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,
96 [EXT4_XATTR_INDEX_HURD] = &ext4_xattr_hurd_handler,
99 const struct xattr_handler *ext4_xattr_handlers[] = {
100 &ext4_xattr_user_handler,
101 &ext4_xattr_trusted_handler,
102 #ifdef CONFIG_EXT4_FS_POSIX_ACL
103 &posix_acl_access_xattr_handler,
104 &posix_acl_default_xattr_handler,
106 #ifdef CONFIG_EXT4_FS_SECURITY
107 &ext4_xattr_security_handler,
109 &ext4_xattr_hurd_handler,
113 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
114 inode->i_sb->s_fs_info)->s_ea_block_cache)
116 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
117 inode->i_sb->s_fs_info)->s_ea_inode_cache)
120 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
121 struct inode *inode);
123 #ifdef CONFIG_LOCKDEP
124 void ext4_xattr_inode_set_class(struct inode *ea_inode)
126 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
130 static __le32 ext4_xattr_block_csum(struct inode *inode,
132 struct ext4_xattr_header *hdr)
134 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
136 __le64 dsk_block_nr = cpu_to_le64(block_nr);
137 __u32 dummy_csum = 0;
138 int offset = offsetof(struct ext4_xattr_header, h_checksum);
140 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
141 sizeof(dsk_block_nr));
142 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
143 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
144 offset += sizeof(dummy_csum);
145 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
146 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
148 return cpu_to_le32(csum);
151 static int ext4_xattr_block_csum_verify(struct inode *inode,
152 struct buffer_head *bh)
154 struct ext4_xattr_header *hdr = BHDR(bh);
157 if (ext4_has_metadata_csum(inode->i_sb)) {
159 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
160 bh->b_blocknr, hdr));
166 static void ext4_xattr_block_csum_set(struct inode *inode,
167 struct buffer_head *bh)
169 if (ext4_has_metadata_csum(inode->i_sb))
170 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
171 bh->b_blocknr, BHDR(bh));
174 static inline const struct xattr_handler *
175 ext4_xattr_handler(int name_index)
177 const struct xattr_handler *handler = NULL;
179 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
180 handler = ext4_xattr_handler_map[name_index];
185 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
188 struct ext4_xattr_entry *e = entry;
190 /* Find the end of the names list */
191 while (!IS_LAST_ENTRY(e)) {
192 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
193 if ((void *)next >= end)
194 return -EFSCORRUPTED;
195 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
196 return -EFSCORRUPTED;
200 /* Check the values */
201 while (!IS_LAST_ENTRY(entry)) {
202 u32 size = le32_to_cpu(entry->e_value_size);
204 if (size > EXT4_XATTR_SIZE_MAX)
205 return -EFSCORRUPTED;
207 if (size != 0 && entry->e_value_inum == 0) {
208 u16 offs = le16_to_cpu(entry->e_value_offs);
212 * The value cannot overlap the names, and the value
213 * with padding cannot extend beyond 'end'. Check both
214 * the padded and unpadded sizes, since the size may
215 * overflow to 0 when adding padding.
217 if (offs > end - value_start)
218 return -EFSCORRUPTED;
219 value = value_start + offs;
220 if (value < (void *)e + sizeof(u32) ||
221 size > end - value ||
222 EXT4_XATTR_SIZE(size) > end - value)
223 return -EFSCORRUPTED;
225 entry = EXT4_XATTR_NEXT(entry);
232 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
233 const char *function, unsigned int line)
235 int error = -EFSCORRUPTED;
237 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
238 BHDR(bh)->h_blocks != cpu_to_le32(1))
240 if (buffer_verified(bh))
244 if (!ext4_xattr_block_csum_verify(inode, bh))
246 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
250 __ext4_error_inode(inode, function, line, 0, -error,
251 "corrupted xattr block %llu",
252 (unsigned long long) bh->b_blocknr);
254 set_buffer_verified(bh);
258 #define ext4_xattr_check_block(inode, bh) \
259 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
263 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
264 void *end, const char *function, unsigned int line)
266 int error = -EFSCORRUPTED;
268 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
269 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
271 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
274 __ext4_error_inode(inode, function, line, 0, -error,
275 "corrupted in-inode xattr");
279 #define xattr_check_inode(inode, header, end) \
280 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
283 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
284 void *end, int name_index, const char *name, int sorted)
286 struct ext4_xattr_entry *entry, *next;
292 name_len = strlen(name);
293 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
294 next = EXT4_XATTR_NEXT(entry);
295 if ((void *) next >= end) {
296 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
297 return -EFSCORRUPTED;
299 cmp = name_index - entry->e_name_index;
301 cmp = name_len - entry->e_name_len;
303 cmp = memcmp(name, entry->e_name, name_len);
304 if (cmp <= 0 && (sorted || cmp == 0))
308 return cmp ? -ENODATA : 0;
312 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
314 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
317 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
319 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
320 (u32) inode_peek_iversion_raw(ea_inode);
323 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
325 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
326 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
329 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
331 return (u32)ea_inode->i_atime.tv_sec;
334 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
336 ea_inode->i_atime.tv_sec = hash;
340 * Read the EA value from an inode.
342 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
344 int blocksize = 1 << ea_inode->i_blkbits;
345 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
346 int tail_size = (size % blocksize) ?: blocksize;
347 struct buffer_head *bhs_inline[8];
348 struct buffer_head **bhs = bhs_inline;
351 if (bh_count > ARRAY_SIZE(bhs_inline)) {
352 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
357 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
358 true /* wait */, bhs);
362 for (i = 0; i < bh_count; i++) {
363 /* There shouldn't be any holes in ea_inode. */
368 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
369 i < bh_count - 1 ? blocksize : tail_size);
373 for (i = 0; i < bh_count; i++)
376 if (bhs != bhs_inline)
381 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
383 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
384 u32 ea_inode_hash, struct inode **ea_inode)
389 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
391 err = PTR_ERR(inode);
392 ext4_error(parent->i_sb,
393 "error while reading EA inode %lu err=%d", ea_ino,
398 if (is_bad_inode(inode)) {
399 ext4_error(parent->i_sb,
400 "error while reading EA inode %lu is_bad_inode",
406 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
407 ext4_error(parent->i_sb,
408 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
414 ext4_xattr_inode_set_class(inode);
417 * Check whether this is an old Lustre-style xattr inode. Lustre
418 * implementation does not have hash validation, rather it has a
419 * backpointer from ea_inode to the parent inode.
421 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
422 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
423 inode->i_generation == parent->i_generation) {
424 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
425 ext4_xattr_inode_set_ref(inode, 1);
428 inode->i_flags |= S_NOQUOTA;
440 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
441 struct ext4_xattr_entry *entry, void *buffer,
446 /* Verify stored hash matches calculated hash. */
447 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
448 if (hash != ext4_xattr_inode_get_hash(ea_inode))
449 return -EFSCORRUPTED;
452 __le32 e_hash, tmp_data;
454 /* Verify entry hash. */
455 tmp_data = cpu_to_le32(hash);
456 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
458 if (e_hash != entry->e_hash)
459 return -EFSCORRUPTED;
465 * Read xattr value from the EA inode.
468 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
469 void *buffer, size_t size)
471 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
472 struct inode *ea_inode;
475 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
476 le32_to_cpu(entry->e_hash), &ea_inode);
482 if (i_size_read(ea_inode) != size) {
483 ext4_warning_inode(ea_inode,
484 "ea_inode file size=%llu entry size=%zu",
485 i_size_read(ea_inode), size);
490 err = ext4_xattr_inode_read(ea_inode, buffer, size);
494 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
495 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
498 ext4_warning_inode(ea_inode,
499 "EA inode hash validation failed");
504 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
505 ext4_xattr_inode_get_hash(ea_inode),
506 ea_inode->i_ino, true /* reusable */);
514 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
515 void *buffer, size_t buffer_size)
517 struct buffer_head *bh = NULL;
518 struct ext4_xattr_entry *entry;
522 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
524 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
525 name_index, name, buffer, (long)buffer_size);
527 if (!EXT4_I(inode)->i_file_acl)
529 ea_idebug(inode, "reading block %llu",
530 (unsigned long long)EXT4_I(inode)->i_file_acl);
531 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
534 ea_bdebug(bh, "b_count=%d, refcount=%d",
535 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
536 error = ext4_xattr_check_block(inode, bh);
539 ext4_xattr_block_cache_insert(ea_block_cache, bh);
541 end = bh->b_data + bh->b_size;
542 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
545 size = le32_to_cpu(entry->e_value_size);
547 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
550 if (size > buffer_size)
552 if (entry->e_value_inum) {
553 error = ext4_xattr_inode_get(inode, entry, buffer,
558 u16 offset = le16_to_cpu(entry->e_value_offs);
559 void *p = bh->b_data + offset;
561 if (unlikely(p + size > end))
563 memcpy(buffer, p, size);
574 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
575 void *buffer, size_t buffer_size)
577 struct ext4_xattr_ibody_header *header;
578 struct ext4_xattr_entry *entry;
579 struct ext4_inode *raw_inode;
580 struct ext4_iloc iloc;
585 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
587 error = ext4_get_inode_loc(inode, &iloc);
590 raw_inode = ext4_raw_inode(&iloc);
591 header = IHDR(inode, raw_inode);
592 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
593 error = xattr_check_inode(inode, header, end);
596 entry = IFIRST(header);
597 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
600 size = le32_to_cpu(entry->e_value_size);
602 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
605 if (size > buffer_size)
607 if (entry->e_value_inum) {
608 error = ext4_xattr_inode_get(inode, entry, buffer,
613 u16 offset = le16_to_cpu(entry->e_value_offs);
614 void *p = (void *)IFIRST(header) + offset;
616 if (unlikely(p + size > end))
618 memcpy(buffer, p, size);
631 * Copy an extended attribute into the buffer
632 * provided, or compute the buffer size required.
633 * Buffer is NULL to compute the size of the buffer required.
635 * Returns a negative error number on failure, or the number of bytes
636 * used / required on success.
639 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
640 void *buffer, size_t buffer_size)
644 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
647 if (strlen(name) > 255)
650 down_read(&EXT4_I(inode)->xattr_sem);
651 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
653 if (error == -ENODATA)
654 error = ext4_xattr_block_get(inode, name_index, name, buffer,
656 up_read(&EXT4_I(inode)->xattr_sem);
661 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
662 char *buffer, size_t buffer_size)
664 size_t rest = buffer_size;
666 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
667 const struct xattr_handler *handler =
668 ext4_xattr_handler(entry->e_name_index);
670 if (handler && (!handler->list || handler->list(dentry))) {
671 const char *prefix = handler->prefix ?: handler->name;
672 size_t prefix_len = strlen(prefix);
673 size_t size = prefix_len + entry->e_name_len + 1;
678 memcpy(buffer, prefix, prefix_len);
679 buffer += prefix_len;
680 memcpy(buffer, entry->e_name, entry->e_name_len);
681 buffer += entry->e_name_len;
687 return buffer_size - rest; /* total size */
691 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
693 struct inode *inode = d_inode(dentry);
694 struct buffer_head *bh = NULL;
697 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
698 buffer, (long)buffer_size);
700 if (!EXT4_I(inode)->i_file_acl)
702 ea_idebug(inode, "reading block %llu",
703 (unsigned long long)EXT4_I(inode)->i_file_acl);
704 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
707 ea_bdebug(bh, "b_count=%d, refcount=%d",
708 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
709 error = ext4_xattr_check_block(inode, bh);
712 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
713 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
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 = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
838 ret = ext4_xattr_check_block(inode, bh);
842 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
843 entry = EXT4_XATTR_NEXT(entry))
844 if (entry->e_value_inum)
847 *usage = ea_inode_refs + 1;
855 static inline size_t round_up_cluster(struct inode *inode, size_t length)
857 struct super_block *sb = inode->i_sb;
858 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
860 size_t mask = ~(cluster_size - 1);
862 return (length + cluster_size - 1) & mask;
865 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
869 err = dquot_alloc_inode(inode);
872 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
874 dquot_free_inode(inode);
878 static void ext4_xattr_inode_free_quota(struct inode *parent,
879 struct inode *ea_inode,
883 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
885 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
886 dquot_free_inode(parent);
889 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
890 struct buffer_head *block_bh, size_t value_len,
897 * 1) Owner inode update
898 * 2) Ref count update on old xattr block
900 * 4) block bitmap update for new xattr block
901 * 5) group descriptor for new xattr block
902 * 6) block bitmap update for old xattr block
903 * 7) group descriptor for old block
905 * 6 & 7 can happen if we have two racing threads T_a and T_b
906 * which are each trying to set an xattr on inodes I_a and I_b
907 * which were both initially sharing an xattr block.
912 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
915 * In case of inline data, we may push out the data to a block,
916 * so we need to reserve credits for this eventuality
918 if (inode && ext4_has_inline_data(inode))
919 credits += ext4_writepage_trans_blocks(inode) + 1;
921 /* We are done if ea_inode feature is not enabled. */
922 if (!ext4_has_feature_ea_inode(sb))
925 /* New ea_inode, inode map, block bitmap, group descriptor. */
929 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
931 /* Indirection block or one level of extent tree. */
934 /* Block bitmap and group descriptor updates for each block. */
935 credits += blocks * 2;
937 /* Blocks themselves. */
941 /* Dereference ea_inode holding old xattr value.
942 * Old ea_inode, inode map, block bitmap, group descriptor.
946 /* Data blocks for old ea_inode. */
947 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
949 /* Indirection block or one level of extent tree for old
954 /* Block bitmap and group descriptor updates for each block. */
955 credits += blocks * 2;
958 /* We may need to clone the existing xattr block in which case we need
959 * to increment ref counts for existing ea_inodes referenced by it.
962 struct ext4_xattr_entry *entry = BFIRST(block_bh);
964 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
965 if (entry->e_value_inum)
966 /* Ref count update on ea_inode. */
972 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
975 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
976 struct ext4_iloc iloc;
981 inode_lock(ea_inode);
983 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
987 ref_count = ext4_xattr_inode_get_ref(ea_inode);
988 ref_count += ref_change;
989 ext4_xattr_inode_set_ref(ea_inode, ref_count);
991 if (ref_change > 0) {
992 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
993 ea_inode->i_ino, ref_count);
995 if (ref_count == 1) {
996 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
997 ea_inode->i_ino, ea_inode->i_nlink);
999 set_nlink(ea_inode, 1);
1000 ext4_orphan_del(handle, ea_inode);
1002 if (ea_inode_cache) {
1003 hash = ext4_xattr_inode_get_hash(ea_inode);
1004 mb_cache_entry_create(ea_inode_cache,
1007 true /* reusable */);
1011 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1012 ea_inode->i_ino, ref_count);
1014 if (ref_count == 0) {
1015 WARN_ONCE(ea_inode->i_nlink != 1,
1016 "EA inode %lu i_nlink=%u",
1017 ea_inode->i_ino, ea_inode->i_nlink);
1019 clear_nlink(ea_inode);
1020 ext4_orphan_add(handle, ea_inode);
1022 if (ea_inode_cache) {
1023 hash = ext4_xattr_inode_get_hash(ea_inode);
1024 mb_cache_entry_delete(ea_inode_cache, hash,
1030 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1032 ext4_warning_inode(ea_inode,
1033 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1035 inode_unlock(ea_inode);
1039 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1041 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1044 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1046 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1049 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1050 struct ext4_xattr_entry *first)
1052 struct inode *ea_inode;
1053 struct ext4_xattr_entry *entry;
1054 struct ext4_xattr_entry *failed_entry;
1055 unsigned int ea_ino;
1058 for (entry = first; !IS_LAST_ENTRY(entry);
1059 entry = EXT4_XATTR_NEXT(entry)) {
1060 if (!entry->e_value_inum)
1062 ea_ino = le32_to_cpu(entry->e_value_inum);
1063 err = ext4_xattr_inode_iget(parent, ea_ino,
1064 le32_to_cpu(entry->e_hash),
1068 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1070 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1080 failed_entry = entry;
1082 for (entry = first; entry != failed_entry;
1083 entry = EXT4_XATTR_NEXT(entry)) {
1084 if (!entry->e_value_inum)
1086 ea_ino = le32_to_cpu(entry->e_value_inum);
1087 err = ext4_xattr_inode_iget(parent, ea_ino,
1088 le32_to_cpu(entry->e_hash),
1091 ext4_warning(parent->i_sb,
1092 "cleanup ea_ino %u iget error %d", ea_ino,
1096 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1098 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1105 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1106 struct buffer_head *bh, bool block_csum, bool dirty)
1112 ext4_xattr_block_csum_set(inode, bh);
1113 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1115 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1124 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1125 struct buffer_head *bh,
1126 struct ext4_xattr_entry *first, bool block_csum,
1127 struct ext4_xattr_inode_array **ea_inode_array,
1128 int extra_credits, bool skip_quota)
1130 struct inode *ea_inode;
1131 struct ext4_xattr_entry *entry;
1133 unsigned int ea_ino;
1137 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1138 credits = 2 + extra_credits;
1140 for (entry = first; !IS_LAST_ENTRY(entry);
1141 entry = EXT4_XATTR_NEXT(entry)) {
1142 if (!entry->e_value_inum)
1144 ea_ino = le32_to_cpu(entry->e_value_inum);
1145 err = ext4_xattr_inode_iget(parent, ea_ino,
1146 le32_to_cpu(entry->e_hash),
1151 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1153 ext4_warning_inode(ea_inode,
1154 "Expand inode array err=%d", err);
1159 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1160 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1161 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1164 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1169 err = ext4_journal_get_write_access(handle, bh);
1171 ext4_warning_inode(ea_inode,
1172 "Re-get write access err=%d",
1178 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1180 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1186 ext4_xattr_inode_free_quota(parent, ea_inode,
1187 le32_to_cpu(entry->e_value_size));
1190 * Forget about ea_inode within the same transaction that
1191 * decrements the ref count. This avoids duplicate decrements in
1192 * case the rest of the work spills over to subsequent
1195 entry->e_value_inum = 0;
1196 entry->e_value_size = 0;
1203 * Note that we are deliberately skipping csum calculation for
1204 * the final update because we do not expect any journal
1205 * restarts until xattr block is freed.
1208 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1210 ext4_warning_inode(parent,
1211 "handle dirty metadata err=%d", err);
1216 * Release the xattr block BH: If the reference count is > 1, decrement it;
1217 * otherwise free the block.
1220 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1221 struct buffer_head *bh,
1222 struct ext4_xattr_inode_array **ea_inode_array,
1225 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1229 BUFFER_TRACE(bh, "get_write_access");
1230 error = ext4_journal_get_write_access(handle, bh);
1235 hash = le32_to_cpu(BHDR(bh)->h_hash);
1236 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1238 ea_bdebug(bh, "refcount now=0; freeing");
1240 * This must happen under buffer lock for
1241 * ext4_xattr_block_set() to reliably detect freed block
1244 mb_cache_entry_delete(ea_block_cache, hash,
1249 if (ext4_has_feature_ea_inode(inode->i_sb))
1250 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1252 true /* block_csum */,
1255 true /* skip_quota */);
1256 ext4_free_blocks(handle, inode, bh, 0, 1,
1257 EXT4_FREE_BLOCKS_METADATA |
1258 EXT4_FREE_BLOCKS_FORGET);
1261 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1262 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1263 struct mb_cache_entry *ce;
1265 if (ea_block_cache) {
1266 ce = mb_cache_entry_get(ea_block_cache, hash,
1270 mb_cache_entry_put(ea_block_cache, ce);
1275 ext4_xattr_block_csum_set(inode, bh);
1277 * Beware of this ugliness: Releasing of xattr block references
1278 * from different inodes can race and so we have to protect
1279 * from a race where someone else frees the block (and releases
1280 * its journal_head) before we are done dirtying the buffer. In
1281 * nojournal mode this race is harmless and we actually cannot
1282 * call ext4_handle_dirty_metadata() with locked buffer as
1283 * that function can call sync_dirty_buffer() so for that case
1284 * we handle the dirtying after unlocking the buffer.
1286 if (ext4_handle_valid(handle))
1287 error = ext4_handle_dirty_metadata(handle, inode, bh);
1289 if (!ext4_handle_valid(handle))
1290 error = ext4_handle_dirty_metadata(handle, inode, bh);
1292 ext4_handle_sync(handle);
1293 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1294 ea_bdebug(bh, "refcount now=%d; releasing",
1295 le32_to_cpu(BHDR(bh)->h_refcount));
1298 ext4_std_error(inode->i_sb, error);
1303 * Find the available free space for EAs. This also returns the total number of
1304 * bytes used by EA entries.
1306 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1307 size_t *min_offs, void *base, int *total)
1309 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1310 if (!last->e_value_inum && last->e_value_size) {
1311 size_t offs = le16_to_cpu(last->e_value_offs);
1312 if (offs < *min_offs)
1316 *total += EXT4_XATTR_LEN(last->e_name_len);
1318 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1322 * Write the value of the EA in an inode.
1324 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1325 const void *buf, int bufsize)
1327 struct buffer_head *bh = NULL;
1328 unsigned long block = 0;
1329 int blocksize = ea_inode->i_sb->s_blocksize;
1330 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1331 int csize, wsize = 0;
1332 int ret = 0, ret2 = 0;
1336 while (ret >= 0 && ret < max_blocks) {
1337 struct ext4_map_blocks map;
1338 map.m_lblk = block += ret;
1339 map.m_len = max_blocks -= ret;
1341 ret = ext4_map_blocks(handle, ea_inode, &map,
1342 EXT4_GET_BLOCKS_CREATE);
1344 ext4_mark_inode_dirty(handle, ea_inode);
1345 if (ret == -ENOSPC &&
1346 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1358 while (wsize < bufsize) {
1360 csize = (bufsize - wsize) > blocksize ? blocksize :
1362 bh = ext4_getblk(handle, ea_inode, block, 0);
1367 EXT4_ERROR_INODE(ea_inode,
1368 "ext4_getblk() return bh = NULL");
1369 return -EFSCORRUPTED;
1371 ret = ext4_journal_get_write_access(handle, bh);
1375 memcpy(bh->b_data, buf, csize);
1376 set_buffer_uptodate(bh);
1377 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1384 inode_lock(ea_inode);
1385 i_size_write(ea_inode, wsize);
1386 ext4_update_i_disksize(ea_inode, wsize);
1387 inode_unlock(ea_inode);
1389 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1390 if (unlikely(ret2 && !ret))
1400 * Create an inode to store the value of a large EA.
1402 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1403 struct inode *inode, u32 hash)
1405 struct inode *ea_inode = NULL;
1406 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1410 * Let the next inode be the goal, so we try and allocate the EA inode
1411 * in the same group, or nearby one.
1413 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1414 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1416 if (!IS_ERR(ea_inode)) {
1417 ea_inode->i_op = &ext4_file_inode_operations;
1418 ea_inode->i_fop = &ext4_file_operations;
1419 ext4_set_aops(ea_inode);
1420 ext4_xattr_inode_set_class(ea_inode);
1421 unlock_new_inode(ea_inode);
1422 ext4_xattr_inode_set_ref(ea_inode, 1);
1423 ext4_xattr_inode_set_hash(ea_inode, hash);
1424 err = ext4_mark_inode_dirty(handle, ea_inode);
1426 err = ext4_inode_attach_jinode(ea_inode);
1429 return ERR_PTR(err);
1433 * Xattr inodes are shared therefore quota charging is performed
1434 * at a higher level.
1436 dquot_free_inode(ea_inode);
1437 dquot_drop(ea_inode);
1438 inode_lock(ea_inode);
1439 ea_inode->i_flags |= S_NOQUOTA;
1440 inode_unlock(ea_inode);
1446 static struct inode *
1447 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1448 size_t value_len, u32 hash)
1450 struct inode *ea_inode;
1451 struct mb_cache_entry *ce;
1452 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1455 if (!ea_inode_cache)
1458 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1462 ea_data = kvmalloc(value_len, GFP_KERNEL);
1464 mb_cache_entry_put(ea_inode_cache, ce);
1469 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1471 if (!IS_ERR(ea_inode) &&
1472 !is_bad_inode(ea_inode) &&
1473 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1474 i_size_read(ea_inode) == value_len &&
1475 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1476 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1478 !memcmp(value, ea_data, value_len)) {
1479 mb_cache_entry_touch(ea_inode_cache, ce);
1480 mb_cache_entry_put(ea_inode_cache, ce);
1485 if (!IS_ERR(ea_inode))
1487 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1494 * Add value of the EA in an inode.
1496 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1497 const void *value, size_t value_len,
1498 struct inode **ret_inode)
1500 struct inode *ea_inode;
1504 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1505 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1507 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1513 *ret_inode = ea_inode;
1517 /* Create an inode for the EA value */
1518 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1519 if (IS_ERR(ea_inode))
1520 return PTR_ERR(ea_inode);
1522 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1524 ext4_xattr_inode_dec_ref(handle, ea_inode);
1529 if (EA_INODE_CACHE(inode))
1530 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1531 ea_inode->i_ino, true /* reusable */);
1533 *ret_inode = ea_inode;
1538 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1539 * feature is enabled.
1541 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1543 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1544 struct ext4_xattr_search *s,
1545 handle_t *handle, struct inode *inode,
1548 struct ext4_xattr_entry *last, *next;
1549 struct ext4_xattr_entry *here = s->here;
1550 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1551 int in_inode = i->in_inode;
1552 struct inode *old_ea_inode = NULL;
1553 struct inode *new_ea_inode = NULL;
1554 size_t old_size, new_size;
1557 /* Space used by old and new values. */
1558 old_size = (!s->not_found && !here->e_value_inum) ?
1559 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1560 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1563 * Optimization for the simple case when old and new values have the
1564 * same padded sizes. Not applicable if external inodes are involved.
1566 if (new_size && new_size == old_size) {
1567 size_t offs = le16_to_cpu(here->e_value_offs);
1568 void *val = s->base + offs;
1570 here->e_value_size = cpu_to_le32(i->value_len);
1571 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1572 memset(val, 0, new_size);
1574 memcpy(val, i->value, i->value_len);
1575 /* Clear padding bytes. */
1576 memset(val + i->value_len, 0, new_size - i->value_len);
1581 /* Compute min_offs and last. */
1583 for (; !IS_LAST_ENTRY(last); last = next) {
1584 next = EXT4_XATTR_NEXT(last);
1585 if ((void *)next >= s->end) {
1586 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1587 ret = -EFSCORRUPTED;
1590 if (!last->e_value_inum && last->e_value_size) {
1591 size_t offs = le16_to_cpu(last->e_value_offs);
1592 if (offs < min_offs)
1597 /* Check whether we have enough space. */
1601 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1603 free += EXT4_XATTR_LEN(name_len) + old_size;
1605 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1611 * If storing the value in an external inode is an option,
1612 * reserve space for xattr entries/names in the external
1613 * attribute block so that a long value does not occupy the
1614 * whole space and prevent futher entries being added.
1616 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1617 new_size && is_block &&
1618 (min_offs + old_size - new_size) <
1619 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1626 * Getting access to old and new ea inodes is subject to failures.
1627 * Finish that work before doing any modifications to the xattr data.
1629 if (!s->not_found && here->e_value_inum) {
1630 ret = ext4_xattr_inode_iget(inode,
1631 le32_to_cpu(here->e_value_inum),
1632 le32_to_cpu(here->e_hash),
1635 old_ea_inode = NULL;
1639 if (i->value && in_inode) {
1640 WARN_ON_ONCE(!i->value_len);
1642 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1646 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1650 new_ea_inode = NULL;
1651 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1657 /* We are ready to release ref count on the old_ea_inode. */
1658 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1660 /* Release newly required ref count on new_ea_inode. */
1664 err = ext4_xattr_inode_dec_ref(handle,
1667 ext4_warning_inode(new_ea_inode,
1668 "dec ref new_ea_inode err=%d",
1670 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1676 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1677 le32_to_cpu(here->e_value_size));
1680 /* No failures allowed past this point. */
1682 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1683 /* Remove the old value. */
1684 void *first_val = s->base + min_offs;
1685 size_t offs = le16_to_cpu(here->e_value_offs);
1686 void *val = s->base + offs;
1688 memmove(first_val + old_size, first_val, val - first_val);
1689 memset(first_val, 0, old_size);
1690 min_offs += old_size;
1692 /* Adjust all value offsets. */
1694 while (!IS_LAST_ENTRY(last)) {
1695 size_t o = le16_to_cpu(last->e_value_offs);
1697 if (!last->e_value_inum &&
1698 last->e_value_size && o < offs)
1699 last->e_value_offs = cpu_to_le16(o + old_size);
1700 last = EXT4_XATTR_NEXT(last);
1705 /* Remove old name. */
1706 size_t size = EXT4_XATTR_LEN(name_len);
1708 last = ENTRY((void *)last - size);
1709 memmove(here, (void *)here + size,
1710 (void *)last - (void *)here + sizeof(__u32));
1711 memset(last, 0, size);
1712 } else if (s->not_found) {
1713 /* Insert new name. */
1714 size_t size = EXT4_XATTR_LEN(name_len);
1715 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1717 memmove((void *)here + size, here, rest);
1718 memset(here, 0, size);
1719 here->e_name_index = i->name_index;
1720 here->e_name_len = name_len;
1721 memcpy(here->e_name, i->name, name_len);
1723 /* This is an update, reset value info. */
1724 here->e_value_inum = 0;
1725 here->e_value_offs = 0;
1726 here->e_value_size = 0;
1730 /* Insert new value. */
1732 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1733 } else if (i->value_len) {
1734 void *val = s->base + min_offs - new_size;
1736 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1737 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1738 memset(val, 0, new_size);
1740 memcpy(val, i->value, i->value_len);
1741 /* Clear padding bytes. */
1742 memset(val + i->value_len, 0,
1743 new_size - i->value_len);
1746 here->e_value_size = cpu_to_le32(i->value_len);
1753 /* Entry hash calculation. */
1758 * Feed crc32c hash instead of the raw value for entry
1759 * hash calculation. This is to avoid walking
1760 * potentially long value buffer again.
1762 crc32c_hash = cpu_to_le32(
1763 ext4_xattr_inode_get_hash(new_ea_inode));
1764 hash = ext4_xattr_hash_entry(here->e_name,
1767 } else if (is_block) {
1768 __le32 *value = s->base + le16_to_cpu(
1769 here->e_value_offs);
1771 hash = ext4_xattr_hash_entry(here->e_name,
1772 here->e_name_len, value,
1775 here->e_hash = hash;
1779 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1788 struct ext4_xattr_block_find {
1789 struct ext4_xattr_search s;
1790 struct buffer_head *bh;
1794 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1795 struct ext4_xattr_block_find *bs)
1797 struct super_block *sb = inode->i_sb;
1800 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1801 i->name_index, i->name, i->value, (long)i->value_len);
1803 if (EXT4_I(inode)->i_file_acl) {
1804 /* The inode already has an extended attribute block. */
1805 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1806 if (IS_ERR(bs->bh)) {
1807 error = PTR_ERR(bs->bh);
1811 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1812 atomic_read(&(bs->bh->b_count)),
1813 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1814 error = ext4_xattr_check_block(inode, bs->bh);
1817 /* Find the named attribute. */
1818 bs->s.base = BHDR(bs->bh);
1819 bs->s.first = BFIRST(bs->bh);
1820 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1821 bs->s.here = bs->s.first;
1822 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1823 i->name_index, i->name, 1);
1824 if (error && error != -ENODATA)
1826 bs->s.not_found = error;
1832 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1833 struct ext4_xattr_info *i,
1834 struct ext4_xattr_block_find *bs)
1836 struct super_block *sb = inode->i_sb;
1837 struct buffer_head *new_bh = NULL;
1838 struct ext4_xattr_search s_copy = bs->s;
1839 struct ext4_xattr_search *s = &s_copy;
1840 struct mb_cache_entry *ce = NULL;
1842 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1843 struct inode *ea_inode = NULL, *tmp_inode;
1844 size_t old_ea_inode_quota = 0;
1845 unsigned int ea_ino;
1848 #define header(x) ((struct ext4_xattr_header *)(x))
1851 BUFFER_TRACE(bs->bh, "get_write_access");
1852 error = ext4_journal_get_write_access(handle, bs->bh);
1855 lock_buffer(bs->bh);
1857 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1858 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1861 * This must happen under buffer lock for
1862 * ext4_xattr_block_set() to reliably detect modified
1866 mb_cache_entry_delete(ea_block_cache, hash,
1868 ea_bdebug(bs->bh, "modifying in-place");
1869 error = ext4_xattr_set_entry(i, s, handle, inode,
1870 true /* is_block */);
1871 ext4_xattr_block_csum_set(inode, bs->bh);
1872 unlock_buffer(bs->bh);
1873 if (error == -EFSCORRUPTED)
1876 error = ext4_handle_dirty_metadata(handle,
1883 int offset = (char *)s->here - bs->bh->b_data;
1885 unlock_buffer(bs->bh);
1886 ea_bdebug(bs->bh, "cloning");
1887 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1889 if (s->base == NULL)
1891 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1892 s->first = ENTRY(header(s->base)+1);
1893 header(s->base)->h_refcount = cpu_to_le32(1);
1894 s->here = ENTRY(s->base + offset);
1895 s->end = s->base + bs->bh->b_size;
1898 * If existing entry points to an xattr inode, we need
1899 * to prevent ext4_xattr_set_entry() from decrementing
1900 * ref count on it because the reference belongs to the
1901 * original block. In this case, make the entry look
1902 * like it has an empty value.
1904 if (!s->not_found && s->here->e_value_inum) {
1905 ea_ino = le32_to_cpu(s->here->e_value_inum);
1906 error = ext4_xattr_inode_iget(inode, ea_ino,
1907 le32_to_cpu(s->here->e_hash),
1912 if (!ext4_test_inode_state(tmp_inode,
1913 EXT4_STATE_LUSTRE_EA_INODE)) {
1915 * Defer quota free call for previous
1916 * inode until success is guaranteed.
1918 old_ea_inode_quota = le32_to_cpu(
1919 s->here->e_value_size);
1923 s->here->e_value_inum = 0;
1924 s->here->e_value_size = 0;
1928 /* Allocate a buffer where we construct the new block. */
1929 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1930 /* assert(header == s->base) */
1932 if (s->base == NULL)
1934 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1935 header(s->base)->h_blocks = cpu_to_le32(1);
1936 header(s->base)->h_refcount = cpu_to_le32(1);
1937 s->first = ENTRY(header(s->base)+1);
1938 s->here = ENTRY(header(s->base)+1);
1939 s->end = s->base + sb->s_blocksize;
1942 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1943 if (error == -EFSCORRUPTED)
1948 if (i->value && s->here->e_value_inum) {
1950 * A ref count on ea_inode has been taken as part of the call to
1951 * ext4_xattr_set_entry() above. We would like to drop this
1952 * extra ref but we have to wait until the xattr block is
1953 * initialized and has its own ref count on the ea_inode.
1955 ea_ino = le32_to_cpu(s->here->e_value_inum);
1956 error = ext4_xattr_inode_iget(inode, ea_ino,
1957 le32_to_cpu(s->here->e_hash),
1966 if (!IS_LAST_ENTRY(s->first)) {
1967 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1970 /* We found an identical block in the cache. */
1971 if (new_bh == bs->bh)
1972 ea_bdebug(new_bh, "keeping");
1976 WARN_ON_ONCE(dquot_initialize_needed(inode));
1978 /* The old block is released after updating
1980 error = dquot_alloc_block(inode,
1981 EXT4_C2B(EXT4_SB(sb), 1));
1984 BUFFER_TRACE(new_bh, "get_write_access");
1985 error = ext4_journal_get_write_access(handle,
1989 lock_buffer(new_bh);
1991 * We have to be careful about races with
1992 * freeing, rehashing or adding references to
1993 * xattr block. Once we hold buffer lock xattr
1994 * block's state is stable so we can check
1995 * whether the block got freed / rehashed or
1996 * not. Since we unhash mbcache entry under
1997 * buffer lock when freeing / rehashing xattr
1998 * block, checking whether entry is still
1999 * hashed is reliable. Same rules hold for
2000 * e_reusable handling.
2002 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2005 * Undo everything and check mbcache
2008 unlock_buffer(new_bh);
2009 dquot_free_block(inode,
2010 EXT4_C2B(EXT4_SB(sb),
2013 mb_cache_entry_put(ea_block_cache, ce);
2018 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2019 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2020 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2022 ea_bdebug(new_bh, "reusing; refcount now=%d",
2024 ext4_xattr_block_csum_set(inode, new_bh);
2025 unlock_buffer(new_bh);
2026 error = ext4_handle_dirty_metadata(handle,
2032 mb_cache_entry_touch(ea_block_cache, ce);
2033 mb_cache_entry_put(ea_block_cache, ce);
2035 } else if (bs->bh && s->base == bs->bh->b_data) {
2036 /* We were modifying this block in-place. */
2037 ea_bdebug(bs->bh, "keeping this block");
2038 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2042 /* We need to allocate a new block */
2043 ext4_fsblk_t goal, block;
2045 WARN_ON_ONCE(dquot_initialize_needed(inode));
2047 goal = ext4_group_first_block_no(sb,
2048 EXT4_I(inode)->i_block_group);
2050 /* non-extent files can't have physical blocks past 2^32 */
2051 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2052 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2054 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2059 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2060 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2062 ea_idebug(inode, "creating block %llu",
2063 (unsigned long long)block);
2065 new_bh = sb_getblk(sb, block);
2066 if (unlikely(!new_bh)) {
2069 ext4_free_blocks(handle, inode, NULL, block, 1,
2070 EXT4_FREE_BLOCKS_METADATA);
2073 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2074 ENTRY(header(s->base)+1));
2078 /* Drop the extra ref on ea_inode. */
2079 error = ext4_xattr_inode_dec_ref(handle,
2082 ext4_warning_inode(ea_inode,
2089 lock_buffer(new_bh);
2090 error = ext4_journal_get_create_access(handle, new_bh);
2092 unlock_buffer(new_bh);
2096 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2097 ext4_xattr_block_csum_set(inode, new_bh);
2098 set_buffer_uptodate(new_bh);
2099 unlock_buffer(new_bh);
2100 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2101 error = ext4_handle_dirty_metadata(handle, inode,
2108 if (old_ea_inode_quota)
2109 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2111 /* Update the inode. */
2112 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2114 /* Drop the previous xattr block. */
2115 if (bs->bh && bs->bh != new_bh) {
2116 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2118 ext4_xattr_release_block(handle, inode, bs->bh,
2120 0 /* extra_credits */);
2121 ext4_xattr_inode_array_free(ea_inode_array);
2129 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2131 ext4_warning_inode(ea_inode, "dec ref error=%d",
2134 /* If there was an error, revert the quota charge. */
2136 ext4_xattr_inode_free_quota(inode, ea_inode,
2137 i_size_read(ea_inode));
2141 mb_cache_entry_put(ea_block_cache, ce);
2143 if (!(bs->bh && s->base == bs->bh->b_data))
2149 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2153 EXT4_ERROR_INODE(inode, "bad block %llu",
2154 EXT4_I(inode)->i_file_acl);
2160 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2161 struct ext4_xattr_ibody_find *is)
2163 struct ext4_xattr_ibody_header *header;
2164 struct ext4_inode *raw_inode;
2167 if (EXT4_I(inode)->i_extra_isize == 0)
2169 raw_inode = ext4_raw_inode(&is->iloc);
2170 header = IHDR(inode, raw_inode);
2171 is->s.base = is->s.first = IFIRST(header);
2172 is->s.here = is->s.first;
2173 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2174 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2175 error = xattr_check_inode(inode, header, is->s.end);
2178 /* Find the named attribute. */
2179 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2180 i->name_index, i->name, 0);
2181 if (error && error != -ENODATA)
2183 is->s.not_found = error;
2188 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2189 struct ext4_xattr_info *i,
2190 struct ext4_xattr_ibody_find *is)
2192 struct ext4_xattr_ibody_header *header;
2193 struct ext4_xattr_search *s = &is->s;
2196 if (EXT4_I(inode)->i_extra_isize == 0)
2198 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2201 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2202 if (!IS_LAST_ENTRY(s->first)) {
2203 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2204 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2206 header->h_magic = cpu_to_le32(0);
2207 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2212 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2213 struct ext4_xattr_info *i,
2214 struct ext4_xattr_ibody_find *is)
2216 struct ext4_xattr_ibody_header *header;
2217 struct ext4_xattr_search *s = &is->s;
2220 if (EXT4_I(inode)->i_extra_isize == 0)
2222 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2225 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2226 if (!IS_LAST_ENTRY(s->first)) {
2227 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2228 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2230 header->h_magic = cpu_to_le32(0);
2231 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2236 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2237 struct ext4_xattr_info *i)
2241 /* When e_value_inum is set the value is stored externally. */
2242 if (s->here->e_value_inum)
2244 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2246 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2247 return !memcmp(value, i->value, i->value_len);
2250 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2252 struct buffer_head *bh;
2255 if (!EXT4_I(inode)->i_file_acl)
2257 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2260 error = ext4_xattr_check_block(inode, bh);
2263 return ERR_PTR(error);
2269 * ext4_xattr_set_handle()
2271 * Create, replace or remove an extended attribute for this inode. Value
2272 * is NULL to remove an existing extended attribute, and non-NULL to
2273 * either replace an existing extended attribute, or create a new extended
2274 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2275 * specify that an extended attribute must exist and must not exist
2276 * previous to the call, respectively.
2278 * Returns 0, or a negative error number on failure.
2281 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2282 const char *name, const void *value, size_t value_len,
2285 struct ext4_xattr_info i = {
2286 .name_index = name_index,
2289 .value_len = value_len,
2292 struct ext4_xattr_ibody_find is = {
2293 .s = { .not_found = -ENODATA, },
2295 struct ext4_xattr_block_find bs = {
2296 .s = { .not_found = -ENODATA, },
2303 if (strlen(name) > 255)
2306 ext4_write_lock_xattr(inode, &no_expand);
2308 /* Check journal credits under write lock. */
2309 if (ext4_handle_valid(handle)) {
2310 struct buffer_head *bh;
2313 bh = ext4_xattr_get_block(inode);
2315 error = PTR_ERR(bh);
2319 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2321 flags & XATTR_CREATE);
2324 if (jbd2_handle_buffer_credits(handle) < credits) {
2330 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2334 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2335 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2336 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2337 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2340 error = ext4_xattr_ibody_find(inode, &i, &is);
2344 error = ext4_xattr_block_find(inode, &i, &bs);
2347 if (is.s.not_found && bs.s.not_found) {
2349 if (flags & XATTR_REPLACE)
2356 if (flags & XATTR_CREATE)
2361 if (!is.s.not_found)
2362 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2363 else if (!bs.s.not_found)
2364 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2367 /* Xattr value did not change? Save us some work and bail out */
2368 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2370 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2373 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2374 (EXT4_XATTR_SIZE(i.value_len) >
2375 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2378 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2379 if (!error && !bs.s.not_found) {
2381 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2382 } else if (error == -ENOSPC) {
2383 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2386 error = ext4_xattr_block_find(inode, &i, &bs);
2390 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2391 if (!error && !is.s.not_found) {
2393 error = ext4_xattr_ibody_set(handle, inode, &i,
2395 } else if (error == -ENOSPC) {
2397 * Xattr does not fit in the block, store at
2398 * external inode if possible.
2400 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2409 ext4_xattr_update_super_block(handle, inode->i_sb);
2410 inode->i_ctime = current_time(inode);
2413 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2415 * The bh is consumed by ext4_mark_iloc_dirty, even with
2420 ext4_handle_sync(handle);
2422 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR);
2427 ext4_write_unlock_xattr(inode, &no_expand);
2431 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2432 bool is_create, int *credits)
2434 struct buffer_head *bh;
2439 if (!EXT4_SB(inode->i_sb)->s_journal)
2442 down_read(&EXT4_I(inode)->xattr_sem);
2444 bh = ext4_xattr_get_block(inode);
2448 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2449 value_len, is_create);
2454 up_read(&EXT4_I(inode)->xattr_sem);
2461 * Like ext4_xattr_set_handle, but start from an inode. This extended
2462 * attribute modification is a filesystem transaction by itself.
2464 * Returns 0, or a negative error number on failure.
2467 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2468 const void *value, size_t value_len, int flags)
2471 struct super_block *sb = inode->i_sb;
2472 int error, retries = 0;
2475 error = dquot_initialize(inode);
2480 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2485 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2486 if (IS_ERR(handle)) {
2487 error = PTR_ERR(handle);
2491 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2492 value, value_len, flags);
2493 error2 = ext4_journal_stop(handle);
2494 if (error == -ENOSPC &&
2495 ext4_should_retry_alloc(sb, &retries))
2500 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR);
2506 * Shift the EA entries in the inode to create space for the increased
2509 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2510 int value_offs_shift, void *to,
2511 void *from, size_t n)
2513 struct ext4_xattr_entry *last = entry;
2516 /* We always shift xattr headers further thus offsets get lower */
2517 BUG_ON(value_offs_shift > 0);
2519 /* Adjust the value offsets of the entries */
2520 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2521 if (!last->e_value_inum && last->e_value_size) {
2522 new_offs = le16_to_cpu(last->e_value_offs) +
2524 last->e_value_offs = cpu_to_le16(new_offs);
2527 /* Shift the entries by n bytes */
2528 memmove(to, from, n);
2532 * Move xattr pointed to by 'entry' from inode into external xattr block
2534 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2535 struct ext4_inode *raw_inode,
2536 struct ext4_xattr_entry *entry)
2538 struct ext4_xattr_ibody_find *is = NULL;
2539 struct ext4_xattr_block_find *bs = NULL;
2540 char *buffer = NULL, *b_entry_name = NULL;
2541 size_t value_size = le32_to_cpu(entry->e_value_size);
2542 struct ext4_xattr_info i = {
2545 .name_index = entry->e_name_index,
2546 .in_inode = !!entry->e_value_inum,
2548 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2551 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2552 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2553 buffer = kmalloc(value_size, GFP_NOFS);
2554 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2555 if (!is || !bs || !buffer || !b_entry_name) {
2560 is->s.not_found = -ENODATA;
2561 bs->s.not_found = -ENODATA;
2565 /* Save the entry name and the entry value */
2566 if (entry->e_value_inum) {
2567 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2571 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2572 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2575 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2576 b_entry_name[entry->e_name_len] = '\0';
2577 i.name = b_entry_name;
2579 error = ext4_get_inode_loc(inode, &is->iloc);
2583 error = ext4_xattr_ibody_find(inode, &i, is);
2587 /* Remove the chosen entry from the inode */
2588 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2593 i.value_len = value_size;
2594 error = ext4_xattr_block_find(inode, &i, bs);
2598 /* Add entry which was removed from the inode into the block */
2599 error = ext4_xattr_block_set(handle, inode, &i, bs);
2604 kfree(b_entry_name);
2607 brelse(is->iloc.bh);
2616 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2617 struct ext4_inode *raw_inode,
2618 int isize_diff, size_t ifree,
2619 size_t bfree, int *total_ino)
2621 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2622 struct ext4_xattr_entry *small_entry;
2623 struct ext4_xattr_entry *entry;
2624 struct ext4_xattr_entry *last;
2625 unsigned int entry_size; /* EA entry size */
2626 unsigned int total_size; /* EA entry size + value size */
2627 unsigned int min_total_size;
2630 while (isize_diff > ifree) {
2633 min_total_size = ~0U;
2634 last = IFIRST(header);
2635 /* Find the entry best suited to be pushed into EA block */
2636 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2637 /* never move system.data out of the inode */
2638 if ((last->e_name_len == 4) &&
2639 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2640 !memcmp(last->e_name, "data", 4))
2642 total_size = EXT4_XATTR_LEN(last->e_name_len);
2643 if (!last->e_value_inum)
2644 total_size += EXT4_XATTR_SIZE(
2645 le32_to_cpu(last->e_value_size));
2646 if (total_size <= bfree &&
2647 total_size < min_total_size) {
2648 if (total_size + ifree < isize_diff) {
2652 min_total_size = total_size;
2657 if (entry == NULL) {
2658 if (small_entry == NULL)
2660 entry = small_entry;
2663 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2664 total_size = entry_size;
2665 if (!entry->e_value_inum)
2666 total_size += EXT4_XATTR_SIZE(
2667 le32_to_cpu(entry->e_value_size));
2668 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2673 *total_ino -= entry_size;
2674 ifree += total_size;
2675 bfree -= total_size;
2682 * Expand an inode by new_extra_isize bytes when EAs are present.
2683 * Returns 0 on success or negative error number on failure.
2685 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2686 struct ext4_inode *raw_inode, handle_t *handle)
2688 struct ext4_xattr_ibody_header *header;
2689 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2690 static unsigned int mnt_count;
2692 size_t ifree, bfree;
2695 int error = 0, tried_min_extra_isize = 0;
2696 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2697 int isize_diff; /* How much do we need to grow i_extra_isize */
2700 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2701 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2704 header = IHDR(inode, raw_inode);
2707 * Check if enough free space is available in the inode to shift the
2708 * entries ahead by new_extra_isize.
2711 base = IFIRST(header);
2712 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2713 min_offs = end - base;
2714 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2716 error = xattr_check_inode(inode, header, end);
2720 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2721 if (ifree >= isize_diff)
2725 * Enough free space isn't available in the inode, check if
2726 * EA block can hold new_extra_isize bytes.
2728 if (EXT4_I(inode)->i_file_acl) {
2729 struct buffer_head *bh;
2731 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2733 error = PTR_ERR(bh);
2736 error = ext4_xattr_check_block(inode, bh);
2742 end = bh->b_data + bh->b_size;
2743 min_offs = end - base;
2744 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2747 if (bfree + ifree < isize_diff) {
2748 if (!tried_min_extra_isize && s_min_extra_isize) {
2749 tried_min_extra_isize++;
2750 new_extra_isize = s_min_extra_isize;
2757 bfree = inode->i_sb->s_blocksize;
2760 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2761 isize_diff, ifree, bfree,
2764 if (error == -ENOSPC && !tried_min_extra_isize &&
2765 s_min_extra_isize) {
2766 tried_min_extra_isize++;
2767 new_extra_isize = s_min_extra_isize;
2773 /* Adjust the offsets and shift the remaining entries ahead */
2774 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2775 - new_extra_isize, (void *)raw_inode +
2776 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2777 (void *)header, total_ino);
2778 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2781 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2782 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2784 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2789 #define EIA_INCR 16 /* must be 2^n */
2790 #define EIA_MASK (EIA_INCR - 1)
2792 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2793 * If @ea_inode_array is new or full it will be grown and the old
2794 * contents copied over.
2797 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2798 struct inode *inode)
2800 if (*ea_inode_array == NULL) {
2802 * Start with 15 inodes, so it fits into a power-of-two size.
2803 * If *ea_inode_array is NULL, this is essentially offsetof()
2806 kmalloc(offsetof(struct ext4_xattr_inode_array,
2809 if (*ea_inode_array == NULL)
2811 (*ea_inode_array)->count = 0;
2812 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2813 /* expand the array once all 15 + n * 16 slots are full */
2814 struct ext4_xattr_inode_array *new_array = NULL;
2815 int count = (*ea_inode_array)->count;
2817 /* if new_array is NULL, this is essentially offsetof() */
2818 new_array = kmalloc(
2819 offsetof(struct ext4_xattr_inode_array,
2820 inodes[count + EIA_INCR]),
2822 if (new_array == NULL)
2824 memcpy(new_array, *ea_inode_array,
2825 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2826 kfree(*ea_inode_array);
2827 *ea_inode_array = new_array;
2829 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2834 * ext4_xattr_delete_inode()
2836 * Free extended attribute resources associated with this inode. Traverse
2837 * all entries and decrement reference on any xattr inodes associated with this
2838 * inode. This is called immediately before an inode is freed. We have exclusive
2839 * access to the inode. If an orphan inode is deleted it will also release its
2840 * references on xattr block and xattr inodes.
2842 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2843 struct ext4_xattr_inode_array **ea_inode_array,
2846 struct buffer_head *bh = NULL;
2847 struct ext4_xattr_ibody_header *header;
2848 struct ext4_iloc iloc = { .bh = NULL };
2849 struct ext4_xattr_entry *entry;
2850 struct inode *ea_inode;
2853 error = ext4_journal_ensure_credits(handle, extra_credits,
2854 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2856 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2860 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2861 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2863 error = ext4_get_inode_loc(inode, &iloc);
2865 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2869 error = ext4_journal_get_write_access(handle, iloc.bh);
2871 EXT4_ERROR_INODE(inode, "write access (error %d)",
2876 header = IHDR(inode, ext4_raw_inode(&iloc));
2877 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2878 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2880 false /* block_csum */,
2883 false /* skip_quota */);
2886 if (EXT4_I(inode)->i_file_acl) {
2887 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2889 error = PTR_ERR(bh);
2890 if (error == -EIO) {
2891 EXT4_ERROR_INODE_ERR(inode, EIO,
2892 "block %llu read error",
2893 EXT4_I(inode)->i_file_acl);
2898 error = ext4_xattr_check_block(inode, bh);
2902 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2903 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2904 entry = EXT4_XATTR_NEXT(entry)) {
2905 if (!entry->e_value_inum)
2907 error = ext4_xattr_inode_iget(inode,
2908 le32_to_cpu(entry->e_value_inum),
2909 le32_to_cpu(entry->e_hash),
2913 ext4_xattr_inode_free_quota(inode, ea_inode,
2914 le32_to_cpu(entry->e_value_size));
2920 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2923 * Update i_file_acl value in the same transaction that releases
2926 EXT4_I(inode)->i_file_acl = 0;
2927 error = ext4_mark_inode_dirty(handle, inode);
2929 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2933 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR);
2942 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2946 if (ea_inode_array == NULL)
2949 for (idx = 0; idx < ea_inode_array->count; ++idx)
2950 iput(ea_inode_array->inodes[idx]);
2951 kfree(ea_inode_array);
2955 * ext4_xattr_block_cache_insert()
2957 * Create a new entry in the extended attribute block cache, and insert
2958 * it unless such an entry is already in the cache.
2960 * Returns 0, or a negative error number on failure.
2963 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2964 struct buffer_head *bh)
2966 struct ext4_xattr_header *header = BHDR(bh);
2967 __u32 hash = le32_to_cpu(header->h_hash);
2968 int reusable = le32_to_cpu(header->h_refcount) <
2969 EXT4_XATTR_REFCOUNT_MAX;
2972 if (!ea_block_cache)
2974 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2975 bh->b_blocknr, reusable);
2977 if (error == -EBUSY)
2978 ea_bdebug(bh, "already in cache");
2980 ea_bdebug(bh, "inserting [%x]", (int)hash);
2986 * Compare two extended attribute blocks for equality.
2988 * Returns 0 if the blocks are equal, 1 if they differ, and
2989 * a negative error number on errors.
2992 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2993 struct ext4_xattr_header *header2)
2995 struct ext4_xattr_entry *entry1, *entry2;
2997 entry1 = ENTRY(header1+1);
2998 entry2 = ENTRY(header2+1);
2999 while (!IS_LAST_ENTRY(entry1)) {
3000 if (IS_LAST_ENTRY(entry2))
3002 if (entry1->e_hash != entry2->e_hash ||
3003 entry1->e_name_index != entry2->e_name_index ||
3004 entry1->e_name_len != entry2->e_name_len ||
3005 entry1->e_value_size != entry2->e_value_size ||
3006 entry1->e_value_inum != entry2->e_value_inum ||
3007 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3009 if (!entry1->e_value_inum &&
3010 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3011 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3012 le32_to_cpu(entry1->e_value_size)))
3015 entry1 = EXT4_XATTR_NEXT(entry1);
3016 entry2 = EXT4_XATTR_NEXT(entry2);
3018 if (!IS_LAST_ENTRY(entry2))
3024 * ext4_xattr_block_cache_find()
3026 * Find an identical extended attribute block.
3028 * Returns a pointer to the block found, or NULL if such a block was
3029 * not found or an error occurred.
3031 static struct buffer_head *
3032 ext4_xattr_block_cache_find(struct inode *inode,
3033 struct ext4_xattr_header *header,
3034 struct mb_cache_entry **pce)
3036 __u32 hash = le32_to_cpu(header->h_hash);
3037 struct mb_cache_entry *ce;
3038 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3040 if (!ea_block_cache)
3042 if (!header->h_hash)
3043 return NULL; /* never share */
3044 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3045 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3047 struct buffer_head *bh;
3049 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3051 if (PTR_ERR(bh) == -ENOMEM)
3054 EXT4_ERROR_INODE(inode, "block %lu read error",
3055 (unsigned long)ce->e_value);
3056 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3061 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3066 #define NAME_HASH_SHIFT 5
3067 #define VALUE_HASH_SHIFT 16
3070 * ext4_xattr_hash_entry()
3072 * Compute the hash of an extended attribute.
3074 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3079 while (name_len--) {
3080 hash = (hash << NAME_HASH_SHIFT) ^
3081 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3084 while (value_count--) {
3085 hash = (hash << VALUE_HASH_SHIFT) ^
3086 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3087 le32_to_cpu(*value++);
3089 return cpu_to_le32(hash);
3092 #undef NAME_HASH_SHIFT
3093 #undef VALUE_HASH_SHIFT
3095 #define BLOCK_HASH_SHIFT 16
3098 * ext4_xattr_rehash()
3100 * Re-compute the extended attribute hash value after an entry has changed.
3102 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3104 struct ext4_xattr_entry *here;
3107 here = ENTRY(header+1);
3108 while (!IS_LAST_ENTRY(here)) {
3109 if (!here->e_hash) {
3110 /* Block is not shared if an entry's hash value == 0 */
3114 hash = (hash << BLOCK_HASH_SHIFT) ^
3115 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3116 le32_to_cpu(here->e_hash);
3117 here = EXT4_XATTR_NEXT(here);
3119 header->h_hash = cpu_to_le32(hash);
3122 #undef BLOCK_HASH_SHIFT
3124 #define HASH_BUCKET_BITS 10
3127 ext4_xattr_create_cache(void)
3129 return mb_cache_create(HASH_BUCKET_BITS);
3132 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3135 mb_cache_destroy(cache);