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] = &nop_posix_acl_access,
92 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &nop_posix_acl_default,
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_SECURITY
105 &ext4_xattr_security_handler,
107 &ext4_xattr_hurd_handler,
111 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
112 inode->i_sb->s_fs_info)->s_ea_block_cache)
114 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
115 inode->i_sb->s_fs_info)->s_ea_inode_cache)
118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119 struct inode *inode);
121 #ifdef CONFIG_LOCKDEP
122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
124 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
128 static __le32 ext4_xattr_block_csum(struct inode *inode,
130 struct ext4_xattr_header *hdr)
132 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
134 __le64 dsk_block_nr = cpu_to_le64(block_nr);
135 __u32 dummy_csum = 0;
136 int offset = offsetof(struct ext4_xattr_header, h_checksum);
138 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
139 sizeof(dsk_block_nr));
140 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
141 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
142 offset += sizeof(dummy_csum);
143 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
144 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
146 return cpu_to_le32(csum);
149 static int ext4_xattr_block_csum_verify(struct inode *inode,
150 struct buffer_head *bh)
152 struct ext4_xattr_header *hdr = BHDR(bh);
155 if (ext4_has_metadata_csum(inode->i_sb)) {
157 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
158 bh->b_blocknr, hdr));
164 static void ext4_xattr_block_csum_set(struct inode *inode,
165 struct buffer_head *bh)
167 if (ext4_has_metadata_csum(inode->i_sb))
168 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
169 bh->b_blocknr, BHDR(bh));
172 static inline const char *ext4_xattr_prefix(int name_index,
173 struct dentry *dentry)
175 const struct xattr_handler *handler = NULL;
177 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
178 handler = ext4_xattr_handler_map[name_index];
180 if (!xattr_handler_can_list(handler, dentry))
183 return xattr_prefix(handler);
187 check_xattrs(struct inode *inode, struct buffer_head *bh,
188 struct ext4_xattr_entry *entry, void *end, void *value_start,
189 const char *function, unsigned int line)
191 struct ext4_xattr_entry *e = entry;
192 int err = -EFSCORRUPTED;
196 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
197 BHDR(bh)->h_blocks != cpu_to_le32(1)) {
198 err_str = "invalid header";
201 if (buffer_verified(bh))
203 if (!ext4_xattr_block_csum_verify(inode, bh)) {
205 err_str = "invalid checksum";
209 struct ext4_xattr_ibody_header *header = value_start;
212 if (end - (void *)header < sizeof(*header) + sizeof(u32)) {
213 err_str = "in-inode xattr block too small";
216 if (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
217 err_str = "bad magic number in in-inode xattr";
222 /* Find the end of the names list */
223 while (!IS_LAST_ENTRY(e)) {
224 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
225 if ((void *)next >= end) {
226 err_str = "e_name out of bounds";
229 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len) {
230 err_str = "bad e_name length";
236 /* Check the values */
237 while (!IS_LAST_ENTRY(entry)) {
238 u32 size = le32_to_cpu(entry->e_value_size);
239 unsigned long ea_ino = le32_to_cpu(entry->e_value_inum);
241 if (!ext4_has_feature_ea_inode(inode->i_sb) && ea_ino) {
242 err_str = "ea_inode specified without ea_inode feature enabled";
245 if (ea_ino && ((ea_ino == EXT4_ROOT_INO) ||
246 !ext4_valid_inum(inode->i_sb, ea_ino))) {
247 err_str = "invalid ea_ino";
250 if (size > EXT4_XATTR_SIZE_MAX) {
251 err_str = "e_value size too large";
255 if (size != 0 && entry->e_value_inum == 0) {
256 u16 offs = le16_to_cpu(entry->e_value_offs);
260 * The value cannot overlap the names, and the value
261 * with padding cannot extend beyond 'end'. Check both
262 * the padded and unpadded sizes, since the size may
263 * overflow to 0 when adding padding.
265 if (offs > end - value_start) {
266 err_str = "e_value out of bounds";
269 value = value_start + offs;
270 if (value < (void *)e + sizeof(u32) ||
271 size > end - value ||
272 EXT4_XATTR_SIZE(size) > end - value) {
273 err_str = "overlapping e_value ";
277 entry = EXT4_XATTR_NEXT(entry);
280 set_buffer_verified(bh);
285 __ext4_error_inode(inode, function, line, 0, -err,
286 "corrupted xattr block %llu: %s",
287 (unsigned long long) bh->b_blocknr,
290 __ext4_error_inode(inode, function, line, 0, -err,
291 "corrupted in-inode xattr: %s", err_str);
296 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
297 const char *function, unsigned int line)
299 return check_xattrs(inode, bh, BFIRST(bh), bh->b_data + bh->b_size,
300 bh->b_data, function, line);
303 #define ext4_xattr_check_block(inode, bh) \
304 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
308 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
309 void *end, const char *function, unsigned int line)
311 return check_xattrs(inode, NULL, IFIRST(header), end, IFIRST(header),
315 #define xattr_check_inode(inode, header, end) \
316 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
319 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
320 void *end, int name_index, const char *name, int sorted)
322 struct ext4_xattr_entry *entry, *next;
328 name_len = strlen(name);
329 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
330 next = EXT4_XATTR_NEXT(entry);
331 if ((void *) next >= end) {
332 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
333 return -EFSCORRUPTED;
335 cmp = name_index - entry->e_name_index;
337 cmp = name_len - entry->e_name_len;
339 cmp = memcmp(name, entry->e_name, name_len);
340 if (cmp <= 0 && (sorted || cmp == 0))
344 return cmp ? -ENODATA : 0;
348 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
350 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
353 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
355 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
356 (u32) inode_peek_iversion_raw(ea_inode);
359 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
361 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
362 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
365 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
367 return (u32)ea_inode->i_atime.tv_sec;
370 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
372 ea_inode->i_atime.tv_sec = hash;
376 * Read the EA value from an inode.
378 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
380 int blocksize = 1 << ea_inode->i_blkbits;
381 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
382 int tail_size = (size % blocksize) ?: blocksize;
383 struct buffer_head *bhs_inline[8];
384 struct buffer_head **bhs = bhs_inline;
387 if (bh_count > ARRAY_SIZE(bhs_inline)) {
388 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
393 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
394 true /* wait */, bhs);
398 for (i = 0; i < bh_count; i++) {
399 /* There shouldn't be any holes in ea_inode. */
404 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
405 i < bh_count - 1 ? blocksize : tail_size);
409 for (i = 0; i < bh_count; i++)
412 if (bhs != bhs_inline)
417 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
419 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
420 u32 ea_inode_hash, struct inode **ea_inode)
426 * We have to check for this corruption early as otherwise
427 * iget_locked() could wait indefinitely for the state of our
430 if (parent->i_ino == ea_ino) {
431 ext4_error(parent->i_sb,
432 "Parent and EA inode have the same ino %lu", ea_ino);
433 return -EFSCORRUPTED;
436 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_EA_INODE);
438 err = PTR_ERR(inode);
439 ext4_error(parent->i_sb,
440 "error while reading EA inode %lu err=%d", ea_ino,
444 ext4_xattr_inode_set_class(inode);
447 * Check whether this is an old Lustre-style xattr inode. Lustre
448 * implementation does not have hash validation, rather it has a
449 * backpointer from ea_inode to the parent inode.
451 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
452 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
453 inode->i_generation == parent->i_generation) {
454 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
455 ext4_xattr_inode_set_ref(inode, 1);
458 inode->i_flags |= S_NOQUOTA;
466 /* Remove entry from mbcache when EA inode is getting evicted */
467 void ext4_evict_ea_inode(struct inode *inode)
469 struct mb_cache_entry *oe;
471 if (!EA_INODE_CACHE(inode))
473 /* Wait for entry to get unused so that we can remove it */
474 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
475 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
476 mb_cache_entry_wait_unused(oe);
477 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
482 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
483 struct ext4_xattr_entry *entry, void *buffer,
488 /* Verify stored hash matches calculated hash. */
489 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
490 if (hash != ext4_xattr_inode_get_hash(ea_inode))
491 return -EFSCORRUPTED;
494 __le32 e_hash, tmp_data;
496 /* Verify entry hash. */
497 tmp_data = cpu_to_le32(hash);
498 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
501 if (e_hash == entry->e_hash)
505 * Not good. Maybe the entry hash was calculated
506 * using the buggy signed char version?
508 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
510 /* Still no match - bad */
511 if (e_hash != entry->e_hash)
512 return -EFSCORRUPTED;
514 /* Let people know about old hash */
515 pr_warn_once("ext4: filesystem with signed xattr name hash");
521 * Read xattr value from the EA inode.
524 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
525 void *buffer, size_t size)
527 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
528 struct inode *ea_inode;
531 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
532 le32_to_cpu(entry->e_hash), &ea_inode);
538 if (i_size_read(ea_inode) != size) {
539 ext4_warning_inode(ea_inode,
540 "ea_inode file size=%llu entry size=%zu",
541 i_size_read(ea_inode), size);
546 err = ext4_xattr_inode_read(ea_inode, buffer, size);
550 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
551 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
554 ext4_warning_inode(ea_inode,
555 "EA inode hash validation failed");
560 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
561 ext4_xattr_inode_get_hash(ea_inode),
562 ea_inode->i_ino, true /* reusable */);
570 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
571 void *buffer, size_t buffer_size)
573 struct buffer_head *bh = NULL;
574 struct ext4_xattr_entry *entry;
578 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
580 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
581 name_index, name, buffer, (long)buffer_size);
583 if (!EXT4_I(inode)->i_file_acl)
585 ea_idebug(inode, "reading block %llu",
586 (unsigned long long)EXT4_I(inode)->i_file_acl);
587 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
590 ea_bdebug(bh, "b_count=%d, refcount=%d",
591 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
592 error = ext4_xattr_check_block(inode, bh);
595 ext4_xattr_block_cache_insert(ea_block_cache, bh);
597 end = bh->b_data + bh->b_size;
598 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
601 size = le32_to_cpu(entry->e_value_size);
603 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
606 if (size > buffer_size)
608 if (entry->e_value_inum) {
609 error = ext4_xattr_inode_get(inode, entry, buffer,
614 u16 offset = le16_to_cpu(entry->e_value_offs);
615 void *p = bh->b_data + offset;
617 if (unlikely(p + size > end))
619 memcpy(buffer, p, size);
630 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
631 void *buffer, size_t buffer_size)
633 struct ext4_xattr_ibody_header *header;
634 struct ext4_xattr_entry *entry;
635 struct ext4_inode *raw_inode;
636 struct ext4_iloc iloc;
641 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
643 error = ext4_get_inode_loc(inode, &iloc);
646 raw_inode = ext4_raw_inode(&iloc);
647 header = IHDR(inode, raw_inode);
648 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
649 error = xattr_check_inode(inode, header, end);
652 entry = IFIRST(header);
653 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
656 size = le32_to_cpu(entry->e_value_size);
658 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
661 if (size > buffer_size)
663 if (entry->e_value_inum) {
664 error = ext4_xattr_inode_get(inode, entry, buffer,
669 u16 offset = le16_to_cpu(entry->e_value_offs);
670 void *p = (void *)IFIRST(header) + offset;
672 if (unlikely(p + size > end))
674 memcpy(buffer, p, size);
687 * Copy an extended attribute into the buffer
688 * provided, or compute the buffer size required.
689 * Buffer is NULL to compute the size of the buffer required.
691 * Returns a negative error number on failure, or the number of bytes
692 * used / required on success.
695 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
696 void *buffer, size_t buffer_size)
700 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
703 if (strlen(name) > 255)
706 down_read(&EXT4_I(inode)->xattr_sem);
707 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
709 if (error == -ENODATA)
710 error = ext4_xattr_block_get(inode, name_index, name, buffer,
712 up_read(&EXT4_I(inode)->xattr_sem);
717 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
718 char *buffer, size_t buffer_size)
720 size_t rest = buffer_size;
722 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
725 prefix = ext4_xattr_prefix(entry->e_name_index, dentry);
727 size_t prefix_len = strlen(prefix);
728 size_t size = prefix_len + entry->e_name_len + 1;
733 memcpy(buffer, prefix, prefix_len);
734 buffer += prefix_len;
735 memcpy(buffer, entry->e_name, entry->e_name_len);
736 buffer += entry->e_name_len;
742 return buffer_size - rest; /* total size */
746 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
748 struct inode *inode = d_inode(dentry);
749 struct buffer_head *bh = NULL;
752 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
753 buffer, (long)buffer_size);
755 if (!EXT4_I(inode)->i_file_acl)
757 ea_idebug(inode, "reading block %llu",
758 (unsigned long long)EXT4_I(inode)->i_file_acl);
759 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
762 ea_bdebug(bh, "b_count=%d, refcount=%d",
763 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
764 error = ext4_xattr_check_block(inode, bh);
767 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
768 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
776 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
778 struct inode *inode = d_inode(dentry);
779 struct ext4_xattr_ibody_header *header;
780 struct ext4_inode *raw_inode;
781 struct ext4_iloc iloc;
785 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
787 error = ext4_get_inode_loc(inode, &iloc);
790 raw_inode = ext4_raw_inode(&iloc);
791 header = IHDR(inode, raw_inode);
792 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
793 error = xattr_check_inode(inode, header, end);
796 error = ext4_xattr_list_entries(dentry, IFIRST(header),
797 buffer, buffer_size);
805 * Inode operation listxattr()
807 * d_inode(dentry)->i_rwsem: don't care
809 * Copy a list of attribute names into the buffer
810 * provided, or compute the buffer size required.
811 * Buffer is NULL to compute the size of the buffer required.
813 * Returns a negative error number on failure, or the number of bytes
814 * used / required on success.
817 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
821 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
822 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
829 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
834 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
839 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
842 static void ext4_xattr_update_super_block(handle_t *handle,
843 struct super_block *sb)
845 if (ext4_has_feature_xattr(sb))
848 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
849 if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
850 EXT4_JTR_NONE) == 0) {
851 lock_buffer(EXT4_SB(sb)->s_sbh);
852 ext4_set_feature_xattr(sb);
853 ext4_superblock_csum_set(sb);
854 unlock_buffer(EXT4_SB(sb)->s_sbh);
855 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
859 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
861 struct ext4_iloc iloc = { .bh = NULL };
862 struct buffer_head *bh = NULL;
863 struct ext4_inode *raw_inode;
864 struct ext4_xattr_ibody_header *header;
865 struct ext4_xattr_entry *entry;
866 qsize_t ea_inode_refs = 0;
870 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
872 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
873 ret = ext4_get_inode_loc(inode, &iloc);
876 raw_inode = ext4_raw_inode(&iloc);
877 header = IHDR(inode, raw_inode);
878 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
879 ret = xattr_check_inode(inode, header, end);
883 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
884 entry = EXT4_XATTR_NEXT(entry))
885 if (entry->e_value_inum)
889 if (EXT4_I(inode)->i_file_acl) {
890 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
897 ret = ext4_xattr_check_block(inode, bh);
901 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
902 entry = EXT4_XATTR_NEXT(entry))
903 if (entry->e_value_inum)
906 *usage = ea_inode_refs + 1;
914 static inline size_t round_up_cluster(struct inode *inode, size_t length)
916 struct super_block *sb = inode->i_sb;
917 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
919 size_t mask = ~(cluster_size - 1);
921 return (length + cluster_size - 1) & mask;
924 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
928 err = dquot_alloc_inode(inode);
931 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
933 dquot_free_inode(inode);
937 static void ext4_xattr_inode_free_quota(struct inode *parent,
938 struct inode *ea_inode,
942 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
944 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
945 dquot_free_inode(parent);
948 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
949 struct buffer_head *block_bh, size_t value_len,
956 * 1) Owner inode update
957 * 2) Ref count update on old xattr block
959 * 4) block bitmap update for new xattr block
960 * 5) group descriptor for new xattr block
961 * 6) block bitmap update for old xattr block
962 * 7) group descriptor for old block
964 * 6 & 7 can happen if we have two racing threads T_a and T_b
965 * which are each trying to set an xattr on inodes I_a and I_b
966 * which were both initially sharing an xattr block.
971 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
974 * In case of inline data, we may push out the data to a block,
975 * so we need to reserve credits for this eventuality
977 if (inode && ext4_has_inline_data(inode))
978 credits += ext4_writepage_trans_blocks(inode) + 1;
980 /* We are done if ea_inode feature is not enabled. */
981 if (!ext4_has_feature_ea_inode(sb))
984 /* New ea_inode, inode map, block bitmap, group descriptor. */
988 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
990 /* Indirection block or one level of extent tree. */
993 /* Block bitmap and group descriptor updates for each block. */
994 credits += blocks * 2;
996 /* Blocks themselves. */
1000 /* Dereference ea_inode holding old xattr value.
1001 * Old ea_inode, inode map, block bitmap, group descriptor.
1005 /* Data blocks for old ea_inode. */
1006 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
1008 /* Indirection block or one level of extent tree for old
1013 /* Block bitmap and group descriptor updates for each block. */
1014 credits += blocks * 2;
1017 /* We may need to clone the existing xattr block in which case we need
1018 * to increment ref counts for existing ea_inodes referenced by it.
1021 struct ext4_xattr_entry *entry = BFIRST(block_bh);
1023 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1024 if (entry->e_value_inum)
1025 /* Ref count update on ea_inode. */
1031 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1034 struct ext4_iloc iloc;
1038 inode_lock(ea_inode);
1040 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1044 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1045 ref_count += ref_change;
1046 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1048 if (ref_change > 0) {
1049 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1050 ea_inode->i_ino, ref_count);
1052 if (ref_count == 1) {
1053 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1054 ea_inode->i_ino, ea_inode->i_nlink);
1056 set_nlink(ea_inode, 1);
1057 ext4_orphan_del(handle, ea_inode);
1060 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1061 ea_inode->i_ino, ref_count);
1063 if (ref_count == 0) {
1064 WARN_ONCE(ea_inode->i_nlink != 1,
1065 "EA inode %lu i_nlink=%u",
1066 ea_inode->i_ino, ea_inode->i_nlink);
1068 clear_nlink(ea_inode);
1069 ext4_orphan_add(handle, ea_inode);
1073 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1075 ext4_warning_inode(ea_inode,
1076 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1078 inode_unlock(ea_inode);
1082 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1084 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1087 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1089 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1092 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1093 struct ext4_xattr_entry *first)
1095 struct inode *ea_inode;
1096 struct ext4_xattr_entry *entry;
1097 struct ext4_xattr_entry *failed_entry;
1098 unsigned int ea_ino;
1101 for (entry = first; !IS_LAST_ENTRY(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),
1111 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1113 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1123 failed_entry = entry;
1125 for (entry = first; entry != failed_entry;
1126 entry = EXT4_XATTR_NEXT(entry)) {
1127 if (!entry->e_value_inum)
1129 ea_ino = le32_to_cpu(entry->e_value_inum);
1130 err = ext4_xattr_inode_iget(parent, ea_ino,
1131 le32_to_cpu(entry->e_hash),
1134 ext4_warning(parent->i_sb,
1135 "cleanup ea_ino %u iget error %d", ea_ino,
1139 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1141 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1148 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1149 struct buffer_head *bh, bool block_csum, bool dirty)
1155 ext4_xattr_block_csum_set(inode, bh);
1156 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1158 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1167 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1168 struct buffer_head *bh,
1169 struct ext4_xattr_entry *first, bool block_csum,
1170 struct ext4_xattr_inode_array **ea_inode_array,
1171 int extra_credits, bool skip_quota)
1173 struct inode *ea_inode;
1174 struct ext4_xattr_entry *entry;
1176 unsigned int ea_ino;
1180 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1181 credits = 2 + extra_credits;
1183 for (entry = first; !IS_LAST_ENTRY(entry);
1184 entry = EXT4_XATTR_NEXT(entry)) {
1185 if (!entry->e_value_inum)
1187 ea_ino = le32_to_cpu(entry->e_value_inum);
1188 err = ext4_xattr_inode_iget(parent, ea_ino,
1189 le32_to_cpu(entry->e_hash),
1194 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1196 ext4_warning_inode(ea_inode,
1197 "Expand inode array err=%d", err);
1202 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1203 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1204 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1207 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1212 err = ext4_journal_get_write_access(handle,
1213 parent->i_sb, bh, EXT4_JTR_NONE);
1215 ext4_warning_inode(ea_inode,
1216 "Re-get write access err=%d",
1222 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1224 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1230 ext4_xattr_inode_free_quota(parent, ea_inode,
1231 le32_to_cpu(entry->e_value_size));
1234 * Forget about ea_inode within the same transaction that
1235 * decrements the ref count. This avoids duplicate decrements in
1236 * case the rest of the work spills over to subsequent
1239 entry->e_value_inum = 0;
1240 entry->e_value_size = 0;
1247 * Note that we are deliberately skipping csum calculation for
1248 * the final update because we do not expect any journal
1249 * restarts until xattr block is freed.
1252 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1254 ext4_warning_inode(parent,
1255 "handle dirty metadata err=%d", err);
1260 * Release the xattr block BH: If the reference count is > 1, decrement it;
1261 * otherwise free the block.
1264 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1265 struct buffer_head *bh,
1266 struct ext4_xattr_inode_array **ea_inode_array,
1269 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1273 BUFFER_TRACE(bh, "get_write_access");
1274 error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1281 hash = le32_to_cpu(BHDR(bh)->h_hash);
1282 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1284 ea_bdebug(bh, "refcount now=0; freeing");
1286 * This must happen under buffer lock for
1287 * ext4_xattr_block_set() to reliably detect freed block
1289 if (ea_block_cache) {
1290 struct mb_cache_entry *oe;
1292 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1296 mb_cache_entry_wait_unused(oe);
1297 mb_cache_entry_put(ea_block_cache, oe);
1304 if (ext4_has_feature_ea_inode(inode->i_sb))
1305 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1307 true /* block_csum */,
1310 true /* skip_quota */);
1311 ext4_free_blocks(handle, inode, bh, 0, 1,
1312 EXT4_FREE_BLOCKS_METADATA |
1313 EXT4_FREE_BLOCKS_FORGET);
1316 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1317 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1318 struct mb_cache_entry *ce;
1320 if (ea_block_cache) {
1321 ce = mb_cache_entry_get(ea_block_cache, hash,
1324 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1325 mb_cache_entry_put(ea_block_cache, ce);
1330 ext4_xattr_block_csum_set(inode, bh);
1332 * Beware of this ugliness: Releasing of xattr block references
1333 * from different inodes can race and so we have to protect
1334 * from a race where someone else frees the block (and releases
1335 * its journal_head) before we are done dirtying the buffer. In
1336 * nojournal mode this race is harmless and we actually cannot
1337 * call ext4_handle_dirty_metadata() with locked buffer as
1338 * that function can call sync_dirty_buffer() so for that case
1339 * we handle the dirtying after unlocking the buffer.
1341 if (ext4_handle_valid(handle))
1342 error = ext4_handle_dirty_metadata(handle, inode, bh);
1344 if (!ext4_handle_valid(handle))
1345 error = ext4_handle_dirty_metadata(handle, inode, bh);
1347 ext4_handle_sync(handle);
1348 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1349 ea_bdebug(bh, "refcount now=%d; releasing",
1350 le32_to_cpu(BHDR(bh)->h_refcount));
1353 ext4_std_error(inode->i_sb, error);
1358 * Find the available free space for EAs. This also returns the total number of
1359 * bytes used by EA entries.
1361 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1362 size_t *min_offs, void *base, int *total)
1364 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1365 if (!last->e_value_inum && last->e_value_size) {
1366 size_t offs = le16_to_cpu(last->e_value_offs);
1367 if (offs < *min_offs)
1371 *total += EXT4_XATTR_LEN(last->e_name_len);
1373 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1377 * Write the value of the EA in an inode.
1379 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1380 const void *buf, int bufsize)
1382 struct buffer_head *bh = NULL;
1383 unsigned long block = 0;
1384 int blocksize = ea_inode->i_sb->s_blocksize;
1385 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1386 int csize, wsize = 0;
1387 int ret = 0, ret2 = 0;
1391 while (ret >= 0 && ret < max_blocks) {
1392 struct ext4_map_blocks map;
1393 map.m_lblk = block += ret;
1394 map.m_len = max_blocks -= ret;
1396 ret = ext4_map_blocks(handle, ea_inode, &map,
1397 EXT4_GET_BLOCKS_CREATE);
1399 ext4_mark_inode_dirty(handle, ea_inode);
1400 if (ret == -ENOSPC &&
1401 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1413 while (wsize < bufsize) {
1415 csize = (bufsize - wsize) > blocksize ? blocksize :
1417 bh = ext4_getblk(handle, ea_inode, block, 0);
1422 EXT4_ERROR_INODE(ea_inode,
1423 "ext4_getblk() return bh = NULL");
1424 return -EFSCORRUPTED;
1426 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1431 memcpy(bh->b_data, buf, csize);
1432 set_buffer_uptodate(bh);
1433 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1440 inode_lock(ea_inode);
1441 i_size_write(ea_inode, wsize);
1442 ext4_update_i_disksize(ea_inode, wsize);
1443 inode_unlock(ea_inode);
1445 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1446 if (unlikely(ret2 && !ret))
1456 * Create an inode to store the value of a large EA.
1458 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1459 struct inode *inode, u32 hash)
1461 struct inode *ea_inode = NULL;
1462 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1465 if (inode->i_sb->s_root == NULL) {
1466 ext4_warning(inode->i_sb,
1467 "refuse to create EA inode when umounting");
1469 return ERR_PTR(-EINVAL);
1473 * Let the next inode be the goal, so we try and allocate the EA inode
1474 * in the same group, or nearby one.
1476 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1477 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1479 if (!IS_ERR(ea_inode)) {
1480 ea_inode->i_op = &ext4_file_inode_operations;
1481 ea_inode->i_fop = &ext4_file_operations;
1482 ext4_set_aops(ea_inode);
1483 ext4_xattr_inode_set_class(ea_inode);
1484 unlock_new_inode(ea_inode);
1485 ext4_xattr_inode_set_ref(ea_inode, 1);
1486 ext4_xattr_inode_set_hash(ea_inode, hash);
1487 err = ext4_mark_inode_dirty(handle, ea_inode);
1489 err = ext4_inode_attach_jinode(ea_inode);
1491 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1492 ext4_warning_inode(ea_inode,
1493 "cleanup dec ref error %d", err);
1495 return ERR_PTR(err);
1499 * Xattr inodes are shared therefore quota charging is performed
1500 * at a higher level.
1502 dquot_free_inode(ea_inode);
1503 dquot_drop(ea_inode);
1504 inode_lock(ea_inode);
1505 ea_inode->i_flags |= S_NOQUOTA;
1506 inode_unlock(ea_inode);
1512 static struct inode *
1513 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1514 size_t value_len, u32 hash)
1516 struct inode *ea_inode;
1517 struct mb_cache_entry *ce;
1518 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1521 if (!ea_inode_cache)
1524 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1528 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1529 !(current->flags & PF_MEMALLOC_NOFS));
1531 ea_data = kvmalloc(value_len, GFP_KERNEL);
1533 mb_cache_entry_put(ea_inode_cache, ce);
1538 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1539 EXT4_IGET_EA_INODE);
1540 if (IS_ERR(ea_inode))
1542 if (i_size_read(ea_inode) == value_len &&
1543 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1544 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1546 !memcmp(value, ea_data, value_len)) {
1547 mb_cache_entry_touch(ea_inode_cache, ce);
1548 mb_cache_entry_put(ea_inode_cache, ce);
1554 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1561 * Add value of the EA in an inode.
1563 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1564 const void *value, size_t value_len,
1565 struct inode **ret_inode)
1567 struct inode *ea_inode;
1571 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1572 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1574 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1580 *ret_inode = ea_inode;
1584 /* Create an inode for the EA value */
1585 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1586 if (IS_ERR(ea_inode))
1587 return PTR_ERR(ea_inode);
1589 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1591 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1592 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1597 if (EA_INODE_CACHE(inode))
1598 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1599 ea_inode->i_ino, true /* reusable */);
1601 *ret_inode = ea_inode;
1606 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1607 * feature is enabled.
1609 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1611 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1612 struct ext4_xattr_search *s,
1613 handle_t *handle, struct inode *inode,
1616 struct ext4_xattr_entry *last, *next;
1617 struct ext4_xattr_entry *here = s->here;
1618 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1619 int in_inode = i->in_inode;
1620 struct inode *old_ea_inode = NULL;
1621 struct inode *new_ea_inode = NULL;
1622 size_t old_size, new_size;
1625 /* Space used by old and new values. */
1626 old_size = (!s->not_found && !here->e_value_inum) ?
1627 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1628 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1631 * Optimization for the simple case when old and new values have the
1632 * same padded sizes. Not applicable if external inodes are involved.
1634 if (new_size && new_size == old_size) {
1635 size_t offs = le16_to_cpu(here->e_value_offs);
1636 void *val = s->base + offs;
1638 here->e_value_size = cpu_to_le32(i->value_len);
1639 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1640 memset(val, 0, new_size);
1642 memcpy(val, i->value, i->value_len);
1643 /* Clear padding bytes. */
1644 memset(val + i->value_len, 0, new_size - i->value_len);
1649 /* Compute min_offs and last. */
1651 for (; !IS_LAST_ENTRY(last); last = next) {
1652 next = EXT4_XATTR_NEXT(last);
1653 if ((void *)next >= s->end) {
1654 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1655 ret = -EFSCORRUPTED;
1658 if (!last->e_value_inum && last->e_value_size) {
1659 size_t offs = le16_to_cpu(last->e_value_offs);
1660 if (offs < min_offs)
1665 /* Check whether we have enough space. */
1669 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1671 free += EXT4_XATTR_LEN(name_len) + old_size;
1673 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1679 * If storing the value in an external inode is an option,
1680 * reserve space for xattr entries/names in the external
1681 * attribute block so that a long value does not occupy the
1682 * whole space and prevent further entries being added.
1684 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1685 new_size && is_block &&
1686 (min_offs + old_size - new_size) <
1687 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1694 * Getting access to old and new ea inodes is subject to failures.
1695 * Finish that work before doing any modifications to the xattr data.
1697 if (!s->not_found && here->e_value_inum) {
1698 ret = ext4_xattr_inode_iget(inode,
1699 le32_to_cpu(here->e_value_inum),
1700 le32_to_cpu(here->e_hash),
1703 old_ea_inode = NULL;
1707 if (i->value && in_inode) {
1708 WARN_ON_ONCE(!i->value_len);
1710 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1714 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1718 new_ea_inode = NULL;
1719 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1725 /* We are ready to release ref count on the old_ea_inode. */
1726 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1728 /* Release newly required ref count on new_ea_inode. */
1732 err = ext4_xattr_inode_dec_ref(handle,
1735 ext4_warning_inode(new_ea_inode,
1736 "dec ref new_ea_inode err=%d",
1738 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1744 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1745 le32_to_cpu(here->e_value_size));
1748 /* No failures allowed past this point. */
1750 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1751 /* Remove the old value. */
1752 void *first_val = s->base + min_offs;
1753 size_t offs = le16_to_cpu(here->e_value_offs);
1754 void *val = s->base + offs;
1756 memmove(first_val + old_size, first_val, val - first_val);
1757 memset(first_val, 0, old_size);
1758 min_offs += old_size;
1760 /* Adjust all value offsets. */
1762 while (!IS_LAST_ENTRY(last)) {
1763 size_t o = le16_to_cpu(last->e_value_offs);
1765 if (!last->e_value_inum &&
1766 last->e_value_size && o < offs)
1767 last->e_value_offs = cpu_to_le16(o + old_size);
1768 last = EXT4_XATTR_NEXT(last);
1773 /* Remove old name. */
1774 size_t size = EXT4_XATTR_LEN(name_len);
1776 last = ENTRY((void *)last - size);
1777 memmove(here, (void *)here + size,
1778 (void *)last - (void *)here + sizeof(__u32));
1779 memset(last, 0, size);
1780 } else if (s->not_found) {
1781 /* Insert new name. */
1782 size_t size = EXT4_XATTR_LEN(name_len);
1783 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1785 memmove((void *)here + size, here, rest);
1786 memset(here, 0, size);
1787 here->e_name_index = i->name_index;
1788 here->e_name_len = name_len;
1789 memcpy(here->e_name, i->name, name_len);
1791 /* This is an update, reset value info. */
1792 here->e_value_inum = 0;
1793 here->e_value_offs = 0;
1794 here->e_value_size = 0;
1798 /* Insert new value. */
1800 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1801 } else if (i->value_len) {
1802 void *val = s->base + min_offs - new_size;
1804 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1805 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1806 memset(val, 0, new_size);
1808 memcpy(val, i->value, i->value_len);
1809 /* Clear padding bytes. */
1810 memset(val + i->value_len, 0,
1811 new_size - i->value_len);
1814 here->e_value_size = cpu_to_le32(i->value_len);
1821 /* Entry hash calculation. */
1826 * Feed crc32c hash instead of the raw value for entry
1827 * hash calculation. This is to avoid walking
1828 * potentially long value buffer again.
1830 crc32c_hash = cpu_to_le32(
1831 ext4_xattr_inode_get_hash(new_ea_inode));
1832 hash = ext4_xattr_hash_entry(here->e_name,
1835 } else if (is_block) {
1836 __le32 *value = s->base + le16_to_cpu(
1837 here->e_value_offs);
1839 hash = ext4_xattr_hash_entry(here->e_name,
1840 here->e_name_len, value,
1843 here->e_hash = hash;
1847 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1856 struct ext4_xattr_block_find {
1857 struct ext4_xattr_search s;
1858 struct buffer_head *bh;
1862 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1863 struct ext4_xattr_block_find *bs)
1865 struct super_block *sb = inode->i_sb;
1868 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1869 i->name_index, i->name, i->value, (long)i->value_len);
1871 if (EXT4_I(inode)->i_file_acl) {
1872 /* The inode already has an extended attribute block. */
1873 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1874 if (IS_ERR(bs->bh)) {
1875 error = PTR_ERR(bs->bh);
1879 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1880 atomic_read(&(bs->bh->b_count)),
1881 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1882 error = ext4_xattr_check_block(inode, bs->bh);
1885 /* Find the named attribute. */
1886 bs->s.base = BHDR(bs->bh);
1887 bs->s.first = BFIRST(bs->bh);
1888 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1889 bs->s.here = bs->s.first;
1890 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1891 i->name_index, i->name, 1);
1892 if (error && error != -ENODATA)
1894 bs->s.not_found = error;
1900 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1901 struct ext4_xattr_info *i,
1902 struct ext4_xattr_block_find *bs)
1904 struct super_block *sb = inode->i_sb;
1905 struct buffer_head *new_bh = NULL;
1906 struct ext4_xattr_search s_copy = bs->s;
1907 struct ext4_xattr_search *s = &s_copy;
1908 struct mb_cache_entry *ce = NULL;
1910 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1911 struct inode *ea_inode = NULL, *tmp_inode;
1912 size_t old_ea_inode_quota = 0;
1913 unsigned int ea_ino;
1916 #define header(x) ((struct ext4_xattr_header *)(x))
1919 int offset = (char *)s->here - bs->bh->b_data;
1921 BUFFER_TRACE(bs->bh, "get_write_access");
1922 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1926 lock_buffer(bs->bh);
1928 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1929 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1932 * This must happen under buffer lock for
1933 * ext4_xattr_block_set() to reliably detect modified
1936 if (ea_block_cache) {
1937 struct mb_cache_entry *oe;
1939 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1940 hash, bs->bh->b_blocknr);
1943 * Xattr block is getting reused. Leave
1946 mb_cache_entry_put(ea_block_cache, oe);
1950 ea_bdebug(bs->bh, "modifying in-place");
1951 error = ext4_xattr_set_entry(i, s, handle, inode,
1952 true /* is_block */);
1953 ext4_xattr_block_csum_set(inode, bs->bh);
1954 unlock_buffer(bs->bh);
1955 if (error == -EFSCORRUPTED)
1958 error = ext4_handle_dirty_metadata(handle,
1966 unlock_buffer(bs->bh);
1967 ea_bdebug(bs->bh, "cloning");
1968 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1970 if (s->base == NULL)
1972 s->first = ENTRY(header(s->base)+1);
1973 header(s->base)->h_refcount = cpu_to_le32(1);
1974 s->here = ENTRY(s->base + offset);
1975 s->end = s->base + bs->bh->b_size;
1978 * If existing entry points to an xattr inode, we need
1979 * to prevent ext4_xattr_set_entry() from decrementing
1980 * ref count on it because the reference belongs to the
1981 * original block. In this case, make the entry look
1982 * like it has an empty value.
1984 if (!s->not_found && s->here->e_value_inum) {
1985 ea_ino = le32_to_cpu(s->here->e_value_inum);
1986 error = ext4_xattr_inode_iget(inode, ea_ino,
1987 le32_to_cpu(s->here->e_hash),
1992 if (!ext4_test_inode_state(tmp_inode,
1993 EXT4_STATE_LUSTRE_EA_INODE)) {
1995 * Defer quota free call for previous
1996 * inode until success is guaranteed.
1998 old_ea_inode_quota = le32_to_cpu(
1999 s->here->e_value_size);
2003 s->here->e_value_inum = 0;
2004 s->here->e_value_size = 0;
2007 /* Allocate a buffer where we construct the new block. */
2008 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
2010 if (s->base == NULL)
2012 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2013 header(s->base)->h_blocks = cpu_to_le32(1);
2014 header(s->base)->h_refcount = cpu_to_le32(1);
2015 s->first = ENTRY(header(s->base)+1);
2016 s->here = ENTRY(header(s->base)+1);
2017 s->end = s->base + sb->s_blocksize;
2020 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
2021 if (error == -EFSCORRUPTED)
2026 if (i->value && s->here->e_value_inum) {
2028 * A ref count on ea_inode has been taken as part of the call to
2029 * ext4_xattr_set_entry() above. We would like to drop this
2030 * extra ref but we have to wait until the xattr block is
2031 * initialized and has its own ref count on the ea_inode.
2033 ea_ino = le32_to_cpu(s->here->e_value_inum);
2034 error = ext4_xattr_inode_iget(inode, ea_ino,
2035 le32_to_cpu(s->here->e_hash),
2044 if (!IS_LAST_ENTRY(s->first)) {
2045 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2048 /* We found an identical block in the cache. */
2049 if (new_bh == bs->bh)
2050 ea_bdebug(new_bh, "keeping");
2054 WARN_ON_ONCE(dquot_initialize_needed(inode));
2056 /* The old block is released after updating
2058 error = dquot_alloc_block(inode,
2059 EXT4_C2B(EXT4_SB(sb), 1));
2062 BUFFER_TRACE(new_bh, "get_write_access");
2063 error = ext4_journal_get_write_access(
2068 lock_buffer(new_bh);
2070 * We have to be careful about races with
2071 * adding references to xattr block. Once we
2072 * hold buffer lock xattr block's state is
2073 * stable so we can check the additional
2076 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2077 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2079 * Undo everything and check mbcache
2082 unlock_buffer(new_bh);
2083 dquot_free_block(inode,
2084 EXT4_C2B(EXT4_SB(sb),
2087 mb_cache_entry_put(ea_block_cache, ce);
2092 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2093 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2094 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2095 ea_bdebug(new_bh, "reusing; refcount now=%d",
2097 ext4_xattr_block_csum_set(inode, new_bh);
2098 unlock_buffer(new_bh);
2099 error = ext4_handle_dirty_metadata(handle,
2105 mb_cache_entry_touch(ea_block_cache, ce);
2106 mb_cache_entry_put(ea_block_cache, ce);
2108 } else if (bs->bh && s->base == bs->bh->b_data) {
2109 /* We were modifying this block in-place. */
2110 ea_bdebug(bs->bh, "keeping this block");
2111 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2115 /* We need to allocate a new block */
2116 ext4_fsblk_t goal, block;
2118 WARN_ON_ONCE(dquot_initialize_needed(inode));
2120 goal = ext4_group_first_block_no(sb,
2121 EXT4_I(inode)->i_block_group);
2122 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2127 ea_idebug(inode, "creating block %llu",
2128 (unsigned long long)block);
2130 new_bh = sb_getblk(sb, block);
2131 if (unlikely(!new_bh)) {
2134 ext4_free_blocks(handle, inode, NULL, block, 1,
2135 EXT4_FREE_BLOCKS_METADATA);
2138 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2139 ENTRY(header(s->base)+1));
2143 /* Drop the extra ref on ea_inode. */
2144 error = ext4_xattr_inode_dec_ref(handle,
2147 ext4_warning_inode(ea_inode,
2154 lock_buffer(new_bh);
2155 error = ext4_journal_get_create_access(handle, sb,
2156 new_bh, EXT4_JTR_NONE);
2158 unlock_buffer(new_bh);
2162 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2163 ext4_xattr_block_csum_set(inode, new_bh);
2164 set_buffer_uptodate(new_bh);
2165 unlock_buffer(new_bh);
2166 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2167 error = ext4_handle_dirty_metadata(handle, inode,
2174 if (old_ea_inode_quota)
2175 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2177 /* Update the inode. */
2178 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2180 /* Drop the previous xattr block. */
2181 if (bs->bh && bs->bh != new_bh) {
2182 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2184 ext4_xattr_release_block(handle, inode, bs->bh,
2186 0 /* extra_credits */);
2187 ext4_xattr_inode_array_free(ea_inode_array);
2195 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2197 ext4_warning_inode(ea_inode, "dec ref error=%d",
2200 /* If there was an error, revert the quota charge. */
2202 ext4_xattr_inode_free_quota(inode, ea_inode,
2203 i_size_read(ea_inode));
2207 mb_cache_entry_put(ea_block_cache, ce);
2209 if (!(bs->bh && s->base == bs->bh->b_data))
2215 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2219 EXT4_ERROR_INODE(inode, "bad block %llu",
2220 EXT4_I(inode)->i_file_acl);
2226 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2227 struct ext4_xattr_ibody_find *is)
2229 struct ext4_xattr_ibody_header *header;
2230 struct ext4_inode *raw_inode;
2233 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2236 raw_inode = ext4_raw_inode(&is->iloc);
2237 header = IHDR(inode, raw_inode);
2238 is->s.base = is->s.first = IFIRST(header);
2239 is->s.here = is->s.first;
2240 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2241 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2242 error = xattr_check_inode(inode, header, is->s.end);
2245 /* Find the named attribute. */
2246 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2247 i->name_index, i->name, 0);
2248 if (error && error != -ENODATA)
2250 is->s.not_found = error;
2255 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2256 struct ext4_xattr_info *i,
2257 struct ext4_xattr_ibody_find *is)
2259 struct ext4_xattr_ibody_header *header;
2260 struct ext4_xattr_search *s = &is->s;
2263 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2266 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2269 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2270 if (!IS_LAST_ENTRY(s->first)) {
2271 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2272 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2274 header->h_magic = cpu_to_le32(0);
2275 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2280 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2281 struct ext4_xattr_info *i)
2285 /* When e_value_inum is set the value is stored externally. */
2286 if (s->here->e_value_inum)
2288 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2290 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2291 return !memcmp(value, i->value, i->value_len);
2294 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2296 struct buffer_head *bh;
2299 if (!EXT4_I(inode)->i_file_acl)
2301 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2304 error = ext4_xattr_check_block(inode, bh);
2307 return ERR_PTR(error);
2313 * ext4_xattr_set_handle()
2315 * Create, replace or remove an extended attribute for this inode. Value
2316 * is NULL to remove an existing extended attribute, and non-NULL to
2317 * either replace an existing extended attribute, or create a new extended
2318 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2319 * specify that an extended attribute must exist and must not exist
2320 * previous to the call, respectively.
2322 * Returns 0, or a negative error number on failure.
2325 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2326 const char *name, const void *value, size_t value_len,
2329 struct ext4_xattr_info i = {
2330 .name_index = name_index,
2333 .value_len = value_len,
2336 struct ext4_xattr_ibody_find is = {
2337 .s = { .not_found = -ENODATA, },
2339 struct ext4_xattr_block_find bs = {
2340 .s = { .not_found = -ENODATA, },
2347 if (strlen(name) > 255)
2350 ext4_write_lock_xattr(inode, &no_expand);
2352 /* Check journal credits under write lock. */
2353 if (ext4_handle_valid(handle)) {
2354 struct buffer_head *bh;
2357 bh = ext4_xattr_get_block(inode);
2359 error = PTR_ERR(bh);
2363 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2365 flags & XATTR_CREATE);
2368 if (jbd2_handle_buffer_credits(handle) < credits) {
2372 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2375 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2379 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2380 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2381 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2382 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2385 error = ext4_xattr_ibody_find(inode, &i, &is);
2389 error = ext4_xattr_block_find(inode, &i, &bs);
2392 if (is.s.not_found && bs.s.not_found) {
2394 if (flags & XATTR_REPLACE)
2401 if (flags & XATTR_CREATE)
2406 if (!is.s.not_found)
2407 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2408 else if (!bs.s.not_found)
2409 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2412 /* Xattr value did not change? Save us some work and bail out */
2413 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2415 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2418 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2419 (EXT4_XATTR_SIZE(i.value_len) >
2420 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2423 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2424 if (!error && !bs.s.not_found) {
2426 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2427 } else if (error == -ENOSPC) {
2428 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2431 error = ext4_xattr_block_find(inode, &i, &bs);
2435 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2436 if (!error && !is.s.not_found) {
2438 error = ext4_xattr_ibody_set(handle, inode, &i,
2440 } else if (error == -ENOSPC) {
2442 * Xattr does not fit in the block, store at
2443 * external inode if possible.
2445 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2446 i.value_len && !i.in_inode) {
2454 ext4_xattr_update_super_block(handle, inode->i_sb);
2455 inode->i_ctime = current_time(inode);
2456 inode_inc_iversion(inode);
2459 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2461 * The bh is consumed by ext4_mark_iloc_dirty, even with
2466 ext4_handle_sync(handle);
2468 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2473 ext4_write_unlock_xattr(inode, &no_expand);
2477 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2478 bool is_create, int *credits)
2480 struct buffer_head *bh;
2485 if (!EXT4_SB(inode->i_sb)->s_journal)
2488 down_read(&EXT4_I(inode)->xattr_sem);
2490 bh = ext4_xattr_get_block(inode);
2494 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2495 value_len, is_create);
2500 up_read(&EXT4_I(inode)->xattr_sem);
2507 * Like ext4_xattr_set_handle, but start from an inode. This extended
2508 * attribute modification is a filesystem transaction by itself.
2510 * Returns 0, or a negative error number on failure.
2513 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2514 const void *value, size_t value_len, int flags)
2517 struct super_block *sb = inode->i_sb;
2518 int error, retries = 0;
2521 error = dquot_initialize(inode);
2526 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2531 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2532 if (IS_ERR(handle)) {
2533 error = PTR_ERR(handle);
2537 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2538 value, value_len, flags);
2539 error2 = ext4_journal_stop(handle);
2540 if (error == -ENOSPC &&
2541 ext4_should_retry_alloc(sb, &retries))
2546 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2552 * Shift the EA entries in the inode to create space for the increased
2555 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2556 int value_offs_shift, void *to,
2557 void *from, size_t n)
2559 struct ext4_xattr_entry *last = entry;
2562 /* We always shift xattr headers further thus offsets get lower */
2563 BUG_ON(value_offs_shift > 0);
2565 /* Adjust the value offsets of the entries */
2566 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2567 if (!last->e_value_inum && last->e_value_size) {
2568 new_offs = le16_to_cpu(last->e_value_offs) +
2570 last->e_value_offs = cpu_to_le16(new_offs);
2573 /* Shift the entries by n bytes */
2574 memmove(to, from, n);
2578 * Move xattr pointed to by 'entry' from inode into external xattr block
2580 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2581 struct ext4_inode *raw_inode,
2582 struct ext4_xattr_entry *entry)
2584 struct ext4_xattr_ibody_find *is = NULL;
2585 struct ext4_xattr_block_find *bs = NULL;
2586 char *buffer = NULL, *b_entry_name = NULL;
2587 size_t value_size = le32_to_cpu(entry->e_value_size);
2588 struct ext4_xattr_info i = {
2591 .name_index = entry->e_name_index,
2592 .in_inode = !!entry->e_value_inum,
2594 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2595 int needs_kvfree = 0;
2598 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2599 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2600 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2601 if (!is || !bs || !b_entry_name) {
2606 is->s.not_found = -ENODATA;
2607 bs->s.not_found = -ENODATA;
2611 /* Save the entry name and the entry value */
2612 if (entry->e_value_inum) {
2613 buffer = kvmalloc(value_size, GFP_NOFS);
2619 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2623 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2624 buffer = (void *)IFIRST(header) + value_offs;
2627 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2628 b_entry_name[entry->e_name_len] = '\0';
2629 i.name = b_entry_name;
2631 error = ext4_get_inode_loc(inode, &is->iloc);
2635 error = ext4_xattr_ibody_find(inode, &i, is);
2640 i.value_len = value_size;
2641 error = ext4_xattr_block_find(inode, &i, bs);
2645 /* Move ea entry from the inode into the block */
2646 error = ext4_xattr_block_set(handle, inode, &i, bs);
2650 /* Remove the chosen entry from the inode */
2653 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2656 kfree(b_entry_name);
2657 if (needs_kvfree && buffer)
2660 brelse(is->iloc.bh);
2669 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2670 struct ext4_inode *raw_inode,
2671 int isize_diff, size_t ifree,
2672 size_t bfree, int *total_ino)
2674 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2675 struct ext4_xattr_entry *small_entry;
2676 struct ext4_xattr_entry *entry;
2677 struct ext4_xattr_entry *last;
2678 unsigned int entry_size; /* EA entry size */
2679 unsigned int total_size; /* EA entry size + value size */
2680 unsigned int min_total_size;
2683 while (isize_diff > ifree) {
2686 min_total_size = ~0U;
2687 last = IFIRST(header);
2688 /* Find the entry best suited to be pushed into EA block */
2689 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2690 /* never move system.data out of the inode */
2691 if ((last->e_name_len == 4) &&
2692 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2693 !memcmp(last->e_name, "data", 4))
2695 total_size = EXT4_XATTR_LEN(last->e_name_len);
2696 if (!last->e_value_inum)
2697 total_size += EXT4_XATTR_SIZE(
2698 le32_to_cpu(last->e_value_size));
2699 if (total_size <= bfree &&
2700 total_size < min_total_size) {
2701 if (total_size + ifree < isize_diff) {
2705 min_total_size = total_size;
2710 if (entry == NULL) {
2711 if (small_entry == NULL)
2713 entry = small_entry;
2716 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2717 total_size = entry_size;
2718 if (!entry->e_value_inum)
2719 total_size += EXT4_XATTR_SIZE(
2720 le32_to_cpu(entry->e_value_size));
2721 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2726 *total_ino -= entry_size;
2727 ifree += total_size;
2728 bfree -= total_size;
2735 * Expand an inode by new_extra_isize bytes when EAs are present.
2736 * Returns 0 on success or negative error number on failure.
2738 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2739 struct ext4_inode *raw_inode, handle_t *handle)
2741 struct ext4_xattr_ibody_header *header;
2742 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2743 static unsigned int mnt_count;
2745 size_t ifree, bfree;
2748 int error = 0, tried_min_extra_isize = 0;
2749 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2750 int isize_diff; /* How much do we need to grow i_extra_isize */
2753 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2754 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2757 header = IHDR(inode, raw_inode);
2760 * Check if enough free space is available in the inode to shift the
2761 * entries ahead by new_extra_isize.
2764 base = IFIRST(header);
2765 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2766 min_offs = end - base;
2767 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2769 error = xattr_check_inode(inode, header, end);
2773 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2774 if (ifree >= isize_diff)
2778 * Enough free space isn't available in the inode, check if
2779 * EA block can hold new_extra_isize bytes.
2781 if (EXT4_I(inode)->i_file_acl) {
2782 struct buffer_head *bh;
2784 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2786 error = PTR_ERR(bh);
2789 error = ext4_xattr_check_block(inode, bh);
2795 end = bh->b_data + bh->b_size;
2796 min_offs = end - base;
2797 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2800 if (bfree + ifree < isize_diff) {
2801 if (!tried_min_extra_isize && s_min_extra_isize) {
2802 tried_min_extra_isize++;
2803 new_extra_isize = s_min_extra_isize;
2810 bfree = inode->i_sb->s_blocksize;
2813 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2814 isize_diff, ifree, bfree,
2817 if (error == -ENOSPC && !tried_min_extra_isize &&
2818 s_min_extra_isize) {
2819 tried_min_extra_isize++;
2820 new_extra_isize = s_min_extra_isize;
2826 /* Adjust the offsets and shift the remaining entries ahead */
2827 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2828 - new_extra_isize, (void *)raw_inode +
2829 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2830 (void *)header, total_ino);
2831 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2833 if (ext4_has_inline_data(inode))
2834 error = ext4_find_inline_data_nolock(inode);
2837 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2838 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2840 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2845 #define EIA_INCR 16 /* must be 2^n */
2846 #define EIA_MASK (EIA_INCR - 1)
2848 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2849 * If @ea_inode_array is new or full it will be grown and the old
2850 * contents copied over.
2853 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2854 struct inode *inode)
2856 if (*ea_inode_array == NULL) {
2858 * Start with 15 inodes, so it fits into a power-of-two size.
2859 * If *ea_inode_array is NULL, this is essentially offsetof()
2862 kmalloc(offsetof(struct ext4_xattr_inode_array,
2865 if (*ea_inode_array == NULL)
2867 (*ea_inode_array)->count = 0;
2868 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2869 /* expand the array once all 15 + n * 16 slots are full */
2870 struct ext4_xattr_inode_array *new_array = NULL;
2871 int count = (*ea_inode_array)->count;
2873 /* if new_array is NULL, this is essentially offsetof() */
2874 new_array = kmalloc(
2875 offsetof(struct ext4_xattr_inode_array,
2876 inodes[count + EIA_INCR]),
2878 if (new_array == NULL)
2880 memcpy(new_array, *ea_inode_array,
2881 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2882 kfree(*ea_inode_array);
2883 *ea_inode_array = new_array;
2885 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2890 * ext4_xattr_delete_inode()
2892 * Free extended attribute resources associated with this inode. Traverse
2893 * all entries and decrement reference on any xattr inodes associated with this
2894 * inode. This is called immediately before an inode is freed. We have exclusive
2895 * access to the inode. If an orphan inode is deleted it will also release its
2896 * references on xattr block and xattr inodes.
2898 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2899 struct ext4_xattr_inode_array **ea_inode_array,
2902 struct buffer_head *bh = NULL;
2903 struct ext4_xattr_ibody_header *header;
2904 struct ext4_iloc iloc = { .bh = NULL };
2905 struct ext4_xattr_entry *entry;
2906 struct inode *ea_inode;
2909 error = ext4_journal_ensure_credits(handle, extra_credits,
2910 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2912 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2916 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2917 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2919 error = ext4_get_inode_loc(inode, &iloc);
2921 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2925 error = ext4_journal_get_write_access(handle, inode->i_sb,
2926 iloc.bh, EXT4_JTR_NONE);
2928 EXT4_ERROR_INODE(inode, "write access (error %d)",
2933 header = IHDR(inode, ext4_raw_inode(&iloc));
2934 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2935 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2937 false /* block_csum */,
2940 false /* skip_quota */);
2943 if (EXT4_I(inode)->i_file_acl) {
2944 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2946 error = PTR_ERR(bh);
2947 if (error == -EIO) {
2948 EXT4_ERROR_INODE_ERR(inode, EIO,
2949 "block %llu read error",
2950 EXT4_I(inode)->i_file_acl);
2955 error = ext4_xattr_check_block(inode, bh);
2959 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2960 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2961 entry = EXT4_XATTR_NEXT(entry)) {
2962 if (!entry->e_value_inum)
2964 error = ext4_xattr_inode_iget(inode,
2965 le32_to_cpu(entry->e_value_inum),
2966 le32_to_cpu(entry->e_hash),
2970 ext4_xattr_inode_free_quota(inode, ea_inode,
2971 le32_to_cpu(entry->e_value_size));
2977 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2980 * Update i_file_acl value in the same transaction that releases
2983 EXT4_I(inode)->i_file_acl = 0;
2984 error = ext4_mark_inode_dirty(handle, inode);
2986 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2990 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2999 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
3003 if (ea_inode_array == NULL)
3006 for (idx = 0; idx < ea_inode_array->count; ++idx)
3007 iput(ea_inode_array->inodes[idx]);
3008 kfree(ea_inode_array);
3012 * ext4_xattr_block_cache_insert()
3014 * Create a new entry in the extended attribute block cache, and insert
3015 * it unless such an entry is already in the cache.
3017 * Returns 0, or a negative error number on failure.
3020 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3021 struct buffer_head *bh)
3023 struct ext4_xattr_header *header = BHDR(bh);
3024 __u32 hash = le32_to_cpu(header->h_hash);
3025 int reusable = le32_to_cpu(header->h_refcount) <
3026 EXT4_XATTR_REFCOUNT_MAX;
3029 if (!ea_block_cache)
3031 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3032 bh->b_blocknr, reusable);
3034 if (error == -EBUSY)
3035 ea_bdebug(bh, "already in cache");
3037 ea_bdebug(bh, "inserting [%x]", (int)hash);
3043 * Compare two extended attribute blocks for equality.
3045 * Returns 0 if the blocks are equal, 1 if they differ, and
3046 * a negative error number on errors.
3049 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3050 struct ext4_xattr_header *header2)
3052 struct ext4_xattr_entry *entry1, *entry2;
3054 entry1 = ENTRY(header1+1);
3055 entry2 = ENTRY(header2+1);
3056 while (!IS_LAST_ENTRY(entry1)) {
3057 if (IS_LAST_ENTRY(entry2))
3059 if (entry1->e_hash != entry2->e_hash ||
3060 entry1->e_name_index != entry2->e_name_index ||
3061 entry1->e_name_len != entry2->e_name_len ||
3062 entry1->e_value_size != entry2->e_value_size ||
3063 entry1->e_value_inum != entry2->e_value_inum ||
3064 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3066 if (!entry1->e_value_inum &&
3067 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3068 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3069 le32_to_cpu(entry1->e_value_size)))
3072 entry1 = EXT4_XATTR_NEXT(entry1);
3073 entry2 = EXT4_XATTR_NEXT(entry2);
3075 if (!IS_LAST_ENTRY(entry2))
3081 * ext4_xattr_block_cache_find()
3083 * Find an identical extended attribute block.
3085 * Returns a pointer to the block found, or NULL if such a block was
3086 * not found or an error occurred.
3088 static struct buffer_head *
3089 ext4_xattr_block_cache_find(struct inode *inode,
3090 struct ext4_xattr_header *header,
3091 struct mb_cache_entry **pce)
3093 __u32 hash = le32_to_cpu(header->h_hash);
3094 struct mb_cache_entry *ce;
3095 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3097 if (!ea_block_cache)
3099 if (!header->h_hash)
3100 return NULL; /* never share */
3101 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3102 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3104 struct buffer_head *bh;
3106 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3108 if (PTR_ERR(bh) == -ENOMEM)
3111 EXT4_ERROR_INODE(inode, "block %lu read error",
3112 (unsigned long)ce->e_value);
3113 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3118 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3123 #define NAME_HASH_SHIFT 5
3124 #define VALUE_HASH_SHIFT 16
3127 * ext4_xattr_hash_entry()
3129 * Compute the hash of an extended attribute.
3131 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3136 while (name_len--) {
3137 hash = (hash << NAME_HASH_SHIFT) ^
3138 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3139 (unsigned char)*name++;
3141 while (value_count--) {
3142 hash = (hash << VALUE_HASH_SHIFT) ^
3143 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3144 le32_to_cpu(*value++);
3146 return cpu_to_le32(hash);
3150 * ext4_xattr_hash_entry_signed()
3152 * Compute the hash of an extended attribute incorrectly.
3154 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3158 while (name_len--) {
3159 hash = (hash << NAME_HASH_SHIFT) ^
3160 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3161 (signed char)*name++;
3163 while (value_count--) {
3164 hash = (hash << VALUE_HASH_SHIFT) ^
3165 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3166 le32_to_cpu(*value++);
3168 return cpu_to_le32(hash);
3171 #undef NAME_HASH_SHIFT
3172 #undef VALUE_HASH_SHIFT
3174 #define BLOCK_HASH_SHIFT 16
3177 * ext4_xattr_rehash()
3179 * Re-compute the extended attribute hash value after an entry has changed.
3181 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3183 struct ext4_xattr_entry *here;
3186 here = ENTRY(header+1);
3187 while (!IS_LAST_ENTRY(here)) {
3188 if (!here->e_hash) {
3189 /* Block is not shared if an entry's hash value == 0 */
3193 hash = (hash << BLOCK_HASH_SHIFT) ^
3194 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3195 le32_to_cpu(here->e_hash);
3196 here = EXT4_XATTR_NEXT(here);
3198 header->h_hash = cpu_to_le32(hash);
3201 #undef BLOCK_HASH_SHIFT
3203 #define HASH_BUCKET_BITS 10
3206 ext4_xattr_create_cache(void)
3208 return mb_cache_create(HASH_BUCKET_BITS);
3211 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3214 mb_cache_destroy(cache);