xref: /openbmc/linux/fs/ext4/xattr.c (revision 132db935)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/ext4/xattr.c
4  *
5  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
6  *
7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9  * Extended attributes for symlinks and special files added per
10  *  suggestion of Luka Renko <luka.renko@hermes.si>.
11  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
12  *  Red Hat Inc.
13  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14  *  and Andreas Gruenbacher <agruen@suse.de>.
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 void ext4_xattr_rehash(struct ext4_xattr_header *);
85 
86 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
87 	[EXT4_XATTR_INDEX_USER]		     = &ext4_xattr_user_handler,
88 #ifdef CONFIG_EXT4_FS_POSIX_ACL
89 	[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
90 	[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
91 #endif
92 	[EXT4_XATTR_INDEX_TRUSTED]	     = &ext4_xattr_trusted_handler,
93 #ifdef CONFIG_EXT4_FS_SECURITY
94 	[EXT4_XATTR_INDEX_SECURITY]	     = &ext4_xattr_security_handler,
95 #endif
96 	[EXT4_XATTR_INDEX_HURD]		     = &ext4_xattr_hurd_handler,
97 };
98 
99 const struct xattr_handler *ext4_xattr_handlers[] = {
100 	&ext4_xattr_user_handler,
101 	&ext4_xattr_trusted_handler,
102 #ifdef CONFIG_EXT4_FS_POSIX_ACL
103 	&posix_acl_access_xattr_handler,
104 	&posix_acl_default_xattr_handler,
105 #endif
106 #ifdef CONFIG_EXT4_FS_SECURITY
107 	&ext4_xattr_security_handler,
108 #endif
109 	&ext4_xattr_hurd_handler,
110 	NULL
111 };
112 
113 #define EA_BLOCK_CACHE(inode)	(((struct ext4_sb_info *) \
114 				inode->i_sb->s_fs_info)->s_ea_block_cache)
115 
116 #define EA_INODE_CACHE(inode)	(((struct ext4_sb_info *) \
117 				inode->i_sb->s_fs_info)->s_ea_inode_cache)
118 
119 static int
120 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
121 			struct inode *inode);
122 
123 #ifdef CONFIG_LOCKDEP
124 void ext4_xattr_inode_set_class(struct inode *ea_inode)
125 {
126 	lockdep_set_subclass(&ea_inode->i_rwsem, 1);
127 }
128 #endif
129 
130 static __le32 ext4_xattr_block_csum(struct inode *inode,
131 				    sector_t block_nr,
132 				    struct ext4_xattr_header *hdr)
133 {
134 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
135 	__u32 csum;
136 	__le64 dsk_block_nr = cpu_to_le64(block_nr);
137 	__u32 dummy_csum = 0;
138 	int offset = offsetof(struct ext4_xattr_header, h_checksum);
139 
140 	csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
141 			   sizeof(dsk_block_nr));
142 	csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
143 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
144 	offset += sizeof(dummy_csum);
145 	csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
146 			   EXT4_BLOCK_SIZE(inode->i_sb) - offset);
147 
148 	return cpu_to_le32(csum);
149 }
150 
151 static int ext4_xattr_block_csum_verify(struct inode *inode,
152 					struct buffer_head *bh)
153 {
154 	struct ext4_xattr_header *hdr = BHDR(bh);
155 	int ret = 1;
156 
157 	if (ext4_has_metadata_csum(inode->i_sb)) {
158 		lock_buffer(bh);
159 		ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
160 							bh->b_blocknr, hdr));
161 		unlock_buffer(bh);
162 	}
163 	return ret;
164 }
165 
166 static void ext4_xattr_block_csum_set(struct inode *inode,
167 				      struct buffer_head *bh)
168 {
169 	if (ext4_has_metadata_csum(inode->i_sb))
170 		BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
171 						bh->b_blocknr, BHDR(bh));
172 }
173 
174 static inline const struct xattr_handler *
175 ext4_xattr_handler(int name_index)
176 {
177 	const struct xattr_handler *handler = NULL;
178 
179 	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
180 		handler = ext4_xattr_handler_map[name_index];
181 	return handler;
182 }
183 
184 static int
185 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
186 			 void *value_start)
187 {
188 	struct ext4_xattr_entry *e = entry;
189 
190 	/* Find the end of the names list */
191 	while (!IS_LAST_ENTRY(e)) {
192 		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
193 		if ((void *)next >= end)
194 			return -EFSCORRUPTED;
195 		if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
196 			return -EFSCORRUPTED;
197 		e = next;
198 	}
199 
200 	/* Check the values */
201 	while (!IS_LAST_ENTRY(entry)) {
202 		u32 size = le32_to_cpu(entry->e_value_size);
203 
204 		if (size > EXT4_XATTR_SIZE_MAX)
205 			return -EFSCORRUPTED;
206 
207 		if (size != 0 && entry->e_value_inum == 0) {
208 			u16 offs = le16_to_cpu(entry->e_value_offs);
209 			void *value;
210 
211 			/*
212 			 * The value cannot overlap the names, and the value
213 			 * with padding cannot extend beyond 'end'.  Check both
214 			 * the padded and unpadded sizes, since the size may
215 			 * overflow to 0 when adding padding.
216 			 */
217 			if (offs > end - value_start)
218 				return -EFSCORRUPTED;
219 			value = value_start + offs;
220 			if (value < (void *)e + sizeof(u32) ||
221 			    size > end - value ||
222 			    EXT4_XATTR_SIZE(size) > end - value)
223 				return -EFSCORRUPTED;
224 		}
225 		entry = EXT4_XATTR_NEXT(entry);
226 	}
227 
228 	return 0;
229 }
230 
231 static inline int
232 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
233 			 const char *function, unsigned int line)
234 {
235 	int error = -EFSCORRUPTED;
236 
237 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
238 	    BHDR(bh)->h_blocks != cpu_to_le32(1))
239 		goto errout;
240 	if (buffer_verified(bh))
241 		return 0;
242 
243 	error = -EFSBADCRC;
244 	if (!ext4_xattr_block_csum_verify(inode, bh))
245 		goto errout;
246 	error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
247 					 bh->b_data);
248 errout:
249 	if (error)
250 		__ext4_error_inode(inode, function, line, 0, -error,
251 				   "corrupted xattr block %llu",
252 				   (unsigned long long) bh->b_blocknr);
253 	else
254 		set_buffer_verified(bh);
255 	return error;
256 }
257 
258 #define ext4_xattr_check_block(inode, bh) \
259 	__ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
260 
261 
262 static int
263 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
264 			 void *end, const char *function, unsigned int line)
265 {
266 	int error = -EFSCORRUPTED;
267 
268 	if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
269 	    (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
270 		goto errout;
271 	error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
272 errout:
273 	if (error)
274 		__ext4_error_inode(inode, function, line, 0, -error,
275 				   "corrupted in-inode xattr");
276 	return error;
277 }
278 
279 #define xattr_check_inode(inode, header, end) \
280 	__xattr_check_inode((inode), (header), (end), __func__, __LINE__)
281 
282 static int
283 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
284 		 void *end, int name_index, const char *name, int sorted)
285 {
286 	struct ext4_xattr_entry *entry, *next;
287 	size_t name_len;
288 	int cmp = 1;
289 
290 	if (name == NULL)
291 		return -EINVAL;
292 	name_len = strlen(name);
293 	for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
294 		next = EXT4_XATTR_NEXT(entry);
295 		if ((void *) next >= end) {
296 			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
297 			return -EFSCORRUPTED;
298 		}
299 		cmp = name_index - entry->e_name_index;
300 		if (!cmp)
301 			cmp = name_len - entry->e_name_len;
302 		if (!cmp)
303 			cmp = memcmp(name, entry->e_name, name_len);
304 		if (cmp <= 0 && (sorted || cmp == 0))
305 			break;
306 	}
307 	*pentry = entry;
308 	return cmp ? -ENODATA : 0;
309 }
310 
311 static u32
312 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
313 {
314 	return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
315 }
316 
317 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
318 {
319 	return ((u64)ea_inode->i_ctime.tv_sec << 32) |
320 		(u32) inode_peek_iversion_raw(ea_inode);
321 }
322 
323 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
324 {
325 	ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
326 	inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
327 }
328 
329 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
330 {
331 	return (u32)ea_inode->i_atime.tv_sec;
332 }
333 
334 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
335 {
336 	ea_inode->i_atime.tv_sec = hash;
337 }
338 
339 /*
340  * Read the EA value from an inode.
341  */
342 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
343 {
344 	int blocksize = 1 << ea_inode->i_blkbits;
345 	int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
346 	int tail_size = (size % blocksize) ?: blocksize;
347 	struct buffer_head *bhs_inline[8];
348 	struct buffer_head **bhs = bhs_inline;
349 	int i, ret;
350 
351 	if (bh_count > ARRAY_SIZE(bhs_inline)) {
352 		bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
353 		if (!bhs)
354 			return -ENOMEM;
355 	}
356 
357 	ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
358 			       true /* wait */, bhs);
359 	if (ret)
360 		goto free_bhs;
361 
362 	for (i = 0; i < bh_count; i++) {
363 		/* There shouldn't be any holes in ea_inode. */
364 		if (!bhs[i]) {
365 			ret = -EFSCORRUPTED;
366 			goto put_bhs;
367 		}
368 		memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
369 		       i < bh_count - 1 ? blocksize : tail_size);
370 	}
371 	ret = 0;
372 put_bhs:
373 	for (i = 0; i < bh_count; i++)
374 		brelse(bhs[i]);
375 free_bhs:
376 	if (bhs != bhs_inline)
377 		kfree(bhs);
378 	return ret;
379 }
380 
381 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
382 
383 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
384 				 u32 ea_inode_hash, struct inode **ea_inode)
385 {
386 	struct inode *inode;
387 	int err;
388 
389 	inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
390 	if (IS_ERR(inode)) {
391 		err = PTR_ERR(inode);
392 		ext4_error(parent->i_sb,
393 			   "error while reading EA inode %lu err=%d", ea_ino,
394 			   err);
395 		return err;
396 	}
397 
398 	if (is_bad_inode(inode)) {
399 		ext4_error(parent->i_sb,
400 			   "error while reading EA inode %lu is_bad_inode",
401 			   ea_ino);
402 		err = -EIO;
403 		goto error;
404 	}
405 
406 	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
407 		ext4_error(parent->i_sb,
408 			   "EA inode %lu does not have EXT4_EA_INODE_FL flag",
409 			    ea_ino);
410 		err = -EINVAL;
411 		goto error;
412 	}
413 
414 	ext4_xattr_inode_set_class(inode);
415 
416 	/*
417 	 * Check whether this is an old Lustre-style xattr inode. Lustre
418 	 * implementation does not have hash validation, rather it has a
419 	 * backpointer from ea_inode to the parent inode.
420 	 */
421 	if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
422 	    EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
423 	    inode->i_generation == parent->i_generation) {
424 		ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
425 		ext4_xattr_inode_set_ref(inode, 1);
426 	} else {
427 		inode_lock(inode);
428 		inode->i_flags |= S_NOQUOTA;
429 		inode_unlock(inode);
430 	}
431 
432 	*ea_inode = inode;
433 	return 0;
434 error:
435 	iput(inode);
436 	return err;
437 }
438 
439 static int
440 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
441 			       struct ext4_xattr_entry *entry, void *buffer,
442 			       size_t size)
443 {
444 	u32 hash;
445 
446 	/* Verify stored hash matches calculated hash. */
447 	hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
448 	if (hash != ext4_xattr_inode_get_hash(ea_inode))
449 		return -EFSCORRUPTED;
450 
451 	if (entry) {
452 		__le32 e_hash, tmp_data;
453 
454 		/* Verify entry hash. */
455 		tmp_data = cpu_to_le32(hash);
456 		e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
457 					       &tmp_data, 1);
458 		if (e_hash != entry->e_hash)
459 			return -EFSCORRUPTED;
460 	}
461 	return 0;
462 }
463 
464 /*
465  * Read xattr value from the EA inode.
466  */
467 static int
468 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
469 		     void *buffer, size_t size)
470 {
471 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
472 	struct inode *ea_inode;
473 	int err;
474 
475 	err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
476 				    le32_to_cpu(entry->e_hash), &ea_inode);
477 	if (err) {
478 		ea_inode = NULL;
479 		goto out;
480 	}
481 
482 	if (i_size_read(ea_inode) != size) {
483 		ext4_warning_inode(ea_inode,
484 				   "ea_inode file size=%llu entry size=%zu",
485 				   i_size_read(ea_inode), size);
486 		err = -EFSCORRUPTED;
487 		goto out;
488 	}
489 
490 	err = ext4_xattr_inode_read(ea_inode, buffer, size);
491 	if (err)
492 		goto out;
493 
494 	if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
495 		err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
496 						     size);
497 		if (err) {
498 			ext4_warning_inode(ea_inode,
499 					   "EA inode hash validation failed");
500 			goto out;
501 		}
502 
503 		if (ea_inode_cache)
504 			mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
505 					ext4_xattr_inode_get_hash(ea_inode),
506 					ea_inode->i_ino, true /* reusable */);
507 	}
508 out:
509 	iput(ea_inode);
510 	return err;
511 }
512 
513 static int
514 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
515 		     void *buffer, size_t buffer_size)
516 {
517 	struct buffer_head *bh = NULL;
518 	struct ext4_xattr_entry *entry;
519 	size_t size;
520 	void *end;
521 	int error;
522 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
523 
524 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
525 		  name_index, name, buffer, (long)buffer_size);
526 
527 	if (!EXT4_I(inode)->i_file_acl)
528 		return -ENODATA;
529 	ea_idebug(inode, "reading block %llu",
530 		  (unsigned long long)EXT4_I(inode)->i_file_acl);
531 	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
532 	if (IS_ERR(bh))
533 		return PTR_ERR(bh);
534 	ea_bdebug(bh, "b_count=%d, refcount=%d",
535 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
536 	error = ext4_xattr_check_block(inode, bh);
537 	if (error)
538 		goto cleanup;
539 	ext4_xattr_block_cache_insert(ea_block_cache, bh);
540 	entry = BFIRST(bh);
541 	end = bh->b_data + bh->b_size;
542 	error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
543 	if (error)
544 		goto cleanup;
545 	size = le32_to_cpu(entry->e_value_size);
546 	error = -ERANGE;
547 	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
548 		goto cleanup;
549 	if (buffer) {
550 		if (size > buffer_size)
551 			goto cleanup;
552 		if (entry->e_value_inum) {
553 			error = ext4_xattr_inode_get(inode, entry, buffer,
554 						     size);
555 			if (error)
556 				goto cleanup;
557 		} else {
558 			u16 offset = le16_to_cpu(entry->e_value_offs);
559 			void *p = bh->b_data + offset;
560 
561 			if (unlikely(p + size > end))
562 				goto cleanup;
563 			memcpy(buffer, p, size);
564 		}
565 	}
566 	error = size;
567 
568 cleanup:
569 	brelse(bh);
570 	return error;
571 }
572 
573 int
574 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
575 		     void *buffer, size_t buffer_size)
576 {
577 	struct ext4_xattr_ibody_header *header;
578 	struct ext4_xattr_entry *entry;
579 	struct ext4_inode *raw_inode;
580 	struct ext4_iloc iloc;
581 	size_t size;
582 	void *end;
583 	int error;
584 
585 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
586 		return -ENODATA;
587 	error = ext4_get_inode_loc(inode, &iloc);
588 	if (error)
589 		return error;
590 	raw_inode = ext4_raw_inode(&iloc);
591 	header = IHDR(inode, raw_inode);
592 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
593 	error = xattr_check_inode(inode, header, end);
594 	if (error)
595 		goto cleanup;
596 	entry = IFIRST(header);
597 	error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
598 	if (error)
599 		goto cleanup;
600 	size = le32_to_cpu(entry->e_value_size);
601 	error = -ERANGE;
602 	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
603 		goto cleanup;
604 	if (buffer) {
605 		if (size > buffer_size)
606 			goto cleanup;
607 		if (entry->e_value_inum) {
608 			error = ext4_xattr_inode_get(inode, entry, buffer,
609 						     size);
610 			if (error)
611 				goto cleanup;
612 		} else {
613 			u16 offset = le16_to_cpu(entry->e_value_offs);
614 			void *p = (void *)IFIRST(header) + offset;
615 
616 			if (unlikely(p + size > end))
617 				goto cleanup;
618 			memcpy(buffer, p, size);
619 		}
620 	}
621 	error = size;
622 
623 cleanup:
624 	brelse(iloc.bh);
625 	return error;
626 }
627 
628 /*
629  * ext4_xattr_get()
630  *
631  * Copy an extended attribute into the buffer
632  * provided, or compute the buffer size required.
633  * Buffer is NULL to compute the size of the buffer required.
634  *
635  * Returns a negative error number on failure, or the number of bytes
636  * used / required on success.
637  */
638 int
639 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
640 	       void *buffer, size_t buffer_size)
641 {
642 	int error;
643 
644 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
645 		return -EIO;
646 
647 	if (strlen(name) > 255)
648 		return -ERANGE;
649 
650 	down_read(&EXT4_I(inode)->xattr_sem);
651 	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
652 				     buffer_size);
653 	if (error == -ENODATA)
654 		error = ext4_xattr_block_get(inode, name_index, name, buffer,
655 					     buffer_size);
656 	up_read(&EXT4_I(inode)->xattr_sem);
657 	return error;
658 }
659 
660 static int
661 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
662 			char *buffer, size_t buffer_size)
663 {
664 	size_t rest = buffer_size;
665 
666 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
667 		const struct xattr_handler *handler =
668 			ext4_xattr_handler(entry->e_name_index);
669 
670 		if (handler && (!handler->list || handler->list(dentry))) {
671 			const char *prefix = handler->prefix ?: handler->name;
672 			size_t prefix_len = strlen(prefix);
673 			size_t size = prefix_len + entry->e_name_len + 1;
674 
675 			if (buffer) {
676 				if (size > rest)
677 					return -ERANGE;
678 				memcpy(buffer, prefix, prefix_len);
679 				buffer += prefix_len;
680 				memcpy(buffer, entry->e_name, entry->e_name_len);
681 				buffer += entry->e_name_len;
682 				*buffer++ = 0;
683 			}
684 			rest -= size;
685 		}
686 	}
687 	return buffer_size - rest;  /* total size */
688 }
689 
690 static int
691 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
692 {
693 	struct inode *inode = d_inode(dentry);
694 	struct buffer_head *bh = NULL;
695 	int error;
696 
697 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
698 		  buffer, (long)buffer_size);
699 
700 	if (!EXT4_I(inode)->i_file_acl)
701 		return 0;
702 	ea_idebug(inode, "reading block %llu",
703 		  (unsigned long long)EXT4_I(inode)->i_file_acl);
704 	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
705 	if (IS_ERR(bh))
706 		return PTR_ERR(bh);
707 	ea_bdebug(bh, "b_count=%d, refcount=%d",
708 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
709 	error = ext4_xattr_check_block(inode, bh);
710 	if (error)
711 		goto cleanup;
712 	ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
713 	error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
714 					buffer_size);
715 cleanup:
716 	brelse(bh);
717 	return error;
718 }
719 
720 static int
721 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
722 {
723 	struct inode *inode = d_inode(dentry);
724 	struct ext4_xattr_ibody_header *header;
725 	struct ext4_inode *raw_inode;
726 	struct ext4_iloc iloc;
727 	void *end;
728 	int error;
729 
730 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
731 		return 0;
732 	error = ext4_get_inode_loc(inode, &iloc);
733 	if (error)
734 		return error;
735 	raw_inode = ext4_raw_inode(&iloc);
736 	header = IHDR(inode, raw_inode);
737 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
738 	error = xattr_check_inode(inode, header, end);
739 	if (error)
740 		goto cleanup;
741 	error = ext4_xattr_list_entries(dentry, IFIRST(header),
742 					buffer, buffer_size);
743 
744 cleanup:
745 	brelse(iloc.bh);
746 	return error;
747 }
748 
749 /*
750  * Inode operation listxattr()
751  *
752  * d_inode(dentry)->i_rwsem: don't care
753  *
754  * Copy a list of attribute names into the buffer
755  * provided, or compute the buffer size required.
756  * Buffer is NULL to compute the size of the buffer required.
757  *
758  * Returns a negative error number on failure, or the number of bytes
759  * used / required on success.
760  */
761 ssize_t
762 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
763 {
764 	int ret, ret2;
765 
766 	down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
767 	ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
768 	if (ret < 0)
769 		goto errout;
770 	if (buffer) {
771 		buffer += ret;
772 		buffer_size -= ret;
773 	}
774 	ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
775 	if (ret < 0)
776 		goto errout;
777 	ret += ret2;
778 errout:
779 	up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
780 	return ret;
781 }
782 
783 /*
784  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
785  * not set, set it.
786  */
787 static void ext4_xattr_update_super_block(handle_t *handle,
788 					  struct super_block *sb)
789 {
790 	if (ext4_has_feature_xattr(sb))
791 		return;
792 
793 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
794 	if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
795 		ext4_set_feature_xattr(sb);
796 		ext4_handle_dirty_super(handle, sb);
797 	}
798 }
799 
800 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
801 {
802 	struct ext4_iloc iloc = { .bh = NULL };
803 	struct buffer_head *bh = NULL;
804 	struct ext4_inode *raw_inode;
805 	struct ext4_xattr_ibody_header *header;
806 	struct ext4_xattr_entry *entry;
807 	qsize_t ea_inode_refs = 0;
808 	void *end;
809 	int ret;
810 
811 	lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
812 
813 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
814 		ret = ext4_get_inode_loc(inode, &iloc);
815 		if (ret)
816 			goto out;
817 		raw_inode = ext4_raw_inode(&iloc);
818 		header = IHDR(inode, raw_inode);
819 		end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
820 		ret = xattr_check_inode(inode, header, end);
821 		if (ret)
822 			goto out;
823 
824 		for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
825 		     entry = EXT4_XATTR_NEXT(entry))
826 			if (entry->e_value_inum)
827 				ea_inode_refs++;
828 	}
829 
830 	if (EXT4_I(inode)->i_file_acl) {
831 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
832 		if (IS_ERR(bh)) {
833 			ret = PTR_ERR(bh);
834 			bh = NULL;
835 			goto out;
836 		}
837 
838 		ret = ext4_xattr_check_block(inode, bh);
839 		if (ret)
840 			goto out;
841 
842 		for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
843 		     entry = EXT4_XATTR_NEXT(entry))
844 			if (entry->e_value_inum)
845 				ea_inode_refs++;
846 	}
847 	*usage = ea_inode_refs + 1;
848 	ret = 0;
849 out:
850 	brelse(iloc.bh);
851 	brelse(bh);
852 	return ret;
853 }
854 
855 static inline size_t round_up_cluster(struct inode *inode, size_t length)
856 {
857 	struct super_block *sb = inode->i_sb;
858 	size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
859 				    inode->i_blkbits);
860 	size_t mask = ~(cluster_size - 1);
861 
862 	return (length + cluster_size - 1) & mask;
863 }
864 
865 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
866 {
867 	int err;
868 
869 	err = dquot_alloc_inode(inode);
870 	if (err)
871 		return err;
872 	err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
873 	if (err)
874 		dquot_free_inode(inode);
875 	return err;
876 }
877 
878 static void ext4_xattr_inode_free_quota(struct inode *parent,
879 					struct inode *ea_inode,
880 					size_t len)
881 {
882 	if (ea_inode &&
883 	    ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
884 		return;
885 	dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
886 	dquot_free_inode(parent);
887 }
888 
889 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
890 			     struct buffer_head *block_bh, size_t value_len,
891 			     bool is_create)
892 {
893 	int credits;
894 	int blocks;
895 
896 	/*
897 	 * 1) Owner inode update
898 	 * 2) Ref count update on old xattr block
899 	 * 3) new xattr block
900 	 * 4) block bitmap update for new xattr block
901 	 * 5) group descriptor for new xattr block
902 	 * 6) block bitmap update for old xattr block
903 	 * 7) group descriptor for old block
904 	 *
905 	 * 6 & 7 can happen if we have two racing threads T_a and T_b
906 	 * which are each trying to set an xattr on inodes I_a and I_b
907 	 * which were both initially sharing an xattr block.
908 	 */
909 	credits = 7;
910 
911 	/* Quota updates. */
912 	credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
913 
914 	/*
915 	 * In case of inline data, we may push out the data to a block,
916 	 * so we need to reserve credits for this eventuality
917 	 */
918 	if (inode && ext4_has_inline_data(inode))
919 		credits += ext4_writepage_trans_blocks(inode) + 1;
920 
921 	/* We are done if ea_inode feature is not enabled. */
922 	if (!ext4_has_feature_ea_inode(sb))
923 		return credits;
924 
925 	/* New ea_inode, inode map, block bitmap, group descriptor. */
926 	credits += 4;
927 
928 	/* Data blocks. */
929 	blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
930 
931 	/* Indirection block or one level of extent tree. */
932 	blocks += 1;
933 
934 	/* Block bitmap and group descriptor updates for each block. */
935 	credits += blocks * 2;
936 
937 	/* Blocks themselves. */
938 	credits += blocks;
939 
940 	if (!is_create) {
941 		/* Dereference ea_inode holding old xattr value.
942 		 * Old ea_inode, inode map, block bitmap, group descriptor.
943 		 */
944 		credits += 4;
945 
946 		/* Data blocks for old ea_inode. */
947 		blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
948 
949 		/* Indirection block or one level of extent tree for old
950 		 * ea_inode.
951 		 */
952 		blocks += 1;
953 
954 		/* Block bitmap and group descriptor updates for each block. */
955 		credits += blocks * 2;
956 	}
957 
958 	/* We may need to clone the existing xattr block in which case we need
959 	 * to increment ref counts for existing ea_inodes referenced by it.
960 	 */
961 	if (block_bh) {
962 		struct ext4_xattr_entry *entry = BFIRST(block_bh);
963 
964 		for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
965 			if (entry->e_value_inum)
966 				/* Ref count update on ea_inode. */
967 				credits += 1;
968 	}
969 	return credits;
970 }
971 
972 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
973 				       int ref_change)
974 {
975 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
976 	struct ext4_iloc iloc;
977 	s64 ref_count;
978 	u32 hash;
979 	int ret;
980 
981 	inode_lock(ea_inode);
982 
983 	ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
984 	if (ret)
985 		goto out;
986 
987 	ref_count = ext4_xattr_inode_get_ref(ea_inode);
988 	ref_count += ref_change;
989 	ext4_xattr_inode_set_ref(ea_inode, ref_count);
990 
991 	if (ref_change > 0) {
992 		WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
993 			  ea_inode->i_ino, ref_count);
994 
995 		if (ref_count == 1) {
996 			WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
997 				  ea_inode->i_ino, ea_inode->i_nlink);
998 
999 			set_nlink(ea_inode, 1);
1000 			ext4_orphan_del(handle, ea_inode);
1001 
1002 			if (ea_inode_cache) {
1003 				hash = ext4_xattr_inode_get_hash(ea_inode);
1004 				mb_cache_entry_create(ea_inode_cache,
1005 						      GFP_NOFS, hash,
1006 						      ea_inode->i_ino,
1007 						      true /* reusable */);
1008 			}
1009 		}
1010 	} else {
1011 		WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1012 			  ea_inode->i_ino, ref_count);
1013 
1014 		if (ref_count == 0) {
1015 			WARN_ONCE(ea_inode->i_nlink != 1,
1016 				  "EA inode %lu i_nlink=%u",
1017 				  ea_inode->i_ino, ea_inode->i_nlink);
1018 
1019 			clear_nlink(ea_inode);
1020 			ext4_orphan_add(handle, ea_inode);
1021 
1022 			if (ea_inode_cache) {
1023 				hash = ext4_xattr_inode_get_hash(ea_inode);
1024 				mb_cache_entry_delete(ea_inode_cache, hash,
1025 						      ea_inode->i_ino);
1026 			}
1027 		}
1028 	}
1029 
1030 	ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1031 	if (ret)
1032 		ext4_warning_inode(ea_inode,
1033 				   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1034 out:
1035 	inode_unlock(ea_inode);
1036 	return ret;
1037 }
1038 
1039 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1040 {
1041 	return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1042 }
1043 
1044 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1045 {
1046 	return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1047 }
1048 
1049 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1050 					struct ext4_xattr_entry *first)
1051 {
1052 	struct inode *ea_inode;
1053 	struct ext4_xattr_entry *entry;
1054 	struct ext4_xattr_entry *failed_entry;
1055 	unsigned int ea_ino;
1056 	int err, saved_err;
1057 
1058 	for (entry = first; !IS_LAST_ENTRY(entry);
1059 	     entry = EXT4_XATTR_NEXT(entry)) {
1060 		if (!entry->e_value_inum)
1061 			continue;
1062 		ea_ino = le32_to_cpu(entry->e_value_inum);
1063 		err = ext4_xattr_inode_iget(parent, ea_ino,
1064 					    le32_to_cpu(entry->e_hash),
1065 					    &ea_inode);
1066 		if (err)
1067 			goto cleanup;
1068 		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1069 		if (err) {
1070 			ext4_warning_inode(ea_inode, "inc ref error %d", err);
1071 			iput(ea_inode);
1072 			goto cleanup;
1073 		}
1074 		iput(ea_inode);
1075 	}
1076 	return 0;
1077 
1078 cleanup:
1079 	saved_err = err;
1080 	failed_entry = entry;
1081 
1082 	for (entry = first; entry != failed_entry;
1083 	     entry = EXT4_XATTR_NEXT(entry)) {
1084 		if (!entry->e_value_inum)
1085 			continue;
1086 		ea_ino = le32_to_cpu(entry->e_value_inum);
1087 		err = ext4_xattr_inode_iget(parent, ea_ino,
1088 					    le32_to_cpu(entry->e_hash),
1089 					    &ea_inode);
1090 		if (err) {
1091 			ext4_warning(parent->i_sb,
1092 				     "cleanup ea_ino %u iget error %d", ea_ino,
1093 				     err);
1094 			continue;
1095 		}
1096 		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1097 		if (err)
1098 			ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1099 					   err);
1100 		iput(ea_inode);
1101 	}
1102 	return saved_err;
1103 }
1104 
1105 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1106 			struct buffer_head *bh, bool block_csum, bool dirty)
1107 {
1108 	int error;
1109 
1110 	if (bh && dirty) {
1111 		if (block_csum)
1112 			ext4_xattr_block_csum_set(inode, bh);
1113 		error = ext4_handle_dirty_metadata(handle, NULL, bh);
1114 		if (error) {
1115 			ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1116 				     error);
1117 			return error;
1118 		}
1119 	}
1120 	return 0;
1121 }
1122 
1123 static void
1124 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1125 			     struct buffer_head *bh,
1126 			     struct ext4_xattr_entry *first, bool block_csum,
1127 			     struct ext4_xattr_inode_array **ea_inode_array,
1128 			     int extra_credits, bool skip_quota)
1129 {
1130 	struct inode *ea_inode;
1131 	struct ext4_xattr_entry *entry;
1132 	bool dirty = false;
1133 	unsigned int ea_ino;
1134 	int err;
1135 	int credits;
1136 
1137 	/* One credit for dec ref on ea_inode, one for orphan list addition, */
1138 	credits = 2 + extra_credits;
1139 
1140 	for (entry = first; !IS_LAST_ENTRY(entry);
1141 	     entry = EXT4_XATTR_NEXT(entry)) {
1142 		if (!entry->e_value_inum)
1143 			continue;
1144 		ea_ino = le32_to_cpu(entry->e_value_inum);
1145 		err = ext4_xattr_inode_iget(parent, ea_ino,
1146 					    le32_to_cpu(entry->e_hash),
1147 					    &ea_inode);
1148 		if (err)
1149 			continue;
1150 
1151 		err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1152 		if (err) {
1153 			ext4_warning_inode(ea_inode,
1154 					   "Expand inode array err=%d", err);
1155 			iput(ea_inode);
1156 			continue;
1157 		}
1158 
1159 		err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1160 			ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1161 			ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1162 					      dirty));
1163 		if (err < 0) {
1164 			ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1165 					   err);
1166 			continue;
1167 		}
1168 		if (err > 0) {
1169 			err = ext4_journal_get_write_access(handle, bh);
1170 			if (err) {
1171 				ext4_warning_inode(ea_inode,
1172 						"Re-get write access err=%d",
1173 						err);
1174 				continue;
1175 			}
1176 		}
1177 
1178 		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1179 		if (err) {
1180 			ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1181 					   err);
1182 			continue;
1183 		}
1184 
1185 		if (!skip_quota)
1186 			ext4_xattr_inode_free_quota(parent, ea_inode,
1187 					      le32_to_cpu(entry->e_value_size));
1188 
1189 		/*
1190 		 * Forget about ea_inode within the same transaction that
1191 		 * decrements the ref count. This avoids duplicate decrements in
1192 		 * case the rest of the work spills over to subsequent
1193 		 * transactions.
1194 		 */
1195 		entry->e_value_inum = 0;
1196 		entry->e_value_size = 0;
1197 
1198 		dirty = true;
1199 	}
1200 
1201 	if (dirty) {
1202 		/*
1203 		 * Note that we are deliberately skipping csum calculation for
1204 		 * the final update because we do not expect any journal
1205 		 * restarts until xattr block is freed.
1206 		 */
1207 
1208 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1209 		if (err)
1210 			ext4_warning_inode(parent,
1211 					   "handle dirty metadata err=%d", err);
1212 	}
1213 }
1214 
1215 /*
1216  * Release the xattr block BH: If the reference count is > 1, decrement it;
1217  * otherwise free the block.
1218  */
1219 static void
1220 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1221 			 struct buffer_head *bh,
1222 			 struct ext4_xattr_inode_array **ea_inode_array,
1223 			 int extra_credits)
1224 {
1225 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1226 	u32 hash, ref;
1227 	int error = 0;
1228 
1229 	BUFFER_TRACE(bh, "get_write_access");
1230 	error = ext4_journal_get_write_access(handle, bh);
1231 	if (error)
1232 		goto out;
1233 
1234 	lock_buffer(bh);
1235 	hash = le32_to_cpu(BHDR(bh)->h_hash);
1236 	ref = le32_to_cpu(BHDR(bh)->h_refcount);
1237 	if (ref == 1) {
1238 		ea_bdebug(bh, "refcount now=0; freeing");
1239 		/*
1240 		 * This must happen under buffer lock for
1241 		 * ext4_xattr_block_set() to reliably detect freed block
1242 		 */
1243 		if (ea_block_cache)
1244 			mb_cache_entry_delete(ea_block_cache, hash,
1245 					      bh->b_blocknr);
1246 		get_bh(bh);
1247 		unlock_buffer(bh);
1248 
1249 		if (ext4_has_feature_ea_inode(inode->i_sb))
1250 			ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1251 						     BFIRST(bh),
1252 						     true /* block_csum */,
1253 						     ea_inode_array,
1254 						     extra_credits,
1255 						     true /* skip_quota */);
1256 		ext4_free_blocks(handle, inode, bh, 0, 1,
1257 				 EXT4_FREE_BLOCKS_METADATA |
1258 				 EXT4_FREE_BLOCKS_FORGET);
1259 	} else {
1260 		ref--;
1261 		BHDR(bh)->h_refcount = cpu_to_le32(ref);
1262 		if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1263 			struct mb_cache_entry *ce;
1264 
1265 			if (ea_block_cache) {
1266 				ce = mb_cache_entry_get(ea_block_cache, hash,
1267 							bh->b_blocknr);
1268 				if (ce) {
1269 					ce->e_reusable = 1;
1270 					mb_cache_entry_put(ea_block_cache, ce);
1271 				}
1272 			}
1273 		}
1274 
1275 		ext4_xattr_block_csum_set(inode, bh);
1276 		/*
1277 		 * Beware of this ugliness: Releasing of xattr block references
1278 		 * from different inodes can race and so we have to protect
1279 		 * from a race where someone else frees the block (and releases
1280 		 * its journal_head) before we are done dirtying the buffer. In
1281 		 * nojournal mode this race is harmless and we actually cannot
1282 		 * call ext4_handle_dirty_metadata() with locked buffer as
1283 		 * that function can call sync_dirty_buffer() so for that case
1284 		 * we handle the dirtying after unlocking the buffer.
1285 		 */
1286 		if (ext4_handle_valid(handle))
1287 			error = ext4_handle_dirty_metadata(handle, inode, bh);
1288 		unlock_buffer(bh);
1289 		if (!ext4_handle_valid(handle))
1290 			error = ext4_handle_dirty_metadata(handle, inode, bh);
1291 		if (IS_SYNC(inode))
1292 			ext4_handle_sync(handle);
1293 		dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1294 		ea_bdebug(bh, "refcount now=%d; releasing",
1295 			  le32_to_cpu(BHDR(bh)->h_refcount));
1296 	}
1297 out:
1298 	ext4_std_error(inode->i_sb, error);
1299 	return;
1300 }
1301 
1302 /*
1303  * Find the available free space for EAs. This also returns the total number of
1304  * bytes used by EA entries.
1305  */
1306 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1307 				    size_t *min_offs, void *base, int *total)
1308 {
1309 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1310 		if (!last->e_value_inum && last->e_value_size) {
1311 			size_t offs = le16_to_cpu(last->e_value_offs);
1312 			if (offs < *min_offs)
1313 				*min_offs = offs;
1314 		}
1315 		if (total)
1316 			*total += EXT4_XATTR_LEN(last->e_name_len);
1317 	}
1318 	return (*min_offs - ((void *)last - base) - sizeof(__u32));
1319 }
1320 
1321 /*
1322  * Write the value of the EA in an inode.
1323  */
1324 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1325 				  const void *buf, int bufsize)
1326 {
1327 	struct buffer_head *bh = NULL;
1328 	unsigned long block = 0;
1329 	int blocksize = ea_inode->i_sb->s_blocksize;
1330 	int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1331 	int csize, wsize = 0;
1332 	int ret = 0, ret2 = 0;
1333 	int retries = 0;
1334 
1335 retry:
1336 	while (ret >= 0 && ret < max_blocks) {
1337 		struct ext4_map_blocks map;
1338 		map.m_lblk = block += ret;
1339 		map.m_len = max_blocks -= ret;
1340 
1341 		ret = ext4_map_blocks(handle, ea_inode, &map,
1342 				      EXT4_GET_BLOCKS_CREATE);
1343 		if (ret <= 0) {
1344 			ext4_mark_inode_dirty(handle, ea_inode);
1345 			if (ret == -ENOSPC &&
1346 			    ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1347 				ret = 0;
1348 				goto retry;
1349 			}
1350 			break;
1351 		}
1352 	}
1353 
1354 	if (ret < 0)
1355 		return ret;
1356 
1357 	block = 0;
1358 	while (wsize < bufsize) {
1359 		if (bh != NULL)
1360 			brelse(bh);
1361 		csize = (bufsize - wsize) > blocksize ? blocksize :
1362 								bufsize - wsize;
1363 		bh = ext4_getblk(handle, ea_inode, block, 0);
1364 		if (IS_ERR(bh))
1365 			return PTR_ERR(bh);
1366 		if (!bh) {
1367 			WARN_ON_ONCE(1);
1368 			EXT4_ERROR_INODE(ea_inode,
1369 					 "ext4_getblk() return bh = NULL");
1370 			return -EFSCORRUPTED;
1371 		}
1372 		ret = ext4_journal_get_write_access(handle, bh);
1373 		if (ret)
1374 			goto out;
1375 
1376 		memcpy(bh->b_data, buf, csize);
1377 		set_buffer_uptodate(bh);
1378 		ext4_handle_dirty_metadata(handle, ea_inode, bh);
1379 
1380 		buf += csize;
1381 		wsize += csize;
1382 		block += 1;
1383 	}
1384 
1385 	inode_lock(ea_inode);
1386 	i_size_write(ea_inode, wsize);
1387 	ext4_update_i_disksize(ea_inode, wsize);
1388 	inode_unlock(ea_inode);
1389 
1390 	ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1391 	if (unlikely(ret2 && !ret))
1392 		ret = ret2;
1393 
1394 out:
1395 	brelse(bh);
1396 
1397 	return ret;
1398 }
1399 
1400 /*
1401  * Create an inode to store the value of a large EA.
1402  */
1403 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1404 					     struct inode *inode, u32 hash)
1405 {
1406 	struct inode *ea_inode = NULL;
1407 	uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1408 	int err;
1409 
1410 	/*
1411 	 * Let the next inode be the goal, so we try and allocate the EA inode
1412 	 * in the same group, or nearby one.
1413 	 */
1414 	ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1415 				  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1416 				  EXT4_EA_INODE_FL);
1417 	if (!IS_ERR(ea_inode)) {
1418 		ea_inode->i_op = &ext4_file_inode_operations;
1419 		ea_inode->i_fop = &ext4_file_operations;
1420 		ext4_set_aops(ea_inode);
1421 		ext4_xattr_inode_set_class(ea_inode);
1422 		unlock_new_inode(ea_inode);
1423 		ext4_xattr_inode_set_ref(ea_inode, 1);
1424 		ext4_xattr_inode_set_hash(ea_inode, hash);
1425 		err = ext4_mark_inode_dirty(handle, ea_inode);
1426 		if (!err)
1427 			err = ext4_inode_attach_jinode(ea_inode);
1428 		if (err) {
1429 			iput(ea_inode);
1430 			return ERR_PTR(err);
1431 		}
1432 
1433 		/*
1434 		 * Xattr inodes are shared therefore quota charging is performed
1435 		 * at a higher level.
1436 		 */
1437 		dquot_free_inode(ea_inode);
1438 		dquot_drop(ea_inode);
1439 		inode_lock(ea_inode);
1440 		ea_inode->i_flags |= S_NOQUOTA;
1441 		inode_unlock(ea_inode);
1442 	}
1443 
1444 	return ea_inode;
1445 }
1446 
1447 static struct inode *
1448 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1449 			    size_t value_len, u32 hash)
1450 {
1451 	struct inode *ea_inode;
1452 	struct mb_cache_entry *ce;
1453 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1454 	void *ea_data;
1455 
1456 	if (!ea_inode_cache)
1457 		return NULL;
1458 
1459 	ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1460 	if (!ce)
1461 		return NULL;
1462 
1463 	ea_data = kvmalloc(value_len, GFP_KERNEL);
1464 	if (!ea_data) {
1465 		mb_cache_entry_put(ea_inode_cache, ce);
1466 		return NULL;
1467 	}
1468 
1469 	while (ce) {
1470 		ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1471 				     EXT4_IGET_NORMAL);
1472 		if (!IS_ERR(ea_inode) &&
1473 		    !is_bad_inode(ea_inode) &&
1474 		    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1475 		    i_size_read(ea_inode) == value_len &&
1476 		    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1477 		    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1478 						    value_len) &&
1479 		    !memcmp(value, ea_data, value_len)) {
1480 			mb_cache_entry_touch(ea_inode_cache, ce);
1481 			mb_cache_entry_put(ea_inode_cache, ce);
1482 			kvfree(ea_data);
1483 			return ea_inode;
1484 		}
1485 
1486 		if (!IS_ERR(ea_inode))
1487 			iput(ea_inode);
1488 		ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1489 	}
1490 	kvfree(ea_data);
1491 	return NULL;
1492 }
1493 
1494 /*
1495  * Add value of the EA in an inode.
1496  */
1497 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1498 					  const void *value, size_t value_len,
1499 					  struct inode **ret_inode)
1500 {
1501 	struct inode *ea_inode;
1502 	u32 hash;
1503 	int err;
1504 
1505 	hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1506 	ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1507 	if (ea_inode) {
1508 		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1509 		if (err) {
1510 			iput(ea_inode);
1511 			return err;
1512 		}
1513 
1514 		*ret_inode = ea_inode;
1515 		return 0;
1516 	}
1517 
1518 	/* Create an inode for the EA value */
1519 	ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1520 	if (IS_ERR(ea_inode))
1521 		return PTR_ERR(ea_inode);
1522 
1523 	err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1524 	if (err) {
1525 		ext4_xattr_inode_dec_ref(handle, ea_inode);
1526 		iput(ea_inode);
1527 		return err;
1528 	}
1529 
1530 	if (EA_INODE_CACHE(inode))
1531 		mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1532 				      ea_inode->i_ino, true /* reusable */);
1533 
1534 	*ret_inode = ea_inode;
1535 	return 0;
1536 }
1537 
1538 /*
1539  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1540  * feature is enabled.
1541  */
1542 #define EXT4_XATTR_BLOCK_RESERVE(inode)	min(i_blocksize(inode)/8, 1024U)
1543 
1544 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1545 				struct ext4_xattr_search *s,
1546 				handle_t *handle, struct inode *inode,
1547 				bool is_block)
1548 {
1549 	struct ext4_xattr_entry *last, *next;
1550 	struct ext4_xattr_entry *here = s->here;
1551 	size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1552 	int in_inode = i->in_inode;
1553 	struct inode *old_ea_inode = NULL;
1554 	struct inode *new_ea_inode = NULL;
1555 	size_t old_size, new_size;
1556 	int ret;
1557 
1558 	/* Space used by old and new values. */
1559 	old_size = (!s->not_found && !here->e_value_inum) ?
1560 			EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1561 	new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1562 
1563 	/*
1564 	 * Optimization for the simple case when old and new values have the
1565 	 * same padded sizes. Not applicable if external inodes are involved.
1566 	 */
1567 	if (new_size && new_size == old_size) {
1568 		size_t offs = le16_to_cpu(here->e_value_offs);
1569 		void *val = s->base + offs;
1570 
1571 		here->e_value_size = cpu_to_le32(i->value_len);
1572 		if (i->value == EXT4_ZERO_XATTR_VALUE) {
1573 			memset(val, 0, new_size);
1574 		} else {
1575 			memcpy(val, i->value, i->value_len);
1576 			/* Clear padding bytes. */
1577 			memset(val + i->value_len, 0, new_size - i->value_len);
1578 		}
1579 		goto update_hash;
1580 	}
1581 
1582 	/* Compute min_offs and last. */
1583 	last = s->first;
1584 	for (; !IS_LAST_ENTRY(last); last = next) {
1585 		next = EXT4_XATTR_NEXT(last);
1586 		if ((void *)next >= s->end) {
1587 			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1588 			ret = -EFSCORRUPTED;
1589 			goto out;
1590 		}
1591 		if (!last->e_value_inum && last->e_value_size) {
1592 			size_t offs = le16_to_cpu(last->e_value_offs);
1593 			if (offs < min_offs)
1594 				min_offs = offs;
1595 		}
1596 	}
1597 
1598 	/* Check whether we have enough space. */
1599 	if (i->value) {
1600 		size_t free;
1601 
1602 		free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1603 		if (!s->not_found)
1604 			free += EXT4_XATTR_LEN(name_len) + old_size;
1605 
1606 		if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1607 			ret = -ENOSPC;
1608 			goto out;
1609 		}
1610 
1611 		/*
1612 		 * If storing the value in an external inode is an option,
1613 		 * reserve space for xattr entries/names in the external
1614 		 * attribute block so that a long value does not occupy the
1615 		 * whole space and prevent futher entries being added.
1616 		 */
1617 		if (ext4_has_feature_ea_inode(inode->i_sb) &&
1618 		    new_size && is_block &&
1619 		    (min_offs + old_size - new_size) <
1620 					EXT4_XATTR_BLOCK_RESERVE(inode)) {
1621 			ret = -ENOSPC;
1622 			goto out;
1623 		}
1624 	}
1625 
1626 	/*
1627 	 * Getting access to old and new ea inodes is subject to failures.
1628 	 * Finish that work before doing any modifications to the xattr data.
1629 	 */
1630 	if (!s->not_found && here->e_value_inum) {
1631 		ret = ext4_xattr_inode_iget(inode,
1632 					    le32_to_cpu(here->e_value_inum),
1633 					    le32_to_cpu(here->e_hash),
1634 					    &old_ea_inode);
1635 		if (ret) {
1636 			old_ea_inode = NULL;
1637 			goto out;
1638 		}
1639 	}
1640 	if (i->value && in_inode) {
1641 		WARN_ON_ONCE(!i->value_len);
1642 
1643 		ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1644 		if (ret)
1645 			goto out;
1646 
1647 		ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1648 						     i->value_len,
1649 						     &new_ea_inode);
1650 		if (ret) {
1651 			new_ea_inode = NULL;
1652 			ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1653 			goto out;
1654 		}
1655 	}
1656 
1657 	if (old_ea_inode) {
1658 		/* We are ready to release ref count on the old_ea_inode. */
1659 		ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1660 		if (ret) {
1661 			/* Release newly required ref count on new_ea_inode. */
1662 			if (new_ea_inode) {
1663 				int err;
1664 
1665 				err = ext4_xattr_inode_dec_ref(handle,
1666 							       new_ea_inode);
1667 				if (err)
1668 					ext4_warning_inode(new_ea_inode,
1669 						  "dec ref new_ea_inode err=%d",
1670 						  err);
1671 				ext4_xattr_inode_free_quota(inode, new_ea_inode,
1672 							    i->value_len);
1673 			}
1674 			goto out;
1675 		}
1676 
1677 		ext4_xattr_inode_free_quota(inode, old_ea_inode,
1678 					    le32_to_cpu(here->e_value_size));
1679 	}
1680 
1681 	/* No failures allowed past this point. */
1682 
1683 	if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1684 		/* Remove the old value. */
1685 		void *first_val = s->base + min_offs;
1686 		size_t offs = le16_to_cpu(here->e_value_offs);
1687 		void *val = s->base + offs;
1688 
1689 		memmove(first_val + old_size, first_val, val - first_val);
1690 		memset(first_val, 0, old_size);
1691 		min_offs += old_size;
1692 
1693 		/* Adjust all value offsets. */
1694 		last = s->first;
1695 		while (!IS_LAST_ENTRY(last)) {
1696 			size_t o = le16_to_cpu(last->e_value_offs);
1697 
1698 			if (!last->e_value_inum &&
1699 			    last->e_value_size && o < offs)
1700 				last->e_value_offs = cpu_to_le16(o + old_size);
1701 			last = EXT4_XATTR_NEXT(last);
1702 		}
1703 	}
1704 
1705 	if (!i->value) {
1706 		/* Remove old name. */
1707 		size_t size = EXT4_XATTR_LEN(name_len);
1708 
1709 		last = ENTRY((void *)last - size);
1710 		memmove(here, (void *)here + size,
1711 			(void *)last - (void *)here + sizeof(__u32));
1712 		memset(last, 0, size);
1713 	} else if (s->not_found) {
1714 		/* Insert new name. */
1715 		size_t size = EXT4_XATTR_LEN(name_len);
1716 		size_t rest = (void *)last - (void *)here + sizeof(__u32);
1717 
1718 		memmove((void *)here + size, here, rest);
1719 		memset(here, 0, size);
1720 		here->e_name_index = i->name_index;
1721 		here->e_name_len = name_len;
1722 		memcpy(here->e_name, i->name, name_len);
1723 	} else {
1724 		/* This is an update, reset value info. */
1725 		here->e_value_inum = 0;
1726 		here->e_value_offs = 0;
1727 		here->e_value_size = 0;
1728 	}
1729 
1730 	if (i->value) {
1731 		/* Insert new value. */
1732 		if (in_inode) {
1733 			here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1734 		} else if (i->value_len) {
1735 			void *val = s->base + min_offs - new_size;
1736 
1737 			here->e_value_offs = cpu_to_le16(min_offs - new_size);
1738 			if (i->value == EXT4_ZERO_XATTR_VALUE) {
1739 				memset(val, 0, new_size);
1740 			} else {
1741 				memcpy(val, i->value, i->value_len);
1742 				/* Clear padding bytes. */
1743 				memset(val + i->value_len, 0,
1744 				       new_size - i->value_len);
1745 			}
1746 		}
1747 		here->e_value_size = cpu_to_le32(i->value_len);
1748 	}
1749 
1750 update_hash:
1751 	if (i->value) {
1752 		__le32 hash = 0;
1753 
1754 		/* Entry hash calculation. */
1755 		if (in_inode) {
1756 			__le32 crc32c_hash;
1757 
1758 			/*
1759 			 * Feed crc32c hash instead of the raw value for entry
1760 			 * hash calculation. This is to avoid walking
1761 			 * potentially long value buffer again.
1762 			 */
1763 			crc32c_hash = cpu_to_le32(
1764 				       ext4_xattr_inode_get_hash(new_ea_inode));
1765 			hash = ext4_xattr_hash_entry(here->e_name,
1766 						     here->e_name_len,
1767 						     &crc32c_hash, 1);
1768 		} else if (is_block) {
1769 			__le32 *value = s->base + le16_to_cpu(
1770 							here->e_value_offs);
1771 
1772 			hash = ext4_xattr_hash_entry(here->e_name,
1773 						     here->e_name_len, value,
1774 						     new_size >> 2);
1775 		}
1776 		here->e_hash = hash;
1777 	}
1778 
1779 	if (is_block)
1780 		ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1781 
1782 	ret = 0;
1783 out:
1784 	iput(old_ea_inode);
1785 	iput(new_ea_inode);
1786 	return ret;
1787 }
1788 
1789 struct ext4_xattr_block_find {
1790 	struct ext4_xattr_search s;
1791 	struct buffer_head *bh;
1792 };
1793 
1794 static int
1795 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1796 		      struct ext4_xattr_block_find *bs)
1797 {
1798 	struct super_block *sb = inode->i_sb;
1799 	int error;
1800 
1801 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1802 		  i->name_index, i->name, i->value, (long)i->value_len);
1803 
1804 	if (EXT4_I(inode)->i_file_acl) {
1805 		/* The inode already has an extended attribute block. */
1806 		bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1807 		if (IS_ERR(bs->bh)) {
1808 			error = PTR_ERR(bs->bh);
1809 			bs->bh = NULL;
1810 			return error;
1811 		}
1812 		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1813 			atomic_read(&(bs->bh->b_count)),
1814 			le32_to_cpu(BHDR(bs->bh)->h_refcount));
1815 		error = ext4_xattr_check_block(inode, bs->bh);
1816 		if (error)
1817 			return error;
1818 		/* Find the named attribute. */
1819 		bs->s.base = BHDR(bs->bh);
1820 		bs->s.first = BFIRST(bs->bh);
1821 		bs->s.end = bs->bh->b_data + bs->bh->b_size;
1822 		bs->s.here = bs->s.first;
1823 		error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1824 					 i->name_index, i->name, 1);
1825 		if (error && error != -ENODATA)
1826 			return error;
1827 		bs->s.not_found = error;
1828 	}
1829 	return 0;
1830 }
1831 
1832 static int
1833 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1834 		     struct ext4_xattr_info *i,
1835 		     struct ext4_xattr_block_find *bs)
1836 {
1837 	struct super_block *sb = inode->i_sb;
1838 	struct buffer_head *new_bh = NULL;
1839 	struct ext4_xattr_search s_copy = bs->s;
1840 	struct ext4_xattr_search *s = &s_copy;
1841 	struct mb_cache_entry *ce = NULL;
1842 	int error = 0;
1843 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1844 	struct inode *ea_inode = NULL, *tmp_inode;
1845 	size_t old_ea_inode_quota = 0;
1846 	unsigned int ea_ino;
1847 
1848 
1849 #define header(x) ((struct ext4_xattr_header *)(x))
1850 
1851 	if (s->base) {
1852 		BUFFER_TRACE(bs->bh, "get_write_access");
1853 		error = ext4_journal_get_write_access(handle, bs->bh);
1854 		if (error)
1855 			goto cleanup;
1856 		lock_buffer(bs->bh);
1857 
1858 		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1859 			__u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1860 
1861 			/*
1862 			 * This must happen under buffer lock for
1863 			 * ext4_xattr_block_set() to reliably detect modified
1864 			 * block
1865 			 */
1866 			if (ea_block_cache)
1867 				mb_cache_entry_delete(ea_block_cache, hash,
1868 						      bs->bh->b_blocknr);
1869 			ea_bdebug(bs->bh, "modifying in-place");
1870 			error = ext4_xattr_set_entry(i, s, handle, inode,
1871 						     true /* is_block */);
1872 			ext4_xattr_block_csum_set(inode, bs->bh);
1873 			unlock_buffer(bs->bh);
1874 			if (error == -EFSCORRUPTED)
1875 				goto bad_block;
1876 			if (!error)
1877 				error = ext4_handle_dirty_metadata(handle,
1878 								   inode,
1879 								   bs->bh);
1880 			if (error)
1881 				goto cleanup;
1882 			goto inserted;
1883 		} else {
1884 			int offset = (char *)s->here - bs->bh->b_data;
1885 
1886 			unlock_buffer(bs->bh);
1887 			ea_bdebug(bs->bh, "cloning");
1888 			s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1889 			error = -ENOMEM;
1890 			if (s->base == NULL)
1891 				goto cleanup;
1892 			memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1893 			s->first = ENTRY(header(s->base)+1);
1894 			header(s->base)->h_refcount = cpu_to_le32(1);
1895 			s->here = ENTRY(s->base + offset);
1896 			s->end = s->base + bs->bh->b_size;
1897 
1898 			/*
1899 			 * If existing entry points to an xattr inode, we need
1900 			 * to prevent ext4_xattr_set_entry() from decrementing
1901 			 * ref count on it because the reference belongs to the
1902 			 * original block. In this case, make the entry look
1903 			 * like it has an empty value.
1904 			 */
1905 			if (!s->not_found && s->here->e_value_inum) {
1906 				ea_ino = le32_to_cpu(s->here->e_value_inum);
1907 				error = ext4_xattr_inode_iget(inode, ea_ino,
1908 					      le32_to_cpu(s->here->e_hash),
1909 					      &tmp_inode);
1910 				if (error)
1911 					goto cleanup;
1912 
1913 				if (!ext4_test_inode_state(tmp_inode,
1914 						EXT4_STATE_LUSTRE_EA_INODE)) {
1915 					/*
1916 					 * Defer quota free call for previous
1917 					 * inode until success is guaranteed.
1918 					 */
1919 					old_ea_inode_quota = le32_to_cpu(
1920 							s->here->e_value_size);
1921 				}
1922 				iput(tmp_inode);
1923 
1924 				s->here->e_value_inum = 0;
1925 				s->here->e_value_size = 0;
1926 			}
1927 		}
1928 	} else {
1929 		/* Allocate a buffer where we construct the new block. */
1930 		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1931 		/* assert(header == s->base) */
1932 		error = -ENOMEM;
1933 		if (s->base == NULL)
1934 			goto cleanup;
1935 		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1936 		header(s->base)->h_blocks = cpu_to_le32(1);
1937 		header(s->base)->h_refcount = cpu_to_le32(1);
1938 		s->first = ENTRY(header(s->base)+1);
1939 		s->here = ENTRY(header(s->base)+1);
1940 		s->end = s->base + sb->s_blocksize;
1941 	}
1942 
1943 	error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1944 	if (error == -EFSCORRUPTED)
1945 		goto bad_block;
1946 	if (error)
1947 		goto cleanup;
1948 
1949 	if (i->value && s->here->e_value_inum) {
1950 		/*
1951 		 * A ref count on ea_inode has been taken as part of the call to
1952 		 * ext4_xattr_set_entry() above. We would like to drop this
1953 		 * extra ref but we have to wait until the xattr block is
1954 		 * initialized and has its own ref count on the ea_inode.
1955 		 */
1956 		ea_ino = le32_to_cpu(s->here->e_value_inum);
1957 		error = ext4_xattr_inode_iget(inode, ea_ino,
1958 					      le32_to_cpu(s->here->e_hash),
1959 					      &ea_inode);
1960 		if (error) {
1961 			ea_inode = NULL;
1962 			goto cleanup;
1963 		}
1964 	}
1965 
1966 inserted:
1967 	if (!IS_LAST_ENTRY(s->first)) {
1968 		new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1969 						     &ce);
1970 		if (new_bh) {
1971 			/* We found an identical block in the cache. */
1972 			if (new_bh == bs->bh)
1973 				ea_bdebug(new_bh, "keeping");
1974 			else {
1975 				u32 ref;
1976 
1977 				WARN_ON_ONCE(dquot_initialize_needed(inode));
1978 
1979 				/* The old block is released after updating
1980 				   the inode. */
1981 				error = dquot_alloc_block(inode,
1982 						EXT4_C2B(EXT4_SB(sb), 1));
1983 				if (error)
1984 					goto cleanup;
1985 				BUFFER_TRACE(new_bh, "get_write_access");
1986 				error = ext4_journal_get_write_access(handle,
1987 								      new_bh);
1988 				if (error)
1989 					goto cleanup_dquot;
1990 				lock_buffer(new_bh);
1991 				/*
1992 				 * We have to be careful about races with
1993 				 * freeing, rehashing or adding references to
1994 				 * xattr block. Once we hold buffer lock xattr
1995 				 * block's state is stable so we can check
1996 				 * whether the block got freed / rehashed or
1997 				 * not.  Since we unhash mbcache entry under
1998 				 * buffer lock when freeing / rehashing xattr
1999 				 * block, checking whether entry is still
2000 				 * hashed is reliable. Same rules hold for
2001 				 * e_reusable handling.
2002 				 */
2003 				if (hlist_bl_unhashed(&ce->e_hash_list) ||
2004 				    !ce->e_reusable) {
2005 					/*
2006 					 * Undo everything and check mbcache
2007 					 * again.
2008 					 */
2009 					unlock_buffer(new_bh);
2010 					dquot_free_block(inode,
2011 							 EXT4_C2B(EXT4_SB(sb),
2012 								  1));
2013 					brelse(new_bh);
2014 					mb_cache_entry_put(ea_block_cache, ce);
2015 					ce = NULL;
2016 					new_bh = NULL;
2017 					goto inserted;
2018 				}
2019 				ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2020 				BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2021 				if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2022 					ce->e_reusable = 0;
2023 				ea_bdebug(new_bh, "reusing; refcount now=%d",
2024 					  ref);
2025 				ext4_xattr_block_csum_set(inode, new_bh);
2026 				unlock_buffer(new_bh);
2027 				error = ext4_handle_dirty_metadata(handle,
2028 								   inode,
2029 								   new_bh);
2030 				if (error)
2031 					goto cleanup_dquot;
2032 			}
2033 			mb_cache_entry_touch(ea_block_cache, ce);
2034 			mb_cache_entry_put(ea_block_cache, ce);
2035 			ce = NULL;
2036 		} else if (bs->bh && s->base == bs->bh->b_data) {
2037 			/* We were modifying this block in-place. */
2038 			ea_bdebug(bs->bh, "keeping this block");
2039 			ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2040 			new_bh = bs->bh;
2041 			get_bh(new_bh);
2042 		} else {
2043 			/* We need to allocate a new block */
2044 			ext4_fsblk_t goal, block;
2045 
2046 			WARN_ON_ONCE(dquot_initialize_needed(inode));
2047 
2048 			goal = ext4_group_first_block_no(sb,
2049 						EXT4_I(inode)->i_block_group);
2050 
2051 			/* non-extent files can't have physical blocks past 2^32 */
2052 			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2053 				goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2054 
2055 			block = ext4_new_meta_blocks(handle, inode, goal, 0,
2056 						     NULL, &error);
2057 			if (error)
2058 				goto cleanup;
2059 
2060 			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2061 				BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2062 
2063 			ea_idebug(inode, "creating block %llu",
2064 				  (unsigned long long)block);
2065 
2066 			new_bh = sb_getblk(sb, block);
2067 			if (unlikely(!new_bh)) {
2068 				error = -ENOMEM;
2069 getblk_failed:
2070 				ext4_free_blocks(handle, inode, NULL, block, 1,
2071 						 EXT4_FREE_BLOCKS_METADATA);
2072 				goto cleanup;
2073 			}
2074 			error = ext4_xattr_inode_inc_ref_all(handle, inode,
2075 						      ENTRY(header(s->base)+1));
2076 			if (error)
2077 				goto getblk_failed;
2078 			if (ea_inode) {
2079 				/* Drop the extra ref on ea_inode. */
2080 				error = ext4_xattr_inode_dec_ref(handle,
2081 								 ea_inode);
2082 				if (error)
2083 					ext4_warning_inode(ea_inode,
2084 							   "dec ref error=%d",
2085 							   error);
2086 				iput(ea_inode);
2087 				ea_inode = NULL;
2088 			}
2089 
2090 			lock_buffer(new_bh);
2091 			error = ext4_journal_get_create_access(handle, new_bh);
2092 			if (error) {
2093 				unlock_buffer(new_bh);
2094 				error = -EIO;
2095 				goto getblk_failed;
2096 			}
2097 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
2098 			ext4_xattr_block_csum_set(inode, new_bh);
2099 			set_buffer_uptodate(new_bh);
2100 			unlock_buffer(new_bh);
2101 			ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2102 			error = ext4_handle_dirty_metadata(handle, inode,
2103 							   new_bh);
2104 			if (error)
2105 				goto cleanup;
2106 		}
2107 	}
2108 
2109 	if (old_ea_inode_quota)
2110 		ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2111 
2112 	/* Update the inode. */
2113 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2114 
2115 	/* Drop the previous xattr block. */
2116 	if (bs->bh && bs->bh != new_bh) {
2117 		struct ext4_xattr_inode_array *ea_inode_array = NULL;
2118 
2119 		ext4_xattr_release_block(handle, inode, bs->bh,
2120 					 &ea_inode_array,
2121 					 0 /* extra_credits */);
2122 		ext4_xattr_inode_array_free(ea_inode_array);
2123 	}
2124 	error = 0;
2125 
2126 cleanup:
2127 	if (ea_inode) {
2128 		int error2;
2129 
2130 		error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2131 		if (error2)
2132 			ext4_warning_inode(ea_inode, "dec ref error=%d",
2133 					   error2);
2134 
2135 		/* If there was an error, revert the quota charge. */
2136 		if (error)
2137 			ext4_xattr_inode_free_quota(inode, ea_inode,
2138 						    i_size_read(ea_inode));
2139 		iput(ea_inode);
2140 	}
2141 	if (ce)
2142 		mb_cache_entry_put(ea_block_cache, ce);
2143 	brelse(new_bh);
2144 	if (!(bs->bh && s->base == bs->bh->b_data))
2145 		kfree(s->base);
2146 
2147 	return error;
2148 
2149 cleanup_dquot:
2150 	dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2151 	goto cleanup;
2152 
2153 bad_block:
2154 	EXT4_ERROR_INODE(inode, "bad block %llu",
2155 			 EXT4_I(inode)->i_file_acl);
2156 	goto cleanup;
2157 
2158 #undef header
2159 }
2160 
2161 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2162 			  struct ext4_xattr_ibody_find *is)
2163 {
2164 	struct ext4_xattr_ibody_header *header;
2165 	struct ext4_inode *raw_inode;
2166 	int error;
2167 
2168 	if (EXT4_I(inode)->i_extra_isize == 0)
2169 		return 0;
2170 	raw_inode = ext4_raw_inode(&is->iloc);
2171 	header = IHDR(inode, raw_inode);
2172 	is->s.base = is->s.first = IFIRST(header);
2173 	is->s.here = is->s.first;
2174 	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2175 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2176 		error = xattr_check_inode(inode, header, is->s.end);
2177 		if (error)
2178 			return error;
2179 		/* Find the named attribute. */
2180 		error = xattr_find_entry(inode, &is->s.here, is->s.end,
2181 					 i->name_index, i->name, 0);
2182 		if (error && error != -ENODATA)
2183 			return error;
2184 		is->s.not_found = error;
2185 	}
2186 	return 0;
2187 }
2188 
2189 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2190 				struct ext4_xattr_info *i,
2191 				struct ext4_xattr_ibody_find *is)
2192 {
2193 	struct ext4_xattr_ibody_header *header;
2194 	struct ext4_xattr_search *s = &is->s;
2195 	int error;
2196 
2197 	if (EXT4_I(inode)->i_extra_isize == 0)
2198 		return -ENOSPC;
2199 	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2200 	if (error)
2201 		return error;
2202 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2203 	if (!IS_LAST_ENTRY(s->first)) {
2204 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2205 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2206 	} else {
2207 		header->h_magic = cpu_to_le32(0);
2208 		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2209 	}
2210 	return 0;
2211 }
2212 
2213 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2214 				struct ext4_xattr_info *i,
2215 				struct ext4_xattr_ibody_find *is)
2216 {
2217 	struct ext4_xattr_ibody_header *header;
2218 	struct ext4_xattr_search *s = &is->s;
2219 	int error;
2220 
2221 	if (EXT4_I(inode)->i_extra_isize == 0)
2222 		return -ENOSPC;
2223 	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2224 	if (error)
2225 		return error;
2226 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2227 	if (!IS_LAST_ENTRY(s->first)) {
2228 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2229 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2230 	} else {
2231 		header->h_magic = cpu_to_le32(0);
2232 		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2233 	}
2234 	return 0;
2235 }
2236 
2237 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2238 				 struct ext4_xattr_info *i)
2239 {
2240 	void *value;
2241 
2242 	/* When e_value_inum is set the value is stored externally. */
2243 	if (s->here->e_value_inum)
2244 		return 0;
2245 	if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2246 		return 0;
2247 	value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2248 	return !memcmp(value, i->value, i->value_len);
2249 }
2250 
2251 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2252 {
2253 	struct buffer_head *bh;
2254 	int error;
2255 
2256 	if (!EXT4_I(inode)->i_file_acl)
2257 		return NULL;
2258 	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2259 	if (IS_ERR(bh))
2260 		return bh;
2261 	error = ext4_xattr_check_block(inode, bh);
2262 	if (error) {
2263 		brelse(bh);
2264 		return ERR_PTR(error);
2265 	}
2266 	return bh;
2267 }
2268 
2269 /*
2270  * ext4_xattr_set_handle()
2271  *
2272  * Create, replace or remove an extended attribute for this inode.  Value
2273  * is NULL to remove an existing extended attribute, and non-NULL to
2274  * either replace an existing extended attribute, or create a new extended
2275  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2276  * specify that an extended attribute must exist and must not exist
2277  * previous to the call, respectively.
2278  *
2279  * Returns 0, or a negative error number on failure.
2280  */
2281 int
2282 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2283 		      const char *name, const void *value, size_t value_len,
2284 		      int flags)
2285 {
2286 	struct ext4_xattr_info i = {
2287 		.name_index = name_index,
2288 		.name = name,
2289 		.value = value,
2290 		.value_len = value_len,
2291 		.in_inode = 0,
2292 	};
2293 	struct ext4_xattr_ibody_find is = {
2294 		.s = { .not_found = -ENODATA, },
2295 	};
2296 	struct ext4_xattr_block_find bs = {
2297 		.s = { .not_found = -ENODATA, },
2298 	};
2299 	int no_expand;
2300 	int error;
2301 
2302 	if (!name)
2303 		return -EINVAL;
2304 	if (strlen(name) > 255)
2305 		return -ERANGE;
2306 
2307 	ext4_write_lock_xattr(inode, &no_expand);
2308 
2309 	/* Check journal credits under write lock. */
2310 	if (ext4_handle_valid(handle)) {
2311 		struct buffer_head *bh;
2312 		int credits;
2313 
2314 		bh = ext4_xattr_get_block(inode);
2315 		if (IS_ERR(bh)) {
2316 			error = PTR_ERR(bh);
2317 			goto cleanup;
2318 		}
2319 
2320 		credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2321 						   value_len,
2322 						   flags & XATTR_CREATE);
2323 		brelse(bh);
2324 
2325 		if (jbd2_handle_buffer_credits(handle) < credits) {
2326 			error = -ENOSPC;
2327 			goto cleanup;
2328 		}
2329 	}
2330 
2331 	error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2332 	if (error)
2333 		goto cleanup;
2334 
2335 	if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2336 		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2337 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2338 		ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2339 	}
2340 
2341 	error = ext4_xattr_ibody_find(inode, &i, &is);
2342 	if (error)
2343 		goto cleanup;
2344 	if (is.s.not_found)
2345 		error = ext4_xattr_block_find(inode, &i, &bs);
2346 	if (error)
2347 		goto cleanup;
2348 	if (is.s.not_found && bs.s.not_found) {
2349 		error = -ENODATA;
2350 		if (flags & XATTR_REPLACE)
2351 			goto cleanup;
2352 		error = 0;
2353 		if (!value)
2354 			goto cleanup;
2355 	} else {
2356 		error = -EEXIST;
2357 		if (flags & XATTR_CREATE)
2358 			goto cleanup;
2359 	}
2360 
2361 	if (!value) {
2362 		if (!is.s.not_found)
2363 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2364 		else if (!bs.s.not_found)
2365 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2366 	} else {
2367 		error = 0;
2368 		/* Xattr value did not change? Save us some work and bail out */
2369 		if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2370 			goto cleanup;
2371 		if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2372 			goto cleanup;
2373 
2374 		if (ext4_has_feature_ea_inode(inode->i_sb) &&
2375 		    (EXT4_XATTR_SIZE(i.value_len) >
2376 			EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2377 			i.in_inode = 1;
2378 retry_inode:
2379 		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2380 		if (!error && !bs.s.not_found) {
2381 			i.value = NULL;
2382 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2383 		} else if (error == -ENOSPC) {
2384 			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2385 				brelse(bs.bh);
2386 				bs.bh = NULL;
2387 				error = ext4_xattr_block_find(inode, &i, &bs);
2388 				if (error)
2389 					goto cleanup;
2390 			}
2391 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2392 			if (!error && !is.s.not_found) {
2393 				i.value = NULL;
2394 				error = ext4_xattr_ibody_set(handle, inode, &i,
2395 							     &is);
2396 			} else if (error == -ENOSPC) {
2397 				/*
2398 				 * Xattr does not fit in the block, store at
2399 				 * external inode if possible.
2400 				 */
2401 				if (ext4_has_feature_ea_inode(inode->i_sb) &&
2402 				    !i.in_inode) {
2403 					i.in_inode = 1;
2404 					goto retry_inode;
2405 				}
2406 			}
2407 		}
2408 	}
2409 	if (!error) {
2410 		ext4_xattr_update_super_block(handle, inode->i_sb);
2411 		inode->i_ctime = current_time(inode);
2412 		if (!value)
2413 			no_expand = 0;
2414 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2415 		/*
2416 		 * The bh is consumed by ext4_mark_iloc_dirty, even with
2417 		 * error != 0.
2418 		 */
2419 		is.iloc.bh = NULL;
2420 		if (IS_SYNC(inode))
2421 			ext4_handle_sync(handle);
2422 	}
2423 
2424 cleanup:
2425 	brelse(is.iloc.bh);
2426 	brelse(bs.bh);
2427 	ext4_write_unlock_xattr(inode, &no_expand);
2428 	return error;
2429 }
2430 
2431 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2432 			   bool is_create, int *credits)
2433 {
2434 	struct buffer_head *bh;
2435 	int err;
2436 
2437 	*credits = 0;
2438 
2439 	if (!EXT4_SB(inode->i_sb)->s_journal)
2440 		return 0;
2441 
2442 	down_read(&EXT4_I(inode)->xattr_sem);
2443 
2444 	bh = ext4_xattr_get_block(inode);
2445 	if (IS_ERR(bh)) {
2446 		err = PTR_ERR(bh);
2447 	} else {
2448 		*credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2449 						    value_len, is_create);
2450 		brelse(bh);
2451 		err = 0;
2452 	}
2453 
2454 	up_read(&EXT4_I(inode)->xattr_sem);
2455 	return err;
2456 }
2457 
2458 /*
2459  * ext4_xattr_set()
2460  *
2461  * Like ext4_xattr_set_handle, but start from an inode. This extended
2462  * attribute modification is a filesystem transaction by itself.
2463  *
2464  * Returns 0, or a negative error number on failure.
2465  */
2466 int
2467 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2468 	       const void *value, size_t value_len, int flags)
2469 {
2470 	handle_t *handle;
2471 	struct super_block *sb = inode->i_sb;
2472 	int error, retries = 0;
2473 	int credits;
2474 
2475 	error = dquot_initialize(inode);
2476 	if (error)
2477 		return error;
2478 
2479 retry:
2480 	error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2481 				       &credits);
2482 	if (error)
2483 		return error;
2484 
2485 	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2486 	if (IS_ERR(handle)) {
2487 		error = PTR_ERR(handle);
2488 	} else {
2489 		int error2;
2490 
2491 		error = ext4_xattr_set_handle(handle, inode, name_index, name,
2492 					      value, value_len, flags);
2493 		error2 = ext4_journal_stop(handle);
2494 		if (error == -ENOSPC &&
2495 		    ext4_should_retry_alloc(sb, &retries))
2496 			goto retry;
2497 		if (error == 0)
2498 			error = error2;
2499 	}
2500 
2501 	return error;
2502 }
2503 
2504 /*
2505  * Shift the EA entries in the inode to create space for the increased
2506  * i_extra_isize.
2507  */
2508 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2509 				     int value_offs_shift, void *to,
2510 				     void *from, size_t n)
2511 {
2512 	struct ext4_xattr_entry *last = entry;
2513 	int new_offs;
2514 
2515 	/* We always shift xattr headers further thus offsets get lower */
2516 	BUG_ON(value_offs_shift > 0);
2517 
2518 	/* Adjust the value offsets of the entries */
2519 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2520 		if (!last->e_value_inum && last->e_value_size) {
2521 			new_offs = le16_to_cpu(last->e_value_offs) +
2522 							value_offs_shift;
2523 			last->e_value_offs = cpu_to_le16(new_offs);
2524 		}
2525 	}
2526 	/* Shift the entries by n bytes */
2527 	memmove(to, from, n);
2528 }
2529 
2530 /*
2531  * Move xattr pointed to by 'entry' from inode into external xattr block
2532  */
2533 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2534 				    struct ext4_inode *raw_inode,
2535 				    struct ext4_xattr_entry *entry)
2536 {
2537 	struct ext4_xattr_ibody_find *is = NULL;
2538 	struct ext4_xattr_block_find *bs = NULL;
2539 	char *buffer = NULL, *b_entry_name = NULL;
2540 	size_t value_size = le32_to_cpu(entry->e_value_size);
2541 	struct ext4_xattr_info i = {
2542 		.value = NULL,
2543 		.value_len = 0,
2544 		.name_index = entry->e_name_index,
2545 		.in_inode = !!entry->e_value_inum,
2546 	};
2547 	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2548 	int error;
2549 
2550 	is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2551 	bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2552 	buffer = kmalloc(value_size, GFP_NOFS);
2553 	b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2554 	if (!is || !bs || !buffer || !b_entry_name) {
2555 		error = -ENOMEM;
2556 		goto out;
2557 	}
2558 
2559 	is->s.not_found = -ENODATA;
2560 	bs->s.not_found = -ENODATA;
2561 	is->iloc.bh = NULL;
2562 	bs->bh = NULL;
2563 
2564 	/* Save the entry name and the entry value */
2565 	if (entry->e_value_inum) {
2566 		error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2567 		if (error)
2568 			goto out;
2569 	} else {
2570 		size_t value_offs = le16_to_cpu(entry->e_value_offs);
2571 		memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2572 	}
2573 
2574 	memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2575 	b_entry_name[entry->e_name_len] = '\0';
2576 	i.name = b_entry_name;
2577 
2578 	error = ext4_get_inode_loc(inode, &is->iloc);
2579 	if (error)
2580 		goto out;
2581 
2582 	error = ext4_xattr_ibody_find(inode, &i, is);
2583 	if (error)
2584 		goto out;
2585 
2586 	/* Remove the chosen entry from the inode */
2587 	error = ext4_xattr_ibody_set(handle, inode, &i, is);
2588 	if (error)
2589 		goto out;
2590 
2591 	i.value = buffer;
2592 	i.value_len = value_size;
2593 	error = ext4_xattr_block_find(inode, &i, bs);
2594 	if (error)
2595 		goto out;
2596 
2597 	/* Add entry which was removed from the inode into the block */
2598 	error = ext4_xattr_block_set(handle, inode, &i, bs);
2599 	if (error)
2600 		goto out;
2601 	error = 0;
2602 out:
2603 	kfree(b_entry_name);
2604 	kfree(buffer);
2605 	if (is)
2606 		brelse(is->iloc.bh);
2607 	if (bs)
2608 		brelse(bs->bh);
2609 	kfree(is);
2610 	kfree(bs);
2611 
2612 	return error;
2613 }
2614 
2615 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2616 				       struct ext4_inode *raw_inode,
2617 				       int isize_diff, size_t ifree,
2618 				       size_t bfree, int *total_ino)
2619 {
2620 	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2621 	struct ext4_xattr_entry *small_entry;
2622 	struct ext4_xattr_entry *entry;
2623 	struct ext4_xattr_entry *last;
2624 	unsigned int entry_size;	/* EA entry size */
2625 	unsigned int total_size;	/* EA entry size + value size */
2626 	unsigned int min_total_size;
2627 	int error;
2628 
2629 	while (isize_diff > ifree) {
2630 		entry = NULL;
2631 		small_entry = NULL;
2632 		min_total_size = ~0U;
2633 		last = IFIRST(header);
2634 		/* Find the entry best suited to be pushed into EA block */
2635 		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2636 			/* never move system.data out of the inode */
2637 			if ((last->e_name_len == 4) &&
2638 			    (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2639 			    !memcmp(last->e_name, "data", 4))
2640 				continue;
2641 			total_size = EXT4_XATTR_LEN(last->e_name_len);
2642 			if (!last->e_value_inum)
2643 				total_size += EXT4_XATTR_SIZE(
2644 					       le32_to_cpu(last->e_value_size));
2645 			if (total_size <= bfree &&
2646 			    total_size < min_total_size) {
2647 				if (total_size + ifree < isize_diff) {
2648 					small_entry = last;
2649 				} else {
2650 					entry = last;
2651 					min_total_size = total_size;
2652 				}
2653 			}
2654 		}
2655 
2656 		if (entry == NULL) {
2657 			if (small_entry == NULL)
2658 				return -ENOSPC;
2659 			entry = small_entry;
2660 		}
2661 
2662 		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2663 		total_size = entry_size;
2664 		if (!entry->e_value_inum)
2665 			total_size += EXT4_XATTR_SIZE(
2666 					      le32_to_cpu(entry->e_value_size));
2667 		error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2668 						 entry);
2669 		if (error)
2670 			return error;
2671 
2672 		*total_ino -= entry_size;
2673 		ifree += total_size;
2674 		bfree -= total_size;
2675 	}
2676 
2677 	return 0;
2678 }
2679 
2680 /*
2681  * Expand an inode by new_extra_isize bytes when EAs are present.
2682  * Returns 0 on success or negative error number on failure.
2683  */
2684 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2685 			       struct ext4_inode *raw_inode, handle_t *handle)
2686 {
2687 	struct ext4_xattr_ibody_header *header;
2688 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2689 	static unsigned int mnt_count;
2690 	size_t min_offs;
2691 	size_t ifree, bfree;
2692 	int total_ino;
2693 	void *base, *end;
2694 	int error = 0, tried_min_extra_isize = 0;
2695 	int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2696 	int isize_diff;	/* How much do we need to grow i_extra_isize */
2697 
2698 retry:
2699 	isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2700 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2701 		return 0;
2702 
2703 	header = IHDR(inode, raw_inode);
2704 
2705 	/*
2706 	 * Check if enough free space is available in the inode to shift the
2707 	 * entries ahead by new_extra_isize.
2708 	 */
2709 
2710 	base = IFIRST(header);
2711 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2712 	min_offs = end - base;
2713 	total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2714 
2715 	error = xattr_check_inode(inode, header, end);
2716 	if (error)
2717 		goto cleanup;
2718 
2719 	ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2720 	if (ifree >= isize_diff)
2721 		goto shift;
2722 
2723 	/*
2724 	 * Enough free space isn't available in the inode, check if
2725 	 * EA block can hold new_extra_isize bytes.
2726 	 */
2727 	if (EXT4_I(inode)->i_file_acl) {
2728 		struct buffer_head *bh;
2729 
2730 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2731 		if (IS_ERR(bh)) {
2732 			error = PTR_ERR(bh);
2733 			goto cleanup;
2734 		}
2735 		error = ext4_xattr_check_block(inode, bh);
2736 		if (error) {
2737 			brelse(bh);
2738 			goto cleanup;
2739 		}
2740 		base = BHDR(bh);
2741 		end = bh->b_data + bh->b_size;
2742 		min_offs = end - base;
2743 		bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2744 					      NULL);
2745 		brelse(bh);
2746 		if (bfree + ifree < isize_diff) {
2747 			if (!tried_min_extra_isize && s_min_extra_isize) {
2748 				tried_min_extra_isize++;
2749 				new_extra_isize = s_min_extra_isize;
2750 				goto retry;
2751 			}
2752 			error = -ENOSPC;
2753 			goto cleanup;
2754 		}
2755 	} else {
2756 		bfree = inode->i_sb->s_blocksize;
2757 	}
2758 
2759 	error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2760 					    isize_diff, ifree, bfree,
2761 					    &total_ino);
2762 	if (error) {
2763 		if (error == -ENOSPC && !tried_min_extra_isize &&
2764 		    s_min_extra_isize) {
2765 			tried_min_extra_isize++;
2766 			new_extra_isize = s_min_extra_isize;
2767 			goto retry;
2768 		}
2769 		goto cleanup;
2770 	}
2771 shift:
2772 	/* Adjust the offsets and shift the remaining entries ahead */
2773 	ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2774 			- new_extra_isize, (void *)raw_inode +
2775 			EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2776 			(void *)header, total_ino);
2777 	EXT4_I(inode)->i_extra_isize = new_extra_isize;
2778 
2779 cleanup:
2780 	if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2781 		ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2782 			     inode->i_ino);
2783 		mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2784 	}
2785 	return error;
2786 }
2787 
2788 #define EIA_INCR 16 /* must be 2^n */
2789 #define EIA_MASK (EIA_INCR - 1)
2790 
2791 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2792  * If @ea_inode_array is new or full it will be grown and the old
2793  * contents copied over.
2794  */
2795 static int
2796 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2797 			struct inode *inode)
2798 {
2799 	if (*ea_inode_array == NULL) {
2800 		/*
2801 		 * Start with 15 inodes, so it fits into a power-of-two size.
2802 		 * If *ea_inode_array is NULL, this is essentially offsetof()
2803 		 */
2804 		(*ea_inode_array) =
2805 			kmalloc(offsetof(struct ext4_xattr_inode_array,
2806 					 inodes[EIA_MASK]),
2807 				GFP_NOFS);
2808 		if (*ea_inode_array == NULL)
2809 			return -ENOMEM;
2810 		(*ea_inode_array)->count = 0;
2811 	} else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2812 		/* expand the array once all 15 + n * 16 slots are full */
2813 		struct ext4_xattr_inode_array *new_array = NULL;
2814 		int count = (*ea_inode_array)->count;
2815 
2816 		/* if new_array is NULL, this is essentially offsetof() */
2817 		new_array = kmalloc(
2818 				offsetof(struct ext4_xattr_inode_array,
2819 					 inodes[count + EIA_INCR]),
2820 				GFP_NOFS);
2821 		if (new_array == NULL)
2822 			return -ENOMEM;
2823 		memcpy(new_array, *ea_inode_array,
2824 		       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2825 		kfree(*ea_inode_array);
2826 		*ea_inode_array = new_array;
2827 	}
2828 	(*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2829 	return 0;
2830 }
2831 
2832 /*
2833  * ext4_xattr_delete_inode()
2834  *
2835  * Free extended attribute resources associated with this inode. Traverse
2836  * all entries and decrement reference on any xattr inodes associated with this
2837  * inode. This is called immediately before an inode is freed. We have exclusive
2838  * access to the inode. If an orphan inode is deleted it will also release its
2839  * references on xattr block and xattr inodes.
2840  */
2841 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2842 			    struct ext4_xattr_inode_array **ea_inode_array,
2843 			    int extra_credits)
2844 {
2845 	struct buffer_head *bh = NULL;
2846 	struct ext4_xattr_ibody_header *header;
2847 	struct ext4_iloc iloc = { .bh = NULL };
2848 	struct ext4_xattr_entry *entry;
2849 	struct inode *ea_inode;
2850 	int error;
2851 
2852 	error = ext4_journal_ensure_credits(handle, extra_credits,
2853 			ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2854 	if (error < 0) {
2855 		EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2856 		goto cleanup;
2857 	}
2858 
2859 	if (ext4_has_feature_ea_inode(inode->i_sb) &&
2860 	    ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2861 
2862 		error = ext4_get_inode_loc(inode, &iloc);
2863 		if (error) {
2864 			EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2865 			goto cleanup;
2866 		}
2867 
2868 		error = ext4_journal_get_write_access(handle, iloc.bh);
2869 		if (error) {
2870 			EXT4_ERROR_INODE(inode, "write access (error %d)",
2871 					 error);
2872 			goto cleanup;
2873 		}
2874 
2875 		header = IHDR(inode, ext4_raw_inode(&iloc));
2876 		if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2877 			ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2878 						     IFIRST(header),
2879 						     false /* block_csum */,
2880 						     ea_inode_array,
2881 						     extra_credits,
2882 						     false /* skip_quota */);
2883 	}
2884 
2885 	if (EXT4_I(inode)->i_file_acl) {
2886 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2887 		if (IS_ERR(bh)) {
2888 			error = PTR_ERR(bh);
2889 			if (error == -EIO) {
2890 				EXT4_ERROR_INODE_ERR(inode, EIO,
2891 						     "block %llu read error",
2892 						     EXT4_I(inode)->i_file_acl);
2893 			}
2894 			bh = NULL;
2895 			goto cleanup;
2896 		}
2897 		error = ext4_xattr_check_block(inode, bh);
2898 		if (error)
2899 			goto cleanup;
2900 
2901 		if (ext4_has_feature_ea_inode(inode->i_sb)) {
2902 			for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2903 			     entry = EXT4_XATTR_NEXT(entry)) {
2904 				if (!entry->e_value_inum)
2905 					continue;
2906 				error = ext4_xattr_inode_iget(inode,
2907 					      le32_to_cpu(entry->e_value_inum),
2908 					      le32_to_cpu(entry->e_hash),
2909 					      &ea_inode);
2910 				if (error)
2911 					continue;
2912 				ext4_xattr_inode_free_quota(inode, ea_inode,
2913 					      le32_to_cpu(entry->e_value_size));
2914 				iput(ea_inode);
2915 			}
2916 
2917 		}
2918 
2919 		ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2920 					 extra_credits);
2921 		/*
2922 		 * Update i_file_acl value in the same transaction that releases
2923 		 * block.
2924 		 */
2925 		EXT4_I(inode)->i_file_acl = 0;
2926 		error = ext4_mark_inode_dirty(handle, inode);
2927 		if (error) {
2928 			EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2929 					 error);
2930 			goto cleanup;
2931 		}
2932 	}
2933 	error = 0;
2934 cleanup:
2935 	brelse(iloc.bh);
2936 	brelse(bh);
2937 	return error;
2938 }
2939 
2940 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2941 {
2942 	int idx;
2943 
2944 	if (ea_inode_array == NULL)
2945 		return;
2946 
2947 	for (idx = 0; idx < ea_inode_array->count; ++idx)
2948 		iput(ea_inode_array->inodes[idx]);
2949 	kfree(ea_inode_array);
2950 }
2951 
2952 /*
2953  * ext4_xattr_block_cache_insert()
2954  *
2955  * Create a new entry in the extended attribute block cache, and insert
2956  * it unless such an entry is already in the cache.
2957  *
2958  * Returns 0, or a negative error number on failure.
2959  */
2960 static void
2961 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2962 			      struct buffer_head *bh)
2963 {
2964 	struct ext4_xattr_header *header = BHDR(bh);
2965 	__u32 hash = le32_to_cpu(header->h_hash);
2966 	int reusable = le32_to_cpu(header->h_refcount) <
2967 		       EXT4_XATTR_REFCOUNT_MAX;
2968 	int error;
2969 
2970 	if (!ea_block_cache)
2971 		return;
2972 	error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2973 				      bh->b_blocknr, reusable);
2974 	if (error) {
2975 		if (error == -EBUSY)
2976 			ea_bdebug(bh, "already in cache");
2977 	} else
2978 		ea_bdebug(bh, "inserting [%x]", (int)hash);
2979 }
2980 
2981 /*
2982  * ext4_xattr_cmp()
2983  *
2984  * Compare two extended attribute blocks for equality.
2985  *
2986  * Returns 0 if the blocks are equal, 1 if they differ, and
2987  * a negative error number on errors.
2988  */
2989 static int
2990 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2991 	       struct ext4_xattr_header *header2)
2992 {
2993 	struct ext4_xattr_entry *entry1, *entry2;
2994 
2995 	entry1 = ENTRY(header1+1);
2996 	entry2 = ENTRY(header2+1);
2997 	while (!IS_LAST_ENTRY(entry1)) {
2998 		if (IS_LAST_ENTRY(entry2))
2999 			return 1;
3000 		if (entry1->e_hash != entry2->e_hash ||
3001 		    entry1->e_name_index != entry2->e_name_index ||
3002 		    entry1->e_name_len != entry2->e_name_len ||
3003 		    entry1->e_value_size != entry2->e_value_size ||
3004 		    entry1->e_value_inum != entry2->e_value_inum ||
3005 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3006 			return 1;
3007 		if (!entry1->e_value_inum &&
3008 		    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3009 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3010 			   le32_to_cpu(entry1->e_value_size)))
3011 			return 1;
3012 
3013 		entry1 = EXT4_XATTR_NEXT(entry1);
3014 		entry2 = EXT4_XATTR_NEXT(entry2);
3015 	}
3016 	if (!IS_LAST_ENTRY(entry2))
3017 		return 1;
3018 	return 0;
3019 }
3020 
3021 /*
3022  * ext4_xattr_block_cache_find()
3023  *
3024  * Find an identical extended attribute block.
3025  *
3026  * Returns a pointer to the block found, or NULL if such a block was
3027  * not found or an error occurred.
3028  */
3029 static struct buffer_head *
3030 ext4_xattr_block_cache_find(struct inode *inode,
3031 			    struct ext4_xattr_header *header,
3032 			    struct mb_cache_entry **pce)
3033 {
3034 	__u32 hash = le32_to_cpu(header->h_hash);
3035 	struct mb_cache_entry *ce;
3036 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3037 
3038 	if (!ea_block_cache)
3039 		return NULL;
3040 	if (!header->h_hash)
3041 		return NULL;  /* never share */
3042 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3043 	ce = mb_cache_entry_find_first(ea_block_cache, hash);
3044 	while (ce) {
3045 		struct buffer_head *bh;
3046 
3047 		bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3048 		if (IS_ERR(bh)) {
3049 			if (PTR_ERR(bh) == -ENOMEM)
3050 				return NULL;
3051 			bh = NULL;
3052 			EXT4_ERROR_INODE(inode, "block %lu read error",
3053 					 (unsigned long)ce->e_value);
3054 		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3055 			*pce = ce;
3056 			return bh;
3057 		}
3058 		brelse(bh);
3059 		ce = mb_cache_entry_find_next(ea_block_cache, ce);
3060 	}
3061 	return NULL;
3062 }
3063 
3064 #define NAME_HASH_SHIFT 5
3065 #define VALUE_HASH_SHIFT 16
3066 
3067 /*
3068  * ext4_xattr_hash_entry()
3069  *
3070  * Compute the hash of an extended attribute.
3071  */
3072 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3073 				    size_t value_count)
3074 {
3075 	__u32 hash = 0;
3076 
3077 	while (name_len--) {
3078 		hash = (hash << NAME_HASH_SHIFT) ^
3079 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3080 		       *name++;
3081 	}
3082 	while (value_count--) {
3083 		hash = (hash << VALUE_HASH_SHIFT) ^
3084 		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3085 		       le32_to_cpu(*value++);
3086 	}
3087 	return cpu_to_le32(hash);
3088 }
3089 
3090 #undef NAME_HASH_SHIFT
3091 #undef VALUE_HASH_SHIFT
3092 
3093 #define BLOCK_HASH_SHIFT 16
3094 
3095 /*
3096  * ext4_xattr_rehash()
3097  *
3098  * Re-compute the extended attribute hash value after an entry has changed.
3099  */
3100 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3101 {
3102 	struct ext4_xattr_entry *here;
3103 	__u32 hash = 0;
3104 
3105 	here = ENTRY(header+1);
3106 	while (!IS_LAST_ENTRY(here)) {
3107 		if (!here->e_hash) {
3108 			/* Block is not shared if an entry's hash value == 0 */
3109 			hash = 0;
3110 			break;
3111 		}
3112 		hash = (hash << BLOCK_HASH_SHIFT) ^
3113 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3114 		       le32_to_cpu(here->e_hash);
3115 		here = EXT4_XATTR_NEXT(here);
3116 	}
3117 	header->h_hash = cpu_to_le32(hash);
3118 }
3119 
3120 #undef BLOCK_HASH_SHIFT
3121 
3122 #define	HASH_BUCKET_BITS	10
3123 
3124 struct mb_cache *
3125 ext4_xattr_create_cache(void)
3126 {
3127 	return mb_cache_create(HASH_BUCKET_BITS);
3128 }
3129 
3130 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3131 {
3132 	if (cache)
3133 		mb_cache_destroy(cache);
3134 }
3135 
3136