xref: /openbmc/linux/fs/ext2/xattr.c (revision f8314dc6)
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 "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 		char b[BDEVNAME_SIZE]; \
81 		printk(KERN_DEBUG "block %s:%lu: ", \
82 			bdevname(bh->b_bdev, b), \
83 			(unsigned long) bh->b_blocknr); \
84 		printk(f); \
85 		printk("\n"); \
86 	} while (0)
87 #else
88 # define ea_idebug(f...)
89 # define ea_bdebug(f...)
90 #endif
91 
92 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
93 			   struct ext2_xattr_header *);
94 
95 static int ext2_xattr_cache_insert(struct buffer_head *);
96 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
97 						 struct ext2_xattr_header *);
98 static void ext2_xattr_rehash(struct ext2_xattr_header *,
99 			      struct ext2_xattr_entry *);
100 
101 static struct mb_cache *ext2_xattr_cache;
102 
103 static struct xattr_handler *ext2_xattr_handler_map[] = {
104 	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
105 #ifdef CONFIG_EXT2_FS_POSIX_ACL
106 	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &ext2_xattr_acl_access_handler,
107 	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext2_xattr_acl_default_handler,
108 #endif
109 	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
110 #ifdef CONFIG_EXT2_FS_SECURITY
111 	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
112 #endif
113 };
114 
115 struct xattr_handler *ext2_xattr_handlers[] = {
116 	&ext2_xattr_user_handler,
117 	&ext2_xattr_trusted_handler,
118 #ifdef CONFIG_EXT2_FS_POSIX_ACL
119 	&ext2_xattr_acl_access_handler,
120 	&ext2_xattr_acl_default_handler,
121 #endif
122 #ifdef CONFIG_EXT2_FS_SECURITY
123 	&ext2_xattr_security_handler,
124 #endif
125 	NULL
126 };
127 
128 static inline struct xattr_handler *
129 ext2_xattr_handler(int name_index)
130 {
131 	struct xattr_handler *handler = NULL;
132 
133 	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
134 		handler = ext2_xattr_handler_map[name_index];
135 	return handler;
136 }
137 
138 /*
139  * ext2_xattr_get()
140  *
141  * Copy an extended attribute into the buffer
142  * provided, or compute the buffer size required.
143  * Buffer is NULL to compute the size of the buffer required.
144  *
145  * Returns a negative error number on failure, or the number of bytes
146  * used / required on success.
147  */
148 int
149 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
150 	       void *buffer, size_t buffer_size)
151 {
152 	struct buffer_head *bh = NULL;
153 	struct ext2_xattr_entry *entry;
154 	size_t name_len, size;
155 	char *end;
156 	int error;
157 
158 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
159 		  name_index, name, buffer, (long)buffer_size);
160 
161 	if (name == NULL)
162 		return -EINVAL;
163 	down_read(&EXT2_I(inode)->xattr_sem);
164 	error = -ENODATA;
165 	if (!EXT2_I(inode)->i_file_acl)
166 		goto cleanup;
167 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
168 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
169 	error = -EIO;
170 	if (!bh)
171 		goto cleanup;
172 	ea_bdebug(bh, "b_count=%d, refcount=%d",
173 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
174 	end = bh->b_data + bh->b_size;
175 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
176 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
177 bad_block:	ext2_error(inode->i_sb, "ext2_xattr_get",
178 			"inode %ld: bad block %d", inode->i_ino,
179 			EXT2_I(inode)->i_file_acl);
180 		error = -EIO;
181 		goto cleanup;
182 	}
183 	/* find named attribute */
184 	name_len = strlen(name);
185 
186 	error = -ERANGE;
187 	if (name_len > 255)
188 		goto cleanup;
189 	entry = FIRST_ENTRY(bh);
190 	while (!IS_LAST_ENTRY(entry)) {
191 		struct ext2_xattr_entry *next =
192 			EXT2_XATTR_NEXT(entry);
193 		if ((char *)next >= end)
194 			goto bad_block;
195 		if (name_index == entry->e_name_index &&
196 		    name_len == entry->e_name_len &&
197 		    memcmp(name, entry->e_name, name_len) == 0)
198 			goto found;
199 		entry = next;
200 	}
201 	/* Check the remaining name entries */
202 	while (!IS_LAST_ENTRY(entry)) {
203 		struct ext2_xattr_entry *next =
204 			EXT2_XATTR_NEXT(entry);
205 		if ((char *)next >= end)
206 			goto bad_block;
207 		entry = next;
208 	}
209 	if (ext2_xattr_cache_insert(bh))
210 		ea_idebug(inode, "cache insert failed");
211 	error = -ENODATA;
212 	goto cleanup;
213 found:
214 	/* check the buffer size */
215 	if (entry->e_value_block != 0)
216 		goto bad_block;
217 	size = le32_to_cpu(entry->e_value_size);
218 	if (size > inode->i_sb->s_blocksize ||
219 	    le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
220 		goto bad_block;
221 
222 	if (ext2_xattr_cache_insert(bh))
223 		ea_idebug(inode, "cache insert failed");
224 	if (buffer) {
225 		error = -ERANGE;
226 		if (size > buffer_size)
227 			goto cleanup;
228 		/* return value of attribute */
229 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
230 			size);
231 	}
232 	error = size;
233 
234 cleanup:
235 	brelse(bh);
236 	up_read(&EXT2_I(inode)->xattr_sem);
237 
238 	return error;
239 }
240 
241 /*
242  * ext2_xattr_list()
243  *
244  * Copy a list of attribute names into the buffer
245  * provided, or compute the buffer size required.
246  * Buffer is NULL to compute the size of the buffer required.
247  *
248  * Returns a negative error number on failure, or the number of bytes
249  * used / required on success.
250  */
251 static int
252 ext2_xattr_list(struct inode *inode, char *buffer, size_t buffer_size)
253 {
254 	struct buffer_head *bh = NULL;
255 	struct ext2_xattr_entry *entry;
256 	char *end;
257 	size_t rest = buffer_size;
258 	int error;
259 
260 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
261 		  buffer, (long)buffer_size);
262 
263 	down_read(&EXT2_I(inode)->xattr_sem);
264 	error = 0;
265 	if (!EXT2_I(inode)->i_file_acl)
266 		goto cleanup;
267 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
268 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
269 	error = -EIO;
270 	if (!bh)
271 		goto cleanup;
272 	ea_bdebug(bh, "b_count=%d, refcount=%d",
273 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
274 	end = bh->b_data + bh->b_size;
275 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
276 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
277 bad_block:	ext2_error(inode->i_sb, "ext2_xattr_list",
278 			"inode %ld: bad block %d", inode->i_ino,
279 			EXT2_I(inode)->i_file_acl);
280 		error = -EIO;
281 		goto cleanup;
282 	}
283 
284 	/* check the on-disk data structure */
285 	entry = FIRST_ENTRY(bh);
286 	while (!IS_LAST_ENTRY(entry)) {
287 		struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
288 
289 		if ((char *)next >= end)
290 			goto bad_block;
291 		entry = next;
292 	}
293 	if (ext2_xattr_cache_insert(bh))
294 		ea_idebug(inode, "cache insert failed");
295 
296 	/* list the attribute names */
297 	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
298 	     entry = EXT2_XATTR_NEXT(entry)) {
299 		struct xattr_handler *handler =
300 			ext2_xattr_handler(entry->e_name_index);
301 
302 		if (handler) {
303 			size_t size = handler->list(inode, buffer, rest,
304 						    entry->e_name,
305 						    entry->e_name_len);
306 			if (buffer) {
307 				if (size > rest) {
308 					error = -ERANGE;
309 					goto cleanup;
310 				}
311 				buffer += size;
312 			}
313 			rest -= size;
314 		}
315 	}
316 	error = buffer_size - rest;  /* total size */
317 
318 cleanup:
319 	brelse(bh);
320 	up_read(&EXT2_I(inode)->xattr_sem);
321 
322 	return error;
323 }
324 
325 /*
326  * Inode operation listxattr()
327  *
328  * dentry->d_inode->i_mutex: don't care
329  */
330 ssize_t
331 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
332 {
333 	return ext2_xattr_list(dentry->d_inode, buffer, size);
334 }
335 
336 /*
337  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
338  * not set, set it.
339  */
340 static void ext2_xattr_update_super_block(struct super_block *sb)
341 {
342 	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
343 		return;
344 
345 	lock_super(sb);
346 	EXT2_SB(sb)->s_es->s_feature_compat |=
347 		cpu_to_le32(EXT2_FEATURE_COMPAT_EXT_ATTR);
348 	sb->s_dirt = 1;
349 	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
350 	unlock_super(sb);
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 (DQUOT_ALLOC_BLOCK(inode, 1)) {
649 					unlock_buffer(new_bh);
650 					goto cleanup;
651 				}
652 				HDR(new_bh)->h_refcount = cpu_to_le32(1 +
653 					le32_to_cpu(HDR(new_bh)->h_refcount));
654 				ea_bdebug(new_bh, "refcount now=%d",
655 					le32_to_cpu(HDR(new_bh)->h_refcount));
656 			}
657 			unlock_buffer(new_bh);
658 		} else if (old_bh && header == HDR(old_bh)) {
659 			/* Keep this block. No need to lock the block as we
660 			   don't need to change the reference count. */
661 			new_bh = old_bh;
662 			get_bh(new_bh);
663 			ext2_xattr_cache_insert(new_bh);
664 		} else {
665 			/* We need to allocate a new block */
666 			int goal = le32_to_cpu(EXT2_SB(sb)->s_es->
667 						           s_first_data_block) +
668 				   EXT2_I(inode)->i_block_group *
669 				   EXT2_BLOCKS_PER_GROUP(sb);
670 			int block = ext2_new_block(inode, goal,
671 						   NULL, NULL, &error);
672 			if (error)
673 				goto cleanup;
674 			ea_idebug(inode, "creating block %d", block);
675 
676 			new_bh = sb_getblk(sb, block);
677 			if (!new_bh) {
678 				ext2_free_blocks(inode, block, 1);
679 				error = -EIO;
680 				goto cleanup;
681 			}
682 			lock_buffer(new_bh);
683 			memcpy(new_bh->b_data, header, new_bh->b_size);
684 			set_buffer_uptodate(new_bh);
685 			unlock_buffer(new_bh);
686 			ext2_xattr_cache_insert(new_bh);
687 
688 			ext2_xattr_update_super_block(sb);
689 		}
690 		mark_buffer_dirty(new_bh);
691 		if (IS_SYNC(inode)) {
692 			sync_dirty_buffer(new_bh);
693 			error = -EIO;
694 			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
695 				goto cleanup;
696 		}
697 	}
698 
699 	/* Update the inode. */
700 	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
701 	inode->i_ctime = CURRENT_TIME_SEC;
702 	if (IS_SYNC(inode)) {
703 		error = ext2_sync_inode (inode);
704 		/* In case sync failed due to ENOSPC the inode was actually
705 		 * written (only some dirty data were not) so we just proceed
706 		 * as if nothing happened and cleanup the unused block */
707 		if (error && error != -ENOSPC) {
708 			if (new_bh && new_bh != old_bh)
709 				DQUOT_FREE_BLOCK(inode, 1);
710 			goto cleanup;
711 		}
712 	} else
713 		mark_inode_dirty(inode);
714 
715 	error = 0;
716 	if (old_bh && old_bh != new_bh) {
717 		struct mb_cache_entry *ce;
718 
719 		/*
720 		 * If there was an old block and we are no longer using it,
721 		 * release the old block.
722 		 */
723 		ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
724 					old_bh->b_blocknr);
725 		lock_buffer(old_bh);
726 		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
727 			/* Free the old block. */
728 			if (ce)
729 				mb_cache_entry_free(ce);
730 			ea_bdebug(old_bh, "freeing");
731 			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
732 			/* We let our caller release old_bh, so we
733 			 * need to duplicate the buffer before. */
734 			get_bh(old_bh);
735 			bforget(old_bh);
736 		} else {
737 			/* Decrement the refcount only. */
738 			HDR(old_bh)->h_refcount = cpu_to_le32(
739 				le32_to_cpu(HDR(old_bh)->h_refcount) - 1);
740 			if (ce)
741 				mb_cache_entry_release(ce);
742 			DQUOT_FREE_BLOCK(inode, 1);
743 			mark_buffer_dirty(old_bh);
744 			ea_bdebug(old_bh, "refcount now=%d",
745 				le32_to_cpu(HDR(old_bh)->h_refcount));
746 		}
747 		unlock_buffer(old_bh);
748 	}
749 
750 cleanup:
751 	brelse(new_bh);
752 
753 	return error;
754 }
755 
756 /*
757  * ext2_xattr_delete_inode()
758  *
759  * Free extended attribute resources associated with this inode. This
760  * is called immediately before an inode is freed.
761  */
762 void
763 ext2_xattr_delete_inode(struct inode *inode)
764 {
765 	struct buffer_head *bh = NULL;
766 	struct mb_cache_entry *ce;
767 
768 	down_write(&EXT2_I(inode)->xattr_sem);
769 	if (!EXT2_I(inode)->i_file_acl)
770 		goto cleanup;
771 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
772 	if (!bh) {
773 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
774 			"inode %ld: block %d read error", inode->i_ino,
775 			EXT2_I(inode)->i_file_acl);
776 		goto cleanup;
777 	}
778 	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
779 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
780 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
781 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
782 			"inode %ld: bad block %d", inode->i_ino,
783 			EXT2_I(inode)->i_file_acl);
784 		goto cleanup;
785 	}
786 	ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
787 	lock_buffer(bh);
788 	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
789 		if (ce)
790 			mb_cache_entry_free(ce);
791 		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
792 		get_bh(bh);
793 		bforget(bh);
794 		unlock_buffer(bh);
795 	} else {
796 		HDR(bh)->h_refcount = cpu_to_le32(
797 			le32_to_cpu(HDR(bh)->h_refcount) - 1);
798 		if (ce)
799 			mb_cache_entry_release(ce);
800 		ea_bdebug(bh, "refcount now=%d",
801 			le32_to_cpu(HDR(bh)->h_refcount));
802 		unlock_buffer(bh);
803 		mark_buffer_dirty(bh);
804 		if (IS_SYNC(inode))
805 			sync_dirty_buffer(bh);
806 		DQUOT_FREE_BLOCK(inode, 1);
807 	}
808 	EXT2_I(inode)->i_file_acl = 0;
809 
810 cleanup:
811 	brelse(bh);
812 	up_write(&EXT2_I(inode)->xattr_sem);
813 }
814 
815 /*
816  * ext2_xattr_put_super()
817  *
818  * This is called when a file system is unmounted.
819  */
820 void
821 ext2_xattr_put_super(struct super_block *sb)
822 {
823 	mb_cache_shrink(sb->s_bdev);
824 }
825 
826 
827 /*
828  * ext2_xattr_cache_insert()
829  *
830  * Create a new entry in the extended attribute cache, and insert
831  * it unless such an entry is already in the cache.
832  *
833  * Returns 0, or a negative error number on failure.
834  */
835 static int
836 ext2_xattr_cache_insert(struct buffer_head *bh)
837 {
838 	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
839 	struct mb_cache_entry *ce;
840 	int error;
841 
842 	ce = mb_cache_entry_alloc(ext2_xattr_cache);
843 	if (!ce)
844 		return -ENOMEM;
845 	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash);
846 	if (error) {
847 		mb_cache_entry_free(ce);
848 		if (error == -EBUSY) {
849 			ea_bdebug(bh, "already in cache (%d cache entries)",
850 				atomic_read(&ext2_xattr_cache->c_entry_count));
851 			error = 0;
852 		}
853 	} else {
854 		ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
855 			  atomic_read(&ext2_xattr_cache->c_entry_count));
856 		mb_cache_entry_release(ce);
857 	}
858 	return error;
859 }
860 
861 /*
862  * ext2_xattr_cmp()
863  *
864  * Compare two extended attribute blocks for equality.
865  *
866  * Returns 0 if the blocks are equal, 1 if they differ, and
867  * a negative error number on errors.
868  */
869 static int
870 ext2_xattr_cmp(struct ext2_xattr_header *header1,
871 	       struct ext2_xattr_header *header2)
872 {
873 	struct ext2_xattr_entry *entry1, *entry2;
874 
875 	entry1 = ENTRY(header1+1);
876 	entry2 = ENTRY(header2+1);
877 	while (!IS_LAST_ENTRY(entry1)) {
878 		if (IS_LAST_ENTRY(entry2))
879 			return 1;
880 		if (entry1->e_hash != entry2->e_hash ||
881 		    entry1->e_name_index != entry2->e_name_index ||
882 		    entry1->e_name_len != entry2->e_name_len ||
883 		    entry1->e_value_size != entry2->e_value_size ||
884 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
885 			return 1;
886 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
887 			return -EIO;
888 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
889 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
890 			   le32_to_cpu(entry1->e_value_size)))
891 			return 1;
892 
893 		entry1 = EXT2_XATTR_NEXT(entry1);
894 		entry2 = EXT2_XATTR_NEXT(entry2);
895 	}
896 	if (!IS_LAST_ENTRY(entry2))
897 		return 1;
898 	return 0;
899 }
900 
901 /*
902  * ext2_xattr_cache_find()
903  *
904  * Find an identical extended attribute block.
905  *
906  * Returns a locked buffer head to the block found, or NULL if such
907  * a block was not found or an error occurred.
908  */
909 static struct buffer_head *
910 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
911 {
912 	__u32 hash = le32_to_cpu(header->h_hash);
913 	struct mb_cache_entry *ce;
914 
915 	if (!header->h_hash)
916 		return NULL;  /* never share */
917 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
918 again:
919 	ce = mb_cache_entry_find_first(ext2_xattr_cache, 0,
920 				       inode->i_sb->s_bdev, hash);
921 	while (ce) {
922 		struct buffer_head *bh;
923 
924 		if (IS_ERR(ce)) {
925 			if (PTR_ERR(ce) == -EAGAIN)
926 				goto again;
927 			break;
928 		}
929 
930 		bh = sb_bread(inode->i_sb, ce->e_block);
931 		if (!bh) {
932 			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
933 				"inode %ld: block %ld read error",
934 				inode->i_ino, (unsigned long) ce->e_block);
935 		} else {
936 			lock_buffer(bh);
937 			if (le32_to_cpu(HDR(bh)->h_refcount) >
938 				   EXT2_XATTR_REFCOUNT_MAX) {
939 				ea_idebug(inode, "block %ld refcount %d>%d",
940 					  (unsigned long) ce->e_block,
941 					  le32_to_cpu(HDR(bh)->h_refcount),
942 					  EXT2_XATTR_REFCOUNT_MAX);
943 			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
944 				ea_bdebug(bh, "b_count=%d",
945 					  atomic_read(&(bh->b_count)));
946 				mb_cache_entry_release(ce);
947 				return bh;
948 			}
949 			unlock_buffer(bh);
950 			brelse(bh);
951 		}
952 		ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash);
953 	}
954 	return NULL;
955 }
956 
957 #define NAME_HASH_SHIFT 5
958 #define VALUE_HASH_SHIFT 16
959 
960 /*
961  * ext2_xattr_hash_entry()
962  *
963  * Compute the hash of an extended attribute.
964  */
965 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
966 					 struct ext2_xattr_entry *entry)
967 {
968 	__u32 hash = 0;
969 	char *name = entry->e_name;
970 	int n;
971 
972 	for (n=0; n < entry->e_name_len; n++) {
973 		hash = (hash << NAME_HASH_SHIFT) ^
974 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
975 		       *name++;
976 	}
977 
978 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
979 		__le32 *value = (__le32 *)((char *)header +
980 			le16_to_cpu(entry->e_value_offs));
981 		for (n = (le32_to_cpu(entry->e_value_size) +
982 		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
983 			hash = (hash << VALUE_HASH_SHIFT) ^
984 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
985 			       le32_to_cpu(*value++);
986 		}
987 	}
988 	entry->e_hash = cpu_to_le32(hash);
989 }
990 
991 #undef NAME_HASH_SHIFT
992 #undef VALUE_HASH_SHIFT
993 
994 #define BLOCK_HASH_SHIFT 16
995 
996 /*
997  * ext2_xattr_rehash()
998  *
999  * Re-compute the extended attribute hash value after an entry has changed.
1000  */
1001 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1002 			      struct ext2_xattr_entry *entry)
1003 {
1004 	struct ext2_xattr_entry *here;
1005 	__u32 hash = 0;
1006 
1007 	ext2_xattr_hash_entry(header, entry);
1008 	here = ENTRY(header+1);
1009 	while (!IS_LAST_ENTRY(here)) {
1010 		if (!here->e_hash) {
1011 			/* Block is not shared if an entry's hash value == 0 */
1012 			hash = 0;
1013 			break;
1014 		}
1015 		hash = (hash << BLOCK_HASH_SHIFT) ^
1016 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1017 		       le32_to_cpu(here->e_hash);
1018 		here = EXT2_XATTR_NEXT(here);
1019 	}
1020 	header->h_hash = cpu_to_le32(hash);
1021 }
1022 
1023 #undef BLOCK_HASH_SHIFT
1024 
1025 int __init
1026 init_ext2_xattr(void)
1027 {
1028 	ext2_xattr_cache = mb_cache_create("ext2_xattr", NULL,
1029 		sizeof(struct mb_cache_entry) +
1030 		sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6);
1031 	if (!ext2_xattr_cache)
1032 		return -ENOMEM;
1033 	return 0;
1034 }
1035 
1036 void
1037 exit_ext2_xattr(void)
1038 {
1039 	mb_cache_destroy(ext2_xattr_cache);
1040 }
1041