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 __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value,
86 static void ext4_xattr_rehash(struct ext4_xattr_header *);
88 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
89 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
90 #ifdef CONFIG_EXT4_FS_POSIX_ACL
91 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
92 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
94 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
95 #ifdef CONFIG_EXT4_FS_SECURITY
96 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
98 [EXT4_XATTR_INDEX_HURD] = &ext4_xattr_hurd_handler,
101 const struct xattr_handler *ext4_xattr_handlers[] = {
102 &ext4_xattr_user_handler,
103 &ext4_xattr_trusted_handler,
104 #ifdef CONFIG_EXT4_FS_POSIX_ACL
105 &posix_acl_access_xattr_handler,
106 &posix_acl_default_xattr_handler,
108 #ifdef CONFIG_EXT4_FS_SECURITY
109 &ext4_xattr_security_handler,
111 &ext4_xattr_hurd_handler,
115 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
116 inode->i_sb->s_fs_info)->s_ea_block_cache)
118 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
119 inode->i_sb->s_fs_info)->s_ea_inode_cache)
122 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
123 struct inode *inode);
125 #ifdef CONFIG_LOCKDEP
126 void ext4_xattr_inode_set_class(struct inode *ea_inode)
128 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
132 static __le32 ext4_xattr_block_csum(struct inode *inode,
134 struct ext4_xattr_header *hdr)
136 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
138 __le64 dsk_block_nr = cpu_to_le64(block_nr);
139 __u32 dummy_csum = 0;
140 int offset = offsetof(struct ext4_xattr_header, h_checksum);
142 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
143 sizeof(dsk_block_nr));
144 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
145 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
146 offset += sizeof(dummy_csum);
147 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
148 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
150 return cpu_to_le32(csum);
153 static int ext4_xattr_block_csum_verify(struct inode *inode,
154 struct buffer_head *bh)
156 struct ext4_xattr_header *hdr = BHDR(bh);
159 if (ext4_has_metadata_csum(inode->i_sb)) {
161 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
162 bh->b_blocknr, hdr));
168 static void ext4_xattr_block_csum_set(struct inode *inode,
169 struct buffer_head *bh)
171 if (ext4_has_metadata_csum(inode->i_sb))
172 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
173 bh->b_blocknr, BHDR(bh));
176 static inline const struct xattr_handler *
177 ext4_xattr_handler(int name_index)
179 const struct xattr_handler *handler = NULL;
181 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
182 handler = ext4_xattr_handler_map[name_index];
187 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
190 struct ext4_xattr_entry *e = entry;
192 /* Find the end of the names list */
193 while (!IS_LAST_ENTRY(e)) {
194 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
195 if ((void *)next >= end)
196 return -EFSCORRUPTED;
197 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
198 return -EFSCORRUPTED;
202 /* Check the values */
203 while (!IS_LAST_ENTRY(entry)) {
204 u32 size = le32_to_cpu(entry->e_value_size);
206 if (size > EXT4_XATTR_SIZE_MAX)
207 return -EFSCORRUPTED;
209 if (size != 0 && entry->e_value_inum == 0) {
210 u16 offs = le16_to_cpu(entry->e_value_offs);
214 * The value cannot overlap the names, and the value
215 * with padding cannot extend beyond 'end'. Check both
216 * the padded and unpadded sizes, since the size may
217 * overflow to 0 when adding padding.
219 if (offs > end - value_start)
220 return -EFSCORRUPTED;
221 value = value_start + offs;
222 if (value < (void *)e + sizeof(u32) ||
223 size > end - value ||
224 EXT4_XATTR_SIZE(size) > end - value)
225 return -EFSCORRUPTED;
227 entry = EXT4_XATTR_NEXT(entry);
234 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
235 const char *function, unsigned int line)
237 int error = -EFSCORRUPTED;
239 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
240 BHDR(bh)->h_blocks != cpu_to_le32(1))
242 if (buffer_verified(bh))
246 if (!ext4_xattr_block_csum_verify(inode, bh))
248 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
252 __ext4_error_inode(inode, function, line, 0, -error,
253 "corrupted xattr block %llu",
254 (unsigned long long) bh->b_blocknr);
256 set_buffer_verified(bh);
260 #define ext4_xattr_check_block(inode, bh) \
261 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
265 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
266 void *end, const char *function, unsigned int line)
268 int error = -EFSCORRUPTED;
270 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
271 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
273 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
276 __ext4_error_inode(inode, function, line, 0, -error,
277 "corrupted in-inode xattr");
281 #define xattr_check_inode(inode, header, end) \
282 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
285 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
286 void *end, int name_index, const char *name, int sorted)
288 struct ext4_xattr_entry *entry, *next;
294 name_len = strlen(name);
295 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
296 next = EXT4_XATTR_NEXT(entry);
297 if ((void *) next >= end) {
298 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
299 return -EFSCORRUPTED;
301 cmp = name_index - entry->e_name_index;
303 cmp = name_len - entry->e_name_len;
305 cmp = memcmp(name, entry->e_name, name_len);
306 if (cmp <= 0 && (sorted || cmp == 0))
310 return cmp ? -ENODATA : 0;
314 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
316 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
319 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
321 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
322 (u32) inode_peek_iversion_raw(ea_inode);
325 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
327 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
328 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
331 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
333 return (u32)ea_inode->i_atime.tv_sec;
336 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
338 ea_inode->i_atime.tv_sec = hash;
342 * Read the EA value from an inode.
344 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
346 int blocksize = 1 << ea_inode->i_blkbits;
347 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
348 int tail_size = (size % blocksize) ?: blocksize;
349 struct buffer_head *bhs_inline[8];
350 struct buffer_head **bhs = bhs_inline;
353 if (bh_count > ARRAY_SIZE(bhs_inline)) {
354 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
359 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
360 true /* wait */, bhs);
364 for (i = 0; i < bh_count; i++) {
365 /* There shouldn't be any holes in ea_inode. */
370 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
371 i < bh_count - 1 ? blocksize : tail_size);
375 for (i = 0; i < bh_count; i++)
378 if (bhs != bhs_inline)
383 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
385 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
386 u32 ea_inode_hash, struct inode **ea_inode)
391 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
393 err = PTR_ERR(inode);
394 ext4_error(parent->i_sb,
395 "error while reading EA inode %lu err=%d", ea_ino,
400 if (is_bad_inode(inode)) {
401 ext4_error(parent->i_sb,
402 "error while reading EA inode %lu is_bad_inode",
408 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
409 ext4_error(parent->i_sb,
410 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
416 ext4_xattr_inode_set_class(inode);
419 * Check whether this is an old Lustre-style xattr inode. Lustre
420 * implementation does not have hash validation, rather it has a
421 * backpointer from ea_inode to the parent inode.
423 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
424 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
425 inode->i_generation == parent->i_generation) {
426 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
427 ext4_xattr_inode_set_ref(inode, 1);
430 inode->i_flags |= S_NOQUOTA;
441 /* Remove entry from mbcache when EA inode is getting evicted */
442 void ext4_evict_ea_inode(struct inode *inode)
444 struct mb_cache_entry *oe;
446 if (!EA_INODE_CACHE(inode))
448 /* Wait for entry to get unused so that we can remove it */
449 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
450 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
451 mb_cache_entry_wait_unused(oe);
452 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
457 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
458 struct ext4_xattr_entry *entry, void *buffer,
463 /* Verify stored hash matches calculated hash. */
464 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
465 if (hash != ext4_xattr_inode_get_hash(ea_inode))
466 return -EFSCORRUPTED;
469 __le32 e_hash, tmp_data;
471 /* Verify entry hash. */
472 tmp_data = cpu_to_le32(hash);
473 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
476 if (e_hash == entry->e_hash)
480 * Not good. Maybe the entry hash was calculated
481 * using the buggy signed char version?
483 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
485 /* Still no match - bad */
486 if (e_hash != entry->e_hash)
487 return -EFSCORRUPTED;
489 /* Let people know about old hash */
490 pr_warn_once("ext4: filesystem with signed xattr name hash");
496 * Read xattr value from the EA inode.
499 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
500 void *buffer, size_t size)
502 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
503 struct inode *ea_inode;
506 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
507 le32_to_cpu(entry->e_hash), &ea_inode);
513 if (i_size_read(ea_inode) != size) {
514 ext4_warning_inode(ea_inode,
515 "ea_inode file size=%llu entry size=%zu",
516 i_size_read(ea_inode), size);
521 err = ext4_xattr_inode_read(ea_inode, buffer, size);
525 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
526 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
529 ext4_warning_inode(ea_inode,
530 "EA inode hash validation failed");
535 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
536 ext4_xattr_inode_get_hash(ea_inode),
537 ea_inode->i_ino, true /* reusable */);
545 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
546 void *buffer, size_t buffer_size)
548 struct buffer_head *bh = NULL;
549 struct ext4_xattr_entry *entry;
553 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
555 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
556 name_index, name, buffer, (long)buffer_size);
558 if (!EXT4_I(inode)->i_file_acl)
560 ea_idebug(inode, "reading block %llu",
561 (unsigned long long)EXT4_I(inode)->i_file_acl);
562 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
565 ea_bdebug(bh, "b_count=%d, refcount=%d",
566 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
567 error = ext4_xattr_check_block(inode, bh);
570 ext4_xattr_block_cache_insert(ea_block_cache, bh);
572 end = bh->b_data + bh->b_size;
573 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
576 size = le32_to_cpu(entry->e_value_size);
578 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
581 if (size > buffer_size)
583 if (entry->e_value_inum) {
584 error = ext4_xattr_inode_get(inode, entry, buffer,
589 u16 offset = le16_to_cpu(entry->e_value_offs);
590 void *p = bh->b_data + offset;
592 if (unlikely(p + size > end))
594 memcpy(buffer, p, size);
605 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
606 void *buffer, size_t buffer_size)
608 struct ext4_xattr_ibody_header *header;
609 struct ext4_xattr_entry *entry;
610 struct ext4_inode *raw_inode;
611 struct ext4_iloc iloc;
616 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
618 error = ext4_get_inode_loc(inode, &iloc);
621 raw_inode = ext4_raw_inode(&iloc);
622 header = IHDR(inode, raw_inode);
623 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
624 error = xattr_check_inode(inode, header, end);
627 entry = IFIRST(header);
628 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
631 size = le32_to_cpu(entry->e_value_size);
633 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
636 if (size > buffer_size)
638 if (entry->e_value_inum) {
639 error = ext4_xattr_inode_get(inode, entry, buffer,
644 u16 offset = le16_to_cpu(entry->e_value_offs);
645 void *p = (void *)IFIRST(header) + offset;
647 if (unlikely(p + size > end))
649 memcpy(buffer, p, size);
662 * Copy an extended attribute into the buffer
663 * provided, or compute the buffer size required.
664 * Buffer is NULL to compute the size of the buffer required.
666 * Returns a negative error number on failure, or the number of bytes
667 * used / required on success.
670 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
671 void *buffer, size_t buffer_size)
675 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
678 if (strlen(name) > 255)
681 down_read(&EXT4_I(inode)->xattr_sem);
682 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
684 if (error == -ENODATA)
685 error = ext4_xattr_block_get(inode, name_index, name, buffer,
687 up_read(&EXT4_I(inode)->xattr_sem);
692 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
693 char *buffer, size_t buffer_size)
695 size_t rest = buffer_size;
697 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
698 const struct xattr_handler *handler =
699 ext4_xattr_handler(entry->e_name_index);
701 if (handler && (!handler->list || handler->list(dentry))) {
702 const char *prefix = handler->prefix ?: handler->name;
703 size_t prefix_len = strlen(prefix);
704 size_t size = prefix_len + entry->e_name_len + 1;
709 memcpy(buffer, prefix, prefix_len);
710 buffer += prefix_len;
711 memcpy(buffer, entry->e_name, entry->e_name_len);
712 buffer += entry->e_name_len;
718 return buffer_size - rest; /* total size */
722 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
724 struct inode *inode = d_inode(dentry);
725 struct buffer_head *bh = NULL;
728 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
729 buffer, (long)buffer_size);
731 if (!EXT4_I(inode)->i_file_acl)
733 ea_idebug(inode, "reading block %llu",
734 (unsigned long long)EXT4_I(inode)->i_file_acl);
735 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
738 ea_bdebug(bh, "b_count=%d, refcount=%d",
739 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
740 error = ext4_xattr_check_block(inode, bh);
743 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
744 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
752 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
754 struct inode *inode = d_inode(dentry);
755 struct ext4_xattr_ibody_header *header;
756 struct ext4_inode *raw_inode;
757 struct ext4_iloc iloc;
761 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
763 error = ext4_get_inode_loc(inode, &iloc);
766 raw_inode = ext4_raw_inode(&iloc);
767 header = IHDR(inode, raw_inode);
768 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
769 error = xattr_check_inode(inode, header, end);
772 error = ext4_xattr_list_entries(dentry, IFIRST(header),
773 buffer, buffer_size);
781 * Inode operation listxattr()
783 * d_inode(dentry)->i_rwsem: don't care
785 * Copy a list of attribute names into the buffer
786 * provided, or compute the buffer size required.
787 * Buffer is NULL to compute the size of the buffer required.
789 * Returns a negative error number on failure, or the number of bytes
790 * used / required on success.
793 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
797 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
798 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
805 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
810 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
815 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
818 static void ext4_xattr_update_super_block(handle_t *handle,
819 struct super_block *sb)
821 if (ext4_has_feature_xattr(sb))
824 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
825 if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
826 EXT4_JTR_NONE) == 0) {
827 lock_buffer(EXT4_SB(sb)->s_sbh);
828 ext4_set_feature_xattr(sb);
829 ext4_superblock_csum_set(sb);
830 unlock_buffer(EXT4_SB(sb)->s_sbh);
831 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
835 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
837 struct ext4_iloc iloc = { .bh = NULL };
838 struct buffer_head *bh = NULL;
839 struct ext4_inode *raw_inode;
840 struct ext4_xattr_ibody_header *header;
841 struct ext4_xattr_entry *entry;
842 qsize_t ea_inode_refs = 0;
846 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
848 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
849 ret = ext4_get_inode_loc(inode, &iloc);
852 raw_inode = ext4_raw_inode(&iloc);
853 header = IHDR(inode, raw_inode);
854 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
855 ret = xattr_check_inode(inode, header, end);
859 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
860 entry = EXT4_XATTR_NEXT(entry))
861 if (entry->e_value_inum)
865 if (EXT4_I(inode)->i_file_acl) {
866 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
873 ret = ext4_xattr_check_block(inode, bh);
877 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
878 entry = EXT4_XATTR_NEXT(entry))
879 if (entry->e_value_inum)
882 *usage = ea_inode_refs + 1;
890 static inline size_t round_up_cluster(struct inode *inode, size_t length)
892 struct super_block *sb = inode->i_sb;
893 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
895 size_t mask = ~(cluster_size - 1);
897 return (length + cluster_size - 1) & mask;
900 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
904 err = dquot_alloc_inode(inode);
907 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
909 dquot_free_inode(inode);
913 static void ext4_xattr_inode_free_quota(struct inode *parent,
914 struct inode *ea_inode,
918 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
920 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
921 dquot_free_inode(parent);
924 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
925 struct buffer_head *block_bh, size_t value_len,
932 * 1) Owner inode update
933 * 2) Ref count update on old xattr block
935 * 4) block bitmap update for new xattr block
936 * 5) group descriptor for new xattr block
937 * 6) block bitmap update for old xattr block
938 * 7) group descriptor for old block
940 * 6 & 7 can happen if we have two racing threads T_a and T_b
941 * which are each trying to set an xattr on inodes I_a and I_b
942 * which were both initially sharing an xattr block.
947 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
950 * In case of inline data, we may push out the data to a block,
951 * so we need to reserve credits for this eventuality
953 if (inode && ext4_has_inline_data(inode))
954 credits += ext4_writepage_trans_blocks(inode) + 1;
956 /* We are done if ea_inode feature is not enabled. */
957 if (!ext4_has_feature_ea_inode(sb))
960 /* New ea_inode, inode map, block bitmap, group descriptor. */
964 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
966 /* Indirection block or one level of extent tree. */
969 /* Block bitmap and group descriptor updates for each block. */
970 credits += blocks * 2;
972 /* Blocks themselves. */
976 /* Dereference ea_inode holding old xattr value.
977 * Old ea_inode, inode map, block bitmap, group descriptor.
981 /* Data blocks for old ea_inode. */
982 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
984 /* Indirection block or one level of extent tree for old
989 /* Block bitmap and group descriptor updates for each block. */
990 credits += blocks * 2;
993 /* We may need to clone the existing xattr block in which case we need
994 * to increment ref counts for existing ea_inodes referenced by it.
997 struct ext4_xattr_entry *entry = BFIRST(block_bh);
999 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1000 if (entry->e_value_inum)
1001 /* Ref count update on ea_inode. */
1007 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1010 struct ext4_iloc iloc;
1014 inode_lock(ea_inode);
1016 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1020 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1021 ref_count += ref_change;
1022 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1024 if (ref_change > 0) {
1025 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1026 ea_inode->i_ino, ref_count);
1028 if (ref_count == 1) {
1029 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1030 ea_inode->i_ino, ea_inode->i_nlink);
1032 set_nlink(ea_inode, 1);
1033 ext4_orphan_del(handle, ea_inode);
1036 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1037 ea_inode->i_ino, ref_count);
1039 if (ref_count == 0) {
1040 WARN_ONCE(ea_inode->i_nlink != 1,
1041 "EA inode %lu i_nlink=%u",
1042 ea_inode->i_ino, ea_inode->i_nlink);
1044 clear_nlink(ea_inode);
1045 ext4_orphan_add(handle, ea_inode);
1049 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1051 ext4_warning_inode(ea_inode,
1052 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1054 inode_unlock(ea_inode);
1058 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1060 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1063 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1065 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1068 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1069 struct ext4_xattr_entry *first)
1071 struct inode *ea_inode;
1072 struct ext4_xattr_entry *entry;
1073 struct ext4_xattr_entry *failed_entry;
1074 unsigned int ea_ino;
1077 for (entry = first; !IS_LAST_ENTRY(entry);
1078 entry = EXT4_XATTR_NEXT(entry)) {
1079 if (!entry->e_value_inum)
1081 ea_ino = le32_to_cpu(entry->e_value_inum);
1082 err = ext4_xattr_inode_iget(parent, ea_ino,
1083 le32_to_cpu(entry->e_hash),
1087 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1089 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1099 failed_entry = entry;
1101 for (entry = first; entry != failed_entry;
1102 entry = EXT4_XATTR_NEXT(entry)) {
1103 if (!entry->e_value_inum)
1105 ea_ino = le32_to_cpu(entry->e_value_inum);
1106 err = ext4_xattr_inode_iget(parent, ea_ino,
1107 le32_to_cpu(entry->e_hash),
1110 ext4_warning(parent->i_sb,
1111 "cleanup ea_ino %u iget error %d", ea_ino,
1115 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1117 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1124 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1125 struct buffer_head *bh, bool block_csum, bool dirty)
1131 ext4_xattr_block_csum_set(inode, bh);
1132 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1134 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1143 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1144 struct buffer_head *bh,
1145 struct ext4_xattr_entry *first, bool block_csum,
1146 struct ext4_xattr_inode_array **ea_inode_array,
1147 int extra_credits, bool skip_quota)
1149 struct inode *ea_inode;
1150 struct ext4_xattr_entry *entry;
1152 unsigned int ea_ino;
1156 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1157 credits = 2 + extra_credits;
1159 for (entry = first; !IS_LAST_ENTRY(entry);
1160 entry = EXT4_XATTR_NEXT(entry)) {
1161 if (!entry->e_value_inum)
1163 ea_ino = le32_to_cpu(entry->e_value_inum);
1164 err = ext4_xattr_inode_iget(parent, ea_ino,
1165 le32_to_cpu(entry->e_hash),
1170 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1172 ext4_warning_inode(ea_inode,
1173 "Expand inode array err=%d", err);
1178 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1179 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1180 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1183 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1188 err = ext4_journal_get_write_access(handle,
1189 parent->i_sb, bh, EXT4_JTR_NONE);
1191 ext4_warning_inode(ea_inode,
1192 "Re-get write access err=%d",
1198 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1200 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1206 ext4_xattr_inode_free_quota(parent, ea_inode,
1207 le32_to_cpu(entry->e_value_size));
1210 * Forget about ea_inode within the same transaction that
1211 * decrements the ref count. This avoids duplicate decrements in
1212 * case the rest of the work spills over to subsequent
1215 entry->e_value_inum = 0;
1216 entry->e_value_size = 0;
1223 * Note that we are deliberately skipping csum calculation for
1224 * the final update because we do not expect any journal
1225 * restarts until xattr block is freed.
1228 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1230 ext4_warning_inode(parent,
1231 "handle dirty metadata err=%d", err);
1236 * Release the xattr block BH: If the reference count is > 1, decrement it;
1237 * otherwise free the block.
1240 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1241 struct buffer_head *bh,
1242 struct ext4_xattr_inode_array **ea_inode_array,
1245 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1249 BUFFER_TRACE(bh, "get_write_access");
1250 error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1257 hash = le32_to_cpu(BHDR(bh)->h_hash);
1258 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1260 ea_bdebug(bh, "refcount now=0; freeing");
1262 * This must happen under buffer lock for
1263 * ext4_xattr_block_set() to reliably detect freed block
1265 if (ea_block_cache) {
1266 struct mb_cache_entry *oe;
1268 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1272 mb_cache_entry_wait_unused(oe);
1273 mb_cache_entry_put(ea_block_cache, oe);
1280 if (ext4_has_feature_ea_inode(inode->i_sb))
1281 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1283 true /* block_csum */,
1286 true /* skip_quota */);
1287 ext4_free_blocks(handle, inode, bh, 0, 1,
1288 EXT4_FREE_BLOCKS_METADATA |
1289 EXT4_FREE_BLOCKS_FORGET);
1292 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1293 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1294 struct mb_cache_entry *ce;
1296 if (ea_block_cache) {
1297 ce = mb_cache_entry_get(ea_block_cache, hash,
1300 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1301 mb_cache_entry_put(ea_block_cache, ce);
1306 ext4_xattr_block_csum_set(inode, bh);
1308 * Beware of this ugliness: Releasing of xattr block references
1309 * from different inodes can race and so we have to protect
1310 * from a race where someone else frees the block (and releases
1311 * its journal_head) before we are done dirtying the buffer. In
1312 * nojournal mode this race is harmless and we actually cannot
1313 * call ext4_handle_dirty_metadata() with locked buffer as
1314 * that function can call sync_dirty_buffer() so for that case
1315 * we handle the dirtying after unlocking the buffer.
1317 if (ext4_handle_valid(handle))
1318 error = ext4_handle_dirty_metadata(handle, inode, bh);
1320 if (!ext4_handle_valid(handle))
1321 error = ext4_handle_dirty_metadata(handle, inode, bh);
1323 ext4_handle_sync(handle);
1324 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1325 ea_bdebug(bh, "refcount now=%d; releasing",
1326 le32_to_cpu(BHDR(bh)->h_refcount));
1329 ext4_std_error(inode->i_sb, error);
1334 * Find the available free space for EAs. This also returns the total number of
1335 * bytes used by EA entries.
1337 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1338 size_t *min_offs, void *base, int *total)
1340 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1341 if (!last->e_value_inum && last->e_value_size) {
1342 size_t offs = le16_to_cpu(last->e_value_offs);
1343 if (offs < *min_offs)
1347 *total += EXT4_XATTR_LEN(last->e_name_len);
1349 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1353 * Write the value of the EA in an inode.
1355 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1356 const void *buf, int bufsize)
1358 struct buffer_head *bh = NULL;
1359 unsigned long block = 0;
1360 int blocksize = ea_inode->i_sb->s_blocksize;
1361 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1362 int csize, wsize = 0;
1363 int ret = 0, ret2 = 0;
1367 while (ret >= 0 && ret < max_blocks) {
1368 struct ext4_map_blocks map;
1369 map.m_lblk = block += ret;
1370 map.m_len = max_blocks -= ret;
1372 ret = ext4_map_blocks(handle, ea_inode, &map,
1373 EXT4_GET_BLOCKS_CREATE);
1375 ext4_mark_inode_dirty(handle, ea_inode);
1376 if (ret == -ENOSPC &&
1377 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1389 while (wsize < bufsize) {
1391 csize = (bufsize - wsize) > blocksize ? blocksize :
1393 bh = ext4_getblk(handle, ea_inode, block, 0);
1398 EXT4_ERROR_INODE(ea_inode,
1399 "ext4_getblk() return bh = NULL");
1400 return -EFSCORRUPTED;
1402 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1407 memcpy(bh->b_data, buf, csize);
1408 set_buffer_uptodate(bh);
1409 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1416 inode_lock(ea_inode);
1417 i_size_write(ea_inode, wsize);
1418 ext4_update_i_disksize(ea_inode, wsize);
1419 inode_unlock(ea_inode);
1421 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1422 if (unlikely(ret2 && !ret))
1432 * Create an inode to store the value of a large EA.
1434 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1435 struct inode *inode, u32 hash)
1437 struct inode *ea_inode = NULL;
1438 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1442 * Let the next inode be the goal, so we try and allocate the EA inode
1443 * in the same group, or nearby one.
1445 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1446 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1448 if (!IS_ERR(ea_inode)) {
1449 ea_inode->i_op = &ext4_file_inode_operations;
1450 ea_inode->i_fop = &ext4_file_operations;
1451 ext4_set_aops(ea_inode);
1452 ext4_xattr_inode_set_class(ea_inode);
1453 unlock_new_inode(ea_inode);
1454 ext4_xattr_inode_set_ref(ea_inode, 1);
1455 ext4_xattr_inode_set_hash(ea_inode, hash);
1456 err = ext4_mark_inode_dirty(handle, ea_inode);
1458 err = ext4_inode_attach_jinode(ea_inode);
1460 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1461 ext4_warning_inode(ea_inode,
1462 "cleanup dec ref error %d", err);
1464 return ERR_PTR(err);
1468 * Xattr inodes are shared therefore quota charging is performed
1469 * at a higher level.
1471 dquot_free_inode(ea_inode);
1472 dquot_drop(ea_inode);
1473 inode_lock(ea_inode);
1474 ea_inode->i_flags |= S_NOQUOTA;
1475 inode_unlock(ea_inode);
1481 static struct inode *
1482 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1483 size_t value_len, u32 hash)
1485 struct inode *ea_inode;
1486 struct mb_cache_entry *ce;
1487 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1490 if (!ea_inode_cache)
1493 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1497 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1498 !(current->flags & PF_MEMALLOC_NOFS));
1500 ea_data = kvmalloc(value_len, GFP_KERNEL);
1502 mb_cache_entry_put(ea_inode_cache, ce);
1507 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1509 if (!IS_ERR(ea_inode) &&
1510 !is_bad_inode(ea_inode) &&
1511 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1512 i_size_read(ea_inode) == value_len &&
1513 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1514 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1516 !memcmp(value, ea_data, value_len)) {
1517 mb_cache_entry_touch(ea_inode_cache, ce);
1518 mb_cache_entry_put(ea_inode_cache, ce);
1523 if (!IS_ERR(ea_inode))
1525 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1532 * Add value of the EA in an inode.
1534 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1535 const void *value, size_t value_len,
1536 struct inode **ret_inode)
1538 struct inode *ea_inode;
1542 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1543 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1545 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1551 *ret_inode = ea_inode;
1555 /* Create an inode for the EA value */
1556 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1557 if (IS_ERR(ea_inode))
1558 return PTR_ERR(ea_inode);
1560 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1562 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1563 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1568 if (EA_INODE_CACHE(inode))
1569 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1570 ea_inode->i_ino, true /* reusable */);
1572 *ret_inode = ea_inode;
1577 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1578 * feature is enabled.
1580 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1582 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1583 struct ext4_xattr_search *s,
1584 handle_t *handle, struct inode *inode,
1587 struct ext4_xattr_entry *last, *next;
1588 struct ext4_xattr_entry *here = s->here;
1589 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1590 int in_inode = i->in_inode;
1591 struct inode *old_ea_inode = NULL;
1592 struct inode *new_ea_inode = NULL;
1593 size_t old_size, new_size;
1596 /* Space used by old and new values. */
1597 old_size = (!s->not_found && !here->e_value_inum) ?
1598 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1599 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1602 * Optimization for the simple case when old and new values have the
1603 * same padded sizes. Not applicable if external inodes are involved.
1605 if (new_size && new_size == old_size) {
1606 size_t offs = le16_to_cpu(here->e_value_offs);
1607 void *val = s->base + offs;
1609 here->e_value_size = cpu_to_le32(i->value_len);
1610 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1611 memset(val, 0, new_size);
1613 memcpy(val, i->value, i->value_len);
1614 /* Clear padding bytes. */
1615 memset(val + i->value_len, 0, new_size - i->value_len);
1620 /* Compute min_offs and last. */
1622 for (; !IS_LAST_ENTRY(last); last = next) {
1623 next = EXT4_XATTR_NEXT(last);
1624 if ((void *)next >= s->end) {
1625 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1626 ret = -EFSCORRUPTED;
1629 if (!last->e_value_inum && last->e_value_size) {
1630 size_t offs = le16_to_cpu(last->e_value_offs);
1631 if (offs < min_offs)
1636 /* Check whether we have enough space. */
1640 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1642 free += EXT4_XATTR_LEN(name_len) + old_size;
1644 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1650 * If storing the value in an external inode is an option,
1651 * reserve space for xattr entries/names in the external
1652 * attribute block so that a long value does not occupy the
1653 * whole space and prevent further entries being added.
1655 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1656 new_size && is_block &&
1657 (min_offs + old_size - new_size) <
1658 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1665 * Getting access to old and new ea inodes is subject to failures.
1666 * Finish that work before doing any modifications to the xattr data.
1668 if (!s->not_found && here->e_value_inum) {
1669 ret = ext4_xattr_inode_iget(inode,
1670 le32_to_cpu(here->e_value_inum),
1671 le32_to_cpu(here->e_hash),
1674 old_ea_inode = NULL;
1678 if (i->value && in_inode) {
1679 WARN_ON_ONCE(!i->value_len);
1681 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1685 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1689 new_ea_inode = NULL;
1690 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1696 /* We are ready to release ref count on the old_ea_inode. */
1697 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1699 /* Release newly required ref count on new_ea_inode. */
1703 err = ext4_xattr_inode_dec_ref(handle,
1706 ext4_warning_inode(new_ea_inode,
1707 "dec ref new_ea_inode err=%d",
1709 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1715 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1716 le32_to_cpu(here->e_value_size));
1719 /* No failures allowed past this point. */
1721 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1722 /* Remove the old value. */
1723 void *first_val = s->base + min_offs;
1724 size_t offs = le16_to_cpu(here->e_value_offs);
1725 void *val = s->base + offs;
1727 memmove(first_val + old_size, first_val, val - first_val);
1728 memset(first_val, 0, old_size);
1729 min_offs += old_size;
1731 /* Adjust all value offsets. */
1733 while (!IS_LAST_ENTRY(last)) {
1734 size_t o = le16_to_cpu(last->e_value_offs);
1736 if (!last->e_value_inum &&
1737 last->e_value_size && o < offs)
1738 last->e_value_offs = cpu_to_le16(o + old_size);
1739 last = EXT4_XATTR_NEXT(last);
1744 /* Remove old name. */
1745 size_t size = EXT4_XATTR_LEN(name_len);
1747 last = ENTRY((void *)last - size);
1748 memmove(here, (void *)here + size,
1749 (void *)last - (void *)here + sizeof(__u32));
1750 memset(last, 0, size);
1751 } else if (s->not_found) {
1752 /* Insert new name. */
1753 size_t size = EXT4_XATTR_LEN(name_len);
1754 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1756 memmove((void *)here + size, here, rest);
1757 memset(here, 0, size);
1758 here->e_name_index = i->name_index;
1759 here->e_name_len = name_len;
1760 memcpy(here->e_name, i->name, name_len);
1762 /* This is an update, reset value info. */
1763 here->e_value_inum = 0;
1764 here->e_value_offs = 0;
1765 here->e_value_size = 0;
1769 /* Insert new value. */
1771 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1772 } else if (i->value_len) {
1773 void *val = s->base + min_offs - new_size;
1775 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1776 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1777 memset(val, 0, new_size);
1779 memcpy(val, i->value, i->value_len);
1780 /* Clear padding bytes. */
1781 memset(val + i->value_len, 0,
1782 new_size - i->value_len);
1785 here->e_value_size = cpu_to_le32(i->value_len);
1792 /* Entry hash calculation. */
1797 * Feed crc32c hash instead of the raw value for entry
1798 * hash calculation. This is to avoid walking
1799 * potentially long value buffer again.
1801 crc32c_hash = cpu_to_le32(
1802 ext4_xattr_inode_get_hash(new_ea_inode));
1803 hash = ext4_xattr_hash_entry(here->e_name,
1806 } else if (is_block) {
1807 __le32 *value = s->base + le16_to_cpu(
1808 here->e_value_offs);
1810 hash = ext4_xattr_hash_entry(here->e_name,
1811 here->e_name_len, value,
1814 here->e_hash = hash;
1818 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1827 struct ext4_xattr_block_find {
1828 struct ext4_xattr_search s;
1829 struct buffer_head *bh;
1833 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1834 struct ext4_xattr_block_find *bs)
1836 struct super_block *sb = inode->i_sb;
1839 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1840 i->name_index, i->name, i->value, (long)i->value_len);
1842 if (EXT4_I(inode)->i_file_acl) {
1843 /* The inode already has an extended attribute block. */
1844 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1845 if (IS_ERR(bs->bh)) {
1846 error = PTR_ERR(bs->bh);
1850 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1851 atomic_read(&(bs->bh->b_count)),
1852 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1853 error = ext4_xattr_check_block(inode, bs->bh);
1856 /* Find the named attribute. */
1857 bs->s.base = BHDR(bs->bh);
1858 bs->s.first = BFIRST(bs->bh);
1859 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1860 bs->s.here = bs->s.first;
1861 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1862 i->name_index, i->name, 1);
1863 if (error && error != -ENODATA)
1865 bs->s.not_found = error;
1871 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1872 struct ext4_xattr_info *i,
1873 struct ext4_xattr_block_find *bs)
1875 struct super_block *sb = inode->i_sb;
1876 struct buffer_head *new_bh = NULL;
1877 struct ext4_xattr_search s_copy = bs->s;
1878 struct ext4_xattr_search *s = &s_copy;
1879 struct mb_cache_entry *ce = NULL;
1881 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1882 struct inode *ea_inode = NULL, *tmp_inode;
1883 size_t old_ea_inode_quota = 0;
1884 unsigned int ea_ino;
1887 #define header(x) ((struct ext4_xattr_header *)(x))
1890 int offset = (char *)s->here - bs->bh->b_data;
1892 BUFFER_TRACE(bs->bh, "get_write_access");
1893 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1897 lock_buffer(bs->bh);
1899 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1900 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1903 * This must happen under buffer lock for
1904 * ext4_xattr_block_set() to reliably detect modified
1907 if (ea_block_cache) {
1908 struct mb_cache_entry *oe;
1910 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1911 hash, bs->bh->b_blocknr);
1914 * Xattr block is getting reused. Leave
1917 mb_cache_entry_put(ea_block_cache, oe);
1921 ea_bdebug(bs->bh, "modifying in-place");
1922 error = ext4_xattr_set_entry(i, s, handle, inode,
1923 true /* is_block */);
1924 ext4_xattr_block_csum_set(inode, bs->bh);
1925 unlock_buffer(bs->bh);
1926 if (error == -EFSCORRUPTED)
1929 error = ext4_handle_dirty_metadata(handle,
1937 unlock_buffer(bs->bh);
1938 ea_bdebug(bs->bh, "cloning");
1939 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1941 if (s->base == NULL)
1943 s->first = ENTRY(header(s->base)+1);
1944 header(s->base)->h_refcount = cpu_to_le32(1);
1945 s->here = ENTRY(s->base + offset);
1946 s->end = s->base + bs->bh->b_size;
1949 * If existing entry points to an xattr inode, we need
1950 * to prevent ext4_xattr_set_entry() from decrementing
1951 * ref count on it because the reference belongs to the
1952 * original block. In this case, make the entry look
1953 * like it has an empty value.
1955 if (!s->not_found && s->here->e_value_inum) {
1956 ea_ino = le32_to_cpu(s->here->e_value_inum);
1957 error = ext4_xattr_inode_iget(inode, ea_ino,
1958 le32_to_cpu(s->here->e_hash),
1963 if (!ext4_test_inode_state(tmp_inode,
1964 EXT4_STATE_LUSTRE_EA_INODE)) {
1966 * Defer quota free call for previous
1967 * inode until success is guaranteed.
1969 old_ea_inode_quota = le32_to_cpu(
1970 s->here->e_value_size);
1974 s->here->e_value_inum = 0;
1975 s->here->e_value_size = 0;
1978 /* Allocate a buffer where we construct the new block. */
1979 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1981 if (s->base == NULL)
1983 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1984 header(s->base)->h_blocks = cpu_to_le32(1);
1985 header(s->base)->h_refcount = cpu_to_le32(1);
1986 s->first = ENTRY(header(s->base)+1);
1987 s->here = ENTRY(header(s->base)+1);
1988 s->end = s->base + sb->s_blocksize;
1991 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1992 if (error == -EFSCORRUPTED)
1997 if (i->value && s->here->e_value_inum) {
1999 * A ref count on ea_inode has been taken as part of the call to
2000 * ext4_xattr_set_entry() above. We would like to drop this
2001 * extra ref but we have to wait until the xattr block is
2002 * initialized and has its own ref count on the ea_inode.
2004 ea_ino = le32_to_cpu(s->here->e_value_inum);
2005 error = ext4_xattr_inode_iget(inode, ea_ino,
2006 le32_to_cpu(s->here->e_hash),
2015 if (!IS_LAST_ENTRY(s->first)) {
2016 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2019 /* We found an identical block in the cache. */
2020 if (new_bh == bs->bh)
2021 ea_bdebug(new_bh, "keeping");
2025 WARN_ON_ONCE(dquot_initialize_needed(inode));
2027 /* The old block is released after updating
2029 error = dquot_alloc_block(inode,
2030 EXT4_C2B(EXT4_SB(sb), 1));
2033 BUFFER_TRACE(new_bh, "get_write_access");
2034 error = ext4_journal_get_write_access(
2039 lock_buffer(new_bh);
2041 * We have to be careful about races with
2042 * adding references to xattr block. Once we
2043 * hold buffer lock xattr block's state is
2044 * stable so we can check the additional
2047 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2048 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2050 * Undo everything and check mbcache
2053 unlock_buffer(new_bh);
2054 dquot_free_block(inode,
2055 EXT4_C2B(EXT4_SB(sb),
2058 mb_cache_entry_put(ea_block_cache, ce);
2063 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2064 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2065 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2066 ea_bdebug(new_bh, "reusing; refcount now=%d",
2068 ext4_xattr_block_csum_set(inode, new_bh);
2069 unlock_buffer(new_bh);
2070 error = ext4_handle_dirty_metadata(handle,
2076 mb_cache_entry_touch(ea_block_cache, ce);
2077 mb_cache_entry_put(ea_block_cache, ce);
2079 } else if (bs->bh && s->base == bs->bh->b_data) {
2080 /* We were modifying this block in-place. */
2081 ea_bdebug(bs->bh, "keeping this block");
2082 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2086 /* We need to allocate a new block */
2087 ext4_fsblk_t goal, block;
2089 WARN_ON_ONCE(dquot_initialize_needed(inode));
2091 goal = ext4_group_first_block_no(sb,
2092 EXT4_I(inode)->i_block_group);
2093 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2098 ea_idebug(inode, "creating block %llu",
2099 (unsigned long long)block);
2101 new_bh = sb_getblk(sb, block);
2102 if (unlikely(!new_bh)) {
2105 ext4_free_blocks(handle, inode, NULL, block, 1,
2106 EXT4_FREE_BLOCKS_METADATA);
2109 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2110 ENTRY(header(s->base)+1));
2114 /* Drop the extra ref on ea_inode. */
2115 error = ext4_xattr_inode_dec_ref(handle,
2118 ext4_warning_inode(ea_inode,
2125 lock_buffer(new_bh);
2126 error = ext4_journal_get_create_access(handle, sb,
2127 new_bh, EXT4_JTR_NONE);
2129 unlock_buffer(new_bh);
2133 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2134 ext4_xattr_block_csum_set(inode, new_bh);
2135 set_buffer_uptodate(new_bh);
2136 unlock_buffer(new_bh);
2137 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2138 error = ext4_handle_dirty_metadata(handle, inode,
2145 if (old_ea_inode_quota)
2146 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2148 /* Update the inode. */
2149 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2151 /* Drop the previous xattr block. */
2152 if (bs->bh && bs->bh != new_bh) {
2153 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2155 ext4_xattr_release_block(handle, inode, bs->bh,
2157 0 /* extra_credits */);
2158 ext4_xattr_inode_array_free(ea_inode_array);
2166 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2168 ext4_warning_inode(ea_inode, "dec ref error=%d",
2171 /* If there was an error, revert the quota charge. */
2173 ext4_xattr_inode_free_quota(inode, ea_inode,
2174 i_size_read(ea_inode));
2178 mb_cache_entry_put(ea_block_cache, ce);
2180 if (!(bs->bh && s->base == bs->bh->b_data))
2186 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2190 EXT4_ERROR_INODE(inode, "bad block %llu",
2191 EXT4_I(inode)->i_file_acl);
2197 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2198 struct ext4_xattr_ibody_find *is)
2200 struct ext4_xattr_ibody_header *header;
2201 struct ext4_inode *raw_inode;
2204 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2207 raw_inode = ext4_raw_inode(&is->iloc);
2208 header = IHDR(inode, raw_inode);
2209 is->s.base = is->s.first = IFIRST(header);
2210 is->s.here = is->s.first;
2211 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2212 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2213 error = xattr_check_inode(inode, header, is->s.end);
2216 /* Find the named attribute. */
2217 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2218 i->name_index, i->name, 0);
2219 if (error && error != -ENODATA)
2221 is->s.not_found = error;
2226 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2227 struct ext4_xattr_info *i,
2228 struct ext4_xattr_ibody_find *is)
2230 struct ext4_xattr_ibody_header *header;
2231 struct ext4_xattr_search *s = &is->s;
2234 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2237 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2240 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2241 if (!IS_LAST_ENTRY(s->first)) {
2242 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2243 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2245 header->h_magic = cpu_to_le32(0);
2246 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2251 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2252 struct ext4_xattr_info *i)
2256 /* When e_value_inum is set the value is stored externally. */
2257 if (s->here->e_value_inum)
2259 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2261 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2262 return !memcmp(value, i->value, i->value_len);
2265 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2267 struct buffer_head *bh;
2270 if (!EXT4_I(inode)->i_file_acl)
2272 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2275 error = ext4_xattr_check_block(inode, bh);
2278 return ERR_PTR(error);
2284 * ext4_xattr_set_handle()
2286 * Create, replace or remove an extended attribute for this inode. Value
2287 * is NULL to remove an existing extended attribute, and non-NULL to
2288 * either replace an existing extended attribute, or create a new extended
2289 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2290 * specify that an extended attribute must exist and must not exist
2291 * previous to the call, respectively.
2293 * Returns 0, or a negative error number on failure.
2296 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2297 const char *name, const void *value, size_t value_len,
2300 struct ext4_xattr_info i = {
2301 .name_index = name_index,
2304 .value_len = value_len,
2307 struct ext4_xattr_ibody_find is = {
2308 .s = { .not_found = -ENODATA, },
2310 struct ext4_xattr_block_find bs = {
2311 .s = { .not_found = -ENODATA, },
2318 if (strlen(name) > 255)
2321 ext4_write_lock_xattr(inode, &no_expand);
2323 /* Check journal credits under write lock. */
2324 if (ext4_handle_valid(handle)) {
2325 struct buffer_head *bh;
2328 bh = ext4_xattr_get_block(inode);
2330 error = PTR_ERR(bh);
2334 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2336 flags & XATTR_CREATE);
2339 if (jbd2_handle_buffer_credits(handle) < credits) {
2343 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2346 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2350 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2351 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2352 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2353 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2356 error = ext4_xattr_ibody_find(inode, &i, &is);
2360 error = ext4_xattr_block_find(inode, &i, &bs);
2363 if (is.s.not_found && bs.s.not_found) {
2365 if (flags & XATTR_REPLACE)
2372 if (flags & XATTR_CREATE)
2377 if (!is.s.not_found)
2378 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2379 else if (!bs.s.not_found)
2380 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2383 /* Xattr value did not change? Save us some work and bail out */
2384 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2386 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2389 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2390 (EXT4_XATTR_SIZE(i.value_len) >
2391 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2394 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2395 if (!error && !bs.s.not_found) {
2397 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2398 } else if (error == -ENOSPC) {
2399 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2402 error = ext4_xattr_block_find(inode, &i, &bs);
2406 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2407 if (!error && !is.s.not_found) {
2409 error = ext4_xattr_ibody_set(handle, inode, &i,
2411 } else if (error == -ENOSPC) {
2413 * Xattr does not fit in the block, store at
2414 * external inode if possible.
2416 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2417 i.value_len && !i.in_inode) {
2425 ext4_xattr_update_super_block(handle, inode->i_sb);
2426 inode->i_ctime = current_time(inode);
2427 inode_inc_iversion(inode);
2430 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2432 * The bh is consumed by ext4_mark_iloc_dirty, even with
2437 ext4_handle_sync(handle);
2439 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2444 ext4_write_unlock_xattr(inode, &no_expand);
2448 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2449 bool is_create, int *credits)
2451 struct buffer_head *bh;
2456 if (!EXT4_SB(inode->i_sb)->s_journal)
2459 down_read(&EXT4_I(inode)->xattr_sem);
2461 bh = ext4_xattr_get_block(inode);
2465 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2466 value_len, is_create);
2471 up_read(&EXT4_I(inode)->xattr_sem);
2478 * Like ext4_xattr_set_handle, but start from an inode. This extended
2479 * attribute modification is a filesystem transaction by itself.
2481 * Returns 0, or a negative error number on failure.
2484 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2485 const void *value, size_t value_len, int flags)
2488 struct super_block *sb = inode->i_sb;
2489 int error, retries = 0;
2492 error = dquot_initialize(inode);
2497 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2502 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2503 if (IS_ERR(handle)) {
2504 error = PTR_ERR(handle);
2508 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2509 value, value_len, flags);
2510 error2 = ext4_journal_stop(handle);
2511 if (error == -ENOSPC &&
2512 ext4_should_retry_alloc(sb, &retries))
2517 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2523 * Shift the EA entries in the inode to create space for the increased
2526 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2527 int value_offs_shift, void *to,
2528 void *from, size_t n)
2530 struct ext4_xattr_entry *last = entry;
2533 /* We always shift xattr headers further thus offsets get lower */
2534 BUG_ON(value_offs_shift > 0);
2536 /* Adjust the value offsets of the entries */
2537 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2538 if (!last->e_value_inum && last->e_value_size) {
2539 new_offs = le16_to_cpu(last->e_value_offs) +
2541 last->e_value_offs = cpu_to_le16(new_offs);
2544 /* Shift the entries by n bytes */
2545 memmove(to, from, n);
2549 * Move xattr pointed to by 'entry' from inode into external xattr block
2551 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2552 struct ext4_inode *raw_inode,
2553 struct ext4_xattr_entry *entry)
2555 struct ext4_xattr_ibody_find *is = NULL;
2556 struct ext4_xattr_block_find *bs = NULL;
2557 char *buffer = NULL, *b_entry_name = NULL;
2558 size_t value_size = le32_to_cpu(entry->e_value_size);
2559 struct ext4_xattr_info i = {
2562 .name_index = entry->e_name_index,
2563 .in_inode = !!entry->e_value_inum,
2565 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2568 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2569 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2570 buffer = kvmalloc(value_size, GFP_NOFS);
2571 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2572 if (!is || !bs || !buffer || !b_entry_name) {
2577 is->s.not_found = -ENODATA;
2578 bs->s.not_found = -ENODATA;
2582 /* Save the entry name and the entry value */
2583 if (entry->e_value_inum) {
2584 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2588 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2589 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2592 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2593 b_entry_name[entry->e_name_len] = '\0';
2594 i.name = b_entry_name;
2596 error = ext4_get_inode_loc(inode, &is->iloc);
2600 error = ext4_xattr_ibody_find(inode, &i, is);
2604 /* Remove the chosen entry from the inode */
2605 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2610 i.value_len = value_size;
2611 error = ext4_xattr_block_find(inode, &i, bs);
2615 /* Add entry which was removed from the inode into the block */
2616 error = ext4_xattr_block_set(handle, inode, &i, bs);
2621 kfree(b_entry_name);
2624 brelse(is->iloc.bh);
2633 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2634 struct ext4_inode *raw_inode,
2635 int isize_diff, size_t ifree,
2636 size_t bfree, int *total_ino)
2638 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2639 struct ext4_xattr_entry *small_entry;
2640 struct ext4_xattr_entry *entry;
2641 struct ext4_xattr_entry *last;
2642 unsigned int entry_size; /* EA entry size */
2643 unsigned int total_size; /* EA entry size + value size */
2644 unsigned int min_total_size;
2647 while (isize_diff > ifree) {
2650 min_total_size = ~0U;
2651 last = IFIRST(header);
2652 /* Find the entry best suited to be pushed into EA block */
2653 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2654 /* never move system.data out of the inode */
2655 if ((last->e_name_len == 4) &&
2656 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2657 !memcmp(last->e_name, "data", 4))
2659 total_size = EXT4_XATTR_LEN(last->e_name_len);
2660 if (!last->e_value_inum)
2661 total_size += EXT4_XATTR_SIZE(
2662 le32_to_cpu(last->e_value_size));
2663 if (total_size <= bfree &&
2664 total_size < min_total_size) {
2665 if (total_size + ifree < isize_diff) {
2669 min_total_size = total_size;
2674 if (entry == NULL) {
2675 if (small_entry == NULL)
2677 entry = small_entry;
2680 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2681 total_size = entry_size;
2682 if (!entry->e_value_inum)
2683 total_size += EXT4_XATTR_SIZE(
2684 le32_to_cpu(entry->e_value_size));
2685 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2690 *total_ino -= entry_size;
2691 ifree += total_size;
2692 bfree -= total_size;
2699 * Expand an inode by new_extra_isize bytes when EAs are present.
2700 * Returns 0 on success or negative error number on failure.
2702 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2703 struct ext4_inode *raw_inode, handle_t *handle)
2705 struct ext4_xattr_ibody_header *header;
2706 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2707 static unsigned int mnt_count;
2709 size_t ifree, bfree;
2712 int error = 0, tried_min_extra_isize = 0;
2713 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2714 int isize_diff; /* How much do we need to grow i_extra_isize */
2717 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2718 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2721 header = IHDR(inode, raw_inode);
2724 * Check if enough free space is available in the inode to shift the
2725 * entries ahead by new_extra_isize.
2728 base = IFIRST(header);
2729 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2730 min_offs = end - base;
2731 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2733 error = xattr_check_inode(inode, header, end);
2737 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2738 if (ifree >= isize_diff)
2742 * Enough free space isn't available in the inode, check if
2743 * EA block can hold new_extra_isize bytes.
2745 if (EXT4_I(inode)->i_file_acl) {
2746 struct buffer_head *bh;
2748 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2750 error = PTR_ERR(bh);
2753 error = ext4_xattr_check_block(inode, bh);
2759 end = bh->b_data + bh->b_size;
2760 min_offs = end - base;
2761 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2764 if (bfree + ifree < isize_diff) {
2765 if (!tried_min_extra_isize && s_min_extra_isize) {
2766 tried_min_extra_isize++;
2767 new_extra_isize = s_min_extra_isize;
2774 bfree = inode->i_sb->s_blocksize;
2777 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2778 isize_diff, ifree, bfree,
2781 if (error == -ENOSPC && !tried_min_extra_isize &&
2782 s_min_extra_isize) {
2783 tried_min_extra_isize++;
2784 new_extra_isize = s_min_extra_isize;
2790 /* Adjust the offsets and shift the remaining entries ahead */
2791 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2792 - new_extra_isize, (void *)raw_inode +
2793 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2794 (void *)header, total_ino);
2795 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2798 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2799 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2801 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2806 #define EIA_INCR 16 /* must be 2^n */
2807 #define EIA_MASK (EIA_INCR - 1)
2809 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2810 * If @ea_inode_array is new or full it will be grown and the old
2811 * contents copied over.
2814 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2815 struct inode *inode)
2817 if (*ea_inode_array == NULL) {
2819 * Start with 15 inodes, so it fits into a power-of-two size.
2820 * If *ea_inode_array is NULL, this is essentially offsetof()
2823 kmalloc(offsetof(struct ext4_xattr_inode_array,
2826 if (*ea_inode_array == NULL)
2828 (*ea_inode_array)->count = 0;
2829 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2830 /* expand the array once all 15 + n * 16 slots are full */
2831 struct ext4_xattr_inode_array *new_array = NULL;
2832 int count = (*ea_inode_array)->count;
2834 /* if new_array is NULL, this is essentially offsetof() */
2835 new_array = kmalloc(
2836 offsetof(struct ext4_xattr_inode_array,
2837 inodes[count + EIA_INCR]),
2839 if (new_array == NULL)
2841 memcpy(new_array, *ea_inode_array,
2842 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2843 kfree(*ea_inode_array);
2844 *ea_inode_array = new_array;
2846 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2851 * ext4_xattr_delete_inode()
2853 * Free extended attribute resources associated with this inode. Traverse
2854 * all entries and decrement reference on any xattr inodes associated with this
2855 * inode. This is called immediately before an inode is freed. We have exclusive
2856 * access to the inode. If an orphan inode is deleted it will also release its
2857 * references on xattr block and xattr inodes.
2859 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2860 struct ext4_xattr_inode_array **ea_inode_array,
2863 struct buffer_head *bh = NULL;
2864 struct ext4_xattr_ibody_header *header;
2865 struct ext4_iloc iloc = { .bh = NULL };
2866 struct ext4_xattr_entry *entry;
2867 struct inode *ea_inode;
2870 error = ext4_journal_ensure_credits(handle, extra_credits,
2871 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2873 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2877 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2878 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2880 error = ext4_get_inode_loc(inode, &iloc);
2882 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2886 error = ext4_journal_get_write_access(handle, inode->i_sb,
2887 iloc.bh, EXT4_JTR_NONE);
2889 EXT4_ERROR_INODE(inode, "write access (error %d)",
2894 header = IHDR(inode, ext4_raw_inode(&iloc));
2895 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2896 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2898 false /* block_csum */,
2901 false /* skip_quota */);
2904 if (EXT4_I(inode)->i_file_acl) {
2905 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2907 error = PTR_ERR(bh);
2908 if (error == -EIO) {
2909 EXT4_ERROR_INODE_ERR(inode, EIO,
2910 "block %llu read error",
2911 EXT4_I(inode)->i_file_acl);
2916 error = ext4_xattr_check_block(inode, bh);
2920 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2921 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2922 entry = EXT4_XATTR_NEXT(entry)) {
2923 if (!entry->e_value_inum)
2925 error = ext4_xattr_inode_iget(inode,
2926 le32_to_cpu(entry->e_value_inum),
2927 le32_to_cpu(entry->e_hash),
2931 ext4_xattr_inode_free_quota(inode, ea_inode,
2932 le32_to_cpu(entry->e_value_size));
2938 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2941 * Update i_file_acl value in the same transaction that releases
2944 EXT4_I(inode)->i_file_acl = 0;
2945 error = ext4_mark_inode_dirty(handle, inode);
2947 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2951 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2960 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2964 if (ea_inode_array == NULL)
2967 for (idx = 0; idx < ea_inode_array->count; ++idx)
2968 iput(ea_inode_array->inodes[idx]);
2969 kfree(ea_inode_array);
2973 * ext4_xattr_block_cache_insert()
2975 * Create a new entry in the extended attribute block cache, and insert
2976 * it unless such an entry is already in the cache.
2978 * Returns 0, or a negative error number on failure.
2981 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2982 struct buffer_head *bh)
2984 struct ext4_xattr_header *header = BHDR(bh);
2985 __u32 hash = le32_to_cpu(header->h_hash);
2986 int reusable = le32_to_cpu(header->h_refcount) <
2987 EXT4_XATTR_REFCOUNT_MAX;
2990 if (!ea_block_cache)
2992 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2993 bh->b_blocknr, reusable);
2995 if (error == -EBUSY)
2996 ea_bdebug(bh, "already in cache");
2998 ea_bdebug(bh, "inserting [%x]", (int)hash);
3004 * Compare two extended attribute blocks for equality.
3006 * Returns 0 if the blocks are equal, 1 if they differ, and
3007 * a negative error number on errors.
3010 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3011 struct ext4_xattr_header *header2)
3013 struct ext4_xattr_entry *entry1, *entry2;
3015 entry1 = ENTRY(header1+1);
3016 entry2 = ENTRY(header2+1);
3017 while (!IS_LAST_ENTRY(entry1)) {
3018 if (IS_LAST_ENTRY(entry2))
3020 if (entry1->e_hash != entry2->e_hash ||
3021 entry1->e_name_index != entry2->e_name_index ||
3022 entry1->e_name_len != entry2->e_name_len ||
3023 entry1->e_value_size != entry2->e_value_size ||
3024 entry1->e_value_inum != entry2->e_value_inum ||
3025 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3027 if (!entry1->e_value_inum &&
3028 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3029 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3030 le32_to_cpu(entry1->e_value_size)))
3033 entry1 = EXT4_XATTR_NEXT(entry1);
3034 entry2 = EXT4_XATTR_NEXT(entry2);
3036 if (!IS_LAST_ENTRY(entry2))
3042 * ext4_xattr_block_cache_find()
3044 * Find an identical extended attribute block.
3046 * Returns a pointer to the block found, or NULL if such a block was
3047 * not found or an error occurred.
3049 static struct buffer_head *
3050 ext4_xattr_block_cache_find(struct inode *inode,
3051 struct ext4_xattr_header *header,
3052 struct mb_cache_entry **pce)
3054 __u32 hash = le32_to_cpu(header->h_hash);
3055 struct mb_cache_entry *ce;
3056 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3058 if (!ea_block_cache)
3060 if (!header->h_hash)
3061 return NULL; /* never share */
3062 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3063 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3065 struct buffer_head *bh;
3067 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3069 if (PTR_ERR(bh) == -ENOMEM)
3072 EXT4_ERROR_INODE(inode, "block %lu read error",
3073 (unsigned long)ce->e_value);
3074 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3079 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3084 #define NAME_HASH_SHIFT 5
3085 #define VALUE_HASH_SHIFT 16
3088 * ext4_xattr_hash_entry()
3090 * Compute the hash of an extended attribute.
3092 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3097 while (name_len--) {
3098 hash = (hash << NAME_HASH_SHIFT) ^
3099 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3100 (unsigned char)*name++;
3102 while (value_count--) {
3103 hash = (hash << VALUE_HASH_SHIFT) ^
3104 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3105 le32_to_cpu(*value++);
3107 return cpu_to_le32(hash);
3111 * ext4_xattr_hash_entry_signed()
3113 * Compute the hash of an extended attribute incorrectly.
3115 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3119 while (name_len--) {
3120 hash = (hash << NAME_HASH_SHIFT) ^
3121 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3122 (signed char)*name++;
3124 while (value_count--) {
3125 hash = (hash << VALUE_HASH_SHIFT) ^
3126 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3127 le32_to_cpu(*value++);
3129 return cpu_to_le32(hash);
3132 #undef NAME_HASH_SHIFT
3133 #undef VALUE_HASH_SHIFT
3135 #define BLOCK_HASH_SHIFT 16
3138 * ext4_xattr_rehash()
3140 * Re-compute the extended attribute hash value after an entry has changed.
3142 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3144 struct ext4_xattr_entry *here;
3147 here = ENTRY(header+1);
3148 while (!IS_LAST_ENTRY(here)) {
3149 if (!here->e_hash) {
3150 /* Block is not shared if an entry's hash value == 0 */
3154 hash = (hash << BLOCK_HASH_SHIFT) ^
3155 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3156 le32_to_cpu(here->e_hash);
3157 here = EXT4_XATTR_NEXT(here);
3159 header->h_hash = cpu_to_le32(hash);
3162 #undef BLOCK_HASH_SHIFT
3164 #define HASH_BUCKET_BITS 10
3167 ext4_xattr_create_cache(void)
3169 return mb_cache_create(HASH_BUCKET_BITS);
3172 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3175 mb_cache_destroy(cache);