xref: /linux/fs/ext2/xattr.c (revision f884ab15afdc5514e88105c92a4e2e1e6539869a)
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 		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 const 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 const 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 const struct xattr_handler *
129 ext2_xattr_handler(int name_index)
130 {
131 	const 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 	name_len = strlen(name);
164 	if (name_len > 255)
165 		return -ERANGE;
166 
167 	down_read(&EXT2_I(inode)->xattr_sem);
168 	error = -ENODATA;
169 	if (!EXT2_I(inode)->i_file_acl)
170 		goto cleanup;
171 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
172 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
173 	error = -EIO;
174 	if (!bh)
175 		goto cleanup;
176 	ea_bdebug(bh, "b_count=%d, refcount=%d",
177 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
178 	end = bh->b_data + bh->b_size;
179 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
180 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
181 bad_block:	ext2_error(inode->i_sb, "ext2_xattr_get",
182 			"inode %ld: bad block %d", inode->i_ino,
183 			EXT2_I(inode)->i_file_acl);
184 		error = -EIO;
185 		goto cleanup;
186 	}
187 
188 	/* find named attribute */
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 	if (ext2_xattr_cache_insert(bh))
202 		ea_idebug(inode, "cache insert failed");
203 	error = -ENODATA;
204 	goto cleanup;
205 found:
206 	/* check the buffer size */
207 	if (entry->e_value_block != 0)
208 		goto bad_block;
209 	size = le32_to_cpu(entry->e_value_size);
210 	if (size > inode->i_sb->s_blocksize ||
211 	    le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
212 		goto bad_block;
213 
214 	if (ext2_xattr_cache_insert(bh))
215 		ea_idebug(inode, "cache insert failed");
216 	if (buffer) {
217 		error = -ERANGE;
218 		if (size > buffer_size)
219 			goto cleanup;
220 		/* return value of attribute */
221 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
222 			size);
223 	}
224 	error = size;
225 
226 cleanup:
227 	brelse(bh);
228 	up_read(&EXT2_I(inode)->xattr_sem);
229 
230 	return error;
231 }
232 
233 /*
234  * ext2_xattr_list()
235  *
236  * Copy a list of attribute names into the buffer
237  * provided, or compute the buffer size required.
238  * Buffer is NULL to compute the size of the buffer required.
239  *
240  * Returns a negative error number on failure, or the number of bytes
241  * used / required on success.
242  */
243 static int
244 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
245 {
246 	struct inode *inode = dentry->d_inode;
247 	struct buffer_head *bh = NULL;
248 	struct ext2_xattr_entry *entry;
249 	char *end;
250 	size_t rest = buffer_size;
251 	int error;
252 
253 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
254 		  buffer, (long)buffer_size);
255 
256 	down_read(&EXT2_I(inode)->xattr_sem);
257 	error = 0;
258 	if (!EXT2_I(inode)->i_file_acl)
259 		goto cleanup;
260 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
261 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
262 	error = -EIO;
263 	if (!bh)
264 		goto cleanup;
265 	ea_bdebug(bh, "b_count=%d, refcount=%d",
266 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
267 	end = bh->b_data + bh->b_size;
268 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
269 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
270 bad_block:	ext2_error(inode->i_sb, "ext2_xattr_list",
271 			"inode %ld: bad block %d", inode->i_ino,
272 			EXT2_I(inode)->i_file_acl);
273 		error = -EIO;
274 		goto cleanup;
275 	}
276 
277 	/* check the on-disk data structure */
278 	entry = FIRST_ENTRY(bh);
279 	while (!IS_LAST_ENTRY(entry)) {
280 		struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
281 
282 		if ((char *)next >= end)
283 			goto bad_block;
284 		entry = next;
285 	}
286 	if (ext2_xattr_cache_insert(bh))
287 		ea_idebug(inode, "cache insert failed");
288 
289 	/* list the attribute names */
290 	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
291 	     entry = EXT2_XATTR_NEXT(entry)) {
292 		const struct xattr_handler *handler =
293 			ext2_xattr_handler(entry->e_name_index);
294 
295 		if (handler) {
296 			size_t size = handler->list(dentry, buffer, rest,
297 						    entry->e_name,
298 						    entry->e_name_len,
299 						    handler->flags);
300 			if (buffer) {
301 				if (size > rest) {
302 					error = -ERANGE;
303 					goto cleanup;
304 				}
305 				buffer += size;
306 			}
307 			rest -= size;
308 		}
309 	}
310 	error = buffer_size - rest;  /* total size */
311 
312 cleanup:
313 	brelse(bh);
314 	up_read(&EXT2_I(inode)->xattr_sem);
315 
316 	return error;
317 }
318 
319 /*
320  * Inode operation listxattr()
321  *
322  * dentry->d_inode->i_mutex: don't care
323  */
324 ssize_t
325 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
326 {
327 	return ext2_xattr_list(dentry, buffer, size);
328 }
329 
330 /*
331  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
332  * not set, set it.
333  */
334 static void ext2_xattr_update_super_block(struct super_block *sb)
335 {
336 	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
337 		return;
338 
339 	spin_lock(&EXT2_SB(sb)->s_lock);
340 	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
341 	spin_unlock(&EXT2_SB(sb)->s_lock);
342 	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
343 }
344 
345 /*
346  * ext2_xattr_set()
347  *
348  * Create, replace or remove an extended attribute for this inode.  Value
349  * is NULL to remove an existing extended attribute, and non-NULL to
350  * either replace an existing extended attribute, or create a new extended
351  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
352  * specify that an extended attribute must exist and must not exist
353  * previous to the call, respectively.
354  *
355  * Returns 0, or a negative error number on failure.
356  */
357 int
358 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
359 	       const void *value, size_t value_len, int flags)
360 {
361 	struct super_block *sb = inode->i_sb;
362 	struct buffer_head *bh = NULL;
363 	struct ext2_xattr_header *header = NULL;
364 	struct ext2_xattr_entry *here, *last;
365 	size_t name_len, free, min_offs = sb->s_blocksize;
366 	int not_found = 1, error;
367 	char *end;
368 
369 	/*
370 	 * header -- Points either into bh, or to a temporarily
371 	 *           allocated buffer.
372 	 * here -- The named entry found, or the place for inserting, within
373 	 *         the block pointed to by header.
374 	 * last -- Points right after the last named entry within the block
375 	 *         pointed to by header.
376 	 * min_offs -- The offset of the first value (values are aligned
377 	 *             towards the end of the block).
378 	 * end -- Points right after the block pointed to by header.
379 	 */
380 
381 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
382 		  name_index, name, value, (long)value_len);
383 
384 	if (value == NULL)
385 		value_len = 0;
386 	if (name == NULL)
387 		return -EINVAL;
388 	name_len = strlen(name);
389 	if (name_len > 255 || value_len > sb->s_blocksize)
390 		return -ERANGE;
391 	down_write(&EXT2_I(inode)->xattr_sem);
392 	if (EXT2_I(inode)->i_file_acl) {
393 		/* The inode already has an extended attribute block. */
394 		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
395 		error = -EIO;
396 		if (!bh)
397 			goto cleanup;
398 		ea_bdebug(bh, "b_count=%d, refcount=%d",
399 			atomic_read(&(bh->b_count)),
400 			le32_to_cpu(HDR(bh)->h_refcount));
401 		header = HDR(bh);
402 		end = bh->b_data + bh->b_size;
403 		if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
404 		    header->h_blocks != cpu_to_le32(1)) {
405 bad_block:		ext2_error(sb, "ext2_xattr_set",
406 				"inode %ld: bad block %d", inode->i_ino,
407 				   EXT2_I(inode)->i_file_acl);
408 			error = -EIO;
409 			goto cleanup;
410 		}
411 		/* Find the named attribute. */
412 		here = FIRST_ENTRY(bh);
413 		while (!IS_LAST_ENTRY(here)) {
414 			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
415 			if ((char *)next >= end)
416 				goto bad_block;
417 			if (!here->e_value_block && here->e_value_size) {
418 				size_t offs = le16_to_cpu(here->e_value_offs);
419 				if (offs < min_offs)
420 					min_offs = offs;
421 			}
422 			not_found = name_index - here->e_name_index;
423 			if (!not_found)
424 				not_found = name_len - here->e_name_len;
425 			if (!not_found)
426 				not_found = memcmp(name, here->e_name,name_len);
427 			if (not_found <= 0)
428 				break;
429 			here = next;
430 		}
431 		last = here;
432 		/* We still need to compute min_offs and last. */
433 		while (!IS_LAST_ENTRY(last)) {
434 			struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
435 			if ((char *)next >= end)
436 				goto bad_block;
437 			if (!last->e_value_block && last->e_value_size) {
438 				size_t offs = le16_to_cpu(last->e_value_offs);
439 				if (offs < min_offs)
440 					min_offs = offs;
441 			}
442 			last = next;
443 		}
444 
445 		/* Check whether we have enough space left. */
446 		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
447 	} else {
448 		/* We will use a new extended attribute block. */
449 		free = sb->s_blocksize -
450 			sizeof(struct ext2_xattr_header) - sizeof(__u32);
451 		here = last = NULL;  /* avoid gcc uninitialized warning. */
452 	}
453 
454 	if (not_found) {
455 		/* Request to remove a nonexistent attribute? */
456 		error = -ENODATA;
457 		if (flags & XATTR_REPLACE)
458 			goto cleanup;
459 		error = 0;
460 		if (value == NULL)
461 			goto cleanup;
462 	} else {
463 		/* Request to create an existing attribute? */
464 		error = -EEXIST;
465 		if (flags & XATTR_CREATE)
466 			goto cleanup;
467 		if (!here->e_value_block && here->e_value_size) {
468 			size_t size = le32_to_cpu(here->e_value_size);
469 
470 			if (le16_to_cpu(here->e_value_offs) + size >
471 			    sb->s_blocksize || size > sb->s_blocksize)
472 				goto bad_block;
473 			free += EXT2_XATTR_SIZE(size);
474 		}
475 		free += EXT2_XATTR_LEN(name_len);
476 	}
477 	error = -ENOSPC;
478 	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
479 		goto cleanup;
480 
481 	/* Here we know that we can set the new attribute. */
482 
483 	if (header) {
484 		struct mb_cache_entry *ce;
485 
486 		/* assert(header == HDR(bh)); */
487 		ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
488 					bh->b_blocknr);
489 		lock_buffer(bh);
490 		if (header->h_refcount == cpu_to_le32(1)) {
491 			ea_bdebug(bh, "modifying in-place");
492 			if (ce)
493 				mb_cache_entry_free(ce);
494 			/* keep the buffer locked while modifying it. */
495 		} else {
496 			int offset;
497 
498 			if (ce)
499 				mb_cache_entry_release(ce);
500 			unlock_buffer(bh);
501 			ea_bdebug(bh, "cloning");
502 			header = kmalloc(bh->b_size, GFP_KERNEL);
503 			error = -ENOMEM;
504 			if (header == NULL)
505 				goto cleanup;
506 			memcpy(header, HDR(bh), bh->b_size);
507 			header->h_refcount = cpu_to_le32(1);
508 
509 			offset = (char *)here - bh->b_data;
510 			here = ENTRY((char *)header + offset);
511 			offset = (char *)last - bh->b_data;
512 			last = ENTRY((char *)header + offset);
513 		}
514 	} else {
515 		/* Allocate a buffer where we construct the new block. */
516 		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
517 		error = -ENOMEM;
518 		if (header == NULL)
519 			goto cleanup;
520 		end = (char *)header + sb->s_blocksize;
521 		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
522 		header->h_blocks = header->h_refcount = cpu_to_le32(1);
523 		last = here = ENTRY(header+1);
524 	}
525 
526 	/* Iff we are modifying the block in-place, bh is locked here. */
527 
528 	if (not_found) {
529 		/* Insert the new name. */
530 		size_t size = EXT2_XATTR_LEN(name_len);
531 		size_t rest = (char *)last - (char *)here;
532 		memmove((char *)here + size, here, rest);
533 		memset(here, 0, size);
534 		here->e_name_index = name_index;
535 		here->e_name_len = name_len;
536 		memcpy(here->e_name, name, name_len);
537 	} else {
538 		if (!here->e_value_block && here->e_value_size) {
539 			char *first_val = (char *)header + min_offs;
540 			size_t offs = le16_to_cpu(here->e_value_offs);
541 			char *val = (char *)header + offs;
542 			size_t size = EXT2_XATTR_SIZE(
543 				le32_to_cpu(here->e_value_size));
544 
545 			if (size == EXT2_XATTR_SIZE(value_len)) {
546 				/* The old and the new value have the same
547 				   size. Just replace. */
548 				here->e_value_size = cpu_to_le32(value_len);
549 				memset(val + size - EXT2_XATTR_PAD, 0,
550 				       EXT2_XATTR_PAD); /* Clear pad bytes. */
551 				memcpy(val, value, value_len);
552 				goto skip_replace;
553 			}
554 
555 			/* Remove the old value. */
556 			memmove(first_val + size, first_val, val - first_val);
557 			memset(first_val, 0, size);
558 			here->e_value_offs = 0;
559 			min_offs += size;
560 
561 			/* Adjust all value offsets. */
562 			last = ENTRY(header+1);
563 			while (!IS_LAST_ENTRY(last)) {
564 				size_t o = le16_to_cpu(last->e_value_offs);
565 				if (!last->e_value_block && o < offs)
566 					last->e_value_offs =
567 						cpu_to_le16(o + size);
568 				last = EXT2_XATTR_NEXT(last);
569 			}
570 		}
571 		if (value == NULL) {
572 			/* Remove the old name. */
573 			size_t size = EXT2_XATTR_LEN(name_len);
574 			last = ENTRY((char *)last - size);
575 			memmove(here, (char*)here + size,
576 				(char*)last - (char*)here);
577 			memset(last, 0, size);
578 		}
579 	}
580 
581 	if (value != NULL) {
582 		/* Insert the new value. */
583 		here->e_value_size = cpu_to_le32(value_len);
584 		if (value_len) {
585 			size_t size = EXT2_XATTR_SIZE(value_len);
586 			char *val = (char *)header + min_offs - size;
587 			here->e_value_offs =
588 				cpu_to_le16((char *)val - (char *)header);
589 			memset(val + size - EXT2_XATTR_PAD, 0,
590 			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
591 			memcpy(val, value, value_len);
592 		}
593 	}
594 
595 skip_replace:
596 	if (IS_LAST_ENTRY(ENTRY(header+1))) {
597 		/* This block is now empty. */
598 		if (bh && header == HDR(bh))
599 			unlock_buffer(bh);  /* we were modifying in-place. */
600 		error = ext2_xattr_set2(inode, bh, NULL);
601 	} else {
602 		ext2_xattr_rehash(header, here);
603 		if (bh && header == HDR(bh))
604 			unlock_buffer(bh);  /* we were modifying in-place. */
605 		error = ext2_xattr_set2(inode, bh, header);
606 	}
607 
608 cleanup:
609 	brelse(bh);
610 	if (!(bh && header == HDR(bh)))
611 		kfree(header);
612 	up_write(&EXT2_I(inode)->xattr_sem);
613 
614 	return error;
615 }
616 
617 /*
618  * Second half of ext2_xattr_set(): Update the file system.
619  */
620 static int
621 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
622 		struct ext2_xattr_header *header)
623 {
624 	struct super_block *sb = inode->i_sb;
625 	struct buffer_head *new_bh = NULL;
626 	int error;
627 
628 	if (header) {
629 		new_bh = ext2_xattr_cache_find(inode, header);
630 		if (new_bh) {
631 			/* We found an identical block in the cache. */
632 			if (new_bh == old_bh) {
633 				ea_bdebug(new_bh, "keeping this block");
634 			} else {
635 				/* The old block is released after updating
636 				   the inode.  */
637 				ea_bdebug(new_bh, "reusing block");
638 
639 				error = dquot_alloc_block(inode, 1);
640 				if (error) {
641 					unlock_buffer(new_bh);
642 					goto cleanup;
643 				}
644 				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
645 				ea_bdebug(new_bh, "refcount now=%d",
646 					le32_to_cpu(HDR(new_bh)->h_refcount));
647 			}
648 			unlock_buffer(new_bh);
649 		} else if (old_bh && header == HDR(old_bh)) {
650 			/* Keep this block. No need to lock the block as we
651 			   don't need to change the reference count. */
652 			new_bh = old_bh;
653 			get_bh(new_bh);
654 			ext2_xattr_cache_insert(new_bh);
655 		} else {
656 			/* We need to allocate a new block */
657 			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
658 						EXT2_I(inode)->i_block_group);
659 			int block = ext2_new_block(inode, goal, &error);
660 			if (error)
661 				goto cleanup;
662 			ea_idebug(inode, "creating block %d", block);
663 
664 			new_bh = sb_getblk(sb, block);
665 			if (unlikely(!new_bh)) {
666 				ext2_free_blocks(inode, block, 1);
667 				mark_inode_dirty(inode);
668 				error = -ENOMEM;
669 				goto cleanup;
670 			}
671 			lock_buffer(new_bh);
672 			memcpy(new_bh->b_data, header, new_bh->b_size);
673 			set_buffer_uptodate(new_bh);
674 			unlock_buffer(new_bh);
675 			ext2_xattr_cache_insert(new_bh);
676 
677 			ext2_xattr_update_super_block(sb);
678 		}
679 		mark_buffer_dirty(new_bh);
680 		if (IS_SYNC(inode)) {
681 			sync_dirty_buffer(new_bh);
682 			error = -EIO;
683 			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
684 				goto cleanup;
685 		}
686 	}
687 
688 	/* Update the inode. */
689 	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
690 	inode->i_ctime = CURRENT_TIME_SEC;
691 	if (IS_SYNC(inode)) {
692 		error = sync_inode_metadata(inode, 1);
693 		/* In case sync failed due to ENOSPC the inode was actually
694 		 * written (only some dirty data were not) so we just proceed
695 		 * as if nothing happened and cleanup the unused block */
696 		if (error && error != -ENOSPC) {
697 			if (new_bh && new_bh != old_bh) {
698 				dquot_free_block_nodirty(inode, 1);
699 				mark_inode_dirty(inode);
700 			}
701 			goto cleanup;
702 		}
703 	} else
704 		mark_inode_dirty(inode);
705 
706 	error = 0;
707 	if (old_bh && old_bh != new_bh) {
708 		struct mb_cache_entry *ce;
709 
710 		/*
711 		 * If there was an old block and we are no longer using it,
712 		 * release the old block.
713 		 */
714 		ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
715 					old_bh->b_blocknr);
716 		lock_buffer(old_bh);
717 		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
718 			/* Free the old block. */
719 			if (ce)
720 				mb_cache_entry_free(ce);
721 			ea_bdebug(old_bh, "freeing");
722 			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
723 			mark_inode_dirty(inode);
724 			/* We let our caller release old_bh, so we
725 			 * need to duplicate the buffer before. */
726 			get_bh(old_bh);
727 			bforget(old_bh);
728 		} else {
729 			/* Decrement the refcount only. */
730 			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
731 			if (ce)
732 				mb_cache_entry_release(ce);
733 			dquot_free_block_nodirty(inode, 1);
734 			mark_inode_dirty(inode);
735 			mark_buffer_dirty(old_bh);
736 			ea_bdebug(old_bh, "refcount now=%d",
737 				le32_to_cpu(HDR(old_bh)->h_refcount));
738 		}
739 		unlock_buffer(old_bh);
740 	}
741 
742 cleanup:
743 	brelse(new_bh);
744 
745 	return error;
746 }
747 
748 /*
749  * ext2_xattr_delete_inode()
750  *
751  * Free extended attribute resources associated with this inode. This
752  * is called immediately before an inode is freed.
753  */
754 void
755 ext2_xattr_delete_inode(struct inode *inode)
756 {
757 	struct buffer_head *bh = NULL;
758 	struct mb_cache_entry *ce;
759 
760 	down_write(&EXT2_I(inode)->xattr_sem);
761 	if (!EXT2_I(inode)->i_file_acl)
762 		goto cleanup;
763 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
764 	if (!bh) {
765 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
766 			"inode %ld: block %d read error", inode->i_ino,
767 			EXT2_I(inode)->i_file_acl);
768 		goto cleanup;
769 	}
770 	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
771 	if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
772 	    HDR(bh)->h_blocks != cpu_to_le32(1)) {
773 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
774 			"inode %ld: bad block %d", inode->i_ino,
775 			EXT2_I(inode)->i_file_acl);
776 		goto cleanup;
777 	}
778 	ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
779 	lock_buffer(bh);
780 	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
781 		if (ce)
782 			mb_cache_entry_free(ce);
783 		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
784 		get_bh(bh);
785 		bforget(bh);
786 		unlock_buffer(bh);
787 	} else {
788 		le32_add_cpu(&HDR(bh)->h_refcount, -1);
789 		if (ce)
790 			mb_cache_entry_release(ce);
791 		ea_bdebug(bh, "refcount now=%d",
792 			le32_to_cpu(HDR(bh)->h_refcount));
793 		unlock_buffer(bh);
794 		mark_buffer_dirty(bh);
795 		if (IS_SYNC(inode))
796 			sync_dirty_buffer(bh);
797 		dquot_free_block_nodirty(inode, 1);
798 	}
799 	EXT2_I(inode)->i_file_acl = 0;
800 
801 cleanup:
802 	brelse(bh);
803 	up_write(&EXT2_I(inode)->xattr_sem);
804 }
805 
806 /*
807  * ext2_xattr_put_super()
808  *
809  * This is called when a file system is unmounted.
810  */
811 void
812 ext2_xattr_put_super(struct super_block *sb)
813 {
814 	mb_cache_shrink(sb->s_bdev);
815 }
816 
817 
818 /*
819  * ext2_xattr_cache_insert()
820  *
821  * Create a new entry in the extended attribute cache, and insert
822  * it unless such an entry is already in the cache.
823  *
824  * Returns 0, or a negative error number on failure.
825  */
826 static int
827 ext2_xattr_cache_insert(struct buffer_head *bh)
828 {
829 	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
830 	struct mb_cache_entry *ce;
831 	int error;
832 
833 	ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS);
834 	if (!ce)
835 		return -ENOMEM;
836 	error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
837 	if (error) {
838 		mb_cache_entry_free(ce);
839 		if (error == -EBUSY) {
840 			ea_bdebug(bh, "already in cache (%d cache entries)",
841 				atomic_read(&ext2_xattr_cache->c_entry_count));
842 			error = 0;
843 		}
844 	} else {
845 		ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
846 			  atomic_read(&ext2_xattr_cache->c_entry_count));
847 		mb_cache_entry_release(ce);
848 	}
849 	return error;
850 }
851 
852 /*
853  * ext2_xattr_cmp()
854  *
855  * Compare two extended attribute blocks for equality.
856  *
857  * Returns 0 if the blocks are equal, 1 if they differ, and
858  * a negative error number on errors.
859  */
860 static int
861 ext2_xattr_cmp(struct ext2_xattr_header *header1,
862 	       struct ext2_xattr_header *header2)
863 {
864 	struct ext2_xattr_entry *entry1, *entry2;
865 
866 	entry1 = ENTRY(header1+1);
867 	entry2 = ENTRY(header2+1);
868 	while (!IS_LAST_ENTRY(entry1)) {
869 		if (IS_LAST_ENTRY(entry2))
870 			return 1;
871 		if (entry1->e_hash != entry2->e_hash ||
872 		    entry1->e_name_index != entry2->e_name_index ||
873 		    entry1->e_name_len != entry2->e_name_len ||
874 		    entry1->e_value_size != entry2->e_value_size ||
875 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
876 			return 1;
877 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
878 			return -EIO;
879 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
880 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
881 			   le32_to_cpu(entry1->e_value_size)))
882 			return 1;
883 
884 		entry1 = EXT2_XATTR_NEXT(entry1);
885 		entry2 = EXT2_XATTR_NEXT(entry2);
886 	}
887 	if (!IS_LAST_ENTRY(entry2))
888 		return 1;
889 	return 0;
890 }
891 
892 /*
893  * ext2_xattr_cache_find()
894  *
895  * Find an identical extended attribute block.
896  *
897  * Returns a locked buffer head to the block found, or NULL if such
898  * a block was not found or an error occurred.
899  */
900 static struct buffer_head *
901 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
902 {
903 	__u32 hash = le32_to_cpu(header->h_hash);
904 	struct mb_cache_entry *ce;
905 
906 	if (!header->h_hash)
907 		return NULL;  /* never share */
908 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
909 again:
910 	ce = mb_cache_entry_find_first(ext2_xattr_cache, inode->i_sb->s_bdev,
911 				       hash);
912 	while (ce) {
913 		struct buffer_head *bh;
914 
915 		if (IS_ERR(ce)) {
916 			if (PTR_ERR(ce) == -EAGAIN)
917 				goto again;
918 			break;
919 		}
920 
921 		bh = sb_bread(inode->i_sb, ce->e_block);
922 		if (!bh) {
923 			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
924 				"inode %ld: block %ld read error",
925 				inode->i_ino, (unsigned long) ce->e_block);
926 		} else {
927 			lock_buffer(bh);
928 			if (le32_to_cpu(HDR(bh)->h_refcount) >
929 				   EXT2_XATTR_REFCOUNT_MAX) {
930 				ea_idebug(inode, "block %ld refcount %d>%d",
931 					  (unsigned long) ce->e_block,
932 					  le32_to_cpu(HDR(bh)->h_refcount),
933 					  EXT2_XATTR_REFCOUNT_MAX);
934 			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
935 				ea_bdebug(bh, "b_count=%d",
936 					  atomic_read(&(bh->b_count)));
937 				mb_cache_entry_release(ce);
938 				return bh;
939 			}
940 			unlock_buffer(bh);
941 			brelse(bh);
942 		}
943 		ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
944 	}
945 	return NULL;
946 }
947 
948 #define NAME_HASH_SHIFT 5
949 #define VALUE_HASH_SHIFT 16
950 
951 /*
952  * ext2_xattr_hash_entry()
953  *
954  * Compute the hash of an extended attribute.
955  */
956 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
957 					 struct ext2_xattr_entry *entry)
958 {
959 	__u32 hash = 0;
960 	char *name = entry->e_name;
961 	int n;
962 
963 	for (n=0; n < entry->e_name_len; n++) {
964 		hash = (hash << NAME_HASH_SHIFT) ^
965 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
966 		       *name++;
967 	}
968 
969 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
970 		__le32 *value = (__le32 *)((char *)header +
971 			le16_to_cpu(entry->e_value_offs));
972 		for (n = (le32_to_cpu(entry->e_value_size) +
973 		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
974 			hash = (hash << VALUE_HASH_SHIFT) ^
975 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
976 			       le32_to_cpu(*value++);
977 		}
978 	}
979 	entry->e_hash = cpu_to_le32(hash);
980 }
981 
982 #undef NAME_HASH_SHIFT
983 #undef VALUE_HASH_SHIFT
984 
985 #define BLOCK_HASH_SHIFT 16
986 
987 /*
988  * ext2_xattr_rehash()
989  *
990  * Re-compute the extended attribute hash value after an entry has changed.
991  */
992 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
993 			      struct ext2_xattr_entry *entry)
994 {
995 	struct ext2_xattr_entry *here;
996 	__u32 hash = 0;
997 
998 	ext2_xattr_hash_entry(header, entry);
999 	here = ENTRY(header+1);
1000 	while (!IS_LAST_ENTRY(here)) {
1001 		if (!here->e_hash) {
1002 			/* Block is not shared if an entry's hash value == 0 */
1003 			hash = 0;
1004 			break;
1005 		}
1006 		hash = (hash << BLOCK_HASH_SHIFT) ^
1007 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1008 		       le32_to_cpu(here->e_hash);
1009 		here = EXT2_XATTR_NEXT(here);
1010 	}
1011 	header->h_hash = cpu_to_le32(hash);
1012 }
1013 
1014 #undef BLOCK_HASH_SHIFT
1015 
1016 int __init
1017 init_ext2_xattr(void)
1018 {
1019 	ext2_xattr_cache = mb_cache_create("ext2_xattr", 6);
1020 	if (!ext2_xattr_cache)
1021 		return -ENOMEM;
1022 	return 0;
1023 }
1024 
1025 void
1026 exit_ext2_xattr(void)
1027 {
1028 	mb_cache_destroy(ext2_xattr_cache);
1029 }
1030