crypto: cavium/nitrox - no need to check return value of debugfs_create functions
[linux/fpc-iii.git] / fs / ext4 / xattr.c
blob86ed9c6862493eb676515ce61cf8d13c94ca4be3
1 // SPDX-License-Identifier: GPL-2.0
2 /*
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>,
12 * Red Hat Inc.
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 * +------------------+
30 * | header |
31 * | entry 1 | |
32 * | entry 2 | | growing downwards
33 * | entry 3 | v
34 * | four null bytes |
35 * | . . . |
36 * | value 1 | ^
37 * | value 3 | | growing upwards
38 * | value 2 | |
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.
45 * Locking strategy
46 * ----------------
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
51 * by the buffer lock.
54 #include <linux/init.h>
55 #include <linux/fs.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"
61 #include "ext4.h"
62 #include "xattr.h"
63 #include "acl.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__)
72 #else
73 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
74 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
75 #endif
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,
83 size_t value_count);
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,
91 #endif
92 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
93 #ifdef CONFIG_EXT4_FS_SECURITY
94 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
95 #endif
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,
104 #endif
105 #ifdef CONFIG_EXT4_FS_SECURITY
106 &ext4_xattr_security_handler,
107 #endif
108 NULL
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)
117 static int
118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119 struct inode *inode);
121 #ifdef CONFIG_LOCKDEP
122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
124 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
126 #endif
128 static __le32 ext4_xattr_block_csum(struct inode *inode,
129 sector_t block_nr,
130 struct ext4_xattr_header *hdr)
132 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
133 __u32 csum;
134 __le64 dsk_block_nr = cpu_to_le64(block_nr);
135 __u32 dummy_csum = 0;
136 int offset = offsetof(struct ext4_xattr_header, h_checksum);
138 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
139 sizeof(dsk_block_nr));
140 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
141 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
142 offset += sizeof(dummy_csum);
143 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
144 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
146 return cpu_to_le32(csum);
149 static int ext4_xattr_block_csum_verify(struct inode *inode,
150 struct buffer_head *bh)
152 struct ext4_xattr_header *hdr = BHDR(bh);
153 int ret = 1;
155 if (ext4_has_metadata_csum(inode->i_sb)) {
156 lock_buffer(bh);
157 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
158 bh->b_blocknr, hdr));
159 unlock_buffer(bh);
161 return ret;
164 static void ext4_xattr_block_csum_set(struct inode *inode,
165 struct buffer_head *bh)
167 if (ext4_has_metadata_csum(inode->i_sb))
168 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
169 bh->b_blocknr, BHDR(bh));
172 static inline const struct xattr_handler *
173 ext4_xattr_handler(int name_index)
175 const struct xattr_handler *handler = NULL;
177 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
178 handler = ext4_xattr_handler_map[name_index];
179 return handler;
182 static int
183 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
184 void *value_start)
186 struct ext4_xattr_entry *e = entry;
188 /* Find the end of the names list */
189 while (!IS_LAST_ENTRY(e)) {
190 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
191 if ((void *)next >= end)
192 return -EFSCORRUPTED;
193 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
194 return -EFSCORRUPTED;
195 e = next;
198 /* Check the values */
199 while (!IS_LAST_ENTRY(entry)) {
200 u32 size = le32_to_cpu(entry->e_value_size);
202 if (size > EXT4_XATTR_SIZE_MAX)
203 return -EFSCORRUPTED;
205 if (size != 0 && entry->e_value_inum == 0) {
206 u16 offs = le16_to_cpu(entry->e_value_offs);
207 void *value;
210 * The value cannot overlap the names, and the value
211 * with padding cannot extend beyond 'end'. Check both
212 * the padded and unpadded sizes, since the size may
213 * overflow to 0 when adding padding.
215 if (offs > end - value_start)
216 return -EFSCORRUPTED;
217 value = value_start + offs;
218 if (value < (void *)e + sizeof(u32) ||
219 size > end - value ||
220 EXT4_XATTR_SIZE(size) > end - value)
221 return -EFSCORRUPTED;
223 entry = EXT4_XATTR_NEXT(entry);
226 return 0;
229 static inline int
230 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
231 const char *function, unsigned int line)
233 int error = -EFSCORRUPTED;
235 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
236 BHDR(bh)->h_blocks != cpu_to_le32(1))
237 goto errout;
238 if (buffer_verified(bh))
239 return 0;
241 error = -EFSBADCRC;
242 if (!ext4_xattr_block_csum_verify(inode, bh))
243 goto errout;
244 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
245 bh->b_data);
246 errout:
247 if (error)
248 __ext4_error_inode(inode, function, line, 0,
249 "corrupted xattr block %llu",
250 (unsigned long long) bh->b_blocknr);
251 else
252 set_buffer_verified(bh);
253 return error;
256 #define ext4_xattr_check_block(inode, bh) \
257 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
260 static int
261 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
262 void *end, const char *function, unsigned int line)
264 int error = -EFSCORRUPTED;
266 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
267 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
268 goto errout;
269 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
270 errout:
271 if (error)
272 __ext4_error_inode(inode, function, line, 0,
273 "corrupted in-inode xattr");
274 return error;
277 #define xattr_check_inode(inode, header, end) \
278 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
280 static int
281 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
282 void *end, int name_index, const char *name, int sorted)
284 struct ext4_xattr_entry *entry, *next;
285 size_t name_len;
286 int cmp = 1;
288 if (name == NULL)
289 return -EINVAL;
290 name_len = strlen(name);
291 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
292 next = EXT4_XATTR_NEXT(entry);
293 if ((void *) next >= end) {
294 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
295 return -EFSCORRUPTED;
297 cmp = name_index - entry->e_name_index;
298 if (!cmp)
299 cmp = name_len - entry->e_name_len;
300 if (!cmp)
301 cmp = memcmp(name, entry->e_name, name_len);
302 if (cmp <= 0 && (sorted || cmp == 0))
303 break;
305 *pentry = entry;
306 return cmp ? -ENODATA : 0;
309 static u32
310 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
312 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
315 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
317 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
318 (u32) inode_peek_iversion_raw(ea_inode);
321 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
323 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
324 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
327 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
329 return (u32)ea_inode->i_atime.tv_sec;
332 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
334 ea_inode->i_atime.tv_sec = hash;
338 * Read the EA value from an inode.
340 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
342 int blocksize = 1 << ea_inode->i_blkbits;
343 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
344 int tail_size = (size % blocksize) ?: blocksize;
345 struct buffer_head *bhs_inline[8];
346 struct buffer_head **bhs = bhs_inline;
347 int i, ret;
349 if (bh_count > ARRAY_SIZE(bhs_inline)) {
350 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
351 if (!bhs)
352 return -ENOMEM;
355 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
356 true /* wait */, bhs);
357 if (ret)
358 goto free_bhs;
360 for (i = 0; i < bh_count; i++) {
361 /* There shouldn't be any holes in ea_inode. */
362 if (!bhs[i]) {
363 ret = -EFSCORRUPTED;
364 goto put_bhs;
366 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
367 i < bh_count - 1 ? blocksize : tail_size);
369 ret = 0;
370 put_bhs:
371 for (i = 0; i < bh_count; i++)
372 brelse(bhs[i]);
373 free_bhs:
374 if (bhs != bhs_inline)
375 kfree(bhs);
376 return ret;
379 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
381 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
382 u32 ea_inode_hash, struct inode **ea_inode)
384 struct inode *inode;
385 int err;
387 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
388 if (IS_ERR(inode)) {
389 err = PTR_ERR(inode);
390 ext4_error(parent->i_sb,
391 "error while reading EA inode %lu err=%d", ea_ino,
392 err);
393 return err;
396 if (is_bad_inode(inode)) {
397 ext4_error(parent->i_sb,
398 "error while reading EA inode %lu is_bad_inode",
399 ea_ino);
400 err = -EIO;
401 goto error;
404 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
405 ext4_error(parent->i_sb,
406 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
407 ea_ino);
408 err = -EINVAL;
409 goto error;
412 ext4_xattr_inode_set_class(inode);
415 * Check whether this is an old Lustre-style xattr inode. Lustre
416 * implementation does not have hash validation, rather it has a
417 * backpointer from ea_inode to the parent inode.
419 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
420 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
421 inode->i_generation == parent->i_generation) {
422 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
423 ext4_xattr_inode_set_ref(inode, 1);
424 } else {
425 inode_lock(inode);
426 inode->i_flags |= S_NOQUOTA;
427 inode_unlock(inode);
430 *ea_inode = inode;
431 return 0;
432 error:
433 iput(inode);
434 return err;
437 static int
438 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
439 struct ext4_xattr_entry *entry, void *buffer,
440 size_t size)
442 u32 hash;
444 /* Verify stored hash matches calculated hash. */
445 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
446 if (hash != ext4_xattr_inode_get_hash(ea_inode))
447 return -EFSCORRUPTED;
449 if (entry) {
450 __le32 e_hash, tmp_data;
452 /* Verify entry hash. */
453 tmp_data = cpu_to_le32(hash);
454 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
455 &tmp_data, 1);
456 if (e_hash != entry->e_hash)
457 return -EFSCORRUPTED;
459 return 0;
463 * Read xattr value from the EA inode.
465 static int
466 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
467 void *buffer, size_t size)
469 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
470 struct inode *ea_inode;
471 int err;
473 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
474 le32_to_cpu(entry->e_hash), &ea_inode);
475 if (err) {
476 ea_inode = NULL;
477 goto out;
480 if (i_size_read(ea_inode) != size) {
481 ext4_warning_inode(ea_inode,
482 "ea_inode file size=%llu entry size=%zu",
483 i_size_read(ea_inode), size);
484 err = -EFSCORRUPTED;
485 goto out;
488 err = ext4_xattr_inode_read(ea_inode, buffer, size);
489 if (err)
490 goto out;
492 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
493 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
494 size);
495 if (err) {
496 ext4_warning_inode(ea_inode,
497 "EA inode hash validation failed");
498 goto out;
501 if (ea_inode_cache)
502 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
503 ext4_xattr_inode_get_hash(ea_inode),
504 ea_inode->i_ino, true /* reusable */);
506 out:
507 iput(ea_inode);
508 return err;
511 static int
512 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
513 void *buffer, size_t buffer_size)
515 struct buffer_head *bh = NULL;
516 struct ext4_xattr_entry *entry;
517 size_t size;
518 void *end;
519 int error;
520 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
522 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
523 name_index, name, buffer, (long)buffer_size);
525 if (!EXT4_I(inode)->i_file_acl)
526 return -ENODATA;
527 ea_idebug(inode, "reading block %llu",
528 (unsigned long long)EXT4_I(inode)->i_file_acl);
529 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
530 if (IS_ERR(bh))
531 return PTR_ERR(bh);
532 ea_bdebug(bh, "b_count=%d, refcount=%d",
533 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
534 error = ext4_xattr_check_block(inode, bh);
535 if (error)
536 goto cleanup;
537 ext4_xattr_block_cache_insert(ea_block_cache, bh);
538 entry = BFIRST(bh);
539 end = bh->b_data + bh->b_size;
540 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
541 if (error)
542 goto cleanup;
543 size = le32_to_cpu(entry->e_value_size);
544 error = -ERANGE;
545 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
546 goto cleanup;
547 if (buffer) {
548 if (size > buffer_size)
549 goto cleanup;
550 if (entry->e_value_inum) {
551 error = ext4_xattr_inode_get(inode, entry, buffer,
552 size);
553 if (error)
554 goto cleanup;
555 } else {
556 u16 offset = le16_to_cpu(entry->e_value_offs);
557 void *p = bh->b_data + offset;
559 if (unlikely(p + size > end))
560 goto cleanup;
561 memcpy(buffer, p, size);
564 error = size;
566 cleanup:
567 brelse(bh);
568 return error;
572 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
573 void *buffer, size_t buffer_size)
575 struct ext4_xattr_ibody_header *header;
576 struct ext4_xattr_entry *entry;
577 struct ext4_inode *raw_inode;
578 struct ext4_iloc iloc;
579 size_t size;
580 void *end;
581 int error;
583 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
584 return -ENODATA;
585 error = ext4_get_inode_loc(inode, &iloc);
586 if (error)
587 return error;
588 raw_inode = ext4_raw_inode(&iloc);
589 header = IHDR(inode, raw_inode);
590 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
591 error = xattr_check_inode(inode, header, end);
592 if (error)
593 goto cleanup;
594 entry = IFIRST(header);
595 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
596 if (error)
597 goto cleanup;
598 size = le32_to_cpu(entry->e_value_size);
599 error = -ERANGE;
600 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
601 goto cleanup;
602 if (buffer) {
603 if (size > buffer_size)
604 goto cleanup;
605 if (entry->e_value_inum) {
606 error = ext4_xattr_inode_get(inode, entry, buffer,
607 size);
608 if (error)
609 goto cleanup;
610 } else {
611 u16 offset = le16_to_cpu(entry->e_value_offs);
612 void *p = (void *)IFIRST(header) + offset;
614 if (unlikely(p + size > end))
615 goto cleanup;
616 memcpy(buffer, p, size);
619 error = size;
621 cleanup:
622 brelse(iloc.bh);
623 return error;
627 * ext4_xattr_get()
629 * Copy an extended attribute into the buffer
630 * provided, or compute the buffer size required.
631 * Buffer is NULL to compute the size of the buffer required.
633 * Returns a negative error number on failure, or the number of bytes
634 * used / required on success.
637 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
638 void *buffer, size_t buffer_size)
640 int error;
642 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
643 return -EIO;
645 if (strlen(name) > 255)
646 return -ERANGE;
648 down_read(&EXT4_I(inode)->xattr_sem);
649 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
650 buffer_size);
651 if (error == -ENODATA)
652 error = ext4_xattr_block_get(inode, name_index, name, buffer,
653 buffer_size);
654 up_read(&EXT4_I(inode)->xattr_sem);
655 return error;
658 static int
659 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
660 char *buffer, size_t buffer_size)
662 size_t rest = buffer_size;
664 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
665 const struct xattr_handler *handler =
666 ext4_xattr_handler(entry->e_name_index);
668 if (handler && (!handler->list || handler->list(dentry))) {
669 const char *prefix = handler->prefix ?: handler->name;
670 size_t prefix_len = strlen(prefix);
671 size_t size = prefix_len + entry->e_name_len + 1;
673 if (buffer) {
674 if (size > rest)
675 return -ERANGE;
676 memcpy(buffer, prefix, prefix_len);
677 buffer += prefix_len;
678 memcpy(buffer, entry->e_name, entry->e_name_len);
679 buffer += entry->e_name_len;
680 *buffer++ = 0;
682 rest -= size;
685 return buffer_size - rest; /* total size */
688 static int
689 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
691 struct inode *inode = d_inode(dentry);
692 struct buffer_head *bh = NULL;
693 int error;
695 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
696 buffer, (long)buffer_size);
698 if (!EXT4_I(inode)->i_file_acl)
699 return 0;
700 ea_idebug(inode, "reading block %llu",
701 (unsigned long long)EXT4_I(inode)->i_file_acl);
702 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
703 if (IS_ERR(bh))
704 return PTR_ERR(bh);
705 ea_bdebug(bh, "b_count=%d, refcount=%d",
706 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
707 error = ext4_xattr_check_block(inode, bh);
708 if (error)
709 goto cleanup;
710 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
711 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
712 buffer_size);
713 cleanup:
714 brelse(bh);
715 return error;
718 static int
719 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
721 struct inode *inode = d_inode(dentry);
722 struct ext4_xattr_ibody_header *header;
723 struct ext4_inode *raw_inode;
724 struct ext4_iloc iloc;
725 void *end;
726 int error;
728 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
729 return 0;
730 error = ext4_get_inode_loc(inode, &iloc);
731 if (error)
732 return error;
733 raw_inode = ext4_raw_inode(&iloc);
734 header = IHDR(inode, raw_inode);
735 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
736 error = xattr_check_inode(inode, header, end);
737 if (error)
738 goto cleanup;
739 error = ext4_xattr_list_entries(dentry, IFIRST(header),
740 buffer, buffer_size);
742 cleanup:
743 brelse(iloc.bh);
744 return error;
748 * Inode operation listxattr()
750 * d_inode(dentry)->i_rwsem: don't care
752 * Copy a list of attribute names into the buffer
753 * provided, or compute the buffer size required.
754 * Buffer is NULL to compute the size of the buffer required.
756 * Returns a negative error number on failure, or the number of bytes
757 * used / required on success.
759 ssize_t
760 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
762 int ret, ret2;
764 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
765 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
766 if (ret < 0)
767 goto errout;
768 if (buffer) {
769 buffer += ret;
770 buffer_size -= ret;
772 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
773 if (ret < 0)
774 goto errout;
775 ret += ret2;
776 errout:
777 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
778 return ret;
782 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
783 * not set, set it.
785 static void ext4_xattr_update_super_block(handle_t *handle,
786 struct super_block *sb)
788 if (ext4_has_feature_xattr(sb))
789 return;
791 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
792 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
793 ext4_set_feature_xattr(sb);
794 ext4_handle_dirty_super(handle, sb);
798 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
800 struct ext4_iloc iloc = { .bh = NULL };
801 struct buffer_head *bh = NULL;
802 struct ext4_inode *raw_inode;
803 struct ext4_xattr_ibody_header *header;
804 struct ext4_xattr_entry *entry;
805 qsize_t ea_inode_refs = 0;
806 void *end;
807 int ret;
809 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
811 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
812 ret = ext4_get_inode_loc(inode, &iloc);
813 if (ret)
814 goto out;
815 raw_inode = ext4_raw_inode(&iloc);
816 header = IHDR(inode, raw_inode);
817 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
818 ret = xattr_check_inode(inode, header, end);
819 if (ret)
820 goto out;
822 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
823 entry = EXT4_XATTR_NEXT(entry))
824 if (entry->e_value_inum)
825 ea_inode_refs++;
828 if (EXT4_I(inode)->i_file_acl) {
829 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
830 if (IS_ERR(bh)) {
831 ret = PTR_ERR(bh);
832 goto out;
835 ret = ext4_xattr_check_block(inode, bh);
836 if (ret)
837 goto out;
839 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
840 entry = EXT4_XATTR_NEXT(entry))
841 if (entry->e_value_inum)
842 ea_inode_refs++;
844 *usage = ea_inode_refs + 1;
845 ret = 0;
846 out:
847 brelse(iloc.bh);
848 brelse(bh);
849 return ret;
852 static inline size_t round_up_cluster(struct inode *inode, size_t length)
854 struct super_block *sb = inode->i_sb;
855 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
856 inode->i_blkbits);
857 size_t mask = ~(cluster_size - 1);
859 return (length + cluster_size - 1) & mask;
862 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
864 int err;
866 err = dquot_alloc_inode(inode);
867 if (err)
868 return err;
869 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
870 if (err)
871 dquot_free_inode(inode);
872 return err;
875 static void ext4_xattr_inode_free_quota(struct inode *parent,
876 struct inode *ea_inode,
877 size_t len)
879 if (ea_inode &&
880 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
881 return;
882 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
883 dquot_free_inode(parent);
886 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
887 struct buffer_head *block_bh, size_t value_len,
888 bool is_create)
890 int credits;
891 int blocks;
894 * 1) Owner inode update
895 * 2) Ref count update on old xattr block
896 * 3) new xattr block
897 * 4) block bitmap update for new xattr block
898 * 5) group descriptor for new xattr block
899 * 6) block bitmap update for old xattr block
900 * 7) group descriptor for old block
902 * 6 & 7 can happen if we have two racing threads T_a and T_b
903 * which are each trying to set an xattr on inodes I_a and I_b
904 * which were both initially sharing an xattr block.
906 credits = 7;
908 /* Quota updates. */
909 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
912 * In case of inline data, we may push out the data to a block,
913 * so we need to reserve credits for this eventuality
915 if (inode && ext4_has_inline_data(inode))
916 credits += ext4_writepage_trans_blocks(inode) + 1;
918 /* We are done if ea_inode feature is not enabled. */
919 if (!ext4_has_feature_ea_inode(sb))
920 return credits;
922 /* New ea_inode, inode map, block bitmap, group descriptor. */
923 credits += 4;
925 /* Data blocks. */
926 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
928 /* Indirection block or one level of extent tree. */
929 blocks += 1;
931 /* Block bitmap and group descriptor updates for each block. */
932 credits += blocks * 2;
934 /* Blocks themselves. */
935 credits += blocks;
937 if (!is_create) {
938 /* Dereference ea_inode holding old xattr value.
939 * Old ea_inode, inode map, block bitmap, group descriptor.
941 credits += 4;
943 /* Data blocks for old ea_inode. */
944 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
946 /* Indirection block or one level of extent tree for old
947 * ea_inode.
949 blocks += 1;
951 /* Block bitmap and group descriptor updates for each block. */
952 credits += blocks * 2;
955 /* We may need to clone the existing xattr block in which case we need
956 * to increment ref counts for existing ea_inodes referenced by it.
958 if (block_bh) {
959 struct ext4_xattr_entry *entry = BFIRST(block_bh);
961 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
962 if (entry->e_value_inum)
963 /* Ref count update on ea_inode. */
964 credits += 1;
966 return credits;
969 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
970 int credits, struct buffer_head *bh,
971 bool dirty, bool block_csum)
973 int error;
975 if (!ext4_handle_valid(handle))
976 return 0;
978 if (handle->h_buffer_credits >= credits)
979 return 0;
981 error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
982 if (!error)
983 return 0;
984 if (error < 0) {
985 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
986 return error;
989 if (bh && dirty) {
990 if (block_csum)
991 ext4_xattr_block_csum_set(inode, bh);
992 error = ext4_handle_dirty_metadata(handle, NULL, bh);
993 if (error) {
994 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
995 error);
996 return error;
1000 error = ext4_journal_restart(handle, credits);
1001 if (error) {
1002 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1003 return error;
1006 if (bh) {
1007 error = ext4_journal_get_write_access(handle, bh);
1008 if (error) {
1009 ext4_warning(inode->i_sb,
1010 "Get write access failed (error %d)",
1011 error);
1012 return error;
1015 return 0;
1018 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1019 int ref_change)
1021 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1022 struct ext4_iloc iloc;
1023 s64 ref_count;
1024 u32 hash;
1025 int ret;
1027 inode_lock(ea_inode);
1029 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1030 if (ret)
1031 goto out;
1033 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1034 ref_count += ref_change;
1035 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1037 if (ref_change > 0) {
1038 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1039 ea_inode->i_ino, ref_count);
1041 if (ref_count == 1) {
1042 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1043 ea_inode->i_ino, ea_inode->i_nlink);
1045 set_nlink(ea_inode, 1);
1046 ext4_orphan_del(handle, ea_inode);
1048 if (ea_inode_cache) {
1049 hash = ext4_xattr_inode_get_hash(ea_inode);
1050 mb_cache_entry_create(ea_inode_cache,
1051 GFP_NOFS, hash,
1052 ea_inode->i_ino,
1053 true /* reusable */);
1056 } else {
1057 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1058 ea_inode->i_ino, ref_count);
1060 if (ref_count == 0) {
1061 WARN_ONCE(ea_inode->i_nlink != 1,
1062 "EA inode %lu i_nlink=%u",
1063 ea_inode->i_ino, ea_inode->i_nlink);
1065 clear_nlink(ea_inode);
1066 ext4_orphan_add(handle, ea_inode);
1068 if (ea_inode_cache) {
1069 hash = ext4_xattr_inode_get_hash(ea_inode);
1070 mb_cache_entry_delete(ea_inode_cache, hash,
1071 ea_inode->i_ino);
1076 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1077 if (ret)
1078 ext4_warning_inode(ea_inode,
1079 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1080 out:
1081 inode_unlock(ea_inode);
1082 return ret;
1085 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1087 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1090 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1092 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1095 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1096 struct ext4_xattr_entry *first)
1098 struct inode *ea_inode;
1099 struct ext4_xattr_entry *entry;
1100 struct ext4_xattr_entry *failed_entry;
1101 unsigned int ea_ino;
1102 int err, saved_err;
1104 for (entry = first; !IS_LAST_ENTRY(entry);
1105 entry = EXT4_XATTR_NEXT(entry)) {
1106 if (!entry->e_value_inum)
1107 continue;
1108 ea_ino = le32_to_cpu(entry->e_value_inum);
1109 err = ext4_xattr_inode_iget(parent, ea_ino,
1110 le32_to_cpu(entry->e_hash),
1111 &ea_inode);
1112 if (err)
1113 goto cleanup;
1114 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1115 if (err) {
1116 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1117 iput(ea_inode);
1118 goto cleanup;
1120 iput(ea_inode);
1122 return 0;
1124 cleanup:
1125 saved_err = err;
1126 failed_entry = entry;
1128 for (entry = first; entry != failed_entry;
1129 entry = EXT4_XATTR_NEXT(entry)) {
1130 if (!entry->e_value_inum)
1131 continue;
1132 ea_ino = le32_to_cpu(entry->e_value_inum);
1133 err = ext4_xattr_inode_iget(parent, ea_ino,
1134 le32_to_cpu(entry->e_hash),
1135 &ea_inode);
1136 if (err) {
1137 ext4_warning(parent->i_sb,
1138 "cleanup ea_ino %u iget error %d", ea_ino,
1139 err);
1140 continue;
1142 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1143 if (err)
1144 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1145 err);
1146 iput(ea_inode);
1148 return saved_err;
1151 static void
1152 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1153 struct buffer_head *bh,
1154 struct ext4_xattr_entry *first, bool block_csum,
1155 struct ext4_xattr_inode_array **ea_inode_array,
1156 int extra_credits, bool skip_quota)
1158 struct inode *ea_inode;
1159 struct ext4_xattr_entry *entry;
1160 bool dirty = false;
1161 unsigned int ea_ino;
1162 int err;
1163 int credits;
1165 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1166 credits = 2 + extra_credits;
1168 for (entry = first; !IS_LAST_ENTRY(entry);
1169 entry = EXT4_XATTR_NEXT(entry)) {
1170 if (!entry->e_value_inum)
1171 continue;
1172 ea_ino = le32_to_cpu(entry->e_value_inum);
1173 err = ext4_xattr_inode_iget(parent, ea_ino,
1174 le32_to_cpu(entry->e_hash),
1175 &ea_inode);
1176 if (err)
1177 continue;
1179 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1180 if (err) {
1181 ext4_warning_inode(ea_inode,
1182 "Expand inode array err=%d", err);
1183 iput(ea_inode);
1184 continue;
1187 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1188 dirty, block_csum);
1189 if (err) {
1190 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1191 err);
1192 continue;
1195 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1196 if (err) {
1197 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1198 err);
1199 continue;
1202 if (!skip_quota)
1203 ext4_xattr_inode_free_quota(parent, ea_inode,
1204 le32_to_cpu(entry->e_value_size));
1207 * Forget about ea_inode within the same transaction that
1208 * decrements the ref count. This avoids duplicate decrements in
1209 * case the rest of the work spills over to subsequent
1210 * transactions.
1212 entry->e_value_inum = 0;
1213 entry->e_value_size = 0;
1215 dirty = true;
1218 if (dirty) {
1220 * Note that we are deliberately skipping csum calculation for
1221 * the final update because we do not expect any journal
1222 * restarts until xattr block is freed.
1225 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1226 if (err)
1227 ext4_warning_inode(parent,
1228 "handle dirty metadata err=%d", err);
1233 * Release the xattr block BH: If the reference count is > 1, decrement it;
1234 * otherwise free the block.
1236 static void
1237 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1238 struct buffer_head *bh,
1239 struct ext4_xattr_inode_array **ea_inode_array,
1240 int extra_credits)
1242 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1243 u32 hash, ref;
1244 int error = 0;
1246 BUFFER_TRACE(bh, "get_write_access");
1247 error = ext4_journal_get_write_access(handle, bh);
1248 if (error)
1249 goto out;
1251 lock_buffer(bh);
1252 hash = le32_to_cpu(BHDR(bh)->h_hash);
1253 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1254 if (ref == 1) {
1255 ea_bdebug(bh, "refcount now=0; freeing");
1257 * This must happen under buffer lock for
1258 * ext4_xattr_block_set() to reliably detect freed block
1260 if (ea_block_cache)
1261 mb_cache_entry_delete(ea_block_cache, hash,
1262 bh->b_blocknr);
1263 get_bh(bh);
1264 unlock_buffer(bh);
1266 if (ext4_has_feature_ea_inode(inode->i_sb))
1267 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1268 BFIRST(bh),
1269 true /* block_csum */,
1270 ea_inode_array,
1271 extra_credits,
1272 true /* skip_quota */);
1273 ext4_free_blocks(handle, inode, bh, 0, 1,
1274 EXT4_FREE_BLOCKS_METADATA |
1275 EXT4_FREE_BLOCKS_FORGET);
1276 } else {
1277 ref--;
1278 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1279 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1280 struct mb_cache_entry *ce;
1282 if (ea_block_cache) {
1283 ce = mb_cache_entry_get(ea_block_cache, hash,
1284 bh->b_blocknr);
1285 if (ce) {
1286 ce->e_reusable = 1;
1287 mb_cache_entry_put(ea_block_cache, ce);
1292 ext4_xattr_block_csum_set(inode, bh);
1294 * Beware of this ugliness: Releasing of xattr block references
1295 * from different inodes can race and so we have to protect
1296 * from a race where someone else frees the block (and releases
1297 * its journal_head) before we are done dirtying the buffer. In
1298 * nojournal mode this race is harmless and we actually cannot
1299 * call ext4_handle_dirty_metadata() with locked buffer as
1300 * that function can call sync_dirty_buffer() so for that case
1301 * we handle the dirtying after unlocking the buffer.
1303 if (ext4_handle_valid(handle))
1304 error = ext4_handle_dirty_metadata(handle, inode, bh);
1305 unlock_buffer(bh);
1306 if (!ext4_handle_valid(handle))
1307 error = ext4_handle_dirty_metadata(handle, inode, bh);
1308 if (IS_SYNC(inode))
1309 ext4_handle_sync(handle);
1310 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1311 ea_bdebug(bh, "refcount now=%d; releasing",
1312 le32_to_cpu(BHDR(bh)->h_refcount));
1314 out:
1315 ext4_std_error(inode->i_sb, error);
1316 return;
1320 * Find the available free space for EAs. This also returns the total number of
1321 * bytes used by EA entries.
1323 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1324 size_t *min_offs, void *base, int *total)
1326 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1327 if (!last->e_value_inum && last->e_value_size) {
1328 size_t offs = le16_to_cpu(last->e_value_offs);
1329 if (offs < *min_offs)
1330 *min_offs = offs;
1332 if (total)
1333 *total += EXT4_XATTR_LEN(last->e_name_len);
1335 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1339 * Write the value of the EA in an inode.
1341 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1342 const void *buf, int bufsize)
1344 struct buffer_head *bh = NULL;
1345 unsigned long block = 0;
1346 int blocksize = ea_inode->i_sb->s_blocksize;
1347 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1348 int csize, wsize = 0;
1349 int ret = 0;
1350 int retries = 0;
1352 retry:
1353 while (ret >= 0 && ret < max_blocks) {
1354 struct ext4_map_blocks map;
1355 map.m_lblk = block += ret;
1356 map.m_len = max_blocks -= ret;
1358 ret = ext4_map_blocks(handle, ea_inode, &map,
1359 EXT4_GET_BLOCKS_CREATE);
1360 if (ret <= 0) {
1361 ext4_mark_inode_dirty(handle, ea_inode);
1362 if (ret == -ENOSPC &&
1363 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1364 ret = 0;
1365 goto retry;
1367 break;
1371 if (ret < 0)
1372 return ret;
1374 block = 0;
1375 while (wsize < bufsize) {
1376 if (bh != NULL)
1377 brelse(bh);
1378 csize = (bufsize - wsize) > blocksize ? blocksize :
1379 bufsize - wsize;
1380 bh = ext4_getblk(handle, ea_inode, block, 0);
1381 if (IS_ERR(bh))
1382 return PTR_ERR(bh);
1383 if (!bh) {
1384 WARN_ON_ONCE(1);
1385 EXT4_ERROR_INODE(ea_inode,
1386 "ext4_getblk() return bh = NULL");
1387 return -EFSCORRUPTED;
1389 ret = ext4_journal_get_write_access(handle, bh);
1390 if (ret)
1391 goto out;
1393 memcpy(bh->b_data, buf, csize);
1394 set_buffer_uptodate(bh);
1395 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1397 buf += csize;
1398 wsize += csize;
1399 block += 1;
1402 inode_lock(ea_inode);
1403 i_size_write(ea_inode, wsize);
1404 ext4_update_i_disksize(ea_inode, wsize);
1405 inode_unlock(ea_inode);
1407 ext4_mark_inode_dirty(handle, ea_inode);
1409 out:
1410 brelse(bh);
1412 return ret;
1416 * Create an inode to store the value of a large EA.
1418 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1419 struct inode *inode, u32 hash)
1421 struct inode *ea_inode = NULL;
1422 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1423 int err;
1426 * Let the next inode be the goal, so we try and allocate the EA inode
1427 * in the same group, or nearby one.
1429 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1430 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1431 EXT4_EA_INODE_FL);
1432 if (!IS_ERR(ea_inode)) {
1433 ea_inode->i_op = &ext4_file_inode_operations;
1434 ea_inode->i_fop = &ext4_file_operations;
1435 ext4_set_aops(ea_inode);
1436 ext4_xattr_inode_set_class(ea_inode);
1437 unlock_new_inode(ea_inode);
1438 ext4_xattr_inode_set_ref(ea_inode, 1);
1439 ext4_xattr_inode_set_hash(ea_inode, hash);
1440 err = ext4_mark_inode_dirty(handle, ea_inode);
1441 if (!err)
1442 err = ext4_inode_attach_jinode(ea_inode);
1443 if (err) {
1444 iput(ea_inode);
1445 return ERR_PTR(err);
1449 * Xattr inodes are shared therefore quota charging is performed
1450 * at a higher level.
1452 dquot_free_inode(ea_inode);
1453 dquot_drop(ea_inode);
1454 inode_lock(ea_inode);
1455 ea_inode->i_flags |= S_NOQUOTA;
1456 inode_unlock(ea_inode);
1459 return ea_inode;
1462 static struct inode *
1463 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1464 size_t value_len, u32 hash)
1466 struct inode *ea_inode;
1467 struct mb_cache_entry *ce;
1468 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1469 void *ea_data;
1471 if (!ea_inode_cache)
1472 return NULL;
1474 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1475 if (!ce)
1476 return NULL;
1478 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1479 if (!ea_data) {
1480 mb_cache_entry_put(ea_inode_cache, ce);
1481 return NULL;
1484 while (ce) {
1485 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1486 EXT4_IGET_NORMAL);
1487 if (!IS_ERR(ea_inode) &&
1488 !is_bad_inode(ea_inode) &&
1489 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1490 i_size_read(ea_inode) == value_len &&
1491 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1492 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1493 value_len) &&
1494 !memcmp(value, ea_data, value_len)) {
1495 mb_cache_entry_touch(ea_inode_cache, ce);
1496 mb_cache_entry_put(ea_inode_cache, ce);
1497 kvfree(ea_data);
1498 return ea_inode;
1501 if (!IS_ERR(ea_inode))
1502 iput(ea_inode);
1503 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1505 kvfree(ea_data);
1506 return NULL;
1510 * Add value of the EA in an inode.
1512 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1513 const void *value, size_t value_len,
1514 struct inode **ret_inode)
1516 struct inode *ea_inode;
1517 u32 hash;
1518 int err;
1520 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1521 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1522 if (ea_inode) {
1523 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1524 if (err) {
1525 iput(ea_inode);
1526 return err;
1529 *ret_inode = ea_inode;
1530 return 0;
1533 /* Create an inode for the EA value */
1534 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1535 if (IS_ERR(ea_inode))
1536 return PTR_ERR(ea_inode);
1538 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1539 if (err) {
1540 ext4_xattr_inode_dec_ref(handle, ea_inode);
1541 iput(ea_inode);
1542 return err;
1545 if (EA_INODE_CACHE(inode))
1546 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1547 ea_inode->i_ino, true /* reusable */);
1549 *ret_inode = ea_inode;
1550 return 0;
1554 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1555 * feature is enabled.
1557 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1559 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1560 struct ext4_xattr_search *s,
1561 handle_t *handle, struct inode *inode,
1562 bool is_block)
1564 struct ext4_xattr_entry *last, *next;
1565 struct ext4_xattr_entry *here = s->here;
1566 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1567 int in_inode = i->in_inode;
1568 struct inode *old_ea_inode = NULL;
1569 struct inode *new_ea_inode = NULL;
1570 size_t old_size, new_size;
1571 int ret;
1573 /* Space used by old and new values. */
1574 old_size = (!s->not_found && !here->e_value_inum) ?
1575 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1576 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1579 * Optimization for the simple case when old and new values have the
1580 * same padded sizes. Not applicable if external inodes are involved.
1582 if (new_size && new_size == old_size) {
1583 size_t offs = le16_to_cpu(here->e_value_offs);
1584 void *val = s->base + offs;
1586 here->e_value_size = cpu_to_le32(i->value_len);
1587 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1588 memset(val, 0, new_size);
1589 } else {
1590 memcpy(val, i->value, i->value_len);
1591 /* Clear padding bytes. */
1592 memset(val + i->value_len, 0, new_size - i->value_len);
1594 goto update_hash;
1597 /* Compute min_offs and last. */
1598 last = s->first;
1599 for (; !IS_LAST_ENTRY(last); last = next) {
1600 next = EXT4_XATTR_NEXT(last);
1601 if ((void *)next >= s->end) {
1602 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1603 ret = -EFSCORRUPTED;
1604 goto out;
1606 if (!last->e_value_inum && last->e_value_size) {
1607 size_t offs = le16_to_cpu(last->e_value_offs);
1608 if (offs < min_offs)
1609 min_offs = offs;
1613 /* Check whether we have enough space. */
1614 if (i->value) {
1615 size_t free;
1617 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1618 if (!s->not_found)
1619 free += EXT4_XATTR_LEN(name_len) + old_size;
1621 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1622 ret = -ENOSPC;
1623 goto out;
1627 * If storing the value in an external inode is an option,
1628 * reserve space for xattr entries/names in the external
1629 * attribute block so that a long value does not occupy the
1630 * whole space and prevent futher entries being added.
1632 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1633 new_size && is_block &&
1634 (min_offs + old_size - new_size) <
1635 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1636 ret = -ENOSPC;
1637 goto out;
1642 * Getting access to old and new ea inodes is subject to failures.
1643 * Finish that work before doing any modifications to the xattr data.
1645 if (!s->not_found && here->e_value_inum) {
1646 ret = ext4_xattr_inode_iget(inode,
1647 le32_to_cpu(here->e_value_inum),
1648 le32_to_cpu(here->e_hash),
1649 &old_ea_inode);
1650 if (ret) {
1651 old_ea_inode = NULL;
1652 goto out;
1655 if (i->value && in_inode) {
1656 WARN_ON_ONCE(!i->value_len);
1658 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1659 if (ret)
1660 goto out;
1662 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1663 i->value_len,
1664 &new_ea_inode);
1665 if (ret) {
1666 new_ea_inode = NULL;
1667 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1668 goto out;
1672 if (old_ea_inode) {
1673 /* We are ready to release ref count on the old_ea_inode. */
1674 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1675 if (ret) {
1676 /* Release newly required ref count on new_ea_inode. */
1677 if (new_ea_inode) {
1678 int err;
1680 err = ext4_xattr_inode_dec_ref(handle,
1681 new_ea_inode);
1682 if (err)
1683 ext4_warning_inode(new_ea_inode,
1684 "dec ref new_ea_inode err=%d",
1685 err);
1686 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1687 i->value_len);
1689 goto out;
1692 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1693 le32_to_cpu(here->e_value_size));
1696 /* No failures allowed past this point. */
1698 if (!s->not_found && here->e_value_size && here->e_value_offs) {
1699 /* Remove the old value. */
1700 void *first_val = s->base + min_offs;
1701 size_t offs = le16_to_cpu(here->e_value_offs);
1702 void *val = s->base + offs;
1704 memmove(first_val + old_size, first_val, val - first_val);
1705 memset(first_val, 0, old_size);
1706 min_offs += old_size;
1708 /* Adjust all value offsets. */
1709 last = s->first;
1710 while (!IS_LAST_ENTRY(last)) {
1711 size_t o = le16_to_cpu(last->e_value_offs);
1713 if (!last->e_value_inum &&
1714 last->e_value_size && o < offs)
1715 last->e_value_offs = cpu_to_le16(o + old_size);
1716 last = EXT4_XATTR_NEXT(last);
1720 if (!i->value) {
1721 /* Remove old name. */
1722 size_t size = EXT4_XATTR_LEN(name_len);
1724 last = ENTRY((void *)last - size);
1725 memmove(here, (void *)here + size,
1726 (void *)last - (void *)here + sizeof(__u32));
1727 memset(last, 0, size);
1728 } else if (s->not_found) {
1729 /* Insert new name. */
1730 size_t size = EXT4_XATTR_LEN(name_len);
1731 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1733 memmove((void *)here + size, here, rest);
1734 memset(here, 0, size);
1735 here->e_name_index = i->name_index;
1736 here->e_name_len = name_len;
1737 memcpy(here->e_name, i->name, name_len);
1738 } else {
1739 /* This is an update, reset value info. */
1740 here->e_value_inum = 0;
1741 here->e_value_offs = 0;
1742 here->e_value_size = 0;
1745 if (i->value) {
1746 /* Insert new value. */
1747 if (in_inode) {
1748 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1749 } else if (i->value_len) {
1750 void *val = s->base + min_offs - new_size;
1752 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1753 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1754 memset(val, 0, new_size);
1755 } else {
1756 memcpy(val, i->value, i->value_len);
1757 /* Clear padding bytes. */
1758 memset(val + i->value_len, 0,
1759 new_size - i->value_len);
1762 here->e_value_size = cpu_to_le32(i->value_len);
1765 update_hash:
1766 if (i->value) {
1767 __le32 hash = 0;
1769 /* Entry hash calculation. */
1770 if (in_inode) {
1771 __le32 crc32c_hash;
1774 * Feed crc32c hash instead of the raw value for entry
1775 * hash calculation. This is to avoid walking
1776 * potentially long value buffer again.
1778 crc32c_hash = cpu_to_le32(
1779 ext4_xattr_inode_get_hash(new_ea_inode));
1780 hash = ext4_xattr_hash_entry(here->e_name,
1781 here->e_name_len,
1782 &crc32c_hash, 1);
1783 } else if (is_block) {
1784 __le32 *value = s->base + le16_to_cpu(
1785 here->e_value_offs);
1787 hash = ext4_xattr_hash_entry(here->e_name,
1788 here->e_name_len, value,
1789 new_size >> 2);
1791 here->e_hash = hash;
1794 if (is_block)
1795 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1797 ret = 0;
1798 out:
1799 iput(old_ea_inode);
1800 iput(new_ea_inode);
1801 return ret;
1804 struct ext4_xattr_block_find {
1805 struct ext4_xattr_search s;
1806 struct buffer_head *bh;
1809 static int
1810 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1811 struct ext4_xattr_block_find *bs)
1813 struct super_block *sb = inode->i_sb;
1814 int error;
1816 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1817 i->name_index, i->name, i->value, (long)i->value_len);
1819 if (EXT4_I(inode)->i_file_acl) {
1820 /* The inode already has an extended attribute block. */
1821 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1822 if (IS_ERR(bs->bh))
1823 return PTR_ERR(bs->bh);
1824 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1825 atomic_read(&(bs->bh->b_count)),
1826 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1827 error = ext4_xattr_check_block(inode, bs->bh);
1828 if (error)
1829 return error;
1830 /* Find the named attribute. */
1831 bs->s.base = BHDR(bs->bh);
1832 bs->s.first = BFIRST(bs->bh);
1833 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1834 bs->s.here = bs->s.first;
1835 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1836 i->name_index, i->name, 1);
1837 if (error && error != -ENODATA)
1838 return error;
1839 bs->s.not_found = error;
1841 return 0;
1844 static int
1845 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1846 struct ext4_xattr_info *i,
1847 struct ext4_xattr_block_find *bs)
1849 struct super_block *sb = inode->i_sb;
1850 struct buffer_head *new_bh = NULL;
1851 struct ext4_xattr_search s_copy = bs->s;
1852 struct ext4_xattr_search *s = &s_copy;
1853 struct mb_cache_entry *ce = NULL;
1854 int error = 0;
1855 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1856 struct inode *ea_inode = NULL, *tmp_inode;
1857 size_t old_ea_inode_quota = 0;
1858 unsigned int ea_ino;
1861 #define header(x) ((struct ext4_xattr_header *)(x))
1863 if (s->base) {
1864 BUFFER_TRACE(bs->bh, "get_write_access");
1865 error = ext4_journal_get_write_access(handle, bs->bh);
1866 if (error)
1867 goto cleanup;
1868 lock_buffer(bs->bh);
1870 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1871 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1874 * This must happen under buffer lock for
1875 * ext4_xattr_block_set() to reliably detect modified
1876 * block
1878 if (ea_block_cache)
1879 mb_cache_entry_delete(ea_block_cache, hash,
1880 bs->bh->b_blocknr);
1881 ea_bdebug(bs->bh, "modifying in-place");
1882 error = ext4_xattr_set_entry(i, s, handle, inode,
1883 true /* is_block */);
1884 ext4_xattr_block_csum_set(inode, bs->bh);
1885 unlock_buffer(bs->bh);
1886 if (error == -EFSCORRUPTED)
1887 goto bad_block;
1888 if (!error)
1889 error = ext4_handle_dirty_metadata(handle,
1890 inode,
1891 bs->bh);
1892 if (error)
1893 goto cleanup;
1894 goto inserted;
1895 } else {
1896 int offset = (char *)s->here - bs->bh->b_data;
1898 unlock_buffer(bs->bh);
1899 ea_bdebug(bs->bh, "cloning");
1900 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1901 error = -ENOMEM;
1902 if (s->base == NULL)
1903 goto cleanup;
1904 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1905 s->first = ENTRY(header(s->base)+1);
1906 header(s->base)->h_refcount = cpu_to_le32(1);
1907 s->here = ENTRY(s->base + offset);
1908 s->end = s->base + bs->bh->b_size;
1911 * If existing entry points to an xattr inode, we need
1912 * to prevent ext4_xattr_set_entry() from decrementing
1913 * ref count on it because the reference belongs to the
1914 * original block. In this case, make the entry look
1915 * like it has an empty value.
1917 if (!s->not_found && s->here->e_value_inum) {
1918 ea_ino = le32_to_cpu(s->here->e_value_inum);
1919 error = ext4_xattr_inode_iget(inode, ea_ino,
1920 le32_to_cpu(s->here->e_hash),
1921 &tmp_inode);
1922 if (error)
1923 goto cleanup;
1925 if (!ext4_test_inode_state(tmp_inode,
1926 EXT4_STATE_LUSTRE_EA_INODE)) {
1928 * Defer quota free call for previous
1929 * inode until success is guaranteed.
1931 old_ea_inode_quota = le32_to_cpu(
1932 s->here->e_value_size);
1934 iput(tmp_inode);
1936 s->here->e_value_inum = 0;
1937 s->here->e_value_size = 0;
1940 } else {
1941 /* Allocate a buffer where we construct the new block. */
1942 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1943 /* assert(header == s->base) */
1944 error = -ENOMEM;
1945 if (s->base == NULL)
1946 goto cleanup;
1947 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1948 header(s->base)->h_blocks = cpu_to_le32(1);
1949 header(s->base)->h_refcount = cpu_to_le32(1);
1950 s->first = ENTRY(header(s->base)+1);
1951 s->here = ENTRY(header(s->base)+1);
1952 s->end = s->base + sb->s_blocksize;
1955 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1956 if (error == -EFSCORRUPTED)
1957 goto bad_block;
1958 if (error)
1959 goto cleanup;
1961 if (i->value && s->here->e_value_inum) {
1963 * A ref count on ea_inode has been taken as part of the call to
1964 * ext4_xattr_set_entry() above. We would like to drop this
1965 * extra ref but we have to wait until the xattr block is
1966 * initialized and has its own ref count on the ea_inode.
1968 ea_ino = le32_to_cpu(s->here->e_value_inum);
1969 error = ext4_xattr_inode_iget(inode, ea_ino,
1970 le32_to_cpu(s->here->e_hash),
1971 &ea_inode);
1972 if (error) {
1973 ea_inode = NULL;
1974 goto cleanup;
1978 inserted:
1979 if (!IS_LAST_ENTRY(s->first)) {
1980 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1981 &ce);
1982 if (new_bh) {
1983 /* We found an identical block in the cache. */
1984 if (new_bh == bs->bh)
1985 ea_bdebug(new_bh, "keeping");
1986 else {
1987 u32 ref;
1989 WARN_ON_ONCE(dquot_initialize_needed(inode));
1991 /* The old block is released after updating
1992 the inode. */
1993 error = dquot_alloc_block(inode,
1994 EXT4_C2B(EXT4_SB(sb), 1));
1995 if (error)
1996 goto cleanup;
1997 BUFFER_TRACE(new_bh, "get_write_access");
1998 error = ext4_journal_get_write_access(handle,
1999 new_bh);
2000 if (error)
2001 goto cleanup_dquot;
2002 lock_buffer(new_bh);
2004 * We have to be careful about races with
2005 * freeing, rehashing or adding references to
2006 * xattr block. Once we hold buffer lock xattr
2007 * block's state is stable so we can check
2008 * whether the block got freed / rehashed or
2009 * not. Since we unhash mbcache entry under
2010 * buffer lock when freeing / rehashing xattr
2011 * block, checking whether entry is still
2012 * hashed is reliable. Same rules hold for
2013 * e_reusable handling.
2015 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2016 !ce->e_reusable) {
2018 * Undo everything and check mbcache
2019 * again.
2021 unlock_buffer(new_bh);
2022 dquot_free_block(inode,
2023 EXT4_C2B(EXT4_SB(sb),
2024 1));
2025 brelse(new_bh);
2026 mb_cache_entry_put(ea_block_cache, ce);
2027 ce = NULL;
2028 new_bh = NULL;
2029 goto inserted;
2031 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2032 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2033 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2034 ce->e_reusable = 0;
2035 ea_bdebug(new_bh, "reusing; refcount now=%d",
2036 ref);
2037 ext4_xattr_block_csum_set(inode, new_bh);
2038 unlock_buffer(new_bh);
2039 error = ext4_handle_dirty_metadata(handle,
2040 inode,
2041 new_bh);
2042 if (error)
2043 goto cleanup_dquot;
2045 mb_cache_entry_touch(ea_block_cache, ce);
2046 mb_cache_entry_put(ea_block_cache, ce);
2047 ce = NULL;
2048 } else if (bs->bh && s->base == bs->bh->b_data) {
2049 /* We were modifying this block in-place. */
2050 ea_bdebug(bs->bh, "keeping this block");
2051 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2052 new_bh = bs->bh;
2053 get_bh(new_bh);
2054 } else {
2055 /* We need to allocate a new block */
2056 ext4_fsblk_t goal, block;
2058 WARN_ON_ONCE(dquot_initialize_needed(inode));
2060 goal = ext4_group_first_block_no(sb,
2061 EXT4_I(inode)->i_block_group);
2063 /* non-extent files can't have physical blocks past 2^32 */
2064 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2065 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2067 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2068 NULL, &error);
2069 if (error)
2070 goto cleanup;
2072 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2073 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2075 ea_idebug(inode, "creating block %llu",
2076 (unsigned long long)block);
2078 new_bh = sb_getblk(sb, block);
2079 if (unlikely(!new_bh)) {
2080 error = -ENOMEM;
2081 getblk_failed:
2082 ext4_free_blocks(handle, inode, NULL, block, 1,
2083 EXT4_FREE_BLOCKS_METADATA);
2084 goto cleanup;
2086 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2087 ENTRY(header(s->base)+1));
2088 if (error)
2089 goto getblk_failed;
2090 if (ea_inode) {
2091 /* Drop the extra ref on ea_inode. */
2092 error = ext4_xattr_inode_dec_ref(handle,
2093 ea_inode);
2094 if (error)
2095 ext4_warning_inode(ea_inode,
2096 "dec ref error=%d",
2097 error);
2098 iput(ea_inode);
2099 ea_inode = NULL;
2102 lock_buffer(new_bh);
2103 error = ext4_journal_get_create_access(handle, new_bh);
2104 if (error) {
2105 unlock_buffer(new_bh);
2106 error = -EIO;
2107 goto getblk_failed;
2109 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2110 ext4_xattr_block_csum_set(inode, new_bh);
2111 set_buffer_uptodate(new_bh);
2112 unlock_buffer(new_bh);
2113 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2114 error = ext4_handle_dirty_metadata(handle, inode,
2115 new_bh);
2116 if (error)
2117 goto cleanup;
2121 if (old_ea_inode_quota)
2122 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2124 /* Update the inode. */
2125 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2127 /* Drop the previous xattr block. */
2128 if (bs->bh && bs->bh != new_bh) {
2129 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2131 ext4_xattr_release_block(handle, inode, bs->bh,
2132 &ea_inode_array,
2133 0 /* extra_credits */);
2134 ext4_xattr_inode_array_free(ea_inode_array);
2136 error = 0;
2138 cleanup:
2139 if (ea_inode) {
2140 int error2;
2142 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2143 if (error2)
2144 ext4_warning_inode(ea_inode, "dec ref error=%d",
2145 error2);
2147 /* If there was an error, revert the quota charge. */
2148 if (error)
2149 ext4_xattr_inode_free_quota(inode, ea_inode,
2150 i_size_read(ea_inode));
2151 iput(ea_inode);
2153 if (ce)
2154 mb_cache_entry_put(ea_block_cache, ce);
2155 brelse(new_bh);
2156 if (!(bs->bh && s->base == bs->bh->b_data))
2157 kfree(s->base);
2159 return error;
2161 cleanup_dquot:
2162 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2163 goto cleanup;
2165 bad_block:
2166 EXT4_ERROR_INODE(inode, "bad block %llu",
2167 EXT4_I(inode)->i_file_acl);
2168 goto cleanup;
2170 #undef header
2173 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2174 struct ext4_xattr_ibody_find *is)
2176 struct ext4_xattr_ibody_header *header;
2177 struct ext4_inode *raw_inode;
2178 int error;
2180 if (EXT4_I(inode)->i_extra_isize == 0)
2181 return 0;
2182 raw_inode = ext4_raw_inode(&is->iloc);
2183 header = IHDR(inode, raw_inode);
2184 is->s.base = is->s.first = IFIRST(header);
2185 is->s.here = is->s.first;
2186 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2187 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2188 error = xattr_check_inode(inode, header, is->s.end);
2189 if (error)
2190 return error;
2191 /* Find the named attribute. */
2192 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2193 i->name_index, i->name, 0);
2194 if (error && error != -ENODATA)
2195 return error;
2196 is->s.not_found = error;
2198 return 0;
2201 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2202 struct ext4_xattr_info *i,
2203 struct ext4_xattr_ibody_find *is)
2205 struct ext4_xattr_ibody_header *header;
2206 struct ext4_xattr_search *s = &is->s;
2207 int error;
2209 if (EXT4_I(inode)->i_extra_isize == 0)
2210 return -ENOSPC;
2211 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2212 if (error)
2213 return error;
2214 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2215 if (!IS_LAST_ENTRY(s->first)) {
2216 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2217 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2218 } else {
2219 header->h_magic = cpu_to_le32(0);
2220 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2222 return 0;
2225 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2226 struct ext4_xattr_info *i,
2227 struct ext4_xattr_ibody_find *is)
2229 struct ext4_xattr_ibody_header *header;
2230 struct ext4_xattr_search *s = &is->s;
2231 int error;
2233 if (EXT4_I(inode)->i_extra_isize == 0)
2234 return -ENOSPC;
2235 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2236 if (error)
2237 return error;
2238 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2239 if (!IS_LAST_ENTRY(s->first)) {
2240 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2241 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2242 } else {
2243 header->h_magic = cpu_to_le32(0);
2244 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2246 return 0;
2249 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2250 struct ext4_xattr_info *i)
2252 void *value;
2254 /* When e_value_inum is set the value is stored externally. */
2255 if (s->here->e_value_inum)
2256 return 0;
2257 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2258 return 0;
2259 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2260 return !memcmp(value, i->value, i->value_len);
2263 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2265 struct buffer_head *bh;
2266 int error;
2268 if (!EXT4_I(inode)->i_file_acl)
2269 return NULL;
2270 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2271 if (IS_ERR(bh))
2272 return bh;
2273 error = ext4_xattr_check_block(inode, bh);
2274 if (error) {
2275 brelse(bh);
2276 return ERR_PTR(error);
2278 return bh;
2282 * ext4_xattr_set_handle()
2284 * Create, replace or remove an extended attribute for this inode. Value
2285 * is NULL to remove an existing extended attribute, and non-NULL to
2286 * either replace an existing extended attribute, or create a new extended
2287 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2288 * specify that an extended attribute must exist and must not exist
2289 * previous to the call, respectively.
2291 * Returns 0, or a negative error number on failure.
2294 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2295 const char *name, const void *value, size_t value_len,
2296 int flags)
2298 struct ext4_xattr_info i = {
2299 .name_index = name_index,
2300 .name = name,
2301 .value = value,
2302 .value_len = value_len,
2303 .in_inode = 0,
2305 struct ext4_xattr_ibody_find is = {
2306 .s = { .not_found = -ENODATA, },
2308 struct ext4_xattr_block_find bs = {
2309 .s = { .not_found = -ENODATA, },
2311 int no_expand;
2312 int error;
2314 if (!name)
2315 return -EINVAL;
2316 if (strlen(name) > 255)
2317 return -ERANGE;
2319 ext4_write_lock_xattr(inode, &no_expand);
2321 /* Check journal credits under write lock. */
2322 if (ext4_handle_valid(handle)) {
2323 struct buffer_head *bh;
2324 int credits;
2326 bh = ext4_xattr_get_block(inode);
2327 if (IS_ERR(bh)) {
2328 error = PTR_ERR(bh);
2329 goto cleanup;
2332 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2333 value_len,
2334 flags & XATTR_CREATE);
2335 brelse(bh);
2337 if (!ext4_handle_has_enough_credits(handle, credits)) {
2338 error = -ENOSPC;
2339 goto cleanup;
2343 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2344 if (error)
2345 goto cleanup;
2347 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2348 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2349 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2350 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2353 error = ext4_xattr_ibody_find(inode, &i, &is);
2354 if (error)
2355 goto cleanup;
2356 if (is.s.not_found)
2357 error = ext4_xattr_block_find(inode, &i, &bs);
2358 if (error)
2359 goto cleanup;
2360 if (is.s.not_found && bs.s.not_found) {
2361 error = -ENODATA;
2362 if (flags & XATTR_REPLACE)
2363 goto cleanup;
2364 error = 0;
2365 if (!value)
2366 goto cleanup;
2367 } else {
2368 error = -EEXIST;
2369 if (flags & XATTR_CREATE)
2370 goto cleanup;
2373 if (!value) {
2374 if (!is.s.not_found)
2375 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2376 else if (!bs.s.not_found)
2377 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2378 } else {
2379 error = 0;
2380 /* Xattr value did not change? Save us some work and bail out */
2381 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2382 goto cleanup;
2383 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2384 goto cleanup;
2386 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2387 (EXT4_XATTR_SIZE(i.value_len) >
2388 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2389 i.in_inode = 1;
2390 retry_inode:
2391 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2392 if (!error && !bs.s.not_found) {
2393 i.value = NULL;
2394 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2395 } else if (error == -ENOSPC) {
2396 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2397 brelse(bs.bh);
2398 bs.bh = NULL;
2399 error = ext4_xattr_block_find(inode, &i, &bs);
2400 if (error)
2401 goto cleanup;
2403 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2404 if (!error && !is.s.not_found) {
2405 i.value = NULL;
2406 error = ext4_xattr_ibody_set(handle, inode, &i,
2407 &is);
2408 } else if (error == -ENOSPC) {
2410 * Xattr does not fit in the block, store at
2411 * external inode if possible.
2413 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2414 !i.in_inode) {
2415 i.in_inode = 1;
2416 goto retry_inode;
2421 if (!error) {
2422 ext4_xattr_update_super_block(handle, inode->i_sb);
2423 inode->i_ctime = current_time(inode);
2424 if (!value)
2425 no_expand = 0;
2426 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2428 * The bh is consumed by ext4_mark_iloc_dirty, even with
2429 * error != 0.
2431 is.iloc.bh = NULL;
2432 if (IS_SYNC(inode))
2433 ext4_handle_sync(handle);
2436 cleanup:
2437 brelse(is.iloc.bh);
2438 brelse(bs.bh);
2439 ext4_write_unlock_xattr(inode, &no_expand);
2440 return error;
2443 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2444 bool is_create, int *credits)
2446 struct buffer_head *bh;
2447 int err;
2449 *credits = 0;
2451 if (!EXT4_SB(inode->i_sb)->s_journal)
2452 return 0;
2454 down_read(&EXT4_I(inode)->xattr_sem);
2456 bh = ext4_xattr_get_block(inode);
2457 if (IS_ERR(bh)) {
2458 err = PTR_ERR(bh);
2459 } else {
2460 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2461 value_len, is_create);
2462 brelse(bh);
2463 err = 0;
2466 up_read(&EXT4_I(inode)->xattr_sem);
2467 return err;
2471 * ext4_xattr_set()
2473 * Like ext4_xattr_set_handle, but start from an inode. This extended
2474 * attribute modification is a filesystem transaction by itself.
2476 * Returns 0, or a negative error number on failure.
2479 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2480 const void *value, size_t value_len, int flags)
2482 handle_t *handle;
2483 struct super_block *sb = inode->i_sb;
2484 int error, retries = 0;
2485 int credits;
2487 error = dquot_initialize(inode);
2488 if (error)
2489 return error;
2491 retry:
2492 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2493 &credits);
2494 if (error)
2495 return error;
2497 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2498 if (IS_ERR(handle)) {
2499 error = PTR_ERR(handle);
2500 } else {
2501 int error2;
2503 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2504 value, value_len, flags);
2505 error2 = ext4_journal_stop(handle);
2506 if (error == -ENOSPC &&
2507 ext4_should_retry_alloc(sb, &retries))
2508 goto retry;
2509 if (error == 0)
2510 error = error2;
2513 return error;
2517 * Shift the EA entries in the inode to create space for the increased
2518 * i_extra_isize.
2520 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2521 int value_offs_shift, void *to,
2522 void *from, size_t n)
2524 struct ext4_xattr_entry *last = entry;
2525 int new_offs;
2527 /* We always shift xattr headers further thus offsets get lower */
2528 BUG_ON(value_offs_shift > 0);
2530 /* Adjust the value offsets of the entries */
2531 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2532 if (!last->e_value_inum && last->e_value_size) {
2533 new_offs = le16_to_cpu(last->e_value_offs) +
2534 value_offs_shift;
2535 last->e_value_offs = cpu_to_le16(new_offs);
2538 /* Shift the entries by n bytes */
2539 memmove(to, from, n);
2543 * Move xattr pointed to by 'entry' from inode into external xattr block
2545 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2546 struct ext4_inode *raw_inode,
2547 struct ext4_xattr_entry *entry)
2549 struct ext4_xattr_ibody_find *is = NULL;
2550 struct ext4_xattr_block_find *bs = NULL;
2551 char *buffer = NULL, *b_entry_name = NULL;
2552 size_t value_size = le32_to_cpu(entry->e_value_size);
2553 struct ext4_xattr_info i = {
2554 .value = NULL,
2555 .value_len = 0,
2556 .name_index = entry->e_name_index,
2557 .in_inode = !!entry->e_value_inum,
2559 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2560 int error;
2562 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2563 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2564 buffer = kmalloc(value_size, GFP_NOFS);
2565 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2566 if (!is || !bs || !buffer || !b_entry_name) {
2567 error = -ENOMEM;
2568 goto out;
2571 is->s.not_found = -ENODATA;
2572 bs->s.not_found = -ENODATA;
2573 is->iloc.bh = NULL;
2574 bs->bh = NULL;
2576 /* Save the entry name and the entry value */
2577 if (entry->e_value_inum) {
2578 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2579 if (error)
2580 goto out;
2581 } else {
2582 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2583 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2586 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2587 b_entry_name[entry->e_name_len] = '\0';
2588 i.name = b_entry_name;
2590 error = ext4_get_inode_loc(inode, &is->iloc);
2591 if (error)
2592 goto out;
2594 error = ext4_xattr_ibody_find(inode, &i, is);
2595 if (error)
2596 goto out;
2598 /* Remove the chosen entry from the inode */
2599 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2600 if (error)
2601 goto out;
2603 i.value = buffer;
2604 i.value_len = value_size;
2605 error = ext4_xattr_block_find(inode, &i, bs);
2606 if (error)
2607 goto out;
2609 /* Add entry which was removed from the inode into the block */
2610 error = ext4_xattr_block_set(handle, inode, &i, bs);
2611 if (error)
2612 goto out;
2613 error = 0;
2614 out:
2615 kfree(b_entry_name);
2616 kfree(buffer);
2617 if (is)
2618 brelse(is->iloc.bh);
2619 if (bs)
2620 brelse(bs->bh);
2621 kfree(is);
2622 kfree(bs);
2624 return error;
2627 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2628 struct ext4_inode *raw_inode,
2629 int isize_diff, size_t ifree,
2630 size_t bfree, int *total_ino)
2632 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2633 struct ext4_xattr_entry *small_entry;
2634 struct ext4_xattr_entry *entry;
2635 struct ext4_xattr_entry *last;
2636 unsigned int entry_size; /* EA entry size */
2637 unsigned int total_size; /* EA entry size + value size */
2638 unsigned int min_total_size;
2639 int error;
2641 while (isize_diff > ifree) {
2642 entry = NULL;
2643 small_entry = NULL;
2644 min_total_size = ~0U;
2645 last = IFIRST(header);
2646 /* Find the entry best suited to be pushed into EA block */
2647 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2648 /* never move system.data out of the inode */
2649 if ((last->e_name_len == 4) &&
2650 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2651 !memcmp(last->e_name, "data", 4))
2652 continue;
2653 total_size = EXT4_XATTR_LEN(last->e_name_len);
2654 if (!last->e_value_inum)
2655 total_size += EXT4_XATTR_SIZE(
2656 le32_to_cpu(last->e_value_size));
2657 if (total_size <= bfree &&
2658 total_size < min_total_size) {
2659 if (total_size + ifree < isize_diff) {
2660 small_entry = last;
2661 } else {
2662 entry = last;
2663 min_total_size = total_size;
2668 if (entry == NULL) {
2669 if (small_entry == NULL)
2670 return -ENOSPC;
2671 entry = small_entry;
2674 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2675 total_size = entry_size;
2676 if (!entry->e_value_inum)
2677 total_size += EXT4_XATTR_SIZE(
2678 le32_to_cpu(entry->e_value_size));
2679 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2680 entry);
2681 if (error)
2682 return error;
2684 *total_ino -= entry_size;
2685 ifree += total_size;
2686 bfree -= total_size;
2689 return 0;
2693 * Expand an inode by new_extra_isize bytes when EAs are present.
2694 * Returns 0 on success or negative error number on failure.
2696 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2697 struct ext4_inode *raw_inode, handle_t *handle)
2699 struct ext4_xattr_ibody_header *header;
2700 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2701 static unsigned int mnt_count;
2702 size_t min_offs;
2703 size_t ifree, bfree;
2704 int total_ino;
2705 void *base, *end;
2706 int error = 0, tried_min_extra_isize = 0;
2707 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2708 int isize_diff; /* How much do we need to grow i_extra_isize */
2710 retry:
2711 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2712 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2713 return 0;
2715 header = IHDR(inode, raw_inode);
2718 * Check if enough free space is available in the inode to shift the
2719 * entries ahead by new_extra_isize.
2722 base = IFIRST(header);
2723 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2724 min_offs = end - base;
2725 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2727 error = xattr_check_inode(inode, header, end);
2728 if (error)
2729 goto cleanup;
2731 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2732 if (ifree >= isize_diff)
2733 goto shift;
2736 * Enough free space isn't available in the inode, check if
2737 * EA block can hold new_extra_isize bytes.
2739 if (EXT4_I(inode)->i_file_acl) {
2740 struct buffer_head *bh;
2742 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2743 if (IS_ERR(bh)) {
2744 error = PTR_ERR(bh);
2745 goto cleanup;
2747 error = ext4_xattr_check_block(inode, bh);
2748 if (error) {
2749 brelse(bh);
2750 goto cleanup;
2752 base = BHDR(bh);
2753 end = bh->b_data + bh->b_size;
2754 min_offs = end - base;
2755 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2756 NULL);
2757 brelse(bh);
2758 if (bfree + ifree < isize_diff) {
2759 if (!tried_min_extra_isize && s_min_extra_isize) {
2760 tried_min_extra_isize++;
2761 new_extra_isize = s_min_extra_isize;
2762 goto retry;
2764 error = -ENOSPC;
2765 goto cleanup;
2767 } else {
2768 bfree = inode->i_sb->s_blocksize;
2771 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2772 isize_diff, ifree, bfree,
2773 &total_ino);
2774 if (error) {
2775 if (error == -ENOSPC && !tried_min_extra_isize &&
2776 s_min_extra_isize) {
2777 tried_min_extra_isize++;
2778 new_extra_isize = s_min_extra_isize;
2779 goto retry;
2781 goto cleanup;
2783 shift:
2784 /* Adjust the offsets and shift the remaining entries ahead */
2785 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2786 - new_extra_isize, (void *)raw_inode +
2787 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2788 (void *)header, total_ino);
2789 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2791 cleanup:
2792 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2793 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2794 inode->i_ino);
2795 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2797 return error;
2800 #define EIA_INCR 16 /* must be 2^n */
2801 #define EIA_MASK (EIA_INCR - 1)
2803 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2804 * If @ea_inode_array is new or full it will be grown and the old
2805 * contents copied over.
2807 static int
2808 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2809 struct inode *inode)
2811 if (*ea_inode_array == NULL) {
2813 * Start with 15 inodes, so it fits into a power-of-two size.
2814 * If *ea_inode_array is NULL, this is essentially offsetof()
2816 (*ea_inode_array) =
2817 kmalloc(offsetof(struct ext4_xattr_inode_array,
2818 inodes[EIA_MASK]),
2819 GFP_NOFS);
2820 if (*ea_inode_array == NULL)
2821 return -ENOMEM;
2822 (*ea_inode_array)->count = 0;
2823 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2824 /* expand the array once all 15 + n * 16 slots are full */
2825 struct ext4_xattr_inode_array *new_array = NULL;
2826 int count = (*ea_inode_array)->count;
2828 /* if new_array is NULL, this is essentially offsetof() */
2829 new_array = kmalloc(
2830 offsetof(struct ext4_xattr_inode_array,
2831 inodes[count + EIA_INCR]),
2832 GFP_NOFS);
2833 if (new_array == NULL)
2834 return -ENOMEM;
2835 memcpy(new_array, *ea_inode_array,
2836 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2837 kfree(*ea_inode_array);
2838 *ea_inode_array = new_array;
2840 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2841 return 0;
2845 * ext4_xattr_delete_inode()
2847 * Free extended attribute resources associated with this inode. Traverse
2848 * all entries and decrement reference on any xattr inodes associated with this
2849 * inode. This is called immediately before an inode is freed. We have exclusive
2850 * access to the inode. If an orphan inode is deleted it will also release its
2851 * references on xattr block and xattr inodes.
2853 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2854 struct ext4_xattr_inode_array **ea_inode_array,
2855 int extra_credits)
2857 struct buffer_head *bh = NULL;
2858 struct ext4_xattr_ibody_header *header;
2859 struct ext4_iloc iloc = { .bh = NULL };
2860 struct ext4_xattr_entry *entry;
2861 struct inode *ea_inode;
2862 int error;
2864 error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2865 NULL /* bh */,
2866 false /* dirty */,
2867 false /* block_csum */);
2868 if (error) {
2869 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2870 goto cleanup;
2873 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2874 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2876 error = ext4_get_inode_loc(inode, &iloc);
2877 if (error) {
2878 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2879 goto cleanup;
2882 error = ext4_journal_get_write_access(handle, iloc.bh);
2883 if (error) {
2884 EXT4_ERROR_INODE(inode, "write access (error %d)",
2885 error);
2886 goto cleanup;
2889 header = IHDR(inode, ext4_raw_inode(&iloc));
2890 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2891 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2892 IFIRST(header),
2893 false /* block_csum */,
2894 ea_inode_array,
2895 extra_credits,
2896 false /* skip_quota */);
2899 if (EXT4_I(inode)->i_file_acl) {
2900 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2901 if (IS_ERR(bh)) {
2902 error = PTR_ERR(bh);
2903 if (error == -EIO)
2904 EXT4_ERROR_INODE(inode, "block %llu read error",
2905 EXT4_I(inode)->i_file_acl);
2906 goto cleanup;
2908 error = ext4_xattr_check_block(inode, bh);
2909 if (error)
2910 goto cleanup;
2912 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2913 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2914 entry = EXT4_XATTR_NEXT(entry)) {
2915 if (!entry->e_value_inum)
2916 continue;
2917 error = ext4_xattr_inode_iget(inode,
2918 le32_to_cpu(entry->e_value_inum),
2919 le32_to_cpu(entry->e_hash),
2920 &ea_inode);
2921 if (error)
2922 continue;
2923 ext4_xattr_inode_free_quota(inode, ea_inode,
2924 le32_to_cpu(entry->e_value_size));
2925 iput(ea_inode);
2930 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2931 extra_credits);
2933 * Update i_file_acl value in the same transaction that releases
2934 * block.
2936 EXT4_I(inode)->i_file_acl = 0;
2937 error = ext4_mark_inode_dirty(handle, inode);
2938 if (error) {
2939 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2940 error);
2941 goto cleanup;
2944 error = 0;
2945 cleanup:
2946 brelse(iloc.bh);
2947 brelse(bh);
2948 return error;
2951 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2953 int idx;
2955 if (ea_inode_array == NULL)
2956 return;
2958 for (idx = 0; idx < ea_inode_array->count; ++idx)
2959 iput(ea_inode_array->inodes[idx]);
2960 kfree(ea_inode_array);
2964 * ext4_xattr_block_cache_insert()
2966 * Create a new entry in the extended attribute block cache, and insert
2967 * it unless such an entry is already in the cache.
2969 * Returns 0, or a negative error number on failure.
2971 static void
2972 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2973 struct buffer_head *bh)
2975 struct ext4_xattr_header *header = BHDR(bh);
2976 __u32 hash = le32_to_cpu(header->h_hash);
2977 int reusable = le32_to_cpu(header->h_refcount) <
2978 EXT4_XATTR_REFCOUNT_MAX;
2979 int error;
2981 if (!ea_block_cache)
2982 return;
2983 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2984 bh->b_blocknr, reusable);
2985 if (error) {
2986 if (error == -EBUSY)
2987 ea_bdebug(bh, "already in cache");
2988 } else
2989 ea_bdebug(bh, "inserting [%x]", (int)hash);
2993 * ext4_xattr_cmp()
2995 * Compare two extended attribute blocks for equality.
2997 * Returns 0 if the blocks are equal, 1 if they differ, and
2998 * a negative error number on errors.
3000 static int
3001 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3002 struct ext4_xattr_header *header2)
3004 struct ext4_xattr_entry *entry1, *entry2;
3006 entry1 = ENTRY(header1+1);
3007 entry2 = ENTRY(header2+1);
3008 while (!IS_LAST_ENTRY(entry1)) {
3009 if (IS_LAST_ENTRY(entry2))
3010 return 1;
3011 if (entry1->e_hash != entry2->e_hash ||
3012 entry1->e_name_index != entry2->e_name_index ||
3013 entry1->e_name_len != entry2->e_name_len ||
3014 entry1->e_value_size != entry2->e_value_size ||
3015 entry1->e_value_inum != entry2->e_value_inum ||
3016 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3017 return 1;
3018 if (!entry1->e_value_inum &&
3019 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3020 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3021 le32_to_cpu(entry1->e_value_size)))
3022 return 1;
3024 entry1 = EXT4_XATTR_NEXT(entry1);
3025 entry2 = EXT4_XATTR_NEXT(entry2);
3027 if (!IS_LAST_ENTRY(entry2))
3028 return 1;
3029 return 0;
3033 * ext4_xattr_block_cache_find()
3035 * Find an identical extended attribute block.
3037 * Returns a pointer to the block found, or NULL if such a block was
3038 * not found or an error occurred.
3040 static struct buffer_head *
3041 ext4_xattr_block_cache_find(struct inode *inode,
3042 struct ext4_xattr_header *header,
3043 struct mb_cache_entry **pce)
3045 __u32 hash = le32_to_cpu(header->h_hash);
3046 struct mb_cache_entry *ce;
3047 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3049 if (!ea_block_cache)
3050 return NULL;
3051 if (!header->h_hash)
3052 return NULL; /* never share */
3053 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3054 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3055 while (ce) {
3056 struct buffer_head *bh;
3058 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3059 if (IS_ERR(bh)) {
3060 if (PTR_ERR(bh) == -ENOMEM)
3061 return NULL;
3062 EXT4_ERROR_INODE(inode, "block %lu read error",
3063 (unsigned long)ce->e_value);
3064 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3065 *pce = ce;
3066 return bh;
3068 brelse(bh);
3069 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3071 return NULL;
3074 #define NAME_HASH_SHIFT 5
3075 #define VALUE_HASH_SHIFT 16
3078 * ext4_xattr_hash_entry()
3080 * Compute the hash of an extended attribute.
3082 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3083 size_t value_count)
3085 __u32 hash = 0;
3087 while (name_len--) {
3088 hash = (hash << NAME_HASH_SHIFT) ^
3089 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3090 *name++;
3092 while (value_count--) {
3093 hash = (hash << VALUE_HASH_SHIFT) ^
3094 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3095 le32_to_cpu(*value++);
3097 return cpu_to_le32(hash);
3100 #undef NAME_HASH_SHIFT
3101 #undef VALUE_HASH_SHIFT
3103 #define BLOCK_HASH_SHIFT 16
3106 * ext4_xattr_rehash()
3108 * Re-compute the extended attribute hash value after an entry has changed.
3110 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3112 struct ext4_xattr_entry *here;
3113 __u32 hash = 0;
3115 here = ENTRY(header+1);
3116 while (!IS_LAST_ENTRY(here)) {
3117 if (!here->e_hash) {
3118 /* Block is not shared if an entry's hash value == 0 */
3119 hash = 0;
3120 break;
3122 hash = (hash << BLOCK_HASH_SHIFT) ^
3123 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3124 le32_to_cpu(here->e_hash);
3125 here = EXT4_XATTR_NEXT(here);
3127 header->h_hash = cpu_to_le32(hash);
3130 #undef BLOCK_HASH_SHIFT
3132 #define HASH_BUCKET_BITS 10
3134 struct mb_cache *
3135 ext4_xattr_create_cache(void)
3137 return mb_cache_create(HASH_BUCKET_BITS);
3140 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3142 if (cache)
3143 mb_cache_destroy(cache);