xref: /openbmc/linux/fs/ext2/xattr.c (revision 1eaf5faa)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/ext2/xattr.c
4  *
5  * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
6  *
7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
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  *
13  */
14 
15 /*
16  * Extended attributes are stored on disk blocks allocated outside of
17  * any inode. The i_file_acl field is then made to point to this allocated
18  * block. If all extended attributes of an inode are identical, these
19  * inodes may share the same extended attribute block. Such situations
20  * are automatically detected by keeping a cache of recent attribute block
21  * numbers and hashes over the block's contents in memory.
22  *
23  *
24  * Extended attribute block layout:
25  *
26  *   +------------------+
27  *   | header           |
28  *   | entry 1          | |
29  *   | entry 2          | | growing downwards
30  *   | entry 3          | v
31  *   | four null bytes  |
32  *   | . . .            |
33  *   | value 1          | ^
34  *   | value 3          | | growing upwards
35  *   | value 2          | |
36  *   +------------------+
37  *
38  * The block header is followed by multiple entry descriptors. These entry
39  * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
40  * byte boundaries. The entry descriptors are sorted by attribute name,
41  * so that two extended attribute blocks can be compared efficiently.
42  *
43  * Attribute values are aligned to the end of the block, stored in
44  * no specific order. They are also padded to EXT2_XATTR_PAD byte
45  * boundaries. No additional gaps are left between them.
46  *
47  * Locking strategy
48  * ----------------
49  * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
50  * EA blocks are only changed if they are exclusive to an inode, so
51  * holding xattr_sem also means that nothing but the EA block's reference
52  * count will change. Multiple writers to an EA block are synchronized
53  * by the bh lock. No more than a single bh lock is held at any time
54  * to avoid deadlocks.
55  */
56 
57 #include <linux/buffer_head.h>
58 #include <linux/init.h>
59 #include <linux/slab.h>
60 #include <linux/mbcache.h>
61 #include <linux/quotaops.h>
62 #include <linux/rwsem.h>
63 #include <linux/security.h>
64 #include "ext2.h"
65 #include "xattr.h"
66 #include "acl.h"
67 
68 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
69 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
70 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
71 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
72 
73 #ifdef EXT2_XATTR_DEBUG
74 # define ea_idebug(inode, f...) do { \
75 		printk(KERN_DEBUG "inode %s:%ld: ", \
76 			inode->i_sb->s_id, inode->i_ino); \
77 		printk(f); \
78 		printk("\n"); \
79 	} while (0)
80 # define ea_bdebug(bh, f...) do { \
81 		printk(KERN_DEBUG "block %pg:%lu: ", \
82 			bh->b_bdev, (unsigned long) bh->b_blocknr); \
83 		printk(f); \
84 		printk("\n"); \
85 	} while (0)
86 #else
87 # define ea_idebug(f...)
88 # define ea_bdebug(f...)
89 #endif
90 
91 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
92 			   struct ext2_xattr_header *);
93 
94 static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
95 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
96 						 struct ext2_xattr_header *);
97 static void ext2_xattr_rehash(struct ext2_xattr_header *,
98 			      struct ext2_xattr_entry *);
99 
100 static const struct xattr_handler *ext2_xattr_handler_map[] = {
101 	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
102 #ifdef CONFIG_EXT2_FS_POSIX_ACL
103 	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
104 	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
105 #endif
106 	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
107 #ifdef CONFIG_EXT2_FS_SECURITY
108 	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
109 #endif
110 };
111 
112 const struct xattr_handler *ext2_xattr_handlers[] = {
113 	&ext2_xattr_user_handler,
114 	&ext2_xattr_trusted_handler,
115 #ifdef CONFIG_EXT2_FS_POSIX_ACL
116 	&posix_acl_access_xattr_handler,
117 	&posix_acl_default_xattr_handler,
118 #endif
119 #ifdef CONFIG_EXT2_FS_SECURITY
120 	&ext2_xattr_security_handler,
121 #endif
122 	NULL
123 };
124 
125 #define EA_BLOCK_CACHE(inode)	(EXT2_SB(inode->i_sb)->s_ea_block_cache)
126 
127 static inline const struct xattr_handler *
128 ext2_xattr_handler(int name_index)
129 {
130 	const struct xattr_handler *handler = NULL;
131 
132 	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
133 		handler = ext2_xattr_handler_map[name_index];
134 	return handler;
135 }
136 
137 static bool
138 ext2_xattr_header_valid(struct ext2_xattr_header *header)
139 {
140 	if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
141 	    header->h_blocks != cpu_to_le32(1))
142 		return false;
143 
144 	return true;
145 }
146 
147 static bool
148 ext2_xattr_entry_valid(struct ext2_xattr_entry *entry,
149 		       char *end, size_t end_offs)
150 {
151 	struct ext2_xattr_entry *next;
152 	size_t size;
153 
154 	next = EXT2_XATTR_NEXT(entry);
155 	if ((char *)next >= end)
156 		return false;
157 
158 	if (entry->e_value_block != 0)
159 		return false;
160 
161 	size = le32_to_cpu(entry->e_value_size);
162 	if (size > end_offs ||
163 	    le16_to_cpu(entry->e_value_offs) + size > end_offs)
164 		return false;
165 
166 	return true;
167 }
168 
169 static int
170 ext2_xattr_cmp_entry(int name_index, size_t name_len, const char *name,
171 		     struct ext2_xattr_entry *entry)
172 {
173 	int cmp;
174 
175 	cmp = name_index - entry->e_name_index;
176 	if (!cmp)
177 		cmp = name_len - entry->e_name_len;
178 	if (!cmp)
179 		cmp = memcmp(name, entry->e_name, name_len);
180 
181 	return cmp;
182 }
183 
184 /*
185  * ext2_xattr_get()
186  *
187  * Copy an extended attribute into the buffer
188  * provided, or compute the buffer size required.
189  * Buffer is NULL to compute the size of the buffer required.
190  *
191  * Returns a negative error number on failure, or the number of bytes
192  * used / required on success.
193  */
194 int
195 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
196 	       void *buffer, size_t buffer_size)
197 {
198 	struct buffer_head *bh = NULL;
199 	struct ext2_xattr_entry *entry;
200 	size_t name_len, size;
201 	char *end;
202 	int error, not_found;
203 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
204 
205 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
206 		  name_index, name, buffer, (long)buffer_size);
207 
208 	if (name == NULL)
209 		return -EINVAL;
210 	name_len = strlen(name);
211 	if (name_len > 255)
212 		return -ERANGE;
213 
214 	down_read(&EXT2_I(inode)->xattr_sem);
215 	error = -ENODATA;
216 	if (!EXT2_I(inode)->i_file_acl)
217 		goto cleanup;
218 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
219 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
220 	error = -EIO;
221 	if (!bh)
222 		goto cleanup;
223 	ea_bdebug(bh, "b_count=%d, refcount=%d",
224 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
225 	end = bh->b_data + bh->b_size;
226 	if (!ext2_xattr_header_valid(HDR(bh))) {
227 bad_block:
228 		ext2_error(inode->i_sb, "ext2_xattr_get",
229 			"inode %ld: bad block %d", inode->i_ino,
230 			EXT2_I(inode)->i_file_acl);
231 		error = -EIO;
232 		goto cleanup;
233 	}
234 
235 	/* find named attribute */
236 	entry = FIRST_ENTRY(bh);
237 	while (!IS_LAST_ENTRY(entry)) {
238 		if (!ext2_xattr_entry_valid(entry, end,
239 		    inode->i_sb->s_blocksize))
240 			goto bad_block;
241 
242 		not_found = ext2_xattr_cmp_entry(name_index, name_len, name,
243 						 entry);
244 		if (!not_found)
245 			goto found;
246 		if (not_found < 0)
247 			break;
248 
249 		entry = EXT2_XATTR_NEXT(entry);
250 	}
251 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
252 		ea_idebug(inode, "cache insert failed");
253 	error = -ENODATA;
254 	goto cleanup;
255 found:
256 	size = le32_to_cpu(entry->e_value_size);
257 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
258 		ea_idebug(inode, "cache insert failed");
259 	if (buffer) {
260 		error = -ERANGE;
261 		if (size > buffer_size)
262 			goto cleanup;
263 		/* return value of attribute */
264 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
265 			size);
266 	}
267 	error = size;
268 
269 cleanup:
270 	brelse(bh);
271 	up_read(&EXT2_I(inode)->xattr_sem);
272 
273 	return error;
274 }
275 
276 /*
277  * ext2_xattr_list()
278  *
279  * Copy a list of attribute names into the buffer
280  * provided, or compute the buffer size required.
281  * Buffer is NULL to compute the size of the buffer required.
282  *
283  * Returns a negative error number on failure, or the number of bytes
284  * used / required on success.
285  */
286 static int
287 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
288 {
289 	struct inode *inode = d_inode(dentry);
290 	struct buffer_head *bh = NULL;
291 	struct ext2_xattr_entry *entry;
292 	char *end;
293 	size_t rest = buffer_size;
294 	int error;
295 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
296 
297 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
298 		  buffer, (long)buffer_size);
299 
300 	down_read(&EXT2_I(inode)->xattr_sem);
301 	error = 0;
302 	if (!EXT2_I(inode)->i_file_acl)
303 		goto cleanup;
304 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
305 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
306 	error = -EIO;
307 	if (!bh)
308 		goto cleanup;
309 	ea_bdebug(bh, "b_count=%d, refcount=%d",
310 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
311 	end = bh->b_data + bh->b_size;
312 	if (!ext2_xattr_header_valid(HDR(bh))) {
313 bad_block:
314 		ext2_error(inode->i_sb, "ext2_xattr_list",
315 			"inode %ld: bad block %d", inode->i_ino,
316 			EXT2_I(inode)->i_file_acl);
317 		error = -EIO;
318 		goto cleanup;
319 	}
320 
321 	/* check the on-disk data structure */
322 	entry = FIRST_ENTRY(bh);
323 	while (!IS_LAST_ENTRY(entry)) {
324 		if (!ext2_xattr_entry_valid(entry, end,
325 		    inode->i_sb->s_blocksize))
326 			goto bad_block;
327 		entry = EXT2_XATTR_NEXT(entry);
328 	}
329 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
330 		ea_idebug(inode, "cache insert failed");
331 
332 	/* list the attribute names */
333 	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
334 	     entry = EXT2_XATTR_NEXT(entry)) {
335 		const struct xattr_handler *handler =
336 			ext2_xattr_handler(entry->e_name_index);
337 
338 		if (handler && (!handler->list || handler->list(dentry))) {
339 			const char *prefix = handler->prefix ?: handler->name;
340 			size_t prefix_len = strlen(prefix);
341 			size_t size = prefix_len + entry->e_name_len + 1;
342 
343 			if (buffer) {
344 				if (size > rest) {
345 					error = -ERANGE;
346 					goto cleanup;
347 				}
348 				memcpy(buffer, prefix, prefix_len);
349 				buffer += prefix_len;
350 				memcpy(buffer, entry->e_name, entry->e_name_len);
351 				buffer += entry->e_name_len;
352 				*buffer++ = 0;
353 			}
354 			rest -= size;
355 		}
356 	}
357 	error = buffer_size - rest;  /* total size */
358 
359 cleanup:
360 	brelse(bh);
361 	up_read(&EXT2_I(inode)->xattr_sem);
362 
363 	return error;
364 }
365 
366 /*
367  * Inode operation listxattr()
368  *
369  * d_inode(dentry)->i_mutex: don't care
370  */
371 ssize_t
372 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
373 {
374 	return ext2_xattr_list(dentry, buffer, size);
375 }
376 
377 /*
378  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
379  * not set, set it.
380  */
381 static void ext2_xattr_update_super_block(struct super_block *sb)
382 {
383 	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
384 		return;
385 
386 	spin_lock(&EXT2_SB(sb)->s_lock);
387 	ext2_update_dynamic_rev(sb);
388 	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
389 	spin_unlock(&EXT2_SB(sb)->s_lock);
390 	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
391 }
392 
393 /*
394  * ext2_xattr_set()
395  *
396  * Create, replace or remove an extended attribute for this inode.  Value
397  * is NULL to remove an existing extended attribute, and non-NULL to
398  * either replace an existing extended attribute, or create a new extended
399  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
400  * specify that an extended attribute must exist and must not exist
401  * previous to the call, respectively.
402  *
403  * Returns 0, or a negative error number on failure.
404  */
405 int
406 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
407 	       const void *value, size_t value_len, int flags)
408 {
409 	struct super_block *sb = inode->i_sb;
410 	struct buffer_head *bh = NULL;
411 	struct ext2_xattr_header *header = NULL;
412 	struct ext2_xattr_entry *here = NULL, *last = NULL;
413 	size_t name_len, free, min_offs = sb->s_blocksize;
414 	int not_found = 1, error;
415 	char *end;
416 
417 	/*
418 	 * header -- Points either into bh, or to a temporarily
419 	 *           allocated buffer.
420 	 * here -- The named entry found, or the place for inserting, within
421 	 *         the block pointed to by header.
422 	 * last -- Points right after the last named entry within the block
423 	 *         pointed to by header.
424 	 * min_offs -- The offset of the first value (values are aligned
425 	 *             towards the end of the block).
426 	 * end -- Points right after the block pointed to by header.
427 	 */
428 
429 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
430 		  name_index, name, value, (long)value_len);
431 
432 	if (value == NULL)
433 		value_len = 0;
434 	if (name == NULL)
435 		return -EINVAL;
436 	name_len = strlen(name);
437 	if (name_len > 255 || value_len > sb->s_blocksize)
438 		return -ERANGE;
439 	down_write(&EXT2_I(inode)->xattr_sem);
440 	if (EXT2_I(inode)->i_file_acl) {
441 		/* The inode already has an extended attribute block. */
442 		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
443 		error = -EIO;
444 		if (!bh)
445 			goto cleanup;
446 		ea_bdebug(bh, "b_count=%d, refcount=%d",
447 			atomic_read(&(bh->b_count)),
448 			le32_to_cpu(HDR(bh)->h_refcount));
449 		header = HDR(bh);
450 		end = bh->b_data + bh->b_size;
451 		if (!ext2_xattr_header_valid(header)) {
452 bad_block:
453 			ext2_error(sb, "ext2_xattr_set",
454 				"inode %ld: bad block %d", inode->i_ino,
455 				   EXT2_I(inode)->i_file_acl);
456 			error = -EIO;
457 			goto cleanup;
458 		}
459 		/*
460 		 * Find the named attribute. If not found, 'here' will point
461 		 * to entry where the new attribute should be inserted to
462 		 * maintain sorting.
463 		 */
464 		last = FIRST_ENTRY(bh);
465 		while (!IS_LAST_ENTRY(last)) {
466 			if (!ext2_xattr_entry_valid(last, end, sb->s_blocksize))
467 				goto bad_block;
468 			if (last->e_value_size) {
469 				size_t offs = le16_to_cpu(last->e_value_offs);
470 				if (offs < min_offs)
471 					min_offs = offs;
472 			}
473 			if (not_found > 0) {
474 				not_found = ext2_xattr_cmp_entry(name_index,
475 								 name_len,
476 								 name, last);
477 				if (not_found <= 0)
478 					here = last;
479 			}
480 			last = EXT2_XATTR_NEXT(last);
481 		}
482 		if (not_found > 0)
483 			here = last;
484 
485 		/* Check whether we have enough space left. */
486 		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
487 	} else {
488 		/* We will use a new extended attribute block. */
489 		free = sb->s_blocksize -
490 			sizeof(struct ext2_xattr_header) - sizeof(__u32);
491 	}
492 
493 	if (not_found) {
494 		/* Request to remove a nonexistent attribute? */
495 		error = -ENODATA;
496 		if (flags & XATTR_REPLACE)
497 			goto cleanup;
498 		error = 0;
499 		if (value == NULL)
500 			goto cleanup;
501 	} else {
502 		/* Request to create an existing attribute? */
503 		error = -EEXIST;
504 		if (flags & XATTR_CREATE)
505 			goto cleanup;
506 		free += EXT2_XATTR_SIZE(le32_to_cpu(here->e_value_size));
507 		free += EXT2_XATTR_LEN(name_len);
508 	}
509 	error = -ENOSPC;
510 	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
511 		goto cleanup;
512 
513 	/* Here we know that we can set the new attribute. */
514 
515 	if (header) {
516 		/* assert(header == HDR(bh)); */
517 		lock_buffer(bh);
518 		if (header->h_refcount == cpu_to_le32(1)) {
519 			__u32 hash = le32_to_cpu(header->h_hash);
520 
521 			ea_bdebug(bh, "modifying in-place");
522 			/*
523 			 * This must happen under buffer lock for
524 			 * ext2_xattr_set2() to reliably detect modified block
525 			 */
526 			mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
527 					      bh->b_blocknr);
528 
529 			/* keep the buffer locked while modifying it. */
530 		} else {
531 			int offset;
532 
533 			unlock_buffer(bh);
534 			ea_bdebug(bh, "cloning");
535 			header = kmalloc(bh->b_size, GFP_KERNEL);
536 			error = -ENOMEM;
537 			if (header == NULL)
538 				goto cleanup;
539 			memcpy(header, HDR(bh), bh->b_size);
540 			header->h_refcount = cpu_to_le32(1);
541 
542 			offset = (char *)here - bh->b_data;
543 			here = ENTRY((char *)header + offset);
544 			offset = (char *)last - bh->b_data;
545 			last = ENTRY((char *)header + offset);
546 		}
547 	} else {
548 		/* Allocate a buffer where we construct the new block. */
549 		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
550 		error = -ENOMEM;
551 		if (header == NULL)
552 			goto cleanup;
553 		end = (char *)header + sb->s_blocksize;
554 		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
555 		header->h_blocks = header->h_refcount = cpu_to_le32(1);
556 		last = here = ENTRY(header+1);
557 	}
558 
559 	/* Iff we are modifying the block in-place, bh is locked here. */
560 
561 	if (not_found) {
562 		/* Insert the new name. */
563 		size_t size = EXT2_XATTR_LEN(name_len);
564 		size_t rest = (char *)last - (char *)here;
565 		memmove((char *)here + size, here, rest);
566 		memset(here, 0, size);
567 		here->e_name_index = name_index;
568 		here->e_name_len = name_len;
569 		memcpy(here->e_name, name, name_len);
570 	} else {
571 		if (here->e_value_size) {
572 			char *first_val = (char *)header + min_offs;
573 			size_t offs = le16_to_cpu(here->e_value_offs);
574 			char *val = (char *)header + offs;
575 			size_t size = EXT2_XATTR_SIZE(
576 				le32_to_cpu(here->e_value_size));
577 
578 			if (size == EXT2_XATTR_SIZE(value_len)) {
579 				/* The old and the new value have the same
580 				   size. Just replace. */
581 				here->e_value_size = cpu_to_le32(value_len);
582 				memset(val + size - EXT2_XATTR_PAD, 0,
583 				       EXT2_XATTR_PAD); /* Clear pad bytes. */
584 				memcpy(val, value, value_len);
585 				goto skip_replace;
586 			}
587 
588 			/* Remove the old value. */
589 			memmove(first_val + size, first_val, val - first_val);
590 			memset(first_val, 0, size);
591 			here->e_value_offs = 0;
592 			min_offs += size;
593 
594 			/* Adjust all value offsets. */
595 			last = ENTRY(header+1);
596 			while (!IS_LAST_ENTRY(last)) {
597 				size_t o = le16_to_cpu(last->e_value_offs);
598 				if (o < offs)
599 					last->e_value_offs =
600 						cpu_to_le16(o + size);
601 				last = EXT2_XATTR_NEXT(last);
602 			}
603 		}
604 		if (value == NULL) {
605 			/* Remove the old name. */
606 			size_t size = EXT2_XATTR_LEN(name_len);
607 			last = ENTRY((char *)last - size);
608 			memmove(here, (char*)here + size,
609 				(char*)last - (char*)here);
610 			memset(last, 0, size);
611 		}
612 	}
613 
614 	if (value != NULL) {
615 		/* Insert the new value. */
616 		here->e_value_size = cpu_to_le32(value_len);
617 		if (value_len) {
618 			size_t size = EXT2_XATTR_SIZE(value_len);
619 			char *val = (char *)header + min_offs - size;
620 			here->e_value_offs =
621 				cpu_to_le16((char *)val - (char *)header);
622 			memset(val + size - EXT2_XATTR_PAD, 0,
623 			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
624 			memcpy(val, value, value_len);
625 		}
626 	}
627 
628 skip_replace:
629 	if (IS_LAST_ENTRY(ENTRY(header+1))) {
630 		/* This block is now empty. */
631 		if (bh && header == HDR(bh))
632 			unlock_buffer(bh);  /* we were modifying in-place. */
633 		error = ext2_xattr_set2(inode, bh, NULL);
634 	} else {
635 		ext2_xattr_rehash(header, here);
636 		if (bh && header == HDR(bh))
637 			unlock_buffer(bh);  /* we were modifying in-place. */
638 		error = ext2_xattr_set2(inode, bh, header);
639 	}
640 
641 cleanup:
642 	if (!(bh && header == HDR(bh)))
643 		kfree(header);
644 	brelse(bh);
645 	up_write(&EXT2_I(inode)->xattr_sem);
646 
647 	return error;
648 }
649 
650 /*
651  * Second half of ext2_xattr_set(): Update the file system.
652  */
653 static int
654 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
655 		struct ext2_xattr_header *header)
656 {
657 	struct super_block *sb = inode->i_sb;
658 	struct buffer_head *new_bh = NULL;
659 	int error;
660 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
661 
662 	if (header) {
663 		new_bh = ext2_xattr_cache_find(inode, header);
664 		if (new_bh) {
665 			/* We found an identical block in the cache. */
666 			if (new_bh == old_bh) {
667 				ea_bdebug(new_bh, "keeping this block");
668 			} else {
669 				/* The old block is released after updating
670 				   the inode.  */
671 				ea_bdebug(new_bh, "reusing block");
672 
673 				error = dquot_alloc_block(inode, 1);
674 				if (error) {
675 					unlock_buffer(new_bh);
676 					goto cleanup;
677 				}
678 				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
679 				ea_bdebug(new_bh, "refcount now=%d",
680 					le32_to_cpu(HDR(new_bh)->h_refcount));
681 			}
682 			unlock_buffer(new_bh);
683 		} else if (old_bh && header == HDR(old_bh)) {
684 			/* Keep this block. No need to lock the block as we
685 			   don't need to change the reference count. */
686 			new_bh = old_bh;
687 			get_bh(new_bh);
688 			ext2_xattr_cache_insert(ea_block_cache, new_bh);
689 		} else {
690 			/* We need to allocate a new block */
691 			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
692 						EXT2_I(inode)->i_block_group);
693 			int block = ext2_new_block(inode, goal, &error);
694 			if (error)
695 				goto cleanup;
696 			ea_idebug(inode, "creating block %d", block);
697 
698 			new_bh = sb_getblk(sb, block);
699 			if (unlikely(!new_bh)) {
700 				ext2_free_blocks(inode, block, 1);
701 				mark_inode_dirty(inode);
702 				error = -ENOMEM;
703 				goto cleanup;
704 			}
705 			lock_buffer(new_bh);
706 			memcpy(new_bh->b_data, header, new_bh->b_size);
707 			set_buffer_uptodate(new_bh);
708 			unlock_buffer(new_bh);
709 			ext2_xattr_cache_insert(ea_block_cache, new_bh);
710 
711 			ext2_xattr_update_super_block(sb);
712 		}
713 		mark_buffer_dirty(new_bh);
714 		if (IS_SYNC(inode)) {
715 			sync_dirty_buffer(new_bh);
716 			error = -EIO;
717 			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
718 				goto cleanup;
719 		}
720 	}
721 
722 	/* Update the inode. */
723 	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
724 	inode->i_ctime = current_time(inode);
725 	if (IS_SYNC(inode)) {
726 		error = sync_inode_metadata(inode, 1);
727 		/* In case sync failed due to ENOSPC the inode was actually
728 		 * written (only some dirty data were not) so we just proceed
729 		 * as if nothing happened and cleanup the unused block */
730 		if (error && error != -ENOSPC) {
731 			if (new_bh && new_bh != old_bh) {
732 				dquot_free_block_nodirty(inode, 1);
733 				mark_inode_dirty(inode);
734 			}
735 			goto cleanup;
736 		}
737 	} else
738 		mark_inode_dirty(inode);
739 
740 	error = 0;
741 	if (old_bh && old_bh != new_bh) {
742 		/*
743 		 * If there was an old block and we are no longer using it,
744 		 * release the old block.
745 		 */
746 		lock_buffer(old_bh);
747 		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
748 			__u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
749 
750 			/*
751 			 * This must happen under buffer lock for
752 			 * ext2_xattr_set2() to reliably detect freed block
753 			 */
754 			mb_cache_entry_delete(ea_block_cache, hash,
755 					      old_bh->b_blocknr);
756 			/* Free the old block. */
757 			ea_bdebug(old_bh, "freeing");
758 			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
759 			mark_inode_dirty(inode);
760 			/* We let our caller release old_bh, so we
761 			 * need to duplicate the buffer before. */
762 			get_bh(old_bh);
763 			bforget(old_bh);
764 		} else {
765 			/* Decrement the refcount only. */
766 			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
767 			dquot_free_block_nodirty(inode, 1);
768 			mark_inode_dirty(inode);
769 			mark_buffer_dirty(old_bh);
770 			ea_bdebug(old_bh, "refcount now=%d",
771 				le32_to_cpu(HDR(old_bh)->h_refcount));
772 		}
773 		unlock_buffer(old_bh);
774 	}
775 
776 cleanup:
777 	brelse(new_bh);
778 
779 	return error;
780 }
781 
782 /*
783  * ext2_xattr_delete_inode()
784  *
785  * Free extended attribute resources associated with this inode. This
786  * is called immediately before an inode is freed.
787  */
788 void
789 ext2_xattr_delete_inode(struct inode *inode)
790 {
791 	struct buffer_head *bh = NULL;
792 	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
793 
794 	down_write(&EXT2_I(inode)->xattr_sem);
795 	if (!EXT2_I(inode)->i_file_acl)
796 		goto cleanup;
797 
798 	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 0)) {
799 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
800 			"inode %ld: xattr block %d is out of data blocks range",
801 			inode->i_ino, EXT2_I(inode)->i_file_acl);
802 		goto cleanup;
803 	}
804 
805 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
806 	if (!bh) {
807 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
808 			"inode %ld: block %d read error", inode->i_ino,
809 			EXT2_I(inode)->i_file_acl);
810 		goto cleanup;
811 	}
812 	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
813 	if (!ext2_xattr_header_valid(HDR(bh))) {
814 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
815 			"inode %ld: bad block %d", inode->i_ino,
816 			EXT2_I(inode)->i_file_acl);
817 		goto cleanup;
818 	}
819 	lock_buffer(bh);
820 	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
821 		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
822 
823 		/*
824 		 * This must happen under buffer lock for ext2_xattr_set2() to
825 		 * reliably detect freed block
826 		 */
827 		mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
828 				      bh->b_blocknr);
829 		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
830 		get_bh(bh);
831 		bforget(bh);
832 		unlock_buffer(bh);
833 	} else {
834 		le32_add_cpu(&HDR(bh)->h_refcount, -1);
835 		ea_bdebug(bh, "refcount now=%d",
836 			le32_to_cpu(HDR(bh)->h_refcount));
837 		unlock_buffer(bh);
838 		mark_buffer_dirty(bh);
839 		if (IS_SYNC(inode))
840 			sync_dirty_buffer(bh);
841 		dquot_free_block_nodirty(inode, 1);
842 	}
843 	EXT2_I(inode)->i_file_acl = 0;
844 
845 cleanup:
846 	brelse(bh);
847 	up_write(&EXT2_I(inode)->xattr_sem);
848 }
849 
850 /*
851  * ext2_xattr_cache_insert()
852  *
853  * Create a new entry in the extended attribute cache, and insert
854  * it unless such an entry is already in the cache.
855  *
856  * Returns 0, or a negative error number on failure.
857  */
858 static int
859 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
860 {
861 	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
862 	int error;
863 
864 	error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr,
865 				      true);
866 	if (error) {
867 		if (error == -EBUSY) {
868 			ea_bdebug(bh, "already in cache (%d cache entries)",
869 				atomic_read(&ext2_xattr_cache->c_entry_count));
870 			error = 0;
871 		}
872 	} else
873 		ea_bdebug(bh, "inserting [%x]", (int)hash);
874 	return error;
875 }
876 
877 /*
878  * ext2_xattr_cmp()
879  *
880  * Compare two extended attribute blocks for equality.
881  *
882  * Returns 0 if the blocks are equal, 1 if they differ, and
883  * a negative error number on errors.
884  */
885 static int
886 ext2_xattr_cmp(struct ext2_xattr_header *header1,
887 	       struct ext2_xattr_header *header2)
888 {
889 	struct ext2_xattr_entry *entry1, *entry2;
890 
891 	entry1 = ENTRY(header1+1);
892 	entry2 = ENTRY(header2+1);
893 	while (!IS_LAST_ENTRY(entry1)) {
894 		if (IS_LAST_ENTRY(entry2))
895 			return 1;
896 		if (entry1->e_hash != entry2->e_hash ||
897 		    entry1->e_name_index != entry2->e_name_index ||
898 		    entry1->e_name_len != entry2->e_name_len ||
899 		    entry1->e_value_size != entry2->e_value_size ||
900 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
901 			return 1;
902 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
903 			return -EIO;
904 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
905 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
906 			   le32_to_cpu(entry1->e_value_size)))
907 			return 1;
908 
909 		entry1 = EXT2_XATTR_NEXT(entry1);
910 		entry2 = EXT2_XATTR_NEXT(entry2);
911 	}
912 	if (!IS_LAST_ENTRY(entry2))
913 		return 1;
914 	return 0;
915 }
916 
917 /*
918  * ext2_xattr_cache_find()
919  *
920  * Find an identical extended attribute block.
921  *
922  * Returns a locked buffer head to the block found, or NULL if such
923  * a block was not found or an error occurred.
924  */
925 static struct buffer_head *
926 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
927 {
928 	__u32 hash = le32_to_cpu(header->h_hash);
929 	struct mb_cache_entry *ce;
930 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
931 
932 	if (!header->h_hash)
933 		return NULL;  /* never share */
934 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
935 again:
936 	ce = mb_cache_entry_find_first(ea_block_cache, hash);
937 	while (ce) {
938 		struct buffer_head *bh;
939 
940 		bh = sb_bread(inode->i_sb, ce->e_value);
941 		if (!bh) {
942 			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
943 				"inode %ld: block %ld read error",
944 				inode->i_ino, (unsigned long) ce->e_value);
945 		} else {
946 			lock_buffer(bh);
947 			/*
948 			 * We have to be careful about races with freeing or
949 			 * rehashing of xattr block. Once we hold buffer lock
950 			 * xattr block's state is stable so we can check
951 			 * whether the block got freed / rehashed or not.
952 			 * Since we unhash mbcache entry under buffer lock when
953 			 * freeing / rehashing xattr block, checking whether
954 			 * entry is still hashed is reliable.
955 			 */
956 			if (hlist_bl_unhashed(&ce->e_hash_list)) {
957 				mb_cache_entry_put(ea_block_cache, ce);
958 				unlock_buffer(bh);
959 				brelse(bh);
960 				goto again;
961 			} else if (le32_to_cpu(HDR(bh)->h_refcount) >
962 				   EXT2_XATTR_REFCOUNT_MAX) {
963 				ea_idebug(inode, "block %ld refcount %d>%d",
964 					  (unsigned long) ce->e_value,
965 					  le32_to_cpu(HDR(bh)->h_refcount),
966 					  EXT2_XATTR_REFCOUNT_MAX);
967 			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
968 				ea_bdebug(bh, "b_count=%d",
969 					  atomic_read(&(bh->b_count)));
970 				mb_cache_entry_touch(ea_block_cache, ce);
971 				mb_cache_entry_put(ea_block_cache, ce);
972 				return bh;
973 			}
974 			unlock_buffer(bh);
975 			brelse(bh);
976 		}
977 		ce = mb_cache_entry_find_next(ea_block_cache, ce);
978 	}
979 	return NULL;
980 }
981 
982 #define NAME_HASH_SHIFT 5
983 #define VALUE_HASH_SHIFT 16
984 
985 /*
986  * ext2_xattr_hash_entry()
987  *
988  * Compute the hash of an extended attribute.
989  */
990 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
991 					 struct ext2_xattr_entry *entry)
992 {
993 	__u32 hash = 0;
994 	char *name = entry->e_name;
995 	int n;
996 
997 	for (n=0; n < entry->e_name_len; n++) {
998 		hash = (hash << NAME_HASH_SHIFT) ^
999 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1000 		       *name++;
1001 	}
1002 
1003 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1004 		__le32 *value = (__le32 *)((char *)header +
1005 			le16_to_cpu(entry->e_value_offs));
1006 		for (n = (le32_to_cpu(entry->e_value_size) +
1007 		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
1008 			hash = (hash << VALUE_HASH_SHIFT) ^
1009 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1010 			       le32_to_cpu(*value++);
1011 		}
1012 	}
1013 	entry->e_hash = cpu_to_le32(hash);
1014 }
1015 
1016 #undef NAME_HASH_SHIFT
1017 #undef VALUE_HASH_SHIFT
1018 
1019 #define BLOCK_HASH_SHIFT 16
1020 
1021 /*
1022  * ext2_xattr_rehash()
1023  *
1024  * Re-compute the extended attribute hash value after an entry has changed.
1025  */
1026 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1027 			      struct ext2_xattr_entry *entry)
1028 {
1029 	struct ext2_xattr_entry *here;
1030 	__u32 hash = 0;
1031 
1032 	ext2_xattr_hash_entry(header, entry);
1033 	here = ENTRY(header+1);
1034 	while (!IS_LAST_ENTRY(here)) {
1035 		if (!here->e_hash) {
1036 			/* Block is not shared if an entry's hash value == 0 */
1037 			hash = 0;
1038 			break;
1039 		}
1040 		hash = (hash << BLOCK_HASH_SHIFT) ^
1041 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1042 		       le32_to_cpu(here->e_hash);
1043 		here = EXT2_XATTR_NEXT(here);
1044 	}
1045 	header->h_hash = cpu_to_le32(hash);
1046 }
1047 
1048 #undef BLOCK_HASH_SHIFT
1049 
1050 #define HASH_BUCKET_BITS 10
1051 
1052 struct mb_cache *ext2_xattr_create_cache(void)
1053 {
1054 	return mb_cache_create(HASH_BUCKET_BITS);
1055 }
1056 
1057 void ext2_xattr_destroy_cache(struct mb_cache *cache)
1058 {
1059 	if (cache)
1060 		mb_cache_destroy(cache);
1061 }
1062