Merge tag 'sti-soc-for-v4.21-round1' of git://git.kernel.org/pub/scm/linux/kernel...
[sfrench/cifs-2.6.git] / fs / ext4 / xattr.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/ext4/xattr.c
4  *
5  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
6  *
7  * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8  * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
9  * Extended attributes for symlinks and special files added per
10  *  suggestion of Luka Renko <luka.renko@hermes.si>.
11  * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
12  *  Red Hat Inc.
13  * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
14  *  and Andreas Gruenbacher <agruen@suse.de>.
15  */
16
17 /*
18  * Extended attributes are stored directly in inodes (on file systems with
19  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
20  * field contains the block number if an inode uses an additional block. All
21  * attributes must fit in the inode and one additional block. Blocks that
22  * contain the identical set of attributes may be shared among several inodes.
23  * Identical blocks are detected by keeping a cache of blocks that have
24  * recently been accessed.
25  *
26  * The attributes in inodes and on blocks have a different header; the entries
27  * are stored in the same format:
28  *
29  *   +------------------+
30  *   | header           |
31  *   | entry 1          | |
32  *   | entry 2          | | growing downwards
33  *   | entry 3          | v
34  *   | four null bytes  |
35  *   | . . .            |
36  *   | value 1          | ^
37  *   | value 3          | | growing upwards
38  *   | value 2          | |
39  *   +------------------+
40  *
41  * The header is followed by multiple entry descriptors. In disk blocks, the
42  * entry descriptors are kept sorted. In inodes, they are unsorted. The
43  * attribute values are aligned to the end of the block in no specific order.
44  *
45  * Locking strategy
46  * ----------------
47  * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
48  * EA blocks are only changed if they are exclusive to an inode, so
49  * holding xattr_sem also means that nothing but the EA block's reference
50  * count can change. Multiple writers to the same block are synchronized
51  * by the buffer lock.
52  */
53
54 #include <linux/init.h>
55 #include <linux/fs.h>
56 #include <linux/slab.h>
57 #include <linux/mbcache.h>
58 #include <linux/quotaops.h>
59 #include <linux/iversion.h>
60 #include "ext4_jbd2.h"
61 #include "ext4.h"
62 #include "xattr.h"
63 #include "acl.h"
64
65 #ifdef EXT4_XATTR_DEBUG
66 # define ea_idebug(inode, fmt, ...)                                     \
67         printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
68                inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
69 # define ea_bdebug(bh, fmt, ...)                                        \
70         printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
71                bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
72 #else
73 # define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
74 # define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
75 #endif
76
77 static void ext4_xattr_block_cache_insert(struct mb_cache *,
78                                           struct buffer_head *);
79 static struct buffer_head *
80 ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
81                             struct mb_cache_entry **);
82 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
83                                     size_t value_count);
84 static void ext4_xattr_rehash(struct ext4_xattr_header *);
85
86 static const struct xattr_handler * const ext4_xattr_handler_map[] = {
87         [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
88 #ifdef CONFIG_EXT4_FS_POSIX_ACL
89         [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
90         [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
91 #endif
92         [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
93 #ifdef CONFIG_EXT4_FS_SECURITY
94         [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
95 #endif
96 };
97
98 const struct xattr_handler *ext4_xattr_handlers[] = {
99         &ext4_xattr_user_handler,
100         &ext4_xattr_trusted_handler,
101 #ifdef CONFIG_EXT4_FS_POSIX_ACL
102         &posix_acl_access_xattr_handler,
103         &posix_acl_default_xattr_handler,
104 #endif
105 #ifdef CONFIG_EXT4_FS_SECURITY
106         &ext4_xattr_security_handler,
107 #endif
108         NULL
109 };
110
111 #define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
112                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
113
114 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
115                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
116
117 static int
118 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
119                         struct inode *inode);
120
121 #ifdef CONFIG_LOCKDEP
122 void ext4_xattr_inode_set_class(struct inode *ea_inode)
123 {
124         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
125 }
126 #endif
127
128 static __le32 ext4_xattr_block_csum(struct inode *inode,
129                                     sector_t block_nr,
130                                     struct ext4_xattr_header *hdr)
131 {
132         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
133         __u32 csum;
134         __le64 dsk_block_nr = cpu_to_le64(block_nr);
135         __u32 dummy_csum = 0;
136         int offset = offsetof(struct ext4_xattr_header, h_checksum);
137
138         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
139                            sizeof(dsk_block_nr));
140         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
141         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
142         offset += sizeof(dummy_csum);
143         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
144                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
145
146         return cpu_to_le32(csum);
147 }
148
149 static int ext4_xattr_block_csum_verify(struct inode *inode,
150                                         struct buffer_head *bh)
151 {
152         struct ext4_xattr_header *hdr = BHDR(bh);
153         int ret = 1;
154
155         if (ext4_has_metadata_csum(inode->i_sb)) {
156                 lock_buffer(bh);
157                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
158                                                         bh->b_blocknr, hdr));
159                 unlock_buffer(bh);
160         }
161         return ret;
162 }
163
164 static void ext4_xattr_block_csum_set(struct inode *inode,
165                                       struct buffer_head *bh)
166 {
167         if (ext4_has_metadata_csum(inode->i_sb))
168                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
169                                                 bh->b_blocknr, BHDR(bh));
170 }
171
172 static inline const struct xattr_handler *
173 ext4_xattr_handler(int name_index)
174 {
175         const struct xattr_handler *handler = NULL;
176
177         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
178                 handler = ext4_xattr_handler_map[name_index];
179         return handler;
180 }
181
182 static int
183 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
184                          void *value_start)
185 {
186         struct ext4_xattr_entry *e = entry;
187
188         /* Find the end of the names list */
189         while (!IS_LAST_ENTRY(e)) {
190                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
191                 if ((void *)next >= end)
192                         return -EFSCORRUPTED;
193                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
194                         return -EFSCORRUPTED;
195                 e = next;
196         }
197
198         /* Check the values */
199         while (!IS_LAST_ENTRY(entry)) {
200                 u32 size = le32_to_cpu(entry->e_value_size);
201
202                 if (size > EXT4_XATTR_SIZE_MAX)
203                         return -EFSCORRUPTED;
204
205                 if (size != 0 && entry->e_value_inum == 0) {
206                         u16 offs = le16_to_cpu(entry->e_value_offs);
207                         void *value;
208
209                         /*
210                          * The value cannot overlap the names, and the value
211                          * with padding cannot extend beyond 'end'.  Check both
212                          * the padded and unpadded sizes, since the size may
213                          * overflow to 0 when adding padding.
214                          */
215                         if (offs > end - value_start)
216                                 return -EFSCORRUPTED;
217                         value = value_start + offs;
218                         if (value < (void *)e + sizeof(u32) ||
219                             size > end - value ||
220                             EXT4_XATTR_SIZE(size) > end - value)
221                                 return -EFSCORRUPTED;
222                 }
223                 entry = EXT4_XATTR_NEXT(entry);
224         }
225
226         return 0;
227 }
228
229 static inline int
230 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
231                          const char *function, unsigned int line)
232 {
233         int error = -EFSCORRUPTED;
234
235         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
236             BHDR(bh)->h_blocks != cpu_to_le32(1))
237                 goto errout;
238         if (buffer_verified(bh))
239                 return 0;
240
241         error = -EFSBADCRC;
242         if (!ext4_xattr_block_csum_verify(inode, bh))
243                 goto errout;
244         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
245                                          bh->b_data);
246 errout:
247         if (error)
248                 __ext4_error_inode(inode, function, line, 0,
249                                    "corrupted xattr block %llu",
250                                    (unsigned long long) bh->b_blocknr);
251         else
252                 set_buffer_verified(bh);
253         return error;
254 }
255
256 #define ext4_xattr_check_block(inode, bh) \
257         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
258
259
260 static int
261 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
262                          void *end, const char *function, unsigned int line)
263 {
264         int error = -EFSCORRUPTED;
265
266         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
267             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
268                 goto errout;
269         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
270 errout:
271         if (error)
272                 __ext4_error_inode(inode, function, line, 0,
273                                    "corrupted in-inode xattr");
274         return error;
275 }
276
277 #define xattr_check_inode(inode, header, end) \
278         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
279
280 static int
281 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
282                  void *end, int name_index, const char *name, int sorted)
283 {
284         struct ext4_xattr_entry *entry, *next;
285         size_t name_len;
286         int cmp = 1;
287
288         if (name == NULL)
289                 return -EINVAL;
290         name_len = strlen(name);
291         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
292                 next = EXT4_XATTR_NEXT(entry);
293                 if ((void *) next >= end) {
294                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
295                         return -EFSCORRUPTED;
296                 }
297                 cmp = name_index - entry->e_name_index;
298                 if (!cmp)
299                         cmp = name_len - entry->e_name_len;
300                 if (!cmp)
301                         cmp = memcmp(name, entry->e_name, name_len);
302                 if (cmp <= 0 && (sorted || cmp == 0))
303                         break;
304         }
305         *pentry = entry;
306         return cmp ? -ENODATA : 0;
307 }
308
309 static u32
310 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
311 {
312         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
313 }
314
315 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
316 {
317         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
318                 (u32) inode_peek_iversion_raw(ea_inode);
319 }
320
321 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
322 {
323         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
324         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
325 }
326
327 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
328 {
329         return (u32)ea_inode->i_atime.tv_sec;
330 }
331
332 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
333 {
334         ea_inode->i_atime.tv_sec = hash;
335 }
336
337 /*
338  * Read the EA value from an inode.
339  */
340 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
341 {
342         int blocksize = 1 << ea_inode->i_blkbits;
343         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
344         int tail_size = (size % blocksize) ?: blocksize;
345         struct buffer_head *bhs_inline[8];
346         struct buffer_head **bhs = bhs_inline;
347         int i, ret;
348
349         if (bh_count > ARRAY_SIZE(bhs_inline)) {
350                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
351                 if (!bhs)
352                         return -ENOMEM;
353         }
354
355         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
356                                true /* wait */, bhs);
357         if (ret)
358                 goto free_bhs;
359
360         for (i = 0; i < bh_count; i++) {
361                 /* There shouldn't be any holes in ea_inode. */
362                 if (!bhs[i]) {
363                         ret = -EFSCORRUPTED;
364                         goto put_bhs;
365                 }
366                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
367                        i < bh_count - 1 ? blocksize : tail_size);
368         }
369         ret = 0;
370 put_bhs:
371         for (i = 0; i < bh_count; i++)
372                 brelse(bhs[i]);
373 free_bhs:
374         if (bhs != bhs_inline)
375                 kfree(bhs);
376         return ret;
377 }
378
379 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
380
381 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
382                                  u32 ea_inode_hash, struct inode **ea_inode)
383 {
384         struct inode *inode;
385         int err;
386
387         inode = ext4_iget(parent->i_sb, ea_ino);
388         if (IS_ERR(inode)) {
389                 err = PTR_ERR(inode);
390                 ext4_error(parent->i_sb,
391                            "error while reading EA inode %lu err=%d", ea_ino,
392                            err);
393                 return err;
394         }
395
396         if (is_bad_inode(inode)) {
397                 ext4_error(parent->i_sb,
398                            "error while reading EA inode %lu is_bad_inode",
399                            ea_ino);
400                 err = -EIO;
401                 goto error;
402         }
403
404         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
405                 ext4_error(parent->i_sb,
406                            "EA inode %lu does not have EXT4_EA_INODE_FL flag",
407                             ea_ino);
408                 err = -EINVAL;
409                 goto error;
410         }
411
412         ext4_xattr_inode_set_class(inode);
413
414         /*
415          * Check whether this is an old Lustre-style xattr inode. Lustre
416          * implementation does not have hash validation, rather it has a
417          * backpointer from ea_inode to the parent inode.
418          */
419         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
420             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
421             inode->i_generation == parent->i_generation) {
422                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
423                 ext4_xattr_inode_set_ref(inode, 1);
424         } else {
425                 inode_lock(inode);
426                 inode->i_flags |= S_NOQUOTA;
427                 inode_unlock(inode);
428         }
429
430         *ea_inode = inode;
431         return 0;
432 error:
433         iput(inode);
434         return err;
435 }
436
437 static int
438 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
439                                struct ext4_xattr_entry *entry, void *buffer,
440                                size_t size)
441 {
442         u32 hash;
443
444         /* Verify stored hash matches calculated hash. */
445         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
446         if (hash != ext4_xattr_inode_get_hash(ea_inode))
447                 return -EFSCORRUPTED;
448
449         if (entry) {
450                 __le32 e_hash, tmp_data;
451
452                 /* Verify entry hash. */
453                 tmp_data = cpu_to_le32(hash);
454                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
455                                                &tmp_data, 1);
456                 if (e_hash != entry->e_hash)
457                         return -EFSCORRUPTED;
458         }
459         return 0;
460 }
461
462 /*
463  * Read xattr value from the EA inode.
464  */
465 static int
466 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
467                      void *buffer, size_t size)
468 {
469         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
470         struct inode *ea_inode;
471         int err;
472
473         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
474                                     le32_to_cpu(entry->e_hash), &ea_inode);
475         if (err) {
476                 ea_inode = NULL;
477                 goto out;
478         }
479
480         if (i_size_read(ea_inode) != size) {
481                 ext4_warning_inode(ea_inode,
482                                    "ea_inode file size=%llu entry size=%zu",
483                                    i_size_read(ea_inode), size);
484                 err = -EFSCORRUPTED;
485                 goto out;
486         }
487
488         err = ext4_xattr_inode_read(ea_inode, buffer, size);
489         if (err)
490                 goto out;
491
492         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
493                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
494                                                      size);
495                 if (err) {
496                         ext4_warning_inode(ea_inode,
497                                            "EA inode hash validation failed");
498                         goto out;
499                 }
500
501                 if (ea_inode_cache)
502                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
503                                         ext4_xattr_inode_get_hash(ea_inode),
504                                         ea_inode->i_ino, true /* reusable */);
505         }
506 out:
507         iput(ea_inode);
508         return err;
509 }
510
511 static int
512 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
513                      void *buffer, size_t buffer_size)
514 {
515         struct buffer_head *bh = NULL;
516         struct ext4_xattr_entry *entry;
517         size_t size;
518         void *end;
519         int error;
520         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
521
522         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
523                   name_index, name, buffer, (long)buffer_size);
524
525         error = -ENODATA;
526         if (!EXT4_I(inode)->i_file_acl)
527                 goto cleanup;
528         ea_idebug(inode, "reading block %llu",
529                   (unsigned long long)EXT4_I(inode)->i_file_acl);
530         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
531         if (!bh)
532                 goto cleanup;
533         ea_bdebug(bh, "b_count=%d, refcount=%d",
534                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
535         error = ext4_xattr_check_block(inode, bh);
536         if (error)
537                 goto cleanup;
538         ext4_xattr_block_cache_insert(ea_block_cache, bh);
539         entry = BFIRST(bh);
540         end = bh->b_data + bh->b_size;
541         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
542         if (error)
543                 goto cleanup;
544         size = le32_to_cpu(entry->e_value_size);
545         error = -ERANGE;
546         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
547                 goto cleanup;
548         if (buffer) {
549                 if (size > buffer_size)
550                         goto cleanup;
551                 if (entry->e_value_inum) {
552                         error = ext4_xattr_inode_get(inode, entry, buffer,
553                                                      size);
554                         if (error)
555                                 goto cleanup;
556                 } else {
557                         u16 offset = le16_to_cpu(entry->e_value_offs);
558                         void *p = bh->b_data + offset;
559
560                         if (unlikely(p + size > end))
561                                 goto cleanup;
562                         memcpy(buffer, p, size);
563                 }
564         }
565         error = size;
566
567 cleanup:
568         brelse(bh);
569         return error;
570 }
571
572 int
573 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
574                      void *buffer, size_t buffer_size)
575 {
576         struct ext4_xattr_ibody_header *header;
577         struct ext4_xattr_entry *entry;
578         struct ext4_inode *raw_inode;
579         struct ext4_iloc iloc;
580         size_t size;
581         void *end;
582         int error;
583
584         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
585                 return -ENODATA;
586         error = ext4_get_inode_loc(inode, &iloc);
587         if (error)
588                 return error;
589         raw_inode = ext4_raw_inode(&iloc);
590         header = IHDR(inode, raw_inode);
591         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
592         error = xattr_check_inode(inode, header, end);
593         if (error)
594                 goto cleanup;
595         entry = IFIRST(header);
596         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
597         if (error)
598                 goto cleanup;
599         size = le32_to_cpu(entry->e_value_size);
600         error = -ERANGE;
601         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
602                 goto cleanup;
603         if (buffer) {
604                 if (size > buffer_size)
605                         goto cleanup;
606                 if (entry->e_value_inum) {
607                         error = ext4_xattr_inode_get(inode, entry, buffer,
608                                                      size);
609                         if (error)
610                                 goto cleanup;
611                 } else {
612                         u16 offset = le16_to_cpu(entry->e_value_offs);
613                         void *p = (void *)IFIRST(header) + offset;
614
615                         if (unlikely(p + size > end))
616                                 goto cleanup;
617                         memcpy(buffer, p, size);
618                 }
619         }
620         error = size;
621
622 cleanup:
623         brelse(iloc.bh);
624         return error;
625 }
626
627 /*
628  * ext4_xattr_get()
629  *
630  * Copy an extended attribute into the buffer
631  * provided, or compute the buffer size required.
632  * Buffer is NULL to compute the size of the buffer required.
633  *
634  * Returns a negative error number on failure, or the number of bytes
635  * used / required on success.
636  */
637 int
638 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
639                void *buffer, size_t buffer_size)
640 {
641         int error;
642
643         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
644                 return -EIO;
645
646         if (strlen(name) > 255)
647                 return -ERANGE;
648
649         down_read(&EXT4_I(inode)->xattr_sem);
650         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
651                                      buffer_size);
652         if (error == -ENODATA)
653                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
654                                              buffer_size);
655         up_read(&EXT4_I(inode)->xattr_sem);
656         return error;
657 }
658
659 static int
660 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
661                         char *buffer, size_t buffer_size)
662 {
663         size_t rest = buffer_size;
664
665         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
666                 const struct xattr_handler *handler =
667                         ext4_xattr_handler(entry->e_name_index);
668
669                 if (handler && (!handler->list || handler->list(dentry))) {
670                         const char *prefix = handler->prefix ?: handler->name;
671                         size_t prefix_len = strlen(prefix);
672                         size_t size = prefix_len + entry->e_name_len + 1;
673
674                         if (buffer) {
675                                 if (size > rest)
676                                         return -ERANGE;
677                                 memcpy(buffer, prefix, prefix_len);
678                                 buffer += prefix_len;
679                                 memcpy(buffer, entry->e_name, entry->e_name_len);
680                                 buffer += entry->e_name_len;
681                                 *buffer++ = 0;
682                         }
683                         rest -= size;
684                 }
685         }
686         return buffer_size - rest;  /* total size */
687 }
688
689 static int
690 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
691 {
692         struct inode *inode = d_inode(dentry);
693         struct buffer_head *bh = NULL;
694         int error;
695
696         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
697                   buffer, (long)buffer_size);
698
699         error = 0;
700         if (!EXT4_I(inode)->i_file_acl)
701                 goto cleanup;
702         ea_idebug(inode, "reading block %llu",
703                   (unsigned long long)EXT4_I(inode)->i_file_acl);
704         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
705         error = -EIO;
706         if (!bh)
707                 goto cleanup;
708         ea_bdebug(bh, "b_count=%d, refcount=%d",
709                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
710         error = ext4_xattr_check_block(inode, bh);
711         if (error)
712                 goto cleanup;
713         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
714         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
715
716 cleanup:
717         brelse(bh);
718
719         return error;
720 }
721
722 static int
723 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
724 {
725         struct inode *inode = d_inode(dentry);
726         struct ext4_xattr_ibody_header *header;
727         struct ext4_inode *raw_inode;
728         struct ext4_iloc iloc;
729         void *end;
730         int error;
731
732         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
733                 return 0;
734         error = ext4_get_inode_loc(inode, &iloc);
735         if (error)
736                 return error;
737         raw_inode = ext4_raw_inode(&iloc);
738         header = IHDR(inode, raw_inode);
739         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
740         error = xattr_check_inode(inode, header, end);
741         if (error)
742                 goto cleanup;
743         error = ext4_xattr_list_entries(dentry, IFIRST(header),
744                                         buffer, buffer_size);
745
746 cleanup:
747         brelse(iloc.bh);
748         return error;
749 }
750
751 /*
752  * Inode operation listxattr()
753  *
754  * d_inode(dentry)->i_rwsem: don't care
755  *
756  * Copy a list of attribute names into the buffer
757  * provided, or compute the buffer size required.
758  * Buffer is NULL to compute the size of the buffer required.
759  *
760  * Returns a negative error number on failure, or the number of bytes
761  * used / required on success.
762  */
763 ssize_t
764 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
765 {
766         int ret, ret2;
767
768         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
769         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
770         if (ret < 0)
771                 goto errout;
772         if (buffer) {
773                 buffer += ret;
774                 buffer_size -= ret;
775         }
776         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
777         if (ret < 0)
778                 goto errout;
779         ret += ret2;
780 errout:
781         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
782         return ret;
783 }
784
785 /*
786  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
787  * not set, set it.
788  */
789 static void ext4_xattr_update_super_block(handle_t *handle,
790                                           struct super_block *sb)
791 {
792         if (ext4_has_feature_xattr(sb))
793                 return;
794
795         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
796         if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
797                 ext4_set_feature_xattr(sb);
798                 ext4_handle_dirty_super(handle, sb);
799         }
800 }
801
802 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
803 {
804         struct ext4_iloc iloc = { .bh = NULL };
805         struct buffer_head *bh = NULL;
806         struct ext4_inode *raw_inode;
807         struct ext4_xattr_ibody_header *header;
808         struct ext4_xattr_entry *entry;
809         qsize_t ea_inode_refs = 0;
810         void *end;
811         int ret;
812
813         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
814
815         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
816                 ret = ext4_get_inode_loc(inode, &iloc);
817                 if (ret)
818                         goto out;
819                 raw_inode = ext4_raw_inode(&iloc);
820                 header = IHDR(inode, raw_inode);
821                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
822                 ret = xattr_check_inode(inode, header, end);
823                 if (ret)
824                         goto out;
825
826                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
827                      entry = EXT4_XATTR_NEXT(entry))
828                         if (entry->e_value_inum)
829                                 ea_inode_refs++;
830         }
831
832         if (EXT4_I(inode)->i_file_acl) {
833                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
834                 if (!bh) {
835                         ret = -EIO;
836                         goto out;
837                 }
838
839                 ret = ext4_xattr_check_block(inode, bh);
840                 if (ret)
841                         goto out;
842
843                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
844                      entry = EXT4_XATTR_NEXT(entry))
845                         if (entry->e_value_inum)
846                                 ea_inode_refs++;
847         }
848         *usage = ea_inode_refs + 1;
849         ret = 0;
850 out:
851         brelse(iloc.bh);
852         brelse(bh);
853         return ret;
854 }
855
856 static inline size_t round_up_cluster(struct inode *inode, size_t length)
857 {
858         struct super_block *sb = inode->i_sb;
859         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
860                                     inode->i_blkbits);
861         size_t mask = ~(cluster_size - 1);
862
863         return (length + cluster_size - 1) & mask;
864 }
865
866 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
867 {
868         int err;
869
870         err = dquot_alloc_inode(inode);
871         if (err)
872                 return err;
873         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
874         if (err)
875                 dquot_free_inode(inode);
876         return err;
877 }
878
879 static void ext4_xattr_inode_free_quota(struct inode *parent,
880                                         struct inode *ea_inode,
881                                         size_t len)
882 {
883         if (ea_inode &&
884             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
885                 return;
886         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
887         dquot_free_inode(parent);
888 }
889
890 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
891                              struct buffer_head *block_bh, size_t value_len,
892                              bool is_create)
893 {
894         int credits;
895         int blocks;
896
897         /*
898          * 1) Owner inode update
899          * 2) Ref count update on old xattr block
900          * 3) new xattr block
901          * 4) block bitmap update for new xattr block
902          * 5) group descriptor for new xattr block
903          * 6) block bitmap update for old xattr block
904          * 7) group descriptor for old block
905          *
906          * 6 & 7 can happen if we have two racing threads T_a and T_b
907          * which are each trying to set an xattr on inodes I_a and I_b
908          * which were both initially sharing an xattr block.
909          */
910         credits = 7;
911
912         /* Quota updates. */
913         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
914
915         /*
916          * In case of inline data, we may push out the data to a block,
917          * so we need to reserve credits for this eventuality
918          */
919         if (inode && ext4_has_inline_data(inode))
920                 credits += ext4_writepage_trans_blocks(inode) + 1;
921
922         /* We are done if ea_inode feature is not enabled. */
923         if (!ext4_has_feature_ea_inode(sb))
924                 return credits;
925
926         /* New ea_inode, inode map, block bitmap, group descriptor. */
927         credits += 4;
928
929         /* Data blocks. */
930         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
931
932         /* Indirection block or one level of extent tree. */
933         blocks += 1;
934
935         /* Block bitmap and group descriptor updates for each block. */
936         credits += blocks * 2;
937
938         /* Blocks themselves. */
939         credits += blocks;
940
941         if (!is_create) {
942                 /* Dereference ea_inode holding old xattr value.
943                  * Old ea_inode, inode map, block bitmap, group descriptor.
944                  */
945                 credits += 4;
946
947                 /* Data blocks for old ea_inode. */
948                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
949
950                 /* Indirection block or one level of extent tree for old
951                  * ea_inode.
952                  */
953                 blocks += 1;
954
955                 /* Block bitmap and group descriptor updates for each block. */
956                 credits += blocks * 2;
957         }
958
959         /* We may need to clone the existing xattr block in which case we need
960          * to increment ref counts for existing ea_inodes referenced by it.
961          */
962         if (block_bh) {
963                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
964
965                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
966                         if (entry->e_value_inum)
967                                 /* Ref count update on ea_inode. */
968                                 credits += 1;
969         }
970         return credits;
971 }
972
973 static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
974                                      int credits, struct buffer_head *bh,
975                                      bool dirty, bool block_csum)
976 {
977         int error;
978
979         if (!ext4_handle_valid(handle))
980                 return 0;
981
982         if (handle->h_buffer_credits >= credits)
983                 return 0;
984
985         error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
986         if (!error)
987                 return 0;
988         if (error < 0) {
989                 ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
990                 return error;
991         }
992
993         if (bh && dirty) {
994                 if (block_csum)
995                         ext4_xattr_block_csum_set(inode, bh);
996                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
997                 if (error) {
998                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
999                                      error);
1000                         return error;
1001                 }
1002         }
1003
1004         error = ext4_journal_restart(handle, credits);
1005         if (error) {
1006                 ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1007                 return error;
1008         }
1009
1010         if (bh) {
1011                 error = ext4_journal_get_write_access(handle, bh);
1012                 if (error) {
1013                         ext4_warning(inode->i_sb,
1014                                      "Get write access failed (error %d)",
1015                                      error);
1016                         return error;
1017                 }
1018         }
1019         return 0;
1020 }
1021
1022 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1023                                        int ref_change)
1024 {
1025         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1026         struct ext4_iloc iloc;
1027         s64 ref_count;
1028         u32 hash;
1029         int ret;
1030
1031         inode_lock(ea_inode);
1032
1033         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1034         if (ret)
1035                 goto out;
1036
1037         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1038         ref_count += ref_change;
1039         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1040
1041         if (ref_change > 0) {
1042                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1043                           ea_inode->i_ino, ref_count);
1044
1045                 if (ref_count == 1) {
1046                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1047                                   ea_inode->i_ino, ea_inode->i_nlink);
1048
1049                         set_nlink(ea_inode, 1);
1050                         ext4_orphan_del(handle, ea_inode);
1051
1052                         if (ea_inode_cache) {
1053                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1054                                 mb_cache_entry_create(ea_inode_cache,
1055                                                       GFP_NOFS, hash,
1056                                                       ea_inode->i_ino,
1057                                                       true /* reusable */);
1058                         }
1059                 }
1060         } else {
1061                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1062                           ea_inode->i_ino, ref_count);
1063
1064                 if (ref_count == 0) {
1065                         WARN_ONCE(ea_inode->i_nlink != 1,
1066                                   "EA inode %lu i_nlink=%u",
1067                                   ea_inode->i_ino, ea_inode->i_nlink);
1068
1069                         clear_nlink(ea_inode);
1070                         ext4_orphan_add(handle, ea_inode);
1071
1072                         if (ea_inode_cache) {
1073                                 hash = ext4_xattr_inode_get_hash(ea_inode);
1074                                 mb_cache_entry_delete(ea_inode_cache, hash,
1075                                                       ea_inode->i_ino);
1076                         }
1077                 }
1078         }
1079
1080         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1081         if (ret)
1082                 ext4_warning_inode(ea_inode,
1083                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1084 out:
1085         inode_unlock(ea_inode);
1086         return ret;
1087 }
1088
1089 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1090 {
1091         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1092 }
1093
1094 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1095 {
1096         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1097 }
1098
1099 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1100                                         struct ext4_xattr_entry *first)
1101 {
1102         struct inode *ea_inode;
1103         struct ext4_xattr_entry *entry;
1104         struct ext4_xattr_entry *failed_entry;
1105         unsigned int ea_ino;
1106         int err, saved_err;
1107
1108         for (entry = first; !IS_LAST_ENTRY(entry);
1109              entry = EXT4_XATTR_NEXT(entry)) {
1110                 if (!entry->e_value_inum)
1111                         continue;
1112                 ea_ino = le32_to_cpu(entry->e_value_inum);
1113                 err = ext4_xattr_inode_iget(parent, ea_ino,
1114                                             le32_to_cpu(entry->e_hash),
1115                                             &ea_inode);
1116                 if (err)
1117                         goto cleanup;
1118                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1119                 if (err) {
1120                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1121                         iput(ea_inode);
1122                         goto cleanup;
1123                 }
1124                 iput(ea_inode);
1125         }
1126         return 0;
1127
1128 cleanup:
1129         saved_err = err;
1130         failed_entry = entry;
1131
1132         for (entry = first; entry != failed_entry;
1133              entry = EXT4_XATTR_NEXT(entry)) {
1134                 if (!entry->e_value_inum)
1135                         continue;
1136                 ea_ino = le32_to_cpu(entry->e_value_inum);
1137                 err = ext4_xattr_inode_iget(parent, ea_ino,
1138                                             le32_to_cpu(entry->e_hash),
1139                                             &ea_inode);
1140                 if (err) {
1141                         ext4_warning(parent->i_sb,
1142                                      "cleanup ea_ino %u iget error %d", ea_ino,
1143                                      err);
1144                         continue;
1145                 }
1146                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1147                 if (err)
1148                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1149                                            err);
1150                 iput(ea_inode);
1151         }
1152         return saved_err;
1153 }
1154
1155 static void
1156 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1157                              struct buffer_head *bh,
1158                              struct ext4_xattr_entry *first, bool block_csum,
1159                              struct ext4_xattr_inode_array **ea_inode_array,
1160                              int extra_credits, bool skip_quota)
1161 {
1162         struct inode *ea_inode;
1163         struct ext4_xattr_entry *entry;
1164         bool dirty = false;
1165         unsigned int ea_ino;
1166         int err;
1167         int credits;
1168
1169         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1170         credits = 2 + extra_credits;
1171
1172         for (entry = first; !IS_LAST_ENTRY(entry);
1173              entry = EXT4_XATTR_NEXT(entry)) {
1174                 if (!entry->e_value_inum)
1175                         continue;
1176                 ea_ino = le32_to_cpu(entry->e_value_inum);
1177                 err = ext4_xattr_inode_iget(parent, ea_ino,
1178                                             le32_to_cpu(entry->e_hash),
1179                                             &ea_inode);
1180                 if (err)
1181                         continue;
1182
1183                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1184                 if (err) {
1185                         ext4_warning_inode(ea_inode,
1186                                            "Expand inode array err=%d", err);
1187                         iput(ea_inode);
1188                         continue;
1189                 }
1190
1191                 err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1192                                                 dirty, block_csum);
1193                 if (err) {
1194                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1195                                            err);
1196                         continue;
1197                 }
1198
1199                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1200                 if (err) {
1201                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1202                                            err);
1203                         continue;
1204                 }
1205
1206                 if (!skip_quota)
1207                         ext4_xattr_inode_free_quota(parent, ea_inode,
1208                                               le32_to_cpu(entry->e_value_size));
1209
1210                 /*
1211                  * Forget about ea_inode within the same transaction that
1212                  * decrements the ref count. This avoids duplicate decrements in
1213                  * case the rest of the work spills over to subsequent
1214                  * transactions.
1215                  */
1216                 entry->e_value_inum = 0;
1217                 entry->e_value_size = 0;
1218
1219                 dirty = true;
1220         }
1221
1222         if (dirty) {
1223                 /*
1224                  * Note that we are deliberately skipping csum calculation for
1225                  * the final update because we do not expect any journal
1226                  * restarts until xattr block is freed.
1227                  */
1228
1229                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1230                 if (err)
1231                         ext4_warning_inode(parent,
1232                                            "handle dirty metadata err=%d", err);
1233         }
1234 }
1235
1236 /*
1237  * Release the xattr block BH: If the reference count is > 1, decrement it;
1238  * otherwise free the block.
1239  */
1240 static void
1241 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1242                          struct buffer_head *bh,
1243                          struct ext4_xattr_inode_array **ea_inode_array,
1244                          int extra_credits)
1245 {
1246         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1247         u32 hash, ref;
1248         int error = 0;
1249
1250         BUFFER_TRACE(bh, "get_write_access");
1251         error = ext4_journal_get_write_access(handle, bh);
1252         if (error)
1253                 goto out;
1254
1255         lock_buffer(bh);
1256         hash = le32_to_cpu(BHDR(bh)->h_hash);
1257         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1258         if (ref == 1) {
1259                 ea_bdebug(bh, "refcount now=0; freeing");
1260                 /*
1261                  * This must happen under buffer lock for
1262                  * ext4_xattr_block_set() to reliably detect freed block
1263                  */
1264                 if (ea_block_cache)
1265                         mb_cache_entry_delete(ea_block_cache, hash,
1266                                               bh->b_blocknr);
1267                 get_bh(bh);
1268                 unlock_buffer(bh);
1269
1270                 if (ext4_has_feature_ea_inode(inode->i_sb))
1271                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1272                                                      BFIRST(bh),
1273                                                      true /* block_csum */,
1274                                                      ea_inode_array,
1275                                                      extra_credits,
1276                                                      true /* skip_quota */);
1277                 ext4_free_blocks(handle, inode, bh, 0, 1,
1278                                  EXT4_FREE_BLOCKS_METADATA |
1279                                  EXT4_FREE_BLOCKS_FORGET);
1280         } else {
1281                 ref--;
1282                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1283                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1284                         struct mb_cache_entry *ce;
1285
1286                         if (ea_block_cache) {
1287                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1288                                                         bh->b_blocknr);
1289                                 if (ce) {
1290                                         ce->e_reusable = 1;
1291                                         mb_cache_entry_put(ea_block_cache, ce);
1292                                 }
1293                         }
1294                 }
1295
1296                 ext4_xattr_block_csum_set(inode, bh);
1297                 /*
1298                  * Beware of this ugliness: Releasing of xattr block references
1299                  * from different inodes can race and so we have to protect
1300                  * from a race where someone else frees the block (and releases
1301                  * its journal_head) before we are done dirtying the buffer. In
1302                  * nojournal mode this race is harmless and we actually cannot
1303                  * call ext4_handle_dirty_metadata() with locked buffer as
1304                  * that function can call sync_dirty_buffer() so for that case
1305                  * we handle the dirtying after unlocking the buffer.
1306                  */
1307                 if (ext4_handle_valid(handle))
1308                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1309                 unlock_buffer(bh);
1310                 if (!ext4_handle_valid(handle))
1311                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1312                 if (IS_SYNC(inode))
1313                         ext4_handle_sync(handle);
1314                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1315                 ea_bdebug(bh, "refcount now=%d; releasing",
1316                           le32_to_cpu(BHDR(bh)->h_refcount));
1317         }
1318 out:
1319         ext4_std_error(inode->i_sb, error);
1320         return;
1321 }
1322
1323 /*
1324  * Find the available free space for EAs. This also returns the total number of
1325  * bytes used by EA entries.
1326  */
1327 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1328                                     size_t *min_offs, void *base, int *total)
1329 {
1330         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1331                 if (!last->e_value_inum && last->e_value_size) {
1332                         size_t offs = le16_to_cpu(last->e_value_offs);
1333                         if (offs < *min_offs)
1334                                 *min_offs = offs;
1335                 }
1336                 if (total)
1337                         *total += EXT4_XATTR_LEN(last->e_name_len);
1338         }
1339         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1340 }
1341
1342 /*
1343  * Write the value of the EA in an inode.
1344  */
1345 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1346                                   const void *buf, int bufsize)
1347 {
1348         struct buffer_head *bh = NULL;
1349         unsigned long block = 0;
1350         int blocksize = ea_inode->i_sb->s_blocksize;
1351         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1352         int csize, wsize = 0;
1353         int ret = 0;
1354         int retries = 0;
1355
1356 retry:
1357         while (ret >= 0 && ret < max_blocks) {
1358                 struct ext4_map_blocks map;
1359                 map.m_lblk = block += ret;
1360                 map.m_len = max_blocks -= ret;
1361
1362                 ret = ext4_map_blocks(handle, ea_inode, &map,
1363                                       EXT4_GET_BLOCKS_CREATE);
1364                 if (ret <= 0) {
1365                         ext4_mark_inode_dirty(handle, ea_inode);
1366                         if (ret == -ENOSPC &&
1367                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1368                                 ret = 0;
1369                                 goto retry;
1370                         }
1371                         break;
1372                 }
1373         }
1374
1375         if (ret < 0)
1376                 return ret;
1377
1378         block = 0;
1379         while (wsize < bufsize) {
1380                 if (bh != NULL)
1381                         brelse(bh);
1382                 csize = (bufsize - wsize) > blocksize ? blocksize :
1383                                                                 bufsize - wsize;
1384                 bh = ext4_getblk(handle, ea_inode, block, 0);
1385                 if (IS_ERR(bh))
1386                         return PTR_ERR(bh);
1387                 if (!bh) {
1388                         WARN_ON_ONCE(1);
1389                         EXT4_ERROR_INODE(ea_inode,
1390                                          "ext4_getblk() return bh = NULL");
1391                         return -EFSCORRUPTED;
1392                 }
1393                 ret = ext4_journal_get_write_access(handle, bh);
1394                 if (ret)
1395                         goto out;
1396
1397                 memcpy(bh->b_data, buf, csize);
1398                 set_buffer_uptodate(bh);
1399                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1400
1401                 buf += csize;
1402                 wsize += csize;
1403                 block += 1;
1404         }
1405
1406         inode_lock(ea_inode);
1407         i_size_write(ea_inode, wsize);
1408         ext4_update_i_disksize(ea_inode, wsize);
1409         inode_unlock(ea_inode);
1410
1411         ext4_mark_inode_dirty(handle, ea_inode);
1412
1413 out:
1414         brelse(bh);
1415
1416         return ret;
1417 }
1418
1419 /*
1420  * Create an inode to store the value of a large EA.
1421  */
1422 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1423                                              struct inode *inode, u32 hash)
1424 {
1425         struct inode *ea_inode = NULL;
1426         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1427         int err;
1428
1429         /*
1430          * Let the next inode be the goal, so we try and allocate the EA inode
1431          * in the same group, or nearby one.
1432          */
1433         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1434                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1435                                   EXT4_EA_INODE_FL);
1436         if (!IS_ERR(ea_inode)) {
1437                 ea_inode->i_op = &ext4_file_inode_operations;
1438                 ea_inode->i_fop = &ext4_file_operations;
1439                 ext4_set_aops(ea_inode);
1440                 ext4_xattr_inode_set_class(ea_inode);
1441                 unlock_new_inode(ea_inode);
1442                 ext4_xattr_inode_set_ref(ea_inode, 1);
1443                 ext4_xattr_inode_set_hash(ea_inode, hash);
1444                 err = ext4_mark_inode_dirty(handle, ea_inode);
1445                 if (!err)
1446                         err = ext4_inode_attach_jinode(ea_inode);
1447                 if (err) {
1448                         iput(ea_inode);
1449                         return ERR_PTR(err);
1450                 }
1451
1452                 /*
1453                  * Xattr inodes are shared therefore quota charging is performed
1454                  * at a higher level.
1455                  */
1456                 dquot_free_inode(ea_inode);
1457                 dquot_drop(ea_inode);
1458                 inode_lock(ea_inode);
1459                 ea_inode->i_flags |= S_NOQUOTA;
1460                 inode_unlock(ea_inode);
1461         }
1462
1463         return ea_inode;
1464 }
1465
1466 static struct inode *
1467 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1468                             size_t value_len, u32 hash)
1469 {
1470         struct inode *ea_inode;
1471         struct mb_cache_entry *ce;
1472         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1473         void *ea_data;
1474
1475         if (!ea_inode_cache)
1476                 return NULL;
1477
1478         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1479         if (!ce)
1480                 return NULL;
1481
1482         ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1483         if (!ea_data) {
1484                 mb_cache_entry_put(ea_inode_cache, ce);
1485                 return NULL;
1486         }
1487
1488         while (ce) {
1489                 ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1490                 if (!IS_ERR(ea_inode) &&
1491                     !is_bad_inode(ea_inode) &&
1492                     (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1493                     i_size_read(ea_inode) == value_len &&
1494                     !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1495                     !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1496                                                     value_len) &&
1497                     !memcmp(value, ea_data, value_len)) {
1498                         mb_cache_entry_touch(ea_inode_cache, ce);
1499                         mb_cache_entry_put(ea_inode_cache, ce);
1500                         kvfree(ea_data);
1501                         return ea_inode;
1502                 }
1503
1504                 if (!IS_ERR(ea_inode))
1505                         iput(ea_inode);
1506                 ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1507         }
1508         kvfree(ea_data);
1509         return NULL;
1510 }
1511
1512 /*
1513  * Add value of the EA in an inode.
1514  */
1515 static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1516                                           const void *value, size_t value_len,
1517                                           struct inode **ret_inode)
1518 {
1519         struct inode *ea_inode;
1520         u32 hash;
1521         int err;
1522
1523         hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1524         ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1525         if (ea_inode) {
1526                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1527                 if (err) {
1528                         iput(ea_inode);
1529                         return err;
1530                 }
1531
1532                 *ret_inode = ea_inode;
1533                 return 0;
1534         }
1535
1536         /* Create an inode for the EA value */
1537         ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1538         if (IS_ERR(ea_inode))
1539                 return PTR_ERR(ea_inode);
1540
1541         err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1542         if (err) {
1543                 ext4_xattr_inode_dec_ref(handle, ea_inode);
1544                 iput(ea_inode);
1545                 return err;
1546         }
1547
1548         if (EA_INODE_CACHE(inode))
1549                 mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1550                                       ea_inode->i_ino, true /* reusable */);
1551
1552         *ret_inode = ea_inode;
1553         return 0;
1554 }
1555
1556 /*
1557  * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1558  * feature is enabled.
1559  */
1560 #define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1561
1562 static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1563                                 struct ext4_xattr_search *s,
1564                                 handle_t *handle, struct inode *inode,
1565                                 bool is_block)
1566 {
1567         struct ext4_xattr_entry *last, *next;
1568         struct ext4_xattr_entry *here = s->here;
1569         size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1570         int in_inode = i->in_inode;
1571         struct inode *old_ea_inode = NULL;
1572         struct inode *new_ea_inode = NULL;
1573         size_t old_size, new_size;
1574         int ret;
1575
1576         /* Space used by old and new values. */
1577         old_size = (!s->not_found && !here->e_value_inum) ?
1578                         EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1579         new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1580
1581         /*
1582          * Optimization for the simple case when old and new values have the
1583          * same padded sizes. Not applicable if external inodes are involved.
1584          */
1585         if (new_size && new_size == old_size) {
1586                 size_t offs = le16_to_cpu(here->e_value_offs);
1587                 void *val = s->base + offs;
1588
1589                 here->e_value_size = cpu_to_le32(i->value_len);
1590                 if (i->value == EXT4_ZERO_XATTR_VALUE) {
1591                         memset(val, 0, new_size);
1592                 } else {
1593                         memcpy(val, i->value, i->value_len);
1594                         /* Clear padding bytes. */
1595                         memset(val + i->value_len, 0, new_size - i->value_len);
1596                 }
1597                 goto update_hash;
1598         }
1599
1600         /* Compute min_offs and last. */
1601         last = s->first;
1602         for (; !IS_LAST_ENTRY(last); last = next) {
1603                 next = EXT4_XATTR_NEXT(last);
1604                 if ((void *)next >= s->end) {
1605                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1606                         ret = -EFSCORRUPTED;
1607                         goto out;
1608                 }
1609                 if (!last->e_value_inum && last->e_value_size) {
1610                         size_t offs = le16_to_cpu(last->e_value_offs);
1611                         if (offs < min_offs)
1612                                 min_offs = offs;
1613                 }
1614         }
1615
1616         /* Check whether we have enough space. */
1617         if (i->value) {
1618                 size_t free;
1619
1620                 free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1621                 if (!s->not_found)
1622                         free += EXT4_XATTR_LEN(name_len) + old_size;
1623
1624                 if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1625                         ret = -ENOSPC;
1626                         goto out;
1627                 }
1628
1629                 /*
1630                  * If storing the value in an external inode is an option,
1631                  * reserve space for xattr entries/names in the external
1632                  * attribute block so that a long value does not occupy the
1633                  * whole space and prevent futher entries being added.
1634                  */
1635                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
1636                     new_size && is_block &&
1637                     (min_offs + old_size - new_size) <
1638                                         EXT4_XATTR_BLOCK_RESERVE(inode)) {
1639                         ret = -ENOSPC;
1640                         goto out;
1641                 }
1642         }
1643
1644         /*
1645          * Getting access to old and new ea inodes is subject to failures.
1646          * Finish that work before doing any modifications to the xattr data.
1647          */
1648         if (!s->not_found && here->e_value_inum) {
1649                 ret = ext4_xattr_inode_iget(inode,
1650                                             le32_to_cpu(here->e_value_inum),
1651                                             le32_to_cpu(here->e_hash),
1652                                             &old_ea_inode);
1653                 if (ret) {
1654                         old_ea_inode = NULL;
1655                         goto out;
1656                 }
1657         }
1658         if (i->value && in_inode) {
1659                 WARN_ON_ONCE(!i->value_len);
1660
1661                 ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1662                 if (ret)
1663                         goto out;
1664
1665                 ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1666                                                      i->value_len,
1667                                                      &new_ea_inode);
1668                 if (ret) {
1669                         new_ea_inode = NULL;
1670                         ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1671                         goto out;
1672                 }
1673         }
1674
1675         if (old_ea_inode) {
1676                 /* We are ready to release ref count on the old_ea_inode. */
1677                 ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1678                 if (ret) {
1679                         /* Release newly required ref count on new_ea_inode. */
1680                         if (new_ea_inode) {
1681                                 int err;
1682
1683                                 err = ext4_xattr_inode_dec_ref(handle,
1684                                                                new_ea_inode);
1685                                 if (err)
1686                                         ext4_warning_inode(new_ea_inode,
1687                                                   "dec ref new_ea_inode err=%d",
1688                                                   err);
1689                                 ext4_xattr_inode_free_quota(inode, new_ea_inode,
1690                                                             i->value_len);
1691                         }
1692                         goto out;
1693                 }
1694
1695                 ext4_xattr_inode_free_quota(inode, old_ea_inode,
1696                                             le32_to_cpu(here->e_value_size));
1697         }
1698
1699         /* No failures allowed past this point. */
1700
1701         if (!s->not_found && here->e_value_size && here->e_value_offs) {
1702                 /* Remove the old value. */
1703                 void *first_val = s->base + min_offs;
1704                 size_t offs = le16_to_cpu(here->e_value_offs);
1705                 void *val = s->base + offs;
1706
1707                 memmove(first_val + old_size, first_val, val - first_val);
1708                 memset(first_val, 0, old_size);
1709                 min_offs += old_size;
1710
1711                 /* Adjust all value offsets. */
1712                 last = s->first;
1713                 while (!IS_LAST_ENTRY(last)) {
1714                         size_t o = le16_to_cpu(last->e_value_offs);
1715
1716                         if (!last->e_value_inum &&
1717                             last->e_value_size && o < offs)
1718                                 last->e_value_offs = cpu_to_le16(o + old_size);
1719                         last = EXT4_XATTR_NEXT(last);
1720                 }
1721         }
1722
1723         if (!i->value) {
1724                 /* Remove old name. */
1725                 size_t size = EXT4_XATTR_LEN(name_len);
1726
1727                 last = ENTRY((void *)last - size);
1728                 memmove(here, (void *)here + size,
1729                         (void *)last - (void *)here + sizeof(__u32));
1730                 memset(last, 0, size);
1731         } else if (s->not_found) {
1732                 /* Insert new name. */
1733                 size_t size = EXT4_XATTR_LEN(name_len);
1734                 size_t rest = (void *)last - (void *)here + sizeof(__u32);
1735
1736                 memmove((void *)here + size, here, rest);
1737                 memset(here, 0, size);
1738                 here->e_name_index = i->name_index;
1739                 here->e_name_len = name_len;
1740                 memcpy(here->e_name, i->name, name_len);
1741         } else {
1742                 /* This is an update, reset value info. */
1743                 here->e_value_inum = 0;
1744                 here->e_value_offs = 0;
1745                 here->e_value_size = 0;
1746         }
1747
1748         if (i->value) {
1749                 /* Insert new value. */
1750                 if (in_inode) {
1751                         here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1752                 } else if (i->value_len) {
1753                         void *val = s->base + min_offs - new_size;
1754
1755                         here->e_value_offs = cpu_to_le16(min_offs - new_size);
1756                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
1757                                 memset(val, 0, new_size);
1758                         } else {
1759                                 memcpy(val, i->value, i->value_len);
1760                                 /* Clear padding bytes. */
1761                                 memset(val + i->value_len, 0,
1762                                        new_size - i->value_len);
1763                         }
1764                 }
1765                 here->e_value_size = cpu_to_le32(i->value_len);
1766         }
1767
1768 update_hash:
1769         if (i->value) {
1770                 __le32 hash = 0;
1771
1772                 /* Entry hash calculation. */
1773                 if (in_inode) {
1774                         __le32 crc32c_hash;
1775
1776                         /*
1777                          * Feed crc32c hash instead of the raw value for entry
1778                          * hash calculation. This is to avoid walking
1779                          * potentially long value buffer again.
1780                          */
1781                         crc32c_hash = cpu_to_le32(
1782                                        ext4_xattr_inode_get_hash(new_ea_inode));
1783                         hash = ext4_xattr_hash_entry(here->e_name,
1784                                                      here->e_name_len,
1785                                                      &crc32c_hash, 1);
1786                 } else if (is_block) {
1787                         __le32 *value = s->base + le16_to_cpu(
1788                                                         here->e_value_offs);
1789
1790                         hash = ext4_xattr_hash_entry(here->e_name,
1791                                                      here->e_name_len, value,
1792                                                      new_size >> 2);
1793                 }
1794                 here->e_hash = hash;
1795         }
1796
1797         if (is_block)
1798                 ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1799
1800         ret = 0;
1801 out:
1802         iput(old_ea_inode);
1803         iput(new_ea_inode);
1804         return ret;
1805 }
1806
1807 struct ext4_xattr_block_find {
1808         struct ext4_xattr_search s;
1809         struct buffer_head *bh;
1810 };
1811
1812 static int
1813 ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1814                       struct ext4_xattr_block_find *bs)
1815 {
1816         struct super_block *sb = inode->i_sb;
1817         int error;
1818
1819         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1820                   i->name_index, i->name, i->value, (long)i->value_len);
1821
1822         if (EXT4_I(inode)->i_file_acl) {
1823                 /* The inode already has an extended attribute block. */
1824                 bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1825                 error = -EIO;
1826                 if (!bs->bh)
1827                         goto cleanup;
1828                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1829                         atomic_read(&(bs->bh->b_count)),
1830                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1831                 error = ext4_xattr_check_block(inode, bs->bh);
1832                 if (error)
1833                         goto cleanup;
1834                 /* Find the named attribute. */
1835                 bs->s.base = BHDR(bs->bh);
1836                 bs->s.first = BFIRST(bs->bh);
1837                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1838                 bs->s.here = bs->s.first;
1839                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1840                                          i->name_index, i->name, 1);
1841                 if (error && error != -ENODATA)
1842                         goto cleanup;
1843                 bs->s.not_found = error;
1844         }
1845         error = 0;
1846
1847 cleanup:
1848         return error;
1849 }
1850
1851 static int
1852 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1853                      struct ext4_xattr_info *i,
1854                      struct ext4_xattr_block_find *bs)
1855 {
1856         struct super_block *sb = inode->i_sb;
1857         struct buffer_head *new_bh = NULL;
1858         struct ext4_xattr_search s_copy = bs->s;
1859         struct ext4_xattr_search *s = &s_copy;
1860         struct mb_cache_entry *ce = NULL;
1861         int error = 0;
1862         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1863         struct inode *ea_inode = NULL, *tmp_inode;
1864         size_t old_ea_inode_quota = 0;
1865         unsigned int ea_ino;
1866
1867
1868 #define header(x) ((struct ext4_xattr_header *)(x))
1869
1870         if (s->base) {
1871                 BUFFER_TRACE(bs->bh, "get_write_access");
1872                 error = ext4_journal_get_write_access(handle, bs->bh);
1873                 if (error)
1874                         goto cleanup;
1875                 lock_buffer(bs->bh);
1876
1877                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1878                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1879
1880                         /*
1881                          * This must happen under buffer lock for
1882                          * ext4_xattr_block_set() to reliably detect modified
1883                          * block
1884                          */
1885                         if (ea_block_cache)
1886                                 mb_cache_entry_delete(ea_block_cache, hash,
1887                                                       bs->bh->b_blocknr);
1888                         ea_bdebug(bs->bh, "modifying in-place");
1889                         error = ext4_xattr_set_entry(i, s, handle, inode,
1890                                                      true /* is_block */);
1891                         ext4_xattr_block_csum_set(inode, bs->bh);
1892                         unlock_buffer(bs->bh);
1893                         if (error == -EFSCORRUPTED)
1894                                 goto bad_block;
1895                         if (!error)
1896                                 error = ext4_handle_dirty_metadata(handle,
1897                                                                    inode,
1898                                                                    bs->bh);
1899                         if (error)
1900                                 goto cleanup;
1901                         goto inserted;
1902                 } else {
1903                         int offset = (char *)s->here - bs->bh->b_data;
1904
1905                         unlock_buffer(bs->bh);
1906                         ea_bdebug(bs->bh, "cloning");
1907                         s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1908                         error = -ENOMEM;
1909                         if (s->base == NULL)
1910                                 goto cleanup;
1911                         memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1912                         s->first = ENTRY(header(s->base)+1);
1913                         header(s->base)->h_refcount = cpu_to_le32(1);
1914                         s->here = ENTRY(s->base + offset);
1915                         s->end = s->base + bs->bh->b_size;
1916
1917                         /*
1918                          * If existing entry points to an xattr inode, we need
1919                          * to prevent ext4_xattr_set_entry() from decrementing
1920                          * ref count on it because the reference belongs to the
1921                          * original block. In this case, make the entry look
1922                          * like it has an empty value.
1923                          */
1924                         if (!s->not_found && s->here->e_value_inum) {
1925                                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1926                                 error = ext4_xattr_inode_iget(inode, ea_ino,
1927                                               le32_to_cpu(s->here->e_hash),
1928                                               &tmp_inode);
1929                                 if (error)
1930                                         goto cleanup;
1931
1932                                 if (!ext4_test_inode_state(tmp_inode,
1933                                                 EXT4_STATE_LUSTRE_EA_INODE)) {
1934                                         /*
1935                                          * Defer quota free call for previous
1936                                          * inode until success is guaranteed.
1937                                          */
1938                                         old_ea_inode_quota = le32_to_cpu(
1939                                                         s->here->e_value_size);
1940                                 }
1941                                 iput(tmp_inode);
1942
1943                                 s->here->e_value_inum = 0;
1944                                 s->here->e_value_size = 0;
1945                         }
1946                 }
1947         } else {
1948                 /* Allocate a buffer where we construct the new block. */
1949                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1950                 /* assert(header == s->base) */
1951                 error = -ENOMEM;
1952                 if (s->base == NULL)
1953                         goto cleanup;
1954                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1955                 header(s->base)->h_blocks = cpu_to_le32(1);
1956                 header(s->base)->h_refcount = cpu_to_le32(1);
1957                 s->first = ENTRY(header(s->base)+1);
1958                 s->here = ENTRY(header(s->base)+1);
1959                 s->end = s->base + sb->s_blocksize;
1960         }
1961
1962         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1963         if (error == -EFSCORRUPTED)
1964                 goto bad_block;
1965         if (error)
1966                 goto cleanup;
1967
1968         if (i->value && s->here->e_value_inum) {
1969                 /*
1970                  * A ref count on ea_inode has been taken as part of the call to
1971                  * ext4_xattr_set_entry() above. We would like to drop this
1972                  * extra ref but we have to wait until the xattr block is
1973                  * initialized and has its own ref count on the ea_inode.
1974                  */
1975                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1976                 error = ext4_xattr_inode_iget(inode, ea_ino,
1977                                               le32_to_cpu(s->here->e_hash),
1978                                               &ea_inode);
1979                 if (error) {
1980                         ea_inode = NULL;
1981                         goto cleanup;
1982                 }
1983         }
1984
1985 inserted:
1986         if (!IS_LAST_ENTRY(s->first)) {
1987                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1988                                                      &ce);
1989                 if (new_bh) {
1990                         /* We found an identical block in the cache. */
1991                         if (new_bh == bs->bh)
1992                                 ea_bdebug(new_bh, "keeping");
1993                         else {
1994                                 u32 ref;
1995
1996                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
1997
1998                                 /* The old block is released after updating
1999                                    the inode. */
2000                                 error = dquot_alloc_block(inode,
2001                                                 EXT4_C2B(EXT4_SB(sb), 1));
2002                                 if (error)
2003                                         goto cleanup;
2004                                 BUFFER_TRACE(new_bh, "get_write_access");
2005                                 error = ext4_journal_get_write_access(handle,
2006                                                                       new_bh);
2007                                 if (error)
2008                                         goto cleanup_dquot;
2009                                 lock_buffer(new_bh);
2010                                 /*
2011                                  * We have to be careful about races with
2012                                  * freeing, rehashing or adding references to
2013                                  * xattr block. Once we hold buffer lock xattr
2014                                  * block's state is stable so we can check
2015                                  * whether the block got freed / rehashed or
2016                                  * not.  Since we unhash mbcache entry under
2017                                  * buffer lock when freeing / rehashing xattr
2018                                  * block, checking whether entry is still
2019                                  * hashed is reliable. Same rules hold for
2020                                  * e_reusable handling.
2021                                  */
2022                                 if (hlist_bl_unhashed(&ce->e_hash_list) ||
2023                                     !ce->e_reusable) {
2024                                         /*
2025                                          * Undo everything and check mbcache
2026                                          * again.
2027                                          */
2028                                         unlock_buffer(new_bh);
2029                                         dquot_free_block(inode,
2030                                                          EXT4_C2B(EXT4_SB(sb),
2031                                                                   1));
2032                                         brelse(new_bh);
2033                                         mb_cache_entry_put(ea_block_cache, ce);
2034                                         ce = NULL;
2035                                         new_bh = NULL;
2036                                         goto inserted;
2037                                 }
2038                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2039                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2040                                 if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2041                                         ce->e_reusable = 0;
2042                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2043                                           ref);
2044                                 ext4_xattr_block_csum_set(inode, new_bh);
2045                                 unlock_buffer(new_bh);
2046                                 error = ext4_handle_dirty_metadata(handle,
2047                                                                    inode,
2048                                                                    new_bh);
2049                                 if (error)
2050                                         goto cleanup_dquot;
2051                         }
2052                         mb_cache_entry_touch(ea_block_cache, ce);
2053                         mb_cache_entry_put(ea_block_cache, ce);
2054                         ce = NULL;
2055                 } else if (bs->bh && s->base == bs->bh->b_data) {
2056                         /* We were modifying this block in-place. */
2057                         ea_bdebug(bs->bh, "keeping this block");
2058                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2059                         new_bh = bs->bh;
2060                         get_bh(new_bh);
2061                 } else {
2062                         /* We need to allocate a new block */
2063                         ext4_fsblk_t goal, block;
2064
2065                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2066
2067                         goal = ext4_group_first_block_no(sb,
2068                                                 EXT4_I(inode)->i_block_group);
2069
2070                         /* non-extent files can't have physical blocks past 2^32 */
2071                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2072                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2073
2074                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2075                                                      NULL, &error);
2076                         if (error)
2077                                 goto cleanup;
2078
2079                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2080                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2081
2082                         ea_idebug(inode, "creating block %llu",
2083                                   (unsigned long long)block);
2084
2085                         new_bh = sb_getblk(sb, block);
2086                         if (unlikely(!new_bh)) {
2087                                 error = -ENOMEM;
2088 getblk_failed:
2089                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2090                                                  EXT4_FREE_BLOCKS_METADATA);
2091                                 goto cleanup;
2092                         }
2093                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2094                                                       ENTRY(header(s->base)+1));
2095                         if (error)
2096                                 goto getblk_failed;
2097                         if (ea_inode) {
2098                                 /* Drop the extra ref on ea_inode. */
2099                                 error = ext4_xattr_inode_dec_ref(handle,
2100                                                                  ea_inode);
2101                                 if (error)
2102                                         ext4_warning_inode(ea_inode,
2103                                                            "dec ref error=%d",
2104                                                            error);
2105                                 iput(ea_inode);
2106                                 ea_inode = NULL;
2107                         }
2108
2109                         lock_buffer(new_bh);
2110                         error = ext4_journal_get_create_access(handle, new_bh);
2111                         if (error) {
2112                                 unlock_buffer(new_bh);
2113                                 error = -EIO;
2114                                 goto getblk_failed;
2115                         }
2116                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2117                         ext4_xattr_block_csum_set(inode, new_bh);
2118                         set_buffer_uptodate(new_bh);
2119                         unlock_buffer(new_bh);
2120                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2121                         error = ext4_handle_dirty_metadata(handle, inode,
2122                                                            new_bh);
2123                         if (error)
2124                                 goto cleanup;
2125                 }
2126         }
2127
2128         if (old_ea_inode_quota)
2129                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2130
2131         /* Update the inode. */
2132         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2133
2134         /* Drop the previous xattr block. */
2135         if (bs->bh && bs->bh != new_bh) {
2136                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2137
2138                 ext4_xattr_release_block(handle, inode, bs->bh,
2139                                          &ea_inode_array,
2140                                          0 /* extra_credits */);
2141                 ext4_xattr_inode_array_free(ea_inode_array);
2142         }
2143         error = 0;
2144
2145 cleanup:
2146         if (ea_inode) {
2147                 int error2;
2148
2149                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2150                 if (error2)
2151                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2152                                            error2);
2153
2154                 /* If there was an error, revert the quota charge. */
2155                 if (error)
2156                         ext4_xattr_inode_free_quota(inode, ea_inode,
2157                                                     i_size_read(ea_inode));
2158                 iput(ea_inode);
2159         }
2160         if (ce)
2161                 mb_cache_entry_put(ea_block_cache, ce);
2162         brelse(new_bh);
2163         if (!(bs->bh && s->base == bs->bh->b_data))
2164                 kfree(s->base);
2165
2166         return error;
2167
2168 cleanup_dquot:
2169         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2170         goto cleanup;
2171
2172 bad_block:
2173         EXT4_ERROR_INODE(inode, "bad block %llu",
2174                          EXT4_I(inode)->i_file_acl);
2175         goto cleanup;
2176
2177 #undef header
2178 }
2179
2180 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2181                           struct ext4_xattr_ibody_find *is)
2182 {
2183         struct ext4_xattr_ibody_header *header;
2184         struct ext4_inode *raw_inode;
2185         int error;
2186
2187         if (EXT4_I(inode)->i_extra_isize == 0)
2188                 return 0;
2189         raw_inode = ext4_raw_inode(&is->iloc);
2190         header = IHDR(inode, raw_inode);
2191         is->s.base = is->s.first = IFIRST(header);
2192         is->s.here = is->s.first;
2193         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2194         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2195                 error = xattr_check_inode(inode, header, is->s.end);
2196                 if (error)
2197                         return error;
2198                 /* Find the named attribute. */
2199                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2200                                          i->name_index, i->name, 0);
2201                 if (error && error != -ENODATA)
2202                         return error;
2203                 is->s.not_found = error;
2204         }
2205         return 0;
2206 }
2207
2208 int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2209                                 struct ext4_xattr_info *i,
2210                                 struct ext4_xattr_ibody_find *is)
2211 {
2212         struct ext4_xattr_ibody_header *header;
2213         struct ext4_xattr_search *s = &is->s;
2214         int error;
2215
2216         if (EXT4_I(inode)->i_extra_isize == 0)
2217                 return -ENOSPC;
2218         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2219         if (error)
2220                 return error;
2221         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2222         if (!IS_LAST_ENTRY(s->first)) {
2223                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2224                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2225         } else {
2226                 header->h_magic = cpu_to_le32(0);
2227                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2228         }
2229         return 0;
2230 }
2231
2232 static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2233                                 struct ext4_xattr_info *i,
2234                                 struct ext4_xattr_ibody_find *is)
2235 {
2236         struct ext4_xattr_ibody_header *header;
2237         struct ext4_xattr_search *s = &is->s;
2238         int error;
2239
2240         if (EXT4_I(inode)->i_extra_isize == 0)
2241                 return -ENOSPC;
2242         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2243         if (error)
2244                 return error;
2245         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2246         if (!IS_LAST_ENTRY(s->first)) {
2247                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2248                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2249         } else {
2250                 header->h_magic = cpu_to_le32(0);
2251                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2252         }
2253         return 0;
2254 }
2255
2256 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2257                                  struct ext4_xattr_info *i)
2258 {
2259         void *value;
2260
2261         /* When e_value_inum is set the value is stored externally. */
2262         if (s->here->e_value_inum)
2263                 return 0;
2264         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2265                 return 0;
2266         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2267         return !memcmp(value, i->value, i->value_len);
2268 }
2269
2270 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2271 {
2272         struct buffer_head *bh;
2273         int error;
2274
2275         if (!EXT4_I(inode)->i_file_acl)
2276                 return NULL;
2277         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2278         if (!bh)
2279                 return ERR_PTR(-EIO);
2280         error = ext4_xattr_check_block(inode, bh);
2281         if (error) {
2282                 brelse(bh);
2283                 return ERR_PTR(error);
2284         }
2285         return bh;
2286 }
2287
2288 /*
2289  * ext4_xattr_set_handle()
2290  *
2291  * Create, replace or remove an extended attribute for this inode.  Value
2292  * is NULL to remove an existing extended attribute, and non-NULL to
2293  * either replace an existing extended attribute, or create a new extended
2294  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2295  * specify that an extended attribute must exist and must not exist
2296  * previous to the call, respectively.
2297  *
2298  * Returns 0, or a negative error number on failure.
2299  */
2300 int
2301 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2302                       const char *name, const void *value, size_t value_len,
2303                       int flags)
2304 {
2305         struct ext4_xattr_info i = {
2306                 .name_index = name_index,
2307                 .name = name,
2308                 .value = value,
2309                 .value_len = value_len,
2310                 .in_inode = 0,
2311         };
2312         struct ext4_xattr_ibody_find is = {
2313                 .s = { .not_found = -ENODATA, },
2314         };
2315         struct ext4_xattr_block_find bs = {
2316                 .s = { .not_found = -ENODATA, },
2317         };
2318         int no_expand;
2319         int error;
2320
2321         if (!name)
2322                 return -EINVAL;
2323         if (strlen(name) > 255)
2324                 return -ERANGE;
2325
2326         ext4_write_lock_xattr(inode, &no_expand);
2327
2328         /* Check journal credits under write lock. */
2329         if (ext4_handle_valid(handle)) {
2330                 struct buffer_head *bh;
2331                 int credits;
2332
2333                 bh = ext4_xattr_get_block(inode);
2334                 if (IS_ERR(bh)) {
2335                         error = PTR_ERR(bh);
2336                         goto cleanup;
2337                 }
2338
2339                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2340                                                    value_len,
2341                                                    flags & XATTR_CREATE);
2342                 brelse(bh);
2343
2344                 if (!ext4_handle_has_enough_credits(handle, credits)) {
2345                         error = -ENOSPC;
2346                         goto cleanup;
2347                 }
2348         }
2349
2350         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2351         if (error)
2352                 goto cleanup;
2353
2354         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2355                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2356                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2357                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2358         }
2359
2360         error = ext4_xattr_ibody_find(inode, &i, &is);
2361         if (error)
2362                 goto cleanup;
2363         if (is.s.not_found)
2364                 error = ext4_xattr_block_find(inode, &i, &bs);
2365         if (error)
2366                 goto cleanup;
2367         if (is.s.not_found && bs.s.not_found) {
2368                 error = -ENODATA;
2369                 if (flags & XATTR_REPLACE)
2370                         goto cleanup;
2371                 error = 0;
2372                 if (!value)
2373                         goto cleanup;
2374         } else {
2375                 error = -EEXIST;
2376                 if (flags & XATTR_CREATE)
2377                         goto cleanup;
2378         }
2379
2380         if (!value) {
2381                 if (!is.s.not_found)
2382                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2383                 else if (!bs.s.not_found)
2384                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2385         } else {
2386                 error = 0;
2387                 /* Xattr value did not change? Save us some work and bail out */
2388                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2389                         goto cleanup;
2390                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2391                         goto cleanup;
2392
2393                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2394                     (EXT4_XATTR_SIZE(i.value_len) >
2395                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2396                         i.in_inode = 1;
2397 retry_inode:
2398                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2399                 if (!error && !bs.s.not_found) {
2400                         i.value = NULL;
2401                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2402                 } else if (error == -ENOSPC) {
2403                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2404                                 brelse(bs.bh);
2405                                 bs.bh = NULL;
2406                                 error = ext4_xattr_block_find(inode, &i, &bs);
2407                                 if (error)
2408                                         goto cleanup;
2409                         }
2410                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2411                         if (!error && !is.s.not_found) {
2412                                 i.value = NULL;
2413                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2414                                                              &is);
2415                         } else if (error == -ENOSPC) {
2416                                 /*
2417                                  * Xattr does not fit in the block, store at
2418                                  * external inode if possible.
2419                                  */
2420                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2421                                     !i.in_inode) {
2422                                         i.in_inode = 1;
2423                                         goto retry_inode;
2424                                 }
2425                         }
2426                 }
2427         }
2428         if (!error) {
2429                 ext4_xattr_update_super_block(handle, inode->i_sb);
2430                 inode->i_ctime = current_time(inode);
2431                 if (!value)
2432                         no_expand = 0;
2433                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2434                 /*
2435                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2436                  * error != 0.
2437                  */
2438                 is.iloc.bh = NULL;
2439                 if (IS_SYNC(inode))
2440                         ext4_handle_sync(handle);
2441         }
2442
2443 cleanup:
2444         brelse(is.iloc.bh);
2445         brelse(bs.bh);
2446         ext4_write_unlock_xattr(inode, &no_expand);
2447         return error;
2448 }
2449
2450 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2451                            bool is_create, int *credits)
2452 {
2453         struct buffer_head *bh;
2454         int err;
2455
2456         *credits = 0;
2457
2458         if (!EXT4_SB(inode->i_sb)->s_journal)
2459                 return 0;
2460
2461         down_read(&EXT4_I(inode)->xattr_sem);
2462
2463         bh = ext4_xattr_get_block(inode);
2464         if (IS_ERR(bh)) {
2465                 err = PTR_ERR(bh);
2466         } else {
2467                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2468                                                     value_len, is_create);
2469                 brelse(bh);
2470                 err = 0;
2471         }
2472
2473         up_read(&EXT4_I(inode)->xattr_sem);
2474         return err;
2475 }
2476
2477 /*
2478  * ext4_xattr_set()
2479  *
2480  * Like ext4_xattr_set_handle, but start from an inode. This extended
2481  * attribute modification is a filesystem transaction by itself.
2482  *
2483  * Returns 0, or a negative error number on failure.
2484  */
2485 int
2486 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2487                const void *value, size_t value_len, int flags)
2488 {
2489         handle_t *handle;
2490         struct super_block *sb = inode->i_sb;
2491         int error, retries = 0;
2492         int credits;
2493
2494         error = dquot_initialize(inode);
2495         if (error)
2496                 return error;
2497
2498 retry:
2499         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2500                                        &credits);
2501         if (error)
2502                 return error;
2503
2504         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2505         if (IS_ERR(handle)) {
2506                 error = PTR_ERR(handle);
2507         } else {
2508                 int error2;
2509
2510                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2511                                               value, value_len, flags);
2512                 error2 = ext4_journal_stop(handle);
2513                 if (error == -ENOSPC &&
2514                     ext4_should_retry_alloc(sb, &retries))
2515                         goto retry;
2516                 if (error == 0)
2517                         error = error2;
2518         }
2519
2520         return error;
2521 }
2522
2523 /*
2524  * Shift the EA entries in the inode to create space for the increased
2525  * i_extra_isize.
2526  */
2527 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2528                                      int value_offs_shift, void *to,
2529                                      void *from, size_t n)
2530 {
2531         struct ext4_xattr_entry *last = entry;
2532         int new_offs;
2533
2534         /* We always shift xattr headers further thus offsets get lower */
2535         BUG_ON(value_offs_shift > 0);
2536
2537         /* Adjust the value offsets of the entries */
2538         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2539                 if (!last->e_value_inum && last->e_value_size) {
2540                         new_offs = le16_to_cpu(last->e_value_offs) +
2541                                                         value_offs_shift;
2542                         last->e_value_offs = cpu_to_le16(new_offs);
2543                 }
2544         }
2545         /* Shift the entries by n bytes */
2546         memmove(to, from, n);
2547 }
2548
2549 /*
2550  * Move xattr pointed to by 'entry' from inode into external xattr block
2551  */
2552 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2553                                     struct ext4_inode *raw_inode,
2554                                     struct ext4_xattr_entry *entry)
2555 {
2556         struct ext4_xattr_ibody_find *is = NULL;
2557         struct ext4_xattr_block_find *bs = NULL;
2558         char *buffer = NULL, *b_entry_name = NULL;
2559         size_t value_size = le32_to_cpu(entry->e_value_size);
2560         struct ext4_xattr_info i = {
2561                 .value = NULL,
2562                 .value_len = 0,
2563                 .name_index = entry->e_name_index,
2564                 .in_inode = !!entry->e_value_inum,
2565         };
2566         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2567         int error;
2568
2569         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2570         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2571         buffer = kmalloc(value_size, GFP_NOFS);
2572         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2573         if (!is || !bs || !buffer || !b_entry_name) {
2574                 error = -ENOMEM;
2575                 goto out;
2576         }
2577
2578         is->s.not_found = -ENODATA;
2579         bs->s.not_found = -ENODATA;
2580         is->iloc.bh = NULL;
2581         bs->bh = NULL;
2582
2583         /* Save the entry name and the entry value */
2584         if (entry->e_value_inum) {
2585                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2586                 if (error)
2587                         goto out;
2588         } else {
2589                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2590                 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2591         }
2592
2593         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2594         b_entry_name[entry->e_name_len] = '\0';
2595         i.name = b_entry_name;
2596
2597         error = ext4_get_inode_loc(inode, &is->iloc);
2598         if (error)
2599                 goto out;
2600
2601         error = ext4_xattr_ibody_find(inode, &i, is);
2602         if (error)
2603                 goto out;
2604
2605         /* Remove the chosen entry from the inode */
2606         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2607         if (error)
2608                 goto out;
2609
2610         i.value = buffer;
2611         i.value_len = value_size;
2612         error = ext4_xattr_block_find(inode, &i, bs);
2613         if (error)
2614                 goto out;
2615
2616         /* Add entry which was removed from the inode into the block */
2617         error = ext4_xattr_block_set(handle, inode, &i, bs);
2618         if (error)
2619                 goto out;
2620         error = 0;
2621 out:
2622         kfree(b_entry_name);
2623         kfree(buffer);
2624         if (is)
2625                 brelse(is->iloc.bh);
2626         if (bs)
2627                 brelse(bs->bh);
2628         kfree(is);
2629         kfree(bs);
2630
2631         return error;
2632 }
2633
2634 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2635                                        struct ext4_inode *raw_inode,
2636                                        int isize_diff, size_t ifree,
2637                                        size_t bfree, int *total_ino)
2638 {
2639         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2640         struct ext4_xattr_entry *small_entry;
2641         struct ext4_xattr_entry *entry;
2642         struct ext4_xattr_entry *last;
2643         unsigned int entry_size;        /* EA entry size */
2644         unsigned int total_size;        /* EA entry size + value size */
2645         unsigned int min_total_size;
2646         int error;
2647
2648         while (isize_diff > ifree) {
2649                 entry = NULL;
2650                 small_entry = NULL;
2651                 min_total_size = ~0U;
2652                 last = IFIRST(header);
2653                 /* Find the entry best suited to be pushed into EA block */
2654                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2655                         /* never move system.data out of the inode */
2656                         if ((last->e_name_len == 4) &&
2657                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2658                             !memcmp(last->e_name, "data", 4))
2659                                 continue;
2660                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2661                         if (!last->e_value_inum)
2662                                 total_size += EXT4_XATTR_SIZE(
2663                                                le32_to_cpu(last->e_value_size));
2664                         if (total_size <= bfree &&
2665                             total_size < min_total_size) {
2666                                 if (total_size + ifree < isize_diff) {
2667                                         small_entry = last;
2668                                 } else {
2669                                         entry = last;
2670                                         min_total_size = total_size;
2671                                 }
2672                         }
2673                 }
2674
2675                 if (entry == NULL) {
2676                         if (small_entry == NULL)
2677                                 return -ENOSPC;
2678                         entry = small_entry;
2679                 }
2680
2681                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2682                 total_size = entry_size;
2683                 if (!entry->e_value_inum)
2684                         total_size += EXT4_XATTR_SIZE(
2685                                               le32_to_cpu(entry->e_value_size));
2686                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2687                                                  entry);
2688                 if (error)
2689                         return error;
2690
2691                 *total_ino -= entry_size;
2692                 ifree += total_size;
2693                 bfree -= total_size;
2694         }
2695
2696         return 0;
2697 }
2698
2699 /*
2700  * Expand an inode by new_extra_isize bytes when EAs are present.
2701  * Returns 0 on success or negative error number on failure.
2702  */
2703 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2704                                struct ext4_inode *raw_inode, handle_t *handle)
2705 {
2706         struct ext4_xattr_ibody_header *header;
2707         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2708         static unsigned int mnt_count;
2709         size_t min_offs;
2710         size_t ifree, bfree;
2711         int total_ino;
2712         void *base, *end;
2713         int error = 0, tried_min_extra_isize = 0;
2714         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2715         int isize_diff; /* How much do we need to grow i_extra_isize */
2716
2717 retry:
2718         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2719         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2720                 return 0;
2721
2722         header = IHDR(inode, raw_inode);
2723
2724         /*
2725          * Check if enough free space is available in the inode to shift the
2726          * entries ahead by new_extra_isize.
2727          */
2728
2729         base = IFIRST(header);
2730         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2731         min_offs = end - base;
2732         total_ino = sizeof(struct ext4_xattr_ibody_header);
2733
2734         error = xattr_check_inode(inode, header, end);
2735         if (error)
2736                 goto cleanup;
2737
2738         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2739         if (ifree >= isize_diff)
2740                 goto shift;
2741
2742         /*
2743          * Enough free space isn't available in the inode, check if
2744          * EA block can hold new_extra_isize bytes.
2745          */
2746         if (EXT4_I(inode)->i_file_acl) {
2747                 struct buffer_head *bh;
2748
2749                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2750                 error = -EIO;
2751                 if (!bh)
2752                         goto cleanup;
2753                 error = ext4_xattr_check_block(inode, bh);
2754                 if (error) {
2755                         brelse(bh);
2756                         goto cleanup;
2757                 }
2758                 base = BHDR(bh);
2759                 end = bh->b_data + bh->b_size;
2760                 min_offs = end - base;
2761                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2762                                               NULL);
2763                 brelse(bh);
2764                 if (bfree + ifree < isize_diff) {
2765                         if (!tried_min_extra_isize && s_min_extra_isize) {
2766                                 tried_min_extra_isize++;
2767                                 new_extra_isize = s_min_extra_isize;
2768                                 goto retry;
2769                         }
2770                         error = -ENOSPC;
2771                         goto cleanup;
2772                 }
2773         } else {
2774                 bfree = inode->i_sb->s_blocksize;
2775         }
2776
2777         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2778                                             isize_diff, ifree, bfree,
2779                                             &total_ino);
2780         if (error) {
2781                 if (error == -ENOSPC && !tried_min_extra_isize &&
2782                     s_min_extra_isize) {
2783                         tried_min_extra_isize++;
2784                         new_extra_isize = s_min_extra_isize;
2785                         goto retry;
2786                 }
2787                 goto cleanup;
2788         }
2789 shift:
2790         /* Adjust the offsets and shift the remaining entries ahead */
2791         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2792                         - new_extra_isize, (void *)raw_inode +
2793                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2794                         (void *)header, total_ino);
2795         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2796
2797 cleanup:
2798         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2799                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2800                              inode->i_ino);
2801                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2802         }
2803         return error;
2804 }
2805
2806 #define EIA_INCR 16 /* must be 2^n */
2807 #define EIA_MASK (EIA_INCR - 1)
2808
2809 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2810  * If @ea_inode_array is new or full it will be grown and the old
2811  * contents copied over.
2812  */
2813 static int
2814 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2815                         struct inode *inode)
2816 {
2817         if (*ea_inode_array == NULL) {
2818                 /*
2819                  * Start with 15 inodes, so it fits into a power-of-two size.
2820                  * If *ea_inode_array is NULL, this is essentially offsetof()
2821                  */
2822                 (*ea_inode_array) =
2823                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2824                                          inodes[EIA_MASK]),
2825                                 GFP_NOFS);
2826                 if (*ea_inode_array == NULL)
2827                         return -ENOMEM;
2828                 (*ea_inode_array)->count = 0;
2829         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2830                 /* expand the array once all 15 + n * 16 slots are full */
2831                 struct ext4_xattr_inode_array *new_array = NULL;
2832                 int count = (*ea_inode_array)->count;
2833
2834                 /* if new_array is NULL, this is essentially offsetof() */
2835                 new_array = kmalloc(
2836                                 offsetof(struct ext4_xattr_inode_array,
2837                                          inodes[count + EIA_INCR]),
2838                                 GFP_NOFS);
2839                 if (new_array == NULL)
2840                         return -ENOMEM;
2841                 memcpy(new_array, *ea_inode_array,
2842                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2843                 kfree(*ea_inode_array);
2844                 *ea_inode_array = new_array;
2845         }
2846         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2847         return 0;
2848 }
2849
2850 /*
2851  * ext4_xattr_delete_inode()
2852  *
2853  * Free extended attribute resources associated with this inode. Traverse
2854  * all entries and decrement reference on any xattr inodes associated with this
2855  * inode. This is called immediately before an inode is freed. We have exclusive
2856  * access to the inode. If an orphan inode is deleted it will also release its
2857  * references on xattr block and xattr inodes.
2858  */
2859 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2860                             struct ext4_xattr_inode_array **ea_inode_array,
2861                             int extra_credits)
2862 {
2863         struct buffer_head *bh = NULL;
2864         struct ext4_xattr_ibody_header *header;
2865         struct ext4_iloc iloc = { .bh = NULL };
2866         struct ext4_xattr_entry *entry;
2867         struct inode *ea_inode;
2868         int error;
2869
2870         error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2871                                           NULL /* bh */,
2872                                           false /* dirty */,
2873                                           false /* block_csum */);
2874         if (error) {
2875                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2876                 goto cleanup;
2877         }
2878
2879         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2880             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2881
2882                 error = ext4_get_inode_loc(inode, &iloc);
2883                 if (error) {
2884                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2885                         goto cleanup;
2886                 }
2887
2888                 error = ext4_journal_get_write_access(handle, iloc.bh);
2889                 if (error) {
2890                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2891                                          error);
2892                         goto cleanup;
2893                 }
2894
2895                 header = IHDR(inode, ext4_raw_inode(&iloc));
2896                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2897                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2898                                                      IFIRST(header),
2899                                                      false /* block_csum */,
2900                                                      ea_inode_array,
2901                                                      extra_credits,
2902                                                      false /* skip_quota */);
2903         }
2904
2905         if (EXT4_I(inode)->i_file_acl) {
2906                 bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2907                 if (!bh) {
2908                         EXT4_ERROR_INODE(inode, "block %llu read error",
2909                                          EXT4_I(inode)->i_file_acl);
2910                         error = -EIO;
2911                         goto cleanup;
2912                 }
2913                 error = ext4_xattr_check_block(inode, bh);
2914                 if (error)
2915                         goto cleanup;
2916
2917                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2918                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2919                              entry = EXT4_XATTR_NEXT(entry)) {
2920                                 if (!entry->e_value_inum)
2921                                         continue;
2922                                 error = ext4_xattr_inode_iget(inode,
2923                                               le32_to_cpu(entry->e_value_inum),
2924                                               le32_to_cpu(entry->e_hash),
2925                                               &ea_inode);
2926                                 if (error)
2927                                         continue;
2928                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2929                                               le32_to_cpu(entry->e_value_size));
2930                                 iput(ea_inode);
2931                         }
2932
2933                 }
2934
2935                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2936                                          extra_credits);
2937                 /*
2938                  * Update i_file_acl value in the same transaction that releases
2939                  * block.
2940                  */
2941                 EXT4_I(inode)->i_file_acl = 0;
2942                 error = ext4_mark_inode_dirty(handle, inode);
2943                 if (error) {
2944                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2945                                          error);
2946                         goto cleanup;
2947                 }
2948         }
2949         error = 0;
2950 cleanup:
2951         brelse(iloc.bh);
2952         brelse(bh);
2953         return error;
2954 }
2955
2956 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2957 {
2958         int idx;
2959
2960         if (ea_inode_array == NULL)
2961                 return;
2962
2963         for (idx = 0; idx < ea_inode_array->count; ++idx)
2964                 iput(ea_inode_array->inodes[idx]);
2965         kfree(ea_inode_array);
2966 }
2967
2968 /*
2969  * ext4_xattr_block_cache_insert()
2970  *
2971  * Create a new entry in the extended attribute block cache, and insert
2972  * it unless such an entry is already in the cache.
2973  *
2974  * Returns 0, or a negative error number on failure.
2975  */
2976 static void
2977 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2978                               struct buffer_head *bh)
2979 {
2980         struct ext4_xattr_header *header = BHDR(bh);
2981         __u32 hash = le32_to_cpu(header->h_hash);
2982         int reusable = le32_to_cpu(header->h_refcount) <
2983                        EXT4_XATTR_REFCOUNT_MAX;
2984         int error;
2985
2986         if (!ea_block_cache)
2987                 return;
2988         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2989                                       bh->b_blocknr, reusable);
2990         if (error) {
2991                 if (error == -EBUSY)
2992                         ea_bdebug(bh, "already in cache");
2993         } else
2994                 ea_bdebug(bh, "inserting [%x]", (int)hash);
2995 }
2996
2997 /*
2998  * ext4_xattr_cmp()
2999  *
3000  * Compare two extended attribute blocks for equality.
3001  *
3002  * Returns 0 if the blocks are equal, 1 if they differ, and
3003  * a negative error number on errors.
3004  */
3005 static int
3006 ext4_xattr_cmp(struct ext4_xattr_header *header1,
3007                struct ext4_xattr_header *header2)
3008 {
3009         struct ext4_xattr_entry *entry1, *entry2;
3010
3011         entry1 = ENTRY(header1+1);
3012         entry2 = ENTRY(header2+1);
3013         while (!IS_LAST_ENTRY(entry1)) {
3014                 if (IS_LAST_ENTRY(entry2))
3015                         return 1;
3016                 if (entry1->e_hash != entry2->e_hash ||
3017                     entry1->e_name_index != entry2->e_name_index ||
3018                     entry1->e_name_len != entry2->e_name_len ||
3019                     entry1->e_value_size != entry2->e_value_size ||
3020                     entry1->e_value_inum != entry2->e_value_inum ||
3021                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3022                         return 1;
3023                 if (!entry1->e_value_inum &&
3024                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3025                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3026                            le32_to_cpu(entry1->e_value_size)))
3027                         return 1;
3028
3029                 entry1 = EXT4_XATTR_NEXT(entry1);
3030                 entry2 = EXT4_XATTR_NEXT(entry2);
3031         }
3032         if (!IS_LAST_ENTRY(entry2))
3033                 return 1;
3034         return 0;
3035 }
3036
3037 /*
3038  * ext4_xattr_block_cache_find()
3039  *
3040  * Find an identical extended attribute block.
3041  *
3042  * Returns a pointer to the block found, or NULL if such a block was
3043  * not found or an error occurred.
3044  */
3045 static struct buffer_head *
3046 ext4_xattr_block_cache_find(struct inode *inode,
3047                             struct ext4_xattr_header *header,
3048                             struct mb_cache_entry **pce)
3049 {
3050         __u32 hash = le32_to_cpu(header->h_hash);
3051         struct mb_cache_entry *ce;
3052         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3053
3054         if (!ea_block_cache)
3055                 return NULL;
3056         if (!header->h_hash)
3057                 return NULL;  /* never share */
3058         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3059         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3060         while (ce) {
3061                 struct buffer_head *bh;
3062
3063                 bh = sb_bread(inode->i_sb, ce->e_value);
3064                 if (!bh) {
3065                         EXT4_ERROR_INODE(inode, "block %lu read error",
3066                                          (unsigned long)ce->e_value);
3067                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3068                         *pce = ce;
3069                         return bh;
3070                 }
3071                 brelse(bh);
3072                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3073         }
3074         return NULL;
3075 }
3076
3077 #define NAME_HASH_SHIFT 5
3078 #define VALUE_HASH_SHIFT 16
3079
3080 /*
3081  * ext4_xattr_hash_entry()
3082  *
3083  * Compute the hash of an extended attribute.
3084  */
3085 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3086                                     size_t value_count)
3087 {
3088         __u32 hash = 0;
3089
3090         while (name_len--) {
3091                 hash = (hash << NAME_HASH_SHIFT) ^
3092                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3093                        *name++;
3094         }
3095         while (value_count--) {
3096                 hash = (hash << VALUE_HASH_SHIFT) ^
3097                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3098                        le32_to_cpu(*value++);
3099         }
3100         return cpu_to_le32(hash);
3101 }
3102
3103 #undef NAME_HASH_SHIFT
3104 #undef VALUE_HASH_SHIFT
3105
3106 #define BLOCK_HASH_SHIFT 16
3107
3108 /*
3109  * ext4_xattr_rehash()
3110  *
3111  * Re-compute the extended attribute hash value after an entry has changed.
3112  */
3113 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3114 {
3115         struct ext4_xattr_entry *here;
3116         __u32 hash = 0;
3117
3118         here = ENTRY(header+1);
3119         while (!IS_LAST_ENTRY(here)) {
3120                 if (!here->e_hash) {
3121                         /* Block is not shared if an entry's hash value == 0 */
3122                         hash = 0;
3123                         break;
3124                 }
3125                 hash = (hash << BLOCK_HASH_SHIFT) ^
3126                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3127                        le32_to_cpu(here->e_hash);
3128                 here = EXT4_XATTR_NEXT(here);
3129         }
3130         header->h_hash = cpu_to_le32(hash);
3131 }
3132
3133 #undef BLOCK_HASH_SHIFT
3134
3135 #define HASH_BUCKET_BITS        10
3136
3137 struct mb_cache *
3138 ext4_xattr_create_cache(void)
3139 {
3140         return mb_cache_create(HASH_BUCKET_BITS);
3141 }
3142
3143 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3144 {
3145         if (cache)
3146                 mb_cache_destroy(cache);
3147 }
3148