xref: /openbmc/linux/fs/ext2/xattr.c (revision cf21eb6a)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/ext2/xattr.c
4  *
5  * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
6  *
7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8  * Extended attributes for symlinks and special files added per
9  *  suggestion of Luka Renko <luka.renko@hermes.si>.
10  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
11  *  Red Hat Inc.
12  *
13  */
14 
15 /*
16  * Extended attributes are stored on disk blocks allocated outside of
17  * any inode. The i_file_acl field is then made to point to this allocated
18  * block. If all extended attributes of an inode are identical, these
19  * inodes may share the same extended attribute block. Such situations
20  * are automatically detected by keeping a cache of recent attribute block
21  * numbers and hashes over the block's contents in memory.
22  *
23  *
24  * Extended attribute block layout:
25  *
26  *   +------------------+
27  *   | header           |
28  *   | entry 1          | |
29  *   | entry 2          | | growing downwards
30  *   | entry 3          | v
31  *   | four null bytes  |
32  *   | . . .            |
33  *   | value 1          | ^
34  *   | value 3          | | growing upwards
35  *   | value 2          | |
36  *   +------------------+
37  *
38  * The block header is followed by multiple entry descriptors. These entry
39  * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
40  * byte boundaries. The entry descriptors are sorted by attribute name,
41  * so that two extended attribute blocks can be compared efficiently.
42  *
43  * Attribute values are aligned to the end of the block, stored in
44  * no specific order. They are also padded to EXT2_XATTR_PAD byte
45  * boundaries. No additional gaps are left between them.
46  *
47  * Locking strategy
48  * ----------------
49  * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
50  * EA blocks are only changed if they are exclusive to an inode, so
51  * holding xattr_sem also means that nothing but the EA block's reference
52  * count will change. Multiple writers to an EA block are synchronized
53  * by the bh lock. No more than a single bh lock is held at any time
54  * to avoid deadlocks.
55  */
56 
57 #include <linux/buffer_head.h>
58 #include <linux/init.h>
59 #include <linux/printk.h>
60 #include <linux/slab.h>
61 #include <linux/mbcache.h>
62 #include <linux/quotaops.h>
63 #include <linux/rwsem.h>
64 #include <linux/security.h>
65 #include "ext2.h"
66 #include "xattr.h"
67 #include "acl.h"
68 
69 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
70 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
71 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
72 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
73 
74 #ifdef EXT2_XATTR_DEBUG
75 # define ea_idebug(inode, f...) do { \
76 		printk(KERN_DEBUG "inode %s:%ld: ", \
77 			inode->i_sb->s_id, inode->i_ino); \
78 		printk(f); \
79 		printk("\n"); \
80 	} while (0)
81 # define ea_bdebug(bh, f...) do { \
82 		printk(KERN_DEBUG "block %pg:%lu: ", \
83 			bh->b_bdev, (unsigned long) bh->b_blocknr); \
84 		printk(f); \
85 		printk("\n"); \
86 	} while (0)
87 #else
88 # define ea_idebug(inode, f...)	no_printk(f)
89 # define ea_bdebug(bh, f...)	no_printk(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 mb_cache *, 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 const struct xattr_handler *ext2_xattr_handler_map[] = {
102 	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
103 #ifdef CONFIG_EXT2_FS_POSIX_ACL
104 	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &nop_posix_acl_access,
105 	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &nop_posix_acl_default,
106 #endif
107 	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
108 #ifdef CONFIG_EXT2_FS_SECURITY
109 	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
110 #endif
111 };
112 
113 const struct xattr_handler *ext2_xattr_handlers[] = {
114 	&ext2_xattr_user_handler,
115 	&ext2_xattr_trusted_handler,
116 #ifdef CONFIG_EXT2_FS_SECURITY
117 	&ext2_xattr_security_handler,
118 #endif
119 	NULL
120 };
121 
122 #define EA_BLOCK_CACHE(inode)	(EXT2_SB(inode->i_sb)->s_ea_block_cache)
123 
124 static inline const char *ext2_xattr_prefix(int name_index,
125 					    struct dentry *dentry)
126 {
127 	const struct xattr_handler *handler = NULL;
128 
129 	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
130 		handler = ext2_xattr_handler_map[name_index];
131 
132 	if (!xattr_handler_can_list(handler, dentry))
133 		return NULL;
134 
135 	return xattr_prefix(handler);
136 }
137 
138 static bool
139 ext2_xattr_header_valid(struct ext2_xattr_header *header)
140 {
141 	if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
142 	    header->h_blocks != cpu_to_le32(1))
143 		return false;
144 
145 	return true;
146 }
147 
148 static bool
149 ext2_xattr_entry_valid(struct ext2_xattr_entry *entry,
150 		       char *end, size_t end_offs)
151 {
152 	struct ext2_xattr_entry *next;
153 	size_t size;
154 
155 	next = EXT2_XATTR_NEXT(entry);
156 	if ((char *)next >= end)
157 		return false;
158 
159 	if (entry->e_value_block != 0)
160 		return false;
161 
162 	size = le32_to_cpu(entry->e_value_size);
163 	if (size > end_offs ||
164 	    le16_to_cpu(entry->e_value_offs) + size > end_offs)
165 		return false;
166 
167 	return true;
168 }
169 
170 static int
171 ext2_xattr_cmp_entry(int name_index, size_t name_len, const char *name,
172 		     struct ext2_xattr_entry *entry)
173 {
174 	int cmp;
175 
176 	cmp = name_index - entry->e_name_index;
177 	if (!cmp)
178 		cmp = name_len - entry->e_name_len;
179 	if (!cmp)
180 		cmp = memcmp(name, entry->e_name, name_len);
181 
182 	return cmp;
183 }
184 
185 /*
186  * ext2_xattr_get()
187  *
188  * Copy an extended attribute into the buffer
189  * provided, or compute the buffer size required.
190  * Buffer is NULL to compute the size of the buffer required.
191  *
192  * Returns a negative error number on failure, or the number of bytes
193  * used / required on success.
194  */
195 int
196 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
197 	       void *buffer, size_t buffer_size)
198 {
199 	struct buffer_head *bh = NULL;
200 	struct ext2_xattr_entry *entry;
201 	size_t name_len, size;
202 	char *end;
203 	int error, not_found;
204 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
205 
206 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
207 		  name_index, name, buffer, (long)buffer_size);
208 
209 	if (name == NULL)
210 		return -EINVAL;
211 	name_len = strlen(name);
212 	if (name_len > 255)
213 		return -ERANGE;
214 
215 	down_read(&EXT2_I(inode)->xattr_sem);
216 	error = -ENODATA;
217 	if (!EXT2_I(inode)->i_file_acl)
218 		goto cleanup;
219 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
220 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
221 	error = -EIO;
222 	if (!bh)
223 		goto cleanup;
224 	ea_bdebug(bh, "b_count=%d, refcount=%d",
225 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
226 	end = bh->b_data + bh->b_size;
227 	if (!ext2_xattr_header_valid(HDR(bh))) {
228 bad_block:
229 		ext2_error(inode->i_sb, "ext2_xattr_get",
230 			"inode %ld: bad block %d", inode->i_ino,
231 			EXT2_I(inode)->i_file_acl);
232 		error = -EIO;
233 		goto cleanup;
234 	}
235 
236 	/* find named attribute */
237 	entry = FIRST_ENTRY(bh);
238 	while (!IS_LAST_ENTRY(entry)) {
239 		if (!ext2_xattr_entry_valid(entry, end,
240 		    inode->i_sb->s_blocksize))
241 			goto bad_block;
242 
243 		not_found = ext2_xattr_cmp_entry(name_index, name_len, name,
244 						 entry);
245 		if (!not_found)
246 			goto found;
247 		if (not_found < 0)
248 			break;
249 
250 		entry = EXT2_XATTR_NEXT(entry);
251 	}
252 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
253 		ea_idebug(inode, "cache insert failed");
254 	error = -ENODATA;
255 	goto cleanup;
256 found:
257 	size = le32_to_cpu(entry->e_value_size);
258 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
259 		ea_idebug(inode, "cache insert failed");
260 	if (buffer) {
261 		error = -ERANGE;
262 		if (size > buffer_size)
263 			goto cleanup;
264 		/* return value of attribute */
265 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
266 			size);
267 	}
268 	error = size;
269 
270 cleanup:
271 	brelse(bh);
272 	up_read(&EXT2_I(inode)->xattr_sem);
273 
274 	return error;
275 }
276 
277 /*
278  * ext2_xattr_list()
279  *
280  * Copy a list of attribute names into the buffer
281  * provided, or compute the buffer size required.
282  * Buffer is NULL to compute the size of the buffer required.
283  *
284  * Returns a negative error number on failure, or the number of bytes
285  * used / required on success.
286  */
287 static int
288 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
289 {
290 	struct inode *inode = d_inode(dentry);
291 	struct buffer_head *bh = NULL;
292 	struct ext2_xattr_entry *entry;
293 	char *end;
294 	size_t rest = buffer_size;
295 	int error;
296 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
297 
298 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
299 		  buffer, (long)buffer_size);
300 
301 	down_read(&EXT2_I(inode)->xattr_sem);
302 	error = 0;
303 	if (!EXT2_I(inode)->i_file_acl)
304 		goto cleanup;
305 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
306 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
307 	error = -EIO;
308 	if (!bh)
309 		goto cleanup;
310 	ea_bdebug(bh, "b_count=%d, refcount=%d",
311 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
312 	end = bh->b_data + bh->b_size;
313 	if (!ext2_xattr_header_valid(HDR(bh))) {
314 bad_block:
315 		ext2_error(inode->i_sb, "ext2_xattr_list",
316 			"inode %ld: bad block %d", inode->i_ino,
317 			EXT2_I(inode)->i_file_acl);
318 		error = -EIO;
319 		goto cleanup;
320 	}
321 
322 	/* check the on-disk data structure */
323 	entry = FIRST_ENTRY(bh);
324 	while (!IS_LAST_ENTRY(entry)) {
325 		if (!ext2_xattr_entry_valid(entry, end,
326 		    inode->i_sb->s_blocksize))
327 			goto bad_block;
328 		entry = EXT2_XATTR_NEXT(entry);
329 	}
330 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
331 		ea_idebug(inode, "cache insert failed");
332 
333 	/* list the attribute names */
334 	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
335 	     entry = EXT2_XATTR_NEXT(entry)) {
336 		const char *prefix;
337 
338 		prefix = ext2_xattr_prefix(entry->e_name_index, dentry);
339 		if (prefix) {
340 			size_t prefix_len = strlen(prefix);
341 			size_t size = prefix_len + entry->e_name_len + 1;
342 
343 			if (buffer) {
344 				if (size > rest) {
345 					error = -ERANGE;
346 					goto cleanup;
347 				}
348 				memcpy(buffer, prefix, prefix_len);
349 				buffer += prefix_len;
350 				memcpy(buffer, entry->e_name, entry->e_name_len);
351 				buffer += entry->e_name_len;
352 				*buffer++ = 0;
353 			}
354 			rest -= size;
355 		}
356 	}
357 	error = buffer_size - rest;  /* total size */
358 
359 cleanup:
360 	brelse(bh);
361 	up_read(&EXT2_I(inode)->xattr_sem);
362 
363 	return error;
364 }
365 
366 /*
367  * Inode operation listxattr()
368  *
369  * d_inode(dentry)->i_mutex: don't care
370  */
371 ssize_t
372 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
373 {
374 	return ext2_xattr_list(dentry, buffer, size);
375 }
376 
377 /*
378  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
379  * not set, set it.
380  */
381 static void ext2_xattr_update_super_block(struct super_block *sb)
382 {
383 	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
384 		return;
385 
386 	spin_lock(&EXT2_SB(sb)->s_lock);
387 	ext2_update_dynamic_rev(sb);
388 	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
389 	spin_unlock(&EXT2_SB(sb)->s_lock);
390 	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
391 }
392 
393 /*
394  * ext2_xattr_set()
395  *
396  * Create, replace or remove an extended attribute for this inode.  Value
397  * is NULL to remove an existing extended attribute, and non-NULL to
398  * either replace an existing extended attribute, or create a new extended
399  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
400  * specify that an extended attribute must exist and must not exist
401  * previous to the call, respectively.
402  *
403  * Returns 0, or a negative error number on failure.
404  */
405 int
406 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
407 	       const void *value, size_t value_len, int flags)
408 {
409 	struct super_block *sb = inode->i_sb;
410 	struct buffer_head *bh = NULL;
411 	struct ext2_xattr_header *header = NULL;
412 	struct ext2_xattr_entry *here = NULL, *last = NULL;
413 	size_t name_len, free, min_offs = sb->s_blocksize;
414 	int not_found = 1, error;
415 	char *end;
416 
417 	/*
418 	 * header -- Points either into bh, or to a temporarily
419 	 *           allocated buffer.
420 	 * here -- The named entry found, or the place for inserting, within
421 	 *         the block pointed to by header.
422 	 * last -- Points right after the last named entry within the block
423 	 *         pointed to by header.
424 	 * min_offs -- The offset of the first value (values are aligned
425 	 *             towards the end of the block).
426 	 * end -- Points right after the block pointed to by header.
427 	 */
428 
429 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
430 		  name_index, name, value, (long)value_len);
431 
432 	if (value == NULL)
433 		value_len = 0;
434 	if (name == NULL)
435 		return -EINVAL;
436 	name_len = strlen(name);
437 	if (name_len > 255 || value_len > sb->s_blocksize)
438 		return -ERANGE;
439 	error = dquot_initialize(inode);
440 	if (error)
441 		return error;
442 	down_write(&EXT2_I(inode)->xattr_sem);
443 	if (EXT2_I(inode)->i_file_acl) {
444 		/* The inode already has an extended attribute block. */
445 		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
446 		error = -EIO;
447 		if (!bh)
448 			goto cleanup;
449 		ea_bdebug(bh, "b_count=%d, refcount=%d",
450 			atomic_read(&(bh->b_count)),
451 			le32_to_cpu(HDR(bh)->h_refcount));
452 		header = HDR(bh);
453 		end = bh->b_data + bh->b_size;
454 		if (!ext2_xattr_header_valid(header)) {
455 bad_block:
456 			ext2_error(sb, "ext2_xattr_set",
457 				"inode %ld: bad block %d", inode->i_ino,
458 				   EXT2_I(inode)->i_file_acl);
459 			error = -EIO;
460 			goto cleanup;
461 		}
462 		/*
463 		 * Find the named attribute. If not found, 'here' will point
464 		 * to entry where the new attribute should be inserted to
465 		 * maintain sorting.
466 		 */
467 		last = FIRST_ENTRY(bh);
468 		while (!IS_LAST_ENTRY(last)) {
469 			if (!ext2_xattr_entry_valid(last, end, sb->s_blocksize))
470 				goto bad_block;
471 			if (last->e_value_size) {
472 				size_t offs = le16_to_cpu(last->e_value_offs);
473 				if (offs < min_offs)
474 					min_offs = offs;
475 			}
476 			if (not_found > 0) {
477 				not_found = ext2_xattr_cmp_entry(name_index,
478 								 name_len,
479 								 name, last);
480 				if (not_found <= 0)
481 					here = last;
482 			}
483 			last = EXT2_XATTR_NEXT(last);
484 		}
485 		if (not_found > 0)
486 			here = last;
487 
488 		/* Check whether we have enough space left. */
489 		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
490 	} else {
491 		/* We will use a new extended attribute block. */
492 		free = sb->s_blocksize -
493 			sizeof(struct ext2_xattr_header) - sizeof(__u32);
494 	}
495 
496 	if (not_found) {
497 		/* Request to remove a nonexistent attribute? */
498 		error = -ENODATA;
499 		if (flags & XATTR_REPLACE)
500 			goto cleanup;
501 		error = 0;
502 		if (value == NULL)
503 			goto cleanup;
504 	} else {
505 		/* Request to create an existing attribute? */
506 		error = -EEXIST;
507 		if (flags & XATTR_CREATE)
508 			goto cleanup;
509 		free += EXT2_XATTR_SIZE(le32_to_cpu(here->e_value_size));
510 		free += EXT2_XATTR_LEN(name_len);
511 	}
512 	error = -ENOSPC;
513 	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
514 		goto cleanup;
515 
516 	/* Here we know that we can set the new attribute. */
517 
518 	if (header) {
519 		int offset;
520 
521 		lock_buffer(bh);
522 		if (header->h_refcount == cpu_to_le32(1)) {
523 			__u32 hash = le32_to_cpu(header->h_hash);
524 			struct mb_cache_entry *oe;
525 
526 			oe = mb_cache_entry_delete_or_get(EA_BLOCK_CACHE(inode),
527 					hash, bh->b_blocknr);
528 			if (!oe) {
529 				ea_bdebug(bh, "modifying in-place");
530 				goto update_block;
531 			}
532 			/*
533 			 * Someone is trying to reuse the block, leave it alone
534 			 */
535 			mb_cache_entry_put(EA_BLOCK_CACHE(inode), oe);
536 		}
537 		unlock_buffer(bh);
538 		ea_bdebug(bh, "cloning");
539 		header = kmemdup(HDR(bh), bh->b_size, GFP_KERNEL);
540 		error = -ENOMEM;
541 		if (header == NULL)
542 			goto cleanup;
543 		header->h_refcount = cpu_to_le32(1);
544 
545 		offset = (char *)here - bh->b_data;
546 		here = ENTRY((char *)header + offset);
547 		offset = (char *)last - bh->b_data;
548 		last = ENTRY((char *)header + offset);
549 	} else {
550 		/* Allocate a buffer where we construct the new block. */
551 		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
552 		error = -ENOMEM;
553 		if (header == NULL)
554 			goto cleanup;
555 		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
556 		header->h_blocks = header->h_refcount = cpu_to_le32(1);
557 		last = here = ENTRY(header+1);
558 	}
559 
560 update_block:
561 	/* Iff we are modifying the block in-place, bh is locked here. */
562 
563 	if (not_found) {
564 		/* Insert the new name. */
565 		size_t size = EXT2_XATTR_LEN(name_len);
566 		size_t rest = (char *)last - (char *)here;
567 		memmove((char *)here + size, here, rest);
568 		memset(here, 0, size);
569 		here->e_name_index = name_index;
570 		here->e_name_len = name_len;
571 		memcpy(here->e_name, name, name_len);
572 	} else {
573 		if (here->e_value_size) {
574 			char *first_val = (char *)header + min_offs;
575 			size_t offs = le16_to_cpu(here->e_value_offs);
576 			char *val = (char *)header + offs;
577 			size_t size = EXT2_XATTR_SIZE(
578 				le32_to_cpu(here->e_value_size));
579 
580 			if (size == EXT2_XATTR_SIZE(value_len)) {
581 				/* The old and the new value have the same
582 				   size. Just replace. */
583 				here->e_value_size = cpu_to_le32(value_len);
584 				memset(val + size - EXT2_XATTR_PAD, 0,
585 				       EXT2_XATTR_PAD); /* Clear pad bytes. */
586 				memcpy(val, value, value_len);
587 				goto skip_replace;
588 			}
589 
590 			/* Remove the old value. */
591 			memmove(first_val + size, first_val, val - first_val);
592 			memset(first_val, 0, size);
593 			min_offs += size;
594 
595 			/* Adjust all value offsets. */
596 			last = ENTRY(header+1);
597 			while (!IS_LAST_ENTRY(last)) {
598 				size_t o = le16_to_cpu(last->e_value_offs);
599 				if (o < offs)
600 					last->e_value_offs =
601 						cpu_to_le16(o + size);
602 				last = EXT2_XATTR_NEXT(last);
603 			}
604 
605 			here->e_value_offs = 0;
606 		}
607 		if (value == NULL) {
608 			/* Remove the old name. */
609 			size_t size = EXT2_XATTR_LEN(name_len);
610 			last = ENTRY((char *)last - size);
611 			memmove(here, (char*)here + size,
612 				(char*)last - (char*)here);
613 			memset(last, 0, size);
614 		}
615 	}
616 
617 	if (value != NULL) {
618 		/* Insert the new value. */
619 		here->e_value_size = cpu_to_le32(value_len);
620 		if (value_len) {
621 			size_t size = EXT2_XATTR_SIZE(value_len);
622 			char *val = (char *)header + min_offs - size;
623 			here->e_value_offs =
624 				cpu_to_le16((char *)val - (char *)header);
625 			memset(val + size - EXT2_XATTR_PAD, 0,
626 			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
627 			memcpy(val, value, value_len);
628 		}
629 	}
630 
631 skip_replace:
632 	if (IS_LAST_ENTRY(ENTRY(header+1))) {
633 		/* This block is now empty. */
634 		if (bh && header == HDR(bh))
635 			unlock_buffer(bh);  /* we were modifying in-place. */
636 		error = ext2_xattr_set2(inode, bh, NULL);
637 	} else {
638 		ext2_xattr_rehash(header, here);
639 		if (bh && header == HDR(bh))
640 			unlock_buffer(bh);  /* we were modifying in-place. */
641 		error = ext2_xattr_set2(inode, bh, header);
642 	}
643 
644 cleanup:
645 	if (!(bh && header == HDR(bh)))
646 		kfree(header);
647 	brelse(bh);
648 	up_write(&EXT2_I(inode)->xattr_sem);
649 
650 	return error;
651 }
652 
653 static void ext2_xattr_release_block(struct inode *inode,
654 				     struct buffer_head *bh)
655 {
656 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
657 
658 retry_ref:
659 	lock_buffer(bh);
660 	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
661 		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
662 		struct mb_cache_entry *oe;
663 
664 		/*
665 		 * This must happen under buffer lock to properly
666 		 * serialize with ext2_xattr_set() reusing the block.
667 		 */
668 		oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
669 						  bh->b_blocknr);
670 		if (oe) {
671 			/*
672 			 * Someone is trying to reuse the block. Wait
673 			 * and retry.
674 			 */
675 			unlock_buffer(bh);
676 			mb_cache_entry_wait_unused(oe);
677 			mb_cache_entry_put(ea_block_cache, oe);
678 			goto retry_ref;
679 		}
680 
681 		/* Free the old block. */
682 		ea_bdebug(bh, "freeing");
683 		ext2_free_blocks(inode, bh->b_blocknr, 1);
684 		/* We let our caller release bh, so we
685 		 * need to duplicate the buffer before. */
686 		get_bh(bh);
687 		bforget(bh);
688 		unlock_buffer(bh);
689 	} else {
690 		/* Decrement the refcount only. */
691 		le32_add_cpu(&HDR(bh)->h_refcount, -1);
692 		dquot_free_block(inode, 1);
693 		mark_buffer_dirty(bh);
694 		unlock_buffer(bh);
695 		ea_bdebug(bh, "refcount now=%d",
696 			le32_to_cpu(HDR(bh)->h_refcount));
697 		if (IS_SYNC(inode))
698 			sync_dirty_buffer(bh);
699 	}
700 }
701 
702 /*
703  * Second half of ext2_xattr_set(): Update the file system.
704  */
705 static int
706 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
707 		struct ext2_xattr_header *header)
708 {
709 	struct super_block *sb = inode->i_sb;
710 	struct buffer_head *new_bh = NULL;
711 	int error;
712 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
713 
714 	if (header) {
715 		new_bh = ext2_xattr_cache_find(inode, header);
716 		if (new_bh) {
717 			/* We found an identical block in the cache. */
718 			if (new_bh == old_bh) {
719 				ea_bdebug(new_bh, "keeping this block");
720 			} else {
721 				/* The old block is released after updating
722 				   the inode.  */
723 				ea_bdebug(new_bh, "reusing block");
724 
725 				error = dquot_alloc_block(inode, 1);
726 				if (error) {
727 					unlock_buffer(new_bh);
728 					goto cleanup;
729 				}
730 				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
731 				ea_bdebug(new_bh, "refcount now=%d",
732 					le32_to_cpu(HDR(new_bh)->h_refcount));
733 			}
734 			unlock_buffer(new_bh);
735 		} else if (old_bh && header == HDR(old_bh)) {
736 			/* Keep this block. No need to lock the block as we
737 			   don't need to change the reference count. */
738 			new_bh = old_bh;
739 			get_bh(new_bh);
740 			ext2_xattr_cache_insert(ea_block_cache, new_bh);
741 		} else {
742 			/* We need to allocate a new block */
743 			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
744 						EXT2_I(inode)->i_block_group);
745 			unsigned long count = 1;
746 			ext2_fsblk_t block = ext2_new_blocks(inode, goal,
747 						&count, &error,
748 						EXT2_ALLOC_NORESERVE);
749 			if (error)
750 				goto cleanup;
751 			ea_idebug(inode, "creating block %lu", block);
752 
753 			new_bh = sb_getblk(sb, block);
754 			if (unlikely(!new_bh)) {
755 				ext2_free_blocks(inode, block, 1);
756 				mark_inode_dirty(inode);
757 				error = -ENOMEM;
758 				goto cleanup;
759 			}
760 			lock_buffer(new_bh);
761 			memcpy(new_bh->b_data, header, new_bh->b_size);
762 			set_buffer_uptodate(new_bh);
763 			unlock_buffer(new_bh);
764 			ext2_xattr_cache_insert(ea_block_cache, new_bh);
765 
766 			ext2_xattr_update_super_block(sb);
767 		}
768 		mark_buffer_dirty(new_bh);
769 		if (IS_SYNC(inode)) {
770 			sync_dirty_buffer(new_bh);
771 			error = -EIO;
772 			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
773 				goto cleanup;
774 		}
775 	}
776 
777 	/* Update the inode. */
778 	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
779 	inode_set_ctime_current(inode);
780 	if (IS_SYNC(inode)) {
781 		error = sync_inode_metadata(inode, 1);
782 		/* In case sync failed due to ENOSPC the inode was actually
783 		 * written (only some dirty data were not) so we just proceed
784 		 * as if nothing happened and cleanup the unused block */
785 		if (error && error != -ENOSPC) {
786 			if (new_bh && new_bh != old_bh) {
787 				dquot_free_block_nodirty(inode, 1);
788 				mark_inode_dirty(inode);
789 			}
790 			goto cleanup;
791 		}
792 	} else
793 		mark_inode_dirty(inode);
794 
795 	error = 0;
796 	if (old_bh && old_bh != new_bh) {
797 		/*
798 		 * If there was an old block and we are no longer using it,
799 		 * release the old block.
800 		 */
801 		ext2_xattr_release_block(inode, old_bh);
802 	}
803 
804 cleanup:
805 	brelse(new_bh);
806 
807 	return error;
808 }
809 
810 /*
811  * ext2_xattr_delete_inode()
812  *
813  * Free extended attribute resources associated with this inode. This
814  * is called immediately before an inode is freed.
815  */
816 void
817 ext2_xattr_delete_inode(struct inode *inode)
818 {
819 	struct buffer_head *bh = NULL;
820 	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
821 
822 	/*
823 	 * We are the only ones holding inode reference. The xattr_sem should
824 	 * better be unlocked! We could as well just not acquire xattr_sem at
825 	 * all but this makes the code more futureproof. OTOH we need trylock
826 	 * here to avoid false-positive warning from lockdep about reclaim
827 	 * circular dependency.
828 	 */
829 	if (WARN_ON_ONCE(!down_write_trylock(&EXT2_I(inode)->xattr_sem)))
830 		return;
831 	if (!EXT2_I(inode)->i_file_acl)
832 		goto cleanup;
833 
834 	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 1)) {
835 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
836 			"inode %ld: xattr block %d is out of data blocks range",
837 			inode->i_ino, EXT2_I(inode)->i_file_acl);
838 		goto cleanup;
839 	}
840 
841 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
842 	if (!bh) {
843 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
844 			"inode %ld: block %d read error", inode->i_ino,
845 			EXT2_I(inode)->i_file_acl);
846 		goto cleanup;
847 	}
848 	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
849 	if (!ext2_xattr_header_valid(HDR(bh))) {
850 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
851 			"inode %ld: bad block %d", inode->i_ino,
852 			EXT2_I(inode)->i_file_acl);
853 		goto cleanup;
854 	}
855 	ext2_xattr_release_block(inode, bh);
856 	EXT2_I(inode)->i_file_acl = 0;
857 
858 cleanup:
859 	brelse(bh);
860 	up_write(&EXT2_I(inode)->xattr_sem);
861 }
862 
863 /*
864  * ext2_xattr_cache_insert()
865  *
866  * Create a new entry in the extended attribute cache, and insert
867  * it unless such an entry is already in the cache.
868  *
869  * Returns 0, or a negative error number on failure.
870  */
871 static int
872 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
873 {
874 	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
875 	int error;
876 
877 	error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr,
878 				      true);
879 	if (error) {
880 		if (error == -EBUSY) {
881 			ea_bdebug(bh, "already in cache");
882 			error = 0;
883 		}
884 	} else
885 		ea_bdebug(bh, "inserting [%x]", (int)hash);
886 	return error;
887 }
888 
889 /*
890  * ext2_xattr_cmp()
891  *
892  * Compare two extended attribute blocks for equality.
893  *
894  * Returns 0 if the blocks are equal, 1 if they differ, and
895  * a negative error number on errors.
896  */
897 static int
898 ext2_xattr_cmp(struct ext2_xattr_header *header1,
899 	       struct ext2_xattr_header *header2)
900 {
901 	struct ext2_xattr_entry *entry1, *entry2;
902 
903 	entry1 = ENTRY(header1+1);
904 	entry2 = ENTRY(header2+1);
905 	while (!IS_LAST_ENTRY(entry1)) {
906 		if (IS_LAST_ENTRY(entry2))
907 			return 1;
908 		if (entry1->e_hash != entry2->e_hash ||
909 		    entry1->e_name_index != entry2->e_name_index ||
910 		    entry1->e_name_len != entry2->e_name_len ||
911 		    entry1->e_value_size != entry2->e_value_size ||
912 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
913 			return 1;
914 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
915 			return -EIO;
916 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
917 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
918 			   le32_to_cpu(entry1->e_value_size)))
919 			return 1;
920 
921 		entry1 = EXT2_XATTR_NEXT(entry1);
922 		entry2 = EXT2_XATTR_NEXT(entry2);
923 	}
924 	if (!IS_LAST_ENTRY(entry2))
925 		return 1;
926 	return 0;
927 }
928 
929 /*
930  * ext2_xattr_cache_find()
931  *
932  * Find an identical extended attribute block.
933  *
934  * Returns a locked buffer head to the block found, or NULL if such
935  * a block was not found or an error occurred.
936  */
937 static struct buffer_head *
938 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
939 {
940 	__u32 hash = le32_to_cpu(header->h_hash);
941 	struct mb_cache_entry *ce;
942 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
943 
944 	if (!header->h_hash)
945 		return NULL;  /* never share */
946 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
947 
948 	ce = mb_cache_entry_find_first(ea_block_cache, hash);
949 	while (ce) {
950 		struct buffer_head *bh;
951 
952 		bh = sb_bread(inode->i_sb, ce->e_value);
953 		if (!bh) {
954 			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
955 				"inode %ld: block %ld read error",
956 				inode->i_ino, (unsigned long) ce->e_value);
957 		} else {
958 			lock_buffer(bh);
959 			if (le32_to_cpu(HDR(bh)->h_refcount) >
960 			    EXT2_XATTR_REFCOUNT_MAX) {
961 				ea_idebug(inode, "block %ld refcount %d>%d",
962 					  (unsigned long) ce->e_value,
963 					  le32_to_cpu(HDR(bh)->h_refcount),
964 					  EXT2_XATTR_REFCOUNT_MAX);
965 			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
966 				ea_bdebug(bh, "b_count=%d",
967 					  atomic_read(&(bh->b_count)));
968 				mb_cache_entry_touch(ea_block_cache, ce);
969 				mb_cache_entry_put(ea_block_cache, ce);
970 				return bh;
971 			}
972 			unlock_buffer(bh);
973 			brelse(bh);
974 		}
975 		ce = mb_cache_entry_find_next(ea_block_cache, ce);
976 	}
977 	return NULL;
978 }
979 
980 #define NAME_HASH_SHIFT 5
981 #define VALUE_HASH_SHIFT 16
982 
983 /*
984  * ext2_xattr_hash_entry()
985  *
986  * Compute the hash of an extended attribute.
987  */
988 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
989 					 struct ext2_xattr_entry *entry)
990 {
991 	__u32 hash = 0;
992 	char *name = entry->e_name;
993 	int n;
994 
995 	for (n=0; n < entry->e_name_len; n++) {
996 		hash = (hash << NAME_HASH_SHIFT) ^
997 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
998 		       *name++;
999 	}
1000 
1001 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1002 		__le32 *value = (__le32 *)((char *)header +
1003 			le16_to_cpu(entry->e_value_offs));
1004 		for (n = (le32_to_cpu(entry->e_value_size) +
1005 		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
1006 			hash = (hash << VALUE_HASH_SHIFT) ^
1007 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1008 			       le32_to_cpu(*value++);
1009 		}
1010 	}
1011 	entry->e_hash = cpu_to_le32(hash);
1012 }
1013 
1014 #undef NAME_HASH_SHIFT
1015 #undef VALUE_HASH_SHIFT
1016 
1017 #define BLOCK_HASH_SHIFT 16
1018 
1019 /*
1020  * ext2_xattr_rehash()
1021  *
1022  * Re-compute the extended attribute hash value after an entry has changed.
1023  */
1024 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1025 			      struct ext2_xattr_entry *entry)
1026 {
1027 	struct ext2_xattr_entry *here;
1028 	__u32 hash = 0;
1029 
1030 	ext2_xattr_hash_entry(header, entry);
1031 	here = ENTRY(header+1);
1032 	while (!IS_LAST_ENTRY(here)) {
1033 		if (!here->e_hash) {
1034 			/* Block is not shared if an entry's hash value == 0 */
1035 			hash = 0;
1036 			break;
1037 		}
1038 		hash = (hash << BLOCK_HASH_SHIFT) ^
1039 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1040 		       le32_to_cpu(here->e_hash);
1041 		here = EXT2_XATTR_NEXT(here);
1042 	}
1043 	header->h_hash = cpu_to_le32(hash);
1044 }
1045 
1046 #undef BLOCK_HASH_SHIFT
1047 
1048 #define HASH_BUCKET_BITS 10
1049 
1050 struct mb_cache *ext2_xattr_create_cache(void)
1051 {
1052 	return mb_cache_create(HASH_BUCKET_BITS);
1053 }
1054 
1055 void ext2_xattr_destroy_cache(struct mb_cache *cache)
1056 {
1057 	if (cache)
1058 		mb_cache_destroy(cache);
1059 }
1060