]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - fs/ext4/xattr.c
ARM: 8567/1: cache-uniphier: activate ways for secondary CPUs
[karo-tx-linux.git] / fs / ext4 / xattr.c
1 /*
2  * linux/fs/ext4/xattr.c
3  *
4  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
5  *
6  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
7  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
8  * Extended attributes for symlinks and special files added per
9  *  suggestion of Luka Renko <luka.renko@hermes.si>.
10  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
11  *  Red Hat Inc.
12  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
13  *  and Andreas Gruenbacher <agruen@suse.de>.
14  */
15
16 /*
17  * Extended attributes are stored directly in inodes (on file systems with
18  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
19  * field contains the block number if an inode uses an additional block. All
20  * attributes must fit in the inode and one additional block. Blocks that
21  * contain the identical set of attributes may be shared among several inodes.
22  * Identical blocks are detected by keeping a cache of blocks that have
23  * recently been accessed.
24  *
25  * The attributes in inodes and on blocks have a different header; the entries
26  * are stored in the same format:
27  *
28  *   +------------------+
29  *   | header           |
30  *   | entry 1          | |
31  *   | entry 2          | | growing downwards
32  *   | entry 3          | v
33  *   | four null bytes  |
34  *   | . . .            |
35  *   | value 1          | ^
36  *   | value 3          | | growing upwards
37  *   | value 2          | |
38  *   +------------------+
39  *
40  * The header is followed by multiple entry descriptors. In disk blocks, the
41  * entry descriptors are kept sorted. In inodes, they are unsorted. The
42  * attribute values are aligned to the end of the block in no specific order.
43  *
44  * Locking strategy
45  * ----------------
46  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
47  * EA blocks are only changed if they are exclusive to an inode, so
48  * holding xattr_sem also means that nothing but the EA block's reference
49  * count can change. Multiple writers to the same block are synchronized
50  * by the buffer lock.
51  */
52
53 #include <linux/init.h>
54 #include <linux/fs.h>
55 #include <linux/slab.h>
56 #include <linux/mbcache.h>
57 #include <linux/quotaops.h>
58 #include "ext4_jbd2.h"
59 #include "ext4.h"
60 #include "xattr.h"
61 #include "acl.h"
62
63 #ifdef EXT4_XATTR_DEBUG
64 # define ea_idebug(inode, f...) do { \
65                 printk(KERN_DEBUG "inode %s:%lu: ", \
66                         inode->i_sb->s_id, inode->i_ino); \
67                 printk(f); \
68                 printk("\n"); \
69         } while (0)
70 # define ea_bdebug(bh, f...) do { \
71                 printk(KERN_DEBUG "block %pg:%lu: ",               \
72                        bh->b_bdev, (unsigned long) bh->b_blocknr); \
73                 printk(f); \
74                 printk("\n"); \
75         } while (0)
76 #else
77 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
78 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
79 #endif
80
81 static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
82 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
83                                                  struct ext4_xattr_header *,
84                                                  struct mb_cache_entry **);
85 static void ext4_xattr_rehash(struct ext4_xattr_header *,
86                               struct ext4_xattr_entry *);
87 static int ext4_xattr_list(struct dentry *dentry, char *buffer,
88                            size_t buffer_size);
89
90 static const struct xattr_handler *ext4_xattr_handler_map[] = {
91         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
92 #ifdef CONFIG_EXT4_FS_POSIX_ACL
93         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
94         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
95 #endif
96         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
97 #ifdef CONFIG_EXT4_FS_SECURITY
98         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
99 #endif
100 };
101
102 const struct xattr_handler *ext4_xattr_handlers[] = {
103         &ext4_xattr_user_handler,
104         &ext4_xattr_trusted_handler,
105 #ifdef CONFIG_EXT4_FS_POSIX_ACL
106         &posix_acl_access_xattr_handler,
107         &posix_acl_default_xattr_handler,
108 #endif
109 #ifdef CONFIG_EXT4_FS_SECURITY
110         &ext4_xattr_security_handler,
111 #endif
112         NULL
113 };
114
115 #define EXT4_GET_MB_CACHE(inode)        (((struct ext4_sb_info *) \
116                                 inode->i_sb->s_fs_info)->s_mb_cache)
117
118 static __le32 ext4_xattr_block_csum(struct inode *inode,
119                                     sector_t block_nr,
120                                     struct ext4_xattr_header *hdr)
121 {
122         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
123         __u32 csum;
124         __le32 save_csum;
125         __le64 dsk_block_nr = cpu_to_le64(block_nr);
126
127         save_csum = hdr->h_checksum;
128         hdr->h_checksum = 0;
129         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
130                            sizeof(dsk_block_nr));
131         csum = ext4_chksum(sbi, csum, (__u8 *)hdr,
132                            EXT4_BLOCK_SIZE(inode->i_sb));
133
134         hdr->h_checksum = save_csum;
135         return cpu_to_le32(csum);
136 }
137
138 static int ext4_xattr_block_csum_verify(struct inode *inode,
139                                         sector_t block_nr,
140                                         struct ext4_xattr_header *hdr)
141 {
142         if (ext4_has_metadata_csum(inode->i_sb) &&
143             (hdr->h_checksum != ext4_xattr_block_csum(inode, block_nr, hdr)))
144                 return 0;
145         return 1;
146 }
147
148 static void ext4_xattr_block_csum_set(struct inode *inode,
149                                       sector_t block_nr,
150                                       struct ext4_xattr_header *hdr)
151 {
152         if (!ext4_has_metadata_csum(inode->i_sb))
153                 return;
154
155         hdr->h_checksum = ext4_xattr_block_csum(inode, block_nr, hdr);
156 }
157
158 static inline int ext4_handle_dirty_xattr_block(handle_t *handle,
159                                                 struct inode *inode,
160                                                 struct buffer_head *bh)
161 {
162         ext4_xattr_block_csum_set(inode, bh->b_blocknr, BHDR(bh));
163         return ext4_handle_dirty_metadata(handle, inode, bh);
164 }
165
166 static inline const struct xattr_handler *
167 ext4_xattr_handler(int name_index)
168 {
169         const struct xattr_handler *handler = NULL;
170
171         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
172                 handler = ext4_xattr_handler_map[name_index];
173         return handler;
174 }
175
176 /*
177  * Inode operation listxattr()
178  *
179  * d_inode(dentry)->i_mutex: don't care
180  */
181 ssize_t
182 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
183 {
184         return ext4_xattr_list(dentry, buffer, size);
185 }
186
187 static int
188 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
189                        void *value_start)
190 {
191         struct ext4_xattr_entry *e = entry;
192
193         while (!IS_LAST_ENTRY(e)) {
194                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
195                 if ((void *)next >= end)
196                         return -EFSCORRUPTED;
197                 e = next;
198         }
199
200         while (!IS_LAST_ENTRY(entry)) {
201                 if (entry->e_value_size != 0 &&
202                     (value_start + le16_to_cpu(entry->e_value_offs) <
203                      (void *)e + sizeof(__u32) ||
204                      value_start + le16_to_cpu(entry->e_value_offs) +
205                     le32_to_cpu(entry->e_value_size) > end))
206                         return -EFSCORRUPTED;
207                 entry = EXT4_XATTR_NEXT(entry);
208         }
209
210         return 0;
211 }
212
213 static inline int
214 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
215 {
216         int error;
217
218         if (buffer_verified(bh))
219                 return 0;
220
221         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
222             BHDR(bh)->h_blocks != cpu_to_le32(1))
223                 return -EFSCORRUPTED;
224         if (!ext4_xattr_block_csum_verify(inode, bh->b_blocknr, BHDR(bh)))
225                 return -EFSBADCRC;
226         error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size,
227                                        bh->b_data);
228         if (!error)
229                 set_buffer_verified(bh);
230         return error;
231 }
232
233 static inline int
234 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
235 {
236         size_t value_size = le32_to_cpu(entry->e_value_size);
237
238         if (entry->e_value_block != 0 || value_size > size ||
239             le16_to_cpu(entry->e_value_offs) + value_size > size)
240                 return -EFSCORRUPTED;
241         return 0;
242 }
243
244 static int
245 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
246                       const char *name, size_t size, int sorted)
247 {
248         struct ext4_xattr_entry *entry;
249         size_t name_len;
250         int cmp = 1;
251
252         if (name == NULL)
253                 return -EINVAL;
254         name_len = strlen(name);
255         entry = *pentry;
256         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
257                 cmp = name_index - entry->e_name_index;
258                 if (!cmp)
259                         cmp = name_len - entry->e_name_len;
260                 if (!cmp)
261                         cmp = memcmp(name, entry->e_name, name_len);
262                 if (cmp <= 0 && (sorted || cmp == 0))
263                         break;
264         }
265         *pentry = entry;
266         if (!cmp && ext4_xattr_check_entry(entry, size))
267                 return -EFSCORRUPTED;
268         return cmp ? -ENODATA : 0;
269 }
270
271 static int
272 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
273                      void *buffer, size_t buffer_size)
274 {
275         struct buffer_head *bh = NULL;
276         struct ext4_xattr_entry *entry;
277         size_t size;
278         int error;
279         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
280
281         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
282                   name_index, name, buffer, (long)buffer_size);
283
284         error = -ENODATA;
285         if (!EXT4_I(inode)->i_file_acl)
286                 goto cleanup;
287         ea_idebug(inode, "reading block %llu",
288                   (unsigned long long)EXT4_I(inode)->i_file_acl);
289         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
290         if (!bh)
291                 goto cleanup;
292         ea_bdebug(bh, "b_count=%d, refcount=%d",
293                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
294         if (ext4_xattr_check_block(inode, bh)) {
295 bad_block:
296                 EXT4_ERROR_INODE(inode, "bad block %llu",
297                                  EXT4_I(inode)->i_file_acl);
298                 error = -EFSCORRUPTED;
299                 goto cleanup;
300         }
301         ext4_xattr_cache_insert(ext4_mb_cache, bh);
302         entry = BFIRST(bh);
303         error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
304         if (error == -EFSCORRUPTED)
305                 goto bad_block;
306         if (error)
307                 goto cleanup;
308         size = le32_to_cpu(entry->e_value_size);
309         if (buffer) {
310                 error = -ERANGE;
311                 if (size > buffer_size)
312                         goto cleanup;
313                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
314                        size);
315         }
316         error = size;
317
318 cleanup:
319         brelse(bh);
320         return error;
321 }
322
323 int
324 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
325                      void *buffer, size_t buffer_size)
326 {
327         struct ext4_xattr_ibody_header *header;
328         struct ext4_xattr_entry *entry;
329         struct ext4_inode *raw_inode;
330         struct ext4_iloc iloc;
331         size_t size;
332         void *end;
333         int error;
334
335         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
336                 return -ENODATA;
337         error = ext4_get_inode_loc(inode, &iloc);
338         if (error)
339                 return error;
340         raw_inode = ext4_raw_inode(&iloc);
341         header = IHDR(inode, raw_inode);
342         entry = IFIRST(header);
343         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
344         error = ext4_xattr_check_names(entry, end, entry);
345         if (error)
346                 goto cleanup;
347         error = ext4_xattr_find_entry(&entry, name_index, name,
348                                       end - (void *)entry, 0);
349         if (error)
350                 goto cleanup;
351         size = le32_to_cpu(entry->e_value_size);
352         if (buffer) {
353                 error = -ERANGE;
354                 if (size > buffer_size)
355                         goto cleanup;
356                 memcpy(buffer, (void *)IFIRST(header) +
357                        le16_to_cpu(entry->e_value_offs), size);
358         }
359         error = size;
360
361 cleanup:
362         brelse(iloc.bh);
363         return error;
364 }
365
366 /*
367  * ext4_xattr_get()
368  *
369  * Copy an extended attribute into the buffer
370  * provided, or compute the buffer size required.
371  * Buffer is NULL to compute the size of the buffer required.
372  *
373  * Returns a negative error number on failure, or the number of bytes
374  * used / required on success.
375  */
376 int
377 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
378                void *buffer, size_t buffer_size)
379 {
380         int error;
381
382         if (strlen(name) > 255)
383                 return -ERANGE;
384
385         down_read(&EXT4_I(inode)->xattr_sem);
386         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
387                                      buffer_size);
388         if (error == -ENODATA)
389                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
390                                              buffer_size);
391         up_read(&EXT4_I(inode)->xattr_sem);
392         return error;
393 }
394
395 static int
396 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
397                         char *buffer, size_t buffer_size)
398 {
399         size_t rest = buffer_size;
400
401         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
402                 const struct xattr_handler *handler =
403                         ext4_xattr_handler(entry->e_name_index);
404
405                 if (handler && (!handler->list || handler->list(dentry))) {
406                         const char *prefix = handler->prefix ?: handler->name;
407                         size_t prefix_len = strlen(prefix);
408                         size_t size = prefix_len + entry->e_name_len + 1;
409
410                         if (buffer) {
411                                 if (size > rest)
412                                         return -ERANGE;
413                                 memcpy(buffer, prefix, prefix_len);
414                                 buffer += prefix_len;
415                                 memcpy(buffer, entry->e_name, entry->e_name_len);
416                                 buffer += entry->e_name_len;
417                                 *buffer++ = 0;
418                         }
419                         rest -= size;
420                 }
421         }
422         return buffer_size - rest;  /* total size */
423 }
424
425 static int
426 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
427 {
428         struct inode *inode = d_inode(dentry);
429         struct buffer_head *bh = NULL;
430         int error;
431         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
432
433         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
434                   buffer, (long)buffer_size);
435
436         error = 0;
437         if (!EXT4_I(inode)->i_file_acl)
438                 goto cleanup;
439         ea_idebug(inode, "reading block %llu",
440                   (unsigned long long)EXT4_I(inode)->i_file_acl);
441         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
442         error = -EIO;
443         if (!bh)
444                 goto cleanup;
445         ea_bdebug(bh, "b_count=%d, refcount=%d",
446                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
447         if (ext4_xattr_check_block(inode, bh)) {
448                 EXT4_ERROR_INODE(inode, "bad block %llu",
449                                  EXT4_I(inode)->i_file_acl);
450                 error = -EFSCORRUPTED;
451                 goto cleanup;
452         }
453         ext4_xattr_cache_insert(ext4_mb_cache, bh);
454         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
455
456 cleanup:
457         brelse(bh);
458
459         return error;
460 }
461
462 static int
463 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
464 {
465         struct inode *inode = d_inode(dentry);
466         struct ext4_xattr_ibody_header *header;
467         struct ext4_inode *raw_inode;
468         struct ext4_iloc iloc;
469         void *end;
470         int error;
471
472         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
473                 return 0;
474         error = ext4_get_inode_loc(inode, &iloc);
475         if (error)
476                 return error;
477         raw_inode = ext4_raw_inode(&iloc);
478         header = IHDR(inode, raw_inode);
479         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
480         error = ext4_xattr_check_names(IFIRST(header), end, IFIRST(header));
481         if (error)
482                 goto cleanup;
483         error = ext4_xattr_list_entries(dentry, IFIRST(header),
484                                         buffer, buffer_size);
485
486 cleanup:
487         brelse(iloc.bh);
488         return error;
489 }
490
491 /*
492  * ext4_xattr_list()
493  *
494  * Copy a list of attribute names into the buffer
495  * provided, or compute the buffer size required.
496  * Buffer is NULL to compute the size of the buffer required.
497  *
498  * Returns a negative error number on failure, or the number of bytes
499  * used / required on success.
500  */
501 static int
502 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
503 {
504         int ret, ret2;
505
506         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
507         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
508         if (ret < 0)
509                 goto errout;
510         if (buffer) {
511                 buffer += ret;
512                 buffer_size -= ret;
513         }
514         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
515         if (ret < 0)
516                 goto errout;
517         ret += ret2;
518 errout:
519         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
520         return ret;
521 }
522
523 /*
524  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
525  * not set, set it.
526  */
527 static void ext4_xattr_update_super_block(handle_t *handle,
528                                           struct super_block *sb)
529 {
530         if (ext4_has_feature_xattr(sb))
531                 return;
532
533         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
534         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
535                 ext4_set_feature_xattr(sb);
536                 ext4_handle_dirty_super(handle, sb);
537         }
538 }
539
540 /*
541  * Release the xattr block BH: If the reference count is > 1, decrement it;
542  * otherwise free the block.
543  */
544 static void
545 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
546                          struct buffer_head *bh)
547 {
548         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
549         u32 hash, ref;
550         int error = 0;
551
552         BUFFER_TRACE(bh, "get_write_access");
553         error = ext4_journal_get_write_access(handle, bh);
554         if (error)
555                 goto out;
556
557         lock_buffer(bh);
558         hash = le32_to_cpu(BHDR(bh)->h_hash);
559         ref = le32_to_cpu(BHDR(bh)->h_refcount);
560         if (ref == 1) {
561                 ea_bdebug(bh, "refcount now=0; freeing");
562                 /*
563                  * This must happen under buffer lock for
564                  * ext4_xattr_block_set() to reliably detect freed block
565                  */
566                 mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
567                 get_bh(bh);
568                 unlock_buffer(bh);
569                 ext4_free_blocks(handle, inode, bh, 0, 1,
570                                  EXT4_FREE_BLOCKS_METADATA |
571                                  EXT4_FREE_BLOCKS_FORGET);
572         } else {
573                 ref--;
574                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
575                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
576                         struct mb_cache_entry *ce;
577
578                         ce = mb_cache_entry_get(ext4_mb_cache, hash,
579                                                 bh->b_blocknr);
580                         if (ce) {
581                                 ce->e_reusable = 1;
582                                 mb_cache_entry_put(ext4_mb_cache, ce);
583                         }
584                 }
585
586                 /*
587                  * Beware of this ugliness: Releasing of xattr block references
588                  * from different inodes can race and so we have to protect
589                  * from a race where someone else frees the block (and releases
590                  * its journal_head) before we are done dirtying the buffer. In
591                  * nojournal mode this race is harmless and we actually cannot
592                  * call ext4_handle_dirty_xattr_block() with locked buffer as
593                  * that function can call sync_dirty_buffer() so for that case
594                  * we handle the dirtying after unlocking the buffer.
595                  */
596                 if (ext4_handle_valid(handle))
597                         error = ext4_handle_dirty_xattr_block(handle, inode,
598                                                               bh);
599                 unlock_buffer(bh);
600                 if (!ext4_handle_valid(handle))
601                         error = ext4_handle_dirty_xattr_block(handle, inode,
602                                                               bh);
603                 if (IS_SYNC(inode))
604                         ext4_handle_sync(handle);
605                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
606                 ea_bdebug(bh, "refcount now=%d; releasing",
607                           le32_to_cpu(BHDR(bh)->h_refcount));
608         }
609 out:
610         ext4_std_error(inode->i_sb, error);
611         return;
612 }
613
614 /*
615  * Find the available free space for EAs. This also returns the total number of
616  * bytes used by EA entries.
617  */
618 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
619                                     size_t *min_offs, void *base, int *total)
620 {
621         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
622                 if (!last->e_value_block && last->e_value_size) {
623                         size_t offs = le16_to_cpu(last->e_value_offs);
624                         if (offs < *min_offs)
625                                 *min_offs = offs;
626                 }
627                 if (total)
628                         *total += EXT4_XATTR_LEN(last->e_name_len);
629         }
630         return (*min_offs - ((void *)last - base) - sizeof(__u32));
631 }
632
633 static int
634 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
635 {
636         struct ext4_xattr_entry *last;
637         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
638
639         /* Compute min_offs and last. */
640         last = s->first;
641         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
642                 if (!last->e_value_block && last->e_value_size) {
643                         size_t offs = le16_to_cpu(last->e_value_offs);
644                         if (offs < min_offs)
645                                 min_offs = offs;
646                 }
647         }
648         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
649         if (!s->not_found) {
650                 if (!s->here->e_value_block && s->here->e_value_size) {
651                         size_t size = le32_to_cpu(s->here->e_value_size);
652                         free += EXT4_XATTR_SIZE(size);
653                 }
654                 free += EXT4_XATTR_LEN(name_len);
655         }
656         if (i->value) {
657                 if (free < EXT4_XATTR_LEN(name_len) +
658                            EXT4_XATTR_SIZE(i->value_len))
659                         return -ENOSPC;
660         }
661
662         if (i->value && s->not_found) {
663                 /* Insert the new name. */
664                 size_t size = EXT4_XATTR_LEN(name_len);
665                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
666                 memmove((void *)s->here + size, s->here, rest);
667                 memset(s->here, 0, size);
668                 s->here->e_name_index = i->name_index;
669                 s->here->e_name_len = name_len;
670                 memcpy(s->here->e_name, i->name, name_len);
671         } else {
672                 if (!s->here->e_value_block && s->here->e_value_size) {
673                         void *first_val = s->base + min_offs;
674                         size_t offs = le16_to_cpu(s->here->e_value_offs);
675                         void *val = s->base + offs;
676                         size_t size = EXT4_XATTR_SIZE(
677                                 le32_to_cpu(s->here->e_value_size));
678
679                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
680                                 /* The old and the new value have the same
681                                    size. Just replace. */
682                                 s->here->e_value_size =
683                                         cpu_to_le32(i->value_len);
684                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
685                                         memset(val, 0, size);
686                                 } else {
687                                         /* Clear pad bytes first. */
688                                         memset(val + size - EXT4_XATTR_PAD, 0,
689                                                EXT4_XATTR_PAD);
690                                         memcpy(val, i->value, i->value_len);
691                                 }
692                                 return 0;
693                         }
694
695                         /* Remove the old value. */
696                         memmove(first_val + size, first_val, val - first_val);
697                         memset(first_val, 0, size);
698                         s->here->e_value_size = 0;
699                         s->here->e_value_offs = 0;
700                         min_offs += size;
701
702                         /* Adjust all value offsets. */
703                         last = s->first;
704                         while (!IS_LAST_ENTRY(last)) {
705                                 size_t o = le16_to_cpu(last->e_value_offs);
706                                 if (!last->e_value_block &&
707                                     last->e_value_size && o < offs)
708                                         last->e_value_offs =
709                                                 cpu_to_le16(o + size);
710                                 last = EXT4_XATTR_NEXT(last);
711                         }
712                 }
713                 if (!i->value) {
714                         /* Remove the old name. */
715                         size_t size = EXT4_XATTR_LEN(name_len);
716                         last = ENTRY((void *)last - size);
717                         memmove(s->here, (void *)s->here + size,
718                                 (void *)last - (void *)s->here + sizeof(__u32));
719                         memset(last, 0, size);
720                 }
721         }
722
723         if (i->value) {
724                 /* Insert the new value. */
725                 s->here->e_value_size = cpu_to_le32(i->value_len);
726                 if (i->value_len) {
727                         size_t size = EXT4_XATTR_SIZE(i->value_len);
728                         void *val = s->base + min_offs - size;
729                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
730                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
731                                 memset(val, 0, size);
732                         } else {
733                                 /* Clear the pad bytes first. */
734                                 memset(val + size - EXT4_XATTR_PAD, 0,
735                                        EXT4_XATTR_PAD);
736                                 memcpy(val, i->value, i->value_len);
737                         }
738                 }
739         }
740         return 0;
741 }
742
743 struct ext4_xattr_block_find {
744         struct ext4_xattr_search s;
745         struct buffer_head *bh;
746 };
747
748 static int
749 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
750                       struct ext4_xattr_block_find *bs)
751 {
752         struct super_block *sb = inode->i_sb;
753         int error;
754
755         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
756                   i->name_index, i->name, i->value, (long)i->value_len);
757
758         if (EXT4_I(inode)->i_file_acl) {
759                 /* The inode already has an extended attribute block. */
760                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
761                 error = -EIO;
762                 if (!bs->bh)
763                         goto cleanup;
764                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
765                         atomic_read(&(bs->bh->b_count)),
766                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
767                 if (ext4_xattr_check_block(inode, bs->bh)) {
768                         EXT4_ERROR_INODE(inode, "bad block %llu",
769                                          EXT4_I(inode)->i_file_acl);
770                         error = -EFSCORRUPTED;
771                         goto cleanup;
772                 }
773                 /* Find the named attribute. */
774                 bs->s.base = BHDR(bs->bh);
775                 bs->s.first = BFIRST(bs->bh);
776                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
777                 bs->s.here = bs->s.first;
778                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
779                                               i->name, bs->bh->b_size, 1);
780                 if (error && error != -ENODATA)
781                         goto cleanup;
782                 bs->s.not_found = error;
783         }
784         error = 0;
785
786 cleanup:
787         return error;
788 }
789
790 static int
791 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
792                      struct ext4_xattr_info *i,
793                      struct ext4_xattr_block_find *bs)
794 {
795         struct super_block *sb = inode->i_sb;
796         struct buffer_head *new_bh = NULL;
797         struct ext4_xattr_search *s = &bs->s;
798         struct mb_cache_entry *ce = NULL;
799         int error = 0;
800         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
801
802 #define header(x) ((struct ext4_xattr_header *)(x))
803
804         if (i->value && i->value_len > sb->s_blocksize)
805                 return -ENOSPC;
806         if (s->base) {
807                 BUFFER_TRACE(bs->bh, "get_write_access");
808                 error = ext4_journal_get_write_access(handle, bs->bh);
809                 if (error)
810                         goto cleanup;
811                 lock_buffer(bs->bh);
812
813                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
814                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
815
816                         /*
817                          * This must happen under buffer lock for
818                          * ext4_xattr_block_set() to reliably detect modified
819                          * block
820                          */
821                         mb_cache_entry_delete_block(ext4_mb_cache, hash,
822                                                     bs->bh->b_blocknr);
823                         ea_bdebug(bs->bh, "modifying in-place");
824                         error = ext4_xattr_set_entry(i, s);
825                         if (!error) {
826                                 if (!IS_LAST_ENTRY(s->first))
827                                         ext4_xattr_rehash(header(s->base),
828                                                           s->here);
829                                 ext4_xattr_cache_insert(ext4_mb_cache,
830                                         bs->bh);
831                         }
832                         unlock_buffer(bs->bh);
833                         if (error == -EFSCORRUPTED)
834                                 goto bad_block;
835                         if (!error)
836                                 error = ext4_handle_dirty_xattr_block(handle,
837                                                                       inode,
838                                                                       bs->bh);
839                         if (error)
840                                 goto cleanup;
841                         goto inserted;
842                 } else {
843                         int offset = (char *)s->here - bs->bh->b_data;
844
845                         unlock_buffer(bs->bh);
846                         ea_bdebug(bs->bh, "cloning");
847                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
848                         error = -ENOMEM;
849                         if (s->base == NULL)
850                                 goto cleanup;
851                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
852                         s->first = ENTRY(header(s->base)+1);
853                         header(s->base)->h_refcount = cpu_to_le32(1);
854                         s->here = ENTRY(s->base + offset);
855                         s->end = s->base + bs->bh->b_size;
856                 }
857         } else {
858                 /* Allocate a buffer where we construct the new block. */
859                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
860                 /* assert(header == s->base) */
861                 error = -ENOMEM;
862                 if (s->base == NULL)
863                         goto cleanup;
864                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
865                 header(s->base)->h_blocks = cpu_to_le32(1);
866                 header(s->base)->h_refcount = cpu_to_le32(1);
867                 s->first = ENTRY(header(s->base)+1);
868                 s->here = ENTRY(header(s->base)+1);
869                 s->end = s->base + sb->s_blocksize;
870         }
871
872         error = ext4_xattr_set_entry(i, s);
873         if (error == -EFSCORRUPTED)
874                 goto bad_block;
875         if (error)
876                 goto cleanup;
877         if (!IS_LAST_ENTRY(s->first))
878                 ext4_xattr_rehash(header(s->base), s->here);
879
880 inserted:
881         if (!IS_LAST_ENTRY(s->first)) {
882                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
883                 if (new_bh) {
884                         /* We found an identical block in the cache. */
885                         if (new_bh == bs->bh)
886                                 ea_bdebug(new_bh, "keeping");
887                         else {
888                                 u32 ref;
889
890                                 /* The old block is released after updating
891                                    the inode. */
892                                 error = dquot_alloc_block(inode,
893                                                 EXT4_C2B(EXT4_SB(sb), 1));
894                                 if (error)
895                                         goto cleanup;
896                                 BUFFER_TRACE(new_bh, "get_write_access");
897                                 error = ext4_journal_get_write_access(handle,
898                                                                       new_bh);
899                                 if (error)
900                                         goto cleanup_dquot;
901                                 lock_buffer(new_bh);
902                                 /*
903                                  * We have to be careful about races with
904                                  * freeing, rehashing or adding references to
905                                  * xattr block. Once we hold buffer lock xattr
906                                  * block's state is stable so we can check
907                                  * whether the block got freed / rehashed or
908                                  * not.  Since we unhash mbcache entry under
909                                  * buffer lock when freeing / rehashing xattr
910                                  * block, checking whether entry is still
911                                  * hashed is reliable. Same rules hold for
912                                  * e_reusable handling.
913                                  */
914                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
915                                     !ce->e_reusable) {
916                                         /*
917                                          * Undo everything and check mbcache
918                                          * again.
919                                          */
920                                         unlock_buffer(new_bh);
921                                         dquot_free_block(inode,
922                                                          EXT4_C2B(EXT4_SB(sb),
923                                                                   1));
924                                         brelse(new_bh);
925                                         mb_cache_entry_put(ext4_mb_cache, ce);
926                                         ce = NULL;
927                                         new_bh = NULL;
928                                         goto inserted;
929                                 }
930                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
931                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
932                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
933                                         ce->e_reusable = 0;
934                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
935                                           ref);
936                                 unlock_buffer(new_bh);
937                                 error = ext4_handle_dirty_xattr_block(handle,
938                                                                       inode,
939                                                                       new_bh);
940                                 if (error)
941                                         goto cleanup_dquot;
942                         }
943                         mb_cache_entry_touch(ext4_mb_cache, ce);
944                         mb_cache_entry_put(ext4_mb_cache, ce);
945                         ce = NULL;
946                 } else if (bs->bh && s->base == bs->bh->b_data) {
947                         /* We were modifying this block in-place. */
948                         ea_bdebug(bs->bh, "keeping this block");
949                         new_bh = bs->bh;
950                         get_bh(new_bh);
951                 } else {
952                         /* We need to allocate a new block */
953                         ext4_fsblk_t goal, block;
954
955                         goal = ext4_group_first_block_no(sb,
956                                                 EXT4_I(inode)->i_block_group);
957
958                         /* non-extent files can't have physical blocks past 2^32 */
959                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
960                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
961
962                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
963                                                      NULL, &error);
964                         if (error)
965                                 goto cleanup;
966
967                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
968                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
969
970                         ea_idebug(inode, "creating block %llu",
971                                   (unsigned long long)block);
972
973                         new_bh = sb_getblk(sb, block);
974                         if (unlikely(!new_bh)) {
975                                 error = -ENOMEM;
976 getblk_failed:
977                                 ext4_free_blocks(handle, inode, NULL, block, 1,
978                                                  EXT4_FREE_BLOCKS_METADATA);
979                                 goto cleanup;
980                         }
981                         lock_buffer(new_bh);
982                         error = ext4_journal_get_create_access(handle, new_bh);
983                         if (error) {
984                                 unlock_buffer(new_bh);
985                                 error = -EIO;
986                                 goto getblk_failed;
987                         }
988                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
989                         set_buffer_uptodate(new_bh);
990                         unlock_buffer(new_bh);
991                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
992                         error = ext4_handle_dirty_xattr_block(handle,
993                                                               inode, new_bh);
994                         if (error)
995                                 goto cleanup;
996                 }
997         }
998
999         /* Update the inode. */
1000         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
1001
1002         /* Drop the previous xattr block. */
1003         if (bs->bh && bs->bh != new_bh)
1004                 ext4_xattr_release_block(handle, inode, bs->bh);
1005         error = 0;
1006
1007 cleanup:
1008         if (ce)
1009                 mb_cache_entry_put(ext4_mb_cache, ce);
1010         brelse(new_bh);
1011         if (!(bs->bh && s->base == bs->bh->b_data))
1012                 kfree(s->base);
1013
1014         return error;
1015
1016 cleanup_dquot:
1017         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
1018         goto cleanup;
1019
1020 bad_block:
1021         EXT4_ERROR_INODE(inode, "bad block %llu",
1022                          EXT4_I(inode)->i_file_acl);
1023         goto cleanup;
1024
1025 #undef header
1026 }
1027
1028 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1029                           struct ext4_xattr_ibody_find *is)
1030 {
1031         struct ext4_xattr_ibody_header *header;
1032         struct ext4_inode *raw_inode;
1033         int error;
1034
1035         if (EXT4_I(inode)->i_extra_isize == 0)
1036                 return 0;
1037         raw_inode = ext4_raw_inode(&is->iloc);
1038         header = IHDR(inode, raw_inode);
1039         is->s.base = is->s.first = IFIRST(header);
1040         is->s.here = is->s.first;
1041         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1042         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1043                 error = ext4_xattr_check_names(IFIRST(header), is->s.end,
1044                                                IFIRST(header));
1045                 if (error)
1046                         return error;
1047                 /* Find the named attribute. */
1048                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1049                                               i->name, is->s.end -
1050                                               (void *)is->s.base, 0);
1051                 if (error && error != -ENODATA)
1052                         return error;
1053                 is->s.not_found = error;
1054         }
1055         return 0;
1056 }
1057
1058 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1059                                 struct ext4_xattr_info *i,
1060                                 struct ext4_xattr_ibody_find *is)
1061 {
1062         struct ext4_xattr_ibody_header *header;
1063         struct ext4_xattr_search *s = &is->s;
1064         int error;
1065
1066         if (EXT4_I(inode)->i_extra_isize == 0)
1067                 return -ENOSPC;
1068         error = ext4_xattr_set_entry(i, s);
1069         if (error) {
1070                 if (error == -ENOSPC &&
1071                     ext4_has_inline_data(inode)) {
1072                         error = ext4_try_to_evict_inline_data(handle, inode,
1073                                         EXT4_XATTR_LEN(strlen(i->name) +
1074                                         EXT4_XATTR_SIZE(i->value_len)));
1075                         if (error)
1076                                 return error;
1077                         error = ext4_xattr_ibody_find(inode, i, is);
1078                         if (error)
1079                                 return error;
1080                         error = ext4_xattr_set_entry(i, s);
1081                 }
1082                 if (error)
1083                         return error;
1084         }
1085         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1086         if (!IS_LAST_ENTRY(s->first)) {
1087                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1088                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1089         } else {
1090                 header->h_magic = cpu_to_le32(0);
1091                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1092         }
1093         return 0;
1094 }
1095
1096 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
1097                                 struct ext4_xattr_info *i,
1098                                 struct ext4_xattr_ibody_find *is)
1099 {
1100         struct ext4_xattr_ibody_header *header;
1101         struct ext4_xattr_search *s = &is->s;
1102         int error;
1103
1104         if (EXT4_I(inode)->i_extra_isize == 0)
1105                 return -ENOSPC;
1106         error = ext4_xattr_set_entry(i, s);
1107         if (error)
1108                 return error;
1109         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1110         if (!IS_LAST_ENTRY(s->first)) {
1111                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1112                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1113         } else {
1114                 header->h_magic = cpu_to_le32(0);
1115                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1116         }
1117         return 0;
1118 }
1119
1120 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
1121                                  struct ext4_xattr_info *i)
1122 {
1123         void *value;
1124
1125         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
1126                 return 0;
1127         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
1128         return !memcmp(value, i->value, i->value_len);
1129 }
1130
1131 /*
1132  * ext4_xattr_set_handle()
1133  *
1134  * Create, replace or remove an extended attribute for this inode.  Value
1135  * is NULL to remove an existing extended attribute, and non-NULL to
1136  * either replace an existing extended attribute, or create a new extended
1137  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1138  * specify that an extended attribute must exist and must not exist
1139  * previous to the call, respectively.
1140  *
1141  * Returns 0, or a negative error number on failure.
1142  */
1143 int
1144 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1145                       const char *name, const void *value, size_t value_len,
1146                       int flags)
1147 {
1148         struct ext4_xattr_info i = {
1149                 .name_index = name_index,
1150                 .name = name,
1151                 .value = value,
1152                 .value_len = value_len,
1153
1154         };
1155         struct ext4_xattr_ibody_find is = {
1156                 .s = { .not_found = -ENODATA, },
1157         };
1158         struct ext4_xattr_block_find bs = {
1159                 .s = { .not_found = -ENODATA, },
1160         };
1161         unsigned long no_expand;
1162         int error;
1163
1164         if (!name)
1165                 return -EINVAL;
1166         if (strlen(name) > 255)
1167                 return -ERANGE;
1168         down_write(&EXT4_I(inode)->xattr_sem);
1169         no_expand = ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND);
1170         ext4_set_inode_state(inode, EXT4_STATE_NO_EXPAND);
1171
1172         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1173         if (error)
1174                 goto cleanup;
1175
1176         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1177                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1178                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1179                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1180         }
1181
1182         error = ext4_xattr_ibody_find(inode, &i, &is);
1183         if (error)
1184                 goto cleanup;
1185         if (is.s.not_found)
1186                 error = ext4_xattr_block_find(inode, &i, &bs);
1187         if (error)
1188                 goto cleanup;
1189         if (is.s.not_found && bs.s.not_found) {
1190                 error = -ENODATA;
1191                 if (flags & XATTR_REPLACE)
1192                         goto cleanup;
1193                 error = 0;
1194                 if (!value)
1195                         goto cleanup;
1196         } else {
1197                 error = -EEXIST;
1198                 if (flags & XATTR_CREATE)
1199                         goto cleanup;
1200         }
1201         if (!value) {
1202                 if (!is.s.not_found)
1203                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1204                 else if (!bs.s.not_found)
1205                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1206         } else {
1207                 error = 0;
1208                 /* Xattr value did not change? Save us some work and bail out */
1209                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
1210                         goto cleanup;
1211                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
1212                         goto cleanup;
1213
1214                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1215                 if (!error && !bs.s.not_found) {
1216                         i.value = NULL;
1217                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1218                 } else if (error == -ENOSPC) {
1219                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1220                                 error = ext4_xattr_block_find(inode, &i, &bs);
1221                                 if (error)
1222                                         goto cleanup;
1223                         }
1224                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1225                         if (error)
1226                                 goto cleanup;
1227                         if (!is.s.not_found) {
1228                                 i.value = NULL;
1229                                 error = ext4_xattr_ibody_set(handle, inode, &i,
1230                                                              &is);
1231                         }
1232                 }
1233         }
1234         if (!error) {
1235                 ext4_xattr_update_super_block(handle, inode->i_sb);
1236                 inode->i_ctime = ext4_current_time(inode);
1237                 if (!value)
1238                         ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1239                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1240                 /*
1241                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1242                  * error != 0.
1243                  */
1244                 is.iloc.bh = NULL;
1245                 if (IS_SYNC(inode))
1246                         ext4_handle_sync(handle);
1247         }
1248
1249 cleanup:
1250         brelse(is.iloc.bh);
1251         brelse(bs.bh);
1252         if (no_expand == 0)
1253                 ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1254         up_write(&EXT4_I(inode)->xattr_sem);
1255         return error;
1256 }
1257
1258 /*
1259  * ext4_xattr_set()
1260  *
1261  * Like ext4_xattr_set_handle, but start from an inode. This extended
1262  * attribute modification is a filesystem transaction by itself.
1263  *
1264  * Returns 0, or a negative error number on failure.
1265  */
1266 int
1267 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1268                const void *value, size_t value_len, int flags)
1269 {
1270         handle_t *handle;
1271         int error, retries = 0;
1272         int credits = ext4_jbd2_credits_xattr(inode);
1273
1274 retry:
1275         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1276         if (IS_ERR(handle)) {
1277                 error = PTR_ERR(handle);
1278         } else {
1279                 int error2;
1280
1281                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1282                                               value, value_len, flags);
1283                 error2 = ext4_journal_stop(handle);
1284                 if (error == -ENOSPC &&
1285                     ext4_should_retry_alloc(inode->i_sb, &retries))
1286                         goto retry;
1287                 if (error == 0)
1288                         error = error2;
1289         }
1290
1291         return error;
1292 }
1293
1294 /*
1295  * Shift the EA entries in the inode to create space for the increased
1296  * i_extra_isize.
1297  */
1298 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1299                                      int value_offs_shift, void *to,
1300                                      void *from, size_t n, int blocksize)
1301 {
1302         struct ext4_xattr_entry *last = entry;
1303         int new_offs;
1304
1305         /* Adjust the value offsets of the entries */
1306         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1307                 if (!last->e_value_block && last->e_value_size) {
1308                         new_offs = le16_to_cpu(last->e_value_offs) +
1309                                                         value_offs_shift;
1310                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
1311                                  > blocksize);
1312                         last->e_value_offs = cpu_to_le16(new_offs);
1313                 }
1314         }
1315         /* Shift the entries by n bytes */
1316         memmove(to, from, n);
1317 }
1318
1319 /*
1320  * Expand an inode by new_extra_isize bytes when EAs are present.
1321  * Returns 0 on success or negative error number on failure.
1322  */
1323 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1324                                struct ext4_inode *raw_inode, handle_t *handle)
1325 {
1326         struct ext4_xattr_ibody_header *header;
1327         struct ext4_xattr_entry *entry, *last, *first;
1328         struct buffer_head *bh = NULL;
1329         struct ext4_xattr_ibody_find *is = NULL;
1330         struct ext4_xattr_block_find *bs = NULL;
1331         char *buffer = NULL, *b_entry_name = NULL;
1332         size_t min_offs, free;
1333         int total_ino;
1334         void *base, *start, *end;
1335         int extra_isize = 0, error = 0, tried_min_extra_isize = 0;
1336         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1337
1338         down_write(&EXT4_I(inode)->xattr_sem);
1339 retry:
1340         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) {
1341                 up_write(&EXT4_I(inode)->xattr_sem);
1342                 return 0;
1343         }
1344
1345         header = IHDR(inode, raw_inode);
1346         entry = IFIRST(header);
1347
1348         /*
1349          * Check if enough free space is available in the inode to shift the
1350          * entries ahead by new_extra_isize.
1351          */
1352
1353         base = start = entry;
1354         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1355         min_offs = end - base;
1356         last = entry;
1357         total_ino = sizeof(struct ext4_xattr_ibody_header);
1358
1359         free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
1360         if (free >= new_extra_isize) {
1361                 entry = IFIRST(header);
1362                 ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize
1363                                 - new_extra_isize, (void *)raw_inode +
1364                                 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1365                                 (void *)header, total_ino,
1366                                 inode->i_sb->s_blocksize);
1367                 EXT4_I(inode)->i_extra_isize = new_extra_isize;
1368                 error = 0;
1369                 goto cleanup;
1370         }
1371
1372         /*
1373          * Enough free space isn't available in the inode, check if
1374          * EA block can hold new_extra_isize bytes.
1375          */
1376         if (EXT4_I(inode)->i_file_acl) {
1377                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1378                 error = -EIO;
1379                 if (!bh)
1380                         goto cleanup;
1381                 if (ext4_xattr_check_block(inode, bh)) {
1382                         EXT4_ERROR_INODE(inode, "bad block %llu",
1383                                          EXT4_I(inode)->i_file_acl);
1384                         error = -EFSCORRUPTED;
1385                         goto cleanup;
1386                 }
1387                 base = BHDR(bh);
1388                 first = BFIRST(bh);
1389                 end = bh->b_data + bh->b_size;
1390                 min_offs = end - base;
1391                 free = ext4_xattr_free_space(first, &min_offs, base, NULL);
1392                 if (free < new_extra_isize) {
1393                         if (!tried_min_extra_isize && s_min_extra_isize) {
1394                                 tried_min_extra_isize++;
1395                                 new_extra_isize = s_min_extra_isize;
1396                                 brelse(bh);
1397                                 goto retry;
1398                         }
1399                         error = -1;
1400                         goto cleanup;
1401                 }
1402         } else {
1403                 free = inode->i_sb->s_blocksize;
1404         }
1405
1406         while (new_extra_isize > 0) {
1407                 size_t offs, size, entry_size;
1408                 struct ext4_xattr_entry *small_entry = NULL;
1409                 struct ext4_xattr_info i = {
1410                         .value = NULL,
1411                         .value_len = 0,
1412                 };
1413                 unsigned int total_size;  /* EA entry size + value size */
1414                 unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
1415                 unsigned int min_total_size = ~0U;
1416
1417                 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1418                 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1419                 if (!is || !bs) {
1420                         error = -ENOMEM;
1421                         goto cleanup;
1422                 }
1423
1424                 is->s.not_found = -ENODATA;
1425                 bs->s.not_found = -ENODATA;
1426                 is->iloc.bh = NULL;
1427                 bs->bh = NULL;
1428
1429                 last = IFIRST(header);
1430                 /* Find the entry best suited to be pushed into EA block */
1431                 entry = NULL;
1432                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1433                         total_size =
1434                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1435                                         EXT4_XATTR_LEN(last->e_name_len);
1436                         if (total_size <= free && total_size < min_total_size) {
1437                                 if (total_size < new_extra_isize) {
1438                                         small_entry = last;
1439                                 } else {
1440                                         entry = last;
1441                                         min_total_size = total_size;
1442                                 }
1443                         }
1444                 }
1445
1446                 if (entry == NULL) {
1447                         if (small_entry) {
1448                                 entry = small_entry;
1449                         } else {
1450                                 if (!tried_min_extra_isize &&
1451                                     s_min_extra_isize) {
1452                                         tried_min_extra_isize++;
1453                                         new_extra_isize = s_min_extra_isize;
1454                                         kfree(is); is = NULL;
1455                                         kfree(bs); bs = NULL;
1456                                         brelse(bh);
1457                                         goto retry;
1458                                 }
1459                                 error = -1;
1460                                 goto cleanup;
1461                         }
1462                 }
1463                 offs = le16_to_cpu(entry->e_value_offs);
1464                 size = le32_to_cpu(entry->e_value_size);
1465                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1466                 i.name_index = entry->e_name_index,
1467                 buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
1468                 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1469                 if (!buffer || !b_entry_name) {
1470                         error = -ENOMEM;
1471                         goto cleanup;
1472                 }
1473                 /* Save the entry name and the entry value */
1474                 memcpy(buffer, (void *)IFIRST(header) + offs,
1475                        EXT4_XATTR_SIZE(size));
1476                 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1477                 b_entry_name[entry->e_name_len] = '\0';
1478                 i.name = b_entry_name;
1479
1480                 error = ext4_get_inode_loc(inode, &is->iloc);
1481                 if (error)
1482                         goto cleanup;
1483
1484                 error = ext4_xattr_ibody_find(inode, &i, is);
1485                 if (error)
1486                         goto cleanup;
1487
1488                 /* Remove the chosen entry from the inode */
1489                 error = ext4_xattr_ibody_set(handle, inode, &i, is);
1490                 if (error)
1491                         goto cleanup;
1492
1493                 entry = IFIRST(header);
1494                 if (entry_size + EXT4_XATTR_SIZE(size) >= new_extra_isize)
1495                         shift_bytes = new_extra_isize;
1496                 else
1497                         shift_bytes = entry_size + size;
1498                 /* Adjust the offsets and shift the remaining entries ahead */
1499                 ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize -
1500                         shift_bytes, (void *)raw_inode +
1501                         EXT4_GOOD_OLD_INODE_SIZE + extra_isize + shift_bytes,
1502                         (void *)header, total_ino - entry_size,
1503                         inode->i_sb->s_blocksize);
1504
1505                 extra_isize += shift_bytes;
1506                 new_extra_isize -= shift_bytes;
1507                 EXT4_I(inode)->i_extra_isize = extra_isize;
1508
1509                 i.name = b_entry_name;
1510                 i.value = buffer;
1511                 i.value_len = size;
1512                 error = ext4_xattr_block_find(inode, &i, bs);
1513                 if (error)
1514                         goto cleanup;
1515
1516                 /* Add entry which was removed from the inode into the block */
1517                 error = ext4_xattr_block_set(handle, inode, &i, bs);
1518                 if (error)
1519                         goto cleanup;
1520                 kfree(b_entry_name);
1521                 kfree(buffer);
1522                 b_entry_name = NULL;
1523                 buffer = NULL;
1524                 brelse(is->iloc.bh);
1525                 kfree(is);
1526                 kfree(bs);
1527         }
1528         brelse(bh);
1529         up_write(&EXT4_I(inode)->xattr_sem);
1530         return 0;
1531
1532 cleanup:
1533         kfree(b_entry_name);
1534         kfree(buffer);
1535         if (is)
1536                 brelse(is->iloc.bh);
1537         kfree(is);
1538         kfree(bs);
1539         brelse(bh);
1540         up_write(&EXT4_I(inode)->xattr_sem);
1541         return error;
1542 }
1543
1544
1545
1546 /*
1547  * ext4_xattr_delete_inode()
1548  *
1549  * Free extended attribute resources associated with this inode. This
1550  * is called immediately before an inode is freed. We have exclusive
1551  * access to the inode.
1552  */
1553 void
1554 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1555 {
1556         struct buffer_head *bh = NULL;
1557
1558         if (!EXT4_I(inode)->i_file_acl)
1559                 goto cleanup;
1560         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1561         if (!bh) {
1562                 EXT4_ERROR_INODE(inode, "block %llu read error",
1563                                  EXT4_I(inode)->i_file_acl);
1564                 goto cleanup;
1565         }
1566         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1567             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1568                 EXT4_ERROR_INODE(inode, "bad block %llu",
1569                                  EXT4_I(inode)->i_file_acl);
1570                 goto cleanup;
1571         }
1572         ext4_xattr_release_block(handle, inode, bh);
1573         EXT4_I(inode)->i_file_acl = 0;
1574
1575 cleanup:
1576         brelse(bh);
1577 }
1578
1579 /*
1580  * ext4_xattr_cache_insert()
1581  *
1582  * Create a new entry in the extended attribute cache, and insert
1583  * it unless such an entry is already in the cache.
1584  *
1585  * Returns 0, or a negative error number on failure.
1586  */
1587 static void
1588 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1589 {
1590         struct ext4_xattr_header *header = BHDR(bh);
1591         __u32 hash = le32_to_cpu(header->h_hash);
1592         int reusable = le32_to_cpu(header->h_refcount) <
1593                        EXT4_XATTR_REFCOUNT_MAX;
1594         int error;
1595
1596         error = mb_cache_entry_create(ext4_mb_cache, GFP_NOFS, hash,
1597                                       bh->b_blocknr, reusable);
1598         if (error) {
1599                 if (error == -EBUSY)
1600                         ea_bdebug(bh, "already in cache");
1601         } else
1602                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1603 }
1604
1605 /*
1606  * ext4_xattr_cmp()
1607  *
1608  * Compare two extended attribute blocks for equality.
1609  *
1610  * Returns 0 if the blocks are equal, 1 if they differ, and
1611  * a negative error number on errors.
1612  */
1613 static int
1614 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1615                struct ext4_xattr_header *header2)
1616 {
1617         struct ext4_xattr_entry *entry1, *entry2;
1618
1619         entry1 = ENTRY(header1+1);
1620         entry2 = ENTRY(header2+1);
1621         while (!IS_LAST_ENTRY(entry1)) {
1622                 if (IS_LAST_ENTRY(entry2))
1623                         return 1;
1624                 if (entry1->e_hash != entry2->e_hash ||
1625                     entry1->e_name_index != entry2->e_name_index ||
1626                     entry1->e_name_len != entry2->e_name_len ||
1627                     entry1->e_value_size != entry2->e_value_size ||
1628                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1629                         return 1;
1630                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1631                         return -EFSCORRUPTED;
1632                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1633                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1634                            le32_to_cpu(entry1->e_value_size)))
1635                         return 1;
1636
1637                 entry1 = EXT4_XATTR_NEXT(entry1);
1638                 entry2 = EXT4_XATTR_NEXT(entry2);
1639         }
1640         if (!IS_LAST_ENTRY(entry2))
1641                 return 1;
1642         return 0;
1643 }
1644
1645 /*
1646  * ext4_xattr_cache_find()
1647  *
1648  * Find an identical extended attribute block.
1649  *
1650  * Returns a pointer to the block found, or NULL if such a block was
1651  * not found or an error occurred.
1652  */
1653 static struct buffer_head *
1654 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1655                       struct mb_cache_entry **pce)
1656 {
1657         __u32 hash = le32_to_cpu(header->h_hash);
1658         struct mb_cache_entry *ce;
1659         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1660
1661         if (!header->h_hash)
1662                 return NULL;  /* never share */
1663         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1664         ce = mb_cache_entry_find_first(ext4_mb_cache, hash);
1665         while (ce) {
1666                 struct buffer_head *bh;
1667
1668                 bh = sb_bread(inode->i_sb, ce->e_block);
1669                 if (!bh) {
1670                         EXT4_ERROR_INODE(inode, "block %lu read error",
1671                                          (unsigned long) ce->e_block);
1672                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1673                         *pce = ce;
1674                         return bh;
1675                 }
1676                 brelse(bh);
1677                 ce = mb_cache_entry_find_next(ext4_mb_cache, ce);
1678         }
1679         return NULL;
1680 }
1681
1682 #define NAME_HASH_SHIFT 5
1683 #define VALUE_HASH_SHIFT 16
1684
1685 /*
1686  * ext4_xattr_hash_entry()
1687  *
1688  * Compute the hash of an extended attribute.
1689  */
1690 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1691                                          struct ext4_xattr_entry *entry)
1692 {
1693         __u32 hash = 0;
1694         char *name = entry->e_name;
1695         int n;
1696
1697         for (n = 0; n < entry->e_name_len; n++) {
1698                 hash = (hash << NAME_HASH_SHIFT) ^
1699                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1700                        *name++;
1701         }
1702
1703         if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1704                 __le32 *value = (__le32 *)((char *)header +
1705                         le16_to_cpu(entry->e_value_offs));
1706                 for (n = (le32_to_cpu(entry->e_value_size) +
1707                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1708                         hash = (hash << VALUE_HASH_SHIFT) ^
1709                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1710                                le32_to_cpu(*value++);
1711                 }
1712         }
1713         entry->e_hash = cpu_to_le32(hash);
1714 }
1715
1716 #undef NAME_HASH_SHIFT
1717 #undef VALUE_HASH_SHIFT
1718
1719 #define BLOCK_HASH_SHIFT 16
1720
1721 /*
1722  * ext4_xattr_rehash()
1723  *
1724  * Re-compute the extended attribute hash value after an entry has changed.
1725  */
1726 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1727                               struct ext4_xattr_entry *entry)
1728 {
1729         struct ext4_xattr_entry *here;
1730         __u32 hash = 0;
1731
1732         ext4_xattr_hash_entry(header, entry);
1733         here = ENTRY(header+1);
1734         while (!IS_LAST_ENTRY(here)) {
1735                 if (!here->e_hash) {
1736                         /* Block is not shared if an entry's hash value == 0 */
1737                         hash = 0;
1738                         break;
1739                 }
1740                 hash = (hash << BLOCK_HASH_SHIFT) ^
1741                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1742                        le32_to_cpu(here->e_hash);
1743                 here = EXT4_XATTR_NEXT(here);
1744         }
1745         header->h_hash = cpu_to_le32(hash);
1746 }
1747
1748 #undef BLOCK_HASH_SHIFT
1749
1750 #define HASH_BUCKET_BITS        10
1751
1752 struct mb_cache *
1753 ext4_xattr_create_cache(void)
1754 {
1755         return mb_cache_create(HASH_BUCKET_BITS);
1756 }
1757
1758 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1759 {
1760         if (cache)
1761                 mb_cache_destroy(cache);
1762 }
1763