change? Save us some work and bail out */
2382 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2384 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2387 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2388 (EXT4_XATTR_SIZE(i.value_len) >
2389 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2392 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2393 if (!error && !bs.s.not_found) {
2395 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2396 } else if (error == -ENOSPC) {
2397 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2398 error = ext4_xattr_block_find(inode, &i, &bs);
2402 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2403 if (!error && !is.s.not_found) {
2405 error = ext4_xattr_ibody_set(handle, inode, &i,
2407 } else if (error == -ENOSPC) {
2409 * Xattr does not fit in the block, store at
2410 * external inode if possible.
2412 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2421 ext4_xattr_update_super_block(handle, inode->i_sb);
2422 inode->i_ctime = current_time(inode);
2425 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2427 * The bh is consumed by ext4_mark_iloc_dirty, even with
2432 ext4_handle_sync(handle);
2438 ext4_write_unlock_xattr(inode, &no_expand);
2442 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2443 bool is_create, int *credits)
2445 struct buffer_head *bh;
2450 if (!EXT4_SB(inode->i_sb)->s_journal)
2453 down_read(&EXT4_I(inode)->xattr_sem);
2455 bh = ext4_xattr_get_block(inode);
2459 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2460 value_len, is_create);
2465 up_read(&EXT4_I(inode)->xattr_sem);
2472 * Like ext4_xattr_set_handle, but start from an inode. This extended
2473 * attribute modification is a filesystem transaction by itself.
2475 * Returns 0, or a negative error number on failure.
2478 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2479 const void *value, size_t value_len, int flags)
2482 struct super_block *sb = inode->i_sb;
2483 int error, retries = 0;
2486 error = dquot_initialize(inode);
2491 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2496 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2497 if (IS_ERR(handle)) {
2498 error = PTR_ERR(handle);
2502 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2503 value, value_len, flags);
2504 error2 = ext4_journal_stop(handle);
2505 if (error == -ENOSPC &&
2506 ext4_should_retry_alloc(sb, &retries))
2516 * Shift the EA entries in the inode to create space for the increased
2519 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2520 int value_offs_shift, void *to,
2521 void *from, size_t n)
2523 struct ext4_xattr_entry *last = entry;
2526 /* We always shift xattr headers further thus offsets get lower */
2527 BUG_ON(value_offs_shift > 0);
2529 /* Adjust the value offsets of the entries */
2530 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2531 if (!last->e_value_inum && last->e_value_size) {
2532 new_offs = le16_to_cpu(last->e_value_offs) +
2534 last->e_value_offs = cpu_to_le16(new_offs);
2537 /* Shift the entries by n bytes */
2538 memmove(to, from, n);
2542 * Move xattr pointed to by 'entry' from inode into external xattr block
2544 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2545 struct ext4_inode *raw_inode,
2546 struct ext4_xattr_entry *entry)
2548 struct ext4_xattr_ibody_find *is = NULL;
2549 struct ext4_xattr_block_find *bs = NULL;
2550 char *buffer = NULL, *b_entry_name = NULL;
2551 size_t value_size = le32_to_cpu(entry->e_value_size);
2552 struct ext4_xattr_info i = {
2555 .name_index = entry->e_name_index,
2556 .in_inode = !!entry->e_value_inum,
2558 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2561 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2562 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2563 buffer = kmalloc(value_size, GFP_NOFS);
2564 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2565 if (!is || !bs || !buffer || !b_entry_name) {
2570 is->s.not_found = -ENODATA;
2571 bs->s.not_found = -ENODATA;
2575 /* Save the entry name and the entry value */
2576 if (entry->e_value_inum) {
2577 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2581 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2582 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2585 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2586 b_entry_name[entry->e_name_len] = '\0';
2587 i.name = b_entry_name;
2589 error = ext4_get_inode_loc(inode, &is->iloc);
2593 error = ext4_xattr_ibody_find(inode, &i, is);
2597 /* Remove the chosen entry from the inode */
2598 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2603 i.value_len = value_size;
2604 error = ext4_xattr_block_find(inode, &i, bs);
2608 /* Add entry which was removed from the inode into the block */
2609 error = ext4_xattr_block_set(handle, inode, &i, bs);
2614 kfree(b_entry_name);
2617 brelse(is->iloc.bh);
2624 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2625 struct ext4_inode *raw_inode,
2626 int isize_diff, size_t ifree,
2627 size_t bfree, int *total_ino)
2629 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2630 struct ext4_xattr_entry *small_entry;
2631 struct ext4_xattr_entry *entry;
2632 struct ext4_xattr_entry *last;
2633 unsigned int entry_size; /* EA entry size */
2634 unsigned int total_size; /* EA entry size + value size */
2635 unsigned int min_total_size;
2638 while (isize_diff > ifree) {
2641 min_total_size = ~0U;
2642 last = IFIRST(header);
2643 /* Find the entry best suited to be pushed into EA block */
2644 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2645 /* never move system.data out of the inode */
2646 if ((last->e_name_len == 4) &&
2647 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2648 !memcmp(last->e_name, "data", 4))
2650 total_size = EXT4_XATTR_LEN(last->e_name_len);
2651 if (!last->e_value_inum)
2652 total_size += EXT4_XATTR_SIZE(
2653 le32_to_cpu(last->e_value_size));
2654 if (total_size <= bfree &&
2655 total_size < min_total_size) {
2656 if (total_size + ifree < isize_diff) {
2660 min_total_size = total_size;
2665 if (entry == NULL) {
2666 if (small_entry == NULL)
2668 entry = small_entry;
2671 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2672 total_size = entry_size;
2673 if (!entry->e_value_inum)
2674 total_size += EXT4_XATTR_SIZE(
2675 le32_to_cpu(entry->e_value_size));
2676 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2681 *total_ino -= entry_size;
2682 ifree += total_size;
2683 bfree -= total_size;
2690 * Expand an inode by new_extra_isize bytes when EAs are present.
2691 * Returns 0 on success or negative error number on failure.
2693 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2694 struct ext4_inode *raw_inode, handle_t *handle)
2696 struct ext4_xattr_ibody_header *header;
2697 struct buffer_head *bh;
2698 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2699 static unsigned int mnt_count;
2701 size_t ifree, bfree;
2704 int error = 0, tried_min_extra_isize = 0;
2705 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2706 int isize_diff; /* How much do we need to grow i_extra_isize */
2709 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2710 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2713 header = IHDR(inode, raw_inode);
2716 * Check if enough free space is available in the inode to shift the
2717 * entries ahead by new_extra_isize.
2720 base = IFIRST(header);
2721 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2722 min_offs = end - base;
2723 total_ino = sizeof(struct ext4_xattr_ibody_header);
2725 error = xattr_check_inode(inode, header, end);
2729 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2730 if (ifree >= isize_diff)
2734 * Enough free space isn't available in the inode, check if
2735 * EA block can hold new_extra_isize bytes.
2737 if (EXT4_I(inode)->i_file_acl) {
2738 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2742 error = ext4_xattr_check_block(inode, bh);
2746 end = bh->b_data + bh->b_size;
2747 min_offs = end - base;
2748 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2751 if (bfree + ifree < isize_diff) {
2752 if (!tried_min_extra_isize && s_min_extra_isize) {
2753 tried_min_extra_isize++;
2754 new_extra_isize = s_min_extra_isize;
2761 bfree = inode->i_sb->s_blocksize;
2764 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2765 isize_diff, ifree, bfree,
2768 if (error == -ENOSPC && !tried_min_extra_isize &&
2769 s_min_extra_isize) {
2770 tried_min_extra_isize++;
2771 new_extra_isize = s_min_extra_isize;
2777 /* Adjust the offsets and shift the remaining entries ahead */
2778 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2779 - new_extra_isize, (void *)raw_inode +
2780 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2781 (void *)header, total_ino);
2782 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2785 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2786 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2788 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2793 #define EIA_INCR 16 /* must be 2^n */
2794 #define EIA_MASK (EIA_INCR - 1)
2796 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2797 * If @ea_inode_array is new or full it will be grown and the old
2798 * contents copied over.
2801 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2802 struct inode *inode)
2804 if (*ea_inode_array == NULL) {
2806 * Start with 15 inodes, so it fits into a power-of-two size.
2807 * If *ea_inode_array is NULL, this is essentially offsetof()
2810 kmalloc(offsetof(struct ext4_xattr_inode_array,
2813 if (*ea_inode_array == NULL)
2815 (*ea_inode_array)->count = 0;
2816 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2817 /* expand the array once all 15 + n * 16 slots are full */
2818 struct ext4_xattr_inode_array *new_array = NULL;
2819 int count = (*ea_inode_array)->count;
2821 /* if new_array is NULL, this is essentially offsetof() */
2822 new_array = kmalloc(
2823 offsetof(struct ext4_xattr_inode_array,
2824 inodes[count + EIA_INCR]),
2826 if (new_array == NULL)
2828 memcpy(new_array, *ea_inode_array,
2829 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2830 kfree(*ea_inode_array);
2831 *ea_inode_array = new_array;
2833 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2838 * ext4_xattr_delete_inode()
2840 * Free extended attribute resources associated with this inode. Traverse
2841 * all entries and decrement reference on any xattr inodes associated with this
2842 * inode. This is called immediately before an inode is freed. We have exclusive
2843 * access to the inode. If an orphan inode is deleted it will also release its
2844 * references on xattr block and xattr inodes.
2846 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2847 struct ext4_xattr_inode_array **ea_inode_array,
2850 struct buffer_head *bh = NULL;
2851 struct ext4_xattr_ibody_header *header;
2852 struct ext4_iloc iloc = { .bh = NULL };
2853 struct ext4_xattr_entry *entry;
2854 struct inode *ea_inode;
2857 error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2860 false /* block_csum */);
2862 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2866 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2867 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2869 error = ext4_get_inode_loc(inode, &iloc);
2871 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2875 error = ext4_journal_get_write_access(handle, iloc.bh);
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 = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2895 EXT4_ERROR_INODE(inode, "block %llu read error",
2896 EXT4_I(inode)->i_file_acl);
2900 error = ext4_xattr_check_block(inode, bh);
2904 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2905 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2906 entry = EXT4_XATTR_NEXT(entry)) {
2907 if (!entry->e_value_inum)
2909 error = ext4_xattr_inode_iget(inode,
2910 le32_to_cpu(entry->e_value_inum),
2911 le32_to_cpu(entry->e_hash),
2915 ext4_xattr_inode_free_quota(inode, ea_inode,
2916 le32_to_cpu(entry->e_value_size));
2922 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2925 * Update i_file_acl value in the same transaction that releases
2928 EXT4_I(inode)->i_file_acl = 0;
2929 error = ext4_mark_inode_dirty(handle, inode);
2931 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2943 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2947 if (ea_inode_array == NULL)
2950 for (idx = 0; idx < ea_inode_array->count; ++idx)
2951 iput(ea_inode_array->inodes[idx]);
2952 kfree(ea_inode_array);
2956 * ext4_xattr_block_cache_insert()
2958 * Create a new entry in the extended attribute block cache, and insert
2959 * it unless such an entry is already in the cache.
2961 * Returns 0, or a negative error number on failure.
2964 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2965 struct buffer_head *bh)
2967 struct ext4_xattr_header *header = BHDR(bh);
2968 __u32 hash = le32_to_cpu(header->h_hash);
2969 int reusable = le32_to_cpu(header->h_refcount) <
2970 EXT4_XATTR_REFCOUNT_MAX;
2973 if (!ea_block_cache)
2975 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2976 bh->b_blocknr, reusable);
2978 if (error == -EBUSY)
2979 ea_bdebug(bh, "already in cache");
2981 ea_bdebug(bh, "inserting [%x]", (int)hash);
2987 * Compare two extended attribute blocks for equality.
2989 * Returns 0 if the blocks are equal, 1 if they differ, and
2990 * a negative error number on errors.
2993 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2994 struct ext4_xattr_header *header2)
2996 struct ext4_xattr_entry *entry1, *entry2;
2998 entry1 = ENTRY(header1+1);
2999 entry2 = ENTRY(header2+1);
3000 while (!IS_LAST_ENTRY(entry1)) {
3001 if (IS_LAST_ENTRY(entry2))
3003 if (entry1->e_hash != entry2->e_hash ||
3004 entry1->e_name_index != entry2->e_name_index ||
3005 entry1->e_name_len != entry2->e_name_len ||
3006 entry1->e_value_size != entry2->e_value_size ||
3007 entry1->e_value_inum != entry2->e_value_inum ||
3008 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3010 if (!entry1->e_value_inum &&
3011 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3012 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3013 le32_to_cpu(entry1->e_value_size)))
3016 entry1 = EXT4_XATTR_NEXT(entry1);
3017 entry2 = EXT4_XATTR_NEXT(entry2);
3019 if (!IS_LAST_ENTRY(entry2))
3025 * ext4_xattr_block_cache_find()
3027 * Find an identical extended attribute block.
3029 * Returns a pointer to the block found, or NULL if such a block was
3030 * not found or an error occurred.
3032 static struct buffer_head *
3033 ext4_xattr_block_cache_find(struct inode *inode,
3034 struct ext4_xattr_header *header,
3035 struct mb_cache_entry **pce)
3037 __u32 hash = le32_to_cpu(header->h_hash);
3038 struct mb_cache_entry *ce;
3039 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3041 if (!ea_block_cache)
3043 if (!header->h_hash)
3044 return NULL; /* never share */
3045 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3046 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3048 struct buffer_head *bh;
3050 bh = sb_bread(inode->i_sb, ce->e_value);
3052 EXT4_ERROR_INODE(inode, "block %lu read error",
3053 (unsigned long)ce->e_value);
3054 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3059 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3064 #define NAME_HASH_SHIFT 5
3065 #define VALUE_HASH_SHIFT 16
3068 * ext4_xattr_hash_entry()
3070 * Compute the hash of an extended attribute.
3072 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3077 while (name_len--) {
3078 hash = (hash << NAME_HASH_SHIFT) ^
3079 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3082 while (value_count--) {
3083 hash = (hash << VALUE_HASH_SHIFT) ^
3084 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3085 le32_to_cpu(*value++);
3087 return cpu_to_le32(hash);
3090 #undef NAME_HASH_SHIFT
3091 #undef VALUE_HASH_SHIFT
3093 #define BLOCK_HASH_SHIFT 16
3096 * ext4_xattr_rehash()
3098 * Re-compute the extended attribute hash value after an entry has changed.
3100 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3102 struct ext4_xattr_entry *here;
3105 here = ENTRY(header+1);
3106 while (!IS_LAST_ENTRY(here)) {
3107 if (!here->e_hash) {
3108 /* Block is not shared if an entry's hash value == 0 */
3112 hash = (hash << BLOCK_HASH_SHIFT) ^
3113 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3114 le32_to_cpu(here->e_hash);
3115 here = EXT4_XATTR_NEXT(here);
3117 header->h_hash = cpu_to_le32(hash);
3120 #undef BLOCK_HASH_SHIFT
3122 #define HASH_BUCKET_BITS 10
3125 ext4_xattr_create_cache(void)
3127 return mb_cache_create(HASH_BUCKET_BITS);
3130 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3133 mb_cache_destroy(cache);