xref: /linux/fs/ext2/xattr.c (revision 80d443e8876602be2c130f79c4de81e12e2a700d)
1 /*
2  * linux/fs/ext2/xattr.c
3  *
4  * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
5  *
6  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
7  * Extended attributes for symlinks and special files added per
8  *  suggestion of Luka Renko <luka.renko@hermes.si>.
9  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
10  *  Red Hat Inc.
11  *
12  */
13 
14 /*
15  * Extended attributes are stored on disk blocks allocated outside of
16  * any inode. The i_file_acl field is then made to point to this allocated
17  * block. If all extended attributes of an inode are identical, these
18  * inodes may share the same extended attribute block. Such situations
19  * are automatically detected by keeping a cache of recent attribute block
20  * numbers and hashes over the block's contents in memory.
21  *
22  *
23  * Extended attribute block layout:
24  *
25  *   +------------------+
26  *   | header           |
27  *   | entry 1          | |
28  *   | entry 2          | | growing downwards
29  *   | entry 3          | v
30  *   | four null bytes  |
31  *   | . . .            |
32  *   | value 1          | ^
33  *   | value 3          | | growing upwards
34  *   | value 2          | |
35  *   +------------------+
36  *
37  * The block header is followed by multiple entry descriptors. These entry
38  * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
39  * byte boundaries. The entry descriptors are sorted by attribute name,
40  * so that two extended attribute blocks can be compared efficiently.
41  *
42  * Attribute values are aligned to the end of the block, stored in
43  * no specific order. They are also padded to EXT2_XATTR_PAD byte
44  * boundaries. No additional gaps are left between them.
45  *
46  * Locking strategy
47  * ----------------
48  * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
49  * EA blocks are only changed if they are exclusive to an inode, so
50  * holding xattr_sem also means that nothing but the EA block's reference
51  * count will change. Multiple writers to an EA block are synchronized
52  * by the bh lock. No more than a single bh lock is held at any time
53  * to avoid deadlocks.
54  */
55 
56 #include <linux/buffer_head.h>
57 #include <linux/init.h>
58 #include <linux/slab.h>
59 #include <linux/mbcache.h>
60 #include <linux/quotaops.h>
61 #include <linux/rwsem.h>
62 #include <linux/security.h>
63 #include "ext2.h"
64 #include "xattr.h"
65 #include "acl.h"
66 
67 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
68 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
69 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
70 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
71 
72 #ifdef EXT2_XATTR_DEBUG
73 # define ea_idebug(inode, f...) do { \
74 		printk(KERN_DEBUG "inode %s:%ld: ", \
75 			inode->i_sb->s_id, inode->i_ino); \
76 		printk(f); \
77 		printk("\n"); \
78 	} while (0)
79 # define ea_bdebug(bh, f...) do { \
80 		printk(KERN_DEBUG "block %pg:%lu: ", \
81 			bh->b_bdev, (unsigned long) bh->b_blocknr); \
82 		printk(f); \
83 		printk("\n"); \
84 	} while (0)
85 #else
86 # define ea_idebug(f...)
87 # define ea_bdebug(f...)
88 #endif
89 
90 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
91 			   struct ext2_xattr_header *);
92 
93 static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
94 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
95 						 struct ext2_xattr_header *);
96 static void ext2_xattr_rehash(struct ext2_xattr_header *,
97 			      struct ext2_xattr_entry *);
98 
99 static const struct xattr_handler *ext2_xattr_handler_map[] = {
100 	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
101 #ifdef CONFIG_EXT2_FS_POSIX_ACL
102 	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
103 	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
104 #endif
105 	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
106 #ifdef CONFIG_EXT2_FS_SECURITY
107 	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
108 #endif
109 };
110 
111 const struct xattr_handler *ext2_xattr_handlers[] = {
112 	&ext2_xattr_user_handler,
113 	&ext2_xattr_trusted_handler,
114 #ifdef CONFIG_EXT2_FS_POSIX_ACL
115 	&posix_acl_access_xattr_handler,
116 	&posix_acl_default_xattr_handler,
117 #endif
118 #ifdef CONFIG_EXT2_FS_SECURITY
119 	&ext2_xattr_security_handler,
120 #endif
121 	NULL
122 };
123 
124 static inline const struct xattr_handler *
125 ext2_xattr_handler(int name_index)
126 {
127 	const struct xattr_handler *handler = NULL;
128 
129 	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
130 		handler = ext2_xattr_handler_map[name_index];
131 	return handler;
132 }
133 
134 /*
135  * ext2_xattr_get()
136  *
137  * Copy an extended attribute into the buffer
138  * provided, or compute the buffer size required.
139  * Buffer is NULL to compute the size of the buffer required.
140  *
141  * Returns a negative error number on failure, or the number of bytes
142  * used / required on success.
143  */
144 int
145 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
146 	       void *buffer, size_t buffer_size)
147 {
148 	struct buffer_head *bh = NULL;
149 	struct ext2_xattr_entry *entry;
150 	size_t name_len, size;
151 	char *end;
152 	int error;
153 	struct mb_cache *ext2_mb_cache = EXT2_SB(inode->i_sb)->s_mb_cache;
154 
155 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
156 		  name_index, name, buffer, (long)buffer_size);
157 
158 	if (name == NULL)
159 		return -EINVAL;
160 	name_len = strlen(name);
161 	if (name_len > 255)
162 		return -ERANGE;
163 
164 	down_read(&EXT2_I(inode)->xattr_sem);
165 	error = -ENODATA;
166 	if (!EXT2_I(inode)->i_file_acl)
167 		goto cleanup;
168 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
169 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
170 	error = -EIO;
171 	if (!bh)
172 		goto cleanup;
173 	ea_bdebug(bh, "b_count=%d, refcount=%d",
174 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
175 	end = bh->b_data + bh->b_size;
176 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
177 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
178 bad_block:	ext2_error(inode->i_sb, "ext2_xattr_get",
179 			"inode %ld: bad block %d", inode->i_ino,
180 			EXT2_I(inode)->i_file_acl);
181 		error = -EIO;
182 		goto cleanup;
183 	}
184 
185 	/* find named attribute */
186 	entry = FIRST_ENTRY(bh);
187 	while (!IS_LAST_ENTRY(entry)) {
188 		struct ext2_xattr_entry *next =
189 			EXT2_XATTR_NEXT(entry);
190 		if ((char *)next >= end)
191 			goto bad_block;
192 		if (name_index == entry->e_name_index &&
193 		    name_len == entry->e_name_len &&
194 		    memcmp(name, entry->e_name, name_len) == 0)
195 			goto found;
196 		entry = next;
197 	}
198 	if (ext2_xattr_cache_insert(ext2_mb_cache, bh))
199 		ea_idebug(inode, "cache insert failed");
200 	error = -ENODATA;
201 	goto cleanup;
202 found:
203 	/* check the buffer size */
204 	if (entry->e_value_block != 0)
205 		goto bad_block;
206 	size = le32_to_cpu(entry->e_value_size);
207 	if (size > inode->i_sb->s_blocksize ||
208 	    le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
209 		goto bad_block;
210 
211 	if (ext2_xattr_cache_insert(ext2_mb_cache, bh))
212 		ea_idebug(inode, "cache insert failed");
213 	if (buffer) {
214 		error = -ERANGE;
215 		if (size > buffer_size)
216 			goto cleanup;
217 		/* return value of attribute */
218 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
219 			size);
220 	}
221 	error = size;
222 
223 cleanup:
224 	brelse(bh);
225 	up_read(&EXT2_I(inode)->xattr_sem);
226 
227 	return error;
228 }
229 
230 /*
231  * ext2_xattr_list()
232  *
233  * Copy a list of attribute names into the buffer
234  * provided, or compute the buffer size required.
235  * Buffer is NULL to compute the size of the buffer required.
236  *
237  * Returns a negative error number on failure, or the number of bytes
238  * used / required on success.
239  */
240 static int
241 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
242 {
243 	struct inode *inode = d_inode(dentry);
244 	struct buffer_head *bh = NULL;
245 	struct ext2_xattr_entry *entry;
246 	char *end;
247 	size_t rest = buffer_size;
248 	int error;
249 	struct mb_cache *ext2_mb_cache = EXT2_SB(inode->i_sb)->s_mb_cache;
250 
251 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
252 		  buffer, (long)buffer_size);
253 
254 	down_read(&EXT2_I(inode)->xattr_sem);
255 	error = 0;
256 	if (!EXT2_I(inode)->i_file_acl)
257 		goto cleanup;
258 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
259 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
260 	error = -EIO;
261 	if (!bh)
262 		goto cleanup;
263 	ea_bdebug(bh, "b_count=%d, refcount=%d",
264 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
265 	end = bh->b_data + bh->b_size;
266 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
267 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
268 bad_block:	ext2_error(inode->i_sb, "ext2_xattr_list",
269 			"inode %ld: bad block %d", inode->i_ino,
270 			EXT2_I(inode)->i_file_acl);
271 		error = -EIO;
272 		goto cleanup;
273 	}
274 
275 	/* check the on-disk data structure */
276 	entry = FIRST_ENTRY(bh);
277 	while (!IS_LAST_ENTRY(entry)) {
278 		struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
279 
280 		if ((char *)next >= end)
281 			goto bad_block;
282 		entry = next;
283 	}
284 	if (ext2_xattr_cache_insert(ext2_mb_cache, bh))
285 		ea_idebug(inode, "cache insert failed");
286 
287 	/* list the attribute names */
288 	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
289 	     entry = EXT2_XATTR_NEXT(entry)) {
290 		const struct xattr_handler *handler =
291 			ext2_xattr_handler(entry->e_name_index);
292 
293 		if (handler && (!handler->list || handler->list(dentry))) {
294 			const char *prefix = handler->prefix ?: handler->name;
295 			size_t prefix_len = strlen(prefix);
296 			size_t size = prefix_len + entry->e_name_len + 1;
297 
298 			if (buffer) {
299 				if (size > rest) {
300 					error = -ERANGE;
301 					goto cleanup;
302 				}
303 				memcpy(buffer, prefix, prefix_len);
304 				buffer += prefix_len;
305 				memcpy(buffer, entry->e_name, entry->e_name_len);
306 				buffer += entry->e_name_len;
307 				*buffer++ = 0;
308 			}
309 			rest -= size;
310 		}
311 	}
312 	error = buffer_size - rest;  /* total size */
313 
314 cleanup:
315 	brelse(bh);
316 	up_read(&EXT2_I(inode)->xattr_sem);
317 
318 	return error;
319 }
320 
321 /*
322  * Inode operation listxattr()
323  *
324  * d_inode(dentry)->i_mutex: don't care
325  */
326 ssize_t
327 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
328 {
329 	return ext2_xattr_list(dentry, buffer, size);
330 }
331 
332 /*
333  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
334  * not set, set it.
335  */
336 static void ext2_xattr_update_super_block(struct super_block *sb)
337 {
338 	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
339 		return;
340 
341 	spin_lock(&EXT2_SB(sb)->s_lock);
342 	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
343 	spin_unlock(&EXT2_SB(sb)->s_lock);
344 	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
345 }
346 
347 /*
348  * ext2_xattr_set()
349  *
350  * Create, replace or remove an extended attribute for this inode.  Value
351  * is NULL to remove an existing extended attribute, and non-NULL to
352  * either replace an existing extended attribute, or create a new extended
353  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
354  * specify that an extended attribute must exist and must not exist
355  * previous to the call, respectively.
356  *
357  * Returns 0, or a negative error number on failure.
358  */
359 int
360 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
361 	       const void *value, size_t value_len, int flags)
362 {
363 	struct super_block *sb = inode->i_sb;
364 	struct buffer_head *bh = NULL;
365 	struct ext2_xattr_header *header = NULL;
366 	struct ext2_xattr_entry *here, *last;
367 	size_t name_len, free, min_offs = sb->s_blocksize;
368 	int not_found = 1, error;
369 	char *end;
370 
371 	/*
372 	 * header -- Points either into bh, or to a temporarily
373 	 *           allocated buffer.
374 	 * here -- The named entry found, or the place for inserting, within
375 	 *         the block pointed to by header.
376 	 * last -- Points right after the last named entry within the block
377 	 *         pointed to by header.
378 	 * min_offs -- The offset of the first value (values are aligned
379 	 *             towards the end of the block).
380 	 * end -- Points right after the block pointed to by header.
381 	 */
382 
383 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
384 		  name_index, name, value, (long)value_len);
385 
386 	if (value == NULL)
387 		value_len = 0;
388 	if (name == NULL)
389 		return -EINVAL;
390 	name_len = strlen(name);
391 	if (name_len > 255 || value_len > sb->s_blocksize)
392 		return -ERANGE;
393 	down_write(&EXT2_I(inode)->xattr_sem);
394 	if (EXT2_I(inode)->i_file_acl) {
395 		/* The inode already has an extended attribute block. */
396 		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
397 		error = -EIO;
398 		if (!bh)
399 			goto cleanup;
400 		ea_bdebug(bh, "b_count=%d, refcount=%d",
401 			atomic_read(&(bh->b_count)),
402 			le32_to_cpu(HDR(bh)->h_refcount));
403 		header = HDR(bh);
404 		end = bh->b_data + bh->b_size;
405 		if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
406 		    header->h_blocks != cpu_to_le32(1)) {
407 bad_block:		ext2_error(sb, "ext2_xattr_set",
408 				"inode %ld: bad block %d", inode->i_ino,
409 				   EXT2_I(inode)->i_file_acl);
410 			error = -EIO;
411 			goto cleanup;
412 		}
413 		/* Find the named attribute. */
414 		here = FIRST_ENTRY(bh);
415 		while (!IS_LAST_ENTRY(here)) {
416 			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
417 			if ((char *)next >= end)
418 				goto bad_block;
419 			if (!here->e_value_block && here->e_value_size) {
420 				size_t offs = le16_to_cpu(here->e_value_offs);
421 				if (offs < min_offs)
422 					min_offs = offs;
423 			}
424 			not_found = name_index - here->e_name_index;
425 			if (!not_found)
426 				not_found = name_len - here->e_name_len;
427 			if (!not_found)
428 				not_found = memcmp(name, here->e_name,name_len);
429 			if (not_found <= 0)
430 				break;
431 			here = next;
432 		}
433 		last = here;
434 		/* We still need to compute min_offs and last. */
435 		while (!IS_LAST_ENTRY(last)) {
436 			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
437 			if ((char *)next >= end)
438 				goto bad_block;
439 			if (!last->e_value_block && last->e_value_size) {
440 				size_t offs = le16_to_cpu(last->e_value_offs);
441 				if (offs < min_offs)
442 					min_offs = offs;
443 			}
444 			last = next;
445 		}
446 
447 		/* Check whether we have enough space left. */
448 		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
449 	} else {
450 		/* We will use a new extended attribute block. */
451 		free = sb->s_blocksize -
452 			sizeof(struct ext2_xattr_header) - sizeof(__u32);
453 		here = last = NULL;  /* avoid gcc uninitialized warning. */
454 	}
455 
456 	if (not_found) {
457 		/* Request to remove a nonexistent attribute? */
458 		error = -ENODATA;
459 		if (flags & XATTR_REPLACE)
460 			goto cleanup;
461 		error = 0;
462 		if (value == NULL)
463 			goto cleanup;
464 	} else {
465 		/* Request to create an existing attribute? */
466 		error = -EEXIST;
467 		if (flags & XATTR_CREATE)
468 			goto cleanup;
469 		if (!here->e_value_block && here->e_value_size) {
470 			size_t size = le32_to_cpu(here->e_value_size);
471 
472 			if (le16_to_cpu(here->e_value_offs) + size >
473 			    sb->s_blocksize || size > sb->s_blocksize)
474 				goto bad_block;
475 			free += EXT2_XATTR_SIZE(size);
476 		}
477 		free += EXT2_XATTR_LEN(name_len);
478 	}
479 	error = -ENOSPC;
480 	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
481 		goto cleanup;
482 
483 	/* Here we know that we can set the new attribute. */
484 
485 	if (header) {
486 		/* assert(header == HDR(bh)); */
487 		lock_buffer(bh);
488 		if (header->h_refcount == cpu_to_le32(1)) {
489 			__u32 hash = le32_to_cpu(header->h_hash);
490 
491 			ea_bdebug(bh, "modifying in-place");
492 			/*
493 			 * This must happen under buffer lock for
494 			 * ext2_xattr_set2() to reliably detect modified block
495 			 */
496 			mb_cache_entry_delete_block(EXT2_SB(sb)->s_mb_cache,
497 						    hash, bh->b_blocknr);
498 
499 			/* keep the buffer locked while modifying it. */
500 		} else {
501 			int offset;
502 
503 			unlock_buffer(bh);
504 			ea_bdebug(bh, "cloning");
505 			header = kmalloc(bh->b_size, GFP_KERNEL);
506 			error = -ENOMEM;
507 			if (header == NULL)
508 				goto cleanup;
509 			memcpy(header, HDR(bh), bh->b_size);
510 			header->h_refcount = cpu_to_le32(1);
511 
512 			offset = (char *)here - bh->b_data;
513 			here = ENTRY((char *)header + offset);
514 			offset = (char *)last - bh->b_data;
515 			last = ENTRY((char *)header + offset);
516 		}
517 	} else {
518 		/* Allocate a buffer where we construct the new block. */
519 		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
520 		error = -ENOMEM;
521 		if (header == NULL)
522 			goto cleanup;
523 		end = (char *)header + sb->s_blocksize;
524 		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
525 		header->h_blocks = header->h_refcount = cpu_to_le32(1);
526 		last = here = ENTRY(header+1);
527 	}
528 
529 	/* Iff we are modifying the block in-place, bh is locked here. */
530 
531 	if (not_found) {
532 		/* Insert the new name. */
533 		size_t size = EXT2_XATTR_LEN(name_len);
534 		size_t rest = (char *)last - (char *)here;
535 		memmove((char *)here + size, here, rest);
536 		memset(here, 0, size);
537 		here->e_name_index = name_index;
538 		here->e_name_len = name_len;
539 		memcpy(here->e_name, name, name_len);
540 	} else {
541 		if (!here->e_value_block && here->e_value_size) {
542 			char *first_val = (char *)header + min_offs;
543 			size_t offs = le16_to_cpu(here->e_value_offs);
544 			char *val = (char *)header + offs;
545 			size_t size = EXT2_XATTR_SIZE(
546 				le32_to_cpu(here->e_value_size));
547 
548 			if (size == EXT2_XATTR_SIZE(value_len)) {
549 				/* The old and the new value have the same
550 				   size. Just replace. */
551 				here->e_value_size = cpu_to_le32(value_len);
552 				memset(val + size - EXT2_XATTR_PAD, 0,
553 				       EXT2_XATTR_PAD); /* Clear pad bytes. */
554 				memcpy(val, value, value_len);
555 				goto skip_replace;
556 			}
557 
558 			/* Remove the old value. */
559 			memmove(first_val + size, first_val, val - first_val);
560 			memset(first_val, 0, size);
561 			here->e_value_offs = 0;
562 			min_offs += size;
563 
564 			/* Adjust all value offsets. */
565 			last = ENTRY(header+1);
566 			while (!IS_LAST_ENTRY(last)) {
567 				size_t o = le16_to_cpu(last->e_value_offs);
568 				if (!last->e_value_block && o < offs)
569 					last->e_value_offs =
570 						cpu_to_le16(o + size);
571 				last = EXT2_XATTR_NEXT(last);
572 			}
573 		}
574 		if (value == NULL) {
575 			/* Remove the old name. */
576 			size_t size = EXT2_XATTR_LEN(name_len);
577 			last = ENTRY((char *)last - size);
578 			memmove(here, (char*)here + size,
579 				(char*)last - (char*)here);
580 			memset(last, 0, size);
581 		}
582 	}
583 
584 	if (value != NULL) {
585 		/* Insert the new value. */
586 		here->e_value_size = cpu_to_le32(value_len);
587 		if (value_len) {
588 			size_t size = EXT2_XATTR_SIZE(value_len);
589 			char *val = (char *)header + min_offs - size;
590 			here->e_value_offs =
591 				cpu_to_le16((char *)val - (char *)header);
592 			memset(val + size - EXT2_XATTR_PAD, 0,
593 			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
594 			memcpy(val, value, value_len);
595 		}
596 	}
597 
598 skip_replace:
599 	if (IS_LAST_ENTRY(ENTRY(header+1))) {
600 		/* This block is now empty. */
601 		if (bh && header == HDR(bh))
602 			unlock_buffer(bh);  /* we were modifying in-place. */
603 		error = ext2_xattr_set2(inode, bh, NULL);
604 	} else {
605 		ext2_xattr_rehash(header, here);
606 		if (bh && header == HDR(bh))
607 			unlock_buffer(bh);  /* we were modifying in-place. */
608 		error = ext2_xattr_set2(inode, bh, header);
609 	}
610 
611 cleanup:
612 	brelse(bh);
613 	if (!(bh && header == HDR(bh)))
614 		kfree(header);
615 	up_write(&EXT2_I(inode)->xattr_sem);
616 
617 	return error;
618 }
619 
620 /*
621  * Second half of ext2_xattr_set(): Update the file system.
622  */
623 static int
624 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
625 		struct ext2_xattr_header *header)
626 {
627 	struct super_block *sb = inode->i_sb;
628 	struct buffer_head *new_bh = NULL;
629 	int error;
630 	struct mb_cache *ext2_mb_cache = EXT2_SB(sb)->s_mb_cache;
631 
632 	if (header) {
633 		new_bh = ext2_xattr_cache_find(inode, header);
634 		if (new_bh) {
635 			/* We found an identical block in the cache. */
636 			if (new_bh == old_bh) {
637 				ea_bdebug(new_bh, "keeping this block");
638 			} else {
639 				/* The old block is released after updating
640 				   the inode.  */
641 				ea_bdebug(new_bh, "reusing block");
642 
643 				error = dquot_alloc_block(inode, 1);
644 				if (error) {
645 					unlock_buffer(new_bh);
646 					goto cleanup;
647 				}
648 				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
649 				ea_bdebug(new_bh, "refcount now=%d",
650 					le32_to_cpu(HDR(new_bh)->h_refcount));
651 			}
652 			unlock_buffer(new_bh);
653 		} else if (old_bh && header == HDR(old_bh)) {
654 			/* Keep this block. No need to lock the block as we
655 			   don't need to change the reference count. */
656 			new_bh = old_bh;
657 			get_bh(new_bh);
658 			ext2_xattr_cache_insert(ext2_mb_cache, new_bh);
659 		} else {
660 			/* We need to allocate a new block */
661 			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
662 						EXT2_I(inode)->i_block_group);
663 			int block = ext2_new_block(inode, goal, &error);
664 			if (error)
665 				goto cleanup;
666 			ea_idebug(inode, "creating block %d", block);
667 
668 			new_bh = sb_getblk(sb, block);
669 			if (unlikely(!new_bh)) {
670 				ext2_free_blocks(inode, block, 1);
671 				mark_inode_dirty(inode);
672 				error = -ENOMEM;
673 				goto cleanup;
674 			}
675 			lock_buffer(new_bh);
676 			memcpy(new_bh->b_data, header, new_bh->b_size);
677 			set_buffer_uptodate(new_bh);
678 			unlock_buffer(new_bh);
679 			ext2_xattr_cache_insert(ext2_mb_cache, new_bh);
680 
681 			ext2_xattr_update_super_block(sb);
682 		}
683 		mark_buffer_dirty(new_bh);
684 		if (IS_SYNC(inode)) {
685 			sync_dirty_buffer(new_bh);
686 			error = -EIO;
687 			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
688 				goto cleanup;
689 		}
690 	}
691 
692 	/* Update the inode. */
693 	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
694 	inode->i_ctime = current_time(inode);
695 	if (IS_SYNC(inode)) {
696 		error = sync_inode_metadata(inode, 1);
697 		/* In case sync failed due to ENOSPC the inode was actually
698 		 * written (only some dirty data were not) so we just proceed
699 		 * as if nothing happened and cleanup the unused block */
700 		if (error && error != -ENOSPC) {
701 			if (new_bh && new_bh != old_bh) {
702 				dquot_free_block_nodirty(inode, 1);
703 				mark_inode_dirty(inode);
704 			}
705 			goto cleanup;
706 		}
707 	} else
708 		mark_inode_dirty(inode);
709 
710 	error = 0;
711 	if (old_bh && old_bh != new_bh) {
712 		/*
713 		 * If there was an old block and we are no longer using it,
714 		 * release the old block.
715 		 */
716 		lock_buffer(old_bh);
717 		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
718 			__u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
719 
720 			/*
721 			 * This must happen under buffer lock for
722 			 * ext2_xattr_set2() to reliably detect freed block
723 			 */
724 			mb_cache_entry_delete_block(ext2_mb_cache,
725 						    hash, old_bh->b_blocknr);
726 			/* Free the old block. */
727 			ea_bdebug(old_bh, "freeing");
728 			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
729 			mark_inode_dirty(inode);
730 			/* We let our caller release old_bh, so we
731 			 * need to duplicate the buffer before. */
732 			get_bh(old_bh);
733 			bforget(old_bh);
734 		} else {
735 			/* Decrement the refcount only. */
736 			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
737 			dquot_free_block_nodirty(inode, 1);
738 			mark_inode_dirty(inode);
739 			mark_buffer_dirty(old_bh);
740 			ea_bdebug(old_bh, "refcount now=%d",
741 				le32_to_cpu(HDR(old_bh)->h_refcount));
742 		}
743 		unlock_buffer(old_bh);
744 	}
745 
746 cleanup:
747 	brelse(new_bh);
748 
749 	return error;
750 }
751 
752 /*
753  * ext2_xattr_delete_inode()
754  *
755  * Free extended attribute resources associated with this inode. This
756  * is called immediately before an inode is freed.
757  */
758 void
759 ext2_xattr_delete_inode(struct inode *inode)
760 {
761 	struct buffer_head *bh = NULL;
762 	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
763 
764 	down_write(&EXT2_I(inode)->xattr_sem);
765 	if (!EXT2_I(inode)->i_file_acl)
766 		goto cleanup;
767 
768 	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 0)) {
769 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
770 			"inode %ld: xattr block %d is out of data blocks range",
771 			inode->i_ino, EXT2_I(inode)->i_file_acl);
772 		goto cleanup;
773 	}
774 
775 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
776 	if (!bh) {
777 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
778 			"inode %ld: block %d read error", inode->i_ino,
779 			EXT2_I(inode)->i_file_acl);
780 		goto cleanup;
781 	}
782 	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
783 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
784 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
785 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
786 			"inode %ld: bad block %d", inode->i_ino,
787 			EXT2_I(inode)->i_file_acl);
788 		goto cleanup;
789 	}
790 	lock_buffer(bh);
791 	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
792 		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
793 
794 		/*
795 		 * This must happen under buffer lock for ext2_xattr_set2() to
796 		 * reliably detect freed block
797 		 */
798 		mb_cache_entry_delete_block(EXT2_SB(inode->i_sb)->s_mb_cache,
799 					    hash, bh->b_blocknr);
800 		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
801 		get_bh(bh);
802 		bforget(bh);
803 		unlock_buffer(bh);
804 	} else {
805 		le32_add_cpu(&HDR(bh)->h_refcount, -1);
806 		ea_bdebug(bh, "refcount now=%d",
807 			le32_to_cpu(HDR(bh)->h_refcount));
808 		unlock_buffer(bh);
809 		mark_buffer_dirty(bh);
810 		if (IS_SYNC(inode))
811 			sync_dirty_buffer(bh);
812 		dquot_free_block_nodirty(inode, 1);
813 	}
814 	EXT2_I(inode)->i_file_acl = 0;
815 
816 cleanup:
817 	brelse(bh);
818 	up_write(&EXT2_I(inode)->xattr_sem);
819 }
820 
821 /*
822  * ext2_xattr_cache_insert()
823  *
824  * Create a new entry in the extended attribute cache, and insert
825  * it unless such an entry is already in the cache.
826  *
827  * Returns 0, or a negative error number on failure.
828  */
829 static int
830 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
831 {
832 	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
833 	int error;
834 
835 	error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr, 1);
836 	if (error) {
837 		if (error == -EBUSY) {
838 			ea_bdebug(bh, "already in cache (%d cache entries)",
839 				atomic_read(&ext2_xattr_cache->c_entry_count));
840 			error = 0;
841 		}
842 	} else
843 		ea_bdebug(bh, "inserting [%x]", (int)hash);
844 	return error;
845 }
846 
847 /*
848  * ext2_xattr_cmp()
849  *
850  * Compare two extended attribute blocks for equality.
851  *
852  * Returns 0 if the blocks are equal, 1 if they differ, and
853  * a negative error number on errors.
854  */
855 static int
856 ext2_xattr_cmp(struct ext2_xattr_header *header1,
857 	       struct ext2_xattr_header *header2)
858 {
859 	struct ext2_xattr_entry *entry1, *entry2;
860 
861 	entry1 = ENTRY(header1+1);
862 	entry2 = ENTRY(header2+1);
863 	while (!IS_LAST_ENTRY(entry1)) {
864 		if (IS_LAST_ENTRY(entry2))
865 			return 1;
866 		if (entry1->e_hash != entry2->e_hash ||
867 		    entry1->e_name_index != entry2->e_name_index ||
868 		    entry1->e_name_len != entry2->e_name_len ||
869 		    entry1->e_value_size != entry2->e_value_size ||
870 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
871 			return 1;
872 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
873 			return -EIO;
874 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
875 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
876 			   le32_to_cpu(entry1->e_value_size)))
877 			return 1;
878 
879 		entry1 = EXT2_XATTR_NEXT(entry1);
880 		entry2 = EXT2_XATTR_NEXT(entry2);
881 	}
882 	if (!IS_LAST_ENTRY(entry2))
883 		return 1;
884 	return 0;
885 }
886 
887 /*
888  * ext2_xattr_cache_find()
889  *
890  * Find an identical extended attribute block.
891  *
892  * Returns a locked buffer head to the block found, or NULL if such
893  * a block was not found or an error occurred.
894  */
895 static struct buffer_head *
896 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
897 {
898 	__u32 hash = le32_to_cpu(header->h_hash);
899 	struct mb_cache_entry *ce;
900 	struct mb_cache *ext2_mb_cache = EXT2_SB(inode->i_sb)->s_mb_cache;
901 
902 	if (!header->h_hash)
903 		return NULL;  /* never share */
904 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
905 again:
906 	ce = mb_cache_entry_find_first(ext2_mb_cache, hash);
907 	while (ce) {
908 		struct buffer_head *bh;
909 
910 		bh = sb_bread(inode->i_sb, ce->e_block);
911 		if (!bh) {
912 			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
913 				"inode %ld: block %ld read error",
914 				inode->i_ino, (unsigned long) ce->e_block);
915 		} else {
916 			lock_buffer(bh);
917 			/*
918 			 * We have to be careful about races with freeing or
919 			 * rehashing of xattr block. Once we hold buffer lock
920 			 * xattr block's state is stable so we can check
921 			 * whether the block got freed / rehashed or not.
922 			 * Since we unhash mbcache entry under buffer lock when
923 			 * freeing / rehashing xattr block, checking whether
924 			 * entry is still hashed is reliable.
925 			 */
926 			if (hlist_bl_unhashed(&ce->e_hash_list)) {
927 				mb_cache_entry_put(ext2_mb_cache, ce);
928 				unlock_buffer(bh);
929 				brelse(bh);
930 				goto again;
931 			} else if (le32_to_cpu(HDR(bh)->h_refcount) >
932 				   EXT2_XATTR_REFCOUNT_MAX) {
933 				ea_idebug(inode, "block %ld refcount %d>%d",
934 					  (unsigned long) ce->e_block,
935 					  le32_to_cpu(HDR(bh)->h_refcount),
936 					  EXT2_XATTR_REFCOUNT_MAX);
937 			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
938 				ea_bdebug(bh, "b_count=%d",
939 					  atomic_read(&(bh->b_count)));
940 				mb_cache_entry_touch(ext2_mb_cache, ce);
941 				mb_cache_entry_put(ext2_mb_cache, ce);
942 				return bh;
943 			}
944 			unlock_buffer(bh);
945 			brelse(bh);
946 		}
947 		ce = mb_cache_entry_find_next(ext2_mb_cache, ce);
948 	}
949 	return NULL;
950 }
951 
952 #define NAME_HASH_SHIFT 5
953 #define VALUE_HASH_SHIFT 16
954 
955 /*
956  * ext2_xattr_hash_entry()
957  *
958  * Compute the hash of an extended attribute.
959  */
960 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
961 					 struct ext2_xattr_entry *entry)
962 {
963 	__u32 hash = 0;
964 	char *name = entry->e_name;
965 	int n;
966 
967 	for (n=0; n < entry->e_name_len; n++) {
968 		hash = (hash << NAME_HASH_SHIFT) ^
969 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
970 		       *name++;
971 	}
972 
973 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
974 		__le32 *value = (__le32 *)((char *)header +
975 			le16_to_cpu(entry->e_value_offs));
976 		for (n = (le32_to_cpu(entry->e_value_size) +
977 		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
978 			hash = (hash << VALUE_HASH_SHIFT) ^
979 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
980 			       le32_to_cpu(*value++);
981 		}
982 	}
983 	entry->e_hash = cpu_to_le32(hash);
984 }
985 
986 #undef NAME_HASH_SHIFT
987 #undef VALUE_HASH_SHIFT
988 
989 #define BLOCK_HASH_SHIFT 16
990 
991 /*
992  * ext2_xattr_rehash()
993  *
994  * Re-compute the extended attribute hash value after an entry has changed.
995  */
996 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
997 			      struct ext2_xattr_entry *entry)
998 {
999 	struct ext2_xattr_entry *here;
1000 	__u32 hash = 0;
1001 
1002 	ext2_xattr_hash_entry(header, entry);
1003 	here = ENTRY(header+1);
1004 	while (!IS_LAST_ENTRY(here)) {
1005 		if (!here->e_hash) {
1006 			/* Block is not shared if an entry's hash value == 0 */
1007 			hash = 0;
1008 			break;
1009 		}
1010 		hash = (hash << BLOCK_HASH_SHIFT) ^
1011 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1012 		       le32_to_cpu(here->e_hash);
1013 		here = EXT2_XATTR_NEXT(here);
1014 	}
1015 	header->h_hash = cpu_to_le32(hash);
1016 }
1017 
1018 #undef BLOCK_HASH_SHIFT
1019 
1020 #define HASH_BUCKET_BITS 10
1021 
1022 struct mb_cache *ext2_xattr_create_cache(void)
1023 {
1024 	return mb_cache_create(HASH_BUCKET_BITS);
1025 }
1026 
1027 void ext2_xattr_destroy_cache(struct mb_cache *cache)
1028 {
1029 	if (cache)
1030 		mb_cache_destroy(cache);
1031 }
1032