ext4: fix invalid free tracking in ext4_xattr_move_to_block()
[linux-2.6-block.git] / fs / ext4 / xattr.c
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
122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
123 {
124         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
125 }
126 #endif
127
128 static __le32 ext4_xattr_block_csum(struct inode *inode,
129                                     sector_t block_nr,
130                                     struct ext4_xattr_header *hdr)
131 {
132         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
133         __u32 csum;
134         __le64 dsk_block_nr = cpu_to_le64(block_nr);
135         __u32 dummy_csum = 0;
136         int offset = offsetof(struct ext4_xattr_header, h_checksum);
137
138         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
139                            sizeof(dsk_block_nr));
140         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
141         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
142         offset += sizeof(dummy_csum);
143         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
144                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
145
146         return cpu_to_le32(csum);
147 }
148
149 static int ext4_xattr_block_csum_verify(struct inode *inode,
150                                         struct buffer_head *bh)
151 {
152         struct ext4_xattr_header *hdr = BHDR(bh);
153         int ret = 1;
154
155         if (ext4_has_metadata_csum(inode->i_sb)) {
156                 lock_buffer(bh);
157                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
158                                                         bh->b_blocknr, hdr));
159                 unlock_buffer(bh);
160         }
161         return ret;
162 }
163
164 static void ext4_xattr_block_csum_set(struct inode *inode,
165                                       struct buffer_head *bh)
166 {
167         if (ext4_has_metadata_csum(inode->i_sb))
168                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
169                                                 bh->b_blocknr, BHDR(bh));
170 }
171
172 static inline const char *ext4_xattr_prefix(int name_index,
173                                             struct dentry *dentry)
174 {
175         const struct xattr_handler *handler = NULL;
176
177         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
178                 handler = ext4_xattr_handler_map[name_index];
179
180         if (!xattr_handler_can_list(handler, dentry))
181                 return NULL;
182
183         return xattr_prefix(handler);
184 }
185
186 static int
187 check_xattrs(struct inode *inode, struct buffer_head *bh,
188              struct ext4_xattr_entry *entry, void *end, void *value_start,
189              const char *function, unsigned int line)
190 {
191         struct ext4_xattr_entry *e = entry;
192         int err = -EFSCORRUPTED;
193         char *err_str;
194
195         if (bh) {
196                 if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
197                     BHDR(bh)->h_blocks != cpu_to_le32(1)) {
198                         err_str = "invalid header";
199                         goto errout;
200                 }
201                 if (buffer_verified(bh))
202                         return 0;
203                 if (!ext4_xattr_block_csum_verify(inode, bh)) {
204                         err = -EFSBADCRC;
205                         err_str = "invalid checksum";
206                         goto errout;
207                 }
208         } else {
209                 struct ext4_xattr_ibody_header *header = value_start;
210
211                 header -= 1;
212                 if (end - (void *)header < sizeof(*header) + sizeof(u32)) {
213                         err_str = "in-inode xattr block too small";
214                         goto errout;
215                 }
216                 if (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
217                         err_str = "bad magic number in in-inode xattr";
218                         goto errout;
219                 }
220         }
221
222         /* Find the end of the names list */
223         while (!IS_LAST_ENTRY(e)) {
224                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
225                 if ((void *)next >= end) {
226                         err_str = "e_name out of bounds";
227                         goto errout;
228                 }
229                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len) {
230                         err_str = "bad e_name length";
231                         goto errout;
232                 }
233                 e = next;
234         }
235
236         /* Check the values */
237         while (!IS_LAST_ENTRY(entry)) {
238                 u32 size = le32_to_cpu(entry->e_value_size);
239                 unsigned long ea_ino = le32_to_cpu(entry->e_value_inum);
240
241                 if (!ext4_has_feature_ea_inode(inode->i_sb) && ea_ino) {
242                         err_str = "ea_inode specified without ea_inode feature enabled";
243                         goto errout;
244                 }
245                 if (ea_ino && ((ea_ino == EXT4_ROOT_INO) ||
246                                !ext4_valid_inum(inode->i_sb, ea_ino))) {
247                         err_str = "invalid ea_ino";
248                         goto errout;
249                 }
250                 if (size > EXT4_XATTR_SIZE_MAX) {
251                         err_str = "e_value size too large";
252                         goto errout;
253                 }
254
255                 if (size != 0 && entry->e_value_inum == 0) {
256                         u16 offs = le16_to_cpu(entry->e_value_offs);
257                         void *value;
258
259                         /*
260                          * The value cannot overlap the names, and the value
261                          * with padding cannot extend beyond 'end'.  Check both
262                          * the padded and unpadded sizes, since the size may
263                          * overflow to 0 when adding padding.
264                          */
265                         if (offs > end - value_start) {
266                                 err_str = "e_value out of bounds";
267                                 goto errout;
268                         }
269                         value = value_start + offs;
270                         if (value < (void *)e + sizeof(u32) ||
271                             size > end - value ||
272                             EXT4_XATTR_SIZE(size) > end - value) {
273                                 err_str = "overlapping e_value ";
274                                 goto errout;
275                         }
276                 }
277                 entry = EXT4_XATTR_NEXT(entry);
278         }
279         if (bh)
280                 set_buffer_verified(bh);
281         return 0;
282
283 errout:
284         if (bh)
285                 __ext4_error_inode(inode, function, line, 0, -err,
286                                    "corrupted xattr block %llu: %s",
287                                    (unsigned long long) bh->b_blocknr,
288                                    err_str);
289         else
290                 __ext4_error_inode(inode, function, line, 0, -err,
291                                    "corrupted in-inode xattr: %s", err_str);
292         return err;
293 }
294
295 static inline int
296 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
297                          const char *function, unsigned int line)
298 {
299         return check_xattrs(inode, bh, BFIRST(bh), bh->b_data + bh->b_size,
300                             bh->b_data, function, line);
301 }
302
303 #define ext4_xattr_check_block(inode, bh) \
304         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
305
306
307 static inline int
308 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
309                          void *end, const char *function, unsigned int line)
310 {
311         return check_xattrs(inode, NULL, IFIRST(header), end, IFIRST(header),
312                             function, line);
313 }
314
315 #define xattr_check_inode(inode, header, end) \
316         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
317
318 static int
319 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
320                  void *end, int name_index, const char *name, int sorted)
321 {
322         struct ext4_xattr_entry *entry, *next;
323         size_t name_len;
324         int cmp = 1;
325
326         if (name == NULL)
327                 return -EINVAL;
328         name_len = strlen(name);
329         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
330                 next = EXT4_XATTR_NEXT(entry);
331                 if ((void *) next >= end) {
332                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
333                         return -EFSCORRUPTED;
334                 }
335                 cmp = name_index - entry->e_name_index;
336                 if (!cmp)
337                         cmp = name_len - entry->e_name_len;
338                 if (!cmp)
339                         cmp = memcmp(name, entry->e_name, name_len);
340                 if (cmp <= 0 && (sorted || cmp == 0))
341                         break;
342         }
343         *pentry = entry;
344         return cmp ? -ENODATA : 0;
345 }
346
347 static u32
348 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
349 {
350         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
351 }
352
353 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
354 {
355         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
356                 (u32) inode_peek_iversion_raw(ea_inode);
357 }
358
359 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
360 {
361         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
362         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
363 }
364
365 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
366 {
367         return (u32)ea_inode->i_atime.tv_sec;
368 }
369
370 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
371 {
372         ea_inode->i_atime.tv_sec = hash;
373 }
374
375 /*
376  * Read the EA value from an inode.
377  */
378 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
379 {
380         int blocksize = 1 << ea_inode->i_blkbits;
381         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
382         int tail_size = (size % blocksize) ?: blocksize;
383         struct buffer_head *bhs_inline[8];
384         struct buffer_head **bhs = bhs_inline;
385         int i, ret;
386
387         if (bh_count > ARRAY_SIZE(bhs_inline)) {
388                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
389                 if (!bhs)
390                         return -ENOMEM;
391         }
392
393         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
394                                true /* wait */, bhs);
395         if (ret)
396                 goto free_bhs;
397
398         for (i = 0; i < bh_count; i++) {
399                 /* There shouldn't be any holes in ea_inode. */
400                 if (!bhs[i]) {
401                         ret = -EFSCORRUPTED;
402                         goto put_bhs;
403                 }
404                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
405                        i < bh_count - 1 ? blocksize : tail_size);
406         }
407         ret = 0;
408 put_bhs:
409         for (i = 0; i < bh_count; i++)
410                 brelse(bhs[i]);
411 free_bhs:
412         if (bhs != bhs_inline)
413                 kfree(bhs);
414         return ret;
415 }
416
417 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
418
419 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
420                                  u32 ea_inode_hash, struct inode **ea_inode)
421 {
422         struct inode *inode;
423         int err;
424
425         /*
426          * We have to check for this corruption early as otherwise
427          * iget_locked() could wait indefinitely for the state of our
428          * parent inode.
429          */
430         if (parent->i_ino == ea_ino) {
431                 ext4_error(parent->i_sb,
432                            "Parent and EA inode have the same ino %lu", ea_ino);
433                 return -EFSCORRUPTED;
434         }
435
436         inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
437         if (IS_ERR(inode)) {
438                 err = PTR_ERR(inode);
439                 ext4_error(parent->i_sb,
440                            "error while reading EA inode %lu err=%d", ea_ino,
441                            err);
442                 return err;
443         }
444
445         if (is_bad_inode(inode)) {
446                 ext4_error(parent->i_sb,
447                            "error while reading EA inode %lu is_bad_inode",
448                            ea_ino);
449                 err = -EIO;
450                 goto error;
451         }
452
453         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
454                 ext4_error(parent->i_sb,
455                            "EA inode %lu does not have EXT4_EA_INODE_FL flag",
456                             ea_ino);
457                 err = -EINVAL;
458                 goto error;
459         }
460
461         ext4_xattr_inode_set_class(inode);
462
463         /*
464          * Check whether this is an old Lustre-style xattr inode. Lustre
465          * implementation does not have hash validation, rather it has a
466          * backpointer from ea_inode to the parent inode.
467          */
468         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
469             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
470             inode->i_generation == parent->i_generation) {
471                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
472                 ext4_xattr_inode_set_ref(inode, 1);
473         } else {
474                 inode_lock(inode);
475                 inode->i_flags |= S_NOQUOTA;
476                 inode_unlock(inode);
477         }
478
479         *ea_inode = inode;
480         return 0;
481 error:
482         iput(inode);
483         return err;
484 }
485
486 /* Remove entry from mbcache when EA inode is getting evicted */
487 void ext4_evict_ea_inode(struct inode *inode)
488 {
489         struct mb_cache_entry *oe;
490
491         if (!EA_INODE_CACHE(inode))
492                 return;
493         /* Wait for entry to get unused so that we can remove it */
494         while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
495                         ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
496                 mb_cache_entry_wait_unused(oe);
497                 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
498         }
499 }
500
501 static int
502 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
503                                struct ext4_xattr_entry *entry, void *buffer,
504                                size_t size)
505 {
506         u32 hash;
507
508         /* Verify stored hash matches calculated hash. */
509         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
510         if (hash != ext4_xattr_inode_get_hash(ea_inode))
511                 return -EFSCORRUPTED;
512
513         if (entry) {
514                 __le32 e_hash, tmp_data;
515
516                 /* Verify entry hash. */
517                 tmp_data = cpu_to_le32(hash);
518                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
519                                                &tmp_data, 1);
520                 /* All good? */
521                 if (e_hash == entry->e_hash)
522                         return 0;
523
524                 /*
525                  * Not good. Maybe the entry hash was calculated
526                  * using the buggy signed char version?
527                  */
528                 e_hash = ext4_xattr_hash_entry_signed(entry->e_name, entry->e_name_len,
529                                                         &tmp_data, 1);
530                 /* Still no match - bad */
531                 if (e_hash != entry->e_hash)
532                         return -EFSCORRUPTED;
533
534                 /* Let people know about old hash */
535                 pr_warn_once("ext4: filesystem with signed xattr name hash");
536         }
537         return 0;
538 }
539
540 /*
541  * Read xattr value from the EA inode.
542  */
543 static int
544 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
545                      void *buffer, size_t size)
546 {
547         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
548         struct inode *ea_inode;
549         int err;
550
551         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
552                                     le32_to_cpu(entry->e_hash), &ea_inode);
553         if (err) {
554                 ea_inode = NULL;
555                 goto out;
556         }
557
558         if (i_size_read(ea_inode) != size) {
559                 ext4_warning_inode(ea_inode,
560                                    "ea_inode file size=%llu entry size=%zu",
561                                    i_size_read(ea_inode), size);
562                 err = -EFSCORRUPTED;
563                 goto out;
564         }
565
566         err = ext4_xattr_inode_read(ea_inode, buffer, size);
567         if (err)
568                 goto out;
569
570         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
571                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
572                                                      size);
573                 if (err) {
574                         ext4_warning_inode(ea_inode,
575                                            "EA inode hash validation failed");
576                         goto out;
577                 }
578
579                 if (ea_inode_cache)
580                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
581                                         ext4_xattr_inode_get_hash(ea_inode),
582                                         ea_inode->i_ino, true /* reusable */);
583         }
584 out:
585         iput(ea_inode);
586         return err;
587 }
588
589 static int
590 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
591                      void *buffer, size_t buffer_size)
592 {
593         struct buffer_head *bh = NULL;
594         struct ext4_xattr_entry *entry;
595         size_t size;
596         void *end;
597         int error;
598         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
599
600         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
601                   name_index, name, buffer, (long)buffer_size);
602
603         if (!EXT4_I(inode)->i_file_acl)
604                 return -ENODATA;
605         ea_idebug(inode, "reading block %llu",
606                   (unsigned long long)EXT4_I(inode)->i_file_acl);
607         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
608         if (IS_ERR(bh))
609                 return PTR_ERR(bh);
610         ea_bdebug(bh, "b_count=%d, refcount=%d",
611                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
612         error = ext4_xattr_check_block(inode, bh);
613         if (error)
614                 goto cleanup;
615         ext4_xattr_block_cache_insert(ea_block_cache, bh);
616         entry = BFIRST(bh);
617         end = bh->b_data + bh->b_size;
618         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
619         if (error)
620                 goto cleanup;
621         size = le32_to_cpu(entry->e_value_size);
622         error = -ERANGE;
623         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
624                 goto cleanup;
625         if (buffer) {
626                 if (size > buffer_size)
627                         goto cleanup;
628                 if (entry->e_value_inum) {
629                         error = ext4_xattr_inode_get(inode, entry, buffer,
630                                                      size);
631                         if (error)
632                                 goto cleanup;
633                 } else {
634                         u16 offset = le16_to_cpu(entry->e_value_offs);
635                         void *p = bh->b_data + offset;
636
637                         if (unlikely(p + size > end))
638                                 goto cleanup;
639                         memcpy(buffer, p, size);
640                 }
641         }
642         error = size;
643
644 cleanup:
645         brelse(bh);
646         return error;
647 }
648
649 int
650 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
651                      void *buffer, size_t buffer_size)
652 {
653         struct ext4_xattr_ibody_header *header;
654         struct ext4_xattr_entry *entry;
655         struct ext4_inode *raw_inode;
656         struct ext4_iloc iloc;
657         size_t size;
658         void *end;
659         int error;
660
661         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
662                 return -ENODATA;
663         error = ext4_get_inode_loc(inode, &iloc);
664         if (error)
665                 return error;
666         raw_inode = ext4_raw_inode(&iloc);
667         header = IHDR(inode, raw_inode);
668         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
669         error = xattr_check_inode(inode, header, end);
670         if (error)
671                 goto cleanup;
672         entry = IFIRST(header);
673         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
674         if (error)
675                 goto cleanup;
676         size = le32_to_cpu(entry->e_value_size);
677         error = -ERANGE;
678         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
679                 goto cleanup;
680         if (buffer) {
681                 if (size > buffer_size)
682                         goto cleanup;
683                 if (entry->e_value_inum) {
684                         error = ext4_xattr_inode_get(inode, entry, buffer,
685                                                      size);
686                         if (error)
687                                 goto cleanup;
688                 } else {
689                         u16 offset = le16_to_cpu(entry->e_value_offs);
690                         void *p = (void *)IFIRST(header) + offset;
691
692                         if (unlikely(p + size > end))
693                                 goto cleanup;
694                         memcpy(buffer, p, size);
695                 }
696         }
697         error = size;
698
699 cleanup:
700         brelse(iloc.bh);
701         return error;
702 }
703
704 /*
705  * ext4_xattr_get()
706  *
707  * Copy an extended attribute into the buffer
708  * provided, or compute the buffer size required.
709  * Buffer is NULL to compute the size of the buffer required.
710  *
711  * Returns a negative error number on failure, or the number of bytes
712  * used / required on success.
713  */
714 int
715 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
716                void *buffer, size_t buffer_size)
717 {
718         int error;
719
720         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
721                 return -EIO;
722
723         if (strlen(name) > 255)
724                 return -ERANGE;
725
726         down_read(&EXT4_I(inode)->xattr_sem);
727         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
728                                      buffer_size);
729         if (error == -ENODATA)
730                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
731                                              buffer_size);
732         up_read(&EXT4_I(inode)->xattr_sem);
733         return error;
734 }
735
736 static int
737 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
738                         char *buffer, size_t buffer_size)
739 {
740         size_t rest = buffer_size;
741
742         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
743                 const char *prefix;
744
745                 prefix = ext4_xattr_prefix(entry->e_name_index, dentry);
746                 if (prefix) {
747                         size_t prefix_len = strlen(prefix);
748                         size_t size = prefix_len + entry->e_name_len + 1;
749
750                         if (buffer) {
751                                 if (size > rest)
752                                         return -ERANGE;
753                                 memcpy(buffer, prefix, prefix_len);
754                                 buffer += prefix_len;
755                                 memcpy(buffer, entry->e_name, entry->e_name_len);
756                                 buffer += entry->e_name_len;
757                                 *buffer++ = 0;
758                         }
759                         rest -= size;
760                 }
761         }
762         return buffer_size - rest;  /* total size */
763 }
764
765 static int
766 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
767 {
768         struct inode *inode = d_inode(dentry);
769         struct buffer_head *bh = NULL;
770         int error;
771
772         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
773                   buffer, (long)buffer_size);
774
775         if (!EXT4_I(inode)->i_file_acl)
776                 return 0;
777         ea_idebug(inode, "reading block %llu",
778                   (unsigned long long)EXT4_I(inode)->i_file_acl);
779         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
780         if (IS_ERR(bh))
781                 return PTR_ERR(bh);
782         ea_bdebug(bh, "b_count=%d, refcount=%d",
783                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
784         error = ext4_xattr_check_block(inode, bh);
785         if (error)
786                 goto cleanup;
787         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
788         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
789                                         buffer_size);
790 cleanup:
791         brelse(bh);
792         return error;
793 }
794
795 static int
796 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
797 {
798         struct inode *inode = d_inode(dentry);
799         struct ext4_xattr_ibody_header *header;
800         struct ext4_inode *raw_inode;
801         struct ext4_iloc iloc;
802         void *end;
803         int error;
804
805         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
806                 return 0;
807         error = ext4_get_inode_loc(inode, &iloc);
808         if (error)
809                 return error;
810         raw_inode = ext4_raw_inode(&iloc);
811         header = IHDR(inode, raw_inode);
812         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
813         error = xattr_check_inode(inode, header, end);
814         if (error)
815                 goto cleanup;
816         error = ext4_xattr_list_entries(dentry, IFIRST(header),
817                                         buffer, buffer_size);
818
819 cleanup:
820         brelse(iloc.bh);
821         return error;
822 }
823
824 /*
825  * Inode operation listxattr()
826  *
827  * d_inode(dentry)->i_rwsem: don't care
828  *
829  * Copy a list of attribute names into the buffer
830  * provided, or compute the buffer size required.
831  * Buffer is NULL to compute the size of the buffer required.
832  *
833  * Returns a negative error number on failure, or the number of bytes
834  * used / required on success.
835  */
836 ssize_t
837 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
838 {
839         int ret, ret2;
840
841         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
842         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
843         if (ret < 0)
844                 goto errout;
845         if (buffer) {
846                 buffer += ret;
847                 buffer_size -= ret;
848         }
849         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
850         if (ret < 0)
851                 goto errout;
852         ret += ret2;
853 errout:
854         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
855         return ret;
856 }
857
858 /*
859  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
860  * not set, set it.
861  */
862 static void ext4_xattr_update_super_block(handle_t *handle,
863                                           struct super_block *sb)
864 {
865         if (ext4_has_feature_xattr(sb))
866                 return;
867
868         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
869         if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
870                                           EXT4_JTR_NONE) == 0) {
871                 lock_buffer(EXT4_SB(sb)->s_sbh);
872                 ext4_set_feature_xattr(sb);
873                 ext4_superblock_csum_set(sb);
874                 unlock_buffer(EXT4_SB(sb)->s_sbh);
875                 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
876         }
877 }
878
879 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
880 {
881         struct ext4_iloc iloc = { .bh = NULL };
882         struct buffer_head *bh = NULL;
883         struct ext4_inode *raw_inode;
884         struct ext4_xattr_ibody_header *header;
885         struct ext4_xattr_entry *entry;
886         qsize_t ea_inode_refs = 0;
887         void *end;
888         int ret;
889
890         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
891
892         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
893                 ret = ext4_get_inode_loc(inode, &iloc);
894                 if (ret)
895                         goto out;
896                 raw_inode = ext4_raw_inode(&iloc);
897                 header = IHDR(inode, raw_inode);
898                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
899                 ret = xattr_check_inode(inode, header, end);
900                 if (ret)
901                         goto out;
902
903                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
904                      entry = EXT4_XATTR_NEXT(entry))
905                         if (entry->e_value_inum)
906                                 ea_inode_refs++;
907         }
908
909         if (EXT4_I(inode)->i_file_acl) {
910                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
911                 if (IS_ERR(bh)) {
912                         ret = PTR_ERR(bh);
913                         bh = NULL;
914                         goto out;
915                 }
916
917                 ret = ext4_xattr_check_block(inode, bh);
918                 if (ret)
919                         goto out;
920
921                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
922                      entry = EXT4_XATTR_NEXT(entry))
923                         if (entry->e_value_inum)
924                                 ea_inode_refs++;
925         }
926         *usage = ea_inode_refs + 1;
927         ret = 0;
928 out:
929         brelse(iloc.bh);
930         brelse(bh);
931         return ret;
932 }
933
934 static inline size_t round_up_cluster(struct inode *inode, size_t length)
935 {
936         struct super_block *sb = inode->i_sb;
937         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
938                                     inode->i_blkbits);
939         size_t mask = ~(cluster_size - 1);
940
941         return (length + cluster_size - 1) & mask;
942 }
943
944 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
945 {
946         int err;
947
948         err = dquot_alloc_inode(inode);
949         if (err)
950                 return err;
951         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
952         if (err)
953                 dquot_free_inode(inode);
954         return err;
955 }
956
957 static void ext4_xattr_inode_free_quota(struct inode *parent,
958                                         struct inode *ea_inode,
959                                         size_t len)
960 {
961         if (ea_inode &&
962             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
963                 return;
964         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
965         dquot_free_inode(parent);
966 }
967
968 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
969                              struct buffer_head *block_bh, size_t value_len,
970                              bool is_create)
971 {
972         int credits;
973         int blocks;
974
975         /*
976          * 1) Owner inode update
977          * 2) Ref count update on old xattr block
978          * 3) new xattr block
979          * 4) block bitmap update for new xattr block
980          * 5) group descriptor for new xattr block
981          * 6) block bitmap update for old xattr block
982          * 7) group descriptor for old block
983          *
984          * 6 & 7 can happen if we have two racing threads T_a and T_b
985          * which are each trying to set an xattr on inodes I_a and I_b
986          * which were both initially sharing an xattr block.
987          */
988         credits = 7;
989
990         /* Quota updates. */
991         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
992
993         /*
994          * In case of inline data, we may push out the data to a block,
995          * so we need to reserve credits for this eventuality
996          */
997         if (inode && ext4_has_inline_data(inode))
998                 credits += ext4_writepage_trans_blocks(inode) + 1;
999
1000         /* We are done if ea_inode feature is not enabled. */
1001         if (!ext4_has_feature_ea_inode(sb))
1002                 return credits;
1003
1004         /* New ea_inode, inode map, block bitmap, group descriptor. */
1005         credits += 4;
1006
1007         /* Data blocks. */
1008         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1009
1010         /* Indirection block or one level of extent tree. */
1011         blocks += 1;
1012
1013         /* Block bitmap and group descriptor updates for each block. */
1014         credits += blocks * 2;
1015
1016         /* Blocks themselves. */
1017         credits += blocks;
1018
1019         if (!is_create) {
1020                 /* Dereference ea_inode holding old xattr value.
1021                  * Old ea_inode, inode map, block bitmap, group descriptor.
1022                  */
1023                 credits += 4;
1024
1025                 /* Data blocks for old ea_inode. */
1026                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
1027
1028                 /* Indirection block or one level of extent tree for old
1029                  * ea_inode.
1030                  */
1031                 blocks += 1;
1032
1033                 /* Block bitmap and group descriptor updates for each block. */
1034                 credits += blocks * 2;
1035         }
1036
1037         /* We may need to clone the existing xattr block in which case we need
1038          * to increment ref counts for existing ea_inodes referenced by it.
1039          */
1040         if (block_bh) {
1041                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
1042
1043                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1044                         if (entry->e_value_inum)
1045                                 /* Ref count update on ea_inode. */
1046                                 credits += 1;
1047         }
1048         return credits;
1049 }
1050
1051 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1052                                        int ref_change)
1053 {
1054         struct ext4_iloc iloc;
1055         s64 ref_count;
1056         int ret;
1057
1058         inode_lock(ea_inode);
1059
1060         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1061         if (ret)
1062                 goto out;
1063
1064         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1065         ref_count += ref_change;
1066         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1067
1068         if (ref_change > 0) {
1069                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1070                           ea_inode->i_ino, ref_count);
1071
1072                 if (ref_count == 1) {
1073                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1074                                   ea_inode->i_ino, ea_inode->i_nlink);
1075
1076                         set_nlink(ea_inode, 1);
1077                         ext4_orphan_del(handle, ea_inode);
1078                 }
1079         } else {
1080                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1081                           ea_inode->i_ino, ref_count);
1082
1083                 if (ref_count == 0) {
1084                         WARN_ONCE(ea_inode->i_nlink != 1,
1085                                   "EA inode %lu i_nlink=%u",
1086                                   ea_inode->i_ino, ea_inode->i_nlink);
1087
1088                         clear_nlink(ea_inode);
1089                         ext4_orphan_add(handle, ea_inode);
1090                 }
1091         }
1092
1093         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1094         if (ret)
1095                 ext4_warning_inode(ea_inode,
1096                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1097 out:
1098         inode_unlock(ea_inode);
1099         return ret;
1100 }
1101
1102 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1103 {
1104         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1105 }
1106
1107 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1108 {
1109         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1110 }
1111
1112 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1113                                         struct ext4_xattr_entry *first)
1114 {
1115         struct inode *ea_inode;
1116         struct ext4_xattr_entry *entry;
1117         struct ext4_xattr_entry *failed_entry;
1118         unsigned int ea_ino;
1119         int err, saved_err;
1120
1121         for (entry = first; !IS_LAST_ENTRY(entry);
1122              entry = EXT4_XATTR_NEXT(entry)) {
1123                 if (!entry->e_value_inum)
1124                         continue;
1125                 ea_ino = le32_to_cpu(entry->e_value_inum);
1126                 err = ext4_xattr_inode_iget(parent, ea_ino,
1127                                             le32_to_cpu(entry->e_hash),
1128                                             &ea_inode);
1129                 if (err)
1130                         goto cleanup;
1131                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1132                 if (err) {
1133                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1134                         iput(ea_inode);
1135                         goto cleanup;
1136                 }
1137                 iput(ea_inode);
1138         }
1139         return 0;
1140
1141 cleanup:
1142         saved_err = err;
1143         failed_entry = entry;
1144
1145         for (entry = first; entry != failed_entry;
1146              entry = EXT4_XATTR_NEXT(entry)) {
1147                 if (!entry->e_value_inum)
1148                         continue;
1149                 ea_ino = le32_to_cpu(entry->e_value_inum);
1150                 err = ext4_xattr_inode_iget(parent, ea_ino,
1151                                             le32_to_cpu(entry->e_hash),
1152                                             &ea_inode);
1153                 if (err) {
1154                         ext4_warning(parent->i_sb,
1155                                      "cleanup ea_ino %u iget error %d", ea_ino,
1156                                      err);
1157                         continue;
1158                 }
1159                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1160                 if (err)
1161                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1162                                            err);
1163                 iput(ea_inode);
1164         }
1165         return saved_err;
1166 }
1167
1168 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1169                         struct buffer_head *bh, bool block_csum, bool dirty)
1170 {
1171         int error;
1172
1173         if (bh && dirty) {
1174                 if (block_csum)
1175                         ext4_xattr_block_csum_set(inode, bh);
1176                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1177                 if (error) {
1178                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1179                                      error);
1180                         return error;
1181                 }
1182         }
1183         return 0;
1184 }
1185
1186 static void
1187 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1188                              struct buffer_head *bh,
1189                              struct ext4_xattr_entry *first, bool block_csum,
1190                              struct ext4_xattr_inode_array **ea_inode_array,
1191                              int extra_credits, bool skip_quota)
1192 {
1193         struct inode *ea_inode;
1194         struct ext4_xattr_entry *entry;
1195         bool dirty = false;
1196         unsigned int ea_ino;
1197         int err;
1198         int credits;
1199
1200         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1201         credits = 2 + extra_credits;
1202
1203         for (entry = first; !IS_LAST_ENTRY(entry);
1204              entry = EXT4_XATTR_NEXT(entry)) {
1205                 if (!entry->e_value_inum)
1206                         continue;
1207                 ea_ino = le32_to_cpu(entry->e_value_inum);
1208                 err = ext4_xattr_inode_iget(parent, ea_ino,
1209                                             le32_to_cpu(entry->e_hash),
1210                                             &ea_inode);
1211                 if (err)
1212                         continue;
1213
1214                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1215                 if (err) {
1216                         ext4_warning_inode(ea_inode,
1217                                            "Expand inode array err=%d", err);
1218                         iput(ea_inode);
1219                         continue;
1220                 }
1221
1222                 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1223                         ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1224                         ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1225                                               dirty));
1226                 if (err < 0) {
1227                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1228                                            err);
1229                         continue;
1230                 }
1231                 if (err > 0) {
1232                         err = ext4_journal_get_write_access(handle,
1233                                         parent->i_sb, bh, EXT4_JTR_NONE);
1234                         if (err) {
1235                                 ext4_warning_inode(ea_inode,
1236                                                 "Re-get write access err=%d",
1237                                                 err);
1238                                 continue;
1239                         }
1240                 }
1241
1242                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1243                 if (err) {
1244                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1245                                            err);
1246                         continue;
1247                 }
1248
1249                 if (!skip_quota)
1250                         ext4_xattr_inode_free_quota(parent, ea_inode,
1251                                               le32_to_cpu(entry->e_value_size));
1252
1253                 /*
1254                  * Forget about ea_inode within the same transaction that
1255                  * decrements the ref count. This avoids duplicate decrements in
1256                  * case the rest of the work spills over to subsequent
1257                  * transactions.
1258                  */
1259                 entry->e_value_inum = 0;
1260                 entry->e_value_size = 0;
1261
1262                 dirty = true;
1263         }
1264
1265         if (dirty) {
1266                 /*
1267                  * Note that we are deliberately skipping csum calculation for
1268                  * the final update because we do not expect any journal
1269                  * restarts until xattr block is freed.
1270                  */
1271
1272                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1273                 if (err)
1274                         ext4_warning_inode(parent,
1275                                            "handle dirty metadata err=%d", err);
1276         }
1277 }
1278
1279 /*
1280  * Release the xattr block BH: If the reference count is > 1, decrement it;
1281  * otherwise free the block.
1282  */
1283 static void
1284 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1285                          struct buffer_head *bh,
1286                          struct ext4_xattr_inode_array **ea_inode_array,
1287                          int extra_credits)
1288 {
1289         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1290         u32 hash, ref;
1291         int error = 0;
1292
1293         BUFFER_TRACE(bh, "get_write_access");
1294         error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1295                                               EXT4_JTR_NONE);
1296         if (error)
1297                 goto out;
1298
1299 retry_ref:
1300         lock_buffer(bh);
1301         hash = le32_to_cpu(BHDR(bh)->h_hash);
1302         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1303         if (ref == 1) {
1304                 ea_bdebug(bh, "refcount now=0; freeing");
1305                 /*
1306                  * This must happen under buffer lock for
1307                  * ext4_xattr_block_set() to reliably detect freed block
1308                  */
1309                 if (ea_block_cache) {
1310                         struct mb_cache_entry *oe;
1311
1312                         oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1313                                                           bh->b_blocknr);
1314                         if (oe) {
1315                                 unlock_buffer(bh);
1316                                 mb_cache_entry_wait_unused(oe);
1317                                 mb_cache_entry_put(ea_block_cache, oe);
1318                                 goto retry_ref;
1319                         }
1320                 }
1321                 get_bh(bh);
1322                 unlock_buffer(bh);
1323
1324                 if (ext4_has_feature_ea_inode(inode->i_sb))
1325                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1326                                                      BFIRST(bh),
1327                                                      true /* block_csum */,
1328                                                      ea_inode_array,
1329                                                      extra_credits,
1330                                                      true /* skip_quota */);
1331                 ext4_free_blocks(handle, inode, bh, 0, 1,
1332                                  EXT4_FREE_BLOCKS_METADATA |
1333                                  EXT4_FREE_BLOCKS_FORGET);
1334         } else {
1335                 ref--;
1336                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1337                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1338                         struct mb_cache_entry *ce;
1339
1340                         if (ea_block_cache) {
1341                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1342                                                         bh->b_blocknr);
1343                                 if (ce) {
1344                                         set_bit(MBE_REUSABLE_B, &ce->e_flags);
1345                                         mb_cache_entry_put(ea_block_cache, ce);
1346                                 }
1347                         }
1348                 }
1349
1350                 ext4_xattr_block_csum_set(inode, bh);
1351                 /*
1352                  * Beware of this ugliness: Releasing of xattr block references
1353                  * from different inodes can race and so we have to protect
1354                  * from a race where someone else frees the block (and releases
1355                  * its journal_head) before we are done dirtying the buffer. In
1356                  * nojournal mode this race is harmless and we actually cannot
1357                  * call ext4_handle_dirty_metadata() with locked buffer as
1358                  * that function can call sync_dirty_buffer() so for that case
1359                  * we handle the dirtying after unlocking the buffer.
1360                  */
1361                 if (ext4_handle_valid(handle))
1362                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1363                 unlock_buffer(bh);
1364                 if (!ext4_handle_valid(handle))
1365                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1366                 if (IS_SYNC(inode))
1367                         ext4_handle_sync(handle);
1368                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1369                 ea_bdebug(bh, "refcount now=%d; releasing",
1370                           le32_to_cpu(BHDR(bh)->h_refcount));
1371         }
1372 out:
1373         ext4_std_error(inode->i_sb, error);
1374         return;
1375 }
1376
1377 /*
1378  * Find the available free space for EAs. This also returns the total number of
1379  * bytes used by EA entries.
1380  */
1381 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1382                                     size_t *min_offs, void *base, int *total)
1383 {
1384         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1385                 if (!last->e_value_inum && last->e_value_size) {
1386                         size_t offs = le16_to_cpu(last->e_value_offs);
1387                         if (offs < *min_offs)
1388                                 *min_offs = offs;
1389                 }
1390                 if (total)
1391                         *total += EXT4_XATTR_LEN(last->e_name_len);
1392         }
1393         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1394 }
1395
1396 /*
1397  * Write the value of the EA in an inode.
1398  */
1399 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1400                                   const void *buf, int bufsize)
1401 {
1402         struct buffer_head *bh = NULL;
1403         unsigned long block = 0;
1404         int blocksize = ea_inode->i_sb->s_blocksize;
1405         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1406         int csize, wsize = 0;
1407         int ret = 0, ret2 = 0;
1408         int retries = 0;
1409
1410 retry:
1411         while (ret >= 0 && ret < max_blocks) {
1412                 struct ext4_map_blocks map;
1413                 map.m_lblk = block += ret;
1414                 map.m_len = max_blocks -= ret;
1415
1416                 ret = ext4_map_blocks(handle, ea_inode, &map,
1417                                       EXT4_GET_BLOCKS_CREATE);
1418                 if (ret <= 0) {
1419                         ext4_mark_inode_dirty(handle, ea_inode);
1420                         if (ret == -ENOSPC &&
1421                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1422                                 ret = 0;
1423                                 goto retry;
1424                         }
1425                         break;
1426                 }
1427         }
1428
1429         if (ret < 0)
1430                 return ret;
1431
1432         block = 0;
1433         while (wsize < bufsize) {
1434                 brelse(bh);
1435                 csize = (bufsize - wsize) > blocksize ? blocksize :
1436                                                                 bufsize - wsize;
1437                 bh = ext4_getblk(handle, ea_inode, block, 0);
1438                 if (IS_ERR(bh))
1439                         return PTR_ERR(bh);
1440                 if (!bh) {
1441                         WARN_ON_ONCE(1);
1442                         EXT4_ERROR_INODE(ea_inode,
1443                                          "ext4_getblk() return bh = NULL");
1444                         return -EFSCORRUPTED;
1445                 }
1446                 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1447                                                    EXT4_JTR_NONE);
1448                 if (ret)
1449                         goto out;
1450
1451                 memcpy(bh->b_data, buf, csize);
1452                 set_buffer_uptodate(bh);
1453                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1454
1455                 buf += csize;
1456                 wsize += csize;
1457                 block += 1;
1458         }
1459
1460         inode_lock(ea_inode);
1461         i_size_write(ea_inode, wsize);
1462         ext4_update_i_disksize(ea_inode, wsize);
1463         inode_unlock(ea_inode);
1464
1465         ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1466         if (unlikely(ret2 && !ret))
1467                 ret = ret2;
1468
1469 out:
1470         brelse(bh);
1471
1472         return ret;
1473 }
1474
1475 /*
1476  * Create an inode to store the value of a large EA.
1477  */
1478 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1479                                              struct inode *inode, u32 hash)
1480 {
1481         struct inode *ea_inode = NULL;
1482         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1483         int err;
1484
1485         if (inode->i_sb->s_root == NULL) {
1486                 ext4_warning(inode->i_sb,
1487                              "refuse to create EA inode when umounting");
1488                 WARN_ON(1);
1489                 return ERR_PTR(-EINVAL);
1490         }
1491
1492         /*
1493          * Let the next inode be the goal, so we try and allocate the EA inode
1494          * in the same group, or nearby one.
1495          */
1496         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1497                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1498                                   EXT4_EA_INODE_FL);
1499         if (!IS_ERR(ea_inode)) {
1500                 ea_inode->i_op = &ext4_file_inode_operations;
1501                 ea_inode->i_fop = &ext4_file_operations;
1502                 ext4_set_aops(ea_inode);
1503                 ext4_xattr_inode_set_class(ea_inode);
1504                 unlock_new_inode(ea_inode);
1505                 ext4_xattr_inode_set_ref(ea_inode, 1);
1506                 ext4_xattr_inode_set_hash(ea_inode, hash);
1507                 err = ext4_mark_inode_dirty(handle, ea_inode);
1508                 if (!err)
1509                         err = ext4_inode_attach_jinode(ea_inode);
1510                 if (err) {
1511                         if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1512                                 ext4_warning_inode(ea_inode,
1513                                         "cleanup dec ref error %d", err);
1514                         iput(ea_inode);
1515                         return ERR_PTR(err);
1516                 }
1517
1518                 /*
1519                  * Xattr inodes are shared therefore quota charging is performed
1520                  * at a higher level.
1521                  */
1522                 dquot_free_inode(ea_inode);
1523                 dquot_drop(ea_inode);
1524                 inode_lock(ea_inode);
1525                 ea_inode->i_flags |= S_NOQUOTA;
1526                 inode_unlock(ea_inode);
1527         }
1528
1529         return ea_inode;
1530 }
1531
1532 static struct inode *
1533 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1534                             size_t value_len, u32 hash)
1535 {
1536         struct inode *ea_inode;
1537         struct mb_cache_entry *ce;
1538         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1539         void *ea_data;
1540
1541         if (!ea_inode_cache)
1542                 return NULL;
1543
1544         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1545         if (!ce)
1546                 return NULL;
1547
1548         WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1549                      !(current->flags & PF_MEMALLOC_NOFS));
1550
1551         ea_data = kvmalloc(value_len, GFP_KERNEL);
1552         if (!ea_data) {
1553                 mb_cache_entry_put(ea_inode_cache, ce);
1554                 return NULL;
1555         }
1556
1557         while (ce) {
1558                 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1559                                      EXT4_IGET_NORMAL);
1560                 if (!IS_ERR(ea_inode) &&
1561                     !is_bad_inode(ea_inode) &&
1562                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1563                     i_size_read(ea_inode) == value_len &&
1564                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1565                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1566                                                     value_len) &&
1567                     !memcmp(value, ea_data, value_len)) {
1568                         mb_cache_entry_touch(ea_inode_cache, ce);
1569                         mb_cache_entry_put(ea_inode_cache, ce);
1570                         kvfree(ea_data);
1571                         return ea_inode;
1572                 }
1573
1574                 if (!IS_ERR(ea_inode))
1575                         iput(ea_inode);
1576                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1577         }
1578         kvfree(ea_data);
1579         return NULL;
1580 }
1581
1582 /*
1583  * Add value of the EA in an inode.
1584  */
1585 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1586                                           const void *value, size_t value_len,
1587                                           struct inode **ret_inode)
1588 {
1589         struct inode *ea_inode;
1590         u32 hash;
1591         int err;
1592
1593         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1594         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1595         if (ea_inode) {
1596                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1597                 if (err) {
1598                         iput(ea_inode);
1599                         return err;
1600                 }
1601
1602                 *ret_inode = ea_inode;
1603                 return 0;
1604         }
1605
1606         /* Create an inode for the EA value */
1607         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1608         if (IS_ERR(ea_inode))
1609                 return PTR_ERR(ea_inode);
1610
1611         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1612         if (err) {
1613                 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1614                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1615                 iput(ea_inode);
1616                 return err;
1617         }
1618
1619         if (EA_INODE_CACHE(inode))
1620                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1621                                       ea_inode->i_ino, true /* reusable */);
1622
1623         *ret_inode = ea_inode;
1624         return 0;
1625 }
1626
1627 /*
1628  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1629  * feature is enabled.
1630  */
1631 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1632
1633 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1634                                 struct ext4_xattr_search *s,
1635                                 handle_t *handle, struct inode *inode,
1636                                 bool is_block)
1637 {
1638         struct ext4_xattr_entry *last, *next;
1639         struct ext4_xattr_entry *here = s->here;
1640         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1641         int in_inode = i->in_inode;
1642         struct inode *old_ea_inode = NULL;
1643         struct inode *new_ea_inode = NULL;
1644         size_t old_size, new_size;
1645         int ret;
1646
1647         /* Space used by old and new values. */
1648         old_size = (!s->not_found && !here->e_value_inum) ?
1649                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1650         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1651
1652         /*
1653          * Optimization for the simple case when old and new values have the
1654          * same padded sizes. Not applicable if external inodes are involved.
1655          */
1656         if (new_size && new_size == old_size) {
1657                 size_t offs = le16_to_cpu(here->e_value_offs);
1658                 void *val = s->base + offs;
1659
1660                 here->e_value_size = cpu_to_le32(i->value_len);
1661                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1662                         memset(val, 0, new_size);
1663                 } else {
1664                         memcpy(val, i->value, i->value_len);
1665                         /* Clear padding bytes. */
1666                         memset(val + i->value_len, 0, new_size - i->value_len);
1667                 }
1668                 goto update_hash;
1669         }
1670
1671         /* Compute min_offs and last. */
1672         last = s->first;
1673         for (; !IS_LAST_ENTRY(last); last = next) {
1674                 next = EXT4_XATTR_NEXT(last);
1675                 if ((void *)next >= s->end) {
1676                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1677                         ret = -EFSCORRUPTED;
1678                         goto out;
1679                 }
1680                 if (!last->e_value_inum && last->e_value_size) {
1681                         size_t offs = le16_to_cpu(last->e_value_offs);
1682                         if (offs < min_offs)
1683                                 min_offs = offs;
1684                 }
1685         }
1686
1687         /* Check whether we have enough space. */
1688         if (i->value) {
1689                 size_t free;
1690
1691                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1692                 if (!s->not_found)
1693                         free += EXT4_XATTR_LEN(name_len) + old_size;
1694
1695                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1696                         ret = -ENOSPC;
1697                         goto out;
1698                 }
1699
1700                 /*
1701                  * If storing the value in an external inode is an option,
1702                  * reserve space for xattr entries/names in the external
1703                  * attribute block so that a long value does not occupy the
1704                  * whole space and prevent further entries being added.
1705                  */
1706                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1707                     new_size && is_block &&
1708                     (min_offs + old_size - new_size) <
1709                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1710                         ret = -ENOSPC;
1711                         goto out;
1712                 }
1713         }
1714
1715         /*
1716          * Getting access to old and new ea inodes is subject to failures.
1717          * Finish that work before doing any modifications to the xattr data.
1718          */
1719         if (!s->not_found && here->e_value_inum) {
1720                 ret = ext4_xattr_inode_iget(inode,
1721                                             le32_to_cpu(here->e_value_inum),
1722                                             le32_to_cpu(here->e_hash),
1723                                             &old_ea_inode);
1724                 if (ret) {
1725                         old_ea_inode = NULL;
1726                         goto out;
1727                 }
1728         }
1729         if (i->value && in_inode) {
1730                 WARN_ON_ONCE(!i->value_len);
1731
1732                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1733                 if (ret)
1734                         goto out;
1735
1736                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1737                                                      i->value_len,
1738                                                      &new_ea_inode);
1739                 if (ret) {
1740                         new_ea_inode = NULL;
1741                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1742                         goto out;
1743                 }
1744         }
1745
1746         if (old_ea_inode) {
1747                 /* We are ready to release ref count on the old_ea_inode. */
1748                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1749                 if (ret) {
1750                         /* Release newly required ref count on new_ea_inode. */
1751                         if (new_ea_inode) {
1752                                 int err;
1753
1754                                 err = ext4_xattr_inode_dec_ref(handle,
1755                                                                new_ea_inode);
1756                                 if (err)
1757                                         ext4_warning_inode(new_ea_inode,
1758                                                   "dec ref new_ea_inode err=%d",
1759                                                   err);
1760                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1761                                                             i->value_len);
1762                         }
1763                         goto out;
1764                 }
1765
1766                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1767                                             le32_to_cpu(here->e_value_size));
1768         }
1769
1770         /* No failures allowed past this point. */
1771
1772         if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1773                 /* Remove the old value. */
1774                 void *first_val = s->base + min_offs;
1775                 size_t offs = le16_to_cpu(here->e_value_offs);
1776                 void *val = s->base + offs;
1777
1778                 memmove(first_val + old_size, first_val, val - first_val);
1779                 memset(first_val, 0, old_size);
1780                 min_offs += old_size;
1781
1782                 /* Adjust all value offsets. */
1783                 last = s->first;
1784                 while (!IS_LAST_ENTRY(last)) {
1785                         size_t o = le16_to_cpu(last->e_value_offs);
1786
1787                         if (!last->e_value_inum &&
1788                             last->e_value_size && o < offs)
1789                                 last->e_value_offs = cpu_to_le16(o + old_size);
1790                         last = EXT4_XATTR_NEXT(last);
1791                 }
1792         }
1793
1794         if (!i->value) {
1795                 /* Remove old name. */
1796                 size_t size = EXT4_XATTR_LEN(name_len);
1797
1798                 last = ENTRY((void *)last - size);
1799                 memmove(here, (void *)here + size,
1800                         (void *)last - (void *)here + sizeof(__u32));
1801                 memset(last, 0, size);
1802         } else if (s->not_found) {
1803                 /* Insert new name. */
1804                 size_t size = EXT4_XATTR_LEN(name_len);
1805                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1806
1807                 memmove((void *)here + size, here, rest);
1808                 memset(here, 0, size);
1809                 here->e_name_index = i->name_index;
1810                 here->e_name_len = name_len;
1811                 memcpy(here->e_name, i->name, name_len);
1812         } else {
1813                 /* This is an update, reset value info. */
1814                 here->e_value_inum = 0;
1815                 here->e_value_offs = 0;
1816                 here->e_value_size = 0;
1817         }
1818
1819         if (i->value) {
1820                 /* Insert new value. */
1821                 if (in_inode) {
1822                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1823                 } else if (i->value_len) {
1824                         void *val = s->base + min_offs - new_size;
1825
1826                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1827                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1828                                 memset(val, 0, new_size);
1829                         } else {
1830                                 memcpy(val, i->value, i->value_len);
1831                                 /* Clear padding bytes. */
1832                                 memset(val + i->value_len, 0,
1833                                        new_size - i->value_len);
1834                         }
1835                 }
1836                 here->e_value_size = cpu_to_le32(i->value_len);
1837         }
1838
1839 update_hash:
1840         if (i->value) {
1841                 __le32 hash = 0;
1842
1843                 /* Entry hash calculation. */
1844                 if (in_inode) {
1845                         __le32 crc32c_hash;
1846
1847                         /*
1848                          * Feed crc32c hash instead of the raw value for entry
1849                          * hash calculation. This is to avoid walking
1850                          * potentially long value buffer again.
1851                          */
1852                         crc32c_hash = cpu_to_le32(
1853                                        ext4_xattr_inode_get_hash(new_ea_inode));
1854                         hash = ext4_xattr_hash_entry(here->e_name,
1855                                                      here->e_name_len,
1856                                                      &crc32c_hash, 1);
1857                 } else if (is_block) {
1858                         __le32 *value = s->base + le16_to_cpu(
1859                                                         here->e_value_offs);
1860
1861                         hash = ext4_xattr_hash_entry(here->e_name,
1862                                                      here->e_name_len, value,
1863                                                      new_size >> 2);
1864                 }
1865                 here->e_hash = hash;
1866         }
1867
1868         if (is_block)
1869                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1870
1871         ret = 0;
1872 out:
1873         iput(old_ea_inode);
1874         iput(new_ea_inode);
1875         return ret;
1876 }
1877
1878 struct ext4_xattr_block_find {
1879         struct ext4_xattr_search s;
1880         struct buffer_head *bh;
1881 };
1882
1883 static int
1884 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1885                       struct ext4_xattr_block_find *bs)
1886 {
1887         struct super_block *sb = inode->i_sb;
1888         int error;
1889
1890         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1891                   i->name_index, i->name, i->value, (long)i->value_len);
1892
1893         if (EXT4_I(inode)->i_file_acl) {
1894                 /* The inode already has an extended attribute block. */
1895                 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1896                 if (IS_ERR(bs->bh)) {
1897                         error = PTR_ERR(bs->bh);
1898                         bs->bh = NULL;
1899                         return error;
1900                 }
1901                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1902                         atomic_read(&(bs->bh->b_count)),
1903                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1904                 error = ext4_xattr_check_block(inode, bs->bh);
1905                 if (error)
1906                         return error;
1907                 /* Find the named attribute. */
1908                 bs->s.base = BHDR(bs->bh);
1909                 bs->s.first = BFIRST(bs->bh);
1910                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1911                 bs->s.here = bs->s.first;
1912                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1913                                          i->name_index, i->name, 1);
1914                 if (error && error != -ENODATA)
1915                         return error;
1916                 bs->s.not_found = error;
1917         }
1918         return 0;
1919 }
1920
1921 static int
1922 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1923                      struct ext4_xattr_info *i,
1924                      struct ext4_xattr_block_find *bs)
1925 {
1926         struct super_block *sb = inode->i_sb;
1927         struct buffer_head *new_bh = NULL;
1928         struct ext4_xattr_search s_copy = bs->s;
1929         struct ext4_xattr_search *s = &s_copy;
1930         struct mb_cache_entry *ce = NULL;
1931         int error = 0;
1932         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1933         struct inode *ea_inode = NULL, *tmp_inode;
1934         size_t old_ea_inode_quota = 0;
1935         unsigned int ea_ino;
1936
1937
1938 #define header(x) ((struct ext4_xattr_header *)(x))
1939
1940         if (s->base) {
1941                 int offset = (char *)s->here - bs->bh->b_data;
1942
1943                 BUFFER_TRACE(bs->bh, "get_write_access");
1944                 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1945                                                       EXT4_JTR_NONE);
1946                 if (error)
1947                         goto cleanup;
1948                 lock_buffer(bs->bh);
1949
1950                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1951                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1952
1953                         /*
1954                          * This must happen under buffer lock for
1955                          * ext4_xattr_block_set() to reliably detect modified
1956                          * block
1957                          */
1958                         if (ea_block_cache) {
1959                                 struct mb_cache_entry *oe;
1960
1961                                 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1962                                         hash, bs->bh->b_blocknr);
1963                                 if (oe) {
1964                                         /*
1965                                          * Xattr block is getting reused. Leave
1966                                          * it alone.
1967                                          */
1968                                         mb_cache_entry_put(ea_block_cache, oe);
1969                                         goto clone_block;
1970                                 }
1971                         }
1972                         ea_bdebug(bs->bh, "modifying in-place");
1973                         error = ext4_xattr_set_entry(i, s, handle, inode,
1974                                                      true /* is_block */);
1975                         ext4_xattr_block_csum_set(inode, bs->bh);
1976                         unlock_buffer(bs->bh);
1977                         if (error == -EFSCORRUPTED)
1978                                 goto bad_block;
1979                         if (!error)
1980                                 error = ext4_handle_dirty_metadata(handle,
1981                                                                    inode,
1982                                                                    bs->bh);
1983                         if (error)
1984                                 goto cleanup;
1985                         goto inserted;
1986                 }
1987 clone_block:
1988                 unlock_buffer(bs->bh);
1989                 ea_bdebug(bs->bh, "cloning");
1990                 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1991                 error = -ENOMEM;
1992                 if (s->base == NULL)
1993                         goto cleanup;
1994                 s->first = ENTRY(header(s->base)+1);
1995                 header(s->base)->h_refcount = cpu_to_le32(1);
1996                 s->here = ENTRY(s->base + offset);
1997                 s->end = s->base + bs->bh->b_size;
1998
1999                 /*
2000                  * If existing entry points to an xattr inode, we need
2001                  * to prevent ext4_xattr_set_entry() from decrementing
2002                  * ref count on it because the reference belongs to the
2003                  * original block. In this case, make the entry look
2004                  * like it has an empty value.
2005                  */
2006                 if (!s->not_found && s->here->e_value_inum) {
2007                         ea_ino = le32_to_cpu(s->here->e_value_inum);
2008                         error = ext4_xattr_inode_iget(inode, ea_ino,
2009                                       le32_to_cpu(s->here->e_hash),
2010                                       &tmp_inode);
2011                         if (error)
2012                                 goto cleanup;
2013
2014                         if (!ext4_test_inode_state(tmp_inode,
2015                                         EXT4_STATE_LUSTRE_EA_INODE)) {
2016                                 /*
2017                                  * Defer quota free call for previous
2018                                  * inode until success is guaranteed.
2019                                  */
2020                                 old_ea_inode_quota = le32_to_cpu(
2021                                                 s->here->e_value_size);
2022                         }
2023                         iput(tmp_inode);
2024
2025                         s->here->e_value_inum = 0;
2026                         s->here->e_value_size = 0;
2027                 }
2028         } else {
2029                 /* Allocate a buffer where we construct the new block. */
2030                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
2031                 error = -ENOMEM;
2032                 if (s->base == NULL)
2033                         goto cleanup;
2034                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2035                 header(s->base)->h_blocks = cpu_to_le32(1);
2036                 header(s->base)->h_refcount = cpu_to_le32(1);
2037                 s->first = ENTRY(header(s->base)+1);
2038                 s->here = ENTRY(header(s->base)+1);
2039                 s->end = s->base + sb->s_blocksize;
2040         }
2041
2042         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
2043         if (error == -EFSCORRUPTED)
2044                 goto bad_block;
2045         if (error)
2046                 goto cleanup;
2047
2048         if (i->value && s->here->e_value_inum) {
2049                 /*
2050                  * A ref count on ea_inode has been taken as part of the call to
2051                  * ext4_xattr_set_entry() above. We would like to drop this
2052                  * extra ref but we have to wait until the xattr block is
2053                  * initialized and has its own ref count on the ea_inode.
2054                  */
2055                 ea_ino = le32_to_cpu(s->here->e_value_inum);
2056                 error = ext4_xattr_inode_iget(inode, ea_ino,
2057                                               le32_to_cpu(s->here->e_hash),
2058                                               &ea_inode);
2059                 if (error) {
2060                         ea_inode = NULL;
2061                         goto cleanup;
2062                 }
2063         }
2064
2065 inserted:
2066         if (!IS_LAST_ENTRY(s->first)) {
2067                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2068                                                      &ce);
2069                 if (new_bh) {
2070                         /* We found an identical block in the cache. */
2071                         if (new_bh == bs->bh)
2072                                 ea_bdebug(new_bh, "keeping");
2073                         else {
2074                                 u32 ref;
2075
2076                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
2077
2078                                 /* The old block is released after updating
2079                                    the inode. */
2080                                 error = dquot_alloc_block(inode,
2081                                                 EXT4_C2B(EXT4_SB(sb), 1));
2082                                 if (error)
2083                                         goto cleanup;
2084                                 BUFFER_TRACE(new_bh, "get_write_access");
2085                                 error = ext4_journal_get_write_access(
2086                                                 handle, sb, new_bh,
2087                                                 EXT4_JTR_NONE);
2088                                 if (error)
2089                                         goto cleanup_dquot;
2090                                 lock_buffer(new_bh);
2091                                 /*
2092                                  * We have to be careful about races with
2093                                  * adding references to xattr block. Once we
2094                                  * hold buffer lock xattr block's state is
2095                                  * stable so we can check the additional
2096                                  * reference fits.
2097                                  */
2098                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2099                                 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2100                                         /*
2101                                          * Undo everything and check mbcache
2102                                          * again.
2103                                          */
2104                                         unlock_buffer(new_bh);
2105                                         dquot_free_block(inode,
2106                                                          EXT4_C2B(EXT4_SB(sb),
2107                                                                   1));
2108                                         brelse(new_bh);
2109                                         mb_cache_entry_put(ea_block_cache, ce);
2110                                         ce = NULL;
2111                                         new_bh = NULL;
2112                                         goto inserted;
2113                                 }
2114                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2115                                 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2116                                         clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2117                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2118                                           ref);
2119                                 ext4_xattr_block_csum_set(inode, new_bh);
2120                                 unlock_buffer(new_bh);
2121                                 error = ext4_handle_dirty_metadata(handle,
2122                                                                    inode,
2123                                                                    new_bh);
2124                                 if (error)
2125                                         goto cleanup_dquot;
2126                         }
2127                         mb_cache_entry_touch(ea_block_cache, ce);
2128                         mb_cache_entry_put(ea_block_cache, ce);
2129                         ce = NULL;
2130                 } else if (bs->bh && s->base == bs->bh->b_data) {
2131                         /* We were modifying this block in-place. */
2132                         ea_bdebug(bs->bh, "keeping this block");
2133                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2134                         new_bh = bs->bh;
2135                         get_bh(new_bh);
2136                 } else {
2137                         /* We need to allocate a new block */
2138                         ext4_fsblk_t goal, block;
2139
2140                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2141
2142                         goal = ext4_group_first_block_no(sb,
2143                                                 EXT4_I(inode)->i_block_group);
2144                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2145                                                      NULL, &error);
2146                         if (error)
2147                                 goto cleanup;
2148
2149                         ea_idebug(inode, "creating block %llu",
2150                                   (unsigned long long)block);
2151
2152                         new_bh = sb_getblk(sb, block);
2153                         if (unlikely(!new_bh)) {
2154                                 error = -ENOMEM;
2155 getblk_failed:
2156                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2157                                                  EXT4_FREE_BLOCKS_METADATA);
2158                                 goto cleanup;
2159                         }
2160                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2161                                                       ENTRY(header(s->base)+1));
2162                         if (error)
2163                                 goto getblk_failed;
2164                         if (ea_inode) {
2165                                 /* Drop the extra ref on ea_inode. */
2166                                 error = ext4_xattr_inode_dec_ref(handle,
2167                                                                  ea_inode);
2168                                 if (error)
2169                                         ext4_warning_inode(ea_inode,
2170                                                            "dec ref error=%d",
2171                                                            error);
2172                                 iput(ea_inode);
2173                                 ea_inode = NULL;
2174                         }
2175
2176                         lock_buffer(new_bh);
2177                         error = ext4_journal_get_create_access(handle, sb,
2178                                                         new_bh, EXT4_JTR_NONE);
2179                         if (error) {
2180                                 unlock_buffer(new_bh);
2181                                 error = -EIO;
2182                                 goto getblk_failed;
2183                         }
2184                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2185                         ext4_xattr_block_csum_set(inode, new_bh);
2186                         set_buffer_uptodate(new_bh);
2187                         unlock_buffer(new_bh);
2188                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2189                         error = ext4_handle_dirty_metadata(handle, inode,
2190                                                            new_bh);
2191                         if (error)
2192                                 goto cleanup;
2193                 }
2194         }
2195
2196         if (old_ea_inode_quota)
2197                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2198
2199         /* Update the inode. */
2200         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2201
2202         /* Drop the previous xattr block. */
2203         if (bs->bh && bs->bh != new_bh) {
2204                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2205
2206                 ext4_xattr_release_block(handle, inode, bs->bh,
2207                                          &ea_inode_array,
2208                                          0 /* extra_credits */);
2209                 ext4_xattr_inode_array_free(ea_inode_array);
2210         }
2211         error = 0;
2212
2213 cleanup:
2214         if (ea_inode) {
2215                 int error2;
2216
2217                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2218                 if (error2)
2219                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2220                                            error2);
2221
2222                 /* If there was an error, revert the quota charge. */
2223                 if (error)
2224                         ext4_xattr_inode_free_quota(inode, ea_inode,
2225                                                     i_size_read(ea_inode));
2226                 iput(ea_inode);
2227         }
2228         if (ce)
2229                 mb_cache_entry_put(ea_block_cache, ce);
2230         brelse(new_bh);
2231         if (!(bs->bh && s->base == bs->bh->b_data))
2232                 kfree(s->base);
2233
2234         return error;
2235
2236 cleanup_dquot:
2237         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2238         goto cleanup;
2239
2240 bad_block:
2241         EXT4_ERROR_INODE(inode, "bad block %llu",
2242                          EXT4_I(inode)->i_file_acl);
2243         goto cleanup;
2244
2245 #undef header
2246 }
2247
2248 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2249                           struct ext4_xattr_ibody_find *is)
2250 {
2251         struct ext4_xattr_ibody_header *header;
2252         struct ext4_inode *raw_inode;
2253         int error;
2254
2255         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2256                 return 0;
2257
2258         raw_inode = ext4_raw_inode(&is->iloc);
2259         header = IHDR(inode, raw_inode);
2260         is->s.base = is->s.first = IFIRST(header);
2261         is->s.here = is->s.first;
2262         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2263         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2264                 error = xattr_check_inode(inode, header, is->s.end);
2265                 if (error)
2266                         return error;
2267                 /* Find the named attribute. */
2268                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2269                                          i->name_index, i->name, 0);
2270                 if (error && error != -ENODATA)
2271                         return error;
2272                 is->s.not_found = error;
2273         }
2274         return 0;
2275 }
2276
2277 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2278                                 struct ext4_xattr_info *i,
2279                                 struct ext4_xattr_ibody_find *is)
2280 {
2281         struct ext4_xattr_ibody_header *header;
2282         struct ext4_xattr_search *s = &is->s;
2283         int error;
2284
2285         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2286                 return -ENOSPC;
2287
2288         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2289         if (error)
2290                 return error;
2291         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2292         if (!IS_LAST_ENTRY(s->first)) {
2293                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2294                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2295         } else {
2296                 header->h_magic = cpu_to_le32(0);
2297                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2298         }
2299         return 0;
2300 }
2301
2302 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2303                                  struct ext4_xattr_info *i)
2304 {
2305         void *value;
2306
2307         /* When e_value_inum is set the value is stored externally. */
2308         if (s->here->e_value_inum)
2309                 return 0;
2310         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2311                 return 0;
2312         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2313         return !memcmp(value, i->value, i->value_len);
2314 }
2315
2316 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2317 {
2318         struct buffer_head *bh;
2319         int error;
2320
2321         if (!EXT4_I(inode)->i_file_acl)
2322                 return NULL;
2323         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2324         if (IS_ERR(bh))
2325                 return bh;
2326         error = ext4_xattr_check_block(inode, bh);
2327         if (error) {
2328                 brelse(bh);
2329                 return ERR_PTR(error);
2330         }
2331         return bh;
2332 }
2333
2334 /*
2335  * ext4_xattr_set_handle()
2336  *
2337  * Create, replace or remove an extended attribute for this inode.  Value
2338  * is NULL to remove an existing extended attribute, and non-NULL to
2339  * either replace an existing extended attribute, or create a new extended
2340  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2341  * specify that an extended attribute must exist and must not exist
2342  * previous to the call, respectively.
2343  *
2344  * Returns 0, or a negative error number on failure.
2345  */
2346 int
2347 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2348                       const char *name, const void *value, size_t value_len,
2349                       int flags)
2350 {
2351         struct ext4_xattr_info i = {
2352                 .name_index = name_index,
2353                 .name = name,
2354                 .value = value,
2355                 .value_len = value_len,
2356                 .in_inode = 0,
2357         };
2358         struct ext4_xattr_ibody_find is = {
2359                 .s = { .not_found = -ENODATA, },
2360         };
2361         struct ext4_xattr_block_find bs = {
2362                 .s = { .not_found = -ENODATA, },
2363         };
2364         int no_expand;
2365         int error;
2366
2367         if (!name)
2368                 return -EINVAL;
2369         if (strlen(name) > 255)
2370                 return -ERANGE;
2371
2372         ext4_write_lock_xattr(inode, &no_expand);
2373
2374         /* Check journal credits under write lock. */
2375         if (ext4_handle_valid(handle)) {
2376                 struct buffer_head *bh;
2377                 int credits;
2378
2379                 bh = ext4_xattr_get_block(inode);
2380                 if (IS_ERR(bh)) {
2381                         error = PTR_ERR(bh);
2382                         goto cleanup;
2383                 }
2384
2385                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2386                                                    value_len,
2387                                                    flags & XATTR_CREATE);
2388                 brelse(bh);
2389
2390                 if (jbd2_handle_buffer_credits(handle) < credits) {
2391                         error = -ENOSPC;
2392                         goto cleanup;
2393                 }
2394                 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2395         }
2396
2397         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2398         if (error)
2399                 goto cleanup;
2400
2401         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2402                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2403                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2404                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2405         }
2406
2407         error = ext4_xattr_ibody_find(inode, &i, &is);
2408         if (error)
2409                 goto cleanup;
2410         if (is.s.not_found)
2411                 error = ext4_xattr_block_find(inode, &i, &bs);
2412         if (error)
2413                 goto cleanup;
2414         if (is.s.not_found && bs.s.not_found) {
2415                 error = -ENODATA;
2416                 if (flags & XATTR_REPLACE)
2417                         goto cleanup;
2418                 error = 0;
2419                 if (!value)
2420                         goto cleanup;
2421         } else {
2422                 error = -EEXIST;
2423                 if (flags & XATTR_CREATE)
2424                         goto cleanup;
2425         }
2426
2427         if (!value) {
2428                 if (!is.s.not_found)
2429                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2430                 else if (!bs.s.not_found)
2431                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2432         } else {
2433                 error = 0;
2434                 /* Xattr value did not change? Save us some work and bail out */
2435                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2436                         goto cleanup;
2437                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2438                         goto cleanup;
2439
2440                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2441                     (EXT4_XATTR_SIZE(i.value_len) >
2442                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2443                         i.in_inode = 1;
2444 retry_inode:
2445                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2446                 if (!error && !bs.s.not_found) {
2447                         i.value = NULL;
2448                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2449                 } else if (error == -ENOSPC) {
2450                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2451                                 brelse(bs.bh);
2452                                 bs.bh = NULL;
2453                                 error = ext4_xattr_block_find(inode, &i, &bs);
2454                                 if (error)
2455                                         goto cleanup;
2456                         }
2457                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2458                         if (!error && !is.s.not_found) {
2459                                 i.value = NULL;
2460                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2461                                                              &is);
2462                         } else if (error == -ENOSPC) {
2463                                 /*
2464                                  * Xattr does not fit in the block, store at
2465                                  * external inode if possible.
2466                                  */
2467                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2468                                     i.value_len && !i.in_inode) {
2469                                         i.in_inode = 1;
2470                                         goto retry_inode;
2471                                 }
2472                         }
2473                 }
2474         }
2475         if (!error) {
2476                 ext4_xattr_update_super_block(handle, inode->i_sb);
2477                 inode->i_ctime = current_time(inode);
2478                 inode_inc_iversion(inode);
2479                 if (!value)
2480                         no_expand = 0;
2481                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2482                 /*
2483                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2484                  * error != 0.
2485                  */
2486                 is.iloc.bh = NULL;
2487                 if (IS_SYNC(inode))
2488                         ext4_handle_sync(handle);
2489         }
2490         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2491
2492 cleanup:
2493         brelse(is.iloc.bh);
2494         brelse(bs.bh);
2495         ext4_write_unlock_xattr(inode, &no_expand);
2496         return error;
2497 }
2498
2499 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2500                            bool is_create, int *credits)
2501 {
2502         struct buffer_head *bh;
2503         int err;
2504
2505         *credits = 0;
2506
2507         if (!EXT4_SB(inode->i_sb)->s_journal)
2508                 return 0;
2509
2510         down_read(&EXT4_I(inode)->xattr_sem);
2511
2512         bh = ext4_xattr_get_block(inode);
2513         if (IS_ERR(bh)) {
2514                 err = PTR_ERR(bh);
2515         } else {
2516                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2517                                                     value_len, is_create);
2518                 brelse(bh);
2519                 err = 0;
2520         }
2521
2522         up_read(&EXT4_I(inode)->xattr_sem);
2523         return err;
2524 }
2525
2526 /*
2527  * ext4_xattr_set()
2528  *
2529  * Like ext4_xattr_set_handle, but start from an inode. This extended
2530  * attribute modification is a filesystem transaction by itself.
2531  *
2532  * Returns 0, or a negative error number on failure.
2533  */
2534 int
2535 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2536                const void *value, size_t value_len, int flags)
2537 {
2538         handle_t *handle;
2539         struct super_block *sb = inode->i_sb;
2540         int error, retries = 0;
2541         int credits;
2542
2543         error = dquot_initialize(inode);
2544         if (error)
2545                 return error;
2546
2547 retry:
2548         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2549                                        &credits);
2550         if (error)
2551                 return error;
2552
2553         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2554         if (IS_ERR(handle)) {
2555                 error = PTR_ERR(handle);
2556         } else {
2557                 int error2;
2558
2559                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2560                                               value, value_len, flags);
2561                 error2 = ext4_journal_stop(handle);
2562                 if (error == -ENOSPC &&
2563                     ext4_should_retry_alloc(sb, &retries))
2564                         goto retry;
2565                 if (error == 0)
2566                         error = error2;
2567         }
2568         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2569
2570         return error;
2571 }
2572
2573 /*
2574  * Shift the EA entries in the inode to create space for the increased
2575  * i_extra_isize.
2576  */
2577 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2578                                      int value_offs_shift, void *to,
2579                                      void *from, size_t n)
2580 {
2581         struct ext4_xattr_entry *last = entry;
2582         int new_offs;
2583
2584         /* We always shift xattr headers further thus offsets get lower */
2585         BUG_ON(value_offs_shift > 0);
2586
2587         /* Adjust the value offsets of the entries */
2588         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2589                 if (!last->e_value_inum && last->e_value_size) {
2590                         new_offs = le16_to_cpu(last->e_value_offs) +
2591                                                         value_offs_shift;
2592                         last->e_value_offs = cpu_to_le16(new_offs);
2593                 }
2594         }
2595         /* Shift the entries by n bytes */
2596         memmove(to, from, n);
2597 }
2598
2599 /*
2600  * Move xattr pointed to by 'entry' from inode into external xattr block
2601  */
2602 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2603                                     struct ext4_inode *raw_inode,
2604                                     struct ext4_xattr_entry *entry)
2605 {
2606         struct ext4_xattr_ibody_find *is = NULL;
2607         struct ext4_xattr_block_find *bs = NULL;
2608         char *buffer = NULL, *b_entry_name = NULL;
2609         size_t value_size = le32_to_cpu(entry->e_value_size);
2610         struct ext4_xattr_info i = {
2611                 .value = NULL,
2612                 .value_len = 0,
2613                 .name_index = entry->e_name_index,
2614                 .in_inode = !!entry->e_value_inum,
2615         };
2616         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2617         int needs_kvfree = 0;
2618         int error;
2619
2620         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2621         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2622         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2623         if (!is || !bs || !b_entry_name) {
2624                 error = -ENOMEM;
2625                 goto out;
2626         }
2627
2628         is->s.not_found = -ENODATA;
2629         bs->s.not_found = -ENODATA;
2630         is->iloc.bh = NULL;
2631         bs->bh = NULL;
2632
2633         /* Save the entry name and the entry value */
2634         if (entry->e_value_inum) {
2635                 buffer = kvmalloc(value_size, GFP_NOFS);
2636                 if (!buffer) {
2637                         error = -ENOMEM;
2638                         goto out;
2639                 }
2640                 needs_kvfree = 1;
2641                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2642                 if (error)
2643                         goto out;
2644         } else {
2645                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2646                 buffer = (void *)IFIRST(header) + value_offs;
2647         }
2648
2649         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2650         b_entry_name[entry->e_name_len] = '\0';
2651         i.name = b_entry_name;
2652
2653         error = ext4_get_inode_loc(inode, &is->iloc);
2654         if (error)
2655                 goto out;
2656
2657         error = ext4_xattr_ibody_find(inode, &i, is);
2658         if (error)
2659                 goto out;
2660
2661         i.value = buffer;
2662         i.value_len = value_size;
2663         error = ext4_xattr_block_find(inode, &i, bs);
2664         if (error)
2665                 goto out;
2666
2667         /* Move ea entry from the inode into the block */
2668         error = ext4_xattr_block_set(handle, inode, &i, bs);
2669         if (error)
2670                 goto out;
2671
2672         /* Remove the chosen entry from the inode */
2673         i.value = NULL;
2674         i.value_len = 0;
2675         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2676
2677 out:
2678         kfree(b_entry_name);
2679         if (needs_kvfree && buffer)
2680                 kvfree(buffer);
2681         if (is)
2682                 brelse(is->iloc.bh);
2683         if (bs)
2684                 brelse(bs->bh);
2685         kfree(is);
2686         kfree(bs);
2687
2688         return error;
2689 }
2690
2691 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2692                                        struct ext4_inode *raw_inode,
2693                                        int isize_diff, size_t ifree,
2694                                        size_t bfree, int *total_ino)
2695 {
2696         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2697         struct ext4_xattr_entry *small_entry;
2698         struct ext4_xattr_entry *entry;
2699         struct ext4_xattr_entry *last;
2700         unsigned int entry_size;        /* EA entry size */
2701         unsigned int total_size;        /* EA entry size + value size */
2702         unsigned int min_total_size;
2703         int error;
2704
2705         while (isize_diff > ifree) {
2706                 entry = NULL;
2707                 small_entry = NULL;
2708                 min_total_size = ~0U;
2709                 last = IFIRST(header);
2710                 /* Find the entry best suited to be pushed into EA block */
2711                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2712                         /* never move system.data out of the inode */
2713                         if ((last->e_name_len == 4) &&
2714                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2715                             !memcmp(last->e_name, "data", 4))
2716                                 continue;
2717                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2718                         if (!last->e_value_inum)
2719                                 total_size += EXT4_XATTR_SIZE(
2720                                                le32_to_cpu(last->e_value_size));
2721                         if (total_size <= bfree &&
2722                             total_size < min_total_size) {
2723                                 if (total_size + ifree < isize_diff) {
2724                                         small_entry = last;
2725                                 } else {
2726                                         entry = last;
2727                                         min_total_size = total_size;
2728                                 }
2729                         }
2730                 }
2731
2732                 if (entry == NULL) {
2733                         if (small_entry == NULL)
2734                                 return -ENOSPC;
2735                         entry = small_entry;
2736                 }
2737
2738                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2739                 total_size = entry_size;
2740                 if (!entry->e_value_inum)
2741                         total_size += EXT4_XATTR_SIZE(
2742                                               le32_to_cpu(entry->e_value_size));
2743                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2744                                                  entry);
2745                 if (error)
2746                         return error;
2747
2748                 *total_ino -= entry_size;
2749                 ifree += total_size;
2750                 bfree -= total_size;
2751         }
2752
2753         return 0;
2754 }
2755
2756 /*
2757  * Expand an inode by new_extra_isize bytes when EAs are present.
2758  * Returns 0 on success or negative error number on failure.
2759  */
2760 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2761                                struct ext4_inode *raw_inode, handle_t *handle)
2762 {
2763         struct ext4_xattr_ibody_header *header;
2764         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2765         static unsigned int mnt_count;
2766         size_t min_offs;
2767         size_t ifree, bfree;
2768         int total_ino;
2769         void *base, *end;
2770         int error = 0, tried_min_extra_isize = 0;
2771         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2772         int isize_diff; /* How much do we need to grow i_extra_isize */
2773
2774 retry:
2775         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2776         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2777                 return 0;
2778
2779         header = IHDR(inode, raw_inode);
2780
2781         /*
2782          * Check if enough free space is available in the inode to shift the
2783          * entries ahead by new_extra_isize.
2784          */
2785
2786         base = IFIRST(header);
2787         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2788         min_offs = end - base;
2789         total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2790
2791         error = xattr_check_inode(inode, header, end);
2792         if (error)
2793                 goto cleanup;
2794
2795         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2796         if (ifree >= isize_diff)
2797                 goto shift;
2798
2799         /*
2800          * Enough free space isn't available in the inode, check if
2801          * EA block can hold new_extra_isize bytes.
2802          */
2803         if (EXT4_I(inode)->i_file_acl) {
2804                 struct buffer_head *bh;
2805
2806                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2807                 if (IS_ERR(bh)) {
2808                         error = PTR_ERR(bh);
2809                         goto cleanup;
2810                 }
2811                 error = ext4_xattr_check_block(inode, bh);
2812                 if (error) {
2813                         brelse(bh);
2814                         goto cleanup;
2815                 }
2816                 base = BHDR(bh);
2817                 end = bh->b_data + bh->b_size;
2818                 min_offs = end - base;
2819                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2820                                               NULL);
2821                 brelse(bh);
2822                 if (bfree + ifree < isize_diff) {
2823                         if (!tried_min_extra_isize && s_min_extra_isize) {
2824                                 tried_min_extra_isize++;
2825                                 new_extra_isize = s_min_extra_isize;
2826                                 goto retry;
2827                         }
2828                         error = -ENOSPC;
2829                         goto cleanup;
2830                 }
2831         } else {
2832                 bfree = inode->i_sb->s_blocksize;
2833         }
2834
2835         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2836                                             isize_diff, ifree, bfree,
2837                                             &total_ino);
2838         if (error) {
2839                 if (error == -ENOSPC && !tried_min_extra_isize &&
2840                     s_min_extra_isize) {
2841                         tried_min_extra_isize++;
2842                         new_extra_isize = s_min_extra_isize;
2843                         goto retry;
2844                 }
2845                 goto cleanup;
2846         }
2847 shift:
2848         /* Adjust the offsets and shift the remaining entries ahead */
2849         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2850                         - new_extra_isize, (void *)raw_inode +
2851                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2852                         (void *)header, total_ino);
2853         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2854
2855         if (ext4_has_inline_data(inode))
2856                 error = ext4_find_inline_data_nolock(inode);
2857
2858 cleanup:
2859         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2860                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2861                              inode->i_ino);
2862                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2863         }
2864         return error;
2865 }
2866
2867 #define EIA_INCR 16 /* must be 2^n */
2868 #define EIA_MASK (EIA_INCR - 1)
2869
2870 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2871  * If @ea_inode_array is new or full it will be grown and the old
2872  * contents copied over.
2873  */
2874 static int
2875 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2876                         struct inode *inode)
2877 {
2878         if (*ea_inode_array == NULL) {
2879                 /*
2880                  * Start with 15 inodes, so it fits into a power-of-two size.
2881                  * If *ea_inode_array is NULL, this is essentially offsetof()
2882                  */
2883                 (*ea_inode_array) =
2884                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2885                                          inodes[EIA_MASK]),
2886                                 GFP_NOFS);
2887                 if (*ea_inode_array == NULL)
2888                         return -ENOMEM;
2889                 (*ea_inode_array)->count = 0;
2890         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2891                 /* expand the array once all 15 + n * 16 slots are full */
2892                 struct ext4_xattr_inode_array *new_array = NULL;
2893                 int count = (*ea_inode_array)->count;
2894
2895                 /* if new_array is NULL, this is essentially offsetof() */
2896                 new_array = kmalloc(
2897                                 offsetof(struct ext4_xattr_inode_array,
2898                                          inodes[count + EIA_INCR]),
2899                                 GFP_NOFS);
2900                 if (new_array == NULL)
2901                         return -ENOMEM;
2902                 memcpy(new_array, *ea_inode_array,
2903                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2904                 kfree(*ea_inode_array);
2905                 *ea_inode_array = new_array;
2906         }
2907         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2908         return 0;
2909 }
2910
2911 /*
2912  * ext4_xattr_delete_inode()
2913  *
2914  * Free extended attribute resources associated with this inode. Traverse
2915  * all entries and decrement reference on any xattr inodes associated with this
2916  * inode. This is called immediately before an inode is freed. We have exclusive
2917  * access to the inode. If an orphan inode is deleted it will also release its
2918  * references on xattr block and xattr inodes.
2919  */
2920 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2921                             struct ext4_xattr_inode_array **ea_inode_array,
2922                             int extra_credits)
2923 {
2924         struct buffer_head *bh = NULL;
2925         struct ext4_xattr_ibody_header *header;
2926         struct ext4_iloc iloc = { .bh = NULL };
2927         struct ext4_xattr_entry *entry;
2928         struct inode *ea_inode;
2929         int error;
2930
2931         error = ext4_journal_ensure_credits(handle, extra_credits,
2932                         ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2933         if (error < 0) {
2934                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2935                 goto cleanup;
2936         }
2937
2938         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2939             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2940
2941                 error = ext4_get_inode_loc(inode, &iloc);
2942                 if (error) {
2943                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2944                         goto cleanup;
2945                 }
2946
2947                 error = ext4_journal_get_write_access(handle, inode->i_sb,
2948                                                 iloc.bh, EXT4_JTR_NONE);
2949                 if (error) {
2950                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2951                                          error);
2952                         goto cleanup;
2953                 }
2954
2955                 header = IHDR(inode, ext4_raw_inode(&iloc));
2956                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2957                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2958                                                      IFIRST(header),
2959                                                      false /* block_csum */,
2960                                                      ea_inode_array,
2961                                                      extra_credits,
2962                                                      false /* skip_quota */);
2963         }
2964
2965         if (EXT4_I(inode)->i_file_acl) {
2966                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2967                 if (IS_ERR(bh)) {
2968                         error = PTR_ERR(bh);
2969                         if (error == -EIO) {
2970                                 EXT4_ERROR_INODE_ERR(inode, EIO,
2971                                                      "block %llu read error",
2972                                                      EXT4_I(inode)->i_file_acl);
2973                         }
2974                         bh = NULL;
2975                         goto cleanup;
2976                 }
2977                 error = ext4_xattr_check_block(inode, bh);
2978                 if (error)
2979                         goto cleanup;
2980
2981                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2982                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2983                              entry = EXT4_XATTR_NEXT(entry)) {
2984                                 if (!entry->e_value_inum)
2985                                         continue;
2986                                 error = ext4_xattr_inode_iget(inode,
2987                                               le32_to_cpu(entry->e_value_inum),
2988                                               le32_to_cpu(entry->e_hash),
2989                                               &ea_inode);
2990                                 if (error)
2991                                         continue;
2992                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2993                                               le32_to_cpu(entry->e_value_size));
2994                                 iput(ea_inode);
2995                         }
2996
2997                 }
2998
2999                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
3000                                          extra_credits);
3001                 /*
3002                  * Update i_file_acl value in the same transaction that releases
3003                  * block.
3004                  */
3005                 EXT4_I(inode)->i_file_acl = 0;
3006                 error = ext4_mark_inode_dirty(handle, inode);
3007                 if (error) {
3008                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
3009                                          error);
3010                         goto cleanup;
3011                 }
3012                 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
3013         }
3014         error = 0;
3015 cleanup:
3016         brelse(iloc.bh);
3017         brelse(bh);
3018         return error;
3019 }
3020
3021 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
3022 {
3023         int idx;
3024
3025         if (ea_inode_array == NULL)
3026                 return;
3027
3028         for (idx = 0; idx < ea_inode_array->count; ++idx)
3029                 iput(ea_inode_array->inodes[idx]);
3030         kfree(ea_inode_array);
3031 }
3032
3033 /*
3034  * ext4_xattr_block_cache_insert()
3035  *
3036  * Create a new entry in the extended attribute block cache, and insert
3037  * it unless such an entry is already in the cache.
3038  *
3039  * Returns 0, or a negative error number on failure.
3040  */
3041 static void
3042 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3043                               struct buffer_head *bh)
3044 {
3045         struct ext4_xattr_header *header = BHDR(bh);
3046         __u32 hash = le32_to_cpu(header->h_hash);
3047         int reusable = le32_to_cpu(header->h_refcount) <
3048                        EXT4_XATTR_REFCOUNT_MAX;
3049         int error;
3050
3051         if (!ea_block_cache)
3052                 return;
3053         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3054                                       bh->b_blocknr, reusable);
3055         if (error) {
3056                 if (error == -EBUSY)
3057                         ea_bdebug(bh, "already in cache");
3058         } else
3059                 ea_bdebug(bh, "inserting [%x]", (int)hash);
3060 }
3061
3062 /*
3063  * ext4_xattr_cmp()
3064  *
3065  * Compare two extended attribute blocks for equality.
3066  *
3067  * Returns 0 if the blocks are equal, 1 if they differ, and
3068  * a negative error number on errors.
3069  */
3070 static int
3071 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3072                struct ext4_xattr_header *header2)
3073 {
3074         struct ext4_xattr_entry *entry1, *entry2;
3075
3076         entry1 = ENTRY(header1+1);
3077         entry2 = ENTRY(header2+1);
3078         while (!IS_LAST_ENTRY(entry1)) {
3079                 if (IS_LAST_ENTRY(entry2))
3080                         return 1;
3081                 if (entry1->e_hash != entry2->e_hash ||
3082                     entry1->e_name_index != entry2->e_name_index ||
3083                     entry1->e_name_len != entry2->e_name_len ||
3084                     entry1->e_value_size != entry2->e_value_size ||
3085                     entry1->e_value_inum != entry2->e_value_inum ||
3086                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3087                         return 1;
3088                 if (!entry1->e_value_inum &&
3089                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3090                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3091                            le32_to_cpu(entry1->e_value_size)))
3092                         return 1;
3093
3094                 entry1 = EXT4_XATTR_NEXT(entry1);
3095                 entry2 = EXT4_XATTR_NEXT(entry2);
3096         }
3097         if (!IS_LAST_ENTRY(entry2))
3098                 return 1;
3099         return 0;
3100 }
3101
3102 /*
3103  * ext4_xattr_block_cache_find()
3104  *
3105  * Find an identical extended attribute block.
3106  *
3107  * Returns a pointer to the block found, or NULL if such a block was
3108  * not found or an error occurred.
3109  */
3110 static struct buffer_head *
3111 ext4_xattr_block_cache_find(struct inode *inode,
3112                             struct ext4_xattr_header *header,
3113                             struct mb_cache_entry **pce)
3114 {
3115         __u32 hash = le32_to_cpu(header->h_hash);
3116         struct mb_cache_entry *ce;
3117         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3118
3119         if (!ea_block_cache)
3120                 return NULL;
3121         if (!header->h_hash)
3122                 return NULL;  /* never share */
3123         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3124         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3125         while (ce) {
3126                 struct buffer_head *bh;
3127
3128                 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3129                 if (IS_ERR(bh)) {
3130                         if (PTR_ERR(bh) == -ENOMEM)
3131                                 return NULL;
3132                         bh = NULL;
3133                         EXT4_ERROR_INODE(inode, "block %lu read error",
3134                                          (unsigned long)ce->e_value);
3135                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3136                         *pce = ce;
3137                         return bh;
3138                 }
3139                 brelse(bh);
3140                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3141         }
3142         return NULL;
3143 }
3144
3145 #define NAME_HASH_SHIFT 5
3146 #define VALUE_HASH_SHIFT 16
3147
3148 /*
3149  * ext4_xattr_hash_entry()
3150  *
3151  * Compute the hash of an extended attribute.
3152  */
3153 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3154                                     size_t value_count)
3155 {
3156         __u32 hash = 0;
3157
3158         while (name_len--) {
3159                 hash = (hash << NAME_HASH_SHIFT) ^
3160                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3161                        (unsigned char)*name++;
3162         }
3163         while (value_count--) {
3164                 hash = (hash << VALUE_HASH_SHIFT) ^
3165                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3166                        le32_to_cpu(*value++);
3167         }
3168         return cpu_to_le32(hash);
3169 }
3170
3171 /*
3172  * ext4_xattr_hash_entry_signed()
3173  *
3174  * Compute the hash of an extended attribute incorrectly.
3175  */
3176 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3177 {
3178         __u32 hash = 0;
3179
3180         while (name_len--) {
3181                 hash = (hash << NAME_HASH_SHIFT) ^
3182                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3183                        (signed char)*name++;
3184         }
3185         while (value_count--) {
3186                 hash = (hash << VALUE_HASH_SHIFT) ^
3187                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3188                        le32_to_cpu(*value++);
3189         }
3190         return cpu_to_le32(hash);
3191 }
3192
3193 #undef NAME_HASH_SHIFT
3194 #undef VALUE_HASH_SHIFT
3195
3196 #define BLOCK_HASH_SHIFT 16
3197
3198 /*
3199  * ext4_xattr_rehash()
3200  *
3201  * Re-compute the extended attribute hash value after an entry has changed.
3202  */
3203 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3204 {
3205         struct ext4_xattr_entry *here;
3206         __u32 hash = 0;
3207
3208         here = ENTRY(header+1);
3209         while (!IS_LAST_ENTRY(here)) {
3210                 if (!here->e_hash) {
3211                         /* Block is not shared if an entry's hash value == 0 */
3212                         hash = 0;
3213                         break;
3214                 }
3215                 hash = (hash << BLOCK_HASH_SHIFT) ^
3216                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3217                        le32_to_cpu(here->e_hash);
3218                 here = EXT4_XATTR_NEXT(here);
3219         }
3220         header->h_hash = cpu_to_le32(hash);
3221 }
3222
3223 #undef BLOCK_HASH_SHIFT
3224
3225 #define HASH_BUCKET_BITS        10
3226
3227 struct mb_cache *
3228 ext4_xattr_create_cache(void)
3229 {
3230         return mb_cache_create(HASH_BUCKET_BITS);
3231 }
3232
3233 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3234 {
3235         if (cache)
3236                 mb_cache_destroy(cache);
3237 }
3238