xref: /openbmc/linux/fs/ext2/xattr.c (revision d561d4dd)
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/slab.h>
60 #include <linux/mbcache.h>
61 #include <linux/quotaops.h>
62 #include <linux/rwsem.h>
63 #include <linux/security.h>
64 #include "ext2.h"
65 #include "xattr.h"
66 #include "acl.h"
67 
68 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
69 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
70 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
71 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
72 
73 #ifdef EXT2_XATTR_DEBUG
74 # define ea_idebug(inode, f...) do { \
75 		printk(KERN_DEBUG "inode %s:%ld: ", \
76 			inode->i_sb->s_id, inode->i_ino); \
77 		printk(f); \
78 		printk("\n"); \
79 	} while (0)
80 # define ea_bdebug(bh, f...) do { \
81 		printk(KERN_DEBUG "block %pg:%lu: ", \
82 			bh->b_bdev, (unsigned long) bh->b_blocknr); \
83 		printk(f); \
84 		printk("\n"); \
85 	} while (0)
86 #else
87 # define ea_idebug(f...)
88 # define ea_bdebug(f...)
89 #endif
90 
91 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
92 			   struct ext2_xattr_header *);
93 
94 static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
95 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
96 						 struct ext2_xattr_header *);
97 static void ext2_xattr_rehash(struct ext2_xattr_header *,
98 			      struct ext2_xattr_entry *);
99 
100 static const struct xattr_handler *ext2_xattr_handler_map[] = {
101 	[EXT2_XATTR_INDEX_USER]		     = &ext2_xattr_user_handler,
102 #ifdef CONFIG_EXT2_FS_POSIX_ACL
103 	[EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
104 	[EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
105 #endif
106 	[EXT2_XATTR_INDEX_TRUSTED]	     = &ext2_xattr_trusted_handler,
107 #ifdef CONFIG_EXT2_FS_SECURITY
108 	[EXT2_XATTR_INDEX_SECURITY]	     = &ext2_xattr_security_handler,
109 #endif
110 };
111 
112 const struct xattr_handler *ext2_xattr_handlers[] = {
113 	&ext2_xattr_user_handler,
114 	&ext2_xattr_trusted_handler,
115 #ifdef CONFIG_EXT2_FS_POSIX_ACL
116 	&posix_acl_access_xattr_handler,
117 	&posix_acl_default_xattr_handler,
118 #endif
119 #ifdef CONFIG_EXT2_FS_SECURITY
120 	&ext2_xattr_security_handler,
121 #endif
122 	NULL
123 };
124 
125 #define EA_BLOCK_CACHE(inode)	(EXT2_SB(inode->i_sb)->s_ea_block_cache)
126 
127 static inline const struct xattr_handler *
128 ext2_xattr_handler(int name_index)
129 {
130 	const struct xattr_handler *handler = NULL;
131 
132 	if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
133 		handler = ext2_xattr_handler_map[name_index];
134 	return handler;
135 }
136 
137 static bool
138 ext2_xattr_header_valid(struct ext2_xattr_header *header)
139 {
140 	if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
141 	    header->h_blocks != cpu_to_le32(1))
142 		return false;
143 
144 	return true;
145 }
146 
147 static bool
148 ext2_xattr_entry_valid(struct ext2_xattr_entry *entry,
149 		       char *end, size_t end_offs)
150 {
151 	struct ext2_xattr_entry *next;
152 	size_t size;
153 
154 	next = EXT2_XATTR_NEXT(entry);
155 	if ((char *)next >= end)
156 		return false;
157 
158 	if (entry->e_value_block != 0)
159 		return false;
160 
161 	size = le32_to_cpu(entry->e_value_size);
162 	if (size > end_offs ||
163 	    le16_to_cpu(entry->e_value_offs) + size > end_offs)
164 		return false;
165 
166 	return true;
167 }
168 
169 static int
170 ext2_xattr_cmp_entry(int name_index, size_t name_len, const char *name,
171 		     struct ext2_xattr_entry *entry)
172 {
173 	int cmp;
174 
175 	cmp = name_index - entry->e_name_index;
176 	if (!cmp)
177 		cmp = name_len - entry->e_name_len;
178 	if (!cmp)
179 		cmp = memcmp(name, entry->e_name, name_len);
180 
181 	return cmp;
182 }
183 
184 /*
185  * ext2_xattr_get()
186  *
187  * Copy an extended attribute into the buffer
188  * provided, or compute the buffer size required.
189  * Buffer is NULL to compute the size of the buffer required.
190  *
191  * Returns a negative error number on failure, or the number of bytes
192  * used / required on success.
193  */
194 int
195 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
196 	       void *buffer, size_t buffer_size)
197 {
198 	struct buffer_head *bh = NULL;
199 	struct ext2_xattr_entry *entry;
200 	size_t name_len, size;
201 	char *end;
202 	int error;
203 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
204 
205 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
206 		  name_index, name, buffer, (long)buffer_size);
207 
208 	if (name == NULL)
209 		return -EINVAL;
210 	name_len = strlen(name);
211 	if (name_len > 255)
212 		return -ERANGE;
213 
214 	down_read(&EXT2_I(inode)->xattr_sem);
215 	error = -ENODATA;
216 	if (!EXT2_I(inode)->i_file_acl)
217 		goto cleanup;
218 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
219 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
220 	error = -EIO;
221 	if (!bh)
222 		goto cleanup;
223 	ea_bdebug(bh, "b_count=%d, refcount=%d",
224 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
225 	end = bh->b_data + bh->b_size;
226 	if (!ext2_xattr_header_valid(HDR(bh))) {
227 bad_block:
228 		ext2_error(inode->i_sb, "ext2_xattr_get",
229 			"inode %ld: bad block %d", inode->i_ino,
230 			EXT2_I(inode)->i_file_acl);
231 		error = -EIO;
232 		goto cleanup;
233 	}
234 
235 	/* find named attribute */
236 	entry = FIRST_ENTRY(bh);
237 	while (!IS_LAST_ENTRY(entry)) {
238 		if (!ext2_xattr_entry_valid(entry, end,
239 		    inode->i_sb->s_blocksize))
240 			goto bad_block;
241 		if (name_index == entry->e_name_index &&
242 		    name_len == entry->e_name_len &&
243 		    memcmp(name, entry->e_name, name_len) == 0)
244 			goto found;
245 		entry = EXT2_XATTR_NEXT(entry);
246 	}
247 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
248 		ea_idebug(inode, "cache insert failed");
249 	error = -ENODATA;
250 	goto cleanup;
251 found:
252 	size = le32_to_cpu(entry->e_value_size);
253 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
254 		ea_idebug(inode, "cache insert failed");
255 	if (buffer) {
256 		error = -ERANGE;
257 		if (size > buffer_size)
258 			goto cleanup;
259 		/* return value of attribute */
260 		memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
261 			size);
262 	}
263 	error = size;
264 
265 cleanup:
266 	brelse(bh);
267 	up_read(&EXT2_I(inode)->xattr_sem);
268 
269 	return error;
270 }
271 
272 /*
273  * ext2_xattr_list()
274  *
275  * Copy a list of attribute names into the buffer
276  * provided, or compute the buffer size required.
277  * Buffer is NULL to compute the size of the buffer required.
278  *
279  * Returns a negative error number on failure, or the number of bytes
280  * used / required on success.
281  */
282 static int
283 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
284 {
285 	struct inode *inode = d_inode(dentry);
286 	struct buffer_head *bh = NULL;
287 	struct ext2_xattr_entry *entry;
288 	char *end;
289 	size_t rest = buffer_size;
290 	int error;
291 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
292 
293 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
294 		  buffer, (long)buffer_size);
295 
296 	down_read(&EXT2_I(inode)->xattr_sem);
297 	error = 0;
298 	if (!EXT2_I(inode)->i_file_acl)
299 		goto cleanup;
300 	ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
301 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
302 	error = -EIO;
303 	if (!bh)
304 		goto cleanup;
305 	ea_bdebug(bh, "b_count=%d, refcount=%d",
306 		atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
307 	end = bh->b_data + bh->b_size;
308 	if (!ext2_xattr_header_valid(HDR(bh))) {
309 bad_block:
310 		ext2_error(inode->i_sb, "ext2_xattr_list",
311 			"inode %ld: bad block %d", inode->i_ino,
312 			EXT2_I(inode)->i_file_acl);
313 		error = -EIO;
314 		goto cleanup;
315 	}
316 
317 	/* check the on-disk data structure */
318 	entry = FIRST_ENTRY(bh);
319 	while (!IS_LAST_ENTRY(entry)) {
320 		if (!ext2_xattr_entry_valid(entry, end,
321 		    inode->i_sb->s_blocksize))
322 			goto bad_block;
323 		entry = EXT2_XATTR_NEXT(entry);
324 	}
325 	if (ext2_xattr_cache_insert(ea_block_cache, bh))
326 		ea_idebug(inode, "cache insert failed");
327 
328 	/* list the attribute names */
329 	for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
330 	     entry = EXT2_XATTR_NEXT(entry)) {
331 		const struct xattr_handler *handler =
332 			ext2_xattr_handler(entry->e_name_index);
333 
334 		if (handler && (!handler->list || handler->list(dentry))) {
335 			const char *prefix = handler->prefix ?: handler->name;
336 			size_t prefix_len = strlen(prefix);
337 			size_t size = prefix_len + entry->e_name_len + 1;
338 
339 			if (buffer) {
340 				if (size > rest) {
341 					error = -ERANGE;
342 					goto cleanup;
343 				}
344 				memcpy(buffer, prefix, prefix_len);
345 				buffer += prefix_len;
346 				memcpy(buffer, entry->e_name, entry->e_name_len);
347 				buffer += entry->e_name_len;
348 				*buffer++ = 0;
349 			}
350 			rest -= size;
351 		}
352 	}
353 	error = buffer_size - rest;  /* total size */
354 
355 cleanup:
356 	brelse(bh);
357 	up_read(&EXT2_I(inode)->xattr_sem);
358 
359 	return error;
360 }
361 
362 /*
363  * Inode operation listxattr()
364  *
365  * d_inode(dentry)->i_mutex: don't care
366  */
367 ssize_t
368 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
369 {
370 	return ext2_xattr_list(dentry, buffer, size);
371 }
372 
373 /*
374  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
375  * not set, set it.
376  */
377 static void ext2_xattr_update_super_block(struct super_block *sb)
378 {
379 	if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
380 		return;
381 
382 	spin_lock(&EXT2_SB(sb)->s_lock);
383 	ext2_update_dynamic_rev(sb);
384 	EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
385 	spin_unlock(&EXT2_SB(sb)->s_lock);
386 	mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
387 }
388 
389 /*
390  * ext2_xattr_set()
391  *
392  * Create, replace or remove an extended attribute for this inode.  Value
393  * is NULL to remove an existing extended attribute, and non-NULL to
394  * either replace an existing extended attribute, or create a new extended
395  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
396  * specify that an extended attribute must exist and must not exist
397  * previous to the call, respectively.
398  *
399  * Returns 0, or a negative error number on failure.
400  */
401 int
402 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
403 	       const void *value, size_t value_len, int flags)
404 {
405 	struct super_block *sb = inode->i_sb;
406 	struct buffer_head *bh = NULL;
407 	struct ext2_xattr_header *header = NULL;
408 	struct ext2_xattr_entry *here = NULL, *last = NULL;
409 	size_t name_len, free, min_offs = sb->s_blocksize;
410 	int not_found = 1, error;
411 	char *end;
412 
413 	/*
414 	 * header -- Points either into bh, or to a temporarily
415 	 *           allocated buffer.
416 	 * here -- The named entry found, or the place for inserting, within
417 	 *         the block pointed to by header.
418 	 * last -- Points right after the last named entry within the block
419 	 *         pointed to by header.
420 	 * min_offs -- The offset of the first value (values are aligned
421 	 *             towards the end of the block).
422 	 * end -- Points right after the block pointed to by header.
423 	 */
424 
425 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
426 		  name_index, name, value, (long)value_len);
427 
428 	if (value == NULL)
429 		value_len = 0;
430 	if (name == NULL)
431 		return -EINVAL;
432 	name_len = strlen(name);
433 	if (name_len > 255 || value_len > sb->s_blocksize)
434 		return -ERANGE;
435 	down_write(&EXT2_I(inode)->xattr_sem);
436 	if (EXT2_I(inode)->i_file_acl) {
437 		/* The inode already has an extended attribute block. */
438 		bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
439 		error = -EIO;
440 		if (!bh)
441 			goto cleanup;
442 		ea_bdebug(bh, "b_count=%d, refcount=%d",
443 			atomic_read(&(bh->b_count)),
444 			le32_to_cpu(HDR(bh)->h_refcount));
445 		header = HDR(bh);
446 		end = bh->b_data + bh->b_size;
447 		if (!ext2_xattr_header_valid(header)) {
448 bad_block:
449 			ext2_error(sb, "ext2_xattr_set",
450 				"inode %ld: bad block %d", inode->i_ino,
451 				   EXT2_I(inode)->i_file_acl);
452 			error = -EIO;
453 			goto cleanup;
454 		}
455 		/*
456 		 * Find the named attribute. If not found, 'here' will point
457 		 * to entry where the new attribute should be inserted to
458 		 * maintain sorting.
459 		 */
460 		last = FIRST_ENTRY(bh);
461 		while (!IS_LAST_ENTRY(last)) {
462 			if (!ext2_xattr_entry_valid(last, end, sb->s_blocksize))
463 				goto bad_block;
464 			if (last->e_value_size) {
465 				size_t offs = le16_to_cpu(last->e_value_offs);
466 				if (offs < min_offs)
467 					min_offs = offs;
468 			}
469 			if (not_found > 0) {
470 				not_found = ext2_xattr_cmp_entry(name_index,
471 								 name_len,
472 								 name, last);
473 				if (not_found <= 0)
474 					here = last;
475 			}
476 			last = EXT2_XATTR_NEXT(last);
477 		}
478 		if (not_found > 0)
479 			here = last;
480 
481 		/* Check whether we have enough space left. */
482 		free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
483 	} else {
484 		/* We will use a new extended attribute block. */
485 		free = sb->s_blocksize -
486 			sizeof(struct ext2_xattr_header) - sizeof(__u32);
487 	}
488 
489 	if (not_found) {
490 		/* Request to remove a nonexistent attribute? */
491 		error = -ENODATA;
492 		if (flags & XATTR_REPLACE)
493 			goto cleanup;
494 		error = 0;
495 		if (value == NULL)
496 			goto cleanup;
497 	} else {
498 		/* Request to create an existing attribute? */
499 		error = -EEXIST;
500 		if (flags & XATTR_CREATE)
501 			goto cleanup;
502 		free += EXT2_XATTR_SIZE(le32_to_cpu(here->e_value_size));
503 		free += EXT2_XATTR_LEN(name_len);
504 	}
505 	error = -ENOSPC;
506 	if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
507 		goto cleanup;
508 
509 	/* Here we know that we can set the new attribute. */
510 
511 	if (header) {
512 		/* assert(header == HDR(bh)); */
513 		lock_buffer(bh);
514 		if (header->h_refcount == cpu_to_le32(1)) {
515 			__u32 hash = le32_to_cpu(header->h_hash);
516 
517 			ea_bdebug(bh, "modifying in-place");
518 			/*
519 			 * This must happen under buffer lock for
520 			 * ext2_xattr_set2() to reliably detect modified block
521 			 */
522 			mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
523 					      bh->b_blocknr);
524 
525 			/* keep the buffer locked while modifying it. */
526 		} else {
527 			int offset;
528 
529 			unlock_buffer(bh);
530 			ea_bdebug(bh, "cloning");
531 			header = kmalloc(bh->b_size, GFP_KERNEL);
532 			error = -ENOMEM;
533 			if (header == NULL)
534 				goto cleanup;
535 			memcpy(header, HDR(bh), bh->b_size);
536 			header->h_refcount = cpu_to_le32(1);
537 
538 			offset = (char *)here - bh->b_data;
539 			here = ENTRY((char *)header + offset);
540 			offset = (char *)last - bh->b_data;
541 			last = ENTRY((char *)header + offset);
542 		}
543 	} else {
544 		/* Allocate a buffer where we construct the new block. */
545 		header = kzalloc(sb->s_blocksize, GFP_KERNEL);
546 		error = -ENOMEM;
547 		if (header == NULL)
548 			goto cleanup;
549 		end = (char *)header + sb->s_blocksize;
550 		header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
551 		header->h_blocks = header->h_refcount = cpu_to_le32(1);
552 		last = here = ENTRY(header+1);
553 	}
554 
555 	/* Iff we are modifying the block in-place, bh is locked here. */
556 
557 	if (not_found) {
558 		/* Insert the new name. */
559 		size_t size = EXT2_XATTR_LEN(name_len);
560 		size_t rest = (char *)last - (char *)here;
561 		memmove((char *)here + size, here, rest);
562 		memset(here, 0, size);
563 		here->e_name_index = name_index;
564 		here->e_name_len = name_len;
565 		memcpy(here->e_name, name, name_len);
566 	} else {
567 		if (here->e_value_size) {
568 			char *first_val = (char *)header + min_offs;
569 			size_t offs = le16_to_cpu(here->e_value_offs);
570 			char *val = (char *)header + offs;
571 			size_t size = EXT2_XATTR_SIZE(
572 				le32_to_cpu(here->e_value_size));
573 
574 			if (size == EXT2_XATTR_SIZE(value_len)) {
575 				/* The old and the new value have the same
576 				   size. Just replace. */
577 				here->e_value_size = cpu_to_le32(value_len);
578 				memset(val + size - EXT2_XATTR_PAD, 0,
579 				       EXT2_XATTR_PAD); /* Clear pad bytes. */
580 				memcpy(val, value, value_len);
581 				goto skip_replace;
582 			}
583 
584 			/* Remove the old value. */
585 			memmove(first_val + size, first_val, val - first_val);
586 			memset(first_val, 0, size);
587 			here->e_value_offs = 0;
588 			min_offs += size;
589 
590 			/* Adjust all value offsets. */
591 			last = ENTRY(header+1);
592 			while (!IS_LAST_ENTRY(last)) {
593 				size_t o = le16_to_cpu(last->e_value_offs);
594 				if (o < offs)
595 					last->e_value_offs =
596 						cpu_to_le16(o + size);
597 				last = EXT2_XATTR_NEXT(last);
598 			}
599 		}
600 		if (value == NULL) {
601 			/* Remove the old name. */
602 			size_t size = EXT2_XATTR_LEN(name_len);
603 			last = ENTRY((char *)last - size);
604 			memmove(here, (char*)here + size,
605 				(char*)last - (char*)here);
606 			memset(last, 0, size);
607 		}
608 	}
609 
610 	if (value != NULL) {
611 		/* Insert the new value. */
612 		here->e_value_size = cpu_to_le32(value_len);
613 		if (value_len) {
614 			size_t size = EXT2_XATTR_SIZE(value_len);
615 			char *val = (char *)header + min_offs - size;
616 			here->e_value_offs =
617 				cpu_to_le16((char *)val - (char *)header);
618 			memset(val + size - EXT2_XATTR_PAD, 0,
619 			       EXT2_XATTR_PAD); /* Clear the pad bytes. */
620 			memcpy(val, value, value_len);
621 		}
622 	}
623 
624 skip_replace:
625 	if (IS_LAST_ENTRY(ENTRY(header+1))) {
626 		/* This block is now empty. */
627 		if (bh && header == HDR(bh))
628 			unlock_buffer(bh);  /* we were modifying in-place. */
629 		error = ext2_xattr_set2(inode, bh, NULL);
630 	} else {
631 		ext2_xattr_rehash(header, here);
632 		if (bh && header == HDR(bh))
633 			unlock_buffer(bh);  /* we were modifying in-place. */
634 		error = ext2_xattr_set2(inode, bh, header);
635 	}
636 
637 cleanup:
638 	if (!(bh && header == HDR(bh)))
639 		kfree(header);
640 	brelse(bh);
641 	up_write(&EXT2_I(inode)->xattr_sem);
642 
643 	return error;
644 }
645 
646 /*
647  * Second half of ext2_xattr_set(): Update the file system.
648  */
649 static int
650 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
651 		struct ext2_xattr_header *header)
652 {
653 	struct super_block *sb = inode->i_sb;
654 	struct buffer_head *new_bh = NULL;
655 	int error;
656 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
657 
658 	if (header) {
659 		new_bh = ext2_xattr_cache_find(inode, header);
660 		if (new_bh) {
661 			/* We found an identical block in the cache. */
662 			if (new_bh == old_bh) {
663 				ea_bdebug(new_bh, "keeping this block");
664 			} else {
665 				/* The old block is released after updating
666 				   the inode.  */
667 				ea_bdebug(new_bh, "reusing block");
668 
669 				error = dquot_alloc_block(inode, 1);
670 				if (error) {
671 					unlock_buffer(new_bh);
672 					goto cleanup;
673 				}
674 				le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
675 				ea_bdebug(new_bh, "refcount now=%d",
676 					le32_to_cpu(HDR(new_bh)->h_refcount));
677 			}
678 			unlock_buffer(new_bh);
679 		} else if (old_bh && header == HDR(old_bh)) {
680 			/* Keep this block. No need to lock the block as we
681 			   don't need to change the reference count. */
682 			new_bh = old_bh;
683 			get_bh(new_bh);
684 			ext2_xattr_cache_insert(ea_block_cache, new_bh);
685 		} else {
686 			/* We need to allocate a new block */
687 			ext2_fsblk_t goal = ext2_group_first_block_no(sb,
688 						EXT2_I(inode)->i_block_group);
689 			int block = ext2_new_block(inode, goal, &error);
690 			if (error)
691 				goto cleanup;
692 			ea_idebug(inode, "creating block %d", block);
693 
694 			new_bh = sb_getblk(sb, block);
695 			if (unlikely(!new_bh)) {
696 				ext2_free_blocks(inode, block, 1);
697 				mark_inode_dirty(inode);
698 				error = -ENOMEM;
699 				goto cleanup;
700 			}
701 			lock_buffer(new_bh);
702 			memcpy(new_bh->b_data, header, new_bh->b_size);
703 			set_buffer_uptodate(new_bh);
704 			unlock_buffer(new_bh);
705 			ext2_xattr_cache_insert(ea_block_cache, new_bh);
706 
707 			ext2_xattr_update_super_block(sb);
708 		}
709 		mark_buffer_dirty(new_bh);
710 		if (IS_SYNC(inode)) {
711 			sync_dirty_buffer(new_bh);
712 			error = -EIO;
713 			if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
714 				goto cleanup;
715 		}
716 	}
717 
718 	/* Update the inode. */
719 	EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
720 	inode->i_ctime = current_time(inode);
721 	if (IS_SYNC(inode)) {
722 		error = sync_inode_metadata(inode, 1);
723 		/* In case sync failed due to ENOSPC the inode was actually
724 		 * written (only some dirty data were not) so we just proceed
725 		 * as if nothing happened and cleanup the unused block */
726 		if (error && error != -ENOSPC) {
727 			if (new_bh && new_bh != old_bh) {
728 				dquot_free_block_nodirty(inode, 1);
729 				mark_inode_dirty(inode);
730 			}
731 			goto cleanup;
732 		}
733 	} else
734 		mark_inode_dirty(inode);
735 
736 	error = 0;
737 	if (old_bh && old_bh != new_bh) {
738 		/*
739 		 * If there was an old block and we are no longer using it,
740 		 * release the old block.
741 		 */
742 		lock_buffer(old_bh);
743 		if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
744 			__u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
745 
746 			/*
747 			 * This must happen under buffer lock for
748 			 * ext2_xattr_set2() to reliably detect freed block
749 			 */
750 			mb_cache_entry_delete(ea_block_cache, hash,
751 					      old_bh->b_blocknr);
752 			/* Free the old block. */
753 			ea_bdebug(old_bh, "freeing");
754 			ext2_free_blocks(inode, old_bh->b_blocknr, 1);
755 			mark_inode_dirty(inode);
756 			/* We let our caller release old_bh, so we
757 			 * need to duplicate the buffer before. */
758 			get_bh(old_bh);
759 			bforget(old_bh);
760 		} else {
761 			/* Decrement the refcount only. */
762 			le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
763 			dquot_free_block_nodirty(inode, 1);
764 			mark_inode_dirty(inode);
765 			mark_buffer_dirty(old_bh);
766 			ea_bdebug(old_bh, "refcount now=%d",
767 				le32_to_cpu(HDR(old_bh)->h_refcount));
768 		}
769 		unlock_buffer(old_bh);
770 	}
771 
772 cleanup:
773 	brelse(new_bh);
774 
775 	return error;
776 }
777 
778 /*
779  * ext2_xattr_delete_inode()
780  *
781  * Free extended attribute resources associated with this inode. This
782  * is called immediately before an inode is freed.
783  */
784 void
785 ext2_xattr_delete_inode(struct inode *inode)
786 {
787 	struct buffer_head *bh = NULL;
788 	struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
789 
790 	down_write(&EXT2_I(inode)->xattr_sem);
791 	if (!EXT2_I(inode)->i_file_acl)
792 		goto cleanup;
793 
794 	if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 0)) {
795 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
796 			"inode %ld: xattr block %d is out of data blocks range",
797 			inode->i_ino, EXT2_I(inode)->i_file_acl);
798 		goto cleanup;
799 	}
800 
801 	bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
802 	if (!bh) {
803 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
804 			"inode %ld: block %d read error", inode->i_ino,
805 			EXT2_I(inode)->i_file_acl);
806 		goto cleanup;
807 	}
808 	ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
809 	if (!ext2_xattr_header_valid(HDR(bh))) {
810 		ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
811 			"inode %ld: bad block %d", inode->i_ino,
812 			EXT2_I(inode)->i_file_acl);
813 		goto cleanup;
814 	}
815 	lock_buffer(bh);
816 	if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
817 		__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
818 
819 		/*
820 		 * This must happen under buffer lock for ext2_xattr_set2() to
821 		 * reliably detect freed block
822 		 */
823 		mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
824 				      bh->b_blocknr);
825 		ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
826 		get_bh(bh);
827 		bforget(bh);
828 		unlock_buffer(bh);
829 	} else {
830 		le32_add_cpu(&HDR(bh)->h_refcount, -1);
831 		ea_bdebug(bh, "refcount now=%d",
832 			le32_to_cpu(HDR(bh)->h_refcount));
833 		unlock_buffer(bh);
834 		mark_buffer_dirty(bh);
835 		if (IS_SYNC(inode))
836 			sync_dirty_buffer(bh);
837 		dquot_free_block_nodirty(inode, 1);
838 	}
839 	EXT2_I(inode)->i_file_acl = 0;
840 
841 cleanup:
842 	brelse(bh);
843 	up_write(&EXT2_I(inode)->xattr_sem);
844 }
845 
846 /*
847  * ext2_xattr_cache_insert()
848  *
849  * Create a new entry in the extended attribute cache, and insert
850  * it unless such an entry is already in the cache.
851  *
852  * Returns 0, or a negative error number on failure.
853  */
854 static int
855 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
856 {
857 	__u32 hash = le32_to_cpu(HDR(bh)->h_hash);
858 	int error;
859 
860 	error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr,
861 				      true);
862 	if (error) {
863 		if (error == -EBUSY) {
864 			ea_bdebug(bh, "already in cache (%d cache entries)",
865 				atomic_read(&ext2_xattr_cache->c_entry_count));
866 			error = 0;
867 		}
868 	} else
869 		ea_bdebug(bh, "inserting [%x]", (int)hash);
870 	return error;
871 }
872 
873 /*
874  * ext2_xattr_cmp()
875  *
876  * Compare two extended attribute blocks for equality.
877  *
878  * Returns 0 if the blocks are equal, 1 if they differ, and
879  * a negative error number on errors.
880  */
881 static int
882 ext2_xattr_cmp(struct ext2_xattr_header *header1,
883 	       struct ext2_xattr_header *header2)
884 {
885 	struct ext2_xattr_entry *entry1, *entry2;
886 
887 	entry1 = ENTRY(header1+1);
888 	entry2 = ENTRY(header2+1);
889 	while (!IS_LAST_ENTRY(entry1)) {
890 		if (IS_LAST_ENTRY(entry2))
891 			return 1;
892 		if (entry1->e_hash != entry2->e_hash ||
893 		    entry1->e_name_index != entry2->e_name_index ||
894 		    entry1->e_name_len != entry2->e_name_len ||
895 		    entry1->e_value_size != entry2->e_value_size ||
896 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
897 			return 1;
898 		if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
899 			return -EIO;
900 		if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
901 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
902 			   le32_to_cpu(entry1->e_value_size)))
903 			return 1;
904 
905 		entry1 = EXT2_XATTR_NEXT(entry1);
906 		entry2 = EXT2_XATTR_NEXT(entry2);
907 	}
908 	if (!IS_LAST_ENTRY(entry2))
909 		return 1;
910 	return 0;
911 }
912 
913 /*
914  * ext2_xattr_cache_find()
915  *
916  * Find an identical extended attribute block.
917  *
918  * Returns a locked buffer head to the block found, or NULL if such
919  * a block was not found or an error occurred.
920  */
921 static struct buffer_head *
922 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
923 {
924 	__u32 hash = le32_to_cpu(header->h_hash);
925 	struct mb_cache_entry *ce;
926 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
927 
928 	if (!header->h_hash)
929 		return NULL;  /* never share */
930 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
931 again:
932 	ce = mb_cache_entry_find_first(ea_block_cache, hash);
933 	while (ce) {
934 		struct buffer_head *bh;
935 
936 		bh = sb_bread(inode->i_sb, ce->e_value);
937 		if (!bh) {
938 			ext2_error(inode->i_sb, "ext2_xattr_cache_find",
939 				"inode %ld: block %ld read error",
940 				inode->i_ino, (unsigned long) ce->e_value);
941 		} else {
942 			lock_buffer(bh);
943 			/*
944 			 * We have to be careful about races with freeing or
945 			 * rehashing of xattr block. Once we hold buffer lock
946 			 * xattr block's state is stable so we can check
947 			 * whether the block got freed / rehashed or not.
948 			 * Since we unhash mbcache entry under buffer lock when
949 			 * freeing / rehashing xattr block, checking whether
950 			 * entry is still hashed is reliable.
951 			 */
952 			if (hlist_bl_unhashed(&ce->e_hash_list)) {
953 				mb_cache_entry_put(ea_block_cache, ce);
954 				unlock_buffer(bh);
955 				brelse(bh);
956 				goto again;
957 			} else if (le32_to_cpu(HDR(bh)->h_refcount) >
958 				   EXT2_XATTR_REFCOUNT_MAX) {
959 				ea_idebug(inode, "block %ld refcount %d>%d",
960 					  (unsigned long) ce->e_value,
961 					  le32_to_cpu(HDR(bh)->h_refcount),
962 					  EXT2_XATTR_REFCOUNT_MAX);
963 			} else if (!ext2_xattr_cmp(header, HDR(bh))) {
964 				ea_bdebug(bh, "b_count=%d",
965 					  atomic_read(&(bh->b_count)));
966 				mb_cache_entry_touch(ea_block_cache, ce);
967 				mb_cache_entry_put(ea_block_cache, ce);
968 				return bh;
969 			}
970 			unlock_buffer(bh);
971 			brelse(bh);
972 		}
973 		ce = mb_cache_entry_find_next(ea_block_cache, ce);
974 	}
975 	return NULL;
976 }
977 
978 #define NAME_HASH_SHIFT 5
979 #define VALUE_HASH_SHIFT 16
980 
981 /*
982  * ext2_xattr_hash_entry()
983  *
984  * Compute the hash of an extended attribute.
985  */
986 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
987 					 struct ext2_xattr_entry *entry)
988 {
989 	__u32 hash = 0;
990 	char *name = entry->e_name;
991 	int n;
992 
993 	for (n=0; n < entry->e_name_len; n++) {
994 		hash = (hash << NAME_HASH_SHIFT) ^
995 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
996 		       *name++;
997 	}
998 
999 	if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1000 		__le32 *value = (__le32 *)((char *)header +
1001 			le16_to_cpu(entry->e_value_offs));
1002 		for (n = (le32_to_cpu(entry->e_value_size) +
1003 		     EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
1004 			hash = (hash << VALUE_HASH_SHIFT) ^
1005 			       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1006 			       le32_to_cpu(*value++);
1007 		}
1008 	}
1009 	entry->e_hash = cpu_to_le32(hash);
1010 }
1011 
1012 #undef NAME_HASH_SHIFT
1013 #undef VALUE_HASH_SHIFT
1014 
1015 #define BLOCK_HASH_SHIFT 16
1016 
1017 /*
1018  * ext2_xattr_rehash()
1019  *
1020  * Re-compute the extended attribute hash value after an entry has changed.
1021  */
1022 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1023 			      struct ext2_xattr_entry *entry)
1024 {
1025 	struct ext2_xattr_entry *here;
1026 	__u32 hash = 0;
1027 
1028 	ext2_xattr_hash_entry(header, entry);
1029 	here = ENTRY(header+1);
1030 	while (!IS_LAST_ENTRY(here)) {
1031 		if (!here->e_hash) {
1032 			/* Block is not shared if an entry's hash value == 0 */
1033 			hash = 0;
1034 			break;
1035 		}
1036 		hash = (hash << BLOCK_HASH_SHIFT) ^
1037 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1038 		       le32_to_cpu(here->e_hash);
1039 		here = EXT2_XATTR_NEXT(here);
1040 	}
1041 	header->h_hash = cpu_to_le32(hash);
1042 }
1043 
1044 #undef BLOCK_HASH_SHIFT
1045 
1046 #define HASH_BUCKET_BITS 10
1047 
1048 struct mb_cache *ext2_xattr_create_cache(void)
1049 {
1050 	return mb_cache_create(HASH_BUCKET_BITS);
1051 }
1052 
1053 void ext2_xattr_destroy_cache(struct mb_cache *cache)
1054 {
1055 	if (cache)
1056 		mb_cache_destroy(cache);
1057 }
1058