xref: /openbmc/linux/fs/ext4/xattr.c (revision c21b37f6)
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/ext4_jbd2.h>
57 #include <linux/ext4_fs.h>
58 #include <linux/mbcache.h>
59 #include <linux/quotaops.h>
60 #include <linux/rwsem.h>
61 #include "xattr.h"
62 #include "acl.h"
63 
64 #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
65 #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
66 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
67 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
68 
69 #ifdef EXT4_XATTR_DEBUG
70 # define ea_idebug(inode, f...) do { \
71 		printk(KERN_DEBUG "inode %s:%lu: ", \
72 			inode->i_sb->s_id, inode->i_ino); \
73 		printk(f); \
74 		printk("\n"); \
75 	} while (0)
76 # define ea_bdebug(bh, f...) do { \
77 		char b[BDEVNAME_SIZE]; \
78 		printk(KERN_DEBUG "block %s:%lu: ", \
79 			bdevname(bh->b_bdev, b), \
80 			(unsigned long) bh->b_blocknr); \
81 		printk(f); \
82 		printk("\n"); \
83 	} while (0)
84 #else
85 # define ea_idebug(f...)
86 # define ea_bdebug(f...)
87 #endif
88 
89 static void ext4_xattr_cache_insert(struct buffer_head *);
90 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
91 						 struct ext4_xattr_header *,
92 						 struct mb_cache_entry **);
93 static void ext4_xattr_rehash(struct ext4_xattr_header *,
94 			      struct ext4_xattr_entry *);
95 
96 static struct mb_cache *ext4_xattr_cache;
97 
98 static struct xattr_handler *ext4_xattr_handler_map[] = {
99 	[EXT4_XATTR_INDEX_USER]		     = &ext4_xattr_user_handler,
100 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
101 	[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &ext4_xattr_acl_access_handler,
102 	[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext4_xattr_acl_default_handler,
103 #endif
104 	[EXT4_XATTR_INDEX_TRUSTED]	     = &ext4_xattr_trusted_handler,
105 #ifdef CONFIG_EXT4DEV_FS_SECURITY
106 	[EXT4_XATTR_INDEX_SECURITY]	     = &ext4_xattr_security_handler,
107 #endif
108 };
109 
110 struct xattr_handler *ext4_xattr_handlers[] = {
111 	&ext4_xattr_user_handler,
112 	&ext4_xattr_trusted_handler,
113 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
114 	&ext4_xattr_acl_access_handler,
115 	&ext4_xattr_acl_default_handler,
116 #endif
117 #ifdef CONFIG_EXT4DEV_FS_SECURITY
118 	&ext4_xattr_security_handler,
119 #endif
120 	NULL
121 };
122 
123 static inline struct xattr_handler *
124 ext4_xattr_handler(int name_index)
125 {
126 	struct xattr_handler *handler = NULL;
127 
128 	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
129 		handler = ext4_xattr_handler_map[name_index];
130 	return handler;
131 }
132 
133 /*
134  * Inode operation listxattr()
135  *
136  * dentry->d_inode->i_mutex: don't care
137  */
138 ssize_t
139 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
140 {
141 	return ext4_xattr_list(dentry->d_inode, buffer, size);
142 }
143 
144 static int
145 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end)
146 {
147 	while (!IS_LAST_ENTRY(entry)) {
148 		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(entry);
149 		if ((void *)next >= end)
150 			return -EIO;
151 		entry = next;
152 	}
153 	return 0;
154 }
155 
156 static inline int
157 ext4_xattr_check_block(struct buffer_head *bh)
158 {
159 	int error;
160 
161 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
162 	    BHDR(bh)->h_blocks != cpu_to_le32(1))
163 		return -EIO;
164 	error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size);
165 	return error;
166 }
167 
168 static inline int
169 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
170 {
171 	size_t value_size = le32_to_cpu(entry->e_value_size);
172 
173 	if (entry->e_value_block != 0 || value_size > size ||
174 	    le16_to_cpu(entry->e_value_offs) + value_size > size)
175 		return -EIO;
176 	return 0;
177 }
178 
179 static int
180 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
181 		      const char *name, size_t size, int sorted)
182 {
183 	struct ext4_xattr_entry *entry;
184 	size_t name_len;
185 	int cmp = 1;
186 
187 	if (name == NULL)
188 		return -EINVAL;
189 	name_len = strlen(name);
190 	entry = *pentry;
191 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
192 		cmp = name_index - entry->e_name_index;
193 		if (!cmp)
194 			cmp = name_len - entry->e_name_len;
195 		if (!cmp)
196 			cmp = memcmp(name, entry->e_name, name_len);
197 		if (cmp <= 0 && (sorted || cmp == 0))
198 			break;
199 	}
200 	*pentry = entry;
201 	if (!cmp && ext4_xattr_check_entry(entry, size))
202 			return -EIO;
203 	return cmp ? -ENODATA : 0;
204 }
205 
206 static int
207 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
208 		     void *buffer, size_t buffer_size)
209 {
210 	struct buffer_head *bh = NULL;
211 	struct ext4_xattr_entry *entry;
212 	size_t size;
213 	int error;
214 
215 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
216 		  name_index, name, buffer, (long)buffer_size);
217 
218 	error = -ENODATA;
219 	if (!EXT4_I(inode)->i_file_acl)
220 		goto cleanup;
221 	ea_idebug(inode, "reading block %u", EXT4_I(inode)->i_file_acl);
222 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
223 	if (!bh)
224 		goto cleanup;
225 	ea_bdebug(bh, "b_count=%d, refcount=%d",
226 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
227 	if (ext4_xattr_check_block(bh)) {
228 bad_block:	ext4_error(inode->i_sb, __FUNCTION__,
229 			   "inode %lu: bad block %llu", inode->i_ino,
230 			   EXT4_I(inode)->i_file_acl);
231 		error = -EIO;
232 		goto cleanup;
233 	}
234 	ext4_xattr_cache_insert(bh);
235 	entry = BFIRST(bh);
236 	error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
237 	if (error == -EIO)
238 		goto bad_block;
239 	if (error)
240 		goto cleanup;
241 	size = le32_to_cpu(entry->e_value_size);
242 	if (buffer) {
243 		error = -ERANGE;
244 		if (size > buffer_size)
245 			goto cleanup;
246 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
247 		       size);
248 	}
249 	error = size;
250 
251 cleanup:
252 	brelse(bh);
253 	return error;
254 }
255 
256 static int
257 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
258 		     void *buffer, size_t buffer_size)
259 {
260 	struct ext4_xattr_ibody_header *header;
261 	struct ext4_xattr_entry *entry;
262 	struct ext4_inode *raw_inode;
263 	struct ext4_iloc iloc;
264 	size_t size;
265 	void *end;
266 	int error;
267 
268 	if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR))
269 		return -ENODATA;
270 	error = ext4_get_inode_loc(inode, &iloc);
271 	if (error)
272 		return error;
273 	raw_inode = ext4_raw_inode(&iloc);
274 	header = IHDR(inode, raw_inode);
275 	entry = IFIRST(header);
276 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
277 	error = ext4_xattr_check_names(entry, end);
278 	if (error)
279 		goto cleanup;
280 	error = ext4_xattr_find_entry(&entry, name_index, name,
281 				      end - (void *)entry, 0);
282 	if (error)
283 		goto cleanup;
284 	size = le32_to_cpu(entry->e_value_size);
285 	if (buffer) {
286 		error = -ERANGE;
287 		if (size > buffer_size)
288 			goto cleanup;
289 		memcpy(buffer, (void *)IFIRST(header) +
290 		       le16_to_cpu(entry->e_value_offs), size);
291 	}
292 	error = size;
293 
294 cleanup:
295 	brelse(iloc.bh);
296 	return error;
297 }
298 
299 /*
300  * ext4_xattr_get()
301  *
302  * Copy an extended attribute into the buffer
303  * provided, or compute the buffer size required.
304  * Buffer is NULL to compute the size of the buffer required.
305  *
306  * Returns a negative error number on failure, or the number of bytes
307  * used / required on success.
308  */
309 int
310 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
311 	       void *buffer, size_t buffer_size)
312 {
313 	int error;
314 
315 	down_read(&EXT4_I(inode)->xattr_sem);
316 	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
317 				     buffer_size);
318 	if (error == -ENODATA)
319 		error = ext4_xattr_block_get(inode, name_index, name, buffer,
320 					     buffer_size);
321 	up_read(&EXT4_I(inode)->xattr_sem);
322 	return error;
323 }
324 
325 static int
326 ext4_xattr_list_entries(struct inode *inode, struct ext4_xattr_entry *entry,
327 			char *buffer, size_t buffer_size)
328 {
329 	size_t rest = buffer_size;
330 
331 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
332 		struct xattr_handler *handler =
333 			ext4_xattr_handler(entry->e_name_index);
334 
335 		if (handler) {
336 			size_t size = handler->list(inode, buffer, rest,
337 						    entry->e_name,
338 						    entry->e_name_len);
339 			if (buffer) {
340 				if (size > rest)
341 					return -ERANGE;
342 				buffer += size;
343 			}
344 			rest -= size;
345 		}
346 	}
347 	return buffer_size - rest;
348 }
349 
350 static int
351 ext4_xattr_block_list(struct inode *inode, char *buffer, size_t buffer_size)
352 {
353 	struct buffer_head *bh = NULL;
354 	int error;
355 
356 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
357 		  buffer, (long)buffer_size);
358 
359 	error = 0;
360 	if (!EXT4_I(inode)->i_file_acl)
361 		goto cleanup;
362 	ea_idebug(inode, "reading block %u", EXT4_I(inode)->i_file_acl);
363 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
364 	error = -EIO;
365 	if (!bh)
366 		goto cleanup;
367 	ea_bdebug(bh, "b_count=%d, refcount=%d",
368 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
369 	if (ext4_xattr_check_block(bh)) {
370 		ext4_error(inode->i_sb, __FUNCTION__,
371 			   "inode %lu: bad block %llu", inode->i_ino,
372 			   EXT4_I(inode)->i_file_acl);
373 		error = -EIO;
374 		goto cleanup;
375 	}
376 	ext4_xattr_cache_insert(bh);
377 	error = ext4_xattr_list_entries(inode, BFIRST(bh), buffer, buffer_size);
378 
379 cleanup:
380 	brelse(bh);
381 
382 	return error;
383 }
384 
385 static int
386 ext4_xattr_ibody_list(struct inode *inode, char *buffer, size_t buffer_size)
387 {
388 	struct ext4_xattr_ibody_header *header;
389 	struct ext4_inode *raw_inode;
390 	struct ext4_iloc iloc;
391 	void *end;
392 	int error;
393 
394 	if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR))
395 		return 0;
396 	error = ext4_get_inode_loc(inode, &iloc);
397 	if (error)
398 		return error;
399 	raw_inode = ext4_raw_inode(&iloc);
400 	header = IHDR(inode, raw_inode);
401 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
402 	error = ext4_xattr_check_names(IFIRST(header), end);
403 	if (error)
404 		goto cleanup;
405 	error = ext4_xattr_list_entries(inode, IFIRST(header),
406 					buffer, buffer_size);
407 
408 cleanup:
409 	brelse(iloc.bh);
410 	return error;
411 }
412 
413 /*
414  * ext4_xattr_list()
415  *
416  * Copy a list of attribute names into the buffer
417  * provided, or compute the buffer size required.
418  * Buffer is NULL to compute the size of the buffer required.
419  *
420  * Returns a negative error number on failure, or the number of bytes
421  * used / required on success.
422  */
423 int
424 ext4_xattr_list(struct inode *inode, char *buffer, size_t buffer_size)
425 {
426 	int i_error, b_error;
427 
428 	down_read(&EXT4_I(inode)->xattr_sem);
429 	i_error = ext4_xattr_ibody_list(inode, buffer, buffer_size);
430 	if (i_error < 0) {
431 		b_error = 0;
432 	} else {
433 		if (buffer) {
434 			buffer += i_error;
435 			buffer_size -= i_error;
436 		}
437 		b_error = ext4_xattr_block_list(inode, buffer, buffer_size);
438 		if (b_error < 0)
439 			i_error = 0;
440 	}
441 	up_read(&EXT4_I(inode)->xattr_sem);
442 	return i_error + b_error;
443 }
444 
445 /*
446  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
447  * not set, set it.
448  */
449 static void ext4_xattr_update_super_block(handle_t *handle,
450 					  struct super_block *sb)
451 {
452 	if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_EXT_ATTR))
453 		return;
454 
455 	if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
456 		EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_EXT_ATTR);
457 		sb->s_dirt = 1;
458 		ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
459 	}
460 }
461 
462 /*
463  * Release the xattr block BH: If the reference count is > 1, decrement
464  * it; otherwise free the block.
465  */
466 static void
467 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
468 			 struct buffer_head *bh)
469 {
470 	struct mb_cache_entry *ce = NULL;
471 	int error = 0;
472 
473 	ce = mb_cache_entry_get(ext4_xattr_cache, bh->b_bdev, bh->b_blocknr);
474 	error = ext4_journal_get_write_access(handle, bh);
475 	if (error)
476 		goto out;
477 
478 	lock_buffer(bh);
479 	if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
480 		ea_bdebug(bh, "refcount now=0; freeing");
481 		if (ce)
482 			mb_cache_entry_free(ce);
483 		ext4_free_blocks(handle, inode, bh->b_blocknr, 1);
484 		get_bh(bh);
485 		ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
486 	} else {
487 		BHDR(bh)->h_refcount = cpu_to_le32(
488 				le32_to_cpu(BHDR(bh)->h_refcount) - 1);
489 		error = ext4_journal_dirty_metadata(handle, bh);
490 		if (IS_SYNC(inode))
491 			handle->h_sync = 1;
492 		DQUOT_FREE_BLOCK(inode, 1);
493 		ea_bdebug(bh, "refcount now=%d; releasing",
494 			  le32_to_cpu(BHDR(bh)->h_refcount));
495 		if (ce)
496 			mb_cache_entry_release(ce);
497 	}
498 	unlock_buffer(bh);
499 out:
500 	ext4_std_error(inode->i_sb, error);
501 	return;
502 }
503 
504 /*
505  * Find the available free space for EAs. This also returns the total number of
506  * bytes used by EA entries.
507  */
508 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
509 				    size_t *min_offs, void *base, int *total)
510 {
511 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
512 		*total += EXT4_XATTR_LEN(last->e_name_len);
513 		if (!last->e_value_block && last->e_value_size) {
514 			size_t offs = le16_to_cpu(last->e_value_offs);
515 			if (offs < *min_offs)
516 				*min_offs = offs;
517 		}
518 	}
519 	return (*min_offs - ((void *)last - base) - sizeof(__u32));
520 }
521 
522 struct ext4_xattr_info {
523 	int name_index;
524 	const char *name;
525 	const void *value;
526 	size_t value_len;
527 };
528 
529 struct ext4_xattr_search {
530 	struct ext4_xattr_entry *first;
531 	void *base;
532 	void *end;
533 	struct ext4_xattr_entry *here;
534 	int not_found;
535 };
536 
537 static int
538 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
539 {
540 	struct ext4_xattr_entry *last;
541 	size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
542 
543 	/* Compute min_offs and last. */
544 	last = s->first;
545 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
546 		if (!last->e_value_block && last->e_value_size) {
547 			size_t offs = le16_to_cpu(last->e_value_offs);
548 			if (offs < min_offs)
549 				min_offs = offs;
550 		}
551 	}
552 	free = min_offs - ((void *)last - s->base) - sizeof(__u32);
553 	if (!s->not_found) {
554 		if (!s->here->e_value_block && s->here->e_value_size) {
555 			size_t size = le32_to_cpu(s->here->e_value_size);
556 			free += EXT4_XATTR_SIZE(size);
557 		}
558 		free += EXT4_XATTR_LEN(name_len);
559 	}
560 	if (i->value) {
561 		if (free < EXT4_XATTR_SIZE(i->value_len) ||
562 		    free < EXT4_XATTR_LEN(name_len) +
563 			   EXT4_XATTR_SIZE(i->value_len))
564 			return -ENOSPC;
565 	}
566 
567 	if (i->value && s->not_found) {
568 		/* Insert the new name. */
569 		size_t size = EXT4_XATTR_LEN(name_len);
570 		size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
571 		memmove((void *)s->here + size, s->here, rest);
572 		memset(s->here, 0, size);
573 		s->here->e_name_index = i->name_index;
574 		s->here->e_name_len = name_len;
575 		memcpy(s->here->e_name, i->name, name_len);
576 	} else {
577 		if (!s->here->e_value_block && s->here->e_value_size) {
578 			void *first_val = s->base + min_offs;
579 			size_t offs = le16_to_cpu(s->here->e_value_offs);
580 			void *val = s->base + offs;
581 			size_t size = EXT4_XATTR_SIZE(
582 				le32_to_cpu(s->here->e_value_size));
583 
584 			if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
585 				/* The old and the new value have the same
586 				   size. Just replace. */
587 				s->here->e_value_size =
588 					cpu_to_le32(i->value_len);
589 				memset(val + size - EXT4_XATTR_PAD, 0,
590 				       EXT4_XATTR_PAD); /* Clear pad bytes. */
591 				memcpy(val, i->value, i->value_len);
592 				return 0;
593 			}
594 
595 			/* Remove the old value. */
596 			memmove(first_val + size, first_val, val - first_val);
597 			memset(first_val, 0, size);
598 			s->here->e_value_size = 0;
599 			s->here->e_value_offs = 0;
600 			min_offs += size;
601 
602 			/* Adjust all value offsets. */
603 			last = s->first;
604 			while (!IS_LAST_ENTRY(last)) {
605 				size_t o = le16_to_cpu(last->e_value_offs);
606 				if (!last->e_value_block &&
607 				    last->e_value_size && o < offs)
608 					last->e_value_offs =
609 						cpu_to_le16(o + size);
610 				last = EXT4_XATTR_NEXT(last);
611 			}
612 		}
613 		if (!i->value) {
614 			/* Remove the old name. */
615 			size_t size = EXT4_XATTR_LEN(name_len);
616 			last = ENTRY((void *)last - size);
617 			memmove(s->here, (void *)s->here + size,
618 				(void *)last - (void *)s->here + sizeof(__u32));
619 			memset(last, 0, size);
620 		}
621 	}
622 
623 	if (i->value) {
624 		/* Insert the new value. */
625 		s->here->e_value_size = cpu_to_le32(i->value_len);
626 		if (i->value_len) {
627 			size_t size = EXT4_XATTR_SIZE(i->value_len);
628 			void *val = s->base + min_offs - size;
629 			s->here->e_value_offs = cpu_to_le16(min_offs - size);
630 			memset(val + size - EXT4_XATTR_PAD, 0,
631 			       EXT4_XATTR_PAD); /* Clear the pad bytes. */
632 			memcpy(val, i->value, i->value_len);
633 		}
634 	}
635 	return 0;
636 }
637 
638 struct ext4_xattr_block_find {
639 	struct ext4_xattr_search s;
640 	struct buffer_head *bh;
641 };
642 
643 static int
644 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
645 		      struct ext4_xattr_block_find *bs)
646 {
647 	struct super_block *sb = inode->i_sb;
648 	int error;
649 
650 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
651 		  i->name_index, i->name, i->value, (long)i->value_len);
652 
653 	if (EXT4_I(inode)->i_file_acl) {
654 		/* The inode already has an extended attribute block. */
655 		bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
656 		error = -EIO;
657 		if (!bs->bh)
658 			goto cleanup;
659 		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
660 			atomic_read(&(bs->bh->b_count)),
661 			le32_to_cpu(BHDR(bs->bh)->h_refcount));
662 		if (ext4_xattr_check_block(bs->bh)) {
663 			ext4_error(sb, __FUNCTION__,
664 				"inode %lu: bad block %llu", inode->i_ino,
665 				EXT4_I(inode)->i_file_acl);
666 			error = -EIO;
667 			goto cleanup;
668 		}
669 		/* Find the named attribute. */
670 		bs->s.base = BHDR(bs->bh);
671 		bs->s.first = BFIRST(bs->bh);
672 		bs->s.end = bs->bh->b_data + bs->bh->b_size;
673 		bs->s.here = bs->s.first;
674 		error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
675 					      i->name, bs->bh->b_size, 1);
676 		if (error && error != -ENODATA)
677 			goto cleanup;
678 		bs->s.not_found = error;
679 	}
680 	error = 0;
681 
682 cleanup:
683 	return error;
684 }
685 
686 static int
687 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
688 		     struct ext4_xattr_info *i,
689 		     struct ext4_xattr_block_find *bs)
690 {
691 	struct super_block *sb = inode->i_sb;
692 	struct buffer_head *new_bh = NULL;
693 	struct ext4_xattr_search *s = &bs->s;
694 	struct mb_cache_entry *ce = NULL;
695 	int error = 0;
696 
697 #define header(x) ((struct ext4_xattr_header *)(x))
698 
699 	if (i->value && i->value_len > sb->s_blocksize)
700 		return -ENOSPC;
701 	if (s->base) {
702 		ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
703 					bs->bh->b_blocknr);
704 		error = ext4_journal_get_write_access(handle, bs->bh);
705 		if (error)
706 			goto cleanup;
707 		lock_buffer(bs->bh);
708 
709 		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
710 			if (ce) {
711 				mb_cache_entry_free(ce);
712 				ce = NULL;
713 			}
714 			ea_bdebug(bs->bh, "modifying in-place");
715 			error = ext4_xattr_set_entry(i, s);
716 			if (!error) {
717 				if (!IS_LAST_ENTRY(s->first))
718 					ext4_xattr_rehash(header(s->base),
719 							  s->here);
720 				ext4_xattr_cache_insert(bs->bh);
721 			}
722 			unlock_buffer(bs->bh);
723 			if (error == -EIO)
724 				goto bad_block;
725 			if (!error)
726 				error = ext4_journal_dirty_metadata(handle,
727 								    bs->bh);
728 			if (error)
729 				goto cleanup;
730 			goto inserted;
731 		} else {
732 			int offset = (char *)s->here - bs->bh->b_data;
733 
734 			unlock_buffer(bs->bh);
735 			jbd2_journal_release_buffer(handle, bs->bh);
736 			if (ce) {
737 				mb_cache_entry_release(ce);
738 				ce = NULL;
739 			}
740 			ea_bdebug(bs->bh, "cloning");
741 			s->base = kmalloc(bs->bh->b_size, GFP_KERNEL);
742 			error = -ENOMEM;
743 			if (s->base == NULL)
744 				goto cleanup;
745 			memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
746 			s->first = ENTRY(header(s->base)+1);
747 			header(s->base)->h_refcount = cpu_to_le32(1);
748 			s->here = ENTRY(s->base + offset);
749 			s->end = s->base + bs->bh->b_size;
750 		}
751 	} else {
752 		/* Allocate a buffer where we construct the new block. */
753 		s->base = kmalloc(sb->s_blocksize, GFP_KERNEL);
754 		/* assert(header == s->base) */
755 		error = -ENOMEM;
756 		if (s->base == NULL)
757 			goto cleanup;
758 		memset(s->base, 0, sb->s_blocksize);
759 		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
760 		header(s->base)->h_blocks = cpu_to_le32(1);
761 		header(s->base)->h_refcount = cpu_to_le32(1);
762 		s->first = ENTRY(header(s->base)+1);
763 		s->here = ENTRY(header(s->base)+1);
764 		s->end = s->base + sb->s_blocksize;
765 	}
766 
767 	error = ext4_xattr_set_entry(i, s);
768 	if (error == -EIO)
769 		goto bad_block;
770 	if (error)
771 		goto cleanup;
772 	if (!IS_LAST_ENTRY(s->first))
773 		ext4_xattr_rehash(header(s->base), s->here);
774 
775 inserted:
776 	if (!IS_LAST_ENTRY(s->first)) {
777 		new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
778 		if (new_bh) {
779 			/* We found an identical block in the cache. */
780 			if (new_bh == bs->bh)
781 				ea_bdebug(new_bh, "keeping");
782 			else {
783 				/* The old block is released after updating
784 				   the inode. */
785 				error = -EDQUOT;
786 				if (DQUOT_ALLOC_BLOCK(inode, 1))
787 					goto cleanup;
788 				error = ext4_journal_get_write_access(handle,
789 								      new_bh);
790 				if (error)
791 					goto cleanup_dquot;
792 				lock_buffer(new_bh);
793 				BHDR(new_bh)->h_refcount = cpu_to_le32(1 +
794 					le32_to_cpu(BHDR(new_bh)->h_refcount));
795 				ea_bdebug(new_bh, "reusing; refcount now=%d",
796 					le32_to_cpu(BHDR(new_bh)->h_refcount));
797 				unlock_buffer(new_bh);
798 				error = ext4_journal_dirty_metadata(handle,
799 								    new_bh);
800 				if (error)
801 					goto cleanup_dquot;
802 			}
803 			mb_cache_entry_release(ce);
804 			ce = NULL;
805 		} else if (bs->bh && s->base == bs->bh->b_data) {
806 			/* We were modifying this block in-place. */
807 			ea_bdebug(bs->bh, "keeping this block");
808 			new_bh = bs->bh;
809 			get_bh(new_bh);
810 		} else {
811 			/* We need to allocate a new block */
812 			ext4_fsblk_t goal = le32_to_cpu(
813 					EXT4_SB(sb)->s_es->s_first_data_block) +
814 				(ext4_fsblk_t)EXT4_I(inode)->i_block_group *
815 				EXT4_BLOCKS_PER_GROUP(sb);
816 			ext4_fsblk_t block = ext4_new_block(handle, inode,
817 							goal, &error);
818 			if (error)
819 				goto cleanup;
820 			ea_idebug(inode, "creating block %d", block);
821 
822 			new_bh = sb_getblk(sb, block);
823 			if (!new_bh) {
824 getblk_failed:
825 				ext4_free_blocks(handle, inode, block, 1);
826 				error = -EIO;
827 				goto cleanup;
828 			}
829 			lock_buffer(new_bh);
830 			error = ext4_journal_get_create_access(handle, new_bh);
831 			if (error) {
832 				unlock_buffer(new_bh);
833 				goto getblk_failed;
834 			}
835 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
836 			set_buffer_uptodate(new_bh);
837 			unlock_buffer(new_bh);
838 			ext4_xattr_cache_insert(new_bh);
839 			error = ext4_journal_dirty_metadata(handle, new_bh);
840 			if (error)
841 				goto cleanup;
842 		}
843 	}
844 
845 	/* Update the inode. */
846 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
847 
848 	/* Drop the previous xattr block. */
849 	if (bs->bh && bs->bh != new_bh)
850 		ext4_xattr_release_block(handle, inode, bs->bh);
851 	error = 0;
852 
853 cleanup:
854 	if (ce)
855 		mb_cache_entry_release(ce);
856 	brelse(new_bh);
857 	if (!(bs->bh && s->base == bs->bh->b_data))
858 		kfree(s->base);
859 
860 	return error;
861 
862 cleanup_dquot:
863 	DQUOT_FREE_BLOCK(inode, 1);
864 	goto cleanup;
865 
866 bad_block:
867 	ext4_error(inode->i_sb, __FUNCTION__,
868 		   "inode %lu: bad block %llu", inode->i_ino,
869 		   EXT4_I(inode)->i_file_acl);
870 	goto cleanup;
871 
872 #undef header
873 }
874 
875 struct ext4_xattr_ibody_find {
876 	struct ext4_xattr_search s;
877 	struct ext4_iloc iloc;
878 };
879 
880 static int
881 ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
882 		      struct ext4_xattr_ibody_find *is)
883 {
884 	struct ext4_xattr_ibody_header *header;
885 	struct ext4_inode *raw_inode;
886 	int error;
887 
888 	if (EXT4_I(inode)->i_extra_isize == 0)
889 		return 0;
890 	raw_inode = ext4_raw_inode(&is->iloc);
891 	header = IHDR(inode, raw_inode);
892 	is->s.base = is->s.first = IFIRST(header);
893 	is->s.here = is->s.first;
894 	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
895 	if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
896 		error = ext4_xattr_check_names(IFIRST(header), is->s.end);
897 		if (error)
898 			return error;
899 		/* Find the named attribute. */
900 		error = ext4_xattr_find_entry(&is->s.here, i->name_index,
901 					      i->name, is->s.end -
902 					      (void *)is->s.base, 0);
903 		if (error && error != -ENODATA)
904 			return error;
905 		is->s.not_found = error;
906 	}
907 	return 0;
908 }
909 
910 static int
911 ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
912 		     struct ext4_xattr_info *i,
913 		     struct ext4_xattr_ibody_find *is)
914 {
915 	struct ext4_xattr_ibody_header *header;
916 	struct ext4_xattr_search *s = &is->s;
917 	int error;
918 
919 	if (EXT4_I(inode)->i_extra_isize == 0)
920 		return -ENOSPC;
921 	error = ext4_xattr_set_entry(i, s);
922 	if (error)
923 		return error;
924 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
925 	if (!IS_LAST_ENTRY(s->first)) {
926 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
927 		EXT4_I(inode)->i_state |= EXT4_STATE_XATTR;
928 	} else {
929 		header->h_magic = cpu_to_le32(0);
930 		EXT4_I(inode)->i_state &= ~EXT4_STATE_XATTR;
931 	}
932 	return 0;
933 }
934 
935 /*
936  * ext4_xattr_set_handle()
937  *
938  * Create, replace or remove an extended attribute for this inode. Buffer
939  * is NULL to remove an existing extended attribute, and non-NULL to
940  * either replace an existing extended attribute, or create a new extended
941  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
942  * specify that an extended attribute must exist and must not exist
943  * previous to the call, respectively.
944  *
945  * Returns 0, or a negative error number on failure.
946  */
947 int
948 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
949 		      const char *name, const void *value, size_t value_len,
950 		      int flags)
951 {
952 	struct ext4_xattr_info i = {
953 		.name_index = name_index,
954 		.name = name,
955 		.value = value,
956 		.value_len = value_len,
957 
958 	};
959 	struct ext4_xattr_ibody_find is = {
960 		.s = { .not_found = -ENODATA, },
961 	};
962 	struct ext4_xattr_block_find bs = {
963 		.s = { .not_found = -ENODATA, },
964 	};
965 	int error;
966 
967 	if (!name)
968 		return -EINVAL;
969 	if (strlen(name) > 255)
970 		return -ERANGE;
971 	down_write(&EXT4_I(inode)->xattr_sem);
972 	error = ext4_get_inode_loc(inode, &is.iloc);
973 	if (error)
974 		goto cleanup;
975 
976 	if (EXT4_I(inode)->i_state & EXT4_STATE_NEW) {
977 		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
978 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
979 		EXT4_I(inode)->i_state &= ~EXT4_STATE_NEW;
980 	}
981 
982 	error = ext4_xattr_ibody_find(inode, &i, &is);
983 	if (error)
984 		goto cleanup;
985 	if (is.s.not_found)
986 		error = ext4_xattr_block_find(inode, &i, &bs);
987 	if (error)
988 		goto cleanup;
989 	if (is.s.not_found && bs.s.not_found) {
990 		error = -ENODATA;
991 		if (flags & XATTR_REPLACE)
992 			goto cleanup;
993 		error = 0;
994 		if (!value)
995 			goto cleanup;
996 	} else {
997 		error = -EEXIST;
998 		if (flags & XATTR_CREATE)
999 			goto cleanup;
1000 	}
1001 	error = ext4_journal_get_write_access(handle, is.iloc.bh);
1002 	if (error)
1003 		goto cleanup;
1004 	if (!value) {
1005 		if (!is.s.not_found)
1006 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1007 		else if (!bs.s.not_found)
1008 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1009 	} else {
1010 		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1011 		if (!error && !bs.s.not_found) {
1012 			i.value = NULL;
1013 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1014 		} else if (error == -ENOSPC) {
1015 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1016 			if (error)
1017 				goto cleanup;
1018 			if (!is.s.not_found) {
1019 				i.value = NULL;
1020 				error = ext4_xattr_ibody_set(handle, inode, &i,
1021 							     &is);
1022 			}
1023 		}
1024 	}
1025 	if (!error) {
1026 		ext4_xattr_update_super_block(handle, inode->i_sb);
1027 		inode->i_ctime = ext4_current_time(inode);
1028 		if (!value)
1029 			EXT4_I(inode)->i_state &= ~EXT4_STATE_NO_EXPAND;
1030 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1031 		/*
1032 		 * The bh is consumed by ext4_mark_iloc_dirty, even with
1033 		 * error != 0.
1034 		 */
1035 		is.iloc.bh = NULL;
1036 		if (IS_SYNC(inode))
1037 			handle->h_sync = 1;
1038 	}
1039 
1040 cleanup:
1041 	brelse(is.iloc.bh);
1042 	brelse(bs.bh);
1043 	up_write(&EXT4_I(inode)->xattr_sem);
1044 	return error;
1045 }
1046 
1047 /*
1048  * ext4_xattr_set()
1049  *
1050  * Like ext4_xattr_set_handle, but start from an inode. This extended
1051  * attribute modification is a filesystem transaction by itself.
1052  *
1053  * Returns 0, or a negative error number on failure.
1054  */
1055 int
1056 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1057 	       const void *value, size_t value_len, int flags)
1058 {
1059 	handle_t *handle;
1060 	int error, retries = 0;
1061 
1062 retry:
1063 	handle = ext4_journal_start(inode, EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
1064 	if (IS_ERR(handle)) {
1065 		error = PTR_ERR(handle);
1066 	} else {
1067 		int error2;
1068 
1069 		error = ext4_xattr_set_handle(handle, inode, name_index, name,
1070 					      value, value_len, flags);
1071 		error2 = ext4_journal_stop(handle);
1072 		if (error == -ENOSPC &&
1073 		    ext4_should_retry_alloc(inode->i_sb, &retries))
1074 			goto retry;
1075 		if (error == 0)
1076 			error = error2;
1077 	}
1078 
1079 	return error;
1080 }
1081 
1082 /*
1083  * Shift the EA entries in the inode to create space for the increased
1084  * i_extra_isize.
1085  */
1086 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1087 				     int value_offs_shift, void *to,
1088 				     void *from, size_t n, int blocksize)
1089 {
1090 	struct ext4_xattr_entry *last = entry;
1091 	int new_offs;
1092 
1093 	/* Adjust the value offsets of the entries */
1094 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1095 		if (!last->e_value_block && last->e_value_size) {
1096 			new_offs = le16_to_cpu(last->e_value_offs) +
1097 							value_offs_shift;
1098 			BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
1099 				 > blocksize);
1100 			last->e_value_offs = cpu_to_le16(new_offs);
1101 		}
1102 	}
1103 	/* Shift the entries by n bytes */
1104 	memmove(to, from, n);
1105 }
1106 
1107 /*
1108  * Expand an inode by new_extra_isize bytes when EAs are present.
1109  * Returns 0 on success or negative error number on failure.
1110  */
1111 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1112 			       struct ext4_inode *raw_inode, handle_t *handle)
1113 {
1114 	struct ext4_xattr_ibody_header *header;
1115 	struct ext4_xattr_entry *entry, *last, *first;
1116 	struct buffer_head *bh = NULL;
1117 	struct ext4_xattr_ibody_find *is = NULL;
1118 	struct ext4_xattr_block_find *bs = NULL;
1119 	char *buffer = NULL, *b_entry_name = NULL;
1120 	size_t min_offs, free;
1121 	int total_ino, total_blk;
1122 	void *base, *start, *end;
1123 	int extra_isize = 0, error = 0, tried_min_extra_isize = 0;
1124 	int s_min_extra_isize = EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize;
1125 
1126 	down_write(&EXT4_I(inode)->xattr_sem);
1127 retry:
1128 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) {
1129 		up_write(&EXT4_I(inode)->xattr_sem);
1130 		return 0;
1131 	}
1132 
1133 	header = IHDR(inode, raw_inode);
1134 	entry = IFIRST(header);
1135 
1136 	/*
1137 	 * Check if enough free space is available in the inode to shift the
1138 	 * entries ahead by new_extra_isize.
1139 	 */
1140 
1141 	base = start = entry;
1142 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1143 	min_offs = end - base;
1144 	last = entry;
1145 	total_ino = sizeof(struct ext4_xattr_ibody_header);
1146 
1147 	free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
1148 	if (free >= new_extra_isize) {
1149 		entry = IFIRST(header);
1150 		ext4_xattr_shift_entries(entry,	EXT4_I(inode)->i_extra_isize
1151 				- new_extra_isize, (void *)raw_inode +
1152 				EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1153 				(void *)header, total_ino,
1154 				inode->i_sb->s_blocksize);
1155 		EXT4_I(inode)->i_extra_isize = new_extra_isize;
1156 		error = 0;
1157 		goto cleanup;
1158 	}
1159 
1160 	/*
1161 	 * Enough free space isn't available in the inode, check if
1162 	 * EA block can hold new_extra_isize bytes.
1163 	 */
1164 	if (EXT4_I(inode)->i_file_acl) {
1165 		bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1166 		error = -EIO;
1167 		if (!bh)
1168 			goto cleanup;
1169 		if (ext4_xattr_check_block(bh)) {
1170 			ext4_error(inode->i_sb, __FUNCTION__,
1171 				"inode %lu: bad block %llu", inode->i_ino,
1172 				EXT4_I(inode)->i_file_acl);
1173 			error = -EIO;
1174 			goto cleanup;
1175 		}
1176 		base = BHDR(bh);
1177 		first = BFIRST(bh);
1178 		end = bh->b_data + bh->b_size;
1179 		min_offs = end - base;
1180 		free = ext4_xattr_free_space(first, &min_offs, base,
1181 					     &total_blk);
1182 		if (free < new_extra_isize) {
1183 			if (!tried_min_extra_isize && s_min_extra_isize) {
1184 				tried_min_extra_isize++;
1185 				new_extra_isize = s_min_extra_isize;
1186 				brelse(bh);
1187 				goto retry;
1188 			}
1189 			error = -1;
1190 			goto cleanup;
1191 		}
1192 	} else {
1193 		free = inode->i_sb->s_blocksize;
1194 	}
1195 
1196 	while (new_extra_isize > 0) {
1197 		size_t offs, size, entry_size;
1198 		struct ext4_xattr_entry *small_entry = NULL;
1199 		struct ext4_xattr_info i = {
1200 			.value = NULL,
1201 			.value_len = 0,
1202 		};
1203 		unsigned int total_size;  /* EA entry size + value size */
1204 		unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
1205 		unsigned int min_total_size = ~0U;
1206 
1207 		is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1208 		bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1209 		if (!is || !bs) {
1210 			error = -ENOMEM;
1211 			goto cleanup;
1212 		}
1213 
1214 		is->s.not_found = -ENODATA;
1215 		bs->s.not_found = -ENODATA;
1216 		is->iloc.bh = NULL;
1217 		bs->bh = NULL;
1218 
1219 		last = IFIRST(header);
1220 		/* Find the entry best suited to be pushed into EA block */
1221 		entry = NULL;
1222 		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1223 			total_size =
1224 			EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1225 					EXT4_XATTR_LEN(last->e_name_len);
1226 			if (total_size <= free && total_size < min_total_size) {
1227 				if (total_size < new_extra_isize) {
1228 					small_entry = last;
1229 				} else {
1230 					entry = last;
1231 					min_total_size = total_size;
1232 				}
1233 			}
1234 		}
1235 
1236 		if (entry == NULL) {
1237 			if (small_entry) {
1238 				entry = small_entry;
1239 			} else {
1240 				if (!tried_min_extra_isize &&
1241 				    s_min_extra_isize) {
1242 					tried_min_extra_isize++;
1243 					new_extra_isize = s_min_extra_isize;
1244 					goto retry;
1245 				}
1246 				error = -1;
1247 				goto cleanup;
1248 			}
1249 		}
1250 		offs = le16_to_cpu(entry->e_value_offs);
1251 		size = le32_to_cpu(entry->e_value_size);
1252 		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1253 		i.name_index = entry->e_name_index,
1254 		buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
1255 		b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1256 		if (!buffer || !b_entry_name) {
1257 			error = -ENOMEM;
1258 			goto cleanup;
1259 		}
1260 		/* Save the entry name and the entry value */
1261 		memcpy(buffer, (void *)IFIRST(header) + offs,
1262 		       EXT4_XATTR_SIZE(size));
1263 		memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1264 		b_entry_name[entry->e_name_len] = '\0';
1265 		i.name = b_entry_name;
1266 
1267 		error = ext4_get_inode_loc(inode, &is->iloc);
1268 		if (error)
1269 			goto cleanup;
1270 
1271 		error = ext4_xattr_ibody_find(inode, &i, is);
1272 		if (error)
1273 			goto cleanup;
1274 
1275 		/* Remove the chosen entry from the inode */
1276 		error = ext4_xattr_ibody_set(handle, inode, &i, is);
1277 
1278 		entry = IFIRST(header);
1279 		if (entry_size + EXT4_XATTR_SIZE(size) >= new_extra_isize)
1280 			shift_bytes = new_extra_isize;
1281 		else
1282 			shift_bytes = entry_size + size;
1283 		/* Adjust the offsets and shift the remaining entries ahead */
1284 		ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize -
1285 			shift_bytes, (void *)raw_inode +
1286 			EXT4_GOOD_OLD_INODE_SIZE + extra_isize + shift_bytes,
1287 			(void *)header, total_ino - entry_size,
1288 			inode->i_sb->s_blocksize);
1289 
1290 		extra_isize += shift_bytes;
1291 		new_extra_isize -= shift_bytes;
1292 		EXT4_I(inode)->i_extra_isize = extra_isize;
1293 
1294 		i.name = b_entry_name;
1295 		i.value = buffer;
1296 		i.value_len = cpu_to_le32(size);
1297 		error = ext4_xattr_block_find(inode, &i, bs);
1298 		if (error)
1299 			goto cleanup;
1300 
1301 		/* Add entry which was removed from the inode into the block */
1302 		error = ext4_xattr_block_set(handle, inode, &i, bs);
1303 		if (error)
1304 			goto cleanup;
1305 		kfree(b_entry_name);
1306 		kfree(buffer);
1307 		brelse(is->iloc.bh);
1308 		kfree(is);
1309 		kfree(bs);
1310 	}
1311 	brelse(bh);
1312 	up_write(&EXT4_I(inode)->xattr_sem);
1313 	return 0;
1314 
1315 cleanup:
1316 	kfree(b_entry_name);
1317 	kfree(buffer);
1318 	if (is)
1319 		brelse(is->iloc.bh);
1320 	kfree(is);
1321 	kfree(bs);
1322 	brelse(bh);
1323 	up_write(&EXT4_I(inode)->xattr_sem);
1324 	return error;
1325 }
1326 
1327 
1328 
1329 /*
1330  * ext4_xattr_delete_inode()
1331  *
1332  * Free extended attribute resources associated with this inode. This
1333  * is called immediately before an inode is freed. We have exclusive
1334  * access to the inode.
1335  */
1336 void
1337 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1338 {
1339 	struct buffer_head *bh = NULL;
1340 
1341 	if (!EXT4_I(inode)->i_file_acl)
1342 		goto cleanup;
1343 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1344 	if (!bh) {
1345 		ext4_error(inode->i_sb, __FUNCTION__,
1346 			"inode %lu: block %llu read error", inode->i_ino,
1347 			EXT4_I(inode)->i_file_acl);
1348 		goto cleanup;
1349 	}
1350 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1351 	    BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1352 		ext4_error(inode->i_sb, __FUNCTION__,
1353 			"inode %lu: bad block %llu", inode->i_ino,
1354 			EXT4_I(inode)->i_file_acl);
1355 		goto cleanup;
1356 	}
1357 	ext4_xattr_release_block(handle, inode, bh);
1358 	EXT4_I(inode)->i_file_acl = 0;
1359 
1360 cleanup:
1361 	brelse(bh);
1362 }
1363 
1364 /*
1365  * ext4_xattr_put_super()
1366  *
1367  * This is called when a file system is unmounted.
1368  */
1369 void
1370 ext4_xattr_put_super(struct super_block *sb)
1371 {
1372 	mb_cache_shrink(sb->s_bdev);
1373 }
1374 
1375 /*
1376  * ext4_xattr_cache_insert()
1377  *
1378  * Create a new entry in the extended attribute cache, and insert
1379  * it unless such an entry is already in the cache.
1380  *
1381  * Returns 0, or a negative error number on failure.
1382  */
1383 static void
1384 ext4_xattr_cache_insert(struct buffer_head *bh)
1385 {
1386 	__u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1387 	struct mb_cache_entry *ce;
1388 	int error;
1389 
1390 	ce = mb_cache_entry_alloc(ext4_xattr_cache);
1391 	if (!ce) {
1392 		ea_bdebug(bh, "out of memory");
1393 		return;
1394 	}
1395 	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash);
1396 	if (error) {
1397 		mb_cache_entry_free(ce);
1398 		if (error == -EBUSY) {
1399 			ea_bdebug(bh, "already in cache");
1400 			error = 0;
1401 		}
1402 	} else {
1403 		ea_bdebug(bh, "inserting [%x]", (int)hash);
1404 		mb_cache_entry_release(ce);
1405 	}
1406 }
1407 
1408 /*
1409  * ext4_xattr_cmp()
1410  *
1411  * Compare two extended attribute blocks for equality.
1412  *
1413  * Returns 0 if the blocks are equal, 1 if they differ, and
1414  * a negative error number on errors.
1415  */
1416 static int
1417 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1418 	       struct ext4_xattr_header *header2)
1419 {
1420 	struct ext4_xattr_entry *entry1, *entry2;
1421 
1422 	entry1 = ENTRY(header1+1);
1423 	entry2 = ENTRY(header2+1);
1424 	while (!IS_LAST_ENTRY(entry1)) {
1425 		if (IS_LAST_ENTRY(entry2))
1426 			return 1;
1427 		if (entry1->e_hash != entry2->e_hash ||
1428 		    entry1->e_name_index != entry2->e_name_index ||
1429 		    entry1->e_name_len != entry2->e_name_len ||
1430 		    entry1->e_value_size != entry2->e_value_size ||
1431 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1432 			return 1;
1433 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1434 			return -EIO;
1435 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1436 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1437 			   le32_to_cpu(entry1->e_value_size)))
1438 			return 1;
1439 
1440 		entry1 = EXT4_XATTR_NEXT(entry1);
1441 		entry2 = EXT4_XATTR_NEXT(entry2);
1442 	}
1443 	if (!IS_LAST_ENTRY(entry2))
1444 		return 1;
1445 	return 0;
1446 }
1447 
1448 /*
1449  * ext4_xattr_cache_find()
1450  *
1451  * Find an identical extended attribute block.
1452  *
1453  * Returns a pointer to the block found, or NULL if such a block was
1454  * not found or an error occurred.
1455  */
1456 static struct buffer_head *
1457 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1458 		      struct mb_cache_entry **pce)
1459 {
1460 	__u32 hash = le32_to_cpu(header->h_hash);
1461 	struct mb_cache_entry *ce;
1462 
1463 	if (!header->h_hash)
1464 		return NULL;  /* never share */
1465 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1466 again:
1467 	ce = mb_cache_entry_find_first(ext4_xattr_cache, 0,
1468 				       inode->i_sb->s_bdev, hash);
1469 	while (ce) {
1470 		struct buffer_head *bh;
1471 
1472 		if (IS_ERR(ce)) {
1473 			if (PTR_ERR(ce) == -EAGAIN)
1474 				goto again;
1475 			break;
1476 		}
1477 		bh = sb_bread(inode->i_sb, ce->e_block);
1478 		if (!bh) {
1479 			ext4_error(inode->i_sb, __FUNCTION__,
1480 				"inode %lu: block %lu read error",
1481 				inode->i_ino, (unsigned long) ce->e_block);
1482 		} else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1483 				EXT4_XATTR_REFCOUNT_MAX) {
1484 			ea_idebug(inode, "block %lu refcount %d>=%d",
1485 				  (unsigned long) ce->e_block,
1486 				  le32_to_cpu(BHDR(bh)->h_refcount),
1487 					  EXT4_XATTR_REFCOUNT_MAX);
1488 		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1489 			*pce = ce;
1490 			return bh;
1491 		}
1492 		brelse(bh);
1493 		ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash);
1494 	}
1495 	return NULL;
1496 }
1497 
1498 #define NAME_HASH_SHIFT 5
1499 #define VALUE_HASH_SHIFT 16
1500 
1501 /*
1502  * ext4_xattr_hash_entry()
1503  *
1504  * Compute the hash of an extended attribute.
1505  */
1506 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1507 					 struct ext4_xattr_entry *entry)
1508 {
1509 	__u32 hash = 0;
1510 	char *name = entry->e_name;
1511 	int n;
1512 
1513 	for (n=0; n < entry->e_name_len; n++) {
1514 		hash = (hash << NAME_HASH_SHIFT) ^
1515 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1516 		       *name++;
1517 	}
1518 
1519 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1520 		__le32 *value = (__le32 *)((char *)header +
1521 			le16_to_cpu(entry->e_value_offs));
1522 		for (n = (le32_to_cpu(entry->e_value_size) +
1523 		     EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1524 			hash = (hash << VALUE_HASH_SHIFT) ^
1525 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1526 			       le32_to_cpu(*value++);
1527 		}
1528 	}
1529 	entry->e_hash = cpu_to_le32(hash);
1530 }
1531 
1532 #undef NAME_HASH_SHIFT
1533 #undef VALUE_HASH_SHIFT
1534 
1535 #define BLOCK_HASH_SHIFT 16
1536 
1537 /*
1538  * ext4_xattr_rehash()
1539  *
1540  * Re-compute the extended attribute hash value after an entry has changed.
1541  */
1542 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1543 			      struct ext4_xattr_entry *entry)
1544 {
1545 	struct ext4_xattr_entry *here;
1546 	__u32 hash = 0;
1547 
1548 	ext4_xattr_hash_entry(header, entry);
1549 	here = ENTRY(header+1);
1550 	while (!IS_LAST_ENTRY(here)) {
1551 		if (!here->e_hash) {
1552 			/* Block is not shared if an entry's hash value == 0 */
1553 			hash = 0;
1554 			break;
1555 		}
1556 		hash = (hash << BLOCK_HASH_SHIFT) ^
1557 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1558 		       le32_to_cpu(here->e_hash);
1559 		here = EXT4_XATTR_NEXT(here);
1560 	}
1561 	header->h_hash = cpu_to_le32(hash);
1562 }
1563 
1564 #undef BLOCK_HASH_SHIFT
1565 
1566 int __init
1567 init_ext4_xattr(void)
1568 {
1569 	ext4_xattr_cache = mb_cache_create("ext4_xattr", NULL,
1570 		sizeof(struct mb_cache_entry) +
1571 		sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6);
1572 	if (!ext4_xattr_cache)
1573 		return -ENOMEM;
1574 	return 0;
1575 }
1576 
1577 void
1578 exit_ext4_xattr(void)
1579 {
1580 	if (ext4_xattr_cache)
1581 		mb_cache_destroy(ext4_xattr_cache);
1582 	ext4_xattr_cache = NULL;
1583 }
1584