xref: /openbmc/linux/fs/ext2/xattr.c (revision 6391503b)
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 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 struct mb_cache *ext2_xattr_cache;
100 
101 static const struct xattr_handler *ext2_xattr_handler_map[] = {
102 	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
103 #ifdef CONFIG_EXT2_FS_POSIX_ACL
104 	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
105 	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
106 #endif
107 	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
108 #ifdef CONFIG_EXT2_FS_SECURITY
109 	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
110 #endif
111 };
112 
113 const struct xattr_handler *ext2_xattr_handlers[] = {
114 	&ext2_xattr_user_handler,
115 	&ext2_xattr_trusted_handler,
116 #ifdef CONFIG_EXT2_FS_POSIX_ACL
117 	&posix_acl_access_xattr_handler,
118 	&posix_acl_default_xattr_handler,
119 #endif
120 #ifdef CONFIG_EXT2_FS_SECURITY
121 	&ext2_xattr_security_handler,
122 #endif
123 	NULL
124 };
125 
126 static inline const struct xattr_handler *
127 ext2_xattr_handler(int name_index)
128 {
129 	const struct xattr_handler *handler = NULL;
130 
131 	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
132 		handler = ext2_xattr_handler_map[name_index];
133 	return handler;
134 }
135 
136 /*
137  * ext2_xattr_get()
138  *
139  * Copy an extended attribute into the buffer
140  * provided, or compute the buffer size required.
141  * Buffer is NULL to compute the size of the buffer required.
142  *
143  * Returns a negative error number on failure, or the number of bytes
144  * used / required on success.
145  */
146 int
147 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
148 	       void *buffer, size_t buffer_size)
149 {
150 	struct buffer_head *bh = NULL;
151 	struct ext2_xattr_entry *entry;
152 	size_t name_len, size;
153 	char *end;
154 	int error;
155 
156 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
157 		  name_index, name, buffer, (long)buffer_size);
158 
159 	if (name == NULL)
160 		return -EINVAL;
161 	name_len = strlen(name);
162 	if (name_len > 255)
163 		return -ERANGE;
164 
165 	down_read(&EXT2_I(inode)->xattr_sem);
166 	error = -ENODATA;
167 	if (!EXT2_I(inode)->i_file_acl)
168 		goto cleanup;
169 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
170 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
171 	error = -EIO;
172 	if (!bh)
173 		goto cleanup;
174 	ea_bdebug(bh, "b_count=%d, refcount=%d",
175 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
176 	end = bh->b_data + bh->b_size;
177 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
178 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
179 bad_block:	ext2_error(inode->i_sb, "ext2_xattr_get",
180 			"inode %ld: bad block %d", inode->i_ino,
181 			EXT2_I(inode)->i_file_acl);
182 		error = -EIO;
183 		goto cleanup;
184 	}
185 
186 	/* find named attribute */
187 	entry = FIRST_ENTRY(bh);
188 	while (!IS_LAST_ENTRY(entry)) {
189 		struct ext2_xattr_entry *next =
190 			EXT2_XATTR_NEXT(entry);
191 		if ((char *)next >= end)
192 			goto bad_block;
193 		if (name_index == entry->e_name_index &&
194 		    name_len == entry->e_name_len &&
195 		    memcmp(name, entry->e_name, name_len) == 0)
196 			goto found;
197 		entry = next;
198 	}
199 	if (ext2_xattr_cache_insert(bh))
200 		ea_idebug(inode, "cache insert failed");
201 	error = -ENODATA;
202 	goto cleanup;
203 found:
204 	/* check the buffer size */
205 	if (entry->e_value_block != 0)
206 		goto bad_block;
207 	size = le32_to_cpu(entry->e_value_size);
208 	if (size > inode->i_sb->s_blocksize ||
209 	    le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
210 		goto bad_block;
211 
212 	if (ext2_xattr_cache_insert(bh))
213 		ea_idebug(inode, "cache insert failed");
214 	if (buffer) {
215 		error = -ERANGE;
216 		if (size > buffer_size)
217 			goto cleanup;
218 		/* return value of attribute */
219 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
220 			size);
221 	}
222 	error = size;
223 
224 cleanup:
225 	brelse(bh);
226 	up_read(&EXT2_I(inode)->xattr_sem);
227 
228 	return error;
229 }
230 
231 /*
232  * ext2_xattr_list()
233  *
234  * Copy a list of attribute names into the buffer
235  * provided, or compute the buffer size required.
236  * Buffer is NULL to compute the size of the buffer required.
237  *
238  * Returns a negative error number on failure, or the number of bytes
239  * used / required on success.
240  */
241 static int
242 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
243 {
244 	struct inode *inode = d_inode(dentry);
245 	struct buffer_head *bh = NULL;
246 	struct ext2_xattr_entry *entry;
247 	char *end;
248 	size_t rest = buffer_size;
249 	int error;
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(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 		struct mb_cache_entry *ce;
487 
488 		/* assert(header == HDR(bh)); */
489 		ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
490 					bh->b_blocknr);
491 		lock_buffer(bh);
492 		if (header->h_refcount == cpu_to_le32(1)) {
493 			ea_bdebug(bh, "modifying in-place");
494 			if (ce)
495 				mb_cache_entry_free(ce);
496 			/* keep the buffer locked while modifying it. */
497 		} else {
498 			int offset;
499 
500 			if (ce)
501 				mb_cache_entry_release(ce);
502 			unlock_buffer(bh);
503 			ea_bdebug(bh, "cloning");
504 			header = kmalloc(bh->b_size, GFP_KERNEL);
505 			error = -ENOMEM;
506 			if (header == NULL)
507 				goto cleanup;
508 			memcpy(header, HDR(bh), bh->b_size);
509 			header->h_refcount = cpu_to_le32(1);
510 
511 			offset = (char *)here - bh->b_data;
512 			here = ENTRY((char *)header + offset);
513 			offset = (char *)last - bh->b_data;
514 			last = ENTRY((char *)header + offset);
515 		}
516 	} else {
517 		/* Allocate a buffer where we construct the new block. */
518 		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
519 		error = -ENOMEM;
520 		if (header == NULL)
521 			goto cleanup;
522 		end = (char *)header + sb->s_blocksize;
523 		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
524 		header->h_blocks = header->h_refcount = cpu_to_le32(1);
525 		last = here = ENTRY(header+1);
526 	}
527 
528 	/* Iff we are modifying the block in-place, bh is locked here. */
529 
530 	if (not_found) {
531 		/* Insert the new name. */
532 		size_t size = EXT2_XATTR_LEN(name_len);
533 		size_t rest = (char *)last - (char *)here;
534 		memmove((char *)here + size, here, rest);
535 		memset(here, 0, size);
536 		here->e_name_index = name_index;
537 		here->e_name_len = name_len;
538 		memcpy(here->e_name, name, name_len);
539 	} else {
540 		if (!here->e_value_block && here->e_value_size) {
541 			char *first_val = (char *)header + min_offs;
542 			size_t offs = le16_to_cpu(here->e_value_offs);
543 			char *val = (char *)header + offs;
544 			size_t size = EXT2_XATTR_SIZE(
545 				le32_to_cpu(here->e_value_size));
546 
547 			if (size == EXT2_XATTR_SIZE(value_len)) {
548 				/* The old and the new value have the same
549 				   size. Just replace. */
550 				here->e_value_size = cpu_to_le32(value_len);
551 				memset(val + size - EXT2_XATTR_PAD, 0,
552 				       EXT2_XATTR_PAD); /* Clear pad bytes. */
553 				memcpy(val, value, value_len);
554 				goto skip_replace;
555 			}
556 
557 			/* Remove the old value. */
558 			memmove(first_val + size, first_val, val - first_val);
559 			memset(first_val, 0, size);
560 			here->e_value_offs = 0;
561 			min_offs += size;
562 
563 			/* Adjust all value offsets. */
564 			last = ENTRY(header+1);
565 			while (!IS_LAST_ENTRY(last)) {
566 				size_t o = le16_to_cpu(last->e_value_offs);
567 				if (!last->e_value_block && o < offs)
568 					last->e_value_offs =
569 						cpu_to_le16(o + size);
570 				last = EXT2_XATTR_NEXT(last);
571 			}
572 		}
573 		if (value == NULL) {
574 			/* Remove the old name. */
575 			size_t size = EXT2_XATTR_LEN(name_len);
576 			last = ENTRY((char *)last - size);
577 			memmove(here, (char*)here + size,
578 				(char*)last - (char*)here);
579 			memset(last, 0, size);
580 		}
581 	}
582 
583 	if (value != NULL) {
584 		/* Insert the new value. */
585 		here->e_value_size = cpu_to_le32(value_len);
586 		if (value_len) {
587 			size_t size = EXT2_XATTR_SIZE(value_len);
588 			char *val = (char *)header + min_offs - size;
589 			here->e_value_offs =
590 				cpu_to_le16((char *)val - (char *)header);
591 			memset(val + size - EXT2_XATTR_PAD, 0,
592 			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
593 			memcpy(val, value, value_len);
594 		}
595 	}
596 
597 skip_replace:
598 	if (IS_LAST_ENTRY(ENTRY(header+1))) {
599 		/* This block is now empty. */
600 		if (bh && header == HDR(bh))
601 			unlock_buffer(bh);  /* we were modifying in-place. */
602 		error = ext2_xattr_set2(inode, bh, NULL);
603 	} else {
604 		ext2_xattr_rehash(header, here);
605 		if (bh && header == HDR(bh))
606 			unlock_buffer(bh);  /* we were modifying in-place. */
607 		error = ext2_xattr_set2(inode, bh, header);
608 	}
609 
610 cleanup:
611 	brelse(bh);
612 	if (!(bh && header == HDR(bh)))
613 		kfree(header);
614 	up_write(&EXT2_I(inode)->xattr_sem);
615 
616 	return error;
617 }
618 
619 /*
620  * Second half of ext2_xattr_set(): Update the file system.
621  */
622 static int
623 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
624 		struct ext2_xattr_header *header)
625 {
626 	struct super_block *sb = inode->i_sb;
627 	struct buffer_head *new_bh = NULL;
628 	int error;
629 
630 	if (header) {
631 		new_bh = ext2_xattr_cache_find(inode, header);
632 		if (new_bh) {
633 			/* We found an identical block in the cache. */
634 			if (new_bh == old_bh) {
635 				ea_bdebug(new_bh, "keeping this block");
636 			} else {
637 				/* The old block is released after updating
638 				   the inode.  */
639 				ea_bdebug(new_bh, "reusing block");
640 
641 				error = dquot_alloc_block(inode, 1);
642 				if (error) {
643 					unlock_buffer(new_bh);
644 					goto cleanup;
645 				}
646 				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
647 				ea_bdebug(new_bh, "refcount now=%d",
648 					le32_to_cpu(HDR(new_bh)->h_refcount));
649 			}
650 			unlock_buffer(new_bh);
651 		} else if (old_bh && header == HDR(old_bh)) {
652 			/* Keep this block. No need to lock the block as we
653 			   don't need to change the reference count. */
654 			new_bh = old_bh;
655 			get_bh(new_bh);
656 			ext2_xattr_cache_insert(new_bh);
657 		} else {
658 			/* We need to allocate a new block */
659 			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
660 						EXT2_I(inode)->i_block_group);
661 			int block = ext2_new_block(inode, goal, &error);
662 			if (error)
663 				goto cleanup;
664 			ea_idebug(inode, "creating block %d", block);
665 
666 			new_bh = sb_getblk(sb, block);
667 			if (unlikely(!new_bh)) {
668 				ext2_free_blocks(inode, block, 1);
669 				mark_inode_dirty(inode);
670 				error = -ENOMEM;
671 				goto cleanup;
672 			}
673 			lock_buffer(new_bh);
674 			memcpy(new_bh->b_data, header, new_bh->b_size);
675 			set_buffer_uptodate(new_bh);
676 			unlock_buffer(new_bh);
677 			ext2_xattr_cache_insert(new_bh);
678 
679 			ext2_xattr_update_super_block(sb);
680 		}
681 		mark_buffer_dirty(new_bh);
682 		if (IS_SYNC(inode)) {
683 			sync_dirty_buffer(new_bh);
684 			error = -EIO;
685 			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
686 				goto cleanup;
687 		}
688 	}
689 
690 	/* Update the inode. */
691 	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
692 	inode->i_ctime = CURRENT_TIME_SEC;
693 	if (IS_SYNC(inode)) {
694 		error = sync_inode_metadata(inode, 1);
695 		/* In case sync failed due to ENOSPC the inode was actually
696 		 * written (only some dirty data were not) so we just proceed
697 		 * as if nothing happened and cleanup the unused block */
698 		if (error && error != -ENOSPC) {
699 			if (new_bh && new_bh != old_bh) {
700 				dquot_free_block_nodirty(inode, 1);
701 				mark_inode_dirty(inode);
702 			}
703 			goto cleanup;
704 		}
705 	} else
706 		mark_inode_dirty(inode);
707 
708 	error = 0;
709 	if (old_bh && old_bh != new_bh) {
710 		struct mb_cache_entry *ce;
711 
712 		/*
713 		 * If there was an old block and we are no longer using it,
714 		 * release the old block.
715 		 */
716 		ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
717 					old_bh->b_blocknr);
718 		lock_buffer(old_bh);
719 		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
720 			/* Free the old block. */
721 			if (ce)
722 				mb_cache_entry_free(ce);
723 			ea_bdebug(old_bh, "freeing");
724 			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
725 			mark_inode_dirty(inode);
726 			/* We let our caller release old_bh, so we
727 			 * need to duplicate the buffer before. */
728 			get_bh(old_bh);
729 			bforget(old_bh);
730 		} else {
731 			/* Decrement the refcount only. */
732 			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
733 			if (ce)
734 				mb_cache_entry_release(ce);
735 			dquot_free_block_nodirty(inode, 1);
736 			mark_inode_dirty(inode);
737 			mark_buffer_dirty(old_bh);
738 			ea_bdebug(old_bh, "refcount now=%d",
739 				le32_to_cpu(HDR(old_bh)->h_refcount));
740 		}
741 		unlock_buffer(old_bh);
742 	}
743 
744 cleanup:
745 	brelse(new_bh);
746 
747 	return error;
748 }
749 
750 /*
751  * ext2_xattr_delete_inode()
752  *
753  * Free extended attribute resources associated with this inode. This
754  * is called immediately before an inode is freed.
755  */
756 void
757 ext2_xattr_delete_inode(struct inode *inode)
758 {
759 	struct buffer_head *bh = NULL;
760 	struct mb_cache_entry *ce;
761 
762 	down_write(&EXT2_I(inode)->xattr_sem);
763 	if (!EXT2_I(inode)->i_file_acl)
764 		goto cleanup;
765 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
766 	if (!bh) {
767 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
768 			"inode %ld: block %d read error", inode->i_ino,
769 			EXT2_I(inode)->i_file_acl);
770 		goto cleanup;
771 	}
772 	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
773 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
774 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
775 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
776 			"inode %ld: bad block %d", inode->i_ino,
777 			EXT2_I(inode)->i_file_acl);
778 		goto cleanup;
779 	}
780 	ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
781 	lock_buffer(bh);
782 	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
783 		if (ce)
784 			mb_cache_entry_free(ce);
785 		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
786 		get_bh(bh);
787 		bforget(bh);
788 		unlock_buffer(bh);
789 	} else {
790 		le32_add_cpu(&HDR(bh)->h_refcount, -1);
791 		if (ce)
792 			mb_cache_entry_release(ce);
793 		ea_bdebug(bh, "refcount now=%d",
794 			le32_to_cpu(HDR(bh)->h_refcount));
795 		unlock_buffer(bh);
796 		mark_buffer_dirty(bh);
797 		if (IS_SYNC(inode))
798 			sync_dirty_buffer(bh);
799 		dquot_free_block_nodirty(inode, 1);
800 	}
801 	EXT2_I(inode)->i_file_acl = 0;
802 
803 cleanup:
804 	brelse(bh);
805 	up_write(&EXT2_I(inode)->xattr_sem);
806 }
807 
808 /*
809  * ext2_xattr_put_super()
810  *
811  * This is called when a file system is unmounted.
812  */
813 void
814 ext2_xattr_put_super(struct super_block *sb)
815 {
816 	mb_cache_shrink(sb->s_bdev);
817 }
818 
819 
820 /*
821  * ext2_xattr_cache_insert()
822  *
823  * Create a new entry in the extended attribute cache, and insert
824  * it unless such an entry is already in the cache.
825  *
826  * Returns 0, or a negative error number on failure.
827  */
828 static int
829 ext2_xattr_cache_insert(struct buffer_head *bh)
830 {
831 	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
832 	struct mb_cache_entry *ce;
833 	int error;
834 
835 	ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS);
836 	if (!ce)
837 		return -ENOMEM;
838 	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
839 	if (error) {
840 		mb_cache_entry_free(ce);
841 		if (error == -EBUSY) {
842 			ea_bdebug(bh, "already in cache (%d cache entries)",
843 				atomic_read(&ext2_xattr_cache->c_entry_count));
844 			error = 0;
845 		}
846 	} else {
847 		ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
848 			  atomic_read(&ext2_xattr_cache->c_entry_count));
849 		mb_cache_entry_release(ce);
850 	}
851 	return error;
852 }
853 
854 /*
855  * ext2_xattr_cmp()
856  *
857  * Compare two extended attribute blocks for equality.
858  *
859  * Returns 0 if the blocks are equal, 1 if they differ, and
860  * a negative error number on errors.
861  */
862 static int
863 ext2_xattr_cmp(struct ext2_xattr_header *header1,
864 	       struct ext2_xattr_header *header2)
865 {
866 	struct ext2_xattr_entry *entry1, *entry2;
867 
868 	entry1 = ENTRY(header1+1);
869 	entry2 = ENTRY(header2+1);
870 	while (!IS_LAST_ENTRY(entry1)) {
871 		if (IS_LAST_ENTRY(entry2))
872 			return 1;
873 		if (entry1->e_hash != entry2->e_hash ||
874 		    entry1->e_name_index != entry2->e_name_index ||
875 		    entry1->e_name_len != entry2->e_name_len ||
876 		    entry1->e_value_size != entry2->e_value_size ||
877 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
878 			return 1;
879 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
880 			return -EIO;
881 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
882 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
883 			   le32_to_cpu(entry1->e_value_size)))
884 			return 1;
885 
886 		entry1 = EXT2_XATTR_NEXT(entry1);
887 		entry2 = EXT2_XATTR_NEXT(entry2);
888 	}
889 	if (!IS_LAST_ENTRY(entry2))
890 		return 1;
891 	return 0;
892 }
893 
894 /*
895  * ext2_xattr_cache_find()
896  *
897  * Find an identical extended attribute block.
898  *
899  * Returns a locked buffer head to the block found, or NULL if such
900  * a block was not found or an error occurred.
901  */
902 static struct buffer_head *
903 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
904 {
905 	__u32 hash = le32_to_cpu(header->h_hash);
906 	struct mb_cache_entry *ce;
907 
908 	if (!header->h_hash)
909 		return NULL;  /* never share */
910 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
911 again:
912 	ce = mb_cache_entry_find_first(ext2_xattr_cache, inode->i_sb->s_bdev,
913 				       hash);
914 	while (ce) {
915 		struct buffer_head *bh;
916 
917 		if (IS_ERR(ce)) {
918 			if (PTR_ERR(ce) == -EAGAIN)
919 				goto again;
920 			break;
921 		}
922 
923 		bh = sb_bread(inode->i_sb, ce->e_block);
924 		if (!bh) {
925 			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
926 				"inode %ld: block %ld read error",
927 				inode->i_ino, (unsigned long) ce->e_block);
928 		} else {
929 			lock_buffer(bh);
930 			if (le32_to_cpu(HDR(bh)->h_refcount) >
931 				   EXT2_XATTR_REFCOUNT_MAX) {
932 				ea_idebug(inode, "block %ld refcount %d>%d",
933 					  (unsigned long) ce->e_block,
934 					  le32_to_cpu(HDR(bh)->h_refcount),
935 					  EXT2_XATTR_REFCOUNT_MAX);
936 			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
937 				ea_bdebug(bh, "b_count=%d",
938 					  atomic_read(&(bh->b_count)));
939 				mb_cache_entry_release(ce);
940 				return bh;
941 			}
942 			unlock_buffer(bh);
943 			brelse(bh);
944 		}
945 		ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
946 	}
947 	return NULL;
948 }
949 
950 #define NAME_HASH_SHIFT 5
951 #define VALUE_HASH_SHIFT 16
952 
953 /*
954  * ext2_xattr_hash_entry()
955  *
956  * Compute the hash of an extended attribute.
957  */
958 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
959 					 struct ext2_xattr_entry *entry)
960 {
961 	__u32 hash = 0;
962 	char *name = entry->e_name;
963 	int n;
964 
965 	for (n=0; n < entry->e_name_len; n++) {
966 		hash = (hash << NAME_HASH_SHIFT) ^
967 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
968 		       *name++;
969 	}
970 
971 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
972 		__le32 *value = (__le32 *)((char *)header +
973 			le16_to_cpu(entry->e_value_offs));
974 		for (n = (le32_to_cpu(entry->e_value_size) +
975 		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
976 			hash = (hash << VALUE_HASH_SHIFT) ^
977 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
978 			       le32_to_cpu(*value++);
979 		}
980 	}
981 	entry->e_hash = cpu_to_le32(hash);
982 }
983 
984 #undef NAME_HASH_SHIFT
985 #undef VALUE_HASH_SHIFT
986 
987 #define BLOCK_HASH_SHIFT 16
988 
989 /*
990  * ext2_xattr_rehash()
991  *
992  * Re-compute the extended attribute hash value after an entry has changed.
993  */
994 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
995 			      struct ext2_xattr_entry *entry)
996 {
997 	struct ext2_xattr_entry *here;
998 	__u32 hash = 0;
999 
1000 	ext2_xattr_hash_entry(header, entry);
1001 	here = ENTRY(header+1);
1002 	while (!IS_LAST_ENTRY(here)) {
1003 		if (!here->e_hash) {
1004 			/* Block is not shared if an entry's hash value == 0 */
1005 			hash = 0;
1006 			break;
1007 		}
1008 		hash = (hash << BLOCK_HASH_SHIFT) ^
1009 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1010 		       le32_to_cpu(here->e_hash);
1011 		here = EXT2_XATTR_NEXT(here);
1012 	}
1013 	header->h_hash = cpu_to_le32(hash);
1014 }
1015 
1016 #undef BLOCK_HASH_SHIFT
1017 
1018 int __init
1019 init_ext2_xattr(void)
1020 {
1021 	ext2_xattr_cache = mb_cache_create("ext2_xattr", 6);
1022 	if (!ext2_xattr_cache)
1023 		return -ENOMEM;
1024 	return 0;
1025 }
1026 
1027 void
1028 exit_ext2_xattr(void)
1029 {
1030 	mb_cache_destroy(ext2_xattr_cache);
1031 }
1032