xref: /openbmc/linux/fs/ext4/xattr.c (revision 545e4006)
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 <linux/rwsem.h>
59 #include "ext4_jbd2.h"
60 #include "ext4.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 static int ext4_xattr_list(struct inode *inode, char *buffer,
96 			   size_t buffer_size);
97 
98 static struct mb_cache *ext4_xattr_cache;
99 
100 static struct xattr_handler *ext4_xattr_handler_map[] = {
101 	[EXT4_XATTR_INDEX_USER]		     = &ext4_xattr_user_handler,
102 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
103 	[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &ext4_xattr_acl_access_handler,
104 	[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext4_xattr_acl_default_handler,
105 #endif
106 	[EXT4_XATTR_INDEX_TRUSTED]	     = &ext4_xattr_trusted_handler,
107 #ifdef CONFIG_EXT4DEV_FS_SECURITY
108 	[EXT4_XATTR_INDEX_SECURITY]	     = &ext4_xattr_security_handler,
109 #endif
110 };
111 
112 struct xattr_handler *ext4_xattr_handlers[] = {
113 	&ext4_xattr_user_handler,
114 	&ext4_xattr_trusted_handler,
115 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
116 	&ext4_xattr_acl_access_handler,
117 	&ext4_xattr_acl_default_handler,
118 #endif
119 #ifdef CONFIG_EXT4DEV_FS_SECURITY
120 	&ext4_xattr_security_handler,
121 #endif
122 	NULL
123 };
124 
125 static inline struct xattr_handler *
126 ext4_xattr_handler(int name_index)
127 {
128 	struct xattr_handler *handler = NULL;
129 
130 	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
131 		handler = ext4_xattr_handler_map[name_index];
132 	return handler;
133 }
134 
135 /*
136  * Inode operation listxattr()
137  *
138  * dentry->d_inode->i_mutex: don't care
139  */
140 ssize_t
141 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
142 {
143 	return ext4_xattr_list(dentry->d_inode, buffer, size);
144 }
145 
146 static int
147 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end)
148 {
149 	while (!IS_LAST_ENTRY(entry)) {
150 		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(entry);
151 		if ((void *)next >= end)
152 			return -EIO;
153 		entry = next;
154 	}
155 	return 0;
156 }
157 
158 static inline int
159 ext4_xattr_check_block(struct buffer_head *bh)
160 {
161 	int error;
162 
163 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
164 	    BHDR(bh)->h_blocks != cpu_to_le32(1))
165 		return -EIO;
166 	error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size);
167 	return error;
168 }
169 
170 static inline int
171 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
172 {
173 	size_t value_size = le32_to_cpu(entry->e_value_size);
174 
175 	if (entry->e_value_block != 0 || value_size > size ||
176 	    le16_to_cpu(entry->e_value_offs) + value_size > size)
177 		return -EIO;
178 	return 0;
179 }
180 
181 static int
182 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
183 		      const char *name, size_t size, int sorted)
184 {
185 	struct ext4_xattr_entry *entry;
186 	size_t name_len;
187 	int cmp = 1;
188 
189 	if (name == NULL)
190 		return -EINVAL;
191 	name_len = strlen(name);
192 	entry = *pentry;
193 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
194 		cmp = name_index - entry->e_name_index;
195 		if (!cmp)
196 			cmp = name_len - entry->e_name_len;
197 		if (!cmp)
198 			cmp = memcmp(name, entry->e_name, name_len);
199 		if (cmp <= 0 && (sorted || cmp == 0))
200 			break;
201 	}
202 	*pentry = entry;
203 	if (!cmp && ext4_xattr_check_entry(entry, size))
204 			return -EIO;
205 	return cmp ? -ENODATA : 0;
206 }
207 
208 static int
209 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
210 		     void *buffer, size_t buffer_size)
211 {
212 	struct buffer_head *bh = NULL;
213 	struct ext4_xattr_entry *entry;
214 	size_t size;
215 	int error;
216 
217 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
218 		  name_index, name, buffer, (long)buffer_size);
219 
220 	error = -ENODATA;
221 	if (!EXT4_I(inode)->i_file_acl)
222 		goto cleanup;
223 	ea_idebug(inode, "reading block %u", EXT4_I(inode)->i_file_acl);
224 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
225 	if (!bh)
226 		goto cleanup;
227 	ea_bdebug(bh, "b_count=%d, refcount=%d",
228 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
229 	if (ext4_xattr_check_block(bh)) {
230 bad_block:	ext4_error(inode->i_sb, __func__,
231 			   "inode %lu: bad block %llu", inode->i_ino,
232 			   EXT4_I(inode)->i_file_acl);
233 		error = -EIO;
234 		goto cleanup;
235 	}
236 	ext4_xattr_cache_insert(bh);
237 	entry = BFIRST(bh);
238 	error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
239 	if (error == -EIO)
240 		goto bad_block;
241 	if (error)
242 		goto cleanup;
243 	size = le32_to_cpu(entry->e_value_size);
244 	if (buffer) {
245 		error = -ERANGE;
246 		if (size > buffer_size)
247 			goto cleanup;
248 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
249 		       size);
250 	}
251 	error = size;
252 
253 cleanup:
254 	brelse(bh);
255 	return error;
256 }
257 
258 static int
259 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
260 		     void *buffer, size_t buffer_size)
261 {
262 	struct ext4_xattr_ibody_header *header;
263 	struct ext4_xattr_entry *entry;
264 	struct ext4_inode *raw_inode;
265 	struct ext4_iloc iloc;
266 	size_t size;
267 	void *end;
268 	int error;
269 
270 	if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR))
271 		return -ENODATA;
272 	error = ext4_get_inode_loc(inode, &iloc);
273 	if (error)
274 		return error;
275 	raw_inode = ext4_raw_inode(&iloc);
276 	header = IHDR(inode, raw_inode);
277 	entry = IFIRST(header);
278 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
279 	error = ext4_xattr_check_names(entry, end);
280 	if (error)
281 		goto cleanup;
282 	error = ext4_xattr_find_entry(&entry, name_index, name,
283 				      end - (void *)entry, 0);
284 	if (error)
285 		goto cleanup;
286 	size = le32_to_cpu(entry->e_value_size);
287 	if (buffer) {
288 		error = -ERANGE;
289 		if (size > buffer_size)
290 			goto cleanup;
291 		memcpy(buffer, (void *)IFIRST(header) +
292 		       le16_to_cpu(entry->e_value_offs), size);
293 	}
294 	error = size;
295 
296 cleanup:
297 	brelse(iloc.bh);
298 	return error;
299 }
300 
301 /*
302  * ext4_xattr_get()
303  *
304  * Copy an extended attribute into the buffer
305  * provided, or compute the buffer size required.
306  * Buffer is NULL to compute the size of the buffer required.
307  *
308  * Returns a negative error number on failure, or the number of bytes
309  * used / required on success.
310  */
311 int
312 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
313 	       void *buffer, size_t buffer_size)
314 {
315 	int error;
316 
317 	down_read(&EXT4_I(inode)->xattr_sem);
318 	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
319 				     buffer_size);
320 	if (error == -ENODATA)
321 		error = ext4_xattr_block_get(inode, name_index, name, buffer,
322 					     buffer_size);
323 	up_read(&EXT4_I(inode)->xattr_sem);
324 	return error;
325 }
326 
327 static int
328 ext4_xattr_list_entries(struct inode *inode, struct ext4_xattr_entry *entry,
329 			char *buffer, size_t buffer_size)
330 {
331 	size_t rest = buffer_size;
332 
333 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
334 		struct xattr_handler *handler =
335 			ext4_xattr_handler(entry->e_name_index);
336 
337 		if (handler) {
338 			size_t size = handler->list(inode, buffer, rest,
339 						    entry->e_name,
340 						    entry->e_name_len);
341 			if (buffer) {
342 				if (size > rest)
343 					return -ERANGE;
344 				buffer += size;
345 			}
346 			rest -= size;
347 		}
348 	}
349 	return buffer_size - rest;
350 }
351 
352 static int
353 ext4_xattr_block_list(struct inode *inode, char *buffer, size_t buffer_size)
354 {
355 	struct buffer_head *bh = NULL;
356 	int error;
357 
358 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
359 		  buffer, (long)buffer_size);
360 
361 	error = 0;
362 	if (!EXT4_I(inode)->i_file_acl)
363 		goto cleanup;
364 	ea_idebug(inode, "reading block %u", EXT4_I(inode)->i_file_acl);
365 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
366 	error = -EIO;
367 	if (!bh)
368 		goto cleanup;
369 	ea_bdebug(bh, "b_count=%d, refcount=%d",
370 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
371 	if (ext4_xattr_check_block(bh)) {
372 		ext4_error(inode->i_sb, __func__,
373 			   "inode %lu: bad block %llu", inode->i_ino,
374 			   EXT4_I(inode)->i_file_acl);
375 		error = -EIO;
376 		goto cleanup;
377 	}
378 	ext4_xattr_cache_insert(bh);
379 	error = ext4_xattr_list_entries(inode, BFIRST(bh), buffer, buffer_size);
380 
381 cleanup:
382 	brelse(bh);
383 
384 	return error;
385 }
386 
387 static int
388 ext4_xattr_ibody_list(struct inode *inode, char *buffer, size_t buffer_size)
389 {
390 	struct ext4_xattr_ibody_header *header;
391 	struct ext4_inode *raw_inode;
392 	struct ext4_iloc iloc;
393 	void *end;
394 	int error;
395 
396 	if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR))
397 		return 0;
398 	error = ext4_get_inode_loc(inode, &iloc);
399 	if (error)
400 		return error;
401 	raw_inode = ext4_raw_inode(&iloc);
402 	header = IHDR(inode, raw_inode);
403 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
404 	error = ext4_xattr_check_names(IFIRST(header), end);
405 	if (error)
406 		goto cleanup;
407 	error = ext4_xattr_list_entries(inode, IFIRST(header),
408 					buffer, buffer_size);
409 
410 cleanup:
411 	brelse(iloc.bh);
412 	return error;
413 }
414 
415 /*
416  * ext4_xattr_list()
417  *
418  * Copy a list of attribute names into the buffer
419  * provided, or compute the buffer size required.
420  * Buffer is NULL to compute the size of the buffer required.
421  *
422  * Returns a negative error number on failure, or the number of bytes
423  * used / required on success.
424  */
425 static int
426 ext4_xattr_list(struct inode *inode, char *buffer, size_t buffer_size)
427 {
428 	int i_error, b_error;
429 
430 	down_read(&EXT4_I(inode)->xattr_sem);
431 	i_error = ext4_xattr_ibody_list(inode, buffer, buffer_size);
432 	if (i_error < 0) {
433 		b_error = 0;
434 	} else {
435 		if (buffer) {
436 			buffer += i_error;
437 			buffer_size -= i_error;
438 		}
439 		b_error = ext4_xattr_block_list(inode, buffer, buffer_size);
440 		if (b_error < 0)
441 			i_error = 0;
442 	}
443 	up_read(&EXT4_I(inode)->xattr_sem);
444 	return i_error + b_error;
445 }
446 
447 /*
448  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
449  * not set, set it.
450  */
451 static void ext4_xattr_update_super_block(handle_t *handle,
452 					  struct super_block *sb)
453 {
454 	if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_EXT_ATTR))
455 		return;
456 
457 	if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
458 		EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_EXT_ATTR);
459 		sb->s_dirt = 1;
460 		ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
461 	}
462 }
463 
464 /*
465  * Release the xattr block BH: If the reference count is > 1, decrement
466  * it; otherwise free the block.
467  */
468 static void
469 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
470 			 struct buffer_head *bh)
471 {
472 	struct mb_cache_entry *ce = NULL;
473 	int error = 0;
474 
475 	ce = mb_cache_entry_get(ext4_xattr_cache, bh->b_bdev, bh->b_blocknr);
476 	error = ext4_journal_get_write_access(handle, bh);
477 	if (error)
478 		goto out;
479 
480 	lock_buffer(bh);
481 	if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
482 		ea_bdebug(bh, "refcount now=0; freeing");
483 		if (ce)
484 			mb_cache_entry_free(ce);
485 		ext4_free_blocks(handle, inode, bh->b_blocknr, 1, 1);
486 		get_bh(bh);
487 		ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
488 	} else {
489 		le32_add_cpu(&BHDR(bh)->h_refcount, -1);
490 		error = ext4_journal_dirty_metadata(handle, bh);
491 		if (IS_SYNC(inode))
492 			handle->h_sync = 1;
493 		DQUOT_FREE_BLOCK(inode, 1);
494 		ea_bdebug(bh, "refcount now=%d; releasing",
495 			  le32_to_cpu(BHDR(bh)->h_refcount));
496 		if (ce)
497 			mb_cache_entry_release(ce);
498 	}
499 	unlock_buffer(bh);
500 out:
501 	ext4_std_error(inode->i_sb, error);
502 	return;
503 }
504 
505 /*
506  * Find the available free space for EAs. This also returns the total number of
507  * bytes used by EA entries.
508  */
509 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
510 				    size_t *min_offs, void *base, int *total)
511 {
512 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
513 		*total += EXT4_XATTR_LEN(last->e_name_len);
514 		if (!last->e_value_block && last->e_value_size) {
515 			size_t offs = le16_to_cpu(last->e_value_offs);
516 			if (offs < *min_offs)
517 				*min_offs = offs;
518 		}
519 	}
520 	return (*min_offs - ((void *)last - base) - sizeof(__u32));
521 }
522 
523 struct ext4_xattr_info {
524 	int name_index;
525 	const char *name;
526 	const void *value;
527 	size_t value_len;
528 };
529 
530 struct ext4_xattr_search {
531 	struct ext4_xattr_entry *first;
532 	void *base;
533 	void *end;
534 	struct ext4_xattr_entry *here;
535 	int not_found;
536 };
537 
538 static int
539 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
540 {
541 	struct ext4_xattr_entry *last;
542 	size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
543 
544 	/* Compute min_offs and last. */
545 	last = s->first;
546 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
547 		if (!last->e_value_block && last->e_value_size) {
548 			size_t offs = le16_to_cpu(last->e_value_offs);
549 			if (offs < min_offs)
550 				min_offs = offs;
551 		}
552 	}
553 	free = min_offs - ((void *)last - s->base) - sizeof(__u32);
554 	if (!s->not_found) {
555 		if (!s->here->e_value_block && s->here->e_value_size) {
556 			size_t size = le32_to_cpu(s->here->e_value_size);
557 			free += EXT4_XATTR_SIZE(size);
558 		}
559 		free += EXT4_XATTR_LEN(name_len);
560 	}
561 	if (i->value) {
562 		if (free < EXT4_XATTR_SIZE(i->value_len) ||
563 		    free < EXT4_XATTR_LEN(name_len) +
564 			   EXT4_XATTR_SIZE(i->value_len))
565 			return -ENOSPC;
566 	}
567 
568 	if (i->value && s->not_found) {
569 		/* Insert the new name. */
570 		size_t size = EXT4_XATTR_LEN(name_len);
571 		size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
572 		memmove((void *)s->here + size, s->here, rest);
573 		memset(s->here, 0, size);
574 		s->here->e_name_index = i->name_index;
575 		s->here->e_name_len = name_len;
576 		memcpy(s->here->e_name, i->name, name_len);
577 	} else {
578 		if (!s->here->e_value_block && s->here->e_value_size) {
579 			void *first_val = s->base + min_offs;
580 			size_t offs = le16_to_cpu(s->here->e_value_offs);
581 			void *val = s->base + offs;
582 			size_t size = EXT4_XATTR_SIZE(
583 				le32_to_cpu(s->here->e_value_size));
584 
585 			if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
586 				/* The old and the new value have the same
587 				   size. Just replace. */
588 				s->here->e_value_size =
589 					cpu_to_le32(i->value_len);
590 				memset(val + size - EXT4_XATTR_PAD, 0,
591 				       EXT4_XATTR_PAD); /* Clear pad bytes. */
592 				memcpy(val, i->value, i->value_len);
593 				return 0;
594 			}
595 
596 			/* Remove the old value. */
597 			memmove(first_val + size, first_val, val - first_val);
598 			memset(first_val, 0, size);
599 			s->here->e_value_size = 0;
600 			s->here->e_value_offs = 0;
601 			min_offs += size;
602 
603 			/* Adjust all value offsets. */
604 			last = s->first;
605 			while (!IS_LAST_ENTRY(last)) {
606 				size_t o = le16_to_cpu(last->e_value_offs);
607 				if (!last->e_value_block &&
608 				    last->e_value_size && o < offs)
609 					last->e_value_offs =
610 						cpu_to_le16(o + size);
611 				last = EXT4_XATTR_NEXT(last);
612 			}
613 		}
614 		if (!i->value) {
615 			/* Remove the old name. */
616 			size_t size = EXT4_XATTR_LEN(name_len);
617 			last = ENTRY((void *)last - size);
618 			memmove(s->here, (void *)s->here + size,
619 				(void *)last - (void *)s->here + sizeof(__u32));
620 			memset(last, 0, size);
621 		}
622 	}
623 
624 	if (i->value) {
625 		/* Insert the new value. */
626 		s->here->e_value_size = cpu_to_le32(i->value_len);
627 		if (i->value_len) {
628 			size_t size = EXT4_XATTR_SIZE(i->value_len);
629 			void *val = s->base + min_offs - size;
630 			s->here->e_value_offs = cpu_to_le16(min_offs - size);
631 			memset(val + size - EXT4_XATTR_PAD, 0,
632 			       EXT4_XATTR_PAD); /* Clear the pad bytes. */
633 			memcpy(val, i->value, i->value_len);
634 		}
635 	}
636 	return 0;
637 }
638 
639 struct ext4_xattr_block_find {
640 	struct ext4_xattr_search s;
641 	struct buffer_head *bh;
642 };
643 
644 static int
645 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
646 		      struct ext4_xattr_block_find *bs)
647 {
648 	struct super_block *sb = inode->i_sb;
649 	int error;
650 
651 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
652 		  i->name_index, i->name, i->value, (long)i->value_len);
653 
654 	if (EXT4_I(inode)->i_file_acl) {
655 		/* The inode already has an extended attribute block. */
656 		bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
657 		error = -EIO;
658 		if (!bs->bh)
659 			goto cleanup;
660 		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
661 			atomic_read(&(bs->bh->b_count)),
662 			le32_to_cpu(BHDR(bs->bh)->h_refcount));
663 		if (ext4_xattr_check_block(bs->bh)) {
664 			ext4_error(sb, __func__,
665 				"inode %lu: bad block %llu", inode->i_ino,
666 				EXT4_I(inode)->i_file_acl);
667 			error = -EIO;
668 			goto cleanup;
669 		}
670 		/* Find the named attribute. */
671 		bs->s.base = BHDR(bs->bh);
672 		bs->s.first = BFIRST(bs->bh);
673 		bs->s.end = bs->bh->b_data + bs->bh->b_size;
674 		bs->s.here = bs->s.first;
675 		error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
676 					      i->name, bs->bh->b_size, 1);
677 		if (error && error != -ENODATA)
678 			goto cleanup;
679 		bs->s.not_found = error;
680 	}
681 	error = 0;
682 
683 cleanup:
684 	return error;
685 }
686 
687 static int
688 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
689 		     struct ext4_xattr_info *i,
690 		     struct ext4_xattr_block_find *bs)
691 {
692 	struct super_block *sb = inode->i_sb;
693 	struct buffer_head *new_bh = NULL;
694 	struct ext4_xattr_search *s = &bs->s;
695 	struct mb_cache_entry *ce = NULL;
696 	int error = 0;
697 
698 #define header(x) ((struct ext4_xattr_header *)(x))
699 
700 	if (i->value && i->value_len > sb->s_blocksize)
701 		return -ENOSPC;
702 	if (s->base) {
703 		ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
704 					bs->bh->b_blocknr);
705 		error = ext4_journal_get_write_access(handle, bs->bh);
706 		if (error)
707 			goto cleanup;
708 		lock_buffer(bs->bh);
709 
710 		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
711 			if (ce) {
712 				mb_cache_entry_free(ce);
713 				ce = NULL;
714 			}
715 			ea_bdebug(bs->bh, "modifying in-place");
716 			error = ext4_xattr_set_entry(i, s);
717 			if (!error) {
718 				if (!IS_LAST_ENTRY(s->first))
719 					ext4_xattr_rehash(header(s->base),
720 							  s->here);
721 				ext4_xattr_cache_insert(bs->bh);
722 			}
723 			unlock_buffer(bs->bh);
724 			if (error == -EIO)
725 				goto bad_block;
726 			if (!error)
727 				error = ext4_journal_dirty_metadata(handle,
728 								    bs->bh);
729 			if (error)
730 				goto cleanup;
731 			goto inserted;
732 		} else {
733 			int offset = (char *)s->here - bs->bh->b_data;
734 
735 			unlock_buffer(bs->bh);
736 			jbd2_journal_release_buffer(handle, bs->bh);
737 			if (ce) {
738 				mb_cache_entry_release(ce);
739 				ce = NULL;
740 			}
741 			ea_bdebug(bs->bh, "cloning");
742 			s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
743 			error = -ENOMEM;
744 			if (s->base == NULL)
745 				goto cleanup;
746 			memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
747 			s->first = ENTRY(header(s->base)+1);
748 			header(s->base)->h_refcount = cpu_to_le32(1);
749 			s->here = ENTRY(s->base + offset);
750 			s->end = s->base + bs->bh->b_size;
751 		}
752 	} else {
753 		/* Allocate a buffer where we construct the new block. */
754 		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
755 		/* assert(header == s->base) */
756 		error = -ENOMEM;
757 		if (s->base == NULL)
758 			goto cleanup;
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 				le32_add_cpu(&BHDR(new_bh)->h_refcount, 1);
794 				ea_bdebug(new_bh, "reusing; refcount now=%d",
795 					le32_to_cpu(BHDR(new_bh)->h_refcount));
796 				unlock_buffer(new_bh);
797 				error = ext4_journal_dirty_metadata(handle,
798 								    new_bh);
799 				if (error)
800 					goto cleanup_dquot;
801 			}
802 			mb_cache_entry_release(ce);
803 			ce = NULL;
804 		} else if (bs->bh && s->base == bs->bh->b_data) {
805 			/* We were modifying this block in-place. */
806 			ea_bdebug(bs->bh, "keeping this block");
807 			new_bh = bs->bh;
808 			get_bh(new_bh);
809 		} else {
810 			/* We need to allocate a new block */
811 			ext4_fsblk_t goal = ext4_group_first_block_no(sb,
812 						EXT4_I(inode)->i_block_group);
813 			ext4_fsblk_t block = ext4_new_meta_block(handle, inode,
814 							goal, &error);
815 			if (error)
816 				goto cleanup;
817 			ea_idebug(inode, "creating block %d", block);
818 
819 			new_bh = sb_getblk(sb, block);
820 			if (!new_bh) {
821 getblk_failed:
822 				ext4_free_blocks(handle, inode, block, 1, 1);
823 				error = -EIO;
824 				goto cleanup;
825 			}
826 			lock_buffer(new_bh);
827 			error = ext4_journal_get_create_access(handle, new_bh);
828 			if (error) {
829 				unlock_buffer(new_bh);
830 				goto getblk_failed;
831 			}
832 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
833 			set_buffer_uptodate(new_bh);
834 			unlock_buffer(new_bh);
835 			ext4_xattr_cache_insert(new_bh);
836 			error = ext4_journal_dirty_metadata(handle, new_bh);
837 			if (error)
838 				goto cleanup;
839 		}
840 	}
841 
842 	/* Update the inode. */
843 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
844 
845 	/* Drop the previous xattr block. */
846 	if (bs->bh && bs->bh != new_bh)
847 		ext4_xattr_release_block(handle, inode, bs->bh);
848 	error = 0;
849 
850 cleanup:
851 	if (ce)
852 		mb_cache_entry_release(ce);
853 	brelse(new_bh);
854 	if (!(bs->bh && s->base == bs->bh->b_data))
855 		kfree(s->base);
856 
857 	return error;
858 
859 cleanup_dquot:
860 	DQUOT_FREE_BLOCK(inode, 1);
861 	goto cleanup;
862 
863 bad_block:
864 	ext4_error(inode->i_sb, __func__,
865 		   "inode %lu: bad block %llu", inode->i_ino,
866 		   EXT4_I(inode)->i_file_acl);
867 	goto cleanup;
868 
869 #undef header
870 }
871 
872 struct ext4_xattr_ibody_find {
873 	struct ext4_xattr_search s;
874 	struct ext4_iloc iloc;
875 };
876 
877 static int
878 ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
879 		      struct ext4_xattr_ibody_find *is)
880 {
881 	struct ext4_xattr_ibody_header *header;
882 	struct ext4_inode *raw_inode;
883 	int error;
884 
885 	if (EXT4_I(inode)->i_extra_isize == 0)
886 		return 0;
887 	raw_inode = ext4_raw_inode(&is->iloc);
888 	header = IHDR(inode, raw_inode);
889 	is->s.base = is->s.first = IFIRST(header);
890 	is->s.here = is->s.first;
891 	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
892 	if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
893 		error = ext4_xattr_check_names(IFIRST(header), is->s.end);
894 		if (error)
895 			return error;
896 		/* Find the named attribute. */
897 		error = ext4_xattr_find_entry(&is->s.here, i->name_index,
898 					      i->name, is->s.end -
899 					      (void *)is->s.base, 0);
900 		if (error && error != -ENODATA)
901 			return error;
902 		is->s.not_found = error;
903 	}
904 	return 0;
905 }
906 
907 static int
908 ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
909 		     struct ext4_xattr_info *i,
910 		     struct ext4_xattr_ibody_find *is)
911 {
912 	struct ext4_xattr_ibody_header *header;
913 	struct ext4_xattr_search *s = &is->s;
914 	int error;
915 
916 	if (EXT4_I(inode)->i_extra_isize == 0)
917 		return -ENOSPC;
918 	error = ext4_xattr_set_entry(i, s);
919 	if (error)
920 		return error;
921 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
922 	if (!IS_LAST_ENTRY(s->first)) {
923 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
924 		EXT4_I(inode)->i_state |= EXT4_STATE_XATTR;
925 	} else {
926 		header->h_magic = cpu_to_le32(0);
927 		EXT4_I(inode)->i_state &= ~EXT4_STATE_XATTR;
928 	}
929 	return 0;
930 }
931 
932 /*
933  * ext4_xattr_set_handle()
934  *
935  * Create, replace or remove an extended attribute for this inode. Buffer
936  * is NULL to remove an existing extended attribute, and non-NULL to
937  * either replace an existing extended attribute, or create a new extended
938  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
939  * specify that an extended attribute must exist and must not exist
940  * previous to the call, respectively.
941  *
942  * Returns 0, or a negative error number on failure.
943  */
944 int
945 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
946 		      const char *name, const void *value, size_t value_len,
947 		      int flags)
948 {
949 	struct ext4_xattr_info i = {
950 		.name_index = name_index,
951 		.name = name,
952 		.value = value,
953 		.value_len = value_len,
954 
955 	};
956 	struct ext4_xattr_ibody_find is = {
957 		.s = { .not_found = -ENODATA, },
958 	};
959 	struct ext4_xattr_block_find bs = {
960 		.s = { .not_found = -ENODATA, },
961 	};
962 	int error;
963 
964 	if (!name)
965 		return -EINVAL;
966 	if (strlen(name) > 255)
967 		return -ERANGE;
968 	down_write(&EXT4_I(inode)->xattr_sem);
969 	error = ext4_get_inode_loc(inode, &is.iloc);
970 	if (error)
971 		goto cleanup;
972 
973 	if (EXT4_I(inode)->i_state & EXT4_STATE_NEW) {
974 		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
975 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
976 		EXT4_I(inode)->i_state &= ~EXT4_STATE_NEW;
977 	}
978 
979 	error = ext4_xattr_ibody_find(inode, &i, &is);
980 	if (error)
981 		goto cleanup;
982 	if (is.s.not_found)
983 		error = ext4_xattr_block_find(inode, &i, &bs);
984 	if (error)
985 		goto cleanup;
986 	if (is.s.not_found && bs.s.not_found) {
987 		error = -ENODATA;
988 		if (flags & XATTR_REPLACE)
989 			goto cleanup;
990 		error = 0;
991 		if (!value)
992 			goto cleanup;
993 	} else {
994 		error = -EEXIST;
995 		if (flags & XATTR_CREATE)
996 			goto cleanup;
997 	}
998 	error = ext4_journal_get_write_access(handle, is.iloc.bh);
999 	if (error)
1000 		goto cleanup;
1001 	if (!value) {
1002 		if (!is.s.not_found)
1003 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1004 		else if (!bs.s.not_found)
1005 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1006 	} else {
1007 		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1008 		if (!error && !bs.s.not_found) {
1009 			i.value = NULL;
1010 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1011 		} else if (error == -ENOSPC) {
1012 			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1013 				error = ext4_xattr_block_find(inode, &i, &bs);
1014 				if (error)
1015 					goto cleanup;
1016 			}
1017 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1018 			if (error)
1019 				goto cleanup;
1020 			if (!is.s.not_found) {
1021 				i.value = NULL;
1022 				error = ext4_xattr_ibody_set(handle, inode, &i,
1023 							     &is);
1024 			}
1025 		}
1026 	}
1027 	if (!error) {
1028 		ext4_xattr_update_super_block(handle, inode->i_sb);
1029 		inode->i_ctime = ext4_current_time(inode);
1030 		if (!value)
1031 			EXT4_I(inode)->i_state &= ~EXT4_STATE_NO_EXPAND;
1032 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1033 		/*
1034 		 * The bh is consumed by ext4_mark_iloc_dirty, even with
1035 		 * error != 0.
1036 		 */
1037 		is.iloc.bh = NULL;
1038 		if (IS_SYNC(inode))
1039 			handle->h_sync = 1;
1040 	}
1041 
1042 cleanup:
1043 	brelse(is.iloc.bh);
1044 	brelse(bs.bh);
1045 	up_write(&EXT4_I(inode)->xattr_sem);
1046 	return error;
1047 }
1048 
1049 /*
1050  * ext4_xattr_set()
1051  *
1052  * Like ext4_xattr_set_handle, but start from an inode. This extended
1053  * attribute modification is a filesystem transaction by itself.
1054  *
1055  * Returns 0, or a negative error number on failure.
1056  */
1057 int
1058 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1059 	       const void *value, size_t value_len, int flags)
1060 {
1061 	handle_t *handle;
1062 	int error, retries = 0;
1063 
1064 retry:
1065 	handle = ext4_journal_start(inode, EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
1066 	if (IS_ERR(handle)) {
1067 		error = PTR_ERR(handle);
1068 	} else {
1069 		int error2;
1070 
1071 		error = ext4_xattr_set_handle(handle, inode, name_index, name,
1072 					      value, value_len, flags);
1073 		error2 = ext4_journal_stop(handle);
1074 		if (error == -ENOSPC &&
1075 		    ext4_should_retry_alloc(inode->i_sb, &retries))
1076 			goto retry;
1077 		if (error == 0)
1078 			error = error2;
1079 	}
1080 
1081 	return error;
1082 }
1083 
1084 /*
1085  * Shift the EA entries in the inode to create space for the increased
1086  * i_extra_isize.
1087  */
1088 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1089 				     int value_offs_shift, void *to,
1090 				     void *from, size_t n, int blocksize)
1091 {
1092 	struct ext4_xattr_entry *last = entry;
1093 	int new_offs;
1094 
1095 	/* Adjust the value offsets of the entries */
1096 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1097 		if (!last->e_value_block && last->e_value_size) {
1098 			new_offs = le16_to_cpu(last->e_value_offs) +
1099 							value_offs_shift;
1100 			BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
1101 				 > blocksize);
1102 			last->e_value_offs = cpu_to_le16(new_offs);
1103 		}
1104 	}
1105 	/* Shift the entries by n bytes */
1106 	memmove(to, from, n);
1107 }
1108 
1109 /*
1110  * Expand an inode by new_extra_isize bytes when EAs are present.
1111  * Returns 0 on success or negative error number on failure.
1112  */
1113 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1114 			       struct ext4_inode *raw_inode, handle_t *handle)
1115 {
1116 	struct ext4_xattr_ibody_header *header;
1117 	struct ext4_xattr_entry *entry, *last, *first;
1118 	struct buffer_head *bh = NULL;
1119 	struct ext4_xattr_ibody_find *is = NULL;
1120 	struct ext4_xattr_block_find *bs = NULL;
1121 	char *buffer = NULL, *b_entry_name = NULL;
1122 	size_t min_offs, free;
1123 	int total_ino, total_blk;
1124 	void *base, *start, *end;
1125 	int extra_isize = 0, error = 0, tried_min_extra_isize = 0;
1126 	int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1127 
1128 	down_write(&EXT4_I(inode)->xattr_sem);
1129 retry:
1130 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) {
1131 		up_write(&EXT4_I(inode)->xattr_sem);
1132 		return 0;
1133 	}
1134 
1135 	header = IHDR(inode, raw_inode);
1136 	entry = IFIRST(header);
1137 
1138 	/*
1139 	 * Check if enough free space is available in the inode to shift the
1140 	 * entries ahead by new_extra_isize.
1141 	 */
1142 
1143 	base = start = entry;
1144 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1145 	min_offs = end - base;
1146 	last = entry;
1147 	total_ino = sizeof(struct ext4_xattr_ibody_header);
1148 
1149 	free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
1150 	if (free >= new_extra_isize) {
1151 		entry = IFIRST(header);
1152 		ext4_xattr_shift_entries(entry,	EXT4_I(inode)->i_extra_isize
1153 				- new_extra_isize, (void *)raw_inode +
1154 				EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1155 				(void *)header, total_ino,
1156 				inode->i_sb->s_blocksize);
1157 		EXT4_I(inode)->i_extra_isize = new_extra_isize;
1158 		error = 0;
1159 		goto cleanup;
1160 	}
1161 
1162 	/*
1163 	 * Enough free space isn't available in the inode, check if
1164 	 * EA block can hold new_extra_isize bytes.
1165 	 */
1166 	if (EXT4_I(inode)->i_file_acl) {
1167 		bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1168 		error = -EIO;
1169 		if (!bh)
1170 			goto cleanup;
1171 		if (ext4_xattr_check_block(bh)) {
1172 			ext4_error(inode->i_sb, __func__,
1173 				"inode %lu: bad block %llu", inode->i_ino,
1174 				EXT4_I(inode)->i_file_acl);
1175 			error = -EIO;
1176 			goto cleanup;
1177 		}
1178 		base = BHDR(bh);
1179 		first = BFIRST(bh);
1180 		end = bh->b_data + bh->b_size;
1181 		min_offs = end - base;
1182 		free = ext4_xattr_free_space(first, &min_offs, base,
1183 					     &total_blk);
1184 		if (free < new_extra_isize) {
1185 			if (!tried_min_extra_isize && s_min_extra_isize) {
1186 				tried_min_extra_isize++;
1187 				new_extra_isize = s_min_extra_isize;
1188 				brelse(bh);
1189 				goto retry;
1190 			}
1191 			error = -1;
1192 			goto cleanup;
1193 		}
1194 	} else {
1195 		free = inode->i_sb->s_blocksize;
1196 	}
1197 
1198 	while (new_extra_isize > 0) {
1199 		size_t offs, size, entry_size;
1200 		struct ext4_xattr_entry *small_entry = NULL;
1201 		struct ext4_xattr_info i = {
1202 			.value = NULL,
1203 			.value_len = 0,
1204 		};
1205 		unsigned int total_size;  /* EA entry size + value size */
1206 		unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
1207 		unsigned int min_total_size = ~0U;
1208 
1209 		is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1210 		bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1211 		if (!is || !bs) {
1212 			error = -ENOMEM;
1213 			goto cleanup;
1214 		}
1215 
1216 		is->s.not_found = -ENODATA;
1217 		bs->s.not_found = -ENODATA;
1218 		is->iloc.bh = NULL;
1219 		bs->bh = NULL;
1220 
1221 		last = IFIRST(header);
1222 		/* Find the entry best suited to be pushed into EA block */
1223 		entry = NULL;
1224 		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1225 			total_size =
1226 			EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1227 					EXT4_XATTR_LEN(last->e_name_len);
1228 			if (total_size <= free && total_size < min_total_size) {
1229 				if (total_size < new_extra_isize) {
1230 					small_entry = last;
1231 				} else {
1232 					entry = last;
1233 					min_total_size = total_size;
1234 				}
1235 			}
1236 		}
1237 
1238 		if (entry == NULL) {
1239 			if (small_entry) {
1240 				entry = small_entry;
1241 			} else {
1242 				if (!tried_min_extra_isize &&
1243 				    s_min_extra_isize) {
1244 					tried_min_extra_isize++;
1245 					new_extra_isize = s_min_extra_isize;
1246 					goto retry;
1247 				}
1248 				error = -1;
1249 				goto cleanup;
1250 			}
1251 		}
1252 		offs = le16_to_cpu(entry->e_value_offs);
1253 		size = le32_to_cpu(entry->e_value_size);
1254 		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1255 		i.name_index = entry->e_name_index,
1256 		buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
1257 		b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1258 		if (!buffer || !b_entry_name) {
1259 			error = -ENOMEM;
1260 			goto cleanup;
1261 		}
1262 		/* Save the entry name and the entry value */
1263 		memcpy(buffer, (void *)IFIRST(header) + offs,
1264 		       EXT4_XATTR_SIZE(size));
1265 		memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1266 		b_entry_name[entry->e_name_len] = '\0';
1267 		i.name = b_entry_name;
1268 
1269 		error = ext4_get_inode_loc(inode, &is->iloc);
1270 		if (error)
1271 			goto cleanup;
1272 
1273 		error = ext4_xattr_ibody_find(inode, &i, is);
1274 		if (error)
1275 			goto cleanup;
1276 
1277 		/* Remove the chosen entry from the inode */
1278 		error = ext4_xattr_ibody_set(handle, inode, &i, is);
1279 
1280 		entry = IFIRST(header);
1281 		if (entry_size + EXT4_XATTR_SIZE(size) >= new_extra_isize)
1282 			shift_bytes = new_extra_isize;
1283 		else
1284 			shift_bytes = entry_size + size;
1285 		/* Adjust the offsets and shift the remaining entries ahead */
1286 		ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize -
1287 			shift_bytes, (void *)raw_inode +
1288 			EXT4_GOOD_OLD_INODE_SIZE + extra_isize + shift_bytes,
1289 			(void *)header, total_ino - entry_size,
1290 			inode->i_sb->s_blocksize);
1291 
1292 		extra_isize += shift_bytes;
1293 		new_extra_isize -= shift_bytes;
1294 		EXT4_I(inode)->i_extra_isize = extra_isize;
1295 
1296 		i.name = b_entry_name;
1297 		i.value = buffer;
1298 		i.value_len = size;
1299 		error = ext4_xattr_block_find(inode, &i, bs);
1300 		if (error)
1301 			goto cleanup;
1302 
1303 		/* Add entry which was removed from the inode into the block */
1304 		error = ext4_xattr_block_set(handle, inode, &i, bs);
1305 		if (error)
1306 			goto cleanup;
1307 		kfree(b_entry_name);
1308 		kfree(buffer);
1309 		brelse(is->iloc.bh);
1310 		kfree(is);
1311 		kfree(bs);
1312 	}
1313 	brelse(bh);
1314 	up_write(&EXT4_I(inode)->xattr_sem);
1315 	return 0;
1316 
1317 cleanup:
1318 	kfree(b_entry_name);
1319 	kfree(buffer);
1320 	if (is)
1321 		brelse(is->iloc.bh);
1322 	kfree(is);
1323 	kfree(bs);
1324 	brelse(bh);
1325 	up_write(&EXT4_I(inode)->xattr_sem);
1326 	return error;
1327 }
1328 
1329 
1330 
1331 /*
1332  * ext4_xattr_delete_inode()
1333  *
1334  * Free extended attribute resources associated with this inode. This
1335  * is called immediately before an inode is freed. We have exclusive
1336  * access to the inode.
1337  */
1338 void
1339 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1340 {
1341 	struct buffer_head *bh = NULL;
1342 
1343 	if (!EXT4_I(inode)->i_file_acl)
1344 		goto cleanup;
1345 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1346 	if (!bh) {
1347 		ext4_error(inode->i_sb, __func__,
1348 			"inode %lu: block %llu read error", inode->i_ino,
1349 			EXT4_I(inode)->i_file_acl);
1350 		goto cleanup;
1351 	}
1352 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1353 	    BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1354 		ext4_error(inode->i_sb, __func__,
1355 			"inode %lu: bad block %llu", inode->i_ino,
1356 			EXT4_I(inode)->i_file_acl);
1357 		goto cleanup;
1358 	}
1359 	ext4_xattr_release_block(handle, inode, bh);
1360 	EXT4_I(inode)->i_file_acl = 0;
1361 
1362 cleanup:
1363 	brelse(bh);
1364 }
1365 
1366 /*
1367  * ext4_xattr_put_super()
1368  *
1369  * This is called when a file system is unmounted.
1370  */
1371 void
1372 ext4_xattr_put_super(struct super_block *sb)
1373 {
1374 	mb_cache_shrink(sb->s_bdev);
1375 }
1376 
1377 /*
1378  * ext4_xattr_cache_insert()
1379  *
1380  * Create a new entry in the extended attribute cache, and insert
1381  * it unless such an entry is already in the cache.
1382  *
1383  * Returns 0, or a negative error number on failure.
1384  */
1385 static void
1386 ext4_xattr_cache_insert(struct buffer_head *bh)
1387 {
1388 	__u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1389 	struct mb_cache_entry *ce;
1390 	int error;
1391 
1392 	ce = mb_cache_entry_alloc(ext4_xattr_cache, GFP_NOFS);
1393 	if (!ce) {
1394 		ea_bdebug(bh, "out of memory");
1395 		return;
1396 	}
1397 	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash);
1398 	if (error) {
1399 		mb_cache_entry_free(ce);
1400 		if (error == -EBUSY) {
1401 			ea_bdebug(bh, "already in cache");
1402 			error = 0;
1403 		}
1404 	} else {
1405 		ea_bdebug(bh, "inserting [%x]", (int)hash);
1406 		mb_cache_entry_release(ce);
1407 	}
1408 }
1409 
1410 /*
1411  * ext4_xattr_cmp()
1412  *
1413  * Compare two extended attribute blocks for equality.
1414  *
1415  * Returns 0 if the blocks are equal, 1 if they differ, and
1416  * a negative error number on errors.
1417  */
1418 static int
1419 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1420 	       struct ext4_xattr_header *header2)
1421 {
1422 	struct ext4_xattr_entry *entry1, *entry2;
1423 
1424 	entry1 = ENTRY(header1+1);
1425 	entry2 = ENTRY(header2+1);
1426 	while (!IS_LAST_ENTRY(entry1)) {
1427 		if (IS_LAST_ENTRY(entry2))
1428 			return 1;
1429 		if (entry1->e_hash != entry2->e_hash ||
1430 		    entry1->e_name_index != entry2->e_name_index ||
1431 		    entry1->e_name_len != entry2->e_name_len ||
1432 		    entry1->e_value_size != entry2->e_value_size ||
1433 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1434 			return 1;
1435 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1436 			return -EIO;
1437 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1438 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1439 			   le32_to_cpu(entry1->e_value_size)))
1440 			return 1;
1441 
1442 		entry1 = EXT4_XATTR_NEXT(entry1);
1443 		entry2 = EXT4_XATTR_NEXT(entry2);
1444 	}
1445 	if (!IS_LAST_ENTRY(entry2))
1446 		return 1;
1447 	return 0;
1448 }
1449 
1450 /*
1451  * ext4_xattr_cache_find()
1452  *
1453  * Find an identical extended attribute block.
1454  *
1455  * Returns a pointer to the block found, or NULL if such a block was
1456  * not found or an error occurred.
1457  */
1458 static struct buffer_head *
1459 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1460 		      struct mb_cache_entry **pce)
1461 {
1462 	__u32 hash = le32_to_cpu(header->h_hash);
1463 	struct mb_cache_entry *ce;
1464 
1465 	if (!header->h_hash)
1466 		return NULL;  /* never share */
1467 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1468 again:
1469 	ce = mb_cache_entry_find_first(ext4_xattr_cache, 0,
1470 				       inode->i_sb->s_bdev, hash);
1471 	while (ce) {
1472 		struct buffer_head *bh;
1473 
1474 		if (IS_ERR(ce)) {
1475 			if (PTR_ERR(ce) == -EAGAIN)
1476 				goto again;
1477 			break;
1478 		}
1479 		bh = sb_bread(inode->i_sb, ce->e_block);
1480 		if (!bh) {
1481 			ext4_error(inode->i_sb, __func__,
1482 				"inode %lu: block %lu read error",
1483 				inode->i_ino, (unsigned long) ce->e_block);
1484 		} else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1485 				EXT4_XATTR_REFCOUNT_MAX) {
1486 			ea_idebug(inode, "block %lu refcount %d>=%d",
1487 				  (unsigned long) ce->e_block,
1488 				  le32_to_cpu(BHDR(bh)->h_refcount),
1489 					  EXT4_XATTR_REFCOUNT_MAX);
1490 		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1491 			*pce = ce;
1492 			return bh;
1493 		}
1494 		brelse(bh);
1495 		ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash);
1496 	}
1497 	return NULL;
1498 }
1499 
1500 #define NAME_HASH_SHIFT 5
1501 #define VALUE_HASH_SHIFT 16
1502 
1503 /*
1504  * ext4_xattr_hash_entry()
1505  *
1506  * Compute the hash of an extended attribute.
1507  */
1508 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1509 					 struct ext4_xattr_entry *entry)
1510 {
1511 	__u32 hash = 0;
1512 	char *name = entry->e_name;
1513 	int n;
1514 
1515 	for (n=0; n < entry->e_name_len; n++) {
1516 		hash = (hash << NAME_HASH_SHIFT) ^
1517 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1518 		       *name++;
1519 	}
1520 
1521 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1522 		__le32 *value = (__le32 *)((char *)header +
1523 			le16_to_cpu(entry->e_value_offs));
1524 		for (n = (le32_to_cpu(entry->e_value_size) +
1525 		     EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1526 			hash = (hash << VALUE_HASH_SHIFT) ^
1527 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1528 			       le32_to_cpu(*value++);
1529 		}
1530 	}
1531 	entry->e_hash = cpu_to_le32(hash);
1532 }
1533 
1534 #undef NAME_HASH_SHIFT
1535 #undef VALUE_HASH_SHIFT
1536 
1537 #define BLOCK_HASH_SHIFT 16
1538 
1539 /*
1540  * ext4_xattr_rehash()
1541  *
1542  * Re-compute the extended attribute hash value after an entry has changed.
1543  */
1544 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1545 			      struct ext4_xattr_entry *entry)
1546 {
1547 	struct ext4_xattr_entry *here;
1548 	__u32 hash = 0;
1549 
1550 	ext4_xattr_hash_entry(header, entry);
1551 	here = ENTRY(header+1);
1552 	while (!IS_LAST_ENTRY(here)) {
1553 		if (!here->e_hash) {
1554 			/* Block is not shared if an entry's hash value == 0 */
1555 			hash = 0;
1556 			break;
1557 		}
1558 		hash = (hash << BLOCK_HASH_SHIFT) ^
1559 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1560 		       le32_to_cpu(here->e_hash);
1561 		here = EXT4_XATTR_NEXT(here);
1562 	}
1563 	header->h_hash = cpu_to_le32(hash);
1564 }
1565 
1566 #undef BLOCK_HASH_SHIFT
1567 
1568 int __init
1569 init_ext4_xattr(void)
1570 {
1571 	ext4_xattr_cache = mb_cache_create("ext4_xattr", NULL,
1572 		sizeof(struct mb_cache_entry) +
1573 		sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6);
1574 	if (!ext4_xattr_cache)
1575 		return -ENOMEM;
1576 	return 0;
1577 }
1578 
1579 void
1580 exit_ext4_xattr(void)
1581 {
1582 	if (ext4_xattr_cache)
1583 		mb_cache_destroy(ext4_xattr_cache);
1584 	ext4_xattr_cache = NULL;
1585 }
1586