Merge tag 'devicetree-for-4.10' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-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 (strlen(name) > 255)
415                 return -ERANGE;
416
417         down_read(&EXT4_I(inode)->xattr_sem);
418         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
419                                      buffer_size);
420         if (error == -ENODATA)
421                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
422                                              buffer_size);
423         up_read(&EXT4_I(inode)->xattr_sem);
424         return error;
425 }
426
427 static int
428 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
429                         char *buffer, size_t buffer_size)
430 {
431         size_t rest = buffer_size;
432
433         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
434                 const struct xattr_handler *handler =
435                         ext4_xattr_handler(entry->e_name_index);
436
437                 if (handler && (!handler->list || handler->list(dentry))) {
438                         const char *prefix = handler->prefix ?: handler->name;
439                         size_t prefix_len = strlen(prefix);
440                         size_t size = prefix_len + entry->e_name_len + 1;
441
442                         if (buffer) {
443                                 if (size > rest)
444                                         return -ERANGE;
445                                 memcpy(buffer, prefix, prefix_len);
446                                 buffer += prefix_len;
447                                 memcpy(buffer, entry->e_name, entry->e_name_len);
448                                 buffer += entry->e_name_len;
449                                 *buffer++ = 0;
450                         }
451                         rest -= size;
452                 }
453         }
454         return buffer_size - rest;  /* total size */
455 }
456
457 static int
458 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
459 {
460         struct inode *inode = d_inode(dentry);
461         struct buffer_head *bh = NULL;
462         int error;
463         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
464
465         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
466                   buffer, (long)buffer_size);
467
468         error = 0;
469         if (!EXT4_I(inode)->i_file_acl)
470                 goto cleanup;
471         ea_idebug(inode, "reading block %llu",
472                   (unsigned long long)EXT4_I(inode)->i_file_acl);
473         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
474         error = -EIO;
475         if (!bh)
476                 goto cleanup;
477         ea_bdebug(bh, "b_count=%d, refcount=%d",
478                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
479         if (ext4_xattr_check_block(inode, bh)) {
480                 EXT4_ERROR_INODE(inode, "bad block %llu",
481                                  EXT4_I(inode)->i_file_acl);
482                 error = -EFSCORRUPTED;
483                 goto cleanup;
484         }
485         ext4_xattr_cache_insert(ext4_mb_cache, bh);
486         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
487
488 cleanup:
489         brelse(bh);
490
491         return error;
492 }
493
494 static int
495 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
496 {
497         struct inode *inode = d_inode(dentry);
498         struct ext4_xattr_ibody_header *header;
499         struct ext4_inode *raw_inode;
500         struct ext4_iloc iloc;
501         void *end;
502         int error;
503
504         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
505                 return 0;
506         error = ext4_get_inode_loc(inode, &iloc);
507         if (error)
508                 return error;
509         raw_inode = ext4_raw_inode(&iloc);
510         header = IHDR(inode, raw_inode);
511         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
512         error = xattr_check_inode(inode, header, end);
513         if (error)
514                 goto cleanup;
515         error = ext4_xattr_list_entries(dentry, IFIRST(header),
516                                         buffer, buffer_size);
517
518 cleanup:
519         brelse(iloc.bh);
520         return error;
521 }
522
523 /*
524  * ext4_xattr_list()
525  *
526  * Copy a list of attribute names into the buffer
527  * provided, or compute the buffer size required.
528  * Buffer is NULL to compute the size of the buffer required.
529  *
530  * Returns a negative error number on failure, or the number of bytes
531  * used / required on success.
532  */
533 static int
534 ext4_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
535 {
536         int ret, ret2;
537
538         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
539         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
540         if (ret < 0)
541                 goto errout;
542         if (buffer) {
543                 buffer += ret;
544                 buffer_size -= ret;
545         }
546         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
547         if (ret < 0)
548                 goto errout;
549         ret += ret2;
550 errout:
551         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
552         return ret;
553 }
554
555 /*
556  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
557  * not set, set it.
558  */
559 static void ext4_xattr_update_super_block(handle_t *handle,
560                                           struct super_block *sb)
561 {
562         if (ext4_has_feature_xattr(sb))
563                 return;
564
565         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
566         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
567                 ext4_set_feature_xattr(sb);
568                 ext4_handle_dirty_super(handle, sb);
569         }
570 }
571
572 /*
573  * Release the xattr block BH: If the reference count is > 1, decrement it;
574  * otherwise free the block.
575  */
576 static void
577 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
578                          struct buffer_head *bh)
579 {
580         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
581         u32 hash, ref;
582         int error = 0;
583
584         BUFFER_TRACE(bh, "get_write_access");
585         error = ext4_journal_get_write_access(handle, bh);
586         if (error)
587                 goto out;
588
589         lock_buffer(bh);
590         hash = le32_to_cpu(BHDR(bh)->h_hash);
591         ref = le32_to_cpu(BHDR(bh)->h_refcount);
592         if (ref == 1) {
593                 ea_bdebug(bh, "refcount now=0; freeing");
594                 /*
595                  * This must happen under buffer lock for
596                  * ext4_xattr_block_set() to reliably detect freed block
597                  */
598                 mb_cache_entry_delete_block(ext4_mb_cache, hash, bh->b_blocknr);
599                 get_bh(bh);
600                 unlock_buffer(bh);
601                 ext4_free_blocks(handle, inode, bh, 0, 1,
602                                  EXT4_FREE_BLOCKS_METADATA |
603                                  EXT4_FREE_BLOCKS_FORGET);
604         } else {
605                 ref--;
606                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
607                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
608                         struct mb_cache_entry *ce;
609
610                         ce = mb_cache_entry_get(ext4_mb_cache, hash,
611                                                 bh->b_blocknr);
612                         if (ce) {
613                                 ce->e_reusable = 1;
614                                 mb_cache_entry_put(ext4_mb_cache, ce);
615                         }
616                 }
617
618                 /*
619                  * Beware of this ugliness: Releasing of xattr block references
620                  * from different inodes can race and so we have to protect
621                  * from a race where someone else frees the block (and releases
622                  * its journal_head) before we are done dirtying the buffer. In
623                  * nojournal mode this race is harmless and we actually cannot
624                  * call ext4_handle_dirty_xattr_block() with locked buffer as
625                  * that function can call sync_dirty_buffer() so for that case
626                  * we handle the dirtying after unlocking the buffer.
627                  */
628                 if (ext4_handle_valid(handle))
629                         error = ext4_handle_dirty_xattr_block(handle, inode,
630                                                               bh);
631                 unlock_buffer(bh);
632                 if (!ext4_handle_valid(handle))
633                         error = ext4_handle_dirty_xattr_block(handle, inode,
634                                                               bh);
635                 if (IS_SYNC(inode))
636                         ext4_handle_sync(handle);
637                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
638                 ea_bdebug(bh, "refcount now=%d; releasing",
639                           le32_to_cpu(BHDR(bh)->h_refcount));
640         }
641 out:
642         ext4_std_error(inode->i_sb, error);
643         return;
644 }
645
646 /*
647  * Find the available free space for EAs. This also returns the total number of
648  * bytes used by EA entries.
649  */
650 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
651                                     size_t *min_offs, void *base, int *total)
652 {
653         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
654                 if (last->e_value_size) {
655                         size_t offs = le16_to_cpu(last->e_value_offs);
656                         if (offs < *min_offs)
657                                 *min_offs = offs;
658                 }
659                 if (total)
660                         *total += EXT4_XATTR_LEN(last->e_name_len);
661         }
662         return (*min_offs - ((void *)last - base) - sizeof(__u32));
663 }
664
665 static int
666 ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
667 {
668         struct ext4_xattr_entry *last;
669         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
670
671         /* Compute min_offs and last. */
672         last = s->first;
673         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
674                 if (last->e_value_size) {
675                         size_t offs = le16_to_cpu(last->e_value_offs);
676                         if (offs < min_offs)
677                                 min_offs = offs;
678                 }
679         }
680         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
681         if (!s->not_found) {
682                 if (s->here->e_value_size) {
683                         size_t size = le32_to_cpu(s->here->e_value_size);
684                         free += EXT4_XATTR_SIZE(size);
685                 }
686                 free += EXT4_XATTR_LEN(name_len);
687         }
688         if (i->value) {
689                 if (free < EXT4_XATTR_LEN(name_len) +
690                            EXT4_XATTR_SIZE(i->value_len))
691                         return -ENOSPC;
692         }
693
694         if (i->value && s->not_found) {
695                 /* Insert the new name. */
696                 size_t size = EXT4_XATTR_LEN(name_len);
697                 size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
698                 memmove((void *)s->here + size, s->here, rest);
699                 memset(s->here, 0, size);
700                 s->here->e_name_index = i->name_index;
701                 s->here->e_name_len = name_len;
702                 memcpy(s->here->e_name, i->name, name_len);
703         } else {
704                 if (s->here->e_value_size) {
705                         void *first_val = s->base + min_offs;
706                         size_t offs = le16_to_cpu(s->here->e_value_offs);
707                         void *val = s->base + offs;
708                         size_t size = EXT4_XATTR_SIZE(
709                                 le32_to_cpu(s->here->e_value_size));
710
711                         if (i->value && size == EXT4_XATTR_SIZE(i->value_len)) {
712                                 /* The old and the new value have the same
713                                    size. Just replace. */
714                                 s->here->e_value_size =
715                                         cpu_to_le32(i->value_len);
716                                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
717                                         memset(val, 0, size);
718                                 } else {
719                                         /* Clear pad bytes first. */
720                                         memset(val + size - EXT4_XATTR_PAD, 0,
721                                                EXT4_XATTR_PAD);
722                                         memcpy(val, i->value, i->value_len);
723                                 }
724                                 return 0;
725                         }
726
727                         /* Remove the old value. */
728                         memmove(first_val + size, first_val, val - first_val);
729                         memset(first_val, 0, size);
730                         s->here->e_value_size = 0;
731                         s->here->e_value_offs = 0;
732                         min_offs += size;
733
734                         /* Adjust all value offsets. */
735                         last = s->first;
736                         while (!IS_LAST_ENTRY(last)) {
737                                 size_t o = le16_to_cpu(last->e_value_offs);
738                                 if (last->e_value_size && o < offs)
739                                         last->e_value_offs =
740                                                 cpu_to_le16(o + size);
741                                 last = EXT4_XATTR_NEXT(last);
742                         }
743                 }
744                 if (!i->value) {
745                         /* Remove the old name. */
746                         size_t size = EXT4_XATTR_LEN(name_len);
747                         last = ENTRY((void *)last - size);
748                         memmove(s->here, (void *)s->here + size,
749                                 (void *)last - (void *)s->here + sizeof(__u32));
750                         memset(last, 0, size);
751                 }
752         }
753
754         if (i->value) {
755                 /* Insert the new value. */
756                 s->here->e_value_size = cpu_to_le32(i->value_len);
757                 if (i->value_len) {
758                         size_t size = EXT4_XATTR_SIZE(i->value_len);
759                         void *val = s->base + min_offs - size;
760                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
761                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
762                                 memset(val, 0, size);
763                         } else {
764                                 /* Clear the pad bytes first. */
765                                 memset(val + size - EXT4_XATTR_PAD, 0,
766                                        EXT4_XATTR_PAD);
767                                 memcpy(val, i->value, i->value_len);
768                         }
769                 }
770         }
771         return 0;
772 }
773
774 struct ext4_xattr_block_find {
775         struct ext4_xattr_search s;
776         struct buffer_head *bh;
777 };
778
779 static int
780 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
781                       struct ext4_xattr_block_find *bs)
782 {
783         struct super_block *sb = inode->i_sb;
784         int error;
785
786         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
787                   i->name_index, i->name, i->value, (long)i->value_len);
788
789         if (EXT4_I(inode)->i_file_acl) {
790                 /* The inode already has an extended attribute block. */
791                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
792                 error = -EIO;
793                 if (!bs->bh)
794                         goto cleanup;
795                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
796                         atomic_read(&(bs->bh->b_count)),
797                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
798                 if (ext4_xattr_check_block(inode, bs->bh)) {
799                         EXT4_ERROR_INODE(inode, "bad block %llu",
800                                          EXT4_I(inode)->i_file_acl);
801                         error = -EFSCORRUPTED;
802                         goto cleanup;
803                 }
804                 /* Find the named attribute. */
805                 bs->s.base = BHDR(bs->bh);
806                 bs->s.first = BFIRST(bs->bh);
807                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
808                 bs->s.here = bs->s.first;
809                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
810                                               i->name, bs->bh->b_size, 1);
811                 if (error && error != -ENODATA)
812                         goto cleanup;
813                 bs->s.not_found = error;
814         }
815         error = 0;
816
817 cleanup:
818         return error;
819 }
820
821 static int
822 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
823                      struct ext4_xattr_info *i,
824                      struct ext4_xattr_block_find *bs)
825 {
826         struct super_block *sb = inode->i_sb;
827         struct buffer_head *new_bh = NULL;
828         struct ext4_xattr_search *s = &bs->s;
829         struct mb_cache_entry *ce = NULL;
830         int error = 0;
831         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
832
833 #define header(x) ((struct ext4_xattr_header *)(x))
834
835         if (i->value && i->value_len > sb->s_blocksize)
836                 return -ENOSPC;
837         if (s->base) {
838                 BUFFER_TRACE(bs->bh, "get_write_access");
839                 error = ext4_journal_get_write_access(handle, bs->bh);
840                 if (error)
841                         goto cleanup;
842                 lock_buffer(bs->bh);
843
844                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
845                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
846
847                         /*
848                          * This must happen under buffer lock for
849                          * ext4_xattr_block_set() to reliably detect modified
850                          * block
851                          */
852                         mb_cache_entry_delete_block(ext4_mb_cache, hash,
853                                                     bs->bh->b_blocknr);
854                         ea_bdebug(bs->bh, "modifying in-place");
855                         error = ext4_xattr_set_entry(i, s);
856                         if (!error) {
857                                 if (!IS_LAST_ENTRY(s->first))
858                                         ext4_xattr_rehash(header(s->base),
859                                                           s->here);
860                                 ext4_xattr_cache_insert(ext4_mb_cache,
861                                         bs->bh);
862                         }
863                         unlock_buffer(bs->bh);
864                         if (error == -EFSCORRUPTED)
865                                 goto bad_block;
866                         if (!error)
867                                 error = ext4_handle_dirty_xattr_block(handle,
868                                                                       inode,
869                                                                       bs->bh);
870                         if (error)
871                                 goto cleanup;
872                         goto inserted;
873                 } else {
874                         int offset = (char *)s->here - bs->bh->b_data;
875
876                         unlock_buffer(bs->bh);
877                         ea_bdebug(bs->bh, "cloning");
878                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
879                         error = -ENOMEM;
880                         if (s->base == NULL)
881                                 goto cleanup;
882                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
883                         s->first = ENTRY(header(s->base)+1);
884                         header(s->base)->h_refcount = cpu_to_le32(1);
885                         s->here = ENTRY(s->base + offset);
886                         s->end = s->base + bs->bh->b_size;
887                 }
888         } else {
889                 /* Allocate a buffer where we construct the new block. */
890                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
891                 /* assert(header == s->base) */
892                 error = -ENOMEM;
893                 if (s->base == NULL)
894                         goto cleanup;
895                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
896                 header(s->base)->h_blocks = cpu_to_le32(1);
897                 header(s->base)->h_refcount = cpu_to_le32(1);
898                 s->first = ENTRY(header(s->base)+1);
899                 s->here = ENTRY(header(s->base)+1);
900                 s->end = s->base + sb->s_blocksize;
901         }
902
903         error = ext4_xattr_set_entry(i, s);
904         if (error == -EFSCORRUPTED)
905                 goto bad_block;
906         if (error)
907                 goto cleanup;
908         if (!IS_LAST_ENTRY(s->first))
909                 ext4_xattr_rehash(header(s->base), s->here);
910
911 inserted:
912         if (!IS_LAST_ENTRY(s->first)) {
913                 new_bh = ext4_xattr_cache_find(inode, header(s->base), &ce);
914                 if (new_bh) {
915                         /* We found an identical block in the cache. */
916                         if (new_bh == bs->bh)
917                                 ea_bdebug(new_bh, "keeping");
918                         else {
919                                 u32 ref;
920
921                                 /* The old block is released after updating
922                                    the inode. */
923                                 error = dquot_alloc_block(inode,
924                                                 EXT4_C2B(EXT4_SB(sb), 1));
925                                 if (error)
926                                         goto cleanup;
927                                 BUFFER_TRACE(new_bh, "get_write_access");
928                                 error = ext4_journal_get_write_access(handle,
929                                                                       new_bh);
930                                 if (error)
931                                         goto cleanup_dquot;
932                                 lock_buffer(new_bh);
933                                 /*
934                                  * We have to be careful about races with
935                                  * freeing, rehashing or adding references to
936                                  * xattr block. Once we hold buffer lock xattr
937                                  * block's state is stable so we can check
938                                  * whether the block got freed / rehashed or
939                                  * not.  Since we unhash mbcache entry under
940                                  * buffer lock when freeing / rehashing xattr
941                                  * block, checking whether entry is still
942                                  * hashed is reliable. Same rules hold for
943                                  * e_reusable handling.
944                                  */
945                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
946                                     !ce->e_reusable) {
947                                         /*
948                                          * Undo everything and check mbcache
949                                          * again.
950                                          */
951                                         unlock_buffer(new_bh);
952                                         dquot_free_block(inode,
953                                                          EXT4_C2B(EXT4_SB(sb),
954                                                                   1));
955                                         brelse(new_bh);
956                                         mb_cache_entry_put(ext4_mb_cache, ce);
957                                         ce = NULL;
958                                         new_bh = NULL;
959                                         goto inserted;
960                                 }
961                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
962                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
963                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
964                                         ce->e_reusable = 0;
965                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
966                                           ref);
967                                 unlock_buffer(new_bh);
968                                 error = ext4_handle_dirty_xattr_block(handle,
969                                                                       inode,
970                                                                       new_bh);
971                                 if (error)
972                                         goto cleanup_dquot;
973                         }
974                         mb_cache_entry_touch(ext4_mb_cache, ce);
975                         mb_cache_entry_put(ext4_mb_cache, ce);
976                         ce = NULL;
977                 } else if (bs->bh && s->base == bs->bh->b_data) {
978                         /* We were modifying this block in-place. */
979                         ea_bdebug(bs->bh, "keeping this block");
980                         new_bh = bs->bh;
981                         get_bh(new_bh);
982                 } else {
983                         /* We need to allocate a new block */
984                         ext4_fsblk_t goal, block;
985
986                         goal = ext4_group_first_block_no(sb,
987                                                 EXT4_I(inode)->i_block_group);
988
989                         /* non-extent files can't have physical blocks past 2^32 */
990                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
991                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
992
993                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
994                                                      NULL, &error);
995                         if (error)
996                                 goto cleanup;
997
998                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
999                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
1000
1001                         ea_idebug(inode, "creating block %llu",
1002                                   (unsigned long long)block);
1003
1004                         new_bh = sb_getblk(sb, block);
1005                         if (unlikely(!new_bh)) {
1006                                 error = -ENOMEM;
1007 getblk_failed:
1008                                 ext4_free_blocks(handle, inode, NULL, block, 1,
1009                                                  EXT4_FREE_BLOCKS_METADATA);
1010                                 goto cleanup;
1011                         }
1012                         lock_buffer(new_bh);
1013                         error = ext4_journal_get_create_access(handle, new_bh);
1014                         if (error) {
1015                                 unlock_buffer(new_bh);
1016                                 error = -EIO;
1017                                 goto getblk_failed;
1018                         }
1019                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
1020                         set_buffer_uptodate(new_bh);
1021                         unlock_buffer(new_bh);
1022                         ext4_xattr_cache_insert(ext4_mb_cache, new_bh);
1023                         error = ext4_handle_dirty_xattr_block(handle,
1024                                                               inode, new_bh);
1025                         if (error)
1026                                 goto cleanup;
1027                 }
1028         }
1029
1030         /* Update the inode. */
1031         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
1032
1033         /* Drop the previous xattr block. */
1034         if (bs->bh && bs->bh != new_bh)
1035                 ext4_xattr_release_block(handle, inode, bs->bh);
1036         error = 0;
1037
1038 cleanup:
1039         if (ce)
1040                 mb_cache_entry_put(ext4_mb_cache, ce);
1041         brelse(new_bh);
1042         if (!(bs->bh && s->base == bs->bh->b_data))
1043                 kfree(s->base);
1044
1045         return error;
1046
1047 cleanup_dquot:
1048         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
1049         goto cleanup;
1050
1051 bad_block:
1052         EXT4_ERROR_INODE(inode, "bad block %llu",
1053                          EXT4_I(inode)->i_file_acl);
1054         goto cleanup;
1055
1056 #undef header
1057 }
1058
1059 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
1060                           struct ext4_xattr_ibody_find *is)
1061 {
1062         struct ext4_xattr_ibody_header *header;
1063         struct ext4_inode *raw_inode;
1064         int error;
1065
1066         if (EXT4_I(inode)->i_extra_isize == 0)
1067                 return 0;
1068         raw_inode = ext4_raw_inode(&is->iloc);
1069         header = IHDR(inode, raw_inode);
1070         is->s.base = is->s.first = IFIRST(header);
1071         is->s.here = is->s.first;
1072         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1073         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
1074                 error = xattr_check_inode(inode, header, is->s.end);
1075                 if (error)
1076                         return error;
1077                 /* Find the named attribute. */
1078                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
1079                                               i->name, is->s.end -
1080                                               (void *)is->s.base, 0);
1081                 if (error && error != -ENODATA)
1082                         return error;
1083                 is->s.not_found = error;
1084         }
1085         return 0;
1086 }
1087
1088 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
1089                                 struct ext4_xattr_info *i,
1090                                 struct ext4_xattr_ibody_find *is)
1091 {
1092         struct ext4_xattr_ibody_header *header;
1093         struct ext4_xattr_search *s = &is->s;
1094         int error;
1095
1096         if (EXT4_I(inode)->i_extra_isize == 0)
1097                 return -ENOSPC;
1098         error = ext4_xattr_set_entry(i, s);
1099         if (error) {
1100                 if (error == -ENOSPC &&
1101                     ext4_has_inline_data(inode)) {
1102                         error = ext4_try_to_evict_inline_data(handle, inode,
1103                                         EXT4_XATTR_LEN(strlen(i->name) +
1104                                         EXT4_XATTR_SIZE(i->value_len)));
1105                         if (error)
1106                                 return error;
1107                         error = ext4_xattr_ibody_find(inode, i, is);
1108                         if (error)
1109                                 return error;
1110                         error = ext4_xattr_set_entry(i, s);
1111                 }
1112                 if (error)
1113                         return error;
1114         }
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_ibody_set(struct inode *inode,
1127                                 struct ext4_xattr_info *i,
1128                                 struct ext4_xattr_ibody_find *is)
1129 {
1130         struct ext4_xattr_ibody_header *header;
1131         struct ext4_xattr_search *s = &is->s;
1132         int error;
1133
1134         if (EXT4_I(inode)->i_extra_isize == 0)
1135                 return -ENOSPC;
1136         error = ext4_xattr_set_entry(i, s);
1137         if (error)
1138                 return error;
1139         header = IHDR(inode, ext4_raw_inode(&is->iloc));
1140         if (!IS_LAST_ENTRY(s->first)) {
1141                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1142                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
1143         } else {
1144                 header->h_magic = cpu_to_le32(0);
1145                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
1146         }
1147         return 0;
1148 }
1149
1150 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
1151                                  struct ext4_xattr_info *i)
1152 {
1153         void *value;
1154
1155         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
1156                 return 0;
1157         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
1158         return !memcmp(value, i->value, i->value_len);
1159 }
1160
1161 /*
1162  * ext4_xattr_set_handle()
1163  *
1164  * Create, replace or remove an extended attribute for this inode.  Value
1165  * is NULL to remove an existing extended attribute, and non-NULL to
1166  * either replace an existing extended attribute, or create a new extended
1167  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
1168  * specify that an extended attribute must exist and must not exist
1169  * previous to the call, respectively.
1170  *
1171  * Returns 0, or a negative error number on failure.
1172  */
1173 int
1174 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
1175                       const char *name, const void *value, size_t value_len,
1176                       int flags)
1177 {
1178         struct ext4_xattr_info i = {
1179                 .name_index = name_index,
1180                 .name = name,
1181                 .value = value,
1182                 .value_len = value_len,
1183
1184         };
1185         struct ext4_xattr_ibody_find is = {
1186                 .s = { .not_found = -ENODATA, },
1187         };
1188         struct ext4_xattr_block_find bs = {
1189                 .s = { .not_found = -ENODATA, },
1190         };
1191         unsigned long no_expand;
1192         int error;
1193
1194         if (!name)
1195                 return -EINVAL;
1196         if (strlen(name) > 255)
1197                 return -ERANGE;
1198         down_write(&EXT4_I(inode)->xattr_sem);
1199         no_expand = ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND);
1200         ext4_set_inode_state(inode, EXT4_STATE_NO_EXPAND);
1201
1202         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
1203         if (error)
1204                 goto cleanup;
1205
1206         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
1207                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
1208                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
1209                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
1210         }
1211
1212         error = ext4_xattr_ibody_find(inode, &i, &is);
1213         if (error)
1214                 goto cleanup;
1215         if (is.s.not_found)
1216                 error = ext4_xattr_block_find(inode, &i, &bs);
1217         if (error)
1218                 goto cleanup;
1219         if (is.s.not_found && bs.s.not_found) {
1220                 error = -ENODATA;
1221                 if (flags & XATTR_REPLACE)
1222                         goto cleanup;
1223                 error = 0;
1224                 if (!value)
1225                         goto cleanup;
1226         } else {
1227                 error = -EEXIST;
1228                 if (flags & XATTR_CREATE)
1229                         goto cleanup;
1230         }
1231         if (!value) {
1232                 if (!is.s.not_found)
1233                         error = ext4_xattr_ibody_set(inode, &i, &is);
1234                 else if (!bs.s.not_found)
1235                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1236         } else {
1237                 error = 0;
1238                 /* Xattr value did not change? Save us some work and bail out */
1239                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
1240                         goto cleanup;
1241                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
1242                         goto cleanup;
1243
1244                 error = ext4_xattr_ibody_set(inode, &i, &is);
1245                 if (!error && !bs.s.not_found) {
1246                         i.value = NULL;
1247                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1248                 } else if (error == -ENOSPC) {
1249                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
1250                                 error = ext4_xattr_block_find(inode, &i, &bs);
1251                                 if (error)
1252                                         goto cleanup;
1253                         }
1254                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
1255                         if (error)
1256                                 goto cleanup;
1257                         if (!is.s.not_found) {
1258                                 i.value = NULL;
1259                                 error = ext4_xattr_ibody_set(inode, &i, &is);
1260                         }
1261                 }
1262         }
1263         if (!error) {
1264                 ext4_xattr_update_super_block(handle, inode->i_sb);
1265                 inode->i_ctime = current_time(inode);
1266                 if (!value)
1267                         ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1268                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
1269                 /*
1270                  * The bh is consumed by ext4_mark_iloc_dirty, even with
1271                  * error != 0.
1272                  */
1273                 is.iloc.bh = NULL;
1274                 if (IS_SYNC(inode))
1275                         ext4_handle_sync(handle);
1276         }
1277
1278 cleanup:
1279         brelse(is.iloc.bh);
1280         brelse(bs.bh);
1281         if (no_expand == 0)
1282                 ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1283         up_write(&EXT4_I(inode)->xattr_sem);
1284         return error;
1285 }
1286
1287 /*
1288  * ext4_xattr_set()
1289  *
1290  * Like ext4_xattr_set_handle, but start from an inode. This extended
1291  * attribute modification is a filesystem transaction by itself.
1292  *
1293  * Returns 0, or a negative error number on failure.
1294  */
1295 int
1296 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
1297                const void *value, size_t value_len, int flags)
1298 {
1299         handle_t *handle;
1300         int error, retries = 0;
1301         int credits = ext4_jbd2_credits_xattr(inode);
1302
1303 retry:
1304         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
1305         if (IS_ERR(handle)) {
1306                 error = PTR_ERR(handle);
1307         } else {
1308                 int error2;
1309
1310                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
1311                                               value, value_len, flags);
1312                 error2 = ext4_journal_stop(handle);
1313                 if (error == -ENOSPC &&
1314                     ext4_should_retry_alloc(inode->i_sb, &retries))
1315                         goto retry;
1316                 if (error == 0)
1317                         error = error2;
1318         }
1319
1320         return error;
1321 }
1322
1323 /*
1324  * Shift the EA entries in the inode to create space for the increased
1325  * i_extra_isize.
1326  */
1327 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
1328                                      int value_offs_shift, void *to,
1329                                      void *from, size_t n)
1330 {
1331         struct ext4_xattr_entry *last = entry;
1332         int new_offs;
1333
1334         /* We always shift xattr headers further thus offsets get lower */
1335         BUG_ON(value_offs_shift > 0);
1336
1337         /* Adjust the value offsets of the entries */
1338         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1339                 if (last->e_value_size) {
1340                         new_offs = le16_to_cpu(last->e_value_offs) +
1341                                                         value_offs_shift;
1342                         last->e_value_offs = cpu_to_le16(new_offs);
1343                 }
1344         }
1345         /* Shift the entries by n bytes */
1346         memmove(to, from, n);
1347 }
1348
1349 /*
1350  * Move xattr pointed to by 'entry' from inode into external xattr block
1351  */
1352 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
1353                                     struct ext4_inode *raw_inode,
1354                                     struct ext4_xattr_entry *entry)
1355 {
1356         struct ext4_xattr_ibody_find *is = NULL;
1357         struct ext4_xattr_block_find *bs = NULL;
1358         char *buffer = NULL, *b_entry_name = NULL;
1359         size_t value_offs, value_size;
1360         struct ext4_xattr_info i = {
1361                 .value = NULL,
1362                 .value_len = 0,
1363                 .name_index = entry->e_name_index,
1364         };
1365         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1366         int error;
1367
1368         value_offs = le16_to_cpu(entry->e_value_offs);
1369         value_size = le32_to_cpu(entry->e_value_size);
1370
1371         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
1372         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
1373         buffer = kmalloc(value_size, GFP_NOFS);
1374         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
1375         if (!is || !bs || !buffer || !b_entry_name) {
1376                 error = -ENOMEM;
1377                 goto out;
1378         }
1379
1380         is->s.not_found = -ENODATA;
1381         bs->s.not_found = -ENODATA;
1382         is->iloc.bh = NULL;
1383         bs->bh = NULL;
1384
1385         /* Save the entry name and the entry value */
1386         memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
1387         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
1388         b_entry_name[entry->e_name_len] = '\0';
1389         i.name = b_entry_name;
1390
1391         error = ext4_get_inode_loc(inode, &is->iloc);
1392         if (error)
1393                 goto out;
1394
1395         error = ext4_xattr_ibody_find(inode, &i, is);
1396         if (error)
1397                 goto out;
1398
1399         /* Remove the chosen entry from the inode */
1400         error = ext4_xattr_ibody_set(inode, &i, is);
1401         if (error)
1402                 goto out;
1403
1404         i.name = b_entry_name;
1405         i.value = buffer;
1406         i.value_len = value_size;
1407         error = ext4_xattr_block_find(inode, &i, bs);
1408         if (error)
1409                 goto out;
1410
1411         /* Add entry which was removed from the inode into the block */
1412         error = ext4_xattr_block_set(handle, inode, &i, bs);
1413         if (error)
1414                 goto out;
1415         error = 0;
1416 out:
1417         kfree(b_entry_name);
1418         kfree(buffer);
1419         if (is)
1420                 brelse(is->iloc.bh);
1421         kfree(is);
1422         kfree(bs);
1423
1424         return error;
1425 }
1426
1427 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
1428                                        struct ext4_inode *raw_inode,
1429                                        int isize_diff, size_t ifree,
1430                                        size_t bfree, int *total_ino)
1431 {
1432         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
1433         struct ext4_xattr_entry *small_entry;
1434         struct ext4_xattr_entry *entry;
1435         struct ext4_xattr_entry *last;
1436         unsigned int entry_size;        /* EA entry size */
1437         unsigned int total_size;        /* EA entry size + value size */
1438         unsigned int min_total_size;
1439         int error;
1440
1441         while (isize_diff > ifree) {
1442                 entry = NULL;
1443                 small_entry = NULL;
1444                 min_total_size = ~0U;
1445                 last = IFIRST(header);
1446                 /* Find the entry best suited to be pushed into EA block */
1447                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1448                         total_size =
1449                         EXT4_XATTR_SIZE(le32_to_cpu(last->e_value_size)) +
1450                                         EXT4_XATTR_LEN(last->e_name_len);
1451                         if (total_size <= bfree &&
1452                             total_size < min_total_size) {
1453                                 if (total_size + ifree < isize_diff) {
1454                                         small_entry = last;
1455                                 } else {
1456                                         entry = last;
1457                                         min_total_size = total_size;
1458                                 }
1459                         }
1460                 }
1461
1462                 if (entry == NULL) {
1463                         if (small_entry == NULL)
1464                                 return -ENOSPC;
1465                         entry = small_entry;
1466                 }
1467
1468                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
1469                 total_size = entry_size +
1470                         EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
1471                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
1472                                                  entry);
1473                 if (error)
1474                         return error;
1475
1476                 *total_ino -= entry_size;
1477                 ifree += total_size;
1478                 bfree -= total_size;
1479         }
1480
1481         return 0;
1482 }
1483
1484 /*
1485  * Expand an inode by new_extra_isize bytes when EAs are present.
1486  * Returns 0 on success or negative error number on failure.
1487  */
1488 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1489                                struct ext4_inode *raw_inode, handle_t *handle)
1490 {
1491         struct ext4_xattr_ibody_header *header;
1492         struct buffer_head *bh = NULL;
1493         size_t min_offs;
1494         size_t ifree, bfree;
1495         int total_ino;
1496         void *base, *end;
1497         int error = 0, tried_min_extra_isize = 0;
1498         int s_min_extra_isize = le16_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_min_extra_isize);
1499         int isize_diff; /* How much do we need to grow i_extra_isize */
1500
1501         down_write(&EXT4_I(inode)->xattr_sem);
1502         /*
1503          * Set EXT4_STATE_NO_EXPAND to avoid recursion when marking inode dirty
1504          */
1505         ext4_set_inode_state(inode, EXT4_STATE_NO_EXPAND);
1506 retry:
1507         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
1508         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
1509                 goto out;
1510
1511         header = IHDR(inode, raw_inode);
1512
1513         /*
1514          * Check if enough free space is available in the inode to shift the
1515          * entries ahead by new_extra_isize.
1516          */
1517
1518         base = IFIRST(header);
1519         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
1520         min_offs = end - base;
1521         total_ino = sizeof(struct ext4_xattr_ibody_header);
1522
1523         error = xattr_check_inode(inode, header, end);
1524         if (error)
1525                 goto cleanup;
1526
1527         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
1528         if (ifree >= isize_diff)
1529                 goto shift;
1530
1531         /*
1532          * Enough free space isn't available in the inode, check if
1533          * EA block can hold new_extra_isize bytes.
1534          */
1535         if (EXT4_I(inode)->i_file_acl) {
1536                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1537                 error = -EIO;
1538                 if (!bh)
1539                         goto cleanup;
1540                 if (ext4_xattr_check_block(inode, bh)) {
1541                         EXT4_ERROR_INODE(inode, "bad block %llu",
1542                                          EXT4_I(inode)->i_file_acl);
1543                         error = -EFSCORRUPTED;
1544                         goto cleanup;
1545                 }
1546                 base = BHDR(bh);
1547                 end = bh->b_data + bh->b_size;
1548                 min_offs = end - base;
1549                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
1550                                               NULL);
1551                 if (bfree + ifree < isize_diff) {
1552                         if (!tried_min_extra_isize && s_min_extra_isize) {
1553                                 tried_min_extra_isize++;
1554                                 new_extra_isize = s_min_extra_isize;
1555                                 brelse(bh);
1556                                 goto retry;
1557                         }
1558                         error = -ENOSPC;
1559                         goto cleanup;
1560                 }
1561         } else {
1562                 bfree = inode->i_sb->s_blocksize;
1563         }
1564
1565         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
1566                                             isize_diff, ifree, bfree,
1567                                             &total_ino);
1568         if (error) {
1569                 if (error == -ENOSPC && !tried_min_extra_isize &&
1570                     s_min_extra_isize) {
1571                         tried_min_extra_isize++;
1572                         new_extra_isize = s_min_extra_isize;
1573                         brelse(bh);
1574                         goto retry;
1575                 }
1576                 goto cleanup;
1577         }
1578 shift:
1579         /* Adjust the offsets and shift the remaining entries ahead */
1580         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
1581                         - new_extra_isize, (void *)raw_inode +
1582                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
1583                         (void *)header, total_ino);
1584         EXT4_I(inode)->i_extra_isize = new_extra_isize;
1585         brelse(bh);
1586 out:
1587         ext4_clear_inode_state(inode, EXT4_STATE_NO_EXPAND);
1588         up_write(&EXT4_I(inode)->xattr_sem);
1589         return 0;
1590
1591 cleanup:
1592         brelse(bh);
1593         /*
1594          * We deliberately leave EXT4_STATE_NO_EXPAND set here since inode
1595          * size expansion failed.
1596          */
1597         up_write(&EXT4_I(inode)->xattr_sem);
1598         return error;
1599 }
1600
1601
1602
1603 /*
1604  * ext4_xattr_delete_inode()
1605  *
1606  * Free extended attribute resources associated with this inode. This
1607  * is called immediately before an inode is freed. We have exclusive
1608  * access to the inode.
1609  */
1610 void
1611 ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1612 {
1613         struct buffer_head *bh = NULL;
1614
1615         if (!EXT4_I(inode)->i_file_acl)
1616                 goto cleanup;
1617         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
1618         if (!bh) {
1619                 EXT4_ERROR_INODE(inode, "block %llu read error",
1620                                  EXT4_I(inode)->i_file_acl);
1621                 goto cleanup;
1622         }
1623         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
1624             BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1625                 EXT4_ERROR_INODE(inode, "bad block %llu",
1626                                  EXT4_I(inode)->i_file_acl);
1627                 goto cleanup;
1628         }
1629         ext4_xattr_release_block(handle, inode, bh);
1630         EXT4_I(inode)->i_file_acl = 0;
1631
1632 cleanup:
1633         brelse(bh);
1634 }
1635
1636 /*
1637  * ext4_xattr_cache_insert()
1638  *
1639  * Create a new entry in the extended attribute cache, and insert
1640  * it unless such an entry is already in the cache.
1641  *
1642  * Returns 0, or a negative error number on failure.
1643  */
1644 static void
1645 ext4_xattr_cache_insert(struct mb_cache *ext4_mb_cache, struct buffer_head *bh)
1646 {
1647         struct ext4_xattr_header *header = BHDR(bh);
1648         __u32 hash = le32_to_cpu(header->h_hash);
1649         int reusable = le32_to_cpu(header->h_refcount) <
1650                        EXT4_XATTR_REFCOUNT_MAX;
1651         int error;
1652
1653         error = mb_cache_entry_create(ext4_mb_cache, GFP_NOFS, hash,
1654                                       bh->b_blocknr, reusable);
1655         if (error) {
1656                 if (error == -EBUSY)
1657                         ea_bdebug(bh, "already in cache");
1658         } else
1659                 ea_bdebug(bh, "inserting [%x]", (int)hash);
1660 }
1661
1662 /*
1663  * ext4_xattr_cmp()
1664  *
1665  * Compare two extended attribute blocks for equality.
1666  *
1667  * Returns 0 if the blocks are equal, 1 if they differ, and
1668  * a negative error number on errors.
1669  */
1670 static int
1671 ext4_xattr_cmp(struct ext4_xattr_header *header1,
1672                struct ext4_xattr_header *header2)
1673 {
1674         struct ext4_xattr_entry *entry1, *entry2;
1675
1676         entry1 = ENTRY(header1+1);
1677         entry2 = ENTRY(header2+1);
1678         while (!IS_LAST_ENTRY(entry1)) {
1679                 if (IS_LAST_ENTRY(entry2))
1680                         return 1;
1681                 if (entry1->e_hash != entry2->e_hash ||
1682                     entry1->e_name_index != entry2->e_name_index ||
1683                     entry1->e_name_len != entry2->e_name_len ||
1684                     entry1->e_value_size != entry2->e_value_size ||
1685                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1686                         return 1;
1687                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1688                         return -EFSCORRUPTED;
1689                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1690                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1691                            le32_to_cpu(entry1->e_value_size)))
1692                         return 1;
1693
1694                 entry1 = EXT4_XATTR_NEXT(entry1);
1695                 entry2 = EXT4_XATTR_NEXT(entry2);
1696         }
1697         if (!IS_LAST_ENTRY(entry2))
1698                 return 1;
1699         return 0;
1700 }
1701
1702 /*
1703  * ext4_xattr_cache_find()
1704  *
1705  * Find an identical extended attribute block.
1706  *
1707  * Returns a pointer to the block found, or NULL if such a block was
1708  * not found or an error occurred.
1709  */
1710 static struct buffer_head *
1711 ext4_xattr_cache_find(struct inode *inode, struct ext4_xattr_header *header,
1712                       struct mb_cache_entry **pce)
1713 {
1714         __u32 hash = le32_to_cpu(header->h_hash);
1715         struct mb_cache_entry *ce;
1716         struct mb_cache *ext4_mb_cache = EXT4_GET_MB_CACHE(inode);
1717
1718         if (!header->h_hash)
1719                 return NULL;  /* never share */
1720         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1721         ce = mb_cache_entry_find_first(ext4_mb_cache, hash);
1722         while (ce) {
1723                 struct buffer_head *bh;
1724
1725                 bh = sb_bread(inode->i_sb, ce->e_block);
1726                 if (!bh) {
1727                         EXT4_ERROR_INODE(inode, "block %lu read error",
1728                                          (unsigned long) ce->e_block);
1729                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
1730                         *pce = ce;
1731                         return bh;
1732                 }
1733                 brelse(bh);
1734                 ce = mb_cache_entry_find_next(ext4_mb_cache, ce);
1735         }
1736         return NULL;
1737 }
1738
1739 #define NAME_HASH_SHIFT 5
1740 #define VALUE_HASH_SHIFT 16
1741
1742 /*
1743  * ext4_xattr_hash_entry()
1744  *
1745  * Compute the hash of an extended attribute.
1746  */
1747 static inline void ext4_xattr_hash_entry(struct ext4_xattr_header *header,
1748                                          struct ext4_xattr_entry *entry)
1749 {
1750         __u32 hash = 0;
1751         char *name = entry->e_name;
1752         int n;
1753
1754         for (n = 0; n < entry->e_name_len; n++) {
1755                 hash = (hash << NAME_HASH_SHIFT) ^
1756                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1757                        *name++;
1758         }
1759
1760         if (entry->e_value_size != 0) {
1761                 __le32 *value = (__le32 *)((char *)header +
1762                         le16_to_cpu(entry->e_value_offs));
1763                 for (n = (le32_to_cpu(entry->e_value_size) +
1764                      EXT4_XATTR_ROUND) >> EXT4_XATTR_PAD_BITS; n; n--) {
1765                         hash = (hash << VALUE_HASH_SHIFT) ^
1766                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1767                                le32_to_cpu(*value++);
1768                 }
1769         }
1770         entry->e_hash = cpu_to_le32(hash);
1771 }
1772
1773 #undef NAME_HASH_SHIFT
1774 #undef VALUE_HASH_SHIFT
1775
1776 #define BLOCK_HASH_SHIFT 16
1777
1778 /*
1779  * ext4_xattr_rehash()
1780  *
1781  * Re-compute the extended attribute hash value after an entry has changed.
1782  */
1783 static void ext4_xattr_rehash(struct ext4_xattr_header *header,
1784                               struct ext4_xattr_entry *entry)
1785 {
1786         struct ext4_xattr_entry *here;
1787         __u32 hash = 0;
1788
1789         ext4_xattr_hash_entry(header, entry);
1790         here = ENTRY(header+1);
1791         while (!IS_LAST_ENTRY(here)) {
1792                 if (!here->e_hash) {
1793                         /* Block is not shared if an entry's hash value == 0 */
1794                         hash = 0;
1795                         break;
1796                 }
1797                 hash = (hash << BLOCK_HASH_SHIFT) ^
1798                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1799                        le32_to_cpu(here->e_hash);
1800                 here = EXT4_XATTR_NEXT(here);
1801         }
1802         header->h_hash = cpu_to_le32(hash);
1803 }
1804
1805 #undef BLOCK_HASH_SHIFT
1806
1807 #define HASH_BUCKET_BITS        10
1808
1809 struct mb_cache *
1810 ext4_xattr_create_cache(void)
1811 {
1812         return mb_cache_create(HASH_BUCKET_BITS);
1813 }
1814
1815 void ext4_xattr_destroy_cache(struct mb_cache *cache)
1816 {
1817         if (cache)
1818                 mb_cache_destroy(cache);
1819 }
1820