xref: /openbmc/linux/fs/ext4/xattr.c (revision 4bce6fce)
1 /*
2  * linux/fs/ext4/xattr.c
3  *
4  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
5  *
6  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
7  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
8  * Extended attributes for symlinks and special files added per
9  *  suggestion of Luka Renko <luka.renko@hermes.si>.
10  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
11  *  Red Hat Inc.
12  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
13  *  and Andreas Gruenbacher <agruen@suse.de>.
14  */
15 
16 /*
17  * Extended attributes are stored directly in inodes (on file systems with
18  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
19  * field contains the block number if an inode uses an additional block. All
20  * attributes must fit in the inode and one additional block. Blocks that
21  * contain the identical set of attributes may be shared among several inodes.
22  * Identical blocks are detected by keeping a cache of blocks that have
23  * recently been accessed.
24  *
25  * The attributes in inodes and on blocks have a different header; the entries
26  * are stored in the same format:
27  *
28  *   +------------------+
29  *   | header           |
30  *   | entry 1          | |
31  *   | entry 2          | | growing downwards
32  *   | entry 3          | v
33  *   | four null bytes  |
34  *   | . . .            |
35  *   | value 1          | ^
36  *   | value 3          | | growing upwards
37  *   | value 2          | |
38  *   +------------------+
39  *
40  * The header is followed by multiple entry descriptors. In disk blocks, the
41  * entry descriptors are kept sorted. In inodes, they are unsorted. The
42  * attribute values are aligned to the end of the block in no specific order.
43  *
44  * Locking strategy
45  * ----------------
46  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
47  * EA blocks are only changed if they are exclusive to an inode, so
48  * holding xattr_sem also means that nothing but the EA block's reference
49  * count can change. Multiple writers to the same block are synchronized
50  * by the buffer lock.
51  */
52 
53 #include <linux/init.h>
54 #include <linux/fs.h>
55 #include <linux/slab.h>
56 #include <linux/mbcache.h>
57 #include <linux/quotaops.h>
58 #include "ext4_jbd2.h"
59 #include "ext4.h"
60 #include "xattr.h"
61 #include "acl.h"
62 
63 #ifdef EXT4_XATTR_DEBUG
64 # define ea_idebug(inode, f...) do { \
65 		printk(KERN_DEBUG "inode %s:%lu: ", \
66 			inode->i_sb->s_id, inode->i_ino); \
67 		printk(f); \
68 		printk("\n"); \
69 	} while (0)
70 # define ea_bdebug(bh, f...) do { \
71 		char b[BDEVNAME_SIZE]; \
72 		printk(KERN_DEBUG "block %s:%lu: ", \
73 			bdevname(bh->b_bdev, b), \
74 			(unsigned long) bh->b_blocknr); \
75 		printk(f); \
76 		printk("\n"); \
77 	} while (0)
78 #else
79 # define ea_idebug(inode, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
80 # define ea_bdebug(bh, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
81 #endif
82 
83 static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
84 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
85 						 struct ext4_xattr_header *,
86 						 struct mb_cache_entry **);
87 static void ext4_xattr_rehash(struct ext4_xattr_header *,
88 			      struct ext4_xattr_entry *);
89 static int ext4_xattr_list(struct dentry *dentry, char *buffer,
90 			   size_t buffer_size);
91 
92 static const struct xattr_handler *ext4_xattr_handler_map[] = {
93 	[EXT4_XATTR_INDEX_USER]		     = &ext4_xattr_user_handler,
94 #ifdef CONFIG_EXT4_FS_POSIX_ACL
95 	[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
96 	[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
97 #endif
98 	[EXT4_XATTR_INDEX_TRUSTED]	     = &ext4_xattr_trusted_handler,
99 #ifdef CONFIG_EXT4_FS_SECURITY
100 	[EXT4_XATTR_INDEX_SECURITY]	     = &ext4_xattr_security_handler,
101 #endif
102 };
103 
104 const struct xattr_handler *ext4_xattr_handlers[] = {
105 	&ext4_xattr_user_handler,
106 	&ext4_xattr_trusted_handler,
107 #ifdef CONFIG_EXT4_FS_POSIX_ACL
108 	&posix_acl_access_xattr_handler,
109 	&posix_acl_default_xattr_handler,
110 #endif
111 #ifdef CONFIG_EXT4_FS_SECURITY
112 	&ext4_xattr_security_handler,
113 #endif
114 	NULL
115 };
116 
117 #define EXT4_GET_MB_CACHE(inode)	(((struct ext4_sb_info *) \
118 				inode->i_sb->s_fs_info)->s_mb_cache)
119 
120 static __le32 ext4_xattr_block_csum(struct inode *inode,
121 				    sector_t block_nr,
122 				    struct ext4_xattr_header *hdr)
123 {
124 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
125 	__u32 csum;
126 	__le32 save_csum;
127 	__le64 dsk_block_nr = cpu_to_le64(block_nr);
128 
129 	save_csum = hdr->h_checksum;
130 	hdr->h_checksum = 0;
131 	csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
132 			   sizeof(dsk_block_nr));
133 	csum = ext4_chksum(sbi, csum, (__u8 *)hdr,
134 			   EXT4_BLOCK_SIZE(inode->i_sb));
135 
136 	hdr->h_checksum = save_csum;
137 	return cpu_to_le32(csum);
138 }
139 
140 static int ext4_xattr_block_csum_verify(struct inode *inode,
141 					sector_t block_nr,
142 					struct ext4_xattr_header *hdr)
143 {
144 	if (ext4_has_metadata_csum(inode->i_sb) &&
145 	    (hdr->h_checksum != ext4_xattr_block_csum(inode, block_nr, hdr)))
146 		return 0;
147 	return 1;
148 }
149 
150 static void ext4_xattr_block_csum_set(struct inode *inode,
151 				      sector_t block_nr,
152 				      struct ext4_xattr_header *hdr)
153 {
154 	if (!ext4_has_metadata_csum(inode->i_sb))
155 		return;
156 
157 	hdr->h_checksum = ext4_xattr_block_csum(inode, block_nr, hdr);
158 }
159 
160 static inline int ext4_handle_dirty_xattr_block(handle_t *handle,
161 						struct inode *inode,
162 						struct buffer_head *bh)
163 {
164 	ext4_xattr_block_csum_set(inode, bh->b_blocknr, BHDR(bh));
165 	return ext4_handle_dirty_metadata(handle, inode, bh);
166 }
167 
168 static inline const struct xattr_handler *
169 ext4_xattr_handler(int name_index)
170 {
171 	const struct xattr_handler *handler = NULL;
172 
173 	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
174 		handler = ext4_xattr_handler_map[name_index];
175 	return handler;
176 }
177 
178 /*
179  * Inode operation listxattr()
180  *
181  * d_inode(dentry)->i_mutex: don't care
182  */
183 ssize_t
184 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
185 {
186 	return ext4_xattr_list(dentry, buffer, size);
187 }
188 
189 static int
190 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
191 		       void *value_start)
192 {
193 	struct ext4_xattr_entry *e = entry;
194 
195 	while (!IS_LAST_ENTRY(e)) {
196 		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
197 		if ((void *)next >= end)
198 			return -EIO;
199 		e = next;
200 	}
201 
202 	while (!IS_LAST_ENTRY(entry)) {
203 		if (entry->e_value_size != 0 &&
204 		    (value_start + le16_to_cpu(entry->e_value_offs) <
205 		     (void *)e + sizeof(__u32) ||
206 		     value_start + le16_to_cpu(entry->e_value_offs) +
207 		    le32_to_cpu(entry->e_value_size) > end))
208 			return -EIO;
209 		entry = EXT4_XATTR_NEXT(entry);
210 	}
211 
212 	return 0;
213 }
214 
215 static inline int
216 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
217 {
218 	int error;
219 
220 	if (buffer_verified(bh))
221 		return 0;
222 
223 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
224 	    BHDR(bh)->h_blocks != cpu_to_le32(1))
225 		return -EIO;
226 	if (!ext4_xattr_block_csum_verify(inode, bh->b_blocknr, BHDR(bh)))
227 		return -EIO;
228 	error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size,
229 				       bh->b_data);
230 	if (!error)
231 		set_buffer_verified(bh);
232 	return error;
233 }
234 
235 static inline int
236 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
237 {
238 	size_t value_size = le32_to_cpu(entry->e_value_size);
239 
240 	if (entry->e_value_block != 0 || value_size > size ||
241 	    le16_to_cpu(entry->e_value_offs) + value_size > size)
242 		return -EIO;
243 	return 0;
244 }
245 
246 static int
247 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
248 		      const char *name, size_t size, int sorted)
249 {
250 	struct ext4_xattr_entry *entry;
251 	size_t name_len;
252 	int cmp = 1;
253 
254 	if (name == NULL)
255 		return -EINVAL;
256 	name_len = strlen(name);
257 	entry = *pentry;
258 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
259 		cmp = name_index - entry->e_name_index;
260 		if (!cmp)
261 			cmp = name_len - entry->e_name_len;
262 		if (!cmp)
263 			cmp = memcmp(name, entry->e_name, name_len);
264 		if (cmp <= 0 && (sorted || cmp == 0))
265 			break;
266 	}
267 	*pentry = entry;
268 	if (!cmp && ext4_xattr_check_entry(entry, size))
269 			return -EIO;
270 	return cmp ? -ENODATA : 0;
271 }
272 
273 static int
274 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
275 		     void *buffer, size_t buffer_size)
276 {
277 	struct buffer_head *bh = NULL;
278 	struct ext4_xattr_entry *entry;
279 	size_t size;
280 	int error;
281 	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
282 
283 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
284 		  name_index, name, buffer, (long)buffer_size);
285 
286 	error = -ENODATA;
287 	if (!EXT4_I(inode)->i_file_acl)
288 		goto cleanup;
289 	ea_idebug(inode, "reading block %llu",
290 		  (unsigned long long)EXT4_I(inode)->i_file_acl);
291 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
292 	if (!bh)
293 		goto cleanup;
294 	ea_bdebug(bh, "b_count=%d, refcount=%d",
295 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
296 	if (ext4_xattr_check_block(inode, bh)) {
297 bad_block:
298 		EXT4_ERROR_INODE(inode, "bad block %llu",
299 				 EXT4_I(inode)->i_file_acl);
300 		error = -EIO;
301 		goto cleanup;
302 	}
303 	ext4_xattr_cache_insert(ext4_mb_cache, bh);
304 	entry = BFIRST(bh);
305 	error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
306 	if (error == -EIO)
307 		goto bad_block;
308 	if (error)
309 		goto cleanup;
310 	size = le32_to_cpu(entry->e_value_size);
311 	if (buffer) {
312 		error = -ERANGE;
313 		if (size > buffer_size)
314 			goto cleanup;
315 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
316 		       size);
317 	}
318 	error = size;
319 
320 cleanup:
321 	brelse(bh);
322 	return error;
323 }
324 
325 int
326 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
327 		     void *buffer, size_t buffer_size)
328 {
329 	struct ext4_xattr_ibody_header *header;
330 	struct ext4_xattr_entry *entry;
331 	struct ext4_inode *raw_inode;
332 	struct ext4_iloc iloc;
333 	size_t size;
334 	void *end;
335 	int error;
336 
337 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
338 		return -ENODATA;
339 	error = ext4_get_inode_loc(inode, &iloc);
340 	if (error)
341 		return error;
342 	raw_inode = ext4_raw_inode(&iloc);
343 	header = IHDR(inode, raw_inode);
344 	entry = IFIRST(header);
345 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
346 	error = ext4_xattr_check_names(entry, end, entry);
347 	if (error)
348 		goto cleanup;
349 	error = ext4_xattr_find_entry(&entry, name_index, name,
350 				      end - (void *)entry, 0);
351 	if (error)
352 		goto cleanup;
353 	size = le32_to_cpu(entry->e_value_size);
354 	if (buffer) {
355 		error = -ERANGE;
356 		if (size > buffer_size)
357 			goto cleanup;
358 		memcpy(buffer, (void *)IFIRST(header) +
359 		       le16_to_cpu(entry->e_value_offs), size);
360 	}
361 	error = size;
362 
363 cleanup:
364 	brelse(iloc.bh);
365 	return error;
366 }
367 
368 /*
369  * ext4_xattr_get()
370  *
371  * Copy an extended attribute into the buffer
372  * provided, or compute the buffer size required.
373  * Buffer is NULL to compute the size of the buffer required.
374  *
375  * Returns a negative error number on failure, or the number of bytes
376  * used / required on success.
377  */
378 int
379 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
380 	       void *buffer, size_t buffer_size)
381 {
382 	int error;
383 
384 	if (strlen(name) > 255)
385 		return -ERANGE;
386 
387 	down_read(&EXT4_I(inode)->xattr_sem);
388 	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
389 				     buffer_size);
390 	if (error == -ENODATA)
391 		error = ext4_xattr_block_get(inode, name_index, name, buffer,
392 					     buffer_size);
393 	up_read(&EXT4_I(inode)->xattr_sem);
394 	return error;
395 }
396 
397 static int
398 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
399 			char *buffer, size_t buffer_size)
400 {
401 	size_t rest = buffer_size;
402 
403 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
404 		const struct xattr_handler *handler =
405 			ext4_xattr_handler(entry->e_name_index);
406 
407 		if (handler) {
408 			size_t size = handler->list(dentry, buffer, rest,
409 						    entry->e_name,
410 						    entry->e_name_len,
411 						    handler->flags);
412 			if (buffer) {
413 				if (size > rest)
414 					return -ERANGE;
415 				buffer += size;
416 			}
417 			rest -= size;
418 		}
419 	}
420 	return buffer_size - rest;
421 }
422 
423 static int
424 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
425 {
426 	struct inode *inode = d_inode(dentry);
427 	struct buffer_head *bh = NULL;
428 	int error;
429 	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
430 
431 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
432 		  buffer, (long)buffer_size);
433 
434 	error = 0;
435 	if (!EXT4_I(inode)->i_file_acl)
436 		goto cleanup;
437 	ea_idebug(inode, "reading block %llu",
438 		  (unsigned long long)EXT4_I(inode)->i_file_acl);
439 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
440 	error = -EIO;
441 	if (!bh)
442 		goto cleanup;
443 	ea_bdebug(bh, "b_count=%d, refcount=%d",
444 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
445 	if (ext4_xattr_check_block(inode, bh)) {
446 		EXT4_ERROR_INODE(inode, "bad block %llu",
447 				 EXT4_I(inode)->i_file_acl);
448 		error = -EIO;
449 		goto cleanup;
450 	}
451 	ext4_xattr_cache_insert(ext4_mb_cache, bh);
452 	error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
453 
454 cleanup:
455 	brelse(bh);
456 
457 	return error;
458 }
459 
460 static int
461 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
462 {
463 	struct inode *inode = d_inode(dentry);
464 	struct ext4_xattr_ibody_header *header;
465 	struct ext4_inode *raw_inode;
466 	struct ext4_iloc iloc;
467 	void *end;
468 	int error;
469 
470 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
471 		return 0;
472 	error = ext4_get_inode_loc(inode, &iloc);
473 	if (error)
474 		return error;
475 	raw_inode = ext4_raw_inode(&iloc);
476 	header = IHDR(inode, raw_inode);
477 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
478 	error = ext4_xattr_check_names(IFIRST(header), end, IFIRST(header));
479 	if (error)
480 		goto cleanup;
481 	error = ext4_xattr_list_entries(dentry, IFIRST(header),
482 					buffer, buffer_size);
483 
484 cleanup:
485 	brelse(iloc.bh);
486 	return error;
487 }
488 
489 /*
490  * ext4_xattr_list()
491  *
492  * Copy a list of attribute names into the buffer
493  * provided, or compute the buffer size required.
494  * Buffer is NULL to compute the size of the buffer required.
495  *
496  * Returns a negative error number on failure, or the number of bytes
497  * used / required on success.
498  */
499 static int
500 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
501 {
502 	int ret, ret2;
503 
504 	down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
505 	ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
506 	if (ret < 0)
507 		goto errout;
508 	if (buffer) {
509 		buffer += ret;
510 		buffer_size -= ret;
511 	}
512 	ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
513 	if (ret < 0)
514 		goto errout;
515 	ret += ret2;
516 errout:
517 	up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
518 	return ret;
519 }
520 
521 /*
522  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
523  * not set, set it.
524  */
525 static void ext4_xattr_update_super_block(handle_t *handle,
526 					  struct super_block *sb)
527 {
528 	if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_EXT_ATTR))
529 		return;
530 
531 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
532 	if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
533 		EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_EXT_ATTR);
534 		ext4_handle_dirty_super(handle, sb);
535 	}
536 }
537 
538 /*
539  * Release the xattr block BH: If the reference count is > 1, decrement it;
540  * otherwise free the block.
541  */
542 static void
543 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
544 			 struct buffer_head *bh)
545 {
546 	struct mb_cache_entry *ce = NULL;
547 	int error = 0;
548 	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
549 
550 	ce = mb_cache_entry_get(ext4_mb_cache, bh->b_bdev, bh->b_blocknr);
551 	BUFFER_TRACE(bh, "get_write_access");
552 	error = ext4_journal_get_write_access(handle, bh);
553 	if (error)
554 		goto out;
555 
556 	lock_buffer(bh);
557 	if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
558 		ea_bdebug(bh, "refcount now=0; freeing");
559 		if (ce)
560 			mb_cache_entry_free(ce);
561 		get_bh(bh);
562 		unlock_buffer(bh);
563 		ext4_free_blocks(handle, inode, bh, 0, 1,
564 				 EXT4_FREE_BLOCKS_METADATA |
565 				 EXT4_FREE_BLOCKS_FORGET);
566 	} else {
567 		le32_add_cpu(&BHDR(bh)->h_refcount, -1);
568 		if (ce)
569 			mb_cache_entry_release(ce);
570 		/*
571 		 * Beware of this ugliness: Releasing of xattr block references
572 		 * from different inodes can race and so we have to protect
573 		 * from a race where someone else frees the block (and releases
574 		 * its journal_head) before we are done dirtying the buffer. In
575 		 * nojournal mode this race is harmless and we actually cannot
576 		 * call ext4_handle_dirty_xattr_block() with locked buffer as
577 		 * that function can call sync_dirty_buffer() so for that case
578 		 * we handle the dirtying after unlocking the buffer.
579 		 */
580 		if (ext4_handle_valid(handle))
581 			error = ext4_handle_dirty_xattr_block(handle, inode,
582 							      bh);
583 		unlock_buffer(bh);
584 		if (!ext4_handle_valid(handle))
585 			error = ext4_handle_dirty_xattr_block(handle, inode,
586 							      bh);
587 		if (IS_SYNC(inode))
588 			ext4_handle_sync(handle);
589 		dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
590 		ea_bdebug(bh, "refcount now=%d; releasing",
591 			  le32_to_cpu(BHDR(bh)->h_refcount));
592 	}
593 out:
594 	ext4_std_error(inode->i_sb, error);
595 	return;
596 }
597 
598 /*
599  * Find the available free space for EAs. This also returns the total number of
600  * bytes used by EA entries.
601  */
602 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
603 				    size_t *min_offs, void *base, int *total)
604 {
605 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
606 		if (!last->e_value_block && last->e_value_size) {
607 			size_t offs = le16_to_cpu(last->e_value_offs);
608 			if (offs < *min_offs)
609 				*min_offs = offs;
610 		}
611 		if (total)
612 			*total += EXT4_XATTR_LEN(last->e_name_len);
613 	}
614 	return (*min_offs - ((void *)last - base) - sizeof(__u32));
615 }
616 
617 static int
618 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
619 {
620 	struct ext4_xattr_entry *last;
621 	size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
622 
623 	/* Compute min_offs and last. */
624 	last = s->first;
625 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
626 		if (!last->e_value_block && last->e_value_size) {
627 			size_t offs = le16_to_cpu(last->e_value_offs);
628 			if (offs < min_offs)
629 				min_offs = offs;
630 		}
631 	}
632 	free = min_offs - ((void *)last - s->base) - sizeof(__u32);
633 	if (!s->not_found) {
634 		if (!s->here->e_value_block && s->here->e_value_size) {
635 			size_t size = le32_to_cpu(s->here->e_value_size);
636 			free += EXT4_XATTR_SIZE(size);
637 		}
638 		free += EXT4_XATTR_LEN(name_len);
639 	}
640 	if (i->value) {
641 		if (free < EXT4_XATTR_LEN(name_len) +
642 			   EXT4_XATTR_SIZE(i->value_len))
643 			return -ENOSPC;
644 	}
645 
646 	if (i->value && s->not_found) {
647 		/* Insert the new name. */
648 		size_t size = EXT4_XATTR_LEN(name_len);
649 		size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
650 		memmove((void *)s->here + size, s->here, rest);
651 		memset(s->here, 0, size);
652 		s->here->e_name_index = i->name_index;
653 		s->here->e_name_len = name_len;
654 		memcpy(s->here->e_name, i->name, name_len);
655 	} else {
656 		if (!s->here->e_value_block && s->here->e_value_size) {
657 			void *first_val = s->base + min_offs;
658 			size_t offs = le16_to_cpu(s->here->e_value_offs);
659 			void *val = s->base + offs;
660 			size_t size = EXT4_XATTR_SIZE(
661 				le32_to_cpu(s->here->e_value_size));
662 
663 			if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
664 				/* The old and the new value have the same
665 				   size. Just replace. */
666 				s->here->e_value_size =
667 					cpu_to_le32(i->value_len);
668 				if (i->value == EXT4_ZERO_XATTR_VALUE) {
669 					memset(val, 0, size);
670 				} else {
671 					/* Clear pad bytes first. */
672 					memset(val + size - EXT4_XATTR_PAD, 0,
673 					       EXT4_XATTR_PAD);
674 					memcpy(val, i->value, i->value_len);
675 				}
676 				return 0;
677 			}
678 
679 			/* Remove the old value. */
680 			memmove(first_val + size, first_val, val - first_val);
681 			memset(first_val, 0, size);
682 			s->here->e_value_size = 0;
683 			s->here->e_value_offs = 0;
684 			min_offs += size;
685 
686 			/* Adjust all value offsets. */
687 			last = s->first;
688 			while (!IS_LAST_ENTRY(last)) {
689 				size_t o = le16_to_cpu(last->e_value_offs);
690 				if (!last->e_value_block &&
691 				    last->e_value_size && o < offs)
692 					last->e_value_offs =
693 						cpu_to_le16(o + size);
694 				last = EXT4_XATTR_NEXT(last);
695 			}
696 		}
697 		if (!i->value) {
698 			/* Remove the old name. */
699 			size_t size = EXT4_XATTR_LEN(name_len);
700 			last = ENTRY((void *)last - size);
701 			memmove(s->here, (void *)s->here + size,
702 				(void *)last - (void *)s->here + sizeof(__u32));
703 			memset(last, 0, size);
704 		}
705 	}
706 
707 	if (i->value) {
708 		/* Insert the new value. */
709 		s->here->e_value_size = cpu_to_le32(i->value_len);
710 		if (i->value_len) {
711 			size_t size = EXT4_XATTR_SIZE(i->value_len);
712 			void *val = s->base + min_offs - size;
713 			s->here->e_value_offs = cpu_to_le16(min_offs - size);
714 			if (i->value == EXT4_ZERO_XATTR_VALUE) {
715 				memset(val, 0, size);
716 			} else {
717 				/* Clear the pad bytes first. */
718 				memset(val + size - EXT4_XATTR_PAD, 0,
719 				       EXT4_XATTR_PAD);
720 				memcpy(val, i->value, i->value_len);
721 			}
722 		}
723 	}
724 	return 0;
725 }
726 
727 struct ext4_xattr_block_find {
728 	struct ext4_xattr_search s;
729 	struct buffer_head *bh;
730 };
731 
732 static int
733 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
734 		      struct ext4_xattr_block_find *bs)
735 {
736 	struct super_block *sb = inode->i_sb;
737 	int error;
738 
739 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
740 		  i->name_index, i->name, i->value, (long)i->value_len);
741 
742 	if (EXT4_I(inode)->i_file_acl) {
743 		/* The inode already has an extended attribute block. */
744 		bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
745 		error = -EIO;
746 		if (!bs->bh)
747 			goto cleanup;
748 		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
749 			atomic_read(&(bs->bh->b_count)),
750 			le32_to_cpu(BHDR(bs->bh)->h_refcount));
751 		if (ext4_xattr_check_block(inode, bs->bh)) {
752 			EXT4_ERROR_INODE(inode, "bad block %llu",
753 					 EXT4_I(inode)->i_file_acl);
754 			error = -EIO;
755 			goto cleanup;
756 		}
757 		/* Find the named attribute. */
758 		bs->s.base = BHDR(bs->bh);
759 		bs->s.first = BFIRST(bs->bh);
760 		bs->s.end = bs->bh->b_data + bs->bh->b_size;
761 		bs->s.here = bs->s.first;
762 		error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
763 					      i->name, bs->bh->b_size, 1);
764 		if (error && error != -ENODATA)
765 			goto cleanup;
766 		bs->s.not_found = error;
767 	}
768 	error = 0;
769 
770 cleanup:
771 	return error;
772 }
773 
774 static int
775 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
776 		     struct ext4_xattr_info *i,
777 		     struct ext4_xattr_block_find *bs)
778 {
779 	struct super_block *sb = inode->i_sb;
780 	struct buffer_head *new_bh = NULL;
781 	struct ext4_xattr_search *s = &bs->s;
782 	struct mb_cache_entry *ce = NULL;
783 	int error = 0;
784 	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
785 
786 #define header(x) ((struct ext4_xattr_header *)(x))
787 
788 	if (i->value && i->value_len > sb->s_blocksize)
789 		return -ENOSPC;
790 	if (s->base) {
791 		ce = mb_cache_entry_get(ext4_mb_cache, bs->bh->b_bdev,
792 					bs->bh->b_blocknr);
793 		BUFFER_TRACE(bs->bh, "get_write_access");
794 		error = ext4_journal_get_write_access(handle, bs->bh);
795 		if (error)
796 			goto cleanup;
797 		lock_buffer(bs->bh);
798 
799 		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
800 			if (ce) {
801 				mb_cache_entry_free(ce);
802 				ce = NULL;
803 			}
804 			ea_bdebug(bs->bh, "modifying in-place");
805 			error = ext4_xattr_set_entry(i, s);
806 			if (!error) {
807 				if (!IS_LAST_ENTRY(s->first))
808 					ext4_xattr_rehash(header(s->base),
809 							  s->here);
810 				ext4_xattr_cache_insert(ext4_mb_cache,
811 					bs->bh);
812 			}
813 			unlock_buffer(bs->bh);
814 			if (error == -EIO)
815 				goto bad_block;
816 			if (!error)
817 				error = ext4_handle_dirty_xattr_block(handle,
818 								      inode,
819 								      bs->bh);
820 			if (error)
821 				goto cleanup;
822 			goto inserted;
823 		} else {
824 			int offset = (char *)s->here - bs->bh->b_data;
825 
826 			unlock_buffer(bs->bh);
827 			if (ce) {
828 				mb_cache_entry_release(ce);
829 				ce = NULL;
830 			}
831 			ea_bdebug(bs->bh, "cloning");
832 			s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
833 			error = -ENOMEM;
834 			if (s->base == NULL)
835 				goto cleanup;
836 			memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
837 			s->first = ENTRY(header(s->base)+1);
838 			header(s->base)->h_refcount = cpu_to_le32(1);
839 			s->here = ENTRY(s->base + offset);
840 			s->end = s->base + bs->bh->b_size;
841 		}
842 	} else {
843 		/* Allocate a buffer where we construct the new block. */
844 		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
845 		/* assert(header == s->base) */
846 		error = -ENOMEM;
847 		if (s->base == NULL)
848 			goto cleanup;
849 		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
850 		header(s->base)->h_blocks = cpu_to_le32(1);
851 		header(s->base)->h_refcount = cpu_to_le32(1);
852 		s->first = ENTRY(header(s->base)+1);
853 		s->here = ENTRY(header(s->base)+1);
854 		s->end = s->base + sb->s_blocksize;
855 	}
856 
857 	error = ext4_xattr_set_entry(i, s);
858 	if (error == -EIO)
859 		goto bad_block;
860 	if (error)
861 		goto cleanup;
862 	if (!IS_LAST_ENTRY(s->first))
863 		ext4_xattr_rehash(header(s->base), s->here);
864 
865 inserted:
866 	if (!IS_LAST_ENTRY(s->first)) {
867 		new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
868 		if (new_bh) {
869 			/* We found an identical block in the cache. */
870 			if (new_bh == bs->bh)
871 				ea_bdebug(new_bh, "keeping");
872 			else {
873 				/* The old block is released after updating
874 				   the inode. */
875 				error = dquot_alloc_block(inode,
876 						EXT4_C2B(EXT4_SB(sb), 1));
877 				if (error)
878 					goto cleanup;
879 				BUFFER_TRACE(new_bh, "get_write_access");
880 				error = ext4_journal_get_write_access(handle,
881 								      new_bh);
882 				if (error)
883 					goto cleanup_dquot;
884 				lock_buffer(new_bh);
885 				le32_add_cpu(&BHDR(new_bh)->h_refcount, 1);
886 				ea_bdebug(new_bh, "reusing; refcount now=%d",
887 					le32_to_cpu(BHDR(new_bh)->h_refcount));
888 				unlock_buffer(new_bh);
889 				error = ext4_handle_dirty_xattr_block(handle,
890 								      inode,
891 								      new_bh);
892 				if (error)
893 					goto cleanup_dquot;
894 			}
895 			mb_cache_entry_release(ce);
896 			ce = NULL;
897 		} else if (bs->bh && s->base == bs->bh->b_data) {
898 			/* We were modifying this block in-place. */
899 			ea_bdebug(bs->bh, "keeping this block");
900 			new_bh = bs->bh;
901 			get_bh(new_bh);
902 		} else {
903 			/* We need to allocate a new block */
904 			ext4_fsblk_t goal, block;
905 
906 			goal = ext4_group_first_block_no(sb,
907 						EXT4_I(inode)->i_block_group);
908 
909 			/* non-extent files can't have physical blocks past 2^32 */
910 			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
911 				goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
912 
913 			block = ext4_new_meta_blocks(handle, inode, goal, 0,
914 						     NULL, &error);
915 			if (error)
916 				goto cleanup;
917 
918 			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
919 				BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
920 
921 			ea_idebug(inode, "creating block %llu",
922 				  (unsigned long long)block);
923 
924 			new_bh = sb_getblk(sb, block);
925 			if (unlikely(!new_bh)) {
926 				error = -ENOMEM;
927 getblk_failed:
928 				ext4_free_blocks(handle, inode, NULL, block, 1,
929 						 EXT4_FREE_BLOCKS_METADATA);
930 				goto cleanup;
931 			}
932 			lock_buffer(new_bh);
933 			error = ext4_journal_get_create_access(handle, new_bh);
934 			if (error) {
935 				unlock_buffer(new_bh);
936 				error = -EIO;
937 				goto getblk_failed;
938 			}
939 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
940 			set_buffer_uptodate(new_bh);
941 			unlock_buffer(new_bh);
942 			ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
943 			error = ext4_handle_dirty_xattr_block(handle,
944 							      inode, new_bh);
945 			if (error)
946 				goto cleanup;
947 		}
948 	}
949 
950 	/* Update the inode. */
951 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
952 
953 	/* Drop the previous xattr block. */
954 	if (bs->bh && bs->bh != new_bh)
955 		ext4_xattr_release_block(handle, inode, bs->bh);
956 	error = 0;
957 
958 cleanup:
959 	if (ce)
960 		mb_cache_entry_release(ce);
961 	brelse(new_bh);
962 	if (!(bs->bh && s->base == bs->bh->b_data))
963 		kfree(s->base);
964 
965 	return error;
966 
967 cleanup_dquot:
968 	dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
969 	goto cleanup;
970 
971 bad_block:
972 	EXT4_ERROR_INODE(inode, "bad block %llu",
973 			 EXT4_I(inode)->i_file_acl);
974 	goto cleanup;
975 
976 #undef header
977 }
978 
979 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
980 			  struct ext4_xattr_ibody_find *is)
981 {
982 	struct ext4_xattr_ibody_header *header;
983 	struct ext4_inode *raw_inode;
984 	int error;
985 
986 	if (EXT4_I(inode)->i_extra_isize == 0)
987 		return 0;
988 	raw_inode = ext4_raw_inode(&is->iloc);
989 	header = IHDR(inode, raw_inode);
990 	is->s.base = is->s.first = IFIRST(header);
991 	is->s.here = is->s.first;
992 	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
993 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
994 		error = ext4_xattr_check_names(IFIRST(header), is->s.end,
995 					       IFIRST(header));
996 		if (error)
997 			return error;
998 		/* Find the named attribute. */
999 		error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1000 					      i->name, is->s.end -
1001 					      (void *)is->s.base, 0);
1002 		if (error && error != -ENODATA)
1003 			return error;
1004 		is->s.not_found = error;
1005 	}
1006 	return 0;
1007 }
1008 
1009 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1010 				struct ext4_xattr_info *i,
1011 				struct ext4_xattr_ibody_find *is)
1012 {
1013 	struct ext4_xattr_ibody_header *header;
1014 	struct ext4_xattr_search *s = &is->s;
1015 	int error;
1016 
1017 	if (EXT4_I(inode)->i_extra_isize == 0)
1018 		return -ENOSPC;
1019 	error = ext4_xattr_set_entry(i, s);
1020 	if (error) {
1021 		if (error == -ENOSPC &&
1022 		    ext4_has_inline_data(inode)) {
1023 			error = ext4_try_to_evict_inline_data(handle, inode,
1024 					EXT4_XATTR_LEN(strlen(i->name) +
1025 					EXT4_XATTR_SIZE(i->value_len)));
1026 			if (error)
1027 				return error;
1028 			error = ext4_xattr_ibody_find(inode, i, is);
1029 			if (error)
1030 				return error;
1031 			error = ext4_xattr_set_entry(i, s);
1032 		}
1033 		if (error)
1034 			return error;
1035 	}
1036 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
1037 	if (!IS_LAST_ENTRY(s->first)) {
1038 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1039 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1040 	} else {
1041 		header->h_magic = cpu_to_le32(0);
1042 		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1043 	}
1044 	return 0;
1045 }
1046 
1047 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
1048 				struct ext4_xattr_info *i,
1049 				struct ext4_xattr_ibody_find *is)
1050 {
1051 	struct ext4_xattr_ibody_header *header;
1052 	struct ext4_xattr_search *s = &is->s;
1053 	int error;
1054 
1055 	if (EXT4_I(inode)->i_extra_isize == 0)
1056 		return -ENOSPC;
1057 	error = ext4_xattr_set_entry(i, s);
1058 	if (error)
1059 		return error;
1060 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
1061 	if (!IS_LAST_ENTRY(s->first)) {
1062 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1063 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1064 	} else {
1065 		header->h_magic = cpu_to_le32(0);
1066 		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1067 	}
1068 	return 0;
1069 }
1070 
1071 /*
1072  * ext4_xattr_set_handle()
1073  *
1074  * Create, replace or remove an extended attribute for this inode.  Value
1075  * is NULL to remove an existing extended attribute, and non-NULL to
1076  * either replace an existing extended attribute, or create a new extended
1077  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1078  * specify that an extended attribute must exist and must not exist
1079  * previous to the call, respectively.
1080  *
1081  * Returns 0, or a negative error number on failure.
1082  */
1083 int
1084 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1085 		      const char *name, const void *value, size_t value_len,
1086 		      int flags)
1087 {
1088 	struct ext4_xattr_info i = {
1089 		.name_index = name_index,
1090 		.name = name,
1091 		.value = value,
1092 		.value_len = value_len,
1093 
1094 	};
1095 	struct ext4_xattr_ibody_find is = {
1096 		.s = { .not_found = -ENODATA, },
1097 	};
1098 	struct ext4_xattr_block_find bs = {
1099 		.s = { .not_found = -ENODATA, },
1100 	};
1101 	unsigned long no_expand;
1102 	int error;
1103 
1104 	if (!name)
1105 		return -EINVAL;
1106 	if (strlen(name) > 255)
1107 		return -ERANGE;
1108 	down_write(&EXT4_I(inode)->xattr_sem);
1109 	no_expand = ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND);
1110 	ext4_set_inode_state(inode, EXT4_STATE_NO_EXPAND);
1111 
1112 	error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1113 	if (error)
1114 		goto cleanup;
1115 
1116 	if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1117 		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1118 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1119 		ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1120 	}
1121 
1122 	error = ext4_xattr_ibody_find(inode, &i, &is);
1123 	if (error)
1124 		goto cleanup;
1125 	if (is.s.not_found)
1126 		error = ext4_xattr_block_find(inode, &i, &bs);
1127 	if (error)
1128 		goto cleanup;
1129 	if (is.s.not_found && bs.s.not_found) {
1130 		error = -ENODATA;
1131 		if (flags & XATTR_REPLACE)
1132 			goto cleanup;
1133 		error = 0;
1134 		if (!value)
1135 			goto cleanup;
1136 	} else {
1137 		error = -EEXIST;
1138 		if (flags & XATTR_CREATE)
1139 			goto cleanup;
1140 	}
1141 	if (!value) {
1142 		if (!is.s.not_found)
1143 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1144 		else if (!bs.s.not_found)
1145 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1146 	} else {
1147 		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1148 		if (!error && !bs.s.not_found) {
1149 			i.value = NULL;
1150 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1151 		} else if (error == -ENOSPC) {
1152 			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1153 				error = ext4_xattr_block_find(inode, &i, &bs);
1154 				if (error)
1155 					goto cleanup;
1156 			}
1157 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1158 			if (error)
1159 				goto cleanup;
1160 			if (!is.s.not_found) {
1161 				i.value = NULL;
1162 				error = ext4_xattr_ibody_set(handle, inode, &i,
1163 							     &is);
1164 			}
1165 		}
1166 	}
1167 	if (!error) {
1168 		ext4_xattr_update_super_block(handle, inode->i_sb);
1169 		inode->i_ctime = ext4_current_time(inode);
1170 		if (!value)
1171 			ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1172 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1173 		/*
1174 		 * The bh is consumed by ext4_mark_iloc_dirty, even with
1175 		 * error != 0.
1176 		 */
1177 		is.iloc.bh = NULL;
1178 		if (IS_SYNC(inode))
1179 			ext4_handle_sync(handle);
1180 	}
1181 
1182 cleanup:
1183 	brelse(is.iloc.bh);
1184 	brelse(bs.bh);
1185 	if (no_expand == 0)
1186 		ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1187 	up_write(&EXT4_I(inode)->xattr_sem);
1188 	return error;
1189 }
1190 
1191 /*
1192  * ext4_xattr_set()
1193  *
1194  * Like ext4_xattr_set_handle, but start from an inode. This extended
1195  * attribute modification is a filesystem transaction by itself.
1196  *
1197  * Returns 0, or a negative error number on failure.
1198  */
1199 int
1200 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1201 	       const void *value, size_t value_len, int flags)
1202 {
1203 	handle_t *handle;
1204 	int error, retries = 0;
1205 	int credits = ext4_jbd2_credits_xattr(inode);
1206 
1207 retry:
1208 	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1209 	if (IS_ERR(handle)) {
1210 		error = PTR_ERR(handle);
1211 	} else {
1212 		int error2;
1213 
1214 		error = ext4_xattr_set_handle(handle, inode, name_index, name,
1215 					      value, value_len, flags);
1216 		error2 = ext4_journal_stop(handle);
1217 		if (error == -ENOSPC &&
1218 		    ext4_should_retry_alloc(inode->i_sb, &retries))
1219 			goto retry;
1220 		if (error == 0)
1221 			error = error2;
1222 	}
1223 
1224 	return error;
1225 }
1226 
1227 /*
1228  * Shift the EA entries in the inode to create space for the increased
1229  * i_extra_isize.
1230  */
1231 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1232 				     int value_offs_shift, void *to,
1233 				     void *from, size_t n, int blocksize)
1234 {
1235 	struct ext4_xattr_entry *last = entry;
1236 	int new_offs;
1237 
1238 	/* Adjust the value offsets of the entries */
1239 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1240 		if (!last->e_value_block && last->e_value_size) {
1241 			new_offs = le16_to_cpu(last->e_value_offs) +
1242 							value_offs_shift;
1243 			BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
1244 				 > blocksize);
1245 			last->e_value_offs = cpu_to_le16(new_offs);
1246 		}
1247 	}
1248 	/* Shift the entries by n bytes */
1249 	memmove(to, from, n);
1250 }
1251 
1252 /*
1253  * Expand an inode by new_extra_isize bytes when EAs are present.
1254  * Returns 0 on success or negative error number on failure.
1255  */
1256 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1257 			       struct ext4_inode *raw_inode, handle_t *handle)
1258 {
1259 	struct ext4_xattr_ibody_header *header;
1260 	struct ext4_xattr_entry *entry, *last, *first;
1261 	struct buffer_head *bh = NULL;
1262 	struct ext4_xattr_ibody_find *is = NULL;
1263 	struct ext4_xattr_block_find *bs = NULL;
1264 	char *buffer = NULL, *b_entry_name = NULL;
1265 	size_t min_offs, free;
1266 	int total_ino;
1267 	void *base, *start, *end;
1268 	int extra_isize = 0, error = 0, tried_min_extra_isize = 0;
1269 	int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1270 
1271 	down_write(&EXT4_I(inode)->xattr_sem);
1272 retry:
1273 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) {
1274 		up_write(&EXT4_I(inode)->xattr_sem);
1275 		return 0;
1276 	}
1277 
1278 	header = IHDR(inode, raw_inode);
1279 	entry = IFIRST(header);
1280 
1281 	/*
1282 	 * Check if enough free space is available in the inode to shift the
1283 	 * entries ahead by new_extra_isize.
1284 	 */
1285 
1286 	base = start = entry;
1287 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1288 	min_offs = end - base;
1289 	last = entry;
1290 	total_ino = sizeof(struct ext4_xattr_ibody_header);
1291 
1292 	free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
1293 	if (free >= new_extra_isize) {
1294 		entry = IFIRST(header);
1295 		ext4_xattr_shift_entries(entry,	EXT4_I(inode)->i_extra_isize
1296 				- new_extra_isize, (void *)raw_inode +
1297 				EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1298 				(void *)header, total_ino,
1299 				inode->i_sb->s_blocksize);
1300 		EXT4_I(inode)->i_extra_isize = new_extra_isize;
1301 		error = 0;
1302 		goto cleanup;
1303 	}
1304 
1305 	/*
1306 	 * Enough free space isn't available in the inode, check if
1307 	 * EA block can hold new_extra_isize bytes.
1308 	 */
1309 	if (EXT4_I(inode)->i_file_acl) {
1310 		bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1311 		error = -EIO;
1312 		if (!bh)
1313 			goto cleanup;
1314 		if (ext4_xattr_check_block(inode, bh)) {
1315 			EXT4_ERROR_INODE(inode, "bad block %llu",
1316 					 EXT4_I(inode)->i_file_acl);
1317 			error = -EIO;
1318 			goto cleanup;
1319 		}
1320 		base = BHDR(bh);
1321 		first = BFIRST(bh);
1322 		end = bh->b_data + bh->b_size;
1323 		min_offs = end - base;
1324 		free = ext4_xattr_free_space(first, &min_offs, base, NULL);
1325 		if (free < new_extra_isize) {
1326 			if (!tried_min_extra_isize && s_min_extra_isize) {
1327 				tried_min_extra_isize++;
1328 				new_extra_isize = s_min_extra_isize;
1329 				brelse(bh);
1330 				goto retry;
1331 			}
1332 			error = -1;
1333 			goto cleanup;
1334 		}
1335 	} else {
1336 		free = inode->i_sb->s_blocksize;
1337 	}
1338 
1339 	while (new_extra_isize > 0) {
1340 		size_t offs, size, entry_size;
1341 		struct ext4_xattr_entry *small_entry = NULL;
1342 		struct ext4_xattr_info i = {
1343 			.value = NULL,
1344 			.value_len = 0,
1345 		};
1346 		unsigned int total_size;  /* EA entry size + value size */
1347 		unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
1348 		unsigned int min_total_size = ~0U;
1349 
1350 		is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1351 		bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1352 		if (!is || !bs) {
1353 			error = -ENOMEM;
1354 			goto cleanup;
1355 		}
1356 
1357 		is->s.not_found = -ENODATA;
1358 		bs->s.not_found = -ENODATA;
1359 		is->iloc.bh = NULL;
1360 		bs->bh = NULL;
1361 
1362 		last = IFIRST(header);
1363 		/* Find the entry best suited to be pushed into EA block */
1364 		entry = NULL;
1365 		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1366 			total_size =
1367 			EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1368 					EXT4_XATTR_LEN(last->e_name_len);
1369 			if (total_size <= free && total_size < min_total_size) {
1370 				if (total_size < new_extra_isize) {
1371 					small_entry = last;
1372 				} else {
1373 					entry = last;
1374 					min_total_size = total_size;
1375 				}
1376 			}
1377 		}
1378 
1379 		if (entry == NULL) {
1380 			if (small_entry) {
1381 				entry = small_entry;
1382 			} else {
1383 				if (!tried_min_extra_isize &&
1384 				    s_min_extra_isize) {
1385 					tried_min_extra_isize++;
1386 					new_extra_isize = s_min_extra_isize;
1387 					kfree(is); is = NULL;
1388 					kfree(bs); bs = NULL;
1389 					brelse(bh);
1390 					goto retry;
1391 				}
1392 				error = -1;
1393 				goto cleanup;
1394 			}
1395 		}
1396 		offs = le16_to_cpu(entry->e_value_offs);
1397 		size = le32_to_cpu(entry->e_value_size);
1398 		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1399 		i.name_index = entry->e_name_index,
1400 		buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
1401 		b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1402 		if (!buffer || !b_entry_name) {
1403 			error = -ENOMEM;
1404 			goto cleanup;
1405 		}
1406 		/* Save the entry name and the entry value */
1407 		memcpy(buffer, (void *)IFIRST(header) + offs,
1408 		       EXT4_XATTR_SIZE(size));
1409 		memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1410 		b_entry_name[entry->e_name_len] = '\0';
1411 		i.name = b_entry_name;
1412 
1413 		error = ext4_get_inode_loc(inode, &is->iloc);
1414 		if (error)
1415 			goto cleanup;
1416 
1417 		error = ext4_xattr_ibody_find(inode, &i, is);
1418 		if (error)
1419 			goto cleanup;
1420 
1421 		/* Remove the chosen entry from the inode */
1422 		error = ext4_xattr_ibody_set(handle, inode, &i, is);
1423 		if (error)
1424 			goto cleanup;
1425 
1426 		entry = IFIRST(header);
1427 		if (entry_size + EXT4_XATTR_SIZE(size) >= new_extra_isize)
1428 			shift_bytes = new_extra_isize;
1429 		else
1430 			shift_bytes = entry_size + size;
1431 		/* Adjust the offsets and shift the remaining entries ahead */
1432 		ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize -
1433 			shift_bytes, (void *)raw_inode +
1434 			EXT4_GOOD_OLD_INODE_SIZE + extra_isize + shift_bytes,
1435 			(void *)header, total_ino - entry_size,
1436 			inode->i_sb->s_blocksize);
1437 
1438 		extra_isize += shift_bytes;
1439 		new_extra_isize -= shift_bytes;
1440 		EXT4_I(inode)->i_extra_isize = extra_isize;
1441 
1442 		i.name = b_entry_name;
1443 		i.value = buffer;
1444 		i.value_len = size;
1445 		error = ext4_xattr_block_find(inode, &i, bs);
1446 		if (error)
1447 			goto cleanup;
1448 
1449 		/* Add entry which was removed from the inode into the block */
1450 		error = ext4_xattr_block_set(handle, inode, &i, bs);
1451 		if (error)
1452 			goto cleanup;
1453 		kfree(b_entry_name);
1454 		kfree(buffer);
1455 		b_entry_name = NULL;
1456 		buffer = NULL;
1457 		brelse(is->iloc.bh);
1458 		kfree(is);
1459 		kfree(bs);
1460 	}
1461 	brelse(bh);
1462 	up_write(&EXT4_I(inode)->xattr_sem);
1463 	return 0;
1464 
1465 cleanup:
1466 	kfree(b_entry_name);
1467 	kfree(buffer);
1468 	if (is)
1469 		brelse(is->iloc.bh);
1470 	kfree(is);
1471 	kfree(bs);
1472 	brelse(bh);
1473 	up_write(&EXT4_I(inode)->xattr_sem);
1474 	return error;
1475 }
1476 
1477 
1478 
1479 /*
1480  * ext4_xattr_delete_inode()
1481  *
1482  * Free extended attribute resources associated with this inode. This
1483  * is called immediately before an inode is freed. We have exclusive
1484  * access to the inode.
1485  */
1486 void
1487 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1488 {
1489 	struct buffer_head *bh = NULL;
1490 
1491 	if (!EXT4_I(inode)->i_file_acl)
1492 		goto cleanup;
1493 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1494 	if (!bh) {
1495 		EXT4_ERROR_INODE(inode, "block %llu read error",
1496 				 EXT4_I(inode)->i_file_acl);
1497 		goto cleanup;
1498 	}
1499 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1500 	    BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1501 		EXT4_ERROR_INODE(inode, "bad block %llu",
1502 				 EXT4_I(inode)->i_file_acl);
1503 		goto cleanup;
1504 	}
1505 	ext4_xattr_release_block(handle, inode, bh);
1506 	EXT4_I(inode)->i_file_acl = 0;
1507 
1508 cleanup:
1509 	brelse(bh);
1510 }
1511 
1512 /*
1513  * ext4_xattr_put_super()
1514  *
1515  * This is called when a file system is unmounted.
1516  */
1517 void
1518 ext4_xattr_put_super(struct super_block *sb)
1519 {
1520 	mb_cache_shrink(sb->s_bdev);
1521 }
1522 
1523 /*
1524  * ext4_xattr_cache_insert()
1525  *
1526  * Create a new entry in the extended attribute cache, and insert
1527  * it unless such an entry is already in the cache.
1528  *
1529  * Returns 0, or a negative error number on failure.
1530  */
1531 static void
1532 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1533 {
1534 	__u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1535 	struct mb_cache_entry *ce;
1536 	int error;
1537 
1538 	ce = mb_cache_entry_alloc(ext4_mb_cache, GFP_NOFS);
1539 	if (!ce) {
1540 		ea_bdebug(bh, "out of memory");
1541 		return;
1542 	}
1543 	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
1544 	if (error) {
1545 		mb_cache_entry_free(ce);
1546 		if (error == -EBUSY) {
1547 			ea_bdebug(bh, "already in cache");
1548 			error = 0;
1549 		}
1550 	} else {
1551 		ea_bdebug(bh, "inserting [%x]", (int)hash);
1552 		mb_cache_entry_release(ce);
1553 	}
1554 }
1555 
1556 /*
1557  * ext4_xattr_cmp()
1558  *
1559  * Compare two extended attribute blocks for equality.
1560  *
1561  * Returns 0 if the blocks are equal, 1 if they differ, and
1562  * a negative error number on errors.
1563  */
1564 static int
1565 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1566 	       struct ext4_xattr_header *header2)
1567 {
1568 	struct ext4_xattr_entry *entry1, *entry2;
1569 
1570 	entry1 = ENTRY(header1+1);
1571 	entry2 = ENTRY(header2+1);
1572 	while (!IS_LAST_ENTRY(entry1)) {
1573 		if (IS_LAST_ENTRY(entry2))
1574 			return 1;
1575 		if (entry1->e_hash != entry2->e_hash ||
1576 		    entry1->e_name_index != entry2->e_name_index ||
1577 		    entry1->e_name_len != entry2->e_name_len ||
1578 		    entry1->e_value_size != entry2->e_value_size ||
1579 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1580 			return 1;
1581 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1582 			return -EIO;
1583 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1584 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1585 			   le32_to_cpu(entry1->e_value_size)))
1586 			return 1;
1587 
1588 		entry1 = EXT4_XATTR_NEXT(entry1);
1589 		entry2 = EXT4_XATTR_NEXT(entry2);
1590 	}
1591 	if (!IS_LAST_ENTRY(entry2))
1592 		return 1;
1593 	return 0;
1594 }
1595 
1596 /*
1597  * ext4_xattr_cache_find()
1598  *
1599  * Find an identical extended attribute block.
1600  *
1601  * Returns a pointer to the block found, or NULL if such a block was
1602  * not found or an error occurred.
1603  */
1604 static struct buffer_head *
1605 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1606 		      struct mb_cache_entry **pce)
1607 {
1608 	__u32 hash = le32_to_cpu(header->h_hash);
1609 	struct mb_cache_entry *ce;
1610 	struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1611 
1612 	if (!header->h_hash)
1613 		return NULL;  /* never share */
1614 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1615 again:
1616 	ce = mb_cache_entry_find_first(ext4_mb_cache, inode->i_sb->s_bdev,
1617 				       hash);
1618 	while (ce) {
1619 		struct buffer_head *bh;
1620 
1621 		if (IS_ERR(ce)) {
1622 			if (PTR_ERR(ce) == -EAGAIN)
1623 				goto again;
1624 			break;
1625 		}
1626 		bh = sb_bread(inode->i_sb, ce->e_block);
1627 		if (!bh) {
1628 			EXT4_ERROR_INODE(inode, "block %lu read error",
1629 					 (unsigned long) ce->e_block);
1630 		} else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1631 				EXT4_XATTR_REFCOUNT_MAX) {
1632 			ea_idebug(inode, "block %lu refcount %d>=%d",
1633 				  (unsigned long) ce->e_block,
1634 				  le32_to_cpu(BHDR(bh)->h_refcount),
1635 					  EXT4_XATTR_REFCOUNT_MAX);
1636 		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1637 			*pce = ce;
1638 			return bh;
1639 		}
1640 		brelse(bh);
1641 		ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
1642 	}
1643 	return NULL;
1644 }
1645 
1646 #define NAME_HASH_SHIFT 5
1647 #define VALUE_HASH_SHIFT 16
1648 
1649 /*
1650  * ext4_xattr_hash_entry()
1651  *
1652  * Compute the hash of an extended attribute.
1653  */
1654 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1655 					 struct ext4_xattr_entry *entry)
1656 {
1657 	__u32 hash = 0;
1658 	char *name = entry->e_name;
1659 	int n;
1660 
1661 	for (n = 0; n < entry->e_name_len; n++) {
1662 		hash = (hash << NAME_HASH_SHIFT) ^
1663 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1664 		       *name++;
1665 	}
1666 
1667 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1668 		__le32 *value = (__le32 *)((char *)header +
1669 			le16_to_cpu(entry->e_value_offs));
1670 		for (n = (le32_to_cpu(entry->e_value_size) +
1671 		     EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1672 			hash = (hash << VALUE_HASH_SHIFT) ^
1673 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1674 			       le32_to_cpu(*value++);
1675 		}
1676 	}
1677 	entry->e_hash = cpu_to_le32(hash);
1678 }
1679 
1680 #undef NAME_HASH_SHIFT
1681 #undef VALUE_HASH_SHIFT
1682 
1683 #define BLOCK_HASH_SHIFT 16
1684 
1685 /*
1686  * ext4_xattr_rehash()
1687  *
1688  * Re-compute the extended attribute hash value after an entry has changed.
1689  */
1690 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1691 			      struct ext4_xattr_entry *entry)
1692 {
1693 	struct ext4_xattr_entry *here;
1694 	__u32 hash = 0;
1695 
1696 	ext4_xattr_hash_entry(header, entry);
1697 	here = ENTRY(header+1);
1698 	while (!IS_LAST_ENTRY(here)) {
1699 		if (!here->e_hash) {
1700 			/* Block is not shared if an entry's hash value == 0 */
1701 			hash = 0;
1702 			break;
1703 		}
1704 		hash = (hash << BLOCK_HASH_SHIFT) ^
1705 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1706 		       le32_to_cpu(here->e_hash);
1707 		here = EXT4_XATTR_NEXT(here);
1708 	}
1709 	header->h_hash = cpu_to_le32(hash);
1710 }
1711 
1712 #undef BLOCK_HASH_SHIFT
1713 
1714 #define	HASH_BUCKET_BITS	10
1715 
1716 struct mb_cache *
1717 ext4_xattr_create_cache(char *name)
1718 {
1719 	return mb_cache_create(name, HASH_BUCKET_BITS);
1720 }
1721 
1722 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1723 {
1724 	if (cache)
1725 		mb_cache_destroy(cache);
1726 }
1727 
1728