xref: /linux/fs/ext4/xattr.c (revision 273b281fa22c293963ee3e6eec418f5dda2dbc83)
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_EXT4_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_EXT4_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_EXT4_FS_POSIX_ACL
116 	&ext4_xattr_acl_access_handler,
117 	&ext4_xattr_acl_default_handler,
118 #endif
119 #ifdef CONFIG_EXT4_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_handle_dirty_metadata(handle, NULL, 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 		get_bh(bh);
486 		ext4_free_blocks(handle, inode, bh, 0, 1,
487 				 EXT4_FREE_BLOCKS_METADATA |
488 				 EXT4_FREE_BLOCKS_FORGET);
489 	} else {
490 		le32_add_cpu(&BHDR(bh)->h_refcount, -1);
491 		error = ext4_handle_dirty_metadata(handle, inode, bh);
492 		if (IS_SYNC(inode))
493 			ext4_handle_sync(handle);
494 		vfs_dq_free_block(inode, 1);
495 		ea_bdebug(bh, "refcount now=%d; releasing",
496 			  le32_to_cpu(BHDR(bh)->h_refcount));
497 		if (ce)
498 			mb_cache_entry_release(ce);
499 	}
500 	unlock_buffer(bh);
501 out:
502 	ext4_std_error(inode->i_sb, error);
503 	return;
504 }
505 
506 /*
507  * Find the available free space for EAs. This also returns the total number of
508  * bytes used by EA entries.
509  */
510 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
511 				    size_t *min_offs, void *base, int *total)
512 {
513 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
514 		*total += EXT4_XATTR_LEN(last->e_name_len);
515 		if (!last->e_value_block && last->e_value_size) {
516 			size_t offs = le16_to_cpu(last->e_value_offs);
517 			if (offs < *min_offs)
518 				*min_offs = offs;
519 		}
520 	}
521 	return (*min_offs - ((void *)last - base) - sizeof(__u32));
522 }
523 
524 struct ext4_xattr_info {
525 	int name_index;
526 	const char *name;
527 	const void *value;
528 	size_t value_len;
529 };
530 
531 struct ext4_xattr_search {
532 	struct ext4_xattr_entry *first;
533 	void *base;
534 	void *end;
535 	struct ext4_xattr_entry *here;
536 	int not_found;
537 };
538 
539 static int
540 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
541 {
542 	struct ext4_xattr_entry *last;
543 	size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
544 
545 	/* Compute min_offs and last. */
546 	last = s->first;
547 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
548 		if (!last->e_value_block && last->e_value_size) {
549 			size_t offs = le16_to_cpu(last->e_value_offs);
550 			if (offs < min_offs)
551 				min_offs = offs;
552 		}
553 	}
554 	free = min_offs - ((void *)last - s->base) - sizeof(__u32);
555 	if (!s->not_found) {
556 		if (!s->here->e_value_block && s->here->e_value_size) {
557 			size_t size = le32_to_cpu(s->here->e_value_size);
558 			free += EXT4_XATTR_SIZE(size);
559 		}
560 		free += EXT4_XATTR_LEN(name_len);
561 	}
562 	if (i->value) {
563 		if (free < EXT4_XATTR_SIZE(i->value_len) ||
564 		    free < EXT4_XATTR_LEN(name_len) +
565 			   EXT4_XATTR_SIZE(i->value_len))
566 			return -ENOSPC;
567 	}
568 
569 	if (i->value && s->not_found) {
570 		/* Insert the new name. */
571 		size_t size = EXT4_XATTR_LEN(name_len);
572 		size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
573 		memmove((void *)s->here + size, s->here, rest);
574 		memset(s->here, 0, size);
575 		s->here->e_name_index = i->name_index;
576 		s->here->e_name_len = name_len;
577 		memcpy(s->here->e_name, i->name, name_len);
578 	} else {
579 		if (!s->here->e_value_block && s->here->e_value_size) {
580 			void *first_val = s->base + min_offs;
581 			size_t offs = le16_to_cpu(s->here->e_value_offs);
582 			void *val = s->base + offs;
583 			size_t size = EXT4_XATTR_SIZE(
584 				le32_to_cpu(s->here->e_value_size));
585 
586 			if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
587 				/* The old and the new value have the same
588 				   size. Just replace. */
589 				s->here->e_value_size =
590 					cpu_to_le32(i->value_len);
591 				memset(val + size - EXT4_XATTR_PAD, 0,
592 				       EXT4_XATTR_PAD); /* Clear pad bytes. */
593 				memcpy(val, i->value, i->value_len);
594 				return 0;
595 			}
596 
597 			/* Remove the old value. */
598 			memmove(first_val + size, first_val, val - first_val);
599 			memset(first_val, 0, size);
600 			s->here->e_value_size = 0;
601 			s->here->e_value_offs = 0;
602 			min_offs += size;
603 
604 			/* Adjust all value offsets. */
605 			last = s->first;
606 			while (!IS_LAST_ENTRY(last)) {
607 				size_t o = le16_to_cpu(last->e_value_offs);
608 				if (!last->e_value_block &&
609 				    last->e_value_size && o < offs)
610 					last->e_value_offs =
611 						cpu_to_le16(o + size);
612 				last = EXT4_XATTR_NEXT(last);
613 			}
614 		}
615 		if (!i->value) {
616 			/* Remove the old name. */
617 			size_t size = EXT4_XATTR_LEN(name_len);
618 			last = ENTRY((void *)last - size);
619 			memmove(s->here, (void *)s->here + size,
620 				(void *)last - (void *)s->here + sizeof(__u32));
621 			memset(last, 0, size);
622 		}
623 	}
624 
625 	if (i->value) {
626 		/* Insert the new value. */
627 		s->here->e_value_size = cpu_to_le32(i->value_len);
628 		if (i->value_len) {
629 			size_t size = EXT4_XATTR_SIZE(i->value_len);
630 			void *val = s->base + min_offs - size;
631 			s->here->e_value_offs = cpu_to_le16(min_offs - size);
632 			memset(val + size - EXT4_XATTR_PAD, 0,
633 			       EXT4_XATTR_PAD); /* Clear the pad bytes. */
634 			memcpy(val, i->value, i->value_len);
635 		}
636 	}
637 	return 0;
638 }
639 
640 struct ext4_xattr_block_find {
641 	struct ext4_xattr_search s;
642 	struct buffer_head *bh;
643 };
644 
645 static int
646 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
647 		      struct ext4_xattr_block_find *bs)
648 {
649 	struct super_block *sb = inode->i_sb;
650 	int error;
651 
652 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
653 		  i->name_index, i->name, i->value, (long)i->value_len);
654 
655 	if (EXT4_I(inode)->i_file_acl) {
656 		/* The inode already has an extended attribute block. */
657 		bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
658 		error = -EIO;
659 		if (!bs->bh)
660 			goto cleanup;
661 		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
662 			atomic_read(&(bs->bh->b_count)),
663 			le32_to_cpu(BHDR(bs->bh)->h_refcount));
664 		if (ext4_xattr_check_block(bs->bh)) {
665 			ext4_error(sb, __func__,
666 				"inode %lu: bad block %llu", inode->i_ino,
667 				EXT4_I(inode)->i_file_acl);
668 			error = -EIO;
669 			goto cleanup;
670 		}
671 		/* Find the named attribute. */
672 		bs->s.base = BHDR(bs->bh);
673 		bs->s.first = BFIRST(bs->bh);
674 		bs->s.end = bs->bh->b_data + bs->bh->b_size;
675 		bs->s.here = bs->s.first;
676 		error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
677 					      i->name, bs->bh->b_size, 1);
678 		if (error && error != -ENODATA)
679 			goto cleanup;
680 		bs->s.not_found = error;
681 	}
682 	error = 0;
683 
684 cleanup:
685 	return error;
686 }
687 
688 static int
689 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
690 		     struct ext4_xattr_info *i,
691 		     struct ext4_xattr_block_find *bs)
692 {
693 	struct super_block *sb = inode->i_sb;
694 	struct buffer_head *new_bh = NULL;
695 	struct ext4_xattr_search *s = &bs->s;
696 	struct mb_cache_entry *ce = NULL;
697 	int error = 0;
698 
699 #define header(x) ((struct ext4_xattr_header *)(x))
700 
701 	if (i->value && i->value_len > sb->s_blocksize)
702 		return -ENOSPC;
703 	if (s->base) {
704 		ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
705 					bs->bh->b_blocknr);
706 		error = ext4_journal_get_write_access(handle, bs->bh);
707 		if (error)
708 			goto cleanup;
709 		lock_buffer(bs->bh);
710 
711 		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
712 			if (ce) {
713 				mb_cache_entry_free(ce);
714 				ce = NULL;
715 			}
716 			ea_bdebug(bs->bh, "modifying in-place");
717 			error = ext4_xattr_set_entry(i, s);
718 			if (!error) {
719 				if (!IS_LAST_ENTRY(s->first))
720 					ext4_xattr_rehash(header(s->base),
721 							  s->here);
722 				ext4_xattr_cache_insert(bs->bh);
723 			}
724 			unlock_buffer(bs->bh);
725 			if (error == -EIO)
726 				goto bad_block;
727 			if (!error)
728 				error = ext4_handle_dirty_metadata(handle,
729 								   inode,
730 								   bs->bh);
731 			if (error)
732 				goto cleanup;
733 			goto inserted;
734 		} else {
735 			int offset = (char *)s->here - bs->bh->b_data;
736 
737 			unlock_buffer(bs->bh);
738 			jbd2_journal_release_buffer(handle, bs->bh);
739 			if (ce) {
740 				mb_cache_entry_release(ce);
741 				ce = NULL;
742 			}
743 			ea_bdebug(bs->bh, "cloning");
744 			s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
745 			error = -ENOMEM;
746 			if (s->base == NULL)
747 				goto cleanup;
748 			memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
749 			s->first = ENTRY(header(s->base)+1);
750 			header(s->base)->h_refcount = cpu_to_le32(1);
751 			s->here = ENTRY(s->base + offset);
752 			s->end = s->base + bs->bh->b_size;
753 		}
754 	} else {
755 		/* Allocate a buffer where we construct the new block. */
756 		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
757 		/* assert(header == s->base) */
758 		error = -ENOMEM;
759 		if (s->base == NULL)
760 			goto cleanup;
761 		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
762 		header(s->base)->h_blocks = cpu_to_le32(1);
763 		header(s->base)->h_refcount = cpu_to_le32(1);
764 		s->first = ENTRY(header(s->base)+1);
765 		s->here = ENTRY(header(s->base)+1);
766 		s->end = s->base + sb->s_blocksize;
767 	}
768 
769 	error = ext4_xattr_set_entry(i, s);
770 	if (error == -EIO)
771 		goto bad_block;
772 	if (error)
773 		goto cleanup;
774 	if (!IS_LAST_ENTRY(s->first))
775 		ext4_xattr_rehash(header(s->base), s->here);
776 
777 inserted:
778 	if (!IS_LAST_ENTRY(s->first)) {
779 		new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
780 		if (new_bh) {
781 			/* We found an identical block in the cache. */
782 			if (new_bh == bs->bh)
783 				ea_bdebug(new_bh, "keeping");
784 			else {
785 				/* The old block is released after updating
786 				   the inode. */
787 				error = -EDQUOT;
788 				if (vfs_dq_alloc_block(inode, 1))
789 					goto cleanup;
790 				error = ext4_journal_get_write_access(handle,
791 								      new_bh);
792 				if (error)
793 					goto cleanup_dquot;
794 				lock_buffer(new_bh);
795 				le32_add_cpu(&BHDR(new_bh)->h_refcount, 1);
796 				ea_bdebug(new_bh, "reusing; refcount now=%d",
797 					le32_to_cpu(BHDR(new_bh)->h_refcount));
798 				unlock_buffer(new_bh);
799 				error = ext4_handle_dirty_metadata(handle,
800 								   inode,
801 								   new_bh);
802 				if (error)
803 					goto cleanup_dquot;
804 			}
805 			mb_cache_entry_release(ce);
806 			ce = NULL;
807 		} else if (bs->bh && s->base == bs->bh->b_data) {
808 			/* We were modifying this block in-place. */
809 			ea_bdebug(bs->bh, "keeping this block");
810 			new_bh = bs->bh;
811 			get_bh(new_bh);
812 		} else {
813 			/* We need to allocate a new block */
814 			ext4_fsblk_t goal, block;
815 
816 			goal = ext4_group_first_block_no(sb,
817 						EXT4_I(inode)->i_block_group);
818 
819 			/* non-extent files can't have physical blocks past 2^32 */
820 			if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
821 				goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
822 
823 			block = ext4_new_meta_blocks(handle, inode,
824 						  goal, NULL, &error);
825 			if (error)
826 				goto cleanup;
827 
828 			if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
829 				BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
830 
831 			ea_idebug(inode, "creating block %d", block);
832 
833 			new_bh = sb_getblk(sb, block);
834 			if (!new_bh) {
835 getblk_failed:
836 				ext4_free_blocks(handle, inode, 0, block, 1,
837 						 EXT4_FREE_BLOCKS_METADATA);
838 				error = -EIO;
839 				goto cleanup;
840 			}
841 			lock_buffer(new_bh);
842 			error = ext4_journal_get_create_access(handle, new_bh);
843 			if (error) {
844 				unlock_buffer(new_bh);
845 				goto getblk_failed;
846 			}
847 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
848 			set_buffer_uptodate(new_bh);
849 			unlock_buffer(new_bh);
850 			ext4_xattr_cache_insert(new_bh);
851 			error = ext4_handle_dirty_metadata(handle,
852 							   inode, new_bh);
853 			if (error)
854 				goto cleanup;
855 		}
856 	}
857 
858 	/* Update the inode. */
859 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
860 
861 	/* Drop the previous xattr block. */
862 	if (bs->bh && bs->bh != new_bh)
863 		ext4_xattr_release_block(handle, inode, bs->bh);
864 	error = 0;
865 
866 cleanup:
867 	if (ce)
868 		mb_cache_entry_release(ce);
869 	brelse(new_bh);
870 	if (!(bs->bh && s->base == bs->bh->b_data))
871 		kfree(s->base);
872 
873 	return error;
874 
875 cleanup_dquot:
876 	vfs_dq_free_block(inode, 1);
877 	goto cleanup;
878 
879 bad_block:
880 	ext4_error(inode->i_sb, __func__,
881 		   "inode %lu: bad block %llu", inode->i_ino,
882 		   EXT4_I(inode)->i_file_acl);
883 	goto cleanup;
884 
885 #undef header
886 }
887 
888 struct ext4_xattr_ibody_find {
889 	struct ext4_xattr_search s;
890 	struct ext4_iloc iloc;
891 };
892 
893 static int
894 ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
895 		      struct ext4_xattr_ibody_find *is)
896 {
897 	struct ext4_xattr_ibody_header *header;
898 	struct ext4_inode *raw_inode;
899 	int error;
900 
901 	if (EXT4_I(inode)->i_extra_isize == 0)
902 		return 0;
903 	raw_inode = ext4_raw_inode(&is->iloc);
904 	header = IHDR(inode, raw_inode);
905 	is->s.base = is->s.first = IFIRST(header);
906 	is->s.here = is->s.first;
907 	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
908 	if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
909 		error = ext4_xattr_check_names(IFIRST(header), is->s.end);
910 		if (error)
911 			return error;
912 		/* Find the named attribute. */
913 		error = ext4_xattr_find_entry(&is->s.here, i->name_index,
914 					      i->name, is->s.end -
915 					      (void *)is->s.base, 0);
916 		if (error && error != -ENODATA)
917 			return error;
918 		is->s.not_found = error;
919 	}
920 	return 0;
921 }
922 
923 static int
924 ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
925 		     struct ext4_xattr_info *i,
926 		     struct ext4_xattr_ibody_find *is)
927 {
928 	struct ext4_xattr_ibody_header *header;
929 	struct ext4_xattr_search *s = &is->s;
930 	int error;
931 
932 	if (EXT4_I(inode)->i_extra_isize == 0)
933 		return -ENOSPC;
934 	error = ext4_xattr_set_entry(i, s);
935 	if (error)
936 		return error;
937 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
938 	if (!IS_LAST_ENTRY(s->first)) {
939 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
940 		EXT4_I(inode)->i_state |= EXT4_STATE_XATTR;
941 	} else {
942 		header->h_magic = cpu_to_le32(0);
943 		EXT4_I(inode)->i_state &= ~EXT4_STATE_XATTR;
944 	}
945 	return 0;
946 }
947 
948 /*
949  * ext4_xattr_set_handle()
950  *
951  * Create, replace or remove an extended attribute for this inode. Buffer
952  * is NULL to remove an existing extended attribute, and non-NULL to
953  * either replace an existing extended attribute, or create a new extended
954  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
955  * specify that an extended attribute must exist and must not exist
956  * previous to the call, respectively.
957  *
958  * Returns 0, or a negative error number on failure.
959  */
960 int
961 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
962 		      const char *name, const void *value, size_t value_len,
963 		      int flags)
964 {
965 	struct ext4_xattr_info i = {
966 		.name_index = name_index,
967 		.name = name,
968 		.value = value,
969 		.value_len = value_len,
970 
971 	};
972 	struct ext4_xattr_ibody_find is = {
973 		.s = { .not_found = -ENODATA, },
974 	};
975 	struct ext4_xattr_block_find bs = {
976 		.s = { .not_found = -ENODATA, },
977 	};
978 	unsigned long no_expand;
979 	int error;
980 
981 	if (!name)
982 		return -EINVAL;
983 	if (strlen(name) > 255)
984 		return -ERANGE;
985 	down_write(&EXT4_I(inode)->xattr_sem);
986 	no_expand = EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND;
987 	EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
988 
989 	error = ext4_get_inode_loc(inode, &is.iloc);
990 	if (error)
991 		goto cleanup;
992 
993 	error = ext4_journal_get_write_access(handle, is.iloc.bh);
994 	if (error)
995 		goto cleanup;
996 
997 	if (EXT4_I(inode)->i_state & EXT4_STATE_NEW) {
998 		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
999 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1000 		EXT4_I(inode)->i_state &= ~EXT4_STATE_NEW;
1001 	}
1002 
1003 	error = ext4_xattr_ibody_find(inode, &i, &is);
1004 	if (error)
1005 		goto cleanup;
1006 	if (is.s.not_found)
1007 		error = ext4_xattr_block_find(inode, &i, &bs);
1008 	if (error)
1009 		goto cleanup;
1010 	if (is.s.not_found && bs.s.not_found) {
1011 		error = -ENODATA;
1012 		if (flags & XATTR_REPLACE)
1013 			goto cleanup;
1014 		error = 0;
1015 		if (!value)
1016 			goto cleanup;
1017 	} else {
1018 		error = -EEXIST;
1019 		if (flags & XATTR_CREATE)
1020 			goto cleanup;
1021 	}
1022 	if (!value) {
1023 		if (!is.s.not_found)
1024 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1025 		else if (!bs.s.not_found)
1026 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1027 	} else {
1028 		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1029 		if (!error && !bs.s.not_found) {
1030 			i.value = NULL;
1031 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1032 		} else if (error == -ENOSPC) {
1033 			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1034 				error = ext4_xattr_block_find(inode, &i, &bs);
1035 				if (error)
1036 					goto cleanup;
1037 			}
1038 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
1039 			if (error)
1040 				goto cleanup;
1041 			if (!is.s.not_found) {
1042 				i.value = NULL;
1043 				error = ext4_xattr_ibody_set(handle, inode, &i,
1044 							     &is);
1045 			}
1046 		}
1047 	}
1048 	if (!error) {
1049 		ext4_xattr_update_super_block(handle, inode->i_sb);
1050 		inode->i_ctime = ext4_current_time(inode);
1051 		if (!value)
1052 			EXT4_I(inode)->i_state &= ~EXT4_STATE_NO_EXPAND;
1053 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1054 		/*
1055 		 * The bh is consumed by ext4_mark_iloc_dirty, even with
1056 		 * error != 0.
1057 		 */
1058 		is.iloc.bh = NULL;
1059 		if (IS_SYNC(inode))
1060 			ext4_handle_sync(handle);
1061 	}
1062 
1063 cleanup:
1064 	brelse(is.iloc.bh);
1065 	brelse(bs.bh);
1066 	if (no_expand == 0)
1067 		EXT4_I(inode)->i_state &= ~EXT4_STATE_NO_EXPAND;
1068 	up_write(&EXT4_I(inode)->xattr_sem);
1069 	return error;
1070 }
1071 
1072 /*
1073  * ext4_xattr_set()
1074  *
1075  * Like ext4_xattr_set_handle, but start from an inode. This extended
1076  * attribute modification is a filesystem transaction by itself.
1077  *
1078  * Returns 0, or a negative error number on failure.
1079  */
1080 int
1081 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1082 	       const void *value, size_t value_len, int flags)
1083 {
1084 	handle_t *handle;
1085 	int error, retries = 0;
1086 
1087 retry:
1088 	handle = ext4_journal_start(inode, EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
1089 	if (IS_ERR(handle)) {
1090 		error = PTR_ERR(handle);
1091 	} else {
1092 		int error2;
1093 
1094 		error = ext4_xattr_set_handle(handle, inode, name_index, name,
1095 					      value, value_len, flags);
1096 		error2 = ext4_journal_stop(handle);
1097 		if (error == -ENOSPC &&
1098 		    ext4_should_retry_alloc(inode->i_sb, &retries))
1099 			goto retry;
1100 		if (error == 0)
1101 			error = error2;
1102 	}
1103 
1104 	return error;
1105 }
1106 
1107 /*
1108  * Shift the EA entries in the inode to create space for the increased
1109  * i_extra_isize.
1110  */
1111 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1112 				     int value_offs_shift, void *to,
1113 				     void *from, size_t n, int blocksize)
1114 {
1115 	struct ext4_xattr_entry *last = entry;
1116 	int new_offs;
1117 
1118 	/* Adjust the value offsets of the entries */
1119 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1120 		if (!last->e_value_block && last->e_value_size) {
1121 			new_offs = le16_to_cpu(last->e_value_offs) +
1122 							value_offs_shift;
1123 			BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
1124 				 > blocksize);
1125 			last->e_value_offs = cpu_to_le16(new_offs);
1126 		}
1127 	}
1128 	/* Shift the entries by n bytes */
1129 	memmove(to, from, n);
1130 }
1131 
1132 /*
1133  * Expand an inode by new_extra_isize bytes when EAs are present.
1134  * Returns 0 on success or negative error number on failure.
1135  */
1136 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1137 			       struct ext4_inode *raw_inode, handle_t *handle)
1138 {
1139 	struct ext4_xattr_ibody_header *header;
1140 	struct ext4_xattr_entry *entry, *last, *first;
1141 	struct buffer_head *bh = NULL;
1142 	struct ext4_xattr_ibody_find *is = NULL;
1143 	struct ext4_xattr_block_find *bs = NULL;
1144 	char *buffer = NULL, *b_entry_name = NULL;
1145 	size_t min_offs, free;
1146 	int total_ino, total_blk;
1147 	void *base, *start, *end;
1148 	int extra_isize = 0, error = 0, tried_min_extra_isize = 0;
1149 	int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1150 
1151 	down_write(&EXT4_I(inode)->xattr_sem);
1152 retry:
1153 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) {
1154 		up_write(&EXT4_I(inode)->xattr_sem);
1155 		return 0;
1156 	}
1157 
1158 	header = IHDR(inode, raw_inode);
1159 	entry = IFIRST(header);
1160 
1161 	/*
1162 	 * Check if enough free space is available in the inode to shift the
1163 	 * entries ahead by new_extra_isize.
1164 	 */
1165 
1166 	base = start = entry;
1167 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1168 	min_offs = end - base;
1169 	last = entry;
1170 	total_ino = sizeof(struct ext4_xattr_ibody_header);
1171 
1172 	free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
1173 	if (free >= new_extra_isize) {
1174 		entry = IFIRST(header);
1175 		ext4_xattr_shift_entries(entry,	EXT4_I(inode)->i_extra_isize
1176 				- new_extra_isize, (void *)raw_inode +
1177 				EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1178 				(void *)header, total_ino,
1179 				inode->i_sb->s_blocksize);
1180 		EXT4_I(inode)->i_extra_isize = new_extra_isize;
1181 		error = 0;
1182 		goto cleanup;
1183 	}
1184 
1185 	/*
1186 	 * Enough free space isn't available in the inode, check if
1187 	 * EA block can hold new_extra_isize bytes.
1188 	 */
1189 	if (EXT4_I(inode)->i_file_acl) {
1190 		bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1191 		error = -EIO;
1192 		if (!bh)
1193 			goto cleanup;
1194 		if (ext4_xattr_check_block(bh)) {
1195 			ext4_error(inode->i_sb, __func__,
1196 				"inode %lu: bad block %llu", inode->i_ino,
1197 				EXT4_I(inode)->i_file_acl);
1198 			error = -EIO;
1199 			goto cleanup;
1200 		}
1201 		base = BHDR(bh);
1202 		first = BFIRST(bh);
1203 		end = bh->b_data + bh->b_size;
1204 		min_offs = end - base;
1205 		free = ext4_xattr_free_space(first, &min_offs, base,
1206 					     &total_blk);
1207 		if (free < new_extra_isize) {
1208 			if (!tried_min_extra_isize && s_min_extra_isize) {
1209 				tried_min_extra_isize++;
1210 				new_extra_isize = s_min_extra_isize;
1211 				brelse(bh);
1212 				goto retry;
1213 			}
1214 			error = -1;
1215 			goto cleanup;
1216 		}
1217 	} else {
1218 		free = inode->i_sb->s_blocksize;
1219 	}
1220 
1221 	while (new_extra_isize > 0) {
1222 		size_t offs, size, entry_size;
1223 		struct ext4_xattr_entry *small_entry = NULL;
1224 		struct ext4_xattr_info i = {
1225 			.value = NULL,
1226 			.value_len = 0,
1227 		};
1228 		unsigned int total_size;  /* EA entry size + value size */
1229 		unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
1230 		unsigned int min_total_size = ~0U;
1231 
1232 		is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1233 		bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1234 		if (!is || !bs) {
1235 			error = -ENOMEM;
1236 			goto cleanup;
1237 		}
1238 
1239 		is->s.not_found = -ENODATA;
1240 		bs->s.not_found = -ENODATA;
1241 		is->iloc.bh = NULL;
1242 		bs->bh = NULL;
1243 
1244 		last = IFIRST(header);
1245 		/* Find the entry best suited to be pushed into EA block */
1246 		entry = NULL;
1247 		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1248 			total_size =
1249 			EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1250 					EXT4_XATTR_LEN(last->e_name_len);
1251 			if (total_size <= free && total_size < min_total_size) {
1252 				if (total_size < new_extra_isize) {
1253 					small_entry = last;
1254 				} else {
1255 					entry = last;
1256 					min_total_size = total_size;
1257 				}
1258 			}
1259 		}
1260 
1261 		if (entry == NULL) {
1262 			if (small_entry) {
1263 				entry = small_entry;
1264 			} else {
1265 				if (!tried_min_extra_isize &&
1266 				    s_min_extra_isize) {
1267 					tried_min_extra_isize++;
1268 					new_extra_isize = s_min_extra_isize;
1269 					goto retry;
1270 				}
1271 				error = -1;
1272 				goto cleanup;
1273 			}
1274 		}
1275 		offs = le16_to_cpu(entry->e_value_offs);
1276 		size = le32_to_cpu(entry->e_value_size);
1277 		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1278 		i.name_index = entry->e_name_index,
1279 		buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
1280 		b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1281 		if (!buffer || !b_entry_name) {
1282 			error = -ENOMEM;
1283 			goto cleanup;
1284 		}
1285 		/* Save the entry name and the entry value */
1286 		memcpy(buffer, (void *)IFIRST(header) + offs,
1287 		       EXT4_XATTR_SIZE(size));
1288 		memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1289 		b_entry_name[entry->e_name_len] = '\0';
1290 		i.name = b_entry_name;
1291 
1292 		error = ext4_get_inode_loc(inode, &is->iloc);
1293 		if (error)
1294 			goto cleanup;
1295 
1296 		error = ext4_xattr_ibody_find(inode, &i, is);
1297 		if (error)
1298 			goto cleanup;
1299 
1300 		/* Remove the chosen entry from the inode */
1301 		error = ext4_xattr_ibody_set(handle, inode, &i, is);
1302 
1303 		entry = IFIRST(header);
1304 		if (entry_size + EXT4_XATTR_SIZE(size) >= new_extra_isize)
1305 			shift_bytes = new_extra_isize;
1306 		else
1307 			shift_bytes = entry_size + size;
1308 		/* Adjust the offsets and shift the remaining entries ahead */
1309 		ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize -
1310 			shift_bytes, (void *)raw_inode +
1311 			EXT4_GOOD_OLD_INODE_SIZE + extra_isize + shift_bytes,
1312 			(void *)header, total_ino - entry_size,
1313 			inode->i_sb->s_blocksize);
1314 
1315 		extra_isize += shift_bytes;
1316 		new_extra_isize -= shift_bytes;
1317 		EXT4_I(inode)->i_extra_isize = extra_isize;
1318 
1319 		i.name = b_entry_name;
1320 		i.value = buffer;
1321 		i.value_len = size;
1322 		error = ext4_xattr_block_find(inode, &i, bs);
1323 		if (error)
1324 			goto cleanup;
1325 
1326 		/* Add entry which was removed from the inode into the block */
1327 		error = ext4_xattr_block_set(handle, inode, &i, bs);
1328 		if (error)
1329 			goto cleanup;
1330 		kfree(b_entry_name);
1331 		kfree(buffer);
1332 		brelse(is->iloc.bh);
1333 		kfree(is);
1334 		kfree(bs);
1335 	}
1336 	brelse(bh);
1337 	up_write(&EXT4_I(inode)->xattr_sem);
1338 	return 0;
1339 
1340 cleanup:
1341 	kfree(b_entry_name);
1342 	kfree(buffer);
1343 	if (is)
1344 		brelse(is->iloc.bh);
1345 	kfree(is);
1346 	kfree(bs);
1347 	brelse(bh);
1348 	up_write(&EXT4_I(inode)->xattr_sem);
1349 	return error;
1350 }
1351 
1352 
1353 
1354 /*
1355  * ext4_xattr_delete_inode()
1356  *
1357  * Free extended attribute resources associated with this inode. This
1358  * is called immediately before an inode is freed. We have exclusive
1359  * access to the inode.
1360  */
1361 void
1362 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1363 {
1364 	struct buffer_head *bh = NULL;
1365 
1366 	if (!EXT4_I(inode)->i_file_acl)
1367 		goto cleanup;
1368 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1369 	if (!bh) {
1370 		ext4_error(inode->i_sb, __func__,
1371 			"inode %lu: block %llu read error", inode->i_ino,
1372 			EXT4_I(inode)->i_file_acl);
1373 		goto cleanup;
1374 	}
1375 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1376 	    BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1377 		ext4_error(inode->i_sb, __func__,
1378 			"inode %lu: bad block %llu", inode->i_ino,
1379 			EXT4_I(inode)->i_file_acl);
1380 		goto cleanup;
1381 	}
1382 	ext4_xattr_release_block(handle, inode, bh);
1383 	EXT4_I(inode)->i_file_acl = 0;
1384 
1385 cleanup:
1386 	brelse(bh);
1387 }
1388 
1389 /*
1390  * ext4_xattr_put_super()
1391  *
1392  * This is called when a file system is unmounted.
1393  */
1394 void
1395 ext4_xattr_put_super(struct super_block *sb)
1396 {
1397 	mb_cache_shrink(sb->s_bdev);
1398 }
1399 
1400 /*
1401  * ext4_xattr_cache_insert()
1402  *
1403  * Create a new entry in the extended attribute cache, and insert
1404  * it unless such an entry is already in the cache.
1405  *
1406  * Returns 0, or a negative error number on failure.
1407  */
1408 static void
1409 ext4_xattr_cache_insert(struct buffer_head *bh)
1410 {
1411 	__u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1412 	struct mb_cache_entry *ce;
1413 	int error;
1414 
1415 	ce = mb_cache_entry_alloc(ext4_xattr_cache, GFP_NOFS);
1416 	if (!ce) {
1417 		ea_bdebug(bh, "out of memory");
1418 		return;
1419 	}
1420 	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash);
1421 	if (error) {
1422 		mb_cache_entry_free(ce);
1423 		if (error == -EBUSY) {
1424 			ea_bdebug(bh, "already in cache");
1425 			error = 0;
1426 		}
1427 	} else {
1428 		ea_bdebug(bh, "inserting [%x]", (int)hash);
1429 		mb_cache_entry_release(ce);
1430 	}
1431 }
1432 
1433 /*
1434  * ext4_xattr_cmp()
1435  *
1436  * Compare two extended attribute blocks for equality.
1437  *
1438  * Returns 0 if the blocks are equal, 1 if they differ, and
1439  * a negative error number on errors.
1440  */
1441 static int
1442 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1443 	       struct ext4_xattr_header *header2)
1444 {
1445 	struct ext4_xattr_entry *entry1, *entry2;
1446 
1447 	entry1 = ENTRY(header1+1);
1448 	entry2 = ENTRY(header2+1);
1449 	while (!IS_LAST_ENTRY(entry1)) {
1450 		if (IS_LAST_ENTRY(entry2))
1451 			return 1;
1452 		if (entry1->e_hash != entry2->e_hash ||
1453 		    entry1->e_name_index != entry2->e_name_index ||
1454 		    entry1->e_name_len != entry2->e_name_len ||
1455 		    entry1->e_value_size != entry2->e_value_size ||
1456 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1457 			return 1;
1458 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1459 			return -EIO;
1460 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1461 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1462 			   le32_to_cpu(entry1->e_value_size)))
1463 			return 1;
1464 
1465 		entry1 = EXT4_XATTR_NEXT(entry1);
1466 		entry2 = EXT4_XATTR_NEXT(entry2);
1467 	}
1468 	if (!IS_LAST_ENTRY(entry2))
1469 		return 1;
1470 	return 0;
1471 }
1472 
1473 /*
1474  * ext4_xattr_cache_find()
1475  *
1476  * Find an identical extended attribute block.
1477  *
1478  * Returns a pointer to the block found, or NULL if such a block was
1479  * not found or an error occurred.
1480  */
1481 static struct buffer_head *
1482 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1483 		      struct mb_cache_entry **pce)
1484 {
1485 	__u32 hash = le32_to_cpu(header->h_hash);
1486 	struct mb_cache_entry *ce;
1487 
1488 	if (!header->h_hash)
1489 		return NULL;  /* never share */
1490 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1491 again:
1492 	ce = mb_cache_entry_find_first(ext4_xattr_cache, 0,
1493 				       inode->i_sb->s_bdev, hash);
1494 	while (ce) {
1495 		struct buffer_head *bh;
1496 
1497 		if (IS_ERR(ce)) {
1498 			if (PTR_ERR(ce) == -EAGAIN)
1499 				goto again;
1500 			break;
1501 		}
1502 		bh = sb_bread(inode->i_sb, ce->e_block);
1503 		if (!bh) {
1504 			ext4_error(inode->i_sb, __func__,
1505 				"inode %lu: block %lu read error",
1506 				inode->i_ino, (unsigned long) ce->e_block);
1507 		} else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1508 				EXT4_XATTR_REFCOUNT_MAX) {
1509 			ea_idebug(inode, "block %lu refcount %d>=%d",
1510 				  (unsigned long) ce->e_block,
1511 				  le32_to_cpu(BHDR(bh)->h_refcount),
1512 					  EXT4_XATTR_REFCOUNT_MAX);
1513 		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1514 			*pce = ce;
1515 			return bh;
1516 		}
1517 		brelse(bh);
1518 		ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash);
1519 	}
1520 	return NULL;
1521 }
1522 
1523 #define NAME_HASH_SHIFT 5
1524 #define VALUE_HASH_SHIFT 16
1525 
1526 /*
1527  * ext4_xattr_hash_entry()
1528  *
1529  * Compute the hash of an extended attribute.
1530  */
1531 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1532 					 struct ext4_xattr_entry *entry)
1533 {
1534 	__u32 hash = 0;
1535 	char *name = entry->e_name;
1536 	int n;
1537 
1538 	for (n = 0; n < entry->e_name_len; n++) {
1539 		hash = (hash << NAME_HASH_SHIFT) ^
1540 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1541 		       *name++;
1542 	}
1543 
1544 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1545 		__le32 *value = (__le32 *)((char *)header +
1546 			le16_to_cpu(entry->e_value_offs));
1547 		for (n = (le32_to_cpu(entry->e_value_size) +
1548 		     EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1549 			hash = (hash << VALUE_HASH_SHIFT) ^
1550 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1551 			       le32_to_cpu(*value++);
1552 		}
1553 	}
1554 	entry->e_hash = cpu_to_le32(hash);
1555 }
1556 
1557 #undef NAME_HASH_SHIFT
1558 #undef VALUE_HASH_SHIFT
1559 
1560 #define BLOCK_HASH_SHIFT 16
1561 
1562 /*
1563  * ext4_xattr_rehash()
1564  *
1565  * Re-compute the extended attribute hash value after an entry has changed.
1566  */
1567 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1568 			      struct ext4_xattr_entry *entry)
1569 {
1570 	struct ext4_xattr_entry *here;
1571 	__u32 hash = 0;
1572 
1573 	ext4_xattr_hash_entry(header, entry);
1574 	here = ENTRY(header+1);
1575 	while (!IS_LAST_ENTRY(here)) {
1576 		if (!here->e_hash) {
1577 			/* Block is not shared if an entry's hash value == 0 */
1578 			hash = 0;
1579 			break;
1580 		}
1581 		hash = (hash << BLOCK_HASH_SHIFT) ^
1582 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1583 		       le32_to_cpu(here->e_hash);
1584 		here = EXT4_XATTR_NEXT(here);
1585 	}
1586 	header->h_hash = cpu_to_le32(hash);
1587 }
1588 
1589 #undef BLOCK_HASH_SHIFT
1590 
1591 int __init
1592 init_ext4_xattr(void)
1593 {
1594 	ext4_xattr_cache = mb_cache_create("ext4_xattr", NULL,
1595 		sizeof(struct mb_cache_entry) +
1596 		sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6);
1597 	if (!ext4_xattr_cache)
1598 		return -ENOMEM;
1599 	return 0;
1600 }
1601 
1602 void
1603 exit_ext4_xattr(void)
1604 {
1605 	if (ext4_xattr_cache)
1606 		mb_cache_destroy(ext4_xattr_cache);
1607 	ext4_xattr_cache = NULL;
1608 }
1609