Merge tag 'rust-fixes-6.3-rc1' of https://github.com/Rust-for-Linux/linux
[sfrench/cifs-2.6.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]  = &posix_acl_access_xattr_handler,
92         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
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_POSIX_ACL
105         &posix_acl_access_xattr_handler,
106         &posix_acl_default_xattr_handler,
107 #endif
108 #ifdef CONFIG_EXT4_FS_SECURITY
109         &ext4_xattr_security_handler,
110 #endif
111         &ext4_xattr_hurd_handler,
112         NULL
113 };
114
115 #define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
116                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
117
118 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
119                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
120
121 static int
122 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
123                         struct inode *inode);
124
125 #ifdef CONFIG_LOCKDEP
126 void ext4_xattr_inode_set_class(struct inode *ea_inode)
127 {
128         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
129 }
130 #endif
131
132 static __le32 ext4_xattr_block_csum(struct inode *inode,
133                                     sector_t block_nr,
134                                     struct ext4_xattr_header *hdr)
135 {
136         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
137         __u32 csum;
138         __le64 dsk_block_nr = cpu_to_le64(block_nr);
139         __u32 dummy_csum = 0;
140         int offset = offsetof(struct ext4_xattr_header, h_checksum);
141
142         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
143                            sizeof(dsk_block_nr));
144         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
145         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
146         offset += sizeof(dummy_csum);
147         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
148                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
149
150         return cpu_to_le32(csum);
151 }
152
153 static int ext4_xattr_block_csum_verify(struct inode *inode,
154                                         struct buffer_head *bh)
155 {
156         struct ext4_xattr_header *hdr = BHDR(bh);
157         int ret = 1;
158
159         if (ext4_has_metadata_csum(inode->i_sb)) {
160                 lock_buffer(bh);
161                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
162                                                         bh->b_blocknr, hdr));
163                 unlock_buffer(bh);
164         }
165         return ret;
166 }
167
168 static void ext4_xattr_block_csum_set(struct inode *inode,
169                                       struct buffer_head *bh)
170 {
171         if (ext4_has_metadata_csum(inode->i_sb))
172                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
173                                                 bh->b_blocknr, BHDR(bh));
174 }
175
176 static inline const struct xattr_handler *
177 ext4_xattr_handler(int name_index)
178 {
179         const struct xattr_handler *handler = NULL;
180
181         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
182                 handler = ext4_xattr_handler_map[name_index];
183         return 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 struct xattr_handler *handler =
744                         ext4_xattr_handler(entry->e_name_index);
745
746                 if (handler && (!handler->list || handler->list(dentry))) {
747                         const char *prefix = handler->prefix ?: handler->name;
748                         size_t prefix_len = strlen(prefix);
749                         size_t size = prefix_len + entry->e_name_len + 1;
750
751                         if (buffer) {
752                                 if (size > rest)
753                                         return -ERANGE;
754                                 memcpy(buffer, prefix, prefix_len);
755                                 buffer += prefix_len;
756                                 memcpy(buffer, entry->e_name, entry->e_name_len);
757                                 buffer += entry->e_name_len;
758                                 *buffer++ = 0;
759                         }
760                         rest -= size;
761                 }
762         }
763         return buffer_size - rest;  /* total size */
764 }
765
766 static int
767 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
768 {
769         struct inode *inode = d_inode(dentry);
770         struct buffer_head *bh = NULL;
771         int error;
772
773         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
774                   buffer, (long)buffer_size);
775
776         if (!EXT4_I(inode)->i_file_acl)
777                 return 0;
778         ea_idebug(inode, "reading block %llu",
779                   (unsigned long long)EXT4_I(inode)->i_file_acl);
780         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
781         if (IS_ERR(bh))
782                 return PTR_ERR(bh);
783         ea_bdebug(bh, "b_count=%d, refcount=%d",
784                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
785         error = ext4_xattr_check_block(inode, bh);
786         if (error)
787                 goto cleanup;
788         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
789         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
790                                         buffer_size);
791 cleanup:
792         brelse(bh);
793         return error;
794 }
795
796 static int
797 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
798 {
799         struct inode *inode = d_inode(dentry);
800         struct ext4_xattr_ibody_header *header;
801         struct ext4_inode *raw_inode;
802         struct ext4_iloc iloc;
803         void *end;
804         int error;
805
806         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
807                 return 0;
808         error = ext4_get_inode_loc(inode, &iloc);
809         if (error)
810                 return error;
811         raw_inode = ext4_raw_inode(&iloc);
812         header = IHDR(inode, raw_inode);
813         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
814         error = xattr_check_inode(inode, header, end);
815         if (error)
816                 goto cleanup;
817         error = ext4_xattr_list_entries(dentry, IFIRST(header),
818                                         buffer, buffer_size);
819
820 cleanup:
821         brelse(iloc.bh);
822         return error;
823 }
824
825 /*
826  * Inode operation listxattr()
827  *
828  * d_inode(dentry)->i_rwsem: don't care
829  *
830  * Copy a list of attribute names into the buffer
831  * provided, or compute the buffer size required.
832  * Buffer is NULL to compute the size of the buffer required.
833  *
834  * Returns a negative error number on failure, or the number of bytes
835  * used / required on success.
836  */
837 ssize_t
838 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
839 {
840         int ret, ret2;
841
842         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
843         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
844         if (ret < 0)
845                 goto errout;
846         if (buffer) {
847                 buffer += ret;
848                 buffer_size -= ret;
849         }
850         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
851         if (ret < 0)
852                 goto errout;
853         ret += ret2;
854 errout:
855         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
856         return ret;
857 }
858
859 /*
860  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
861  * not set, set it.
862  */
863 static void ext4_xattr_update_super_block(handle_t *handle,
864                                           struct super_block *sb)
865 {
866         if (ext4_has_feature_xattr(sb))
867                 return;
868
869         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
870         if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
871                                           EXT4_JTR_NONE) == 0) {
872                 lock_buffer(EXT4_SB(sb)->s_sbh);
873                 ext4_set_feature_xattr(sb);
874                 ext4_superblock_csum_set(sb);
875                 unlock_buffer(EXT4_SB(sb)->s_sbh);
876                 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
877         }
878 }
879
880 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
881 {
882         struct ext4_iloc iloc = { .bh = NULL };
883         struct buffer_head *bh = NULL;
884         struct ext4_inode *raw_inode;
885         struct ext4_xattr_ibody_header *header;
886         struct ext4_xattr_entry *entry;
887         qsize_t ea_inode_refs = 0;
888         void *end;
889         int ret;
890
891         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
892
893         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
894                 ret = ext4_get_inode_loc(inode, &iloc);
895                 if (ret)
896                         goto out;
897                 raw_inode = ext4_raw_inode(&iloc);
898                 header = IHDR(inode, raw_inode);
899                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
900                 ret = xattr_check_inode(inode, header, end);
901                 if (ret)
902                         goto out;
903
904                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
905                      entry = EXT4_XATTR_NEXT(entry))
906                         if (entry->e_value_inum)
907                                 ea_inode_refs++;
908         }
909
910         if (EXT4_I(inode)->i_file_acl) {
911                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
912                 if (IS_ERR(bh)) {
913                         ret = PTR_ERR(bh);
914                         bh = NULL;
915                         goto out;
916                 }
917
918                 ret = ext4_xattr_check_block(inode, bh);
919                 if (ret)
920                         goto out;
921
922                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
923                      entry = EXT4_XATTR_NEXT(entry))
924                         if (entry->e_value_inum)
925                                 ea_inode_refs++;
926         }
927         *usage = ea_inode_refs + 1;
928         ret = 0;
929 out:
930         brelse(iloc.bh);
931         brelse(bh);
932         return ret;
933 }
934
935 static inline size_t round_up_cluster(struct inode *inode, size_t length)
936 {
937         struct super_block *sb = inode->i_sb;
938         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
939                                     inode->i_blkbits);
940         size_t mask = ~(cluster_size - 1);
941
942         return (length + cluster_size - 1) & mask;
943 }
944
945 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
946 {
947         int err;
948
949         err = dquot_alloc_inode(inode);
950         if (err)
951                 return err;
952         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
953         if (err)
954                 dquot_free_inode(inode);
955         return err;
956 }
957
958 static void ext4_xattr_inode_free_quota(struct inode *parent,
959                                         struct inode *ea_inode,
960                                         size_t len)
961 {
962         if (ea_inode &&
963             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
964                 return;
965         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
966         dquot_free_inode(parent);
967 }
968
969 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
970                              struct buffer_head *block_bh, size_t value_len,
971                              bool is_create)
972 {
973         int credits;
974         int blocks;
975
976         /*
977          * 1) Owner inode update
978          * 2) Ref count update on old xattr block
979          * 3) new xattr block
980          * 4) block bitmap update for new xattr block
981          * 5) group descriptor for new xattr block
982          * 6) block bitmap update for old xattr block
983          * 7) group descriptor for old block
984          *
985          * 6 & 7 can happen if we have two racing threads T_a and T_b
986          * which are each trying to set an xattr on inodes I_a and I_b
987          * which were both initially sharing an xattr block.
988          */
989         credits = 7;
990
991         /* Quota updates. */
992         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
993
994         /*
995          * In case of inline data, we may push out the data to a block,
996          * so we need to reserve credits for this eventuality
997          */
998         if (inode && ext4_has_inline_data(inode))
999                 credits += ext4_writepage_trans_blocks(inode) + 1;
1000
1001         /* We are done if ea_inode feature is not enabled. */
1002         if (!ext4_has_feature_ea_inode(sb))
1003                 return credits;
1004
1005         /* New ea_inode, inode map, block bitmap, group descriptor. */
1006         credits += 4;
1007
1008         /* Data blocks. */
1009         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1010
1011         /* Indirection block or one level of extent tree. */
1012         blocks += 1;
1013
1014         /* Block bitmap and group descriptor updates for each block. */
1015         credits += blocks * 2;
1016
1017         /* Blocks themselves. */
1018         credits += blocks;
1019
1020         if (!is_create) {
1021                 /* Dereference ea_inode holding old xattr value.
1022                  * Old ea_inode, inode map, block bitmap, group descriptor.
1023                  */
1024                 credits += 4;
1025
1026                 /* Data blocks for old ea_inode. */
1027                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
1028
1029                 /* Indirection block or one level of extent tree for old
1030                  * ea_inode.
1031                  */
1032                 blocks += 1;
1033
1034                 /* Block bitmap and group descriptor updates for each block. */
1035                 credits += blocks * 2;
1036         }
1037
1038         /* We may need to clone the existing xattr block in which case we need
1039          * to increment ref counts for existing ea_inodes referenced by it.
1040          */
1041         if (block_bh) {
1042                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
1043
1044                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
1045                         if (entry->e_value_inum)
1046                                 /* Ref count update on ea_inode. */
1047                                 credits += 1;
1048         }
1049         return credits;
1050 }
1051
1052 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1053                                        int ref_change)
1054 {
1055         struct ext4_iloc iloc;
1056         s64 ref_count;
1057         int ret;
1058
1059         inode_lock(ea_inode);
1060
1061         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1062         if (ret)
1063                 goto out;
1064
1065         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1066         ref_count += ref_change;
1067         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1068
1069         if (ref_change > 0) {
1070                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1071                           ea_inode->i_ino, ref_count);
1072
1073                 if (ref_count == 1) {
1074                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1075                                   ea_inode->i_ino, ea_inode->i_nlink);
1076
1077                         set_nlink(ea_inode, 1);
1078                         ext4_orphan_del(handle, ea_inode);
1079                 }
1080         } else {
1081                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1082                           ea_inode->i_ino, ref_count);
1083
1084                 if (ref_count == 0) {
1085                         WARN_ONCE(ea_inode->i_nlink != 1,
1086                                   "EA inode %lu i_nlink=%u",
1087                                   ea_inode->i_ino, ea_inode->i_nlink);
1088
1089                         clear_nlink(ea_inode);
1090                         ext4_orphan_add(handle, ea_inode);
1091                 }
1092         }
1093
1094         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1095         if (ret)
1096                 ext4_warning_inode(ea_inode,
1097                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1098 out:
1099         inode_unlock(ea_inode);
1100         return ret;
1101 }
1102
1103 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1104 {
1105         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1106 }
1107
1108 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1109 {
1110         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1111 }
1112
1113 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1114                                         struct ext4_xattr_entry *first)
1115 {
1116         struct inode *ea_inode;
1117         struct ext4_xattr_entry *entry;
1118         struct ext4_xattr_entry *failed_entry;
1119         unsigned int ea_ino;
1120         int err, saved_err;
1121
1122         for (entry = first; !IS_LAST_ENTRY(entry);
1123              entry = EXT4_XATTR_NEXT(entry)) {
1124                 if (!entry->e_value_inum)
1125                         continue;
1126                 ea_ino = le32_to_cpu(entry->e_value_inum);
1127                 err = ext4_xattr_inode_iget(parent, ea_ino,
1128                                             le32_to_cpu(entry->e_hash),
1129                                             &ea_inode);
1130                 if (err)
1131                         goto cleanup;
1132                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1133                 if (err) {
1134                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1135                         iput(ea_inode);
1136                         goto cleanup;
1137                 }
1138                 iput(ea_inode);
1139         }
1140         return 0;
1141
1142 cleanup:
1143         saved_err = err;
1144         failed_entry = entry;
1145
1146         for (entry = first; entry != failed_entry;
1147              entry = EXT4_XATTR_NEXT(entry)) {
1148                 if (!entry->e_value_inum)
1149                         continue;
1150                 ea_ino = le32_to_cpu(entry->e_value_inum);
1151                 err = ext4_xattr_inode_iget(parent, ea_ino,
1152                                             le32_to_cpu(entry->e_hash),
1153                                             &ea_inode);
1154                 if (err) {
1155                         ext4_warning(parent->i_sb,
1156                                      "cleanup ea_ino %u iget error %d", ea_ino,
1157                                      err);
1158                         continue;
1159                 }
1160                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1161                 if (err)
1162                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1163                                            err);
1164                 iput(ea_inode);
1165         }
1166         return saved_err;
1167 }
1168
1169 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1170                         struct buffer_head *bh, bool block_csum, bool dirty)
1171 {
1172         int error;
1173
1174         if (bh && dirty) {
1175                 if (block_csum)
1176                         ext4_xattr_block_csum_set(inode, bh);
1177                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1178                 if (error) {
1179                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1180                                      error);
1181                         return error;
1182                 }
1183         }
1184         return 0;
1185 }
1186
1187 static void
1188 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1189                              struct buffer_head *bh,
1190                              struct ext4_xattr_entry *first, bool block_csum,
1191                              struct ext4_xattr_inode_array **ea_inode_array,
1192                              int extra_credits, bool skip_quota)
1193 {
1194         struct inode *ea_inode;
1195         struct ext4_xattr_entry *entry;
1196         bool dirty = false;
1197         unsigned int ea_ino;
1198         int err;
1199         int credits;
1200
1201         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1202         credits = 2 + extra_credits;
1203
1204         for (entry = first; !IS_LAST_ENTRY(entry);
1205              entry = EXT4_XATTR_NEXT(entry)) {
1206                 if (!entry->e_value_inum)
1207                         continue;
1208                 ea_ino = le32_to_cpu(entry->e_value_inum);
1209                 err = ext4_xattr_inode_iget(parent, ea_ino,
1210                                             le32_to_cpu(entry->e_hash),
1211                                             &ea_inode);
1212                 if (err)
1213                         continue;
1214
1215                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1216                 if (err) {
1217                         ext4_warning_inode(ea_inode,
1218                                            "Expand inode array err=%d", err);
1219                         iput(ea_inode);
1220                         continue;
1221                 }
1222
1223                 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1224                         ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1225                         ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1226                                               dirty));
1227                 if (err < 0) {
1228                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1229                                            err);
1230                         continue;
1231                 }
1232                 if (err > 0) {
1233                         err = ext4_journal_get_write_access(handle,
1234                                         parent->i_sb, bh, EXT4_JTR_NONE);
1235                         if (err) {
1236                                 ext4_warning_inode(ea_inode,
1237                                                 "Re-get write access err=%d",
1238                                                 err);
1239                                 continue;
1240                         }
1241                 }
1242
1243                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1244                 if (err) {
1245                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1246                                            err);
1247                         continue;
1248                 }
1249
1250                 if (!skip_quota)
1251                         ext4_xattr_inode_free_quota(parent, ea_inode,
1252                                               le32_to_cpu(entry->e_value_size));
1253
1254                 /*
1255                  * Forget about ea_inode within the same transaction that
1256                  * decrements the ref count. This avoids duplicate decrements in
1257                  * case the rest of the work spills over to subsequent
1258                  * transactions.
1259                  */
1260                 entry->e_value_inum = 0;
1261                 entry->e_value_size = 0;
1262
1263                 dirty = true;
1264         }
1265
1266         if (dirty) {
1267                 /*
1268                  * Note that we are deliberately skipping csum calculation for
1269                  * the final update because we do not expect any journal
1270                  * restarts until xattr block is freed.
1271                  */
1272
1273                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1274                 if (err)
1275                         ext4_warning_inode(parent,
1276                                            "handle dirty metadata err=%d", err);
1277         }
1278 }
1279
1280 /*
1281  * Release the xattr block BH: If the reference count is > 1, decrement it;
1282  * otherwise free the block.
1283  */
1284 static void
1285 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1286                          struct buffer_head *bh,
1287                          struct ext4_xattr_inode_array **ea_inode_array,
1288                          int extra_credits)
1289 {
1290         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1291         u32 hash, ref;
1292         int error = 0;
1293
1294         BUFFER_TRACE(bh, "get_write_access");
1295         error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1296                                               EXT4_JTR_NONE);
1297         if (error)
1298                 goto out;
1299
1300 retry_ref:
1301         lock_buffer(bh);
1302         hash = le32_to_cpu(BHDR(bh)->h_hash);
1303         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1304         if (ref == 1) {
1305                 ea_bdebug(bh, "refcount now=0; freeing");
1306                 /*
1307                  * This must happen under buffer lock for
1308                  * ext4_xattr_block_set() to reliably detect freed block
1309                  */
1310                 if (ea_block_cache) {
1311                         struct mb_cache_entry *oe;
1312
1313                         oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1314                                                           bh->b_blocknr);
1315                         if (oe) {
1316                                 unlock_buffer(bh);
1317                                 mb_cache_entry_wait_unused(oe);
1318                                 mb_cache_entry_put(ea_block_cache, oe);
1319                                 goto retry_ref;
1320                         }
1321                 }
1322                 get_bh(bh);
1323                 unlock_buffer(bh);
1324
1325                 if (ext4_has_feature_ea_inode(inode->i_sb))
1326                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1327                                                      BFIRST(bh),
1328                                                      true /* block_csum */,
1329                                                      ea_inode_array,
1330                                                      extra_credits,
1331                                                      true /* skip_quota */);
1332                 ext4_free_blocks(handle, inode, bh, 0, 1,
1333                                  EXT4_FREE_BLOCKS_METADATA |
1334                                  EXT4_FREE_BLOCKS_FORGET);
1335         } else {
1336                 ref--;
1337                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1338                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1339                         struct mb_cache_entry *ce;
1340
1341                         if (ea_block_cache) {
1342                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1343                                                         bh->b_blocknr);
1344                                 if (ce) {
1345                                         set_bit(MBE_REUSABLE_B, &ce->e_flags);
1346                                         mb_cache_entry_put(ea_block_cache, ce);
1347                                 }
1348                         }
1349                 }
1350
1351                 ext4_xattr_block_csum_set(inode, bh);
1352                 /*
1353                  * Beware of this ugliness: Releasing of xattr block references
1354                  * from different inodes can race and so we have to protect
1355                  * from a race where someone else frees the block (and releases
1356                  * its journal_head) before we are done dirtying the buffer. In
1357                  * nojournal mode this race is harmless and we actually cannot
1358                  * call ext4_handle_dirty_metadata() with locked buffer as
1359                  * that function can call sync_dirty_buffer() so for that case
1360                  * we handle the dirtying after unlocking the buffer.
1361                  */
1362                 if (ext4_handle_valid(handle))
1363                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1364                 unlock_buffer(bh);
1365                 if (!ext4_handle_valid(handle))
1366                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1367                 if (IS_SYNC(inode))
1368                         ext4_handle_sync(handle);
1369                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1370                 ea_bdebug(bh, "refcount now=%d; releasing",
1371                           le32_to_cpu(BHDR(bh)->h_refcount));
1372         }
1373 out:
1374         ext4_std_error(inode->i_sb, error);
1375         return;
1376 }
1377
1378 /*
1379  * Find the available free space for EAs. This also returns the total number of
1380  * bytes used by EA entries.
1381  */
1382 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1383                                     size_t *min_offs, void *base, int *total)
1384 {
1385         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1386                 if (!last->e_value_inum && last->e_value_size) {
1387                         size_t offs = le16_to_cpu(last->e_value_offs);
1388                         if (offs < *min_offs)
1389                                 *min_offs = offs;
1390                 }
1391                 if (total)
1392                         *total += EXT4_XATTR_LEN(last->e_name_len);
1393         }
1394         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1395 }
1396
1397 /*
1398  * Write the value of the EA in an inode.
1399  */
1400 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1401                                   const void *buf, int bufsize)
1402 {
1403         struct buffer_head *bh = NULL;
1404         unsigned long block = 0;
1405         int blocksize = ea_inode->i_sb->s_blocksize;
1406         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1407         int csize, wsize = 0;
1408         int ret = 0, ret2 = 0;
1409         int retries = 0;
1410
1411 retry:
1412         while (ret >= 0 && ret < max_blocks) {
1413                 struct ext4_map_blocks map;
1414                 map.m_lblk = block += ret;
1415                 map.m_len = max_blocks -= ret;
1416
1417                 ret = ext4_map_blocks(handle, ea_inode, &map,
1418                                       EXT4_GET_BLOCKS_CREATE);
1419                 if (ret <= 0) {
1420                         ext4_mark_inode_dirty(handle, ea_inode);
1421                         if (ret == -ENOSPC &&
1422                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1423                                 ret = 0;
1424                                 goto retry;
1425                         }
1426                         break;
1427                 }
1428         }
1429
1430         if (ret < 0)
1431                 return ret;
1432
1433         block = 0;
1434         while (wsize < bufsize) {
1435                 brelse(bh);
1436                 csize = (bufsize - wsize) > blocksize ? blocksize :
1437                                                                 bufsize - wsize;
1438                 bh = ext4_getblk(handle, ea_inode, block, 0);
1439                 if (IS_ERR(bh))
1440                         return PTR_ERR(bh);
1441                 if (!bh) {
1442                         WARN_ON_ONCE(1);
1443                         EXT4_ERROR_INODE(ea_inode,
1444                                          "ext4_getblk() return bh = NULL");
1445                         return -EFSCORRUPTED;
1446                 }
1447                 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1448                                                    EXT4_JTR_NONE);
1449                 if (ret)
1450                         goto out;
1451
1452                 memcpy(bh->b_data, buf, csize);
1453                 set_buffer_uptodate(bh);
1454                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1455
1456                 buf += csize;
1457                 wsize += csize;
1458                 block += 1;
1459         }
1460
1461         inode_lock(ea_inode);
1462         i_size_write(ea_inode, wsize);
1463         ext4_update_i_disksize(ea_inode, wsize);
1464         inode_unlock(ea_inode);
1465
1466         ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1467         if (unlikely(ret2 && !ret))
1468                 ret = ret2;
1469
1470 out:
1471         brelse(bh);
1472
1473         return ret;
1474 }
1475
1476 /*
1477  * Create an inode to store the value of a large EA.
1478  */
1479 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1480                                              struct inode *inode, u32 hash)
1481 {
1482         struct inode *ea_inode = NULL;
1483         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1484         int err;
1485
1486         if (inode->i_sb->s_root == NULL) {
1487                 ext4_warning(inode->i_sb,
1488                              "refuse to create EA inode when umounting");
1489                 WARN_ON(1);
1490                 return ERR_PTR(-EINVAL);
1491         }
1492
1493         /*
1494          * Let the next inode be the goal, so we try and allocate the EA inode
1495          * in the same group, or nearby one.
1496          */
1497         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1498                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1499                                   EXT4_EA_INODE_FL);
1500         if (!IS_ERR(ea_inode)) {
1501                 ea_inode->i_op = &ext4_file_inode_operations;
1502                 ea_inode->i_fop = &ext4_file_operations;
1503                 ext4_set_aops(ea_inode);
1504                 ext4_xattr_inode_set_class(ea_inode);
1505                 unlock_new_inode(ea_inode);
1506                 ext4_xattr_inode_set_ref(ea_inode, 1);
1507                 ext4_xattr_inode_set_hash(ea_inode, hash);
1508                 err = ext4_mark_inode_dirty(handle, ea_inode);
1509                 if (!err)
1510                         err = ext4_inode_attach_jinode(ea_inode);
1511                 if (err) {
1512                         if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1513                                 ext4_warning_inode(ea_inode,
1514                                         "cleanup dec ref error %d", err);
1515                         iput(ea_inode);
1516                         return ERR_PTR(err);
1517                 }
1518
1519                 /*
1520                  * Xattr inodes are shared therefore quota charging is performed
1521                  * at a higher level.
1522                  */
1523                 dquot_free_inode(ea_inode);
1524                 dquot_drop(ea_inode);
1525                 inode_lock(ea_inode);
1526                 ea_inode->i_flags |= S_NOQUOTA;
1527                 inode_unlock(ea_inode);
1528         }
1529
1530         return ea_inode;
1531 }
1532
1533 static struct inode *
1534 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1535                             size_t value_len, u32 hash)
1536 {
1537         struct inode *ea_inode;
1538         struct mb_cache_entry *ce;
1539         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1540         void *ea_data;
1541
1542         if (!ea_inode_cache)
1543                 return NULL;
1544
1545         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1546         if (!ce)
1547                 return NULL;
1548
1549         WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1550                      !(current->flags & PF_MEMALLOC_NOFS));
1551
1552         ea_data = kvmalloc(value_len, GFP_KERNEL);
1553         if (!ea_data) {
1554                 mb_cache_entry_put(ea_inode_cache, ce);
1555                 return NULL;
1556         }
1557
1558         while (ce) {
1559                 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1560                                      EXT4_IGET_NORMAL);
1561                 if (!IS_ERR(ea_inode) &&
1562                     !is_bad_inode(ea_inode) &&
1563                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1564                     i_size_read(ea_inode) == value_len &&
1565                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1566                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1567                                                     value_len) &&
1568                     !memcmp(value, ea_data, value_len)) {
1569                         mb_cache_entry_touch(ea_inode_cache, ce);
1570                         mb_cache_entry_put(ea_inode_cache, ce);
1571                         kvfree(ea_data);
1572                         return ea_inode;
1573                 }
1574
1575                 if (!IS_ERR(ea_inode))
1576                         iput(ea_inode);
1577                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1578         }
1579         kvfree(ea_data);
1580         return NULL;
1581 }
1582
1583 /*
1584  * Add value of the EA in an inode.
1585  */
1586 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1587                                           const void *value, size_t value_len,
1588                                           struct inode **ret_inode)
1589 {
1590         struct inode *ea_inode;
1591         u32 hash;
1592         int err;
1593
1594         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1595         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1596         if (ea_inode) {
1597                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1598                 if (err) {
1599                         iput(ea_inode);
1600                         return err;
1601                 }
1602
1603                 *ret_inode = ea_inode;
1604                 return 0;
1605         }
1606
1607         /* Create an inode for the EA value */
1608         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1609         if (IS_ERR(ea_inode))
1610                 return PTR_ERR(ea_inode);
1611
1612         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1613         if (err) {
1614                 if (ext4_xattr_inode_dec_ref(handle, ea_inode))
1615                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d", err);
1616                 iput(ea_inode);
1617                 return err;
1618         }
1619
1620         if (EA_INODE_CACHE(inode))
1621                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1622                                       ea_inode->i_ino, true /* reusable */);
1623
1624         *ret_inode = ea_inode;
1625         return 0;
1626 }
1627
1628 /*
1629  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1630  * feature is enabled.
1631  */
1632 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1633
1634 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1635                                 struct ext4_xattr_search *s,
1636                                 handle_t *handle, struct inode *inode,
1637                                 bool is_block)
1638 {
1639         struct ext4_xattr_entry *last, *next;
1640         struct ext4_xattr_entry *here = s->here;
1641         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1642         int in_inode = i->in_inode;
1643         struct inode *old_ea_inode = NULL;
1644         struct inode *new_ea_inode = NULL;
1645         size_t old_size, new_size;
1646         int ret;
1647
1648         /* Space used by old and new values. */
1649         old_size = (!s->not_found && !here->e_value_inum) ?
1650                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1651         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1652
1653         /*
1654          * Optimization for the simple case when old and new values have the
1655          * same padded sizes. Not applicable if external inodes are involved.
1656          */
1657         if (new_size && new_size == old_size) {
1658                 size_t offs = le16_to_cpu(here->e_value_offs);
1659                 void *val = s->base + offs;
1660
1661                 here->e_value_size = cpu_to_le32(i->value_len);
1662                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1663                         memset(val, 0, new_size);
1664                 } else {
1665                         memcpy(val, i->value, i->value_len);
1666                         /* Clear padding bytes. */
1667                         memset(val + i->value_len, 0, new_size - i->value_len);
1668                 }
1669                 goto update_hash;
1670         }
1671
1672         /* Compute min_offs and last. */
1673         last = s->first;
1674         for (; !IS_LAST_ENTRY(last); last = next) {
1675                 next = EXT4_XATTR_NEXT(last);
1676                 if ((void *)next >= s->end) {
1677                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1678                         ret = -EFSCORRUPTED;
1679                         goto out;
1680                 }
1681                 if (!last->e_value_inum && last->e_value_size) {
1682                         size_t offs = le16_to_cpu(last->e_value_offs);
1683                         if (offs < min_offs)
1684                                 min_offs = offs;
1685                 }
1686         }
1687
1688         /* Check whether we have enough space. */
1689         if (i->value) {
1690                 size_t free;
1691
1692                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1693                 if (!s->not_found)
1694                         free += EXT4_XATTR_LEN(name_len) + old_size;
1695
1696                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1697                         ret = -ENOSPC;
1698                         goto out;
1699                 }
1700
1701                 /*
1702                  * If storing the value in an external inode is an option,
1703                  * reserve space for xattr entries/names in the external
1704                  * attribute block so that a long value does not occupy the
1705                  * whole space and prevent further entries being added.
1706                  */
1707                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1708                     new_size && is_block &&
1709                     (min_offs + old_size - new_size) <
1710                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1711                         ret = -ENOSPC;
1712                         goto out;
1713                 }
1714         }
1715
1716         /*
1717          * Getting access to old and new ea inodes is subject to failures.
1718          * Finish that work before doing any modifications to the xattr data.
1719          */
1720         if (!s->not_found && here->e_value_inum) {
1721                 ret = ext4_xattr_inode_iget(inode,
1722                                             le32_to_cpu(here->e_value_inum),
1723                                             le32_to_cpu(here->e_hash),
1724                                             &old_ea_inode);
1725                 if (ret) {
1726                         old_ea_inode = NULL;
1727                         goto out;
1728                 }
1729         }
1730         if (i->value && in_inode) {
1731                 WARN_ON_ONCE(!i->value_len);
1732
1733                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1734                 if (ret)
1735                         goto out;
1736
1737                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1738                                                      i->value_len,
1739                                                      &new_ea_inode);
1740                 if (ret) {
1741                         new_ea_inode = NULL;
1742                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1743                         goto out;
1744                 }
1745         }
1746
1747         if (old_ea_inode) {
1748                 /* We are ready to release ref count on the old_ea_inode. */
1749                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1750                 if (ret) {
1751                         /* Release newly required ref count on new_ea_inode. */
1752                         if (new_ea_inode) {
1753                                 int err;
1754
1755                                 err = ext4_xattr_inode_dec_ref(handle,
1756                                                                new_ea_inode);
1757                                 if (err)
1758                                         ext4_warning_inode(new_ea_inode,
1759                                                   "dec ref new_ea_inode err=%d",
1760                                                   err);
1761                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1762                                                             i->value_len);
1763                         }
1764                         goto out;
1765                 }
1766
1767                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1768                                             le32_to_cpu(here->e_value_size));
1769         }
1770
1771         /* No failures allowed past this point. */
1772
1773         if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1774                 /* Remove the old value. */
1775                 void *first_val = s->base + min_offs;
1776                 size_t offs = le16_to_cpu(here->e_value_offs);
1777                 void *val = s->base + offs;
1778
1779                 memmove(first_val + old_size, first_val, val - first_val);
1780                 memset(first_val, 0, old_size);
1781                 min_offs += old_size;
1782
1783                 /* Adjust all value offsets. */
1784                 last = s->first;
1785                 while (!IS_LAST_ENTRY(last)) {
1786                         size_t o = le16_to_cpu(last->e_value_offs);
1787
1788                         if (!last->e_value_inum &&
1789                             last->e_value_size && o < offs)
1790                                 last->e_value_offs = cpu_to_le16(o + old_size);
1791                         last = EXT4_XATTR_NEXT(last);
1792                 }
1793         }
1794
1795         if (!i->value) {
1796                 /* Remove old name. */
1797                 size_t size = EXT4_XATTR_LEN(name_len);
1798
1799                 last = ENTRY((void *)last - size);
1800                 memmove(here, (void *)here + size,
1801                         (void *)last - (void *)here + sizeof(__u32));
1802                 memset(last, 0, size);
1803         } else if (s->not_found) {
1804                 /* Insert new name. */
1805                 size_t size = EXT4_XATTR_LEN(name_len);
1806                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1807
1808                 memmove((void *)here + size, here, rest);
1809                 memset(here, 0, size);
1810                 here->e_name_index = i->name_index;
1811                 here->e_name_len = name_len;
1812                 memcpy(here->e_name, i->name, name_len);
1813         } else {
1814                 /* This is an update, reset value info. */
1815                 here->e_value_inum = 0;
1816                 here->e_value_offs = 0;
1817                 here->e_value_size = 0;
1818         }
1819
1820         if (i->value) {
1821                 /* Insert new value. */
1822                 if (in_inode) {
1823                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1824                 } else if (i->value_len) {
1825                         void *val = s->base + min_offs - new_size;
1826
1827                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1828                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1829                                 memset(val, 0, new_size);
1830                         } else {
1831                                 memcpy(val, i->value, i->value_len);
1832                                 /* Clear padding bytes. */
1833                                 memset(val + i->value_len, 0,
1834                                        new_size - i->value_len);
1835                         }
1836                 }
1837                 here->e_value_size = cpu_to_le32(i->value_len);
1838         }
1839
1840 update_hash:
1841         if (i->value) {
1842                 __le32 hash = 0;
1843
1844                 /* Entry hash calculation. */
1845                 if (in_inode) {
1846                         __le32 crc32c_hash;
1847
1848                         /*
1849                          * Feed crc32c hash instead of the raw value for entry
1850                          * hash calculation. This is to avoid walking
1851                          * potentially long value buffer again.
1852                          */
1853                         crc32c_hash = cpu_to_le32(
1854                                        ext4_xattr_inode_get_hash(new_ea_inode));
1855                         hash = ext4_xattr_hash_entry(here->e_name,
1856                                                      here->e_name_len,
1857                                                      &crc32c_hash, 1);
1858                 } else if (is_block) {
1859                         __le32 *value = s->base + le16_to_cpu(
1860                                                         here->e_value_offs);
1861
1862                         hash = ext4_xattr_hash_entry(here->e_name,
1863                                                      here->e_name_len, value,
1864                                                      new_size >> 2);
1865                 }
1866                 here->e_hash = hash;
1867         }
1868
1869         if (is_block)
1870                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1871
1872         ret = 0;
1873 out:
1874         iput(old_ea_inode);
1875         iput(new_ea_inode);
1876         return ret;
1877 }
1878
1879 struct ext4_xattr_block_find {
1880         struct ext4_xattr_search s;
1881         struct buffer_head *bh;
1882 };
1883
1884 static int
1885 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1886                       struct ext4_xattr_block_find *bs)
1887 {
1888         struct super_block *sb = inode->i_sb;
1889         int error;
1890
1891         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1892                   i->name_index, i->name, i->value, (long)i->value_len);
1893
1894         if (EXT4_I(inode)->i_file_acl) {
1895                 /* The inode already has an extended attribute block. */
1896                 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1897                 if (IS_ERR(bs->bh)) {
1898                         error = PTR_ERR(bs->bh);
1899                         bs->bh = NULL;
1900                         return error;
1901                 }
1902                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1903                         atomic_read(&(bs->bh->b_count)),
1904                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1905                 error = ext4_xattr_check_block(inode, bs->bh);
1906                 if (error)
1907                         return error;
1908                 /* Find the named attribute. */
1909                 bs->s.base = BHDR(bs->bh);
1910                 bs->s.first = BFIRST(bs->bh);
1911                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1912                 bs->s.here = bs->s.first;
1913                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1914                                          i->name_index, i->name, 1);
1915                 if (error && error != -ENODATA)
1916                         return error;
1917                 bs->s.not_found = error;
1918         }
1919         return 0;
1920 }
1921
1922 static int
1923 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1924                      struct ext4_xattr_info *i,
1925                      struct ext4_xattr_block_find *bs)
1926 {
1927         struct super_block *sb = inode->i_sb;
1928         struct buffer_head *new_bh = NULL;
1929         struct ext4_xattr_search s_copy = bs->s;
1930         struct ext4_xattr_search *s = &s_copy;
1931         struct mb_cache_entry *ce = NULL;
1932         int error = 0;
1933         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1934         struct inode *ea_inode = NULL, *tmp_inode;
1935         size_t old_ea_inode_quota = 0;
1936         unsigned int ea_ino;
1937
1938
1939 #define header(x) ((struct ext4_xattr_header *)(x))
1940
1941         if (s->base) {
1942                 int offset = (char *)s->here - bs->bh->b_data;
1943
1944                 BUFFER_TRACE(bs->bh, "get_write_access");
1945                 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1946                                                       EXT4_JTR_NONE);
1947                 if (error)
1948                         goto cleanup;
1949                 lock_buffer(bs->bh);
1950
1951                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1952                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1953
1954                         /*
1955                          * This must happen under buffer lock for
1956                          * ext4_xattr_block_set() to reliably detect modified
1957                          * block
1958                          */
1959                         if (ea_block_cache) {
1960                                 struct mb_cache_entry *oe;
1961
1962                                 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1963                                         hash, bs->bh->b_blocknr);
1964                                 if (oe) {
1965                                         /*
1966                                          * Xattr block is getting reused. Leave
1967                                          * it alone.
1968                                          */
1969                                         mb_cache_entry_put(ea_block_cache, oe);
1970                                         goto clone_block;
1971                                 }
1972                         }
1973                         ea_bdebug(bs->bh, "modifying in-place");
1974                         error = ext4_xattr_set_entry(i, s, handle, inode,
1975                                                      true /* is_block */);
1976                         ext4_xattr_block_csum_set(inode, bs->bh);
1977                         unlock_buffer(bs->bh);
1978                         if (error == -EFSCORRUPTED)
1979                                 goto bad_block;
1980                         if (!error)
1981                                 error = ext4_handle_dirty_metadata(handle,
1982                                                                    inode,
1983                                                                    bs->bh);
1984                         if (error)
1985                                 goto cleanup;
1986                         goto inserted;
1987                 }
1988 clone_block:
1989                 unlock_buffer(bs->bh);
1990                 ea_bdebug(bs->bh, "cloning");
1991                 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1992                 error = -ENOMEM;
1993                 if (s->base == NULL)
1994                         goto cleanup;
1995                 s->first = ENTRY(header(s->base)+1);
1996                 header(s->base)->h_refcount = cpu_to_le32(1);
1997                 s->here = ENTRY(s->base + offset);
1998                 s->end = s->base + bs->bh->b_size;
1999
2000                 /*
2001                  * If existing entry points to an xattr inode, we need
2002                  * to prevent ext4_xattr_set_entry() from decrementing
2003                  * ref count on it because the reference belongs to the
2004                  * original block. In this case, make the entry look
2005                  * like it has an empty value.
2006                  */
2007                 if (!s->not_found && s->here->e_value_inum) {
2008                         ea_ino = le32_to_cpu(s->here->e_value_inum);
2009                         error = ext4_xattr_inode_iget(inode, ea_ino,
2010                                       le32_to_cpu(s->here->e_hash),
2011                                       &tmp_inode);
2012                         if (error)
2013                                 goto cleanup;
2014
2015                         if (!ext4_test_inode_state(tmp_inode,
2016                                         EXT4_STATE_LUSTRE_EA_INODE)) {
2017                                 /*
2018                                  * Defer quota free call for previous
2019                                  * inode until success is guaranteed.
2020                                  */
2021                                 old_ea_inode_quota = le32_to_cpu(
2022                                                 s->here->e_value_size);
2023                         }
2024                         iput(tmp_inode);
2025
2026                         s->here->e_value_inum = 0;
2027                         s->here->e_value_size = 0;
2028                 }
2029         } else {
2030                 /* Allocate a buffer where we construct the new block. */
2031                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
2032                 error = -ENOMEM;
2033                 if (s->base == NULL)
2034                         goto cleanup;
2035                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2036                 header(s->base)->h_blocks = cpu_to_le32(1);
2037                 header(s->base)->h_refcount = cpu_to_le32(1);
2038                 s->first = ENTRY(header(s->base)+1);
2039                 s->here = ENTRY(header(s->base)+1);
2040                 s->end = s->base + sb->s_blocksize;
2041         }
2042
2043         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
2044         if (error == -EFSCORRUPTED)
2045                 goto bad_block;
2046         if (error)
2047                 goto cleanup;
2048
2049         if (i->value && s->here->e_value_inum) {
2050                 /*
2051                  * A ref count on ea_inode has been taken as part of the call to
2052                  * ext4_xattr_set_entry() above. We would like to drop this
2053                  * extra ref but we have to wait until the xattr block is
2054                  * initialized and has its own ref count on the ea_inode.
2055                  */
2056                 ea_ino = le32_to_cpu(s->here->e_value_inum);
2057                 error = ext4_xattr_inode_iget(inode, ea_ino,
2058                                               le32_to_cpu(s->here->e_hash),
2059                                               &ea_inode);
2060                 if (error) {
2061                         ea_inode = NULL;
2062                         goto cleanup;
2063                 }
2064         }
2065
2066 inserted:
2067         if (!IS_LAST_ENTRY(s->first)) {
2068                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
2069                                                      &ce);
2070                 if (new_bh) {
2071                         /* We found an identical block in the cache. */
2072                         if (new_bh == bs->bh)
2073                                 ea_bdebug(new_bh, "keeping");
2074                         else {
2075                                 u32 ref;
2076
2077                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
2078
2079                                 /* The old block is released after updating
2080                                    the inode. */
2081                                 error = dquot_alloc_block(inode,
2082                                                 EXT4_C2B(EXT4_SB(sb), 1));
2083                                 if (error)
2084                                         goto cleanup;
2085                                 BUFFER_TRACE(new_bh, "get_write_access");
2086                                 error = ext4_journal_get_write_access(
2087                                                 handle, sb, new_bh,
2088                                                 EXT4_JTR_NONE);
2089                                 if (error)
2090                                         goto cleanup_dquot;
2091                                 lock_buffer(new_bh);
2092                                 /*
2093                                  * We have to be careful about races with
2094                                  * adding references to xattr block. Once we
2095                                  * hold buffer lock xattr block's state is
2096                                  * stable so we can check the additional
2097                                  * reference fits.
2098                                  */
2099                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2100                                 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2101                                         /*
2102                                          * Undo everything and check mbcache
2103                                          * again.
2104                                          */
2105                                         unlock_buffer(new_bh);
2106                                         dquot_free_block(inode,
2107                                                          EXT4_C2B(EXT4_SB(sb),
2108                                                                   1));
2109                                         brelse(new_bh);
2110                                         mb_cache_entry_put(ea_block_cache, ce);
2111                                         ce = NULL;
2112                                         new_bh = NULL;
2113                                         goto inserted;
2114                                 }
2115                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2116                                 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2117                                         clear_bit(MBE_REUSABLE_B, &ce->e_flags);
2118                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2119                                           ref);
2120                                 ext4_xattr_block_csum_set(inode, new_bh);
2121                                 unlock_buffer(new_bh);
2122                                 error = ext4_handle_dirty_metadata(handle,
2123                                                                    inode,
2124                                                                    new_bh);
2125                                 if (error)
2126                                         goto cleanup_dquot;
2127                         }
2128                         mb_cache_entry_touch(ea_block_cache, ce);
2129                         mb_cache_entry_put(ea_block_cache, ce);
2130                         ce = NULL;
2131                 } else if (bs->bh && s->base == bs->bh->b_data) {
2132                         /* We were modifying this block in-place. */
2133                         ea_bdebug(bs->bh, "keeping this block");
2134                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2135                         new_bh = bs->bh;
2136                         get_bh(new_bh);
2137                 } else {
2138                         /* We need to allocate a new block */
2139                         ext4_fsblk_t goal, block;
2140
2141                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2142
2143                         goal = ext4_group_first_block_no(sb,
2144                                                 EXT4_I(inode)->i_block_group);
2145                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2146                                                      NULL, &error);
2147                         if (error)
2148                                 goto cleanup;
2149
2150                         ea_idebug(inode, "creating block %llu",
2151                                   (unsigned long long)block);
2152
2153                         new_bh = sb_getblk(sb, block);
2154                         if (unlikely(!new_bh)) {
2155                                 error = -ENOMEM;
2156 getblk_failed:
2157                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2158                                                  EXT4_FREE_BLOCKS_METADATA);
2159                                 goto cleanup;
2160                         }
2161                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2162                                                       ENTRY(header(s->base)+1));
2163                         if (error)
2164                                 goto getblk_failed;
2165                         if (ea_inode) {
2166                                 /* Drop the extra ref on ea_inode. */
2167                                 error = ext4_xattr_inode_dec_ref(handle,
2168                                                                  ea_inode);
2169                                 if (error)
2170                                         ext4_warning_inode(ea_inode,
2171                                                            "dec ref error=%d",
2172                                                            error);
2173                                 iput(ea_inode);
2174                                 ea_inode = NULL;
2175                         }
2176
2177                         lock_buffer(new_bh);
2178                         error = ext4_journal_get_create_access(handle, sb,
2179                                                         new_bh, EXT4_JTR_NONE);
2180                         if (error) {
2181                                 unlock_buffer(new_bh);
2182                                 error = -EIO;
2183                                 goto getblk_failed;
2184                         }
2185                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2186                         ext4_xattr_block_csum_set(inode, new_bh);
2187                         set_buffer_uptodate(new_bh);
2188                         unlock_buffer(new_bh);
2189                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2190                         error = ext4_handle_dirty_metadata(handle, inode,
2191                                                            new_bh);
2192                         if (error)
2193                                 goto cleanup;
2194                 }
2195         }
2196
2197         if (old_ea_inode_quota)
2198                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2199
2200         /* Update the inode. */
2201         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2202
2203         /* Drop the previous xattr block. */
2204         if (bs->bh && bs->bh != new_bh) {
2205                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2206
2207                 ext4_xattr_release_block(handle, inode, bs->bh,
2208                                          &ea_inode_array,
2209                                          0 /* extra_credits */);
2210                 ext4_xattr_inode_array_free(ea_inode_array);
2211         }
2212         error = 0;
2213
2214 cleanup:
2215         if (ea_inode) {
2216                 int error2;
2217
2218                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2219                 if (error2)
2220                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2221                                            error2);
2222
2223                 /* If there was an error, revert the quota charge. */
2224                 if (error)
2225                         ext4_xattr_inode_free_quota(inode, ea_inode,
2226                                                     i_size_read(ea_inode));
2227                 iput(ea_inode);
2228         }
2229         if (ce)
2230                 mb_cache_entry_put(ea_block_cache, ce);
2231         brelse(new_bh);
2232         if (!(bs->bh && s->base == bs->bh->b_data))
2233                 kfree(s->base);
2234
2235         return error;
2236
2237 cleanup_dquot:
2238         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2239         goto cleanup;
2240
2241 bad_block:
2242         EXT4_ERROR_INODE(inode, "bad block %llu",
2243                          EXT4_I(inode)->i_file_acl);
2244         goto cleanup;
2245
2246 #undef header
2247 }
2248
2249 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2250                           struct ext4_xattr_ibody_find *is)
2251 {
2252         struct ext4_xattr_ibody_header *header;
2253         struct ext4_inode *raw_inode;
2254         int error;
2255
2256         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2257                 return 0;
2258
2259         raw_inode = ext4_raw_inode(&is->iloc);
2260         header = IHDR(inode, raw_inode);
2261         is->s.base = is->s.first = IFIRST(header);
2262         is->s.here = is->s.first;
2263         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2264         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2265                 error = xattr_check_inode(inode, header, is->s.end);
2266                 if (error)
2267                         return error;
2268                 /* Find the named attribute. */
2269                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2270                                          i->name_index, i->name, 0);
2271                 if (error && error != -ENODATA)
2272                         return error;
2273                 is->s.not_found = error;
2274         }
2275         return 0;
2276 }
2277
2278 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2279                                 struct ext4_xattr_info *i,
2280                                 struct ext4_xattr_ibody_find *is)
2281 {
2282         struct ext4_xattr_ibody_header *header;
2283         struct ext4_xattr_search *s = &is->s;
2284         int error;
2285
2286         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2287                 return -ENOSPC;
2288
2289         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2290         if (error)
2291                 return error;
2292         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2293         if (!IS_LAST_ENTRY(s->first)) {
2294                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2295                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2296         } else {
2297                 header->h_magic = cpu_to_le32(0);
2298                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2299         }
2300         return 0;
2301 }
2302
2303 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2304                                  struct ext4_xattr_info *i)
2305 {
2306         void *value;
2307
2308         /* When e_value_inum is set the value is stored externally. */
2309         if (s->here->e_value_inum)
2310                 return 0;
2311         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2312                 return 0;
2313         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2314         return !memcmp(value, i->value, i->value_len);
2315 }
2316
2317 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2318 {
2319         struct buffer_head *bh;
2320         int error;
2321
2322         if (!EXT4_I(inode)->i_file_acl)
2323                 return NULL;
2324         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2325         if (IS_ERR(bh))
2326                 return bh;
2327         error = ext4_xattr_check_block(inode, bh);
2328         if (error) {
2329                 brelse(bh);
2330                 return ERR_PTR(error);
2331         }
2332         return bh;
2333 }
2334
2335 /*
2336  * ext4_xattr_set_handle()
2337  *
2338  * Create, replace or remove an extended attribute for this inode.  Value
2339  * is NULL to remove an existing extended attribute, and non-NULL to
2340  * either replace an existing extended attribute, or create a new extended
2341  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2342  * specify that an extended attribute must exist and must not exist
2343  * previous to the call, respectively.
2344  *
2345  * Returns 0, or a negative error number on failure.
2346  */
2347 int
2348 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2349                       const char *name, const void *value, size_t value_len,
2350                       int flags)
2351 {
2352         struct ext4_xattr_info i = {
2353                 .name_index = name_index,
2354                 .name = name,
2355                 .value = value,
2356                 .value_len = value_len,
2357                 .in_inode = 0,
2358         };
2359         struct ext4_xattr_ibody_find is = {
2360                 .s = { .not_found = -ENODATA, },
2361         };
2362         struct ext4_xattr_block_find bs = {
2363                 .s = { .not_found = -ENODATA, },
2364         };
2365         int no_expand;
2366         int error;
2367
2368         if (!name)
2369                 return -EINVAL;
2370         if (strlen(name) > 255)
2371                 return -ERANGE;
2372
2373         ext4_write_lock_xattr(inode, &no_expand);
2374
2375         /* Check journal credits under write lock. */
2376         if (ext4_handle_valid(handle)) {
2377                 struct buffer_head *bh;
2378                 int credits;
2379
2380                 bh = ext4_xattr_get_block(inode);
2381                 if (IS_ERR(bh)) {
2382                         error = PTR_ERR(bh);
2383                         goto cleanup;
2384                 }
2385
2386                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2387                                                    value_len,
2388                                                    flags & XATTR_CREATE);
2389                 brelse(bh);
2390
2391                 if (jbd2_handle_buffer_credits(handle) < credits) {
2392                         error = -ENOSPC;
2393                         goto cleanup;
2394                 }
2395                 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2396         }
2397
2398         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2399         if (error)
2400                 goto cleanup;
2401
2402         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2403                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2404                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2405                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2406         }
2407
2408         error = ext4_xattr_ibody_find(inode, &i, &is);
2409         if (error)
2410                 goto cleanup;
2411         if (is.s.not_found)
2412                 error = ext4_xattr_block_find(inode, &i, &bs);
2413         if (error)
2414                 goto cleanup;
2415         if (is.s.not_found && bs.s.not_found) {
2416                 error = -ENODATA;
2417                 if (flags & XATTR_REPLACE)
2418                         goto cleanup;
2419                 error = 0;
2420                 if (!value)
2421                         goto cleanup;
2422         } else {
2423                 error = -EEXIST;
2424                 if (flags & XATTR_CREATE)
2425                         goto cleanup;
2426         }
2427
2428         if (!value) {
2429                 if (!is.s.not_found)
2430                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2431                 else if (!bs.s.not_found)
2432                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2433         } else {
2434                 error = 0;
2435                 /* Xattr value did not change? Save us some work and bail out */
2436                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2437                         goto cleanup;
2438                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2439                         goto cleanup;
2440
2441                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2442                     (EXT4_XATTR_SIZE(i.value_len) >
2443                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2444                         i.in_inode = 1;
2445 retry_inode:
2446                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2447                 if (!error && !bs.s.not_found) {
2448                         i.value = NULL;
2449                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2450                 } else if (error == -ENOSPC) {
2451                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2452                                 brelse(bs.bh);
2453                                 bs.bh = NULL;
2454                                 error = ext4_xattr_block_find(inode, &i, &bs);
2455                                 if (error)
2456                                         goto cleanup;
2457                         }
2458                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2459                         if (!error && !is.s.not_found) {
2460                                 i.value = NULL;
2461                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2462                                                              &is);
2463                         } else if (error == -ENOSPC) {
2464                                 /*
2465                                  * Xattr does not fit in the block, store at
2466                                  * external inode if possible.
2467                                  */
2468                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2469                                     i.value_len && !i.in_inode) {
2470                                         i.in_inode = 1;
2471                                         goto retry_inode;
2472                                 }
2473                         }
2474                 }
2475         }
2476         if (!error) {
2477                 ext4_xattr_update_super_block(handle, inode->i_sb);
2478                 inode->i_ctime = current_time(inode);
2479                 inode_inc_iversion(inode);
2480                 if (!value)
2481                         no_expand = 0;
2482                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2483                 /*
2484                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2485                  * error != 0.
2486                  */
2487                 is.iloc.bh = NULL;
2488                 if (IS_SYNC(inode))
2489                         ext4_handle_sync(handle);
2490         }
2491         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2492
2493 cleanup:
2494         brelse(is.iloc.bh);
2495         brelse(bs.bh);
2496         ext4_write_unlock_xattr(inode, &no_expand);
2497         return error;
2498 }
2499
2500 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2501                            bool is_create, int *credits)
2502 {
2503         struct buffer_head *bh;
2504         int err;
2505
2506         *credits = 0;
2507
2508         if (!EXT4_SB(inode->i_sb)->s_journal)
2509                 return 0;
2510
2511         down_read(&EXT4_I(inode)->xattr_sem);
2512
2513         bh = ext4_xattr_get_block(inode);
2514         if (IS_ERR(bh)) {
2515                 err = PTR_ERR(bh);
2516         } else {
2517                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2518                                                     value_len, is_create);
2519                 brelse(bh);
2520                 err = 0;
2521         }
2522
2523         up_read(&EXT4_I(inode)->xattr_sem);
2524         return err;
2525 }
2526
2527 /*
2528  * ext4_xattr_set()
2529  *
2530  * Like ext4_xattr_set_handle, but start from an inode. This extended
2531  * attribute modification is a filesystem transaction by itself.
2532  *
2533  * Returns 0, or a negative error number on failure.
2534  */
2535 int
2536 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2537                const void *value, size_t value_len, int flags)
2538 {
2539         handle_t *handle;
2540         struct super_block *sb = inode->i_sb;
2541         int error, retries = 0;
2542         int credits;
2543
2544         error = dquot_initialize(inode);
2545         if (error)
2546                 return error;
2547
2548 retry:
2549         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2550                                        &credits);
2551         if (error)
2552                 return error;
2553
2554         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2555         if (IS_ERR(handle)) {
2556                 error = PTR_ERR(handle);
2557         } else {
2558                 int error2;
2559
2560                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2561                                               value, value_len, flags);
2562                 error2 = ext4_journal_stop(handle);
2563                 if (error == -ENOSPC &&
2564                     ext4_should_retry_alloc(sb, &retries))
2565                         goto retry;
2566                 if (error == 0)
2567                         error = error2;
2568         }
2569         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2570
2571         return error;
2572 }
2573
2574 /*
2575  * Shift the EA entries in the inode to create space for the increased
2576  * i_extra_isize.
2577  */
2578 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2579                                      int value_offs_shift, void *to,
2580                                      void *from, size_t n)
2581 {
2582         struct ext4_xattr_entry *last = entry;
2583         int new_offs;
2584
2585         /* We always shift xattr headers further thus offsets get lower */
2586         BUG_ON(value_offs_shift > 0);
2587
2588         /* Adjust the value offsets of the entries */
2589         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2590                 if (!last->e_value_inum && last->e_value_size) {
2591                         new_offs = le16_to_cpu(last->e_value_offs) +
2592                                                         value_offs_shift;
2593                         last->e_value_offs = cpu_to_le16(new_offs);
2594                 }
2595         }
2596         /* Shift the entries by n bytes */
2597         memmove(to, from, n);
2598 }
2599
2600 /*
2601  * Move xattr pointed to by 'entry' from inode into external xattr block
2602  */
2603 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2604                                     struct ext4_inode *raw_inode,
2605                                     struct ext4_xattr_entry *entry)
2606 {
2607         struct ext4_xattr_ibody_find *is = NULL;
2608         struct ext4_xattr_block_find *bs = NULL;
2609         char *buffer = NULL, *b_entry_name = NULL;
2610         size_t value_size = le32_to_cpu(entry->e_value_size);
2611         struct ext4_xattr_info i = {
2612                 .value = NULL,
2613                 .value_len = 0,
2614                 .name_index = entry->e_name_index,
2615                 .in_inode = !!entry->e_value_inum,
2616         };
2617         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
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
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 (entry->e_value_inum && 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 cleanup:
2856         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2857                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2858                              inode->i_ino);
2859                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2860         }
2861         return error;
2862 }
2863
2864 #define EIA_INCR 16 /* must be 2^n */
2865 #define EIA_MASK (EIA_INCR - 1)
2866
2867 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2868  * If @ea_inode_array is new or full it will be grown and the old
2869  * contents copied over.
2870  */
2871 static int
2872 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2873                         struct inode *inode)
2874 {
2875         if (*ea_inode_array == NULL) {
2876                 /*
2877                  * Start with 15 inodes, so it fits into a power-of-two size.
2878                  * If *ea_inode_array is NULL, this is essentially offsetof()
2879                  */
2880                 (*ea_inode_array) =
2881                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2882                                          inodes[EIA_MASK]),
2883                                 GFP_NOFS);
2884                 if (*ea_inode_array == NULL)
2885                         return -ENOMEM;
2886                 (*ea_inode_array)->count = 0;
2887         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2888                 /* expand the array once all 15 + n * 16 slots are full */
2889                 struct ext4_xattr_inode_array *new_array = NULL;
2890                 int count = (*ea_inode_array)->count;
2891
2892                 /* if new_array is NULL, this is essentially offsetof() */
2893                 new_array = kmalloc(
2894                                 offsetof(struct ext4_xattr_inode_array,
2895                                          inodes[count + EIA_INCR]),
2896                                 GFP_NOFS);
2897                 if (new_array == NULL)
2898                         return -ENOMEM;
2899                 memcpy(new_array, *ea_inode_array,
2900                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2901                 kfree(*ea_inode_array);
2902                 *ea_inode_array = new_array;
2903         }
2904         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2905         return 0;
2906 }
2907
2908 /*
2909  * ext4_xattr_delete_inode()
2910  *
2911  * Free extended attribute resources associated with this inode. Traverse
2912  * all entries and decrement reference on any xattr inodes associated with this
2913  * inode. This is called immediately before an inode is freed. We have exclusive
2914  * access to the inode. If an orphan inode is deleted it will also release its
2915  * references on xattr block and xattr inodes.
2916  */
2917 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2918                             struct ext4_xattr_inode_array **ea_inode_array,
2919                             int extra_credits)
2920 {
2921         struct buffer_head *bh = NULL;
2922         struct ext4_xattr_ibody_header *header;
2923         struct ext4_iloc iloc = { .bh = NULL };
2924         struct ext4_xattr_entry *entry;
2925         struct inode *ea_inode;
2926         int error;
2927
2928         error = ext4_journal_ensure_credits(handle, extra_credits,
2929                         ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2930         if (error < 0) {
2931                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2932                 goto cleanup;
2933         }
2934
2935         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2936             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2937
2938                 error = ext4_get_inode_loc(inode, &iloc);
2939                 if (error) {
2940                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2941                         goto cleanup;
2942                 }
2943
2944                 error = ext4_journal_get_write_access(handle, inode->i_sb,
2945                                                 iloc.bh, EXT4_JTR_NONE);
2946                 if (error) {
2947                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2948                                          error);
2949                         goto cleanup;
2950                 }
2951
2952                 header = IHDR(inode, ext4_raw_inode(&iloc));
2953                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2954                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2955                                                      IFIRST(header),
2956                                                      false /* block_csum */,
2957                                                      ea_inode_array,
2958                                                      extra_credits,
2959                                                      false /* skip_quota */);
2960         }
2961
2962         if (EXT4_I(inode)->i_file_acl) {
2963                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2964                 if (IS_ERR(bh)) {
2965                         error = PTR_ERR(bh);
2966                         if (error == -EIO) {
2967                                 EXT4_ERROR_INODE_ERR(inode, EIO,
2968                                                      "block %llu read error",
2969                                                      EXT4_I(inode)->i_file_acl);
2970                         }
2971                         bh = NULL;
2972                         goto cleanup;
2973                 }
2974                 error = ext4_xattr_check_block(inode, bh);
2975                 if (error)
2976                         goto cleanup;
2977
2978                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2979                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2980                              entry = EXT4_XATTR_NEXT(entry)) {
2981                                 if (!entry->e_value_inum)
2982                                         continue;
2983                                 error = ext4_xattr_inode_iget(inode,
2984                                               le32_to_cpu(entry->e_value_inum),
2985                                               le32_to_cpu(entry->e_hash),
2986                                               &ea_inode);
2987                                 if (error)
2988                                         continue;
2989                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2990                                               le32_to_cpu(entry->e_value_size));
2991                                 iput(ea_inode);
2992                         }
2993
2994                 }
2995
2996                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2997                                          extra_credits);
2998                 /*
2999                  * Update i_file_acl value in the same transaction that releases
3000                  * block.
3001                  */
3002                 EXT4_I(inode)->i_file_acl = 0;
3003                 error = ext4_mark_inode_dirty(handle, inode);
3004                 if (error) {
3005                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
3006                                          error);
3007                         goto cleanup;
3008                 }
3009                 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
3010         }
3011         error = 0;
3012 cleanup:
3013         brelse(iloc.bh);
3014         brelse(bh);
3015         return error;
3016 }
3017
3018 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
3019 {
3020         int idx;
3021
3022         if (ea_inode_array == NULL)
3023                 return;
3024
3025         for (idx = 0; idx < ea_inode_array->count; ++idx)
3026                 iput(ea_inode_array->inodes[idx]);
3027         kfree(ea_inode_array);
3028 }
3029
3030 /*
3031  * ext4_xattr_block_cache_insert()
3032  *
3033  * Create a new entry in the extended attribute block cache, and insert
3034  * it unless such an entry is already in the cache.
3035  *
3036  * Returns 0, or a negative error number on failure.
3037  */
3038 static void
3039 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
3040                               struct buffer_head *bh)
3041 {
3042         struct ext4_xattr_header *header = BHDR(bh);
3043         __u32 hash = le32_to_cpu(header->h_hash);
3044         int reusable = le32_to_cpu(header->h_refcount) <
3045                        EXT4_XATTR_REFCOUNT_MAX;
3046         int error;
3047
3048         if (!ea_block_cache)
3049                 return;
3050         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
3051                                       bh->b_blocknr, reusable);
3052         if (error) {
3053                 if (error == -EBUSY)
3054                         ea_bdebug(bh, "already in cache");
3055         } else
3056                 ea_bdebug(bh, "inserting [%x]", (int)hash);
3057 }
3058
3059 /*
3060  * ext4_xattr_cmp()
3061  *
3062  * Compare two extended attribute blocks for equality.
3063  *
3064  * Returns 0 if the blocks are equal, 1 if they differ, and
3065  * a negative error number on errors.
3066  */
3067 static int
3068 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3069                struct ext4_xattr_header *header2)
3070 {
3071         struct ext4_xattr_entry *entry1, *entry2;
3072
3073         entry1 = ENTRY(header1+1);
3074         entry2 = ENTRY(header2+1);
3075         while (!IS_LAST_ENTRY(entry1)) {
3076                 if (IS_LAST_ENTRY(entry2))
3077                         return 1;
3078                 if (entry1->e_hash != entry2->e_hash ||
3079                     entry1->e_name_index != entry2->e_name_index ||
3080                     entry1->e_name_len != entry2->e_name_len ||
3081                     entry1->e_value_size != entry2->e_value_size ||
3082                     entry1->e_value_inum != entry2->e_value_inum ||
3083                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3084                         return 1;
3085                 if (!entry1->e_value_inum &&
3086                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3087                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3088                            le32_to_cpu(entry1->e_value_size)))
3089                         return 1;
3090
3091                 entry1 = EXT4_XATTR_NEXT(entry1);
3092                 entry2 = EXT4_XATTR_NEXT(entry2);
3093         }
3094         if (!IS_LAST_ENTRY(entry2))
3095                 return 1;
3096         return 0;
3097 }
3098
3099 /*
3100  * ext4_xattr_block_cache_find()
3101  *
3102  * Find an identical extended attribute block.
3103  *
3104  * Returns a pointer to the block found, or NULL if such a block was
3105  * not found or an error occurred.
3106  */
3107 static struct buffer_head *
3108 ext4_xattr_block_cache_find(struct inode *inode,
3109                             struct ext4_xattr_header *header,
3110                             struct mb_cache_entry **pce)
3111 {
3112         __u32 hash = le32_to_cpu(header->h_hash);
3113         struct mb_cache_entry *ce;
3114         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3115
3116         if (!ea_block_cache)
3117                 return NULL;
3118         if (!header->h_hash)
3119                 return NULL;  /* never share */
3120         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3121         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3122         while (ce) {
3123                 struct buffer_head *bh;
3124
3125                 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3126                 if (IS_ERR(bh)) {
3127                         if (PTR_ERR(bh) == -ENOMEM)
3128                                 return NULL;
3129                         bh = NULL;
3130                         EXT4_ERROR_INODE(inode, "block %lu read error",
3131                                          (unsigned long)ce->e_value);
3132                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3133                         *pce = ce;
3134                         return bh;
3135                 }
3136                 brelse(bh);
3137                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3138         }
3139         return NULL;
3140 }
3141
3142 #define NAME_HASH_SHIFT 5
3143 #define VALUE_HASH_SHIFT 16
3144
3145 /*
3146  * ext4_xattr_hash_entry()
3147  *
3148  * Compute the hash of an extended attribute.
3149  */
3150 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3151                                     size_t value_count)
3152 {
3153         __u32 hash = 0;
3154
3155         while (name_len--) {
3156                 hash = (hash << NAME_HASH_SHIFT) ^
3157                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3158                        (unsigned char)*name++;
3159         }
3160         while (value_count--) {
3161                 hash = (hash << VALUE_HASH_SHIFT) ^
3162                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3163                        le32_to_cpu(*value++);
3164         }
3165         return cpu_to_le32(hash);
3166 }
3167
3168 /*
3169  * ext4_xattr_hash_entry_signed()
3170  *
3171  * Compute the hash of an extended attribute incorrectly.
3172  */
3173 static __le32 ext4_xattr_hash_entry_signed(char *name, size_t name_len, __le32 *value, size_t value_count)
3174 {
3175         __u32 hash = 0;
3176
3177         while (name_len--) {
3178                 hash = (hash << NAME_HASH_SHIFT) ^
3179                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3180                        (signed char)*name++;
3181         }
3182         while (value_count--) {
3183                 hash = (hash << VALUE_HASH_SHIFT) ^
3184                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3185                        le32_to_cpu(*value++);
3186         }
3187         return cpu_to_le32(hash);
3188 }
3189
3190 #undef NAME_HASH_SHIFT
3191 #undef VALUE_HASH_SHIFT
3192
3193 #define BLOCK_HASH_SHIFT 16
3194
3195 /*
3196  * ext4_xattr_rehash()
3197  *
3198  * Re-compute the extended attribute hash value after an entry has changed.
3199  */
3200 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3201 {
3202         struct ext4_xattr_entry *here;
3203         __u32 hash = 0;
3204
3205         here = ENTRY(header+1);
3206         while (!IS_LAST_ENTRY(here)) {
3207                 if (!here->e_hash) {
3208                         /* Block is not shared if an entry's hash value == 0 */
3209                         hash = 0;
3210                         break;
3211                 }
3212                 hash = (hash << BLOCK_HASH_SHIFT) ^
3213                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3214                        le32_to_cpu(here->e_hash);
3215                 here = EXT4_XATTR_NEXT(here);
3216         }
3217         header->h_hash = cpu_to_le32(hash);
3218 }
3219
3220 #undef BLOCK_HASH_SHIFT
3221
3222 #define HASH_BUCKET_BITS        10
3223
3224 struct mb_cache *
3225 ext4_xattr_create_cache(void)
3226 {
3227         return mb_cache_create(HASH_BUCKET_BITS);
3228 }
3229
3230 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3231 {
3232         if (cache)
3233                 mb_cache_destroy(cache);
3234 }
3235