1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/xattr.c
5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9 * Extended attributes for symlinks and special files added per
10 * suggestion of Luka Renko <luka.renko@hermes.si>.
11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14 * and Andreas Gruenbacher <agruen@suse.de>.
18 * Extended attributes are stored directly in inodes (on file systems with
19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
20 * field contains the block number if an inode uses an additional block. All
21 * attributes must fit in the inode and one additional block. Blocks that
22 * contain the identical set of attributes may be shared among several inodes.
23 * Identical blocks are detected by keeping a cache of blocks that have
24 * recently been accessed.
26 * The attributes in inodes and on blocks have a different header; the entries
27 * are stored in the same format:
29 * +------------------+
32 * | entry 2 | | growing downwards
37 * | value 3 | | growing upwards
39 * +------------------+
41 * The header is followed by multiple entry descriptors. In disk blocks, the
42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
43 * attribute values are aligned to the end of the block in no specific order.
47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
48 * EA blocks are only changed if they are exclusive to an inode, so
49 * holding xattr_sem also means that nothing but the EA block's reference
50 * count can change. Multiple writers to the same block are synchronized
54 #include <linux/init.h>
56 #include <linux/slab.h>
57 #include <linux/mbcache.h>
58 #include <linux/quotaops.h>
59 #include <linux/iversion.h>
60 #include "ext4_jbd2.h"
65 #ifdef EXT4_XATTR_DEBUG
66 # define ea_idebug(inode, fmt, ...) \
67 printk(KERN_DEBUG "inode %s:%lu: " fmt "\n", \
68 inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
69 # define ea_bdebug(bh, fmt, ...) \
70 printk(KERN_DEBUG "block %pg:%lu: " fmt "\n", \
71 bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
73 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
74 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
77 static void ext4_xattr_block_cache_insert(struct mb_cache *,
78 struct buffer_head *);
79 static struct buffer_head *
80 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
81 struct mb_cache_entry **);
82 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
84 static void ext4_xattr_rehash(struct ext4_xattr_header *);
86 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
87 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
88 #ifdef CONFIG_EXT4_FS_POSIX_ACL
89 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
90 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
92 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
93 #ifdef CONFIG_EXT4_FS_SECURITY
94 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
98 const struct xattr_handler *ext4_xattr_handlers[] = {
99 &ext4_xattr_user_handler,
100 &ext4_xattr_trusted_handler,
101 #ifdef CONFIG_EXT4_FS_POSIX_ACL
102 &posix_acl_access_xattr_handler,
103 &posix_acl_default_xattr_handler,
105 #ifdef CONFIG_EXT4_FS_SECURITY
106 &ext4_xattr_security_handler,
111 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
112 inode->i_sb->s_fs_info)->s_ea_block_cache)
114 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
115 inode->i_sb->s_fs_info)->s_ea_inode_cache)
118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119 struct inode *inode);
121 #ifdef CONFIG_LOCKDEP
122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
124 struct ext4_inode_info *ei = EXT4_I(ea_inode);
126 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
127 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
128 lockdep_set_subclass(&ei->i_data_sem, I_DATA_SEM_EA);
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,
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,
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)
392 * We have to check for this corruption early as otherwise
393 * iget_locked() could wait indefinitely for the state of our
396 if (parent->i_ino == ea_ino) {
397 ext4_error(parent->i_sb,
398 "Parent and EA inode have the same ino %lu", ea_ino);
399 return -EFSCORRUPTED;
402 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_EA_INODE);
404 err = PTR_ERR(inode);
405 ext4_error(parent->i_sb,
406 "error while reading EA inode %lu err=%d", ea_ino,
410 ext4_xattr_inode_set_class(inode);
413 * Check whether this is an old Lustre-style xattr inode. Lustre
414 * implementation does not have hash validation, rather it has a
415 * backpointer from ea_inode to the parent inode.
417 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
418 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
419 inode->i_generation == parent->i_generation) {
420 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
421 ext4_xattr_inode_set_ref(inode, 1);
424 inode->i_flags |= S_NOQUOTA;
432 /* Remove entry from mbcache when EA inode is getting evicted */
433 void ext4_evict_ea_inode(struct inode *inode)
435 struct mb_cache_entry *oe;
437 if (!EA_INODE_CACHE(inode))
439 /* Wait for entry to get unused so that we can remove it */
440 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
441 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
442 mb_cache_entry_wait_unused(oe);
443 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
448 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
449 struct ext4_xattr_entry *entry, void *buffer,
454 /* Verify stored hash matches calculated hash. */
455 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
456 if (hash != ext4_xattr_inode_get_hash(ea_inode))
457 return -EFSCORRUPTED;
460 __le32 e_hash, tmp_data;
462 /* Verify entry hash. */
463 tmp_data = cpu_to_le32(hash);
464 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
466 if (e_hash != entry->e_hash)
467 return -EFSCORRUPTED;
473 * Read xattr value from the EA inode.
476 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
477 void *buffer, size_t size)
479 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
480 struct inode *ea_inode;
483 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
484 le32_to_cpu(entry->e_hash), &ea_inode);
490 if (i_size_read(ea_inode) != size) {
491 ext4_warning_inode(ea_inode,
492 "ea_inode file size=%llu entry size=%zu",
493 i_size_read(ea_inode), size);
498 err = ext4_xattr_inode_read(ea_inode, buffer, size);
502 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
503 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
506 ext4_warning_inode(ea_inode,
507 "EA inode hash validation failed");
512 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
513 ext4_xattr_inode_get_hash(ea_inode),
514 ea_inode->i_ino, true /* reusable */);
522 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
523 void *buffer, size_t buffer_size)
525 struct buffer_head *bh = NULL;
526 struct ext4_xattr_entry *entry;
530 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
532 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
533 name_index, name, buffer, (long)buffer_size);
535 if (!EXT4_I(inode)->i_file_acl)
537 ea_idebug(inode, "reading block %llu",
538 (unsigned long long)EXT4_I(inode)->i_file_acl);
539 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
542 ea_bdebug(bh, "b_count=%d, refcount=%d",
543 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
544 error = ext4_xattr_check_block(inode, bh);
547 ext4_xattr_block_cache_insert(ea_block_cache, bh);
549 end = bh->b_data + bh->b_size;
550 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
553 size = le32_to_cpu(entry->e_value_size);
555 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
558 if (size > buffer_size)
560 if (entry->e_value_inum) {
561 error = ext4_xattr_inode_get(inode, entry, buffer,
566 u16 offset = le16_to_cpu(entry->e_value_offs);
567 void *p = bh->b_data + offset;
569 if (unlikely(p + size > end))
571 memcpy(buffer, p, size);
582 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
583 void *buffer, size_t buffer_size)
585 struct ext4_xattr_ibody_header *header;
586 struct ext4_xattr_entry *entry;
587 struct ext4_inode *raw_inode;
588 struct ext4_iloc iloc;
593 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
595 error = ext4_get_inode_loc(inode, &iloc);
598 raw_inode = ext4_raw_inode(&iloc);
599 header = IHDR(inode, raw_inode);
600 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
601 error = xattr_check_inode(inode, header, end);
604 entry = IFIRST(header);
605 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
608 size = le32_to_cpu(entry->e_value_size);
610 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
613 if (size > buffer_size)
615 if (entry->e_value_inum) {
616 error = ext4_xattr_inode_get(inode, entry, buffer,
621 u16 offset = le16_to_cpu(entry->e_value_offs);
622 void *p = (void *)IFIRST(header) + offset;
624 if (unlikely(p + size > end))
626 memcpy(buffer, p, size);
639 * Copy an extended attribute into the buffer
640 * provided, or compute the buffer size required.
641 * Buffer is NULL to compute the size of the buffer required.
643 * Returns a negative error number on failure, or the number of bytes
644 * used / required on success.
647 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
648 void *buffer, size_t buffer_size)
652 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
655 if (strlen(name) > 255)
658 down_read(&EXT4_I(inode)->xattr_sem);
659 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
661 if (error == -ENODATA)
662 error = ext4_xattr_block_get(inode, name_index, name, buffer,
664 up_read(&EXT4_I(inode)->xattr_sem);
669 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
670 char *buffer, size_t buffer_size)
672 size_t rest = buffer_size;
674 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
675 const struct xattr_handler *handler =
676 ext4_xattr_handler(entry->e_name_index);
678 if (handler && (!handler->list || handler->list(dentry))) {
679 const char *prefix = handler->prefix ?: handler->name;
680 size_t prefix_len = strlen(prefix);
681 size_t size = prefix_len + entry->e_name_len + 1;
686 memcpy(buffer, prefix, prefix_len);
687 buffer += prefix_len;
688 memcpy(buffer, entry->e_name, entry->e_name_len);
689 buffer += entry->e_name_len;
695 return buffer_size - rest; /* total size */
699 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
701 struct inode *inode = d_inode(dentry);
702 struct buffer_head *bh = NULL;
705 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
706 buffer, (long)buffer_size);
708 if (!EXT4_I(inode)->i_file_acl)
710 ea_idebug(inode, "reading block %llu",
711 (unsigned long long)EXT4_I(inode)->i_file_acl);
712 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
715 ea_bdebug(bh, "b_count=%d, refcount=%d",
716 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
717 error = ext4_xattr_check_block(inode, bh);
720 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
721 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
729 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
731 struct inode *inode = d_inode(dentry);
732 struct ext4_xattr_ibody_header *header;
733 struct ext4_inode *raw_inode;
734 struct ext4_iloc iloc;
738 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
740 error = ext4_get_inode_loc(inode, &iloc);
743 raw_inode = ext4_raw_inode(&iloc);
744 header = IHDR(inode, raw_inode);
745 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
746 error = xattr_check_inode(inode, header, end);
749 error = ext4_xattr_list_entries(dentry, IFIRST(header),
750 buffer, buffer_size);
758 * Inode operation listxattr()
760 * d_inode(dentry)->i_rwsem: don't care
762 * Copy a list of attribute names into the buffer
763 * provided, or compute the buffer size required.
764 * Buffer is NULL to compute the size of the buffer required.
766 * Returns a negative error number on failure, or the number of bytes
767 * used / required on success.
770 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
774 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
775 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
782 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
787 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
792 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
795 static void ext4_xattr_update_super_block(handle_t *handle,
796 struct super_block *sb)
798 if (ext4_has_feature_xattr(sb))
801 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
802 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
803 ext4_set_feature_xattr(sb);
804 ext4_handle_dirty_super(handle, sb);
808 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
810 struct ext4_iloc iloc = { .bh = NULL };
811 struct buffer_head *bh = NULL;
812 struct ext4_inode *raw_inode;
813 struct ext4_xattr_ibody_header *header;
814 struct ext4_xattr_entry *entry;
815 qsize_t ea_inode_refs = 0;
819 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
821 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
822 ret = ext4_get_inode_loc(inode, &iloc);
825 raw_inode = ext4_raw_inode(&iloc);
826 header = IHDR(inode, raw_inode);
827 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
828 ret = xattr_check_inode(inode, header, end);
832 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
833 entry = EXT4_XATTR_NEXT(entry))
834 if (entry->e_value_inum)
838 if (EXT4_I(inode)->i_file_acl) {
839 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
846 ret = ext4_xattr_check_block(inode, bh);
850 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
851 entry = EXT4_XATTR_NEXT(entry))
852 if (entry->e_value_inum)
855 *usage = ea_inode_refs + 1;
863 static inline size_t round_up_cluster(struct inode *inode, size_t length)
865 struct super_block *sb = inode->i_sb;
866 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
868 size_t mask = ~(cluster_size - 1);
870 return (length + cluster_size - 1) & mask;
873 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
877 err = dquot_alloc_inode(inode);
880 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
882 dquot_free_inode(inode);
886 static void ext4_xattr_inode_free_quota(struct inode *parent,
887 struct inode *ea_inode,
891 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
893 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
894 dquot_free_inode(parent);
897 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
898 struct buffer_head *block_bh, size_t value_len,
905 * 1) Owner inode update
906 * 2) Ref count update on old xattr block
908 * 4) block bitmap update for new xattr block
909 * 5) group descriptor for new xattr block
910 * 6) block bitmap update for old xattr block
911 * 7) group descriptor for old block
913 * 6 & 7 can happen if we have two racing threads T_a and T_b
914 * which are each trying to set an xattr on inodes I_a and I_b
915 * which were both initially sharing an xattr block.
920 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
923 * In case of inline data, we may push out the data to a block,
924 * so we need to reserve credits for this eventuality
926 if (inode && ext4_has_inline_data(inode))
927 credits += ext4_writepage_trans_blocks(inode) + 1;
929 /* We are done if ea_inode feature is not enabled. */
930 if (!ext4_has_feature_ea_inode(sb))
933 /* New ea_inode, inode map, block bitmap, group descriptor. */
937 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
939 /* Indirection block or one level of extent tree. */
942 /* Block bitmap and group descriptor updates for each block. */
943 credits += blocks * 2;
945 /* Blocks themselves. */
949 /* Dereference ea_inode holding old xattr value.
950 * Old ea_inode, inode map, block bitmap, group descriptor.
954 /* Data blocks for old ea_inode. */
955 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
957 /* Indirection block or one level of extent tree for old
962 /* Block bitmap and group descriptor updates for each block. */
963 credits += blocks * 2;
966 /* We may need to clone the existing xattr block in which case we need
967 * to increment ref counts for existing ea_inodes referenced by it.
970 struct ext4_xattr_entry *entry = BFIRST(block_bh);
972 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
973 if (entry->e_value_inum)
974 /* Ref count update on ea_inode. */
980 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
981 int credits, struct buffer_head *bh,
982 bool dirty, bool block_csum)
986 if (!ext4_handle_valid(handle))
989 if (handle->h_buffer_credits >= credits)
992 error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
996 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
1002 ext4_xattr_block_csum_set(inode, bh);
1003 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1005 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1011 error = ext4_journal_restart(handle, credits);
1013 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1018 error = ext4_journal_get_write_access(handle, bh);
1020 ext4_warning(inode->i_sb,
1021 "Get write access failed (error %d)",
1029 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1032 struct ext4_iloc iloc;
1036 inode_lock(ea_inode);
1038 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1042 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1043 ref_count += ref_change;
1044 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1046 if (ref_change > 0) {
1047 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1048 ea_inode->i_ino, ref_count);
1050 if (ref_count == 1) {
1051 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1052 ea_inode->i_ino, ea_inode->i_nlink);
1054 set_nlink(ea_inode, 1);
1055 ext4_orphan_del(handle, ea_inode);
1058 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1059 ea_inode->i_ino, ref_count);
1061 if (ref_count == 0) {
1062 WARN_ONCE(ea_inode->i_nlink != 1,
1063 "EA inode %lu i_nlink=%u",
1064 ea_inode->i_ino, ea_inode->i_nlink);
1066 clear_nlink(ea_inode);
1067 ext4_orphan_add(handle, ea_inode);
1071 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1073 ext4_warning_inode(ea_inode,
1074 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1076 inode_unlock(ea_inode);
1080 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1082 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1085 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1087 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1090 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1091 struct ext4_xattr_entry *first)
1093 struct inode *ea_inode;
1094 struct ext4_xattr_entry *entry;
1095 struct ext4_xattr_entry *failed_entry;
1096 unsigned int ea_ino;
1099 for (entry = first; !IS_LAST_ENTRY(entry);
1100 entry = EXT4_XATTR_NEXT(entry)) {
1101 if (!entry->e_value_inum)
1103 ea_ino = le32_to_cpu(entry->e_value_inum);
1104 err = ext4_xattr_inode_iget(parent, ea_ino,
1105 le32_to_cpu(entry->e_hash),
1109 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1111 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1121 failed_entry = entry;
1123 for (entry = first; entry != failed_entry;
1124 entry = EXT4_XATTR_NEXT(entry)) {
1125 if (!entry->e_value_inum)
1127 ea_ino = le32_to_cpu(entry->e_value_inum);
1128 err = ext4_xattr_inode_iget(parent, ea_ino,
1129 le32_to_cpu(entry->e_hash),
1132 ext4_warning(parent->i_sb,
1133 "cleanup ea_ino %u iget error %d", ea_ino,
1137 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1139 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1147 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1148 struct buffer_head *bh,
1149 struct ext4_xattr_entry *first, bool block_csum,
1150 struct ext4_xattr_inode_array **ea_inode_array,
1151 int extra_credits, bool skip_quota)
1153 struct inode *ea_inode;
1154 struct ext4_xattr_entry *entry;
1156 unsigned int ea_ino;
1160 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1161 credits = 2 + extra_credits;
1163 for (entry = first; !IS_LAST_ENTRY(entry);
1164 entry = EXT4_XATTR_NEXT(entry)) {
1165 if (!entry->e_value_inum)
1167 ea_ino = le32_to_cpu(entry->e_value_inum);
1168 err = ext4_xattr_inode_iget(parent, ea_ino,
1169 le32_to_cpu(entry->e_hash),
1174 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1176 ext4_warning_inode(ea_inode,
1177 "Expand inode array err=%d", err);
1182 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1185 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1190 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1192 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1198 ext4_xattr_inode_free_quota(parent, ea_inode,
1199 le32_to_cpu(entry->e_value_size));
1202 * Forget about ea_inode within the same transaction that
1203 * decrements the ref count. This avoids duplicate decrements in
1204 * case the rest of the work spills over to subsequent
1207 entry->e_value_inum = 0;
1208 entry->e_value_size = 0;
1215 * Note that we are deliberately skipping csum calculation for
1216 * the final update because we do not expect any journal
1217 * restarts until xattr block is freed.
1220 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1222 ext4_warning_inode(parent,
1223 "handle dirty metadata err=%d", err);
1228 * Release the xattr block BH: If the reference count is > 1, decrement it;
1229 * otherwise free the block.
1232 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1233 struct buffer_head *bh,
1234 struct ext4_xattr_inode_array **ea_inode_array,
1237 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1241 BUFFER_TRACE(bh, "get_write_access");
1242 error = ext4_journal_get_write_access(handle, bh);
1248 hash = le32_to_cpu(BHDR(bh)->h_hash);
1249 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1251 ea_bdebug(bh, "refcount now=0; freeing");
1253 * This must happen under buffer lock for
1254 * ext4_xattr_block_set() to reliably detect freed block
1256 if (ea_block_cache) {
1257 struct mb_cache_entry *oe;
1259 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1263 mb_cache_entry_wait_unused(oe);
1264 mb_cache_entry_put(ea_block_cache, oe);
1271 if (ext4_has_feature_ea_inode(inode->i_sb))
1272 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1274 true /* block_csum */,
1277 true /* skip_quota */);
1278 ext4_free_blocks(handle, inode, bh, 0, 1,
1279 EXT4_FREE_BLOCKS_METADATA |
1280 EXT4_FREE_BLOCKS_FORGET);
1283 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1284 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1285 struct mb_cache_entry *ce;
1287 if (ea_block_cache) {
1288 ce = mb_cache_entry_get(ea_block_cache, hash,
1291 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1292 mb_cache_entry_put(ea_block_cache, ce);
1297 ext4_xattr_block_csum_set(inode, bh);
1299 * Beware of this ugliness: Releasing of xattr block references
1300 * from different inodes can race and so we have to protect
1301 * from a race where someone else frees the block (and releases
1302 * its journal_head) before we are done dirtying the buffer. In
1303 * nojournal mode this race is harmless and we actually cannot
1304 * call ext4_handle_dirty_metadata() with locked buffer as
1305 * that function can call sync_dirty_buffer() so for that case
1306 * we handle the dirtying after unlocking the buffer.
1308 if (ext4_handle_valid(handle))
1309 error = ext4_handle_dirty_metadata(handle, inode, bh);
1311 if (!ext4_handle_valid(handle))
1312 error = ext4_handle_dirty_metadata(handle, inode, bh);
1314 ext4_handle_sync(handle);
1315 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1316 ea_bdebug(bh, "refcount now=%d; releasing",
1317 le32_to_cpu(BHDR(bh)->h_refcount));
1320 ext4_std_error(inode->i_sb, error);
1325 * Find the available free space for EAs. This also returns the total number of
1326 * bytes used by EA entries.
1328 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1329 size_t *min_offs, void *base, int *total)
1331 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1332 if (!last->e_value_inum && last->e_value_size) {
1333 size_t offs = le16_to_cpu(last->e_value_offs);
1334 if (offs < *min_offs)
1338 *total += EXT4_XATTR_LEN(last->e_name_len);
1340 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1344 * Write the value of the EA in an inode.
1346 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1347 const void *buf, int bufsize)
1349 struct buffer_head *bh = NULL;
1350 unsigned long block = 0;
1351 int blocksize = ea_inode->i_sb->s_blocksize;
1352 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1353 int csize, wsize = 0;
1358 while (ret >= 0 && ret < max_blocks) {
1359 struct ext4_map_blocks map;
1360 map.m_lblk = block += ret;
1361 map.m_len = max_blocks -= ret;
1363 ret = ext4_map_blocks(handle, ea_inode, &map,
1364 EXT4_GET_BLOCKS_CREATE);
1366 ext4_mark_inode_dirty(handle, ea_inode);
1367 if (ret == -ENOSPC &&
1368 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1380 while (wsize < bufsize) {
1383 csize = (bufsize - wsize) > blocksize ? blocksize :
1385 bh = ext4_getblk(handle, ea_inode, block, 0);
1390 EXT4_ERROR_INODE(ea_inode,
1391 "ext4_getblk() return bh = NULL");
1392 return -EFSCORRUPTED;
1394 ret = ext4_journal_get_write_access(handle, bh);
1398 memcpy(bh->b_data, buf, csize);
1399 set_buffer_uptodate(bh);
1400 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1407 inode_lock(ea_inode);
1408 i_size_write(ea_inode, wsize);
1409 ext4_update_i_disksize(ea_inode, wsize);
1410 inode_unlock(ea_inode);
1412 ext4_mark_inode_dirty(handle, ea_inode);
1421 * Create an inode to store the value of a large EA.
1423 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1424 struct inode *inode, u32 hash)
1426 struct inode *ea_inode = NULL;
1427 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1430 if (inode->i_sb->s_root == NULL) {
1431 ext4_warning(inode->i_sb,
1432 "refuse to create EA inode when umounting");
1434 return ERR_PTR(-EINVAL);
1438 * Let the next inode be the goal, so we try and allocate the EA inode
1439 * in the same group, or nearby one.
1441 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1442 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1444 if (!IS_ERR(ea_inode)) {
1445 ea_inode->i_op = &ext4_file_inode_operations;
1446 ea_inode->i_fop = &ext4_file_operations;
1447 ext4_set_aops(ea_inode);
1448 ext4_xattr_inode_set_class(ea_inode);
1449 unlock_new_inode(ea_inode);
1450 ext4_xattr_inode_set_ref(ea_inode, 1);
1451 ext4_xattr_inode_set_hash(ea_inode, hash);
1452 err = ext4_mark_inode_dirty(handle, ea_inode);
1454 err = ext4_inode_attach_jinode(ea_inode);
1456 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1457 ext4_warning_inode(ea_inode,
1458 "cleanup dec ref error %d", err);
1460 return ERR_PTR(err);
1464 * Xattr inodes are shared therefore quota charging is performed
1465 * at a higher level.
1467 dquot_free_inode(ea_inode);
1468 dquot_drop(ea_inode);
1469 inode_lock(ea_inode);
1470 ea_inode->i_flags |= S_NOQUOTA;
1471 inode_unlock(ea_inode);
1477 static struct inode *
1478 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1479 size_t value_len, u32 hash)
1481 struct inode *ea_inode;
1482 struct mb_cache_entry *ce;
1483 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1486 if (!ea_inode_cache)
1489 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1493 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1494 !(current->flags & PF_MEMALLOC_NOFS));
1496 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1498 mb_cache_entry_put(ea_inode_cache, ce);
1503 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1504 EXT4_IGET_EA_INODE);
1505 if (IS_ERR(ea_inode))
1507 ext4_xattr_inode_set_class(ea_inode);
1508 if (i_size_read(ea_inode) == value_len &&
1509 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1510 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1512 !memcmp(value, ea_data, value_len)) {
1513 mb_cache_entry_touch(ea_inode_cache, ce);
1514 mb_cache_entry_put(ea_inode_cache, ce);
1520 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1527 * Add value of the EA in an inode.
1529 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1530 const void *value, size_t value_len,
1531 struct inode **ret_inode)
1533 struct inode *ea_inode;
1537 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1538 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1540 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1546 *ret_inode = ea_inode;
1550 /* Create an inode for the EA value */
1551 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1552 if (IS_ERR(ea_inode))
1553 return PTR_ERR(ea_inode);
1555 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1557 ext4_xattr_inode_dec_ref(handle, ea_inode);
1562 if (EA_INODE_CACHE(inode))
1563 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1564 ea_inode->i_ino, true /* reusable */);
1566 *ret_inode = ea_inode;
1571 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1572 * feature is enabled.
1574 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1576 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1577 struct ext4_xattr_search *s,
1578 handle_t *handle, struct inode *inode,
1581 struct ext4_xattr_entry *last, *next;
1582 struct ext4_xattr_entry *here = s->here;
1583 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1584 int in_inode = i->in_inode;
1585 struct inode *old_ea_inode = NULL;
1586 struct inode *new_ea_inode = NULL;
1587 size_t old_size, new_size;
1590 /* Space used by old and new values. */
1591 old_size = (!s->not_found && !here->e_value_inum) ?
1592 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1593 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1596 * Optimization for the simple case when old and new values have the
1597 * same padded sizes. Not applicable if external inodes are involved.
1599 if (new_size && new_size == old_size) {
1600 size_t offs = le16_to_cpu(here->e_value_offs);
1601 void *val = s->base + offs;
1603 here->e_value_size = cpu_to_le32(i->value_len);
1604 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1605 memset(val, 0, new_size);
1607 memcpy(val, i->value, i->value_len);
1608 /* Clear padding bytes. */
1609 memset(val + i->value_len, 0, new_size - i->value_len);
1614 /* Compute min_offs and last. */
1616 for (; !IS_LAST_ENTRY(last); last = next) {
1617 next = EXT4_XATTR_NEXT(last);
1618 if ((void *)next >= s->end) {
1619 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1620 ret = -EFSCORRUPTED;
1623 if (!last->e_value_inum && last->e_value_size) {
1624 size_t offs = le16_to_cpu(last->e_value_offs);
1625 if (offs < min_offs)
1630 /* Check whether we have enough space. */
1634 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1636 free += EXT4_XATTR_LEN(name_len) + old_size;
1638 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1644 * If storing the value in an external inode is an option,
1645 * reserve space for xattr entries/names in the external
1646 * attribute block so that a long value does not occupy the
1647 * whole space and prevent futher entries being added.
1649 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1650 new_size && is_block &&
1651 (min_offs + old_size - new_size) <
1652 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1659 * Getting access to old and new ea inodes is subject to failures.
1660 * Finish that work before doing any modifications to the xattr data.
1662 if (!s->not_found && here->e_value_inum) {
1663 ret = ext4_xattr_inode_iget(inode,
1664 le32_to_cpu(here->e_value_inum),
1665 le32_to_cpu(here->e_hash),
1668 old_ea_inode = NULL;
1672 if (i->value && in_inode) {
1673 WARN_ON_ONCE(!i->value_len);
1675 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1679 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1683 new_ea_inode = NULL;
1684 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1690 /* We are ready to release ref count on the old_ea_inode. */
1691 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1693 /* Release newly required ref count on new_ea_inode. */
1697 err = ext4_xattr_inode_dec_ref(handle,
1700 ext4_warning_inode(new_ea_inode,
1701 "dec ref new_ea_inode err=%d",
1703 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1709 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1710 le32_to_cpu(here->e_value_size));
1713 /* No failures allowed past this point. */
1715 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1716 /* Remove the old value. */
1717 void *first_val = s->base + min_offs;
1718 size_t offs = le16_to_cpu(here->e_value_offs);
1719 void *val = s->base + offs;
1721 memmove(first_val + old_size, first_val, val - first_val);
1722 memset(first_val, 0, old_size);
1723 min_offs += old_size;
1725 /* Adjust all value offsets. */
1727 while (!IS_LAST_ENTRY(last)) {
1728 size_t o = le16_to_cpu(last->e_value_offs);
1730 if (!last->e_value_inum &&
1731 last->e_value_size && o < offs)
1732 last->e_value_offs = cpu_to_le16(o + old_size);
1733 last = EXT4_XATTR_NEXT(last);
1738 /* Remove old name. */
1739 size_t size = EXT4_XATTR_LEN(name_len);
1741 last = ENTRY((void *)last - size);
1742 memmove(here, (void *)here + size,
1743 (void *)last - (void *)here + sizeof(__u32));
1744 memset(last, 0, size);
1747 * Update i_inline_off - moved ibody region might contain
1748 * system.data attribute. Handling a failure here won't
1749 * cause other complications for setting an xattr.
1751 if (!is_block && ext4_has_inline_data(inode)) {
1752 ret = ext4_find_inline_data_nolock(inode);
1754 ext4_warning_inode(inode,
1755 "unable to update i_inline_off");
1759 } else if (s->not_found) {
1760 /* Insert new name. */
1761 size_t size = EXT4_XATTR_LEN(name_len);
1762 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1764 memmove((void *)here + size, here, rest);
1765 memset(here, 0, size);
1766 here->e_name_index = i->name_index;
1767 here->e_name_len = name_len;
1768 memcpy(here->e_name, i->name, name_len);
1770 /* This is an update, reset value info. */
1771 here->e_value_inum = 0;
1772 here->e_value_offs = 0;
1773 here->e_value_size = 0;
1777 /* Insert new value. */
1779 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1780 } else if (i->value_len) {
1781 void *val = s->base + min_offs - new_size;
1783 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1784 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1785 memset(val, 0, new_size);
1787 memcpy(val, i->value, i->value_len);
1788 /* Clear padding bytes. */
1789 memset(val + i->value_len, 0,
1790 new_size - i->value_len);
1793 here->e_value_size = cpu_to_le32(i->value_len);
1800 /* Entry hash calculation. */
1805 * Feed crc32c hash instead of the raw value for entry
1806 * hash calculation. This is to avoid walking
1807 * potentially long value buffer again.
1809 crc32c_hash = cpu_to_le32(
1810 ext4_xattr_inode_get_hash(new_ea_inode));
1811 hash = ext4_xattr_hash_entry(here->e_name,
1814 } else if (is_block) {
1815 __le32 *value = s->base + le16_to_cpu(
1816 here->e_value_offs);
1818 hash = ext4_xattr_hash_entry(here->e_name,
1819 here->e_name_len, value,
1822 here->e_hash = hash;
1826 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1835 struct ext4_xattr_block_find {
1836 struct ext4_xattr_search s;
1837 struct buffer_head *bh;
1841 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1842 struct ext4_xattr_block_find *bs)
1844 struct super_block *sb = inode->i_sb;
1847 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1848 i->name_index, i->name, i->value, (long)i->value_len);
1850 if (EXT4_I(inode)->i_file_acl) {
1851 /* The inode already has an extended attribute block. */
1852 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1853 if (IS_ERR(bs->bh)) {
1854 error = PTR_ERR(bs->bh);
1858 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1859 atomic_read(&(bs->bh->b_count)),
1860 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1861 error = ext4_xattr_check_block(inode, bs->bh);
1864 /* Find the named attribute. */
1865 bs->s.base = BHDR(bs->bh);
1866 bs->s.first = BFIRST(bs->bh);
1867 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1868 bs->s.here = bs->s.first;
1869 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1870 i->name_index, i->name, 1);
1871 if (error && error != -ENODATA)
1873 bs->s.not_found = error;
1879 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1880 struct ext4_xattr_info *i,
1881 struct ext4_xattr_block_find *bs)
1883 struct super_block *sb = inode->i_sb;
1884 struct buffer_head *new_bh = NULL;
1885 struct ext4_xattr_search s_copy = bs->s;
1886 struct ext4_xattr_search *s = &s_copy;
1887 struct mb_cache_entry *ce = NULL;
1889 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1890 struct inode *ea_inode = NULL, *tmp_inode;
1891 size_t old_ea_inode_quota = 0;
1892 unsigned int ea_ino;
1895 #define header(x) ((struct ext4_xattr_header *)(x))
1898 int offset = (char *)s->here - bs->bh->b_data;
1900 BUFFER_TRACE(bs->bh, "get_write_access");
1901 error = ext4_journal_get_write_access(handle, bs->bh);
1904 lock_buffer(bs->bh);
1906 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1907 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1910 * This must happen under buffer lock for
1911 * ext4_xattr_block_set() to reliably detect modified
1914 if (ea_block_cache) {
1915 struct mb_cache_entry *oe;
1917 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1918 hash, bs->bh->b_blocknr);
1921 * Xattr block is getting reused. Leave
1924 mb_cache_entry_put(ea_block_cache, oe);
1928 ea_bdebug(bs->bh, "modifying in-place");
1929 error = ext4_xattr_set_entry(i, s, handle, inode,
1930 true /* is_block */);
1931 ext4_xattr_block_csum_set(inode, bs->bh);
1932 unlock_buffer(bs->bh);
1933 if (error == -EFSCORRUPTED)
1936 error = ext4_handle_dirty_metadata(handle,
1944 unlock_buffer(bs->bh);
1945 ea_bdebug(bs->bh, "cloning");
1946 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1948 if (s->base == NULL)
1950 s->first = ENTRY(header(s->base)+1);
1951 header(s->base)->h_refcount = cpu_to_le32(1);
1952 s->here = ENTRY(s->base + offset);
1953 s->end = s->base + bs->bh->b_size;
1956 * If existing entry points to an xattr inode, we need
1957 * to prevent ext4_xattr_set_entry() from decrementing
1958 * ref count on it because the reference belongs to the
1959 * original block. In this case, make the entry look
1960 * like it has an empty value.
1962 if (!s->not_found && s->here->e_value_inum) {
1963 ea_ino = le32_to_cpu(s->here->e_value_inum);
1964 error = ext4_xattr_inode_iget(inode, ea_ino,
1965 le32_to_cpu(s->here->e_hash),
1970 if (!ext4_test_inode_state(tmp_inode,
1971 EXT4_STATE_LUSTRE_EA_INODE)) {
1973 * Defer quota free call for previous
1974 * inode until success is guaranteed.
1976 old_ea_inode_quota = le32_to_cpu(
1977 s->here->e_value_size);
1981 s->here->e_value_inum = 0;
1982 s->here->e_value_size = 0;
1985 /* Allocate a buffer where we construct the new block. */
1986 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1987 /* assert(header == s->base) */
1989 if (s->base == NULL)
1991 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1992 header(s->base)->h_blocks = cpu_to_le32(1);
1993 header(s->base)->h_refcount = cpu_to_le32(1);
1994 s->first = ENTRY(header(s->base)+1);
1995 s->here = ENTRY(header(s->base)+1);
1996 s->end = s->base + sb->s_blocksize;
1999 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
2000 if (error == -EFSCORRUPTED)
2005 if (i->value && s->here->e_value_inum) {
2007 * A ref count on ea_inode has been taken as part of the call to
2008 * ext4_xattr_set_entry() above. We would like to drop this
2009 * extra ref but we have to wait until the xattr block is
2010 * initialized and has its own ref count on the ea_inode.
2012 ea_ino = le32_to_cpu(s->here->e_value_inum);
2013 error = ext4_xattr_inode_iget(inode, ea_ino,
2014 le32_to_cpu(s->here->e_hash),
2023 if (!IS_LAST_ENTRY(s->first)) {
2024 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2027 /* We found an identical block in the cache. */
2028 if (new_bh == bs->bh)
2029 ea_bdebug(new_bh, "keeping");
2033 #ifdef EXT4_XATTR_DEBUG
2034 WARN_ON_ONCE(dquot_initialize_needed(inode));
2036 /* The old block is released after updating
2038 error = dquot_alloc_block(inode,
2039 EXT4_C2B(EXT4_SB(sb), 1));
2042 BUFFER_TRACE(new_bh, "get_write_access");
2043 error = ext4_journal_get_write_access(handle,
2047 lock_buffer(new_bh);
2049 * We have to be careful about races with
2050 * adding references to xattr block. Once we
2051 * hold buffer lock xattr block's state is
2052 * stable so we can check the additional
2055 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2056 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2058 * Undo everything and check mbcache
2061 unlock_buffer(new_bh);
2062 dquot_free_block(inode,
2063 EXT4_C2B(EXT4_SB(sb),
2066 mb_cache_entry_put(ea_block_cache, ce);
2071 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2072 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2073 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2074 ea_bdebug(new_bh, "reusing; refcount now=%d",
2076 ext4_xattr_block_csum_set(inode, new_bh);
2077 unlock_buffer(new_bh);
2078 error = ext4_handle_dirty_metadata(handle,
2084 mb_cache_entry_touch(ea_block_cache, ce);
2085 mb_cache_entry_put(ea_block_cache, ce);
2087 } else if (bs->bh && s->base == bs->bh->b_data) {
2088 /* We were modifying this block in-place. */
2089 ea_bdebug(bs->bh, "keeping this block");
2090 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2094 /* We need to allocate a new block */
2095 ext4_fsblk_t goal, block;
2097 #ifdef EXT4_XATTR_DEBUG
2098 WARN_ON_ONCE(dquot_initialize_needed(inode));
2100 goal = ext4_group_first_block_no(sb,
2101 EXT4_I(inode)->i_block_group);
2102 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2107 ea_idebug(inode, "creating block %llu",
2108 (unsigned long long)block);
2110 new_bh = sb_getblk(sb, block);
2111 if (unlikely(!new_bh)) {
2114 ext4_free_blocks(handle, inode, NULL, block, 1,
2115 EXT4_FREE_BLOCKS_METADATA);
2118 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2119 ENTRY(header(s->base)+1));
2123 /* Drop the extra ref on ea_inode. */
2124 error = ext4_xattr_inode_dec_ref(handle,
2127 ext4_warning_inode(ea_inode,
2134 lock_buffer(new_bh);
2135 error = ext4_journal_get_create_access(handle, new_bh);
2137 unlock_buffer(new_bh);
2141 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2142 ext4_xattr_block_csum_set(inode, new_bh);
2143 set_buffer_uptodate(new_bh);
2144 unlock_buffer(new_bh);
2145 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2146 error = ext4_handle_dirty_metadata(handle, inode,
2153 if (old_ea_inode_quota)
2154 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2156 /* Update the inode. */
2157 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2159 /* Drop the previous xattr block. */
2160 if (bs->bh && bs->bh != new_bh) {
2161 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2163 ext4_xattr_release_block(handle, inode, bs->bh,
2165 0 /* extra_credits */);
2166 ext4_xattr_inode_array_free(ea_inode_array);
2174 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2176 ext4_warning_inode(ea_inode, "dec ref error=%d",
2179 /* If there was an error, revert the quota charge. */
2181 ext4_xattr_inode_free_quota(inode, ea_inode,
2182 i_size_read(ea_inode));
2186 mb_cache_entry_put(ea_block_cache, ce);
2188 if (!(bs->bh && s->base == bs->bh->b_data))
2194 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2198 EXT4_ERROR_INODE(inode, "bad block %llu",
2199 EXT4_I(inode)->i_file_acl);
2205 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2206 struct ext4_xattr_ibody_find *is)
2208 struct ext4_xattr_ibody_header *header;
2209 struct ext4_inode *raw_inode;
2212 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2215 raw_inode = ext4_raw_inode(&is->iloc);
2216 header = IHDR(inode, raw_inode);
2217 is->s.base = is->s.first = IFIRST(header);
2218 is->s.here = is->s.first;
2219 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2220 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2221 error = xattr_check_inode(inode, header, is->s.end);
2224 /* Find the named attribute. */
2225 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2226 i->name_index, i->name, 0);
2227 if (error && error != -ENODATA)
2229 is->s.not_found = error;
2234 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2235 struct ext4_xattr_info *i,
2236 struct ext4_xattr_ibody_find *is)
2238 struct ext4_xattr_ibody_header *header;
2239 struct ext4_xattr_search *s = &is->s;
2242 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2245 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2248 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2249 if (!IS_LAST_ENTRY(s->first)) {
2250 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2251 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2253 header->h_magic = cpu_to_le32(0);
2254 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2259 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2260 struct ext4_xattr_info *i)
2264 /* When e_value_inum is set the value is stored externally. */
2265 if (s->here->e_value_inum)
2267 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2269 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2270 return !memcmp(value, i->value, i->value_len);
2273 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2275 struct buffer_head *bh;
2278 if (!EXT4_I(inode)->i_file_acl)
2280 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2283 error = ext4_xattr_check_block(inode, bh);
2286 return ERR_PTR(error);
2292 * ext4_xattr_set_handle()
2294 * Create, replace or remove an extended attribute for this inode. Value
2295 * is NULL to remove an existing extended attribute, and non-NULL to
2296 * either replace an existing extended attribute, or create a new extended
2297 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2298 * specify that an extended attribute must exist and must not exist
2299 * previous to the call, respectively.
2301 * Returns 0, or a negative error number on failure.
2304 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2305 const char *name, const void *value, size_t value_len,
2308 struct ext4_xattr_info i = {
2309 .name_index = name_index,
2312 .value_len = value_len,
2315 struct ext4_xattr_ibody_find is = {
2316 .s = { .not_found = -ENODATA, },
2318 struct ext4_xattr_block_find bs = {
2319 .s = { .not_found = -ENODATA, },
2326 if (strlen(name) > 255)
2329 ext4_write_lock_xattr(inode, &no_expand);
2331 /* Check journal credits under write lock. */
2332 if (ext4_handle_valid(handle)) {
2333 struct buffer_head *bh;
2336 bh = ext4_xattr_get_block(inode);
2338 error = PTR_ERR(bh);
2342 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2344 flags & XATTR_CREATE);
2347 if (!ext4_handle_has_enough_credits(handle, credits)) {
2351 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2354 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2358 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2359 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2360 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2361 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2364 error = ext4_xattr_ibody_find(inode, &i, &is);
2368 error = ext4_xattr_block_find(inode, &i, &bs);
2371 if (is.s.not_found && bs.s.not_found) {
2373 if (flags & XATTR_REPLACE)
2380 if (flags & XATTR_CREATE)
2385 if (!is.s.not_found)
2386 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2387 else if (!bs.s.not_found)
2388 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2391 /* Xattr value did not change? Save us some work and bail out */
2392 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2394 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2397 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2398 (EXT4_XATTR_SIZE(i.value_len) >
2399 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2402 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2403 if (!error && !bs.s.not_found) {
2405 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2406 } else if (error == -ENOSPC) {
2407 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2410 error = ext4_xattr_block_find(inode, &i, &bs);
2414 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2415 if (!error && !is.s.not_found) {
2417 error = ext4_xattr_ibody_set(handle, inode, &i,
2419 } else if (error == -ENOSPC) {
2421 * Xattr does not fit in the block, store at
2422 * external inode if possible.
2424 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2425 i.value_len && !i.in_inode) {
2433 ext4_xattr_update_super_block(handle, inode->i_sb);
2434 inode->i_ctime = current_time(inode);
2437 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2439 * The bh is consumed by ext4_mark_iloc_dirty, even with
2444 ext4_handle_sync(handle);
2450 ext4_write_unlock_xattr(inode, &no_expand);
2454 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2455 bool is_create, int *credits)
2457 struct buffer_head *bh;
2462 if (!EXT4_SB(inode->i_sb)->s_journal)
2465 down_read(&EXT4_I(inode)->xattr_sem);
2467 bh = ext4_xattr_get_block(inode);
2471 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2472 value_len, is_create);
2477 up_read(&EXT4_I(inode)->xattr_sem);
2484 * Like ext4_xattr_set_handle, but start from an inode. This extended
2485 * attribute modification is a filesystem transaction by itself.
2487 * Returns 0, or a negative error number on failure.
2490 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2491 const void *value, size_t value_len, int flags)
2494 struct super_block *sb = inode->i_sb;
2495 int error, retries = 0;
2498 error = dquot_initialize(inode);
2503 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2508 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2509 if (IS_ERR(handle)) {
2510 error = PTR_ERR(handle);
2514 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2515 value, value_len, flags);
2516 error2 = ext4_journal_stop(handle);
2517 if (error == -ENOSPC &&
2518 ext4_should_retry_alloc(sb, &retries))
2528 * Shift the EA entries in the inode to create space for the increased
2531 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2532 int value_offs_shift, void *to,
2533 void *from, size_t n)
2535 struct ext4_xattr_entry *last = entry;
2538 /* We always shift xattr headers further thus offsets get lower */
2539 BUG_ON(value_offs_shift > 0);
2541 /* Adjust the value offsets of the entries */
2542 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2543 if (!last->e_value_inum && last->e_value_size) {
2544 new_offs = le16_to_cpu(last->e_value_offs) +
2546 last->e_value_offs = cpu_to_le16(new_offs);
2549 /* Shift the entries by n bytes */
2550 memmove(to, from, n);
2554 * Move xattr pointed to by 'entry' from inode into external xattr block
2556 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2557 struct ext4_inode *raw_inode,
2558 struct ext4_xattr_entry *entry)
2560 struct ext4_xattr_ibody_find *is = NULL;
2561 struct ext4_xattr_block_find *bs = NULL;
2562 char *buffer = NULL, *b_entry_name = NULL;
2563 size_t value_size = le32_to_cpu(entry->e_value_size);
2564 struct ext4_xattr_info i = {
2567 .name_index = entry->e_name_index,
2568 .in_inode = !!entry->e_value_inum,
2570 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2571 int needs_kvfree = 0;
2574 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2575 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2576 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2577 if (!is || !bs || !b_entry_name) {
2582 is->s.not_found = -ENODATA;
2583 bs->s.not_found = -ENODATA;
2587 /* Save the entry name and the entry value */
2588 if (entry->e_value_inum) {
2589 buffer = kvmalloc(value_size, GFP_NOFS);
2595 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2599 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2600 buffer = (void *)IFIRST(header) + value_offs;
2603 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2604 b_entry_name[entry->e_name_len] = '\0';
2605 i.name = b_entry_name;
2607 error = ext4_get_inode_loc(inode, &is->iloc);
2611 error = ext4_xattr_ibody_find(inode, &i, is);
2616 i.value_len = value_size;
2617 error = ext4_xattr_block_find(inode, &i, bs);
2621 /* Move ea entry from the inode into the block */
2622 error = ext4_xattr_block_set(handle, inode, &i, bs);
2626 /* Remove the chosen entry from the inode */
2629 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2632 kfree(b_entry_name);
2633 if (needs_kvfree && buffer)
2636 brelse(is->iloc.bh);
2645 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2646 struct ext4_inode *raw_inode,
2647 int isize_diff, size_t ifree,
2648 size_t bfree, int *total_ino)
2650 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2651 struct ext4_xattr_entry *small_entry;
2652 struct ext4_xattr_entry *entry;
2653 struct ext4_xattr_entry *last;
2654 unsigned int entry_size; /* EA entry size */
2655 unsigned int total_size; /* EA entry size + value size */
2656 unsigned int min_total_size;
2659 while (isize_diff > ifree) {
2662 min_total_size = ~0U;
2663 last = IFIRST(header);
2664 /* Find the entry best suited to be pushed into EA block */
2665 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2666 /* never move system.data out of the inode */
2667 if ((last->e_name_len == 4) &&
2668 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2669 !memcmp(last->e_name, "data", 4))
2671 total_size = EXT4_XATTR_LEN(last->e_name_len);
2672 if (!last->e_value_inum)
2673 total_size += EXT4_XATTR_SIZE(
2674 le32_to_cpu(last->e_value_size));
2675 if (total_size <= bfree &&
2676 total_size < min_total_size) {
2677 if (total_size + ifree < isize_diff) {
2681 min_total_size = total_size;
2686 if (entry == NULL) {
2687 if (small_entry == NULL)
2689 entry = small_entry;
2692 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2693 total_size = entry_size;
2694 if (!entry->e_value_inum)
2695 total_size += EXT4_XATTR_SIZE(
2696 le32_to_cpu(entry->e_value_size));
2697 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2702 *total_ino -= entry_size;
2703 ifree += total_size;
2704 bfree -= total_size;
2711 * Expand an inode by new_extra_isize bytes when EAs are present.
2712 * Returns 0 on success or negative error number on failure.
2714 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2715 struct ext4_inode *raw_inode, handle_t *handle)
2717 struct ext4_xattr_ibody_header *header;
2718 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2719 static unsigned int mnt_count;
2721 size_t ifree, bfree;
2724 int error = 0, tried_min_extra_isize = 0;
2725 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2726 int isize_diff; /* How much do we need to grow i_extra_isize */
2729 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2730 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2733 header = IHDR(inode, raw_inode);
2736 * Check if enough free space is available in the inode to shift the
2737 * entries ahead by new_extra_isize.
2740 base = IFIRST(header);
2741 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2742 min_offs = end - base;
2743 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2745 error = xattr_check_inode(inode, header, end);
2749 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2750 if (ifree >= isize_diff)
2754 * Enough free space isn't available in the inode, check if
2755 * EA block can hold new_extra_isize bytes.
2757 if (EXT4_I(inode)->i_file_acl) {
2758 struct buffer_head *bh;
2760 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2762 error = PTR_ERR(bh);
2765 error = ext4_xattr_check_block(inode, bh);
2771 end = bh->b_data + bh->b_size;
2772 min_offs = end - base;
2773 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2776 if (bfree + ifree < isize_diff) {
2777 if (!tried_min_extra_isize && s_min_extra_isize) {
2778 tried_min_extra_isize++;
2779 new_extra_isize = s_min_extra_isize;
2786 bfree = inode->i_sb->s_blocksize;
2789 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2790 isize_diff, ifree, bfree,
2793 if (error == -ENOSPC && !tried_min_extra_isize &&
2794 s_min_extra_isize) {
2795 tried_min_extra_isize++;
2796 new_extra_isize = s_min_extra_isize;
2802 /* Adjust the offsets and shift the remaining entries ahead */
2803 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2804 - new_extra_isize, (void *)raw_inode +
2805 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2806 (void *)header, total_ino);
2807 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2809 if (ext4_has_inline_data(inode))
2810 error = ext4_find_inline_data_nolock(inode);
2813 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2814 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2816 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2821 #define EIA_INCR 16 /* must be 2^n */
2822 #define EIA_MASK (EIA_INCR - 1)
2824 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2825 * If @ea_inode_array is new or full it will be grown and the old
2826 * contents copied over.
2829 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2830 struct inode *inode)
2832 if (*ea_inode_array == NULL) {
2834 * Start with 15 inodes, so it fits into a power-of-two size.
2835 * If *ea_inode_array is NULL, this is essentially offsetof()
2838 kmalloc(offsetof(struct ext4_xattr_inode_array,
2841 if (*ea_inode_array == NULL)
2843 (*ea_inode_array)->count = 0;
2844 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2845 /* expand the array once all 15 + n * 16 slots are full */
2846 struct ext4_xattr_inode_array *new_array = NULL;
2847 int count = (*ea_inode_array)->count;
2849 /* if new_array is NULL, this is essentially offsetof() */
2850 new_array = kmalloc(
2851 offsetof(struct ext4_xattr_inode_array,
2852 inodes[count + EIA_INCR]),
2854 if (new_array == NULL)
2856 memcpy(new_array, *ea_inode_array,
2857 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2858 kfree(*ea_inode_array);
2859 *ea_inode_array = new_array;
2861 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2866 * ext4_xattr_delete_inode()
2868 * Free extended attribute resources associated with this inode. Traverse
2869 * all entries and decrement reference on any xattr inodes associated with this
2870 * inode. This is called immediately before an inode is freed. We have exclusive
2871 * access to the inode. If an orphan inode is deleted it will also release its
2872 * references on xattr block and xattr inodes.
2874 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2875 struct ext4_xattr_inode_array **ea_inode_array,
2878 struct buffer_head *bh = NULL;
2879 struct ext4_xattr_ibody_header *header;
2880 struct ext4_iloc iloc = { .bh = NULL };
2881 struct ext4_xattr_entry *entry;
2882 struct inode *ea_inode;
2885 error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2888 false /* block_csum */);
2890 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2894 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2895 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2897 error = ext4_get_inode_loc(inode, &iloc);
2899 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2903 error = ext4_journal_get_write_access(handle, iloc.bh);
2905 EXT4_ERROR_INODE(inode, "write access (error %d)",
2910 header = IHDR(inode, ext4_raw_inode(&iloc));
2911 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2912 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2914 false /* block_csum */,
2917 false /* skip_quota */);
2920 if (EXT4_I(inode)->i_file_acl) {
2921 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2923 error = PTR_ERR(bh);
2925 EXT4_ERROR_INODE(inode, "block %llu read error",
2926 EXT4_I(inode)->i_file_acl);
2930 error = ext4_xattr_check_block(inode, bh);
2934 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2935 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2936 entry = EXT4_XATTR_NEXT(entry)) {
2937 if (!entry->e_value_inum)
2939 error = ext4_xattr_inode_iget(inode,
2940 le32_to_cpu(entry->e_value_inum),
2941 le32_to_cpu(entry->e_hash),
2945 ext4_xattr_inode_free_quota(inode, ea_inode,
2946 le32_to_cpu(entry->e_value_size));
2952 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2955 * Update i_file_acl value in the same transaction that releases
2958 EXT4_I(inode)->i_file_acl = 0;
2959 error = ext4_mark_inode_dirty(handle, inode);
2961 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2973 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2977 if (ea_inode_array == NULL)
2980 for (idx = 0; idx < ea_inode_array->count; ++idx)
2981 iput(ea_inode_array->inodes[idx]);
2982 kfree(ea_inode_array);
2986 * ext4_xattr_block_cache_insert()
2988 * Create a new entry in the extended attribute block cache, and insert
2989 * it unless such an entry is already in the cache.
2991 * Returns 0, or a negative error number on failure.
2994 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2995 struct buffer_head *bh)
2997 struct ext4_xattr_header *header = BHDR(bh);
2998 __u32 hash = le32_to_cpu(header->h_hash);
2999 int reusable = le32_to_cpu(header->h_refcount) <
3000 EXT4_XATTR_REFCOUNT_MAX;
3003 if (!ea_block_cache)
3005 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3006 bh->b_blocknr, reusable);
3008 if (error == -EBUSY)
3009 ea_bdebug(bh, "already in cache");
3011 ea_bdebug(bh, "inserting [%x]", (int)hash);
3017 * Compare two extended attribute blocks for equality.
3019 * Returns 0 if the blocks are equal, 1 if they differ, and
3020 * a negative error number on errors.
3023 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3024 struct ext4_xattr_header *header2)
3026 struct ext4_xattr_entry *entry1, *entry2;
3028 entry1 = ENTRY(header1+1);
3029 entry2 = ENTRY(header2+1);
3030 while (!IS_LAST_ENTRY(entry1)) {
3031 if (IS_LAST_ENTRY(entry2))
3033 if (entry1->e_hash != entry2->e_hash ||
3034 entry1->e_name_index != entry2->e_name_index ||
3035 entry1->e_name_len != entry2->e_name_len ||
3036 entry1->e_value_size != entry2->e_value_size ||
3037 entry1->e_value_inum != entry2->e_value_inum ||
3038 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3040 if (!entry1->e_value_inum &&
3041 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3042 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3043 le32_to_cpu(entry1->e_value_size)))
3046 entry1 = EXT4_XATTR_NEXT(entry1);
3047 entry2 = EXT4_XATTR_NEXT(entry2);
3049 if (!IS_LAST_ENTRY(entry2))
3055 * ext4_xattr_block_cache_find()
3057 * Find an identical extended attribute block.
3059 * Returns a pointer to the block found, or NULL if such a block was
3060 * not found or an error occurred.
3062 static struct buffer_head *
3063 ext4_xattr_block_cache_find(struct inode *inode,
3064 struct ext4_xattr_header *header,
3065 struct mb_cache_entry **pce)
3067 __u32 hash = le32_to_cpu(header->h_hash);
3068 struct mb_cache_entry *ce;
3069 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3071 if (!ea_block_cache)
3073 if (!header->h_hash)
3074 return NULL; /* never share */
3075 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3076 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3078 struct buffer_head *bh;
3080 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3082 if (PTR_ERR(bh) == -ENOMEM)
3085 EXT4_ERROR_INODE(inode, "block %lu read error",
3086 (unsigned long)ce->e_value);
3087 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3092 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3097 #define NAME_HASH_SHIFT 5
3098 #define VALUE_HASH_SHIFT 16
3101 * ext4_xattr_hash_entry()
3103 * Compute the hash of an extended attribute.
3105 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3110 while (name_len--) {
3111 hash = (hash << NAME_HASH_SHIFT) ^
3112 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3115 while (value_count--) {
3116 hash = (hash << VALUE_HASH_SHIFT) ^
3117 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3118 le32_to_cpu(*value++);
3120 return cpu_to_le32(hash);
3123 #undef NAME_HASH_SHIFT
3124 #undef VALUE_HASH_SHIFT
3126 #define BLOCK_HASH_SHIFT 16
3129 * ext4_xattr_rehash()
3131 * Re-compute the extended attribute hash value after an entry has changed.
3133 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3135 struct ext4_xattr_entry *here;
3138 here = ENTRY(header+1);
3139 while (!IS_LAST_ENTRY(here)) {
3140 if (!here->e_hash) {
3141 /* Block is not shared if an entry's hash value == 0 */
3145 hash = (hash << BLOCK_HASH_SHIFT) ^
3146 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3147 le32_to_cpu(here->e_hash);
3148 here = EXT4_XATTR_NEXT(here);
3150 header->h_hash = cpu_to_le32(hash);
3153 #undef BLOCK_HASH_SHIFT
3155 #define HASH_BUCKET_BITS 10
3158 ext4_xattr_create_cache(void)
3160 return mb_cache_create(HASH_BUCKET_BITS);
3163 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3166 mb_cache_destroy(cache);