net/mlx5e: Implement DCBNL IEEE max rate
[cascardo/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_entry *ce = NULL;
549         int error = 0;
550         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
551
552         ce = mb_cache_entry_get(ext4_mb_cache, bh->b_bdev, bh->b_blocknr);
553         BUFFER_TRACE(bh, "get_write_access");
554         error = ext4_journal_get_write_access(handle, bh);
555         if (error)
556                 goto out;
557
558         lock_buffer(bh);
559         if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
560                 ea_bdebug(bh, "refcount now=0; freeing");
561                 if (ce)
562                         mb_cache_entry_free(ce);
563                 get_bh(bh);
564                 unlock_buffer(bh);
565                 ext4_free_blocks(handle, inode, bh, 0, 1,
566                                  EXT4_FREE_BLOCKS_METADATA |
567                                  EXT4_FREE_BLOCKS_FORGET);
568         } else {
569                 le32_add_cpu(&BHDR(bh)->h_refcount, -1);
570                 if (ce)
571                         mb_cache_entry_release(ce);
572                 /*
573                  * Beware of this ugliness: Releasing of xattr block references
574                  * from different inodes can race and so we have to protect
575                  * from a race where someone else frees the block (and releases
576                  * its journal_head) before we are done dirtying the buffer. In
577                  * nojournal mode this race is harmless and we actually cannot
578                  * call ext4_handle_dirty_xattr_block() with locked buffer as
579                  * that function can call sync_dirty_buffer() so for that case
580                  * we handle the dirtying after unlocking the buffer.
581                  */
582                 if (ext4_handle_valid(handle))
583                         error = ext4_handle_dirty_xattr_block(handle, inode,
584                                                               bh);
585                 unlock_buffer(bh);
586                 if (!ext4_handle_valid(handle))
587                         error = ext4_handle_dirty_xattr_block(handle, inode,
588                                                               bh);
589                 if (IS_SYNC(inode))
590                         ext4_handle_sync(handle);
591                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
592                 ea_bdebug(bh, "refcount now=%d; releasing",
593                           le32_to_cpu(BHDR(bh)->h_refcount));
594         }
595 out:
596         ext4_std_error(inode->i_sb, error);
597         return;
598 }
599
600 /*
601  * Find the available free space for EAs. This also returns the total number of
602  * bytes used by EA entries.
603  */
604 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
605                                     size_t *min_offs, void *base, int *total)
606 {
607         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
608                 if (!last->e_value_block && last->e_value_size) {
609                         size_t offs = le16_to_cpu(last->e_value_offs);
610                         if (offs < *min_offs)
611                                 *min_offs = offs;
612                 }
613                 if (total)
614                         *total += EXT4_XATTR_LEN(last->e_name_len);
615         }
616         return (*min_offs - ((void *)last - base) - sizeof(__u32));
617 }
618
619 static int
620 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
621 {
622         struct ext4_xattr_entry *last;
623         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
624
625         /* Compute min_offs and last. */
626         last = s->first;
627         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
628                 if (!last->e_value_block && last->e_value_size) {
629                         size_t offs = le16_to_cpu(last->e_value_offs);
630                         if (offs < min_offs)
631                                 min_offs = offs;
632                 }
633         }
634         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
635         if (!s->not_found) {
636                 if (!s->here->e_value_block && s->here->e_value_size) {
637                         size_t size = le32_to_cpu(s->here->e_value_size);
638                         free += EXT4_XATTR_SIZE(size);
639                 }
640                 free += EXT4_XATTR_LEN(name_len);
641         }
642         if (i->value) {
643                 if (free < EXT4_XATTR_LEN(name_len) +
644                            EXT4_XATTR_SIZE(i->value_len))
645                         return -ENOSPC;
646         }
647
648         if (i->value && s->not_found) {
649                 /* Insert the new name. */
650                 size_t size = EXT4_XATTR_LEN(name_len);
651                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
652                 memmove((void *)s->here + size, s->here, rest);
653                 memset(s->here, 0, size);
654                 s->here->e_name_index = i->name_index;
655                 s->here->e_name_len = name_len;
656                 memcpy(s->here->e_name, i->name, name_len);
657         } else {
658                 if (!s->here->e_value_block && s->here->e_value_size) {
659                         void *first_val = s->base + min_offs;
660                         size_t offs = le16_to_cpu(s->here->e_value_offs);
661                         void *val = s->base + offs;
662                         size_t size = EXT4_XATTR_SIZE(
663                                 le32_to_cpu(s->here->e_value_size));
664
665                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
666                                 /* The old and the new value have the same
667                                    size. Just replace. */
668                                 s->here->e_value_size =
669                                         cpu_to_le32(i->value_len);
670                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
671                                         memset(val, 0, size);
672                                 } else {
673                                         /* Clear pad bytes first. */
674                                         memset(val + size - EXT4_XATTR_PAD, 0,
675                                                EXT4_XATTR_PAD);
676                                         memcpy(val, i->value, i->value_len);
677                                 }
678                                 return 0;
679                         }
680
681                         /* Remove the old value. */
682                         memmove(first_val + size, first_val, val - first_val);
683                         memset(first_val, 0, size);
684                         s->here->e_value_size = 0;
685                         s->here->e_value_offs = 0;
686                         min_offs += size;
687
688                         /* Adjust all value offsets. */
689                         last = s->first;
690                         while (!IS_LAST_ENTRY(last)) {
691                                 size_t o = le16_to_cpu(last->e_value_offs);
692                                 if (!last->e_value_block &&
693                                     last->e_value_size && o < offs)
694                                         last->e_value_offs =
695                                                 cpu_to_le16(o + size);
696                                 last = EXT4_XATTR_NEXT(last);
697                         }
698                 }
699                 if (!i->value) {
700                         /* Remove the old name. */
701                         size_t size = EXT4_XATTR_LEN(name_len);
702                         last = ENTRY((void *)last - size);
703                         memmove(s->here, (void *)s->here + size,
704                                 (void *)last - (void *)s->here + sizeof(__u32));
705                         memset(last, 0, size);
706                 }
707         }
708
709         if (i->value) {
710                 /* Insert the new value. */
711                 s->here->e_value_size = cpu_to_le32(i->value_len);
712                 if (i->value_len) {
713                         size_t size = EXT4_XATTR_SIZE(i->value_len);
714                         void *val = s->base + min_offs - size;
715                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
716                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
717                                 memset(val, 0, size);
718                         } else {
719                                 /* Clear the pad bytes first. */
720                                 memset(val + size - EXT4_XATTR_PAD, 0,
721                                        EXT4_XATTR_PAD);
722                                 memcpy(val, i->value, i->value_len);
723                         }
724                 }
725         }
726         return 0;
727 }
728
729 struct ext4_xattr_block_find {
730         struct ext4_xattr_search s;
731         struct buffer_head *bh;
732 };
733
734 static int
735 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
736                       struct ext4_xattr_block_find *bs)
737 {
738         struct super_block *sb = inode->i_sb;
739         int error;
740
741         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
742                   i->name_index, i->name, i->value, (long)i->value_len);
743
744         if (EXT4_I(inode)->i_file_acl) {
745                 /* The inode already has an extended attribute block. */
746                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
747                 error = -EIO;
748                 if (!bs->bh)
749                         goto cleanup;
750                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
751                         atomic_read(&(bs->bh->b_count)),
752                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
753                 if (ext4_xattr_check_block(inode, bs->bh)) {
754                         EXT4_ERROR_INODE(inode, "bad block %llu",
755                                          EXT4_I(inode)->i_file_acl);
756                         error = -EFSCORRUPTED;
757                         goto cleanup;
758                 }
759                 /* Find the named attribute. */
760                 bs->s.base = BHDR(bs->bh);
761                 bs->s.first = BFIRST(bs->bh);
762                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
763                 bs->s.here = bs->s.first;
764                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
765                                               i->name, bs->bh->b_size, 1);
766                 if (error && error != -ENODATA)
767                         goto cleanup;
768                 bs->s.not_found = error;
769         }
770         error = 0;
771
772 cleanup:
773         return error;
774 }
775
776 static int
777 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
778                      struct ext4_xattr_info *i,
779                      struct ext4_xattr_block_find *bs)
780 {
781         struct super_block *sb = inode->i_sb;
782         struct buffer_head *new_bh = NULL;
783         struct ext4_xattr_search *s = &bs->s;
784         struct mb_cache_entry *ce = NULL;
785         int error = 0;
786         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
787
788 #define header(x) ((struct ext4_xattr_header *)(x))
789
790         if (i->value && i->value_len > sb->s_blocksize)
791                 return -ENOSPC;
792         if (s->base) {
793                 ce = mb_cache_entry_get(ext4_mb_cache, bs->bh->b_bdev,
794                                         bs->bh->b_blocknr);
795                 BUFFER_TRACE(bs->bh, "get_write_access");
796                 error = ext4_journal_get_write_access(handle, bs->bh);
797                 if (error)
798                         goto cleanup;
799                 lock_buffer(bs->bh);
800
801                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
802                         if (ce) {
803                                 mb_cache_entry_free(ce);
804                                 ce = NULL;
805                         }
806                         ea_bdebug(bs->bh, "modifying in-place");
807                         error = ext4_xattr_set_entry(i, s);
808                         if (!error) {
809                                 if (!IS_LAST_ENTRY(s->first))
810                                         ext4_xattr_rehash(header(s->base),
811                                                           s->here);
812                                 ext4_xattr_cache_insert(ext4_mb_cache,
813                                         bs->bh);
814                         }
815                         unlock_buffer(bs->bh);
816                         if (error == -EFSCORRUPTED)
817                                 goto bad_block;
818                         if (!error)
819                                 error = ext4_handle_dirty_xattr_block(handle,
820                                                                       inode,
821                                                                       bs->bh);
822                         if (error)
823                                 goto cleanup;
824                         goto inserted;
825                 } else {
826                         int offset = (char *)s->here - bs->bh->b_data;
827
828                         unlock_buffer(bs->bh);
829                         if (ce) {
830                                 mb_cache_entry_release(ce);
831                                 ce = NULL;
832                         }
833                         ea_bdebug(bs->bh, "cloning");
834                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
835                         error = -ENOMEM;
836                         if (s->base == NULL)
837                                 goto cleanup;
838                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
839                         s->first = ENTRY(header(s->base)+1);
840                         header(s->base)->h_refcount = cpu_to_le32(1);
841                         s->here = ENTRY(s->base + offset);
842                         s->end = s->base + bs->bh->b_size;
843                 }
844         } else {
845                 /* Allocate a buffer where we construct the new block. */
846                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
847                 /* assert(header == s->base) */
848                 error = -ENOMEM;
849                 if (s->base == NULL)
850                         goto cleanup;
851                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
852                 header(s->base)->h_blocks = cpu_to_le32(1);
853                 header(s->base)->h_refcount = cpu_to_le32(1);
854                 s->first = ENTRY(header(s->base)+1);
855                 s->here = ENTRY(header(s->base)+1);
856                 s->end = s->base + sb->s_blocksize;
857         }
858
859         error = ext4_xattr_set_entry(i, s);
860         if (error == -EFSCORRUPTED)
861                 goto bad_block;
862         if (error)
863                 goto cleanup;
864         if (!IS_LAST_ENTRY(s->first))
865                 ext4_xattr_rehash(header(s->base), s->here);
866
867 inserted:
868         if (!IS_LAST_ENTRY(s->first)) {
869                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
870                 if (new_bh) {
871                         /* We found an identical block in the cache. */
872                         if (new_bh == bs->bh)
873                                 ea_bdebug(new_bh, "keeping");
874                         else {
875                                 /* The old block is released after updating
876                                    the inode. */
877                                 error = dquot_alloc_block(inode,
878                                                 EXT4_C2B(EXT4_SB(sb), 1));
879                                 if (error)
880                                         goto cleanup;
881                                 BUFFER_TRACE(new_bh, "get_write_access");
882                                 error = ext4_journal_get_write_access(handle,
883                                                                       new_bh);
884                                 if (error)
885                                         goto cleanup_dquot;
886                                 lock_buffer(new_bh);
887                                 le32_add_cpu(&BHDR(new_bh)->h_refcount, 1);
888                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
889                                         le32_to_cpu(BHDR(new_bh)->h_refcount));
890                                 unlock_buffer(new_bh);
891                                 error = ext4_handle_dirty_xattr_block(handle,
892                                                                       inode,
893                                                                       new_bh);
894                                 if (error)
895                                         goto cleanup_dquot;
896                         }
897                         mb_cache_entry_release(ce);
898                         ce = NULL;
899                 } else if (bs->bh && s->base == bs->bh->b_data) {
900                         /* We were modifying this block in-place. */
901                         ea_bdebug(bs->bh, "keeping this block");
902                         new_bh = bs->bh;
903                         get_bh(new_bh);
904                 } else {
905                         /* We need to allocate a new block */
906                         ext4_fsblk_t goal, block;
907
908                         goal = ext4_group_first_block_no(sb,
909                                                 EXT4_I(inode)->i_block_group);
910
911                         /* non-extent files can't have physical blocks past 2^32 */
912                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
913                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
914
915                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
916                                                      NULL, &error);
917                         if (error)
918                                 goto cleanup;
919
920                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
921                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
922
923                         ea_idebug(inode, "creating block %llu",
924                                   (unsigned long long)block);
925
926                         new_bh = sb_getblk(sb, block);
927                         if (unlikely(!new_bh)) {
928                                 error = -ENOMEM;
929 getblk_failed:
930                                 ext4_free_blocks(handle, inode, NULL, block, 1,
931                                                  EXT4_FREE_BLOCKS_METADATA);
932                                 goto cleanup;
933                         }
934                         lock_buffer(new_bh);
935                         error = ext4_journal_get_create_access(handle, new_bh);
936                         if (error) {
937                                 unlock_buffer(new_bh);
938                                 error = -EIO;
939                                 goto getblk_failed;
940                         }
941                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
942                         set_buffer_uptodate(new_bh);
943                         unlock_buffer(new_bh);
944                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
945                         error = ext4_handle_dirty_xattr_block(handle,
946                                                               inode, new_bh);
947                         if (error)
948                                 goto cleanup;
949                 }
950         }
951
952         /* Update the inode. */
953         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
954
955         /* Drop the previous xattr block. */
956         if (bs->bh && bs->bh != new_bh)
957                 ext4_xattr_release_block(handle, inode, bs->bh);
958         error = 0;
959
960 cleanup:
961         if (ce)
962                 mb_cache_entry_release(ce);
963         brelse(new_bh);
964         if (!(bs->bh && s->base == bs->bh->b_data))
965                 kfree(s->base);
966
967         return error;
968
969 cleanup_dquot:
970         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
971         goto cleanup;
972
973 bad_block:
974         EXT4_ERROR_INODE(inode, "bad block %llu",
975                          EXT4_I(inode)->i_file_acl);
976         goto cleanup;
977
978 #undef header
979 }
980
981 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
982                           struct ext4_xattr_ibody_find *is)
983 {
984         struct ext4_xattr_ibody_header *header;
985         struct ext4_inode *raw_inode;
986         int error;
987
988         if (EXT4_I(inode)->i_extra_isize == 0)
989                 return 0;
990         raw_inode = ext4_raw_inode(&is->iloc);
991         header = IHDR(inode, raw_inode);
992         is->s.base = is->s.first = IFIRST(header);
993         is->s.here = is->s.first;
994         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
995         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
996                 error = ext4_xattr_check_names(IFIRST(header), is->s.end,
997                                                IFIRST(header));
998                 if (error)
999                         return error;
1000                 /* Find the named attribute. */
1001                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1002                                               i->name, is->s.end -
1003                                               (void *)is->s.base, 0);
1004                 if (error && error != -ENODATA)
1005                         return error;
1006                 is->s.not_found = error;
1007         }
1008         return 0;
1009 }
1010
1011 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1012                                 struct ext4_xattr_info *i,
1013                                 struct ext4_xattr_ibody_find *is)
1014 {
1015         struct ext4_xattr_ibody_header *header;
1016         struct ext4_xattr_search *s = &is->s;
1017         int error;
1018
1019         if (EXT4_I(inode)->i_extra_isize == 0)
1020                 return -ENOSPC;
1021         error = ext4_xattr_set_entry(i, s);
1022         if (error) {
1023                 if (error == -ENOSPC &&
1024                     ext4_has_inline_data(inode)) {
1025                         error = ext4_try_to_evict_inline_data(handle, inode,
1026                                         EXT4_XATTR_LEN(strlen(i->name) +
1027                                         EXT4_XATTR_SIZE(i->value_len)));
1028                         if (error)
1029                                 return error;
1030                         error = ext4_xattr_ibody_find(inode, i, is);
1031                         if (error)
1032                                 return error;
1033                         error = ext4_xattr_set_entry(i, s);
1034                 }
1035                 if (error)
1036                         return error;
1037         }
1038         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1039         if (!IS_LAST_ENTRY(s->first)) {
1040                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1041                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1042         } else {
1043                 header->h_magic = cpu_to_le32(0);
1044                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1045         }
1046         return 0;
1047 }
1048
1049 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
1050                                 struct ext4_xattr_info *i,
1051                                 struct ext4_xattr_ibody_find *is)
1052 {
1053         struct ext4_xattr_ibody_header *header;
1054         struct ext4_xattr_search *s = &is->s;
1055         int error;
1056
1057         if (EXT4_I(inode)->i_extra_isize == 0)
1058                 return -ENOSPC;
1059         error = ext4_xattr_set_entry(i, s);
1060         if (error)
1061                 return error;
1062         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1063         if (!IS_LAST_ENTRY(s->first)) {
1064                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1065                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1066         } else {
1067                 header->h_magic = cpu_to_le32(0);
1068                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1069         }
1070         return 0;
1071 }
1072
1073 /*
1074  * ext4_xattr_set_handle()
1075  *
1076  * Create, replace or remove an extended attribute for this inode.  Value
1077  * is NULL to remove an existing extended attribute, and non-NULL to
1078  * either replace an existing extended attribute, or create a new extended
1079  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1080  * specify that an extended attribute must exist and must not exist
1081  * previous to the call, respectively.
1082  *
1083  * Returns 0, or a negative error number on failure.
1084  */
1085 int
1086 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1087                       const char *name, const void *value, size_t value_len,
1088                       int flags)
1089 {
1090         struct ext4_xattr_info i = {
1091                 .name_index = name_index,
1092                 .name = name,
1093                 .value = value,
1094                 .value_len = value_len,
1095
1096         };
1097         struct ext4_xattr_ibody_find is = {
1098                 .s = { .not_found = -ENODATA, },
1099         };
1100         struct ext4_xattr_block_find bs = {
1101                 .s = { .not_found = -ENODATA, },
1102         };
1103         unsigned long no_expand;
1104         int error;
1105
1106         if (!name)
1107                 return -EINVAL;
1108         if (strlen(name) > 255)
1109                 return -ERANGE;
1110         down_write(&EXT4_I(inode)->xattr_sem);
1111         no_expand = ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND);
1112         ext4_set_inode_state(inode, EXT4_STATE_NO_EXPAND);
1113
1114         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1115         if (error)
1116                 goto cleanup;
1117
1118         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1119                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1120                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1121                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1122         }
1123
1124         error = ext4_xattr_ibody_find(inode, &i, &is);
1125         if (error)
1126                 goto cleanup;
1127         if (is.s.not_found)
1128                 error = ext4_xattr_block_find(inode, &i, &bs);
1129         if (error)
1130                 goto cleanup;
1131         if (is.s.not_found && bs.s.not_found) {
1132                 error = -ENODATA;
1133                 if (flags & XATTR_REPLACE)
1134                         goto cleanup;
1135                 error = 0;
1136                 if (!value)
1137                         goto cleanup;
1138         } else {
1139                 error = -EEXIST;
1140                 if (flags & XATTR_CREATE)
1141                         goto cleanup;
1142         }
1143         if (!value) {
1144                 if (!is.s.not_found)
1145                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1146                 else if (!bs.s.not_found)
1147                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1148         } else {
1149                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1150                 if (!error && !bs.s.not_found) {
1151                         i.value = NULL;
1152                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1153                 } else if (error == -ENOSPC) {
1154                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1155                                 error = ext4_xattr_block_find(inode, &i, &bs);
1156                                 if (error)
1157                                         goto cleanup;
1158                         }
1159                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1160                         if (error)
1161                                 goto cleanup;
1162                         if (!is.s.not_found) {
1163                                 i.value = NULL;
1164                                 error = ext4_xattr_ibody_set(handle, inode, &i,
1165                                                              &is);
1166                         }
1167                 }
1168         }
1169         if (!error) {
1170                 ext4_xattr_update_super_block(handle, inode->i_sb);
1171                 inode->i_ctime = ext4_current_time(inode);
1172                 if (!value)
1173                         ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1174                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1175                 /*
1176                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1177                  * error != 0.
1178                  */
1179                 is.iloc.bh = NULL;
1180                 if (IS_SYNC(inode))
1181                         ext4_handle_sync(handle);
1182         }
1183
1184 cleanup:
1185         brelse(is.iloc.bh);
1186         brelse(bs.bh);
1187         if (no_expand == 0)
1188                 ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1189         up_write(&EXT4_I(inode)->xattr_sem);
1190         return error;
1191 }
1192
1193 /*
1194  * ext4_xattr_set()
1195  *
1196  * Like ext4_xattr_set_handle, but start from an inode. This extended
1197  * attribute modification is a filesystem transaction by itself.
1198  *
1199  * Returns 0, or a negative error number on failure.
1200  */
1201 int
1202 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1203                const void *value, size_t value_len, int flags)
1204 {
1205         handle_t *handle;
1206         int error, retries = 0;
1207         int credits = ext4_jbd2_credits_xattr(inode);
1208
1209 retry:
1210         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1211         if (IS_ERR(handle)) {
1212                 error = PTR_ERR(handle);
1213         } else {
1214                 int error2;
1215
1216                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1217                                               value, value_len, flags);
1218                 error2 = ext4_journal_stop(handle);
1219                 if (error == -ENOSPC &&
1220                     ext4_should_retry_alloc(inode->i_sb, &retries))
1221                         goto retry;
1222                 if (error == 0)
1223                         error = error2;
1224         }
1225
1226         return error;
1227 }
1228
1229 /*
1230  * Shift the EA entries in the inode to create space for the increased
1231  * i_extra_isize.
1232  */
1233 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1234                                      int value_offs_shift, void *to,
1235                                      void *from, size_t n, int blocksize)
1236 {
1237         struct ext4_xattr_entry *last = entry;
1238         int new_offs;
1239
1240         /* Adjust the value offsets of the entries */
1241         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1242                 if (!last->e_value_block && last->e_value_size) {
1243                         new_offs = le16_to_cpu(last->e_value_offs) +
1244                                                         value_offs_shift;
1245                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
1246                                  > blocksize);
1247                         last->e_value_offs = cpu_to_le16(new_offs);
1248                 }
1249         }
1250         /* Shift the entries by n bytes */
1251         memmove(to, from, n);
1252 }
1253
1254 /*
1255  * Expand an inode by new_extra_isize bytes when EAs are present.
1256  * Returns 0 on success or negative error number on failure.
1257  */
1258 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1259                                struct ext4_inode *raw_inode, handle_t *handle)
1260 {
1261         struct ext4_xattr_ibody_header *header;
1262         struct ext4_xattr_entry *entry, *last, *first;
1263         struct buffer_head *bh = NULL;
1264         struct ext4_xattr_ibody_find *is = NULL;
1265         struct ext4_xattr_block_find *bs = NULL;
1266         char *buffer = NULL, *b_entry_name = NULL;
1267         size_t min_offs, free;
1268         int total_ino;
1269         void *base, *start, *end;
1270         int extra_isize = 0, error = 0, tried_min_extra_isize = 0;
1271         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1272
1273         down_write(&EXT4_I(inode)->xattr_sem);
1274 retry:
1275         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) {
1276                 up_write(&EXT4_I(inode)->xattr_sem);
1277                 return 0;
1278         }
1279
1280         header = IHDR(inode, raw_inode);
1281         entry = IFIRST(header);
1282
1283         /*
1284          * Check if enough free space is available in the inode to shift the
1285          * entries ahead by new_extra_isize.
1286          */
1287
1288         base = start = entry;
1289         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1290         min_offs = end - base;
1291         last = entry;
1292         total_ino = sizeof(struct ext4_xattr_ibody_header);
1293
1294         free = ext4_xattr_free_space(last, &min_offs, base, &total_ino);
1295         if (free >= new_extra_isize) {
1296                 entry = IFIRST(header);
1297                 ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize
1298                                 - new_extra_isize, (void *)raw_inode +
1299                                 EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1300                                 (void *)header, total_ino,
1301                                 inode->i_sb->s_blocksize);
1302                 EXT4_I(inode)->i_extra_isize = new_extra_isize;
1303                 error = 0;
1304                 goto cleanup;
1305         }
1306
1307         /*
1308          * Enough free space isn't available in the inode, check if
1309          * EA block can hold new_extra_isize bytes.
1310          */
1311         if (EXT4_I(inode)->i_file_acl) {
1312                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1313                 error = -EIO;
1314                 if (!bh)
1315                         goto cleanup;
1316                 if (ext4_xattr_check_block(inode, bh)) {
1317                         EXT4_ERROR_INODE(inode, "bad block %llu",
1318                                          EXT4_I(inode)->i_file_acl);
1319                         error = -EFSCORRUPTED;
1320                         goto cleanup;
1321                 }
1322                 base = BHDR(bh);
1323                 first = BFIRST(bh);
1324                 end = bh->b_data + bh->b_size;
1325                 min_offs = end - base;
1326                 free = ext4_xattr_free_space(first, &min_offs, base, NULL);
1327                 if (free < new_extra_isize) {
1328                         if (!tried_min_extra_isize && s_min_extra_isize) {
1329                                 tried_min_extra_isize++;
1330                                 new_extra_isize = s_min_extra_isize;
1331                                 brelse(bh);
1332                                 goto retry;
1333                         }
1334                         error = -1;
1335                         goto cleanup;
1336                 }
1337         } else {
1338                 free = inode->i_sb->s_blocksize;
1339         }
1340
1341         while (new_extra_isize > 0) {
1342                 size_t offs, size, entry_size;
1343                 struct ext4_xattr_entry *small_entry = NULL;
1344                 struct ext4_xattr_info i = {
1345                         .value = NULL,
1346                         .value_len = 0,
1347                 };
1348                 unsigned int total_size;  /* EA entry size + value size */
1349                 unsigned int shift_bytes; /* No. of bytes to shift EAs by? */
1350                 unsigned int min_total_size = ~0U;
1351
1352                 is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1353                 bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1354                 if (!is || !bs) {
1355                         error = -ENOMEM;
1356                         goto cleanup;
1357                 }
1358
1359                 is->s.not_found = -ENODATA;
1360                 bs->s.not_found = -ENODATA;
1361                 is->iloc.bh = NULL;
1362                 bs->bh = NULL;
1363
1364                 last = IFIRST(header);
1365                 /* Find the entry best suited to be pushed into EA block */
1366                 entry = NULL;
1367                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1368                         total_size =
1369                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1370                                         EXT4_XATTR_LEN(last->e_name_len);
1371                         if (total_size <= free && total_size < min_total_size) {
1372                                 if (total_size < new_extra_isize) {
1373                                         small_entry = last;
1374                                 } else {
1375                                         entry = last;
1376                                         min_total_size = total_size;
1377                                 }
1378                         }
1379                 }
1380
1381                 if (entry == NULL) {
1382                         if (small_entry) {
1383                                 entry = small_entry;
1384                         } else {
1385                                 if (!tried_min_extra_isize &&
1386                                     s_min_extra_isize) {
1387                                         tried_min_extra_isize++;
1388                                         new_extra_isize = s_min_extra_isize;
1389                                         kfree(is); is = NULL;
1390                                         kfree(bs); bs = NULL;
1391                                         brelse(bh);
1392                                         goto retry;
1393                                 }
1394                                 error = -1;
1395                                 goto cleanup;
1396                         }
1397                 }
1398                 offs = le16_to_cpu(entry->e_value_offs);
1399                 size = le32_to_cpu(entry->e_value_size);
1400                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1401                 i.name_index = entry->e_name_index,
1402                 buffer = kmalloc(EXT4_XATTR_SIZE(size), GFP_NOFS);
1403                 b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1404                 if (!buffer || !b_entry_name) {
1405                         error = -ENOMEM;
1406                         goto cleanup;
1407                 }
1408                 /* Save the entry name and the entry value */
1409                 memcpy(buffer, (void *)IFIRST(header) + offs,
1410                        EXT4_XATTR_SIZE(size));
1411                 memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1412                 b_entry_name[entry->e_name_len] = '\0';
1413                 i.name = b_entry_name;
1414
1415                 error = ext4_get_inode_loc(inode, &is->iloc);
1416                 if (error)
1417                         goto cleanup;
1418
1419                 error = ext4_xattr_ibody_find(inode, &i, is);
1420                 if (error)
1421                         goto cleanup;
1422
1423                 /* Remove the chosen entry from the inode */
1424                 error = ext4_xattr_ibody_set(handle, inode, &i, is);
1425                 if (error)
1426                         goto cleanup;
1427
1428                 entry = IFIRST(header);
1429                 if (entry_size + EXT4_XATTR_SIZE(size) >= new_extra_isize)
1430                         shift_bytes = new_extra_isize;
1431                 else
1432                         shift_bytes = entry_size + size;
1433                 /* Adjust the offsets and shift the remaining entries ahead */
1434                 ext4_xattr_shift_entries(entry, EXT4_I(inode)->i_extra_isize -
1435                         shift_bytes, (void *)raw_inode +
1436                         EXT4_GOOD_OLD_INODE_SIZE + extra_isize + shift_bytes,
1437                         (void *)header, total_ino - entry_size,
1438                         inode->i_sb->s_blocksize);
1439
1440                 extra_isize += shift_bytes;
1441                 new_extra_isize -= shift_bytes;
1442                 EXT4_I(inode)->i_extra_isize = extra_isize;
1443
1444                 i.name = b_entry_name;
1445                 i.value = buffer;
1446                 i.value_len = size;
1447                 error = ext4_xattr_block_find(inode, &i, bs);
1448                 if (error)
1449                         goto cleanup;
1450
1451                 /* Add entry which was removed from the inode into the block */
1452                 error = ext4_xattr_block_set(handle, inode, &i, bs);
1453                 if (error)
1454                         goto cleanup;
1455                 kfree(b_entry_name);
1456                 kfree(buffer);
1457                 b_entry_name = NULL;
1458                 buffer = NULL;
1459                 brelse(is->iloc.bh);
1460                 kfree(is);
1461                 kfree(bs);
1462         }
1463         brelse(bh);
1464         up_write(&EXT4_I(inode)->xattr_sem);
1465         return 0;
1466
1467 cleanup:
1468         kfree(b_entry_name);
1469         kfree(buffer);
1470         if (is)
1471                 brelse(is->iloc.bh);
1472         kfree(is);
1473         kfree(bs);
1474         brelse(bh);
1475         up_write(&EXT4_I(inode)->xattr_sem);
1476         return error;
1477 }
1478
1479
1480
1481 /*
1482  * ext4_xattr_delete_inode()
1483  *
1484  * Free extended attribute resources associated with this inode. This
1485  * is called immediately before an inode is freed. We have exclusive
1486  * access to the inode.
1487  */
1488 void
1489 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1490 {
1491         struct buffer_head *bh = NULL;
1492
1493         if (!EXT4_I(inode)->i_file_acl)
1494                 goto cleanup;
1495         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1496         if (!bh) {
1497                 EXT4_ERROR_INODE(inode, "block %llu read error",
1498                                  EXT4_I(inode)->i_file_acl);
1499                 goto cleanup;
1500         }
1501         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1502             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1503                 EXT4_ERROR_INODE(inode, "bad block %llu",
1504                                  EXT4_I(inode)->i_file_acl);
1505                 goto cleanup;
1506         }
1507         ext4_xattr_release_block(handle, inode, bh);
1508         EXT4_I(inode)->i_file_acl = 0;
1509
1510 cleanup:
1511         brelse(bh);
1512 }
1513
1514 /*
1515  * ext4_xattr_put_super()
1516  *
1517  * This is called when a file system is unmounted.
1518  */
1519 void
1520 ext4_xattr_put_super(struct super_block *sb)
1521 {
1522         mb_cache_shrink(sb->s_bdev);
1523 }
1524
1525 /*
1526  * ext4_xattr_cache_insert()
1527  *
1528  * Create a new entry in the extended attribute cache, and insert
1529  * it unless such an entry is already in the cache.
1530  *
1531  * Returns 0, or a negative error number on failure.
1532  */
1533 static void
1534 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1535 {
1536         __u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1537         struct mb_cache_entry *ce;
1538         int error;
1539
1540         ce = mb_cache_entry_alloc(ext4_mb_cache, GFP_NOFS);
1541         if (!ce) {
1542                 ea_bdebug(bh, "out of memory");
1543                 return;
1544         }
1545         error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
1546         if (error) {
1547                 mb_cache_entry_free(ce);
1548                 if (error == -EBUSY) {
1549                         ea_bdebug(bh, "already in cache");
1550                         error = 0;
1551                 }
1552         } else {
1553                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1554                 mb_cache_entry_release(ce);
1555         }
1556 }
1557
1558 /*
1559  * ext4_xattr_cmp()
1560  *
1561  * Compare two extended attribute blocks for equality.
1562  *
1563  * Returns 0 if the blocks are equal, 1 if they differ, and
1564  * a negative error number on errors.
1565  */
1566 static int
1567 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1568                struct ext4_xattr_header *header2)
1569 {
1570         struct ext4_xattr_entry *entry1, *entry2;
1571
1572         entry1 = ENTRY(header1+1);
1573         entry2 = ENTRY(header2+1);
1574         while (!IS_LAST_ENTRY(entry1)) {
1575                 if (IS_LAST_ENTRY(entry2))
1576                         return 1;
1577                 if (entry1->e_hash != entry2->e_hash ||
1578                     entry1->e_name_index != entry2->e_name_index ||
1579                     entry1->e_name_len != entry2->e_name_len ||
1580                     entry1->e_value_size != entry2->e_value_size ||
1581                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1582                         return 1;
1583                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1584                         return -EFSCORRUPTED;
1585                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1586                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1587                            le32_to_cpu(entry1->e_value_size)))
1588                         return 1;
1589
1590                 entry1 = EXT4_XATTR_NEXT(entry1);
1591                 entry2 = EXT4_XATTR_NEXT(entry2);
1592         }
1593         if (!IS_LAST_ENTRY(entry2))
1594                 return 1;
1595         return 0;
1596 }
1597
1598 /*
1599  * ext4_xattr_cache_find()
1600  *
1601  * Find an identical extended attribute block.
1602  *
1603  * Returns a pointer to the block found, or NULL if such a block was
1604  * not found or an error occurred.
1605  */
1606 static struct buffer_head *
1607 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1608                       struct mb_cache_entry **pce)
1609 {
1610         __u32 hash = le32_to_cpu(header->h_hash);
1611         struct mb_cache_entry *ce;
1612         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1613
1614         if (!header->h_hash)
1615                 return NULL;  /* never share */
1616         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1617 again:
1618         ce = mb_cache_entry_find_first(ext4_mb_cache, inode->i_sb->s_bdev,
1619                                        hash);
1620         while (ce) {
1621                 struct buffer_head *bh;
1622
1623                 if (IS_ERR(ce)) {
1624                         if (PTR_ERR(ce) == -EAGAIN)
1625                                 goto again;
1626                         break;
1627                 }
1628                 bh = sb_bread(inode->i_sb, ce->e_block);
1629                 if (!bh) {
1630                         EXT4_ERROR_INODE(inode, "block %lu read error",
1631                                          (unsigned long) ce->e_block);
1632                 } else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1633                                 EXT4_XATTR_REFCOUNT_MAX) {
1634                         ea_idebug(inode, "block %lu refcount %d>=%d",
1635                                   (unsigned long) ce->e_block,
1636                                   le32_to_cpu(BHDR(bh)->h_refcount),
1637                                           EXT4_XATTR_REFCOUNT_MAX);
1638                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1639                         *pce = ce;
1640                         return bh;
1641                 }
1642                 brelse(bh);
1643                 ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
1644         }
1645         return NULL;
1646 }
1647
1648 #define NAME_HASH_SHIFT 5
1649 #define VALUE_HASH_SHIFT 16
1650
1651 /*
1652  * ext4_xattr_hash_entry()
1653  *
1654  * Compute the hash of an extended attribute.
1655  */
1656 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1657                                          struct ext4_xattr_entry *entry)
1658 {
1659         __u32 hash = 0;
1660         char *name = entry->e_name;
1661         int n;
1662
1663         for (n = 0; n < entry->e_name_len; n++) {
1664                 hash = (hash << NAME_HASH_SHIFT) ^
1665                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1666                        *name++;
1667         }
1668
1669         if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1670                 __le32 *value = (__le32 *)((char *)header +
1671                         le16_to_cpu(entry->e_value_offs));
1672                 for (n = (le32_to_cpu(entry->e_value_size) +
1673                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1674                         hash = (hash << VALUE_HASH_SHIFT) ^
1675                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1676                                le32_to_cpu(*value++);
1677                 }
1678         }
1679         entry->e_hash = cpu_to_le32(hash);
1680 }
1681
1682 #undef NAME_HASH_SHIFT
1683 #undef VALUE_HASH_SHIFT
1684
1685 #define BLOCK_HASH_SHIFT 16
1686
1687 /*
1688  * ext4_xattr_rehash()
1689  *
1690  * Re-compute the extended attribute hash value after an entry has changed.
1691  */
1692 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1693                               struct ext4_xattr_entry *entry)
1694 {
1695         struct ext4_xattr_entry *here;
1696         __u32 hash = 0;
1697
1698         ext4_xattr_hash_entry(header, entry);
1699         here = ENTRY(header+1);
1700         while (!IS_LAST_ENTRY(here)) {
1701                 if (!here->e_hash) {
1702                         /* Block is not shared if an entry's hash value == 0 */
1703                         hash = 0;
1704                         break;
1705                 }
1706                 hash = (hash << BLOCK_HASH_SHIFT) ^
1707                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1708                        le32_to_cpu(here->e_hash);
1709                 here = EXT4_XATTR_NEXT(here);
1710         }
1711         header->h_hash = cpu_to_le32(hash);
1712 }
1713
1714 #undef BLOCK_HASH_SHIFT
1715
1716 #define HASH_BUCKET_BITS        10
1717
1718 struct mb_cache *
1719 ext4_xattr_create_cache(char *name)
1720 {
1721         return mb_cache_create(name, HASH_BUCKET_BITS);
1722 }
1723
1724 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1725 {
1726         if (cache)
1727                 mb_cache_destroy(cache);
1728 }
1729