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_NORMAL);
404 err = PTR_ERR(inode);
405 ext4_error(parent->i_sb,
406 "error while reading EA inode %lu err=%d", ea_ino,
411 if (is_bad_inode(inode)) {
412 ext4_error(parent->i_sb,
413 "error while reading EA inode %lu is_bad_inode",
419 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
420 ext4_error(parent->i_sb,
421 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
427 ext4_xattr_inode_set_class(inode);
430 * Check whether this is an old Lustre-style xattr inode. Lustre
431 * implementation does not have hash validation, rather it has a
432 * backpointer from ea_inode to the parent inode.
434 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
435 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
436 inode->i_generation == parent->i_generation) {
437 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
438 ext4_xattr_inode_set_ref(inode, 1);
441 inode->i_flags |= S_NOQUOTA;
452 /* Remove entry from mbcache when EA inode is getting evicted */
453 void ext4_evict_ea_inode(struct inode *inode)
455 struct mb_cache_entry *oe;
457 if (!EA_INODE_CACHE(inode))
459 /* Wait for entry to get unused so that we can remove it */
460 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
461 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
462 mb_cache_entry_wait_unused(oe);
463 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
468 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
469 struct ext4_xattr_entry *entry, void *buffer,
474 /* Verify stored hash matches calculated hash. */
475 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
476 if (hash != ext4_xattr_inode_get_hash(ea_inode))
477 return -EFSCORRUPTED;
480 __le32 e_hash, tmp_data;
482 /* Verify entry hash. */
483 tmp_data = cpu_to_le32(hash);
484 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
486 if (e_hash != entry->e_hash)
487 return -EFSCORRUPTED;
493 * Read xattr value from the EA inode.
496 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
497 void *buffer, size_t size)
499 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
500 struct inode *ea_inode;
503 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
504 le32_to_cpu(entry->e_hash), &ea_inode);
510 if (i_size_read(ea_inode) != size) {
511 ext4_warning_inode(ea_inode,
512 "ea_inode file size=%llu entry size=%zu",
513 i_size_read(ea_inode), size);
518 err = ext4_xattr_inode_read(ea_inode, buffer, size);
522 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
523 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
526 ext4_warning_inode(ea_inode,
527 "EA inode hash validation failed");
532 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
533 ext4_xattr_inode_get_hash(ea_inode),
534 ea_inode->i_ino, true /* reusable */);
542 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
543 void *buffer, size_t buffer_size)
545 struct buffer_head *bh = NULL;
546 struct ext4_xattr_entry *entry;
550 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
552 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
553 name_index, name, buffer, (long)buffer_size);
555 if (!EXT4_I(inode)->i_file_acl)
557 ea_idebug(inode, "reading block %llu",
558 (unsigned long long)EXT4_I(inode)->i_file_acl);
559 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
562 ea_bdebug(bh, "b_count=%d, refcount=%d",
563 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
564 error = ext4_xattr_check_block(inode, bh);
567 ext4_xattr_block_cache_insert(ea_block_cache, bh);
569 end = bh->b_data + bh->b_size;
570 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
573 size = le32_to_cpu(entry->e_value_size);
575 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
578 if (size > buffer_size)
580 if (entry->e_value_inum) {
581 error = ext4_xattr_inode_get(inode, entry, buffer,
586 u16 offset = le16_to_cpu(entry->e_value_offs);
587 void *p = bh->b_data + offset;
589 if (unlikely(p + size > end))
591 memcpy(buffer, p, size);
602 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
603 void *buffer, size_t buffer_size)
605 struct ext4_xattr_ibody_header *header;
606 struct ext4_xattr_entry *entry;
607 struct ext4_inode *raw_inode;
608 struct ext4_iloc iloc;
613 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
615 error = ext4_get_inode_loc(inode, &iloc);
618 raw_inode = ext4_raw_inode(&iloc);
619 header = IHDR(inode, raw_inode);
620 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
621 error = xattr_check_inode(inode, header, end);
624 entry = IFIRST(header);
625 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
628 size = le32_to_cpu(entry->e_value_size);
630 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
633 if (size > buffer_size)
635 if (entry->e_value_inum) {
636 error = ext4_xattr_inode_get(inode, entry, buffer,
641 u16 offset = le16_to_cpu(entry->e_value_offs);
642 void *p = (void *)IFIRST(header) + offset;
644 if (unlikely(p + size > end))
646 memcpy(buffer, p, size);
659 * Copy an extended attribute into the buffer
660 * provided, or compute the buffer size required.
661 * Buffer is NULL to compute the size of the buffer required.
663 * Returns a negative error number on failure, or the number of bytes
664 * used / required on success.
667 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
668 void *buffer, size_t buffer_size)
672 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
675 if (strlen(name) > 255)
678 down_read(&EXT4_I(inode)->xattr_sem);
679 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
681 if (error == -ENODATA)
682 error = ext4_xattr_block_get(inode, name_index, name, buffer,
684 up_read(&EXT4_I(inode)->xattr_sem);
689 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
690 char *buffer, size_t buffer_size)
692 size_t rest = buffer_size;
694 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
695 const struct xattr_handler *handler =
696 ext4_xattr_handler(entry->e_name_index);
698 if (handler && (!handler->list || handler->list(dentry))) {
699 const char *prefix = handler->prefix ?: handler->name;
700 size_t prefix_len = strlen(prefix);
701 size_t size = prefix_len + entry->e_name_len + 1;
706 memcpy(buffer, prefix, prefix_len);
707 buffer += prefix_len;
708 memcpy(buffer, entry->e_name, entry->e_name_len);
709 buffer += entry->e_name_len;
715 return buffer_size - rest; /* total size */
719 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
721 struct inode *inode = d_inode(dentry);
722 struct buffer_head *bh = NULL;
725 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
726 buffer, (long)buffer_size);
728 if (!EXT4_I(inode)->i_file_acl)
730 ea_idebug(inode, "reading block %llu",
731 (unsigned long long)EXT4_I(inode)->i_file_acl);
732 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
735 ea_bdebug(bh, "b_count=%d, refcount=%d",
736 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
737 error = ext4_xattr_check_block(inode, bh);
740 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
741 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
749 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
751 struct inode *inode = d_inode(dentry);
752 struct ext4_xattr_ibody_header *header;
753 struct ext4_inode *raw_inode;
754 struct ext4_iloc iloc;
758 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
760 error = ext4_get_inode_loc(inode, &iloc);
763 raw_inode = ext4_raw_inode(&iloc);
764 header = IHDR(inode, raw_inode);
765 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
766 error = xattr_check_inode(inode, header, end);
769 error = ext4_xattr_list_entries(dentry, IFIRST(header),
770 buffer, buffer_size);
778 * Inode operation listxattr()
780 * d_inode(dentry)->i_rwsem: don't care
782 * Copy a list of attribute names into the buffer
783 * provided, or compute the buffer size required.
784 * Buffer is NULL to compute the size of the buffer required.
786 * Returns a negative error number on failure, or the number of bytes
787 * used / required on success.
790 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
794 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
795 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
802 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
807 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
812 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
815 static void ext4_xattr_update_super_block(handle_t *handle,
816 struct super_block *sb)
818 if (ext4_has_feature_xattr(sb))
821 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
822 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
823 ext4_set_feature_xattr(sb);
824 ext4_handle_dirty_super(handle, sb);
828 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
830 struct ext4_iloc iloc = { .bh = NULL };
831 struct buffer_head *bh = NULL;
832 struct ext4_inode *raw_inode;
833 struct ext4_xattr_ibody_header *header;
834 struct ext4_xattr_entry *entry;
835 qsize_t ea_inode_refs = 0;
839 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
841 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
842 ret = ext4_get_inode_loc(inode, &iloc);
845 raw_inode = ext4_raw_inode(&iloc);
846 header = IHDR(inode, raw_inode);
847 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
848 ret = xattr_check_inode(inode, header, end);
852 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
853 entry = EXT4_XATTR_NEXT(entry))
854 if (entry->e_value_inum)
858 if (EXT4_I(inode)->i_file_acl) {
859 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
866 ret = ext4_xattr_check_block(inode, bh);
870 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
871 entry = EXT4_XATTR_NEXT(entry))
872 if (entry->e_value_inum)
875 *usage = ea_inode_refs + 1;
883 static inline size_t round_up_cluster(struct inode *inode, size_t length)
885 struct super_block *sb = inode->i_sb;
886 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
888 size_t mask = ~(cluster_size - 1);
890 return (length + cluster_size - 1) & mask;
893 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
897 err = dquot_alloc_inode(inode);
900 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
902 dquot_free_inode(inode);
906 static void ext4_xattr_inode_free_quota(struct inode *parent,
907 struct inode *ea_inode,
911 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
913 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
914 dquot_free_inode(parent);
917 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
918 struct buffer_head *block_bh, size_t value_len,
925 * 1) Owner inode update
926 * 2) Ref count update on old xattr block
928 * 4) block bitmap update for new xattr block
929 * 5) group descriptor for new xattr block
930 * 6) block bitmap update for old xattr block
931 * 7) group descriptor for old block
933 * 6 & 7 can happen if we have two racing threads T_a and T_b
934 * which are each trying to set an xattr on inodes I_a and I_b
935 * which were both initially sharing an xattr block.
940 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
943 * In case of inline data, we may push out the data to a block,
944 * so we need to reserve credits for this eventuality
946 if (inode && ext4_has_inline_data(inode))
947 credits += ext4_writepage_trans_blocks(inode) + 1;
949 /* We are done if ea_inode feature is not enabled. */
950 if (!ext4_has_feature_ea_inode(sb))
953 /* New ea_inode, inode map, block bitmap, group descriptor. */
957 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
959 /* Indirection block or one level of extent tree. */
962 /* Block bitmap and group descriptor updates for each block. */
963 credits += blocks * 2;
965 /* Blocks themselves. */
969 /* Dereference ea_inode holding old xattr value.
970 * Old ea_inode, inode map, block bitmap, group descriptor.
974 /* Data blocks for old ea_inode. */
975 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
977 /* Indirection block or one level of extent tree for old
982 /* Block bitmap and group descriptor updates for each block. */
983 credits += blocks * 2;
986 /* We may need to clone the existing xattr block in which case we need
987 * to increment ref counts for existing ea_inodes referenced by it.
990 struct ext4_xattr_entry *entry = BFIRST(block_bh);
992 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
993 if (entry->e_value_inum)
994 /* Ref count update on ea_inode. */
1000 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
1001 int credits, struct buffer_head *bh,
1002 bool dirty, bool block_csum)
1006 if (!ext4_handle_valid(handle))
1009 if (handle->h_buffer_credits >= credits)
1012 error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
1016 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
1022 ext4_xattr_block_csum_set(inode, bh);
1023 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1025 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1031 error = ext4_journal_restart(handle, credits);
1033 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1038 error = ext4_journal_get_write_access(handle, bh);
1040 ext4_warning(inode->i_sb,
1041 "Get write access failed (error %d)",
1049 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1052 struct ext4_iloc iloc;
1056 inode_lock(ea_inode);
1058 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1064 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1065 ref_count += ref_change;
1066 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1068 if (ref_change > 0) {
1069 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1070 ea_inode->i_ino, ref_count);
1072 if (ref_count == 1) {
1073 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1074 ea_inode->i_ino, ea_inode->i_nlink);
1076 set_nlink(ea_inode, 1);
1077 ext4_orphan_del(handle, ea_inode);
1080 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1081 ea_inode->i_ino, ref_count);
1083 if (ref_count == 0) {
1084 WARN_ONCE(ea_inode->i_nlink != 1,
1085 "EA inode %lu i_nlink=%u",
1086 ea_inode->i_ino, ea_inode->i_nlink);
1088 clear_nlink(ea_inode);
1089 ext4_orphan_add(handle, ea_inode);
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);
1272 hash = le32_to_cpu(BHDR(bh)->h_hash);
1273 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1275 ea_bdebug(bh, "refcount now=0; freeing");
1277 * This must happen under buffer lock for
1278 * ext4_xattr_block_set() to reliably detect freed block
1280 if (ea_block_cache) {
1281 struct mb_cache_entry *oe;
1283 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1287 mb_cache_entry_wait_unused(oe);
1288 mb_cache_entry_put(ea_block_cache, oe);
1295 if (ext4_has_feature_ea_inode(inode->i_sb))
1296 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1298 true /* block_csum */,
1301 true /* skip_quota */);
1302 ext4_free_blocks(handle, inode, bh, 0, 1,
1303 EXT4_FREE_BLOCKS_METADATA |
1304 EXT4_FREE_BLOCKS_FORGET);
1307 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1308 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1309 struct mb_cache_entry *ce;
1311 if (ea_block_cache) {
1312 ce = mb_cache_entry_get(ea_block_cache, hash,
1315 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1316 mb_cache_entry_put(ea_block_cache, ce);
1321 ext4_xattr_block_csum_set(inode, bh);
1323 * Beware of this ugliness: Releasing of xattr block references
1324 * from different inodes can race and so we have to protect
1325 * from a race where someone else frees the block (and releases
1326 * its journal_head) before we are done dirtying the buffer. In
1327 * nojournal mode this race is harmless and we actually cannot
1328 * call ext4_handle_dirty_metadata() with locked buffer as
1329 * that function can call sync_dirty_buffer() so for that case
1330 * we handle the dirtying after unlocking the buffer.
1332 if (ext4_handle_valid(handle))
1333 error = ext4_handle_dirty_metadata(handle, inode, bh);
1335 if (!ext4_handle_valid(handle))
1336 error = ext4_handle_dirty_metadata(handle, inode, bh);
1338 ext4_handle_sync(handle);
1339 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1340 ea_bdebug(bh, "refcount now=%d; releasing",
1341 le32_to_cpu(BHDR(bh)->h_refcount));
1344 ext4_std_error(inode->i_sb, error);
1349 * Find the available free space for EAs. This also returns the total number of
1350 * bytes used by EA entries.
1352 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1353 size_t *min_offs, void *base, int *total)
1355 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1356 if (!last->e_value_inum && last->e_value_size) {
1357 size_t offs = le16_to_cpu(last->e_value_offs);
1358 if (offs < *min_offs)
1362 *total += EXT4_XATTR_LEN(last->e_name_len);
1364 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1368 * Write the value of the EA in an inode.
1370 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1371 const void *buf, int bufsize)
1373 struct buffer_head *bh = NULL;
1374 unsigned long block = 0;
1375 int blocksize = ea_inode->i_sb->s_blocksize;
1376 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1377 int csize, wsize = 0;
1382 while (ret >= 0 && ret < max_blocks) {
1383 struct ext4_map_blocks map;
1384 map.m_lblk = block += ret;
1385 map.m_len = max_blocks -= ret;
1387 ret = ext4_map_blocks(handle, ea_inode, &map,
1388 EXT4_GET_BLOCKS_CREATE);
1390 ext4_mark_inode_dirty(handle, ea_inode);
1391 if (ret == -ENOSPC &&
1392 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1404 while (wsize < bufsize) {
1407 csize = (bufsize - wsize) > blocksize ? blocksize :
1409 bh = ext4_getblk(handle, ea_inode, block, 0);
1414 EXT4_ERROR_INODE(ea_inode,
1415 "ext4_getblk() return bh = NULL");
1416 return -EFSCORRUPTED;
1418 ret = ext4_journal_get_write_access(handle, bh);
1422 memcpy(bh->b_data, buf, csize);
1423 set_buffer_uptodate(bh);
1424 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1431 inode_lock(ea_inode);
1432 i_size_write(ea_inode, wsize);
1433 ext4_update_i_disksize(ea_inode, wsize);
1434 inode_unlock(ea_inode);
1436 ext4_mark_inode_dirty(handle, ea_inode);
1445 * Create an inode to store the value of a large EA.
1447 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1448 struct inode *inode, u32 hash)
1450 struct inode *ea_inode = NULL;
1451 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1454 if (inode->i_sb->s_root == NULL) {
1455 ext4_warning(inode->i_sb,
1456 "refuse to create EA inode when umounting");
1458 return ERR_PTR(-EINVAL);
1462 * Let the next inode be the goal, so we try and allocate the EA inode
1463 * in the same group, or nearby one.
1465 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1466 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1468 if (!IS_ERR(ea_inode)) {
1469 ea_inode->i_op = &ext4_file_inode_operations;
1470 ea_inode->i_fop = &ext4_file_operations;
1471 ext4_set_aops(ea_inode);
1472 ext4_xattr_inode_set_class(ea_inode);
1473 unlock_new_inode(ea_inode);
1474 ext4_xattr_inode_set_ref(ea_inode, 1);
1475 ext4_xattr_inode_set_hash(ea_inode, hash);
1476 err = ext4_mark_inode_dirty(handle, ea_inode);
1478 err = ext4_inode_attach_jinode(ea_inode);
1480 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1481 ext4_warning_inode(ea_inode,
1482 "cleanup dec ref error %d", err);
1484 return ERR_PTR(err);
1488 * Xattr inodes are shared therefore quota charging is performed
1489 * at a higher level.
1491 dquot_free_inode(ea_inode);
1492 dquot_drop(ea_inode);
1493 inode_lock(ea_inode);
1494 ea_inode->i_flags |= S_NOQUOTA;
1495 inode_unlock(ea_inode);
1501 static struct inode *
1502 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1503 size_t value_len, u32 hash)
1505 struct inode *ea_inode;
1506 struct mb_cache_entry *ce;
1507 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1510 if (!ea_inode_cache)
1513 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1517 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1518 !(current->flags & PF_MEMALLOC_NOFS));
1520 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1522 mb_cache_entry_put(ea_inode_cache, ce);
1527 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1529 if (!IS_ERR(ea_inode) &&
1530 !is_bad_inode(ea_inode) &&
1531 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1532 i_size_read(ea_inode) == value_len &&
1533 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1534 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1536 !memcmp(value, ea_data, value_len)) {
1537 mb_cache_entry_touch(ea_inode_cache, ce);
1538 mb_cache_entry_put(ea_inode_cache, ce);
1543 if (!IS_ERR(ea_inode))
1545 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1552 * Add value of the EA in an inode.
1554 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1555 const void *value, size_t value_len,
1556 struct inode **ret_inode)
1558 struct inode *ea_inode;
1562 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1563 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1565 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1571 *ret_inode = ea_inode;
1575 /* Create an inode for the EA value */
1576 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1577 if (IS_ERR(ea_inode))
1578 return PTR_ERR(ea_inode);
1580 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1582 ext4_xattr_inode_dec_ref(handle, ea_inode);
1587 if (EA_INODE_CACHE(inode))
1588 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1589 ea_inode->i_ino, true /* reusable */);
1591 *ret_inode = ea_inode;
1596 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1597 * feature is enabled.
1599 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1601 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1602 struct ext4_xattr_search *s,
1603 handle_t *handle, struct inode *inode,
1606 struct ext4_xattr_entry *last, *next;
1607 struct ext4_xattr_entry *here = s->here;
1608 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1609 int in_inode = i->in_inode;
1610 struct inode *old_ea_inode = NULL;
1611 struct inode *new_ea_inode = NULL;
1612 size_t old_size, new_size;
1615 /* Space used by old and new values. */
1616 old_size = (!s->not_found && !here->e_value_inum) ?
1617 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1618 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1621 * Optimization for the simple case when old and new values have the
1622 * same padded sizes. Not applicable if external inodes are involved.
1624 if (new_size && new_size == old_size) {
1625 size_t offs = le16_to_cpu(here->e_value_offs);
1626 void *val = s->base + offs;
1628 here->e_value_size = cpu_to_le32(i->value_len);
1629 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1630 memset(val, 0, new_size);
1632 memcpy(val, i->value, i->value_len);
1633 /* Clear padding bytes. */
1634 memset(val + i->value_len, 0, new_size - i->value_len);
1639 /* Compute min_offs and last. */
1641 for (; !IS_LAST_ENTRY(last); last = next) {
1642 next = EXT4_XATTR_NEXT(last);
1643 if ((void *)next >= s->end) {
1644 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1645 ret = -EFSCORRUPTED;
1648 if (!last->e_value_inum && last->e_value_size) {
1649 size_t offs = le16_to_cpu(last->e_value_offs);
1650 if (offs < min_offs)
1655 /* Check whether we have enough space. */
1659 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1661 free += EXT4_XATTR_LEN(name_len) + old_size;
1663 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1669 * If storing the value in an external inode is an option,
1670 * reserve space for xattr entries/names in the external
1671 * attribute block so that a long value does not occupy the
1672 * whole space and prevent futher entries being added.
1674 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1675 new_size && is_block &&
1676 (min_offs + old_size - new_size) <
1677 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1684 * Getting access to old and new ea inodes is subject to failures.
1685 * Finish that work before doing any modifications to the xattr data.
1687 if (!s->not_found && here->e_value_inum) {
1688 ret = ext4_xattr_inode_iget(inode,
1689 le32_to_cpu(here->e_value_inum),
1690 le32_to_cpu(here->e_hash),
1693 old_ea_inode = NULL;
1697 if (i->value && in_inode) {
1698 WARN_ON_ONCE(!i->value_len);
1700 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1704 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1708 new_ea_inode = NULL;
1709 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1715 /* We are ready to release ref count on the old_ea_inode. */
1716 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1718 /* Release newly required ref count on new_ea_inode. */
1722 err = ext4_xattr_inode_dec_ref(handle,
1725 ext4_warning_inode(new_ea_inode,
1726 "dec ref new_ea_inode err=%d",
1728 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1734 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1735 le32_to_cpu(here->e_value_size));
1738 /* No failures allowed past this point. */
1740 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1741 /* Remove the old value. */
1742 void *first_val = s->base + min_offs;
1743 size_t offs = le16_to_cpu(here->e_value_offs);
1744 void *val = s->base + offs;
1746 memmove(first_val + old_size, first_val, val - first_val);
1747 memset(first_val, 0, old_size);
1748 min_offs += old_size;
1750 /* Adjust all value offsets. */
1752 while (!IS_LAST_ENTRY(last)) {
1753 size_t o = le16_to_cpu(last->e_value_offs);
1755 if (!last->e_value_inum &&
1756 last->e_value_size && o < offs)
1757 last->e_value_offs = cpu_to_le16(o + old_size);
1758 last = EXT4_XATTR_NEXT(last);
1763 /* Remove old name. */
1764 size_t size = EXT4_XATTR_LEN(name_len);
1766 last = ENTRY((void *)last - size);
1767 memmove(here, (void *)here + size,
1768 (void *)last - (void *)here + sizeof(__u32));
1769 memset(last, 0, size);
1770 } else if (s->not_found) {
1771 /* Insert new name. */
1772 size_t size = EXT4_XATTR_LEN(name_len);
1773 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1775 memmove((void *)here + size, here, rest);
1776 memset(here, 0, size);
1777 here->e_name_index = i->name_index;
1778 here->e_name_len = name_len;
1779 memcpy(here->e_name, i->name, name_len);
1781 /* This is an update, reset value info. */
1782 here->e_value_inum = 0;
1783 here->e_value_offs = 0;
1784 here->e_value_size = 0;
1788 /* Insert new value. */
1790 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1791 } else if (i->value_len) {
1792 void *val = s->base + min_offs - new_size;
1794 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1795 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1796 memset(val, 0, new_size);
1798 memcpy(val, i->value, i->value_len);
1799 /* Clear padding bytes. */
1800 memset(val + i->value_len, 0,
1801 new_size - i->value_len);
1804 here->e_value_size = cpu_to_le32(i->value_len);
1811 /* Entry hash calculation. */
1816 * Feed crc32c hash instead of the raw value for entry
1817 * hash calculation. This is to avoid walking
1818 * potentially long value buffer again.
1820 crc32c_hash = cpu_to_le32(
1821 ext4_xattr_inode_get_hash(new_ea_inode));
1822 hash = ext4_xattr_hash_entry(here->e_name,
1825 } else if (is_block) {
1826 __le32 *value = s->base + le16_to_cpu(
1827 here->e_value_offs);
1829 hash = ext4_xattr_hash_entry(here->e_name,
1830 here->e_name_len, value,
1833 here->e_hash = hash;
1837 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1846 struct ext4_xattr_block_find {
1847 struct ext4_xattr_search s;
1848 struct buffer_head *bh;
1852 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1853 struct ext4_xattr_block_find *bs)
1855 struct super_block *sb = inode->i_sb;
1858 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1859 i->name_index, i->name, i->value, (long)i->value_len);
1861 if (EXT4_I(inode)->i_file_acl) {
1862 /* The inode already has an extended attribute block. */
1863 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1864 if (IS_ERR(bs->bh)) {
1865 error = PTR_ERR(bs->bh);
1869 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1870 atomic_read(&(bs->bh->b_count)),
1871 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1872 error = ext4_xattr_check_block(inode, bs->bh);
1875 /* Find the named attribute. */
1876 bs->s.base = BHDR(bs->bh);
1877 bs->s.first = BFIRST(bs->bh);
1878 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1879 bs->s.here = bs->s.first;
1880 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1881 i->name_index, i->name, 1);
1882 if (error && error != -ENODATA)
1884 bs->s.not_found = error;
1890 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1891 struct ext4_xattr_info *i,
1892 struct ext4_xattr_block_find *bs)
1894 struct super_block *sb = inode->i_sb;
1895 struct buffer_head *new_bh = NULL;
1896 struct ext4_xattr_search s_copy = bs->s;
1897 struct ext4_xattr_search *s = &s_copy;
1898 struct mb_cache_entry *ce = NULL;
1900 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1901 struct inode *ea_inode = NULL, *tmp_inode;
1902 size_t old_ea_inode_quota = 0;
1903 unsigned int ea_ino;
1906 #define header(x) ((struct ext4_xattr_header *)(x))
1909 int offset = (char *)s->here - bs->bh->b_data;
1911 BUFFER_TRACE(bs->bh, "get_write_access");
1912 error = ext4_journal_get_write_access(handle, bs->bh);
1915 lock_buffer(bs->bh);
1917 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1918 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1921 * This must happen under buffer lock for
1922 * ext4_xattr_block_set() to reliably detect modified
1925 if (ea_block_cache) {
1926 struct mb_cache_entry *oe;
1928 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1929 hash, bs->bh->b_blocknr);
1932 * Xattr block is getting reused. Leave
1935 mb_cache_entry_put(ea_block_cache, oe);
1939 ea_bdebug(bs->bh, "modifying in-place");
1940 error = ext4_xattr_set_entry(i, s, handle, inode,
1941 true /* is_block */);
1942 ext4_xattr_block_csum_set(inode, bs->bh);
1943 unlock_buffer(bs->bh);
1944 if (error == -EFSCORRUPTED)
1947 error = ext4_handle_dirty_metadata(handle,
1955 unlock_buffer(bs->bh);
1956 ea_bdebug(bs->bh, "cloning");
1957 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1959 if (s->base == NULL)
1961 s->first = ENTRY(header(s->base)+1);
1962 header(s->base)->h_refcount = cpu_to_le32(1);
1963 s->here = ENTRY(s->base + offset);
1964 s->end = s->base + bs->bh->b_size;
1967 * If existing entry points to an xattr inode, we need
1968 * to prevent ext4_xattr_set_entry() from decrementing
1969 * ref count on it because the reference belongs to the
1970 * original block. In this case, make the entry look
1971 * like it has an empty value.
1973 if (!s->not_found && s->here->e_value_inum) {
1974 ea_ino = le32_to_cpu(s->here->e_value_inum);
1975 error = ext4_xattr_inode_iget(inode, ea_ino,
1976 le32_to_cpu(s->here->e_hash),
1981 if (!ext4_test_inode_state(tmp_inode,
1982 EXT4_STATE_LUSTRE_EA_INODE)) {
1984 * Defer quota free call for previous
1985 * inode until success is guaranteed.
1987 old_ea_inode_quota = le32_to_cpu(
1988 s->here->e_value_size);
1992 s->here->e_value_inum = 0;
1993 s->here->e_value_size = 0;
1996 /* Allocate a buffer where we construct the new block. */
1997 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1998 /* assert(header == s->base) */
2000 if (s->base == NULL)
2002 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2003 header(s->base)->h_blocks = cpu_to_le32(1);
2004 header(s->base)->h_refcount = cpu_to_le32(1);
2005 s->first = ENTRY(header(s->base)+1);
2006 s->here = ENTRY(header(s->base)+1);
2007 s->end = s->base + sb->s_blocksize;
2010 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
2011 if (error == -EFSCORRUPTED)
2016 if (i->value && s->here->e_value_inum) {
2018 * A ref count on ea_inode has been taken as part of the call to
2019 * ext4_xattr_set_entry() above. We would like to drop this
2020 * extra ref but we have to wait until the xattr block is
2021 * initialized and has its own ref count on the ea_inode.
2023 ea_ino = le32_to_cpu(s->here->e_value_inum);
2024 error = ext4_xattr_inode_iget(inode, ea_ino,
2025 le32_to_cpu(s->here->e_hash),
2034 if (!IS_LAST_ENTRY(s->first)) {
2035 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2038 /* We found an identical block in the cache. */
2039 if (new_bh == bs->bh)
2040 ea_bdebug(new_bh, "keeping");
2044 #ifdef EXT4_XATTR_DEBUG
2045 WARN_ON_ONCE(dquot_initialize_needed(inode));
2047 /* The old block is released after updating
2049 error = dquot_alloc_block(inode,
2050 EXT4_C2B(EXT4_SB(sb), 1));
2053 BUFFER_TRACE(new_bh, "get_write_access");
2054 error = ext4_journal_get_write_access(handle,
2058 lock_buffer(new_bh);
2060 * We have to be careful about races with
2061 * adding references to xattr block. Once we
2062 * hold buffer lock xattr block's state is
2063 * stable so we can check the additional
2066 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2067 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2069 * Undo everything and check mbcache
2072 unlock_buffer(new_bh);
2073 dquot_free_block(inode,
2074 EXT4_C2B(EXT4_SB(sb),
2077 mb_cache_entry_put(ea_block_cache, ce);
2082 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2083 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2084 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2085 ea_bdebug(new_bh, "reusing; refcount now=%d",
2087 ext4_xattr_block_csum_set(inode, new_bh);
2088 unlock_buffer(new_bh);
2089 error = ext4_handle_dirty_metadata(handle,
2095 mb_cache_entry_touch(ea_block_cache, ce);
2096 mb_cache_entry_put(ea_block_cache, ce);
2098 } else if (bs->bh && s->base == bs->bh->b_data) {
2099 /* We were modifying this block in-place. */
2100 ea_bdebug(bs->bh, "keeping this block");
2101 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2105 /* We need to allocate a new block */
2106 ext4_fsblk_t goal, block;
2108 #ifdef EXT4_XATTR_DEBUG
2109 WARN_ON_ONCE(dquot_initialize_needed(inode));
2111 goal = ext4_group_first_block_no(sb,
2112 EXT4_I(inode)->i_block_group);
2113 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2118 ea_idebug(inode, "creating block %llu",
2119 (unsigned long long)block);
2121 new_bh = sb_getblk(sb, block);
2122 if (unlikely(!new_bh)) {
2125 ext4_free_blocks(handle, inode, NULL, block, 1,
2126 EXT4_FREE_BLOCKS_METADATA);
2129 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2130 ENTRY(header(s->base)+1));
2134 /* Drop the extra ref on ea_inode. */
2135 error = ext4_xattr_inode_dec_ref(handle,
2138 ext4_warning_inode(ea_inode,
2145 lock_buffer(new_bh);
2146 error = ext4_journal_get_create_access(handle, new_bh);
2148 unlock_buffer(new_bh);
2152 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2153 ext4_xattr_block_csum_set(inode, new_bh);
2154 set_buffer_uptodate(new_bh);
2155 unlock_buffer(new_bh);
2156 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2157 error = ext4_handle_dirty_metadata(handle, inode,
2164 if (old_ea_inode_quota)
2165 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2167 /* Update the inode. */
2168 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2170 /* Drop the previous xattr block. */
2171 if (bs->bh && bs->bh != new_bh) {
2172 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2174 ext4_xattr_release_block(handle, inode, bs->bh,
2176 0 /* extra_credits */);
2177 ext4_xattr_inode_array_free(ea_inode_array);
2185 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2187 ext4_warning_inode(ea_inode, "dec ref error=%d",
2190 /* If there was an error, revert the quota charge. */
2192 ext4_xattr_inode_free_quota(inode, ea_inode,
2193 i_size_read(ea_inode));
2197 mb_cache_entry_put(ea_block_cache, ce);
2199 if (!(bs->bh && s->base == bs->bh->b_data))
2205 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2209 EXT4_ERROR_INODE(inode, "bad block %llu",
2210 EXT4_I(inode)->i_file_acl);
2216 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2217 struct ext4_xattr_ibody_find *is)
2219 struct ext4_xattr_ibody_header *header;
2220 struct ext4_inode *raw_inode;
2223 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2226 raw_inode = ext4_raw_inode(&is->iloc);
2227 header = IHDR(inode, raw_inode);
2228 is->s.base = is->s.first = IFIRST(header);
2229 is->s.here = is->s.first;
2230 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2231 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2232 error = xattr_check_inode(inode, header, is->s.end);
2235 /* Find the named attribute. */
2236 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2237 i->name_index, i->name, 0);
2238 if (error && error != -ENODATA)
2240 is->s.not_found = error;
2245 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2246 struct ext4_xattr_info *i,
2247 struct ext4_xattr_ibody_find *is)
2249 struct ext4_xattr_ibody_header *header;
2250 struct ext4_xattr_search *s = &is->s;
2253 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2256 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2259 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2260 if (!IS_LAST_ENTRY(s->first)) {
2261 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2262 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2264 header->h_magic = cpu_to_le32(0);
2265 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2270 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2271 struct ext4_xattr_info *i)
2275 /* When e_value_inum is set the value is stored externally. */
2276 if (s->here->e_value_inum)
2278 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2280 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2281 return !memcmp(value, i->value, i->value_len);
2284 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2286 struct buffer_head *bh;
2289 if (!EXT4_I(inode)->i_file_acl)
2291 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2294 error = ext4_xattr_check_block(inode, bh);
2297 return ERR_PTR(error);
2303 * ext4_xattr_set_handle()
2305 * Create, replace or remove an extended attribute for this inode. Value
2306 * is NULL to remove an existing extended attribute, and non-NULL to
2307 * either replace an existing extended attribute, or create a new extended
2308 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2309 * specify that an extended attribute must exist and must not exist
2310 * previous to the call, respectively.
2312 * Returns 0, or a negative error number on failure.
2315 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2316 const char *name, const void *value, size_t value_len,
2319 struct ext4_xattr_info i = {
2320 .name_index = name_index,
2323 .value_len = value_len,
2326 struct ext4_xattr_ibody_find is = {
2327 .s = { .not_found = -ENODATA, },
2329 struct ext4_xattr_block_find bs = {
2330 .s = { .not_found = -ENODATA, },
2337 if (strlen(name) > 255)
2340 ext4_write_lock_xattr(inode, &no_expand);
2342 /* Check journal credits under write lock. */
2343 if (ext4_handle_valid(handle)) {
2344 struct buffer_head *bh;
2347 bh = ext4_xattr_get_block(inode);
2349 error = PTR_ERR(bh);
2353 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2355 flags & XATTR_CREATE);
2358 if (!ext4_handle_has_enough_credits(handle, credits)) {
2362 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2365 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2369 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2370 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2371 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2372 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2375 error = ext4_xattr_ibody_find(inode, &i, &is);
2379 error = ext4_xattr_block_find(inode, &i, &bs);
2382 if (is.s.not_found && bs.s.not_found) {
2384 if (flags & XATTR_REPLACE)
2391 if (flags & XATTR_CREATE)
2396 if (!is.s.not_found)
2397 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2398 else if (!bs.s.not_found)
2399 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2402 /* Xattr value did not change? Save us some work and bail out */
2403 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2405 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2408 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2409 (EXT4_XATTR_SIZE(i.value_len) >
2410 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2413 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2414 if (!error && !bs.s.not_found) {
2416 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2417 } else if (error == -ENOSPC) {
2418 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2421 error = ext4_xattr_block_find(inode, &i, &bs);
2425 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2426 if (!error && !is.s.not_found) {
2428 error = ext4_xattr_ibody_set(handle, inode, &i,
2430 } else if (error == -ENOSPC) {
2432 * Xattr does not fit in the block, store at
2433 * external inode if possible.
2435 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2436 i.value_len && !i.in_inode) {
2444 ext4_xattr_update_super_block(handle, inode->i_sb);
2445 inode->i_ctime = current_time(inode);
2448 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2450 * The bh is consumed by ext4_mark_iloc_dirty, even with
2455 ext4_handle_sync(handle);
2461 ext4_write_unlock_xattr(inode, &no_expand);
2465 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2466 bool is_create, int *credits)
2468 struct buffer_head *bh;
2473 if (!EXT4_SB(inode->i_sb)->s_journal)
2476 down_read(&EXT4_I(inode)->xattr_sem);
2478 bh = ext4_xattr_get_block(inode);
2482 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2483 value_len, is_create);
2488 up_read(&EXT4_I(inode)->xattr_sem);
2495 * Like ext4_xattr_set_handle, but start from an inode. This extended
2496 * attribute modification is a filesystem transaction by itself.
2498 * Returns 0, or a negative error number on failure.
2501 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2502 const void *value, size_t value_len, int flags)
2505 struct super_block *sb = inode->i_sb;
2506 int error, retries = 0;
2509 error = dquot_initialize(inode);
2514 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2519 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2520 if (IS_ERR(handle)) {
2521 error = PTR_ERR(handle);
2525 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2526 value, value_len, flags);
2527 error2 = ext4_journal_stop(handle);
2528 if (error == -ENOSPC &&
2529 ext4_should_retry_alloc(sb, &retries))
2539 * Shift the EA entries in the inode to create space for the increased
2542 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2543 int value_offs_shift, void *to,
2544 void *from, size_t n)
2546 struct ext4_xattr_entry *last = entry;
2549 /* We always shift xattr headers further thus offsets get lower */
2550 BUG_ON(value_offs_shift > 0);
2552 /* Adjust the value offsets of the entries */
2553 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2554 if (!last->e_value_inum && last->e_value_size) {
2555 new_offs = le16_to_cpu(last->e_value_offs) +
2557 last->e_value_offs = cpu_to_le16(new_offs);
2560 /* Shift the entries by n bytes */
2561 memmove(to, from, n);
2565 * Move xattr pointed to by 'entry' from inode into external xattr block
2567 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2568 struct ext4_inode *raw_inode,
2569 struct ext4_xattr_entry *entry)
2571 struct ext4_xattr_ibody_find *is = NULL;
2572 struct ext4_xattr_block_find *bs = NULL;
2573 char *buffer = NULL, *b_entry_name = NULL;
2574 size_t value_size = le32_to_cpu(entry->e_value_size);
2575 struct ext4_xattr_info i = {
2578 .name_index = entry->e_name_index,
2579 .in_inode = !!entry->e_value_inum,
2581 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2582 int needs_kvfree = 0;
2585 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2586 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2587 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2588 if (!is || !bs || !b_entry_name) {
2593 is->s.not_found = -ENODATA;
2594 bs->s.not_found = -ENODATA;
2598 /* Save the entry name and the entry value */
2599 if (entry->e_value_inum) {
2600 buffer = kvmalloc(value_size, GFP_NOFS);
2606 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2610 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2611 buffer = (void *)IFIRST(header) + value_offs;
2614 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2615 b_entry_name[entry->e_name_len] = '\0';
2616 i.name = b_entry_name;
2618 error = ext4_get_inode_loc(inode, &is->iloc);
2622 error = ext4_xattr_ibody_find(inode, &i, is);
2627 i.value_len = value_size;
2628 error = ext4_xattr_block_find(inode, &i, bs);
2632 /* Move ea entry from the inode into the block */
2633 error = ext4_xattr_block_set(handle, inode, &i, bs);
2637 /* Remove the chosen entry from the inode */
2640 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2643 kfree(b_entry_name);
2644 if (needs_kvfree && buffer)
2647 brelse(is->iloc.bh);
2656 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2657 struct ext4_inode *raw_inode,
2658 int isize_diff, size_t ifree,
2659 size_t bfree, int *total_ino)
2661 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2662 struct ext4_xattr_entry *small_entry;
2663 struct ext4_xattr_entry *entry;
2664 struct ext4_xattr_entry *last;
2665 unsigned int entry_size; /* EA entry size */
2666 unsigned int total_size; /* EA entry size + value size */
2667 unsigned int min_total_size;
2670 while (isize_diff > ifree) {
2673 min_total_size = ~0U;
2674 last = IFIRST(header);
2675 /* Find the entry best suited to be pushed into EA block */
2676 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2677 /* never move system.data out of the inode */
2678 if ((last->e_name_len == 4) &&
2679 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2680 !memcmp(last->e_name, "data", 4))
2682 total_size = EXT4_XATTR_LEN(last->e_name_len);
2683 if (!last->e_value_inum)
2684 total_size += EXT4_XATTR_SIZE(
2685 le32_to_cpu(last->e_value_size));
2686 if (total_size <= bfree &&
2687 total_size < min_total_size) {
2688 if (total_size + ifree < isize_diff) {
2692 min_total_size = total_size;
2697 if (entry == NULL) {
2698 if (small_entry == NULL)
2700 entry = small_entry;
2703 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2704 total_size = entry_size;
2705 if (!entry->e_value_inum)
2706 total_size += EXT4_XATTR_SIZE(
2707 le32_to_cpu(entry->e_value_size));
2708 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2713 *total_ino -= entry_size;
2714 ifree += total_size;
2715 bfree -= total_size;
2722 * Expand an inode by new_extra_isize bytes when EAs are present.
2723 * Returns 0 on success or negative error number on failure.
2725 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2726 struct ext4_inode *raw_inode, handle_t *handle)
2728 struct ext4_xattr_ibody_header *header;
2729 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2730 static unsigned int mnt_count;
2732 size_t ifree, bfree;
2735 int error = 0, tried_min_extra_isize = 0;
2736 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2737 int isize_diff; /* How much do we need to grow i_extra_isize */
2740 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2741 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2744 header = IHDR(inode, raw_inode);
2747 * Check if enough free space is available in the inode to shift the
2748 * entries ahead by new_extra_isize.
2751 base = IFIRST(header);
2752 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2753 min_offs = end - base;
2754 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2756 error = xattr_check_inode(inode, header, end);
2760 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2761 if (ifree >= isize_diff)
2765 * Enough free space isn't available in the inode, check if
2766 * EA block can hold new_extra_isize bytes.
2768 if (EXT4_I(inode)->i_file_acl) {
2769 struct buffer_head *bh;
2771 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2773 error = PTR_ERR(bh);
2776 error = ext4_xattr_check_block(inode, bh);
2782 end = bh->b_data + bh->b_size;
2783 min_offs = end - base;
2784 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2787 if (bfree + ifree < isize_diff) {
2788 if (!tried_min_extra_isize && s_min_extra_isize) {
2789 tried_min_extra_isize++;
2790 new_extra_isize = s_min_extra_isize;
2797 bfree = inode->i_sb->s_blocksize;
2800 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2801 isize_diff, ifree, bfree,
2804 if (error == -ENOSPC && !tried_min_extra_isize &&
2805 s_min_extra_isize) {
2806 tried_min_extra_isize++;
2807 new_extra_isize = s_min_extra_isize;
2813 /* Adjust the offsets and shift the remaining entries ahead */
2814 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2815 - new_extra_isize, (void *)raw_inode +
2816 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2817 (void *)header, total_ino);
2818 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2820 if (ext4_has_inline_data(inode))
2821 error = ext4_find_inline_data_nolock(inode);
2824 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2825 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2827 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2832 #define EIA_INCR 16 /* must be 2^n */
2833 #define EIA_MASK (EIA_INCR - 1)
2835 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2836 * If @ea_inode_array is new or full it will be grown and the old
2837 * contents copied over.
2840 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2841 struct inode *inode)
2843 if (*ea_inode_array == NULL) {
2845 * Start with 15 inodes, so it fits into a power-of-two size.
2846 * If *ea_inode_array is NULL, this is essentially offsetof()
2849 kmalloc(offsetof(struct ext4_xattr_inode_array,
2852 if (*ea_inode_array == NULL)
2854 (*ea_inode_array)->count = 0;
2855 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2856 /* expand the array once all 15 + n * 16 slots are full */
2857 struct ext4_xattr_inode_array *new_array = NULL;
2858 int count = (*ea_inode_array)->count;
2860 /* if new_array is NULL, this is essentially offsetof() */
2861 new_array = kmalloc(
2862 offsetof(struct ext4_xattr_inode_array,
2863 inodes[count + EIA_INCR]),
2865 if (new_array == NULL)
2867 memcpy(new_array, *ea_inode_array,
2868 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2869 kfree(*ea_inode_array);
2870 *ea_inode_array = new_array;
2872 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2877 * ext4_xattr_delete_inode()
2879 * Free extended attribute resources associated with this inode. Traverse
2880 * all entries and decrement reference on any xattr inodes associated with this
2881 * inode. This is called immediately before an inode is freed. We have exclusive
2882 * access to the inode. If an orphan inode is deleted it will also release its
2883 * references on xattr block and xattr inodes.
2885 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2886 struct ext4_xattr_inode_array **ea_inode_array,
2889 struct buffer_head *bh = NULL;
2890 struct ext4_xattr_ibody_header *header;
2891 struct ext4_iloc iloc = { .bh = NULL };
2892 struct ext4_xattr_entry *entry;
2893 struct inode *ea_inode;
2896 error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2899 false /* block_csum */);
2901 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2905 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2906 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2908 error = ext4_get_inode_loc(inode, &iloc);
2910 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2914 error = ext4_journal_get_write_access(handle, iloc.bh);
2916 EXT4_ERROR_INODE(inode, "write access (error %d)",
2921 header = IHDR(inode, ext4_raw_inode(&iloc));
2922 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2923 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2925 false /* block_csum */,
2928 false /* skip_quota */);
2931 if (EXT4_I(inode)->i_file_acl) {
2932 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2934 error = PTR_ERR(bh);
2936 EXT4_ERROR_INODE(inode, "block %llu read error",
2937 EXT4_I(inode)->i_file_acl);
2941 error = ext4_xattr_check_block(inode, bh);
2945 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2946 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2947 entry = EXT4_XATTR_NEXT(entry)) {
2948 if (!entry->e_value_inum)
2950 error = ext4_xattr_inode_iget(inode,
2951 le32_to_cpu(entry->e_value_inum),
2952 le32_to_cpu(entry->e_hash),
2956 ext4_xattr_inode_free_quota(inode, ea_inode,
2957 le32_to_cpu(entry->e_value_size));
2963 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2966 * Update i_file_acl value in the same transaction that releases
2969 EXT4_I(inode)->i_file_acl = 0;
2970 error = ext4_mark_inode_dirty(handle, inode);
2972 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2984 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2988 if (ea_inode_array == NULL)
2991 for (idx = 0; idx < ea_inode_array->count; ++idx)
2992 iput(ea_inode_array->inodes[idx]);
2993 kfree(ea_inode_array);
2997 * ext4_xattr_block_cache_insert()
2999 * Create a new entry in the extended attribute block cache, and insert
3000 * it unless such an entry is already in the cache.
3002 * Returns 0, or a negative error number on failure.
3005 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3006 struct buffer_head *bh)
3008 struct ext4_xattr_header *header = BHDR(bh);
3009 __u32 hash = le32_to_cpu(header->h_hash);
3010 int reusable = le32_to_cpu(header->h_refcount) <
3011 EXT4_XATTR_REFCOUNT_MAX;
3014 if (!ea_block_cache)
3016 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3017 bh->b_blocknr, reusable);
3019 if (error == -EBUSY)
3020 ea_bdebug(bh, "already in cache");
3022 ea_bdebug(bh, "inserting [%x]", (int)hash);
3028 * Compare two extended attribute blocks for equality.
3030 * Returns 0 if the blocks are equal, 1 if they differ, and
3031 * a negative error number on errors.
3034 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3035 struct ext4_xattr_header *header2)
3037 struct ext4_xattr_entry *entry1, *entry2;
3039 entry1 = ENTRY(header1+1);
3040 entry2 = ENTRY(header2+1);
3041 while (!IS_LAST_ENTRY(entry1)) {
3042 if (IS_LAST_ENTRY(entry2))
3044 if (entry1->e_hash != entry2->e_hash ||
3045 entry1->e_name_index != entry2->e_name_index ||
3046 entry1->e_name_len != entry2->e_name_len ||
3047 entry1->e_value_size != entry2->e_value_size ||
3048 entry1->e_value_inum != entry2->e_value_inum ||
3049 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3051 if (!entry1->e_value_inum &&
3052 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3053 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3054 le32_to_cpu(entry1->e_value_size)))
3057 entry1 = EXT4_XATTR_NEXT(entry1);
3058 entry2 = EXT4_XATTR_NEXT(entry2);
3060 if (!IS_LAST_ENTRY(entry2))
3066 * ext4_xattr_block_cache_find()
3068 * Find an identical extended attribute block.
3070 * Returns a pointer to the block found, or NULL if such a block was
3071 * not found or an error occurred.
3073 static struct buffer_head *
3074 ext4_xattr_block_cache_find(struct inode *inode,
3075 struct ext4_xattr_header *header,
3076 struct mb_cache_entry **pce)
3078 __u32 hash = le32_to_cpu(header->h_hash);
3079 struct mb_cache_entry *ce;
3080 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3082 if (!ea_block_cache)
3084 if (!header->h_hash)
3085 return NULL; /* never share */
3086 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3087 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3089 struct buffer_head *bh;
3091 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3093 if (PTR_ERR(bh) == -ENOMEM)
3096 EXT4_ERROR_INODE(inode, "block %lu read error",
3097 (unsigned long)ce->e_value);
3098 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3103 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3108 #define NAME_HASH_SHIFT 5
3109 #define VALUE_HASH_SHIFT 16
3112 * ext4_xattr_hash_entry()
3114 * Compute the hash of an extended attribute.
3116 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3121 while (name_len--) {
3122 hash = (hash << NAME_HASH_SHIFT) ^
3123 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3126 while (value_count--) {
3127 hash = (hash << VALUE_HASH_SHIFT) ^
3128 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3129 le32_to_cpu(*value++);
3131 return cpu_to_le32(hash);
3134 #undef NAME_HASH_SHIFT
3135 #undef VALUE_HASH_SHIFT
3137 #define BLOCK_HASH_SHIFT 16
3140 * ext4_xattr_rehash()
3142 * Re-compute the extended attribute hash value after an entry has changed.
3144 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3146 struct ext4_xattr_entry *here;
3149 here = ENTRY(header+1);
3150 while (!IS_LAST_ENTRY(here)) {
3151 if (!here->e_hash) {
3152 /* Block is not shared if an entry's hash value == 0 */
3156 hash = (hash << BLOCK_HASH_SHIFT) ^
3157 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3158 le32_to_cpu(here->e_hash);
3159 here = EXT4_XATTR_NEXT(here);
3161 header->h_hash = cpu_to_le32(hash);
3164 #undef BLOCK_HASH_SHIFT
3166 #define HASH_BUCKET_BITS 10
3169 ext4_xattr_create_cache(void)
3171 return mb_cache_create(HASH_BUCKET_BITS);
3174 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3177 mb_cache_destroy(cache);