xref: /openbmc/linux/fs/ext2/xattr.c (revision 431547b3)
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 alligned 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/module.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 		char b[BDEVNAME_SIZE]; \
82 		printk(KERN_DEBUG "block %s:%lu: ", \
83 			bdevname(bh->b_bdev, b), \
84 			(unsigned long) bh->b_blocknr); \
85 		printk(f); \
86 		printk("\n"); \
87 	} while (0)
88 #else
89 # define ea_idebug(f...)
90 # define ea_bdebug(f...)
91 #endif
92 
93 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
94 			   struct ext2_xattr_header *);
95 
96 static int ext2_xattr_cache_insert(struct buffer_head *);
97 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
98 						 struct ext2_xattr_header *);
99 static void ext2_xattr_rehash(struct ext2_xattr_header *,
100 			      struct ext2_xattr_entry *);
101 
102 static struct mb_cache *ext2_xattr_cache;
103 
104 static struct xattr_handler *ext2_xattr_handler_map[] = {
105 	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
106 #ifdef CONFIG_EXT2_FS_POSIX_ACL
107 	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &ext2_xattr_acl_access_handler,
108 	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext2_xattr_acl_default_handler,
109 #endif
110 	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
111 #ifdef CONFIG_EXT2_FS_SECURITY
112 	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
113 #endif
114 };
115 
116 struct xattr_handler *ext2_xattr_handlers[] = {
117 	&ext2_xattr_user_handler,
118 	&ext2_xattr_trusted_handler,
119 #ifdef CONFIG_EXT2_FS_POSIX_ACL
120 	&ext2_xattr_acl_access_handler,
121 	&ext2_xattr_acl_default_handler,
122 #endif
123 #ifdef CONFIG_EXT2_FS_SECURITY
124 	&ext2_xattr_security_handler,
125 #endif
126 	NULL
127 };
128 
129 static inline struct xattr_handler *
130 ext2_xattr_handler(int name_index)
131 {
132 	struct xattr_handler *handler = NULL;
133 
134 	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
135 		handler = ext2_xattr_handler_map[name_index];
136 	return handler;
137 }
138 
139 /*
140  * ext2_xattr_get()
141  *
142  * Copy an extended attribute into the buffer
143  * provided, or compute the buffer size required.
144  * Buffer is NULL to compute the size of the buffer required.
145  *
146  * Returns a negative error number on failure, or the number of bytes
147  * used / required on success.
148  */
149 int
150 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
151 	       void *buffer, size_t buffer_size)
152 {
153 	struct buffer_head *bh = NULL;
154 	struct ext2_xattr_entry *entry;
155 	size_t name_len, size;
156 	char *end;
157 	int error;
158 
159 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
160 		  name_index, name, buffer, (long)buffer_size);
161 
162 	if (name == NULL)
163 		return -EINVAL;
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 	/* find named attribute */
185 	name_len = strlen(name);
186 
187 	error = -ERANGE;
188 	if (name_len > 255)
189 		goto cleanup;
190 	entry = FIRST_ENTRY(bh);
191 	while (!IS_LAST_ENTRY(entry)) {
192 		struct ext2_xattr_entry *next =
193 			EXT2_XATTR_NEXT(entry);
194 		if ((char *)next >= end)
195 			goto bad_block;
196 		if (name_index == entry->e_name_index &&
197 		    name_len == entry->e_name_len &&
198 		    memcmp(name, entry->e_name, name_len) == 0)
199 			goto found;
200 		entry = next;
201 	}
202 	/* Check the remaining name entries */
203 	while (!IS_LAST_ENTRY(entry)) {
204 		struct ext2_xattr_entry *next =
205 			EXT2_XATTR_NEXT(entry);
206 		if ((char *)next >= end)
207 			goto bad_block;
208 		entry = next;
209 	}
210 	if (ext2_xattr_cache_insert(bh))
211 		ea_idebug(inode, "cache insert failed");
212 	error = -ENODATA;
213 	goto cleanup;
214 found:
215 	/* check the buffer size */
216 	if (entry->e_value_block != 0)
217 		goto bad_block;
218 	size = le32_to_cpu(entry->e_value_size);
219 	if (size > inode->i_sb->s_blocksize ||
220 	    le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
221 		goto bad_block;
222 
223 	if (ext2_xattr_cache_insert(bh))
224 		ea_idebug(inode, "cache insert failed");
225 	if (buffer) {
226 		error = -ERANGE;
227 		if (size > buffer_size)
228 			goto cleanup;
229 		/* return value of attribute */
230 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
231 			size);
232 	}
233 	error = size;
234 
235 cleanup:
236 	brelse(bh);
237 	up_read(&EXT2_I(inode)->xattr_sem);
238 
239 	return error;
240 }
241 
242 /*
243  * ext2_xattr_list()
244  *
245  * Copy a list of attribute names into the buffer
246  * provided, or compute the buffer size required.
247  * Buffer is NULL to compute the size of the buffer required.
248  *
249  * Returns a negative error number on failure, or the number of bytes
250  * used / required on success.
251  */
252 static int
253 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
254 {
255 	struct inode *inode = dentry->d_inode;
256 	struct buffer_head *bh = NULL;
257 	struct ext2_xattr_entry *entry;
258 	char *end;
259 	size_t rest = buffer_size;
260 	int error;
261 
262 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
263 		  buffer, (long)buffer_size);
264 
265 	down_read(&EXT2_I(inode)->xattr_sem);
266 	error = 0;
267 	if (!EXT2_I(inode)->i_file_acl)
268 		goto cleanup;
269 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
270 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
271 	error = -EIO;
272 	if (!bh)
273 		goto cleanup;
274 	ea_bdebug(bh, "b_count=%d, refcount=%d",
275 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
276 	end = bh->b_data + bh->b_size;
277 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
278 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
279 bad_block:	ext2_error(inode->i_sb, "ext2_xattr_list",
280 			"inode %ld: bad block %d", inode->i_ino,
281 			EXT2_I(inode)->i_file_acl);
282 		error = -EIO;
283 		goto cleanup;
284 	}
285 
286 	/* check the on-disk data structure */
287 	entry = FIRST_ENTRY(bh);
288 	while (!IS_LAST_ENTRY(entry)) {
289 		struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
290 
291 		if ((char *)next >= end)
292 			goto bad_block;
293 		entry = next;
294 	}
295 	if (ext2_xattr_cache_insert(bh))
296 		ea_idebug(inode, "cache insert failed");
297 
298 	/* list the attribute names */
299 	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
300 	     entry = EXT2_XATTR_NEXT(entry)) {
301 		struct xattr_handler *handler =
302 			ext2_xattr_handler(entry->e_name_index);
303 
304 		if (handler) {
305 			size_t size = handler->list(dentry, buffer, rest,
306 						    entry->e_name,
307 						    entry->e_name_len,
308 						    handler->flags);
309 			if (buffer) {
310 				if (size > rest) {
311 					error = -ERANGE;
312 					goto cleanup;
313 				}
314 				buffer += size;
315 			}
316 			rest -= size;
317 		}
318 	}
319 	error = buffer_size - rest;  /* total size */
320 
321 cleanup:
322 	brelse(bh);
323 	up_read(&EXT2_I(inode)->xattr_sem);
324 
325 	return error;
326 }
327 
328 /*
329  * Inode operation listxattr()
330  *
331  * dentry->d_inode->i_mutex: don't care
332  */
333 ssize_t
334 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
335 {
336 	return ext2_xattr_list(dentry, buffer, size);
337 }
338 
339 /*
340  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
341  * not set, set it.
342  */
343 static void ext2_xattr_update_super_block(struct super_block *sb)
344 {
345 	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
346 		return;
347 
348 	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
349 	sb->s_dirt = 1;
350 	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
351 }
352 
353 /*
354  * ext2_xattr_set()
355  *
356  * Create, replace or remove an extended attribute for this inode. Buffer
357  * is NULL to remove an existing extended attribute, and non-NULL to
358  * either replace an existing extended attribute, or create a new extended
359  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
360  * specify that an extended attribute must exist and must not exist
361  * previous to the call, respectively.
362  *
363  * Returns 0, or a negative error number on failure.
364  */
365 int
366 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
367 	       const void *value, size_t value_len, int flags)
368 {
369 	struct super_block *sb = inode->i_sb;
370 	struct buffer_head *bh = NULL;
371 	struct ext2_xattr_header *header = NULL;
372 	struct ext2_xattr_entry *here, *last;
373 	size_t name_len, free, min_offs = sb->s_blocksize;
374 	int not_found = 1, error;
375 	char *end;
376 
377 	/*
378 	 * header -- Points either into bh, or to a temporarily
379 	 *           allocated buffer.
380 	 * here -- The named entry found, or the place for inserting, within
381 	 *         the block pointed to by header.
382 	 * last -- Points right after the last named entry within the block
383 	 *         pointed to by header.
384 	 * min_offs -- The offset of the first value (values are aligned
385 	 *             towards the end of the block).
386 	 * end -- Points right after the block pointed to by header.
387 	 */
388 
389 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
390 		  name_index, name, value, (long)value_len);
391 
392 	if (value == NULL)
393 		value_len = 0;
394 	if (name == NULL)
395 		return -EINVAL;
396 	name_len = strlen(name);
397 	if (name_len > 255 || value_len > sb->s_blocksize)
398 		return -ERANGE;
399 	down_write(&EXT2_I(inode)->xattr_sem);
400 	if (EXT2_I(inode)->i_file_acl) {
401 		/* The inode already has an extended attribute block. */
402 		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
403 		error = -EIO;
404 		if (!bh)
405 			goto cleanup;
406 		ea_bdebug(bh, "b_count=%d, refcount=%d",
407 			atomic_read(&(bh->b_count)),
408 			le32_to_cpu(HDR(bh)->h_refcount));
409 		header = HDR(bh);
410 		end = bh->b_data + bh->b_size;
411 		if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
412 		    header->h_blocks != cpu_to_le32(1)) {
413 bad_block:		ext2_error(sb, "ext2_xattr_set",
414 				"inode %ld: bad block %d", inode->i_ino,
415 				   EXT2_I(inode)->i_file_acl);
416 			error = -EIO;
417 			goto cleanup;
418 		}
419 		/* Find the named attribute. */
420 		here = FIRST_ENTRY(bh);
421 		while (!IS_LAST_ENTRY(here)) {
422 			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
423 			if ((char *)next >= end)
424 				goto bad_block;
425 			if (!here->e_value_block && here->e_value_size) {
426 				size_t offs = le16_to_cpu(here->e_value_offs);
427 				if (offs < min_offs)
428 					min_offs = offs;
429 			}
430 			not_found = name_index - here->e_name_index;
431 			if (!not_found)
432 				not_found = name_len - here->e_name_len;
433 			if (!not_found)
434 				not_found = memcmp(name, here->e_name,name_len);
435 			if (not_found <= 0)
436 				break;
437 			here = next;
438 		}
439 		last = here;
440 		/* We still need to compute min_offs and last. */
441 		while (!IS_LAST_ENTRY(last)) {
442 			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
443 			if ((char *)next >= end)
444 				goto bad_block;
445 			if (!last->e_value_block && last->e_value_size) {
446 				size_t offs = le16_to_cpu(last->e_value_offs);
447 				if (offs < min_offs)
448 					min_offs = offs;
449 			}
450 			last = next;
451 		}
452 
453 		/* Check whether we have enough space left. */
454 		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
455 	} else {
456 		/* We will use a new extended attribute block. */
457 		free = sb->s_blocksize -
458 			sizeof(struct ext2_xattr_header) - sizeof(__u32);
459 		here = last = NULL;  /* avoid gcc uninitialized warning. */
460 	}
461 
462 	if (not_found) {
463 		/* Request to remove a nonexistent attribute? */
464 		error = -ENODATA;
465 		if (flags & XATTR_REPLACE)
466 			goto cleanup;
467 		error = 0;
468 		if (value == NULL)
469 			goto cleanup;
470 	} else {
471 		/* Request to create an existing attribute? */
472 		error = -EEXIST;
473 		if (flags & XATTR_CREATE)
474 			goto cleanup;
475 		if (!here->e_value_block && here->e_value_size) {
476 			size_t size = le32_to_cpu(here->e_value_size);
477 
478 			if (le16_to_cpu(here->e_value_offs) + size >
479 			    sb->s_blocksize || size > sb->s_blocksize)
480 				goto bad_block;
481 			free += EXT2_XATTR_SIZE(size);
482 		}
483 		free += EXT2_XATTR_LEN(name_len);
484 	}
485 	error = -ENOSPC;
486 	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
487 		goto cleanup;
488 
489 	/* Here we know that we can set the new attribute. */
490 
491 	if (header) {
492 		struct mb_cache_entry *ce;
493 
494 		/* assert(header == HDR(bh)); */
495 		ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
496 					bh->b_blocknr);
497 		lock_buffer(bh);
498 		if (header->h_refcount == cpu_to_le32(1)) {
499 			ea_bdebug(bh, "modifying in-place");
500 			if (ce)
501 				mb_cache_entry_free(ce);
502 			/* keep the buffer locked while modifying it. */
503 		} else {
504 			int offset;
505 
506 			if (ce)
507 				mb_cache_entry_release(ce);
508 			unlock_buffer(bh);
509 			ea_bdebug(bh, "cloning");
510 			header = kmalloc(bh->b_size, GFP_KERNEL);
511 			error = -ENOMEM;
512 			if (header == NULL)
513 				goto cleanup;
514 			memcpy(header, HDR(bh), bh->b_size);
515 			header->h_refcount = cpu_to_le32(1);
516 
517 			offset = (char *)here - bh->b_data;
518 			here = ENTRY((char *)header + offset);
519 			offset = (char *)last - bh->b_data;
520 			last = ENTRY((char *)header + offset);
521 		}
522 	} else {
523 		/* Allocate a buffer where we construct the new block. */
524 		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
525 		error = -ENOMEM;
526 		if (header == NULL)
527 			goto cleanup;
528 		end = (char *)header + sb->s_blocksize;
529 		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
530 		header->h_blocks = header->h_refcount = cpu_to_le32(1);
531 		last = here = ENTRY(header+1);
532 	}
533 
534 	/* Iff we are modifying the block in-place, bh is locked here. */
535 
536 	if (not_found) {
537 		/* Insert the new name. */
538 		size_t size = EXT2_XATTR_LEN(name_len);
539 		size_t rest = (char *)last - (char *)here;
540 		memmove((char *)here + size, here, rest);
541 		memset(here, 0, size);
542 		here->e_name_index = name_index;
543 		here->e_name_len = name_len;
544 		memcpy(here->e_name, name, name_len);
545 	} else {
546 		if (!here->e_value_block && here->e_value_size) {
547 			char *first_val = (char *)header + min_offs;
548 			size_t offs = le16_to_cpu(here->e_value_offs);
549 			char *val = (char *)header + offs;
550 			size_t size = EXT2_XATTR_SIZE(
551 				le32_to_cpu(here->e_value_size));
552 
553 			if (size == EXT2_XATTR_SIZE(value_len)) {
554 				/* The old and the new value have the same
555 				   size. Just replace. */
556 				here->e_value_size = cpu_to_le32(value_len);
557 				memset(val + size - EXT2_XATTR_PAD, 0,
558 				       EXT2_XATTR_PAD); /* Clear pad bytes. */
559 				memcpy(val, value, value_len);
560 				goto skip_replace;
561 			}
562 
563 			/* Remove the old value. */
564 			memmove(first_val + size, first_val, val - first_val);
565 			memset(first_val, 0, size);
566 			here->e_value_offs = 0;
567 			min_offs += size;
568 
569 			/* Adjust all value offsets. */
570 			last = ENTRY(header+1);
571 			while (!IS_LAST_ENTRY(last)) {
572 				size_t o = le16_to_cpu(last->e_value_offs);
573 				if (!last->e_value_block && o < offs)
574 					last->e_value_offs =
575 						cpu_to_le16(o + size);
576 				last = EXT2_XATTR_NEXT(last);
577 			}
578 		}
579 		if (value == NULL) {
580 			/* Remove the old name. */
581 			size_t size = EXT2_XATTR_LEN(name_len);
582 			last = ENTRY((char *)last - size);
583 			memmove(here, (char*)here + size,
584 				(char*)last - (char*)here);
585 			memset(last, 0, size);
586 		}
587 	}
588 
589 	if (value != NULL) {
590 		/* Insert the new value. */
591 		here->e_value_size = cpu_to_le32(value_len);
592 		if (value_len) {
593 			size_t size = EXT2_XATTR_SIZE(value_len);
594 			char *val = (char *)header + min_offs - size;
595 			here->e_value_offs =
596 				cpu_to_le16((char *)val - (char *)header);
597 			memset(val + size - EXT2_XATTR_PAD, 0,
598 			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
599 			memcpy(val, value, value_len);
600 		}
601 	}
602 
603 skip_replace:
604 	if (IS_LAST_ENTRY(ENTRY(header+1))) {
605 		/* This block is now empty. */
606 		if (bh && header == HDR(bh))
607 			unlock_buffer(bh);  /* we were modifying in-place. */
608 		error = ext2_xattr_set2(inode, bh, NULL);
609 	} else {
610 		ext2_xattr_rehash(header, here);
611 		if (bh && header == HDR(bh))
612 			unlock_buffer(bh);  /* we were modifying in-place. */
613 		error = ext2_xattr_set2(inode, bh, header);
614 	}
615 
616 cleanup:
617 	brelse(bh);
618 	if (!(bh && header == HDR(bh)))
619 		kfree(header);
620 	up_write(&EXT2_I(inode)->xattr_sem);
621 
622 	return error;
623 }
624 
625 /*
626  * Second half of ext2_xattr_set(): Update the file system.
627  */
628 static int
629 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
630 		struct ext2_xattr_header *header)
631 {
632 	struct super_block *sb = inode->i_sb;
633 	struct buffer_head *new_bh = NULL;
634 	int error;
635 
636 	if (header) {
637 		new_bh = ext2_xattr_cache_find(inode, header);
638 		if (new_bh) {
639 			/* We found an identical block in the cache. */
640 			if (new_bh == old_bh) {
641 				ea_bdebug(new_bh, "keeping this block");
642 			} else {
643 				/* The old block is released after updating
644 				   the inode.  */
645 				ea_bdebug(new_bh, "reusing block");
646 
647 				error = -EDQUOT;
648 				if (vfs_dq_alloc_block(inode, 1)) {
649 					unlock_buffer(new_bh);
650 					goto cleanup;
651 				}
652 				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
653 				ea_bdebug(new_bh, "refcount now=%d",
654 					le32_to_cpu(HDR(new_bh)->h_refcount));
655 			}
656 			unlock_buffer(new_bh);
657 		} else if (old_bh && header == HDR(old_bh)) {
658 			/* Keep this block. No need to lock the block as we
659 			   don't need to change the reference count. */
660 			new_bh = old_bh;
661 			get_bh(new_bh);
662 			ext2_xattr_cache_insert(new_bh);
663 		} else {
664 			/* We need to allocate a new block */
665 			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
666 						EXT2_I(inode)->i_block_group);
667 			int block = ext2_new_block(inode, goal, &error);
668 			if (error)
669 				goto cleanup;
670 			ea_idebug(inode, "creating block %d", block);
671 
672 			new_bh = sb_getblk(sb, block);
673 			if (!new_bh) {
674 				ext2_free_blocks(inode, block, 1);
675 				error = -EIO;
676 				goto cleanup;
677 			}
678 			lock_buffer(new_bh);
679 			memcpy(new_bh->b_data, header, new_bh->b_size);
680 			set_buffer_uptodate(new_bh);
681 			unlock_buffer(new_bh);
682 			ext2_xattr_cache_insert(new_bh);
683 
684 			ext2_xattr_update_super_block(sb);
685 		}
686 		mark_buffer_dirty(new_bh);
687 		if (IS_SYNC(inode)) {
688 			sync_dirty_buffer(new_bh);
689 			error = -EIO;
690 			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
691 				goto cleanup;
692 		}
693 	}
694 
695 	/* Update the inode. */
696 	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
697 	inode->i_ctime = CURRENT_TIME_SEC;
698 	if (IS_SYNC(inode)) {
699 		error = ext2_sync_inode (inode);
700 		/* In case sync failed due to ENOSPC the inode was actually
701 		 * written (only some dirty data were not) so we just proceed
702 		 * as if nothing happened and cleanup the unused block */
703 		if (error && error != -ENOSPC) {
704 			if (new_bh && new_bh != old_bh)
705 				vfs_dq_free_block(inode, 1);
706 			goto cleanup;
707 		}
708 	} else
709 		mark_inode_dirty(inode);
710 
711 	error = 0;
712 	if (old_bh && old_bh != new_bh) {
713 		struct mb_cache_entry *ce;
714 
715 		/*
716 		 * If there was an old block and we are no longer using it,
717 		 * release the old block.
718 		 */
719 		ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
720 					old_bh->b_blocknr);
721 		lock_buffer(old_bh);
722 		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
723 			/* Free the old block. */
724 			if (ce)
725 				mb_cache_entry_free(ce);
726 			ea_bdebug(old_bh, "freeing");
727 			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
728 			/* We let our caller release old_bh, so we
729 			 * need to duplicate the buffer before. */
730 			get_bh(old_bh);
731 			bforget(old_bh);
732 		} else {
733 			/* Decrement the refcount only. */
734 			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
735 			if (ce)
736 				mb_cache_entry_release(ce);
737 			vfs_dq_free_block(inode, 1);
738 			mark_buffer_dirty(old_bh);
739 			ea_bdebug(old_bh, "refcount now=%d",
740 				le32_to_cpu(HDR(old_bh)->h_refcount));
741 		}
742 		unlock_buffer(old_bh);
743 	}
744 
745 cleanup:
746 	brelse(new_bh);
747 
748 	return error;
749 }
750 
751 /*
752  * ext2_xattr_delete_inode()
753  *
754  * Free extended attribute resources associated with this inode. This
755  * is called immediately before an inode is freed.
756  */
757 void
758 ext2_xattr_delete_inode(struct inode *inode)
759 {
760 	struct buffer_head *bh = NULL;
761 	struct mb_cache_entry *ce;
762 
763 	down_write(&EXT2_I(inode)->xattr_sem);
764 	if (!EXT2_I(inode)->i_file_acl)
765 		goto cleanup;
766 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
767 	if (!bh) {
768 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
769 			"inode %ld: block %d read error", inode->i_ino,
770 			EXT2_I(inode)->i_file_acl);
771 		goto cleanup;
772 	}
773 	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
774 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
775 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
776 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
777 			"inode %ld: bad block %d", inode->i_ino,
778 			EXT2_I(inode)->i_file_acl);
779 		goto cleanup;
780 	}
781 	ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
782 	lock_buffer(bh);
783 	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
784 		if (ce)
785 			mb_cache_entry_free(ce);
786 		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
787 		get_bh(bh);
788 		bforget(bh);
789 		unlock_buffer(bh);
790 	} else {
791 		le32_add_cpu(&HDR(bh)->h_refcount, -1);
792 		if (ce)
793 			mb_cache_entry_release(ce);
794 		ea_bdebug(bh, "refcount now=%d",
795 			le32_to_cpu(HDR(bh)->h_refcount));
796 		unlock_buffer(bh);
797 		mark_buffer_dirty(bh);
798 		if (IS_SYNC(inode))
799 			sync_dirty_buffer(bh);
800 		vfs_dq_free_block(inode, 1);
801 	}
802 	EXT2_I(inode)->i_file_acl = 0;
803 
804 cleanup:
805 	brelse(bh);
806 	up_write(&EXT2_I(inode)->xattr_sem);
807 }
808 
809 /*
810  * ext2_xattr_put_super()
811  *
812  * This is called when a file system is unmounted.
813  */
814 void
815 ext2_xattr_put_super(struct super_block *sb)
816 {
817 	mb_cache_shrink(sb->s_bdev);
818 }
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 buffer_head *bh)
831 {
832 	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
833 	struct mb_cache_entry *ce;
834 	int error;
835 
836 	ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS);
837 	if (!ce)
838 		return -ENOMEM;
839 	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash);
840 	if (error) {
841 		mb_cache_entry_free(ce);
842 		if (error == -EBUSY) {
843 			ea_bdebug(bh, "already in cache (%d cache entries)",
844 				atomic_read(&ext2_xattr_cache->c_entry_count));
845 			error = 0;
846 		}
847 	} else {
848 		ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
849 			  atomic_read(&ext2_xattr_cache->c_entry_count));
850 		mb_cache_entry_release(ce);
851 	}
852 	return error;
853 }
854 
855 /*
856  * ext2_xattr_cmp()
857  *
858  * Compare two extended attribute blocks for equality.
859  *
860  * Returns 0 if the blocks are equal, 1 if they differ, and
861  * a negative error number on errors.
862  */
863 static int
864 ext2_xattr_cmp(struct ext2_xattr_header *header1,
865 	       struct ext2_xattr_header *header2)
866 {
867 	struct ext2_xattr_entry *entry1, *entry2;
868 
869 	entry1 = ENTRY(header1+1);
870 	entry2 = ENTRY(header2+1);
871 	while (!IS_LAST_ENTRY(entry1)) {
872 		if (IS_LAST_ENTRY(entry2))
873 			return 1;
874 		if (entry1->e_hash != entry2->e_hash ||
875 		    entry1->e_name_index != entry2->e_name_index ||
876 		    entry1->e_name_len != entry2->e_name_len ||
877 		    entry1->e_value_size != entry2->e_value_size ||
878 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
879 			return 1;
880 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
881 			return -EIO;
882 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
883 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
884 			   le32_to_cpu(entry1->e_value_size)))
885 			return 1;
886 
887 		entry1 = EXT2_XATTR_NEXT(entry1);
888 		entry2 = EXT2_XATTR_NEXT(entry2);
889 	}
890 	if (!IS_LAST_ENTRY(entry2))
891 		return 1;
892 	return 0;
893 }
894 
895 /*
896  * ext2_xattr_cache_find()
897  *
898  * Find an identical extended attribute block.
899  *
900  * Returns a locked buffer head to the block found, or NULL if such
901  * a block was not found or an error occurred.
902  */
903 static struct buffer_head *
904 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
905 {
906 	__u32 hash = le32_to_cpu(header->h_hash);
907 	struct mb_cache_entry *ce;
908 
909 	if (!header->h_hash)
910 		return NULL;  /* never share */
911 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
912 again:
913 	ce = mb_cache_entry_find_first(ext2_xattr_cache, 0,
914 				       inode->i_sb->s_bdev, hash);
915 	while (ce) {
916 		struct buffer_head *bh;
917 
918 		if (IS_ERR(ce)) {
919 			if (PTR_ERR(ce) == -EAGAIN)
920 				goto again;
921 			break;
922 		}
923 
924 		bh = sb_bread(inode->i_sb, ce->e_block);
925 		if (!bh) {
926 			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
927 				"inode %ld: block %ld read error",
928 				inode->i_ino, (unsigned long) ce->e_block);
929 		} else {
930 			lock_buffer(bh);
931 			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_release(ce);
941 				return bh;
942 			}
943 			unlock_buffer(bh);
944 			brelse(bh);
945 		}
946 		ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash);
947 	}
948 	return NULL;
949 }
950 
951 #define NAME_HASH_SHIFT 5
952 #define VALUE_HASH_SHIFT 16
953 
954 /*
955  * ext2_xattr_hash_entry()
956  *
957  * Compute the hash of an extended attribute.
958  */
959 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
960 					 struct ext2_xattr_entry *entry)
961 {
962 	__u32 hash = 0;
963 	char *name = entry->e_name;
964 	int n;
965 
966 	for (n=0; n < entry->e_name_len; n++) {
967 		hash = (hash << NAME_HASH_SHIFT) ^
968 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
969 		       *name++;
970 	}
971 
972 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
973 		__le32 *value = (__le32 *)((char *)header +
974 			le16_to_cpu(entry->e_value_offs));
975 		for (n = (le32_to_cpu(entry->e_value_size) +
976 		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
977 			hash = (hash << VALUE_HASH_SHIFT) ^
978 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
979 			       le32_to_cpu(*value++);
980 		}
981 	}
982 	entry->e_hash = cpu_to_le32(hash);
983 }
984 
985 #undef NAME_HASH_SHIFT
986 #undef VALUE_HASH_SHIFT
987 
988 #define BLOCK_HASH_SHIFT 16
989 
990 /*
991  * ext2_xattr_rehash()
992  *
993  * Re-compute the extended attribute hash value after an entry has changed.
994  */
995 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
996 			      struct ext2_xattr_entry *entry)
997 {
998 	struct ext2_xattr_entry *here;
999 	__u32 hash = 0;
1000 
1001 	ext2_xattr_hash_entry(header, entry);
1002 	here = ENTRY(header+1);
1003 	while (!IS_LAST_ENTRY(here)) {
1004 		if (!here->e_hash) {
1005 			/* Block is not shared if an entry's hash value == 0 */
1006 			hash = 0;
1007 			break;
1008 		}
1009 		hash = (hash << BLOCK_HASH_SHIFT) ^
1010 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1011 		       le32_to_cpu(here->e_hash);
1012 		here = EXT2_XATTR_NEXT(here);
1013 	}
1014 	header->h_hash = cpu_to_le32(hash);
1015 }
1016 
1017 #undef BLOCK_HASH_SHIFT
1018 
1019 int __init
1020 init_ext2_xattr(void)
1021 {
1022 	ext2_xattr_cache = mb_cache_create("ext2_xattr", NULL,
1023 		sizeof(struct mb_cache_entry) +
1024 		sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6);
1025 	if (!ext2_xattr_cache)
1026 		return -ENOMEM;
1027 	return 0;
1028 }
1029 
1030 void
1031 exit_ext2_xattr(void)
1032 {
1033 	mb_cache_destroy(ext2_xattr_cache);
1034 }
1035