0b968868352626fd8aed6066eae9c43a0c1040e6
[sfrench/cifs-2.6.git] / fs / ext4 / xattr.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/ext4/xattr.c
4  *
5  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
6  *
7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9  * Extended attributes for symlinks and special files added per
10  *  suggestion of Luka Renko <luka.renko@hermes.si>.
11  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
12  *  Red Hat Inc.
13  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14  *  and Andreas Gruenbacher <agruen@suse.de>.
15  */
16
17 /*
18  * Extended attributes are stored directly in inodes (on file systems with
19  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
20  * field contains the block number if an inode uses an additional block. All
21  * attributes must fit in the inode and one additional block. Blocks that
22  * contain the identical set of attributes may be shared among several inodes.
23  * Identical blocks are detected by keeping a cache of blocks that have
24  * recently been accessed.
25  *
26  * The attributes in inodes and on blocks have a different header; the entries
27  * are stored in the same format:
28  *
29  *   +------------------+
30  *   | header           |
31  *   | entry 1          | |
32  *   | entry 2          | | growing downwards
33  *   | entry 3          | v
34  *   | four null bytes  |
35  *   | . . .            |
36  *   | value 1          | ^
37  *   | value 3          | | growing upwards
38  *   | value 2          | |
39  *   +------------------+
40  *
41  * The header is followed by multiple entry descriptors. In disk blocks, the
42  * entry descriptors are kept sorted. In inodes, they are unsorted. The
43  * attribute values are aligned to the end of the block in no specific order.
44  *
45  * Locking strategy
46  * ----------------
47  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
48  * EA blocks are only changed if they are exclusive to an inode, so
49  * holding xattr_sem also means that nothing but the EA block's reference
50  * count can change. Multiple writers to the same block are synchronized
51  * by the buffer lock.
52  */
53
54 #include <linux/init.h>
55 #include <linux/fs.h>
56 #include <linux/slab.h>
57 #include <linux/mbcache.h>
58 #include <linux/quotaops.h>
59 #include <linux/iversion.h>
60 #include "ext4_jbd2.h"
61 #include "ext4.h"
62 #include "xattr.h"
63 #include "acl.h"
64
65 #ifdef EXT4_XATTR_DEBUG
66 # define ea_idebug(inode, fmt, ...)                                     \
67         printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
68                inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
69 # define ea_bdebug(bh, fmt, ...)                                        \
70         printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
71                bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
72 #else
73 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
74 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
75 #endif
76
77 static void ext4_xattr_block_cache_insert(struct mb_cache *,
78                                           struct buffer_head *);
79 static struct buffer_head *
80 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
81                             struct mb_cache_entry **);
82 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
83                                     size_t value_count);
84 static void ext4_xattr_rehash(struct ext4_xattr_header *);
85
86 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
87         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
88 #ifdef CONFIG_EXT4_FS_POSIX_ACL
89         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
90         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
91 #endif
92         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
93 #ifdef CONFIG_EXT4_FS_SECURITY
94         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
95 #endif
96 };
97
98 const struct xattr_handler *ext4_xattr_handlers[] = {
99         &ext4_xattr_user_handler,
100         &ext4_xattr_trusted_handler,
101 #ifdef CONFIG_EXT4_FS_POSIX_ACL
102         &posix_acl_access_xattr_handler,
103         &posix_acl_default_xattr_handler,
104 #endif
105 #ifdef CONFIG_EXT4_FS_SECURITY
106         &ext4_xattr_security_handler,
107 #endif
108         NULL
109 };
110
111 #define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
112                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
113
114 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
115                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
116
117 static int
118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119                         struct inode *inode);
120
121 #ifdef CONFIG_LOCKDEP
122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
123 {
124         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
125 }
126 #endif
127
128 static __le32 ext4_xattr_block_csum(struct inode *inode,
129                                     sector_t block_nr,
130                                     struct ext4_xattr_header *hdr)
131 {
132         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
133         __u32 csum;
134         __le64 dsk_block_nr = cpu_to_le64(block_nr);
135         __u32 dummy_csum = 0;
136         int offset = offsetof(struct ext4_xattr_header, h_checksum);
137
138         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
139                            sizeof(dsk_block_nr));
140         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
141         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
142         offset += sizeof(dummy_csum);
143         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
144                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
145
146         return cpu_to_le32(csum);
147 }
148
149 static int ext4_xattr_block_csum_verify(struct inode *inode,
150                                         struct buffer_head *bh)
151 {
152         struct ext4_xattr_header *hdr = BHDR(bh);
153         int ret = 1;
154
155         if (ext4_has_metadata_csum(inode->i_sb)) {
156                 lock_buffer(bh);
157                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
158                                                         bh->b_blocknr, hdr));
159                 unlock_buffer(bh);
160         }
161         return ret;
162 }
163
164 static void ext4_xattr_block_csum_set(struct inode *inode,
165                                       struct buffer_head *bh)
166 {
167         if (ext4_has_metadata_csum(inode->i_sb))
168                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
169                                                 bh->b_blocknr, BHDR(bh));
170 }
171
172 static inline const struct xattr_handler *
173 ext4_xattr_handler(int name_index)
174 {
175         const struct xattr_handler *handler = NULL;
176
177         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
178                 handler = ext4_xattr_handler_map[name_index];
179         return handler;
180 }
181
182 static int
183 ext4_xattr_check_entries(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                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
194                         return -EFSCORRUPTED;
195                 e = next;
196         }
197
198         /* Check the values */
199         while (!IS_LAST_ENTRY(entry)) {
200                 u32 size = le32_to_cpu(entry->e_value_size);
201
202                 if (size > EXT4_XATTR_SIZE_MAX)
203                         return -EFSCORRUPTED;
204
205                 if (size != 0 && entry->e_value_inum == 0) {
206                         u16 offs = le16_to_cpu(entry->e_value_offs);
207                         void *value;
208
209                         /*
210                          * The value cannot overlap the names, and the value
211                          * with padding cannot extend beyond 'end'.  Check both
212                          * the padded and unpadded sizes, since the size may
213                          * overflow to 0 when adding padding.
214                          */
215                         if (offs > end - value_start)
216                                 return -EFSCORRUPTED;
217                         value = value_start + offs;
218                         if (value < (void *)e + sizeof(u32) ||
219                             size > end - value ||
220                             EXT4_XATTR_SIZE(size) > end - value)
221                                 return -EFSCORRUPTED;
222                 }
223                 entry = EXT4_XATTR_NEXT(entry);
224         }
225
226         return 0;
227 }
228
229 static inline int
230 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
231                          const char *function, unsigned int line)
232 {
233         int error = -EFSCORRUPTED;
234
235         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
236             BHDR(bh)->h_blocks != cpu_to_le32(1))
237                 goto errout;
238         if (buffer_verified(bh))
239                 return 0;
240
241         error = -EFSBADCRC;
242         if (!ext4_xattr_block_csum_verify(inode, bh))
243                 goto errout;
244         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
245                                          bh->b_data);
246 errout:
247         if (error)
248                 __ext4_error_inode(inode, function, line, 0,
249                                    "corrupted xattr block %llu",
250                                    (unsigned long long) bh->b_blocknr);
251         else
252                 set_buffer_verified(bh);
253         return error;
254 }
255
256 #define ext4_xattr_check_block(inode, bh) \
257         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
258
259
260 static int
261 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
262                          void *end, const char *function, unsigned int line)
263 {
264         int error = -EFSCORRUPTED;
265
266         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
267             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
268                 goto errout;
269         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
270 errout:
271         if (error)
272                 __ext4_error_inode(inode, function, line, 0,
273                                    "corrupted in-inode xattr");
274         return error;
275 }
276
277 #define xattr_check_inode(inode, header, end) \
278         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
279
280 static int
281 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
282                  void *end, int name_index, const char *name, int sorted)
283 {
284         struct ext4_xattr_entry *entry, *next;
285         size_t name_len;
286         int cmp = 1;
287
288         if (name == NULL)
289                 return -EINVAL;
290         name_len = strlen(name);
291         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
292                 next = EXT4_XATTR_NEXT(entry);
293                 if ((void *) next >= end) {
294                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
295                         return -EFSCORRUPTED;
296                 }
297                 cmp = name_index - entry->e_name_index;
298                 if (!cmp)
299                         cmp = name_len - entry->e_name_len;
300                 if (!cmp)
301                         cmp = memcmp(name, entry->e_name, name_len);
302                 if (cmp <= 0 && (sorted || cmp == 0))
303                         break;
304         }
305         *pentry = entry;
306         return cmp ? -ENODATA : 0;
307 }
308
309 static u32
310 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
311 {
312         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
313 }
314
315 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
316 {
317         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
318                 (u32) inode_peek_iversion_raw(ea_inode);
319 }
320
321 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
322 {
323         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
324         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
325 }
326
327 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
328 {
329         return (u32)ea_inode->i_atime.tv_sec;
330 }
331
332 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
333 {
334         ea_inode->i_atime.tv_sec = hash;
335 }
336
337 /*
338  * Read the EA value from an inode.
339  */
340 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
341 {
342         int blocksize = 1 << ea_inode->i_blkbits;
343         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
344         int tail_size = (size % blocksize) ?: blocksize;
345         struct buffer_head *bhs_inline[8];
346         struct buffer_head **bhs = bhs_inline;
347         int i, ret;
348
349         if (bh_count > ARRAY_SIZE(bhs_inline)) {
350                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
351                 if (!bhs)
352                         return -ENOMEM;
353         }
354
355         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
356                                true /* wait */, bhs);
357         if (ret)
358                 goto free_bhs;
359
360         for (i = 0; i < bh_count; i++) {
361                 /* There shouldn't be any holes in ea_inode. */
362                 if (!bhs[i]) {
363                         ret = -EFSCORRUPTED;
364                         goto put_bhs;
365                 }
366                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
367                        i < bh_count - 1 ? blocksize : tail_size);
368         }
369         ret = 0;
370 put_bhs:
371         for (i = 0; i < bh_count; i++)
372                 brelse(bhs[i]);
373 free_bhs:
374         if (bhs != bhs_inline)
375                 kfree(bhs);
376         return ret;
377 }
378
379 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
380
381 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
382                                  u32 ea_inode_hash, struct inode **ea_inode)
383 {
384         struct inode *inode;
385         int err;
386
387         inode = ext4_iget(parent->i_sb, ea_ino);
388         if (IS_ERR(inode)) {
389                 err = PTR_ERR(inode);
390                 ext4_error(parent->i_sb,
391                            "error while reading EA inode %lu err=%d", ea_ino,
392                            err);
393                 return err;
394         }
395
396         if (is_bad_inode(inode)) {
397                 ext4_error(parent->i_sb,
398                            "error while reading EA inode %lu is_bad_inode",
399                            ea_ino);
400                 err = -EIO;
401                 goto error;
402         }
403
404         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
405                 ext4_error(parent->i_sb,
406                            "EA inode %lu does not have EXT4_EA_INODE_FL flag",
407                             ea_ino);
408                 err = -EINVAL;
409                 goto error;
410         }
411
412         ext4_xattr_inode_set_class(inode);
413
414         /*
415          * Check whether this is an old Lustre-style xattr inode. Lustre
416          * implementation does not have hash validation, rather it has a
417          * backpointer from ea_inode to the parent inode.
418          */
419         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
420             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
421             inode->i_generation == parent->i_generation) {
422                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
423                 ext4_xattr_inode_set_ref(inode, 1);
424         } else {
425                 inode_lock(inode);
426                 inode->i_flags |= S_NOQUOTA;
427                 inode_unlock(inode);
428         }
429
430         *ea_inode = inode;
431         return 0;
432 error:
433         iput(inode);
434         return err;
435 }
436
437 static int
438 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
439                                struct ext4_xattr_entry *entry, void *buffer,
440                                size_t size)
441 {
442         u32 hash;
443
444         /* Verify stored hash matches calculated hash. */
445         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
446         if (hash != ext4_xattr_inode_get_hash(ea_inode))
447                 return -EFSCORRUPTED;
448
449         if (entry) {
450                 __le32 e_hash, tmp_data;
451
452                 /* Verify entry hash. */
453                 tmp_data = cpu_to_le32(hash);
454                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
455                                                &tmp_data, 1);
456                 if (e_hash != entry->e_hash)
457                         return -EFSCORRUPTED;
458         }
459         return 0;
460 }
461
462 /*
463  * Read xattr value from the EA inode.
464  */
465 static int
466 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
467                      void *buffer, size_t size)
468 {
469         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
470         struct inode *ea_inode;
471         int err;
472
473         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
474                                     le32_to_cpu(entry->e_hash), &ea_inode);
475         if (err) {
476                 ea_inode = NULL;
477                 goto out;
478         }
479
480         if (i_size_read(ea_inode) != size) {
481                 ext4_warning_inode(ea_inode,
482                                    "ea_inode file size=%llu entry size=%zu",
483                                    i_size_read(ea_inode), size);
484                 err = -EFSCORRUPTED;
485                 goto out;
486         }
487
488         err = ext4_xattr_inode_read(ea_inode, buffer, size);
489         if (err)
490                 goto out;
491
492         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
493                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
494                                                      size);
495                 if (err) {
496                         ext4_warning_inode(ea_inode,
497                                            "EA inode hash validation failed");
498                         goto out;
499                 }
500
501                 if (ea_inode_cache)
502                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
503                                         ext4_xattr_inode_get_hash(ea_inode),
504                                         ea_inode->i_ino, true /* reusable */);
505         }
506 out:
507         iput(ea_inode);
508         return err;
509 }
510
511 static int
512 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
513                      void *buffer, size_t buffer_size)
514 {
515         struct buffer_head *bh = NULL;
516         struct ext4_xattr_entry *entry;
517         size_t size;
518         void *end;
519         int error;
520         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
521
522         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
523                   name_index, name, buffer, (long)buffer_size);
524
525         error = -ENODATA;
526         if (!EXT4_I(inode)->i_file_acl)
527                 goto cleanup;
528         ea_idebug(inode, "reading block %llu",
529                   (unsigned long long)EXT4_I(inode)->i_file_acl);
530         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
531         if (!bh)
532                 goto cleanup;
533         ea_bdebug(bh, "b_count=%d, refcount=%d",
534                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
535         error = ext4_xattr_check_block(inode, bh);
536         if (error)
537                 goto cleanup;
538         ext4_xattr_block_cache_insert(ea_block_cache, bh);
539         entry = BFIRST(bh);
540         end = bh->b_data + bh->b_size;
541         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
542         if (error)
543                 goto cleanup;
544         size = le32_to_cpu(entry->e_value_size);
545         error = -ERANGE;
546         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
547                 goto cleanup;
548         if (buffer) {
549                 if (size > buffer_size)
550                         goto cleanup;
551                 if (entry->e_value_inum) {
552                         error = ext4_xattr_inode_get(inode, entry, buffer,
553                                                      size);
554                         if (error)
555                                 goto cleanup;
556                 } else {
557                         u16 offset = le16_to_cpu(entry->e_value_offs);
558                         void *p = bh->b_data + offset;
559
560                         if (unlikely(p + size > end))
561                                 goto cleanup;
562                         memcpy(buffer, p, size);
563                 }
564         }
565         error = size;
566
567 cleanup:
568         brelse(bh);
569         return error;
570 }
571
572 int
573 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
574                      void *buffer, size_t buffer_size)
575 {
576         struct ext4_xattr_ibody_header *header;
577         struct ext4_xattr_entry *entry;
578         struct ext4_inode *raw_inode;
579         struct ext4_iloc iloc;
580         size_t size;
581         void *end;
582         int error;
583
584         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
585                 return -ENODATA;
586         error = ext4_get_inode_loc(inode, &iloc);
587         if (error)
588                 return error;
589         raw_inode = ext4_raw_inode(&iloc);
590         header = IHDR(inode, raw_inode);
591         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
592         error = xattr_check_inode(inode, header, end);
593         if (error)
594                 goto cleanup;
595         entry = IFIRST(header);
596         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
597         if (error)
598                 goto cleanup;
599         size = le32_to_cpu(entry->e_value_size);
600         error = -ERANGE;
601         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
602                 goto cleanup;
603         if (buffer) {
604                 if (size > buffer_size)
605                         goto cleanup;
606                 if (entry->e_value_inum) {
607                         error = ext4_xattr_inode_get(inode, entry, buffer,
608                                                      size);
609                         if (error)
610                                 goto cleanup;
611                 } else {
612                         u16 offset = le16_to_cpu(entry->e_value_offs);
613                         void *p = (void *)IFIRST(header) + offset;
614
615                         if (unlikely(p + size > end))
616                                 goto cleanup;
617                         memcpy(buffer, p, size);
618                 }
619         }
620         error = size;
621
622 cleanup:
623         brelse(iloc.bh);
624         return error;
625 }
626
627 /*
628  * ext4_xattr_get()
629  *
630  * Copy an extended attribute into the buffer
631  * provided, or compute the buffer size required.
632  * Buffer is NULL to compute the size of the buffer required.
633  *
634  * Returns a negative error number on failure, or the number of bytes
635  * used / required on success.
636  */
637 int
638 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
639                void *buffer, size_t buffer_size)
640 {
641         int error;
642
643         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
644                 return -EIO;
645
646         if (strlen(name) > 255)
647                 return -ERANGE;
648
649         down_read(&EXT4_I(inode)->xattr_sem);
650         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
651                                      buffer_size);
652         if (error == -ENODATA)
653                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
654                                              buffer_size);
655         up_read(&EXT4_I(inode)->xattr_sem);
656         return error;
657 }
658
659 static int
660 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
661                         char *buffer, size_t buffer_size)
662 {
663         size_t rest = buffer_size;
664
665         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
666                 const struct xattr_handler *handler =
667                         ext4_xattr_handler(entry->e_name_index);
668
669                 if (handler && (!handler->list || handler->list(dentry))) {
670                         const char *prefix = handler->prefix ?: handler->name;
671                         size_t prefix_len = strlen(prefix);
672                         size_t size = prefix_len + entry->e_name_len + 1;
673
674                         if (buffer) {
675                                 if (size > rest)
676                                         return -ERANGE;
677                                 memcpy(buffer, prefix, prefix_len);
678                                 buffer += prefix_len;
679                                 memcpy(buffer, entry->e_name, entry->e_name_len);
680                                 buffer += entry->e_name_len;
681                                 *buffer++ = 0;
682                         }
683                         rest -= size;
684                 }
685         }
686         return buffer_size - rest;  /* total size */
687 }
688
689 static int
690 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
691 {
692         struct inode *inode = d_inode(dentry);
693         struct buffer_head *bh = NULL;
694         int error;
695
696         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
697                   buffer, (long)buffer_size);
698
699         error = 0;
700         if (!EXT4_I(inode)->i_file_acl)
701                 goto cleanup;
702         ea_idebug(inode, "reading block %llu",
703                   (unsigned long long)EXT4_I(inode)->i_file_acl);
704         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
705         error = -EIO;
706         if (!bh)
707                 goto cleanup;
708         ea_bdebug(bh, "b_count=%d, refcount=%d",
709                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
710         error = ext4_xattr_check_block(inode, bh);
711         if (error)
712                 goto cleanup;
713         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
714         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
715
716 cleanup:
717         brelse(bh);
718
719         return error;
720 }
721
722 static int
723 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
724 {
725         struct inode *inode = d_inode(dentry);
726         struct ext4_xattr_ibody_header *header;
727         struct ext4_inode *raw_inode;
728         struct ext4_iloc iloc;
729         void *end;
730         int error;
731
732         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
733                 return 0;
734         error = ext4_get_inode_loc(inode, &iloc);
735         if (error)
736                 return error;
737         raw_inode = ext4_raw_inode(&iloc);
738         header = IHDR(inode, raw_inode);
739         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
740         error = xattr_check_inode(inode, header, end);
741         if (error)
742                 goto cleanup;
743         error = ext4_xattr_list_entries(dentry, IFIRST(header),
744                                         buffer, buffer_size);
745
746 cleanup:
747         brelse(iloc.bh);
748         return error;
749 }
750
751 /*
752  * Inode operation listxattr()
753  *
754  * d_inode(dentry)->i_rwsem: don't care
755  *
756  * Copy a list of attribute names into the buffer
757  * provided, or compute the buffer size required.
758  * Buffer is NULL to compute the size of the buffer required.
759  *
760  * Returns a negative error number on failure, or the number of bytes
761  * used / required on success.
762  */
763 ssize_t
764 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
765 {
766         int ret, ret2;
767
768         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
769         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
770         if (ret < 0)
771                 goto errout;
772         if (buffer) {
773                 buffer += ret;
774                 buffer_size -= ret;
775         }
776         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
777         if (ret < 0)
778                 goto errout;
779         ret += ret2;
780 errout:
781         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
782         return ret;
783 }
784
785 /*
786  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
787  * not set, set it.
788  */
789 static void ext4_xattr_update_super_block(handle_t *handle,
790                                           struct super_block *sb)
791 {
792         if (ext4_has_feature_xattr(sb))
793                 return;
794
795         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
796         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
797                 ext4_set_feature_xattr(sb);
798                 ext4_handle_dirty_super(handle, sb);
799         }
800 }
801
802 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
803 {
804         struct ext4_iloc iloc = { .bh = NULL };
805         struct buffer_head *bh = NULL;
806         struct ext4_inode *raw_inode;
807         struct ext4_xattr_ibody_header *header;
808         struct ext4_xattr_entry *entry;
809         qsize_t ea_inode_refs = 0;
810         void *end;
811         int ret;
812
813         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
814
815         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
816                 ret = ext4_get_inode_loc(inode, &iloc);
817                 if (ret)
818                         goto out;
819                 raw_inode = ext4_raw_inode(&iloc);
820                 header = IHDR(inode, raw_inode);
821                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
822                 ret = xattr_check_inode(inode, header, end);
823                 if (ret)
824                         goto out;
825
826                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
827                      entry = EXT4_XATTR_NEXT(entry))
828                         if (entry->e_value_inum)
829                                 ea_inode_refs++;
830         }
831
832         if (EXT4_I(inode)->i_file_acl) {
833                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
834                 if (!bh) {
835                         ret = -EIO;
836                         goto out;
837                 }
838
839                 ret = ext4_xattr_check_block(inode, bh);
840                 if (ret)
841                         goto out;
842
843                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
844                      entry = EXT4_XATTR_NEXT(entry))
845                         if (entry->e_value_inum)
846                                 ea_inode_refs++;
847         }
848         *usage = ea_inode_refs + 1;
849         ret = 0;
850 out:
851         brelse(iloc.bh);
852         brelse(bh);
853         return ret;
854 }
855
856 static inline size_t round_up_cluster(struct inode *inode, size_t length)
857 {
858         struct super_block *sb = inode->i_sb;
859         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
860                                     inode->i_blkbits);
861         size_t mask = ~(cluster_size - 1);
862
863         return (length + cluster_size - 1) & mask;
864 }
865
866 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
867 {
868         int err;
869
870         err = dquot_alloc_inode(inode);
871         if (err)
872                 return err;
873         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
874         if (err)
875                 dquot_free_inode(inode);
876         return err;
877 }
878
879 static void ext4_xattr_inode_free_quota(struct inode *parent,
880                                         struct inode *ea_inode,
881                                         size_t len)
882 {
883         if (ea_inode &&
884             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
885                 return;
886         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
887         dquot_free_inode(parent);
888 }
889
890 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
891                              struct buffer_head *block_bh, size_t value_len,
892                              bool is_create)
893 {
894         int credits;
895         int blocks;
896
897         /*
898          * 1) Owner inode update
899          * 2) Ref count update on old xattr block
900          * 3) new xattr block
901          * 4) block bitmap update for new xattr block
902          * 5) group descriptor for new xattr block
903          * 6) block bitmap update for old xattr block
904          * 7) group descriptor for old block
905          *
906          * 6 & 7 can happen if we have two racing threads T_a and T_b
907          * which are each trying to set an xattr on inodes I_a and I_b
908          * which were both initially sharing an xattr block.
909          */
910         credits = 7;
911
912         /* Quota updates. */
913         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
914
915         /*
916          * In case of inline data, we may push out the data to a block,
917          * so we need to reserve credits for this eventuality
918          */
919         if (inode && ext4_has_inline_data(inode))
920                 credits += ext4_writepage_trans_blocks(inode) + 1;
921
922         /* We are done if ea_inode feature is not enabled. */
923         if (!ext4_has_feature_ea_inode(sb))
924                 return credits;
925
926         /* New ea_inode, inode map, block bitmap, group descriptor. */
927         credits += 4;
928
929         /* Data blocks. */
930         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
931
932         /* Indirection block or one level of extent tree. */
933         blocks += 1;
934
935         /* Block bitmap and group descriptor updates for each block. */
936         credits += blocks * 2;
937
938         /* Blocks themselves. */
939         credits += blocks;
940
941         if (!is_create) {
942                 /* Dereference ea_inode holding old xattr value.
943                  * Old ea_inode, inode map, block bitmap, group descriptor.
944                  */
945                 credits += 4;
946
947                 /* Data blocks for old ea_inode. */
948                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
949
950                 /* Indirection block or one level of extent tree for old
951                  * ea_inode.
952                  */
953                 blocks += 1;
954
955                 /* Block bitmap and group descriptor updates for each block. */
956                 credits += blocks * 2;
957         }
958
959         /* We may need to clone the existing xattr block in which case we need
960          * to increment ref counts for existing ea_inodes referenced by it.
961          */
962         if (block_bh) {
963                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
964
965                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
966                         if (entry->e_value_inum)
967                                 /* Ref count update on ea_inode. */
968                                 credits += 1;
969         }
970         return credits;
971 }
972
973 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
974                                      int credits, struct buffer_head *bh,
975                                      bool dirty, bool block_csum)
976 {
977         int error;
978
979         if (!ext4_handle_valid(handle))
980                 return 0;
981
982         if (handle->h_buffer_credits >= credits)
983                 return 0;
984
985         error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
986         if (!error)
987                 return 0;
988         if (error < 0) {
989                 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
990                 return error;
991         }
992
993         if (bh && dirty) {
994                 if (block_csum)
995                         ext4_xattr_block_csum_set(inode, bh);
996                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
997                 if (error) {
998                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
999                                      error);
1000                         return error;
1001                 }
1002         }
1003
1004         error = ext4_journal_restart(handle, credits);
1005         if (error) {
1006                 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1007                 return error;
1008         }
1009
1010         if (bh) {
1011                 error = ext4_journal_get_write_access(handle, bh);
1012                 if (error) {
1013                         ext4_warning(inode->i_sb,
1014                                      "Get write access failed (error %d)",
1015                                      error);
1016                         return error;
1017                 }
1018         }
1019         return 0;
1020 }
1021
1022 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1023                                        int ref_change)
1024 {
1025         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1026         struct ext4_iloc iloc;
1027         s64 ref_count;
1028         u32 hash;
1029         int ret;
1030
1031         inode_lock(ea_inode);
1032
1033         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1034         if (ret)
1035                 goto out;
1036
1037         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1038         ref_count += ref_change;
1039         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1040
1041         if (ref_change > 0) {
1042                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1043                           ea_inode->i_ino, ref_count);
1044
1045                 if (ref_count == 1) {
1046                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1047                                   ea_inode->i_ino, ea_inode->i_nlink);
1048
1049                         set_nlink(ea_inode, 1);
1050                         ext4_orphan_del(handle, ea_inode);
1051
1052                         if (ea_inode_cache) {
1053                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1054                                 mb_cache_entry_create(ea_inode_cache,
1055                                                       GFP_NOFS, hash,
1056                                                       ea_inode->i_ino,
1057                                                       true /* reusable */);
1058                         }
1059                 }
1060         } else {
1061                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1062                           ea_inode->i_ino, ref_count);
1063
1064                 if (ref_count == 0) {
1065                         WARN_ONCE(ea_inode->i_nlink != 1,
1066                                   "EA inode %lu i_nlink=%u",
1067                                   ea_inode->i_ino, ea_inode->i_nlink);
1068
1069                         clear_nlink(ea_inode);
1070                         ext4_orphan_add(handle, ea_inode);
1071
1072                         if (ea_inode_cache) {
1073                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1074                                 mb_cache_entry_delete(ea_inode_cache, hash,
1075                                                       ea_inode->i_ino);
1076                         }
1077                 }
1078         }
1079
1080         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1081         if (ret)
1082                 ext4_warning_inode(ea_inode,
1083                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1084 out:
1085         inode_unlock(ea_inode);
1086         return ret;
1087 }
1088
1089 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1090 {
1091         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1092 }
1093
1094 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1095 {
1096         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1097 }
1098
1099 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1100                                         struct ext4_xattr_entry *first)
1101 {
1102         struct inode *ea_inode;
1103         struct ext4_xattr_entry *entry;
1104         struct ext4_xattr_entry *failed_entry;
1105         unsigned int ea_ino;
1106         int err, saved_err;
1107
1108         for (entry = first; !IS_LAST_ENTRY(entry);
1109              entry = EXT4_XATTR_NEXT(entry)) {
1110                 if (!entry->e_value_inum)
1111                         continue;
1112                 ea_ino = le32_to_cpu(entry->e_value_inum);
1113                 err = ext4_xattr_inode_iget(parent, ea_ino,
1114                                             le32_to_cpu(entry->e_hash),
1115                                             &ea_inode);
1116                 if (err)
1117                         goto cleanup;
1118                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1119                 if (err) {
1120                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1121                         iput(ea_inode);
1122                         goto cleanup;
1123                 }
1124                 iput(ea_inode);
1125         }
1126         return 0;
1127
1128 cleanup:
1129         saved_err = err;
1130         failed_entry = entry;
1131
1132         for (entry = first; entry != failed_entry;
1133              entry = EXT4_XATTR_NEXT(entry)) {
1134                 if (!entry->e_value_inum)
1135                         continue;
1136                 ea_ino = le32_to_cpu(entry->e_value_inum);
1137                 err = ext4_xattr_inode_iget(parent, ea_ino,
1138                                             le32_to_cpu(entry->e_hash),
1139                                             &ea_inode);
1140                 if (err) {
1141                         ext4_warning(parent->i_sb,
1142                                      "cleanup ea_ino %u iget error %d", ea_ino,
1143                                      err);
1144                         continue;
1145                 }
1146                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1147                 if (err)
1148                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1149                                            err);
1150                 iput(ea_inode);
1151         }
1152         return saved_err;
1153 }
1154
1155 static void
1156 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1157                              struct buffer_head *bh,
1158                              struct ext4_xattr_entry *first, bool block_csum,
1159                              struct ext4_xattr_inode_array **ea_inode_array,
1160                              int extra_credits, bool skip_quota)
1161 {
1162         struct inode *ea_inode;
1163         struct ext4_xattr_entry *entry;
1164         bool dirty = false;
1165         unsigned int ea_ino;
1166         int err;
1167         int credits;
1168
1169         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1170         credits = 2 + extra_credits;
1171
1172         for (entry = first; !IS_LAST_ENTRY(entry);
1173              entry = EXT4_XATTR_NEXT(entry)) {
1174                 if (!entry->e_value_inum)
1175                         continue;
1176                 ea_ino = le32_to_cpu(entry->e_value_inum);
1177                 err = ext4_xattr_inode_iget(parent, ea_ino,
1178                                             le32_to_cpu(entry->e_hash),
1179                                             &ea_inode);
1180                 if (err)
1181                         continue;
1182
1183                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1184                 if (err) {
1185                         ext4_warning_inode(ea_inode,
1186                                            "Expand inode array err=%d", err);
1187                         iput(ea_inode);
1188                         continue;
1189                 }
1190
1191                 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1192                                                 dirty, block_csum);
1193                 if (err) {
1194                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1195                                            err);
1196                         continue;
1197                 }
1198
1199                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1200                 if (err) {
1201                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1202                                            err);
1203                         continue;
1204                 }
1205
1206                 if (!skip_quota)
1207                         ext4_xattr_inode_free_quota(parent, ea_inode,
1208                                               le32_to_cpu(entry->e_value_size));
1209
1210                 /*
1211                  * Forget about ea_inode within the same transaction that
1212                  * decrements the ref count. This avoids duplicate decrements in
1213                  * case the rest of the work spills over to subsequent
1214                  * transactions.
1215                  */
1216                 entry->e_value_inum = 0;
1217                 entry->e_value_size = 0;
1218
1219                 dirty = true;
1220         }
1221
1222         if (dirty) {
1223                 /*
1224                  * Note that we are deliberately skipping csum calculation for
1225                  * the final update because we do not expect any journal
1226                  * restarts until xattr block is freed.
1227                  */
1228
1229                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1230                 if (err)
1231                         ext4_warning_inode(parent,
1232                                            "handle dirty metadata err=%d", err);
1233         }
1234 }
1235
1236 /*
1237  * Release the xattr block BH: If the reference count is > 1, decrement it;
1238  * otherwise free the block.
1239  */
1240 static void
1241 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1242                          struct buffer_head *bh,
1243                          struct ext4_xattr_inode_array **ea_inode_array,
1244                          int extra_credits)
1245 {
1246         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1247         u32 hash, ref;
1248         int error = 0;
1249
1250         BUFFER_TRACE(bh, "get_write_access");
1251         error = ext4_journal_get_write_access(handle, bh);
1252         if (error)
1253                 goto out;
1254
1255         lock_buffer(bh);
1256         hash = le32_to_cpu(BHDR(bh)->h_hash);
1257         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1258         if (ref == 1) {
1259                 ea_bdebug(bh, "refcount now=0; freeing");
1260                 /*
1261                  * This must happen under buffer lock for
1262                  * ext4_xattr_block_set() to reliably detect freed block
1263                  */
1264                 if (ea_block_cache)
1265                         mb_cache_entry_delete(ea_block_cache, hash,
1266                                               bh->b_blocknr);
1267                 get_bh(bh);
1268                 unlock_buffer(bh);
1269
1270                 if (ext4_has_feature_ea_inode(inode->i_sb))
1271                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1272                                                      BFIRST(bh),
1273                                                      true /* block_csum */,
1274                                                      ea_inode_array,
1275                                                      extra_credits,
1276                                                      true /* skip_quota */);
1277                 ext4_free_blocks(handle, inode, bh, 0, 1,
1278                                  EXT4_FREE_BLOCKS_METADATA |
1279                                  EXT4_FREE_BLOCKS_FORGET);
1280         } else {
1281                 ref--;
1282                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1283                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1284                         struct mb_cache_entry *ce;
1285
1286                         if (ea_block_cache) {
1287                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1288                                                         bh->b_blocknr);
1289                                 if (ce) {
1290                                         ce->e_reusable = 1;
1291                                         mb_cache_entry_put(ea_block_cache, ce);
1292                                 }
1293                         }
1294                 }
1295
1296                 ext4_xattr_block_csum_set(inode, bh);
1297                 /*
1298                  * Beware of this ugliness: Releasing of xattr block references
1299                  * from different inodes can race and so we have to protect
1300                  * from a race where someone else frees the block (and releases
1301                  * its journal_head) before we are done dirtying the buffer. In
1302                  * nojournal mode this race is harmless and we actually cannot
1303                  * call ext4_handle_dirty_metadata() with locked buffer as
1304                  * that function can call sync_dirty_buffer() so for that case
1305                  * we handle the dirtying after unlocking the buffer.
1306                  */
1307                 if (ext4_handle_valid(handle))
1308                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1309                 unlock_buffer(bh);
1310                 if (!ext4_handle_valid(handle))
1311                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1312                 if (IS_SYNC(inode))
1313                         ext4_handle_sync(handle);
1314                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1315                 ea_bdebug(bh, "refcount now=%d; releasing",
1316                           le32_to_cpu(BHDR(bh)->h_refcount));
1317         }
1318 out:
1319         ext4_std_error(inode->i_sb, error);
1320         return;
1321 }
1322
1323 /*
1324  * Find the available free space for EAs. This also returns the total number of
1325  * bytes used by EA entries.
1326  */
1327 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1328                                     size_t *min_offs, void *base, int *total)
1329 {
1330         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1331                 if (!last->e_value_inum && last->e_value_size) {
1332                         size_t offs = le16_to_cpu(last->e_value_offs);
1333                         if (offs < *min_offs)
1334                                 *min_offs = offs;
1335                 }
1336                 if (total)
1337                         *total += EXT4_XATTR_LEN(last->e_name_len);
1338         }
1339         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1340 }
1341
1342 /*
1343  * Write the value of the EA in an inode.
1344  */
1345 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1346                                   const void *buf, int bufsize)
1347 {
1348         struct buffer_head *bh = NULL;
1349         unsigned long block = 0;
1350         int blocksize = ea_inode->i_sb->s_blocksize;
1351         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1352         int csize, wsize = 0;
1353         int ret = 0;
1354         int retries = 0;
1355
1356 retry:
1357         while (ret >= 0 && ret < max_blocks) {
1358                 struct ext4_map_blocks map;
1359                 map.m_lblk = block += ret;
1360                 map.m_len = max_blocks -= ret;
1361
1362                 ret = ext4_map_blocks(handle, ea_inode, &map,
1363                                       EXT4_GET_BLOCKS_CREATE);
1364                 if (ret <= 0) {
1365                         ext4_mark_inode_dirty(handle, ea_inode);
1366                         if (ret == -ENOSPC &&
1367                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1368                                 ret = 0;
1369                                 goto retry;
1370                         }
1371                         break;
1372                 }
1373         }
1374
1375         if (ret < 0)
1376                 return ret;
1377
1378         block = 0;
1379         while (wsize < bufsize) {
1380                 if (bh != NULL)
1381                         brelse(bh);
1382                 csize = (bufsize - wsize) > blocksize ? blocksize :
1383                                                                 bufsize - wsize;
1384                 bh = ext4_getblk(handle, ea_inode, block, 0);
1385                 if (IS_ERR(bh))
1386                         return PTR_ERR(bh);
1387                 ret = ext4_journal_get_write_access(handle, bh);
1388                 if (ret)
1389                         goto out;
1390
1391                 memcpy(bh->b_data, buf, csize);
1392                 set_buffer_uptodate(bh);
1393                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1394
1395                 buf += csize;
1396                 wsize += csize;
1397                 block += 1;
1398         }
1399
1400         inode_lock(ea_inode);
1401         i_size_write(ea_inode, wsize);
1402         ext4_update_i_disksize(ea_inode, wsize);
1403         inode_unlock(ea_inode);
1404
1405         ext4_mark_inode_dirty(handle, ea_inode);
1406
1407 out:
1408         brelse(bh);
1409
1410         return ret;
1411 }
1412
1413 /*
1414  * Create an inode to store the value of a large EA.
1415  */
1416 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1417                                              struct inode *inode, u32 hash)
1418 {
1419         struct inode *ea_inode = NULL;
1420         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1421         int err;
1422
1423         /*
1424          * Let the next inode be the goal, so we try and allocate the EA inode
1425          * in the same group, or nearby one.
1426          */
1427         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1428                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1429                                   EXT4_EA_INODE_FL);
1430         if (!IS_ERR(ea_inode)) {
1431                 ea_inode->i_op = &ext4_file_inode_operations;
1432                 ea_inode->i_fop = &ext4_file_operations;
1433                 ext4_set_aops(ea_inode);
1434                 ext4_xattr_inode_set_class(ea_inode);
1435                 unlock_new_inode(ea_inode);
1436                 ext4_xattr_inode_set_ref(ea_inode, 1);
1437                 ext4_xattr_inode_set_hash(ea_inode, hash);
1438                 err = ext4_mark_inode_dirty(handle, ea_inode);
1439                 if (!err)
1440                         err = ext4_inode_attach_jinode(ea_inode);
1441                 if (err) {
1442                         iput(ea_inode);
1443                         return ERR_PTR(err);
1444                 }
1445
1446                 /*
1447                  * Xattr inodes are shared therefore quota charging is performed
1448                  * at a higher level.
1449                  */
1450                 dquot_free_inode(ea_inode);
1451                 dquot_drop(ea_inode);
1452                 inode_lock(ea_inode);
1453                 ea_inode->i_flags |= S_NOQUOTA;
1454                 inode_unlock(ea_inode);
1455         }
1456
1457         return ea_inode;
1458 }
1459
1460 static struct inode *
1461 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1462                             size_t value_len, u32 hash)
1463 {
1464         struct inode *ea_inode;
1465         struct mb_cache_entry *ce;
1466         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1467         void *ea_data;
1468
1469         if (!ea_inode_cache)
1470                 return NULL;
1471
1472         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1473         if (!ce)
1474                 return NULL;
1475
1476         ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1477         if (!ea_data) {
1478                 mb_cache_entry_put(ea_inode_cache, ce);
1479                 return NULL;
1480         }
1481
1482         while (ce) {
1483                 ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1484                 if (!IS_ERR(ea_inode) &&
1485                     !is_bad_inode(ea_inode) &&
1486                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1487                     i_size_read(ea_inode) == value_len &&
1488                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1489                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1490                                                     value_len) &&
1491                     !memcmp(value, ea_data, value_len)) {
1492                         mb_cache_entry_touch(ea_inode_cache, ce);
1493                         mb_cache_entry_put(ea_inode_cache, ce);
1494                         kvfree(ea_data);
1495                         return ea_inode;
1496                 }
1497
1498                 if (!IS_ERR(ea_inode))
1499                         iput(ea_inode);
1500                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1501         }
1502         kvfree(ea_data);
1503         return NULL;
1504 }
1505
1506 /*
1507  * Add value of the EA in an inode.
1508  */
1509 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1510                                           const void *value, size_t value_len,
1511                                           struct inode **ret_inode)
1512 {
1513         struct inode *ea_inode;
1514         u32 hash;
1515         int err;
1516
1517         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1518         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1519         if (ea_inode) {
1520                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1521                 if (err) {
1522                         iput(ea_inode);
1523                         return err;
1524                 }
1525
1526                 *ret_inode = ea_inode;
1527                 return 0;
1528         }
1529
1530         /* Create an inode for the EA value */
1531         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1532         if (IS_ERR(ea_inode))
1533                 return PTR_ERR(ea_inode);
1534
1535         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1536         if (err) {
1537                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1538                 iput(ea_inode);
1539                 return err;
1540         }
1541
1542         if (EA_INODE_CACHE(inode))
1543                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1544                                       ea_inode->i_ino, true /* reusable */);
1545
1546         *ret_inode = ea_inode;
1547         return 0;
1548 }
1549
1550 /*
1551  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1552  * feature is enabled.
1553  */
1554 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1555
1556 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1557                                 struct ext4_xattr_search *s,
1558                                 handle_t *handle, struct inode *inode,
1559                                 bool is_block)
1560 {
1561         struct ext4_xattr_entry *last, *next;
1562         struct ext4_xattr_entry *here = s->here;
1563         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1564         int in_inode = i->in_inode;
1565         struct inode *old_ea_inode = NULL;
1566         struct inode *new_ea_inode = NULL;
1567         size_t old_size, new_size;
1568         int ret;
1569
1570         /* Space used by old and new values. */
1571         old_size = (!s->not_found && !here->e_value_inum) ?
1572                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1573         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1574
1575         /*
1576          * Optimization for the simple case when old and new values have the
1577          * same padded sizes. Not applicable if external inodes are involved.
1578          */
1579         if (new_size && new_size == old_size) {
1580                 size_t offs = le16_to_cpu(here->e_value_offs);
1581                 void *val = s->base + offs;
1582
1583                 here->e_value_size = cpu_to_le32(i->value_len);
1584                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1585                         memset(val, 0, new_size);
1586                 } else {
1587                         memcpy(val, i->value, i->value_len);
1588                         /* Clear padding bytes. */
1589                         memset(val + i->value_len, 0, new_size - i->value_len);
1590                 }
1591                 goto update_hash;
1592         }
1593
1594         /* Compute min_offs and last. */
1595         last = s->first;
1596         for (; !IS_LAST_ENTRY(last); last = next) {
1597                 next = EXT4_XATTR_NEXT(last);
1598                 if ((void *)next >= s->end) {
1599                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1600                         ret = -EFSCORRUPTED;
1601                         goto out;
1602                 }
1603                 if (!last->e_value_inum && last->e_value_size) {
1604                         size_t offs = le16_to_cpu(last->e_value_offs);
1605                         if (offs < min_offs)
1606                                 min_offs = offs;
1607                 }
1608         }
1609
1610         /* Check whether we have enough space. */
1611         if (i->value) {
1612                 size_t free;
1613
1614                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1615                 if (!s->not_found)
1616                         free += EXT4_XATTR_LEN(name_len) + old_size;
1617
1618                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1619                         ret = -ENOSPC;
1620                         goto out;
1621                 }
1622
1623                 /*
1624                  * If storing the value in an external inode is an option,
1625                  * reserve space for xattr entries/names in the external
1626                  * attribute block so that a long value does not occupy the
1627                  * whole space and prevent futher entries being added.
1628                  */
1629                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1630                     new_size && is_block &&
1631                     (min_offs + old_size - new_size) <
1632                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1633                         ret = -ENOSPC;
1634                         goto out;
1635                 }
1636         }
1637
1638         /*
1639          * Getting access to old and new ea inodes is subject to failures.
1640          * Finish that work before doing any modifications to the xattr data.
1641          */
1642         if (!s->not_found && here->e_value_inum) {
1643                 ret = ext4_xattr_inode_iget(inode,
1644                                             le32_to_cpu(here->e_value_inum),
1645                                             le32_to_cpu(here->e_hash),
1646                                             &old_ea_inode);
1647                 if (ret) {
1648                         old_ea_inode = NULL;
1649                         goto out;
1650                 }
1651         }
1652         if (i->value && in_inode) {
1653                 WARN_ON_ONCE(!i->value_len);
1654
1655                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1656                 if (ret)
1657                         goto out;
1658
1659                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1660                                                      i->value_len,
1661                                                      &new_ea_inode);
1662                 if (ret) {
1663                         new_ea_inode = NULL;
1664                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1665                         goto out;
1666                 }
1667         }
1668
1669         if (old_ea_inode) {
1670                 /* We are ready to release ref count on the old_ea_inode. */
1671                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1672                 if (ret) {
1673                         /* Release newly required ref count on new_ea_inode. */
1674                         if (new_ea_inode) {
1675                                 int err;
1676
1677                                 err = ext4_xattr_inode_dec_ref(handle,
1678                                                                new_ea_inode);
1679                                 if (err)
1680                                         ext4_warning_inode(new_ea_inode,
1681                                                   "dec ref new_ea_inode err=%d",
1682                                                   err);
1683                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1684                                                             i->value_len);
1685                         }
1686                         goto out;
1687                 }
1688
1689                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1690                                             le32_to_cpu(here->e_value_size));
1691         }
1692
1693         /* No failures allowed past this point. */
1694
1695         if (!s->not_found && here->e_value_size && here->e_value_offs) {
1696                 /* Remove the old value. */
1697                 void *first_val = s->base + min_offs;
1698                 size_t offs = le16_to_cpu(here->e_value_offs);
1699                 void *val = s->base + offs;
1700
1701                 memmove(first_val + old_size, first_val, val - first_val);
1702                 memset(first_val, 0, old_size);
1703                 min_offs += old_size;
1704
1705                 /* Adjust all value offsets. */
1706                 last = s->first;
1707                 while (!IS_LAST_ENTRY(last)) {
1708                         size_t o = le16_to_cpu(last->e_value_offs);
1709
1710                         if (!last->e_value_inum &&
1711                             last->e_value_size && o < offs)
1712                                 last->e_value_offs = cpu_to_le16(o + old_size);
1713                         last = EXT4_XATTR_NEXT(last);
1714                 }
1715         }
1716
1717         if (!i->value) {
1718                 /* Remove old name. */
1719                 size_t size = EXT4_XATTR_LEN(name_len);
1720
1721                 last = ENTRY((void *)last - size);
1722                 memmove(here, (void *)here + size,
1723                         (void *)last - (void *)here + sizeof(__u32));
1724                 memset(last, 0, size);
1725         } else if (s->not_found) {
1726                 /* Insert new name. */
1727                 size_t size = EXT4_XATTR_LEN(name_len);
1728                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1729
1730                 memmove((void *)here + size, here, rest);
1731                 memset(here, 0, size);
1732                 here->e_name_index = i->name_index;
1733                 here->e_name_len = name_len;
1734                 memcpy(here->e_name, i->name, name_len);
1735         } else {
1736                 /* This is an update, reset value info. */
1737                 here->e_value_inum = 0;
1738                 here->e_value_offs = 0;
1739                 here->e_value_size = 0;
1740         }
1741
1742         if (i->value) {
1743                 /* Insert new value. */
1744                 if (in_inode) {
1745                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1746                 } else if (i->value_len) {
1747                         void *val = s->base + min_offs - new_size;
1748
1749                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1750                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1751                                 memset(val, 0, new_size);
1752                         } else {
1753                                 memcpy(val, i->value, i->value_len);
1754                                 /* Clear padding bytes. */
1755                                 memset(val + i->value_len, 0,
1756                                        new_size - i->value_len);
1757                         }
1758                 }
1759                 here->e_value_size = cpu_to_le32(i->value_len);
1760         }
1761
1762 update_hash:
1763         if (i->value) {
1764                 __le32 hash = 0;
1765
1766                 /* Entry hash calculation. */
1767                 if (in_inode) {
1768                         __le32 crc32c_hash;
1769
1770                         /*
1771                          * Feed crc32c hash instead of the raw value for entry
1772                          * hash calculation. This is to avoid walking
1773                          * potentially long value buffer again.
1774                          */
1775                         crc32c_hash = cpu_to_le32(
1776                                        ext4_xattr_inode_get_hash(new_ea_inode));
1777                         hash = ext4_xattr_hash_entry(here->e_name,
1778                                                      here->e_name_len,
1779                                                      &crc32c_hash, 1);
1780                 } else if (is_block) {
1781                         __le32 *value = s->base + le16_to_cpu(
1782                                                         here->e_value_offs);
1783
1784                         hash = ext4_xattr_hash_entry(here->e_name,
1785                                                      here->e_name_len, value,
1786                                                      new_size >> 2);
1787                 }
1788                 here->e_hash = hash;
1789         }
1790
1791         if (is_block)
1792                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1793
1794         ret = 0;
1795 out:
1796         iput(old_ea_inode);
1797         iput(new_ea_inode);
1798         return ret;
1799 }
1800
1801 struct ext4_xattr_block_find {
1802         struct ext4_xattr_search s;
1803         struct buffer_head *bh;
1804 };
1805
1806 static int
1807 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1808                       struct ext4_xattr_block_find *bs)
1809 {
1810         struct super_block *sb = inode->i_sb;
1811         int error;
1812
1813         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1814                   i->name_index, i->name, i->value, (long)i->value_len);
1815
1816         if (EXT4_I(inode)->i_file_acl) {
1817                 /* The inode already has an extended attribute block. */
1818                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1819                 error = -EIO;
1820                 if (!bs->bh)
1821                         goto cleanup;
1822                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1823                         atomic_read(&(bs->bh->b_count)),
1824                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1825                 error = ext4_xattr_check_block(inode, bs->bh);
1826                 if (error)
1827                         goto cleanup;
1828                 /* Find the named attribute. */
1829                 bs->s.base = BHDR(bs->bh);
1830                 bs->s.first = BFIRST(bs->bh);
1831                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1832                 bs->s.here = bs->s.first;
1833                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1834                                          i->name_index, i->name, 1);
1835                 if (error && error != -ENODATA)
1836                         goto cleanup;
1837                 bs->s.not_found = error;
1838         }
1839         error = 0;
1840
1841 cleanup:
1842         return error;
1843 }
1844
1845 static int
1846 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1847                      struct ext4_xattr_info *i,
1848                      struct ext4_xattr_block_find *bs)
1849 {
1850         struct super_block *sb = inode->i_sb;
1851         struct buffer_head *new_bh = NULL;
1852         struct ext4_xattr_search s_copy = bs->s;
1853         struct ext4_xattr_search *s = &s_copy;
1854         struct mb_cache_entry *ce = NULL;
1855         int error = 0;
1856         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1857         struct inode *ea_inode = NULL, *tmp_inode;
1858         size_t old_ea_inode_quota = 0;
1859         unsigned int ea_ino;
1860
1861
1862 #define header(x) ((struct ext4_xattr_header *)(x))
1863
1864         if (s->base) {
1865                 BUFFER_TRACE(bs->bh, "get_write_access");
1866                 error = ext4_journal_get_write_access(handle, bs->bh);
1867                 if (error)
1868                         goto cleanup;
1869                 lock_buffer(bs->bh);
1870
1871                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1872                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1873
1874                         /*
1875                          * This must happen under buffer lock for
1876                          * ext4_xattr_block_set() to reliably detect modified
1877                          * block
1878                          */
1879                         if (ea_block_cache)
1880                                 mb_cache_entry_delete(ea_block_cache, hash,
1881                                                       bs->bh->b_blocknr);
1882                         ea_bdebug(bs->bh, "modifying in-place");
1883                         error = ext4_xattr_set_entry(i, s, handle, inode,
1884                                                      true /* is_block */);
1885                         ext4_xattr_block_csum_set(inode, bs->bh);
1886                         unlock_buffer(bs->bh);
1887                         if (error == -EFSCORRUPTED)
1888                                 goto bad_block;
1889                         if (!error)
1890                                 error = ext4_handle_dirty_metadata(handle,
1891                                                                    inode,
1892                                                                    bs->bh);
1893                         if (error)
1894                                 goto cleanup;
1895                         goto inserted;
1896                 } else {
1897                         int offset = (char *)s->here - bs->bh->b_data;
1898
1899                         unlock_buffer(bs->bh);
1900                         ea_bdebug(bs->bh, "cloning");
1901                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1902                         error = -ENOMEM;
1903                         if (s->base == NULL)
1904                                 goto cleanup;
1905                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1906                         s->first = ENTRY(header(s->base)+1);
1907                         header(s->base)->h_refcount = cpu_to_le32(1);
1908                         s->here = ENTRY(s->base + offset);
1909                         s->end = s->base + bs->bh->b_size;
1910
1911                         /*
1912                          * If existing entry points to an xattr inode, we need
1913                          * to prevent ext4_xattr_set_entry() from decrementing
1914                          * ref count on it because the reference belongs to the
1915                          * original block. In this case, make the entry look
1916                          * like it has an empty value.
1917                          */
1918                         if (!s->not_found && s->here->e_value_inum) {
1919                                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1920                                 error = ext4_xattr_inode_iget(inode, ea_ino,
1921                                               le32_to_cpu(s->here->e_hash),
1922                                               &tmp_inode);
1923                                 if (error)
1924                                         goto cleanup;
1925
1926                                 if (!ext4_test_inode_state(tmp_inode,
1927                                                 EXT4_STATE_LUSTRE_EA_INODE)) {
1928                                         /*
1929                                          * Defer quota free call for previous
1930                                          * inode until success is guaranteed.
1931                                          */
1932                                         old_ea_inode_quota = le32_to_cpu(
1933                                                         s->here->e_value_size);
1934                                 }
1935                                 iput(tmp_inode);
1936
1937                                 s->here->e_value_inum = 0;
1938                                 s->here->e_value_size = 0;
1939                         }
1940                 }
1941         } else {
1942                 /* Allocate a buffer where we construct the new block. */
1943                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1944                 /* assert(header == s->base) */
1945                 error = -ENOMEM;
1946                 if (s->base == NULL)
1947                         goto cleanup;
1948                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1949                 header(s->base)->h_blocks = cpu_to_le32(1);
1950                 header(s->base)->h_refcount = cpu_to_le32(1);
1951                 s->first = ENTRY(header(s->base)+1);
1952                 s->here = ENTRY(header(s->base)+1);
1953                 s->end = s->base + sb->s_blocksize;
1954         }
1955
1956         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1957         if (error == -EFSCORRUPTED)
1958                 goto bad_block;
1959         if (error)
1960                 goto cleanup;
1961
1962         if (i->value && s->here->e_value_inum) {
1963                 /*
1964                  * A ref count on ea_inode has been taken as part of the call to
1965                  * ext4_xattr_set_entry() above. We would like to drop this
1966                  * extra ref but we have to wait until the xattr block is
1967                  * initialized and has its own ref count on the ea_inode.
1968                  */
1969                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1970                 error = ext4_xattr_inode_iget(inode, ea_ino,
1971                                               le32_to_cpu(s->here->e_hash),
1972                                               &ea_inode);
1973                 if (error) {
1974                         ea_inode = NULL;
1975                         goto cleanup;
1976                 }
1977         }
1978
1979 inserted:
1980         if (!IS_LAST_ENTRY(s->first)) {
1981                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1982                                                      &ce);
1983                 if (new_bh) {
1984                         /* We found an identical block in the cache. */
1985                         if (new_bh == bs->bh)
1986                                 ea_bdebug(new_bh, "keeping");
1987                         else {
1988                                 u32 ref;
1989
1990                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
1991
1992                                 /* The old block is released after updating
1993                                    the inode. */
1994                                 error = dquot_alloc_block(inode,
1995                                                 EXT4_C2B(EXT4_SB(sb), 1));
1996                                 if (error)
1997                                         goto cleanup;
1998                                 BUFFER_TRACE(new_bh, "get_write_access");
1999                                 error = ext4_journal_get_write_access(handle,
2000                                                                       new_bh);
2001                                 if (error)
2002                                         goto cleanup_dquot;
2003                                 lock_buffer(new_bh);
2004                                 /*
2005                                  * We have to be careful about races with
2006                                  * freeing, rehashing or adding references to
2007                                  * xattr block. Once we hold buffer lock xattr
2008                                  * block's state is stable so we can check
2009                                  * whether the block got freed / rehashed or
2010                                  * not.  Since we unhash mbcache entry under
2011                                  * buffer lock when freeing / rehashing xattr
2012                                  * block, checking whether entry is still
2013                                  * hashed is reliable. Same rules hold for
2014                                  * e_reusable handling.
2015                                  */
2016                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2017                                     !ce->e_reusable) {
2018                                         /*
2019                                          * Undo everything and check mbcache
2020                                          * again.
2021                                          */
2022                                         unlock_buffer(new_bh);
2023                                         dquot_free_block(inode,
2024                                                          EXT4_C2B(EXT4_SB(sb),
2025                                                                   1));
2026                                         brelse(new_bh);
2027                                         mb_cache_entry_put(ea_block_cache, ce);
2028                                         ce = NULL;
2029                                         new_bh = NULL;
2030                                         goto inserted;
2031                                 }
2032                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2033                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2034                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2035                                         ce->e_reusable = 0;
2036                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2037                                           ref);
2038                                 ext4_xattr_block_csum_set(inode, new_bh);
2039                                 unlock_buffer(new_bh);
2040                                 error = ext4_handle_dirty_metadata(handle,
2041                                                                    inode,
2042                                                                    new_bh);
2043                                 if (error)
2044                                         goto cleanup_dquot;
2045                         }
2046                         mb_cache_entry_touch(ea_block_cache, ce);
2047                         mb_cache_entry_put(ea_block_cache, ce);
2048                         ce = NULL;
2049                 } else if (bs->bh && s->base == bs->bh->b_data) {
2050                         /* We were modifying this block in-place. */
2051                         ea_bdebug(bs->bh, "keeping this block");
2052                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2053                         new_bh = bs->bh;
2054                         get_bh(new_bh);
2055                 } else {
2056                         /* We need to allocate a new block */
2057                         ext4_fsblk_t goal, block;
2058
2059                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2060
2061                         goal = ext4_group_first_block_no(sb,
2062                                                 EXT4_I(inode)->i_block_group);
2063
2064                         /* non-extent files can't have physical blocks past 2^32 */
2065                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2066                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2067
2068                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2069                                                      NULL, &error);
2070                         if (error)
2071                                 goto cleanup;
2072
2073                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2074                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2075
2076                         ea_idebug(inode, "creating block %llu",
2077                                   (unsigned long long)block);
2078
2079                         new_bh = sb_getblk(sb, block);
2080                         if (unlikely(!new_bh)) {
2081                                 error = -ENOMEM;
2082 getblk_failed:
2083                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2084                                                  EXT4_FREE_BLOCKS_METADATA);
2085                                 goto cleanup;
2086                         }
2087                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2088                                                       ENTRY(header(s->base)+1));
2089                         if (error)
2090                                 goto getblk_failed;
2091                         if (ea_inode) {
2092                                 /* Drop the extra ref on ea_inode. */
2093                                 error = ext4_xattr_inode_dec_ref(handle,
2094                                                                  ea_inode);
2095                                 if (error)
2096                                         ext4_warning_inode(ea_inode,
2097                                                            "dec ref error=%d",
2098                                                            error);
2099                                 iput(ea_inode);
2100                                 ea_inode = NULL;
2101                         }
2102
2103                         lock_buffer(new_bh);
2104                         error = ext4_journal_get_create_access(handle, new_bh);
2105                         if (error) {
2106                                 unlock_buffer(new_bh);
2107                                 error = -EIO;
2108                                 goto getblk_failed;
2109                         }
2110                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2111                         ext4_xattr_block_csum_set(inode, new_bh);
2112                         set_buffer_uptodate(new_bh);
2113                         unlock_buffer(new_bh);
2114                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2115                         error = ext4_handle_dirty_metadata(handle, inode,
2116                                                            new_bh);
2117                         if (error)
2118                                 goto cleanup;
2119                 }
2120         }
2121
2122         if (old_ea_inode_quota)
2123                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2124
2125         /* Update the inode. */
2126         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2127
2128         /* Drop the previous xattr block. */
2129         if (bs->bh && bs->bh != new_bh) {
2130                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2131
2132                 ext4_xattr_release_block(handle, inode, bs->bh,
2133                                          &ea_inode_array,
2134                                          0 /* extra_credits */);
2135                 ext4_xattr_inode_array_free(ea_inode_array);
2136         }
2137         error = 0;
2138
2139 cleanup:
2140         if (ea_inode) {
2141                 int error2;
2142
2143                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2144                 if (error2)
2145                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2146                                            error2);
2147
2148                 /* If there was an error, revert the quota charge. */
2149                 if (error)
2150                         ext4_xattr_inode_free_quota(inode, ea_inode,
2151                                                     i_size_read(ea_inode));
2152                 iput(ea_inode);
2153         }
2154         if (ce)
2155                 mb_cache_entry_put(ea_block_cache, ce);
2156         brelse(new_bh);
2157         if (!(bs->bh && s->base == bs->bh->b_data))
2158                 kfree(s->base);
2159
2160         return error;
2161
2162 cleanup_dquot:
2163         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2164         goto cleanup;
2165
2166 bad_block:
2167         EXT4_ERROR_INODE(inode, "bad block %llu",
2168                          EXT4_I(inode)->i_file_acl);
2169         goto cleanup;
2170
2171 #undef header
2172 }
2173
2174 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2175                           struct ext4_xattr_ibody_find *is)
2176 {
2177         struct ext4_xattr_ibody_header *header;
2178         struct ext4_inode *raw_inode;
2179         int error;
2180
2181         if (EXT4_I(inode)->i_extra_isize == 0)
2182                 return 0;
2183         raw_inode = ext4_raw_inode(&is->iloc);
2184         header = IHDR(inode, raw_inode);
2185         is->s.base = is->s.first = IFIRST(header);
2186         is->s.here = is->s.first;
2187         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2188         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2189                 error = xattr_check_inode(inode, header, is->s.end);
2190                 if (error)
2191                         return error;
2192                 /* Find the named attribute. */
2193                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2194                                          i->name_index, i->name, 0);
2195                 if (error && error != -ENODATA)
2196                         return error;
2197                 is->s.not_found = error;
2198         }
2199         return 0;
2200 }
2201
2202 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2203                                 struct ext4_xattr_info *i,
2204                                 struct ext4_xattr_ibody_find *is)
2205 {
2206         struct ext4_xattr_ibody_header *header;
2207         struct ext4_xattr_search *s = &is->s;
2208         int error;
2209
2210         if (EXT4_I(inode)->i_extra_isize == 0)
2211                 return -ENOSPC;
2212         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2213         if (error)
2214                 return error;
2215         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2216         if (!IS_LAST_ENTRY(s->first)) {
2217                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2218                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2219         } else {
2220                 header->h_magic = cpu_to_le32(0);
2221                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2222         }
2223         return 0;
2224 }
2225
2226 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2227                                 struct ext4_xattr_info *i,
2228                                 struct ext4_xattr_ibody_find *is)
2229 {
2230         struct ext4_xattr_ibody_header *header;
2231         struct ext4_xattr_search *s = &is->s;
2232         int error;
2233
2234         if (EXT4_I(inode)->i_extra_isize == 0)
2235                 return -ENOSPC;
2236         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2237         if (error)
2238                 return error;
2239         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2240         if (!IS_LAST_ENTRY(s->first)) {
2241                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2242                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2243         } else {
2244                 header->h_magic = cpu_to_le32(0);
2245                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2246         }
2247         return 0;
2248 }
2249
2250 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2251                                  struct ext4_xattr_info *i)
2252 {
2253         void *value;
2254
2255         /* When e_value_inum is set the value is stored externally. */
2256         if (s->here->e_value_inum)
2257                 return 0;
2258         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2259                 return 0;
2260         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2261         return !memcmp(value, i->value, i->value_len);
2262 }
2263
2264 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2265 {
2266         struct buffer_head *bh;
2267         int error;
2268
2269         if (!EXT4_I(inode)->i_file_acl)
2270                 return NULL;
2271         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2272         if (!bh)
2273                 return ERR_PTR(-EIO);
2274         error = ext4_xattr_check_block(inode, bh);
2275         if (error) {
2276                 brelse(bh);
2277                 return ERR_PTR(error);
2278         }
2279         return bh;
2280 }
2281
2282 /*
2283  * ext4_xattr_set_handle()
2284  *
2285  * Create, replace or remove an extended attribute for this inode.  Value
2286  * is NULL to remove an existing extended attribute, and non-NULL to
2287  * either replace an existing extended attribute, or create a new extended
2288  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2289  * specify that an extended attribute must exist and must not exist
2290  * previous to the call, respectively.
2291  *
2292  * Returns 0, or a negative error number on failure.
2293  */
2294 int
2295 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2296                       const char *name, const void *value, size_t value_len,
2297                       int flags)
2298 {
2299         struct ext4_xattr_info i = {
2300                 .name_index = name_index,
2301                 .name = name,
2302                 .value = value,
2303                 .value_len = value_len,
2304                 .in_inode = 0,
2305         };
2306         struct ext4_xattr_ibody_find is = {
2307                 .s = { .not_found = -ENODATA, },
2308         };
2309         struct ext4_xattr_block_find bs = {
2310                 .s = { .not_found = -ENODATA, },
2311         };
2312         int no_expand;
2313         int error;
2314
2315         if (!name)
2316                 return -EINVAL;
2317         if (strlen(name) > 255)
2318                 return -ERANGE;
2319
2320         ext4_write_lock_xattr(inode, &no_expand);
2321
2322         /* Check journal credits under write lock. */
2323         if (ext4_handle_valid(handle)) {
2324                 struct buffer_head *bh;
2325                 int credits;
2326
2327                 bh = ext4_xattr_get_block(inode);
2328                 if (IS_ERR(bh)) {
2329                         error = PTR_ERR(bh);
2330                         goto cleanup;
2331                 }
2332
2333                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2334                                                    value_len,
2335                                                    flags & XATTR_CREATE);
2336                 brelse(bh);
2337
2338                 if (!ext4_handle_has_enough_credits(handle, credits)) {
2339                         error = -ENOSPC;
2340                         goto cleanup;
2341                 }
2342         }
2343
2344         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2345         if (error)
2346                 goto cleanup;
2347
2348         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2349                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2350                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2351                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2352         }
2353
2354         error = ext4_xattr_ibody_find(inode, &i, &is);
2355         if (error)
2356                 goto cleanup;
2357         if (is.s.not_found)
2358                 error = ext4_xattr_block_find(inode, &i, &bs);
2359         if (error)
2360                 goto cleanup;
2361         if (is.s.not_found && bs.s.not_found) {
2362                 error = -ENODATA;
2363                 if (flags & XATTR_REPLACE)
2364                         goto cleanup;
2365                 error = 0;
2366                 if (!value)
2367                         goto cleanup;
2368         } else {
2369                 error = -EEXIST;
2370                 if (flags & XATTR_CREATE)
2371                         goto cleanup;
2372         }
2373
2374         if (!value) {
2375                 if (!is.s.not_found)
2376                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2377                 else if (!bs.s.not_found)
2378                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2379         } else {
2380                 error = 0;
2381                 /* Xattr value did not change? Save us some work and bail out */
2382                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2383                         goto cleanup;
2384                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2385                         goto cleanup;
2386
2387                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2388                     (EXT4_XATTR_SIZE(i.value_len) >
2389                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2390                         i.in_inode = 1;
2391 retry_inode:
2392                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2393                 if (!error && !bs.s.not_found) {
2394                         i.value = NULL;
2395                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2396                 } else if (error == -ENOSPC) {
2397                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2398                                 brelse(bs.bh);
2399                                 bs.bh = NULL;
2400                                 error = ext4_xattr_block_find(inode, &i, &bs);
2401                                 if (error)
2402                                         goto cleanup;
2403                         }
2404                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2405                         if (!error && !is.s.not_found) {
2406                                 i.value = NULL;
2407                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2408                                                              &is);
2409                         } else if (error == -ENOSPC) {
2410                                 /*
2411                                  * Xattr does not fit in the block, store at
2412                                  * external inode if possible.
2413                                  */
2414                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2415                                     !i.in_inode) {
2416                                         i.in_inode = 1;
2417                                         goto retry_inode;
2418                                 }
2419                         }
2420                 }
2421         }
2422         if (!error) {
2423                 ext4_xattr_update_super_block(handle, inode->i_sb);
2424                 inode->i_ctime = current_time(inode);
2425                 if (!value)
2426                         no_expand = 0;
2427                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2428                 /*
2429                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2430                  * error != 0.
2431                  */
2432                 is.iloc.bh = NULL;
2433                 if (IS_SYNC(inode))
2434                         ext4_handle_sync(handle);
2435         }
2436
2437 cleanup:
2438         brelse(is.iloc.bh);
2439         brelse(bs.bh);
2440         ext4_write_unlock_xattr(inode, &no_expand);
2441         return error;
2442 }
2443
2444 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2445                            bool is_create, int *credits)
2446 {
2447         struct buffer_head *bh;
2448         int err;
2449
2450         *credits = 0;
2451
2452         if (!EXT4_SB(inode->i_sb)->s_journal)
2453                 return 0;
2454
2455         down_read(&EXT4_I(inode)->xattr_sem);
2456
2457         bh = ext4_xattr_get_block(inode);
2458         if (IS_ERR(bh)) {
2459                 err = PTR_ERR(bh);
2460         } else {
2461                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2462                                                     value_len, is_create);
2463                 brelse(bh);
2464                 err = 0;
2465         }
2466
2467         up_read(&EXT4_I(inode)->xattr_sem);
2468         return err;
2469 }
2470
2471 /*
2472  * ext4_xattr_set()
2473  *
2474  * Like ext4_xattr_set_handle, but start from an inode. This extended
2475  * attribute modification is a filesystem transaction by itself.
2476  *
2477  * Returns 0, or a negative error number on failure.
2478  */
2479 int
2480 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2481                const void *value, size_t value_len, int flags)
2482 {
2483         handle_t *handle;
2484         struct super_block *sb = inode->i_sb;
2485         int error, retries = 0;
2486         int credits;
2487
2488         error = dquot_initialize(inode);
2489         if (error)
2490                 return error;
2491
2492 retry:
2493         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2494                                        &credits);
2495         if (error)
2496                 return error;
2497
2498         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2499         if (IS_ERR(handle)) {
2500                 error = PTR_ERR(handle);
2501         } else {
2502                 int error2;
2503
2504                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2505                                               value, value_len, flags);
2506                 error2 = ext4_journal_stop(handle);
2507                 if (error == -ENOSPC &&
2508                     ext4_should_retry_alloc(sb, &retries))
2509                         goto retry;
2510                 if (error == 0)
2511                         error = error2;
2512         }
2513
2514         return error;
2515 }
2516
2517 /*
2518  * Shift the EA entries in the inode to create space for the increased
2519  * i_extra_isize.
2520  */
2521 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2522                                      int value_offs_shift, void *to,
2523                                      void *from, size_t n)
2524 {
2525         struct ext4_xattr_entry *last = entry;
2526         int new_offs;
2527
2528         /* We always shift xattr headers further thus offsets get lower */
2529         BUG_ON(value_offs_shift > 0);
2530
2531         /* Adjust the value offsets of the entries */
2532         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2533                 if (!last->e_value_inum && last->e_value_size) {
2534                         new_offs = le16_to_cpu(last->e_value_offs) +
2535                                                         value_offs_shift;
2536                         last->e_value_offs = cpu_to_le16(new_offs);
2537                 }
2538         }
2539         /* Shift the entries by n bytes */
2540         memmove(to, from, n);
2541 }
2542
2543 /*
2544  * Move xattr pointed to by 'entry' from inode into external xattr block
2545  */
2546 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2547                                     struct ext4_inode *raw_inode,
2548                                     struct ext4_xattr_entry *entry)
2549 {
2550         struct ext4_xattr_ibody_find *is = NULL;
2551         struct ext4_xattr_block_find *bs = NULL;
2552         char *buffer = NULL, *b_entry_name = NULL;
2553         size_t value_size = le32_to_cpu(entry->e_value_size);
2554         struct ext4_xattr_info i = {
2555                 .value = NULL,
2556                 .value_len = 0,
2557                 .name_index = entry->e_name_index,
2558                 .in_inode = !!entry->e_value_inum,
2559         };
2560         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2561         int error;
2562
2563         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2564         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2565         buffer = kmalloc(value_size, GFP_NOFS);
2566         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2567         if (!is || !bs || !buffer || !b_entry_name) {
2568                 error = -ENOMEM;
2569                 goto out;
2570         }
2571
2572         is->s.not_found = -ENODATA;
2573         bs->s.not_found = -ENODATA;
2574         is->iloc.bh = NULL;
2575         bs->bh = NULL;
2576
2577         /* Save the entry name and the entry value */
2578         if (entry->e_value_inum) {
2579                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2580                 if (error)
2581                         goto out;
2582         } else {
2583                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2584                 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2585         }
2586
2587         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2588         b_entry_name[entry->e_name_len] = '\0';
2589         i.name = b_entry_name;
2590
2591         error = ext4_get_inode_loc(inode, &is->iloc);
2592         if (error)
2593                 goto out;
2594
2595         error = ext4_xattr_ibody_find(inode, &i, is);
2596         if (error)
2597                 goto out;
2598
2599         /* Remove the chosen entry from the inode */
2600         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2601         if (error)
2602                 goto out;
2603
2604         i.value = buffer;
2605         i.value_len = value_size;
2606         error = ext4_xattr_block_find(inode, &i, bs);
2607         if (error)
2608                 goto out;
2609
2610         /* Add entry which was removed from the inode into the block */
2611         error = ext4_xattr_block_set(handle, inode, &i, bs);
2612         if (error)
2613                 goto out;
2614         error = 0;
2615 out:
2616         kfree(b_entry_name);
2617         kfree(buffer);
2618         if (is)
2619                 brelse(is->iloc.bh);
2620         if (bs)
2621                 brelse(bs->bh);
2622         kfree(is);
2623         kfree(bs);
2624
2625         return error;
2626 }
2627
2628 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2629                                        struct ext4_inode *raw_inode,
2630                                        int isize_diff, size_t ifree,
2631                                        size_t bfree, int *total_ino)
2632 {
2633         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2634         struct ext4_xattr_entry *small_entry;
2635         struct ext4_xattr_entry *entry;
2636         struct ext4_xattr_entry *last;
2637         unsigned int entry_size;        /* EA entry size */
2638         unsigned int total_size;        /* EA entry size + value size */
2639         unsigned int min_total_size;
2640         int error;
2641
2642         while (isize_diff > ifree) {
2643                 entry = NULL;
2644                 small_entry = NULL;
2645                 min_total_size = ~0U;
2646                 last = IFIRST(header);
2647                 /* Find the entry best suited to be pushed into EA block */
2648                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2649                         /* never move system.data out of the inode */
2650                         if ((last->e_name_len == 4) &&
2651                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2652                             !memcmp(last->e_name, "data", 4))
2653                                 continue;
2654                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2655                         if (!last->e_value_inum)
2656                                 total_size += EXT4_XATTR_SIZE(
2657                                                le32_to_cpu(last->e_value_size));
2658                         if (total_size <= bfree &&
2659                             total_size < min_total_size) {
2660                                 if (total_size + ifree < isize_diff) {
2661                                         small_entry = last;
2662                                 } else {
2663                                         entry = last;
2664                                         min_total_size = total_size;
2665                                 }
2666                         }
2667                 }
2668
2669                 if (entry == NULL) {
2670                         if (small_entry == NULL)
2671                                 return -ENOSPC;
2672                         entry = small_entry;
2673                 }
2674
2675                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2676                 total_size = entry_size;
2677                 if (!entry->e_value_inum)
2678                         total_size += EXT4_XATTR_SIZE(
2679                                               le32_to_cpu(entry->e_value_size));
2680                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2681                                                  entry);
2682                 if (error)
2683                         return error;
2684
2685                 *total_ino -= entry_size;
2686                 ifree += total_size;
2687                 bfree -= total_size;
2688         }
2689
2690         return 0;
2691 }
2692
2693 /*
2694  * Expand an inode by new_extra_isize bytes when EAs are present.
2695  * Returns 0 on success or negative error number on failure.
2696  */
2697 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2698                                struct ext4_inode *raw_inode, handle_t *handle)
2699 {
2700         struct ext4_xattr_ibody_header *header;
2701         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2702         static unsigned int mnt_count;
2703         size_t min_offs;
2704         size_t ifree, bfree;
2705         int total_ino;
2706         void *base, *end;
2707         int error = 0, tried_min_extra_isize = 0;
2708         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2709         int isize_diff; /* How much do we need to grow i_extra_isize */
2710
2711 retry:
2712         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2713         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2714                 return 0;
2715
2716         header = IHDR(inode, raw_inode);
2717
2718         /*
2719          * Check if enough free space is available in the inode to shift the
2720          * entries ahead by new_extra_isize.
2721          */
2722
2723         base = IFIRST(header);
2724         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2725         min_offs = end - base;
2726         total_ino = sizeof(struct ext4_xattr_ibody_header);
2727
2728         error = xattr_check_inode(inode, header, end);
2729         if (error)
2730                 goto cleanup;
2731
2732         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2733         if (ifree >= isize_diff)
2734                 goto shift;
2735
2736         /*
2737          * Enough free space isn't available in the inode, check if
2738          * EA block can hold new_extra_isize bytes.
2739          */
2740         if (EXT4_I(inode)->i_file_acl) {
2741                 struct buffer_head *bh;
2742
2743                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2744                 error = -EIO;
2745                 if (!bh)
2746                         goto cleanup;
2747                 error = ext4_xattr_check_block(inode, bh);
2748                 if (error) {
2749                         brelse(bh);
2750                         goto cleanup;
2751                 }
2752                 base = BHDR(bh);
2753                 end = bh->b_data + bh->b_size;
2754                 min_offs = end - base;
2755                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2756                                               NULL);
2757                 brelse(bh);
2758                 if (bfree + ifree < isize_diff) {
2759                         if (!tried_min_extra_isize && s_min_extra_isize) {
2760                                 tried_min_extra_isize++;
2761                                 new_extra_isize = s_min_extra_isize;
2762                                 goto retry;
2763                         }
2764                         error = -ENOSPC;
2765                         goto cleanup;
2766                 }
2767         } else {
2768                 bfree = inode->i_sb->s_blocksize;
2769         }
2770
2771         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2772                                             isize_diff, ifree, bfree,
2773                                             &total_ino);
2774         if (error) {
2775                 if (error == -ENOSPC && !tried_min_extra_isize &&
2776                     s_min_extra_isize) {
2777                         tried_min_extra_isize++;
2778                         new_extra_isize = s_min_extra_isize;
2779                         goto retry;
2780                 }
2781                 goto cleanup;
2782         }
2783 shift:
2784         /* Adjust the offsets and shift the remaining entries ahead */
2785         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2786                         - new_extra_isize, (void *)raw_inode +
2787                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2788                         (void *)header, total_ino);
2789         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2790
2791 cleanup:
2792         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2793                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2794                              inode->i_ino);
2795                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2796         }
2797         return error;
2798 }
2799
2800 #define EIA_INCR 16 /* must be 2^n */
2801 #define EIA_MASK (EIA_INCR - 1)
2802
2803 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2804  * If @ea_inode_array is new or full it will be grown and the old
2805  * contents copied over.
2806  */
2807 static int
2808 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2809                         struct inode *inode)
2810 {
2811         if (*ea_inode_array == NULL) {
2812                 /*
2813                  * Start with 15 inodes, so it fits into a power-of-two size.
2814                  * If *ea_inode_array is NULL, this is essentially offsetof()
2815                  */
2816                 (*ea_inode_array) =
2817                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2818                                          inodes[EIA_MASK]),
2819                                 GFP_NOFS);
2820                 if (*ea_inode_array == NULL)
2821                         return -ENOMEM;
2822                 (*ea_inode_array)->count = 0;
2823         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2824                 /* expand the array once all 15 + n * 16 slots are full */
2825                 struct ext4_xattr_inode_array *new_array = NULL;
2826                 int count = (*ea_inode_array)->count;
2827
2828                 /* if new_array is NULL, this is essentially offsetof() */
2829                 new_array = kmalloc(
2830                                 offsetof(struct ext4_xattr_inode_array,
2831                                          inodes[count + EIA_INCR]),
2832                                 GFP_NOFS);
2833                 if (new_array == NULL)
2834                         return -ENOMEM;
2835                 memcpy(new_array, *ea_inode_array,
2836                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2837                 kfree(*ea_inode_array);
2838                 *ea_inode_array = new_array;
2839         }
2840         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2841         return 0;
2842 }
2843
2844 /*
2845  * ext4_xattr_delete_inode()
2846  *
2847  * Free extended attribute resources associated with this inode. Traverse
2848  * all entries and decrement reference on any xattr inodes associated with this
2849  * inode. This is called immediately before an inode is freed. We have exclusive
2850  * access to the inode. If an orphan inode is deleted it will also release its
2851  * references on xattr block and xattr inodes.
2852  */
2853 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2854                             struct ext4_xattr_inode_array **ea_inode_array,
2855                             int extra_credits)
2856 {
2857         struct buffer_head *bh = NULL;
2858         struct ext4_xattr_ibody_header *header;
2859         struct ext4_iloc iloc = { .bh = NULL };
2860         struct ext4_xattr_entry *entry;
2861         struct inode *ea_inode;
2862         int error;
2863
2864         error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2865                                           NULL /* bh */,
2866                                           false /* dirty */,
2867                                           false /* block_csum */);
2868         if (error) {
2869                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2870                 goto cleanup;
2871         }
2872
2873         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2874             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2875
2876                 error = ext4_get_inode_loc(inode, &iloc);
2877                 if (error) {
2878                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2879                         goto cleanup;
2880                 }
2881
2882                 error = ext4_journal_get_write_access(handle, iloc.bh);
2883                 if (error) {
2884                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2885                                          error);
2886                         goto cleanup;
2887                 }
2888
2889                 header = IHDR(inode, ext4_raw_inode(&iloc));
2890                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2891                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2892                                                      IFIRST(header),
2893                                                      false /* block_csum */,
2894                                                      ea_inode_array,
2895                                                      extra_credits,
2896                                                      false /* skip_quota */);
2897         }
2898
2899         if (EXT4_I(inode)->i_file_acl) {
2900                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2901                 if (!bh) {
2902                         EXT4_ERROR_INODE(inode, "block %llu read error",
2903                                          EXT4_I(inode)->i_file_acl);
2904                         error = -EIO;
2905                         goto cleanup;
2906                 }
2907                 error = ext4_xattr_check_block(inode, bh);
2908                 if (error)
2909                         goto cleanup;
2910
2911                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2912                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2913                              entry = EXT4_XATTR_NEXT(entry)) {
2914                                 if (!entry->e_value_inum)
2915                                         continue;
2916                                 error = ext4_xattr_inode_iget(inode,
2917                                               le32_to_cpu(entry->e_value_inum),
2918                                               le32_to_cpu(entry->e_hash),
2919                                               &ea_inode);
2920                                 if (error)
2921                                         continue;
2922                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2923                                               le32_to_cpu(entry->e_value_size));
2924                                 iput(ea_inode);
2925                         }
2926
2927                 }
2928
2929                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2930                                          extra_credits);
2931                 /*
2932                  * Update i_file_acl value in the same transaction that releases
2933                  * block.
2934                  */
2935                 EXT4_I(inode)->i_file_acl = 0;
2936                 error = ext4_mark_inode_dirty(handle, inode);
2937                 if (error) {
2938                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2939                                          error);
2940                         goto cleanup;
2941                 }
2942         }
2943         error = 0;
2944 cleanup:
2945         brelse(iloc.bh);
2946         brelse(bh);
2947         return error;
2948 }
2949
2950 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2951 {
2952         int idx;
2953
2954         if (ea_inode_array == NULL)
2955                 return;
2956
2957         for (idx = 0; idx < ea_inode_array->count; ++idx)
2958                 iput(ea_inode_array->inodes[idx]);
2959         kfree(ea_inode_array);
2960 }
2961
2962 /*
2963  * ext4_xattr_block_cache_insert()
2964  *
2965  * Create a new entry in the extended attribute block cache, and insert
2966  * it unless such an entry is already in the cache.
2967  *
2968  * Returns 0, or a negative error number on failure.
2969  */
2970 static void
2971 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2972                               struct buffer_head *bh)
2973 {
2974         struct ext4_xattr_header *header = BHDR(bh);
2975         __u32 hash = le32_to_cpu(header->h_hash);
2976         int reusable = le32_to_cpu(header->h_refcount) <
2977                        EXT4_XATTR_REFCOUNT_MAX;
2978         int error;
2979
2980         if (!ea_block_cache)
2981                 return;
2982         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2983                                       bh->b_blocknr, reusable);
2984         if (error) {
2985                 if (error == -EBUSY)
2986                         ea_bdebug(bh, "already in cache");
2987         } else
2988                 ea_bdebug(bh, "inserting [%x]", (int)hash);
2989 }
2990
2991 /*
2992  * ext4_xattr_cmp()
2993  *
2994  * Compare two extended attribute blocks for equality.
2995  *
2996  * Returns 0 if the blocks are equal, 1 if they differ, and
2997  * a negative error number on errors.
2998  */
2999 static int
3000 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3001                struct ext4_xattr_header *header2)
3002 {
3003         struct ext4_xattr_entry *entry1, *entry2;
3004
3005         entry1 = ENTRY(header1+1);
3006         entry2 = ENTRY(header2+1);
3007         while (!IS_LAST_ENTRY(entry1)) {
3008                 if (IS_LAST_ENTRY(entry2))
3009                         return 1;
3010                 if (entry1->e_hash != entry2->e_hash ||
3011                     entry1->e_name_index != entry2->e_name_index ||
3012                     entry1->e_name_len != entry2->e_name_len ||
3013                     entry1->e_value_size != entry2->e_value_size ||
3014                     entry1->e_value_inum != entry2->e_value_inum ||
3015                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3016                         return 1;
3017                 if (!entry1->e_value_inum &&
3018                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3019                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3020                            le32_to_cpu(entry1->e_value_size)))
3021                         return 1;
3022
3023                 entry1 = EXT4_XATTR_NEXT(entry1);
3024                 entry2 = EXT4_XATTR_NEXT(entry2);
3025         }
3026         if (!IS_LAST_ENTRY(entry2))
3027                 return 1;
3028         return 0;
3029 }
3030
3031 /*
3032  * ext4_xattr_block_cache_find()
3033  *
3034  * Find an identical extended attribute block.
3035  *
3036  * Returns a pointer to the block found, or NULL if such a block was
3037  * not found or an error occurred.
3038  */
3039 static struct buffer_head *
3040 ext4_xattr_block_cache_find(struct inode *inode,
3041                             struct ext4_xattr_header *header,
3042                             struct mb_cache_entry **pce)
3043 {
3044         __u32 hash = le32_to_cpu(header->h_hash);
3045         struct mb_cache_entry *ce;
3046         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3047
3048         if (!ea_block_cache)
3049                 return NULL;
3050         if (!header->h_hash)
3051                 return NULL;  /* never share */
3052         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3053         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3054         while (ce) {
3055                 struct buffer_head *bh;
3056
3057                 bh = sb_bread(inode->i_sb, ce->e_value);
3058                 if (!bh) {
3059                         EXT4_ERROR_INODE(inode, "block %lu read error",
3060                                          (unsigned long)ce->e_value);
3061                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3062                         *pce = ce;
3063                         return bh;
3064                 }
3065                 brelse(bh);
3066                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3067         }
3068         return NULL;
3069 }
3070
3071 #define NAME_HASH_SHIFT 5
3072 #define VALUE_HASH_SHIFT 16
3073
3074 /*
3075  * ext4_xattr_hash_entry()
3076  *
3077  * Compute the hash of an extended attribute.
3078  */
3079 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3080                                     size_t value_count)
3081 {
3082         __u32 hash = 0;
3083
3084         while (name_len--) {
3085                 hash = (hash << NAME_HASH_SHIFT) ^
3086                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3087                        *name++;
3088         }
3089         while (value_count--) {
3090                 hash = (hash << VALUE_HASH_SHIFT) ^
3091                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3092                        le32_to_cpu(*value++);
3093         }
3094         return cpu_to_le32(hash);
3095 }
3096
3097 #undef NAME_HASH_SHIFT
3098 #undef VALUE_HASH_SHIFT
3099
3100 #define BLOCK_HASH_SHIFT 16
3101
3102 /*
3103  * ext4_xattr_rehash()
3104  *
3105  * Re-compute the extended attribute hash value after an entry has changed.
3106  */
3107 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3108 {
3109         struct ext4_xattr_entry *here;
3110         __u32 hash = 0;
3111
3112         here = ENTRY(header+1);
3113         while (!IS_LAST_ENTRY(here)) {
3114                 if (!here->e_hash) {
3115                         /* Block is not shared if an entry's hash value == 0 */
3116                         hash = 0;
3117                         break;
3118                 }
3119                 hash = (hash << BLOCK_HASH_SHIFT) ^
3120                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3121                        le32_to_cpu(here->e_hash);
3122                 here = EXT4_XATTR_NEXT(here);
3123         }
3124         header->h_hash = cpu_to_le32(hash);
3125 }
3126
3127 #undef BLOCK_HASH_SHIFT
3128
3129 #define HASH_BUCKET_BITS        10
3130
3131 struct mb_cache *
3132 ext4_xattr_create_cache(void)
3133 {
3134         return mb_cache_create(HASH_BUCKET_BITS);
3135 }
3136
3137 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3138 {
3139         if (cache)
3140                 mb_cache_destroy(cache);
3141 }
3142