GNU Linux-libre 6.0.2-gnu
[releases.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         [EXT4_XATTR_INDEX_HURD]              = &ext4_xattr_hurd_handler,
97 };
98
99 const struct xattr_handler *ext4_xattr_handlers[] = {
100         &ext4_xattr_user_handler,
101         &ext4_xattr_trusted_handler,
102 #ifdef CONFIG_EXT4_FS_POSIX_ACL
103         &posix_acl_access_xattr_handler,
104         &posix_acl_default_xattr_handler,
105 #endif
106 #ifdef CONFIG_EXT4_FS_SECURITY
107         &ext4_xattr_security_handler,
108 #endif
109         &ext4_xattr_hurd_handler,
110         NULL
111 };
112
113 #define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
114                                 inode->i_sb->s_fs_info)->s_ea_block_cache)
115
116 #define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
117                                 inode->i_sb->s_fs_info)->s_ea_inode_cache)
118
119 static int
120 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
121                         struct inode *inode);
122
123 #ifdef CONFIG_LOCKDEP
124 void ext4_xattr_inode_set_class(struct inode *ea_inode)
125 {
126         lockdep_set_subclass(&ea_inode->i_rwsem, 1);
127 }
128 #endif
129
130 static __le32 ext4_xattr_block_csum(struct inode *inode,
131                                     sector_t block_nr,
132                                     struct ext4_xattr_header *hdr)
133 {
134         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
135         __u32 csum;
136         __le64 dsk_block_nr = cpu_to_le64(block_nr);
137         __u32 dummy_csum = 0;
138         int offset = offsetof(struct ext4_xattr_header, h_checksum);
139
140         csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
141                            sizeof(dsk_block_nr));
142         csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
143         csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
144         offset += sizeof(dummy_csum);
145         csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
146                            EXT4_BLOCK_SIZE(inode->i_sb) - offset);
147
148         return cpu_to_le32(csum);
149 }
150
151 static int ext4_xattr_block_csum_verify(struct inode *inode,
152                                         struct buffer_head *bh)
153 {
154         struct ext4_xattr_header *hdr = BHDR(bh);
155         int ret = 1;
156
157         if (ext4_has_metadata_csum(inode->i_sb)) {
158                 lock_buffer(bh);
159                 ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
160                                                         bh->b_blocknr, hdr));
161                 unlock_buffer(bh);
162         }
163         return ret;
164 }
165
166 static void ext4_xattr_block_csum_set(struct inode *inode,
167                                       struct buffer_head *bh)
168 {
169         if (ext4_has_metadata_csum(inode->i_sb))
170                 BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
171                                                 bh->b_blocknr, BHDR(bh));
172 }
173
174 static inline const struct xattr_handler *
175 ext4_xattr_handler(int name_index)
176 {
177         const struct xattr_handler *handler = NULL;
178
179         if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
180                 handler = ext4_xattr_handler_map[name_index];
181         return handler;
182 }
183
184 static int
185 ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
186                          void *value_start)
187 {
188         struct ext4_xattr_entry *e = entry;
189
190         /* Find the end of the names list */
191         while (!IS_LAST_ENTRY(e)) {
192                 struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
193                 if ((void *)next >= end)
194                         return -EFSCORRUPTED;
195                 if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
196                         return -EFSCORRUPTED;
197                 e = next;
198         }
199
200         /* Check the values */
201         while (!IS_LAST_ENTRY(entry)) {
202                 u32 size = le32_to_cpu(entry->e_value_size);
203
204                 if (size > EXT4_XATTR_SIZE_MAX)
205                         return -EFSCORRUPTED;
206
207                 if (size != 0 && entry->e_value_inum == 0) {
208                         u16 offs = le16_to_cpu(entry->e_value_offs);
209                         void *value;
210
211                         /*
212                          * The value cannot overlap the names, and the value
213                          * with padding cannot extend beyond 'end'.  Check both
214                          * the padded and unpadded sizes, since the size may
215                          * overflow to 0 when adding padding.
216                          */
217                         if (offs > end - value_start)
218                                 return -EFSCORRUPTED;
219                         value = value_start + offs;
220                         if (value < (void *)e + sizeof(u32) ||
221                             size > end - value ||
222                             EXT4_XATTR_SIZE(size) > end - value)
223                                 return -EFSCORRUPTED;
224                 }
225                 entry = EXT4_XATTR_NEXT(entry);
226         }
227
228         return 0;
229 }
230
231 static inline int
232 __ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
233                          const char *function, unsigned int line)
234 {
235         int error = -EFSCORRUPTED;
236
237         if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
238             BHDR(bh)->h_blocks != cpu_to_le32(1))
239                 goto errout;
240         if (buffer_verified(bh))
241                 return 0;
242
243         error = -EFSBADCRC;
244         if (!ext4_xattr_block_csum_verify(inode, bh))
245                 goto errout;
246         error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
247                                          bh->b_data);
248 errout:
249         if (error)
250                 __ext4_error_inode(inode, function, line, 0, -error,
251                                    "corrupted xattr block %llu",
252                                    (unsigned long long) bh->b_blocknr);
253         else
254                 set_buffer_verified(bh);
255         return error;
256 }
257
258 #define ext4_xattr_check_block(inode, bh) \
259         __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
260
261
262 static int
263 __xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
264                          void *end, const char *function, unsigned int line)
265 {
266         int error = -EFSCORRUPTED;
267
268         if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
269             (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
270                 goto errout;
271         error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
272 errout:
273         if (error)
274                 __ext4_error_inode(inode, function, line, 0, -error,
275                                    "corrupted in-inode xattr");
276         return error;
277 }
278
279 #define xattr_check_inode(inode, header, end) \
280         __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
281
282 static int
283 xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
284                  void *end, int name_index, const char *name, int sorted)
285 {
286         struct ext4_xattr_entry *entry, *next;
287         size_t name_len;
288         int cmp = 1;
289
290         if (name == NULL)
291                 return -EINVAL;
292         name_len = strlen(name);
293         for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
294                 next = EXT4_XATTR_NEXT(entry);
295                 if ((void *) next >= end) {
296                         EXT4_ERROR_INODE(inode, "corrupted xattr entries");
297                         return -EFSCORRUPTED;
298                 }
299                 cmp = name_index - entry->e_name_index;
300                 if (!cmp)
301                         cmp = name_len - entry->e_name_len;
302                 if (!cmp)
303                         cmp = memcmp(name, entry->e_name, name_len);
304                 if (cmp <= 0 && (sorted || cmp == 0))
305                         break;
306         }
307         *pentry = entry;
308         return cmp ? -ENODATA : 0;
309 }
310
311 static u32
312 ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
313 {
314         return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
315 }
316
317 static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
318 {
319         return ((u64)ea_inode->i_ctime.tv_sec << 32) |
320                 (u32) inode_peek_iversion_raw(ea_inode);
321 }
322
323 static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
324 {
325         ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
326         inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
327 }
328
329 static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
330 {
331         return (u32)ea_inode->i_atime.tv_sec;
332 }
333
334 static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
335 {
336         ea_inode->i_atime.tv_sec = hash;
337 }
338
339 /*
340  * Read the EA value from an inode.
341  */
342 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
343 {
344         int blocksize = 1 << ea_inode->i_blkbits;
345         int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
346         int tail_size = (size % blocksize) ?: blocksize;
347         struct buffer_head *bhs_inline[8];
348         struct buffer_head **bhs = bhs_inline;
349         int i, ret;
350
351         if (bh_count > ARRAY_SIZE(bhs_inline)) {
352                 bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
353                 if (!bhs)
354                         return -ENOMEM;
355         }
356
357         ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
358                                true /* wait */, bhs);
359         if (ret)
360                 goto free_bhs;
361
362         for (i = 0; i < bh_count; i++) {
363                 /* There shouldn't be any holes in ea_inode. */
364                 if (!bhs[i]) {
365                         ret = -EFSCORRUPTED;
366                         goto put_bhs;
367                 }
368                 memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
369                        i < bh_count - 1 ? blocksize : tail_size);
370         }
371         ret = 0;
372 put_bhs:
373         for (i = 0; i < bh_count; i++)
374                 brelse(bhs[i]);
375 free_bhs:
376         if (bhs != bhs_inline)
377                 kfree(bhs);
378         return ret;
379 }
380
381 #define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
382
383 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
384                                  u32 ea_inode_hash, struct inode **ea_inode)
385 {
386         struct inode *inode;
387         int err;
388
389         inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
390         if (IS_ERR(inode)) {
391                 err = PTR_ERR(inode);
392                 ext4_error(parent->i_sb,
393                            "error while reading EA inode %lu err=%d", ea_ino,
394                            err);
395                 return err;
396         }
397
398         if (is_bad_inode(inode)) {
399                 ext4_error(parent->i_sb,
400                            "error while reading EA inode %lu is_bad_inode",
401                            ea_ino);
402                 err = -EIO;
403                 goto error;
404         }
405
406         if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
407                 ext4_error(parent->i_sb,
408                            "EA inode %lu does not have EXT4_EA_INODE_FL flag",
409                             ea_ino);
410                 err = -EINVAL;
411                 goto error;
412         }
413
414         ext4_xattr_inode_set_class(inode);
415
416         /*
417          * Check whether this is an old Lustre-style xattr inode. Lustre
418          * implementation does not have hash validation, rather it has a
419          * backpointer from ea_inode to the parent inode.
420          */
421         if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
422             EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
423             inode->i_generation == parent->i_generation) {
424                 ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
425                 ext4_xattr_inode_set_ref(inode, 1);
426         } else {
427                 inode_lock(inode);
428                 inode->i_flags |= S_NOQUOTA;
429                 inode_unlock(inode);
430         }
431
432         *ea_inode = inode;
433         return 0;
434 error:
435         iput(inode);
436         return err;
437 }
438
439 /* Remove entry from mbcache when EA inode is getting evicted */
440 void ext4_evict_ea_inode(struct inode *inode)
441 {
442         struct mb_cache_entry *oe;
443
444         if (!EA_INODE_CACHE(inode))
445                 return;
446         /* Wait for entry to get unused so that we can remove it */
447         while ((oe = mb_cache_entry_delete_or_get(EA_INODE_CACHE(inode),
448                         ext4_xattr_inode_get_hash(inode), inode->i_ino))) {
449                 mb_cache_entry_wait_unused(oe);
450                 mb_cache_entry_put(EA_INODE_CACHE(inode), oe);
451         }
452 }
453
454 static int
455 ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
456                                struct ext4_xattr_entry *entry, void *buffer,
457                                size_t size)
458 {
459         u32 hash;
460
461         /* Verify stored hash matches calculated hash. */
462         hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
463         if (hash != ext4_xattr_inode_get_hash(ea_inode))
464                 return -EFSCORRUPTED;
465
466         if (entry) {
467                 __le32 e_hash, tmp_data;
468
469                 /* Verify entry hash. */
470                 tmp_data = cpu_to_le32(hash);
471                 e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
472                                                &tmp_data, 1);
473                 if (e_hash != entry->e_hash)
474                         return -EFSCORRUPTED;
475         }
476         return 0;
477 }
478
479 /*
480  * Read xattr value from the EA inode.
481  */
482 static int
483 ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
484                      void *buffer, size_t size)
485 {
486         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
487         struct inode *ea_inode;
488         int err;
489
490         err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
491                                     le32_to_cpu(entry->e_hash), &ea_inode);
492         if (err) {
493                 ea_inode = NULL;
494                 goto out;
495         }
496
497         if (i_size_read(ea_inode) != size) {
498                 ext4_warning_inode(ea_inode,
499                                    "ea_inode file size=%llu entry size=%zu",
500                                    i_size_read(ea_inode), size);
501                 err = -EFSCORRUPTED;
502                 goto out;
503         }
504
505         err = ext4_xattr_inode_read(ea_inode, buffer, size);
506         if (err)
507                 goto out;
508
509         if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
510                 err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
511                                                      size);
512                 if (err) {
513                         ext4_warning_inode(ea_inode,
514                                            "EA inode hash validation failed");
515                         goto out;
516                 }
517
518                 if (ea_inode_cache)
519                         mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
520                                         ext4_xattr_inode_get_hash(ea_inode),
521                                         ea_inode->i_ino, true /* reusable */);
522         }
523 out:
524         iput(ea_inode);
525         return err;
526 }
527
528 static int
529 ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
530                      void *buffer, size_t buffer_size)
531 {
532         struct buffer_head *bh = NULL;
533         struct ext4_xattr_entry *entry;
534         size_t size;
535         void *end;
536         int error;
537         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
538
539         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
540                   name_index, name, buffer, (long)buffer_size);
541
542         if (!EXT4_I(inode)->i_file_acl)
543                 return -ENODATA;
544         ea_idebug(inode, "reading block %llu",
545                   (unsigned long long)EXT4_I(inode)->i_file_acl);
546         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
547         if (IS_ERR(bh))
548                 return PTR_ERR(bh);
549         ea_bdebug(bh, "b_count=%d, refcount=%d",
550                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
551         error = ext4_xattr_check_block(inode, bh);
552         if (error)
553                 goto cleanup;
554         ext4_xattr_block_cache_insert(ea_block_cache, bh);
555         entry = BFIRST(bh);
556         end = bh->b_data + bh->b_size;
557         error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
558         if (error)
559                 goto cleanup;
560         size = le32_to_cpu(entry->e_value_size);
561         error = -ERANGE;
562         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
563                 goto cleanup;
564         if (buffer) {
565                 if (size > buffer_size)
566                         goto cleanup;
567                 if (entry->e_value_inum) {
568                         error = ext4_xattr_inode_get(inode, entry, buffer,
569                                                      size);
570                         if (error)
571                                 goto cleanup;
572                 } else {
573                         u16 offset = le16_to_cpu(entry->e_value_offs);
574                         void *p = bh->b_data + offset;
575
576                         if (unlikely(p + size > end))
577                                 goto cleanup;
578                         memcpy(buffer, p, size);
579                 }
580         }
581         error = size;
582
583 cleanup:
584         brelse(bh);
585         return error;
586 }
587
588 int
589 ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
590                      void *buffer, size_t buffer_size)
591 {
592         struct ext4_xattr_ibody_header *header;
593         struct ext4_xattr_entry *entry;
594         struct ext4_inode *raw_inode;
595         struct ext4_iloc iloc;
596         size_t size;
597         void *end;
598         int error;
599
600         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
601                 return -ENODATA;
602         error = ext4_get_inode_loc(inode, &iloc);
603         if (error)
604                 return error;
605         raw_inode = ext4_raw_inode(&iloc);
606         header = IHDR(inode, raw_inode);
607         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
608         error = xattr_check_inode(inode, header, end);
609         if (error)
610                 goto cleanup;
611         entry = IFIRST(header);
612         error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
613         if (error)
614                 goto cleanup;
615         size = le32_to_cpu(entry->e_value_size);
616         error = -ERANGE;
617         if (unlikely(size > EXT4_XATTR_SIZE_MAX))
618                 goto cleanup;
619         if (buffer) {
620                 if (size > buffer_size)
621                         goto cleanup;
622                 if (entry->e_value_inum) {
623                         error = ext4_xattr_inode_get(inode, entry, buffer,
624                                                      size);
625                         if (error)
626                                 goto cleanup;
627                 } else {
628                         u16 offset = le16_to_cpu(entry->e_value_offs);
629                         void *p = (void *)IFIRST(header) + offset;
630
631                         if (unlikely(p + size > end))
632                                 goto cleanup;
633                         memcpy(buffer, p, size);
634                 }
635         }
636         error = size;
637
638 cleanup:
639         brelse(iloc.bh);
640         return error;
641 }
642
643 /*
644  * ext4_xattr_get()
645  *
646  * Copy an extended attribute into the buffer
647  * provided, or compute the buffer size required.
648  * Buffer is NULL to compute the size of the buffer required.
649  *
650  * Returns a negative error number on failure, or the number of bytes
651  * used / required on success.
652  */
653 int
654 ext4_xattr_get(struct inode *inode, int name_index, const char *name,
655                void *buffer, size_t buffer_size)
656 {
657         int error;
658
659         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
660                 return -EIO;
661
662         if (strlen(name) > 255)
663                 return -ERANGE;
664
665         down_read(&EXT4_I(inode)->xattr_sem);
666         error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
667                                      buffer_size);
668         if (error == -ENODATA)
669                 error = ext4_xattr_block_get(inode, name_index, name, buffer,
670                                              buffer_size);
671         up_read(&EXT4_I(inode)->xattr_sem);
672         return error;
673 }
674
675 static int
676 ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
677                         char *buffer, size_t buffer_size)
678 {
679         size_t rest = buffer_size;
680
681         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
682                 const struct xattr_handler *handler =
683                         ext4_xattr_handler(entry->e_name_index);
684
685                 if (handler && (!handler->list || handler->list(dentry))) {
686                         const char *prefix = handler->prefix ?: handler->name;
687                         size_t prefix_len = strlen(prefix);
688                         size_t size = prefix_len + entry->e_name_len + 1;
689
690                         if (buffer) {
691                                 if (size > rest)
692                                         return -ERANGE;
693                                 memcpy(buffer, prefix, prefix_len);
694                                 buffer += prefix_len;
695                                 memcpy(buffer, entry->e_name, entry->e_name_len);
696                                 buffer += entry->e_name_len;
697                                 *buffer++ = 0;
698                         }
699                         rest -= size;
700                 }
701         }
702         return buffer_size - rest;  /* total size */
703 }
704
705 static int
706 ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
707 {
708         struct inode *inode = d_inode(dentry);
709         struct buffer_head *bh = NULL;
710         int error;
711
712         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
713                   buffer, (long)buffer_size);
714
715         if (!EXT4_I(inode)->i_file_acl)
716                 return 0;
717         ea_idebug(inode, "reading block %llu",
718                   (unsigned long long)EXT4_I(inode)->i_file_acl);
719         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
720         if (IS_ERR(bh))
721                 return PTR_ERR(bh);
722         ea_bdebug(bh, "b_count=%d, refcount=%d",
723                 atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
724         error = ext4_xattr_check_block(inode, bh);
725         if (error)
726                 goto cleanup;
727         ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
728         error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
729                                         buffer_size);
730 cleanup:
731         brelse(bh);
732         return error;
733 }
734
735 static int
736 ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
737 {
738         struct inode *inode = d_inode(dentry);
739         struct ext4_xattr_ibody_header *header;
740         struct ext4_inode *raw_inode;
741         struct ext4_iloc iloc;
742         void *end;
743         int error;
744
745         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
746                 return 0;
747         error = ext4_get_inode_loc(inode, &iloc);
748         if (error)
749                 return error;
750         raw_inode = ext4_raw_inode(&iloc);
751         header = IHDR(inode, raw_inode);
752         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
753         error = xattr_check_inode(inode, header, end);
754         if (error)
755                 goto cleanup;
756         error = ext4_xattr_list_entries(dentry, IFIRST(header),
757                                         buffer, buffer_size);
758
759 cleanup:
760         brelse(iloc.bh);
761         return error;
762 }
763
764 /*
765  * Inode operation listxattr()
766  *
767  * d_inode(dentry)->i_rwsem: don't care
768  *
769  * Copy a list of attribute names into the buffer
770  * provided, or compute the buffer size required.
771  * Buffer is NULL to compute the size of the buffer required.
772  *
773  * Returns a negative error number on failure, or the number of bytes
774  * used / required on success.
775  */
776 ssize_t
777 ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
778 {
779         int ret, ret2;
780
781         down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
782         ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
783         if (ret < 0)
784                 goto errout;
785         if (buffer) {
786                 buffer += ret;
787                 buffer_size -= ret;
788         }
789         ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
790         if (ret < 0)
791                 goto errout;
792         ret += ret2;
793 errout:
794         up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
795         return ret;
796 }
797
798 /*
799  * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
800  * not set, set it.
801  */
802 static void ext4_xattr_update_super_block(handle_t *handle,
803                                           struct super_block *sb)
804 {
805         if (ext4_has_feature_xattr(sb))
806                 return;
807
808         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
809         if (ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
810                                           EXT4_JTR_NONE) == 0) {
811                 lock_buffer(EXT4_SB(sb)->s_sbh);
812                 ext4_set_feature_xattr(sb);
813                 ext4_superblock_csum_set(sb);
814                 unlock_buffer(EXT4_SB(sb)->s_sbh);
815                 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
816         }
817 }
818
819 int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
820 {
821         struct ext4_iloc iloc = { .bh = NULL };
822         struct buffer_head *bh = NULL;
823         struct ext4_inode *raw_inode;
824         struct ext4_xattr_ibody_header *header;
825         struct ext4_xattr_entry *entry;
826         qsize_t ea_inode_refs = 0;
827         void *end;
828         int ret;
829
830         lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
831
832         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
833                 ret = ext4_get_inode_loc(inode, &iloc);
834                 if (ret)
835                         goto out;
836                 raw_inode = ext4_raw_inode(&iloc);
837                 header = IHDR(inode, raw_inode);
838                 end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
839                 ret = xattr_check_inode(inode, header, end);
840                 if (ret)
841                         goto out;
842
843                 for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
844                      entry = EXT4_XATTR_NEXT(entry))
845                         if (entry->e_value_inum)
846                                 ea_inode_refs++;
847         }
848
849         if (EXT4_I(inode)->i_file_acl) {
850                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
851                 if (IS_ERR(bh)) {
852                         ret = PTR_ERR(bh);
853                         bh = NULL;
854                         goto out;
855                 }
856
857                 ret = ext4_xattr_check_block(inode, bh);
858                 if (ret)
859                         goto out;
860
861                 for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
862                      entry = EXT4_XATTR_NEXT(entry))
863                         if (entry->e_value_inum)
864                                 ea_inode_refs++;
865         }
866         *usage = ea_inode_refs + 1;
867         ret = 0;
868 out:
869         brelse(iloc.bh);
870         brelse(bh);
871         return ret;
872 }
873
874 static inline size_t round_up_cluster(struct inode *inode, size_t length)
875 {
876         struct super_block *sb = inode->i_sb;
877         size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
878                                     inode->i_blkbits);
879         size_t mask = ~(cluster_size - 1);
880
881         return (length + cluster_size - 1) & mask;
882 }
883
884 static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
885 {
886         int err;
887
888         err = dquot_alloc_inode(inode);
889         if (err)
890                 return err;
891         err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
892         if (err)
893                 dquot_free_inode(inode);
894         return err;
895 }
896
897 static void ext4_xattr_inode_free_quota(struct inode *parent,
898                                         struct inode *ea_inode,
899                                         size_t len)
900 {
901         if (ea_inode &&
902             ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
903                 return;
904         dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
905         dquot_free_inode(parent);
906 }
907
908 int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
909                              struct buffer_head *block_bh, size_t value_len,
910                              bool is_create)
911 {
912         int credits;
913         int blocks;
914
915         /*
916          * 1) Owner inode update
917          * 2) Ref count update on old xattr block
918          * 3) new xattr block
919          * 4) block bitmap update for new xattr block
920          * 5) group descriptor for new xattr block
921          * 6) block bitmap update for old xattr block
922          * 7) group descriptor for old block
923          *
924          * 6 & 7 can happen if we have two racing threads T_a and T_b
925          * which are each trying to set an xattr on inodes I_a and I_b
926          * which were both initially sharing an xattr block.
927          */
928         credits = 7;
929
930         /* Quota updates. */
931         credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
932
933         /*
934          * In case of inline data, we may push out the data to a block,
935          * so we need to reserve credits for this eventuality
936          */
937         if (inode && ext4_has_inline_data(inode))
938                 credits += ext4_writepage_trans_blocks(inode) + 1;
939
940         /* We are done if ea_inode feature is not enabled. */
941         if (!ext4_has_feature_ea_inode(sb))
942                 return credits;
943
944         /* New ea_inode, inode map, block bitmap, group descriptor. */
945         credits += 4;
946
947         /* Data blocks. */
948         blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
949
950         /* Indirection block or one level of extent tree. */
951         blocks += 1;
952
953         /* Block bitmap and group descriptor updates for each block. */
954         credits += blocks * 2;
955
956         /* Blocks themselves. */
957         credits += blocks;
958
959         if (!is_create) {
960                 /* Dereference ea_inode holding old xattr value.
961                  * Old ea_inode, inode map, block bitmap, group descriptor.
962                  */
963                 credits += 4;
964
965                 /* Data blocks for old ea_inode. */
966                 blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
967
968                 /* Indirection block or one level of extent tree for old
969                  * ea_inode.
970                  */
971                 blocks += 1;
972
973                 /* Block bitmap and group descriptor updates for each block. */
974                 credits += blocks * 2;
975         }
976
977         /* We may need to clone the existing xattr block in which case we need
978          * to increment ref counts for existing ea_inodes referenced by it.
979          */
980         if (block_bh) {
981                 struct ext4_xattr_entry *entry = BFIRST(block_bh);
982
983                 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
984                         if (entry->e_value_inum)
985                                 /* Ref count update on ea_inode. */
986                                 credits += 1;
987         }
988         return credits;
989 }
990
991 static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
992                                        int ref_change)
993 {
994         struct ext4_iloc iloc;
995         s64 ref_count;
996         int ret;
997
998         inode_lock(ea_inode);
999
1000         ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1001         if (ret)
1002                 goto out;
1003
1004         ref_count = ext4_xattr_inode_get_ref(ea_inode);
1005         ref_count += ref_change;
1006         ext4_xattr_inode_set_ref(ea_inode, ref_count);
1007
1008         if (ref_change > 0) {
1009                 WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1010                           ea_inode->i_ino, ref_count);
1011
1012                 if (ref_count == 1) {
1013                         WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1014                                   ea_inode->i_ino, ea_inode->i_nlink);
1015
1016                         set_nlink(ea_inode, 1);
1017                         ext4_orphan_del(handle, ea_inode);
1018                 }
1019         } else {
1020                 WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1021                           ea_inode->i_ino, ref_count);
1022
1023                 if (ref_count == 0) {
1024                         WARN_ONCE(ea_inode->i_nlink != 1,
1025                                   "EA inode %lu i_nlink=%u",
1026                                   ea_inode->i_ino, ea_inode->i_nlink);
1027
1028                         clear_nlink(ea_inode);
1029                         ext4_orphan_add(handle, ea_inode);
1030                 }
1031         }
1032
1033         ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1034         if (ret)
1035                 ext4_warning_inode(ea_inode,
1036                                    "ext4_mark_iloc_dirty() failed ret=%d", ret);
1037 out:
1038         inode_unlock(ea_inode);
1039         return ret;
1040 }
1041
1042 static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1043 {
1044         return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1045 }
1046
1047 static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1048 {
1049         return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1050 }
1051
1052 static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1053                                         struct ext4_xattr_entry *first)
1054 {
1055         struct inode *ea_inode;
1056         struct ext4_xattr_entry *entry;
1057         struct ext4_xattr_entry *failed_entry;
1058         unsigned int ea_ino;
1059         int err, saved_err;
1060
1061         for (entry = first; !IS_LAST_ENTRY(entry);
1062              entry = EXT4_XATTR_NEXT(entry)) {
1063                 if (!entry->e_value_inum)
1064                         continue;
1065                 ea_ino = le32_to_cpu(entry->e_value_inum);
1066                 err = ext4_xattr_inode_iget(parent, ea_ino,
1067                                             le32_to_cpu(entry->e_hash),
1068                                             &ea_inode);
1069                 if (err)
1070                         goto cleanup;
1071                 err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1072                 if (err) {
1073                         ext4_warning_inode(ea_inode, "inc ref error %d", err);
1074                         iput(ea_inode);
1075                         goto cleanup;
1076                 }
1077                 iput(ea_inode);
1078         }
1079         return 0;
1080
1081 cleanup:
1082         saved_err = err;
1083         failed_entry = entry;
1084
1085         for (entry = first; entry != failed_entry;
1086              entry = EXT4_XATTR_NEXT(entry)) {
1087                 if (!entry->e_value_inum)
1088                         continue;
1089                 ea_ino = le32_to_cpu(entry->e_value_inum);
1090                 err = ext4_xattr_inode_iget(parent, ea_ino,
1091                                             le32_to_cpu(entry->e_hash),
1092                                             &ea_inode);
1093                 if (err) {
1094                         ext4_warning(parent->i_sb,
1095                                      "cleanup ea_ino %u iget error %d", ea_ino,
1096                                      err);
1097                         continue;
1098                 }
1099                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1100                 if (err)
1101                         ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1102                                            err);
1103                 iput(ea_inode);
1104         }
1105         return saved_err;
1106 }
1107
1108 static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1109                         struct buffer_head *bh, bool block_csum, bool dirty)
1110 {
1111         int error;
1112
1113         if (bh && dirty) {
1114                 if (block_csum)
1115                         ext4_xattr_block_csum_set(inode, bh);
1116                 error = ext4_handle_dirty_metadata(handle, NULL, bh);
1117                 if (error) {
1118                         ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1119                                      error);
1120                         return error;
1121                 }
1122         }
1123         return 0;
1124 }
1125
1126 static void
1127 ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1128                              struct buffer_head *bh,
1129                              struct ext4_xattr_entry *first, bool block_csum,
1130                              struct ext4_xattr_inode_array **ea_inode_array,
1131                              int extra_credits, bool skip_quota)
1132 {
1133         struct inode *ea_inode;
1134         struct ext4_xattr_entry *entry;
1135         bool dirty = false;
1136         unsigned int ea_ino;
1137         int err;
1138         int credits;
1139
1140         /* One credit for dec ref on ea_inode, one for orphan list addition, */
1141         credits = 2 + extra_credits;
1142
1143         for (entry = first; !IS_LAST_ENTRY(entry);
1144              entry = EXT4_XATTR_NEXT(entry)) {
1145                 if (!entry->e_value_inum)
1146                         continue;
1147                 ea_ino = le32_to_cpu(entry->e_value_inum);
1148                 err = ext4_xattr_inode_iget(parent, ea_ino,
1149                                             le32_to_cpu(entry->e_hash),
1150                                             &ea_inode);
1151                 if (err)
1152                         continue;
1153
1154                 err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1155                 if (err) {
1156                         ext4_warning_inode(ea_inode,
1157                                            "Expand inode array err=%d", err);
1158                         iput(ea_inode);
1159                         continue;
1160                 }
1161
1162                 err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1163                         ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1164                         ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1165                                               dirty));
1166                 if (err < 0) {
1167                         ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1168                                            err);
1169                         continue;
1170                 }
1171                 if (err > 0) {
1172                         err = ext4_journal_get_write_access(handle,
1173                                         parent->i_sb, bh, EXT4_JTR_NONE);
1174                         if (err) {
1175                                 ext4_warning_inode(ea_inode,
1176                                                 "Re-get write access err=%d",
1177                                                 err);
1178                                 continue;
1179                         }
1180                 }
1181
1182                 err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1183                 if (err) {
1184                         ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1185                                            err);
1186                         continue;
1187                 }
1188
1189                 if (!skip_quota)
1190                         ext4_xattr_inode_free_quota(parent, ea_inode,
1191                                               le32_to_cpu(entry->e_value_size));
1192
1193                 /*
1194                  * Forget about ea_inode within the same transaction that
1195                  * decrements the ref count. This avoids duplicate decrements in
1196                  * case the rest of the work spills over to subsequent
1197                  * transactions.
1198                  */
1199                 entry->e_value_inum = 0;
1200                 entry->e_value_size = 0;
1201
1202                 dirty = true;
1203         }
1204
1205         if (dirty) {
1206                 /*
1207                  * Note that we are deliberately skipping csum calculation for
1208                  * the final update because we do not expect any journal
1209                  * restarts until xattr block is freed.
1210                  */
1211
1212                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1213                 if (err)
1214                         ext4_warning_inode(parent,
1215                                            "handle dirty metadata err=%d", err);
1216         }
1217 }
1218
1219 /*
1220  * Release the xattr block BH: If the reference count is > 1, decrement it;
1221  * otherwise free the block.
1222  */
1223 static void
1224 ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1225                          struct buffer_head *bh,
1226                          struct ext4_xattr_inode_array **ea_inode_array,
1227                          int extra_credits)
1228 {
1229         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1230         u32 hash, ref;
1231         int error = 0;
1232
1233         BUFFER_TRACE(bh, "get_write_access");
1234         error = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1235                                               EXT4_JTR_NONE);
1236         if (error)
1237                 goto out;
1238
1239 retry_ref:
1240         lock_buffer(bh);
1241         hash = le32_to_cpu(BHDR(bh)->h_hash);
1242         ref = le32_to_cpu(BHDR(bh)->h_refcount);
1243         if (ref == 1) {
1244                 ea_bdebug(bh, "refcount now=0; freeing");
1245                 /*
1246                  * This must happen under buffer lock for
1247                  * ext4_xattr_block_set() to reliably detect freed block
1248                  */
1249                 if (ea_block_cache) {
1250                         struct mb_cache_entry *oe;
1251
1252                         oe = mb_cache_entry_delete_or_get(ea_block_cache, hash,
1253                                                           bh->b_blocknr);
1254                         if (oe) {
1255                                 unlock_buffer(bh);
1256                                 mb_cache_entry_wait_unused(oe);
1257                                 mb_cache_entry_put(ea_block_cache, oe);
1258                                 goto retry_ref;
1259                         }
1260                 }
1261                 get_bh(bh);
1262                 unlock_buffer(bh);
1263
1264                 if (ext4_has_feature_ea_inode(inode->i_sb))
1265                         ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1266                                                      BFIRST(bh),
1267                                                      true /* block_csum */,
1268                                                      ea_inode_array,
1269                                                      extra_credits,
1270                                                      true /* skip_quota */);
1271                 ext4_free_blocks(handle, inode, bh, 0, 1,
1272                                  EXT4_FREE_BLOCKS_METADATA |
1273                                  EXT4_FREE_BLOCKS_FORGET);
1274         } else {
1275                 ref--;
1276                 BHDR(bh)->h_refcount = cpu_to_le32(ref);
1277                 if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1278                         struct mb_cache_entry *ce;
1279
1280                         if (ea_block_cache) {
1281                                 ce = mb_cache_entry_get(ea_block_cache, hash,
1282                                                         bh->b_blocknr);
1283                                 if (ce) {
1284                                         ce->e_reusable = 1;
1285                                         mb_cache_entry_put(ea_block_cache, ce);
1286                                 }
1287                         }
1288                 }
1289
1290                 ext4_xattr_block_csum_set(inode, bh);
1291                 /*
1292                  * Beware of this ugliness: Releasing of xattr block references
1293                  * from different inodes can race and so we have to protect
1294                  * from a race where someone else frees the block (and releases
1295                  * its journal_head) before we are done dirtying the buffer. In
1296                  * nojournal mode this race is harmless and we actually cannot
1297                  * call ext4_handle_dirty_metadata() with locked buffer as
1298                  * that function can call sync_dirty_buffer() so for that case
1299                  * we handle the dirtying after unlocking the buffer.
1300                  */
1301                 if (ext4_handle_valid(handle))
1302                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1303                 unlock_buffer(bh);
1304                 if (!ext4_handle_valid(handle))
1305                         error = ext4_handle_dirty_metadata(handle, inode, bh);
1306                 if (IS_SYNC(inode))
1307                         ext4_handle_sync(handle);
1308                 dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1309                 ea_bdebug(bh, "refcount now=%d; releasing",
1310                           le32_to_cpu(BHDR(bh)->h_refcount));
1311         }
1312 out:
1313         ext4_std_error(inode->i_sb, error);
1314         return;
1315 }
1316
1317 /*
1318  * Find the available free space for EAs. This also returns the total number of
1319  * bytes used by EA entries.
1320  */
1321 static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1322                                     size_t *min_offs, void *base, int *total)
1323 {
1324         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1325                 if (!last->e_value_inum && last->e_value_size) {
1326                         size_t offs = le16_to_cpu(last->e_value_offs);
1327                         if (offs < *min_offs)
1328                                 *min_offs = offs;
1329                 }
1330                 if (total)
1331                         *total += EXT4_XATTR_LEN(last->e_name_len);
1332         }
1333         return (*min_offs - ((void *)last - base) - sizeof(__u32));
1334 }
1335
1336 /*
1337  * Write the value of the EA in an inode.
1338  */
1339 static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1340                                   const void *buf, int bufsize)
1341 {
1342         struct buffer_head *bh = NULL;
1343         unsigned long block = 0;
1344         int blocksize = ea_inode->i_sb->s_blocksize;
1345         int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1346         int csize, wsize = 0;
1347         int ret = 0, ret2 = 0;
1348         int retries = 0;
1349
1350 retry:
1351         while (ret >= 0 && ret < max_blocks) {
1352                 struct ext4_map_blocks map;
1353                 map.m_lblk = block += ret;
1354                 map.m_len = max_blocks -= ret;
1355
1356                 ret = ext4_map_blocks(handle, ea_inode, &map,
1357                                       EXT4_GET_BLOCKS_CREATE);
1358                 if (ret <= 0) {
1359                         ext4_mark_inode_dirty(handle, ea_inode);
1360                         if (ret == -ENOSPC &&
1361                             ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1362                                 ret = 0;
1363                                 goto retry;
1364                         }
1365                         break;
1366                 }
1367         }
1368
1369         if (ret < 0)
1370                 return ret;
1371
1372         block = 0;
1373         while (wsize < bufsize) {
1374                 brelse(bh);
1375                 csize = (bufsize - wsize) > blocksize ? blocksize :
1376                                                                 bufsize - wsize;
1377                 bh = ext4_getblk(handle, ea_inode, block, 0);
1378                 if (IS_ERR(bh))
1379                         return PTR_ERR(bh);
1380                 if (!bh) {
1381                         WARN_ON_ONCE(1);
1382                         EXT4_ERROR_INODE(ea_inode,
1383                                          "ext4_getblk() return bh = NULL");
1384                         return -EFSCORRUPTED;
1385                 }
1386                 ret = ext4_journal_get_write_access(handle, ea_inode->i_sb, bh,
1387                                                    EXT4_JTR_NONE);
1388                 if (ret)
1389                         goto out;
1390
1391                 memcpy(bh->b_data, buf, csize);
1392                 set_buffer_uptodate(bh);
1393                 ext4_handle_dirty_metadata(handle, ea_inode, bh);
1394
1395                 buf += csize;
1396                 wsize += csize;
1397                 block += 1;
1398         }
1399
1400         inode_lock(ea_inode);
1401         i_size_write(ea_inode, wsize);
1402         ext4_update_i_disksize(ea_inode, wsize);
1403         inode_unlock(ea_inode);
1404
1405         ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1406         if (unlikely(ret2 && !ret))
1407                 ret = ret2;
1408
1409 out:
1410         brelse(bh);
1411
1412         return ret;
1413 }
1414
1415 /*
1416  * Create an inode to store the value of a large EA.
1417  */
1418 static struct inode *ext4_xattr_inode_create(handle_t *handle,
1419                                              struct inode *inode, u32 hash)
1420 {
1421         struct inode *ea_inode = NULL;
1422         uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1423         int err;
1424
1425         /*
1426          * Let the next inode be the goal, so we try and allocate the EA inode
1427          * in the same group, or nearby one.
1428          */
1429         ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1430                                   S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1431                                   EXT4_EA_INODE_FL);
1432         if (!IS_ERR(ea_inode)) {
1433                 ea_inode->i_op = &ext4_file_inode_operations;
1434                 ea_inode->i_fop = &ext4_file_operations;
1435                 ext4_set_aops(ea_inode);
1436                 ext4_xattr_inode_set_class(ea_inode);
1437                 unlock_new_inode(ea_inode);
1438                 ext4_xattr_inode_set_ref(ea_inode, 1);
1439                 ext4_xattr_inode_set_hash(ea_inode, hash);
1440                 err = ext4_mark_inode_dirty(handle, ea_inode);
1441                 if (!err)
1442                         err = ext4_inode_attach_jinode(ea_inode);
1443                 if (err) {
1444                         iput(ea_inode);
1445                         return ERR_PTR(err);
1446                 }
1447
1448                 /*
1449                  * Xattr inodes are shared therefore quota charging is performed
1450                  * at a higher level.
1451                  */
1452                 dquot_free_inode(ea_inode);
1453                 dquot_drop(ea_inode);
1454                 inode_lock(ea_inode);
1455                 ea_inode->i_flags |= S_NOQUOTA;
1456                 inode_unlock(ea_inode);
1457         }
1458
1459         return ea_inode;
1460 }
1461
1462 static struct inode *
1463 ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1464                             size_t value_len, u32 hash)
1465 {
1466         struct inode *ea_inode;
1467         struct mb_cache_entry *ce;
1468         struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1469         void *ea_data;
1470
1471         if (!ea_inode_cache)
1472                 return NULL;
1473
1474         ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1475         if (!ce)
1476                 return NULL;
1477
1478         WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1479                      !(current->flags & PF_MEMALLOC_NOFS));
1480
1481         ea_data = kvmalloc(value_len, GFP_KERNEL);
1482         if (!ea_data) {
1483                 mb_cache_entry_put(ea_inode_cache, ce);
1484                 return NULL;
1485         }
1486
1487         while (ce) {
1488                 ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1489                                      EXT4_IGET_NORMAL);
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 further 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_inum) {
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 = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1825                 if (IS_ERR(bs->bh)) {
1826                         error = PTR_ERR(bs->bh);
1827                         bs->bh = NULL;
1828                         return error;
1829                 }
1830                 ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1831                         atomic_read(&(bs->bh->b_count)),
1832                         le32_to_cpu(BHDR(bs->bh)->h_refcount));
1833                 error = ext4_xattr_check_block(inode, bs->bh);
1834                 if (error)
1835                         return error;
1836                 /* Find the named attribute. */
1837                 bs->s.base = BHDR(bs->bh);
1838                 bs->s.first = BFIRST(bs->bh);
1839                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
1840                 bs->s.here = bs->s.first;
1841                 error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1842                                          i->name_index, i->name, 1);
1843                 if (error && error != -ENODATA)
1844                         return error;
1845                 bs->s.not_found = error;
1846         }
1847         return 0;
1848 }
1849
1850 static int
1851 ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1852                      struct ext4_xattr_info *i,
1853                      struct ext4_xattr_block_find *bs)
1854 {
1855         struct super_block *sb = inode->i_sb;
1856         struct buffer_head *new_bh = NULL;
1857         struct ext4_xattr_search s_copy = bs->s;
1858         struct ext4_xattr_search *s = &s_copy;
1859         struct mb_cache_entry *ce = NULL;
1860         int error = 0;
1861         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1862         struct inode *ea_inode = NULL, *tmp_inode;
1863         size_t old_ea_inode_quota = 0;
1864         unsigned int ea_ino;
1865
1866
1867 #define header(x) ((struct ext4_xattr_header *)(x))
1868
1869         if (s->base) {
1870                 int offset = (char *)s->here - bs->bh->b_data;
1871
1872                 BUFFER_TRACE(bs->bh, "get_write_access");
1873                 error = ext4_journal_get_write_access(handle, sb, bs->bh,
1874                                                       EXT4_JTR_NONE);
1875                 if (error)
1876                         goto cleanup;
1877                 lock_buffer(bs->bh);
1878
1879                 if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1880                         __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1881
1882                         /*
1883                          * This must happen under buffer lock for
1884                          * ext4_xattr_block_set() to reliably detect modified
1885                          * block
1886                          */
1887                         if (ea_block_cache) {
1888                                 struct mb_cache_entry *oe;
1889
1890                                 oe = mb_cache_entry_delete_or_get(ea_block_cache,
1891                                         hash, bs->bh->b_blocknr);
1892                                 if (oe) {
1893                                         /*
1894                                          * Xattr block is getting reused. Leave
1895                                          * it alone.
1896                                          */
1897                                         mb_cache_entry_put(ea_block_cache, oe);
1898                                         goto clone_block;
1899                                 }
1900                         }
1901                         ea_bdebug(bs->bh, "modifying in-place");
1902                         error = ext4_xattr_set_entry(i, s, handle, inode,
1903                                                      true /* is_block */);
1904                         ext4_xattr_block_csum_set(inode, bs->bh);
1905                         unlock_buffer(bs->bh);
1906                         if (error == -EFSCORRUPTED)
1907                                 goto bad_block;
1908                         if (!error)
1909                                 error = ext4_handle_dirty_metadata(handle,
1910                                                                    inode,
1911                                                                    bs->bh);
1912                         if (error)
1913                                 goto cleanup;
1914                         goto inserted;
1915                 }
1916 clone_block:
1917                 unlock_buffer(bs->bh);
1918                 ea_bdebug(bs->bh, "cloning");
1919                 s->base = kmemdup(BHDR(bs->bh), bs->bh->b_size, GFP_NOFS);
1920                 error = -ENOMEM;
1921                 if (s->base == NULL)
1922                         goto cleanup;
1923                 s->first = ENTRY(header(s->base)+1);
1924                 header(s->base)->h_refcount = cpu_to_le32(1);
1925                 s->here = ENTRY(s->base + offset);
1926                 s->end = s->base + bs->bh->b_size;
1927
1928                 /*
1929                  * If existing entry points to an xattr inode, we need
1930                  * to prevent ext4_xattr_set_entry() from decrementing
1931                  * ref count on it because the reference belongs to the
1932                  * original block. In this case, make the entry look
1933                  * like it has an empty value.
1934                  */
1935                 if (!s->not_found && s->here->e_value_inum) {
1936                         ea_ino = le32_to_cpu(s->here->e_value_inum);
1937                         error = ext4_xattr_inode_iget(inode, ea_ino,
1938                                       le32_to_cpu(s->here->e_hash),
1939                                       &tmp_inode);
1940                         if (error)
1941                                 goto cleanup;
1942
1943                         if (!ext4_test_inode_state(tmp_inode,
1944                                         EXT4_STATE_LUSTRE_EA_INODE)) {
1945                                 /*
1946                                  * Defer quota free call for previous
1947                                  * inode until success is guaranteed.
1948                                  */
1949                                 old_ea_inode_quota = le32_to_cpu(
1950                                                 s->here->e_value_size);
1951                         }
1952                         iput(tmp_inode);
1953
1954                         s->here->e_value_inum = 0;
1955                         s->here->e_value_size = 0;
1956                 }
1957         } else {
1958                 /* Allocate a buffer where we construct the new block. */
1959                 s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1960                 error = -ENOMEM;
1961                 if (s->base == NULL)
1962                         goto cleanup;
1963                 header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1964                 header(s->base)->h_blocks = cpu_to_le32(1);
1965                 header(s->base)->h_refcount = cpu_to_le32(1);
1966                 s->first = ENTRY(header(s->base)+1);
1967                 s->here = ENTRY(header(s->base)+1);
1968                 s->end = s->base + sb->s_blocksize;
1969         }
1970
1971         error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1972         if (error == -EFSCORRUPTED)
1973                 goto bad_block;
1974         if (error)
1975                 goto cleanup;
1976
1977         if (i->value && s->here->e_value_inum) {
1978                 /*
1979                  * A ref count on ea_inode has been taken as part of the call to
1980                  * ext4_xattr_set_entry() above. We would like to drop this
1981                  * extra ref but we have to wait until the xattr block is
1982                  * initialized and has its own ref count on the ea_inode.
1983                  */
1984                 ea_ino = le32_to_cpu(s->here->e_value_inum);
1985                 error = ext4_xattr_inode_iget(inode, ea_ino,
1986                                               le32_to_cpu(s->here->e_hash),
1987                                               &ea_inode);
1988                 if (error) {
1989                         ea_inode = NULL;
1990                         goto cleanup;
1991                 }
1992         }
1993
1994 inserted:
1995         if (!IS_LAST_ENTRY(s->first)) {
1996                 new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1997                                                      &ce);
1998                 if (new_bh) {
1999                         /* We found an identical block in the cache. */
2000                         if (new_bh == bs->bh)
2001                                 ea_bdebug(new_bh, "keeping");
2002                         else {
2003                                 u32 ref;
2004
2005                                 WARN_ON_ONCE(dquot_initialize_needed(inode));
2006
2007                                 /* The old block is released after updating
2008                                    the inode. */
2009                                 error = dquot_alloc_block(inode,
2010                                                 EXT4_C2B(EXT4_SB(sb), 1));
2011                                 if (error)
2012                                         goto cleanup;
2013                                 BUFFER_TRACE(new_bh, "get_write_access");
2014                                 error = ext4_journal_get_write_access(
2015                                                 handle, sb, new_bh,
2016                                                 EXT4_JTR_NONE);
2017                                 if (error)
2018                                         goto cleanup_dquot;
2019                                 lock_buffer(new_bh);
2020                                 /*
2021                                  * We have to be careful about races with
2022                                  * adding references to xattr block. Once we
2023                                  * hold buffer lock xattr block's state is
2024                                  * stable so we can check the additional
2025                                  * reference fits.
2026                                  */
2027                                 ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2028                                 if (ref > EXT4_XATTR_REFCOUNT_MAX) {
2029                                         /*
2030                                          * Undo everything and check mbcache
2031                                          * again.
2032                                          */
2033                                         unlock_buffer(new_bh);
2034                                         dquot_free_block(inode,
2035                                                          EXT4_C2B(EXT4_SB(sb),
2036                                                                   1));
2037                                         brelse(new_bh);
2038                                         mb_cache_entry_put(ea_block_cache, ce);
2039                                         ce = NULL;
2040                                         new_bh = NULL;
2041                                         goto inserted;
2042                                 }
2043                                 BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2044                                 if (ref == EXT4_XATTR_REFCOUNT_MAX)
2045                                         ce->e_reusable = 0;
2046                                 ea_bdebug(new_bh, "reusing; refcount now=%d",
2047                                           ref);
2048                                 ext4_xattr_block_csum_set(inode, new_bh);
2049                                 unlock_buffer(new_bh);
2050                                 error = ext4_handle_dirty_metadata(handle,
2051                                                                    inode,
2052                                                                    new_bh);
2053                                 if (error)
2054                                         goto cleanup_dquot;
2055                         }
2056                         mb_cache_entry_touch(ea_block_cache, ce);
2057                         mb_cache_entry_put(ea_block_cache, ce);
2058                         ce = NULL;
2059                 } else if (bs->bh && s->base == bs->bh->b_data) {
2060                         /* We were modifying this block in-place. */
2061                         ea_bdebug(bs->bh, "keeping this block");
2062                         ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2063                         new_bh = bs->bh;
2064                         get_bh(new_bh);
2065                 } else {
2066                         /* We need to allocate a new block */
2067                         ext4_fsblk_t goal, block;
2068
2069                         WARN_ON_ONCE(dquot_initialize_needed(inode));
2070
2071                         goal = ext4_group_first_block_no(sb,
2072                                                 EXT4_I(inode)->i_block_group);
2073
2074                         /* non-extent files can't have physical blocks past 2^32 */
2075                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2076                                 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2077
2078                         block = ext4_new_meta_blocks(handle, inode, goal, 0,
2079                                                      NULL, &error);
2080                         if (error)
2081                                 goto cleanup;
2082
2083                         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2084                                 BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2085
2086                         ea_idebug(inode, "creating block %llu",
2087                                   (unsigned long long)block);
2088
2089                         new_bh = sb_getblk(sb, block);
2090                         if (unlikely(!new_bh)) {
2091                                 error = -ENOMEM;
2092 getblk_failed:
2093                                 ext4_free_blocks(handle, inode, NULL, block, 1,
2094                                                  EXT4_FREE_BLOCKS_METADATA);
2095                                 goto cleanup;
2096                         }
2097                         error = ext4_xattr_inode_inc_ref_all(handle, inode,
2098                                                       ENTRY(header(s->base)+1));
2099                         if (error)
2100                                 goto getblk_failed;
2101                         if (ea_inode) {
2102                                 /* Drop the extra ref on ea_inode. */
2103                                 error = ext4_xattr_inode_dec_ref(handle,
2104                                                                  ea_inode);
2105                                 if (error)
2106                                         ext4_warning_inode(ea_inode,
2107                                                            "dec ref error=%d",
2108                                                            error);
2109                                 iput(ea_inode);
2110                                 ea_inode = NULL;
2111                         }
2112
2113                         lock_buffer(new_bh);
2114                         error = ext4_journal_get_create_access(handle, sb,
2115                                                         new_bh, EXT4_JTR_NONE);
2116                         if (error) {
2117                                 unlock_buffer(new_bh);
2118                                 error = -EIO;
2119                                 goto getblk_failed;
2120                         }
2121                         memcpy(new_bh->b_data, s->base, new_bh->b_size);
2122                         ext4_xattr_block_csum_set(inode, new_bh);
2123                         set_buffer_uptodate(new_bh);
2124                         unlock_buffer(new_bh);
2125                         ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2126                         error = ext4_handle_dirty_metadata(handle, inode,
2127                                                            new_bh);
2128                         if (error)
2129                                 goto cleanup;
2130                 }
2131         }
2132
2133         if (old_ea_inode_quota)
2134                 ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2135
2136         /* Update the inode. */
2137         EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2138
2139         /* Drop the previous xattr block. */
2140         if (bs->bh && bs->bh != new_bh) {
2141                 struct ext4_xattr_inode_array *ea_inode_array = NULL;
2142
2143                 ext4_xattr_release_block(handle, inode, bs->bh,
2144                                          &ea_inode_array,
2145                                          0 /* extra_credits */);
2146                 ext4_xattr_inode_array_free(ea_inode_array);
2147         }
2148         error = 0;
2149
2150 cleanup:
2151         if (ea_inode) {
2152                 int error2;
2153
2154                 error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2155                 if (error2)
2156                         ext4_warning_inode(ea_inode, "dec ref error=%d",
2157                                            error2);
2158
2159                 /* If there was an error, revert the quota charge. */
2160                 if (error)
2161                         ext4_xattr_inode_free_quota(inode, ea_inode,
2162                                                     i_size_read(ea_inode));
2163                 iput(ea_inode);
2164         }
2165         if (ce)
2166                 mb_cache_entry_put(ea_block_cache, ce);
2167         brelse(new_bh);
2168         if (!(bs->bh && s->base == bs->bh->b_data))
2169                 kfree(s->base);
2170
2171         return error;
2172
2173 cleanup_dquot:
2174         dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2175         goto cleanup;
2176
2177 bad_block:
2178         EXT4_ERROR_INODE(inode, "bad block %llu",
2179                          EXT4_I(inode)->i_file_acl);
2180         goto cleanup;
2181
2182 #undef header
2183 }
2184
2185 int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2186                           struct ext4_xattr_ibody_find *is)
2187 {
2188         struct ext4_xattr_ibody_header *header;
2189         struct ext4_inode *raw_inode;
2190         int error;
2191
2192         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2193                 return 0;
2194
2195         raw_inode = ext4_raw_inode(&is->iloc);
2196         header = IHDR(inode, raw_inode);
2197         is->s.base = is->s.first = IFIRST(header);
2198         is->s.here = is->s.first;
2199         is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2200         if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2201                 error = xattr_check_inode(inode, header, is->s.end);
2202                 if (error)
2203                         return error;
2204                 /* Find the named attribute. */
2205                 error = xattr_find_entry(inode, &is->s.here, is->s.end,
2206                                          i->name_index, i->name, 0);
2207                 if (error && error != -ENODATA)
2208                         return error;
2209                 is->s.not_found = error;
2210         }
2211         return 0;
2212 }
2213
2214 int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2215                                 struct ext4_xattr_info *i,
2216                                 struct ext4_xattr_ibody_find *is)
2217 {
2218         struct ext4_xattr_ibody_header *header;
2219         struct ext4_xattr_search *s = &is->s;
2220         int error;
2221
2222         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
2223                 return -ENOSPC;
2224
2225         error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2226         if (error)
2227                 return error;
2228         header = IHDR(inode, ext4_raw_inode(&is->iloc));
2229         if (!IS_LAST_ENTRY(s->first)) {
2230                 header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2231                 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2232         } else {
2233                 header->h_magic = cpu_to_le32(0);
2234                 ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2235         }
2236         return 0;
2237 }
2238
2239 static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2240                                  struct ext4_xattr_info *i)
2241 {
2242         void *value;
2243
2244         /* When e_value_inum is set the value is stored externally. */
2245         if (s->here->e_value_inum)
2246                 return 0;
2247         if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2248                 return 0;
2249         value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2250         return !memcmp(value, i->value, i->value_len);
2251 }
2252
2253 static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2254 {
2255         struct buffer_head *bh;
2256         int error;
2257
2258         if (!EXT4_I(inode)->i_file_acl)
2259                 return NULL;
2260         bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2261         if (IS_ERR(bh))
2262                 return bh;
2263         error = ext4_xattr_check_block(inode, bh);
2264         if (error) {
2265                 brelse(bh);
2266                 return ERR_PTR(error);
2267         }
2268         return bh;
2269 }
2270
2271 /*
2272  * ext4_xattr_set_handle()
2273  *
2274  * Create, replace or remove an extended attribute for this inode.  Value
2275  * is NULL to remove an existing extended attribute, and non-NULL to
2276  * either replace an existing extended attribute, or create a new extended
2277  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2278  * specify that an extended attribute must exist and must not exist
2279  * previous to the call, respectively.
2280  *
2281  * Returns 0, or a negative error number on failure.
2282  */
2283 int
2284 ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2285                       const char *name, const void *value, size_t value_len,
2286                       int flags)
2287 {
2288         struct ext4_xattr_info i = {
2289                 .name_index = name_index,
2290                 .name = name,
2291                 .value = value,
2292                 .value_len = value_len,
2293                 .in_inode = 0,
2294         };
2295         struct ext4_xattr_ibody_find is = {
2296                 .s = { .not_found = -ENODATA, },
2297         };
2298         struct ext4_xattr_block_find bs = {
2299                 .s = { .not_found = -ENODATA, },
2300         };
2301         int no_expand;
2302         int error;
2303
2304         if (!name)
2305                 return -EINVAL;
2306         if (strlen(name) > 255)
2307                 return -ERANGE;
2308
2309         ext4_write_lock_xattr(inode, &no_expand);
2310
2311         /* Check journal credits under write lock. */
2312         if (ext4_handle_valid(handle)) {
2313                 struct buffer_head *bh;
2314                 int credits;
2315
2316                 bh = ext4_xattr_get_block(inode);
2317                 if (IS_ERR(bh)) {
2318                         error = PTR_ERR(bh);
2319                         goto cleanup;
2320                 }
2321
2322                 credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2323                                                    value_len,
2324                                                    flags & XATTR_CREATE);
2325                 brelse(bh);
2326
2327                 if (jbd2_handle_buffer_credits(handle) < credits) {
2328                         error = -ENOSPC;
2329                         goto cleanup;
2330                 }
2331                 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2332         }
2333
2334         error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2335         if (error)
2336                 goto cleanup;
2337
2338         if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2339                 struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2340                 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2341                 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2342         }
2343
2344         error = ext4_xattr_ibody_find(inode, &i, &is);
2345         if (error)
2346                 goto cleanup;
2347         if (is.s.not_found)
2348                 error = ext4_xattr_block_find(inode, &i, &bs);
2349         if (error)
2350                 goto cleanup;
2351         if (is.s.not_found && bs.s.not_found) {
2352                 error = -ENODATA;
2353                 if (flags & XATTR_REPLACE)
2354                         goto cleanup;
2355                 error = 0;
2356                 if (!value)
2357                         goto cleanup;
2358         } else {
2359                 error = -EEXIST;
2360                 if (flags & XATTR_CREATE)
2361                         goto cleanup;
2362         }
2363
2364         if (!value) {
2365                 if (!is.s.not_found)
2366                         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2367                 else if (!bs.s.not_found)
2368                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2369         } else {
2370                 error = 0;
2371                 /* Xattr value did not change? Save us some work and bail out */
2372                 if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2373                         goto cleanup;
2374                 if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2375                         goto cleanup;
2376
2377                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2378                     (EXT4_XATTR_SIZE(i.value_len) >
2379                         EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2380                         i.in_inode = 1;
2381 retry_inode:
2382                 error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2383                 if (!error && !bs.s.not_found) {
2384                         i.value = NULL;
2385                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2386                 } else if (error == -ENOSPC) {
2387                         if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2388                                 brelse(bs.bh);
2389                                 bs.bh = NULL;
2390                                 error = ext4_xattr_block_find(inode, &i, &bs);
2391                                 if (error)
2392                                         goto cleanup;
2393                         }
2394                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
2395                         if (!error && !is.s.not_found) {
2396                                 i.value = NULL;
2397                                 error = ext4_xattr_ibody_set(handle, inode, &i,
2398                                                              &is);
2399                         } else if (error == -ENOSPC) {
2400                                 /*
2401                                  * Xattr does not fit in the block, store at
2402                                  * external inode if possible.
2403                                  */
2404                                 if (ext4_has_feature_ea_inode(inode->i_sb) &&
2405                                     i.value_len && !i.in_inode) {
2406                                         i.in_inode = 1;
2407                                         goto retry_inode;
2408                                 }
2409                         }
2410                 }
2411         }
2412         if (!error) {
2413                 ext4_xattr_update_super_block(handle, inode->i_sb);
2414                 inode->i_ctime = current_time(inode);
2415                 if (!value)
2416                         no_expand = 0;
2417                 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2418                 /*
2419                  * The bh is consumed by ext4_mark_iloc_dirty, even with
2420                  * error != 0.
2421                  */
2422                 is.iloc.bh = NULL;
2423                 if (IS_SYNC(inode))
2424                         ext4_handle_sync(handle);
2425         }
2426         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2427
2428 cleanup:
2429         brelse(is.iloc.bh);
2430         brelse(bs.bh);
2431         ext4_write_unlock_xattr(inode, &no_expand);
2432         return error;
2433 }
2434
2435 int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2436                            bool is_create, int *credits)
2437 {
2438         struct buffer_head *bh;
2439         int err;
2440
2441         *credits = 0;
2442
2443         if (!EXT4_SB(inode->i_sb)->s_journal)
2444                 return 0;
2445
2446         down_read(&EXT4_I(inode)->xattr_sem);
2447
2448         bh = ext4_xattr_get_block(inode);
2449         if (IS_ERR(bh)) {
2450                 err = PTR_ERR(bh);
2451         } else {
2452                 *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2453                                                     value_len, is_create);
2454                 brelse(bh);
2455                 err = 0;
2456         }
2457
2458         up_read(&EXT4_I(inode)->xattr_sem);
2459         return err;
2460 }
2461
2462 /*
2463  * ext4_xattr_set()
2464  *
2465  * Like ext4_xattr_set_handle, but start from an inode. This extended
2466  * attribute modification is a filesystem transaction by itself.
2467  *
2468  * Returns 0, or a negative error number on failure.
2469  */
2470 int
2471 ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2472                const void *value, size_t value_len, int flags)
2473 {
2474         handle_t *handle;
2475         struct super_block *sb = inode->i_sb;
2476         int error, retries = 0;
2477         int credits;
2478
2479         error = dquot_initialize(inode);
2480         if (error)
2481                 return error;
2482
2483 retry:
2484         error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2485                                        &credits);
2486         if (error)
2487                 return error;
2488
2489         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2490         if (IS_ERR(handle)) {
2491                 error = PTR_ERR(handle);
2492         } else {
2493                 int error2;
2494
2495                 error = ext4_xattr_set_handle(handle, inode, name_index, name,
2496                                               value, value_len, flags);
2497                 error2 = ext4_journal_stop(handle);
2498                 if (error == -ENOSPC &&
2499                     ext4_should_retry_alloc(sb, &retries))
2500                         goto retry;
2501                 if (error == 0)
2502                         error = error2;
2503         }
2504         ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, NULL);
2505
2506         return error;
2507 }
2508
2509 /*
2510  * Shift the EA entries in the inode to create space for the increased
2511  * i_extra_isize.
2512  */
2513 static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2514                                      int value_offs_shift, void *to,
2515                                      void *from, size_t n)
2516 {
2517         struct ext4_xattr_entry *last = entry;
2518         int new_offs;
2519
2520         /* We always shift xattr headers further thus offsets get lower */
2521         BUG_ON(value_offs_shift > 0);
2522
2523         /* Adjust the value offsets of the entries */
2524         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2525                 if (!last->e_value_inum && last->e_value_size) {
2526                         new_offs = le16_to_cpu(last->e_value_offs) +
2527                                                         value_offs_shift;
2528                         last->e_value_offs = cpu_to_le16(new_offs);
2529                 }
2530         }
2531         /* Shift the entries by n bytes */
2532         memmove(to, from, n);
2533 }
2534
2535 /*
2536  * Move xattr pointed to by 'entry' from inode into external xattr block
2537  */
2538 static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2539                                     struct ext4_inode *raw_inode,
2540                                     struct ext4_xattr_entry *entry)
2541 {
2542         struct ext4_xattr_ibody_find *is = NULL;
2543         struct ext4_xattr_block_find *bs = NULL;
2544         char *buffer = NULL, *b_entry_name = NULL;
2545         size_t value_size = le32_to_cpu(entry->e_value_size);
2546         struct ext4_xattr_info i = {
2547                 .value = NULL,
2548                 .value_len = 0,
2549                 .name_index = entry->e_name_index,
2550                 .in_inode = !!entry->e_value_inum,
2551         };
2552         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2553         int error;
2554
2555         is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2556         bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2557         buffer = kmalloc(value_size, GFP_NOFS);
2558         b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2559         if (!is || !bs || !buffer || !b_entry_name) {
2560                 error = -ENOMEM;
2561                 goto out;
2562         }
2563
2564         is->s.not_found = -ENODATA;
2565         bs->s.not_found = -ENODATA;
2566         is->iloc.bh = NULL;
2567         bs->bh = NULL;
2568
2569         /* Save the entry name and the entry value */
2570         if (entry->e_value_inum) {
2571                 error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2572                 if (error)
2573                         goto out;
2574         } else {
2575                 size_t value_offs = le16_to_cpu(entry->e_value_offs);
2576                 memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2577         }
2578
2579         memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2580         b_entry_name[entry->e_name_len] = '\0';
2581         i.name = b_entry_name;
2582
2583         error = ext4_get_inode_loc(inode, &is->iloc);
2584         if (error)
2585                 goto out;
2586
2587         error = ext4_xattr_ibody_find(inode, &i, is);
2588         if (error)
2589                 goto out;
2590
2591         /* Remove the chosen entry from the inode */
2592         error = ext4_xattr_ibody_set(handle, inode, &i, is);
2593         if (error)
2594                 goto out;
2595
2596         i.value = buffer;
2597         i.value_len = value_size;
2598         error = ext4_xattr_block_find(inode, &i, bs);
2599         if (error)
2600                 goto out;
2601
2602         /* Add entry which was removed from the inode into the block */
2603         error = ext4_xattr_block_set(handle, inode, &i, bs);
2604         if (error)
2605                 goto out;
2606         error = 0;
2607 out:
2608         kfree(b_entry_name);
2609         kfree(buffer);
2610         if (is)
2611                 brelse(is->iloc.bh);
2612         if (bs)
2613                 brelse(bs->bh);
2614         kfree(is);
2615         kfree(bs);
2616
2617         return error;
2618 }
2619
2620 static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2621                                        struct ext4_inode *raw_inode,
2622                                        int isize_diff, size_t ifree,
2623                                        size_t bfree, int *total_ino)
2624 {
2625         struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2626         struct ext4_xattr_entry *small_entry;
2627         struct ext4_xattr_entry *entry;
2628         struct ext4_xattr_entry *last;
2629         unsigned int entry_size;        /* EA entry size */
2630         unsigned int total_size;        /* EA entry size + value size */
2631         unsigned int min_total_size;
2632         int error;
2633
2634         while (isize_diff > ifree) {
2635                 entry = NULL;
2636                 small_entry = NULL;
2637                 min_total_size = ~0U;
2638                 last = IFIRST(header);
2639                 /* Find the entry best suited to be pushed into EA block */
2640                 for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2641                         /* never move system.data out of the inode */
2642                         if ((last->e_name_len == 4) &&
2643                             (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2644                             !memcmp(last->e_name, "data", 4))
2645                                 continue;
2646                         total_size = EXT4_XATTR_LEN(last->e_name_len);
2647                         if (!last->e_value_inum)
2648                                 total_size += EXT4_XATTR_SIZE(
2649                                                le32_to_cpu(last->e_value_size));
2650                         if (total_size <= bfree &&
2651                             total_size < min_total_size) {
2652                                 if (total_size + ifree < isize_diff) {
2653                                         small_entry = last;
2654                                 } else {
2655                                         entry = last;
2656                                         min_total_size = total_size;
2657                                 }
2658                         }
2659                 }
2660
2661                 if (entry == NULL) {
2662                         if (small_entry == NULL)
2663                                 return -ENOSPC;
2664                         entry = small_entry;
2665                 }
2666
2667                 entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2668                 total_size = entry_size;
2669                 if (!entry->e_value_inum)
2670                         total_size += EXT4_XATTR_SIZE(
2671                                               le32_to_cpu(entry->e_value_size));
2672                 error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2673                                                  entry);
2674                 if (error)
2675                         return error;
2676
2677                 *total_ino -= entry_size;
2678                 ifree += total_size;
2679                 bfree -= total_size;
2680         }
2681
2682         return 0;
2683 }
2684
2685 /*
2686  * Expand an inode by new_extra_isize bytes when EAs are present.
2687  * Returns 0 on success or negative error number on failure.
2688  */
2689 int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2690                                struct ext4_inode *raw_inode, handle_t *handle)
2691 {
2692         struct ext4_xattr_ibody_header *header;
2693         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2694         static unsigned int mnt_count;
2695         size_t min_offs;
2696         size_t ifree, bfree;
2697         int total_ino;
2698         void *base, *end;
2699         int error = 0, tried_min_extra_isize = 0;
2700         int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2701         int isize_diff; /* How much do we need to grow i_extra_isize */
2702
2703 retry:
2704         isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2705         if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2706                 return 0;
2707
2708         header = IHDR(inode, raw_inode);
2709
2710         /*
2711          * Check if enough free space is available in the inode to shift the
2712          * entries ahead by new_extra_isize.
2713          */
2714
2715         base = IFIRST(header);
2716         end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2717         min_offs = end - base;
2718         total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2719
2720         error = xattr_check_inode(inode, header, end);
2721         if (error)
2722                 goto cleanup;
2723
2724         ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2725         if (ifree >= isize_diff)
2726                 goto shift;
2727
2728         /*
2729          * Enough free space isn't available in the inode, check if
2730          * EA block can hold new_extra_isize bytes.
2731          */
2732         if (EXT4_I(inode)->i_file_acl) {
2733                 struct buffer_head *bh;
2734
2735                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2736                 if (IS_ERR(bh)) {
2737                         error = PTR_ERR(bh);
2738                         goto cleanup;
2739                 }
2740                 error = ext4_xattr_check_block(inode, bh);
2741                 if (error) {
2742                         brelse(bh);
2743                         goto cleanup;
2744                 }
2745                 base = BHDR(bh);
2746                 end = bh->b_data + bh->b_size;
2747                 min_offs = end - base;
2748                 bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2749                                               NULL);
2750                 brelse(bh);
2751                 if (bfree + ifree < isize_diff) {
2752                         if (!tried_min_extra_isize && s_min_extra_isize) {
2753                                 tried_min_extra_isize++;
2754                                 new_extra_isize = s_min_extra_isize;
2755                                 goto retry;
2756                         }
2757                         error = -ENOSPC;
2758                         goto cleanup;
2759                 }
2760         } else {
2761                 bfree = inode->i_sb->s_blocksize;
2762         }
2763
2764         error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2765                                             isize_diff, ifree, bfree,
2766                                             &total_ino);
2767         if (error) {
2768                 if (error == -ENOSPC && !tried_min_extra_isize &&
2769                     s_min_extra_isize) {
2770                         tried_min_extra_isize++;
2771                         new_extra_isize = s_min_extra_isize;
2772                         goto retry;
2773                 }
2774                 goto cleanup;
2775         }
2776 shift:
2777         /* Adjust the offsets and shift the remaining entries ahead */
2778         ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2779                         - new_extra_isize, (void *)raw_inode +
2780                         EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2781                         (void *)header, total_ino);
2782         EXT4_I(inode)->i_extra_isize = new_extra_isize;
2783
2784 cleanup:
2785         if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2786                 ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2787                              inode->i_ino);
2788                 mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2789         }
2790         return error;
2791 }
2792
2793 #define EIA_INCR 16 /* must be 2^n */
2794 #define EIA_MASK (EIA_INCR - 1)
2795
2796 /* Add the large xattr @inode into @ea_inode_array for deferred iput().
2797  * If @ea_inode_array is new or full it will be grown and the old
2798  * contents copied over.
2799  */
2800 static int
2801 ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2802                         struct inode *inode)
2803 {
2804         if (*ea_inode_array == NULL) {
2805                 /*
2806                  * Start with 15 inodes, so it fits into a power-of-two size.
2807                  * If *ea_inode_array is NULL, this is essentially offsetof()
2808                  */
2809                 (*ea_inode_array) =
2810                         kmalloc(offsetof(struct ext4_xattr_inode_array,
2811                                          inodes[EIA_MASK]),
2812                                 GFP_NOFS);
2813                 if (*ea_inode_array == NULL)
2814                         return -ENOMEM;
2815                 (*ea_inode_array)->count = 0;
2816         } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2817                 /* expand the array once all 15 + n * 16 slots are full */
2818                 struct ext4_xattr_inode_array *new_array = NULL;
2819                 int count = (*ea_inode_array)->count;
2820
2821                 /* if new_array is NULL, this is essentially offsetof() */
2822                 new_array = kmalloc(
2823                                 offsetof(struct ext4_xattr_inode_array,
2824                                          inodes[count + EIA_INCR]),
2825                                 GFP_NOFS);
2826                 if (new_array == NULL)
2827                         return -ENOMEM;
2828                 memcpy(new_array, *ea_inode_array,
2829                        offsetof(struct ext4_xattr_inode_array, inodes[count]));
2830                 kfree(*ea_inode_array);
2831                 *ea_inode_array = new_array;
2832         }
2833         (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2834         return 0;
2835 }
2836
2837 /*
2838  * ext4_xattr_delete_inode()
2839  *
2840  * Free extended attribute resources associated with this inode. Traverse
2841  * all entries and decrement reference on any xattr inodes associated with this
2842  * inode. This is called immediately before an inode is freed. We have exclusive
2843  * access to the inode. If an orphan inode is deleted it will also release its
2844  * references on xattr block and xattr inodes.
2845  */
2846 int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2847                             struct ext4_xattr_inode_array **ea_inode_array,
2848                             int extra_credits)
2849 {
2850         struct buffer_head *bh = NULL;
2851         struct ext4_xattr_ibody_header *header;
2852         struct ext4_iloc iloc = { .bh = NULL };
2853         struct ext4_xattr_entry *entry;
2854         struct inode *ea_inode;
2855         int error;
2856
2857         error = ext4_journal_ensure_credits(handle, extra_credits,
2858                         ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2859         if (error < 0) {
2860                 EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2861                 goto cleanup;
2862         }
2863
2864         if (ext4_has_feature_ea_inode(inode->i_sb) &&
2865             ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2866
2867                 error = ext4_get_inode_loc(inode, &iloc);
2868                 if (error) {
2869                         EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2870                         goto cleanup;
2871                 }
2872
2873                 error = ext4_journal_get_write_access(handle, inode->i_sb,
2874                                                 iloc.bh, EXT4_JTR_NONE);
2875                 if (error) {
2876                         EXT4_ERROR_INODE(inode, "write access (error %d)",
2877                                          error);
2878                         goto cleanup;
2879                 }
2880
2881                 header = IHDR(inode, ext4_raw_inode(&iloc));
2882                 if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2883                         ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2884                                                      IFIRST(header),
2885                                                      false /* block_csum */,
2886                                                      ea_inode_array,
2887                                                      extra_credits,
2888                                                      false /* skip_quota */);
2889         }
2890
2891         if (EXT4_I(inode)->i_file_acl) {
2892                 bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2893                 if (IS_ERR(bh)) {
2894                         error = PTR_ERR(bh);
2895                         if (error == -EIO) {
2896                                 EXT4_ERROR_INODE_ERR(inode, EIO,
2897                                                      "block %llu read error",
2898                                                      EXT4_I(inode)->i_file_acl);
2899                         }
2900                         bh = NULL;
2901                         goto cleanup;
2902                 }
2903                 error = ext4_xattr_check_block(inode, bh);
2904                 if (error)
2905                         goto cleanup;
2906
2907                 if (ext4_has_feature_ea_inode(inode->i_sb)) {
2908                         for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2909                              entry = EXT4_XATTR_NEXT(entry)) {
2910                                 if (!entry->e_value_inum)
2911                                         continue;
2912                                 error = ext4_xattr_inode_iget(inode,
2913                                               le32_to_cpu(entry->e_value_inum),
2914                                               le32_to_cpu(entry->e_hash),
2915                                               &ea_inode);
2916                                 if (error)
2917                                         continue;
2918                                 ext4_xattr_inode_free_quota(inode, ea_inode,
2919                                               le32_to_cpu(entry->e_value_size));
2920                                 iput(ea_inode);
2921                         }
2922
2923                 }
2924
2925                 ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2926                                          extra_credits);
2927                 /*
2928                  * Update i_file_acl value in the same transaction that releases
2929                  * block.
2930                  */
2931                 EXT4_I(inode)->i_file_acl = 0;
2932                 error = ext4_mark_inode_dirty(handle, inode);
2933                 if (error) {
2934                         EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2935                                          error);
2936                         goto cleanup;
2937                 }
2938                 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR, handle);
2939         }
2940         error = 0;
2941 cleanup:
2942         brelse(iloc.bh);
2943         brelse(bh);
2944         return error;
2945 }
2946
2947 void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2948 {
2949         int idx;
2950
2951         if (ea_inode_array == NULL)
2952                 return;
2953
2954         for (idx = 0; idx < ea_inode_array->count; ++idx)
2955                 iput(ea_inode_array->inodes[idx]);
2956         kfree(ea_inode_array);
2957 }
2958
2959 /*
2960  * ext4_xattr_block_cache_insert()
2961  *
2962  * Create a new entry in the extended attribute block cache, and insert
2963  * it unless such an entry is already in the cache.
2964  *
2965  * Returns 0, or a negative error number on failure.
2966  */
2967 static void
2968 ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2969                               struct buffer_head *bh)
2970 {
2971         struct ext4_xattr_header *header = BHDR(bh);
2972         __u32 hash = le32_to_cpu(header->h_hash);
2973         int reusable = le32_to_cpu(header->h_refcount) <
2974                        EXT4_XATTR_REFCOUNT_MAX;
2975         int error;
2976
2977         if (!ea_block_cache)
2978                 return;
2979         error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2980                                       bh->b_blocknr, reusable);
2981         if (error) {
2982                 if (error == -EBUSY)
2983                         ea_bdebug(bh, "already in cache");
2984         } else
2985                 ea_bdebug(bh, "inserting [%x]", (int)hash);
2986 }
2987
2988 /*
2989  * ext4_xattr_cmp()
2990  *
2991  * Compare two extended attribute blocks for equality.
2992  *
2993  * Returns 0 if the blocks are equal, 1 if they differ, and
2994  * a negative error number on errors.
2995  */
2996 static int
2997 ext4_xattr_cmp(struct ext4_xattr_header *header1,
2998                struct ext4_xattr_header *header2)
2999 {
3000         struct ext4_xattr_entry *entry1, *entry2;
3001
3002         entry1 = ENTRY(header1+1);
3003         entry2 = ENTRY(header2+1);
3004         while (!IS_LAST_ENTRY(entry1)) {
3005                 if (IS_LAST_ENTRY(entry2))
3006                         return 1;
3007                 if (entry1->e_hash != entry2->e_hash ||
3008                     entry1->e_name_index != entry2->e_name_index ||
3009                     entry1->e_name_len != entry2->e_name_len ||
3010                     entry1->e_value_size != entry2->e_value_size ||
3011                     entry1->e_value_inum != entry2->e_value_inum ||
3012                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3013                         return 1;
3014                 if (!entry1->e_value_inum &&
3015                     memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3016                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3017                            le32_to_cpu(entry1->e_value_size)))
3018                         return 1;
3019
3020                 entry1 = EXT4_XATTR_NEXT(entry1);
3021                 entry2 = EXT4_XATTR_NEXT(entry2);
3022         }
3023         if (!IS_LAST_ENTRY(entry2))
3024                 return 1;
3025         return 0;
3026 }
3027
3028 /*
3029  * ext4_xattr_block_cache_find()
3030  *
3031  * Find an identical extended attribute block.
3032  *
3033  * Returns a pointer to the block found, or NULL if such a block was
3034  * not found or an error occurred.
3035  */
3036 static struct buffer_head *
3037 ext4_xattr_block_cache_find(struct inode *inode,
3038                             struct ext4_xattr_header *header,
3039                             struct mb_cache_entry **pce)
3040 {
3041         __u32 hash = le32_to_cpu(header->h_hash);
3042         struct mb_cache_entry *ce;
3043         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3044
3045         if (!ea_block_cache)
3046                 return NULL;
3047         if (!header->h_hash)
3048                 return NULL;  /* never share */
3049         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3050         ce = mb_cache_entry_find_first(ea_block_cache, hash);
3051         while (ce) {
3052                 struct buffer_head *bh;
3053
3054                 bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3055                 if (IS_ERR(bh)) {
3056                         if (PTR_ERR(bh) == -ENOMEM)
3057                                 return NULL;
3058                         bh = NULL;
3059                         EXT4_ERROR_INODE(inode, "block %lu read error",
3060                                          (unsigned long)ce->e_value);
3061                 } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3062                         *pce = ce;
3063                         return bh;
3064                 }
3065                 brelse(bh);
3066                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
3067         }
3068         return NULL;
3069 }
3070
3071 #define NAME_HASH_SHIFT 5
3072 #define VALUE_HASH_SHIFT 16
3073
3074 /*
3075  * ext4_xattr_hash_entry()
3076  *
3077  * Compute the hash of an extended attribute.
3078  */
3079 static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3080                                     size_t value_count)
3081 {
3082         __u32 hash = 0;
3083
3084         while (name_len--) {
3085                 hash = (hash << NAME_HASH_SHIFT) ^
3086                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3087                        *name++;
3088         }
3089         while (value_count--) {
3090                 hash = (hash << VALUE_HASH_SHIFT) ^
3091                        (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3092                        le32_to_cpu(*value++);
3093         }
3094         return cpu_to_le32(hash);
3095 }
3096
3097 #undef NAME_HASH_SHIFT
3098 #undef VALUE_HASH_SHIFT
3099
3100 #define BLOCK_HASH_SHIFT 16
3101
3102 /*
3103  * ext4_xattr_rehash()
3104  *
3105  * Re-compute the extended attribute hash value after an entry has changed.
3106  */
3107 static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3108 {
3109         struct ext4_xattr_entry *here;
3110         __u32 hash = 0;
3111
3112         here = ENTRY(header+1);
3113         while (!IS_LAST_ENTRY(here)) {
3114                 if (!here->e_hash) {
3115                         /* Block is not shared if an entry's hash value == 0 */
3116                         hash = 0;
3117                         break;
3118                 }
3119                 hash = (hash << BLOCK_HASH_SHIFT) ^
3120                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3121                        le32_to_cpu(here->e_hash);
3122                 here = EXT4_XATTR_NEXT(here);
3123         }
3124         header->h_hash = cpu_to_le32(hash);
3125 }
3126
3127 #undef BLOCK_HASH_SHIFT
3128
3129 #define HASH_BUCKET_BITS        10
3130
3131 struct mb_cache *
3132 ext4_xattr_create_cache(void)
3133 {
3134         return mb_cache_create(HASH_BUCKET_BITS);
3135 }
3136
3137 void ext4_xattr_destroy_cache(struct mb_cache *cache)
3138 {
3139         if (cache)
3140                 mb_cache_destroy(cache);
3141 }
3142