mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / ext4 / xattr.c
blob823b01f6b0f0a7a9d006c7f3c24df79a4fd06d32
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 "ext4_jbd2.h"
60 #include "ext4.h"
61 #include "xattr.h"
62 #include "acl.h"
64 #ifdef EXT4_XATTR_DEBUG
65 # define ea_idebug(inode, fmt, ...) \
66 printk(KERN_DEBUG "inode %s:%lu: " fmt "\n", \
67 inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
68 # define ea_bdebug(bh, fmt, ...) \
69 printk(KERN_DEBUG "block %pg:%lu: " fmt "\n", \
70 bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
71 #else
72 # define ea_idebug(inode, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
73 # define ea_bdebug(bh, fmt, ...) no_printk(fmt, ##__VA_ARGS__)
74 #endif
76 static void ext4_xattr_block_cache_insert(struct mb_cache *,
77 struct buffer_head *);
78 static struct buffer_head *
79 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
80 struct mb_cache_entry **);
81 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
82 size_t value_count);
83 static void ext4_xattr_rehash(struct ext4_xattr_header *);
85 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
86 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
87 #ifdef CONFIG_EXT4_FS_POSIX_ACL
88 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
89 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
90 #endif
91 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
92 #ifdef CONFIG_EXT4_FS_SECURITY
93 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
94 #endif
97 const struct xattr_handler *ext4_xattr_handlers[] = {
98 &ext4_xattr_user_handler,
99 &ext4_xattr_trusted_handler,
100 #ifdef CONFIG_EXT4_FS_POSIX_ACL
101 &posix_acl_access_xattr_handler,
102 &posix_acl_default_xattr_handler,
103 #endif
104 #ifdef CONFIG_EXT4_FS_SECURITY
105 &ext4_xattr_security_handler,
106 #endif
107 NULL
110 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
111 inode->i_sb->s_fs_info)->s_ea_block_cache)
113 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
114 inode->i_sb->s_fs_info)->s_ea_inode_cache)
116 static int
117 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
118 struct inode *inode);
120 #ifdef CONFIG_LOCKDEP
121 void ext4_xattr_inode_set_class(struct inode *ea_inode)
123 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
125 #endif
127 static __le32 ext4_xattr_block_csum(struct inode *inode,
128 sector_t block_nr,
129 struct ext4_xattr_header *hdr)
131 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
132 __u32 csum;
133 __le64 dsk_block_nr = cpu_to_le64(block_nr);
134 __u32 dummy_csum = 0;
135 int offset = offsetof(struct ext4_xattr_header, h_checksum);
137 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
138 sizeof(dsk_block_nr));
139 csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
140 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
141 offset += sizeof(dummy_csum);
142 csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
143 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
145 return cpu_to_le32(csum);
148 static int ext4_xattr_block_csum_verify(struct inode *inode,
149 struct buffer_head *bh)
151 struct ext4_xattr_header *hdr = BHDR(bh);
152 int ret = 1;
154 if (ext4_has_metadata_csum(inode->i_sb)) {
155 lock_buffer(bh);
156 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
157 bh->b_blocknr, hdr));
158 unlock_buffer(bh);
160 return ret;
163 static void ext4_xattr_block_csum_set(struct inode *inode,
164 struct buffer_head *bh)
166 if (ext4_has_metadata_csum(inode->i_sb))
167 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
168 bh->b_blocknr, BHDR(bh));
171 static inline const struct xattr_handler *
172 ext4_xattr_handler(int name_index)
174 const struct xattr_handler *handler = NULL;
176 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
177 handler = ext4_xattr_handler_map[name_index];
178 return handler;
181 static int
182 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
183 void *value_start)
185 struct ext4_xattr_entry *e = entry;
187 /* Find the end of the names list */
188 while (!IS_LAST_ENTRY(e)) {
189 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
190 if ((void *)next >= end)
191 return -EFSCORRUPTED;
192 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
193 return -EFSCORRUPTED;
194 e = next;
197 /* Check the values */
198 while (!IS_LAST_ENTRY(entry)) {
199 u32 size = le32_to_cpu(entry->e_value_size);
201 if (size > EXT4_XATTR_SIZE_MAX)
202 return -EFSCORRUPTED;
204 if (size != 0 && entry->e_value_inum == 0) {
205 u16 offs = le16_to_cpu(entry->e_value_offs);
206 void *value;
209 * The value cannot overlap the names, and the value
210 * with padding cannot extend beyond 'end'. Check both
211 * the padded and unpadded sizes, since the size may
212 * overflow to 0 when adding padding.
214 if (offs > end - value_start)
215 return -EFSCORRUPTED;
216 value = value_start + offs;
217 if (value < (void *)e + sizeof(u32) ||
218 size > end - value ||
219 EXT4_XATTR_SIZE(size) > end - value)
220 return -EFSCORRUPTED;
222 entry = EXT4_XATTR_NEXT(entry);
225 return 0;
228 static inline int
229 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
230 const char *function, unsigned int line)
232 int error = -EFSCORRUPTED;
234 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
235 BHDR(bh)->h_blocks != cpu_to_le32(1))
236 goto errout;
237 if (buffer_verified(bh))
238 return 0;
240 error = -EFSBADCRC;
241 if (!ext4_xattr_block_csum_verify(inode, bh))
242 goto errout;
243 error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
244 bh->b_data);
245 errout:
246 if (error)
247 __ext4_error_inode(inode, function, line, 0,
248 "corrupted xattr block %llu",
249 (unsigned long long) bh->b_blocknr);
250 else
251 set_buffer_verified(bh);
252 return error;
255 #define ext4_xattr_check_block(inode, bh) \
256 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
259 static int
260 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
261 void *end, const char *function, unsigned int line)
263 int error = -EFSCORRUPTED;
265 if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
266 (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
267 goto errout;
268 error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
269 errout:
270 if (error)
271 __ext4_error_inode(inode, function, line, 0,
272 "corrupted in-inode xattr");
273 return error;
276 #define xattr_check_inode(inode, header, end) \
277 __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
279 static int
280 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
281 void *end, int name_index, const char *name, int sorted)
283 struct ext4_xattr_entry *entry, *next;
284 size_t name_len;
285 int cmp = 1;
287 if (name == NULL)
288 return -EINVAL;
289 name_len = strlen(name);
290 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
291 next = EXT4_XATTR_NEXT(entry);
292 if ((void *) next >= end) {
293 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
294 return -EFSCORRUPTED;
296 cmp = name_index - entry->e_name_index;
297 if (!cmp)
298 cmp = name_len - entry->e_name_len;
299 if (!cmp)
300 cmp = memcmp(name, entry->e_name, name_len);
301 if (cmp <= 0 && (sorted || cmp == 0))
302 break;
304 *pentry = entry;
305 return cmp ? -ENODATA : 0;
308 static u32
309 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
311 return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
314 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
316 return ((u64)ea_inode->i_ctime.tv_sec << 32) |
317 ((u32)ea_inode->i_version);
320 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
322 ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
323 ea_inode->i_version = (u32)ref_count;
326 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
328 return (u32)ea_inode->i_atime.tv_sec;
331 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
333 ea_inode->i_atime.tv_sec = hash;
337 * Read the EA value from an inode.
339 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
341 int blocksize = 1 << ea_inode->i_blkbits;
342 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
343 int tail_size = (size % blocksize) ?: blocksize;
344 struct buffer_head *bhs_inline[8];
345 struct buffer_head **bhs = bhs_inline;
346 int i, ret;
348 if (bh_count > ARRAY_SIZE(bhs_inline)) {
349 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
350 if (!bhs)
351 return -ENOMEM;
354 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
355 true /* wait */, bhs);
356 if (ret)
357 goto free_bhs;
359 for (i = 0; i < bh_count; i++) {
360 /* There shouldn't be any holes in ea_inode. */
361 if (!bhs[i]) {
362 ret = -EFSCORRUPTED;
363 goto put_bhs;
365 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
366 i < bh_count - 1 ? blocksize : tail_size);
368 ret = 0;
369 put_bhs:
370 for (i = 0; i < bh_count; i++)
371 brelse(bhs[i]);
372 free_bhs:
373 if (bhs != bhs_inline)
374 kfree(bhs);
375 return ret;
378 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
380 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
381 u32 ea_inode_hash, struct inode **ea_inode)
383 struct inode *inode;
384 int err;
386 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
387 if (IS_ERR(inode)) {
388 err = PTR_ERR(inode);
389 ext4_error(parent->i_sb,
390 "error while reading EA inode %lu err=%d", ea_ino,
391 err);
392 return err;
395 if (is_bad_inode(inode)) {
396 ext4_error(parent->i_sb,
397 "error while reading EA inode %lu is_bad_inode",
398 ea_ino);
399 err = -EIO;
400 goto error;
403 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
404 ext4_error(parent->i_sb,
405 "EA inode %lu does not have EXT4_EA_INODE_FL flag",
406 ea_ino);
407 err = -EINVAL;
408 goto error;
411 ext4_xattr_inode_set_class(inode);
414 * Check whether this is an old Lustre-style xattr inode. Lustre
415 * implementation does not have hash validation, rather it has a
416 * backpointer from ea_inode to the parent inode.
418 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
419 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
420 inode->i_generation == parent->i_generation) {
421 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
422 ext4_xattr_inode_set_ref(inode, 1);
423 } else {
424 inode_lock(inode);
425 inode->i_flags |= S_NOQUOTA;
426 inode_unlock(inode);
429 *ea_inode = inode;
430 return 0;
431 error:
432 iput(inode);
433 return err;
436 static int
437 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
438 struct ext4_xattr_entry *entry, void *buffer,
439 size_t size)
441 u32 hash;
443 /* Verify stored hash matches calculated hash. */
444 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
445 if (hash != ext4_xattr_inode_get_hash(ea_inode))
446 return -EFSCORRUPTED;
448 if (entry) {
449 __le32 e_hash, tmp_data;
451 /* Verify entry hash. */
452 tmp_data = cpu_to_le32(hash);
453 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
454 &tmp_data, 1);
455 if (e_hash != entry->e_hash)
456 return -EFSCORRUPTED;
458 return 0;
462 * Read xattr value from the EA inode.
464 static int
465 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
466 void *buffer, size_t size)
468 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
469 struct inode *ea_inode;
470 int err;
472 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
473 le32_to_cpu(entry->e_hash), &ea_inode);
474 if (err) {
475 ea_inode = NULL;
476 goto out;
479 if (i_size_read(ea_inode) != size) {
480 ext4_warning_inode(ea_inode,
481 "ea_inode file size=%llu entry size=%zu",
482 i_size_read(ea_inode), size);
483 err = -EFSCORRUPTED;
484 goto out;
487 err = ext4_xattr_inode_read(ea_inode, buffer, size);
488 if (err)
489 goto out;
491 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
492 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
493 size);
494 if (err) {
495 ext4_warning_inode(ea_inode,
496 "EA inode hash validation failed");
497 goto out;
500 if (ea_inode_cache)
501 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
502 ext4_xattr_inode_get_hash(ea_inode),
503 ea_inode->i_ino, true /* reusable */);
505 out:
506 iput(ea_inode);
507 return err;
510 static int
511 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
512 void *buffer, size_t buffer_size)
514 struct buffer_head *bh = NULL;
515 struct ext4_xattr_entry *entry;
516 size_t size;
517 void *end;
518 int error;
519 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
521 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
522 name_index, name, buffer, (long)buffer_size);
524 if (!EXT4_I(inode)->i_file_acl)
525 return -ENODATA;
526 ea_idebug(inode, "reading block %llu",
527 (unsigned long long)EXT4_I(inode)->i_file_acl);
528 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
529 if (IS_ERR(bh))
530 return PTR_ERR(bh);
531 ea_bdebug(bh, "b_count=%d, refcount=%d",
532 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
533 error = ext4_xattr_check_block(inode, bh);
534 if (error)
535 goto cleanup;
536 ext4_xattr_block_cache_insert(ea_block_cache, bh);
537 entry = BFIRST(bh);
538 end = bh->b_data + bh->b_size;
539 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
540 if (error)
541 goto cleanup;
542 size = le32_to_cpu(entry->e_value_size);
543 error = -ERANGE;
544 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
545 goto cleanup;
546 if (buffer) {
547 if (size > buffer_size)
548 goto cleanup;
549 if (entry->e_value_inum) {
550 error = ext4_xattr_inode_get(inode, entry, buffer,
551 size);
552 if (error)
553 goto cleanup;
554 } else {
555 u16 offset = le16_to_cpu(entry->e_value_offs);
556 void *p = bh->b_data + offset;
558 if (unlikely(p + size > end))
559 goto cleanup;
560 memcpy(buffer, p, size);
563 error = size;
565 cleanup:
566 brelse(bh);
567 return error;
571 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
572 void *buffer, size_t buffer_size)
574 struct ext4_xattr_ibody_header *header;
575 struct ext4_xattr_entry *entry;
576 struct ext4_inode *raw_inode;
577 struct ext4_iloc iloc;
578 size_t size;
579 void *end;
580 int error;
582 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
583 return -ENODATA;
584 error = ext4_get_inode_loc(inode, &iloc);
585 if (error)
586 return error;
587 raw_inode = ext4_raw_inode(&iloc);
588 header = IHDR(inode, raw_inode);
589 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
590 error = xattr_check_inode(inode, header, end);
591 if (error)
592 goto cleanup;
593 entry = IFIRST(header);
594 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
595 if (error)
596 goto cleanup;
597 size = le32_to_cpu(entry->e_value_size);
598 error = -ERANGE;
599 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
600 goto cleanup;
601 if (buffer) {
602 if (size > buffer_size)
603 goto cleanup;
604 if (entry->e_value_inum) {
605 error = ext4_xattr_inode_get(inode, entry, buffer,
606 size);
607 if (error)
608 goto cleanup;
609 } else {
610 u16 offset = le16_to_cpu(entry->e_value_offs);
611 void *p = (void *)IFIRST(header) + offset;
613 if (unlikely(p + size > end))
614 goto cleanup;
615 memcpy(buffer, p, size);
618 error = size;
620 cleanup:
621 brelse(iloc.bh);
622 return error;
626 * ext4_xattr_get()
628 * Copy an extended attribute into the buffer
629 * provided, or compute the buffer size required.
630 * Buffer is NULL to compute the size of the buffer required.
632 * Returns a negative error number on failure, or the number of bytes
633 * used / required on success.
636 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
637 void *buffer, size_t buffer_size)
639 int error;
641 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
642 return -EIO;
644 if (strlen(name) > 255)
645 return -ERANGE;
647 down_read(&EXT4_I(inode)->xattr_sem);
648 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
649 buffer_size);
650 if (error == -ENODATA)
651 error = ext4_xattr_block_get(inode, name_index, name, buffer,
652 buffer_size);
653 up_read(&EXT4_I(inode)->xattr_sem);
654 return error;
657 static int
658 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
659 char *buffer, size_t buffer_size)
661 size_t rest = buffer_size;
663 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
664 const struct xattr_handler *handler =
665 ext4_xattr_handler(entry->e_name_index);
667 if (handler && (!handler->list || handler->list(dentry))) {
668 const char *prefix = handler->prefix ?: handler->name;
669 size_t prefix_len = strlen(prefix);
670 size_t size = prefix_len + entry->e_name_len + 1;
672 if (buffer) {
673 if (size > rest)
674 return -ERANGE;
675 memcpy(buffer, prefix, prefix_len);
676 buffer += prefix_len;
677 memcpy(buffer, entry->e_name, entry->e_name_len);
678 buffer += entry->e_name_len;
679 *buffer++ = 0;
681 rest -= size;
684 return buffer_size - rest; /* total size */
687 static int
688 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
690 struct inode *inode = d_inode(dentry);
691 struct buffer_head *bh = NULL;
692 int error;
694 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
695 buffer, (long)buffer_size);
697 if (!EXT4_I(inode)->i_file_acl)
698 return 0;
699 ea_idebug(inode, "reading block %llu",
700 (unsigned long long)EXT4_I(inode)->i_file_acl);
701 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
702 if (IS_ERR(bh))
703 return PTR_ERR(bh);
704 ea_bdebug(bh, "b_count=%d, refcount=%d",
705 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
706 error = ext4_xattr_check_block(inode, bh);
707 if (error)
708 goto cleanup;
709 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
710 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
711 buffer_size);
712 cleanup:
713 brelse(bh);
714 return error;
717 static int
718 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
720 struct inode *inode = d_inode(dentry);
721 struct ext4_xattr_ibody_header *header;
722 struct ext4_inode *raw_inode;
723 struct ext4_iloc iloc;
724 void *end;
725 int error;
727 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
728 return 0;
729 error = ext4_get_inode_loc(inode, &iloc);
730 if (error)
731 return error;
732 raw_inode = ext4_raw_inode(&iloc);
733 header = IHDR(inode, raw_inode);
734 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
735 error = xattr_check_inode(inode, header, end);
736 if (error)
737 goto cleanup;
738 error = ext4_xattr_list_entries(dentry, IFIRST(header),
739 buffer, buffer_size);
741 cleanup:
742 brelse(iloc.bh);
743 return error;
747 * Inode operation listxattr()
749 * d_inode(dentry)->i_rwsem: don't care
751 * Copy a list of attribute names into the buffer
752 * provided, or compute the buffer size required.
753 * Buffer is NULL to compute the size of the buffer required.
755 * Returns a negative error number on failure, or the number of bytes
756 * used / required on success.
758 ssize_t
759 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
761 int ret, ret2;
763 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
764 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
765 if (ret < 0)
766 goto errout;
767 if (buffer) {
768 buffer += ret;
769 buffer_size -= ret;
771 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
772 if (ret < 0)
773 goto errout;
774 ret += ret2;
775 errout:
776 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
777 return ret;
781 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
782 * not set, set it.
784 static void ext4_xattr_update_super_block(handle_t *handle,
785 struct super_block *sb)
787 if (ext4_has_feature_xattr(sb))
788 return;
790 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
791 if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
792 ext4_set_feature_xattr(sb);
793 ext4_handle_dirty_super(handle, sb);
797 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
799 struct ext4_iloc iloc = { .bh = NULL };
800 struct buffer_head *bh = NULL;
801 struct ext4_inode *raw_inode;
802 struct ext4_xattr_ibody_header *header;
803 struct ext4_xattr_entry *entry;
804 qsize_t ea_inode_refs = 0;
805 void *end;
806 int ret;
808 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
810 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
811 ret = ext4_get_inode_loc(inode, &iloc);
812 if (ret)
813 goto out;
814 raw_inode = ext4_raw_inode(&iloc);
815 header = IHDR(inode, raw_inode);
816 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
817 ret = xattr_check_inode(inode, header, end);
818 if (ret)
819 goto out;
821 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
822 entry = EXT4_XATTR_NEXT(entry))
823 if (entry->e_value_inum)
824 ea_inode_refs++;
827 if (EXT4_I(inode)->i_file_acl) {
828 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
829 if (IS_ERR(bh)) {
830 ret = PTR_ERR(bh);
831 bh = NULL;
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 iloc.bh = NULL;
1032 goto out;
1035 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1036 ref_count += ref_change;
1037 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1039 if (ref_change > 0) {
1040 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1041 ea_inode->i_ino, ref_count);
1043 if (ref_count == 1) {
1044 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1045 ea_inode->i_ino, ea_inode->i_nlink);
1047 set_nlink(ea_inode, 1);
1048 ext4_orphan_del(handle, ea_inode);
1050 if (ea_inode_cache) {
1051 hash = ext4_xattr_inode_get_hash(ea_inode);
1052 mb_cache_entry_create(ea_inode_cache,
1053 GFP_NOFS, hash,
1054 ea_inode->i_ino,
1055 true /* reusable */);
1058 } else {
1059 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1060 ea_inode->i_ino, ref_count);
1062 if (ref_count == 0) {
1063 WARN_ONCE(ea_inode->i_nlink != 1,
1064 "EA inode %lu i_nlink=%u",
1065 ea_inode->i_ino, ea_inode->i_nlink);
1067 clear_nlink(ea_inode);
1068 ext4_orphan_add(handle, ea_inode);
1070 if (ea_inode_cache) {
1071 hash = ext4_xattr_inode_get_hash(ea_inode);
1072 mb_cache_entry_delete(ea_inode_cache, hash,
1073 ea_inode->i_ino);
1078 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1079 iloc.bh = NULL;
1080 if (ret)
1081 ext4_warning_inode(ea_inode,
1082 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1083 out:
1084 brelse(iloc.bh);
1085 inode_unlock(ea_inode);
1086 return ret;
1089 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1091 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1094 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1096 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1099 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1100 struct ext4_xattr_entry *first)
1102 struct inode *ea_inode;
1103 struct ext4_xattr_entry *entry;
1104 struct ext4_xattr_entry *failed_entry;
1105 unsigned int ea_ino;
1106 int err, saved_err;
1108 for (entry = first; !IS_LAST_ENTRY(entry);
1109 entry = EXT4_XATTR_NEXT(entry)) {
1110 if (!entry->e_value_inum)
1111 continue;
1112 ea_ino = le32_to_cpu(entry->e_value_inum);
1113 err = ext4_xattr_inode_iget(parent, ea_ino,
1114 le32_to_cpu(entry->e_hash),
1115 &ea_inode);
1116 if (err)
1117 goto cleanup;
1118 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1119 if (err) {
1120 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1121 iput(ea_inode);
1122 goto cleanup;
1124 iput(ea_inode);
1126 return 0;
1128 cleanup:
1129 saved_err = err;
1130 failed_entry = entry;
1132 for (entry = first; entry != failed_entry;
1133 entry = EXT4_XATTR_NEXT(entry)) {
1134 if (!entry->e_value_inum)
1135 continue;
1136 ea_ino = le32_to_cpu(entry->e_value_inum);
1137 err = ext4_xattr_inode_iget(parent, ea_ino,
1138 le32_to_cpu(entry->e_hash),
1139 &ea_inode);
1140 if (err) {
1141 ext4_warning(parent->i_sb,
1142 "cleanup ea_ino %u iget error %d", ea_ino,
1143 err);
1144 continue;
1146 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1147 if (err)
1148 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1149 err);
1150 iput(ea_inode);
1152 return saved_err;
1155 static void
1156 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1157 struct buffer_head *bh,
1158 struct ext4_xattr_entry *first, bool block_csum,
1159 struct ext4_xattr_inode_array **ea_inode_array,
1160 int extra_credits, bool skip_quota)
1162 struct inode *ea_inode;
1163 struct ext4_xattr_entry *entry;
1164 bool dirty = false;
1165 unsigned int ea_ino;
1166 int err;
1167 int credits;
1169 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1170 credits = 2 + extra_credits;
1172 for (entry = first; !IS_LAST_ENTRY(entry);
1173 entry = EXT4_XATTR_NEXT(entry)) {
1174 if (!entry->e_value_inum)
1175 continue;
1176 ea_ino = le32_to_cpu(entry->e_value_inum);
1177 err = ext4_xattr_inode_iget(parent, ea_ino,
1178 le32_to_cpu(entry->e_hash),
1179 &ea_inode);
1180 if (err)
1181 continue;
1183 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1184 if (err) {
1185 ext4_warning_inode(ea_inode,
1186 "Expand inode array err=%d", err);
1187 iput(ea_inode);
1188 continue;
1191 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1192 dirty, block_csum);
1193 if (err) {
1194 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1195 err);
1196 continue;
1199 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1200 if (err) {
1201 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1202 err);
1203 continue;
1206 if (!skip_quota)
1207 ext4_xattr_inode_free_quota(parent, ea_inode,
1208 le32_to_cpu(entry->e_value_size));
1211 * Forget about ea_inode within the same transaction that
1212 * decrements the ref count. This avoids duplicate decrements in
1213 * case the rest of the work spills over to subsequent
1214 * transactions.
1216 entry->e_value_inum = 0;
1217 entry->e_value_size = 0;
1219 dirty = true;
1222 if (dirty) {
1224 * Note that we are deliberately skipping csum calculation for
1225 * the final update because we do not expect any journal
1226 * restarts until xattr block is freed.
1229 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1230 if (err)
1231 ext4_warning_inode(parent,
1232 "handle dirty metadata err=%d", err);
1237 * Release the xattr block BH: If the reference count is > 1, decrement it;
1238 * otherwise free the block.
1240 static void
1241 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1242 struct buffer_head *bh,
1243 struct ext4_xattr_inode_array **ea_inode_array,
1244 int extra_credits)
1246 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1247 u32 hash, ref;
1248 int error = 0;
1250 BUFFER_TRACE(bh, "get_write_access");
1251 error = ext4_journal_get_write_access(handle, bh);
1252 if (error)
1253 goto out;
1255 lock_buffer(bh);
1256 hash = le32_to_cpu(BHDR(bh)->h_hash);
1257 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1258 if (ref == 1) {
1259 ea_bdebug(bh, "refcount now=0; freeing");
1261 * This must happen under buffer lock for
1262 * ext4_xattr_block_set() to reliably detect freed block
1264 if (ea_block_cache)
1265 mb_cache_entry_delete(ea_block_cache, hash,
1266 bh->b_blocknr);
1267 get_bh(bh);
1268 unlock_buffer(bh);
1270 if (ext4_has_feature_ea_inode(inode->i_sb))
1271 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1272 BFIRST(bh),
1273 true /* block_csum */,
1274 ea_inode_array,
1275 extra_credits,
1276 true /* skip_quota */);
1277 ext4_free_blocks(handle, inode, bh, 0, 1,
1278 EXT4_FREE_BLOCKS_METADATA |
1279 EXT4_FREE_BLOCKS_FORGET);
1280 } else {
1281 ref--;
1282 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1283 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1284 struct mb_cache_entry *ce;
1286 if (ea_block_cache) {
1287 ce = mb_cache_entry_get(ea_block_cache, hash,
1288 bh->b_blocknr);
1289 if (ce) {
1290 ce->e_reusable = 1;
1291 mb_cache_entry_put(ea_block_cache, ce);
1296 ext4_xattr_block_csum_set(inode, bh);
1298 * Beware of this ugliness: Releasing of xattr block references
1299 * from different inodes can race and so we have to protect
1300 * from a race where someone else frees the block (and releases
1301 * its journal_head) before we are done dirtying the buffer. In
1302 * nojournal mode this race is harmless and we actually cannot
1303 * call ext4_handle_dirty_metadata() with locked buffer as
1304 * that function can call sync_dirty_buffer() so for that case
1305 * we handle the dirtying after unlocking the buffer.
1307 if (ext4_handle_valid(handle))
1308 error = ext4_handle_dirty_metadata(handle, inode, bh);
1309 unlock_buffer(bh);
1310 if (!ext4_handle_valid(handle))
1311 error = ext4_handle_dirty_metadata(handle, inode, bh);
1312 if (IS_SYNC(inode))
1313 ext4_handle_sync(handle);
1314 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1315 ea_bdebug(bh, "refcount now=%d; releasing",
1316 le32_to_cpu(BHDR(bh)->h_refcount));
1318 out:
1319 ext4_std_error(inode->i_sb, error);
1320 return;
1324 * Find the available free space for EAs. This also returns the total number of
1325 * bytes used by EA entries.
1327 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1328 size_t *min_offs, void *base, int *total)
1330 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1331 if (!last->e_value_inum && last->e_value_size) {
1332 size_t offs = le16_to_cpu(last->e_value_offs);
1333 if (offs < *min_offs)
1334 *min_offs = offs;
1336 if (total)
1337 *total += EXT4_XATTR_LEN(last->e_name_len);
1339 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1343 * Write the value of the EA in an inode.
1345 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1346 const void *buf, int bufsize)
1348 struct buffer_head *bh = NULL;
1349 unsigned long block = 0;
1350 int blocksize = ea_inode->i_sb->s_blocksize;
1351 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1352 int csize, wsize = 0;
1353 int ret = 0;
1354 int retries = 0;
1356 retry:
1357 while (ret >= 0 && ret < max_blocks) {
1358 struct ext4_map_blocks map;
1359 map.m_lblk = block += ret;
1360 map.m_len = max_blocks -= ret;
1362 ret = ext4_map_blocks(handle, ea_inode, &map,
1363 EXT4_GET_BLOCKS_CREATE);
1364 if (ret <= 0) {
1365 ext4_mark_inode_dirty(handle, ea_inode);
1366 if (ret == -ENOSPC &&
1367 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1368 ret = 0;
1369 goto retry;
1371 break;
1375 if (ret < 0)
1376 return ret;
1378 block = 0;
1379 while (wsize < bufsize) {
1380 if (bh != NULL)
1381 brelse(bh);
1382 csize = (bufsize - wsize) > blocksize ? blocksize :
1383 bufsize - wsize;
1384 bh = ext4_getblk(handle, ea_inode, block, 0);
1385 if (IS_ERR(bh))
1386 return PTR_ERR(bh);
1387 if (!bh) {
1388 WARN_ON_ONCE(1);
1389 EXT4_ERROR_INODE(ea_inode,
1390 "ext4_getblk() return bh = NULL");
1391 return -EFSCORRUPTED;
1393 ret = ext4_journal_get_write_access(handle, bh);
1394 if (ret)
1395 goto out;
1397 memcpy(bh->b_data, buf, csize);
1398 set_buffer_uptodate(bh);
1399 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1401 buf += csize;
1402 wsize += csize;
1403 block += 1;
1406 inode_lock(ea_inode);
1407 i_size_write(ea_inode, wsize);
1408 ext4_update_i_disksize(ea_inode, wsize);
1409 inode_unlock(ea_inode);
1411 ext4_mark_inode_dirty(handle, ea_inode);
1413 out:
1414 brelse(bh);
1416 return ret;
1420 * Create an inode to store the value of a large EA.
1422 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1423 struct inode *inode, u32 hash)
1425 struct inode *ea_inode = NULL;
1426 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1427 int err;
1430 * Let the next inode be the goal, so we try and allocate the EA inode
1431 * in the same group, or nearby one.
1433 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1434 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1435 EXT4_EA_INODE_FL);
1436 if (!IS_ERR(ea_inode)) {
1437 ea_inode->i_op = &ext4_file_inode_operations;
1438 ea_inode->i_fop = &ext4_file_operations;
1439 ext4_set_aops(ea_inode);
1440 ext4_xattr_inode_set_class(ea_inode);
1441 unlock_new_inode(ea_inode);
1442 ext4_xattr_inode_set_ref(ea_inode, 1);
1443 ext4_xattr_inode_set_hash(ea_inode, hash);
1444 err = ext4_mark_inode_dirty(handle, ea_inode);
1445 if (!err)
1446 err = ext4_inode_attach_jinode(ea_inode);
1447 if (err) {
1448 iput(ea_inode);
1449 return ERR_PTR(err);
1453 * Xattr inodes are shared therefore quota charging is performed
1454 * at a higher level.
1456 dquot_free_inode(ea_inode);
1457 dquot_drop(ea_inode);
1458 inode_lock(ea_inode);
1459 ea_inode->i_flags |= S_NOQUOTA;
1460 inode_unlock(ea_inode);
1463 return ea_inode;
1466 static struct inode *
1467 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1468 size_t value_len, u32 hash)
1470 struct inode *ea_inode;
1471 struct mb_cache_entry *ce;
1472 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1473 void *ea_data;
1475 if (!ea_inode_cache)
1476 return NULL;
1478 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1479 if (!ce)
1480 return NULL;
1482 ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1483 if (!ea_data) {
1484 mb_cache_entry_put(ea_inode_cache, ce);
1485 return NULL;
1488 while (ce) {
1489 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1490 EXT4_IGET_NORMAL);
1491 if (!IS_ERR(ea_inode) &&
1492 !is_bad_inode(ea_inode) &&
1493 (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1494 i_size_read(ea_inode) == value_len &&
1495 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1496 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1497 value_len) &&
1498 !memcmp(value, ea_data, value_len)) {
1499 mb_cache_entry_touch(ea_inode_cache, ce);
1500 mb_cache_entry_put(ea_inode_cache, ce);
1501 kvfree(ea_data);
1502 return ea_inode;
1505 if (!IS_ERR(ea_inode))
1506 iput(ea_inode);
1507 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1509 kvfree(ea_data);
1510 return NULL;
1514 * Add value of the EA in an inode.
1516 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1517 const void *value, size_t value_len,
1518 struct inode **ret_inode)
1520 struct inode *ea_inode;
1521 u32 hash;
1522 int err;
1524 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1525 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1526 if (ea_inode) {
1527 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1528 if (err) {
1529 iput(ea_inode);
1530 return err;
1533 *ret_inode = ea_inode;
1534 return 0;
1537 /* Create an inode for the EA value */
1538 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1539 if (IS_ERR(ea_inode))
1540 return PTR_ERR(ea_inode);
1542 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1543 if (err) {
1544 ext4_xattr_inode_dec_ref(handle, ea_inode);
1545 iput(ea_inode);
1546 return err;
1549 if (EA_INODE_CACHE(inode))
1550 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1551 ea_inode->i_ino, true /* reusable */);
1553 *ret_inode = ea_inode;
1554 return 0;
1558 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1559 * feature is enabled.
1561 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1563 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1564 struct ext4_xattr_search *s,
1565 handle_t *handle, struct inode *inode,
1566 bool is_block)
1568 struct ext4_xattr_entry *last, *next;
1569 struct ext4_xattr_entry *here = s->here;
1570 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1571 int in_inode = i->in_inode;
1572 struct inode *old_ea_inode = NULL;
1573 struct inode *new_ea_inode = NULL;
1574 size_t old_size, new_size;
1575 int ret;
1577 /* Space used by old and new values. */
1578 old_size = (!s->not_found && !here->e_value_inum) ?
1579 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1580 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1583 * Optimization for the simple case when old and new values have the
1584 * same padded sizes. Not applicable if external inodes are involved.
1586 if (new_size && new_size == old_size) {
1587 size_t offs = le16_to_cpu(here->e_value_offs);
1588 void *val = s->base + offs;
1590 here->e_value_size = cpu_to_le32(i->value_len);
1591 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1592 memset(val, 0, new_size);
1593 } else {
1594 memcpy(val, i->value, i->value_len);
1595 /* Clear padding bytes. */
1596 memset(val + i->value_len, 0, new_size - i->value_len);
1598 goto update_hash;
1601 /* Compute min_offs and last. */
1602 last = s->first;
1603 for (; !IS_LAST_ENTRY(last); last = next) {
1604 next = EXT4_XATTR_NEXT(last);
1605 if ((void *)next >= s->end) {
1606 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1607 ret = -EFSCORRUPTED;
1608 goto out;
1610 if (!last->e_value_inum && last->e_value_size) {
1611 size_t offs = le16_to_cpu(last->e_value_offs);
1612 if (offs < min_offs)
1613 min_offs = offs;
1617 /* Check whether we have enough space. */
1618 if (i->value) {
1619 size_t free;
1621 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1622 if (!s->not_found)
1623 free += EXT4_XATTR_LEN(name_len) + old_size;
1625 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1626 ret = -ENOSPC;
1627 goto out;
1631 * If storing the value in an external inode is an option,
1632 * reserve space for xattr entries/names in the external
1633 * attribute block so that a long value does not occupy the
1634 * whole space and prevent futher entries being added.
1636 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1637 new_size && is_block &&
1638 (min_offs + old_size - new_size) <
1639 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1640 ret = -ENOSPC;
1641 goto out;
1646 * Getting access to old and new ea inodes is subject to failures.
1647 * Finish that work before doing any modifications to the xattr data.
1649 if (!s->not_found && here->e_value_inum) {
1650 ret = ext4_xattr_inode_iget(inode,
1651 le32_to_cpu(here->e_value_inum),
1652 le32_to_cpu(here->e_hash),
1653 &old_ea_inode);
1654 if (ret) {
1655 old_ea_inode = NULL;
1656 goto out;
1659 if (i->value && in_inode) {
1660 WARN_ON_ONCE(!i->value_len);
1662 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1663 if (ret)
1664 goto out;
1666 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1667 i->value_len,
1668 &new_ea_inode);
1669 if (ret) {
1670 new_ea_inode = NULL;
1671 ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1672 goto out;
1676 if (old_ea_inode) {
1677 /* We are ready to release ref count on the old_ea_inode. */
1678 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1679 if (ret) {
1680 /* Release newly required ref count on new_ea_inode. */
1681 if (new_ea_inode) {
1682 int err;
1684 err = ext4_xattr_inode_dec_ref(handle,
1685 new_ea_inode);
1686 if (err)
1687 ext4_warning_inode(new_ea_inode,
1688 "dec ref new_ea_inode err=%d",
1689 err);
1690 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1691 i->value_len);
1693 goto out;
1696 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1697 le32_to_cpu(here->e_value_size));
1700 /* No failures allowed past this point. */
1702 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1703 /* Remove the old value. */
1704 void *first_val = s->base + min_offs;
1705 size_t offs = le16_to_cpu(here->e_value_offs);
1706 void *val = s->base + offs;
1708 memmove(first_val + old_size, first_val, val - first_val);
1709 memset(first_val, 0, old_size);
1710 min_offs += old_size;
1712 /* Adjust all value offsets. */
1713 last = s->first;
1714 while (!IS_LAST_ENTRY(last)) {
1715 size_t o = le16_to_cpu(last->e_value_offs);
1717 if (!last->e_value_inum &&
1718 last->e_value_size && o < offs)
1719 last->e_value_offs = cpu_to_le16(o + old_size);
1720 last = EXT4_XATTR_NEXT(last);
1724 if (!i->value) {
1725 /* Remove old name. */
1726 size_t size = EXT4_XATTR_LEN(name_len);
1728 last = ENTRY((void *)last - size);
1729 memmove(here, (void *)here + size,
1730 (void *)last - (void *)here + sizeof(__u32));
1731 memset(last, 0, size);
1732 } else if (s->not_found) {
1733 /* Insert new name. */
1734 size_t size = EXT4_XATTR_LEN(name_len);
1735 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1737 memmove((void *)here + size, here, rest);
1738 memset(here, 0, size);
1739 here->e_name_index = i->name_index;
1740 here->e_name_len = name_len;
1741 memcpy(here->e_name, i->name, name_len);
1742 } else {
1743 /* This is an update, reset value info. */
1744 here->e_value_inum = 0;
1745 here->e_value_offs = 0;
1746 here->e_value_size = 0;
1749 if (i->value) {
1750 /* Insert new value. */
1751 if (in_inode) {
1752 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1753 } else if (i->value_len) {
1754 void *val = s->base + min_offs - new_size;
1756 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1757 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1758 memset(val, 0, new_size);
1759 } else {
1760 memcpy(val, i->value, i->value_len);
1761 /* Clear padding bytes. */
1762 memset(val + i->value_len, 0,
1763 new_size - i->value_len);
1766 here->e_value_size = cpu_to_le32(i->value_len);
1769 update_hash:
1770 if (i->value) {
1771 __le32 hash = 0;
1773 /* Entry hash calculation. */
1774 if (in_inode) {
1775 __le32 crc32c_hash;
1778 * Feed crc32c hash instead of the raw value for entry
1779 * hash calculation. This is to avoid walking
1780 * potentially long value buffer again.
1782 crc32c_hash = cpu_to_le32(
1783 ext4_xattr_inode_get_hash(new_ea_inode));
1784 hash = ext4_xattr_hash_entry(here->e_name,
1785 here->e_name_len,
1786 &crc32c_hash, 1);
1787 } else if (is_block) {
1788 __le32 *value = s->base + le16_to_cpu(
1789 here->e_value_offs);
1791 hash = ext4_xattr_hash_entry(here->e_name,
1792 here->e_name_len, value,
1793 new_size >> 2);
1795 here->e_hash = hash;
1798 if (is_block)
1799 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1801 ret = 0;
1802 out:
1803 iput(old_ea_inode);
1804 iput(new_ea_inode);
1805 return ret;
1808 struct ext4_xattr_block_find {
1809 struct ext4_xattr_search s;
1810 struct buffer_head *bh;
1813 static int
1814 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1815 struct ext4_xattr_block_find *bs)
1817 struct super_block *sb = inode->i_sb;
1818 int error;
1820 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1821 i->name_index, i->name, i->value, (long)i->value_len);
1823 if (EXT4_I(inode)->i_file_acl) {
1824 /* The inode already has an extended attribute block. */
1825 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1826 if (IS_ERR(bs->bh)) {
1827 error = PTR_ERR(bs->bh);
1828 bs->bh = NULL;
1829 return error;
1831 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1832 atomic_read(&(bs->bh->b_count)),
1833 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1834 error = ext4_xattr_check_block(inode, bs->bh);
1835 if (error)
1836 return error;
1837 /* Find the named attribute. */
1838 bs->s.base = BHDR(bs->bh);
1839 bs->s.first = BFIRST(bs->bh);
1840 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1841 bs->s.here = bs->s.first;
1842 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1843 i->name_index, i->name, 1);
1844 if (error && error != -ENODATA)
1845 return error;
1846 bs->s.not_found = error;
1848 return 0;
1851 static int
1852 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1853 struct ext4_xattr_info *i,
1854 struct ext4_xattr_block_find *bs)
1856 struct super_block *sb = inode->i_sb;
1857 struct buffer_head *new_bh = NULL;
1858 struct ext4_xattr_search s_copy = bs->s;
1859 struct ext4_xattr_search *s = &s_copy;
1860 struct mb_cache_entry *ce = NULL;
1861 int error = 0;
1862 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1863 struct inode *ea_inode = NULL, *tmp_inode;
1864 size_t old_ea_inode_quota = 0;
1865 unsigned int ea_ino;
1868 #define header(x) ((struct ext4_xattr_header *)(x))
1870 if (s->base) {
1871 BUFFER_TRACE(bs->bh, "get_write_access");
1872 error = ext4_journal_get_write_access(handle, bs->bh);
1873 if (error)
1874 goto cleanup;
1875 lock_buffer(bs->bh);
1877 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1878 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1881 * This must happen under buffer lock for
1882 * ext4_xattr_block_set() to reliably detect modified
1883 * block
1885 if (ea_block_cache)
1886 mb_cache_entry_delete(ea_block_cache, hash,
1887 bs->bh->b_blocknr);
1888 ea_bdebug(bs->bh, "modifying in-place");
1889 error = ext4_xattr_set_entry(i, s, handle, inode,
1890 true /* is_block */);
1891 ext4_xattr_block_csum_set(inode, bs->bh);
1892 unlock_buffer(bs->bh);
1893 if (error == -EFSCORRUPTED)
1894 goto bad_block;
1895 if (!error)
1896 error = ext4_handle_dirty_metadata(handle,
1897 inode,
1898 bs->bh);
1899 if (error)
1900 goto cleanup;
1901 goto inserted;
1902 } else {
1903 int offset = (char *)s->here - bs->bh->b_data;
1905 unlock_buffer(bs->bh);
1906 ea_bdebug(bs->bh, "cloning");
1907 s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1908 error = -ENOMEM;
1909 if (s->base == NULL)
1910 goto cleanup;
1911 memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1912 s->first = ENTRY(header(s->base)+1);
1913 header(s->base)->h_refcount = cpu_to_le32(1);
1914 s->here = ENTRY(s->base + offset);
1915 s->end = s->base + bs->bh->b_size;
1918 * If existing entry points to an xattr inode, we need
1919 * to prevent ext4_xattr_set_entry() from decrementing
1920 * ref count on it because the reference belongs to the
1921 * original block. In this case, make the entry look
1922 * like it has an empty value.
1924 if (!s->not_found && s->here->e_value_inum) {
1925 ea_ino = le32_to_cpu(s->here->e_value_inum);
1926 error = ext4_xattr_inode_iget(inode, ea_ino,
1927 le32_to_cpu(s->here->e_hash),
1928 &tmp_inode);
1929 if (error)
1930 goto cleanup;
1932 if (!ext4_test_inode_state(tmp_inode,
1933 EXT4_STATE_LUSTRE_EA_INODE)) {
1935 * Defer quota free call for previous
1936 * inode until success is guaranteed.
1938 old_ea_inode_quota = le32_to_cpu(
1939 s->here->e_value_size);
1941 iput(tmp_inode);
1943 s->here->e_value_inum = 0;
1944 s->here->e_value_size = 0;
1947 } else {
1948 /* Allocate a buffer where we construct the new block. */
1949 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1950 /* assert(header == s->base) */
1951 error = -ENOMEM;
1952 if (s->base == NULL)
1953 goto cleanup;
1954 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1955 header(s->base)->h_blocks = cpu_to_le32(1);
1956 header(s->base)->h_refcount = cpu_to_le32(1);
1957 s->first = ENTRY(header(s->base)+1);
1958 s->here = ENTRY(header(s->base)+1);
1959 s->end = s->base + sb->s_blocksize;
1962 error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1963 if (error == -EFSCORRUPTED)
1964 goto bad_block;
1965 if (error)
1966 goto cleanup;
1968 if (i->value && s->here->e_value_inum) {
1970 * A ref count on ea_inode has been taken as part of the call to
1971 * ext4_xattr_set_entry() above. We would like to drop this
1972 * extra ref but we have to wait until the xattr block is
1973 * initialized and has its own ref count on the ea_inode.
1975 ea_ino = le32_to_cpu(s->here->e_value_inum);
1976 error = ext4_xattr_inode_iget(inode, ea_ino,
1977 le32_to_cpu(s->here->e_hash),
1978 &ea_inode);
1979 if (error) {
1980 ea_inode = NULL;
1981 goto cleanup;
1985 inserted:
1986 if (!IS_LAST_ENTRY(s->first)) {
1987 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1988 &ce);
1989 if (new_bh) {
1990 /* We found an identical block in the cache. */
1991 if (new_bh == bs->bh)
1992 ea_bdebug(new_bh, "keeping");
1993 else {
1994 u32 ref;
1996 WARN_ON_ONCE(dquot_initialize_needed(inode));
1998 /* The old block is released after updating
1999 the inode. */
2000 error = dquot_alloc_block(inode,
2001 EXT4_C2B(EXT4_SB(sb), 1));
2002 if (error)
2003 goto cleanup;
2004 BUFFER_TRACE(new_bh, "get_write_access");
2005 error = ext4_journal_get_write_access(handle,
2006 new_bh);
2007 if (error)
2008 goto cleanup_dquot;
2009 lock_buffer(new_bh);
2011 * We have to be careful about races with
2012 * freeing, rehashing or adding references to
2013 * xattr block. Once we hold buffer lock xattr
2014 * block's state is stable so we can check
2015 * whether the block got freed / rehashed or
2016 * not. Since we unhash mbcache entry under
2017 * buffer lock when freeing / rehashing xattr
2018 * block, checking whether entry is still
2019 * hashed is reliable. Same rules hold for
2020 * e_reusable handling.
2022 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2023 !ce->e_reusable) {
2025 * Undo everything and check mbcache
2026 * again.
2028 unlock_buffer(new_bh);
2029 dquot_free_block(inode,
2030 EXT4_C2B(EXT4_SB(sb),
2031 1));
2032 brelse(new_bh);
2033 mb_cache_entry_put(ea_block_cache, ce);
2034 ce = NULL;
2035 new_bh = NULL;
2036 goto inserted;
2038 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2039 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2040 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2041 ce->e_reusable = 0;
2042 ea_bdebug(new_bh, "reusing; refcount now=%d",
2043 ref);
2044 ext4_xattr_block_csum_set(inode, new_bh);
2045 unlock_buffer(new_bh);
2046 error = ext4_handle_dirty_metadata(handle,
2047 inode,
2048 new_bh);
2049 if (error)
2050 goto cleanup_dquot;
2052 mb_cache_entry_touch(ea_block_cache, ce);
2053 mb_cache_entry_put(ea_block_cache, ce);
2054 ce = NULL;
2055 } else if (bs->bh && s->base == bs->bh->b_data) {
2056 /* We were modifying this block in-place. */
2057 ea_bdebug(bs->bh, "keeping this block");
2058 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2059 new_bh = bs->bh;
2060 get_bh(new_bh);
2061 } else {
2062 /* We need to allocate a new block */
2063 ext4_fsblk_t goal, block;
2065 WARN_ON_ONCE(dquot_initialize_needed(inode));
2067 goal = ext4_group_first_block_no(sb,
2068 EXT4_I(inode)->i_block_group);
2070 /* non-extent files can't have physical blocks past 2^32 */
2071 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2072 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2074 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2075 NULL, &error);
2076 if (error)
2077 goto cleanup;
2079 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2080 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2082 ea_idebug(inode, "creating block %llu",
2083 (unsigned long long)block);
2085 new_bh = sb_getblk(sb, block);
2086 if (unlikely(!new_bh)) {
2087 error = -ENOMEM;
2088 getblk_failed:
2089 ext4_free_blocks(handle, inode, NULL, block, 1,
2090 EXT4_FREE_BLOCKS_METADATA);
2091 goto cleanup;
2093 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2094 ENTRY(header(s->base)+1));
2095 if (error)
2096 goto getblk_failed;
2097 if (ea_inode) {
2098 /* Drop the extra ref on ea_inode. */
2099 error = ext4_xattr_inode_dec_ref(handle,
2100 ea_inode);
2101 if (error)
2102 ext4_warning_inode(ea_inode,
2103 "dec ref error=%d",
2104 error);
2105 iput(ea_inode);
2106 ea_inode = NULL;
2109 lock_buffer(new_bh);
2110 error = ext4_journal_get_create_access(handle, new_bh);
2111 if (error) {
2112 unlock_buffer(new_bh);
2113 error = -EIO;
2114 goto getblk_failed;
2116 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2117 ext4_xattr_block_csum_set(inode, new_bh);
2118 set_buffer_uptodate(new_bh);
2119 unlock_buffer(new_bh);
2120 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2121 error = ext4_handle_dirty_metadata(handle, inode,
2122 new_bh);
2123 if (error)
2124 goto cleanup;
2128 if (old_ea_inode_quota)
2129 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2131 /* Update the inode. */
2132 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2134 /* Drop the previous xattr block. */
2135 if (bs->bh && bs->bh != new_bh) {
2136 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2138 ext4_xattr_release_block(handle, inode, bs->bh,
2139 &ea_inode_array,
2140 0 /* extra_credits */);
2141 ext4_xattr_inode_array_free(ea_inode_array);
2143 error = 0;
2145 cleanup:
2146 if (ea_inode) {
2147 int error2;
2149 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2150 if (error2)
2151 ext4_warning_inode(ea_inode, "dec ref error=%d",
2152 error2);
2154 /* If there was an error, revert the quota charge. */
2155 if (error)
2156 ext4_xattr_inode_free_quota(inode, ea_inode,
2157 i_size_read(ea_inode));
2158 iput(ea_inode);
2160 if (ce)
2161 mb_cache_entry_put(ea_block_cache, ce);
2162 brelse(new_bh);
2163 if (!(bs->bh && s->base == bs->bh->b_data))
2164 kfree(s->base);
2166 return error;
2168 cleanup_dquot:
2169 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2170 goto cleanup;
2172 bad_block:
2173 EXT4_ERROR_INODE(inode, "bad block %llu",
2174 EXT4_I(inode)->i_file_acl);
2175 goto cleanup;
2177 #undef header
2180 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2181 struct ext4_xattr_ibody_find *is)
2183 struct ext4_xattr_ibody_header *header;
2184 struct ext4_inode *raw_inode;
2185 int error;
2187 if (EXT4_I(inode)->i_extra_isize == 0)
2188 return 0;
2189 raw_inode = ext4_raw_inode(&is->iloc);
2190 header = IHDR(inode, raw_inode);
2191 is->s.base = is->s.first = IFIRST(header);
2192 is->s.here = is->s.first;
2193 is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2194 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2195 error = xattr_check_inode(inode, header, is->s.end);
2196 if (error)
2197 return error;
2198 /* Find the named attribute. */
2199 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2200 i->name_index, i->name, 0);
2201 if (error && error != -ENODATA)
2202 return error;
2203 is->s.not_found = error;
2205 return 0;
2208 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2209 struct ext4_xattr_info *i,
2210 struct ext4_xattr_ibody_find *is)
2212 struct ext4_xattr_ibody_header *header;
2213 struct ext4_xattr_search *s = &is->s;
2214 int error;
2216 if (EXT4_I(inode)->i_extra_isize == 0)
2217 return -ENOSPC;
2218 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2219 if (error)
2220 return error;
2221 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2222 if (!IS_LAST_ENTRY(s->first)) {
2223 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2224 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2225 } else {
2226 header->h_magic = cpu_to_le32(0);
2227 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2229 return 0;
2232 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2233 struct ext4_xattr_info *i,
2234 struct ext4_xattr_ibody_find *is)
2236 struct ext4_xattr_ibody_header *header;
2237 struct ext4_xattr_search *s = &is->s;
2238 int error;
2240 if (EXT4_I(inode)->i_extra_isize == 0)
2241 return -ENOSPC;
2242 error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2243 if (error)
2244 return error;
2245 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2246 if (!IS_LAST_ENTRY(s->first)) {
2247 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2248 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2249 } else {
2250 header->h_magic = cpu_to_le32(0);
2251 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2253 return 0;
2256 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2257 struct ext4_xattr_info *i)
2259 void *value;
2261 /* When e_value_inum is set the value is stored externally. */
2262 if (s->here->e_value_inum)
2263 return 0;
2264 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2265 return 0;
2266 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2267 return !memcmp(value, i->value, i->value_len);
2270 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2272 struct buffer_head *bh;
2273 int error;
2275 if (!EXT4_I(inode)->i_file_acl)
2276 return NULL;
2277 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2278 if (IS_ERR(bh))
2279 return bh;
2280 error = ext4_xattr_check_block(inode, bh);
2281 if (error) {
2282 brelse(bh);
2283 return ERR_PTR(error);
2285 return bh;
2289 * ext4_xattr_set_handle()
2291 * Create, replace or remove an extended attribute for this inode. Value
2292 * is NULL to remove an existing extended attribute, and non-NULL to
2293 * either replace an existing extended attribute, or create a new extended
2294 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2295 * specify that an extended attribute must exist and must not exist
2296 * previous to the call, respectively.
2298 * Returns 0, or a negative error number on failure.
2301 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2302 const char *name, const void *value, size_t value_len,
2303 int flags)
2305 struct ext4_xattr_info i = {
2306 .name_index = name_index,
2307 .name = name,
2308 .value = value,
2309 .value_len = value_len,
2310 .in_inode = 0,
2312 struct ext4_xattr_ibody_find is = {
2313 .s = { .not_found = -ENODATA, },
2315 struct ext4_xattr_block_find bs = {
2316 .s = { .not_found = -ENODATA, },
2318 int no_expand;
2319 int error;
2321 if (!name)
2322 return -EINVAL;
2323 if (strlen(name) > 255)
2324 return -ERANGE;
2326 ext4_write_lock_xattr(inode, &no_expand);
2328 /* Check journal credits under write lock. */
2329 if (ext4_handle_valid(handle)) {
2330 struct buffer_head *bh;
2331 int credits;
2333 bh = ext4_xattr_get_block(inode);
2334 if (IS_ERR(bh)) {
2335 error = PTR_ERR(bh);
2336 goto cleanup;
2339 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2340 value_len,
2341 flags & XATTR_CREATE);
2342 brelse(bh);
2344 if (!ext4_handle_has_enough_credits(handle, credits)) {
2345 error = -ENOSPC;
2346 goto cleanup;
2350 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2351 if (error)
2352 goto cleanup;
2354 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2355 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2356 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2357 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2360 error = ext4_xattr_ibody_find(inode, &i, &is);
2361 if (error)
2362 goto cleanup;
2363 if (is.s.not_found)
2364 error = ext4_xattr_block_find(inode, &i, &bs);
2365 if (error)
2366 goto cleanup;
2367 if (is.s.not_found && bs.s.not_found) {
2368 error = -ENODATA;
2369 if (flags & XATTR_REPLACE)
2370 goto cleanup;
2371 error = 0;
2372 if (!value)
2373 goto cleanup;
2374 } else {
2375 error = -EEXIST;
2376 if (flags & XATTR_CREATE)
2377 goto cleanup;
2380 if (!value) {
2381 if (!is.s.not_found)
2382 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2383 else if (!bs.s.not_found)
2384 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2385 } else {
2386 error = 0;
2387 /* Xattr value did not change? Save us some work and bail out */
2388 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2389 goto cleanup;
2390 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2391 goto cleanup;
2393 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2394 (EXT4_XATTR_SIZE(i.value_len) >
2395 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2396 i.in_inode = 1;
2397 retry_inode:
2398 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2399 if (!error && !bs.s.not_found) {
2400 i.value = NULL;
2401 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2402 } else if (error == -ENOSPC) {
2403 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2404 brelse(bs.bh);
2405 bs.bh = NULL;
2406 error = ext4_xattr_block_find(inode, &i, &bs);
2407 if (error)
2408 goto cleanup;
2410 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2411 if (!error && !is.s.not_found) {
2412 i.value = NULL;
2413 error = ext4_xattr_ibody_set(handle, inode, &i,
2414 &is);
2415 } else if (error == -ENOSPC) {
2417 * Xattr does not fit in the block, store at
2418 * external inode if possible.
2420 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2421 !i.in_inode) {
2422 i.in_inode = 1;
2423 goto retry_inode;
2428 if (!error) {
2429 ext4_xattr_update_super_block(handle, inode->i_sb);
2430 inode->i_ctime = current_time(inode);
2431 if (!value)
2432 no_expand = 0;
2433 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2435 * The bh is consumed by ext4_mark_iloc_dirty, even with
2436 * error != 0.
2438 is.iloc.bh = NULL;
2439 if (IS_SYNC(inode))
2440 ext4_handle_sync(handle);
2443 cleanup:
2444 brelse(is.iloc.bh);
2445 brelse(bs.bh);
2446 ext4_write_unlock_xattr(inode, &no_expand);
2447 return error;
2450 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2451 bool is_create, int *credits)
2453 struct buffer_head *bh;
2454 int err;
2456 *credits = 0;
2458 if (!EXT4_SB(inode->i_sb)->s_journal)
2459 return 0;
2461 down_read(&EXT4_I(inode)->xattr_sem);
2463 bh = ext4_xattr_get_block(inode);
2464 if (IS_ERR(bh)) {
2465 err = PTR_ERR(bh);
2466 } else {
2467 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2468 value_len, is_create);
2469 brelse(bh);
2470 err = 0;
2473 up_read(&EXT4_I(inode)->xattr_sem);
2474 return err;
2478 * ext4_xattr_set()
2480 * Like ext4_xattr_set_handle, but start from an inode. This extended
2481 * attribute modification is a filesystem transaction by itself.
2483 * Returns 0, or a negative error number on failure.
2486 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2487 const void *value, size_t value_len, int flags)
2489 handle_t *handle;
2490 struct super_block *sb = inode->i_sb;
2491 int error, retries = 0;
2492 int credits;
2494 error = dquot_initialize(inode);
2495 if (error)
2496 return error;
2498 retry:
2499 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2500 &credits);
2501 if (error)
2502 return error;
2504 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2505 if (IS_ERR(handle)) {
2506 error = PTR_ERR(handle);
2507 } else {
2508 int error2;
2510 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2511 value, value_len, flags);
2512 error2 = ext4_journal_stop(handle);
2513 if (error == -ENOSPC &&
2514 ext4_should_retry_alloc(sb, &retries))
2515 goto retry;
2516 if (error == 0)
2517 error = error2;
2520 return error;
2524 * Shift the EA entries in the inode to create space for the increased
2525 * i_extra_isize.
2527 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2528 int value_offs_shift, void *to,
2529 void *from, size_t n)
2531 struct ext4_xattr_entry *last = entry;
2532 int new_offs;
2534 /* We always shift xattr headers further thus offsets get lower */
2535 BUG_ON(value_offs_shift > 0);
2537 /* Adjust the value offsets of the entries */
2538 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2539 if (!last->e_value_inum && last->e_value_size) {
2540 new_offs = le16_to_cpu(last->e_value_offs) +
2541 value_offs_shift;
2542 last->e_value_offs = cpu_to_le16(new_offs);
2545 /* Shift the entries by n bytes */
2546 memmove(to, from, n);
2550 * Move xattr pointed to by 'entry' from inode into external xattr block
2552 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2553 struct ext4_inode *raw_inode,
2554 struct ext4_xattr_entry *entry)
2556 struct ext4_xattr_ibody_find *is = NULL;
2557 struct ext4_xattr_block_find *bs = NULL;
2558 char *buffer = NULL, *b_entry_name = NULL;
2559 size_t value_size = le32_to_cpu(entry->e_value_size);
2560 struct ext4_xattr_info i = {
2561 .value = NULL,
2562 .value_len = 0,
2563 .name_index = entry->e_name_index,
2564 .in_inode = !!entry->e_value_inum,
2566 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2567 int error;
2569 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2570 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2571 buffer = kmalloc(value_size, GFP_NOFS);
2572 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2573 if (!is || !bs || !buffer || !b_entry_name) {
2574 error = -ENOMEM;
2575 goto out;
2578 is->s.not_found = -ENODATA;
2579 bs->s.not_found = -ENODATA;
2580 is->iloc.bh = NULL;
2581 bs->bh = NULL;
2583 /* Save the entry name and the entry value */
2584 if (entry->e_value_inum) {
2585 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2586 if (error)
2587 goto out;
2588 } else {
2589 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2590 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2593 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2594 b_entry_name[entry->e_name_len] = '\0';
2595 i.name = b_entry_name;
2597 error = ext4_get_inode_loc(inode, &is->iloc);
2598 if (error)
2599 goto out;
2601 error = ext4_xattr_ibody_find(inode, &i, is);
2602 if (error)
2603 goto out;
2605 /* Remove the chosen entry from the inode */
2606 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2607 if (error)
2608 goto out;
2610 i.value = buffer;
2611 i.value_len = value_size;
2612 error = ext4_xattr_block_find(inode, &i, bs);
2613 if (error)
2614 goto out;
2616 /* Add entry which was removed from the inode into the block */
2617 error = ext4_xattr_block_set(handle, inode, &i, bs);
2618 if (error)
2619 goto out;
2620 error = 0;
2621 out:
2622 kfree(b_entry_name);
2623 kfree(buffer);
2624 if (is)
2625 brelse(is->iloc.bh);
2626 if (bs)
2627 brelse(bs->bh);
2628 kfree(is);
2629 kfree(bs);
2631 return error;
2634 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2635 struct ext4_inode *raw_inode,
2636 int isize_diff, size_t ifree,
2637 size_t bfree, int *total_ino)
2639 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2640 struct ext4_xattr_entry *small_entry;
2641 struct ext4_xattr_entry *entry;
2642 struct ext4_xattr_entry *last;
2643 unsigned int entry_size; /* EA entry size */
2644 unsigned int total_size; /* EA entry size + value size */
2645 unsigned int min_total_size;
2646 int error;
2648 while (isize_diff > ifree) {
2649 entry = NULL;
2650 small_entry = NULL;
2651 min_total_size = ~0U;
2652 last = IFIRST(header);
2653 /* Find the entry best suited to be pushed into EA block */
2654 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2655 /* never move system.data out of the inode */
2656 if ((last->e_name_len == 4) &&
2657 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2658 !memcmp(last->e_name, "data", 4))
2659 continue;
2660 total_size = EXT4_XATTR_LEN(last->e_name_len);
2661 if (!last->e_value_inum)
2662 total_size += EXT4_XATTR_SIZE(
2663 le32_to_cpu(last->e_value_size));
2664 if (total_size <= bfree &&
2665 total_size < min_total_size) {
2666 if (total_size + ifree < isize_diff) {
2667 small_entry = last;
2668 } else {
2669 entry = last;
2670 min_total_size = total_size;
2675 if (entry == NULL) {
2676 if (small_entry == NULL)
2677 return -ENOSPC;
2678 entry = small_entry;
2681 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2682 total_size = entry_size;
2683 if (!entry->e_value_inum)
2684 total_size += EXT4_XATTR_SIZE(
2685 le32_to_cpu(entry->e_value_size));
2686 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2687 entry);
2688 if (error)
2689 return error;
2691 *total_ino -= entry_size;
2692 ifree += total_size;
2693 bfree -= total_size;
2696 return 0;
2700 * Expand an inode by new_extra_isize bytes when EAs are present.
2701 * Returns 0 on success or negative error number on failure.
2703 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2704 struct ext4_inode *raw_inode, handle_t *handle)
2706 struct ext4_xattr_ibody_header *header;
2707 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2708 static unsigned int mnt_count;
2709 size_t min_offs;
2710 size_t ifree, bfree;
2711 int total_ino;
2712 void *base, *end;
2713 int error = 0, tried_min_extra_isize = 0;
2714 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2715 int isize_diff; /* How much do we need to grow i_extra_isize */
2717 retry:
2718 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2719 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2720 return 0;
2722 header = IHDR(inode, raw_inode);
2725 * Check if enough free space is available in the inode to shift the
2726 * entries ahead by new_extra_isize.
2729 base = IFIRST(header);
2730 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2731 min_offs = end - base;
2732 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2734 error = xattr_check_inode(inode, header, end);
2735 if (error)
2736 goto cleanup;
2738 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2739 if (ifree >= isize_diff)
2740 goto shift;
2743 * Enough free space isn't available in the inode, check if
2744 * EA block can hold new_extra_isize bytes.
2746 if (EXT4_I(inode)->i_file_acl) {
2747 struct buffer_head *bh;
2749 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2750 if (IS_ERR(bh)) {
2751 error = PTR_ERR(bh);
2752 goto cleanup;
2754 error = ext4_xattr_check_block(inode, bh);
2755 if (error) {
2756 brelse(bh);
2757 goto cleanup;
2759 base = BHDR(bh);
2760 end = bh->b_data + bh->b_size;
2761 min_offs = end - base;
2762 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2763 NULL);
2764 brelse(bh);
2765 if (bfree + ifree < isize_diff) {
2766 if (!tried_min_extra_isize && s_min_extra_isize) {
2767 tried_min_extra_isize++;
2768 new_extra_isize = s_min_extra_isize;
2769 goto retry;
2771 error = -ENOSPC;
2772 goto cleanup;
2774 } else {
2775 bfree = inode->i_sb->s_blocksize;
2778 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2779 isize_diff, ifree, bfree,
2780 &total_ino);
2781 if (error) {
2782 if (error == -ENOSPC && !tried_min_extra_isize &&
2783 s_min_extra_isize) {
2784 tried_min_extra_isize++;
2785 new_extra_isize = s_min_extra_isize;
2786 goto retry;
2788 goto cleanup;
2790 shift:
2791 /* Adjust the offsets and shift the remaining entries ahead */
2792 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2793 - new_extra_isize, (void *)raw_inode +
2794 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2795 (void *)header, total_ino);
2796 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2798 cleanup:
2799 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2800 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2801 inode->i_ino);
2802 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2804 return error;
2807 #define EIA_INCR 16 /* must be 2^n */
2808 #define EIA_MASK (EIA_INCR - 1)
2810 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2811 * If @ea_inode_array is new or full it will be grown and the old
2812 * contents copied over.
2814 static int
2815 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2816 struct inode *inode)
2818 if (*ea_inode_array == NULL) {
2820 * Start with 15 inodes, so it fits into a power-of-two size.
2821 * If *ea_inode_array is NULL, this is essentially offsetof()
2823 (*ea_inode_array) =
2824 kmalloc(offsetof(struct ext4_xattr_inode_array,
2825 inodes[EIA_MASK]),
2826 GFP_NOFS);
2827 if (*ea_inode_array == NULL)
2828 return -ENOMEM;
2829 (*ea_inode_array)->count = 0;
2830 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2831 /* expand the array once all 15 + n * 16 slots are full */
2832 struct ext4_xattr_inode_array *new_array = NULL;
2833 int count = (*ea_inode_array)->count;
2835 /* if new_array is NULL, this is essentially offsetof() */
2836 new_array = kmalloc(
2837 offsetof(struct ext4_xattr_inode_array,
2838 inodes[count + EIA_INCR]),
2839 GFP_NOFS);
2840 if (new_array == NULL)
2841 return -ENOMEM;
2842 memcpy(new_array, *ea_inode_array,
2843 offsetof(struct ext4_xattr_inode_array, inodes[count]));
2844 kfree(*ea_inode_array);
2845 *ea_inode_array = new_array;
2847 (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2848 return 0;
2852 * ext4_xattr_delete_inode()
2854 * Free extended attribute resources associated with this inode. Traverse
2855 * all entries and decrement reference on any xattr inodes associated with this
2856 * inode. This is called immediately before an inode is freed. We have exclusive
2857 * access to the inode. If an orphan inode is deleted it will also release its
2858 * references on xattr block and xattr inodes.
2860 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2861 struct ext4_xattr_inode_array **ea_inode_array,
2862 int extra_credits)
2864 struct buffer_head *bh = NULL;
2865 struct ext4_xattr_ibody_header *header;
2866 struct ext4_iloc iloc = { .bh = NULL };
2867 struct ext4_xattr_entry *entry;
2868 struct inode *ea_inode;
2869 int error;
2871 error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2872 NULL /* bh */,
2873 false /* dirty */,
2874 false /* block_csum */);
2875 if (error) {
2876 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2877 goto cleanup;
2880 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2881 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2883 error = ext4_get_inode_loc(inode, &iloc);
2884 if (error) {
2885 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2886 goto cleanup;
2889 error = ext4_journal_get_write_access(handle, iloc.bh);
2890 if (error) {
2891 EXT4_ERROR_INODE(inode, "write access (error %d)",
2892 error);
2893 goto cleanup;
2896 header = IHDR(inode, ext4_raw_inode(&iloc));
2897 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2898 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2899 IFIRST(header),
2900 false /* block_csum */,
2901 ea_inode_array,
2902 extra_credits,
2903 false /* skip_quota */);
2906 if (EXT4_I(inode)->i_file_acl) {
2907 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2908 if (IS_ERR(bh)) {
2909 error = PTR_ERR(bh);
2910 if (error == -EIO)
2911 EXT4_ERROR_INODE(inode, "block %llu read error",
2912 EXT4_I(inode)->i_file_acl);
2913 bh = NULL;
2914 goto cleanup;
2916 error = ext4_xattr_check_block(inode, bh);
2917 if (error)
2918 goto cleanup;
2920 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2921 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2922 entry = EXT4_XATTR_NEXT(entry)) {
2923 if (!entry->e_value_inum)
2924 continue;
2925 error = ext4_xattr_inode_iget(inode,
2926 le32_to_cpu(entry->e_value_inum),
2927 le32_to_cpu(entry->e_hash),
2928 &ea_inode);
2929 if (error)
2930 continue;
2931 ext4_xattr_inode_free_quota(inode, ea_inode,
2932 le32_to_cpu(entry->e_value_size));
2933 iput(ea_inode);
2938 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2939 extra_credits);
2941 * Update i_file_acl value in the same transaction that releases
2942 * block.
2944 EXT4_I(inode)->i_file_acl = 0;
2945 error = ext4_mark_inode_dirty(handle, inode);
2946 if (error) {
2947 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2948 error);
2949 goto cleanup;
2952 error = 0;
2953 cleanup:
2954 brelse(iloc.bh);
2955 brelse(bh);
2956 return error;
2959 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2961 int idx;
2963 if (ea_inode_array == NULL)
2964 return;
2966 for (idx = 0; idx < ea_inode_array->count; ++idx)
2967 iput(ea_inode_array->inodes[idx]);
2968 kfree(ea_inode_array);
2972 * ext4_xattr_block_cache_insert()
2974 * Create a new entry in the extended attribute block cache, and insert
2975 * it unless such an entry is already in the cache.
2977 * Returns 0, or a negative error number on failure.
2979 static void
2980 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2981 struct buffer_head *bh)
2983 struct ext4_xattr_header *header = BHDR(bh);
2984 __u32 hash = le32_to_cpu(header->h_hash);
2985 int reusable = le32_to_cpu(header->h_refcount) <
2986 EXT4_XATTR_REFCOUNT_MAX;
2987 int error;
2989 if (!ea_block_cache)
2990 return;
2991 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2992 bh->b_blocknr, reusable);
2993 if (error) {
2994 if (error == -EBUSY)
2995 ea_bdebug(bh, "already in cache");
2996 } else
2997 ea_bdebug(bh, "inserting [%x]", (int)hash);
3001 * ext4_xattr_cmp()
3003 * Compare two extended attribute blocks for equality.
3005 * Returns 0 if the blocks are equal, 1 if they differ, and
3006 * a negative error number on errors.
3008 static int
3009 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3010 struct ext4_xattr_header *header2)
3012 struct ext4_xattr_entry *entry1, *entry2;
3014 entry1 = ENTRY(header1+1);
3015 entry2 = ENTRY(header2+1);
3016 while (!IS_LAST_ENTRY(entry1)) {
3017 if (IS_LAST_ENTRY(entry2))
3018 return 1;
3019 if (entry1->e_hash != entry2->e_hash ||
3020 entry1->e_name_index != entry2->e_name_index ||
3021 entry1->e_name_len != entry2->e_name_len ||
3022 entry1->e_value_size != entry2->e_value_size ||
3023 entry1->e_value_inum != entry2->e_value_inum ||
3024 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3025 return 1;
3026 if (!entry1->e_value_inum &&
3027 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3028 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3029 le32_to_cpu(entry1->e_value_size)))
3030 return 1;
3032 entry1 = EXT4_XATTR_NEXT(entry1);
3033 entry2 = EXT4_XATTR_NEXT(entry2);
3035 if (!IS_LAST_ENTRY(entry2))
3036 return 1;
3037 return 0;
3041 * ext4_xattr_block_cache_find()
3043 * Find an identical extended attribute block.
3045 * Returns a pointer to the block found, or NULL if such a block was
3046 * not found or an error occurred.
3048 static struct buffer_head *
3049 ext4_xattr_block_cache_find(struct inode *inode,
3050 struct ext4_xattr_header *header,
3051 struct mb_cache_entry **pce)
3053 __u32 hash = le32_to_cpu(header->h_hash);
3054 struct mb_cache_entry *ce;
3055 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3057 if (!ea_block_cache)
3058 return NULL;
3059 if (!header->h_hash)
3060 return NULL; /* never share */
3061 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3062 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3063 while (ce) {
3064 struct buffer_head *bh;
3066 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3067 if (IS_ERR(bh)) {
3068 if (PTR_ERR(bh) == -ENOMEM)
3069 return NULL;
3070 bh = NULL;
3071 EXT4_ERROR_INODE(inode, "block %lu read error",
3072 (unsigned long)ce->e_value);
3073 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3074 *pce = ce;
3075 return bh;
3077 brelse(bh);
3078 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3080 return NULL;
3083 #define NAME_HASH_SHIFT 5
3084 #define VALUE_HASH_SHIFT 16
3087 * ext4_xattr_hash_entry()
3089 * Compute the hash of an extended attribute.
3091 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3092 size_t value_count)
3094 __u32 hash = 0;
3096 while (name_len--) {
3097 hash = (hash << NAME_HASH_SHIFT) ^
3098 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3099 *name++;
3101 while (value_count--) {
3102 hash = (hash << VALUE_HASH_SHIFT) ^
3103 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3104 le32_to_cpu(*value++);
3106 return cpu_to_le32(hash);
3109 #undef NAME_HASH_SHIFT
3110 #undef VALUE_HASH_SHIFT
3112 #define BLOCK_HASH_SHIFT 16
3115 * ext4_xattr_rehash()
3117 * Re-compute the extended attribute hash value after an entry has changed.
3119 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3121 struct ext4_xattr_entry *here;
3122 __u32 hash = 0;
3124 here = ENTRY(header+1);
3125 while (!IS_LAST_ENTRY(here)) {
3126 if (!here->e_hash) {
3127 /* Block is not shared if an entry's hash value == 0 */
3128 hash = 0;
3129 break;
3131 hash = (hash << BLOCK_HASH_SHIFT) ^
3132 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3133 le32_to_cpu(here->e_hash);
3134 here = EXT4_XATTR_NEXT(here);
3136 header->h_hash = cpu_to_le32(hash);
3139 #undef BLOCK_HASH_SHIFT
3141 #define HASH_BUCKET_BITS 10
3143 struct mb_cache *
3144 ext4_xattr_create_cache(void)
3146 return mb_cache_create(HASH_BUCKET_BITS);
3149 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3151 if (cache)
3152 mb_cache_destroy(cache);