ext4: add shutdown bit and check for it
[linux-block.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                                         sector_t block_nr,
135                                         struct ext4_xattr_header *hdr)
136 {
137         if (ext4_has_metadata_csum(inode->i_sb) &&
138             (hdr->h_checksum != ext4_xattr_block_csum(inode, block_nr, hdr)))
139                 return 0;
140         return 1;
141 }
142
143 static void ext4_xattr_block_csum_set(struct inode *inode,
144                                       sector_t block_nr,
145                                       struct ext4_xattr_header *hdr)
146 {
147         if (!ext4_has_metadata_csum(inode->i_sb))
148                 return;
149
150         hdr->h_checksum = ext4_xattr_block_csum(inode, block_nr, hdr);
151 }
152
153 static inline int ext4_handle_dirty_xattr_block(handle_t *handle,
154                                                 struct inode *inode,
155                                                 struct buffer_head *bh)
156 {
157         ext4_xattr_block_csum_set(inode, bh->b_blocknr, BHDR(bh));
158         return ext4_handle_dirty_metadata(handle, inode, bh);
159 }
160
161 static inline const struct xattr_handler *
162 ext4_xattr_handler(int name_index)
163 {
164         const struct xattr_handler *handler = NULL;
165
166         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
167                 handler = ext4_xattr_handler_map[name_index];
168         return handler;
169 }
170
171 /*
172  * Inode operation listxattr()
173  *
174  * d_inode(dentry)->i_mutex: don't care
175  */
176 ssize_t
177 ext4_listxattr(struct dentry *dentry, char *buffer, size_t size)
178 {
179         return ext4_xattr_list(dentry, buffer, size);
180 }
181
182 static int
183 ext4_xattr_check_names(struct ext4_xattr_entry *entry, void *end,
184                        void *value_start)
185 {
186         struct ext4_xattr_entry *e = entry;
187
188         /* Find the end of the names list */
189         while (!IS_LAST_ENTRY(e)) {
190                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
191                 if ((void *)next >= end)
192                         return -EFSCORRUPTED;
193                 e = next;
194         }
195
196         /* Check the values */
197         while (!IS_LAST_ENTRY(entry)) {
198                 if (entry->e_value_block != 0)
199                         return -EFSCORRUPTED;
200                 if (entry->e_value_size != 0) {
201                         u16 offs = le16_to_cpu(entry->e_value_offs);
202                         u32 size = le32_to_cpu(entry->e_value_size);
203                         void *value;
204
205                         /*
206                          * The value cannot overlap the names, and the value
207                          * with padding cannot extend beyond 'end'.  Check both
208                          * the padded and unpadded sizes, since the size may
209                          * overflow to 0 when adding padding.
210                          */
211                         if (offs > end - value_start)
212                                 return -EFSCORRUPTED;
213                         value = value_start + offs;
214                         if (value < (void *)e + sizeof(u32) ||
215                             size > end - value ||
216                             EXT4_XATTR_SIZE(size) > end - value)
217                                 return -EFSCORRUPTED;
218                 }
219                 entry = EXT4_XATTR_NEXT(entry);
220         }
221
222         return 0;
223 }
224
225 static inline int
226 ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
227 {
228         int error;
229
230         if (buffer_verified(bh))
231                 return 0;
232
233         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
234             BHDR(bh)->h_blocks != cpu_to_le32(1))
235                 return -EFSCORRUPTED;
236         if (!ext4_xattr_block_csum_verify(inode, bh->b_blocknr, BHDR(bh)))
237                 return -EFSBADCRC;
238         error = ext4_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size,
239                                        bh->b_data);
240         if (!error)
241                 set_buffer_verified(bh);
242         return error;
243 }
244
245 static int
246 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
247                          void *end, const char *function, unsigned int line)
248 {
249         int error = -EFSCORRUPTED;
250
251         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
252             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
253                 goto errout;
254         error = ext4_xattr_check_names(IFIRST(header), end, IFIRST(header));
255 errout:
256         if (error)
257                 __ext4_error_inode(inode, function, line, 0,
258                                    "corrupted in-inode xattr");
259         return error;
260 }
261
262 #define xattr_check_inode(inode, header, end) \
263         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
264
265 static inline int
266 ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
267 {
268         size_t value_size = le32_to_cpu(entry->e_value_size);
269
270         if (entry->e_value_block != 0 || value_size > size ||
271             le16_to_cpu(entry->e_value_offs) + value_size > size)
272                 return -EFSCORRUPTED;
273         return 0;
274 }
275
276 static int
277 ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
278                       const char *name, size_t size, int sorted)
279 {
280         struct ext4_xattr_entry *entry;
281         size_t name_len;
282         int cmp = 1;
283
284         if (name == NULL)
285                 return -EINVAL;
286         name_len = strlen(name);
287         entry = *pentry;
288         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
289                 cmp = name_index - entry->e_name_index;
290                 if (!cmp)
291                         cmp = name_len - entry->e_name_len;
292                 if (!cmp)
293                         cmp = memcmp(name, entry->e_name, name_len);
294                 if (cmp <= 0 && (sorted || cmp == 0))
295                         break;
296         }
297         *pentry = entry;
298         if (!cmp && ext4_xattr_check_entry(entry, size))
299                 return -EFSCORRUPTED;
300         return cmp ? -ENODATA : 0;
301 }
302
303 static int
304 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
305                      void *buffer, size_t buffer_size)
306 {
307         struct buffer_head *bh = NULL;
308         struct ext4_xattr_entry *entry;
309         size_t size;
310         int error;
311         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
312
313         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
314                   name_index, name, buffer, (long)buffer_size);
315
316         error = -ENODATA;
317         if (!EXT4_I(inode)->i_file_acl)
318                 goto cleanup;
319         ea_idebug(inode, "reading block %llu",
320                   (unsigned long long)EXT4_I(inode)->i_file_acl);
321         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
322         if (!bh)
323                 goto cleanup;
324         ea_bdebug(bh, "b_count=%d, refcount=%d",
325                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
326         if (ext4_xattr_check_block(inode, bh)) {
327 bad_block:
328                 EXT4_ERROR_INODE(inode, "bad block %llu",
329                                  EXT4_I(inode)->i_file_acl);
330                 error = -EFSCORRUPTED;
331                 goto cleanup;
332         }
333         ext4_xattr_cache_insert(ext4_mb_cache, bh);
334         entry = BFIRST(bh);
335         error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
336         if (error == -EFSCORRUPTED)
337                 goto bad_block;
338         if (error)
339                 goto cleanup;
340         size = le32_to_cpu(entry->e_value_size);
341         if (buffer) {
342                 error = -ERANGE;
343                 if (size > buffer_size)
344                         goto cleanup;
345                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
346                        size);
347         }
348         error = size;
349
350 cleanup:
351         brelse(bh);
352         return error;
353 }
354
355 int
356 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
357                      void *buffer, size_t buffer_size)
358 {
359         struct ext4_xattr_ibody_header *header;
360         struct ext4_xattr_entry *entry;
361         struct ext4_inode *raw_inode;
362         struct ext4_iloc iloc;
363         size_t size;
364         void *end;
365         int error;
366
367         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
368                 return -ENODATA;
369         error = ext4_get_inode_loc(inode, &iloc);
370         if (error)
371                 return error;
372         raw_inode = ext4_raw_inode(&iloc);
373         header = IHDR(inode, raw_inode);
374         entry = IFIRST(header);
375         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
376         error = xattr_check_inode(inode, header, end);
377         if (error)
378                 goto cleanup;
379         error = ext4_xattr_find_entry(&entry, name_index, name,
380                                       end - (void *)entry, 0);
381         if (error)
382                 goto cleanup;
383         size = le32_to_cpu(entry->e_value_size);
384         if (buffer) {
385                 error = -ERANGE;
386                 if (size > buffer_size)
387                         goto cleanup;
388                 memcpy(buffer, (void *)IFIRST(header) +
389                        le16_to_cpu(entry->e_value_offs), size);
390         }
391         error = size;
392
393 cleanup:
394         brelse(iloc.bh);
395         return error;
396 }
397
398 /*
399  * ext4_xattr_get()
400  *
401  * Copy an extended attribute into the buffer
402  * provided, or compute the buffer size required.
403  * Buffer is NULL to compute the size of the buffer required.
404  *
405  * Returns a negative error number on failure, or the number of bytes
406  * used / required on success.
407  */
408 int
409 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
410                void *buffer, size_t buffer_size)
411 {
412         int error;
413
414         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
415                 return -EIO;
416
417         if (strlen(name) > 255)
418                 return -ERANGE;
419
420         down_read(&EXT4_I(inode)->xattr_sem);
421         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
422                                      buffer_size);
423         if (error == -ENODATA)
424                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
425                                              buffer_size);
426         up_read(&EXT4_I(inode)->xattr_sem);
427         return error;
428 }
429
430 static int
431 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
432                         char *buffer, size_t buffer_size)
433 {
434         size_t rest = buffer_size;
435
436         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
437                 const struct xattr_handler *handler =
438                         ext4_xattr_handler(entry->e_name_index);
439
440                 if (handler && (!handler->list || handler->list(dentry))) {
441                         const char *prefix = handler->prefix ?: handler->name;
442                         size_t prefix_len = strlen(prefix);
443                         size_t size = prefix_len + entry->e_name_len + 1;
444
445                         if (buffer) {
446                                 if (size > rest)
447                                         return -ERANGE;
448                                 memcpy(buffer, prefix, prefix_len);
449                                 buffer += prefix_len;
450                                 memcpy(buffer, entry->e_name, entry->e_name_len);
451                                 buffer += entry->e_name_len;
452                                 *buffer++ = 0;
453                         }
454                         rest -= size;
455                 }
456         }
457         return buffer_size - rest;  /* total size */
458 }
459
460 static int
461 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
462 {
463         struct inode *inode = d_inode(dentry);
464         struct buffer_head *bh = NULL;
465         int error;
466         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
467
468         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
469                   buffer, (long)buffer_size);
470
471         error = 0;
472         if (!EXT4_I(inode)->i_file_acl)
473                 goto cleanup;
474         ea_idebug(inode, "reading block %llu",
475                   (unsigned long long)EXT4_I(inode)->i_file_acl);
476         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
477         error = -EIO;
478         if (!bh)
479                 goto cleanup;
480         ea_bdebug(bh, "b_count=%d, refcount=%d",
481                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
482         if (ext4_xattr_check_block(inode, bh)) {
483                 EXT4_ERROR_INODE(inode, "bad block %llu",
484                                  EXT4_I(inode)->i_file_acl);
485                 error = -EFSCORRUPTED;
486                 goto cleanup;
487         }
488         ext4_xattr_cache_insert(ext4_mb_cache, bh);
489         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
490
491 cleanup:
492         brelse(bh);
493
494         return error;
495 }
496
497 static int
498 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
499 {
500         struct inode *inode = d_inode(dentry);
501         struct ext4_xattr_ibody_header *header;
502         struct ext4_inode *raw_inode;
503         struct ext4_iloc iloc;
504         void *end;
505         int error;
506
507         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
508                 return 0;
509         error = ext4_get_inode_loc(inode, &iloc);
510         if (error)
511                 return error;
512         raw_inode = ext4_raw_inode(&iloc);
513         header = IHDR(inode, raw_inode);
514         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
515         error = xattr_check_inode(inode, header, end);
516         if (error)
517                 goto cleanup;
518         error = ext4_xattr_list_entries(dentry, IFIRST(header),
519                                         buffer, buffer_size);
520
521 cleanup:
522         brelse(iloc.bh);
523         return error;
524 }
525
526 /*
527  * ext4_xattr_list()
528  *
529  * Copy a list of attribute names into the buffer
530  * provided, or compute the buffer size required.
531  * Buffer is NULL to compute the size of the buffer required.
532  *
533  * Returns a negative error number on failure, or the number of bytes
534  * used / required on success.
535  */
536 static int
537 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
538 {
539         int ret, ret2;
540
541         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
542         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
543         if (ret < 0)
544                 goto errout;
545         if (buffer) {
546                 buffer += ret;
547                 buffer_size -= ret;
548         }
549         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
550         if (ret < 0)
551                 goto errout;
552         ret += ret2;
553 errout:
554         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
555         return ret;
556 }
557
558 /*
559  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
560  * not set, set it.
561  */
562 static void ext4_xattr_update_super_block(handle_t *handle,
563                                           struct super_block *sb)
564 {
565         if (ext4_has_feature_xattr(sb))
566                 return;
567
568         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
569         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
570                 ext4_set_feature_xattr(sb);
571                 ext4_handle_dirty_super(handle, sb);
572         }
573 }
574
575 /*
576  * Release the xattr block BH: If the reference count is > 1, decrement it;
577  * otherwise free the block.
578  */
579 static void
580 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
581                          struct buffer_head *bh)
582 {
583         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
584         u32 hash, ref;
585         int error = 0;
586
587         BUFFER_TRACE(bh, "get_write_access");
588         error = ext4_journal_get_write_access(handle, bh);
589         if (error)
590                 goto out;
591
592         lock_buffer(bh);
593         hash = le32_to_cpu(BHDR(bh)->h_hash);
594         ref = le32_to_cpu(BHDR(bh)->h_refcount);
595         if (ref == 1) {
596                 ea_bdebug(bh, "refcount now=0; freeing");
597                 /*
598                  * This must happen under buffer lock for
599                  * ext4_xattr_block_set() to reliably detect freed block
600                  */
601                 mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
602                 get_bh(bh);
603                 unlock_buffer(bh);
604                 ext4_free_blocks(handle, inode, bh, 0, 1,
605                                  EXT4_FREE_BLOCKS_METADATA |
606                                  EXT4_FREE_BLOCKS_FORGET);
607         } else {
608                 ref--;
609                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
610                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
611                         struct mb_cache_entry *ce;
612
613                         ce = mb_cache_entry_get(ext4_mb_cache, hash,
614                                                 bh->b_blocknr);
615                         if (ce) {
616                                 ce->e_reusable = 1;
617                                 mb_cache_entry_put(ext4_mb_cache, ce);
618                         }
619                 }
620
621                 /*
622                  * Beware of this ugliness: Releasing of xattr block references
623                  * from different inodes can race and so we have to protect
624                  * from a race where someone else frees the block (and releases
625                  * its journal_head) before we are done dirtying the buffer. In
626                  * nojournal mode this race is harmless and we actually cannot
627                  * call ext4_handle_dirty_xattr_block() with locked buffer as
628                  * that function can call sync_dirty_buffer() so for that case
629                  * we handle the dirtying after unlocking the buffer.
630                  */
631                 if (ext4_handle_valid(handle))
632                         error = ext4_handle_dirty_xattr_block(handle, inode,
633                                                               bh);
634                 unlock_buffer(bh);
635                 if (!ext4_handle_valid(handle))
636                         error = ext4_handle_dirty_xattr_block(handle, inode,
637                                                               bh);
638                 if (IS_SYNC(inode))
639                         ext4_handle_sync(handle);
640                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
641                 ea_bdebug(bh, "refcount now=%d; releasing",
642                           le32_to_cpu(BHDR(bh)->h_refcount));
643         }
644 out:
645         ext4_std_error(inode->i_sb, error);
646         return;
647 }
648
649 /*
650  * Find the available free space for EAs. This also returns the total number of
651  * bytes used by EA entries.
652  */
653 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
654                                     size_t *min_offs, void *base, int *total)
655 {
656         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
657                 if (last->e_value_size) {
658                         size_t offs = le16_to_cpu(last->e_value_offs);
659                         if (offs < *min_offs)
660                                 *min_offs = offs;
661                 }
662                 if (total)
663                         *total += EXT4_XATTR_LEN(last->e_name_len);
664         }
665         return (*min_offs - ((void *)last - base) - sizeof(__u32));
666 }
667
668 static int
669 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
670 {
671         struct ext4_xattr_entry *last;
672         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
673
674         /* Compute min_offs and last. */
675         last = s->first;
676         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
677                 if (last->e_value_size) {
678                         size_t offs = le16_to_cpu(last->e_value_offs);
679                         if (offs < min_offs)
680                                 min_offs = offs;
681                 }
682         }
683         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
684         if (!s->not_found) {
685                 if (s->here->e_value_size) {
686                         size_t size = le32_to_cpu(s->here->e_value_size);
687                         free += EXT4_XATTR_SIZE(size);
688                 }
689                 free += EXT4_XATTR_LEN(name_len);
690         }
691         if (i->value) {
692                 if (free < EXT4_XATTR_LEN(name_len) +
693                            EXT4_XATTR_SIZE(i->value_len))
694                         return -ENOSPC;
695         }
696
697         if (i->value && s->not_found) {
698                 /* Insert the new name. */
699                 size_t size = EXT4_XATTR_LEN(name_len);
700                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
701                 memmove((void *)s->here + size, s->here, rest);
702                 memset(s->here, 0, size);
703                 s->here->e_name_index = i->name_index;
704                 s->here->e_name_len = name_len;
705                 memcpy(s->here->e_name, i->name, name_len);
706         } else {
707                 if (s->here->e_value_size) {
708                         void *first_val = s->base + min_offs;
709                         size_t offs = le16_to_cpu(s->here->e_value_offs);
710                         void *val = s->base + offs;
711                         size_t size = EXT4_XATTR_SIZE(
712                                 le32_to_cpu(s->here->e_value_size));
713
714                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
715                                 /* The old and the new value have the same
716                                    size. Just replace. */
717                                 s->here->e_value_size =
718                                         cpu_to_le32(i->value_len);
719                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
720                                         memset(val, 0, size);
721                                 } else {
722                                         /* Clear pad bytes first. */
723                                         memset(val + size - EXT4_XATTR_PAD, 0,
724                                                EXT4_XATTR_PAD);
725                                         memcpy(val, i->value, i->value_len);
726                                 }
727                                 return 0;
728                         }
729
730                         /* Remove the old value. */
731                         memmove(first_val + size, first_val, val - first_val);
732                         memset(first_val, 0, size);
733                         s->here->e_value_size = 0;
734                         s->here->e_value_offs = 0;
735                         min_offs += size;
736
737                         /* Adjust all value offsets. */
738                         last = s->first;
739                         while (!IS_LAST_ENTRY(last)) {
740                                 size_t o = le16_to_cpu(last->e_value_offs);
741                                 if (last->e_value_size && o < offs)
742                                         last->e_value_offs =
743                                                 cpu_to_le16(o + size);
744                                 last = EXT4_XATTR_NEXT(last);
745                         }
746                 }
747                 if (!i->value) {
748                         /* Remove the old name. */
749                         size_t size = EXT4_XATTR_LEN(name_len);
750                         last = ENTRY((void *)last - size);
751                         memmove(s->here, (void *)s->here + size,
752                                 (void *)last - (void *)s->here + sizeof(__u32));
753                         memset(last, 0, size);
754                 }
755         }
756
757         if (i->value) {
758                 /* Insert the new value. */
759                 s->here->e_value_size = cpu_to_le32(i->value_len);
760                 if (i->value_len) {
761                         size_t size = EXT4_XATTR_SIZE(i->value_len);
762                         void *val = s->base + min_offs - size;
763                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
764                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
765                                 memset(val, 0, size);
766                         } else {
767                                 /* Clear the pad bytes first. */
768                                 memset(val + size - EXT4_XATTR_PAD, 0,
769                                        EXT4_XATTR_PAD);
770                                 memcpy(val, i->value, i->value_len);
771                         }
772                 }
773         }
774         return 0;
775 }
776
777 struct ext4_xattr_block_find {
778         struct ext4_xattr_search s;
779         struct buffer_head *bh;
780 };
781
782 static int
783 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
784                       struct ext4_xattr_block_find *bs)
785 {
786         struct super_block *sb = inode->i_sb;
787         int error;
788
789         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
790                   i->name_index, i->name, i->value, (long)i->value_len);
791
792         if (EXT4_I(inode)->i_file_acl) {
793                 /* The inode already has an extended attribute block. */
794                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
795                 error = -EIO;
796                 if (!bs->bh)
797                         goto cleanup;
798                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
799                         atomic_read(&(bs->bh->b_count)),
800                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
801                 if (ext4_xattr_check_block(inode, bs->bh)) {
802                         EXT4_ERROR_INODE(inode, "bad block %llu",
803                                          EXT4_I(inode)->i_file_acl);
804                         error = -EFSCORRUPTED;
805                         goto cleanup;
806                 }
807                 /* Find the named attribute. */
808                 bs->s.base = BHDR(bs->bh);
809                 bs->s.first = BFIRST(bs->bh);
810                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
811                 bs->s.here = bs->s.first;
812                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
813                                               i->name, bs->bh->b_size, 1);
814                 if (error && error != -ENODATA)
815                         goto cleanup;
816                 bs->s.not_found = error;
817         }
818         error = 0;
819
820 cleanup:
821         return error;
822 }
823
824 static int
825 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
826                      struct ext4_xattr_info *i,
827                      struct ext4_xattr_block_find *bs)
828 {
829         struct super_block *sb = inode->i_sb;
830         struct buffer_head *new_bh = NULL;
831         struct ext4_xattr_search *s = &bs->s;
832         struct mb_cache_entry *ce = NULL;
833         int error = 0;
834         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
835
836 #define header(x) ((struct ext4_xattr_header *)(x))
837
838         if (i->value && i->value_len > sb->s_blocksize)
839                 return -ENOSPC;
840         if (s->base) {
841                 BUFFER_TRACE(bs->bh, "get_write_access");
842                 error = ext4_journal_get_write_access(handle, bs->bh);
843                 if (error)
844                         goto cleanup;
845                 lock_buffer(bs->bh);
846
847                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
848                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
849
850                         /*
851                          * This must happen under buffer lock for
852                          * ext4_xattr_block_set() to reliably detect modified
853                          * block
854                          */
855                         mb_cache_entry_delete_block(ext4_mb_cache, hash,
856                                                     bs->bh->b_blocknr);
857                         ea_bdebug(bs->bh, "modifying in-place");
858                         error = ext4_xattr_set_entry(i, s);
859                         if (!error) {
860                                 if (!IS_LAST_ENTRY(s->first))
861                                         ext4_xattr_rehash(header(s->base),
862                                                           s->here);
863                                 ext4_xattr_cache_insert(ext4_mb_cache,
864                                         bs->bh);
865                         }
866                         unlock_buffer(bs->bh);
867                         if (error == -EFSCORRUPTED)
868                                 goto bad_block;
869                         if (!error)
870                                 error = ext4_handle_dirty_xattr_block(handle,
871                                                                       inode,
872                                                                       bs->bh);
873                         if (error)
874                                 goto cleanup;
875                         goto inserted;
876                 } else {
877                         int offset = (char *)s->here - bs->bh->b_data;
878
879                         unlock_buffer(bs->bh);
880                         ea_bdebug(bs->bh, "cloning");
881                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
882                         error = -ENOMEM;
883                         if (s->base == NULL)
884                                 goto cleanup;
885                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
886                         s->first = ENTRY(header(s->base)+1);
887                         header(s->base)->h_refcount = cpu_to_le32(1);
888                         s->here = ENTRY(s->base + offset);
889                         s->end = s->base + bs->bh->b_size;
890                 }
891         } else {
892                 /* Allocate a buffer where we construct the new block. */
893                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
894                 /* assert(header == s->base) */
895                 error = -ENOMEM;
896                 if (s->base == NULL)
897                         goto cleanup;
898                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
899                 header(s->base)->h_blocks = cpu_to_le32(1);
900                 header(s->base)->h_refcount = cpu_to_le32(1);
901                 s->first = ENTRY(header(s->base)+1);
902                 s->here = ENTRY(header(s->base)+1);
903                 s->end = s->base + sb->s_blocksize;
904         }
905
906         error = ext4_xattr_set_entry(i, s);
907         if (error == -EFSCORRUPTED)
908                 goto bad_block;
909         if (error)
910                 goto cleanup;
911         if (!IS_LAST_ENTRY(s->first))
912                 ext4_xattr_rehash(header(s->base), s->here);
913
914 inserted:
915         if (!IS_LAST_ENTRY(s->first)) {
916                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
917                 if (new_bh) {
918                         /* We found an identical block in the cache. */
919                         if (new_bh == bs->bh)
920                                 ea_bdebug(new_bh, "keeping");
921                         else {
922                                 u32 ref;
923
924                                 /* The old block is released after updating
925                                    the inode. */
926                                 error = dquot_alloc_block(inode,
927                                                 EXT4_C2B(EXT4_SB(sb), 1));
928                                 if (error)
929                                         goto cleanup;
930                                 BUFFER_TRACE(new_bh, "get_write_access");
931                                 error = ext4_journal_get_write_access(handle,
932                                                                       new_bh);
933                                 if (error)
934                                         goto cleanup_dquot;
935                                 lock_buffer(new_bh);
936                                 /*
937                                  * We have to be careful about races with
938                                  * freeing, rehashing or adding references to
939                                  * xattr block. Once we hold buffer lock xattr
940                                  * block's state is stable so we can check
941                                  * whether the block got freed / rehashed or
942                                  * not.  Since we unhash mbcache entry under
943                                  * buffer lock when freeing / rehashing xattr
944                                  * block, checking whether entry is still
945                                  * hashed is reliable. Same rules hold for
946                                  * e_reusable handling.
947                                  */
948                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
949                                     !ce->e_reusable) {
950                                         /*
951                                          * Undo everything and check mbcache
952                                          * again.
953                                          */
954                                         unlock_buffer(new_bh);
955                                         dquot_free_block(inode,
956                                                          EXT4_C2B(EXT4_SB(sb),
957                                                                   1));
958                                         brelse(new_bh);
959                                         mb_cache_entry_put(ext4_mb_cache, ce);
960                                         ce = NULL;
961                                         new_bh = NULL;
962                                         goto inserted;
963                                 }
964                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
965                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
966                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
967                                         ce->e_reusable = 0;
968                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
969                                           ref);
970                                 unlock_buffer(new_bh);
971                                 error = ext4_handle_dirty_xattr_block(handle,
972                                                                       inode,
973                                                                       new_bh);
974                                 if (error)
975                                         goto cleanup_dquot;
976                         }
977                         mb_cache_entry_touch(ext4_mb_cache, ce);
978                         mb_cache_entry_put(ext4_mb_cache, ce);
979                         ce = NULL;
980                 } else if (bs->bh && s->base == bs->bh->b_data) {
981                         /* We were modifying this block in-place. */
982                         ea_bdebug(bs->bh, "keeping this block");
983                         new_bh = bs->bh;
984                         get_bh(new_bh);
985                 } else {
986                         /* We need to allocate a new block */
987                         ext4_fsblk_t goal, block;
988
989                         goal = ext4_group_first_block_no(sb,
990                                                 EXT4_I(inode)->i_block_group);
991
992                         /* non-extent files can't have physical blocks past 2^32 */
993                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
994                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
995
996                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
997                                                      NULL, &error);
998                         if (error)
999                                 goto cleanup;
1000
1001                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
1002                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
1003
1004                         ea_idebug(inode, "creating block %llu",
1005                                   (unsigned long long)block);
1006
1007                         new_bh = sb_getblk(sb, block);
1008                         if (unlikely(!new_bh)) {
1009                                 error = -ENOMEM;
1010 getblk_failed:
1011                                 ext4_free_blocks(handle, inode, NULL, block, 1,
1012                                                  EXT4_FREE_BLOCKS_METADATA);
1013                                 goto cleanup;
1014                         }
1015                         lock_buffer(new_bh);
1016                         error = ext4_journal_get_create_access(handle, new_bh);
1017                         if (error) {
1018                                 unlock_buffer(new_bh);
1019                                 error = -EIO;
1020                                 goto getblk_failed;
1021                         }
1022                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
1023                         set_buffer_uptodate(new_bh);
1024                         unlock_buffer(new_bh);
1025                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
1026                         error = ext4_handle_dirty_xattr_block(handle,
1027                                                               inode, new_bh);
1028                         if (error)
1029                                 goto cleanup;
1030                 }
1031         }
1032
1033         /* Update the inode. */
1034         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
1035
1036         /* Drop the previous xattr block. */
1037         if (bs->bh && bs->bh != new_bh)
1038                 ext4_xattr_release_block(handle, inode, bs->bh);
1039         error = 0;
1040
1041 cleanup:
1042         if (ce)
1043                 mb_cache_entry_put(ext4_mb_cache, ce);
1044         brelse(new_bh);
1045         if (!(bs->bh && s->base == bs->bh->b_data))
1046                 kfree(s->base);
1047
1048         return error;
1049
1050 cleanup_dquot:
1051         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
1052         goto cleanup;
1053
1054 bad_block:
1055         EXT4_ERROR_INODE(inode, "bad block %llu",
1056                          EXT4_I(inode)->i_file_acl);
1057         goto cleanup;
1058
1059 #undef header
1060 }
1061
1062 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1063                           struct ext4_xattr_ibody_find *is)
1064 {
1065         struct ext4_xattr_ibody_header *header;
1066         struct ext4_inode *raw_inode;
1067         int error;
1068
1069         if (EXT4_I(inode)->i_extra_isize == 0)
1070                 return 0;
1071         raw_inode = ext4_raw_inode(&is->iloc);
1072         header = IHDR(inode, raw_inode);
1073         is->s.base = is->s.first = IFIRST(header);
1074         is->s.here = is->s.first;
1075         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1076         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1077                 error = xattr_check_inode(inode, header, is->s.end);
1078                 if (error)
1079                         return error;
1080                 /* Find the named attribute. */
1081                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1082                                               i->name, is->s.end -
1083                                               (void *)is->s.base, 0);
1084                 if (error && error != -ENODATA)
1085                         return error;
1086                 is->s.not_found = error;
1087         }
1088         return 0;
1089 }
1090
1091 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1092                                 struct ext4_xattr_info *i,
1093                                 struct ext4_xattr_ibody_find *is)
1094 {
1095         struct ext4_xattr_ibody_header *header;
1096         struct ext4_xattr_search *s = &is->s;
1097         int error;
1098
1099         if (EXT4_I(inode)->i_extra_isize == 0)
1100                 return -ENOSPC;
1101         error = ext4_xattr_set_entry(i, s);
1102         if (error) {
1103                 if (error == -ENOSPC &&
1104                     ext4_has_inline_data(inode)) {
1105                         error = ext4_try_to_evict_inline_data(handle, inode,
1106                                         EXT4_XATTR_LEN(strlen(i->name) +
1107                                         EXT4_XATTR_SIZE(i->value_len)));
1108                         if (error)
1109                                 return error;
1110                         error = ext4_xattr_ibody_find(inode, i, is);
1111                         if (error)
1112                                 return error;
1113                         error = ext4_xattr_set_entry(i, s);
1114                 }
1115                 if (error)
1116                         return error;
1117         }
1118         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1119         if (!IS_LAST_ENTRY(s->first)) {
1120                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1121                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1122         } else {
1123                 header->h_magic = cpu_to_le32(0);
1124                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1125         }
1126         return 0;
1127 }
1128
1129 static int ext4_xattr_ibody_set(struct inode *inode,
1130                                 struct ext4_xattr_info *i,
1131                                 struct ext4_xattr_ibody_find *is)
1132 {
1133         struct ext4_xattr_ibody_header *header;
1134         struct ext4_xattr_search *s = &is->s;
1135         int error;
1136
1137         if (EXT4_I(inode)->i_extra_isize == 0)
1138                 return -ENOSPC;
1139         error = ext4_xattr_set_entry(i, s);
1140         if (error)
1141                 return error;
1142         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1143         if (!IS_LAST_ENTRY(s->first)) {
1144                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1145                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1146         } else {
1147                 header->h_magic = cpu_to_le32(0);
1148                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1149         }
1150         return 0;
1151 }
1152
1153 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
1154                                  struct ext4_xattr_info *i)
1155 {
1156         void *value;
1157
1158         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
1159                 return 0;
1160         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
1161         return !memcmp(value, i->value, i->value_len);
1162 }
1163
1164 /*
1165  * ext4_xattr_set_handle()
1166  *
1167  * Create, replace or remove an extended attribute for this inode.  Value
1168  * is NULL to remove an existing extended attribute, and non-NULL to
1169  * either replace an existing extended attribute, or create a new extended
1170  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1171  * specify that an extended attribute must exist and must not exist
1172  * previous to the call, respectively.
1173  *
1174  * Returns 0, or a negative error number on failure.
1175  */
1176 int
1177 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1178                       const char *name, const void *value, size_t value_len,
1179                       int flags)
1180 {
1181         struct ext4_xattr_info i = {
1182                 .name_index = name_index,
1183                 .name = name,
1184                 .value = value,
1185                 .value_len = value_len,
1186
1187         };
1188         struct ext4_xattr_ibody_find is = {
1189                 .s = { .not_found = -ENODATA, },
1190         };
1191         struct ext4_xattr_block_find bs = {
1192                 .s = { .not_found = -ENODATA, },
1193         };
1194         int no_expand;
1195         int error;
1196
1197         if (!name)
1198                 return -EINVAL;
1199         if (strlen(name) > 255)
1200                 return -ERANGE;
1201         ext4_write_lock_xattr(inode, &no_expand);
1202
1203         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1204         if (error)
1205                 goto cleanup;
1206
1207         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1208                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1209                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1210                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1211         }
1212
1213         error = ext4_xattr_ibody_find(inode, &i, &is);
1214         if (error)
1215                 goto cleanup;
1216         if (is.s.not_found)
1217                 error = ext4_xattr_block_find(inode, &i, &bs);
1218         if (error)
1219                 goto cleanup;
1220         if (is.s.not_found && bs.s.not_found) {
1221                 error = -ENODATA;
1222                 if (flags & XATTR_REPLACE)
1223                         goto cleanup;
1224                 error = 0;
1225                 if (!value)
1226                         goto cleanup;
1227         } else {
1228                 error = -EEXIST;
1229                 if (flags & XATTR_CREATE)
1230                         goto cleanup;
1231         }
1232         if (!value) {
1233                 if (!is.s.not_found)
1234                         error = ext4_xattr_ibody_set(inode, &i, &is);
1235                 else if (!bs.s.not_found)
1236                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1237         } else {
1238                 error = 0;
1239                 /* Xattr value did not change? Save us some work and bail out */
1240                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
1241                         goto cleanup;
1242                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
1243                         goto cleanup;
1244
1245                 error = ext4_xattr_ibody_set(inode, &i, &is);
1246                 if (!error && !bs.s.not_found) {
1247                         i.value = NULL;
1248                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1249                 } else if (error == -ENOSPC) {
1250                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1251                                 error = ext4_xattr_block_find(inode, &i, &bs);
1252                                 if (error)
1253                                         goto cleanup;
1254                         }
1255                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1256                         if (error)
1257                                 goto cleanup;
1258                         if (!is.s.not_found) {
1259                                 i.value = NULL;
1260                                 error = ext4_xattr_ibody_set(inode, &i, &is);
1261                         }
1262                 }
1263         }
1264         if (!error) {
1265                 ext4_xattr_update_super_block(handle, inode->i_sb);
1266                 inode->i_ctime = current_time(inode);
1267                 if (!value)
1268                         no_expand = 0;
1269                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1270                 /*
1271                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1272                  * error != 0.
1273                  */
1274                 is.iloc.bh = NULL;
1275                 if (IS_SYNC(inode))
1276                         ext4_handle_sync(handle);
1277         }
1278
1279 cleanup:
1280         brelse(is.iloc.bh);
1281         brelse(bs.bh);
1282         ext4_write_unlock_xattr(inode, &no_expand);
1283         return error;
1284 }
1285
1286 /*
1287  * ext4_xattr_set()
1288  *
1289  * Like ext4_xattr_set_handle, but start from an inode. This extended
1290  * attribute modification is a filesystem transaction by itself.
1291  *
1292  * Returns 0, or a negative error number on failure.
1293  */
1294 int
1295 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1296                const void *value, size_t value_len, int flags)
1297 {
1298         handle_t *handle;
1299         int error, retries = 0;
1300         int credits = ext4_jbd2_credits_xattr(inode);
1301
1302 retry:
1303         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1304         if (IS_ERR(handle)) {
1305                 error = PTR_ERR(handle);
1306         } else {
1307                 int error2;
1308
1309                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1310                                               value, value_len, flags);
1311                 error2 = ext4_journal_stop(handle);
1312                 if (error == -ENOSPC &&
1313                     ext4_should_retry_alloc(inode->i_sb, &retries))
1314                         goto retry;
1315                 if (error == 0)
1316                         error = error2;
1317         }
1318
1319         return error;
1320 }
1321
1322 /*
1323  * Shift the EA entries in the inode to create space for the increased
1324  * i_extra_isize.
1325  */
1326 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1327                                      int value_offs_shift, void *to,
1328                                      void *from, size_t n)
1329 {
1330         struct ext4_xattr_entry *last = entry;
1331         int new_offs;
1332
1333         /* We always shift xattr headers further thus offsets get lower */
1334         BUG_ON(value_offs_shift > 0);
1335
1336         /* Adjust the value offsets of the entries */
1337         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1338                 if (last->e_value_size) {
1339                         new_offs = le16_to_cpu(last->e_value_offs) +
1340                                                         value_offs_shift;
1341                         last->e_value_offs = cpu_to_le16(new_offs);
1342                 }
1343         }
1344         /* Shift the entries by n bytes */
1345         memmove(to, from, n);
1346 }
1347
1348 /*
1349  * Move xattr pointed to by 'entry' from inode into external xattr block
1350  */
1351 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
1352                                     struct ext4_inode *raw_inode,
1353                                     struct ext4_xattr_entry *entry)
1354 {
1355         struct ext4_xattr_ibody_find *is = NULL;
1356         struct ext4_xattr_block_find *bs = NULL;
1357         char *buffer = NULL, *b_entry_name = NULL;
1358         size_t value_offs, value_size;
1359         struct ext4_xattr_info i = {
1360                 .value = NULL,
1361                 .value_len = 0,
1362                 .name_index = entry->e_name_index,
1363         };
1364         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1365         int error;
1366
1367         value_offs = le16_to_cpu(entry->e_value_offs);
1368         value_size = le32_to_cpu(entry->e_value_size);
1369
1370         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1371         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1372         buffer = kmalloc(value_size, GFP_NOFS);
1373         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1374         if (!is || !bs || !buffer || !b_entry_name) {
1375                 error = -ENOMEM;
1376                 goto out;
1377         }
1378
1379         is->s.not_found = -ENODATA;
1380         bs->s.not_found = -ENODATA;
1381         is->iloc.bh = NULL;
1382         bs->bh = NULL;
1383
1384         /* Save the entry name and the entry value */
1385         memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
1386         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1387         b_entry_name[entry->e_name_len] = '\0';
1388         i.name = b_entry_name;
1389
1390         error = ext4_get_inode_loc(inode, &is->iloc);
1391         if (error)
1392                 goto out;
1393
1394         error = ext4_xattr_ibody_find(inode, &i, is);
1395         if (error)
1396                 goto out;
1397
1398         /* Remove the chosen entry from the inode */
1399         error = ext4_xattr_ibody_set(inode, &i, is);
1400         if (error)
1401                 goto out;
1402
1403         i.name = b_entry_name;
1404         i.value = buffer;
1405         i.value_len = value_size;
1406         error = ext4_xattr_block_find(inode, &i, bs);
1407         if (error)
1408                 goto out;
1409
1410         /* Add entry which was removed from the inode into the block */
1411         error = ext4_xattr_block_set(handle, inode, &i, bs);
1412         if (error)
1413                 goto out;
1414         error = 0;
1415 out:
1416         kfree(b_entry_name);
1417         kfree(buffer);
1418         if (is)
1419                 brelse(is->iloc.bh);
1420         kfree(is);
1421         kfree(bs);
1422
1423         return error;
1424 }
1425
1426 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
1427                                        struct ext4_inode *raw_inode,
1428                                        int isize_diff, size_t ifree,
1429                                        size_t bfree, int *total_ino)
1430 {
1431         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1432         struct ext4_xattr_entry *small_entry;
1433         struct ext4_xattr_entry *entry;
1434         struct ext4_xattr_entry *last;
1435         unsigned int entry_size;        /* EA entry size */
1436         unsigned int total_size;        /* EA entry size + value size */
1437         unsigned int min_total_size;
1438         int error;
1439
1440         while (isize_diff > ifree) {
1441                 entry = NULL;
1442                 small_entry = NULL;
1443                 min_total_size = ~0U;
1444                 last = IFIRST(header);
1445                 /* Find the entry best suited to be pushed into EA block */
1446                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1447                         total_size =
1448                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1449                                         EXT4_XATTR_LEN(last->e_name_len);
1450                         if (total_size <= bfree &&
1451                             total_size < min_total_size) {
1452                                 if (total_size + ifree < isize_diff) {
1453                                         small_entry = last;
1454                                 } else {
1455                                         entry = last;
1456                                         min_total_size = total_size;
1457                                 }
1458                         }
1459                 }
1460
1461                 if (entry == NULL) {
1462                         if (small_entry == NULL)
1463                                 return -ENOSPC;
1464                         entry = small_entry;
1465                 }
1466
1467                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1468                 total_size = entry_size +
1469                         EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
1470                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
1471                                                  entry);
1472                 if (error)
1473                         return error;
1474
1475                 *total_ino -= entry_size;
1476                 ifree += total_size;
1477                 bfree -= total_size;
1478         }
1479
1480         return 0;
1481 }
1482
1483 /*
1484  * Expand an inode by new_extra_isize bytes when EAs are present.
1485  * Returns 0 on success or negative error number on failure.
1486  */
1487 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1488                                struct ext4_inode *raw_inode, handle_t *handle)
1489 {
1490         struct ext4_xattr_ibody_header *header;
1491         struct buffer_head *bh = NULL;
1492         size_t min_offs;
1493         size_t ifree, bfree;
1494         int total_ino;
1495         void *base, *end;
1496         int error = 0, tried_min_extra_isize = 0;
1497         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1498         int isize_diff; /* How much do we need to grow i_extra_isize */
1499         int no_expand;
1500
1501         if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
1502                 return 0;
1503
1504 retry:
1505         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
1506         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
1507                 goto out;
1508
1509         header = IHDR(inode, raw_inode);
1510
1511         /*
1512          * Check if enough free space is available in the inode to shift the
1513          * entries ahead by new_extra_isize.
1514          */
1515
1516         base = IFIRST(header);
1517         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1518         min_offs = end - base;
1519         total_ino = sizeof(struct ext4_xattr_ibody_header);
1520
1521         error = xattr_check_inode(inode, header, end);
1522         if (error)
1523                 goto cleanup;
1524
1525         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
1526         if (ifree >= isize_diff)
1527                 goto shift;
1528
1529         /*
1530          * Enough free space isn't available in the inode, check if
1531          * EA block can hold new_extra_isize bytes.
1532          */
1533         if (EXT4_I(inode)->i_file_acl) {
1534                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1535                 error = -EIO;
1536                 if (!bh)
1537                         goto cleanup;
1538                 if (ext4_xattr_check_block(inode, bh)) {
1539                         EXT4_ERROR_INODE(inode, "bad block %llu",
1540                                          EXT4_I(inode)->i_file_acl);
1541                         error = -EFSCORRUPTED;
1542                         goto cleanup;
1543                 }
1544                 base = BHDR(bh);
1545                 end = bh->b_data + bh->b_size;
1546                 min_offs = end - base;
1547                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
1548                                               NULL);
1549                 if (bfree + ifree < isize_diff) {
1550                         if (!tried_min_extra_isize && s_min_extra_isize) {
1551                                 tried_min_extra_isize++;
1552                                 new_extra_isize = s_min_extra_isize;
1553                                 brelse(bh);
1554                                 goto retry;
1555                         }
1556                         error = -ENOSPC;
1557                         goto cleanup;
1558                 }
1559         } else {
1560                 bfree = inode->i_sb->s_blocksize;
1561         }
1562
1563         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
1564                                             isize_diff, ifree, bfree,
1565                                             &total_ino);
1566         if (error) {
1567                 if (error == -ENOSPC && !tried_min_extra_isize &&
1568                     s_min_extra_isize) {
1569                         tried_min_extra_isize++;
1570                         new_extra_isize = s_min_extra_isize;
1571                         brelse(bh);
1572                         goto retry;
1573                 }
1574                 goto cleanup;
1575         }
1576 shift:
1577         /* Adjust the offsets and shift the remaining entries ahead */
1578         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
1579                         - new_extra_isize, (void *)raw_inode +
1580                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1581                         (void *)header, total_ino);
1582         EXT4_I(inode)->i_extra_isize = new_extra_isize;
1583         brelse(bh);
1584 out:
1585         ext4_write_unlock_xattr(inode, &no_expand);
1586         return 0;
1587
1588 cleanup:
1589         brelse(bh);
1590         /*
1591          * Inode size expansion failed; don't try again
1592          */
1593         no_expand = 1;
1594         ext4_write_unlock_xattr(inode, &no_expand);
1595         return error;
1596 }
1597
1598
1599
1600 /*
1601  * ext4_xattr_delete_inode()
1602  *
1603  * Free extended attribute resources associated with this inode. This
1604  * is called immediately before an inode is freed. We have exclusive
1605  * access to the inode.
1606  */
1607 void
1608 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1609 {
1610         struct buffer_head *bh = NULL;
1611
1612         if (!EXT4_I(inode)->i_file_acl)
1613                 goto cleanup;
1614         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1615         if (!bh) {
1616                 EXT4_ERROR_INODE(inode, "block %llu read error",
1617                                  EXT4_I(inode)->i_file_acl);
1618                 goto cleanup;
1619         }
1620         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1621             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1622                 EXT4_ERROR_INODE(inode, "bad block %llu",
1623                                  EXT4_I(inode)->i_file_acl);
1624                 goto cleanup;
1625         }
1626         ext4_xattr_release_block(handle, inode, bh);
1627         EXT4_I(inode)->i_file_acl = 0;
1628
1629 cleanup:
1630         brelse(bh);
1631 }
1632
1633 /*
1634  * ext4_xattr_cache_insert()
1635  *
1636  * Create a new entry in the extended attribute cache, and insert
1637  * it unless such an entry is already in the cache.
1638  *
1639  * Returns 0, or a negative error number on failure.
1640  */
1641 static void
1642 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1643 {
1644         struct ext4_xattr_header *header = BHDR(bh);
1645         __u32 hash = le32_to_cpu(header->h_hash);
1646         int reusable = le32_to_cpu(header->h_refcount) <
1647                        EXT4_XATTR_REFCOUNT_MAX;
1648         int error;
1649
1650         error = mb_cache_entry_create(ext4_mb_cache, GFP_NOFS, hash,
1651                                       bh->b_blocknr, reusable);
1652         if (error) {
1653                 if (error == -EBUSY)
1654                         ea_bdebug(bh, "already in cache");
1655         } else
1656                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1657 }
1658
1659 /*
1660  * ext4_xattr_cmp()
1661  *
1662  * Compare two extended attribute blocks for equality.
1663  *
1664  * Returns 0 if the blocks are equal, 1 if they differ, and
1665  * a negative error number on errors.
1666  */
1667 static int
1668 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1669                struct ext4_xattr_header *header2)
1670 {
1671         struct ext4_xattr_entry *entry1, *entry2;
1672
1673         entry1 = ENTRY(header1+1);
1674         entry2 = ENTRY(header2+1);
1675         while (!IS_LAST_ENTRY(entry1)) {
1676                 if (IS_LAST_ENTRY(entry2))
1677                         return 1;
1678                 if (entry1->e_hash != entry2->e_hash ||
1679                     entry1->e_name_index != entry2->e_name_index ||
1680                     entry1->e_name_len != entry2->e_name_len ||
1681                     entry1->e_value_size != entry2->e_value_size ||
1682                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1683                         return 1;
1684                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1685                         return -EFSCORRUPTED;
1686                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1687                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1688                            le32_to_cpu(entry1->e_value_size)))
1689                         return 1;
1690
1691                 entry1 = EXT4_XATTR_NEXT(entry1);
1692                 entry2 = EXT4_XATTR_NEXT(entry2);
1693         }
1694         if (!IS_LAST_ENTRY(entry2))
1695                 return 1;
1696         return 0;
1697 }
1698
1699 /*
1700  * ext4_xattr_cache_find()
1701  *
1702  * Find an identical extended attribute block.
1703  *
1704  * Returns a pointer to the block found, or NULL if such a block was
1705  * not found or an error occurred.
1706  */
1707 static struct buffer_head *
1708 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1709                       struct mb_cache_entry **pce)
1710 {
1711         __u32 hash = le32_to_cpu(header->h_hash);
1712         struct mb_cache_entry *ce;
1713         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1714
1715         if (!header->h_hash)
1716                 return NULL;  /* never share */
1717         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1718         ce = mb_cache_entry_find_first(ext4_mb_cache, hash);
1719         while (ce) {
1720                 struct buffer_head *bh;
1721
1722                 bh = sb_bread(inode->i_sb, ce->e_block);
1723                 if (!bh) {
1724                         EXT4_ERROR_INODE(inode, "block %lu read error",
1725                                          (unsigned long) ce->e_block);
1726                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1727                         *pce = ce;
1728                         return bh;
1729                 }
1730                 brelse(bh);
1731                 ce = mb_cache_entry_find_next(ext4_mb_cache, ce);
1732         }
1733         return NULL;
1734 }
1735
1736 #define NAME_HASH_SHIFT 5
1737 #define VALUE_HASH_SHIFT 16
1738
1739 /*
1740  * ext4_xattr_hash_entry()
1741  *
1742  * Compute the hash of an extended attribute.
1743  */
1744 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1745                                          struct ext4_xattr_entry *entry)
1746 {
1747         __u32 hash = 0;
1748         char *name = entry->e_name;
1749         int n;
1750
1751         for (n = 0; n < entry->e_name_len; n++) {
1752                 hash = (hash << NAME_HASH_SHIFT) ^
1753                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1754                        *name++;
1755         }
1756
1757         if (entry->e_value_size != 0) {
1758                 __le32 *value = (__le32 *)((char *)header +
1759                         le16_to_cpu(entry->e_value_offs));
1760                 for (n = (le32_to_cpu(entry->e_value_size) +
1761                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1762                         hash = (hash << VALUE_HASH_SHIFT) ^
1763                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1764                                le32_to_cpu(*value++);
1765                 }
1766         }
1767         entry->e_hash = cpu_to_le32(hash);
1768 }
1769
1770 #undef NAME_HASH_SHIFT
1771 #undef VALUE_HASH_SHIFT
1772
1773 #define BLOCK_HASH_SHIFT 16
1774
1775 /*
1776  * ext4_xattr_rehash()
1777  *
1778  * Re-compute the extended attribute hash value after an entry has changed.
1779  */
1780 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1781                               struct ext4_xattr_entry *entry)
1782 {
1783         struct ext4_xattr_entry *here;
1784         __u32 hash = 0;
1785
1786         ext4_xattr_hash_entry(header, entry);
1787         here = ENTRY(header+1);
1788         while (!IS_LAST_ENTRY(here)) {
1789                 if (!here->e_hash) {
1790                         /* Block is not shared if an entry's hash value == 0 */
1791                         hash = 0;
1792                         break;
1793                 }
1794                 hash = (hash << BLOCK_HASH_SHIFT) ^
1795                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1796                        le32_to_cpu(here->e_hash);
1797                 here = EXT4_XATTR_NEXT(here);
1798         }
1799         header->h_hash = cpu_to_le32(hash);
1800 }
1801
1802 #undef BLOCK_HASH_SHIFT
1803
1804 #define HASH_BUCKET_BITS        10
1805
1806 struct mb_cache *
1807 ext4_xattr_create_cache(void)
1808 {
1809         return mb_cache_create(HASH_BUCKET_BITS);
1810 }
1811
1812 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1813 {
1814         if (cache)
1815                 mb_cache_destroy(cache);
1816 }
1817