1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/xattr.c
5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9 * Extended attributes for symlinks and special files added per
10 * suggestion of Luka Renko <luka.renko@hermes.si>.
11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14 * and Andreas Gruenbacher <agruen@suse.de>.
18 * Extended attributes are stored directly in inodes (on file systems with
19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
20 * field contains the block number if an inode uses an additional block. All
21 * attributes must fit in the inode and one additional block. Blocks that
22 * contain the identical set of attributes may be shared among several inodes.
23 * Identical blocks are detected by keeping a cache of blocks that have
24 * recently been accessed.
26 * The attributes in inodes and on blocks have a different header; the entries
27 * are stored in the same format:
29 * +------------------+
32 * | entry 2 | | growing downwards
37 * | value 3 | | growing upwards
39 * +------------------+
41 * The header is followed by multiple entry descriptors. In disk blocks, the
42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
43 * attribute values are aligned to the end of the block in no specific order.
47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
48 * EA blocks are only changed if they are exclusive to an inode, so
49 * holding xattr_sem also means that nothing but the EA block's reference
50 * count can change. Multiple writers to the same block are synchronized
54 #include <linux/init.h>
56 #include <linux/slab.h>
57 #include <linux/mbcache.h>
58 #include <linux/quotaops.h>
59 #include "ext4_jbd2.h"
64 #ifdef EXT4_XATTR_DEBUG
65 # define ea_idebug(inode, fmt, ...) \
66 printk(KERN_DEBUG "inode %s:%lu: " fmt "\n", \
67 inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
68 # define ea_bdebug(bh, fmt, ...) \
69 printk(KERN_DEBUG "block %pg:%lu: " fmt "\n", \
70 bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
72 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
73 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
76 static void ext4_xattr_block_cache_insert(struct mb_cache *,
77 struct buffer_head *);
78 static struct buffer_head *
79 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
80 struct mb_cache_entry **);
81 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
83 static void ext4_xattr_rehash(struct ext4_xattr_header *);
85 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
86 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
87 #ifdef CONFIG_EXT4_FS_POSIX_ACL
88 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
89 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
91 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
92 #ifdef CONFIG_EXT4_FS_SECURITY
93 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
97 const struct xattr_handler *ext4_xattr_handlers[] = {
98 &ext4_xattr_user_handler,
99 &ext4_xattr_trusted_handler,
100 #ifdef CONFIG_EXT4_FS_POSIX_ACL
101 &posix_acl_access_xattr_handler,
102 &posix_acl_default_xattr_handler,
104 #ifdef CONFIG_EXT4_FS_SECURITY
105 &ext4_xattr_security_handler,
110 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
111 inode->i_sb->s_fs_info)->s_ea_block_cache)
113 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
114 inode->i_sb->s_fs_info)->s_ea_inode_cache)
117 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
118 struct inode *inode);
120 #ifdef CONFIG_LOCKDEP
121 void ext4_xattr_inode_set_class(struct inode *ea_inode)
123 struct ext4_inode_info *ei = EXT4_I(ea_inode);
125 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
126 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
127 lockdep_set_subclass(&ei->i_data_sem, I_DATA_SEM_EA);
131 static __le32 ext4_xattr_block_csum(struct inode *inode,
133 struct ext4_xattr_header *hdr)
135 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
137 __le64 dsk_block_nr = cpu_to_le64(block_nr);
138 __u32 dummy_csum = 0;
139 int offset = offsetof(struct ext4_xattr_header, h_checksum);
141 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
142 sizeof(dsk_block_nr));
143 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
144 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
145 offset += sizeof(dummy_csum);
146 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
147 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
149 return cpu_to_le32(csum);
152 static int ext4_xattr_block_csum_verify(struct inode *inode,
153 struct buffer_head *bh)
155 struct ext4_xattr_header *hdr = BHDR(bh);
158 if (ext4_has_metadata_csum(inode->i_sb)) {
160 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
161 bh->b_blocknr, hdr));
167 static void ext4_xattr_block_csum_set(struct inode *inode,
168 struct buffer_head *bh)
170 if (ext4_has_metadata_csum(inode->i_sb))
171 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
172 bh->b_blocknr, BHDR(bh));
175 static inline const struct xattr_handler *
176 ext4_xattr_handler(int name_index)
178 const struct xattr_handler *handler = NULL;
180 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
181 handler = ext4_xattr_handler_map[name_index];
186 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
189 struct ext4_xattr_entry *e = entry;
191 /* Find the end of the names list */
192 while (!IS_LAST_ENTRY(e)) {
193 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
194 if ((void *)next >= end)
195 return -EFSCORRUPTED;
196 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
197 return -EFSCORRUPTED;
201 /* Check the values */
202 while (!IS_LAST_ENTRY(entry)) {
203 u32 size = le32_to_cpu(entry->e_value_size);
205 if (size > EXT4_XATTR_SIZE_MAX)
206 return -EFSCORRUPTED;
208 if (size != 0 && entry->e_value_inum == 0) {
209 u16 offs = le16_to_cpu(entry->e_value_offs);
213 * The value cannot overlap the names, and the value
214 * with padding cannot extend beyond 'end'. Check both
215 * the padded and unpadded sizes, since the size may
216 * overflow to 0 when adding padding.
218 if (offs > end - value_start)
219 return -EFSCORRUPTED;
220 value = value_start + offs;
221 if (value < (void *)e + sizeof(u32) ||
222 size > end - value ||
223 EXT4_XATTR_SIZE(size) > end - value)
224 return -EFSCORRUPTED;
226 entry = EXT4_XATTR_NEXT(entry);
233 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
234 const char *function, unsigned int line)
236 int error = -EFSCORRUPTED;
238 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
239 BHDR(bh)->h_blocks != cpu_to_le32(1))
241 if (buffer_verified(bh))
245 if (!ext4_xattr_block_csum_verify(inode, bh))
247 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
251 __ext4_error_inode(inode, function, line, 0,
252 "corrupted xattr block %llu",
253 (unsigned long long) bh->b_blocknr);
255 set_buffer_verified(bh);
259 #define ext4_xattr_check_block(inode, bh) \
260 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
264 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
265 void *end, const char *function, unsigned int line)
267 int error = -EFSCORRUPTED;
269 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
270 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
272 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
275 __ext4_error_inode(inode, function, line, 0,
276 "corrupted in-inode xattr");
280 #define xattr_check_inode(inode, header, end) \
281 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
284 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
285 void *end, int name_index, const char *name, int sorted)
287 struct ext4_xattr_entry *entry, *next;
293 name_len = strlen(name);
294 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
295 next = EXT4_XATTR_NEXT(entry);
296 if ((void *) next >= end) {
297 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
298 return -EFSCORRUPTED;
300 cmp = name_index - entry->e_name_index;
302 cmp = name_len - entry->e_name_len;
304 cmp = memcmp(name, entry->e_name, name_len);
305 if (cmp <= 0 && (sorted || cmp == 0))
309 return cmp ? -ENODATA : 0;
313 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
315 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
318 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
320 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
321 ((u32)ea_inode->i_version);
324 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
326 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
327 ea_inode->i_version = (u32)ref_count;
330 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
332 return (u32)ea_inode->i_atime.tv_sec;
335 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
337 ea_inode->i_atime.tv_sec = hash;
341 * Read the EA value from an inode.
343 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
345 int blocksize = 1 << ea_inode->i_blkbits;
346 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
347 int tail_size = (size % blocksize) ?: blocksize;
348 struct buffer_head *bhs_inline[8];
349 struct buffer_head **bhs = bhs_inline;
352 if (bh_count > ARRAY_SIZE(bhs_inline)) {
353 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
358 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
359 true /* wait */, bhs);
363 for (i = 0; i < bh_count; i++) {
364 /* There shouldn't be any holes in ea_inode. */
369 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
370 i < bh_count - 1 ? blocksize : tail_size);
374 for (i = 0; i < bh_count; i++)
377 if (bhs != bhs_inline)
382 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
384 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
385 u32 ea_inode_hash, struct inode **ea_inode)
391 * We have to check for this corruption early as otherwise
392 * iget_locked() could wait indefinitely for the state of our
395 if (parent->i_ino == ea_ino) {
396 ext4_error(parent->i_sb,
397 "Parent and EA inode have the same ino %lu", ea_ino);
398 return -EFSCORRUPTED;
401 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
403 err = PTR_ERR(inode);
404 ext4_error(parent->i_sb,
405 "error while reading EA inode %lu err=%d", ea_ino,
410 if (is_bad_inode(inode)) {
411 ext4_error(parent->i_sb,
412 "error while reading EA inode %lu is_bad_inode",
418 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
419 ext4_error(parent->i_sb,
420 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
426 ext4_xattr_inode_set_class(inode);
429 * Check whether this is an old Lustre-style xattr inode. Lustre
430 * implementation does not have hash validation, rather it has a
431 * backpointer from ea_inode to the parent inode.
433 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
434 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
435 inode->i_generation == parent->i_generation) {
436 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
437 ext4_xattr_inode_set_ref(inode, 1);
440 inode->i_flags |= S_NOQUOTA;
452 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
453 struct ext4_xattr_entry *entry, void *buffer,
458 /* Verify stored hash matches calculated hash. */
459 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
460 if (hash != ext4_xattr_inode_get_hash(ea_inode))
461 return -EFSCORRUPTED;
464 __le32 e_hash, tmp_data;
466 /* Verify entry hash. */
467 tmp_data = cpu_to_le32(hash);
468 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
470 if (e_hash != entry->e_hash)
471 return -EFSCORRUPTED;
477 * Read xattr value from the EA inode.
480 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
481 void *buffer, size_t size)
483 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
484 struct inode *ea_inode;
487 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
488 le32_to_cpu(entry->e_hash), &ea_inode);
494 if (i_size_read(ea_inode) != size) {
495 ext4_warning_inode(ea_inode,
496 "ea_inode file size=%llu entry size=%zu",
497 i_size_read(ea_inode), size);
502 err = ext4_xattr_inode_read(ea_inode, buffer, size);
506 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
507 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
510 ext4_warning_inode(ea_inode,
511 "EA inode hash validation failed");
516 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
517 ext4_xattr_inode_get_hash(ea_inode),
518 ea_inode->i_ino, true /* reusable */);
526 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
527 void *buffer, size_t buffer_size)
529 struct buffer_head *bh = NULL;
530 struct ext4_xattr_entry *entry;
534 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
536 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
537 name_index, name, buffer, (long)buffer_size);
539 if (!EXT4_I(inode)->i_file_acl)
541 ea_idebug(inode, "reading block %llu",
542 (unsigned long long)EXT4_I(inode)->i_file_acl);
543 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
546 ea_bdebug(bh, "b_count=%d, refcount=%d",
547 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
548 error = ext4_xattr_check_block(inode, bh);
551 ext4_xattr_block_cache_insert(ea_block_cache, bh);
553 end = bh->b_data + bh->b_size;
554 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
557 size = le32_to_cpu(entry->e_value_size);
559 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
562 if (size > buffer_size)
564 if (entry->e_value_inum) {
565 error = ext4_xattr_inode_get(inode, entry, buffer,
570 u16 offset = le16_to_cpu(entry->e_value_offs);
571 void *p = bh->b_data + offset;
573 if (unlikely(p + size > end))
575 memcpy(buffer, p, size);
586 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
587 void *buffer, size_t buffer_size)
589 struct ext4_xattr_ibody_header *header;
590 struct ext4_xattr_entry *entry;
591 struct ext4_inode *raw_inode;
592 struct ext4_iloc iloc;
597 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
599 error = ext4_get_inode_loc(inode, &iloc);
602 raw_inode = ext4_raw_inode(&iloc);
603 header = IHDR(inode, raw_inode);
604 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
605 error = xattr_check_inode(inode, header, end);
608 entry = IFIRST(header);
609 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
612 size = le32_to_cpu(entry->e_value_size);
614 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
617 if (size > buffer_size)
619 if (entry->e_value_inum) {
620 error = ext4_xattr_inode_get(inode, entry, buffer,
625 u16 offset = le16_to_cpu(entry->e_value_offs);
626 void *p = (void *)IFIRST(header) + offset;
628 if (unlikely(p + size > end))
630 memcpy(buffer, p, size);
643 * Copy an extended attribute into the buffer
644 * provided, or compute the buffer size required.
645 * Buffer is NULL to compute the size of the buffer required.
647 * Returns a negative error number on failure, or the number of bytes
648 * used / required on success.
651 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
652 void *buffer, size_t buffer_size)
656 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
659 if (strlen(name) > 255)
662 down_read(&EXT4_I(inode)->xattr_sem);
663 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
665 if (error == -ENODATA)
666 error = ext4_xattr_block_get(inode, name_index, name, buffer,
668 up_read(&EXT4_I(inode)->xattr_sem);
673 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
674 char *buffer, size_t buffer_size)
676 size_t rest = buffer_size;
678 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
679 const struct xattr_handler *handler =
680 ext4_xattr_handler(entry->e_name_index);
682 if (handler && (!handler->list || handler->list(dentry))) {
683 const char *prefix = handler->prefix ?: handler->name;
684 size_t prefix_len = strlen(prefix);
685 size_t size = prefix_len + entry->e_name_len + 1;
690 memcpy(buffer, prefix, prefix_len);
691 buffer += prefix_len;
692 memcpy(buffer, entry->e_name, entry->e_name_len);
693 buffer += entry->e_name_len;
699 return buffer_size - rest; /* total size */
703 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
705 struct inode *inode = d_inode(dentry);
706 struct buffer_head *bh = NULL;
709 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
710 buffer, (long)buffer_size);
712 if (!EXT4_I(inode)->i_file_acl)
714 ea_idebug(inode, "reading block %llu",
715 (unsigned long long)EXT4_I(inode)->i_file_acl);
716 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
719 ea_bdebug(bh, "b_count=%d, refcount=%d",
720 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
721 error = ext4_xattr_check_block(inode, bh);
724 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
725 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
733 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
735 struct inode *inode = d_inode(dentry);
736 struct ext4_xattr_ibody_header *header;
737 struct ext4_inode *raw_inode;
738 struct ext4_iloc iloc;
742 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
744 error = ext4_get_inode_loc(inode, &iloc);
747 raw_inode = ext4_raw_inode(&iloc);
748 header = IHDR(inode, raw_inode);
749 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
750 error = xattr_check_inode(inode, header, end);
753 error = ext4_xattr_list_entries(dentry, IFIRST(header),
754 buffer, buffer_size);
762 * Inode operation listxattr()
764 * d_inode(dentry)->i_rwsem: don't care
766 * Copy a list of attribute names into the buffer
767 * provided, or compute the buffer size required.
768 * Buffer is NULL to compute the size of the buffer required.
770 * Returns a negative error number on failure, or the number of bytes
771 * used / required on success.
774 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
778 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
779 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
786 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
791 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
796 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
799 static void ext4_xattr_update_super_block(handle_t *handle,
800 struct super_block *sb)
802 if (ext4_has_feature_xattr(sb))
805 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
806 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
807 ext4_set_feature_xattr(sb);
808 ext4_handle_dirty_super(handle, sb);
812 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
814 struct ext4_iloc iloc = { .bh = NULL };
815 struct buffer_head *bh = NULL;
816 struct ext4_inode *raw_inode;
817 struct ext4_xattr_ibody_header *header;
818 struct ext4_xattr_entry *entry;
819 qsize_t ea_inode_refs = 0;
823 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
825 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
826 ret = ext4_get_inode_loc(inode, &iloc);
829 raw_inode = ext4_raw_inode(&iloc);
830 header = IHDR(inode, raw_inode);
831 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
832 ret = xattr_check_inode(inode, header, end);
836 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
837 entry = EXT4_XATTR_NEXT(entry))
838 if (entry->e_value_inum)
842 if (EXT4_I(inode)->i_file_acl) {
843 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
850 ret = ext4_xattr_check_block(inode, bh);
854 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
855 entry = EXT4_XATTR_NEXT(entry))
856 if (entry->e_value_inum)
859 *usage = ea_inode_refs + 1;
867 static inline size_t round_up_cluster(struct inode *inode, size_t length)
869 struct super_block *sb = inode->i_sb;
870 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
872 size_t mask = ~(cluster_size - 1);
874 return (length + cluster_size - 1) & mask;
877 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
881 err = dquot_alloc_inode(inode);
884 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
886 dquot_free_inode(inode);
890 static void ext4_xattr_inode_free_quota(struct inode *parent,
891 struct inode *ea_inode,
895 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
897 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
898 dquot_free_inode(parent);
901 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
902 struct buffer_head *block_bh, size_t value_len,
909 * 1) Owner inode update
910 * 2) Ref count update on old xattr block
912 * 4) block bitmap update for new xattr block
913 * 5) group descriptor for new xattr block
914 * 6) block bitmap update for old xattr block
915 * 7) group descriptor for old block
917 * 6 & 7 can happen if we have two racing threads T_a and T_b
918 * which are each trying to set an xattr on inodes I_a and I_b
919 * which were both initially sharing an xattr block.
924 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
927 * In case of inline data, we may push out the data to a block,
928 * so we need to reserve credits for this eventuality
930 if (inode && ext4_has_inline_data(inode))
931 credits += ext4_writepage_trans_blocks(inode) + 1;
933 /* We are done if ea_inode feature is not enabled. */
934 if (!ext4_has_feature_ea_inode(sb))
937 /* New ea_inode, inode map, block bitmap, group descriptor. */
941 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
943 /* Indirection block or one level of extent tree. */
946 /* Block bitmap and group descriptor updates for each block. */
947 credits += blocks * 2;
949 /* Blocks themselves. */
953 /* Dereference ea_inode holding old xattr value.
954 * Old ea_inode, inode map, block bitmap, group descriptor.
958 /* Data blocks for old ea_inode. */
959 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
961 /* Indirection block or one level of extent tree for old
966 /* Block bitmap and group descriptor updates for each block. */
967 credits += blocks * 2;
970 /* We may need to clone the existing xattr block in which case we need
971 * to increment ref counts for existing ea_inodes referenced by it.
974 struct ext4_xattr_entry *entry = BFIRST(block_bh);
976 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
977 if (entry->e_value_inum)
978 /* Ref count update on ea_inode. */
984 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
985 int credits, struct buffer_head *bh,
986 bool dirty, bool block_csum)
990 if (!ext4_handle_valid(handle))
993 if (handle->h_buffer_credits >= credits)
996 error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
1000 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
1006 ext4_xattr_block_csum_set(inode, bh);
1007 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1009 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1015 error = ext4_journal_restart(handle, credits);
1017 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1022 error = ext4_journal_get_write_access(handle, bh);
1024 ext4_warning(inode->i_sb,
1025 "Get write access failed (error %d)",
1033 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1036 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1037 struct ext4_iloc iloc;
1042 inode_lock(ea_inode);
1044 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1050 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1051 ref_count += ref_change;
1052 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1054 if (ref_change > 0) {
1055 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1056 ea_inode->i_ino, ref_count);
1058 if (ref_count == 1) {
1059 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1060 ea_inode->i_ino, ea_inode->i_nlink);
1062 set_nlink(ea_inode, 1);
1063 ext4_orphan_del(handle, ea_inode);
1065 if (ea_inode_cache) {
1066 hash = ext4_xattr_inode_get_hash(ea_inode);
1067 mb_cache_entry_create(ea_inode_cache,
1070 true /* reusable */);
1074 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1075 ea_inode->i_ino, ref_count);
1077 if (ref_count == 0) {
1078 WARN_ONCE(ea_inode->i_nlink != 1,
1079 "EA inode %lu i_nlink=%u",
1080 ea_inode->i_ino, ea_inode->i_nlink);
1082 clear_nlink(ea_inode);
1083 ext4_orphan_add(handle, ea_inode);
1085 if (ea_inode_cache) {
1086 hash = ext4_xattr_inode_get_hash(ea_inode);
1087 mb_cache_entry_delete(ea_inode_cache, hash,
1093 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1096 ext4_warning_inode(ea_inode,
1097 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1100 inode_unlock(ea_inode);
1104 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1106 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1109 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1111 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1114 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1115 struct ext4_xattr_entry *first)
1117 struct inode *ea_inode;
1118 struct ext4_xattr_entry *entry;
1119 struct ext4_xattr_entry *failed_entry;
1120 unsigned int ea_ino;
1123 for (entry = first; !IS_LAST_ENTRY(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),
1133 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1135 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1145 failed_entry = entry;
1147 for (entry = first; entry != failed_entry;
1148 entry = EXT4_XATTR_NEXT(entry)) {
1149 if (!entry->e_value_inum)
1151 ea_ino = le32_to_cpu(entry->e_value_inum);
1152 err = ext4_xattr_inode_iget(parent, ea_ino,
1153 le32_to_cpu(entry->e_hash),
1156 ext4_warning(parent->i_sb,
1157 "cleanup ea_ino %u iget error %d", ea_ino,
1161 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1163 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1171 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1172 struct buffer_head *bh,
1173 struct ext4_xattr_entry *first, bool block_csum,
1174 struct ext4_xattr_inode_array **ea_inode_array,
1175 int extra_credits, bool skip_quota)
1177 struct inode *ea_inode;
1178 struct ext4_xattr_entry *entry;
1180 unsigned int ea_ino;
1184 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1185 credits = 2 + extra_credits;
1187 for (entry = first; !IS_LAST_ENTRY(entry);
1188 entry = EXT4_XATTR_NEXT(entry)) {
1189 if (!entry->e_value_inum)
1191 ea_ino = le32_to_cpu(entry->e_value_inum);
1192 err = ext4_xattr_inode_iget(parent, ea_ino,
1193 le32_to_cpu(entry->e_hash),
1198 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1200 ext4_warning_inode(ea_inode,
1201 "Expand inode array err=%d", err);
1206 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1209 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1214 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1216 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1222 ext4_xattr_inode_free_quota(parent, ea_inode,
1223 le32_to_cpu(entry->e_value_size));
1226 * Forget about ea_inode within the same transaction that
1227 * decrements the ref count. This avoids duplicate decrements in
1228 * case the rest of the work spills over to subsequent
1231 entry->e_value_inum = 0;
1232 entry->e_value_size = 0;
1239 * Note that we are deliberately skipping csum calculation for
1240 * the final update because we do not expect any journal
1241 * restarts until xattr block is freed.
1244 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1246 ext4_warning_inode(parent,
1247 "handle dirty metadata err=%d", err);
1252 * Release the xattr block BH: If the reference count is > 1, decrement it;
1253 * otherwise free the block.
1256 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1257 struct buffer_head *bh,
1258 struct ext4_xattr_inode_array **ea_inode_array,
1261 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1265 BUFFER_TRACE(bh, "get_write_access");
1266 error = ext4_journal_get_write_access(handle, bh);
1271 hash = le32_to_cpu(BHDR(bh)->h_hash);
1272 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1274 ea_bdebug(bh, "refcount now=0; freeing");
1276 * This must happen under buffer lock for
1277 * ext4_xattr_block_set() to reliably detect freed block
1280 mb_cache_entry_delete(ea_block_cache, hash,
1285 if (ext4_has_feature_ea_inode(inode->i_sb))
1286 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1288 true /* block_csum */,
1291 true /* skip_quota */);
1292 ext4_free_blocks(handle, inode, bh, 0, 1,
1293 EXT4_FREE_BLOCKS_METADATA |
1294 EXT4_FREE_BLOCKS_FORGET);
1297 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1298 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1299 struct mb_cache_entry *ce;
1301 if (ea_block_cache) {
1302 ce = mb_cache_entry_get(ea_block_cache, hash,
1306 mb_cache_entry_put(ea_block_cache, ce);
1311 ext4_xattr_block_csum_set(inode, bh);
1313 * Beware of this ugliness: Releasing of xattr block references
1314 * from different inodes can race and so we have to protect
1315 * from a race where someone else frees the block (and releases
1316 * its journal_head) before we are done dirtying the buffer. In
1317 * nojournal mode this race is harmless and we actually cannot
1318 * call ext4_handle_dirty_metadata() with locked buffer as
1319 * that function can call sync_dirty_buffer() so for that case
1320 * we handle the dirtying after unlocking the buffer.
1322 if (ext4_handle_valid(handle))
1323 error = ext4_handle_dirty_metadata(handle, inode, bh);
1325 if (!ext4_handle_valid(handle))
1326 error = ext4_handle_dirty_metadata(handle, inode, bh);
1328 ext4_handle_sync(handle);
1329 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1330 ea_bdebug(bh, "refcount now=%d; releasing",
1331 le32_to_cpu(BHDR(bh)->h_refcount));
1334 ext4_std_error(inode->i_sb, error);
1339 * Find the available free space for EAs. This also returns the total number of
1340 * bytes used by EA entries.
1342 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1343 size_t *min_offs, void *base, int *total)
1345 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1346 if (!last->e_value_inum && last->e_value_size) {
1347 size_t offs = le16_to_cpu(last->e_value_offs);
1348 if (offs < *min_offs)
1352 *total += EXT4_XATTR_LEN(last->e_name_len);
1354 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1358 * Write the value of the EA in an inode.
1360 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1361 const void *buf, int bufsize)
1363 struct buffer_head *bh = NULL;
1364 unsigned long block = 0;
1365 int blocksize = ea_inode->i_sb->s_blocksize;
1366 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1367 int csize, wsize = 0;
1372 while (ret >= 0 && ret < max_blocks) {
1373 struct ext4_map_blocks map;
1374 map.m_lblk = block += ret;
1375 map.m_len = max_blocks -= ret;
1377 ret = ext4_map_blocks(handle, ea_inode, &map,
1378 EXT4_GET_BLOCKS_CREATE);
1380 ext4_mark_inode_dirty(handle, ea_inode);
1381 if (ret == -ENOSPC &&
1382 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1394 while (wsize < bufsize) {
1397 csize = (bufsize - wsize) > blocksize ? blocksize :
1399 bh = ext4_getblk(handle, ea_inode, block, 0);
1404 EXT4_ERROR_INODE(ea_inode,
1405 "ext4_getblk() return bh = NULL");
1406 return -EFSCORRUPTED;
1408 ret = ext4_journal_get_write_access(handle, bh);
1412 memcpy(bh->b_data, buf, csize);
1413 set_buffer_uptodate(bh);
1414 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1421 inode_lock(ea_inode);
1422 i_size_write(ea_inode, wsize);
1423 ext4_update_i_disksize(ea_inode, wsize);
1424 inode_unlock(ea_inode);
1426 ext4_mark_inode_dirty(handle, ea_inode);
1435 * Create an inode to store the value of a large EA.
1437 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1438 struct inode *inode, u32 hash)
1440 struct inode *ea_inode = NULL;
1441 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1444 if (inode->i_sb->s_root == NULL) {
1445 ext4_warning(inode->i_sb,
1446 "refuse to create EA inode when umounting");
1448 return ERR_PTR(-EINVAL);
1452 * Let the next inode be the goal, so we try and allocate the EA inode
1453 * in the same group, or nearby one.
1455 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1456 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1458 if (!IS_ERR(ea_inode)) {
1459 ea_inode->i_op = &ext4_file_inode_operations;
1460 ea_inode->i_fop = &ext4_file_operations;
1461 ext4_set_aops(ea_inode);
1462 ext4_xattr_inode_set_class(ea_inode);
1463 unlock_new_inode(ea_inode);
1464 ext4_xattr_inode_set_ref(ea_inode, 1);
1465 ext4_xattr_inode_set_hash(ea_inode, hash);
1466 err = ext4_mark_inode_dirty(handle, ea_inode);
1468 err = ext4_inode_attach_jinode(ea_inode);
1470 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1471 ext4_warning_inode(ea_inode,
1472 "cleanup dec ref error %d", err);
1474 return ERR_PTR(err);
1478 * Xattr inodes are shared therefore quota charging is performed
1479 * at a higher level.
1481 dquot_free_inode(ea_inode);
1482 dquot_drop(ea_inode);
1483 inode_lock(ea_inode);
1484 ea_inode->i_flags |= S_NOQUOTA;
1485 inode_unlock(ea_inode);
1491 static struct inode *
1492 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1493 size_t value_len, u32 hash)
1495 struct inode *ea_inode;
1496 struct mb_cache_entry *ce;
1497 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1500 if (!ea_inode_cache)
1503 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1507 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1508 !(current->flags & PF_MEMALLOC_NOFS));
1510 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1512 mb_cache_entry_put(ea_inode_cache, ce);
1517 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1519 if (!IS_ERR(ea_inode) &&
1520 !is_bad_inode(ea_inode) &&
1521 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1522 i_size_read(ea_inode) == value_len &&
1523 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1524 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1526 !memcmp(value, ea_data, value_len)) {
1527 mb_cache_entry_touch(ea_inode_cache, ce);
1528 mb_cache_entry_put(ea_inode_cache, ce);
1533 if (!IS_ERR(ea_inode))
1535 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1542 * Add value of the EA in an inode.
1544 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1545 const void *value, size_t value_len,
1546 struct inode **ret_inode)
1548 struct inode *ea_inode;
1552 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1553 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1555 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1561 *ret_inode = ea_inode;
1565 /* Create an inode for the EA value */
1566 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1567 if (IS_ERR(ea_inode))
1568 return PTR_ERR(ea_inode);
1570 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1572 ext4_xattr_inode_dec_ref(handle, ea_inode);
1577 if (EA_INODE_CACHE(inode))
1578 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1579 ea_inode->i_ino, true /* reusable */);
1581 *ret_inode = ea_inode;
1586 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1587 * feature is enabled.
1589 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1591 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1592 struct ext4_xattr_search *s,
1593 handle_t *handle, struct inode *inode,
1596 struct ext4_xattr_entry *last, *next;
1597 struct ext4_xattr_entry *here = s->here;
1598 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1599 int in_inode = i->in_inode;
1600 struct inode *old_ea_inode = NULL;
1601 struct inode *new_ea_inode = NULL;
1602 size_t old_size, new_size;
1605 /* Space used by old and new values. */
1606 old_size = (!s->not_found && !here->e_value_inum) ?
1607 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1608 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1611 * Optimization for the simple case when old and new values have the
1612 * same padded sizes. Not applicable if external inodes are involved.
1614 if (new_size && new_size == old_size) {
1615 size_t offs = le16_to_cpu(here->e_value_offs);
1616 void *val = s->base + offs;
1618 here->e_value_size = cpu_to_le32(i->value_len);
1619 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1620 memset(val, 0, new_size);
1622 memcpy(val, i->value, i->value_len);
1623 /* Clear padding bytes. */
1624 memset(val + i->value_len, 0, new_size - i->value_len);
1629 /* Compute min_offs and last. */
1631 for (; !IS_LAST_ENTRY(last); last = next) {
1632 next = EXT4_XATTR_NEXT(last);
1633 if ((void *)next >= s->end) {
1634 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1635 ret = -EFSCORRUPTED;
1638 if (!last->e_value_inum && last->e_value_size) {
1639 size_t offs = le16_to_cpu(last->e_value_offs);
1640 if (offs < min_offs)
1645 /* Check whether we have enough space. */
1649 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1651 free += EXT4_XATTR_LEN(name_len) + old_size;
1653 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1659 * If storing the value in an external inode is an option,
1660 * reserve space for xattr entries/names in the external
1661 * attribute block so that a long value does not occupy the
1662 * whole space and prevent futher entries being added.
1664 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1665 new_size && is_block &&
1666 (min_offs + old_size - new_size) <
1667 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1674 * Getting access to old and new ea inodes is subject to failures.
1675 * Finish that work before doing any modifications to the xattr data.
1677 if (!s->not_found && here->e_value_inum) {
1678 ret = ext4_xattr_inode_iget(inode,
1679 le32_to_cpu(here->e_value_inum),
1680 le32_to_cpu(here->e_hash),
1683 old_ea_inode = NULL;
1687 if (i->value && in_inode) {
1688 WARN_ON_ONCE(!i->value_len);
1690 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1694 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1698 new_ea_inode = NULL;
1699 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1705 /* We are ready to release ref count on the old_ea_inode. */
1706 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1708 /* Release newly required ref count on new_ea_inode. */
1712 err = ext4_xattr_inode_dec_ref(handle,
1715 ext4_warning_inode(new_ea_inode,
1716 "dec ref new_ea_inode err=%d",
1718 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1724 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1725 le32_to_cpu(here->e_value_size));
1728 /* No failures allowed past this point. */
1730 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1731 /* Remove the old value. */
1732 void *first_val = s->base + min_offs;
1733 size_t offs = le16_to_cpu(here->e_value_offs);
1734 void *val = s->base + offs;
1736 memmove(first_val + old_size, first_val, val - first_val);
1737 memset(first_val, 0, old_size);
1738 min_offs += old_size;
1740 /* Adjust all value offsets. */
1742 while (!IS_LAST_ENTRY(last)) {
1743 size_t o = le16_to_cpu(last->e_value_offs);
1745 if (!last->e_value_inum &&
1746 last->e_value_size && o < offs)
1747 last->e_value_offs = cpu_to_le16(o + old_size);
1748 last = EXT4_XATTR_NEXT(last);
1753 /* Remove old name. */
1754 size_t size = EXT4_XATTR_LEN(name_len);
1756 last = ENTRY((void *)last - size);
1757 memmove(here, (void *)here + size,
1758 (void *)last - (void *)here + sizeof(__u32));
1759 memset(last, 0, size);
1760 } else if (s->not_found) {
1761 /* Insert new name. */
1762 size_t size = EXT4_XATTR_LEN(name_len);
1763 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1765 memmove((void *)here + size, here, rest);
1766 memset(here, 0, size);
1767 here->e_name_index = i->name_index;
1768 here->e_name_len = name_len;
1769 memcpy(here->e_name, i->name, name_len);
1771 /* This is an update, reset value info. */
1772 here->e_value_inum = 0;
1773 here->e_value_offs = 0;
1774 here->e_value_size = 0;
1778 /* Insert new value. */
1780 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1781 } else if (i->value_len) {
1782 void *val = s->base + min_offs - new_size;
1784 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1785 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1786 memset(val, 0, new_size);
1788 memcpy(val, i->value, i->value_len);
1789 /* Clear padding bytes. */
1790 memset(val + i->value_len, 0,
1791 new_size - i->value_len);
1794 here->e_value_size = cpu_to_le32(i->value_len);
1801 /* Entry hash calculation. */
1806 * Feed crc32c hash instead of the raw value for entry
1807 * hash calculation. This is to avoid walking
1808 * potentially long value buffer again.
1810 crc32c_hash = cpu_to_le32(
1811 ext4_xattr_inode_get_hash(new_ea_inode));
1812 hash = ext4_xattr_hash_entry(here->e_name,
1815 } else if (is_block) {
1816 __le32 *value = s->base + le16_to_cpu(
1817 here->e_value_offs);
1819 hash = ext4_xattr_hash_entry(here->e_name,
1820 here->e_name_len, value,
1823 here->e_hash = hash;
1827 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1836 struct ext4_xattr_block_find {
1837 struct ext4_xattr_search s;
1838 struct buffer_head *bh;
1842 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1843 struct ext4_xattr_block_find *bs)
1845 struct super_block *sb = inode->i_sb;
1848 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1849 i->name_index, i->name, i->value, (long)i->value_len);
1851 if (EXT4_I(inode)->i_file_acl) {
1852 /* The inode already has an extended attribute block. */
1853 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1854 if (IS_ERR(bs->bh)) {
1855 error = PTR_ERR(bs->bh);
1859 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1860 atomic_read(&(bs->bh->b_count)),
1861 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1862 error = ext4_xattr_check_block(inode, bs->bh);
1865 /* Find the named attribute. */
1866 bs->s.base = BHDR(bs->bh);
1867 bs->s.first = BFIRST(bs->bh);
1868 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1869 bs->s.here = bs->s.first;
1870 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1871 i->name_index, i->name, 1);
1872 if (error && error != -ENODATA)
1874 bs->s.not_found = error;
1880 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1881 struct ext4_xattr_info *i,
1882 struct ext4_xattr_block_find *bs)
1884 struct super_block *sb = inode->i_sb;
1885 struct buffer_head *new_bh = NULL;
1886 struct ext4_xattr_search s_copy = bs->s;
1887 struct ext4_xattr_search *s = &s_copy;
1888 struct mb_cache_entry *ce = NULL;
1890 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1891 struct inode *ea_inode = NULL, *tmp_inode;
1892 size_t old_ea_inode_quota = 0;
1893 unsigned int ea_ino;
1896 #define header(x) ((struct ext4_xattr_header *)(x))
1899 BUFFER_TRACE(bs->bh, "get_write_access");
1900 error = ext4_journal_get_write_access(handle, bs->bh);
1903 lock_buffer(bs->bh);
1905 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1906 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1909 * This must happen under buffer lock for
1910 * ext4_xattr_block_set() to reliably detect modified
1914 mb_cache_entry_delete(ea_block_cache, hash,
1916 ea_bdebug(bs->bh, "modifying in-place");
1917 error = ext4_xattr_set_entry(i, s, handle, inode,
1918 true /* is_block */);
1919 ext4_xattr_block_csum_set(inode, bs->bh);
1920 unlock_buffer(bs->bh);
1921 if (error == -EFSCORRUPTED)
1924 error = ext4_handle_dirty_metadata(handle,
1931 int offset = (char *)s->here - bs->bh->b_data;
1933 unlock_buffer(bs->bh);
1934 ea_bdebug(bs->bh, "cloning");
1935 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1937 if (s->base == NULL)
1939 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1940 s->first = ENTRY(header(s->base)+1);
1941 header(s->base)->h_refcount = cpu_to_le32(1);
1942 s->here = ENTRY(s->base + offset);
1943 s->end = s->base + bs->bh->b_size;
1946 * If existing entry points to an xattr inode, we need
1947 * to prevent ext4_xattr_set_entry() from decrementing
1948 * ref count on it because the reference belongs to the
1949 * original block. In this case, make the entry look
1950 * like it has an empty value.
1952 if (!s->not_found && s->here->e_value_inum) {
1953 ea_ino = le32_to_cpu(s->here->e_value_inum);
1954 error = ext4_xattr_inode_iget(inode, ea_ino,
1955 le32_to_cpu(s->here->e_hash),
1960 if (!ext4_test_inode_state(tmp_inode,
1961 EXT4_STATE_LUSTRE_EA_INODE)) {
1963 * Defer quota free call for previous
1964 * inode until success is guaranteed.
1966 old_ea_inode_quota = le32_to_cpu(
1967 s->here->e_value_size);
1971 s->here->e_value_inum = 0;
1972 s->here->e_value_size = 0;
1976 /* Allocate a buffer where we construct the new block. */
1977 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1978 /* assert(header == s->base) */
1980 if (s->base == NULL)
1982 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1983 header(s->base)->h_blocks = cpu_to_le32(1);
1984 header(s->base)->h_refcount = cpu_to_le32(1);
1985 s->first = ENTRY(header(s->base)+1);
1986 s->here = ENTRY(header(s->base)+1);
1987 s->end = s->base + sb->s_blocksize;
1990 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1991 if (error == -EFSCORRUPTED)
1996 if (i->value && s->here->e_value_inum) {
1998 * A ref count on ea_inode has been taken as part of the call to
1999 * ext4_xattr_set_entry() above. We would like to drop this
2000 * extra ref but we have to wait until the xattr block is
2001 * initialized and has its own ref count on the ea_inode.
2003 ea_ino = le32_to_cpu(s->here->e_value_inum);
2004 error = ext4_xattr_inode_iget(inode, ea_ino,
2005 le32_to_cpu(s->here->e_hash),
2014 if (!IS_LAST_ENTRY(s->first)) {
2015 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2018 /* We found an identical block in the cache. */
2019 if (new_bh == bs->bh)
2020 ea_bdebug(new_bh, "keeping");
2024 #ifdef EXT4_XATTR_DEBUG
2025 WARN_ON_ONCE(dquot_initialize_needed(inode));
2027 /* The old block is released after updating
2029 error = dquot_alloc_block(inode,
2030 EXT4_C2B(EXT4_SB(sb), 1));
2033 BUFFER_TRACE(new_bh, "get_write_access");
2034 error = ext4_journal_get_write_access(handle,
2038 lock_buffer(new_bh);
2040 * We have to be careful about races with
2041 * freeing, rehashing or adding references to
2042 * xattr block. Once we hold buffer lock xattr
2043 * block's state is stable so we can check
2044 * whether the block got freed / rehashed or
2045 * not. Since we unhash mbcache entry under
2046 * buffer lock when freeing / rehashing xattr
2047 * block, checking whether entry is still
2048 * hashed is reliable. Same rules hold for
2049 * e_reusable handling.
2051 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2054 * Undo everything and check mbcache
2057 unlock_buffer(new_bh);
2058 dquot_free_block(inode,
2059 EXT4_C2B(EXT4_SB(sb),
2062 mb_cache_entry_put(ea_block_cache, ce);
2067 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2068 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2069 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2071 ea_bdebug(new_bh, "reusing; refcount now=%d",
2073 ext4_xattr_block_csum_set(inode, new_bh);
2074 unlock_buffer(new_bh);
2075 error = ext4_handle_dirty_metadata(handle,
2081 mb_cache_entry_touch(ea_block_cache, ce);
2082 mb_cache_entry_put(ea_block_cache, ce);
2084 } else if (bs->bh && s->base == bs->bh->b_data) {
2085 /* We were modifying this block in-place. */
2086 ea_bdebug(bs->bh, "keeping this block");
2087 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2091 /* We need to allocate a new block */
2092 ext4_fsblk_t goal, block;
2094 #ifdef EXT4_XATTR_DEBUG
2095 WARN_ON_ONCE(dquot_initialize_needed(inode));
2097 goal = ext4_group_first_block_no(sb,
2098 EXT4_I(inode)->i_block_group);
2099 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2104 ea_idebug(inode, "creating block %llu",
2105 (unsigned long long)block);
2107 new_bh = sb_getblk(sb, block);
2108 if (unlikely(!new_bh)) {
2111 ext4_free_blocks(handle, inode, NULL, block, 1,
2112 EXT4_FREE_BLOCKS_METADATA);
2115 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2116 ENTRY(header(s->base)+1));
2120 /* Drop the extra ref on ea_inode. */
2121 error = ext4_xattr_inode_dec_ref(handle,
2124 ext4_warning_inode(ea_inode,
2131 lock_buffer(new_bh);
2132 error = ext4_journal_get_create_access(handle, new_bh);
2134 unlock_buffer(new_bh);
2138 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2139 ext4_xattr_block_csum_set(inode, new_bh);
2140 set_buffer_uptodate(new_bh);
2141 unlock_buffer(new_bh);
2142 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2143 error = ext4_handle_dirty_metadata(handle, inode,
2150 if (old_ea_inode_quota)
2151 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2153 /* Update the inode. */
2154 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2156 /* Drop the previous xattr block. */
2157 if (bs->bh && bs->bh != new_bh) {
2158 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2160 ext4_xattr_release_block(handle, inode, bs->bh,
2162 0 /* extra_credits */);
2163 ext4_xattr_inode_array_free(ea_inode_array);
2171 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2173 ext4_warning_inode(ea_inode, "dec ref error=%d",
2176 /* If there was an error, revert the quota charge. */
2178 ext4_xattr_inode_free_quota(inode, ea_inode,
2179 i_size_read(ea_inode));
2183 mb_cache_entry_put(ea_block_cache, ce);
2185 if (!(bs->bh && s->base == bs->bh->b_data))
2191 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2195 EXT4_ERROR_INODE(inode, "bad block %llu",
2196 EXT4_I(inode)->i_file_acl);
2202 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2203 struct ext4_xattr_ibody_find *is)
2205 struct ext4_xattr_ibody_header *header;
2206 struct ext4_inode *raw_inode;
2209 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2212 raw_inode = ext4_raw_inode(&is->iloc);
2213 header = IHDR(inode, raw_inode);
2214 is->s.base = is->s.first = IFIRST(header);
2215 is->s.here = is->s.first;
2216 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2217 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2218 error = xattr_check_inode(inode, header, is->s.end);
2221 /* Find the named attribute. */
2222 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2223 i->name_index, i->name, 0);
2224 if (error && error != -ENODATA)
2226 is->s.not_found = error;
2231 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2232 struct ext4_xattr_info *i,
2233 struct ext4_xattr_ibody_find *is)
2235 struct ext4_xattr_ibody_header *header;
2236 struct ext4_xattr_search *s = &is->s;
2239 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2242 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2245 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2246 if (!IS_LAST_ENTRY(s->first)) {
2247 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2248 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2250 header->h_magic = cpu_to_le32(0);
2251 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2256 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2257 struct ext4_xattr_info *i)
2261 /* When e_value_inum is set the value is stored externally. */
2262 if (s->here->e_value_inum)
2264 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2266 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2267 return !memcmp(value, i->value, i->value_len);
2270 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2272 struct buffer_head *bh;
2275 if (!EXT4_I(inode)->i_file_acl)
2277 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2280 error = ext4_xattr_check_block(inode, bh);
2283 return ERR_PTR(error);
2289 * ext4_xattr_set_handle()
2291 * Create, replace or remove an extended attribute for this inode. Value
2292 * is NULL to remove an existing extended attribute, and non-NULL to
2293 * either replace an existing extended attribute, or create a new extended
2294 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2295 * specify that an extended attribute must exist and must not exist
2296 * previous to the call, respectively.
2298 * Returns 0, or a negative error number on failure.
2301 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2302 const char *name, const void *value, size_t value_len,
2305 struct ext4_xattr_info i = {
2306 .name_index = name_index,
2309 .value_len = value_len,
2312 struct ext4_xattr_ibody_find is = {
2313 .s = { .not_found = -ENODATA, },
2315 struct ext4_xattr_block_find bs = {
2316 .s = { .not_found = -ENODATA, },
2323 if (strlen(name) > 255)
2326 ext4_write_lock_xattr(inode, &no_expand);
2328 /* Check journal credits under write lock. */
2329 if (ext4_handle_valid(handle)) {
2330 struct buffer_head *bh;
2333 bh = ext4_xattr_get_block(inode);
2335 error = PTR_ERR(bh);
2339 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2341 flags & XATTR_CREATE);
2344 if (!ext4_handle_has_enough_credits(handle, credits)) {
2348 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2351 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2355 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2356 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2357 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2358 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2361 error = ext4_xattr_ibody_find(inode, &i, &is);
2365 error = ext4_xattr_block_find(inode, &i, &bs);
2368 if (is.s.not_found && bs.s.not_found) {
2370 if (flags & XATTR_REPLACE)
2377 if (flags & XATTR_CREATE)
2382 if (!is.s.not_found)
2383 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2384 else if (!bs.s.not_found)
2385 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2388 /* Xattr value did not change? Save us some work and bail out */
2389 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2391 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2394 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2395 (EXT4_XATTR_SIZE(i.value_len) >
2396 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2399 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2400 if (!error && !bs.s.not_found) {
2402 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2403 } else if (error == -ENOSPC) {
2404 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2407 error = ext4_xattr_block_find(inode, &i, &bs);
2411 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2412 if (!error && !is.s.not_found) {
2414 error = ext4_xattr_ibody_set(handle, inode, &i,
2416 } else if (error == -ENOSPC) {
2418 * Xattr does not fit in the block, store at
2419 * external inode if possible.
2421 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2422 i.value_len && !i.in_inode) {
2430 ext4_xattr_update_super_block(handle, inode->i_sb);
2431 inode->i_ctime = current_time(inode);
2434 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2436 * The bh is consumed by ext4_mark_iloc_dirty, even with
2441 ext4_handle_sync(handle);
2447 ext4_write_unlock_xattr(inode, &no_expand);
2451 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2452 bool is_create, int *credits)
2454 struct buffer_head *bh;
2459 if (!EXT4_SB(inode->i_sb)->s_journal)
2462 down_read(&EXT4_I(inode)->xattr_sem);
2464 bh = ext4_xattr_get_block(inode);
2468 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2469 value_len, is_create);
2474 up_read(&EXT4_I(inode)->xattr_sem);
2481 * Like ext4_xattr_set_handle, but start from an inode. This extended
2482 * attribute modification is a filesystem transaction by itself.
2484 * Returns 0, or a negative error number on failure.
2487 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2488 const void *value, size_t value_len, int flags)
2491 struct super_block *sb = inode->i_sb;
2492 int error, retries = 0;
2495 error = dquot_initialize(inode);
2500 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2505 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2506 if (IS_ERR(handle)) {
2507 error = PTR_ERR(handle);
2511 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2512 value, value_len, flags);
2513 error2 = ext4_journal_stop(handle);
2514 if (error == -ENOSPC &&
2515 ext4_should_retry_alloc(sb, &retries))
2525 * Shift the EA entries in the inode to create space for the increased
2528 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2529 int value_offs_shift, void *to,
2530 void *from, size_t n)
2532 struct ext4_xattr_entry *last = entry;
2535 /* We always shift xattr headers further thus offsets get lower */
2536 BUG_ON(value_offs_shift > 0);
2538 /* Adjust the value offsets of the entries */
2539 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2540 if (!last->e_value_inum && last->e_value_size) {
2541 new_offs = le16_to_cpu(last->e_value_offs) +
2543 last->e_value_offs = cpu_to_le16(new_offs);
2546 /* Shift the entries by n bytes */
2547 memmove(to, from, n);
2551 * Move xattr pointed to by 'entry' from inode into external xattr block
2553 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2554 struct ext4_inode *raw_inode,
2555 struct ext4_xattr_entry *entry)
2557 struct ext4_xattr_ibody_find *is = NULL;
2558 struct ext4_xattr_block_find *bs = NULL;
2559 char *buffer = NULL, *b_entry_name = NULL;
2560 size_t value_size = le32_to_cpu(entry->e_value_size);
2561 struct ext4_xattr_info i = {
2564 .name_index = entry->e_name_index,
2565 .in_inode = !!entry->e_value_inum,
2567 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2568 int needs_kvfree = 0;
2571 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2572 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2573 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2574 if (!is || !bs || !b_entry_name) {
2579 is->s.not_found = -ENODATA;
2580 bs->s.not_found = -ENODATA;
2584 /* Save the entry name and the entry value */
2585 if (entry->e_value_inum) {
2586 buffer = kvmalloc(value_size, GFP_NOFS);
2592 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2596 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2597 buffer = (void *)IFIRST(header) + value_offs;
2600 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2601 b_entry_name[entry->e_name_len] = '\0';
2602 i.name = b_entry_name;
2604 error = ext4_get_inode_loc(inode, &is->iloc);
2608 error = ext4_xattr_ibody_find(inode, &i, is);
2613 i.value_len = value_size;
2614 error = ext4_xattr_block_find(inode, &i, bs);
2618 /* Move ea entry from the inode into the block */
2619 error = ext4_xattr_block_set(handle, inode, &i, bs);
2623 /* Remove the chosen entry from the inode */
2626 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2629 kfree(b_entry_name);
2630 if (needs_kvfree && buffer)
2633 brelse(is->iloc.bh);
2642 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2643 struct ext4_inode *raw_inode,
2644 int isize_diff, size_t ifree,
2645 size_t bfree, int *total_ino)
2647 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2648 struct ext4_xattr_entry *small_entry;
2649 struct ext4_xattr_entry *entry;
2650 struct ext4_xattr_entry *last;
2651 unsigned int entry_size; /* EA entry size */
2652 unsigned int total_size; /* EA entry size + value size */
2653 unsigned int min_total_size;
2656 while (isize_diff > ifree) {
2659 min_total_size = ~0U;
2660 last = IFIRST(header);
2661 /* Find the entry best suited to be pushed into EA block */
2662 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2663 /* never move system.data out of the inode */
2664 if ((last->e_name_len == 4) &&
2665 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2666 !memcmp(last->e_name, "data", 4))
2668 total_size = EXT4_XATTR_LEN(last->e_name_len);
2669 if (!last->e_value_inum)
2670 total_size += EXT4_XATTR_SIZE(
2671 le32_to_cpu(last->e_value_size));
2672 if (total_size <= bfree &&
2673 total_size < min_total_size) {
2674 if (total_size + ifree < isize_diff) {
2678 min_total_size = total_size;
2683 if (entry == NULL) {
2684 if (small_entry == NULL)
2686 entry = small_entry;
2689 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2690 total_size = entry_size;
2691 if (!entry->e_value_inum)
2692 total_size += EXT4_XATTR_SIZE(
2693 le32_to_cpu(entry->e_value_size));
2694 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2699 *total_ino -= entry_size;
2700 ifree += total_size;
2701 bfree -= total_size;
2708 * Expand an inode by new_extra_isize bytes when EAs are present.
2709 * Returns 0 on success or negative error number on failure.
2711 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2712 struct ext4_inode *raw_inode, handle_t *handle)
2714 struct ext4_xattr_ibody_header *header;
2715 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2716 static unsigned int mnt_count;
2718 size_t ifree, bfree;
2721 int error = 0, tried_min_extra_isize = 0;
2722 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2723 int isize_diff; /* How much do we need to grow i_extra_isize */
2726 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2727 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2730 header = IHDR(inode, raw_inode);
2733 * Check if enough free space is available in the inode to shift the
2734 * entries ahead by new_extra_isize.
2737 base = IFIRST(header);
2738 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2739 min_offs = end - base;
2740 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2742 error = xattr_check_inode(inode, header, end);
2746 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2747 if (ifree >= isize_diff)
2751 * Enough free space isn't available in the inode, check if
2752 * EA block can hold new_extra_isize bytes.
2754 if (EXT4_I(inode)->i_file_acl) {
2755 struct buffer_head *bh;
2757 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2759 error = PTR_ERR(bh);
2762 error = ext4_xattr_check_block(inode, bh);
2768 end = bh->b_data + bh->b_size;
2769 min_offs = end - base;
2770 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2773 if (bfree + ifree < isize_diff) {
2774 if (!tried_min_extra_isize && s_min_extra_isize) {
2775 tried_min_extra_isize++;
2776 new_extra_isize = s_min_extra_isize;
2783 bfree = inode->i_sb->s_blocksize;
2786 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2787 isize_diff, ifree, bfree,
2790 if (error == -ENOSPC && !tried_min_extra_isize &&
2791 s_min_extra_isize) {
2792 tried_min_extra_isize++;
2793 new_extra_isize = s_min_extra_isize;
2799 /* Adjust the offsets and shift the remaining entries ahead */
2800 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2801 - new_extra_isize, (void *)raw_inode +
2802 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2803 (void *)header, total_ino);
2804 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2806 if (ext4_has_inline_data(inode))
2807 error = ext4_find_inline_data_nolock(inode);
2810 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2811 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2813 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2818 #define EIA_INCR 16 /* must be 2^n */
2819 #define EIA_MASK (EIA_INCR - 1)
2821 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2822 * If @ea_inode_array is new or full it will be grown and the old
2823 * contents copied over.
2826 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2827 struct inode *inode)
2829 if (*ea_inode_array == NULL) {
2831 * Start with 15 inodes, so it fits into a power-of-two size.
2832 * If *ea_inode_array is NULL, this is essentially offsetof()
2835 kmalloc(offsetof(struct ext4_xattr_inode_array,
2838 if (*ea_inode_array == NULL)
2840 (*ea_inode_array)->count = 0;
2841 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2842 /* expand the array once all 15 + n * 16 slots are full */
2843 struct ext4_xattr_inode_array *new_array = NULL;
2844 int count = (*ea_inode_array)->count;
2846 /* if new_array is NULL, this is essentially offsetof() */
2847 new_array = kmalloc(
2848 offsetof(struct ext4_xattr_inode_array,
2849 inodes[count + EIA_INCR]),
2851 if (new_array == NULL)
2853 memcpy(new_array, *ea_inode_array,
2854 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2855 kfree(*ea_inode_array);
2856 *ea_inode_array = new_array;
2858 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2863 * ext4_xattr_delete_inode()
2865 * Free extended attribute resources associated with this inode. Traverse
2866 * all entries and decrement reference on any xattr inodes associated with this
2867 * inode. This is called immediately before an inode is freed. We have exclusive
2868 * access to the inode. If an orphan inode is deleted it will also release its
2869 * references on xattr block and xattr inodes.
2871 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2872 struct ext4_xattr_inode_array **ea_inode_array,
2875 struct buffer_head *bh = NULL;
2876 struct ext4_xattr_ibody_header *header;
2877 struct ext4_iloc iloc = { .bh = NULL };
2878 struct ext4_xattr_entry *entry;
2879 struct inode *ea_inode;
2882 error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2885 false /* block_csum */);
2887 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2891 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2892 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2894 error = ext4_get_inode_loc(inode, &iloc);
2896 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2900 error = ext4_journal_get_write_access(handle, iloc.bh);
2902 EXT4_ERROR_INODE(inode, "write access (error %d)",
2907 header = IHDR(inode, ext4_raw_inode(&iloc));
2908 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2909 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2911 false /* block_csum */,
2914 false /* skip_quota */);
2917 if (EXT4_I(inode)->i_file_acl) {
2918 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2920 error = PTR_ERR(bh);
2922 EXT4_ERROR_INODE(inode, "block %llu read error",
2923 EXT4_I(inode)->i_file_acl);
2927 error = ext4_xattr_check_block(inode, bh);
2931 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2932 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2933 entry = EXT4_XATTR_NEXT(entry)) {
2934 if (!entry->e_value_inum)
2936 error = ext4_xattr_inode_iget(inode,
2937 le32_to_cpu(entry->e_value_inum),
2938 le32_to_cpu(entry->e_hash),
2942 ext4_xattr_inode_free_quota(inode, ea_inode,
2943 le32_to_cpu(entry->e_value_size));
2949 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2952 * Update i_file_acl value in the same transaction that releases
2955 EXT4_I(inode)->i_file_acl = 0;
2956 error = ext4_mark_inode_dirty(handle, inode);
2958 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2970 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2974 if (ea_inode_array == NULL)
2977 for (idx = 0; idx < ea_inode_array->count; ++idx)
2978 iput(ea_inode_array->inodes[idx]);
2979 kfree(ea_inode_array);
2983 * ext4_xattr_block_cache_insert()
2985 * Create a new entry in the extended attribute block cache, and insert
2986 * it unless such an entry is already in the cache.
2988 * Returns 0, or a negative error number on failure.
2991 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2992 struct buffer_head *bh)
2994 struct ext4_xattr_header *header = BHDR(bh);
2995 __u32 hash = le32_to_cpu(header->h_hash);
2996 int reusable = le32_to_cpu(header->h_refcount) <
2997 EXT4_XATTR_REFCOUNT_MAX;
3000 if (!ea_block_cache)
3002 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3003 bh->b_blocknr, reusable);
3005 if (error == -EBUSY)
3006 ea_bdebug(bh, "already in cache");
3008 ea_bdebug(bh, "inserting [%x]", (int)hash);
3014 * Compare two extended attribute blocks for equality.
3016 * Returns 0 if the blocks are equal, 1 if they differ, and
3017 * a negative error number on errors.
3020 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3021 struct ext4_xattr_header *header2)
3023 struct ext4_xattr_entry *entry1, *entry2;
3025 entry1 = ENTRY(header1+1);
3026 entry2 = ENTRY(header2+1);
3027 while (!IS_LAST_ENTRY(entry1)) {
3028 if (IS_LAST_ENTRY(entry2))
3030 if (entry1->e_hash != entry2->e_hash ||
3031 entry1->e_name_index != entry2->e_name_index ||
3032 entry1->e_name_len != entry2->e_name_len ||
3033 entry1->e_value_size != entry2->e_value_size ||
3034 entry1->e_value_inum != entry2->e_value_inum ||
3035 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3037 if (!entry1->e_value_inum &&
3038 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3039 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3040 le32_to_cpu(entry1->e_value_size)))
3043 entry1 = EXT4_XATTR_NEXT(entry1);
3044 entry2 = EXT4_XATTR_NEXT(entry2);
3046 if (!IS_LAST_ENTRY(entry2))
3052 * ext4_xattr_block_cache_find()
3054 * Find an identical extended attribute block.
3056 * Returns a pointer to the block found, or NULL if such a block was
3057 * not found or an error occurred.
3059 static struct buffer_head *
3060 ext4_xattr_block_cache_find(struct inode *inode,
3061 struct ext4_xattr_header *header,
3062 struct mb_cache_entry **pce)
3064 __u32 hash = le32_to_cpu(header->h_hash);
3065 struct mb_cache_entry *ce;
3066 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3068 if (!ea_block_cache)
3070 if (!header->h_hash)
3071 return NULL; /* never share */
3072 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3073 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3075 struct buffer_head *bh;
3077 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3079 if (PTR_ERR(bh) == -ENOMEM)
3082 EXT4_ERROR_INODE(inode, "block %lu read error",
3083 (unsigned long)ce->e_value);
3084 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3089 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3094 #define NAME_HASH_SHIFT 5
3095 #define VALUE_HASH_SHIFT 16
3098 * ext4_xattr_hash_entry()
3100 * Compute the hash of an extended attribute.
3102 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3107 while (name_len--) {
3108 hash = (hash << NAME_HASH_SHIFT) ^
3109 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3112 while (value_count--) {
3113 hash = (hash << VALUE_HASH_SHIFT) ^
3114 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3115 le32_to_cpu(*value++);
3117 return cpu_to_le32(hash);
3120 #undef NAME_HASH_SHIFT
3121 #undef VALUE_HASH_SHIFT
3123 #define BLOCK_HASH_SHIFT 16
3126 * ext4_xattr_rehash()
3128 * Re-compute the extended attribute hash value after an entry has changed.
3130 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3132 struct ext4_xattr_entry *here;
3135 here = ENTRY(header+1);
3136 while (!IS_LAST_ENTRY(here)) {
3137 if (!here->e_hash) {
3138 /* Block is not shared if an entry's hash value == 0 */
3142 hash = (hash << BLOCK_HASH_SHIFT) ^
3143 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3144 le32_to_cpu(here->e_hash);
3145 here = EXT4_XATTR_NEXT(here);
3147 header->h_hash = cpu_to_le32(hash);
3150 #undef BLOCK_HASH_SHIFT
3152 #define HASH_BUCKET_BITS 10
3155 ext4_xattr_create_cache(void)
3157 return mb_cache_create(HASH_BUCKET_BITS);
3160 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3163 mb_cache_destroy(cache);