Merge tag 'core-rcu-2020-06-01' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-microblaze.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, -error,
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, -error,
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, EXT4_IGET_NORMAL);
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         if (!EXT4_I(inode)->i_file_acl)
526                 return -ENODATA;
527         ea_idebug(inode, "reading block %llu",
528                   (unsigned long long)EXT4_I(inode)->i_file_acl);
529         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
530         if (IS_ERR(bh))
531                 return PTR_ERR(bh);
532         ea_bdebug(bh, "b_count=%d, refcount=%d",
533                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
534         error = ext4_xattr_check_block(inode, bh);
535         if (error)
536                 goto cleanup;
537         ext4_xattr_block_cache_insert(ea_block_cache, bh);
538         entry = BFIRST(bh);
539         end = bh->b_data + bh->b_size;
540         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
541         if (error)
542                 goto cleanup;
543         size = le32_to_cpu(entry->e_value_size);
544         error = -ERANGE;
545         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
546                 goto cleanup;
547         if (buffer) {
548                 if (size > buffer_size)
549                         goto cleanup;
550                 if (entry->e_value_inum) {
551                         error = ext4_xattr_inode_get(inode, entry, buffer,
552                                                      size);
553                         if (error)
554                                 goto cleanup;
555                 } else {
556                         u16 offset = le16_to_cpu(entry->e_value_offs);
557                         void *p = bh->b_data + offset;
558
559                         if (unlikely(p + size > end))
560                                 goto cleanup;
561                         memcpy(buffer, p, size);
562                 }
563         }
564         error = size;
565
566 cleanup:
567         brelse(bh);
568         return error;
569 }
570
571 int
572 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
573                      void *buffer, size_t buffer_size)
574 {
575         struct ext4_xattr_ibody_header *header;
576         struct ext4_xattr_entry *entry;
577         struct ext4_inode *raw_inode;
578         struct ext4_iloc iloc;
579         size_t size;
580         void *end;
581         int error;
582
583         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
584                 return -ENODATA;
585         error = ext4_get_inode_loc(inode, &iloc);
586         if (error)
587                 return error;
588         raw_inode = ext4_raw_inode(&iloc);
589         header = IHDR(inode, raw_inode);
590         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
591         error = xattr_check_inode(inode, header, end);
592         if (error)
593                 goto cleanup;
594         entry = IFIRST(header);
595         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
596         if (error)
597                 goto cleanup;
598         size = le32_to_cpu(entry->e_value_size);
599         error = -ERANGE;
600         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
601                 goto cleanup;
602         if (buffer) {
603                 if (size > buffer_size)
604                         goto cleanup;
605                 if (entry->e_value_inum) {
606                         error = ext4_xattr_inode_get(inode, entry, buffer,
607                                                      size);
608                         if (error)
609                                 goto cleanup;
610                 } else {
611                         u16 offset = le16_to_cpu(entry->e_value_offs);
612                         void *p = (void *)IFIRST(header) + offset;
613
614                         if (unlikely(p + size > end))
615                                 goto cleanup;
616                         memcpy(buffer, p, size);
617                 }
618         }
619         error = size;
620
621 cleanup:
622         brelse(iloc.bh);
623         return error;
624 }
625
626 /*
627  * ext4_xattr_get()
628  *
629  * Copy an extended attribute into the buffer
630  * provided, or compute the buffer size required.
631  * Buffer is NULL to compute the size of the buffer required.
632  *
633  * Returns a negative error number on failure, or the number of bytes
634  * used / required on success.
635  */
636 int
637 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
638                void *buffer, size_t buffer_size)
639 {
640         int error;
641
642         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
643                 return -EIO;
644
645         if (strlen(name) > 255)
646                 return -ERANGE;
647
648         down_read(&EXT4_I(inode)->xattr_sem);
649         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
650                                      buffer_size);
651         if (error == -ENODATA)
652                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
653                                              buffer_size);
654         up_read(&EXT4_I(inode)->xattr_sem);
655         return error;
656 }
657
658 static int
659 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
660                         char *buffer, size_t buffer_size)
661 {
662         size_t rest = buffer_size;
663
664         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
665                 const struct xattr_handler *handler =
666                         ext4_xattr_handler(entry->e_name_index);
667
668                 if (handler && (!handler->list || handler->list(dentry))) {
669                         const char *prefix = handler->prefix ?: handler->name;
670                         size_t prefix_len = strlen(prefix);
671                         size_t size = prefix_len + entry->e_name_len + 1;
672
673                         if (buffer) {
674                                 if (size > rest)
675                                         return -ERANGE;
676                                 memcpy(buffer, prefix, prefix_len);
677                                 buffer += prefix_len;
678                                 memcpy(buffer, entry->e_name, entry->e_name_len);
679                                 buffer += entry->e_name_len;
680                                 *buffer++ = 0;
681                         }
682                         rest -= size;
683                 }
684         }
685         return buffer_size - rest;  /* total size */
686 }
687
688 static int
689 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
690 {
691         struct inode *inode = d_inode(dentry);
692         struct buffer_head *bh = NULL;
693         int error;
694
695         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
696                   buffer, (long)buffer_size);
697
698         if (!EXT4_I(inode)->i_file_acl)
699                 return 0;
700         ea_idebug(inode, "reading block %llu",
701                   (unsigned long long)EXT4_I(inode)->i_file_acl);
702         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
703         if (IS_ERR(bh))
704                 return PTR_ERR(bh);
705         ea_bdebug(bh, "b_count=%d, refcount=%d",
706                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
707         error = ext4_xattr_check_block(inode, bh);
708         if (error)
709                 goto cleanup;
710         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
711         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
712                                         buffer_size);
713 cleanup:
714         brelse(bh);
715         return error;
716 }
717
718 static int
719 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
720 {
721         struct inode *inode = d_inode(dentry);
722         struct ext4_xattr_ibody_header *header;
723         struct ext4_inode *raw_inode;
724         struct ext4_iloc iloc;
725         void *end;
726         int error;
727
728         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
729                 return 0;
730         error = ext4_get_inode_loc(inode, &iloc);
731         if (error)
732                 return error;
733         raw_inode = ext4_raw_inode(&iloc);
734         header = IHDR(inode, raw_inode);
735         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
736         error = xattr_check_inode(inode, header, end);
737         if (error)
738                 goto cleanup;
739         error = ext4_xattr_list_entries(dentry, IFIRST(header),
740                                         buffer, buffer_size);
741
742 cleanup:
743         brelse(iloc.bh);
744         return error;
745 }
746
747 /*
748  * Inode operation listxattr()
749  *
750  * d_inode(dentry)->i_rwsem: don't care
751  *
752  * Copy a list of attribute names into the buffer
753  * provided, or compute the buffer size required.
754  * Buffer is NULL to compute the size of the buffer required.
755  *
756  * Returns a negative error number on failure, or the number of bytes
757  * used / required on success.
758  */
759 ssize_t
760 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
761 {
762         int ret, ret2;
763
764         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
765         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
766         if (ret < 0)
767                 goto errout;
768         if (buffer) {
769                 buffer += ret;
770                 buffer_size -= ret;
771         }
772         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
773         if (ret < 0)
774                 goto errout;
775         ret += ret2;
776 errout:
777         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
778         return ret;
779 }
780
781 /*
782  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
783  * not set, set it.
784  */
785 static void ext4_xattr_update_super_block(handle_t *handle,
786                                           struct super_block *sb)
787 {
788         if (ext4_has_feature_xattr(sb))
789                 return;
790
791         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
792         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
793                 ext4_set_feature_xattr(sb);
794                 ext4_handle_dirty_super(handle, sb);
795         }
796 }
797
798 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
799 {
800         struct ext4_iloc iloc = { .bh = NULL };
801         struct buffer_head *bh = NULL;
802         struct ext4_inode *raw_inode;
803         struct ext4_xattr_ibody_header *header;
804         struct ext4_xattr_entry *entry;
805         qsize_t ea_inode_refs = 0;
806         void *end;
807         int ret;
808
809         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
810
811         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
812                 ret = ext4_get_inode_loc(inode, &iloc);
813                 if (ret)
814                         goto out;
815                 raw_inode = ext4_raw_inode(&iloc);
816                 header = IHDR(inode, raw_inode);
817                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
818                 ret = xattr_check_inode(inode, header, end);
819                 if (ret)
820                         goto out;
821
822                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
823                      entry = EXT4_XATTR_NEXT(entry))
824                         if (entry->e_value_inum)
825                                 ea_inode_refs++;
826         }
827
828         if (EXT4_I(inode)->i_file_acl) {
829                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
830                 if (IS_ERR(bh)) {
831                         ret = PTR_ERR(bh);
832                         bh = NULL;
833                         goto out;
834                 }
835
836                 ret = ext4_xattr_check_block(inode, bh);
837                 if (ret)
838                         goto out;
839
840                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
841                      entry = EXT4_XATTR_NEXT(entry))
842                         if (entry->e_value_inum)
843                                 ea_inode_refs++;
844         }
845         *usage = ea_inode_refs + 1;
846         ret = 0;
847 out:
848         brelse(iloc.bh);
849         brelse(bh);
850         return ret;
851 }
852
853 static inline size_t round_up_cluster(struct inode *inode, size_t length)
854 {
855         struct super_block *sb = inode->i_sb;
856         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
857                                     inode->i_blkbits);
858         size_t mask = ~(cluster_size - 1);
859
860         return (length + cluster_size - 1) & mask;
861 }
862
863 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
864 {
865         int err;
866
867         err = dquot_alloc_inode(inode);
868         if (err)
869                 return err;
870         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
871         if (err)
872                 dquot_free_inode(inode);
873         return err;
874 }
875
876 static void ext4_xattr_inode_free_quota(struct inode *parent,
877                                         struct inode *ea_inode,
878                                         size_t len)
879 {
880         if (ea_inode &&
881             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
882                 return;
883         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
884         dquot_free_inode(parent);
885 }
886
887 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
888                              struct buffer_head *block_bh, size_t value_len,
889                              bool is_create)
890 {
891         int credits;
892         int blocks;
893
894         /*
895          * 1) Owner inode update
896          * 2) Ref count update on old xattr block
897          * 3) new xattr block
898          * 4) block bitmap update for new xattr block
899          * 5) group descriptor for new xattr block
900          * 6) block bitmap update for old xattr block
901          * 7) group descriptor for old block
902          *
903          * 6 & 7 can happen if we have two racing threads T_a and T_b
904          * which are each trying to set an xattr on inodes I_a and I_b
905          * which were both initially sharing an xattr block.
906          */
907         credits = 7;
908
909         /* Quota updates. */
910         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
911
912         /*
913          * In case of inline data, we may push out the data to a block,
914          * so we need to reserve credits for this eventuality
915          */
916         if (inode && ext4_has_inline_data(inode))
917                 credits += ext4_writepage_trans_blocks(inode) + 1;
918
919         /* We are done if ea_inode feature is not enabled. */
920         if (!ext4_has_feature_ea_inode(sb))
921                 return credits;
922
923         /* New ea_inode, inode map, block bitmap, group descriptor. */
924         credits += 4;
925
926         /* Data blocks. */
927         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
928
929         /* Indirection block or one level of extent tree. */
930         blocks += 1;
931
932         /* Block bitmap and group descriptor updates for each block. */
933         credits += blocks * 2;
934
935         /* Blocks themselves. */
936         credits += blocks;
937
938         if (!is_create) {
939                 /* Dereference ea_inode holding old xattr value.
940                  * Old ea_inode, inode map, block bitmap, group descriptor.
941                  */
942                 credits += 4;
943
944                 /* Data blocks for old ea_inode. */
945                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
946
947                 /* Indirection block or one level of extent tree for old
948                  * ea_inode.
949                  */
950                 blocks += 1;
951
952                 /* Block bitmap and group descriptor updates for each block. */
953                 credits += blocks * 2;
954         }
955
956         /* We may need to clone the existing xattr block in which case we need
957          * to increment ref counts for existing ea_inodes referenced by it.
958          */
959         if (block_bh) {
960                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
961
962                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
963                         if (entry->e_value_inum)
964                                 /* Ref count update on ea_inode. */
965                                 credits += 1;
966         }
967         return credits;
968 }
969
970 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
971                                        int ref_change)
972 {
973         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
974         struct ext4_iloc iloc;
975         s64 ref_count;
976         u32 hash;
977         int ret;
978
979         inode_lock(ea_inode);
980
981         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
982         if (ret)
983                 goto out;
984
985         ref_count = ext4_xattr_inode_get_ref(ea_inode);
986         ref_count += ref_change;
987         ext4_xattr_inode_set_ref(ea_inode, ref_count);
988
989         if (ref_change > 0) {
990                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
991                           ea_inode->i_ino, ref_count);
992
993                 if (ref_count == 1) {
994                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
995                                   ea_inode->i_ino, ea_inode->i_nlink);
996
997                         set_nlink(ea_inode, 1);
998                         ext4_orphan_del(handle, ea_inode);
999
1000                         if (ea_inode_cache) {
1001                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1002                                 mb_cache_entry_create(ea_inode_cache,
1003                                                       GFP_NOFS, hash,
1004                                                       ea_inode->i_ino,
1005                                                       true /* reusable */);
1006                         }
1007                 }
1008         } else {
1009                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1010                           ea_inode->i_ino, ref_count);
1011
1012                 if (ref_count == 0) {
1013                         WARN_ONCE(ea_inode->i_nlink != 1,
1014                                   "EA inode %lu i_nlink=%u",
1015                                   ea_inode->i_ino, ea_inode->i_nlink);
1016
1017                         clear_nlink(ea_inode);
1018                         ext4_orphan_add(handle, ea_inode);
1019
1020                         if (ea_inode_cache) {
1021                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1022                                 mb_cache_entry_delete(ea_inode_cache, hash,
1023                                                       ea_inode->i_ino);
1024                         }
1025                 }
1026         }
1027
1028         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1029         if (ret)
1030                 ext4_warning_inode(ea_inode,
1031                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1032 out:
1033         inode_unlock(ea_inode);
1034         return ret;
1035 }
1036
1037 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1038 {
1039         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1040 }
1041
1042 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1043 {
1044         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1045 }
1046
1047 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1048                                         struct ext4_xattr_entry *first)
1049 {
1050         struct inode *ea_inode;
1051         struct ext4_xattr_entry *entry;
1052         struct ext4_xattr_entry *failed_entry;
1053         unsigned int ea_ino;
1054         int err, saved_err;
1055
1056         for (entry = first; !IS_LAST_ENTRY(entry);
1057              entry = EXT4_XATTR_NEXT(entry)) {
1058                 if (!entry->e_value_inum)
1059                         continue;
1060                 ea_ino = le32_to_cpu(entry->e_value_inum);
1061                 err = ext4_xattr_inode_iget(parent, ea_ino,
1062                                             le32_to_cpu(entry->e_hash),
1063                                             &ea_inode);
1064                 if (err)
1065                         goto cleanup;
1066                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1067                 if (err) {
1068                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1069                         iput(ea_inode);
1070                         goto cleanup;
1071                 }
1072                 iput(ea_inode);
1073         }
1074         return 0;
1075
1076 cleanup:
1077         saved_err = err;
1078         failed_entry = entry;
1079
1080         for (entry = first; entry != failed_entry;
1081              entry = EXT4_XATTR_NEXT(entry)) {
1082                 if (!entry->e_value_inum)
1083                         continue;
1084                 ea_ino = le32_to_cpu(entry->e_value_inum);
1085                 err = ext4_xattr_inode_iget(parent, ea_ino,
1086                                             le32_to_cpu(entry->e_hash),
1087                                             &ea_inode);
1088                 if (err) {
1089                         ext4_warning(parent->i_sb,
1090                                      "cleanup ea_ino %u iget error %d", ea_ino,
1091                                      err);
1092                         continue;
1093                 }
1094                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1095                 if (err)
1096                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1097                                            err);
1098                 iput(ea_inode);
1099         }
1100         return saved_err;
1101 }
1102
1103 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1104                         struct buffer_head *bh, bool block_csum, bool dirty)
1105 {
1106         int error;
1107
1108         if (bh && dirty) {
1109                 if (block_csum)
1110                         ext4_xattr_block_csum_set(inode, bh);
1111                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1112                 if (error) {
1113                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1114                                      error);
1115                         return error;
1116                 }
1117         }
1118         return 0;
1119 }
1120
1121 static void
1122 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1123                              struct buffer_head *bh,
1124                              struct ext4_xattr_entry *first, bool block_csum,
1125                              struct ext4_xattr_inode_array **ea_inode_array,
1126                              int extra_credits, bool skip_quota)
1127 {
1128         struct inode *ea_inode;
1129         struct ext4_xattr_entry *entry;
1130         bool dirty = false;
1131         unsigned int ea_ino;
1132         int err;
1133         int credits;
1134
1135         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1136         credits = 2 + extra_credits;
1137
1138         for (entry = first; !IS_LAST_ENTRY(entry);
1139              entry = EXT4_XATTR_NEXT(entry)) {
1140                 if (!entry->e_value_inum)
1141                         continue;
1142                 ea_ino = le32_to_cpu(entry->e_value_inum);
1143                 err = ext4_xattr_inode_iget(parent, ea_ino,
1144                                             le32_to_cpu(entry->e_hash),
1145                                             &ea_inode);
1146                 if (err)
1147                         continue;
1148
1149                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1150                 if (err) {
1151                         ext4_warning_inode(ea_inode,
1152                                            "Expand inode array err=%d", err);
1153                         iput(ea_inode);
1154                         continue;
1155                 }
1156
1157                 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1158                         ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1159                         ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1160                                               dirty));
1161                 if (err < 0) {
1162                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1163                                            err);
1164                         continue;
1165                 }
1166                 if (err > 0) {
1167                         err = ext4_journal_get_write_access(handle, bh);
1168                         if (err) {
1169                                 ext4_warning_inode(ea_inode,
1170                                                 "Re-get write access err=%d",
1171                                                 err);
1172                                 continue;
1173                         }
1174                 }
1175
1176                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1177                 if (err) {
1178                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1179                                            err);
1180                         continue;
1181                 }
1182
1183                 if (!skip_quota)
1184                         ext4_xattr_inode_free_quota(parent, ea_inode,
1185                                               le32_to_cpu(entry->e_value_size));
1186
1187                 /*
1188                  * Forget about ea_inode within the same transaction that
1189                  * decrements the ref count. This avoids duplicate decrements in
1190                  * case the rest of the work spills over to subsequent
1191                  * transactions.
1192                  */
1193                 entry->e_value_inum = 0;
1194                 entry->e_value_size = 0;
1195
1196                 dirty = true;
1197         }
1198
1199         if (dirty) {
1200                 /*
1201                  * Note that we are deliberately skipping csum calculation for
1202                  * the final update because we do not expect any journal
1203                  * restarts until xattr block is freed.
1204                  */
1205
1206                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1207                 if (err)
1208                         ext4_warning_inode(parent,
1209                                            "handle dirty metadata err=%d", err);
1210         }
1211 }
1212
1213 /*
1214  * Release the xattr block BH: If the reference count is > 1, decrement it;
1215  * otherwise free the block.
1216  */
1217 static void
1218 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1219                          struct buffer_head *bh,
1220                          struct ext4_xattr_inode_array **ea_inode_array,
1221                          int extra_credits)
1222 {
1223         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1224         u32 hash, ref;
1225         int error = 0;
1226
1227         BUFFER_TRACE(bh, "get_write_access");
1228         error = ext4_journal_get_write_access(handle, bh);
1229         if (error)
1230                 goto out;
1231
1232         lock_buffer(bh);
1233         hash = le32_to_cpu(BHDR(bh)->h_hash);
1234         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1235         if (ref == 1) {
1236                 ea_bdebug(bh, "refcount now=0; freeing");
1237                 /*
1238                  * This must happen under buffer lock for
1239                  * ext4_xattr_block_set() to reliably detect freed block
1240                  */
1241                 if (ea_block_cache)
1242                         mb_cache_entry_delete(ea_block_cache, hash,
1243                                               bh->b_blocknr);
1244                 get_bh(bh);
1245                 unlock_buffer(bh);
1246
1247                 if (ext4_has_feature_ea_inode(inode->i_sb))
1248                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1249                                                      BFIRST(bh),
1250                                                      true /* block_csum */,
1251                                                      ea_inode_array,
1252                                                      extra_credits,
1253                                                      true /* skip_quota */);
1254                 ext4_free_blocks(handle, inode, bh, 0, 1,
1255                                  EXT4_FREE_BLOCKS_METADATA |
1256                                  EXT4_FREE_BLOCKS_FORGET);
1257         } else {
1258                 ref--;
1259                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1260                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1261                         struct mb_cache_entry *ce;
1262
1263                         if (ea_block_cache) {
1264                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1265                                                         bh->b_blocknr);
1266                                 if (ce) {
1267                                         ce->e_reusable = 1;
1268                                         mb_cache_entry_put(ea_block_cache, ce);
1269                                 }
1270                         }
1271                 }
1272
1273                 ext4_xattr_block_csum_set(inode, bh);
1274                 /*
1275                  * Beware of this ugliness: Releasing of xattr block references
1276                  * from different inodes can race and so we have to protect
1277                  * from a race where someone else frees the block (and releases
1278                  * its journal_head) before we are done dirtying the buffer. In
1279                  * nojournal mode this race is harmless and we actually cannot
1280                  * call ext4_handle_dirty_metadata() with locked buffer as
1281                  * that function can call sync_dirty_buffer() so for that case
1282                  * we handle the dirtying after unlocking the buffer.
1283                  */
1284                 if (ext4_handle_valid(handle))
1285                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1286                 unlock_buffer(bh);
1287                 if (!ext4_handle_valid(handle))
1288                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1289                 if (IS_SYNC(inode))
1290                         ext4_handle_sync(handle);
1291                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1292                 ea_bdebug(bh, "refcount now=%d; releasing",
1293                           le32_to_cpu(BHDR(bh)->h_refcount));
1294         }
1295 out:
1296         ext4_std_error(inode->i_sb, error);
1297         return;
1298 }
1299
1300 /*
1301  * Find the available free space for EAs. This also returns the total number of
1302  * bytes used by EA entries.
1303  */
1304 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1305                                     size_t *min_offs, void *base, int *total)
1306 {
1307         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1308                 if (!last->e_value_inum && last->e_value_size) {
1309                         size_t offs = le16_to_cpu(last->e_value_offs);
1310                         if (offs < *min_offs)
1311                                 *min_offs = offs;
1312                 }
1313                 if (total)
1314                         *total += EXT4_XATTR_LEN(last->e_name_len);
1315         }
1316         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1317 }
1318
1319 /*
1320  * Write the value of the EA in an inode.
1321  */
1322 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1323                                   const void *buf, int bufsize)
1324 {
1325         struct buffer_head *bh = NULL;
1326         unsigned long block = 0;
1327         int blocksize = ea_inode->i_sb->s_blocksize;
1328         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1329         int csize, wsize = 0;
1330         int ret = 0;
1331         int retries = 0;
1332
1333 retry:
1334         while (ret >= 0 && ret < max_blocks) {
1335                 struct ext4_map_blocks map;
1336                 map.m_lblk = block += ret;
1337                 map.m_len = max_blocks -= ret;
1338
1339                 ret = ext4_map_blocks(handle, ea_inode, &map,
1340                                       EXT4_GET_BLOCKS_CREATE);
1341                 if (ret <= 0) {
1342                         ext4_mark_inode_dirty(handle, ea_inode);
1343                         if (ret == -ENOSPC &&
1344                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1345                                 ret = 0;
1346                                 goto retry;
1347                         }
1348                         break;
1349                 }
1350         }
1351
1352         if (ret < 0)
1353                 return ret;
1354
1355         block = 0;
1356         while (wsize < bufsize) {
1357                 if (bh != NULL)
1358                         brelse(bh);
1359                 csize = (bufsize - wsize) > blocksize ? blocksize :
1360                                                                 bufsize - wsize;
1361                 bh = ext4_getblk(handle, ea_inode, block, 0);
1362                 if (IS_ERR(bh))
1363                         return PTR_ERR(bh);
1364                 if (!bh) {
1365                         WARN_ON_ONCE(1);
1366                         EXT4_ERROR_INODE(ea_inode,
1367                                          "ext4_getblk() return bh = NULL");
1368                         return -EFSCORRUPTED;
1369                 }
1370                 ret = ext4_journal_get_write_access(handle, bh);
1371                 if (ret)
1372                         goto out;
1373
1374                 memcpy(bh->b_data, buf, csize);
1375                 set_buffer_uptodate(bh);
1376                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1377
1378                 buf += csize;
1379                 wsize += csize;
1380                 block += 1;
1381         }
1382
1383         inode_lock(ea_inode);
1384         i_size_write(ea_inode, wsize);
1385         ext4_update_i_disksize(ea_inode, wsize);
1386         inode_unlock(ea_inode);
1387
1388         ext4_mark_inode_dirty(handle, ea_inode);
1389
1390 out:
1391         brelse(bh);
1392
1393         return ret;
1394 }
1395
1396 /*
1397  * Create an inode to store the value of a large EA.
1398  */
1399 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1400                                              struct inode *inode, u32 hash)
1401 {
1402         struct inode *ea_inode = NULL;
1403         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1404         int err;
1405
1406         /*
1407          * Let the next inode be the goal, so we try and allocate the EA inode
1408          * in the same group, or nearby one.
1409          */
1410         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1411                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1412                                   EXT4_EA_INODE_FL);
1413         if (!IS_ERR(ea_inode)) {
1414                 ea_inode->i_op = &ext4_file_inode_operations;
1415                 ea_inode->i_fop = &ext4_file_operations;
1416                 ext4_set_aops(ea_inode);
1417                 ext4_xattr_inode_set_class(ea_inode);
1418                 unlock_new_inode(ea_inode);
1419                 ext4_xattr_inode_set_ref(ea_inode, 1);
1420                 ext4_xattr_inode_set_hash(ea_inode, hash);
1421                 err = ext4_mark_inode_dirty(handle, ea_inode);
1422                 if (!err)
1423                         err = ext4_inode_attach_jinode(ea_inode);
1424                 if (err) {
1425                         iput(ea_inode);
1426                         return ERR_PTR(err);
1427                 }
1428
1429                 /*
1430                  * Xattr inodes are shared therefore quota charging is performed
1431                  * at a higher level.
1432                  */
1433                 dquot_free_inode(ea_inode);
1434                 dquot_drop(ea_inode);
1435                 inode_lock(ea_inode);
1436                 ea_inode->i_flags |= S_NOQUOTA;
1437                 inode_unlock(ea_inode);
1438         }
1439
1440         return ea_inode;
1441 }
1442
1443 static struct inode *
1444 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1445                             size_t value_len, u32 hash)
1446 {
1447         struct inode *ea_inode;
1448         struct mb_cache_entry *ce;
1449         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1450         void *ea_data;
1451
1452         if (!ea_inode_cache)
1453                 return NULL;
1454
1455         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1456         if (!ce)
1457                 return NULL;
1458
1459         ea_data = kvmalloc(value_len, GFP_KERNEL);
1460         if (!ea_data) {
1461                 mb_cache_entry_put(ea_inode_cache, ce);
1462                 return NULL;
1463         }
1464
1465         while (ce) {
1466                 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1467                                      EXT4_IGET_NORMAL);
1468                 if (!IS_ERR(ea_inode) &&
1469                     !is_bad_inode(ea_inode) &&
1470                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1471                     i_size_read(ea_inode) == value_len &&
1472                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1473                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1474                                                     value_len) &&
1475                     !memcmp(value, ea_data, value_len)) {
1476                         mb_cache_entry_touch(ea_inode_cache, ce);
1477                         mb_cache_entry_put(ea_inode_cache, ce);
1478                         kvfree(ea_data);
1479                         return ea_inode;
1480                 }
1481
1482                 if (!IS_ERR(ea_inode))
1483                         iput(ea_inode);
1484                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1485         }
1486         kvfree(ea_data);
1487         return NULL;
1488 }
1489
1490 /*
1491  * Add value of the EA in an inode.
1492  */
1493 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1494                                           const void *value, size_t value_len,
1495                                           struct inode **ret_inode)
1496 {
1497         struct inode *ea_inode;
1498         u32 hash;
1499         int err;
1500
1501         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1502         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1503         if (ea_inode) {
1504                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1505                 if (err) {
1506                         iput(ea_inode);
1507                         return err;
1508                 }
1509
1510                 *ret_inode = ea_inode;
1511                 return 0;
1512         }
1513
1514         /* Create an inode for the EA value */
1515         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1516         if (IS_ERR(ea_inode))
1517                 return PTR_ERR(ea_inode);
1518
1519         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1520         if (err) {
1521                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1522                 iput(ea_inode);
1523                 return err;
1524         }
1525
1526         if (EA_INODE_CACHE(inode))
1527                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1528                                       ea_inode->i_ino, true /* reusable */);
1529
1530         *ret_inode = ea_inode;
1531         return 0;
1532 }
1533
1534 /*
1535  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1536  * feature is enabled.
1537  */
1538 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1539
1540 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1541                                 struct ext4_xattr_search *s,
1542                                 handle_t *handle, struct inode *inode,
1543                                 bool is_block)
1544 {
1545         struct ext4_xattr_entry *last, *next;
1546         struct ext4_xattr_entry *here = s->here;
1547         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1548         int in_inode = i->in_inode;
1549         struct inode *old_ea_inode = NULL;
1550         struct inode *new_ea_inode = NULL;
1551         size_t old_size, new_size;
1552         int ret;
1553
1554         /* Space used by old and new values. */
1555         old_size = (!s->not_found && !here->e_value_inum) ?
1556                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1557         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1558
1559         /*
1560          * Optimization for the simple case when old and new values have the
1561          * same padded sizes. Not applicable if external inodes are involved.
1562          */
1563         if (new_size && new_size == old_size) {
1564                 size_t offs = le16_to_cpu(here->e_value_offs);
1565                 void *val = s->base + offs;
1566
1567                 here->e_value_size = cpu_to_le32(i->value_len);
1568                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1569                         memset(val, 0, new_size);
1570                 } else {
1571                         memcpy(val, i->value, i->value_len);
1572                         /* Clear padding bytes. */
1573                         memset(val + i->value_len, 0, new_size - i->value_len);
1574                 }
1575                 goto update_hash;
1576         }
1577
1578         /* Compute min_offs and last. */
1579         last = s->first;
1580         for (; !IS_LAST_ENTRY(last); last = next) {
1581                 next = EXT4_XATTR_NEXT(last);
1582                 if ((void *)next >= s->end) {
1583                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1584                         ret = -EFSCORRUPTED;
1585                         goto out;
1586                 }
1587                 if (!last->e_value_inum && last->e_value_size) {
1588                         size_t offs = le16_to_cpu(last->e_value_offs);
1589                         if (offs < min_offs)
1590                                 min_offs = offs;
1591                 }
1592         }
1593
1594         /* Check whether we have enough space. */
1595         if (i->value) {
1596                 size_t free;
1597
1598                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1599                 if (!s->not_found)
1600                         free += EXT4_XATTR_LEN(name_len) + old_size;
1601
1602                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1603                         ret = -ENOSPC;
1604                         goto out;
1605                 }
1606
1607                 /*
1608                  * If storing the value in an external inode is an option,
1609                  * reserve space for xattr entries/names in the external
1610                  * attribute block so that a long value does not occupy the
1611                  * whole space and prevent futher entries being added.
1612                  */
1613                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1614                     new_size && is_block &&
1615                     (min_offs + old_size - new_size) <
1616                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1617                         ret = -ENOSPC;
1618                         goto out;
1619                 }
1620         }
1621
1622         /*
1623          * Getting access to old and new ea inodes is subject to failures.
1624          * Finish that work before doing any modifications to the xattr data.
1625          */
1626         if (!s->not_found && here->e_value_inum) {
1627                 ret = ext4_xattr_inode_iget(inode,
1628                                             le32_to_cpu(here->e_value_inum),
1629                                             le32_to_cpu(here->e_hash),
1630                                             &old_ea_inode);
1631                 if (ret) {
1632                         old_ea_inode = NULL;
1633                         goto out;
1634                 }
1635         }
1636         if (i->value && in_inode) {
1637                 WARN_ON_ONCE(!i->value_len);
1638
1639                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1640                 if (ret)
1641                         goto out;
1642
1643                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1644                                                      i->value_len,
1645                                                      &new_ea_inode);
1646                 if (ret) {
1647                         new_ea_inode = NULL;
1648                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1649                         goto out;
1650                 }
1651         }
1652
1653         if (old_ea_inode) {
1654                 /* We are ready to release ref count on the old_ea_inode. */
1655                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1656                 if (ret) {
1657                         /* Release newly required ref count on new_ea_inode. */
1658                         if (new_ea_inode) {
1659                                 int err;
1660
1661                                 err = ext4_xattr_inode_dec_ref(handle,
1662                                                                new_ea_inode);
1663                                 if (err)
1664                                         ext4_warning_inode(new_ea_inode,
1665                                                   "dec ref new_ea_inode err=%d",
1666                                                   err);
1667                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1668                                                             i->value_len);
1669                         }
1670                         goto out;
1671                 }
1672
1673                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1674                                             le32_to_cpu(here->e_value_size));
1675         }
1676
1677         /* No failures allowed past this point. */
1678
1679         if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1680                 /* Remove the old value. */
1681                 void *first_val = s->base + min_offs;
1682                 size_t offs = le16_to_cpu(here->e_value_offs);
1683                 void *val = s->base + offs;
1684
1685                 memmove(first_val + old_size, first_val, val - first_val);
1686                 memset(first_val, 0, old_size);
1687                 min_offs += old_size;
1688
1689                 /* Adjust all value offsets. */
1690                 last = s->first;
1691                 while (!IS_LAST_ENTRY(last)) {
1692                         size_t o = le16_to_cpu(last->e_value_offs);
1693
1694                         if (!last->e_value_inum &&
1695                             last->e_value_size && o < offs)
1696                                 last->e_value_offs = cpu_to_le16(o + old_size);
1697                         last = EXT4_XATTR_NEXT(last);
1698                 }
1699         }
1700
1701         if (!i->value) {
1702                 /* Remove old name. */
1703                 size_t size = EXT4_XATTR_LEN(name_len);
1704
1705                 last = ENTRY((void *)last - size);
1706                 memmove(here, (void *)here + size,
1707                         (void *)last - (void *)here + sizeof(__u32));
1708                 memset(last, 0, size);
1709         } else if (s->not_found) {
1710                 /* Insert new name. */
1711                 size_t size = EXT4_XATTR_LEN(name_len);
1712                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1713
1714                 memmove((void *)here + size, here, rest);
1715                 memset(here, 0, size);
1716                 here->e_name_index = i->name_index;
1717                 here->e_name_len = name_len;
1718                 memcpy(here->e_name, i->name, name_len);
1719         } else {
1720                 /* This is an update, reset value info. */
1721                 here->e_value_inum = 0;
1722                 here->e_value_offs = 0;
1723                 here->e_value_size = 0;
1724         }
1725
1726         if (i->value) {
1727                 /* Insert new value. */
1728                 if (in_inode) {
1729                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1730                 } else if (i->value_len) {
1731                         void *val = s->base + min_offs - new_size;
1732
1733                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1734                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1735                                 memset(val, 0, new_size);
1736                         } else {
1737                                 memcpy(val, i->value, i->value_len);
1738                                 /* Clear padding bytes. */
1739                                 memset(val + i->value_len, 0,
1740                                        new_size - i->value_len);
1741                         }
1742                 }
1743                 here->e_value_size = cpu_to_le32(i->value_len);
1744         }
1745
1746 update_hash:
1747         if (i->value) {
1748                 __le32 hash = 0;
1749
1750                 /* Entry hash calculation. */
1751                 if (in_inode) {
1752                         __le32 crc32c_hash;
1753
1754                         /*
1755                          * Feed crc32c hash instead of the raw value for entry
1756                          * hash calculation. This is to avoid walking
1757                          * potentially long value buffer again.
1758                          */
1759                         crc32c_hash = cpu_to_le32(
1760                                        ext4_xattr_inode_get_hash(new_ea_inode));
1761                         hash = ext4_xattr_hash_entry(here->e_name,
1762                                                      here->e_name_len,
1763                                                      &crc32c_hash, 1);
1764                 } else if (is_block) {
1765                         __le32 *value = s->base + le16_to_cpu(
1766                                                         here->e_value_offs);
1767
1768                         hash = ext4_xattr_hash_entry(here->e_name,
1769                                                      here->e_name_len, value,
1770                                                      new_size >> 2);
1771                 }
1772                 here->e_hash = hash;
1773         }
1774
1775         if (is_block)
1776                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1777
1778         ret = 0;
1779 out:
1780         iput(old_ea_inode);
1781         iput(new_ea_inode);
1782         return ret;
1783 }
1784
1785 struct ext4_xattr_block_find {
1786         struct ext4_xattr_search s;
1787         struct buffer_head *bh;
1788 };
1789
1790 static int
1791 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1792                       struct ext4_xattr_block_find *bs)
1793 {
1794         struct super_block *sb = inode->i_sb;
1795         int error;
1796
1797         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1798                   i->name_index, i->name, i->value, (long)i->value_len);
1799
1800         if (EXT4_I(inode)->i_file_acl) {
1801                 /* The inode already has an extended attribute block. */
1802                 bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1803                 if (IS_ERR(bs->bh))
1804                         return PTR_ERR(bs->bh);
1805                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1806                         atomic_read(&(bs->bh->b_count)),
1807                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1808                 error = ext4_xattr_check_block(inode, bs->bh);
1809                 if (error)
1810                         return error;
1811                 /* Find the named attribute. */
1812                 bs->s.base = BHDR(bs->bh);
1813                 bs->s.first = BFIRST(bs->bh);
1814                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1815                 bs->s.here = bs->s.first;
1816                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1817                                          i->name_index, i->name, 1);
1818                 if (error && error != -ENODATA)
1819                         return error;
1820                 bs->s.not_found = error;
1821         }
1822         return 0;
1823 }
1824
1825 static int
1826 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1827                      struct ext4_xattr_info *i,
1828                      struct ext4_xattr_block_find *bs)
1829 {
1830         struct super_block *sb = inode->i_sb;
1831         struct buffer_head *new_bh = NULL;
1832         struct ext4_xattr_search s_copy = bs->s;
1833         struct ext4_xattr_search *s = &s_copy;
1834         struct mb_cache_entry *ce = NULL;
1835         int error = 0;
1836         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1837         struct inode *ea_inode = NULL, *tmp_inode;
1838         size_t old_ea_inode_quota = 0;
1839         unsigned int ea_ino;
1840
1841
1842 #define header(x) ((struct ext4_xattr_header *)(x))
1843
1844         if (s->base) {
1845                 BUFFER_TRACE(bs->bh, "get_write_access");
1846                 error = ext4_journal_get_write_access(handle, bs->bh);
1847                 if (error)
1848                         goto cleanup;
1849                 lock_buffer(bs->bh);
1850
1851                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1852                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1853
1854                         /*
1855                          * This must happen under buffer lock for
1856                          * ext4_xattr_block_set() to reliably detect modified
1857                          * block
1858                          */
1859                         if (ea_block_cache)
1860                                 mb_cache_entry_delete(ea_block_cache, hash,
1861                                                       bs->bh->b_blocknr);
1862                         ea_bdebug(bs->bh, "modifying in-place");
1863                         error = ext4_xattr_set_entry(i, s, handle, inode,
1864                                                      true /* is_block */);
1865                         ext4_xattr_block_csum_set(inode, bs->bh);
1866                         unlock_buffer(bs->bh);
1867                         if (error == -EFSCORRUPTED)
1868                                 goto bad_block;
1869                         if (!error)
1870                                 error = ext4_handle_dirty_metadata(handle,
1871                                                                    inode,
1872                                                                    bs->bh);
1873                         if (error)
1874                                 goto cleanup;
1875                         goto inserted;
1876                 } else {
1877                         int offset = (char *)s->here - bs->bh->b_data;
1878
1879                         unlock_buffer(bs->bh);
1880                         ea_bdebug(bs->bh, "cloning");
1881                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1882                         error = -ENOMEM;
1883                         if (s->base == NULL)
1884                                 goto cleanup;
1885                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1886                         s->first = ENTRY(header(s->base)+1);
1887                         header(s->base)->h_refcount = cpu_to_le32(1);
1888                         s->here = ENTRY(s->base + offset);
1889                         s->end = s->base + bs->bh->b_size;
1890
1891                         /*
1892                          * If existing entry points to an xattr inode, we need
1893                          * to prevent ext4_xattr_set_entry() from decrementing
1894                          * ref count on it because the reference belongs to the
1895                          * original block. In this case, make the entry look
1896                          * like it has an empty value.
1897                          */
1898                         if (!s->not_found && s->here->e_value_inum) {
1899                                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1900                                 error = ext4_xattr_inode_iget(inode, ea_ino,
1901                                               le32_to_cpu(s->here->e_hash),
1902                                               &tmp_inode);
1903                                 if (error)
1904                                         goto cleanup;
1905
1906                                 if (!ext4_test_inode_state(tmp_inode,
1907                                                 EXT4_STATE_LUSTRE_EA_INODE)) {
1908                                         /*
1909                                          * Defer quota free call for previous
1910                                          * inode until success is guaranteed.
1911                                          */
1912                                         old_ea_inode_quota = le32_to_cpu(
1913                                                         s->here->e_value_size);
1914                                 }
1915                                 iput(tmp_inode);
1916
1917                                 s->here->e_value_inum = 0;
1918                                 s->here->e_value_size = 0;
1919                         }
1920                 }
1921         } else {
1922                 /* Allocate a buffer where we construct the new block. */
1923                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1924                 /* assert(header == s->base) */
1925                 error = -ENOMEM;
1926                 if (s->base == NULL)
1927                         goto cleanup;
1928                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1929                 header(s->base)->h_blocks = cpu_to_le32(1);
1930                 header(s->base)->h_refcount = cpu_to_le32(1);
1931                 s->first = ENTRY(header(s->base)+1);
1932                 s->here = ENTRY(header(s->base)+1);
1933                 s->end = s->base + sb->s_blocksize;
1934         }
1935
1936         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1937         if (error == -EFSCORRUPTED)
1938                 goto bad_block;
1939         if (error)
1940                 goto cleanup;
1941
1942         if (i->value && s->here->e_value_inum) {
1943                 /*
1944                  * A ref count on ea_inode has been taken as part of the call to
1945                  * ext4_xattr_set_entry() above. We would like to drop this
1946                  * extra ref but we have to wait until the xattr block is
1947                  * initialized and has its own ref count on the ea_inode.
1948                  */
1949                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1950                 error = ext4_xattr_inode_iget(inode, ea_ino,
1951                                               le32_to_cpu(s->here->e_hash),
1952                                               &ea_inode);
1953                 if (error) {
1954                         ea_inode = NULL;
1955                         goto cleanup;
1956                 }
1957         }
1958
1959 inserted:
1960         if (!IS_LAST_ENTRY(s->first)) {
1961                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1962                                                      &ce);
1963                 if (new_bh) {
1964                         /* We found an identical block in the cache. */
1965                         if (new_bh == bs->bh)
1966                                 ea_bdebug(new_bh, "keeping");
1967                         else {
1968                                 u32 ref;
1969
1970                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
1971
1972                                 /* The old block is released after updating
1973                                    the inode. */
1974                                 error = dquot_alloc_block(inode,
1975                                                 EXT4_C2B(EXT4_SB(sb), 1));
1976                                 if (error)
1977                                         goto cleanup;
1978                                 BUFFER_TRACE(new_bh, "get_write_access");
1979                                 error = ext4_journal_get_write_access(handle,
1980                                                                       new_bh);
1981                                 if (error)
1982                                         goto cleanup_dquot;
1983                                 lock_buffer(new_bh);
1984                                 /*
1985                                  * We have to be careful about races with
1986                                  * freeing, rehashing or adding references to
1987                                  * xattr block. Once we hold buffer lock xattr
1988                                  * block's state is stable so we can check
1989                                  * whether the block got freed / rehashed or
1990                                  * not.  Since we unhash mbcache entry under
1991                                  * buffer lock when freeing / rehashing xattr
1992                                  * block, checking whether entry is still
1993                                  * hashed is reliable. Same rules hold for
1994                                  * e_reusable handling.
1995                                  */
1996                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
1997                                     !ce->e_reusable) {
1998                                         /*
1999                                          * Undo everything and check mbcache
2000                                          * again.
2001                                          */
2002                                         unlock_buffer(new_bh);
2003                                         dquot_free_block(inode,
2004                                                          EXT4_C2B(EXT4_SB(sb),
2005                                                                   1));
2006                                         brelse(new_bh);
2007                                         mb_cache_entry_put(ea_block_cache, ce);
2008                                         ce = NULL;
2009                                         new_bh = NULL;
2010                                         goto inserted;
2011                                 }
2012                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2013                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2014                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2015                                         ce->e_reusable = 0;
2016                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2017                                           ref);
2018                                 ext4_xattr_block_csum_set(inode, new_bh);
2019                                 unlock_buffer(new_bh);
2020                                 error = ext4_handle_dirty_metadata(handle,
2021                                                                    inode,
2022                                                                    new_bh);
2023                                 if (error)
2024                                         goto cleanup_dquot;
2025                         }
2026                         mb_cache_entry_touch(ea_block_cache, ce);
2027                         mb_cache_entry_put(ea_block_cache, ce);
2028                         ce = NULL;
2029                 } else if (bs->bh && s->base == bs->bh->b_data) {
2030                         /* We were modifying this block in-place. */
2031                         ea_bdebug(bs->bh, "keeping this block");
2032                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2033                         new_bh = bs->bh;
2034                         get_bh(new_bh);
2035                 } else {
2036                         /* We need to allocate a new block */
2037                         ext4_fsblk_t goal, block;
2038
2039                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2040
2041                         goal = ext4_group_first_block_no(sb,
2042                                                 EXT4_I(inode)->i_block_group);
2043
2044                         /* non-extent files can't have physical blocks past 2^32 */
2045                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2046                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2047
2048                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2049                                                      NULL, &error);
2050                         if (error)
2051                                 goto cleanup;
2052
2053                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2054                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2055
2056                         ea_idebug(inode, "creating block %llu",
2057                                   (unsigned long long)block);
2058
2059                         new_bh = sb_getblk(sb, block);
2060                         if (unlikely(!new_bh)) {
2061                                 error = -ENOMEM;
2062 getblk_failed:
2063                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2064                                                  EXT4_FREE_BLOCKS_METADATA);
2065                                 goto cleanup;
2066                         }
2067                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2068                                                       ENTRY(header(s->base)+1));
2069                         if (error)
2070                                 goto getblk_failed;
2071                         if (ea_inode) {
2072                                 /* Drop the extra ref on ea_inode. */
2073                                 error = ext4_xattr_inode_dec_ref(handle,
2074                                                                  ea_inode);
2075                                 if (error)
2076                                         ext4_warning_inode(ea_inode,
2077                                                            "dec ref error=%d",
2078                                                            error);
2079                                 iput(ea_inode);
2080                                 ea_inode = NULL;
2081                         }
2082
2083                         lock_buffer(new_bh);
2084                         error = ext4_journal_get_create_access(handle, new_bh);
2085                         if (error) {
2086                                 unlock_buffer(new_bh);
2087                                 error = -EIO;
2088                                 goto getblk_failed;
2089                         }
2090                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2091                         ext4_xattr_block_csum_set(inode, new_bh);
2092                         set_buffer_uptodate(new_bh);
2093                         unlock_buffer(new_bh);
2094                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2095                         error = ext4_handle_dirty_metadata(handle, inode,
2096                                                            new_bh);
2097                         if (error)
2098                                 goto cleanup;
2099                 }
2100         }
2101
2102         if (old_ea_inode_quota)
2103                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2104
2105         /* Update the inode. */
2106         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2107
2108         /* Drop the previous xattr block. */
2109         if (bs->bh && bs->bh != new_bh) {
2110                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2111
2112                 ext4_xattr_release_block(handle, inode, bs->bh,
2113                                          &ea_inode_array,
2114                                          0 /* extra_credits */);
2115                 ext4_xattr_inode_array_free(ea_inode_array);
2116         }
2117         error = 0;
2118
2119 cleanup:
2120         if (ea_inode) {
2121                 int error2;
2122
2123                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2124                 if (error2)
2125                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2126                                            error2);
2127
2128                 /* If there was an error, revert the quota charge. */
2129                 if (error)
2130                         ext4_xattr_inode_free_quota(inode, ea_inode,
2131                                                     i_size_read(ea_inode));
2132                 iput(ea_inode);
2133         }
2134         if (ce)
2135                 mb_cache_entry_put(ea_block_cache, ce);
2136         brelse(new_bh);
2137         if (!(bs->bh && s->base == bs->bh->b_data))
2138                 kfree(s->base);
2139
2140         return error;
2141
2142 cleanup_dquot:
2143         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2144         goto cleanup;
2145
2146 bad_block:
2147         EXT4_ERROR_INODE(inode, "bad block %llu",
2148                          EXT4_I(inode)->i_file_acl);
2149         goto cleanup;
2150
2151 #undef header
2152 }
2153
2154 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2155                           struct ext4_xattr_ibody_find *is)
2156 {
2157         struct ext4_xattr_ibody_header *header;
2158         struct ext4_inode *raw_inode;
2159         int error;
2160
2161         if (EXT4_I(inode)->i_extra_isize == 0)
2162                 return 0;
2163         raw_inode = ext4_raw_inode(&is->iloc);
2164         header = IHDR(inode, raw_inode);
2165         is->s.base = is->s.first = IFIRST(header);
2166         is->s.here = is->s.first;
2167         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2168         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2169                 error = xattr_check_inode(inode, header, is->s.end);
2170                 if (error)
2171                         return error;
2172                 /* Find the named attribute. */
2173                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2174                                          i->name_index, i->name, 0);
2175                 if (error && error != -ENODATA)
2176                         return error;
2177                 is->s.not_found = error;
2178         }
2179         return 0;
2180 }
2181
2182 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2183                                 struct ext4_xattr_info *i,
2184                                 struct ext4_xattr_ibody_find *is)
2185 {
2186         struct ext4_xattr_ibody_header *header;
2187         struct ext4_xattr_search *s = &is->s;
2188         int error;
2189
2190         if (EXT4_I(inode)->i_extra_isize == 0)
2191                 return -ENOSPC;
2192         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2193         if (error)
2194                 return error;
2195         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2196         if (!IS_LAST_ENTRY(s->first)) {
2197                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2198                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2199         } else {
2200                 header->h_magic = cpu_to_le32(0);
2201                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2202         }
2203         return 0;
2204 }
2205
2206 static int ext4_xattr_ibody_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_value_same(struct ext4_xattr_search *s,
2231                                  struct ext4_xattr_info *i)
2232 {
2233         void *value;
2234
2235         /* When e_value_inum is set the value is stored externally. */
2236         if (s->here->e_value_inum)
2237                 return 0;
2238         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2239                 return 0;
2240         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2241         return !memcmp(value, i->value, i->value_len);
2242 }
2243
2244 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2245 {
2246         struct buffer_head *bh;
2247         int error;
2248
2249         if (!EXT4_I(inode)->i_file_acl)
2250                 return NULL;
2251         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2252         if (IS_ERR(bh))
2253                 return bh;
2254         error = ext4_xattr_check_block(inode, bh);
2255         if (error) {
2256                 brelse(bh);
2257                 return ERR_PTR(error);
2258         }
2259         return bh;
2260 }
2261
2262 /*
2263  * ext4_xattr_set_handle()
2264  *
2265  * Create, replace or remove an extended attribute for this inode.  Value
2266  * is NULL to remove an existing extended attribute, and non-NULL to
2267  * either replace an existing extended attribute, or create a new extended
2268  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2269  * specify that an extended attribute must exist and must not exist
2270  * previous to the call, respectively.
2271  *
2272  * Returns 0, or a negative error number on failure.
2273  */
2274 int
2275 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2276                       const char *name, const void *value, size_t value_len,
2277                       int flags)
2278 {
2279         struct ext4_xattr_info i = {
2280                 .name_index = name_index,
2281                 .name = name,
2282                 .value = value,
2283                 .value_len = value_len,
2284                 .in_inode = 0,
2285         };
2286         struct ext4_xattr_ibody_find is = {
2287                 .s = { .not_found = -ENODATA, },
2288         };
2289         struct ext4_xattr_block_find bs = {
2290                 .s = { .not_found = -ENODATA, },
2291         };
2292         int no_expand;
2293         int error;
2294
2295         if (!name)
2296                 return -EINVAL;
2297         if (strlen(name) > 255)
2298                 return -ERANGE;
2299
2300         ext4_write_lock_xattr(inode, &no_expand);
2301
2302         /* Check journal credits under write lock. */
2303         if (ext4_handle_valid(handle)) {
2304                 struct buffer_head *bh;
2305                 int credits;
2306
2307                 bh = ext4_xattr_get_block(inode);
2308                 if (IS_ERR(bh)) {
2309                         error = PTR_ERR(bh);
2310                         goto cleanup;
2311                 }
2312
2313                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2314                                                    value_len,
2315                                                    flags & XATTR_CREATE);
2316                 brelse(bh);
2317
2318                 if (jbd2_handle_buffer_credits(handle) < credits) {
2319                         error = -ENOSPC;
2320                         goto cleanup;
2321                 }
2322         }
2323
2324         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2325         if (error)
2326                 goto cleanup;
2327
2328         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2329                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2330                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2331                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2332         }
2333
2334         error = ext4_xattr_ibody_find(inode, &i, &is);
2335         if (error)
2336                 goto cleanup;
2337         if (is.s.not_found)
2338                 error = ext4_xattr_block_find(inode, &i, &bs);
2339         if (error)
2340                 goto cleanup;
2341         if (is.s.not_found && bs.s.not_found) {
2342                 error = -ENODATA;
2343                 if (flags & XATTR_REPLACE)
2344                         goto cleanup;
2345                 error = 0;
2346                 if (!value)
2347                         goto cleanup;
2348         } else {
2349                 error = -EEXIST;
2350                 if (flags & XATTR_CREATE)
2351                         goto cleanup;
2352         }
2353
2354         if (!value) {
2355                 if (!is.s.not_found)
2356                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2357                 else if (!bs.s.not_found)
2358                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2359         } else {
2360                 error = 0;
2361                 /* Xattr value did not change? Save us some work and bail out */
2362                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2363                         goto cleanup;
2364                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2365                         goto cleanup;
2366
2367                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2368                     (EXT4_XATTR_SIZE(i.value_len) >
2369                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2370                         i.in_inode = 1;
2371 retry_inode:
2372                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2373                 if (!error && !bs.s.not_found) {
2374                         i.value = NULL;
2375                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2376                 } else if (error == -ENOSPC) {
2377                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2378                                 brelse(bs.bh);
2379                                 bs.bh = NULL;
2380                                 error = ext4_xattr_block_find(inode, &i, &bs);
2381                                 if (error)
2382                                         goto cleanup;
2383                         }
2384                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2385                         if (!error && !is.s.not_found) {
2386                                 i.value = NULL;
2387                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2388                                                              &is);
2389                         } else if (error == -ENOSPC) {
2390                                 /*
2391                                  * Xattr does not fit in the block, store at
2392                                  * external inode if possible.
2393                                  */
2394                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2395                                     !i.in_inode) {
2396                                         i.in_inode = 1;
2397                                         goto retry_inode;
2398                                 }
2399                         }
2400                 }
2401         }
2402         if (!error) {
2403                 ext4_xattr_update_super_block(handle, inode->i_sb);
2404                 inode->i_ctime = current_time(inode);
2405                 if (!value)
2406                         no_expand = 0;
2407                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2408                 /*
2409                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2410                  * error != 0.
2411                  */
2412                 is.iloc.bh = NULL;
2413                 if (IS_SYNC(inode))
2414                         ext4_handle_sync(handle);
2415         }
2416
2417 cleanup:
2418         brelse(is.iloc.bh);
2419         brelse(bs.bh);
2420         ext4_write_unlock_xattr(inode, &no_expand);
2421         return error;
2422 }
2423
2424 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2425                            bool is_create, int *credits)
2426 {
2427         struct buffer_head *bh;
2428         int err;
2429
2430         *credits = 0;
2431
2432         if (!EXT4_SB(inode->i_sb)->s_journal)
2433                 return 0;
2434
2435         down_read(&EXT4_I(inode)->xattr_sem);
2436
2437         bh = ext4_xattr_get_block(inode);
2438         if (IS_ERR(bh)) {
2439                 err = PTR_ERR(bh);
2440         } else {
2441                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2442                                                     value_len, is_create);
2443                 brelse(bh);
2444                 err = 0;
2445         }
2446
2447         up_read(&EXT4_I(inode)->xattr_sem);
2448         return err;
2449 }
2450
2451 /*
2452  * ext4_xattr_set()
2453  *
2454  * Like ext4_xattr_set_handle, but start from an inode. This extended
2455  * attribute modification is a filesystem transaction by itself.
2456  *
2457  * Returns 0, or a negative error number on failure.
2458  */
2459 int
2460 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2461                const void *value, size_t value_len, int flags)
2462 {
2463         handle_t *handle;
2464         struct super_block *sb = inode->i_sb;
2465         int error, retries = 0;
2466         int credits;
2467
2468         error = dquot_initialize(inode);
2469         if (error)
2470                 return error;
2471
2472 retry:
2473         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2474                                        &credits);
2475         if (error)
2476                 return error;
2477
2478         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2479         if (IS_ERR(handle)) {
2480                 error = PTR_ERR(handle);
2481         } else {
2482                 int error2;
2483
2484                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2485                                               value, value_len, flags);
2486                 error2 = ext4_journal_stop(handle);
2487                 if (error == -ENOSPC &&
2488                     ext4_should_retry_alloc(sb, &retries))
2489                         goto retry;
2490                 if (error == 0)
2491                         error = error2;
2492         }
2493
2494         return error;
2495 }
2496
2497 /*
2498  * Shift the EA entries in the inode to create space for the increased
2499  * i_extra_isize.
2500  */
2501 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2502                                      int value_offs_shift, void *to,
2503                                      void *from, size_t n)
2504 {
2505         struct ext4_xattr_entry *last = entry;
2506         int new_offs;
2507
2508         /* We always shift xattr headers further thus offsets get lower */
2509         BUG_ON(value_offs_shift > 0);
2510
2511         /* Adjust the value offsets of the entries */
2512         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2513                 if (!last->e_value_inum && last->e_value_size) {
2514                         new_offs = le16_to_cpu(last->e_value_offs) +
2515                                                         value_offs_shift;
2516                         last->e_value_offs = cpu_to_le16(new_offs);
2517                 }
2518         }
2519         /* Shift the entries by n bytes */
2520         memmove(to, from, n);
2521 }
2522
2523 /*
2524  * Move xattr pointed to by 'entry' from inode into external xattr block
2525  */
2526 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2527                                     struct ext4_inode *raw_inode,
2528                                     struct ext4_xattr_entry *entry)
2529 {
2530         struct ext4_xattr_ibody_find *is = NULL;
2531         struct ext4_xattr_block_find *bs = NULL;
2532         char *buffer = NULL, *b_entry_name = NULL;
2533         size_t value_size = le32_to_cpu(entry->e_value_size);
2534         struct ext4_xattr_info i = {
2535                 .value = NULL,
2536                 .value_len = 0,
2537                 .name_index = entry->e_name_index,
2538                 .in_inode = !!entry->e_value_inum,
2539         };
2540         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2541         int error;
2542
2543         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2544         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2545         buffer = kmalloc(value_size, GFP_NOFS);
2546         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2547         if (!is || !bs || !buffer || !b_entry_name) {
2548                 error = -ENOMEM;
2549                 goto out;
2550         }
2551
2552         is->s.not_found = -ENODATA;
2553         bs->s.not_found = -ENODATA;
2554         is->iloc.bh = NULL;
2555         bs->bh = NULL;
2556
2557         /* Save the entry name and the entry value */
2558         if (entry->e_value_inum) {
2559                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2560                 if (error)
2561                         goto out;
2562         } else {
2563                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2564                 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2565         }
2566
2567         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2568         b_entry_name[entry->e_name_len] = '\0';
2569         i.name = b_entry_name;
2570
2571         error = ext4_get_inode_loc(inode, &is->iloc);
2572         if (error)
2573                 goto out;
2574
2575         error = ext4_xattr_ibody_find(inode, &i, is);
2576         if (error)
2577                 goto out;
2578
2579         /* Remove the chosen entry from the inode */
2580         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2581         if (error)
2582                 goto out;
2583
2584         i.value = buffer;
2585         i.value_len = value_size;
2586         error = ext4_xattr_block_find(inode, &i, bs);
2587         if (error)
2588                 goto out;
2589
2590         /* Add entry which was removed from the inode into the block */
2591         error = ext4_xattr_block_set(handle, inode, &i, bs);
2592         if (error)
2593                 goto out;
2594         error = 0;
2595 out:
2596         kfree(b_entry_name);
2597         kfree(buffer);
2598         if (is)
2599                 brelse(is->iloc.bh);
2600         if (bs)
2601                 brelse(bs->bh);
2602         kfree(is);
2603         kfree(bs);
2604
2605         return error;
2606 }
2607
2608 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2609                                        struct ext4_inode *raw_inode,
2610                                        int isize_diff, size_t ifree,
2611                                        size_t bfree, int *total_ino)
2612 {
2613         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2614         struct ext4_xattr_entry *small_entry;
2615         struct ext4_xattr_entry *entry;
2616         struct ext4_xattr_entry *last;
2617         unsigned int entry_size;        /* EA entry size */
2618         unsigned int total_size;        /* EA entry size + value size */
2619         unsigned int min_total_size;
2620         int error;
2621
2622         while (isize_diff > ifree) {
2623                 entry = NULL;
2624                 small_entry = NULL;
2625                 min_total_size = ~0U;
2626                 last = IFIRST(header);
2627                 /* Find the entry best suited to be pushed into EA block */
2628                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2629                         /* never move system.data out of the inode */
2630                         if ((last->e_name_len == 4) &&
2631                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2632                             !memcmp(last->e_name, "data", 4))
2633                                 continue;
2634                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2635                         if (!last->e_value_inum)
2636                                 total_size += EXT4_XATTR_SIZE(
2637                                                le32_to_cpu(last->e_value_size));
2638                         if (total_size <= bfree &&
2639                             total_size < min_total_size) {
2640                                 if (total_size + ifree < isize_diff) {
2641                                         small_entry = last;
2642                                 } else {
2643                                         entry = last;
2644                                         min_total_size = total_size;
2645                                 }
2646                         }
2647                 }
2648
2649                 if (entry == NULL) {
2650                         if (small_entry == NULL)
2651                                 return -ENOSPC;
2652                         entry = small_entry;
2653                 }
2654
2655                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2656                 total_size = entry_size;
2657                 if (!entry->e_value_inum)
2658                         total_size += EXT4_XATTR_SIZE(
2659                                               le32_to_cpu(entry->e_value_size));
2660                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2661                                                  entry);
2662                 if (error)
2663                         return error;
2664
2665                 *total_ino -= entry_size;
2666                 ifree += total_size;
2667                 bfree -= total_size;
2668         }
2669
2670         return 0;
2671 }
2672
2673 /*
2674  * Expand an inode by new_extra_isize bytes when EAs are present.
2675  * Returns 0 on success or negative error number on failure.
2676  */
2677 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2678                                struct ext4_inode *raw_inode, handle_t *handle)
2679 {
2680         struct ext4_xattr_ibody_header *header;
2681         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2682         static unsigned int mnt_count;
2683         size_t min_offs;
2684         size_t ifree, bfree;
2685         int total_ino;
2686         void *base, *end;
2687         int error = 0, tried_min_extra_isize = 0;
2688         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2689         int isize_diff; /* How much do we need to grow i_extra_isize */
2690
2691 retry:
2692         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2693         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2694                 return 0;
2695
2696         header = IHDR(inode, raw_inode);
2697
2698         /*
2699          * Check if enough free space is available in the inode to shift the
2700          * entries ahead by new_extra_isize.
2701          */
2702
2703         base = IFIRST(header);
2704         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2705         min_offs = end - base;
2706         total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2707
2708         error = xattr_check_inode(inode, header, end);
2709         if (error)
2710                 goto cleanup;
2711
2712         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2713         if (ifree >= isize_diff)
2714                 goto shift;
2715
2716         /*
2717          * Enough free space isn't available in the inode, check if
2718          * EA block can hold new_extra_isize bytes.
2719          */
2720         if (EXT4_I(inode)->i_file_acl) {
2721                 struct buffer_head *bh;
2722
2723                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2724                 if (IS_ERR(bh)) {
2725                         error = PTR_ERR(bh);
2726                         goto cleanup;
2727                 }
2728                 error = ext4_xattr_check_block(inode, bh);
2729                 if (error) {
2730                         brelse(bh);
2731                         goto cleanup;
2732                 }
2733                 base = BHDR(bh);
2734                 end = bh->b_data + bh->b_size;
2735                 min_offs = end - base;
2736                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2737                                               NULL);
2738                 brelse(bh);
2739                 if (bfree + ifree < isize_diff) {
2740                         if (!tried_min_extra_isize && s_min_extra_isize) {
2741                                 tried_min_extra_isize++;
2742                                 new_extra_isize = s_min_extra_isize;
2743                                 goto retry;
2744                         }
2745                         error = -ENOSPC;
2746                         goto cleanup;
2747                 }
2748         } else {
2749                 bfree = inode->i_sb->s_blocksize;
2750         }
2751
2752         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2753                                             isize_diff, ifree, bfree,
2754                                             &total_ino);
2755         if (error) {
2756                 if (error == -ENOSPC && !tried_min_extra_isize &&
2757                     s_min_extra_isize) {
2758                         tried_min_extra_isize++;
2759                         new_extra_isize = s_min_extra_isize;
2760                         goto retry;
2761                 }
2762                 goto cleanup;
2763         }
2764 shift:
2765         /* Adjust the offsets and shift the remaining entries ahead */
2766         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2767                         - new_extra_isize, (void *)raw_inode +
2768                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2769                         (void *)header, total_ino);
2770         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2771
2772 cleanup:
2773         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2774                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2775                              inode->i_ino);
2776                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2777         }
2778         return error;
2779 }
2780
2781 #define EIA_INCR 16 /* must be 2^n */
2782 #define EIA_MASK (EIA_INCR - 1)
2783
2784 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2785  * If @ea_inode_array is new or full it will be grown and the old
2786  * contents copied over.
2787  */
2788 static int
2789 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2790                         struct inode *inode)
2791 {
2792         if (*ea_inode_array == NULL) {
2793                 /*
2794                  * Start with 15 inodes, so it fits into a power-of-two size.
2795                  * If *ea_inode_array is NULL, this is essentially offsetof()
2796                  */
2797                 (*ea_inode_array) =
2798                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2799                                          inodes[EIA_MASK]),
2800                                 GFP_NOFS);
2801                 if (*ea_inode_array == NULL)
2802                         return -ENOMEM;
2803                 (*ea_inode_array)->count = 0;
2804         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2805                 /* expand the array once all 15 + n * 16 slots are full */
2806                 struct ext4_xattr_inode_array *new_array = NULL;
2807                 int count = (*ea_inode_array)->count;
2808
2809                 /* if new_array is NULL, this is essentially offsetof() */
2810                 new_array = kmalloc(
2811                                 offsetof(struct ext4_xattr_inode_array,
2812                                          inodes[count + EIA_INCR]),
2813                                 GFP_NOFS);
2814                 if (new_array == NULL)
2815                         return -ENOMEM;
2816                 memcpy(new_array, *ea_inode_array,
2817                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2818                 kfree(*ea_inode_array);
2819                 *ea_inode_array = new_array;
2820         }
2821         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2822         return 0;
2823 }
2824
2825 /*
2826  * ext4_xattr_delete_inode()
2827  *
2828  * Free extended attribute resources associated with this inode. Traverse
2829  * all entries and decrement reference on any xattr inodes associated with this
2830  * inode. This is called immediately before an inode is freed. We have exclusive
2831  * access to the inode. If an orphan inode is deleted it will also release its
2832  * references on xattr block and xattr inodes.
2833  */
2834 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2835                             struct ext4_xattr_inode_array **ea_inode_array,
2836                             int extra_credits)
2837 {
2838         struct buffer_head *bh = NULL;
2839         struct ext4_xattr_ibody_header *header;
2840         struct ext4_iloc iloc = { .bh = NULL };
2841         struct ext4_xattr_entry *entry;
2842         struct inode *ea_inode;
2843         int error;
2844
2845         error = ext4_journal_ensure_credits(handle, extra_credits,
2846                         ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2847         if (error < 0) {
2848                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2849                 goto cleanup;
2850         }
2851
2852         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2853             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2854
2855                 error = ext4_get_inode_loc(inode, &iloc);
2856                 if (error) {
2857                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2858                         goto cleanup;
2859                 }
2860
2861                 error = ext4_journal_get_write_access(handle, iloc.bh);
2862                 if (error) {
2863                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2864                                          error);
2865                         goto cleanup;
2866                 }
2867
2868                 header = IHDR(inode, ext4_raw_inode(&iloc));
2869                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2870                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2871                                                      IFIRST(header),
2872                                                      false /* block_csum */,
2873                                                      ea_inode_array,
2874                                                      extra_credits,
2875                                                      false /* skip_quota */);
2876         }
2877
2878         if (EXT4_I(inode)->i_file_acl) {
2879                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2880                 if (IS_ERR(bh)) {
2881                         error = PTR_ERR(bh);
2882                         if (error == -EIO) {
2883                                 EXT4_ERROR_INODE_ERR(inode, EIO,
2884                                                      "block %llu read error",
2885                                                      EXT4_I(inode)->i_file_acl);
2886                         }
2887                         bh = NULL;
2888                         goto cleanup;
2889                 }
2890                 error = ext4_xattr_check_block(inode, bh);
2891                 if (error)
2892                         goto cleanup;
2893
2894                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2895                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2896                              entry = EXT4_XATTR_NEXT(entry)) {
2897                                 if (!entry->e_value_inum)
2898                                         continue;
2899                                 error = ext4_xattr_inode_iget(inode,
2900                                               le32_to_cpu(entry->e_value_inum),
2901                                               le32_to_cpu(entry->e_hash),
2902                                               &ea_inode);
2903                                 if (error)
2904                                         continue;
2905                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2906                                               le32_to_cpu(entry->e_value_size));
2907                                 iput(ea_inode);
2908                         }
2909
2910                 }
2911
2912                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2913                                          extra_credits);
2914                 /*
2915                  * Update i_file_acl value in the same transaction that releases
2916                  * block.
2917                  */
2918                 EXT4_I(inode)->i_file_acl = 0;
2919                 error = ext4_mark_inode_dirty(handle, inode);
2920                 if (error) {
2921                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2922                                          error);
2923                         goto cleanup;
2924                 }
2925         }
2926         error = 0;
2927 cleanup:
2928         brelse(iloc.bh);
2929         brelse(bh);
2930         return error;
2931 }
2932
2933 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2934 {
2935         int idx;
2936
2937         if (ea_inode_array == NULL)
2938                 return;
2939
2940         for (idx = 0; idx < ea_inode_array->count; ++idx)
2941                 iput(ea_inode_array->inodes[idx]);
2942         kfree(ea_inode_array);
2943 }
2944
2945 /*
2946  * ext4_xattr_block_cache_insert()
2947  *
2948  * Create a new entry in the extended attribute block cache, and insert
2949  * it unless such an entry is already in the cache.
2950  *
2951  * Returns 0, or a negative error number on failure.
2952  */
2953 static void
2954 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2955                               struct buffer_head *bh)
2956 {
2957         struct ext4_xattr_header *header = BHDR(bh);
2958         __u32 hash = le32_to_cpu(header->h_hash);
2959         int reusable = le32_to_cpu(header->h_refcount) <
2960                        EXT4_XATTR_REFCOUNT_MAX;
2961         int error;
2962
2963         if (!ea_block_cache)
2964                 return;
2965         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2966                                       bh->b_blocknr, reusable);
2967         if (error) {
2968                 if (error == -EBUSY)
2969                         ea_bdebug(bh, "already in cache");
2970         } else
2971                 ea_bdebug(bh, "inserting [%x]", (int)hash);
2972 }
2973
2974 /*
2975  * ext4_xattr_cmp()
2976  *
2977  * Compare two extended attribute blocks for equality.
2978  *
2979  * Returns 0 if the blocks are equal, 1 if they differ, and
2980  * a negative error number on errors.
2981  */
2982 static int
2983 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2984                struct ext4_xattr_header *header2)
2985 {
2986         struct ext4_xattr_entry *entry1, *entry2;
2987
2988         entry1 = ENTRY(header1+1);
2989         entry2 = ENTRY(header2+1);
2990         while (!IS_LAST_ENTRY(entry1)) {
2991                 if (IS_LAST_ENTRY(entry2))
2992                         return 1;
2993                 if (entry1->e_hash != entry2->e_hash ||
2994                     entry1->e_name_index != entry2->e_name_index ||
2995                     entry1->e_name_len != entry2->e_name_len ||
2996                     entry1->e_value_size != entry2->e_value_size ||
2997                     entry1->e_value_inum != entry2->e_value_inum ||
2998                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
2999                         return 1;
3000                 if (!entry1->e_value_inum &&
3001                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3002                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3003                            le32_to_cpu(entry1->e_value_size)))
3004                         return 1;
3005
3006                 entry1 = EXT4_XATTR_NEXT(entry1);
3007                 entry2 = EXT4_XATTR_NEXT(entry2);
3008         }
3009         if (!IS_LAST_ENTRY(entry2))
3010                 return 1;
3011         return 0;
3012 }
3013
3014 /*
3015  * ext4_xattr_block_cache_find()
3016  *
3017  * Find an identical extended attribute block.
3018  *
3019  * Returns a pointer to the block found, or NULL if such a block was
3020  * not found or an error occurred.
3021  */
3022 static struct buffer_head *
3023 ext4_xattr_block_cache_find(struct inode *inode,
3024                             struct ext4_xattr_header *header,
3025                             struct mb_cache_entry **pce)
3026 {
3027         __u32 hash = le32_to_cpu(header->h_hash);
3028         struct mb_cache_entry *ce;
3029         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3030
3031         if (!ea_block_cache)
3032                 return NULL;
3033         if (!header->h_hash)
3034                 return NULL;  /* never share */
3035         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3036         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3037         while (ce) {
3038                 struct buffer_head *bh;
3039
3040                 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3041                 if (IS_ERR(bh)) {
3042                         if (PTR_ERR(bh) == -ENOMEM)
3043                                 return NULL;
3044                         bh = NULL;
3045                         EXT4_ERROR_INODE(inode, "block %lu read error",
3046                                          (unsigned long)ce->e_value);
3047                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3048                         *pce = ce;
3049                         return bh;
3050                 }
3051                 brelse(bh);
3052                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3053         }
3054         return NULL;
3055 }
3056
3057 #define NAME_HASH_SHIFT 5
3058 #define VALUE_HASH_SHIFT 16
3059
3060 /*
3061  * ext4_xattr_hash_entry()
3062  *
3063  * Compute the hash of an extended attribute.
3064  */
3065 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3066                                     size_t value_count)
3067 {
3068         __u32 hash = 0;
3069
3070         while (name_len--) {
3071                 hash = (hash << NAME_HASH_SHIFT) ^
3072                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3073                        *name++;
3074         }
3075         while (value_count--) {
3076                 hash = (hash << VALUE_HASH_SHIFT) ^
3077                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3078                        le32_to_cpu(*value++);
3079         }
3080         return cpu_to_le32(hash);
3081 }
3082
3083 #undef NAME_HASH_SHIFT
3084 #undef VALUE_HASH_SHIFT
3085
3086 #define BLOCK_HASH_SHIFT 16
3087
3088 /*
3089  * ext4_xattr_rehash()
3090  *
3091  * Re-compute the extended attribute hash value after an entry has changed.
3092  */
3093 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3094 {
3095         struct ext4_xattr_entry *here;
3096         __u32 hash = 0;
3097
3098         here = ENTRY(header+1);
3099         while (!IS_LAST_ENTRY(here)) {
3100                 if (!here->e_hash) {
3101                         /* Block is not shared if an entry's hash value == 0 */
3102                         hash = 0;
3103                         break;
3104                 }
3105                 hash = (hash << BLOCK_HASH_SHIFT) ^
3106                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3107                        le32_to_cpu(here->e_hash);
3108                 here = EXT4_XATTR_NEXT(here);
3109         }
3110         header->h_hash = cpu_to_le32(hash);
3111 }
3112
3113 #undef BLOCK_HASH_SHIFT
3114
3115 #define HASH_BUCKET_BITS        10
3116
3117 struct mb_cache *
3118 ext4_xattr_create_cache(void)
3119 {
3120         return mb_cache_create(HASH_BUCKET_BITS);
3121 }
3122
3123 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3124 {
3125         if (cache)
3126                 mb_cache_destroy(cache);
3127 }
3128