Merge tag 'f2fs-for-4.19' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk...
[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                 iloc.bh = NULL;
1036                 goto out;
1037         }
1038
1039         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1040         ref_count += ref_change;
1041         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1042
1043         if (ref_change > 0) {
1044                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1045                           ea_inode->i_ino, ref_count);
1046
1047                 if (ref_count == 1) {
1048                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1049                                   ea_inode->i_ino, ea_inode->i_nlink);
1050
1051                         set_nlink(ea_inode, 1);
1052                         ext4_orphan_del(handle, ea_inode);
1053
1054                         if (ea_inode_cache) {
1055                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1056                                 mb_cache_entry_create(ea_inode_cache,
1057                                                       GFP_NOFS, hash,
1058                                                       ea_inode->i_ino,
1059                                                       true /* reusable */);
1060                         }
1061                 }
1062         } else {
1063                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1064                           ea_inode->i_ino, ref_count);
1065
1066                 if (ref_count == 0) {
1067                         WARN_ONCE(ea_inode->i_nlink != 1,
1068                                   "EA inode %lu i_nlink=%u",
1069                                   ea_inode->i_ino, ea_inode->i_nlink);
1070
1071                         clear_nlink(ea_inode);
1072                         ext4_orphan_add(handle, ea_inode);
1073
1074                         if (ea_inode_cache) {
1075                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1076                                 mb_cache_entry_delete(ea_inode_cache, hash,
1077                                                       ea_inode->i_ino);
1078                         }
1079                 }
1080         }
1081
1082         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1083         iloc.bh = NULL;
1084         if (ret)
1085                 ext4_warning_inode(ea_inode,
1086                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1087 out:
1088         brelse(iloc.bh);
1089         inode_unlock(ea_inode);
1090         return ret;
1091 }
1092
1093 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1094 {
1095         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1096 }
1097
1098 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1099 {
1100         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1101 }
1102
1103 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1104                                         struct ext4_xattr_entry *first)
1105 {
1106         struct inode *ea_inode;
1107         struct ext4_xattr_entry *entry;
1108         struct ext4_xattr_entry *failed_entry;
1109         unsigned int ea_ino;
1110         int err, saved_err;
1111
1112         for (entry = first; !IS_LAST_ENTRY(entry);
1113              entry = EXT4_XATTR_NEXT(entry)) {
1114                 if (!entry->e_value_inum)
1115                         continue;
1116                 ea_ino = le32_to_cpu(entry->e_value_inum);
1117                 err = ext4_xattr_inode_iget(parent, ea_ino,
1118                                             le32_to_cpu(entry->e_hash),
1119                                             &ea_inode);
1120                 if (err)
1121                         goto cleanup;
1122                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1123                 if (err) {
1124                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1125                         iput(ea_inode);
1126                         goto cleanup;
1127                 }
1128                 iput(ea_inode);
1129         }
1130         return 0;
1131
1132 cleanup:
1133         saved_err = err;
1134         failed_entry = entry;
1135
1136         for (entry = first; entry != failed_entry;
1137              entry = EXT4_XATTR_NEXT(entry)) {
1138                 if (!entry->e_value_inum)
1139                         continue;
1140                 ea_ino = le32_to_cpu(entry->e_value_inum);
1141                 err = ext4_xattr_inode_iget(parent, ea_ino,
1142                                             le32_to_cpu(entry->e_hash),
1143                                             &ea_inode);
1144                 if (err) {
1145                         ext4_warning(parent->i_sb,
1146                                      "cleanup ea_ino %u iget error %d", ea_ino,
1147                                      err);
1148                         continue;
1149                 }
1150                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1151                 if (err)
1152                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1153                                            err);
1154                 iput(ea_inode);
1155         }
1156         return saved_err;
1157 }
1158
1159 static void
1160 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1161                              struct buffer_head *bh,
1162                              struct ext4_xattr_entry *first, bool block_csum,
1163                              struct ext4_xattr_inode_array **ea_inode_array,
1164                              int extra_credits, bool skip_quota)
1165 {
1166         struct inode *ea_inode;
1167         struct ext4_xattr_entry *entry;
1168         bool dirty = false;
1169         unsigned int ea_ino;
1170         int err;
1171         int credits;
1172
1173         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1174         credits = 2 + extra_credits;
1175
1176         for (entry = first; !IS_LAST_ENTRY(entry);
1177              entry = EXT4_XATTR_NEXT(entry)) {
1178                 if (!entry->e_value_inum)
1179                         continue;
1180                 ea_ino = le32_to_cpu(entry->e_value_inum);
1181                 err = ext4_xattr_inode_iget(parent, ea_ino,
1182                                             le32_to_cpu(entry->e_hash),
1183                                             &ea_inode);
1184                 if (err)
1185                         continue;
1186
1187                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1188                 if (err) {
1189                         ext4_warning_inode(ea_inode,
1190                                            "Expand inode array err=%d", err);
1191                         iput(ea_inode);
1192                         continue;
1193                 }
1194
1195                 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1196                                                 dirty, block_csum);
1197                 if (err) {
1198                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1199                                            err);
1200                         continue;
1201                 }
1202
1203                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1204                 if (err) {
1205                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1206                                            err);
1207                         continue;
1208                 }
1209
1210                 if (!skip_quota)
1211                         ext4_xattr_inode_free_quota(parent, ea_inode,
1212                                               le32_to_cpu(entry->e_value_size));
1213
1214                 /*
1215                  * Forget about ea_inode within the same transaction that
1216                  * decrements the ref count. This avoids duplicate decrements in
1217                  * case the rest of the work spills over to subsequent
1218                  * transactions.
1219                  */
1220                 entry->e_value_inum = 0;
1221                 entry->e_value_size = 0;
1222
1223                 dirty = true;
1224         }
1225
1226         if (dirty) {
1227                 /*
1228                  * Note that we are deliberately skipping csum calculation for
1229                  * the final update because we do not expect any journal
1230                  * restarts until xattr block is freed.
1231                  */
1232
1233                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1234                 if (err)
1235                         ext4_warning_inode(parent,
1236                                            "handle dirty metadata err=%d", err);
1237         }
1238 }
1239
1240 /*
1241  * Release the xattr block BH: If the reference count is > 1, decrement it;
1242  * otherwise free the block.
1243  */
1244 static void
1245 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1246                          struct buffer_head *bh,
1247                          struct ext4_xattr_inode_array **ea_inode_array,
1248                          int extra_credits)
1249 {
1250         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1251         u32 hash, ref;
1252         int error = 0;
1253
1254         BUFFER_TRACE(bh, "get_write_access");
1255         error = ext4_journal_get_write_access(handle, bh);
1256         if (error)
1257                 goto out;
1258
1259         lock_buffer(bh);
1260         hash = le32_to_cpu(BHDR(bh)->h_hash);
1261         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1262         if (ref == 1) {
1263                 ea_bdebug(bh, "refcount now=0; freeing");
1264                 /*
1265                  * This must happen under buffer lock for
1266                  * ext4_xattr_block_set() to reliably detect freed block
1267                  */
1268                 if (ea_block_cache)
1269                         mb_cache_entry_delete(ea_block_cache, hash,
1270                                               bh->b_blocknr);
1271                 get_bh(bh);
1272                 unlock_buffer(bh);
1273
1274                 if (ext4_has_feature_ea_inode(inode->i_sb))
1275                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1276                                                      BFIRST(bh),
1277                                                      true /* block_csum */,
1278                                                      ea_inode_array,
1279                                                      extra_credits,
1280                                                      true /* skip_quota */);
1281                 ext4_free_blocks(handle, inode, bh, 0, 1,
1282                                  EXT4_FREE_BLOCKS_METADATA |
1283                                  EXT4_FREE_BLOCKS_FORGET);
1284         } else {
1285                 ref--;
1286                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1287                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1288                         struct mb_cache_entry *ce;
1289
1290                         if (ea_block_cache) {
1291                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1292                                                         bh->b_blocknr);
1293                                 if (ce) {
1294                                         ce->e_reusable = 1;
1295                                         mb_cache_entry_put(ea_block_cache, ce);
1296                                 }
1297                         }
1298                 }
1299
1300                 ext4_xattr_block_csum_set(inode, bh);
1301                 /*
1302                  * Beware of this ugliness: Releasing of xattr block references
1303                  * from different inodes can race and so we have to protect
1304                  * from a race where someone else frees the block (and releases
1305                  * its journal_head) before we are done dirtying the buffer. In
1306                  * nojournal mode this race is harmless and we actually cannot
1307                  * call ext4_handle_dirty_metadata() with locked buffer as
1308                  * that function can call sync_dirty_buffer() so for that case
1309                  * we handle the dirtying after unlocking the buffer.
1310                  */
1311                 if (ext4_handle_valid(handle))
1312                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1313                 unlock_buffer(bh);
1314                 if (!ext4_handle_valid(handle))
1315                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1316                 if (IS_SYNC(inode))
1317                         ext4_handle_sync(handle);
1318                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1319                 ea_bdebug(bh, "refcount now=%d; releasing",
1320                           le32_to_cpu(BHDR(bh)->h_refcount));
1321         }
1322 out:
1323         ext4_std_error(inode->i_sb, error);
1324         return;
1325 }
1326
1327 /*
1328  * Find the available free space for EAs. This also returns the total number of
1329  * bytes used by EA entries.
1330  */
1331 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1332                                     size_t *min_offs, void *base, int *total)
1333 {
1334         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1335                 if (!last->e_value_inum && last->e_value_size) {
1336                         size_t offs = le16_to_cpu(last->e_value_offs);
1337                         if (offs < *min_offs)
1338                                 *min_offs = offs;
1339                 }
1340                 if (total)
1341                         *total += EXT4_XATTR_LEN(last->e_name_len);
1342         }
1343         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1344 }
1345
1346 /*
1347  * Write the value of the EA in an inode.
1348  */
1349 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1350                                   const void *buf, int bufsize)
1351 {
1352         struct buffer_head *bh = NULL;
1353         unsigned long block = 0;
1354         int blocksize = ea_inode->i_sb->s_blocksize;
1355         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1356         int csize, wsize = 0;
1357         int ret = 0;
1358         int retries = 0;
1359
1360 retry:
1361         while (ret >= 0 && ret < max_blocks) {
1362                 struct ext4_map_blocks map;
1363                 map.m_lblk = block += ret;
1364                 map.m_len = max_blocks -= ret;
1365
1366                 ret = ext4_map_blocks(handle, ea_inode, &map,
1367                                       EXT4_GET_BLOCKS_CREATE);
1368                 if (ret <= 0) {
1369                         ext4_mark_inode_dirty(handle, ea_inode);
1370                         if (ret == -ENOSPC &&
1371                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1372                                 ret = 0;
1373                                 goto retry;
1374                         }
1375                         break;
1376                 }
1377         }
1378
1379         if (ret < 0)
1380                 return ret;
1381
1382         block = 0;
1383         while (wsize < bufsize) {
1384                 if (bh != NULL)
1385                         brelse(bh);
1386                 csize = (bufsize - wsize) > blocksize ? blocksize :
1387                                                                 bufsize - wsize;
1388                 bh = ext4_getblk(handle, ea_inode, block, 0);
1389                 if (IS_ERR(bh))
1390                         return PTR_ERR(bh);
1391                 ret = ext4_journal_get_write_access(handle, bh);
1392                 if (ret)
1393                         goto out;
1394
1395                 memcpy(bh->b_data, buf, csize);
1396                 set_buffer_uptodate(bh);
1397                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1398
1399                 buf += csize;
1400                 wsize += csize;
1401                 block += 1;
1402         }
1403
1404         inode_lock(ea_inode);
1405         i_size_write(ea_inode, wsize);
1406         ext4_update_i_disksize(ea_inode, wsize);
1407         inode_unlock(ea_inode);
1408
1409         ext4_mark_inode_dirty(handle, ea_inode);
1410
1411 out:
1412         brelse(bh);
1413
1414         return ret;
1415 }
1416
1417 /*
1418  * Create an inode to store the value of a large EA.
1419  */
1420 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1421                                              struct inode *inode, u32 hash)
1422 {
1423         struct inode *ea_inode = NULL;
1424         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1425         int err;
1426
1427         /*
1428          * Let the next inode be the goal, so we try and allocate the EA inode
1429          * in the same group, or nearby one.
1430          */
1431         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1432                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1433                                   EXT4_EA_INODE_FL);
1434         if (!IS_ERR(ea_inode)) {
1435                 ea_inode->i_op = &ext4_file_inode_operations;
1436                 ea_inode->i_fop = &ext4_file_operations;
1437                 ext4_set_aops(ea_inode);
1438                 ext4_xattr_inode_set_class(ea_inode);
1439                 unlock_new_inode(ea_inode);
1440                 ext4_xattr_inode_set_ref(ea_inode, 1);
1441                 ext4_xattr_inode_set_hash(ea_inode, hash);
1442                 err = ext4_mark_inode_dirty(handle, ea_inode);
1443                 if (!err)
1444                         err = ext4_inode_attach_jinode(ea_inode);
1445                 if (err) {
1446                         iput(ea_inode);
1447                         return ERR_PTR(err);
1448                 }
1449
1450                 /*
1451                  * Xattr inodes are shared therefore quota charging is performed
1452                  * at a higher level.
1453                  */
1454                 dquot_free_inode(ea_inode);
1455                 dquot_drop(ea_inode);
1456                 inode_lock(ea_inode);
1457                 ea_inode->i_flags |= S_NOQUOTA;
1458                 inode_unlock(ea_inode);
1459         }
1460
1461         return ea_inode;
1462 }
1463
1464 static struct inode *
1465 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1466                             size_t value_len, u32 hash)
1467 {
1468         struct inode *ea_inode;
1469         struct mb_cache_entry *ce;
1470         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1471         void *ea_data;
1472
1473         if (!ea_inode_cache)
1474                 return NULL;
1475
1476         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1477         if (!ce)
1478                 return NULL;
1479
1480         ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1481         if (!ea_data) {
1482                 mb_cache_entry_put(ea_inode_cache, ce);
1483                 return NULL;
1484         }
1485
1486         while (ce) {
1487                 ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1488                 if (!IS_ERR(ea_inode) &&
1489                     !is_bad_inode(ea_inode) &&
1490                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1491                     i_size_read(ea_inode) == value_len &&
1492                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1493                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1494                                                     value_len) &&
1495                     !memcmp(value, ea_data, value_len)) {
1496                         mb_cache_entry_touch(ea_inode_cache, ce);
1497                         mb_cache_entry_put(ea_inode_cache, ce);
1498                         kvfree(ea_data);
1499                         return ea_inode;
1500                 }
1501
1502                 if (!IS_ERR(ea_inode))
1503                         iput(ea_inode);
1504                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1505         }
1506         kvfree(ea_data);
1507         return NULL;
1508 }
1509
1510 /*
1511  * Add value of the EA in an inode.
1512  */
1513 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1514                                           const void *value, size_t value_len,
1515                                           struct inode **ret_inode)
1516 {
1517         struct inode *ea_inode;
1518         u32 hash;
1519         int err;
1520
1521         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1522         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1523         if (ea_inode) {
1524                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1525                 if (err) {
1526                         iput(ea_inode);
1527                         return err;
1528                 }
1529
1530                 *ret_inode = ea_inode;
1531                 return 0;
1532         }
1533
1534         /* Create an inode for the EA value */
1535         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1536         if (IS_ERR(ea_inode))
1537                 return PTR_ERR(ea_inode);
1538
1539         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1540         if (err) {
1541                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1542                 iput(ea_inode);
1543                 return err;
1544         }
1545
1546         if (EA_INODE_CACHE(inode))
1547                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1548                                       ea_inode->i_ino, true /* reusable */);
1549
1550         *ret_inode = ea_inode;
1551         return 0;
1552 }
1553
1554 /*
1555  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1556  * feature is enabled.
1557  */
1558 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1559
1560 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1561                                 struct ext4_xattr_search *s,
1562                                 handle_t *handle, struct inode *inode,
1563                                 bool is_block)
1564 {
1565         struct ext4_xattr_entry *last, *next;
1566         struct ext4_xattr_entry *here = s->here;
1567         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1568         int in_inode = i->in_inode;
1569         struct inode *old_ea_inode = NULL;
1570         struct inode *new_ea_inode = NULL;
1571         size_t old_size, new_size;
1572         int ret;
1573
1574         /* Space used by old and new values. */
1575         old_size = (!s->not_found && !here->e_value_inum) ?
1576                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1577         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1578
1579         /*
1580          * Optimization for the simple case when old and new values have the
1581          * same padded sizes. Not applicable if external inodes are involved.
1582          */
1583         if (new_size && new_size == old_size) {
1584                 size_t offs = le16_to_cpu(here->e_value_offs);
1585                 void *val = s->base + offs;
1586
1587                 here->e_value_size = cpu_to_le32(i->value_len);
1588                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1589                         memset(val, 0, new_size);
1590                 } else {
1591                         memcpy(val, i->value, i->value_len);
1592                         /* Clear padding bytes. */
1593                         memset(val + i->value_len, 0, new_size - i->value_len);
1594                 }
1595                 goto update_hash;
1596         }
1597
1598         /* Compute min_offs and last. */
1599         last = s->first;
1600         for (; !IS_LAST_ENTRY(last); last = next) {
1601                 next = EXT4_XATTR_NEXT(last);
1602                 if ((void *)next >= s->end) {
1603                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1604                         ret = -EFSCORRUPTED;
1605                         goto out;
1606                 }
1607                 if (!last->e_value_inum && last->e_value_size) {
1608                         size_t offs = le16_to_cpu(last->e_value_offs);
1609                         if (offs < min_offs)
1610                                 min_offs = offs;
1611                 }
1612         }
1613
1614         /* Check whether we have enough space. */
1615         if (i->value) {
1616                 size_t free;
1617
1618                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1619                 if (!s->not_found)
1620                         free += EXT4_XATTR_LEN(name_len) + old_size;
1621
1622                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1623                         ret = -ENOSPC;
1624                         goto out;
1625                 }
1626
1627                 /*
1628                  * If storing the value in an external inode is an option,
1629                  * reserve space for xattr entries/names in the external
1630                  * attribute block so that a long value does not occupy the
1631                  * whole space and prevent futher entries being added.
1632                  */
1633                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1634                     new_size && is_block &&
1635                     (min_offs + old_size - new_size) <
1636                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1637                         ret = -ENOSPC;
1638                         goto out;
1639                 }
1640         }
1641
1642         /*
1643          * Getting access to old and new ea inodes is subject to failures.
1644          * Finish that work before doing any modifications to the xattr data.
1645          */
1646         if (!s->not_found && here->e_value_inum) {
1647                 ret = ext4_xattr_inode_iget(inode,
1648                                             le32_to_cpu(here->e_value_inum),
1649                                             le32_to_cpu(here->e_hash),
1650                                             &old_ea_inode);
1651                 if (ret) {
1652                         old_ea_inode = NULL;
1653                         goto out;
1654                 }
1655         }
1656         if (i->value && in_inode) {
1657                 WARN_ON_ONCE(!i->value_len);
1658
1659                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1660                 if (ret)
1661                         goto out;
1662
1663                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1664                                                      i->value_len,
1665                                                      &new_ea_inode);
1666                 if (ret) {
1667                         new_ea_inode = NULL;
1668                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1669                         goto out;
1670                 }
1671         }
1672
1673         if (old_ea_inode) {
1674                 /* We are ready to release ref count on the old_ea_inode. */
1675                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1676                 if (ret) {
1677                         /* Release newly required ref count on new_ea_inode. */
1678                         if (new_ea_inode) {
1679                                 int err;
1680
1681                                 err = ext4_xattr_inode_dec_ref(handle,
1682                                                                new_ea_inode);
1683                                 if (err)
1684                                         ext4_warning_inode(new_ea_inode,
1685                                                   "dec ref new_ea_inode err=%d",
1686                                                   err);
1687                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1688                                                             i->value_len);
1689                         }
1690                         goto out;
1691                 }
1692
1693                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1694                                             le32_to_cpu(here->e_value_size));
1695         }
1696
1697         /* No failures allowed past this point. */
1698
1699         if (!s->not_found && here->e_value_size && here->e_value_offs) {
1700                 /* Remove the old value. */
1701                 void *first_val = s->base + min_offs;
1702                 size_t offs = le16_to_cpu(here->e_value_offs);
1703                 void *val = s->base + offs;
1704
1705                 memmove(first_val + old_size, first_val, val - first_val);
1706                 memset(first_val, 0, old_size);
1707                 min_offs += old_size;
1708
1709                 /* Adjust all value offsets. */
1710                 last = s->first;
1711                 while (!IS_LAST_ENTRY(last)) {
1712                         size_t o = le16_to_cpu(last->e_value_offs);
1713
1714                         if (!last->e_value_inum &&
1715                             last->e_value_size && o < offs)
1716                                 last->e_value_offs = cpu_to_le16(o + old_size);
1717                         last = EXT4_XATTR_NEXT(last);
1718                 }
1719         }
1720
1721         if (!i->value) {
1722                 /* Remove old name. */
1723                 size_t size = EXT4_XATTR_LEN(name_len);
1724
1725                 last = ENTRY((void *)last - size);
1726                 memmove(here, (void *)here + size,
1727                         (void *)last - (void *)here + sizeof(__u32));
1728                 memset(last, 0, size);
1729         } else if (s->not_found) {
1730                 /* Insert new name. */
1731                 size_t size = EXT4_XATTR_LEN(name_len);
1732                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1733
1734                 memmove((void *)here + size, here, rest);
1735                 memset(here, 0, size);
1736                 here->e_name_index = i->name_index;
1737                 here->e_name_len = name_len;
1738                 memcpy(here->e_name, i->name, name_len);
1739         } else {
1740                 /* This is an update, reset value info. */
1741                 here->e_value_inum = 0;
1742                 here->e_value_offs = 0;
1743                 here->e_value_size = 0;
1744         }
1745
1746         if (i->value) {
1747                 /* Insert new value. */
1748                 if (in_inode) {
1749                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1750                 } else if (i->value_len) {
1751                         void *val = s->base + min_offs - new_size;
1752
1753                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1754                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1755                                 memset(val, 0, new_size);
1756                         } else {
1757                                 memcpy(val, i->value, i->value_len);
1758                                 /* Clear padding bytes. */
1759                                 memset(val + i->value_len, 0,
1760                                        new_size - i->value_len);
1761                         }
1762                 }
1763                 here->e_value_size = cpu_to_le32(i->value_len);
1764         }
1765
1766 update_hash:
1767         if (i->value) {
1768                 __le32 hash = 0;
1769
1770                 /* Entry hash calculation. */
1771                 if (in_inode) {
1772                         __le32 crc32c_hash;
1773
1774                         /*
1775                          * Feed crc32c hash instead of the raw value for entry
1776                          * hash calculation. This is to avoid walking
1777                          * potentially long value buffer again.
1778                          */
1779                         crc32c_hash = cpu_to_le32(
1780                                        ext4_xattr_inode_get_hash(new_ea_inode));
1781                         hash = ext4_xattr_hash_entry(here->e_name,
1782                                                      here->e_name_len,
1783                                                      &crc32c_hash, 1);
1784                 } else if (is_block) {
1785                         __le32 *value = s->base + le16_to_cpu(
1786                                                         here->e_value_offs);
1787
1788                         hash = ext4_xattr_hash_entry(here->e_name,
1789                                                      here->e_name_len, value,
1790                                                      new_size >> 2);
1791                 }
1792                 here->e_hash = hash;
1793         }
1794
1795         if (is_block)
1796                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1797
1798         ret = 0;
1799 out:
1800         iput(old_ea_inode);
1801         iput(new_ea_inode);
1802         return ret;
1803 }
1804
1805 struct ext4_xattr_block_find {
1806         struct ext4_xattr_search s;
1807         struct buffer_head *bh;
1808 };
1809
1810 static int
1811 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1812                       struct ext4_xattr_block_find *bs)
1813 {
1814         struct super_block *sb = inode->i_sb;
1815         int error;
1816
1817         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1818                   i->name_index, i->name, i->value, (long)i->value_len);
1819
1820         if (EXT4_I(inode)->i_file_acl) {
1821                 /* The inode already has an extended attribute block. */
1822                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1823                 error = -EIO;
1824                 if (!bs->bh)
1825                         goto cleanup;
1826                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1827                         atomic_read(&(bs->bh->b_count)),
1828                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1829                 error = ext4_xattr_check_block(inode, bs->bh);
1830                 if (error)
1831                         goto cleanup;
1832                 /* Find the named attribute. */
1833                 bs->s.base = BHDR(bs->bh);
1834                 bs->s.first = BFIRST(bs->bh);
1835                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1836                 bs->s.here = bs->s.first;
1837                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1838                                          i->name_index, i->name, 1);
1839                 if (error && error != -ENODATA)
1840                         goto cleanup;
1841                 bs->s.not_found = error;
1842         }
1843         error = 0;
1844
1845 cleanup:
1846         return error;
1847 }
1848
1849 static int
1850 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1851                      struct ext4_xattr_info *i,
1852                      struct ext4_xattr_block_find *bs)
1853 {
1854         struct super_block *sb = inode->i_sb;
1855         struct buffer_head *new_bh = NULL;
1856         struct ext4_xattr_search s_copy = bs->s;
1857         struct ext4_xattr_search *s = &s_copy;
1858         struct mb_cache_entry *ce = NULL;
1859         int error = 0;
1860         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1861         struct inode *ea_inode = NULL, *tmp_inode;
1862         size_t old_ea_inode_quota = 0;
1863         unsigned int ea_ino;
1864
1865
1866 #define header(x) ((struct ext4_xattr_header *)(x))
1867
1868         if (s->base) {
1869                 BUFFER_TRACE(bs->bh, "get_write_access");
1870                 error = ext4_journal_get_write_access(handle, bs->bh);
1871                 if (error)
1872                         goto cleanup;
1873                 lock_buffer(bs->bh);
1874
1875                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1876                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1877
1878                         /*
1879                          * This must happen under buffer lock for
1880                          * ext4_xattr_block_set() to reliably detect modified
1881                          * block
1882                          */
1883                         if (ea_block_cache)
1884                                 mb_cache_entry_delete(ea_block_cache, hash,
1885                                                       bs->bh->b_blocknr);
1886                         ea_bdebug(bs->bh, "modifying in-place");
1887                         error = ext4_xattr_set_entry(i, s, handle, inode,
1888                                                      true /* is_block */);
1889                         ext4_xattr_block_csum_set(inode, bs->bh);
1890                         unlock_buffer(bs->bh);
1891                         if (error == -EFSCORRUPTED)
1892                                 goto bad_block;
1893                         if (!error)
1894                                 error = ext4_handle_dirty_metadata(handle,
1895                                                                    inode,
1896                                                                    bs->bh);
1897                         if (error)
1898                                 goto cleanup;
1899                         goto inserted;
1900                 } else {
1901                         int offset = (char *)s->here - bs->bh->b_data;
1902
1903                         unlock_buffer(bs->bh);
1904                         ea_bdebug(bs->bh, "cloning");
1905                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1906                         error = -ENOMEM;
1907                         if (s->base == NULL)
1908                                 goto cleanup;
1909                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1910                         s->first = ENTRY(header(s->base)+1);
1911                         header(s->base)->h_refcount = cpu_to_le32(1);
1912                         s->here = ENTRY(s->base + offset);
1913                         s->end = s->base + bs->bh->b_size;
1914
1915                         /*
1916                          * If existing entry points to an xattr inode, we need
1917                          * to prevent ext4_xattr_set_entry() from decrementing
1918                          * ref count on it because the reference belongs to the
1919                          * original block. In this case, make the entry look
1920                          * like it has an empty value.
1921                          */
1922                         if (!s->not_found && s->here->e_value_inum) {
1923                                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1924                                 error = ext4_xattr_inode_iget(inode, ea_ino,
1925                                               le32_to_cpu(s->here->e_hash),
1926                                               &tmp_inode);
1927                                 if (error)
1928                                         goto cleanup;
1929
1930                                 if (!ext4_test_inode_state(tmp_inode,
1931                                                 EXT4_STATE_LUSTRE_EA_INODE)) {
1932                                         /*
1933                                          * Defer quota free call for previous
1934                                          * inode until success is guaranteed.
1935                                          */
1936                                         old_ea_inode_quota = le32_to_cpu(
1937                                                         s->here->e_value_size);
1938                                 }
1939                                 iput(tmp_inode);
1940
1941                                 s->here->e_value_inum = 0;
1942                                 s->here->e_value_size = 0;
1943                         }
1944                 }
1945         } else {
1946                 /* Allocate a buffer where we construct the new block. */
1947                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1948                 /* assert(header == s->base) */
1949                 error = -ENOMEM;
1950                 if (s->base == NULL)
1951                         goto cleanup;
1952                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1953                 header(s->base)->h_blocks = cpu_to_le32(1);
1954                 header(s->base)->h_refcount = cpu_to_le32(1);
1955                 s->first = ENTRY(header(s->base)+1);
1956                 s->here = ENTRY(header(s->base)+1);
1957                 s->end = s->base + sb->s_blocksize;
1958         }
1959
1960         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1961         if (error == -EFSCORRUPTED)
1962                 goto bad_block;
1963         if (error)
1964                 goto cleanup;
1965
1966         if (i->value && s->here->e_value_inum) {
1967                 /*
1968                  * A ref count on ea_inode has been taken as part of the call to
1969                  * ext4_xattr_set_entry() above. We would like to drop this
1970                  * extra ref but we have to wait until the xattr block is
1971                  * initialized and has its own ref count on the ea_inode.
1972                  */
1973                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1974                 error = ext4_xattr_inode_iget(inode, ea_ino,
1975                                               le32_to_cpu(s->here->e_hash),
1976                                               &ea_inode);
1977                 if (error) {
1978                         ea_inode = NULL;
1979                         goto cleanup;
1980                 }
1981         }
1982
1983 inserted:
1984         if (!IS_LAST_ENTRY(s->first)) {
1985                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1986                                                      &ce);
1987                 if (new_bh) {
1988                         /* We found an identical block in the cache. */
1989                         if (new_bh == bs->bh)
1990                                 ea_bdebug(new_bh, "keeping");
1991                         else {
1992                                 u32 ref;
1993
1994                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
1995
1996                                 /* The old block is released after updating
1997                                    the inode. */
1998                                 error = dquot_alloc_block(inode,
1999                                                 EXT4_C2B(EXT4_SB(sb), 1));
2000                                 if (error)
2001                                         goto cleanup;
2002                                 BUFFER_TRACE(new_bh, "get_write_access");
2003                                 error = ext4_journal_get_write_access(handle,
2004                                                                       new_bh);
2005                                 if (error)
2006                                         goto cleanup_dquot;
2007                                 lock_buffer(new_bh);
2008                                 /*
2009                                  * We have to be careful about races with
2010                                  * freeing, rehashing or adding references to
2011                                  * xattr block. Once we hold buffer lock xattr
2012                                  * block's state is stable so we can check
2013                                  * whether the block got freed / rehashed or
2014                                  * not.  Since we unhash mbcache entry under
2015                                  * buffer lock when freeing / rehashing xattr
2016                                  * block, checking whether entry is still
2017                                  * hashed is reliable. Same rules hold for
2018                                  * e_reusable handling.
2019                                  */
2020                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2021                                     !ce->e_reusable) {
2022                                         /*
2023                                          * Undo everything and check mbcache
2024                                          * again.
2025                                          */
2026                                         unlock_buffer(new_bh);
2027                                         dquot_free_block(inode,
2028                                                          EXT4_C2B(EXT4_SB(sb),
2029                                                                   1));
2030                                         brelse(new_bh);
2031                                         mb_cache_entry_put(ea_block_cache, ce);
2032                                         ce = NULL;
2033                                         new_bh = NULL;
2034                                         goto inserted;
2035                                 }
2036                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2037                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2038                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2039                                         ce->e_reusable = 0;
2040                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2041                                           ref);
2042                                 ext4_xattr_block_csum_set(inode, new_bh);
2043                                 unlock_buffer(new_bh);
2044                                 error = ext4_handle_dirty_metadata(handle,
2045                                                                    inode,
2046                                                                    new_bh);
2047                                 if (error)
2048                                         goto cleanup_dquot;
2049                         }
2050                         mb_cache_entry_touch(ea_block_cache, ce);
2051                         mb_cache_entry_put(ea_block_cache, ce);
2052                         ce = NULL;
2053                 } else if (bs->bh && s->base == bs->bh->b_data) {
2054                         /* We were modifying this block in-place. */
2055                         ea_bdebug(bs->bh, "keeping this block");
2056                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2057                         new_bh = bs->bh;
2058                         get_bh(new_bh);
2059                 } else {
2060                         /* We need to allocate a new block */
2061                         ext4_fsblk_t goal, block;
2062
2063                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2064
2065                         goal = ext4_group_first_block_no(sb,
2066                                                 EXT4_I(inode)->i_block_group);
2067
2068                         /* non-extent files can't have physical blocks past 2^32 */
2069                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2070                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2071
2072                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2073                                                      NULL, &error);
2074                         if (error)
2075                                 goto cleanup;
2076
2077                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2078                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2079
2080                         ea_idebug(inode, "creating block %llu",
2081                                   (unsigned long long)block);
2082
2083                         new_bh = sb_getblk(sb, block);
2084                         if (unlikely(!new_bh)) {
2085                                 error = -ENOMEM;
2086 getblk_failed:
2087                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2088                                                  EXT4_FREE_BLOCKS_METADATA);
2089                                 goto cleanup;
2090                         }
2091                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2092                                                       ENTRY(header(s->base)+1));
2093                         if (error)
2094                                 goto getblk_failed;
2095                         if (ea_inode) {
2096                                 /* Drop the extra ref on ea_inode. */
2097                                 error = ext4_xattr_inode_dec_ref(handle,
2098                                                                  ea_inode);
2099                                 if (error)
2100                                         ext4_warning_inode(ea_inode,
2101                                                            "dec ref error=%d",
2102                                                            error);
2103                                 iput(ea_inode);
2104                                 ea_inode = NULL;
2105                         }
2106
2107                         lock_buffer(new_bh);
2108                         error = ext4_journal_get_create_access(handle, new_bh);
2109                         if (error) {
2110                                 unlock_buffer(new_bh);
2111                                 error = -EIO;
2112                                 goto getblk_failed;
2113                         }
2114                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2115                         ext4_xattr_block_csum_set(inode, new_bh);
2116                         set_buffer_uptodate(new_bh);
2117                         unlock_buffer(new_bh);
2118                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2119                         error = ext4_handle_dirty_metadata(handle, inode,
2120                                                            new_bh);
2121                         if (error)
2122                                 goto cleanup;
2123                 }
2124         }
2125
2126         if (old_ea_inode_quota)
2127                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2128
2129         /* Update the inode. */
2130         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2131
2132         /* Drop the previous xattr block. */
2133         if (bs->bh && bs->bh != new_bh) {
2134                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2135
2136                 ext4_xattr_release_block(handle, inode, bs->bh,
2137                                          &ea_inode_array,
2138                                          0 /* extra_credits */);
2139                 ext4_xattr_inode_array_free(ea_inode_array);
2140         }
2141         error = 0;
2142
2143 cleanup:
2144         if (ea_inode) {
2145                 int error2;
2146
2147                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2148                 if (error2)
2149                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2150                                            error2);
2151
2152                 /* If there was an error, revert the quota charge. */
2153                 if (error)
2154                         ext4_xattr_inode_free_quota(inode, ea_inode,
2155                                                     i_size_read(ea_inode));
2156                 iput(ea_inode);
2157         }
2158         if (ce)
2159                 mb_cache_entry_put(ea_block_cache, ce);
2160         brelse(new_bh);
2161         if (!(bs->bh && s->base == bs->bh->b_data))
2162                 kfree(s->base);
2163
2164         return error;
2165
2166 cleanup_dquot:
2167         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2168         goto cleanup;
2169
2170 bad_block:
2171         EXT4_ERROR_INODE(inode, "bad block %llu",
2172                          EXT4_I(inode)->i_file_acl);
2173         goto cleanup;
2174
2175 #undef header
2176 }
2177
2178 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2179                           struct ext4_xattr_ibody_find *is)
2180 {
2181         struct ext4_xattr_ibody_header *header;
2182         struct ext4_inode *raw_inode;
2183         int error;
2184
2185         if (EXT4_I(inode)->i_extra_isize == 0)
2186                 return 0;
2187         raw_inode = ext4_raw_inode(&is->iloc);
2188         header = IHDR(inode, raw_inode);
2189         is->s.base = is->s.first = IFIRST(header);
2190         is->s.here = is->s.first;
2191         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2192         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2193                 error = xattr_check_inode(inode, header, is->s.end);
2194                 if (error)
2195                         return error;
2196                 /* Find the named attribute. */
2197                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2198                                          i->name_index, i->name, 0);
2199                 if (error && error != -ENODATA)
2200                         return error;
2201                 is->s.not_found = error;
2202         }
2203         return 0;
2204 }
2205
2206 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2207                                 struct ext4_xattr_info *i,
2208                                 struct ext4_xattr_ibody_find *is)
2209 {
2210         struct ext4_xattr_ibody_header *header;
2211         struct ext4_xattr_search *s = &is->s;
2212         int error;
2213
2214         if (EXT4_I(inode)->i_extra_isize == 0)
2215                 return -ENOSPC;
2216         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2217         if (error)
2218                 return error;
2219         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2220         if (!IS_LAST_ENTRY(s->first)) {
2221                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2222                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2223         } else {
2224                 header->h_magic = cpu_to_le32(0);
2225                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2226         }
2227         return 0;
2228 }
2229
2230 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2231                                 struct ext4_xattr_info *i,
2232                                 struct ext4_xattr_ibody_find *is)
2233 {
2234         struct ext4_xattr_ibody_header *header;
2235         struct ext4_xattr_search *s = &is->s;
2236         int error;
2237
2238         if (EXT4_I(inode)->i_extra_isize == 0)
2239                 return -ENOSPC;
2240         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2241         if (error)
2242                 return error;
2243         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2244         if (!IS_LAST_ENTRY(s->first)) {
2245                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2246                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2247         } else {
2248                 header->h_magic = cpu_to_le32(0);
2249                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2250         }
2251         return 0;
2252 }
2253
2254 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2255                                  struct ext4_xattr_info *i)
2256 {
2257         void *value;
2258
2259         /* When e_value_inum is set the value is stored externally. */
2260         if (s->here->e_value_inum)
2261                 return 0;
2262         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2263                 return 0;
2264         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2265         return !memcmp(value, i->value, i->value_len);
2266 }
2267
2268 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2269 {
2270         struct buffer_head *bh;
2271         int error;
2272
2273         if (!EXT4_I(inode)->i_file_acl)
2274                 return NULL;
2275         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2276         if (!bh)
2277                 return ERR_PTR(-EIO);
2278         error = ext4_xattr_check_block(inode, bh);
2279         if (error)
2280                 return ERR_PTR(error);
2281         return bh;
2282 }
2283
2284 /*
2285  * ext4_xattr_set_handle()
2286  *
2287  * Create, replace or remove an extended attribute for this inode.  Value
2288  * is NULL to remove an existing extended attribute, and non-NULL to
2289  * either replace an existing extended attribute, or create a new extended
2290  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2291  * specify that an extended attribute must exist and must not exist
2292  * previous to the call, respectively.
2293  *
2294  * Returns 0, or a negative error number on failure.
2295  */
2296 int
2297 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2298                       const char *name, const void *value, size_t value_len,
2299                       int flags)
2300 {
2301         struct ext4_xattr_info i = {
2302                 .name_index = name_index,
2303                 .name = name,
2304                 .value = value,
2305                 .value_len = value_len,
2306                 .in_inode = 0,
2307         };
2308         struct ext4_xattr_ibody_find is = {
2309                 .s = { .not_found = -ENODATA, },
2310         };
2311         struct ext4_xattr_block_find bs = {
2312                 .s = { .not_found = -ENODATA, },
2313         };
2314         int no_expand;
2315         int error;
2316
2317         if (!name)
2318                 return -EINVAL;
2319         if (strlen(name) > 255)
2320                 return -ERANGE;
2321
2322         ext4_write_lock_xattr(inode, &no_expand);
2323
2324         /* Check journal credits under write lock. */
2325         if (ext4_handle_valid(handle)) {
2326                 struct buffer_head *bh;
2327                 int credits;
2328
2329                 bh = ext4_xattr_get_block(inode);
2330                 if (IS_ERR(bh)) {
2331                         error = PTR_ERR(bh);
2332                         goto cleanup;
2333                 }
2334
2335                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2336                                                    value_len,
2337                                                    flags & XATTR_CREATE);
2338                 brelse(bh);
2339
2340                 if (!ext4_handle_has_enough_credits(handle, credits)) {
2341                         error = -ENOSPC;
2342                         goto cleanup;
2343                 }
2344         }
2345
2346         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2347         if (error)
2348                 goto cleanup;
2349
2350         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2351                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2352                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2353                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2354         }
2355
2356         error = ext4_xattr_ibody_find(inode, &i, &is);
2357         if (error)
2358                 goto cleanup;
2359         if (is.s.not_found)
2360                 error = ext4_xattr_block_find(inode, &i, &bs);
2361         if (error)
2362                 goto cleanup;
2363         if (is.s.not_found && bs.s.not_found) {
2364                 error = -ENODATA;
2365                 if (flags & XATTR_REPLACE)
2366                         goto cleanup;
2367                 error = 0;
2368                 if (!value)
2369                         goto cleanup;
2370         } else {
2371                 error = -EEXIST;
2372                 if (flags & XATTR_CREATE)
2373                         goto cleanup;
2374         }
2375
2376         if (!value) {
2377                 if (!is.s.not_found)
2378                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2379                 else if (!bs.s.not_found)
2380                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2381         } else {
2382                 error = 0;
2383                 /* Xattr value did not change? Save us some work and bail out */
2384                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2385                         goto cleanup;
2386                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2387                         goto cleanup;
2388
2389                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2390                     (EXT4_XATTR_SIZE(i.value_len) >
2391                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2392                         i.in_inode = 1;
2393 retry_inode:
2394                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2395                 if (!error && !bs.s.not_found) {
2396                         i.value = NULL;
2397                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2398                 } else if (error == -ENOSPC) {
2399                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
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         kfree(is);
2621         kfree(bs);
2622
2623         return error;
2624 }
2625
2626 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2627                                        struct ext4_inode *raw_inode,
2628                                        int isize_diff, size_t ifree,
2629                                        size_t bfree, int *total_ino)
2630 {
2631         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2632         struct ext4_xattr_entry *small_entry;
2633         struct ext4_xattr_entry *entry;
2634         struct ext4_xattr_entry *last;
2635         unsigned int entry_size;        /* EA entry size */
2636         unsigned int total_size;        /* EA entry size + value size */
2637         unsigned int min_total_size;
2638         int error;
2639
2640         while (isize_diff > ifree) {
2641                 entry = NULL;
2642                 small_entry = NULL;
2643                 min_total_size = ~0U;
2644                 last = IFIRST(header);
2645                 /* Find the entry best suited to be pushed into EA block */
2646                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2647                         /* never move system.data out of the inode */
2648                         if ((last->e_name_len == 4) &&
2649                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2650                             !memcmp(last->e_name, "data", 4))
2651                                 continue;
2652                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2653                         if (!last->e_value_inum)
2654                                 total_size += EXT4_XATTR_SIZE(
2655                                                le32_to_cpu(last->e_value_size));
2656                         if (total_size <= bfree &&
2657                             total_size < min_total_size) {
2658                                 if (total_size + ifree < isize_diff) {
2659                                         small_entry = last;
2660                                 } else {
2661                                         entry = last;
2662                                         min_total_size = total_size;
2663                                 }
2664                         }
2665                 }
2666
2667                 if (entry == NULL) {
2668                         if (small_entry == NULL)
2669                                 return -ENOSPC;
2670                         entry = small_entry;
2671                 }
2672
2673                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2674                 total_size = entry_size;
2675                 if (!entry->e_value_inum)
2676                         total_size += EXT4_XATTR_SIZE(
2677                                               le32_to_cpu(entry->e_value_size));
2678                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2679                                                  entry);
2680                 if (error)
2681                         return error;
2682
2683                 *total_ino -= entry_size;
2684                 ifree += total_size;
2685                 bfree -= total_size;
2686         }
2687
2688         return 0;
2689 }
2690
2691 /*
2692  * Expand an inode by new_extra_isize bytes when EAs are present.
2693  * Returns 0 on success or negative error number on failure.
2694  */
2695 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2696                                struct ext4_inode *raw_inode, handle_t *handle)
2697 {
2698         struct ext4_xattr_ibody_header *header;
2699         struct buffer_head *bh;
2700         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2701         static unsigned int mnt_count;
2702         size_t min_offs;
2703         size_t ifree, bfree;
2704         int total_ino;
2705         void *base, *end;
2706         int error = 0, tried_min_extra_isize = 0;
2707         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2708         int isize_diff; /* How much do we need to grow i_extra_isize */
2709
2710 retry:
2711         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2712         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2713                 return 0;
2714
2715         header = IHDR(inode, raw_inode);
2716
2717         /*
2718          * Check if enough free space is available in the inode to shift the
2719          * entries ahead by new_extra_isize.
2720          */
2721
2722         base = IFIRST(header);
2723         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2724         min_offs = end - base;
2725         total_ino = sizeof(struct ext4_xattr_ibody_header);
2726
2727         error = xattr_check_inode(inode, header, end);
2728         if (error)
2729                 goto cleanup;
2730
2731         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2732         if (ifree >= isize_diff)
2733                 goto shift;
2734
2735         /*
2736          * Enough free space isn't available in the inode, check if
2737          * EA block can hold new_extra_isize bytes.
2738          */
2739         if (EXT4_I(inode)->i_file_acl) {
2740                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2741                 error = -EIO;
2742                 if (!bh)
2743                         goto cleanup;
2744                 error = ext4_xattr_check_block(inode, bh);
2745                 if (error)
2746                         goto cleanup;
2747                 base = BHDR(bh);
2748                 end = bh->b_data + bh->b_size;
2749                 min_offs = end - base;
2750                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2751                                               NULL);
2752                 brelse(bh);
2753                 if (bfree + ifree < isize_diff) {
2754                         if (!tried_min_extra_isize && s_min_extra_isize) {
2755                                 tried_min_extra_isize++;
2756                                 new_extra_isize = s_min_extra_isize;
2757                                 goto retry;
2758                         }
2759                         error = -ENOSPC;
2760                         goto cleanup;
2761                 }
2762         } else {
2763                 bfree = inode->i_sb->s_blocksize;
2764         }
2765
2766         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2767                                             isize_diff, ifree, bfree,
2768                                             &total_ino);
2769         if (error) {
2770                 if (error == -ENOSPC && !tried_min_extra_isize &&
2771                     s_min_extra_isize) {
2772                         tried_min_extra_isize++;
2773                         new_extra_isize = s_min_extra_isize;
2774                         goto retry;
2775                 }
2776                 goto cleanup;
2777         }
2778 shift:
2779         /* Adjust the offsets and shift the remaining entries ahead */
2780         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2781                         - new_extra_isize, (void *)raw_inode +
2782                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2783                         (void *)header, total_ino);
2784         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2785
2786 cleanup:
2787         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2788                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2789                              inode->i_ino);
2790                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2791         }
2792         return error;
2793 }
2794
2795 #define EIA_INCR 16 /* must be 2^n */
2796 #define EIA_MASK (EIA_INCR - 1)
2797
2798 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2799  * If @ea_inode_array is new or full it will be grown and the old
2800  * contents copied over.
2801  */
2802 static int
2803 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2804                         struct inode *inode)
2805 {
2806         if (*ea_inode_array == NULL) {
2807                 /*
2808                  * Start with 15 inodes, so it fits into a power-of-two size.
2809                  * If *ea_inode_array is NULL, this is essentially offsetof()
2810                  */
2811                 (*ea_inode_array) =
2812                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2813                                          inodes[EIA_MASK]),
2814                                 GFP_NOFS);
2815                 if (*ea_inode_array == NULL)
2816                         return -ENOMEM;
2817                 (*ea_inode_array)->count = 0;
2818         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2819                 /* expand the array once all 15 + n * 16 slots are full */
2820                 struct ext4_xattr_inode_array *new_array = NULL;
2821                 int count = (*ea_inode_array)->count;
2822
2823                 /* if new_array is NULL, this is essentially offsetof() */
2824                 new_array = kmalloc(
2825                                 offsetof(struct ext4_xattr_inode_array,
2826                                          inodes[count + EIA_INCR]),
2827                                 GFP_NOFS);
2828                 if (new_array == NULL)
2829                         return -ENOMEM;
2830                 memcpy(new_array, *ea_inode_array,
2831                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2832                 kfree(*ea_inode_array);
2833                 *ea_inode_array = new_array;
2834         }
2835         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2836         return 0;
2837 }
2838
2839 /*
2840  * ext4_xattr_delete_inode()
2841  *
2842  * Free extended attribute resources associated with this inode. Traverse
2843  * all entries and decrement reference on any xattr inodes associated with this
2844  * inode. This is called immediately before an inode is freed. We have exclusive
2845  * access to the inode. If an orphan inode is deleted it will also release its
2846  * references on xattr block and xattr inodes.
2847  */
2848 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2849                             struct ext4_xattr_inode_array **ea_inode_array,
2850                             int extra_credits)
2851 {
2852         struct buffer_head *bh = NULL;
2853         struct ext4_xattr_ibody_header *header;
2854         struct ext4_iloc iloc = { .bh = NULL };
2855         struct ext4_xattr_entry *entry;
2856         struct inode *ea_inode;
2857         int error;
2858
2859         error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2860                                           NULL /* bh */,
2861                                           false /* dirty */,
2862                                           false /* block_csum */);
2863         if (error) {
2864                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2865                 goto cleanup;
2866         }
2867
2868         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2869             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2870
2871                 error = ext4_get_inode_loc(inode, &iloc);
2872                 if (error) {
2873                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2874                         goto cleanup;
2875                 }
2876
2877                 error = ext4_journal_get_write_access(handle, iloc.bh);
2878                 if (error) {
2879                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2880                                          error);
2881                         goto cleanup;
2882                 }
2883
2884                 header = IHDR(inode, ext4_raw_inode(&iloc));
2885                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2886                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2887                                                      IFIRST(header),
2888                                                      false /* block_csum */,
2889                                                      ea_inode_array,
2890                                                      extra_credits,
2891                                                      false /* skip_quota */);
2892         }
2893
2894         if (EXT4_I(inode)->i_file_acl) {
2895                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2896                 if (!bh) {
2897                         EXT4_ERROR_INODE(inode, "block %llu read error",
2898                                          EXT4_I(inode)->i_file_acl);
2899                         error = -EIO;
2900                         goto cleanup;
2901                 }
2902                 error = ext4_xattr_check_block(inode, bh);
2903                 if (error)
2904                         goto cleanup;
2905
2906                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2907                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2908                              entry = EXT4_XATTR_NEXT(entry)) {
2909                                 if (!entry->e_value_inum)
2910                                         continue;
2911                                 error = ext4_xattr_inode_iget(inode,
2912                                               le32_to_cpu(entry->e_value_inum),
2913                                               le32_to_cpu(entry->e_hash),
2914                                               &ea_inode);
2915                                 if (error)
2916                                         continue;
2917                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2918                                               le32_to_cpu(entry->e_value_size));
2919                                 iput(ea_inode);
2920                         }
2921
2922                 }
2923
2924                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2925                                          extra_credits);
2926                 /*
2927                  * Update i_file_acl value in the same transaction that releases
2928                  * block.
2929                  */
2930                 EXT4_I(inode)->i_file_acl = 0;
2931                 error = ext4_mark_inode_dirty(handle, inode);
2932                 if (error) {
2933                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2934                                          error);
2935                         goto cleanup;
2936                 }
2937         }
2938         error = 0;
2939 cleanup:
2940         brelse(iloc.bh);
2941         brelse(bh);
2942         return error;
2943 }
2944
2945 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2946 {
2947         int idx;
2948
2949         if (ea_inode_array == NULL)
2950                 return;
2951
2952         for (idx = 0; idx < ea_inode_array->count; ++idx)
2953                 iput(ea_inode_array->inodes[idx]);
2954         kfree(ea_inode_array);
2955 }
2956
2957 /*
2958  * ext4_xattr_block_cache_insert()
2959  *
2960  * Create a new entry in the extended attribute block cache, and insert
2961  * it unless such an entry is already in the cache.
2962  *
2963  * Returns 0, or a negative error number on failure.
2964  */
2965 static void
2966 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2967                               struct buffer_head *bh)
2968 {
2969         struct ext4_xattr_header *header = BHDR(bh);
2970         __u32 hash = le32_to_cpu(header->h_hash);
2971         int reusable = le32_to_cpu(header->h_refcount) <
2972                        EXT4_XATTR_REFCOUNT_MAX;
2973         int error;
2974
2975         if (!ea_block_cache)
2976                 return;
2977         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2978                                       bh->b_blocknr, reusable);
2979         if (error) {
2980                 if (error == -EBUSY)
2981                         ea_bdebug(bh, "already in cache");
2982         } else
2983                 ea_bdebug(bh, "inserting [%x]", (int)hash);
2984 }
2985
2986 /*
2987  * ext4_xattr_cmp()
2988  *
2989  * Compare two extended attribute blocks for equality.
2990  *
2991  * Returns 0 if the blocks are equal, 1 if they differ, and
2992  * a negative error number on errors.
2993  */
2994 static int
2995 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2996                struct ext4_xattr_header *header2)
2997 {
2998         struct ext4_xattr_entry *entry1, *entry2;
2999
3000         entry1 = ENTRY(header1+1);
3001         entry2 = ENTRY(header2+1);
3002         while (!IS_LAST_ENTRY(entry1)) {
3003                 if (IS_LAST_ENTRY(entry2))
3004                         return 1;
3005                 if (entry1->e_hash != entry2->e_hash ||
3006                     entry1->e_name_index != entry2->e_name_index ||
3007                     entry1->e_name_len != entry2->e_name_len ||
3008                     entry1->e_value_size != entry2->e_value_size ||
3009                     entry1->e_value_inum != entry2->e_value_inum ||
3010                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3011                         return 1;
3012                 if (!entry1->e_value_inum &&
3013                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3014                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3015                            le32_to_cpu(entry1->e_value_size)))
3016                         return 1;
3017
3018                 entry1 = EXT4_XATTR_NEXT(entry1);
3019                 entry2 = EXT4_XATTR_NEXT(entry2);
3020         }
3021         if (!IS_LAST_ENTRY(entry2))
3022                 return 1;
3023         return 0;
3024 }
3025
3026 /*
3027  * ext4_xattr_block_cache_find()
3028  *
3029  * Find an identical extended attribute block.
3030  *
3031  * Returns a pointer to the block found, or NULL if such a block was
3032  * not found or an error occurred.
3033  */
3034 static struct buffer_head *
3035 ext4_xattr_block_cache_find(struct inode *inode,
3036                             struct ext4_xattr_header *header,
3037                             struct mb_cache_entry **pce)
3038 {
3039         __u32 hash = le32_to_cpu(header->h_hash);
3040         struct mb_cache_entry *ce;
3041         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3042
3043         if (!ea_block_cache)
3044                 return NULL;
3045         if (!header->h_hash)
3046                 return NULL;  /* never share */
3047         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3048         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3049         while (ce) {
3050                 struct buffer_head *bh;
3051
3052                 bh = sb_bread(inode->i_sb, ce->e_value);
3053                 if (!bh) {
3054                         EXT4_ERROR_INODE(inode, "block %lu read error",
3055                                          (unsigned long)ce->e_value);
3056                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3057                         *pce = ce;
3058                         return bh;
3059                 }
3060                 brelse(bh);
3061                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3062         }
3063         return NULL;
3064 }
3065
3066 #define NAME_HASH_SHIFT 5
3067 #define VALUE_HASH_SHIFT 16
3068
3069 /*
3070  * ext4_xattr_hash_entry()
3071  *
3072  * Compute the hash of an extended attribute.
3073  */
3074 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3075                                     size_t value_count)
3076 {
3077         __u32 hash = 0;
3078
3079         while (name_len--) {
3080                 hash = (hash << NAME_HASH_SHIFT) ^
3081                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3082                        *name++;
3083         }
3084         while (value_count--) {
3085                 hash = (hash << VALUE_HASH_SHIFT) ^
3086                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3087                        le32_to_cpu(*value++);
3088         }
3089         return cpu_to_le32(hash);
3090 }
3091
3092 #undef NAME_HASH_SHIFT
3093 #undef VALUE_HASH_SHIFT
3094
3095 #define BLOCK_HASH_SHIFT 16
3096
3097 /*
3098  * ext4_xattr_rehash()
3099  *
3100  * Re-compute the extended attribute hash value after an entry has changed.
3101  */
3102 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3103 {
3104         struct ext4_xattr_entry *here;
3105         __u32 hash = 0;
3106
3107         here = ENTRY(header+1);
3108         while (!IS_LAST_ENTRY(here)) {
3109                 if (!here->e_hash) {
3110                         /* Block is not shared if an entry's hash value == 0 */
3111                         hash = 0;
3112                         break;
3113                 }
3114                 hash = (hash << BLOCK_HASH_SHIFT) ^
3115                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3116                        le32_to_cpu(here->e_hash);
3117                 here = EXT4_XATTR_NEXT(here);
3118         }
3119         header->h_hash = cpu_to_le32(hash);
3120 }
3121
3122 #undef BLOCK_HASH_SHIFT
3123
3124 #define HASH_BUCKET_BITS        10
3125
3126 struct mb_cache *
3127 ext4_xattr_create_cache(void)
3128 {
3129         return mb_cache_create(HASH_BUCKET_BITS);
3130 }
3131
3132 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3133 {
3134         if (cache)
3135                 mb_cache_destroy(cache);
3136 }
3137