xref: /openbmc/linux/fs/ext4/xattr.c (revision d3741027)
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, sb, EXT4_SB(sb)->s_sbh,
795 					  EXT4_JTR_NONE) == 0) {
796 		lock_buffer(EXT4_SB(sb)->s_sbh);
797 		ext4_set_feature_xattr(sb);
798 		ext4_superblock_csum_set(sb);
799 		unlock_buffer(EXT4_SB(sb)->s_sbh);
800 		ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
801 	}
802 }
803 
804 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
805 {
806 	struct ext4_iloc iloc = { .bh = NULL };
807 	struct buffer_head *bh = NULL;
808 	struct ext4_inode *raw_inode;
809 	struct ext4_xattr_ibody_header *header;
810 	struct ext4_xattr_entry *entry;
811 	qsize_t ea_inode_refs = 0;
812 	void *end;
813 	int ret;
814 
815 	lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
816 
817 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
818 		ret = ext4_get_inode_loc(inode, &iloc);
819 		if (ret)
820 			goto out;
821 		raw_inode = ext4_raw_inode(&iloc);
822 		header = IHDR(inode, raw_inode);
823 		end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
824 		ret = xattr_check_inode(inode, header, end);
825 		if (ret)
826 			goto out;
827 
828 		for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
829 		     entry = EXT4_XATTR_NEXT(entry))
830 			if (entry->e_value_inum)
831 				ea_inode_refs++;
832 	}
833 
834 	if (EXT4_I(inode)->i_file_acl) {
835 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
836 		if (IS_ERR(bh)) {
837 			ret = PTR_ERR(bh);
838 			bh = NULL;
839 			goto out;
840 		}
841 
842 		ret = ext4_xattr_check_block(inode, bh);
843 		if (ret)
844 			goto out;
845 
846 		for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
847 		     entry = EXT4_XATTR_NEXT(entry))
848 			if (entry->e_value_inum)
849 				ea_inode_refs++;
850 	}
851 	*usage = ea_inode_refs + 1;
852 	ret = 0;
853 out:
854 	brelse(iloc.bh);
855 	brelse(bh);
856 	return ret;
857 }
858 
859 static inline size_t round_up_cluster(struct inode *inode, size_t length)
860 {
861 	struct super_block *sb = inode->i_sb;
862 	size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
863 				    inode->i_blkbits);
864 	size_t mask = ~(cluster_size - 1);
865 
866 	return (length + cluster_size - 1) & mask;
867 }
868 
869 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
870 {
871 	int err;
872 
873 	err = dquot_alloc_inode(inode);
874 	if (err)
875 		return err;
876 	err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
877 	if (err)
878 		dquot_free_inode(inode);
879 	return err;
880 }
881 
882 static void ext4_xattr_inode_free_quota(struct inode *parent,
883 					struct inode *ea_inode,
884 					size_t len)
885 {
886 	if (ea_inode &&
887 	    ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
888 		return;
889 	dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
890 	dquot_free_inode(parent);
891 }
892 
893 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
894 			     struct buffer_head *block_bh, size_t value_len,
895 			     bool is_create)
896 {
897 	int credits;
898 	int blocks;
899 
900 	/*
901 	 * 1) Owner inode update
902 	 * 2) Ref count update on old xattr block
903 	 * 3) new xattr block
904 	 * 4) block bitmap update for new xattr block
905 	 * 5) group descriptor for new xattr block
906 	 * 6) block bitmap update for old xattr block
907 	 * 7) group descriptor for old block
908 	 *
909 	 * 6 & 7 can happen if we have two racing threads T_a and T_b
910 	 * which are each trying to set an xattr on inodes I_a and I_b
911 	 * which were both initially sharing an xattr block.
912 	 */
913 	credits = 7;
914 
915 	/* Quota updates. */
916 	credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
917 
918 	/*
919 	 * In case of inline data, we may push out the data to a block,
920 	 * so we need to reserve credits for this eventuality
921 	 */
922 	if (inode && ext4_has_inline_data(inode))
923 		credits += ext4_writepage_trans_blocks(inode) + 1;
924 
925 	/* We are done if ea_inode feature is not enabled. */
926 	if (!ext4_has_feature_ea_inode(sb))
927 		return credits;
928 
929 	/* New ea_inode, inode map, block bitmap, group descriptor. */
930 	credits += 4;
931 
932 	/* Data blocks. */
933 	blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
934 
935 	/* Indirection block or one level of extent tree. */
936 	blocks += 1;
937 
938 	/* Block bitmap and group descriptor updates for each block. */
939 	credits += blocks * 2;
940 
941 	/* Blocks themselves. */
942 	credits += blocks;
943 
944 	if (!is_create) {
945 		/* Dereference ea_inode holding old xattr value.
946 		 * Old ea_inode, inode map, block bitmap, group descriptor.
947 		 */
948 		credits += 4;
949 
950 		/* Data blocks for old ea_inode. */
951 		blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
952 
953 		/* Indirection block or one level of extent tree for old
954 		 * ea_inode.
955 		 */
956 		blocks += 1;
957 
958 		/* Block bitmap and group descriptor updates for each block. */
959 		credits += blocks * 2;
960 	}
961 
962 	/* We may need to clone the existing xattr block in which case we need
963 	 * to increment ref counts for existing ea_inodes referenced by it.
964 	 */
965 	if (block_bh) {
966 		struct ext4_xattr_entry *entry = BFIRST(block_bh);
967 
968 		for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
969 			if (entry->e_value_inum)
970 				/* Ref count update on ea_inode. */
971 				credits += 1;
972 	}
973 	return credits;
974 }
975 
976 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
977 				       int ref_change)
978 {
979 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
980 	struct ext4_iloc iloc;
981 	s64 ref_count;
982 	u32 hash;
983 	int ret;
984 
985 	inode_lock(ea_inode);
986 
987 	ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
988 	if (ret)
989 		goto out;
990 
991 	ref_count = ext4_xattr_inode_get_ref(ea_inode);
992 	ref_count += ref_change;
993 	ext4_xattr_inode_set_ref(ea_inode, ref_count);
994 
995 	if (ref_change > 0) {
996 		WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
997 			  ea_inode->i_ino, ref_count);
998 
999 		if (ref_count == 1) {
1000 			WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1001 				  ea_inode->i_ino, ea_inode->i_nlink);
1002 
1003 			set_nlink(ea_inode, 1);
1004 			ext4_orphan_del(handle, ea_inode);
1005 
1006 			if (ea_inode_cache) {
1007 				hash = ext4_xattr_inode_get_hash(ea_inode);
1008 				mb_cache_entry_create(ea_inode_cache,
1009 						      GFP_NOFS, hash,
1010 						      ea_inode->i_ino,
1011 						      true /* reusable */);
1012 			}
1013 		}
1014 	} else {
1015 		WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1016 			  ea_inode->i_ino, ref_count);
1017 
1018 		if (ref_count == 0) {
1019 			WARN_ONCE(ea_inode->i_nlink != 1,
1020 				  "EA inode %lu i_nlink=%u",
1021 				  ea_inode->i_ino, ea_inode->i_nlink);
1022 
1023 			clear_nlink(ea_inode);
1024 			ext4_orphan_add(handle, ea_inode);
1025 
1026 			if (ea_inode_cache) {
1027 				hash = ext4_xattr_inode_get_hash(ea_inode);
1028 				mb_cache_entry_delete(ea_inode_cache, hash,
1029 						      ea_inode->i_ino);
1030 			}
1031 		}
1032 	}
1033 
1034 	ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1035 	if (ret)
1036 		ext4_warning_inode(ea_inode,
1037 				   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1038 out:
1039 	inode_unlock(ea_inode);
1040 	return ret;
1041 }
1042 
1043 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1044 {
1045 	return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1046 }
1047 
1048 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1049 {
1050 	return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1051 }
1052 
1053 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1054 					struct ext4_xattr_entry *first)
1055 {
1056 	struct inode *ea_inode;
1057 	struct ext4_xattr_entry *entry;
1058 	struct ext4_xattr_entry *failed_entry;
1059 	unsigned int ea_ino;
1060 	int err, saved_err;
1061 
1062 	for (entry = first; !IS_LAST_ENTRY(entry);
1063 	     entry = EXT4_XATTR_NEXT(entry)) {
1064 		if (!entry->e_value_inum)
1065 			continue;
1066 		ea_ino = le32_to_cpu(entry->e_value_inum);
1067 		err = ext4_xattr_inode_iget(parent, ea_ino,
1068 					    le32_to_cpu(entry->e_hash),
1069 					    &ea_inode);
1070 		if (err)
1071 			goto cleanup;
1072 		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1073 		if (err) {
1074 			ext4_warning_inode(ea_inode, "inc ref error %d", err);
1075 			iput(ea_inode);
1076 			goto cleanup;
1077 		}
1078 		iput(ea_inode);
1079 	}
1080 	return 0;
1081 
1082 cleanup:
1083 	saved_err = err;
1084 	failed_entry = entry;
1085 
1086 	for (entry = first; entry != failed_entry;
1087 	     entry = EXT4_XATTR_NEXT(entry)) {
1088 		if (!entry->e_value_inum)
1089 			continue;
1090 		ea_ino = le32_to_cpu(entry->e_value_inum);
1091 		err = ext4_xattr_inode_iget(parent, ea_ino,
1092 					    le32_to_cpu(entry->e_hash),
1093 					    &ea_inode);
1094 		if (err) {
1095 			ext4_warning(parent->i_sb,
1096 				     "cleanup ea_ino %u iget error %d", ea_ino,
1097 				     err);
1098 			continue;
1099 		}
1100 		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1101 		if (err)
1102 			ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1103 					   err);
1104 		iput(ea_inode);
1105 	}
1106 	return saved_err;
1107 }
1108 
1109 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1110 			struct buffer_head *bh, bool block_csum, bool dirty)
1111 {
1112 	int error;
1113 
1114 	if (bh && dirty) {
1115 		if (block_csum)
1116 			ext4_xattr_block_csum_set(inode, bh);
1117 		error = ext4_handle_dirty_metadata(handle, NULL, bh);
1118 		if (error) {
1119 			ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1120 				     error);
1121 			return error;
1122 		}
1123 	}
1124 	return 0;
1125 }
1126 
1127 static void
1128 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1129 			     struct buffer_head *bh,
1130 			     struct ext4_xattr_entry *first, bool block_csum,
1131 			     struct ext4_xattr_inode_array **ea_inode_array,
1132 			     int extra_credits, bool skip_quota)
1133 {
1134 	struct inode *ea_inode;
1135 	struct ext4_xattr_entry *entry;
1136 	bool dirty = false;
1137 	unsigned int ea_ino;
1138 	int err;
1139 	int credits;
1140 
1141 	/* One credit for dec ref on ea_inode, one for orphan list addition, */
1142 	credits = 2 + extra_credits;
1143 
1144 	for (entry = first; !IS_LAST_ENTRY(entry);
1145 	     entry = EXT4_XATTR_NEXT(entry)) {
1146 		if (!entry->e_value_inum)
1147 			continue;
1148 		ea_ino = le32_to_cpu(entry->e_value_inum);
1149 		err = ext4_xattr_inode_iget(parent, ea_ino,
1150 					    le32_to_cpu(entry->e_hash),
1151 					    &ea_inode);
1152 		if (err)
1153 			continue;
1154 
1155 		err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1156 		if (err) {
1157 			ext4_warning_inode(ea_inode,
1158 					   "Expand inode array err=%d", err);
1159 			iput(ea_inode);
1160 			continue;
1161 		}
1162 
1163 		err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1164 			ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1165 			ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1166 					      dirty));
1167 		if (err < 0) {
1168 			ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1169 					   err);
1170 			continue;
1171 		}
1172 		if (err > 0) {
1173 			err = ext4_journal_get_write_access(handle,
1174 					parent->i_sb, bh, EXT4_JTR_NONE);
1175 			if (err) {
1176 				ext4_warning_inode(ea_inode,
1177 						"Re-get write access err=%d",
1178 						err);
1179 				continue;
1180 			}
1181 		}
1182 
1183 		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1184 		if (err) {
1185 			ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1186 					   err);
1187 			continue;
1188 		}
1189 
1190 		if (!skip_quota)
1191 			ext4_xattr_inode_free_quota(parent, ea_inode,
1192 					      le32_to_cpu(entry->e_value_size));
1193 
1194 		/*
1195 		 * Forget about ea_inode within the same transaction that
1196 		 * decrements the ref count. This avoids duplicate decrements in
1197 		 * case the rest of the work spills over to subsequent
1198 		 * transactions.
1199 		 */
1200 		entry->e_value_inum = 0;
1201 		entry->e_value_size = 0;
1202 
1203 		dirty = true;
1204 	}
1205 
1206 	if (dirty) {
1207 		/*
1208 		 * Note that we are deliberately skipping csum calculation for
1209 		 * the final update because we do not expect any journal
1210 		 * restarts until xattr block is freed.
1211 		 */
1212 
1213 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1214 		if (err)
1215 			ext4_warning_inode(parent,
1216 					   "handle dirty metadata err=%d", err);
1217 	}
1218 }
1219 
1220 /*
1221  * Release the xattr block BH: If the reference count is > 1, decrement it;
1222  * otherwise free the block.
1223  */
1224 static void
1225 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1226 			 struct buffer_head *bh,
1227 			 struct ext4_xattr_inode_array **ea_inode_array,
1228 			 int extra_credits)
1229 {
1230 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1231 	u32 hash, ref;
1232 	int error = 0;
1233 
1234 	BUFFER_TRACE(bh, "get_write_access");
1235 	error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1236 					      EXT4_JTR_NONE);
1237 	if (error)
1238 		goto out;
1239 
1240 	lock_buffer(bh);
1241 	hash = le32_to_cpu(BHDR(bh)->h_hash);
1242 	ref = le32_to_cpu(BHDR(bh)->h_refcount);
1243 	if (ref == 1) {
1244 		ea_bdebug(bh, "refcount now=0; freeing");
1245 		/*
1246 		 * This must happen under buffer lock for
1247 		 * ext4_xattr_block_set() to reliably detect freed block
1248 		 */
1249 		if (ea_block_cache)
1250 			mb_cache_entry_delete(ea_block_cache, hash,
1251 					      bh->b_blocknr);
1252 		get_bh(bh);
1253 		unlock_buffer(bh);
1254 
1255 		if (ext4_has_feature_ea_inode(inode->i_sb))
1256 			ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1257 						     BFIRST(bh),
1258 						     true /* block_csum */,
1259 						     ea_inode_array,
1260 						     extra_credits,
1261 						     true /* skip_quota */);
1262 		ext4_free_blocks(handle, inode, bh, 0, 1,
1263 				 EXT4_FREE_BLOCKS_METADATA |
1264 				 EXT4_FREE_BLOCKS_FORGET);
1265 	} else {
1266 		ref--;
1267 		BHDR(bh)->h_refcount = cpu_to_le32(ref);
1268 		if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1269 			struct mb_cache_entry *ce;
1270 
1271 			if (ea_block_cache) {
1272 				ce = mb_cache_entry_get(ea_block_cache, hash,
1273 							bh->b_blocknr);
1274 				if (ce) {
1275 					ce->e_reusable = 1;
1276 					mb_cache_entry_put(ea_block_cache, ce);
1277 				}
1278 			}
1279 		}
1280 
1281 		ext4_xattr_block_csum_set(inode, bh);
1282 		/*
1283 		 * Beware of this ugliness: Releasing of xattr block references
1284 		 * from different inodes can race and so we have to protect
1285 		 * from a race where someone else frees the block (and releases
1286 		 * its journal_head) before we are done dirtying the buffer. In
1287 		 * nojournal mode this race is harmless and we actually cannot
1288 		 * call ext4_handle_dirty_metadata() with locked buffer as
1289 		 * that function can call sync_dirty_buffer() so for that case
1290 		 * we handle the dirtying after unlocking the buffer.
1291 		 */
1292 		if (ext4_handle_valid(handle))
1293 			error = ext4_handle_dirty_metadata(handle, inode, bh);
1294 		unlock_buffer(bh);
1295 		if (!ext4_handle_valid(handle))
1296 			error = ext4_handle_dirty_metadata(handle, inode, bh);
1297 		if (IS_SYNC(inode))
1298 			ext4_handle_sync(handle);
1299 		dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1300 		ea_bdebug(bh, "refcount now=%d; releasing",
1301 			  le32_to_cpu(BHDR(bh)->h_refcount));
1302 	}
1303 out:
1304 	ext4_std_error(inode->i_sb, error);
1305 	return;
1306 }
1307 
1308 /*
1309  * Find the available free space for EAs. This also returns the total number of
1310  * bytes used by EA entries.
1311  */
1312 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1313 				    size_t *min_offs, void *base, int *total)
1314 {
1315 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1316 		if (!last->e_value_inum && last->e_value_size) {
1317 			size_t offs = le16_to_cpu(last->e_value_offs);
1318 			if (offs < *min_offs)
1319 				*min_offs = offs;
1320 		}
1321 		if (total)
1322 			*total += EXT4_XATTR_LEN(last->e_name_len);
1323 	}
1324 	return (*min_offs - ((void *)last - base) - sizeof(__u32));
1325 }
1326 
1327 /*
1328  * Write the value of the EA in an inode.
1329  */
1330 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1331 				  const void *buf, int bufsize)
1332 {
1333 	struct buffer_head *bh = NULL;
1334 	unsigned long block = 0;
1335 	int blocksize = ea_inode->i_sb->s_blocksize;
1336 	int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1337 	int csize, wsize = 0;
1338 	int ret = 0, ret2 = 0;
1339 	int retries = 0;
1340 
1341 retry:
1342 	while (ret >= 0 && ret < max_blocks) {
1343 		struct ext4_map_blocks map;
1344 		map.m_lblk = block += ret;
1345 		map.m_len = max_blocks -= ret;
1346 
1347 		ret = ext4_map_blocks(handle, ea_inode, &map,
1348 				      EXT4_GET_BLOCKS_CREATE);
1349 		if (ret <= 0) {
1350 			ext4_mark_inode_dirty(handle, ea_inode);
1351 			if (ret == -ENOSPC &&
1352 			    ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1353 				ret = 0;
1354 				goto retry;
1355 			}
1356 			break;
1357 		}
1358 	}
1359 
1360 	if (ret < 0)
1361 		return ret;
1362 
1363 	block = 0;
1364 	while (wsize < bufsize) {
1365 		brelse(bh);
1366 		csize = (bufsize - wsize) > blocksize ? blocksize :
1367 								bufsize - wsize;
1368 		bh = ext4_getblk(handle, ea_inode, block, 0);
1369 		if (IS_ERR(bh))
1370 			return PTR_ERR(bh);
1371 		if (!bh) {
1372 			WARN_ON_ONCE(1);
1373 			EXT4_ERROR_INODE(ea_inode,
1374 					 "ext4_getblk() return bh = NULL");
1375 			return -EFSCORRUPTED;
1376 		}
1377 		ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1378 						   EXT4_JTR_NONE);
1379 		if (ret)
1380 			goto out;
1381 
1382 		memcpy(bh->b_data, buf, csize);
1383 		set_buffer_uptodate(bh);
1384 		ext4_handle_dirty_metadata(handle, ea_inode, bh);
1385 
1386 		buf += csize;
1387 		wsize += csize;
1388 		block += 1;
1389 	}
1390 
1391 	inode_lock(ea_inode);
1392 	i_size_write(ea_inode, wsize);
1393 	ext4_update_i_disksize(ea_inode, wsize);
1394 	inode_unlock(ea_inode);
1395 
1396 	ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1397 	if (unlikely(ret2 && !ret))
1398 		ret = ret2;
1399 
1400 out:
1401 	brelse(bh);
1402 
1403 	return ret;
1404 }
1405 
1406 /*
1407  * Create an inode to store the value of a large EA.
1408  */
1409 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1410 					     struct inode *inode, u32 hash)
1411 {
1412 	struct inode *ea_inode = NULL;
1413 	uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1414 	int err;
1415 
1416 	/*
1417 	 * Let the next inode be the goal, so we try and allocate the EA inode
1418 	 * in the same group, or nearby one.
1419 	 */
1420 	ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1421 				  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1422 				  EXT4_EA_INODE_FL);
1423 	if (!IS_ERR(ea_inode)) {
1424 		ea_inode->i_op = &ext4_file_inode_operations;
1425 		ea_inode->i_fop = &ext4_file_operations;
1426 		ext4_set_aops(ea_inode);
1427 		ext4_xattr_inode_set_class(ea_inode);
1428 		unlock_new_inode(ea_inode);
1429 		ext4_xattr_inode_set_ref(ea_inode, 1);
1430 		ext4_xattr_inode_set_hash(ea_inode, hash);
1431 		err = ext4_mark_inode_dirty(handle, ea_inode);
1432 		if (!err)
1433 			err = ext4_inode_attach_jinode(ea_inode);
1434 		if (err) {
1435 			iput(ea_inode);
1436 			return ERR_PTR(err);
1437 		}
1438 
1439 		/*
1440 		 * Xattr inodes are shared therefore quota charging is performed
1441 		 * at a higher level.
1442 		 */
1443 		dquot_free_inode(ea_inode);
1444 		dquot_drop(ea_inode);
1445 		inode_lock(ea_inode);
1446 		ea_inode->i_flags |= S_NOQUOTA;
1447 		inode_unlock(ea_inode);
1448 	}
1449 
1450 	return ea_inode;
1451 }
1452 
1453 static struct inode *
1454 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1455 			    size_t value_len, u32 hash)
1456 {
1457 	struct inode *ea_inode;
1458 	struct mb_cache_entry *ce;
1459 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1460 	void *ea_data;
1461 
1462 	if (!ea_inode_cache)
1463 		return NULL;
1464 
1465 	ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1466 	if (!ce)
1467 		return NULL;
1468 
1469 	WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1470 		     !(current->flags & PF_MEMALLOC_NOFS));
1471 
1472 	ea_data = kvmalloc(value_len, GFP_KERNEL);
1473 	if (!ea_data) {
1474 		mb_cache_entry_put(ea_inode_cache, ce);
1475 		return NULL;
1476 	}
1477 
1478 	while (ce) {
1479 		ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1480 				     EXT4_IGET_NORMAL);
1481 		if (!IS_ERR(ea_inode) &&
1482 		    !is_bad_inode(ea_inode) &&
1483 		    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1484 		    i_size_read(ea_inode) == value_len &&
1485 		    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1486 		    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1487 						    value_len) &&
1488 		    !memcmp(value, ea_data, value_len)) {
1489 			mb_cache_entry_touch(ea_inode_cache, ce);
1490 			mb_cache_entry_put(ea_inode_cache, ce);
1491 			kvfree(ea_data);
1492 			return ea_inode;
1493 		}
1494 
1495 		if (!IS_ERR(ea_inode))
1496 			iput(ea_inode);
1497 		ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1498 	}
1499 	kvfree(ea_data);
1500 	return NULL;
1501 }
1502 
1503 /*
1504  * Add value of the EA in an inode.
1505  */
1506 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1507 					  const void *value, size_t value_len,
1508 					  struct inode **ret_inode)
1509 {
1510 	struct inode *ea_inode;
1511 	u32 hash;
1512 	int err;
1513 
1514 	hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1515 	ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1516 	if (ea_inode) {
1517 		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1518 		if (err) {
1519 			iput(ea_inode);
1520 			return err;
1521 		}
1522 
1523 		*ret_inode = ea_inode;
1524 		return 0;
1525 	}
1526 
1527 	/* Create an inode for the EA value */
1528 	ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1529 	if (IS_ERR(ea_inode))
1530 		return PTR_ERR(ea_inode);
1531 
1532 	err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1533 	if (err) {
1534 		ext4_xattr_inode_dec_ref(handle, ea_inode);
1535 		iput(ea_inode);
1536 		return err;
1537 	}
1538 
1539 	if (EA_INODE_CACHE(inode))
1540 		mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1541 				      ea_inode->i_ino, true /* reusable */);
1542 
1543 	*ret_inode = ea_inode;
1544 	return 0;
1545 }
1546 
1547 /*
1548  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1549  * feature is enabled.
1550  */
1551 #define EXT4_XATTR_BLOCK_RESERVE(inode)	min(i_blocksize(inode)/8, 1024U)
1552 
1553 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1554 				struct ext4_xattr_search *s,
1555 				handle_t *handle, struct inode *inode,
1556 				bool is_block)
1557 {
1558 	struct ext4_xattr_entry *last, *next;
1559 	struct ext4_xattr_entry *here = s->here;
1560 	size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1561 	int in_inode = i->in_inode;
1562 	struct inode *old_ea_inode = NULL;
1563 	struct inode *new_ea_inode = NULL;
1564 	size_t old_size, new_size;
1565 	int ret;
1566 
1567 	/* Space used by old and new values. */
1568 	old_size = (!s->not_found && !here->e_value_inum) ?
1569 			EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1570 	new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1571 
1572 	/*
1573 	 * Optimization for the simple case when old and new values have the
1574 	 * same padded sizes. Not applicable if external inodes are involved.
1575 	 */
1576 	if (new_size && new_size == old_size) {
1577 		size_t offs = le16_to_cpu(here->e_value_offs);
1578 		void *val = s->base + offs;
1579 
1580 		here->e_value_size = cpu_to_le32(i->value_len);
1581 		if (i->value == EXT4_ZERO_XATTR_VALUE) {
1582 			memset(val, 0, new_size);
1583 		} else {
1584 			memcpy(val, i->value, i->value_len);
1585 			/* Clear padding bytes. */
1586 			memset(val + i->value_len, 0, new_size - i->value_len);
1587 		}
1588 		goto update_hash;
1589 	}
1590 
1591 	/* Compute min_offs and last. */
1592 	last = s->first;
1593 	for (; !IS_LAST_ENTRY(last); last = next) {
1594 		next = EXT4_XATTR_NEXT(last);
1595 		if ((void *)next >= s->end) {
1596 			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1597 			ret = -EFSCORRUPTED;
1598 			goto out;
1599 		}
1600 		if (!last->e_value_inum && last->e_value_size) {
1601 			size_t offs = le16_to_cpu(last->e_value_offs);
1602 			if (offs < min_offs)
1603 				min_offs = offs;
1604 		}
1605 	}
1606 
1607 	/* Check whether we have enough space. */
1608 	if (i->value) {
1609 		size_t free;
1610 
1611 		free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1612 		if (!s->not_found)
1613 			free += EXT4_XATTR_LEN(name_len) + old_size;
1614 
1615 		if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1616 			ret = -ENOSPC;
1617 			goto out;
1618 		}
1619 
1620 		/*
1621 		 * If storing the value in an external inode is an option,
1622 		 * reserve space for xattr entries/names in the external
1623 		 * attribute block so that a long value does not occupy the
1624 		 * whole space and prevent further entries being added.
1625 		 */
1626 		if (ext4_has_feature_ea_inode(inode->i_sb) &&
1627 		    new_size && is_block &&
1628 		    (min_offs + old_size - new_size) <
1629 					EXT4_XATTR_BLOCK_RESERVE(inode)) {
1630 			ret = -ENOSPC;
1631 			goto out;
1632 		}
1633 	}
1634 
1635 	/*
1636 	 * Getting access to old and new ea inodes is subject to failures.
1637 	 * Finish that work before doing any modifications to the xattr data.
1638 	 */
1639 	if (!s->not_found && here->e_value_inum) {
1640 		ret = ext4_xattr_inode_iget(inode,
1641 					    le32_to_cpu(here->e_value_inum),
1642 					    le32_to_cpu(here->e_hash),
1643 					    &old_ea_inode);
1644 		if (ret) {
1645 			old_ea_inode = NULL;
1646 			goto out;
1647 		}
1648 	}
1649 	if (i->value && in_inode) {
1650 		WARN_ON_ONCE(!i->value_len);
1651 
1652 		ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1653 		if (ret)
1654 			goto out;
1655 
1656 		ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1657 						     i->value_len,
1658 						     &new_ea_inode);
1659 		if (ret) {
1660 			new_ea_inode = NULL;
1661 			ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1662 			goto out;
1663 		}
1664 	}
1665 
1666 	if (old_ea_inode) {
1667 		/* We are ready to release ref count on the old_ea_inode. */
1668 		ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1669 		if (ret) {
1670 			/* Release newly required ref count on new_ea_inode. */
1671 			if (new_ea_inode) {
1672 				int err;
1673 
1674 				err = ext4_xattr_inode_dec_ref(handle,
1675 							       new_ea_inode);
1676 				if (err)
1677 					ext4_warning_inode(new_ea_inode,
1678 						  "dec ref new_ea_inode err=%d",
1679 						  err);
1680 				ext4_xattr_inode_free_quota(inode, new_ea_inode,
1681 							    i->value_len);
1682 			}
1683 			goto out;
1684 		}
1685 
1686 		ext4_xattr_inode_free_quota(inode, old_ea_inode,
1687 					    le32_to_cpu(here->e_value_size));
1688 	}
1689 
1690 	/* No failures allowed past this point. */
1691 
1692 	if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1693 		/* Remove the old value. */
1694 		void *first_val = s->base + min_offs;
1695 		size_t offs = le16_to_cpu(here->e_value_offs);
1696 		void *val = s->base + offs;
1697 
1698 		memmove(first_val + old_size, first_val, val - first_val);
1699 		memset(first_val, 0, old_size);
1700 		min_offs += old_size;
1701 
1702 		/* Adjust all value offsets. */
1703 		last = s->first;
1704 		while (!IS_LAST_ENTRY(last)) {
1705 			size_t o = le16_to_cpu(last->e_value_offs);
1706 
1707 			if (!last->e_value_inum &&
1708 			    last->e_value_size && o < offs)
1709 				last->e_value_offs = cpu_to_le16(o + old_size);
1710 			last = EXT4_XATTR_NEXT(last);
1711 		}
1712 	}
1713 
1714 	if (!i->value) {
1715 		/* Remove old name. */
1716 		size_t size = EXT4_XATTR_LEN(name_len);
1717 
1718 		last = ENTRY((void *)last - size);
1719 		memmove(here, (void *)here + size,
1720 			(void *)last - (void *)here + sizeof(__u32));
1721 		memset(last, 0, size);
1722 	} else if (s->not_found) {
1723 		/* Insert new name. */
1724 		size_t size = EXT4_XATTR_LEN(name_len);
1725 		size_t rest = (void *)last - (void *)here + sizeof(__u32);
1726 
1727 		memmove((void *)here + size, here, rest);
1728 		memset(here, 0, size);
1729 		here->e_name_index = i->name_index;
1730 		here->e_name_len = name_len;
1731 		memcpy(here->e_name, i->name, name_len);
1732 	} else {
1733 		/* This is an update, reset value info. */
1734 		here->e_value_inum = 0;
1735 		here->e_value_offs = 0;
1736 		here->e_value_size = 0;
1737 	}
1738 
1739 	if (i->value) {
1740 		/* Insert new value. */
1741 		if (in_inode) {
1742 			here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1743 		} else if (i->value_len) {
1744 			void *val = s->base + min_offs - new_size;
1745 
1746 			here->e_value_offs = cpu_to_le16(min_offs - new_size);
1747 			if (i->value == EXT4_ZERO_XATTR_VALUE) {
1748 				memset(val, 0, new_size);
1749 			} else {
1750 				memcpy(val, i->value, i->value_len);
1751 				/* Clear padding bytes. */
1752 				memset(val + i->value_len, 0,
1753 				       new_size - i->value_len);
1754 			}
1755 		}
1756 		here->e_value_size = cpu_to_le32(i->value_len);
1757 	}
1758 
1759 update_hash:
1760 	if (i->value) {
1761 		__le32 hash = 0;
1762 
1763 		/* Entry hash calculation. */
1764 		if (in_inode) {
1765 			__le32 crc32c_hash;
1766 
1767 			/*
1768 			 * Feed crc32c hash instead of the raw value for entry
1769 			 * hash calculation. This is to avoid walking
1770 			 * potentially long value buffer again.
1771 			 */
1772 			crc32c_hash = cpu_to_le32(
1773 				       ext4_xattr_inode_get_hash(new_ea_inode));
1774 			hash = ext4_xattr_hash_entry(here->e_name,
1775 						     here->e_name_len,
1776 						     &crc32c_hash, 1);
1777 		} else if (is_block) {
1778 			__le32 *value = s->base + le16_to_cpu(
1779 							here->e_value_offs);
1780 
1781 			hash = ext4_xattr_hash_entry(here->e_name,
1782 						     here->e_name_len, value,
1783 						     new_size >> 2);
1784 		}
1785 		here->e_hash = hash;
1786 	}
1787 
1788 	if (is_block)
1789 		ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1790 
1791 	ret = 0;
1792 out:
1793 	iput(old_ea_inode);
1794 	iput(new_ea_inode);
1795 	return ret;
1796 }
1797 
1798 struct ext4_xattr_block_find {
1799 	struct ext4_xattr_search s;
1800 	struct buffer_head *bh;
1801 };
1802 
1803 static int
1804 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1805 		      struct ext4_xattr_block_find *bs)
1806 {
1807 	struct super_block *sb = inode->i_sb;
1808 	int error;
1809 
1810 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1811 		  i->name_index, i->name, i->value, (long)i->value_len);
1812 
1813 	if (EXT4_I(inode)->i_file_acl) {
1814 		/* The inode already has an extended attribute block. */
1815 		bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1816 		if (IS_ERR(bs->bh)) {
1817 			error = PTR_ERR(bs->bh);
1818 			bs->bh = NULL;
1819 			return error;
1820 		}
1821 		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1822 			atomic_read(&(bs->bh->b_count)),
1823 			le32_to_cpu(BHDR(bs->bh)->h_refcount));
1824 		error = ext4_xattr_check_block(inode, bs->bh);
1825 		if (error)
1826 			return error;
1827 		/* Find the named attribute. */
1828 		bs->s.base = BHDR(bs->bh);
1829 		bs->s.first = BFIRST(bs->bh);
1830 		bs->s.end = bs->bh->b_data + bs->bh->b_size;
1831 		bs->s.here = bs->s.first;
1832 		error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1833 					 i->name_index, i->name, 1);
1834 		if (error && error != -ENODATA)
1835 			return error;
1836 		bs->s.not_found = error;
1837 	}
1838 	return 0;
1839 }
1840 
1841 static int
1842 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1843 		     struct ext4_xattr_info *i,
1844 		     struct ext4_xattr_block_find *bs)
1845 {
1846 	struct super_block *sb = inode->i_sb;
1847 	struct buffer_head *new_bh = NULL;
1848 	struct ext4_xattr_search s_copy = bs->s;
1849 	struct ext4_xattr_search *s = &s_copy;
1850 	struct mb_cache_entry *ce = NULL;
1851 	int error = 0;
1852 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1853 	struct inode *ea_inode = NULL, *tmp_inode;
1854 	size_t old_ea_inode_quota = 0;
1855 	unsigned int ea_ino;
1856 
1857 
1858 #define header(x) ((struct ext4_xattr_header *)(x))
1859 
1860 	if (s->base) {
1861 		BUFFER_TRACE(bs->bh, "get_write_access");
1862 		error = ext4_journal_get_write_access(handle, sb, bs->bh,
1863 						      EXT4_JTR_NONE);
1864 		if (error)
1865 			goto cleanup;
1866 		lock_buffer(bs->bh);
1867 
1868 		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1869 			__u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1870 
1871 			/*
1872 			 * This must happen under buffer lock for
1873 			 * ext4_xattr_block_set() to reliably detect modified
1874 			 * block
1875 			 */
1876 			if (ea_block_cache)
1877 				mb_cache_entry_delete(ea_block_cache, hash,
1878 						      bs->bh->b_blocknr);
1879 			ea_bdebug(bs->bh, "modifying in-place");
1880 			error = ext4_xattr_set_entry(i, s, handle, inode,
1881 						     true /* is_block */);
1882 			ext4_xattr_block_csum_set(inode, bs->bh);
1883 			unlock_buffer(bs->bh);
1884 			if (error == -EFSCORRUPTED)
1885 				goto bad_block;
1886 			if (!error)
1887 				error = ext4_handle_dirty_metadata(handle,
1888 								   inode,
1889 								   bs->bh);
1890 			if (error)
1891 				goto cleanup;
1892 			goto inserted;
1893 		} else {
1894 			int offset = (char *)s->here - bs->bh->b_data;
1895 
1896 			unlock_buffer(bs->bh);
1897 			ea_bdebug(bs->bh, "cloning");
1898 			s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1899 			error = -ENOMEM;
1900 			if (s->base == NULL)
1901 				goto cleanup;
1902 			s->first = ENTRY(header(s->base)+1);
1903 			header(s->base)->h_refcount = cpu_to_le32(1);
1904 			s->here = ENTRY(s->base + offset);
1905 			s->end = s->base + bs->bh->b_size;
1906 
1907 			/*
1908 			 * If existing entry points to an xattr inode, we need
1909 			 * to prevent ext4_xattr_set_entry() from decrementing
1910 			 * ref count on it because the reference belongs to the
1911 			 * original block. In this case, make the entry look
1912 			 * like it has an empty value.
1913 			 */
1914 			if (!s->not_found && s->here->e_value_inum) {
1915 				ea_ino = le32_to_cpu(s->here->e_value_inum);
1916 				error = ext4_xattr_inode_iget(inode, ea_ino,
1917 					      le32_to_cpu(s->here->e_hash),
1918 					      &tmp_inode);
1919 				if (error)
1920 					goto cleanup;
1921 
1922 				if (!ext4_test_inode_state(tmp_inode,
1923 						EXT4_STATE_LUSTRE_EA_INODE)) {
1924 					/*
1925 					 * Defer quota free call for previous
1926 					 * inode until success is guaranteed.
1927 					 */
1928 					old_ea_inode_quota = le32_to_cpu(
1929 							s->here->e_value_size);
1930 				}
1931 				iput(tmp_inode);
1932 
1933 				s->here->e_value_inum = 0;
1934 				s->here->e_value_size = 0;
1935 			}
1936 		}
1937 	} else {
1938 		/* Allocate a buffer where we construct the new block. */
1939 		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1940 		error = -ENOMEM;
1941 		if (s->base == NULL)
1942 			goto cleanup;
1943 		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1944 		header(s->base)->h_blocks = cpu_to_le32(1);
1945 		header(s->base)->h_refcount = cpu_to_le32(1);
1946 		s->first = ENTRY(header(s->base)+1);
1947 		s->here = ENTRY(header(s->base)+1);
1948 		s->end = s->base + sb->s_blocksize;
1949 	}
1950 
1951 	error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1952 	if (error == -EFSCORRUPTED)
1953 		goto bad_block;
1954 	if (error)
1955 		goto cleanup;
1956 
1957 	if (i->value && s->here->e_value_inum) {
1958 		/*
1959 		 * A ref count on ea_inode has been taken as part of the call to
1960 		 * ext4_xattr_set_entry() above. We would like to drop this
1961 		 * extra ref but we have to wait until the xattr block is
1962 		 * initialized and has its own ref count on the ea_inode.
1963 		 */
1964 		ea_ino = le32_to_cpu(s->here->e_value_inum);
1965 		error = ext4_xattr_inode_iget(inode, ea_ino,
1966 					      le32_to_cpu(s->here->e_hash),
1967 					      &ea_inode);
1968 		if (error) {
1969 			ea_inode = NULL;
1970 			goto cleanup;
1971 		}
1972 	}
1973 
1974 inserted:
1975 	if (!IS_LAST_ENTRY(s->first)) {
1976 		new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1977 						     &ce);
1978 		if (new_bh) {
1979 			/* We found an identical block in the cache. */
1980 			if (new_bh == bs->bh)
1981 				ea_bdebug(new_bh, "keeping");
1982 			else {
1983 				u32 ref;
1984 
1985 				WARN_ON_ONCE(dquot_initialize_needed(inode));
1986 
1987 				/* The old block is released after updating
1988 				   the inode. */
1989 				error = dquot_alloc_block(inode,
1990 						EXT4_C2B(EXT4_SB(sb), 1));
1991 				if (error)
1992 					goto cleanup;
1993 				BUFFER_TRACE(new_bh, "get_write_access");
1994 				error = ext4_journal_get_write_access(
1995 						handle, sb, new_bh,
1996 						EXT4_JTR_NONE);
1997 				if (error)
1998 					goto cleanup_dquot;
1999 				lock_buffer(new_bh);
2000 				/*
2001 				 * We have to be careful about races with
2002 				 * freeing, rehashing or adding references to
2003 				 * xattr block. Once we hold buffer lock xattr
2004 				 * block's state is stable so we can check
2005 				 * whether the block got freed / rehashed or
2006 				 * not.  Since we unhash mbcache entry under
2007 				 * buffer lock when freeing / rehashing xattr
2008 				 * block, checking whether entry is still
2009 				 * hashed is reliable. Same rules hold for
2010 				 * e_reusable handling.
2011 				 */
2012 				if (hlist_bl_unhashed(&ce->e_hash_list) ||
2013 				    !ce->e_reusable) {
2014 					/*
2015 					 * Undo everything and check mbcache
2016 					 * again.
2017 					 */
2018 					unlock_buffer(new_bh);
2019 					dquot_free_block(inode,
2020 							 EXT4_C2B(EXT4_SB(sb),
2021 								  1));
2022 					brelse(new_bh);
2023 					mb_cache_entry_put(ea_block_cache, ce);
2024 					ce = NULL;
2025 					new_bh = NULL;
2026 					goto inserted;
2027 				}
2028 				ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2029 				BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2030 				if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2031 					ce->e_reusable = 0;
2032 				ea_bdebug(new_bh, "reusing; refcount now=%d",
2033 					  ref);
2034 				ext4_xattr_block_csum_set(inode, new_bh);
2035 				unlock_buffer(new_bh);
2036 				error = ext4_handle_dirty_metadata(handle,
2037 								   inode,
2038 								   new_bh);
2039 				if (error)
2040 					goto cleanup_dquot;
2041 			}
2042 			mb_cache_entry_touch(ea_block_cache, ce);
2043 			mb_cache_entry_put(ea_block_cache, ce);
2044 			ce = NULL;
2045 		} else if (bs->bh && s->base == bs->bh->b_data) {
2046 			/* We were modifying this block in-place. */
2047 			ea_bdebug(bs->bh, "keeping this block");
2048 			ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2049 			new_bh = bs->bh;
2050 			get_bh(new_bh);
2051 		} else {
2052 			/* We need to allocate a new block */
2053 			ext4_fsblk_t goal, block;
2054 
2055 			WARN_ON_ONCE(dquot_initialize_needed(inode));
2056 
2057 			goal = ext4_group_first_block_no(sb,
2058 						EXT4_I(inode)->i_block_group);
2059 
2060 			/* non-extent files can't have physical blocks past 2^32 */
2061 			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2062 				goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2063 
2064 			block = ext4_new_meta_blocks(handle, inode, goal, 0,
2065 						     NULL, &error);
2066 			if (error)
2067 				goto cleanup;
2068 
2069 			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2070 				BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2071 
2072 			ea_idebug(inode, "creating block %llu",
2073 				  (unsigned long long)block);
2074 
2075 			new_bh = sb_getblk(sb, block);
2076 			if (unlikely(!new_bh)) {
2077 				error = -ENOMEM;
2078 getblk_failed:
2079 				ext4_free_blocks(handle, inode, NULL, block, 1,
2080 						 EXT4_FREE_BLOCKS_METADATA);
2081 				goto cleanup;
2082 			}
2083 			error = ext4_xattr_inode_inc_ref_all(handle, inode,
2084 						      ENTRY(header(s->base)+1));
2085 			if (error)
2086 				goto getblk_failed;
2087 			if (ea_inode) {
2088 				/* Drop the extra ref on ea_inode. */
2089 				error = ext4_xattr_inode_dec_ref(handle,
2090 								 ea_inode);
2091 				if (error)
2092 					ext4_warning_inode(ea_inode,
2093 							   "dec ref error=%d",
2094 							   error);
2095 				iput(ea_inode);
2096 				ea_inode = NULL;
2097 			}
2098 
2099 			lock_buffer(new_bh);
2100 			error = ext4_journal_get_create_access(handle, sb,
2101 							new_bh, EXT4_JTR_NONE);
2102 			if (error) {
2103 				unlock_buffer(new_bh);
2104 				error = -EIO;
2105 				goto getblk_failed;
2106 			}
2107 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
2108 			ext4_xattr_block_csum_set(inode, new_bh);
2109 			set_buffer_uptodate(new_bh);
2110 			unlock_buffer(new_bh);
2111 			ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2112 			error = ext4_handle_dirty_metadata(handle, inode,
2113 							   new_bh);
2114 			if (error)
2115 				goto cleanup;
2116 		}
2117 	}
2118 
2119 	if (old_ea_inode_quota)
2120 		ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2121 
2122 	/* Update the inode. */
2123 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2124 
2125 	/* Drop the previous xattr block. */
2126 	if (bs->bh && bs->bh != new_bh) {
2127 		struct ext4_xattr_inode_array *ea_inode_array = NULL;
2128 
2129 		ext4_xattr_release_block(handle, inode, bs->bh,
2130 					 &ea_inode_array,
2131 					 0 /* extra_credits */);
2132 		ext4_xattr_inode_array_free(ea_inode_array);
2133 	}
2134 	error = 0;
2135 
2136 cleanup:
2137 	if (ea_inode) {
2138 		int error2;
2139 
2140 		error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2141 		if (error2)
2142 			ext4_warning_inode(ea_inode, "dec ref error=%d",
2143 					   error2);
2144 
2145 		/* If there was an error, revert the quota charge. */
2146 		if (error)
2147 			ext4_xattr_inode_free_quota(inode, ea_inode,
2148 						    i_size_read(ea_inode));
2149 		iput(ea_inode);
2150 	}
2151 	if (ce)
2152 		mb_cache_entry_put(ea_block_cache, ce);
2153 	brelse(new_bh);
2154 	if (!(bs->bh && s->base == bs->bh->b_data))
2155 		kfree(s->base);
2156 
2157 	return error;
2158 
2159 cleanup_dquot:
2160 	dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2161 	goto cleanup;
2162 
2163 bad_block:
2164 	EXT4_ERROR_INODE(inode, "bad block %llu",
2165 			 EXT4_I(inode)->i_file_acl);
2166 	goto cleanup;
2167 
2168 #undef header
2169 }
2170 
2171 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2172 			  struct ext4_xattr_ibody_find *is)
2173 {
2174 	struct ext4_xattr_ibody_header *header;
2175 	struct ext4_inode *raw_inode;
2176 	int error;
2177 
2178 	if (EXT4_I(inode)->i_extra_isize == 0)
2179 		return 0;
2180 	raw_inode = ext4_raw_inode(&is->iloc);
2181 	header = IHDR(inode, raw_inode);
2182 	is->s.base = is->s.first = IFIRST(header);
2183 	is->s.here = is->s.first;
2184 	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2185 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2186 		error = xattr_check_inode(inode, header, is->s.end);
2187 		if (error)
2188 			return error;
2189 		/* Find the named attribute. */
2190 		error = xattr_find_entry(inode, &is->s.here, is->s.end,
2191 					 i->name_index, i->name, 0);
2192 		if (error && error != -ENODATA)
2193 			return error;
2194 		is->s.not_found = error;
2195 	}
2196 	return 0;
2197 }
2198 
2199 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2200 				struct ext4_xattr_info *i,
2201 				struct ext4_xattr_ibody_find *is)
2202 {
2203 	struct ext4_xattr_ibody_header *header;
2204 	struct ext4_xattr_search *s = &is->s;
2205 	int error;
2206 
2207 	if (EXT4_I(inode)->i_extra_isize == 0)
2208 		return -ENOSPC;
2209 	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2210 	if (error)
2211 		return error;
2212 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2213 	if (!IS_LAST_ENTRY(s->first)) {
2214 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2215 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2216 	} else {
2217 		header->h_magic = cpu_to_le32(0);
2218 		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2219 	}
2220 	return 0;
2221 }
2222 
2223 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2224 				 struct ext4_xattr_info *i)
2225 {
2226 	void *value;
2227 
2228 	/* When e_value_inum is set the value is stored externally. */
2229 	if (s->here->e_value_inum)
2230 		return 0;
2231 	if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2232 		return 0;
2233 	value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2234 	return !memcmp(value, i->value, i->value_len);
2235 }
2236 
2237 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2238 {
2239 	struct buffer_head *bh;
2240 	int error;
2241 
2242 	if (!EXT4_I(inode)->i_file_acl)
2243 		return NULL;
2244 	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2245 	if (IS_ERR(bh))
2246 		return bh;
2247 	error = ext4_xattr_check_block(inode, bh);
2248 	if (error) {
2249 		brelse(bh);
2250 		return ERR_PTR(error);
2251 	}
2252 	return bh;
2253 }
2254 
2255 /*
2256  * ext4_xattr_set_handle()
2257  *
2258  * Create, replace or remove an extended attribute for this inode.  Value
2259  * is NULL to remove an existing extended attribute, and non-NULL to
2260  * either replace an existing extended attribute, or create a new extended
2261  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2262  * specify that an extended attribute must exist and must not exist
2263  * previous to the call, respectively.
2264  *
2265  * Returns 0, or a negative error number on failure.
2266  */
2267 int
2268 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2269 		      const char *name, const void *value, size_t value_len,
2270 		      int flags)
2271 {
2272 	struct ext4_xattr_info i = {
2273 		.name_index = name_index,
2274 		.name = name,
2275 		.value = value,
2276 		.value_len = value_len,
2277 		.in_inode = 0,
2278 	};
2279 	struct ext4_xattr_ibody_find is = {
2280 		.s = { .not_found = -ENODATA, },
2281 	};
2282 	struct ext4_xattr_block_find bs = {
2283 		.s = { .not_found = -ENODATA, },
2284 	};
2285 	int no_expand;
2286 	int error;
2287 
2288 	if (!name)
2289 		return -EINVAL;
2290 	if (strlen(name) > 255)
2291 		return -ERANGE;
2292 
2293 	ext4_write_lock_xattr(inode, &no_expand);
2294 
2295 	/* Check journal credits under write lock. */
2296 	if (ext4_handle_valid(handle)) {
2297 		struct buffer_head *bh;
2298 		int credits;
2299 
2300 		bh = ext4_xattr_get_block(inode);
2301 		if (IS_ERR(bh)) {
2302 			error = PTR_ERR(bh);
2303 			goto cleanup;
2304 		}
2305 
2306 		credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2307 						   value_len,
2308 						   flags & XATTR_CREATE);
2309 		brelse(bh);
2310 
2311 		if (jbd2_handle_buffer_credits(handle) < credits) {
2312 			error = -ENOSPC;
2313 			goto cleanup;
2314 		}
2315 		WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2316 	}
2317 
2318 	error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2319 	if (error)
2320 		goto cleanup;
2321 
2322 	if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2323 		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2324 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2325 		ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2326 	}
2327 
2328 	error = ext4_xattr_ibody_find(inode, &i, &is);
2329 	if (error)
2330 		goto cleanup;
2331 	if (is.s.not_found)
2332 		error = ext4_xattr_block_find(inode, &i, &bs);
2333 	if (error)
2334 		goto cleanup;
2335 	if (is.s.not_found && bs.s.not_found) {
2336 		error = -ENODATA;
2337 		if (flags & XATTR_REPLACE)
2338 			goto cleanup;
2339 		error = 0;
2340 		if (!value)
2341 			goto cleanup;
2342 	} else {
2343 		error = -EEXIST;
2344 		if (flags & XATTR_CREATE)
2345 			goto cleanup;
2346 	}
2347 
2348 	if (!value) {
2349 		if (!is.s.not_found)
2350 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2351 		else if (!bs.s.not_found)
2352 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2353 	} else {
2354 		error = 0;
2355 		/* Xattr value did not change? Save us some work and bail out */
2356 		if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2357 			goto cleanup;
2358 		if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2359 			goto cleanup;
2360 
2361 		if (ext4_has_feature_ea_inode(inode->i_sb) &&
2362 		    (EXT4_XATTR_SIZE(i.value_len) >
2363 			EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2364 			i.in_inode = 1;
2365 retry_inode:
2366 		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2367 		if (!error && !bs.s.not_found) {
2368 			i.value = NULL;
2369 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2370 		} else if (error == -ENOSPC) {
2371 			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2372 				brelse(bs.bh);
2373 				bs.bh = NULL;
2374 				error = ext4_xattr_block_find(inode, &i, &bs);
2375 				if (error)
2376 					goto cleanup;
2377 			}
2378 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2379 			if (!error && !is.s.not_found) {
2380 				i.value = NULL;
2381 				error = ext4_xattr_ibody_set(handle, inode, &i,
2382 							     &is);
2383 			} else if (error == -ENOSPC) {
2384 				/*
2385 				 * Xattr does not fit in the block, store at
2386 				 * external inode if possible.
2387 				 */
2388 				if (ext4_has_feature_ea_inode(inode->i_sb) &&
2389 				    i.value_len && !i.in_inode) {
2390 					i.in_inode = 1;
2391 					goto retry_inode;
2392 				}
2393 			}
2394 		}
2395 	}
2396 	if (!error) {
2397 		ext4_xattr_update_super_block(handle, inode->i_sb);
2398 		inode->i_ctime = current_time(inode);
2399 		if (!value)
2400 			no_expand = 0;
2401 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2402 		/*
2403 		 * The bh is consumed by ext4_mark_iloc_dirty, even with
2404 		 * error != 0.
2405 		 */
2406 		is.iloc.bh = NULL;
2407 		if (IS_SYNC(inode))
2408 			ext4_handle_sync(handle);
2409 	}
2410 	ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2411 
2412 cleanup:
2413 	brelse(is.iloc.bh);
2414 	brelse(bs.bh);
2415 	ext4_write_unlock_xattr(inode, &no_expand);
2416 	return error;
2417 }
2418 
2419 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2420 			   bool is_create, int *credits)
2421 {
2422 	struct buffer_head *bh;
2423 	int err;
2424 
2425 	*credits = 0;
2426 
2427 	if (!EXT4_SB(inode->i_sb)->s_journal)
2428 		return 0;
2429 
2430 	down_read(&EXT4_I(inode)->xattr_sem);
2431 
2432 	bh = ext4_xattr_get_block(inode);
2433 	if (IS_ERR(bh)) {
2434 		err = PTR_ERR(bh);
2435 	} else {
2436 		*credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2437 						    value_len, is_create);
2438 		brelse(bh);
2439 		err = 0;
2440 	}
2441 
2442 	up_read(&EXT4_I(inode)->xattr_sem);
2443 	return err;
2444 }
2445 
2446 /*
2447  * ext4_xattr_set()
2448  *
2449  * Like ext4_xattr_set_handle, but start from an inode. This extended
2450  * attribute modification is a filesystem transaction by itself.
2451  *
2452  * Returns 0, or a negative error number on failure.
2453  */
2454 int
2455 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2456 	       const void *value, size_t value_len, int flags)
2457 {
2458 	handle_t *handle;
2459 	struct super_block *sb = inode->i_sb;
2460 	int error, retries = 0;
2461 	int credits;
2462 
2463 	error = dquot_initialize(inode);
2464 	if (error)
2465 		return error;
2466 
2467 retry:
2468 	error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2469 				       &credits);
2470 	if (error)
2471 		return error;
2472 
2473 	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2474 	if (IS_ERR(handle)) {
2475 		error = PTR_ERR(handle);
2476 	} else {
2477 		int error2;
2478 
2479 		error = ext4_xattr_set_handle(handle, inode, name_index, name,
2480 					      value, value_len, flags);
2481 		error2 = ext4_journal_stop(handle);
2482 		if (error == -ENOSPC &&
2483 		    ext4_should_retry_alloc(sb, &retries))
2484 			goto retry;
2485 		if (error == 0)
2486 			error = error2;
2487 	}
2488 	ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2489 
2490 	return error;
2491 }
2492 
2493 /*
2494  * Shift the EA entries in the inode to create space for the increased
2495  * i_extra_isize.
2496  */
2497 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2498 				     int value_offs_shift, void *to,
2499 				     void *from, size_t n)
2500 {
2501 	struct ext4_xattr_entry *last = entry;
2502 	int new_offs;
2503 
2504 	/* We always shift xattr headers further thus offsets get lower */
2505 	BUG_ON(value_offs_shift > 0);
2506 
2507 	/* Adjust the value offsets of the entries */
2508 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2509 		if (!last->e_value_inum && last->e_value_size) {
2510 			new_offs = le16_to_cpu(last->e_value_offs) +
2511 							value_offs_shift;
2512 			last->e_value_offs = cpu_to_le16(new_offs);
2513 		}
2514 	}
2515 	/* Shift the entries by n bytes */
2516 	memmove(to, from, n);
2517 }
2518 
2519 /*
2520  * Move xattr pointed to by 'entry' from inode into external xattr block
2521  */
2522 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2523 				    struct ext4_inode *raw_inode,
2524 				    struct ext4_xattr_entry *entry)
2525 {
2526 	struct ext4_xattr_ibody_find *is = NULL;
2527 	struct ext4_xattr_block_find *bs = NULL;
2528 	char *buffer = NULL, *b_entry_name = NULL;
2529 	size_t value_size = le32_to_cpu(entry->e_value_size);
2530 	struct ext4_xattr_info i = {
2531 		.value = NULL,
2532 		.value_len = 0,
2533 		.name_index = entry->e_name_index,
2534 		.in_inode = !!entry->e_value_inum,
2535 	};
2536 	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2537 	int error;
2538 
2539 	is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2540 	bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2541 	buffer = kmalloc(value_size, GFP_NOFS);
2542 	b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2543 	if (!is || !bs || !buffer || !b_entry_name) {
2544 		error = -ENOMEM;
2545 		goto out;
2546 	}
2547 
2548 	is->s.not_found = -ENODATA;
2549 	bs->s.not_found = -ENODATA;
2550 	is->iloc.bh = NULL;
2551 	bs->bh = NULL;
2552 
2553 	/* Save the entry name and the entry value */
2554 	if (entry->e_value_inum) {
2555 		error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2556 		if (error)
2557 			goto out;
2558 	} else {
2559 		size_t value_offs = le16_to_cpu(entry->e_value_offs);
2560 		memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2561 	}
2562 
2563 	memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2564 	b_entry_name[entry->e_name_len] = '\0';
2565 	i.name = b_entry_name;
2566 
2567 	error = ext4_get_inode_loc(inode, &is->iloc);
2568 	if (error)
2569 		goto out;
2570 
2571 	error = ext4_xattr_ibody_find(inode, &i, is);
2572 	if (error)
2573 		goto out;
2574 
2575 	/* Remove the chosen entry from the inode */
2576 	error = ext4_xattr_ibody_set(handle, inode, &i, is);
2577 	if (error)
2578 		goto out;
2579 
2580 	i.value = buffer;
2581 	i.value_len = value_size;
2582 	error = ext4_xattr_block_find(inode, &i, bs);
2583 	if (error)
2584 		goto out;
2585 
2586 	/* Add entry which was removed from the inode into the block */
2587 	error = ext4_xattr_block_set(handle, inode, &i, bs);
2588 	if (error)
2589 		goto out;
2590 	error = 0;
2591 out:
2592 	kfree(b_entry_name);
2593 	kfree(buffer);
2594 	if (is)
2595 		brelse(is->iloc.bh);
2596 	if (bs)
2597 		brelse(bs->bh);
2598 	kfree(is);
2599 	kfree(bs);
2600 
2601 	return error;
2602 }
2603 
2604 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2605 				       struct ext4_inode *raw_inode,
2606 				       int isize_diff, size_t ifree,
2607 				       size_t bfree, int *total_ino)
2608 {
2609 	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2610 	struct ext4_xattr_entry *small_entry;
2611 	struct ext4_xattr_entry *entry;
2612 	struct ext4_xattr_entry *last;
2613 	unsigned int entry_size;	/* EA entry size */
2614 	unsigned int total_size;	/* EA entry size + value size */
2615 	unsigned int min_total_size;
2616 	int error;
2617 
2618 	while (isize_diff > ifree) {
2619 		entry = NULL;
2620 		small_entry = NULL;
2621 		min_total_size = ~0U;
2622 		last = IFIRST(header);
2623 		/* Find the entry best suited to be pushed into EA block */
2624 		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2625 			/* never move system.data out of the inode */
2626 			if ((last->e_name_len == 4) &&
2627 			    (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2628 			    !memcmp(last->e_name, "data", 4))
2629 				continue;
2630 			total_size = EXT4_XATTR_LEN(last->e_name_len);
2631 			if (!last->e_value_inum)
2632 				total_size += EXT4_XATTR_SIZE(
2633 					       le32_to_cpu(last->e_value_size));
2634 			if (total_size <= bfree &&
2635 			    total_size < min_total_size) {
2636 				if (total_size + ifree < isize_diff) {
2637 					small_entry = last;
2638 				} else {
2639 					entry = last;
2640 					min_total_size = total_size;
2641 				}
2642 			}
2643 		}
2644 
2645 		if (entry == NULL) {
2646 			if (small_entry == NULL)
2647 				return -ENOSPC;
2648 			entry = small_entry;
2649 		}
2650 
2651 		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2652 		total_size = entry_size;
2653 		if (!entry->e_value_inum)
2654 			total_size += EXT4_XATTR_SIZE(
2655 					      le32_to_cpu(entry->e_value_size));
2656 		error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2657 						 entry);
2658 		if (error)
2659 			return error;
2660 
2661 		*total_ino -= entry_size;
2662 		ifree += total_size;
2663 		bfree -= total_size;
2664 	}
2665 
2666 	return 0;
2667 }
2668 
2669 /*
2670  * Expand an inode by new_extra_isize bytes when EAs are present.
2671  * Returns 0 on success or negative error number on failure.
2672  */
2673 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2674 			       struct ext4_inode *raw_inode, handle_t *handle)
2675 {
2676 	struct ext4_xattr_ibody_header *header;
2677 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2678 	static unsigned int mnt_count;
2679 	size_t min_offs;
2680 	size_t ifree, bfree;
2681 	int total_ino;
2682 	void *base, *end;
2683 	int error = 0, tried_min_extra_isize = 0;
2684 	int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2685 	int isize_diff;	/* How much do we need to grow i_extra_isize */
2686 
2687 retry:
2688 	isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2689 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2690 		return 0;
2691 
2692 	header = IHDR(inode, raw_inode);
2693 
2694 	/*
2695 	 * Check if enough free space is available in the inode to shift the
2696 	 * entries ahead by new_extra_isize.
2697 	 */
2698 
2699 	base = IFIRST(header);
2700 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2701 	min_offs = end - base;
2702 	total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2703 
2704 	error = xattr_check_inode(inode, header, end);
2705 	if (error)
2706 		goto cleanup;
2707 
2708 	ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2709 	if (ifree >= isize_diff)
2710 		goto shift;
2711 
2712 	/*
2713 	 * Enough free space isn't available in the inode, check if
2714 	 * EA block can hold new_extra_isize bytes.
2715 	 */
2716 	if (EXT4_I(inode)->i_file_acl) {
2717 		struct buffer_head *bh;
2718 
2719 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2720 		if (IS_ERR(bh)) {
2721 			error = PTR_ERR(bh);
2722 			goto cleanup;
2723 		}
2724 		error = ext4_xattr_check_block(inode, bh);
2725 		if (error) {
2726 			brelse(bh);
2727 			goto cleanup;
2728 		}
2729 		base = BHDR(bh);
2730 		end = bh->b_data + bh->b_size;
2731 		min_offs = end - base;
2732 		bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2733 					      NULL);
2734 		brelse(bh);
2735 		if (bfree + ifree < isize_diff) {
2736 			if (!tried_min_extra_isize && s_min_extra_isize) {
2737 				tried_min_extra_isize++;
2738 				new_extra_isize = s_min_extra_isize;
2739 				goto retry;
2740 			}
2741 			error = -ENOSPC;
2742 			goto cleanup;
2743 		}
2744 	} else {
2745 		bfree = inode->i_sb->s_blocksize;
2746 	}
2747 
2748 	error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2749 					    isize_diff, ifree, bfree,
2750 					    &total_ino);
2751 	if (error) {
2752 		if (error == -ENOSPC && !tried_min_extra_isize &&
2753 		    s_min_extra_isize) {
2754 			tried_min_extra_isize++;
2755 			new_extra_isize = s_min_extra_isize;
2756 			goto retry;
2757 		}
2758 		goto cleanup;
2759 	}
2760 shift:
2761 	/* Adjust the offsets and shift the remaining entries ahead */
2762 	ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2763 			- new_extra_isize, (void *)raw_inode +
2764 			EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2765 			(void *)header, total_ino);
2766 	EXT4_I(inode)->i_extra_isize = new_extra_isize;
2767 
2768 cleanup:
2769 	if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2770 		ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2771 			     inode->i_ino);
2772 		mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2773 	}
2774 	return error;
2775 }
2776 
2777 #define EIA_INCR 16 /* must be 2^n */
2778 #define EIA_MASK (EIA_INCR - 1)
2779 
2780 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2781  * If @ea_inode_array is new or full it will be grown and the old
2782  * contents copied over.
2783  */
2784 static int
2785 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2786 			struct inode *inode)
2787 {
2788 	if (*ea_inode_array == NULL) {
2789 		/*
2790 		 * Start with 15 inodes, so it fits into a power-of-two size.
2791 		 * If *ea_inode_array is NULL, this is essentially offsetof()
2792 		 */
2793 		(*ea_inode_array) =
2794 			kmalloc(offsetof(struct ext4_xattr_inode_array,
2795 					 inodes[EIA_MASK]),
2796 				GFP_NOFS);
2797 		if (*ea_inode_array == NULL)
2798 			return -ENOMEM;
2799 		(*ea_inode_array)->count = 0;
2800 	} else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2801 		/* expand the array once all 15 + n * 16 slots are full */
2802 		struct ext4_xattr_inode_array *new_array = NULL;
2803 		int count = (*ea_inode_array)->count;
2804 
2805 		/* if new_array is NULL, this is essentially offsetof() */
2806 		new_array = kmalloc(
2807 				offsetof(struct ext4_xattr_inode_array,
2808 					 inodes[count + EIA_INCR]),
2809 				GFP_NOFS);
2810 		if (new_array == NULL)
2811 			return -ENOMEM;
2812 		memcpy(new_array, *ea_inode_array,
2813 		       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2814 		kfree(*ea_inode_array);
2815 		*ea_inode_array = new_array;
2816 	}
2817 	(*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2818 	return 0;
2819 }
2820 
2821 /*
2822  * ext4_xattr_delete_inode()
2823  *
2824  * Free extended attribute resources associated with this inode. Traverse
2825  * all entries and decrement reference on any xattr inodes associated with this
2826  * inode. This is called immediately before an inode is freed. We have exclusive
2827  * access to the inode. If an orphan inode is deleted it will also release its
2828  * references on xattr block and xattr inodes.
2829  */
2830 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2831 			    struct ext4_xattr_inode_array **ea_inode_array,
2832 			    int extra_credits)
2833 {
2834 	struct buffer_head *bh = NULL;
2835 	struct ext4_xattr_ibody_header *header;
2836 	struct ext4_iloc iloc = { .bh = NULL };
2837 	struct ext4_xattr_entry *entry;
2838 	struct inode *ea_inode;
2839 	int error;
2840 
2841 	error = ext4_journal_ensure_credits(handle, extra_credits,
2842 			ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2843 	if (error < 0) {
2844 		EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2845 		goto cleanup;
2846 	}
2847 
2848 	if (ext4_has_feature_ea_inode(inode->i_sb) &&
2849 	    ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2850 
2851 		error = ext4_get_inode_loc(inode, &iloc);
2852 		if (error) {
2853 			EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2854 			goto cleanup;
2855 		}
2856 
2857 		error = ext4_journal_get_write_access(handle, inode->i_sb,
2858 						iloc.bh, EXT4_JTR_NONE);
2859 		if (error) {
2860 			EXT4_ERROR_INODE(inode, "write access (error %d)",
2861 					 error);
2862 			goto cleanup;
2863 		}
2864 
2865 		header = IHDR(inode, ext4_raw_inode(&iloc));
2866 		if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2867 			ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2868 						     IFIRST(header),
2869 						     false /* block_csum */,
2870 						     ea_inode_array,
2871 						     extra_credits,
2872 						     false /* skip_quota */);
2873 	}
2874 
2875 	if (EXT4_I(inode)->i_file_acl) {
2876 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2877 		if (IS_ERR(bh)) {
2878 			error = PTR_ERR(bh);
2879 			if (error == -EIO) {
2880 				EXT4_ERROR_INODE_ERR(inode, EIO,
2881 						     "block %llu read error",
2882 						     EXT4_I(inode)->i_file_acl);
2883 			}
2884 			bh = NULL;
2885 			goto cleanup;
2886 		}
2887 		error = ext4_xattr_check_block(inode, bh);
2888 		if (error)
2889 			goto cleanup;
2890 
2891 		if (ext4_has_feature_ea_inode(inode->i_sb)) {
2892 			for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2893 			     entry = EXT4_XATTR_NEXT(entry)) {
2894 				if (!entry->e_value_inum)
2895 					continue;
2896 				error = ext4_xattr_inode_iget(inode,
2897 					      le32_to_cpu(entry->e_value_inum),
2898 					      le32_to_cpu(entry->e_hash),
2899 					      &ea_inode);
2900 				if (error)
2901 					continue;
2902 				ext4_xattr_inode_free_quota(inode, ea_inode,
2903 					      le32_to_cpu(entry->e_value_size));
2904 				iput(ea_inode);
2905 			}
2906 
2907 		}
2908 
2909 		ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2910 					 extra_credits);
2911 		/*
2912 		 * Update i_file_acl value in the same transaction that releases
2913 		 * block.
2914 		 */
2915 		EXT4_I(inode)->i_file_acl = 0;
2916 		error = ext4_mark_inode_dirty(handle, inode);
2917 		if (error) {
2918 			EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2919 					 error);
2920 			goto cleanup;
2921 		}
2922 		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2923 	}
2924 	error = 0;
2925 cleanup:
2926 	brelse(iloc.bh);
2927 	brelse(bh);
2928 	return error;
2929 }
2930 
2931 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2932 {
2933 	int idx;
2934 
2935 	if (ea_inode_array == NULL)
2936 		return;
2937 
2938 	for (idx = 0; idx < ea_inode_array->count; ++idx)
2939 		iput(ea_inode_array->inodes[idx]);
2940 	kfree(ea_inode_array);
2941 }
2942 
2943 /*
2944  * ext4_xattr_block_cache_insert()
2945  *
2946  * Create a new entry in the extended attribute block cache, and insert
2947  * it unless such an entry is already in the cache.
2948  *
2949  * Returns 0, or a negative error number on failure.
2950  */
2951 static void
2952 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2953 			      struct buffer_head *bh)
2954 {
2955 	struct ext4_xattr_header *header = BHDR(bh);
2956 	__u32 hash = le32_to_cpu(header->h_hash);
2957 	int reusable = le32_to_cpu(header->h_refcount) <
2958 		       EXT4_XATTR_REFCOUNT_MAX;
2959 	int error;
2960 
2961 	if (!ea_block_cache)
2962 		return;
2963 	error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2964 				      bh->b_blocknr, reusable);
2965 	if (error) {
2966 		if (error == -EBUSY)
2967 			ea_bdebug(bh, "already in cache");
2968 	} else
2969 		ea_bdebug(bh, "inserting [%x]", (int)hash);
2970 }
2971 
2972 /*
2973  * ext4_xattr_cmp()
2974  *
2975  * Compare two extended attribute blocks for equality.
2976  *
2977  * Returns 0 if the blocks are equal, 1 if they differ, and
2978  * a negative error number on errors.
2979  */
2980 static int
2981 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2982 	       struct ext4_xattr_header *header2)
2983 {
2984 	struct ext4_xattr_entry *entry1, *entry2;
2985 
2986 	entry1 = ENTRY(header1+1);
2987 	entry2 = ENTRY(header2+1);
2988 	while (!IS_LAST_ENTRY(entry1)) {
2989 		if (IS_LAST_ENTRY(entry2))
2990 			return 1;
2991 		if (entry1->e_hash != entry2->e_hash ||
2992 		    entry1->e_name_index != entry2->e_name_index ||
2993 		    entry1->e_name_len != entry2->e_name_len ||
2994 		    entry1->e_value_size != entry2->e_value_size ||
2995 		    entry1->e_value_inum != entry2->e_value_inum ||
2996 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
2997 			return 1;
2998 		if (!entry1->e_value_inum &&
2999 		    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3000 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3001 			   le32_to_cpu(entry1->e_value_size)))
3002 			return 1;
3003 
3004 		entry1 = EXT4_XATTR_NEXT(entry1);
3005 		entry2 = EXT4_XATTR_NEXT(entry2);
3006 	}
3007 	if (!IS_LAST_ENTRY(entry2))
3008 		return 1;
3009 	return 0;
3010 }
3011 
3012 /*
3013  * ext4_xattr_block_cache_find()
3014  *
3015  * Find an identical extended attribute block.
3016  *
3017  * Returns a pointer to the block found, or NULL if such a block was
3018  * not found or an error occurred.
3019  */
3020 static struct buffer_head *
3021 ext4_xattr_block_cache_find(struct inode *inode,
3022 			    struct ext4_xattr_header *header,
3023 			    struct mb_cache_entry **pce)
3024 {
3025 	__u32 hash = le32_to_cpu(header->h_hash);
3026 	struct mb_cache_entry *ce;
3027 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3028 
3029 	if (!ea_block_cache)
3030 		return NULL;
3031 	if (!header->h_hash)
3032 		return NULL;  /* never share */
3033 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3034 	ce = mb_cache_entry_find_first(ea_block_cache, hash);
3035 	while (ce) {
3036 		struct buffer_head *bh;
3037 
3038 		bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3039 		if (IS_ERR(bh)) {
3040 			if (PTR_ERR(bh) == -ENOMEM)
3041 				return NULL;
3042 			bh = NULL;
3043 			EXT4_ERROR_INODE(inode, "block %lu read error",
3044 					 (unsigned long)ce->e_value);
3045 		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3046 			*pce = ce;
3047 			return bh;
3048 		}
3049 		brelse(bh);
3050 		ce = mb_cache_entry_find_next(ea_block_cache, ce);
3051 	}
3052 	return NULL;
3053 }
3054 
3055 #define NAME_HASH_SHIFT 5
3056 #define VALUE_HASH_SHIFT 16
3057 
3058 /*
3059  * ext4_xattr_hash_entry()
3060  *
3061  * Compute the hash of an extended attribute.
3062  */
3063 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3064 				    size_t value_count)
3065 {
3066 	__u32 hash = 0;
3067 
3068 	while (name_len--) {
3069 		hash = (hash << NAME_HASH_SHIFT) ^
3070 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3071 		       *name++;
3072 	}
3073 	while (value_count--) {
3074 		hash = (hash << VALUE_HASH_SHIFT) ^
3075 		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3076 		       le32_to_cpu(*value++);
3077 	}
3078 	return cpu_to_le32(hash);
3079 }
3080 
3081 #undef NAME_HASH_SHIFT
3082 #undef VALUE_HASH_SHIFT
3083 
3084 #define BLOCK_HASH_SHIFT 16
3085 
3086 /*
3087  * ext4_xattr_rehash()
3088  *
3089  * Re-compute the extended attribute hash value after an entry has changed.
3090  */
3091 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3092 {
3093 	struct ext4_xattr_entry *here;
3094 	__u32 hash = 0;
3095 
3096 	here = ENTRY(header+1);
3097 	while (!IS_LAST_ENTRY(here)) {
3098 		if (!here->e_hash) {
3099 			/* Block is not shared if an entry's hash value == 0 */
3100 			hash = 0;
3101 			break;
3102 		}
3103 		hash = (hash << BLOCK_HASH_SHIFT) ^
3104 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3105 		       le32_to_cpu(here->e_hash);
3106 		here = EXT4_XATTR_NEXT(here);
3107 	}
3108 	header->h_hash = cpu_to_le32(hash);
3109 }
3110 
3111 #undef BLOCK_HASH_SHIFT
3112 
3113 #define	HASH_BUCKET_BITS	10
3114 
3115 struct mb_cache *
3116 ext4_xattr_create_cache(void)
3117 {
3118 	return mb_cache_create(HASH_BUCKET_BITS);
3119 }
3120 
3121 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3122 {
3123 	if (cache)
3124 		mb_cache_destroy(cache);
3125 }
3126 
3127