GNU Linux-libre 4.9.318-gnu1
[releases.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, fmt, ...)                                     \
65         printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
66                inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
67 # define ea_bdebug(bh, fmt, ...)                                        \
68         printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
69                bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
70 #else
71 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
72 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
73 #endif
74
75 static void ext4_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
76 static struct buffer_head *ext4_xattr_cache_find(struct inode *,
77                                                  struct ext4_xattr_header *,
78                                                  struct mb_cache_entry **);
79 static void ext4_xattr_rehash(struct ext4_xattr_header *,
80                               struct ext4_xattr_entry *);
81 static int ext4_xattr_list(struct dentry *dentry, char *buffer,
82                            size_t buffer_size);
83
84 static const struct xattr_handler *ext4_xattr_handler_map[] = {
85         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
86 #ifdef CONFIG_EXT4_FS_POSIX_ACL
87         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
88         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
89 #endif
90         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
91 #ifdef CONFIG_EXT4_FS_SECURITY
92         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
93 #endif
94 };
95
96 const struct xattr_handler *ext4_xattr_handlers[] = {
97         &ext4_xattr_user_handler,
98         &ext4_xattr_trusted_handler,
99 #ifdef CONFIG_EXT4_FS_POSIX_ACL
100         &posix_acl_access_xattr_handler,
101         &posix_acl_default_xattr_handler,
102 #endif
103 #ifdef CONFIG_EXT4_FS_SECURITY
104         &ext4_xattr_security_handler,
105 #endif
106         NULL
107 };
108
109 #define EXT4_GET_MB_CACHE(inode)        (((struct ext4_sb_info *) \
110                                 inode->i_sb->s_fs_info)->s_mb_cache)
111
112 static __le32 ext4_xattr_block_csum(struct inode *inode,
113                                     sector_t block_nr,
114                                     struct ext4_xattr_header *hdr)
115 {
116         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
117         __u32 csum;
118         __le64 dsk_block_nr = cpu_to_le64(block_nr);
119         __u32 dummy_csum = 0;
120         int offset = offsetof(struct ext4_xattr_header, h_checksum);
121
122         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
123                            sizeof(dsk_block_nr));
124         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
125         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
126         offset += sizeof(dummy_csum);
127         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
128                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
129
130         return cpu_to_le32(csum);
131 }
132
133 static int ext4_xattr_block_csum_verify(struct inode *inode,
134                                         struct buffer_head *bh)
135 {
136         struct ext4_xattr_header *hdr = BHDR(bh);
137         int ret = 1;
138
139         if (ext4_has_metadata_csum(inode->i_sb)) {
140                 lock_buffer(bh);
141                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
142                                                         bh->b_blocknr, hdr));
143                 unlock_buffer(bh);
144         }
145         return ret;
146 }
147
148 static void ext4_xattr_block_csum_set(struct inode *inode,
149                                       struct buffer_head *bh)
150 {
151         if (ext4_has_metadata_csum(inode->i_sb))
152                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
153                                                 bh->b_blocknr, BHDR(bh));
154 }
155
156 static inline const struct xattr_handler *
157 ext4_xattr_handler(int name_index)
158 {
159         const struct xattr_handler *handler = NULL;
160
161         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
162                 handler = ext4_xattr_handler_map[name_index];
163         return handler;
164 }
165
166 /*
167  * Inode operation listxattr()
168  *
169  * d_inode(dentry)->i_mutex: don't care
170  */
171 ssize_t
172 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
173 {
174         return ext4_xattr_list(dentry, buffer, size);
175 }
176
177 static int
178 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
179                        void *value_start)
180 {
181         struct ext4_xattr_entry *e = entry;
182
183         while (!IS_LAST_ENTRY(e)) {
184                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
185                 if ((void *)next >= end)
186                         return -EFSCORRUPTED;
187                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
188                         return -EFSCORRUPTED;
189                 e = next;
190         }
191
192         while (!IS_LAST_ENTRY(entry)) {
193                 if (entry->e_value_block != 0)
194                         return -EFSCORRUPTED;
195                 if (entry->e_value_size != 0 &&
196                     (value_start + le16_to_cpu(entry->e_value_offs) <
197                      (void *)e + sizeof(__u32) ||
198                      value_start + le16_to_cpu(entry->e_value_offs) +
199                     le32_to_cpu(entry->e_value_size) > end))
200                         return -EFSCORRUPTED;
201                 entry = EXT4_XATTR_NEXT(entry);
202         }
203
204         return 0;
205 }
206
207 static inline int
208 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
209 {
210         int error;
211
212         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
213             BHDR(bh)->h_blocks != cpu_to_le32(1))
214                 return -EFSCORRUPTED;
215         if (buffer_verified(bh))
216                 return 0;
217
218         if (!ext4_xattr_block_csum_verify(inode, bh))
219                 return -EFSBADCRC;
220         error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size,
221                                        bh->b_data);
222         if (!error)
223                 set_buffer_verified(bh);
224         return error;
225 }
226
227 static int
228 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
229                          void *end, const char *function, unsigned int line)
230 {
231         struct ext4_xattr_entry *entry = IFIRST(header);
232         int error = -EFSCORRUPTED;
233
234         if (((void *) header >= end) ||
235             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
236                 goto errout;
237         error = ext4_xattr_check_names(entry, end, entry);
238 errout:
239         if (error)
240                 __ext4_error_inode(inode, function, line, 0,
241                                    "corrupted in-inode xattr");
242         return error;
243 }
244
245 #define xattr_check_inode(inode, header, end) \
246         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
247
248 static inline int
249 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
250 {
251         size_t value_size = le32_to_cpu(entry->e_value_size);
252
253         if (entry->e_value_block != 0 || value_size > size ||
254             le16_to_cpu(entry->e_value_offs) + value_size > size)
255                 return -EFSCORRUPTED;
256         return 0;
257 }
258
259 static int
260 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
261                       const char *name, size_t size, int sorted)
262 {
263         struct ext4_xattr_entry *entry;
264         size_t name_len;
265         int cmp = 1;
266
267         if (name == NULL)
268                 return -EINVAL;
269         name_len = strlen(name);
270         entry = *pentry;
271         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
272                 cmp = name_index - entry->e_name_index;
273                 if (!cmp)
274                         cmp = name_len - entry->e_name_len;
275                 if (!cmp)
276                         cmp = memcmp(name, entry->e_name, name_len);
277                 if (cmp <= 0 && (sorted || cmp == 0))
278                         break;
279         }
280         *pentry = entry;
281         if (!cmp && ext4_xattr_check_entry(entry, size))
282                 return -EFSCORRUPTED;
283         return cmp ? -ENODATA : 0;
284 }
285
286 static int
287 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
288                      void *buffer, size_t buffer_size)
289 {
290         struct buffer_head *bh = NULL;
291         struct ext4_xattr_entry *entry;
292         size_t size;
293         int error;
294         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
295
296         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
297                   name_index, name, buffer, (long)buffer_size);
298
299         error = -ENODATA;
300         if (!EXT4_I(inode)->i_file_acl)
301                 goto cleanup;
302         ea_idebug(inode, "reading block %llu",
303                   (unsigned long long)EXT4_I(inode)->i_file_acl);
304         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
305         if (!bh)
306                 goto cleanup;
307         ea_bdebug(bh, "b_count=%d, refcount=%d",
308                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
309         if (ext4_xattr_check_block(inode, bh)) {
310 bad_block:
311                 EXT4_ERROR_INODE(inode, "bad block %llu",
312                                  EXT4_I(inode)->i_file_acl);
313                 error = -EFSCORRUPTED;
314                 goto cleanup;
315         }
316         ext4_xattr_cache_insert(ext4_mb_cache, bh);
317         entry = BFIRST(bh);
318         error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
319         if (error == -EFSCORRUPTED)
320                 goto bad_block;
321         if (error)
322                 goto cleanup;
323         size = le32_to_cpu(entry->e_value_size);
324         if (buffer) {
325                 error = -ERANGE;
326                 if (size > buffer_size)
327                         goto cleanup;
328                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
329                        size);
330         }
331         error = size;
332
333 cleanup:
334         brelse(bh);
335         return error;
336 }
337
338 int
339 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
340                      void *buffer, size_t buffer_size)
341 {
342         struct ext4_xattr_ibody_header *header;
343         struct ext4_xattr_entry *entry;
344         struct ext4_inode *raw_inode;
345         struct ext4_iloc iloc;
346         size_t size;
347         void *end;
348         int error;
349
350         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
351                 return -ENODATA;
352         error = ext4_get_inode_loc(inode, &iloc);
353         if (error)
354                 return error;
355         raw_inode = ext4_raw_inode(&iloc);
356         header = IHDR(inode, raw_inode);
357         entry = IFIRST(header);
358         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
359         error = xattr_check_inode(inode, header, end);
360         if (error)
361                 goto cleanup;
362         error = ext4_xattr_find_entry(&entry, name_index, name,
363                                       end - (void *)entry, 0);
364         if (error)
365                 goto cleanup;
366         size = le32_to_cpu(entry->e_value_size);
367         if (buffer) {
368                 error = -ERANGE;
369                 if (size > buffer_size)
370                         goto cleanup;
371                 memcpy(buffer, (void *)IFIRST(header) +
372                        le16_to_cpu(entry->e_value_offs), size);
373         }
374         error = size;
375
376 cleanup:
377         brelse(iloc.bh);
378         return error;
379 }
380
381 /*
382  * ext4_xattr_get()
383  *
384  * Copy an extended attribute into the buffer
385  * provided, or compute the buffer size required.
386  * Buffer is NULL to compute the size of the buffer required.
387  *
388  * Returns a negative error number on failure, or the number of bytes
389  * used / required on success.
390  */
391 int
392 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
393                void *buffer, size_t buffer_size)
394 {
395         int error;
396
397         if (strlen(name) > 255)
398                 return -ERANGE;
399
400         down_read(&EXT4_I(inode)->xattr_sem);
401         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
402                                      buffer_size);
403         if (error == -ENODATA)
404                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
405                                              buffer_size);
406         up_read(&EXT4_I(inode)->xattr_sem);
407         return error;
408 }
409
410 static int
411 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
412                         char *buffer, size_t buffer_size)
413 {
414         size_t rest = buffer_size;
415
416         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
417                 const struct xattr_handler *handler =
418                         ext4_xattr_handler(entry->e_name_index);
419
420                 if (handler && (!handler->list || handler->list(dentry))) {
421                         const char *prefix = handler->prefix ?: handler->name;
422                         size_t prefix_len = strlen(prefix);
423                         size_t size = prefix_len + entry->e_name_len + 1;
424
425                         if (buffer) {
426                                 if (size > rest)
427                                         return -ERANGE;
428                                 memcpy(buffer, prefix, prefix_len);
429                                 buffer += prefix_len;
430                                 memcpy(buffer, entry->e_name, entry->e_name_len);
431                                 buffer += entry->e_name_len;
432                                 *buffer++ = 0;
433                         }
434                         rest -= size;
435                 }
436         }
437         return buffer_size - rest;  /* total size */
438 }
439
440 static int
441 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
442 {
443         struct inode *inode = d_inode(dentry);
444         struct buffer_head *bh = NULL;
445         int error;
446         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
447
448         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
449                   buffer, (long)buffer_size);
450
451         error = 0;
452         if (!EXT4_I(inode)->i_file_acl)
453                 goto cleanup;
454         ea_idebug(inode, "reading block %llu",
455                   (unsigned long long)EXT4_I(inode)->i_file_acl);
456         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
457         error = -EIO;
458         if (!bh)
459                 goto cleanup;
460         ea_bdebug(bh, "b_count=%d, refcount=%d",
461                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
462         if (ext4_xattr_check_block(inode, bh)) {
463                 EXT4_ERROR_INODE(inode, "bad block %llu",
464                                  EXT4_I(inode)->i_file_acl);
465                 error = -EFSCORRUPTED;
466                 goto cleanup;
467         }
468         ext4_xattr_cache_insert(ext4_mb_cache, bh);
469         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
470
471 cleanup:
472         brelse(bh);
473
474         return error;
475 }
476
477 static int
478 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
479 {
480         struct inode *inode = d_inode(dentry);
481         struct ext4_xattr_ibody_header *header;
482         struct ext4_inode *raw_inode;
483         struct ext4_iloc iloc;
484         void *end;
485         int error;
486
487         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
488                 return 0;
489         error = ext4_get_inode_loc(inode, &iloc);
490         if (error)
491                 return error;
492         raw_inode = ext4_raw_inode(&iloc);
493         header = IHDR(inode, raw_inode);
494         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
495         error = xattr_check_inode(inode, header, end);
496         if (error)
497                 goto cleanup;
498         error = ext4_xattr_list_entries(dentry, IFIRST(header),
499                                         buffer, buffer_size);
500
501 cleanup:
502         brelse(iloc.bh);
503         return error;
504 }
505
506 /*
507  * ext4_xattr_list()
508  *
509  * Copy a list of attribute names into the buffer
510  * provided, or compute the buffer size required.
511  * Buffer is NULL to compute the size of the buffer required.
512  *
513  * Returns a negative error number on failure, or the number of bytes
514  * used / required on success.
515  */
516 static int
517 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
518 {
519         int ret, ret2;
520
521         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
522         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
523         if (ret < 0)
524                 goto errout;
525         if (buffer) {
526                 buffer += ret;
527                 buffer_size -= ret;
528         }
529         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
530         if (ret < 0)
531                 goto errout;
532         ret += ret2;
533 errout:
534         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
535         return ret;
536 }
537
538 /*
539  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
540  * not set, set it.
541  */
542 static void ext4_xattr_update_super_block(handle_t *handle,
543                                           struct super_block *sb)
544 {
545         if (ext4_has_feature_xattr(sb))
546                 return;
547
548         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
549         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
550                 ext4_set_feature_xattr(sb);
551                 ext4_handle_dirty_super(handle, sb);
552         }
553 }
554
555 /*
556  * Release the xattr block BH: If the reference count is > 1, decrement it;
557  * otherwise free the block.
558  */
559 static void
560 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
561                          struct buffer_head *bh)
562 {
563         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
564         u32 hash, ref;
565         int error = 0;
566
567         BUFFER_TRACE(bh, "get_write_access");
568         error = ext4_journal_get_write_access(handle, bh);
569         if (error)
570                 goto out;
571
572         lock_buffer(bh);
573         hash = le32_to_cpu(BHDR(bh)->h_hash);
574         ref = le32_to_cpu(BHDR(bh)->h_refcount);
575         if (ref == 1) {
576                 ea_bdebug(bh, "refcount now=0; freeing");
577                 /*
578                  * This must happen under buffer lock for
579                  * ext4_xattr_block_set() to reliably detect freed block
580                  */
581                 mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
582                 get_bh(bh);
583                 unlock_buffer(bh);
584                 ext4_free_blocks(handle, inode, bh, 0, 1,
585                                  EXT4_FREE_BLOCKS_METADATA |
586                                  EXT4_FREE_BLOCKS_FORGET);
587         } else {
588                 ref--;
589                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
590                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
591                         struct mb_cache_entry *ce;
592
593                         ce = mb_cache_entry_get(ext4_mb_cache, hash,
594                                                 bh->b_blocknr);
595                         if (ce) {
596                                 ce->e_reusable = 1;
597                                 mb_cache_entry_put(ext4_mb_cache, ce);
598                         }
599                 }
600
601                 ext4_xattr_block_csum_set(inode, bh);
602                 /*
603                  * Beware of this ugliness: Releasing of xattr block references
604                  * from different inodes can race and so we have to protect
605                  * from a race where someone else frees the block (and releases
606                  * its journal_head) before we are done dirtying the buffer. In
607                  * nojournal mode this race is harmless and we actually cannot
608                  * call ext4_handle_dirty_metadata() with locked buffer as
609                  * that function can call sync_dirty_buffer() so for that case
610                  * we handle the dirtying after unlocking the buffer.
611                  */
612                 if (ext4_handle_valid(handle))
613                         error = ext4_handle_dirty_metadata(handle, inode, bh);
614                 unlock_buffer(bh);
615                 if (!ext4_handle_valid(handle))
616                         error = ext4_handle_dirty_metadata(handle, inode, bh);
617                 if (IS_SYNC(inode))
618                         ext4_handle_sync(handle);
619                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
620                 ea_bdebug(bh, "refcount now=%d; releasing",
621                           le32_to_cpu(BHDR(bh)->h_refcount));
622         }
623 out:
624         ext4_std_error(inode->i_sb, error);
625         return;
626 }
627
628 /*
629  * Find the available free space for EAs. This also returns the total number of
630  * bytes used by EA entries.
631  */
632 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
633                                     size_t *min_offs, void *base, int *total)
634 {
635         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
636                 if (last->e_value_size) {
637                         size_t offs = le16_to_cpu(last->e_value_offs);
638                         if (offs < *min_offs)
639                                 *min_offs = offs;
640                 }
641                 if (total)
642                         *total += EXT4_XATTR_LEN(last->e_name_len);
643         }
644         return (*min_offs - ((void *)last - base) - sizeof(__u32));
645 }
646
647 static int
648 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
649                      struct inode *inode)
650 {
651         struct ext4_xattr_entry *last, *next;
652         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
653
654         /* Compute min_offs and last. */
655         last = s->first;
656         for (; !IS_LAST_ENTRY(last); last = next) {
657                 next = EXT4_XATTR_NEXT(last);
658                 if ((void *)next >= s->end) {
659                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
660                         return -EFSCORRUPTED;
661                 }
662                 if (last->e_value_size) {
663                         size_t offs = le16_to_cpu(last->e_value_offs);
664                         if (offs < min_offs)
665                                 min_offs = offs;
666                 }
667         }
668         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
669         if (!s->not_found) {
670                 if (s->here->e_value_size) {
671                         size_t size = le32_to_cpu(s->here->e_value_size);
672                         free += EXT4_XATTR_SIZE(size);
673                 }
674                 free += EXT4_XATTR_LEN(name_len);
675         }
676         if (i->value) {
677                 if (free < EXT4_XATTR_LEN(name_len) +
678                            EXT4_XATTR_SIZE(i->value_len))
679                         return -ENOSPC;
680         }
681
682         if (i->value && s->not_found) {
683                 /* Insert the new name. */
684                 size_t size = EXT4_XATTR_LEN(name_len);
685                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
686                 memmove((void *)s->here + size, s->here, rest);
687                 memset(s->here, 0, size);
688                 s->here->e_name_index = i->name_index;
689                 s->here->e_name_len = name_len;
690                 memcpy(s->here->e_name, i->name, name_len);
691         } else {
692                 if (s->here->e_value_size) {
693                         void *first_val = s->base + min_offs;
694                         size_t offs = le16_to_cpu(s->here->e_value_offs);
695                         void *val = s->base + offs;
696                         size_t size = EXT4_XATTR_SIZE(
697                                 le32_to_cpu(s->here->e_value_size));
698
699                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
700                                 /* The old and the new value have the same
701                                    size. Just replace. */
702                                 s->here->e_value_size =
703                                         cpu_to_le32(i->value_len);
704                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
705                                         memset(val, 0, size);
706                                 } else {
707                                         /* Clear pad bytes first. */
708                                         memset(val + size - EXT4_XATTR_PAD, 0,
709                                                EXT4_XATTR_PAD);
710                                         memcpy(val, i->value, i->value_len);
711                                 }
712                                 return 0;
713                         }
714
715                         /* Remove the old value. */
716                         memmove(first_val + size, first_val, val - first_val);
717                         memset(first_val, 0, size);
718                         s->here->e_value_size = 0;
719                         s->here->e_value_offs = 0;
720                         min_offs += size;
721
722                         /* Adjust all value offsets. */
723                         last = s->first;
724                         while (!IS_LAST_ENTRY(last)) {
725                                 size_t o = le16_to_cpu(last->e_value_offs);
726                                 if (last->e_value_size && o < offs)
727                                         last->e_value_offs =
728                                                 cpu_to_le16(o + size);
729                                 last = EXT4_XATTR_NEXT(last);
730                         }
731                 }
732                 if (!i->value) {
733                         /* Remove the old name. */
734                         size_t size = EXT4_XATTR_LEN(name_len);
735                         last = ENTRY((void *)last - size);
736                         memmove(s->here, (void *)s->here + size,
737                                 (void *)last - (void *)s->here + sizeof(__u32));
738                         memset(last, 0, size);
739                 }
740         }
741
742         if (i->value) {
743                 /* Insert the new value. */
744                 s->here->e_value_size = cpu_to_le32(i->value_len);
745                 if (i->value_len) {
746                         size_t size = EXT4_XATTR_SIZE(i->value_len);
747                         void *val = s->base + min_offs - size;
748                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
749                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
750                                 memset(val, 0, size);
751                         } else {
752                                 /* Clear the pad bytes first. */
753                                 memset(val + size - EXT4_XATTR_PAD, 0,
754                                        EXT4_XATTR_PAD);
755                                 memcpy(val, i->value, i->value_len);
756                         }
757                 }
758         }
759         return 0;
760 }
761
762 struct ext4_xattr_block_find {
763         struct ext4_xattr_search s;
764         struct buffer_head *bh;
765 };
766
767 static int
768 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
769                       struct ext4_xattr_block_find *bs)
770 {
771         struct super_block *sb = inode->i_sb;
772         int error;
773
774         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
775                   i->name_index, i->name, i->value, (long)i->value_len);
776
777         if (EXT4_I(inode)->i_file_acl) {
778                 /* The inode already has an extended attribute block. */
779                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
780                 error = -EIO;
781                 if (!bs->bh)
782                         goto cleanup;
783                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
784                         atomic_read(&(bs->bh->b_count)),
785                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
786                 if (ext4_xattr_check_block(inode, bs->bh)) {
787                         EXT4_ERROR_INODE(inode, "bad block %llu",
788                                          EXT4_I(inode)->i_file_acl);
789                         error = -EFSCORRUPTED;
790                         goto cleanup;
791                 }
792                 /* Find the named attribute. */
793                 bs->s.base = BHDR(bs->bh);
794                 bs->s.first = BFIRST(bs->bh);
795                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
796                 bs->s.here = bs->s.first;
797                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
798                                               i->name, bs->bh->b_size, 1);
799                 if (error && error != -ENODATA)
800                         goto cleanup;
801                 bs->s.not_found = error;
802         }
803         error = 0;
804
805 cleanup:
806         return error;
807 }
808
809 static int
810 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
811                      struct ext4_xattr_info *i,
812                      struct ext4_xattr_block_find *bs)
813 {
814         struct super_block *sb = inode->i_sb;
815         struct buffer_head *new_bh = NULL;
816         struct ext4_xattr_search *s = &bs->s;
817         struct mb_cache_entry *ce = NULL;
818         int error = 0;
819         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
820
821 #define header(x) ((struct ext4_xattr_header *)(x))
822
823         if (i->value && i->value_len > sb->s_blocksize)
824                 return -ENOSPC;
825         if (s->base) {
826                 BUFFER_TRACE(bs->bh, "get_write_access");
827                 error = ext4_journal_get_write_access(handle, bs->bh);
828                 if (error)
829                         goto cleanup;
830                 lock_buffer(bs->bh);
831
832                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
833                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
834
835                         /*
836                          * This must happen under buffer lock for
837                          * ext4_xattr_block_set() to reliably detect modified
838                          * block
839                          */
840                         mb_cache_entry_delete_block(ext4_mb_cache, hash,
841                                                     bs->bh->b_blocknr);
842                         ea_bdebug(bs->bh, "modifying in-place");
843                         error = ext4_xattr_set_entry(i, s, inode);
844                         if (!error) {
845                                 if (!IS_LAST_ENTRY(s->first))
846                                         ext4_xattr_rehash(header(s->base),
847                                                           s->here);
848                         }
849                         ext4_xattr_block_csum_set(inode, bs->bh);
850                         unlock_buffer(bs->bh);
851                         if (error == -EFSCORRUPTED)
852                                 goto bad_block;
853                         if (!error)
854                                 error = ext4_handle_dirty_metadata(handle,
855                                                                    inode,
856                                                                    bs->bh);
857                         if (error)
858                                 goto cleanup;
859                         goto inserted;
860                 } else {
861                         int offset = (char *)s->here - bs->bh->b_data;
862
863                         unlock_buffer(bs->bh);
864                         ea_bdebug(bs->bh, "cloning");
865                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
866                         error = -ENOMEM;
867                         if (s->base == NULL)
868                                 goto cleanup;
869                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
870                         s->first = ENTRY(header(s->base)+1);
871                         header(s->base)->h_refcount = cpu_to_le32(1);
872                         s->here = ENTRY(s->base + offset);
873                         s->end = s->base + bs->bh->b_size;
874                 }
875         } else {
876                 /* Allocate a buffer where we construct the new block. */
877                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
878                 /* assert(header == s->base) */
879                 error = -ENOMEM;
880                 if (s->base == NULL)
881                         goto cleanup;
882                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
883                 header(s->base)->h_blocks = cpu_to_le32(1);
884                 header(s->base)->h_refcount = cpu_to_le32(1);
885                 s->first = ENTRY(header(s->base)+1);
886                 s->here = ENTRY(header(s->base)+1);
887                 s->end = s->base + sb->s_blocksize;
888         }
889
890         error = ext4_xattr_set_entry(i, s, inode);
891         if (error == -EFSCORRUPTED)
892                 goto bad_block;
893         if (error)
894                 goto cleanup;
895         if (!IS_LAST_ENTRY(s->first))
896                 ext4_xattr_rehash(header(s->base), s->here);
897
898 inserted:
899         if (!IS_LAST_ENTRY(s->first)) {
900                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
901                 if (new_bh) {
902                         /* We found an identical block in the cache. */
903                         if (new_bh == bs->bh)
904                                 ea_bdebug(new_bh, "keeping");
905                         else {
906                                 u32 ref;
907
908                                 /* The old block is released after updating
909                                    the inode. */
910                                 error = dquot_alloc_block(inode,
911                                                 EXT4_C2B(EXT4_SB(sb), 1));
912                                 if (error)
913                                         goto cleanup;
914                                 BUFFER_TRACE(new_bh, "get_write_access");
915                                 error = ext4_journal_get_write_access(handle,
916                                                                       new_bh);
917                                 if (error)
918                                         goto cleanup_dquot;
919                                 lock_buffer(new_bh);
920                                 /*
921                                  * We have to be careful about races with
922                                  * freeing, rehashing or adding references to
923                                  * xattr block. Once we hold buffer lock xattr
924                                  * block's state is stable so we can check
925                                  * whether the block got freed / rehashed or
926                                  * not.  Since we unhash mbcache entry under
927                                  * buffer lock when freeing / rehashing xattr
928                                  * block, checking whether entry is still
929                                  * hashed is reliable. Same rules hold for
930                                  * e_reusable handling.
931                                  */
932                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
933                                     !ce->e_reusable) {
934                                         /*
935                                          * Undo everything and check mbcache
936                                          * again.
937                                          */
938                                         unlock_buffer(new_bh);
939                                         dquot_free_block(inode,
940                                                          EXT4_C2B(EXT4_SB(sb),
941                                                                   1));
942                                         brelse(new_bh);
943                                         mb_cache_entry_put(ext4_mb_cache, ce);
944                                         ce = NULL;
945                                         new_bh = NULL;
946                                         goto inserted;
947                                 }
948                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
949                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
950                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
951                                         ce->e_reusable = 0;
952                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
953                                           ref);
954                                 ext4_xattr_block_csum_set(inode, new_bh);
955                                 unlock_buffer(new_bh);
956                                 error = ext4_handle_dirty_metadata(handle,
957                                                                    inode,
958                                                                    new_bh);
959                                 if (error)
960                                         goto cleanup_dquot;
961                         }
962                         mb_cache_entry_touch(ext4_mb_cache, ce);
963                         mb_cache_entry_put(ext4_mb_cache, ce);
964                         ce = NULL;
965                 } else if (bs->bh && s->base == bs->bh->b_data) {
966                         /* We were modifying this block in-place. */
967                         ea_bdebug(bs->bh, "keeping this block");
968                         ext4_xattr_cache_insert(ext4_mb_cache, bs->bh);
969                         new_bh = bs->bh;
970                         get_bh(new_bh);
971                 } else {
972                         /* We need to allocate a new block */
973                         ext4_fsblk_t goal, block;
974
975                         goal = ext4_group_first_block_no(sb,
976                                                 EXT4_I(inode)->i_block_group);
977
978                         /* non-extent files can't have physical blocks past 2^32 */
979                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
980                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
981
982                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
983                                                      NULL, &error);
984                         if (error)
985                                 goto cleanup;
986
987                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
988                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
989
990                         ea_idebug(inode, "creating block %llu",
991                                   (unsigned long long)block);
992
993                         new_bh = sb_getblk(sb, block);
994                         if (unlikely(!new_bh)) {
995                                 error = -ENOMEM;
996 getblk_failed:
997                                 ext4_free_blocks(handle, inode, NULL, block, 1,
998                                                  EXT4_FREE_BLOCKS_METADATA);
999                                 goto cleanup;
1000                         }
1001                         lock_buffer(new_bh);
1002                         error = ext4_journal_get_create_access(handle, new_bh);
1003                         if (error) {
1004                                 unlock_buffer(new_bh);
1005                                 error = -EIO;
1006                                 goto getblk_failed;
1007                         }
1008                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
1009                         ext4_xattr_block_csum_set(inode, new_bh);
1010                         set_buffer_uptodate(new_bh);
1011                         unlock_buffer(new_bh);
1012                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
1013                         error = ext4_handle_dirty_metadata(handle, inode,
1014                                                            new_bh);
1015                         if (error)
1016                                 goto cleanup;
1017                 }
1018         }
1019
1020         /* Update the inode. */
1021         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
1022
1023         /* Drop the previous xattr block. */
1024         if (bs->bh && bs->bh != new_bh)
1025                 ext4_xattr_release_block(handle, inode, bs->bh);
1026         error = 0;
1027
1028 cleanup:
1029         if (ce)
1030                 mb_cache_entry_put(ext4_mb_cache, ce);
1031         brelse(new_bh);
1032         if (!(bs->bh && s->base == bs->bh->b_data))
1033                 kfree(s->base);
1034
1035         return error;
1036
1037 cleanup_dquot:
1038         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
1039         goto cleanup;
1040
1041 bad_block:
1042         EXT4_ERROR_INODE(inode, "bad block %llu",
1043                          EXT4_I(inode)->i_file_acl);
1044         goto cleanup;
1045
1046 #undef header
1047 }
1048
1049 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1050                           struct ext4_xattr_ibody_find *is)
1051 {
1052         struct ext4_xattr_ibody_header *header;
1053         struct ext4_inode *raw_inode;
1054         int error;
1055
1056         if (EXT4_I(inode)->i_extra_isize == 0)
1057                 return 0;
1058         raw_inode = ext4_raw_inode(&is->iloc);
1059         header = IHDR(inode, raw_inode);
1060         is->s.base = is->s.first = IFIRST(header);
1061         is->s.here = is->s.first;
1062         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1063         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1064                 error = xattr_check_inode(inode, header, is->s.end);
1065                 if (error)
1066                         return error;
1067                 /* Find the named attribute. */
1068                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1069                                               i->name, is->s.end -
1070                                               (void *)is->s.base, 0);
1071                 if (error && error != -ENODATA)
1072                         return error;
1073                 is->s.not_found = error;
1074         }
1075         return 0;
1076 }
1077
1078 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1079                                 struct ext4_xattr_info *i,
1080                                 struct ext4_xattr_ibody_find *is)
1081 {
1082         struct ext4_xattr_ibody_header *header;
1083         struct ext4_xattr_search *s = &is->s;
1084         int error;
1085
1086         if (EXT4_I(inode)->i_extra_isize == 0)
1087                 return -ENOSPC;
1088         error = ext4_xattr_set_entry(i, s, inode);
1089         if (error)
1090                 return error;
1091         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1092         if (!IS_LAST_ENTRY(s->first)) {
1093                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1094                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1095         } else {
1096                 header->h_magic = cpu_to_le32(0);
1097                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1098         }
1099         return 0;
1100 }
1101
1102 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
1103                                 struct ext4_xattr_info *i,
1104                                 struct ext4_xattr_ibody_find *is)
1105 {
1106         struct ext4_xattr_ibody_header *header;
1107         struct ext4_xattr_search *s = &is->s;
1108         int error;
1109
1110         if (EXT4_I(inode)->i_extra_isize == 0)
1111                 return -ENOSPC;
1112         error = ext4_xattr_set_entry(i, s, inode);
1113         if (error)
1114                 return error;
1115         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1116         if (!IS_LAST_ENTRY(s->first)) {
1117                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1118                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1119         } else {
1120                 header->h_magic = cpu_to_le32(0);
1121                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1122         }
1123         return 0;
1124 }
1125
1126 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
1127                                  struct ext4_xattr_info *i)
1128 {
1129         void *value;
1130
1131         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
1132                 return 0;
1133         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
1134         return !memcmp(value, i->value, i->value_len);
1135 }
1136
1137 /*
1138  * ext4_xattr_set_handle()
1139  *
1140  * Create, replace or remove an extended attribute for this inode.  Value
1141  * is NULL to remove an existing extended attribute, and non-NULL to
1142  * either replace an existing extended attribute, or create a new extended
1143  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1144  * specify that an extended attribute must exist and must not exist
1145  * previous to the call, respectively.
1146  *
1147  * Returns 0, or a negative error number on failure.
1148  */
1149 int
1150 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1151                       const char *name, const void *value, size_t value_len,
1152                       int flags)
1153 {
1154         struct ext4_xattr_info i = {
1155                 .name_index = name_index,
1156                 .name = name,
1157                 .value = value,
1158                 .value_len = value_len,
1159
1160         };
1161         struct ext4_xattr_ibody_find is = {
1162                 .s = { .not_found = -ENODATA, },
1163         };
1164         struct ext4_xattr_block_find bs = {
1165                 .s = { .not_found = -ENODATA, },
1166         };
1167         int no_expand;
1168         int error;
1169
1170         if (!name)
1171                 return -EINVAL;
1172         if (strlen(name) > 255)
1173                 return -ERANGE;
1174         ext4_write_lock_xattr(inode, &no_expand);
1175
1176         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1177         if (error)
1178                 goto cleanup;
1179
1180         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1181                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1182                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1183                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1184         }
1185
1186         error = ext4_xattr_ibody_find(inode, &i, &is);
1187         if (error)
1188                 goto cleanup;
1189         if (is.s.not_found)
1190                 error = ext4_xattr_block_find(inode, &i, &bs);
1191         if (error)
1192                 goto cleanup;
1193         if (is.s.not_found && bs.s.not_found) {
1194                 error = -ENODATA;
1195                 if (flags & XATTR_REPLACE)
1196                         goto cleanup;
1197                 error = 0;
1198                 if (!value)
1199                         goto cleanup;
1200         } else {
1201                 error = -EEXIST;
1202                 if (flags & XATTR_CREATE)
1203                         goto cleanup;
1204         }
1205         if (!value) {
1206                 if (!is.s.not_found)
1207                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1208                 else if (!bs.s.not_found)
1209                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1210         } else {
1211                 error = 0;
1212                 /* Xattr value did not change? Save us some work and bail out */
1213                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
1214                         goto cleanup;
1215                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
1216                         goto cleanup;
1217
1218                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
1219                 if (!error && !bs.s.not_found) {
1220                         i.value = NULL;
1221                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1222                 } else if (error == -ENOSPC) {
1223                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1224                                 brelse(bs.bh);
1225                                 bs.bh = NULL;
1226                                 error = ext4_xattr_block_find(inode, &i, &bs);
1227                                 if (error)
1228                                         goto cleanup;
1229                         }
1230                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1231                         if (error)
1232                                 goto cleanup;
1233                         if (!is.s.not_found) {
1234                                 i.value = NULL;
1235                                 error = ext4_xattr_ibody_set(handle, inode, &i,
1236                                                              &is);
1237                         }
1238                 }
1239         }
1240         if (!error) {
1241                 ext4_xattr_update_super_block(handle, inode->i_sb);
1242                 inode->i_ctime = ext4_current_time(inode);
1243                 if (!value)
1244                         no_expand = 0;
1245                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1246                 /*
1247                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1248                  * error != 0.
1249                  */
1250                 is.iloc.bh = NULL;
1251                 if (IS_SYNC(inode))
1252                         ext4_handle_sync(handle);
1253         }
1254
1255 cleanup:
1256         brelse(is.iloc.bh);
1257         brelse(bs.bh);
1258         ext4_write_unlock_xattr(inode, &no_expand);
1259         return error;
1260 }
1261
1262 /*
1263  * ext4_xattr_set()
1264  *
1265  * Like ext4_xattr_set_handle, but start from an inode. This extended
1266  * attribute modification is a filesystem transaction by itself.
1267  *
1268  * Returns 0, or a negative error number on failure.
1269  */
1270 int
1271 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1272                const void *value, size_t value_len, int flags)
1273 {
1274         handle_t *handle;
1275         int error, retries = 0;
1276         int credits = ext4_jbd2_credits_xattr(inode);
1277
1278 retry:
1279         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1280         if (IS_ERR(handle)) {
1281                 error = PTR_ERR(handle);
1282         } else {
1283                 int error2;
1284
1285                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1286                                               value, value_len, flags);
1287                 error2 = ext4_journal_stop(handle);
1288                 if (error == -ENOSPC &&
1289                     ext4_should_retry_alloc(inode->i_sb, &retries))
1290                         goto retry;
1291                 if (error == 0)
1292                         error = error2;
1293         }
1294
1295         return error;
1296 }
1297
1298 /*
1299  * Shift the EA entries in the inode to create space for the increased
1300  * i_extra_isize.
1301  */
1302 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1303                                      int value_offs_shift, void *to,
1304                                      void *from, size_t n)
1305 {
1306         struct ext4_xattr_entry *last = entry;
1307         int new_offs;
1308
1309         /* We always shift xattr headers further thus offsets get lower */
1310         BUG_ON(value_offs_shift > 0);
1311
1312         /* Adjust the value offsets of the entries */
1313         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1314                 if (last->e_value_size) {
1315                         new_offs = le16_to_cpu(last->e_value_offs) +
1316                                                         value_offs_shift;
1317                         last->e_value_offs = cpu_to_le16(new_offs);
1318                 }
1319         }
1320         /* Shift the entries by n bytes */
1321         memmove(to, from, n);
1322 }
1323
1324 /*
1325  * Move xattr pointed to by 'entry' from inode into external xattr block
1326  */
1327 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
1328                                     struct ext4_inode *raw_inode,
1329                                     struct ext4_xattr_entry *entry)
1330 {
1331         struct ext4_xattr_ibody_find *is = NULL;
1332         struct ext4_xattr_block_find *bs = NULL;
1333         char *buffer = NULL, *b_entry_name = NULL;
1334         size_t value_offs, value_size;
1335         struct ext4_xattr_info i = {
1336                 .value = NULL,
1337                 .value_len = 0,
1338                 .name_index = entry->e_name_index,
1339         };
1340         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1341         int error;
1342
1343         value_offs = le16_to_cpu(entry->e_value_offs);
1344         value_size = le32_to_cpu(entry->e_value_size);
1345
1346         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1347         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1348         buffer = kmalloc(value_size, GFP_NOFS);
1349         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1350         if (!is || !bs || !buffer || !b_entry_name) {
1351                 error = -ENOMEM;
1352                 goto out;
1353         }
1354
1355         is->s.not_found = -ENODATA;
1356         bs->s.not_found = -ENODATA;
1357         is->iloc.bh = NULL;
1358         bs->bh = NULL;
1359
1360         /* Save the entry name and the entry value */
1361         memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
1362         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1363         b_entry_name[entry->e_name_len] = '\0';
1364         i.name = b_entry_name;
1365
1366         error = ext4_get_inode_loc(inode, &is->iloc);
1367         if (error)
1368                 goto out;
1369
1370         error = ext4_xattr_ibody_find(inode, &i, is);
1371         if (error)
1372                 goto out;
1373
1374         /* Remove the chosen entry from the inode */
1375         error = ext4_xattr_ibody_set(handle, inode, &i, is);
1376         if (error)
1377                 goto out;
1378
1379         i.name = b_entry_name;
1380         i.value = buffer;
1381         i.value_len = value_size;
1382         error = ext4_xattr_block_find(inode, &i, bs);
1383         if (error)
1384                 goto out;
1385
1386         /* Add entry which was removed from the inode into the block */
1387         error = ext4_xattr_block_set(handle, inode, &i, bs);
1388         if (error)
1389                 goto out;
1390         error = 0;
1391 out:
1392         kfree(b_entry_name);
1393         kfree(buffer);
1394         if (is)
1395                 brelse(is->iloc.bh);
1396         if (bs)
1397                 brelse(bs->bh);
1398         kfree(is);
1399         kfree(bs);
1400
1401         return error;
1402 }
1403
1404 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
1405                                        struct ext4_inode *raw_inode,
1406                                        int isize_diff, size_t ifree,
1407                                        size_t bfree, int *total_ino)
1408 {
1409         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1410         struct ext4_xattr_entry *small_entry;
1411         struct ext4_xattr_entry *entry;
1412         struct ext4_xattr_entry *last;
1413         unsigned int entry_size;        /* EA entry size */
1414         unsigned int total_size;        /* EA entry size + value size */
1415         unsigned int min_total_size;
1416         int error;
1417
1418         while (isize_diff > ifree) {
1419                 entry = NULL;
1420                 small_entry = NULL;
1421                 min_total_size = ~0U;
1422                 last = IFIRST(header);
1423                 /* Find the entry best suited to be pushed into EA block */
1424                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1425                         /* never move system.data out of the inode */
1426                         if ((last->e_name_len == 4) &&
1427                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
1428                             !memcmp(last->e_name, "data", 4))
1429                                 continue;
1430                         total_size =
1431                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1432                                         EXT4_XATTR_LEN(last->e_name_len);
1433                         if (total_size <= bfree &&
1434                             total_size < min_total_size) {
1435                                 if (total_size + ifree < isize_diff) {
1436                                         small_entry = last;
1437                                 } else {
1438                                         entry = last;
1439                                         min_total_size = total_size;
1440                                 }
1441                         }
1442                 }
1443
1444                 if (entry == NULL) {
1445                         if (small_entry == NULL)
1446                                 return -ENOSPC;
1447                         entry = small_entry;
1448                 }
1449
1450                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1451                 total_size = entry_size +
1452                         EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
1453                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
1454                                                  entry);
1455                 if (error)
1456                         return error;
1457
1458                 *total_ino -= entry_size;
1459                 ifree += total_size;
1460                 bfree -= total_size;
1461         }
1462
1463         return 0;
1464 }
1465
1466 /*
1467  * Expand an inode by new_extra_isize bytes when EAs are present.
1468  * Returns 0 on success or negative error number on failure.
1469  */
1470 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1471                                struct ext4_inode *raw_inode, handle_t *handle)
1472 {
1473         struct ext4_xattr_ibody_header *header;
1474         struct buffer_head *bh = NULL;
1475         size_t min_offs;
1476         size_t ifree, bfree;
1477         int total_ino;
1478         void *base, *end;
1479         int error = 0, tried_min_extra_isize = 0;
1480         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1481         int isize_diff; /* How much do we need to grow i_extra_isize */
1482         int no_expand;
1483
1484         if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
1485                 return 0;
1486
1487 retry:
1488         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
1489         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
1490                 goto out;
1491
1492         header = IHDR(inode, raw_inode);
1493
1494         /*
1495          * Check if enough free space is available in the inode to shift the
1496          * entries ahead by new_extra_isize.
1497          */
1498
1499         base = IFIRST(header);
1500         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1501         min_offs = end - base;
1502         total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
1503
1504         error = xattr_check_inode(inode, header, end);
1505         if (error)
1506                 goto cleanup;
1507
1508         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
1509         if (ifree >= isize_diff)
1510                 goto shift;
1511
1512         /*
1513          * Enough free space isn't available in the inode, check if
1514          * EA block can hold new_extra_isize bytes.
1515          */
1516         if (EXT4_I(inode)->i_file_acl) {
1517                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1518                 error = -EIO;
1519                 if (!bh)
1520                         goto cleanup;
1521                 if (ext4_xattr_check_block(inode, bh)) {
1522                         EXT4_ERROR_INODE(inode, "bad block %llu",
1523                                          EXT4_I(inode)->i_file_acl);
1524                         error = -EFSCORRUPTED;
1525                         goto cleanup;
1526                 }
1527                 base = BHDR(bh);
1528                 end = bh->b_data + bh->b_size;
1529                 min_offs = end - base;
1530                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
1531                                               NULL);
1532                 if (bfree + ifree < isize_diff) {
1533                         if (!tried_min_extra_isize && s_min_extra_isize) {
1534                                 tried_min_extra_isize++;
1535                                 new_extra_isize = s_min_extra_isize;
1536                                 brelse(bh);
1537                                 goto retry;
1538                         }
1539                         error = -ENOSPC;
1540                         goto cleanup;
1541                 }
1542         } else {
1543                 bfree = inode->i_sb->s_blocksize;
1544         }
1545
1546         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
1547                                             isize_diff, ifree, bfree,
1548                                             &total_ino);
1549         if (error) {
1550                 if (error == -ENOSPC && !tried_min_extra_isize &&
1551                     s_min_extra_isize) {
1552                         tried_min_extra_isize++;
1553                         new_extra_isize = s_min_extra_isize;
1554                         brelse(bh);
1555                         goto retry;
1556                 }
1557                 goto cleanup;
1558         }
1559 shift:
1560         /* Adjust the offsets and shift the remaining entries ahead */
1561         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
1562                         - new_extra_isize, (void *)raw_inode +
1563                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1564                         (void *)header, total_ino);
1565         EXT4_I(inode)->i_extra_isize = new_extra_isize;
1566         brelse(bh);
1567 out:
1568         ext4_write_unlock_xattr(inode, &no_expand);
1569         return 0;
1570
1571 cleanup:
1572         brelse(bh);
1573         /*
1574          * Inode size expansion failed; don't try again
1575          */
1576         no_expand = 1;
1577         ext4_write_unlock_xattr(inode, &no_expand);
1578         return error;
1579 }
1580
1581
1582
1583 /*
1584  * ext4_xattr_delete_inode()
1585  *
1586  * Free extended attribute resources associated with this inode. This
1587  * is called immediately before an inode is freed. We have exclusive
1588  * access to the inode.
1589  */
1590 void
1591 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1592 {
1593         struct buffer_head *bh = NULL;
1594
1595         if (!EXT4_I(inode)->i_file_acl)
1596                 goto cleanup;
1597         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1598         if (!bh) {
1599                 EXT4_ERROR_INODE(inode, "block %llu read error",
1600                                  EXT4_I(inode)->i_file_acl);
1601                 goto cleanup;
1602         }
1603         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1604             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1605                 EXT4_ERROR_INODE(inode, "bad block %llu",
1606                                  EXT4_I(inode)->i_file_acl);
1607                 goto cleanup;
1608         }
1609         ext4_xattr_release_block(handle, inode, bh);
1610         EXT4_I(inode)->i_file_acl = 0;
1611
1612 cleanup:
1613         brelse(bh);
1614 }
1615
1616 /*
1617  * ext4_xattr_cache_insert()
1618  *
1619  * Create a new entry in the extended attribute cache, and insert
1620  * it unless such an entry is already in the cache.
1621  *
1622  * Returns 0, or a negative error number on failure.
1623  */
1624 static void
1625 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1626 {
1627         struct ext4_xattr_header *header = BHDR(bh);
1628         __u32 hash = le32_to_cpu(header->h_hash);
1629         int reusable = le32_to_cpu(header->h_refcount) <
1630                        EXT4_XATTR_REFCOUNT_MAX;
1631         int error;
1632
1633         error = mb_cache_entry_create(ext4_mb_cache, GFP_NOFS, hash,
1634                                       bh->b_blocknr, reusable);
1635         if (error) {
1636                 if (error == -EBUSY)
1637                         ea_bdebug(bh, "already in cache");
1638         } else
1639                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1640 }
1641
1642 /*
1643  * ext4_xattr_cmp()
1644  *
1645  * Compare two extended attribute blocks for equality.
1646  *
1647  * Returns 0 if the blocks are equal, 1 if they differ, and
1648  * a negative error number on errors.
1649  */
1650 static int
1651 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1652                struct ext4_xattr_header *header2)
1653 {
1654         struct ext4_xattr_entry *entry1, *entry2;
1655
1656         entry1 = ENTRY(header1+1);
1657         entry2 = ENTRY(header2+1);
1658         while (!IS_LAST_ENTRY(entry1)) {
1659                 if (IS_LAST_ENTRY(entry2))
1660                         return 1;
1661                 if (entry1->e_hash != entry2->e_hash ||
1662                     entry1->e_name_index != entry2->e_name_index ||
1663                     entry1->e_name_len != entry2->e_name_len ||
1664                     entry1->e_value_size != entry2->e_value_size ||
1665                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1666                         return 1;
1667                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1668                         return -EFSCORRUPTED;
1669                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1670                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1671                            le32_to_cpu(entry1->e_value_size)))
1672                         return 1;
1673
1674                 entry1 = EXT4_XATTR_NEXT(entry1);
1675                 entry2 = EXT4_XATTR_NEXT(entry2);
1676         }
1677         if (!IS_LAST_ENTRY(entry2))
1678                 return 1;
1679         return 0;
1680 }
1681
1682 /*
1683  * ext4_xattr_cache_find()
1684  *
1685  * Find an identical extended attribute block.
1686  *
1687  * Returns a pointer to the block found, or NULL if such a block was
1688  * not found or an error occurred.
1689  */
1690 static struct buffer_head *
1691 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1692                       struct mb_cache_entry **pce)
1693 {
1694         __u32 hash = le32_to_cpu(header->h_hash);
1695         struct mb_cache_entry *ce;
1696         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1697
1698         if (!header->h_hash)
1699                 return NULL;  /* never share */
1700         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1701         ce = mb_cache_entry_find_first(ext4_mb_cache, hash);
1702         while (ce) {
1703                 struct buffer_head *bh;
1704
1705                 bh = sb_bread(inode->i_sb, ce->e_block);
1706                 if (!bh) {
1707                         EXT4_ERROR_INODE(inode, "block %lu read error",
1708                                          (unsigned long) ce->e_block);
1709                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1710                         *pce = ce;
1711                         return bh;
1712                 }
1713                 brelse(bh);
1714                 ce = mb_cache_entry_find_next(ext4_mb_cache, ce);
1715         }
1716         return NULL;
1717 }
1718
1719 #define NAME_HASH_SHIFT 5
1720 #define VALUE_HASH_SHIFT 16
1721
1722 /*
1723  * ext4_xattr_hash_entry()
1724  *
1725  * Compute the hash of an extended attribute.
1726  */
1727 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1728                                          struct ext4_xattr_entry *entry)
1729 {
1730         __u32 hash = 0;
1731         char *name = entry->e_name;
1732         int n;
1733
1734         for (n = 0; n < entry->e_name_len; n++) {
1735                 hash = (hash << NAME_HASH_SHIFT) ^
1736                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1737                        *name++;
1738         }
1739
1740         if (entry->e_value_size != 0) {
1741                 __le32 *value = (__le32 *)((char *)header +
1742                         le16_to_cpu(entry->e_value_offs));
1743                 for (n = (le32_to_cpu(entry->e_value_size) +
1744                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1745                         hash = (hash << VALUE_HASH_SHIFT) ^
1746                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1747                                le32_to_cpu(*value++);
1748                 }
1749         }
1750         entry->e_hash = cpu_to_le32(hash);
1751 }
1752
1753 #undef NAME_HASH_SHIFT
1754 #undef VALUE_HASH_SHIFT
1755
1756 #define BLOCK_HASH_SHIFT 16
1757
1758 /*
1759  * ext4_xattr_rehash()
1760  *
1761  * Re-compute the extended attribute hash value after an entry has changed.
1762  */
1763 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1764                               struct ext4_xattr_entry *entry)
1765 {
1766         struct ext4_xattr_entry *here;
1767         __u32 hash = 0;
1768
1769         ext4_xattr_hash_entry(header, entry);
1770         here = ENTRY(header+1);
1771         while (!IS_LAST_ENTRY(here)) {
1772                 if (!here->e_hash) {
1773                         /* Block is not shared if an entry's hash value == 0 */
1774                         hash = 0;
1775                         break;
1776                 }
1777                 hash = (hash << BLOCK_HASH_SHIFT) ^
1778                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1779                        le32_to_cpu(here->e_hash);
1780                 here = EXT4_XATTR_NEXT(here);
1781         }
1782         header->h_hash = cpu_to_le32(hash);
1783 }
1784
1785 #undef BLOCK_HASH_SHIFT
1786
1787 #define HASH_BUCKET_BITS        10
1788
1789 struct mb_cache *
1790 ext4_xattr_create_cache(void)
1791 {
1792         return mb_cache_create(HASH_BUCKET_BITS);
1793 }
1794
1795 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1796 {
1797         if (cache)
1798                 mb_cache_destroy(cache);
1799 }
1800