Debian Codegraph @

fs/ext4/xattr.c

raw
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/xattr.c
4 *
5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <[email protected]>
6 *
7 * Fix by Harrison Xing <[email protected]>.
8 * Ext4 code with a lot of help from Eric Jarman <[email protected]>.
9 * Extended attributes for symlinks and special files added per
10 * suggestion of Luka Renko <[email protected]>.
11 * xattr consolidation Copyright (c) 2004 James Morris <[email protected]>,
12 * Red Hat Inc.
13 * ea-in-inode support by Alex Tomas <[email protected]> aka bzzz
14 * and Andreas Gruenbacher <[email protected]>.
15 */
16
17 /*
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.
25 *
26 * The attributes in inodes and on blocks have a different header; the entries
27 * are stored in the same format:
28 *
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 * +------------------+
40 *
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.
44 *
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.
52 */
53
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"
64
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
76
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 __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value,
85 size_t value_count);
86 static void ext4_xattr_rehash(struct ext4_xattr_header *);
87
88 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
89 [EXT4_XATTR_INDEX_USER] = &ext4_xattr_user_handler,
90 #ifdef CONFIG_EXT4_FS_POSIX_ACL
91 [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS] = &nop_posix_acl_access,
92 [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &nop_posix_acl_default,
93 #endif
94 [EXT4_XATTR_INDEX_TRUSTED] = &ext4_xattr_trusted_handler,
95 #ifdef CONFIG_EXT4_FS_SECURITY
96 [EXT4_XATTR_INDEX_SECURITY] = &ext4_xattr_security_handler,
97 #endif
98 [EXT4_XATTR_INDEX_HURD] = &ext4_xattr_hurd_handler,
99 };
100
101 const struct xattr_handler * const ext4_xattr_handlers[] = {
102 &ext4_xattr_user_handler,
103 &ext4_xattr_trusted_handler,
104 #ifdef CONFIG_EXT4_FS_SECURITY
105 &ext4_xattr_security_handler,
106 #endif
107 &ext4_xattr_hurd_handler,
108 NULL
109 };
110
111 #define EA_BLOCK_CACHE(inode) (((struct ext4_sb_info *) \
112 inode->i_sb->s_fs_info)->s_ea_block_cache)
113
114 #define EA_INODE_CACHE(inode) (((struct ext4_sb_info *) \
115 inode->i_sb->s_fs_info)->s_ea_inode_cache)
116
117 static int
118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119 struct inode *inode);
120
121 #ifdef CONFIG_LOCKDEP
122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
123 {
124 struct ext4_inode_info *ei = EXT4_I(ea_inode);
125
126 lockdep_set_subclass(&ea_inode->i_rwsem, 1);
127 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
128 lockdep_set_subclass(&ei->i_data_sem, I_DATA_SEM_EA);
129 }
130 #endif
131
132 static __le32 ext4_xattr_block_csum(struct inode *inode,
133 sector_t block_nr,
134 struct ext4_xattr_header *hdr)
135 {
136 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
137 __u32 csum;
138 __le64 dsk_block_nr = cpu_to_le64(block_nr);
139 __u32 dummy_csum = 0;
140 int offset = offsetof(struct ext4_xattr_header, h_checksum);
141
142 csum = ext4_chksum(sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
143 sizeof(dsk_block_nr));
144 csum = ext4_chksum(csum, (__u8 *)hdr, offset);
145 csum = ext4_chksum(csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
146 offset += sizeof(dummy_csum);
147 csum = ext4_chksum(csum, (__u8 *)hdr + offset,
148 EXT4_BLOCK_SIZE(inode->i_sb) - offset);
149
150 return cpu_to_le32(csum);
151 }
152
153 static int ext4_xattr_block_csum_verify(struct inode *inode,
154 struct buffer_head *bh)
155 {
156 struct ext4_xattr_header *hdr = BHDR(bh);
157 int ret = 1;
158
159 if (ext4_has_feature_metadata_csum(inode->i_sb)) {
160 lock_buffer(bh);
161 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
162 bh->b_blocknr, hdr));
163 unlock_buffer(bh);
164 }
165 return ret;
166 }
167
168 static void ext4_xattr_block_csum_set(struct inode *inode,
169 struct buffer_head *bh)
170 {
171 if (ext4_has_feature_metadata_csum(inode->i_sb))
172 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
173 bh->b_blocknr, BHDR(bh));
174 }
175
176 static inline const char *ext4_xattr_prefix(int name_index,
177 struct dentry *dentry)
178 {
179 const struct xattr_handler *handler = NULL;
180
181 if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
182 handler = ext4_xattr_handler_map[name_index];
183
184 if (!xattr_handler_can_list(handler, dentry))
185 return NULL;
186
187 return xattr_prefix(handler);
188 }
189
190 static int
191 check_xattrs(struct inode *inode, struct buffer_head *bh,
192 struct ext4_xattr_entry *entry, void *end, void *value_start,
193 const char *function, unsigned int line)
194 {
195 struct ext4_xattr_entry *e = entry;
196 int err = -EFSCORRUPTED;
197 char *err_str;
198
199 if (bh) {
200 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
201 BHDR(bh)->h_blocks != cpu_to_le32(1)) {
202 err_str = "invalid header";
203 goto errout;
204 }
205 if (buffer_verified(bh))
206 return 0;
207 if (!ext4_xattr_block_csum_verify(inode, bh)) {
208 err = -EFSBADCRC;
209 err_str = "invalid checksum";
210 goto errout;
211 }
212 } else {
213 struct ext4_xattr_ibody_header *header = value_start;
214
215 header -= 1;
216 if (end - (void *)header < sizeof(*header) + sizeof(u32)) {
217 err_str = "in-inode xattr block too small";
218 goto errout;
219 }
220 if (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
221 err_str = "bad magic number in in-inode xattr";
222 goto errout;
223 }
224 }
225
226 /* Find the end of the names list */
227 while (!IS_LAST_ENTRY(e)) {
228 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
229 if ((void *)next + sizeof(u32) > end) {
230 err_str = "e_name out of bounds";
231 goto errout;
232 }
233 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len) {
234 err_str = "bad e_name length";
235 goto errout;
236 }
237 e = next;
238 }
239
240 /* Check the values */
241 while (!IS_LAST_ENTRY(entry)) {
242 u32 size = le32_to_cpu(entry->e_value_size);
243 unsigned long ea_ino = le32_to_cpu(entry->e_value_inum);
244
245 if (!ext4_has_feature_ea_inode(inode->i_sb) && ea_ino) {
246 err_str = "ea_inode specified without ea_inode feature enabled";
247 goto errout;
248 }
249 if (ea_ino && ((ea_ino == EXT4_ROOT_INO) ||
250 !ext4_valid_inum(inode->i_sb, ea_ino))) {
251 err_str = "invalid ea_ino";
252 goto errout;
253 }
254 if (ea_ino && !size) {
255 err_str = "invalid size in ea xattr";
256 goto errout;
257 }
258 if (size > EXT4_XATTR_SIZE_MAX) {
259 err_str = "e_value size too large";
260 goto errout;
261 }
262
263 if (size != 0 && entry->e_value_inum == 0) {
264 u16 offs = le16_to_cpu(entry->e_value_offs);
265 void *value;
266
267 /*
268 * The value cannot overlap the names, and the value
269 * with padding cannot extend beyond 'end'. Check both
270 * the padded and unpadded sizes, since the size may
271 * overflow to 0 when adding padding.
272 */
273 if (offs > end - value_start) {
274 err_str = "e_value out of bounds";
275 goto errout;
276 }
277 value = value_start + offs;
278 if (value < (void *)e + sizeof(u32) ||
279 size > end - value ||
280 EXT4_XATTR_SIZE(size) > end - value) {
281 err_str = "overlapping e_value ";
282 goto errout;
283 }
284 }
285 entry = EXT4_XATTR_NEXT(entry);
286 }
287 if (bh)
288 set_buffer_verified(bh);
289 return 0;
290
291 errout:
292 if (bh)
293 __ext4_error_inode(inode, function, line, 0, -err,
294 "corrupted xattr block %llu: %s",
295 (unsigned long long) bh->b_blocknr,
296 err_str);
297 else
298 __ext4_error_inode(inode, function, line, 0, -err,
299 "corrupted in-inode xattr: %s", err_str);
300 return err;
301 }
302
303 static inline int
304 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
305 const char *function, unsigned int line)
306 {
307 return check_xattrs(inode, bh, BFIRST(bh), bh->b_data + bh->b_size,
308 bh->b_data, function, line);
309 }
310
311 #define ext4_xattr_check_block(inode, bh) \
312 __ext4_xattr_check_block((inode), (bh), __func__, __LINE__)
313
314
315 int
316 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
317 void *end, const char *function, unsigned int line)
318 {
319 return check_xattrs(inode, NULL, IFIRST(header), end, IFIRST(header),
320 function, line);
321 }
322
323 static int
324 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
325 void *end, int name_index, const char *name, int sorted)
326 {
327 struct ext4_xattr_entry *entry, *next;
328 size_t name_len;
329 int cmp = 1;
330
331 if (name == NULL)
332 return -EINVAL;
333 name_len = strlen(name);
334 for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
335 next = EXT4_XATTR_NEXT(entry);
336 if ((void *) next >= end) {
337 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
338 return -EFSCORRUPTED;
339 }
340 cmp = name_index - entry->e_name_index;
341 if (!cmp)
342 cmp = name_len - entry->e_name_len;
343 if (!cmp)
344 cmp = memcmp(name, entry->e_name, name_len);
345 if (!cmp || (cmp < 0 && sorted))
346 break;
347 }
348 *pentry = entry;
349 return cmp ? -ENODATA : 0;
350 }
351
352 static u32
353 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
354 {
355 return ext4_chksum(sbi->s_csum_seed, buffer, size);
356 }
357
358 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
359 {
360 return ((u64) inode_get_ctime_sec(ea_inode) << 32) |
361 (u32) inode_peek_iversion_raw(ea_inode);
362 }
363
364 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
365 {
366 inode_set_ctime(ea_inode, (u32)(ref_count >> 32), 0);
367 inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
368 }
369
370 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
371 {
372 return (u32) inode_get_atime_sec(ea_inode);
373 }
374
375 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
376 {
377 inode_set_atime(ea_inode, hash, 0);
378 }
379
380 /*
381 * Read the EA value from an inode.
382 */
383 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
384 {
385 int blocksize = 1 << ea_inode->i_blkbits;
386 int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
387 int tail_size = (size % blocksize) ?: blocksize;
388 struct buffer_head *bhs_inline[8];
389 struct buffer_head **bhs = bhs_inline;
390 int i, ret;
391
392 if (bh_count > ARRAY_SIZE(bhs_inline)) {
393 bhs = kmalloc_objs(*bhs, bh_count, GFP_NOFS);
394 if (!bhs)
395 return -ENOMEM;
396 }
397
398 ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
399 true /* wait */, bhs);
400 if (ret)
401 goto free_bhs;
402
403 for (i = 0; i < bh_count; i++) {
404 /* There shouldn't be any holes in ea_inode. */
405 if (!bhs[i]) {
406 ret = -EFSCORRUPTED;
407 goto put_bhs;
408 }
409 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
410 i < bh_count - 1 ? blocksize : tail_size);
411 }
412 ret = 0;
413 put_bhs:
414 for (i = 0; i < bh_count; i++)
415 brelse(bhs[i]);
416 free_bhs:
417 if (bhs != bhs_inline)
418 kfree(bhs);
419 return ret;
420 }
421
422 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode_get_mtime_sec(inode)))
423
424 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
425 u32 ea_inode_hash, struct inode **ea_inode)
426 {
427 struct inode *inode;
428 int err;
429
430 /*
431 * We have to check for this corruption early as otherwise
432 * iget_locked() could wait indefinitely for the state of our
433 * parent inode.
434 */
435 if (parent->i_ino == ea_ino) {
436 ext4_error(parent->i_sb,
437 "Parent and EA inode have the same ino %lu", ea_ino);
438 return -EFSCORRUPTED;
439 }
440
441 inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_EA_INODE);
442 if (IS_ERR(inode)) {
443 err = PTR_ERR(inode);
444 ext4_error(parent->i_sb,
445 "error while reading EA inode %lu err=%d", ea_ino,
446 err);
447 return err;
448 }
449 ext4_xattr_inode_set_class(inode);
450
451 /*
452 * Check whether this is an old Lustre-style xattr inode. Lustre
453 * implementation does not have hash validation, rather it has a
454 * backpointer from ea_inode to the parent inode.
455 */
456 if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
457 EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
458 inode->i_generation == parent->i_generation) {
459 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
460 ext4_xattr_inode_set_ref(inode, 1);
461 } else {
462 inode_lock_nested(inode, I_MUTEX_XATTR);
463 inode->i_flags |= S_NOQUOTA;
464 inode_unlock(inode);
465 }
466
467 *ea_inode = inode;
468 return 0;
469 }
470
471 /* Remove entry from mbcache when EA inode is getting evicted */
472 void ext4_evict_ea_inode(struct inode *inode)
473 {
474 struct mb_cache_entry *oe;
475
476 if (!EA_INODE_CACHE(inode))
477 return;
478 /* Wait for entry to get unused so that we can remove it */
479 while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
480 ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
481 mb_cache_entry_wait_unused(oe);
482 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
483 }
484 }
485
486 static int
487 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
488 struct ext4_xattr_entry *entry, void *buffer,
489 size_t size)
490 {
491 u32 hash;
492
493 /* Verify stored hash matches calculated hash. */
494 hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
495 if (hash != ext4_xattr_inode_get_hash(ea_inode))
496 return -EFSCORRUPTED;
497
498 if (entry) {
499 __le32 e_hash, tmp_data;
500
501 /* Verify entry hash. */
502 tmp_data = cpu_to_le32(hash);
503 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
504 &tmp_data, 1);
505 /* All good? */
506 if (e_hash == entry->e_hash)
507 return 0;
508
509 /*
510 * Not good. Maybe the entry hash was calculated
511 * using the buggy signed char version?
512 */
513 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
514 &tmp_data, 1);
515 /* Still no match - bad */
516 if (e_hash != entry->e_hash)
517 return -EFSCORRUPTED;
518
519 /* Let people know about old hash */
520 pr_warn_once("ext4: filesystem with signed xattr name hash");
521 }
522 return 0;
523 }
524
525 /*
526 * Read xattr value from the EA inode.
527 */
528 static int
529 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
530 void *buffer, size_t size)
531 {
532 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
533 struct inode *ea_inode;
534 int err;
535
536 err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
537 le32_to_cpu(entry->e_hash), &ea_inode);
538 if (err) {
539 ea_inode = NULL;
540 goto out;
541 }
542
543 if (i_size_read(ea_inode) != size) {
544 ext4_warning_inode(ea_inode,
545 "ea_inode file size=%llu entry size=%zu",
546 i_size_read(ea_inode), size);
547 err = -EFSCORRUPTED;
548 goto out;
549 }
550
551 err = ext4_xattr_inode_read(ea_inode, buffer, size);
552 if (err)
553 goto out;
554
555 if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
556 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
557 size);
558 if (err) {
559 ext4_warning_inode(ea_inode,
560 "EA inode hash validation failed");
561 goto out;
562 }
563
564 if (ea_inode_cache)
565 mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
566 ext4_xattr_inode_get_hash(ea_inode),
567 ea_inode->i_ino, true /* reusable */);
568 }
569 out:
570 iput(ea_inode);
571 return err;
572 }
573
574 static int
575 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
576 void *buffer, size_t buffer_size)
577 {
578 struct buffer_head *bh = NULL;
579 struct ext4_xattr_entry *entry;
580 size_t size;
581 void *end;
582 int error;
583 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
584
585 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
586 name_index, name, buffer, (long)buffer_size);
587
588 if (!EXT4_I(inode)->i_file_acl)
589 return -ENODATA;
590 ea_idebug(inode, "reading block %llu",
591 (unsigned long long)EXT4_I(inode)->i_file_acl);
592 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
593 if (IS_ERR(bh))
594 return PTR_ERR(bh);
595 ea_bdebug(bh, "b_count=%d, refcount=%d",
596 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
597 error = ext4_xattr_check_block(inode, bh);
598 if (error)
599 goto cleanup;
600 ext4_xattr_block_cache_insert(ea_block_cache, bh);
601 entry = BFIRST(bh);
602 end = bh->b_data + bh->b_size;
603 error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
604 if (error)
605 goto cleanup;
606 size = le32_to_cpu(entry->e_value_size);
607 error = -ERANGE;
608 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
609 goto cleanup;
610 if (buffer) {
611 if (size > buffer_size)
612 goto cleanup;
613 if (entry->e_value_inum) {
614 error = ext4_xattr_inode_get(inode, entry, buffer,
615 size);
616 if (error)
617 goto cleanup;
618 } else {
619 u16 offset = le16_to_cpu(entry->e_value_offs);
620 void *p = bh->b_data + offset;
621
622 if (unlikely(p + size > end))
623 goto cleanup;
624 memcpy(buffer, p, size);
625 }
626 }
627 error = size;
628
629 cleanup:
630 brelse(bh);
631 return error;
632 }
633
634 int
635 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
636 void *buffer, size_t buffer_size)
637 {
638 struct ext4_xattr_ibody_header *header;
639 struct ext4_xattr_entry *entry;
640 struct ext4_inode *raw_inode;
641 struct ext4_iloc iloc;
642 size_t size;
643 void *end;
644 int error;
645
646 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
647 return -ENODATA;
648 error = ext4_get_inode_loc(inode, &iloc);
649 if (error)
650 return error;
651 raw_inode = ext4_raw_inode(&iloc);
652 header = IHDR(inode, raw_inode);
653 end = ITAIL(inode, raw_inode);
654 entry = IFIRST(header);
655 error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
656 if (error)
657 goto cleanup;
658 size = le32_to_cpu(entry->e_value_size);
659 error = -ERANGE;
660 if (unlikely(size > EXT4_XATTR_SIZE_MAX))
661 goto cleanup;
662 if (buffer) {
663 if (size > buffer_size)
664 goto cleanup;
665 if (entry->e_value_inum) {
666 error = ext4_xattr_inode_get(inode, entry, buffer,
667 size);
668 if (error)
669 goto cleanup;
670 } else {
671 u16 offset = le16_to_cpu(entry->e_value_offs);
672 void *p = (void *)IFIRST(header) + offset;
673
674 if (unlikely(p + size > end))
675 goto cleanup;
676 memcpy(buffer, p, size);
677 }
678 }
679 error = size;
680
681 cleanup:
682 brelse(iloc.bh);
683 return error;
684 }
685
686 /*
687 * ext4_xattr_get()
688 *
689 * Copy an extended attribute into the buffer
690 * provided, or compute the buffer size required.
691 * Buffer is NULL to compute the size of the buffer required.
692 *
693 * Returns a negative error number on failure, or the number of bytes
694 * used / required on success.
695 */
696 int
697 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
698 void *buffer, size_t buffer_size)
699 {
700 int error;
701
702 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
703 return -EIO;
704
705 if (strlen(name) > 255)
706 return -ERANGE;
707
708 down_read(&EXT4_I(inode)->xattr_sem);
709 error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
710 buffer_size);
711 if (error == -ENODATA)
712 error = ext4_xattr_block_get(inode, name_index, name, buffer,
713 buffer_size);
714 up_read(&EXT4_I(inode)->xattr_sem);
715 return error;
716 }
717
718 static int
719 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
720 char *buffer, size_t buffer_size)
721 {
722 size_t rest = buffer_size;
723
724 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
725 const char *prefix;
726
727 prefix = ext4_xattr_prefix(entry->e_name_index, dentry);
728 if (prefix) {
729 size_t prefix_len = strlen(prefix);
730 size_t size = prefix_len + entry->e_name_len + 1;
731
732 if (buffer) {
733 if (size > rest)
734 return -ERANGE;
735 memcpy(buffer, prefix, prefix_len);
736 buffer += prefix_len;
737 memcpy(buffer, entry->e_name, entry->e_name_len);
738 buffer += entry->e_name_len;
739 *buffer++ = 0;
740 }
741 rest -= size;
742 }
743 }
744 return buffer_size - rest; /* total size */
745 }
746
747 static int
748 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
749 {
750 struct inode *inode = d_inode(dentry);
751 struct buffer_head *bh = NULL;
752 int error;
753
754 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
755 buffer, (long)buffer_size);
756
757 if (!EXT4_I(inode)->i_file_acl)
758 return 0;
759 ea_idebug(inode, "reading block %llu",
760 (unsigned long long)EXT4_I(inode)->i_file_acl);
761 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
762 if (IS_ERR(bh))
763 return PTR_ERR(bh);
764 ea_bdebug(bh, "b_count=%d, refcount=%d",
765 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
766 error = ext4_xattr_check_block(inode, bh);
767 if (error)
768 goto cleanup;
769 ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
770 error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
771 buffer_size);
772 cleanup:
773 brelse(bh);
774 return error;
775 }
776
777 static int
778 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
779 {
780 struct inode *inode = d_inode(dentry);
781 struct ext4_xattr_ibody_header *header;
782 struct ext4_inode *raw_inode;
783 struct ext4_iloc iloc;
784 int error;
785
786 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
787 return 0;
788 error = ext4_get_inode_loc(inode, &iloc);
789 if (error)
790 return error;
791 raw_inode = ext4_raw_inode(&iloc);
792 header = IHDR(inode, raw_inode);
793 error = ext4_xattr_list_entries(dentry, IFIRST(header),
794 buffer, buffer_size);
795
796 brelse(iloc.bh);
797 return error;
798 }
799
800 /*
801 * Inode operation listxattr()
802 *
803 * d_inode(dentry)->i_rwsem: don't care
804 *
805 * Copy a list of attribute names into the buffer
806 * provided, or compute the buffer size required.
807 * Buffer is NULL to compute the size of the buffer required.
808 *
809 * Returns a negative error number on failure, or the number of bytes
810 * used / required on success.
811 */
812 ssize_t
813 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
814 {
815 int ret, ret2;
816
817 down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
818 ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
819 if (ret < 0)
820 goto errout;
821 if (buffer) {
822 buffer += ret;
823 buffer_size -= ret;
824 }
825 ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
826 if (ret < 0)
827 goto errout;
828 ret += ret2;
829 errout:
830 up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
831 return ret;
832 }
833
834 /*
835 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
836 * not set, set it.
837 */
838 static void ext4_xattr_update_super_block(handle_t *handle,
839 struct super_block *sb)
840 {
841 if (ext4_has_feature_xattr(sb))
842 return;
843
844 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
845 if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
846 EXT4_JTR_NONE) == 0) {
847 lock_buffer(EXT4_SB(sb)->s_sbh);
848 ext4_set_feature_xattr(sb);
849 ext4_superblock_csum_set(sb);
850 unlock_buffer(EXT4_SB(sb)->s_sbh);
851 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
852 }
853 }
854
855 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
856 {
857 struct ext4_iloc iloc = { .bh = NULL };
858 struct buffer_head *bh = NULL;
859 struct ext4_inode *raw_inode;
860 struct ext4_xattr_ibody_header *header;
861 struct ext4_xattr_entry *entry;
862 qsize_t ea_inode_refs = 0;
863 int ret;
864
865 lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
866
867 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
868 ret = ext4_get_inode_loc(inode, &iloc);
869 if (ret)
870 goto out;
871 raw_inode = ext4_raw_inode(&iloc);
872 header = IHDR(inode, raw_inode);
873
874 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
875 entry = EXT4_XATTR_NEXT(entry))
876 if (entry->e_value_inum)
877 ea_inode_refs++;
878 }
879
880 if (EXT4_I(inode)->i_file_acl) {
881 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
882 if (IS_ERR(bh)) {
883 ret = PTR_ERR(bh);
884 bh = NULL;
885 goto out;
886 }
887
888 ret = ext4_xattr_check_block(inode, bh);
889 if (ret)
890 goto out;
891
892 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
893 entry = EXT4_XATTR_NEXT(entry))
894 if (entry->e_value_inum)
895 ea_inode_refs++;
896 }
897 *usage = ea_inode_refs + 1;
898 ret = 0;
899 out:
900 brelse(iloc.bh);
901 brelse(bh);
902 return ret;
903 }
904
905 static inline size_t round_up_cluster(struct inode *inode, size_t length)
906 {
907 struct super_block *sb = inode->i_sb;
908 size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
909 inode->i_blkbits);
910 size_t mask = ~(cluster_size - 1);
911
912 return (length + cluster_size - 1) & mask;
913 }
914
915 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
916 {
917 int err;
918
919 err = dquot_alloc_inode(inode);
920 if (err)
921 return err;
922 err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
923 if (err)
924 dquot_free_inode(inode);
925 return err;
926 }
927
928 static void ext4_xattr_inode_free_quota(struct inode *parent,
929 struct inode *ea_inode,
930 size_t len)
931 {
932 if (ea_inode &&
933 ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
934 return;
935 dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
936 dquot_free_inode(parent);
937 }
938
939 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
940 struct buffer_head *block_bh, size_t value_len,
941 bool is_create)
942 {
943 int credits;
944 int blocks;
945
946 /*
947 * 1) Owner inode update
948 * 2) Ref count update on old xattr block
949 * 3) new xattr block
950 * 4) block bitmap update for new xattr block
951 * 5) group descriptor for new xattr block
952 * 6) block bitmap update for old xattr block
953 * 7) group descriptor for old block
954 *
955 * 6 & 7 can happen if we have two racing threads T_a and T_b
956 * which are each trying to set an xattr on inodes I_a and I_b
957 * which were both initially sharing an xattr block.
958 */
959 credits = 7;
960
961 /* Quota updates. */
962 credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
963
964 /*
965 * In case of inline data, we may push out the data to a block,
966 * so we need to reserve credits for this eventuality
967 */
968 if (inode && ext4_has_inline_data(inode))
969 credits += ext4_chunk_trans_extent(inode, 1) + 1;
970
971 /* We are done if ea_inode feature is not enabled. */
972 if (!ext4_has_feature_ea_inode(sb))
973 return credits;
974
975 /* New ea_inode, inode map, block bitmap, group descriptor. */
976 credits += 4;
977
978 /* Data blocks. */
979 blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
980
981 /* Indirection block or one level of extent tree. */
982 blocks += 1;
983
984 /* Block bitmap and group descriptor updates for each block. */
985 credits += blocks * 2;
986
987 /* Blocks themselves. */
988 credits += blocks;
989
990 if (!is_create) {
991 /* Dereference ea_inode holding old xattr value.
992 * Old ea_inode, inode map, block bitmap, group descriptor.
993 */
994 credits += 4;
995
996 /* Data blocks for old ea_inode. */
997 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
998
999 /* Indirection block or one level of extent tree for old
1000 * ea_inode.
1001 */
1002 blocks += 1;
1003
1004 /* Block bitmap and group descriptor updates for each block. */
1005 credits += blocks * 2;
1006 }
1007
1008 /* We may need to clone the existing xattr block in which case we need
1009 * to increment ref counts for existing ea_inodes referenced by it.
1010 */
1011 if (block_bh) {
1012 struct ext4_xattr_entry *entry = BFIRST(block_bh);
1013
1014 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1015 if (entry->e_value_inum)
1016 /* Ref count update on ea_inode. */
1017 credits += 1;
1018 }
1019 return credits;
1020 }
1021
1022 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1023 int ref_change)
1024 {
1025 struct ext4_iloc iloc;
1026 u64 ref_count;
1027 int ret;
1028
1029 inode_lock_nested(ea_inode, I_MUTEX_XATTR);
1030
1031 ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1032 if (ret)
1033 goto out;
1034
1035 ref_count = ext4_xattr_inode_get_ref(ea_inode);
1036 if ((ref_count == 0 && ref_change < 0) || (ref_count == U64_MAX && ref_change > 0)) {
1037 ext4_error_inode(ea_inode, __func__, __LINE__, 0,
1038 "EA inode %lu ref wraparound: ref_count=%lld ref_change=%d",
1039 ea_inode->i_ino, ref_count, ref_change);
1040 brelse(iloc.bh);
1041 ret = -EFSCORRUPTED;
1042 goto out;
1043 }
1044 ref_count += ref_change;
1045 ext4_xattr_inode_set_ref(ea_inode, ref_count);
1046
1047 if (ref_change > 0) {
1048 if (ref_count == 1) {
1049 WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1050 ea_inode->i_ino, ea_inode->i_nlink);
1051
1052 set_nlink(ea_inode, 1);
1053 ext4_orphan_del(handle, ea_inode);
1054 }
1055 } else {
1056 if (ref_count == 0) {
1057 WARN_ONCE(ea_inode->i_nlink != 1,
1058 "EA inode %lu i_nlink=%u",
1059 ea_inode->i_ino, ea_inode->i_nlink);
1060
1061 clear_nlink(ea_inode);
1062 ext4_orphan_add(handle, ea_inode);
1063 }
1064 }
1065
1066 ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1067 if (ret)
1068 ext4_warning_inode(ea_inode,
1069 "ext4_mark_iloc_dirty() failed ret=%d", ret);
1070 out:
1071 inode_unlock(ea_inode);
1072 return ret;
1073 }
1074
1075 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1076 {
1077 return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1078 }
1079
1080 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1081 {
1082 return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1083 }
1084
1085 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1086 struct ext4_xattr_entry *first)
1087 {
1088 struct inode *ea_inode;
1089 struct ext4_xattr_entry *entry;
1090 struct ext4_xattr_entry *failed_entry;
1091 unsigned int ea_ino;
1092 int err, saved_err;
1093
1094 for (entry = first; !IS_LAST_ENTRY(entry);
1095 entry = EXT4_XATTR_NEXT(entry)) {
1096 if (!entry->e_value_inum)
1097 continue;
1098 ea_ino = le32_to_cpu(entry->e_value_inum);
1099 err = ext4_xattr_inode_iget(parent, ea_ino,
1100 le32_to_cpu(entry->e_hash),
1101 &ea_inode);
1102 if (err)
1103 goto cleanup;
1104 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1105 if (err) {
1106 ext4_warning_inode(ea_inode, "inc ref error %d", err);
1107 iput(ea_inode);
1108 goto cleanup;
1109 }
1110 iput(ea_inode);
1111 }
1112 return 0;
1113
1114 cleanup:
1115 saved_err = err;
1116 failed_entry = entry;
1117
1118 for (entry = first; entry != failed_entry;
1119 entry = EXT4_XATTR_NEXT(entry)) {
1120 if (!entry->e_value_inum)
1121 continue;
1122 ea_ino = le32_to_cpu(entry->e_value_inum);
1123 err = ext4_xattr_inode_iget(parent, ea_ino,
1124 le32_to_cpu(entry->e_hash),
1125 &ea_inode);
1126 if (err) {
1127 ext4_warning(parent->i_sb,
1128 "cleanup ea_ino %u iget error %d", ea_ino,
1129 err);
1130 continue;
1131 }
1132 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1133 if (err)
1134 ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1135 err);
1136 iput(ea_inode);
1137 }
1138 return saved_err;
1139 }
1140
1141 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1142 struct buffer_head *bh, bool block_csum, bool dirty)
1143 {
1144 int error;
1145
1146 if (bh && dirty) {
1147 if (block_csum)
1148 ext4_xattr_block_csum_set(inode, bh);
1149 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1150 if (error) {
1151 ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1152 error);
1153 return error;
1154 }
1155 }
1156 return 0;
1157 }
1158
1159 static void
1160 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1161 struct buffer_head *bh,
1162 struct ext4_xattr_entry *first, bool block_csum,
1163 struct ext4_xattr_inode_array **ea_inode_array,
1164 int extra_credits, bool skip_quota)
1165 {
1166 struct inode *ea_inode;
1167 struct ext4_xattr_entry *entry;
1168 struct ext4_iloc iloc = { .bh = NULL };
1169 bool dirty = false;
1170 unsigned int ea_ino;
1171 int err;
1172 int credits;
1173 void *end;
1174
1175 if (block_csum)
1176 end = (void *)bh->b_data + bh->b_size;
1177 else {
1178 err = ext4_get_inode_loc(parent, &iloc);
1179 if (err) {
1180 EXT4_ERROR_INODE(parent, "parent inode loc (error %d)", err);
1181 return;
1182 }
1183 end = (void *)ext4_raw_inode(&iloc) + EXT4_SB(parent->i_sb)->s_inode_size;
1184 }
1185
1186 /* One credit for dec ref on ea_inode, one for orphan list addition, */
1187 credits = 2 + extra_credits;
1188
1189 for (entry = first; (void *)entry < end && !IS_LAST_ENTRY(entry);
1190 entry = EXT4_XATTR_NEXT(entry)) {
1191 if (!entry->e_value_inum)
1192 continue;
1193 ea_ino = le32_to_cpu(entry->e_value_inum);
1194 err = ext4_xattr_inode_iget(parent, ea_ino,
1195 le32_to_cpu(entry->e_hash),
1196 &ea_inode);
1197 if (err)
1198 continue;
1199
1200 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1201 if (err) {
1202 ext4_warning_inode(ea_inode,
1203 "Expand inode array err=%d", err);
1204 iput(ea_inode);
1205 continue;
1206 }
1207
1208 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1209 ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1210 ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1211 dirty));
1212 if (err < 0) {
1213 ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1214 err);
1215 continue;
1216 }
1217 if (err > 0) {
1218 err = ext4_journal_get_write_access(handle,
1219 parent->i_sb, bh, EXT4_JTR_NONE);
1220 if (err) {
1221 ext4_warning_inode(ea_inode,
1222 "Re-get write access err=%d",
1223 err);
1224 continue;
1225 }
1226 }
1227
1228 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1229 if (err) {
1230 ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1231 err);
1232 continue;
1233 }
1234
1235 if (!skip_quota)
1236 ext4_xattr_inode_free_quota(parent, ea_inode,
1237 le32_to_cpu(entry->e_value_size));
1238
1239 /*
1240 * Forget about ea_inode within the same transaction that
1241 * decrements the ref count. This avoids duplicate decrements in
1242 * case the rest of the work spills over to subsequent
1243 * transactions.
1244 */
1245 entry->e_value_inum = 0;
1246 entry->e_value_size = 0;
1247
1248 dirty = true;
1249 }
1250
1251 if (dirty) {
1252 /*
1253 * Note that we are deliberately skipping csum calculation for
1254 * the final update because we do not expect any journal
1255 * restarts until xattr block is freed.
1256 */
1257
1258 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1259 if (err)
1260 ext4_warning_inode(parent,
1261 "handle dirty metadata err=%d", err);
1262 }
1263
1264 brelse(iloc.bh);
1265 }
1266
1267 /*
1268 * Release the xattr block BH: If the reference count is > 1, decrement it;
1269 * otherwise free the block.
1270 */
1271 static void
1272 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1273 struct buffer_head *bh,
1274 struct ext4_xattr_inode_array **ea_inode_array,
1275 int extra_credits)
1276 {
1277 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1278 u32 hash, ref;
1279 int error = 0;
1280
1281 BUFFER_TRACE(bh, "get_write_access");
1282 error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1283 EXT4_JTR_NONE);
1284 if (error)
1285 goto out;
1286
1287 retry_ref:
1288 lock_buffer(bh);
1289 hash = le32_to_cpu(BHDR(bh)->h_hash);
1290 ref = le32_to_cpu(BHDR(bh)->h_refcount);
1291 if (ref == 1) {
1292 ea_bdebug(bh, "refcount now=0; freeing");
1293 /*
1294 * This must happen under buffer lock for
1295 * ext4_xattr_block_set() to reliably detect freed block
1296 */
1297 if (ea_block_cache) {
1298 struct mb_cache_entry *oe;
1299
1300 oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1301 bh->b_blocknr);
1302 if (oe) {
1303 unlock_buffer(bh);
1304 mb_cache_entry_wait_unused(oe);
1305 mb_cache_entry_put(ea_block_cache, oe);
1306 goto retry_ref;
1307 }
1308 }
1309 get_bh(bh);
1310 unlock_buffer(bh);
1311
1312 if (ext4_has_feature_ea_inode(inode->i_sb))
1313 ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1314 BFIRST(bh),
1315 true /* block_csum */,
1316 ea_inode_array,
1317 extra_credits,
1318 true /* skip_quota */);
1319 ext4_free_blocks(handle, inode, bh, 0, 1,
1320 EXT4_FREE_BLOCKS_METADATA |
1321 EXT4_FREE_BLOCKS_FORGET);
1322 } else {
1323 ref--;
1324 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1325 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1326 struct mb_cache_entry *ce;
1327
1328 if (ea_block_cache) {
1329 ce = mb_cache_entry_get(ea_block_cache, hash,
1330 bh->b_blocknr);
1331 if (ce) {
1332 set_bit(MBE_REUSABLE_B, &ce->e_flags);
1333 mb_cache_entry_put(ea_block_cache, ce);
1334 }
1335 }
1336 }
1337
1338 ext4_xattr_block_csum_set(inode, bh);
1339 /*
1340 * Beware of this ugliness: Releasing of xattr block references
1341 * from different inodes can race and so we have to protect
1342 * from a race where someone else frees the block (and releases
1343 * its journal_head) before we are done dirtying the buffer. In
1344 * nojournal mode this race is harmless and we actually cannot
1345 * call ext4_handle_dirty_metadata() with locked buffer as
1346 * that function can call sync_dirty_buffer() so for that case
1347 * we handle the dirtying after unlocking the buffer.
1348 */
1349 if (ext4_handle_valid(handle))
1350 error = ext4_handle_dirty_metadata(handle, inode, bh);
1351 unlock_buffer(bh);
1352 if (!ext4_handle_valid(handle))
1353 error = ext4_handle_dirty_metadata(handle, inode, bh);
1354 if (IS_SYNC(inode))
1355 ext4_handle_sync(handle);
1356 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1357 ea_bdebug(bh, "refcount now=%d; releasing",
1358 le32_to_cpu(BHDR(bh)->h_refcount));
1359 }
1360 out:
1361 ext4_std_error(inode->i_sb, error);
1362 return;
1363 }
1364
1365 /*
1366 * Find the available free space for EAs. This also returns the total number of
1367 * bytes used by EA entries.
1368 */
1369 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1370 size_t *min_offs, void *base, int *total)
1371 {
1372 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1373 if (!last->e_value_inum && last->e_value_size) {
1374 size_t offs = le16_to_cpu(last->e_value_offs);
1375 if (offs < *min_offs)
1376 *min_offs = offs;
1377 }
1378 if (total)
1379 *total += EXT4_XATTR_LEN(last->e_name_len);
1380 }
1381 return (*min_offs - ((void *)last - base) - sizeof(__u32));
1382 }
1383
1384 /*
1385 * Write the value of the EA in an inode.
1386 */
1387 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1388 const void *buf, int bufsize)
1389 {
1390 struct buffer_head *bh = NULL;
1391 unsigned long block = 0;
1392 int blocksize = ea_inode->i_sb->s_blocksize;
1393 int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1394 int csize, wsize = 0;
1395 int ret = 0, ret2 = 0;
1396 int retries = 0;
1397
1398 retry:
1399 while (ret >= 0 && ret < max_blocks) {
1400 struct ext4_map_blocks map;
1401 map.m_lblk = block += ret;
1402 map.m_len = max_blocks -= ret;
1403
1404 ret = ext4_map_blocks(handle, ea_inode, &map,
1405 EXT4_GET_BLOCKS_CREATE);
1406 if (ret <= 0) {
1407 ext4_mark_inode_dirty(handle, ea_inode);
1408 if (ret == -ENOSPC &&
1409 ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1410 ret = 0;
1411 goto retry;
1412 }
1413 break;
1414 }
1415 }
1416
1417 if (ret < 0)
1418 return ret;
1419
1420 block = 0;
1421 while (wsize < bufsize) {
1422 brelse(bh);
1423 csize = (bufsize - wsize) > blocksize ? blocksize :
1424 bufsize - wsize;
1425 bh = ext4_getblk(handle, ea_inode, block, 0);
1426 if (IS_ERR(bh))
1427 return PTR_ERR(bh);
1428 if (!bh) {
1429 WARN_ON_ONCE(1);
1430 EXT4_ERROR_INODE(ea_inode,
1431 "ext4_getblk() return bh = NULL");
1432 return -EFSCORRUPTED;
1433 }
1434 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1435 EXT4_JTR_NONE);
1436 if (ret)
1437 goto out;
1438
1439 memcpy(bh->b_data, buf, csize);
1440 /*
1441 * Zero out block tail to avoid writing uninitialized memory
1442 * to disk.
1443 */
1444 if (csize < blocksize)
1445 memset(bh->b_data + csize, 0, blocksize - csize);
1446 set_buffer_uptodate(bh);
1447 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1448
1449 buf += csize;
1450 wsize += csize;
1451 block += 1;
1452 }
1453
1454 inode_lock(ea_inode);
1455 i_size_write(ea_inode, wsize);
1456 ext4_update_i_disksize(ea_inode, wsize);
1457 inode_unlock(ea_inode);
1458
1459 ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1460 if (unlikely(ret2 && !ret))
1461 ret = ret2;
1462
1463 out:
1464 brelse(bh);
1465
1466 return ret;
1467 }
1468
1469 /*
1470 * Create an inode to store the value of a large EA.
1471 */
1472 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1473 struct inode *inode, u32 hash)
1474 {
1475 struct inode *ea_inode = NULL;
1476 uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1477 int err;
1478
1479 if (inode->i_sb->s_root == NULL) {
1480 ext4_warning(inode->i_sb,
1481 "refuse to create EA inode when umounting");
1482 WARN_ON(1);
1483 return ERR_PTR(-EINVAL);
1484 }
1485
1486 /*
1487 * Let the next inode be the goal, so we try and allocate the EA inode
1488 * in the same group, or nearby one.
1489 */
1490 ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1491 S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1492 EXT4_EA_INODE_FL);
1493 if (!IS_ERR(ea_inode)) {
1494 ea_inode->i_op = &ext4_file_inode_operations;
1495 ea_inode->i_fop = &ext4_file_operations;
1496 ext4_set_aops(ea_inode);
1497 ext4_xattr_inode_set_class(ea_inode);
1498 unlock_new_inode(ea_inode);
1499 ext4_xattr_inode_set_ref(ea_inode, 1);
1500 ext4_xattr_inode_set_hash(ea_inode, hash);
1501 err = ext4_mark_inode_dirty(handle, ea_inode);
1502 if (!err)
1503 err = ext4_inode_attach_jinode(ea_inode);
1504 if (err) {
1505 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1506 ext4_warning_inode(ea_inode,
1507 "cleanup dec ref error %d", err);
1508 iput(ea_inode);
1509 return ERR_PTR(err);
1510 }
1511
1512 /*
1513 * Xattr inodes are shared therefore quota charging is performed
1514 * at a higher level.
1515 */
1516 dquot_free_inode(ea_inode);
1517 dquot_drop(ea_inode);
1518 inode_lock(ea_inode);
1519 ea_inode->i_flags |= S_NOQUOTA;
1520 inode_unlock(ea_inode);
1521 }
1522
1523 return ea_inode;
1524 }
1525
1526 static struct inode *
1527 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1528 size_t value_len, u32 hash)
1529 {
1530 struct inode *ea_inode;
1531 struct mb_cache_entry *ce;
1532 struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1533 void *ea_data;
1534
1535 if (!ea_inode_cache)
1536 return NULL;
1537
1538 ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1539 if (!ce)
1540 return NULL;
1541
1542 WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1543 !(current->flags & PF_MEMALLOC_NOFS));
1544
1545 ea_data = kvmalloc(value_len, GFP_NOFS);
1546 if (!ea_data) {
1547 mb_cache_entry_put(ea_inode_cache, ce);
1548 return NULL;
1549 }
1550
1551 while (ce) {
1552 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1553 EXT4_IGET_EA_INODE);
1554 if (IS_ERR(ea_inode))
1555 goto next_entry;
1556 ext4_xattr_inode_set_class(ea_inode);
1557 if (i_size_read(ea_inode) == value_len &&
1558 !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1559 !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1560 value_len) &&
1561 !memcmp(value, ea_data, value_len)) {
1562 mb_cache_entry_touch(ea_inode_cache, ce);
1563 mb_cache_entry_put(ea_inode_cache, ce);
1564 kvfree(ea_data);
1565 return ea_inode;
1566 }
1567 iput(ea_inode);
1568 next_entry:
1569 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1570 }
1571 kvfree(ea_data);
1572 return NULL;
1573 }
1574
1575 /*
1576 * Add value of the EA in an inode.
1577 */
1578 static struct inode *ext4_xattr_inode_lookup_create(handle_t *handle,
1579 struct inode *inode, const void *value, size_t value_len)
1580 {
1581 struct inode *ea_inode;
1582 u32 hash;
1583 int err;
1584
1585 /* Account inode & space to quota even if sharing... */
1586 err = ext4_xattr_inode_alloc_quota(inode, value_len);
1587 if (err)
1588 return ERR_PTR(err);
1589
1590 hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1591 ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1592 if (ea_inode) {
1593 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1594 if (err)
1595 goto out_err;
1596 return ea_inode;
1597 }
1598
1599 /* Create an inode for the EA value */
1600 ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1601 if (IS_ERR(ea_inode)) {
1602 ext4_xattr_inode_free_quota(inode, NULL, value_len);
1603 return ea_inode;
1604 }
1605
1606 err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1607 if (err) {
1608 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1609 ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1610 goto out_err;
1611 }
1612
1613 if (EA_INODE_CACHE(inode))
1614 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1615 ea_inode->i_ino, true /* reusable */);
1616 return ea_inode;
1617 out_err:
1618 iput(ea_inode);
1619 ext4_xattr_inode_free_quota(inode, NULL, value_len);
1620 return ERR_PTR(err);
1621 }
1622
1623 /*
1624 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1625 * feature is enabled.
1626 */
1627 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1628
1629 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1630 struct ext4_xattr_search *s,
1631 handle_t *handle, struct inode *inode,
1632 struct inode *new_ea_inode,
1633 bool is_block)
1634 {
1635 struct ext4_xattr_entry *last, *next;
1636 struct ext4_xattr_entry *here = s->here;
1637 size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1638 int in_inode = i->in_inode;
1639 struct inode *old_ea_inode = NULL;
1640 size_t old_size, new_size;
1641 int ret;
1642
1643 /* Space used by old and new values. */
1644 old_size = (!s->not_found && !here->e_value_inum) ?
1645 EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1646 new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1647
1648 /*
1649 * Optimization for the simple case when old and new values have the
1650 * same padded sizes. Not applicable if external inodes are involved.
1651 */
1652 if (new_size && new_size == old_size) {
1653 size_t offs = le16_to_cpu(here->e_value_offs);
1654 void *val = s->base + offs;
1655
1656 here->e_value_size = cpu_to_le32(i->value_len);
1657 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1658 memset(val, 0, new_size);
1659 } else {
1660 memcpy(val, i->value, i->value_len);
1661 /* Clear padding bytes. */
1662 memset(val + i->value_len, 0, new_size - i->value_len);
1663 }
1664 goto update_hash;
1665 }
1666
1667 /* Compute min_offs and last. */
1668 last = s->first;
1669 for (; !IS_LAST_ENTRY(last); last = next) {
1670 next = EXT4_XATTR_NEXT(last);
1671 if ((void *)next >= s->end) {
1672 EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1673 ret = -EFSCORRUPTED;
1674 goto out;
1675 }
1676 if (!last->e_value_inum && last->e_value_size) {
1677 size_t offs = le16_to_cpu(last->e_value_offs);
1678 if (offs < min_offs)
1679 min_offs = offs;
1680 }
1681 }
1682
1683 /* Check whether we have enough space. */
1684 if (i->value) {
1685 size_t free;
1686
1687 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1688 if (!s->not_found)
1689 free += EXT4_XATTR_LEN(name_len) + old_size;
1690
1691 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1692 ret = -ENOSPC;
1693 goto out;
1694 }
1695
1696 /*
1697 * If storing the value in an external inode is an option,
1698 * reserve space for xattr entries/names in the external
1699 * attribute block so that a long value does not occupy the
1700 * whole space and prevent further entries being added.
1701 */
1702 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1703 new_size && is_block &&
1704 (min_offs + old_size - new_size) <
1705 EXT4_XATTR_BLOCK_RESERVE(inode)) {
1706 ret = -ENOSPC;
1707 goto out;
1708 }
1709 }
1710
1711 /*
1712 * Getting access to old and new ea inodes is subject to failures.
1713 * Finish that work before doing any modifications to the xattr data.
1714 */
1715 if (!s->not_found && here->e_value_inum) {
1716 ret = ext4_xattr_inode_iget(inode,
1717 le32_to_cpu(here->e_value_inum),
1718 le32_to_cpu(here->e_hash),
1719 &old_ea_inode);
1720 if (ret) {
1721 old_ea_inode = NULL;
1722 goto out;
1723 }
1724
1725 /* We are ready to release ref count on the old_ea_inode. */
1726 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1727 if (ret)
1728 goto out;
1729
1730 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1731 le32_to_cpu(here->e_value_size));
1732 }
1733
1734 /* No failures allowed past this point. */
1735
1736 if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1737 /* Remove the old value. */
1738 void *first_val = s->base + min_offs;
1739 size_t offs = le16_to_cpu(here->e_value_offs);
1740 void *val = s->base + offs;
1741
1742 memmove(first_val + old_size, first_val, val - first_val);
1743 memset(first_val, 0, old_size);
1744 min_offs += old_size;
1745
1746 /* Adjust all value offsets. */
1747 last = s->first;
1748 while (!IS_LAST_ENTRY(last)) {
1749 size_t o = le16_to_cpu(last->e_value_offs);
1750
1751 if (!last->e_value_inum &&
1752 last->e_value_size && o < offs)
1753 last->e_value_offs = cpu_to_le16(o + old_size);
1754 last = EXT4_XATTR_NEXT(last);
1755 }
1756 }
1757
1758 if (!i->value) {
1759 /* Remove old name. */
1760 size_t size = EXT4_XATTR_LEN(name_len);
1761
1762 last = ENTRY((void *)last - size);
1763 memmove(here, (void *)here + size,
1764 (void *)last - (void *)here + sizeof(__u32));
1765 memset(last, 0, size);
1766
1767 /*
1768 * Update i_inline_off - moved ibody region might contain
1769 * system.data attribute. Handling a failure here won't
1770 * cause other complications for setting an xattr.
1771 */
1772 if (!is_block && ext4_has_inline_data(inode)) {
1773 ret = ext4_find_inline_data_nolock(inode);
1774 if (ret) {
1775 ext4_warning_inode(inode,
1776 "unable to update i_inline_off");
1777 goto out;
1778 }
1779 }
1780 } else if (s->not_found) {
1781 /* Insert new name. */
1782 size_t size = EXT4_XATTR_LEN(name_len);
1783 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1784
1785 memmove((void *)here + size, here, rest);
1786 memset(here, 0, size);
1787 here->e_name_index = i->name_index;
1788 here->e_name_len = name_len;
1789 memcpy(here->e_name, i->name, name_len);
1790 } else {
1791 /* This is an update, reset value info. */
1792 here->e_value_inum = 0;
1793 here->e_value_offs = 0;
1794 here->e_value_size = 0;
1795 }
1796
1797 if (i->value) {
1798 /* Insert new value. */
1799 if (in_inode) {
1800 here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1801 } else if (i->value_len) {
1802 void *val = s->base + min_offs - new_size;
1803
1804 here->e_value_offs = cpu_to_le16(min_offs - new_size);
1805 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1806 memset(val, 0, new_size);
1807 } else {
1808 memcpy(val, i->value, i->value_len);
1809 /* Clear padding bytes. */
1810 memset(val + i->value_len, 0,
1811 new_size - i->value_len);
1812 }
1813 }
1814 here->e_value_size = cpu_to_le32(i->value_len);
1815 }
1816
1817 update_hash:
1818 if (i->value) {
1819 __le32 hash = 0;
1820
1821 /* Entry hash calculation. */
1822 if (in_inode) {
1823 __le32 crc32c_hash;
1824
1825 /*
1826 * Feed crc32c hash instead of the raw value for entry
1827 * hash calculation. This is to avoid walking
1828 * potentially long value buffer again.
1829 */
1830 crc32c_hash = cpu_to_le32(
1831 ext4_xattr_inode_get_hash(new_ea_inode));
1832 hash = ext4_xattr_hash_entry(here->e_name,
1833 here->e_name_len,
1834 &crc32c_hash, 1);
1835 } else if (is_block) {
1836 __le32 *value = s->base + le16_to_cpu(
1837 here->e_value_offs);
1838
1839 hash = ext4_xattr_hash_entry(here->e_name,
1840 here->e_name_len, value,
1841 new_size >> 2);
1842 }
1843 here->e_hash = hash;
1844 }
1845
1846 if (is_block)
1847 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1848
1849 ret = 0;
1850 out:
1851 iput(old_ea_inode);
1852 return ret;
1853 }
1854
1855 struct ext4_xattr_block_find {
1856 struct ext4_xattr_search s;
1857 struct buffer_head *bh;
1858 };
1859
1860 static int
1861 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1862 struct ext4_xattr_block_find *bs)
1863 {
1864 struct super_block *sb = inode->i_sb;
1865 int error;
1866
1867 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1868 i->name_index, i->name, i->value, (long)i->value_len);
1869
1870 if (EXT4_I(inode)->i_file_acl) {
1871 /* The inode already has an extended attribute block. */
1872 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1873 if (IS_ERR(bs->bh)) {
1874 error = PTR_ERR(bs->bh);
1875 bs->bh = NULL;
1876 return error;
1877 }
1878 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1879 atomic_read(&(bs->bh->b_count)),
1880 le32_to_cpu(BHDR(bs->bh)->h_refcount));
1881 error = ext4_xattr_check_block(inode, bs->bh);
1882 if (error)
1883 return error;
1884 /* Find the named attribute. */
1885 bs->s.base = BHDR(bs->bh);
1886 bs->s.first = BFIRST(bs->bh);
1887 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1888 bs->s.here = bs->s.first;
1889 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1890 i->name_index, i->name, 1);
1891 if (error && error != -ENODATA)
1892 return error;
1893 bs->s.not_found = error;
1894 }
1895 return 0;
1896 }
1897
1898 static int
1899 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1900 struct ext4_xattr_info *i,
1901 struct ext4_xattr_block_find *bs)
1902 {
1903 struct super_block *sb = inode->i_sb;
1904 struct buffer_head *new_bh = NULL;
1905 struct ext4_xattr_search s_copy = bs->s;
1906 struct ext4_xattr_search *s = &s_copy;
1907 struct mb_cache_entry *ce = NULL;
1908 int error = 0;
1909 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1910 struct inode *ea_inode = NULL, *tmp_inode;
1911 size_t old_ea_inode_quota = 0;
1912 unsigned int ea_ino;
1913
1914 #define header(x) ((struct ext4_xattr_header *)(x))
1915
1916 /* If we need EA inode, prepare it before locking the buffer */
1917 if (i->value && i->in_inode) {
1918 WARN_ON_ONCE(!i->value_len);
1919
1920 ea_inode = ext4_xattr_inode_lookup_create(handle, inode,
1921 i->value, i->value_len);
1922 if (IS_ERR(ea_inode)) {
1923 error = PTR_ERR(ea_inode);
1924 ea_inode = NULL;
1925 goto cleanup;
1926 }
1927 }
1928
1929 if (s->base) {
1930 int offset = (char *)s->here - bs->bh->b_data;
1931
1932 BUFFER_TRACE(bs->bh, "get_write_access");
1933 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1934 EXT4_JTR_NONE);
1935 if (error)
1936 goto cleanup;
1937
1938 lock_buffer(bs->bh);
1939
1940 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1941 __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1942
1943 /*
1944 * This must happen under buffer lock for
1945 * ext4_xattr_block_set() to reliably detect modified
1946 * block
1947 */
1948 if (ea_block_cache) {
1949 struct mb_cache_entry *oe;
1950
1951 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1952 hash, bs->bh->b_blocknr);
1953 if (oe) {
1954 /*
1955 * Xattr block is getting reused. Leave
1956 * it alone.
1957 */
1958 mb_cache_entry_put(ea_block_cache, oe);
1959 goto clone_block;
1960 }
1961 }
1962 ea_bdebug(bs->bh, "modifying in-place");
1963 error = ext4_xattr_set_entry(i, s, handle, inode,
1964 ea_inode, true /* is_block */);
1965 ext4_xattr_block_csum_set(inode, bs->bh);
1966 unlock_buffer(bs->bh);
1967 if (error == -EFSCORRUPTED)
1968 goto bad_block;
1969 if (!error)
1970 error = ext4_handle_dirty_metadata(handle,
1971 inode,
1972 bs->bh);
1973 if (error)
1974 goto cleanup;
1975 goto inserted;
1976 }
1977 clone_block:
1978 unlock_buffer(bs->bh);
1979 ea_bdebug(bs->bh, "cloning");
1980 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1981 error = -ENOMEM;
1982 if (s->base == NULL)
1983 goto cleanup;
1984 s->first = ENTRY(header(s->base)+1);
1985 header(s->base)->h_refcount = cpu_to_le32(1);
1986 s->here = ENTRY(s->base + offset);
1987 s->end = s->base + bs->bh->b_size;
1988
1989 /*
1990 * If existing entry points to an xattr inode, we need
1991 * to prevent ext4_xattr_set_entry() from decrementing
1992 * ref count on it because the reference belongs to the
1993 * original block. In this case, make the entry look
1994 * like it has an empty value.
1995 */
1996 if (!s->not_found && s->here->e_value_inum) {
1997 ea_ino = le32_to_cpu(s->here->e_value_inum);
1998 error = ext4_xattr_inode_iget(inode, ea_ino,
1999 le32_to_cpu(s->here->e_hash),
2000 &tmp_inode);
2001 if (error)
2002 goto cleanup;
2003
2004 if (!ext4_test_inode_state(tmp_inode,
2005 EXT4_STATE_LUSTRE_EA_INODE)) {
2006 /*
2007 * Defer quota free call for previous
2008 * inode until success is guaranteed.
2009 */
2010 old_ea_inode_quota = le32_to_cpu(
2011 s->here->e_value_size);
2012 }
2013 iput(tmp_inode);
2014
2015 s->here->e_value_inum = 0;
2016 s->here->e_value_size = 0;
2017 }
2018 } else {
2019 /* Allocate a buffer where we construct the new block. */
2020 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
2021 error = -ENOMEM;
2022 if (s->base == NULL)
2023 goto cleanup;
2024 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2025 header(s->base)->h_blocks = cpu_to_le32(1);
2026 header(s->base)->h_refcount = cpu_to_le32(1);
2027 s->first = ENTRY(header(s->base)+1);
2028 s->here = ENTRY(header(s->base)+1);
2029 s->end = s->base + sb->s_blocksize;
2030 }
2031
2032 error = ext4_xattr_set_entry(i, s, handle, inode, ea_inode,
2033 true /* is_block */);
2034 if (error == -EFSCORRUPTED)
2035 goto bad_block;
2036 if (error)
2037 goto cleanup;
2038
2039 inserted:
2040 if (!IS_LAST_ENTRY(s->first)) {
2041 new_bh = ext4_xattr_block_cache_find(inode, header(s->base), &ce);
2042 if (IS_ERR(new_bh)) {
2043 error = PTR_ERR(new_bh);
2044 new_bh = NULL;
2045 goto cleanup;
2046 }
2047
2048 if (new_bh) {
2049 /* We found an identical block in the cache. */
2050 if (new_bh == bs->bh)
2051 ea_bdebug(new_bh, "keeping");
2052 else {
2053 u32 ref;
2054
2055 #ifdef EXT4_XATTR_DEBUG
2056 WARN_ON_ONCE(dquot_initialize_needed(inode));
2057 #endif
2058 /* The old block is released after updating
2059 the inode. */
2060 error = dquot_alloc_block(inode,
2061 EXT4_C2B(EXT4_SB(sb), 1));
2062 if (error)
2063 goto cleanup;
2064 BUFFER_TRACE(new_bh, "get_write_access");
2065 error = ext4_journal_get_write_access(
2066 handle, sb, new_bh,
2067 EXT4_JTR_NONE);
2068 if (error)
2069 goto cleanup_dquot;
2070 lock_buffer(new_bh);
2071 /*
2072 * We have to be careful about races with
2073 * adding references to xattr block. Once we
2074 * hold buffer lock xattr block's state is
2075 * stable so we can check the additional
2076 * reference fits.
2077 */
2078 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2079 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2080 /*
2081 * Undo everything and check mbcache
2082 * again.
2083 */
2084 unlock_buffer(new_bh);
2085 dquot_free_block(inode,
2086 EXT4_C2B(EXT4_SB(sb),
2087 1));
2088 brelse(new_bh);
2089 mb_cache_entry_put(ea_block_cache, ce);
2090 ce = NULL;
2091 new_bh = NULL;
2092 goto inserted;
2093 }
2094 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2095 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2096 clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2097 ea_bdebug(new_bh, "reusing; refcount now=%d",
2098 ref);
2099 ext4_xattr_block_csum_set(inode, new_bh);
2100 unlock_buffer(new_bh);
2101 error = ext4_handle_dirty_metadata(handle,
2102 inode,
2103 new_bh);
2104 if (error)
2105 goto cleanup_dquot;
2106 }
2107 mb_cache_entry_touch(ea_block_cache, ce);
2108 mb_cache_entry_put(ea_block_cache, ce);
2109 ce = NULL;
2110 } else if (bs->bh && s->base == bs->bh->b_data) {
2111 /* We were modifying this block in-place. */
2112 ea_bdebug(bs->bh, "keeping this block");
2113 ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2114 new_bh = bs->bh;
2115 get_bh(new_bh);
2116 } else {
2117 /* We need to allocate a new block */
2118 ext4_fsblk_t goal, block;
2119
2120 #ifdef EXT4_XATTR_DEBUG
2121 WARN_ON_ONCE(dquot_initialize_needed(inode));
2122 #endif
2123 goal = ext4_group_first_block_no(sb,
2124 EXT4_I(inode)->i_block_group);
2125 block = ext4_new_meta_blocks(handle, inode, goal, 0,
2126 NULL, &error);
2127 if (error)
2128 goto cleanup;
2129
2130 ea_idebug(inode, "creating block %llu",
2131 (unsigned long long)block);
2132
2133 new_bh = sb_getblk(sb, block);
2134 if (unlikely(!new_bh)) {
2135 error = -ENOMEM;
2136 getblk_failed:
2137 ext4_free_blocks(handle, inode, NULL, block, 1,
2138 EXT4_FREE_BLOCKS_METADATA);
2139 goto cleanup;
2140 }
2141 error = ext4_xattr_inode_inc_ref_all(handle, inode,
2142 ENTRY(header(s->base)+1));
2143 if (error)
2144 goto getblk_failed;
2145 if (ea_inode) {
2146 /* Drop the extra ref on ea_inode. */
2147 error = ext4_xattr_inode_dec_ref(handle,
2148 ea_inode);
2149 if (error)
2150 ext4_warning_inode(ea_inode,
2151 "dec ref error=%d",
2152 error);
2153 iput(ea_inode);
2154 ea_inode = NULL;
2155 }
2156
2157 lock_buffer(new_bh);
2158 error = ext4_journal_get_create_access(handle, sb,
2159 new_bh, EXT4_JTR_NONE);
2160 if (error) {
2161 unlock_buffer(new_bh);
2162 error = -EIO;
2163 goto getblk_failed;
2164 }
2165 memcpy(new_bh->b_data, s->base, new_bh->b_size);
2166 ext4_xattr_block_csum_set(inode, new_bh);
2167 set_buffer_uptodate(new_bh);
2168 unlock_buffer(new_bh);
2169 ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2170 error = ext4_handle_dirty_metadata(handle, inode,
2171 new_bh);
2172 if (error)
2173 goto cleanup;
2174 }
2175 }
2176
2177 if (old_ea_inode_quota)
2178 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2179
2180 /* Update the inode. */
2181 EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2182
2183 /* Drop the previous xattr block. */
2184 if (bs->bh && bs->bh != new_bh) {
2185 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2186
2187 ext4_xattr_release_block(handle, inode, bs->bh,
2188 &ea_inode_array,
2189 0 /* extra_credits */);
2190 ext4_xattr_inode_array_free(ea_inode_array);
2191 }
2192 error = 0;
2193
2194 cleanup:
2195 if (ea_inode) {
2196 if (error) {
2197 int error2;
2198
2199 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2200 if (error2)
2201 ext4_warning_inode(ea_inode, "dec ref error=%d",
2202 error2);
2203 ext4_xattr_inode_free_quota(inode, ea_inode,
2204 i_size_read(ea_inode));
2205 }
2206 iput(ea_inode);
2207 }
2208 if (ce)
2209 mb_cache_entry_put(ea_block_cache, ce);
2210 brelse(new_bh);
2211 if (!(bs->bh && s->base == bs->bh->b_data))
2212 kfree(s->base);
2213
2214 return error;
2215
2216 cleanup_dquot:
2217 dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2218 goto cleanup;
2219
2220 bad_block:
2221 EXT4_ERROR_INODE(inode, "bad block %llu",
2222 EXT4_I(inode)->i_file_acl);
2223 goto cleanup;
2224
2225 #undef header
2226 }
2227
2228 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2229 struct ext4_xattr_ibody_find *is)
2230 {
2231 struct ext4_xattr_ibody_header *header;
2232 struct ext4_inode *raw_inode;
2233 int error;
2234
2235 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2236 return 0;
2237
2238 raw_inode = ext4_raw_inode(&is->iloc);
2239 header = IHDR(inode, raw_inode);
2240 is->s.base = is->s.first = IFIRST(header);
2241 is->s.here = is->s.first;
2242 is->s.end = ITAIL(inode, raw_inode);
2243 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2244 /* Find the named attribute. */
2245 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2246 i->name_index, i->name, 0);
2247 if (error && error != -ENODATA)
2248 return error;
2249 is->s.not_found = error;
2250 }
2251 return 0;
2252 }
2253
2254 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2255 struct ext4_xattr_info *i,
2256 struct ext4_xattr_ibody_find *is)
2257 {
2258 struct ext4_xattr_ibody_header *header;
2259 struct ext4_xattr_search *s = &is->s;
2260 struct inode *ea_inode = NULL;
2261 int error;
2262
2263 if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2264 return -ENOSPC;
2265
2266 /* If we need EA inode, prepare it before locking the buffer */
2267 if (i->value && i->in_inode) {
2268 WARN_ON_ONCE(!i->value_len);
2269
2270 ea_inode = ext4_xattr_inode_lookup_create(handle, inode,
2271 i->value, i->value_len);
2272 if (IS_ERR(ea_inode))
2273 return PTR_ERR(ea_inode);
2274 }
2275 error = ext4_xattr_set_entry(i, s, handle, inode, ea_inode,
2276 false /* is_block */);
2277 if (error) {
2278 if (ea_inode) {
2279 int error2;
2280
2281 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2282 if (error2)
2283 ext4_warning_inode(ea_inode, "dec ref error=%d",
2284 error2);
2285
2286 ext4_xattr_inode_free_quota(inode, ea_inode,
2287 i_size_read(ea_inode));
2288 iput(ea_inode);
2289 }
2290 return error;
2291 }
2292 header = IHDR(inode, ext4_raw_inode(&is->iloc));
2293 if (!IS_LAST_ENTRY(s->first)) {
2294 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2295 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2296 } else {
2297 header->h_magic = cpu_to_le32(0);
2298 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2299 }
2300 iput(ea_inode);
2301 return 0;
2302 }
2303
2304 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2305 struct ext4_xattr_info *i)
2306 {
2307 void *value;
2308
2309 /* When e_value_inum is set the value is stored externally. */
2310 if (s->here->e_value_inum)
2311 return 0;
2312 if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2313 return 0;
2314 value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2315 return !memcmp(value, i->value, i->value_len);
2316 }
2317
2318 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2319 {
2320 struct buffer_head *bh;
2321 int error;
2322
2323 if (!EXT4_I(inode)->i_file_acl)
2324 return NULL;
2325 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2326 if (IS_ERR(bh))
2327 return bh;
2328 error = ext4_xattr_check_block(inode, bh);
2329 if (error) {
2330 brelse(bh);
2331 return ERR_PTR(error);
2332 }
2333 return bh;
2334 }
2335
2336 /*
2337 * ext4_xattr_set_handle()
2338 *
2339 * Create, replace or remove an extended attribute for this inode. Value
2340 * is NULL to remove an existing extended attribute, and non-NULL to
2341 * either replace an existing extended attribute, or create a new extended
2342 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2343 * specify that an extended attribute must exist and must not exist
2344 * previous to the call, respectively.
2345 *
2346 * Returns 0, or a negative error number on failure.
2347 */
2348 int
2349 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2350 const char *name, const void *value, size_t value_len,
2351 int flags)
2352 {
2353 struct ext4_xattr_info i = {
2354 .name_index = name_index,
2355 .name = name,
2356 .value = value,
2357 .value_len = value_len,
2358 .in_inode = 0,
2359 };
2360 struct ext4_xattr_ibody_find is = {
2361 .s = { .not_found = -ENODATA, },
2362 };
2363 struct ext4_xattr_block_find bs = {
2364 .s = { .not_found = -ENODATA, },
2365 };
2366 int no_expand;
2367 int error;
2368
2369 if (!name)
2370 return -EINVAL;
2371 if (strlen(name) > 255)
2372 return -ERANGE;
2373
2374 ext4_write_lock_xattr(inode, &no_expand);
2375
2376 /* Check journal credits under write lock. */
2377 if (ext4_handle_valid(handle)) {
2378 struct buffer_head *bh;
2379 int credits;
2380
2381 bh = ext4_xattr_get_block(inode);
2382 if (IS_ERR(bh)) {
2383 error = PTR_ERR(bh);
2384 goto cleanup;
2385 }
2386
2387 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2388 value_len,
2389 flags & XATTR_CREATE);
2390 brelse(bh);
2391
2392 if (jbd2_handle_buffer_credits(handle) < credits) {
2393 error = -ENOSPC;
2394 goto cleanup;
2395 }
2396 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2397 }
2398
2399 error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2400 if (error)
2401 goto cleanup;
2402
2403 if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2404 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2405 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2406 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2407 }
2408
2409 error = ext4_xattr_ibody_find(inode, &i, &is);
2410 if (error)
2411 goto cleanup;
2412 if (is.s.not_found)
2413 error = ext4_xattr_block_find(inode, &i, &bs);
2414 if (error)
2415 goto cleanup;
2416 if (is.s.not_found && bs.s.not_found) {
2417 error = -ENODATA;
2418 if (flags & XATTR_REPLACE)
2419 goto cleanup;
2420 error = 0;
2421 if (!value)
2422 goto cleanup;
2423 } else {
2424 error = -EEXIST;
2425 if (flags & XATTR_CREATE)
2426 goto cleanup;
2427 }
2428
2429 if (!value) {
2430 if (!is.s.not_found)
2431 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2432 else if (!bs.s.not_found)
2433 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2434 } else {
2435 error = 0;
2436 /* Xattr value did not change? Save us some work and bail out */
2437 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2438 goto cleanup;
2439 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2440 goto cleanup;
2441
2442 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2443 (EXT4_XATTR_SIZE(i.value_len) >
2444 EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2445 i.in_inode = 1;
2446 retry_inode:
2447 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2448 if (!error && !bs.s.not_found) {
2449 i.value = NULL;
2450 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2451 } else if (error == -ENOSPC) {
2452 if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2453 brelse(bs.bh);
2454 bs.bh = NULL;
2455 error = ext4_xattr_block_find(inode, &i, &bs);
2456 if (error)
2457 goto cleanup;
2458 }
2459 error = ext4_xattr_block_set(handle, inode, &i, &bs);
2460 if (!error && !is.s.not_found) {
2461 i.value = NULL;
2462 error = ext4_xattr_ibody_set(handle, inode, &i,
2463 &is);
2464 } else if (error == -ENOSPC) {
2465 /*
2466 * Xattr does not fit in the block, store at
2467 * external inode if possible.
2468 */
2469 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2470 i.value_len && !i.in_inode) {
2471 i.in_inode = 1;
2472 goto retry_inode;
2473 }
2474 }
2475 }
2476 }
2477 if (!error) {
2478 ext4_xattr_update_super_block(handle, inode->i_sb);
2479 inode_set_ctime_current(inode);
2480 inode_inc_iversion(inode);
2481 if (!value)
2482 no_expand = 0;
2483 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2484 /*
2485 * The bh is consumed by ext4_mark_iloc_dirty, even with
2486 * error != 0.
2487 */
2488 is.iloc.bh = NULL;
2489 if (IS_SYNC(inode))
2490 ext4_handle_sync(handle);
2491 }
2492 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2493
2494 cleanup:
2495 brelse(is.iloc.bh);
2496 brelse(bs.bh);
2497 ext4_write_unlock_xattr(inode, &no_expand);
2498 return error;
2499 }
2500
2501 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2502 bool is_create, int *credits)
2503 {
2504 struct buffer_head *bh;
2505 int err;
2506
2507 *credits = 0;
2508
2509 if (!EXT4_SB(inode->i_sb)->s_journal)
2510 return 0;
2511
2512 down_read(&EXT4_I(inode)->xattr_sem);
2513
2514 bh = ext4_xattr_get_block(inode);
2515 if (IS_ERR(bh)) {
2516 err = PTR_ERR(bh);
2517 } else {
2518 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2519 value_len, is_create);
2520 brelse(bh);
2521 err = 0;
2522 }
2523
2524 up_read(&EXT4_I(inode)->xattr_sem);
2525 return err;
2526 }
2527
2528 /*
2529 * ext4_xattr_set()
2530 *
2531 * Like ext4_xattr_set_handle, but start from an inode. This extended
2532 * attribute modification is a filesystem transaction by itself.
2533 *
2534 * Returns 0, or a negative error number on failure.
2535 */
2536 int
2537 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2538 const void *value, size_t value_len, int flags)
2539 {
2540 handle_t *handle;
2541 struct super_block *sb = inode->i_sb;
2542 int error, retries = 0;
2543 int credits;
2544
2545 error = dquot_initialize(inode);
2546 if (error)
2547 return error;
2548
2549 retry:
2550 error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2551 &credits);
2552 if (error)
2553 return error;
2554
2555 handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2556 if (IS_ERR(handle)) {
2557 error = PTR_ERR(handle);
2558 } else {
2559 int error2;
2560
2561 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2562 value, value_len, flags);
2563 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR,
2564 handle);
2565 error2 = ext4_journal_stop(handle);
2566 if (error == -ENOSPC &&
2567 ext4_should_retry_alloc(sb, &retries))
2568 goto retry;
2569 if (error == 0)
2570 error = error2;
2571 }
2572
2573 return error;
2574 }
2575
2576 /*
2577 * Shift the EA entries in the inode to create space for the increased
2578 * i_extra_isize.
2579 */
2580 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2581 int value_offs_shift, void *to,
2582 void *from, size_t n)
2583 {
2584 struct ext4_xattr_entry *last = entry;
2585 int new_offs;
2586
2587 /* We always shift xattr headers further thus offsets get lower */
2588 BUG_ON(value_offs_shift > 0);
2589
2590 /* Adjust the value offsets of the entries */
2591 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2592 if (!last->e_value_inum && last->e_value_size) {
2593 new_offs = le16_to_cpu(last->e_value_offs) +
2594 value_offs_shift;
2595 last->e_value_offs = cpu_to_le16(new_offs);
2596 }
2597 }
2598 /* Shift the entries by n bytes */
2599 memmove(to, from, n);
2600 }
2601
2602 /*
2603 * Move xattr pointed to by 'entry' from inode into external xattr block
2604 */
2605 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2606 struct ext4_inode *raw_inode,
2607 struct ext4_xattr_entry *entry)
2608 {
2609 struct ext4_xattr_ibody_find *is = NULL;
2610 struct ext4_xattr_block_find *bs = NULL;
2611 char *buffer = NULL, *b_entry_name = NULL;
2612 size_t value_size = le32_to_cpu(entry->e_value_size);
2613 struct ext4_xattr_info i = {
2614 .value = NULL,
2615 .value_len = 0,
2616 .name_index = entry->e_name_index,
2617 .in_inode = !!entry->e_value_inum,
2618 };
2619 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2620 int needs_kvfree = 0;
2621 int error;
2622
2623 is = kzalloc_obj(struct ext4_xattr_ibody_find, GFP_NOFS);
2624 bs = kzalloc_obj(struct ext4_xattr_block_find, GFP_NOFS);
2625 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2626 if (!is || !bs || !b_entry_name) {
2627 error = -ENOMEM;
2628 goto out;
2629 }
2630
2631 is->s.not_found = -ENODATA;
2632 bs->s.not_found = -ENODATA;
2633 is->iloc.bh = NULL;
2634 bs->bh = NULL;
2635
2636 /* Save the entry name and the entry value */
2637 if (entry->e_value_inum) {
2638 buffer = kvmalloc(value_size, GFP_NOFS);
2639 if (!buffer) {
2640 error = -ENOMEM;
2641 goto out;
2642 }
2643 needs_kvfree = 1;
2644 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2645 if (error)
2646 goto out;
2647 } else {
2648 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2649 buffer = (void *)IFIRST(header) + value_offs;
2650 }
2651
2652 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2653 b_entry_name[entry->e_name_len] = '\0';
2654 i.name = b_entry_name;
2655
2656 error = ext4_get_inode_loc(inode, &is->iloc);
2657 if (error)
2658 goto out;
2659
2660 error = ext4_xattr_ibody_find(inode, &i, is);
2661 if (error)
2662 goto out;
2663
2664 i.value = buffer;
2665 i.value_len = value_size;
2666 error = ext4_xattr_block_find(inode, &i, bs);
2667 if (error)
2668 goto out;
2669
2670 /* Move ea entry from the inode into the block */
2671 error = ext4_xattr_block_set(handle, inode, &i, bs);
2672 if (error)
2673 goto out;
2674
2675 /* Remove the chosen entry from the inode */
2676 i.value = NULL;
2677 i.value_len = 0;
2678 error = ext4_xattr_ibody_set(handle, inode, &i, is);
2679
2680 out:
2681 kfree(b_entry_name);
2682 if (needs_kvfree && buffer)
2683 kvfree(buffer);
2684 if (is)
2685 brelse(is->iloc.bh);
2686 if (bs)
2687 brelse(bs->bh);
2688 kfree(is);
2689 kfree(bs);
2690
2691 return error;
2692 }
2693
2694 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2695 struct ext4_inode *raw_inode,
2696 int isize_diff, size_t ifree,
2697 size_t bfree, int *total_ino)
2698 {
2699 struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2700 struct ext4_xattr_entry *small_entry;
2701 struct ext4_xattr_entry *entry;
2702 struct ext4_xattr_entry *last;
2703 unsigned int entry_size; /* EA entry size */
2704 unsigned int total_size; /* EA entry size + value size */
2705 unsigned int min_total_size;
2706 int error;
2707
2708 while (isize_diff > ifree) {
2709 entry = NULL;
2710 small_entry = NULL;
2711 min_total_size = ~0U;
2712 last = IFIRST(header);
2713 /* Find the entry best suited to be pushed into EA block */
2714 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2715 /* never move system.data out of the inode */
2716 if ((last->e_name_len == 4) &&
2717 (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2718 !memcmp(last->e_name, "data", 4))
2719 continue;
2720 total_size = EXT4_XATTR_LEN(last->e_name_len);
2721 if (!last->e_value_inum)
2722 total_size += EXT4_XATTR_SIZE(
2723 le32_to_cpu(last->e_value_size));
2724 if (total_size <= bfree &&
2725 total_size < min_total_size) {
2726 if (total_size + ifree < isize_diff) {
2727 small_entry = last;
2728 } else {
2729 entry = last;
2730 min_total_size = total_size;
2731 }
2732 }
2733 }
2734
2735 if (entry == NULL) {
2736 if (small_entry == NULL)
2737 return -ENOSPC;
2738 entry = small_entry;
2739 }
2740
2741 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2742 total_size = entry_size;
2743 if (!entry->e_value_inum)
2744 total_size += EXT4_XATTR_SIZE(
2745 le32_to_cpu(entry->e_value_size));
2746 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2747 entry);
2748 if (error)
2749 return error;
2750
2751 *total_ino -= entry_size;
2752 ifree += total_size;
2753 bfree -= total_size;
2754 }
2755
2756 return 0;
2757 }
2758
2759 /*
2760 * Expand an inode by new_extra_isize bytes when EAs are present.
2761 * Returns 0 on success or negative error number on failure.
2762 */
2763 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2764 struct ext4_inode *raw_inode, handle_t *handle)
2765 {
2766 struct ext4_xattr_ibody_header *header;
2767 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2768 static unsigned int mnt_count;
2769 size_t min_offs;
2770 size_t ifree, bfree;
2771 int total_ino;
2772 void *base, *end;
2773 int error = 0, tried_min_extra_isize = 0;
2774 int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2775 int isize_diff; /* How much do we need to grow i_extra_isize */
2776
2777 retry:
2778 isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2779 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2780 return 0;
2781
2782 header = IHDR(inode, raw_inode);
2783
2784 /*
2785 * Check if enough free space is available in the inode to shift the
2786 * entries ahead by new_extra_isize.
2787 */
2788
2789 base = IFIRST(header);
2790 end = ITAIL(inode, raw_inode);
2791 min_offs = end - base;
2792 total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2793
2794 ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2795 if (ifree >= isize_diff)
2796 goto shift;
2797
2798 /*
2799 * Enough free space isn't available in the inode, check if
2800 * EA block can hold new_extra_isize bytes.
2801 */
2802 if (EXT4_I(inode)->i_file_acl) {
2803 struct buffer_head *bh;
2804
2805 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2806 if (IS_ERR(bh)) {
2807 error = PTR_ERR(bh);
2808 goto cleanup;
2809 }
2810 error = ext4_xattr_check_block(inode, bh);
2811 if (error) {
2812 brelse(bh);
2813 goto cleanup;
2814 }
2815 base = BHDR(bh);
2816 end = bh->b_data + bh->b_size;
2817 min_offs = end - base;
2818 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2819 NULL);
2820 brelse(bh);
2821 if (bfree + ifree < isize_diff) {
2822 if (!tried_min_extra_isize && s_min_extra_isize) {
2823 tried_min_extra_isize++;
2824 new_extra_isize = s_min_extra_isize;
2825 goto retry;
2826 }
2827 error = -ENOSPC;
2828 goto cleanup;
2829 }
2830 } else {
2831 bfree = inode->i_sb->s_blocksize;
2832 }
2833
2834 error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2835 isize_diff, ifree, bfree,
2836 &total_ino);
2837 if (error) {
2838 if (error == -ENOSPC && !tried_min_extra_isize &&
2839 s_min_extra_isize) {
2840 tried_min_extra_isize++;
2841 new_extra_isize = s_min_extra_isize;
2842 goto retry;
2843 }
2844 goto cleanup;
2845 }
2846 shift:
2847 /* Adjust the offsets and shift the remaining entries ahead */
2848 ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2849 - new_extra_isize, (void *)raw_inode +
2850 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2851 (void *)header, total_ino);
2852 EXT4_I(inode)->i_extra_isize = new_extra_isize;
2853
2854 if (ext4_has_inline_data(inode))
2855 error = ext4_find_inline_data_nolock(inode);
2856
2857 cleanup:
2858 if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2859 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2860 inode->i_ino);
2861 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2862 }
2863 return error;
2864 }
2865
2866 #define EIA_INCR 16 /* must be 2^n */
2867 #define EIA_MASK (EIA_INCR - 1)
2868
2869 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2870 * If @ea_inode_array is new or full it will be grown and the old
2871 * contents copied over.
2872 */
2873 static int
2874 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2875 struct inode *inode)
2876 {
2877 if (*ea_inode_array == NULL) {
2878 /*
2879 * Start with 15 inodes, so it fits into a power-of-two size.
2880 */
2881 (*ea_inode_array) = kmalloc_flex(**ea_inode_array, inodes,
2882 EIA_MASK, GFP_NOFS);
2883 if (*ea_inode_array == NULL)
2884 return -ENOMEM;
2885 (*ea_inode_array)->count = 0;
2886 } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2887 /* expand the array once all 15 + n * 16 slots are full */
2888 struct ext4_xattr_inode_array *new_array = NULL;
2889
2890 new_array = kmalloc_flex(**ea_inode_array, inodes,
2891 (*ea_inode_array)->count + EIA_INCR,
2892 GFP_NOFS);
2893 if (new_array == NULL)
2894 return -ENOMEM;
2895 memcpy(new_array, *ea_inode_array,
2896 struct_size(*ea_inode_array, inodes,
2897 (*ea_inode_array)->count));
2898 kfree(*ea_inode_array);
2899 *ea_inode_array = new_array;
2900 }
2901 (*ea_inode_array)->count++;
2902 (*ea_inode_array)->inodes[(*ea_inode_array)->count - 1] = inode;
2903 return 0;
2904 }
2905
2906 /*
2907 * ext4_xattr_delete_inode()
2908 *
2909 * Free extended attribute resources associated with this inode. Traverse
2910 * all entries and decrement reference on any xattr inodes associated with this
2911 * inode. This is called immediately before an inode is freed. We have exclusive
2912 * access to the inode. If an orphan inode is deleted it will also release its
2913 * references on xattr block and xattr inodes.
2914 */
2915 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2916 struct ext4_xattr_inode_array **ea_inode_array,
2917 int extra_credits)
2918 {
2919 struct buffer_head *bh = NULL;
2920 struct ext4_xattr_ibody_header *header;
2921 struct ext4_iloc iloc = { .bh = NULL };
2922 struct ext4_xattr_entry *entry;
2923 struct inode *ea_inode;
2924 int error;
2925
2926 error = ext4_journal_ensure_credits(handle, extra_credits,
2927 ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2928 if (error < 0) {
2929 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2930 goto cleanup;
2931 }
2932
2933 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2934 ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2935
2936 error = ext4_get_inode_loc(inode, &iloc);
2937 if (error) {
2938 EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2939 goto cleanup;
2940 }
2941
2942 error = ext4_journal_get_write_access(handle, inode->i_sb,
2943 iloc.bh, EXT4_JTR_NONE);
2944 if (error) {
2945 EXT4_ERROR_INODE(inode, "write access (error %d)",
2946 error);
2947 goto cleanup;
2948 }
2949
2950 header = IHDR(inode, ext4_raw_inode(&iloc));
2951 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2952 ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2953 IFIRST(header),
2954 false /* block_csum */,
2955 ea_inode_array,
2956 extra_credits,
2957 false /* skip_quota */);
2958 }
2959
2960 if (EXT4_I(inode)->i_file_acl) {
2961 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2962 if (IS_ERR(bh)) {
2963 error = PTR_ERR(bh);
2964 if (error == -EIO) {
2965 EXT4_ERROR_INODE_ERR(inode, EIO,
2966 "block %llu read error",
2967 EXT4_I(inode)->i_file_acl);
2968 }
2969 bh = NULL;
2970 goto cleanup;
2971 }
2972 error = ext4_xattr_check_block(inode, bh);
2973 if (error)
2974 goto cleanup;
2975
2976 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2977 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2978 entry = EXT4_XATTR_NEXT(entry)) {
2979 if (!entry->e_value_inum)
2980 continue;
2981 error = ext4_xattr_inode_iget(inode,
2982 le32_to_cpu(entry->e_value_inum),
2983 le32_to_cpu(entry->e_hash),
2984 &ea_inode);
2985 if (error)
2986 continue;
2987 ext4_xattr_inode_free_quota(inode, ea_inode,
2988 le32_to_cpu(entry->e_value_size));
2989 iput(ea_inode);
2990 }
2991
2992 }
2993
2994 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2995 extra_credits);
2996 /*
2997 * Update i_file_acl value in the same transaction that releases
2998 * block.
2999 */
3000 EXT4_I(inode)->i_file_acl = 0;
3001 error = ext4_mark_inode_dirty(handle, inode);
3002 if (error) {
3003 EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
3004 error);
3005 goto cleanup;
3006 }
3007 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
3008 }
3009 error = 0;
3010 cleanup:
3011 brelse(iloc.bh);
3012 brelse(bh);
3013 return error;
3014 }
3015
3016 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
3017 {
3018 int idx;
3019
3020 if (ea_inode_array == NULL)
3021 return;
3022
3023 for (idx = 0; idx < ea_inode_array->count; ++idx)
3024 iput(ea_inode_array->inodes[idx]);
3025 kfree(ea_inode_array);
3026 }
3027
3028 /*
3029 * ext4_xattr_block_cache_insert()
3030 *
3031 * Create a new entry in the extended attribute block cache, and insert
3032 * it unless such an entry is already in the cache.
3033 */
3034 static void
3035 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3036 struct buffer_head *bh)
3037 {
3038 struct ext4_xattr_header *header = BHDR(bh);
3039 __u32 hash = le32_to_cpu(header->h_hash);
3040 int reusable = le32_to_cpu(header->h_refcount) <
3041 EXT4_XATTR_REFCOUNT_MAX;
3042 int error;
3043
3044 if (!ea_block_cache)
3045 return;
3046 error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3047 bh->b_blocknr, reusable);
3048 if (error) {
3049 if (error == -EBUSY)
3050 ea_bdebug(bh, "already in cache");
3051 } else
3052 ea_bdebug(bh, "inserting [%x]", (int)hash);
3053 }
3054
3055 /*
3056 * ext4_xattr_cmp()
3057 *
3058 * Compare two extended attribute blocks for equality.
3059 *
3060 * Returns 0 if the blocks are equal, 1 if they differ.
3061 */
3062 static int
3063 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3064 struct ext4_xattr_header *header2)
3065 {
3066 struct ext4_xattr_entry *entry1, *entry2;
3067
3068 entry1 = ENTRY(header1+1);
3069 entry2 = ENTRY(header2+1);
3070 while (!IS_LAST_ENTRY(entry1)) {
3071 if (IS_LAST_ENTRY(entry2))
3072 return 1;
3073 if (entry1->e_hash != entry2->e_hash ||
3074 entry1->e_name_index != entry2->e_name_index ||
3075 entry1->e_name_len != entry2->e_name_len ||
3076 entry1->e_value_size != entry2->e_value_size ||
3077 entry1->e_value_inum != entry2->e_value_inum ||
3078 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3079 return 1;
3080 if (!entry1->e_value_inum &&
3081 memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3082 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3083 le32_to_cpu(entry1->e_value_size)))
3084 return 1;
3085
3086 entry1 = EXT4_XATTR_NEXT(entry1);
3087 entry2 = EXT4_XATTR_NEXT(entry2);
3088 }
3089 if (!IS_LAST_ENTRY(entry2))
3090 return 1;
3091 return 0;
3092 }
3093
3094 /*
3095 * ext4_xattr_block_cache_find()
3096 *
3097 * Find an identical extended attribute block.
3098 *
3099 * Returns a pointer to the block found, or NULL if such a block was not
3100 * found, or an error pointer if an error occurred while reading ea block.
3101 */
3102 static struct buffer_head *
3103 ext4_xattr_block_cache_find(struct inode *inode,
3104 struct ext4_xattr_header *header,
3105 struct mb_cache_entry **pce)
3106 {
3107 __u32 hash = le32_to_cpu(header->h_hash);
3108 struct mb_cache_entry *ce;
3109 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3110
3111 if (!ea_block_cache)
3112 return NULL;
3113 if (!header->h_hash)
3114 return NULL; /* never share */
3115 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3116 ce = mb_cache_entry_find_first(ea_block_cache, hash);
3117 while (ce) {
3118 struct buffer_head *bh;
3119
3120 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3121 if (IS_ERR(bh)) {
3122 if (PTR_ERR(bh) != -ENOMEM)
3123 EXT4_ERROR_INODE(inode, "block %lu read error",
3124 (unsigned long)ce->e_value);
3125 mb_cache_entry_put(ea_block_cache, ce);
3126 return bh;
3127 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3128 *pce = ce;
3129 return bh;
3130 }
3131 brelse(bh);
3132 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3133 }
3134 return NULL;
3135 }
3136
3137 #define NAME_HASH_SHIFT 5
3138 #define VALUE_HASH_SHIFT 16
3139
3140 /*
3141 * ext4_xattr_hash_entry()
3142 *
3143 * Compute the hash of an extended attribute.
3144 */
3145 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3146 size_t value_count)
3147 {
3148 __u32 hash = 0;
3149
3150 while (name_len--) {
3151 hash = (hash << NAME_HASH_SHIFT) ^
3152 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3153 (unsigned char)*name++;
3154 }
3155 while (value_count--) {
3156 hash = (hash << VALUE_HASH_SHIFT) ^
3157 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3158 le32_to_cpu(*value++);
3159 }
3160 return cpu_to_le32(hash);
3161 }
3162
3163 /*
3164 * ext4_xattr_hash_entry_signed()
3165 *
3166 * Compute the hash of an extended attribute incorrectly.
3167 */
3168 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3169 {
3170 __u32 hash = 0;
3171
3172 while (name_len--) {
3173 hash = (hash << NAME_HASH_SHIFT) ^
3174 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3175 (signed char)*name++;
3176 }
3177 while (value_count--) {
3178 hash = (hash << VALUE_HASH_SHIFT) ^
3179 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3180 le32_to_cpu(*value++);
3181 }
3182 return cpu_to_le32(hash);
3183 }
3184
3185 #undef NAME_HASH_SHIFT
3186 #undef VALUE_HASH_SHIFT
3187
3188 #define BLOCK_HASH_SHIFT 16
3189
3190 /*
3191 * ext4_xattr_rehash()
3192 *
3193 * Re-compute the extended attribute hash value after an entry has changed.
3194 */
3195 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3196 {
3197 struct ext4_xattr_entry *here;
3198 __u32 hash = 0;
3199
3200 here = ENTRY(header+1);
3201 while (!IS_LAST_ENTRY(here)) {
3202 if (!here->e_hash) {
3203 /* Block is not shared if an entry's hash value == 0 */
3204 hash = 0;
3205 break;
3206 }
3207 hash = (hash << BLOCK_HASH_SHIFT) ^
3208 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3209 le32_to_cpu(here->e_hash);
3210 here = EXT4_XATTR_NEXT(here);
3211 }
3212 header->h_hash = cpu_to_le32(hash);
3213 }
3214
3215 #undef BLOCK_HASH_SHIFT
3216
3217 #define HASH_BUCKET_BITS 10
3218
3219 struct mb_cache *
3220 ext4_xattr_create_cache(void)
3221 {
3222 return mb_cache_create(HASH_BUCKET_BITS);
3223 }
3224
3225 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3226 {
3227 if (cache)
3228 mb_cache_destroy(cache);
3229 }
3230