xref: /openbmc/linux/fs/ext4/xattr.c (revision c8dbaa22)
1 /*
2  * linux/fs/ext4/xattr.c
3  *
4  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
5  *
6  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
7  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
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  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
13  *  and Andreas Gruenbacher <agruen@suse.de>.
14  */
15 
16 /*
17  * Extended attributes are stored directly in inodes (on file systems with
18  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
19  * field contains the block number if an inode uses an additional block. All
20  * attributes must fit in the inode and one additional block. Blocks that
21  * contain the identical set of attributes may be shared among several inodes.
22  * Identical blocks are detected by keeping a cache of blocks that have
23  * recently been accessed.
24  *
25  * The attributes in inodes and on blocks have a different header; the entries
26  * are stored in the same format:
27  *
28  *   +------------------+
29  *   | header           |
30  *   | entry 1          | |
31  *   | entry 2          | | growing downwards
32  *   | entry 3          | v
33  *   | four null bytes  |
34  *   | . . .            |
35  *   | value 1          | ^
36  *   | value 3          | | growing upwards
37  *   | value 2          | |
38  *   +------------------+
39  *
40  * The header is followed by multiple entry descriptors. In disk blocks, the
41  * entry descriptors are kept sorted. In inodes, they are unsorted. The
42  * attribute values are aligned to the end of the block in no specific order.
43  *
44  * Locking strategy
45  * ----------------
46  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
47  * EA blocks are only changed if they are exclusive to an inode, so
48  * holding xattr_sem also means that nothing but the EA block's reference
49  * count can change. Multiple writers to the same block are synchronized
50  * by the buffer lock.
51  */
52 
53 #include <linux/init.h>
54 #include <linux/fs.h>
55 #include <linux/slab.h>
56 #include <linux/mbcache.h>
57 #include <linux/quotaops.h>
58 #include "ext4_jbd2.h"
59 #include "ext4.h"
60 #include "xattr.h"
61 #include "acl.h"
62 
63 #ifdef EXT4_XATTR_DEBUG
64 # define ea_idebug(inode, fmt, ...)					\
65 	printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",			\
66 	       inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
67 # define ea_bdebug(bh, fmt, ...)					\
68 	printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",			\
69 	       bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
70 #else
71 # define ea_idebug(inode, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
72 # define ea_bdebug(bh, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
73 #endif
74 
75 static void ext4_xattr_block_cache_insert(struct mb_cache *,
76 					  struct buffer_head *);
77 static struct buffer_head *
78 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
79 			    struct mb_cache_entry **);
80 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
81 				    size_t value_count);
82 static void ext4_xattr_rehash(struct ext4_xattr_header *);
83 
84 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
85 	[EXT4_XATTR_INDEX_USER]		     = &ext4_xattr_user_handler,
86 #ifdef CONFIG_EXT4_FS_POSIX_ACL
87 	[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
88 	[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
89 #endif
90 	[EXT4_XATTR_INDEX_TRUSTED]	     = &ext4_xattr_trusted_handler,
91 #ifdef CONFIG_EXT4_FS_SECURITY
92 	[EXT4_XATTR_INDEX_SECURITY]	     = &ext4_xattr_security_handler,
93 #endif
94 };
95 
96 const struct xattr_handler *ext4_xattr_handlers[] = {
97 	&ext4_xattr_user_handler,
98 	&ext4_xattr_trusted_handler,
99 #ifdef CONFIG_EXT4_FS_POSIX_ACL
100 	&posix_acl_access_xattr_handler,
101 	&posix_acl_default_xattr_handler,
102 #endif
103 #ifdef CONFIG_EXT4_FS_SECURITY
104 	&ext4_xattr_security_handler,
105 #endif
106 	NULL
107 };
108 
109 #define EA_BLOCK_CACHE(inode)	(((struct ext4_sb_info *) \
110 				inode->i_sb->s_fs_info)->s_ea_block_cache)
111 
112 #define EA_INODE_CACHE(inode)	(((struct ext4_sb_info *) \
113 				inode->i_sb->s_fs_info)->s_ea_inode_cache)
114 
115 static int
116 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
117 			struct inode *inode);
118 
119 #ifdef CONFIG_LOCKDEP
120 void ext4_xattr_inode_set_class(struct inode *ea_inode)
121 {
122 	lockdep_set_subclass(&ea_inode->i_rwsem, 1);
123 }
124 #endif
125 
126 static __le32 ext4_xattr_block_csum(struct inode *inode,
127 				    sector_t block_nr,
128 				    struct ext4_xattr_header *hdr)
129 {
130 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
131 	__u32 csum;
132 	__le64 dsk_block_nr = cpu_to_le64(block_nr);
133 	__u32 dummy_csum = 0;
134 	int offset = offsetof(struct ext4_xattr_header, h_checksum);
135 
136 	csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
137 			   sizeof(dsk_block_nr));
138 	csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
139 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
140 	offset += sizeof(dummy_csum);
141 	csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
142 			   EXT4_BLOCK_SIZE(inode->i_sb) - offset);
143 
144 	return cpu_to_le32(csum);
145 }
146 
147 static int ext4_xattr_block_csum_verify(struct inode *inode,
148 					struct buffer_head *bh)
149 {
150 	struct ext4_xattr_header *hdr = BHDR(bh);
151 	int ret = 1;
152 
153 	if (ext4_has_metadata_csum(inode->i_sb)) {
154 		lock_buffer(bh);
155 		ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
156 							bh->b_blocknr, hdr));
157 		unlock_buffer(bh);
158 	}
159 	return ret;
160 }
161 
162 static void ext4_xattr_block_csum_set(struct inode *inode,
163 				      struct buffer_head *bh)
164 {
165 	if (ext4_has_metadata_csum(inode->i_sb))
166 		BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
167 						bh->b_blocknr, BHDR(bh));
168 }
169 
170 static inline const struct xattr_handler *
171 ext4_xattr_handler(int name_index)
172 {
173 	const struct xattr_handler *handler = NULL;
174 
175 	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
176 		handler = ext4_xattr_handler_map[name_index];
177 	return handler;
178 }
179 
180 static int
181 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
182 			 void *value_start)
183 {
184 	struct ext4_xattr_entry *e = entry;
185 
186 	/* Find the end of the names list */
187 	while (!IS_LAST_ENTRY(e)) {
188 		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
189 		if ((void *)next >= end)
190 			return -EFSCORRUPTED;
191 		e = next;
192 	}
193 
194 	/* Check the values */
195 	while (!IS_LAST_ENTRY(entry)) {
196 		if (entry->e_value_size != 0 &&
197 		    entry->e_value_inum == 0) {
198 			u16 offs = le16_to_cpu(entry->e_value_offs);
199 			u32 size = le32_to_cpu(entry->e_value_size);
200 			void *value;
201 
202 			/*
203 			 * The value cannot overlap the names, and the value
204 			 * with padding cannot extend beyond 'end'.  Check both
205 			 * the padded and unpadded sizes, since the size may
206 			 * overflow to 0 when adding padding.
207 			 */
208 			if (offs > end - value_start)
209 				return -EFSCORRUPTED;
210 			value = value_start + offs;
211 			if (value < (void *)e + sizeof(u32) ||
212 			    size > end - value ||
213 			    EXT4_XATTR_SIZE(size) > end - value)
214 				return -EFSCORRUPTED;
215 		}
216 		entry = EXT4_XATTR_NEXT(entry);
217 	}
218 
219 	return 0;
220 }
221 
222 static inline int
223 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
224 {
225 	int error;
226 
227 	if (buffer_verified(bh))
228 		return 0;
229 
230 	if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
231 	    BHDR(bh)->h_blocks != cpu_to_le32(1))
232 		return -EFSCORRUPTED;
233 	if (!ext4_xattr_block_csum_verify(inode, bh))
234 		return -EFSBADCRC;
235 	error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
236 					 bh->b_data);
237 	if (!error)
238 		set_buffer_verified(bh);
239 	return error;
240 }
241 
242 static int
243 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
244 			 void *end, const char *function, unsigned int line)
245 {
246 	int error = -EFSCORRUPTED;
247 
248 	if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
249 	    (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
250 		goto errout;
251 	error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
252 errout:
253 	if (error)
254 		__ext4_error_inode(inode, function, line, 0,
255 				   "corrupted in-inode xattr");
256 	return error;
257 }
258 
259 #define xattr_check_inode(inode, header, end) \
260 	__xattr_check_inode((inode), (header), (end), __func__, __LINE__)
261 
262 static int
263 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
264 		      const char *name, int sorted)
265 {
266 	struct ext4_xattr_entry *entry;
267 	size_t name_len;
268 	int cmp = 1;
269 
270 	if (name == NULL)
271 		return -EINVAL;
272 	name_len = strlen(name);
273 	entry = *pentry;
274 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
275 		cmp = name_index - entry->e_name_index;
276 		if (!cmp)
277 			cmp = name_len - entry->e_name_len;
278 		if (!cmp)
279 			cmp = memcmp(name, entry->e_name, name_len);
280 		if (cmp <= 0 && (sorted || cmp == 0))
281 			break;
282 	}
283 	*pentry = entry;
284 	return cmp ? -ENODATA : 0;
285 }
286 
287 static u32
288 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
289 {
290 	return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
291 }
292 
293 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
294 {
295 	return ((u64)ea_inode->i_ctime.tv_sec << 32) |
296 	       ((u32)ea_inode->i_version);
297 }
298 
299 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
300 {
301 	ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
302 	ea_inode->i_version = (u32)ref_count;
303 }
304 
305 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
306 {
307 	return (u32)ea_inode->i_atime.tv_sec;
308 }
309 
310 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
311 {
312 	ea_inode->i_atime.tv_sec = hash;
313 }
314 
315 /*
316  * Read the EA value from an inode.
317  */
318 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
319 {
320 	int blocksize = 1 << ea_inode->i_blkbits;
321 	int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
322 	int tail_size = (size % blocksize) ?: blocksize;
323 	struct buffer_head *bhs_inline[8];
324 	struct buffer_head **bhs = bhs_inline;
325 	int i, ret;
326 
327 	if (bh_count > ARRAY_SIZE(bhs_inline)) {
328 		bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
329 		if (!bhs)
330 			return -ENOMEM;
331 	}
332 
333 	ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
334 			       true /* wait */, bhs);
335 	if (ret)
336 		goto free_bhs;
337 
338 	for (i = 0; i < bh_count; i++) {
339 		/* There shouldn't be any holes in ea_inode. */
340 		if (!bhs[i]) {
341 			ret = -EFSCORRUPTED;
342 			goto put_bhs;
343 		}
344 		memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
345 		       i < bh_count - 1 ? blocksize : tail_size);
346 	}
347 	ret = 0;
348 put_bhs:
349 	for (i = 0; i < bh_count; i++)
350 		brelse(bhs[i]);
351 free_bhs:
352 	if (bhs != bhs_inline)
353 		kfree(bhs);
354 	return ret;
355 }
356 
357 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
358 				 struct inode **ea_inode)
359 {
360 	struct inode *inode;
361 	int err;
362 
363 	inode = ext4_iget(parent->i_sb, ea_ino);
364 	if (IS_ERR(inode)) {
365 		err = PTR_ERR(inode);
366 		ext4_error(parent->i_sb,
367 			   "error while reading EA inode %lu err=%d", ea_ino,
368 			   err);
369 		return err;
370 	}
371 
372 	if (is_bad_inode(inode)) {
373 		ext4_error(parent->i_sb,
374 			   "error while reading EA inode %lu is_bad_inode",
375 			   ea_ino);
376 		err = -EIO;
377 		goto error;
378 	}
379 
380 	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
381 		ext4_error(parent->i_sb,
382 			   "EA inode %lu does not have EXT4_EA_INODE_FL flag",
383 			    ea_ino);
384 		err = -EINVAL;
385 		goto error;
386 	}
387 
388 	*ea_inode = inode;
389 	return 0;
390 error:
391 	iput(inode);
392 	return err;
393 }
394 
395 static int
396 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
397 			       struct ext4_xattr_entry *entry, void *buffer,
398 			       size_t size)
399 {
400 	u32 hash;
401 
402 	/* Verify stored hash matches calculated hash. */
403 	hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
404 	if (hash != ext4_xattr_inode_get_hash(ea_inode))
405 		return -EFSCORRUPTED;
406 
407 	if (entry) {
408 		__le32 e_hash, tmp_data;
409 
410 		/* Verify entry hash. */
411 		tmp_data = cpu_to_le32(hash);
412 		e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
413 					       &tmp_data, 1);
414 		if (e_hash != entry->e_hash)
415 			return -EFSCORRUPTED;
416 	}
417 	return 0;
418 }
419 
420 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
421 
422 /*
423  * Read xattr value from the EA inode.
424  */
425 static int
426 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
427 		     void *buffer, size_t size)
428 {
429 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
430 	struct inode *ea_inode;
431 	int err;
432 
433 	err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
434 				    &ea_inode);
435 	if (err) {
436 		ea_inode = NULL;
437 		goto out;
438 	}
439 
440 	if (i_size_read(ea_inode) != size) {
441 		ext4_warning_inode(ea_inode,
442 				   "ea_inode file size=%llu entry size=%zu",
443 				   i_size_read(ea_inode), size);
444 		err = -EFSCORRUPTED;
445 		goto out;
446 	}
447 
448 	err = ext4_xattr_inode_read(ea_inode, buffer, size);
449 	if (err)
450 		goto out;
451 
452 	err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer, size);
453 	/*
454 	 * Compatibility check for old Lustre ea_inode implementation. Old
455 	 * version does not have hash validation, but it has a backpointer
456 	 * from ea_inode to the parent inode.
457 	 */
458 	if (err == -EFSCORRUPTED) {
459 		if (EXT4_XATTR_INODE_GET_PARENT(ea_inode) != inode->i_ino ||
460 		    ea_inode->i_generation != inode->i_generation) {
461 			ext4_warning_inode(ea_inode,
462 					   "EA inode hash validation failed");
463 			goto out;
464 		}
465 		/* Do not add ea_inode to the cache. */
466 		ea_inode_cache = NULL;
467 		err = 0;
468 	} else if (err)
469 		goto out;
470 
471 	if (ea_inode_cache)
472 		mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
473 				      ext4_xattr_inode_get_hash(ea_inode),
474 				      ea_inode->i_ino, true /* reusable */);
475 out:
476 	iput(ea_inode);
477 	return err;
478 }
479 
480 static int
481 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
482 		     void *buffer, size_t buffer_size)
483 {
484 	struct buffer_head *bh = NULL;
485 	struct ext4_xattr_entry *entry;
486 	size_t size;
487 	int error;
488 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
489 
490 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
491 		  name_index, name, buffer, (long)buffer_size);
492 
493 	error = -ENODATA;
494 	if (!EXT4_I(inode)->i_file_acl)
495 		goto cleanup;
496 	ea_idebug(inode, "reading block %llu",
497 		  (unsigned long long)EXT4_I(inode)->i_file_acl);
498 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
499 	if (!bh)
500 		goto cleanup;
501 	ea_bdebug(bh, "b_count=%d, refcount=%d",
502 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
503 	if (ext4_xattr_check_block(inode, bh)) {
504 		EXT4_ERROR_INODE(inode, "bad block %llu",
505 				 EXT4_I(inode)->i_file_acl);
506 		error = -EFSCORRUPTED;
507 		goto cleanup;
508 	}
509 	ext4_xattr_block_cache_insert(ea_block_cache, bh);
510 	entry = BFIRST(bh);
511 	error = ext4_xattr_find_entry(&entry, name_index, name, 1);
512 	if (error)
513 		goto cleanup;
514 	size = le32_to_cpu(entry->e_value_size);
515 	if (buffer) {
516 		error = -ERANGE;
517 		if (size > buffer_size)
518 			goto cleanup;
519 		if (entry->e_value_inum) {
520 			error = ext4_xattr_inode_get(inode, entry, buffer,
521 						     size);
522 			if (error)
523 				goto cleanup;
524 		} else {
525 			memcpy(buffer, bh->b_data +
526 			       le16_to_cpu(entry->e_value_offs), size);
527 		}
528 	}
529 	error = size;
530 
531 cleanup:
532 	brelse(bh);
533 	return error;
534 }
535 
536 int
537 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
538 		     void *buffer, size_t buffer_size)
539 {
540 	struct ext4_xattr_ibody_header *header;
541 	struct ext4_xattr_entry *entry;
542 	struct ext4_inode *raw_inode;
543 	struct ext4_iloc iloc;
544 	size_t size;
545 	void *end;
546 	int error;
547 
548 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
549 		return -ENODATA;
550 	error = ext4_get_inode_loc(inode, &iloc);
551 	if (error)
552 		return error;
553 	raw_inode = ext4_raw_inode(&iloc);
554 	header = IHDR(inode, raw_inode);
555 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
556 	error = xattr_check_inode(inode, header, end);
557 	if (error)
558 		goto cleanup;
559 	entry = IFIRST(header);
560 	error = ext4_xattr_find_entry(&entry, name_index, name, 0);
561 	if (error)
562 		goto cleanup;
563 	size = le32_to_cpu(entry->e_value_size);
564 	if (buffer) {
565 		error = -ERANGE;
566 		if (size > buffer_size)
567 			goto cleanup;
568 		if (entry->e_value_inum) {
569 			error = ext4_xattr_inode_get(inode, entry, buffer,
570 						     size);
571 			if (error)
572 				goto cleanup;
573 		} else {
574 			memcpy(buffer, (void *)IFIRST(header) +
575 			       le16_to_cpu(entry->e_value_offs), size);
576 		}
577 	}
578 	error = size;
579 
580 cleanup:
581 	brelse(iloc.bh);
582 	return error;
583 }
584 
585 /*
586  * ext4_xattr_get()
587  *
588  * Copy an extended attribute into the buffer
589  * provided, or compute the buffer size required.
590  * Buffer is NULL to compute the size of the buffer required.
591  *
592  * Returns a negative error number on failure, or the number of bytes
593  * used / required on success.
594  */
595 int
596 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
597 	       void *buffer, size_t buffer_size)
598 {
599 	int error;
600 
601 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
602 		return -EIO;
603 
604 	if (strlen(name) > 255)
605 		return -ERANGE;
606 
607 	down_read(&EXT4_I(inode)->xattr_sem);
608 	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
609 				     buffer_size);
610 	if (error == -ENODATA)
611 		error = ext4_xattr_block_get(inode, name_index, name, buffer,
612 					     buffer_size);
613 	up_read(&EXT4_I(inode)->xattr_sem);
614 	return error;
615 }
616 
617 static int
618 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
619 			char *buffer, size_t buffer_size)
620 {
621 	size_t rest = buffer_size;
622 
623 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
624 		const struct xattr_handler *handler =
625 			ext4_xattr_handler(entry->e_name_index);
626 
627 		if (handler && (!handler->list || handler->list(dentry))) {
628 			const char *prefix = handler->prefix ?: handler->name;
629 			size_t prefix_len = strlen(prefix);
630 			size_t size = prefix_len + entry->e_name_len + 1;
631 
632 			if (buffer) {
633 				if (size > rest)
634 					return -ERANGE;
635 				memcpy(buffer, prefix, prefix_len);
636 				buffer += prefix_len;
637 				memcpy(buffer, entry->e_name, entry->e_name_len);
638 				buffer += entry->e_name_len;
639 				*buffer++ = 0;
640 			}
641 			rest -= size;
642 		}
643 	}
644 	return buffer_size - rest;  /* total size */
645 }
646 
647 static int
648 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
649 {
650 	struct inode *inode = d_inode(dentry);
651 	struct buffer_head *bh = NULL;
652 	int error;
653 
654 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
655 		  buffer, (long)buffer_size);
656 
657 	error = 0;
658 	if (!EXT4_I(inode)->i_file_acl)
659 		goto cleanup;
660 	ea_idebug(inode, "reading block %llu",
661 		  (unsigned long long)EXT4_I(inode)->i_file_acl);
662 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
663 	error = -EIO;
664 	if (!bh)
665 		goto cleanup;
666 	ea_bdebug(bh, "b_count=%d, refcount=%d",
667 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
668 	if (ext4_xattr_check_block(inode, bh)) {
669 		EXT4_ERROR_INODE(inode, "bad block %llu",
670 				 EXT4_I(inode)->i_file_acl);
671 		error = -EFSCORRUPTED;
672 		goto cleanup;
673 	}
674 	ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
675 	error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
676 
677 cleanup:
678 	brelse(bh);
679 
680 	return error;
681 }
682 
683 static int
684 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
685 {
686 	struct inode *inode = d_inode(dentry);
687 	struct ext4_xattr_ibody_header *header;
688 	struct ext4_inode *raw_inode;
689 	struct ext4_iloc iloc;
690 	void *end;
691 	int error;
692 
693 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
694 		return 0;
695 	error = ext4_get_inode_loc(inode, &iloc);
696 	if (error)
697 		return error;
698 	raw_inode = ext4_raw_inode(&iloc);
699 	header = IHDR(inode, raw_inode);
700 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
701 	error = xattr_check_inode(inode, header, end);
702 	if (error)
703 		goto cleanup;
704 	error = ext4_xattr_list_entries(dentry, IFIRST(header),
705 					buffer, buffer_size);
706 
707 cleanup:
708 	brelse(iloc.bh);
709 	return error;
710 }
711 
712 /*
713  * Inode operation listxattr()
714  *
715  * d_inode(dentry)->i_rwsem: don't care
716  *
717  * Copy a list of attribute names into the buffer
718  * provided, or compute the buffer size required.
719  * Buffer is NULL to compute the size of the buffer required.
720  *
721  * Returns a negative error number on failure, or the number of bytes
722  * used / required on success.
723  */
724 ssize_t
725 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
726 {
727 	int ret, ret2;
728 
729 	down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
730 	ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
731 	if (ret < 0)
732 		goto errout;
733 	if (buffer) {
734 		buffer += ret;
735 		buffer_size -= ret;
736 	}
737 	ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
738 	if (ret < 0)
739 		goto errout;
740 	ret += ret2;
741 errout:
742 	up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
743 	return ret;
744 }
745 
746 /*
747  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
748  * not set, set it.
749  */
750 static void ext4_xattr_update_super_block(handle_t *handle,
751 					  struct super_block *sb)
752 {
753 	if (ext4_has_feature_xattr(sb))
754 		return;
755 
756 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
757 	if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
758 		ext4_set_feature_xattr(sb);
759 		ext4_handle_dirty_super(handle, sb);
760 	}
761 }
762 
763 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
764 {
765 	struct ext4_iloc iloc = { .bh = NULL };
766 	struct buffer_head *bh = NULL;
767 	struct ext4_inode *raw_inode;
768 	struct ext4_xattr_ibody_header *header;
769 	struct ext4_xattr_entry *entry;
770 	qsize_t ea_inode_refs = 0;
771 	void *end;
772 	int ret;
773 
774 	lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
775 
776 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
777 		ret = ext4_get_inode_loc(inode, &iloc);
778 		if (ret)
779 			goto out;
780 		raw_inode = ext4_raw_inode(&iloc);
781 		header = IHDR(inode, raw_inode);
782 		end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
783 		ret = xattr_check_inode(inode, header, end);
784 		if (ret)
785 			goto out;
786 
787 		for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
788 		     entry = EXT4_XATTR_NEXT(entry))
789 			if (entry->e_value_inum)
790 				ea_inode_refs++;
791 	}
792 
793 	if (EXT4_I(inode)->i_file_acl) {
794 		bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
795 		if (!bh) {
796 			ret = -EIO;
797 			goto out;
798 		}
799 
800 		if (ext4_xattr_check_block(inode, bh)) {
801 			ret = -EFSCORRUPTED;
802 			goto out;
803 		}
804 
805 		for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
806 		     entry = EXT4_XATTR_NEXT(entry))
807 			if (entry->e_value_inum)
808 				ea_inode_refs++;
809 	}
810 	*usage = ea_inode_refs + 1;
811 	ret = 0;
812 out:
813 	brelse(iloc.bh);
814 	brelse(bh);
815 	return ret;
816 }
817 
818 static inline size_t round_up_cluster(struct inode *inode, size_t length)
819 {
820 	struct super_block *sb = inode->i_sb;
821 	size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
822 				    inode->i_blkbits);
823 	size_t mask = ~(cluster_size - 1);
824 
825 	return (length + cluster_size - 1) & mask;
826 }
827 
828 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
829 {
830 	int err;
831 
832 	err = dquot_alloc_inode(inode);
833 	if (err)
834 		return err;
835 	err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
836 	if (err)
837 		dquot_free_inode(inode);
838 	return err;
839 }
840 
841 static void ext4_xattr_inode_free_quota(struct inode *inode, size_t len)
842 {
843 	dquot_free_space_nodirty(inode, round_up_cluster(inode, len));
844 	dquot_free_inode(inode);
845 }
846 
847 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
848 			     struct buffer_head *block_bh, size_t value_len,
849 			     bool is_create)
850 {
851 	int credits;
852 	int blocks;
853 
854 	/*
855 	 * 1) Owner inode update
856 	 * 2) Ref count update on old xattr block
857 	 * 3) new xattr block
858 	 * 4) block bitmap update for new xattr block
859 	 * 5) group descriptor for new xattr block
860 	 * 6) block bitmap update for old xattr block
861 	 * 7) group descriptor for old block
862 	 *
863 	 * 6 & 7 can happen if we have two racing threads T_a and T_b
864 	 * which are each trying to set an xattr on inodes I_a and I_b
865 	 * which were both initially sharing an xattr block.
866 	 */
867 	credits = 7;
868 
869 	/* Quota updates. */
870 	credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
871 
872 	/*
873 	 * In case of inline data, we may push out the data to a block,
874 	 * so we need to reserve credits for this eventuality
875 	 */
876 	if (inode && ext4_has_inline_data(inode))
877 		credits += ext4_writepage_trans_blocks(inode) + 1;
878 
879 	/* We are done if ea_inode feature is not enabled. */
880 	if (!ext4_has_feature_ea_inode(sb))
881 		return credits;
882 
883 	/* New ea_inode, inode map, block bitmap, group descriptor. */
884 	credits += 4;
885 
886 	/* Data blocks. */
887 	blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
888 
889 	/* Indirection block or one level of extent tree. */
890 	blocks += 1;
891 
892 	/* Block bitmap and group descriptor updates for each block. */
893 	credits += blocks * 2;
894 
895 	/* Blocks themselves. */
896 	credits += blocks;
897 
898 	if (!is_create) {
899 		/* Dereference ea_inode holding old xattr value.
900 		 * Old ea_inode, inode map, block bitmap, group descriptor.
901 		 */
902 		credits += 4;
903 
904 		/* Data blocks for old ea_inode. */
905 		blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
906 
907 		/* Indirection block or one level of extent tree for old
908 		 * ea_inode.
909 		 */
910 		blocks += 1;
911 
912 		/* Block bitmap and group descriptor updates for each block. */
913 		credits += blocks * 2;
914 	}
915 
916 	/* We may need to clone the existing xattr block in which case we need
917 	 * to increment ref counts for existing ea_inodes referenced by it.
918 	 */
919 	if (block_bh) {
920 		struct ext4_xattr_entry *entry = BFIRST(block_bh);
921 
922 		for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
923 			if (entry->e_value_inum)
924 				/* Ref count update on ea_inode. */
925 				credits += 1;
926 	}
927 	return credits;
928 }
929 
930 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
931 				     int credits, struct buffer_head *bh,
932 				     bool dirty, bool block_csum)
933 {
934 	int error;
935 
936 	if (!ext4_handle_valid(handle))
937 		return 0;
938 
939 	if (handle->h_buffer_credits >= credits)
940 		return 0;
941 
942 	error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
943 	if (!error)
944 		return 0;
945 	if (error < 0) {
946 		ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
947 		return error;
948 	}
949 
950 	if (bh && dirty) {
951 		if (block_csum)
952 			ext4_xattr_block_csum_set(inode, bh);
953 		error = ext4_handle_dirty_metadata(handle, NULL, bh);
954 		if (error) {
955 			ext4_warning(inode->i_sb, "Handle metadata (error %d)",
956 				     error);
957 			return error;
958 		}
959 	}
960 
961 	error = ext4_journal_restart(handle, credits);
962 	if (error) {
963 		ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
964 		return error;
965 	}
966 
967 	if (bh) {
968 		error = ext4_journal_get_write_access(handle, bh);
969 		if (error) {
970 			ext4_warning(inode->i_sb,
971 				     "Get write access failed (error %d)",
972 				     error);
973 			return error;
974 		}
975 	}
976 	return 0;
977 }
978 
979 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
980 				       int ref_change)
981 {
982 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
983 	struct ext4_iloc iloc;
984 	s64 ref_count;
985 	u32 hash;
986 	int ret;
987 
988 	inode_lock(ea_inode);
989 
990 	ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
991 	if (ret) {
992 		iloc.bh = NULL;
993 		goto out;
994 	}
995 
996 	ref_count = ext4_xattr_inode_get_ref(ea_inode);
997 	ref_count += ref_change;
998 	ext4_xattr_inode_set_ref(ea_inode, ref_count);
999 
1000 	if (ref_change > 0) {
1001 		WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1002 			  ea_inode->i_ino, ref_count);
1003 
1004 		if (ref_count == 1) {
1005 			WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1006 				  ea_inode->i_ino, ea_inode->i_nlink);
1007 
1008 			set_nlink(ea_inode, 1);
1009 			ext4_orphan_del(handle, ea_inode);
1010 
1011 			if (ea_inode_cache) {
1012 				hash = ext4_xattr_inode_get_hash(ea_inode);
1013 				mb_cache_entry_create(ea_inode_cache,
1014 						      GFP_NOFS, hash,
1015 						      ea_inode->i_ino,
1016 						      true /* reusable */);
1017 			}
1018 		}
1019 	} else {
1020 		WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1021 			  ea_inode->i_ino, ref_count);
1022 
1023 		if (ref_count == 0) {
1024 			WARN_ONCE(ea_inode->i_nlink != 1,
1025 				  "EA inode %lu i_nlink=%u",
1026 				  ea_inode->i_ino, ea_inode->i_nlink);
1027 
1028 			clear_nlink(ea_inode);
1029 			ext4_orphan_add(handle, ea_inode);
1030 
1031 			if (ea_inode_cache) {
1032 				hash = ext4_xattr_inode_get_hash(ea_inode);
1033 				mb_cache_entry_delete(ea_inode_cache, hash,
1034 						      ea_inode->i_ino);
1035 			}
1036 		}
1037 	}
1038 
1039 	ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1040 	iloc.bh = NULL;
1041 	if (ret)
1042 		ext4_warning_inode(ea_inode,
1043 				   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1044 out:
1045 	brelse(iloc.bh);
1046 	inode_unlock(ea_inode);
1047 	return ret;
1048 }
1049 
1050 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1051 {
1052 	return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1053 }
1054 
1055 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1056 {
1057 	return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1058 }
1059 
1060 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1061 					struct ext4_xattr_entry *first)
1062 {
1063 	struct inode *ea_inode;
1064 	struct ext4_xattr_entry *entry;
1065 	struct ext4_xattr_entry *failed_entry;
1066 	unsigned int ea_ino;
1067 	int err, saved_err;
1068 
1069 	for (entry = first; !IS_LAST_ENTRY(entry);
1070 	     entry = EXT4_XATTR_NEXT(entry)) {
1071 		if (!entry->e_value_inum)
1072 			continue;
1073 		ea_ino = le32_to_cpu(entry->e_value_inum);
1074 		err = ext4_xattr_inode_iget(parent, ea_ino, &ea_inode);
1075 		if (err)
1076 			goto cleanup;
1077 		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1078 		if (err) {
1079 			ext4_warning_inode(ea_inode, "inc ref error %d", err);
1080 			iput(ea_inode);
1081 			goto cleanup;
1082 		}
1083 		iput(ea_inode);
1084 	}
1085 	return 0;
1086 
1087 cleanup:
1088 	saved_err = err;
1089 	failed_entry = entry;
1090 
1091 	for (entry = first; entry != failed_entry;
1092 	     entry = EXT4_XATTR_NEXT(entry)) {
1093 		if (!entry->e_value_inum)
1094 			continue;
1095 		ea_ino = le32_to_cpu(entry->e_value_inum);
1096 		err = ext4_xattr_inode_iget(parent, ea_ino, &ea_inode);
1097 		if (err) {
1098 			ext4_warning(parent->i_sb,
1099 				     "cleanup ea_ino %u iget error %d", ea_ino,
1100 				     err);
1101 			continue;
1102 		}
1103 		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1104 		if (err)
1105 			ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1106 					   err);
1107 		iput(ea_inode);
1108 	}
1109 	return saved_err;
1110 }
1111 
1112 static void
1113 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1114 			     struct buffer_head *bh,
1115 			     struct ext4_xattr_entry *first, bool block_csum,
1116 			     struct ext4_xattr_inode_array **ea_inode_array,
1117 			     int extra_credits, bool skip_quota)
1118 {
1119 	struct inode *ea_inode;
1120 	struct ext4_xattr_entry *entry;
1121 	bool dirty = false;
1122 	unsigned int ea_ino;
1123 	int err;
1124 	int credits;
1125 
1126 	/* One credit for dec ref on ea_inode, one for orphan list addition, */
1127 	credits = 2 + extra_credits;
1128 
1129 	for (entry = first; !IS_LAST_ENTRY(entry);
1130 	     entry = EXT4_XATTR_NEXT(entry)) {
1131 		if (!entry->e_value_inum)
1132 			continue;
1133 		ea_ino = le32_to_cpu(entry->e_value_inum);
1134 		err = ext4_xattr_inode_iget(parent, ea_ino, &ea_inode);
1135 		if (err)
1136 			continue;
1137 
1138 		err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1139 		if (err) {
1140 			ext4_warning_inode(ea_inode,
1141 					   "Expand inode array err=%d", err);
1142 			iput(ea_inode);
1143 			continue;
1144 		}
1145 
1146 		err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1147 						dirty, block_csum);
1148 		if (err) {
1149 			ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1150 					   err);
1151 			continue;
1152 		}
1153 
1154 		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1155 		if (err) {
1156 			ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1157 					   err);
1158 			continue;
1159 		}
1160 
1161 		if (!skip_quota)
1162 			ext4_xattr_inode_free_quota(parent,
1163 					      le32_to_cpu(entry->e_value_size));
1164 
1165 		/*
1166 		 * Forget about ea_inode within the same transaction that
1167 		 * decrements the ref count. This avoids duplicate decrements in
1168 		 * case the rest of the work spills over to subsequent
1169 		 * transactions.
1170 		 */
1171 		entry->e_value_inum = 0;
1172 		entry->e_value_size = 0;
1173 
1174 		dirty = true;
1175 	}
1176 
1177 	if (dirty) {
1178 		/*
1179 		 * Note that we are deliberately skipping csum calculation for
1180 		 * the final update because we do not expect any journal
1181 		 * restarts until xattr block is freed.
1182 		 */
1183 
1184 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1185 		if (err)
1186 			ext4_warning_inode(parent,
1187 					   "handle dirty metadata err=%d", err);
1188 	}
1189 }
1190 
1191 /*
1192  * Release the xattr block BH: If the reference count is > 1, decrement it;
1193  * otherwise free the block.
1194  */
1195 static void
1196 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1197 			 struct buffer_head *bh,
1198 			 struct ext4_xattr_inode_array **ea_inode_array,
1199 			 int extra_credits)
1200 {
1201 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1202 	u32 hash, ref;
1203 	int error = 0;
1204 
1205 	BUFFER_TRACE(bh, "get_write_access");
1206 	error = ext4_journal_get_write_access(handle, bh);
1207 	if (error)
1208 		goto out;
1209 
1210 	lock_buffer(bh);
1211 	hash = le32_to_cpu(BHDR(bh)->h_hash);
1212 	ref = le32_to_cpu(BHDR(bh)->h_refcount);
1213 	if (ref == 1) {
1214 		ea_bdebug(bh, "refcount now=0; freeing");
1215 		/*
1216 		 * This must happen under buffer lock for
1217 		 * ext4_xattr_block_set() to reliably detect freed block
1218 		 */
1219 		if (ea_block_cache)
1220 			mb_cache_entry_delete(ea_block_cache, hash,
1221 					      bh->b_blocknr);
1222 		get_bh(bh);
1223 		unlock_buffer(bh);
1224 
1225 		if (ext4_has_feature_ea_inode(inode->i_sb))
1226 			ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1227 						     BFIRST(bh),
1228 						     true /* block_csum */,
1229 						     ea_inode_array,
1230 						     extra_credits,
1231 						     true /* skip_quota */);
1232 		ext4_free_blocks(handle, inode, bh, 0, 1,
1233 				 EXT4_FREE_BLOCKS_METADATA |
1234 				 EXT4_FREE_BLOCKS_FORGET);
1235 	} else {
1236 		ref--;
1237 		BHDR(bh)->h_refcount = cpu_to_le32(ref);
1238 		if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1239 			struct mb_cache_entry *ce;
1240 
1241 			if (ea_block_cache) {
1242 				ce = mb_cache_entry_get(ea_block_cache, hash,
1243 							bh->b_blocknr);
1244 				if (ce) {
1245 					ce->e_reusable = 1;
1246 					mb_cache_entry_put(ea_block_cache, ce);
1247 				}
1248 			}
1249 		}
1250 
1251 		ext4_xattr_block_csum_set(inode, bh);
1252 		/*
1253 		 * Beware of this ugliness: Releasing of xattr block references
1254 		 * from different inodes can race and so we have to protect
1255 		 * from a race where someone else frees the block (and releases
1256 		 * its journal_head) before we are done dirtying the buffer. In
1257 		 * nojournal mode this race is harmless and we actually cannot
1258 		 * call ext4_handle_dirty_metadata() with locked buffer as
1259 		 * that function can call sync_dirty_buffer() so for that case
1260 		 * we handle the dirtying after unlocking the buffer.
1261 		 */
1262 		if (ext4_handle_valid(handle))
1263 			error = ext4_handle_dirty_metadata(handle, inode, bh);
1264 		unlock_buffer(bh);
1265 		if (!ext4_handle_valid(handle))
1266 			error = ext4_handle_dirty_metadata(handle, inode, bh);
1267 		if (IS_SYNC(inode))
1268 			ext4_handle_sync(handle);
1269 		dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1270 		ea_bdebug(bh, "refcount now=%d; releasing",
1271 			  le32_to_cpu(BHDR(bh)->h_refcount));
1272 	}
1273 out:
1274 	ext4_std_error(inode->i_sb, error);
1275 	return;
1276 }
1277 
1278 /*
1279  * Find the available free space for EAs. This also returns the total number of
1280  * bytes used by EA entries.
1281  */
1282 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1283 				    size_t *min_offs, void *base, int *total)
1284 {
1285 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1286 		if (!last->e_value_inum && last->e_value_size) {
1287 			size_t offs = le16_to_cpu(last->e_value_offs);
1288 			if (offs < *min_offs)
1289 				*min_offs = offs;
1290 		}
1291 		if (total)
1292 			*total += EXT4_XATTR_LEN(last->e_name_len);
1293 	}
1294 	return (*min_offs - ((void *)last - base) - sizeof(__u32));
1295 }
1296 
1297 /*
1298  * Write the value of the EA in an inode.
1299  */
1300 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1301 				  const void *buf, int bufsize)
1302 {
1303 	struct buffer_head *bh = NULL;
1304 	unsigned long block = 0;
1305 	int blocksize = ea_inode->i_sb->s_blocksize;
1306 	int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1307 	int csize, wsize = 0;
1308 	int ret = 0;
1309 	int retries = 0;
1310 
1311 retry:
1312 	while (ret >= 0 && ret < max_blocks) {
1313 		struct ext4_map_blocks map;
1314 		map.m_lblk = block += ret;
1315 		map.m_len = max_blocks -= ret;
1316 
1317 		ret = ext4_map_blocks(handle, ea_inode, &map,
1318 				      EXT4_GET_BLOCKS_CREATE);
1319 		if (ret <= 0) {
1320 			ext4_mark_inode_dirty(handle, ea_inode);
1321 			if (ret == -ENOSPC &&
1322 			    ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1323 				ret = 0;
1324 				goto retry;
1325 			}
1326 			break;
1327 		}
1328 	}
1329 
1330 	if (ret < 0)
1331 		return ret;
1332 
1333 	block = 0;
1334 	while (wsize < bufsize) {
1335 		if (bh != NULL)
1336 			brelse(bh);
1337 		csize = (bufsize - wsize) > blocksize ? blocksize :
1338 								bufsize - wsize;
1339 		bh = ext4_getblk(handle, ea_inode, block, 0);
1340 		if (IS_ERR(bh))
1341 			return PTR_ERR(bh);
1342 		ret = ext4_journal_get_write_access(handle, bh);
1343 		if (ret)
1344 			goto out;
1345 
1346 		memcpy(bh->b_data, buf, csize);
1347 		set_buffer_uptodate(bh);
1348 		ext4_handle_dirty_metadata(handle, ea_inode, bh);
1349 
1350 		buf += csize;
1351 		wsize += csize;
1352 		block += 1;
1353 	}
1354 
1355 	inode_lock(ea_inode);
1356 	i_size_write(ea_inode, wsize);
1357 	ext4_update_i_disksize(ea_inode, wsize);
1358 	inode_unlock(ea_inode);
1359 
1360 	ext4_mark_inode_dirty(handle, ea_inode);
1361 
1362 out:
1363 	brelse(bh);
1364 
1365 	return ret;
1366 }
1367 
1368 /*
1369  * Create an inode to store the value of a large EA.
1370  */
1371 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1372 					     struct inode *inode, u32 hash)
1373 {
1374 	struct inode *ea_inode = NULL;
1375 	uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1376 	int err;
1377 
1378 	/*
1379 	 * Let the next inode be the goal, so we try and allocate the EA inode
1380 	 * in the same group, or nearby one.
1381 	 */
1382 	ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1383 				  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1384 				  EXT4_EA_INODE_FL);
1385 	if (!IS_ERR(ea_inode)) {
1386 		ea_inode->i_op = &ext4_file_inode_operations;
1387 		ea_inode->i_fop = &ext4_file_operations;
1388 		ext4_set_aops(ea_inode);
1389 		ext4_xattr_inode_set_class(ea_inode);
1390 		unlock_new_inode(ea_inode);
1391 		ext4_xattr_inode_set_ref(ea_inode, 1);
1392 		ext4_xattr_inode_set_hash(ea_inode, hash);
1393 		err = ext4_mark_inode_dirty(handle, ea_inode);
1394 		if (!err)
1395 			err = ext4_inode_attach_jinode(ea_inode);
1396 		if (err) {
1397 			iput(ea_inode);
1398 			return ERR_PTR(err);
1399 		}
1400 
1401 		/*
1402 		 * Xattr inodes are shared therefore quota charging is performed
1403 		 * at a higher level.
1404 		 */
1405 		dquot_free_inode(ea_inode);
1406 		dquot_drop(ea_inode);
1407 		inode_lock(ea_inode);
1408 		ea_inode->i_flags |= S_NOQUOTA;
1409 		inode_unlock(ea_inode);
1410 	}
1411 
1412 	return ea_inode;
1413 }
1414 
1415 static struct inode *
1416 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1417 			    size_t value_len, u32 hash)
1418 {
1419 	struct inode *ea_inode;
1420 	struct mb_cache_entry *ce;
1421 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1422 	void *ea_data;
1423 
1424 	if (!ea_inode_cache)
1425 		return NULL;
1426 
1427 	ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1428 	if (!ce)
1429 		return NULL;
1430 
1431 	ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1432 	if (!ea_data) {
1433 		mb_cache_entry_put(ea_inode_cache, ce);
1434 		return NULL;
1435 	}
1436 
1437 	while (ce) {
1438 		ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1439 		if (!IS_ERR(ea_inode) &&
1440 		    !is_bad_inode(ea_inode) &&
1441 		    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1442 		    i_size_read(ea_inode) == value_len &&
1443 		    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1444 		    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1445 						    value_len) &&
1446 		    !memcmp(value, ea_data, value_len)) {
1447 			mb_cache_entry_touch(ea_inode_cache, ce);
1448 			mb_cache_entry_put(ea_inode_cache, ce);
1449 			kvfree(ea_data);
1450 			return ea_inode;
1451 		}
1452 
1453 		if (!IS_ERR(ea_inode))
1454 			iput(ea_inode);
1455 		ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1456 	}
1457 	kvfree(ea_data);
1458 	return NULL;
1459 }
1460 
1461 /*
1462  * Add value of the EA in an inode.
1463  */
1464 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1465 					  const void *value, size_t value_len,
1466 					  struct inode **ret_inode)
1467 {
1468 	struct inode *ea_inode;
1469 	u32 hash;
1470 	int err;
1471 
1472 	hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1473 	ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1474 	if (ea_inode) {
1475 		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1476 		if (err) {
1477 			iput(ea_inode);
1478 			return err;
1479 		}
1480 
1481 		*ret_inode = ea_inode;
1482 		return 0;
1483 	}
1484 
1485 	/* Create an inode for the EA value */
1486 	ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1487 	if (IS_ERR(ea_inode))
1488 		return PTR_ERR(ea_inode);
1489 
1490 	err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1491 	if (err) {
1492 		ext4_xattr_inode_dec_ref(handle, ea_inode);
1493 		iput(ea_inode);
1494 		return err;
1495 	}
1496 
1497 	if (EA_INODE_CACHE(inode))
1498 		mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1499 				      ea_inode->i_ino, true /* reusable */);
1500 
1501 	*ret_inode = ea_inode;
1502 	return 0;
1503 }
1504 
1505 /*
1506  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1507  * feature is enabled.
1508  */
1509 #define EXT4_XATTR_BLOCK_RESERVE(inode)	min(i_blocksize(inode)/8, 1024U)
1510 
1511 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1512 				struct ext4_xattr_search *s,
1513 				handle_t *handle, struct inode *inode,
1514 				bool is_block)
1515 {
1516 	struct ext4_xattr_entry *last;
1517 	struct ext4_xattr_entry *here = s->here;
1518 	size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1519 	int in_inode = i->in_inode;
1520 	struct inode *old_ea_inode = NULL;
1521 	struct inode *new_ea_inode = NULL;
1522 	size_t old_size, new_size;
1523 	int ret;
1524 
1525 	/* Space used by old and new values. */
1526 	old_size = (!s->not_found && !here->e_value_inum) ?
1527 			EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1528 	new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1529 
1530 	/*
1531 	 * Optimization for the simple case when old and new values have the
1532 	 * same padded sizes. Not applicable if external inodes are involved.
1533 	 */
1534 	if (new_size && new_size == old_size) {
1535 		size_t offs = le16_to_cpu(here->e_value_offs);
1536 		void *val = s->base + offs;
1537 
1538 		here->e_value_size = cpu_to_le32(i->value_len);
1539 		if (i->value == EXT4_ZERO_XATTR_VALUE) {
1540 			memset(val, 0, new_size);
1541 		} else {
1542 			memcpy(val, i->value, i->value_len);
1543 			/* Clear padding bytes. */
1544 			memset(val + i->value_len, 0, new_size - i->value_len);
1545 		}
1546 		return 0;
1547 	}
1548 
1549 	/* Compute min_offs and last. */
1550 	last = s->first;
1551 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1552 		if (!last->e_value_inum && last->e_value_size) {
1553 			size_t offs = le16_to_cpu(last->e_value_offs);
1554 			if (offs < min_offs)
1555 				min_offs = offs;
1556 		}
1557 	}
1558 
1559 	/* Check whether we have enough space. */
1560 	if (i->value) {
1561 		size_t free;
1562 
1563 		free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1564 		if (!s->not_found)
1565 			free += EXT4_XATTR_LEN(name_len) + old_size;
1566 
1567 		if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1568 			ret = -ENOSPC;
1569 			goto out;
1570 		}
1571 
1572 		/*
1573 		 * If storing the value in an external inode is an option,
1574 		 * reserve space for xattr entries/names in the external
1575 		 * attribute block so that a long value does not occupy the
1576 		 * whole space and prevent futher entries being added.
1577 		 */
1578 		if (ext4_has_feature_ea_inode(inode->i_sb) &&
1579 		    new_size && is_block &&
1580 		    (min_offs + old_size - new_size) <
1581 					EXT4_XATTR_BLOCK_RESERVE(inode)) {
1582 			ret = -ENOSPC;
1583 			goto out;
1584 		}
1585 	}
1586 
1587 	/*
1588 	 * Getting access to old and new ea inodes is subject to failures.
1589 	 * Finish that work before doing any modifications to the xattr data.
1590 	 */
1591 	if (!s->not_found && here->e_value_inum) {
1592 		ret = ext4_xattr_inode_iget(inode,
1593 					    le32_to_cpu(here->e_value_inum),
1594 					    &old_ea_inode);
1595 		if (ret) {
1596 			old_ea_inode = NULL;
1597 			goto out;
1598 		}
1599 	}
1600 	if (i->value && in_inode) {
1601 		WARN_ON_ONCE(!i->value_len);
1602 
1603 		ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1604 		if (ret)
1605 			goto out;
1606 
1607 		ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1608 						     i->value_len,
1609 						     &new_ea_inode);
1610 		if (ret) {
1611 			new_ea_inode = NULL;
1612 			ext4_xattr_inode_free_quota(inode, i->value_len);
1613 			goto out;
1614 		}
1615 	}
1616 
1617 	if (old_ea_inode) {
1618 		/* We are ready to release ref count on the old_ea_inode. */
1619 		ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1620 		if (ret) {
1621 			/* Release newly required ref count on new_ea_inode. */
1622 			if (new_ea_inode) {
1623 				int err;
1624 
1625 				err = ext4_xattr_inode_dec_ref(handle,
1626 							       new_ea_inode);
1627 				if (err)
1628 					ext4_warning_inode(new_ea_inode,
1629 						  "dec ref new_ea_inode err=%d",
1630 						  err);
1631 				ext4_xattr_inode_free_quota(inode,
1632 							    i->value_len);
1633 			}
1634 			goto out;
1635 		}
1636 
1637 		ext4_xattr_inode_free_quota(inode,
1638 					    le32_to_cpu(here->e_value_size));
1639 	}
1640 
1641 	/* No failures allowed past this point. */
1642 
1643 	if (!s->not_found && here->e_value_offs) {
1644 		/* Remove the old value. */
1645 		void *first_val = s->base + min_offs;
1646 		size_t offs = le16_to_cpu(here->e_value_offs);
1647 		void *val = s->base + offs;
1648 
1649 		memmove(first_val + old_size, first_val, val - first_val);
1650 		memset(first_val, 0, old_size);
1651 		min_offs += old_size;
1652 
1653 		/* Adjust all value offsets. */
1654 		last = s->first;
1655 		while (!IS_LAST_ENTRY(last)) {
1656 			size_t o = le16_to_cpu(last->e_value_offs);
1657 
1658 			if (!last->e_value_inum &&
1659 			    last->e_value_size && o < offs)
1660 				last->e_value_offs = cpu_to_le16(o + old_size);
1661 			last = EXT4_XATTR_NEXT(last);
1662 		}
1663 	}
1664 
1665 	if (!i->value) {
1666 		/* Remove old name. */
1667 		size_t size = EXT4_XATTR_LEN(name_len);
1668 
1669 		last = ENTRY((void *)last - size);
1670 		memmove(here, (void *)here + size,
1671 			(void *)last - (void *)here + sizeof(__u32));
1672 		memset(last, 0, size);
1673 	} else if (s->not_found) {
1674 		/* Insert new name. */
1675 		size_t size = EXT4_XATTR_LEN(name_len);
1676 		size_t rest = (void *)last - (void *)here + sizeof(__u32);
1677 
1678 		memmove((void *)here + size, here, rest);
1679 		memset(here, 0, size);
1680 		here->e_name_index = i->name_index;
1681 		here->e_name_len = name_len;
1682 		memcpy(here->e_name, i->name, name_len);
1683 	} else {
1684 		/* This is an update, reset value info. */
1685 		here->e_value_inum = 0;
1686 		here->e_value_offs = 0;
1687 		here->e_value_size = 0;
1688 	}
1689 
1690 	if (i->value) {
1691 		/* Insert new value. */
1692 		if (in_inode) {
1693 			here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1694 		} else if (i->value_len) {
1695 			void *val = s->base + min_offs - new_size;
1696 
1697 			here->e_value_offs = cpu_to_le16(min_offs - new_size);
1698 			if (i->value == EXT4_ZERO_XATTR_VALUE) {
1699 				memset(val, 0, new_size);
1700 			} else {
1701 				memcpy(val, i->value, i->value_len);
1702 				/* Clear padding bytes. */
1703 				memset(val + i->value_len, 0,
1704 				       new_size - i->value_len);
1705 			}
1706 		}
1707 		here->e_value_size = cpu_to_le32(i->value_len);
1708 	}
1709 
1710 	if (i->value) {
1711 		__le32 hash = 0;
1712 
1713 		/* Entry hash calculation. */
1714 		if (in_inode) {
1715 			__le32 crc32c_hash;
1716 
1717 			/*
1718 			 * Feed crc32c hash instead of the raw value for entry
1719 			 * hash calculation. This is to avoid walking
1720 			 * potentially long value buffer again.
1721 			 */
1722 			crc32c_hash = cpu_to_le32(
1723 				       ext4_xattr_inode_get_hash(new_ea_inode));
1724 			hash = ext4_xattr_hash_entry(here->e_name,
1725 						     here->e_name_len,
1726 						     &crc32c_hash, 1);
1727 		} else if (is_block) {
1728 			__le32 *value = s->base + min_offs - new_size;
1729 
1730 			hash = ext4_xattr_hash_entry(here->e_name,
1731 						     here->e_name_len, value,
1732 						     new_size >> 2);
1733 		}
1734 		here->e_hash = hash;
1735 	}
1736 
1737 	if (is_block)
1738 		ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1739 
1740 	ret = 0;
1741 out:
1742 	iput(old_ea_inode);
1743 	iput(new_ea_inode);
1744 	return ret;
1745 }
1746 
1747 struct ext4_xattr_block_find {
1748 	struct ext4_xattr_search s;
1749 	struct buffer_head *bh;
1750 };
1751 
1752 static int
1753 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1754 		      struct ext4_xattr_block_find *bs)
1755 {
1756 	struct super_block *sb = inode->i_sb;
1757 	int error;
1758 
1759 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1760 		  i->name_index, i->name, i->value, (long)i->value_len);
1761 
1762 	if (EXT4_I(inode)->i_file_acl) {
1763 		/* The inode already has an extended attribute block. */
1764 		bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1765 		error = -EIO;
1766 		if (!bs->bh)
1767 			goto cleanup;
1768 		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1769 			atomic_read(&(bs->bh->b_count)),
1770 			le32_to_cpu(BHDR(bs->bh)->h_refcount));
1771 		if (ext4_xattr_check_block(inode, bs->bh)) {
1772 			EXT4_ERROR_INODE(inode, "bad block %llu",
1773 					 EXT4_I(inode)->i_file_acl);
1774 			error = -EFSCORRUPTED;
1775 			goto cleanup;
1776 		}
1777 		/* Find the named attribute. */
1778 		bs->s.base = BHDR(bs->bh);
1779 		bs->s.first = BFIRST(bs->bh);
1780 		bs->s.end = bs->bh->b_data + bs->bh->b_size;
1781 		bs->s.here = bs->s.first;
1782 		error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
1783 					      i->name, 1);
1784 		if (error && error != -ENODATA)
1785 			goto cleanup;
1786 		bs->s.not_found = error;
1787 	}
1788 	error = 0;
1789 
1790 cleanup:
1791 	return error;
1792 }
1793 
1794 static int
1795 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1796 		     struct ext4_xattr_info *i,
1797 		     struct ext4_xattr_block_find *bs)
1798 {
1799 	struct super_block *sb = inode->i_sb;
1800 	struct buffer_head *new_bh = NULL;
1801 	struct ext4_xattr_search s_copy = bs->s;
1802 	struct ext4_xattr_search *s = &s_copy;
1803 	struct mb_cache_entry *ce = NULL;
1804 	int error = 0;
1805 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1806 	struct inode *ea_inode = NULL;
1807 	size_t old_ea_inode_size = 0;
1808 
1809 #define header(x) ((struct ext4_xattr_header *)(x))
1810 
1811 	if (s->base) {
1812 		BUFFER_TRACE(bs->bh, "get_write_access");
1813 		error = ext4_journal_get_write_access(handle, bs->bh);
1814 		if (error)
1815 			goto cleanup;
1816 		lock_buffer(bs->bh);
1817 
1818 		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1819 			__u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1820 
1821 			/*
1822 			 * This must happen under buffer lock for
1823 			 * ext4_xattr_block_set() to reliably detect modified
1824 			 * block
1825 			 */
1826 			if (ea_block_cache)
1827 				mb_cache_entry_delete(ea_block_cache, hash,
1828 						      bs->bh->b_blocknr);
1829 			ea_bdebug(bs->bh, "modifying in-place");
1830 			error = ext4_xattr_set_entry(i, s, handle, inode,
1831 						     true /* is_block */);
1832 			ext4_xattr_block_csum_set(inode, bs->bh);
1833 			unlock_buffer(bs->bh);
1834 			if (error == -EFSCORRUPTED)
1835 				goto bad_block;
1836 			if (!error)
1837 				error = ext4_handle_dirty_metadata(handle,
1838 								   inode,
1839 								   bs->bh);
1840 			if (error)
1841 				goto cleanup;
1842 			goto inserted;
1843 		} else {
1844 			int offset = (char *)s->here - bs->bh->b_data;
1845 
1846 			unlock_buffer(bs->bh);
1847 			ea_bdebug(bs->bh, "cloning");
1848 			s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1849 			error = -ENOMEM;
1850 			if (s->base == NULL)
1851 				goto cleanup;
1852 			memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1853 			s->first = ENTRY(header(s->base)+1);
1854 			header(s->base)->h_refcount = cpu_to_le32(1);
1855 			s->here = ENTRY(s->base + offset);
1856 			s->end = s->base + bs->bh->b_size;
1857 
1858 			/*
1859 			 * If existing entry points to an xattr inode, we need
1860 			 * to prevent ext4_xattr_set_entry() from decrementing
1861 			 * ref count on it because the reference belongs to the
1862 			 * original block. In this case, make the entry look
1863 			 * like it has an empty value.
1864 			 */
1865 			if (!s->not_found && s->here->e_value_inum) {
1866 				/*
1867 				 * Defer quota free call for previous inode
1868 				 * until success is guaranteed.
1869 				 */
1870 				old_ea_inode_size = le32_to_cpu(
1871 							s->here->e_value_size);
1872 				s->here->e_value_inum = 0;
1873 				s->here->e_value_size = 0;
1874 			}
1875 		}
1876 	} else {
1877 		/* Allocate a buffer where we construct the new block. */
1878 		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1879 		/* assert(header == s->base) */
1880 		error = -ENOMEM;
1881 		if (s->base == NULL)
1882 			goto cleanup;
1883 		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1884 		header(s->base)->h_blocks = cpu_to_le32(1);
1885 		header(s->base)->h_refcount = cpu_to_le32(1);
1886 		s->first = ENTRY(header(s->base)+1);
1887 		s->here = ENTRY(header(s->base)+1);
1888 		s->end = s->base + sb->s_blocksize;
1889 	}
1890 
1891 	error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1892 	if (error == -EFSCORRUPTED)
1893 		goto bad_block;
1894 	if (error)
1895 		goto cleanup;
1896 
1897 	if (i->value && s->here->e_value_inum) {
1898 		unsigned int ea_ino;
1899 
1900 		/*
1901 		 * A ref count on ea_inode has been taken as part of the call to
1902 		 * ext4_xattr_set_entry() above. We would like to drop this
1903 		 * extra ref but we have to wait until the xattr block is
1904 		 * initialized and has its own ref count on the ea_inode.
1905 		 */
1906 		ea_ino = le32_to_cpu(s->here->e_value_inum);
1907 		error = ext4_xattr_inode_iget(inode, ea_ino, &ea_inode);
1908 		if (error) {
1909 			ea_inode = NULL;
1910 			goto cleanup;
1911 		}
1912 	}
1913 
1914 inserted:
1915 	if (!IS_LAST_ENTRY(s->first)) {
1916 		new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1917 						     &ce);
1918 		if (new_bh) {
1919 			/* We found an identical block in the cache. */
1920 			if (new_bh == bs->bh)
1921 				ea_bdebug(new_bh, "keeping");
1922 			else {
1923 				u32 ref;
1924 
1925 				WARN_ON_ONCE(dquot_initialize_needed(inode));
1926 
1927 				/* The old block is released after updating
1928 				   the inode. */
1929 				error = dquot_alloc_block(inode,
1930 						EXT4_C2B(EXT4_SB(sb), 1));
1931 				if (error)
1932 					goto cleanup;
1933 				BUFFER_TRACE(new_bh, "get_write_access");
1934 				error = ext4_journal_get_write_access(handle,
1935 								      new_bh);
1936 				if (error)
1937 					goto cleanup_dquot;
1938 				lock_buffer(new_bh);
1939 				/*
1940 				 * We have to be careful about races with
1941 				 * freeing, rehashing or adding references to
1942 				 * xattr block. Once we hold buffer lock xattr
1943 				 * block's state is stable so we can check
1944 				 * whether the block got freed / rehashed or
1945 				 * not.  Since we unhash mbcache entry under
1946 				 * buffer lock when freeing / rehashing xattr
1947 				 * block, checking whether entry is still
1948 				 * hashed is reliable. Same rules hold for
1949 				 * e_reusable handling.
1950 				 */
1951 				if (hlist_bl_unhashed(&ce->e_hash_list) ||
1952 				    !ce->e_reusable) {
1953 					/*
1954 					 * Undo everything and check mbcache
1955 					 * again.
1956 					 */
1957 					unlock_buffer(new_bh);
1958 					dquot_free_block(inode,
1959 							 EXT4_C2B(EXT4_SB(sb),
1960 								  1));
1961 					brelse(new_bh);
1962 					mb_cache_entry_put(ea_block_cache, ce);
1963 					ce = NULL;
1964 					new_bh = NULL;
1965 					goto inserted;
1966 				}
1967 				ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
1968 				BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
1969 				if (ref >= EXT4_XATTR_REFCOUNT_MAX)
1970 					ce->e_reusable = 0;
1971 				ea_bdebug(new_bh, "reusing; refcount now=%d",
1972 					  ref);
1973 				ext4_xattr_block_csum_set(inode, new_bh);
1974 				unlock_buffer(new_bh);
1975 				error = ext4_handle_dirty_metadata(handle,
1976 								   inode,
1977 								   new_bh);
1978 				if (error)
1979 					goto cleanup_dquot;
1980 			}
1981 			mb_cache_entry_touch(ea_block_cache, ce);
1982 			mb_cache_entry_put(ea_block_cache, ce);
1983 			ce = NULL;
1984 		} else if (bs->bh && s->base == bs->bh->b_data) {
1985 			/* We were modifying this block in-place. */
1986 			ea_bdebug(bs->bh, "keeping this block");
1987 			ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
1988 			new_bh = bs->bh;
1989 			get_bh(new_bh);
1990 		} else {
1991 			/* We need to allocate a new block */
1992 			ext4_fsblk_t goal, block;
1993 
1994 			WARN_ON_ONCE(dquot_initialize_needed(inode));
1995 
1996 			goal = ext4_group_first_block_no(sb,
1997 						EXT4_I(inode)->i_block_group);
1998 
1999 			/* non-extent files can't have physical blocks past 2^32 */
2000 			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2001 				goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2002 
2003 			block = ext4_new_meta_blocks(handle, inode, goal, 0,
2004 						     NULL, &error);
2005 			if (error)
2006 				goto cleanup;
2007 
2008 			if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2009 				BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2010 
2011 			ea_idebug(inode, "creating block %llu",
2012 				  (unsigned long long)block);
2013 
2014 			new_bh = sb_getblk(sb, block);
2015 			if (unlikely(!new_bh)) {
2016 				error = -ENOMEM;
2017 getblk_failed:
2018 				ext4_free_blocks(handle, inode, NULL, block, 1,
2019 						 EXT4_FREE_BLOCKS_METADATA);
2020 				goto cleanup;
2021 			}
2022 			error = ext4_xattr_inode_inc_ref_all(handle, inode,
2023 						      ENTRY(header(s->base)+1));
2024 			if (error)
2025 				goto getblk_failed;
2026 			if (ea_inode) {
2027 				/* Drop the extra ref on ea_inode. */
2028 				error = ext4_xattr_inode_dec_ref(handle,
2029 								 ea_inode);
2030 				if (error)
2031 					ext4_warning_inode(ea_inode,
2032 							   "dec ref error=%d",
2033 							   error);
2034 				iput(ea_inode);
2035 				ea_inode = NULL;
2036 			}
2037 
2038 			lock_buffer(new_bh);
2039 			error = ext4_journal_get_create_access(handle, new_bh);
2040 			if (error) {
2041 				unlock_buffer(new_bh);
2042 				error = -EIO;
2043 				goto getblk_failed;
2044 			}
2045 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
2046 			ext4_xattr_block_csum_set(inode, new_bh);
2047 			set_buffer_uptodate(new_bh);
2048 			unlock_buffer(new_bh);
2049 			ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2050 			error = ext4_handle_dirty_metadata(handle, inode,
2051 							   new_bh);
2052 			if (error)
2053 				goto cleanup;
2054 		}
2055 	}
2056 
2057 	if (old_ea_inode_size)
2058 		ext4_xattr_inode_free_quota(inode, old_ea_inode_size);
2059 
2060 	/* Update the inode. */
2061 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2062 
2063 	/* Drop the previous xattr block. */
2064 	if (bs->bh && bs->bh != new_bh) {
2065 		struct ext4_xattr_inode_array *ea_inode_array = NULL;
2066 
2067 		ext4_xattr_release_block(handle, inode, bs->bh,
2068 					 &ea_inode_array,
2069 					 0 /* extra_credits */);
2070 		ext4_xattr_inode_array_free(ea_inode_array);
2071 	}
2072 	error = 0;
2073 
2074 cleanup:
2075 	if (ea_inode) {
2076 		int error2;
2077 
2078 		error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2079 		if (error2)
2080 			ext4_warning_inode(ea_inode, "dec ref error=%d",
2081 					   error2);
2082 
2083 		/* If there was an error, revert the quota charge. */
2084 		if (error)
2085 			ext4_xattr_inode_free_quota(inode,
2086 						    i_size_read(ea_inode));
2087 		iput(ea_inode);
2088 	}
2089 	if (ce)
2090 		mb_cache_entry_put(ea_block_cache, ce);
2091 	brelse(new_bh);
2092 	if (!(bs->bh && s->base == bs->bh->b_data))
2093 		kfree(s->base);
2094 
2095 	return error;
2096 
2097 cleanup_dquot:
2098 	dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2099 	goto cleanup;
2100 
2101 bad_block:
2102 	EXT4_ERROR_INODE(inode, "bad block %llu",
2103 			 EXT4_I(inode)->i_file_acl);
2104 	goto cleanup;
2105 
2106 #undef header
2107 }
2108 
2109 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2110 			  struct ext4_xattr_ibody_find *is)
2111 {
2112 	struct ext4_xattr_ibody_header *header;
2113 	struct ext4_inode *raw_inode;
2114 	int error;
2115 
2116 	if (EXT4_I(inode)->i_extra_isize == 0)
2117 		return 0;
2118 	raw_inode = ext4_raw_inode(&is->iloc);
2119 	header = IHDR(inode, raw_inode);
2120 	is->s.base = is->s.first = IFIRST(header);
2121 	is->s.here = is->s.first;
2122 	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2123 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2124 		error = xattr_check_inode(inode, header, is->s.end);
2125 		if (error)
2126 			return error;
2127 		/* Find the named attribute. */
2128 		error = ext4_xattr_find_entry(&is->s.here, i->name_index,
2129 					      i->name, 0);
2130 		if (error && error != -ENODATA)
2131 			return error;
2132 		is->s.not_found = error;
2133 	}
2134 	return 0;
2135 }
2136 
2137 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2138 				struct ext4_xattr_info *i,
2139 				struct ext4_xattr_ibody_find *is)
2140 {
2141 	struct ext4_xattr_ibody_header *header;
2142 	struct ext4_xattr_search *s = &is->s;
2143 	int error;
2144 
2145 	if (EXT4_I(inode)->i_extra_isize == 0)
2146 		return -ENOSPC;
2147 	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2148 	if (error) {
2149 		if (error == -ENOSPC &&
2150 		    ext4_has_inline_data(inode)) {
2151 			error = ext4_try_to_evict_inline_data(handle, inode,
2152 					EXT4_XATTR_LEN(strlen(i->name) +
2153 					EXT4_XATTR_SIZE(i->value_len)));
2154 			if (error)
2155 				return error;
2156 			error = ext4_xattr_ibody_find(inode, i, is);
2157 			if (error)
2158 				return error;
2159 			error = ext4_xattr_set_entry(i, s, handle, inode,
2160 						     false /* is_block */);
2161 		}
2162 		if (error)
2163 			return error;
2164 	}
2165 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2166 	if (!IS_LAST_ENTRY(s->first)) {
2167 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2168 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2169 	} else {
2170 		header->h_magic = cpu_to_le32(0);
2171 		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2172 	}
2173 	return 0;
2174 }
2175 
2176 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2177 				struct ext4_xattr_info *i,
2178 				struct ext4_xattr_ibody_find *is)
2179 {
2180 	struct ext4_xattr_ibody_header *header;
2181 	struct ext4_xattr_search *s = &is->s;
2182 	int error;
2183 
2184 	if (EXT4_I(inode)->i_extra_isize == 0)
2185 		return -ENOSPC;
2186 	error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2187 	if (error)
2188 		return error;
2189 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2190 	if (!IS_LAST_ENTRY(s->first)) {
2191 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2192 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2193 	} else {
2194 		header->h_magic = cpu_to_le32(0);
2195 		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2196 	}
2197 	return 0;
2198 }
2199 
2200 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2201 				 struct ext4_xattr_info *i)
2202 {
2203 	void *value;
2204 
2205 	/* When e_value_inum is set the value is stored externally. */
2206 	if (s->here->e_value_inum)
2207 		return 0;
2208 	if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2209 		return 0;
2210 	value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2211 	return !memcmp(value, i->value, i->value_len);
2212 }
2213 
2214 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2215 {
2216 	struct buffer_head *bh;
2217 	int error;
2218 
2219 	if (!EXT4_I(inode)->i_file_acl)
2220 		return NULL;
2221 	bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2222 	if (!bh)
2223 		return ERR_PTR(-EIO);
2224 	error = ext4_xattr_check_block(inode, bh);
2225 	if (error)
2226 		return ERR_PTR(error);
2227 	return bh;
2228 }
2229 
2230 /*
2231  * ext4_xattr_set_handle()
2232  *
2233  * Create, replace or remove an extended attribute for this inode.  Value
2234  * is NULL to remove an existing extended attribute, and non-NULL to
2235  * either replace an existing extended attribute, or create a new extended
2236  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2237  * specify that an extended attribute must exist and must not exist
2238  * previous to the call, respectively.
2239  *
2240  * Returns 0, or a negative error number on failure.
2241  */
2242 int
2243 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2244 		      const char *name, const void *value, size_t value_len,
2245 		      int flags)
2246 {
2247 	struct ext4_xattr_info i = {
2248 		.name_index = name_index,
2249 		.name = name,
2250 		.value = value,
2251 		.value_len = value_len,
2252 		.in_inode = 0,
2253 	};
2254 	struct ext4_xattr_ibody_find is = {
2255 		.s = { .not_found = -ENODATA, },
2256 	};
2257 	struct ext4_xattr_block_find bs = {
2258 		.s = { .not_found = -ENODATA, },
2259 	};
2260 	int no_expand;
2261 	int error;
2262 
2263 	if (!name)
2264 		return -EINVAL;
2265 	if (strlen(name) > 255)
2266 		return -ERANGE;
2267 
2268 	ext4_write_lock_xattr(inode, &no_expand);
2269 
2270 	/* Check journal credits under write lock. */
2271 	if (ext4_handle_valid(handle)) {
2272 		struct buffer_head *bh;
2273 		int credits;
2274 
2275 		bh = ext4_xattr_get_block(inode);
2276 		if (IS_ERR(bh)) {
2277 			error = PTR_ERR(bh);
2278 			goto cleanup;
2279 		}
2280 
2281 		credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2282 						   value_len,
2283 						   flags & XATTR_CREATE);
2284 		brelse(bh);
2285 
2286 		if (!ext4_handle_has_enough_credits(handle, credits)) {
2287 			error = -ENOSPC;
2288 			goto cleanup;
2289 		}
2290 	}
2291 
2292 	error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2293 	if (error)
2294 		goto cleanup;
2295 
2296 	if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2297 		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2298 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2299 		ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2300 	}
2301 
2302 	error = ext4_xattr_ibody_find(inode, &i, &is);
2303 	if (error)
2304 		goto cleanup;
2305 	if (is.s.not_found)
2306 		error = ext4_xattr_block_find(inode, &i, &bs);
2307 	if (error)
2308 		goto cleanup;
2309 	if (is.s.not_found && bs.s.not_found) {
2310 		error = -ENODATA;
2311 		if (flags & XATTR_REPLACE)
2312 			goto cleanup;
2313 		error = 0;
2314 		if (!value)
2315 			goto cleanup;
2316 	} else {
2317 		error = -EEXIST;
2318 		if (flags & XATTR_CREATE)
2319 			goto cleanup;
2320 	}
2321 
2322 	if (!value) {
2323 		if (!is.s.not_found)
2324 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2325 		else if (!bs.s.not_found)
2326 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2327 	} else {
2328 		error = 0;
2329 		/* Xattr value did not change? Save us some work and bail out */
2330 		if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2331 			goto cleanup;
2332 		if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2333 			goto cleanup;
2334 
2335 		if (ext4_has_feature_ea_inode(inode->i_sb) &&
2336 		    (EXT4_XATTR_SIZE(i.value_len) >
2337 			EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2338 			i.in_inode = 1;
2339 retry_inode:
2340 		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2341 		if (!error && !bs.s.not_found) {
2342 			i.value = NULL;
2343 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2344 		} else if (error == -ENOSPC) {
2345 			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2346 				error = ext4_xattr_block_find(inode, &i, &bs);
2347 				if (error)
2348 					goto cleanup;
2349 			}
2350 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2351 			if (!error && !is.s.not_found) {
2352 				i.value = NULL;
2353 				error = ext4_xattr_ibody_set(handle, inode, &i,
2354 							     &is);
2355 			} else if (error == -ENOSPC) {
2356 				/*
2357 				 * Xattr does not fit in the block, store at
2358 				 * external inode if possible.
2359 				 */
2360 				if (ext4_has_feature_ea_inode(inode->i_sb) &&
2361 				    !i.in_inode) {
2362 					i.in_inode = 1;
2363 					goto retry_inode;
2364 				}
2365 			}
2366 		}
2367 	}
2368 	if (!error) {
2369 		ext4_xattr_update_super_block(handle, inode->i_sb);
2370 		inode->i_ctime = current_time(inode);
2371 		if (!value)
2372 			no_expand = 0;
2373 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2374 		/*
2375 		 * The bh is consumed by ext4_mark_iloc_dirty, even with
2376 		 * error != 0.
2377 		 */
2378 		is.iloc.bh = NULL;
2379 		if (IS_SYNC(inode))
2380 			ext4_handle_sync(handle);
2381 	}
2382 
2383 cleanup:
2384 	brelse(is.iloc.bh);
2385 	brelse(bs.bh);
2386 	ext4_write_unlock_xattr(inode, &no_expand);
2387 	return error;
2388 }
2389 
2390 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2391 			   bool is_create, int *credits)
2392 {
2393 	struct buffer_head *bh;
2394 	int err;
2395 
2396 	*credits = 0;
2397 
2398 	if (!EXT4_SB(inode->i_sb)->s_journal)
2399 		return 0;
2400 
2401 	down_read(&EXT4_I(inode)->xattr_sem);
2402 
2403 	bh = ext4_xattr_get_block(inode);
2404 	if (IS_ERR(bh)) {
2405 		err = PTR_ERR(bh);
2406 	} else {
2407 		*credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2408 						    value_len, is_create);
2409 		brelse(bh);
2410 		err = 0;
2411 	}
2412 
2413 	up_read(&EXT4_I(inode)->xattr_sem);
2414 	return err;
2415 }
2416 
2417 /*
2418  * ext4_xattr_set()
2419  *
2420  * Like ext4_xattr_set_handle, but start from an inode. This extended
2421  * attribute modification is a filesystem transaction by itself.
2422  *
2423  * Returns 0, or a negative error number on failure.
2424  */
2425 int
2426 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2427 	       const void *value, size_t value_len, int flags)
2428 {
2429 	handle_t *handle;
2430 	struct super_block *sb = inode->i_sb;
2431 	int error, retries = 0;
2432 	int credits;
2433 
2434 	error = dquot_initialize(inode);
2435 	if (error)
2436 		return error;
2437 
2438 retry:
2439 	error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2440 				       &credits);
2441 	if (error)
2442 		return error;
2443 
2444 	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2445 	if (IS_ERR(handle)) {
2446 		error = PTR_ERR(handle);
2447 	} else {
2448 		int error2;
2449 
2450 		error = ext4_xattr_set_handle(handle, inode, name_index, name,
2451 					      value, value_len, flags);
2452 		error2 = ext4_journal_stop(handle);
2453 		if (error == -ENOSPC &&
2454 		    ext4_should_retry_alloc(sb, &retries))
2455 			goto retry;
2456 		if (error == 0)
2457 			error = error2;
2458 	}
2459 
2460 	return error;
2461 }
2462 
2463 /*
2464  * Shift the EA entries in the inode to create space for the increased
2465  * i_extra_isize.
2466  */
2467 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2468 				     int value_offs_shift, void *to,
2469 				     void *from, size_t n)
2470 {
2471 	struct ext4_xattr_entry *last = entry;
2472 	int new_offs;
2473 
2474 	/* We always shift xattr headers further thus offsets get lower */
2475 	BUG_ON(value_offs_shift > 0);
2476 
2477 	/* Adjust the value offsets of the entries */
2478 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2479 		if (!last->e_value_inum && last->e_value_size) {
2480 			new_offs = le16_to_cpu(last->e_value_offs) +
2481 							value_offs_shift;
2482 			last->e_value_offs = cpu_to_le16(new_offs);
2483 		}
2484 	}
2485 	/* Shift the entries by n bytes */
2486 	memmove(to, from, n);
2487 }
2488 
2489 /*
2490  * Move xattr pointed to by 'entry' from inode into external xattr block
2491  */
2492 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2493 				    struct ext4_inode *raw_inode,
2494 				    struct ext4_xattr_entry *entry)
2495 {
2496 	struct ext4_xattr_ibody_find *is = NULL;
2497 	struct ext4_xattr_block_find *bs = NULL;
2498 	char *buffer = NULL, *b_entry_name = NULL;
2499 	size_t value_size = le32_to_cpu(entry->e_value_size);
2500 	struct ext4_xattr_info i = {
2501 		.value = NULL,
2502 		.value_len = 0,
2503 		.name_index = entry->e_name_index,
2504 		.in_inode = !!entry->e_value_inum,
2505 	};
2506 	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2507 	int error;
2508 
2509 	is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2510 	bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2511 	buffer = kmalloc(value_size, GFP_NOFS);
2512 	b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2513 	if (!is || !bs || !buffer || !b_entry_name) {
2514 		error = -ENOMEM;
2515 		goto out;
2516 	}
2517 
2518 	is->s.not_found = -ENODATA;
2519 	bs->s.not_found = -ENODATA;
2520 	is->iloc.bh = NULL;
2521 	bs->bh = NULL;
2522 
2523 	/* Save the entry name and the entry value */
2524 	if (entry->e_value_inum) {
2525 		error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2526 		if (error)
2527 			goto out;
2528 	} else {
2529 		size_t value_offs = le16_to_cpu(entry->e_value_offs);
2530 		memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2531 	}
2532 
2533 	memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2534 	b_entry_name[entry->e_name_len] = '\0';
2535 	i.name = b_entry_name;
2536 
2537 	error = ext4_get_inode_loc(inode, &is->iloc);
2538 	if (error)
2539 		goto out;
2540 
2541 	error = ext4_xattr_ibody_find(inode, &i, is);
2542 	if (error)
2543 		goto out;
2544 
2545 	/* Remove the chosen entry from the inode */
2546 	error = ext4_xattr_ibody_set(handle, inode, &i, is);
2547 	if (error)
2548 		goto out;
2549 
2550 	i.value = buffer;
2551 	i.value_len = value_size;
2552 	error = ext4_xattr_block_find(inode, &i, bs);
2553 	if (error)
2554 		goto out;
2555 
2556 	/* Add entry which was removed from the inode into the block */
2557 	error = ext4_xattr_block_set(handle, inode, &i, bs);
2558 	if (error)
2559 		goto out;
2560 	error = 0;
2561 out:
2562 	kfree(b_entry_name);
2563 	kfree(buffer);
2564 	if (is)
2565 		brelse(is->iloc.bh);
2566 	kfree(is);
2567 	kfree(bs);
2568 
2569 	return error;
2570 }
2571 
2572 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2573 				       struct ext4_inode *raw_inode,
2574 				       int isize_diff, size_t ifree,
2575 				       size_t bfree, int *total_ino)
2576 {
2577 	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2578 	struct ext4_xattr_entry *small_entry;
2579 	struct ext4_xattr_entry *entry;
2580 	struct ext4_xattr_entry *last;
2581 	unsigned int entry_size;	/* EA entry size */
2582 	unsigned int total_size;	/* EA entry size + value size */
2583 	unsigned int min_total_size;
2584 	int error;
2585 
2586 	while (isize_diff > ifree) {
2587 		entry = NULL;
2588 		small_entry = NULL;
2589 		min_total_size = ~0U;
2590 		last = IFIRST(header);
2591 		/* Find the entry best suited to be pushed into EA block */
2592 		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2593 			total_size = EXT4_XATTR_LEN(last->e_name_len);
2594 			if (!last->e_value_inum)
2595 				total_size += EXT4_XATTR_SIZE(
2596 					       le32_to_cpu(last->e_value_size));
2597 			if (total_size <= bfree &&
2598 			    total_size < min_total_size) {
2599 				if (total_size + ifree < isize_diff) {
2600 					small_entry = last;
2601 				} else {
2602 					entry = last;
2603 					min_total_size = total_size;
2604 				}
2605 			}
2606 		}
2607 
2608 		if (entry == NULL) {
2609 			if (small_entry == NULL)
2610 				return -ENOSPC;
2611 			entry = small_entry;
2612 		}
2613 
2614 		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2615 		total_size = entry_size;
2616 		if (!entry->e_value_inum)
2617 			total_size += EXT4_XATTR_SIZE(
2618 					      le32_to_cpu(entry->e_value_size));
2619 		error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2620 						 entry);
2621 		if (error)
2622 			return error;
2623 
2624 		*total_ino -= entry_size;
2625 		ifree += total_size;
2626 		bfree -= total_size;
2627 	}
2628 
2629 	return 0;
2630 }
2631 
2632 /*
2633  * Expand an inode by new_extra_isize bytes when EAs are present.
2634  * Returns 0 on success or negative error number on failure.
2635  */
2636 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2637 			       struct ext4_inode *raw_inode, handle_t *handle)
2638 {
2639 	struct ext4_xattr_ibody_header *header;
2640 	struct buffer_head *bh;
2641 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2642 	static unsigned int mnt_count;
2643 	size_t min_offs;
2644 	size_t ifree, bfree;
2645 	int total_ino;
2646 	void *base, *end;
2647 	int error = 0, tried_min_extra_isize = 0;
2648 	int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2649 	int isize_diff;	/* How much do we need to grow i_extra_isize */
2650 
2651 retry:
2652 	isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2653 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2654 		return 0;
2655 
2656 	header = IHDR(inode, raw_inode);
2657 
2658 	/*
2659 	 * Check if enough free space is available in the inode to shift the
2660 	 * entries ahead by new_extra_isize.
2661 	 */
2662 
2663 	base = IFIRST(header);
2664 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2665 	min_offs = end - base;
2666 	total_ino = sizeof(struct ext4_xattr_ibody_header);
2667 
2668 	error = xattr_check_inode(inode, header, end);
2669 	if (error)
2670 		goto cleanup;
2671 
2672 	ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2673 	if (ifree >= isize_diff)
2674 		goto shift;
2675 
2676 	/*
2677 	 * Enough free space isn't available in the inode, check if
2678 	 * EA block can hold new_extra_isize bytes.
2679 	 */
2680 	if (EXT4_I(inode)->i_file_acl) {
2681 		bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2682 		error = -EIO;
2683 		if (!bh)
2684 			goto cleanup;
2685 		if (ext4_xattr_check_block(inode, bh)) {
2686 			EXT4_ERROR_INODE(inode, "bad block %llu",
2687 					 EXT4_I(inode)->i_file_acl);
2688 			error = -EFSCORRUPTED;
2689 			brelse(bh);
2690 			goto cleanup;
2691 		}
2692 		base = BHDR(bh);
2693 		end = bh->b_data + bh->b_size;
2694 		min_offs = end - base;
2695 		bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2696 					      NULL);
2697 		brelse(bh);
2698 		if (bfree + ifree < isize_diff) {
2699 			if (!tried_min_extra_isize && s_min_extra_isize) {
2700 				tried_min_extra_isize++;
2701 				new_extra_isize = s_min_extra_isize;
2702 				goto retry;
2703 			}
2704 			error = -ENOSPC;
2705 			goto cleanup;
2706 		}
2707 	} else {
2708 		bfree = inode->i_sb->s_blocksize;
2709 	}
2710 
2711 	error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2712 					    isize_diff, ifree, bfree,
2713 					    &total_ino);
2714 	if (error) {
2715 		if (error == -ENOSPC && !tried_min_extra_isize &&
2716 		    s_min_extra_isize) {
2717 			tried_min_extra_isize++;
2718 			new_extra_isize = s_min_extra_isize;
2719 			goto retry;
2720 		}
2721 		goto cleanup;
2722 	}
2723 shift:
2724 	/* Adjust the offsets and shift the remaining entries ahead */
2725 	ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2726 			- new_extra_isize, (void *)raw_inode +
2727 			EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2728 			(void *)header, total_ino);
2729 	EXT4_I(inode)->i_extra_isize = new_extra_isize;
2730 
2731 cleanup:
2732 	if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2733 		ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2734 			     inode->i_ino);
2735 		mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2736 	}
2737 	return error;
2738 }
2739 
2740 #define EIA_INCR 16 /* must be 2^n */
2741 #define EIA_MASK (EIA_INCR - 1)
2742 
2743 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2744  * If @ea_inode_array is new or full it will be grown and the old
2745  * contents copied over.
2746  */
2747 static int
2748 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2749 			struct inode *inode)
2750 {
2751 	if (*ea_inode_array == NULL) {
2752 		/*
2753 		 * Start with 15 inodes, so it fits into a power-of-two size.
2754 		 * If *ea_inode_array is NULL, this is essentially offsetof()
2755 		 */
2756 		(*ea_inode_array) =
2757 			kmalloc(offsetof(struct ext4_xattr_inode_array,
2758 					 inodes[EIA_MASK]),
2759 				GFP_NOFS);
2760 		if (*ea_inode_array == NULL)
2761 			return -ENOMEM;
2762 		(*ea_inode_array)->count = 0;
2763 	} else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2764 		/* expand the array once all 15 + n * 16 slots are full */
2765 		struct ext4_xattr_inode_array *new_array = NULL;
2766 		int count = (*ea_inode_array)->count;
2767 
2768 		/* if new_array is NULL, this is essentially offsetof() */
2769 		new_array = kmalloc(
2770 				offsetof(struct ext4_xattr_inode_array,
2771 					 inodes[count + EIA_INCR]),
2772 				GFP_NOFS);
2773 		if (new_array == NULL)
2774 			return -ENOMEM;
2775 		memcpy(new_array, *ea_inode_array,
2776 		       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2777 		kfree(*ea_inode_array);
2778 		*ea_inode_array = new_array;
2779 	}
2780 	(*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2781 	return 0;
2782 }
2783 
2784 /*
2785  * ext4_xattr_delete_inode()
2786  *
2787  * Free extended attribute resources associated with this inode. Traverse
2788  * all entries and decrement reference on any xattr inodes associated with this
2789  * inode. This is called immediately before an inode is freed. We have exclusive
2790  * access to the inode. If an orphan inode is deleted it will also release its
2791  * references on xattr block and xattr inodes.
2792  */
2793 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2794 			    struct ext4_xattr_inode_array **ea_inode_array,
2795 			    int extra_credits)
2796 {
2797 	struct buffer_head *bh = NULL;
2798 	struct ext4_xattr_ibody_header *header;
2799 	struct ext4_iloc iloc = { .bh = NULL };
2800 	struct ext4_xattr_entry *entry;
2801 	int error;
2802 
2803 	error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2804 					  NULL /* bh */,
2805 					  false /* dirty */,
2806 					  false /* block_csum */);
2807 	if (error) {
2808 		EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2809 		goto cleanup;
2810 	}
2811 
2812 	if (ext4_has_feature_ea_inode(inode->i_sb) &&
2813 	    ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2814 
2815 		error = ext4_get_inode_loc(inode, &iloc);
2816 		if (error) {
2817 			EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2818 			goto cleanup;
2819 		}
2820 
2821 		error = ext4_journal_get_write_access(handle, iloc.bh);
2822 		if (error) {
2823 			EXT4_ERROR_INODE(inode, "write access (error %d)",
2824 					 error);
2825 			goto cleanup;
2826 		}
2827 
2828 		header = IHDR(inode, ext4_raw_inode(&iloc));
2829 		if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2830 			ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2831 						     IFIRST(header),
2832 						     false /* block_csum */,
2833 						     ea_inode_array,
2834 						     extra_credits,
2835 						     false /* skip_quota */);
2836 	}
2837 
2838 	if (EXT4_I(inode)->i_file_acl) {
2839 		bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2840 		if (!bh) {
2841 			EXT4_ERROR_INODE(inode, "block %llu read error",
2842 					 EXT4_I(inode)->i_file_acl);
2843 			error = -EIO;
2844 			goto cleanup;
2845 		}
2846 		error = ext4_xattr_check_block(inode, bh);
2847 		if (error) {
2848 			EXT4_ERROR_INODE(inode, "bad block %llu (error %d)",
2849 					 EXT4_I(inode)->i_file_acl, error);
2850 			goto cleanup;
2851 		}
2852 
2853 		if (ext4_has_feature_ea_inode(inode->i_sb)) {
2854 			for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2855 			     entry = EXT4_XATTR_NEXT(entry))
2856 				if (entry->e_value_inum)
2857 					ext4_xattr_inode_free_quota(inode,
2858 					      le32_to_cpu(entry->e_value_size));
2859 
2860 		}
2861 
2862 		ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2863 					 extra_credits);
2864 		/*
2865 		 * Update i_file_acl value in the same transaction that releases
2866 		 * block.
2867 		 */
2868 		EXT4_I(inode)->i_file_acl = 0;
2869 		error = ext4_mark_inode_dirty(handle, inode);
2870 		if (error) {
2871 			EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2872 					 error);
2873 			goto cleanup;
2874 		}
2875 	}
2876 	error = 0;
2877 cleanup:
2878 	brelse(iloc.bh);
2879 	brelse(bh);
2880 	return error;
2881 }
2882 
2883 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2884 {
2885 	int idx;
2886 
2887 	if (ea_inode_array == NULL)
2888 		return;
2889 
2890 	for (idx = 0; idx < ea_inode_array->count; ++idx)
2891 		iput(ea_inode_array->inodes[idx]);
2892 	kfree(ea_inode_array);
2893 }
2894 
2895 /*
2896  * ext4_xattr_block_cache_insert()
2897  *
2898  * Create a new entry in the extended attribute block cache, and insert
2899  * it unless such an entry is already in the cache.
2900  *
2901  * Returns 0, or a negative error number on failure.
2902  */
2903 static void
2904 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2905 			      struct buffer_head *bh)
2906 {
2907 	struct ext4_xattr_header *header = BHDR(bh);
2908 	__u32 hash = le32_to_cpu(header->h_hash);
2909 	int reusable = le32_to_cpu(header->h_refcount) <
2910 		       EXT4_XATTR_REFCOUNT_MAX;
2911 	int error;
2912 
2913 	if (!ea_block_cache)
2914 		return;
2915 	error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2916 				      bh->b_blocknr, reusable);
2917 	if (error) {
2918 		if (error == -EBUSY)
2919 			ea_bdebug(bh, "already in cache");
2920 	} else
2921 		ea_bdebug(bh, "inserting [%x]", (int)hash);
2922 }
2923 
2924 /*
2925  * ext4_xattr_cmp()
2926  *
2927  * Compare two extended attribute blocks for equality.
2928  *
2929  * Returns 0 if the blocks are equal, 1 if they differ, and
2930  * a negative error number on errors.
2931  */
2932 static int
2933 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2934 	       struct ext4_xattr_header *header2)
2935 {
2936 	struct ext4_xattr_entry *entry1, *entry2;
2937 
2938 	entry1 = ENTRY(header1+1);
2939 	entry2 = ENTRY(header2+1);
2940 	while (!IS_LAST_ENTRY(entry1)) {
2941 		if (IS_LAST_ENTRY(entry2))
2942 			return 1;
2943 		if (entry1->e_hash != entry2->e_hash ||
2944 		    entry1->e_name_index != entry2->e_name_index ||
2945 		    entry1->e_name_len != entry2->e_name_len ||
2946 		    entry1->e_value_size != entry2->e_value_size ||
2947 		    entry1->e_value_inum != entry2->e_value_inum ||
2948 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
2949 			return 1;
2950 		if (!entry1->e_value_inum &&
2951 		    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
2952 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
2953 			   le32_to_cpu(entry1->e_value_size)))
2954 			return 1;
2955 
2956 		entry1 = EXT4_XATTR_NEXT(entry1);
2957 		entry2 = EXT4_XATTR_NEXT(entry2);
2958 	}
2959 	if (!IS_LAST_ENTRY(entry2))
2960 		return 1;
2961 	return 0;
2962 }
2963 
2964 /*
2965  * ext4_xattr_block_cache_find()
2966  *
2967  * Find an identical extended attribute block.
2968  *
2969  * Returns a pointer to the block found, or NULL if such a block was
2970  * not found or an error occurred.
2971  */
2972 static struct buffer_head *
2973 ext4_xattr_block_cache_find(struct inode *inode,
2974 			    struct ext4_xattr_header *header,
2975 			    struct mb_cache_entry **pce)
2976 {
2977 	__u32 hash = le32_to_cpu(header->h_hash);
2978 	struct mb_cache_entry *ce;
2979 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
2980 
2981 	if (!ea_block_cache)
2982 		return NULL;
2983 	if (!header->h_hash)
2984 		return NULL;  /* never share */
2985 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
2986 	ce = mb_cache_entry_find_first(ea_block_cache, hash);
2987 	while (ce) {
2988 		struct buffer_head *bh;
2989 
2990 		bh = sb_bread(inode->i_sb, ce->e_value);
2991 		if (!bh) {
2992 			EXT4_ERROR_INODE(inode, "block %lu read error",
2993 					 (unsigned long)ce->e_value);
2994 		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
2995 			*pce = ce;
2996 			return bh;
2997 		}
2998 		brelse(bh);
2999 		ce = mb_cache_entry_find_next(ea_block_cache, ce);
3000 	}
3001 	return NULL;
3002 }
3003 
3004 #define NAME_HASH_SHIFT 5
3005 #define VALUE_HASH_SHIFT 16
3006 
3007 /*
3008  * ext4_xattr_hash_entry()
3009  *
3010  * Compute the hash of an extended attribute.
3011  */
3012 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3013 				    size_t value_count)
3014 {
3015 	__u32 hash = 0;
3016 
3017 	while (name_len--) {
3018 		hash = (hash << NAME_HASH_SHIFT) ^
3019 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3020 		       *name++;
3021 	}
3022 	while (value_count--) {
3023 		hash = (hash << VALUE_HASH_SHIFT) ^
3024 		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3025 		       le32_to_cpu(*value++);
3026 	}
3027 	return cpu_to_le32(hash);
3028 }
3029 
3030 #undef NAME_HASH_SHIFT
3031 #undef VALUE_HASH_SHIFT
3032 
3033 #define BLOCK_HASH_SHIFT 16
3034 
3035 /*
3036  * ext4_xattr_rehash()
3037  *
3038  * Re-compute the extended attribute hash value after an entry has changed.
3039  */
3040 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3041 {
3042 	struct ext4_xattr_entry *here;
3043 	__u32 hash = 0;
3044 
3045 	here = ENTRY(header+1);
3046 	while (!IS_LAST_ENTRY(here)) {
3047 		if (!here->e_hash) {
3048 			/* Block is not shared if an entry's hash value == 0 */
3049 			hash = 0;
3050 			break;
3051 		}
3052 		hash = (hash << BLOCK_HASH_SHIFT) ^
3053 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3054 		       le32_to_cpu(here->e_hash);
3055 		here = EXT4_XATTR_NEXT(here);
3056 	}
3057 	header->h_hash = cpu_to_le32(hash);
3058 }
3059 
3060 #undef BLOCK_HASH_SHIFT
3061 
3062 #define	HASH_BUCKET_BITS	10
3063 
3064 struct mb_cache *
3065 ext4_xattr_create_cache(void)
3066 {
3067 	return mb_cache_create(HASH_BUCKET_BITS);
3068 }
3069 
3070 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3071 {
3072 	if (cache)
3073 		mb_cache_destroy(cache);
3074 }
3075 
3076