1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/xattr.c
5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9 * Extended attributes for symlinks and special files added per
10 * suggestion of Luka Renko <luka.renko@hermes.si>.
11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14 * and Andreas Gruenbacher <agruen@suse.de>.
18 * Extended attributes are stored directly in inodes (on file systems with
19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
20 * field contains the block number if an inode uses an additional block. All
21 * attributes must fit in the inode and one additional block. Blocks that
22 * contain the identical set of attributes may be shared among several inodes.
23 * Identical blocks are detected by keeping a cache of blocks that have
24 * recently been accessed.
26 * The attributes in inodes and on blocks have a different header; the entries
27 * are stored in the same format:
29 * +------------------+
32 * | entry 2 | | growing downwards
37 * | value 3 | | growing upwards
39 * +------------------+
41 * The header is followed by multiple entry descriptors. In disk blocks, the
42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
43 * attribute values are aligned to the end of the block in no specific order.
47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
48 * EA blocks are only changed if they are exclusive to an inode, so
49 * holding xattr_sem also means that nothing but the EA block's reference
50 * count can change. Multiple writers to the same block are synchronized
54 #include <linux/init.h>
56 #include <linux/slab.h>
57 #include <linux/mbcache.h>
58 #include <linux/quotaops.h>
59 #include "ext4_jbd2.h"
64 #ifdef EXT4_XATTR_DEBUG
65 # define ea_idebug(inode, fmt, ...) \
66 printk(KERN_DEBUG "inode %s:%lu: " fmt "\n", \
67 inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
68 # define ea_bdebug(bh, fmt, ...) \
69 printk(KERN_DEBUG "block %pg:%lu: " fmt "\n", \
70 bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
72 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
73 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
76 static void ext4_xattr_block_cache_insert(struct mb_cache *,
77 struct buffer_head *);
78 static struct buffer_head *
79 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
80 struct mb_cache_entry **);
81 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
83 static void ext4_xattr_rehash(struct ext4_xattr_header *);
85 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
86 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
87 #ifdef CONFIG_EXT4_FS_POSIX_ACL
88 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
89 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
91 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
92 #ifdef CONFIG_EXT4_FS_SECURITY
93 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
97 const struct xattr_handler *ext4_xattr_handlers[] = {
98 &ext4_xattr_user_handler,
99 &ext4_xattr_trusted_handler,
100 #ifdef CONFIG_EXT4_FS_POSIX_ACL
101 &posix_acl_access_xattr_handler,
102 &posix_acl_default_xattr_handler,
104 #ifdef CONFIG_EXT4_FS_SECURITY
105 &ext4_xattr_security_handler,
110 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
111 inode->i_sb->s_fs_info)->s_ea_block_cache)
113 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
114 inode->i_sb->s_fs_info)->s_ea_inode_cache)
117 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
118 struct inode *inode);
120 #ifdef CONFIG_LOCKDEP
121 void ext4_xattr_inode_set_class(struct inode *ea_inode)
123 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
127 static __le32 ext4_xattr_block_csum(struct inode *inode,
129 struct ext4_xattr_header *hdr)
131 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
133 __le64 dsk_block_nr = cpu_to_le64(block_nr);
134 __u32 dummy_csum = 0;
135 int offset = offsetof(struct ext4_xattr_header, h_checksum);
137 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
138 sizeof(dsk_block_nr));
139 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
140 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
141 offset += sizeof(dummy_csum);
142 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
143 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
145 return cpu_to_le32(csum);
148 static int ext4_xattr_block_csum_verify(struct inode *inode,
149 struct buffer_head *bh)
151 struct ext4_xattr_header *hdr = BHDR(bh);
154 if (ext4_has_metadata_csum(inode->i_sb)) {
156 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
157 bh->b_blocknr, hdr));
163 static void ext4_xattr_block_csum_set(struct inode *inode,
164 struct buffer_head *bh)
166 if (ext4_has_metadata_csum(inode->i_sb))
167 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
168 bh->b_blocknr, BHDR(bh));
171 static inline const struct xattr_handler *
172 ext4_xattr_handler(int name_index)
174 const struct xattr_handler *handler = NULL;
176 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
177 handler = ext4_xattr_handler_map[name_index];
182 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
185 struct ext4_xattr_entry *e = entry;
187 /* Find the end of the names list */
188 while (!IS_LAST_ENTRY(e)) {
189 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
190 if ((void *)next >= end)
191 return -EFSCORRUPTED;
192 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
193 return -EFSCORRUPTED;
197 /* Check the values */
198 while (!IS_LAST_ENTRY(entry)) {
199 u32 size = le32_to_cpu(entry->e_value_size);
201 if (size > EXT4_XATTR_SIZE_MAX)
202 return -EFSCORRUPTED;
204 if (size != 0 && entry->e_value_inum == 0) {
205 u16 offs = le16_to_cpu(entry->e_value_offs);
209 * The value cannot overlap the names, and the value
210 * with padding cannot extend beyond 'end'. Check both
211 * the padded and unpadded sizes, since the size may
212 * overflow to 0 when adding padding.
214 if (offs > end - value_start)
215 return -EFSCORRUPTED;
216 value = value_start + offs;
217 if (value < (void *)e + sizeof(u32) ||
218 size > end - value ||
219 EXT4_XATTR_SIZE(size) > end - value)
220 return -EFSCORRUPTED;
222 entry = EXT4_XATTR_NEXT(entry);
229 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
230 const char *function, unsigned int line)
232 int error = -EFSCORRUPTED;
234 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
235 BHDR(bh)->h_blocks != cpu_to_le32(1))
237 if (buffer_verified(bh))
241 if (!ext4_xattr_block_csum_verify(inode, bh))
243 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
247 __ext4_error_inode(inode, function, line, 0,
248 "corrupted xattr block %llu",
249 (unsigned long long) bh->b_blocknr);
251 set_buffer_verified(bh);
255 #define ext4_xattr_check_block(inode, bh) \
256 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
260 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
261 void *end, const char *function, unsigned int line)
263 int error = -EFSCORRUPTED;
265 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
266 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
268 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
271 __ext4_error_inode(inode, function, line, 0,
272 "corrupted in-inode xattr");
276 #define xattr_check_inode(inode, header, end) \
277 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
280 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
281 void *end, int name_index, const char *name, int sorted)
283 struct ext4_xattr_entry *entry, *next;
289 name_len = strlen(name);
290 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
291 next = EXT4_XATTR_NEXT(entry);
292 if ((void *) next >= end) {
293 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
294 return -EFSCORRUPTED;
296 cmp = name_index - entry->e_name_index;
298 cmp = name_len - entry->e_name_len;
300 cmp = memcmp(name, entry->e_name, name_len);
301 if (cmp <= 0 && (sorted || cmp == 0))
305 return cmp ? -ENODATA : 0;
309 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
311 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
314 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
316 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
317 ((u32)ea_inode->i_version);
320 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
322 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
323 ea_inode->i_version = (u32)ref_count;
326 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
328 return (u32)ea_inode->i_atime.tv_sec;
331 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
333 ea_inode->i_atime.tv_sec = hash;
337 * Read the EA value from an inode.
339 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
341 int blocksize = 1 << ea_inode->i_blkbits;
342 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
343 int tail_size = (size % blocksize) ?: blocksize;
344 struct buffer_head *bhs_inline[8];
345 struct buffer_head **bhs = bhs_inline;
348 if (bh_count > ARRAY_SIZE(bhs_inline)) {
349 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
354 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
355 true /* wait */, bhs);
359 for (i = 0; i < bh_count; i++) {
360 /* There shouldn't be any holes in ea_inode. */
365 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
366 i < bh_count - 1 ? blocksize : tail_size);
370 for (i = 0; i < bh_count; i++)
373 if (bhs != bhs_inline)
378 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
380 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
381 u32 ea_inode_hash, struct inode **ea_inode)
386 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
388 err = PTR_ERR(inode);
389 ext4_error(parent->i_sb,
390 "error while reading EA inode %lu err=%d", ea_ino,
395 if (is_bad_inode(inode)) {
396 ext4_error(parent->i_sb,
397 "error while reading EA inode %lu is_bad_inode",
403 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
404 ext4_error(parent->i_sb,
405 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
411 ext4_xattr_inode_set_class(inode);
414 * Check whether this is an old Lustre-style xattr inode. Lustre
415 * implementation does not have hash validation, rather it has a
416 * backpointer from ea_inode to the parent inode.
418 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
419 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
420 inode->i_generation == parent->i_generation) {
421 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
422 ext4_xattr_inode_set_ref(inode, 1);
425 inode->i_flags |= S_NOQUOTA;
437 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
438 struct ext4_xattr_entry *entry, void *buffer,
443 /* Verify stored hash matches calculated hash. */
444 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
445 if (hash != ext4_xattr_inode_get_hash(ea_inode))
446 return -EFSCORRUPTED;
449 __le32 e_hash, tmp_data;
451 /* Verify entry hash. */
452 tmp_data = cpu_to_le32(hash);
453 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
455 if (e_hash != entry->e_hash)
456 return -EFSCORRUPTED;
462 * Read xattr value from the EA inode.
465 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
466 void *buffer, size_t size)
468 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
469 struct inode *ea_inode;
472 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
473 le32_to_cpu(entry->e_hash), &ea_inode);
479 if (i_size_read(ea_inode) != size) {
480 ext4_warning_inode(ea_inode,
481 "ea_inode file size=%llu entry size=%zu",
482 i_size_read(ea_inode), size);
487 err = ext4_xattr_inode_read(ea_inode, buffer, size);
491 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
492 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
495 ext4_warning_inode(ea_inode,
496 "EA inode hash validation failed");
501 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
502 ext4_xattr_inode_get_hash(ea_inode),
503 ea_inode->i_ino, true /* reusable */);
511 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
512 void *buffer, size_t buffer_size)
514 struct buffer_head *bh = NULL;
515 struct ext4_xattr_entry *entry;
519 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
521 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
522 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 = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
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);
697 if (!EXT4_I(inode)->i_file_acl)
699 ea_idebug(inode, "reading block %llu",
700 (unsigned long long)EXT4_I(inode)->i_file_acl);
701 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
704 ea_bdebug(bh, "b_count=%d, refcount=%d",
705 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
706 error = ext4_xattr_check_block(inode, bh);
709 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
710 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
718 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
720 struct inode *inode = d_inode(dentry);
721 struct ext4_xattr_ibody_header *header;
722 struct ext4_inode *raw_inode;
723 struct ext4_iloc iloc;
727 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
729 error = ext4_get_inode_loc(inode, &iloc);
732 raw_inode = ext4_raw_inode(&iloc);
733 header = IHDR(inode, raw_inode);
734 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
735 error = xattr_check_inode(inode, header, end);
738 error = ext4_xattr_list_entries(dentry, IFIRST(header),
739 buffer, buffer_size);
747 * Inode operation listxattr()
749 * d_inode(dentry)->i_rwsem: don't care
751 * Copy a list of attribute names into the buffer
752 * provided, or compute the buffer size required.
753 * Buffer is NULL to compute the size of the buffer required.
755 * Returns a negative error number on failure, or the number of bytes
756 * used / required on success.
759 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
763 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
764 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
771 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
776 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
781 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
784 static void ext4_xattr_update_super_block(handle_t *handle,
785 struct super_block *sb)
787 if (ext4_has_feature_xattr(sb))
790 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
791 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
792 ext4_set_feature_xattr(sb);
793 ext4_handle_dirty_super(handle, sb);
797 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
799 struct ext4_iloc iloc = { .bh = NULL };
800 struct buffer_head *bh = NULL;
801 struct ext4_inode *raw_inode;
802 struct ext4_xattr_ibody_header *header;
803 struct ext4_xattr_entry *entry;
804 qsize_t ea_inode_refs = 0;
808 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
810 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
811 ret = ext4_get_inode_loc(inode, &iloc);
814 raw_inode = ext4_raw_inode(&iloc);
815 header = IHDR(inode, raw_inode);
816 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
817 ret = xattr_check_inode(inode, header, end);
821 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
822 entry = EXT4_XATTR_NEXT(entry))
823 if (entry->e_value_inum)
827 if (EXT4_I(inode)->i_file_acl) {
828 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
835 ret = ext4_xattr_check_block(inode, bh);
839 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
840 entry = EXT4_XATTR_NEXT(entry))
841 if (entry->e_value_inum)
844 *usage = ea_inode_refs + 1;
852 static inline size_t round_up_cluster(struct inode *inode, size_t length)
854 struct super_block *sb = inode->i_sb;
855 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
857 size_t mask = ~(cluster_size - 1);
859 return (length + cluster_size - 1) & mask;
862 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
866 err = dquot_alloc_inode(inode);
869 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
871 dquot_free_inode(inode);
875 static void ext4_xattr_inode_free_quota(struct inode *parent,
876 struct inode *ea_inode,
880 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
882 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
883 dquot_free_inode(parent);
886 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
887 struct buffer_head *block_bh, size_t value_len,
894 * 1) Owner inode update
895 * 2) Ref count update on old xattr block
897 * 4) block bitmap update for new xattr block
898 * 5) group descriptor for new xattr block
899 * 6) block bitmap update for old xattr block
900 * 7) group descriptor for old block
902 * 6 & 7 can happen if we have two racing threads T_a and T_b
903 * which are each trying to set an xattr on inodes I_a and I_b
904 * which were both initially sharing an xattr block.
909 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
912 * In case of inline data, we may push out the data to a block,
913 * so we need to reserve credits for this eventuality
915 if (inode && ext4_has_inline_data(inode))
916 credits += ext4_writepage_trans_blocks(inode) + 1;
918 /* We are done if ea_inode feature is not enabled. */
919 if (!ext4_has_feature_ea_inode(sb))
922 /* New ea_inode, inode map, block bitmap, group descriptor. */
926 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
928 /* Indirection block or one level of extent tree. */
931 /* Block bitmap and group descriptor updates for each block. */
932 credits += blocks * 2;
934 /* Blocks themselves. */
938 /* Dereference ea_inode holding old xattr value.
939 * Old ea_inode, inode map, block bitmap, group descriptor.
943 /* Data blocks for old ea_inode. */
944 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
946 /* Indirection block or one level of extent tree for old
951 /* Block bitmap and group descriptor updates for each block. */
952 credits += blocks * 2;
955 /* We may need to clone the existing xattr block in which case we need
956 * to increment ref counts for existing ea_inodes referenced by it.
959 struct ext4_xattr_entry *entry = BFIRST(block_bh);
961 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
962 if (entry->e_value_inum)
963 /* Ref count update on ea_inode. */
969 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
970 int credits, struct buffer_head *bh,
971 bool dirty, bool block_csum)
975 if (!ext4_handle_valid(handle))
978 if (handle->h_buffer_credits >= credits)
981 error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
985 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
991 ext4_xattr_block_csum_set(inode, bh);
992 error = ext4_handle_dirty_metadata(handle, NULL, bh);
994 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1000 error = ext4_journal_restart(handle, credits);
1002 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1007 error = ext4_journal_get_write_access(handle, bh);
1009 ext4_warning(inode->i_sb,
1010 "Get write access failed (error %d)",
1018 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1021 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1022 struct ext4_iloc iloc;
1027 inode_lock(ea_inode);
1029 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1035 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1036 ref_count += ref_change;
1037 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1039 if (ref_change > 0) {
1040 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1041 ea_inode->i_ino, ref_count);
1043 if (ref_count == 1) {
1044 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1045 ea_inode->i_ino, ea_inode->i_nlink);
1047 set_nlink(ea_inode, 1);
1048 ext4_orphan_del(handle, ea_inode);
1050 if (ea_inode_cache) {
1051 hash = ext4_xattr_inode_get_hash(ea_inode);
1052 mb_cache_entry_create(ea_inode_cache,
1055 true /* reusable */);
1059 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1060 ea_inode->i_ino, ref_count);
1062 if (ref_count == 0) {
1063 WARN_ONCE(ea_inode->i_nlink != 1,
1064 "EA inode %lu i_nlink=%u",
1065 ea_inode->i_ino, ea_inode->i_nlink);
1067 clear_nlink(ea_inode);
1068 ext4_orphan_add(handle, ea_inode);
1070 if (ea_inode_cache) {
1071 hash = ext4_xattr_inode_get_hash(ea_inode);
1072 mb_cache_entry_delete(ea_inode_cache, hash,
1078 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1081 ext4_warning_inode(ea_inode,
1082 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1085 inode_unlock(ea_inode);
1089 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1091 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1094 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1096 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1099 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1100 struct ext4_xattr_entry *first)
1102 struct inode *ea_inode;
1103 struct ext4_xattr_entry *entry;
1104 struct ext4_xattr_entry *failed_entry;
1105 unsigned int ea_ino;
1108 for (entry = first; !IS_LAST_ENTRY(entry);
1109 entry = EXT4_XATTR_NEXT(entry)) {
1110 if (!entry->e_value_inum)
1112 ea_ino = le32_to_cpu(entry->e_value_inum);
1113 err = ext4_xattr_inode_iget(parent, ea_ino,
1114 le32_to_cpu(entry->e_hash),
1118 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1120 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1130 failed_entry = entry;
1132 for (entry = first; entry != failed_entry;
1133 entry = EXT4_XATTR_NEXT(entry)) {
1134 if (!entry->e_value_inum)
1136 ea_ino = le32_to_cpu(entry->e_value_inum);
1137 err = ext4_xattr_inode_iget(parent, ea_ino,
1138 le32_to_cpu(entry->e_hash),
1141 ext4_warning(parent->i_sb,
1142 "cleanup ea_ino %u iget error %d", ea_ino,
1146 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1148 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1156 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1157 struct buffer_head *bh,
1158 struct ext4_xattr_entry *first, bool block_csum,
1159 struct ext4_xattr_inode_array **ea_inode_array,
1160 int extra_credits, bool skip_quota)
1162 struct inode *ea_inode;
1163 struct ext4_xattr_entry *entry;
1165 unsigned int ea_ino;
1169 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1170 credits = 2 + extra_credits;
1172 for (entry = first; !IS_LAST_ENTRY(entry);
1173 entry = EXT4_XATTR_NEXT(entry)) {
1174 if (!entry->e_value_inum)
1176 ea_ino = le32_to_cpu(entry->e_value_inum);
1177 err = ext4_xattr_inode_iget(parent, ea_ino,
1178 le32_to_cpu(entry->e_hash),
1183 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1185 ext4_warning_inode(ea_inode,
1186 "Expand inode array err=%d", err);
1191 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1194 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1199 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1201 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1207 ext4_xattr_inode_free_quota(parent, ea_inode,
1208 le32_to_cpu(entry->e_value_size));
1211 * Forget about ea_inode within the same transaction that
1212 * decrements the ref count. This avoids duplicate decrements in
1213 * case the rest of the work spills over to subsequent
1216 entry->e_value_inum = 0;
1217 entry->e_value_size = 0;
1224 * Note that we are deliberately skipping csum calculation for
1225 * the final update because we do not expect any journal
1226 * restarts until xattr block is freed.
1229 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1231 ext4_warning_inode(parent,
1232 "handle dirty metadata err=%d", err);
1237 * Release the xattr block BH: If the reference count is > 1, decrement it;
1238 * otherwise free the block.
1241 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1242 struct buffer_head *bh,
1243 struct ext4_xattr_inode_array **ea_inode_array,
1246 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1250 BUFFER_TRACE(bh, "get_write_access");
1251 error = ext4_journal_get_write_access(handle, bh);
1256 hash = le32_to_cpu(BHDR(bh)->h_hash);
1257 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1259 ea_bdebug(bh, "refcount now=0; freeing");
1261 * This must happen under buffer lock for
1262 * ext4_xattr_block_set() to reliably detect freed block
1265 mb_cache_entry_delete(ea_block_cache, hash,
1270 if (ext4_has_feature_ea_inode(inode->i_sb))
1271 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1273 true /* block_csum */,
1276 true /* skip_quota */);
1277 ext4_free_blocks(handle, inode, bh, 0, 1,
1278 EXT4_FREE_BLOCKS_METADATA |
1279 EXT4_FREE_BLOCKS_FORGET);
1282 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1283 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1284 struct mb_cache_entry *ce;
1286 if (ea_block_cache) {
1287 ce = mb_cache_entry_get(ea_block_cache, hash,
1291 mb_cache_entry_put(ea_block_cache, ce);
1296 ext4_xattr_block_csum_set(inode, bh);
1298 * Beware of this ugliness: Releasing of xattr block references
1299 * from different inodes can race and so we have to protect
1300 * from a race where someone else frees the block (and releases
1301 * its journal_head) before we are done dirtying the buffer. In
1302 * nojournal mode this race is harmless and we actually cannot
1303 * call ext4_handle_dirty_metadata() with locked buffer as
1304 * that function can call sync_dirty_buffer() so for that case
1305 * we handle the dirtying after unlocking the buffer.
1307 if (ext4_handle_valid(handle))
1308 error = ext4_handle_dirty_metadata(handle, inode, bh);
1310 if (!ext4_handle_valid(handle))
1311 error = ext4_handle_dirty_metadata(handle, inode, bh);
1313 ext4_handle_sync(handle);
1314 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1315 ea_bdebug(bh, "refcount now=%d; releasing",
1316 le32_to_cpu(BHDR(bh)->h_refcount));
1319 ext4_std_error(inode->i_sb, error);
1324 * Find the available free space for EAs. This also returns the total number of
1325 * bytes used by EA entries.
1327 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1328 size_t *min_offs, void *base, int *total)
1330 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1331 if (!last->e_value_inum && last->e_value_size) {
1332 size_t offs = le16_to_cpu(last->e_value_offs);
1333 if (offs < *min_offs)
1337 *total += EXT4_XATTR_LEN(last->e_name_len);
1339 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1343 * Write the value of the EA in an inode.
1345 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1346 const void *buf, int bufsize)
1348 struct buffer_head *bh = NULL;
1349 unsigned long block = 0;
1350 int blocksize = ea_inode->i_sb->s_blocksize;
1351 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1352 int csize, wsize = 0;
1357 while (ret >= 0 && ret < max_blocks) {
1358 struct ext4_map_blocks map;
1359 map.m_lblk = block += ret;
1360 map.m_len = max_blocks -= ret;
1362 ret = ext4_map_blocks(handle, ea_inode, &map,
1363 EXT4_GET_BLOCKS_CREATE);
1365 ext4_mark_inode_dirty(handle, ea_inode);
1366 if (ret == -ENOSPC &&
1367 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1379 while (wsize < bufsize) {
1382 csize = (bufsize - wsize) > blocksize ? blocksize :
1384 bh = ext4_getblk(handle, ea_inode, block, 0);
1389 EXT4_ERROR_INODE(ea_inode,
1390 "ext4_getblk() return bh = NULL");
1391 return -EFSCORRUPTED;
1393 ret = ext4_journal_get_write_access(handle, bh);
1397 memcpy(bh->b_data, buf, csize);
1398 set_buffer_uptodate(bh);
1399 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1406 inode_lock(ea_inode);
1407 i_size_write(ea_inode, wsize);
1408 ext4_update_i_disksize(ea_inode, wsize);
1409 inode_unlock(ea_inode);
1411 ext4_mark_inode_dirty(handle, ea_inode);
1420 * Create an inode to store the value of a large EA.
1422 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1423 struct inode *inode, u32 hash)
1425 struct inode *ea_inode = NULL;
1426 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1430 * Let the next inode be the goal, so we try and allocate the EA inode
1431 * in the same group, or nearby one.
1433 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1434 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1436 if (!IS_ERR(ea_inode)) {
1437 ea_inode->i_op = &ext4_file_inode_operations;
1438 ea_inode->i_fop = &ext4_file_operations;
1439 ext4_set_aops(ea_inode);
1440 ext4_xattr_inode_set_class(ea_inode);
1441 unlock_new_inode(ea_inode);
1442 ext4_xattr_inode_set_ref(ea_inode, 1);
1443 ext4_xattr_inode_set_hash(ea_inode, hash);
1444 err = ext4_mark_inode_dirty(handle, ea_inode);
1446 err = ext4_inode_attach_jinode(ea_inode);
1449 return ERR_PTR(err);
1453 * Xattr inodes are shared therefore quota charging is performed
1454 * at a higher level.
1456 dquot_free_inode(ea_inode);
1457 dquot_drop(ea_inode);
1458 inode_lock(ea_inode);
1459 ea_inode->i_flags |= S_NOQUOTA;
1460 inode_unlock(ea_inode);
1466 static struct inode *
1467 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1468 size_t value_len, u32 hash)
1470 struct inode *ea_inode;
1471 struct mb_cache_entry *ce;
1472 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1475 if (!ea_inode_cache)
1478 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1482 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1483 !(current->flags & PF_MEMALLOC_NOFS));
1485 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1487 mb_cache_entry_put(ea_inode_cache, ce);
1492 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1494 if (!IS_ERR(ea_inode) &&
1495 !is_bad_inode(ea_inode) &&
1496 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1497 i_size_read(ea_inode) == value_len &&
1498 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1499 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1501 !memcmp(value, ea_data, value_len)) {
1502 mb_cache_entry_touch(ea_inode_cache, ce);
1503 mb_cache_entry_put(ea_inode_cache, ce);
1508 if (!IS_ERR(ea_inode))
1510 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1517 * Add value of the EA in an inode.
1519 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1520 const void *value, size_t value_len,
1521 struct inode **ret_inode)
1523 struct inode *ea_inode;
1527 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1528 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1530 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1536 *ret_inode = ea_inode;
1540 /* Create an inode for the EA value */
1541 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1542 if (IS_ERR(ea_inode))
1543 return PTR_ERR(ea_inode);
1545 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1547 ext4_xattr_inode_dec_ref(handle, ea_inode);
1552 if (EA_INODE_CACHE(inode))
1553 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1554 ea_inode->i_ino, true /* reusable */);
1556 *ret_inode = ea_inode;
1561 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1562 * feature is enabled.
1564 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1566 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1567 struct ext4_xattr_search *s,
1568 handle_t *handle, struct inode *inode,
1571 struct ext4_xattr_entry *last, *next;
1572 struct ext4_xattr_entry *here = s->here;
1573 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1574 int in_inode = i->in_inode;
1575 struct inode *old_ea_inode = NULL;
1576 struct inode *new_ea_inode = NULL;
1577 size_t old_size, new_size;
1580 /* Space used by old and new values. */
1581 old_size = (!s->not_found && !here->e_value_inum) ?
1582 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1583 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1586 * Optimization for the simple case when old and new values have the
1587 * same padded sizes. Not applicable if external inodes are involved.
1589 if (new_size && new_size == old_size) {
1590 size_t offs = le16_to_cpu(here->e_value_offs);
1591 void *val = s->base + offs;
1593 here->e_value_size = cpu_to_le32(i->value_len);
1594 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1595 memset(val, 0, new_size);
1597 memcpy(val, i->value, i->value_len);
1598 /* Clear padding bytes. */
1599 memset(val + i->value_len, 0, new_size - i->value_len);
1604 /* Compute min_offs and last. */
1606 for (; !IS_LAST_ENTRY(last); last = next) {
1607 next = EXT4_XATTR_NEXT(last);
1608 if ((void *)next >= s->end) {
1609 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1610 ret = -EFSCORRUPTED;
1613 if (!last->e_value_inum && last->e_value_size) {
1614 size_t offs = le16_to_cpu(last->e_value_offs);
1615 if (offs < min_offs)
1620 /* Check whether we have enough space. */
1624 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1626 free += EXT4_XATTR_LEN(name_len) + old_size;
1628 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1634 * If storing the value in an external inode is an option,
1635 * reserve space for xattr entries/names in the external
1636 * attribute block so that a long value does not occupy the
1637 * whole space and prevent futher entries being added.
1639 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1640 new_size && is_block &&
1641 (min_offs + old_size - new_size) <
1642 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1649 * Getting access to old and new ea inodes is subject to failures.
1650 * Finish that work before doing any modifications to the xattr data.
1652 if (!s->not_found && here->e_value_inum) {
1653 ret = ext4_xattr_inode_iget(inode,
1654 le32_to_cpu(here->e_value_inum),
1655 le32_to_cpu(here->e_hash),
1658 old_ea_inode = NULL;
1662 if (i->value && in_inode) {
1663 WARN_ON_ONCE(!i->value_len);
1665 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1669 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1673 new_ea_inode = NULL;
1674 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1680 /* We are ready to release ref count on the old_ea_inode. */
1681 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1683 /* Release newly required ref count on new_ea_inode. */
1687 err = ext4_xattr_inode_dec_ref(handle,
1690 ext4_warning_inode(new_ea_inode,
1691 "dec ref new_ea_inode err=%d",
1693 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1699 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1700 le32_to_cpu(here->e_value_size));
1703 /* No failures allowed past this point. */
1705 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1706 /* Remove the old value. */
1707 void *first_val = s->base + min_offs;
1708 size_t offs = le16_to_cpu(here->e_value_offs);
1709 void *val = s->base + offs;
1711 memmove(first_val + old_size, first_val, val - first_val);
1712 memset(first_val, 0, old_size);
1713 min_offs += old_size;
1715 /* Adjust all value offsets. */
1717 while (!IS_LAST_ENTRY(last)) {
1718 size_t o = le16_to_cpu(last->e_value_offs);
1720 if (!last->e_value_inum &&
1721 last->e_value_size && o < offs)
1722 last->e_value_offs = cpu_to_le16(o + old_size);
1723 last = EXT4_XATTR_NEXT(last);
1728 /* Remove old name. */
1729 size_t size = EXT4_XATTR_LEN(name_len);
1731 last = ENTRY((void *)last - size);
1732 memmove(here, (void *)here + size,
1733 (void *)last - (void *)here + sizeof(__u32));
1734 memset(last, 0, size);
1735 } else if (s->not_found) {
1736 /* Insert new name. */
1737 size_t size = EXT4_XATTR_LEN(name_len);
1738 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1740 memmove((void *)here + size, here, rest);
1741 memset(here, 0, size);
1742 here->e_name_index = i->name_index;
1743 here->e_name_len = name_len;
1744 memcpy(here->e_name, i->name, name_len);
1746 /* This is an update, reset value info. */
1747 here->e_value_inum = 0;
1748 here->e_value_offs = 0;
1749 here->e_value_size = 0;
1753 /* Insert new value. */
1755 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1756 } else if (i->value_len) {
1757 void *val = s->base + min_offs - new_size;
1759 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1760 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1761 memset(val, 0, new_size);
1763 memcpy(val, i->value, i->value_len);
1764 /* Clear padding bytes. */
1765 memset(val + i->value_len, 0,
1766 new_size - i->value_len);
1769 here->e_value_size = cpu_to_le32(i->value_len);
1776 /* Entry hash calculation. */
1781 * Feed crc32c hash instead of the raw value for entry
1782 * hash calculation. This is to avoid walking
1783 * potentially long value buffer again.
1785 crc32c_hash = cpu_to_le32(
1786 ext4_xattr_inode_get_hash(new_ea_inode));
1787 hash = ext4_xattr_hash_entry(here->e_name,
1790 } else if (is_block) {
1791 __le32 *value = s->base + le16_to_cpu(
1792 here->e_value_offs);
1794 hash = ext4_xattr_hash_entry(here->e_name,
1795 here->e_name_len, value,
1798 here->e_hash = hash;
1802 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1811 struct ext4_xattr_block_find {
1812 struct ext4_xattr_search s;
1813 struct buffer_head *bh;
1817 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1818 struct ext4_xattr_block_find *bs)
1820 struct super_block *sb = inode->i_sb;
1823 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1824 i->name_index, i->name, i->value, (long)i->value_len);
1826 if (EXT4_I(inode)->i_file_acl) {
1827 /* The inode already has an extended attribute block. */
1828 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1829 if (IS_ERR(bs->bh)) {
1830 error = PTR_ERR(bs->bh);
1834 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1835 atomic_read(&(bs->bh->b_count)),
1836 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1837 error = ext4_xattr_check_block(inode, bs->bh);
1840 /* Find the named attribute. */
1841 bs->s.base = BHDR(bs->bh);
1842 bs->s.first = BFIRST(bs->bh);
1843 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1844 bs->s.here = bs->s.first;
1845 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1846 i->name_index, i->name, 1);
1847 if (error && error != -ENODATA)
1849 bs->s.not_found = error;
1855 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1856 struct ext4_xattr_info *i,
1857 struct ext4_xattr_block_find *bs)
1859 struct super_block *sb = inode->i_sb;
1860 struct buffer_head *new_bh = NULL;
1861 struct ext4_xattr_search s_copy = bs->s;
1862 struct ext4_xattr_search *s = &s_copy;
1863 struct mb_cache_entry *ce = NULL;
1865 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1866 struct inode *ea_inode = NULL, *tmp_inode;
1867 size_t old_ea_inode_quota = 0;
1868 unsigned int ea_ino;
1871 #define header(x) ((struct ext4_xattr_header *)(x))
1874 BUFFER_TRACE(bs->bh, "get_write_access");
1875 error = ext4_journal_get_write_access(handle, bs->bh);
1878 lock_buffer(bs->bh);
1880 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1881 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1884 * This must happen under buffer lock for
1885 * ext4_xattr_block_set() to reliably detect modified
1889 mb_cache_entry_delete(ea_block_cache, hash,
1891 ea_bdebug(bs->bh, "modifying in-place");
1892 error = ext4_xattr_set_entry(i, s, handle, inode,
1893 true /* is_block */);
1894 ext4_xattr_block_csum_set(inode, bs->bh);
1895 unlock_buffer(bs->bh);
1896 if (error == -EFSCORRUPTED)
1899 error = ext4_handle_dirty_metadata(handle,
1906 int offset = (char *)s->here - bs->bh->b_data;
1908 unlock_buffer(bs->bh);
1909 ea_bdebug(bs->bh, "cloning");
1910 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1912 if (s->base == NULL)
1914 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1915 s->first = ENTRY(header(s->base)+1);
1916 header(s->base)->h_refcount = cpu_to_le32(1);
1917 s->here = ENTRY(s->base + offset);
1918 s->end = s->base + bs->bh->b_size;
1921 * If existing entry points to an xattr inode, we need
1922 * to prevent ext4_xattr_set_entry() from decrementing
1923 * ref count on it because the reference belongs to the
1924 * original block. In this case, make the entry look
1925 * like it has an empty value.
1927 if (!s->not_found && s->here->e_value_inum) {
1928 ea_ino = le32_to_cpu(s->here->e_value_inum);
1929 error = ext4_xattr_inode_iget(inode, ea_ino,
1930 le32_to_cpu(s->here->e_hash),
1935 if (!ext4_test_inode_state(tmp_inode,
1936 EXT4_STATE_LUSTRE_EA_INODE)) {
1938 * Defer quota free call for previous
1939 * inode until success is guaranteed.
1941 old_ea_inode_quota = le32_to_cpu(
1942 s->here->e_value_size);
1946 s->here->e_value_inum = 0;
1947 s->here->e_value_size = 0;
1951 /* Allocate a buffer where we construct the new block. */
1952 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1953 /* assert(header == s->base) */
1955 if (s->base == NULL)
1957 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1958 header(s->base)->h_blocks = cpu_to_le32(1);
1959 header(s->base)->h_refcount = cpu_to_le32(1);
1960 s->first = ENTRY(header(s->base)+1);
1961 s->here = ENTRY(header(s->base)+1);
1962 s->end = s->base + sb->s_blocksize;
1965 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1966 if (error == -EFSCORRUPTED)
1971 if (i->value && s->here->e_value_inum) {
1973 * A ref count on ea_inode has been taken as part of the call to
1974 * ext4_xattr_set_entry() above. We would like to drop this
1975 * extra ref but we have to wait until the xattr block is
1976 * initialized and has its own ref count on the ea_inode.
1978 ea_ino = le32_to_cpu(s->here->e_value_inum);
1979 error = ext4_xattr_inode_iget(inode, ea_ino,
1980 le32_to_cpu(s->here->e_hash),
1989 if (!IS_LAST_ENTRY(s->first)) {
1990 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1993 /* We found an identical block in the cache. */
1994 if (new_bh == bs->bh)
1995 ea_bdebug(new_bh, "keeping");
1999 WARN_ON_ONCE(dquot_initialize_needed(inode));
2001 /* The old block is released after updating
2003 error = dquot_alloc_block(inode,
2004 EXT4_C2B(EXT4_SB(sb), 1));
2007 BUFFER_TRACE(new_bh, "get_write_access");
2008 error = ext4_journal_get_write_access(handle,
2012 lock_buffer(new_bh);
2014 * We have to be careful about races with
2015 * freeing, rehashing or adding references to
2016 * xattr block. Once we hold buffer lock xattr
2017 * block's state is stable so we can check
2018 * whether the block got freed / rehashed or
2019 * not. Since we unhash mbcache entry under
2020 * buffer lock when freeing / rehashing xattr
2021 * block, checking whether entry is still
2022 * hashed is reliable. Same rules hold for
2023 * e_reusable handling.
2025 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2028 * Undo everything and check mbcache
2031 unlock_buffer(new_bh);
2032 dquot_free_block(inode,
2033 EXT4_C2B(EXT4_SB(sb),
2036 mb_cache_entry_put(ea_block_cache, ce);
2041 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2042 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2043 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2045 ea_bdebug(new_bh, "reusing; refcount now=%d",
2047 ext4_xattr_block_csum_set(inode, new_bh);
2048 unlock_buffer(new_bh);
2049 error = ext4_handle_dirty_metadata(handle,
2055 mb_cache_entry_touch(ea_block_cache, ce);
2056 mb_cache_entry_put(ea_block_cache, ce);
2058 } else if (bs->bh && s->base == bs->bh->b_data) {
2059 /* We were modifying this block in-place. */
2060 ea_bdebug(bs->bh, "keeping this block");
2061 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2065 /* We need to allocate a new block */
2066 ext4_fsblk_t goal, block;
2068 WARN_ON_ONCE(dquot_initialize_needed(inode));
2070 goal = ext4_group_first_block_no(sb,
2071 EXT4_I(inode)->i_block_group);
2073 /* non-extent files can't have physical blocks past 2^32 */
2074 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2075 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2077 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2082 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2083 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2085 ea_idebug(inode, "creating block %llu",
2086 (unsigned long long)block);
2088 new_bh = sb_getblk(sb, block);
2089 if (unlikely(!new_bh)) {
2092 ext4_free_blocks(handle, inode, NULL, block, 1,
2093 EXT4_FREE_BLOCKS_METADATA);
2096 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2097 ENTRY(header(s->base)+1));
2101 /* Drop the extra ref on ea_inode. */
2102 error = ext4_xattr_inode_dec_ref(handle,
2105 ext4_warning_inode(ea_inode,
2112 lock_buffer(new_bh);
2113 error = ext4_journal_get_create_access(handle, new_bh);
2115 unlock_buffer(new_bh);
2119 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2120 ext4_xattr_block_csum_set(inode, new_bh);
2121 set_buffer_uptodate(new_bh);
2122 unlock_buffer(new_bh);
2123 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2124 error = ext4_handle_dirty_metadata(handle, inode,
2131 if (old_ea_inode_quota)
2132 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2134 /* Update the inode. */
2135 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2137 /* Drop the previous xattr block. */
2138 if (bs->bh && bs->bh != new_bh) {
2139 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2141 ext4_xattr_release_block(handle, inode, bs->bh,
2143 0 /* extra_credits */);
2144 ext4_xattr_inode_array_free(ea_inode_array);
2152 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2154 ext4_warning_inode(ea_inode, "dec ref error=%d",
2157 /* If there was an error, revert the quota charge. */
2159 ext4_xattr_inode_free_quota(inode, ea_inode,
2160 i_size_read(ea_inode));
2164 mb_cache_entry_put(ea_block_cache, ce);
2166 if (!(bs->bh && s->base == bs->bh->b_data))
2172 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2176 EXT4_ERROR_INODE(inode, "bad block %llu",
2177 EXT4_I(inode)->i_file_acl);
2183 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2184 struct ext4_xattr_ibody_find *is)
2186 struct ext4_xattr_ibody_header *header;
2187 struct ext4_inode *raw_inode;
2190 if (EXT4_I(inode)->i_extra_isize == 0)
2192 raw_inode = ext4_raw_inode(&is->iloc);
2193 header = IHDR(inode, raw_inode);
2194 is->s.base = is->s.first = IFIRST(header);
2195 is->s.here = is->s.first;
2196 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2197 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2198 error = xattr_check_inode(inode, header, is->s.end);
2201 /* Find the named attribute. */
2202 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2203 i->name_index, i->name, 0);
2204 if (error && error != -ENODATA)
2206 is->s.not_found = error;
2211 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2212 struct ext4_xattr_info *i,
2213 struct ext4_xattr_ibody_find *is)
2215 struct ext4_xattr_ibody_header *header;
2216 struct ext4_xattr_search *s = &is->s;
2219 if (EXT4_I(inode)->i_extra_isize == 0)
2221 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2224 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2225 if (!IS_LAST_ENTRY(s->first)) {
2226 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2227 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2229 header->h_magic = cpu_to_le32(0);
2230 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2235 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2236 struct ext4_xattr_info *i,
2237 struct ext4_xattr_ibody_find *is)
2239 struct ext4_xattr_ibody_header *header;
2240 struct ext4_xattr_search *s = &is->s;
2243 if (EXT4_I(inode)->i_extra_isize == 0)
2245 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2248 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2249 if (!IS_LAST_ENTRY(s->first)) {
2250 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2251 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2253 header->h_magic = cpu_to_le32(0);
2254 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2259 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2260 struct ext4_xattr_info *i)
2264 /* When e_value_inum is set the value is stored externally. */
2265 if (s->here->e_value_inum)
2267 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2269 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2270 return !memcmp(value, i->value, i->value_len);
2273 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2275 struct buffer_head *bh;
2278 if (!EXT4_I(inode)->i_file_acl)
2280 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2283 error = ext4_xattr_check_block(inode, bh);
2286 return ERR_PTR(error);
2292 * ext4_xattr_set_handle()
2294 * Create, replace or remove an extended attribute for this inode. Value
2295 * is NULL to remove an existing extended attribute, and non-NULL to
2296 * either replace an existing extended attribute, or create a new extended
2297 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2298 * specify that an extended attribute must exist and must not exist
2299 * previous to the call, respectively.
2301 * Returns 0, or a negative error number on failure.
2304 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2305 const char *name, const void *value, size_t value_len,
2308 struct ext4_xattr_info i = {
2309 .name_index = name_index,
2312 .value_len = value_len,
2315 struct ext4_xattr_ibody_find is = {
2316 .s = { .not_found = -ENODATA, },
2318 struct ext4_xattr_block_find bs = {
2319 .s = { .not_found = -ENODATA, },
2326 if (strlen(name) > 255)
2329 ext4_write_lock_xattr(inode, &no_expand);
2331 /* Check journal credits under write lock. */
2332 if (ext4_handle_valid(handle)) {
2333 struct buffer_head *bh;
2336 bh = ext4_xattr_get_block(inode);
2338 error = PTR_ERR(bh);
2342 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2344 flags & XATTR_CREATE);
2347 if (!ext4_handle_has_enough_credits(handle, credits)) {
2351 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2354 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2358 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2359 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2360 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2361 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2364 error = ext4_xattr_ibody_find(inode, &i, &is);
2368 error = ext4_xattr_block_find(inode, &i, &bs);
2371 if (is.s.not_found && bs.s.not_found) {
2373 if (flags & XATTR_REPLACE)
2380 if (flags & XATTR_CREATE)
2385 if (!is.s.not_found)
2386 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2387 else if (!bs.s.not_found)
2388 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2391 /* Xattr value did not change? Save us some work and bail out */
2392 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2394 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2397 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2398 (EXT4_XATTR_SIZE(i.value_len) >
2399 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2402 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2403 if (!error && !bs.s.not_found) {
2405 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2406 } else if (error == -ENOSPC) {
2407 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2410 error = ext4_xattr_block_find(inode, &i, &bs);
2414 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2415 if (!error && !is.s.not_found) {
2417 error = ext4_xattr_ibody_set(handle, inode, &i,
2419 } else if (error == -ENOSPC) {
2421 * Xattr does not fit in the block, store at
2422 * external inode if possible.
2424 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2425 i.value_len && !i.in_inode) {
2433 ext4_xattr_update_super_block(handle, inode->i_sb);
2434 inode->i_ctime = current_time(inode);
2437 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2439 * The bh is consumed by ext4_mark_iloc_dirty, even with
2444 ext4_handle_sync(handle);
2450 ext4_write_unlock_xattr(inode, &no_expand);
2454 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2455 bool is_create, int *credits)
2457 struct buffer_head *bh;
2462 if (!EXT4_SB(inode->i_sb)->s_journal)
2465 down_read(&EXT4_I(inode)->xattr_sem);
2467 bh = ext4_xattr_get_block(inode);
2471 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2472 value_len, is_create);
2477 up_read(&EXT4_I(inode)->xattr_sem);
2484 * Like ext4_xattr_set_handle, but start from an inode. This extended
2485 * attribute modification is a filesystem transaction by itself.
2487 * Returns 0, or a negative error number on failure.
2490 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2491 const void *value, size_t value_len, int flags)
2494 struct super_block *sb = inode->i_sb;
2495 int error, retries = 0;
2498 error = dquot_initialize(inode);
2503 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2508 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2509 if (IS_ERR(handle)) {
2510 error = PTR_ERR(handle);
2514 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2515 value, value_len, flags);
2516 error2 = ext4_journal_stop(handle);
2517 if (error == -ENOSPC &&
2518 ext4_should_retry_alloc(sb, &retries))
2528 * Shift the EA entries in the inode to create space for the increased
2531 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2532 int value_offs_shift, void *to,
2533 void *from, size_t n)
2535 struct ext4_xattr_entry *last = entry;
2538 /* We always shift xattr headers further thus offsets get lower */
2539 BUG_ON(value_offs_shift > 0);
2541 /* Adjust the value offsets of the entries */
2542 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2543 if (!last->e_value_inum && last->e_value_size) {
2544 new_offs = le16_to_cpu(last->e_value_offs) +
2546 last->e_value_offs = cpu_to_le16(new_offs);
2549 /* Shift the entries by n bytes */
2550 memmove(to, from, n);
2554 * Move xattr pointed to by 'entry' from inode into external xattr block
2556 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2557 struct ext4_inode *raw_inode,
2558 struct ext4_xattr_entry *entry)
2560 struct ext4_xattr_ibody_find *is = NULL;
2561 struct ext4_xattr_block_find *bs = NULL;
2562 char *buffer = NULL, *b_entry_name = NULL;
2563 size_t value_size = le32_to_cpu(entry->e_value_size);
2564 struct ext4_xattr_info i = {
2567 .name_index = entry->e_name_index,
2568 .in_inode = !!entry->e_value_inum,
2570 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2573 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2574 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2575 buffer = kmalloc(value_size, GFP_NOFS);
2576 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2577 if (!is || !bs || !buffer || !b_entry_name) {
2582 is->s.not_found = -ENODATA;
2583 bs->s.not_found = -ENODATA;
2587 /* Save the entry name and the entry value */
2588 if (entry->e_value_inum) {
2589 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2593 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2594 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2597 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2598 b_entry_name[entry->e_name_len] = '\0';
2599 i.name = b_entry_name;
2601 error = ext4_get_inode_loc(inode, &is->iloc);
2605 error = ext4_xattr_ibody_find(inode, &i, is);
2609 /* Remove the chosen entry from the inode */
2610 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2615 i.value_len = value_size;
2616 error = ext4_xattr_block_find(inode, &i, bs);
2620 /* Add entry which was removed from the inode into the block */
2621 error = ext4_xattr_block_set(handle, inode, &i, bs);
2626 kfree(b_entry_name);
2629 brelse(is->iloc.bh);
2638 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2639 struct ext4_inode *raw_inode,
2640 int isize_diff, size_t ifree,
2641 size_t bfree, int *total_ino)
2643 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2644 struct ext4_xattr_entry *small_entry;
2645 struct ext4_xattr_entry *entry;
2646 struct ext4_xattr_entry *last;
2647 unsigned int entry_size; /* EA entry size */
2648 unsigned int total_size; /* EA entry size + value size */
2649 unsigned int min_total_size;
2652 while (isize_diff > ifree) {
2655 min_total_size = ~0U;
2656 last = IFIRST(header);
2657 /* Find the entry best suited to be pushed into EA block */
2658 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2659 /* never move system.data out of the inode */
2660 if ((last->e_name_len == 4) &&
2661 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2662 !memcmp(last->e_name, "data", 4))
2664 total_size = EXT4_XATTR_LEN(last->e_name_len);
2665 if (!last->e_value_inum)
2666 total_size += EXT4_XATTR_SIZE(
2667 le32_to_cpu(last->e_value_size));
2668 if (total_size <= bfree &&
2669 total_size < min_total_size) {
2670 if (total_size + ifree < isize_diff) {
2674 min_total_size = total_size;
2679 if (entry == NULL) {
2680 if (small_entry == NULL)
2682 entry = small_entry;
2685 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2686 total_size = entry_size;
2687 if (!entry->e_value_inum)
2688 total_size += EXT4_XATTR_SIZE(
2689 le32_to_cpu(entry->e_value_size));
2690 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2695 *total_ino -= entry_size;
2696 ifree += total_size;
2697 bfree -= total_size;
2704 * Expand an inode by new_extra_isize bytes when EAs are present.
2705 * Returns 0 on success or negative error number on failure.
2707 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2708 struct ext4_inode *raw_inode, handle_t *handle)
2710 struct ext4_xattr_ibody_header *header;
2711 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2712 static unsigned int mnt_count;
2714 size_t ifree, bfree;
2717 int error = 0, tried_min_extra_isize = 0;
2718 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2719 int isize_diff; /* How much do we need to grow i_extra_isize */
2722 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2723 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2726 header = IHDR(inode, raw_inode);
2729 * Check if enough free space is available in the inode to shift the
2730 * entries ahead by new_extra_isize.
2733 base = IFIRST(header);
2734 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2735 min_offs = end - base;
2736 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2738 error = xattr_check_inode(inode, header, end);
2742 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2743 if (ifree >= isize_diff)
2747 * Enough free space isn't available in the inode, check if
2748 * EA block can hold new_extra_isize bytes.
2750 if (EXT4_I(inode)->i_file_acl) {
2751 struct buffer_head *bh;
2753 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2755 error = PTR_ERR(bh);
2758 error = ext4_xattr_check_block(inode, bh);
2764 end = bh->b_data + bh->b_size;
2765 min_offs = end - base;
2766 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2769 if (bfree + ifree < isize_diff) {
2770 if (!tried_min_extra_isize && s_min_extra_isize) {
2771 tried_min_extra_isize++;
2772 new_extra_isize = s_min_extra_isize;
2779 bfree = inode->i_sb->s_blocksize;
2782 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2783 isize_diff, ifree, bfree,
2786 if (error == -ENOSPC && !tried_min_extra_isize &&
2787 s_min_extra_isize) {
2788 tried_min_extra_isize++;
2789 new_extra_isize = s_min_extra_isize;
2795 /* Adjust the offsets and shift the remaining entries ahead */
2796 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2797 - new_extra_isize, (void *)raw_inode +
2798 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2799 (void *)header, total_ino);
2800 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2803 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2804 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2806 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2811 #define EIA_INCR 16 /* must be 2^n */
2812 #define EIA_MASK (EIA_INCR - 1)
2814 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2815 * If @ea_inode_array is new or full it will be grown and the old
2816 * contents copied over.
2819 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2820 struct inode *inode)
2822 if (*ea_inode_array == NULL) {
2824 * Start with 15 inodes, so it fits into a power-of-two size.
2825 * If *ea_inode_array is NULL, this is essentially offsetof()
2828 kmalloc(offsetof(struct ext4_xattr_inode_array,
2831 if (*ea_inode_array == NULL)
2833 (*ea_inode_array)->count = 0;
2834 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2835 /* expand the array once all 15 + n * 16 slots are full */
2836 struct ext4_xattr_inode_array *new_array = NULL;
2837 int count = (*ea_inode_array)->count;
2839 /* if new_array is NULL, this is essentially offsetof() */
2840 new_array = kmalloc(
2841 offsetof(struct ext4_xattr_inode_array,
2842 inodes[count + EIA_INCR]),
2844 if (new_array == NULL)
2846 memcpy(new_array, *ea_inode_array,
2847 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2848 kfree(*ea_inode_array);
2849 *ea_inode_array = new_array;
2851 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2856 * ext4_xattr_delete_inode()
2858 * Free extended attribute resources associated with this inode. Traverse
2859 * all entries and decrement reference on any xattr inodes associated with this
2860 * inode. This is called immediately before an inode is freed. We have exclusive
2861 * access to the inode. If an orphan inode is deleted it will also release its
2862 * references on xattr block and xattr inodes.
2864 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2865 struct ext4_xattr_inode_array **ea_inode_array,
2868 struct buffer_head *bh = NULL;
2869 struct ext4_xattr_ibody_header *header;
2870 struct ext4_iloc iloc = { .bh = NULL };
2871 struct ext4_xattr_entry *entry;
2872 struct inode *ea_inode;
2875 error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2878 false /* block_csum */);
2880 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2884 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2885 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2887 error = ext4_get_inode_loc(inode, &iloc);
2889 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2893 error = ext4_journal_get_write_access(handle, iloc.bh);
2895 EXT4_ERROR_INODE(inode, "write access (error %d)",
2900 header = IHDR(inode, ext4_raw_inode(&iloc));
2901 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2902 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2904 false /* block_csum */,
2907 false /* skip_quota */);
2910 if (EXT4_I(inode)->i_file_acl) {
2911 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2913 error = PTR_ERR(bh);
2915 EXT4_ERROR_INODE(inode, "block %llu read error",
2916 EXT4_I(inode)->i_file_acl);
2920 error = ext4_xattr_check_block(inode, bh);
2924 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2925 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2926 entry = EXT4_XATTR_NEXT(entry)) {
2927 if (!entry->e_value_inum)
2929 error = ext4_xattr_inode_iget(inode,
2930 le32_to_cpu(entry->e_value_inum),
2931 le32_to_cpu(entry->e_hash),
2935 ext4_xattr_inode_free_quota(inode, ea_inode,
2936 le32_to_cpu(entry->e_value_size));
2942 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2945 * Update i_file_acl value in the same transaction that releases
2948 EXT4_I(inode)->i_file_acl = 0;
2949 error = ext4_mark_inode_dirty(handle, inode);
2951 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2963 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2967 if (ea_inode_array == NULL)
2970 for (idx = 0; idx < ea_inode_array->count; ++idx)
2971 iput(ea_inode_array->inodes[idx]);
2972 kfree(ea_inode_array);
2976 * ext4_xattr_block_cache_insert()
2978 * Create a new entry in the extended attribute block cache, and insert
2979 * it unless such an entry is already in the cache.
2981 * Returns 0, or a negative error number on failure.
2984 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2985 struct buffer_head *bh)
2987 struct ext4_xattr_header *header = BHDR(bh);
2988 __u32 hash = le32_to_cpu(header->h_hash);
2989 int reusable = le32_to_cpu(header->h_refcount) <
2990 EXT4_XATTR_REFCOUNT_MAX;
2993 if (!ea_block_cache)
2995 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2996 bh->b_blocknr, reusable);
2998 if (error == -EBUSY)
2999 ea_bdebug(bh, "already in cache");
3001 ea_bdebug(bh, "inserting [%x]", (int)hash);
3007 * Compare two extended attribute blocks for equality.
3009 * Returns 0 if the blocks are equal, 1 if they differ, and
3010 * a negative error number on errors.
3013 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3014 struct ext4_xattr_header *header2)
3016 struct ext4_xattr_entry *entry1, *entry2;
3018 entry1 = ENTRY(header1+1);
3019 entry2 = ENTRY(header2+1);
3020 while (!IS_LAST_ENTRY(entry1)) {
3021 if (IS_LAST_ENTRY(entry2))
3023 if (entry1->e_hash != entry2->e_hash ||
3024 entry1->e_name_index != entry2->e_name_index ||
3025 entry1->e_name_len != entry2->e_name_len ||
3026 entry1->e_value_size != entry2->e_value_size ||
3027 entry1->e_value_inum != entry2->e_value_inum ||
3028 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3030 if (!entry1->e_value_inum &&
3031 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3032 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3033 le32_to_cpu(entry1->e_value_size)))
3036 entry1 = EXT4_XATTR_NEXT(entry1);
3037 entry2 = EXT4_XATTR_NEXT(entry2);
3039 if (!IS_LAST_ENTRY(entry2))
3045 * ext4_xattr_block_cache_find()
3047 * Find an identical extended attribute block.
3049 * Returns a pointer to the block found, or NULL if such a block was
3050 * not found or an error occurred.
3052 static struct buffer_head *
3053 ext4_xattr_block_cache_find(struct inode *inode,
3054 struct ext4_xattr_header *header,
3055 struct mb_cache_entry **pce)
3057 __u32 hash = le32_to_cpu(header->h_hash);
3058 struct mb_cache_entry *ce;
3059 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3061 if (!ea_block_cache)
3063 if (!header->h_hash)
3064 return NULL; /* never share */
3065 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3066 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3068 struct buffer_head *bh;
3070 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3072 if (PTR_ERR(bh) == -ENOMEM)
3075 EXT4_ERROR_INODE(inode, "block %lu read error",
3076 (unsigned long)ce->e_value);
3077 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3082 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3087 #define NAME_HASH_SHIFT 5
3088 #define VALUE_HASH_SHIFT 16
3091 * ext4_xattr_hash_entry()
3093 * Compute the hash of an extended attribute.
3095 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3100 while (name_len--) {
3101 hash = (hash << NAME_HASH_SHIFT) ^
3102 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3105 while (value_count--) {
3106 hash = (hash << VALUE_HASH_SHIFT) ^
3107 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3108 le32_to_cpu(*value++);
3110 return cpu_to_le32(hash);
3113 #undef NAME_HASH_SHIFT
3114 #undef VALUE_HASH_SHIFT
3116 #define BLOCK_HASH_SHIFT 16
3119 * ext4_xattr_rehash()
3121 * Re-compute the extended attribute hash value after an entry has changed.
3123 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3125 struct ext4_xattr_entry *here;
3128 here = ENTRY(header+1);
3129 while (!IS_LAST_ENTRY(here)) {
3130 if (!here->e_hash) {
3131 /* Block is not shared if an entry's hash value == 0 */
3135 hash = (hash << BLOCK_HASH_SHIFT) ^
3136 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3137 le32_to_cpu(here->e_hash);
3138 here = EXT4_XATTR_NEXT(here);
3140 header->h_hash = cpu_to_le32(hash);
3143 #undef BLOCK_HASH_SHIFT
3145 #define HASH_BUCKET_BITS 10
3148 ext4_xattr_create_cache(void)
3150 return mb_cache_create(HASH_BUCKET_BITS);
3153 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3156 mb_cache_destroy(cache);