xref: /openbmc/linux/fs/ext4/xattr.c (revision e789b7fd)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/ext4/xattr.c
4  *
5  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
6  *
7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9  * Extended attributes for symlinks and special files added per
10  *  suggestion of Luka Renko <luka.renko@hermes.si>.
11  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
12  *  Red Hat Inc.
13  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14  *  and Andreas Gruenbacher <agruen@suse.de>.
15  */
16 
17 /*
18  * Extended attributes are stored directly in inodes (on file systems with
19  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
20  * field contains the block number if an inode uses an additional block. All
21  * attributes must fit in the inode and one additional block. Blocks that
22  * contain the identical set of attributes may be shared among several inodes.
23  * Identical blocks are detected by keeping a cache of blocks that have
24  * recently been accessed.
25  *
26  * The attributes in inodes and on blocks have a different header; the entries
27  * are stored in the same format:
28  *
29  *   +------------------+
30  *   | header           |
31  *   | entry 1          | |
32  *   | entry 2          | | growing downwards
33  *   | entry 3          | v
34  *   | four null bytes  |
35  *   | . . .            |
36  *   | value 1          | ^
37  *   | value 3          | | growing upwards
38  *   | value 2          | |
39  *   +------------------+
40  *
41  * The header is followed by multiple entry descriptors. In disk blocks, the
42  * entry descriptors are kept sorted. In inodes, they are unsorted. The
43  * attribute values are aligned to the end of the block in no specific order.
44  *
45  * Locking strategy
46  * ----------------
47  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
48  * EA blocks are only changed if they are exclusive to an inode, so
49  * holding xattr_sem also means that nothing but the EA block's reference
50  * count can change. Multiple writers to the same block are synchronized
51  * by the buffer lock.
52  */
53 
54 #include <linux/init.h>
55 #include <linux/fs.h>
56 #include <linux/slab.h>
57 #include <linux/mbcache.h>
58 #include <linux/quotaops.h>
59 #include <linux/iversion.h>
60 #include "ext4_jbd2.h"
61 #include "ext4.h"
62 #include "xattr.h"
63 #include "acl.h"
64 
65 #ifdef EXT4_XATTR_DEBUG
66 # define ea_idebug(inode, fmt, ...)					\
67 	printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",			\
68 	       inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
69 # define ea_bdebug(bh, fmt, ...)					\
70 	printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",			\
71 	       bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
72 #else
73 # define ea_idebug(inode, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
74 # define ea_bdebug(bh, fmt, ...)	no_printk(fmt, ##__VA_ARGS__)
75 #endif
76 
77 static void ext4_xattr_block_cache_insert(struct mb_cache *,
78 					  struct buffer_head *);
79 static struct buffer_head *
80 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
81 			    struct mb_cache_entry **);
82 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
83 				    size_t value_count);
84 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value,
85 				    size_t value_count);
86 static void ext4_xattr_rehash(struct ext4_xattr_header *);
87 
88 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
89 	[EXT4_XATTR_INDEX_USER]		     = &ext4_xattr_user_handler,
90 #ifdef CONFIG_EXT4_FS_POSIX_ACL
91 	[EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &nop_posix_acl_access,
92 	[EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &nop_posix_acl_default,
93 #endif
94 	[EXT4_XATTR_INDEX_TRUSTED]	     = &ext4_xattr_trusted_handler,
95 #ifdef CONFIG_EXT4_FS_SECURITY
96 	[EXT4_XATTR_INDEX_SECURITY]	     = &ext4_xattr_security_handler,
97 #endif
98 	[EXT4_XATTR_INDEX_HURD]		     = &ext4_xattr_hurd_handler,
99 };
100 
101 const struct xattr_handler *ext4_xattr_handlers[] = {
102 	&ext4_xattr_user_handler,
103 	&ext4_xattr_trusted_handler,
104 #ifdef CONFIG_EXT4_FS_SECURITY
105 	&ext4_xattr_security_handler,
106 #endif
107 	&ext4_xattr_hurd_handler,
108 	NULL
109 };
110 
111 #define EA_BLOCK_CACHE(inode)	(((struct ext4_sb_info *) \
112 				inode->i_sb->s_fs_info)->s_ea_block_cache)
113 
114 #define EA_INODE_CACHE(inode)	(((struct ext4_sb_info *) \
115 				inode->i_sb->s_fs_info)->s_ea_inode_cache)
116 
117 static int
118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119 			struct inode *inode);
120 
121 #ifdef CONFIG_LOCKDEP
ext4_xattr_inode_set_class(struct inode * ea_inode)122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
123 {
124 	struct ext4_inode_info *ei = EXT4_I(ea_inode);
125 
126 	lockdep_set_subclass(&ea_inode->i_rwsem, 1);
127 	(void) ei;	/* shut up clang warning if !CONFIG_LOCKDEP */
128 	lockdep_set_subclass(&ei->i_data_sem, I_DATA_SEM_EA);
129 }
130 #endif
131 
ext4_xattr_block_csum(struct inode * inode,sector_t block_nr,struct ext4_xattr_header * hdr)132 static __le32 ext4_xattr_block_csum(struct inode *inode,
133 				    sector_t block_nr,
134 				    struct ext4_xattr_header *hdr)
135 {
136 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
137 	__u32 csum;
138 	__le64 dsk_block_nr = cpu_to_le64(block_nr);
139 	__u32 dummy_csum = 0;
140 	int offset = offsetof(struct ext4_xattr_header, h_checksum);
141 
142 	csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
143 			   sizeof(dsk_block_nr));
144 	csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
145 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
146 	offset += sizeof(dummy_csum);
147 	csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
148 			   EXT4_BLOCK_SIZE(inode->i_sb) - offset);
149 
150 	return cpu_to_le32(csum);
151 }
152 
ext4_xattr_block_csum_verify(struct inode * inode,struct buffer_head * bh)153 static int ext4_xattr_block_csum_verify(struct inode *inode,
154 					struct buffer_head *bh)
155 {
156 	struct ext4_xattr_header *hdr = BHDR(bh);
157 	int ret = 1;
158 
159 	if (ext4_has_metadata_csum(inode->i_sb)) {
160 		lock_buffer(bh);
161 		ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
162 							bh->b_blocknr, hdr));
163 		unlock_buffer(bh);
164 	}
165 	return ret;
166 }
167 
ext4_xattr_block_csum_set(struct inode * inode,struct buffer_head * bh)168 static void ext4_xattr_block_csum_set(struct inode *inode,
169 				      struct buffer_head *bh)
170 {
171 	if (ext4_has_metadata_csum(inode->i_sb))
172 		BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
173 						bh->b_blocknr, BHDR(bh));
174 }
175 
ext4_xattr_prefix(int name_index,struct dentry * dentry)176 static inline const char *ext4_xattr_prefix(int name_index,
177 					    struct dentry *dentry)
178 {
179 	const struct xattr_handler *handler = NULL;
180 
181 	if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
182 		handler = ext4_xattr_handler_map[name_index];
183 
184 	if (!xattr_handler_can_list(handler, dentry))
185 		return NULL;
186 
187 	return xattr_prefix(handler);
188 }
189 
190 static int
check_xattrs(struct inode * inode,struct buffer_head * bh,struct ext4_xattr_entry * entry,void * end,void * value_start,const char * function,unsigned int line)191 check_xattrs(struct inode *inode, struct buffer_head *bh,
192 	     struct ext4_xattr_entry *entry, void *end, void *value_start,
193 	     const char *function, unsigned int line)
194 {
195 	struct ext4_xattr_entry *e = entry;
196 	int err = -EFSCORRUPTED;
197 	char *err_str;
198 
199 	if (bh) {
200 		if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
201 		    BHDR(bh)->h_blocks != cpu_to_le32(1)) {
202 			err_str = "invalid header";
203 			goto errout;
204 		}
205 		if (buffer_verified(bh))
206 			return 0;
207 		if (!ext4_xattr_block_csum_verify(inode, bh)) {
208 			err = -EFSBADCRC;
209 			err_str = "invalid checksum";
210 			goto errout;
211 		}
212 	} else {
213 		struct ext4_xattr_ibody_header *header = value_start;
214 
215 		header -= 1;
216 		if (end - (void *)header < sizeof(*header) + sizeof(u32)) {
217 			err_str = "in-inode xattr block too small";
218 			goto errout;
219 		}
220 		if (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
221 			err_str = "bad magic number in in-inode xattr";
222 			goto errout;
223 		}
224 	}
225 
226 	/* Find the end of the names list */
227 	while (!IS_LAST_ENTRY(e)) {
228 		struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
229 		if ((void *)next >= end) {
230 			err_str = "e_name out of bounds";
231 			goto errout;
232 		}
233 		if (strnlen(e->e_name, e->e_name_len) != e->e_name_len) {
234 			err_str = "bad e_name length";
235 			goto errout;
236 		}
237 		e = next;
238 	}
239 
240 	/* Check the values */
241 	while (!IS_LAST_ENTRY(entry)) {
242 		u32 size = le32_to_cpu(entry->e_value_size);
243 		unsigned long ea_ino = le32_to_cpu(entry->e_value_inum);
244 
245 		if (!ext4_has_feature_ea_inode(inode->i_sb) && ea_ino) {
246 			err_str = "ea_inode specified without ea_inode feature enabled";
247 			goto errout;
248 		}
249 		if (ea_ino && ((ea_ino == EXT4_ROOT_INO) ||
250 			       !ext4_valid_inum(inode->i_sb, ea_ino))) {
251 			err_str = "invalid ea_ino";
252 			goto errout;
253 		}
254 		if (size > EXT4_XATTR_SIZE_MAX) {
255 			err_str = "e_value size too large";
256 			goto errout;
257 		}
258 
259 		if (size != 0 && entry->e_value_inum == 0) {
260 			u16 offs = le16_to_cpu(entry->e_value_offs);
261 			void *value;
262 
263 			/*
264 			 * The value cannot overlap the names, and the value
265 			 * with padding cannot extend beyond 'end'.  Check both
266 			 * the padded and unpadded sizes, since the size may
267 			 * overflow to 0 when adding padding.
268 			 */
269 			if (offs > end - value_start) {
270 				err_str = "e_value out of bounds";
271 				goto errout;
272 			}
273 			value = value_start + offs;
274 			if (value < (void *)e + sizeof(u32) ||
275 			    size > end - value ||
276 			    EXT4_XATTR_SIZE(size) > end - value) {
277 				err_str = "overlapping e_value ";
278 				goto errout;
279 			}
280 		}
281 		entry = EXT4_XATTR_NEXT(entry);
282 	}
283 	if (bh)
284 		set_buffer_verified(bh);
285 	return 0;
286 
287 errout:
288 	if (bh)
289 		__ext4_error_inode(inode, function, line, 0, -err,
290 				   "corrupted xattr block %llu: %s",
291 				   (unsigned long long) bh->b_blocknr,
292 				   err_str);
293 	else
294 		__ext4_error_inode(inode, function, line, 0, -err,
295 				   "corrupted in-inode xattr: %s", err_str);
296 	return err;
297 }
298 
299 static inline int
__ext4_xattr_check_block(struct inode * inode,struct buffer_head * bh,const char * function,unsigned int line)300 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
301 			 const char *function, unsigned int line)
302 {
303 	return check_xattrs(inode, bh, BFIRST(bh), bh->b_data + bh->b_size,
304 			    bh->b_data, function, line);
305 }
306 
307 #define ext4_xattr_check_block(inode, bh) \
308 	__ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
309 
310 
311 static inline int
__xattr_check_inode(struct inode * inode,struct ext4_xattr_ibody_header * header,void * end,const char * function,unsigned int line)312 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
313 			 void *end, const char *function, unsigned int line)
314 {
315 	return check_xattrs(inode, NULL, IFIRST(header), end, IFIRST(header),
316 			    function, line);
317 }
318 
319 #define xattr_check_inode(inode, header, end) \
320 	__xattr_check_inode((inode), (header), (end), __func__, __LINE__)
321 
322 static int
xattr_find_entry(struct inode * inode,struct ext4_xattr_entry ** pentry,void * end,int name_index,const char * name,int sorted)323 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
324 		 void *end, int name_index, const char *name, int sorted)
325 {
326 	struct ext4_xattr_entry *entry, *next;
327 	size_t name_len;
328 	int cmp = 1;
329 
330 	if (name == NULL)
331 		return -EINVAL;
332 	name_len = strlen(name);
333 	for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
334 		next = EXT4_XATTR_NEXT(entry);
335 		if ((void *) next >= end) {
336 			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
337 			return -EFSCORRUPTED;
338 		}
339 		cmp = name_index - entry->e_name_index;
340 		if (!cmp)
341 			cmp = name_len - entry->e_name_len;
342 		if (!cmp)
343 			cmp = memcmp(name, entry->e_name, name_len);
344 		if (cmp <= 0 && (sorted || cmp == 0))
345 			break;
346 	}
347 	*pentry = entry;
348 	return cmp ? -ENODATA : 0;
349 }
350 
351 static u32
ext4_xattr_inode_hash(struct ext4_sb_info * sbi,const void * buffer,size_t size)352 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
353 {
354 	return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
355 }
356 
ext4_xattr_inode_get_ref(struct inode * ea_inode)357 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
358 {
359 	return ((u64) inode_get_ctime(ea_inode).tv_sec << 32) |
360 		(u32) inode_peek_iversion_raw(ea_inode);
361 }
362 
ext4_xattr_inode_set_ref(struct inode * ea_inode,u64 ref_count)363 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
364 {
365 	inode_set_ctime(ea_inode, (u32)(ref_count >> 32), 0);
366 	inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
367 }
368 
ext4_xattr_inode_get_hash(struct inode * ea_inode)369 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
370 {
371 	return (u32)ea_inode->i_atime.tv_sec;
372 }
373 
ext4_xattr_inode_set_hash(struct inode * ea_inode,u32 hash)374 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
375 {
376 	ea_inode->i_atime.tv_sec = hash;
377 }
378 
379 /*
380  * Read the EA value from an inode.
381  */
ext4_xattr_inode_read(struct inode * ea_inode,void * buf,size_t size)382 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
383 {
384 	int blocksize = 1 << ea_inode->i_blkbits;
385 	int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
386 	int tail_size = (size % blocksize) ?: blocksize;
387 	struct buffer_head *bhs_inline[8];
388 	struct buffer_head **bhs = bhs_inline;
389 	int i, ret;
390 
391 	if (bh_count > ARRAY_SIZE(bhs_inline)) {
392 		bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
393 		if (!bhs)
394 			return -ENOMEM;
395 	}
396 
397 	ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
398 			       true /* wait */, bhs);
399 	if (ret)
400 		goto free_bhs;
401 
402 	for (i = 0; i < bh_count; i++) {
403 		/* There shouldn't be any holes in ea_inode. */
404 		if (!bhs[i]) {
405 			ret = -EFSCORRUPTED;
406 			goto put_bhs;
407 		}
408 		memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
409 		       i < bh_count - 1 ? blocksize : tail_size);
410 	}
411 	ret = 0;
412 put_bhs:
413 	for (i = 0; i < bh_count; i++)
414 		brelse(bhs[i]);
415 free_bhs:
416 	if (bhs != bhs_inline)
417 		kfree(bhs);
418 	return ret;
419 }
420 
421 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
422 
ext4_xattr_inode_iget(struct inode * parent,unsigned long ea_ino,u32 ea_inode_hash,struct inode ** ea_inode)423 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
424 				 u32 ea_inode_hash, struct inode **ea_inode)
425 {
426 	struct inode *inode;
427 	int err;
428 
429 	/*
430 	 * We have to check for this corruption early as otherwise
431 	 * iget_locked() could wait indefinitely for the state of our
432 	 * parent inode.
433 	 */
434 	if (parent->i_ino == ea_ino) {
435 		ext4_error(parent->i_sb,
436 			   "Parent and EA inode have the same ino %lu", ea_ino);
437 		return -EFSCORRUPTED;
438 	}
439 
440 	inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_EA_INODE);
441 	if (IS_ERR(inode)) {
442 		err = PTR_ERR(inode);
443 		ext4_error(parent->i_sb,
444 			   "error while reading EA inode %lu err=%d", ea_ino,
445 			   err);
446 		return err;
447 	}
448 	ext4_xattr_inode_set_class(inode);
449 
450 	/*
451 	 * Check whether this is an old Lustre-style xattr inode. Lustre
452 	 * implementation does not have hash validation, rather it has a
453 	 * backpointer from ea_inode to the parent inode.
454 	 */
455 	if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
456 	    EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
457 	    inode->i_generation == parent->i_generation) {
458 		ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
459 		ext4_xattr_inode_set_ref(inode, 1);
460 	} else {
461 		inode_lock_nested(inode, I_MUTEX_XATTR);
462 		inode->i_flags |= S_NOQUOTA;
463 		inode_unlock(inode);
464 	}
465 
466 	*ea_inode = inode;
467 	return 0;
468 }
469 
470 /* Remove entry from mbcache when EA inode is getting evicted */
ext4_evict_ea_inode(struct inode * inode)471 void ext4_evict_ea_inode(struct inode *inode)
472 {
473 	struct mb_cache_entry *oe;
474 
475 	if (!EA_INODE_CACHE(inode))
476 		return;
477 	/* Wait for entry to get unused so that we can remove it */
478 	while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
479 			ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
480 		mb_cache_entry_wait_unused(oe);
481 		mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
482 	}
483 }
484 
485 static int
ext4_xattr_inode_verify_hashes(struct inode * ea_inode,struct ext4_xattr_entry * entry,void * buffer,size_t size)486 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
487 			       struct ext4_xattr_entry *entry, void *buffer,
488 			       size_t size)
489 {
490 	u32 hash;
491 
492 	/* Verify stored hash matches calculated hash. */
493 	hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
494 	if (hash != ext4_xattr_inode_get_hash(ea_inode))
495 		return -EFSCORRUPTED;
496 
497 	if (entry) {
498 		__le32 e_hash, tmp_data;
499 
500 		/* Verify entry hash. */
501 		tmp_data = cpu_to_le32(hash);
502 		e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
503 					       &tmp_data, 1);
504 		/* All good? */
505 		if (e_hash == entry->e_hash)
506 			return 0;
507 
508 		/*
509 		 * Not good. Maybe the entry hash was calculated
510 		 * using the buggy signed char version?
511 		 */
512 		e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
513 							&tmp_data, 1);
514 		/* Still no match - bad */
515 		if (e_hash != entry->e_hash)
516 			return -EFSCORRUPTED;
517 
518 		/* Let people know about old hash */
519 		pr_warn_once("ext4: filesystem with signed xattr name hash");
520 	}
521 	return 0;
522 }
523 
524 /*
525  * Read xattr value from the EA inode.
526  */
527 static int
ext4_xattr_inode_get(struct inode * inode,struct ext4_xattr_entry * entry,void * buffer,size_t size)528 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
529 		     void *buffer, size_t size)
530 {
531 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
532 	struct inode *ea_inode;
533 	int err;
534 
535 	err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
536 				    le32_to_cpu(entry->e_hash), &ea_inode);
537 	if (err) {
538 		ea_inode = NULL;
539 		goto out;
540 	}
541 
542 	if (i_size_read(ea_inode) != size) {
543 		ext4_warning_inode(ea_inode,
544 				   "ea_inode file size=%llu entry size=%zu",
545 				   i_size_read(ea_inode), size);
546 		err = -EFSCORRUPTED;
547 		goto out;
548 	}
549 
550 	err = ext4_xattr_inode_read(ea_inode, buffer, size);
551 	if (err)
552 		goto out;
553 
554 	if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
555 		err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
556 						     size);
557 		if (err) {
558 			ext4_warning_inode(ea_inode,
559 					   "EA inode hash validation failed");
560 			goto out;
561 		}
562 
563 		if (ea_inode_cache)
564 			mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
565 					ext4_xattr_inode_get_hash(ea_inode),
566 					ea_inode->i_ino, true /* reusable */);
567 	}
568 out:
569 	iput(ea_inode);
570 	return err;
571 }
572 
573 static int
ext4_xattr_block_get(struct inode * inode,int name_index,const char * name,void * buffer,size_t buffer_size)574 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
575 		     void *buffer, size_t buffer_size)
576 {
577 	struct buffer_head *bh = NULL;
578 	struct ext4_xattr_entry *entry;
579 	size_t size;
580 	void *end;
581 	int error;
582 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
583 
584 	ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
585 		  name_index, name, buffer, (long)buffer_size);
586 
587 	if (!EXT4_I(inode)->i_file_acl)
588 		return -ENODATA;
589 	ea_idebug(inode, "reading block %llu",
590 		  (unsigned long long)EXT4_I(inode)->i_file_acl);
591 	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
592 	if (IS_ERR(bh))
593 		return PTR_ERR(bh);
594 	ea_bdebug(bh, "b_count=%d, refcount=%d",
595 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
596 	error = ext4_xattr_check_block(inode, bh);
597 	if (error)
598 		goto cleanup;
599 	ext4_xattr_block_cache_insert(ea_block_cache, bh);
600 	entry = BFIRST(bh);
601 	end = bh->b_data + bh->b_size;
602 	error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
603 	if (error)
604 		goto cleanup;
605 	size = le32_to_cpu(entry->e_value_size);
606 	error = -ERANGE;
607 	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
608 		goto cleanup;
609 	if (buffer) {
610 		if (size > buffer_size)
611 			goto cleanup;
612 		if (entry->e_value_inum) {
613 			error = ext4_xattr_inode_get(inode, entry, buffer,
614 						     size);
615 			if (error)
616 				goto cleanup;
617 		} else {
618 			u16 offset = le16_to_cpu(entry->e_value_offs);
619 			void *p = bh->b_data + offset;
620 
621 			if (unlikely(p + size > end))
622 				goto cleanup;
623 			memcpy(buffer, p, size);
624 		}
625 	}
626 	error = size;
627 
628 cleanup:
629 	brelse(bh);
630 	return error;
631 }
632 
633 int
ext4_xattr_ibody_get(struct inode * inode,int name_index,const char * name,void * buffer,size_t buffer_size)634 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
635 		     void *buffer, size_t buffer_size)
636 {
637 	struct ext4_xattr_ibody_header *header;
638 	struct ext4_xattr_entry *entry;
639 	struct ext4_inode *raw_inode;
640 	struct ext4_iloc iloc;
641 	size_t size;
642 	void *end;
643 	int error;
644 
645 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
646 		return -ENODATA;
647 	error = ext4_get_inode_loc(inode, &iloc);
648 	if (error)
649 		return error;
650 	raw_inode = ext4_raw_inode(&iloc);
651 	header = IHDR(inode, raw_inode);
652 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
653 	error = xattr_check_inode(inode, header, end);
654 	if (error)
655 		goto cleanup;
656 	entry = IFIRST(header);
657 	error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
658 	if (error)
659 		goto cleanup;
660 	size = le32_to_cpu(entry->e_value_size);
661 	error = -ERANGE;
662 	if (unlikely(size > EXT4_XATTR_SIZE_MAX))
663 		goto cleanup;
664 	if (buffer) {
665 		if (size > buffer_size)
666 			goto cleanup;
667 		if (entry->e_value_inum) {
668 			error = ext4_xattr_inode_get(inode, entry, buffer,
669 						     size);
670 			if (error)
671 				goto cleanup;
672 		} else {
673 			u16 offset = le16_to_cpu(entry->e_value_offs);
674 			void *p = (void *)IFIRST(header) + offset;
675 
676 			if (unlikely(p + size > end))
677 				goto cleanup;
678 			memcpy(buffer, p, size);
679 		}
680 	}
681 	error = size;
682 
683 cleanup:
684 	brelse(iloc.bh);
685 	return error;
686 }
687 
688 /*
689  * ext4_xattr_get()
690  *
691  * Copy an extended attribute into the buffer
692  * provided, or compute the buffer size required.
693  * Buffer is NULL to compute the size of the buffer required.
694  *
695  * Returns a negative error number on failure, or the number of bytes
696  * used / required on success.
697  */
698 int
ext4_xattr_get(struct inode * inode,int name_index,const char * name,void * buffer,size_t buffer_size)699 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
700 	       void *buffer, size_t buffer_size)
701 {
702 	int error;
703 
704 	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
705 		return -EIO;
706 
707 	if (strlen(name) > 255)
708 		return -ERANGE;
709 
710 	down_read(&EXT4_I(inode)->xattr_sem);
711 	error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
712 				     buffer_size);
713 	if (error == -ENODATA)
714 		error = ext4_xattr_block_get(inode, name_index, name, buffer,
715 					     buffer_size);
716 	up_read(&EXT4_I(inode)->xattr_sem);
717 	return error;
718 }
719 
720 static int
ext4_xattr_list_entries(struct dentry * dentry,struct ext4_xattr_entry * entry,char * buffer,size_t buffer_size)721 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
722 			char *buffer, size_t buffer_size)
723 {
724 	size_t rest = buffer_size;
725 
726 	for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
727 		const char *prefix;
728 
729 		prefix = ext4_xattr_prefix(entry->e_name_index, dentry);
730 		if (prefix) {
731 			size_t prefix_len = strlen(prefix);
732 			size_t size = prefix_len + entry->e_name_len + 1;
733 
734 			if (buffer) {
735 				if (size > rest)
736 					return -ERANGE;
737 				memcpy(buffer, prefix, prefix_len);
738 				buffer += prefix_len;
739 				memcpy(buffer, entry->e_name, entry->e_name_len);
740 				buffer += entry->e_name_len;
741 				*buffer++ = 0;
742 			}
743 			rest -= size;
744 		}
745 	}
746 	return buffer_size - rest;  /* total size */
747 }
748 
749 static int
ext4_xattr_block_list(struct dentry * dentry,char * buffer,size_t buffer_size)750 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
751 {
752 	struct inode *inode = d_inode(dentry);
753 	struct buffer_head *bh = NULL;
754 	int error;
755 
756 	ea_idebug(inode, "buffer=%p, buffer_size=%ld",
757 		  buffer, (long)buffer_size);
758 
759 	if (!EXT4_I(inode)->i_file_acl)
760 		return 0;
761 	ea_idebug(inode, "reading block %llu",
762 		  (unsigned long long)EXT4_I(inode)->i_file_acl);
763 	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
764 	if (IS_ERR(bh))
765 		return PTR_ERR(bh);
766 	ea_bdebug(bh, "b_count=%d, refcount=%d",
767 		atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
768 	error = ext4_xattr_check_block(inode, bh);
769 	if (error)
770 		goto cleanup;
771 	ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
772 	error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
773 					buffer_size);
774 cleanup:
775 	brelse(bh);
776 	return error;
777 }
778 
779 static int
ext4_xattr_ibody_list(struct dentry * dentry,char * buffer,size_t buffer_size)780 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
781 {
782 	struct inode *inode = d_inode(dentry);
783 	struct ext4_xattr_ibody_header *header;
784 	struct ext4_inode *raw_inode;
785 	struct ext4_iloc iloc;
786 	void *end;
787 	int error;
788 
789 	if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
790 		return 0;
791 	error = ext4_get_inode_loc(inode, &iloc);
792 	if (error)
793 		return error;
794 	raw_inode = ext4_raw_inode(&iloc);
795 	header = IHDR(inode, raw_inode);
796 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
797 	error = xattr_check_inode(inode, header, end);
798 	if (error)
799 		goto cleanup;
800 	error = ext4_xattr_list_entries(dentry, IFIRST(header),
801 					buffer, buffer_size);
802 
803 cleanup:
804 	brelse(iloc.bh);
805 	return error;
806 }
807 
808 /*
809  * Inode operation listxattr()
810  *
811  * d_inode(dentry)->i_rwsem: don't care
812  *
813  * Copy a list of attribute names into the buffer
814  * provided, or compute the buffer size required.
815  * Buffer is NULL to compute the size of the buffer required.
816  *
817  * Returns a negative error number on failure, or the number of bytes
818  * used / required on success.
819  */
820 ssize_t
ext4_listxattr(struct dentry * dentry,char * buffer,size_t buffer_size)821 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
822 {
823 	int ret, ret2;
824 
825 	down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
826 	ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
827 	if (ret < 0)
828 		goto errout;
829 	if (buffer) {
830 		buffer += ret;
831 		buffer_size -= ret;
832 	}
833 	ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
834 	if (ret < 0)
835 		goto errout;
836 	ret += ret2;
837 errout:
838 	up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
839 	return ret;
840 }
841 
842 /*
843  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
844  * not set, set it.
845  */
ext4_xattr_update_super_block(handle_t * handle,struct super_block * sb)846 static void ext4_xattr_update_super_block(handle_t *handle,
847 					  struct super_block *sb)
848 {
849 	if (ext4_has_feature_xattr(sb))
850 		return;
851 
852 	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
853 	if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
854 					  EXT4_JTR_NONE) == 0) {
855 		lock_buffer(EXT4_SB(sb)->s_sbh);
856 		ext4_set_feature_xattr(sb);
857 		ext4_superblock_csum_set(sb);
858 		unlock_buffer(EXT4_SB(sb)->s_sbh);
859 		ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
860 	}
861 }
862 
ext4_get_inode_usage(struct inode * inode,qsize_t * usage)863 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
864 {
865 	struct ext4_iloc iloc = { .bh = NULL };
866 	struct buffer_head *bh = NULL;
867 	struct ext4_inode *raw_inode;
868 	struct ext4_xattr_ibody_header *header;
869 	struct ext4_xattr_entry *entry;
870 	qsize_t ea_inode_refs = 0;
871 	void *end;
872 	int ret;
873 
874 	lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
875 
876 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
877 		ret = ext4_get_inode_loc(inode, &iloc);
878 		if (ret)
879 			goto out;
880 		raw_inode = ext4_raw_inode(&iloc);
881 		header = IHDR(inode, raw_inode);
882 		end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
883 		ret = xattr_check_inode(inode, header, end);
884 		if (ret)
885 			goto out;
886 
887 		for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
888 		     entry = EXT4_XATTR_NEXT(entry))
889 			if (entry->e_value_inum)
890 				ea_inode_refs++;
891 	}
892 
893 	if (EXT4_I(inode)->i_file_acl) {
894 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
895 		if (IS_ERR(bh)) {
896 			ret = PTR_ERR(bh);
897 			bh = NULL;
898 			goto out;
899 		}
900 
901 		ret = ext4_xattr_check_block(inode, bh);
902 		if (ret)
903 			goto out;
904 
905 		for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
906 		     entry = EXT4_XATTR_NEXT(entry))
907 			if (entry->e_value_inum)
908 				ea_inode_refs++;
909 	}
910 	*usage = ea_inode_refs + 1;
911 	ret = 0;
912 out:
913 	brelse(iloc.bh);
914 	brelse(bh);
915 	return ret;
916 }
917 
round_up_cluster(struct inode * inode,size_t length)918 static inline size_t round_up_cluster(struct inode *inode, size_t length)
919 {
920 	struct super_block *sb = inode->i_sb;
921 	size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
922 				    inode->i_blkbits);
923 	size_t mask = ~(cluster_size - 1);
924 
925 	return (length + cluster_size - 1) & mask;
926 }
927 
ext4_xattr_inode_alloc_quota(struct inode * inode,size_t len)928 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
929 {
930 	int err;
931 
932 	err = dquot_alloc_inode(inode);
933 	if (err)
934 		return err;
935 	err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
936 	if (err)
937 		dquot_free_inode(inode);
938 	return err;
939 }
940 
ext4_xattr_inode_free_quota(struct inode * parent,struct inode * ea_inode,size_t len)941 static void ext4_xattr_inode_free_quota(struct inode *parent,
942 					struct inode *ea_inode,
943 					size_t len)
944 {
945 	if (ea_inode &&
946 	    ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
947 		return;
948 	dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
949 	dquot_free_inode(parent);
950 }
951 
__ext4_xattr_set_credits(struct super_block * sb,struct inode * inode,struct buffer_head * block_bh,size_t value_len,bool is_create)952 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
953 			     struct buffer_head *block_bh, size_t value_len,
954 			     bool is_create)
955 {
956 	int credits;
957 	int blocks;
958 
959 	/*
960 	 * 1) Owner inode update
961 	 * 2) Ref count update on old xattr block
962 	 * 3) new xattr block
963 	 * 4) block bitmap update for new xattr block
964 	 * 5) group descriptor for new xattr block
965 	 * 6) block bitmap update for old xattr block
966 	 * 7) group descriptor for old block
967 	 *
968 	 * 6 & 7 can happen if we have two racing threads T_a and T_b
969 	 * which are each trying to set an xattr on inodes I_a and I_b
970 	 * which were both initially sharing an xattr block.
971 	 */
972 	credits = 7;
973 
974 	/* Quota updates. */
975 	credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
976 
977 	/*
978 	 * In case of inline data, we may push out the data to a block,
979 	 * so we need to reserve credits for this eventuality
980 	 */
981 	if (inode && ext4_has_inline_data(inode))
982 		credits += ext4_writepage_trans_blocks(inode) + 1;
983 
984 	/* We are done if ea_inode feature is not enabled. */
985 	if (!ext4_has_feature_ea_inode(sb))
986 		return credits;
987 
988 	/* New ea_inode, inode map, block bitmap, group descriptor. */
989 	credits += 4;
990 
991 	/* Data blocks. */
992 	blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
993 
994 	/* Indirection block or one level of extent tree. */
995 	blocks += 1;
996 
997 	/* Block bitmap and group descriptor updates for each block. */
998 	credits += blocks * 2;
999 
1000 	/* Blocks themselves. */
1001 	credits += blocks;
1002 
1003 	if (!is_create) {
1004 		/* Dereference ea_inode holding old xattr value.
1005 		 * Old ea_inode, inode map, block bitmap, group descriptor.
1006 		 */
1007 		credits += 4;
1008 
1009 		/* Data blocks for old ea_inode. */
1010 		blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
1011 
1012 		/* Indirection block or one level of extent tree for old
1013 		 * ea_inode.
1014 		 */
1015 		blocks += 1;
1016 
1017 		/* Block bitmap and group descriptor updates for each block. */
1018 		credits += blocks * 2;
1019 	}
1020 
1021 	/* We may need to clone the existing xattr block in which case we need
1022 	 * to increment ref counts for existing ea_inodes referenced by it.
1023 	 */
1024 	if (block_bh) {
1025 		struct ext4_xattr_entry *entry = BFIRST(block_bh);
1026 
1027 		for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1028 			if (entry->e_value_inum)
1029 				/* Ref count update on ea_inode. */
1030 				credits += 1;
1031 	}
1032 	return credits;
1033 }
1034 
ext4_xattr_inode_update_ref(handle_t * handle,struct inode * ea_inode,int ref_change)1035 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1036 				       int ref_change)
1037 {
1038 	struct ext4_iloc iloc;
1039 	s64 ref_count;
1040 	int ret;
1041 
1042 	inode_lock_nested(ea_inode, I_MUTEX_XATTR);
1043 
1044 	ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1045 	if (ret)
1046 		goto out;
1047 
1048 	ref_count = ext4_xattr_inode_get_ref(ea_inode);
1049 	ref_count += ref_change;
1050 	ext4_xattr_inode_set_ref(ea_inode, ref_count);
1051 
1052 	if (ref_change > 0) {
1053 		WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1054 			  ea_inode->i_ino, ref_count);
1055 
1056 		if (ref_count == 1) {
1057 			WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1058 				  ea_inode->i_ino, ea_inode->i_nlink);
1059 
1060 			set_nlink(ea_inode, 1);
1061 			ext4_orphan_del(handle, ea_inode);
1062 		}
1063 	} else {
1064 		WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1065 			  ea_inode->i_ino, ref_count);
1066 
1067 		if (ref_count == 0) {
1068 			WARN_ONCE(ea_inode->i_nlink != 1,
1069 				  "EA inode %lu i_nlink=%u",
1070 				  ea_inode->i_ino, ea_inode->i_nlink);
1071 
1072 			clear_nlink(ea_inode);
1073 			ext4_orphan_add(handle, ea_inode);
1074 		}
1075 	}
1076 
1077 	ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1078 	if (ret)
1079 		ext4_warning_inode(ea_inode,
1080 				   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1081 out:
1082 	inode_unlock(ea_inode);
1083 	return ret;
1084 }
1085 
ext4_xattr_inode_inc_ref(handle_t * handle,struct inode * ea_inode)1086 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1087 {
1088 	return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1089 }
1090 
ext4_xattr_inode_dec_ref(handle_t * handle,struct inode * ea_inode)1091 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1092 {
1093 	return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1094 }
1095 
ext4_xattr_inode_inc_ref_all(handle_t * handle,struct inode * parent,struct ext4_xattr_entry * first)1096 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1097 					struct ext4_xattr_entry *first)
1098 {
1099 	struct inode *ea_inode;
1100 	struct ext4_xattr_entry *entry;
1101 	struct ext4_xattr_entry *failed_entry;
1102 	unsigned int ea_ino;
1103 	int err, saved_err;
1104 
1105 	for (entry = first; !IS_LAST_ENTRY(entry);
1106 	     entry = EXT4_XATTR_NEXT(entry)) {
1107 		if (!entry->e_value_inum)
1108 			continue;
1109 		ea_ino = le32_to_cpu(entry->e_value_inum);
1110 		err = ext4_xattr_inode_iget(parent, ea_ino,
1111 					    le32_to_cpu(entry->e_hash),
1112 					    &ea_inode);
1113 		if (err)
1114 			goto cleanup;
1115 		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1116 		if (err) {
1117 			ext4_warning_inode(ea_inode, "inc ref error %d", err);
1118 			iput(ea_inode);
1119 			goto cleanup;
1120 		}
1121 		iput(ea_inode);
1122 	}
1123 	return 0;
1124 
1125 cleanup:
1126 	saved_err = err;
1127 	failed_entry = entry;
1128 
1129 	for (entry = first; entry != failed_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,
1135 					    le32_to_cpu(entry->e_hash),
1136 					    &ea_inode);
1137 		if (err) {
1138 			ext4_warning(parent->i_sb,
1139 				     "cleanup ea_ino %u iget error %d", ea_ino,
1140 				     err);
1141 			continue;
1142 		}
1143 		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1144 		if (err)
1145 			ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1146 					   err);
1147 		iput(ea_inode);
1148 	}
1149 	return saved_err;
1150 }
1151 
ext4_xattr_restart_fn(handle_t * handle,struct inode * inode,struct buffer_head * bh,bool block_csum,bool dirty)1152 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1153 			struct buffer_head *bh, bool block_csum, bool dirty)
1154 {
1155 	int error;
1156 
1157 	if (bh && dirty) {
1158 		if (block_csum)
1159 			ext4_xattr_block_csum_set(inode, bh);
1160 		error = ext4_handle_dirty_metadata(handle, NULL, bh);
1161 		if (error) {
1162 			ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1163 				     error);
1164 			return error;
1165 		}
1166 	}
1167 	return 0;
1168 }
1169 
1170 static void
ext4_xattr_inode_dec_ref_all(handle_t * handle,struct inode * parent,struct buffer_head * bh,struct ext4_xattr_entry * first,bool block_csum,struct ext4_xattr_inode_array ** ea_inode_array,int extra_credits,bool skip_quota)1171 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1172 			     struct buffer_head *bh,
1173 			     struct ext4_xattr_entry *first, bool block_csum,
1174 			     struct ext4_xattr_inode_array **ea_inode_array,
1175 			     int extra_credits, bool skip_quota)
1176 {
1177 	struct inode *ea_inode;
1178 	struct ext4_xattr_entry *entry;
1179 	bool dirty = false;
1180 	unsigned int ea_ino;
1181 	int err;
1182 	int credits;
1183 
1184 	/* One credit for dec ref on ea_inode, one for orphan list addition, */
1185 	credits = 2 + extra_credits;
1186 
1187 	for (entry = first; !IS_LAST_ENTRY(entry);
1188 	     entry = EXT4_XATTR_NEXT(entry)) {
1189 		if (!entry->e_value_inum)
1190 			continue;
1191 		ea_ino = le32_to_cpu(entry->e_value_inum);
1192 		err = ext4_xattr_inode_iget(parent, ea_ino,
1193 					    le32_to_cpu(entry->e_hash),
1194 					    &ea_inode);
1195 		if (err)
1196 			continue;
1197 
1198 		err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1199 		if (err) {
1200 			ext4_warning_inode(ea_inode,
1201 					   "Expand inode array err=%d", err);
1202 			iput(ea_inode);
1203 			continue;
1204 		}
1205 
1206 		err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1207 			ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1208 			ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1209 					      dirty));
1210 		if (err < 0) {
1211 			ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1212 					   err);
1213 			continue;
1214 		}
1215 		if (err > 0) {
1216 			err = ext4_journal_get_write_access(handle,
1217 					parent->i_sb, bh, EXT4_JTR_NONE);
1218 			if (err) {
1219 				ext4_warning_inode(ea_inode,
1220 						"Re-get write access err=%d",
1221 						err);
1222 				continue;
1223 			}
1224 		}
1225 
1226 		err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1227 		if (err) {
1228 			ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1229 					   err);
1230 			continue;
1231 		}
1232 
1233 		if (!skip_quota)
1234 			ext4_xattr_inode_free_quota(parent, ea_inode,
1235 					      le32_to_cpu(entry->e_value_size));
1236 
1237 		/*
1238 		 * Forget about ea_inode within the same transaction that
1239 		 * decrements the ref count. This avoids duplicate decrements in
1240 		 * case the rest of the work spills over to subsequent
1241 		 * transactions.
1242 		 */
1243 		entry->e_value_inum = 0;
1244 		entry->e_value_size = 0;
1245 
1246 		dirty = true;
1247 	}
1248 
1249 	if (dirty) {
1250 		/*
1251 		 * Note that we are deliberately skipping csum calculation for
1252 		 * the final update because we do not expect any journal
1253 		 * restarts until xattr block is freed.
1254 		 */
1255 
1256 		err = ext4_handle_dirty_metadata(handle, NULL, bh);
1257 		if (err)
1258 			ext4_warning_inode(parent,
1259 					   "handle dirty metadata err=%d", err);
1260 	}
1261 }
1262 
1263 /*
1264  * Release the xattr block BH: If the reference count is > 1, decrement it;
1265  * otherwise free the block.
1266  */
1267 static void
ext4_xattr_release_block(handle_t * handle,struct inode * inode,struct buffer_head * bh,struct ext4_xattr_inode_array ** ea_inode_array,int extra_credits)1268 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1269 			 struct buffer_head *bh,
1270 			 struct ext4_xattr_inode_array **ea_inode_array,
1271 			 int extra_credits)
1272 {
1273 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1274 	u32 hash, ref;
1275 	int error = 0;
1276 
1277 	BUFFER_TRACE(bh, "get_write_access");
1278 	error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1279 					      EXT4_JTR_NONE);
1280 	if (error)
1281 		goto out;
1282 
1283 retry_ref:
1284 	lock_buffer(bh);
1285 	hash = le32_to_cpu(BHDR(bh)->h_hash);
1286 	ref = le32_to_cpu(BHDR(bh)->h_refcount);
1287 	if (ref == 1) {
1288 		ea_bdebug(bh, "refcount now=0; freeing");
1289 		/*
1290 		 * This must happen under buffer lock for
1291 		 * ext4_xattr_block_set() to reliably detect freed block
1292 		 */
1293 		if (ea_block_cache) {
1294 			struct mb_cache_entry *oe;
1295 
1296 			oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1297 							  bh->b_blocknr);
1298 			if (oe) {
1299 				unlock_buffer(bh);
1300 				mb_cache_entry_wait_unused(oe);
1301 				mb_cache_entry_put(ea_block_cache, oe);
1302 				goto retry_ref;
1303 			}
1304 		}
1305 		get_bh(bh);
1306 		unlock_buffer(bh);
1307 
1308 		if (ext4_has_feature_ea_inode(inode->i_sb))
1309 			ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1310 						     BFIRST(bh),
1311 						     true /* block_csum */,
1312 						     ea_inode_array,
1313 						     extra_credits,
1314 						     true /* skip_quota */);
1315 		ext4_free_blocks(handle, inode, bh, 0, 1,
1316 				 EXT4_FREE_BLOCKS_METADATA |
1317 				 EXT4_FREE_BLOCKS_FORGET);
1318 	} else {
1319 		ref--;
1320 		BHDR(bh)->h_refcount = cpu_to_le32(ref);
1321 		if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1322 			struct mb_cache_entry *ce;
1323 
1324 			if (ea_block_cache) {
1325 				ce = mb_cache_entry_get(ea_block_cache, hash,
1326 							bh->b_blocknr);
1327 				if (ce) {
1328 					set_bit(MBE_REUSABLE_B, &ce->e_flags);
1329 					mb_cache_entry_put(ea_block_cache, ce);
1330 				}
1331 			}
1332 		}
1333 
1334 		ext4_xattr_block_csum_set(inode, bh);
1335 		/*
1336 		 * Beware of this ugliness: Releasing of xattr block references
1337 		 * from different inodes can race and so we have to protect
1338 		 * from a race where someone else frees the block (and releases
1339 		 * its journal_head) before we are done dirtying the buffer. In
1340 		 * nojournal mode this race is harmless and we actually cannot
1341 		 * call ext4_handle_dirty_metadata() with locked buffer as
1342 		 * that function can call sync_dirty_buffer() so for that case
1343 		 * we handle the dirtying after unlocking the buffer.
1344 		 */
1345 		if (ext4_handle_valid(handle))
1346 			error = ext4_handle_dirty_metadata(handle, inode, bh);
1347 		unlock_buffer(bh);
1348 		if (!ext4_handle_valid(handle))
1349 			error = ext4_handle_dirty_metadata(handle, inode, bh);
1350 		if (IS_SYNC(inode))
1351 			ext4_handle_sync(handle);
1352 		dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1353 		ea_bdebug(bh, "refcount now=%d; releasing",
1354 			  le32_to_cpu(BHDR(bh)->h_refcount));
1355 	}
1356 out:
1357 	ext4_std_error(inode->i_sb, error);
1358 	return;
1359 }
1360 
1361 /*
1362  * Find the available free space for EAs. This also returns the total number of
1363  * bytes used by EA entries.
1364  */
ext4_xattr_free_space(struct ext4_xattr_entry * last,size_t * min_offs,void * base,int * total)1365 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1366 				    size_t *min_offs, void *base, int *total)
1367 {
1368 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1369 		if (!last->e_value_inum && last->e_value_size) {
1370 			size_t offs = le16_to_cpu(last->e_value_offs);
1371 			if (offs < *min_offs)
1372 				*min_offs = offs;
1373 		}
1374 		if (total)
1375 			*total += EXT4_XATTR_LEN(last->e_name_len);
1376 	}
1377 	return (*min_offs - ((void *)last - base) - sizeof(__u32));
1378 }
1379 
1380 /*
1381  * Write the value of the EA in an inode.
1382  */
ext4_xattr_inode_write(handle_t * handle,struct inode * ea_inode,const void * buf,int bufsize)1383 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1384 				  const void *buf, int bufsize)
1385 {
1386 	struct buffer_head *bh = NULL;
1387 	unsigned long block = 0;
1388 	int blocksize = ea_inode->i_sb->s_blocksize;
1389 	int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1390 	int csize, wsize = 0;
1391 	int ret = 0, ret2 = 0;
1392 	int retries = 0;
1393 
1394 retry:
1395 	while (ret >= 0 && ret < max_blocks) {
1396 		struct ext4_map_blocks map;
1397 		map.m_lblk = block += ret;
1398 		map.m_len = max_blocks -= ret;
1399 
1400 		ret = ext4_map_blocks(handle, ea_inode, &map,
1401 				      EXT4_GET_BLOCKS_CREATE);
1402 		if (ret <= 0) {
1403 			ext4_mark_inode_dirty(handle, ea_inode);
1404 			if (ret == -ENOSPC &&
1405 			    ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1406 				ret = 0;
1407 				goto retry;
1408 			}
1409 			break;
1410 		}
1411 	}
1412 
1413 	if (ret < 0)
1414 		return ret;
1415 
1416 	block = 0;
1417 	while (wsize < bufsize) {
1418 		brelse(bh);
1419 		csize = (bufsize - wsize) > blocksize ? blocksize :
1420 								bufsize - wsize;
1421 		bh = ext4_getblk(handle, ea_inode, block, 0);
1422 		if (IS_ERR(bh))
1423 			return PTR_ERR(bh);
1424 		if (!bh) {
1425 			WARN_ON_ONCE(1);
1426 			EXT4_ERROR_INODE(ea_inode,
1427 					 "ext4_getblk() return bh = NULL");
1428 			return -EFSCORRUPTED;
1429 		}
1430 		ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1431 						   EXT4_JTR_NONE);
1432 		if (ret)
1433 			goto out;
1434 
1435 		memcpy(bh->b_data, buf, csize);
1436 		/*
1437 		 * Zero out block tail to avoid writing uninitialized memory
1438 		 * to disk.
1439 		 */
1440 		if (csize < blocksize)
1441 			memset(bh->b_data + csize, 0, blocksize - csize);
1442 		set_buffer_uptodate(bh);
1443 		ext4_handle_dirty_metadata(handle, ea_inode, bh);
1444 
1445 		buf += csize;
1446 		wsize += csize;
1447 		block += 1;
1448 	}
1449 
1450 	inode_lock(ea_inode);
1451 	i_size_write(ea_inode, wsize);
1452 	ext4_update_i_disksize(ea_inode, wsize);
1453 	inode_unlock(ea_inode);
1454 
1455 	ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1456 	if (unlikely(ret2 && !ret))
1457 		ret = ret2;
1458 
1459 out:
1460 	brelse(bh);
1461 
1462 	return ret;
1463 }
1464 
1465 /*
1466  * Create an inode to store the value of a large EA.
1467  */
ext4_xattr_inode_create(handle_t * handle,struct inode * inode,u32 hash)1468 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1469 					     struct inode *inode, u32 hash)
1470 {
1471 	struct inode *ea_inode = NULL;
1472 	uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1473 	int err;
1474 
1475 	if (inode->i_sb->s_root == NULL) {
1476 		ext4_warning(inode->i_sb,
1477 			     "refuse to create EA inode when umounting");
1478 		WARN_ON(1);
1479 		return ERR_PTR(-EINVAL);
1480 	}
1481 
1482 	/*
1483 	 * Let the next inode be the goal, so we try and allocate the EA inode
1484 	 * in the same group, or nearby one.
1485 	 */
1486 	ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1487 				  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1488 				  EXT4_EA_INODE_FL);
1489 	if (!IS_ERR(ea_inode)) {
1490 		ea_inode->i_op = &ext4_file_inode_operations;
1491 		ea_inode->i_fop = &ext4_file_operations;
1492 		ext4_set_aops(ea_inode);
1493 		ext4_xattr_inode_set_class(ea_inode);
1494 		unlock_new_inode(ea_inode);
1495 		ext4_xattr_inode_set_ref(ea_inode, 1);
1496 		ext4_xattr_inode_set_hash(ea_inode, hash);
1497 		err = ext4_mark_inode_dirty(handle, ea_inode);
1498 		if (!err)
1499 			err = ext4_inode_attach_jinode(ea_inode);
1500 		if (err) {
1501 			if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1502 				ext4_warning_inode(ea_inode,
1503 					"cleanup dec ref error %d", err);
1504 			iput(ea_inode);
1505 			return ERR_PTR(err);
1506 		}
1507 
1508 		/*
1509 		 * Xattr inodes are shared therefore quota charging is performed
1510 		 * at a higher level.
1511 		 */
1512 		dquot_free_inode(ea_inode);
1513 		dquot_drop(ea_inode);
1514 		inode_lock(ea_inode);
1515 		ea_inode->i_flags |= S_NOQUOTA;
1516 		inode_unlock(ea_inode);
1517 	}
1518 
1519 	return ea_inode;
1520 }
1521 
1522 static struct inode *
ext4_xattr_inode_cache_find(struct inode * inode,const void * value,size_t value_len,u32 hash)1523 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1524 			    size_t value_len, u32 hash)
1525 {
1526 	struct inode *ea_inode;
1527 	struct mb_cache_entry *ce;
1528 	struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1529 	void *ea_data;
1530 
1531 	if (!ea_inode_cache)
1532 		return NULL;
1533 
1534 	ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1535 	if (!ce)
1536 		return NULL;
1537 
1538 	WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1539 		     !(current->flags & PF_MEMALLOC_NOFS));
1540 
1541 	ea_data = kvmalloc(value_len, GFP_KERNEL);
1542 	if (!ea_data) {
1543 		mb_cache_entry_put(ea_inode_cache, ce);
1544 		return NULL;
1545 	}
1546 
1547 	while (ce) {
1548 		ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1549 				     EXT4_IGET_EA_INODE);
1550 		if (IS_ERR(ea_inode))
1551 			goto next_entry;
1552 		ext4_xattr_inode_set_class(ea_inode);
1553 		if (i_size_read(ea_inode) == value_len &&
1554 		    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1555 		    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1556 						    value_len) &&
1557 		    !memcmp(value, ea_data, value_len)) {
1558 			mb_cache_entry_touch(ea_inode_cache, ce);
1559 			mb_cache_entry_put(ea_inode_cache, ce);
1560 			kvfree(ea_data);
1561 			return ea_inode;
1562 		}
1563 		iput(ea_inode);
1564 	next_entry:
1565 		ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1566 	}
1567 	kvfree(ea_data);
1568 	return NULL;
1569 }
1570 
1571 /*
1572  * Add value of the EA in an inode.
1573  */
ext4_xattr_inode_lookup_create(handle_t * handle,struct inode * inode,const void * value,size_t value_len)1574 static struct inode *ext4_xattr_inode_lookup_create(handle_t *handle,
1575 		struct inode *inode, const void *value, size_t value_len)
1576 {
1577 	struct inode *ea_inode;
1578 	u32 hash;
1579 	int err;
1580 
1581 	/* Account inode & space to quota even if sharing... */
1582 	err = ext4_xattr_inode_alloc_quota(inode, value_len);
1583 	if (err)
1584 		return ERR_PTR(err);
1585 
1586 	hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1587 	ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1588 	if (ea_inode) {
1589 		err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1590 		if (err)
1591 			goto out_err;
1592 		return ea_inode;
1593 	}
1594 
1595 	/* Create an inode for the EA value */
1596 	ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1597 	if (IS_ERR(ea_inode)) {
1598 		ext4_xattr_inode_free_quota(inode, NULL, value_len);
1599 		return ea_inode;
1600 	}
1601 
1602 	err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1603 	if (err) {
1604 		if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1605 			ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1606 		goto out_err;
1607 	}
1608 
1609 	if (EA_INODE_CACHE(inode))
1610 		mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1611 				      ea_inode->i_ino, true /* reusable */);
1612 	return ea_inode;
1613 out_err:
1614 	iput(ea_inode);
1615 	ext4_xattr_inode_free_quota(inode, NULL, value_len);
1616 	return ERR_PTR(err);
1617 }
1618 
1619 /*
1620  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1621  * feature is enabled.
1622  */
1623 #define EXT4_XATTR_BLOCK_RESERVE(inode)	min(i_blocksize(inode)/8, 1024U)
1624 
ext4_xattr_set_entry(struct ext4_xattr_info * i,struct ext4_xattr_search * s,handle_t * handle,struct inode * inode,struct inode * new_ea_inode,bool is_block)1625 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1626 				struct ext4_xattr_search *s,
1627 				handle_t *handle, struct inode *inode,
1628 				struct inode *new_ea_inode,
1629 				bool is_block)
1630 {
1631 	struct ext4_xattr_entry *last, *next;
1632 	struct ext4_xattr_entry *here = s->here;
1633 	size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1634 	int in_inode = i->in_inode;
1635 	struct inode *old_ea_inode = NULL;
1636 	size_t old_size, new_size;
1637 	int ret;
1638 
1639 	/* Space used by old and new values. */
1640 	old_size = (!s->not_found && !here->e_value_inum) ?
1641 			EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1642 	new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1643 
1644 	/*
1645 	 * Optimization for the simple case when old and new values have the
1646 	 * same padded sizes. Not applicable if external inodes are involved.
1647 	 */
1648 	if (new_size && new_size == old_size) {
1649 		size_t offs = le16_to_cpu(here->e_value_offs);
1650 		void *val = s->base + offs;
1651 
1652 		here->e_value_size = cpu_to_le32(i->value_len);
1653 		if (i->value == EXT4_ZERO_XATTR_VALUE) {
1654 			memset(val, 0, new_size);
1655 		} else {
1656 			memcpy(val, i->value, i->value_len);
1657 			/* Clear padding bytes. */
1658 			memset(val + i->value_len, 0, new_size - i->value_len);
1659 		}
1660 		goto update_hash;
1661 	}
1662 
1663 	/* Compute min_offs and last. */
1664 	last = s->first;
1665 	for (; !IS_LAST_ENTRY(last); last = next) {
1666 		next = EXT4_XATTR_NEXT(last);
1667 		if ((void *)next >= s->end) {
1668 			EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1669 			ret = -EFSCORRUPTED;
1670 			goto out;
1671 		}
1672 		if (!last->e_value_inum && last->e_value_size) {
1673 			size_t offs = le16_to_cpu(last->e_value_offs);
1674 			if (offs < min_offs)
1675 				min_offs = offs;
1676 		}
1677 	}
1678 
1679 	/* Check whether we have enough space. */
1680 	if (i->value) {
1681 		size_t free;
1682 
1683 		free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1684 		if (!s->not_found)
1685 			free += EXT4_XATTR_LEN(name_len) + old_size;
1686 
1687 		if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1688 			ret = -ENOSPC;
1689 			goto out;
1690 		}
1691 
1692 		/*
1693 		 * If storing the value in an external inode is an option,
1694 		 * reserve space for xattr entries/names in the external
1695 		 * attribute block so that a long value does not occupy the
1696 		 * whole space and prevent further entries being added.
1697 		 */
1698 		if (ext4_has_feature_ea_inode(inode->i_sb) &&
1699 		    new_size && is_block &&
1700 		    (min_offs + old_size - new_size) <
1701 					EXT4_XATTR_BLOCK_RESERVE(inode)) {
1702 			ret = -ENOSPC;
1703 			goto out;
1704 		}
1705 	}
1706 
1707 	/*
1708 	 * Getting access to old and new ea inodes is subject to failures.
1709 	 * Finish that work before doing any modifications to the xattr data.
1710 	 */
1711 	if (!s->not_found && here->e_value_inum) {
1712 		ret = ext4_xattr_inode_iget(inode,
1713 					    le32_to_cpu(here->e_value_inum),
1714 					    le32_to_cpu(here->e_hash),
1715 					    &old_ea_inode);
1716 		if (ret) {
1717 			old_ea_inode = NULL;
1718 			goto out;
1719 		}
1720 
1721 		/* We are ready to release ref count on the old_ea_inode. */
1722 		ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1723 		if (ret)
1724 			goto out;
1725 
1726 		ext4_xattr_inode_free_quota(inode, old_ea_inode,
1727 					    le32_to_cpu(here->e_value_size));
1728 	}
1729 
1730 	/* No failures allowed past this point. */
1731 
1732 	if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1733 		/* Remove the old value. */
1734 		void *first_val = s->base + min_offs;
1735 		size_t offs = le16_to_cpu(here->e_value_offs);
1736 		void *val = s->base + offs;
1737 
1738 		memmove(first_val + old_size, first_val, val - first_val);
1739 		memset(first_val, 0, old_size);
1740 		min_offs += old_size;
1741 
1742 		/* Adjust all value offsets. */
1743 		last = s->first;
1744 		while (!IS_LAST_ENTRY(last)) {
1745 			size_t o = le16_to_cpu(last->e_value_offs);
1746 
1747 			if (!last->e_value_inum &&
1748 			    last->e_value_size && o < offs)
1749 				last->e_value_offs = cpu_to_le16(o + old_size);
1750 			last = EXT4_XATTR_NEXT(last);
1751 		}
1752 	}
1753 
1754 	if (!i->value) {
1755 		/* Remove old name. */
1756 		size_t size = EXT4_XATTR_LEN(name_len);
1757 
1758 		last = ENTRY((void *)last - size);
1759 		memmove(here, (void *)here + size,
1760 			(void *)last - (void *)here + sizeof(__u32));
1761 		memset(last, 0, size);
1762 
1763 		/*
1764 		 * Update i_inline_off - moved ibody region might contain
1765 		 * system.data attribute.  Handling a failure here won't
1766 		 * cause other complications for setting an xattr.
1767 		 */
1768 		if (!is_block && ext4_has_inline_data(inode)) {
1769 			ret = ext4_find_inline_data_nolock(inode);
1770 			if (ret) {
1771 				ext4_warning_inode(inode,
1772 					"unable to update i_inline_off");
1773 				goto out;
1774 			}
1775 		}
1776 	} else if (s->not_found) {
1777 		/* Insert new name. */
1778 		size_t size = EXT4_XATTR_LEN(name_len);
1779 		size_t rest = (void *)last - (void *)here + sizeof(__u32);
1780 
1781 		memmove((void *)here + size, here, rest);
1782 		memset(here, 0, size);
1783 		here->e_name_index = i->name_index;
1784 		here->e_name_len = name_len;
1785 		memcpy(here->e_name, i->name, name_len);
1786 	} else {
1787 		/* This is an update, reset value info. */
1788 		here->e_value_inum = 0;
1789 		here->e_value_offs = 0;
1790 		here->e_value_size = 0;
1791 	}
1792 
1793 	if (i->value) {
1794 		/* Insert new value. */
1795 		if (in_inode) {
1796 			here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1797 		} else if (i->value_len) {
1798 			void *val = s->base + min_offs - new_size;
1799 
1800 			here->e_value_offs = cpu_to_le16(min_offs - new_size);
1801 			if (i->value == EXT4_ZERO_XATTR_VALUE) {
1802 				memset(val, 0, new_size);
1803 			} else {
1804 				memcpy(val, i->value, i->value_len);
1805 				/* Clear padding bytes. */
1806 				memset(val + i->value_len, 0,
1807 				       new_size - i->value_len);
1808 			}
1809 		}
1810 		here->e_value_size = cpu_to_le32(i->value_len);
1811 	}
1812 
1813 update_hash:
1814 	if (i->value) {
1815 		__le32 hash = 0;
1816 
1817 		/* Entry hash calculation. */
1818 		if (in_inode) {
1819 			__le32 crc32c_hash;
1820 
1821 			/*
1822 			 * Feed crc32c hash instead of the raw value for entry
1823 			 * hash calculation. This is to avoid walking
1824 			 * potentially long value buffer again.
1825 			 */
1826 			crc32c_hash = cpu_to_le32(
1827 				       ext4_xattr_inode_get_hash(new_ea_inode));
1828 			hash = ext4_xattr_hash_entry(here->e_name,
1829 						     here->e_name_len,
1830 						     &crc32c_hash, 1);
1831 		} else if (is_block) {
1832 			__le32 *value = s->base + le16_to_cpu(
1833 							here->e_value_offs);
1834 
1835 			hash = ext4_xattr_hash_entry(here->e_name,
1836 						     here->e_name_len, value,
1837 						     new_size >> 2);
1838 		}
1839 		here->e_hash = hash;
1840 	}
1841 
1842 	if (is_block)
1843 		ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1844 
1845 	ret = 0;
1846 out:
1847 	iput(old_ea_inode);
1848 	return ret;
1849 }
1850 
1851 struct ext4_xattr_block_find {
1852 	struct ext4_xattr_search s;
1853 	struct buffer_head *bh;
1854 };
1855 
1856 static int
ext4_xattr_block_find(struct inode * inode,struct ext4_xattr_info * i,struct ext4_xattr_block_find * bs)1857 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1858 		      struct ext4_xattr_block_find *bs)
1859 {
1860 	struct super_block *sb = inode->i_sb;
1861 	int error;
1862 
1863 	ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1864 		  i->name_index, i->name, i->value, (long)i->value_len);
1865 
1866 	if (EXT4_I(inode)->i_file_acl) {
1867 		/* The inode already has an extended attribute block. */
1868 		bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1869 		if (IS_ERR(bs->bh)) {
1870 			error = PTR_ERR(bs->bh);
1871 			bs->bh = NULL;
1872 			return error;
1873 		}
1874 		ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1875 			atomic_read(&(bs->bh->b_count)),
1876 			le32_to_cpu(BHDR(bs->bh)->h_refcount));
1877 		error = ext4_xattr_check_block(inode, bs->bh);
1878 		if (error)
1879 			return error;
1880 		/* Find the named attribute. */
1881 		bs->s.base = BHDR(bs->bh);
1882 		bs->s.first = BFIRST(bs->bh);
1883 		bs->s.end = bs->bh->b_data + bs->bh->b_size;
1884 		bs->s.here = bs->s.first;
1885 		error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1886 					 i->name_index, i->name, 1);
1887 		if (error && error != -ENODATA)
1888 			return error;
1889 		bs->s.not_found = error;
1890 	}
1891 	return 0;
1892 }
1893 
1894 static int
ext4_xattr_block_set(handle_t * handle,struct inode * inode,struct ext4_xattr_info * i,struct ext4_xattr_block_find * bs)1895 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1896 		     struct ext4_xattr_info *i,
1897 		     struct ext4_xattr_block_find *bs)
1898 {
1899 	struct super_block *sb = inode->i_sb;
1900 	struct buffer_head *new_bh = NULL;
1901 	struct ext4_xattr_search s_copy = bs->s;
1902 	struct ext4_xattr_search *s = &s_copy;
1903 	struct mb_cache_entry *ce = NULL;
1904 	int error = 0;
1905 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1906 	struct inode *ea_inode = NULL, *tmp_inode;
1907 	size_t old_ea_inode_quota = 0;
1908 	unsigned int ea_ino;
1909 
1910 #define header(x) ((struct ext4_xattr_header *)(x))
1911 
1912 	/* If we need EA inode, prepare it before locking the buffer */
1913 	if (i->value && i->in_inode) {
1914 		WARN_ON_ONCE(!i->value_len);
1915 
1916 		ea_inode = ext4_xattr_inode_lookup_create(handle, inode,
1917 					i->value, i->value_len);
1918 		if (IS_ERR(ea_inode)) {
1919 			error = PTR_ERR(ea_inode);
1920 			ea_inode = NULL;
1921 			goto cleanup;
1922 		}
1923 	}
1924 
1925 	if (s->base) {
1926 		int offset = (char *)s->here - bs->bh->b_data;
1927 
1928 		BUFFER_TRACE(bs->bh, "get_write_access");
1929 		error = ext4_journal_get_write_access(handle, sb, bs->bh,
1930 						      EXT4_JTR_NONE);
1931 		if (error)
1932 			goto cleanup;
1933 
1934 		lock_buffer(bs->bh);
1935 
1936 		if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1937 			__u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1938 
1939 			/*
1940 			 * This must happen under buffer lock for
1941 			 * ext4_xattr_block_set() to reliably detect modified
1942 			 * block
1943 			 */
1944 			if (ea_block_cache) {
1945 				struct mb_cache_entry *oe;
1946 
1947 				oe = mb_cache_entry_delete_or_get(ea_block_cache,
1948 					hash, bs->bh->b_blocknr);
1949 				if (oe) {
1950 					/*
1951 					 * Xattr block is getting reused. Leave
1952 					 * it alone.
1953 					 */
1954 					mb_cache_entry_put(ea_block_cache, oe);
1955 					goto clone_block;
1956 				}
1957 			}
1958 			ea_bdebug(bs->bh, "modifying in-place");
1959 			error = ext4_xattr_set_entry(i, s, handle, inode,
1960 					     ea_inode, true /* is_block */);
1961 			ext4_xattr_block_csum_set(inode, bs->bh);
1962 			unlock_buffer(bs->bh);
1963 			if (error == -EFSCORRUPTED)
1964 				goto bad_block;
1965 			if (!error)
1966 				error = ext4_handle_dirty_metadata(handle,
1967 								   inode,
1968 								   bs->bh);
1969 			if (error)
1970 				goto cleanup;
1971 			goto inserted;
1972 		}
1973 clone_block:
1974 		unlock_buffer(bs->bh);
1975 		ea_bdebug(bs->bh, "cloning");
1976 		s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1977 		error = -ENOMEM;
1978 		if (s->base == NULL)
1979 			goto cleanup;
1980 		s->first = ENTRY(header(s->base)+1);
1981 		header(s->base)->h_refcount = cpu_to_le32(1);
1982 		s->here = ENTRY(s->base + offset);
1983 		s->end = s->base + bs->bh->b_size;
1984 
1985 		/*
1986 		 * If existing entry points to an xattr inode, we need
1987 		 * to prevent ext4_xattr_set_entry() from decrementing
1988 		 * ref count on it because the reference belongs to the
1989 		 * original block. In this case, make the entry look
1990 		 * like it has an empty value.
1991 		 */
1992 		if (!s->not_found && s->here->e_value_inum) {
1993 			ea_ino = le32_to_cpu(s->here->e_value_inum);
1994 			error = ext4_xattr_inode_iget(inode, ea_ino,
1995 				      le32_to_cpu(s->here->e_hash),
1996 				      &tmp_inode);
1997 			if (error)
1998 				goto cleanup;
1999 
2000 			if (!ext4_test_inode_state(tmp_inode,
2001 					EXT4_STATE_LUSTRE_EA_INODE)) {
2002 				/*
2003 				 * Defer quota free call for previous
2004 				 * inode until success is guaranteed.
2005 				 */
2006 				old_ea_inode_quota = le32_to_cpu(
2007 						s->here->e_value_size);
2008 			}
2009 			iput(tmp_inode);
2010 
2011 			s->here->e_value_inum = 0;
2012 			s->here->e_value_size = 0;
2013 		}
2014 	} else {
2015 		/* Allocate a buffer where we construct the new block. */
2016 		s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
2017 		error = -ENOMEM;
2018 		if (s->base == NULL)
2019 			goto cleanup;
2020 		header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2021 		header(s->base)->h_blocks = cpu_to_le32(1);
2022 		header(s->base)->h_refcount = cpu_to_le32(1);
2023 		s->first = ENTRY(header(s->base)+1);
2024 		s->here = ENTRY(header(s->base)+1);
2025 		s->end = s->base + sb->s_blocksize;
2026 	}
2027 
2028 	error = ext4_xattr_set_entry(i, s, handle, inode, ea_inode,
2029 				     true /* is_block */);
2030 	if (error == -EFSCORRUPTED)
2031 		goto bad_block;
2032 	if (error)
2033 		goto cleanup;
2034 
2035 inserted:
2036 	if (!IS_LAST_ENTRY(s->first)) {
2037 		new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2038 						     &ce);
2039 		if (new_bh) {
2040 			/* We found an identical block in the cache. */
2041 			if (new_bh == bs->bh)
2042 				ea_bdebug(new_bh, "keeping");
2043 			else {
2044 				u32 ref;
2045 
2046 #ifdef EXT4_XATTR_DEBUG
2047 				WARN_ON_ONCE(dquot_initialize_needed(inode));
2048 #endif
2049 				/* The old block is released after updating
2050 				   the inode. */
2051 				error = dquot_alloc_block(inode,
2052 						EXT4_C2B(EXT4_SB(sb), 1));
2053 				if (error)
2054 					goto cleanup;
2055 				BUFFER_TRACE(new_bh, "get_write_access");
2056 				error = ext4_journal_get_write_access(
2057 						handle, sb, new_bh,
2058 						EXT4_JTR_NONE);
2059 				if (error)
2060 					goto cleanup_dquot;
2061 				lock_buffer(new_bh);
2062 				/*
2063 				 * We have to be careful about races with
2064 				 * adding references to xattr block. Once we
2065 				 * hold buffer lock xattr block's state is
2066 				 * stable so we can check the additional
2067 				 * reference fits.
2068 				 */
2069 				ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2070 				if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2071 					/*
2072 					 * Undo everything and check mbcache
2073 					 * again.
2074 					 */
2075 					unlock_buffer(new_bh);
2076 					dquot_free_block(inode,
2077 							 EXT4_C2B(EXT4_SB(sb),
2078 								  1));
2079 					brelse(new_bh);
2080 					mb_cache_entry_put(ea_block_cache, ce);
2081 					ce = NULL;
2082 					new_bh = NULL;
2083 					goto inserted;
2084 				}
2085 				BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2086 				if (ref == EXT4_XATTR_REFCOUNT_MAX)
2087 					clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2088 				ea_bdebug(new_bh, "reusing; refcount now=%d",
2089 					  ref);
2090 				ext4_xattr_block_csum_set(inode, new_bh);
2091 				unlock_buffer(new_bh);
2092 				error = ext4_handle_dirty_metadata(handle,
2093 								   inode,
2094 								   new_bh);
2095 				if (error)
2096 					goto cleanup_dquot;
2097 			}
2098 			mb_cache_entry_touch(ea_block_cache, ce);
2099 			mb_cache_entry_put(ea_block_cache, ce);
2100 			ce = NULL;
2101 		} else if (bs->bh && s->base == bs->bh->b_data) {
2102 			/* We were modifying this block in-place. */
2103 			ea_bdebug(bs->bh, "keeping this block");
2104 			ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2105 			new_bh = bs->bh;
2106 			get_bh(new_bh);
2107 		} else {
2108 			/* We need to allocate a new block */
2109 			ext4_fsblk_t goal, block;
2110 
2111 #ifdef EXT4_XATTR_DEBUG
2112 			WARN_ON_ONCE(dquot_initialize_needed(inode));
2113 #endif
2114 			goal = ext4_group_first_block_no(sb,
2115 						EXT4_I(inode)->i_block_group);
2116 			block = ext4_new_meta_blocks(handle, inode, goal, 0,
2117 						     NULL, &error);
2118 			if (error)
2119 				goto cleanup;
2120 
2121 			ea_idebug(inode, "creating block %llu",
2122 				  (unsigned long long)block);
2123 
2124 			new_bh = sb_getblk(sb, block);
2125 			if (unlikely(!new_bh)) {
2126 				error = -ENOMEM;
2127 getblk_failed:
2128 				ext4_free_blocks(handle, inode, NULL, block, 1,
2129 						 EXT4_FREE_BLOCKS_METADATA);
2130 				goto cleanup;
2131 			}
2132 			error = ext4_xattr_inode_inc_ref_all(handle, inode,
2133 						      ENTRY(header(s->base)+1));
2134 			if (error)
2135 				goto getblk_failed;
2136 			if (ea_inode) {
2137 				/* Drop the extra ref on ea_inode. */
2138 				error = ext4_xattr_inode_dec_ref(handle,
2139 								 ea_inode);
2140 				if (error)
2141 					ext4_warning_inode(ea_inode,
2142 							   "dec ref error=%d",
2143 							   error);
2144 				iput(ea_inode);
2145 				ea_inode = NULL;
2146 			}
2147 
2148 			lock_buffer(new_bh);
2149 			error = ext4_journal_get_create_access(handle, sb,
2150 							new_bh, EXT4_JTR_NONE);
2151 			if (error) {
2152 				unlock_buffer(new_bh);
2153 				error = -EIO;
2154 				goto getblk_failed;
2155 			}
2156 			memcpy(new_bh->b_data, s->base, new_bh->b_size);
2157 			ext4_xattr_block_csum_set(inode, new_bh);
2158 			set_buffer_uptodate(new_bh);
2159 			unlock_buffer(new_bh);
2160 			ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2161 			error = ext4_handle_dirty_metadata(handle, inode,
2162 							   new_bh);
2163 			if (error)
2164 				goto cleanup;
2165 		}
2166 	}
2167 
2168 	if (old_ea_inode_quota)
2169 		ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2170 
2171 	/* Update the inode. */
2172 	EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2173 
2174 	/* Drop the previous xattr block. */
2175 	if (bs->bh && bs->bh != new_bh) {
2176 		struct ext4_xattr_inode_array *ea_inode_array = NULL;
2177 
2178 		ext4_xattr_release_block(handle, inode, bs->bh,
2179 					 &ea_inode_array,
2180 					 0 /* extra_credits */);
2181 		ext4_xattr_inode_array_free(ea_inode_array);
2182 	}
2183 	error = 0;
2184 
2185 cleanup:
2186 	if (ea_inode) {
2187 		if (error) {
2188 			int error2;
2189 
2190 			error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2191 			if (error2)
2192 				ext4_warning_inode(ea_inode, "dec ref error=%d",
2193 						   error2);
2194 			ext4_xattr_inode_free_quota(inode, ea_inode,
2195 						    i_size_read(ea_inode));
2196 		}
2197 		iput(ea_inode);
2198 	}
2199 	if (ce)
2200 		mb_cache_entry_put(ea_block_cache, ce);
2201 	brelse(new_bh);
2202 	if (!(bs->bh && s->base == bs->bh->b_data))
2203 		kfree(s->base);
2204 
2205 	return error;
2206 
2207 cleanup_dquot:
2208 	dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2209 	goto cleanup;
2210 
2211 bad_block:
2212 	EXT4_ERROR_INODE(inode, "bad block %llu",
2213 			 EXT4_I(inode)->i_file_acl);
2214 	goto cleanup;
2215 
2216 #undef header
2217 }
2218 
ext4_xattr_ibody_find(struct inode * inode,struct ext4_xattr_info * i,struct ext4_xattr_ibody_find * is)2219 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2220 			  struct ext4_xattr_ibody_find *is)
2221 {
2222 	struct ext4_xattr_ibody_header *header;
2223 	struct ext4_inode *raw_inode;
2224 	int error;
2225 
2226 	if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2227 		return 0;
2228 
2229 	raw_inode = ext4_raw_inode(&is->iloc);
2230 	header = IHDR(inode, raw_inode);
2231 	is->s.base = is->s.first = IFIRST(header);
2232 	is->s.here = is->s.first;
2233 	is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2234 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2235 		error = xattr_check_inode(inode, header, is->s.end);
2236 		if (error)
2237 			return error;
2238 		/* Find the named attribute. */
2239 		error = xattr_find_entry(inode, &is->s.here, is->s.end,
2240 					 i->name_index, i->name, 0);
2241 		if (error && error != -ENODATA)
2242 			return error;
2243 		is->s.not_found = error;
2244 	}
2245 	return 0;
2246 }
2247 
ext4_xattr_ibody_set(handle_t * handle,struct inode * inode,struct ext4_xattr_info * i,struct ext4_xattr_ibody_find * is)2248 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2249 				struct ext4_xattr_info *i,
2250 				struct ext4_xattr_ibody_find *is)
2251 {
2252 	struct ext4_xattr_ibody_header *header;
2253 	struct ext4_xattr_search *s = &is->s;
2254 	struct inode *ea_inode = NULL;
2255 	int error;
2256 
2257 	if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2258 		return -ENOSPC;
2259 
2260 	/* If we need EA inode, prepare it before locking the buffer */
2261 	if (i->value && i->in_inode) {
2262 		WARN_ON_ONCE(!i->value_len);
2263 
2264 		ea_inode = ext4_xattr_inode_lookup_create(handle, inode,
2265 					i->value, i->value_len);
2266 		if (IS_ERR(ea_inode))
2267 			return PTR_ERR(ea_inode);
2268 	}
2269 	error = ext4_xattr_set_entry(i, s, handle, inode, ea_inode,
2270 				     false /* is_block */);
2271 	if (error) {
2272 		if (ea_inode) {
2273 			int error2;
2274 
2275 			error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2276 			if (error2)
2277 				ext4_warning_inode(ea_inode, "dec ref error=%d",
2278 						   error2);
2279 
2280 			ext4_xattr_inode_free_quota(inode, ea_inode,
2281 						    i_size_read(ea_inode));
2282 			iput(ea_inode);
2283 		}
2284 		return error;
2285 	}
2286 	header = IHDR(inode, ext4_raw_inode(&is->iloc));
2287 	if (!IS_LAST_ENTRY(s->first)) {
2288 		header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2289 		ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2290 	} else {
2291 		header->h_magic = cpu_to_le32(0);
2292 		ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2293 	}
2294 	iput(ea_inode);
2295 	return 0;
2296 }
2297 
ext4_xattr_value_same(struct ext4_xattr_search * s,struct ext4_xattr_info * i)2298 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2299 				 struct ext4_xattr_info *i)
2300 {
2301 	void *value;
2302 
2303 	/* When e_value_inum is set the value is stored externally. */
2304 	if (s->here->e_value_inum)
2305 		return 0;
2306 	if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2307 		return 0;
2308 	value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2309 	return !memcmp(value, i->value, i->value_len);
2310 }
2311 
ext4_xattr_get_block(struct inode * inode)2312 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2313 {
2314 	struct buffer_head *bh;
2315 	int error;
2316 
2317 	if (!EXT4_I(inode)->i_file_acl)
2318 		return NULL;
2319 	bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2320 	if (IS_ERR(bh))
2321 		return bh;
2322 	error = ext4_xattr_check_block(inode, bh);
2323 	if (error) {
2324 		brelse(bh);
2325 		return ERR_PTR(error);
2326 	}
2327 	return bh;
2328 }
2329 
2330 /*
2331  * ext4_xattr_set_handle()
2332  *
2333  * Create, replace or remove an extended attribute for this inode.  Value
2334  * is NULL to remove an existing extended attribute, and non-NULL to
2335  * either replace an existing extended attribute, or create a new extended
2336  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2337  * specify that an extended attribute must exist and must not exist
2338  * previous to the call, respectively.
2339  *
2340  * Returns 0, or a negative error number on failure.
2341  */
2342 int
ext4_xattr_set_handle(handle_t * handle,struct inode * inode,int name_index,const char * name,const void * value,size_t value_len,int flags)2343 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2344 		      const char *name, const void *value, size_t value_len,
2345 		      int flags)
2346 {
2347 	struct ext4_xattr_info i = {
2348 		.name_index = name_index,
2349 		.name = name,
2350 		.value = value,
2351 		.value_len = value_len,
2352 		.in_inode = 0,
2353 	};
2354 	struct ext4_xattr_ibody_find is = {
2355 		.s = { .not_found = -ENODATA, },
2356 	};
2357 	struct ext4_xattr_block_find bs = {
2358 		.s = { .not_found = -ENODATA, },
2359 	};
2360 	int no_expand;
2361 	int error;
2362 
2363 	if (!name)
2364 		return -EINVAL;
2365 	if (strlen(name) > 255)
2366 		return -ERANGE;
2367 
2368 	ext4_write_lock_xattr(inode, &no_expand);
2369 
2370 	/* Check journal credits under write lock. */
2371 	if (ext4_handle_valid(handle)) {
2372 		struct buffer_head *bh;
2373 		int credits;
2374 
2375 		bh = ext4_xattr_get_block(inode);
2376 		if (IS_ERR(bh)) {
2377 			error = PTR_ERR(bh);
2378 			goto cleanup;
2379 		}
2380 
2381 		credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2382 						   value_len,
2383 						   flags & XATTR_CREATE);
2384 		brelse(bh);
2385 
2386 		if (jbd2_handle_buffer_credits(handle) < credits) {
2387 			error = -ENOSPC;
2388 			goto cleanup;
2389 		}
2390 		WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2391 	}
2392 
2393 	error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2394 	if (error)
2395 		goto cleanup;
2396 
2397 	if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2398 		struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2399 		memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2400 		ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2401 	}
2402 
2403 	error = ext4_xattr_ibody_find(inode, &i, &is);
2404 	if (error)
2405 		goto cleanup;
2406 	if (is.s.not_found)
2407 		error = ext4_xattr_block_find(inode, &i, &bs);
2408 	if (error)
2409 		goto cleanup;
2410 	if (is.s.not_found && bs.s.not_found) {
2411 		error = -ENODATA;
2412 		if (flags & XATTR_REPLACE)
2413 			goto cleanup;
2414 		error = 0;
2415 		if (!value)
2416 			goto cleanup;
2417 	} else {
2418 		error = -EEXIST;
2419 		if (flags & XATTR_CREATE)
2420 			goto cleanup;
2421 	}
2422 
2423 	if (!value) {
2424 		if (!is.s.not_found)
2425 			error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2426 		else if (!bs.s.not_found)
2427 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2428 	} else {
2429 		error = 0;
2430 		/* Xattr value did not change? Save us some work and bail out */
2431 		if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2432 			goto cleanup;
2433 		if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2434 			goto cleanup;
2435 
2436 		if (ext4_has_feature_ea_inode(inode->i_sb) &&
2437 		    (EXT4_XATTR_SIZE(i.value_len) >
2438 			EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2439 			i.in_inode = 1;
2440 retry_inode:
2441 		error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2442 		if (!error && !bs.s.not_found) {
2443 			i.value = NULL;
2444 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2445 		} else if (error == -ENOSPC) {
2446 			if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2447 				brelse(bs.bh);
2448 				bs.bh = NULL;
2449 				error = ext4_xattr_block_find(inode, &i, &bs);
2450 				if (error)
2451 					goto cleanup;
2452 			}
2453 			error = ext4_xattr_block_set(handle, inode, &i, &bs);
2454 			if (!error && !is.s.not_found) {
2455 				i.value = NULL;
2456 				error = ext4_xattr_ibody_set(handle, inode, &i,
2457 							     &is);
2458 			} else if (error == -ENOSPC) {
2459 				/*
2460 				 * Xattr does not fit in the block, store at
2461 				 * external inode if possible.
2462 				 */
2463 				if (ext4_has_feature_ea_inode(inode->i_sb) &&
2464 				    i.value_len && !i.in_inode) {
2465 					i.in_inode = 1;
2466 					goto retry_inode;
2467 				}
2468 			}
2469 		}
2470 	}
2471 	if (!error) {
2472 		ext4_xattr_update_super_block(handle, inode->i_sb);
2473 		inode_set_ctime_current(inode);
2474 		inode_inc_iversion(inode);
2475 		if (!value)
2476 			no_expand = 0;
2477 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2478 		/*
2479 		 * The bh is consumed by ext4_mark_iloc_dirty, even with
2480 		 * error != 0.
2481 		 */
2482 		is.iloc.bh = NULL;
2483 		if (IS_SYNC(inode))
2484 			ext4_handle_sync(handle);
2485 	}
2486 	ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2487 
2488 cleanup:
2489 	brelse(is.iloc.bh);
2490 	brelse(bs.bh);
2491 	ext4_write_unlock_xattr(inode, &no_expand);
2492 	return error;
2493 }
2494 
ext4_xattr_set_credits(struct inode * inode,size_t value_len,bool is_create,int * credits)2495 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2496 			   bool is_create, int *credits)
2497 {
2498 	struct buffer_head *bh;
2499 	int err;
2500 
2501 	*credits = 0;
2502 
2503 	if (!EXT4_SB(inode->i_sb)->s_journal)
2504 		return 0;
2505 
2506 	down_read(&EXT4_I(inode)->xattr_sem);
2507 
2508 	bh = ext4_xattr_get_block(inode);
2509 	if (IS_ERR(bh)) {
2510 		err = PTR_ERR(bh);
2511 	} else {
2512 		*credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2513 						    value_len, is_create);
2514 		brelse(bh);
2515 		err = 0;
2516 	}
2517 
2518 	up_read(&EXT4_I(inode)->xattr_sem);
2519 	return err;
2520 }
2521 
2522 /*
2523  * ext4_xattr_set()
2524  *
2525  * Like ext4_xattr_set_handle, but start from an inode. This extended
2526  * attribute modification is a filesystem transaction by itself.
2527  *
2528  * Returns 0, or a negative error number on failure.
2529  */
2530 int
ext4_xattr_set(struct inode * inode,int name_index,const char * name,const void * value,size_t value_len,int flags)2531 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2532 	       const void *value, size_t value_len, int flags)
2533 {
2534 	handle_t *handle;
2535 	struct super_block *sb = inode->i_sb;
2536 	int error, retries = 0;
2537 	int credits;
2538 
2539 	error = dquot_initialize(inode);
2540 	if (error)
2541 		return error;
2542 
2543 retry:
2544 	error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2545 				       &credits);
2546 	if (error)
2547 		return error;
2548 
2549 	handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2550 	if (IS_ERR(handle)) {
2551 		error = PTR_ERR(handle);
2552 	} else {
2553 		int error2;
2554 
2555 		error = ext4_xattr_set_handle(handle, inode, name_index, name,
2556 					      value, value_len, flags);
2557 		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR,
2558 					handle);
2559 		error2 = ext4_journal_stop(handle);
2560 		if (error == -ENOSPC &&
2561 		    ext4_should_retry_alloc(sb, &retries))
2562 			goto retry;
2563 		if (error == 0)
2564 			error = error2;
2565 	}
2566 
2567 	return error;
2568 }
2569 
2570 /*
2571  * Shift the EA entries in the inode to create space for the increased
2572  * i_extra_isize.
2573  */
ext4_xattr_shift_entries(struct ext4_xattr_entry * entry,int value_offs_shift,void * to,void * from,size_t n)2574 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2575 				     int value_offs_shift, void *to,
2576 				     void *from, size_t n)
2577 {
2578 	struct ext4_xattr_entry *last = entry;
2579 	int new_offs;
2580 
2581 	/* We always shift xattr headers further thus offsets get lower */
2582 	BUG_ON(value_offs_shift > 0);
2583 
2584 	/* Adjust the value offsets of the entries */
2585 	for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2586 		if (!last->e_value_inum && last->e_value_size) {
2587 			new_offs = le16_to_cpu(last->e_value_offs) +
2588 							value_offs_shift;
2589 			last->e_value_offs = cpu_to_le16(new_offs);
2590 		}
2591 	}
2592 	/* Shift the entries by n bytes */
2593 	memmove(to, from, n);
2594 }
2595 
2596 /*
2597  * Move xattr pointed to by 'entry' from inode into external xattr block
2598  */
ext4_xattr_move_to_block(handle_t * handle,struct inode * inode,struct ext4_inode * raw_inode,struct ext4_xattr_entry * entry)2599 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2600 				    struct ext4_inode *raw_inode,
2601 				    struct ext4_xattr_entry *entry)
2602 {
2603 	struct ext4_xattr_ibody_find *is = NULL;
2604 	struct ext4_xattr_block_find *bs = NULL;
2605 	char *buffer = NULL, *b_entry_name = NULL;
2606 	size_t value_size = le32_to_cpu(entry->e_value_size);
2607 	struct ext4_xattr_info i = {
2608 		.value = NULL,
2609 		.value_len = 0,
2610 		.name_index = entry->e_name_index,
2611 		.in_inode = !!entry->e_value_inum,
2612 	};
2613 	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2614 	int needs_kvfree = 0;
2615 	int error;
2616 
2617 	is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2618 	bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2619 	b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2620 	if (!is || !bs || !b_entry_name) {
2621 		error = -ENOMEM;
2622 		goto out;
2623 	}
2624 
2625 	is->s.not_found = -ENODATA;
2626 	bs->s.not_found = -ENODATA;
2627 	is->iloc.bh = NULL;
2628 	bs->bh = NULL;
2629 
2630 	/* Save the entry name and the entry value */
2631 	if (entry->e_value_inum) {
2632 		buffer = kvmalloc(value_size, GFP_NOFS);
2633 		if (!buffer) {
2634 			error = -ENOMEM;
2635 			goto out;
2636 		}
2637 		needs_kvfree = 1;
2638 		error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2639 		if (error)
2640 			goto out;
2641 	} else {
2642 		size_t value_offs = le16_to_cpu(entry->e_value_offs);
2643 		buffer = (void *)IFIRST(header) + value_offs;
2644 	}
2645 
2646 	memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2647 	b_entry_name[entry->e_name_len] = '\0';
2648 	i.name = b_entry_name;
2649 
2650 	error = ext4_get_inode_loc(inode, &is->iloc);
2651 	if (error)
2652 		goto out;
2653 
2654 	error = ext4_xattr_ibody_find(inode, &i, is);
2655 	if (error)
2656 		goto out;
2657 
2658 	i.value = buffer;
2659 	i.value_len = value_size;
2660 	error = ext4_xattr_block_find(inode, &i, bs);
2661 	if (error)
2662 		goto out;
2663 
2664 	/* Move ea entry from the inode into the block */
2665 	error = ext4_xattr_block_set(handle, inode, &i, bs);
2666 	if (error)
2667 		goto out;
2668 
2669 	/* Remove the chosen entry from the inode */
2670 	i.value = NULL;
2671 	i.value_len = 0;
2672 	error = ext4_xattr_ibody_set(handle, inode, &i, is);
2673 
2674 out:
2675 	kfree(b_entry_name);
2676 	if (needs_kvfree && buffer)
2677 		kvfree(buffer);
2678 	if (is)
2679 		brelse(is->iloc.bh);
2680 	if (bs)
2681 		brelse(bs->bh);
2682 	kfree(is);
2683 	kfree(bs);
2684 
2685 	return error;
2686 }
2687 
ext4_xattr_make_inode_space(handle_t * handle,struct inode * inode,struct ext4_inode * raw_inode,int isize_diff,size_t ifree,size_t bfree,int * total_ino)2688 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2689 				       struct ext4_inode *raw_inode,
2690 				       int isize_diff, size_t ifree,
2691 				       size_t bfree, int *total_ino)
2692 {
2693 	struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2694 	struct ext4_xattr_entry *small_entry;
2695 	struct ext4_xattr_entry *entry;
2696 	struct ext4_xattr_entry *last;
2697 	unsigned int entry_size;	/* EA entry size */
2698 	unsigned int total_size;	/* EA entry size + value size */
2699 	unsigned int min_total_size;
2700 	int error;
2701 
2702 	while (isize_diff > ifree) {
2703 		entry = NULL;
2704 		small_entry = NULL;
2705 		min_total_size = ~0U;
2706 		last = IFIRST(header);
2707 		/* Find the entry best suited to be pushed into EA block */
2708 		for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2709 			/* never move system.data out of the inode */
2710 			if ((last->e_name_len == 4) &&
2711 			    (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2712 			    !memcmp(last->e_name, "data", 4))
2713 				continue;
2714 			total_size = EXT4_XATTR_LEN(last->e_name_len);
2715 			if (!last->e_value_inum)
2716 				total_size += EXT4_XATTR_SIZE(
2717 					       le32_to_cpu(last->e_value_size));
2718 			if (total_size <= bfree &&
2719 			    total_size < min_total_size) {
2720 				if (total_size + ifree < isize_diff) {
2721 					small_entry = last;
2722 				} else {
2723 					entry = last;
2724 					min_total_size = total_size;
2725 				}
2726 			}
2727 		}
2728 
2729 		if (entry == NULL) {
2730 			if (small_entry == NULL)
2731 				return -ENOSPC;
2732 			entry = small_entry;
2733 		}
2734 
2735 		entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2736 		total_size = entry_size;
2737 		if (!entry->e_value_inum)
2738 			total_size += EXT4_XATTR_SIZE(
2739 					      le32_to_cpu(entry->e_value_size));
2740 		error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2741 						 entry);
2742 		if (error)
2743 			return error;
2744 
2745 		*total_ino -= entry_size;
2746 		ifree += total_size;
2747 		bfree -= total_size;
2748 	}
2749 
2750 	return 0;
2751 }
2752 
2753 /*
2754  * Expand an inode by new_extra_isize bytes when EAs are present.
2755  * Returns 0 on success or negative error number on failure.
2756  */
ext4_expand_extra_isize_ea(struct inode * inode,int new_extra_isize,struct ext4_inode * raw_inode,handle_t * handle)2757 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2758 			       struct ext4_inode *raw_inode, handle_t *handle)
2759 {
2760 	struct ext4_xattr_ibody_header *header;
2761 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2762 	static unsigned int mnt_count;
2763 	size_t min_offs;
2764 	size_t ifree, bfree;
2765 	int total_ino;
2766 	void *base, *end;
2767 	int error = 0, tried_min_extra_isize = 0;
2768 	int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2769 	int isize_diff;	/* How much do we need to grow i_extra_isize */
2770 
2771 retry:
2772 	isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2773 	if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2774 		return 0;
2775 
2776 	header = IHDR(inode, raw_inode);
2777 
2778 	/*
2779 	 * Check if enough free space is available in the inode to shift the
2780 	 * entries ahead by new_extra_isize.
2781 	 */
2782 
2783 	base = IFIRST(header);
2784 	end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2785 	min_offs = end - base;
2786 	total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2787 
2788 	error = xattr_check_inode(inode, header, end);
2789 	if (error)
2790 		goto cleanup;
2791 
2792 	ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2793 	if (ifree >= isize_diff)
2794 		goto shift;
2795 
2796 	/*
2797 	 * Enough free space isn't available in the inode, check if
2798 	 * EA block can hold new_extra_isize bytes.
2799 	 */
2800 	if (EXT4_I(inode)->i_file_acl) {
2801 		struct buffer_head *bh;
2802 
2803 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2804 		if (IS_ERR(bh)) {
2805 			error = PTR_ERR(bh);
2806 			goto cleanup;
2807 		}
2808 		error = ext4_xattr_check_block(inode, bh);
2809 		if (error) {
2810 			brelse(bh);
2811 			goto cleanup;
2812 		}
2813 		base = BHDR(bh);
2814 		end = bh->b_data + bh->b_size;
2815 		min_offs = end - base;
2816 		bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2817 					      NULL);
2818 		brelse(bh);
2819 		if (bfree + ifree < isize_diff) {
2820 			if (!tried_min_extra_isize && s_min_extra_isize) {
2821 				tried_min_extra_isize++;
2822 				new_extra_isize = s_min_extra_isize;
2823 				goto retry;
2824 			}
2825 			error = -ENOSPC;
2826 			goto cleanup;
2827 		}
2828 	} else {
2829 		bfree = inode->i_sb->s_blocksize;
2830 	}
2831 
2832 	error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2833 					    isize_diff, ifree, bfree,
2834 					    &total_ino);
2835 	if (error) {
2836 		if (error == -ENOSPC && !tried_min_extra_isize &&
2837 		    s_min_extra_isize) {
2838 			tried_min_extra_isize++;
2839 			new_extra_isize = s_min_extra_isize;
2840 			goto retry;
2841 		}
2842 		goto cleanup;
2843 	}
2844 shift:
2845 	/* Adjust the offsets and shift the remaining entries ahead */
2846 	ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2847 			- new_extra_isize, (void *)raw_inode +
2848 			EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2849 			(void *)header, total_ino);
2850 	EXT4_I(inode)->i_extra_isize = new_extra_isize;
2851 
2852 	if (ext4_has_inline_data(inode))
2853 		error = ext4_find_inline_data_nolock(inode);
2854 
2855 cleanup:
2856 	if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2857 		ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2858 			     inode->i_ino);
2859 		mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2860 	}
2861 	return error;
2862 }
2863 
2864 #define EIA_INCR 16 /* must be 2^n */
2865 #define EIA_MASK (EIA_INCR - 1)
2866 
2867 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2868  * If @ea_inode_array is new or full it will be grown and the old
2869  * contents copied over.
2870  */
2871 static int
ext4_expand_inode_array(struct ext4_xattr_inode_array ** ea_inode_array,struct inode * inode)2872 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2873 			struct inode *inode)
2874 {
2875 	if (*ea_inode_array == NULL) {
2876 		/*
2877 		 * Start with 15 inodes, so it fits into a power-of-two size.
2878 		 * If *ea_inode_array is NULL, this is essentially offsetof()
2879 		 */
2880 		(*ea_inode_array) =
2881 			kmalloc(offsetof(struct ext4_xattr_inode_array,
2882 					 inodes[EIA_MASK]),
2883 				GFP_NOFS);
2884 		if (*ea_inode_array == NULL)
2885 			return -ENOMEM;
2886 		(*ea_inode_array)->count = 0;
2887 	} else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2888 		/* expand the array once all 15 + n * 16 slots are full */
2889 		struct ext4_xattr_inode_array *new_array = NULL;
2890 		int count = (*ea_inode_array)->count;
2891 
2892 		/* if new_array is NULL, this is essentially offsetof() */
2893 		new_array = kmalloc(
2894 				offsetof(struct ext4_xattr_inode_array,
2895 					 inodes[count + EIA_INCR]),
2896 				GFP_NOFS);
2897 		if (new_array == NULL)
2898 			return -ENOMEM;
2899 		memcpy(new_array, *ea_inode_array,
2900 		       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2901 		kfree(*ea_inode_array);
2902 		*ea_inode_array = new_array;
2903 	}
2904 	(*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2905 	return 0;
2906 }
2907 
2908 /*
2909  * ext4_xattr_delete_inode()
2910  *
2911  * Free extended attribute resources associated with this inode. Traverse
2912  * all entries and decrement reference on any xattr inodes associated with this
2913  * inode. This is called immediately before an inode is freed. We have exclusive
2914  * access to the inode. If an orphan inode is deleted it will also release its
2915  * references on xattr block and xattr inodes.
2916  */
ext4_xattr_delete_inode(handle_t * handle,struct inode * inode,struct ext4_xattr_inode_array ** ea_inode_array,int extra_credits)2917 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2918 			    struct ext4_xattr_inode_array **ea_inode_array,
2919 			    int extra_credits)
2920 {
2921 	struct buffer_head *bh = NULL;
2922 	struct ext4_xattr_ibody_header *header;
2923 	struct ext4_iloc iloc = { .bh = NULL };
2924 	struct ext4_xattr_entry *entry;
2925 	struct inode *ea_inode;
2926 	int error;
2927 
2928 	error = ext4_journal_ensure_credits(handle, extra_credits,
2929 			ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2930 	if (error < 0) {
2931 		EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2932 		goto cleanup;
2933 	}
2934 
2935 	if (ext4_has_feature_ea_inode(inode->i_sb) &&
2936 	    ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2937 
2938 		error = ext4_get_inode_loc(inode, &iloc);
2939 		if (error) {
2940 			EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2941 			goto cleanup;
2942 		}
2943 
2944 		error = ext4_journal_get_write_access(handle, inode->i_sb,
2945 						iloc.bh, EXT4_JTR_NONE);
2946 		if (error) {
2947 			EXT4_ERROR_INODE(inode, "write access (error %d)",
2948 					 error);
2949 			goto cleanup;
2950 		}
2951 
2952 		header = IHDR(inode, ext4_raw_inode(&iloc));
2953 		if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2954 			ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2955 						     IFIRST(header),
2956 						     false /* block_csum */,
2957 						     ea_inode_array,
2958 						     extra_credits,
2959 						     false /* skip_quota */);
2960 	}
2961 
2962 	if (EXT4_I(inode)->i_file_acl) {
2963 		bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2964 		if (IS_ERR(bh)) {
2965 			error = PTR_ERR(bh);
2966 			if (error == -EIO) {
2967 				EXT4_ERROR_INODE_ERR(inode, EIO,
2968 						     "block %llu read error",
2969 						     EXT4_I(inode)->i_file_acl);
2970 			}
2971 			bh = NULL;
2972 			goto cleanup;
2973 		}
2974 		error = ext4_xattr_check_block(inode, bh);
2975 		if (error)
2976 			goto cleanup;
2977 
2978 		if (ext4_has_feature_ea_inode(inode->i_sb)) {
2979 			for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2980 			     entry = EXT4_XATTR_NEXT(entry)) {
2981 				if (!entry->e_value_inum)
2982 					continue;
2983 				error = ext4_xattr_inode_iget(inode,
2984 					      le32_to_cpu(entry->e_value_inum),
2985 					      le32_to_cpu(entry->e_hash),
2986 					      &ea_inode);
2987 				if (error)
2988 					continue;
2989 				ext4_xattr_inode_free_quota(inode, ea_inode,
2990 					      le32_to_cpu(entry->e_value_size));
2991 				iput(ea_inode);
2992 			}
2993 
2994 		}
2995 
2996 		ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2997 					 extra_credits);
2998 		/*
2999 		 * Update i_file_acl value in the same transaction that releases
3000 		 * block.
3001 		 */
3002 		EXT4_I(inode)->i_file_acl = 0;
3003 		error = ext4_mark_inode_dirty(handle, inode);
3004 		if (error) {
3005 			EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
3006 					 error);
3007 			goto cleanup;
3008 		}
3009 		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
3010 	}
3011 	error = 0;
3012 cleanup:
3013 	brelse(iloc.bh);
3014 	brelse(bh);
3015 	return error;
3016 }
3017 
ext4_xattr_inode_array_free(struct ext4_xattr_inode_array * ea_inode_array)3018 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
3019 {
3020 	int idx;
3021 
3022 	if (ea_inode_array == NULL)
3023 		return;
3024 
3025 	for (idx = 0; idx < ea_inode_array->count; ++idx)
3026 		iput(ea_inode_array->inodes[idx]);
3027 	kfree(ea_inode_array);
3028 }
3029 
3030 /*
3031  * ext4_xattr_block_cache_insert()
3032  *
3033  * Create a new entry in the extended attribute block cache, and insert
3034  * it unless such an entry is already in the cache.
3035  *
3036  * Returns 0, or a negative error number on failure.
3037  */
3038 static void
ext4_xattr_block_cache_insert(struct mb_cache * ea_block_cache,struct buffer_head * bh)3039 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3040 			      struct buffer_head *bh)
3041 {
3042 	struct ext4_xattr_header *header = BHDR(bh);
3043 	__u32 hash = le32_to_cpu(header->h_hash);
3044 	int reusable = le32_to_cpu(header->h_refcount) <
3045 		       EXT4_XATTR_REFCOUNT_MAX;
3046 	int error;
3047 
3048 	if (!ea_block_cache)
3049 		return;
3050 	error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3051 				      bh->b_blocknr, reusable);
3052 	if (error) {
3053 		if (error == -EBUSY)
3054 			ea_bdebug(bh, "already in cache");
3055 	} else
3056 		ea_bdebug(bh, "inserting [%x]", (int)hash);
3057 }
3058 
3059 /*
3060  * ext4_xattr_cmp()
3061  *
3062  * Compare two extended attribute blocks for equality.
3063  *
3064  * Returns 0 if the blocks are equal, 1 if they differ, and
3065  * a negative error number on errors.
3066  */
3067 static int
ext4_xattr_cmp(struct ext4_xattr_header * header1,struct ext4_xattr_header * header2)3068 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3069 	       struct ext4_xattr_header *header2)
3070 {
3071 	struct ext4_xattr_entry *entry1, *entry2;
3072 
3073 	entry1 = ENTRY(header1+1);
3074 	entry2 = ENTRY(header2+1);
3075 	while (!IS_LAST_ENTRY(entry1)) {
3076 		if (IS_LAST_ENTRY(entry2))
3077 			return 1;
3078 		if (entry1->e_hash != entry2->e_hash ||
3079 		    entry1->e_name_index != entry2->e_name_index ||
3080 		    entry1->e_name_len != entry2->e_name_len ||
3081 		    entry1->e_value_size != entry2->e_value_size ||
3082 		    entry1->e_value_inum != entry2->e_value_inum ||
3083 		    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3084 			return 1;
3085 		if (!entry1->e_value_inum &&
3086 		    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3087 			   (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3088 			   le32_to_cpu(entry1->e_value_size)))
3089 			return 1;
3090 
3091 		entry1 = EXT4_XATTR_NEXT(entry1);
3092 		entry2 = EXT4_XATTR_NEXT(entry2);
3093 	}
3094 	if (!IS_LAST_ENTRY(entry2))
3095 		return 1;
3096 	return 0;
3097 }
3098 
3099 /*
3100  * ext4_xattr_block_cache_find()
3101  *
3102  * Find an identical extended attribute block.
3103  *
3104  * Returns a pointer to the block found, or NULL if such a block was
3105  * not found or an error occurred.
3106  */
3107 static struct buffer_head *
ext4_xattr_block_cache_find(struct inode * inode,struct ext4_xattr_header * header,struct mb_cache_entry ** pce)3108 ext4_xattr_block_cache_find(struct inode *inode,
3109 			    struct ext4_xattr_header *header,
3110 			    struct mb_cache_entry **pce)
3111 {
3112 	__u32 hash = le32_to_cpu(header->h_hash);
3113 	struct mb_cache_entry *ce;
3114 	struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3115 
3116 	if (!ea_block_cache)
3117 		return NULL;
3118 	if (!header->h_hash)
3119 		return NULL;  /* never share */
3120 	ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3121 	ce = mb_cache_entry_find_first(ea_block_cache, hash);
3122 	while (ce) {
3123 		struct buffer_head *bh;
3124 
3125 		bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3126 		if (IS_ERR(bh)) {
3127 			if (PTR_ERR(bh) == -ENOMEM) {
3128 				mb_cache_entry_put(ea_block_cache, ce);
3129 				return NULL;
3130 			}
3131 			bh = NULL;
3132 			EXT4_ERROR_INODE(inode, "block %lu read error",
3133 					 (unsigned long)ce->e_value);
3134 		} else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3135 			*pce = ce;
3136 			return bh;
3137 		}
3138 		brelse(bh);
3139 		ce = mb_cache_entry_find_next(ea_block_cache, ce);
3140 	}
3141 	return NULL;
3142 }
3143 
3144 #define NAME_HASH_SHIFT 5
3145 #define VALUE_HASH_SHIFT 16
3146 
3147 /*
3148  * ext4_xattr_hash_entry()
3149  *
3150  * Compute the hash of an extended attribute.
3151  */
ext4_xattr_hash_entry(char * name,size_t name_len,__le32 * value,size_t value_count)3152 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3153 				    size_t value_count)
3154 {
3155 	__u32 hash = 0;
3156 
3157 	while (name_len--) {
3158 		hash = (hash << NAME_HASH_SHIFT) ^
3159 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3160 		       (unsigned char)*name++;
3161 	}
3162 	while (value_count--) {
3163 		hash = (hash << VALUE_HASH_SHIFT) ^
3164 		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3165 		       le32_to_cpu(*value++);
3166 	}
3167 	return cpu_to_le32(hash);
3168 }
3169 
3170 /*
3171  * ext4_xattr_hash_entry_signed()
3172  *
3173  * Compute the hash of an extended attribute incorrectly.
3174  */
ext4_xattr_hash_entry_signed(char * name,size_t name_len,__le32 * value,size_t value_count)3175 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3176 {
3177 	__u32 hash = 0;
3178 
3179 	while (name_len--) {
3180 		hash = (hash << NAME_HASH_SHIFT) ^
3181 		       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3182 		       (signed char)*name++;
3183 	}
3184 	while (value_count--) {
3185 		hash = (hash << VALUE_HASH_SHIFT) ^
3186 		       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3187 		       le32_to_cpu(*value++);
3188 	}
3189 	return cpu_to_le32(hash);
3190 }
3191 
3192 #undef NAME_HASH_SHIFT
3193 #undef VALUE_HASH_SHIFT
3194 
3195 #define BLOCK_HASH_SHIFT 16
3196 
3197 /*
3198  * ext4_xattr_rehash()
3199  *
3200  * Re-compute the extended attribute hash value after an entry has changed.
3201  */
ext4_xattr_rehash(struct ext4_xattr_header * header)3202 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3203 {
3204 	struct ext4_xattr_entry *here;
3205 	__u32 hash = 0;
3206 
3207 	here = ENTRY(header+1);
3208 	while (!IS_LAST_ENTRY(here)) {
3209 		if (!here->e_hash) {
3210 			/* Block is not shared if an entry's hash value == 0 */
3211 			hash = 0;
3212 			break;
3213 		}
3214 		hash = (hash << BLOCK_HASH_SHIFT) ^
3215 		       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3216 		       le32_to_cpu(here->e_hash);
3217 		here = EXT4_XATTR_NEXT(here);
3218 	}
3219 	header->h_hash = cpu_to_le32(hash);
3220 }
3221 
3222 #undef BLOCK_HASH_SHIFT
3223 
3224 #define	HASH_BUCKET_BITS	10
3225 
3226 struct mb_cache *
ext4_xattr_create_cache(void)3227 ext4_xattr_create_cache(void)
3228 {
3229 	return mb_cache_create(HASH_BUCKET_BITS);
3230 }
3231 
ext4_xattr_destroy_cache(struct mb_cache * cache)3232 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3233 {
3234 	if (cache)
3235 		mb_cache_destroy(cache);
3236 }
3237 
3238